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> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
041d3a8c | 60 | #include <linux/migrate.h> |
c0a32fc5 | 61 | #include <linux/page-debug-flags.h> |
949f7ec5 | 62 | #include <linux/hugetlb.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
1da177e4 | 64 | |
7ee3d4e8 | 65 | #include <asm/sections.h> |
1da177e4 | 66 | #include <asm/tlbflush.h> |
ac924c60 | 67 | #include <asm/div64.h> |
1da177e4 LT |
68 | #include "internal.h" |
69 | ||
c8e251fa CS |
70 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
71 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
72 | ||
72812019 LS |
73 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
74 | DEFINE_PER_CPU(int, numa_node); | |
75 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
76 | #endif | |
77 | ||
7aac7898 LS |
78 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
79 | /* | |
80 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
81 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
82 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
83 | * defined in <linux/topology.h>. | |
84 | */ | |
85 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
86 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
87 | #endif | |
88 | ||
1da177e4 | 89 | /* |
13808910 | 90 | * Array of node states. |
1da177e4 | 91 | */ |
13808910 CL |
92 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
93 | [N_POSSIBLE] = NODE_MASK_ALL, | |
94 | [N_ONLINE] = { { [0] = 1UL } }, | |
95 | #ifndef CONFIG_NUMA | |
96 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
97 | #ifdef CONFIG_HIGHMEM | |
98 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
99 | #endif |
100 | #ifdef CONFIG_MOVABLE_NODE | |
101 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
102 | #endif |
103 | [N_CPU] = { { [0] = 1UL } }, | |
104 | #endif /* NUMA */ | |
105 | }; | |
106 | EXPORT_SYMBOL(node_states); | |
107 | ||
c3d5f5f0 JL |
108 | /* Protect totalram_pages and zone->managed_pages */ |
109 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
110 | ||
6c231b7b | 111 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 112 | unsigned long totalreserve_pages __read_mostly; |
ab8fabd4 JW |
113 | /* |
114 | * When calculating the number of globally allowed dirty pages, there | |
115 | * is a certain number of per-zone reserves that should not be | |
116 | * considered dirtyable memory. This is the sum of those reserves | |
117 | * over all existing zones that contribute dirtyable memory. | |
118 | */ | |
119 | unsigned long dirty_balance_reserve __read_mostly; | |
120 | ||
1b76b02f | 121 | int percpu_pagelist_fraction; |
dcce284a | 122 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 123 | |
452aa699 RW |
124 | #ifdef CONFIG_PM_SLEEP |
125 | /* | |
126 | * The following functions are used by the suspend/hibernate code to temporarily | |
127 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
128 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
129 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
130 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
131 | * guaranteed not to run in parallel with that modification). | |
132 | */ | |
c9e664f1 RW |
133 | |
134 | static gfp_t saved_gfp_mask; | |
135 | ||
136 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
137 | { |
138 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
139 | if (saved_gfp_mask) { |
140 | gfp_allowed_mask = saved_gfp_mask; | |
141 | saved_gfp_mask = 0; | |
142 | } | |
452aa699 RW |
143 | } |
144 | ||
c9e664f1 | 145 | void pm_restrict_gfp_mask(void) |
452aa699 | 146 | { |
452aa699 | 147 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
148 | WARN_ON(saved_gfp_mask); |
149 | saved_gfp_mask = gfp_allowed_mask; | |
150 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 | 151 | } |
f90ac398 MG |
152 | |
153 | bool pm_suspended_storage(void) | |
154 | { | |
155 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) | |
156 | return false; | |
157 | return true; | |
158 | } | |
452aa699 RW |
159 | #endif /* CONFIG_PM_SLEEP */ |
160 | ||
d9c23400 MG |
161 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
162 | int pageblock_order __read_mostly; | |
163 | #endif | |
164 | ||
d98c7a09 | 165 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 166 | |
1da177e4 LT |
167 | /* |
168 | * results with 256, 32 in the lowmem_reserve sysctl: | |
169 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
170 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
171 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
172 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
173 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
174 | * |
175 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
176 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 177 | */ |
2f1b6248 | 178 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 179 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 180 | 256, |
4b51d669 | 181 | #endif |
fb0e7942 | 182 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 183 | 256, |
fb0e7942 | 184 | #endif |
e53ef38d | 185 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 186 | 32, |
e53ef38d | 187 | #endif |
2a1e274a | 188 | 32, |
2f1b6248 | 189 | }; |
1da177e4 LT |
190 | |
191 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 192 | |
15ad7cdc | 193 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 194 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 195 | "DMA", |
4b51d669 | 196 | #endif |
fb0e7942 | 197 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 198 | "DMA32", |
fb0e7942 | 199 | #endif |
2f1b6248 | 200 | "Normal", |
e53ef38d | 201 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 202 | "HighMem", |
e53ef38d | 203 | #endif |
2a1e274a | 204 | "Movable", |
2f1b6248 CL |
205 | }; |
206 | ||
1da177e4 | 207 | int min_free_kbytes = 1024; |
5f12733e | 208 | int user_min_free_kbytes; |
1da177e4 | 209 | |
2c85f51d JB |
210 | static unsigned long __meminitdata nr_kernel_pages; |
211 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 212 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 213 | |
0ee332c1 TH |
214 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
215 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
216 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
217 | static unsigned long __initdata required_kernelcore; | |
218 | static unsigned long __initdata required_movablecore; | |
219 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
220 | ||
221 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
222 | int movable_zone; | |
223 | EXPORT_SYMBOL(movable_zone); | |
224 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 225 | |
418508c1 MS |
226 | #if MAX_NUMNODES > 1 |
227 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 228 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 229 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 230 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
231 | #endif |
232 | ||
9ef9acb0 MG |
233 | int page_group_by_mobility_disabled __read_mostly; |
234 | ||
ee6f509c | 235 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 236 | { |
5d0f3f72 KM |
237 | if (unlikely(page_group_by_mobility_disabled && |
238 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
239 | migratetype = MIGRATE_UNMOVABLE; |
240 | ||
b2a0ac88 MG |
241 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
242 | PB_migrate, PB_migrate_end); | |
243 | } | |
244 | ||
7f33d49a RW |
245 | bool oom_killer_disabled __read_mostly; |
246 | ||
13e7444b | 247 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 248 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 249 | { |
bdc8cb98 DH |
250 | int ret = 0; |
251 | unsigned seq; | |
252 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 253 | unsigned long sp, start_pfn; |
c6a57e19 | 254 | |
bdc8cb98 DH |
255 | do { |
256 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
257 | start_pfn = zone->zone_start_pfn; |
258 | sp = zone->spanned_pages; | |
108bcc96 | 259 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
260 | ret = 1; |
261 | } while (zone_span_seqretry(zone, seq)); | |
262 | ||
b5e6a5a2 CS |
263 | if (ret) |
264 | pr_err("page %lu outside zone [ %lu - %lu ]\n", | |
265 | pfn, start_pfn, start_pfn + sp); | |
266 | ||
bdc8cb98 | 267 | return ret; |
c6a57e19 DH |
268 | } |
269 | ||
270 | static int page_is_consistent(struct zone *zone, struct page *page) | |
271 | { | |
14e07298 | 272 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 273 | return 0; |
1da177e4 | 274 | if (zone != page_zone(page)) |
c6a57e19 DH |
275 | return 0; |
276 | ||
277 | return 1; | |
278 | } | |
279 | /* | |
280 | * Temporary debugging check for pages not lying within a given zone. | |
281 | */ | |
282 | static int bad_range(struct zone *zone, struct page *page) | |
283 | { | |
284 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 285 | return 1; |
c6a57e19 DH |
286 | if (!page_is_consistent(zone, page)) |
287 | return 1; | |
288 | ||
1da177e4 LT |
289 | return 0; |
290 | } | |
13e7444b NP |
291 | #else |
292 | static inline int bad_range(struct zone *zone, struct page *page) | |
293 | { | |
294 | return 0; | |
295 | } | |
296 | #endif | |
297 | ||
224abf92 | 298 | static void bad_page(struct page *page) |
1da177e4 | 299 | { |
d936cf9b HD |
300 | static unsigned long resume; |
301 | static unsigned long nr_shown; | |
302 | static unsigned long nr_unshown; | |
303 | ||
2a7684a2 WF |
304 | /* Don't complain about poisoned pages */ |
305 | if (PageHWPoison(page)) { | |
22b751c3 | 306 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
307 | return; |
308 | } | |
309 | ||
d936cf9b HD |
310 | /* |
311 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
312 | * or allow a steady drip of one report per second. | |
313 | */ | |
314 | if (nr_shown == 60) { | |
315 | if (time_before(jiffies, resume)) { | |
316 | nr_unshown++; | |
317 | goto out; | |
318 | } | |
319 | if (nr_unshown) { | |
1e9e6365 HD |
320 | printk(KERN_ALERT |
321 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
322 | nr_unshown); |
323 | nr_unshown = 0; | |
324 | } | |
325 | nr_shown = 0; | |
326 | } | |
327 | if (nr_shown++ == 0) | |
328 | resume = jiffies + 60 * HZ; | |
329 | ||
1e9e6365 | 330 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 331 | current->comm, page_to_pfn(page)); |
718a3821 | 332 | dump_page(page); |
3dc14741 | 333 | |
4f31888c | 334 | print_modules(); |
1da177e4 | 335 | dump_stack(); |
d936cf9b | 336 | out: |
8cc3b392 | 337 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 338 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 339 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
340 | } |
341 | ||
1da177e4 LT |
342 | /* |
343 | * Higher-order pages are called "compound pages". They are structured thusly: | |
344 | * | |
345 | * The first PAGE_SIZE page is called the "head page". | |
346 | * | |
347 | * The remaining PAGE_SIZE pages are called "tail pages". | |
348 | * | |
6416b9fa WSH |
349 | * All pages have PG_compound set. All tail pages have their ->first_page |
350 | * pointing at the head page. | |
1da177e4 | 351 | * |
41d78ba5 HD |
352 | * The first tail page's ->lru.next holds the address of the compound page's |
353 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
354 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 355 | */ |
d98c7a09 HD |
356 | |
357 | static void free_compound_page(struct page *page) | |
358 | { | |
d85f3385 | 359 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
360 | } |
361 | ||
01ad1c08 | 362 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
363 | { |
364 | int i; | |
365 | int nr_pages = 1 << order; | |
366 | ||
367 | set_compound_page_dtor(page, free_compound_page); | |
368 | set_compound_order(page, order); | |
369 | __SetPageHead(page); | |
370 | for (i = 1; i < nr_pages; i++) { | |
371 | struct page *p = page + i; | |
18229df5 | 372 | __SetPageTail(p); |
58a84aa9 | 373 | set_page_count(p, 0); |
18229df5 AW |
374 | p->first_page = page; |
375 | } | |
376 | } | |
377 | ||
59ff4216 | 378 | /* update __split_huge_page_refcount if you change this function */ |
8cc3b392 | 379 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
380 | { |
381 | int i; | |
382 | int nr_pages = 1 << order; | |
8cc3b392 | 383 | int bad = 0; |
1da177e4 | 384 | |
0bb2c763 | 385 | if (unlikely(compound_order(page) != order)) { |
224abf92 | 386 | bad_page(page); |
8cc3b392 HD |
387 | bad++; |
388 | } | |
1da177e4 | 389 | |
6d777953 | 390 | __ClearPageHead(page); |
8cc3b392 | 391 | |
18229df5 AW |
392 | for (i = 1; i < nr_pages; i++) { |
393 | struct page *p = page + i; | |
1da177e4 | 394 | |
e713a21d | 395 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 396 | bad_page(page); |
8cc3b392 HD |
397 | bad++; |
398 | } | |
d85f3385 | 399 | __ClearPageTail(p); |
1da177e4 | 400 | } |
8cc3b392 HD |
401 | |
402 | return bad; | |
1da177e4 | 403 | } |
1da177e4 | 404 | |
17cf4406 NP |
405 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
406 | { | |
407 | int i; | |
408 | ||
6626c5d5 AM |
409 | /* |
410 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
411 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
412 | */ | |
725d704e | 413 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
414 | for (i = 0; i < (1 << order); i++) |
415 | clear_highpage(page + i); | |
416 | } | |
417 | ||
c0a32fc5 SG |
418 | #ifdef CONFIG_DEBUG_PAGEALLOC |
419 | unsigned int _debug_guardpage_minorder; | |
420 | ||
421 | static int __init debug_guardpage_minorder_setup(char *buf) | |
422 | { | |
423 | unsigned long res; | |
424 | ||
425 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
426 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
427 | return 0; | |
428 | } | |
429 | _debug_guardpage_minorder = res; | |
430 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
431 | return 0; | |
432 | } | |
433 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
434 | ||
435 | static inline void set_page_guard_flag(struct page *page) | |
436 | { | |
437 | __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
438 | } | |
439 | ||
440 | static inline void clear_page_guard_flag(struct page *page) | |
441 | { | |
442 | __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
443 | } | |
444 | #else | |
445 | static inline void set_page_guard_flag(struct page *page) { } | |
446 | static inline void clear_page_guard_flag(struct page *page) { } | |
447 | #endif | |
448 | ||
6aa3001b AM |
449 | static inline void set_page_order(struct page *page, int order) |
450 | { | |
4c21e2f2 | 451 | set_page_private(page, order); |
676165a8 | 452 | __SetPageBuddy(page); |
1da177e4 LT |
453 | } |
454 | ||
455 | static inline void rmv_page_order(struct page *page) | |
456 | { | |
676165a8 | 457 | __ClearPageBuddy(page); |
4c21e2f2 | 458 | set_page_private(page, 0); |
1da177e4 LT |
459 | } |
460 | ||
461 | /* | |
462 | * Locate the struct page for both the matching buddy in our | |
463 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
464 | * | |
465 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
466 | * the following equation: | |
467 | * B2 = B1 ^ (1 << O) | |
468 | * For example, if the starting buddy (buddy2) is #8 its order | |
469 | * 1 buddy is #10: | |
470 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
471 | * | |
472 | * 2) Any buddy B will have an order O+1 parent P which | |
473 | * satisfies the following equation: | |
474 | * P = B & ~(1 << O) | |
475 | * | |
d6e05edc | 476 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 | 477 | */ |
1da177e4 | 478 | static inline unsigned long |
43506fad | 479 | __find_buddy_index(unsigned long page_idx, unsigned int order) |
1da177e4 | 480 | { |
43506fad | 481 | return page_idx ^ (1 << order); |
1da177e4 LT |
482 | } |
483 | ||
484 | /* | |
485 | * This function checks whether a page is free && is the buddy | |
486 | * we can do coalesce a page and its buddy if | |
13e7444b | 487 | * (a) the buddy is not in a hole && |
676165a8 | 488 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
489 | * (c) a page and its buddy have the same order && |
490 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 491 | * |
cf6fe945 WSH |
492 | * For recording whether a page is in the buddy system, we set ->_mapcount |
493 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
494 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
495 | * serialized by zone->lock. | |
1da177e4 | 496 | * |
676165a8 | 497 | * For recording page's order, we use page_private(page). |
1da177e4 | 498 | */ |
cb2b95e1 AW |
499 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
500 | int order) | |
1da177e4 | 501 | { |
14e07298 | 502 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 503 | return 0; |
13e7444b | 504 | |
cb2b95e1 AW |
505 | if (page_zone_id(page) != page_zone_id(buddy)) |
506 | return 0; | |
507 | ||
c0a32fc5 SG |
508 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
509 | VM_BUG_ON(page_count(buddy) != 0); | |
510 | return 1; | |
511 | } | |
512 | ||
cb2b95e1 | 513 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
a3af9c38 | 514 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 515 | return 1; |
676165a8 | 516 | } |
6aa3001b | 517 | return 0; |
1da177e4 LT |
518 | } |
519 | ||
520 | /* | |
521 | * Freeing function for a buddy system allocator. | |
522 | * | |
523 | * The concept of a buddy system is to maintain direct-mapped table | |
524 | * (containing bit values) for memory blocks of various "orders". | |
525 | * The bottom level table contains the map for the smallest allocatable | |
526 | * units of memory (here, pages), and each level above it describes | |
527 | * pairs of units from the levels below, hence, "buddies". | |
528 | * At a high level, all that happens here is marking the table entry | |
529 | * at the bottom level available, and propagating the changes upward | |
530 | * as necessary, plus some accounting needed to play nicely with other | |
531 | * parts of the VM system. | |
532 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
533 | * free pages of length of (1 << order) and marked with _mapcount |
534 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
535 | * field. | |
1da177e4 | 536 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
537 | * other. That is, if we allocate a small block, and both were |
538 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 539 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 540 | * triggers coalescing into a block of larger size. |
1da177e4 | 541 | * |
6d49e352 | 542 | * -- nyc |
1da177e4 LT |
543 | */ |
544 | ||
48db57f8 | 545 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
546 | struct zone *zone, unsigned int order, |
547 | int migratetype) | |
1da177e4 LT |
548 | { |
549 | unsigned long page_idx; | |
6dda9d55 | 550 | unsigned long combined_idx; |
43506fad | 551 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 552 | struct page *buddy; |
1da177e4 | 553 | |
d29bb978 CS |
554 | VM_BUG_ON(!zone_is_initialized(zone)); |
555 | ||
224abf92 | 556 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
557 | if (unlikely(destroy_compound_page(page, order))) |
558 | return; | |
1da177e4 | 559 | |
ed0ae21d MG |
560 | VM_BUG_ON(migratetype == -1); |
561 | ||
1da177e4 LT |
562 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
563 | ||
f2260e6b | 564 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 565 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 566 | |
1da177e4 | 567 | while (order < MAX_ORDER-1) { |
43506fad KC |
568 | buddy_idx = __find_buddy_index(page_idx, order); |
569 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 570 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 571 | break; |
c0a32fc5 SG |
572 | /* |
573 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
574 | * merge with it and move up one order. | |
575 | */ | |
576 | if (page_is_guard(buddy)) { | |
577 | clear_page_guard_flag(buddy); | |
578 | set_page_private(page, 0); | |
d1ce749a BZ |
579 | __mod_zone_freepage_state(zone, 1 << order, |
580 | migratetype); | |
c0a32fc5 SG |
581 | } else { |
582 | list_del(&buddy->lru); | |
583 | zone->free_area[order].nr_free--; | |
584 | rmv_page_order(buddy); | |
585 | } | |
43506fad | 586 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
587 | page = page + (combined_idx - page_idx); |
588 | page_idx = combined_idx; | |
589 | order++; | |
590 | } | |
591 | set_page_order(page, order); | |
6dda9d55 CZ |
592 | |
593 | /* | |
594 | * If this is not the largest possible page, check if the buddy | |
595 | * of the next-highest order is free. If it is, it's possible | |
596 | * that pages are being freed that will coalesce soon. In case, | |
597 | * that is happening, add the free page to the tail of the list | |
598 | * so it's less likely to be used soon and more likely to be merged | |
599 | * as a higher order page | |
600 | */ | |
b7f50cfa | 601 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 602 | struct page *higher_page, *higher_buddy; |
43506fad KC |
603 | combined_idx = buddy_idx & page_idx; |
604 | higher_page = page + (combined_idx - page_idx); | |
605 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 606 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
607 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
608 | list_add_tail(&page->lru, | |
609 | &zone->free_area[order].free_list[migratetype]); | |
610 | goto out; | |
611 | } | |
612 | } | |
613 | ||
614 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
615 | out: | |
1da177e4 LT |
616 | zone->free_area[order].nr_free++; |
617 | } | |
618 | ||
224abf92 | 619 | static inline int free_pages_check(struct page *page) |
1da177e4 | 620 | { |
92be2e33 NP |
621 | if (unlikely(page_mapcount(page) | |
622 | (page->mapping != NULL) | | |
a3af9c38 | 623 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
624 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE) | |
625 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 626 | bad_page(page); |
79f4b7bf | 627 | return 1; |
8cc3b392 | 628 | } |
90572890 | 629 | page_cpupid_reset_last(page); |
79f4b7bf HD |
630 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
631 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
632 | return 0; | |
1da177e4 LT |
633 | } |
634 | ||
635 | /* | |
5f8dcc21 | 636 | * Frees a number of pages from the PCP lists |
1da177e4 | 637 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 638 | * count is the number of pages to free. |
1da177e4 LT |
639 | * |
640 | * If the zone was previously in an "all pages pinned" state then look to | |
641 | * see if this freeing clears that state. | |
642 | * | |
643 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
644 | * pinned" detection logic. | |
645 | */ | |
5f8dcc21 MG |
646 | static void free_pcppages_bulk(struct zone *zone, int count, |
647 | struct per_cpu_pages *pcp) | |
1da177e4 | 648 | { |
5f8dcc21 | 649 | int migratetype = 0; |
a6f9edd6 | 650 | int batch_free = 0; |
72853e29 | 651 | int to_free = count; |
5f8dcc21 | 652 | |
c54ad30c | 653 | spin_lock(&zone->lock); |
1da177e4 | 654 | zone->pages_scanned = 0; |
f2260e6b | 655 | |
72853e29 | 656 | while (to_free) { |
48db57f8 | 657 | struct page *page; |
5f8dcc21 MG |
658 | struct list_head *list; |
659 | ||
660 | /* | |
a6f9edd6 MG |
661 | * Remove pages from lists in a round-robin fashion. A |
662 | * batch_free count is maintained that is incremented when an | |
663 | * empty list is encountered. This is so more pages are freed | |
664 | * off fuller lists instead of spinning excessively around empty | |
665 | * lists | |
5f8dcc21 MG |
666 | */ |
667 | do { | |
a6f9edd6 | 668 | batch_free++; |
5f8dcc21 MG |
669 | if (++migratetype == MIGRATE_PCPTYPES) |
670 | migratetype = 0; | |
671 | list = &pcp->lists[migratetype]; | |
672 | } while (list_empty(list)); | |
48db57f8 | 673 | |
1d16871d NK |
674 | /* This is the only non-empty list. Free them all. */ |
675 | if (batch_free == MIGRATE_PCPTYPES) | |
676 | batch_free = to_free; | |
677 | ||
a6f9edd6 | 678 | do { |
770c8aaa BZ |
679 | int mt; /* migratetype of the to-be-freed page */ |
680 | ||
a6f9edd6 MG |
681 | page = list_entry(list->prev, struct page, lru); |
682 | /* must delete as __free_one_page list manipulates */ | |
683 | list_del(&page->lru); | |
b12c4ad1 | 684 | mt = get_freepage_migratetype(page); |
a7016235 | 685 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
770c8aaa BZ |
686 | __free_one_page(page, zone, 0, mt); |
687 | trace_mm_page_pcpu_drain(page, 0, mt); | |
194159fb | 688 | if (likely(!is_migrate_isolate_page(page))) { |
97d0da22 WC |
689 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1); |
690 | if (is_migrate_cma(mt)) | |
691 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1); | |
692 | } | |
72853e29 | 693 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 694 | } |
c54ad30c | 695 | spin_unlock(&zone->lock); |
1da177e4 LT |
696 | } |
697 | ||
ed0ae21d MG |
698 | static void free_one_page(struct zone *zone, struct page *page, int order, |
699 | int migratetype) | |
1da177e4 | 700 | { |
006d22d9 | 701 | spin_lock(&zone->lock); |
006d22d9 | 702 | zone->pages_scanned = 0; |
f2260e6b | 703 | |
ed0ae21d | 704 | __free_one_page(page, zone, order, migratetype); |
194159fb | 705 | if (unlikely(!is_migrate_isolate(migratetype))) |
d1ce749a | 706 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
006d22d9 | 707 | spin_unlock(&zone->lock); |
48db57f8 NP |
708 | } |
709 | ||
ec95f53a | 710 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 711 | { |
1da177e4 | 712 | int i; |
8cc3b392 | 713 | int bad = 0; |
1da177e4 | 714 | |
b413d48a | 715 | trace_mm_page_free(page, order); |
b1eeab67 VN |
716 | kmemcheck_free_shadow(page, order); |
717 | ||
8dd60a3a AA |
718 | if (PageAnon(page)) |
719 | page->mapping = NULL; | |
720 | for (i = 0; i < (1 << order); i++) | |
721 | bad += free_pages_check(page + i); | |
8cc3b392 | 722 | if (bad) |
ec95f53a | 723 | return false; |
689bcebf | 724 | |
3ac7fe5a | 725 | if (!PageHighMem(page)) { |
b8af2941 PK |
726 | debug_check_no_locks_freed(page_address(page), |
727 | PAGE_SIZE << order); | |
3ac7fe5a TG |
728 | debug_check_no_obj_freed(page_address(page), |
729 | PAGE_SIZE << order); | |
730 | } | |
dafb1367 | 731 | arch_free_page(page, order); |
48db57f8 | 732 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 733 | |
ec95f53a KM |
734 | return true; |
735 | } | |
736 | ||
737 | static void __free_pages_ok(struct page *page, unsigned int order) | |
738 | { | |
739 | unsigned long flags; | |
95e34412 | 740 | int migratetype; |
ec95f53a KM |
741 | |
742 | if (!free_pages_prepare(page, order)) | |
743 | return; | |
744 | ||
c54ad30c | 745 | local_irq_save(flags); |
f8891e5e | 746 | __count_vm_events(PGFREE, 1 << order); |
95e34412 MK |
747 | migratetype = get_pageblock_migratetype(page); |
748 | set_freepage_migratetype(page, migratetype); | |
749 | free_one_page(page_zone(page), page, order, migratetype); | |
c54ad30c | 750 | local_irq_restore(flags); |
1da177e4 LT |
751 | } |
752 | ||
170a5a7e | 753 | void __init __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 | 754 | { |
c3993076 | 755 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 756 | struct page *p = page; |
c3993076 | 757 | unsigned int loop; |
a226f6c8 | 758 | |
e2d0bd2b YL |
759 | prefetchw(p); |
760 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
761 | prefetchw(p + 1); | |
c3993076 JW |
762 | __ClearPageReserved(p); |
763 | set_page_count(p, 0); | |
a226f6c8 | 764 | } |
e2d0bd2b YL |
765 | __ClearPageReserved(p); |
766 | set_page_count(p, 0); | |
c3993076 | 767 | |
e2d0bd2b | 768 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
769 | set_page_refcounted(page); |
770 | __free_pages(page, order); | |
a226f6c8 DH |
771 | } |
772 | ||
47118af0 | 773 | #ifdef CONFIG_CMA |
9cf510a5 | 774 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
775 | void __init init_cma_reserved_pageblock(struct page *page) |
776 | { | |
777 | unsigned i = pageblock_nr_pages; | |
778 | struct page *p = page; | |
779 | ||
780 | do { | |
781 | __ClearPageReserved(p); | |
782 | set_page_count(p, 0); | |
783 | } while (++p, --i); | |
784 | ||
785 | set_page_refcounted(page); | |
786 | set_pageblock_migratetype(page, MIGRATE_CMA); | |
787 | __free_pages(page, pageblock_order); | |
3dcc0571 | 788 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
789 | } |
790 | #endif | |
1da177e4 LT |
791 | |
792 | /* | |
793 | * The order of subdivision here is critical for the IO subsystem. | |
794 | * Please do not alter this order without good reasons and regression | |
795 | * testing. Specifically, as large blocks of memory are subdivided, | |
796 | * the order in which smaller blocks are delivered depends on the order | |
797 | * they're subdivided in this function. This is the primary factor | |
798 | * influencing the order in which pages are delivered to the IO | |
799 | * subsystem according to empirical testing, and this is also justified | |
800 | * by considering the behavior of a buddy system containing a single | |
801 | * large block of memory acted on by a series of small allocations. | |
802 | * This behavior is a critical factor in sglist merging's success. | |
803 | * | |
6d49e352 | 804 | * -- nyc |
1da177e4 | 805 | */ |
085cc7d5 | 806 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
807 | int low, int high, struct free_area *area, |
808 | int migratetype) | |
1da177e4 LT |
809 | { |
810 | unsigned long size = 1 << high; | |
811 | ||
812 | while (high > low) { | |
813 | area--; | |
814 | high--; | |
815 | size >>= 1; | |
725d704e | 816 | VM_BUG_ON(bad_range(zone, &page[size])); |
c0a32fc5 SG |
817 | |
818 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
819 | if (high < debug_guardpage_minorder()) { | |
820 | /* | |
821 | * Mark as guard pages (or page), that will allow to | |
822 | * merge back to allocator when buddy will be freed. | |
823 | * Corresponding page table entries will not be touched, | |
824 | * pages will stay not present in virtual address space | |
825 | */ | |
826 | INIT_LIST_HEAD(&page[size].lru); | |
827 | set_page_guard_flag(&page[size]); | |
828 | set_page_private(&page[size], high); | |
829 | /* Guard pages are not available for any usage */ | |
d1ce749a BZ |
830 | __mod_zone_freepage_state(zone, -(1 << high), |
831 | migratetype); | |
c0a32fc5 SG |
832 | continue; |
833 | } | |
834 | #endif | |
b2a0ac88 | 835 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
836 | area->nr_free++; |
837 | set_page_order(&page[size], high); | |
838 | } | |
1da177e4 LT |
839 | } |
840 | ||
1da177e4 LT |
841 | /* |
842 | * This page is about to be returned from the page allocator | |
843 | */ | |
2a7684a2 | 844 | static inline int check_new_page(struct page *page) |
1da177e4 | 845 | { |
92be2e33 NP |
846 | if (unlikely(page_mapcount(page) | |
847 | (page->mapping != NULL) | | |
a3af9c38 | 848 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
849 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP) | |
850 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 851 | bad_page(page); |
689bcebf | 852 | return 1; |
8cc3b392 | 853 | } |
2a7684a2 WF |
854 | return 0; |
855 | } | |
856 | ||
857 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
858 | { | |
859 | int i; | |
860 | ||
861 | for (i = 0; i < (1 << order); i++) { | |
862 | struct page *p = page + i; | |
863 | if (unlikely(check_new_page(p))) | |
864 | return 1; | |
865 | } | |
689bcebf | 866 | |
4c21e2f2 | 867 | set_page_private(page, 0); |
7835e98b | 868 | set_page_refcounted(page); |
cc102509 NP |
869 | |
870 | arch_alloc_page(page, order); | |
1da177e4 | 871 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
872 | |
873 | if (gfp_flags & __GFP_ZERO) | |
874 | prep_zero_page(page, order, gfp_flags); | |
875 | ||
876 | if (order && (gfp_flags & __GFP_COMP)) | |
877 | prep_compound_page(page, order); | |
878 | ||
689bcebf | 879 | return 0; |
1da177e4 LT |
880 | } |
881 | ||
56fd56b8 MG |
882 | /* |
883 | * Go through the free lists for the given migratetype and remove | |
884 | * the smallest available page from the freelists | |
885 | */ | |
728ec980 MG |
886 | static inline |
887 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
888 | int migratetype) |
889 | { | |
890 | unsigned int current_order; | |
b8af2941 | 891 | struct free_area *area; |
56fd56b8 MG |
892 | struct page *page; |
893 | ||
894 | /* Find a page of the appropriate size in the preferred list */ | |
895 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
896 | area = &(zone->free_area[current_order]); | |
897 | if (list_empty(&area->free_list[migratetype])) | |
898 | continue; | |
899 | ||
900 | page = list_entry(area->free_list[migratetype].next, | |
901 | struct page, lru); | |
902 | list_del(&page->lru); | |
903 | rmv_page_order(page); | |
904 | area->nr_free--; | |
56fd56b8 MG |
905 | expand(zone, page, order, current_order, area, migratetype); |
906 | return page; | |
907 | } | |
908 | ||
909 | return NULL; | |
910 | } | |
911 | ||
912 | ||
b2a0ac88 MG |
913 | /* |
914 | * This array describes the order lists are fallen back to when | |
915 | * the free lists for the desirable migrate type are depleted | |
916 | */ | |
47118af0 MN |
917 | static int fallbacks[MIGRATE_TYPES][4] = { |
918 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
919 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
920 | #ifdef CONFIG_CMA | |
921 | [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
922 | [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */ | |
923 | #else | |
924 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
925 | #endif | |
6d4a4916 | 926 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 927 | #ifdef CONFIG_MEMORY_ISOLATION |
6d4a4916 | 928 | [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 929 | #endif |
b2a0ac88 MG |
930 | }; |
931 | ||
c361be55 MG |
932 | /* |
933 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 934 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
935 | * boundary. If alignment is required, use move_freepages_block() |
936 | */ | |
435b405c | 937 | int move_freepages(struct zone *zone, |
b69a7288 AB |
938 | struct page *start_page, struct page *end_page, |
939 | int migratetype) | |
c361be55 MG |
940 | { |
941 | struct page *page; | |
942 | unsigned long order; | |
d100313f | 943 | int pages_moved = 0; |
c361be55 MG |
944 | |
945 | #ifndef CONFIG_HOLES_IN_ZONE | |
946 | /* | |
947 | * page_zone is not safe to call in this context when | |
948 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
949 | * anyway as we check zone boundaries in move_freepages_block(). | |
950 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 951 | * grouping pages by mobility |
c361be55 MG |
952 | */ |
953 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
954 | #endif | |
955 | ||
956 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
957 | /* Make sure we are not inadvertently changing nodes */ |
958 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
959 | ||
c361be55 MG |
960 | if (!pfn_valid_within(page_to_pfn(page))) { |
961 | page++; | |
962 | continue; | |
963 | } | |
964 | ||
965 | if (!PageBuddy(page)) { | |
966 | page++; | |
967 | continue; | |
968 | } | |
969 | ||
970 | order = page_order(page); | |
84be48d8 KS |
971 | list_move(&page->lru, |
972 | &zone->free_area[order].free_list[migratetype]); | |
95e34412 | 973 | set_freepage_migratetype(page, migratetype); |
c361be55 | 974 | page += 1 << order; |
d100313f | 975 | pages_moved += 1 << order; |
c361be55 MG |
976 | } |
977 | ||
d100313f | 978 | return pages_moved; |
c361be55 MG |
979 | } |
980 | ||
ee6f509c | 981 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 982 | int migratetype) |
c361be55 MG |
983 | { |
984 | unsigned long start_pfn, end_pfn; | |
985 | struct page *start_page, *end_page; | |
986 | ||
987 | start_pfn = page_to_pfn(page); | |
d9c23400 | 988 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 989 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
990 | end_page = start_page + pageblock_nr_pages - 1; |
991 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
992 | |
993 | /* Do not cross zone boundaries */ | |
108bcc96 | 994 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 995 | start_page = page; |
108bcc96 | 996 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
997 | return 0; |
998 | ||
999 | return move_freepages(zone, start_page, end_page, migratetype); | |
1000 | } | |
1001 | ||
2f66a68f MG |
1002 | static void change_pageblock_range(struct page *pageblock_page, |
1003 | int start_order, int migratetype) | |
1004 | { | |
1005 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1006 | ||
1007 | while (nr_pageblocks--) { | |
1008 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1009 | pageblock_page += pageblock_nr_pages; | |
1010 | } | |
1011 | } | |
1012 | ||
fef903ef SB |
1013 | /* |
1014 | * If breaking a large block of pages, move all free pages to the preferred | |
1015 | * allocation list. If falling back for a reclaimable kernel allocation, be | |
1016 | * more aggressive about taking ownership of free pages. | |
1017 | * | |
1018 | * On the other hand, never change migration type of MIGRATE_CMA pageblocks | |
1019 | * nor move CMA pages to different free lists. We don't want unmovable pages | |
1020 | * to be allocated from MIGRATE_CMA areas. | |
1021 | * | |
1022 | * Returns the new migratetype of the pageblock (or the same old migratetype | |
1023 | * if it was unchanged). | |
1024 | */ | |
1025 | static int try_to_steal_freepages(struct zone *zone, struct page *page, | |
1026 | int start_type, int fallback_type) | |
1027 | { | |
1028 | int current_order = page_order(page); | |
1029 | ||
0cbef29a KM |
1030 | /* |
1031 | * When borrowing from MIGRATE_CMA, we need to release the excess | |
1032 | * buddy pages to CMA itself. | |
1033 | */ | |
fef903ef SB |
1034 | if (is_migrate_cma(fallback_type)) |
1035 | return fallback_type; | |
1036 | ||
1037 | /* Take ownership for orders >= pageblock_order */ | |
1038 | if (current_order >= pageblock_order) { | |
1039 | change_pageblock_range(page, current_order, start_type); | |
1040 | return start_type; | |
1041 | } | |
1042 | ||
1043 | if (current_order >= pageblock_order / 2 || | |
1044 | start_type == MIGRATE_RECLAIMABLE || | |
1045 | page_group_by_mobility_disabled) { | |
1046 | int pages; | |
1047 | ||
1048 | pages = move_freepages_block(zone, page, start_type); | |
1049 | ||
1050 | /* Claim the whole block if over half of it is free */ | |
1051 | if (pages >= (1 << (pageblock_order-1)) || | |
1052 | page_group_by_mobility_disabled) { | |
1053 | ||
1054 | set_pageblock_migratetype(page, start_type); | |
1055 | return start_type; | |
1056 | } | |
1057 | ||
1058 | } | |
1059 | ||
1060 | return fallback_type; | |
1061 | } | |
1062 | ||
b2a0ac88 | 1063 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
1064 | static inline struct page * |
1065 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 | 1066 | { |
b8af2941 | 1067 | struct free_area *area; |
b2a0ac88 MG |
1068 | int current_order; |
1069 | struct page *page; | |
fef903ef | 1070 | int migratetype, new_type, i; |
b2a0ac88 MG |
1071 | |
1072 | /* Find the largest possible block of pages in the other list */ | |
1073 | for (current_order = MAX_ORDER-1; current_order >= order; | |
1074 | --current_order) { | |
6d4a4916 | 1075 | for (i = 0;; i++) { |
b2a0ac88 MG |
1076 | migratetype = fallbacks[start_migratetype][i]; |
1077 | ||
56fd56b8 MG |
1078 | /* MIGRATE_RESERVE handled later if necessary */ |
1079 | if (migratetype == MIGRATE_RESERVE) | |
6d4a4916 | 1080 | break; |
e010487d | 1081 | |
b2a0ac88 MG |
1082 | area = &(zone->free_area[current_order]); |
1083 | if (list_empty(&area->free_list[migratetype])) | |
1084 | continue; | |
1085 | ||
1086 | page = list_entry(area->free_list[migratetype].next, | |
1087 | struct page, lru); | |
1088 | area->nr_free--; | |
1089 | ||
fef903ef SB |
1090 | new_type = try_to_steal_freepages(zone, page, |
1091 | start_migratetype, | |
1092 | migratetype); | |
b2a0ac88 MG |
1093 | |
1094 | /* Remove the page from the freelists */ | |
1095 | list_del(&page->lru); | |
1096 | rmv_page_order(page); | |
b2a0ac88 | 1097 | |
47118af0 | 1098 | expand(zone, page, order, current_order, area, |
0cbef29a | 1099 | new_type); |
e0fff1bd | 1100 | |
52c8f6a5 KM |
1101 | trace_mm_page_alloc_extfrag(page, order, current_order, |
1102 | start_migratetype, migratetype, new_type); | |
e0fff1bd | 1103 | |
b2a0ac88 MG |
1104 | return page; |
1105 | } | |
1106 | } | |
1107 | ||
728ec980 | 1108 | return NULL; |
b2a0ac88 MG |
1109 | } |
1110 | ||
56fd56b8 | 1111 | /* |
1da177e4 LT |
1112 | * Do the hard work of removing an element from the buddy allocator. |
1113 | * Call me with the zone->lock already held. | |
1114 | */ | |
b2a0ac88 MG |
1115 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1116 | int migratetype) | |
1da177e4 | 1117 | { |
1da177e4 LT |
1118 | struct page *page; |
1119 | ||
728ec980 | 1120 | retry_reserve: |
56fd56b8 | 1121 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1122 | |
728ec980 | 1123 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 1124 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 1125 | |
728ec980 MG |
1126 | /* |
1127 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1128 | * is used because __rmqueue_smallest is an inline function | |
1129 | * and we want just one call site | |
1130 | */ | |
1131 | if (!page) { | |
1132 | migratetype = MIGRATE_RESERVE; | |
1133 | goto retry_reserve; | |
1134 | } | |
1135 | } | |
1136 | ||
0d3d062a | 1137 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1138 | return page; |
1da177e4 LT |
1139 | } |
1140 | ||
5f63b720 | 1141 | /* |
1da177e4 LT |
1142 | * Obtain a specified number of elements from the buddy allocator, all under |
1143 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1144 | * Returns the number of new pages which were placed at *list. | |
1145 | */ | |
5f63b720 | 1146 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1147 | unsigned long count, struct list_head *list, |
e084b2d9 | 1148 | int migratetype, int cold) |
1da177e4 | 1149 | { |
47118af0 | 1150 | int mt = migratetype, i; |
5f63b720 | 1151 | |
c54ad30c | 1152 | spin_lock(&zone->lock); |
1da177e4 | 1153 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1154 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1155 | if (unlikely(page == NULL)) |
1da177e4 | 1156 | break; |
81eabcbe MG |
1157 | |
1158 | /* | |
1159 | * Split buddy pages returned by expand() are received here | |
1160 | * in physical page order. The page is added to the callers and | |
1161 | * list and the list head then moves forward. From the callers | |
1162 | * perspective, the linked list is ordered by page number in | |
1163 | * some conditions. This is useful for IO devices that can | |
1164 | * merge IO requests if the physical pages are ordered | |
1165 | * properly. | |
1166 | */ | |
e084b2d9 MG |
1167 | if (likely(cold == 0)) |
1168 | list_add(&page->lru, list); | |
1169 | else | |
1170 | list_add_tail(&page->lru, list); | |
47118af0 MN |
1171 | if (IS_ENABLED(CONFIG_CMA)) { |
1172 | mt = get_pageblock_migratetype(page); | |
194159fb | 1173 | if (!is_migrate_cma(mt) && !is_migrate_isolate(mt)) |
47118af0 MN |
1174 | mt = migratetype; |
1175 | } | |
b12c4ad1 | 1176 | set_freepage_migratetype(page, mt); |
81eabcbe | 1177 | list = &page->lru; |
d1ce749a BZ |
1178 | if (is_migrate_cma(mt)) |
1179 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, | |
1180 | -(1 << order)); | |
1da177e4 | 1181 | } |
f2260e6b | 1182 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1183 | spin_unlock(&zone->lock); |
085cc7d5 | 1184 | return i; |
1da177e4 LT |
1185 | } |
1186 | ||
4ae7c039 | 1187 | #ifdef CONFIG_NUMA |
8fce4d8e | 1188 | /* |
4037d452 CL |
1189 | * Called from the vmstat counter updater to drain pagesets of this |
1190 | * currently executing processor on remote nodes after they have | |
1191 | * expired. | |
1192 | * | |
879336c3 CL |
1193 | * Note that this function must be called with the thread pinned to |
1194 | * a single processor. | |
8fce4d8e | 1195 | */ |
4037d452 | 1196 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1197 | { |
4ae7c039 | 1198 | unsigned long flags; |
4037d452 | 1199 | int to_drain; |
998d39cb | 1200 | unsigned long batch; |
4ae7c039 | 1201 | |
4037d452 | 1202 | local_irq_save(flags); |
998d39cb CS |
1203 | batch = ACCESS_ONCE(pcp->batch); |
1204 | if (pcp->count >= batch) | |
1205 | to_drain = batch; | |
4037d452 CL |
1206 | else |
1207 | to_drain = pcp->count; | |
2a13515c KM |
1208 | if (to_drain > 0) { |
1209 | free_pcppages_bulk(zone, to_drain, pcp); | |
1210 | pcp->count -= to_drain; | |
1211 | } | |
4037d452 | 1212 | local_irq_restore(flags); |
4ae7c039 CL |
1213 | } |
1214 | #endif | |
1215 | ||
9f8f2172 CL |
1216 | /* |
1217 | * Drain pages of the indicated processor. | |
1218 | * | |
1219 | * The processor must either be the current processor and the | |
1220 | * thread pinned to the current processor or a processor that | |
1221 | * is not online. | |
1222 | */ | |
1223 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1224 | { |
c54ad30c | 1225 | unsigned long flags; |
1da177e4 | 1226 | struct zone *zone; |
1da177e4 | 1227 | |
ee99c71c | 1228 | for_each_populated_zone(zone) { |
1da177e4 | 1229 | struct per_cpu_pageset *pset; |
3dfa5721 | 1230 | struct per_cpu_pages *pcp; |
1da177e4 | 1231 | |
99dcc3e5 CL |
1232 | local_irq_save(flags); |
1233 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3dfa5721 CL |
1234 | |
1235 | pcp = &pset->pcp; | |
2ff754fa DR |
1236 | if (pcp->count) { |
1237 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1238 | pcp->count = 0; | |
1239 | } | |
3dfa5721 | 1240 | local_irq_restore(flags); |
1da177e4 LT |
1241 | } |
1242 | } | |
1da177e4 | 1243 | |
9f8f2172 CL |
1244 | /* |
1245 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1246 | */ | |
1247 | void drain_local_pages(void *arg) | |
1248 | { | |
1249 | drain_pages(smp_processor_id()); | |
1250 | } | |
1251 | ||
1252 | /* | |
74046494 GBY |
1253 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
1254 | * | |
1255 | * Note that this code is protected against sending an IPI to an offline | |
1256 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
1257 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
1258 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
1259 | * before the call to on_each_cpu_mask(). | |
9f8f2172 CL |
1260 | */ |
1261 | void drain_all_pages(void) | |
1262 | { | |
74046494 GBY |
1263 | int cpu; |
1264 | struct per_cpu_pageset *pcp; | |
1265 | struct zone *zone; | |
1266 | ||
1267 | /* | |
1268 | * Allocate in the BSS so we wont require allocation in | |
1269 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
1270 | */ | |
1271 | static cpumask_t cpus_with_pcps; | |
1272 | ||
1273 | /* | |
1274 | * We don't care about racing with CPU hotplug event | |
1275 | * as offline notification will cause the notified | |
1276 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
1277 | * disables preemption as part of its processing | |
1278 | */ | |
1279 | for_each_online_cpu(cpu) { | |
1280 | bool has_pcps = false; | |
1281 | for_each_populated_zone(zone) { | |
1282 | pcp = per_cpu_ptr(zone->pageset, cpu); | |
1283 | if (pcp->pcp.count) { | |
1284 | has_pcps = true; | |
1285 | break; | |
1286 | } | |
1287 | } | |
1288 | if (has_pcps) | |
1289 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
1290 | else | |
1291 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
1292 | } | |
1293 | on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1); | |
9f8f2172 CL |
1294 | } |
1295 | ||
296699de | 1296 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1297 | |
1298 | void mark_free_pages(struct zone *zone) | |
1299 | { | |
f623f0db RW |
1300 | unsigned long pfn, max_zone_pfn; |
1301 | unsigned long flags; | |
b2a0ac88 | 1302 | int order, t; |
1da177e4 LT |
1303 | struct list_head *curr; |
1304 | ||
8080fc03 | 1305 | if (zone_is_empty(zone)) |
1da177e4 LT |
1306 | return; |
1307 | ||
1308 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 1309 | |
108bcc96 | 1310 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
1311 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
1312 | if (pfn_valid(pfn)) { | |
1313 | struct page *page = pfn_to_page(pfn); | |
1314 | ||
7be98234 RW |
1315 | if (!swsusp_page_is_forbidden(page)) |
1316 | swsusp_unset_page_free(page); | |
f623f0db | 1317 | } |
1da177e4 | 1318 | |
b2a0ac88 MG |
1319 | for_each_migratetype_order(order, t) { |
1320 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1321 | unsigned long i; |
1da177e4 | 1322 | |
f623f0db RW |
1323 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1324 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1325 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1326 | } |
b2a0ac88 | 1327 | } |
1da177e4 LT |
1328 | spin_unlock_irqrestore(&zone->lock, flags); |
1329 | } | |
e2c55dc8 | 1330 | #endif /* CONFIG_PM */ |
1da177e4 | 1331 | |
1da177e4 LT |
1332 | /* |
1333 | * Free a 0-order page | |
fc91668e | 1334 | * cold == 1 ? free a cold page : free a hot page |
1da177e4 | 1335 | */ |
fc91668e | 1336 | void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1337 | { |
1338 | struct zone *zone = page_zone(page); | |
1339 | struct per_cpu_pages *pcp; | |
1340 | unsigned long flags; | |
5f8dcc21 | 1341 | int migratetype; |
1da177e4 | 1342 | |
ec95f53a | 1343 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1344 | return; |
1345 | ||
5f8dcc21 | 1346 | migratetype = get_pageblock_migratetype(page); |
b12c4ad1 | 1347 | set_freepage_migratetype(page, migratetype); |
1da177e4 | 1348 | local_irq_save(flags); |
f8891e5e | 1349 | __count_vm_event(PGFREE); |
da456f14 | 1350 | |
5f8dcc21 MG |
1351 | /* |
1352 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1353 | * Free ISOLATE pages back to the allocator because they are being | |
1354 | * offlined but treat RESERVE as movable pages so we can get those | |
1355 | * areas back if necessary. Otherwise, we may have to free | |
1356 | * excessively into the page allocator | |
1357 | */ | |
1358 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 1359 | if (unlikely(is_migrate_isolate(migratetype))) { |
5f8dcc21 MG |
1360 | free_one_page(zone, page, 0, migratetype); |
1361 | goto out; | |
1362 | } | |
1363 | migratetype = MIGRATE_MOVABLE; | |
1364 | } | |
1365 | ||
99dcc3e5 | 1366 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3dfa5721 | 1367 | if (cold) |
5f8dcc21 | 1368 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1369 | else |
5f8dcc21 | 1370 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1371 | pcp->count++; |
48db57f8 | 1372 | if (pcp->count >= pcp->high) { |
998d39cb CS |
1373 | unsigned long batch = ACCESS_ONCE(pcp->batch); |
1374 | free_pcppages_bulk(zone, batch, pcp); | |
1375 | pcp->count -= batch; | |
48db57f8 | 1376 | } |
5f8dcc21 MG |
1377 | |
1378 | out: | |
1da177e4 | 1379 | local_irq_restore(flags); |
1da177e4 LT |
1380 | } |
1381 | ||
cc59850e KK |
1382 | /* |
1383 | * Free a list of 0-order pages | |
1384 | */ | |
1385 | void free_hot_cold_page_list(struct list_head *list, int cold) | |
1386 | { | |
1387 | struct page *page, *next; | |
1388 | ||
1389 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1390 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1391 | free_hot_cold_page(page, cold); |
1392 | } | |
1393 | } | |
1394 | ||
8dfcc9ba NP |
1395 | /* |
1396 | * split_page takes a non-compound higher-order page, and splits it into | |
1397 | * n (1<<order) sub-pages: page[0..n] | |
1398 | * Each sub-page must be freed individually. | |
1399 | * | |
1400 | * Note: this is probably too low level an operation for use in drivers. | |
1401 | * Please consult with lkml before using this in your driver. | |
1402 | */ | |
1403 | void split_page(struct page *page, unsigned int order) | |
1404 | { | |
1405 | int i; | |
1406 | ||
725d704e NP |
1407 | VM_BUG_ON(PageCompound(page)); |
1408 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1409 | |
1410 | #ifdef CONFIG_KMEMCHECK | |
1411 | /* | |
1412 | * Split shadow pages too, because free(page[0]) would | |
1413 | * otherwise free the whole shadow. | |
1414 | */ | |
1415 | if (kmemcheck_page_is_tracked(page)) | |
1416 | split_page(virt_to_page(page[0].shadow), order); | |
1417 | #endif | |
1418 | ||
7835e98b NP |
1419 | for (i = 1; i < (1 << order); i++) |
1420 | set_page_refcounted(page + i); | |
8dfcc9ba | 1421 | } |
5853ff23 | 1422 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 1423 | |
8fb74b9f | 1424 | static int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 1425 | { |
748446bb MG |
1426 | unsigned long watermark; |
1427 | struct zone *zone; | |
2139cbe6 | 1428 | int mt; |
748446bb MG |
1429 | |
1430 | BUG_ON(!PageBuddy(page)); | |
1431 | ||
1432 | zone = page_zone(page); | |
2e30abd1 | 1433 | mt = get_pageblock_migratetype(page); |
748446bb | 1434 | |
194159fb | 1435 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
1436 | /* Obey watermarks as if the page was being allocated */ |
1437 | watermark = low_wmark_pages(zone) + (1 << order); | |
1438 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1439 | return 0; | |
1440 | ||
8fb74b9f | 1441 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 1442 | } |
748446bb MG |
1443 | |
1444 | /* Remove page from free list */ | |
1445 | list_del(&page->lru); | |
1446 | zone->free_area[order].nr_free--; | |
1447 | rmv_page_order(page); | |
2139cbe6 | 1448 | |
8fb74b9f | 1449 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
1450 | if (order >= pageblock_order - 1) { |
1451 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
1452 | for (; page < endpage; page += pageblock_nr_pages) { |
1453 | int mt = get_pageblock_migratetype(page); | |
194159fb | 1454 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
1455 | set_pageblock_migratetype(page, |
1456 | MIGRATE_MOVABLE); | |
1457 | } | |
748446bb MG |
1458 | } |
1459 | ||
8fb74b9f | 1460 | return 1UL << order; |
1fb3f8ca MG |
1461 | } |
1462 | ||
1463 | /* | |
1464 | * Similar to split_page except the page is already free. As this is only | |
1465 | * being used for migration, the migratetype of the block also changes. | |
1466 | * As this is called with interrupts disabled, the caller is responsible | |
1467 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1468 | * are enabled. | |
1469 | * | |
1470 | * Note: this is probably too low level an operation for use in drivers. | |
1471 | * Please consult with lkml before using this in your driver. | |
1472 | */ | |
1473 | int split_free_page(struct page *page) | |
1474 | { | |
1475 | unsigned int order; | |
1476 | int nr_pages; | |
1477 | ||
1fb3f8ca MG |
1478 | order = page_order(page); |
1479 | ||
8fb74b9f | 1480 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
1481 | if (!nr_pages) |
1482 | return 0; | |
1483 | ||
1484 | /* Split into individual pages */ | |
1485 | set_page_refcounted(page); | |
1486 | split_page(page, order); | |
1487 | return nr_pages; | |
748446bb MG |
1488 | } |
1489 | ||
1da177e4 LT |
1490 | /* |
1491 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1492 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1493 | * or two. | |
1494 | */ | |
0a15c3e9 MG |
1495 | static inline |
1496 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1497 | struct zone *zone, int order, gfp_t gfp_flags, |
1498 | int migratetype) | |
1da177e4 LT |
1499 | { |
1500 | unsigned long flags; | |
689bcebf | 1501 | struct page *page; |
1da177e4 LT |
1502 | int cold = !!(gfp_flags & __GFP_COLD); |
1503 | ||
689bcebf | 1504 | again: |
48db57f8 | 1505 | if (likely(order == 0)) { |
1da177e4 | 1506 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1507 | struct list_head *list; |
1da177e4 | 1508 | |
1da177e4 | 1509 | local_irq_save(flags); |
99dcc3e5 CL |
1510 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1511 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1512 | if (list_empty(list)) { |
535131e6 | 1513 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1514 | pcp->batch, list, |
e084b2d9 | 1515 | migratetype, cold); |
5f8dcc21 | 1516 | if (unlikely(list_empty(list))) |
6fb332fa | 1517 | goto failed; |
535131e6 | 1518 | } |
b92a6edd | 1519 | |
5f8dcc21 MG |
1520 | if (cold) |
1521 | page = list_entry(list->prev, struct page, lru); | |
1522 | else | |
1523 | page = list_entry(list->next, struct page, lru); | |
1524 | ||
b92a6edd MG |
1525 | list_del(&page->lru); |
1526 | pcp->count--; | |
7fb1d9fc | 1527 | } else { |
dab48dab AM |
1528 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1529 | /* | |
1530 | * __GFP_NOFAIL is not to be used in new code. | |
1531 | * | |
1532 | * All __GFP_NOFAIL callers should be fixed so that they | |
1533 | * properly detect and handle allocation failures. | |
1534 | * | |
1535 | * We most definitely don't want callers attempting to | |
4923abf9 | 1536 | * allocate greater than order-1 page units with |
dab48dab AM |
1537 | * __GFP_NOFAIL. |
1538 | */ | |
4923abf9 | 1539 | WARN_ON_ONCE(order > 1); |
dab48dab | 1540 | } |
1da177e4 | 1541 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1542 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1543 | spin_unlock(&zone->lock); |
1544 | if (!page) | |
1545 | goto failed; | |
d1ce749a BZ |
1546 | __mod_zone_freepage_state(zone, -(1 << order), |
1547 | get_pageblock_migratetype(page)); | |
1da177e4 LT |
1548 | } |
1549 | ||
81c0a2bb | 1550 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
f8891e5e | 1551 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1552 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1553 | local_irq_restore(flags); |
1da177e4 | 1554 | |
725d704e | 1555 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1556 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1557 | goto again; |
1da177e4 | 1558 | return page; |
a74609fa NP |
1559 | |
1560 | failed: | |
1561 | local_irq_restore(flags); | |
a74609fa | 1562 | return NULL; |
1da177e4 LT |
1563 | } |
1564 | ||
933e312e AM |
1565 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1566 | ||
b2588c4b | 1567 | static struct { |
933e312e AM |
1568 | struct fault_attr attr; |
1569 | ||
1570 | u32 ignore_gfp_highmem; | |
1571 | u32 ignore_gfp_wait; | |
54114994 | 1572 | u32 min_order; |
933e312e AM |
1573 | } fail_page_alloc = { |
1574 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1575 | .ignore_gfp_wait = 1, |
1576 | .ignore_gfp_highmem = 1, | |
54114994 | 1577 | .min_order = 1, |
933e312e AM |
1578 | }; |
1579 | ||
1580 | static int __init setup_fail_page_alloc(char *str) | |
1581 | { | |
1582 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1583 | } | |
1584 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1585 | ||
deaf386e | 1586 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1587 | { |
54114994 | 1588 | if (order < fail_page_alloc.min_order) |
deaf386e | 1589 | return false; |
933e312e | 1590 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 1591 | return false; |
933e312e | 1592 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 1593 | return false; |
933e312e | 1594 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
deaf386e | 1595 | return false; |
933e312e AM |
1596 | |
1597 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1598 | } | |
1599 | ||
1600 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1601 | ||
1602 | static int __init fail_page_alloc_debugfs(void) | |
1603 | { | |
f4ae40a6 | 1604 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1605 | struct dentry *dir; |
933e312e | 1606 | |
dd48c085 AM |
1607 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1608 | &fail_page_alloc.attr); | |
1609 | if (IS_ERR(dir)) | |
1610 | return PTR_ERR(dir); | |
933e312e | 1611 | |
b2588c4b AM |
1612 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1613 | &fail_page_alloc.ignore_gfp_wait)) | |
1614 | goto fail; | |
1615 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1616 | &fail_page_alloc.ignore_gfp_highmem)) | |
1617 | goto fail; | |
1618 | if (!debugfs_create_u32("min-order", mode, dir, | |
1619 | &fail_page_alloc.min_order)) | |
1620 | goto fail; | |
1621 | ||
1622 | return 0; | |
1623 | fail: | |
dd48c085 | 1624 | debugfs_remove_recursive(dir); |
933e312e | 1625 | |
b2588c4b | 1626 | return -ENOMEM; |
933e312e AM |
1627 | } |
1628 | ||
1629 | late_initcall(fail_page_alloc_debugfs); | |
1630 | ||
1631 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1632 | ||
1633 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1634 | ||
deaf386e | 1635 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1636 | { |
deaf386e | 1637 | return false; |
933e312e AM |
1638 | } |
1639 | ||
1640 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1641 | ||
1da177e4 | 1642 | /* |
88f5acf8 | 1643 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1644 | * of the allocation. |
1645 | */ | |
88f5acf8 MG |
1646 | static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
1647 | int classzone_idx, int alloc_flags, long free_pages) | |
1da177e4 LT |
1648 | { |
1649 | /* free_pages my go negative - that's OK */ | |
d23ad423 | 1650 | long min = mark; |
2cfed075 | 1651 | long lowmem_reserve = z->lowmem_reserve[classzone_idx]; |
1da177e4 | 1652 | int o; |
026b0814 | 1653 | long free_cma = 0; |
1da177e4 | 1654 | |
df0a6daa | 1655 | free_pages -= (1 << order) - 1; |
7fb1d9fc | 1656 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1657 | min -= min / 2; |
7fb1d9fc | 1658 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 | 1659 | min -= min / 4; |
d95ea5d1 BZ |
1660 | #ifdef CONFIG_CMA |
1661 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
1662 | if (!(alloc_flags & ALLOC_CMA)) | |
026b0814 | 1663 | free_cma = zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 1664 | #endif |
026b0814 TS |
1665 | |
1666 | if (free_pages - free_cma <= min + lowmem_reserve) | |
88f5acf8 | 1667 | return false; |
1da177e4 LT |
1668 | for (o = 0; o < order; o++) { |
1669 | /* At the next order, this order's pages become unavailable */ | |
1670 | free_pages -= z->free_area[o].nr_free << o; | |
1671 | ||
1672 | /* Require fewer higher order pages to be free */ | |
1673 | min >>= 1; | |
1674 | ||
1675 | if (free_pages <= min) | |
88f5acf8 | 1676 | return false; |
1da177e4 | 1677 | } |
88f5acf8 MG |
1678 | return true; |
1679 | } | |
1680 | ||
1681 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
1682 | int classzone_idx, int alloc_flags) | |
1683 | { | |
1684 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1685 | zone_page_state(z, NR_FREE_PAGES)); | |
1686 | } | |
1687 | ||
1688 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, | |
1689 | int classzone_idx, int alloc_flags) | |
1690 | { | |
1691 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1692 | ||
1693 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1694 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1695 | ||
1696 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1697 | free_pages); | |
1da177e4 LT |
1698 | } |
1699 | ||
9276b1bc PJ |
1700 | #ifdef CONFIG_NUMA |
1701 | /* | |
1702 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1703 | * skip over zones that are not allowed by the cpuset, or that have | |
1704 | * been recently (in last second) found to be nearly full. See further | |
1705 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1706 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc | 1707 | * |
a1aeb65a | 1708 | * If the zonelist cache is present in the passed zonelist, then |
9276b1bc | 1709 | * returns a pointer to the allowed node mask (either the current |
4b0ef1fe | 1710 | * tasks mems_allowed, or node_states[N_MEMORY].) |
9276b1bc PJ |
1711 | * |
1712 | * If the zonelist cache is not available for this zonelist, does | |
1713 | * nothing and returns NULL. | |
1714 | * | |
1715 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1716 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1717 | * | |
1718 | * We hold off even calling zlc_setup, until after we've checked the | |
1719 | * first zone in the zonelist, on the theory that most allocations will | |
1720 | * be satisfied from that first zone, so best to examine that zone as | |
1721 | * quickly as we can. | |
1722 | */ | |
1723 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1724 | { | |
1725 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1726 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1727 | ||
1728 | zlc = zonelist->zlcache_ptr; | |
1729 | if (!zlc) | |
1730 | return NULL; | |
1731 | ||
f05111f5 | 1732 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1733 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1734 | zlc->last_full_zap = jiffies; | |
1735 | } | |
1736 | ||
1737 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1738 | &cpuset_current_mems_allowed : | |
4b0ef1fe | 1739 | &node_states[N_MEMORY]; |
9276b1bc PJ |
1740 | return allowednodes; |
1741 | } | |
1742 | ||
1743 | /* | |
1744 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1745 | * if it is worth looking at further for free memory: | |
1746 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1747 | * bit set in the zonelist_cache fullzones BITMAP). | |
1748 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1749 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1750 | * Return true (non-zero) if zone is worth looking at further, or | |
1751 | * else return false (zero) if it is not. | |
1752 | * | |
1753 | * This check -ignores- the distinction between various watermarks, | |
1754 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1755 | * found to be full for any variation of these watermarks, it will | |
1756 | * be considered full for up to one second by all requests, unless | |
1757 | * we are so low on memory on all allowed nodes that we are forced | |
1758 | * into the second scan of the zonelist. | |
1759 | * | |
1760 | * In the second scan we ignore this zonelist cache and exactly | |
1761 | * apply the watermarks to all zones, even it is slower to do so. | |
1762 | * We are low on memory in the second scan, and should leave no stone | |
1763 | * unturned looking for a free page. | |
1764 | */ | |
dd1a239f | 1765 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1766 | nodemask_t *allowednodes) |
1767 | { | |
1768 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1769 | int i; /* index of *z in zonelist zones */ | |
1770 | int n; /* node that zone *z is on */ | |
1771 | ||
1772 | zlc = zonelist->zlcache_ptr; | |
1773 | if (!zlc) | |
1774 | return 1; | |
1775 | ||
dd1a239f | 1776 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1777 | n = zlc->z_to_n[i]; |
1778 | ||
1779 | /* This zone is worth trying if it is allowed but not full */ | |
1780 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1781 | } | |
1782 | ||
1783 | /* | |
1784 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1785 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1786 | * from that zone don't waste time re-examining it. | |
1787 | */ | |
dd1a239f | 1788 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1789 | { |
1790 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1791 | int i; /* index of *z in zonelist zones */ | |
1792 | ||
1793 | zlc = zonelist->zlcache_ptr; | |
1794 | if (!zlc) | |
1795 | return; | |
1796 | ||
dd1a239f | 1797 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1798 | |
1799 | set_bit(i, zlc->fullzones); | |
1800 | } | |
1801 | ||
76d3fbf8 MG |
1802 | /* |
1803 | * clear all zones full, called after direct reclaim makes progress so that | |
1804 | * a zone that was recently full is not skipped over for up to a second | |
1805 | */ | |
1806 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1807 | { | |
1808 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1809 | ||
1810 | zlc = zonelist->zlcache_ptr; | |
1811 | if (!zlc) | |
1812 | return; | |
1813 | ||
1814 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1815 | } | |
1816 | ||
81c0a2bb JW |
1817 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
1818 | { | |
1819 | return node_distance(local_zone->node, zone->node) == LOCAL_DISTANCE; | |
1820 | } | |
1821 | ||
957f822a DR |
1822 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1823 | { | |
1824 | return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes); | |
1825 | } | |
1826 | ||
1827 | static void __paginginit init_zone_allows_reclaim(int nid) | |
1828 | { | |
1829 | int i; | |
1830 | ||
1831 | for_each_online_node(i) | |
6b187d02 | 1832 | if (node_distance(nid, i) <= RECLAIM_DISTANCE) |
957f822a | 1833 | node_set(i, NODE_DATA(nid)->reclaim_nodes); |
6b187d02 | 1834 | else |
957f822a | 1835 | zone_reclaim_mode = 1; |
957f822a DR |
1836 | } |
1837 | ||
9276b1bc PJ |
1838 | #else /* CONFIG_NUMA */ |
1839 | ||
1840 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1841 | { | |
1842 | return NULL; | |
1843 | } | |
1844 | ||
dd1a239f | 1845 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1846 | nodemask_t *allowednodes) |
1847 | { | |
1848 | return 1; | |
1849 | } | |
1850 | ||
dd1a239f | 1851 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1852 | { |
1853 | } | |
76d3fbf8 MG |
1854 | |
1855 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1856 | { | |
1857 | } | |
957f822a | 1858 | |
81c0a2bb JW |
1859 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
1860 | { | |
1861 | return true; | |
1862 | } | |
1863 | ||
957f822a DR |
1864 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1865 | { | |
1866 | return true; | |
1867 | } | |
1868 | ||
1869 | static inline void init_zone_allows_reclaim(int nid) | |
1870 | { | |
1871 | } | |
9276b1bc PJ |
1872 | #endif /* CONFIG_NUMA */ |
1873 | ||
7fb1d9fc | 1874 | /* |
0798e519 | 1875 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1876 | * a page. |
1877 | */ | |
1878 | static struct page * | |
19770b32 | 1879 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1880 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1881 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1882 | { |
dd1a239f | 1883 | struct zoneref *z; |
7fb1d9fc | 1884 | struct page *page = NULL; |
54a6eb5c | 1885 | int classzone_idx; |
5117f45d | 1886 | struct zone *zone; |
9276b1bc PJ |
1887 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1888 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1889 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1890 | |
19770b32 | 1891 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1892 | zonelist_scan: |
7fb1d9fc | 1893 | /* |
9276b1bc | 1894 | * Scan zonelist, looking for a zone with enough free. |
3b11f0aa | 1895 | * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c. |
7fb1d9fc | 1896 | */ |
19770b32 MG |
1897 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1898 | high_zoneidx, nodemask) { | |
e085dbc5 JW |
1899 | unsigned long mark; |
1900 | ||
e5adfffc | 1901 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
9276b1bc PJ |
1902 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1903 | continue; | |
7fb1d9fc | 1904 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1905 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
cd38b115 | 1906 | continue; |
e085dbc5 | 1907 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
e66f0972 | 1908 | if (unlikely(alloc_flags & ALLOC_NO_WATERMARKS)) |
e085dbc5 | 1909 | goto try_this_zone; |
81c0a2bb JW |
1910 | /* |
1911 | * Distribute pages in proportion to the individual | |
1912 | * zone size to ensure fair page aging. The zone a | |
1913 | * page was allocated in should have no effect on the | |
1914 | * time the page has in memory before being reclaimed. | |
1915 | * | |
1916 | * When zone_reclaim_mode is enabled, try to stay in | |
1917 | * local zones in the fastpath. If that fails, the | |
1918 | * slowpath is entered, which will do another pass | |
1919 | * starting with the local zones, but ultimately fall | |
1920 | * back to remote zones that do not partake in the | |
1921 | * fairness round-robin cycle of this zonelist. | |
1922 | */ | |
1923 | if (alloc_flags & ALLOC_WMARK_LOW) { | |
1924 | if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0) | |
1925 | continue; | |
1926 | if (zone_reclaim_mode && | |
1927 | !zone_local(preferred_zone, zone)) | |
1928 | continue; | |
1929 | } | |
a756cf59 JW |
1930 | /* |
1931 | * When allocating a page cache page for writing, we | |
1932 | * want to get it from a zone that is within its dirty | |
1933 | * limit, such that no single zone holds more than its | |
1934 | * proportional share of globally allowed dirty pages. | |
1935 | * The dirty limits take into account the zone's | |
1936 | * lowmem reserves and high watermark so that kswapd | |
1937 | * should be able to balance it without having to | |
1938 | * write pages from its LRU list. | |
1939 | * | |
1940 | * This may look like it could increase pressure on | |
1941 | * lower zones by failing allocations in higher zones | |
1942 | * before they are full. But the pages that do spill | |
1943 | * over are limited as the lower zones are protected | |
1944 | * by this very same mechanism. It should not become | |
1945 | * a practical burden to them. | |
1946 | * | |
1947 | * XXX: For now, allow allocations to potentially | |
1948 | * exceed the per-zone dirty limit in the slowpath | |
1949 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
1950 | * which is important when on a NUMA setup the allowed | |
1951 | * zones are together not big enough to reach the | |
1952 | * global limit. The proper fix for these situations | |
1953 | * will require awareness of zones in the | |
1954 | * dirty-throttling and the flusher threads. | |
1955 | */ | |
1956 | if ((alloc_flags & ALLOC_WMARK_LOW) && | |
1957 | (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone)) | |
1958 | goto this_zone_full; | |
7fb1d9fc | 1959 | |
e085dbc5 JW |
1960 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
1961 | if (!zone_watermark_ok(zone, order, mark, | |
1962 | classzone_idx, alloc_flags)) { | |
fa5e084e MG |
1963 | int ret; |
1964 | ||
e5adfffc KS |
1965 | if (IS_ENABLED(CONFIG_NUMA) && |
1966 | !did_zlc_setup && nr_online_nodes > 1) { | |
cd38b115 MG |
1967 | /* |
1968 | * we do zlc_setup if there are multiple nodes | |
1969 | * and before considering the first zone allowed | |
1970 | * by the cpuset. | |
1971 | */ | |
1972 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1973 | zlc_active = 1; | |
1974 | did_zlc_setup = 1; | |
1975 | } | |
1976 | ||
957f822a DR |
1977 | if (zone_reclaim_mode == 0 || |
1978 | !zone_allows_reclaim(preferred_zone, zone)) | |
fa5e084e MG |
1979 | goto this_zone_full; |
1980 | ||
cd38b115 MG |
1981 | /* |
1982 | * As we may have just activated ZLC, check if the first | |
1983 | * eligible zone has failed zone_reclaim recently. | |
1984 | */ | |
e5adfffc | 1985 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
cd38b115 MG |
1986 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1987 | continue; | |
1988 | ||
fa5e084e MG |
1989 | ret = zone_reclaim(zone, gfp_mask, order); |
1990 | switch (ret) { | |
1991 | case ZONE_RECLAIM_NOSCAN: | |
1992 | /* did not scan */ | |
cd38b115 | 1993 | continue; |
fa5e084e MG |
1994 | case ZONE_RECLAIM_FULL: |
1995 | /* scanned but unreclaimable */ | |
cd38b115 | 1996 | continue; |
fa5e084e MG |
1997 | default: |
1998 | /* did we reclaim enough */ | |
fed2719e | 1999 | if (zone_watermark_ok(zone, order, mark, |
fa5e084e | 2000 | classzone_idx, alloc_flags)) |
fed2719e MG |
2001 | goto try_this_zone; |
2002 | ||
2003 | /* | |
2004 | * Failed to reclaim enough to meet watermark. | |
2005 | * Only mark the zone full if checking the min | |
2006 | * watermark or if we failed to reclaim just | |
2007 | * 1<<order pages or else the page allocator | |
2008 | * fastpath will prematurely mark zones full | |
2009 | * when the watermark is between the low and | |
2010 | * min watermarks. | |
2011 | */ | |
2012 | if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) || | |
2013 | ret == ZONE_RECLAIM_SOME) | |
9276b1bc | 2014 | goto this_zone_full; |
fed2719e MG |
2015 | |
2016 | continue; | |
0798e519 | 2017 | } |
7fb1d9fc RS |
2018 | } |
2019 | ||
fa5e084e | 2020 | try_this_zone: |
3dd28266 MG |
2021 | page = buffered_rmqueue(preferred_zone, zone, order, |
2022 | gfp_mask, migratetype); | |
0798e519 | 2023 | if (page) |
7fb1d9fc | 2024 | break; |
9276b1bc | 2025 | this_zone_full: |
e5adfffc | 2026 | if (IS_ENABLED(CONFIG_NUMA)) |
9276b1bc | 2027 | zlc_mark_zone_full(zonelist, z); |
54a6eb5c | 2028 | } |
9276b1bc | 2029 | |
e5adfffc | 2030 | if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) { |
9276b1bc PJ |
2031 | /* Disable zlc cache for second zonelist scan */ |
2032 | zlc_active = 0; | |
2033 | goto zonelist_scan; | |
2034 | } | |
b121186a AS |
2035 | |
2036 | if (page) | |
2037 | /* | |
2038 | * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was | |
2039 | * necessary to allocate the page. The expectation is | |
2040 | * that the caller is taking steps that will free more | |
2041 | * memory. The caller should avoid the page being used | |
2042 | * for !PFMEMALLOC purposes. | |
2043 | */ | |
2044 | page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS); | |
2045 | ||
7fb1d9fc | 2046 | return page; |
753ee728 MH |
2047 | } |
2048 | ||
29423e77 DR |
2049 | /* |
2050 | * Large machines with many possible nodes should not always dump per-node | |
2051 | * meminfo in irq context. | |
2052 | */ | |
2053 | static inline bool should_suppress_show_mem(void) | |
2054 | { | |
2055 | bool ret = false; | |
2056 | ||
2057 | #if NODES_SHIFT > 8 | |
2058 | ret = in_interrupt(); | |
2059 | #endif | |
2060 | return ret; | |
2061 | } | |
2062 | ||
a238ab5b DH |
2063 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2064 | DEFAULT_RATELIMIT_INTERVAL, | |
2065 | DEFAULT_RATELIMIT_BURST); | |
2066 | ||
2067 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
2068 | { | |
a238ab5b DH |
2069 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2070 | ||
c0a32fc5 SG |
2071 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2072 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2073 | return; |
2074 | ||
4b59e6c4 DR |
2075 | /* |
2076 | * Walking all memory to count page types is very expensive and should | |
2077 | * be inhibited in non-blockable contexts. | |
2078 | */ | |
2079 | if (!(gfp_mask & __GFP_WAIT)) | |
2080 | filter |= SHOW_MEM_FILTER_PAGE_COUNT; | |
2081 | ||
a238ab5b DH |
2082 | /* |
2083 | * This documents exceptions given to allocations in certain | |
2084 | * contexts that are allowed to allocate outside current's set | |
2085 | * of allowed nodes. | |
2086 | */ | |
2087 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2088 | if (test_thread_flag(TIF_MEMDIE) || | |
2089 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2090 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2091 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
2092 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2093 | ||
2094 | if (fmt) { | |
3ee9a4f0 JP |
2095 | struct va_format vaf; |
2096 | va_list args; | |
2097 | ||
a238ab5b | 2098 | va_start(args, fmt); |
3ee9a4f0 JP |
2099 | |
2100 | vaf.fmt = fmt; | |
2101 | vaf.va = &args; | |
2102 | ||
2103 | pr_warn("%pV", &vaf); | |
2104 | ||
a238ab5b DH |
2105 | va_end(args); |
2106 | } | |
2107 | ||
3ee9a4f0 JP |
2108 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
2109 | current->comm, order, gfp_mask); | |
a238ab5b DH |
2110 | |
2111 | dump_stack(); | |
2112 | if (!should_suppress_show_mem()) | |
2113 | show_mem(filter); | |
2114 | } | |
2115 | ||
11e33f6a MG |
2116 | static inline int |
2117 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
f90ac398 | 2118 | unsigned long did_some_progress, |
11e33f6a | 2119 | unsigned long pages_reclaimed) |
1da177e4 | 2120 | { |
11e33f6a MG |
2121 | /* Do not loop if specifically requested */ |
2122 | if (gfp_mask & __GFP_NORETRY) | |
2123 | return 0; | |
1da177e4 | 2124 | |
f90ac398 MG |
2125 | /* Always retry if specifically requested */ |
2126 | if (gfp_mask & __GFP_NOFAIL) | |
2127 | return 1; | |
2128 | ||
2129 | /* | |
2130 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim | |
2131 | * making forward progress without invoking OOM. Suspend also disables | |
2132 | * storage devices so kswapd will not help. Bail if we are suspending. | |
2133 | */ | |
2134 | if (!did_some_progress && pm_suspended_storage()) | |
2135 | return 0; | |
2136 | ||
11e33f6a MG |
2137 | /* |
2138 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
2139 | * means __GFP_NOFAIL, but that may not be true in other | |
2140 | * implementations. | |
2141 | */ | |
2142 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
2143 | return 1; | |
2144 | ||
2145 | /* | |
2146 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
2147 | * specified, then we retry until we no longer reclaim any pages | |
2148 | * (above), or we've reclaimed an order of pages at least as | |
2149 | * large as the allocation's order. In both cases, if the | |
2150 | * allocation still fails, we stop retrying. | |
2151 | */ | |
2152 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
2153 | return 1; | |
cf40bd16 | 2154 | |
11e33f6a MG |
2155 | return 0; |
2156 | } | |
933e312e | 2157 | |
11e33f6a MG |
2158 | static inline struct page * |
2159 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
2160 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2161 | nodemask_t *nodemask, struct zone *preferred_zone, |
2162 | int migratetype) | |
11e33f6a MG |
2163 | { |
2164 | struct page *page; | |
2165 | ||
2166 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
ff321fea | 2167 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
11e33f6a | 2168 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2169 | return NULL; |
2170 | } | |
6b1de916 | 2171 | |
11e33f6a MG |
2172 | /* |
2173 | * Go through the zonelist yet one more time, keep very high watermark | |
2174 | * here, this is only to catch a parallel oom killing, we must fail if | |
2175 | * we're still under heavy pressure. | |
2176 | */ | |
2177 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
2178 | order, zonelist, high_zoneidx, | |
5117f45d | 2179 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 2180 | preferred_zone, migratetype); |
7fb1d9fc | 2181 | if (page) |
11e33f6a MG |
2182 | goto out; |
2183 | ||
4365a567 KH |
2184 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2185 | /* The OOM killer will not help higher order allocs */ | |
2186 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2187 | goto out; | |
03668b3c DR |
2188 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
2189 | if (high_zoneidx < ZONE_NORMAL) | |
2190 | goto out; | |
4365a567 KH |
2191 | /* |
2192 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
2193 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
2194 | * The caller should handle page allocation failure by itself if | |
2195 | * it specifies __GFP_THISNODE. | |
2196 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
2197 | */ | |
2198 | if (gfp_mask & __GFP_THISNODE) | |
2199 | goto out; | |
2200 | } | |
11e33f6a | 2201 | /* Exhausted what can be done so it's blamo time */ |
08ab9b10 | 2202 | out_of_memory(zonelist, gfp_mask, order, nodemask, false); |
11e33f6a MG |
2203 | |
2204 | out: | |
2205 | clear_zonelist_oom(zonelist, gfp_mask); | |
2206 | return page; | |
2207 | } | |
2208 | ||
56de7263 MG |
2209 | #ifdef CONFIG_COMPACTION |
2210 | /* Try memory compaction for high-order allocations before reclaim */ | |
2211 | static struct page * | |
2212 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2213 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2214 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
66199712 | 2215 | int migratetype, bool sync_migration, |
c67fe375 | 2216 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2217 | unsigned long *did_some_progress) |
56de7263 | 2218 | { |
66199712 | 2219 | if (!order) |
56de7263 MG |
2220 | return NULL; |
2221 | ||
aff62249 | 2222 | if (compaction_deferred(preferred_zone, order)) { |
66199712 MG |
2223 | *deferred_compaction = true; |
2224 | return NULL; | |
2225 | } | |
2226 | ||
c06b1fca | 2227 | current->flags |= PF_MEMALLOC; |
56de7263 | 2228 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
c67fe375 | 2229 | nodemask, sync_migration, |
8fb74b9f | 2230 | contended_compaction); |
c06b1fca | 2231 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2232 | |
1fb3f8ca | 2233 | if (*did_some_progress != COMPACT_SKIPPED) { |
8fb74b9f MG |
2234 | struct page *page; |
2235 | ||
56de7263 MG |
2236 | /* Page migration frees to the PCP lists but we want merging */ |
2237 | drain_pages(get_cpu()); | |
2238 | put_cpu(); | |
2239 | ||
2240 | page = get_page_from_freelist(gfp_mask, nodemask, | |
2241 | order, zonelist, high_zoneidx, | |
cfd19c5a MG |
2242 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2243 | preferred_zone, migratetype); | |
56de7263 | 2244 | if (page) { |
62997027 | 2245 | preferred_zone->compact_blockskip_flush = false; |
4f92e258 MG |
2246 | preferred_zone->compact_considered = 0; |
2247 | preferred_zone->compact_defer_shift = 0; | |
aff62249 RR |
2248 | if (order >= preferred_zone->compact_order_failed) |
2249 | preferred_zone->compact_order_failed = order + 1; | |
56de7263 MG |
2250 | count_vm_event(COMPACTSUCCESS); |
2251 | return page; | |
2252 | } | |
2253 | ||
2254 | /* | |
2255 | * It's bad if compaction run occurs and fails. | |
2256 | * The most likely reason is that pages exist, | |
2257 | * but not enough to satisfy watermarks. | |
2258 | */ | |
2259 | count_vm_event(COMPACTFAIL); | |
66199712 MG |
2260 | |
2261 | /* | |
2262 | * As async compaction considers a subset of pageblocks, only | |
2263 | * defer if the failure was a sync compaction failure. | |
2264 | */ | |
2265 | if (sync_migration) | |
aff62249 | 2266 | defer_compaction(preferred_zone, order); |
56de7263 MG |
2267 | |
2268 | cond_resched(); | |
2269 | } | |
2270 | ||
2271 | return NULL; | |
2272 | } | |
2273 | #else | |
2274 | static inline struct page * | |
2275 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2276 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2277 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
66199712 | 2278 | int migratetype, bool sync_migration, |
c67fe375 | 2279 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2280 | unsigned long *did_some_progress) |
56de7263 MG |
2281 | { |
2282 | return NULL; | |
2283 | } | |
2284 | #endif /* CONFIG_COMPACTION */ | |
2285 | ||
bba90710 MS |
2286 | /* Perform direct synchronous page reclaim */ |
2287 | static int | |
2288 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, | |
2289 | nodemask_t *nodemask) | |
11e33f6a | 2290 | { |
11e33f6a | 2291 | struct reclaim_state reclaim_state; |
bba90710 | 2292 | int progress; |
11e33f6a MG |
2293 | |
2294 | cond_resched(); | |
2295 | ||
2296 | /* We now go into synchronous reclaim */ | |
2297 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2298 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2299 | lockdep_set_current_reclaim_state(gfp_mask); |
2300 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2301 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2302 | |
bba90710 | 2303 | progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
11e33f6a | 2304 | |
c06b1fca | 2305 | current->reclaim_state = NULL; |
11e33f6a | 2306 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2307 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2308 | |
2309 | cond_resched(); | |
2310 | ||
bba90710 MS |
2311 | return progress; |
2312 | } | |
2313 | ||
2314 | /* The really slow allocator path where we enter direct reclaim */ | |
2315 | static inline struct page * | |
2316 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
2317 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2318 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
2319 | int migratetype, unsigned long *did_some_progress) | |
2320 | { | |
2321 | struct page *page = NULL; | |
2322 | bool drained = false; | |
2323 | ||
2324 | *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist, | |
2325 | nodemask); | |
9ee493ce MG |
2326 | if (unlikely(!(*did_some_progress))) |
2327 | return NULL; | |
11e33f6a | 2328 | |
76d3fbf8 | 2329 | /* After successful reclaim, reconsider all zones for allocation */ |
e5adfffc | 2330 | if (IS_ENABLED(CONFIG_NUMA)) |
76d3fbf8 MG |
2331 | zlc_clear_zones_full(zonelist); |
2332 | ||
9ee493ce MG |
2333 | retry: |
2334 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2335 | zonelist, high_zoneidx, |
cfd19c5a MG |
2336 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2337 | preferred_zone, migratetype); | |
9ee493ce MG |
2338 | |
2339 | /* | |
2340 | * If an allocation failed after direct reclaim, it could be because | |
2341 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2342 | */ | |
2343 | if (!page && !drained) { | |
2344 | drain_all_pages(); | |
2345 | drained = true; | |
2346 | goto retry; | |
2347 | } | |
2348 | ||
11e33f6a MG |
2349 | return page; |
2350 | } | |
2351 | ||
1da177e4 | 2352 | /* |
11e33f6a MG |
2353 | * This is called in the allocator slow-path if the allocation request is of |
2354 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2355 | */ |
11e33f6a MG |
2356 | static inline struct page * |
2357 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
2358 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2359 | nodemask_t *nodemask, struct zone *preferred_zone, |
2360 | int migratetype) | |
11e33f6a MG |
2361 | { |
2362 | struct page *page; | |
2363 | ||
2364 | do { | |
2365 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2366 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 2367 | preferred_zone, migratetype); |
11e33f6a MG |
2368 | |
2369 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 2370 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
2371 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2372 | ||
2373 | return page; | |
2374 | } | |
2375 | ||
81c0a2bb JW |
2376 | static void prepare_slowpath(gfp_t gfp_mask, unsigned int order, |
2377 | struct zonelist *zonelist, | |
2378 | enum zone_type high_zoneidx, | |
2379 | struct zone *preferred_zone) | |
1da177e4 | 2380 | { |
dd1a239f MG |
2381 | struct zoneref *z; |
2382 | struct zone *zone; | |
1da177e4 | 2383 | |
81c0a2bb JW |
2384 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
2385 | if (!(gfp_mask & __GFP_NO_KSWAPD)) | |
2386 | wakeup_kswapd(zone, order, zone_idx(preferred_zone)); | |
2387 | /* | |
2388 | * Only reset the batches of zones that were actually | |
2389 | * considered in the fast path, we don't want to | |
2390 | * thrash fairness information for zones that are not | |
2391 | * actually part of this zonelist's round-robin cycle. | |
2392 | */ | |
2393 | if (zone_reclaim_mode && !zone_local(preferred_zone, zone)) | |
2394 | continue; | |
2395 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
2396 | high_wmark_pages(zone) - | |
2397 | low_wmark_pages(zone) - | |
2398 | zone_page_state(zone, NR_ALLOC_BATCH)); | |
2399 | } | |
11e33f6a | 2400 | } |
cf40bd16 | 2401 | |
341ce06f PZ |
2402 | static inline int |
2403 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2404 | { | |
341ce06f PZ |
2405 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
2406 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 2407 | |
a56f57ff | 2408 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2409 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2410 | |
341ce06f PZ |
2411 | /* |
2412 | * The caller may dip into page reserves a bit more if the caller | |
2413 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2414 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
2415 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
2416 | */ | |
e6223a3b | 2417 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2418 | |
341ce06f | 2419 | if (!wait) { |
5c3240d9 AA |
2420 | /* |
2421 | * Not worth trying to allocate harder for | |
2422 | * __GFP_NOMEMALLOC even if it can't schedule. | |
2423 | */ | |
2424 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2425 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 2426 | /* |
341ce06f PZ |
2427 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
2428 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 2429 | */ |
341ce06f | 2430 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2431 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2432 | alloc_flags |= ALLOC_HARDER; |
2433 | ||
b37f1dd0 MG |
2434 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
2435 | if (gfp_mask & __GFP_MEMALLOC) | |
2436 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
2437 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
2438 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
2439 | else if (!in_interrupt() && | |
2440 | ((current->flags & PF_MEMALLOC) || | |
2441 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 2442 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 2443 | } |
d95ea5d1 BZ |
2444 | #ifdef CONFIG_CMA |
2445 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2446 | alloc_flags |= ALLOC_CMA; | |
2447 | #endif | |
341ce06f PZ |
2448 | return alloc_flags; |
2449 | } | |
2450 | ||
072bb0aa MG |
2451 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
2452 | { | |
b37f1dd0 | 2453 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
2454 | } |
2455 | ||
11e33f6a MG |
2456 | static inline struct page * |
2457 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2458 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2459 | nodemask_t *nodemask, struct zone *preferred_zone, |
2460 | int migratetype) | |
11e33f6a MG |
2461 | { |
2462 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2463 | struct page *page = NULL; | |
2464 | int alloc_flags; | |
2465 | unsigned long pages_reclaimed = 0; | |
2466 | unsigned long did_some_progress; | |
77f1fe6b | 2467 | bool sync_migration = false; |
66199712 | 2468 | bool deferred_compaction = false; |
c67fe375 | 2469 | bool contended_compaction = false; |
1da177e4 | 2470 | |
72807a74 MG |
2471 | /* |
2472 | * In the slowpath, we sanity check order to avoid ever trying to | |
2473 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2474 | * be using allocators in order of preference for an area that is | |
2475 | * too large. | |
2476 | */ | |
1fc28b70 MG |
2477 | if (order >= MAX_ORDER) { |
2478 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2479 | return NULL; |
1fc28b70 | 2480 | } |
1da177e4 | 2481 | |
952f3b51 CL |
2482 | /* |
2483 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2484 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2485 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2486 | * using a larger set of nodes after it has established that the | |
2487 | * allowed per node queues are empty and that nodes are | |
2488 | * over allocated. | |
2489 | */ | |
e5adfffc KS |
2490 | if (IS_ENABLED(CONFIG_NUMA) && |
2491 | (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
952f3b51 CL |
2492 | goto nopage; |
2493 | ||
cc4a6851 | 2494 | restart: |
81c0a2bb JW |
2495 | prepare_slowpath(gfp_mask, order, zonelist, |
2496 | high_zoneidx, preferred_zone); | |
1da177e4 | 2497 | |
9bf2229f | 2498 | /* |
7fb1d9fc RS |
2499 | * OK, we're below the kswapd watermark and have kicked background |
2500 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2501 | * to how we want to proceed. | |
9bf2229f | 2502 | */ |
341ce06f | 2503 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2504 | |
f33261d7 DR |
2505 | /* |
2506 | * Find the true preferred zone if the allocation is unconstrained by | |
2507 | * cpusets. | |
2508 | */ | |
2509 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) | |
2510 | first_zones_zonelist(zonelist, high_zoneidx, NULL, | |
2511 | &preferred_zone); | |
2512 | ||
cfa54a0f | 2513 | rebalance: |
341ce06f | 2514 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2515 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
2516 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
2517 | preferred_zone, migratetype); | |
7fb1d9fc RS |
2518 | if (page) |
2519 | goto got_pg; | |
1da177e4 | 2520 | |
11e33f6a | 2521 | /* Allocate without watermarks if the context allows */ |
341ce06f | 2522 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
2523 | /* |
2524 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
2525 | * the allocation is high priority and these type of | |
2526 | * allocations are system rather than user orientated | |
2527 | */ | |
2528 | zonelist = node_zonelist(numa_node_id(), gfp_mask); | |
2529 | ||
341ce06f PZ |
2530 | page = __alloc_pages_high_priority(gfp_mask, order, |
2531 | zonelist, high_zoneidx, nodemask, | |
2532 | preferred_zone, migratetype); | |
cfd19c5a | 2533 | if (page) { |
341ce06f | 2534 | goto got_pg; |
cfd19c5a | 2535 | } |
1da177e4 LT |
2536 | } |
2537 | ||
2538 | /* Atomic allocations - we can't balance anything */ | |
2539 | if (!wait) | |
2540 | goto nopage; | |
2541 | ||
341ce06f | 2542 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2543 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2544 | goto nopage; |
2545 | ||
6583bb64 DR |
2546 | /* Avoid allocations with no watermarks from looping endlessly */ |
2547 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2548 | goto nopage; | |
2549 | ||
77f1fe6b MG |
2550 | /* |
2551 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2552 | * attempts after direct reclaim are synchronous | |
2553 | */ | |
56de7263 MG |
2554 | page = __alloc_pages_direct_compact(gfp_mask, order, |
2555 | zonelist, high_zoneidx, | |
2556 | nodemask, | |
2557 | alloc_flags, preferred_zone, | |
66199712 | 2558 | migratetype, sync_migration, |
c67fe375 | 2559 | &contended_compaction, |
66199712 MG |
2560 | &deferred_compaction, |
2561 | &did_some_progress); | |
56de7263 MG |
2562 | if (page) |
2563 | goto got_pg; | |
c6a140bf | 2564 | sync_migration = true; |
56de7263 | 2565 | |
31f8d42d LT |
2566 | /* |
2567 | * If compaction is deferred for high-order allocations, it is because | |
2568 | * sync compaction recently failed. In this is the case and the caller | |
2569 | * requested a movable allocation that does not heavily disrupt the | |
2570 | * system then fail the allocation instead of entering direct reclaim. | |
2571 | */ | |
2572 | if ((deferred_compaction || contended_compaction) && | |
caf49191 | 2573 | (gfp_mask & __GFP_NO_KSWAPD)) |
31f8d42d | 2574 | goto nopage; |
66199712 | 2575 | |
11e33f6a MG |
2576 | /* Try direct reclaim and then allocating */ |
2577 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2578 | zonelist, high_zoneidx, | |
2579 | nodemask, | |
5117f45d | 2580 | alloc_flags, preferred_zone, |
3dd28266 | 2581 | migratetype, &did_some_progress); |
11e33f6a MG |
2582 | if (page) |
2583 | goto got_pg; | |
1da177e4 | 2584 | |
e33c3b5e | 2585 | /* |
11e33f6a MG |
2586 | * If we failed to make any progress reclaiming, then we are |
2587 | * running out of options and have to consider going OOM | |
e33c3b5e | 2588 | */ |
11e33f6a | 2589 | if (!did_some_progress) { |
b9921ecd | 2590 | if (oom_gfp_allowed(gfp_mask)) { |
7f33d49a RW |
2591 | if (oom_killer_disabled) |
2592 | goto nopage; | |
29fd66d2 DR |
2593 | /* Coredumps can quickly deplete all memory reserves */ |
2594 | if ((current->flags & PF_DUMPCORE) && | |
2595 | !(gfp_mask & __GFP_NOFAIL)) | |
2596 | goto nopage; | |
11e33f6a MG |
2597 | page = __alloc_pages_may_oom(gfp_mask, order, |
2598 | zonelist, high_zoneidx, | |
3dd28266 MG |
2599 | nodemask, preferred_zone, |
2600 | migratetype); | |
11e33f6a MG |
2601 | if (page) |
2602 | goto got_pg; | |
1da177e4 | 2603 | |
03668b3c DR |
2604 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2605 | /* | |
2606 | * The oom killer is not called for high-order | |
2607 | * allocations that may fail, so if no progress | |
2608 | * is being made, there are no other options and | |
2609 | * retrying is unlikely to help. | |
2610 | */ | |
2611 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2612 | goto nopage; | |
2613 | /* | |
2614 | * The oom killer is not called for lowmem | |
2615 | * allocations to prevent needlessly killing | |
2616 | * innocent tasks. | |
2617 | */ | |
2618 | if (high_zoneidx < ZONE_NORMAL) | |
2619 | goto nopage; | |
2620 | } | |
e2c55dc8 | 2621 | |
ff0ceb9d DR |
2622 | goto restart; |
2623 | } | |
1da177e4 LT |
2624 | } |
2625 | ||
11e33f6a | 2626 | /* Check if we should retry the allocation */ |
a41f24ea | 2627 | pages_reclaimed += did_some_progress; |
f90ac398 MG |
2628 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
2629 | pages_reclaimed)) { | |
11e33f6a | 2630 | /* Wait for some write requests to complete then retry */ |
0e093d99 | 2631 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2632 | goto rebalance; |
3e7d3449 MG |
2633 | } else { |
2634 | /* | |
2635 | * High-order allocations do not necessarily loop after | |
2636 | * direct reclaim and reclaim/compaction depends on compaction | |
2637 | * being called after reclaim so call directly if necessary | |
2638 | */ | |
2639 | page = __alloc_pages_direct_compact(gfp_mask, order, | |
2640 | zonelist, high_zoneidx, | |
2641 | nodemask, | |
2642 | alloc_flags, preferred_zone, | |
66199712 | 2643 | migratetype, sync_migration, |
c67fe375 | 2644 | &contended_compaction, |
66199712 MG |
2645 | &deferred_compaction, |
2646 | &did_some_progress); | |
3e7d3449 MG |
2647 | if (page) |
2648 | goto got_pg; | |
1da177e4 LT |
2649 | } |
2650 | ||
2651 | nopage: | |
a238ab5b | 2652 | warn_alloc_failed(gfp_mask, order, NULL); |
b1eeab67 | 2653 | return page; |
1da177e4 | 2654 | got_pg: |
b1eeab67 VN |
2655 | if (kmemcheck_enabled) |
2656 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
11e33f6a | 2657 | |
072bb0aa | 2658 | return page; |
1da177e4 | 2659 | } |
11e33f6a MG |
2660 | |
2661 | /* | |
2662 | * This is the 'heart' of the zoned buddy allocator. | |
2663 | */ | |
2664 | struct page * | |
2665 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2666 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2667 | { | |
2668 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2669 | struct zone *preferred_zone; |
cc9a6c87 | 2670 | struct page *page = NULL; |
3dd28266 | 2671 | int migratetype = allocflags_to_migratetype(gfp_mask); |
cc9a6c87 | 2672 | unsigned int cpuset_mems_cookie; |
d95ea5d1 | 2673 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET; |
6a1a0d3b | 2674 | struct mem_cgroup *memcg = NULL; |
11e33f6a | 2675 | |
dcce284a BH |
2676 | gfp_mask &= gfp_allowed_mask; |
2677 | ||
11e33f6a MG |
2678 | lockdep_trace_alloc(gfp_mask); |
2679 | ||
2680 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2681 | ||
2682 | if (should_fail_alloc_page(gfp_mask, order)) | |
2683 | return NULL; | |
2684 | ||
2685 | /* | |
2686 | * Check the zones suitable for the gfp_mask contain at least one | |
2687 | * valid zone. It's possible to have an empty zonelist as a result | |
2688 | * of GFP_THISNODE and a memoryless node | |
2689 | */ | |
2690 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2691 | return NULL; | |
2692 | ||
6a1a0d3b GC |
2693 | /* |
2694 | * Will only have any effect when __GFP_KMEMCG is set. This is | |
2695 | * verified in the (always inline) callee | |
2696 | */ | |
2697 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
2698 | return NULL; | |
2699 | ||
cc9a6c87 MG |
2700 | retry_cpuset: |
2701 | cpuset_mems_cookie = get_mems_allowed(); | |
2702 | ||
5117f45d | 2703 | /* The preferred zone is used for statistics later */ |
f33261d7 DR |
2704 | first_zones_zonelist(zonelist, high_zoneidx, |
2705 | nodemask ? : &cpuset_current_mems_allowed, | |
2706 | &preferred_zone); | |
cc9a6c87 MG |
2707 | if (!preferred_zone) |
2708 | goto out; | |
5117f45d | 2709 | |
d95ea5d1 BZ |
2710 | #ifdef CONFIG_CMA |
2711 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2712 | alloc_flags |= ALLOC_CMA; | |
2713 | #endif | |
5117f45d | 2714 | /* First allocation attempt */ |
11e33f6a | 2715 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
d95ea5d1 | 2716 | zonelist, high_zoneidx, alloc_flags, |
3dd28266 | 2717 | preferred_zone, migratetype); |
21caf2fc ML |
2718 | if (unlikely(!page)) { |
2719 | /* | |
2720 | * Runtime PM, block IO and its error handling path | |
2721 | * can deadlock because I/O on the device might not | |
2722 | * complete. | |
2723 | */ | |
2724 | gfp_mask = memalloc_noio_flags(gfp_mask); | |
11e33f6a | 2725 | page = __alloc_pages_slowpath(gfp_mask, order, |
5117f45d | 2726 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 2727 | preferred_zone, migratetype); |
21caf2fc | 2728 | } |
11e33f6a | 2729 | |
4b4f278c | 2730 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
cc9a6c87 MG |
2731 | |
2732 | out: | |
2733 | /* | |
2734 | * When updating a task's mems_allowed, it is possible to race with | |
2735 | * parallel threads in such a way that an allocation can fail while | |
2736 | * the mask is being updated. If a page allocation is about to fail, | |
2737 | * check if the cpuset changed during allocation and if so, retry. | |
2738 | */ | |
2739 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) | |
2740 | goto retry_cpuset; | |
2741 | ||
6a1a0d3b GC |
2742 | memcg_kmem_commit_charge(page, memcg, order); |
2743 | ||
11e33f6a | 2744 | return page; |
1da177e4 | 2745 | } |
d239171e | 2746 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2747 | |
2748 | /* | |
2749 | * Common helper functions. | |
2750 | */ | |
920c7a5d | 2751 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2752 | { |
945a1113 AM |
2753 | struct page *page; |
2754 | ||
2755 | /* | |
2756 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2757 | * a highmem page | |
2758 | */ | |
2759 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2760 | ||
1da177e4 LT |
2761 | page = alloc_pages(gfp_mask, order); |
2762 | if (!page) | |
2763 | return 0; | |
2764 | return (unsigned long) page_address(page); | |
2765 | } | |
1da177e4 LT |
2766 | EXPORT_SYMBOL(__get_free_pages); |
2767 | ||
920c7a5d | 2768 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2769 | { |
945a1113 | 2770 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2771 | } |
1da177e4 LT |
2772 | EXPORT_SYMBOL(get_zeroed_page); |
2773 | ||
920c7a5d | 2774 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2775 | { |
b5810039 | 2776 | if (put_page_testzero(page)) { |
1da177e4 | 2777 | if (order == 0) |
fc91668e | 2778 | free_hot_cold_page(page, 0); |
1da177e4 LT |
2779 | else |
2780 | __free_pages_ok(page, order); | |
2781 | } | |
2782 | } | |
2783 | ||
2784 | EXPORT_SYMBOL(__free_pages); | |
2785 | ||
920c7a5d | 2786 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2787 | { |
2788 | if (addr != 0) { | |
725d704e | 2789 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2790 | __free_pages(virt_to_page((void *)addr), order); |
2791 | } | |
2792 | } | |
2793 | ||
2794 | EXPORT_SYMBOL(free_pages); | |
2795 | ||
6a1a0d3b GC |
2796 | /* |
2797 | * __free_memcg_kmem_pages and free_memcg_kmem_pages will free | |
2798 | * pages allocated with __GFP_KMEMCG. | |
2799 | * | |
2800 | * Those pages are accounted to a particular memcg, embedded in the | |
2801 | * corresponding page_cgroup. To avoid adding a hit in the allocator to search | |
2802 | * for that information only to find out that it is NULL for users who have no | |
2803 | * interest in that whatsoever, we provide these functions. | |
2804 | * | |
2805 | * The caller knows better which flags it relies on. | |
2806 | */ | |
2807 | void __free_memcg_kmem_pages(struct page *page, unsigned int order) | |
2808 | { | |
2809 | memcg_kmem_uncharge_pages(page, order); | |
2810 | __free_pages(page, order); | |
2811 | } | |
2812 | ||
2813 | void free_memcg_kmem_pages(unsigned long addr, unsigned int order) | |
2814 | { | |
2815 | if (addr != 0) { | |
2816 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
2817 | __free_memcg_kmem_pages(virt_to_page((void *)addr), order); | |
2818 | } | |
2819 | } | |
2820 | ||
ee85c2e1 AK |
2821 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
2822 | { | |
2823 | if (addr) { | |
2824 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2825 | unsigned long used = addr + PAGE_ALIGN(size); | |
2826 | ||
2827 | split_page(virt_to_page((void *)addr), order); | |
2828 | while (used < alloc_end) { | |
2829 | free_page(used); | |
2830 | used += PAGE_SIZE; | |
2831 | } | |
2832 | } | |
2833 | return (void *)addr; | |
2834 | } | |
2835 | ||
2be0ffe2 TT |
2836 | /** |
2837 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2838 | * @size: the number of bytes to allocate | |
2839 | * @gfp_mask: GFP flags for the allocation | |
2840 | * | |
2841 | * This function is similar to alloc_pages(), except that it allocates the | |
2842 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2843 | * allocate memory in power-of-two pages. | |
2844 | * | |
2845 | * This function is also limited by MAX_ORDER. | |
2846 | * | |
2847 | * Memory allocated by this function must be released by free_pages_exact(). | |
2848 | */ | |
2849 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2850 | { | |
2851 | unsigned int order = get_order(size); | |
2852 | unsigned long addr; | |
2853 | ||
2854 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 2855 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
2856 | } |
2857 | EXPORT_SYMBOL(alloc_pages_exact); | |
2858 | ||
ee85c2e1 AK |
2859 | /** |
2860 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
2861 | * pages on a node. | |
b5e6ab58 | 2862 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
2863 | * @size: the number of bytes to allocate |
2864 | * @gfp_mask: GFP flags for the allocation | |
2865 | * | |
2866 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
2867 | * back. | |
2868 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
2869 | * but is not exact. | |
2870 | */ | |
2871 | void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) | |
2872 | { | |
2873 | unsigned order = get_order(size); | |
2874 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
2875 | if (!p) | |
2876 | return NULL; | |
2877 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
2878 | } | |
2879 | EXPORT_SYMBOL(alloc_pages_exact_nid); | |
2880 | ||
2be0ffe2 TT |
2881 | /** |
2882 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2883 | * @virt: the value returned by alloc_pages_exact. | |
2884 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2885 | * | |
2886 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2887 | */ | |
2888 | void free_pages_exact(void *virt, size_t size) | |
2889 | { | |
2890 | unsigned long addr = (unsigned long)virt; | |
2891 | unsigned long end = addr + PAGE_ALIGN(size); | |
2892 | ||
2893 | while (addr < end) { | |
2894 | free_page(addr); | |
2895 | addr += PAGE_SIZE; | |
2896 | } | |
2897 | } | |
2898 | EXPORT_SYMBOL(free_pages_exact); | |
2899 | ||
e0fb5815 ZY |
2900 | /** |
2901 | * nr_free_zone_pages - count number of pages beyond high watermark | |
2902 | * @offset: The zone index of the highest zone | |
2903 | * | |
2904 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
2905 | * high watermark within all zones at or below a given zone index. For each | |
2906 | * zone, the number of pages is calculated as: | |
834405c3 | 2907 | * managed_pages - high_pages |
e0fb5815 | 2908 | */ |
ebec3862 | 2909 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 2910 | { |
dd1a239f | 2911 | struct zoneref *z; |
54a6eb5c MG |
2912 | struct zone *zone; |
2913 | ||
e310fd43 | 2914 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 2915 | unsigned long sum = 0; |
1da177e4 | 2916 | |
0e88460d | 2917 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2918 | |
54a6eb5c | 2919 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 2920 | unsigned long size = zone->managed_pages; |
41858966 | 2921 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2922 | if (size > high) |
2923 | sum += size - high; | |
1da177e4 LT |
2924 | } |
2925 | ||
2926 | return sum; | |
2927 | } | |
2928 | ||
e0fb5815 ZY |
2929 | /** |
2930 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
2931 | * | |
2932 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
2933 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 2934 | */ |
ebec3862 | 2935 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 2936 | { |
af4ca457 | 2937 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2938 | } |
c2f1a551 | 2939 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 2940 | |
e0fb5815 ZY |
2941 | /** |
2942 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
2943 | * | |
2944 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
2945 | * high watermark within all zones. | |
1da177e4 | 2946 | */ |
ebec3862 | 2947 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 2948 | { |
2a1e274a | 2949 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2950 | } |
08e0f6a9 CL |
2951 | |
2952 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2953 | { |
e5adfffc | 2954 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 2955 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2956 | } |
1da177e4 | 2957 | |
1da177e4 LT |
2958 | void si_meminfo(struct sysinfo *val) |
2959 | { | |
2960 | val->totalram = totalram_pages; | |
2961 | val->sharedram = 0; | |
d23ad423 | 2962 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2963 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2964 | val->totalhigh = totalhigh_pages; |
2965 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2966 | val->mem_unit = PAGE_SIZE; |
2967 | } | |
2968 | ||
2969 | EXPORT_SYMBOL(si_meminfo); | |
2970 | ||
2971 | #ifdef CONFIG_NUMA | |
2972 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2973 | { | |
cdd91a77 JL |
2974 | int zone_type; /* needs to be signed */ |
2975 | unsigned long managed_pages = 0; | |
1da177e4 LT |
2976 | pg_data_t *pgdat = NODE_DATA(nid); |
2977 | ||
cdd91a77 JL |
2978 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
2979 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
2980 | val->totalram = managed_pages; | |
d23ad423 | 2981 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2982 | #ifdef CONFIG_HIGHMEM |
b40da049 | 2983 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages; |
d23ad423 CL |
2984 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2985 | NR_FREE_PAGES); | |
98d2b0eb CL |
2986 | #else |
2987 | val->totalhigh = 0; | |
2988 | val->freehigh = 0; | |
2989 | #endif | |
1da177e4 LT |
2990 | val->mem_unit = PAGE_SIZE; |
2991 | } | |
2992 | #endif | |
2993 | ||
ddd588b5 | 2994 | /* |
7bf02ea2 DR |
2995 | * Determine whether the node should be displayed or not, depending on whether |
2996 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 2997 | */ |
7bf02ea2 | 2998 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
2999 | { |
3000 | bool ret = false; | |
cc9a6c87 | 3001 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
3002 | |
3003 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
3004 | goto out; | |
3005 | ||
cc9a6c87 MG |
3006 | do { |
3007 | cpuset_mems_cookie = get_mems_allowed(); | |
3008 | ret = !node_isset(nid, cpuset_current_mems_allowed); | |
3009 | } while (!put_mems_allowed(cpuset_mems_cookie)); | |
ddd588b5 DR |
3010 | out: |
3011 | return ret; | |
3012 | } | |
3013 | ||
1da177e4 LT |
3014 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
3015 | ||
377e4f16 RV |
3016 | static void show_migration_types(unsigned char type) |
3017 | { | |
3018 | static const char types[MIGRATE_TYPES] = { | |
3019 | [MIGRATE_UNMOVABLE] = 'U', | |
3020 | [MIGRATE_RECLAIMABLE] = 'E', | |
3021 | [MIGRATE_MOVABLE] = 'M', | |
3022 | [MIGRATE_RESERVE] = 'R', | |
3023 | #ifdef CONFIG_CMA | |
3024 | [MIGRATE_CMA] = 'C', | |
3025 | #endif | |
194159fb | 3026 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 3027 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 3028 | #endif |
377e4f16 RV |
3029 | }; |
3030 | char tmp[MIGRATE_TYPES + 1]; | |
3031 | char *p = tmp; | |
3032 | int i; | |
3033 | ||
3034 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
3035 | if (type & (1 << i)) | |
3036 | *p++ = types[i]; | |
3037 | } | |
3038 | ||
3039 | *p = '\0'; | |
3040 | printk("(%s) ", tmp); | |
3041 | } | |
3042 | ||
1da177e4 LT |
3043 | /* |
3044 | * Show free area list (used inside shift_scroll-lock stuff) | |
3045 | * We also calculate the percentage fragmentation. We do this by counting the | |
3046 | * memory on each free list with the exception of the first item on the list. | |
ddd588b5 DR |
3047 | * Suppresses nodes that are not allowed by current's cpuset if |
3048 | * SHOW_MEM_FILTER_NODES is passed. | |
1da177e4 | 3049 | */ |
7bf02ea2 | 3050 | void show_free_areas(unsigned int filter) |
1da177e4 | 3051 | { |
c7241913 | 3052 | int cpu; |
1da177e4 LT |
3053 | struct zone *zone; |
3054 | ||
ee99c71c | 3055 | for_each_populated_zone(zone) { |
7bf02ea2 | 3056 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3057 | continue; |
c7241913 JS |
3058 | show_node(zone); |
3059 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 3060 | |
6b482c67 | 3061 | for_each_online_cpu(cpu) { |
1da177e4 LT |
3062 | struct per_cpu_pageset *pageset; |
3063 | ||
99dcc3e5 | 3064 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 3065 | |
3dfa5721 CL |
3066 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
3067 | cpu, pageset->pcp.high, | |
3068 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
3069 | } |
3070 | } | |
3071 | ||
a731286d KM |
3072 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
3073 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 3074 | " unevictable:%lu" |
b76146ed | 3075 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 3076 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a BZ |
3077 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
3078 | " free_cma:%lu\n", | |
4f98a2fe | 3079 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 3080 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
3081 | global_page_state(NR_ISOLATED_ANON), |
3082 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 3083 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 3084 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 3085 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 3086 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 3087 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 3088 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 3089 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
3090 | global_page_state(NR_SLAB_RECLAIMABLE), |
3091 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 3092 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 3093 | global_page_state(NR_SHMEM), |
a25700a5 | 3094 | global_page_state(NR_PAGETABLE), |
d1ce749a BZ |
3095 | global_page_state(NR_BOUNCE), |
3096 | global_page_state(NR_FREE_CMA_PAGES)); | |
1da177e4 | 3097 | |
ee99c71c | 3098 | for_each_populated_zone(zone) { |
1da177e4 LT |
3099 | int i; |
3100 | ||
7bf02ea2 | 3101 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3102 | continue; |
1da177e4 LT |
3103 | show_node(zone); |
3104 | printk("%s" | |
3105 | " free:%lukB" | |
3106 | " min:%lukB" | |
3107 | " low:%lukB" | |
3108 | " high:%lukB" | |
4f98a2fe RR |
3109 | " active_anon:%lukB" |
3110 | " inactive_anon:%lukB" | |
3111 | " active_file:%lukB" | |
3112 | " inactive_file:%lukB" | |
7b854121 | 3113 | " unevictable:%lukB" |
a731286d KM |
3114 | " isolated(anon):%lukB" |
3115 | " isolated(file):%lukB" | |
1da177e4 | 3116 | " present:%lukB" |
9feedc9d | 3117 | " managed:%lukB" |
4a0aa73f KM |
3118 | " mlocked:%lukB" |
3119 | " dirty:%lukB" | |
3120 | " writeback:%lukB" | |
3121 | " mapped:%lukB" | |
4b02108a | 3122 | " shmem:%lukB" |
4a0aa73f KM |
3123 | " slab_reclaimable:%lukB" |
3124 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3125 | " kernel_stack:%lukB" |
4a0aa73f KM |
3126 | " pagetables:%lukB" |
3127 | " unstable:%lukB" | |
3128 | " bounce:%lukB" | |
d1ce749a | 3129 | " free_cma:%lukB" |
4a0aa73f | 3130 | " writeback_tmp:%lukB" |
1da177e4 LT |
3131 | " pages_scanned:%lu" |
3132 | " all_unreclaimable? %s" | |
3133 | "\n", | |
3134 | zone->name, | |
88f5acf8 | 3135 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3136 | K(min_wmark_pages(zone)), |
3137 | K(low_wmark_pages(zone)), | |
3138 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3139 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3140 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3141 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3142 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3143 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3144 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3145 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3146 | K(zone->present_pages), |
9feedc9d | 3147 | K(zone->managed_pages), |
4a0aa73f KM |
3148 | K(zone_page_state(zone, NR_MLOCK)), |
3149 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3150 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3151 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3152 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3153 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3154 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3155 | zone_page_state(zone, NR_KERNEL_STACK) * |
3156 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3157 | K(zone_page_state(zone, NR_PAGETABLE)), |
3158 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3159 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1ce749a | 3160 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3161 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
1da177e4 | 3162 | zone->pages_scanned, |
6e543d57 | 3163 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
3164 | ); |
3165 | printk("lowmem_reserve[]:"); | |
3166 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3167 | printk(" %lu", zone->lowmem_reserve[i]); | |
3168 | printk("\n"); | |
3169 | } | |
3170 | ||
ee99c71c | 3171 | for_each_populated_zone(zone) { |
b8af2941 | 3172 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
377e4f16 | 3173 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3174 | |
7bf02ea2 | 3175 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3176 | continue; |
1da177e4 LT |
3177 | show_node(zone); |
3178 | printk("%s: ", zone->name); | |
1da177e4 LT |
3179 | |
3180 | spin_lock_irqsave(&zone->lock, flags); | |
3181 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3182 | struct free_area *area = &zone->free_area[order]; |
3183 | int type; | |
3184 | ||
3185 | nr[order] = area->nr_free; | |
8f9de51a | 3186 | total += nr[order] << order; |
377e4f16 RV |
3187 | |
3188 | types[order] = 0; | |
3189 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3190 | if (!list_empty(&area->free_list[type])) | |
3191 | types[order] |= 1 << type; | |
3192 | } | |
1da177e4 LT |
3193 | } |
3194 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3195 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3196 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3197 | if (nr[order]) |
3198 | show_migration_types(types[order]); | |
3199 | } | |
1da177e4 LT |
3200 | printk("= %lukB\n", K(total)); |
3201 | } | |
3202 | ||
949f7ec5 DR |
3203 | hugetlb_show_meminfo(); |
3204 | ||
e6f3602d LW |
3205 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
3206 | ||
1da177e4 LT |
3207 | show_swap_cache_info(); |
3208 | } | |
3209 | ||
19770b32 MG |
3210 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
3211 | { | |
3212 | zoneref->zone = zone; | |
3213 | zoneref->zone_idx = zone_idx(zone); | |
3214 | } | |
3215 | ||
1da177e4 LT |
3216 | /* |
3217 | * Builds allocation fallback zone lists. | |
1a93205b CL |
3218 | * |
3219 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 3220 | */ |
f0c0b2b8 | 3221 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 3222 | int nr_zones) |
1da177e4 | 3223 | { |
1a93205b | 3224 | struct zone *zone; |
bc732f1d | 3225 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
3226 | |
3227 | do { | |
2f6726e5 | 3228 | zone_type--; |
070f8032 | 3229 | zone = pgdat->node_zones + zone_type; |
1a93205b | 3230 | if (populated_zone(zone)) { |
dd1a239f MG |
3231 | zoneref_set_zone(zone, |
3232 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 3233 | check_highest_zone(zone_type); |
1da177e4 | 3234 | } |
2f6726e5 | 3235 | } while (zone_type); |
bc732f1d | 3236 | |
070f8032 | 3237 | return nr_zones; |
1da177e4 LT |
3238 | } |
3239 | ||
f0c0b2b8 KH |
3240 | |
3241 | /* | |
3242 | * zonelist_order: | |
3243 | * 0 = automatic detection of better ordering. | |
3244 | * 1 = order by ([node] distance, -zonetype) | |
3245 | * 2 = order by (-zonetype, [node] distance) | |
3246 | * | |
3247 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
3248 | * the same zonelist. So only NUMA can configure this param. | |
3249 | */ | |
3250 | #define ZONELIST_ORDER_DEFAULT 0 | |
3251 | #define ZONELIST_ORDER_NODE 1 | |
3252 | #define ZONELIST_ORDER_ZONE 2 | |
3253 | ||
3254 | /* zonelist order in the kernel. | |
3255 | * set_zonelist_order() will set this to NODE or ZONE. | |
3256 | */ | |
3257 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3258 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
3259 | ||
3260 | ||
1da177e4 | 3261 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
3262 | /* The value user specified ....changed by config */ |
3263 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3264 | /* string for sysctl */ | |
3265 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
3266 | char numa_zonelist_order[16] = "default"; | |
3267 | ||
3268 | /* | |
3269 | * interface for configure zonelist ordering. | |
3270 | * command line option "numa_zonelist_order" | |
3271 | * = "[dD]efault - default, automatic configuration. | |
3272 | * = "[nN]ode - order by node locality, then by zone within node | |
3273 | * = "[zZ]one - order by zone, then by locality within zone | |
3274 | */ | |
3275 | ||
3276 | static int __parse_numa_zonelist_order(char *s) | |
3277 | { | |
3278 | if (*s == 'd' || *s == 'D') { | |
3279 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3280 | } else if (*s == 'n' || *s == 'N') { | |
3281 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
3282 | } else if (*s == 'z' || *s == 'Z') { | |
3283 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
3284 | } else { | |
3285 | printk(KERN_WARNING | |
3286 | "Ignoring invalid numa_zonelist_order value: " | |
3287 | "%s\n", s); | |
3288 | return -EINVAL; | |
3289 | } | |
3290 | return 0; | |
3291 | } | |
3292 | ||
3293 | static __init int setup_numa_zonelist_order(char *s) | |
3294 | { | |
ecb256f8 VL |
3295 | int ret; |
3296 | ||
3297 | if (!s) | |
3298 | return 0; | |
3299 | ||
3300 | ret = __parse_numa_zonelist_order(s); | |
3301 | if (ret == 0) | |
3302 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
3303 | ||
3304 | return ret; | |
f0c0b2b8 KH |
3305 | } |
3306 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
3307 | ||
3308 | /* | |
3309 | * sysctl handler for numa_zonelist_order | |
3310 | */ | |
3311 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 3312 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
3313 | loff_t *ppos) |
3314 | { | |
3315 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
3316 | int ret; | |
443c6f14 | 3317 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 3318 | |
443c6f14 | 3319 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
3320 | if (write) { |
3321 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
3322 | ret = -EINVAL; | |
3323 | goto out; | |
3324 | } | |
3325 | strcpy(saved_string, (char *)table->data); | |
3326 | } | |
8d65af78 | 3327 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 3328 | if (ret) |
443c6f14 | 3329 | goto out; |
f0c0b2b8 KH |
3330 | if (write) { |
3331 | int oldval = user_zonelist_order; | |
dacbde09 CG |
3332 | |
3333 | ret = __parse_numa_zonelist_order((char *)table->data); | |
3334 | if (ret) { | |
f0c0b2b8 KH |
3335 | /* |
3336 | * bogus value. restore saved string | |
3337 | */ | |
dacbde09 | 3338 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
3339 | NUMA_ZONELIST_ORDER_LEN); |
3340 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
3341 | } else if (oldval != user_zonelist_order) { |
3342 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 3343 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
3344 | mutex_unlock(&zonelists_mutex); |
3345 | } | |
f0c0b2b8 | 3346 | } |
443c6f14 AK |
3347 | out: |
3348 | mutex_unlock(&zl_order_mutex); | |
3349 | return ret; | |
f0c0b2b8 KH |
3350 | } |
3351 | ||
3352 | ||
62bc62a8 | 3353 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
3354 | static int node_load[MAX_NUMNODES]; |
3355 | ||
1da177e4 | 3356 | /** |
4dc3b16b | 3357 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
3358 | * @node: node whose fallback list we're appending |
3359 | * @used_node_mask: nodemask_t of already used nodes | |
3360 | * | |
3361 | * We use a number of factors to determine which is the next node that should | |
3362 | * appear on a given node's fallback list. The node should not have appeared | |
3363 | * already in @node's fallback list, and it should be the next closest node | |
3364 | * according to the distance array (which contains arbitrary distance values | |
3365 | * from each node to each node in the system), and should also prefer nodes | |
3366 | * with no CPUs, since presumably they'll have very little allocation pressure | |
3367 | * on them otherwise. | |
3368 | * It returns -1 if no node is found. | |
3369 | */ | |
f0c0b2b8 | 3370 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 3371 | { |
4cf808eb | 3372 | int n, val; |
1da177e4 | 3373 | int min_val = INT_MAX; |
00ef2d2f | 3374 | int best_node = NUMA_NO_NODE; |
a70f7302 | 3375 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 3376 | |
4cf808eb LT |
3377 | /* Use the local node if we haven't already */ |
3378 | if (!node_isset(node, *used_node_mask)) { | |
3379 | node_set(node, *used_node_mask); | |
3380 | return node; | |
3381 | } | |
1da177e4 | 3382 | |
4b0ef1fe | 3383 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
3384 | |
3385 | /* Don't want a node to appear more than once */ | |
3386 | if (node_isset(n, *used_node_mask)) | |
3387 | continue; | |
3388 | ||
1da177e4 LT |
3389 | /* Use the distance array to find the distance */ |
3390 | val = node_distance(node, n); | |
3391 | ||
4cf808eb LT |
3392 | /* Penalize nodes under us ("prefer the next node") */ |
3393 | val += (n < node); | |
3394 | ||
1da177e4 | 3395 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
3396 | tmp = cpumask_of_node(n); |
3397 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
3398 | val += PENALTY_FOR_NODE_WITH_CPUS; |
3399 | ||
3400 | /* Slight preference for less loaded node */ | |
3401 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
3402 | val += node_load[n]; | |
3403 | ||
3404 | if (val < min_val) { | |
3405 | min_val = val; | |
3406 | best_node = n; | |
3407 | } | |
3408 | } | |
3409 | ||
3410 | if (best_node >= 0) | |
3411 | node_set(best_node, *used_node_mask); | |
3412 | ||
3413 | return best_node; | |
3414 | } | |
3415 | ||
f0c0b2b8 KH |
3416 | |
3417 | /* | |
3418 | * Build zonelists ordered by node and zones within node. | |
3419 | * This results in maximum locality--normal zone overflows into local | |
3420 | * DMA zone, if any--but risks exhausting DMA zone. | |
3421 | */ | |
3422 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 3423 | { |
f0c0b2b8 | 3424 | int j; |
1da177e4 | 3425 | struct zonelist *zonelist; |
f0c0b2b8 | 3426 | |
54a6eb5c | 3427 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 3428 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 3429 | ; |
bc732f1d | 3430 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
3431 | zonelist->_zonerefs[j].zone = NULL; |
3432 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
3433 | } |
3434 | ||
523b9458 CL |
3435 | /* |
3436 | * Build gfp_thisnode zonelists | |
3437 | */ | |
3438 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3439 | { | |
523b9458 CL |
3440 | int j; |
3441 | struct zonelist *zonelist; | |
3442 | ||
54a6eb5c | 3443 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 3444 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
3445 | zonelist->_zonerefs[j].zone = NULL; |
3446 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3447 | } |
3448 | ||
f0c0b2b8 KH |
3449 | /* |
3450 | * Build zonelists ordered by zone and nodes within zones. | |
3451 | * This results in conserving DMA zone[s] until all Normal memory is | |
3452 | * exhausted, but results in overflowing to remote node while memory | |
3453 | * may still exist in local DMA zone. | |
3454 | */ | |
3455 | static int node_order[MAX_NUMNODES]; | |
3456 | ||
3457 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3458 | { | |
f0c0b2b8 KH |
3459 | int pos, j, node; |
3460 | int zone_type; /* needs to be signed */ | |
3461 | struct zone *z; | |
3462 | struct zonelist *zonelist; | |
3463 | ||
54a6eb5c MG |
3464 | zonelist = &pgdat->node_zonelists[0]; |
3465 | pos = 0; | |
3466 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3467 | for (j = 0; j < nr_nodes; j++) { | |
3468 | node = node_order[j]; | |
3469 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3470 | if (populated_zone(z)) { | |
dd1a239f MG |
3471 | zoneref_set_zone(z, |
3472 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3473 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3474 | } |
3475 | } | |
f0c0b2b8 | 3476 | } |
dd1a239f MG |
3477 | zonelist->_zonerefs[pos].zone = NULL; |
3478 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3479 | } |
3480 | ||
3481 | static int default_zonelist_order(void) | |
3482 | { | |
3483 | int nid, zone_type; | |
b8af2941 | 3484 | unsigned long low_kmem_size, total_size; |
f0c0b2b8 KH |
3485 | struct zone *z; |
3486 | int average_size; | |
3487 | /* | |
b8af2941 | 3488 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
f0c0b2b8 KH |
3489 | * If they are really small and used heavily, the system can fall |
3490 | * into OOM very easily. | |
e325c90f | 3491 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
f0c0b2b8 KH |
3492 | */ |
3493 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
3494 | low_kmem_size = 0; | |
3495 | total_size = 0; | |
3496 | for_each_online_node(nid) { | |
3497 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3498 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3499 | if (populated_zone(z)) { | |
3500 | if (zone_type < ZONE_NORMAL) | |
4f9f4774 JL |
3501 | low_kmem_size += z->managed_pages; |
3502 | total_size += z->managed_pages; | |
e325c90f DR |
3503 | } else if (zone_type == ZONE_NORMAL) { |
3504 | /* | |
3505 | * If any node has only lowmem, then node order | |
3506 | * is preferred to allow kernel allocations | |
3507 | * locally; otherwise, they can easily infringe | |
3508 | * on other nodes when there is an abundance of | |
3509 | * lowmem available to allocate from. | |
3510 | */ | |
3511 | return ZONELIST_ORDER_NODE; | |
f0c0b2b8 KH |
3512 | } |
3513 | } | |
3514 | } | |
3515 | if (!low_kmem_size || /* there are no DMA area. */ | |
3516 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
3517 | return ZONELIST_ORDER_NODE; | |
3518 | /* | |
3519 | * look into each node's config. | |
b8af2941 PK |
3520 | * If there is a node whose DMA/DMA32 memory is very big area on |
3521 | * local memory, NODE_ORDER may be suitable. | |
3522 | */ | |
37b07e41 | 3523 | average_size = total_size / |
4b0ef1fe | 3524 | (nodes_weight(node_states[N_MEMORY]) + 1); |
f0c0b2b8 KH |
3525 | for_each_online_node(nid) { |
3526 | low_kmem_size = 0; | |
3527 | total_size = 0; | |
3528 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3529 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3530 | if (populated_zone(z)) { | |
3531 | if (zone_type < ZONE_NORMAL) | |
3532 | low_kmem_size += z->present_pages; | |
3533 | total_size += z->present_pages; | |
3534 | } | |
3535 | } | |
3536 | if (low_kmem_size && | |
3537 | total_size > average_size && /* ignore small node */ | |
3538 | low_kmem_size > total_size * 70/100) | |
3539 | return ZONELIST_ORDER_NODE; | |
3540 | } | |
3541 | return ZONELIST_ORDER_ZONE; | |
3542 | } | |
3543 | ||
3544 | static void set_zonelist_order(void) | |
3545 | { | |
3546 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3547 | current_zonelist_order = default_zonelist_order(); | |
3548 | else | |
3549 | current_zonelist_order = user_zonelist_order; | |
3550 | } | |
3551 | ||
3552 | static void build_zonelists(pg_data_t *pgdat) | |
3553 | { | |
3554 | int j, node, load; | |
3555 | enum zone_type i; | |
1da177e4 | 3556 | nodemask_t used_mask; |
f0c0b2b8 KH |
3557 | int local_node, prev_node; |
3558 | struct zonelist *zonelist; | |
3559 | int order = current_zonelist_order; | |
1da177e4 LT |
3560 | |
3561 | /* initialize zonelists */ | |
523b9458 | 3562 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3563 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3564 | zonelist->_zonerefs[0].zone = NULL; |
3565 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3566 | } |
3567 | ||
3568 | /* NUMA-aware ordering of nodes */ | |
3569 | local_node = pgdat->node_id; | |
62bc62a8 | 3570 | load = nr_online_nodes; |
1da177e4 LT |
3571 | prev_node = local_node; |
3572 | nodes_clear(used_mask); | |
f0c0b2b8 | 3573 | |
f0c0b2b8 KH |
3574 | memset(node_order, 0, sizeof(node_order)); |
3575 | j = 0; | |
3576 | ||
1da177e4 LT |
3577 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
3578 | /* | |
3579 | * We don't want to pressure a particular node. | |
3580 | * So adding penalty to the first node in same | |
3581 | * distance group to make it round-robin. | |
3582 | */ | |
957f822a DR |
3583 | if (node_distance(local_node, node) != |
3584 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
3585 | node_load[node] = load; |
3586 | ||
1da177e4 LT |
3587 | prev_node = node; |
3588 | load--; | |
f0c0b2b8 KH |
3589 | if (order == ZONELIST_ORDER_NODE) |
3590 | build_zonelists_in_node_order(pgdat, node); | |
3591 | else | |
3592 | node_order[j++] = node; /* remember order */ | |
3593 | } | |
1da177e4 | 3594 | |
f0c0b2b8 KH |
3595 | if (order == ZONELIST_ORDER_ZONE) { |
3596 | /* calculate node order -- i.e., DMA last! */ | |
3597 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3598 | } |
523b9458 CL |
3599 | |
3600 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3601 | } |
3602 | ||
9276b1bc | 3603 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3604 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3605 | { |
54a6eb5c MG |
3606 | struct zonelist *zonelist; |
3607 | struct zonelist_cache *zlc; | |
dd1a239f | 3608 | struct zoneref *z; |
9276b1bc | 3609 | |
54a6eb5c MG |
3610 | zonelist = &pgdat->node_zonelists[0]; |
3611 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3612 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3613 | for (z = zonelist->_zonerefs; z->zone; z++) |
3614 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
3615 | } |
3616 | ||
7aac7898 LS |
3617 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3618 | /* | |
3619 | * Return node id of node used for "local" allocations. | |
3620 | * I.e., first node id of first zone in arg node's generic zonelist. | |
3621 | * Used for initializing percpu 'numa_mem', which is used primarily | |
3622 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
3623 | */ | |
3624 | int local_memory_node(int node) | |
3625 | { | |
3626 | struct zone *zone; | |
3627 | ||
3628 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
3629 | gfp_zone(GFP_KERNEL), | |
3630 | NULL, | |
3631 | &zone); | |
3632 | return zone->node; | |
3633 | } | |
3634 | #endif | |
f0c0b2b8 | 3635 | |
1da177e4 LT |
3636 | #else /* CONFIG_NUMA */ |
3637 | ||
f0c0b2b8 KH |
3638 | static void set_zonelist_order(void) |
3639 | { | |
3640 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
3641 | } | |
3642 | ||
3643 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 3644 | { |
19655d34 | 3645 | int node, local_node; |
54a6eb5c MG |
3646 | enum zone_type j; |
3647 | struct zonelist *zonelist; | |
1da177e4 LT |
3648 | |
3649 | local_node = pgdat->node_id; | |
1da177e4 | 3650 | |
54a6eb5c | 3651 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 3652 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 3653 | |
54a6eb5c MG |
3654 | /* |
3655 | * Now we build the zonelist so that it contains the zones | |
3656 | * of all the other nodes. | |
3657 | * We don't want to pressure a particular node, so when | |
3658 | * building the zones for node N, we make sure that the | |
3659 | * zones coming right after the local ones are those from | |
3660 | * node N+1 (modulo N) | |
3661 | */ | |
3662 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3663 | if (!node_online(node)) | |
3664 | continue; | |
bc732f1d | 3665 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 3666 | } |
54a6eb5c MG |
3667 | for (node = 0; node < local_node; node++) { |
3668 | if (!node_online(node)) | |
3669 | continue; | |
bc732f1d | 3670 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
3671 | } |
3672 | ||
dd1a239f MG |
3673 | zonelist->_zonerefs[j].zone = NULL; |
3674 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3675 | } |
3676 | ||
9276b1bc | 3677 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3678 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3679 | { |
54a6eb5c | 3680 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3681 | } |
3682 | ||
1da177e4 LT |
3683 | #endif /* CONFIG_NUMA */ |
3684 | ||
99dcc3e5 CL |
3685 | /* |
3686 | * Boot pageset table. One per cpu which is going to be used for all | |
3687 | * zones and all nodes. The parameters will be set in such a way | |
3688 | * that an item put on a list will immediately be handed over to | |
3689 | * the buddy list. This is safe since pageset manipulation is done | |
3690 | * with interrupts disabled. | |
3691 | * | |
3692 | * The boot_pagesets must be kept even after bootup is complete for | |
3693 | * unused processors and/or zones. They do play a role for bootstrapping | |
3694 | * hotplugged processors. | |
3695 | * | |
3696 | * zoneinfo_show() and maybe other functions do | |
3697 | * not check if the processor is online before following the pageset pointer. | |
3698 | * Other parts of the kernel may not check if the zone is available. | |
3699 | */ | |
3700 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3701 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3702 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3703 | |
4eaf3f64 HL |
3704 | /* |
3705 | * Global mutex to protect against size modification of zonelists | |
3706 | * as well as to serialize pageset setup for the new populated zone. | |
3707 | */ | |
3708 | DEFINE_MUTEX(zonelists_mutex); | |
3709 | ||
9b1a4d38 | 3710 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 3711 | static int __build_all_zonelists(void *data) |
1da177e4 | 3712 | { |
6811378e | 3713 | int nid; |
99dcc3e5 | 3714 | int cpu; |
9adb62a5 | 3715 | pg_data_t *self = data; |
9276b1bc | 3716 | |
7f9cfb31 BL |
3717 | #ifdef CONFIG_NUMA |
3718 | memset(node_load, 0, sizeof(node_load)); | |
3719 | #endif | |
9adb62a5 JL |
3720 | |
3721 | if (self && !node_online(self->node_id)) { | |
3722 | build_zonelists(self); | |
3723 | build_zonelist_cache(self); | |
3724 | } | |
3725 | ||
9276b1bc | 3726 | for_each_online_node(nid) { |
7ea1530a CL |
3727 | pg_data_t *pgdat = NODE_DATA(nid); |
3728 | ||
3729 | build_zonelists(pgdat); | |
3730 | build_zonelist_cache(pgdat); | |
9276b1bc | 3731 | } |
99dcc3e5 CL |
3732 | |
3733 | /* | |
3734 | * Initialize the boot_pagesets that are going to be used | |
3735 | * for bootstrapping processors. The real pagesets for | |
3736 | * each zone will be allocated later when the per cpu | |
3737 | * allocator is available. | |
3738 | * | |
3739 | * boot_pagesets are used also for bootstrapping offline | |
3740 | * cpus if the system is already booted because the pagesets | |
3741 | * are needed to initialize allocators on a specific cpu too. | |
3742 | * F.e. the percpu allocator needs the page allocator which | |
3743 | * needs the percpu allocator in order to allocate its pagesets | |
3744 | * (a chicken-egg dilemma). | |
3745 | */ | |
7aac7898 | 3746 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3747 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3748 | ||
7aac7898 LS |
3749 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3750 | /* | |
3751 | * We now know the "local memory node" for each node-- | |
3752 | * i.e., the node of the first zone in the generic zonelist. | |
3753 | * Set up numa_mem percpu variable for on-line cpus. During | |
3754 | * boot, only the boot cpu should be on-line; we'll init the | |
3755 | * secondary cpus' numa_mem as they come on-line. During | |
3756 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3757 | */ | |
3758 | if (cpu_online(cpu)) | |
3759 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3760 | #endif | |
3761 | } | |
3762 | ||
6811378e YG |
3763 | return 0; |
3764 | } | |
3765 | ||
4eaf3f64 HL |
3766 | /* |
3767 | * Called with zonelists_mutex held always | |
3768 | * unless system_state == SYSTEM_BOOTING. | |
3769 | */ | |
9adb62a5 | 3770 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 3771 | { |
f0c0b2b8 KH |
3772 | set_zonelist_order(); |
3773 | ||
6811378e | 3774 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3775 | __build_all_zonelists(NULL); |
68ad8df4 | 3776 | mminit_verify_zonelist(); |
6811378e YG |
3777 | cpuset_init_current_mems_allowed(); |
3778 | } else { | |
e9959f0f | 3779 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
3780 | if (zone) |
3781 | setup_zone_pageset(zone); | |
e9959f0f | 3782 | #endif |
dd1895e2 CS |
3783 | /* we have to stop all cpus to guarantee there is no user |
3784 | of zonelist */ | |
9adb62a5 | 3785 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
3786 | /* cpuset refresh routine should be here */ |
3787 | } | |
bd1e22b8 | 3788 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3789 | /* |
3790 | * Disable grouping by mobility if the number of pages in the | |
3791 | * system is too low to allow the mechanism to work. It would be | |
3792 | * more accurate, but expensive to check per-zone. This check is | |
3793 | * made on memory-hotadd so a system can start with mobility | |
3794 | * disabled and enable it later | |
3795 | */ | |
d9c23400 | 3796 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3797 | page_group_by_mobility_disabled = 1; |
3798 | else | |
3799 | page_group_by_mobility_disabled = 0; | |
3800 | ||
3801 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
3802 | "Total pages: %ld\n", | |
62bc62a8 | 3803 | nr_online_nodes, |
f0c0b2b8 | 3804 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3805 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3806 | vm_total_pages); |
3807 | #ifdef CONFIG_NUMA | |
3808 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
3809 | #endif | |
1da177e4 LT |
3810 | } |
3811 | ||
3812 | /* | |
3813 | * Helper functions to size the waitqueue hash table. | |
3814 | * Essentially these want to choose hash table sizes sufficiently | |
3815 | * large so that collisions trying to wait on pages are rare. | |
3816 | * But in fact, the number of active page waitqueues on typical | |
3817 | * systems is ridiculously low, less than 200. So this is even | |
3818 | * conservative, even though it seems large. | |
3819 | * | |
3820 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
3821 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
3822 | */ | |
3823 | #define PAGES_PER_WAITQUEUE 256 | |
3824 | ||
cca448fe | 3825 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 3826 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
3827 | { |
3828 | unsigned long size = 1; | |
3829 | ||
3830 | pages /= PAGES_PER_WAITQUEUE; | |
3831 | ||
3832 | while (size < pages) | |
3833 | size <<= 1; | |
3834 | ||
3835 | /* | |
3836 | * Once we have dozens or even hundreds of threads sleeping | |
3837 | * on IO we've got bigger problems than wait queue collision. | |
3838 | * Limit the size of the wait table to a reasonable size. | |
3839 | */ | |
3840 | size = min(size, 4096UL); | |
3841 | ||
3842 | return max(size, 4UL); | |
3843 | } | |
cca448fe YG |
3844 | #else |
3845 | /* | |
3846 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
3847 | * a suitable size for its wait_table. So we use the maximum size now. | |
3848 | * | |
3849 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
3850 | * | |
3851 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
3852 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
3853 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
3854 | * | |
3855 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
3856 | * or more by the traditional way. (See above). It equals: | |
3857 | * | |
3858 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
3859 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
3860 | * powerpc (64K page size) : = (32G +16M)byte. | |
3861 | */ | |
3862 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
3863 | { | |
3864 | return 4096UL; | |
3865 | } | |
3866 | #endif | |
1da177e4 LT |
3867 | |
3868 | /* | |
3869 | * This is an integer logarithm so that shifts can be used later | |
3870 | * to extract the more random high bits from the multiplicative | |
3871 | * hash function before the remainder is taken. | |
3872 | */ | |
3873 | static inline unsigned long wait_table_bits(unsigned long size) | |
3874 | { | |
3875 | return ffz(~size); | |
3876 | } | |
3877 | ||
6d3163ce AH |
3878 | /* |
3879 | * Check if a pageblock contains reserved pages | |
3880 | */ | |
3881 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
3882 | { | |
3883 | unsigned long pfn; | |
3884 | ||
3885 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
3886 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
3887 | return 1; | |
3888 | } | |
3889 | return 0; | |
3890 | } | |
3891 | ||
56fd56b8 | 3892 | /* |
d9c23400 | 3893 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
3894 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
3895 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
3896 | * higher will lead to a bigger reserve which will get freed as contiguous |
3897 | * blocks as reclaim kicks in | |
3898 | */ | |
3899 | static void setup_zone_migrate_reserve(struct zone *zone) | |
3900 | { | |
6d3163ce | 3901 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 3902 | struct page *page; |
78986a67 MG |
3903 | unsigned long block_migratetype; |
3904 | int reserve; | |
56fd56b8 | 3905 | |
d0215638 MH |
3906 | /* |
3907 | * Get the start pfn, end pfn and the number of blocks to reserve | |
3908 | * We have to be careful to be aligned to pageblock_nr_pages to | |
3909 | * make sure that we always check pfn_valid for the first page in | |
3910 | * the block. | |
3911 | */ | |
56fd56b8 | 3912 | start_pfn = zone->zone_start_pfn; |
108bcc96 | 3913 | end_pfn = zone_end_pfn(zone); |
d0215638 | 3914 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 3915 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 3916 | pageblock_order; |
56fd56b8 | 3917 | |
78986a67 MG |
3918 | /* |
3919 | * Reserve blocks are generally in place to help high-order atomic | |
3920 | * allocations that are short-lived. A min_free_kbytes value that | |
3921 | * would result in more than 2 reserve blocks for atomic allocations | |
3922 | * is assumed to be in place to help anti-fragmentation for the | |
3923 | * future allocation of hugepages at runtime. | |
3924 | */ | |
3925 | reserve = min(2, reserve); | |
3926 | ||
d9c23400 | 3927 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
3928 | if (!pfn_valid(pfn)) |
3929 | continue; | |
3930 | page = pfn_to_page(pfn); | |
3931 | ||
344c790e AL |
3932 | /* Watch out for overlapping nodes */ |
3933 | if (page_to_nid(page) != zone_to_nid(zone)) | |
3934 | continue; | |
3935 | ||
56fd56b8 MG |
3936 | block_migratetype = get_pageblock_migratetype(page); |
3937 | ||
938929f1 MG |
3938 | /* Only test what is necessary when the reserves are not met */ |
3939 | if (reserve > 0) { | |
3940 | /* | |
3941 | * Blocks with reserved pages will never free, skip | |
3942 | * them. | |
3943 | */ | |
3944 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
3945 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
3946 | continue; | |
56fd56b8 | 3947 | |
938929f1 MG |
3948 | /* If this block is reserved, account for it */ |
3949 | if (block_migratetype == MIGRATE_RESERVE) { | |
3950 | reserve--; | |
3951 | continue; | |
3952 | } | |
3953 | ||
3954 | /* Suitable for reserving if this block is movable */ | |
3955 | if (block_migratetype == MIGRATE_MOVABLE) { | |
3956 | set_pageblock_migratetype(page, | |
3957 | MIGRATE_RESERVE); | |
3958 | move_freepages_block(zone, page, | |
3959 | MIGRATE_RESERVE); | |
3960 | reserve--; | |
3961 | continue; | |
3962 | } | |
56fd56b8 MG |
3963 | } |
3964 | ||
3965 | /* | |
3966 | * If the reserve is met and this is a previous reserved block, | |
3967 | * take it back | |
3968 | */ | |
3969 | if (block_migratetype == MIGRATE_RESERVE) { | |
3970 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
3971 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
3972 | } | |
3973 | } | |
3974 | } | |
ac0e5b7a | 3975 | |
1da177e4 LT |
3976 | /* |
3977 | * Initially all pages are reserved - free ones are freed | |
3978 | * up by free_all_bootmem() once the early boot process is | |
3979 | * done. Non-atomic initialization, single-pass. | |
3980 | */ | |
c09b4240 | 3981 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 3982 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 3983 | { |
1da177e4 | 3984 | struct page *page; |
29751f69 AW |
3985 | unsigned long end_pfn = start_pfn + size; |
3986 | unsigned long pfn; | |
86051ca5 | 3987 | struct zone *z; |
1da177e4 | 3988 | |
22b31eec HD |
3989 | if (highest_memmap_pfn < end_pfn - 1) |
3990 | highest_memmap_pfn = end_pfn - 1; | |
3991 | ||
86051ca5 | 3992 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 3993 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
3994 | /* |
3995 | * There can be holes in boot-time mem_map[]s | |
3996 | * handed to this function. They do not | |
3997 | * exist on hotplugged memory. | |
3998 | */ | |
3999 | if (context == MEMMAP_EARLY) { | |
4000 | if (!early_pfn_valid(pfn)) | |
4001 | continue; | |
4002 | if (!early_pfn_in_nid(pfn, nid)) | |
4003 | continue; | |
4004 | } | |
d41dee36 AW |
4005 | page = pfn_to_page(pfn); |
4006 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 4007 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 4008 | init_page_count(page); |
22b751c3 | 4009 | page_mapcount_reset(page); |
90572890 | 4010 | page_cpupid_reset_last(page); |
1da177e4 | 4011 | SetPageReserved(page); |
b2a0ac88 MG |
4012 | /* |
4013 | * Mark the block movable so that blocks are reserved for | |
4014 | * movable at startup. This will force kernel allocations | |
4015 | * to reserve their blocks rather than leaking throughout | |
4016 | * the address space during boot when many long-lived | |
56fd56b8 MG |
4017 | * kernel allocations are made. Later some blocks near |
4018 | * the start are marked MIGRATE_RESERVE by | |
4019 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
4020 | * |
4021 | * bitmap is created for zone's valid pfn range. but memmap | |
4022 | * can be created for invalid pages (for alignment) | |
4023 | * check here not to call set_pageblock_migratetype() against | |
4024 | * pfn out of zone. | |
b2a0ac88 | 4025 | */ |
86051ca5 | 4026 | if ((z->zone_start_pfn <= pfn) |
108bcc96 | 4027 | && (pfn < zone_end_pfn(z)) |
86051ca5 | 4028 | && !(pfn & (pageblock_nr_pages - 1))) |
56fd56b8 | 4029 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 4030 | |
1da177e4 LT |
4031 | INIT_LIST_HEAD(&page->lru); |
4032 | #ifdef WANT_PAGE_VIRTUAL | |
4033 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
4034 | if (!is_highmem_idx(zone)) | |
3212c6be | 4035 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 4036 | #endif |
1da177e4 LT |
4037 | } |
4038 | } | |
4039 | ||
1e548deb | 4040 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 4041 | { |
b2a0ac88 MG |
4042 | int order, t; |
4043 | for_each_migratetype_order(order, t) { | |
4044 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
4045 | zone->free_area[order].nr_free = 0; |
4046 | } | |
4047 | } | |
4048 | ||
4049 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
4050 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 4051 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
4052 | #endif |
4053 | ||
4ed7e022 | 4054 | static int __meminit zone_batchsize(struct zone *zone) |
e7c8d5c9 | 4055 | { |
3a6be87f | 4056 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
4057 | int batch; |
4058 | ||
4059 | /* | |
4060 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 4061 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
4062 | * |
4063 | * OK, so we don't know how big the cache is. So guess. | |
4064 | */ | |
b40da049 | 4065 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
4066 | if (batch * PAGE_SIZE > 512 * 1024) |
4067 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
4068 | batch /= 4; /* We effectively *= 4 below */ |
4069 | if (batch < 1) | |
4070 | batch = 1; | |
4071 | ||
4072 | /* | |
0ceaacc9 NP |
4073 | * Clamp the batch to a 2^n - 1 value. Having a power |
4074 | * of 2 value was found to be more likely to have | |
4075 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 4076 | * |
0ceaacc9 NP |
4077 | * For example if 2 tasks are alternately allocating |
4078 | * batches of pages, one task can end up with a lot | |
4079 | * of pages of one half of the possible page colors | |
4080 | * and the other with pages of the other colors. | |
e7c8d5c9 | 4081 | */ |
9155203a | 4082 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 4083 | |
e7c8d5c9 | 4084 | return batch; |
3a6be87f DH |
4085 | |
4086 | #else | |
4087 | /* The deferral and batching of frees should be suppressed under NOMMU | |
4088 | * conditions. | |
4089 | * | |
4090 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4091 | * of contiguous memory as there's no hardware page translation to | |
4092 | * assemble apparent contiguous memory from discontiguous pages. | |
4093 | * | |
4094 | * Queueing large contiguous runs of pages for batching, however, | |
4095 | * causes the pages to actually be freed in smaller chunks. As there | |
4096 | * can be a significant delay between the individual batches being | |
4097 | * recycled, this leads to the once large chunks of space being | |
4098 | * fragmented and becoming unavailable for high-order allocations. | |
4099 | */ | |
4100 | return 0; | |
4101 | #endif | |
e7c8d5c9 CL |
4102 | } |
4103 | ||
8d7a8fa9 CS |
4104 | /* |
4105 | * pcp->high and pcp->batch values are related and dependent on one another: | |
4106 | * ->batch must never be higher then ->high. | |
4107 | * The following function updates them in a safe manner without read side | |
4108 | * locking. | |
4109 | * | |
4110 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
4111 | * those fields changing asynchronously (acording the the above rule). | |
4112 | * | |
4113 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
4114 | * outside of boot time (or some other assurance that no concurrent updaters | |
4115 | * exist). | |
4116 | */ | |
4117 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
4118 | unsigned long batch) | |
4119 | { | |
4120 | /* start with a fail safe value for batch */ | |
4121 | pcp->batch = 1; | |
4122 | smp_wmb(); | |
4123 | ||
4124 | /* Update high, then batch, in order */ | |
4125 | pcp->high = high; | |
4126 | smp_wmb(); | |
4127 | ||
4128 | pcp->batch = batch; | |
4129 | } | |
4130 | ||
3664033c | 4131 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
4132 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
4133 | { | |
8d7a8fa9 | 4134 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
4135 | } |
4136 | ||
88c90dbc | 4137 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
4138 | { |
4139 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4140 | int migratetype; |
2caaad41 | 4141 | |
1c6fe946 MD |
4142 | memset(p, 0, sizeof(*p)); |
4143 | ||
3dfa5721 | 4144 | pcp = &p->pcp; |
2caaad41 | 4145 | pcp->count = 0; |
5f8dcc21 MG |
4146 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4147 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4148 | } |
4149 | ||
88c90dbc CS |
4150 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
4151 | { | |
4152 | pageset_init(p); | |
4153 | pageset_set_batch(p, batch); | |
4154 | } | |
4155 | ||
8ad4b1fb | 4156 | /* |
3664033c | 4157 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
4158 | * to the value high for the pageset p. |
4159 | */ | |
3664033c | 4160 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
4161 | unsigned long high) |
4162 | { | |
8d7a8fa9 CS |
4163 | unsigned long batch = max(1UL, high / 4); |
4164 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
4165 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 4166 | |
8d7a8fa9 | 4167 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
4168 | } |
4169 | ||
169f6c19 CS |
4170 | static void __meminit pageset_set_high_and_batch(struct zone *zone, |
4171 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 4172 | { |
56cef2b8 | 4173 | if (percpu_pagelist_fraction) |
3664033c | 4174 | pageset_set_high(pcp, |
56cef2b8 CS |
4175 | (zone->managed_pages / |
4176 | percpu_pagelist_fraction)); | |
4177 | else | |
4178 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
4179 | } | |
4180 | ||
169f6c19 CS |
4181 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
4182 | { | |
4183 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4184 | ||
4185 | pageset_init(pcp); | |
4186 | pageset_set_high_and_batch(zone, pcp); | |
4187 | } | |
4188 | ||
4ed7e022 | 4189 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4190 | { |
4191 | int cpu; | |
319774e2 | 4192 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
4193 | for_each_possible_cpu(cpu) |
4194 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
4195 | } |
4196 | ||
2caaad41 | 4197 | /* |
99dcc3e5 CL |
4198 | * Allocate per cpu pagesets and initialize them. |
4199 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4200 | */ |
99dcc3e5 | 4201 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4202 | { |
99dcc3e5 | 4203 | struct zone *zone; |
e7c8d5c9 | 4204 | |
319774e2 WF |
4205 | for_each_populated_zone(zone) |
4206 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4207 | } |
4208 | ||
577a32f6 | 4209 | static noinline __init_refok |
cca448fe | 4210 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4211 | { |
4212 | int i; | |
4213 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 4214 | size_t alloc_size; |
ed8ece2e DH |
4215 | |
4216 | /* | |
4217 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4218 | * per zone. | |
4219 | */ | |
02b694de YG |
4220 | zone->wait_table_hash_nr_entries = |
4221 | wait_table_hash_nr_entries(zone_size_pages); | |
4222 | zone->wait_table_bits = | |
4223 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4224 | alloc_size = zone->wait_table_hash_nr_entries |
4225 | * sizeof(wait_queue_head_t); | |
4226 | ||
cd94b9db | 4227 | if (!slab_is_available()) { |
cca448fe | 4228 | zone->wait_table = (wait_queue_head_t *) |
8f389a99 | 4229 | alloc_bootmem_node_nopanic(pgdat, alloc_size); |
cca448fe YG |
4230 | } else { |
4231 | /* | |
4232 | * This case means that a zone whose size was 0 gets new memory | |
4233 | * via memory hot-add. | |
4234 | * But it may be the case that a new node was hot-added. In | |
4235 | * this case vmalloc() will not be able to use this new node's | |
4236 | * memory - this wait_table must be initialized to use this new | |
4237 | * node itself as well. | |
4238 | * To use this new node's memory, further consideration will be | |
4239 | * necessary. | |
4240 | */ | |
8691f3a7 | 4241 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4242 | } |
4243 | if (!zone->wait_table) | |
4244 | return -ENOMEM; | |
ed8ece2e | 4245 | |
b8af2941 | 4246 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4247 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4248 | |
4249 | return 0; | |
ed8ece2e DH |
4250 | } |
4251 | ||
c09b4240 | 4252 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4253 | { |
99dcc3e5 CL |
4254 | /* |
4255 | * per cpu subsystem is not up at this point. The following code | |
4256 | * relies on the ability of the linker to provide the | |
4257 | * offset of a (static) per cpu variable into the per cpu area. | |
4258 | */ | |
4259 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4260 | |
b38a8725 | 4261 | if (populated_zone(zone)) |
99dcc3e5 CL |
4262 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4263 | zone->name, zone->present_pages, | |
4264 | zone_batchsize(zone)); | |
ed8ece2e DH |
4265 | } |
4266 | ||
4ed7e022 | 4267 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4268 | unsigned long zone_start_pfn, |
a2f3aa02 DH |
4269 | unsigned long size, |
4270 | enum memmap_context context) | |
ed8ece2e DH |
4271 | { |
4272 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4273 | int ret; |
4274 | ret = zone_wait_table_init(zone, size); | |
4275 | if (ret) | |
4276 | return ret; | |
ed8ece2e DH |
4277 | pgdat->nr_zones = zone_idx(zone) + 1; |
4278 | ||
ed8ece2e DH |
4279 | zone->zone_start_pfn = zone_start_pfn; |
4280 | ||
708614e6 MG |
4281 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4282 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4283 | pgdat->node_id, | |
4284 | (unsigned long)zone_idx(zone), | |
4285 | zone_start_pfn, (zone_start_pfn + size)); | |
4286 | ||
1e548deb | 4287 | zone_init_free_lists(zone); |
718127cc YG |
4288 | |
4289 | return 0; | |
ed8ece2e DH |
4290 | } |
4291 | ||
0ee332c1 | 4292 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
4293 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
4294 | /* | |
4295 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
4296 | * Architectures may implement their own version but if add_active_range() | |
4297 | * was used and there are no special requirements, this is a convenient | |
4298 | * alternative | |
4299 | */ | |
f2dbcfa7 | 4300 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d | 4301 | { |
c13291a5 | 4302 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 4303 | int nid; |
7c243c71 RA |
4304 | /* |
4305 | * NOTE: The following SMP-unsafe globals are only used early in boot | |
4306 | * when the kernel is running single-threaded. | |
4307 | */ | |
4308 | static unsigned long __meminitdata last_start_pfn, last_end_pfn; | |
4309 | static int __meminitdata last_nid; | |
4310 | ||
4311 | if (last_start_pfn <= pfn && pfn < last_end_pfn) | |
4312 | return last_nid; | |
c713216d | 4313 | |
e76b63f8 YL |
4314 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
4315 | if (nid != -1) { | |
4316 | last_start_pfn = start_pfn; | |
4317 | last_end_pfn = end_pfn; | |
4318 | last_nid = nid; | |
4319 | } | |
4320 | ||
4321 | return nid; | |
c713216d MG |
4322 | } |
4323 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
4324 | ||
f2dbcfa7 KH |
4325 | int __meminit early_pfn_to_nid(unsigned long pfn) |
4326 | { | |
cc2559bc KH |
4327 | int nid; |
4328 | ||
4329 | nid = __early_pfn_to_nid(pfn); | |
4330 | if (nid >= 0) | |
4331 | return nid; | |
4332 | /* just returns 0 */ | |
4333 | return 0; | |
f2dbcfa7 KH |
4334 | } |
4335 | ||
cc2559bc KH |
4336 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
4337 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
4338 | { | |
4339 | int nid; | |
4340 | ||
4341 | nid = __early_pfn_to_nid(pfn); | |
4342 | if (nid >= 0 && nid != node) | |
4343 | return false; | |
4344 | return true; | |
4345 | } | |
4346 | #endif | |
f2dbcfa7 | 4347 | |
c713216d MG |
4348 | /** |
4349 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
4350 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
4351 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
4352 | * |
4353 | * If an architecture guarantees that all ranges registered with | |
4354 | * add_active_ranges() contain no holes and may be freed, this | |
4355 | * this function may be used instead of calling free_bootmem() manually. | |
4356 | */ | |
c13291a5 | 4357 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 4358 | { |
c13291a5 TH |
4359 | unsigned long start_pfn, end_pfn; |
4360 | int i, this_nid; | |
edbe7d23 | 4361 | |
c13291a5 TH |
4362 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
4363 | start_pfn = min(start_pfn, max_low_pfn); | |
4364 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 4365 | |
c13291a5 TH |
4366 | if (start_pfn < end_pfn) |
4367 | free_bootmem_node(NODE_DATA(this_nid), | |
4368 | PFN_PHYS(start_pfn), | |
4369 | (end_pfn - start_pfn) << PAGE_SHIFT); | |
edbe7d23 | 4370 | } |
edbe7d23 | 4371 | } |
edbe7d23 | 4372 | |
c713216d MG |
4373 | /** |
4374 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 4375 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
4376 | * |
4377 | * If an architecture guarantees that all ranges registered with | |
4378 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 4379 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
4380 | */ |
4381 | void __init sparse_memory_present_with_active_regions(int nid) | |
4382 | { | |
c13291a5 TH |
4383 | unsigned long start_pfn, end_pfn; |
4384 | int i, this_nid; | |
c713216d | 4385 | |
c13291a5 TH |
4386 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
4387 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
4388 | } |
4389 | ||
4390 | /** | |
4391 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
4392 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
4393 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
4394 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
4395 | * |
4396 | * It returns the start and end page frame of a node based on information | |
4397 | * provided by an arch calling add_active_range(). If called for a node | |
4398 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 4399 | * PFNs will be 0. |
c713216d | 4400 | */ |
a3142c8e | 4401 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
4402 | unsigned long *start_pfn, unsigned long *end_pfn) |
4403 | { | |
c13291a5 | 4404 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 4405 | int i; |
c13291a5 | 4406 | |
c713216d MG |
4407 | *start_pfn = -1UL; |
4408 | *end_pfn = 0; | |
4409 | ||
c13291a5 TH |
4410 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
4411 | *start_pfn = min(*start_pfn, this_start_pfn); | |
4412 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
4413 | } |
4414 | ||
633c0666 | 4415 | if (*start_pfn == -1UL) |
c713216d | 4416 | *start_pfn = 0; |
c713216d MG |
4417 | } |
4418 | ||
2a1e274a MG |
4419 | /* |
4420 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
4421 | * assumption is made that zones within a node are ordered in monotonic | |
4422 | * increasing memory addresses so that the "highest" populated zone is used | |
4423 | */ | |
b69a7288 | 4424 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
4425 | { |
4426 | int zone_index; | |
4427 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
4428 | if (zone_index == ZONE_MOVABLE) | |
4429 | continue; | |
4430 | ||
4431 | if (arch_zone_highest_possible_pfn[zone_index] > | |
4432 | arch_zone_lowest_possible_pfn[zone_index]) | |
4433 | break; | |
4434 | } | |
4435 | ||
4436 | VM_BUG_ON(zone_index == -1); | |
4437 | movable_zone = zone_index; | |
4438 | } | |
4439 | ||
4440 | /* | |
4441 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 4442 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
4443 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
4444 | * in each node depending on the size of each node and how evenly kernelcore | |
4445 | * is distributed. This helper function adjusts the zone ranges | |
4446 | * provided by the architecture for a given node by using the end of the | |
4447 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
4448 | * zones within a node are in order of monotonic increases memory addresses | |
4449 | */ | |
b69a7288 | 4450 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4451 | unsigned long zone_type, |
4452 | unsigned long node_start_pfn, | |
4453 | unsigned long node_end_pfn, | |
4454 | unsigned long *zone_start_pfn, | |
4455 | unsigned long *zone_end_pfn) | |
4456 | { | |
4457 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4458 | if (zone_movable_pfn[nid]) { | |
4459 | /* Size ZONE_MOVABLE */ | |
4460 | if (zone_type == ZONE_MOVABLE) { | |
4461 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4462 | *zone_end_pfn = min(node_end_pfn, | |
4463 | arch_zone_highest_possible_pfn[movable_zone]); | |
4464 | ||
4465 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4466 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4467 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4468 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4469 | ||
4470 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4471 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4472 | *zone_start_pfn = *zone_end_pfn; | |
4473 | } | |
4474 | } | |
4475 | ||
c713216d MG |
4476 | /* |
4477 | * Return the number of pages a zone spans in a node, including holes | |
4478 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4479 | */ | |
6ea6e688 | 4480 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4481 | unsigned long zone_type, |
7960aedd ZY |
4482 | unsigned long node_start_pfn, |
4483 | unsigned long node_end_pfn, | |
c713216d MG |
4484 | unsigned long *ignored) |
4485 | { | |
c713216d MG |
4486 | unsigned long zone_start_pfn, zone_end_pfn; |
4487 | ||
7960aedd | 4488 | /* Get the start and end of the zone */ |
c713216d MG |
4489 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
4490 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4491 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4492 | node_start_pfn, node_end_pfn, | |
4493 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4494 | |
4495 | /* Check that this node has pages within the zone's required range */ | |
4496 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4497 | return 0; | |
4498 | ||
4499 | /* Move the zone boundaries inside the node if necessary */ | |
4500 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4501 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4502 | ||
4503 | /* Return the spanned pages */ | |
4504 | return zone_end_pfn - zone_start_pfn; | |
4505 | } | |
4506 | ||
4507 | /* | |
4508 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4509 | * then all holes in the requested range will be accounted for. |
c713216d | 4510 | */ |
32996250 | 4511 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4512 | unsigned long range_start_pfn, |
4513 | unsigned long range_end_pfn) | |
4514 | { | |
96e907d1 TH |
4515 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4516 | unsigned long start_pfn, end_pfn; | |
4517 | int i; | |
c713216d | 4518 | |
96e907d1 TH |
4519 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4520 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4521 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4522 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4523 | } |
96e907d1 | 4524 | return nr_absent; |
c713216d MG |
4525 | } |
4526 | ||
4527 | /** | |
4528 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4529 | * @start_pfn: The start PFN to start searching for holes | |
4530 | * @end_pfn: The end PFN to stop searching for holes | |
4531 | * | |
88ca3b94 | 4532 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4533 | */ |
4534 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4535 | unsigned long end_pfn) | |
4536 | { | |
4537 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4538 | } | |
4539 | ||
4540 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4541 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4542 | unsigned long zone_type, |
7960aedd ZY |
4543 | unsigned long node_start_pfn, |
4544 | unsigned long node_end_pfn, | |
c713216d MG |
4545 | unsigned long *ignored) |
4546 | { | |
96e907d1 TH |
4547 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4548 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4549 | unsigned long zone_start_pfn, zone_end_pfn; |
4550 | ||
96e907d1 TH |
4551 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4552 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4553 | |
2a1e274a MG |
4554 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4555 | node_start_pfn, node_end_pfn, | |
4556 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4557 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4558 | } |
0e0b864e | 4559 | |
0ee332c1 | 4560 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4561 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4562 | unsigned long zone_type, |
7960aedd ZY |
4563 | unsigned long node_start_pfn, |
4564 | unsigned long node_end_pfn, | |
c713216d MG |
4565 | unsigned long *zones_size) |
4566 | { | |
4567 | return zones_size[zone_type]; | |
4568 | } | |
4569 | ||
6ea6e688 | 4570 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4571 | unsigned long zone_type, |
7960aedd ZY |
4572 | unsigned long node_start_pfn, |
4573 | unsigned long node_end_pfn, | |
c713216d MG |
4574 | unsigned long *zholes_size) |
4575 | { | |
4576 | if (!zholes_size) | |
4577 | return 0; | |
4578 | ||
4579 | return zholes_size[zone_type]; | |
4580 | } | |
20e6926d | 4581 | |
0ee332c1 | 4582 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4583 | |
a3142c8e | 4584 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
4585 | unsigned long node_start_pfn, |
4586 | unsigned long node_end_pfn, | |
4587 | unsigned long *zones_size, | |
4588 | unsigned long *zholes_size) | |
c713216d MG |
4589 | { |
4590 | unsigned long realtotalpages, totalpages = 0; | |
4591 | enum zone_type i; | |
4592 | ||
4593 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4594 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4595 | node_start_pfn, |
4596 | node_end_pfn, | |
4597 | zones_size); | |
c713216d MG |
4598 | pgdat->node_spanned_pages = totalpages; |
4599 | ||
4600 | realtotalpages = totalpages; | |
4601 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4602 | realtotalpages -= | |
4603 | zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4604 | node_start_pfn, node_end_pfn, |
4605 | zholes_size); | |
c713216d MG |
4606 | pgdat->node_present_pages = realtotalpages; |
4607 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4608 | realtotalpages); | |
4609 | } | |
4610 | ||
835c134e MG |
4611 | #ifndef CONFIG_SPARSEMEM |
4612 | /* | |
4613 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4614 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4615 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4616 | * round what is now in bits to nearest long in bits, then return it in |
4617 | * bytes. | |
4618 | */ | |
7c45512d | 4619 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
4620 | { |
4621 | unsigned long usemapsize; | |
4622 | ||
7c45512d | 4623 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
4624 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4625 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4626 | usemapsize *= NR_PAGEBLOCK_BITS; |
4627 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4628 | ||
4629 | return usemapsize / 8; | |
4630 | } | |
4631 | ||
4632 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
4633 | struct zone *zone, |
4634 | unsigned long zone_start_pfn, | |
4635 | unsigned long zonesize) | |
835c134e | 4636 | { |
7c45512d | 4637 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 4638 | zone->pageblock_flags = NULL; |
58a01a45 | 4639 | if (usemapsize) |
8f389a99 YL |
4640 | zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat, |
4641 | usemapsize); | |
835c134e MG |
4642 | } |
4643 | #else | |
7c45512d LT |
4644 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
4645 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
4646 | #endif /* CONFIG_SPARSEMEM */ |
4647 | ||
d9c23400 | 4648 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 4649 | |
d9c23400 | 4650 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 4651 | void __paginginit set_pageblock_order(void) |
d9c23400 | 4652 | { |
955c1cd7 AM |
4653 | unsigned int order; |
4654 | ||
d9c23400 MG |
4655 | /* Check that pageblock_nr_pages has not already been setup */ |
4656 | if (pageblock_order) | |
4657 | return; | |
4658 | ||
955c1cd7 AM |
4659 | if (HPAGE_SHIFT > PAGE_SHIFT) |
4660 | order = HUGETLB_PAGE_ORDER; | |
4661 | else | |
4662 | order = MAX_ORDER - 1; | |
4663 | ||
d9c23400 MG |
4664 | /* |
4665 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
4666 | * This value may be variable depending on boot parameters on IA64 and |
4667 | * powerpc. | |
d9c23400 MG |
4668 | */ |
4669 | pageblock_order = order; | |
4670 | } | |
4671 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4672 | ||
ba72cb8c MG |
4673 | /* |
4674 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
4675 | * is unused as pageblock_order is set at compile-time. See |
4676 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
4677 | * the kernel config | |
ba72cb8c | 4678 | */ |
15ca220e | 4679 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 4680 | { |
ba72cb8c | 4681 | } |
d9c23400 MG |
4682 | |
4683 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4684 | ||
01cefaef JL |
4685 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
4686 | unsigned long present_pages) | |
4687 | { | |
4688 | unsigned long pages = spanned_pages; | |
4689 | ||
4690 | /* | |
4691 | * Provide a more accurate estimation if there are holes within | |
4692 | * the zone and SPARSEMEM is in use. If there are holes within the | |
4693 | * zone, each populated memory region may cost us one or two extra | |
4694 | * memmap pages due to alignment because memmap pages for each | |
4695 | * populated regions may not naturally algined on page boundary. | |
4696 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
4697 | */ | |
4698 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
4699 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
4700 | pages = present_pages; | |
4701 | ||
4702 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
4703 | } | |
4704 | ||
1da177e4 LT |
4705 | /* |
4706 | * Set up the zone data structures: | |
4707 | * - mark all pages reserved | |
4708 | * - mark all memory queues empty | |
4709 | * - clear the memory bitmaps | |
6527af5d MK |
4710 | * |
4711 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 4712 | */ |
b5a0e011 | 4713 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
7960aedd | 4714 | unsigned long node_start_pfn, unsigned long node_end_pfn, |
1da177e4 LT |
4715 | unsigned long *zones_size, unsigned long *zholes_size) |
4716 | { | |
2f1b6248 | 4717 | enum zone_type j; |
ed8ece2e | 4718 | int nid = pgdat->node_id; |
1da177e4 | 4719 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4720 | int ret; |
1da177e4 | 4721 | |
208d54e5 | 4722 | pgdat_resize_init(pgdat); |
8177a420 AA |
4723 | #ifdef CONFIG_NUMA_BALANCING |
4724 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
4725 | pgdat->numabalancing_migrate_nr_pages = 0; | |
4726 | pgdat->numabalancing_migrate_next_window = jiffies; | |
4727 | #endif | |
1da177e4 | 4728 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 4729 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
52d4b9ac | 4730 | pgdat_page_cgroup_init(pgdat); |
5f63b720 | 4731 | |
1da177e4 LT |
4732 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4733 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 4734 | unsigned long size, realsize, freesize, memmap_pages; |
1da177e4 | 4735 | |
7960aedd ZY |
4736 | size = zone_spanned_pages_in_node(nid, j, node_start_pfn, |
4737 | node_end_pfn, zones_size); | |
9feedc9d | 4738 | realsize = freesize = size - zone_absent_pages_in_node(nid, j, |
7960aedd ZY |
4739 | node_start_pfn, |
4740 | node_end_pfn, | |
c713216d | 4741 | zholes_size); |
1da177e4 | 4742 | |
0e0b864e | 4743 | /* |
9feedc9d | 4744 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
4745 | * is used by this zone for memmap. This affects the watermark |
4746 | * and per-cpu initialisations | |
4747 | */ | |
01cefaef | 4748 | memmap_pages = calc_memmap_size(size, realsize); |
9feedc9d JL |
4749 | if (freesize >= memmap_pages) { |
4750 | freesize -= memmap_pages; | |
5594c8c8 YL |
4751 | if (memmap_pages) |
4752 | printk(KERN_DEBUG | |
4753 | " %s zone: %lu pages used for memmap\n", | |
4754 | zone_names[j], memmap_pages); | |
0e0b864e MG |
4755 | } else |
4756 | printk(KERN_WARNING | |
9feedc9d JL |
4757 | " %s zone: %lu pages exceeds freesize %lu\n", |
4758 | zone_names[j], memmap_pages, freesize); | |
0e0b864e | 4759 | |
6267276f | 4760 | /* Account for reserved pages */ |
9feedc9d JL |
4761 | if (j == 0 && freesize > dma_reserve) { |
4762 | freesize -= dma_reserve; | |
d903ef9f | 4763 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4764 | zone_names[0], dma_reserve); |
0e0b864e MG |
4765 | } |
4766 | ||
98d2b0eb | 4767 | if (!is_highmem_idx(j)) |
9feedc9d | 4768 | nr_kernel_pages += freesize; |
01cefaef JL |
4769 | /* Charge for highmem memmap if there are enough kernel pages */ |
4770 | else if (nr_kernel_pages > memmap_pages * 2) | |
4771 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 4772 | nr_all_pages += freesize; |
1da177e4 LT |
4773 | |
4774 | zone->spanned_pages = size; | |
306f2e9e | 4775 | zone->present_pages = realsize; |
9feedc9d JL |
4776 | /* |
4777 | * Set an approximate value for lowmem here, it will be adjusted | |
4778 | * when the bootmem allocator frees pages into the buddy system. | |
4779 | * And all highmem pages will be managed by the buddy system. | |
4780 | */ | |
4781 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 4782 | #ifdef CONFIG_NUMA |
d5f541ed | 4783 | zone->node = nid; |
9feedc9d | 4784 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 4785 | / 100; |
9feedc9d | 4786 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 4787 | #endif |
1da177e4 LT |
4788 | zone->name = zone_names[j]; |
4789 | spin_lock_init(&zone->lock); | |
4790 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4791 | zone_seqlock_init(zone); |
1da177e4 | 4792 | zone->zone_pgdat = pgdat; |
ed8ece2e | 4793 | zone_pcp_init(zone); |
81c0a2bb JW |
4794 | |
4795 | /* For bootup, initialized properly in watermark setup */ | |
4796 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
4797 | ||
bea8c150 | 4798 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
4799 | if (!size) |
4800 | continue; | |
4801 | ||
955c1cd7 | 4802 | set_pageblock_order(); |
7c45512d | 4803 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
a2f3aa02 DH |
4804 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4805 | size, MEMMAP_EARLY); | |
718127cc | 4806 | BUG_ON(ret); |
76cdd58e | 4807 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 4808 | zone_start_pfn += size; |
1da177e4 LT |
4809 | } |
4810 | } | |
4811 | ||
577a32f6 | 4812 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 4813 | { |
1da177e4 LT |
4814 | /* Skip empty nodes */ |
4815 | if (!pgdat->node_spanned_pages) | |
4816 | return; | |
4817 | ||
d41dee36 | 4818 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
4819 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
4820 | if (!pgdat->node_mem_map) { | |
e984bb43 | 4821 | unsigned long size, start, end; |
d41dee36 AW |
4822 | struct page *map; |
4823 | ||
e984bb43 BP |
4824 | /* |
4825 | * The zone's endpoints aren't required to be MAX_ORDER | |
4826 | * aligned but the node_mem_map endpoints must be in order | |
4827 | * for the buddy allocator to function correctly. | |
4828 | */ | |
4829 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
108bcc96 | 4830 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
4831 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
4832 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
4833 | map = alloc_remap(pgdat->node_id, size); |
4834 | if (!map) | |
8f389a99 | 4835 | map = alloc_bootmem_node_nopanic(pgdat, size); |
e984bb43 | 4836 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 4837 | } |
12d810c1 | 4838 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
4839 | /* |
4840 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
4841 | */ | |
c713216d | 4842 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 4843 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 4844 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4845 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 4846 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 4847 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4848 | } |
1da177e4 | 4849 | #endif |
d41dee36 | 4850 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
4851 | } |
4852 | ||
9109fb7b JW |
4853 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
4854 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 4855 | { |
9109fb7b | 4856 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
4857 | unsigned long start_pfn = 0; |
4858 | unsigned long end_pfn = 0; | |
9109fb7b | 4859 | |
88fdf75d | 4860 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 4861 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 4862 | |
1da177e4 LT |
4863 | pgdat->node_id = nid; |
4864 | pgdat->node_start_pfn = node_start_pfn; | |
957f822a | 4865 | init_zone_allows_reclaim(nid); |
7960aedd ZY |
4866 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
4867 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
4868 | #endif | |
4869 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
4870 | zones_size, zholes_size); | |
1da177e4 LT |
4871 | |
4872 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
4873 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
4874 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
4875 | nid, (unsigned long)pgdat, | |
4876 | (unsigned long)pgdat->node_mem_map); | |
4877 | #endif | |
1da177e4 | 4878 | |
7960aedd ZY |
4879 | free_area_init_core(pgdat, start_pfn, end_pfn, |
4880 | zones_size, zholes_size); | |
1da177e4 LT |
4881 | } |
4882 | ||
0ee332c1 | 4883 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
4884 | |
4885 | #if MAX_NUMNODES > 1 | |
4886 | /* | |
4887 | * Figure out the number of possible node ids. | |
4888 | */ | |
f9872caf | 4889 | void __init setup_nr_node_ids(void) |
418508c1 MS |
4890 | { |
4891 | unsigned int node; | |
4892 | unsigned int highest = 0; | |
4893 | ||
4894 | for_each_node_mask(node, node_possible_map) | |
4895 | highest = node; | |
4896 | nr_node_ids = highest + 1; | |
4897 | } | |
418508c1 MS |
4898 | #endif |
4899 | ||
1e01979c TH |
4900 | /** |
4901 | * node_map_pfn_alignment - determine the maximum internode alignment | |
4902 | * | |
4903 | * This function should be called after node map is populated and sorted. | |
4904 | * It calculates the maximum power of two alignment which can distinguish | |
4905 | * all the nodes. | |
4906 | * | |
4907 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
4908 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
4909 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
4910 | * shifted, 1GiB is enough and this function will indicate so. | |
4911 | * | |
4912 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
4913 | * model has fine enough granularity to avoid incorrect mapping for the | |
4914 | * populated node map. | |
4915 | * | |
4916 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
4917 | * requirement (single node). | |
4918 | */ | |
4919 | unsigned long __init node_map_pfn_alignment(void) | |
4920 | { | |
4921 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 4922 | unsigned long start, end, mask; |
1e01979c | 4923 | int last_nid = -1; |
c13291a5 | 4924 | int i, nid; |
1e01979c | 4925 | |
c13291a5 | 4926 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
4927 | if (!start || last_nid < 0 || last_nid == nid) { |
4928 | last_nid = nid; | |
4929 | last_end = end; | |
4930 | continue; | |
4931 | } | |
4932 | ||
4933 | /* | |
4934 | * Start with a mask granular enough to pin-point to the | |
4935 | * start pfn and tick off bits one-by-one until it becomes | |
4936 | * too coarse to separate the current node from the last. | |
4937 | */ | |
4938 | mask = ~((1 << __ffs(start)) - 1); | |
4939 | while (mask && last_end <= (start & (mask << 1))) | |
4940 | mask <<= 1; | |
4941 | ||
4942 | /* accumulate all internode masks */ | |
4943 | accl_mask |= mask; | |
4944 | } | |
4945 | ||
4946 | /* convert mask to number of pages */ | |
4947 | return ~accl_mask + 1; | |
4948 | } | |
4949 | ||
a6af2bc3 | 4950 | /* Find the lowest pfn for a node */ |
b69a7288 | 4951 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 4952 | { |
a6af2bc3 | 4953 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
4954 | unsigned long start_pfn; |
4955 | int i; | |
1abbfb41 | 4956 | |
c13291a5 TH |
4957 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
4958 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 4959 | |
a6af2bc3 MG |
4960 | if (min_pfn == ULONG_MAX) { |
4961 | printk(KERN_WARNING | |
2bc0d261 | 4962 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4963 | return 0; |
4964 | } | |
4965 | ||
4966 | return min_pfn; | |
c713216d MG |
4967 | } |
4968 | ||
4969 | /** | |
4970 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4971 | * | |
4972 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4973 | * add_active_range(). |
c713216d MG |
4974 | */ |
4975 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4976 | { | |
4977 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4978 | } | |
4979 | ||
37b07e41 LS |
4980 | /* |
4981 | * early_calculate_totalpages() | |
4982 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 4983 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 4984 | */ |
484f51f8 | 4985 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 4986 | { |
7e63efef | 4987 | unsigned long totalpages = 0; |
c13291a5 TH |
4988 | unsigned long start_pfn, end_pfn; |
4989 | int i, nid; | |
4990 | ||
4991 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
4992 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 4993 | |
37b07e41 LS |
4994 | totalpages += pages; |
4995 | if (pages) | |
4b0ef1fe | 4996 | node_set_state(nid, N_MEMORY); |
37b07e41 | 4997 | } |
b8af2941 | 4998 | return totalpages; |
7e63efef MG |
4999 | } |
5000 | ||
2a1e274a MG |
5001 | /* |
5002 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
5003 | * is spread evenly between nodes as long as the nodes have enough | |
5004 | * memory. When they don't, some nodes will have more kernelcore than | |
5005 | * others | |
5006 | */ | |
b224ef85 | 5007 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
5008 | { |
5009 | int i, nid; | |
5010 | unsigned long usable_startpfn; | |
5011 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 5012 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 5013 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 5014 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 5015 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
2a1e274a | 5016 | |
7e63efef MG |
5017 | /* |
5018 | * If movablecore was specified, calculate what size of | |
5019 | * kernelcore that corresponds so that memory usable for | |
5020 | * any allocation type is evenly spread. If both kernelcore | |
5021 | * and movablecore are specified, then the value of kernelcore | |
5022 | * will be used for required_kernelcore if it's greater than | |
5023 | * what movablecore would have allowed. | |
5024 | */ | |
5025 | if (required_movablecore) { | |
7e63efef MG |
5026 | unsigned long corepages; |
5027 | ||
5028 | /* | |
5029 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
5030 | * was requested by the user | |
5031 | */ | |
5032 | required_movablecore = | |
5033 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
5034 | corepages = totalpages - required_movablecore; | |
5035 | ||
5036 | required_kernelcore = max(required_kernelcore, corepages); | |
5037 | } | |
5038 | ||
20e6926d YL |
5039 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
5040 | if (!required_kernelcore) | |
66918dcd | 5041 | goto out; |
2a1e274a MG |
5042 | |
5043 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
20e6926d | 5044 | find_usable_zone_for_movable(); |
2a1e274a MG |
5045 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
5046 | ||
5047 | restart: | |
5048 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
5049 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 5050 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
5051 | unsigned long start_pfn, end_pfn; |
5052 | ||
2a1e274a MG |
5053 | /* |
5054 | * Recalculate kernelcore_node if the division per node | |
5055 | * now exceeds what is necessary to satisfy the requested | |
5056 | * amount of memory for the kernel | |
5057 | */ | |
5058 | if (required_kernelcore < kernelcore_node) | |
5059 | kernelcore_node = required_kernelcore / usable_nodes; | |
5060 | ||
5061 | /* | |
5062 | * As the map is walked, we track how much memory is usable | |
5063 | * by the kernel using kernelcore_remaining. When it is | |
5064 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
5065 | */ | |
5066 | kernelcore_remaining = kernelcore_node; | |
5067 | ||
5068 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 5069 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
5070 | unsigned long size_pages; |
5071 | ||
c13291a5 | 5072 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
5073 | if (start_pfn >= end_pfn) |
5074 | continue; | |
5075 | ||
5076 | /* Account for what is only usable for kernelcore */ | |
5077 | if (start_pfn < usable_startpfn) { | |
5078 | unsigned long kernel_pages; | |
5079 | kernel_pages = min(end_pfn, usable_startpfn) | |
5080 | - start_pfn; | |
5081 | ||
5082 | kernelcore_remaining -= min(kernel_pages, | |
5083 | kernelcore_remaining); | |
5084 | required_kernelcore -= min(kernel_pages, | |
5085 | required_kernelcore); | |
5086 | ||
5087 | /* Continue if range is now fully accounted */ | |
5088 | if (end_pfn <= usable_startpfn) { | |
5089 | ||
5090 | /* | |
5091 | * Push zone_movable_pfn to the end so | |
5092 | * that if we have to rebalance | |
5093 | * kernelcore across nodes, we will | |
5094 | * not double account here | |
5095 | */ | |
5096 | zone_movable_pfn[nid] = end_pfn; | |
5097 | continue; | |
5098 | } | |
5099 | start_pfn = usable_startpfn; | |
5100 | } | |
5101 | ||
5102 | /* | |
5103 | * The usable PFN range for ZONE_MOVABLE is from | |
5104 | * start_pfn->end_pfn. Calculate size_pages as the | |
5105 | * number of pages used as kernelcore | |
5106 | */ | |
5107 | size_pages = end_pfn - start_pfn; | |
5108 | if (size_pages > kernelcore_remaining) | |
5109 | size_pages = kernelcore_remaining; | |
5110 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5111 | ||
5112 | /* | |
5113 | * Some kernelcore has been met, update counts and | |
5114 | * break if the kernelcore for this node has been | |
b8af2941 | 5115 | * satisfied |
2a1e274a MG |
5116 | */ |
5117 | required_kernelcore -= min(required_kernelcore, | |
5118 | size_pages); | |
5119 | kernelcore_remaining -= size_pages; | |
5120 | if (!kernelcore_remaining) | |
5121 | break; | |
5122 | } | |
5123 | } | |
5124 | ||
5125 | /* | |
5126 | * If there is still required_kernelcore, we do another pass with one | |
5127 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5128 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 5129 | * satisfied |
2a1e274a MG |
5130 | */ |
5131 | usable_nodes--; | |
5132 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5133 | goto restart; | |
5134 | ||
5135 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
5136 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5137 | zone_movable_pfn[nid] = | |
5138 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5139 | |
20e6926d | 5140 | out: |
66918dcd | 5141 | /* restore the node_state */ |
4b0ef1fe | 5142 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5143 | } |
5144 | ||
4b0ef1fe LJ |
5145 | /* Any regular or high memory on that node ? */ |
5146 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5147 | { |
37b07e41 LS |
5148 | enum zone_type zone_type; |
5149 | ||
4b0ef1fe LJ |
5150 | if (N_MEMORY == N_NORMAL_MEMORY) |
5151 | return; | |
5152 | ||
5153 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5154 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 5155 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
5156 | node_set_state(nid, N_HIGH_MEMORY); |
5157 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5158 | zone_type <= ZONE_NORMAL) | |
5159 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5160 | break; |
5161 | } | |
37b07e41 | 5162 | } |
37b07e41 LS |
5163 | } |
5164 | ||
c713216d MG |
5165 | /** |
5166 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 5167 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
5168 | * |
5169 | * This will call free_area_init_node() for each active node in the system. | |
5170 | * Using the page ranges provided by add_active_range(), the size of each | |
5171 | * zone in each node and their holes is calculated. If the maximum PFN | |
5172 | * between two adjacent zones match, it is assumed that the zone is empty. | |
5173 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
5174 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
5175 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
5176 | * at arch_max_dma_pfn. | |
5177 | */ | |
5178 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
5179 | { | |
c13291a5 TH |
5180 | unsigned long start_pfn, end_pfn; |
5181 | int i, nid; | |
a6af2bc3 | 5182 | |
c713216d MG |
5183 | /* Record where the zone boundaries are */ |
5184 | memset(arch_zone_lowest_possible_pfn, 0, | |
5185 | sizeof(arch_zone_lowest_possible_pfn)); | |
5186 | memset(arch_zone_highest_possible_pfn, 0, | |
5187 | sizeof(arch_zone_highest_possible_pfn)); | |
5188 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
5189 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
5190 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
5191 | if (i == ZONE_MOVABLE) |
5192 | continue; | |
c713216d MG |
5193 | arch_zone_lowest_possible_pfn[i] = |
5194 | arch_zone_highest_possible_pfn[i-1]; | |
5195 | arch_zone_highest_possible_pfn[i] = | |
5196 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
5197 | } | |
2a1e274a MG |
5198 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
5199 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
5200 | ||
5201 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
5202 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 5203 | find_zone_movable_pfns_for_nodes(); |
c713216d | 5204 | |
c713216d | 5205 | /* Print out the zone ranges */ |
a62e2f4f | 5206 | printk("Zone ranges:\n"); |
2a1e274a MG |
5207 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5208 | if (i == ZONE_MOVABLE) | |
5209 | continue; | |
155cbfc8 | 5210 | printk(KERN_CONT " %-8s ", zone_names[i]); |
72f0ba02 DR |
5211 | if (arch_zone_lowest_possible_pfn[i] == |
5212 | arch_zone_highest_possible_pfn[i]) | |
155cbfc8 | 5213 | printk(KERN_CONT "empty\n"); |
72f0ba02 | 5214 | else |
a62e2f4f BH |
5215 | printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n", |
5216 | arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT, | |
5217 | (arch_zone_highest_possible_pfn[i] | |
5218 | << PAGE_SHIFT) - 1); | |
2a1e274a MG |
5219 | } |
5220 | ||
5221 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
a62e2f4f | 5222 | printk("Movable zone start for each node\n"); |
2a1e274a MG |
5223 | for (i = 0; i < MAX_NUMNODES; i++) { |
5224 | if (zone_movable_pfn[i]) | |
a62e2f4f BH |
5225 | printk(" Node %d: %#010lx\n", i, |
5226 | zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 5227 | } |
c713216d | 5228 | |
f2d52fe5 | 5229 | /* Print out the early node map */ |
a62e2f4f | 5230 | printk("Early memory node ranges\n"); |
c13291a5 | 5231 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
a62e2f4f BH |
5232 | printk(" node %3d: [mem %#010lx-%#010lx]\n", nid, |
5233 | start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
5234 | |
5235 | /* Initialise every node */ | |
708614e6 | 5236 | mminit_verify_pageflags_layout(); |
8ef82866 | 5237 | setup_nr_node_ids(); |
c713216d MG |
5238 | for_each_online_node(nid) { |
5239 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 5240 | free_area_init_node(nid, NULL, |
c713216d | 5241 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
5242 | |
5243 | /* Any memory on that node */ | |
5244 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
5245 | node_set_state(nid, N_MEMORY); |
5246 | check_for_memory(pgdat, nid); | |
c713216d MG |
5247 | } |
5248 | } | |
2a1e274a | 5249 | |
7e63efef | 5250 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
5251 | { |
5252 | unsigned long long coremem; | |
5253 | if (!p) | |
5254 | return -EINVAL; | |
5255 | ||
5256 | coremem = memparse(p, &p); | |
7e63efef | 5257 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 5258 | |
7e63efef | 5259 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
5260 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
5261 | ||
5262 | return 0; | |
5263 | } | |
ed7ed365 | 5264 | |
7e63efef MG |
5265 | /* |
5266 | * kernelcore=size sets the amount of memory for use for allocations that | |
5267 | * cannot be reclaimed or migrated. | |
5268 | */ | |
5269 | static int __init cmdline_parse_kernelcore(char *p) | |
5270 | { | |
5271 | return cmdline_parse_core(p, &required_kernelcore); | |
5272 | } | |
5273 | ||
5274 | /* | |
5275 | * movablecore=size sets the amount of memory for use for allocations that | |
5276 | * can be reclaimed or migrated. | |
5277 | */ | |
5278 | static int __init cmdline_parse_movablecore(char *p) | |
5279 | { | |
5280 | return cmdline_parse_core(p, &required_movablecore); | |
5281 | } | |
5282 | ||
ed7ed365 | 5283 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 5284 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 5285 | |
0ee332c1 | 5286 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5287 | |
c3d5f5f0 JL |
5288 | void adjust_managed_page_count(struct page *page, long count) |
5289 | { | |
5290 | spin_lock(&managed_page_count_lock); | |
5291 | page_zone(page)->managed_pages += count; | |
5292 | totalram_pages += count; | |
3dcc0571 JL |
5293 | #ifdef CONFIG_HIGHMEM |
5294 | if (PageHighMem(page)) | |
5295 | totalhigh_pages += count; | |
5296 | #endif | |
c3d5f5f0 JL |
5297 | spin_unlock(&managed_page_count_lock); |
5298 | } | |
3dcc0571 | 5299 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 5300 | |
11199692 | 5301 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 5302 | { |
11199692 JL |
5303 | void *pos; |
5304 | unsigned long pages = 0; | |
69afade7 | 5305 | |
11199692 JL |
5306 | start = (void *)PAGE_ALIGN((unsigned long)start); |
5307 | end = (void *)((unsigned long)end & PAGE_MASK); | |
5308 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 5309 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
5310 | memset(pos, poison, PAGE_SIZE); |
5311 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
5312 | } |
5313 | ||
5314 | if (pages && s) | |
11199692 | 5315 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
5316 | s, pages << (PAGE_SHIFT - 10), start, end); |
5317 | ||
5318 | return pages; | |
5319 | } | |
11199692 | 5320 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 5321 | |
cfa11e08 JL |
5322 | #ifdef CONFIG_HIGHMEM |
5323 | void free_highmem_page(struct page *page) | |
5324 | { | |
5325 | __free_reserved_page(page); | |
5326 | totalram_pages++; | |
7b4b2a0d | 5327 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
5328 | totalhigh_pages++; |
5329 | } | |
5330 | #endif | |
5331 | ||
7ee3d4e8 JL |
5332 | |
5333 | void __init mem_init_print_info(const char *str) | |
5334 | { | |
5335 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
5336 | unsigned long init_code_size, init_data_size; | |
5337 | ||
5338 | physpages = get_num_physpages(); | |
5339 | codesize = _etext - _stext; | |
5340 | datasize = _edata - _sdata; | |
5341 | rosize = __end_rodata - __start_rodata; | |
5342 | bss_size = __bss_stop - __bss_start; | |
5343 | init_data_size = __init_end - __init_begin; | |
5344 | init_code_size = _einittext - _sinittext; | |
5345 | ||
5346 | /* | |
5347 | * Detect special cases and adjust section sizes accordingly: | |
5348 | * 1) .init.* may be embedded into .data sections | |
5349 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
5350 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
5351 | * 3) .rodata.* may be embedded into .text or .data sections. | |
5352 | */ | |
5353 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
5354 | do { \ |
5355 | if (start <= pos && pos < end && size > adj) \ | |
5356 | size -= adj; \ | |
5357 | } while (0) | |
7ee3d4e8 JL |
5358 | |
5359 | adj_init_size(__init_begin, __init_end, init_data_size, | |
5360 | _sinittext, init_code_size); | |
5361 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
5362 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
5363 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
5364 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
5365 | ||
5366 | #undef adj_init_size | |
5367 | ||
5368 | printk("Memory: %luK/%luK available " | |
5369 | "(%luK kernel code, %luK rwdata, %luK rodata, " | |
5370 | "%luK init, %luK bss, %luK reserved" | |
5371 | #ifdef CONFIG_HIGHMEM | |
5372 | ", %luK highmem" | |
5373 | #endif | |
5374 | "%s%s)\n", | |
5375 | nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10), | |
5376 | codesize >> 10, datasize >> 10, rosize >> 10, | |
5377 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
5378 | (physpages - totalram_pages) << (PAGE_SHIFT-10), | |
5379 | #ifdef CONFIG_HIGHMEM | |
5380 | totalhigh_pages << (PAGE_SHIFT-10), | |
5381 | #endif | |
5382 | str ? ", " : "", str ? str : ""); | |
5383 | } | |
5384 | ||
0e0b864e | 5385 | /** |
88ca3b94 RD |
5386 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
5387 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
5388 | * |
5389 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
5390 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
5391 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
5392 | * function may optionally be used to account for unfreeable pages in the |
5393 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
5394 | * smaller per-cpu batchsize. | |
0e0b864e MG |
5395 | */ |
5396 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
5397 | { | |
5398 | dma_reserve = new_dma_reserve; | |
5399 | } | |
5400 | ||
1da177e4 LT |
5401 | void __init free_area_init(unsigned long *zones_size) |
5402 | { | |
9109fb7b | 5403 | free_area_init_node(0, zones_size, |
1da177e4 LT |
5404 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
5405 | } | |
1da177e4 | 5406 | |
1da177e4 LT |
5407 | static int page_alloc_cpu_notify(struct notifier_block *self, |
5408 | unsigned long action, void *hcpu) | |
5409 | { | |
5410 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 5411 | |
8bb78442 | 5412 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 5413 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
5414 | drain_pages(cpu); |
5415 | ||
5416 | /* | |
5417 | * Spill the event counters of the dead processor | |
5418 | * into the current processors event counters. | |
5419 | * This artificially elevates the count of the current | |
5420 | * processor. | |
5421 | */ | |
f8891e5e | 5422 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
5423 | |
5424 | /* | |
5425 | * Zero the differential counters of the dead processor | |
5426 | * so that the vm statistics are consistent. | |
5427 | * | |
5428 | * This is only okay since the processor is dead and cannot | |
5429 | * race with what we are doing. | |
5430 | */ | |
2bb921e5 | 5431 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
5432 | } |
5433 | return NOTIFY_OK; | |
5434 | } | |
1da177e4 LT |
5435 | |
5436 | void __init page_alloc_init(void) | |
5437 | { | |
5438 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
5439 | } | |
5440 | ||
cb45b0e9 HA |
5441 | /* |
5442 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
5443 | * or min_free_kbytes changes. | |
5444 | */ | |
5445 | static void calculate_totalreserve_pages(void) | |
5446 | { | |
5447 | struct pglist_data *pgdat; | |
5448 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5449 | enum zone_type i, j; |
cb45b0e9 HA |
5450 | |
5451 | for_each_online_pgdat(pgdat) { | |
5452 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
5453 | struct zone *zone = pgdat->node_zones + i; | |
5454 | unsigned long max = 0; | |
5455 | ||
5456 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5457 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5458 | if (zone->lowmem_reserve[j] > max) | |
5459 | max = zone->lowmem_reserve[j]; | |
5460 | } | |
5461 | ||
41858966 MG |
5462 | /* we treat the high watermark as reserved pages. */ |
5463 | max += high_wmark_pages(zone); | |
cb45b0e9 | 5464 | |
b40da049 JL |
5465 | if (max > zone->managed_pages) |
5466 | max = zone->managed_pages; | |
cb45b0e9 | 5467 | reserve_pages += max; |
ab8fabd4 JW |
5468 | /* |
5469 | * Lowmem reserves are not available to | |
5470 | * GFP_HIGHUSER page cache allocations and | |
5471 | * kswapd tries to balance zones to their high | |
5472 | * watermark. As a result, neither should be | |
5473 | * regarded as dirtyable memory, to prevent a | |
5474 | * situation where reclaim has to clean pages | |
5475 | * in order to balance the zones. | |
5476 | */ | |
5477 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
5478 | } |
5479 | } | |
ab8fabd4 | 5480 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
5481 | totalreserve_pages = reserve_pages; |
5482 | } | |
5483 | ||
1da177e4 LT |
5484 | /* |
5485 | * setup_per_zone_lowmem_reserve - called whenever | |
5486 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
5487 | * has a correct pages reserved value, so an adequate number of | |
5488 | * pages are left in the zone after a successful __alloc_pages(). | |
5489 | */ | |
5490 | static void setup_per_zone_lowmem_reserve(void) | |
5491 | { | |
5492 | struct pglist_data *pgdat; | |
2f6726e5 | 5493 | enum zone_type j, idx; |
1da177e4 | 5494 | |
ec936fc5 | 5495 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
5496 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5497 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 5498 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
5499 | |
5500 | zone->lowmem_reserve[j] = 0; | |
5501 | ||
2f6726e5 CL |
5502 | idx = j; |
5503 | while (idx) { | |
1da177e4 LT |
5504 | struct zone *lower_zone; |
5505 | ||
2f6726e5 CL |
5506 | idx--; |
5507 | ||
1da177e4 LT |
5508 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
5509 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
5510 | ||
5511 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 5512 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 5513 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 5514 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
5515 | } |
5516 | } | |
5517 | } | |
cb45b0e9 HA |
5518 | |
5519 | /* update totalreserve_pages */ | |
5520 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5521 | } |
5522 | ||
cfd3da1e | 5523 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5524 | { |
5525 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5526 | unsigned long lowmem_pages = 0; | |
5527 | struct zone *zone; | |
5528 | unsigned long flags; | |
5529 | ||
5530 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5531 | for_each_zone(zone) { | |
5532 | if (!is_highmem(zone)) | |
b40da049 | 5533 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
5534 | } |
5535 | ||
5536 | for_each_zone(zone) { | |
ac924c60 AM |
5537 | u64 tmp; |
5538 | ||
1125b4e3 | 5539 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 5540 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 5541 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
5542 | if (is_highmem(zone)) { |
5543 | /* | |
669ed175 NP |
5544 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5545 | * need highmem pages, so cap pages_min to a small | |
5546 | * value here. | |
5547 | * | |
41858966 | 5548 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
5549 | * deltas controls asynch page reclaim, and so should |
5550 | * not be capped for highmem. | |
1da177e4 | 5551 | */ |
90ae8d67 | 5552 | unsigned long min_pages; |
1da177e4 | 5553 | |
b40da049 | 5554 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 5555 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 5556 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5557 | } else { |
669ed175 NP |
5558 | /* |
5559 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5560 | * proportionate to the zone's size. |
5561 | */ | |
41858966 | 5562 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5563 | } |
5564 | ||
41858966 MG |
5565 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
5566 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
49f223a9 | 5567 | |
81c0a2bb JW |
5568 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
5569 | high_wmark_pages(zone) - | |
5570 | low_wmark_pages(zone) - | |
5571 | zone_page_state(zone, NR_ALLOC_BATCH)); | |
5572 | ||
56fd56b8 | 5573 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 5574 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5575 | } |
cb45b0e9 HA |
5576 | |
5577 | /* update totalreserve_pages */ | |
5578 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5579 | } |
5580 | ||
cfd3da1e MG |
5581 | /** |
5582 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5583 | * or when memory is hot-{added|removed} | |
5584 | * | |
5585 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5586 | * correctly with respect to min_free_kbytes. | |
5587 | */ | |
5588 | void setup_per_zone_wmarks(void) | |
5589 | { | |
5590 | mutex_lock(&zonelists_mutex); | |
5591 | __setup_per_zone_wmarks(); | |
5592 | mutex_unlock(&zonelists_mutex); | |
5593 | } | |
5594 | ||
55a4462a | 5595 | /* |
556adecb RR |
5596 | * The inactive anon list should be small enough that the VM never has to |
5597 | * do too much work, but large enough that each inactive page has a chance | |
5598 | * to be referenced again before it is swapped out. | |
5599 | * | |
5600 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
5601 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
5602 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
5603 | * the anonymous pages are kept on the inactive list. | |
5604 | * | |
5605 | * total target max | |
5606 | * memory ratio inactive anon | |
5607 | * ------------------------------------- | |
5608 | * 10MB 1 5MB | |
5609 | * 100MB 1 50MB | |
5610 | * 1GB 3 250MB | |
5611 | * 10GB 10 0.9GB | |
5612 | * 100GB 31 3GB | |
5613 | * 1TB 101 10GB | |
5614 | * 10TB 320 32GB | |
5615 | */ | |
1b79acc9 | 5616 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5617 | { |
96cb4df5 | 5618 | unsigned int gb, ratio; |
556adecb | 5619 | |
96cb4df5 | 5620 | /* Zone size in gigabytes */ |
b40da049 | 5621 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 5622 | if (gb) |
556adecb | 5623 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5624 | else |
5625 | ratio = 1; | |
556adecb | 5626 | |
96cb4df5 MK |
5627 | zone->inactive_ratio = ratio; |
5628 | } | |
556adecb | 5629 | |
839a4fcc | 5630 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
5631 | { |
5632 | struct zone *zone; | |
5633 | ||
5634 | for_each_zone(zone) | |
5635 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5636 | } |
5637 | ||
1da177e4 LT |
5638 | /* |
5639 | * Initialise min_free_kbytes. | |
5640 | * | |
5641 | * For small machines we want it small (128k min). For large machines | |
5642 | * we want it large (64MB max). But it is not linear, because network | |
5643 | * bandwidth does not increase linearly with machine size. We use | |
5644 | * | |
b8af2941 | 5645 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
5646 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
5647 | * | |
5648 | * which yields | |
5649 | * | |
5650 | * 16MB: 512k | |
5651 | * 32MB: 724k | |
5652 | * 64MB: 1024k | |
5653 | * 128MB: 1448k | |
5654 | * 256MB: 2048k | |
5655 | * 512MB: 2896k | |
5656 | * 1024MB: 4096k | |
5657 | * 2048MB: 5792k | |
5658 | * 4096MB: 8192k | |
5659 | * 8192MB: 11584k | |
5660 | * 16384MB: 16384k | |
5661 | */ | |
1b79acc9 | 5662 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
5663 | { |
5664 | unsigned long lowmem_kbytes; | |
5f12733e | 5665 | int new_min_free_kbytes; |
1da177e4 LT |
5666 | |
5667 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
5668 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
5669 | ||
5670 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
5671 | min_free_kbytes = new_min_free_kbytes; | |
5672 | if (min_free_kbytes < 128) | |
5673 | min_free_kbytes = 128; | |
5674 | if (min_free_kbytes > 65536) | |
5675 | min_free_kbytes = 65536; | |
5676 | } else { | |
5677 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
5678 | new_min_free_kbytes, user_min_free_kbytes); | |
5679 | } | |
bc75d33f | 5680 | setup_per_zone_wmarks(); |
a6cccdc3 | 5681 | refresh_zone_stat_thresholds(); |
1da177e4 | 5682 | setup_per_zone_lowmem_reserve(); |
556adecb | 5683 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5684 | return 0; |
5685 | } | |
bc75d33f | 5686 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5687 | |
5688 | /* | |
b8af2941 | 5689 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
5690 | * that we can call two helper functions whenever min_free_kbytes |
5691 | * changes. | |
5692 | */ | |
b8af2941 | 5693 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, |
8d65af78 | 5694 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5695 | { |
8d65af78 | 5696 | proc_dointvec(table, write, buffer, length, ppos); |
5f12733e MH |
5697 | if (write) { |
5698 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 5699 | setup_per_zone_wmarks(); |
5f12733e | 5700 | } |
1da177e4 LT |
5701 | return 0; |
5702 | } | |
5703 | ||
9614634f CL |
5704 | #ifdef CONFIG_NUMA |
5705 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5706 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5707 | { |
5708 | struct zone *zone; | |
5709 | int rc; | |
5710 | ||
8d65af78 | 5711 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5712 | if (rc) |
5713 | return rc; | |
5714 | ||
5715 | for_each_zone(zone) | |
b40da049 | 5716 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
5717 | sysctl_min_unmapped_ratio) / 100; |
5718 | return 0; | |
5719 | } | |
0ff38490 CL |
5720 | |
5721 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5722 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5723 | { |
5724 | struct zone *zone; | |
5725 | int rc; | |
5726 | ||
8d65af78 | 5727 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5728 | if (rc) |
5729 | return rc; | |
5730 | ||
5731 | for_each_zone(zone) | |
b40da049 | 5732 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
5733 | sysctl_min_slab_ratio) / 100; |
5734 | return 0; | |
5735 | } | |
9614634f CL |
5736 | #endif |
5737 | ||
1da177e4 LT |
5738 | /* |
5739 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5740 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5741 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5742 | * | |
5743 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5744 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5745 | * if in function of the boot time zone sizes. |
5746 | */ | |
5747 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5748 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5749 | { |
8d65af78 | 5750 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5751 | setup_per_zone_lowmem_reserve(); |
5752 | return 0; | |
5753 | } | |
5754 | ||
8ad4b1fb RS |
5755 | /* |
5756 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
5757 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
5758 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 5759 | */ |
8ad4b1fb | 5760 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, |
8d65af78 | 5761 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5762 | { |
5763 | struct zone *zone; | |
5764 | unsigned int cpu; | |
5765 | int ret; | |
5766 | ||
8d65af78 | 5767 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
93278814 | 5768 | if (!write || (ret < 0)) |
8ad4b1fb | 5769 | return ret; |
c8e251fa CS |
5770 | |
5771 | mutex_lock(&pcp_batch_high_lock); | |
364df0eb | 5772 | for_each_populated_zone(zone) { |
22a7f12b CS |
5773 | unsigned long high; |
5774 | high = zone->managed_pages / percpu_pagelist_fraction; | |
5775 | for_each_possible_cpu(cpu) | |
3664033c CS |
5776 | pageset_set_high(per_cpu_ptr(zone->pageset, cpu), |
5777 | high); | |
8ad4b1fb | 5778 | } |
c8e251fa | 5779 | mutex_unlock(&pcp_batch_high_lock); |
8ad4b1fb RS |
5780 | return 0; |
5781 | } | |
5782 | ||
f034b5d4 | 5783 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
5784 | |
5785 | #ifdef CONFIG_NUMA | |
5786 | static int __init set_hashdist(char *str) | |
5787 | { | |
5788 | if (!str) | |
5789 | return 0; | |
5790 | hashdist = simple_strtoul(str, &str, 0); | |
5791 | return 1; | |
5792 | } | |
5793 | __setup("hashdist=", set_hashdist); | |
5794 | #endif | |
5795 | ||
5796 | /* | |
5797 | * allocate a large system hash table from bootmem | |
5798 | * - it is assumed that the hash table must contain an exact power-of-2 | |
5799 | * quantity of entries | |
5800 | * - limit is the number of hash buckets, not the total allocation size | |
5801 | */ | |
5802 | void *__init alloc_large_system_hash(const char *tablename, | |
5803 | unsigned long bucketsize, | |
5804 | unsigned long numentries, | |
5805 | int scale, | |
5806 | int flags, | |
5807 | unsigned int *_hash_shift, | |
5808 | unsigned int *_hash_mask, | |
31fe62b9 TB |
5809 | unsigned long low_limit, |
5810 | unsigned long high_limit) | |
1da177e4 | 5811 | { |
31fe62b9 | 5812 | unsigned long long max = high_limit; |
1da177e4 LT |
5813 | unsigned long log2qty, size; |
5814 | void *table = NULL; | |
5815 | ||
5816 | /* allow the kernel cmdline to have a say */ | |
5817 | if (!numentries) { | |
5818 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 5819 | numentries = nr_kernel_pages; |
a7e83318 JZ |
5820 | |
5821 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
5822 | if (PAGE_SHIFT < 20) | |
5823 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
5824 | |
5825 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
5826 | if (scale > PAGE_SHIFT) | |
5827 | numentries >>= (scale - PAGE_SHIFT); | |
5828 | else | |
5829 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
5830 | |
5831 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
5832 | if (unlikely(flags & HASH_SMALL)) { |
5833 | /* Makes no sense without HASH_EARLY */ | |
5834 | WARN_ON(!(flags & HASH_EARLY)); | |
5835 | if (!(numentries >> *_hash_shift)) { | |
5836 | numentries = 1UL << *_hash_shift; | |
5837 | BUG_ON(!numentries); | |
5838 | } | |
5839 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 5840 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 5841 | } |
6e692ed3 | 5842 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
5843 | |
5844 | /* limit allocation size to 1/16 total memory by default */ | |
5845 | if (max == 0) { | |
5846 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
5847 | do_div(max, bucketsize); | |
5848 | } | |
074b8517 | 5849 | max = min(max, 0x80000000ULL); |
1da177e4 | 5850 | |
31fe62b9 TB |
5851 | if (numentries < low_limit) |
5852 | numentries = low_limit; | |
1da177e4 LT |
5853 | if (numentries > max) |
5854 | numentries = max; | |
5855 | ||
f0d1b0b3 | 5856 | log2qty = ilog2(numentries); |
1da177e4 LT |
5857 | |
5858 | do { | |
5859 | size = bucketsize << log2qty; | |
5860 | if (flags & HASH_EARLY) | |
74768ed8 | 5861 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
5862 | else if (hashdist) |
5863 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
5864 | else { | |
1037b83b ED |
5865 | /* |
5866 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
5867 | * some pages at the end of hash table which |
5868 | * alloc_pages_exact() automatically does | |
1037b83b | 5869 | */ |
264ef8a9 | 5870 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 5871 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
5872 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
5873 | } | |
1da177e4 LT |
5874 | } |
5875 | } while (!table && size > PAGE_SIZE && --log2qty); | |
5876 | ||
5877 | if (!table) | |
5878 | panic("Failed to allocate %s hash table\n", tablename); | |
5879 | ||
f241e660 | 5880 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 5881 | tablename, |
f241e660 | 5882 | (1UL << log2qty), |
f0d1b0b3 | 5883 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
5884 | size); |
5885 | ||
5886 | if (_hash_shift) | |
5887 | *_hash_shift = log2qty; | |
5888 | if (_hash_mask) | |
5889 | *_hash_mask = (1 << log2qty) - 1; | |
5890 | ||
5891 | return table; | |
5892 | } | |
a117e66e | 5893 | |
835c134e MG |
5894 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
5895 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
5896 | unsigned long pfn) | |
5897 | { | |
5898 | #ifdef CONFIG_SPARSEMEM | |
5899 | return __pfn_to_section(pfn)->pageblock_flags; | |
5900 | #else | |
5901 | return zone->pageblock_flags; | |
5902 | #endif /* CONFIG_SPARSEMEM */ | |
5903 | } | |
5904 | ||
5905 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
5906 | { | |
5907 | #ifdef CONFIG_SPARSEMEM | |
5908 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 5909 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 5910 | #else |
c060f943 | 5911 | pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 5912 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5913 | #endif /* CONFIG_SPARSEMEM */ |
5914 | } | |
5915 | ||
5916 | /** | |
d9c23400 | 5917 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
5918 | * @page: The page within the block of interest |
5919 | * @start_bitidx: The first bit of interest to retrieve | |
5920 | * @end_bitidx: The last bit of interest | |
5921 | * returns pageblock_bits flags | |
5922 | */ | |
5923 | unsigned long get_pageblock_flags_group(struct page *page, | |
5924 | int start_bitidx, int end_bitidx) | |
5925 | { | |
5926 | struct zone *zone; | |
5927 | unsigned long *bitmap; | |
5928 | unsigned long pfn, bitidx; | |
5929 | unsigned long flags = 0; | |
5930 | unsigned long value = 1; | |
5931 | ||
5932 | zone = page_zone(page); | |
5933 | pfn = page_to_pfn(page); | |
5934 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5935 | bitidx = pfn_to_bitidx(zone, pfn); | |
5936 | ||
5937 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5938 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
5939 | flags |= value; | |
6220ec78 | 5940 | |
835c134e MG |
5941 | return flags; |
5942 | } | |
5943 | ||
5944 | /** | |
d9c23400 | 5945 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
5946 | * @page: The page within the block of interest |
5947 | * @start_bitidx: The first bit of interest | |
5948 | * @end_bitidx: The last bit of interest | |
5949 | * @flags: The flags to set | |
5950 | */ | |
5951 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
5952 | int start_bitidx, int end_bitidx) | |
5953 | { | |
5954 | struct zone *zone; | |
5955 | unsigned long *bitmap; | |
5956 | unsigned long pfn, bitidx; | |
5957 | unsigned long value = 1; | |
5958 | ||
5959 | zone = page_zone(page); | |
5960 | pfn = page_to_pfn(page); | |
5961 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5962 | bitidx = pfn_to_bitidx(zone, pfn); | |
108bcc96 | 5963 | VM_BUG_ON(!zone_spans_pfn(zone, pfn)); |
835c134e MG |
5964 | |
5965 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5966 | if (flags & value) | |
5967 | __set_bit(bitidx + start_bitidx, bitmap); | |
5968 | else | |
5969 | __clear_bit(bitidx + start_bitidx, bitmap); | |
5970 | } | |
a5d76b54 KH |
5971 | |
5972 | /* | |
80934513 MK |
5973 | * This function checks whether pageblock includes unmovable pages or not. |
5974 | * If @count is not zero, it is okay to include less @count unmovable pages | |
5975 | * | |
b8af2941 | 5976 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
5977 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
5978 | * expect this function should be exact. | |
a5d76b54 | 5979 | */ |
b023f468 WC |
5980 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
5981 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
5982 | { |
5983 | unsigned long pfn, iter, found; | |
47118af0 MN |
5984 | int mt; |
5985 | ||
49ac8255 KH |
5986 | /* |
5987 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 5988 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
5989 | */ |
5990 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 5991 | return false; |
47118af0 MN |
5992 | mt = get_pageblock_migratetype(page); |
5993 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 5994 | return false; |
49ac8255 KH |
5995 | |
5996 | pfn = page_to_pfn(page); | |
5997 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
5998 | unsigned long check = pfn + iter; | |
5999 | ||
29723fcc | 6000 | if (!pfn_valid_within(check)) |
49ac8255 | 6001 | continue; |
29723fcc | 6002 | |
49ac8255 | 6003 | page = pfn_to_page(check); |
c8721bbb NH |
6004 | |
6005 | /* | |
6006 | * Hugepages are not in LRU lists, but they're movable. | |
6007 | * We need not scan over tail pages bacause we don't | |
6008 | * handle each tail page individually in migration. | |
6009 | */ | |
6010 | if (PageHuge(page)) { | |
6011 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
6012 | continue; | |
6013 | } | |
6014 | ||
97d255c8 MK |
6015 | /* |
6016 | * We can't use page_count without pin a page | |
6017 | * because another CPU can free compound page. | |
6018 | * This check already skips compound tails of THP | |
6019 | * because their page->_count is zero at all time. | |
6020 | */ | |
6021 | if (!atomic_read(&page->_count)) { | |
49ac8255 KH |
6022 | if (PageBuddy(page)) |
6023 | iter += (1 << page_order(page)) - 1; | |
6024 | continue; | |
6025 | } | |
97d255c8 | 6026 | |
b023f468 WC |
6027 | /* |
6028 | * The HWPoisoned page may be not in buddy system, and | |
6029 | * page_count() is not 0. | |
6030 | */ | |
6031 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6032 | continue; | |
6033 | ||
49ac8255 KH |
6034 | if (!PageLRU(page)) |
6035 | found++; | |
6036 | /* | |
6037 | * If there are RECLAIMABLE pages, we need to check it. | |
6038 | * But now, memory offline itself doesn't call shrink_slab() | |
6039 | * and it still to be fixed. | |
6040 | */ | |
6041 | /* | |
6042 | * If the page is not RAM, page_count()should be 0. | |
6043 | * we don't need more check. This is an _used_ not-movable page. | |
6044 | * | |
6045 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6046 | * is set to both of a memory hole page and a _used_ kernel | |
6047 | * page at boot. | |
6048 | */ | |
6049 | if (found > count) | |
80934513 | 6050 | return true; |
49ac8255 | 6051 | } |
80934513 | 6052 | return false; |
49ac8255 KH |
6053 | } |
6054 | ||
6055 | bool is_pageblock_removable_nolock(struct page *page) | |
6056 | { | |
656a0706 MH |
6057 | struct zone *zone; |
6058 | unsigned long pfn; | |
687875fb MH |
6059 | |
6060 | /* | |
6061 | * We have to be careful here because we are iterating over memory | |
6062 | * sections which are not zone aware so we might end up outside of | |
6063 | * the zone but still within the section. | |
656a0706 MH |
6064 | * We have to take care about the node as well. If the node is offline |
6065 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 6066 | */ |
656a0706 MH |
6067 | if (!node_online(page_to_nid(page))) |
6068 | return false; | |
6069 | ||
6070 | zone = page_zone(page); | |
6071 | pfn = page_to_pfn(page); | |
108bcc96 | 6072 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
6073 | return false; |
6074 | ||
b023f468 | 6075 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6076 | } |
0c0e6195 | 6077 | |
041d3a8c MN |
6078 | #ifdef CONFIG_CMA |
6079 | ||
6080 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6081 | { | |
6082 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6083 | pageblock_nr_pages) - 1); | |
6084 | } | |
6085 | ||
6086 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6087 | { | |
6088 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6089 | pageblock_nr_pages)); | |
6090 | } | |
6091 | ||
041d3a8c | 6092 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6093 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6094 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6095 | { |
6096 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6097 | unsigned long nr_reclaimed; |
041d3a8c MN |
6098 | unsigned long pfn = start; |
6099 | unsigned int tries = 0; | |
6100 | int ret = 0; | |
6101 | ||
be49a6e1 | 6102 | migrate_prep(); |
041d3a8c | 6103 | |
bb13ffeb | 6104 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6105 | if (fatal_signal_pending(current)) { |
6106 | ret = -EINTR; | |
6107 | break; | |
6108 | } | |
6109 | ||
bb13ffeb MG |
6110 | if (list_empty(&cc->migratepages)) { |
6111 | cc->nr_migratepages = 0; | |
6112 | pfn = isolate_migratepages_range(cc->zone, cc, | |
e46a2879 | 6113 | pfn, end, true); |
041d3a8c MN |
6114 | if (!pfn) { |
6115 | ret = -EINTR; | |
6116 | break; | |
6117 | } | |
6118 | tries = 0; | |
6119 | } else if (++tries == 5) { | |
6120 | ret = ret < 0 ? ret : -EBUSY; | |
6121 | break; | |
6122 | } | |
6123 | ||
beb51eaa MK |
6124 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6125 | &cc->migratepages); | |
6126 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6127 | |
9c620e2b HD |
6128 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
6129 | 0, MIGRATE_SYNC, MR_CMA); | |
041d3a8c | 6130 | } |
2a6f5124 SP |
6131 | if (ret < 0) { |
6132 | putback_movable_pages(&cc->migratepages); | |
6133 | return ret; | |
6134 | } | |
6135 | return 0; | |
041d3a8c MN |
6136 | } |
6137 | ||
6138 | /** | |
6139 | * alloc_contig_range() -- tries to allocate given range of pages | |
6140 | * @start: start PFN to allocate | |
6141 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
6142 | * @migratetype: migratetype of the underlaying pageblocks (either |
6143 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
6144 | * in range must have the same migratetype and it must | |
6145 | * be either of the two. | |
041d3a8c MN |
6146 | * |
6147 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
6148 | * aligned, however it's the caller's responsibility to guarantee that | |
6149 | * we are the only thread that changes migrate type of pageblocks the | |
6150 | * pages fall in. | |
6151 | * | |
6152 | * The PFN range must belong to a single zone. | |
6153 | * | |
6154 | * Returns zero on success or negative error code. On success all | |
6155 | * pages which PFN is in [start, end) are allocated for the caller and | |
6156 | * need to be freed with free_contig_range(). | |
6157 | */ | |
0815f3d8 MN |
6158 | int alloc_contig_range(unsigned long start, unsigned long end, |
6159 | unsigned migratetype) | |
041d3a8c | 6160 | { |
041d3a8c MN |
6161 | unsigned long outer_start, outer_end; |
6162 | int ret = 0, order; | |
6163 | ||
bb13ffeb MG |
6164 | struct compact_control cc = { |
6165 | .nr_migratepages = 0, | |
6166 | .order = -1, | |
6167 | .zone = page_zone(pfn_to_page(start)), | |
6168 | .sync = true, | |
6169 | .ignore_skip_hint = true, | |
6170 | }; | |
6171 | INIT_LIST_HEAD(&cc.migratepages); | |
6172 | ||
041d3a8c MN |
6173 | /* |
6174 | * What we do here is we mark all pageblocks in range as | |
6175 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6176 | * have different sizes, and due to the way page allocator | |
6177 | * work, we align the range to biggest of the two pages so | |
6178 | * that page allocator won't try to merge buddies from | |
6179 | * different pageblocks and change MIGRATE_ISOLATE to some | |
6180 | * other migration type. | |
6181 | * | |
6182 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6183 | * migrate the pages from an unaligned range (ie. pages that | |
6184 | * we are interested in). This will put all the pages in | |
6185 | * range back to page allocator as MIGRATE_ISOLATE. | |
6186 | * | |
6187 | * When this is done, we take the pages in range from page | |
6188 | * allocator removing them from the buddy system. This way | |
6189 | * page allocator will never consider using them. | |
6190 | * | |
6191 | * This lets us mark the pageblocks back as | |
6192 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6193 | * aligned range but not in the unaligned, original range are | |
6194 | * put back to page allocator so that buddy can use them. | |
6195 | */ | |
6196 | ||
6197 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
6198 | pfn_max_align_up(end), migratetype, |
6199 | false); | |
041d3a8c | 6200 | if (ret) |
86a595f9 | 6201 | return ret; |
041d3a8c | 6202 | |
bb13ffeb | 6203 | ret = __alloc_contig_migrate_range(&cc, start, end); |
041d3a8c MN |
6204 | if (ret) |
6205 | goto done; | |
6206 | ||
6207 | /* | |
6208 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
6209 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
6210 | * more, all pages in [start, end) are free in page allocator. | |
6211 | * What we are going to do is to allocate all pages from | |
6212 | * [start, end) (that is remove them from page allocator). | |
6213 | * | |
6214 | * The only problem is that pages at the beginning and at the | |
6215 | * end of interesting range may be not aligned with pages that | |
6216 | * page allocator holds, ie. they can be part of higher order | |
6217 | * pages. Because of this, we reserve the bigger range and | |
6218 | * once this is done free the pages we are not interested in. | |
6219 | * | |
6220 | * We don't have to hold zone->lock here because the pages are | |
6221 | * isolated thus they won't get removed from buddy. | |
6222 | */ | |
6223 | ||
6224 | lru_add_drain_all(); | |
6225 | drain_all_pages(); | |
6226 | ||
6227 | order = 0; | |
6228 | outer_start = start; | |
6229 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
6230 | if (++order >= MAX_ORDER) { | |
6231 | ret = -EBUSY; | |
6232 | goto done; | |
6233 | } | |
6234 | outer_start &= ~0UL << order; | |
6235 | } | |
6236 | ||
6237 | /* Make sure the range is really isolated. */ | |
b023f468 | 6238 | if (test_pages_isolated(outer_start, end, false)) { |
041d3a8c MN |
6239 | pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n", |
6240 | outer_start, end); | |
6241 | ret = -EBUSY; | |
6242 | goto done; | |
6243 | } | |
6244 | ||
49f223a9 MS |
6245 | |
6246 | /* Grab isolated pages from freelists. */ | |
bb13ffeb | 6247 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6248 | if (!outer_end) { |
6249 | ret = -EBUSY; | |
6250 | goto done; | |
6251 | } | |
6252 | ||
6253 | /* Free head and tail (if any) */ | |
6254 | if (start != outer_start) | |
6255 | free_contig_range(outer_start, start - outer_start); | |
6256 | if (end != outer_end) | |
6257 | free_contig_range(end, outer_end - end); | |
6258 | ||
6259 | done: | |
6260 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 6261 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
6262 | return ret; |
6263 | } | |
6264 | ||
6265 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
6266 | { | |
bcc2b02f MS |
6267 | unsigned int count = 0; |
6268 | ||
6269 | for (; nr_pages--; pfn++) { | |
6270 | struct page *page = pfn_to_page(pfn); | |
6271 | ||
6272 | count += page_count(page) != 1; | |
6273 | __free_page(page); | |
6274 | } | |
6275 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
6276 | } |
6277 | #endif | |
6278 | ||
4ed7e022 | 6279 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
6280 | /* |
6281 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
6282 | * page high values need to be recalulated. | |
6283 | */ | |
4ed7e022 JL |
6284 | void __meminit zone_pcp_update(struct zone *zone) |
6285 | { | |
0a647f38 | 6286 | unsigned cpu; |
c8e251fa | 6287 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 6288 | for_each_possible_cpu(cpu) |
169f6c19 CS |
6289 | pageset_set_high_and_batch(zone, |
6290 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 6291 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
6292 | } |
6293 | #endif | |
6294 | ||
340175b7 JL |
6295 | void zone_pcp_reset(struct zone *zone) |
6296 | { | |
6297 | unsigned long flags; | |
5a883813 MK |
6298 | int cpu; |
6299 | struct per_cpu_pageset *pset; | |
340175b7 JL |
6300 | |
6301 | /* avoid races with drain_pages() */ | |
6302 | local_irq_save(flags); | |
6303 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
6304 | for_each_online_cpu(cpu) { |
6305 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6306 | drain_zonestat(zone, pset); | |
6307 | } | |
340175b7 JL |
6308 | free_percpu(zone->pageset); |
6309 | zone->pageset = &boot_pageset; | |
6310 | } | |
6311 | local_irq_restore(flags); | |
6312 | } | |
6313 | ||
6dcd73d7 | 6314 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
6315 | /* |
6316 | * All pages in the range must be isolated before calling this. | |
6317 | */ | |
6318 | void | |
6319 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
6320 | { | |
6321 | struct page *page; | |
6322 | struct zone *zone; | |
6323 | int order, i; | |
6324 | unsigned long pfn; | |
6325 | unsigned long flags; | |
6326 | /* find the first valid pfn */ | |
6327 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
6328 | if (pfn_valid(pfn)) | |
6329 | break; | |
6330 | if (pfn == end_pfn) | |
6331 | return; | |
6332 | zone = page_zone(pfn_to_page(pfn)); | |
6333 | spin_lock_irqsave(&zone->lock, flags); | |
6334 | pfn = start_pfn; | |
6335 | while (pfn < end_pfn) { | |
6336 | if (!pfn_valid(pfn)) { | |
6337 | pfn++; | |
6338 | continue; | |
6339 | } | |
6340 | page = pfn_to_page(pfn); | |
b023f468 WC |
6341 | /* |
6342 | * The HWPoisoned page may be not in buddy system, and | |
6343 | * page_count() is not 0. | |
6344 | */ | |
6345 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6346 | pfn++; | |
6347 | SetPageReserved(page); | |
6348 | continue; | |
6349 | } | |
6350 | ||
0c0e6195 KH |
6351 | BUG_ON(page_count(page)); |
6352 | BUG_ON(!PageBuddy(page)); | |
6353 | order = page_order(page); | |
6354 | #ifdef CONFIG_DEBUG_VM | |
6355 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
6356 | pfn, 1 << order, end_pfn); | |
6357 | #endif | |
6358 | list_del(&page->lru); | |
6359 | rmv_page_order(page); | |
6360 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
6361 | for (i = 0; i < (1 << order); i++) |
6362 | SetPageReserved((page+i)); | |
6363 | pfn += (1 << order); | |
6364 | } | |
6365 | spin_unlock_irqrestore(&zone->lock, flags); | |
6366 | } | |
6367 | #endif | |
8d22ba1b WF |
6368 | |
6369 | #ifdef CONFIG_MEMORY_FAILURE | |
6370 | bool is_free_buddy_page(struct page *page) | |
6371 | { | |
6372 | struct zone *zone = page_zone(page); | |
6373 | unsigned long pfn = page_to_pfn(page); | |
6374 | unsigned long flags; | |
6375 | int order; | |
6376 | ||
6377 | spin_lock_irqsave(&zone->lock, flags); | |
6378 | for (order = 0; order < MAX_ORDER; order++) { | |
6379 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
6380 | ||
6381 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
6382 | break; | |
6383 | } | |
6384 | spin_unlock_irqrestore(&zone->lock, flags); | |
6385 | ||
6386 | return order < MAX_ORDER; | |
6387 | } | |
6388 | #endif | |
718a3821 | 6389 | |
51300cef | 6390 | static const struct trace_print_flags pageflag_names[] = { |
718a3821 WF |
6391 | {1UL << PG_locked, "locked" }, |
6392 | {1UL << PG_error, "error" }, | |
6393 | {1UL << PG_referenced, "referenced" }, | |
6394 | {1UL << PG_uptodate, "uptodate" }, | |
6395 | {1UL << PG_dirty, "dirty" }, | |
6396 | {1UL << PG_lru, "lru" }, | |
6397 | {1UL << PG_active, "active" }, | |
6398 | {1UL << PG_slab, "slab" }, | |
6399 | {1UL << PG_owner_priv_1, "owner_priv_1" }, | |
6400 | {1UL << PG_arch_1, "arch_1" }, | |
6401 | {1UL << PG_reserved, "reserved" }, | |
6402 | {1UL << PG_private, "private" }, | |
6403 | {1UL << PG_private_2, "private_2" }, | |
6404 | {1UL << PG_writeback, "writeback" }, | |
6405 | #ifdef CONFIG_PAGEFLAGS_EXTENDED | |
6406 | {1UL << PG_head, "head" }, | |
6407 | {1UL << PG_tail, "tail" }, | |
6408 | #else | |
6409 | {1UL << PG_compound, "compound" }, | |
6410 | #endif | |
6411 | {1UL << PG_swapcache, "swapcache" }, | |
6412 | {1UL << PG_mappedtodisk, "mappedtodisk" }, | |
6413 | {1UL << PG_reclaim, "reclaim" }, | |
718a3821 WF |
6414 | {1UL << PG_swapbacked, "swapbacked" }, |
6415 | {1UL << PG_unevictable, "unevictable" }, | |
6416 | #ifdef CONFIG_MMU | |
6417 | {1UL << PG_mlocked, "mlocked" }, | |
6418 | #endif | |
6419 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED | |
6420 | {1UL << PG_uncached, "uncached" }, | |
6421 | #endif | |
6422 | #ifdef CONFIG_MEMORY_FAILURE | |
6423 | {1UL << PG_hwpoison, "hwpoison" }, | |
be9cd873 GS |
6424 | #endif |
6425 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6426 | {1UL << PG_compound_lock, "compound_lock" }, | |
718a3821 | 6427 | #endif |
718a3821 WF |
6428 | }; |
6429 | ||
6430 | static void dump_page_flags(unsigned long flags) | |
6431 | { | |
6432 | const char *delim = ""; | |
6433 | unsigned long mask; | |
6434 | int i; | |
6435 | ||
51300cef | 6436 | BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS); |
acc50c11 | 6437 | |
718a3821 WF |
6438 | printk(KERN_ALERT "page flags: %#lx(", flags); |
6439 | ||
6440 | /* remove zone id */ | |
6441 | flags &= (1UL << NR_PAGEFLAGS) - 1; | |
6442 | ||
51300cef | 6443 | for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) { |
718a3821 WF |
6444 | |
6445 | mask = pageflag_names[i].mask; | |
6446 | if ((flags & mask) != mask) | |
6447 | continue; | |
6448 | ||
6449 | flags &= ~mask; | |
6450 | printk("%s%s", delim, pageflag_names[i].name); | |
6451 | delim = "|"; | |
6452 | } | |
6453 | ||
6454 | /* check for left over flags */ | |
6455 | if (flags) | |
6456 | printk("%s%#lx", delim, flags); | |
6457 | ||
6458 | printk(")\n"); | |
6459 | } | |
6460 | ||
6461 | void dump_page(struct page *page) | |
6462 | { | |
6463 | printk(KERN_ALERT | |
6464 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", | |
4e9f64c4 | 6465 | page, atomic_read(&page->_count), page_mapcount(page), |
718a3821 WF |
6466 | page->mapping, page->index); |
6467 | dump_page_flags(page->flags); | |
f212ad7c | 6468 | mem_cgroup_print_bad_page(page); |
718a3821 | 6469 | } |