]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
1da177e4 LT |
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> | |
10ed273f | 22 | #include <linux/jiffies.h> |
1da177e4 | 23 | #include <linux/bootmem.h> |
edbe7d23 | 24 | #include <linux/memblock.h> |
1da177e4 | 25 | #include <linux/compiler.h> |
9f158333 | 26 | #include <linux/kernel.h> |
b1eeab67 | 27 | #include <linux/kmemcheck.h> |
1da177e4 LT |
28 | #include <linux/module.h> |
29 | #include <linux/suspend.h> | |
30 | #include <linux/pagevec.h> | |
31 | #include <linux/blkdev.h> | |
32 | #include <linux/slab.h> | |
a238ab5b | 33 | #include <linux/ratelimit.h> |
5a3135c2 | 34 | #include <linux/oom.h> |
1da177e4 LT |
35 | #include <linux/notifier.h> |
36 | #include <linux/topology.h> | |
37 | #include <linux/sysctl.h> | |
38 | #include <linux/cpu.h> | |
39 | #include <linux/cpuset.h> | |
bdc8cb98 | 40 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
41 | #include <linux/nodemask.h> |
42 | #include <linux/vmalloc.h> | |
a6cccdc3 | 43 | #include <linux/vmstat.h> |
4be38e35 | 44 | #include <linux/mempolicy.h> |
6811378e | 45 | #include <linux/stop_machine.h> |
c713216d MG |
46 | #include <linux/sort.h> |
47 | #include <linux/pfn.h> | |
3fcfab16 | 48 | #include <linux/backing-dev.h> |
933e312e | 49 | #include <linux/fault-inject.h> |
a5d76b54 | 50 | #include <linux/page-isolation.h> |
52d4b9ac | 51 | #include <linux/page_cgroup.h> |
3ac7fe5a | 52 | #include <linux/debugobjects.h> |
dbb1f81c | 53 | #include <linux/kmemleak.h> |
56de7263 | 54 | #include <linux/compaction.h> |
0d3d062a | 55 | #include <trace/events/kmem.h> |
718a3821 | 56 | #include <linux/ftrace_event.h> |
f212ad7c | 57 | #include <linux/memcontrol.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
041d3a8c | 59 | #include <linux/migrate.h> |
c0a32fc5 | 60 | #include <linux/page-debug-flags.h> |
8bd75c77 | 61 | #include <linux/sched/rt.h> |
1da177e4 LT |
62 | |
63 | #include <asm/tlbflush.h> | |
ac924c60 | 64 | #include <asm/div64.h> |
1da177e4 LT |
65 | #include "internal.h" |
66 | ||
72812019 LS |
67 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
68 | DEFINE_PER_CPU(int, numa_node); | |
69 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
70 | #endif | |
71 | ||
7aac7898 LS |
72 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
73 | /* | |
74 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
75 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
76 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
77 | * defined in <linux/topology.h>. | |
78 | */ | |
79 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
80 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
81 | #endif | |
82 | ||
1da177e4 | 83 | /* |
13808910 | 84 | * Array of node states. |
1da177e4 | 85 | */ |
13808910 CL |
86 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
87 | [N_POSSIBLE] = NODE_MASK_ALL, | |
88 | [N_ONLINE] = { { [0] = 1UL } }, | |
89 | #ifndef CONFIG_NUMA | |
90 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
91 | #ifdef CONFIG_HIGHMEM | |
92 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
93 | #endif |
94 | #ifdef CONFIG_MOVABLE_NODE | |
95 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
96 | #endif |
97 | [N_CPU] = { { [0] = 1UL } }, | |
98 | #endif /* NUMA */ | |
99 | }; | |
100 | EXPORT_SYMBOL(node_states); | |
101 | ||
6c231b7b | 102 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 103 | unsigned long totalreserve_pages __read_mostly; |
ab8fabd4 JW |
104 | /* |
105 | * When calculating the number of globally allowed dirty pages, there | |
106 | * is a certain number of per-zone reserves that should not be | |
107 | * considered dirtyable memory. This is the sum of those reserves | |
108 | * over all existing zones that contribute dirtyable memory. | |
109 | */ | |
110 | unsigned long dirty_balance_reserve __read_mostly; | |
111 | ||
1b76b02f | 112 | int percpu_pagelist_fraction; |
dcce284a | 113 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 114 | |
452aa699 RW |
115 | #ifdef CONFIG_PM_SLEEP |
116 | /* | |
117 | * The following functions are used by the suspend/hibernate code to temporarily | |
118 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
119 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
120 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
121 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
122 | * guaranteed not to run in parallel with that modification). | |
123 | */ | |
c9e664f1 RW |
124 | |
125 | static gfp_t saved_gfp_mask; | |
126 | ||
127 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
128 | { |
129 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
130 | if (saved_gfp_mask) { |
131 | gfp_allowed_mask = saved_gfp_mask; | |
132 | saved_gfp_mask = 0; | |
133 | } | |
452aa699 RW |
134 | } |
135 | ||
c9e664f1 | 136 | void pm_restrict_gfp_mask(void) |
452aa699 | 137 | { |
452aa699 | 138 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
139 | WARN_ON(saved_gfp_mask); |
140 | saved_gfp_mask = gfp_allowed_mask; | |
141 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 | 142 | } |
f90ac398 MG |
143 | |
144 | bool pm_suspended_storage(void) | |
145 | { | |
146 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) | |
147 | return false; | |
148 | return true; | |
149 | } | |
452aa699 RW |
150 | #endif /* CONFIG_PM_SLEEP */ |
151 | ||
d9c23400 MG |
152 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
153 | int pageblock_order __read_mostly; | |
154 | #endif | |
155 | ||
d98c7a09 | 156 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 157 | |
1da177e4 LT |
158 | /* |
159 | * results with 256, 32 in the lowmem_reserve sysctl: | |
160 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
161 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
162 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
163 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
164 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
165 | * |
166 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
167 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 168 | */ |
2f1b6248 | 169 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 170 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 171 | 256, |
4b51d669 | 172 | #endif |
fb0e7942 | 173 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 174 | 256, |
fb0e7942 | 175 | #endif |
e53ef38d | 176 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 177 | 32, |
e53ef38d | 178 | #endif |
2a1e274a | 179 | 32, |
2f1b6248 | 180 | }; |
1da177e4 LT |
181 | |
182 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 183 | |
15ad7cdc | 184 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 185 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 186 | "DMA", |
4b51d669 | 187 | #endif |
fb0e7942 | 188 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 189 | "DMA32", |
fb0e7942 | 190 | #endif |
2f1b6248 | 191 | "Normal", |
e53ef38d | 192 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 193 | "HighMem", |
e53ef38d | 194 | #endif |
2a1e274a | 195 | "Movable", |
2f1b6248 CL |
196 | }; |
197 | ||
1da177e4 LT |
198 | int min_free_kbytes = 1024; |
199 | ||
2c85f51d JB |
200 | static unsigned long __meminitdata nr_kernel_pages; |
201 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 202 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 203 | |
0ee332c1 | 204 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
34b71f1e TC |
205 | /* Movable memory ranges, will also be used by memblock subsystem. */ |
206 | struct movablemem_map movablemem_map; | |
207 | ||
0ee332c1 TH |
208 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; |
209 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
210 | static unsigned long __initdata required_kernelcore; | |
211 | static unsigned long __initdata required_movablecore; | |
212 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
6981ec31 | 213 | static unsigned long __meminitdata zone_movable_limit[MAX_NUMNODES]; |
0ee332c1 TH |
214 | |
215 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
216 | int movable_zone; | |
217 | EXPORT_SYMBOL(movable_zone); | |
218 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 219 | |
418508c1 MS |
220 | #if MAX_NUMNODES > 1 |
221 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 222 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 223 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 224 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
225 | #endif |
226 | ||
9ef9acb0 MG |
227 | int page_group_by_mobility_disabled __read_mostly; |
228 | ||
ee6f509c | 229 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 230 | { |
49255c61 MG |
231 | |
232 | if (unlikely(page_group_by_mobility_disabled)) | |
233 | migratetype = MIGRATE_UNMOVABLE; | |
234 | ||
b2a0ac88 MG |
235 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
236 | PB_migrate, PB_migrate_end); | |
237 | } | |
238 | ||
7f33d49a RW |
239 | bool oom_killer_disabled __read_mostly; |
240 | ||
13e7444b | 241 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 242 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 243 | { |
bdc8cb98 DH |
244 | int ret = 0; |
245 | unsigned seq; | |
246 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 247 | |
bdc8cb98 DH |
248 | do { |
249 | seq = zone_span_seqbegin(zone); | |
250 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
251 | ret = 1; | |
252 | else if (pfn < zone->zone_start_pfn) | |
253 | ret = 1; | |
254 | } while (zone_span_seqretry(zone, seq)); | |
255 | ||
256 | return ret; | |
c6a57e19 DH |
257 | } |
258 | ||
259 | static int page_is_consistent(struct zone *zone, struct page *page) | |
260 | { | |
14e07298 | 261 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 262 | return 0; |
1da177e4 | 263 | if (zone != page_zone(page)) |
c6a57e19 DH |
264 | return 0; |
265 | ||
266 | return 1; | |
267 | } | |
268 | /* | |
269 | * Temporary debugging check for pages not lying within a given zone. | |
270 | */ | |
271 | static int bad_range(struct zone *zone, struct page *page) | |
272 | { | |
273 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 274 | return 1; |
c6a57e19 DH |
275 | if (!page_is_consistent(zone, page)) |
276 | return 1; | |
277 | ||
1da177e4 LT |
278 | return 0; |
279 | } | |
13e7444b NP |
280 | #else |
281 | static inline int bad_range(struct zone *zone, struct page *page) | |
282 | { | |
283 | return 0; | |
284 | } | |
285 | #endif | |
286 | ||
224abf92 | 287 | static void bad_page(struct page *page) |
1da177e4 | 288 | { |
d936cf9b HD |
289 | static unsigned long resume; |
290 | static unsigned long nr_shown; | |
291 | static unsigned long nr_unshown; | |
292 | ||
2a7684a2 WF |
293 | /* Don't complain about poisoned pages */ |
294 | if (PageHWPoison(page)) { | |
ef2b4b95 | 295 | reset_page_mapcount(page); /* remove PageBuddy */ |
2a7684a2 WF |
296 | return; |
297 | } | |
298 | ||
d936cf9b HD |
299 | /* |
300 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
301 | * or allow a steady drip of one report per second. | |
302 | */ | |
303 | if (nr_shown == 60) { | |
304 | if (time_before(jiffies, resume)) { | |
305 | nr_unshown++; | |
306 | goto out; | |
307 | } | |
308 | if (nr_unshown) { | |
1e9e6365 HD |
309 | printk(KERN_ALERT |
310 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
311 | nr_unshown); |
312 | nr_unshown = 0; | |
313 | } | |
314 | nr_shown = 0; | |
315 | } | |
316 | if (nr_shown++ == 0) | |
317 | resume = jiffies + 60 * HZ; | |
318 | ||
1e9e6365 | 319 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 320 | current->comm, page_to_pfn(page)); |
718a3821 | 321 | dump_page(page); |
3dc14741 | 322 | |
4f31888c | 323 | print_modules(); |
1da177e4 | 324 | dump_stack(); |
d936cf9b | 325 | out: |
8cc3b392 | 326 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
ef2b4b95 | 327 | reset_page_mapcount(page); /* remove PageBuddy */ |
9f158333 | 328 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
329 | } |
330 | ||
1da177e4 LT |
331 | /* |
332 | * Higher-order pages are called "compound pages". They are structured thusly: | |
333 | * | |
334 | * The first PAGE_SIZE page is called the "head page". | |
335 | * | |
336 | * The remaining PAGE_SIZE pages are called "tail pages". | |
337 | * | |
6416b9fa WSH |
338 | * All pages have PG_compound set. All tail pages have their ->first_page |
339 | * pointing at the head page. | |
1da177e4 | 340 | * |
41d78ba5 HD |
341 | * The first tail page's ->lru.next holds the address of the compound page's |
342 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
343 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 344 | */ |
d98c7a09 HD |
345 | |
346 | static void free_compound_page(struct page *page) | |
347 | { | |
d85f3385 | 348 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
349 | } |
350 | ||
01ad1c08 | 351 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
352 | { |
353 | int i; | |
354 | int nr_pages = 1 << order; | |
355 | ||
356 | set_compound_page_dtor(page, free_compound_page); | |
357 | set_compound_order(page, order); | |
358 | __SetPageHead(page); | |
359 | for (i = 1; i < nr_pages; i++) { | |
360 | struct page *p = page + i; | |
18229df5 | 361 | __SetPageTail(p); |
58a84aa9 | 362 | set_page_count(p, 0); |
18229df5 AW |
363 | p->first_page = page; |
364 | } | |
365 | } | |
366 | ||
59ff4216 | 367 | /* update __split_huge_page_refcount if you change this function */ |
8cc3b392 | 368 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
369 | { |
370 | int i; | |
371 | int nr_pages = 1 << order; | |
8cc3b392 | 372 | int bad = 0; |
1da177e4 | 373 | |
0bb2c763 | 374 | if (unlikely(compound_order(page) != order)) { |
224abf92 | 375 | bad_page(page); |
8cc3b392 HD |
376 | bad++; |
377 | } | |
1da177e4 | 378 | |
6d777953 | 379 | __ClearPageHead(page); |
8cc3b392 | 380 | |
18229df5 AW |
381 | for (i = 1; i < nr_pages; i++) { |
382 | struct page *p = page + i; | |
1da177e4 | 383 | |
e713a21d | 384 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 385 | bad_page(page); |
8cc3b392 HD |
386 | bad++; |
387 | } | |
d85f3385 | 388 | __ClearPageTail(p); |
1da177e4 | 389 | } |
8cc3b392 HD |
390 | |
391 | return bad; | |
1da177e4 | 392 | } |
1da177e4 | 393 | |
17cf4406 NP |
394 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
395 | { | |
396 | int i; | |
397 | ||
6626c5d5 AM |
398 | /* |
399 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
400 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
401 | */ | |
725d704e | 402 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
403 | for (i = 0; i < (1 << order); i++) |
404 | clear_highpage(page + i); | |
405 | } | |
406 | ||
c0a32fc5 SG |
407 | #ifdef CONFIG_DEBUG_PAGEALLOC |
408 | unsigned int _debug_guardpage_minorder; | |
409 | ||
410 | static int __init debug_guardpage_minorder_setup(char *buf) | |
411 | { | |
412 | unsigned long res; | |
413 | ||
414 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
415 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
416 | return 0; | |
417 | } | |
418 | _debug_guardpage_minorder = res; | |
419 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
420 | return 0; | |
421 | } | |
422 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
423 | ||
424 | static inline void set_page_guard_flag(struct page *page) | |
425 | { | |
426 | __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
427 | } | |
428 | ||
429 | static inline void clear_page_guard_flag(struct page *page) | |
430 | { | |
431 | __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
432 | } | |
433 | #else | |
434 | static inline void set_page_guard_flag(struct page *page) { } | |
435 | static inline void clear_page_guard_flag(struct page *page) { } | |
436 | #endif | |
437 | ||
6aa3001b AM |
438 | static inline void set_page_order(struct page *page, int order) |
439 | { | |
4c21e2f2 | 440 | set_page_private(page, order); |
676165a8 | 441 | __SetPageBuddy(page); |
1da177e4 LT |
442 | } |
443 | ||
444 | static inline void rmv_page_order(struct page *page) | |
445 | { | |
676165a8 | 446 | __ClearPageBuddy(page); |
4c21e2f2 | 447 | set_page_private(page, 0); |
1da177e4 LT |
448 | } |
449 | ||
450 | /* | |
451 | * Locate the struct page for both the matching buddy in our | |
452 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
453 | * | |
454 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
455 | * the following equation: | |
456 | * B2 = B1 ^ (1 << O) | |
457 | * For example, if the starting buddy (buddy2) is #8 its order | |
458 | * 1 buddy is #10: | |
459 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
460 | * | |
461 | * 2) Any buddy B will have an order O+1 parent P which | |
462 | * satisfies the following equation: | |
463 | * P = B & ~(1 << O) | |
464 | * | |
d6e05edc | 465 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 | 466 | */ |
1da177e4 | 467 | static inline unsigned long |
43506fad | 468 | __find_buddy_index(unsigned long page_idx, unsigned int order) |
1da177e4 | 469 | { |
43506fad | 470 | return page_idx ^ (1 << order); |
1da177e4 LT |
471 | } |
472 | ||
473 | /* | |
474 | * This function checks whether a page is free && is the buddy | |
475 | * we can do coalesce a page and its buddy if | |
13e7444b | 476 | * (a) the buddy is not in a hole && |
676165a8 | 477 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
478 | * (c) a page and its buddy have the same order && |
479 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 480 | * |
5f24ce5f AA |
481 | * For recording whether a page is in the buddy system, we set ->_mapcount -2. |
482 | * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock. | |
1da177e4 | 483 | * |
676165a8 | 484 | * For recording page's order, we use page_private(page). |
1da177e4 | 485 | */ |
cb2b95e1 AW |
486 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
487 | int order) | |
1da177e4 | 488 | { |
14e07298 | 489 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 490 | return 0; |
13e7444b | 491 | |
cb2b95e1 AW |
492 | if (page_zone_id(page) != page_zone_id(buddy)) |
493 | return 0; | |
494 | ||
c0a32fc5 SG |
495 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
496 | VM_BUG_ON(page_count(buddy) != 0); | |
497 | return 1; | |
498 | } | |
499 | ||
cb2b95e1 | 500 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
a3af9c38 | 501 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 502 | return 1; |
676165a8 | 503 | } |
6aa3001b | 504 | return 0; |
1da177e4 LT |
505 | } |
506 | ||
507 | /* | |
508 | * Freeing function for a buddy system allocator. | |
509 | * | |
510 | * The concept of a buddy system is to maintain direct-mapped table | |
511 | * (containing bit values) for memory blocks of various "orders". | |
512 | * The bottom level table contains the map for the smallest allocatable | |
513 | * units of memory (here, pages), and each level above it describes | |
514 | * pairs of units from the levels below, hence, "buddies". | |
515 | * At a high level, all that happens here is marking the table entry | |
516 | * at the bottom level available, and propagating the changes upward | |
517 | * as necessary, plus some accounting needed to play nicely with other | |
518 | * parts of the VM system. | |
519 | * At each level, we keep a list of pages, which are heads of continuous | |
5f24ce5f | 520 | * free pages of length of (1 << order) and marked with _mapcount -2. Page's |
4c21e2f2 | 521 | * order is recorded in page_private(page) field. |
1da177e4 | 522 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
523 | * other. That is, if we allocate a small block, and both were |
524 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 525 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 526 | * triggers coalescing into a block of larger size. |
1da177e4 | 527 | * |
6d49e352 | 528 | * -- nyc |
1da177e4 LT |
529 | */ |
530 | ||
48db57f8 | 531 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
532 | struct zone *zone, unsigned int order, |
533 | int migratetype) | |
1da177e4 LT |
534 | { |
535 | unsigned long page_idx; | |
6dda9d55 | 536 | unsigned long combined_idx; |
43506fad | 537 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 538 | struct page *buddy; |
1da177e4 | 539 | |
224abf92 | 540 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
541 | if (unlikely(destroy_compound_page(page, order))) |
542 | return; | |
1da177e4 | 543 | |
ed0ae21d MG |
544 | VM_BUG_ON(migratetype == -1); |
545 | ||
1da177e4 LT |
546 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
547 | ||
f2260e6b | 548 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 549 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 550 | |
1da177e4 | 551 | while (order < MAX_ORDER-1) { |
43506fad KC |
552 | buddy_idx = __find_buddy_index(page_idx, order); |
553 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 554 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 555 | break; |
c0a32fc5 SG |
556 | /* |
557 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
558 | * merge with it and move up one order. | |
559 | */ | |
560 | if (page_is_guard(buddy)) { | |
561 | clear_page_guard_flag(buddy); | |
562 | set_page_private(page, 0); | |
d1ce749a BZ |
563 | __mod_zone_freepage_state(zone, 1 << order, |
564 | migratetype); | |
c0a32fc5 SG |
565 | } else { |
566 | list_del(&buddy->lru); | |
567 | zone->free_area[order].nr_free--; | |
568 | rmv_page_order(buddy); | |
569 | } | |
43506fad | 570 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
571 | page = page + (combined_idx - page_idx); |
572 | page_idx = combined_idx; | |
573 | order++; | |
574 | } | |
575 | set_page_order(page, order); | |
6dda9d55 CZ |
576 | |
577 | /* | |
578 | * If this is not the largest possible page, check if the buddy | |
579 | * of the next-highest order is free. If it is, it's possible | |
580 | * that pages are being freed that will coalesce soon. In case, | |
581 | * that is happening, add the free page to the tail of the list | |
582 | * so it's less likely to be used soon and more likely to be merged | |
583 | * as a higher order page | |
584 | */ | |
b7f50cfa | 585 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 586 | struct page *higher_page, *higher_buddy; |
43506fad KC |
587 | combined_idx = buddy_idx & page_idx; |
588 | higher_page = page + (combined_idx - page_idx); | |
589 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 590 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
591 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
592 | list_add_tail(&page->lru, | |
593 | &zone->free_area[order].free_list[migratetype]); | |
594 | goto out; | |
595 | } | |
596 | } | |
597 | ||
598 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
599 | out: | |
1da177e4 LT |
600 | zone->free_area[order].nr_free++; |
601 | } | |
602 | ||
224abf92 | 603 | static inline int free_pages_check(struct page *page) |
1da177e4 | 604 | { |
92be2e33 NP |
605 | if (unlikely(page_mapcount(page) | |
606 | (page->mapping != NULL) | | |
a3af9c38 | 607 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
608 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE) | |
609 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 610 | bad_page(page); |
79f4b7bf | 611 | return 1; |
8cc3b392 | 612 | } |
57e0a030 | 613 | reset_page_last_nid(page); |
79f4b7bf HD |
614 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
615 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
616 | return 0; | |
1da177e4 LT |
617 | } |
618 | ||
619 | /* | |
5f8dcc21 | 620 | * Frees a number of pages from the PCP lists |
1da177e4 | 621 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 622 | * count is the number of pages to free. |
1da177e4 LT |
623 | * |
624 | * If the zone was previously in an "all pages pinned" state then look to | |
625 | * see if this freeing clears that state. | |
626 | * | |
627 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
628 | * pinned" detection logic. | |
629 | */ | |
5f8dcc21 MG |
630 | static void free_pcppages_bulk(struct zone *zone, int count, |
631 | struct per_cpu_pages *pcp) | |
1da177e4 | 632 | { |
5f8dcc21 | 633 | int migratetype = 0; |
a6f9edd6 | 634 | int batch_free = 0; |
72853e29 | 635 | int to_free = count; |
5f8dcc21 | 636 | |
c54ad30c | 637 | spin_lock(&zone->lock); |
93e4a89a | 638 | zone->all_unreclaimable = 0; |
1da177e4 | 639 | zone->pages_scanned = 0; |
f2260e6b | 640 | |
72853e29 | 641 | while (to_free) { |
48db57f8 | 642 | struct page *page; |
5f8dcc21 MG |
643 | struct list_head *list; |
644 | ||
645 | /* | |
a6f9edd6 MG |
646 | * Remove pages from lists in a round-robin fashion. A |
647 | * batch_free count is maintained that is incremented when an | |
648 | * empty list is encountered. This is so more pages are freed | |
649 | * off fuller lists instead of spinning excessively around empty | |
650 | * lists | |
5f8dcc21 MG |
651 | */ |
652 | do { | |
a6f9edd6 | 653 | batch_free++; |
5f8dcc21 MG |
654 | if (++migratetype == MIGRATE_PCPTYPES) |
655 | migratetype = 0; | |
656 | list = &pcp->lists[migratetype]; | |
657 | } while (list_empty(list)); | |
48db57f8 | 658 | |
1d16871d NK |
659 | /* This is the only non-empty list. Free them all. */ |
660 | if (batch_free == MIGRATE_PCPTYPES) | |
661 | batch_free = to_free; | |
662 | ||
a6f9edd6 | 663 | do { |
770c8aaa BZ |
664 | int mt; /* migratetype of the to-be-freed page */ |
665 | ||
a6f9edd6 MG |
666 | page = list_entry(list->prev, struct page, lru); |
667 | /* must delete as __free_one_page list manipulates */ | |
668 | list_del(&page->lru); | |
b12c4ad1 | 669 | mt = get_freepage_migratetype(page); |
a7016235 | 670 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
770c8aaa BZ |
671 | __free_one_page(page, zone, 0, mt); |
672 | trace_mm_page_pcpu_drain(page, 0, mt); | |
97d0da22 WC |
673 | if (likely(get_pageblock_migratetype(page) != MIGRATE_ISOLATE)) { |
674 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1); | |
675 | if (is_migrate_cma(mt)) | |
676 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1); | |
677 | } | |
72853e29 | 678 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 679 | } |
c54ad30c | 680 | spin_unlock(&zone->lock); |
1da177e4 LT |
681 | } |
682 | ||
ed0ae21d MG |
683 | static void free_one_page(struct zone *zone, struct page *page, int order, |
684 | int migratetype) | |
1da177e4 | 685 | { |
006d22d9 | 686 | spin_lock(&zone->lock); |
93e4a89a | 687 | zone->all_unreclaimable = 0; |
006d22d9 | 688 | zone->pages_scanned = 0; |
f2260e6b | 689 | |
ed0ae21d | 690 | __free_one_page(page, zone, order, migratetype); |
2139cbe6 | 691 | if (unlikely(migratetype != MIGRATE_ISOLATE)) |
d1ce749a | 692 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
006d22d9 | 693 | spin_unlock(&zone->lock); |
48db57f8 NP |
694 | } |
695 | ||
ec95f53a | 696 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 697 | { |
1da177e4 | 698 | int i; |
8cc3b392 | 699 | int bad = 0; |
1da177e4 | 700 | |
b413d48a | 701 | trace_mm_page_free(page, order); |
b1eeab67 VN |
702 | kmemcheck_free_shadow(page, order); |
703 | ||
8dd60a3a AA |
704 | if (PageAnon(page)) |
705 | page->mapping = NULL; | |
706 | for (i = 0; i < (1 << order); i++) | |
707 | bad += free_pages_check(page + i); | |
8cc3b392 | 708 | if (bad) |
ec95f53a | 709 | return false; |
689bcebf | 710 | |
3ac7fe5a | 711 | if (!PageHighMem(page)) { |
9858db50 | 712 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
713 | debug_check_no_obj_freed(page_address(page), |
714 | PAGE_SIZE << order); | |
715 | } | |
dafb1367 | 716 | arch_free_page(page, order); |
48db57f8 | 717 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 718 | |
ec95f53a KM |
719 | return true; |
720 | } | |
721 | ||
722 | static void __free_pages_ok(struct page *page, unsigned int order) | |
723 | { | |
724 | unsigned long flags; | |
95e34412 | 725 | int migratetype; |
ec95f53a KM |
726 | |
727 | if (!free_pages_prepare(page, order)) | |
728 | return; | |
729 | ||
c54ad30c | 730 | local_irq_save(flags); |
f8891e5e | 731 | __count_vm_events(PGFREE, 1 << order); |
95e34412 MK |
732 | migratetype = get_pageblock_migratetype(page); |
733 | set_freepage_migratetype(page, migratetype); | |
734 | free_one_page(page_zone(page), page, order, migratetype); | |
c54ad30c | 735 | local_irq_restore(flags); |
1da177e4 LT |
736 | } |
737 | ||
9feedc9d JL |
738 | /* |
739 | * Read access to zone->managed_pages is safe because it's unsigned long, | |
740 | * but we still need to serialize writers. Currently all callers of | |
741 | * __free_pages_bootmem() except put_page_bootmem() should only be used | |
742 | * at boot time. So for shorter boot time, we shift the burden to | |
743 | * put_page_bootmem() to serialize writers. | |
744 | */ | |
af370fb8 | 745 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 | 746 | { |
c3993076 JW |
747 | unsigned int nr_pages = 1 << order; |
748 | unsigned int loop; | |
a226f6c8 | 749 | |
c3993076 JW |
750 | prefetchw(page); |
751 | for (loop = 0; loop < nr_pages; loop++) { | |
752 | struct page *p = &page[loop]; | |
753 | ||
754 | if (loop + 1 < nr_pages) | |
755 | prefetchw(p + 1); | |
756 | __ClearPageReserved(p); | |
757 | set_page_count(p, 0); | |
a226f6c8 | 758 | } |
c3993076 | 759 | |
9feedc9d | 760 | page_zone(page)->managed_pages += 1 << order; |
c3993076 JW |
761 | set_page_refcounted(page); |
762 | __free_pages(page, order); | |
a226f6c8 DH |
763 | } |
764 | ||
47118af0 MN |
765 | #ifdef CONFIG_CMA |
766 | /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */ | |
767 | void __init init_cma_reserved_pageblock(struct page *page) | |
768 | { | |
769 | unsigned i = pageblock_nr_pages; | |
770 | struct page *p = page; | |
771 | ||
772 | do { | |
773 | __ClearPageReserved(p); | |
774 | set_page_count(p, 0); | |
775 | } while (++p, --i); | |
776 | ||
777 | set_page_refcounted(page); | |
778 | set_pageblock_migratetype(page, MIGRATE_CMA); | |
779 | __free_pages(page, pageblock_order); | |
780 | totalram_pages += pageblock_nr_pages; | |
41a79734 MS |
781 | #ifdef CONFIG_HIGHMEM |
782 | if (PageHighMem(page)) | |
783 | totalhigh_pages += pageblock_nr_pages; | |
784 | #endif | |
47118af0 MN |
785 | } |
786 | #endif | |
1da177e4 LT |
787 | |
788 | /* | |
789 | * The order of subdivision here is critical for the IO subsystem. | |
790 | * Please do not alter this order without good reasons and regression | |
791 | * testing. Specifically, as large blocks of memory are subdivided, | |
792 | * the order in which smaller blocks are delivered depends on the order | |
793 | * they're subdivided in this function. This is the primary factor | |
794 | * influencing the order in which pages are delivered to the IO | |
795 | * subsystem according to empirical testing, and this is also justified | |
796 | * by considering the behavior of a buddy system containing a single | |
797 | * large block of memory acted on by a series of small allocations. | |
798 | * This behavior is a critical factor in sglist merging's success. | |
799 | * | |
6d49e352 | 800 | * -- nyc |
1da177e4 | 801 | */ |
085cc7d5 | 802 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
803 | int low, int high, struct free_area *area, |
804 | int migratetype) | |
1da177e4 LT |
805 | { |
806 | unsigned long size = 1 << high; | |
807 | ||
808 | while (high > low) { | |
809 | area--; | |
810 | high--; | |
811 | size >>= 1; | |
725d704e | 812 | VM_BUG_ON(bad_range(zone, &page[size])); |
c0a32fc5 SG |
813 | |
814 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
815 | if (high < debug_guardpage_minorder()) { | |
816 | /* | |
817 | * Mark as guard pages (or page), that will allow to | |
818 | * merge back to allocator when buddy will be freed. | |
819 | * Corresponding page table entries will not be touched, | |
820 | * pages will stay not present in virtual address space | |
821 | */ | |
822 | INIT_LIST_HEAD(&page[size].lru); | |
823 | set_page_guard_flag(&page[size]); | |
824 | set_page_private(&page[size], high); | |
825 | /* Guard pages are not available for any usage */ | |
d1ce749a BZ |
826 | __mod_zone_freepage_state(zone, -(1 << high), |
827 | migratetype); | |
c0a32fc5 SG |
828 | continue; |
829 | } | |
830 | #endif | |
b2a0ac88 | 831 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
832 | area->nr_free++; |
833 | set_page_order(&page[size], high); | |
834 | } | |
1da177e4 LT |
835 | } |
836 | ||
1da177e4 LT |
837 | /* |
838 | * This page is about to be returned from the page allocator | |
839 | */ | |
2a7684a2 | 840 | static inline int check_new_page(struct page *page) |
1da177e4 | 841 | { |
92be2e33 NP |
842 | if (unlikely(page_mapcount(page) | |
843 | (page->mapping != NULL) | | |
a3af9c38 | 844 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
845 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP) | |
846 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 847 | bad_page(page); |
689bcebf | 848 | return 1; |
8cc3b392 | 849 | } |
2a7684a2 WF |
850 | return 0; |
851 | } | |
852 | ||
853 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
854 | { | |
855 | int i; | |
856 | ||
857 | for (i = 0; i < (1 << order); i++) { | |
858 | struct page *p = page + i; | |
859 | if (unlikely(check_new_page(p))) | |
860 | return 1; | |
861 | } | |
689bcebf | 862 | |
4c21e2f2 | 863 | set_page_private(page, 0); |
7835e98b | 864 | set_page_refcounted(page); |
cc102509 NP |
865 | |
866 | arch_alloc_page(page, order); | |
1da177e4 | 867 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
868 | |
869 | if (gfp_flags & __GFP_ZERO) | |
870 | prep_zero_page(page, order, gfp_flags); | |
871 | ||
872 | if (order && (gfp_flags & __GFP_COMP)) | |
873 | prep_compound_page(page, order); | |
874 | ||
689bcebf | 875 | return 0; |
1da177e4 LT |
876 | } |
877 | ||
56fd56b8 MG |
878 | /* |
879 | * Go through the free lists for the given migratetype and remove | |
880 | * the smallest available page from the freelists | |
881 | */ | |
728ec980 MG |
882 | static inline |
883 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
884 | int migratetype) |
885 | { | |
886 | unsigned int current_order; | |
887 | struct free_area * area; | |
888 | struct page *page; | |
889 | ||
890 | /* Find a page of the appropriate size in the preferred list */ | |
891 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
892 | area = &(zone->free_area[current_order]); | |
893 | if (list_empty(&area->free_list[migratetype])) | |
894 | continue; | |
895 | ||
896 | page = list_entry(area->free_list[migratetype].next, | |
897 | struct page, lru); | |
898 | list_del(&page->lru); | |
899 | rmv_page_order(page); | |
900 | area->nr_free--; | |
56fd56b8 MG |
901 | expand(zone, page, order, current_order, area, migratetype); |
902 | return page; | |
903 | } | |
904 | ||
905 | return NULL; | |
906 | } | |
907 | ||
908 | ||
b2a0ac88 MG |
909 | /* |
910 | * This array describes the order lists are fallen back to when | |
911 | * the free lists for the desirable migrate type are depleted | |
912 | */ | |
47118af0 MN |
913 | static int fallbacks[MIGRATE_TYPES][4] = { |
914 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
915 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
916 | #ifdef CONFIG_CMA | |
917 | [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
918 | [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */ | |
919 | #else | |
920 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
921 | #endif | |
6d4a4916 MN |
922 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */ |
923 | [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
924 | }; |
925 | ||
c361be55 MG |
926 | /* |
927 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 928 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
929 | * boundary. If alignment is required, use move_freepages_block() |
930 | */ | |
435b405c | 931 | int move_freepages(struct zone *zone, |
b69a7288 AB |
932 | struct page *start_page, struct page *end_page, |
933 | int migratetype) | |
c361be55 MG |
934 | { |
935 | struct page *page; | |
936 | unsigned long order; | |
d100313f | 937 | int pages_moved = 0; |
c361be55 MG |
938 | |
939 | #ifndef CONFIG_HOLES_IN_ZONE | |
940 | /* | |
941 | * page_zone is not safe to call in this context when | |
942 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
943 | * anyway as we check zone boundaries in move_freepages_block(). | |
944 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 945 | * grouping pages by mobility |
c361be55 MG |
946 | */ |
947 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
948 | #endif | |
949 | ||
950 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
951 | /* Make sure we are not inadvertently changing nodes */ |
952 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
953 | ||
c361be55 MG |
954 | if (!pfn_valid_within(page_to_pfn(page))) { |
955 | page++; | |
956 | continue; | |
957 | } | |
958 | ||
959 | if (!PageBuddy(page)) { | |
960 | page++; | |
961 | continue; | |
962 | } | |
963 | ||
964 | order = page_order(page); | |
84be48d8 KS |
965 | list_move(&page->lru, |
966 | &zone->free_area[order].free_list[migratetype]); | |
95e34412 | 967 | set_freepage_migratetype(page, migratetype); |
c361be55 | 968 | page += 1 << order; |
d100313f | 969 | pages_moved += 1 << order; |
c361be55 MG |
970 | } |
971 | ||
d100313f | 972 | return pages_moved; |
c361be55 MG |
973 | } |
974 | ||
ee6f509c | 975 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 976 | int migratetype) |
c361be55 MG |
977 | { |
978 | unsigned long start_pfn, end_pfn; | |
979 | struct page *start_page, *end_page; | |
980 | ||
981 | start_pfn = page_to_pfn(page); | |
d9c23400 | 982 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 983 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
984 | end_page = start_page + pageblock_nr_pages - 1; |
985 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
986 | |
987 | /* Do not cross zone boundaries */ | |
988 | if (start_pfn < zone->zone_start_pfn) | |
989 | start_page = page; | |
990 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
991 | return 0; | |
992 | ||
993 | return move_freepages(zone, start_page, end_page, migratetype); | |
994 | } | |
995 | ||
2f66a68f MG |
996 | static void change_pageblock_range(struct page *pageblock_page, |
997 | int start_order, int migratetype) | |
998 | { | |
999 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1000 | ||
1001 | while (nr_pageblocks--) { | |
1002 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1003 | pageblock_page += pageblock_nr_pages; | |
1004 | } | |
1005 | } | |
1006 | ||
b2a0ac88 | 1007 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
1008 | static inline struct page * |
1009 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 MG |
1010 | { |
1011 | struct free_area * area; | |
1012 | int current_order; | |
1013 | struct page *page; | |
1014 | int migratetype, i; | |
1015 | ||
1016 | /* Find the largest possible block of pages in the other list */ | |
1017 | for (current_order = MAX_ORDER-1; current_order >= order; | |
1018 | --current_order) { | |
6d4a4916 | 1019 | for (i = 0;; i++) { |
b2a0ac88 MG |
1020 | migratetype = fallbacks[start_migratetype][i]; |
1021 | ||
56fd56b8 MG |
1022 | /* MIGRATE_RESERVE handled later if necessary */ |
1023 | if (migratetype == MIGRATE_RESERVE) | |
6d4a4916 | 1024 | break; |
e010487d | 1025 | |
b2a0ac88 MG |
1026 | area = &(zone->free_area[current_order]); |
1027 | if (list_empty(&area->free_list[migratetype])) | |
1028 | continue; | |
1029 | ||
1030 | page = list_entry(area->free_list[migratetype].next, | |
1031 | struct page, lru); | |
1032 | area->nr_free--; | |
1033 | ||
1034 | /* | |
c361be55 | 1035 | * If breaking a large block of pages, move all free |
46dafbca MG |
1036 | * pages to the preferred allocation list. If falling |
1037 | * back for a reclaimable kernel allocation, be more | |
25985edc | 1038 | * aggressive about taking ownership of free pages |
47118af0 MN |
1039 | * |
1040 | * On the other hand, never change migration | |
1041 | * type of MIGRATE_CMA pageblocks nor move CMA | |
1042 | * pages on different free lists. We don't | |
1043 | * want unmovable pages to be allocated from | |
1044 | * MIGRATE_CMA areas. | |
b2a0ac88 | 1045 | */ |
47118af0 MN |
1046 | if (!is_migrate_cma(migratetype) && |
1047 | (unlikely(current_order >= pageblock_order / 2) || | |
1048 | start_migratetype == MIGRATE_RECLAIMABLE || | |
1049 | page_group_by_mobility_disabled)) { | |
1050 | int pages; | |
46dafbca MG |
1051 | pages = move_freepages_block(zone, page, |
1052 | start_migratetype); | |
1053 | ||
1054 | /* Claim the whole block if over half of it is free */ | |
dd5d241e MG |
1055 | if (pages >= (1 << (pageblock_order-1)) || |
1056 | page_group_by_mobility_disabled) | |
46dafbca MG |
1057 | set_pageblock_migratetype(page, |
1058 | start_migratetype); | |
1059 | ||
b2a0ac88 | 1060 | migratetype = start_migratetype; |
c361be55 | 1061 | } |
b2a0ac88 MG |
1062 | |
1063 | /* Remove the page from the freelists */ | |
1064 | list_del(&page->lru); | |
1065 | rmv_page_order(page); | |
b2a0ac88 | 1066 | |
2f66a68f | 1067 | /* Take ownership for orders >= pageblock_order */ |
47118af0 MN |
1068 | if (current_order >= pageblock_order && |
1069 | !is_migrate_cma(migratetype)) | |
2f66a68f | 1070 | change_pageblock_range(page, current_order, |
b2a0ac88 MG |
1071 | start_migratetype); |
1072 | ||
47118af0 MN |
1073 | expand(zone, page, order, current_order, area, |
1074 | is_migrate_cma(migratetype) | |
1075 | ? migratetype : start_migratetype); | |
e0fff1bd MG |
1076 | |
1077 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
1078 | start_migratetype, migratetype); | |
1079 | ||
b2a0ac88 MG |
1080 | return page; |
1081 | } | |
1082 | } | |
1083 | ||
728ec980 | 1084 | return NULL; |
b2a0ac88 MG |
1085 | } |
1086 | ||
56fd56b8 | 1087 | /* |
1da177e4 LT |
1088 | * Do the hard work of removing an element from the buddy allocator. |
1089 | * Call me with the zone->lock already held. | |
1090 | */ | |
b2a0ac88 MG |
1091 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1092 | int migratetype) | |
1da177e4 | 1093 | { |
1da177e4 LT |
1094 | struct page *page; |
1095 | ||
728ec980 | 1096 | retry_reserve: |
56fd56b8 | 1097 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1098 | |
728ec980 | 1099 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 1100 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 1101 | |
728ec980 MG |
1102 | /* |
1103 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1104 | * is used because __rmqueue_smallest is an inline function | |
1105 | * and we want just one call site | |
1106 | */ | |
1107 | if (!page) { | |
1108 | migratetype = MIGRATE_RESERVE; | |
1109 | goto retry_reserve; | |
1110 | } | |
1111 | } | |
1112 | ||
0d3d062a | 1113 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1114 | return page; |
1da177e4 LT |
1115 | } |
1116 | ||
5f63b720 | 1117 | /* |
1da177e4 LT |
1118 | * Obtain a specified number of elements from the buddy allocator, all under |
1119 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1120 | * Returns the number of new pages which were placed at *list. | |
1121 | */ | |
5f63b720 | 1122 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1123 | unsigned long count, struct list_head *list, |
e084b2d9 | 1124 | int migratetype, int cold) |
1da177e4 | 1125 | { |
47118af0 | 1126 | int mt = migratetype, i; |
5f63b720 | 1127 | |
c54ad30c | 1128 | spin_lock(&zone->lock); |
1da177e4 | 1129 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1130 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1131 | if (unlikely(page == NULL)) |
1da177e4 | 1132 | break; |
81eabcbe MG |
1133 | |
1134 | /* | |
1135 | * Split buddy pages returned by expand() are received here | |
1136 | * in physical page order. The page is added to the callers and | |
1137 | * list and the list head then moves forward. From the callers | |
1138 | * perspective, the linked list is ordered by page number in | |
1139 | * some conditions. This is useful for IO devices that can | |
1140 | * merge IO requests if the physical pages are ordered | |
1141 | * properly. | |
1142 | */ | |
e084b2d9 MG |
1143 | if (likely(cold == 0)) |
1144 | list_add(&page->lru, list); | |
1145 | else | |
1146 | list_add_tail(&page->lru, list); | |
47118af0 MN |
1147 | if (IS_ENABLED(CONFIG_CMA)) { |
1148 | mt = get_pageblock_migratetype(page); | |
1149 | if (!is_migrate_cma(mt) && mt != MIGRATE_ISOLATE) | |
1150 | mt = migratetype; | |
1151 | } | |
b12c4ad1 | 1152 | set_freepage_migratetype(page, mt); |
81eabcbe | 1153 | list = &page->lru; |
d1ce749a BZ |
1154 | if (is_migrate_cma(mt)) |
1155 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, | |
1156 | -(1 << order)); | |
1da177e4 | 1157 | } |
f2260e6b | 1158 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1159 | spin_unlock(&zone->lock); |
085cc7d5 | 1160 | return i; |
1da177e4 LT |
1161 | } |
1162 | ||
4ae7c039 | 1163 | #ifdef CONFIG_NUMA |
8fce4d8e | 1164 | /* |
4037d452 CL |
1165 | * Called from the vmstat counter updater to drain pagesets of this |
1166 | * currently executing processor on remote nodes after they have | |
1167 | * expired. | |
1168 | * | |
879336c3 CL |
1169 | * Note that this function must be called with the thread pinned to |
1170 | * a single processor. | |
8fce4d8e | 1171 | */ |
4037d452 | 1172 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1173 | { |
4ae7c039 | 1174 | unsigned long flags; |
4037d452 | 1175 | int to_drain; |
4ae7c039 | 1176 | |
4037d452 CL |
1177 | local_irq_save(flags); |
1178 | if (pcp->count >= pcp->batch) | |
1179 | to_drain = pcp->batch; | |
1180 | else | |
1181 | to_drain = pcp->count; | |
2a13515c KM |
1182 | if (to_drain > 0) { |
1183 | free_pcppages_bulk(zone, to_drain, pcp); | |
1184 | pcp->count -= to_drain; | |
1185 | } | |
4037d452 | 1186 | local_irq_restore(flags); |
4ae7c039 CL |
1187 | } |
1188 | #endif | |
1189 | ||
9f8f2172 CL |
1190 | /* |
1191 | * Drain pages of the indicated processor. | |
1192 | * | |
1193 | * The processor must either be the current processor and the | |
1194 | * thread pinned to the current processor or a processor that | |
1195 | * is not online. | |
1196 | */ | |
1197 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1198 | { |
c54ad30c | 1199 | unsigned long flags; |
1da177e4 | 1200 | struct zone *zone; |
1da177e4 | 1201 | |
ee99c71c | 1202 | for_each_populated_zone(zone) { |
1da177e4 | 1203 | struct per_cpu_pageset *pset; |
3dfa5721 | 1204 | struct per_cpu_pages *pcp; |
1da177e4 | 1205 | |
99dcc3e5 CL |
1206 | local_irq_save(flags); |
1207 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3dfa5721 CL |
1208 | |
1209 | pcp = &pset->pcp; | |
2ff754fa DR |
1210 | if (pcp->count) { |
1211 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1212 | pcp->count = 0; | |
1213 | } | |
3dfa5721 | 1214 | local_irq_restore(flags); |
1da177e4 LT |
1215 | } |
1216 | } | |
1da177e4 | 1217 | |
9f8f2172 CL |
1218 | /* |
1219 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1220 | */ | |
1221 | void drain_local_pages(void *arg) | |
1222 | { | |
1223 | drain_pages(smp_processor_id()); | |
1224 | } | |
1225 | ||
1226 | /* | |
74046494 GBY |
1227 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
1228 | * | |
1229 | * Note that this code is protected against sending an IPI to an offline | |
1230 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
1231 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
1232 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
1233 | * before the call to on_each_cpu_mask(). | |
9f8f2172 CL |
1234 | */ |
1235 | void drain_all_pages(void) | |
1236 | { | |
74046494 GBY |
1237 | int cpu; |
1238 | struct per_cpu_pageset *pcp; | |
1239 | struct zone *zone; | |
1240 | ||
1241 | /* | |
1242 | * Allocate in the BSS so we wont require allocation in | |
1243 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
1244 | */ | |
1245 | static cpumask_t cpus_with_pcps; | |
1246 | ||
1247 | /* | |
1248 | * We don't care about racing with CPU hotplug event | |
1249 | * as offline notification will cause the notified | |
1250 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
1251 | * disables preemption as part of its processing | |
1252 | */ | |
1253 | for_each_online_cpu(cpu) { | |
1254 | bool has_pcps = false; | |
1255 | for_each_populated_zone(zone) { | |
1256 | pcp = per_cpu_ptr(zone->pageset, cpu); | |
1257 | if (pcp->pcp.count) { | |
1258 | has_pcps = true; | |
1259 | break; | |
1260 | } | |
1261 | } | |
1262 | if (has_pcps) | |
1263 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
1264 | else | |
1265 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
1266 | } | |
1267 | on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1); | |
9f8f2172 CL |
1268 | } |
1269 | ||
296699de | 1270 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1271 | |
1272 | void mark_free_pages(struct zone *zone) | |
1273 | { | |
f623f0db RW |
1274 | unsigned long pfn, max_zone_pfn; |
1275 | unsigned long flags; | |
b2a0ac88 | 1276 | int order, t; |
1da177e4 LT |
1277 | struct list_head *curr; |
1278 | ||
1279 | if (!zone->spanned_pages) | |
1280 | return; | |
1281 | ||
1282 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
1283 | |
1284 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
1285 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1286 | if (pfn_valid(pfn)) { | |
1287 | struct page *page = pfn_to_page(pfn); | |
1288 | ||
7be98234 RW |
1289 | if (!swsusp_page_is_forbidden(page)) |
1290 | swsusp_unset_page_free(page); | |
f623f0db | 1291 | } |
1da177e4 | 1292 | |
b2a0ac88 MG |
1293 | for_each_migratetype_order(order, t) { |
1294 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1295 | unsigned long i; |
1da177e4 | 1296 | |
f623f0db RW |
1297 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1298 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1299 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1300 | } |
b2a0ac88 | 1301 | } |
1da177e4 LT |
1302 | spin_unlock_irqrestore(&zone->lock, flags); |
1303 | } | |
e2c55dc8 | 1304 | #endif /* CONFIG_PM */ |
1da177e4 | 1305 | |
1da177e4 LT |
1306 | /* |
1307 | * Free a 0-order page | |
fc91668e | 1308 | * cold == 1 ? free a cold page : free a hot page |
1da177e4 | 1309 | */ |
fc91668e | 1310 | void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1311 | { |
1312 | struct zone *zone = page_zone(page); | |
1313 | struct per_cpu_pages *pcp; | |
1314 | unsigned long flags; | |
5f8dcc21 | 1315 | int migratetype; |
1da177e4 | 1316 | |
ec95f53a | 1317 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1318 | return; |
1319 | ||
5f8dcc21 | 1320 | migratetype = get_pageblock_migratetype(page); |
b12c4ad1 | 1321 | set_freepage_migratetype(page, migratetype); |
1da177e4 | 1322 | local_irq_save(flags); |
f8891e5e | 1323 | __count_vm_event(PGFREE); |
da456f14 | 1324 | |
5f8dcc21 MG |
1325 | /* |
1326 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1327 | * Free ISOLATE pages back to the allocator because they are being | |
1328 | * offlined but treat RESERVE as movable pages so we can get those | |
1329 | * areas back if necessary. Otherwise, we may have to free | |
1330 | * excessively into the page allocator | |
1331 | */ | |
1332 | if (migratetype >= MIGRATE_PCPTYPES) { | |
1333 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { | |
1334 | free_one_page(zone, page, 0, migratetype); | |
1335 | goto out; | |
1336 | } | |
1337 | migratetype = MIGRATE_MOVABLE; | |
1338 | } | |
1339 | ||
99dcc3e5 | 1340 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3dfa5721 | 1341 | if (cold) |
5f8dcc21 | 1342 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1343 | else |
5f8dcc21 | 1344 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1345 | pcp->count++; |
48db57f8 | 1346 | if (pcp->count >= pcp->high) { |
5f8dcc21 | 1347 | free_pcppages_bulk(zone, pcp->batch, pcp); |
48db57f8 NP |
1348 | pcp->count -= pcp->batch; |
1349 | } | |
5f8dcc21 MG |
1350 | |
1351 | out: | |
1da177e4 | 1352 | local_irq_restore(flags); |
1da177e4 LT |
1353 | } |
1354 | ||
cc59850e KK |
1355 | /* |
1356 | * Free a list of 0-order pages | |
1357 | */ | |
1358 | void free_hot_cold_page_list(struct list_head *list, int cold) | |
1359 | { | |
1360 | struct page *page, *next; | |
1361 | ||
1362 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1363 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1364 | free_hot_cold_page(page, cold); |
1365 | } | |
1366 | } | |
1367 | ||
8dfcc9ba NP |
1368 | /* |
1369 | * split_page takes a non-compound higher-order page, and splits it into | |
1370 | * n (1<<order) sub-pages: page[0..n] | |
1371 | * Each sub-page must be freed individually. | |
1372 | * | |
1373 | * Note: this is probably too low level an operation for use in drivers. | |
1374 | * Please consult with lkml before using this in your driver. | |
1375 | */ | |
1376 | void split_page(struct page *page, unsigned int order) | |
1377 | { | |
1378 | int i; | |
1379 | ||
725d704e NP |
1380 | VM_BUG_ON(PageCompound(page)); |
1381 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1382 | |
1383 | #ifdef CONFIG_KMEMCHECK | |
1384 | /* | |
1385 | * Split shadow pages too, because free(page[0]) would | |
1386 | * otherwise free the whole shadow. | |
1387 | */ | |
1388 | if (kmemcheck_page_is_tracked(page)) | |
1389 | split_page(virt_to_page(page[0].shadow), order); | |
1390 | #endif | |
1391 | ||
7835e98b NP |
1392 | for (i = 1; i < (1 << order); i++) |
1393 | set_page_refcounted(page + i); | |
8dfcc9ba | 1394 | } |
8dfcc9ba | 1395 | |
8fb74b9f | 1396 | static int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 1397 | { |
748446bb MG |
1398 | unsigned long watermark; |
1399 | struct zone *zone; | |
2139cbe6 | 1400 | int mt; |
748446bb MG |
1401 | |
1402 | BUG_ON(!PageBuddy(page)); | |
1403 | ||
1404 | zone = page_zone(page); | |
2e30abd1 | 1405 | mt = get_pageblock_migratetype(page); |
748446bb | 1406 | |
2e30abd1 MS |
1407 | if (mt != MIGRATE_ISOLATE) { |
1408 | /* Obey watermarks as if the page was being allocated */ | |
1409 | watermark = low_wmark_pages(zone) + (1 << order); | |
1410 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1411 | return 0; | |
1412 | ||
8fb74b9f | 1413 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 1414 | } |
748446bb MG |
1415 | |
1416 | /* Remove page from free list */ | |
1417 | list_del(&page->lru); | |
1418 | zone->free_area[order].nr_free--; | |
1419 | rmv_page_order(page); | |
2139cbe6 | 1420 | |
8fb74b9f | 1421 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
1422 | if (order >= pageblock_order - 1) { |
1423 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
1424 | for (; page < endpage; page += pageblock_nr_pages) { |
1425 | int mt = get_pageblock_migratetype(page); | |
1426 | if (mt != MIGRATE_ISOLATE && !is_migrate_cma(mt)) | |
1427 | set_pageblock_migratetype(page, | |
1428 | MIGRATE_MOVABLE); | |
1429 | } | |
748446bb MG |
1430 | } |
1431 | ||
8fb74b9f | 1432 | return 1UL << order; |
1fb3f8ca MG |
1433 | } |
1434 | ||
1435 | /* | |
1436 | * Similar to split_page except the page is already free. As this is only | |
1437 | * being used for migration, the migratetype of the block also changes. | |
1438 | * As this is called with interrupts disabled, the caller is responsible | |
1439 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1440 | * are enabled. | |
1441 | * | |
1442 | * Note: this is probably too low level an operation for use in drivers. | |
1443 | * Please consult with lkml before using this in your driver. | |
1444 | */ | |
1445 | int split_free_page(struct page *page) | |
1446 | { | |
1447 | unsigned int order; | |
1448 | int nr_pages; | |
1449 | ||
1fb3f8ca MG |
1450 | order = page_order(page); |
1451 | ||
8fb74b9f | 1452 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
1453 | if (!nr_pages) |
1454 | return 0; | |
1455 | ||
1456 | /* Split into individual pages */ | |
1457 | set_page_refcounted(page); | |
1458 | split_page(page, order); | |
1459 | return nr_pages; | |
748446bb MG |
1460 | } |
1461 | ||
1da177e4 LT |
1462 | /* |
1463 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1464 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1465 | * or two. | |
1466 | */ | |
0a15c3e9 MG |
1467 | static inline |
1468 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1469 | struct zone *zone, int order, gfp_t gfp_flags, |
1470 | int migratetype) | |
1da177e4 LT |
1471 | { |
1472 | unsigned long flags; | |
689bcebf | 1473 | struct page *page; |
1da177e4 LT |
1474 | int cold = !!(gfp_flags & __GFP_COLD); |
1475 | ||
689bcebf | 1476 | again: |
48db57f8 | 1477 | if (likely(order == 0)) { |
1da177e4 | 1478 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1479 | struct list_head *list; |
1da177e4 | 1480 | |
1da177e4 | 1481 | local_irq_save(flags); |
99dcc3e5 CL |
1482 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1483 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1484 | if (list_empty(list)) { |
535131e6 | 1485 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1486 | pcp->batch, list, |
e084b2d9 | 1487 | migratetype, cold); |
5f8dcc21 | 1488 | if (unlikely(list_empty(list))) |
6fb332fa | 1489 | goto failed; |
535131e6 | 1490 | } |
b92a6edd | 1491 | |
5f8dcc21 MG |
1492 | if (cold) |
1493 | page = list_entry(list->prev, struct page, lru); | |
1494 | else | |
1495 | page = list_entry(list->next, struct page, lru); | |
1496 | ||
b92a6edd MG |
1497 | list_del(&page->lru); |
1498 | pcp->count--; | |
7fb1d9fc | 1499 | } else { |
dab48dab AM |
1500 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1501 | /* | |
1502 | * __GFP_NOFAIL is not to be used in new code. | |
1503 | * | |
1504 | * All __GFP_NOFAIL callers should be fixed so that they | |
1505 | * properly detect and handle allocation failures. | |
1506 | * | |
1507 | * We most definitely don't want callers attempting to | |
4923abf9 | 1508 | * allocate greater than order-1 page units with |
dab48dab AM |
1509 | * __GFP_NOFAIL. |
1510 | */ | |
4923abf9 | 1511 | WARN_ON_ONCE(order > 1); |
dab48dab | 1512 | } |
1da177e4 | 1513 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1514 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1515 | spin_unlock(&zone->lock); |
1516 | if (!page) | |
1517 | goto failed; | |
d1ce749a BZ |
1518 | __mod_zone_freepage_state(zone, -(1 << order), |
1519 | get_pageblock_migratetype(page)); | |
1da177e4 LT |
1520 | } |
1521 | ||
f8891e5e | 1522 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1523 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1524 | local_irq_restore(flags); |
1da177e4 | 1525 | |
725d704e | 1526 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1527 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1528 | goto again; |
1da177e4 | 1529 | return page; |
a74609fa NP |
1530 | |
1531 | failed: | |
1532 | local_irq_restore(flags); | |
a74609fa | 1533 | return NULL; |
1da177e4 LT |
1534 | } |
1535 | ||
933e312e AM |
1536 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1537 | ||
b2588c4b | 1538 | static struct { |
933e312e AM |
1539 | struct fault_attr attr; |
1540 | ||
1541 | u32 ignore_gfp_highmem; | |
1542 | u32 ignore_gfp_wait; | |
54114994 | 1543 | u32 min_order; |
933e312e AM |
1544 | } fail_page_alloc = { |
1545 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1546 | .ignore_gfp_wait = 1, |
1547 | .ignore_gfp_highmem = 1, | |
54114994 | 1548 | .min_order = 1, |
933e312e AM |
1549 | }; |
1550 | ||
1551 | static int __init setup_fail_page_alloc(char *str) | |
1552 | { | |
1553 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1554 | } | |
1555 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1556 | ||
deaf386e | 1557 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1558 | { |
54114994 | 1559 | if (order < fail_page_alloc.min_order) |
deaf386e | 1560 | return false; |
933e312e | 1561 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 1562 | return false; |
933e312e | 1563 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 1564 | return false; |
933e312e | 1565 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
deaf386e | 1566 | return false; |
933e312e AM |
1567 | |
1568 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1569 | } | |
1570 | ||
1571 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1572 | ||
1573 | static int __init fail_page_alloc_debugfs(void) | |
1574 | { | |
f4ae40a6 | 1575 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1576 | struct dentry *dir; |
933e312e | 1577 | |
dd48c085 AM |
1578 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1579 | &fail_page_alloc.attr); | |
1580 | if (IS_ERR(dir)) | |
1581 | return PTR_ERR(dir); | |
933e312e | 1582 | |
b2588c4b AM |
1583 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1584 | &fail_page_alloc.ignore_gfp_wait)) | |
1585 | goto fail; | |
1586 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1587 | &fail_page_alloc.ignore_gfp_highmem)) | |
1588 | goto fail; | |
1589 | if (!debugfs_create_u32("min-order", mode, dir, | |
1590 | &fail_page_alloc.min_order)) | |
1591 | goto fail; | |
1592 | ||
1593 | return 0; | |
1594 | fail: | |
dd48c085 | 1595 | debugfs_remove_recursive(dir); |
933e312e | 1596 | |
b2588c4b | 1597 | return -ENOMEM; |
933e312e AM |
1598 | } |
1599 | ||
1600 | late_initcall(fail_page_alloc_debugfs); | |
1601 | ||
1602 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1603 | ||
1604 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1605 | ||
deaf386e | 1606 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1607 | { |
deaf386e | 1608 | return false; |
933e312e AM |
1609 | } |
1610 | ||
1611 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1612 | ||
1da177e4 | 1613 | /* |
88f5acf8 | 1614 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1615 | * of the allocation. |
1616 | */ | |
88f5acf8 MG |
1617 | static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
1618 | int classzone_idx, int alloc_flags, long free_pages) | |
1da177e4 LT |
1619 | { |
1620 | /* free_pages my go negative - that's OK */ | |
d23ad423 | 1621 | long min = mark; |
2cfed075 | 1622 | long lowmem_reserve = z->lowmem_reserve[classzone_idx]; |
1da177e4 LT |
1623 | int o; |
1624 | ||
df0a6daa | 1625 | free_pages -= (1 << order) - 1; |
7fb1d9fc | 1626 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1627 | min -= min / 2; |
7fb1d9fc | 1628 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 | 1629 | min -= min / 4; |
d95ea5d1 BZ |
1630 | #ifdef CONFIG_CMA |
1631 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
1632 | if (!(alloc_flags & ALLOC_CMA)) | |
1633 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); | |
1634 | #endif | |
2cfed075 | 1635 | if (free_pages <= min + lowmem_reserve) |
88f5acf8 | 1636 | return false; |
1da177e4 LT |
1637 | for (o = 0; o < order; o++) { |
1638 | /* At the next order, this order's pages become unavailable */ | |
1639 | free_pages -= z->free_area[o].nr_free << o; | |
1640 | ||
1641 | /* Require fewer higher order pages to be free */ | |
1642 | min >>= 1; | |
1643 | ||
1644 | if (free_pages <= min) | |
88f5acf8 | 1645 | return false; |
1da177e4 | 1646 | } |
88f5acf8 MG |
1647 | return true; |
1648 | } | |
1649 | ||
1650 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
1651 | int classzone_idx, int alloc_flags) | |
1652 | { | |
1653 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1654 | zone_page_state(z, NR_FREE_PAGES)); | |
1655 | } | |
1656 | ||
1657 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, | |
1658 | int classzone_idx, int alloc_flags) | |
1659 | { | |
1660 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1661 | ||
1662 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1663 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1664 | ||
1665 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1666 | free_pages); | |
1da177e4 LT |
1667 | } |
1668 | ||
9276b1bc PJ |
1669 | #ifdef CONFIG_NUMA |
1670 | /* | |
1671 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1672 | * skip over zones that are not allowed by the cpuset, or that have | |
1673 | * been recently (in last second) found to be nearly full. See further | |
1674 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1675 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1676 | * |
1677 | * If the zonelist cache is present in the passed in zonelist, then | |
1678 | * returns a pointer to the allowed node mask (either the current | |
4b0ef1fe | 1679 | * tasks mems_allowed, or node_states[N_MEMORY].) |
9276b1bc PJ |
1680 | * |
1681 | * If the zonelist cache is not available for this zonelist, does | |
1682 | * nothing and returns NULL. | |
1683 | * | |
1684 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1685 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1686 | * | |
1687 | * We hold off even calling zlc_setup, until after we've checked the | |
1688 | * first zone in the zonelist, on the theory that most allocations will | |
1689 | * be satisfied from that first zone, so best to examine that zone as | |
1690 | * quickly as we can. | |
1691 | */ | |
1692 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1693 | { | |
1694 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1695 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1696 | ||
1697 | zlc = zonelist->zlcache_ptr; | |
1698 | if (!zlc) | |
1699 | return NULL; | |
1700 | ||
f05111f5 | 1701 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1702 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1703 | zlc->last_full_zap = jiffies; | |
1704 | } | |
1705 | ||
1706 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1707 | &cpuset_current_mems_allowed : | |
4b0ef1fe | 1708 | &node_states[N_MEMORY]; |
9276b1bc PJ |
1709 | return allowednodes; |
1710 | } | |
1711 | ||
1712 | /* | |
1713 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1714 | * if it is worth looking at further for free memory: | |
1715 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1716 | * bit set in the zonelist_cache fullzones BITMAP). | |
1717 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1718 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1719 | * Return true (non-zero) if zone is worth looking at further, or | |
1720 | * else return false (zero) if it is not. | |
1721 | * | |
1722 | * This check -ignores- the distinction between various watermarks, | |
1723 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1724 | * found to be full for any variation of these watermarks, it will | |
1725 | * be considered full for up to one second by all requests, unless | |
1726 | * we are so low on memory on all allowed nodes that we are forced | |
1727 | * into the second scan of the zonelist. | |
1728 | * | |
1729 | * In the second scan we ignore this zonelist cache and exactly | |
1730 | * apply the watermarks to all zones, even it is slower to do so. | |
1731 | * We are low on memory in the second scan, and should leave no stone | |
1732 | * unturned looking for a free page. | |
1733 | */ | |
dd1a239f | 1734 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1735 | nodemask_t *allowednodes) |
1736 | { | |
1737 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1738 | int i; /* index of *z in zonelist zones */ | |
1739 | int n; /* node that zone *z is on */ | |
1740 | ||
1741 | zlc = zonelist->zlcache_ptr; | |
1742 | if (!zlc) | |
1743 | return 1; | |
1744 | ||
dd1a239f | 1745 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1746 | n = zlc->z_to_n[i]; |
1747 | ||
1748 | /* This zone is worth trying if it is allowed but not full */ | |
1749 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1750 | } | |
1751 | ||
1752 | /* | |
1753 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1754 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1755 | * from that zone don't waste time re-examining it. | |
1756 | */ | |
dd1a239f | 1757 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1758 | { |
1759 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1760 | int i; /* index of *z in zonelist zones */ | |
1761 | ||
1762 | zlc = zonelist->zlcache_ptr; | |
1763 | if (!zlc) | |
1764 | return; | |
1765 | ||
dd1a239f | 1766 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1767 | |
1768 | set_bit(i, zlc->fullzones); | |
1769 | } | |
1770 | ||
76d3fbf8 MG |
1771 | /* |
1772 | * clear all zones full, called after direct reclaim makes progress so that | |
1773 | * a zone that was recently full is not skipped over for up to a second | |
1774 | */ | |
1775 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1776 | { | |
1777 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1778 | ||
1779 | zlc = zonelist->zlcache_ptr; | |
1780 | if (!zlc) | |
1781 | return; | |
1782 | ||
1783 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1784 | } | |
1785 | ||
957f822a DR |
1786 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1787 | { | |
1788 | return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes); | |
1789 | } | |
1790 | ||
1791 | static void __paginginit init_zone_allows_reclaim(int nid) | |
1792 | { | |
1793 | int i; | |
1794 | ||
1795 | for_each_online_node(i) | |
6b187d02 | 1796 | if (node_distance(nid, i) <= RECLAIM_DISTANCE) |
957f822a | 1797 | node_set(i, NODE_DATA(nid)->reclaim_nodes); |
6b187d02 | 1798 | else |
957f822a | 1799 | zone_reclaim_mode = 1; |
957f822a DR |
1800 | } |
1801 | ||
9276b1bc PJ |
1802 | #else /* CONFIG_NUMA */ |
1803 | ||
1804 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1805 | { | |
1806 | return NULL; | |
1807 | } | |
1808 | ||
dd1a239f | 1809 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1810 | nodemask_t *allowednodes) |
1811 | { | |
1812 | return 1; | |
1813 | } | |
1814 | ||
dd1a239f | 1815 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1816 | { |
1817 | } | |
76d3fbf8 MG |
1818 | |
1819 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1820 | { | |
1821 | } | |
957f822a DR |
1822 | |
1823 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) | |
1824 | { | |
1825 | return true; | |
1826 | } | |
1827 | ||
1828 | static inline void init_zone_allows_reclaim(int nid) | |
1829 | { | |
1830 | } | |
9276b1bc PJ |
1831 | #endif /* CONFIG_NUMA */ |
1832 | ||
7fb1d9fc | 1833 | /* |
0798e519 | 1834 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1835 | * a page. |
1836 | */ | |
1837 | static struct page * | |
19770b32 | 1838 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1839 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1840 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1841 | { |
dd1a239f | 1842 | struct zoneref *z; |
7fb1d9fc | 1843 | struct page *page = NULL; |
54a6eb5c | 1844 | int classzone_idx; |
5117f45d | 1845 | struct zone *zone; |
9276b1bc PJ |
1846 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1847 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1848 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1849 | |
19770b32 | 1850 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1851 | zonelist_scan: |
7fb1d9fc | 1852 | /* |
9276b1bc | 1853 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1854 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1855 | */ | |
19770b32 MG |
1856 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1857 | high_zoneidx, nodemask) { | |
e5adfffc | 1858 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
9276b1bc PJ |
1859 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1860 | continue; | |
7fb1d9fc | 1861 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1862 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
cd38b115 | 1863 | continue; |
a756cf59 JW |
1864 | /* |
1865 | * When allocating a page cache page for writing, we | |
1866 | * want to get it from a zone that is within its dirty | |
1867 | * limit, such that no single zone holds more than its | |
1868 | * proportional share of globally allowed dirty pages. | |
1869 | * The dirty limits take into account the zone's | |
1870 | * lowmem reserves and high watermark so that kswapd | |
1871 | * should be able to balance it without having to | |
1872 | * write pages from its LRU list. | |
1873 | * | |
1874 | * This may look like it could increase pressure on | |
1875 | * lower zones by failing allocations in higher zones | |
1876 | * before they are full. But the pages that do spill | |
1877 | * over are limited as the lower zones are protected | |
1878 | * by this very same mechanism. It should not become | |
1879 | * a practical burden to them. | |
1880 | * | |
1881 | * XXX: For now, allow allocations to potentially | |
1882 | * exceed the per-zone dirty limit in the slowpath | |
1883 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
1884 | * which is important when on a NUMA setup the allowed | |
1885 | * zones are together not big enough to reach the | |
1886 | * global limit. The proper fix for these situations | |
1887 | * will require awareness of zones in the | |
1888 | * dirty-throttling and the flusher threads. | |
1889 | */ | |
1890 | if ((alloc_flags & ALLOC_WMARK_LOW) && | |
1891 | (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone)) | |
1892 | goto this_zone_full; | |
7fb1d9fc | 1893 | |
41858966 | 1894 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
7fb1d9fc | 1895 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
3148890b | 1896 | unsigned long mark; |
fa5e084e MG |
1897 | int ret; |
1898 | ||
41858966 | 1899 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
fa5e084e MG |
1900 | if (zone_watermark_ok(zone, order, mark, |
1901 | classzone_idx, alloc_flags)) | |
1902 | goto try_this_zone; | |
1903 | ||
e5adfffc KS |
1904 | if (IS_ENABLED(CONFIG_NUMA) && |
1905 | !did_zlc_setup && nr_online_nodes > 1) { | |
cd38b115 MG |
1906 | /* |
1907 | * we do zlc_setup if there are multiple nodes | |
1908 | * and before considering the first zone allowed | |
1909 | * by the cpuset. | |
1910 | */ | |
1911 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1912 | zlc_active = 1; | |
1913 | did_zlc_setup = 1; | |
1914 | } | |
1915 | ||
957f822a DR |
1916 | if (zone_reclaim_mode == 0 || |
1917 | !zone_allows_reclaim(preferred_zone, zone)) | |
fa5e084e MG |
1918 | goto this_zone_full; |
1919 | ||
cd38b115 MG |
1920 | /* |
1921 | * As we may have just activated ZLC, check if the first | |
1922 | * eligible zone has failed zone_reclaim recently. | |
1923 | */ | |
e5adfffc | 1924 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
cd38b115 MG |
1925 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1926 | continue; | |
1927 | ||
fa5e084e MG |
1928 | ret = zone_reclaim(zone, gfp_mask, order); |
1929 | switch (ret) { | |
1930 | case ZONE_RECLAIM_NOSCAN: | |
1931 | /* did not scan */ | |
cd38b115 | 1932 | continue; |
fa5e084e MG |
1933 | case ZONE_RECLAIM_FULL: |
1934 | /* scanned but unreclaimable */ | |
cd38b115 | 1935 | continue; |
fa5e084e MG |
1936 | default: |
1937 | /* did we reclaim enough */ | |
1938 | if (!zone_watermark_ok(zone, order, mark, | |
1939 | classzone_idx, alloc_flags)) | |
9276b1bc | 1940 | goto this_zone_full; |
0798e519 | 1941 | } |
7fb1d9fc RS |
1942 | } |
1943 | ||
fa5e084e | 1944 | try_this_zone: |
3dd28266 MG |
1945 | page = buffered_rmqueue(preferred_zone, zone, order, |
1946 | gfp_mask, migratetype); | |
0798e519 | 1947 | if (page) |
7fb1d9fc | 1948 | break; |
9276b1bc | 1949 | this_zone_full: |
e5adfffc | 1950 | if (IS_ENABLED(CONFIG_NUMA)) |
9276b1bc | 1951 | zlc_mark_zone_full(zonelist, z); |
54a6eb5c | 1952 | } |
9276b1bc | 1953 | |
e5adfffc | 1954 | if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) { |
9276b1bc PJ |
1955 | /* Disable zlc cache for second zonelist scan */ |
1956 | zlc_active = 0; | |
1957 | goto zonelist_scan; | |
1958 | } | |
b121186a AS |
1959 | |
1960 | if (page) | |
1961 | /* | |
1962 | * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was | |
1963 | * necessary to allocate the page. The expectation is | |
1964 | * that the caller is taking steps that will free more | |
1965 | * memory. The caller should avoid the page being used | |
1966 | * for !PFMEMALLOC purposes. | |
1967 | */ | |
1968 | page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS); | |
1969 | ||
7fb1d9fc | 1970 | return page; |
753ee728 MH |
1971 | } |
1972 | ||
29423e77 DR |
1973 | /* |
1974 | * Large machines with many possible nodes should not always dump per-node | |
1975 | * meminfo in irq context. | |
1976 | */ | |
1977 | static inline bool should_suppress_show_mem(void) | |
1978 | { | |
1979 | bool ret = false; | |
1980 | ||
1981 | #if NODES_SHIFT > 8 | |
1982 | ret = in_interrupt(); | |
1983 | #endif | |
1984 | return ret; | |
1985 | } | |
1986 | ||
a238ab5b DH |
1987 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
1988 | DEFAULT_RATELIMIT_INTERVAL, | |
1989 | DEFAULT_RATELIMIT_BURST); | |
1990 | ||
1991 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
1992 | { | |
a238ab5b DH |
1993 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
1994 | ||
c0a32fc5 SG |
1995 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
1996 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
1997 | return; |
1998 | ||
1999 | /* | |
2000 | * This documents exceptions given to allocations in certain | |
2001 | * contexts that are allowed to allocate outside current's set | |
2002 | * of allowed nodes. | |
2003 | */ | |
2004 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2005 | if (test_thread_flag(TIF_MEMDIE) || | |
2006 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2007 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2008 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
2009 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2010 | ||
2011 | if (fmt) { | |
3ee9a4f0 JP |
2012 | struct va_format vaf; |
2013 | va_list args; | |
2014 | ||
a238ab5b | 2015 | va_start(args, fmt); |
3ee9a4f0 JP |
2016 | |
2017 | vaf.fmt = fmt; | |
2018 | vaf.va = &args; | |
2019 | ||
2020 | pr_warn("%pV", &vaf); | |
2021 | ||
a238ab5b DH |
2022 | va_end(args); |
2023 | } | |
2024 | ||
3ee9a4f0 JP |
2025 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
2026 | current->comm, order, gfp_mask); | |
a238ab5b DH |
2027 | |
2028 | dump_stack(); | |
2029 | if (!should_suppress_show_mem()) | |
2030 | show_mem(filter); | |
2031 | } | |
2032 | ||
11e33f6a MG |
2033 | static inline int |
2034 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
f90ac398 | 2035 | unsigned long did_some_progress, |
11e33f6a | 2036 | unsigned long pages_reclaimed) |
1da177e4 | 2037 | { |
11e33f6a MG |
2038 | /* Do not loop if specifically requested */ |
2039 | if (gfp_mask & __GFP_NORETRY) | |
2040 | return 0; | |
1da177e4 | 2041 | |
f90ac398 MG |
2042 | /* Always retry if specifically requested */ |
2043 | if (gfp_mask & __GFP_NOFAIL) | |
2044 | return 1; | |
2045 | ||
2046 | /* | |
2047 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim | |
2048 | * making forward progress without invoking OOM. Suspend also disables | |
2049 | * storage devices so kswapd will not help. Bail if we are suspending. | |
2050 | */ | |
2051 | if (!did_some_progress && pm_suspended_storage()) | |
2052 | return 0; | |
2053 | ||
11e33f6a MG |
2054 | /* |
2055 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
2056 | * means __GFP_NOFAIL, but that may not be true in other | |
2057 | * implementations. | |
2058 | */ | |
2059 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
2060 | return 1; | |
2061 | ||
2062 | /* | |
2063 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
2064 | * specified, then we retry until we no longer reclaim any pages | |
2065 | * (above), or we've reclaimed an order of pages at least as | |
2066 | * large as the allocation's order. In both cases, if the | |
2067 | * allocation still fails, we stop retrying. | |
2068 | */ | |
2069 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
2070 | return 1; | |
cf40bd16 | 2071 | |
11e33f6a MG |
2072 | return 0; |
2073 | } | |
933e312e | 2074 | |
11e33f6a MG |
2075 | static inline struct page * |
2076 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
2077 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2078 | nodemask_t *nodemask, struct zone *preferred_zone, |
2079 | int migratetype) | |
11e33f6a MG |
2080 | { |
2081 | struct page *page; | |
2082 | ||
2083 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
ff321fea | 2084 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
11e33f6a | 2085 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2086 | return NULL; |
2087 | } | |
6b1de916 | 2088 | |
11e33f6a MG |
2089 | /* |
2090 | * Go through the zonelist yet one more time, keep very high watermark | |
2091 | * here, this is only to catch a parallel oom killing, we must fail if | |
2092 | * we're still under heavy pressure. | |
2093 | */ | |
2094 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
2095 | order, zonelist, high_zoneidx, | |
5117f45d | 2096 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 2097 | preferred_zone, migratetype); |
7fb1d9fc | 2098 | if (page) |
11e33f6a MG |
2099 | goto out; |
2100 | ||
4365a567 KH |
2101 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2102 | /* The OOM killer will not help higher order allocs */ | |
2103 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2104 | goto out; | |
03668b3c DR |
2105 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
2106 | if (high_zoneidx < ZONE_NORMAL) | |
2107 | goto out; | |
4365a567 KH |
2108 | /* |
2109 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
2110 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
2111 | * The caller should handle page allocation failure by itself if | |
2112 | * it specifies __GFP_THISNODE. | |
2113 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
2114 | */ | |
2115 | if (gfp_mask & __GFP_THISNODE) | |
2116 | goto out; | |
2117 | } | |
11e33f6a | 2118 | /* Exhausted what can be done so it's blamo time */ |
08ab9b10 | 2119 | out_of_memory(zonelist, gfp_mask, order, nodemask, false); |
11e33f6a MG |
2120 | |
2121 | out: | |
2122 | clear_zonelist_oom(zonelist, gfp_mask); | |
2123 | return page; | |
2124 | } | |
2125 | ||
56de7263 MG |
2126 | #ifdef CONFIG_COMPACTION |
2127 | /* Try memory compaction for high-order allocations before reclaim */ | |
2128 | static struct page * | |
2129 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2130 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2131 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
66199712 | 2132 | int migratetype, bool sync_migration, |
c67fe375 | 2133 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2134 | unsigned long *did_some_progress) |
56de7263 | 2135 | { |
66199712 | 2136 | if (!order) |
56de7263 MG |
2137 | return NULL; |
2138 | ||
aff62249 | 2139 | if (compaction_deferred(preferred_zone, order)) { |
66199712 MG |
2140 | *deferred_compaction = true; |
2141 | return NULL; | |
2142 | } | |
2143 | ||
c06b1fca | 2144 | current->flags |= PF_MEMALLOC; |
56de7263 | 2145 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
c67fe375 | 2146 | nodemask, sync_migration, |
8fb74b9f | 2147 | contended_compaction); |
c06b1fca | 2148 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2149 | |
1fb3f8ca | 2150 | if (*did_some_progress != COMPACT_SKIPPED) { |
8fb74b9f MG |
2151 | struct page *page; |
2152 | ||
56de7263 MG |
2153 | /* Page migration frees to the PCP lists but we want merging */ |
2154 | drain_pages(get_cpu()); | |
2155 | put_cpu(); | |
2156 | ||
2157 | page = get_page_from_freelist(gfp_mask, nodemask, | |
2158 | order, zonelist, high_zoneidx, | |
cfd19c5a MG |
2159 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2160 | preferred_zone, migratetype); | |
56de7263 | 2161 | if (page) { |
62997027 | 2162 | preferred_zone->compact_blockskip_flush = false; |
4f92e258 MG |
2163 | preferred_zone->compact_considered = 0; |
2164 | preferred_zone->compact_defer_shift = 0; | |
aff62249 RR |
2165 | if (order >= preferred_zone->compact_order_failed) |
2166 | preferred_zone->compact_order_failed = order + 1; | |
56de7263 MG |
2167 | count_vm_event(COMPACTSUCCESS); |
2168 | return page; | |
2169 | } | |
2170 | ||
2171 | /* | |
2172 | * It's bad if compaction run occurs and fails. | |
2173 | * The most likely reason is that pages exist, | |
2174 | * but not enough to satisfy watermarks. | |
2175 | */ | |
2176 | count_vm_event(COMPACTFAIL); | |
66199712 MG |
2177 | |
2178 | /* | |
2179 | * As async compaction considers a subset of pageblocks, only | |
2180 | * defer if the failure was a sync compaction failure. | |
2181 | */ | |
2182 | if (sync_migration) | |
aff62249 | 2183 | defer_compaction(preferred_zone, order); |
56de7263 MG |
2184 | |
2185 | cond_resched(); | |
2186 | } | |
2187 | ||
2188 | return NULL; | |
2189 | } | |
2190 | #else | |
2191 | static inline struct page * | |
2192 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2193 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2194 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
66199712 | 2195 | int migratetype, bool sync_migration, |
c67fe375 | 2196 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2197 | unsigned long *did_some_progress) |
56de7263 MG |
2198 | { |
2199 | return NULL; | |
2200 | } | |
2201 | #endif /* CONFIG_COMPACTION */ | |
2202 | ||
bba90710 MS |
2203 | /* Perform direct synchronous page reclaim */ |
2204 | static int | |
2205 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, | |
2206 | nodemask_t *nodemask) | |
11e33f6a | 2207 | { |
11e33f6a | 2208 | struct reclaim_state reclaim_state; |
bba90710 | 2209 | int progress; |
11e33f6a MG |
2210 | |
2211 | cond_resched(); | |
2212 | ||
2213 | /* We now go into synchronous reclaim */ | |
2214 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2215 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2216 | lockdep_set_current_reclaim_state(gfp_mask); |
2217 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2218 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2219 | |
bba90710 | 2220 | progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
11e33f6a | 2221 | |
c06b1fca | 2222 | current->reclaim_state = NULL; |
11e33f6a | 2223 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2224 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2225 | |
2226 | cond_resched(); | |
2227 | ||
bba90710 MS |
2228 | return progress; |
2229 | } | |
2230 | ||
2231 | /* The really slow allocator path where we enter direct reclaim */ | |
2232 | static inline struct page * | |
2233 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
2234 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2235 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
2236 | int migratetype, unsigned long *did_some_progress) | |
2237 | { | |
2238 | struct page *page = NULL; | |
2239 | bool drained = false; | |
2240 | ||
2241 | *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist, | |
2242 | nodemask); | |
9ee493ce MG |
2243 | if (unlikely(!(*did_some_progress))) |
2244 | return NULL; | |
11e33f6a | 2245 | |
76d3fbf8 | 2246 | /* After successful reclaim, reconsider all zones for allocation */ |
e5adfffc | 2247 | if (IS_ENABLED(CONFIG_NUMA)) |
76d3fbf8 MG |
2248 | zlc_clear_zones_full(zonelist); |
2249 | ||
9ee493ce MG |
2250 | retry: |
2251 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2252 | zonelist, high_zoneidx, |
cfd19c5a MG |
2253 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2254 | preferred_zone, migratetype); | |
9ee493ce MG |
2255 | |
2256 | /* | |
2257 | * If an allocation failed after direct reclaim, it could be because | |
2258 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2259 | */ | |
2260 | if (!page && !drained) { | |
2261 | drain_all_pages(); | |
2262 | drained = true; | |
2263 | goto retry; | |
2264 | } | |
2265 | ||
11e33f6a MG |
2266 | return page; |
2267 | } | |
2268 | ||
1da177e4 | 2269 | /* |
11e33f6a MG |
2270 | * This is called in the allocator slow-path if the allocation request is of |
2271 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2272 | */ |
11e33f6a MG |
2273 | static inline struct page * |
2274 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
2275 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2276 | nodemask_t *nodemask, struct zone *preferred_zone, |
2277 | int migratetype) | |
11e33f6a MG |
2278 | { |
2279 | struct page *page; | |
2280 | ||
2281 | do { | |
2282 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2283 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 2284 | preferred_zone, migratetype); |
11e33f6a MG |
2285 | |
2286 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 2287 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
2288 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2289 | ||
2290 | return page; | |
2291 | } | |
2292 | ||
2293 | static inline | |
2294 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
99504748 MG |
2295 | enum zone_type high_zoneidx, |
2296 | enum zone_type classzone_idx) | |
1da177e4 | 2297 | { |
dd1a239f MG |
2298 | struct zoneref *z; |
2299 | struct zone *zone; | |
1da177e4 | 2300 | |
11e33f6a | 2301 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
99504748 | 2302 | wakeup_kswapd(zone, order, classzone_idx); |
11e33f6a | 2303 | } |
cf40bd16 | 2304 | |
341ce06f PZ |
2305 | static inline int |
2306 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2307 | { | |
341ce06f PZ |
2308 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
2309 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 2310 | |
a56f57ff | 2311 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2312 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2313 | |
341ce06f PZ |
2314 | /* |
2315 | * The caller may dip into page reserves a bit more if the caller | |
2316 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2317 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
2318 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
2319 | */ | |
e6223a3b | 2320 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2321 | |
341ce06f | 2322 | if (!wait) { |
5c3240d9 AA |
2323 | /* |
2324 | * Not worth trying to allocate harder for | |
2325 | * __GFP_NOMEMALLOC even if it can't schedule. | |
2326 | */ | |
2327 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2328 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 2329 | /* |
341ce06f PZ |
2330 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
2331 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 2332 | */ |
341ce06f | 2333 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2334 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2335 | alloc_flags |= ALLOC_HARDER; |
2336 | ||
b37f1dd0 MG |
2337 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
2338 | if (gfp_mask & __GFP_MEMALLOC) | |
2339 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
2340 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
2341 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
2342 | else if (!in_interrupt() && | |
2343 | ((current->flags & PF_MEMALLOC) || | |
2344 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 2345 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 2346 | } |
d95ea5d1 BZ |
2347 | #ifdef CONFIG_CMA |
2348 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2349 | alloc_flags |= ALLOC_CMA; | |
2350 | #endif | |
341ce06f PZ |
2351 | return alloc_flags; |
2352 | } | |
2353 | ||
072bb0aa MG |
2354 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
2355 | { | |
b37f1dd0 | 2356 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
2357 | } |
2358 | ||
11e33f6a MG |
2359 | static inline struct page * |
2360 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2361 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2362 | nodemask_t *nodemask, struct zone *preferred_zone, |
2363 | int migratetype) | |
11e33f6a MG |
2364 | { |
2365 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2366 | struct page *page = NULL; | |
2367 | int alloc_flags; | |
2368 | unsigned long pages_reclaimed = 0; | |
2369 | unsigned long did_some_progress; | |
77f1fe6b | 2370 | bool sync_migration = false; |
66199712 | 2371 | bool deferred_compaction = false; |
c67fe375 | 2372 | bool contended_compaction = false; |
1da177e4 | 2373 | |
72807a74 MG |
2374 | /* |
2375 | * In the slowpath, we sanity check order to avoid ever trying to | |
2376 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2377 | * be using allocators in order of preference for an area that is | |
2378 | * too large. | |
2379 | */ | |
1fc28b70 MG |
2380 | if (order >= MAX_ORDER) { |
2381 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2382 | return NULL; |
1fc28b70 | 2383 | } |
1da177e4 | 2384 | |
952f3b51 CL |
2385 | /* |
2386 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2387 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2388 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2389 | * using a larger set of nodes after it has established that the | |
2390 | * allowed per node queues are empty and that nodes are | |
2391 | * over allocated. | |
2392 | */ | |
e5adfffc KS |
2393 | if (IS_ENABLED(CONFIG_NUMA) && |
2394 | (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
952f3b51 CL |
2395 | goto nopage; |
2396 | ||
cc4a6851 | 2397 | restart: |
caf49191 LT |
2398 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
2399 | wake_all_kswapd(order, zonelist, high_zoneidx, | |
2400 | zone_idx(preferred_zone)); | |
1da177e4 | 2401 | |
9bf2229f | 2402 | /* |
7fb1d9fc RS |
2403 | * OK, we're below the kswapd watermark and have kicked background |
2404 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2405 | * to how we want to proceed. | |
9bf2229f | 2406 | */ |
341ce06f | 2407 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2408 | |
f33261d7 DR |
2409 | /* |
2410 | * Find the true preferred zone if the allocation is unconstrained by | |
2411 | * cpusets. | |
2412 | */ | |
2413 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) | |
2414 | first_zones_zonelist(zonelist, high_zoneidx, NULL, | |
2415 | &preferred_zone); | |
2416 | ||
cfa54a0f | 2417 | rebalance: |
341ce06f | 2418 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2419 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
2420 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
2421 | preferred_zone, migratetype); | |
7fb1d9fc RS |
2422 | if (page) |
2423 | goto got_pg; | |
1da177e4 | 2424 | |
11e33f6a | 2425 | /* Allocate without watermarks if the context allows */ |
341ce06f | 2426 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
2427 | /* |
2428 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
2429 | * the allocation is high priority and these type of | |
2430 | * allocations are system rather than user orientated | |
2431 | */ | |
2432 | zonelist = node_zonelist(numa_node_id(), gfp_mask); | |
2433 | ||
341ce06f PZ |
2434 | page = __alloc_pages_high_priority(gfp_mask, order, |
2435 | zonelist, high_zoneidx, nodemask, | |
2436 | preferred_zone, migratetype); | |
cfd19c5a | 2437 | if (page) { |
341ce06f | 2438 | goto got_pg; |
cfd19c5a | 2439 | } |
1da177e4 LT |
2440 | } |
2441 | ||
2442 | /* Atomic allocations - we can't balance anything */ | |
2443 | if (!wait) | |
2444 | goto nopage; | |
2445 | ||
341ce06f | 2446 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2447 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2448 | goto nopage; |
2449 | ||
6583bb64 DR |
2450 | /* Avoid allocations with no watermarks from looping endlessly */ |
2451 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2452 | goto nopage; | |
2453 | ||
77f1fe6b MG |
2454 | /* |
2455 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2456 | * attempts after direct reclaim are synchronous | |
2457 | */ | |
56de7263 MG |
2458 | page = __alloc_pages_direct_compact(gfp_mask, order, |
2459 | zonelist, high_zoneidx, | |
2460 | nodemask, | |
2461 | alloc_flags, preferred_zone, | |
66199712 | 2462 | migratetype, sync_migration, |
c67fe375 | 2463 | &contended_compaction, |
66199712 MG |
2464 | &deferred_compaction, |
2465 | &did_some_progress); | |
56de7263 MG |
2466 | if (page) |
2467 | goto got_pg; | |
c6a140bf | 2468 | sync_migration = true; |
56de7263 | 2469 | |
31f8d42d LT |
2470 | /* |
2471 | * If compaction is deferred for high-order allocations, it is because | |
2472 | * sync compaction recently failed. In this is the case and the caller | |
2473 | * requested a movable allocation that does not heavily disrupt the | |
2474 | * system then fail the allocation instead of entering direct reclaim. | |
2475 | */ | |
2476 | if ((deferred_compaction || contended_compaction) && | |
caf49191 | 2477 | (gfp_mask & __GFP_NO_KSWAPD)) |
31f8d42d | 2478 | goto nopage; |
66199712 | 2479 | |
11e33f6a MG |
2480 | /* Try direct reclaim and then allocating */ |
2481 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2482 | zonelist, high_zoneidx, | |
2483 | nodemask, | |
5117f45d | 2484 | alloc_flags, preferred_zone, |
3dd28266 | 2485 | migratetype, &did_some_progress); |
11e33f6a MG |
2486 | if (page) |
2487 | goto got_pg; | |
1da177e4 | 2488 | |
e33c3b5e | 2489 | /* |
11e33f6a MG |
2490 | * If we failed to make any progress reclaiming, then we are |
2491 | * running out of options and have to consider going OOM | |
e33c3b5e | 2492 | */ |
11e33f6a MG |
2493 | if (!did_some_progress) { |
2494 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
7f33d49a RW |
2495 | if (oom_killer_disabled) |
2496 | goto nopage; | |
29fd66d2 DR |
2497 | /* Coredumps can quickly deplete all memory reserves */ |
2498 | if ((current->flags & PF_DUMPCORE) && | |
2499 | !(gfp_mask & __GFP_NOFAIL)) | |
2500 | goto nopage; | |
11e33f6a MG |
2501 | page = __alloc_pages_may_oom(gfp_mask, order, |
2502 | zonelist, high_zoneidx, | |
3dd28266 MG |
2503 | nodemask, preferred_zone, |
2504 | migratetype); | |
11e33f6a MG |
2505 | if (page) |
2506 | goto got_pg; | |
1da177e4 | 2507 | |
03668b3c DR |
2508 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2509 | /* | |
2510 | * The oom killer is not called for high-order | |
2511 | * allocations that may fail, so if no progress | |
2512 | * is being made, there are no other options and | |
2513 | * retrying is unlikely to help. | |
2514 | */ | |
2515 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2516 | goto nopage; | |
2517 | /* | |
2518 | * The oom killer is not called for lowmem | |
2519 | * allocations to prevent needlessly killing | |
2520 | * innocent tasks. | |
2521 | */ | |
2522 | if (high_zoneidx < ZONE_NORMAL) | |
2523 | goto nopage; | |
2524 | } | |
e2c55dc8 | 2525 | |
ff0ceb9d DR |
2526 | goto restart; |
2527 | } | |
1da177e4 LT |
2528 | } |
2529 | ||
11e33f6a | 2530 | /* Check if we should retry the allocation */ |
a41f24ea | 2531 | pages_reclaimed += did_some_progress; |
f90ac398 MG |
2532 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
2533 | pages_reclaimed)) { | |
11e33f6a | 2534 | /* Wait for some write requests to complete then retry */ |
0e093d99 | 2535 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2536 | goto rebalance; |
3e7d3449 MG |
2537 | } else { |
2538 | /* | |
2539 | * High-order allocations do not necessarily loop after | |
2540 | * direct reclaim and reclaim/compaction depends on compaction | |
2541 | * being called after reclaim so call directly if necessary | |
2542 | */ | |
2543 | page = __alloc_pages_direct_compact(gfp_mask, order, | |
2544 | zonelist, high_zoneidx, | |
2545 | nodemask, | |
2546 | alloc_flags, preferred_zone, | |
66199712 | 2547 | migratetype, sync_migration, |
c67fe375 | 2548 | &contended_compaction, |
66199712 MG |
2549 | &deferred_compaction, |
2550 | &did_some_progress); | |
3e7d3449 MG |
2551 | if (page) |
2552 | goto got_pg; | |
1da177e4 LT |
2553 | } |
2554 | ||
2555 | nopage: | |
a238ab5b | 2556 | warn_alloc_failed(gfp_mask, order, NULL); |
b1eeab67 | 2557 | return page; |
1da177e4 | 2558 | got_pg: |
b1eeab67 VN |
2559 | if (kmemcheck_enabled) |
2560 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
11e33f6a | 2561 | |
072bb0aa | 2562 | return page; |
1da177e4 | 2563 | } |
11e33f6a MG |
2564 | |
2565 | /* | |
2566 | * This is the 'heart' of the zoned buddy allocator. | |
2567 | */ | |
2568 | struct page * | |
2569 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2570 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2571 | { | |
2572 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2573 | struct zone *preferred_zone; |
cc9a6c87 | 2574 | struct page *page = NULL; |
3dd28266 | 2575 | int migratetype = allocflags_to_migratetype(gfp_mask); |
cc9a6c87 | 2576 | unsigned int cpuset_mems_cookie; |
d95ea5d1 | 2577 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET; |
6a1a0d3b | 2578 | struct mem_cgroup *memcg = NULL; |
11e33f6a | 2579 | |
dcce284a BH |
2580 | gfp_mask &= gfp_allowed_mask; |
2581 | ||
11e33f6a MG |
2582 | lockdep_trace_alloc(gfp_mask); |
2583 | ||
2584 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2585 | ||
2586 | if (should_fail_alloc_page(gfp_mask, order)) | |
2587 | return NULL; | |
2588 | ||
2589 | /* | |
2590 | * Check the zones suitable for the gfp_mask contain at least one | |
2591 | * valid zone. It's possible to have an empty zonelist as a result | |
2592 | * of GFP_THISNODE and a memoryless node | |
2593 | */ | |
2594 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2595 | return NULL; | |
2596 | ||
6a1a0d3b GC |
2597 | /* |
2598 | * Will only have any effect when __GFP_KMEMCG is set. This is | |
2599 | * verified in the (always inline) callee | |
2600 | */ | |
2601 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
2602 | return NULL; | |
2603 | ||
cc9a6c87 MG |
2604 | retry_cpuset: |
2605 | cpuset_mems_cookie = get_mems_allowed(); | |
2606 | ||
5117f45d | 2607 | /* The preferred zone is used for statistics later */ |
f33261d7 DR |
2608 | first_zones_zonelist(zonelist, high_zoneidx, |
2609 | nodemask ? : &cpuset_current_mems_allowed, | |
2610 | &preferred_zone); | |
cc9a6c87 MG |
2611 | if (!preferred_zone) |
2612 | goto out; | |
5117f45d | 2613 | |
d95ea5d1 BZ |
2614 | #ifdef CONFIG_CMA |
2615 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2616 | alloc_flags |= ALLOC_CMA; | |
2617 | #endif | |
5117f45d | 2618 | /* First allocation attempt */ |
11e33f6a | 2619 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
d95ea5d1 | 2620 | zonelist, high_zoneidx, alloc_flags, |
3dd28266 | 2621 | preferred_zone, migratetype); |
11e33f6a MG |
2622 | if (unlikely(!page)) |
2623 | page = __alloc_pages_slowpath(gfp_mask, order, | |
5117f45d | 2624 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 2625 | preferred_zone, migratetype); |
11e33f6a | 2626 | |
4b4f278c | 2627 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
cc9a6c87 MG |
2628 | |
2629 | out: | |
2630 | /* | |
2631 | * When updating a task's mems_allowed, it is possible to race with | |
2632 | * parallel threads in such a way that an allocation can fail while | |
2633 | * the mask is being updated. If a page allocation is about to fail, | |
2634 | * check if the cpuset changed during allocation and if so, retry. | |
2635 | */ | |
2636 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) | |
2637 | goto retry_cpuset; | |
2638 | ||
6a1a0d3b GC |
2639 | memcg_kmem_commit_charge(page, memcg, order); |
2640 | ||
11e33f6a | 2641 | return page; |
1da177e4 | 2642 | } |
d239171e | 2643 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2644 | |
2645 | /* | |
2646 | * Common helper functions. | |
2647 | */ | |
920c7a5d | 2648 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2649 | { |
945a1113 AM |
2650 | struct page *page; |
2651 | ||
2652 | /* | |
2653 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2654 | * a highmem page | |
2655 | */ | |
2656 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2657 | ||
1da177e4 LT |
2658 | page = alloc_pages(gfp_mask, order); |
2659 | if (!page) | |
2660 | return 0; | |
2661 | return (unsigned long) page_address(page); | |
2662 | } | |
1da177e4 LT |
2663 | EXPORT_SYMBOL(__get_free_pages); |
2664 | ||
920c7a5d | 2665 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2666 | { |
945a1113 | 2667 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2668 | } |
1da177e4 LT |
2669 | EXPORT_SYMBOL(get_zeroed_page); |
2670 | ||
920c7a5d | 2671 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2672 | { |
b5810039 | 2673 | if (put_page_testzero(page)) { |
1da177e4 | 2674 | if (order == 0) |
fc91668e | 2675 | free_hot_cold_page(page, 0); |
1da177e4 LT |
2676 | else |
2677 | __free_pages_ok(page, order); | |
2678 | } | |
2679 | } | |
2680 | ||
2681 | EXPORT_SYMBOL(__free_pages); | |
2682 | ||
920c7a5d | 2683 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2684 | { |
2685 | if (addr != 0) { | |
725d704e | 2686 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2687 | __free_pages(virt_to_page((void *)addr), order); |
2688 | } | |
2689 | } | |
2690 | ||
2691 | EXPORT_SYMBOL(free_pages); | |
2692 | ||
6a1a0d3b GC |
2693 | /* |
2694 | * __free_memcg_kmem_pages and free_memcg_kmem_pages will free | |
2695 | * pages allocated with __GFP_KMEMCG. | |
2696 | * | |
2697 | * Those pages are accounted to a particular memcg, embedded in the | |
2698 | * corresponding page_cgroup. To avoid adding a hit in the allocator to search | |
2699 | * for that information only to find out that it is NULL for users who have no | |
2700 | * interest in that whatsoever, we provide these functions. | |
2701 | * | |
2702 | * The caller knows better which flags it relies on. | |
2703 | */ | |
2704 | void __free_memcg_kmem_pages(struct page *page, unsigned int order) | |
2705 | { | |
2706 | memcg_kmem_uncharge_pages(page, order); | |
2707 | __free_pages(page, order); | |
2708 | } | |
2709 | ||
2710 | void free_memcg_kmem_pages(unsigned long addr, unsigned int order) | |
2711 | { | |
2712 | if (addr != 0) { | |
2713 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
2714 | __free_memcg_kmem_pages(virt_to_page((void *)addr), order); | |
2715 | } | |
2716 | } | |
2717 | ||
ee85c2e1 AK |
2718 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
2719 | { | |
2720 | if (addr) { | |
2721 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2722 | unsigned long used = addr + PAGE_ALIGN(size); | |
2723 | ||
2724 | split_page(virt_to_page((void *)addr), order); | |
2725 | while (used < alloc_end) { | |
2726 | free_page(used); | |
2727 | used += PAGE_SIZE; | |
2728 | } | |
2729 | } | |
2730 | return (void *)addr; | |
2731 | } | |
2732 | ||
2be0ffe2 TT |
2733 | /** |
2734 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2735 | * @size: the number of bytes to allocate | |
2736 | * @gfp_mask: GFP flags for the allocation | |
2737 | * | |
2738 | * This function is similar to alloc_pages(), except that it allocates the | |
2739 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2740 | * allocate memory in power-of-two pages. | |
2741 | * | |
2742 | * This function is also limited by MAX_ORDER. | |
2743 | * | |
2744 | * Memory allocated by this function must be released by free_pages_exact(). | |
2745 | */ | |
2746 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2747 | { | |
2748 | unsigned int order = get_order(size); | |
2749 | unsigned long addr; | |
2750 | ||
2751 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 2752 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
2753 | } |
2754 | EXPORT_SYMBOL(alloc_pages_exact); | |
2755 | ||
ee85c2e1 AK |
2756 | /** |
2757 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
2758 | * pages on a node. | |
b5e6ab58 | 2759 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
2760 | * @size: the number of bytes to allocate |
2761 | * @gfp_mask: GFP flags for the allocation | |
2762 | * | |
2763 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
2764 | * back. | |
2765 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
2766 | * but is not exact. | |
2767 | */ | |
2768 | void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) | |
2769 | { | |
2770 | unsigned order = get_order(size); | |
2771 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
2772 | if (!p) | |
2773 | return NULL; | |
2774 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
2775 | } | |
2776 | EXPORT_SYMBOL(alloc_pages_exact_nid); | |
2777 | ||
2be0ffe2 TT |
2778 | /** |
2779 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2780 | * @virt: the value returned by alloc_pages_exact. | |
2781 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2782 | * | |
2783 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2784 | */ | |
2785 | void free_pages_exact(void *virt, size_t size) | |
2786 | { | |
2787 | unsigned long addr = (unsigned long)virt; | |
2788 | unsigned long end = addr + PAGE_ALIGN(size); | |
2789 | ||
2790 | while (addr < end) { | |
2791 | free_page(addr); | |
2792 | addr += PAGE_SIZE; | |
2793 | } | |
2794 | } | |
2795 | EXPORT_SYMBOL(free_pages_exact); | |
2796 | ||
1da177e4 LT |
2797 | static unsigned int nr_free_zone_pages(int offset) |
2798 | { | |
dd1a239f | 2799 | struct zoneref *z; |
54a6eb5c MG |
2800 | struct zone *zone; |
2801 | ||
e310fd43 | 2802 | /* Just pick one node, since fallback list is circular */ |
1da177e4 LT |
2803 | unsigned int sum = 0; |
2804 | ||
0e88460d | 2805 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2806 | |
54a6eb5c | 2807 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
e310fd43 | 2808 | unsigned long size = zone->present_pages; |
41858966 | 2809 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2810 | if (size > high) |
2811 | sum += size - high; | |
1da177e4 LT |
2812 | } |
2813 | ||
2814 | return sum; | |
2815 | } | |
2816 | ||
2817 | /* | |
2818 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
2819 | */ | |
2820 | unsigned int nr_free_buffer_pages(void) | |
2821 | { | |
af4ca457 | 2822 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2823 | } |
c2f1a551 | 2824 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
2825 | |
2826 | /* | |
2827 | * Amount of free RAM allocatable within all zones | |
2828 | */ | |
2829 | unsigned int nr_free_pagecache_pages(void) | |
2830 | { | |
2a1e274a | 2831 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2832 | } |
08e0f6a9 CL |
2833 | |
2834 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2835 | { |
e5adfffc | 2836 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 2837 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2838 | } |
1da177e4 | 2839 | |
1da177e4 LT |
2840 | void si_meminfo(struct sysinfo *val) |
2841 | { | |
2842 | val->totalram = totalram_pages; | |
2843 | val->sharedram = 0; | |
d23ad423 | 2844 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2845 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2846 | val->totalhigh = totalhigh_pages; |
2847 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2848 | val->mem_unit = PAGE_SIZE; |
2849 | } | |
2850 | ||
2851 | EXPORT_SYMBOL(si_meminfo); | |
2852 | ||
2853 | #ifdef CONFIG_NUMA | |
2854 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2855 | { | |
2856 | pg_data_t *pgdat = NODE_DATA(nid); | |
2857 | ||
2858 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 2859 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2860 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2861 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
2862 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2863 | NR_FREE_PAGES); | |
98d2b0eb CL |
2864 | #else |
2865 | val->totalhigh = 0; | |
2866 | val->freehigh = 0; | |
2867 | #endif | |
1da177e4 LT |
2868 | val->mem_unit = PAGE_SIZE; |
2869 | } | |
2870 | #endif | |
2871 | ||
ddd588b5 | 2872 | /* |
7bf02ea2 DR |
2873 | * Determine whether the node should be displayed or not, depending on whether |
2874 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 2875 | */ |
7bf02ea2 | 2876 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
2877 | { |
2878 | bool ret = false; | |
cc9a6c87 | 2879 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
2880 | |
2881 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
2882 | goto out; | |
2883 | ||
cc9a6c87 MG |
2884 | do { |
2885 | cpuset_mems_cookie = get_mems_allowed(); | |
2886 | ret = !node_isset(nid, cpuset_current_mems_allowed); | |
2887 | } while (!put_mems_allowed(cpuset_mems_cookie)); | |
ddd588b5 DR |
2888 | out: |
2889 | return ret; | |
2890 | } | |
2891 | ||
1da177e4 LT |
2892 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
2893 | ||
377e4f16 RV |
2894 | static void show_migration_types(unsigned char type) |
2895 | { | |
2896 | static const char types[MIGRATE_TYPES] = { | |
2897 | [MIGRATE_UNMOVABLE] = 'U', | |
2898 | [MIGRATE_RECLAIMABLE] = 'E', | |
2899 | [MIGRATE_MOVABLE] = 'M', | |
2900 | [MIGRATE_RESERVE] = 'R', | |
2901 | #ifdef CONFIG_CMA | |
2902 | [MIGRATE_CMA] = 'C', | |
2903 | #endif | |
2904 | [MIGRATE_ISOLATE] = 'I', | |
2905 | }; | |
2906 | char tmp[MIGRATE_TYPES + 1]; | |
2907 | char *p = tmp; | |
2908 | int i; | |
2909 | ||
2910 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
2911 | if (type & (1 << i)) | |
2912 | *p++ = types[i]; | |
2913 | } | |
2914 | ||
2915 | *p = '\0'; | |
2916 | printk("(%s) ", tmp); | |
2917 | } | |
2918 | ||
1da177e4 LT |
2919 | /* |
2920 | * Show free area list (used inside shift_scroll-lock stuff) | |
2921 | * We also calculate the percentage fragmentation. We do this by counting the | |
2922 | * memory on each free list with the exception of the first item on the list. | |
ddd588b5 DR |
2923 | * Suppresses nodes that are not allowed by current's cpuset if |
2924 | * SHOW_MEM_FILTER_NODES is passed. | |
1da177e4 | 2925 | */ |
7bf02ea2 | 2926 | void show_free_areas(unsigned int filter) |
1da177e4 | 2927 | { |
c7241913 | 2928 | int cpu; |
1da177e4 LT |
2929 | struct zone *zone; |
2930 | ||
ee99c71c | 2931 | for_each_populated_zone(zone) { |
7bf02ea2 | 2932 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 2933 | continue; |
c7241913 JS |
2934 | show_node(zone); |
2935 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 2936 | |
6b482c67 | 2937 | for_each_online_cpu(cpu) { |
1da177e4 LT |
2938 | struct per_cpu_pageset *pageset; |
2939 | ||
99dcc3e5 | 2940 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 2941 | |
3dfa5721 CL |
2942 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
2943 | cpu, pageset->pcp.high, | |
2944 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
2945 | } |
2946 | } | |
2947 | ||
a731286d KM |
2948 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
2949 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 2950 | " unevictable:%lu" |
b76146ed | 2951 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 2952 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a BZ |
2953 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
2954 | " free_cma:%lu\n", | |
4f98a2fe | 2955 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 2956 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
2957 | global_page_state(NR_ISOLATED_ANON), |
2958 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 2959 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 2960 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 2961 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 2962 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2963 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 2964 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 2965 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
2966 | global_page_state(NR_SLAB_RECLAIMABLE), |
2967 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 2968 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 2969 | global_page_state(NR_SHMEM), |
a25700a5 | 2970 | global_page_state(NR_PAGETABLE), |
d1ce749a BZ |
2971 | global_page_state(NR_BOUNCE), |
2972 | global_page_state(NR_FREE_CMA_PAGES)); | |
1da177e4 | 2973 | |
ee99c71c | 2974 | for_each_populated_zone(zone) { |
1da177e4 LT |
2975 | int i; |
2976 | ||
7bf02ea2 | 2977 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 2978 | continue; |
1da177e4 LT |
2979 | show_node(zone); |
2980 | printk("%s" | |
2981 | " free:%lukB" | |
2982 | " min:%lukB" | |
2983 | " low:%lukB" | |
2984 | " high:%lukB" | |
4f98a2fe RR |
2985 | " active_anon:%lukB" |
2986 | " inactive_anon:%lukB" | |
2987 | " active_file:%lukB" | |
2988 | " inactive_file:%lukB" | |
7b854121 | 2989 | " unevictable:%lukB" |
a731286d KM |
2990 | " isolated(anon):%lukB" |
2991 | " isolated(file):%lukB" | |
1da177e4 | 2992 | " present:%lukB" |
9feedc9d | 2993 | " managed:%lukB" |
4a0aa73f KM |
2994 | " mlocked:%lukB" |
2995 | " dirty:%lukB" | |
2996 | " writeback:%lukB" | |
2997 | " mapped:%lukB" | |
4b02108a | 2998 | " shmem:%lukB" |
4a0aa73f KM |
2999 | " slab_reclaimable:%lukB" |
3000 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3001 | " kernel_stack:%lukB" |
4a0aa73f KM |
3002 | " pagetables:%lukB" |
3003 | " unstable:%lukB" | |
3004 | " bounce:%lukB" | |
d1ce749a | 3005 | " free_cma:%lukB" |
4a0aa73f | 3006 | " writeback_tmp:%lukB" |
1da177e4 LT |
3007 | " pages_scanned:%lu" |
3008 | " all_unreclaimable? %s" | |
3009 | "\n", | |
3010 | zone->name, | |
88f5acf8 | 3011 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3012 | K(min_wmark_pages(zone)), |
3013 | K(low_wmark_pages(zone)), | |
3014 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3015 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3016 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3017 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3018 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3019 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3020 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3021 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3022 | K(zone->present_pages), |
9feedc9d | 3023 | K(zone->managed_pages), |
4a0aa73f KM |
3024 | K(zone_page_state(zone, NR_MLOCK)), |
3025 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3026 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3027 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3028 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3029 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3030 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3031 | zone_page_state(zone, NR_KERNEL_STACK) * |
3032 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3033 | K(zone_page_state(zone, NR_PAGETABLE)), |
3034 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3035 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1ce749a | 3036 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3037 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
1da177e4 | 3038 | zone->pages_scanned, |
93e4a89a | 3039 | (zone->all_unreclaimable ? "yes" : "no") |
1da177e4 LT |
3040 | ); |
3041 | printk("lowmem_reserve[]:"); | |
3042 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3043 | printk(" %lu", zone->lowmem_reserve[i]); | |
3044 | printk("\n"); | |
3045 | } | |
3046 | ||
ee99c71c | 3047 | for_each_populated_zone(zone) { |
8f9de51a | 3048 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
377e4f16 | 3049 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3050 | |
7bf02ea2 | 3051 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3052 | continue; |
1da177e4 LT |
3053 | show_node(zone); |
3054 | printk("%s: ", zone->name); | |
1da177e4 LT |
3055 | |
3056 | spin_lock_irqsave(&zone->lock, flags); | |
3057 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3058 | struct free_area *area = &zone->free_area[order]; |
3059 | int type; | |
3060 | ||
3061 | nr[order] = area->nr_free; | |
8f9de51a | 3062 | total += nr[order] << order; |
377e4f16 RV |
3063 | |
3064 | types[order] = 0; | |
3065 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3066 | if (!list_empty(&area->free_list[type])) | |
3067 | types[order] |= 1 << type; | |
3068 | } | |
1da177e4 LT |
3069 | } |
3070 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3071 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3072 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3073 | if (nr[order]) |
3074 | show_migration_types(types[order]); | |
3075 | } | |
1da177e4 LT |
3076 | printk("= %lukB\n", K(total)); |
3077 | } | |
3078 | ||
e6f3602d LW |
3079 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
3080 | ||
1da177e4 LT |
3081 | show_swap_cache_info(); |
3082 | } | |
3083 | ||
19770b32 MG |
3084 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
3085 | { | |
3086 | zoneref->zone = zone; | |
3087 | zoneref->zone_idx = zone_idx(zone); | |
3088 | } | |
3089 | ||
1da177e4 LT |
3090 | /* |
3091 | * Builds allocation fallback zone lists. | |
1a93205b CL |
3092 | * |
3093 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 3094 | */ |
f0c0b2b8 KH |
3095 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
3096 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 3097 | { |
1a93205b CL |
3098 | struct zone *zone; |
3099 | ||
98d2b0eb | 3100 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 3101 | zone_type++; |
02a68a5e CL |
3102 | |
3103 | do { | |
2f6726e5 | 3104 | zone_type--; |
070f8032 | 3105 | zone = pgdat->node_zones + zone_type; |
1a93205b | 3106 | if (populated_zone(zone)) { |
dd1a239f MG |
3107 | zoneref_set_zone(zone, |
3108 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 3109 | check_highest_zone(zone_type); |
1da177e4 | 3110 | } |
02a68a5e | 3111 | |
2f6726e5 | 3112 | } while (zone_type); |
070f8032 | 3113 | return nr_zones; |
1da177e4 LT |
3114 | } |
3115 | ||
f0c0b2b8 KH |
3116 | |
3117 | /* | |
3118 | * zonelist_order: | |
3119 | * 0 = automatic detection of better ordering. | |
3120 | * 1 = order by ([node] distance, -zonetype) | |
3121 | * 2 = order by (-zonetype, [node] distance) | |
3122 | * | |
3123 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
3124 | * the same zonelist. So only NUMA can configure this param. | |
3125 | */ | |
3126 | #define ZONELIST_ORDER_DEFAULT 0 | |
3127 | #define ZONELIST_ORDER_NODE 1 | |
3128 | #define ZONELIST_ORDER_ZONE 2 | |
3129 | ||
3130 | /* zonelist order in the kernel. | |
3131 | * set_zonelist_order() will set this to NODE or ZONE. | |
3132 | */ | |
3133 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3134 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
3135 | ||
3136 | ||
1da177e4 | 3137 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
3138 | /* The value user specified ....changed by config */ |
3139 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3140 | /* string for sysctl */ | |
3141 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
3142 | char numa_zonelist_order[16] = "default"; | |
3143 | ||
3144 | /* | |
3145 | * interface for configure zonelist ordering. | |
3146 | * command line option "numa_zonelist_order" | |
3147 | * = "[dD]efault - default, automatic configuration. | |
3148 | * = "[nN]ode - order by node locality, then by zone within node | |
3149 | * = "[zZ]one - order by zone, then by locality within zone | |
3150 | */ | |
3151 | ||
3152 | static int __parse_numa_zonelist_order(char *s) | |
3153 | { | |
3154 | if (*s == 'd' || *s == 'D') { | |
3155 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3156 | } else if (*s == 'n' || *s == 'N') { | |
3157 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
3158 | } else if (*s == 'z' || *s == 'Z') { | |
3159 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
3160 | } else { | |
3161 | printk(KERN_WARNING | |
3162 | "Ignoring invalid numa_zonelist_order value: " | |
3163 | "%s\n", s); | |
3164 | return -EINVAL; | |
3165 | } | |
3166 | return 0; | |
3167 | } | |
3168 | ||
3169 | static __init int setup_numa_zonelist_order(char *s) | |
3170 | { | |
ecb256f8 VL |
3171 | int ret; |
3172 | ||
3173 | if (!s) | |
3174 | return 0; | |
3175 | ||
3176 | ret = __parse_numa_zonelist_order(s); | |
3177 | if (ret == 0) | |
3178 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
3179 | ||
3180 | return ret; | |
f0c0b2b8 KH |
3181 | } |
3182 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
3183 | ||
3184 | /* | |
3185 | * sysctl handler for numa_zonelist_order | |
3186 | */ | |
3187 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 3188 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
3189 | loff_t *ppos) |
3190 | { | |
3191 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
3192 | int ret; | |
443c6f14 | 3193 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 3194 | |
443c6f14 | 3195 | mutex_lock(&zl_order_mutex); |
f0c0b2b8 | 3196 | if (write) |
443c6f14 | 3197 | strcpy(saved_string, (char*)table->data); |
8d65af78 | 3198 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 3199 | if (ret) |
443c6f14 | 3200 | goto out; |
f0c0b2b8 KH |
3201 | if (write) { |
3202 | int oldval = user_zonelist_order; | |
3203 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
3204 | /* | |
3205 | * bogus value. restore saved string | |
3206 | */ | |
3207 | strncpy((char*)table->data, saved_string, | |
3208 | NUMA_ZONELIST_ORDER_LEN); | |
3209 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
3210 | } else if (oldval != user_zonelist_order) { |
3211 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 3212 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
3213 | mutex_unlock(&zonelists_mutex); |
3214 | } | |
f0c0b2b8 | 3215 | } |
443c6f14 AK |
3216 | out: |
3217 | mutex_unlock(&zl_order_mutex); | |
3218 | return ret; | |
f0c0b2b8 KH |
3219 | } |
3220 | ||
3221 | ||
62bc62a8 | 3222 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
3223 | static int node_load[MAX_NUMNODES]; |
3224 | ||
1da177e4 | 3225 | /** |
4dc3b16b | 3226 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
3227 | * @node: node whose fallback list we're appending |
3228 | * @used_node_mask: nodemask_t of already used nodes | |
3229 | * | |
3230 | * We use a number of factors to determine which is the next node that should | |
3231 | * appear on a given node's fallback list. The node should not have appeared | |
3232 | * already in @node's fallback list, and it should be the next closest node | |
3233 | * according to the distance array (which contains arbitrary distance values | |
3234 | * from each node to each node in the system), and should also prefer nodes | |
3235 | * with no CPUs, since presumably they'll have very little allocation pressure | |
3236 | * on them otherwise. | |
3237 | * It returns -1 if no node is found. | |
3238 | */ | |
f0c0b2b8 | 3239 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 3240 | { |
4cf808eb | 3241 | int n, val; |
1da177e4 LT |
3242 | int min_val = INT_MAX; |
3243 | int best_node = -1; | |
a70f7302 | 3244 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 3245 | |
4cf808eb LT |
3246 | /* Use the local node if we haven't already */ |
3247 | if (!node_isset(node, *used_node_mask)) { | |
3248 | node_set(node, *used_node_mask); | |
3249 | return node; | |
3250 | } | |
1da177e4 | 3251 | |
4b0ef1fe | 3252 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
3253 | |
3254 | /* Don't want a node to appear more than once */ | |
3255 | if (node_isset(n, *used_node_mask)) | |
3256 | continue; | |
3257 | ||
1da177e4 LT |
3258 | /* Use the distance array to find the distance */ |
3259 | val = node_distance(node, n); | |
3260 | ||
4cf808eb LT |
3261 | /* Penalize nodes under us ("prefer the next node") */ |
3262 | val += (n < node); | |
3263 | ||
1da177e4 | 3264 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
3265 | tmp = cpumask_of_node(n); |
3266 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
3267 | val += PENALTY_FOR_NODE_WITH_CPUS; |
3268 | ||
3269 | /* Slight preference for less loaded node */ | |
3270 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
3271 | val += node_load[n]; | |
3272 | ||
3273 | if (val < min_val) { | |
3274 | min_val = val; | |
3275 | best_node = n; | |
3276 | } | |
3277 | } | |
3278 | ||
3279 | if (best_node >= 0) | |
3280 | node_set(best_node, *used_node_mask); | |
3281 | ||
3282 | return best_node; | |
3283 | } | |
3284 | ||
f0c0b2b8 KH |
3285 | |
3286 | /* | |
3287 | * Build zonelists ordered by node and zones within node. | |
3288 | * This results in maximum locality--normal zone overflows into local | |
3289 | * DMA zone, if any--but risks exhausting DMA zone. | |
3290 | */ | |
3291 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 3292 | { |
f0c0b2b8 | 3293 | int j; |
1da177e4 | 3294 | struct zonelist *zonelist; |
f0c0b2b8 | 3295 | |
54a6eb5c | 3296 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 3297 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
3298 | ; |
3299 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3300 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
3301 | zonelist->_zonerefs[j].zone = NULL; |
3302 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
3303 | } |
3304 | ||
523b9458 CL |
3305 | /* |
3306 | * Build gfp_thisnode zonelists | |
3307 | */ | |
3308 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3309 | { | |
523b9458 CL |
3310 | int j; |
3311 | struct zonelist *zonelist; | |
3312 | ||
54a6eb5c MG |
3313 | zonelist = &pgdat->node_zonelists[1]; |
3314 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
3315 | zonelist->_zonerefs[j].zone = NULL; |
3316 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3317 | } |
3318 | ||
f0c0b2b8 KH |
3319 | /* |
3320 | * Build zonelists ordered by zone and nodes within zones. | |
3321 | * This results in conserving DMA zone[s] until all Normal memory is | |
3322 | * exhausted, but results in overflowing to remote node while memory | |
3323 | * may still exist in local DMA zone. | |
3324 | */ | |
3325 | static int node_order[MAX_NUMNODES]; | |
3326 | ||
3327 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3328 | { | |
f0c0b2b8 KH |
3329 | int pos, j, node; |
3330 | int zone_type; /* needs to be signed */ | |
3331 | struct zone *z; | |
3332 | struct zonelist *zonelist; | |
3333 | ||
54a6eb5c MG |
3334 | zonelist = &pgdat->node_zonelists[0]; |
3335 | pos = 0; | |
3336 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3337 | for (j = 0; j < nr_nodes; j++) { | |
3338 | node = node_order[j]; | |
3339 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3340 | if (populated_zone(z)) { | |
dd1a239f MG |
3341 | zoneref_set_zone(z, |
3342 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3343 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3344 | } |
3345 | } | |
f0c0b2b8 | 3346 | } |
dd1a239f MG |
3347 | zonelist->_zonerefs[pos].zone = NULL; |
3348 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3349 | } |
3350 | ||
3351 | static int default_zonelist_order(void) | |
3352 | { | |
3353 | int nid, zone_type; | |
3354 | unsigned long low_kmem_size,total_size; | |
3355 | struct zone *z; | |
3356 | int average_size; | |
3357 | /* | |
88393161 | 3358 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
f0c0b2b8 KH |
3359 | * If they are really small and used heavily, the system can fall |
3360 | * into OOM very easily. | |
e325c90f | 3361 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
f0c0b2b8 KH |
3362 | */ |
3363 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
3364 | low_kmem_size = 0; | |
3365 | total_size = 0; | |
3366 | for_each_online_node(nid) { | |
3367 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3368 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3369 | if (populated_zone(z)) { | |
3370 | if (zone_type < ZONE_NORMAL) | |
3371 | low_kmem_size += z->present_pages; | |
3372 | total_size += z->present_pages; | |
e325c90f DR |
3373 | } else if (zone_type == ZONE_NORMAL) { |
3374 | /* | |
3375 | * If any node has only lowmem, then node order | |
3376 | * is preferred to allow kernel allocations | |
3377 | * locally; otherwise, they can easily infringe | |
3378 | * on other nodes when there is an abundance of | |
3379 | * lowmem available to allocate from. | |
3380 | */ | |
3381 | return ZONELIST_ORDER_NODE; | |
f0c0b2b8 KH |
3382 | } |
3383 | } | |
3384 | } | |
3385 | if (!low_kmem_size || /* there are no DMA area. */ | |
3386 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
3387 | return ZONELIST_ORDER_NODE; | |
3388 | /* | |
3389 | * look into each node's config. | |
3390 | * If there is a node whose DMA/DMA32 memory is very big area on | |
3391 | * local memory, NODE_ORDER may be suitable. | |
3392 | */ | |
37b07e41 | 3393 | average_size = total_size / |
4b0ef1fe | 3394 | (nodes_weight(node_states[N_MEMORY]) + 1); |
f0c0b2b8 KH |
3395 | for_each_online_node(nid) { |
3396 | low_kmem_size = 0; | |
3397 | total_size = 0; | |
3398 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3399 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3400 | if (populated_zone(z)) { | |
3401 | if (zone_type < ZONE_NORMAL) | |
3402 | low_kmem_size += z->present_pages; | |
3403 | total_size += z->present_pages; | |
3404 | } | |
3405 | } | |
3406 | if (low_kmem_size && | |
3407 | total_size > average_size && /* ignore small node */ | |
3408 | low_kmem_size > total_size * 70/100) | |
3409 | return ZONELIST_ORDER_NODE; | |
3410 | } | |
3411 | return ZONELIST_ORDER_ZONE; | |
3412 | } | |
3413 | ||
3414 | static void set_zonelist_order(void) | |
3415 | { | |
3416 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3417 | current_zonelist_order = default_zonelist_order(); | |
3418 | else | |
3419 | current_zonelist_order = user_zonelist_order; | |
3420 | } | |
3421 | ||
3422 | static void build_zonelists(pg_data_t *pgdat) | |
3423 | { | |
3424 | int j, node, load; | |
3425 | enum zone_type i; | |
1da177e4 | 3426 | nodemask_t used_mask; |
f0c0b2b8 KH |
3427 | int local_node, prev_node; |
3428 | struct zonelist *zonelist; | |
3429 | int order = current_zonelist_order; | |
1da177e4 LT |
3430 | |
3431 | /* initialize zonelists */ | |
523b9458 | 3432 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3433 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3434 | zonelist->_zonerefs[0].zone = NULL; |
3435 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3436 | } |
3437 | ||
3438 | /* NUMA-aware ordering of nodes */ | |
3439 | local_node = pgdat->node_id; | |
62bc62a8 | 3440 | load = nr_online_nodes; |
1da177e4 LT |
3441 | prev_node = local_node; |
3442 | nodes_clear(used_mask); | |
f0c0b2b8 | 3443 | |
f0c0b2b8 KH |
3444 | memset(node_order, 0, sizeof(node_order)); |
3445 | j = 0; | |
3446 | ||
1da177e4 LT |
3447 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
3448 | /* | |
3449 | * We don't want to pressure a particular node. | |
3450 | * So adding penalty to the first node in same | |
3451 | * distance group to make it round-robin. | |
3452 | */ | |
957f822a DR |
3453 | if (node_distance(local_node, node) != |
3454 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
3455 | node_load[node] = load; |
3456 | ||
1da177e4 LT |
3457 | prev_node = node; |
3458 | load--; | |
f0c0b2b8 KH |
3459 | if (order == ZONELIST_ORDER_NODE) |
3460 | build_zonelists_in_node_order(pgdat, node); | |
3461 | else | |
3462 | node_order[j++] = node; /* remember order */ | |
3463 | } | |
1da177e4 | 3464 | |
f0c0b2b8 KH |
3465 | if (order == ZONELIST_ORDER_ZONE) { |
3466 | /* calculate node order -- i.e., DMA last! */ | |
3467 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3468 | } |
523b9458 CL |
3469 | |
3470 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3471 | } |
3472 | ||
9276b1bc | 3473 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3474 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3475 | { |
54a6eb5c MG |
3476 | struct zonelist *zonelist; |
3477 | struct zonelist_cache *zlc; | |
dd1a239f | 3478 | struct zoneref *z; |
9276b1bc | 3479 | |
54a6eb5c MG |
3480 | zonelist = &pgdat->node_zonelists[0]; |
3481 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3482 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3483 | for (z = zonelist->_zonerefs; z->zone; z++) |
3484 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
3485 | } |
3486 | ||
7aac7898 LS |
3487 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3488 | /* | |
3489 | * Return node id of node used for "local" allocations. | |
3490 | * I.e., first node id of first zone in arg node's generic zonelist. | |
3491 | * Used for initializing percpu 'numa_mem', which is used primarily | |
3492 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
3493 | */ | |
3494 | int local_memory_node(int node) | |
3495 | { | |
3496 | struct zone *zone; | |
3497 | ||
3498 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
3499 | gfp_zone(GFP_KERNEL), | |
3500 | NULL, | |
3501 | &zone); | |
3502 | return zone->node; | |
3503 | } | |
3504 | #endif | |
f0c0b2b8 | 3505 | |
1da177e4 LT |
3506 | #else /* CONFIG_NUMA */ |
3507 | ||
f0c0b2b8 KH |
3508 | static void set_zonelist_order(void) |
3509 | { | |
3510 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
3511 | } | |
3512 | ||
3513 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 3514 | { |
19655d34 | 3515 | int node, local_node; |
54a6eb5c MG |
3516 | enum zone_type j; |
3517 | struct zonelist *zonelist; | |
1da177e4 LT |
3518 | |
3519 | local_node = pgdat->node_id; | |
1da177e4 | 3520 | |
54a6eb5c MG |
3521 | zonelist = &pgdat->node_zonelists[0]; |
3522 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 3523 | |
54a6eb5c MG |
3524 | /* |
3525 | * Now we build the zonelist so that it contains the zones | |
3526 | * of all the other nodes. | |
3527 | * We don't want to pressure a particular node, so when | |
3528 | * building the zones for node N, we make sure that the | |
3529 | * zones coming right after the local ones are those from | |
3530 | * node N+1 (modulo N) | |
3531 | */ | |
3532 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3533 | if (!node_online(node)) | |
3534 | continue; | |
3535 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3536 | MAX_NR_ZONES - 1); | |
1da177e4 | 3537 | } |
54a6eb5c MG |
3538 | for (node = 0; node < local_node; node++) { |
3539 | if (!node_online(node)) | |
3540 | continue; | |
3541 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3542 | MAX_NR_ZONES - 1); | |
3543 | } | |
3544 | ||
dd1a239f MG |
3545 | zonelist->_zonerefs[j].zone = NULL; |
3546 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3547 | } |
3548 | ||
9276b1bc | 3549 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3550 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3551 | { |
54a6eb5c | 3552 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3553 | } |
3554 | ||
1da177e4 LT |
3555 | #endif /* CONFIG_NUMA */ |
3556 | ||
99dcc3e5 CL |
3557 | /* |
3558 | * Boot pageset table. One per cpu which is going to be used for all | |
3559 | * zones and all nodes. The parameters will be set in such a way | |
3560 | * that an item put on a list will immediately be handed over to | |
3561 | * the buddy list. This is safe since pageset manipulation is done | |
3562 | * with interrupts disabled. | |
3563 | * | |
3564 | * The boot_pagesets must be kept even after bootup is complete for | |
3565 | * unused processors and/or zones. They do play a role for bootstrapping | |
3566 | * hotplugged processors. | |
3567 | * | |
3568 | * zoneinfo_show() and maybe other functions do | |
3569 | * not check if the processor is online before following the pageset pointer. | |
3570 | * Other parts of the kernel may not check if the zone is available. | |
3571 | */ | |
3572 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3573 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3574 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3575 | |
4eaf3f64 HL |
3576 | /* |
3577 | * Global mutex to protect against size modification of zonelists | |
3578 | * as well as to serialize pageset setup for the new populated zone. | |
3579 | */ | |
3580 | DEFINE_MUTEX(zonelists_mutex); | |
3581 | ||
9b1a4d38 | 3582 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 3583 | static int __build_all_zonelists(void *data) |
1da177e4 | 3584 | { |
6811378e | 3585 | int nid; |
99dcc3e5 | 3586 | int cpu; |
9adb62a5 | 3587 | pg_data_t *self = data; |
9276b1bc | 3588 | |
7f9cfb31 BL |
3589 | #ifdef CONFIG_NUMA |
3590 | memset(node_load, 0, sizeof(node_load)); | |
3591 | #endif | |
9adb62a5 JL |
3592 | |
3593 | if (self && !node_online(self->node_id)) { | |
3594 | build_zonelists(self); | |
3595 | build_zonelist_cache(self); | |
3596 | } | |
3597 | ||
9276b1bc | 3598 | for_each_online_node(nid) { |
7ea1530a CL |
3599 | pg_data_t *pgdat = NODE_DATA(nid); |
3600 | ||
3601 | build_zonelists(pgdat); | |
3602 | build_zonelist_cache(pgdat); | |
9276b1bc | 3603 | } |
99dcc3e5 CL |
3604 | |
3605 | /* | |
3606 | * Initialize the boot_pagesets that are going to be used | |
3607 | * for bootstrapping processors. The real pagesets for | |
3608 | * each zone will be allocated later when the per cpu | |
3609 | * allocator is available. | |
3610 | * | |
3611 | * boot_pagesets are used also for bootstrapping offline | |
3612 | * cpus if the system is already booted because the pagesets | |
3613 | * are needed to initialize allocators on a specific cpu too. | |
3614 | * F.e. the percpu allocator needs the page allocator which | |
3615 | * needs the percpu allocator in order to allocate its pagesets | |
3616 | * (a chicken-egg dilemma). | |
3617 | */ | |
7aac7898 | 3618 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3619 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3620 | ||
7aac7898 LS |
3621 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3622 | /* | |
3623 | * We now know the "local memory node" for each node-- | |
3624 | * i.e., the node of the first zone in the generic zonelist. | |
3625 | * Set up numa_mem percpu variable for on-line cpus. During | |
3626 | * boot, only the boot cpu should be on-line; we'll init the | |
3627 | * secondary cpus' numa_mem as they come on-line. During | |
3628 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3629 | */ | |
3630 | if (cpu_online(cpu)) | |
3631 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3632 | #endif | |
3633 | } | |
3634 | ||
6811378e YG |
3635 | return 0; |
3636 | } | |
3637 | ||
4eaf3f64 HL |
3638 | /* |
3639 | * Called with zonelists_mutex held always | |
3640 | * unless system_state == SYSTEM_BOOTING. | |
3641 | */ | |
9adb62a5 | 3642 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 3643 | { |
f0c0b2b8 KH |
3644 | set_zonelist_order(); |
3645 | ||
6811378e | 3646 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3647 | __build_all_zonelists(NULL); |
68ad8df4 | 3648 | mminit_verify_zonelist(); |
6811378e YG |
3649 | cpuset_init_current_mems_allowed(); |
3650 | } else { | |
183ff22b | 3651 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 3652 | of zonelist */ |
e9959f0f | 3653 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
3654 | if (zone) |
3655 | setup_zone_pageset(zone); | |
e9959f0f | 3656 | #endif |
9adb62a5 | 3657 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
3658 | /* cpuset refresh routine should be here */ |
3659 | } | |
bd1e22b8 | 3660 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3661 | /* |
3662 | * Disable grouping by mobility if the number of pages in the | |
3663 | * system is too low to allow the mechanism to work. It would be | |
3664 | * more accurate, but expensive to check per-zone. This check is | |
3665 | * made on memory-hotadd so a system can start with mobility | |
3666 | * disabled and enable it later | |
3667 | */ | |
d9c23400 | 3668 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3669 | page_group_by_mobility_disabled = 1; |
3670 | else | |
3671 | page_group_by_mobility_disabled = 0; | |
3672 | ||
3673 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
3674 | "Total pages: %ld\n", | |
62bc62a8 | 3675 | nr_online_nodes, |
f0c0b2b8 | 3676 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3677 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3678 | vm_total_pages); |
3679 | #ifdef CONFIG_NUMA | |
3680 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
3681 | #endif | |
1da177e4 LT |
3682 | } |
3683 | ||
3684 | /* | |
3685 | * Helper functions to size the waitqueue hash table. | |
3686 | * Essentially these want to choose hash table sizes sufficiently | |
3687 | * large so that collisions trying to wait on pages are rare. | |
3688 | * But in fact, the number of active page waitqueues on typical | |
3689 | * systems is ridiculously low, less than 200. So this is even | |
3690 | * conservative, even though it seems large. | |
3691 | * | |
3692 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
3693 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
3694 | */ | |
3695 | #define PAGES_PER_WAITQUEUE 256 | |
3696 | ||
cca448fe | 3697 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 3698 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
3699 | { |
3700 | unsigned long size = 1; | |
3701 | ||
3702 | pages /= PAGES_PER_WAITQUEUE; | |
3703 | ||
3704 | while (size < pages) | |
3705 | size <<= 1; | |
3706 | ||
3707 | /* | |
3708 | * Once we have dozens or even hundreds of threads sleeping | |
3709 | * on IO we've got bigger problems than wait queue collision. | |
3710 | * Limit the size of the wait table to a reasonable size. | |
3711 | */ | |
3712 | size = min(size, 4096UL); | |
3713 | ||
3714 | return max(size, 4UL); | |
3715 | } | |
cca448fe YG |
3716 | #else |
3717 | /* | |
3718 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
3719 | * a suitable size for its wait_table. So we use the maximum size now. | |
3720 | * | |
3721 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
3722 | * | |
3723 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
3724 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
3725 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
3726 | * | |
3727 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
3728 | * or more by the traditional way. (See above). It equals: | |
3729 | * | |
3730 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
3731 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
3732 | * powerpc (64K page size) : = (32G +16M)byte. | |
3733 | */ | |
3734 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
3735 | { | |
3736 | return 4096UL; | |
3737 | } | |
3738 | #endif | |
1da177e4 LT |
3739 | |
3740 | /* | |
3741 | * This is an integer logarithm so that shifts can be used later | |
3742 | * to extract the more random high bits from the multiplicative | |
3743 | * hash function before the remainder is taken. | |
3744 | */ | |
3745 | static inline unsigned long wait_table_bits(unsigned long size) | |
3746 | { | |
3747 | return ffz(~size); | |
3748 | } | |
3749 | ||
3750 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
3751 | ||
6d3163ce AH |
3752 | /* |
3753 | * Check if a pageblock contains reserved pages | |
3754 | */ | |
3755 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
3756 | { | |
3757 | unsigned long pfn; | |
3758 | ||
3759 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
3760 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
3761 | return 1; | |
3762 | } | |
3763 | return 0; | |
3764 | } | |
3765 | ||
56fd56b8 | 3766 | /* |
d9c23400 | 3767 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
3768 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
3769 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
3770 | * higher will lead to a bigger reserve which will get freed as contiguous |
3771 | * blocks as reclaim kicks in | |
3772 | */ | |
3773 | static void setup_zone_migrate_reserve(struct zone *zone) | |
3774 | { | |
6d3163ce | 3775 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 3776 | struct page *page; |
78986a67 MG |
3777 | unsigned long block_migratetype; |
3778 | int reserve; | |
56fd56b8 | 3779 | |
d0215638 MH |
3780 | /* |
3781 | * Get the start pfn, end pfn and the number of blocks to reserve | |
3782 | * We have to be careful to be aligned to pageblock_nr_pages to | |
3783 | * make sure that we always check pfn_valid for the first page in | |
3784 | * the block. | |
3785 | */ | |
56fd56b8 MG |
3786 | start_pfn = zone->zone_start_pfn; |
3787 | end_pfn = start_pfn + zone->spanned_pages; | |
d0215638 | 3788 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 3789 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 3790 | pageblock_order; |
56fd56b8 | 3791 | |
78986a67 MG |
3792 | /* |
3793 | * Reserve blocks are generally in place to help high-order atomic | |
3794 | * allocations that are short-lived. A min_free_kbytes value that | |
3795 | * would result in more than 2 reserve blocks for atomic allocations | |
3796 | * is assumed to be in place to help anti-fragmentation for the | |
3797 | * future allocation of hugepages at runtime. | |
3798 | */ | |
3799 | reserve = min(2, reserve); | |
3800 | ||
d9c23400 | 3801 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
3802 | if (!pfn_valid(pfn)) |
3803 | continue; | |
3804 | page = pfn_to_page(pfn); | |
3805 | ||
344c790e AL |
3806 | /* Watch out for overlapping nodes */ |
3807 | if (page_to_nid(page) != zone_to_nid(zone)) | |
3808 | continue; | |
3809 | ||
56fd56b8 MG |
3810 | block_migratetype = get_pageblock_migratetype(page); |
3811 | ||
938929f1 MG |
3812 | /* Only test what is necessary when the reserves are not met */ |
3813 | if (reserve > 0) { | |
3814 | /* | |
3815 | * Blocks with reserved pages will never free, skip | |
3816 | * them. | |
3817 | */ | |
3818 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
3819 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
3820 | continue; | |
56fd56b8 | 3821 | |
938929f1 MG |
3822 | /* If this block is reserved, account for it */ |
3823 | if (block_migratetype == MIGRATE_RESERVE) { | |
3824 | reserve--; | |
3825 | continue; | |
3826 | } | |
3827 | ||
3828 | /* Suitable for reserving if this block is movable */ | |
3829 | if (block_migratetype == MIGRATE_MOVABLE) { | |
3830 | set_pageblock_migratetype(page, | |
3831 | MIGRATE_RESERVE); | |
3832 | move_freepages_block(zone, page, | |
3833 | MIGRATE_RESERVE); | |
3834 | reserve--; | |
3835 | continue; | |
3836 | } | |
56fd56b8 MG |
3837 | } |
3838 | ||
3839 | /* | |
3840 | * If the reserve is met and this is a previous reserved block, | |
3841 | * take it back | |
3842 | */ | |
3843 | if (block_migratetype == MIGRATE_RESERVE) { | |
3844 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
3845 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
3846 | } | |
3847 | } | |
3848 | } | |
ac0e5b7a | 3849 | |
1da177e4 LT |
3850 | /* |
3851 | * Initially all pages are reserved - free ones are freed | |
3852 | * up by free_all_bootmem() once the early boot process is | |
3853 | * done. Non-atomic initialization, single-pass. | |
3854 | */ | |
c09b4240 | 3855 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 3856 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 3857 | { |
1da177e4 | 3858 | struct page *page; |
29751f69 AW |
3859 | unsigned long end_pfn = start_pfn + size; |
3860 | unsigned long pfn; | |
86051ca5 | 3861 | struct zone *z; |
1da177e4 | 3862 | |
22b31eec HD |
3863 | if (highest_memmap_pfn < end_pfn - 1) |
3864 | highest_memmap_pfn = end_pfn - 1; | |
3865 | ||
86051ca5 | 3866 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 3867 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
3868 | /* |
3869 | * There can be holes in boot-time mem_map[]s | |
3870 | * handed to this function. They do not | |
3871 | * exist on hotplugged memory. | |
3872 | */ | |
3873 | if (context == MEMMAP_EARLY) { | |
3874 | if (!early_pfn_valid(pfn)) | |
3875 | continue; | |
3876 | if (!early_pfn_in_nid(pfn, nid)) | |
3877 | continue; | |
3878 | } | |
d41dee36 AW |
3879 | page = pfn_to_page(pfn); |
3880 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 3881 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 3882 | init_page_count(page); |
1da177e4 | 3883 | reset_page_mapcount(page); |
57e0a030 | 3884 | reset_page_last_nid(page); |
1da177e4 | 3885 | SetPageReserved(page); |
b2a0ac88 MG |
3886 | /* |
3887 | * Mark the block movable so that blocks are reserved for | |
3888 | * movable at startup. This will force kernel allocations | |
3889 | * to reserve their blocks rather than leaking throughout | |
3890 | * the address space during boot when many long-lived | |
56fd56b8 MG |
3891 | * kernel allocations are made. Later some blocks near |
3892 | * the start are marked MIGRATE_RESERVE by | |
3893 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
3894 | * |
3895 | * bitmap is created for zone's valid pfn range. but memmap | |
3896 | * can be created for invalid pages (for alignment) | |
3897 | * check here not to call set_pageblock_migratetype() against | |
3898 | * pfn out of zone. | |
b2a0ac88 | 3899 | */ |
86051ca5 KH |
3900 | if ((z->zone_start_pfn <= pfn) |
3901 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
3902 | && !(pfn & (pageblock_nr_pages - 1))) | |
56fd56b8 | 3903 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 3904 | |
1da177e4 LT |
3905 | INIT_LIST_HEAD(&page->lru); |
3906 | #ifdef WANT_PAGE_VIRTUAL | |
3907 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
3908 | if (!is_highmem_idx(zone)) | |
3212c6be | 3909 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 3910 | #endif |
1da177e4 LT |
3911 | } |
3912 | } | |
3913 | ||
1e548deb | 3914 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 3915 | { |
b2a0ac88 MG |
3916 | int order, t; |
3917 | for_each_migratetype_order(order, t) { | |
3918 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
3919 | zone->free_area[order].nr_free = 0; |
3920 | } | |
3921 | } | |
3922 | ||
3923 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
3924 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 3925 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
3926 | #endif |
3927 | ||
4ed7e022 | 3928 | static int __meminit zone_batchsize(struct zone *zone) |
e7c8d5c9 | 3929 | { |
3a6be87f | 3930 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
3931 | int batch; |
3932 | ||
3933 | /* | |
3934 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 3935 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
3936 | * |
3937 | * OK, so we don't know how big the cache is. So guess. | |
3938 | */ | |
3939 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
3940 | if (batch * PAGE_SIZE > 512 * 1024) |
3941 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
3942 | batch /= 4; /* We effectively *= 4 below */ |
3943 | if (batch < 1) | |
3944 | batch = 1; | |
3945 | ||
3946 | /* | |
0ceaacc9 NP |
3947 | * Clamp the batch to a 2^n - 1 value. Having a power |
3948 | * of 2 value was found to be more likely to have | |
3949 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 3950 | * |
0ceaacc9 NP |
3951 | * For example if 2 tasks are alternately allocating |
3952 | * batches of pages, one task can end up with a lot | |
3953 | * of pages of one half of the possible page colors | |
3954 | * and the other with pages of the other colors. | |
e7c8d5c9 | 3955 | */ |
9155203a | 3956 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 3957 | |
e7c8d5c9 | 3958 | return batch; |
3a6be87f DH |
3959 | |
3960 | #else | |
3961 | /* The deferral and batching of frees should be suppressed under NOMMU | |
3962 | * conditions. | |
3963 | * | |
3964 | * The problem is that NOMMU needs to be able to allocate large chunks | |
3965 | * of contiguous memory as there's no hardware page translation to | |
3966 | * assemble apparent contiguous memory from discontiguous pages. | |
3967 | * | |
3968 | * Queueing large contiguous runs of pages for batching, however, | |
3969 | * causes the pages to actually be freed in smaller chunks. As there | |
3970 | * can be a significant delay between the individual batches being | |
3971 | * recycled, this leads to the once large chunks of space being | |
3972 | * fragmented and becoming unavailable for high-order allocations. | |
3973 | */ | |
3974 | return 0; | |
3975 | #endif | |
e7c8d5c9 CL |
3976 | } |
3977 | ||
b69a7288 | 3978 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
3979 | { |
3980 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 3981 | int migratetype; |
2caaad41 | 3982 | |
1c6fe946 MD |
3983 | memset(p, 0, sizeof(*p)); |
3984 | ||
3dfa5721 | 3985 | pcp = &p->pcp; |
2caaad41 | 3986 | pcp->count = 0; |
2caaad41 CL |
3987 | pcp->high = 6 * batch; |
3988 | pcp->batch = max(1UL, 1 * batch); | |
5f8dcc21 MG |
3989 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
3990 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
3991 | } |
3992 | ||
8ad4b1fb RS |
3993 | /* |
3994 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
3995 | * to the value high for the pageset p. | |
3996 | */ | |
3997 | ||
3998 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
3999 | unsigned long high) | |
4000 | { | |
4001 | struct per_cpu_pages *pcp; | |
4002 | ||
3dfa5721 | 4003 | pcp = &p->pcp; |
8ad4b1fb RS |
4004 | pcp->high = high; |
4005 | pcp->batch = max(1UL, high/4); | |
4006 | if ((high/4) > (PAGE_SHIFT * 8)) | |
4007 | pcp->batch = PAGE_SHIFT * 8; | |
4008 | } | |
4009 | ||
4ed7e022 | 4010 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4011 | { |
4012 | int cpu; | |
4013 | ||
4014 | zone->pageset = alloc_percpu(struct per_cpu_pageset); | |
4015 | ||
4016 | for_each_possible_cpu(cpu) { | |
4017 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4018 | ||
4019 | setup_pageset(pcp, zone_batchsize(zone)); | |
4020 | ||
4021 | if (percpu_pagelist_fraction) | |
4022 | setup_pagelist_highmark(pcp, | |
4023 | (zone->present_pages / | |
4024 | percpu_pagelist_fraction)); | |
4025 | } | |
4026 | } | |
4027 | ||
2caaad41 | 4028 | /* |
99dcc3e5 CL |
4029 | * Allocate per cpu pagesets and initialize them. |
4030 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4031 | */ |
99dcc3e5 | 4032 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4033 | { |
99dcc3e5 | 4034 | struct zone *zone; |
e7c8d5c9 | 4035 | |
319774e2 WF |
4036 | for_each_populated_zone(zone) |
4037 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4038 | } |
4039 | ||
577a32f6 | 4040 | static noinline __init_refok |
cca448fe | 4041 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4042 | { |
4043 | int i; | |
4044 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 4045 | size_t alloc_size; |
ed8ece2e DH |
4046 | |
4047 | /* | |
4048 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4049 | * per zone. | |
4050 | */ | |
02b694de YG |
4051 | zone->wait_table_hash_nr_entries = |
4052 | wait_table_hash_nr_entries(zone_size_pages); | |
4053 | zone->wait_table_bits = | |
4054 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4055 | alloc_size = zone->wait_table_hash_nr_entries |
4056 | * sizeof(wait_queue_head_t); | |
4057 | ||
cd94b9db | 4058 | if (!slab_is_available()) { |
cca448fe | 4059 | zone->wait_table = (wait_queue_head_t *) |
8f389a99 | 4060 | alloc_bootmem_node_nopanic(pgdat, alloc_size); |
cca448fe YG |
4061 | } else { |
4062 | /* | |
4063 | * This case means that a zone whose size was 0 gets new memory | |
4064 | * via memory hot-add. | |
4065 | * But it may be the case that a new node was hot-added. In | |
4066 | * this case vmalloc() will not be able to use this new node's | |
4067 | * memory - this wait_table must be initialized to use this new | |
4068 | * node itself as well. | |
4069 | * To use this new node's memory, further consideration will be | |
4070 | * necessary. | |
4071 | */ | |
8691f3a7 | 4072 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4073 | } |
4074 | if (!zone->wait_table) | |
4075 | return -ENOMEM; | |
ed8ece2e | 4076 | |
02b694de | 4077 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4078 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4079 | |
4080 | return 0; | |
ed8ece2e DH |
4081 | } |
4082 | ||
c09b4240 | 4083 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4084 | { |
99dcc3e5 CL |
4085 | /* |
4086 | * per cpu subsystem is not up at this point. The following code | |
4087 | * relies on the ability of the linker to provide the | |
4088 | * offset of a (static) per cpu variable into the per cpu area. | |
4089 | */ | |
4090 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4091 | |
f5335c0f | 4092 | if (zone->present_pages) |
99dcc3e5 CL |
4093 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4094 | zone->name, zone->present_pages, | |
4095 | zone_batchsize(zone)); | |
ed8ece2e DH |
4096 | } |
4097 | ||
4ed7e022 | 4098 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4099 | unsigned long zone_start_pfn, |
a2f3aa02 DH |
4100 | unsigned long size, |
4101 | enum memmap_context context) | |
ed8ece2e DH |
4102 | { |
4103 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4104 | int ret; |
4105 | ret = zone_wait_table_init(zone, size); | |
4106 | if (ret) | |
4107 | return ret; | |
ed8ece2e DH |
4108 | pgdat->nr_zones = zone_idx(zone) + 1; |
4109 | ||
ed8ece2e DH |
4110 | zone->zone_start_pfn = zone_start_pfn; |
4111 | ||
708614e6 MG |
4112 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4113 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4114 | pgdat->node_id, | |
4115 | (unsigned long)zone_idx(zone), | |
4116 | zone_start_pfn, (zone_start_pfn + size)); | |
4117 | ||
1e548deb | 4118 | zone_init_free_lists(zone); |
718127cc YG |
4119 | |
4120 | return 0; | |
ed8ece2e DH |
4121 | } |
4122 | ||
0ee332c1 | 4123 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
4124 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
4125 | /* | |
4126 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
4127 | * Architectures may implement their own version but if add_active_range() | |
4128 | * was used and there are no special requirements, this is a convenient | |
4129 | * alternative | |
4130 | */ | |
f2dbcfa7 | 4131 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d | 4132 | { |
c13291a5 TH |
4133 | unsigned long start_pfn, end_pfn; |
4134 | int i, nid; | |
c713216d | 4135 | |
c13291a5 | 4136 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
c713216d | 4137 | if (start_pfn <= pfn && pfn < end_pfn) |
c13291a5 | 4138 | return nid; |
cc2559bc KH |
4139 | /* This is a memory hole */ |
4140 | return -1; | |
c713216d MG |
4141 | } |
4142 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
4143 | ||
f2dbcfa7 KH |
4144 | int __meminit early_pfn_to_nid(unsigned long pfn) |
4145 | { | |
cc2559bc KH |
4146 | int nid; |
4147 | ||
4148 | nid = __early_pfn_to_nid(pfn); | |
4149 | if (nid >= 0) | |
4150 | return nid; | |
4151 | /* just returns 0 */ | |
4152 | return 0; | |
f2dbcfa7 KH |
4153 | } |
4154 | ||
cc2559bc KH |
4155 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
4156 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
4157 | { | |
4158 | int nid; | |
4159 | ||
4160 | nid = __early_pfn_to_nid(pfn); | |
4161 | if (nid >= 0 && nid != node) | |
4162 | return false; | |
4163 | return true; | |
4164 | } | |
4165 | #endif | |
f2dbcfa7 | 4166 | |
c713216d MG |
4167 | /** |
4168 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
4169 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
4170 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
4171 | * |
4172 | * If an architecture guarantees that all ranges registered with | |
4173 | * add_active_ranges() contain no holes and may be freed, this | |
4174 | * this function may be used instead of calling free_bootmem() manually. | |
4175 | */ | |
c13291a5 | 4176 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 4177 | { |
c13291a5 TH |
4178 | unsigned long start_pfn, end_pfn; |
4179 | int i, this_nid; | |
edbe7d23 | 4180 | |
c13291a5 TH |
4181 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
4182 | start_pfn = min(start_pfn, max_low_pfn); | |
4183 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 4184 | |
c13291a5 TH |
4185 | if (start_pfn < end_pfn) |
4186 | free_bootmem_node(NODE_DATA(this_nid), | |
4187 | PFN_PHYS(start_pfn), | |
4188 | (end_pfn - start_pfn) << PAGE_SHIFT); | |
edbe7d23 | 4189 | } |
edbe7d23 | 4190 | } |
edbe7d23 | 4191 | |
c713216d MG |
4192 | /** |
4193 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 4194 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
4195 | * |
4196 | * If an architecture guarantees that all ranges registered with | |
4197 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 4198 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
4199 | */ |
4200 | void __init sparse_memory_present_with_active_regions(int nid) | |
4201 | { | |
c13291a5 TH |
4202 | unsigned long start_pfn, end_pfn; |
4203 | int i, this_nid; | |
c713216d | 4204 | |
c13291a5 TH |
4205 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
4206 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
4207 | } |
4208 | ||
4209 | /** | |
4210 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
4211 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
4212 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
4213 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
4214 | * |
4215 | * It returns the start and end page frame of a node based on information | |
4216 | * provided by an arch calling add_active_range(). If called for a node | |
4217 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 4218 | * PFNs will be 0. |
c713216d | 4219 | */ |
a3142c8e | 4220 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
4221 | unsigned long *start_pfn, unsigned long *end_pfn) |
4222 | { | |
c13291a5 | 4223 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 4224 | int i; |
c13291a5 | 4225 | |
c713216d MG |
4226 | *start_pfn = -1UL; |
4227 | *end_pfn = 0; | |
4228 | ||
c13291a5 TH |
4229 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
4230 | *start_pfn = min(*start_pfn, this_start_pfn); | |
4231 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
4232 | } |
4233 | ||
633c0666 | 4234 | if (*start_pfn == -1UL) |
c713216d | 4235 | *start_pfn = 0; |
c713216d MG |
4236 | } |
4237 | ||
2a1e274a MG |
4238 | /* |
4239 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
4240 | * assumption is made that zones within a node are ordered in monotonic | |
4241 | * increasing memory addresses so that the "highest" populated zone is used | |
4242 | */ | |
b69a7288 | 4243 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
4244 | { |
4245 | int zone_index; | |
4246 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
4247 | if (zone_index == ZONE_MOVABLE) | |
4248 | continue; | |
4249 | ||
4250 | if (arch_zone_highest_possible_pfn[zone_index] > | |
4251 | arch_zone_lowest_possible_pfn[zone_index]) | |
4252 | break; | |
4253 | } | |
4254 | ||
4255 | VM_BUG_ON(zone_index == -1); | |
4256 | movable_zone = zone_index; | |
4257 | } | |
4258 | ||
4259 | /* | |
4260 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 4261 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
4262 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
4263 | * in each node depending on the size of each node and how evenly kernelcore | |
4264 | * is distributed. This helper function adjusts the zone ranges | |
4265 | * provided by the architecture for a given node by using the end of the | |
4266 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
4267 | * zones within a node are in order of monotonic increases memory addresses | |
4268 | */ | |
b69a7288 | 4269 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4270 | unsigned long zone_type, |
4271 | unsigned long node_start_pfn, | |
4272 | unsigned long node_end_pfn, | |
4273 | unsigned long *zone_start_pfn, | |
4274 | unsigned long *zone_end_pfn) | |
4275 | { | |
4276 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4277 | if (zone_movable_pfn[nid]) { | |
4278 | /* Size ZONE_MOVABLE */ | |
4279 | if (zone_type == ZONE_MOVABLE) { | |
4280 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4281 | *zone_end_pfn = min(node_end_pfn, | |
4282 | arch_zone_highest_possible_pfn[movable_zone]); | |
4283 | ||
4284 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4285 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4286 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4287 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4288 | ||
4289 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4290 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4291 | *zone_start_pfn = *zone_end_pfn; | |
4292 | } | |
4293 | } | |
4294 | ||
c713216d MG |
4295 | /* |
4296 | * Return the number of pages a zone spans in a node, including holes | |
4297 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4298 | */ | |
6ea6e688 | 4299 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4300 | unsigned long zone_type, |
4301 | unsigned long *ignored) | |
4302 | { | |
4303 | unsigned long node_start_pfn, node_end_pfn; | |
4304 | unsigned long zone_start_pfn, zone_end_pfn; | |
4305 | ||
4306 | /* Get the start and end of the node and zone */ | |
4307 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
4308 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
4309 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4310 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4311 | node_start_pfn, node_end_pfn, | |
4312 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4313 | |
4314 | /* Check that this node has pages within the zone's required range */ | |
4315 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4316 | return 0; | |
4317 | ||
4318 | /* Move the zone boundaries inside the node if necessary */ | |
4319 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4320 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4321 | ||
4322 | /* Return the spanned pages */ | |
4323 | return zone_end_pfn - zone_start_pfn; | |
4324 | } | |
4325 | ||
4326 | /* | |
4327 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4328 | * then all holes in the requested range will be accounted for. |
c713216d | 4329 | */ |
32996250 | 4330 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4331 | unsigned long range_start_pfn, |
4332 | unsigned long range_end_pfn) | |
4333 | { | |
96e907d1 TH |
4334 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4335 | unsigned long start_pfn, end_pfn; | |
4336 | int i; | |
c713216d | 4337 | |
96e907d1 TH |
4338 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4339 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4340 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4341 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4342 | } |
96e907d1 | 4343 | return nr_absent; |
c713216d MG |
4344 | } |
4345 | ||
4346 | /** | |
4347 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4348 | * @start_pfn: The start PFN to start searching for holes | |
4349 | * @end_pfn: The end PFN to stop searching for holes | |
4350 | * | |
88ca3b94 | 4351 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4352 | */ |
4353 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4354 | unsigned long end_pfn) | |
4355 | { | |
4356 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4357 | } | |
4358 | ||
4359 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4360 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4361 | unsigned long zone_type, |
4362 | unsigned long *ignored) | |
4363 | { | |
96e907d1 TH |
4364 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4365 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4366 | unsigned long node_start_pfn, node_end_pfn; |
4367 | unsigned long zone_start_pfn, zone_end_pfn; | |
4368 | ||
4369 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
96e907d1 TH |
4370 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4371 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4372 | |
2a1e274a MG |
4373 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4374 | node_start_pfn, node_end_pfn, | |
4375 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4376 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4377 | } |
0e0b864e | 4378 | |
6981ec31 TC |
4379 | /** |
4380 | * sanitize_zone_movable_limit - Sanitize the zone_movable_limit array. | |
4381 | * | |
4382 | * zone_movable_limit is initialized as 0. This function will try to get | |
4383 | * the first ZONE_MOVABLE pfn of each node from movablemem_map, and | |
4384 | * assigne them to zone_movable_limit. | |
4385 | * zone_movable_limit[nid] == 0 means no limit for the node. | |
4386 | * | |
4387 | * Note: Each range is represented as [start_pfn, end_pfn) | |
4388 | */ | |
4389 | static void __meminit sanitize_zone_movable_limit(void) | |
4390 | { | |
4391 | int map_pos = 0, i, nid; | |
4392 | unsigned long start_pfn, end_pfn; | |
4393 | ||
4394 | if (!movablemem_map.nr_map) | |
4395 | return; | |
4396 | ||
4397 | /* Iterate all ranges from minimum to maximum */ | |
4398 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
4399 | /* | |
4400 | * If we have found lowest pfn of ZONE_MOVABLE of the node | |
4401 | * specified by user, just go on to check next range. | |
4402 | */ | |
4403 | if (zone_movable_limit[nid]) | |
4404 | continue; | |
4405 | ||
4406 | #ifdef CONFIG_ZONE_DMA | |
4407 | /* Skip DMA memory. */ | |
4408 | if (start_pfn < arch_zone_highest_possible_pfn[ZONE_DMA]) | |
4409 | start_pfn = arch_zone_highest_possible_pfn[ZONE_DMA]; | |
4410 | #endif | |
4411 | ||
4412 | #ifdef CONFIG_ZONE_DMA32 | |
4413 | /* Skip DMA32 memory. */ | |
4414 | if (start_pfn < arch_zone_highest_possible_pfn[ZONE_DMA32]) | |
4415 | start_pfn = arch_zone_highest_possible_pfn[ZONE_DMA32]; | |
4416 | #endif | |
4417 | ||
4418 | #ifdef CONFIG_HIGHMEM | |
4419 | /* Skip lowmem if ZONE_MOVABLE is highmem. */ | |
4420 | if (zone_movable_is_highmem() && | |
4421 | start_pfn < arch_zone_lowest_possible_pfn[ZONE_HIGHMEM]) | |
4422 | start_pfn = arch_zone_lowest_possible_pfn[ZONE_HIGHMEM]; | |
4423 | #endif | |
4424 | ||
4425 | if (start_pfn >= end_pfn) | |
4426 | continue; | |
4427 | ||
4428 | while (map_pos < movablemem_map.nr_map) { | |
4429 | if (end_pfn <= movablemem_map.map[map_pos].start_pfn) | |
4430 | break; | |
4431 | ||
4432 | if (start_pfn >= movablemem_map.map[map_pos].end_pfn) { | |
4433 | map_pos++; | |
4434 | continue; | |
4435 | } | |
4436 | ||
4437 | /* | |
4438 | * The start_pfn of ZONE_MOVABLE is either the minimum | |
4439 | * pfn specified by movablemem_map, or 0, which means | |
4440 | * the node has no ZONE_MOVABLE. | |
4441 | */ | |
4442 | zone_movable_limit[nid] = max(start_pfn, | |
4443 | movablemem_map.map[map_pos].start_pfn); | |
4444 | ||
4445 | break; | |
4446 | } | |
4447 | } | |
4448 | } | |
4449 | ||
0ee332c1 | 4450 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4451 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4452 | unsigned long zone_type, |
4453 | unsigned long *zones_size) | |
4454 | { | |
4455 | return zones_size[zone_type]; | |
4456 | } | |
4457 | ||
6ea6e688 | 4458 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4459 | unsigned long zone_type, |
4460 | unsigned long *zholes_size) | |
4461 | { | |
4462 | if (!zholes_size) | |
4463 | return 0; | |
4464 | ||
4465 | return zholes_size[zone_type]; | |
4466 | } | |
0ee332c1 | 4467 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4468 | |
a3142c8e | 4469 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
4470 | unsigned long *zones_size, unsigned long *zholes_size) |
4471 | { | |
4472 | unsigned long realtotalpages, totalpages = 0; | |
4473 | enum zone_type i; | |
4474 | ||
4475 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4476 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
4477 | zones_size); | |
4478 | pgdat->node_spanned_pages = totalpages; | |
4479 | ||
4480 | realtotalpages = totalpages; | |
4481 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4482 | realtotalpages -= | |
4483 | zone_absent_pages_in_node(pgdat->node_id, i, | |
4484 | zholes_size); | |
4485 | pgdat->node_present_pages = realtotalpages; | |
4486 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4487 | realtotalpages); | |
4488 | } | |
4489 | ||
835c134e MG |
4490 | #ifndef CONFIG_SPARSEMEM |
4491 | /* | |
4492 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4493 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4494 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4495 | * round what is now in bits to nearest long in bits, then return it in |
4496 | * bytes. | |
4497 | */ | |
7c45512d | 4498 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
4499 | { |
4500 | unsigned long usemapsize; | |
4501 | ||
7c45512d | 4502 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
4503 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4504 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4505 | usemapsize *= NR_PAGEBLOCK_BITS; |
4506 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4507 | ||
4508 | return usemapsize / 8; | |
4509 | } | |
4510 | ||
4511 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
4512 | struct zone *zone, |
4513 | unsigned long zone_start_pfn, | |
4514 | unsigned long zonesize) | |
835c134e | 4515 | { |
7c45512d | 4516 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 4517 | zone->pageblock_flags = NULL; |
58a01a45 | 4518 | if (usemapsize) |
8f389a99 YL |
4519 | zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat, |
4520 | usemapsize); | |
835c134e MG |
4521 | } |
4522 | #else | |
7c45512d LT |
4523 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
4524 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
4525 | #endif /* CONFIG_SPARSEMEM */ |
4526 | ||
d9c23400 | 4527 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 4528 | |
d9c23400 | 4529 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
ca57df79 | 4530 | void __init set_pageblock_order(void) |
d9c23400 | 4531 | { |
955c1cd7 AM |
4532 | unsigned int order; |
4533 | ||
d9c23400 MG |
4534 | /* Check that pageblock_nr_pages has not already been setup */ |
4535 | if (pageblock_order) | |
4536 | return; | |
4537 | ||
955c1cd7 AM |
4538 | if (HPAGE_SHIFT > PAGE_SHIFT) |
4539 | order = HUGETLB_PAGE_ORDER; | |
4540 | else | |
4541 | order = MAX_ORDER - 1; | |
4542 | ||
d9c23400 MG |
4543 | /* |
4544 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
4545 | * This value may be variable depending on boot parameters on IA64 and |
4546 | * powerpc. | |
d9c23400 MG |
4547 | */ |
4548 | pageblock_order = order; | |
4549 | } | |
4550 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4551 | ||
ba72cb8c MG |
4552 | /* |
4553 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
4554 | * is unused as pageblock_order is set at compile-time. See |
4555 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
4556 | * the kernel config | |
ba72cb8c | 4557 | */ |
ca57df79 | 4558 | void __init set_pageblock_order(void) |
ba72cb8c | 4559 | { |
ba72cb8c | 4560 | } |
d9c23400 MG |
4561 | |
4562 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4563 | ||
01cefaef JL |
4564 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
4565 | unsigned long present_pages) | |
4566 | { | |
4567 | unsigned long pages = spanned_pages; | |
4568 | ||
4569 | /* | |
4570 | * Provide a more accurate estimation if there are holes within | |
4571 | * the zone and SPARSEMEM is in use. If there are holes within the | |
4572 | * zone, each populated memory region may cost us one or two extra | |
4573 | * memmap pages due to alignment because memmap pages for each | |
4574 | * populated regions may not naturally algined on page boundary. | |
4575 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
4576 | */ | |
4577 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
4578 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
4579 | pages = present_pages; | |
4580 | ||
4581 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
4582 | } | |
4583 | ||
1da177e4 LT |
4584 | /* |
4585 | * Set up the zone data structures: | |
4586 | * - mark all pages reserved | |
4587 | * - mark all memory queues empty | |
4588 | * - clear the memory bitmaps | |
6527af5d MK |
4589 | * |
4590 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 4591 | */ |
b5a0e011 | 4592 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
4593 | unsigned long *zones_size, unsigned long *zholes_size) |
4594 | { | |
2f1b6248 | 4595 | enum zone_type j; |
ed8ece2e | 4596 | int nid = pgdat->node_id; |
1da177e4 | 4597 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4598 | int ret; |
1da177e4 | 4599 | |
208d54e5 | 4600 | pgdat_resize_init(pgdat); |
8177a420 AA |
4601 | #ifdef CONFIG_NUMA_BALANCING |
4602 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
4603 | pgdat->numabalancing_migrate_nr_pages = 0; | |
4604 | pgdat->numabalancing_migrate_next_window = jiffies; | |
4605 | #endif | |
1da177e4 | 4606 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 4607 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
52d4b9ac | 4608 | pgdat_page_cgroup_init(pgdat); |
5f63b720 | 4609 | |
1da177e4 LT |
4610 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4611 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 4612 | unsigned long size, realsize, freesize, memmap_pages; |
1da177e4 | 4613 | |
c713216d | 4614 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
9feedc9d | 4615 | realsize = freesize = size - zone_absent_pages_in_node(nid, j, |
c713216d | 4616 | zholes_size); |
1da177e4 | 4617 | |
0e0b864e | 4618 | /* |
9feedc9d | 4619 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
4620 | * is used by this zone for memmap. This affects the watermark |
4621 | * and per-cpu initialisations | |
4622 | */ | |
01cefaef | 4623 | memmap_pages = calc_memmap_size(size, realsize); |
9feedc9d JL |
4624 | if (freesize >= memmap_pages) { |
4625 | freesize -= memmap_pages; | |
5594c8c8 YL |
4626 | if (memmap_pages) |
4627 | printk(KERN_DEBUG | |
4628 | " %s zone: %lu pages used for memmap\n", | |
4629 | zone_names[j], memmap_pages); | |
0e0b864e MG |
4630 | } else |
4631 | printk(KERN_WARNING | |
9feedc9d JL |
4632 | " %s zone: %lu pages exceeds freesize %lu\n", |
4633 | zone_names[j], memmap_pages, freesize); | |
0e0b864e | 4634 | |
6267276f | 4635 | /* Account for reserved pages */ |
9feedc9d JL |
4636 | if (j == 0 && freesize > dma_reserve) { |
4637 | freesize -= dma_reserve; | |
d903ef9f | 4638 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4639 | zone_names[0], dma_reserve); |
0e0b864e MG |
4640 | } |
4641 | ||
98d2b0eb | 4642 | if (!is_highmem_idx(j)) |
9feedc9d | 4643 | nr_kernel_pages += freesize; |
01cefaef JL |
4644 | /* Charge for highmem memmap if there are enough kernel pages */ |
4645 | else if (nr_kernel_pages > memmap_pages * 2) | |
4646 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 4647 | nr_all_pages += freesize; |
1da177e4 LT |
4648 | |
4649 | zone->spanned_pages = size; | |
9feedc9d JL |
4650 | zone->present_pages = freesize; |
4651 | /* | |
4652 | * Set an approximate value for lowmem here, it will be adjusted | |
4653 | * when the bootmem allocator frees pages into the buddy system. | |
4654 | * And all highmem pages will be managed by the buddy system. | |
4655 | */ | |
4656 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 4657 | #ifdef CONFIG_NUMA |
d5f541ed | 4658 | zone->node = nid; |
9feedc9d | 4659 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 4660 | / 100; |
9feedc9d | 4661 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 4662 | #endif |
1da177e4 LT |
4663 | zone->name = zone_names[j]; |
4664 | spin_lock_init(&zone->lock); | |
4665 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4666 | zone_seqlock_init(zone); |
1da177e4 | 4667 | zone->zone_pgdat = pgdat; |
1da177e4 | 4668 | |
ed8ece2e | 4669 | zone_pcp_init(zone); |
bea8c150 | 4670 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
4671 | if (!size) |
4672 | continue; | |
4673 | ||
955c1cd7 | 4674 | set_pageblock_order(); |
7c45512d | 4675 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
a2f3aa02 DH |
4676 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4677 | size, MEMMAP_EARLY); | |
718127cc | 4678 | BUG_ON(ret); |
76cdd58e | 4679 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 4680 | zone_start_pfn += size; |
1da177e4 LT |
4681 | } |
4682 | } | |
4683 | ||
577a32f6 | 4684 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 4685 | { |
1da177e4 LT |
4686 | /* Skip empty nodes */ |
4687 | if (!pgdat->node_spanned_pages) | |
4688 | return; | |
4689 | ||
d41dee36 | 4690 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
4691 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
4692 | if (!pgdat->node_mem_map) { | |
e984bb43 | 4693 | unsigned long size, start, end; |
d41dee36 AW |
4694 | struct page *map; |
4695 | ||
e984bb43 BP |
4696 | /* |
4697 | * The zone's endpoints aren't required to be MAX_ORDER | |
4698 | * aligned but the node_mem_map endpoints must be in order | |
4699 | * for the buddy allocator to function correctly. | |
4700 | */ | |
4701 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
4702 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
4703 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
4704 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
4705 | map = alloc_remap(pgdat->node_id, size); |
4706 | if (!map) | |
8f389a99 | 4707 | map = alloc_bootmem_node_nopanic(pgdat, size); |
e984bb43 | 4708 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 4709 | } |
12d810c1 | 4710 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
4711 | /* |
4712 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
4713 | */ | |
c713216d | 4714 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 4715 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 4716 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4717 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 4718 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 4719 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4720 | } |
1da177e4 | 4721 | #endif |
d41dee36 | 4722 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
4723 | } |
4724 | ||
9109fb7b JW |
4725 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
4726 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 4727 | { |
9109fb7b JW |
4728 | pg_data_t *pgdat = NODE_DATA(nid); |
4729 | ||
88fdf75d | 4730 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 4731 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 4732 | |
1da177e4 LT |
4733 | pgdat->node_id = nid; |
4734 | pgdat->node_start_pfn = node_start_pfn; | |
957f822a | 4735 | init_zone_allows_reclaim(nid); |
c713216d | 4736 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
4737 | |
4738 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
4739 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
4740 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
4741 | nid, (unsigned long)pgdat, | |
4742 | (unsigned long)pgdat->node_mem_map); | |
4743 | #endif | |
1da177e4 LT |
4744 | |
4745 | free_area_init_core(pgdat, zones_size, zholes_size); | |
4746 | } | |
4747 | ||
0ee332c1 | 4748 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
4749 | |
4750 | #if MAX_NUMNODES > 1 | |
4751 | /* | |
4752 | * Figure out the number of possible node ids. | |
4753 | */ | |
4754 | static void __init setup_nr_node_ids(void) | |
4755 | { | |
4756 | unsigned int node; | |
4757 | unsigned int highest = 0; | |
4758 | ||
4759 | for_each_node_mask(node, node_possible_map) | |
4760 | highest = node; | |
4761 | nr_node_ids = highest + 1; | |
4762 | } | |
4763 | #else | |
4764 | static inline void setup_nr_node_ids(void) | |
4765 | { | |
4766 | } | |
4767 | #endif | |
4768 | ||
1e01979c TH |
4769 | /** |
4770 | * node_map_pfn_alignment - determine the maximum internode alignment | |
4771 | * | |
4772 | * This function should be called after node map is populated and sorted. | |
4773 | * It calculates the maximum power of two alignment which can distinguish | |
4774 | * all the nodes. | |
4775 | * | |
4776 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
4777 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
4778 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
4779 | * shifted, 1GiB is enough and this function will indicate so. | |
4780 | * | |
4781 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
4782 | * model has fine enough granularity to avoid incorrect mapping for the | |
4783 | * populated node map. | |
4784 | * | |
4785 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
4786 | * requirement (single node). | |
4787 | */ | |
4788 | unsigned long __init node_map_pfn_alignment(void) | |
4789 | { | |
4790 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 4791 | unsigned long start, end, mask; |
1e01979c | 4792 | int last_nid = -1; |
c13291a5 | 4793 | int i, nid; |
1e01979c | 4794 | |
c13291a5 | 4795 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
4796 | if (!start || last_nid < 0 || last_nid == nid) { |
4797 | last_nid = nid; | |
4798 | last_end = end; | |
4799 | continue; | |
4800 | } | |
4801 | ||
4802 | /* | |
4803 | * Start with a mask granular enough to pin-point to the | |
4804 | * start pfn and tick off bits one-by-one until it becomes | |
4805 | * too coarse to separate the current node from the last. | |
4806 | */ | |
4807 | mask = ~((1 << __ffs(start)) - 1); | |
4808 | while (mask && last_end <= (start & (mask << 1))) | |
4809 | mask <<= 1; | |
4810 | ||
4811 | /* accumulate all internode masks */ | |
4812 | accl_mask |= mask; | |
4813 | } | |
4814 | ||
4815 | /* convert mask to number of pages */ | |
4816 | return ~accl_mask + 1; | |
4817 | } | |
4818 | ||
a6af2bc3 | 4819 | /* Find the lowest pfn for a node */ |
b69a7288 | 4820 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 4821 | { |
a6af2bc3 | 4822 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
4823 | unsigned long start_pfn; |
4824 | int i; | |
1abbfb41 | 4825 | |
c13291a5 TH |
4826 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
4827 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 4828 | |
a6af2bc3 MG |
4829 | if (min_pfn == ULONG_MAX) { |
4830 | printk(KERN_WARNING | |
2bc0d261 | 4831 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4832 | return 0; |
4833 | } | |
4834 | ||
4835 | return min_pfn; | |
c713216d MG |
4836 | } |
4837 | ||
4838 | /** | |
4839 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4840 | * | |
4841 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4842 | * add_active_range(). |
c713216d MG |
4843 | */ |
4844 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4845 | { | |
4846 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4847 | } | |
4848 | ||
37b07e41 LS |
4849 | /* |
4850 | * early_calculate_totalpages() | |
4851 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 4852 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 4853 | */ |
484f51f8 | 4854 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 4855 | { |
7e63efef | 4856 | unsigned long totalpages = 0; |
c13291a5 TH |
4857 | unsigned long start_pfn, end_pfn; |
4858 | int i, nid; | |
4859 | ||
4860 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
4861 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 4862 | |
37b07e41 LS |
4863 | totalpages += pages; |
4864 | if (pages) | |
4b0ef1fe | 4865 | node_set_state(nid, N_MEMORY); |
37b07e41 LS |
4866 | } |
4867 | return totalpages; | |
7e63efef MG |
4868 | } |
4869 | ||
2a1e274a MG |
4870 | /* |
4871 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4872 | * is spread evenly between nodes as long as the nodes have enough | |
4873 | * memory. When they don't, some nodes will have more kernelcore than | |
4874 | * others | |
4875 | */ | |
b224ef85 | 4876 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
4877 | { |
4878 | int i, nid; | |
4879 | unsigned long usable_startpfn; | |
4880 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 4881 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 4882 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 4883 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 4884 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
2a1e274a | 4885 | |
7e63efef MG |
4886 | /* |
4887 | * If movablecore was specified, calculate what size of | |
4888 | * kernelcore that corresponds so that memory usable for | |
4889 | * any allocation type is evenly spread. If both kernelcore | |
4890 | * and movablecore are specified, then the value of kernelcore | |
4891 | * will be used for required_kernelcore if it's greater than | |
4892 | * what movablecore would have allowed. | |
4893 | */ | |
4894 | if (required_movablecore) { | |
7e63efef MG |
4895 | unsigned long corepages; |
4896 | ||
4897 | /* | |
4898 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4899 | * was requested by the user | |
4900 | */ | |
4901 | required_movablecore = | |
4902 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4903 | corepages = totalpages - required_movablecore; | |
4904 | ||
4905 | required_kernelcore = max(required_kernelcore, corepages); | |
4906 | } | |
4907 | ||
42f47e27 TC |
4908 | /* |
4909 | * If neither kernelcore/movablecore nor movablemem_map is specified, | |
4910 | * there is no ZONE_MOVABLE. But if movablemem_map is specified, the | |
4911 | * start pfn of ZONE_MOVABLE has been stored in zone_movable_limit[]. | |
4912 | */ | |
4913 | if (!required_kernelcore) { | |
4914 | if (movablemem_map.nr_map) | |
4915 | memcpy(zone_movable_pfn, zone_movable_limit, | |
4916 | sizeof(zone_movable_pfn)); | |
66918dcd | 4917 | goto out; |
42f47e27 | 4918 | } |
2a1e274a MG |
4919 | |
4920 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
4921 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
4922 | ||
4923 | restart: | |
4924 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4925 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 4926 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
4927 | unsigned long start_pfn, end_pfn; |
4928 | ||
2a1e274a MG |
4929 | /* |
4930 | * Recalculate kernelcore_node if the division per node | |
4931 | * now exceeds what is necessary to satisfy the requested | |
4932 | * amount of memory for the kernel | |
4933 | */ | |
4934 | if (required_kernelcore < kernelcore_node) | |
4935 | kernelcore_node = required_kernelcore / usable_nodes; | |
4936 | ||
4937 | /* | |
4938 | * As the map is walked, we track how much memory is usable | |
4939 | * by the kernel using kernelcore_remaining. When it is | |
4940 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4941 | */ | |
4942 | kernelcore_remaining = kernelcore_node; | |
4943 | ||
4944 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 4945 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
4946 | unsigned long size_pages; |
4947 | ||
42f47e27 TC |
4948 | /* |
4949 | * Find more memory for kernelcore in | |
4950 | * [zone_movable_pfn[nid], zone_movable_limit[nid]). | |
4951 | */ | |
c13291a5 | 4952 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
4953 | if (start_pfn >= end_pfn) |
4954 | continue; | |
4955 | ||
42f47e27 TC |
4956 | if (zone_movable_limit[nid]) { |
4957 | end_pfn = min(end_pfn, zone_movable_limit[nid]); | |
4958 | /* No range left for kernelcore in this node */ | |
4959 | if (start_pfn >= end_pfn) { | |
4960 | zone_movable_pfn[nid] = | |
4961 | zone_movable_limit[nid]; | |
4962 | break; | |
4963 | } | |
4964 | } | |
4965 | ||
2a1e274a MG |
4966 | /* Account for what is only usable for kernelcore */ |
4967 | if (start_pfn < usable_startpfn) { | |
4968 | unsigned long kernel_pages; | |
4969 | kernel_pages = min(end_pfn, usable_startpfn) | |
4970 | - start_pfn; | |
4971 | ||
4972 | kernelcore_remaining -= min(kernel_pages, | |
4973 | kernelcore_remaining); | |
4974 | required_kernelcore -= min(kernel_pages, | |
4975 | required_kernelcore); | |
4976 | ||
4977 | /* Continue if range is now fully accounted */ | |
4978 | if (end_pfn <= usable_startpfn) { | |
4979 | ||
4980 | /* | |
4981 | * Push zone_movable_pfn to the end so | |
4982 | * that if we have to rebalance | |
4983 | * kernelcore across nodes, we will | |
4984 | * not double account here | |
4985 | */ | |
4986 | zone_movable_pfn[nid] = end_pfn; | |
4987 | continue; | |
4988 | } | |
4989 | start_pfn = usable_startpfn; | |
4990 | } | |
4991 | ||
4992 | /* | |
4993 | * The usable PFN range for ZONE_MOVABLE is from | |
4994 | * start_pfn->end_pfn. Calculate size_pages as the | |
4995 | * number of pages used as kernelcore | |
4996 | */ | |
4997 | size_pages = end_pfn - start_pfn; | |
4998 | if (size_pages > kernelcore_remaining) | |
4999 | size_pages = kernelcore_remaining; | |
5000 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5001 | ||
5002 | /* | |
5003 | * Some kernelcore has been met, update counts and | |
5004 | * break if the kernelcore for this node has been | |
5005 | * satisified | |
5006 | */ | |
5007 | required_kernelcore -= min(required_kernelcore, | |
5008 | size_pages); | |
5009 | kernelcore_remaining -= size_pages; | |
5010 | if (!kernelcore_remaining) | |
5011 | break; | |
5012 | } | |
5013 | } | |
5014 | ||
5015 | /* | |
5016 | * If there is still required_kernelcore, we do another pass with one | |
5017 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5018 | * along on the nodes that still have memory until kernelcore is | |
5019 | * satisified | |
5020 | */ | |
5021 | usable_nodes--; | |
5022 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5023 | goto restart; | |
5024 | ||
42f47e27 | 5025 | out: |
2a1e274a MG |
5026 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
5027 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5028 | zone_movable_pfn[nid] = | |
5029 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5030 | |
66918dcd | 5031 | /* restore the node_state */ |
4b0ef1fe | 5032 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5033 | } |
5034 | ||
4b0ef1fe LJ |
5035 | /* Any regular or high memory on that node ? */ |
5036 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5037 | { |
37b07e41 LS |
5038 | enum zone_type zone_type; |
5039 | ||
4b0ef1fe LJ |
5040 | if (N_MEMORY == N_NORMAL_MEMORY) |
5041 | return; | |
5042 | ||
5043 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5044 | struct zone *zone = &pgdat->node_zones[zone_type]; |
d0048b0e | 5045 | if (zone->present_pages) { |
4b0ef1fe LJ |
5046 | node_set_state(nid, N_HIGH_MEMORY); |
5047 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5048 | zone_type <= ZONE_NORMAL) | |
5049 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5050 | break; |
5051 | } | |
37b07e41 | 5052 | } |
37b07e41 LS |
5053 | } |
5054 | ||
c713216d MG |
5055 | /** |
5056 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 5057 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
5058 | * |
5059 | * This will call free_area_init_node() for each active node in the system. | |
5060 | * Using the page ranges provided by add_active_range(), the size of each | |
5061 | * zone in each node and their holes is calculated. If the maximum PFN | |
5062 | * between two adjacent zones match, it is assumed that the zone is empty. | |
5063 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
5064 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
5065 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
5066 | * at arch_max_dma_pfn. | |
5067 | */ | |
5068 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
5069 | { | |
c13291a5 TH |
5070 | unsigned long start_pfn, end_pfn; |
5071 | int i, nid; | |
a6af2bc3 | 5072 | |
c713216d MG |
5073 | /* Record where the zone boundaries are */ |
5074 | memset(arch_zone_lowest_possible_pfn, 0, | |
5075 | sizeof(arch_zone_lowest_possible_pfn)); | |
5076 | memset(arch_zone_highest_possible_pfn, 0, | |
5077 | sizeof(arch_zone_highest_possible_pfn)); | |
5078 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
5079 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
5080 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
5081 | if (i == ZONE_MOVABLE) |
5082 | continue; | |
c713216d MG |
5083 | arch_zone_lowest_possible_pfn[i] = |
5084 | arch_zone_highest_possible_pfn[i-1]; | |
5085 | arch_zone_highest_possible_pfn[i] = | |
5086 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
5087 | } | |
2a1e274a MG |
5088 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
5089 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
5090 | ||
5091 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
5092 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
6981ec31 TC |
5093 | find_usable_zone_for_movable(); |
5094 | sanitize_zone_movable_limit(); | |
b224ef85 | 5095 | find_zone_movable_pfns_for_nodes(); |
c713216d | 5096 | |
c713216d | 5097 | /* Print out the zone ranges */ |
a62e2f4f | 5098 | printk("Zone ranges:\n"); |
2a1e274a MG |
5099 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5100 | if (i == ZONE_MOVABLE) | |
5101 | continue; | |
155cbfc8 | 5102 | printk(KERN_CONT " %-8s ", zone_names[i]); |
72f0ba02 DR |
5103 | if (arch_zone_lowest_possible_pfn[i] == |
5104 | arch_zone_highest_possible_pfn[i]) | |
155cbfc8 | 5105 | printk(KERN_CONT "empty\n"); |
72f0ba02 | 5106 | else |
a62e2f4f BH |
5107 | printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n", |
5108 | arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT, | |
5109 | (arch_zone_highest_possible_pfn[i] | |
5110 | << PAGE_SHIFT) - 1); | |
2a1e274a MG |
5111 | } |
5112 | ||
5113 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
a62e2f4f | 5114 | printk("Movable zone start for each node\n"); |
2a1e274a MG |
5115 | for (i = 0; i < MAX_NUMNODES; i++) { |
5116 | if (zone_movable_pfn[i]) | |
a62e2f4f BH |
5117 | printk(" Node %d: %#010lx\n", i, |
5118 | zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 5119 | } |
c713216d | 5120 | |
f2d52fe5 | 5121 | /* Print out the early node map */ |
a62e2f4f | 5122 | printk("Early memory node ranges\n"); |
c13291a5 | 5123 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
a62e2f4f BH |
5124 | printk(" node %3d: [mem %#010lx-%#010lx]\n", nid, |
5125 | start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
5126 | |
5127 | /* Initialise every node */ | |
708614e6 | 5128 | mminit_verify_pageflags_layout(); |
8ef82866 | 5129 | setup_nr_node_ids(); |
c713216d MG |
5130 | for_each_online_node(nid) { |
5131 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 5132 | free_area_init_node(nid, NULL, |
c713216d | 5133 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
5134 | |
5135 | /* Any memory on that node */ | |
5136 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
5137 | node_set_state(nid, N_MEMORY); |
5138 | check_for_memory(pgdat, nid); | |
c713216d MG |
5139 | } |
5140 | } | |
2a1e274a | 5141 | |
7e63efef | 5142 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
5143 | { |
5144 | unsigned long long coremem; | |
5145 | if (!p) | |
5146 | return -EINVAL; | |
5147 | ||
5148 | coremem = memparse(p, &p); | |
7e63efef | 5149 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 5150 | |
7e63efef | 5151 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
5152 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
5153 | ||
5154 | return 0; | |
5155 | } | |
ed7ed365 | 5156 | |
7e63efef MG |
5157 | /* |
5158 | * kernelcore=size sets the amount of memory for use for allocations that | |
5159 | * cannot be reclaimed or migrated. | |
5160 | */ | |
5161 | static int __init cmdline_parse_kernelcore(char *p) | |
5162 | { | |
5163 | return cmdline_parse_core(p, &required_kernelcore); | |
5164 | } | |
5165 | ||
5166 | /* | |
5167 | * movablecore=size sets the amount of memory for use for allocations that | |
5168 | * can be reclaimed or migrated. | |
5169 | */ | |
5170 | static int __init cmdline_parse_movablecore(char *p) | |
5171 | { | |
5172 | return cmdline_parse_core(p, &required_movablecore); | |
5173 | } | |
5174 | ||
ed7ed365 | 5175 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 5176 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 5177 | |
27168d38 TC |
5178 | /** |
5179 | * movablemem_map_overlap() - Check if a range overlaps movablemem_map.map[]. | |
5180 | * @start_pfn: start pfn of the range to be checked | |
5181 | * @end_pfn: end pfn of the range to be checked (exclusive) | |
5182 | * | |
5183 | * This function checks if a given memory range [start_pfn, end_pfn) overlaps | |
5184 | * the movablemem_map.map[] array. | |
5185 | * | |
5186 | * Return: index of the first overlapped element in movablemem_map.map[] | |
5187 | * or -1 if they don't overlap each other. | |
5188 | */ | |
5189 | int __init movablemem_map_overlap(unsigned long start_pfn, | |
5190 | unsigned long end_pfn) | |
5191 | { | |
5192 | int overlap; | |
5193 | ||
5194 | if (!movablemem_map.nr_map) | |
5195 | return -1; | |
5196 | ||
5197 | for (overlap = 0; overlap < movablemem_map.nr_map; overlap++) | |
5198 | if (start_pfn < movablemem_map.map[overlap].end_pfn) | |
5199 | break; | |
5200 | ||
5201 | if (overlap == movablemem_map.nr_map || | |
5202 | end_pfn <= movablemem_map.map[overlap].start_pfn) | |
5203 | return -1; | |
5204 | ||
5205 | return overlap; | |
5206 | } | |
5207 | ||
34b71f1e TC |
5208 | /** |
5209 | * insert_movablemem_map - Insert a memory range in to movablemem_map.map. | |
5210 | * @start_pfn: start pfn of the range | |
5211 | * @end_pfn: end pfn of the range | |
5212 | * | |
5213 | * This function will also merge the overlapped ranges, and sort the array | |
5214 | * by start_pfn in monotonic increasing order. | |
5215 | */ | |
27168d38 TC |
5216 | void __init insert_movablemem_map(unsigned long start_pfn, |
5217 | unsigned long end_pfn) | |
34b71f1e TC |
5218 | { |
5219 | int pos, overlap; | |
5220 | ||
5221 | /* | |
5222 | * pos will be at the 1st overlapped range, or the position | |
5223 | * where the element should be inserted. | |
5224 | */ | |
5225 | for (pos = 0; pos < movablemem_map.nr_map; pos++) | |
5226 | if (start_pfn <= movablemem_map.map[pos].end_pfn) | |
5227 | break; | |
5228 | ||
5229 | /* If there is no overlapped range, just insert the element. */ | |
5230 | if (pos == movablemem_map.nr_map || | |
5231 | end_pfn < movablemem_map.map[pos].start_pfn) { | |
5232 | /* | |
5233 | * If pos is not the end of array, we need to move all | |
5234 | * the rest elements backward. | |
5235 | */ | |
5236 | if (pos < movablemem_map.nr_map) | |
5237 | memmove(&movablemem_map.map[pos+1], | |
5238 | &movablemem_map.map[pos], | |
5239 | sizeof(struct movablemem_entry) * | |
5240 | (movablemem_map.nr_map - pos)); | |
5241 | movablemem_map.map[pos].start_pfn = start_pfn; | |
5242 | movablemem_map.map[pos].end_pfn = end_pfn; | |
5243 | movablemem_map.nr_map++; | |
5244 | return; | |
5245 | } | |
5246 | ||
5247 | /* overlap will be at the last overlapped range */ | |
5248 | for (overlap = pos + 1; overlap < movablemem_map.nr_map; overlap++) | |
5249 | if (end_pfn < movablemem_map.map[overlap].start_pfn) | |
5250 | break; | |
5251 | ||
5252 | /* | |
5253 | * If there are more ranges overlapped, we need to merge them, | |
5254 | * and move the rest elements forward. | |
5255 | */ | |
5256 | overlap--; | |
5257 | movablemem_map.map[pos].start_pfn = min(start_pfn, | |
5258 | movablemem_map.map[pos].start_pfn); | |
5259 | movablemem_map.map[pos].end_pfn = max(end_pfn, | |
5260 | movablemem_map.map[overlap].end_pfn); | |
5261 | ||
5262 | if (pos != overlap && overlap + 1 != movablemem_map.nr_map) | |
5263 | memmove(&movablemem_map.map[pos+1], | |
5264 | &movablemem_map.map[overlap+1], | |
5265 | sizeof(struct movablemem_entry) * | |
5266 | (movablemem_map.nr_map - overlap - 1)); | |
5267 | ||
5268 | movablemem_map.nr_map -= overlap - pos; | |
5269 | } | |
5270 | ||
5271 | /** | |
5272 | * movablemem_map_add_region - Add a memory range into movablemem_map. | |
5273 | * @start: physical start address of range | |
5274 | * @end: physical end address of range | |
5275 | * | |
5276 | * This function transform the physical address into pfn, and then add the | |
5277 | * range into movablemem_map by calling insert_movablemem_map(). | |
5278 | */ | |
5279 | static void __init movablemem_map_add_region(u64 start, u64 size) | |
5280 | { | |
5281 | unsigned long start_pfn, end_pfn; | |
5282 | ||
5283 | /* In case size == 0 or start + size overflows */ | |
5284 | if (start + size <= start) | |
5285 | return; | |
5286 | ||
5287 | if (movablemem_map.nr_map >= ARRAY_SIZE(movablemem_map.map)) { | |
5288 | pr_err("movablemem_map: too many entries;" | |
5289 | " ignoring [mem %#010llx-%#010llx]\n", | |
5290 | (unsigned long long) start, | |
5291 | (unsigned long long) (start + size - 1)); | |
5292 | return; | |
5293 | } | |
5294 | ||
5295 | start_pfn = PFN_DOWN(start); | |
5296 | end_pfn = PFN_UP(start + size); | |
5297 | insert_movablemem_map(start_pfn, end_pfn); | |
5298 | } | |
5299 | ||
5300 | /* | |
5301 | * cmdline_parse_movablemem_map - Parse boot option movablemem_map. | |
5302 | * @p: The boot option of the following format: | |
5303 | * movablemem_map=nn[KMG]@ss[KMG] | |
5304 | * | |
5305 | * This option sets the memory range [ss, ss+nn) to be used as movable memory. | |
5306 | * | |
5307 | * Return: 0 on success or -EINVAL on failure. | |
5308 | */ | |
5309 | static int __init cmdline_parse_movablemem_map(char *p) | |
5310 | { | |
5311 | char *oldp; | |
5312 | u64 start_at, mem_size; | |
5313 | ||
5314 | if (!p) | |
5315 | goto err; | |
5316 | ||
5317 | oldp = p; | |
5318 | mem_size = memparse(p, &p); | |
5319 | if (p == oldp) | |
5320 | goto err; | |
5321 | ||
5322 | if (*p == '@') { | |
5323 | oldp = ++p; | |
5324 | start_at = memparse(p, &p); | |
5325 | if (p == oldp || *p != '\0') | |
5326 | goto err; | |
5327 | ||
5328 | movablemem_map_add_region(start_at, mem_size); | |
5329 | return 0; | |
5330 | } | |
5331 | err: | |
5332 | return -EINVAL; | |
5333 | } | |
5334 | early_param("movablemem_map", cmdline_parse_movablemem_map); | |
5335 | ||
0ee332c1 | 5336 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5337 | |
0e0b864e | 5338 | /** |
88ca3b94 RD |
5339 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
5340 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
5341 | * |
5342 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
5343 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
5344 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
5345 | * function may optionally be used to account for unfreeable pages in the |
5346 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
5347 | * smaller per-cpu batchsize. | |
0e0b864e MG |
5348 | */ |
5349 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
5350 | { | |
5351 | dma_reserve = new_dma_reserve; | |
5352 | } | |
5353 | ||
1da177e4 LT |
5354 | void __init free_area_init(unsigned long *zones_size) |
5355 | { | |
9109fb7b | 5356 | free_area_init_node(0, zones_size, |
1da177e4 LT |
5357 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
5358 | } | |
1da177e4 | 5359 | |
1da177e4 LT |
5360 | static int page_alloc_cpu_notify(struct notifier_block *self, |
5361 | unsigned long action, void *hcpu) | |
5362 | { | |
5363 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 5364 | |
8bb78442 | 5365 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 5366 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
5367 | drain_pages(cpu); |
5368 | ||
5369 | /* | |
5370 | * Spill the event counters of the dead processor | |
5371 | * into the current processors event counters. | |
5372 | * This artificially elevates the count of the current | |
5373 | * processor. | |
5374 | */ | |
f8891e5e | 5375 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
5376 | |
5377 | /* | |
5378 | * Zero the differential counters of the dead processor | |
5379 | * so that the vm statistics are consistent. | |
5380 | * | |
5381 | * This is only okay since the processor is dead and cannot | |
5382 | * race with what we are doing. | |
5383 | */ | |
2244b95a | 5384 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
5385 | } |
5386 | return NOTIFY_OK; | |
5387 | } | |
1da177e4 LT |
5388 | |
5389 | void __init page_alloc_init(void) | |
5390 | { | |
5391 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
5392 | } | |
5393 | ||
cb45b0e9 HA |
5394 | /* |
5395 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
5396 | * or min_free_kbytes changes. | |
5397 | */ | |
5398 | static void calculate_totalreserve_pages(void) | |
5399 | { | |
5400 | struct pglist_data *pgdat; | |
5401 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5402 | enum zone_type i, j; |
cb45b0e9 HA |
5403 | |
5404 | for_each_online_pgdat(pgdat) { | |
5405 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
5406 | struct zone *zone = pgdat->node_zones + i; | |
5407 | unsigned long max = 0; | |
5408 | ||
5409 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5410 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5411 | if (zone->lowmem_reserve[j] > max) | |
5412 | max = zone->lowmem_reserve[j]; | |
5413 | } | |
5414 | ||
41858966 MG |
5415 | /* we treat the high watermark as reserved pages. */ |
5416 | max += high_wmark_pages(zone); | |
cb45b0e9 HA |
5417 | |
5418 | if (max > zone->present_pages) | |
5419 | max = zone->present_pages; | |
5420 | reserve_pages += max; | |
ab8fabd4 JW |
5421 | /* |
5422 | * Lowmem reserves are not available to | |
5423 | * GFP_HIGHUSER page cache allocations and | |
5424 | * kswapd tries to balance zones to their high | |
5425 | * watermark. As a result, neither should be | |
5426 | * regarded as dirtyable memory, to prevent a | |
5427 | * situation where reclaim has to clean pages | |
5428 | * in order to balance the zones. | |
5429 | */ | |
5430 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
5431 | } |
5432 | } | |
ab8fabd4 | 5433 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
5434 | totalreserve_pages = reserve_pages; |
5435 | } | |
5436 | ||
1da177e4 LT |
5437 | /* |
5438 | * setup_per_zone_lowmem_reserve - called whenever | |
5439 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
5440 | * has a correct pages reserved value, so an adequate number of | |
5441 | * pages are left in the zone after a successful __alloc_pages(). | |
5442 | */ | |
5443 | static void setup_per_zone_lowmem_reserve(void) | |
5444 | { | |
5445 | struct pglist_data *pgdat; | |
2f6726e5 | 5446 | enum zone_type j, idx; |
1da177e4 | 5447 | |
ec936fc5 | 5448 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
5449 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5450 | struct zone *zone = pgdat->node_zones + j; | |
5451 | unsigned long present_pages = zone->present_pages; | |
5452 | ||
5453 | zone->lowmem_reserve[j] = 0; | |
5454 | ||
2f6726e5 CL |
5455 | idx = j; |
5456 | while (idx) { | |
1da177e4 LT |
5457 | struct zone *lower_zone; |
5458 | ||
2f6726e5 CL |
5459 | idx--; |
5460 | ||
1da177e4 LT |
5461 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
5462 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
5463 | ||
5464 | lower_zone = pgdat->node_zones + idx; | |
5465 | lower_zone->lowmem_reserve[j] = present_pages / | |
5466 | sysctl_lowmem_reserve_ratio[idx]; | |
5467 | present_pages += lower_zone->present_pages; | |
5468 | } | |
5469 | } | |
5470 | } | |
cb45b0e9 HA |
5471 | |
5472 | /* update totalreserve_pages */ | |
5473 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5474 | } |
5475 | ||
cfd3da1e | 5476 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5477 | { |
5478 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5479 | unsigned long lowmem_pages = 0; | |
5480 | struct zone *zone; | |
5481 | unsigned long flags; | |
5482 | ||
5483 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5484 | for_each_zone(zone) { | |
5485 | if (!is_highmem(zone)) | |
5486 | lowmem_pages += zone->present_pages; | |
5487 | } | |
5488 | ||
5489 | for_each_zone(zone) { | |
ac924c60 AM |
5490 | u64 tmp; |
5491 | ||
1125b4e3 | 5492 | spin_lock_irqsave(&zone->lock, flags); |
ac924c60 AM |
5493 | tmp = (u64)pages_min * zone->present_pages; |
5494 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
5495 | if (is_highmem(zone)) { |
5496 | /* | |
669ed175 NP |
5497 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5498 | * need highmem pages, so cap pages_min to a small | |
5499 | * value here. | |
5500 | * | |
41858966 | 5501 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
5502 | * deltas controls asynch page reclaim, and so should |
5503 | * not be capped for highmem. | |
1da177e4 | 5504 | */ |
90ae8d67 | 5505 | unsigned long min_pages; |
1da177e4 LT |
5506 | |
5507 | min_pages = zone->present_pages / 1024; | |
90ae8d67 | 5508 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 5509 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5510 | } else { |
669ed175 NP |
5511 | /* |
5512 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5513 | * proportionate to the zone's size. |
5514 | */ | |
41858966 | 5515 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5516 | } |
5517 | ||
41858966 MG |
5518 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
5519 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
49f223a9 | 5520 | |
56fd56b8 | 5521 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 5522 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5523 | } |
cb45b0e9 HA |
5524 | |
5525 | /* update totalreserve_pages */ | |
5526 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5527 | } |
5528 | ||
cfd3da1e MG |
5529 | /** |
5530 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5531 | * or when memory is hot-{added|removed} | |
5532 | * | |
5533 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5534 | * correctly with respect to min_free_kbytes. | |
5535 | */ | |
5536 | void setup_per_zone_wmarks(void) | |
5537 | { | |
5538 | mutex_lock(&zonelists_mutex); | |
5539 | __setup_per_zone_wmarks(); | |
5540 | mutex_unlock(&zonelists_mutex); | |
5541 | } | |
5542 | ||
55a4462a | 5543 | /* |
556adecb RR |
5544 | * The inactive anon list should be small enough that the VM never has to |
5545 | * do too much work, but large enough that each inactive page has a chance | |
5546 | * to be referenced again before it is swapped out. | |
5547 | * | |
5548 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
5549 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
5550 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
5551 | * the anonymous pages are kept on the inactive list. | |
5552 | * | |
5553 | * total target max | |
5554 | * memory ratio inactive anon | |
5555 | * ------------------------------------- | |
5556 | * 10MB 1 5MB | |
5557 | * 100MB 1 50MB | |
5558 | * 1GB 3 250MB | |
5559 | * 10GB 10 0.9GB | |
5560 | * 100GB 31 3GB | |
5561 | * 1TB 101 10GB | |
5562 | * 10TB 320 32GB | |
5563 | */ | |
1b79acc9 | 5564 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5565 | { |
96cb4df5 | 5566 | unsigned int gb, ratio; |
556adecb | 5567 | |
96cb4df5 MK |
5568 | /* Zone size in gigabytes */ |
5569 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
5570 | if (gb) | |
556adecb | 5571 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5572 | else |
5573 | ratio = 1; | |
556adecb | 5574 | |
96cb4df5 MK |
5575 | zone->inactive_ratio = ratio; |
5576 | } | |
556adecb | 5577 | |
839a4fcc | 5578 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
5579 | { |
5580 | struct zone *zone; | |
5581 | ||
5582 | for_each_zone(zone) | |
5583 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5584 | } |
5585 | ||
1da177e4 LT |
5586 | /* |
5587 | * Initialise min_free_kbytes. | |
5588 | * | |
5589 | * For small machines we want it small (128k min). For large machines | |
5590 | * we want it large (64MB max). But it is not linear, because network | |
5591 | * bandwidth does not increase linearly with machine size. We use | |
5592 | * | |
5593 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
5594 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
5595 | * | |
5596 | * which yields | |
5597 | * | |
5598 | * 16MB: 512k | |
5599 | * 32MB: 724k | |
5600 | * 64MB: 1024k | |
5601 | * 128MB: 1448k | |
5602 | * 256MB: 2048k | |
5603 | * 512MB: 2896k | |
5604 | * 1024MB: 4096k | |
5605 | * 2048MB: 5792k | |
5606 | * 4096MB: 8192k | |
5607 | * 8192MB: 11584k | |
5608 | * 16384MB: 16384k | |
5609 | */ | |
1b79acc9 | 5610 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
5611 | { |
5612 | unsigned long lowmem_kbytes; | |
5613 | ||
5614 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5615 | ||
5616 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
5617 | if (min_free_kbytes < 128) | |
5618 | min_free_kbytes = 128; | |
5619 | if (min_free_kbytes > 65536) | |
5620 | min_free_kbytes = 65536; | |
bc75d33f | 5621 | setup_per_zone_wmarks(); |
a6cccdc3 | 5622 | refresh_zone_stat_thresholds(); |
1da177e4 | 5623 | setup_per_zone_lowmem_reserve(); |
556adecb | 5624 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5625 | return 0; |
5626 | } | |
bc75d33f | 5627 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5628 | |
5629 | /* | |
5630 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
5631 | * that we can call two helper functions whenever min_free_kbytes | |
5632 | * changes. | |
5633 | */ | |
5634 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5635 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5636 | { |
8d65af78 | 5637 | proc_dointvec(table, write, buffer, length, ppos); |
3b1d92c5 | 5638 | if (write) |
bc75d33f | 5639 | setup_per_zone_wmarks(); |
1da177e4 LT |
5640 | return 0; |
5641 | } | |
5642 | ||
9614634f CL |
5643 | #ifdef CONFIG_NUMA |
5644 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5645 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5646 | { |
5647 | struct zone *zone; | |
5648 | int rc; | |
5649 | ||
8d65af78 | 5650 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5651 | if (rc) |
5652 | return rc; | |
5653 | ||
5654 | for_each_zone(zone) | |
8417bba4 | 5655 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
5656 | sysctl_min_unmapped_ratio) / 100; |
5657 | return 0; | |
5658 | } | |
0ff38490 CL |
5659 | |
5660 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5661 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5662 | { |
5663 | struct zone *zone; | |
5664 | int rc; | |
5665 | ||
8d65af78 | 5666 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5667 | if (rc) |
5668 | return rc; | |
5669 | ||
5670 | for_each_zone(zone) | |
5671 | zone->min_slab_pages = (zone->present_pages * | |
5672 | sysctl_min_slab_ratio) / 100; | |
5673 | return 0; | |
5674 | } | |
9614634f CL |
5675 | #endif |
5676 | ||
1da177e4 LT |
5677 | /* |
5678 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5679 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5680 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5681 | * | |
5682 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5683 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5684 | * if in function of the boot time zone sizes. |
5685 | */ | |
5686 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5687 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5688 | { |
8d65af78 | 5689 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5690 | setup_per_zone_lowmem_reserve(); |
5691 | return 0; | |
5692 | } | |
5693 | ||
8ad4b1fb RS |
5694 | /* |
5695 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
5696 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
5697 | * can have before it gets flushed back to buddy allocator. | |
5698 | */ | |
5699 | ||
5700 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5701 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5702 | { |
5703 | struct zone *zone; | |
5704 | unsigned int cpu; | |
5705 | int ret; | |
5706 | ||
8d65af78 | 5707 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
93278814 | 5708 | if (!write || (ret < 0)) |
8ad4b1fb | 5709 | return ret; |
364df0eb | 5710 | for_each_populated_zone(zone) { |
99dcc3e5 | 5711 | for_each_possible_cpu(cpu) { |
8ad4b1fb RS |
5712 | unsigned long high; |
5713 | high = zone->present_pages / percpu_pagelist_fraction; | |
99dcc3e5 CL |
5714 | setup_pagelist_highmark( |
5715 | per_cpu_ptr(zone->pageset, cpu), high); | |
8ad4b1fb RS |
5716 | } |
5717 | } | |
5718 | return 0; | |
5719 | } | |
5720 | ||
f034b5d4 | 5721 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
5722 | |
5723 | #ifdef CONFIG_NUMA | |
5724 | static int __init set_hashdist(char *str) | |
5725 | { | |
5726 | if (!str) | |
5727 | return 0; | |
5728 | hashdist = simple_strtoul(str, &str, 0); | |
5729 | return 1; | |
5730 | } | |
5731 | __setup("hashdist=", set_hashdist); | |
5732 | #endif | |
5733 | ||
5734 | /* | |
5735 | * allocate a large system hash table from bootmem | |
5736 | * - it is assumed that the hash table must contain an exact power-of-2 | |
5737 | * quantity of entries | |
5738 | * - limit is the number of hash buckets, not the total allocation size | |
5739 | */ | |
5740 | void *__init alloc_large_system_hash(const char *tablename, | |
5741 | unsigned long bucketsize, | |
5742 | unsigned long numentries, | |
5743 | int scale, | |
5744 | int flags, | |
5745 | unsigned int *_hash_shift, | |
5746 | unsigned int *_hash_mask, | |
31fe62b9 TB |
5747 | unsigned long low_limit, |
5748 | unsigned long high_limit) | |
1da177e4 | 5749 | { |
31fe62b9 | 5750 | unsigned long long max = high_limit; |
1da177e4 LT |
5751 | unsigned long log2qty, size; |
5752 | void *table = NULL; | |
5753 | ||
5754 | /* allow the kernel cmdline to have a say */ | |
5755 | if (!numentries) { | |
5756 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 5757 | numentries = nr_kernel_pages; |
1da177e4 LT |
5758 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
5759 | numentries >>= 20 - PAGE_SHIFT; | |
5760 | numentries <<= 20 - PAGE_SHIFT; | |
5761 | ||
5762 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
5763 | if (scale > PAGE_SHIFT) | |
5764 | numentries >>= (scale - PAGE_SHIFT); | |
5765 | else | |
5766 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
5767 | |
5768 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
5769 | if (unlikely(flags & HASH_SMALL)) { |
5770 | /* Makes no sense without HASH_EARLY */ | |
5771 | WARN_ON(!(flags & HASH_EARLY)); | |
5772 | if (!(numentries >> *_hash_shift)) { | |
5773 | numentries = 1UL << *_hash_shift; | |
5774 | BUG_ON(!numentries); | |
5775 | } | |
5776 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 5777 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 5778 | } |
6e692ed3 | 5779 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
5780 | |
5781 | /* limit allocation size to 1/16 total memory by default */ | |
5782 | if (max == 0) { | |
5783 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
5784 | do_div(max, bucketsize); | |
5785 | } | |
074b8517 | 5786 | max = min(max, 0x80000000ULL); |
1da177e4 | 5787 | |
31fe62b9 TB |
5788 | if (numentries < low_limit) |
5789 | numentries = low_limit; | |
1da177e4 LT |
5790 | if (numentries > max) |
5791 | numentries = max; | |
5792 | ||
f0d1b0b3 | 5793 | log2qty = ilog2(numentries); |
1da177e4 LT |
5794 | |
5795 | do { | |
5796 | size = bucketsize << log2qty; | |
5797 | if (flags & HASH_EARLY) | |
74768ed8 | 5798 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
5799 | else if (hashdist) |
5800 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
5801 | else { | |
1037b83b ED |
5802 | /* |
5803 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
5804 | * some pages at the end of hash table which |
5805 | * alloc_pages_exact() automatically does | |
1037b83b | 5806 | */ |
264ef8a9 | 5807 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 5808 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
5809 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
5810 | } | |
1da177e4 LT |
5811 | } |
5812 | } while (!table && size > PAGE_SIZE && --log2qty); | |
5813 | ||
5814 | if (!table) | |
5815 | panic("Failed to allocate %s hash table\n", tablename); | |
5816 | ||
f241e660 | 5817 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 5818 | tablename, |
f241e660 | 5819 | (1UL << log2qty), |
f0d1b0b3 | 5820 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
5821 | size); |
5822 | ||
5823 | if (_hash_shift) | |
5824 | *_hash_shift = log2qty; | |
5825 | if (_hash_mask) | |
5826 | *_hash_mask = (1 << log2qty) - 1; | |
5827 | ||
5828 | return table; | |
5829 | } | |
a117e66e | 5830 | |
835c134e MG |
5831 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
5832 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
5833 | unsigned long pfn) | |
5834 | { | |
5835 | #ifdef CONFIG_SPARSEMEM | |
5836 | return __pfn_to_section(pfn)->pageblock_flags; | |
5837 | #else | |
5838 | return zone->pageblock_flags; | |
5839 | #endif /* CONFIG_SPARSEMEM */ | |
5840 | } | |
5841 | ||
5842 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
5843 | { | |
5844 | #ifdef CONFIG_SPARSEMEM | |
5845 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 5846 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 5847 | #else |
c060f943 | 5848 | pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 5849 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5850 | #endif /* CONFIG_SPARSEMEM */ |
5851 | } | |
5852 | ||
5853 | /** | |
d9c23400 | 5854 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
5855 | * @page: The page within the block of interest |
5856 | * @start_bitidx: The first bit of interest to retrieve | |
5857 | * @end_bitidx: The last bit of interest | |
5858 | * returns pageblock_bits flags | |
5859 | */ | |
5860 | unsigned long get_pageblock_flags_group(struct page *page, | |
5861 | int start_bitidx, int end_bitidx) | |
5862 | { | |
5863 | struct zone *zone; | |
5864 | unsigned long *bitmap; | |
5865 | unsigned long pfn, bitidx; | |
5866 | unsigned long flags = 0; | |
5867 | unsigned long value = 1; | |
5868 | ||
5869 | zone = page_zone(page); | |
5870 | pfn = page_to_pfn(page); | |
5871 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5872 | bitidx = pfn_to_bitidx(zone, pfn); | |
5873 | ||
5874 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5875 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
5876 | flags |= value; | |
6220ec78 | 5877 | |
835c134e MG |
5878 | return flags; |
5879 | } | |
5880 | ||
5881 | /** | |
d9c23400 | 5882 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
5883 | * @page: The page within the block of interest |
5884 | * @start_bitidx: The first bit of interest | |
5885 | * @end_bitidx: The last bit of interest | |
5886 | * @flags: The flags to set | |
5887 | */ | |
5888 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
5889 | int start_bitidx, int end_bitidx) | |
5890 | { | |
5891 | struct zone *zone; | |
5892 | unsigned long *bitmap; | |
5893 | unsigned long pfn, bitidx; | |
5894 | unsigned long value = 1; | |
5895 | ||
5896 | zone = page_zone(page); | |
5897 | pfn = page_to_pfn(page); | |
5898 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5899 | bitidx = pfn_to_bitidx(zone, pfn); | |
86051ca5 KH |
5900 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
5901 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
835c134e MG |
5902 | |
5903 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5904 | if (flags & value) | |
5905 | __set_bit(bitidx + start_bitidx, bitmap); | |
5906 | else | |
5907 | __clear_bit(bitidx + start_bitidx, bitmap); | |
5908 | } | |
a5d76b54 KH |
5909 | |
5910 | /* | |
80934513 MK |
5911 | * This function checks whether pageblock includes unmovable pages or not. |
5912 | * If @count is not zero, it is okay to include less @count unmovable pages | |
5913 | * | |
5914 | * PageLRU check wihtout isolation or lru_lock could race so that | |
5915 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't | |
5916 | * expect this function should be exact. | |
a5d76b54 | 5917 | */ |
b023f468 WC |
5918 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
5919 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
5920 | { |
5921 | unsigned long pfn, iter, found; | |
47118af0 MN |
5922 | int mt; |
5923 | ||
49ac8255 KH |
5924 | /* |
5925 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 5926 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
5927 | */ |
5928 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 5929 | return false; |
47118af0 MN |
5930 | mt = get_pageblock_migratetype(page); |
5931 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 5932 | return false; |
49ac8255 KH |
5933 | |
5934 | pfn = page_to_pfn(page); | |
5935 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
5936 | unsigned long check = pfn + iter; | |
5937 | ||
29723fcc | 5938 | if (!pfn_valid_within(check)) |
49ac8255 | 5939 | continue; |
29723fcc | 5940 | |
49ac8255 | 5941 | page = pfn_to_page(check); |
97d255c8 MK |
5942 | /* |
5943 | * We can't use page_count without pin a page | |
5944 | * because another CPU can free compound page. | |
5945 | * This check already skips compound tails of THP | |
5946 | * because their page->_count is zero at all time. | |
5947 | */ | |
5948 | if (!atomic_read(&page->_count)) { | |
49ac8255 KH |
5949 | if (PageBuddy(page)) |
5950 | iter += (1 << page_order(page)) - 1; | |
5951 | continue; | |
5952 | } | |
97d255c8 | 5953 | |
b023f468 WC |
5954 | /* |
5955 | * The HWPoisoned page may be not in buddy system, and | |
5956 | * page_count() is not 0. | |
5957 | */ | |
5958 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
5959 | continue; | |
5960 | ||
49ac8255 KH |
5961 | if (!PageLRU(page)) |
5962 | found++; | |
5963 | /* | |
5964 | * If there are RECLAIMABLE pages, we need to check it. | |
5965 | * But now, memory offline itself doesn't call shrink_slab() | |
5966 | * and it still to be fixed. | |
5967 | */ | |
5968 | /* | |
5969 | * If the page is not RAM, page_count()should be 0. | |
5970 | * we don't need more check. This is an _used_ not-movable page. | |
5971 | * | |
5972 | * The problematic thing here is PG_reserved pages. PG_reserved | |
5973 | * is set to both of a memory hole page and a _used_ kernel | |
5974 | * page at boot. | |
5975 | */ | |
5976 | if (found > count) | |
80934513 | 5977 | return true; |
49ac8255 | 5978 | } |
80934513 | 5979 | return false; |
49ac8255 KH |
5980 | } |
5981 | ||
5982 | bool is_pageblock_removable_nolock(struct page *page) | |
5983 | { | |
656a0706 MH |
5984 | struct zone *zone; |
5985 | unsigned long pfn; | |
687875fb MH |
5986 | |
5987 | /* | |
5988 | * We have to be careful here because we are iterating over memory | |
5989 | * sections which are not zone aware so we might end up outside of | |
5990 | * the zone but still within the section. | |
656a0706 MH |
5991 | * We have to take care about the node as well. If the node is offline |
5992 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 5993 | */ |
656a0706 MH |
5994 | if (!node_online(page_to_nid(page))) |
5995 | return false; | |
5996 | ||
5997 | zone = page_zone(page); | |
5998 | pfn = page_to_pfn(page); | |
5999 | if (zone->zone_start_pfn > pfn || | |
687875fb MH |
6000 | zone->zone_start_pfn + zone->spanned_pages <= pfn) |
6001 | return false; | |
6002 | ||
b023f468 | 6003 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6004 | } |
0c0e6195 | 6005 | |
041d3a8c MN |
6006 | #ifdef CONFIG_CMA |
6007 | ||
6008 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6009 | { | |
6010 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6011 | pageblock_nr_pages) - 1); | |
6012 | } | |
6013 | ||
6014 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6015 | { | |
6016 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6017 | pageblock_nr_pages)); | |
6018 | } | |
6019 | ||
041d3a8c | 6020 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6021 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6022 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6023 | { |
6024 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6025 | unsigned long nr_reclaimed; |
041d3a8c MN |
6026 | unsigned long pfn = start; |
6027 | unsigned int tries = 0; | |
6028 | int ret = 0; | |
6029 | ||
be49a6e1 | 6030 | migrate_prep(); |
041d3a8c | 6031 | |
bb13ffeb | 6032 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6033 | if (fatal_signal_pending(current)) { |
6034 | ret = -EINTR; | |
6035 | break; | |
6036 | } | |
6037 | ||
bb13ffeb MG |
6038 | if (list_empty(&cc->migratepages)) { |
6039 | cc->nr_migratepages = 0; | |
6040 | pfn = isolate_migratepages_range(cc->zone, cc, | |
e46a2879 | 6041 | pfn, end, true); |
041d3a8c MN |
6042 | if (!pfn) { |
6043 | ret = -EINTR; | |
6044 | break; | |
6045 | } | |
6046 | tries = 0; | |
6047 | } else if (++tries == 5) { | |
6048 | ret = ret < 0 ? ret : -EBUSY; | |
6049 | break; | |
6050 | } | |
6051 | ||
beb51eaa MK |
6052 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6053 | &cc->migratepages); | |
6054 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6055 | |
bb13ffeb | 6056 | ret = migrate_pages(&cc->migratepages, |
723a0644 | 6057 | alloc_migrate_target, |
7b2a2d4a MG |
6058 | 0, false, MIGRATE_SYNC, |
6059 | MR_CMA); | |
041d3a8c | 6060 | } |
2a6f5124 SP |
6061 | if (ret < 0) { |
6062 | putback_movable_pages(&cc->migratepages); | |
6063 | return ret; | |
6064 | } | |
6065 | return 0; | |
041d3a8c MN |
6066 | } |
6067 | ||
6068 | /** | |
6069 | * alloc_contig_range() -- tries to allocate given range of pages | |
6070 | * @start: start PFN to allocate | |
6071 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
6072 | * @migratetype: migratetype of the underlaying pageblocks (either |
6073 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
6074 | * in range must have the same migratetype and it must | |
6075 | * be either of the two. | |
041d3a8c MN |
6076 | * |
6077 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
6078 | * aligned, however it's the caller's responsibility to guarantee that | |
6079 | * we are the only thread that changes migrate type of pageblocks the | |
6080 | * pages fall in. | |
6081 | * | |
6082 | * The PFN range must belong to a single zone. | |
6083 | * | |
6084 | * Returns zero on success or negative error code. On success all | |
6085 | * pages which PFN is in [start, end) are allocated for the caller and | |
6086 | * need to be freed with free_contig_range(). | |
6087 | */ | |
0815f3d8 MN |
6088 | int alloc_contig_range(unsigned long start, unsigned long end, |
6089 | unsigned migratetype) | |
041d3a8c | 6090 | { |
041d3a8c MN |
6091 | unsigned long outer_start, outer_end; |
6092 | int ret = 0, order; | |
6093 | ||
bb13ffeb MG |
6094 | struct compact_control cc = { |
6095 | .nr_migratepages = 0, | |
6096 | .order = -1, | |
6097 | .zone = page_zone(pfn_to_page(start)), | |
6098 | .sync = true, | |
6099 | .ignore_skip_hint = true, | |
6100 | }; | |
6101 | INIT_LIST_HEAD(&cc.migratepages); | |
6102 | ||
041d3a8c MN |
6103 | /* |
6104 | * What we do here is we mark all pageblocks in range as | |
6105 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6106 | * have different sizes, and due to the way page allocator | |
6107 | * work, we align the range to biggest of the two pages so | |
6108 | * that page allocator won't try to merge buddies from | |
6109 | * different pageblocks and change MIGRATE_ISOLATE to some | |
6110 | * other migration type. | |
6111 | * | |
6112 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6113 | * migrate the pages from an unaligned range (ie. pages that | |
6114 | * we are interested in). This will put all the pages in | |
6115 | * range back to page allocator as MIGRATE_ISOLATE. | |
6116 | * | |
6117 | * When this is done, we take the pages in range from page | |
6118 | * allocator removing them from the buddy system. This way | |
6119 | * page allocator will never consider using them. | |
6120 | * | |
6121 | * This lets us mark the pageblocks back as | |
6122 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6123 | * aligned range but not in the unaligned, original range are | |
6124 | * put back to page allocator so that buddy can use them. | |
6125 | */ | |
6126 | ||
6127 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
6128 | pfn_max_align_up(end), migratetype, |
6129 | false); | |
041d3a8c | 6130 | if (ret) |
86a595f9 | 6131 | return ret; |
041d3a8c | 6132 | |
bb13ffeb | 6133 | ret = __alloc_contig_migrate_range(&cc, start, end); |
041d3a8c MN |
6134 | if (ret) |
6135 | goto done; | |
6136 | ||
6137 | /* | |
6138 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
6139 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
6140 | * more, all pages in [start, end) are free in page allocator. | |
6141 | * What we are going to do is to allocate all pages from | |
6142 | * [start, end) (that is remove them from page allocator). | |
6143 | * | |
6144 | * The only problem is that pages at the beginning and at the | |
6145 | * end of interesting range may be not aligned with pages that | |
6146 | * page allocator holds, ie. they can be part of higher order | |
6147 | * pages. Because of this, we reserve the bigger range and | |
6148 | * once this is done free the pages we are not interested in. | |
6149 | * | |
6150 | * We don't have to hold zone->lock here because the pages are | |
6151 | * isolated thus they won't get removed from buddy. | |
6152 | */ | |
6153 | ||
6154 | lru_add_drain_all(); | |
6155 | drain_all_pages(); | |
6156 | ||
6157 | order = 0; | |
6158 | outer_start = start; | |
6159 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
6160 | if (++order >= MAX_ORDER) { | |
6161 | ret = -EBUSY; | |
6162 | goto done; | |
6163 | } | |
6164 | outer_start &= ~0UL << order; | |
6165 | } | |
6166 | ||
6167 | /* Make sure the range is really isolated. */ | |
b023f468 | 6168 | if (test_pages_isolated(outer_start, end, false)) { |
041d3a8c MN |
6169 | pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n", |
6170 | outer_start, end); | |
6171 | ret = -EBUSY; | |
6172 | goto done; | |
6173 | } | |
6174 | ||
49f223a9 MS |
6175 | |
6176 | /* Grab isolated pages from freelists. */ | |
bb13ffeb | 6177 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6178 | if (!outer_end) { |
6179 | ret = -EBUSY; | |
6180 | goto done; | |
6181 | } | |
6182 | ||
6183 | /* Free head and tail (if any) */ | |
6184 | if (start != outer_start) | |
6185 | free_contig_range(outer_start, start - outer_start); | |
6186 | if (end != outer_end) | |
6187 | free_contig_range(end, outer_end - end); | |
6188 | ||
6189 | done: | |
6190 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 6191 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
6192 | return ret; |
6193 | } | |
6194 | ||
6195 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
6196 | { | |
bcc2b02f MS |
6197 | unsigned int count = 0; |
6198 | ||
6199 | for (; nr_pages--; pfn++) { | |
6200 | struct page *page = pfn_to_page(pfn); | |
6201 | ||
6202 | count += page_count(page) != 1; | |
6203 | __free_page(page); | |
6204 | } | |
6205 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
6206 | } |
6207 | #endif | |
6208 | ||
4ed7e022 JL |
6209 | #ifdef CONFIG_MEMORY_HOTPLUG |
6210 | static int __meminit __zone_pcp_update(void *data) | |
6211 | { | |
6212 | struct zone *zone = data; | |
6213 | int cpu; | |
6214 | unsigned long batch = zone_batchsize(zone), flags; | |
6215 | ||
6216 | for_each_possible_cpu(cpu) { | |
6217 | struct per_cpu_pageset *pset; | |
6218 | struct per_cpu_pages *pcp; | |
6219 | ||
6220 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6221 | pcp = &pset->pcp; | |
6222 | ||
6223 | local_irq_save(flags); | |
6224 | if (pcp->count > 0) | |
6225 | free_pcppages_bulk(zone, pcp->count, pcp); | |
5a883813 | 6226 | drain_zonestat(zone, pset); |
4ed7e022 JL |
6227 | setup_pageset(pset, batch); |
6228 | local_irq_restore(flags); | |
6229 | } | |
6230 | return 0; | |
6231 | } | |
6232 | ||
6233 | void __meminit zone_pcp_update(struct zone *zone) | |
6234 | { | |
6235 | stop_machine(__zone_pcp_update, zone, NULL); | |
6236 | } | |
6237 | #endif | |
6238 | ||
340175b7 JL |
6239 | void zone_pcp_reset(struct zone *zone) |
6240 | { | |
6241 | unsigned long flags; | |
5a883813 MK |
6242 | int cpu; |
6243 | struct per_cpu_pageset *pset; | |
340175b7 JL |
6244 | |
6245 | /* avoid races with drain_pages() */ | |
6246 | local_irq_save(flags); | |
6247 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
6248 | for_each_online_cpu(cpu) { |
6249 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6250 | drain_zonestat(zone, pset); | |
6251 | } | |
340175b7 JL |
6252 | free_percpu(zone->pageset); |
6253 | zone->pageset = &boot_pageset; | |
6254 | } | |
6255 | local_irq_restore(flags); | |
6256 | } | |
6257 | ||
6dcd73d7 | 6258 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
6259 | /* |
6260 | * All pages in the range must be isolated before calling this. | |
6261 | */ | |
6262 | void | |
6263 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
6264 | { | |
6265 | struct page *page; | |
6266 | struct zone *zone; | |
6267 | int order, i; | |
6268 | unsigned long pfn; | |
6269 | unsigned long flags; | |
6270 | /* find the first valid pfn */ | |
6271 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
6272 | if (pfn_valid(pfn)) | |
6273 | break; | |
6274 | if (pfn == end_pfn) | |
6275 | return; | |
6276 | zone = page_zone(pfn_to_page(pfn)); | |
6277 | spin_lock_irqsave(&zone->lock, flags); | |
6278 | pfn = start_pfn; | |
6279 | while (pfn < end_pfn) { | |
6280 | if (!pfn_valid(pfn)) { | |
6281 | pfn++; | |
6282 | continue; | |
6283 | } | |
6284 | page = pfn_to_page(pfn); | |
b023f468 WC |
6285 | /* |
6286 | * The HWPoisoned page may be not in buddy system, and | |
6287 | * page_count() is not 0. | |
6288 | */ | |
6289 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6290 | pfn++; | |
6291 | SetPageReserved(page); | |
6292 | continue; | |
6293 | } | |
6294 | ||
0c0e6195 KH |
6295 | BUG_ON(page_count(page)); |
6296 | BUG_ON(!PageBuddy(page)); | |
6297 | order = page_order(page); | |
6298 | #ifdef CONFIG_DEBUG_VM | |
6299 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
6300 | pfn, 1 << order, end_pfn); | |
6301 | #endif | |
6302 | list_del(&page->lru); | |
6303 | rmv_page_order(page); | |
6304 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
6305 | for (i = 0; i < (1 << order); i++) |
6306 | SetPageReserved((page+i)); | |
6307 | pfn += (1 << order); | |
6308 | } | |
6309 | spin_unlock_irqrestore(&zone->lock, flags); | |
6310 | } | |
6311 | #endif | |
8d22ba1b WF |
6312 | |
6313 | #ifdef CONFIG_MEMORY_FAILURE | |
6314 | bool is_free_buddy_page(struct page *page) | |
6315 | { | |
6316 | struct zone *zone = page_zone(page); | |
6317 | unsigned long pfn = page_to_pfn(page); | |
6318 | unsigned long flags; | |
6319 | int order; | |
6320 | ||
6321 | spin_lock_irqsave(&zone->lock, flags); | |
6322 | for (order = 0; order < MAX_ORDER; order++) { | |
6323 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
6324 | ||
6325 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
6326 | break; | |
6327 | } | |
6328 | spin_unlock_irqrestore(&zone->lock, flags); | |
6329 | ||
6330 | return order < MAX_ORDER; | |
6331 | } | |
6332 | #endif | |
718a3821 | 6333 | |
51300cef | 6334 | static const struct trace_print_flags pageflag_names[] = { |
718a3821 WF |
6335 | {1UL << PG_locked, "locked" }, |
6336 | {1UL << PG_error, "error" }, | |
6337 | {1UL << PG_referenced, "referenced" }, | |
6338 | {1UL << PG_uptodate, "uptodate" }, | |
6339 | {1UL << PG_dirty, "dirty" }, | |
6340 | {1UL << PG_lru, "lru" }, | |
6341 | {1UL << PG_active, "active" }, | |
6342 | {1UL << PG_slab, "slab" }, | |
6343 | {1UL << PG_owner_priv_1, "owner_priv_1" }, | |
6344 | {1UL << PG_arch_1, "arch_1" }, | |
6345 | {1UL << PG_reserved, "reserved" }, | |
6346 | {1UL << PG_private, "private" }, | |
6347 | {1UL << PG_private_2, "private_2" }, | |
6348 | {1UL << PG_writeback, "writeback" }, | |
6349 | #ifdef CONFIG_PAGEFLAGS_EXTENDED | |
6350 | {1UL << PG_head, "head" }, | |
6351 | {1UL << PG_tail, "tail" }, | |
6352 | #else | |
6353 | {1UL << PG_compound, "compound" }, | |
6354 | #endif | |
6355 | {1UL << PG_swapcache, "swapcache" }, | |
6356 | {1UL << PG_mappedtodisk, "mappedtodisk" }, | |
6357 | {1UL << PG_reclaim, "reclaim" }, | |
718a3821 WF |
6358 | {1UL << PG_swapbacked, "swapbacked" }, |
6359 | {1UL << PG_unevictable, "unevictable" }, | |
6360 | #ifdef CONFIG_MMU | |
6361 | {1UL << PG_mlocked, "mlocked" }, | |
6362 | #endif | |
6363 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED | |
6364 | {1UL << PG_uncached, "uncached" }, | |
6365 | #endif | |
6366 | #ifdef CONFIG_MEMORY_FAILURE | |
6367 | {1UL << PG_hwpoison, "hwpoison" }, | |
be9cd873 GS |
6368 | #endif |
6369 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6370 | {1UL << PG_compound_lock, "compound_lock" }, | |
718a3821 | 6371 | #endif |
718a3821 WF |
6372 | }; |
6373 | ||
6374 | static void dump_page_flags(unsigned long flags) | |
6375 | { | |
6376 | const char *delim = ""; | |
6377 | unsigned long mask; | |
6378 | int i; | |
6379 | ||
51300cef | 6380 | BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS); |
acc50c11 | 6381 | |
718a3821 WF |
6382 | printk(KERN_ALERT "page flags: %#lx(", flags); |
6383 | ||
6384 | /* remove zone id */ | |
6385 | flags &= (1UL << NR_PAGEFLAGS) - 1; | |
6386 | ||
51300cef | 6387 | for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) { |
718a3821 WF |
6388 | |
6389 | mask = pageflag_names[i].mask; | |
6390 | if ((flags & mask) != mask) | |
6391 | continue; | |
6392 | ||
6393 | flags &= ~mask; | |
6394 | printk("%s%s", delim, pageflag_names[i].name); | |
6395 | delim = "|"; | |
6396 | } | |
6397 | ||
6398 | /* check for left over flags */ | |
6399 | if (flags) | |
6400 | printk("%s%#lx", delim, flags); | |
6401 | ||
6402 | printk(")\n"); | |
6403 | } | |
6404 | ||
6405 | void dump_page(struct page *page) | |
6406 | { | |
6407 | printk(KERN_ALERT | |
6408 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", | |
4e9f64c4 | 6409 | page, atomic_read(&page->_count), page_mapcount(page), |
718a3821 WF |
6410 | page->mapping, page->index); |
6411 | dump_page_flags(page->flags); | |
f212ad7c | 6412 | mem_cgroup_print_bad_page(page); |
718a3821 | 6413 | } |