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> |
eefa864b | 51 | #include <linux/page_ext.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> |
268bb0ce | 56 | #include <linux/prefetch.h> |
6e543d57 | 57 | #include <linux/mm_inline.h> |
041d3a8c | 58 | #include <linux/migrate.h> |
e30825f1 | 59 | #include <linux/page_ext.h> |
949f7ec5 | 60 | #include <linux/hugetlb.h> |
8bd75c77 | 61 | #include <linux/sched/rt.h> |
48c96a36 | 62 | #include <linux/page_owner.h> |
1da177e4 | 63 | |
7ee3d4e8 | 64 | #include <asm/sections.h> |
1da177e4 | 65 | #include <asm/tlbflush.h> |
ac924c60 | 66 | #include <asm/div64.h> |
1da177e4 LT |
67 | #include "internal.h" |
68 | ||
c8e251fa CS |
69 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
70 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 71 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 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_); | |
ad2c8144 | 87 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
88 | #endif |
89 | ||
1da177e4 | 90 | /* |
13808910 | 91 | * Array of node states. |
1da177e4 | 92 | */ |
13808910 CL |
93 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
94 | [N_POSSIBLE] = NODE_MASK_ALL, | |
95 | [N_ONLINE] = { { [0] = 1UL } }, | |
96 | #ifndef CONFIG_NUMA | |
97 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
98 | #ifdef CONFIG_HIGHMEM | |
99 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
100 | #endif |
101 | #ifdef CONFIG_MOVABLE_NODE | |
102 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
103 | #endif |
104 | [N_CPU] = { { [0] = 1UL } }, | |
105 | #endif /* NUMA */ | |
106 | }; | |
107 | EXPORT_SYMBOL(node_states); | |
108 | ||
c3d5f5f0 JL |
109 | /* Protect totalram_pages and zone->managed_pages */ |
110 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
111 | ||
6c231b7b | 112 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 113 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 114 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 JW |
115 | /* |
116 | * When calculating the number of globally allowed dirty pages, there | |
117 | * is a certain number of per-zone reserves that should not be | |
118 | * considered dirtyable memory. This is the sum of those reserves | |
119 | * over all existing zones that contribute dirtyable memory. | |
120 | */ | |
121 | unsigned long dirty_balance_reserve __read_mostly; | |
122 | ||
1b76b02f | 123 | int percpu_pagelist_fraction; |
dcce284a | 124 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 125 | |
452aa699 RW |
126 | #ifdef CONFIG_PM_SLEEP |
127 | /* | |
128 | * The following functions are used by the suspend/hibernate code to temporarily | |
129 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
130 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
131 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
132 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
133 | * guaranteed not to run in parallel with that modification). | |
134 | */ | |
c9e664f1 RW |
135 | |
136 | static gfp_t saved_gfp_mask; | |
137 | ||
138 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
139 | { |
140 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
141 | if (saved_gfp_mask) { |
142 | gfp_allowed_mask = saved_gfp_mask; | |
143 | saved_gfp_mask = 0; | |
144 | } | |
452aa699 RW |
145 | } |
146 | ||
c9e664f1 | 147 | void pm_restrict_gfp_mask(void) |
452aa699 | 148 | { |
452aa699 | 149 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
150 | WARN_ON(saved_gfp_mask); |
151 | saved_gfp_mask = gfp_allowed_mask; | |
152 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 | 153 | } |
f90ac398 MG |
154 | |
155 | bool pm_suspended_storage(void) | |
156 | { | |
157 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) | |
158 | return false; | |
159 | return true; | |
160 | } | |
452aa699 RW |
161 | #endif /* CONFIG_PM_SLEEP */ |
162 | ||
d9c23400 MG |
163 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
164 | int pageblock_order __read_mostly; | |
165 | #endif | |
166 | ||
d98c7a09 | 167 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 168 | |
1da177e4 LT |
169 | /* |
170 | * results with 256, 32 in the lowmem_reserve sysctl: | |
171 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
172 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
173 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
174 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
175 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
176 | * |
177 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
178 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 179 | */ |
2f1b6248 | 180 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 181 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 182 | 256, |
4b51d669 | 183 | #endif |
fb0e7942 | 184 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 185 | 256, |
fb0e7942 | 186 | #endif |
e53ef38d | 187 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 188 | 32, |
e53ef38d | 189 | #endif |
2a1e274a | 190 | 32, |
2f1b6248 | 191 | }; |
1da177e4 LT |
192 | |
193 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 194 | |
15ad7cdc | 195 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 196 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 197 | "DMA", |
4b51d669 | 198 | #endif |
fb0e7942 | 199 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 200 | "DMA32", |
fb0e7942 | 201 | #endif |
2f1b6248 | 202 | "Normal", |
e53ef38d | 203 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 204 | "HighMem", |
e53ef38d | 205 | #endif |
2a1e274a | 206 | "Movable", |
2f1b6248 CL |
207 | }; |
208 | ||
1da177e4 | 209 | int min_free_kbytes = 1024; |
42aa83cb | 210 | int user_min_free_kbytes = -1; |
1da177e4 | 211 | |
2c85f51d JB |
212 | static unsigned long __meminitdata nr_kernel_pages; |
213 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 214 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 215 | |
0ee332c1 TH |
216 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
217 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
218 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
219 | static unsigned long __initdata required_kernelcore; | |
220 | static unsigned long __initdata required_movablecore; | |
221 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
222 | ||
223 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
224 | int movable_zone; | |
225 | EXPORT_SYMBOL(movable_zone); | |
226 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 227 | |
418508c1 MS |
228 | #if MAX_NUMNODES > 1 |
229 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 230 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 231 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 232 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
233 | #endif |
234 | ||
9ef9acb0 MG |
235 | int page_group_by_mobility_disabled __read_mostly; |
236 | ||
ee6f509c | 237 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 238 | { |
5d0f3f72 KM |
239 | if (unlikely(page_group_by_mobility_disabled && |
240 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
241 | migratetype = MIGRATE_UNMOVABLE; |
242 | ||
b2a0ac88 MG |
243 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
244 | PB_migrate, PB_migrate_end); | |
245 | } | |
246 | ||
7f33d49a RW |
247 | bool oom_killer_disabled __read_mostly; |
248 | ||
13e7444b | 249 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 250 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 251 | { |
bdc8cb98 DH |
252 | int ret = 0; |
253 | unsigned seq; | |
254 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 255 | unsigned long sp, start_pfn; |
c6a57e19 | 256 | |
bdc8cb98 DH |
257 | do { |
258 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
259 | start_pfn = zone->zone_start_pfn; |
260 | sp = zone->spanned_pages; | |
108bcc96 | 261 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
262 | ret = 1; |
263 | } while (zone_span_seqretry(zone, seq)); | |
264 | ||
b5e6a5a2 | 265 | if (ret) |
613813e8 DH |
266 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
267 | pfn, zone_to_nid(zone), zone->name, | |
268 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 269 | |
bdc8cb98 | 270 | return ret; |
c6a57e19 DH |
271 | } |
272 | ||
273 | static int page_is_consistent(struct zone *zone, struct page *page) | |
274 | { | |
14e07298 | 275 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 276 | return 0; |
1da177e4 | 277 | if (zone != page_zone(page)) |
c6a57e19 DH |
278 | return 0; |
279 | ||
280 | return 1; | |
281 | } | |
282 | /* | |
283 | * Temporary debugging check for pages not lying within a given zone. | |
284 | */ | |
285 | static int bad_range(struct zone *zone, struct page *page) | |
286 | { | |
287 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 288 | return 1; |
c6a57e19 DH |
289 | if (!page_is_consistent(zone, page)) |
290 | return 1; | |
291 | ||
1da177e4 LT |
292 | return 0; |
293 | } | |
13e7444b NP |
294 | #else |
295 | static inline int bad_range(struct zone *zone, struct page *page) | |
296 | { | |
297 | return 0; | |
298 | } | |
299 | #endif | |
300 | ||
d230dec1 KS |
301 | static void bad_page(struct page *page, const char *reason, |
302 | unsigned long bad_flags) | |
1da177e4 | 303 | { |
d936cf9b HD |
304 | static unsigned long resume; |
305 | static unsigned long nr_shown; | |
306 | static unsigned long nr_unshown; | |
307 | ||
2a7684a2 WF |
308 | /* Don't complain about poisoned pages */ |
309 | if (PageHWPoison(page)) { | |
22b751c3 | 310 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
311 | return; |
312 | } | |
313 | ||
d936cf9b HD |
314 | /* |
315 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
316 | * or allow a steady drip of one report per second. | |
317 | */ | |
318 | if (nr_shown == 60) { | |
319 | if (time_before(jiffies, resume)) { | |
320 | nr_unshown++; | |
321 | goto out; | |
322 | } | |
323 | if (nr_unshown) { | |
1e9e6365 HD |
324 | printk(KERN_ALERT |
325 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
326 | nr_unshown); |
327 | nr_unshown = 0; | |
328 | } | |
329 | nr_shown = 0; | |
330 | } | |
331 | if (nr_shown++ == 0) | |
332 | resume = jiffies + 60 * HZ; | |
333 | ||
1e9e6365 | 334 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 335 | current->comm, page_to_pfn(page)); |
f0b791a3 | 336 | dump_page_badflags(page, reason, bad_flags); |
3dc14741 | 337 | |
4f31888c | 338 | print_modules(); |
1da177e4 | 339 | dump_stack(); |
d936cf9b | 340 | out: |
8cc3b392 | 341 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 342 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 343 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
344 | } |
345 | ||
1da177e4 LT |
346 | /* |
347 | * Higher-order pages are called "compound pages". They are structured thusly: | |
348 | * | |
349 | * The first PAGE_SIZE page is called the "head page". | |
350 | * | |
351 | * The remaining PAGE_SIZE pages are called "tail pages". | |
352 | * | |
6416b9fa WSH |
353 | * All pages have PG_compound set. All tail pages have their ->first_page |
354 | * pointing at the head page. | |
1da177e4 | 355 | * |
41d78ba5 HD |
356 | * The first tail page's ->lru.next holds the address of the compound page's |
357 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
358 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 359 | */ |
d98c7a09 HD |
360 | |
361 | static void free_compound_page(struct page *page) | |
362 | { | |
d85f3385 | 363 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
364 | } |
365 | ||
01ad1c08 | 366 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
367 | { |
368 | int i; | |
369 | int nr_pages = 1 << order; | |
370 | ||
371 | set_compound_page_dtor(page, free_compound_page); | |
372 | set_compound_order(page, order); | |
373 | __SetPageHead(page); | |
374 | for (i = 1; i < nr_pages; i++) { | |
375 | struct page *p = page + i; | |
58a84aa9 | 376 | set_page_count(p, 0); |
18229df5 | 377 | p->first_page = page; |
668f9abb DR |
378 | /* Make sure p->first_page is always valid for PageTail() */ |
379 | smp_wmb(); | |
380 | __SetPageTail(p); | |
18229df5 AW |
381 | } |
382 | } | |
383 | ||
59ff4216 | 384 | /* update __split_huge_page_refcount if you change this function */ |
8cc3b392 | 385 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
386 | { |
387 | int i; | |
388 | int nr_pages = 1 << order; | |
8cc3b392 | 389 | int bad = 0; |
1da177e4 | 390 | |
0bb2c763 | 391 | if (unlikely(compound_order(page) != order)) { |
f0b791a3 | 392 | bad_page(page, "wrong compound order", 0); |
8cc3b392 HD |
393 | bad++; |
394 | } | |
1da177e4 | 395 | |
6d777953 | 396 | __ClearPageHead(page); |
8cc3b392 | 397 | |
18229df5 AW |
398 | for (i = 1; i < nr_pages; i++) { |
399 | struct page *p = page + i; | |
1da177e4 | 400 | |
f0b791a3 DH |
401 | if (unlikely(!PageTail(p))) { |
402 | bad_page(page, "PageTail not set", 0); | |
403 | bad++; | |
404 | } else if (unlikely(p->first_page != page)) { | |
405 | bad_page(page, "first_page not consistent", 0); | |
8cc3b392 HD |
406 | bad++; |
407 | } | |
d85f3385 | 408 | __ClearPageTail(p); |
1da177e4 | 409 | } |
8cc3b392 HD |
410 | |
411 | return bad; | |
1da177e4 | 412 | } |
1da177e4 | 413 | |
7aeb09f9 MG |
414 | static inline void prep_zero_page(struct page *page, unsigned int order, |
415 | gfp_t gfp_flags) | |
17cf4406 NP |
416 | { |
417 | int i; | |
418 | ||
6626c5d5 AM |
419 | /* |
420 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
421 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
422 | */ | |
725d704e | 423 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
424 | for (i = 0; i < (1 << order); i++) |
425 | clear_highpage(page + i); | |
426 | } | |
427 | ||
c0a32fc5 SG |
428 | #ifdef CONFIG_DEBUG_PAGEALLOC |
429 | unsigned int _debug_guardpage_minorder; | |
031bc574 | 430 | bool _debug_pagealloc_enabled __read_mostly; |
e30825f1 JK |
431 | bool _debug_guardpage_enabled __read_mostly; |
432 | ||
031bc574 JK |
433 | static int __init early_debug_pagealloc(char *buf) |
434 | { | |
435 | if (!buf) | |
436 | return -EINVAL; | |
437 | ||
438 | if (strcmp(buf, "on") == 0) | |
439 | _debug_pagealloc_enabled = true; | |
440 | ||
441 | return 0; | |
442 | } | |
443 | early_param("debug_pagealloc", early_debug_pagealloc); | |
444 | ||
e30825f1 JK |
445 | static bool need_debug_guardpage(void) |
446 | { | |
031bc574 JK |
447 | /* If we don't use debug_pagealloc, we don't need guard page */ |
448 | if (!debug_pagealloc_enabled()) | |
449 | return false; | |
450 | ||
e30825f1 JK |
451 | return true; |
452 | } | |
453 | ||
454 | static void init_debug_guardpage(void) | |
455 | { | |
031bc574 JK |
456 | if (!debug_pagealloc_enabled()) |
457 | return; | |
458 | ||
e30825f1 JK |
459 | _debug_guardpage_enabled = true; |
460 | } | |
461 | ||
462 | struct page_ext_operations debug_guardpage_ops = { | |
463 | .need = need_debug_guardpage, | |
464 | .init = init_debug_guardpage, | |
465 | }; | |
c0a32fc5 SG |
466 | |
467 | static int __init debug_guardpage_minorder_setup(char *buf) | |
468 | { | |
469 | unsigned long res; | |
470 | ||
471 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
472 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
473 | return 0; | |
474 | } | |
475 | _debug_guardpage_minorder = res; | |
476 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
477 | return 0; | |
478 | } | |
479 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
480 | ||
2847cf95 JK |
481 | static inline void set_page_guard(struct zone *zone, struct page *page, |
482 | unsigned int order, int migratetype) | |
c0a32fc5 | 483 | { |
e30825f1 JK |
484 | struct page_ext *page_ext; |
485 | ||
486 | if (!debug_guardpage_enabled()) | |
487 | return; | |
488 | ||
489 | page_ext = lookup_page_ext(page); | |
490 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
491 | ||
2847cf95 JK |
492 | INIT_LIST_HEAD(&page->lru); |
493 | set_page_private(page, order); | |
494 | /* Guard pages are not available for any usage */ | |
495 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
c0a32fc5 SG |
496 | } |
497 | ||
2847cf95 JK |
498 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
499 | unsigned int order, int migratetype) | |
c0a32fc5 | 500 | { |
e30825f1 JK |
501 | struct page_ext *page_ext; |
502 | ||
503 | if (!debug_guardpage_enabled()) | |
504 | return; | |
505 | ||
506 | page_ext = lookup_page_ext(page); | |
507 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
508 | ||
2847cf95 JK |
509 | set_page_private(page, 0); |
510 | if (!is_migrate_isolate(migratetype)) | |
511 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
512 | } |
513 | #else | |
e30825f1 | 514 | struct page_ext_operations debug_guardpage_ops = { NULL, }; |
2847cf95 JK |
515 | static inline void set_page_guard(struct zone *zone, struct page *page, |
516 | unsigned int order, int migratetype) {} | |
517 | static inline void clear_page_guard(struct zone *zone, struct page *page, | |
518 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
519 | #endif |
520 | ||
7aeb09f9 | 521 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 522 | { |
4c21e2f2 | 523 | set_page_private(page, order); |
676165a8 | 524 | __SetPageBuddy(page); |
1da177e4 LT |
525 | } |
526 | ||
527 | static inline void rmv_page_order(struct page *page) | |
528 | { | |
676165a8 | 529 | __ClearPageBuddy(page); |
4c21e2f2 | 530 | set_page_private(page, 0); |
1da177e4 LT |
531 | } |
532 | ||
1da177e4 LT |
533 | /* |
534 | * This function checks whether a page is free && is the buddy | |
535 | * we can do coalesce a page and its buddy if | |
13e7444b | 536 | * (a) the buddy is not in a hole && |
676165a8 | 537 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
538 | * (c) a page and its buddy have the same order && |
539 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 540 | * |
cf6fe945 WSH |
541 | * For recording whether a page is in the buddy system, we set ->_mapcount |
542 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
543 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
544 | * serialized by zone->lock. | |
1da177e4 | 545 | * |
676165a8 | 546 | * For recording page's order, we use page_private(page). |
1da177e4 | 547 | */ |
cb2b95e1 | 548 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 549 | unsigned int order) |
1da177e4 | 550 | { |
14e07298 | 551 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 552 | return 0; |
13e7444b | 553 | |
c0a32fc5 | 554 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
309381fe | 555 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
d34c5fa0 MG |
556 | |
557 | if (page_zone_id(page) != page_zone_id(buddy)) | |
558 | return 0; | |
559 | ||
c0a32fc5 SG |
560 | return 1; |
561 | } | |
562 | ||
cb2b95e1 | 563 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
309381fe | 564 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
d34c5fa0 MG |
565 | |
566 | /* | |
567 | * zone check is done late to avoid uselessly | |
568 | * calculating zone/node ids for pages that could | |
569 | * never merge. | |
570 | */ | |
571 | if (page_zone_id(page) != page_zone_id(buddy)) | |
572 | return 0; | |
573 | ||
6aa3001b | 574 | return 1; |
676165a8 | 575 | } |
6aa3001b | 576 | return 0; |
1da177e4 LT |
577 | } |
578 | ||
579 | /* | |
580 | * Freeing function for a buddy system allocator. | |
581 | * | |
582 | * The concept of a buddy system is to maintain direct-mapped table | |
583 | * (containing bit values) for memory blocks of various "orders". | |
584 | * The bottom level table contains the map for the smallest allocatable | |
585 | * units of memory (here, pages), and each level above it describes | |
586 | * pairs of units from the levels below, hence, "buddies". | |
587 | * At a high level, all that happens here is marking the table entry | |
588 | * at the bottom level available, and propagating the changes upward | |
589 | * as necessary, plus some accounting needed to play nicely with other | |
590 | * parts of the VM system. | |
591 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
592 | * free pages of length of (1 << order) and marked with _mapcount |
593 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
594 | * field. | |
1da177e4 | 595 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
596 | * other. That is, if we allocate a small block, and both were |
597 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 598 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 599 | * triggers coalescing into a block of larger size. |
1da177e4 | 600 | * |
6d49e352 | 601 | * -- nyc |
1da177e4 LT |
602 | */ |
603 | ||
48db57f8 | 604 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 605 | unsigned long pfn, |
ed0ae21d MG |
606 | struct zone *zone, unsigned int order, |
607 | int migratetype) | |
1da177e4 LT |
608 | { |
609 | unsigned long page_idx; | |
6dda9d55 | 610 | unsigned long combined_idx; |
43506fad | 611 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 612 | struct page *buddy; |
3c605096 | 613 | int max_order = MAX_ORDER; |
1da177e4 | 614 | |
d29bb978 CS |
615 | VM_BUG_ON(!zone_is_initialized(zone)); |
616 | ||
224abf92 | 617 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
618 | if (unlikely(destroy_compound_page(page, order))) |
619 | return; | |
1da177e4 | 620 | |
ed0ae21d | 621 | VM_BUG_ON(migratetype == -1); |
3c605096 JK |
622 | if (is_migrate_isolate(migratetype)) { |
623 | /* | |
624 | * We restrict max order of merging to prevent merge | |
625 | * between freepages on isolate pageblock and normal | |
626 | * pageblock. Without this, pageblock isolation | |
627 | * could cause incorrect freepage accounting. | |
628 | */ | |
629 | max_order = min(MAX_ORDER, pageblock_order + 1); | |
630 | } else { | |
8f82b55d | 631 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
3c605096 | 632 | } |
ed0ae21d | 633 | |
3c605096 | 634 | page_idx = pfn & ((1 << max_order) - 1); |
1da177e4 | 635 | |
309381fe SL |
636 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
637 | VM_BUG_ON_PAGE(bad_range(zone, page), page); | |
1da177e4 | 638 | |
3c605096 | 639 | while (order < max_order - 1) { |
43506fad KC |
640 | buddy_idx = __find_buddy_index(page_idx, order); |
641 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 642 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 643 | break; |
c0a32fc5 SG |
644 | /* |
645 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
646 | * merge with it and move up one order. | |
647 | */ | |
648 | if (page_is_guard(buddy)) { | |
2847cf95 | 649 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
650 | } else { |
651 | list_del(&buddy->lru); | |
652 | zone->free_area[order].nr_free--; | |
653 | rmv_page_order(buddy); | |
654 | } | |
43506fad | 655 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
656 | page = page + (combined_idx - page_idx); |
657 | page_idx = combined_idx; | |
658 | order++; | |
659 | } | |
660 | set_page_order(page, order); | |
6dda9d55 CZ |
661 | |
662 | /* | |
663 | * If this is not the largest possible page, check if the buddy | |
664 | * of the next-highest order is free. If it is, it's possible | |
665 | * that pages are being freed that will coalesce soon. In case, | |
666 | * that is happening, add the free page to the tail of the list | |
667 | * so it's less likely to be used soon and more likely to be merged | |
668 | * as a higher order page | |
669 | */ | |
b7f50cfa | 670 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 671 | struct page *higher_page, *higher_buddy; |
43506fad KC |
672 | combined_idx = buddy_idx & page_idx; |
673 | higher_page = page + (combined_idx - page_idx); | |
674 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 675 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
676 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
677 | list_add_tail(&page->lru, | |
678 | &zone->free_area[order].free_list[migratetype]); | |
679 | goto out; | |
680 | } | |
681 | } | |
682 | ||
683 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
684 | out: | |
1da177e4 LT |
685 | zone->free_area[order].nr_free++; |
686 | } | |
687 | ||
224abf92 | 688 | static inline int free_pages_check(struct page *page) |
1da177e4 | 689 | { |
d230dec1 | 690 | const char *bad_reason = NULL; |
f0b791a3 DH |
691 | unsigned long bad_flags = 0; |
692 | ||
693 | if (unlikely(page_mapcount(page))) | |
694 | bad_reason = "nonzero mapcount"; | |
695 | if (unlikely(page->mapping != NULL)) | |
696 | bad_reason = "non-NULL mapping"; | |
697 | if (unlikely(atomic_read(&page->_count) != 0)) | |
698 | bad_reason = "nonzero _count"; | |
699 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { | |
700 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
701 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
702 | } | |
9edad6ea JW |
703 | #ifdef CONFIG_MEMCG |
704 | if (unlikely(page->mem_cgroup)) | |
705 | bad_reason = "page still charged to cgroup"; | |
706 | #endif | |
f0b791a3 DH |
707 | if (unlikely(bad_reason)) { |
708 | bad_page(page, bad_reason, bad_flags); | |
79f4b7bf | 709 | return 1; |
8cc3b392 | 710 | } |
90572890 | 711 | page_cpupid_reset_last(page); |
79f4b7bf HD |
712 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
713 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
714 | return 0; | |
1da177e4 LT |
715 | } |
716 | ||
717 | /* | |
5f8dcc21 | 718 | * Frees a number of pages from the PCP lists |
1da177e4 | 719 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 720 | * count is the number of pages to free. |
1da177e4 LT |
721 | * |
722 | * If the zone was previously in an "all pages pinned" state then look to | |
723 | * see if this freeing clears that state. | |
724 | * | |
725 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
726 | * pinned" detection logic. | |
727 | */ | |
5f8dcc21 MG |
728 | static void free_pcppages_bulk(struct zone *zone, int count, |
729 | struct per_cpu_pages *pcp) | |
1da177e4 | 730 | { |
5f8dcc21 | 731 | int migratetype = 0; |
a6f9edd6 | 732 | int batch_free = 0; |
72853e29 | 733 | int to_free = count; |
0d5d823a | 734 | unsigned long nr_scanned; |
5f8dcc21 | 735 | |
c54ad30c | 736 | spin_lock(&zone->lock); |
0d5d823a MG |
737 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
738 | if (nr_scanned) | |
739 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 740 | |
72853e29 | 741 | while (to_free) { |
48db57f8 | 742 | struct page *page; |
5f8dcc21 MG |
743 | struct list_head *list; |
744 | ||
745 | /* | |
a6f9edd6 MG |
746 | * Remove pages from lists in a round-robin fashion. A |
747 | * batch_free count is maintained that is incremented when an | |
748 | * empty list is encountered. This is so more pages are freed | |
749 | * off fuller lists instead of spinning excessively around empty | |
750 | * lists | |
5f8dcc21 MG |
751 | */ |
752 | do { | |
a6f9edd6 | 753 | batch_free++; |
5f8dcc21 MG |
754 | if (++migratetype == MIGRATE_PCPTYPES) |
755 | migratetype = 0; | |
756 | list = &pcp->lists[migratetype]; | |
757 | } while (list_empty(list)); | |
48db57f8 | 758 | |
1d16871d NK |
759 | /* This is the only non-empty list. Free them all. */ |
760 | if (batch_free == MIGRATE_PCPTYPES) | |
761 | batch_free = to_free; | |
762 | ||
a6f9edd6 | 763 | do { |
770c8aaa BZ |
764 | int mt; /* migratetype of the to-be-freed page */ |
765 | ||
a6f9edd6 MG |
766 | page = list_entry(list->prev, struct page, lru); |
767 | /* must delete as __free_one_page list manipulates */ | |
768 | list_del(&page->lru); | |
b12c4ad1 | 769 | mt = get_freepage_migratetype(page); |
8f82b55d | 770 | if (unlikely(has_isolate_pageblock(zone))) |
51bb1a40 | 771 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 772 | |
a7016235 | 773 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
dc4b0caf | 774 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 775 | trace_mm_page_pcpu_drain(page, 0, mt); |
72853e29 | 776 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 777 | } |
c54ad30c | 778 | spin_unlock(&zone->lock); |
1da177e4 LT |
779 | } |
780 | ||
dc4b0caf MG |
781 | static void free_one_page(struct zone *zone, |
782 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 783 | unsigned int order, |
ed0ae21d | 784 | int migratetype) |
1da177e4 | 785 | { |
0d5d823a | 786 | unsigned long nr_scanned; |
006d22d9 | 787 | spin_lock(&zone->lock); |
0d5d823a MG |
788 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
789 | if (nr_scanned) | |
790 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 791 | |
ad53f92e JK |
792 | if (unlikely(has_isolate_pageblock(zone) || |
793 | is_migrate_isolate(migratetype))) { | |
794 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 795 | } |
dc4b0caf | 796 | __free_one_page(page, pfn, zone, order, migratetype); |
006d22d9 | 797 | spin_unlock(&zone->lock); |
48db57f8 NP |
798 | } |
799 | ||
ec95f53a | 800 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 801 | { |
1da177e4 | 802 | int i; |
8cc3b392 | 803 | int bad = 0; |
1da177e4 | 804 | |
ab1f306f YZ |
805 | VM_BUG_ON_PAGE(PageTail(page), page); |
806 | VM_BUG_ON_PAGE(PageHead(page) && compound_order(page) != order, page); | |
807 | ||
b413d48a | 808 | trace_mm_page_free(page, order); |
b1eeab67 VN |
809 | kmemcheck_free_shadow(page, order); |
810 | ||
8dd60a3a AA |
811 | if (PageAnon(page)) |
812 | page->mapping = NULL; | |
813 | for (i = 0; i < (1 << order); i++) | |
814 | bad += free_pages_check(page + i); | |
8cc3b392 | 815 | if (bad) |
ec95f53a | 816 | return false; |
689bcebf | 817 | |
48c96a36 JK |
818 | reset_page_owner(page, order); |
819 | ||
3ac7fe5a | 820 | if (!PageHighMem(page)) { |
b8af2941 PK |
821 | debug_check_no_locks_freed(page_address(page), |
822 | PAGE_SIZE << order); | |
3ac7fe5a TG |
823 | debug_check_no_obj_freed(page_address(page), |
824 | PAGE_SIZE << order); | |
825 | } | |
dafb1367 | 826 | arch_free_page(page, order); |
48db57f8 | 827 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 828 | |
ec95f53a KM |
829 | return true; |
830 | } | |
831 | ||
832 | static void __free_pages_ok(struct page *page, unsigned int order) | |
833 | { | |
834 | unsigned long flags; | |
95e34412 | 835 | int migratetype; |
dc4b0caf | 836 | unsigned long pfn = page_to_pfn(page); |
ec95f53a KM |
837 | |
838 | if (!free_pages_prepare(page, order)) | |
839 | return; | |
840 | ||
cfc47a28 | 841 | migratetype = get_pfnblock_migratetype(page, pfn); |
c54ad30c | 842 | local_irq_save(flags); |
f8891e5e | 843 | __count_vm_events(PGFREE, 1 << order); |
95e34412 | 844 | set_freepage_migratetype(page, migratetype); |
dc4b0caf | 845 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
c54ad30c | 846 | local_irq_restore(flags); |
1da177e4 LT |
847 | } |
848 | ||
170a5a7e | 849 | void __init __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 | 850 | { |
c3993076 | 851 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 852 | struct page *p = page; |
c3993076 | 853 | unsigned int loop; |
a226f6c8 | 854 | |
e2d0bd2b YL |
855 | prefetchw(p); |
856 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
857 | prefetchw(p + 1); | |
c3993076 JW |
858 | __ClearPageReserved(p); |
859 | set_page_count(p, 0); | |
a226f6c8 | 860 | } |
e2d0bd2b YL |
861 | __ClearPageReserved(p); |
862 | set_page_count(p, 0); | |
c3993076 | 863 | |
e2d0bd2b | 864 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
865 | set_page_refcounted(page); |
866 | __free_pages(page, order); | |
a226f6c8 DH |
867 | } |
868 | ||
47118af0 | 869 | #ifdef CONFIG_CMA |
9cf510a5 | 870 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
871 | void __init init_cma_reserved_pageblock(struct page *page) |
872 | { | |
873 | unsigned i = pageblock_nr_pages; | |
874 | struct page *p = page; | |
875 | ||
876 | do { | |
877 | __ClearPageReserved(p); | |
878 | set_page_count(p, 0); | |
879 | } while (++p, --i); | |
880 | ||
47118af0 | 881 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
882 | |
883 | if (pageblock_order >= MAX_ORDER) { | |
884 | i = pageblock_nr_pages; | |
885 | p = page; | |
886 | do { | |
887 | set_page_refcounted(p); | |
888 | __free_pages(p, MAX_ORDER - 1); | |
889 | p += MAX_ORDER_NR_PAGES; | |
890 | } while (i -= MAX_ORDER_NR_PAGES); | |
891 | } else { | |
892 | set_page_refcounted(page); | |
893 | __free_pages(page, pageblock_order); | |
894 | } | |
895 | ||
3dcc0571 | 896 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
897 | } |
898 | #endif | |
1da177e4 LT |
899 | |
900 | /* | |
901 | * The order of subdivision here is critical for the IO subsystem. | |
902 | * Please do not alter this order without good reasons and regression | |
903 | * testing. Specifically, as large blocks of memory are subdivided, | |
904 | * the order in which smaller blocks are delivered depends on the order | |
905 | * they're subdivided in this function. This is the primary factor | |
906 | * influencing the order in which pages are delivered to the IO | |
907 | * subsystem according to empirical testing, and this is also justified | |
908 | * by considering the behavior of a buddy system containing a single | |
909 | * large block of memory acted on by a series of small allocations. | |
910 | * This behavior is a critical factor in sglist merging's success. | |
911 | * | |
6d49e352 | 912 | * -- nyc |
1da177e4 | 913 | */ |
085cc7d5 | 914 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
915 | int low, int high, struct free_area *area, |
916 | int migratetype) | |
1da177e4 LT |
917 | { |
918 | unsigned long size = 1 << high; | |
919 | ||
920 | while (high > low) { | |
921 | area--; | |
922 | high--; | |
923 | size >>= 1; | |
309381fe | 924 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 925 | |
2847cf95 | 926 | if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && |
e30825f1 | 927 | debug_guardpage_enabled() && |
2847cf95 | 928 | high < debug_guardpage_minorder()) { |
c0a32fc5 SG |
929 | /* |
930 | * Mark as guard pages (or page), that will allow to | |
931 | * merge back to allocator when buddy will be freed. | |
932 | * Corresponding page table entries will not be touched, | |
933 | * pages will stay not present in virtual address space | |
934 | */ | |
2847cf95 | 935 | set_page_guard(zone, &page[size], high, migratetype); |
c0a32fc5 SG |
936 | continue; |
937 | } | |
b2a0ac88 | 938 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
939 | area->nr_free++; |
940 | set_page_order(&page[size], high); | |
941 | } | |
1da177e4 LT |
942 | } |
943 | ||
1da177e4 LT |
944 | /* |
945 | * This page is about to be returned from the page allocator | |
946 | */ | |
2a7684a2 | 947 | static inline int check_new_page(struct page *page) |
1da177e4 | 948 | { |
d230dec1 | 949 | const char *bad_reason = NULL; |
f0b791a3 DH |
950 | unsigned long bad_flags = 0; |
951 | ||
952 | if (unlikely(page_mapcount(page))) | |
953 | bad_reason = "nonzero mapcount"; | |
954 | if (unlikely(page->mapping != NULL)) | |
955 | bad_reason = "non-NULL mapping"; | |
956 | if (unlikely(atomic_read(&page->_count) != 0)) | |
957 | bad_reason = "nonzero _count"; | |
958 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { | |
959 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
960 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
961 | } | |
9edad6ea JW |
962 | #ifdef CONFIG_MEMCG |
963 | if (unlikely(page->mem_cgroup)) | |
964 | bad_reason = "page still charged to cgroup"; | |
965 | #endif | |
f0b791a3 DH |
966 | if (unlikely(bad_reason)) { |
967 | bad_page(page, bad_reason, bad_flags); | |
689bcebf | 968 | return 1; |
8cc3b392 | 969 | } |
2a7684a2 WF |
970 | return 0; |
971 | } | |
972 | ||
7aeb09f9 | 973 | static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags) |
2a7684a2 WF |
974 | { |
975 | int i; | |
976 | ||
977 | for (i = 0; i < (1 << order); i++) { | |
978 | struct page *p = page + i; | |
979 | if (unlikely(check_new_page(p))) | |
980 | return 1; | |
981 | } | |
689bcebf | 982 | |
4c21e2f2 | 983 | set_page_private(page, 0); |
7835e98b | 984 | set_page_refcounted(page); |
cc102509 NP |
985 | |
986 | arch_alloc_page(page, order); | |
1da177e4 | 987 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
988 | |
989 | if (gfp_flags & __GFP_ZERO) | |
990 | prep_zero_page(page, order, gfp_flags); | |
991 | ||
992 | if (order && (gfp_flags & __GFP_COMP)) | |
993 | prep_compound_page(page, order); | |
994 | ||
48c96a36 JK |
995 | set_page_owner(page, order, gfp_flags); |
996 | ||
689bcebf | 997 | return 0; |
1da177e4 LT |
998 | } |
999 | ||
56fd56b8 MG |
1000 | /* |
1001 | * Go through the free lists for the given migratetype and remove | |
1002 | * the smallest available page from the freelists | |
1003 | */ | |
728ec980 MG |
1004 | static inline |
1005 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1006 | int migratetype) |
1007 | { | |
1008 | unsigned int current_order; | |
b8af2941 | 1009 | struct free_area *area; |
56fd56b8 MG |
1010 | struct page *page; |
1011 | ||
1012 | /* Find a page of the appropriate size in the preferred list */ | |
1013 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1014 | area = &(zone->free_area[current_order]); | |
1015 | if (list_empty(&area->free_list[migratetype])) | |
1016 | continue; | |
1017 | ||
1018 | page = list_entry(area->free_list[migratetype].next, | |
1019 | struct page, lru); | |
1020 | list_del(&page->lru); | |
1021 | rmv_page_order(page); | |
1022 | area->nr_free--; | |
56fd56b8 | 1023 | expand(zone, page, order, current_order, area, migratetype); |
5bcc9f86 | 1024 | set_freepage_migratetype(page, migratetype); |
56fd56b8 MG |
1025 | return page; |
1026 | } | |
1027 | ||
1028 | return NULL; | |
1029 | } | |
1030 | ||
1031 | ||
b2a0ac88 MG |
1032 | /* |
1033 | * This array describes the order lists are fallen back to when | |
1034 | * the free lists for the desirable migrate type are depleted | |
1035 | */ | |
47118af0 MN |
1036 | static int fallbacks[MIGRATE_TYPES][4] = { |
1037 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
1038 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
1039 | #ifdef CONFIG_CMA | |
1040 | [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
1041 | [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */ | |
1042 | #else | |
1043 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
1044 | #endif | |
6d4a4916 | 1045 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 1046 | #ifdef CONFIG_MEMORY_ISOLATION |
6d4a4916 | 1047 | [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 1048 | #endif |
b2a0ac88 MG |
1049 | }; |
1050 | ||
c361be55 MG |
1051 | /* |
1052 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1053 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1054 | * boundary. If alignment is required, use move_freepages_block() |
1055 | */ | |
435b405c | 1056 | int move_freepages(struct zone *zone, |
b69a7288 AB |
1057 | struct page *start_page, struct page *end_page, |
1058 | int migratetype) | |
c361be55 MG |
1059 | { |
1060 | struct page *page; | |
1061 | unsigned long order; | |
d100313f | 1062 | int pages_moved = 0; |
c361be55 MG |
1063 | |
1064 | #ifndef CONFIG_HOLES_IN_ZONE | |
1065 | /* | |
1066 | * page_zone is not safe to call in this context when | |
1067 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1068 | * anyway as we check zone boundaries in move_freepages_block(). | |
1069 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1070 | * grouping pages by mobility |
c361be55 | 1071 | */ |
97ee4ba7 | 1072 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1073 | #endif |
1074 | ||
1075 | for (page = start_page; page <= end_page;) { | |
344c790e | 1076 | /* Make sure we are not inadvertently changing nodes */ |
309381fe | 1077 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
344c790e | 1078 | |
c361be55 MG |
1079 | if (!pfn_valid_within(page_to_pfn(page))) { |
1080 | page++; | |
1081 | continue; | |
1082 | } | |
1083 | ||
1084 | if (!PageBuddy(page)) { | |
1085 | page++; | |
1086 | continue; | |
1087 | } | |
1088 | ||
1089 | order = page_order(page); | |
84be48d8 KS |
1090 | list_move(&page->lru, |
1091 | &zone->free_area[order].free_list[migratetype]); | |
95e34412 | 1092 | set_freepage_migratetype(page, migratetype); |
c361be55 | 1093 | page += 1 << order; |
d100313f | 1094 | pages_moved += 1 << order; |
c361be55 MG |
1095 | } |
1096 | ||
d100313f | 1097 | return pages_moved; |
c361be55 MG |
1098 | } |
1099 | ||
ee6f509c | 1100 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1101 | int migratetype) |
c361be55 MG |
1102 | { |
1103 | unsigned long start_pfn, end_pfn; | |
1104 | struct page *start_page, *end_page; | |
1105 | ||
1106 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1107 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1108 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1109 | end_page = start_page + pageblock_nr_pages - 1; |
1110 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1111 | |
1112 | /* Do not cross zone boundaries */ | |
108bcc96 | 1113 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1114 | start_page = page; |
108bcc96 | 1115 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1116 | return 0; |
1117 | ||
1118 | return move_freepages(zone, start_page, end_page, migratetype); | |
1119 | } | |
1120 | ||
2f66a68f MG |
1121 | static void change_pageblock_range(struct page *pageblock_page, |
1122 | int start_order, int migratetype) | |
1123 | { | |
1124 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1125 | ||
1126 | while (nr_pageblocks--) { | |
1127 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1128 | pageblock_page += pageblock_nr_pages; | |
1129 | } | |
1130 | } | |
1131 | ||
fef903ef SB |
1132 | /* |
1133 | * If breaking a large block of pages, move all free pages to the preferred | |
1134 | * allocation list. If falling back for a reclaimable kernel allocation, be | |
1135 | * more aggressive about taking ownership of free pages. | |
1136 | * | |
1137 | * On the other hand, never change migration type of MIGRATE_CMA pageblocks | |
1138 | * nor move CMA pages to different free lists. We don't want unmovable pages | |
1139 | * to be allocated from MIGRATE_CMA areas. | |
1140 | * | |
1141 | * Returns the new migratetype of the pageblock (or the same old migratetype | |
1142 | * if it was unchanged). | |
1143 | */ | |
1144 | static int try_to_steal_freepages(struct zone *zone, struct page *page, | |
1145 | int start_type, int fallback_type) | |
1146 | { | |
1147 | int current_order = page_order(page); | |
1148 | ||
0cbef29a KM |
1149 | /* |
1150 | * When borrowing from MIGRATE_CMA, we need to release the excess | |
5bcc9f86 VB |
1151 | * buddy pages to CMA itself. We also ensure the freepage_migratetype |
1152 | * is set to CMA so it is returned to the correct freelist in case | |
1153 | * the page ends up being not actually allocated from the pcp lists. | |
0cbef29a | 1154 | */ |
fef903ef SB |
1155 | if (is_migrate_cma(fallback_type)) |
1156 | return fallback_type; | |
1157 | ||
1158 | /* Take ownership for orders >= pageblock_order */ | |
1159 | if (current_order >= pageblock_order) { | |
1160 | change_pageblock_range(page, current_order, start_type); | |
1161 | return start_type; | |
1162 | } | |
1163 | ||
1164 | if (current_order >= pageblock_order / 2 || | |
1165 | start_type == MIGRATE_RECLAIMABLE || | |
1166 | page_group_by_mobility_disabled) { | |
1167 | int pages; | |
1168 | ||
1169 | pages = move_freepages_block(zone, page, start_type); | |
1170 | ||
1171 | /* Claim the whole block if over half of it is free */ | |
1172 | if (pages >= (1 << (pageblock_order-1)) || | |
1173 | page_group_by_mobility_disabled) { | |
1174 | ||
1175 | set_pageblock_migratetype(page, start_type); | |
1176 | return start_type; | |
1177 | } | |
1178 | ||
1179 | } | |
1180 | ||
1181 | return fallback_type; | |
1182 | } | |
1183 | ||
b2a0ac88 | 1184 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 1185 | static inline struct page * |
7aeb09f9 | 1186 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 1187 | { |
b8af2941 | 1188 | struct free_area *area; |
7aeb09f9 | 1189 | unsigned int current_order; |
b2a0ac88 | 1190 | struct page *page; |
fef903ef | 1191 | int migratetype, new_type, i; |
b2a0ac88 MG |
1192 | |
1193 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
1194 | for (current_order = MAX_ORDER-1; |
1195 | current_order >= order && current_order <= MAX_ORDER-1; | |
1196 | --current_order) { | |
6d4a4916 | 1197 | for (i = 0;; i++) { |
b2a0ac88 MG |
1198 | migratetype = fallbacks[start_migratetype][i]; |
1199 | ||
56fd56b8 MG |
1200 | /* MIGRATE_RESERVE handled later if necessary */ |
1201 | if (migratetype == MIGRATE_RESERVE) | |
6d4a4916 | 1202 | break; |
e010487d | 1203 | |
b2a0ac88 MG |
1204 | area = &(zone->free_area[current_order]); |
1205 | if (list_empty(&area->free_list[migratetype])) | |
1206 | continue; | |
1207 | ||
1208 | page = list_entry(area->free_list[migratetype].next, | |
1209 | struct page, lru); | |
1210 | area->nr_free--; | |
1211 | ||
fef903ef SB |
1212 | new_type = try_to_steal_freepages(zone, page, |
1213 | start_migratetype, | |
1214 | migratetype); | |
b2a0ac88 MG |
1215 | |
1216 | /* Remove the page from the freelists */ | |
1217 | list_del(&page->lru); | |
1218 | rmv_page_order(page); | |
b2a0ac88 | 1219 | |
47118af0 | 1220 | expand(zone, page, order, current_order, area, |
0cbef29a | 1221 | new_type); |
5bcc9f86 VB |
1222 | /* The freepage_migratetype may differ from pageblock's |
1223 | * migratetype depending on the decisions in | |
1224 | * try_to_steal_freepages. This is OK as long as it does | |
1225 | * not differ for MIGRATE_CMA type. | |
1226 | */ | |
1227 | set_freepage_migratetype(page, new_type); | |
e0fff1bd | 1228 | |
52c8f6a5 KM |
1229 | trace_mm_page_alloc_extfrag(page, order, current_order, |
1230 | start_migratetype, migratetype, new_type); | |
e0fff1bd | 1231 | |
b2a0ac88 MG |
1232 | return page; |
1233 | } | |
1234 | } | |
1235 | ||
728ec980 | 1236 | return NULL; |
b2a0ac88 MG |
1237 | } |
1238 | ||
56fd56b8 | 1239 | /* |
1da177e4 LT |
1240 | * Do the hard work of removing an element from the buddy allocator. |
1241 | * Call me with the zone->lock already held. | |
1242 | */ | |
b2a0ac88 MG |
1243 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1244 | int migratetype) | |
1da177e4 | 1245 | { |
1da177e4 LT |
1246 | struct page *page; |
1247 | ||
728ec980 | 1248 | retry_reserve: |
56fd56b8 | 1249 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1250 | |
728ec980 | 1251 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 1252 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 1253 | |
728ec980 MG |
1254 | /* |
1255 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1256 | * is used because __rmqueue_smallest is an inline function | |
1257 | * and we want just one call site | |
1258 | */ | |
1259 | if (!page) { | |
1260 | migratetype = MIGRATE_RESERVE; | |
1261 | goto retry_reserve; | |
1262 | } | |
1263 | } | |
1264 | ||
0d3d062a | 1265 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1266 | return page; |
1da177e4 LT |
1267 | } |
1268 | ||
5f63b720 | 1269 | /* |
1da177e4 LT |
1270 | * Obtain a specified number of elements from the buddy allocator, all under |
1271 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1272 | * Returns the number of new pages which were placed at *list. | |
1273 | */ | |
5f63b720 | 1274 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1275 | unsigned long count, struct list_head *list, |
b745bc85 | 1276 | int migratetype, bool cold) |
1da177e4 | 1277 | { |
5bcc9f86 | 1278 | int i; |
5f63b720 | 1279 | |
c54ad30c | 1280 | spin_lock(&zone->lock); |
1da177e4 | 1281 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1282 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1283 | if (unlikely(page == NULL)) |
1da177e4 | 1284 | break; |
81eabcbe MG |
1285 | |
1286 | /* | |
1287 | * Split buddy pages returned by expand() are received here | |
1288 | * in physical page order. The page is added to the callers and | |
1289 | * list and the list head then moves forward. From the callers | |
1290 | * perspective, the linked list is ordered by page number in | |
1291 | * some conditions. This is useful for IO devices that can | |
1292 | * merge IO requests if the physical pages are ordered | |
1293 | * properly. | |
1294 | */ | |
b745bc85 | 1295 | if (likely(!cold)) |
e084b2d9 MG |
1296 | list_add(&page->lru, list); |
1297 | else | |
1298 | list_add_tail(&page->lru, list); | |
81eabcbe | 1299 | list = &page->lru; |
5bcc9f86 | 1300 | if (is_migrate_cma(get_freepage_migratetype(page))) |
d1ce749a BZ |
1301 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
1302 | -(1 << order)); | |
1da177e4 | 1303 | } |
f2260e6b | 1304 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1305 | spin_unlock(&zone->lock); |
085cc7d5 | 1306 | return i; |
1da177e4 LT |
1307 | } |
1308 | ||
4ae7c039 | 1309 | #ifdef CONFIG_NUMA |
8fce4d8e | 1310 | /* |
4037d452 CL |
1311 | * Called from the vmstat counter updater to drain pagesets of this |
1312 | * currently executing processor on remote nodes after they have | |
1313 | * expired. | |
1314 | * | |
879336c3 CL |
1315 | * Note that this function must be called with the thread pinned to |
1316 | * a single processor. | |
8fce4d8e | 1317 | */ |
4037d452 | 1318 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1319 | { |
4ae7c039 | 1320 | unsigned long flags; |
7be12fc9 | 1321 | int to_drain, batch; |
4ae7c039 | 1322 | |
4037d452 | 1323 | local_irq_save(flags); |
998d39cb | 1324 | batch = ACCESS_ONCE(pcp->batch); |
7be12fc9 | 1325 | to_drain = min(pcp->count, batch); |
2a13515c KM |
1326 | if (to_drain > 0) { |
1327 | free_pcppages_bulk(zone, to_drain, pcp); | |
1328 | pcp->count -= to_drain; | |
1329 | } | |
4037d452 | 1330 | local_irq_restore(flags); |
4ae7c039 CL |
1331 | } |
1332 | #endif | |
1333 | ||
9f8f2172 | 1334 | /* |
93481ff0 | 1335 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
1336 | * |
1337 | * The processor must either be the current processor and the | |
1338 | * thread pinned to the current processor or a processor that | |
1339 | * is not online. | |
1340 | */ | |
93481ff0 | 1341 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 1342 | { |
c54ad30c | 1343 | unsigned long flags; |
93481ff0 VB |
1344 | struct per_cpu_pageset *pset; |
1345 | struct per_cpu_pages *pcp; | |
1da177e4 | 1346 | |
93481ff0 VB |
1347 | local_irq_save(flags); |
1348 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 1349 | |
93481ff0 VB |
1350 | pcp = &pset->pcp; |
1351 | if (pcp->count) { | |
1352 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1353 | pcp->count = 0; | |
1354 | } | |
1355 | local_irq_restore(flags); | |
1356 | } | |
3dfa5721 | 1357 | |
93481ff0 VB |
1358 | /* |
1359 | * Drain pcplists of all zones on the indicated processor. | |
1360 | * | |
1361 | * The processor must either be the current processor and the | |
1362 | * thread pinned to the current processor or a processor that | |
1363 | * is not online. | |
1364 | */ | |
1365 | static void drain_pages(unsigned int cpu) | |
1366 | { | |
1367 | struct zone *zone; | |
1368 | ||
1369 | for_each_populated_zone(zone) { | |
1370 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
1371 | } |
1372 | } | |
1da177e4 | 1373 | |
9f8f2172 CL |
1374 | /* |
1375 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
1376 | * |
1377 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
1378 | * the single zone's pages. | |
9f8f2172 | 1379 | */ |
93481ff0 | 1380 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 1381 | { |
93481ff0 VB |
1382 | int cpu = smp_processor_id(); |
1383 | ||
1384 | if (zone) | |
1385 | drain_pages_zone(cpu, zone); | |
1386 | else | |
1387 | drain_pages(cpu); | |
9f8f2172 CL |
1388 | } |
1389 | ||
1390 | /* | |
74046494 GBY |
1391 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
1392 | * | |
93481ff0 VB |
1393 | * When zone parameter is non-NULL, spill just the single zone's pages. |
1394 | * | |
74046494 GBY |
1395 | * Note that this code is protected against sending an IPI to an offline |
1396 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
1397 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
1398 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
1399 | * before the call to on_each_cpu_mask(). | |
9f8f2172 | 1400 | */ |
93481ff0 | 1401 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 1402 | { |
74046494 | 1403 | int cpu; |
74046494 GBY |
1404 | |
1405 | /* | |
1406 | * Allocate in the BSS so we wont require allocation in | |
1407 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
1408 | */ | |
1409 | static cpumask_t cpus_with_pcps; | |
1410 | ||
1411 | /* | |
1412 | * We don't care about racing with CPU hotplug event | |
1413 | * as offline notification will cause the notified | |
1414 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
1415 | * disables preemption as part of its processing | |
1416 | */ | |
1417 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
1418 | struct per_cpu_pageset *pcp; |
1419 | struct zone *z; | |
74046494 | 1420 | bool has_pcps = false; |
93481ff0 VB |
1421 | |
1422 | if (zone) { | |
74046494 | 1423 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 1424 | if (pcp->pcp.count) |
74046494 | 1425 | has_pcps = true; |
93481ff0 VB |
1426 | } else { |
1427 | for_each_populated_zone(z) { | |
1428 | pcp = per_cpu_ptr(z->pageset, cpu); | |
1429 | if (pcp->pcp.count) { | |
1430 | has_pcps = true; | |
1431 | break; | |
1432 | } | |
74046494 GBY |
1433 | } |
1434 | } | |
93481ff0 | 1435 | |
74046494 GBY |
1436 | if (has_pcps) |
1437 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
1438 | else | |
1439 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
1440 | } | |
93481ff0 VB |
1441 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
1442 | zone, 1); | |
9f8f2172 CL |
1443 | } |
1444 | ||
296699de | 1445 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1446 | |
1447 | void mark_free_pages(struct zone *zone) | |
1448 | { | |
f623f0db RW |
1449 | unsigned long pfn, max_zone_pfn; |
1450 | unsigned long flags; | |
7aeb09f9 | 1451 | unsigned int order, t; |
1da177e4 LT |
1452 | struct list_head *curr; |
1453 | ||
8080fc03 | 1454 | if (zone_is_empty(zone)) |
1da177e4 LT |
1455 | return; |
1456 | ||
1457 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 1458 | |
108bcc96 | 1459 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
1460 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
1461 | if (pfn_valid(pfn)) { | |
1462 | struct page *page = pfn_to_page(pfn); | |
1463 | ||
7be98234 RW |
1464 | if (!swsusp_page_is_forbidden(page)) |
1465 | swsusp_unset_page_free(page); | |
f623f0db | 1466 | } |
1da177e4 | 1467 | |
b2a0ac88 MG |
1468 | for_each_migratetype_order(order, t) { |
1469 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1470 | unsigned long i; |
1da177e4 | 1471 | |
f623f0db RW |
1472 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1473 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1474 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1475 | } |
b2a0ac88 | 1476 | } |
1da177e4 LT |
1477 | spin_unlock_irqrestore(&zone->lock, flags); |
1478 | } | |
e2c55dc8 | 1479 | #endif /* CONFIG_PM */ |
1da177e4 | 1480 | |
1da177e4 LT |
1481 | /* |
1482 | * Free a 0-order page | |
b745bc85 | 1483 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 1484 | */ |
b745bc85 | 1485 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
1486 | { |
1487 | struct zone *zone = page_zone(page); | |
1488 | struct per_cpu_pages *pcp; | |
1489 | unsigned long flags; | |
dc4b0caf | 1490 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 1491 | int migratetype; |
1da177e4 | 1492 | |
ec95f53a | 1493 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1494 | return; |
1495 | ||
dc4b0caf | 1496 | migratetype = get_pfnblock_migratetype(page, pfn); |
b12c4ad1 | 1497 | set_freepage_migratetype(page, migratetype); |
1da177e4 | 1498 | local_irq_save(flags); |
f8891e5e | 1499 | __count_vm_event(PGFREE); |
da456f14 | 1500 | |
5f8dcc21 MG |
1501 | /* |
1502 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1503 | * Free ISOLATE pages back to the allocator because they are being | |
1504 | * offlined but treat RESERVE as movable pages so we can get those | |
1505 | * areas back if necessary. Otherwise, we may have to free | |
1506 | * excessively into the page allocator | |
1507 | */ | |
1508 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 1509 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 1510 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
1511 | goto out; |
1512 | } | |
1513 | migratetype = MIGRATE_MOVABLE; | |
1514 | } | |
1515 | ||
99dcc3e5 | 1516 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 1517 | if (!cold) |
5f8dcc21 | 1518 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
1519 | else |
1520 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 1521 | pcp->count++; |
48db57f8 | 1522 | if (pcp->count >= pcp->high) { |
998d39cb CS |
1523 | unsigned long batch = ACCESS_ONCE(pcp->batch); |
1524 | free_pcppages_bulk(zone, batch, pcp); | |
1525 | pcp->count -= batch; | |
48db57f8 | 1526 | } |
5f8dcc21 MG |
1527 | |
1528 | out: | |
1da177e4 | 1529 | local_irq_restore(flags); |
1da177e4 LT |
1530 | } |
1531 | ||
cc59850e KK |
1532 | /* |
1533 | * Free a list of 0-order pages | |
1534 | */ | |
b745bc85 | 1535 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
1536 | { |
1537 | struct page *page, *next; | |
1538 | ||
1539 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1540 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1541 | free_hot_cold_page(page, cold); |
1542 | } | |
1543 | } | |
1544 | ||
8dfcc9ba NP |
1545 | /* |
1546 | * split_page takes a non-compound higher-order page, and splits it into | |
1547 | * n (1<<order) sub-pages: page[0..n] | |
1548 | * Each sub-page must be freed individually. | |
1549 | * | |
1550 | * Note: this is probably too low level an operation for use in drivers. | |
1551 | * Please consult with lkml before using this in your driver. | |
1552 | */ | |
1553 | void split_page(struct page *page, unsigned int order) | |
1554 | { | |
1555 | int i; | |
1556 | ||
309381fe SL |
1557 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1558 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
1559 | |
1560 | #ifdef CONFIG_KMEMCHECK | |
1561 | /* | |
1562 | * Split shadow pages too, because free(page[0]) would | |
1563 | * otherwise free the whole shadow. | |
1564 | */ | |
1565 | if (kmemcheck_page_is_tracked(page)) | |
1566 | split_page(virt_to_page(page[0].shadow), order); | |
1567 | #endif | |
1568 | ||
48c96a36 JK |
1569 | set_page_owner(page, 0, 0); |
1570 | for (i = 1; i < (1 << order); i++) { | |
7835e98b | 1571 | set_page_refcounted(page + i); |
48c96a36 JK |
1572 | set_page_owner(page + i, 0, 0); |
1573 | } | |
8dfcc9ba | 1574 | } |
5853ff23 | 1575 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 1576 | |
3c605096 | 1577 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 1578 | { |
748446bb MG |
1579 | unsigned long watermark; |
1580 | struct zone *zone; | |
2139cbe6 | 1581 | int mt; |
748446bb MG |
1582 | |
1583 | BUG_ON(!PageBuddy(page)); | |
1584 | ||
1585 | zone = page_zone(page); | |
2e30abd1 | 1586 | mt = get_pageblock_migratetype(page); |
748446bb | 1587 | |
194159fb | 1588 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
1589 | /* Obey watermarks as if the page was being allocated */ |
1590 | watermark = low_wmark_pages(zone) + (1 << order); | |
1591 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1592 | return 0; | |
1593 | ||
8fb74b9f | 1594 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 1595 | } |
748446bb MG |
1596 | |
1597 | /* Remove page from free list */ | |
1598 | list_del(&page->lru); | |
1599 | zone->free_area[order].nr_free--; | |
1600 | rmv_page_order(page); | |
2139cbe6 | 1601 | |
8fb74b9f | 1602 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
1603 | if (order >= pageblock_order - 1) { |
1604 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
1605 | for (; page < endpage; page += pageblock_nr_pages) { |
1606 | int mt = get_pageblock_migratetype(page); | |
194159fb | 1607 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
1608 | set_pageblock_migratetype(page, |
1609 | MIGRATE_MOVABLE); | |
1610 | } | |
748446bb MG |
1611 | } |
1612 | ||
48c96a36 | 1613 | set_page_owner(page, order, 0); |
8fb74b9f | 1614 | return 1UL << order; |
1fb3f8ca MG |
1615 | } |
1616 | ||
1617 | /* | |
1618 | * Similar to split_page except the page is already free. As this is only | |
1619 | * being used for migration, the migratetype of the block also changes. | |
1620 | * As this is called with interrupts disabled, the caller is responsible | |
1621 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1622 | * are enabled. | |
1623 | * | |
1624 | * Note: this is probably too low level an operation for use in drivers. | |
1625 | * Please consult with lkml before using this in your driver. | |
1626 | */ | |
1627 | int split_free_page(struct page *page) | |
1628 | { | |
1629 | unsigned int order; | |
1630 | int nr_pages; | |
1631 | ||
1fb3f8ca MG |
1632 | order = page_order(page); |
1633 | ||
8fb74b9f | 1634 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
1635 | if (!nr_pages) |
1636 | return 0; | |
1637 | ||
1638 | /* Split into individual pages */ | |
1639 | set_page_refcounted(page); | |
1640 | split_page(page, order); | |
1641 | return nr_pages; | |
748446bb MG |
1642 | } |
1643 | ||
1da177e4 LT |
1644 | /* |
1645 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1646 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1647 | * or two. | |
1648 | */ | |
0a15c3e9 MG |
1649 | static inline |
1650 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
7aeb09f9 MG |
1651 | struct zone *zone, unsigned int order, |
1652 | gfp_t gfp_flags, int migratetype) | |
1da177e4 LT |
1653 | { |
1654 | unsigned long flags; | |
689bcebf | 1655 | struct page *page; |
b745bc85 | 1656 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
1da177e4 | 1657 | |
689bcebf | 1658 | again: |
48db57f8 | 1659 | if (likely(order == 0)) { |
1da177e4 | 1660 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1661 | struct list_head *list; |
1da177e4 | 1662 | |
1da177e4 | 1663 | local_irq_save(flags); |
99dcc3e5 CL |
1664 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1665 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1666 | if (list_empty(list)) { |
535131e6 | 1667 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1668 | pcp->batch, list, |
e084b2d9 | 1669 | migratetype, cold); |
5f8dcc21 | 1670 | if (unlikely(list_empty(list))) |
6fb332fa | 1671 | goto failed; |
535131e6 | 1672 | } |
b92a6edd | 1673 | |
5f8dcc21 MG |
1674 | if (cold) |
1675 | page = list_entry(list->prev, struct page, lru); | |
1676 | else | |
1677 | page = list_entry(list->next, struct page, lru); | |
1678 | ||
b92a6edd MG |
1679 | list_del(&page->lru); |
1680 | pcp->count--; | |
7fb1d9fc | 1681 | } else { |
dab48dab AM |
1682 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1683 | /* | |
1684 | * __GFP_NOFAIL is not to be used in new code. | |
1685 | * | |
1686 | * All __GFP_NOFAIL callers should be fixed so that they | |
1687 | * properly detect and handle allocation failures. | |
1688 | * | |
1689 | * We most definitely don't want callers attempting to | |
4923abf9 | 1690 | * allocate greater than order-1 page units with |
dab48dab AM |
1691 | * __GFP_NOFAIL. |
1692 | */ | |
4923abf9 | 1693 | WARN_ON_ONCE(order > 1); |
dab48dab | 1694 | } |
1da177e4 | 1695 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1696 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1697 | spin_unlock(&zone->lock); |
1698 | if (!page) | |
1699 | goto failed; | |
d1ce749a | 1700 | __mod_zone_freepage_state(zone, -(1 << order), |
5bcc9f86 | 1701 | get_freepage_migratetype(page)); |
1da177e4 LT |
1702 | } |
1703 | ||
3a025760 | 1704 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
abe5f972 | 1705 | if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 && |
57054651 JW |
1706 | !test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) |
1707 | set_bit(ZONE_FAIR_DEPLETED, &zone->flags); | |
27329369 | 1708 | |
f8891e5e | 1709 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1710 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1711 | local_irq_restore(flags); |
1da177e4 | 1712 | |
309381fe | 1713 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
17cf4406 | 1714 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1715 | goto again; |
1da177e4 | 1716 | return page; |
a74609fa NP |
1717 | |
1718 | failed: | |
1719 | local_irq_restore(flags); | |
a74609fa | 1720 | return NULL; |
1da177e4 LT |
1721 | } |
1722 | ||
933e312e AM |
1723 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1724 | ||
b2588c4b | 1725 | static struct { |
933e312e AM |
1726 | struct fault_attr attr; |
1727 | ||
1728 | u32 ignore_gfp_highmem; | |
1729 | u32 ignore_gfp_wait; | |
54114994 | 1730 | u32 min_order; |
933e312e AM |
1731 | } fail_page_alloc = { |
1732 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1733 | .ignore_gfp_wait = 1, |
1734 | .ignore_gfp_highmem = 1, | |
54114994 | 1735 | .min_order = 1, |
933e312e AM |
1736 | }; |
1737 | ||
1738 | static int __init setup_fail_page_alloc(char *str) | |
1739 | { | |
1740 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1741 | } | |
1742 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1743 | ||
deaf386e | 1744 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1745 | { |
54114994 | 1746 | if (order < fail_page_alloc.min_order) |
deaf386e | 1747 | return false; |
933e312e | 1748 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 1749 | return false; |
933e312e | 1750 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 1751 | return false; |
933e312e | 1752 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
deaf386e | 1753 | return false; |
933e312e AM |
1754 | |
1755 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1756 | } | |
1757 | ||
1758 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1759 | ||
1760 | static int __init fail_page_alloc_debugfs(void) | |
1761 | { | |
f4ae40a6 | 1762 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1763 | struct dentry *dir; |
933e312e | 1764 | |
dd48c085 AM |
1765 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1766 | &fail_page_alloc.attr); | |
1767 | if (IS_ERR(dir)) | |
1768 | return PTR_ERR(dir); | |
933e312e | 1769 | |
b2588c4b AM |
1770 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1771 | &fail_page_alloc.ignore_gfp_wait)) | |
1772 | goto fail; | |
1773 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1774 | &fail_page_alloc.ignore_gfp_highmem)) | |
1775 | goto fail; | |
1776 | if (!debugfs_create_u32("min-order", mode, dir, | |
1777 | &fail_page_alloc.min_order)) | |
1778 | goto fail; | |
1779 | ||
1780 | return 0; | |
1781 | fail: | |
dd48c085 | 1782 | debugfs_remove_recursive(dir); |
933e312e | 1783 | |
b2588c4b | 1784 | return -ENOMEM; |
933e312e AM |
1785 | } |
1786 | ||
1787 | late_initcall(fail_page_alloc_debugfs); | |
1788 | ||
1789 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1790 | ||
1791 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1792 | ||
deaf386e | 1793 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1794 | { |
deaf386e | 1795 | return false; |
933e312e AM |
1796 | } |
1797 | ||
1798 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1799 | ||
1da177e4 | 1800 | /* |
88f5acf8 | 1801 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1802 | * of the allocation. |
1803 | */ | |
7aeb09f9 MG |
1804 | static bool __zone_watermark_ok(struct zone *z, unsigned int order, |
1805 | unsigned long mark, int classzone_idx, int alloc_flags, | |
1806 | long free_pages) | |
1da177e4 | 1807 | { |
26086de3 | 1808 | /* free_pages may go negative - that's OK */ |
d23ad423 | 1809 | long min = mark; |
1da177e4 | 1810 | int o; |
026b0814 | 1811 | long free_cma = 0; |
1da177e4 | 1812 | |
df0a6daa | 1813 | free_pages -= (1 << order) - 1; |
7fb1d9fc | 1814 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1815 | min -= min / 2; |
7fb1d9fc | 1816 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 | 1817 | min -= min / 4; |
d95ea5d1 BZ |
1818 | #ifdef CONFIG_CMA |
1819 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
1820 | if (!(alloc_flags & ALLOC_CMA)) | |
026b0814 | 1821 | free_cma = zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 1822 | #endif |
026b0814 | 1823 | |
3484b2de | 1824 | if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx]) |
88f5acf8 | 1825 | return false; |
1da177e4 LT |
1826 | for (o = 0; o < order; o++) { |
1827 | /* At the next order, this order's pages become unavailable */ | |
1828 | free_pages -= z->free_area[o].nr_free << o; | |
1829 | ||
1830 | /* Require fewer higher order pages to be free */ | |
1831 | min >>= 1; | |
1832 | ||
1833 | if (free_pages <= min) | |
88f5acf8 | 1834 | return false; |
1da177e4 | 1835 | } |
88f5acf8 MG |
1836 | return true; |
1837 | } | |
1838 | ||
7aeb09f9 | 1839 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
88f5acf8 MG |
1840 | int classzone_idx, int alloc_flags) |
1841 | { | |
1842 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1843 | zone_page_state(z, NR_FREE_PAGES)); | |
1844 | } | |
1845 | ||
7aeb09f9 MG |
1846 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
1847 | unsigned long mark, int classzone_idx, int alloc_flags) | |
88f5acf8 MG |
1848 | { |
1849 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1850 | ||
1851 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1852 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1853 | ||
1854 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1855 | free_pages); | |
1da177e4 LT |
1856 | } |
1857 | ||
9276b1bc PJ |
1858 | #ifdef CONFIG_NUMA |
1859 | /* | |
1860 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1861 | * skip over zones that are not allowed by the cpuset, or that have | |
1862 | * been recently (in last second) found to be nearly full. See further | |
1863 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1864 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc | 1865 | * |
a1aeb65a | 1866 | * If the zonelist cache is present in the passed zonelist, then |
9276b1bc | 1867 | * returns a pointer to the allowed node mask (either the current |
4b0ef1fe | 1868 | * tasks mems_allowed, or node_states[N_MEMORY].) |
9276b1bc PJ |
1869 | * |
1870 | * If the zonelist cache is not available for this zonelist, does | |
1871 | * nothing and returns NULL. | |
1872 | * | |
1873 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1874 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1875 | * | |
1876 | * We hold off even calling zlc_setup, until after we've checked the | |
1877 | * first zone in the zonelist, on the theory that most allocations will | |
1878 | * be satisfied from that first zone, so best to examine that zone as | |
1879 | * quickly as we can. | |
1880 | */ | |
1881 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1882 | { | |
1883 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1884 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1885 | ||
1886 | zlc = zonelist->zlcache_ptr; | |
1887 | if (!zlc) | |
1888 | return NULL; | |
1889 | ||
f05111f5 | 1890 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1891 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1892 | zlc->last_full_zap = jiffies; | |
1893 | } | |
1894 | ||
1895 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1896 | &cpuset_current_mems_allowed : | |
4b0ef1fe | 1897 | &node_states[N_MEMORY]; |
9276b1bc PJ |
1898 | return allowednodes; |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1903 | * if it is worth looking at further for free memory: | |
1904 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1905 | * bit set in the zonelist_cache fullzones BITMAP). | |
1906 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1907 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1908 | * Return true (non-zero) if zone is worth looking at further, or | |
1909 | * else return false (zero) if it is not. | |
1910 | * | |
1911 | * This check -ignores- the distinction between various watermarks, | |
1912 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1913 | * found to be full for any variation of these watermarks, it will | |
1914 | * be considered full for up to one second by all requests, unless | |
1915 | * we are so low on memory on all allowed nodes that we are forced | |
1916 | * into the second scan of the zonelist. | |
1917 | * | |
1918 | * In the second scan we ignore this zonelist cache and exactly | |
1919 | * apply the watermarks to all zones, even it is slower to do so. | |
1920 | * We are low on memory in the second scan, and should leave no stone | |
1921 | * unturned looking for a free page. | |
1922 | */ | |
dd1a239f | 1923 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1924 | nodemask_t *allowednodes) |
1925 | { | |
1926 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1927 | int i; /* index of *z in zonelist zones */ | |
1928 | int n; /* node that zone *z is on */ | |
1929 | ||
1930 | zlc = zonelist->zlcache_ptr; | |
1931 | if (!zlc) | |
1932 | return 1; | |
1933 | ||
dd1a239f | 1934 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1935 | n = zlc->z_to_n[i]; |
1936 | ||
1937 | /* This zone is worth trying if it is allowed but not full */ | |
1938 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1939 | } | |
1940 | ||
1941 | /* | |
1942 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1943 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1944 | * from that zone don't waste time re-examining it. | |
1945 | */ | |
dd1a239f | 1946 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1947 | { |
1948 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1949 | int i; /* index of *z in zonelist zones */ | |
1950 | ||
1951 | zlc = zonelist->zlcache_ptr; | |
1952 | if (!zlc) | |
1953 | return; | |
1954 | ||
dd1a239f | 1955 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1956 | |
1957 | set_bit(i, zlc->fullzones); | |
1958 | } | |
1959 | ||
76d3fbf8 MG |
1960 | /* |
1961 | * clear all zones full, called after direct reclaim makes progress so that | |
1962 | * a zone that was recently full is not skipped over for up to a second | |
1963 | */ | |
1964 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1965 | { | |
1966 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1967 | ||
1968 | zlc = zonelist->zlcache_ptr; | |
1969 | if (!zlc) | |
1970 | return; | |
1971 | ||
1972 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1973 | } | |
1974 | ||
81c0a2bb JW |
1975 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
1976 | { | |
fff4068c | 1977 | return local_zone->node == zone->node; |
81c0a2bb JW |
1978 | } |
1979 | ||
957f822a DR |
1980 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1981 | { | |
5f7a75ac MG |
1982 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
1983 | RECLAIM_DISTANCE; | |
957f822a DR |
1984 | } |
1985 | ||
9276b1bc PJ |
1986 | #else /* CONFIG_NUMA */ |
1987 | ||
1988 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1989 | { | |
1990 | return NULL; | |
1991 | } | |
1992 | ||
dd1a239f | 1993 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1994 | nodemask_t *allowednodes) |
1995 | { | |
1996 | return 1; | |
1997 | } | |
1998 | ||
dd1a239f | 1999 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
2000 | { |
2001 | } | |
76d3fbf8 MG |
2002 | |
2003 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
2004 | { | |
2005 | } | |
957f822a | 2006 | |
81c0a2bb JW |
2007 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2008 | { | |
2009 | return true; | |
2010 | } | |
2011 | ||
957f822a DR |
2012 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2013 | { | |
2014 | return true; | |
2015 | } | |
2016 | ||
9276b1bc PJ |
2017 | #endif /* CONFIG_NUMA */ |
2018 | ||
4ffeaf35 MG |
2019 | static void reset_alloc_batches(struct zone *preferred_zone) |
2020 | { | |
2021 | struct zone *zone = preferred_zone->zone_pgdat->node_zones; | |
2022 | ||
2023 | do { | |
2024 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
2025 | high_wmark_pages(zone) - low_wmark_pages(zone) - | |
2026 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
57054651 | 2027 | clear_bit(ZONE_FAIR_DEPLETED, &zone->flags); |
4ffeaf35 MG |
2028 | } while (zone++ != preferred_zone); |
2029 | } | |
2030 | ||
7fb1d9fc | 2031 | /* |
0798e519 | 2032 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
2033 | * a page. |
2034 | */ | |
2035 | static struct page * | |
19770b32 | 2036 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 2037 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
d8846374 | 2038 | struct zone *preferred_zone, int classzone_idx, int migratetype) |
753ee728 | 2039 | { |
dd1a239f | 2040 | struct zoneref *z; |
7fb1d9fc | 2041 | struct page *page = NULL; |
5117f45d | 2042 | struct zone *zone; |
9276b1bc PJ |
2043 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
2044 | int zlc_active = 0; /* set if using zonelist_cache */ | |
2045 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
a6e21b14 MG |
2046 | bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) && |
2047 | (gfp_mask & __GFP_WRITE); | |
4ffeaf35 MG |
2048 | int nr_fair_skipped = 0; |
2049 | bool zonelist_rescan; | |
54a6eb5c | 2050 | |
9276b1bc | 2051 | zonelist_scan: |
4ffeaf35 MG |
2052 | zonelist_rescan = false; |
2053 | ||
7fb1d9fc | 2054 | /* |
9276b1bc | 2055 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 2056 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 2057 | */ |
19770b32 MG |
2058 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
2059 | high_zoneidx, nodemask) { | |
e085dbc5 JW |
2060 | unsigned long mark; |
2061 | ||
e5adfffc | 2062 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
9276b1bc PJ |
2063 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
2064 | continue; | |
664eedde MG |
2065 | if (cpusets_enabled() && |
2066 | (alloc_flags & ALLOC_CPUSET) && | |
344736f2 | 2067 | !cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 2068 | continue; |
81c0a2bb JW |
2069 | /* |
2070 | * Distribute pages in proportion to the individual | |
2071 | * zone size to ensure fair page aging. The zone a | |
2072 | * page was allocated in should have no effect on the | |
2073 | * time the page has in memory before being reclaimed. | |
81c0a2bb | 2074 | */ |
3a025760 | 2075 | if (alloc_flags & ALLOC_FAIR) { |
fff4068c | 2076 | if (!zone_local(preferred_zone, zone)) |
f7b5d647 | 2077 | break; |
57054651 | 2078 | if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) { |
4ffeaf35 | 2079 | nr_fair_skipped++; |
3a025760 | 2080 | continue; |
4ffeaf35 | 2081 | } |
81c0a2bb | 2082 | } |
a756cf59 JW |
2083 | /* |
2084 | * When allocating a page cache page for writing, we | |
2085 | * want to get it from a zone that is within its dirty | |
2086 | * limit, such that no single zone holds more than its | |
2087 | * proportional share of globally allowed dirty pages. | |
2088 | * The dirty limits take into account the zone's | |
2089 | * lowmem reserves and high watermark so that kswapd | |
2090 | * should be able to balance it without having to | |
2091 | * write pages from its LRU list. | |
2092 | * | |
2093 | * This may look like it could increase pressure on | |
2094 | * lower zones by failing allocations in higher zones | |
2095 | * before they are full. But the pages that do spill | |
2096 | * over are limited as the lower zones are protected | |
2097 | * by this very same mechanism. It should not become | |
2098 | * a practical burden to them. | |
2099 | * | |
2100 | * XXX: For now, allow allocations to potentially | |
2101 | * exceed the per-zone dirty limit in the slowpath | |
2102 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
2103 | * which is important when on a NUMA setup the allowed | |
2104 | * zones are together not big enough to reach the | |
2105 | * global limit. The proper fix for these situations | |
2106 | * will require awareness of zones in the | |
2107 | * dirty-throttling and the flusher threads. | |
2108 | */ | |
a6e21b14 | 2109 | if (consider_zone_dirty && !zone_dirty_ok(zone)) |
800a1e75 | 2110 | continue; |
7fb1d9fc | 2111 | |
e085dbc5 JW |
2112 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
2113 | if (!zone_watermark_ok(zone, order, mark, | |
2114 | classzone_idx, alloc_flags)) { | |
fa5e084e MG |
2115 | int ret; |
2116 | ||
5dab2911 MG |
2117 | /* Checked here to keep the fast path fast */ |
2118 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
2119 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
2120 | goto try_this_zone; | |
2121 | ||
e5adfffc KS |
2122 | if (IS_ENABLED(CONFIG_NUMA) && |
2123 | !did_zlc_setup && nr_online_nodes > 1) { | |
cd38b115 MG |
2124 | /* |
2125 | * we do zlc_setup if there are multiple nodes | |
2126 | * and before considering the first zone allowed | |
2127 | * by the cpuset. | |
2128 | */ | |
2129 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
2130 | zlc_active = 1; | |
2131 | did_zlc_setup = 1; | |
2132 | } | |
2133 | ||
957f822a DR |
2134 | if (zone_reclaim_mode == 0 || |
2135 | !zone_allows_reclaim(preferred_zone, zone)) | |
fa5e084e MG |
2136 | goto this_zone_full; |
2137 | ||
cd38b115 MG |
2138 | /* |
2139 | * As we may have just activated ZLC, check if the first | |
2140 | * eligible zone has failed zone_reclaim recently. | |
2141 | */ | |
e5adfffc | 2142 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
cd38b115 MG |
2143 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
2144 | continue; | |
2145 | ||
fa5e084e MG |
2146 | ret = zone_reclaim(zone, gfp_mask, order); |
2147 | switch (ret) { | |
2148 | case ZONE_RECLAIM_NOSCAN: | |
2149 | /* did not scan */ | |
cd38b115 | 2150 | continue; |
fa5e084e MG |
2151 | case ZONE_RECLAIM_FULL: |
2152 | /* scanned but unreclaimable */ | |
cd38b115 | 2153 | continue; |
fa5e084e MG |
2154 | default: |
2155 | /* did we reclaim enough */ | |
fed2719e | 2156 | if (zone_watermark_ok(zone, order, mark, |
fa5e084e | 2157 | classzone_idx, alloc_flags)) |
fed2719e MG |
2158 | goto try_this_zone; |
2159 | ||
2160 | /* | |
2161 | * Failed to reclaim enough to meet watermark. | |
2162 | * Only mark the zone full if checking the min | |
2163 | * watermark or if we failed to reclaim just | |
2164 | * 1<<order pages or else the page allocator | |
2165 | * fastpath will prematurely mark zones full | |
2166 | * when the watermark is between the low and | |
2167 | * min watermarks. | |
2168 | */ | |
2169 | if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) || | |
2170 | ret == ZONE_RECLAIM_SOME) | |
9276b1bc | 2171 | goto this_zone_full; |
fed2719e MG |
2172 | |
2173 | continue; | |
0798e519 | 2174 | } |
7fb1d9fc RS |
2175 | } |
2176 | ||
fa5e084e | 2177 | try_this_zone: |
3dd28266 MG |
2178 | page = buffered_rmqueue(preferred_zone, zone, order, |
2179 | gfp_mask, migratetype); | |
0798e519 | 2180 | if (page) |
7fb1d9fc | 2181 | break; |
9276b1bc | 2182 | this_zone_full: |
65bb3719 | 2183 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active) |
9276b1bc | 2184 | zlc_mark_zone_full(zonelist, z); |
54a6eb5c | 2185 | } |
9276b1bc | 2186 | |
4ffeaf35 | 2187 | if (page) { |
b121186a AS |
2188 | /* |
2189 | * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was | |
2190 | * necessary to allocate the page. The expectation is | |
2191 | * that the caller is taking steps that will free more | |
2192 | * memory. The caller should avoid the page being used | |
2193 | * for !PFMEMALLOC purposes. | |
2194 | */ | |
2195 | page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS); | |
4ffeaf35 MG |
2196 | return page; |
2197 | } | |
b121186a | 2198 | |
4ffeaf35 MG |
2199 | /* |
2200 | * The first pass makes sure allocations are spread fairly within the | |
2201 | * local node. However, the local node might have free pages left | |
2202 | * after the fairness batches are exhausted, and remote zones haven't | |
2203 | * even been considered yet. Try once more without fairness, and | |
2204 | * include remote zones now, before entering the slowpath and waking | |
2205 | * kswapd: prefer spilling to a remote zone over swapping locally. | |
2206 | */ | |
2207 | if (alloc_flags & ALLOC_FAIR) { | |
2208 | alloc_flags &= ~ALLOC_FAIR; | |
2209 | if (nr_fair_skipped) { | |
2210 | zonelist_rescan = true; | |
2211 | reset_alloc_batches(preferred_zone); | |
2212 | } | |
2213 | if (nr_online_nodes > 1) | |
2214 | zonelist_rescan = true; | |
2215 | } | |
2216 | ||
2217 | if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) { | |
2218 | /* Disable zlc cache for second zonelist scan */ | |
2219 | zlc_active = 0; | |
2220 | zonelist_rescan = true; | |
2221 | } | |
2222 | ||
2223 | if (zonelist_rescan) | |
2224 | goto zonelist_scan; | |
2225 | ||
2226 | return NULL; | |
753ee728 MH |
2227 | } |
2228 | ||
29423e77 DR |
2229 | /* |
2230 | * Large machines with many possible nodes should not always dump per-node | |
2231 | * meminfo in irq context. | |
2232 | */ | |
2233 | static inline bool should_suppress_show_mem(void) | |
2234 | { | |
2235 | bool ret = false; | |
2236 | ||
2237 | #if NODES_SHIFT > 8 | |
2238 | ret = in_interrupt(); | |
2239 | #endif | |
2240 | return ret; | |
2241 | } | |
2242 | ||
a238ab5b DH |
2243 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2244 | DEFAULT_RATELIMIT_INTERVAL, | |
2245 | DEFAULT_RATELIMIT_BURST); | |
2246 | ||
2247 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
2248 | { | |
a238ab5b DH |
2249 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2250 | ||
c0a32fc5 SG |
2251 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2252 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2253 | return; |
2254 | ||
2255 | /* | |
2256 | * This documents exceptions given to allocations in certain | |
2257 | * contexts that are allowed to allocate outside current's set | |
2258 | * of allowed nodes. | |
2259 | */ | |
2260 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2261 | if (test_thread_flag(TIF_MEMDIE) || | |
2262 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2263 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2264 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
2265 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2266 | ||
2267 | if (fmt) { | |
3ee9a4f0 JP |
2268 | struct va_format vaf; |
2269 | va_list args; | |
2270 | ||
a238ab5b | 2271 | va_start(args, fmt); |
3ee9a4f0 JP |
2272 | |
2273 | vaf.fmt = fmt; | |
2274 | vaf.va = &args; | |
2275 | ||
2276 | pr_warn("%pV", &vaf); | |
2277 | ||
a238ab5b DH |
2278 | va_end(args); |
2279 | } | |
2280 | ||
3ee9a4f0 JP |
2281 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
2282 | current->comm, order, gfp_mask); | |
a238ab5b DH |
2283 | |
2284 | dump_stack(); | |
2285 | if (!should_suppress_show_mem()) | |
2286 | show_mem(filter); | |
2287 | } | |
2288 | ||
11e33f6a MG |
2289 | static inline int |
2290 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
f90ac398 | 2291 | unsigned long did_some_progress, |
11e33f6a | 2292 | unsigned long pages_reclaimed) |
1da177e4 | 2293 | { |
11e33f6a MG |
2294 | /* Do not loop if specifically requested */ |
2295 | if (gfp_mask & __GFP_NORETRY) | |
2296 | return 0; | |
1da177e4 | 2297 | |
f90ac398 MG |
2298 | /* Always retry if specifically requested */ |
2299 | if (gfp_mask & __GFP_NOFAIL) | |
2300 | return 1; | |
2301 | ||
2302 | /* | |
2303 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim | |
2304 | * making forward progress without invoking OOM. Suspend also disables | |
2305 | * storage devices so kswapd will not help. Bail if we are suspending. | |
2306 | */ | |
2307 | if (!did_some_progress && pm_suspended_storage()) | |
2308 | return 0; | |
2309 | ||
11e33f6a MG |
2310 | /* |
2311 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
2312 | * means __GFP_NOFAIL, but that may not be true in other | |
2313 | * implementations. | |
2314 | */ | |
2315 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
2316 | return 1; | |
2317 | ||
2318 | /* | |
2319 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
2320 | * specified, then we retry until we no longer reclaim any pages | |
2321 | * (above), or we've reclaimed an order of pages at least as | |
2322 | * large as the allocation's order. In both cases, if the | |
2323 | * allocation still fails, we stop retrying. | |
2324 | */ | |
2325 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
2326 | return 1; | |
cf40bd16 | 2327 | |
11e33f6a MG |
2328 | return 0; |
2329 | } | |
933e312e | 2330 | |
11e33f6a MG |
2331 | static inline struct page * |
2332 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
2333 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 | 2334 | nodemask_t *nodemask, struct zone *preferred_zone, |
d8846374 | 2335 | int classzone_idx, int migratetype) |
11e33f6a MG |
2336 | { |
2337 | struct page *page; | |
2338 | ||
e972a070 DR |
2339 | /* Acquire the per-zone oom lock for each zone */ |
2340 | if (!oom_zonelist_trylock(zonelist, gfp_mask)) { | |
11e33f6a | 2341 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2342 | return NULL; |
2343 | } | |
6b1de916 | 2344 | |
5695be14 MH |
2345 | /* |
2346 | * PM-freezer should be notified that there might be an OOM killer on | |
2347 | * its way to kill and wake somebody up. This is too early and we might | |
2348 | * end up not killing anything but false positives are acceptable. | |
2349 | * See freeze_processes. | |
2350 | */ | |
2351 | note_oom_kill(); | |
2352 | ||
11e33f6a MG |
2353 | /* |
2354 | * Go through the zonelist yet one more time, keep very high watermark | |
2355 | * here, this is only to catch a parallel oom killing, we must fail if | |
2356 | * we're still under heavy pressure. | |
2357 | */ | |
2358 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
2359 | order, zonelist, high_zoneidx, | |
5117f45d | 2360 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
d8846374 | 2361 | preferred_zone, classzone_idx, migratetype); |
7fb1d9fc | 2362 | if (page) |
11e33f6a MG |
2363 | goto out; |
2364 | ||
4365a567 KH |
2365 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2366 | /* The OOM killer will not help higher order allocs */ | |
2367 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2368 | goto out; | |
03668b3c DR |
2369 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
2370 | if (high_zoneidx < ZONE_NORMAL) | |
2371 | goto out; | |
4365a567 KH |
2372 | /* |
2373 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
2374 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
2375 | * The caller should handle page allocation failure by itself if | |
2376 | * it specifies __GFP_THISNODE. | |
2377 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
2378 | */ | |
2379 | if (gfp_mask & __GFP_THISNODE) | |
2380 | goto out; | |
2381 | } | |
11e33f6a | 2382 | /* Exhausted what can be done so it's blamo time */ |
08ab9b10 | 2383 | out_of_memory(zonelist, gfp_mask, order, nodemask, false); |
11e33f6a MG |
2384 | |
2385 | out: | |
e972a070 | 2386 | oom_zonelist_unlock(zonelist, gfp_mask); |
11e33f6a MG |
2387 | return page; |
2388 | } | |
2389 | ||
56de7263 MG |
2390 | #ifdef CONFIG_COMPACTION |
2391 | /* Try memory compaction for high-order allocations before reclaim */ | |
2392 | static struct page * | |
2393 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2394 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2395 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
d8846374 | 2396 | int classzone_idx, int migratetype, enum migrate_mode mode, |
1f9efdef | 2397 | int *contended_compaction, bool *deferred_compaction) |
56de7263 | 2398 | { |
53853e2d | 2399 | unsigned long compact_result; |
98dd3b48 | 2400 | struct page *page; |
53853e2d VB |
2401 | |
2402 | if (!order) | |
66199712 | 2403 | return NULL; |
66199712 | 2404 | |
c06b1fca | 2405 | current->flags |= PF_MEMALLOC; |
53853e2d | 2406 | compact_result = try_to_compact_pages(zonelist, order, gfp_mask, |
e0b9daeb | 2407 | nodemask, mode, |
53853e2d | 2408 | contended_compaction, |
97d47a65 | 2409 | alloc_flags, classzone_idx); |
c06b1fca | 2410 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2411 | |
98dd3b48 VB |
2412 | switch (compact_result) { |
2413 | case COMPACT_DEFERRED: | |
53853e2d | 2414 | *deferred_compaction = true; |
98dd3b48 VB |
2415 | /* fall-through */ |
2416 | case COMPACT_SKIPPED: | |
2417 | return NULL; | |
2418 | default: | |
2419 | break; | |
2420 | } | |
53853e2d | 2421 | |
98dd3b48 VB |
2422 | /* |
2423 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
2424 | * count a compaction stall | |
2425 | */ | |
2426 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 2427 | |
98dd3b48 VB |
2428 | page = get_page_from_freelist(gfp_mask, nodemask, |
2429 | order, zonelist, high_zoneidx, | |
2430 | alloc_flags & ~ALLOC_NO_WATERMARKS, | |
2431 | preferred_zone, classzone_idx, migratetype); | |
53853e2d | 2432 | |
98dd3b48 VB |
2433 | if (page) { |
2434 | struct zone *zone = page_zone(page); | |
53853e2d | 2435 | |
98dd3b48 VB |
2436 | zone->compact_blockskip_flush = false; |
2437 | compaction_defer_reset(zone, order, true); | |
2438 | count_vm_event(COMPACTSUCCESS); | |
2439 | return page; | |
2440 | } | |
56de7263 | 2441 | |
98dd3b48 VB |
2442 | /* |
2443 | * It's bad if compaction run occurs and fails. The most likely reason | |
2444 | * is that pages exist, but not enough to satisfy watermarks. | |
2445 | */ | |
2446 | count_vm_event(COMPACTFAIL); | |
66199712 | 2447 | |
98dd3b48 | 2448 | cond_resched(); |
56de7263 MG |
2449 | |
2450 | return NULL; | |
2451 | } | |
2452 | #else | |
2453 | static inline struct page * | |
2454 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2455 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2456 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
53853e2d | 2457 | int classzone_idx, int migratetype, enum migrate_mode mode, |
1f9efdef | 2458 | int *contended_compaction, bool *deferred_compaction) |
56de7263 MG |
2459 | { |
2460 | return NULL; | |
2461 | } | |
2462 | #endif /* CONFIG_COMPACTION */ | |
2463 | ||
bba90710 MS |
2464 | /* Perform direct synchronous page reclaim */ |
2465 | static int | |
2466 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, | |
2467 | nodemask_t *nodemask) | |
11e33f6a | 2468 | { |
11e33f6a | 2469 | struct reclaim_state reclaim_state; |
bba90710 | 2470 | int progress; |
11e33f6a MG |
2471 | |
2472 | cond_resched(); | |
2473 | ||
2474 | /* We now go into synchronous reclaim */ | |
2475 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2476 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2477 | lockdep_set_current_reclaim_state(gfp_mask); |
2478 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2479 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2480 | |
bba90710 | 2481 | progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
11e33f6a | 2482 | |
c06b1fca | 2483 | current->reclaim_state = NULL; |
11e33f6a | 2484 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2485 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2486 | |
2487 | cond_resched(); | |
2488 | ||
bba90710 MS |
2489 | return progress; |
2490 | } | |
2491 | ||
2492 | /* The really slow allocator path where we enter direct reclaim */ | |
2493 | static inline struct page * | |
2494 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
2495 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2496 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
d8846374 | 2497 | int classzone_idx, int migratetype, unsigned long *did_some_progress) |
bba90710 MS |
2498 | { |
2499 | struct page *page = NULL; | |
2500 | bool drained = false; | |
2501 | ||
2502 | *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist, | |
2503 | nodemask); | |
9ee493ce MG |
2504 | if (unlikely(!(*did_some_progress))) |
2505 | return NULL; | |
11e33f6a | 2506 | |
76d3fbf8 | 2507 | /* After successful reclaim, reconsider all zones for allocation */ |
e5adfffc | 2508 | if (IS_ENABLED(CONFIG_NUMA)) |
76d3fbf8 MG |
2509 | zlc_clear_zones_full(zonelist); |
2510 | ||
9ee493ce MG |
2511 | retry: |
2512 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2513 | zonelist, high_zoneidx, |
cfd19c5a | 2514 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
d8846374 MG |
2515 | preferred_zone, classzone_idx, |
2516 | migratetype); | |
9ee493ce MG |
2517 | |
2518 | /* | |
2519 | * If an allocation failed after direct reclaim, it could be because | |
2520 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2521 | */ | |
2522 | if (!page && !drained) { | |
93481ff0 | 2523 | drain_all_pages(NULL); |
9ee493ce MG |
2524 | drained = true; |
2525 | goto retry; | |
2526 | } | |
2527 | ||
11e33f6a MG |
2528 | return page; |
2529 | } | |
2530 | ||
1da177e4 | 2531 | /* |
11e33f6a MG |
2532 | * This is called in the allocator slow-path if the allocation request is of |
2533 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2534 | */ |
11e33f6a MG |
2535 | static inline struct page * |
2536 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
2537 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 | 2538 | nodemask_t *nodemask, struct zone *preferred_zone, |
d8846374 | 2539 | int classzone_idx, int migratetype) |
11e33f6a MG |
2540 | { |
2541 | struct page *page; | |
2542 | ||
2543 | do { | |
2544 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2545 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
d8846374 | 2546 | preferred_zone, classzone_idx, migratetype); |
11e33f6a MG |
2547 | |
2548 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 2549 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
2550 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2551 | ||
2552 | return page; | |
2553 | } | |
2554 | ||
3a025760 JW |
2555 | static void wake_all_kswapds(unsigned int order, |
2556 | struct zonelist *zonelist, | |
2557 | enum zone_type high_zoneidx, | |
7ade3c99 WY |
2558 | struct zone *preferred_zone, |
2559 | nodemask_t *nodemask) | |
3a025760 JW |
2560 | { |
2561 | struct zoneref *z; | |
2562 | struct zone *zone; | |
2563 | ||
7ade3c99 WY |
2564 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
2565 | high_zoneidx, nodemask) | |
3a025760 JW |
2566 | wakeup_kswapd(zone, order, zone_idx(preferred_zone)); |
2567 | } | |
2568 | ||
341ce06f PZ |
2569 | static inline int |
2570 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2571 | { | |
341ce06f | 2572 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
b104a35d | 2573 | const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD)); |
1da177e4 | 2574 | |
a56f57ff | 2575 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2576 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2577 | |
341ce06f PZ |
2578 | /* |
2579 | * The caller may dip into page reserves a bit more if the caller | |
2580 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2581 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
b104a35d | 2582 | * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 2583 | */ |
e6223a3b | 2584 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2585 | |
b104a35d | 2586 | if (atomic) { |
5c3240d9 | 2587 | /* |
b104a35d DR |
2588 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
2589 | * if it can't schedule. | |
5c3240d9 | 2590 | */ |
b104a35d | 2591 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 2592 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 2593 | /* |
b104a35d | 2594 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 2595 | * comment for __cpuset_node_allowed(). |
523b9458 | 2596 | */ |
341ce06f | 2597 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2598 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2599 | alloc_flags |= ALLOC_HARDER; |
2600 | ||
b37f1dd0 MG |
2601 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
2602 | if (gfp_mask & __GFP_MEMALLOC) | |
2603 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
2604 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
2605 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
2606 | else if (!in_interrupt() && | |
2607 | ((current->flags & PF_MEMALLOC) || | |
2608 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 2609 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 2610 | } |
d95ea5d1 | 2611 | #ifdef CONFIG_CMA |
43e7a34d | 2612 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
2613 | alloc_flags |= ALLOC_CMA; |
2614 | #endif | |
341ce06f PZ |
2615 | return alloc_flags; |
2616 | } | |
2617 | ||
072bb0aa MG |
2618 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
2619 | { | |
b37f1dd0 | 2620 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
2621 | } |
2622 | ||
11e33f6a MG |
2623 | static inline struct page * |
2624 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2625 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 | 2626 | nodemask_t *nodemask, struct zone *preferred_zone, |
d8846374 | 2627 | int classzone_idx, int migratetype) |
11e33f6a MG |
2628 | { |
2629 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2630 | struct page *page = NULL; | |
2631 | int alloc_flags; | |
2632 | unsigned long pages_reclaimed = 0; | |
2633 | unsigned long did_some_progress; | |
e0b9daeb | 2634 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
66199712 | 2635 | bool deferred_compaction = false; |
1f9efdef | 2636 | int contended_compaction = COMPACT_CONTENDED_NONE; |
1da177e4 | 2637 | |
72807a74 MG |
2638 | /* |
2639 | * In the slowpath, we sanity check order to avoid ever trying to | |
2640 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2641 | * be using allocators in order of preference for an area that is | |
2642 | * too large. | |
2643 | */ | |
1fc28b70 MG |
2644 | if (order >= MAX_ORDER) { |
2645 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2646 | return NULL; |
1fc28b70 | 2647 | } |
1da177e4 | 2648 | |
952f3b51 CL |
2649 | /* |
2650 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2651 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2652 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2653 | * using a larger set of nodes after it has established that the | |
2654 | * allowed per node queues are empty and that nodes are | |
2655 | * over allocated. | |
2656 | */ | |
3a025760 JW |
2657 | if (IS_ENABLED(CONFIG_NUMA) && |
2658 | (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
952f3b51 CL |
2659 | goto nopage; |
2660 | ||
cc4a6851 | 2661 | restart: |
3a025760 | 2662 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
7ade3c99 WY |
2663 | wake_all_kswapds(order, zonelist, high_zoneidx, |
2664 | preferred_zone, nodemask); | |
1da177e4 | 2665 | |
9bf2229f | 2666 | /* |
7fb1d9fc RS |
2667 | * OK, we're below the kswapd watermark and have kicked background |
2668 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2669 | * to how we want to proceed. | |
9bf2229f | 2670 | */ |
341ce06f | 2671 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2672 | |
f33261d7 DR |
2673 | /* |
2674 | * Find the true preferred zone if the allocation is unconstrained by | |
2675 | * cpusets. | |
2676 | */ | |
d8846374 MG |
2677 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) { |
2678 | struct zoneref *preferred_zoneref; | |
2679 | preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx, | |
2680 | NULL, &preferred_zone); | |
2681 | classzone_idx = zonelist_zone_idx(preferred_zoneref); | |
2682 | } | |
f33261d7 | 2683 | |
cfa54a0f | 2684 | rebalance: |
341ce06f | 2685 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2686 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f | 2687 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
d8846374 | 2688 | preferred_zone, classzone_idx, migratetype); |
7fb1d9fc RS |
2689 | if (page) |
2690 | goto got_pg; | |
1da177e4 | 2691 | |
11e33f6a | 2692 | /* Allocate without watermarks if the context allows */ |
341ce06f | 2693 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
2694 | /* |
2695 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
2696 | * the allocation is high priority and these type of | |
2697 | * allocations are system rather than user orientated | |
2698 | */ | |
2699 | zonelist = node_zonelist(numa_node_id(), gfp_mask); | |
2700 | ||
341ce06f PZ |
2701 | page = __alloc_pages_high_priority(gfp_mask, order, |
2702 | zonelist, high_zoneidx, nodemask, | |
d8846374 | 2703 | preferred_zone, classzone_idx, migratetype); |
cfd19c5a | 2704 | if (page) { |
341ce06f | 2705 | goto got_pg; |
cfd19c5a | 2706 | } |
1da177e4 LT |
2707 | } |
2708 | ||
2709 | /* Atomic allocations - we can't balance anything */ | |
aed0a0e3 DR |
2710 | if (!wait) { |
2711 | /* | |
2712 | * All existing users of the deprecated __GFP_NOFAIL are | |
2713 | * blockable, so warn of any new users that actually allow this | |
2714 | * type of allocation to fail. | |
2715 | */ | |
2716 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 2717 | goto nopage; |
aed0a0e3 | 2718 | } |
1da177e4 | 2719 | |
341ce06f | 2720 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2721 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2722 | goto nopage; |
2723 | ||
6583bb64 DR |
2724 | /* Avoid allocations with no watermarks from looping endlessly */ |
2725 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2726 | goto nopage; | |
2727 | ||
77f1fe6b MG |
2728 | /* |
2729 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2730 | * attempts after direct reclaim are synchronous | |
2731 | */ | |
e0b9daeb DR |
2732 | page = __alloc_pages_direct_compact(gfp_mask, order, zonelist, |
2733 | high_zoneidx, nodemask, alloc_flags, | |
d8846374 MG |
2734 | preferred_zone, |
2735 | classzone_idx, migratetype, | |
e0b9daeb | 2736 | migration_mode, &contended_compaction, |
53853e2d | 2737 | &deferred_compaction); |
56de7263 MG |
2738 | if (page) |
2739 | goto got_pg; | |
75f30861 | 2740 | |
1f9efdef VB |
2741 | /* Checks for THP-specific high-order allocations */ |
2742 | if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) { | |
2743 | /* | |
2744 | * If compaction is deferred for high-order allocations, it is | |
2745 | * because sync compaction recently failed. If this is the case | |
2746 | * and the caller requested a THP allocation, we do not want | |
2747 | * to heavily disrupt the system, so we fail the allocation | |
2748 | * instead of entering direct reclaim. | |
2749 | */ | |
2750 | if (deferred_compaction) | |
2751 | goto nopage; | |
2752 | ||
2753 | /* | |
2754 | * In all zones where compaction was attempted (and not | |
2755 | * deferred or skipped), lock contention has been detected. | |
2756 | * For THP allocation we do not want to disrupt the others | |
2757 | * so we fallback to base pages instead. | |
2758 | */ | |
2759 | if (contended_compaction == COMPACT_CONTENDED_LOCK) | |
2760 | goto nopage; | |
2761 | ||
2762 | /* | |
2763 | * If compaction was aborted due to need_resched(), we do not | |
2764 | * want to further increase allocation latency, unless it is | |
2765 | * khugepaged trying to collapse. | |
2766 | */ | |
2767 | if (contended_compaction == COMPACT_CONTENDED_SCHED | |
2768 | && !(current->flags & PF_KTHREAD)) | |
2769 | goto nopage; | |
2770 | } | |
66199712 | 2771 | |
8fe78048 DR |
2772 | /* |
2773 | * It can become very expensive to allocate transparent hugepages at | |
2774 | * fault, so use asynchronous memory compaction for THP unless it is | |
2775 | * khugepaged trying to collapse. | |
2776 | */ | |
2777 | if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE || | |
2778 | (current->flags & PF_KTHREAD)) | |
2779 | migration_mode = MIGRATE_SYNC_LIGHT; | |
2780 | ||
11e33f6a MG |
2781 | /* Try direct reclaim and then allocating */ |
2782 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2783 | zonelist, high_zoneidx, | |
2784 | nodemask, | |
5117f45d | 2785 | alloc_flags, preferred_zone, |
d8846374 MG |
2786 | classzone_idx, migratetype, |
2787 | &did_some_progress); | |
11e33f6a MG |
2788 | if (page) |
2789 | goto got_pg; | |
1da177e4 | 2790 | |
e33c3b5e | 2791 | /* |
11e33f6a MG |
2792 | * If we failed to make any progress reclaiming, then we are |
2793 | * running out of options and have to consider going OOM | |
e33c3b5e | 2794 | */ |
11e33f6a | 2795 | if (!did_some_progress) { |
b9921ecd | 2796 | if (oom_gfp_allowed(gfp_mask)) { |
7f33d49a RW |
2797 | if (oom_killer_disabled) |
2798 | goto nopage; | |
29fd66d2 DR |
2799 | /* Coredumps can quickly deplete all memory reserves */ |
2800 | if ((current->flags & PF_DUMPCORE) && | |
2801 | !(gfp_mask & __GFP_NOFAIL)) | |
2802 | goto nopage; | |
11e33f6a MG |
2803 | page = __alloc_pages_may_oom(gfp_mask, order, |
2804 | zonelist, high_zoneidx, | |
3dd28266 | 2805 | nodemask, preferred_zone, |
d8846374 | 2806 | classzone_idx, migratetype); |
11e33f6a MG |
2807 | if (page) |
2808 | goto got_pg; | |
1da177e4 | 2809 | |
03668b3c DR |
2810 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2811 | /* | |
2812 | * The oom killer is not called for high-order | |
2813 | * allocations that may fail, so if no progress | |
2814 | * is being made, there are no other options and | |
2815 | * retrying is unlikely to help. | |
2816 | */ | |
2817 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2818 | goto nopage; | |
2819 | /* | |
2820 | * The oom killer is not called for lowmem | |
2821 | * allocations to prevent needlessly killing | |
2822 | * innocent tasks. | |
2823 | */ | |
2824 | if (high_zoneidx < ZONE_NORMAL) | |
2825 | goto nopage; | |
2826 | } | |
e2c55dc8 | 2827 | |
ff0ceb9d DR |
2828 | goto restart; |
2829 | } | |
1da177e4 LT |
2830 | } |
2831 | ||
11e33f6a | 2832 | /* Check if we should retry the allocation */ |
a41f24ea | 2833 | pages_reclaimed += did_some_progress; |
f90ac398 MG |
2834 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
2835 | pages_reclaimed)) { | |
11e33f6a | 2836 | /* Wait for some write requests to complete then retry */ |
0e093d99 | 2837 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2838 | goto rebalance; |
3e7d3449 MG |
2839 | } else { |
2840 | /* | |
2841 | * High-order allocations do not necessarily loop after | |
2842 | * direct reclaim and reclaim/compaction depends on compaction | |
2843 | * being called after reclaim so call directly if necessary | |
2844 | */ | |
e0b9daeb DR |
2845 | page = __alloc_pages_direct_compact(gfp_mask, order, zonelist, |
2846 | high_zoneidx, nodemask, alloc_flags, | |
d8846374 MG |
2847 | preferred_zone, |
2848 | classzone_idx, migratetype, | |
e0b9daeb | 2849 | migration_mode, &contended_compaction, |
53853e2d | 2850 | &deferred_compaction); |
3e7d3449 MG |
2851 | if (page) |
2852 | goto got_pg; | |
1da177e4 LT |
2853 | } |
2854 | ||
2855 | nopage: | |
a238ab5b | 2856 | warn_alloc_failed(gfp_mask, order, NULL); |
b1eeab67 | 2857 | return page; |
1da177e4 | 2858 | got_pg: |
b1eeab67 VN |
2859 | if (kmemcheck_enabled) |
2860 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
11e33f6a | 2861 | |
072bb0aa | 2862 | return page; |
1da177e4 | 2863 | } |
11e33f6a MG |
2864 | |
2865 | /* | |
2866 | * This is the 'heart' of the zoned buddy allocator. | |
2867 | */ | |
2868 | struct page * | |
2869 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2870 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2871 | { | |
2872 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2873 | struct zone *preferred_zone; |
d8846374 | 2874 | struct zoneref *preferred_zoneref; |
cc9a6c87 | 2875 | struct page *page = NULL; |
43e7a34d | 2876 | int migratetype = gfpflags_to_migratetype(gfp_mask); |
cc9a6c87 | 2877 | unsigned int cpuset_mems_cookie; |
3a025760 | 2878 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR; |
d8846374 | 2879 | int classzone_idx; |
11e33f6a | 2880 | |
dcce284a BH |
2881 | gfp_mask &= gfp_allowed_mask; |
2882 | ||
11e33f6a MG |
2883 | lockdep_trace_alloc(gfp_mask); |
2884 | ||
2885 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2886 | ||
2887 | if (should_fail_alloc_page(gfp_mask, order)) | |
2888 | return NULL; | |
2889 | ||
2890 | /* | |
2891 | * Check the zones suitable for the gfp_mask contain at least one | |
2892 | * valid zone. It's possible to have an empty zonelist as a result | |
2893 | * of GFP_THISNODE and a memoryless node | |
2894 | */ | |
2895 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2896 | return NULL; | |
2897 | ||
21bb9bd1 VB |
2898 | if (IS_ENABLED(CONFIG_CMA) && migratetype == MIGRATE_MOVABLE) |
2899 | alloc_flags |= ALLOC_CMA; | |
2900 | ||
cc9a6c87 | 2901 | retry_cpuset: |
d26914d1 | 2902 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 2903 | |
5117f45d | 2904 | /* The preferred zone is used for statistics later */ |
d8846374 | 2905 | preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx, |
f33261d7 DR |
2906 | nodemask ? : &cpuset_current_mems_allowed, |
2907 | &preferred_zone); | |
cc9a6c87 MG |
2908 | if (!preferred_zone) |
2909 | goto out; | |
d8846374 | 2910 | classzone_idx = zonelist_zone_idx(preferred_zoneref); |
5117f45d MG |
2911 | |
2912 | /* First allocation attempt */ | |
11e33f6a | 2913 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
d95ea5d1 | 2914 | zonelist, high_zoneidx, alloc_flags, |
d8846374 | 2915 | preferred_zone, classzone_idx, migratetype); |
21caf2fc ML |
2916 | if (unlikely(!page)) { |
2917 | /* | |
2918 | * Runtime PM, block IO and its error handling path | |
2919 | * can deadlock because I/O on the device might not | |
2920 | * complete. | |
2921 | */ | |
2922 | gfp_mask = memalloc_noio_flags(gfp_mask); | |
11e33f6a | 2923 | page = __alloc_pages_slowpath(gfp_mask, order, |
5117f45d | 2924 | zonelist, high_zoneidx, nodemask, |
d8846374 | 2925 | preferred_zone, classzone_idx, migratetype); |
21caf2fc | 2926 | } |
11e33f6a | 2927 | |
4b4f278c | 2928 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
cc9a6c87 MG |
2929 | |
2930 | out: | |
2931 | /* | |
2932 | * When updating a task's mems_allowed, it is possible to race with | |
2933 | * parallel threads in such a way that an allocation can fail while | |
2934 | * the mask is being updated. If a page allocation is about to fail, | |
2935 | * check if the cpuset changed during allocation and if so, retry. | |
2936 | */ | |
d26914d1 | 2937 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
2938 | goto retry_cpuset; |
2939 | ||
11e33f6a | 2940 | return page; |
1da177e4 | 2941 | } |
d239171e | 2942 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2943 | |
2944 | /* | |
2945 | * Common helper functions. | |
2946 | */ | |
920c7a5d | 2947 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2948 | { |
945a1113 AM |
2949 | struct page *page; |
2950 | ||
2951 | /* | |
2952 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2953 | * a highmem page | |
2954 | */ | |
2955 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2956 | ||
1da177e4 LT |
2957 | page = alloc_pages(gfp_mask, order); |
2958 | if (!page) | |
2959 | return 0; | |
2960 | return (unsigned long) page_address(page); | |
2961 | } | |
1da177e4 LT |
2962 | EXPORT_SYMBOL(__get_free_pages); |
2963 | ||
920c7a5d | 2964 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2965 | { |
945a1113 | 2966 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2967 | } |
1da177e4 LT |
2968 | EXPORT_SYMBOL(get_zeroed_page); |
2969 | ||
920c7a5d | 2970 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2971 | { |
b5810039 | 2972 | if (put_page_testzero(page)) { |
1da177e4 | 2973 | if (order == 0) |
b745bc85 | 2974 | free_hot_cold_page(page, false); |
1da177e4 LT |
2975 | else |
2976 | __free_pages_ok(page, order); | |
2977 | } | |
2978 | } | |
2979 | ||
2980 | EXPORT_SYMBOL(__free_pages); | |
2981 | ||
920c7a5d | 2982 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2983 | { |
2984 | if (addr != 0) { | |
725d704e | 2985 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2986 | __free_pages(virt_to_page((void *)addr), order); |
2987 | } | |
2988 | } | |
2989 | ||
2990 | EXPORT_SYMBOL(free_pages); | |
2991 | ||
6a1a0d3b | 2992 | /* |
52383431 VD |
2993 | * alloc_kmem_pages charges newly allocated pages to the kmem resource counter |
2994 | * of the current memory cgroup. | |
6a1a0d3b | 2995 | * |
52383431 VD |
2996 | * It should be used when the caller would like to use kmalloc, but since the |
2997 | * allocation is large, it has to fall back to the page allocator. | |
2998 | */ | |
2999 | struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order) | |
3000 | { | |
3001 | struct page *page; | |
3002 | struct mem_cgroup *memcg = NULL; | |
3003 | ||
3004 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
3005 | return NULL; | |
3006 | page = alloc_pages(gfp_mask, order); | |
3007 | memcg_kmem_commit_charge(page, memcg, order); | |
3008 | return page; | |
3009 | } | |
3010 | ||
3011 | struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
3012 | { | |
3013 | struct page *page; | |
3014 | struct mem_cgroup *memcg = NULL; | |
3015 | ||
3016 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
3017 | return NULL; | |
3018 | page = alloc_pages_node(nid, gfp_mask, order); | |
3019 | memcg_kmem_commit_charge(page, memcg, order); | |
3020 | return page; | |
3021 | } | |
3022 | ||
3023 | /* | |
3024 | * __free_kmem_pages and free_kmem_pages will free pages allocated with | |
3025 | * alloc_kmem_pages. | |
6a1a0d3b | 3026 | */ |
52383431 | 3027 | void __free_kmem_pages(struct page *page, unsigned int order) |
6a1a0d3b GC |
3028 | { |
3029 | memcg_kmem_uncharge_pages(page, order); | |
3030 | __free_pages(page, order); | |
3031 | } | |
3032 | ||
52383431 | 3033 | void free_kmem_pages(unsigned long addr, unsigned int order) |
6a1a0d3b GC |
3034 | { |
3035 | if (addr != 0) { | |
3036 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
52383431 | 3037 | __free_kmem_pages(virt_to_page((void *)addr), order); |
6a1a0d3b GC |
3038 | } |
3039 | } | |
3040 | ||
ee85c2e1 AK |
3041 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
3042 | { | |
3043 | if (addr) { | |
3044 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
3045 | unsigned long used = addr + PAGE_ALIGN(size); | |
3046 | ||
3047 | split_page(virt_to_page((void *)addr), order); | |
3048 | while (used < alloc_end) { | |
3049 | free_page(used); | |
3050 | used += PAGE_SIZE; | |
3051 | } | |
3052 | } | |
3053 | return (void *)addr; | |
3054 | } | |
3055 | ||
2be0ffe2 TT |
3056 | /** |
3057 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
3058 | * @size: the number of bytes to allocate | |
3059 | * @gfp_mask: GFP flags for the allocation | |
3060 | * | |
3061 | * This function is similar to alloc_pages(), except that it allocates the | |
3062 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
3063 | * allocate memory in power-of-two pages. | |
3064 | * | |
3065 | * This function is also limited by MAX_ORDER. | |
3066 | * | |
3067 | * Memory allocated by this function must be released by free_pages_exact(). | |
3068 | */ | |
3069 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
3070 | { | |
3071 | unsigned int order = get_order(size); | |
3072 | unsigned long addr; | |
3073 | ||
3074 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 3075 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
3076 | } |
3077 | EXPORT_SYMBOL(alloc_pages_exact); | |
3078 | ||
ee85c2e1 AK |
3079 | /** |
3080 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
3081 | * pages on a node. | |
b5e6ab58 | 3082 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
3083 | * @size: the number of bytes to allocate |
3084 | * @gfp_mask: GFP flags for the allocation | |
3085 | * | |
3086 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
3087 | * back. | |
3088 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
3089 | * but is not exact. | |
3090 | */ | |
e1931811 | 3091 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 AK |
3092 | { |
3093 | unsigned order = get_order(size); | |
3094 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
3095 | if (!p) | |
3096 | return NULL; | |
3097 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
3098 | } | |
ee85c2e1 | 3099 | |
2be0ffe2 TT |
3100 | /** |
3101 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
3102 | * @virt: the value returned by alloc_pages_exact. | |
3103 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
3104 | * | |
3105 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
3106 | */ | |
3107 | void free_pages_exact(void *virt, size_t size) | |
3108 | { | |
3109 | unsigned long addr = (unsigned long)virt; | |
3110 | unsigned long end = addr + PAGE_ALIGN(size); | |
3111 | ||
3112 | while (addr < end) { | |
3113 | free_page(addr); | |
3114 | addr += PAGE_SIZE; | |
3115 | } | |
3116 | } | |
3117 | EXPORT_SYMBOL(free_pages_exact); | |
3118 | ||
e0fb5815 ZY |
3119 | /** |
3120 | * nr_free_zone_pages - count number of pages beyond high watermark | |
3121 | * @offset: The zone index of the highest zone | |
3122 | * | |
3123 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
3124 | * high watermark within all zones at or below a given zone index. For each | |
3125 | * zone, the number of pages is calculated as: | |
834405c3 | 3126 | * managed_pages - high_pages |
e0fb5815 | 3127 | */ |
ebec3862 | 3128 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 3129 | { |
dd1a239f | 3130 | struct zoneref *z; |
54a6eb5c MG |
3131 | struct zone *zone; |
3132 | ||
e310fd43 | 3133 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 3134 | unsigned long sum = 0; |
1da177e4 | 3135 | |
0e88460d | 3136 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 3137 | |
54a6eb5c | 3138 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 3139 | unsigned long size = zone->managed_pages; |
41858966 | 3140 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
3141 | if (size > high) |
3142 | sum += size - high; | |
1da177e4 LT |
3143 | } |
3144 | ||
3145 | return sum; | |
3146 | } | |
3147 | ||
e0fb5815 ZY |
3148 | /** |
3149 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
3150 | * | |
3151 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
3152 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 3153 | */ |
ebec3862 | 3154 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 3155 | { |
af4ca457 | 3156 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 3157 | } |
c2f1a551 | 3158 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 3159 | |
e0fb5815 ZY |
3160 | /** |
3161 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
3162 | * | |
3163 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
3164 | * high watermark within all zones. | |
1da177e4 | 3165 | */ |
ebec3862 | 3166 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 3167 | { |
2a1e274a | 3168 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 3169 | } |
08e0f6a9 CL |
3170 | |
3171 | static inline void show_node(struct zone *zone) | |
1da177e4 | 3172 | { |
e5adfffc | 3173 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 3174 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 3175 | } |
1da177e4 | 3176 | |
1da177e4 LT |
3177 | void si_meminfo(struct sysinfo *val) |
3178 | { | |
3179 | val->totalram = totalram_pages; | |
cc7452b6 | 3180 | val->sharedram = global_page_state(NR_SHMEM); |
d23ad423 | 3181 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 3182 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
3183 | val->totalhigh = totalhigh_pages; |
3184 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
3185 | val->mem_unit = PAGE_SIZE; |
3186 | } | |
3187 | ||
3188 | EXPORT_SYMBOL(si_meminfo); | |
3189 | ||
3190 | #ifdef CONFIG_NUMA | |
3191 | void si_meminfo_node(struct sysinfo *val, int nid) | |
3192 | { | |
cdd91a77 JL |
3193 | int zone_type; /* needs to be signed */ |
3194 | unsigned long managed_pages = 0; | |
1da177e4 LT |
3195 | pg_data_t *pgdat = NODE_DATA(nid); |
3196 | ||
cdd91a77 JL |
3197 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
3198 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
3199 | val->totalram = managed_pages; | |
cc7452b6 | 3200 | val->sharedram = node_page_state(nid, NR_SHMEM); |
d23ad423 | 3201 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 3202 | #ifdef CONFIG_HIGHMEM |
b40da049 | 3203 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages; |
d23ad423 CL |
3204 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
3205 | NR_FREE_PAGES); | |
98d2b0eb CL |
3206 | #else |
3207 | val->totalhigh = 0; | |
3208 | val->freehigh = 0; | |
3209 | #endif | |
1da177e4 LT |
3210 | val->mem_unit = PAGE_SIZE; |
3211 | } | |
3212 | #endif | |
3213 | ||
ddd588b5 | 3214 | /* |
7bf02ea2 DR |
3215 | * Determine whether the node should be displayed or not, depending on whether |
3216 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 3217 | */ |
7bf02ea2 | 3218 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
3219 | { |
3220 | bool ret = false; | |
cc9a6c87 | 3221 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
3222 | |
3223 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
3224 | goto out; | |
3225 | ||
cc9a6c87 | 3226 | do { |
d26914d1 | 3227 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3228 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 3229 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
3230 | out: |
3231 | return ret; | |
3232 | } | |
3233 | ||
1da177e4 LT |
3234 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
3235 | ||
377e4f16 RV |
3236 | static void show_migration_types(unsigned char type) |
3237 | { | |
3238 | static const char types[MIGRATE_TYPES] = { | |
3239 | [MIGRATE_UNMOVABLE] = 'U', | |
3240 | [MIGRATE_RECLAIMABLE] = 'E', | |
3241 | [MIGRATE_MOVABLE] = 'M', | |
3242 | [MIGRATE_RESERVE] = 'R', | |
3243 | #ifdef CONFIG_CMA | |
3244 | [MIGRATE_CMA] = 'C', | |
3245 | #endif | |
194159fb | 3246 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 3247 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 3248 | #endif |
377e4f16 RV |
3249 | }; |
3250 | char tmp[MIGRATE_TYPES + 1]; | |
3251 | char *p = tmp; | |
3252 | int i; | |
3253 | ||
3254 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
3255 | if (type & (1 << i)) | |
3256 | *p++ = types[i]; | |
3257 | } | |
3258 | ||
3259 | *p = '\0'; | |
3260 | printk("(%s) ", tmp); | |
3261 | } | |
3262 | ||
1da177e4 LT |
3263 | /* |
3264 | * Show free area list (used inside shift_scroll-lock stuff) | |
3265 | * We also calculate the percentage fragmentation. We do this by counting the | |
3266 | * memory on each free list with the exception of the first item on the list. | |
ddd588b5 DR |
3267 | * Suppresses nodes that are not allowed by current's cpuset if |
3268 | * SHOW_MEM_FILTER_NODES is passed. | |
1da177e4 | 3269 | */ |
7bf02ea2 | 3270 | void show_free_areas(unsigned int filter) |
1da177e4 | 3271 | { |
c7241913 | 3272 | int cpu; |
1da177e4 LT |
3273 | struct zone *zone; |
3274 | ||
ee99c71c | 3275 | for_each_populated_zone(zone) { |
7bf02ea2 | 3276 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3277 | continue; |
c7241913 JS |
3278 | show_node(zone); |
3279 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 3280 | |
6b482c67 | 3281 | for_each_online_cpu(cpu) { |
1da177e4 LT |
3282 | struct per_cpu_pageset *pageset; |
3283 | ||
99dcc3e5 | 3284 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 3285 | |
3dfa5721 CL |
3286 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
3287 | cpu, pageset->pcp.high, | |
3288 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
3289 | } |
3290 | } | |
3291 | ||
a731286d KM |
3292 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
3293 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 3294 | " unevictable:%lu" |
b76146ed | 3295 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 3296 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a BZ |
3297 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
3298 | " free_cma:%lu\n", | |
4f98a2fe | 3299 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 3300 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
3301 | global_page_state(NR_ISOLATED_ANON), |
3302 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 3303 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 3304 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 3305 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 3306 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 3307 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 3308 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 3309 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
3310 | global_page_state(NR_SLAB_RECLAIMABLE), |
3311 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 3312 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 3313 | global_page_state(NR_SHMEM), |
a25700a5 | 3314 | global_page_state(NR_PAGETABLE), |
d1ce749a BZ |
3315 | global_page_state(NR_BOUNCE), |
3316 | global_page_state(NR_FREE_CMA_PAGES)); | |
1da177e4 | 3317 | |
ee99c71c | 3318 | for_each_populated_zone(zone) { |
1da177e4 LT |
3319 | int i; |
3320 | ||
7bf02ea2 | 3321 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3322 | continue; |
1da177e4 LT |
3323 | show_node(zone); |
3324 | printk("%s" | |
3325 | " free:%lukB" | |
3326 | " min:%lukB" | |
3327 | " low:%lukB" | |
3328 | " high:%lukB" | |
4f98a2fe RR |
3329 | " active_anon:%lukB" |
3330 | " inactive_anon:%lukB" | |
3331 | " active_file:%lukB" | |
3332 | " inactive_file:%lukB" | |
7b854121 | 3333 | " unevictable:%lukB" |
a731286d KM |
3334 | " isolated(anon):%lukB" |
3335 | " isolated(file):%lukB" | |
1da177e4 | 3336 | " present:%lukB" |
9feedc9d | 3337 | " managed:%lukB" |
4a0aa73f KM |
3338 | " mlocked:%lukB" |
3339 | " dirty:%lukB" | |
3340 | " writeback:%lukB" | |
3341 | " mapped:%lukB" | |
4b02108a | 3342 | " shmem:%lukB" |
4a0aa73f KM |
3343 | " slab_reclaimable:%lukB" |
3344 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3345 | " kernel_stack:%lukB" |
4a0aa73f KM |
3346 | " pagetables:%lukB" |
3347 | " unstable:%lukB" | |
3348 | " bounce:%lukB" | |
d1ce749a | 3349 | " free_cma:%lukB" |
4a0aa73f | 3350 | " writeback_tmp:%lukB" |
1da177e4 LT |
3351 | " pages_scanned:%lu" |
3352 | " all_unreclaimable? %s" | |
3353 | "\n", | |
3354 | zone->name, | |
88f5acf8 | 3355 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3356 | K(min_wmark_pages(zone)), |
3357 | K(low_wmark_pages(zone)), | |
3358 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3359 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3360 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3361 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3362 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3363 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3364 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3365 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3366 | K(zone->present_pages), |
9feedc9d | 3367 | K(zone->managed_pages), |
4a0aa73f KM |
3368 | K(zone_page_state(zone, NR_MLOCK)), |
3369 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3370 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3371 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3372 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3373 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3374 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3375 | zone_page_state(zone, NR_KERNEL_STACK) * |
3376 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3377 | K(zone_page_state(zone, NR_PAGETABLE)), |
3378 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3379 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1ce749a | 3380 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3381 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
0d5d823a | 3382 | K(zone_page_state(zone, NR_PAGES_SCANNED)), |
6e543d57 | 3383 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
3384 | ); |
3385 | printk("lowmem_reserve[]:"); | |
3386 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3484b2de | 3387 | printk(" %ld", zone->lowmem_reserve[i]); |
1da177e4 LT |
3388 | printk("\n"); |
3389 | } | |
3390 | ||
ee99c71c | 3391 | for_each_populated_zone(zone) { |
b8af2941 | 3392 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
377e4f16 | 3393 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3394 | |
7bf02ea2 | 3395 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3396 | continue; |
1da177e4 LT |
3397 | show_node(zone); |
3398 | printk("%s: ", zone->name); | |
1da177e4 LT |
3399 | |
3400 | spin_lock_irqsave(&zone->lock, flags); | |
3401 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3402 | struct free_area *area = &zone->free_area[order]; |
3403 | int type; | |
3404 | ||
3405 | nr[order] = area->nr_free; | |
8f9de51a | 3406 | total += nr[order] << order; |
377e4f16 RV |
3407 | |
3408 | types[order] = 0; | |
3409 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3410 | if (!list_empty(&area->free_list[type])) | |
3411 | types[order] |= 1 << type; | |
3412 | } | |
1da177e4 LT |
3413 | } |
3414 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3415 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3416 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3417 | if (nr[order]) |
3418 | show_migration_types(types[order]); | |
3419 | } | |
1da177e4 LT |
3420 | printk("= %lukB\n", K(total)); |
3421 | } | |
3422 | ||
949f7ec5 DR |
3423 | hugetlb_show_meminfo(); |
3424 | ||
e6f3602d LW |
3425 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
3426 | ||
1da177e4 LT |
3427 | show_swap_cache_info(); |
3428 | } | |
3429 | ||
19770b32 MG |
3430 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
3431 | { | |
3432 | zoneref->zone = zone; | |
3433 | zoneref->zone_idx = zone_idx(zone); | |
3434 | } | |
3435 | ||
1da177e4 LT |
3436 | /* |
3437 | * Builds allocation fallback zone lists. | |
1a93205b CL |
3438 | * |
3439 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 3440 | */ |
f0c0b2b8 | 3441 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 3442 | int nr_zones) |
1da177e4 | 3443 | { |
1a93205b | 3444 | struct zone *zone; |
bc732f1d | 3445 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
3446 | |
3447 | do { | |
2f6726e5 | 3448 | zone_type--; |
070f8032 | 3449 | zone = pgdat->node_zones + zone_type; |
1a93205b | 3450 | if (populated_zone(zone)) { |
dd1a239f MG |
3451 | zoneref_set_zone(zone, |
3452 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 3453 | check_highest_zone(zone_type); |
1da177e4 | 3454 | } |
2f6726e5 | 3455 | } while (zone_type); |
bc732f1d | 3456 | |
070f8032 | 3457 | return nr_zones; |
1da177e4 LT |
3458 | } |
3459 | ||
f0c0b2b8 KH |
3460 | |
3461 | /* | |
3462 | * zonelist_order: | |
3463 | * 0 = automatic detection of better ordering. | |
3464 | * 1 = order by ([node] distance, -zonetype) | |
3465 | * 2 = order by (-zonetype, [node] distance) | |
3466 | * | |
3467 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
3468 | * the same zonelist. So only NUMA can configure this param. | |
3469 | */ | |
3470 | #define ZONELIST_ORDER_DEFAULT 0 | |
3471 | #define ZONELIST_ORDER_NODE 1 | |
3472 | #define ZONELIST_ORDER_ZONE 2 | |
3473 | ||
3474 | /* zonelist order in the kernel. | |
3475 | * set_zonelist_order() will set this to NODE or ZONE. | |
3476 | */ | |
3477 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3478 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
3479 | ||
3480 | ||
1da177e4 | 3481 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
3482 | /* The value user specified ....changed by config */ |
3483 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3484 | /* string for sysctl */ | |
3485 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
3486 | char numa_zonelist_order[16] = "default"; | |
3487 | ||
3488 | /* | |
3489 | * interface for configure zonelist ordering. | |
3490 | * command line option "numa_zonelist_order" | |
3491 | * = "[dD]efault - default, automatic configuration. | |
3492 | * = "[nN]ode - order by node locality, then by zone within node | |
3493 | * = "[zZ]one - order by zone, then by locality within zone | |
3494 | */ | |
3495 | ||
3496 | static int __parse_numa_zonelist_order(char *s) | |
3497 | { | |
3498 | if (*s == 'd' || *s == 'D') { | |
3499 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3500 | } else if (*s == 'n' || *s == 'N') { | |
3501 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
3502 | } else if (*s == 'z' || *s == 'Z') { | |
3503 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
3504 | } else { | |
3505 | printk(KERN_WARNING | |
3506 | "Ignoring invalid numa_zonelist_order value: " | |
3507 | "%s\n", s); | |
3508 | return -EINVAL; | |
3509 | } | |
3510 | return 0; | |
3511 | } | |
3512 | ||
3513 | static __init int setup_numa_zonelist_order(char *s) | |
3514 | { | |
ecb256f8 VL |
3515 | int ret; |
3516 | ||
3517 | if (!s) | |
3518 | return 0; | |
3519 | ||
3520 | ret = __parse_numa_zonelist_order(s); | |
3521 | if (ret == 0) | |
3522 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
3523 | ||
3524 | return ret; | |
f0c0b2b8 KH |
3525 | } |
3526 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
3527 | ||
3528 | /* | |
3529 | * sysctl handler for numa_zonelist_order | |
3530 | */ | |
cccad5b9 | 3531 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 3532 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
3533 | loff_t *ppos) |
3534 | { | |
3535 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
3536 | int ret; | |
443c6f14 | 3537 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 3538 | |
443c6f14 | 3539 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
3540 | if (write) { |
3541 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
3542 | ret = -EINVAL; | |
3543 | goto out; | |
3544 | } | |
3545 | strcpy(saved_string, (char *)table->data); | |
3546 | } | |
8d65af78 | 3547 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 3548 | if (ret) |
443c6f14 | 3549 | goto out; |
f0c0b2b8 KH |
3550 | if (write) { |
3551 | int oldval = user_zonelist_order; | |
dacbde09 CG |
3552 | |
3553 | ret = __parse_numa_zonelist_order((char *)table->data); | |
3554 | if (ret) { | |
f0c0b2b8 KH |
3555 | /* |
3556 | * bogus value. restore saved string | |
3557 | */ | |
dacbde09 | 3558 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
3559 | NUMA_ZONELIST_ORDER_LEN); |
3560 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
3561 | } else if (oldval != user_zonelist_order) { |
3562 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 3563 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
3564 | mutex_unlock(&zonelists_mutex); |
3565 | } | |
f0c0b2b8 | 3566 | } |
443c6f14 AK |
3567 | out: |
3568 | mutex_unlock(&zl_order_mutex); | |
3569 | return ret; | |
f0c0b2b8 KH |
3570 | } |
3571 | ||
3572 | ||
62bc62a8 | 3573 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
3574 | static int node_load[MAX_NUMNODES]; |
3575 | ||
1da177e4 | 3576 | /** |
4dc3b16b | 3577 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
3578 | * @node: node whose fallback list we're appending |
3579 | * @used_node_mask: nodemask_t of already used nodes | |
3580 | * | |
3581 | * We use a number of factors to determine which is the next node that should | |
3582 | * appear on a given node's fallback list. The node should not have appeared | |
3583 | * already in @node's fallback list, and it should be the next closest node | |
3584 | * according to the distance array (which contains arbitrary distance values | |
3585 | * from each node to each node in the system), and should also prefer nodes | |
3586 | * with no CPUs, since presumably they'll have very little allocation pressure | |
3587 | * on them otherwise. | |
3588 | * It returns -1 if no node is found. | |
3589 | */ | |
f0c0b2b8 | 3590 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 3591 | { |
4cf808eb | 3592 | int n, val; |
1da177e4 | 3593 | int min_val = INT_MAX; |
00ef2d2f | 3594 | int best_node = NUMA_NO_NODE; |
a70f7302 | 3595 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 3596 | |
4cf808eb LT |
3597 | /* Use the local node if we haven't already */ |
3598 | if (!node_isset(node, *used_node_mask)) { | |
3599 | node_set(node, *used_node_mask); | |
3600 | return node; | |
3601 | } | |
1da177e4 | 3602 | |
4b0ef1fe | 3603 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
3604 | |
3605 | /* Don't want a node to appear more than once */ | |
3606 | if (node_isset(n, *used_node_mask)) | |
3607 | continue; | |
3608 | ||
1da177e4 LT |
3609 | /* Use the distance array to find the distance */ |
3610 | val = node_distance(node, n); | |
3611 | ||
4cf808eb LT |
3612 | /* Penalize nodes under us ("prefer the next node") */ |
3613 | val += (n < node); | |
3614 | ||
1da177e4 | 3615 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
3616 | tmp = cpumask_of_node(n); |
3617 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
3618 | val += PENALTY_FOR_NODE_WITH_CPUS; |
3619 | ||
3620 | /* Slight preference for less loaded node */ | |
3621 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
3622 | val += node_load[n]; | |
3623 | ||
3624 | if (val < min_val) { | |
3625 | min_val = val; | |
3626 | best_node = n; | |
3627 | } | |
3628 | } | |
3629 | ||
3630 | if (best_node >= 0) | |
3631 | node_set(best_node, *used_node_mask); | |
3632 | ||
3633 | return best_node; | |
3634 | } | |
3635 | ||
f0c0b2b8 KH |
3636 | |
3637 | /* | |
3638 | * Build zonelists ordered by node and zones within node. | |
3639 | * This results in maximum locality--normal zone overflows into local | |
3640 | * DMA zone, if any--but risks exhausting DMA zone. | |
3641 | */ | |
3642 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 3643 | { |
f0c0b2b8 | 3644 | int j; |
1da177e4 | 3645 | struct zonelist *zonelist; |
f0c0b2b8 | 3646 | |
54a6eb5c | 3647 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 3648 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 3649 | ; |
bc732f1d | 3650 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
3651 | zonelist->_zonerefs[j].zone = NULL; |
3652 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
3653 | } |
3654 | ||
523b9458 CL |
3655 | /* |
3656 | * Build gfp_thisnode zonelists | |
3657 | */ | |
3658 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3659 | { | |
523b9458 CL |
3660 | int j; |
3661 | struct zonelist *zonelist; | |
3662 | ||
54a6eb5c | 3663 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 3664 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
3665 | zonelist->_zonerefs[j].zone = NULL; |
3666 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3667 | } |
3668 | ||
f0c0b2b8 KH |
3669 | /* |
3670 | * Build zonelists ordered by zone and nodes within zones. | |
3671 | * This results in conserving DMA zone[s] until all Normal memory is | |
3672 | * exhausted, but results in overflowing to remote node while memory | |
3673 | * may still exist in local DMA zone. | |
3674 | */ | |
3675 | static int node_order[MAX_NUMNODES]; | |
3676 | ||
3677 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3678 | { | |
f0c0b2b8 KH |
3679 | int pos, j, node; |
3680 | int zone_type; /* needs to be signed */ | |
3681 | struct zone *z; | |
3682 | struct zonelist *zonelist; | |
3683 | ||
54a6eb5c MG |
3684 | zonelist = &pgdat->node_zonelists[0]; |
3685 | pos = 0; | |
3686 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3687 | for (j = 0; j < nr_nodes; j++) { | |
3688 | node = node_order[j]; | |
3689 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3690 | if (populated_zone(z)) { | |
dd1a239f MG |
3691 | zoneref_set_zone(z, |
3692 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3693 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3694 | } |
3695 | } | |
f0c0b2b8 | 3696 | } |
dd1a239f MG |
3697 | zonelist->_zonerefs[pos].zone = NULL; |
3698 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3699 | } |
3700 | ||
3193913c MG |
3701 | #if defined(CONFIG_64BIT) |
3702 | /* | |
3703 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
3704 | * penalty to every machine in case the specialised case applies. Default | |
3705 | * to Node-ordering on 64-bit NUMA machines | |
3706 | */ | |
3707 | static int default_zonelist_order(void) | |
3708 | { | |
3709 | return ZONELIST_ORDER_NODE; | |
3710 | } | |
3711 | #else | |
3712 | /* | |
3713 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
3714 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
3715 | * then reclaim will occur more frequency increasing stalls and potentially | |
3716 | * be easier to OOM if a large percentage of the zone is under writeback or | |
3717 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
3718 | * Hence, default to zone ordering on 32-bit. | |
3719 | */ | |
f0c0b2b8 KH |
3720 | static int default_zonelist_order(void) |
3721 | { | |
f0c0b2b8 KH |
3722 | return ZONELIST_ORDER_ZONE; |
3723 | } | |
3193913c | 3724 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
3725 | |
3726 | static void set_zonelist_order(void) | |
3727 | { | |
3728 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3729 | current_zonelist_order = default_zonelist_order(); | |
3730 | else | |
3731 | current_zonelist_order = user_zonelist_order; | |
3732 | } | |
3733 | ||
3734 | static void build_zonelists(pg_data_t *pgdat) | |
3735 | { | |
3736 | int j, node, load; | |
3737 | enum zone_type i; | |
1da177e4 | 3738 | nodemask_t used_mask; |
f0c0b2b8 KH |
3739 | int local_node, prev_node; |
3740 | struct zonelist *zonelist; | |
3741 | int order = current_zonelist_order; | |
1da177e4 LT |
3742 | |
3743 | /* initialize zonelists */ | |
523b9458 | 3744 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3745 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3746 | zonelist->_zonerefs[0].zone = NULL; |
3747 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3748 | } |
3749 | ||
3750 | /* NUMA-aware ordering of nodes */ | |
3751 | local_node = pgdat->node_id; | |
62bc62a8 | 3752 | load = nr_online_nodes; |
1da177e4 LT |
3753 | prev_node = local_node; |
3754 | nodes_clear(used_mask); | |
f0c0b2b8 | 3755 | |
f0c0b2b8 KH |
3756 | memset(node_order, 0, sizeof(node_order)); |
3757 | j = 0; | |
3758 | ||
1da177e4 LT |
3759 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
3760 | /* | |
3761 | * We don't want to pressure a particular node. | |
3762 | * So adding penalty to the first node in same | |
3763 | * distance group to make it round-robin. | |
3764 | */ | |
957f822a DR |
3765 | if (node_distance(local_node, node) != |
3766 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
3767 | node_load[node] = load; |
3768 | ||
1da177e4 LT |
3769 | prev_node = node; |
3770 | load--; | |
f0c0b2b8 KH |
3771 | if (order == ZONELIST_ORDER_NODE) |
3772 | build_zonelists_in_node_order(pgdat, node); | |
3773 | else | |
3774 | node_order[j++] = node; /* remember order */ | |
3775 | } | |
1da177e4 | 3776 | |
f0c0b2b8 KH |
3777 | if (order == ZONELIST_ORDER_ZONE) { |
3778 | /* calculate node order -- i.e., DMA last! */ | |
3779 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3780 | } |
523b9458 CL |
3781 | |
3782 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3783 | } |
3784 | ||
9276b1bc | 3785 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3786 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3787 | { |
54a6eb5c MG |
3788 | struct zonelist *zonelist; |
3789 | struct zonelist_cache *zlc; | |
dd1a239f | 3790 | struct zoneref *z; |
9276b1bc | 3791 | |
54a6eb5c MG |
3792 | zonelist = &pgdat->node_zonelists[0]; |
3793 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3794 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3795 | for (z = zonelist->_zonerefs; z->zone; z++) |
3796 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
3797 | } |
3798 | ||
7aac7898 LS |
3799 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3800 | /* | |
3801 | * Return node id of node used for "local" allocations. | |
3802 | * I.e., first node id of first zone in arg node's generic zonelist. | |
3803 | * Used for initializing percpu 'numa_mem', which is used primarily | |
3804 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
3805 | */ | |
3806 | int local_memory_node(int node) | |
3807 | { | |
3808 | struct zone *zone; | |
3809 | ||
3810 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
3811 | gfp_zone(GFP_KERNEL), | |
3812 | NULL, | |
3813 | &zone); | |
3814 | return zone->node; | |
3815 | } | |
3816 | #endif | |
f0c0b2b8 | 3817 | |
1da177e4 LT |
3818 | #else /* CONFIG_NUMA */ |
3819 | ||
f0c0b2b8 KH |
3820 | static void set_zonelist_order(void) |
3821 | { | |
3822 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
3823 | } | |
3824 | ||
3825 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 3826 | { |
19655d34 | 3827 | int node, local_node; |
54a6eb5c MG |
3828 | enum zone_type j; |
3829 | struct zonelist *zonelist; | |
1da177e4 LT |
3830 | |
3831 | local_node = pgdat->node_id; | |
1da177e4 | 3832 | |
54a6eb5c | 3833 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 3834 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 3835 | |
54a6eb5c MG |
3836 | /* |
3837 | * Now we build the zonelist so that it contains the zones | |
3838 | * of all the other nodes. | |
3839 | * We don't want to pressure a particular node, so when | |
3840 | * building the zones for node N, we make sure that the | |
3841 | * zones coming right after the local ones are those from | |
3842 | * node N+1 (modulo N) | |
3843 | */ | |
3844 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3845 | if (!node_online(node)) | |
3846 | continue; | |
bc732f1d | 3847 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 3848 | } |
54a6eb5c MG |
3849 | for (node = 0; node < local_node; node++) { |
3850 | if (!node_online(node)) | |
3851 | continue; | |
bc732f1d | 3852 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
3853 | } |
3854 | ||
dd1a239f MG |
3855 | zonelist->_zonerefs[j].zone = NULL; |
3856 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3857 | } |
3858 | ||
9276b1bc | 3859 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3860 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3861 | { |
54a6eb5c | 3862 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3863 | } |
3864 | ||
1da177e4 LT |
3865 | #endif /* CONFIG_NUMA */ |
3866 | ||
99dcc3e5 CL |
3867 | /* |
3868 | * Boot pageset table. One per cpu which is going to be used for all | |
3869 | * zones and all nodes. The parameters will be set in such a way | |
3870 | * that an item put on a list will immediately be handed over to | |
3871 | * the buddy list. This is safe since pageset manipulation is done | |
3872 | * with interrupts disabled. | |
3873 | * | |
3874 | * The boot_pagesets must be kept even after bootup is complete for | |
3875 | * unused processors and/or zones. They do play a role for bootstrapping | |
3876 | * hotplugged processors. | |
3877 | * | |
3878 | * zoneinfo_show() and maybe other functions do | |
3879 | * not check if the processor is online before following the pageset pointer. | |
3880 | * Other parts of the kernel may not check if the zone is available. | |
3881 | */ | |
3882 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3883 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3884 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3885 | |
4eaf3f64 HL |
3886 | /* |
3887 | * Global mutex to protect against size modification of zonelists | |
3888 | * as well as to serialize pageset setup for the new populated zone. | |
3889 | */ | |
3890 | DEFINE_MUTEX(zonelists_mutex); | |
3891 | ||
9b1a4d38 | 3892 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 3893 | static int __build_all_zonelists(void *data) |
1da177e4 | 3894 | { |
6811378e | 3895 | int nid; |
99dcc3e5 | 3896 | int cpu; |
9adb62a5 | 3897 | pg_data_t *self = data; |
9276b1bc | 3898 | |
7f9cfb31 BL |
3899 | #ifdef CONFIG_NUMA |
3900 | memset(node_load, 0, sizeof(node_load)); | |
3901 | #endif | |
9adb62a5 JL |
3902 | |
3903 | if (self && !node_online(self->node_id)) { | |
3904 | build_zonelists(self); | |
3905 | build_zonelist_cache(self); | |
3906 | } | |
3907 | ||
9276b1bc | 3908 | for_each_online_node(nid) { |
7ea1530a CL |
3909 | pg_data_t *pgdat = NODE_DATA(nid); |
3910 | ||
3911 | build_zonelists(pgdat); | |
3912 | build_zonelist_cache(pgdat); | |
9276b1bc | 3913 | } |
99dcc3e5 CL |
3914 | |
3915 | /* | |
3916 | * Initialize the boot_pagesets that are going to be used | |
3917 | * for bootstrapping processors. The real pagesets for | |
3918 | * each zone will be allocated later when the per cpu | |
3919 | * allocator is available. | |
3920 | * | |
3921 | * boot_pagesets are used also for bootstrapping offline | |
3922 | * cpus if the system is already booted because the pagesets | |
3923 | * are needed to initialize allocators on a specific cpu too. | |
3924 | * F.e. the percpu allocator needs the page allocator which | |
3925 | * needs the percpu allocator in order to allocate its pagesets | |
3926 | * (a chicken-egg dilemma). | |
3927 | */ | |
7aac7898 | 3928 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3929 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3930 | ||
7aac7898 LS |
3931 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3932 | /* | |
3933 | * We now know the "local memory node" for each node-- | |
3934 | * i.e., the node of the first zone in the generic zonelist. | |
3935 | * Set up numa_mem percpu variable for on-line cpus. During | |
3936 | * boot, only the boot cpu should be on-line; we'll init the | |
3937 | * secondary cpus' numa_mem as they come on-line. During | |
3938 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3939 | */ | |
3940 | if (cpu_online(cpu)) | |
3941 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3942 | #endif | |
3943 | } | |
3944 | ||
6811378e YG |
3945 | return 0; |
3946 | } | |
3947 | ||
4eaf3f64 HL |
3948 | /* |
3949 | * Called with zonelists_mutex held always | |
3950 | * unless system_state == SYSTEM_BOOTING. | |
3951 | */ | |
9adb62a5 | 3952 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 3953 | { |
f0c0b2b8 KH |
3954 | set_zonelist_order(); |
3955 | ||
6811378e | 3956 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3957 | __build_all_zonelists(NULL); |
68ad8df4 | 3958 | mminit_verify_zonelist(); |
6811378e YG |
3959 | cpuset_init_current_mems_allowed(); |
3960 | } else { | |
e9959f0f | 3961 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
3962 | if (zone) |
3963 | setup_zone_pageset(zone); | |
e9959f0f | 3964 | #endif |
dd1895e2 CS |
3965 | /* we have to stop all cpus to guarantee there is no user |
3966 | of zonelist */ | |
9adb62a5 | 3967 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
3968 | /* cpuset refresh routine should be here */ |
3969 | } | |
bd1e22b8 | 3970 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3971 | /* |
3972 | * Disable grouping by mobility if the number of pages in the | |
3973 | * system is too low to allow the mechanism to work. It would be | |
3974 | * more accurate, but expensive to check per-zone. This check is | |
3975 | * made on memory-hotadd so a system can start with mobility | |
3976 | * disabled and enable it later | |
3977 | */ | |
d9c23400 | 3978 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3979 | page_group_by_mobility_disabled = 1; |
3980 | else | |
3981 | page_group_by_mobility_disabled = 0; | |
3982 | ||
f88dfff5 | 3983 | pr_info("Built %i zonelists in %s order, mobility grouping %s. " |
9ef9acb0 | 3984 | "Total pages: %ld\n", |
62bc62a8 | 3985 | nr_online_nodes, |
f0c0b2b8 | 3986 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3987 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3988 | vm_total_pages); |
3989 | #ifdef CONFIG_NUMA | |
f88dfff5 | 3990 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 3991 | #endif |
1da177e4 LT |
3992 | } |
3993 | ||
3994 | /* | |
3995 | * Helper functions to size the waitqueue hash table. | |
3996 | * Essentially these want to choose hash table sizes sufficiently | |
3997 | * large so that collisions trying to wait on pages are rare. | |
3998 | * But in fact, the number of active page waitqueues on typical | |
3999 | * systems is ridiculously low, less than 200. So this is even | |
4000 | * conservative, even though it seems large. | |
4001 | * | |
4002 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
4003 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
4004 | */ | |
4005 | #define PAGES_PER_WAITQUEUE 256 | |
4006 | ||
cca448fe | 4007 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 4008 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
4009 | { |
4010 | unsigned long size = 1; | |
4011 | ||
4012 | pages /= PAGES_PER_WAITQUEUE; | |
4013 | ||
4014 | while (size < pages) | |
4015 | size <<= 1; | |
4016 | ||
4017 | /* | |
4018 | * Once we have dozens or even hundreds of threads sleeping | |
4019 | * on IO we've got bigger problems than wait queue collision. | |
4020 | * Limit the size of the wait table to a reasonable size. | |
4021 | */ | |
4022 | size = min(size, 4096UL); | |
4023 | ||
4024 | return max(size, 4UL); | |
4025 | } | |
cca448fe YG |
4026 | #else |
4027 | /* | |
4028 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
4029 | * a suitable size for its wait_table. So we use the maximum size now. | |
4030 | * | |
4031 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
4032 | * | |
4033 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
4034 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
4035 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
4036 | * | |
4037 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
4038 | * or more by the traditional way. (See above). It equals: | |
4039 | * | |
4040 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
4041 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
4042 | * powerpc (64K page size) : = (32G +16M)byte. | |
4043 | */ | |
4044 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
4045 | { | |
4046 | return 4096UL; | |
4047 | } | |
4048 | #endif | |
1da177e4 LT |
4049 | |
4050 | /* | |
4051 | * This is an integer logarithm so that shifts can be used later | |
4052 | * to extract the more random high bits from the multiplicative | |
4053 | * hash function before the remainder is taken. | |
4054 | */ | |
4055 | static inline unsigned long wait_table_bits(unsigned long size) | |
4056 | { | |
4057 | return ffz(~size); | |
4058 | } | |
4059 | ||
6d3163ce AH |
4060 | /* |
4061 | * Check if a pageblock contains reserved pages | |
4062 | */ | |
4063 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
4064 | { | |
4065 | unsigned long pfn; | |
4066 | ||
4067 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
4068 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
4069 | return 1; | |
4070 | } | |
4071 | return 0; | |
4072 | } | |
4073 | ||
56fd56b8 | 4074 | /* |
d9c23400 | 4075 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
4076 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
4077 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
4078 | * higher will lead to a bigger reserve which will get freed as contiguous |
4079 | * blocks as reclaim kicks in | |
4080 | */ | |
4081 | static void setup_zone_migrate_reserve(struct zone *zone) | |
4082 | { | |
6d3163ce | 4083 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 4084 | struct page *page; |
78986a67 MG |
4085 | unsigned long block_migratetype; |
4086 | int reserve; | |
943dca1a | 4087 | int old_reserve; |
56fd56b8 | 4088 | |
d0215638 MH |
4089 | /* |
4090 | * Get the start pfn, end pfn and the number of blocks to reserve | |
4091 | * We have to be careful to be aligned to pageblock_nr_pages to | |
4092 | * make sure that we always check pfn_valid for the first page in | |
4093 | * the block. | |
4094 | */ | |
56fd56b8 | 4095 | start_pfn = zone->zone_start_pfn; |
108bcc96 | 4096 | end_pfn = zone_end_pfn(zone); |
d0215638 | 4097 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 4098 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 4099 | pageblock_order; |
56fd56b8 | 4100 | |
78986a67 MG |
4101 | /* |
4102 | * Reserve blocks are generally in place to help high-order atomic | |
4103 | * allocations that are short-lived. A min_free_kbytes value that | |
4104 | * would result in more than 2 reserve blocks for atomic allocations | |
4105 | * is assumed to be in place to help anti-fragmentation for the | |
4106 | * future allocation of hugepages at runtime. | |
4107 | */ | |
4108 | reserve = min(2, reserve); | |
943dca1a YI |
4109 | old_reserve = zone->nr_migrate_reserve_block; |
4110 | ||
4111 | /* When memory hot-add, we almost always need to do nothing */ | |
4112 | if (reserve == old_reserve) | |
4113 | return; | |
4114 | zone->nr_migrate_reserve_block = reserve; | |
78986a67 | 4115 | |
d9c23400 | 4116 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
4117 | if (!pfn_valid(pfn)) |
4118 | continue; | |
4119 | page = pfn_to_page(pfn); | |
4120 | ||
344c790e AL |
4121 | /* Watch out for overlapping nodes */ |
4122 | if (page_to_nid(page) != zone_to_nid(zone)) | |
4123 | continue; | |
4124 | ||
56fd56b8 MG |
4125 | block_migratetype = get_pageblock_migratetype(page); |
4126 | ||
938929f1 MG |
4127 | /* Only test what is necessary when the reserves are not met */ |
4128 | if (reserve > 0) { | |
4129 | /* | |
4130 | * Blocks with reserved pages will never free, skip | |
4131 | * them. | |
4132 | */ | |
4133 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
4134 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
4135 | continue; | |
56fd56b8 | 4136 | |
938929f1 MG |
4137 | /* If this block is reserved, account for it */ |
4138 | if (block_migratetype == MIGRATE_RESERVE) { | |
4139 | reserve--; | |
4140 | continue; | |
4141 | } | |
4142 | ||
4143 | /* Suitable for reserving if this block is movable */ | |
4144 | if (block_migratetype == MIGRATE_MOVABLE) { | |
4145 | set_pageblock_migratetype(page, | |
4146 | MIGRATE_RESERVE); | |
4147 | move_freepages_block(zone, page, | |
4148 | MIGRATE_RESERVE); | |
4149 | reserve--; | |
4150 | continue; | |
4151 | } | |
943dca1a YI |
4152 | } else if (!old_reserve) { |
4153 | /* | |
4154 | * At boot time we don't need to scan the whole zone | |
4155 | * for turning off MIGRATE_RESERVE. | |
4156 | */ | |
4157 | break; | |
56fd56b8 MG |
4158 | } |
4159 | ||
4160 | /* | |
4161 | * If the reserve is met and this is a previous reserved block, | |
4162 | * take it back | |
4163 | */ | |
4164 | if (block_migratetype == MIGRATE_RESERVE) { | |
4165 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4166 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
4167 | } | |
4168 | } | |
4169 | } | |
ac0e5b7a | 4170 | |
1da177e4 LT |
4171 | /* |
4172 | * Initially all pages are reserved - free ones are freed | |
4173 | * up by free_all_bootmem() once the early boot process is | |
4174 | * done. Non-atomic initialization, single-pass. | |
4175 | */ | |
c09b4240 | 4176 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 4177 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 4178 | { |
1da177e4 | 4179 | struct page *page; |
29751f69 AW |
4180 | unsigned long end_pfn = start_pfn + size; |
4181 | unsigned long pfn; | |
86051ca5 | 4182 | struct zone *z; |
1da177e4 | 4183 | |
22b31eec HD |
4184 | if (highest_memmap_pfn < end_pfn - 1) |
4185 | highest_memmap_pfn = end_pfn - 1; | |
4186 | ||
86051ca5 | 4187 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 4188 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
4189 | /* |
4190 | * There can be holes in boot-time mem_map[]s | |
4191 | * handed to this function. They do not | |
4192 | * exist on hotplugged memory. | |
4193 | */ | |
4194 | if (context == MEMMAP_EARLY) { | |
4195 | if (!early_pfn_valid(pfn)) | |
4196 | continue; | |
4197 | if (!early_pfn_in_nid(pfn, nid)) | |
4198 | continue; | |
4199 | } | |
d41dee36 AW |
4200 | page = pfn_to_page(pfn); |
4201 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 4202 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 4203 | init_page_count(page); |
22b751c3 | 4204 | page_mapcount_reset(page); |
90572890 | 4205 | page_cpupid_reset_last(page); |
1da177e4 | 4206 | SetPageReserved(page); |
b2a0ac88 MG |
4207 | /* |
4208 | * Mark the block movable so that blocks are reserved for | |
4209 | * movable at startup. This will force kernel allocations | |
4210 | * to reserve their blocks rather than leaking throughout | |
4211 | * the address space during boot when many long-lived | |
56fd56b8 MG |
4212 | * kernel allocations are made. Later some blocks near |
4213 | * the start are marked MIGRATE_RESERVE by | |
4214 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
4215 | * |
4216 | * bitmap is created for zone's valid pfn range. but memmap | |
4217 | * can be created for invalid pages (for alignment) | |
4218 | * check here not to call set_pageblock_migratetype() against | |
4219 | * pfn out of zone. | |
b2a0ac88 | 4220 | */ |
86051ca5 | 4221 | if ((z->zone_start_pfn <= pfn) |
108bcc96 | 4222 | && (pfn < zone_end_pfn(z)) |
86051ca5 | 4223 | && !(pfn & (pageblock_nr_pages - 1))) |
56fd56b8 | 4224 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 4225 | |
1da177e4 LT |
4226 | INIT_LIST_HEAD(&page->lru); |
4227 | #ifdef WANT_PAGE_VIRTUAL | |
4228 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
4229 | if (!is_highmem_idx(zone)) | |
3212c6be | 4230 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 4231 | #endif |
1da177e4 LT |
4232 | } |
4233 | } | |
4234 | ||
1e548deb | 4235 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 4236 | { |
7aeb09f9 | 4237 | unsigned int order, t; |
b2a0ac88 MG |
4238 | for_each_migratetype_order(order, t) { |
4239 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
4240 | zone->free_area[order].nr_free = 0; |
4241 | } | |
4242 | } | |
4243 | ||
4244 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
4245 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 4246 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
4247 | #endif |
4248 | ||
7cd2b0a3 | 4249 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 4250 | { |
3a6be87f | 4251 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
4252 | int batch; |
4253 | ||
4254 | /* | |
4255 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 4256 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
4257 | * |
4258 | * OK, so we don't know how big the cache is. So guess. | |
4259 | */ | |
b40da049 | 4260 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
4261 | if (batch * PAGE_SIZE > 512 * 1024) |
4262 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
4263 | batch /= 4; /* We effectively *= 4 below */ |
4264 | if (batch < 1) | |
4265 | batch = 1; | |
4266 | ||
4267 | /* | |
0ceaacc9 NP |
4268 | * Clamp the batch to a 2^n - 1 value. Having a power |
4269 | * of 2 value was found to be more likely to have | |
4270 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 4271 | * |
0ceaacc9 NP |
4272 | * For example if 2 tasks are alternately allocating |
4273 | * batches of pages, one task can end up with a lot | |
4274 | * of pages of one half of the possible page colors | |
4275 | * and the other with pages of the other colors. | |
e7c8d5c9 | 4276 | */ |
9155203a | 4277 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 4278 | |
e7c8d5c9 | 4279 | return batch; |
3a6be87f DH |
4280 | |
4281 | #else | |
4282 | /* The deferral and batching of frees should be suppressed under NOMMU | |
4283 | * conditions. | |
4284 | * | |
4285 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4286 | * of contiguous memory as there's no hardware page translation to | |
4287 | * assemble apparent contiguous memory from discontiguous pages. | |
4288 | * | |
4289 | * Queueing large contiguous runs of pages for batching, however, | |
4290 | * causes the pages to actually be freed in smaller chunks. As there | |
4291 | * can be a significant delay between the individual batches being | |
4292 | * recycled, this leads to the once large chunks of space being | |
4293 | * fragmented and becoming unavailable for high-order allocations. | |
4294 | */ | |
4295 | return 0; | |
4296 | #endif | |
e7c8d5c9 CL |
4297 | } |
4298 | ||
8d7a8fa9 CS |
4299 | /* |
4300 | * pcp->high and pcp->batch values are related and dependent on one another: | |
4301 | * ->batch must never be higher then ->high. | |
4302 | * The following function updates them in a safe manner without read side | |
4303 | * locking. | |
4304 | * | |
4305 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
4306 | * those fields changing asynchronously (acording the the above rule). | |
4307 | * | |
4308 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
4309 | * outside of boot time (or some other assurance that no concurrent updaters | |
4310 | * exist). | |
4311 | */ | |
4312 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
4313 | unsigned long batch) | |
4314 | { | |
4315 | /* start with a fail safe value for batch */ | |
4316 | pcp->batch = 1; | |
4317 | smp_wmb(); | |
4318 | ||
4319 | /* Update high, then batch, in order */ | |
4320 | pcp->high = high; | |
4321 | smp_wmb(); | |
4322 | ||
4323 | pcp->batch = batch; | |
4324 | } | |
4325 | ||
3664033c | 4326 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
4327 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
4328 | { | |
8d7a8fa9 | 4329 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
4330 | } |
4331 | ||
88c90dbc | 4332 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
4333 | { |
4334 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4335 | int migratetype; |
2caaad41 | 4336 | |
1c6fe946 MD |
4337 | memset(p, 0, sizeof(*p)); |
4338 | ||
3dfa5721 | 4339 | pcp = &p->pcp; |
2caaad41 | 4340 | pcp->count = 0; |
5f8dcc21 MG |
4341 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4342 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4343 | } |
4344 | ||
88c90dbc CS |
4345 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
4346 | { | |
4347 | pageset_init(p); | |
4348 | pageset_set_batch(p, batch); | |
4349 | } | |
4350 | ||
8ad4b1fb | 4351 | /* |
3664033c | 4352 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
4353 | * to the value high for the pageset p. |
4354 | */ | |
3664033c | 4355 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
4356 | unsigned long high) |
4357 | { | |
8d7a8fa9 CS |
4358 | unsigned long batch = max(1UL, high / 4); |
4359 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
4360 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 4361 | |
8d7a8fa9 | 4362 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
4363 | } |
4364 | ||
7cd2b0a3 DR |
4365 | static void pageset_set_high_and_batch(struct zone *zone, |
4366 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 4367 | { |
56cef2b8 | 4368 | if (percpu_pagelist_fraction) |
3664033c | 4369 | pageset_set_high(pcp, |
56cef2b8 CS |
4370 | (zone->managed_pages / |
4371 | percpu_pagelist_fraction)); | |
4372 | else | |
4373 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
4374 | } | |
4375 | ||
169f6c19 CS |
4376 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
4377 | { | |
4378 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4379 | ||
4380 | pageset_init(pcp); | |
4381 | pageset_set_high_and_batch(zone, pcp); | |
4382 | } | |
4383 | ||
4ed7e022 | 4384 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4385 | { |
4386 | int cpu; | |
319774e2 | 4387 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
4388 | for_each_possible_cpu(cpu) |
4389 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
4390 | } |
4391 | ||
2caaad41 | 4392 | /* |
99dcc3e5 CL |
4393 | * Allocate per cpu pagesets and initialize them. |
4394 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4395 | */ |
99dcc3e5 | 4396 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4397 | { |
99dcc3e5 | 4398 | struct zone *zone; |
e7c8d5c9 | 4399 | |
319774e2 WF |
4400 | for_each_populated_zone(zone) |
4401 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4402 | } |
4403 | ||
577a32f6 | 4404 | static noinline __init_refok |
cca448fe | 4405 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4406 | { |
4407 | int i; | |
cca448fe | 4408 | size_t alloc_size; |
ed8ece2e DH |
4409 | |
4410 | /* | |
4411 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4412 | * per zone. | |
4413 | */ | |
02b694de YG |
4414 | zone->wait_table_hash_nr_entries = |
4415 | wait_table_hash_nr_entries(zone_size_pages); | |
4416 | zone->wait_table_bits = | |
4417 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4418 | alloc_size = zone->wait_table_hash_nr_entries |
4419 | * sizeof(wait_queue_head_t); | |
4420 | ||
cd94b9db | 4421 | if (!slab_is_available()) { |
cca448fe | 4422 | zone->wait_table = (wait_queue_head_t *) |
6782832e SS |
4423 | memblock_virt_alloc_node_nopanic( |
4424 | alloc_size, zone->zone_pgdat->node_id); | |
cca448fe YG |
4425 | } else { |
4426 | /* | |
4427 | * This case means that a zone whose size was 0 gets new memory | |
4428 | * via memory hot-add. | |
4429 | * But it may be the case that a new node was hot-added. In | |
4430 | * this case vmalloc() will not be able to use this new node's | |
4431 | * memory - this wait_table must be initialized to use this new | |
4432 | * node itself as well. | |
4433 | * To use this new node's memory, further consideration will be | |
4434 | * necessary. | |
4435 | */ | |
8691f3a7 | 4436 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4437 | } |
4438 | if (!zone->wait_table) | |
4439 | return -ENOMEM; | |
ed8ece2e | 4440 | |
b8af2941 | 4441 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4442 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4443 | |
4444 | return 0; | |
ed8ece2e DH |
4445 | } |
4446 | ||
c09b4240 | 4447 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4448 | { |
99dcc3e5 CL |
4449 | /* |
4450 | * per cpu subsystem is not up at this point. The following code | |
4451 | * relies on the ability of the linker to provide the | |
4452 | * offset of a (static) per cpu variable into the per cpu area. | |
4453 | */ | |
4454 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4455 | |
b38a8725 | 4456 | if (populated_zone(zone)) |
99dcc3e5 CL |
4457 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4458 | zone->name, zone->present_pages, | |
4459 | zone_batchsize(zone)); | |
ed8ece2e DH |
4460 | } |
4461 | ||
4ed7e022 | 4462 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4463 | unsigned long zone_start_pfn, |
a2f3aa02 DH |
4464 | unsigned long size, |
4465 | enum memmap_context context) | |
ed8ece2e DH |
4466 | { |
4467 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4468 | int ret; |
4469 | ret = zone_wait_table_init(zone, size); | |
4470 | if (ret) | |
4471 | return ret; | |
ed8ece2e DH |
4472 | pgdat->nr_zones = zone_idx(zone) + 1; |
4473 | ||
ed8ece2e DH |
4474 | zone->zone_start_pfn = zone_start_pfn; |
4475 | ||
708614e6 MG |
4476 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4477 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4478 | pgdat->node_id, | |
4479 | (unsigned long)zone_idx(zone), | |
4480 | zone_start_pfn, (zone_start_pfn + size)); | |
4481 | ||
1e548deb | 4482 | zone_init_free_lists(zone); |
718127cc YG |
4483 | |
4484 | return 0; | |
ed8ece2e DH |
4485 | } |
4486 | ||
0ee332c1 | 4487 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
4488 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
4489 | /* | |
4490 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 4491 | */ |
f2dbcfa7 | 4492 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d | 4493 | { |
c13291a5 | 4494 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 4495 | int nid; |
7c243c71 RA |
4496 | /* |
4497 | * NOTE: The following SMP-unsafe globals are only used early in boot | |
4498 | * when the kernel is running single-threaded. | |
4499 | */ | |
4500 | static unsigned long __meminitdata last_start_pfn, last_end_pfn; | |
4501 | static int __meminitdata last_nid; | |
4502 | ||
4503 | if (last_start_pfn <= pfn && pfn < last_end_pfn) | |
4504 | return last_nid; | |
c713216d | 4505 | |
e76b63f8 YL |
4506 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
4507 | if (nid != -1) { | |
4508 | last_start_pfn = start_pfn; | |
4509 | last_end_pfn = end_pfn; | |
4510 | last_nid = nid; | |
4511 | } | |
4512 | ||
4513 | return nid; | |
c713216d MG |
4514 | } |
4515 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
4516 | ||
f2dbcfa7 KH |
4517 | int __meminit early_pfn_to_nid(unsigned long pfn) |
4518 | { | |
cc2559bc KH |
4519 | int nid; |
4520 | ||
4521 | nid = __early_pfn_to_nid(pfn); | |
4522 | if (nid >= 0) | |
4523 | return nid; | |
4524 | /* just returns 0 */ | |
4525 | return 0; | |
f2dbcfa7 KH |
4526 | } |
4527 | ||
cc2559bc KH |
4528 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
4529 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
4530 | { | |
4531 | int nid; | |
4532 | ||
4533 | nid = __early_pfn_to_nid(pfn); | |
4534 | if (nid >= 0 && nid != node) | |
4535 | return false; | |
4536 | return true; | |
4537 | } | |
4538 | #endif | |
f2dbcfa7 | 4539 | |
c713216d | 4540 | /** |
6782832e | 4541 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 4542 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 4543 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 4544 | * |
7d018176 ZZ |
4545 | * If an architecture guarantees that all ranges registered contain no holes |
4546 | * and may be freed, this this function may be used instead of calling | |
4547 | * memblock_free_early_nid() manually. | |
c713216d | 4548 | */ |
c13291a5 | 4549 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 4550 | { |
c13291a5 TH |
4551 | unsigned long start_pfn, end_pfn; |
4552 | int i, this_nid; | |
edbe7d23 | 4553 | |
c13291a5 TH |
4554 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
4555 | start_pfn = min(start_pfn, max_low_pfn); | |
4556 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 4557 | |
c13291a5 | 4558 | if (start_pfn < end_pfn) |
6782832e SS |
4559 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
4560 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
4561 | this_nid); | |
edbe7d23 | 4562 | } |
edbe7d23 | 4563 | } |
edbe7d23 | 4564 | |
c713216d MG |
4565 | /** |
4566 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 4567 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 4568 | * |
7d018176 ZZ |
4569 | * If an architecture guarantees that all ranges registered contain no holes and may |
4570 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
4571 | */ |
4572 | void __init sparse_memory_present_with_active_regions(int nid) | |
4573 | { | |
c13291a5 TH |
4574 | unsigned long start_pfn, end_pfn; |
4575 | int i, this_nid; | |
c713216d | 4576 | |
c13291a5 TH |
4577 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
4578 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
4579 | } |
4580 | ||
4581 | /** | |
4582 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
4583 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
4584 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
4585 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
4586 | * |
4587 | * It returns the start and end page frame of a node based on information | |
7d018176 | 4588 | * provided by memblock_set_node(). If called for a node |
c713216d | 4589 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 4590 | * PFNs will be 0. |
c713216d | 4591 | */ |
a3142c8e | 4592 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
4593 | unsigned long *start_pfn, unsigned long *end_pfn) |
4594 | { | |
c13291a5 | 4595 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 4596 | int i; |
c13291a5 | 4597 | |
c713216d MG |
4598 | *start_pfn = -1UL; |
4599 | *end_pfn = 0; | |
4600 | ||
c13291a5 TH |
4601 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
4602 | *start_pfn = min(*start_pfn, this_start_pfn); | |
4603 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
4604 | } |
4605 | ||
633c0666 | 4606 | if (*start_pfn == -1UL) |
c713216d | 4607 | *start_pfn = 0; |
c713216d MG |
4608 | } |
4609 | ||
2a1e274a MG |
4610 | /* |
4611 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
4612 | * assumption is made that zones within a node are ordered in monotonic | |
4613 | * increasing memory addresses so that the "highest" populated zone is used | |
4614 | */ | |
b69a7288 | 4615 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
4616 | { |
4617 | int zone_index; | |
4618 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
4619 | if (zone_index == ZONE_MOVABLE) | |
4620 | continue; | |
4621 | ||
4622 | if (arch_zone_highest_possible_pfn[zone_index] > | |
4623 | arch_zone_lowest_possible_pfn[zone_index]) | |
4624 | break; | |
4625 | } | |
4626 | ||
4627 | VM_BUG_ON(zone_index == -1); | |
4628 | movable_zone = zone_index; | |
4629 | } | |
4630 | ||
4631 | /* | |
4632 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 4633 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
4634 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
4635 | * in each node depending on the size of each node and how evenly kernelcore | |
4636 | * is distributed. This helper function adjusts the zone ranges | |
4637 | * provided by the architecture for a given node by using the end of the | |
4638 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
4639 | * zones within a node are in order of monotonic increases memory addresses | |
4640 | */ | |
b69a7288 | 4641 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4642 | unsigned long zone_type, |
4643 | unsigned long node_start_pfn, | |
4644 | unsigned long node_end_pfn, | |
4645 | unsigned long *zone_start_pfn, | |
4646 | unsigned long *zone_end_pfn) | |
4647 | { | |
4648 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4649 | if (zone_movable_pfn[nid]) { | |
4650 | /* Size ZONE_MOVABLE */ | |
4651 | if (zone_type == ZONE_MOVABLE) { | |
4652 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4653 | *zone_end_pfn = min(node_end_pfn, | |
4654 | arch_zone_highest_possible_pfn[movable_zone]); | |
4655 | ||
4656 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4657 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4658 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4659 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4660 | ||
4661 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4662 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4663 | *zone_start_pfn = *zone_end_pfn; | |
4664 | } | |
4665 | } | |
4666 | ||
c713216d MG |
4667 | /* |
4668 | * Return the number of pages a zone spans in a node, including holes | |
4669 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4670 | */ | |
6ea6e688 | 4671 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4672 | unsigned long zone_type, |
7960aedd ZY |
4673 | unsigned long node_start_pfn, |
4674 | unsigned long node_end_pfn, | |
c713216d MG |
4675 | unsigned long *ignored) |
4676 | { | |
c713216d MG |
4677 | unsigned long zone_start_pfn, zone_end_pfn; |
4678 | ||
7960aedd | 4679 | /* Get the start and end of the zone */ |
c713216d MG |
4680 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
4681 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4682 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4683 | node_start_pfn, node_end_pfn, | |
4684 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4685 | |
4686 | /* Check that this node has pages within the zone's required range */ | |
4687 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4688 | return 0; | |
4689 | ||
4690 | /* Move the zone boundaries inside the node if necessary */ | |
4691 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4692 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4693 | ||
4694 | /* Return the spanned pages */ | |
4695 | return zone_end_pfn - zone_start_pfn; | |
4696 | } | |
4697 | ||
4698 | /* | |
4699 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4700 | * then all holes in the requested range will be accounted for. |
c713216d | 4701 | */ |
32996250 | 4702 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4703 | unsigned long range_start_pfn, |
4704 | unsigned long range_end_pfn) | |
4705 | { | |
96e907d1 TH |
4706 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4707 | unsigned long start_pfn, end_pfn; | |
4708 | int i; | |
c713216d | 4709 | |
96e907d1 TH |
4710 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4711 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4712 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4713 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4714 | } |
96e907d1 | 4715 | return nr_absent; |
c713216d MG |
4716 | } |
4717 | ||
4718 | /** | |
4719 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4720 | * @start_pfn: The start PFN to start searching for holes | |
4721 | * @end_pfn: The end PFN to stop searching for holes | |
4722 | * | |
88ca3b94 | 4723 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4724 | */ |
4725 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4726 | unsigned long end_pfn) | |
4727 | { | |
4728 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4729 | } | |
4730 | ||
4731 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4732 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4733 | unsigned long zone_type, |
7960aedd ZY |
4734 | unsigned long node_start_pfn, |
4735 | unsigned long node_end_pfn, | |
c713216d MG |
4736 | unsigned long *ignored) |
4737 | { | |
96e907d1 TH |
4738 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4739 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4740 | unsigned long zone_start_pfn, zone_end_pfn; |
4741 | ||
96e907d1 TH |
4742 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4743 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4744 | |
2a1e274a MG |
4745 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4746 | node_start_pfn, node_end_pfn, | |
4747 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4748 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4749 | } |
0e0b864e | 4750 | |
0ee332c1 | 4751 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4752 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4753 | unsigned long zone_type, |
7960aedd ZY |
4754 | unsigned long node_start_pfn, |
4755 | unsigned long node_end_pfn, | |
c713216d MG |
4756 | unsigned long *zones_size) |
4757 | { | |
4758 | return zones_size[zone_type]; | |
4759 | } | |
4760 | ||
6ea6e688 | 4761 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4762 | unsigned long zone_type, |
7960aedd ZY |
4763 | unsigned long node_start_pfn, |
4764 | unsigned long node_end_pfn, | |
c713216d MG |
4765 | unsigned long *zholes_size) |
4766 | { | |
4767 | if (!zholes_size) | |
4768 | return 0; | |
4769 | ||
4770 | return zholes_size[zone_type]; | |
4771 | } | |
20e6926d | 4772 | |
0ee332c1 | 4773 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4774 | |
a3142c8e | 4775 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
4776 | unsigned long node_start_pfn, |
4777 | unsigned long node_end_pfn, | |
4778 | unsigned long *zones_size, | |
4779 | unsigned long *zholes_size) | |
c713216d MG |
4780 | { |
4781 | unsigned long realtotalpages, totalpages = 0; | |
4782 | enum zone_type i; | |
4783 | ||
4784 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4785 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4786 | node_start_pfn, |
4787 | node_end_pfn, | |
4788 | zones_size); | |
c713216d MG |
4789 | pgdat->node_spanned_pages = totalpages; |
4790 | ||
4791 | realtotalpages = totalpages; | |
4792 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4793 | realtotalpages -= | |
4794 | zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4795 | node_start_pfn, node_end_pfn, |
4796 | zholes_size); | |
c713216d MG |
4797 | pgdat->node_present_pages = realtotalpages; |
4798 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4799 | realtotalpages); | |
4800 | } | |
4801 | ||
835c134e MG |
4802 | #ifndef CONFIG_SPARSEMEM |
4803 | /* | |
4804 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4805 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4806 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4807 | * round what is now in bits to nearest long in bits, then return it in |
4808 | * bytes. | |
4809 | */ | |
7c45512d | 4810 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
4811 | { |
4812 | unsigned long usemapsize; | |
4813 | ||
7c45512d | 4814 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
4815 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4816 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4817 | usemapsize *= NR_PAGEBLOCK_BITS; |
4818 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4819 | ||
4820 | return usemapsize / 8; | |
4821 | } | |
4822 | ||
4823 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
4824 | struct zone *zone, |
4825 | unsigned long zone_start_pfn, | |
4826 | unsigned long zonesize) | |
835c134e | 4827 | { |
7c45512d | 4828 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 4829 | zone->pageblock_flags = NULL; |
58a01a45 | 4830 | if (usemapsize) |
6782832e SS |
4831 | zone->pageblock_flags = |
4832 | memblock_virt_alloc_node_nopanic(usemapsize, | |
4833 | pgdat->node_id); | |
835c134e MG |
4834 | } |
4835 | #else | |
7c45512d LT |
4836 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
4837 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
4838 | #endif /* CONFIG_SPARSEMEM */ |
4839 | ||
d9c23400 | 4840 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 4841 | |
d9c23400 | 4842 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 4843 | void __paginginit set_pageblock_order(void) |
d9c23400 | 4844 | { |
955c1cd7 AM |
4845 | unsigned int order; |
4846 | ||
d9c23400 MG |
4847 | /* Check that pageblock_nr_pages has not already been setup */ |
4848 | if (pageblock_order) | |
4849 | return; | |
4850 | ||
955c1cd7 AM |
4851 | if (HPAGE_SHIFT > PAGE_SHIFT) |
4852 | order = HUGETLB_PAGE_ORDER; | |
4853 | else | |
4854 | order = MAX_ORDER - 1; | |
4855 | ||
d9c23400 MG |
4856 | /* |
4857 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
4858 | * This value may be variable depending on boot parameters on IA64 and |
4859 | * powerpc. | |
d9c23400 MG |
4860 | */ |
4861 | pageblock_order = order; | |
4862 | } | |
4863 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4864 | ||
ba72cb8c MG |
4865 | /* |
4866 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
4867 | * is unused as pageblock_order is set at compile-time. See |
4868 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
4869 | * the kernel config | |
ba72cb8c | 4870 | */ |
15ca220e | 4871 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 4872 | { |
ba72cb8c | 4873 | } |
d9c23400 MG |
4874 | |
4875 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4876 | ||
01cefaef JL |
4877 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
4878 | unsigned long present_pages) | |
4879 | { | |
4880 | unsigned long pages = spanned_pages; | |
4881 | ||
4882 | /* | |
4883 | * Provide a more accurate estimation if there are holes within | |
4884 | * the zone and SPARSEMEM is in use. If there are holes within the | |
4885 | * zone, each populated memory region may cost us one or two extra | |
4886 | * memmap pages due to alignment because memmap pages for each | |
4887 | * populated regions may not naturally algined on page boundary. | |
4888 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
4889 | */ | |
4890 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
4891 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
4892 | pages = present_pages; | |
4893 | ||
4894 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
4895 | } | |
4896 | ||
1da177e4 LT |
4897 | /* |
4898 | * Set up the zone data structures: | |
4899 | * - mark all pages reserved | |
4900 | * - mark all memory queues empty | |
4901 | * - clear the memory bitmaps | |
6527af5d MK |
4902 | * |
4903 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 4904 | */ |
b5a0e011 | 4905 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
7960aedd | 4906 | unsigned long node_start_pfn, unsigned long node_end_pfn, |
1da177e4 LT |
4907 | unsigned long *zones_size, unsigned long *zholes_size) |
4908 | { | |
2f1b6248 | 4909 | enum zone_type j; |
ed8ece2e | 4910 | int nid = pgdat->node_id; |
1da177e4 | 4911 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4912 | int ret; |
1da177e4 | 4913 | |
208d54e5 | 4914 | pgdat_resize_init(pgdat); |
8177a420 AA |
4915 | #ifdef CONFIG_NUMA_BALANCING |
4916 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
4917 | pgdat->numabalancing_migrate_nr_pages = 0; | |
4918 | pgdat->numabalancing_migrate_next_window = jiffies; | |
4919 | #endif | |
1da177e4 | 4920 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 4921 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
eefa864b | 4922 | pgdat_page_ext_init(pgdat); |
5f63b720 | 4923 | |
1da177e4 LT |
4924 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4925 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 4926 | unsigned long size, realsize, freesize, memmap_pages; |
1da177e4 | 4927 | |
7960aedd ZY |
4928 | size = zone_spanned_pages_in_node(nid, j, node_start_pfn, |
4929 | node_end_pfn, zones_size); | |
9feedc9d | 4930 | realsize = freesize = size - zone_absent_pages_in_node(nid, j, |
7960aedd ZY |
4931 | node_start_pfn, |
4932 | node_end_pfn, | |
c713216d | 4933 | zholes_size); |
1da177e4 | 4934 | |
0e0b864e | 4935 | /* |
9feedc9d | 4936 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
4937 | * is used by this zone for memmap. This affects the watermark |
4938 | * and per-cpu initialisations | |
4939 | */ | |
01cefaef | 4940 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
4941 | if (!is_highmem_idx(j)) { |
4942 | if (freesize >= memmap_pages) { | |
4943 | freesize -= memmap_pages; | |
4944 | if (memmap_pages) | |
4945 | printk(KERN_DEBUG | |
4946 | " %s zone: %lu pages used for memmap\n", | |
4947 | zone_names[j], memmap_pages); | |
4948 | } else | |
4949 | printk(KERN_WARNING | |
4950 | " %s zone: %lu pages exceeds freesize %lu\n", | |
4951 | zone_names[j], memmap_pages, freesize); | |
4952 | } | |
0e0b864e | 4953 | |
6267276f | 4954 | /* Account for reserved pages */ |
9feedc9d JL |
4955 | if (j == 0 && freesize > dma_reserve) { |
4956 | freesize -= dma_reserve; | |
d903ef9f | 4957 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4958 | zone_names[0], dma_reserve); |
0e0b864e MG |
4959 | } |
4960 | ||
98d2b0eb | 4961 | if (!is_highmem_idx(j)) |
9feedc9d | 4962 | nr_kernel_pages += freesize; |
01cefaef JL |
4963 | /* Charge for highmem memmap if there are enough kernel pages */ |
4964 | else if (nr_kernel_pages > memmap_pages * 2) | |
4965 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 4966 | nr_all_pages += freesize; |
1da177e4 LT |
4967 | |
4968 | zone->spanned_pages = size; | |
306f2e9e | 4969 | zone->present_pages = realsize; |
9feedc9d JL |
4970 | /* |
4971 | * Set an approximate value for lowmem here, it will be adjusted | |
4972 | * when the bootmem allocator frees pages into the buddy system. | |
4973 | * And all highmem pages will be managed by the buddy system. | |
4974 | */ | |
4975 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 4976 | #ifdef CONFIG_NUMA |
d5f541ed | 4977 | zone->node = nid; |
9feedc9d | 4978 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 4979 | / 100; |
9feedc9d | 4980 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 4981 | #endif |
1da177e4 LT |
4982 | zone->name = zone_names[j]; |
4983 | spin_lock_init(&zone->lock); | |
4984 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4985 | zone_seqlock_init(zone); |
1da177e4 | 4986 | zone->zone_pgdat = pgdat; |
ed8ece2e | 4987 | zone_pcp_init(zone); |
81c0a2bb JW |
4988 | |
4989 | /* For bootup, initialized properly in watermark setup */ | |
4990 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
4991 | ||
bea8c150 | 4992 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
4993 | if (!size) |
4994 | continue; | |
4995 | ||
955c1cd7 | 4996 | set_pageblock_order(); |
7c45512d | 4997 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
a2f3aa02 DH |
4998 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4999 | size, MEMMAP_EARLY); | |
718127cc | 5000 | BUG_ON(ret); |
76cdd58e | 5001 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 5002 | zone_start_pfn += size; |
1da177e4 LT |
5003 | } |
5004 | } | |
5005 | ||
577a32f6 | 5006 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 5007 | { |
1da177e4 LT |
5008 | /* Skip empty nodes */ |
5009 | if (!pgdat->node_spanned_pages) | |
5010 | return; | |
5011 | ||
d41dee36 | 5012 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
5013 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
5014 | if (!pgdat->node_mem_map) { | |
e984bb43 | 5015 | unsigned long size, start, end; |
d41dee36 AW |
5016 | struct page *map; |
5017 | ||
e984bb43 BP |
5018 | /* |
5019 | * The zone's endpoints aren't required to be MAX_ORDER | |
5020 | * aligned but the node_mem_map endpoints must be in order | |
5021 | * for the buddy allocator to function correctly. | |
5022 | */ | |
5023 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
108bcc96 | 5024 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
5025 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
5026 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
5027 | map = alloc_remap(pgdat->node_id, size); |
5028 | if (!map) | |
6782832e SS |
5029 | map = memblock_virt_alloc_node_nopanic(size, |
5030 | pgdat->node_id); | |
e984bb43 | 5031 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 5032 | } |
12d810c1 | 5033 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
5034 | /* |
5035 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
5036 | */ | |
c713216d | 5037 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 5038 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 5039 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5040 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 5041 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 5042 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5043 | } |
1da177e4 | 5044 | #endif |
d41dee36 | 5045 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
5046 | } |
5047 | ||
9109fb7b JW |
5048 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
5049 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 5050 | { |
9109fb7b | 5051 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
5052 | unsigned long start_pfn = 0; |
5053 | unsigned long end_pfn = 0; | |
9109fb7b | 5054 | |
88fdf75d | 5055 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 5056 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 5057 | |
1da177e4 LT |
5058 | pgdat->node_id = nid; |
5059 | pgdat->node_start_pfn = node_start_pfn; | |
7960aedd ZY |
5060 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5061 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8b375f64 LC |
5062 | printk(KERN_INFO "Initmem setup node %d [mem %#010Lx-%#010Lx]\n", nid, |
5063 | (u64) start_pfn << PAGE_SHIFT, (u64) (end_pfn << PAGE_SHIFT) - 1); | |
7960aedd ZY |
5064 | #endif |
5065 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
5066 | zones_size, zholes_size); | |
1da177e4 LT |
5067 | |
5068 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
5069 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
5070 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
5071 | nid, (unsigned long)pgdat, | |
5072 | (unsigned long)pgdat->node_mem_map); | |
5073 | #endif | |
1da177e4 | 5074 | |
7960aedd ZY |
5075 | free_area_init_core(pgdat, start_pfn, end_pfn, |
5076 | zones_size, zholes_size); | |
1da177e4 LT |
5077 | } |
5078 | ||
0ee332c1 | 5079 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
5080 | |
5081 | #if MAX_NUMNODES > 1 | |
5082 | /* | |
5083 | * Figure out the number of possible node ids. | |
5084 | */ | |
f9872caf | 5085 | void __init setup_nr_node_ids(void) |
418508c1 MS |
5086 | { |
5087 | unsigned int node; | |
5088 | unsigned int highest = 0; | |
5089 | ||
5090 | for_each_node_mask(node, node_possible_map) | |
5091 | highest = node; | |
5092 | nr_node_ids = highest + 1; | |
5093 | } | |
418508c1 MS |
5094 | #endif |
5095 | ||
1e01979c TH |
5096 | /** |
5097 | * node_map_pfn_alignment - determine the maximum internode alignment | |
5098 | * | |
5099 | * This function should be called after node map is populated and sorted. | |
5100 | * It calculates the maximum power of two alignment which can distinguish | |
5101 | * all the nodes. | |
5102 | * | |
5103 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
5104 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
5105 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
5106 | * shifted, 1GiB is enough and this function will indicate so. | |
5107 | * | |
5108 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
5109 | * model has fine enough granularity to avoid incorrect mapping for the | |
5110 | * populated node map. | |
5111 | * | |
5112 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
5113 | * requirement (single node). | |
5114 | */ | |
5115 | unsigned long __init node_map_pfn_alignment(void) | |
5116 | { | |
5117 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 5118 | unsigned long start, end, mask; |
1e01979c | 5119 | int last_nid = -1; |
c13291a5 | 5120 | int i, nid; |
1e01979c | 5121 | |
c13291a5 | 5122 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5123 | if (!start || last_nid < 0 || last_nid == nid) { |
5124 | last_nid = nid; | |
5125 | last_end = end; | |
5126 | continue; | |
5127 | } | |
5128 | ||
5129 | /* | |
5130 | * Start with a mask granular enough to pin-point to the | |
5131 | * start pfn and tick off bits one-by-one until it becomes | |
5132 | * too coarse to separate the current node from the last. | |
5133 | */ | |
5134 | mask = ~((1 << __ffs(start)) - 1); | |
5135 | while (mask && last_end <= (start & (mask << 1))) | |
5136 | mask <<= 1; | |
5137 | ||
5138 | /* accumulate all internode masks */ | |
5139 | accl_mask |= mask; | |
5140 | } | |
5141 | ||
5142 | /* convert mask to number of pages */ | |
5143 | return ~accl_mask + 1; | |
5144 | } | |
5145 | ||
a6af2bc3 | 5146 | /* Find the lowest pfn for a node */ |
b69a7288 | 5147 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 5148 | { |
a6af2bc3 | 5149 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
5150 | unsigned long start_pfn; |
5151 | int i; | |
1abbfb41 | 5152 | |
c13291a5 TH |
5153 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
5154 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 5155 | |
a6af2bc3 MG |
5156 | if (min_pfn == ULONG_MAX) { |
5157 | printk(KERN_WARNING | |
2bc0d261 | 5158 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
5159 | return 0; |
5160 | } | |
5161 | ||
5162 | return min_pfn; | |
c713216d MG |
5163 | } |
5164 | ||
5165 | /** | |
5166 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
5167 | * | |
5168 | * It returns the minimum PFN based on information provided via | |
7d018176 | 5169 | * memblock_set_node(). |
c713216d MG |
5170 | */ |
5171 | unsigned long __init find_min_pfn_with_active_regions(void) | |
5172 | { | |
5173 | return find_min_pfn_for_node(MAX_NUMNODES); | |
5174 | } | |
5175 | ||
37b07e41 LS |
5176 | /* |
5177 | * early_calculate_totalpages() | |
5178 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 5179 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 5180 | */ |
484f51f8 | 5181 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 5182 | { |
7e63efef | 5183 | unsigned long totalpages = 0; |
c13291a5 TH |
5184 | unsigned long start_pfn, end_pfn; |
5185 | int i, nid; | |
5186 | ||
5187 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
5188 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 5189 | |
37b07e41 LS |
5190 | totalpages += pages; |
5191 | if (pages) | |
4b0ef1fe | 5192 | node_set_state(nid, N_MEMORY); |
37b07e41 | 5193 | } |
b8af2941 | 5194 | return totalpages; |
7e63efef MG |
5195 | } |
5196 | ||
2a1e274a MG |
5197 | /* |
5198 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
5199 | * is spread evenly between nodes as long as the nodes have enough | |
5200 | * memory. When they don't, some nodes will have more kernelcore than | |
5201 | * others | |
5202 | */ | |
b224ef85 | 5203 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
5204 | { |
5205 | int i, nid; | |
5206 | unsigned long usable_startpfn; | |
5207 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 5208 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 5209 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 5210 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 5211 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 5212 | struct memblock_region *r; |
b2f3eebe TC |
5213 | |
5214 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
5215 | find_usable_zone_for_movable(); | |
5216 | ||
5217 | /* | |
5218 | * If movable_node is specified, ignore kernelcore and movablecore | |
5219 | * options. | |
5220 | */ | |
5221 | if (movable_node_is_enabled()) { | |
136199f0 EM |
5222 | for_each_memblock(memory, r) { |
5223 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
5224 | continue; |
5225 | ||
136199f0 | 5226 | nid = r->nid; |
b2f3eebe | 5227 | |
136199f0 | 5228 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
5229 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
5230 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5231 | usable_startpfn; | |
5232 | } | |
5233 | ||
5234 | goto out2; | |
5235 | } | |
2a1e274a | 5236 | |
7e63efef | 5237 | /* |
b2f3eebe | 5238 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
5239 | * kernelcore that corresponds so that memory usable for |
5240 | * any allocation type is evenly spread. If both kernelcore | |
5241 | * and movablecore are specified, then the value of kernelcore | |
5242 | * will be used for required_kernelcore if it's greater than | |
5243 | * what movablecore would have allowed. | |
5244 | */ | |
5245 | if (required_movablecore) { | |
7e63efef MG |
5246 | unsigned long corepages; |
5247 | ||
5248 | /* | |
5249 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
5250 | * was requested by the user | |
5251 | */ | |
5252 | required_movablecore = | |
5253 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
5254 | corepages = totalpages - required_movablecore; | |
5255 | ||
5256 | required_kernelcore = max(required_kernelcore, corepages); | |
5257 | } | |
5258 | ||
20e6926d YL |
5259 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
5260 | if (!required_kernelcore) | |
66918dcd | 5261 | goto out; |
2a1e274a MG |
5262 | |
5263 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
5264 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
5265 | ||
5266 | restart: | |
5267 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
5268 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 5269 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
5270 | unsigned long start_pfn, end_pfn; |
5271 | ||
2a1e274a MG |
5272 | /* |
5273 | * Recalculate kernelcore_node if the division per node | |
5274 | * now exceeds what is necessary to satisfy the requested | |
5275 | * amount of memory for the kernel | |
5276 | */ | |
5277 | if (required_kernelcore < kernelcore_node) | |
5278 | kernelcore_node = required_kernelcore / usable_nodes; | |
5279 | ||
5280 | /* | |
5281 | * As the map is walked, we track how much memory is usable | |
5282 | * by the kernel using kernelcore_remaining. When it is | |
5283 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
5284 | */ | |
5285 | kernelcore_remaining = kernelcore_node; | |
5286 | ||
5287 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 5288 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
5289 | unsigned long size_pages; |
5290 | ||
c13291a5 | 5291 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
5292 | if (start_pfn >= end_pfn) |
5293 | continue; | |
5294 | ||
5295 | /* Account for what is only usable for kernelcore */ | |
5296 | if (start_pfn < usable_startpfn) { | |
5297 | unsigned long kernel_pages; | |
5298 | kernel_pages = min(end_pfn, usable_startpfn) | |
5299 | - start_pfn; | |
5300 | ||
5301 | kernelcore_remaining -= min(kernel_pages, | |
5302 | kernelcore_remaining); | |
5303 | required_kernelcore -= min(kernel_pages, | |
5304 | required_kernelcore); | |
5305 | ||
5306 | /* Continue if range is now fully accounted */ | |
5307 | if (end_pfn <= usable_startpfn) { | |
5308 | ||
5309 | /* | |
5310 | * Push zone_movable_pfn to the end so | |
5311 | * that if we have to rebalance | |
5312 | * kernelcore across nodes, we will | |
5313 | * not double account here | |
5314 | */ | |
5315 | zone_movable_pfn[nid] = end_pfn; | |
5316 | continue; | |
5317 | } | |
5318 | start_pfn = usable_startpfn; | |
5319 | } | |
5320 | ||
5321 | /* | |
5322 | * The usable PFN range for ZONE_MOVABLE is from | |
5323 | * start_pfn->end_pfn. Calculate size_pages as the | |
5324 | * number of pages used as kernelcore | |
5325 | */ | |
5326 | size_pages = end_pfn - start_pfn; | |
5327 | if (size_pages > kernelcore_remaining) | |
5328 | size_pages = kernelcore_remaining; | |
5329 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5330 | ||
5331 | /* | |
5332 | * Some kernelcore has been met, update counts and | |
5333 | * break if the kernelcore for this node has been | |
b8af2941 | 5334 | * satisfied |
2a1e274a MG |
5335 | */ |
5336 | required_kernelcore -= min(required_kernelcore, | |
5337 | size_pages); | |
5338 | kernelcore_remaining -= size_pages; | |
5339 | if (!kernelcore_remaining) | |
5340 | break; | |
5341 | } | |
5342 | } | |
5343 | ||
5344 | /* | |
5345 | * If there is still required_kernelcore, we do another pass with one | |
5346 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5347 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 5348 | * satisfied |
2a1e274a MG |
5349 | */ |
5350 | usable_nodes--; | |
5351 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5352 | goto restart; | |
5353 | ||
b2f3eebe | 5354 | out2: |
2a1e274a MG |
5355 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
5356 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5357 | zone_movable_pfn[nid] = | |
5358 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5359 | |
20e6926d | 5360 | out: |
66918dcd | 5361 | /* restore the node_state */ |
4b0ef1fe | 5362 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5363 | } |
5364 | ||
4b0ef1fe LJ |
5365 | /* Any regular or high memory on that node ? */ |
5366 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5367 | { |
37b07e41 LS |
5368 | enum zone_type zone_type; |
5369 | ||
4b0ef1fe LJ |
5370 | if (N_MEMORY == N_NORMAL_MEMORY) |
5371 | return; | |
5372 | ||
5373 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5374 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 5375 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
5376 | node_set_state(nid, N_HIGH_MEMORY); |
5377 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5378 | zone_type <= ZONE_NORMAL) | |
5379 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5380 | break; |
5381 | } | |
37b07e41 | 5382 | } |
37b07e41 LS |
5383 | } |
5384 | ||
c713216d MG |
5385 | /** |
5386 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 5387 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
5388 | * |
5389 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 5390 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
5391 | * zone in each node and their holes is calculated. If the maximum PFN |
5392 | * between two adjacent zones match, it is assumed that the zone is empty. | |
5393 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
5394 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
5395 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
5396 | * at arch_max_dma_pfn. | |
5397 | */ | |
5398 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
5399 | { | |
c13291a5 TH |
5400 | unsigned long start_pfn, end_pfn; |
5401 | int i, nid; | |
a6af2bc3 | 5402 | |
c713216d MG |
5403 | /* Record where the zone boundaries are */ |
5404 | memset(arch_zone_lowest_possible_pfn, 0, | |
5405 | sizeof(arch_zone_lowest_possible_pfn)); | |
5406 | memset(arch_zone_highest_possible_pfn, 0, | |
5407 | sizeof(arch_zone_highest_possible_pfn)); | |
5408 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
5409 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
5410 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
5411 | if (i == ZONE_MOVABLE) |
5412 | continue; | |
c713216d MG |
5413 | arch_zone_lowest_possible_pfn[i] = |
5414 | arch_zone_highest_possible_pfn[i-1]; | |
5415 | arch_zone_highest_possible_pfn[i] = | |
5416 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
5417 | } | |
2a1e274a MG |
5418 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
5419 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
5420 | ||
5421 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
5422 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 5423 | find_zone_movable_pfns_for_nodes(); |
c713216d | 5424 | |
c713216d | 5425 | /* Print out the zone ranges */ |
f88dfff5 | 5426 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
5427 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5428 | if (i == ZONE_MOVABLE) | |
5429 | continue; | |
f88dfff5 | 5430 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
5431 | if (arch_zone_lowest_possible_pfn[i] == |
5432 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 5433 | pr_cont("empty\n"); |
72f0ba02 | 5434 | else |
f88dfff5 | 5435 | pr_cont("[mem %0#10lx-%0#10lx]\n", |
a62e2f4f BH |
5436 | arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT, |
5437 | (arch_zone_highest_possible_pfn[i] | |
5438 | << PAGE_SHIFT) - 1); | |
2a1e274a MG |
5439 | } |
5440 | ||
5441 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 5442 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
5443 | for (i = 0; i < MAX_NUMNODES; i++) { |
5444 | if (zone_movable_pfn[i]) | |
f88dfff5 | 5445 | pr_info(" Node %d: %#010lx\n", i, |
a62e2f4f | 5446 | zone_movable_pfn[i] << PAGE_SHIFT); |
2a1e274a | 5447 | } |
c713216d | 5448 | |
f2d52fe5 | 5449 | /* Print out the early node map */ |
f88dfff5 | 5450 | pr_info("Early memory node ranges\n"); |
c13291a5 | 5451 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
f88dfff5 | 5452 | pr_info(" node %3d: [mem %#010lx-%#010lx]\n", nid, |
a62e2f4f | 5453 | start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1); |
c713216d MG |
5454 | |
5455 | /* Initialise every node */ | |
708614e6 | 5456 | mminit_verify_pageflags_layout(); |
8ef82866 | 5457 | setup_nr_node_ids(); |
c713216d MG |
5458 | for_each_online_node(nid) { |
5459 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 5460 | free_area_init_node(nid, NULL, |
c713216d | 5461 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
5462 | |
5463 | /* Any memory on that node */ | |
5464 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
5465 | node_set_state(nid, N_MEMORY); |
5466 | check_for_memory(pgdat, nid); | |
c713216d MG |
5467 | } |
5468 | } | |
2a1e274a | 5469 | |
7e63efef | 5470 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
5471 | { |
5472 | unsigned long long coremem; | |
5473 | if (!p) | |
5474 | return -EINVAL; | |
5475 | ||
5476 | coremem = memparse(p, &p); | |
7e63efef | 5477 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 5478 | |
7e63efef | 5479 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
5480 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
5481 | ||
5482 | return 0; | |
5483 | } | |
ed7ed365 | 5484 | |
7e63efef MG |
5485 | /* |
5486 | * kernelcore=size sets the amount of memory for use for allocations that | |
5487 | * cannot be reclaimed or migrated. | |
5488 | */ | |
5489 | static int __init cmdline_parse_kernelcore(char *p) | |
5490 | { | |
5491 | return cmdline_parse_core(p, &required_kernelcore); | |
5492 | } | |
5493 | ||
5494 | /* | |
5495 | * movablecore=size sets the amount of memory for use for allocations that | |
5496 | * can be reclaimed or migrated. | |
5497 | */ | |
5498 | static int __init cmdline_parse_movablecore(char *p) | |
5499 | { | |
5500 | return cmdline_parse_core(p, &required_movablecore); | |
5501 | } | |
5502 | ||
ed7ed365 | 5503 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 5504 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 5505 | |
0ee332c1 | 5506 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5507 | |
c3d5f5f0 JL |
5508 | void adjust_managed_page_count(struct page *page, long count) |
5509 | { | |
5510 | spin_lock(&managed_page_count_lock); | |
5511 | page_zone(page)->managed_pages += count; | |
5512 | totalram_pages += count; | |
3dcc0571 JL |
5513 | #ifdef CONFIG_HIGHMEM |
5514 | if (PageHighMem(page)) | |
5515 | totalhigh_pages += count; | |
5516 | #endif | |
c3d5f5f0 JL |
5517 | spin_unlock(&managed_page_count_lock); |
5518 | } | |
3dcc0571 | 5519 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 5520 | |
11199692 | 5521 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 5522 | { |
11199692 JL |
5523 | void *pos; |
5524 | unsigned long pages = 0; | |
69afade7 | 5525 | |
11199692 JL |
5526 | start = (void *)PAGE_ALIGN((unsigned long)start); |
5527 | end = (void *)((unsigned long)end & PAGE_MASK); | |
5528 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 5529 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
5530 | memset(pos, poison, PAGE_SIZE); |
5531 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
5532 | } |
5533 | ||
5534 | if (pages && s) | |
11199692 | 5535 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
5536 | s, pages << (PAGE_SHIFT - 10), start, end); |
5537 | ||
5538 | return pages; | |
5539 | } | |
11199692 | 5540 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 5541 | |
cfa11e08 JL |
5542 | #ifdef CONFIG_HIGHMEM |
5543 | void free_highmem_page(struct page *page) | |
5544 | { | |
5545 | __free_reserved_page(page); | |
5546 | totalram_pages++; | |
7b4b2a0d | 5547 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
5548 | totalhigh_pages++; |
5549 | } | |
5550 | #endif | |
5551 | ||
7ee3d4e8 JL |
5552 | |
5553 | void __init mem_init_print_info(const char *str) | |
5554 | { | |
5555 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
5556 | unsigned long init_code_size, init_data_size; | |
5557 | ||
5558 | physpages = get_num_physpages(); | |
5559 | codesize = _etext - _stext; | |
5560 | datasize = _edata - _sdata; | |
5561 | rosize = __end_rodata - __start_rodata; | |
5562 | bss_size = __bss_stop - __bss_start; | |
5563 | init_data_size = __init_end - __init_begin; | |
5564 | init_code_size = _einittext - _sinittext; | |
5565 | ||
5566 | /* | |
5567 | * Detect special cases and adjust section sizes accordingly: | |
5568 | * 1) .init.* may be embedded into .data sections | |
5569 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
5570 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
5571 | * 3) .rodata.* may be embedded into .text or .data sections. | |
5572 | */ | |
5573 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
5574 | do { \ |
5575 | if (start <= pos && pos < end && size > adj) \ | |
5576 | size -= adj; \ | |
5577 | } while (0) | |
7ee3d4e8 JL |
5578 | |
5579 | adj_init_size(__init_begin, __init_end, init_data_size, | |
5580 | _sinittext, init_code_size); | |
5581 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
5582 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
5583 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
5584 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
5585 | ||
5586 | #undef adj_init_size | |
5587 | ||
f88dfff5 | 5588 | pr_info("Memory: %luK/%luK available " |
7ee3d4e8 | 5589 | "(%luK kernel code, %luK rwdata, %luK rodata, " |
e48322ab | 5590 | "%luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 JL |
5591 | #ifdef CONFIG_HIGHMEM |
5592 | ", %luK highmem" | |
5593 | #endif | |
5594 | "%s%s)\n", | |
5595 | nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10), | |
5596 | codesize >> 10, datasize >> 10, rosize >> 10, | |
5597 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
e48322ab PK |
5598 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10), |
5599 | totalcma_pages << (PAGE_SHIFT-10), | |
7ee3d4e8 JL |
5600 | #ifdef CONFIG_HIGHMEM |
5601 | totalhigh_pages << (PAGE_SHIFT-10), | |
5602 | #endif | |
5603 | str ? ", " : "", str ? str : ""); | |
5604 | } | |
5605 | ||
0e0b864e | 5606 | /** |
88ca3b94 RD |
5607 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
5608 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
5609 | * |
5610 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
5611 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
5612 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
5613 | * function may optionally be used to account for unfreeable pages in the |
5614 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
5615 | * smaller per-cpu batchsize. | |
0e0b864e MG |
5616 | */ |
5617 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
5618 | { | |
5619 | dma_reserve = new_dma_reserve; | |
5620 | } | |
5621 | ||
1da177e4 LT |
5622 | void __init free_area_init(unsigned long *zones_size) |
5623 | { | |
9109fb7b | 5624 | free_area_init_node(0, zones_size, |
1da177e4 LT |
5625 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
5626 | } | |
1da177e4 | 5627 | |
1da177e4 LT |
5628 | static int page_alloc_cpu_notify(struct notifier_block *self, |
5629 | unsigned long action, void *hcpu) | |
5630 | { | |
5631 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 5632 | |
8bb78442 | 5633 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 5634 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
5635 | drain_pages(cpu); |
5636 | ||
5637 | /* | |
5638 | * Spill the event counters of the dead processor | |
5639 | * into the current processors event counters. | |
5640 | * This artificially elevates the count of the current | |
5641 | * processor. | |
5642 | */ | |
f8891e5e | 5643 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
5644 | |
5645 | /* | |
5646 | * Zero the differential counters of the dead processor | |
5647 | * so that the vm statistics are consistent. | |
5648 | * | |
5649 | * This is only okay since the processor is dead and cannot | |
5650 | * race with what we are doing. | |
5651 | */ | |
2bb921e5 | 5652 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
5653 | } |
5654 | return NOTIFY_OK; | |
5655 | } | |
1da177e4 LT |
5656 | |
5657 | void __init page_alloc_init(void) | |
5658 | { | |
5659 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
5660 | } | |
5661 | ||
cb45b0e9 HA |
5662 | /* |
5663 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
5664 | * or min_free_kbytes changes. | |
5665 | */ | |
5666 | static void calculate_totalreserve_pages(void) | |
5667 | { | |
5668 | struct pglist_data *pgdat; | |
5669 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5670 | enum zone_type i, j; |
cb45b0e9 HA |
5671 | |
5672 | for_each_online_pgdat(pgdat) { | |
5673 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
5674 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 5675 | long max = 0; |
cb45b0e9 HA |
5676 | |
5677 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5678 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5679 | if (zone->lowmem_reserve[j] > max) | |
5680 | max = zone->lowmem_reserve[j]; | |
5681 | } | |
5682 | ||
41858966 MG |
5683 | /* we treat the high watermark as reserved pages. */ |
5684 | max += high_wmark_pages(zone); | |
cb45b0e9 | 5685 | |
b40da049 JL |
5686 | if (max > zone->managed_pages) |
5687 | max = zone->managed_pages; | |
cb45b0e9 | 5688 | reserve_pages += max; |
ab8fabd4 JW |
5689 | /* |
5690 | * Lowmem reserves are not available to | |
5691 | * GFP_HIGHUSER page cache allocations and | |
5692 | * kswapd tries to balance zones to their high | |
5693 | * watermark. As a result, neither should be | |
5694 | * regarded as dirtyable memory, to prevent a | |
5695 | * situation where reclaim has to clean pages | |
5696 | * in order to balance the zones. | |
5697 | */ | |
5698 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
5699 | } |
5700 | } | |
ab8fabd4 | 5701 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
5702 | totalreserve_pages = reserve_pages; |
5703 | } | |
5704 | ||
1da177e4 LT |
5705 | /* |
5706 | * setup_per_zone_lowmem_reserve - called whenever | |
5707 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
5708 | * has a correct pages reserved value, so an adequate number of | |
5709 | * pages are left in the zone after a successful __alloc_pages(). | |
5710 | */ | |
5711 | static void setup_per_zone_lowmem_reserve(void) | |
5712 | { | |
5713 | struct pglist_data *pgdat; | |
2f6726e5 | 5714 | enum zone_type j, idx; |
1da177e4 | 5715 | |
ec936fc5 | 5716 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
5717 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5718 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 5719 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
5720 | |
5721 | zone->lowmem_reserve[j] = 0; | |
5722 | ||
2f6726e5 CL |
5723 | idx = j; |
5724 | while (idx) { | |
1da177e4 LT |
5725 | struct zone *lower_zone; |
5726 | ||
2f6726e5 CL |
5727 | idx--; |
5728 | ||
1da177e4 LT |
5729 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
5730 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
5731 | ||
5732 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 5733 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 5734 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 5735 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
5736 | } |
5737 | } | |
5738 | } | |
cb45b0e9 HA |
5739 | |
5740 | /* update totalreserve_pages */ | |
5741 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5742 | } |
5743 | ||
cfd3da1e | 5744 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5745 | { |
5746 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5747 | unsigned long lowmem_pages = 0; | |
5748 | struct zone *zone; | |
5749 | unsigned long flags; | |
5750 | ||
5751 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5752 | for_each_zone(zone) { | |
5753 | if (!is_highmem(zone)) | |
b40da049 | 5754 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
5755 | } |
5756 | ||
5757 | for_each_zone(zone) { | |
ac924c60 AM |
5758 | u64 tmp; |
5759 | ||
1125b4e3 | 5760 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 5761 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 5762 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
5763 | if (is_highmem(zone)) { |
5764 | /* | |
669ed175 NP |
5765 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5766 | * need highmem pages, so cap pages_min to a small | |
5767 | * value here. | |
5768 | * | |
41858966 | 5769 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
5770 | * deltas controls asynch page reclaim, and so should |
5771 | * not be capped for highmem. | |
1da177e4 | 5772 | */ |
90ae8d67 | 5773 | unsigned long min_pages; |
1da177e4 | 5774 | |
b40da049 | 5775 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 5776 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 5777 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5778 | } else { |
669ed175 NP |
5779 | /* |
5780 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5781 | * proportionate to the zone's size. |
5782 | */ | |
41858966 | 5783 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5784 | } |
5785 | ||
41858966 MG |
5786 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
5787 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
49f223a9 | 5788 | |
81c0a2bb | 5789 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
abe5f972 JW |
5790 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
5791 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
81c0a2bb | 5792 | |
56fd56b8 | 5793 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 5794 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5795 | } |
cb45b0e9 HA |
5796 | |
5797 | /* update totalreserve_pages */ | |
5798 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5799 | } |
5800 | ||
cfd3da1e MG |
5801 | /** |
5802 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5803 | * or when memory is hot-{added|removed} | |
5804 | * | |
5805 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5806 | * correctly with respect to min_free_kbytes. | |
5807 | */ | |
5808 | void setup_per_zone_wmarks(void) | |
5809 | { | |
5810 | mutex_lock(&zonelists_mutex); | |
5811 | __setup_per_zone_wmarks(); | |
5812 | mutex_unlock(&zonelists_mutex); | |
5813 | } | |
5814 | ||
55a4462a | 5815 | /* |
556adecb RR |
5816 | * The inactive anon list should be small enough that the VM never has to |
5817 | * do too much work, but large enough that each inactive page has a chance | |
5818 | * to be referenced again before it is swapped out. | |
5819 | * | |
5820 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
5821 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
5822 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
5823 | * the anonymous pages are kept on the inactive list. | |
5824 | * | |
5825 | * total target max | |
5826 | * memory ratio inactive anon | |
5827 | * ------------------------------------- | |
5828 | * 10MB 1 5MB | |
5829 | * 100MB 1 50MB | |
5830 | * 1GB 3 250MB | |
5831 | * 10GB 10 0.9GB | |
5832 | * 100GB 31 3GB | |
5833 | * 1TB 101 10GB | |
5834 | * 10TB 320 32GB | |
5835 | */ | |
1b79acc9 | 5836 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5837 | { |
96cb4df5 | 5838 | unsigned int gb, ratio; |
556adecb | 5839 | |
96cb4df5 | 5840 | /* Zone size in gigabytes */ |
b40da049 | 5841 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 5842 | if (gb) |
556adecb | 5843 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5844 | else |
5845 | ratio = 1; | |
556adecb | 5846 | |
96cb4df5 MK |
5847 | zone->inactive_ratio = ratio; |
5848 | } | |
556adecb | 5849 | |
839a4fcc | 5850 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
5851 | { |
5852 | struct zone *zone; | |
5853 | ||
5854 | for_each_zone(zone) | |
5855 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5856 | } |
5857 | ||
1da177e4 LT |
5858 | /* |
5859 | * Initialise min_free_kbytes. | |
5860 | * | |
5861 | * For small machines we want it small (128k min). For large machines | |
5862 | * we want it large (64MB max). But it is not linear, because network | |
5863 | * bandwidth does not increase linearly with machine size. We use | |
5864 | * | |
b8af2941 | 5865 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
5866 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
5867 | * | |
5868 | * which yields | |
5869 | * | |
5870 | * 16MB: 512k | |
5871 | * 32MB: 724k | |
5872 | * 64MB: 1024k | |
5873 | * 128MB: 1448k | |
5874 | * 256MB: 2048k | |
5875 | * 512MB: 2896k | |
5876 | * 1024MB: 4096k | |
5877 | * 2048MB: 5792k | |
5878 | * 4096MB: 8192k | |
5879 | * 8192MB: 11584k | |
5880 | * 16384MB: 16384k | |
5881 | */ | |
1b79acc9 | 5882 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
5883 | { |
5884 | unsigned long lowmem_kbytes; | |
5f12733e | 5885 | int new_min_free_kbytes; |
1da177e4 LT |
5886 | |
5887 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
5888 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
5889 | ||
5890 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
5891 | min_free_kbytes = new_min_free_kbytes; | |
5892 | if (min_free_kbytes < 128) | |
5893 | min_free_kbytes = 128; | |
5894 | if (min_free_kbytes > 65536) | |
5895 | min_free_kbytes = 65536; | |
5896 | } else { | |
5897 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
5898 | new_min_free_kbytes, user_min_free_kbytes); | |
5899 | } | |
bc75d33f | 5900 | setup_per_zone_wmarks(); |
a6cccdc3 | 5901 | refresh_zone_stat_thresholds(); |
1da177e4 | 5902 | setup_per_zone_lowmem_reserve(); |
556adecb | 5903 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5904 | return 0; |
5905 | } | |
bc75d33f | 5906 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5907 | |
5908 | /* | |
b8af2941 | 5909 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
5910 | * that we can call two helper functions whenever min_free_kbytes |
5911 | * changes. | |
5912 | */ | |
cccad5b9 | 5913 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5914 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5915 | { |
da8c757b HP |
5916 | int rc; |
5917 | ||
5918 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
5919 | if (rc) | |
5920 | return rc; | |
5921 | ||
5f12733e MH |
5922 | if (write) { |
5923 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 5924 | setup_per_zone_wmarks(); |
5f12733e | 5925 | } |
1da177e4 LT |
5926 | return 0; |
5927 | } | |
5928 | ||
9614634f | 5929 | #ifdef CONFIG_NUMA |
cccad5b9 | 5930 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5931 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5932 | { |
5933 | struct zone *zone; | |
5934 | int rc; | |
5935 | ||
8d65af78 | 5936 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5937 | if (rc) |
5938 | return rc; | |
5939 | ||
5940 | for_each_zone(zone) | |
b40da049 | 5941 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
5942 | sysctl_min_unmapped_ratio) / 100; |
5943 | return 0; | |
5944 | } | |
0ff38490 | 5945 | |
cccad5b9 | 5946 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5947 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5948 | { |
5949 | struct zone *zone; | |
5950 | int rc; | |
5951 | ||
8d65af78 | 5952 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5953 | if (rc) |
5954 | return rc; | |
5955 | ||
5956 | for_each_zone(zone) | |
b40da049 | 5957 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
5958 | sysctl_min_slab_ratio) / 100; |
5959 | return 0; | |
5960 | } | |
9614634f CL |
5961 | #endif |
5962 | ||
1da177e4 LT |
5963 | /* |
5964 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5965 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5966 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5967 | * | |
5968 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5969 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5970 | * if in function of the boot time zone sizes. |
5971 | */ | |
cccad5b9 | 5972 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5973 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5974 | { |
8d65af78 | 5975 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5976 | setup_per_zone_lowmem_reserve(); |
5977 | return 0; | |
5978 | } | |
5979 | ||
8ad4b1fb RS |
5980 | /* |
5981 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
5982 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
5983 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 5984 | */ |
cccad5b9 | 5985 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5986 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5987 | { |
5988 | struct zone *zone; | |
7cd2b0a3 | 5989 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
5990 | int ret; |
5991 | ||
7cd2b0a3 DR |
5992 | mutex_lock(&pcp_batch_high_lock); |
5993 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
5994 | ||
8d65af78 | 5995 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
5996 | if (!write || ret < 0) |
5997 | goto out; | |
5998 | ||
5999 | /* Sanity checking to avoid pcp imbalance */ | |
6000 | if (percpu_pagelist_fraction && | |
6001 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
6002 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
6003 | ret = -EINVAL; | |
6004 | goto out; | |
6005 | } | |
6006 | ||
6007 | /* No change? */ | |
6008 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
6009 | goto out; | |
c8e251fa | 6010 | |
364df0eb | 6011 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
6012 | unsigned int cpu; |
6013 | ||
22a7f12b | 6014 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
6015 | pageset_set_high_and_batch(zone, |
6016 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 6017 | } |
7cd2b0a3 | 6018 | out: |
c8e251fa | 6019 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6020 | return ret; |
8ad4b1fb RS |
6021 | } |
6022 | ||
f034b5d4 | 6023 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
6024 | |
6025 | #ifdef CONFIG_NUMA | |
6026 | static int __init set_hashdist(char *str) | |
6027 | { | |
6028 | if (!str) | |
6029 | return 0; | |
6030 | hashdist = simple_strtoul(str, &str, 0); | |
6031 | return 1; | |
6032 | } | |
6033 | __setup("hashdist=", set_hashdist); | |
6034 | #endif | |
6035 | ||
6036 | /* | |
6037 | * allocate a large system hash table from bootmem | |
6038 | * - it is assumed that the hash table must contain an exact power-of-2 | |
6039 | * quantity of entries | |
6040 | * - limit is the number of hash buckets, not the total allocation size | |
6041 | */ | |
6042 | void *__init alloc_large_system_hash(const char *tablename, | |
6043 | unsigned long bucketsize, | |
6044 | unsigned long numentries, | |
6045 | int scale, | |
6046 | int flags, | |
6047 | unsigned int *_hash_shift, | |
6048 | unsigned int *_hash_mask, | |
31fe62b9 TB |
6049 | unsigned long low_limit, |
6050 | unsigned long high_limit) | |
1da177e4 | 6051 | { |
31fe62b9 | 6052 | unsigned long long max = high_limit; |
1da177e4 LT |
6053 | unsigned long log2qty, size; |
6054 | void *table = NULL; | |
6055 | ||
6056 | /* allow the kernel cmdline to have a say */ | |
6057 | if (!numentries) { | |
6058 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 6059 | numentries = nr_kernel_pages; |
a7e83318 JZ |
6060 | |
6061 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
6062 | if (PAGE_SHIFT < 20) | |
6063 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
6064 | |
6065 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
6066 | if (scale > PAGE_SHIFT) | |
6067 | numentries >>= (scale - PAGE_SHIFT); | |
6068 | else | |
6069 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
6070 | |
6071 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
6072 | if (unlikely(flags & HASH_SMALL)) { |
6073 | /* Makes no sense without HASH_EARLY */ | |
6074 | WARN_ON(!(flags & HASH_EARLY)); | |
6075 | if (!(numentries >> *_hash_shift)) { | |
6076 | numentries = 1UL << *_hash_shift; | |
6077 | BUG_ON(!numentries); | |
6078 | } | |
6079 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 6080 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 6081 | } |
6e692ed3 | 6082 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
6083 | |
6084 | /* limit allocation size to 1/16 total memory by default */ | |
6085 | if (max == 0) { | |
6086 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
6087 | do_div(max, bucketsize); | |
6088 | } | |
074b8517 | 6089 | max = min(max, 0x80000000ULL); |
1da177e4 | 6090 | |
31fe62b9 TB |
6091 | if (numentries < low_limit) |
6092 | numentries = low_limit; | |
1da177e4 LT |
6093 | if (numentries > max) |
6094 | numentries = max; | |
6095 | ||
f0d1b0b3 | 6096 | log2qty = ilog2(numentries); |
1da177e4 LT |
6097 | |
6098 | do { | |
6099 | size = bucketsize << log2qty; | |
6100 | if (flags & HASH_EARLY) | |
6782832e | 6101 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
6102 | else if (hashdist) |
6103 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
6104 | else { | |
1037b83b ED |
6105 | /* |
6106 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
6107 | * some pages at the end of hash table which |
6108 | * alloc_pages_exact() automatically does | |
1037b83b | 6109 | */ |
264ef8a9 | 6110 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 6111 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
6112 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
6113 | } | |
1da177e4 LT |
6114 | } |
6115 | } while (!table && size > PAGE_SIZE && --log2qty); | |
6116 | ||
6117 | if (!table) | |
6118 | panic("Failed to allocate %s hash table\n", tablename); | |
6119 | ||
f241e660 | 6120 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 6121 | tablename, |
f241e660 | 6122 | (1UL << log2qty), |
f0d1b0b3 | 6123 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
6124 | size); |
6125 | ||
6126 | if (_hash_shift) | |
6127 | *_hash_shift = log2qty; | |
6128 | if (_hash_mask) | |
6129 | *_hash_mask = (1 << log2qty) - 1; | |
6130 | ||
6131 | return table; | |
6132 | } | |
a117e66e | 6133 | |
835c134e MG |
6134 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
6135 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
6136 | unsigned long pfn) | |
6137 | { | |
6138 | #ifdef CONFIG_SPARSEMEM | |
6139 | return __pfn_to_section(pfn)->pageblock_flags; | |
6140 | #else | |
6141 | return zone->pageblock_flags; | |
6142 | #endif /* CONFIG_SPARSEMEM */ | |
6143 | } | |
6144 | ||
6145 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
6146 | { | |
6147 | #ifdef CONFIG_SPARSEMEM | |
6148 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 6149 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 6150 | #else |
c060f943 | 6151 | pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 6152 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
6153 | #endif /* CONFIG_SPARSEMEM */ |
6154 | } | |
6155 | ||
6156 | /** | |
1aab4d77 | 6157 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e | 6158 | * @page: The page within the block of interest |
1aab4d77 RD |
6159 | * @pfn: The target page frame number |
6160 | * @end_bitidx: The last bit of interest to retrieve | |
6161 | * @mask: mask of bits that the caller is interested in | |
6162 | * | |
6163 | * Return: pageblock_bits flags | |
835c134e | 6164 | */ |
dc4b0caf | 6165 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
e58469ba MG |
6166 | unsigned long end_bitidx, |
6167 | unsigned long mask) | |
835c134e MG |
6168 | { |
6169 | struct zone *zone; | |
6170 | unsigned long *bitmap; | |
dc4b0caf | 6171 | unsigned long bitidx, word_bitidx; |
e58469ba | 6172 | unsigned long word; |
835c134e MG |
6173 | |
6174 | zone = page_zone(page); | |
835c134e MG |
6175 | bitmap = get_pageblock_bitmap(zone, pfn); |
6176 | bitidx = pfn_to_bitidx(zone, pfn); | |
e58469ba MG |
6177 | word_bitidx = bitidx / BITS_PER_LONG; |
6178 | bitidx &= (BITS_PER_LONG-1); | |
835c134e | 6179 | |
e58469ba MG |
6180 | word = bitmap[word_bitidx]; |
6181 | bitidx += end_bitidx; | |
6182 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
835c134e MG |
6183 | } |
6184 | ||
6185 | /** | |
dc4b0caf | 6186 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e | 6187 | * @page: The page within the block of interest |
835c134e | 6188 | * @flags: The flags to set |
1aab4d77 RD |
6189 | * @pfn: The target page frame number |
6190 | * @end_bitidx: The last bit of interest | |
6191 | * @mask: mask of bits that the caller is interested in | |
835c134e | 6192 | */ |
dc4b0caf MG |
6193 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
6194 | unsigned long pfn, | |
e58469ba MG |
6195 | unsigned long end_bitidx, |
6196 | unsigned long mask) | |
835c134e MG |
6197 | { |
6198 | struct zone *zone; | |
6199 | unsigned long *bitmap; | |
dc4b0caf | 6200 | unsigned long bitidx, word_bitidx; |
e58469ba MG |
6201 | unsigned long old_word, word; |
6202 | ||
6203 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
835c134e MG |
6204 | |
6205 | zone = page_zone(page); | |
835c134e MG |
6206 | bitmap = get_pageblock_bitmap(zone, pfn); |
6207 | bitidx = pfn_to_bitidx(zone, pfn); | |
e58469ba MG |
6208 | word_bitidx = bitidx / BITS_PER_LONG; |
6209 | bitidx &= (BITS_PER_LONG-1); | |
6210 | ||
309381fe | 6211 | VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page); |
835c134e | 6212 | |
e58469ba MG |
6213 | bitidx += end_bitidx; |
6214 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
6215 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
6216 | ||
6217 | word = ACCESS_ONCE(bitmap[word_bitidx]); | |
6218 | for (;;) { | |
6219 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
6220 | if (word == old_word) | |
6221 | break; | |
6222 | word = old_word; | |
6223 | } | |
835c134e | 6224 | } |
a5d76b54 KH |
6225 | |
6226 | /* | |
80934513 MK |
6227 | * This function checks whether pageblock includes unmovable pages or not. |
6228 | * If @count is not zero, it is okay to include less @count unmovable pages | |
6229 | * | |
b8af2941 | 6230 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
6231 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
6232 | * expect this function should be exact. | |
a5d76b54 | 6233 | */ |
b023f468 WC |
6234 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
6235 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
6236 | { |
6237 | unsigned long pfn, iter, found; | |
47118af0 MN |
6238 | int mt; |
6239 | ||
49ac8255 KH |
6240 | /* |
6241 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 6242 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
6243 | */ |
6244 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 6245 | return false; |
47118af0 MN |
6246 | mt = get_pageblock_migratetype(page); |
6247 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 6248 | return false; |
49ac8255 KH |
6249 | |
6250 | pfn = page_to_pfn(page); | |
6251 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
6252 | unsigned long check = pfn + iter; | |
6253 | ||
29723fcc | 6254 | if (!pfn_valid_within(check)) |
49ac8255 | 6255 | continue; |
29723fcc | 6256 | |
49ac8255 | 6257 | page = pfn_to_page(check); |
c8721bbb NH |
6258 | |
6259 | /* | |
6260 | * Hugepages are not in LRU lists, but they're movable. | |
6261 | * We need not scan over tail pages bacause we don't | |
6262 | * handle each tail page individually in migration. | |
6263 | */ | |
6264 | if (PageHuge(page)) { | |
6265 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
6266 | continue; | |
6267 | } | |
6268 | ||
97d255c8 MK |
6269 | /* |
6270 | * We can't use page_count without pin a page | |
6271 | * because another CPU can free compound page. | |
6272 | * This check already skips compound tails of THP | |
6273 | * because their page->_count is zero at all time. | |
6274 | */ | |
6275 | if (!atomic_read(&page->_count)) { | |
49ac8255 KH |
6276 | if (PageBuddy(page)) |
6277 | iter += (1 << page_order(page)) - 1; | |
6278 | continue; | |
6279 | } | |
97d255c8 | 6280 | |
b023f468 WC |
6281 | /* |
6282 | * The HWPoisoned page may be not in buddy system, and | |
6283 | * page_count() is not 0. | |
6284 | */ | |
6285 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6286 | continue; | |
6287 | ||
49ac8255 KH |
6288 | if (!PageLRU(page)) |
6289 | found++; | |
6290 | /* | |
6b4f7799 JW |
6291 | * If there are RECLAIMABLE pages, we need to check |
6292 | * it. But now, memory offline itself doesn't call | |
6293 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
6294 | */ |
6295 | /* | |
6296 | * If the page is not RAM, page_count()should be 0. | |
6297 | * we don't need more check. This is an _used_ not-movable page. | |
6298 | * | |
6299 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6300 | * is set to both of a memory hole page and a _used_ kernel | |
6301 | * page at boot. | |
6302 | */ | |
6303 | if (found > count) | |
80934513 | 6304 | return true; |
49ac8255 | 6305 | } |
80934513 | 6306 | return false; |
49ac8255 KH |
6307 | } |
6308 | ||
6309 | bool is_pageblock_removable_nolock(struct page *page) | |
6310 | { | |
656a0706 MH |
6311 | struct zone *zone; |
6312 | unsigned long pfn; | |
687875fb MH |
6313 | |
6314 | /* | |
6315 | * We have to be careful here because we are iterating over memory | |
6316 | * sections which are not zone aware so we might end up outside of | |
6317 | * the zone but still within the section. | |
656a0706 MH |
6318 | * We have to take care about the node as well. If the node is offline |
6319 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 6320 | */ |
656a0706 MH |
6321 | if (!node_online(page_to_nid(page))) |
6322 | return false; | |
6323 | ||
6324 | zone = page_zone(page); | |
6325 | pfn = page_to_pfn(page); | |
108bcc96 | 6326 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
6327 | return false; |
6328 | ||
b023f468 | 6329 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6330 | } |
0c0e6195 | 6331 | |
041d3a8c MN |
6332 | #ifdef CONFIG_CMA |
6333 | ||
6334 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6335 | { | |
6336 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6337 | pageblock_nr_pages) - 1); | |
6338 | } | |
6339 | ||
6340 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6341 | { | |
6342 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6343 | pageblock_nr_pages)); | |
6344 | } | |
6345 | ||
041d3a8c | 6346 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6347 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6348 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6349 | { |
6350 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6351 | unsigned long nr_reclaimed; |
041d3a8c MN |
6352 | unsigned long pfn = start; |
6353 | unsigned int tries = 0; | |
6354 | int ret = 0; | |
6355 | ||
be49a6e1 | 6356 | migrate_prep(); |
041d3a8c | 6357 | |
bb13ffeb | 6358 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6359 | if (fatal_signal_pending(current)) { |
6360 | ret = -EINTR; | |
6361 | break; | |
6362 | } | |
6363 | ||
bb13ffeb MG |
6364 | if (list_empty(&cc->migratepages)) { |
6365 | cc->nr_migratepages = 0; | |
edc2ca61 | 6366 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
6367 | if (!pfn) { |
6368 | ret = -EINTR; | |
6369 | break; | |
6370 | } | |
6371 | tries = 0; | |
6372 | } else if (++tries == 5) { | |
6373 | ret = ret < 0 ? ret : -EBUSY; | |
6374 | break; | |
6375 | } | |
6376 | ||
beb51eaa MK |
6377 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6378 | &cc->migratepages); | |
6379 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6380 | |
9c620e2b | 6381 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 6382 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 6383 | } |
2a6f5124 SP |
6384 | if (ret < 0) { |
6385 | putback_movable_pages(&cc->migratepages); | |
6386 | return ret; | |
6387 | } | |
6388 | return 0; | |
041d3a8c MN |
6389 | } |
6390 | ||
6391 | /** | |
6392 | * alloc_contig_range() -- tries to allocate given range of pages | |
6393 | * @start: start PFN to allocate | |
6394 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
6395 | * @migratetype: migratetype of the underlaying pageblocks (either |
6396 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
6397 | * in range must have the same migratetype and it must | |
6398 | * be either of the two. | |
041d3a8c MN |
6399 | * |
6400 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
6401 | * aligned, however it's the caller's responsibility to guarantee that | |
6402 | * we are the only thread that changes migrate type of pageblocks the | |
6403 | * pages fall in. | |
6404 | * | |
6405 | * The PFN range must belong to a single zone. | |
6406 | * | |
6407 | * Returns zero on success or negative error code. On success all | |
6408 | * pages which PFN is in [start, end) are allocated for the caller and | |
6409 | * need to be freed with free_contig_range(). | |
6410 | */ | |
0815f3d8 MN |
6411 | int alloc_contig_range(unsigned long start, unsigned long end, |
6412 | unsigned migratetype) | |
041d3a8c | 6413 | { |
041d3a8c MN |
6414 | unsigned long outer_start, outer_end; |
6415 | int ret = 0, order; | |
6416 | ||
bb13ffeb MG |
6417 | struct compact_control cc = { |
6418 | .nr_migratepages = 0, | |
6419 | .order = -1, | |
6420 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 6421 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
6422 | .ignore_skip_hint = true, |
6423 | }; | |
6424 | INIT_LIST_HEAD(&cc.migratepages); | |
6425 | ||
041d3a8c MN |
6426 | /* |
6427 | * What we do here is we mark all pageblocks in range as | |
6428 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6429 | * have different sizes, and due to the way page allocator | |
6430 | * work, we align the range to biggest of the two pages so | |
6431 | * that page allocator won't try to merge buddies from | |
6432 | * different pageblocks and change MIGRATE_ISOLATE to some | |
6433 | * other migration type. | |
6434 | * | |
6435 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6436 | * migrate the pages from an unaligned range (ie. pages that | |
6437 | * we are interested in). This will put all the pages in | |
6438 | * range back to page allocator as MIGRATE_ISOLATE. | |
6439 | * | |
6440 | * When this is done, we take the pages in range from page | |
6441 | * allocator removing them from the buddy system. This way | |
6442 | * page allocator will never consider using them. | |
6443 | * | |
6444 | * This lets us mark the pageblocks back as | |
6445 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6446 | * aligned range but not in the unaligned, original range are | |
6447 | * put back to page allocator so that buddy can use them. | |
6448 | */ | |
6449 | ||
6450 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
6451 | pfn_max_align_up(end), migratetype, |
6452 | false); | |
041d3a8c | 6453 | if (ret) |
86a595f9 | 6454 | return ret; |
041d3a8c | 6455 | |
bb13ffeb | 6456 | ret = __alloc_contig_migrate_range(&cc, start, end); |
041d3a8c MN |
6457 | if (ret) |
6458 | goto done; | |
6459 | ||
6460 | /* | |
6461 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
6462 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
6463 | * more, all pages in [start, end) are free in page allocator. | |
6464 | * What we are going to do is to allocate all pages from | |
6465 | * [start, end) (that is remove them from page allocator). | |
6466 | * | |
6467 | * The only problem is that pages at the beginning and at the | |
6468 | * end of interesting range may be not aligned with pages that | |
6469 | * page allocator holds, ie. they can be part of higher order | |
6470 | * pages. Because of this, we reserve the bigger range and | |
6471 | * once this is done free the pages we are not interested in. | |
6472 | * | |
6473 | * We don't have to hold zone->lock here because the pages are | |
6474 | * isolated thus they won't get removed from buddy. | |
6475 | */ | |
6476 | ||
6477 | lru_add_drain_all(); | |
510f5507 | 6478 | drain_all_pages(cc.zone); |
041d3a8c MN |
6479 | |
6480 | order = 0; | |
6481 | outer_start = start; | |
6482 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
6483 | if (++order >= MAX_ORDER) { | |
6484 | ret = -EBUSY; | |
6485 | goto done; | |
6486 | } | |
6487 | outer_start &= ~0UL << order; | |
6488 | } | |
6489 | ||
6490 | /* Make sure the range is really isolated. */ | |
b023f468 | 6491 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
6492 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
6493 | __func__, outer_start, end); | |
041d3a8c MN |
6494 | ret = -EBUSY; |
6495 | goto done; | |
6496 | } | |
6497 | ||
49f223a9 | 6498 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 6499 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6500 | if (!outer_end) { |
6501 | ret = -EBUSY; | |
6502 | goto done; | |
6503 | } | |
6504 | ||
6505 | /* Free head and tail (if any) */ | |
6506 | if (start != outer_start) | |
6507 | free_contig_range(outer_start, start - outer_start); | |
6508 | if (end != outer_end) | |
6509 | free_contig_range(end, outer_end - end); | |
6510 | ||
6511 | done: | |
6512 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 6513 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
6514 | return ret; |
6515 | } | |
6516 | ||
6517 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
6518 | { | |
bcc2b02f MS |
6519 | unsigned int count = 0; |
6520 | ||
6521 | for (; nr_pages--; pfn++) { | |
6522 | struct page *page = pfn_to_page(pfn); | |
6523 | ||
6524 | count += page_count(page) != 1; | |
6525 | __free_page(page); | |
6526 | } | |
6527 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
6528 | } |
6529 | #endif | |
6530 | ||
4ed7e022 | 6531 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
6532 | /* |
6533 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
6534 | * page high values need to be recalulated. | |
6535 | */ | |
4ed7e022 JL |
6536 | void __meminit zone_pcp_update(struct zone *zone) |
6537 | { | |
0a647f38 | 6538 | unsigned cpu; |
c8e251fa | 6539 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 6540 | for_each_possible_cpu(cpu) |
169f6c19 CS |
6541 | pageset_set_high_and_batch(zone, |
6542 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 6543 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
6544 | } |
6545 | #endif | |
6546 | ||
340175b7 JL |
6547 | void zone_pcp_reset(struct zone *zone) |
6548 | { | |
6549 | unsigned long flags; | |
5a883813 MK |
6550 | int cpu; |
6551 | struct per_cpu_pageset *pset; | |
340175b7 JL |
6552 | |
6553 | /* avoid races with drain_pages() */ | |
6554 | local_irq_save(flags); | |
6555 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
6556 | for_each_online_cpu(cpu) { |
6557 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6558 | drain_zonestat(zone, pset); | |
6559 | } | |
340175b7 JL |
6560 | free_percpu(zone->pageset); |
6561 | zone->pageset = &boot_pageset; | |
6562 | } | |
6563 | local_irq_restore(flags); | |
6564 | } | |
6565 | ||
6dcd73d7 | 6566 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
6567 | /* |
6568 | * All pages in the range must be isolated before calling this. | |
6569 | */ | |
6570 | void | |
6571 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
6572 | { | |
6573 | struct page *page; | |
6574 | struct zone *zone; | |
7aeb09f9 | 6575 | unsigned int order, i; |
0c0e6195 KH |
6576 | unsigned long pfn; |
6577 | unsigned long flags; | |
6578 | /* find the first valid pfn */ | |
6579 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
6580 | if (pfn_valid(pfn)) | |
6581 | break; | |
6582 | if (pfn == end_pfn) | |
6583 | return; | |
6584 | zone = page_zone(pfn_to_page(pfn)); | |
6585 | spin_lock_irqsave(&zone->lock, flags); | |
6586 | pfn = start_pfn; | |
6587 | while (pfn < end_pfn) { | |
6588 | if (!pfn_valid(pfn)) { | |
6589 | pfn++; | |
6590 | continue; | |
6591 | } | |
6592 | page = pfn_to_page(pfn); | |
b023f468 WC |
6593 | /* |
6594 | * The HWPoisoned page may be not in buddy system, and | |
6595 | * page_count() is not 0. | |
6596 | */ | |
6597 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6598 | pfn++; | |
6599 | SetPageReserved(page); | |
6600 | continue; | |
6601 | } | |
6602 | ||
0c0e6195 KH |
6603 | BUG_ON(page_count(page)); |
6604 | BUG_ON(!PageBuddy(page)); | |
6605 | order = page_order(page); | |
6606 | #ifdef CONFIG_DEBUG_VM | |
6607 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
6608 | pfn, 1 << order, end_pfn); | |
6609 | #endif | |
6610 | list_del(&page->lru); | |
6611 | rmv_page_order(page); | |
6612 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
6613 | for (i = 0; i < (1 << order); i++) |
6614 | SetPageReserved((page+i)); | |
6615 | pfn += (1 << order); | |
6616 | } | |
6617 | spin_unlock_irqrestore(&zone->lock, flags); | |
6618 | } | |
6619 | #endif | |
8d22ba1b WF |
6620 | |
6621 | #ifdef CONFIG_MEMORY_FAILURE | |
6622 | bool is_free_buddy_page(struct page *page) | |
6623 | { | |
6624 | struct zone *zone = page_zone(page); | |
6625 | unsigned long pfn = page_to_pfn(page); | |
6626 | unsigned long flags; | |
7aeb09f9 | 6627 | unsigned int order; |
8d22ba1b WF |
6628 | |
6629 | spin_lock_irqsave(&zone->lock, flags); | |
6630 | for (order = 0; order < MAX_ORDER; order++) { | |
6631 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
6632 | ||
6633 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
6634 | break; | |
6635 | } | |
6636 | spin_unlock_irqrestore(&zone->lock, flags); | |
6637 | ||
6638 | return order < MAX_ORDER; | |
6639 | } | |
6640 | #endif |