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> |
b8c73fc2 | 28 | #include <linux/kasan.h> |
1da177e4 LT |
29 | #include <linux/module.h> |
30 | #include <linux/suspend.h> | |
31 | #include <linux/pagevec.h> | |
32 | #include <linux/blkdev.h> | |
33 | #include <linux/slab.h> | |
a238ab5b | 34 | #include <linux/ratelimit.h> |
5a3135c2 | 35 | #include <linux/oom.h> |
1da177e4 LT |
36 | #include <linux/notifier.h> |
37 | #include <linux/topology.h> | |
38 | #include <linux/sysctl.h> | |
39 | #include <linux/cpu.h> | |
40 | #include <linux/cpuset.h> | |
bdc8cb98 | 41 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
42 | #include <linux/nodemask.h> |
43 | #include <linux/vmalloc.h> | |
a6cccdc3 | 44 | #include <linux/vmstat.h> |
4be38e35 | 45 | #include <linux/mempolicy.h> |
6811378e | 46 | #include <linux/stop_machine.h> |
c713216d MG |
47 | #include <linux/sort.h> |
48 | #include <linux/pfn.h> | |
3fcfab16 | 49 | #include <linux/backing-dev.h> |
933e312e | 50 | #include <linux/fault-inject.h> |
a5d76b54 | 51 | #include <linux/page-isolation.h> |
eefa864b | 52 | #include <linux/page_ext.h> |
3ac7fe5a | 53 | #include <linux/debugobjects.h> |
dbb1f81c | 54 | #include <linux/kmemleak.h> |
56de7263 | 55 | #include <linux/compaction.h> |
0d3d062a | 56 | #include <trace/events/kmem.h> |
268bb0ce | 57 | #include <linux/prefetch.h> |
6e543d57 | 58 | #include <linux/mm_inline.h> |
041d3a8c | 59 | #include <linux/migrate.h> |
e30825f1 | 60 | #include <linux/page_ext.h> |
949f7ec5 | 61 | #include <linux/hugetlb.h> |
8bd75c77 | 62 | #include <linux/sched/rt.h> |
48c96a36 | 63 | #include <linux/page_owner.h> |
1da177e4 | 64 | |
7ee3d4e8 | 65 | #include <asm/sections.h> |
1da177e4 | 66 | #include <asm/tlbflush.h> |
ac924c60 | 67 | #include <asm/div64.h> |
1da177e4 LT |
68 | #include "internal.h" |
69 | ||
c8e251fa CS |
70 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
71 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 72 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 73 | |
72812019 LS |
74 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
75 | DEFINE_PER_CPU(int, numa_node); | |
76 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
77 | #endif | |
78 | ||
7aac7898 LS |
79 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
80 | /* | |
81 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
82 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
83 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
84 | * defined in <linux/topology.h>. | |
85 | */ | |
86 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
87 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 88 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
89 | #endif |
90 | ||
1da177e4 | 91 | /* |
13808910 | 92 | * Array of node states. |
1da177e4 | 93 | */ |
13808910 CL |
94 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
95 | [N_POSSIBLE] = NODE_MASK_ALL, | |
96 | [N_ONLINE] = { { [0] = 1UL } }, | |
97 | #ifndef CONFIG_NUMA | |
98 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
99 | #ifdef CONFIG_HIGHMEM | |
100 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
101 | #endif |
102 | #ifdef CONFIG_MOVABLE_NODE | |
103 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
104 | #endif |
105 | [N_CPU] = { { [0] = 1UL } }, | |
106 | #endif /* NUMA */ | |
107 | }; | |
108 | EXPORT_SYMBOL(node_states); | |
109 | ||
c3d5f5f0 JL |
110 | /* Protect totalram_pages and zone->managed_pages */ |
111 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
112 | ||
6c231b7b | 113 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 114 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 115 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 JW |
116 | /* |
117 | * When calculating the number of globally allowed dirty pages, there | |
118 | * is a certain number of per-zone reserves that should not be | |
119 | * considered dirtyable memory. This is the sum of those reserves | |
120 | * over all existing zones that contribute dirtyable memory. | |
121 | */ | |
122 | unsigned long dirty_balance_reserve __read_mostly; | |
123 | ||
1b76b02f | 124 | int percpu_pagelist_fraction; |
dcce284a | 125 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 126 | |
452aa699 RW |
127 | #ifdef CONFIG_PM_SLEEP |
128 | /* | |
129 | * The following functions are used by the suspend/hibernate code to temporarily | |
130 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
131 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
132 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
133 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
134 | * guaranteed not to run in parallel with that modification). | |
135 | */ | |
c9e664f1 RW |
136 | |
137 | static gfp_t saved_gfp_mask; | |
138 | ||
139 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
140 | { |
141 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
142 | if (saved_gfp_mask) { |
143 | gfp_allowed_mask = saved_gfp_mask; | |
144 | saved_gfp_mask = 0; | |
145 | } | |
452aa699 RW |
146 | } |
147 | ||
c9e664f1 | 148 | void pm_restrict_gfp_mask(void) |
452aa699 | 149 | { |
452aa699 | 150 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
151 | WARN_ON(saved_gfp_mask); |
152 | saved_gfp_mask = gfp_allowed_mask; | |
153 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 | 154 | } |
f90ac398 MG |
155 | |
156 | bool pm_suspended_storage(void) | |
157 | { | |
158 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) | |
159 | return false; | |
160 | return true; | |
161 | } | |
452aa699 RW |
162 | #endif /* CONFIG_PM_SLEEP */ |
163 | ||
d9c23400 MG |
164 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
165 | int pageblock_order __read_mostly; | |
166 | #endif | |
167 | ||
d98c7a09 | 168 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 169 | |
1da177e4 LT |
170 | /* |
171 | * results with 256, 32 in the lowmem_reserve sysctl: | |
172 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
173 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
174 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
175 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 176 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
177 | * |
178 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
179 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 180 | */ |
2f1b6248 | 181 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 182 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 183 | 256, |
4b51d669 | 184 | #endif |
fb0e7942 | 185 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 186 | 256, |
fb0e7942 | 187 | #endif |
e53ef38d | 188 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 189 | 32, |
e53ef38d | 190 | #endif |
2a1e274a | 191 | 32, |
2f1b6248 | 192 | }; |
1da177e4 LT |
193 | |
194 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 195 | |
15ad7cdc | 196 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 197 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 198 | "DMA", |
4b51d669 | 199 | #endif |
fb0e7942 | 200 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 201 | "DMA32", |
fb0e7942 | 202 | #endif |
2f1b6248 | 203 | "Normal", |
e53ef38d | 204 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 205 | "HighMem", |
e53ef38d | 206 | #endif |
2a1e274a | 207 | "Movable", |
2f1b6248 CL |
208 | }; |
209 | ||
1da177e4 | 210 | int min_free_kbytes = 1024; |
42aa83cb | 211 | int user_min_free_kbytes = -1; |
1da177e4 | 212 | |
2c85f51d JB |
213 | static unsigned long __meminitdata nr_kernel_pages; |
214 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 215 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 216 | |
0ee332c1 TH |
217 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
218 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
219 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
220 | static unsigned long __initdata required_kernelcore; | |
221 | static unsigned long __initdata required_movablecore; | |
222 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
223 | ||
224 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
225 | int movable_zone; | |
226 | EXPORT_SYMBOL(movable_zone); | |
227 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 228 | |
418508c1 MS |
229 | #if MAX_NUMNODES > 1 |
230 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 231 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 232 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 233 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
234 | #endif |
235 | ||
9ef9acb0 MG |
236 | int page_group_by_mobility_disabled __read_mostly; |
237 | ||
3a80a7fa MG |
238 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
239 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
240 | { | |
241 | pgdat->first_deferred_pfn = ULONG_MAX; | |
242 | } | |
243 | ||
244 | /* Returns true if the struct page for the pfn is uninitialised */ | |
245 | static inline bool __defermem_init early_page_uninitialised(unsigned long pfn) | |
246 | { | |
247 | int nid = early_pfn_to_nid(pfn); | |
248 | ||
249 | if (pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
250 | return true; | |
251 | ||
252 | return false; | |
253 | } | |
254 | ||
255 | /* | |
256 | * Returns false when the remaining initialisation should be deferred until | |
257 | * later in the boot cycle when it can be parallelised. | |
258 | */ | |
259 | static inline bool update_defer_init(pg_data_t *pgdat, | |
260 | unsigned long pfn, unsigned long zone_end, | |
261 | unsigned long *nr_initialised) | |
262 | { | |
263 | /* Always populate low zones for address-contrained allocations */ | |
264 | if (zone_end < pgdat_end_pfn(pgdat)) | |
265 | return true; | |
266 | ||
267 | /* Initialise at least 2G of the highest zone */ | |
268 | (*nr_initialised)++; | |
269 | if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) && | |
270 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { | |
271 | pgdat->first_deferred_pfn = pfn; | |
272 | return false; | |
273 | } | |
274 | ||
275 | return true; | |
276 | } | |
277 | #else | |
278 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
279 | { | |
280 | } | |
281 | ||
282 | static inline bool early_page_uninitialised(unsigned long pfn) | |
283 | { | |
284 | return false; | |
285 | } | |
286 | ||
287 | static inline bool update_defer_init(pg_data_t *pgdat, | |
288 | unsigned long pfn, unsigned long zone_end, | |
289 | unsigned long *nr_initialised) | |
290 | { | |
291 | return true; | |
292 | } | |
293 | #endif | |
294 | ||
295 | ||
ee6f509c | 296 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 297 | { |
5d0f3f72 KM |
298 | if (unlikely(page_group_by_mobility_disabled && |
299 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
300 | migratetype = MIGRATE_UNMOVABLE; |
301 | ||
b2a0ac88 MG |
302 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
303 | PB_migrate, PB_migrate_end); | |
304 | } | |
305 | ||
13e7444b | 306 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 307 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 308 | { |
bdc8cb98 DH |
309 | int ret = 0; |
310 | unsigned seq; | |
311 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 312 | unsigned long sp, start_pfn; |
c6a57e19 | 313 | |
bdc8cb98 DH |
314 | do { |
315 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
316 | start_pfn = zone->zone_start_pfn; |
317 | sp = zone->spanned_pages; | |
108bcc96 | 318 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
319 | ret = 1; |
320 | } while (zone_span_seqretry(zone, seq)); | |
321 | ||
b5e6a5a2 | 322 | if (ret) |
613813e8 DH |
323 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
324 | pfn, zone_to_nid(zone), zone->name, | |
325 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 326 | |
bdc8cb98 | 327 | return ret; |
c6a57e19 DH |
328 | } |
329 | ||
330 | static int page_is_consistent(struct zone *zone, struct page *page) | |
331 | { | |
14e07298 | 332 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 333 | return 0; |
1da177e4 | 334 | if (zone != page_zone(page)) |
c6a57e19 DH |
335 | return 0; |
336 | ||
337 | return 1; | |
338 | } | |
339 | /* | |
340 | * Temporary debugging check for pages not lying within a given zone. | |
341 | */ | |
342 | static int bad_range(struct zone *zone, struct page *page) | |
343 | { | |
344 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 345 | return 1; |
c6a57e19 DH |
346 | if (!page_is_consistent(zone, page)) |
347 | return 1; | |
348 | ||
1da177e4 LT |
349 | return 0; |
350 | } | |
13e7444b NP |
351 | #else |
352 | static inline int bad_range(struct zone *zone, struct page *page) | |
353 | { | |
354 | return 0; | |
355 | } | |
356 | #endif | |
357 | ||
d230dec1 KS |
358 | static void bad_page(struct page *page, const char *reason, |
359 | unsigned long bad_flags) | |
1da177e4 | 360 | { |
d936cf9b HD |
361 | static unsigned long resume; |
362 | static unsigned long nr_shown; | |
363 | static unsigned long nr_unshown; | |
364 | ||
2a7684a2 WF |
365 | /* Don't complain about poisoned pages */ |
366 | if (PageHWPoison(page)) { | |
22b751c3 | 367 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
368 | return; |
369 | } | |
370 | ||
d936cf9b HD |
371 | /* |
372 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
373 | * or allow a steady drip of one report per second. | |
374 | */ | |
375 | if (nr_shown == 60) { | |
376 | if (time_before(jiffies, resume)) { | |
377 | nr_unshown++; | |
378 | goto out; | |
379 | } | |
380 | if (nr_unshown) { | |
1e9e6365 HD |
381 | printk(KERN_ALERT |
382 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
383 | nr_unshown); |
384 | nr_unshown = 0; | |
385 | } | |
386 | nr_shown = 0; | |
387 | } | |
388 | if (nr_shown++ == 0) | |
389 | resume = jiffies + 60 * HZ; | |
390 | ||
1e9e6365 | 391 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 392 | current->comm, page_to_pfn(page)); |
f0b791a3 | 393 | dump_page_badflags(page, reason, bad_flags); |
3dc14741 | 394 | |
4f31888c | 395 | print_modules(); |
1da177e4 | 396 | dump_stack(); |
d936cf9b | 397 | out: |
8cc3b392 | 398 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 399 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 400 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
401 | } |
402 | ||
1da177e4 LT |
403 | /* |
404 | * Higher-order pages are called "compound pages". They are structured thusly: | |
405 | * | |
406 | * The first PAGE_SIZE page is called the "head page". | |
407 | * | |
408 | * The remaining PAGE_SIZE pages are called "tail pages". | |
409 | * | |
6416b9fa WSH |
410 | * All pages have PG_compound set. All tail pages have their ->first_page |
411 | * pointing at the head page. | |
1da177e4 | 412 | * |
41d78ba5 HD |
413 | * The first tail page's ->lru.next holds the address of the compound page's |
414 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
415 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 416 | */ |
d98c7a09 HD |
417 | |
418 | static void free_compound_page(struct page *page) | |
419 | { | |
d85f3385 | 420 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
421 | } |
422 | ||
01ad1c08 | 423 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
424 | { |
425 | int i; | |
426 | int nr_pages = 1 << order; | |
427 | ||
428 | set_compound_page_dtor(page, free_compound_page); | |
429 | set_compound_order(page, order); | |
430 | __SetPageHead(page); | |
431 | for (i = 1; i < nr_pages; i++) { | |
432 | struct page *p = page + i; | |
58a84aa9 | 433 | set_page_count(p, 0); |
18229df5 | 434 | p->first_page = page; |
668f9abb DR |
435 | /* Make sure p->first_page is always valid for PageTail() */ |
436 | smp_wmb(); | |
437 | __SetPageTail(p); | |
18229df5 AW |
438 | } |
439 | } | |
440 | ||
c0a32fc5 SG |
441 | #ifdef CONFIG_DEBUG_PAGEALLOC |
442 | unsigned int _debug_guardpage_minorder; | |
031bc574 | 443 | bool _debug_pagealloc_enabled __read_mostly; |
e30825f1 JK |
444 | bool _debug_guardpage_enabled __read_mostly; |
445 | ||
031bc574 JK |
446 | static int __init early_debug_pagealloc(char *buf) |
447 | { | |
448 | if (!buf) | |
449 | return -EINVAL; | |
450 | ||
451 | if (strcmp(buf, "on") == 0) | |
452 | _debug_pagealloc_enabled = true; | |
453 | ||
454 | return 0; | |
455 | } | |
456 | early_param("debug_pagealloc", early_debug_pagealloc); | |
457 | ||
e30825f1 JK |
458 | static bool need_debug_guardpage(void) |
459 | { | |
031bc574 JK |
460 | /* If we don't use debug_pagealloc, we don't need guard page */ |
461 | if (!debug_pagealloc_enabled()) | |
462 | return false; | |
463 | ||
e30825f1 JK |
464 | return true; |
465 | } | |
466 | ||
467 | static void init_debug_guardpage(void) | |
468 | { | |
031bc574 JK |
469 | if (!debug_pagealloc_enabled()) |
470 | return; | |
471 | ||
e30825f1 JK |
472 | _debug_guardpage_enabled = true; |
473 | } | |
474 | ||
475 | struct page_ext_operations debug_guardpage_ops = { | |
476 | .need = need_debug_guardpage, | |
477 | .init = init_debug_guardpage, | |
478 | }; | |
c0a32fc5 SG |
479 | |
480 | static int __init debug_guardpage_minorder_setup(char *buf) | |
481 | { | |
482 | unsigned long res; | |
483 | ||
484 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
485 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
486 | return 0; | |
487 | } | |
488 | _debug_guardpage_minorder = res; | |
489 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
490 | return 0; | |
491 | } | |
492 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
493 | ||
2847cf95 JK |
494 | static inline void set_page_guard(struct zone *zone, struct page *page, |
495 | unsigned int order, int migratetype) | |
c0a32fc5 | 496 | { |
e30825f1 JK |
497 | struct page_ext *page_ext; |
498 | ||
499 | if (!debug_guardpage_enabled()) | |
500 | return; | |
501 | ||
502 | page_ext = lookup_page_ext(page); | |
503 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
504 | ||
2847cf95 JK |
505 | INIT_LIST_HEAD(&page->lru); |
506 | set_page_private(page, order); | |
507 | /* Guard pages are not available for any usage */ | |
508 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
c0a32fc5 SG |
509 | } |
510 | ||
2847cf95 JK |
511 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
512 | unsigned int order, int migratetype) | |
c0a32fc5 | 513 | { |
e30825f1 JK |
514 | struct page_ext *page_ext; |
515 | ||
516 | if (!debug_guardpage_enabled()) | |
517 | return; | |
518 | ||
519 | page_ext = lookup_page_ext(page); | |
520 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
521 | ||
2847cf95 JK |
522 | set_page_private(page, 0); |
523 | if (!is_migrate_isolate(migratetype)) | |
524 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
525 | } |
526 | #else | |
e30825f1 | 527 | struct page_ext_operations debug_guardpage_ops = { NULL, }; |
2847cf95 JK |
528 | static inline void set_page_guard(struct zone *zone, struct page *page, |
529 | unsigned int order, int migratetype) {} | |
530 | static inline void clear_page_guard(struct zone *zone, struct page *page, | |
531 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
532 | #endif |
533 | ||
7aeb09f9 | 534 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 535 | { |
4c21e2f2 | 536 | set_page_private(page, order); |
676165a8 | 537 | __SetPageBuddy(page); |
1da177e4 LT |
538 | } |
539 | ||
540 | static inline void rmv_page_order(struct page *page) | |
541 | { | |
676165a8 | 542 | __ClearPageBuddy(page); |
4c21e2f2 | 543 | set_page_private(page, 0); |
1da177e4 LT |
544 | } |
545 | ||
1da177e4 LT |
546 | /* |
547 | * This function checks whether a page is free && is the buddy | |
548 | * we can do coalesce a page and its buddy if | |
13e7444b | 549 | * (a) the buddy is not in a hole && |
676165a8 | 550 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
551 | * (c) a page and its buddy have the same order && |
552 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 553 | * |
cf6fe945 WSH |
554 | * For recording whether a page is in the buddy system, we set ->_mapcount |
555 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
556 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
557 | * serialized by zone->lock. | |
1da177e4 | 558 | * |
676165a8 | 559 | * For recording page's order, we use page_private(page). |
1da177e4 | 560 | */ |
cb2b95e1 | 561 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 562 | unsigned int order) |
1da177e4 | 563 | { |
14e07298 | 564 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 565 | return 0; |
13e7444b | 566 | |
c0a32fc5 | 567 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
568 | if (page_zone_id(page) != page_zone_id(buddy)) |
569 | return 0; | |
570 | ||
4c5018ce WY |
571 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
572 | ||
c0a32fc5 SG |
573 | return 1; |
574 | } | |
575 | ||
cb2b95e1 | 576 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
577 | /* |
578 | * zone check is done late to avoid uselessly | |
579 | * calculating zone/node ids for pages that could | |
580 | * never merge. | |
581 | */ | |
582 | if (page_zone_id(page) != page_zone_id(buddy)) | |
583 | return 0; | |
584 | ||
4c5018ce WY |
585 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
586 | ||
6aa3001b | 587 | return 1; |
676165a8 | 588 | } |
6aa3001b | 589 | return 0; |
1da177e4 LT |
590 | } |
591 | ||
592 | /* | |
593 | * Freeing function for a buddy system allocator. | |
594 | * | |
595 | * The concept of a buddy system is to maintain direct-mapped table | |
596 | * (containing bit values) for memory blocks of various "orders". | |
597 | * The bottom level table contains the map for the smallest allocatable | |
598 | * units of memory (here, pages), and each level above it describes | |
599 | * pairs of units from the levels below, hence, "buddies". | |
600 | * At a high level, all that happens here is marking the table entry | |
601 | * at the bottom level available, and propagating the changes upward | |
602 | * as necessary, plus some accounting needed to play nicely with other | |
603 | * parts of the VM system. | |
604 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
605 | * free pages of length of (1 << order) and marked with _mapcount |
606 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
607 | * field. | |
1da177e4 | 608 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
609 | * other. That is, if we allocate a small block, and both were |
610 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 611 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 612 | * triggers coalescing into a block of larger size. |
1da177e4 | 613 | * |
6d49e352 | 614 | * -- nyc |
1da177e4 LT |
615 | */ |
616 | ||
48db57f8 | 617 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 618 | unsigned long pfn, |
ed0ae21d MG |
619 | struct zone *zone, unsigned int order, |
620 | int migratetype) | |
1da177e4 LT |
621 | { |
622 | unsigned long page_idx; | |
6dda9d55 | 623 | unsigned long combined_idx; |
43506fad | 624 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 625 | struct page *buddy; |
3c605096 | 626 | int max_order = MAX_ORDER; |
1da177e4 | 627 | |
d29bb978 | 628 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 629 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 630 | |
ed0ae21d | 631 | VM_BUG_ON(migratetype == -1); |
3c605096 JK |
632 | if (is_migrate_isolate(migratetype)) { |
633 | /* | |
634 | * We restrict max order of merging to prevent merge | |
635 | * between freepages on isolate pageblock and normal | |
636 | * pageblock. Without this, pageblock isolation | |
637 | * could cause incorrect freepage accounting. | |
638 | */ | |
639 | max_order = min(MAX_ORDER, pageblock_order + 1); | |
640 | } else { | |
8f82b55d | 641 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
3c605096 | 642 | } |
ed0ae21d | 643 | |
3c605096 | 644 | page_idx = pfn & ((1 << max_order) - 1); |
1da177e4 | 645 | |
309381fe SL |
646 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
647 | VM_BUG_ON_PAGE(bad_range(zone, page), page); | |
1da177e4 | 648 | |
3c605096 | 649 | while (order < max_order - 1) { |
43506fad KC |
650 | buddy_idx = __find_buddy_index(page_idx, order); |
651 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 652 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 653 | break; |
c0a32fc5 SG |
654 | /* |
655 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
656 | * merge with it and move up one order. | |
657 | */ | |
658 | if (page_is_guard(buddy)) { | |
2847cf95 | 659 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
660 | } else { |
661 | list_del(&buddy->lru); | |
662 | zone->free_area[order].nr_free--; | |
663 | rmv_page_order(buddy); | |
664 | } | |
43506fad | 665 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
666 | page = page + (combined_idx - page_idx); |
667 | page_idx = combined_idx; | |
668 | order++; | |
669 | } | |
670 | set_page_order(page, order); | |
6dda9d55 CZ |
671 | |
672 | /* | |
673 | * If this is not the largest possible page, check if the buddy | |
674 | * of the next-highest order is free. If it is, it's possible | |
675 | * that pages are being freed that will coalesce soon. In case, | |
676 | * that is happening, add the free page to the tail of the list | |
677 | * so it's less likely to be used soon and more likely to be merged | |
678 | * as a higher order page | |
679 | */ | |
b7f50cfa | 680 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 681 | struct page *higher_page, *higher_buddy; |
43506fad KC |
682 | combined_idx = buddy_idx & page_idx; |
683 | higher_page = page + (combined_idx - page_idx); | |
684 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 685 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
686 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
687 | list_add_tail(&page->lru, | |
688 | &zone->free_area[order].free_list[migratetype]); | |
689 | goto out; | |
690 | } | |
691 | } | |
692 | ||
693 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
694 | out: | |
1da177e4 LT |
695 | zone->free_area[order].nr_free++; |
696 | } | |
697 | ||
224abf92 | 698 | static inline int free_pages_check(struct page *page) |
1da177e4 | 699 | { |
d230dec1 | 700 | const char *bad_reason = NULL; |
f0b791a3 DH |
701 | unsigned long bad_flags = 0; |
702 | ||
703 | if (unlikely(page_mapcount(page))) | |
704 | bad_reason = "nonzero mapcount"; | |
705 | if (unlikely(page->mapping != NULL)) | |
706 | bad_reason = "non-NULL mapping"; | |
707 | if (unlikely(atomic_read(&page->_count) != 0)) | |
708 | bad_reason = "nonzero _count"; | |
709 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { | |
710 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
711 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
712 | } | |
9edad6ea JW |
713 | #ifdef CONFIG_MEMCG |
714 | if (unlikely(page->mem_cgroup)) | |
715 | bad_reason = "page still charged to cgroup"; | |
716 | #endif | |
f0b791a3 DH |
717 | if (unlikely(bad_reason)) { |
718 | bad_page(page, bad_reason, bad_flags); | |
79f4b7bf | 719 | return 1; |
8cc3b392 | 720 | } |
90572890 | 721 | page_cpupid_reset_last(page); |
79f4b7bf HD |
722 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
723 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
724 | return 0; | |
1da177e4 LT |
725 | } |
726 | ||
727 | /* | |
5f8dcc21 | 728 | * Frees a number of pages from the PCP lists |
1da177e4 | 729 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 730 | * count is the number of pages to free. |
1da177e4 LT |
731 | * |
732 | * If the zone was previously in an "all pages pinned" state then look to | |
733 | * see if this freeing clears that state. | |
734 | * | |
735 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
736 | * pinned" detection logic. | |
737 | */ | |
5f8dcc21 MG |
738 | static void free_pcppages_bulk(struct zone *zone, int count, |
739 | struct per_cpu_pages *pcp) | |
1da177e4 | 740 | { |
5f8dcc21 | 741 | int migratetype = 0; |
a6f9edd6 | 742 | int batch_free = 0; |
72853e29 | 743 | int to_free = count; |
0d5d823a | 744 | unsigned long nr_scanned; |
5f8dcc21 | 745 | |
c54ad30c | 746 | spin_lock(&zone->lock); |
0d5d823a MG |
747 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
748 | if (nr_scanned) | |
749 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 750 | |
72853e29 | 751 | while (to_free) { |
48db57f8 | 752 | struct page *page; |
5f8dcc21 MG |
753 | struct list_head *list; |
754 | ||
755 | /* | |
a6f9edd6 MG |
756 | * Remove pages from lists in a round-robin fashion. A |
757 | * batch_free count is maintained that is incremented when an | |
758 | * empty list is encountered. This is so more pages are freed | |
759 | * off fuller lists instead of spinning excessively around empty | |
760 | * lists | |
5f8dcc21 MG |
761 | */ |
762 | do { | |
a6f9edd6 | 763 | batch_free++; |
5f8dcc21 MG |
764 | if (++migratetype == MIGRATE_PCPTYPES) |
765 | migratetype = 0; | |
766 | list = &pcp->lists[migratetype]; | |
767 | } while (list_empty(list)); | |
48db57f8 | 768 | |
1d16871d NK |
769 | /* This is the only non-empty list. Free them all. */ |
770 | if (batch_free == MIGRATE_PCPTYPES) | |
771 | batch_free = to_free; | |
772 | ||
a6f9edd6 | 773 | do { |
770c8aaa BZ |
774 | int mt; /* migratetype of the to-be-freed page */ |
775 | ||
a6f9edd6 MG |
776 | page = list_entry(list->prev, struct page, lru); |
777 | /* must delete as __free_one_page list manipulates */ | |
778 | list_del(&page->lru); | |
b12c4ad1 | 779 | mt = get_freepage_migratetype(page); |
8f82b55d | 780 | if (unlikely(has_isolate_pageblock(zone))) |
51bb1a40 | 781 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 782 | |
a7016235 | 783 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
dc4b0caf | 784 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 785 | trace_mm_page_pcpu_drain(page, 0, mt); |
72853e29 | 786 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 787 | } |
c54ad30c | 788 | spin_unlock(&zone->lock); |
1da177e4 LT |
789 | } |
790 | ||
dc4b0caf MG |
791 | static void free_one_page(struct zone *zone, |
792 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 793 | unsigned int order, |
ed0ae21d | 794 | int migratetype) |
1da177e4 | 795 | { |
0d5d823a | 796 | unsigned long nr_scanned; |
006d22d9 | 797 | spin_lock(&zone->lock); |
0d5d823a MG |
798 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
799 | if (nr_scanned) | |
800 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 801 | |
ad53f92e JK |
802 | if (unlikely(has_isolate_pageblock(zone) || |
803 | is_migrate_isolate(migratetype))) { | |
804 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 805 | } |
dc4b0caf | 806 | __free_one_page(page, pfn, zone, order, migratetype); |
006d22d9 | 807 | spin_unlock(&zone->lock); |
48db57f8 NP |
808 | } |
809 | ||
81422f29 KS |
810 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
811 | { | |
812 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) | |
813 | return 0; | |
814 | if (unlikely(!PageTail(page))) { | |
815 | bad_page(page, "PageTail not set", 0); | |
816 | return 1; | |
817 | } | |
818 | if (unlikely(page->first_page != head_page)) { | |
819 | bad_page(page, "first_page not consistent", 0); | |
820 | return 1; | |
821 | } | |
822 | return 0; | |
823 | } | |
824 | ||
1e8ce83c RH |
825 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
826 | unsigned long zone, int nid) | |
827 | { | |
828 | struct zone *z = &NODE_DATA(nid)->node_zones[zone]; | |
829 | ||
830 | set_page_links(page, zone, nid, pfn); | |
831 | mminit_verify_page_links(page, zone, nid, pfn); | |
832 | init_page_count(page); | |
833 | page_mapcount_reset(page); | |
834 | page_cpupid_reset_last(page); | |
1e8ce83c RH |
835 | |
836 | /* | |
837 | * Mark the block movable so that blocks are reserved for | |
838 | * movable at startup. This will force kernel allocations | |
839 | * to reserve their blocks rather than leaking throughout | |
840 | * the address space during boot when many long-lived | |
841 | * kernel allocations are made. Later some blocks near | |
842 | * the start are marked MIGRATE_RESERVE by | |
843 | * setup_zone_migrate_reserve() | |
844 | * | |
845 | * bitmap is created for zone's valid pfn range. but memmap | |
846 | * can be created for invalid pages (for alignment) | |
847 | * check here not to call set_pageblock_migratetype() against | |
848 | * pfn out of zone. | |
849 | */ | |
850 | if ((z->zone_start_pfn <= pfn) | |
851 | && (pfn < zone_end_pfn(z)) | |
852 | && !(pfn & (pageblock_nr_pages - 1))) | |
853 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
854 | ||
855 | INIT_LIST_HEAD(&page->lru); | |
856 | #ifdef WANT_PAGE_VIRTUAL | |
857 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
858 | if (!is_highmem_idx(zone)) | |
859 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
860 | #endif | |
861 | } | |
862 | ||
863 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
864 | int nid) | |
865 | { | |
866 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
867 | } | |
868 | ||
92923ca3 NZ |
869 | /* |
870 | * Initialised pages do not have PageReserved set. This function is | |
871 | * called for each range allocated by the bootmem allocator and | |
872 | * marks the pages PageReserved. The remaining valid pages are later | |
873 | * sent to the buddy page allocator. | |
874 | */ | |
875 | void reserve_bootmem_region(unsigned long start, unsigned long end) | |
876 | { | |
877 | unsigned long start_pfn = PFN_DOWN(start); | |
878 | unsigned long end_pfn = PFN_UP(end); | |
879 | ||
880 | for (; start_pfn < end_pfn; start_pfn++) | |
881 | if (pfn_valid(start_pfn)) | |
882 | SetPageReserved(pfn_to_page(start_pfn)); | |
883 | } | |
884 | ||
ec95f53a | 885 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 886 | { |
81422f29 KS |
887 | bool compound = PageCompound(page); |
888 | int i, bad = 0; | |
1da177e4 | 889 | |
ab1f306f | 890 | VM_BUG_ON_PAGE(PageTail(page), page); |
81422f29 | 891 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); |
ab1f306f | 892 | |
b413d48a | 893 | trace_mm_page_free(page, order); |
b1eeab67 | 894 | kmemcheck_free_shadow(page, order); |
b8c73fc2 | 895 | kasan_free_pages(page, order); |
b1eeab67 | 896 | |
8dd60a3a AA |
897 | if (PageAnon(page)) |
898 | page->mapping = NULL; | |
81422f29 KS |
899 | bad += free_pages_check(page); |
900 | for (i = 1; i < (1 << order); i++) { | |
901 | if (compound) | |
902 | bad += free_tail_pages_check(page, page + i); | |
8dd60a3a | 903 | bad += free_pages_check(page + i); |
81422f29 | 904 | } |
8cc3b392 | 905 | if (bad) |
ec95f53a | 906 | return false; |
689bcebf | 907 | |
48c96a36 JK |
908 | reset_page_owner(page, order); |
909 | ||
3ac7fe5a | 910 | if (!PageHighMem(page)) { |
b8af2941 PK |
911 | debug_check_no_locks_freed(page_address(page), |
912 | PAGE_SIZE << order); | |
3ac7fe5a TG |
913 | debug_check_no_obj_freed(page_address(page), |
914 | PAGE_SIZE << order); | |
915 | } | |
dafb1367 | 916 | arch_free_page(page, order); |
48db57f8 | 917 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 918 | |
ec95f53a KM |
919 | return true; |
920 | } | |
921 | ||
922 | static void __free_pages_ok(struct page *page, unsigned int order) | |
923 | { | |
924 | unsigned long flags; | |
95e34412 | 925 | int migratetype; |
dc4b0caf | 926 | unsigned long pfn = page_to_pfn(page); |
ec95f53a KM |
927 | |
928 | if (!free_pages_prepare(page, order)) | |
929 | return; | |
930 | ||
cfc47a28 | 931 | migratetype = get_pfnblock_migratetype(page, pfn); |
c54ad30c | 932 | local_irq_save(flags); |
f8891e5e | 933 | __count_vm_events(PGFREE, 1 << order); |
95e34412 | 934 | set_freepage_migratetype(page, migratetype); |
dc4b0caf | 935 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
c54ad30c | 936 | local_irq_restore(flags); |
1da177e4 LT |
937 | } |
938 | ||
3a80a7fa MG |
939 | static void __defer_init __free_pages_boot_core(struct page *page, |
940 | unsigned long pfn, unsigned int order) | |
a226f6c8 | 941 | { |
c3993076 | 942 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 943 | struct page *p = page; |
c3993076 | 944 | unsigned int loop; |
a226f6c8 | 945 | |
e2d0bd2b YL |
946 | prefetchw(p); |
947 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
948 | prefetchw(p + 1); | |
c3993076 JW |
949 | __ClearPageReserved(p); |
950 | set_page_count(p, 0); | |
a226f6c8 | 951 | } |
e2d0bd2b YL |
952 | __ClearPageReserved(p); |
953 | set_page_count(p, 0); | |
c3993076 | 954 | |
e2d0bd2b | 955 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
956 | set_page_refcounted(page); |
957 | __free_pages(page, order); | |
a226f6c8 DH |
958 | } |
959 | ||
75a592a4 MG |
960 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
961 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
962 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
963 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; | |
964 | ||
965 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
966 | { | |
967 | int nid; | |
968 | ||
969 | /* The system will behave unpredictably otherwise */ | |
970 | BUG_ON(system_state != SYSTEM_BOOTING); | |
971 | ||
972 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); | |
973 | if (nid >= 0) | |
974 | return nid; | |
975 | /* just returns 0 */ | |
976 | return 0; | |
977 | } | |
978 | #endif | |
979 | ||
980 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
981 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
982 | struct mminit_pfnnid_cache *state) | |
983 | { | |
984 | int nid; | |
985 | ||
986 | nid = __early_pfn_to_nid(pfn, state); | |
987 | if (nid >= 0 && nid != node) | |
988 | return false; | |
989 | return true; | |
990 | } | |
991 | ||
992 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
993 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
994 | { | |
995 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
996 | } | |
997 | ||
998 | #else | |
999 | ||
1000 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1001 | { | |
1002 | return true; | |
1003 | } | |
1004 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1005 | struct mminit_pfnnid_cache *state) | |
1006 | { | |
1007 | return true; | |
1008 | } | |
1009 | #endif | |
1010 | ||
1011 | ||
3a80a7fa MG |
1012 | void __defer_init __free_pages_bootmem(struct page *page, unsigned long pfn, |
1013 | unsigned int order) | |
1014 | { | |
1015 | if (early_page_uninitialised(pfn)) | |
1016 | return; | |
1017 | return __free_pages_boot_core(page, pfn, order); | |
1018 | } | |
1019 | ||
47118af0 | 1020 | #ifdef CONFIG_CMA |
9cf510a5 | 1021 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1022 | void __init init_cma_reserved_pageblock(struct page *page) |
1023 | { | |
1024 | unsigned i = pageblock_nr_pages; | |
1025 | struct page *p = page; | |
1026 | ||
1027 | do { | |
1028 | __ClearPageReserved(p); | |
1029 | set_page_count(p, 0); | |
1030 | } while (++p, --i); | |
1031 | ||
47118af0 | 1032 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1033 | |
1034 | if (pageblock_order >= MAX_ORDER) { | |
1035 | i = pageblock_nr_pages; | |
1036 | p = page; | |
1037 | do { | |
1038 | set_page_refcounted(p); | |
1039 | __free_pages(p, MAX_ORDER - 1); | |
1040 | p += MAX_ORDER_NR_PAGES; | |
1041 | } while (i -= MAX_ORDER_NR_PAGES); | |
1042 | } else { | |
1043 | set_page_refcounted(page); | |
1044 | __free_pages(page, pageblock_order); | |
1045 | } | |
1046 | ||
3dcc0571 | 1047 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1048 | } |
1049 | #endif | |
1da177e4 LT |
1050 | |
1051 | /* | |
1052 | * The order of subdivision here is critical for the IO subsystem. | |
1053 | * Please do not alter this order without good reasons and regression | |
1054 | * testing. Specifically, as large blocks of memory are subdivided, | |
1055 | * the order in which smaller blocks are delivered depends on the order | |
1056 | * they're subdivided in this function. This is the primary factor | |
1057 | * influencing the order in which pages are delivered to the IO | |
1058 | * subsystem according to empirical testing, and this is also justified | |
1059 | * by considering the behavior of a buddy system containing a single | |
1060 | * large block of memory acted on by a series of small allocations. | |
1061 | * This behavior is a critical factor in sglist merging's success. | |
1062 | * | |
6d49e352 | 1063 | * -- nyc |
1da177e4 | 1064 | */ |
085cc7d5 | 1065 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1066 | int low, int high, struct free_area *area, |
1067 | int migratetype) | |
1da177e4 LT |
1068 | { |
1069 | unsigned long size = 1 << high; | |
1070 | ||
1071 | while (high > low) { | |
1072 | area--; | |
1073 | high--; | |
1074 | size >>= 1; | |
309381fe | 1075 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1076 | |
2847cf95 | 1077 | if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && |
e30825f1 | 1078 | debug_guardpage_enabled() && |
2847cf95 | 1079 | high < debug_guardpage_minorder()) { |
c0a32fc5 SG |
1080 | /* |
1081 | * Mark as guard pages (or page), that will allow to | |
1082 | * merge back to allocator when buddy will be freed. | |
1083 | * Corresponding page table entries will not be touched, | |
1084 | * pages will stay not present in virtual address space | |
1085 | */ | |
2847cf95 | 1086 | set_page_guard(zone, &page[size], high, migratetype); |
c0a32fc5 SG |
1087 | continue; |
1088 | } | |
b2a0ac88 | 1089 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1090 | area->nr_free++; |
1091 | set_page_order(&page[size], high); | |
1092 | } | |
1da177e4 LT |
1093 | } |
1094 | ||
1da177e4 LT |
1095 | /* |
1096 | * This page is about to be returned from the page allocator | |
1097 | */ | |
2a7684a2 | 1098 | static inline int check_new_page(struct page *page) |
1da177e4 | 1099 | { |
d230dec1 | 1100 | const char *bad_reason = NULL; |
f0b791a3 DH |
1101 | unsigned long bad_flags = 0; |
1102 | ||
1103 | if (unlikely(page_mapcount(page))) | |
1104 | bad_reason = "nonzero mapcount"; | |
1105 | if (unlikely(page->mapping != NULL)) | |
1106 | bad_reason = "non-NULL mapping"; | |
1107 | if (unlikely(atomic_read(&page->_count) != 0)) | |
1108 | bad_reason = "nonzero _count"; | |
1109 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { | |
1110 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1111 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1112 | } | |
9edad6ea JW |
1113 | #ifdef CONFIG_MEMCG |
1114 | if (unlikely(page->mem_cgroup)) | |
1115 | bad_reason = "page still charged to cgroup"; | |
1116 | #endif | |
f0b791a3 DH |
1117 | if (unlikely(bad_reason)) { |
1118 | bad_page(page, bad_reason, bad_flags); | |
689bcebf | 1119 | return 1; |
8cc3b392 | 1120 | } |
2a7684a2 WF |
1121 | return 0; |
1122 | } | |
1123 | ||
75379191 VB |
1124 | static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
1125 | int alloc_flags) | |
2a7684a2 WF |
1126 | { |
1127 | int i; | |
1128 | ||
1129 | for (i = 0; i < (1 << order); i++) { | |
1130 | struct page *p = page + i; | |
1131 | if (unlikely(check_new_page(p))) | |
1132 | return 1; | |
1133 | } | |
689bcebf | 1134 | |
4c21e2f2 | 1135 | set_page_private(page, 0); |
7835e98b | 1136 | set_page_refcounted(page); |
cc102509 NP |
1137 | |
1138 | arch_alloc_page(page, order); | |
1da177e4 | 1139 | kernel_map_pages(page, 1 << order, 1); |
b8c73fc2 | 1140 | kasan_alloc_pages(page, order); |
17cf4406 NP |
1141 | |
1142 | if (gfp_flags & __GFP_ZERO) | |
f4d2897b AA |
1143 | for (i = 0; i < (1 << order); i++) |
1144 | clear_highpage(page + i); | |
17cf4406 NP |
1145 | |
1146 | if (order && (gfp_flags & __GFP_COMP)) | |
1147 | prep_compound_page(page, order); | |
1148 | ||
48c96a36 JK |
1149 | set_page_owner(page, order, gfp_flags); |
1150 | ||
75379191 VB |
1151 | /* |
1152 | * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was necessary to | |
1153 | * allocate the page. The expectation is that the caller is taking | |
1154 | * steps that will free more memory. The caller should avoid the page | |
1155 | * being used for !PFMEMALLOC purposes. | |
1156 | */ | |
1157 | page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS); | |
1158 | ||
689bcebf | 1159 | return 0; |
1da177e4 LT |
1160 | } |
1161 | ||
56fd56b8 MG |
1162 | /* |
1163 | * Go through the free lists for the given migratetype and remove | |
1164 | * the smallest available page from the freelists | |
1165 | */ | |
728ec980 MG |
1166 | static inline |
1167 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1168 | int migratetype) |
1169 | { | |
1170 | unsigned int current_order; | |
b8af2941 | 1171 | struct free_area *area; |
56fd56b8 MG |
1172 | struct page *page; |
1173 | ||
1174 | /* Find a page of the appropriate size in the preferred list */ | |
1175 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1176 | area = &(zone->free_area[current_order]); | |
1177 | if (list_empty(&area->free_list[migratetype])) | |
1178 | continue; | |
1179 | ||
1180 | page = list_entry(area->free_list[migratetype].next, | |
1181 | struct page, lru); | |
1182 | list_del(&page->lru); | |
1183 | rmv_page_order(page); | |
1184 | area->nr_free--; | |
56fd56b8 | 1185 | expand(zone, page, order, current_order, area, migratetype); |
5bcc9f86 | 1186 | set_freepage_migratetype(page, migratetype); |
56fd56b8 MG |
1187 | return page; |
1188 | } | |
1189 | ||
1190 | return NULL; | |
1191 | } | |
1192 | ||
1193 | ||
b2a0ac88 MG |
1194 | /* |
1195 | * This array describes the order lists are fallen back to when | |
1196 | * the free lists for the desirable migrate type are depleted | |
1197 | */ | |
47118af0 MN |
1198 | static int fallbacks[MIGRATE_TYPES][4] = { |
1199 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
1200 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
dc67647b | 1201 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, |
47118af0 | 1202 | #ifdef CONFIG_CMA |
47118af0 | 1203 | [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */ |
47118af0 | 1204 | #endif |
6d4a4916 | 1205 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 1206 | #ifdef CONFIG_MEMORY_ISOLATION |
6d4a4916 | 1207 | [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 1208 | #endif |
b2a0ac88 MG |
1209 | }; |
1210 | ||
dc67647b JK |
1211 | #ifdef CONFIG_CMA |
1212 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1213 | unsigned int order) | |
1214 | { | |
1215 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1216 | } | |
1217 | #else | |
1218 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1219 | unsigned int order) { return NULL; } | |
1220 | #endif | |
1221 | ||
c361be55 MG |
1222 | /* |
1223 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1224 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1225 | * boundary. If alignment is required, use move_freepages_block() |
1226 | */ | |
435b405c | 1227 | int move_freepages(struct zone *zone, |
b69a7288 AB |
1228 | struct page *start_page, struct page *end_page, |
1229 | int migratetype) | |
c361be55 MG |
1230 | { |
1231 | struct page *page; | |
1232 | unsigned long order; | |
d100313f | 1233 | int pages_moved = 0; |
c361be55 MG |
1234 | |
1235 | #ifndef CONFIG_HOLES_IN_ZONE | |
1236 | /* | |
1237 | * page_zone is not safe to call in this context when | |
1238 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1239 | * anyway as we check zone boundaries in move_freepages_block(). | |
1240 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1241 | * grouping pages by mobility |
c361be55 | 1242 | */ |
97ee4ba7 | 1243 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1244 | #endif |
1245 | ||
1246 | for (page = start_page; page <= end_page;) { | |
344c790e | 1247 | /* Make sure we are not inadvertently changing nodes */ |
309381fe | 1248 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
344c790e | 1249 | |
c361be55 MG |
1250 | if (!pfn_valid_within(page_to_pfn(page))) { |
1251 | page++; | |
1252 | continue; | |
1253 | } | |
1254 | ||
1255 | if (!PageBuddy(page)) { | |
1256 | page++; | |
1257 | continue; | |
1258 | } | |
1259 | ||
1260 | order = page_order(page); | |
84be48d8 KS |
1261 | list_move(&page->lru, |
1262 | &zone->free_area[order].free_list[migratetype]); | |
95e34412 | 1263 | set_freepage_migratetype(page, migratetype); |
c361be55 | 1264 | page += 1 << order; |
d100313f | 1265 | pages_moved += 1 << order; |
c361be55 MG |
1266 | } |
1267 | ||
d100313f | 1268 | return pages_moved; |
c361be55 MG |
1269 | } |
1270 | ||
ee6f509c | 1271 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1272 | int migratetype) |
c361be55 MG |
1273 | { |
1274 | unsigned long start_pfn, end_pfn; | |
1275 | struct page *start_page, *end_page; | |
1276 | ||
1277 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1278 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1279 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1280 | end_page = start_page + pageblock_nr_pages - 1; |
1281 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1282 | |
1283 | /* Do not cross zone boundaries */ | |
108bcc96 | 1284 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1285 | start_page = page; |
108bcc96 | 1286 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1287 | return 0; |
1288 | ||
1289 | return move_freepages(zone, start_page, end_page, migratetype); | |
1290 | } | |
1291 | ||
2f66a68f MG |
1292 | static void change_pageblock_range(struct page *pageblock_page, |
1293 | int start_order, int migratetype) | |
1294 | { | |
1295 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1296 | ||
1297 | while (nr_pageblocks--) { | |
1298 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1299 | pageblock_page += pageblock_nr_pages; | |
1300 | } | |
1301 | } | |
1302 | ||
fef903ef | 1303 | /* |
9c0415eb VB |
1304 | * When we are falling back to another migratetype during allocation, try to |
1305 | * steal extra free pages from the same pageblocks to satisfy further | |
1306 | * allocations, instead of polluting multiple pageblocks. | |
1307 | * | |
1308 | * If we are stealing a relatively large buddy page, it is likely there will | |
1309 | * be more free pages in the pageblock, so try to steal them all. For | |
1310 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1311 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1312 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1313 | * pageblocks. | |
fef903ef | 1314 | */ |
4eb7dce6 JK |
1315 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1316 | { | |
1317 | /* | |
1318 | * Leaving this order check is intended, although there is | |
1319 | * relaxed order check in next check. The reason is that | |
1320 | * we can actually steal whole pageblock if this condition met, | |
1321 | * but, below check doesn't guarantee it and that is just heuristic | |
1322 | * so could be changed anytime. | |
1323 | */ | |
1324 | if (order >= pageblock_order) | |
1325 | return true; | |
1326 | ||
1327 | if (order >= pageblock_order / 2 || | |
1328 | start_mt == MIGRATE_RECLAIMABLE || | |
1329 | start_mt == MIGRATE_UNMOVABLE || | |
1330 | page_group_by_mobility_disabled) | |
1331 | return true; | |
1332 | ||
1333 | return false; | |
1334 | } | |
1335 | ||
1336 | /* | |
1337 | * This function implements actual steal behaviour. If order is large enough, | |
1338 | * we can steal whole pageblock. If not, we first move freepages in this | |
1339 | * pageblock and check whether half of pages are moved or not. If half of | |
1340 | * pages are moved, we can change migratetype of pageblock and permanently | |
1341 | * use it's pages as requested migratetype in the future. | |
1342 | */ | |
1343 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1344 | int start_type) | |
fef903ef SB |
1345 | { |
1346 | int current_order = page_order(page); | |
4eb7dce6 | 1347 | int pages; |
fef903ef | 1348 | |
fef903ef SB |
1349 | /* Take ownership for orders >= pageblock_order */ |
1350 | if (current_order >= pageblock_order) { | |
1351 | change_pageblock_range(page, current_order, start_type); | |
3a1086fb | 1352 | return; |
fef903ef SB |
1353 | } |
1354 | ||
4eb7dce6 | 1355 | pages = move_freepages_block(zone, page, start_type); |
fef903ef | 1356 | |
4eb7dce6 JK |
1357 | /* Claim the whole block if over half of it is free */ |
1358 | if (pages >= (1 << (pageblock_order-1)) || | |
1359 | page_group_by_mobility_disabled) | |
1360 | set_pageblock_migratetype(page, start_type); | |
1361 | } | |
1362 | ||
2149cdae JK |
1363 | /* |
1364 | * Check whether there is a suitable fallback freepage with requested order. | |
1365 | * If only_stealable is true, this function returns fallback_mt only if | |
1366 | * we can steal other freepages all together. This would help to reduce | |
1367 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1368 | */ | |
1369 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1370 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1371 | { |
1372 | int i; | |
1373 | int fallback_mt; | |
1374 | ||
1375 | if (area->nr_free == 0) | |
1376 | return -1; | |
1377 | ||
1378 | *can_steal = false; | |
1379 | for (i = 0;; i++) { | |
1380 | fallback_mt = fallbacks[migratetype][i]; | |
1381 | if (fallback_mt == MIGRATE_RESERVE) | |
1382 | break; | |
1383 | ||
1384 | if (list_empty(&area->free_list[fallback_mt])) | |
1385 | continue; | |
fef903ef | 1386 | |
4eb7dce6 JK |
1387 | if (can_steal_fallback(order, migratetype)) |
1388 | *can_steal = true; | |
1389 | ||
2149cdae JK |
1390 | if (!only_stealable) |
1391 | return fallback_mt; | |
1392 | ||
1393 | if (*can_steal) | |
1394 | return fallback_mt; | |
fef903ef | 1395 | } |
4eb7dce6 JK |
1396 | |
1397 | return -1; | |
fef903ef SB |
1398 | } |
1399 | ||
b2a0ac88 | 1400 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 1401 | static inline struct page * |
7aeb09f9 | 1402 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 1403 | { |
b8af2941 | 1404 | struct free_area *area; |
7aeb09f9 | 1405 | unsigned int current_order; |
b2a0ac88 | 1406 | struct page *page; |
4eb7dce6 JK |
1407 | int fallback_mt; |
1408 | bool can_steal; | |
b2a0ac88 MG |
1409 | |
1410 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
1411 | for (current_order = MAX_ORDER-1; |
1412 | current_order >= order && current_order <= MAX_ORDER-1; | |
1413 | --current_order) { | |
4eb7dce6 JK |
1414 | area = &(zone->free_area[current_order]); |
1415 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 1416 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
1417 | if (fallback_mt == -1) |
1418 | continue; | |
b2a0ac88 | 1419 | |
4eb7dce6 JK |
1420 | page = list_entry(area->free_list[fallback_mt].next, |
1421 | struct page, lru); | |
1422 | if (can_steal) | |
1423 | steal_suitable_fallback(zone, page, start_migratetype); | |
b2a0ac88 | 1424 | |
4eb7dce6 JK |
1425 | /* Remove the page from the freelists */ |
1426 | area->nr_free--; | |
1427 | list_del(&page->lru); | |
1428 | rmv_page_order(page); | |
3a1086fb | 1429 | |
4eb7dce6 JK |
1430 | expand(zone, page, order, current_order, area, |
1431 | start_migratetype); | |
1432 | /* | |
1433 | * The freepage_migratetype may differ from pageblock's | |
1434 | * migratetype depending on the decisions in | |
1435 | * try_to_steal_freepages(). This is OK as long as it | |
1436 | * does not differ for MIGRATE_CMA pageblocks. For CMA | |
1437 | * we need to make sure unallocated pages flushed from | |
1438 | * pcp lists are returned to the correct freelist. | |
1439 | */ | |
1440 | set_freepage_migratetype(page, start_migratetype); | |
e0fff1bd | 1441 | |
4eb7dce6 JK |
1442 | trace_mm_page_alloc_extfrag(page, order, current_order, |
1443 | start_migratetype, fallback_mt); | |
e0fff1bd | 1444 | |
4eb7dce6 | 1445 | return page; |
b2a0ac88 MG |
1446 | } |
1447 | ||
728ec980 | 1448 | return NULL; |
b2a0ac88 MG |
1449 | } |
1450 | ||
56fd56b8 | 1451 | /* |
1da177e4 LT |
1452 | * Do the hard work of removing an element from the buddy allocator. |
1453 | * Call me with the zone->lock already held. | |
1454 | */ | |
b2a0ac88 MG |
1455 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1456 | int migratetype) | |
1da177e4 | 1457 | { |
1da177e4 LT |
1458 | struct page *page; |
1459 | ||
728ec980 | 1460 | retry_reserve: |
56fd56b8 | 1461 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1462 | |
728ec980 | 1463 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
dc67647b JK |
1464 | if (migratetype == MIGRATE_MOVABLE) |
1465 | page = __rmqueue_cma_fallback(zone, order); | |
1466 | ||
1467 | if (!page) | |
1468 | page = __rmqueue_fallback(zone, order, migratetype); | |
b2a0ac88 | 1469 | |
728ec980 MG |
1470 | /* |
1471 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1472 | * is used because __rmqueue_smallest is an inline function | |
1473 | * and we want just one call site | |
1474 | */ | |
1475 | if (!page) { | |
1476 | migratetype = MIGRATE_RESERVE; | |
1477 | goto retry_reserve; | |
1478 | } | |
1479 | } | |
1480 | ||
0d3d062a | 1481 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1482 | return page; |
1da177e4 LT |
1483 | } |
1484 | ||
5f63b720 | 1485 | /* |
1da177e4 LT |
1486 | * Obtain a specified number of elements from the buddy allocator, all under |
1487 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1488 | * Returns the number of new pages which were placed at *list. | |
1489 | */ | |
5f63b720 | 1490 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1491 | unsigned long count, struct list_head *list, |
b745bc85 | 1492 | int migratetype, bool cold) |
1da177e4 | 1493 | { |
5bcc9f86 | 1494 | int i; |
5f63b720 | 1495 | |
c54ad30c | 1496 | spin_lock(&zone->lock); |
1da177e4 | 1497 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1498 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1499 | if (unlikely(page == NULL)) |
1da177e4 | 1500 | break; |
81eabcbe MG |
1501 | |
1502 | /* | |
1503 | * Split buddy pages returned by expand() are received here | |
1504 | * in physical page order. The page is added to the callers and | |
1505 | * list and the list head then moves forward. From the callers | |
1506 | * perspective, the linked list is ordered by page number in | |
1507 | * some conditions. This is useful for IO devices that can | |
1508 | * merge IO requests if the physical pages are ordered | |
1509 | * properly. | |
1510 | */ | |
b745bc85 | 1511 | if (likely(!cold)) |
e084b2d9 MG |
1512 | list_add(&page->lru, list); |
1513 | else | |
1514 | list_add_tail(&page->lru, list); | |
81eabcbe | 1515 | list = &page->lru; |
5bcc9f86 | 1516 | if (is_migrate_cma(get_freepage_migratetype(page))) |
d1ce749a BZ |
1517 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
1518 | -(1 << order)); | |
1da177e4 | 1519 | } |
f2260e6b | 1520 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1521 | spin_unlock(&zone->lock); |
085cc7d5 | 1522 | return i; |
1da177e4 LT |
1523 | } |
1524 | ||
4ae7c039 | 1525 | #ifdef CONFIG_NUMA |
8fce4d8e | 1526 | /* |
4037d452 CL |
1527 | * Called from the vmstat counter updater to drain pagesets of this |
1528 | * currently executing processor on remote nodes after they have | |
1529 | * expired. | |
1530 | * | |
879336c3 CL |
1531 | * Note that this function must be called with the thread pinned to |
1532 | * a single processor. | |
8fce4d8e | 1533 | */ |
4037d452 | 1534 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1535 | { |
4ae7c039 | 1536 | unsigned long flags; |
7be12fc9 | 1537 | int to_drain, batch; |
4ae7c039 | 1538 | |
4037d452 | 1539 | local_irq_save(flags); |
4db0c3c2 | 1540 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 1541 | to_drain = min(pcp->count, batch); |
2a13515c KM |
1542 | if (to_drain > 0) { |
1543 | free_pcppages_bulk(zone, to_drain, pcp); | |
1544 | pcp->count -= to_drain; | |
1545 | } | |
4037d452 | 1546 | local_irq_restore(flags); |
4ae7c039 CL |
1547 | } |
1548 | #endif | |
1549 | ||
9f8f2172 | 1550 | /* |
93481ff0 | 1551 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
1552 | * |
1553 | * The processor must either be the current processor and the | |
1554 | * thread pinned to the current processor or a processor that | |
1555 | * is not online. | |
1556 | */ | |
93481ff0 | 1557 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 1558 | { |
c54ad30c | 1559 | unsigned long flags; |
93481ff0 VB |
1560 | struct per_cpu_pageset *pset; |
1561 | struct per_cpu_pages *pcp; | |
1da177e4 | 1562 | |
93481ff0 VB |
1563 | local_irq_save(flags); |
1564 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 1565 | |
93481ff0 VB |
1566 | pcp = &pset->pcp; |
1567 | if (pcp->count) { | |
1568 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1569 | pcp->count = 0; | |
1570 | } | |
1571 | local_irq_restore(flags); | |
1572 | } | |
3dfa5721 | 1573 | |
93481ff0 VB |
1574 | /* |
1575 | * Drain pcplists of all zones on the indicated processor. | |
1576 | * | |
1577 | * The processor must either be the current processor and the | |
1578 | * thread pinned to the current processor or a processor that | |
1579 | * is not online. | |
1580 | */ | |
1581 | static void drain_pages(unsigned int cpu) | |
1582 | { | |
1583 | struct zone *zone; | |
1584 | ||
1585 | for_each_populated_zone(zone) { | |
1586 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
1587 | } |
1588 | } | |
1da177e4 | 1589 | |
9f8f2172 CL |
1590 | /* |
1591 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
1592 | * |
1593 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
1594 | * the single zone's pages. | |
9f8f2172 | 1595 | */ |
93481ff0 | 1596 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 1597 | { |
93481ff0 VB |
1598 | int cpu = smp_processor_id(); |
1599 | ||
1600 | if (zone) | |
1601 | drain_pages_zone(cpu, zone); | |
1602 | else | |
1603 | drain_pages(cpu); | |
9f8f2172 CL |
1604 | } |
1605 | ||
1606 | /* | |
74046494 GBY |
1607 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
1608 | * | |
93481ff0 VB |
1609 | * When zone parameter is non-NULL, spill just the single zone's pages. |
1610 | * | |
74046494 GBY |
1611 | * Note that this code is protected against sending an IPI to an offline |
1612 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
1613 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
1614 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
1615 | * before the call to on_each_cpu_mask(). | |
9f8f2172 | 1616 | */ |
93481ff0 | 1617 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 1618 | { |
74046494 | 1619 | int cpu; |
74046494 GBY |
1620 | |
1621 | /* | |
1622 | * Allocate in the BSS so we wont require allocation in | |
1623 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
1624 | */ | |
1625 | static cpumask_t cpus_with_pcps; | |
1626 | ||
1627 | /* | |
1628 | * We don't care about racing with CPU hotplug event | |
1629 | * as offline notification will cause the notified | |
1630 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
1631 | * disables preemption as part of its processing | |
1632 | */ | |
1633 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
1634 | struct per_cpu_pageset *pcp; |
1635 | struct zone *z; | |
74046494 | 1636 | bool has_pcps = false; |
93481ff0 VB |
1637 | |
1638 | if (zone) { | |
74046494 | 1639 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 1640 | if (pcp->pcp.count) |
74046494 | 1641 | has_pcps = true; |
93481ff0 VB |
1642 | } else { |
1643 | for_each_populated_zone(z) { | |
1644 | pcp = per_cpu_ptr(z->pageset, cpu); | |
1645 | if (pcp->pcp.count) { | |
1646 | has_pcps = true; | |
1647 | break; | |
1648 | } | |
74046494 GBY |
1649 | } |
1650 | } | |
93481ff0 | 1651 | |
74046494 GBY |
1652 | if (has_pcps) |
1653 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
1654 | else | |
1655 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
1656 | } | |
93481ff0 VB |
1657 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
1658 | zone, 1); | |
9f8f2172 CL |
1659 | } |
1660 | ||
296699de | 1661 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1662 | |
1663 | void mark_free_pages(struct zone *zone) | |
1664 | { | |
f623f0db RW |
1665 | unsigned long pfn, max_zone_pfn; |
1666 | unsigned long flags; | |
7aeb09f9 | 1667 | unsigned int order, t; |
1da177e4 LT |
1668 | struct list_head *curr; |
1669 | ||
8080fc03 | 1670 | if (zone_is_empty(zone)) |
1da177e4 LT |
1671 | return; |
1672 | ||
1673 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 1674 | |
108bcc96 | 1675 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
1676 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
1677 | if (pfn_valid(pfn)) { | |
1678 | struct page *page = pfn_to_page(pfn); | |
1679 | ||
7be98234 RW |
1680 | if (!swsusp_page_is_forbidden(page)) |
1681 | swsusp_unset_page_free(page); | |
f623f0db | 1682 | } |
1da177e4 | 1683 | |
b2a0ac88 MG |
1684 | for_each_migratetype_order(order, t) { |
1685 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1686 | unsigned long i; |
1da177e4 | 1687 | |
f623f0db RW |
1688 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1689 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1690 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1691 | } |
b2a0ac88 | 1692 | } |
1da177e4 LT |
1693 | spin_unlock_irqrestore(&zone->lock, flags); |
1694 | } | |
e2c55dc8 | 1695 | #endif /* CONFIG_PM */ |
1da177e4 | 1696 | |
1da177e4 LT |
1697 | /* |
1698 | * Free a 0-order page | |
b745bc85 | 1699 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 1700 | */ |
b745bc85 | 1701 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
1702 | { |
1703 | struct zone *zone = page_zone(page); | |
1704 | struct per_cpu_pages *pcp; | |
1705 | unsigned long flags; | |
dc4b0caf | 1706 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 1707 | int migratetype; |
1da177e4 | 1708 | |
ec95f53a | 1709 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1710 | return; |
1711 | ||
dc4b0caf | 1712 | migratetype = get_pfnblock_migratetype(page, pfn); |
b12c4ad1 | 1713 | set_freepage_migratetype(page, migratetype); |
1da177e4 | 1714 | local_irq_save(flags); |
f8891e5e | 1715 | __count_vm_event(PGFREE); |
da456f14 | 1716 | |
5f8dcc21 MG |
1717 | /* |
1718 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1719 | * Free ISOLATE pages back to the allocator because they are being | |
1720 | * offlined but treat RESERVE as movable pages so we can get those | |
1721 | * areas back if necessary. Otherwise, we may have to free | |
1722 | * excessively into the page allocator | |
1723 | */ | |
1724 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 1725 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 1726 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
1727 | goto out; |
1728 | } | |
1729 | migratetype = MIGRATE_MOVABLE; | |
1730 | } | |
1731 | ||
99dcc3e5 | 1732 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 1733 | if (!cold) |
5f8dcc21 | 1734 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
1735 | else |
1736 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 1737 | pcp->count++; |
48db57f8 | 1738 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 1739 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
1740 | free_pcppages_bulk(zone, batch, pcp); |
1741 | pcp->count -= batch; | |
48db57f8 | 1742 | } |
5f8dcc21 MG |
1743 | |
1744 | out: | |
1da177e4 | 1745 | local_irq_restore(flags); |
1da177e4 LT |
1746 | } |
1747 | ||
cc59850e KK |
1748 | /* |
1749 | * Free a list of 0-order pages | |
1750 | */ | |
b745bc85 | 1751 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
1752 | { |
1753 | struct page *page, *next; | |
1754 | ||
1755 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1756 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1757 | free_hot_cold_page(page, cold); |
1758 | } | |
1759 | } | |
1760 | ||
8dfcc9ba NP |
1761 | /* |
1762 | * split_page takes a non-compound higher-order page, and splits it into | |
1763 | * n (1<<order) sub-pages: page[0..n] | |
1764 | * Each sub-page must be freed individually. | |
1765 | * | |
1766 | * Note: this is probably too low level an operation for use in drivers. | |
1767 | * Please consult with lkml before using this in your driver. | |
1768 | */ | |
1769 | void split_page(struct page *page, unsigned int order) | |
1770 | { | |
1771 | int i; | |
1772 | ||
309381fe SL |
1773 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1774 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
1775 | |
1776 | #ifdef CONFIG_KMEMCHECK | |
1777 | /* | |
1778 | * Split shadow pages too, because free(page[0]) would | |
1779 | * otherwise free the whole shadow. | |
1780 | */ | |
1781 | if (kmemcheck_page_is_tracked(page)) | |
1782 | split_page(virt_to_page(page[0].shadow), order); | |
1783 | #endif | |
1784 | ||
48c96a36 JK |
1785 | set_page_owner(page, 0, 0); |
1786 | for (i = 1; i < (1 << order); i++) { | |
7835e98b | 1787 | set_page_refcounted(page + i); |
48c96a36 JK |
1788 | set_page_owner(page + i, 0, 0); |
1789 | } | |
8dfcc9ba | 1790 | } |
5853ff23 | 1791 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 1792 | |
3c605096 | 1793 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 1794 | { |
748446bb MG |
1795 | unsigned long watermark; |
1796 | struct zone *zone; | |
2139cbe6 | 1797 | int mt; |
748446bb MG |
1798 | |
1799 | BUG_ON(!PageBuddy(page)); | |
1800 | ||
1801 | zone = page_zone(page); | |
2e30abd1 | 1802 | mt = get_pageblock_migratetype(page); |
748446bb | 1803 | |
194159fb | 1804 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
1805 | /* Obey watermarks as if the page was being allocated */ |
1806 | watermark = low_wmark_pages(zone) + (1 << order); | |
1807 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1808 | return 0; | |
1809 | ||
8fb74b9f | 1810 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 1811 | } |
748446bb MG |
1812 | |
1813 | /* Remove page from free list */ | |
1814 | list_del(&page->lru); | |
1815 | zone->free_area[order].nr_free--; | |
1816 | rmv_page_order(page); | |
2139cbe6 | 1817 | |
8fb74b9f | 1818 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
1819 | if (order >= pageblock_order - 1) { |
1820 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
1821 | for (; page < endpage; page += pageblock_nr_pages) { |
1822 | int mt = get_pageblock_migratetype(page); | |
194159fb | 1823 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
1824 | set_pageblock_migratetype(page, |
1825 | MIGRATE_MOVABLE); | |
1826 | } | |
748446bb MG |
1827 | } |
1828 | ||
48c96a36 | 1829 | set_page_owner(page, order, 0); |
8fb74b9f | 1830 | return 1UL << order; |
1fb3f8ca MG |
1831 | } |
1832 | ||
1833 | /* | |
1834 | * Similar to split_page except the page is already free. As this is only | |
1835 | * being used for migration, the migratetype of the block also changes. | |
1836 | * As this is called with interrupts disabled, the caller is responsible | |
1837 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1838 | * are enabled. | |
1839 | * | |
1840 | * Note: this is probably too low level an operation for use in drivers. | |
1841 | * Please consult with lkml before using this in your driver. | |
1842 | */ | |
1843 | int split_free_page(struct page *page) | |
1844 | { | |
1845 | unsigned int order; | |
1846 | int nr_pages; | |
1847 | ||
1fb3f8ca MG |
1848 | order = page_order(page); |
1849 | ||
8fb74b9f | 1850 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
1851 | if (!nr_pages) |
1852 | return 0; | |
1853 | ||
1854 | /* Split into individual pages */ | |
1855 | set_page_refcounted(page); | |
1856 | split_page(page, order); | |
1857 | return nr_pages; | |
748446bb MG |
1858 | } |
1859 | ||
1da177e4 | 1860 | /* |
75379191 | 1861 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 1862 | */ |
0a15c3e9 MG |
1863 | static inline |
1864 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
7aeb09f9 MG |
1865 | struct zone *zone, unsigned int order, |
1866 | gfp_t gfp_flags, int migratetype) | |
1da177e4 LT |
1867 | { |
1868 | unsigned long flags; | |
689bcebf | 1869 | struct page *page; |
b745bc85 | 1870 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
1da177e4 | 1871 | |
48db57f8 | 1872 | if (likely(order == 0)) { |
1da177e4 | 1873 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1874 | struct list_head *list; |
1da177e4 | 1875 | |
1da177e4 | 1876 | local_irq_save(flags); |
99dcc3e5 CL |
1877 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1878 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1879 | if (list_empty(list)) { |
535131e6 | 1880 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1881 | pcp->batch, list, |
e084b2d9 | 1882 | migratetype, cold); |
5f8dcc21 | 1883 | if (unlikely(list_empty(list))) |
6fb332fa | 1884 | goto failed; |
535131e6 | 1885 | } |
b92a6edd | 1886 | |
5f8dcc21 MG |
1887 | if (cold) |
1888 | page = list_entry(list->prev, struct page, lru); | |
1889 | else | |
1890 | page = list_entry(list->next, struct page, lru); | |
1891 | ||
b92a6edd MG |
1892 | list_del(&page->lru); |
1893 | pcp->count--; | |
7fb1d9fc | 1894 | } else { |
dab48dab AM |
1895 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1896 | /* | |
1897 | * __GFP_NOFAIL is not to be used in new code. | |
1898 | * | |
1899 | * All __GFP_NOFAIL callers should be fixed so that they | |
1900 | * properly detect and handle allocation failures. | |
1901 | * | |
1902 | * We most definitely don't want callers attempting to | |
4923abf9 | 1903 | * allocate greater than order-1 page units with |
dab48dab AM |
1904 | * __GFP_NOFAIL. |
1905 | */ | |
4923abf9 | 1906 | WARN_ON_ONCE(order > 1); |
dab48dab | 1907 | } |
1da177e4 | 1908 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1909 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1910 | spin_unlock(&zone->lock); |
1911 | if (!page) | |
1912 | goto failed; | |
d1ce749a | 1913 | __mod_zone_freepage_state(zone, -(1 << order), |
5bcc9f86 | 1914 | get_freepage_migratetype(page)); |
1da177e4 LT |
1915 | } |
1916 | ||
3a025760 | 1917 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
abe5f972 | 1918 | if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 && |
57054651 JW |
1919 | !test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) |
1920 | set_bit(ZONE_FAIR_DEPLETED, &zone->flags); | |
27329369 | 1921 | |
f8891e5e | 1922 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1923 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1924 | local_irq_restore(flags); |
1da177e4 | 1925 | |
309381fe | 1926 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 1927 | return page; |
a74609fa NP |
1928 | |
1929 | failed: | |
1930 | local_irq_restore(flags); | |
a74609fa | 1931 | return NULL; |
1da177e4 LT |
1932 | } |
1933 | ||
933e312e AM |
1934 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1935 | ||
b2588c4b | 1936 | static struct { |
933e312e AM |
1937 | struct fault_attr attr; |
1938 | ||
1939 | u32 ignore_gfp_highmem; | |
1940 | u32 ignore_gfp_wait; | |
54114994 | 1941 | u32 min_order; |
933e312e AM |
1942 | } fail_page_alloc = { |
1943 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1944 | .ignore_gfp_wait = 1, |
1945 | .ignore_gfp_highmem = 1, | |
54114994 | 1946 | .min_order = 1, |
933e312e AM |
1947 | }; |
1948 | ||
1949 | static int __init setup_fail_page_alloc(char *str) | |
1950 | { | |
1951 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1952 | } | |
1953 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1954 | ||
deaf386e | 1955 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1956 | { |
54114994 | 1957 | if (order < fail_page_alloc.min_order) |
deaf386e | 1958 | return false; |
933e312e | 1959 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 1960 | return false; |
933e312e | 1961 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 1962 | return false; |
933e312e | 1963 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
deaf386e | 1964 | return false; |
933e312e AM |
1965 | |
1966 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1967 | } | |
1968 | ||
1969 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1970 | ||
1971 | static int __init fail_page_alloc_debugfs(void) | |
1972 | { | |
f4ae40a6 | 1973 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1974 | struct dentry *dir; |
933e312e | 1975 | |
dd48c085 AM |
1976 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1977 | &fail_page_alloc.attr); | |
1978 | if (IS_ERR(dir)) | |
1979 | return PTR_ERR(dir); | |
933e312e | 1980 | |
b2588c4b AM |
1981 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1982 | &fail_page_alloc.ignore_gfp_wait)) | |
1983 | goto fail; | |
1984 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1985 | &fail_page_alloc.ignore_gfp_highmem)) | |
1986 | goto fail; | |
1987 | if (!debugfs_create_u32("min-order", mode, dir, | |
1988 | &fail_page_alloc.min_order)) | |
1989 | goto fail; | |
1990 | ||
1991 | return 0; | |
1992 | fail: | |
dd48c085 | 1993 | debugfs_remove_recursive(dir); |
933e312e | 1994 | |
b2588c4b | 1995 | return -ENOMEM; |
933e312e AM |
1996 | } |
1997 | ||
1998 | late_initcall(fail_page_alloc_debugfs); | |
1999 | ||
2000 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2001 | ||
2002 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2003 | ||
deaf386e | 2004 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2005 | { |
deaf386e | 2006 | return false; |
933e312e AM |
2007 | } |
2008 | ||
2009 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2010 | ||
1da177e4 | 2011 | /* |
88f5acf8 | 2012 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
2013 | * of the allocation. |
2014 | */ | |
7aeb09f9 MG |
2015 | static bool __zone_watermark_ok(struct zone *z, unsigned int order, |
2016 | unsigned long mark, int classzone_idx, int alloc_flags, | |
2017 | long free_pages) | |
1da177e4 | 2018 | { |
26086de3 | 2019 | /* free_pages may go negative - that's OK */ |
d23ad423 | 2020 | long min = mark; |
1da177e4 | 2021 | int o; |
026b0814 | 2022 | long free_cma = 0; |
1da177e4 | 2023 | |
df0a6daa | 2024 | free_pages -= (1 << order) - 1; |
7fb1d9fc | 2025 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2026 | min -= min / 2; |
7fb1d9fc | 2027 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 | 2028 | min -= min / 4; |
d95ea5d1 BZ |
2029 | #ifdef CONFIG_CMA |
2030 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2031 | if (!(alloc_flags & ALLOC_CMA)) | |
026b0814 | 2032 | free_cma = zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2033 | #endif |
026b0814 | 2034 | |
3484b2de | 2035 | if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx]) |
88f5acf8 | 2036 | return false; |
1da177e4 LT |
2037 | for (o = 0; o < order; o++) { |
2038 | /* At the next order, this order's pages become unavailable */ | |
2039 | free_pages -= z->free_area[o].nr_free << o; | |
2040 | ||
2041 | /* Require fewer higher order pages to be free */ | |
2042 | min >>= 1; | |
2043 | ||
2044 | if (free_pages <= min) | |
88f5acf8 | 2045 | return false; |
1da177e4 | 2046 | } |
88f5acf8 MG |
2047 | return true; |
2048 | } | |
2049 | ||
7aeb09f9 | 2050 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
88f5acf8 MG |
2051 | int classzone_idx, int alloc_flags) |
2052 | { | |
2053 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2054 | zone_page_state(z, NR_FREE_PAGES)); | |
2055 | } | |
2056 | ||
7aeb09f9 MG |
2057 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
2058 | unsigned long mark, int classzone_idx, int alloc_flags) | |
88f5acf8 MG |
2059 | { |
2060 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2061 | ||
2062 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2063 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2064 | ||
2065 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2066 | free_pages); | |
1da177e4 LT |
2067 | } |
2068 | ||
9276b1bc PJ |
2069 | #ifdef CONFIG_NUMA |
2070 | /* | |
2071 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
2072 | * skip over zones that are not allowed by the cpuset, or that have | |
2073 | * been recently (in last second) found to be nearly full. See further | |
2074 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 2075 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc | 2076 | * |
a1aeb65a | 2077 | * If the zonelist cache is present in the passed zonelist, then |
9276b1bc | 2078 | * returns a pointer to the allowed node mask (either the current |
4b0ef1fe | 2079 | * tasks mems_allowed, or node_states[N_MEMORY].) |
9276b1bc PJ |
2080 | * |
2081 | * If the zonelist cache is not available for this zonelist, does | |
2082 | * nothing and returns NULL. | |
2083 | * | |
2084 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
2085 | * a second since last zap'd) then we zap it out (clear its bits.) | |
2086 | * | |
2087 | * We hold off even calling zlc_setup, until after we've checked the | |
2088 | * first zone in the zonelist, on the theory that most allocations will | |
2089 | * be satisfied from that first zone, so best to examine that zone as | |
2090 | * quickly as we can. | |
2091 | */ | |
2092 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
2093 | { | |
2094 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
2095 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
2096 | ||
2097 | zlc = zonelist->zlcache_ptr; | |
2098 | if (!zlc) | |
2099 | return NULL; | |
2100 | ||
f05111f5 | 2101 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
2102 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
2103 | zlc->last_full_zap = jiffies; | |
2104 | } | |
2105 | ||
2106 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
2107 | &cpuset_current_mems_allowed : | |
4b0ef1fe | 2108 | &node_states[N_MEMORY]; |
9276b1bc PJ |
2109 | return allowednodes; |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
2114 | * if it is worth looking at further for free memory: | |
2115 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
2116 | * bit set in the zonelist_cache fullzones BITMAP). | |
2117 | * 2) Check that the zones node (obtained from the zonelist_cache | |
2118 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
2119 | * Return true (non-zero) if zone is worth looking at further, or | |
2120 | * else return false (zero) if it is not. | |
2121 | * | |
2122 | * This check -ignores- the distinction between various watermarks, | |
2123 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
2124 | * found to be full for any variation of these watermarks, it will | |
2125 | * be considered full for up to one second by all requests, unless | |
2126 | * we are so low on memory on all allowed nodes that we are forced | |
2127 | * into the second scan of the zonelist. | |
2128 | * | |
2129 | * In the second scan we ignore this zonelist cache and exactly | |
2130 | * apply the watermarks to all zones, even it is slower to do so. | |
2131 | * We are low on memory in the second scan, and should leave no stone | |
2132 | * unturned looking for a free page. | |
2133 | */ | |
dd1a239f | 2134 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
2135 | nodemask_t *allowednodes) |
2136 | { | |
2137 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
2138 | int i; /* index of *z in zonelist zones */ | |
2139 | int n; /* node that zone *z is on */ | |
2140 | ||
2141 | zlc = zonelist->zlcache_ptr; | |
2142 | if (!zlc) | |
2143 | return 1; | |
2144 | ||
dd1a239f | 2145 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
2146 | n = zlc->z_to_n[i]; |
2147 | ||
2148 | /* This zone is worth trying if it is allowed but not full */ | |
2149 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
2150 | } | |
2151 | ||
2152 | /* | |
2153 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
2154 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
2155 | * from that zone don't waste time re-examining it. | |
2156 | */ | |
dd1a239f | 2157 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
2158 | { |
2159 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
2160 | int i; /* index of *z in zonelist zones */ | |
2161 | ||
2162 | zlc = zonelist->zlcache_ptr; | |
2163 | if (!zlc) | |
2164 | return; | |
2165 | ||
dd1a239f | 2166 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
2167 | |
2168 | set_bit(i, zlc->fullzones); | |
2169 | } | |
2170 | ||
76d3fbf8 MG |
2171 | /* |
2172 | * clear all zones full, called after direct reclaim makes progress so that | |
2173 | * a zone that was recently full is not skipped over for up to a second | |
2174 | */ | |
2175 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
2176 | { | |
2177 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
2178 | ||
2179 | zlc = zonelist->zlcache_ptr; | |
2180 | if (!zlc) | |
2181 | return; | |
2182 | ||
2183 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
2184 | } | |
2185 | ||
81c0a2bb JW |
2186 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2187 | { | |
fff4068c | 2188 | return local_zone->node == zone->node; |
81c0a2bb JW |
2189 | } |
2190 | ||
957f822a DR |
2191 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2192 | { | |
5f7a75ac MG |
2193 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
2194 | RECLAIM_DISTANCE; | |
957f822a DR |
2195 | } |
2196 | ||
9276b1bc PJ |
2197 | #else /* CONFIG_NUMA */ |
2198 | ||
2199 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
2200 | { | |
2201 | return NULL; | |
2202 | } | |
2203 | ||
dd1a239f | 2204 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
2205 | nodemask_t *allowednodes) |
2206 | { | |
2207 | return 1; | |
2208 | } | |
2209 | ||
dd1a239f | 2210 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
2211 | { |
2212 | } | |
76d3fbf8 MG |
2213 | |
2214 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
2215 | { | |
2216 | } | |
957f822a | 2217 | |
81c0a2bb JW |
2218 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2219 | { | |
2220 | return true; | |
2221 | } | |
2222 | ||
957f822a DR |
2223 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2224 | { | |
2225 | return true; | |
2226 | } | |
2227 | ||
9276b1bc PJ |
2228 | #endif /* CONFIG_NUMA */ |
2229 | ||
4ffeaf35 MG |
2230 | static void reset_alloc_batches(struct zone *preferred_zone) |
2231 | { | |
2232 | struct zone *zone = preferred_zone->zone_pgdat->node_zones; | |
2233 | ||
2234 | do { | |
2235 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
2236 | high_wmark_pages(zone) - low_wmark_pages(zone) - | |
2237 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
57054651 | 2238 | clear_bit(ZONE_FAIR_DEPLETED, &zone->flags); |
4ffeaf35 MG |
2239 | } while (zone++ != preferred_zone); |
2240 | } | |
2241 | ||
7fb1d9fc | 2242 | /* |
0798e519 | 2243 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
2244 | * a page. |
2245 | */ | |
2246 | static struct page * | |
a9263751 VB |
2247 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
2248 | const struct alloc_context *ac) | |
753ee728 | 2249 | { |
a9263751 | 2250 | struct zonelist *zonelist = ac->zonelist; |
dd1a239f | 2251 | struct zoneref *z; |
7fb1d9fc | 2252 | struct page *page = NULL; |
5117f45d | 2253 | struct zone *zone; |
9276b1bc PJ |
2254 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
2255 | int zlc_active = 0; /* set if using zonelist_cache */ | |
2256 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
a6e21b14 MG |
2257 | bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) && |
2258 | (gfp_mask & __GFP_WRITE); | |
4ffeaf35 MG |
2259 | int nr_fair_skipped = 0; |
2260 | bool zonelist_rescan; | |
54a6eb5c | 2261 | |
9276b1bc | 2262 | zonelist_scan: |
4ffeaf35 MG |
2263 | zonelist_rescan = false; |
2264 | ||
7fb1d9fc | 2265 | /* |
9276b1bc | 2266 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 2267 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 2268 | */ |
a9263751 VB |
2269 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, |
2270 | ac->nodemask) { | |
e085dbc5 JW |
2271 | unsigned long mark; |
2272 | ||
e5adfffc | 2273 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
9276b1bc PJ |
2274 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
2275 | continue; | |
664eedde MG |
2276 | if (cpusets_enabled() && |
2277 | (alloc_flags & ALLOC_CPUSET) && | |
344736f2 | 2278 | !cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 2279 | continue; |
81c0a2bb JW |
2280 | /* |
2281 | * Distribute pages in proportion to the individual | |
2282 | * zone size to ensure fair page aging. The zone a | |
2283 | * page was allocated in should have no effect on the | |
2284 | * time the page has in memory before being reclaimed. | |
81c0a2bb | 2285 | */ |
3a025760 | 2286 | if (alloc_flags & ALLOC_FAIR) { |
a9263751 | 2287 | if (!zone_local(ac->preferred_zone, zone)) |
f7b5d647 | 2288 | break; |
57054651 | 2289 | if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) { |
4ffeaf35 | 2290 | nr_fair_skipped++; |
3a025760 | 2291 | continue; |
4ffeaf35 | 2292 | } |
81c0a2bb | 2293 | } |
a756cf59 JW |
2294 | /* |
2295 | * When allocating a page cache page for writing, we | |
2296 | * want to get it from a zone that is within its dirty | |
2297 | * limit, such that no single zone holds more than its | |
2298 | * proportional share of globally allowed dirty pages. | |
2299 | * The dirty limits take into account the zone's | |
2300 | * lowmem reserves and high watermark so that kswapd | |
2301 | * should be able to balance it without having to | |
2302 | * write pages from its LRU list. | |
2303 | * | |
2304 | * This may look like it could increase pressure on | |
2305 | * lower zones by failing allocations in higher zones | |
2306 | * before they are full. But the pages that do spill | |
2307 | * over are limited as the lower zones are protected | |
2308 | * by this very same mechanism. It should not become | |
2309 | * a practical burden to them. | |
2310 | * | |
2311 | * XXX: For now, allow allocations to potentially | |
2312 | * exceed the per-zone dirty limit in the slowpath | |
2313 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
2314 | * which is important when on a NUMA setup the allowed | |
2315 | * zones are together not big enough to reach the | |
2316 | * global limit. The proper fix for these situations | |
2317 | * will require awareness of zones in the | |
2318 | * dirty-throttling and the flusher threads. | |
2319 | */ | |
a6e21b14 | 2320 | if (consider_zone_dirty && !zone_dirty_ok(zone)) |
800a1e75 | 2321 | continue; |
7fb1d9fc | 2322 | |
e085dbc5 JW |
2323 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
2324 | if (!zone_watermark_ok(zone, order, mark, | |
a9263751 | 2325 | ac->classzone_idx, alloc_flags)) { |
fa5e084e MG |
2326 | int ret; |
2327 | ||
5dab2911 MG |
2328 | /* Checked here to keep the fast path fast */ |
2329 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
2330 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
2331 | goto try_this_zone; | |
2332 | ||
e5adfffc KS |
2333 | if (IS_ENABLED(CONFIG_NUMA) && |
2334 | !did_zlc_setup && nr_online_nodes > 1) { | |
cd38b115 MG |
2335 | /* |
2336 | * we do zlc_setup if there are multiple nodes | |
2337 | * and before considering the first zone allowed | |
2338 | * by the cpuset. | |
2339 | */ | |
2340 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
2341 | zlc_active = 1; | |
2342 | did_zlc_setup = 1; | |
2343 | } | |
2344 | ||
957f822a | 2345 | if (zone_reclaim_mode == 0 || |
a9263751 | 2346 | !zone_allows_reclaim(ac->preferred_zone, zone)) |
fa5e084e MG |
2347 | goto this_zone_full; |
2348 | ||
cd38b115 MG |
2349 | /* |
2350 | * As we may have just activated ZLC, check if the first | |
2351 | * eligible zone has failed zone_reclaim recently. | |
2352 | */ | |
e5adfffc | 2353 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
cd38b115 MG |
2354 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
2355 | continue; | |
2356 | ||
fa5e084e MG |
2357 | ret = zone_reclaim(zone, gfp_mask, order); |
2358 | switch (ret) { | |
2359 | case ZONE_RECLAIM_NOSCAN: | |
2360 | /* did not scan */ | |
cd38b115 | 2361 | continue; |
fa5e084e MG |
2362 | case ZONE_RECLAIM_FULL: |
2363 | /* scanned but unreclaimable */ | |
cd38b115 | 2364 | continue; |
fa5e084e MG |
2365 | default: |
2366 | /* did we reclaim enough */ | |
fed2719e | 2367 | if (zone_watermark_ok(zone, order, mark, |
a9263751 | 2368 | ac->classzone_idx, alloc_flags)) |
fed2719e MG |
2369 | goto try_this_zone; |
2370 | ||
2371 | /* | |
2372 | * Failed to reclaim enough to meet watermark. | |
2373 | * Only mark the zone full if checking the min | |
2374 | * watermark or if we failed to reclaim just | |
2375 | * 1<<order pages or else the page allocator | |
2376 | * fastpath will prematurely mark zones full | |
2377 | * when the watermark is between the low and | |
2378 | * min watermarks. | |
2379 | */ | |
2380 | if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) || | |
2381 | ret == ZONE_RECLAIM_SOME) | |
9276b1bc | 2382 | goto this_zone_full; |
fed2719e MG |
2383 | |
2384 | continue; | |
0798e519 | 2385 | } |
7fb1d9fc RS |
2386 | } |
2387 | ||
fa5e084e | 2388 | try_this_zone: |
a9263751 VB |
2389 | page = buffered_rmqueue(ac->preferred_zone, zone, order, |
2390 | gfp_mask, ac->migratetype); | |
75379191 VB |
2391 | if (page) { |
2392 | if (prep_new_page(page, order, gfp_mask, alloc_flags)) | |
2393 | goto try_this_zone; | |
2394 | return page; | |
2395 | } | |
9276b1bc | 2396 | this_zone_full: |
65bb3719 | 2397 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active) |
9276b1bc | 2398 | zlc_mark_zone_full(zonelist, z); |
54a6eb5c | 2399 | } |
9276b1bc | 2400 | |
4ffeaf35 MG |
2401 | /* |
2402 | * The first pass makes sure allocations are spread fairly within the | |
2403 | * local node. However, the local node might have free pages left | |
2404 | * after the fairness batches are exhausted, and remote zones haven't | |
2405 | * even been considered yet. Try once more without fairness, and | |
2406 | * include remote zones now, before entering the slowpath and waking | |
2407 | * kswapd: prefer spilling to a remote zone over swapping locally. | |
2408 | */ | |
2409 | if (alloc_flags & ALLOC_FAIR) { | |
2410 | alloc_flags &= ~ALLOC_FAIR; | |
2411 | if (nr_fair_skipped) { | |
2412 | zonelist_rescan = true; | |
a9263751 | 2413 | reset_alloc_batches(ac->preferred_zone); |
4ffeaf35 MG |
2414 | } |
2415 | if (nr_online_nodes > 1) | |
2416 | zonelist_rescan = true; | |
2417 | } | |
2418 | ||
2419 | if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) { | |
2420 | /* Disable zlc cache for second zonelist scan */ | |
2421 | zlc_active = 0; | |
2422 | zonelist_rescan = true; | |
2423 | } | |
2424 | ||
2425 | if (zonelist_rescan) | |
2426 | goto zonelist_scan; | |
2427 | ||
2428 | return NULL; | |
753ee728 MH |
2429 | } |
2430 | ||
29423e77 DR |
2431 | /* |
2432 | * Large machines with many possible nodes should not always dump per-node | |
2433 | * meminfo in irq context. | |
2434 | */ | |
2435 | static inline bool should_suppress_show_mem(void) | |
2436 | { | |
2437 | bool ret = false; | |
2438 | ||
2439 | #if NODES_SHIFT > 8 | |
2440 | ret = in_interrupt(); | |
2441 | #endif | |
2442 | return ret; | |
2443 | } | |
2444 | ||
a238ab5b DH |
2445 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2446 | DEFAULT_RATELIMIT_INTERVAL, | |
2447 | DEFAULT_RATELIMIT_BURST); | |
2448 | ||
2449 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
2450 | { | |
a238ab5b DH |
2451 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2452 | ||
c0a32fc5 SG |
2453 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2454 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2455 | return; |
2456 | ||
2457 | /* | |
2458 | * This documents exceptions given to allocations in certain | |
2459 | * contexts that are allowed to allocate outside current's set | |
2460 | * of allowed nodes. | |
2461 | */ | |
2462 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2463 | if (test_thread_flag(TIF_MEMDIE) || | |
2464 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2465 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2466 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
2467 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2468 | ||
2469 | if (fmt) { | |
3ee9a4f0 JP |
2470 | struct va_format vaf; |
2471 | va_list args; | |
2472 | ||
a238ab5b | 2473 | va_start(args, fmt); |
3ee9a4f0 JP |
2474 | |
2475 | vaf.fmt = fmt; | |
2476 | vaf.va = &args; | |
2477 | ||
2478 | pr_warn("%pV", &vaf); | |
2479 | ||
a238ab5b DH |
2480 | va_end(args); |
2481 | } | |
2482 | ||
3ee9a4f0 JP |
2483 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
2484 | current->comm, order, gfp_mask); | |
a238ab5b DH |
2485 | |
2486 | dump_stack(); | |
2487 | if (!should_suppress_show_mem()) | |
2488 | show_mem(filter); | |
2489 | } | |
2490 | ||
11e33f6a MG |
2491 | static inline struct page * |
2492 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 2493 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a MG |
2494 | { |
2495 | struct page *page; | |
2496 | ||
9879de73 JW |
2497 | *did_some_progress = 0; |
2498 | ||
9879de73 | 2499 | /* |
dc56401f JW |
2500 | * Acquire the oom lock. If that fails, somebody else is |
2501 | * making progress for us. | |
9879de73 | 2502 | */ |
dc56401f | 2503 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 2504 | *did_some_progress = 1; |
11e33f6a | 2505 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2506 | return NULL; |
2507 | } | |
6b1de916 | 2508 | |
11e33f6a MG |
2509 | /* |
2510 | * Go through the zonelist yet one more time, keep very high watermark | |
2511 | * here, this is only to catch a parallel oom killing, we must fail if | |
2512 | * we're still under heavy pressure. | |
2513 | */ | |
a9263751 VB |
2514 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
2515 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 2516 | if (page) |
11e33f6a MG |
2517 | goto out; |
2518 | ||
4365a567 | 2519 | if (!(gfp_mask & __GFP_NOFAIL)) { |
9879de73 JW |
2520 | /* Coredumps can quickly deplete all memory reserves */ |
2521 | if (current->flags & PF_DUMPCORE) | |
2522 | goto out; | |
4365a567 KH |
2523 | /* The OOM killer will not help higher order allocs */ |
2524 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2525 | goto out; | |
03668b3c | 2526 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
a9263751 | 2527 | if (ac->high_zoneidx < ZONE_NORMAL) |
03668b3c | 2528 | goto out; |
9083905a | 2529 | /* The OOM killer does not compensate for IO-less reclaim */ |
cc873177 JW |
2530 | if (!(gfp_mask & __GFP_FS)) { |
2531 | /* | |
2532 | * XXX: Page reclaim didn't yield anything, | |
2533 | * and the OOM killer can't be invoked, but | |
9083905a | 2534 | * keep looping as per tradition. |
cc873177 JW |
2535 | */ |
2536 | *did_some_progress = 1; | |
9879de73 | 2537 | goto out; |
cc873177 | 2538 | } |
9083905a JW |
2539 | if (pm_suspended_storage()) |
2540 | goto out; | |
4167e9b2 | 2541 | /* The OOM killer may not free memory on a specific node */ |
4365a567 KH |
2542 | if (gfp_mask & __GFP_THISNODE) |
2543 | goto out; | |
2544 | } | |
11e33f6a | 2545 | /* Exhausted what can be done so it's blamo time */ |
e009d5dc MH |
2546 | if (out_of_memory(ac->zonelist, gfp_mask, order, ac->nodemask, false) |
2547 | || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) | |
c32b3cbe | 2548 | *did_some_progress = 1; |
11e33f6a | 2549 | out: |
dc56401f | 2550 | mutex_unlock(&oom_lock); |
11e33f6a MG |
2551 | return page; |
2552 | } | |
2553 | ||
56de7263 MG |
2554 | #ifdef CONFIG_COMPACTION |
2555 | /* Try memory compaction for high-order allocations before reclaim */ | |
2556 | static struct page * | |
2557 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
a9263751 VB |
2558 | int alloc_flags, const struct alloc_context *ac, |
2559 | enum migrate_mode mode, int *contended_compaction, | |
2560 | bool *deferred_compaction) | |
56de7263 | 2561 | { |
53853e2d | 2562 | unsigned long compact_result; |
98dd3b48 | 2563 | struct page *page; |
53853e2d VB |
2564 | |
2565 | if (!order) | |
66199712 | 2566 | return NULL; |
66199712 | 2567 | |
c06b1fca | 2568 | current->flags |= PF_MEMALLOC; |
1a6d53a1 VB |
2569 | compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
2570 | mode, contended_compaction); | |
c06b1fca | 2571 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2572 | |
98dd3b48 VB |
2573 | switch (compact_result) { |
2574 | case COMPACT_DEFERRED: | |
53853e2d | 2575 | *deferred_compaction = true; |
98dd3b48 VB |
2576 | /* fall-through */ |
2577 | case COMPACT_SKIPPED: | |
2578 | return NULL; | |
2579 | default: | |
2580 | break; | |
2581 | } | |
53853e2d | 2582 | |
98dd3b48 VB |
2583 | /* |
2584 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
2585 | * count a compaction stall | |
2586 | */ | |
2587 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 2588 | |
a9263751 VB |
2589 | page = get_page_from_freelist(gfp_mask, order, |
2590 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
53853e2d | 2591 | |
98dd3b48 VB |
2592 | if (page) { |
2593 | struct zone *zone = page_zone(page); | |
53853e2d | 2594 | |
98dd3b48 VB |
2595 | zone->compact_blockskip_flush = false; |
2596 | compaction_defer_reset(zone, order, true); | |
2597 | count_vm_event(COMPACTSUCCESS); | |
2598 | return page; | |
2599 | } | |
56de7263 | 2600 | |
98dd3b48 VB |
2601 | /* |
2602 | * It's bad if compaction run occurs and fails. The most likely reason | |
2603 | * is that pages exist, but not enough to satisfy watermarks. | |
2604 | */ | |
2605 | count_vm_event(COMPACTFAIL); | |
66199712 | 2606 | |
98dd3b48 | 2607 | cond_resched(); |
56de7263 MG |
2608 | |
2609 | return NULL; | |
2610 | } | |
2611 | #else | |
2612 | static inline struct page * | |
2613 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
a9263751 VB |
2614 | int alloc_flags, const struct alloc_context *ac, |
2615 | enum migrate_mode mode, int *contended_compaction, | |
2616 | bool *deferred_compaction) | |
56de7263 MG |
2617 | { |
2618 | return NULL; | |
2619 | } | |
2620 | #endif /* CONFIG_COMPACTION */ | |
2621 | ||
bba90710 MS |
2622 | /* Perform direct synchronous page reclaim */ |
2623 | static int | |
a9263751 VB |
2624 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
2625 | const struct alloc_context *ac) | |
11e33f6a | 2626 | { |
11e33f6a | 2627 | struct reclaim_state reclaim_state; |
bba90710 | 2628 | int progress; |
11e33f6a MG |
2629 | |
2630 | cond_resched(); | |
2631 | ||
2632 | /* We now go into synchronous reclaim */ | |
2633 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2634 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2635 | lockdep_set_current_reclaim_state(gfp_mask); |
2636 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2637 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2638 | |
a9263751 VB |
2639 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
2640 | ac->nodemask); | |
11e33f6a | 2641 | |
c06b1fca | 2642 | current->reclaim_state = NULL; |
11e33f6a | 2643 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2644 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2645 | |
2646 | cond_resched(); | |
2647 | ||
bba90710 MS |
2648 | return progress; |
2649 | } | |
2650 | ||
2651 | /* The really slow allocator path where we enter direct reclaim */ | |
2652 | static inline struct page * | |
2653 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
a9263751 VB |
2654 | int alloc_flags, const struct alloc_context *ac, |
2655 | unsigned long *did_some_progress) | |
bba90710 MS |
2656 | { |
2657 | struct page *page = NULL; | |
2658 | bool drained = false; | |
2659 | ||
a9263751 | 2660 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
2661 | if (unlikely(!(*did_some_progress))) |
2662 | return NULL; | |
11e33f6a | 2663 | |
76d3fbf8 | 2664 | /* After successful reclaim, reconsider all zones for allocation */ |
e5adfffc | 2665 | if (IS_ENABLED(CONFIG_NUMA)) |
a9263751 | 2666 | zlc_clear_zones_full(ac->zonelist); |
76d3fbf8 | 2667 | |
9ee493ce | 2668 | retry: |
a9263751 VB |
2669 | page = get_page_from_freelist(gfp_mask, order, |
2670 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
9ee493ce MG |
2671 | |
2672 | /* | |
2673 | * If an allocation failed after direct reclaim, it could be because | |
2674 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2675 | */ | |
2676 | if (!page && !drained) { | |
93481ff0 | 2677 | drain_all_pages(NULL); |
9ee493ce MG |
2678 | drained = true; |
2679 | goto retry; | |
2680 | } | |
2681 | ||
11e33f6a MG |
2682 | return page; |
2683 | } | |
2684 | ||
1da177e4 | 2685 | /* |
11e33f6a MG |
2686 | * This is called in the allocator slow-path if the allocation request is of |
2687 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2688 | */ |
11e33f6a MG |
2689 | static inline struct page * |
2690 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 2691 | const struct alloc_context *ac) |
11e33f6a MG |
2692 | { |
2693 | struct page *page; | |
2694 | ||
2695 | do { | |
a9263751 VB |
2696 | page = get_page_from_freelist(gfp_mask, order, |
2697 | ALLOC_NO_WATERMARKS, ac); | |
11e33f6a MG |
2698 | |
2699 | if (!page && gfp_mask & __GFP_NOFAIL) | |
a9263751 VB |
2700 | wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, |
2701 | HZ/50); | |
11e33f6a MG |
2702 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2703 | ||
2704 | return page; | |
2705 | } | |
2706 | ||
a9263751 | 2707 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
2708 | { |
2709 | struct zoneref *z; | |
2710 | struct zone *zone; | |
2711 | ||
a9263751 VB |
2712 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
2713 | ac->high_zoneidx, ac->nodemask) | |
2714 | wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone)); | |
3a025760 JW |
2715 | } |
2716 | ||
341ce06f PZ |
2717 | static inline int |
2718 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2719 | { | |
341ce06f | 2720 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
b104a35d | 2721 | const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD)); |
1da177e4 | 2722 | |
a56f57ff | 2723 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2724 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2725 | |
341ce06f PZ |
2726 | /* |
2727 | * The caller may dip into page reserves a bit more if the caller | |
2728 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2729 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
b104a35d | 2730 | * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 2731 | */ |
e6223a3b | 2732 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2733 | |
b104a35d | 2734 | if (atomic) { |
5c3240d9 | 2735 | /* |
b104a35d DR |
2736 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
2737 | * if it can't schedule. | |
5c3240d9 | 2738 | */ |
b104a35d | 2739 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 2740 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 2741 | /* |
b104a35d | 2742 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 2743 | * comment for __cpuset_node_allowed(). |
523b9458 | 2744 | */ |
341ce06f | 2745 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2746 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2747 | alloc_flags |= ALLOC_HARDER; |
2748 | ||
b37f1dd0 MG |
2749 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
2750 | if (gfp_mask & __GFP_MEMALLOC) | |
2751 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
2752 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
2753 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
2754 | else if (!in_interrupt() && | |
2755 | ((current->flags & PF_MEMALLOC) || | |
2756 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 2757 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 2758 | } |
d95ea5d1 | 2759 | #ifdef CONFIG_CMA |
43e7a34d | 2760 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
2761 | alloc_flags |= ALLOC_CMA; |
2762 | #endif | |
341ce06f PZ |
2763 | return alloc_flags; |
2764 | } | |
2765 | ||
072bb0aa MG |
2766 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
2767 | { | |
b37f1dd0 | 2768 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
2769 | } |
2770 | ||
11e33f6a MG |
2771 | static inline struct page * |
2772 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 2773 | struct alloc_context *ac) |
11e33f6a MG |
2774 | { |
2775 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2776 | struct page *page = NULL; | |
2777 | int alloc_flags; | |
2778 | unsigned long pages_reclaimed = 0; | |
2779 | unsigned long did_some_progress; | |
e0b9daeb | 2780 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
66199712 | 2781 | bool deferred_compaction = false; |
1f9efdef | 2782 | int contended_compaction = COMPACT_CONTENDED_NONE; |
1da177e4 | 2783 | |
72807a74 MG |
2784 | /* |
2785 | * In the slowpath, we sanity check order to avoid ever trying to | |
2786 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2787 | * be using allocators in order of preference for an area that is | |
2788 | * too large. | |
2789 | */ | |
1fc28b70 MG |
2790 | if (order >= MAX_ORDER) { |
2791 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2792 | return NULL; |
1fc28b70 | 2793 | } |
1da177e4 | 2794 | |
952f3b51 | 2795 | /* |
4167e9b2 DR |
2796 | * If this allocation cannot block and it is for a specific node, then |
2797 | * fail early. There's no need to wakeup kswapd or retry for a | |
2798 | * speculative node-specific allocation. | |
952f3b51 | 2799 | */ |
4167e9b2 | 2800 | if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !wait) |
952f3b51 CL |
2801 | goto nopage; |
2802 | ||
9879de73 | 2803 | retry: |
3a025760 | 2804 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
a9263751 | 2805 | wake_all_kswapds(order, ac); |
1da177e4 | 2806 | |
9bf2229f | 2807 | /* |
7fb1d9fc RS |
2808 | * OK, we're below the kswapd watermark and have kicked background |
2809 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2810 | * to how we want to proceed. | |
9bf2229f | 2811 | */ |
341ce06f | 2812 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2813 | |
f33261d7 DR |
2814 | /* |
2815 | * Find the true preferred zone if the allocation is unconstrained by | |
2816 | * cpusets. | |
2817 | */ | |
a9263751 | 2818 | if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) { |
d8846374 | 2819 | struct zoneref *preferred_zoneref; |
a9263751 VB |
2820 | preferred_zoneref = first_zones_zonelist(ac->zonelist, |
2821 | ac->high_zoneidx, NULL, &ac->preferred_zone); | |
2822 | ac->classzone_idx = zonelist_zone_idx(preferred_zoneref); | |
d8846374 | 2823 | } |
f33261d7 | 2824 | |
341ce06f | 2825 | /* This is the last chance, in general, before the goto nopage. */ |
a9263751 VB |
2826 | page = get_page_from_freelist(gfp_mask, order, |
2827 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
7fb1d9fc RS |
2828 | if (page) |
2829 | goto got_pg; | |
1da177e4 | 2830 | |
11e33f6a | 2831 | /* Allocate without watermarks if the context allows */ |
341ce06f | 2832 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
2833 | /* |
2834 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
2835 | * the allocation is high priority and these type of | |
2836 | * allocations are system rather than user orientated | |
2837 | */ | |
a9263751 VB |
2838 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
2839 | ||
2840 | page = __alloc_pages_high_priority(gfp_mask, order, ac); | |
183f6371 | 2841 | |
cfd19c5a | 2842 | if (page) { |
341ce06f | 2843 | goto got_pg; |
cfd19c5a | 2844 | } |
1da177e4 LT |
2845 | } |
2846 | ||
2847 | /* Atomic allocations - we can't balance anything */ | |
aed0a0e3 DR |
2848 | if (!wait) { |
2849 | /* | |
2850 | * All existing users of the deprecated __GFP_NOFAIL are | |
2851 | * blockable, so warn of any new users that actually allow this | |
2852 | * type of allocation to fail. | |
2853 | */ | |
2854 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 2855 | goto nopage; |
aed0a0e3 | 2856 | } |
1da177e4 | 2857 | |
341ce06f | 2858 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2859 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2860 | goto nopage; |
2861 | ||
6583bb64 DR |
2862 | /* Avoid allocations with no watermarks from looping endlessly */ |
2863 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2864 | goto nopage; | |
2865 | ||
77f1fe6b MG |
2866 | /* |
2867 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2868 | * attempts after direct reclaim are synchronous | |
2869 | */ | |
a9263751 VB |
2870 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
2871 | migration_mode, | |
2872 | &contended_compaction, | |
53853e2d | 2873 | &deferred_compaction); |
56de7263 MG |
2874 | if (page) |
2875 | goto got_pg; | |
75f30861 | 2876 | |
1f9efdef VB |
2877 | /* Checks for THP-specific high-order allocations */ |
2878 | if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) { | |
2879 | /* | |
2880 | * If compaction is deferred for high-order allocations, it is | |
2881 | * because sync compaction recently failed. If this is the case | |
2882 | * and the caller requested a THP allocation, we do not want | |
2883 | * to heavily disrupt the system, so we fail the allocation | |
2884 | * instead of entering direct reclaim. | |
2885 | */ | |
2886 | if (deferred_compaction) | |
2887 | goto nopage; | |
2888 | ||
2889 | /* | |
2890 | * In all zones where compaction was attempted (and not | |
2891 | * deferred or skipped), lock contention has been detected. | |
2892 | * For THP allocation we do not want to disrupt the others | |
2893 | * so we fallback to base pages instead. | |
2894 | */ | |
2895 | if (contended_compaction == COMPACT_CONTENDED_LOCK) | |
2896 | goto nopage; | |
2897 | ||
2898 | /* | |
2899 | * If compaction was aborted due to need_resched(), we do not | |
2900 | * want to further increase allocation latency, unless it is | |
2901 | * khugepaged trying to collapse. | |
2902 | */ | |
2903 | if (contended_compaction == COMPACT_CONTENDED_SCHED | |
2904 | && !(current->flags & PF_KTHREAD)) | |
2905 | goto nopage; | |
2906 | } | |
66199712 | 2907 | |
8fe78048 DR |
2908 | /* |
2909 | * It can become very expensive to allocate transparent hugepages at | |
2910 | * fault, so use asynchronous memory compaction for THP unless it is | |
2911 | * khugepaged trying to collapse. | |
2912 | */ | |
2913 | if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE || | |
2914 | (current->flags & PF_KTHREAD)) | |
2915 | migration_mode = MIGRATE_SYNC_LIGHT; | |
2916 | ||
11e33f6a | 2917 | /* Try direct reclaim and then allocating */ |
a9263751 VB |
2918 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, |
2919 | &did_some_progress); | |
11e33f6a MG |
2920 | if (page) |
2921 | goto got_pg; | |
1da177e4 | 2922 | |
9083905a JW |
2923 | /* Do not loop if specifically requested */ |
2924 | if (gfp_mask & __GFP_NORETRY) | |
2925 | goto noretry; | |
2926 | ||
2927 | /* Keep reclaiming pages as long as there is reasonable progress */ | |
a41f24ea | 2928 | pages_reclaimed += did_some_progress; |
9083905a JW |
2929 | if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) || |
2930 | ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) { | |
11e33f6a | 2931 | /* Wait for some write requests to complete then retry */ |
a9263751 | 2932 | wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50); |
9879de73 | 2933 | goto retry; |
1da177e4 LT |
2934 | } |
2935 | ||
9083905a JW |
2936 | /* Reclaim has failed us, start killing things */ |
2937 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
2938 | if (page) | |
2939 | goto got_pg; | |
2940 | ||
2941 | /* Retry as long as the OOM killer is making progress */ | |
2942 | if (did_some_progress) | |
2943 | goto retry; | |
2944 | ||
2945 | noretry: | |
2946 | /* | |
2947 | * High-order allocations do not necessarily loop after | |
2948 | * direct reclaim and reclaim/compaction depends on compaction | |
2949 | * being called after reclaim so call directly if necessary | |
2950 | */ | |
2951 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, | |
2952 | ac, migration_mode, | |
2953 | &contended_compaction, | |
2954 | &deferred_compaction); | |
2955 | if (page) | |
2956 | goto got_pg; | |
1da177e4 | 2957 | nopage: |
a238ab5b | 2958 | warn_alloc_failed(gfp_mask, order, NULL); |
1da177e4 | 2959 | got_pg: |
072bb0aa | 2960 | return page; |
1da177e4 | 2961 | } |
11e33f6a MG |
2962 | |
2963 | /* | |
2964 | * This is the 'heart' of the zoned buddy allocator. | |
2965 | */ | |
2966 | struct page * | |
2967 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2968 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2969 | { | |
d8846374 | 2970 | struct zoneref *preferred_zoneref; |
cc9a6c87 | 2971 | struct page *page = NULL; |
cc9a6c87 | 2972 | unsigned int cpuset_mems_cookie; |
3a025760 | 2973 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR; |
91fbdc0f | 2974 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
a9263751 VB |
2975 | struct alloc_context ac = { |
2976 | .high_zoneidx = gfp_zone(gfp_mask), | |
2977 | .nodemask = nodemask, | |
2978 | .migratetype = gfpflags_to_migratetype(gfp_mask), | |
2979 | }; | |
11e33f6a | 2980 | |
dcce284a BH |
2981 | gfp_mask &= gfp_allowed_mask; |
2982 | ||
11e33f6a MG |
2983 | lockdep_trace_alloc(gfp_mask); |
2984 | ||
2985 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2986 | ||
2987 | if (should_fail_alloc_page(gfp_mask, order)) | |
2988 | return NULL; | |
2989 | ||
2990 | /* | |
2991 | * Check the zones suitable for the gfp_mask contain at least one | |
2992 | * valid zone. It's possible to have an empty zonelist as a result | |
4167e9b2 | 2993 | * of __GFP_THISNODE and a memoryless node |
11e33f6a MG |
2994 | */ |
2995 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2996 | return NULL; | |
2997 | ||
a9263751 | 2998 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
21bb9bd1 VB |
2999 | alloc_flags |= ALLOC_CMA; |
3000 | ||
cc9a6c87 | 3001 | retry_cpuset: |
d26914d1 | 3002 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3003 | |
a9263751 VB |
3004 | /* We set it here, as __alloc_pages_slowpath might have changed it */ |
3005 | ac.zonelist = zonelist; | |
5117f45d | 3006 | /* The preferred zone is used for statistics later */ |
a9263751 VB |
3007 | preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx, |
3008 | ac.nodemask ? : &cpuset_current_mems_allowed, | |
3009 | &ac.preferred_zone); | |
3010 | if (!ac.preferred_zone) | |
cc9a6c87 | 3011 | goto out; |
a9263751 | 3012 | ac.classzone_idx = zonelist_zone_idx(preferred_zoneref); |
5117f45d MG |
3013 | |
3014 | /* First allocation attempt */ | |
91fbdc0f | 3015 | alloc_mask = gfp_mask|__GFP_HARDWALL; |
a9263751 | 3016 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
21caf2fc ML |
3017 | if (unlikely(!page)) { |
3018 | /* | |
3019 | * Runtime PM, block IO and its error handling path | |
3020 | * can deadlock because I/O on the device might not | |
3021 | * complete. | |
3022 | */ | |
91fbdc0f AM |
3023 | alloc_mask = memalloc_noio_flags(gfp_mask); |
3024 | ||
a9263751 | 3025 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
21caf2fc | 3026 | } |
11e33f6a | 3027 | |
23f086f9 XQ |
3028 | if (kmemcheck_enabled && page) |
3029 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
3030 | ||
a9263751 | 3031 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
cc9a6c87 MG |
3032 | |
3033 | out: | |
3034 | /* | |
3035 | * When updating a task's mems_allowed, it is possible to race with | |
3036 | * parallel threads in such a way that an allocation can fail while | |
3037 | * the mask is being updated. If a page allocation is about to fail, | |
3038 | * check if the cpuset changed during allocation and if so, retry. | |
3039 | */ | |
d26914d1 | 3040 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
3041 | goto retry_cpuset; |
3042 | ||
11e33f6a | 3043 | return page; |
1da177e4 | 3044 | } |
d239171e | 3045 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
3046 | |
3047 | /* | |
3048 | * Common helper functions. | |
3049 | */ | |
920c7a5d | 3050 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 3051 | { |
945a1113 AM |
3052 | struct page *page; |
3053 | ||
3054 | /* | |
3055 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
3056 | * a highmem page | |
3057 | */ | |
3058 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
3059 | ||
1da177e4 LT |
3060 | page = alloc_pages(gfp_mask, order); |
3061 | if (!page) | |
3062 | return 0; | |
3063 | return (unsigned long) page_address(page); | |
3064 | } | |
1da177e4 LT |
3065 | EXPORT_SYMBOL(__get_free_pages); |
3066 | ||
920c7a5d | 3067 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 3068 | { |
945a1113 | 3069 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 3070 | } |
1da177e4 LT |
3071 | EXPORT_SYMBOL(get_zeroed_page); |
3072 | ||
920c7a5d | 3073 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 3074 | { |
b5810039 | 3075 | if (put_page_testzero(page)) { |
1da177e4 | 3076 | if (order == 0) |
b745bc85 | 3077 | free_hot_cold_page(page, false); |
1da177e4 LT |
3078 | else |
3079 | __free_pages_ok(page, order); | |
3080 | } | |
3081 | } | |
3082 | ||
3083 | EXPORT_SYMBOL(__free_pages); | |
3084 | ||
920c7a5d | 3085 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
3086 | { |
3087 | if (addr != 0) { | |
725d704e | 3088 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
3089 | __free_pages(virt_to_page((void *)addr), order); |
3090 | } | |
3091 | } | |
3092 | ||
3093 | EXPORT_SYMBOL(free_pages); | |
3094 | ||
b63ae8ca AD |
3095 | /* |
3096 | * Page Fragment: | |
3097 | * An arbitrary-length arbitrary-offset area of memory which resides | |
3098 | * within a 0 or higher order page. Multiple fragments within that page | |
3099 | * are individually refcounted, in the page's reference counter. | |
3100 | * | |
3101 | * The page_frag functions below provide a simple allocation framework for | |
3102 | * page fragments. This is used by the network stack and network device | |
3103 | * drivers to provide a backing region of memory for use as either an | |
3104 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
3105 | */ | |
3106 | static struct page *__page_frag_refill(struct page_frag_cache *nc, | |
3107 | gfp_t gfp_mask) | |
3108 | { | |
3109 | struct page *page = NULL; | |
3110 | gfp_t gfp = gfp_mask; | |
3111 | ||
3112 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3113 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
3114 | __GFP_NOMEMALLOC; | |
3115 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
3116 | PAGE_FRAG_CACHE_MAX_ORDER); | |
3117 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
3118 | #endif | |
3119 | if (unlikely(!page)) | |
3120 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
3121 | ||
3122 | nc->va = page ? page_address(page) : NULL; | |
3123 | ||
3124 | return page; | |
3125 | } | |
3126 | ||
3127 | void *__alloc_page_frag(struct page_frag_cache *nc, | |
3128 | unsigned int fragsz, gfp_t gfp_mask) | |
3129 | { | |
3130 | unsigned int size = PAGE_SIZE; | |
3131 | struct page *page; | |
3132 | int offset; | |
3133 | ||
3134 | if (unlikely(!nc->va)) { | |
3135 | refill: | |
3136 | page = __page_frag_refill(nc, gfp_mask); | |
3137 | if (!page) | |
3138 | return NULL; | |
3139 | ||
3140 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3141 | /* if size can vary use size else just use PAGE_SIZE */ | |
3142 | size = nc->size; | |
3143 | #endif | |
3144 | /* Even if we own the page, we do not use atomic_set(). | |
3145 | * This would break get_page_unless_zero() users. | |
3146 | */ | |
3147 | atomic_add(size - 1, &page->_count); | |
3148 | ||
3149 | /* reset page count bias and offset to start of new frag */ | |
3150 | nc->pfmemalloc = page->pfmemalloc; | |
3151 | nc->pagecnt_bias = size; | |
3152 | nc->offset = size; | |
3153 | } | |
3154 | ||
3155 | offset = nc->offset - fragsz; | |
3156 | if (unlikely(offset < 0)) { | |
3157 | page = virt_to_page(nc->va); | |
3158 | ||
3159 | if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count)) | |
3160 | goto refill; | |
3161 | ||
3162 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3163 | /* if size can vary use size else just use PAGE_SIZE */ | |
3164 | size = nc->size; | |
3165 | #endif | |
3166 | /* OK, page count is 0, we can safely set it */ | |
3167 | atomic_set(&page->_count, size); | |
3168 | ||
3169 | /* reset page count bias and offset to start of new frag */ | |
3170 | nc->pagecnt_bias = size; | |
3171 | offset = size - fragsz; | |
3172 | } | |
3173 | ||
3174 | nc->pagecnt_bias--; | |
3175 | nc->offset = offset; | |
3176 | ||
3177 | return nc->va + offset; | |
3178 | } | |
3179 | EXPORT_SYMBOL(__alloc_page_frag); | |
3180 | ||
3181 | /* | |
3182 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
3183 | */ | |
3184 | void __free_page_frag(void *addr) | |
3185 | { | |
3186 | struct page *page = virt_to_head_page(addr); | |
3187 | ||
3188 | if (unlikely(put_page_testzero(page))) | |
3189 | __free_pages_ok(page, compound_order(page)); | |
3190 | } | |
3191 | EXPORT_SYMBOL(__free_page_frag); | |
3192 | ||
6a1a0d3b | 3193 | /* |
52383431 VD |
3194 | * alloc_kmem_pages charges newly allocated pages to the kmem resource counter |
3195 | * of the current memory cgroup. | |
6a1a0d3b | 3196 | * |
52383431 VD |
3197 | * It should be used when the caller would like to use kmalloc, but since the |
3198 | * allocation is large, it has to fall back to the page allocator. | |
3199 | */ | |
3200 | struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order) | |
3201 | { | |
3202 | struct page *page; | |
3203 | struct mem_cgroup *memcg = NULL; | |
3204 | ||
3205 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
3206 | return NULL; | |
3207 | page = alloc_pages(gfp_mask, order); | |
3208 | memcg_kmem_commit_charge(page, memcg, order); | |
3209 | return page; | |
3210 | } | |
3211 | ||
3212 | struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
3213 | { | |
3214 | struct page *page; | |
3215 | struct mem_cgroup *memcg = NULL; | |
3216 | ||
3217 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
3218 | return NULL; | |
3219 | page = alloc_pages_node(nid, gfp_mask, order); | |
3220 | memcg_kmem_commit_charge(page, memcg, order); | |
3221 | return page; | |
3222 | } | |
3223 | ||
3224 | /* | |
3225 | * __free_kmem_pages and free_kmem_pages will free pages allocated with | |
3226 | * alloc_kmem_pages. | |
6a1a0d3b | 3227 | */ |
52383431 | 3228 | void __free_kmem_pages(struct page *page, unsigned int order) |
6a1a0d3b GC |
3229 | { |
3230 | memcg_kmem_uncharge_pages(page, order); | |
3231 | __free_pages(page, order); | |
3232 | } | |
3233 | ||
52383431 | 3234 | void free_kmem_pages(unsigned long addr, unsigned int order) |
6a1a0d3b GC |
3235 | { |
3236 | if (addr != 0) { | |
3237 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
52383431 | 3238 | __free_kmem_pages(virt_to_page((void *)addr), order); |
6a1a0d3b GC |
3239 | } |
3240 | } | |
3241 | ||
ee85c2e1 AK |
3242 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
3243 | { | |
3244 | if (addr) { | |
3245 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
3246 | unsigned long used = addr + PAGE_ALIGN(size); | |
3247 | ||
3248 | split_page(virt_to_page((void *)addr), order); | |
3249 | while (used < alloc_end) { | |
3250 | free_page(used); | |
3251 | used += PAGE_SIZE; | |
3252 | } | |
3253 | } | |
3254 | return (void *)addr; | |
3255 | } | |
3256 | ||
2be0ffe2 TT |
3257 | /** |
3258 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
3259 | * @size: the number of bytes to allocate | |
3260 | * @gfp_mask: GFP flags for the allocation | |
3261 | * | |
3262 | * This function is similar to alloc_pages(), except that it allocates the | |
3263 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
3264 | * allocate memory in power-of-two pages. | |
3265 | * | |
3266 | * This function is also limited by MAX_ORDER. | |
3267 | * | |
3268 | * Memory allocated by this function must be released by free_pages_exact(). | |
3269 | */ | |
3270 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
3271 | { | |
3272 | unsigned int order = get_order(size); | |
3273 | unsigned long addr; | |
3274 | ||
3275 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 3276 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
3277 | } |
3278 | EXPORT_SYMBOL(alloc_pages_exact); | |
3279 | ||
ee85c2e1 AK |
3280 | /** |
3281 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
3282 | * pages on a node. | |
b5e6ab58 | 3283 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
3284 | * @size: the number of bytes to allocate |
3285 | * @gfp_mask: GFP flags for the allocation | |
3286 | * | |
3287 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
3288 | * back. | |
3289 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
3290 | * but is not exact. | |
3291 | */ | |
e1931811 | 3292 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 AK |
3293 | { |
3294 | unsigned order = get_order(size); | |
3295 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
3296 | if (!p) | |
3297 | return NULL; | |
3298 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
3299 | } | |
ee85c2e1 | 3300 | |
2be0ffe2 TT |
3301 | /** |
3302 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
3303 | * @virt: the value returned by alloc_pages_exact. | |
3304 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
3305 | * | |
3306 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
3307 | */ | |
3308 | void free_pages_exact(void *virt, size_t size) | |
3309 | { | |
3310 | unsigned long addr = (unsigned long)virt; | |
3311 | unsigned long end = addr + PAGE_ALIGN(size); | |
3312 | ||
3313 | while (addr < end) { | |
3314 | free_page(addr); | |
3315 | addr += PAGE_SIZE; | |
3316 | } | |
3317 | } | |
3318 | EXPORT_SYMBOL(free_pages_exact); | |
3319 | ||
e0fb5815 ZY |
3320 | /** |
3321 | * nr_free_zone_pages - count number of pages beyond high watermark | |
3322 | * @offset: The zone index of the highest zone | |
3323 | * | |
3324 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
3325 | * high watermark within all zones at or below a given zone index. For each | |
3326 | * zone, the number of pages is calculated as: | |
834405c3 | 3327 | * managed_pages - high_pages |
e0fb5815 | 3328 | */ |
ebec3862 | 3329 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 3330 | { |
dd1a239f | 3331 | struct zoneref *z; |
54a6eb5c MG |
3332 | struct zone *zone; |
3333 | ||
e310fd43 | 3334 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 3335 | unsigned long sum = 0; |
1da177e4 | 3336 | |
0e88460d | 3337 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 3338 | |
54a6eb5c | 3339 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 3340 | unsigned long size = zone->managed_pages; |
41858966 | 3341 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
3342 | if (size > high) |
3343 | sum += size - high; | |
1da177e4 LT |
3344 | } |
3345 | ||
3346 | return sum; | |
3347 | } | |
3348 | ||
e0fb5815 ZY |
3349 | /** |
3350 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
3351 | * | |
3352 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
3353 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 3354 | */ |
ebec3862 | 3355 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 3356 | { |
af4ca457 | 3357 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 3358 | } |
c2f1a551 | 3359 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 3360 | |
e0fb5815 ZY |
3361 | /** |
3362 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
3363 | * | |
3364 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
3365 | * high watermark within all zones. | |
1da177e4 | 3366 | */ |
ebec3862 | 3367 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 3368 | { |
2a1e274a | 3369 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 3370 | } |
08e0f6a9 CL |
3371 | |
3372 | static inline void show_node(struct zone *zone) | |
1da177e4 | 3373 | { |
e5adfffc | 3374 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 3375 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 3376 | } |
1da177e4 | 3377 | |
1da177e4 LT |
3378 | void si_meminfo(struct sysinfo *val) |
3379 | { | |
3380 | val->totalram = totalram_pages; | |
cc7452b6 | 3381 | val->sharedram = global_page_state(NR_SHMEM); |
d23ad423 | 3382 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 3383 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
3384 | val->totalhigh = totalhigh_pages; |
3385 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
3386 | val->mem_unit = PAGE_SIZE; |
3387 | } | |
3388 | ||
3389 | EXPORT_SYMBOL(si_meminfo); | |
3390 | ||
3391 | #ifdef CONFIG_NUMA | |
3392 | void si_meminfo_node(struct sysinfo *val, int nid) | |
3393 | { | |
cdd91a77 JL |
3394 | int zone_type; /* needs to be signed */ |
3395 | unsigned long managed_pages = 0; | |
1da177e4 LT |
3396 | pg_data_t *pgdat = NODE_DATA(nid); |
3397 | ||
cdd91a77 JL |
3398 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
3399 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
3400 | val->totalram = managed_pages; | |
cc7452b6 | 3401 | val->sharedram = node_page_state(nid, NR_SHMEM); |
d23ad423 | 3402 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 3403 | #ifdef CONFIG_HIGHMEM |
b40da049 | 3404 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages; |
d23ad423 CL |
3405 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
3406 | NR_FREE_PAGES); | |
98d2b0eb CL |
3407 | #else |
3408 | val->totalhigh = 0; | |
3409 | val->freehigh = 0; | |
3410 | #endif | |
1da177e4 LT |
3411 | val->mem_unit = PAGE_SIZE; |
3412 | } | |
3413 | #endif | |
3414 | ||
ddd588b5 | 3415 | /* |
7bf02ea2 DR |
3416 | * Determine whether the node should be displayed or not, depending on whether |
3417 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 3418 | */ |
7bf02ea2 | 3419 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
3420 | { |
3421 | bool ret = false; | |
cc9a6c87 | 3422 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
3423 | |
3424 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
3425 | goto out; | |
3426 | ||
cc9a6c87 | 3427 | do { |
d26914d1 | 3428 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3429 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 3430 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
3431 | out: |
3432 | return ret; | |
3433 | } | |
3434 | ||
1da177e4 LT |
3435 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
3436 | ||
377e4f16 RV |
3437 | static void show_migration_types(unsigned char type) |
3438 | { | |
3439 | static const char types[MIGRATE_TYPES] = { | |
3440 | [MIGRATE_UNMOVABLE] = 'U', | |
3441 | [MIGRATE_RECLAIMABLE] = 'E', | |
3442 | [MIGRATE_MOVABLE] = 'M', | |
3443 | [MIGRATE_RESERVE] = 'R', | |
3444 | #ifdef CONFIG_CMA | |
3445 | [MIGRATE_CMA] = 'C', | |
3446 | #endif | |
194159fb | 3447 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 3448 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 3449 | #endif |
377e4f16 RV |
3450 | }; |
3451 | char tmp[MIGRATE_TYPES + 1]; | |
3452 | char *p = tmp; | |
3453 | int i; | |
3454 | ||
3455 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
3456 | if (type & (1 << i)) | |
3457 | *p++ = types[i]; | |
3458 | } | |
3459 | ||
3460 | *p = '\0'; | |
3461 | printk("(%s) ", tmp); | |
3462 | } | |
3463 | ||
1da177e4 LT |
3464 | /* |
3465 | * Show free area list (used inside shift_scroll-lock stuff) | |
3466 | * We also calculate the percentage fragmentation. We do this by counting the | |
3467 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
3468 | * |
3469 | * Bits in @filter: | |
3470 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
3471 | * cpuset. | |
1da177e4 | 3472 | */ |
7bf02ea2 | 3473 | void show_free_areas(unsigned int filter) |
1da177e4 | 3474 | { |
d1bfcdb8 | 3475 | unsigned long free_pcp = 0; |
c7241913 | 3476 | int cpu; |
1da177e4 LT |
3477 | struct zone *zone; |
3478 | ||
ee99c71c | 3479 | for_each_populated_zone(zone) { |
7bf02ea2 | 3480 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3481 | continue; |
d1bfcdb8 | 3482 | |
761b0677 KK |
3483 | for_each_online_cpu(cpu) |
3484 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
3485 | } |
3486 | ||
a731286d KM |
3487 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
3488 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
3489 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
3490 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 3491 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 3492 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
4f98a2fe | 3493 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 3494 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
3495 | global_page_state(NR_ISOLATED_ANON), |
3496 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 3497 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 3498 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 3499 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 3500 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 3501 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 3502 | global_page_state(NR_UNSTABLE_NFS), |
3701b033 KM |
3503 | global_page_state(NR_SLAB_RECLAIMABLE), |
3504 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 3505 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 3506 | global_page_state(NR_SHMEM), |
a25700a5 | 3507 | global_page_state(NR_PAGETABLE), |
d1ce749a | 3508 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
3509 | global_page_state(NR_FREE_PAGES), |
3510 | free_pcp, | |
d1ce749a | 3511 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 3512 | |
ee99c71c | 3513 | for_each_populated_zone(zone) { |
1da177e4 LT |
3514 | int i; |
3515 | ||
7bf02ea2 | 3516 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3517 | continue; |
d1bfcdb8 KK |
3518 | |
3519 | free_pcp = 0; | |
3520 | for_each_online_cpu(cpu) | |
3521 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
3522 | ||
1da177e4 LT |
3523 | show_node(zone); |
3524 | printk("%s" | |
3525 | " free:%lukB" | |
3526 | " min:%lukB" | |
3527 | " low:%lukB" | |
3528 | " high:%lukB" | |
4f98a2fe RR |
3529 | " active_anon:%lukB" |
3530 | " inactive_anon:%lukB" | |
3531 | " active_file:%lukB" | |
3532 | " inactive_file:%lukB" | |
7b854121 | 3533 | " unevictable:%lukB" |
a731286d KM |
3534 | " isolated(anon):%lukB" |
3535 | " isolated(file):%lukB" | |
1da177e4 | 3536 | " present:%lukB" |
9feedc9d | 3537 | " managed:%lukB" |
4a0aa73f KM |
3538 | " mlocked:%lukB" |
3539 | " dirty:%lukB" | |
3540 | " writeback:%lukB" | |
3541 | " mapped:%lukB" | |
4b02108a | 3542 | " shmem:%lukB" |
4a0aa73f KM |
3543 | " slab_reclaimable:%lukB" |
3544 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3545 | " kernel_stack:%lukB" |
4a0aa73f KM |
3546 | " pagetables:%lukB" |
3547 | " unstable:%lukB" | |
3548 | " bounce:%lukB" | |
d1bfcdb8 KK |
3549 | " free_pcp:%lukB" |
3550 | " local_pcp:%ukB" | |
d1ce749a | 3551 | " free_cma:%lukB" |
4a0aa73f | 3552 | " writeback_tmp:%lukB" |
1da177e4 LT |
3553 | " pages_scanned:%lu" |
3554 | " all_unreclaimable? %s" | |
3555 | "\n", | |
3556 | zone->name, | |
88f5acf8 | 3557 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3558 | K(min_wmark_pages(zone)), |
3559 | K(low_wmark_pages(zone)), | |
3560 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3561 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3562 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3563 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3564 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3565 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3566 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3567 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3568 | K(zone->present_pages), |
9feedc9d | 3569 | K(zone->managed_pages), |
4a0aa73f KM |
3570 | K(zone_page_state(zone, NR_MLOCK)), |
3571 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3572 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3573 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3574 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3575 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3576 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3577 | zone_page_state(zone, NR_KERNEL_STACK) * |
3578 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3579 | K(zone_page_state(zone, NR_PAGETABLE)), |
3580 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3581 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1bfcdb8 KK |
3582 | K(free_pcp), |
3583 | K(this_cpu_read(zone->pageset->pcp.count)), | |
d1ce749a | 3584 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3585 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
0d5d823a | 3586 | K(zone_page_state(zone, NR_PAGES_SCANNED)), |
6e543d57 | 3587 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
3588 | ); |
3589 | printk("lowmem_reserve[]:"); | |
3590 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3484b2de | 3591 | printk(" %ld", zone->lowmem_reserve[i]); |
1da177e4 LT |
3592 | printk("\n"); |
3593 | } | |
3594 | ||
ee99c71c | 3595 | for_each_populated_zone(zone) { |
b8af2941 | 3596 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
377e4f16 | 3597 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3598 | |
7bf02ea2 | 3599 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3600 | continue; |
1da177e4 LT |
3601 | show_node(zone); |
3602 | printk("%s: ", zone->name); | |
1da177e4 LT |
3603 | |
3604 | spin_lock_irqsave(&zone->lock, flags); | |
3605 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3606 | struct free_area *area = &zone->free_area[order]; |
3607 | int type; | |
3608 | ||
3609 | nr[order] = area->nr_free; | |
8f9de51a | 3610 | total += nr[order] << order; |
377e4f16 RV |
3611 | |
3612 | types[order] = 0; | |
3613 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3614 | if (!list_empty(&area->free_list[type])) | |
3615 | types[order] |= 1 << type; | |
3616 | } | |
1da177e4 LT |
3617 | } |
3618 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3619 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3620 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3621 | if (nr[order]) |
3622 | show_migration_types(types[order]); | |
3623 | } | |
1da177e4 LT |
3624 | printk("= %lukB\n", K(total)); |
3625 | } | |
3626 | ||
949f7ec5 DR |
3627 | hugetlb_show_meminfo(); |
3628 | ||
e6f3602d LW |
3629 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
3630 | ||
1da177e4 LT |
3631 | show_swap_cache_info(); |
3632 | } | |
3633 | ||
19770b32 MG |
3634 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
3635 | { | |
3636 | zoneref->zone = zone; | |
3637 | zoneref->zone_idx = zone_idx(zone); | |
3638 | } | |
3639 | ||
1da177e4 LT |
3640 | /* |
3641 | * Builds allocation fallback zone lists. | |
1a93205b CL |
3642 | * |
3643 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 3644 | */ |
f0c0b2b8 | 3645 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 3646 | int nr_zones) |
1da177e4 | 3647 | { |
1a93205b | 3648 | struct zone *zone; |
bc732f1d | 3649 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
3650 | |
3651 | do { | |
2f6726e5 | 3652 | zone_type--; |
070f8032 | 3653 | zone = pgdat->node_zones + zone_type; |
1a93205b | 3654 | if (populated_zone(zone)) { |
dd1a239f MG |
3655 | zoneref_set_zone(zone, |
3656 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 3657 | check_highest_zone(zone_type); |
1da177e4 | 3658 | } |
2f6726e5 | 3659 | } while (zone_type); |
bc732f1d | 3660 | |
070f8032 | 3661 | return nr_zones; |
1da177e4 LT |
3662 | } |
3663 | ||
f0c0b2b8 KH |
3664 | |
3665 | /* | |
3666 | * zonelist_order: | |
3667 | * 0 = automatic detection of better ordering. | |
3668 | * 1 = order by ([node] distance, -zonetype) | |
3669 | * 2 = order by (-zonetype, [node] distance) | |
3670 | * | |
3671 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
3672 | * the same zonelist. So only NUMA can configure this param. | |
3673 | */ | |
3674 | #define ZONELIST_ORDER_DEFAULT 0 | |
3675 | #define ZONELIST_ORDER_NODE 1 | |
3676 | #define ZONELIST_ORDER_ZONE 2 | |
3677 | ||
3678 | /* zonelist order in the kernel. | |
3679 | * set_zonelist_order() will set this to NODE or ZONE. | |
3680 | */ | |
3681 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3682 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
3683 | ||
3684 | ||
1da177e4 | 3685 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
3686 | /* The value user specified ....changed by config */ |
3687 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3688 | /* string for sysctl */ | |
3689 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
3690 | char numa_zonelist_order[16] = "default"; | |
3691 | ||
3692 | /* | |
3693 | * interface for configure zonelist ordering. | |
3694 | * command line option "numa_zonelist_order" | |
3695 | * = "[dD]efault - default, automatic configuration. | |
3696 | * = "[nN]ode - order by node locality, then by zone within node | |
3697 | * = "[zZ]one - order by zone, then by locality within zone | |
3698 | */ | |
3699 | ||
3700 | static int __parse_numa_zonelist_order(char *s) | |
3701 | { | |
3702 | if (*s == 'd' || *s == 'D') { | |
3703 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3704 | } else if (*s == 'n' || *s == 'N') { | |
3705 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
3706 | } else if (*s == 'z' || *s == 'Z') { | |
3707 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
3708 | } else { | |
3709 | printk(KERN_WARNING | |
3710 | "Ignoring invalid numa_zonelist_order value: " | |
3711 | "%s\n", s); | |
3712 | return -EINVAL; | |
3713 | } | |
3714 | return 0; | |
3715 | } | |
3716 | ||
3717 | static __init int setup_numa_zonelist_order(char *s) | |
3718 | { | |
ecb256f8 VL |
3719 | int ret; |
3720 | ||
3721 | if (!s) | |
3722 | return 0; | |
3723 | ||
3724 | ret = __parse_numa_zonelist_order(s); | |
3725 | if (ret == 0) | |
3726 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
3727 | ||
3728 | return ret; | |
f0c0b2b8 KH |
3729 | } |
3730 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
3731 | ||
3732 | /* | |
3733 | * sysctl handler for numa_zonelist_order | |
3734 | */ | |
cccad5b9 | 3735 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 3736 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
3737 | loff_t *ppos) |
3738 | { | |
3739 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
3740 | int ret; | |
443c6f14 | 3741 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 3742 | |
443c6f14 | 3743 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
3744 | if (write) { |
3745 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
3746 | ret = -EINVAL; | |
3747 | goto out; | |
3748 | } | |
3749 | strcpy(saved_string, (char *)table->data); | |
3750 | } | |
8d65af78 | 3751 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 3752 | if (ret) |
443c6f14 | 3753 | goto out; |
f0c0b2b8 KH |
3754 | if (write) { |
3755 | int oldval = user_zonelist_order; | |
dacbde09 CG |
3756 | |
3757 | ret = __parse_numa_zonelist_order((char *)table->data); | |
3758 | if (ret) { | |
f0c0b2b8 KH |
3759 | /* |
3760 | * bogus value. restore saved string | |
3761 | */ | |
dacbde09 | 3762 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
3763 | NUMA_ZONELIST_ORDER_LEN); |
3764 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
3765 | } else if (oldval != user_zonelist_order) { |
3766 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 3767 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
3768 | mutex_unlock(&zonelists_mutex); |
3769 | } | |
f0c0b2b8 | 3770 | } |
443c6f14 AK |
3771 | out: |
3772 | mutex_unlock(&zl_order_mutex); | |
3773 | return ret; | |
f0c0b2b8 KH |
3774 | } |
3775 | ||
3776 | ||
62bc62a8 | 3777 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
3778 | static int node_load[MAX_NUMNODES]; |
3779 | ||
1da177e4 | 3780 | /** |
4dc3b16b | 3781 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
3782 | * @node: node whose fallback list we're appending |
3783 | * @used_node_mask: nodemask_t of already used nodes | |
3784 | * | |
3785 | * We use a number of factors to determine which is the next node that should | |
3786 | * appear on a given node's fallback list. The node should not have appeared | |
3787 | * already in @node's fallback list, and it should be the next closest node | |
3788 | * according to the distance array (which contains arbitrary distance values | |
3789 | * from each node to each node in the system), and should also prefer nodes | |
3790 | * with no CPUs, since presumably they'll have very little allocation pressure | |
3791 | * on them otherwise. | |
3792 | * It returns -1 if no node is found. | |
3793 | */ | |
f0c0b2b8 | 3794 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 3795 | { |
4cf808eb | 3796 | int n, val; |
1da177e4 | 3797 | int min_val = INT_MAX; |
00ef2d2f | 3798 | int best_node = NUMA_NO_NODE; |
a70f7302 | 3799 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 3800 | |
4cf808eb LT |
3801 | /* Use the local node if we haven't already */ |
3802 | if (!node_isset(node, *used_node_mask)) { | |
3803 | node_set(node, *used_node_mask); | |
3804 | return node; | |
3805 | } | |
1da177e4 | 3806 | |
4b0ef1fe | 3807 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
3808 | |
3809 | /* Don't want a node to appear more than once */ | |
3810 | if (node_isset(n, *used_node_mask)) | |
3811 | continue; | |
3812 | ||
1da177e4 LT |
3813 | /* Use the distance array to find the distance */ |
3814 | val = node_distance(node, n); | |
3815 | ||
4cf808eb LT |
3816 | /* Penalize nodes under us ("prefer the next node") */ |
3817 | val += (n < node); | |
3818 | ||
1da177e4 | 3819 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
3820 | tmp = cpumask_of_node(n); |
3821 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
3822 | val += PENALTY_FOR_NODE_WITH_CPUS; |
3823 | ||
3824 | /* Slight preference for less loaded node */ | |
3825 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
3826 | val += node_load[n]; | |
3827 | ||
3828 | if (val < min_val) { | |
3829 | min_val = val; | |
3830 | best_node = n; | |
3831 | } | |
3832 | } | |
3833 | ||
3834 | if (best_node >= 0) | |
3835 | node_set(best_node, *used_node_mask); | |
3836 | ||
3837 | return best_node; | |
3838 | } | |
3839 | ||
f0c0b2b8 KH |
3840 | |
3841 | /* | |
3842 | * Build zonelists ordered by node and zones within node. | |
3843 | * This results in maximum locality--normal zone overflows into local | |
3844 | * DMA zone, if any--but risks exhausting DMA zone. | |
3845 | */ | |
3846 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 3847 | { |
f0c0b2b8 | 3848 | int j; |
1da177e4 | 3849 | struct zonelist *zonelist; |
f0c0b2b8 | 3850 | |
54a6eb5c | 3851 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 3852 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 3853 | ; |
bc732f1d | 3854 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
3855 | zonelist->_zonerefs[j].zone = NULL; |
3856 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
3857 | } |
3858 | ||
523b9458 CL |
3859 | /* |
3860 | * Build gfp_thisnode zonelists | |
3861 | */ | |
3862 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3863 | { | |
523b9458 CL |
3864 | int j; |
3865 | struct zonelist *zonelist; | |
3866 | ||
54a6eb5c | 3867 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 3868 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
3869 | zonelist->_zonerefs[j].zone = NULL; |
3870 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3871 | } |
3872 | ||
f0c0b2b8 KH |
3873 | /* |
3874 | * Build zonelists ordered by zone and nodes within zones. | |
3875 | * This results in conserving DMA zone[s] until all Normal memory is | |
3876 | * exhausted, but results in overflowing to remote node while memory | |
3877 | * may still exist in local DMA zone. | |
3878 | */ | |
3879 | static int node_order[MAX_NUMNODES]; | |
3880 | ||
3881 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3882 | { | |
f0c0b2b8 KH |
3883 | int pos, j, node; |
3884 | int zone_type; /* needs to be signed */ | |
3885 | struct zone *z; | |
3886 | struct zonelist *zonelist; | |
3887 | ||
54a6eb5c MG |
3888 | zonelist = &pgdat->node_zonelists[0]; |
3889 | pos = 0; | |
3890 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3891 | for (j = 0; j < nr_nodes; j++) { | |
3892 | node = node_order[j]; | |
3893 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3894 | if (populated_zone(z)) { | |
dd1a239f MG |
3895 | zoneref_set_zone(z, |
3896 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3897 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3898 | } |
3899 | } | |
f0c0b2b8 | 3900 | } |
dd1a239f MG |
3901 | zonelist->_zonerefs[pos].zone = NULL; |
3902 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3903 | } |
3904 | ||
3193913c MG |
3905 | #if defined(CONFIG_64BIT) |
3906 | /* | |
3907 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
3908 | * penalty to every machine in case the specialised case applies. Default | |
3909 | * to Node-ordering on 64-bit NUMA machines | |
3910 | */ | |
3911 | static int default_zonelist_order(void) | |
3912 | { | |
3913 | return ZONELIST_ORDER_NODE; | |
3914 | } | |
3915 | #else | |
3916 | /* | |
3917 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
3918 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
3919 | * then reclaim will occur more frequency increasing stalls and potentially | |
3920 | * be easier to OOM if a large percentage of the zone is under writeback or | |
3921 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
3922 | * Hence, default to zone ordering on 32-bit. | |
3923 | */ | |
f0c0b2b8 KH |
3924 | static int default_zonelist_order(void) |
3925 | { | |
f0c0b2b8 KH |
3926 | return ZONELIST_ORDER_ZONE; |
3927 | } | |
3193913c | 3928 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
3929 | |
3930 | static void set_zonelist_order(void) | |
3931 | { | |
3932 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3933 | current_zonelist_order = default_zonelist_order(); | |
3934 | else | |
3935 | current_zonelist_order = user_zonelist_order; | |
3936 | } | |
3937 | ||
3938 | static void build_zonelists(pg_data_t *pgdat) | |
3939 | { | |
3940 | int j, node, load; | |
3941 | enum zone_type i; | |
1da177e4 | 3942 | nodemask_t used_mask; |
f0c0b2b8 KH |
3943 | int local_node, prev_node; |
3944 | struct zonelist *zonelist; | |
3945 | int order = current_zonelist_order; | |
1da177e4 LT |
3946 | |
3947 | /* initialize zonelists */ | |
523b9458 | 3948 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3949 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3950 | zonelist->_zonerefs[0].zone = NULL; |
3951 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3952 | } |
3953 | ||
3954 | /* NUMA-aware ordering of nodes */ | |
3955 | local_node = pgdat->node_id; | |
62bc62a8 | 3956 | load = nr_online_nodes; |
1da177e4 LT |
3957 | prev_node = local_node; |
3958 | nodes_clear(used_mask); | |
f0c0b2b8 | 3959 | |
f0c0b2b8 KH |
3960 | memset(node_order, 0, sizeof(node_order)); |
3961 | j = 0; | |
3962 | ||
1da177e4 LT |
3963 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
3964 | /* | |
3965 | * We don't want to pressure a particular node. | |
3966 | * So adding penalty to the first node in same | |
3967 | * distance group to make it round-robin. | |
3968 | */ | |
957f822a DR |
3969 | if (node_distance(local_node, node) != |
3970 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
3971 | node_load[node] = load; |
3972 | ||
1da177e4 LT |
3973 | prev_node = node; |
3974 | load--; | |
f0c0b2b8 KH |
3975 | if (order == ZONELIST_ORDER_NODE) |
3976 | build_zonelists_in_node_order(pgdat, node); | |
3977 | else | |
3978 | node_order[j++] = node; /* remember order */ | |
3979 | } | |
1da177e4 | 3980 | |
f0c0b2b8 KH |
3981 | if (order == ZONELIST_ORDER_ZONE) { |
3982 | /* calculate node order -- i.e., DMA last! */ | |
3983 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3984 | } |
523b9458 CL |
3985 | |
3986 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3987 | } |
3988 | ||
9276b1bc | 3989 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3990 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3991 | { |
54a6eb5c MG |
3992 | struct zonelist *zonelist; |
3993 | struct zonelist_cache *zlc; | |
dd1a239f | 3994 | struct zoneref *z; |
9276b1bc | 3995 | |
54a6eb5c MG |
3996 | zonelist = &pgdat->node_zonelists[0]; |
3997 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3998 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3999 | for (z = zonelist->_zonerefs; z->zone; z++) |
4000 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
4001 | } |
4002 | ||
7aac7898 LS |
4003 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4004 | /* | |
4005 | * Return node id of node used for "local" allocations. | |
4006 | * I.e., first node id of first zone in arg node's generic zonelist. | |
4007 | * Used for initializing percpu 'numa_mem', which is used primarily | |
4008 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
4009 | */ | |
4010 | int local_memory_node(int node) | |
4011 | { | |
4012 | struct zone *zone; | |
4013 | ||
4014 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
4015 | gfp_zone(GFP_KERNEL), | |
4016 | NULL, | |
4017 | &zone); | |
4018 | return zone->node; | |
4019 | } | |
4020 | #endif | |
f0c0b2b8 | 4021 | |
1da177e4 LT |
4022 | #else /* CONFIG_NUMA */ |
4023 | ||
f0c0b2b8 KH |
4024 | static void set_zonelist_order(void) |
4025 | { | |
4026 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
4027 | } | |
4028 | ||
4029 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 4030 | { |
19655d34 | 4031 | int node, local_node; |
54a6eb5c MG |
4032 | enum zone_type j; |
4033 | struct zonelist *zonelist; | |
1da177e4 LT |
4034 | |
4035 | local_node = pgdat->node_id; | |
1da177e4 | 4036 | |
54a6eb5c | 4037 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 4038 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 4039 | |
54a6eb5c MG |
4040 | /* |
4041 | * Now we build the zonelist so that it contains the zones | |
4042 | * of all the other nodes. | |
4043 | * We don't want to pressure a particular node, so when | |
4044 | * building the zones for node N, we make sure that the | |
4045 | * zones coming right after the local ones are those from | |
4046 | * node N+1 (modulo N) | |
4047 | */ | |
4048 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
4049 | if (!node_online(node)) | |
4050 | continue; | |
bc732f1d | 4051 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 4052 | } |
54a6eb5c MG |
4053 | for (node = 0; node < local_node; node++) { |
4054 | if (!node_online(node)) | |
4055 | continue; | |
bc732f1d | 4056 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
4057 | } |
4058 | ||
dd1a239f MG |
4059 | zonelist->_zonerefs[j].zone = NULL; |
4060 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
4061 | } |
4062 | ||
9276b1bc | 4063 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 4064 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 4065 | { |
54a6eb5c | 4066 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
4067 | } |
4068 | ||
1da177e4 LT |
4069 | #endif /* CONFIG_NUMA */ |
4070 | ||
99dcc3e5 CL |
4071 | /* |
4072 | * Boot pageset table. One per cpu which is going to be used for all | |
4073 | * zones and all nodes. The parameters will be set in such a way | |
4074 | * that an item put on a list will immediately be handed over to | |
4075 | * the buddy list. This is safe since pageset manipulation is done | |
4076 | * with interrupts disabled. | |
4077 | * | |
4078 | * The boot_pagesets must be kept even after bootup is complete for | |
4079 | * unused processors and/or zones. They do play a role for bootstrapping | |
4080 | * hotplugged processors. | |
4081 | * | |
4082 | * zoneinfo_show() and maybe other functions do | |
4083 | * not check if the processor is online before following the pageset pointer. | |
4084 | * Other parts of the kernel may not check if the zone is available. | |
4085 | */ | |
4086 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
4087 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 4088 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 4089 | |
4eaf3f64 HL |
4090 | /* |
4091 | * Global mutex to protect against size modification of zonelists | |
4092 | * as well as to serialize pageset setup for the new populated zone. | |
4093 | */ | |
4094 | DEFINE_MUTEX(zonelists_mutex); | |
4095 | ||
9b1a4d38 | 4096 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 4097 | static int __build_all_zonelists(void *data) |
1da177e4 | 4098 | { |
6811378e | 4099 | int nid; |
99dcc3e5 | 4100 | int cpu; |
9adb62a5 | 4101 | pg_data_t *self = data; |
9276b1bc | 4102 | |
7f9cfb31 BL |
4103 | #ifdef CONFIG_NUMA |
4104 | memset(node_load, 0, sizeof(node_load)); | |
4105 | #endif | |
9adb62a5 JL |
4106 | |
4107 | if (self && !node_online(self->node_id)) { | |
4108 | build_zonelists(self); | |
4109 | build_zonelist_cache(self); | |
4110 | } | |
4111 | ||
9276b1bc | 4112 | for_each_online_node(nid) { |
7ea1530a CL |
4113 | pg_data_t *pgdat = NODE_DATA(nid); |
4114 | ||
4115 | build_zonelists(pgdat); | |
4116 | build_zonelist_cache(pgdat); | |
9276b1bc | 4117 | } |
99dcc3e5 CL |
4118 | |
4119 | /* | |
4120 | * Initialize the boot_pagesets that are going to be used | |
4121 | * for bootstrapping processors. The real pagesets for | |
4122 | * each zone will be allocated later when the per cpu | |
4123 | * allocator is available. | |
4124 | * | |
4125 | * boot_pagesets are used also for bootstrapping offline | |
4126 | * cpus if the system is already booted because the pagesets | |
4127 | * are needed to initialize allocators on a specific cpu too. | |
4128 | * F.e. the percpu allocator needs the page allocator which | |
4129 | * needs the percpu allocator in order to allocate its pagesets | |
4130 | * (a chicken-egg dilemma). | |
4131 | */ | |
7aac7898 | 4132 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
4133 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
4134 | ||
7aac7898 LS |
4135 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4136 | /* | |
4137 | * We now know the "local memory node" for each node-- | |
4138 | * i.e., the node of the first zone in the generic zonelist. | |
4139 | * Set up numa_mem percpu variable for on-line cpus. During | |
4140 | * boot, only the boot cpu should be on-line; we'll init the | |
4141 | * secondary cpus' numa_mem as they come on-line. During | |
4142 | * node/memory hotplug, we'll fixup all on-line cpus. | |
4143 | */ | |
4144 | if (cpu_online(cpu)) | |
4145 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
4146 | #endif | |
4147 | } | |
4148 | ||
6811378e YG |
4149 | return 0; |
4150 | } | |
4151 | ||
061f67bc RV |
4152 | static noinline void __init |
4153 | build_all_zonelists_init(void) | |
4154 | { | |
4155 | __build_all_zonelists(NULL); | |
4156 | mminit_verify_zonelist(); | |
4157 | cpuset_init_current_mems_allowed(); | |
4158 | } | |
4159 | ||
4eaf3f64 HL |
4160 | /* |
4161 | * Called with zonelists_mutex held always | |
4162 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
4163 | * |
4164 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
4165 | * [we're only called with non-NULL zone through __meminit paths] and | |
4166 | * (2) call of __init annotated helper build_all_zonelists_init | |
4167 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 4168 | */ |
9adb62a5 | 4169 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 4170 | { |
f0c0b2b8 KH |
4171 | set_zonelist_order(); |
4172 | ||
6811378e | 4173 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 4174 | build_all_zonelists_init(); |
6811378e | 4175 | } else { |
e9959f0f | 4176 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
4177 | if (zone) |
4178 | setup_zone_pageset(zone); | |
e9959f0f | 4179 | #endif |
dd1895e2 CS |
4180 | /* we have to stop all cpus to guarantee there is no user |
4181 | of zonelist */ | |
9adb62a5 | 4182 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
4183 | /* cpuset refresh routine should be here */ |
4184 | } | |
bd1e22b8 | 4185 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
4186 | /* |
4187 | * Disable grouping by mobility if the number of pages in the | |
4188 | * system is too low to allow the mechanism to work. It would be | |
4189 | * more accurate, but expensive to check per-zone. This check is | |
4190 | * made on memory-hotadd so a system can start with mobility | |
4191 | * disabled and enable it later | |
4192 | */ | |
d9c23400 | 4193 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
4194 | page_group_by_mobility_disabled = 1; |
4195 | else | |
4196 | page_group_by_mobility_disabled = 0; | |
4197 | ||
f88dfff5 | 4198 | pr_info("Built %i zonelists in %s order, mobility grouping %s. " |
9ef9acb0 | 4199 | "Total pages: %ld\n", |
62bc62a8 | 4200 | nr_online_nodes, |
f0c0b2b8 | 4201 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 4202 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
4203 | vm_total_pages); |
4204 | #ifdef CONFIG_NUMA | |
f88dfff5 | 4205 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 4206 | #endif |
1da177e4 LT |
4207 | } |
4208 | ||
4209 | /* | |
4210 | * Helper functions to size the waitqueue hash table. | |
4211 | * Essentially these want to choose hash table sizes sufficiently | |
4212 | * large so that collisions trying to wait on pages are rare. | |
4213 | * But in fact, the number of active page waitqueues on typical | |
4214 | * systems is ridiculously low, less than 200. So this is even | |
4215 | * conservative, even though it seems large. | |
4216 | * | |
4217 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
4218 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
4219 | */ | |
4220 | #define PAGES_PER_WAITQUEUE 256 | |
4221 | ||
cca448fe | 4222 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 4223 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
4224 | { |
4225 | unsigned long size = 1; | |
4226 | ||
4227 | pages /= PAGES_PER_WAITQUEUE; | |
4228 | ||
4229 | while (size < pages) | |
4230 | size <<= 1; | |
4231 | ||
4232 | /* | |
4233 | * Once we have dozens or even hundreds of threads sleeping | |
4234 | * on IO we've got bigger problems than wait queue collision. | |
4235 | * Limit the size of the wait table to a reasonable size. | |
4236 | */ | |
4237 | size = min(size, 4096UL); | |
4238 | ||
4239 | return max(size, 4UL); | |
4240 | } | |
cca448fe YG |
4241 | #else |
4242 | /* | |
4243 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
4244 | * a suitable size for its wait_table. So we use the maximum size now. | |
4245 | * | |
4246 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
4247 | * | |
4248 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
4249 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
4250 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
4251 | * | |
4252 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
4253 | * or more by the traditional way. (See above). It equals: | |
4254 | * | |
4255 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
4256 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
4257 | * powerpc (64K page size) : = (32G +16M)byte. | |
4258 | */ | |
4259 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
4260 | { | |
4261 | return 4096UL; | |
4262 | } | |
4263 | #endif | |
1da177e4 LT |
4264 | |
4265 | /* | |
4266 | * This is an integer logarithm so that shifts can be used later | |
4267 | * to extract the more random high bits from the multiplicative | |
4268 | * hash function before the remainder is taken. | |
4269 | */ | |
4270 | static inline unsigned long wait_table_bits(unsigned long size) | |
4271 | { | |
4272 | return ffz(~size); | |
4273 | } | |
4274 | ||
6d3163ce AH |
4275 | /* |
4276 | * Check if a pageblock contains reserved pages | |
4277 | */ | |
4278 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
4279 | { | |
4280 | unsigned long pfn; | |
4281 | ||
4282 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
4283 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
4284 | return 1; | |
4285 | } | |
4286 | return 0; | |
4287 | } | |
4288 | ||
56fd56b8 | 4289 | /* |
d9c23400 | 4290 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
4291 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
4292 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
4293 | * higher will lead to a bigger reserve which will get freed as contiguous |
4294 | * blocks as reclaim kicks in | |
4295 | */ | |
4296 | static void setup_zone_migrate_reserve(struct zone *zone) | |
4297 | { | |
6d3163ce | 4298 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 4299 | struct page *page; |
78986a67 MG |
4300 | unsigned long block_migratetype; |
4301 | int reserve; | |
943dca1a | 4302 | int old_reserve; |
56fd56b8 | 4303 | |
d0215638 MH |
4304 | /* |
4305 | * Get the start pfn, end pfn and the number of blocks to reserve | |
4306 | * We have to be careful to be aligned to pageblock_nr_pages to | |
4307 | * make sure that we always check pfn_valid for the first page in | |
4308 | * the block. | |
4309 | */ | |
56fd56b8 | 4310 | start_pfn = zone->zone_start_pfn; |
108bcc96 | 4311 | end_pfn = zone_end_pfn(zone); |
d0215638 | 4312 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 4313 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 4314 | pageblock_order; |
56fd56b8 | 4315 | |
78986a67 MG |
4316 | /* |
4317 | * Reserve blocks are generally in place to help high-order atomic | |
4318 | * allocations that are short-lived. A min_free_kbytes value that | |
4319 | * would result in more than 2 reserve blocks for atomic allocations | |
4320 | * is assumed to be in place to help anti-fragmentation for the | |
4321 | * future allocation of hugepages at runtime. | |
4322 | */ | |
4323 | reserve = min(2, reserve); | |
943dca1a YI |
4324 | old_reserve = zone->nr_migrate_reserve_block; |
4325 | ||
4326 | /* When memory hot-add, we almost always need to do nothing */ | |
4327 | if (reserve == old_reserve) | |
4328 | return; | |
4329 | zone->nr_migrate_reserve_block = reserve; | |
78986a67 | 4330 | |
d9c23400 | 4331 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
4332 | if (!pfn_valid(pfn)) |
4333 | continue; | |
4334 | page = pfn_to_page(pfn); | |
4335 | ||
344c790e AL |
4336 | /* Watch out for overlapping nodes */ |
4337 | if (page_to_nid(page) != zone_to_nid(zone)) | |
4338 | continue; | |
4339 | ||
56fd56b8 MG |
4340 | block_migratetype = get_pageblock_migratetype(page); |
4341 | ||
938929f1 MG |
4342 | /* Only test what is necessary when the reserves are not met */ |
4343 | if (reserve > 0) { | |
4344 | /* | |
4345 | * Blocks with reserved pages will never free, skip | |
4346 | * them. | |
4347 | */ | |
4348 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
4349 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
4350 | continue; | |
56fd56b8 | 4351 | |
938929f1 MG |
4352 | /* If this block is reserved, account for it */ |
4353 | if (block_migratetype == MIGRATE_RESERVE) { | |
4354 | reserve--; | |
4355 | continue; | |
4356 | } | |
4357 | ||
4358 | /* Suitable for reserving if this block is movable */ | |
4359 | if (block_migratetype == MIGRATE_MOVABLE) { | |
4360 | set_pageblock_migratetype(page, | |
4361 | MIGRATE_RESERVE); | |
4362 | move_freepages_block(zone, page, | |
4363 | MIGRATE_RESERVE); | |
4364 | reserve--; | |
4365 | continue; | |
4366 | } | |
943dca1a YI |
4367 | } else if (!old_reserve) { |
4368 | /* | |
4369 | * At boot time we don't need to scan the whole zone | |
4370 | * for turning off MIGRATE_RESERVE. | |
4371 | */ | |
4372 | break; | |
56fd56b8 MG |
4373 | } |
4374 | ||
4375 | /* | |
4376 | * If the reserve is met and this is a previous reserved block, | |
4377 | * take it back | |
4378 | */ | |
4379 | if (block_migratetype == MIGRATE_RESERVE) { | |
4380 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4381 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
4382 | } | |
4383 | } | |
4384 | } | |
ac0e5b7a | 4385 | |
1da177e4 LT |
4386 | /* |
4387 | * Initially all pages are reserved - free ones are freed | |
4388 | * up by free_all_bootmem() once the early boot process is | |
4389 | * done. Non-atomic initialization, single-pass. | |
4390 | */ | |
c09b4240 | 4391 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 4392 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 4393 | { |
3a80a7fa | 4394 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 AW |
4395 | unsigned long end_pfn = start_pfn + size; |
4396 | unsigned long pfn; | |
86051ca5 | 4397 | struct zone *z; |
3a80a7fa | 4398 | unsigned long nr_initialised = 0; |
1da177e4 | 4399 | |
22b31eec HD |
4400 | if (highest_memmap_pfn < end_pfn - 1) |
4401 | highest_memmap_pfn = end_pfn - 1; | |
4402 | ||
3a80a7fa | 4403 | z = &pgdat->node_zones[zone]; |
cbe8dd4a | 4404 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
4405 | /* |
4406 | * There can be holes in boot-time mem_map[]s | |
4407 | * handed to this function. They do not | |
4408 | * exist on hotplugged memory. | |
4409 | */ | |
4410 | if (context == MEMMAP_EARLY) { | |
4411 | if (!early_pfn_valid(pfn)) | |
4412 | continue; | |
4413 | if (!early_pfn_in_nid(pfn, nid)) | |
4414 | continue; | |
3a80a7fa MG |
4415 | if (!update_defer_init(pgdat, pfn, end_pfn, |
4416 | &nr_initialised)) | |
4417 | break; | |
a2f3aa02 | 4418 | } |
1e8ce83c | 4419 | __init_single_pfn(pfn, zone, nid); |
1da177e4 LT |
4420 | } |
4421 | } | |
4422 | ||
1e548deb | 4423 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 4424 | { |
7aeb09f9 | 4425 | unsigned int order, t; |
b2a0ac88 MG |
4426 | for_each_migratetype_order(order, t) { |
4427 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
4428 | zone->free_area[order].nr_free = 0; |
4429 | } | |
4430 | } | |
4431 | ||
4432 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
4433 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 4434 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
4435 | #endif |
4436 | ||
7cd2b0a3 | 4437 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 4438 | { |
3a6be87f | 4439 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
4440 | int batch; |
4441 | ||
4442 | /* | |
4443 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 4444 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
4445 | * |
4446 | * OK, so we don't know how big the cache is. So guess. | |
4447 | */ | |
b40da049 | 4448 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
4449 | if (batch * PAGE_SIZE > 512 * 1024) |
4450 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
4451 | batch /= 4; /* We effectively *= 4 below */ |
4452 | if (batch < 1) | |
4453 | batch = 1; | |
4454 | ||
4455 | /* | |
0ceaacc9 NP |
4456 | * Clamp the batch to a 2^n - 1 value. Having a power |
4457 | * of 2 value was found to be more likely to have | |
4458 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 4459 | * |
0ceaacc9 NP |
4460 | * For example if 2 tasks are alternately allocating |
4461 | * batches of pages, one task can end up with a lot | |
4462 | * of pages of one half of the possible page colors | |
4463 | * and the other with pages of the other colors. | |
e7c8d5c9 | 4464 | */ |
9155203a | 4465 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 4466 | |
e7c8d5c9 | 4467 | return batch; |
3a6be87f DH |
4468 | |
4469 | #else | |
4470 | /* The deferral and batching of frees should be suppressed under NOMMU | |
4471 | * conditions. | |
4472 | * | |
4473 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4474 | * of contiguous memory as there's no hardware page translation to | |
4475 | * assemble apparent contiguous memory from discontiguous pages. | |
4476 | * | |
4477 | * Queueing large contiguous runs of pages for batching, however, | |
4478 | * causes the pages to actually be freed in smaller chunks. As there | |
4479 | * can be a significant delay between the individual batches being | |
4480 | * recycled, this leads to the once large chunks of space being | |
4481 | * fragmented and becoming unavailable for high-order allocations. | |
4482 | */ | |
4483 | return 0; | |
4484 | #endif | |
e7c8d5c9 CL |
4485 | } |
4486 | ||
8d7a8fa9 CS |
4487 | /* |
4488 | * pcp->high and pcp->batch values are related and dependent on one another: | |
4489 | * ->batch must never be higher then ->high. | |
4490 | * The following function updates them in a safe manner without read side | |
4491 | * locking. | |
4492 | * | |
4493 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
4494 | * those fields changing asynchronously (acording the the above rule). | |
4495 | * | |
4496 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
4497 | * outside of boot time (or some other assurance that no concurrent updaters | |
4498 | * exist). | |
4499 | */ | |
4500 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
4501 | unsigned long batch) | |
4502 | { | |
4503 | /* start with a fail safe value for batch */ | |
4504 | pcp->batch = 1; | |
4505 | smp_wmb(); | |
4506 | ||
4507 | /* Update high, then batch, in order */ | |
4508 | pcp->high = high; | |
4509 | smp_wmb(); | |
4510 | ||
4511 | pcp->batch = batch; | |
4512 | } | |
4513 | ||
3664033c | 4514 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
4515 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
4516 | { | |
8d7a8fa9 | 4517 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
4518 | } |
4519 | ||
88c90dbc | 4520 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
4521 | { |
4522 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4523 | int migratetype; |
2caaad41 | 4524 | |
1c6fe946 MD |
4525 | memset(p, 0, sizeof(*p)); |
4526 | ||
3dfa5721 | 4527 | pcp = &p->pcp; |
2caaad41 | 4528 | pcp->count = 0; |
5f8dcc21 MG |
4529 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4530 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4531 | } |
4532 | ||
88c90dbc CS |
4533 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
4534 | { | |
4535 | pageset_init(p); | |
4536 | pageset_set_batch(p, batch); | |
4537 | } | |
4538 | ||
8ad4b1fb | 4539 | /* |
3664033c | 4540 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
4541 | * to the value high for the pageset p. |
4542 | */ | |
3664033c | 4543 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
4544 | unsigned long high) |
4545 | { | |
8d7a8fa9 CS |
4546 | unsigned long batch = max(1UL, high / 4); |
4547 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
4548 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 4549 | |
8d7a8fa9 | 4550 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
4551 | } |
4552 | ||
7cd2b0a3 DR |
4553 | static void pageset_set_high_and_batch(struct zone *zone, |
4554 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 4555 | { |
56cef2b8 | 4556 | if (percpu_pagelist_fraction) |
3664033c | 4557 | pageset_set_high(pcp, |
56cef2b8 CS |
4558 | (zone->managed_pages / |
4559 | percpu_pagelist_fraction)); | |
4560 | else | |
4561 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
4562 | } | |
4563 | ||
169f6c19 CS |
4564 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
4565 | { | |
4566 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4567 | ||
4568 | pageset_init(pcp); | |
4569 | pageset_set_high_and_batch(zone, pcp); | |
4570 | } | |
4571 | ||
4ed7e022 | 4572 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4573 | { |
4574 | int cpu; | |
319774e2 | 4575 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
4576 | for_each_possible_cpu(cpu) |
4577 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
4578 | } |
4579 | ||
2caaad41 | 4580 | /* |
99dcc3e5 CL |
4581 | * Allocate per cpu pagesets and initialize them. |
4582 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4583 | */ |
99dcc3e5 | 4584 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4585 | { |
99dcc3e5 | 4586 | struct zone *zone; |
e7c8d5c9 | 4587 | |
319774e2 WF |
4588 | for_each_populated_zone(zone) |
4589 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4590 | } |
4591 | ||
577a32f6 | 4592 | static noinline __init_refok |
cca448fe | 4593 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4594 | { |
4595 | int i; | |
cca448fe | 4596 | size_t alloc_size; |
ed8ece2e DH |
4597 | |
4598 | /* | |
4599 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4600 | * per zone. | |
4601 | */ | |
02b694de YG |
4602 | zone->wait_table_hash_nr_entries = |
4603 | wait_table_hash_nr_entries(zone_size_pages); | |
4604 | zone->wait_table_bits = | |
4605 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4606 | alloc_size = zone->wait_table_hash_nr_entries |
4607 | * sizeof(wait_queue_head_t); | |
4608 | ||
cd94b9db | 4609 | if (!slab_is_available()) { |
cca448fe | 4610 | zone->wait_table = (wait_queue_head_t *) |
6782832e SS |
4611 | memblock_virt_alloc_node_nopanic( |
4612 | alloc_size, zone->zone_pgdat->node_id); | |
cca448fe YG |
4613 | } else { |
4614 | /* | |
4615 | * This case means that a zone whose size was 0 gets new memory | |
4616 | * via memory hot-add. | |
4617 | * But it may be the case that a new node was hot-added. In | |
4618 | * this case vmalloc() will not be able to use this new node's | |
4619 | * memory - this wait_table must be initialized to use this new | |
4620 | * node itself as well. | |
4621 | * To use this new node's memory, further consideration will be | |
4622 | * necessary. | |
4623 | */ | |
8691f3a7 | 4624 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4625 | } |
4626 | if (!zone->wait_table) | |
4627 | return -ENOMEM; | |
ed8ece2e | 4628 | |
b8af2941 | 4629 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4630 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4631 | |
4632 | return 0; | |
ed8ece2e DH |
4633 | } |
4634 | ||
c09b4240 | 4635 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4636 | { |
99dcc3e5 CL |
4637 | /* |
4638 | * per cpu subsystem is not up at this point. The following code | |
4639 | * relies on the ability of the linker to provide the | |
4640 | * offset of a (static) per cpu variable into the per cpu area. | |
4641 | */ | |
4642 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4643 | |
b38a8725 | 4644 | if (populated_zone(zone)) |
99dcc3e5 CL |
4645 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4646 | zone->name, zone->present_pages, | |
4647 | zone_batchsize(zone)); | |
ed8ece2e DH |
4648 | } |
4649 | ||
4ed7e022 | 4650 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4651 | unsigned long zone_start_pfn, |
a2f3aa02 DH |
4652 | unsigned long size, |
4653 | enum memmap_context context) | |
ed8ece2e DH |
4654 | { |
4655 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4656 | int ret; |
4657 | ret = zone_wait_table_init(zone, size); | |
4658 | if (ret) | |
4659 | return ret; | |
ed8ece2e DH |
4660 | pgdat->nr_zones = zone_idx(zone) + 1; |
4661 | ||
ed8ece2e DH |
4662 | zone->zone_start_pfn = zone_start_pfn; |
4663 | ||
708614e6 MG |
4664 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4665 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4666 | pgdat->node_id, | |
4667 | (unsigned long)zone_idx(zone), | |
4668 | zone_start_pfn, (zone_start_pfn + size)); | |
4669 | ||
1e548deb | 4670 | zone_init_free_lists(zone); |
718127cc YG |
4671 | |
4672 | return 0; | |
ed8ece2e DH |
4673 | } |
4674 | ||
0ee332c1 | 4675 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4676 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 4677 | |
c713216d MG |
4678 | /* |
4679 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 4680 | */ |
8a942fde MG |
4681 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
4682 | struct mminit_pfnnid_cache *state) | |
c713216d | 4683 | { |
c13291a5 | 4684 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 4685 | int nid; |
7c243c71 | 4686 | |
8a942fde MG |
4687 | if (state->last_start <= pfn && pfn < state->last_end) |
4688 | return state->last_nid; | |
c713216d | 4689 | |
e76b63f8 YL |
4690 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
4691 | if (nid != -1) { | |
8a942fde MG |
4692 | state->last_start = start_pfn; |
4693 | state->last_end = end_pfn; | |
4694 | state->last_nid = nid; | |
e76b63f8 YL |
4695 | } |
4696 | ||
4697 | return nid; | |
c713216d MG |
4698 | } |
4699 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
4700 | ||
c713216d | 4701 | /** |
6782832e | 4702 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 4703 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 4704 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 4705 | * |
7d018176 ZZ |
4706 | * If an architecture guarantees that all ranges registered contain no holes |
4707 | * and may be freed, this this function may be used instead of calling | |
4708 | * memblock_free_early_nid() manually. | |
c713216d | 4709 | */ |
c13291a5 | 4710 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 4711 | { |
c13291a5 TH |
4712 | unsigned long start_pfn, end_pfn; |
4713 | int i, this_nid; | |
edbe7d23 | 4714 | |
c13291a5 TH |
4715 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
4716 | start_pfn = min(start_pfn, max_low_pfn); | |
4717 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 4718 | |
c13291a5 | 4719 | if (start_pfn < end_pfn) |
6782832e SS |
4720 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
4721 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
4722 | this_nid); | |
edbe7d23 | 4723 | } |
edbe7d23 | 4724 | } |
edbe7d23 | 4725 | |
c713216d MG |
4726 | /** |
4727 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 4728 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 4729 | * |
7d018176 ZZ |
4730 | * If an architecture guarantees that all ranges registered contain no holes and may |
4731 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
4732 | */ |
4733 | void __init sparse_memory_present_with_active_regions(int nid) | |
4734 | { | |
c13291a5 TH |
4735 | unsigned long start_pfn, end_pfn; |
4736 | int i, this_nid; | |
c713216d | 4737 | |
c13291a5 TH |
4738 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
4739 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
4740 | } |
4741 | ||
4742 | /** | |
4743 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
4744 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
4745 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
4746 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
4747 | * |
4748 | * It returns the start and end page frame of a node based on information | |
7d018176 | 4749 | * provided by memblock_set_node(). If called for a node |
c713216d | 4750 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 4751 | * PFNs will be 0. |
c713216d | 4752 | */ |
a3142c8e | 4753 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
4754 | unsigned long *start_pfn, unsigned long *end_pfn) |
4755 | { | |
c13291a5 | 4756 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 4757 | int i; |
c13291a5 | 4758 | |
c713216d MG |
4759 | *start_pfn = -1UL; |
4760 | *end_pfn = 0; | |
4761 | ||
c13291a5 TH |
4762 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
4763 | *start_pfn = min(*start_pfn, this_start_pfn); | |
4764 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
4765 | } |
4766 | ||
633c0666 | 4767 | if (*start_pfn == -1UL) |
c713216d | 4768 | *start_pfn = 0; |
c713216d MG |
4769 | } |
4770 | ||
2a1e274a MG |
4771 | /* |
4772 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
4773 | * assumption is made that zones within a node are ordered in monotonic | |
4774 | * increasing memory addresses so that the "highest" populated zone is used | |
4775 | */ | |
b69a7288 | 4776 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
4777 | { |
4778 | int zone_index; | |
4779 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
4780 | if (zone_index == ZONE_MOVABLE) | |
4781 | continue; | |
4782 | ||
4783 | if (arch_zone_highest_possible_pfn[zone_index] > | |
4784 | arch_zone_lowest_possible_pfn[zone_index]) | |
4785 | break; | |
4786 | } | |
4787 | ||
4788 | VM_BUG_ON(zone_index == -1); | |
4789 | movable_zone = zone_index; | |
4790 | } | |
4791 | ||
4792 | /* | |
4793 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 4794 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
4795 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
4796 | * in each node depending on the size of each node and how evenly kernelcore | |
4797 | * is distributed. This helper function adjusts the zone ranges | |
4798 | * provided by the architecture for a given node by using the end of the | |
4799 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
4800 | * zones within a node are in order of monotonic increases memory addresses | |
4801 | */ | |
b69a7288 | 4802 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4803 | unsigned long zone_type, |
4804 | unsigned long node_start_pfn, | |
4805 | unsigned long node_end_pfn, | |
4806 | unsigned long *zone_start_pfn, | |
4807 | unsigned long *zone_end_pfn) | |
4808 | { | |
4809 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4810 | if (zone_movable_pfn[nid]) { | |
4811 | /* Size ZONE_MOVABLE */ | |
4812 | if (zone_type == ZONE_MOVABLE) { | |
4813 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4814 | *zone_end_pfn = min(node_end_pfn, | |
4815 | arch_zone_highest_possible_pfn[movable_zone]); | |
4816 | ||
4817 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4818 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4819 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4820 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4821 | ||
4822 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4823 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4824 | *zone_start_pfn = *zone_end_pfn; | |
4825 | } | |
4826 | } | |
4827 | ||
c713216d MG |
4828 | /* |
4829 | * Return the number of pages a zone spans in a node, including holes | |
4830 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4831 | */ | |
6ea6e688 | 4832 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4833 | unsigned long zone_type, |
7960aedd ZY |
4834 | unsigned long node_start_pfn, |
4835 | unsigned long node_end_pfn, | |
c713216d MG |
4836 | unsigned long *ignored) |
4837 | { | |
c713216d MG |
4838 | unsigned long zone_start_pfn, zone_end_pfn; |
4839 | ||
7960aedd | 4840 | /* Get the start and end of the zone */ |
c713216d MG |
4841 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
4842 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4843 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4844 | node_start_pfn, node_end_pfn, | |
4845 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4846 | |
4847 | /* Check that this node has pages within the zone's required range */ | |
4848 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4849 | return 0; | |
4850 | ||
4851 | /* Move the zone boundaries inside the node if necessary */ | |
4852 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4853 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4854 | ||
4855 | /* Return the spanned pages */ | |
4856 | return zone_end_pfn - zone_start_pfn; | |
4857 | } | |
4858 | ||
4859 | /* | |
4860 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4861 | * then all holes in the requested range will be accounted for. |
c713216d | 4862 | */ |
32996250 | 4863 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4864 | unsigned long range_start_pfn, |
4865 | unsigned long range_end_pfn) | |
4866 | { | |
96e907d1 TH |
4867 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4868 | unsigned long start_pfn, end_pfn; | |
4869 | int i; | |
c713216d | 4870 | |
96e907d1 TH |
4871 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4872 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4873 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4874 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4875 | } |
96e907d1 | 4876 | return nr_absent; |
c713216d MG |
4877 | } |
4878 | ||
4879 | /** | |
4880 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4881 | * @start_pfn: The start PFN to start searching for holes | |
4882 | * @end_pfn: The end PFN to stop searching for holes | |
4883 | * | |
88ca3b94 | 4884 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4885 | */ |
4886 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4887 | unsigned long end_pfn) | |
4888 | { | |
4889 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4890 | } | |
4891 | ||
4892 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4893 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4894 | unsigned long zone_type, |
7960aedd ZY |
4895 | unsigned long node_start_pfn, |
4896 | unsigned long node_end_pfn, | |
c713216d MG |
4897 | unsigned long *ignored) |
4898 | { | |
96e907d1 TH |
4899 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4900 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4901 | unsigned long zone_start_pfn, zone_end_pfn; |
4902 | ||
96e907d1 TH |
4903 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4904 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4905 | |
2a1e274a MG |
4906 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4907 | node_start_pfn, node_end_pfn, | |
4908 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4909 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4910 | } |
0e0b864e | 4911 | |
0ee332c1 | 4912 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4913 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4914 | unsigned long zone_type, |
7960aedd ZY |
4915 | unsigned long node_start_pfn, |
4916 | unsigned long node_end_pfn, | |
c713216d MG |
4917 | unsigned long *zones_size) |
4918 | { | |
4919 | return zones_size[zone_type]; | |
4920 | } | |
4921 | ||
6ea6e688 | 4922 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4923 | unsigned long zone_type, |
7960aedd ZY |
4924 | unsigned long node_start_pfn, |
4925 | unsigned long node_end_pfn, | |
c713216d MG |
4926 | unsigned long *zholes_size) |
4927 | { | |
4928 | if (!zholes_size) | |
4929 | return 0; | |
4930 | ||
4931 | return zholes_size[zone_type]; | |
4932 | } | |
20e6926d | 4933 | |
0ee332c1 | 4934 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4935 | |
a3142c8e | 4936 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
4937 | unsigned long node_start_pfn, |
4938 | unsigned long node_end_pfn, | |
4939 | unsigned long *zones_size, | |
4940 | unsigned long *zholes_size) | |
c713216d | 4941 | { |
febd5949 | 4942 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
4943 | enum zone_type i; |
4944 | ||
febd5949 GZ |
4945 | for (i = 0; i < MAX_NR_ZONES; i++) { |
4946 | struct zone *zone = pgdat->node_zones + i; | |
4947 | unsigned long size, real_size; | |
c713216d | 4948 | |
febd5949 GZ |
4949 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
4950 | node_start_pfn, | |
4951 | node_end_pfn, | |
4952 | zones_size); | |
4953 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4954 | node_start_pfn, node_end_pfn, |
4955 | zholes_size); | |
febd5949 GZ |
4956 | zone->spanned_pages = size; |
4957 | zone->present_pages = real_size; | |
4958 | ||
4959 | totalpages += size; | |
4960 | realtotalpages += real_size; | |
4961 | } | |
4962 | ||
4963 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
4964 | pgdat->node_present_pages = realtotalpages; |
4965 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4966 | realtotalpages); | |
4967 | } | |
4968 | ||
835c134e MG |
4969 | #ifndef CONFIG_SPARSEMEM |
4970 | /* | |
4971 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4972 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4973 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4974 | * round what is now in bits to nearest long in bits, then return it in |
4975 | * bytes. | |
4976 | */ | |
7c45512d | 4977 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
4978 | { |
4979 | unsigned long usemapsize; | |
4980 | ||
7c45512d | 4981 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
4982 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4983 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4984 | usemapsize *= NR_PAGEBLOCK_BITS; |
4985 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4986 | ||
4987 | return usemapsize / 8; | |
4988 | } | |
4989 | ||
4990 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
4991 | struct zone *zone, |
4992 | unsigned long zone_start_pfn, | |
4993 | unsigned long zonesize) | |
835c134e | 4994 | { |
7c45512d | 4995 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 4996 | zone->pageblock_flags = NULL; |
58a01a45 | 4997 | if (usemapsize) |
6782832e SS |
4998 | zone->pageblock_flags = |
4999 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5000 | pgdat->node_id); | |
835c134e MG |
5001 | } |
5002 | #else | |
7c45512d LT |
5003 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5004 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5005 | #endif /* CONFIG_SPARSEMEM */ |
5006 | ||
d9c23400 | 5007 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5008 | |
d9c23400 | 5009 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5010 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5011 | { |
955c1cd7 AM |
5012 | unsigned int order; |
5013 | ||
d9c23400 MG |
5014 | /* Check that pageblock_nr_pages has not already been setup */ |
5015 | if (pageblock_order) | |
5016 | return; | |
5017 | ||
955c1cd7 AM |
5018 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5019 | order = HUGETLB_PAGE_ORDER; | |
5020 | else | |
5021 | order = MAX_ORDER - 1; | |
5022 | ||
d9c23400 MG |
5023 | /* |
5024 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5025 | * This value may be variable depending on boot parameters on IA64 and |
5026 | * powerpc. | |
d9c23400 MG |
5027 | */ |
5028 | pageblock_order = order; | |
5029 | } | |
5030 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5031 | ||
ba72cb8c MG |
5032 | /* |
5033 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5034 | * is unused as pageblock_order is set at compile-time. See |
5035 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5036 | * the kernel config | |
ba72cb8c | 5037 | */ |
15ca220e | 5038 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5039 | { |
ba72cb8c | 5040 | } |
d9c23400 MG |
5041 | |
5042 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5043 | ||
01cefaef JL |
5044 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5045 | unsigned long present_pages) | |
5046 | { | |
5047 | unsigned long pages = spanned_pages; | |
5048 | ||
5049 | /* | |
5050 | * Provide a more accurate estimation if there are holes within | |
5051 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5052 | * zone, each populated memory region may cost us one or two extra | |
5053 | * memmap pages due to alignment because memmap pages for each | |
5054 | * populated regions may not naturally algined on page boundary. | |
5055 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
5056 | */ | |
5057 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5058 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5059 | pages = present_pages; | |
5060 | ||
5061 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5062 | } | |
5063 | ||
1da177e4 LT |
5064 | /* |
5065 | * Set up the zone data structures: | |
5066 | * - mark all pages reserved | |
5067 | * - mark all memory queues empty | |
5068 | * - clear the memory bitmaps | |
6527af5d MK |
5069 | * |
5070 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 5071 | */ |
b5a0e011 | 5072 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
febd5949 | 5073 | unsigned long node_start_pfn, unsigned long node_end_pfn) |
1da177e4 | 5074 | { |
2f1b6248 | 5075 | enum zone_type j; |
ed8ece2e | 5076 | int nid = pgdat->node_id; |
1da177e4 | 5077 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 5078 | int ret; |
1da177e4 | 5079 | |
208d54e5 | 5080 | pgdat_resize_init(pgdat); |
8177a420 AA |
5081 | #ifdef CONFIG_NUMA_BALANCING |
5082 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
5083 | pgdat->numabalancing_migrate_nr_pages = 0; | |
5084 | pgdat->numabalancing_migrate_next_window = jiffies; | |
5085 | #endif | |
1da177e4 | 5086 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 5087 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
eefa864b | 5088 | pgdat_page_ext_init(pgdat); |
5f63b720 | 5089 | |
1da177e4 LT |
5090 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5091 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 5092 | unsigned long size, realsize, freesize, memmap_pages; |
1da177e4 | 5093 | |
febd5949 GZ |
5094 | size = zone->spanned_pages; |
5095 | realsize = freesize = zone->present_pages; | |
1da177e4 | 5096 | |
0e0b864e | 5097 | /* |
9feedc9d | 5098 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
5099 | * is used by this zone for memmap. This affects the watermark |
5100 | * and per-cpu initialisations | |
5101 | */ | |
01cefaef | 5102 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
5103 | if (!is_highmem_idx(j)) { |
5104 | if (freesize >= memmap_pages) { | |
5105 | freesize -= memmap_pages; | |
5106 | if (memmap_pages) | |
5107 | printk(KERN_DEBUG | |
5108 | " %s zone: %lu pages used for memmap\n", | |
5109 | zone_names[j], memmap_pages); | |
5110 | } else | |
5111 | printk(KERN_WARNING | |
5112 | " %s zone: %lu pages exceeds freesize %lu\n", | |
5113 | zone_names[j], memmap_pages, freesize); | |
5114 | } | |
0e0b864e | 5115 | |
6267276f | 5116 | /* Account for reserved pages */ |
9feedc9d JL |
5117 | if (j == 0 && freesize > dma_reserve) { |
5118 | freesize -= dma_reserve; | |
d903ef9f | 5119 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 5120 | zone_names[0], dma_reserve); |
0e0b864e MG |
5121 | } |
5122 | ||
98d2b0eb | 5123 | if (!is_highmem_idx(j)) |
9feedc9d | 5124 | nr_kernel_pages += freesize; |
01cefaef JL |
5125 | /* Charge for highmem memmap if there are enough kernel pages */ |
5126 | else if (nr_kernel_pages > memmap_pages * 2) | |
5127 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 5128 | nr_all_pages += freesize; |
1da177e4 | 5129 | |
9feedc9d JL |
5130 | /* |
5131 | * Set an approximate value for lowmem here, it will be adjusted | |
5132 | * when the bootmem allocator frees pages into the buddy system. | |
5133 | * And all highmem pages will be managed by the buddy system. | |
5134 | */ | |
5135 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 5136 | #ifdef CONFIG_NUMA |
d5f541ed | 5137 | zone->node = nid; |
9feedc9d | 5138 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 5139 | / 100; |
9feedc9d | 5140 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 5141 | #endif |
1da177e4 LT |
5142 | zone->name = zone_names[j]; |
5143 | spin_lock_init(&zone->lock); | |
5144 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 5145 | zone_seqlock_init(zone); |
1da177e4 | 5146 | zone->zone_pgdat = pgdat; |
ed8ece2e | 5147 | zone_pcp_init(zone); |
81c0a2bb JW |
5148 | |
5149 | /* For bootup, initialized properly in watermark setup */ | |
5150 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
5151 | ||
bea8c150 | 5152 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
5153 | if (!size) |
5154 | continue; | |
5155 | ||
955c1cd7 | 5156 | set_pageblock_order(); |
7c45512d | 5157 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
a2f3aa02 DH |
5158 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
5159 | size, MEMMAP_EARLY); | |
718127cc | 5160 | BUG_ON(ret); |
76cdd58e | 5161 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 5162 | zone_start_pfn += size; |
1da177e4 LT |
5163 | } |
5164 | } | |
5165 | ||
577a32f6 | 5166 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 5167 | { |
1da177e4 LT |
5168 | /* Skip empty nodes */ |
5169 | if (!pgdat->node_spanned_pages) | |
5170 | return; | |
5171 | ||
d41dee36 | 5172 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
5173 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
5174 | if (!pgdat->node_mem_map) { | |
e984bb43 | 5175 | unsigned long size, start, end; |
d41dee36 AW |
5176 | struct page *map; |
5177 | ||
e984bb43 BP |
5178 | /* |
5179 | * The zone's endpoints aren't required to be MAX_ORDER | |
5180 | * aligned but the node_mem_map endpoints must be in order | |
5181 | * for the buddy allocator to function correctly. | |
5182 | */ | |
5183 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
108bcc96 | 5184 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
5185 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
5186 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
5187 | map = alloc_remap(pgdat->node_id, size); |
5188 | if (!map) | |
6782832e SS |
5189 | map = memblock_virt_alloc_node_nopanic(size, |
5190 | pgdat->node_id); | |
e984bb43 | 5191 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 5192 | } |
12d810c1 | 5193 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
5194 | /* |
5195 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
5196 | */ | |
c713216d | 5197 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 5198 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 5199 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5200 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 5201 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 5202 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5203 | } |
1da177e4 | 5204 | #endif |
d41dee36 | 5205 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
5206 | } |
5207 | ||
9109fb7b JW |
5208 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
5209 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 5210 | { |
9109fb7b | 5211 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
5212 | unsigned long start_pfn = 0; |
5213 | unsigned long end_pfn = 0; | |
9109fb7b | 5214 | |
88fdf75d | 5215 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 5216 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 5217 | |
3a80a7fa | 5218 | reset_deferred_meminit(pgdat); |
1da177e4 LT |
5219 | pgdat->node_id = nid; |
5220 | pgdat->node_start_pfn = node_start_pfn; | |
7960aedd ZY |
5221 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5222 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a JG |
5223 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
5224 | (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1); | |
7960aedd ZY |
5225 | #endif |
5226 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
5227 | zones_size, zholes_size); | |
1da177e4 LT |
5228 | |
5229 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
5230 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
5231 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
5232 | nid, (unsigned long)pgdat, | |
5233 | (unsigned long)pgdat->node_mem_map); | |
5234 | #endif | |
1da177e4 | 5235 | |
febd5949 | 5236 | free_area_init_core(pgdat, start_pfn, end_pfn); |
1da177e4 LT |
5237 | } |
5238 | ||
0ee332c1 | 5239 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
5240 | |
5241 | #if MAX_NUMNODES > 1 | |
5242 | /* | |
5243 | * Figure out the number of possible node ids. | |
5244 | */ | |
f9872caf | 5245 | void __init setup_nr_node_ids(void) |
418508c1 MS |
5246 | { |
5247 | unsigned int node; | |
5248 | unsigned int highest = 0; | |
5249 | ||
5250 | for_each_node_mask(node, node_possible_map) | |
5251 | highest = node; | |
5252 | nr_node_ids = highest + 1; | |
5253 | } | |
418508c1 MS |
5254 | #endif |
5255 | ||
1e01979c TH |
5256 | /** |
5257 | * node_map_pfn_alignment - determine the maximum internode alignment | |
5258 | * | |
5259 | * This function should be called after node map is populated and sorted. | |
5260 | * It calculates the maximum power of two alignment which can distinguish | |
5261 | * all the nodes. | |
5262 | * | |
5263 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
5264 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
5265 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
5266 | * shifted, 1GiB is enough and this function will indicate so. | |
5267 | * | |
5268 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
5269 | * model has fine enough granularity to avoid incorrect mapping for the | |
5270 | * populated node map. | |
5271 | * | |
5272 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
5273 | * requirement (single node). | |
5274 | */ | |
5275 | unsigned long __init node_map_pfn_alignment(void) | |
5276 | { | |
5277 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 5278 | unsigned long start, end, mask; |
1e01979c | 5279 | int last_nid = -1; |
c13291a5 | 5280 | int i, nid; |
1e01979c | 5281 | |
c13291a5 | 5282 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5283 | if (!start || last_nid < 0 || last_nid == nid) { |
5284 | last_nid = nid; | |
5285 | last_end = end; | |
5286 | continue; | |
5287 | } | |
5288 | ||
5289 | /* | |
5290 | * Start with a mask granular enough to pin-point to the | |
5291 | * start pfn and tick off bits one-by-one until it becomes | |
5292 | * too coarse to separate the current node from the last. | |
5293 | */ | |
5294 | mask = ~((1 << __ffs(start)) - 1); | |
5295 | while (mask && last_end <= (start & (mask << 1))) | |
5296 | mask <<= 1; | |
5297 | ||
5298 | /* accumulate all internode masks */ | |
5299 | accl_mask |= mask; | |
5300 | } | |
5301 | ||
5302 | /* convert mask to number of pages */ | |
5303 | return ~accl_mask + 1; | |
5304 | } | |
5305 | ||
a6af2bc3 | 5306 | /* Find the lowest pfn for a node */ |
b69a7288 | 5307 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 5308 | { |
a6af2bc3 | 5309 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
5310 | unsigned long start_pfn; |
5311 | int i; | |
1abbfb41 | 5312 | |
c13291a5 TH |
5313 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
5314 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 5315 | |
a6af2bc3 MG |
5316 | if (min_pfn == ULONG_MAX) { |
5317 | printk(KERN_WARNING | |
2bc0d261 | 5318 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
5319 | return 0; |
5320 | } | |
5321 | ||
5322 | return min_pfn; | |
c713216d MG |
5323 | } |
5324 | ||
5325 | /** | |
5326 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
5327 | * | |
5328 | * It returns the minimum PFN based on information provided via | |
7d018176 | 5329 | * memblock_set_node(). |
c713216d MG |
5330 | */ |
5331 | unsigned long __init find_min_pfn_with_active_regions(void) | |
5332 | { | |
5333 | return find_min_pfn_for_node(MAX_NUMNODES); | |
5334 | } | |
5335 | ||
37b07e41 LS |
5336 | /* |
5337 | * early_calculate_totalpages() | |
5338 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 5339 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 5340 | */ |
484f51f8 | 5341 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 5342 | { |
7e63efef | 5343 | unsigned long totalpages = 0; |
c13291a5 TH |
5344 | unsigned long start_pfn, end_pfn; |
5345 | int i, nid; | |
5346 | ||
5347 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
5348 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 5349 | |
37b07e41 LS |
5350 | totalpages += pages; |
5351 | if (pages) | |
4b0ef1fe | 5352 | node_set_state(nid, N_MEMORY); |
37b07e41 | 5353 | } |
b8af2941 | 5354 | return totalpages; |
7e63efef MG |
5355 | } |
5356 | ||
2a1e274a MG |
5357 | /* |
5358 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
5359 | * is spread evenly between nodes as long as the nodes have enough | |
5360 | * memory. When they don't, some nodes will have more kernelcore than | |
5361 | * others | |
5362 | */ | |
b224ef85 | 5363 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
5364 | { |
5365 | int i, nid; | |
5366 | unsigned long usable_startpfn; | |
5367 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 5368 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 5369 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 5370 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 5371 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 5372 | struct memblock_region *r; |
b2f3eebe TC |
5373 | |
5374 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
5375 | find_usable_zone_for_movable(); | |
5376 | ||
5377 | /* | |
5378 | * If movable_node is specified, ignore kernelcore and movablecore | |
5379 | * options. | |
5380 | */ | |
5381 | if (movable_node_is_enabled()) { | |
136199f0 EM |
5382 | for_each_memblock(memory, r) { |
5383 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
5384 | continue; |
5385 | ||
136199f0 | 5386 | nid = r->nid; |
b2f3eebe | 5387 | |
136199f0 | 5388 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
5389 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
5390 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5391 | usable_startpfn; | |
5392 | } | |
5393 | ||
5394 | goto out2; | |
5395 | } | |
2a1e274a | 5396 | |
7e63efef | 5397 | /* |
b2f3eebe | 5398 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
5399 | * kernelcore that corresponds so that memory usable for |
5400 | * any allocation type is evenly spread. If both kernelcore | |
5401 | * and movablecore are specified, then the value of kernelcore | |
5402 | * will be used for required_kernelcore if it's greater than | |
5403 | * what movablecore would have allowed. | |
5404 | */ | |
5405 | if (required_movablecore) { | |
7e63efef MG |
5406 | unsigned long corepages; |
5407 | ||
5408 | /* | |
5409 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
5410 | * was requested by the user | |
5411 | */ | |
5412 | required_movablecore = | |
5413 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
5414 | corepages = totalpages - required_movablecore; | |
5415 | ||
5416 | required_kernelcore = max(required_kernelcore, corepages); | |
5417 | } | |
5418 | ||
20e6926d YL |
5419 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
5420 | if (!required_kernelcore) | |
66918dcd | 5421 | goto out; |
2a1e274a MG |
5422 | |
5423 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
5424 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
5425 | ||
5426 | restart: | |
5427 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
5428 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 5429 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
5430 | unsigned long start_pfn, end_pfn; |
5431 | ||
2a1e274a MG |
5432 | /* |
5433 | * Recalculate kernelcore_node if the division per node | |
5434 | * now exceeds what is necessary to satisfy the requested | |
5435 | * amount of memory for the kernel | |
5436 | */ | |
5437 | if (required_kernelcore < kernelcore_node) | |
5438 | kernelcore_node = required_kernelcore / usable_nodes; | |
5439 | ||
5440 | /* | |
5441 | * As the map is walked, we track how much memory is usable | |
5442 | * by the kernel using kernelcore_remaining. When it is | |
5443 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
5444 | */ | |
5445 | kernelcore_remaining = kernelcore_node; | |
5446 | ||
5447 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 5448 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
5449 | unsigned long size_pages; |
5450 | ||
c13291a5 | 5451 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
5452 | if (start_pfn >= end_pfn) |
5453 | continue; | |
5454 | ||
5455 | /* Account for what is only usable for kernelcore */ | |
5456 | if (start_pfn < usable_startpfn) { | |
5457 | unsigned long kernel_pages; | |
5458 | kernel_pages = min(end_pfn, usable_startpfn) | |
5459 | - start_pfn; | |
5460 | ||
5461 | kernelcore_remaining -= min(kernel_pages, | |
5462 | kernelcore_remaining); | |
5463 | required_kernelcore -= min(kernel_pages, | |
5464 | required_kernelcore); | |
5465 | ||
5466 | /* Continue if range is now fully accounted */ | |
5467 | if (end_pfn <= usable_startpfn) { | |
5468 | ||
5469 | /* | |
5470 | * Push zone_movable_pfn to the end so | |
5471 | * that if we have to rebalance | |
5472 | * kernelcore across nodes, we will | |
5473 | * not double account here | |
5474 | */ | |
5475 | zone_movable_pfn[nid] = end_pfn; | |
5476 | continue; | |
5477 | } | |
5478 | start_pfn = usable_startpfn; | |
5479 | } | |
5480 | ||
5481 | /* | |
5482 | * The usable PFN range for ZONE_MOVABLE is from | |
5483 | * start_pfn->end_pfn. Calculate size_pages as the | |
5484 | * number of pages used as kernelcore | |
5485 | */ | |
5486 | size_pages = end_pfn - start_pfn; | |
5487 | if (size_pages > kernelcore_remaining) | |
5488 | size_pages = kernelcore_remaining; | |
5489 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5490 | ||
5491 | /* | |
5492 | * Some kernelcore has been met, update counts and | |
5493 | * break if the kernelcore for this node has been | |
b8af2941 | 5494 | * satisfied |
2a1e274a MG |
5495 | */ |
5496 | required_kernelcore -= min(required_kernelcore, | |
5497 | size_pages); | |
5498 | kernelcore_remaining -= size_pages; | |
5499 | if (!kernelcore_remaining) | |
5500 | break; | |
5501 | } | |
5502 | } | |
5503 | ||
5504 | /* | |
5505 | * If there is still required_kernelcore, we do another pass with one | |
5506 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5507 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 5508 | * satisfied |
2a1e274a MG |
5509 | */ |
5510 | usable_nodes--; | |
5511 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5512 | goto restart; | |
5513 | ||
b2f3eebe | 5514 | out2: |
2a1e274a MG |
5515 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
5516 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5517 | zone_movable_pfn[nid] = | |
5518 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5519 | |
20e6926d | 5520 | out: |
66918dcd | 5521 | /* restore the node_state */ |
4b0ef1fe | 5522 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5523 | } |
5524 | ||
4b0ef1fe LJ |
5525 | /* Any regular or high memory on that node ? */ |
5526 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5527 | { |
37b07e41 LS |
5528 | enum zone_type zone_type; |
5529 | ||
4b0ef1fe LJ |
5530 | if (N_MEMORY == N_NORMAL_MEMORY) |
5531 | return; | |
5532 | ||
5533 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5534 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 5535 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
5536 | node_set_state(nid, N_HIGH_MEMORY); |
5537 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5538 | zone_type <= ZONE_NORMAL) | |
5539 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5540 | break; |
5541 | } | |
37b07e41 | 5542 | } |
37b07e41 LS |
5543 | } |
5544 | ||
c713216d MG |
5545 | /** |
5546 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 5547 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
5548 | * |
5549 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 5550 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
5551 | * zone in each node and their holes is calculated. If the maximum PFN |
5552 | * between two adjacent zones match, it is assumed that the zone is empty. | |
5553 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
5554 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
5555 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
5556 | * at arch_max_dma_pfn. | |
5557 | */ | |
5558 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
5559 | { | |
c13291a5 TH |
5560 | unsigned long start_pfn, end_pfn; |
5561 | int i, nid; | |
a6af2bc3 | 5562 | |
c713216d MG |
5563 | /* Record where the zone boundaries are */ |
5564 | memset(arch_zone_lowest_possible_pfn, 0, | |
5565 | sizeof(arch_zone_lowest_possible_pfn)); | |
5566 | memset(arch_zone_highest_possible_pfn, 0, | |
5567 | sizeof(arch_zone_highest_possible_pfn)); | |
5568 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
5569 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
5570 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
5571 | if (i == ZONE_MOVABLE) |
5572 | continue; | |
c713216d MG |
5573 | arch_zone_lowest_possible_pfn[i] = |
5574 | arch_zone_highest_possible_pfn[i-1]; | |
5575 | arch_zone_highest_possible_pfn[i] = | |
5576 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
5577 | } | |
2a1e274a MG |
5578 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
5579 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
5580 | ||
5581 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
5582 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 5583 | find_zone_movable_pfns_for_nodes(); |
c713216d | 5584 | |
c713216d | 5585 | /* Print out the zone ranges */ |
f88dfff5 | 5586 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
5587 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5588 | if (i == ZONE_MOVABLE) | |
5589 | continue; | |
f88dfff5 | 5590 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
5591 | if (arch_zone_lowest_possible_pfn[i] == |
5592 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 5593 | pr_cont("empty\n"); |
72f0ba02 | 5594 | else |
8d29e18a JG |
5595 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
5596 | (u64)arch_zone_lowest_possible_pfn[i] | |
5597 | << PAGE_SHIFT, | |
5598 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 5599 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
5600 | } |
5601 | ||
5602 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 5603 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
5604 | for (i = 0; i < MAX_NUMNODES; i++) { |
5605 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
5606 | pr_info(" Node %d: %#018Lx\n", i, |
5607 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 5608 | } |
c713216d | 5609 | |
f2d52fe5 | 5610 | /* Print out the early node map */ |
f88dfff5 | 5611 | pr_info("Early memory node ranges\n"); |
c13291a5 | 5612 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
5613 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
5614 | (u64)start_pfn << PAGE_SHIFT, | |
5615 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
5616 | |
5617 | /* Initialise every node */ | |
708614e6 | 5618 | mminit_verify_pageflags_layout(); |
8ef82866 | 5619 | setup_nr_node_ids(); |
c713216d MG |
5620 | for_each_online_node(nid) { |
5621 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 5622 | free_area_init_node(nid, NULL, |
c713216d | 5623 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
5624 | |
5625 | /* Any memory on that node */ | |
5626 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
5627 | node_set_state(nid, N_MEMORY); |
5628 | check_for_memory(pgdat, nid); | |
c713216d MG |
5629 | } |
5630 | } | |
2a1e274a | 5631 | |
7e63efef | 5632 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
5633 | { |
5634 | unsigned long long coremem; | |
5635 | if (!p) | |
5636 | return -EINVAL; | |
5637 | ||
5638 | coremem = memparse(p, &p); | |
7e63efef | 5639 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 5640 | |
7e63efef | 5641 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
5642 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
5643 | ||
5644 | return 0; | |
5645 | } | |
ed7ed365 | 5646 | |
7e63efef MG |
5647 | /* |
5648 | * kernelcore=size sets the amount of memory for use for allocations that | |
5649 | * cannot be reclaimed or migrated. | |
5650 | */ | |
5651 | static int __init cmdline_parse_kernelcore(char *p) | |
5652 | { | |
5653 | return cmdline_parse_core(p, &required_kernelcore); | |
5654 | } | |
5655 | ||
5656 | /* | |
5657 | * movablecore=size sets the amount of memory for use for allocations that | |
5658 | * can be reclaimed or migrated. | |
5659 | */ | |
5660 | static int __init cmdline_parse_movablecore(char *p) | |
5661 | { | |
5662 | return cmdline_parse_core(p, &required_movablecore); | |
5663 | } | |
5664 | ||
ed7ed365 | 5665 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 5666 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 5667 | |
0ee332c1 | 5668 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5669 | |
c3d5f5f0 JL |
5670 | void adjust_managed_page_count(struct page *page, long count) |
5671 | { | |
5672 | spin_lock(&managed_page_count_lock); | |
5673 | page_zone(page)->managed_pages += count; | |
5674 | totalram_pages += count; | |
3dcc0571 JL |
5675 | #ifdef CONFIG_HIGHMEM |
5676 | if (PageHighMem(page)) | |
5677 | totalhigh_pages += count; | |
5678 | #endif | |
c3d5f5f0 JL |
5679 | spin_unlock(&managed_page_count_lock); |
5680 | } | |
3dcc0571 | 5681 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 5682 | |
11199692 | 5683 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 5684 | { |
11199692 JL |
5685 | void *pos; |
5686 | unsigned long pages = 0; | |
69afade7 | 5687 | |
11199692 JL |
5688 | start = (void *)PAGE_ALIGN((unsigned long)start); |
5689 | end = (void *)((unsigned long)end & PAGE_MASK); | |
5690 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 5691 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
5692 | memset(pos, poison, PAGE_SIZE); |
5693 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
5694 | } |
5695 | ||
5696 | if (pages && s) | |
11199692 | 5697 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
5698 | s, pages << (PAGE_SHIFT - 10), start, end); |
5699 | ||
5700 | return pages; | |
5701 | } | |
11199692 | 5702 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 5703 | |
cfa11e08 JL |
5704 | #ifdef CONFIG_HIGHMEM |
5705 | void free_highmem_page(struct page *page) | |
5706 | { | |
5707 | __free_reserved_page(page); | |
5708 | totalram_pages++; | |
7b4b2a0d | 5709 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
5710 | totalhigh_pages++; |
5711 | } | |
5712 | #endif | |
5713 | ||
7ee3d4e8 JL |
5714 | |
5715 | void __init mem_init_print_info(const char *str) | |
5716 | { | |
5717 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
5718 | unsigned long init_code_size, init_data_size; | |
5719 | ||
5720 | physpages = get_num_physpages(); | |
5721 | codesize = _etext - _stext; | |
5722 | datasize = _edata - _sdata; | |
5723 | rosize = __end_rodata - __start_rodata; | |
5724 | bss_size = __bss_stop - __bss_start; | |
5725 | init_data_size = __init_end - __init_begin; | |
5726 | init_code_size = _einittext - _sinittext; | |
5727 | ||
5728 | /* | |
5729 | * Detect special cases and adjust section sizes accordingly: | |
5730 | * 1) .init.* may be embedded into .data sections | |
5731 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
5732 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
5733 | * 3) .rodata.* may be embedded into .text or .data sections. | |
5734 | */ | |
5735 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
5736 | do { \ |
5737 | if (start <= pos && pos < end && size > adj) \ | |
5738 | size -= adj; \ | |
5739 | } while (0) | |
7ee3d4e8 JL |
5740 | |
5741 | adj_init_size(__init_begin, __init_end, init_data_size, | |
5742 | _sinittext, init_code_size); | |
5743 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
5744 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
5745 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
5746 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
5747 | ||
5748 | #undef adj_init_size | |
5749 | ||
f88dfff5 | 5750 | pr_info("Memory: %luK/%luK available " |
7ee3d4e8 | 5751 | "(%luK kernel code, %luK rwdata, %luK rodata, " |
e48322ab | 5752 | "%luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 JL |
5753 | #ifdef CONFIG_HIGHMEM |
5754 | ", %luK highmem" | |
5755 | #endif | |
5756 | "%s%s)\n", | |
5757 | nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10), | |
5758 | codesize >> 10, datasize >> 10, rosize >> 10, | |
5759 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
e48322ab PK |
5760 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10), |
5761 | totalcma_pages << (PAGE_SHIFT-10), | |
7ee3d4e8 JL |
5762 | #ifdef CONFIG_HIGHMEM |
5763 | totalhigh_pages << (PAGE_SHIFT-10), | |
5764 | #endif | |
5765 | str ? ", " : "", str ? str : ""); | |
5766 | } | |
5767 | ||
0e0b864e | 5768 | /** |
88ca3b94 RD |
5769 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
5770 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
5771 | * |
5772 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
5773 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
5774 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
5775 | * function may optionally be used to account for unfreeable pages in the |
5776 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
5777 | * smaller per-cpu batchsize. | |
0e0b864e MG |
5778 | */ |
5779 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
5780 | { | |
5781 | dma_reserve = new_dma_reserve; | |
5782 | } | |
5783 | ||
1da177e4 LT |
5784 | void __init free_area_init(unsigned long *zones_size) |
5785 | { | |
9109fb7b | 5786 | free_area_init_node(0, zones_size, |
1da177e4 LT |
5787 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
5788 | } | |
1da177e4 | 5789 | |
1da177e4 LT |
5790 | static int page_alloc_cpu_notify(struct notifier_block *self, |
5791 | unsigned long action, void *hcpu) | |
5792 | { | |
5793 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 5794 | |
8bb78442 | 5795 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 5796 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
5797 | drain_pages(cpu); |
5798 | ||
5799 | /* | |
5800 | * Spill the event counters of the dead processor | |
5801 | * into the current processors event counters. | |
5802 | * This artificially elevates the count of the current | |
5803 | * processor. | |
5804 | */ | |
f8891e5e | 5805 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
5806 | |
5807 | /* | |
5808 | * Zero the differential counters of the dead processor | |
5809 | * so that the vm statistics are consistent. | |
5810 | * | |
5811 | * This is only okay since the processor is dead and cannot | |
5812 | * race with what we are doing. | |
5813 | */ | |
2bb921e5 | 5814 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
5815 | } |
5816 | return NOTIFY_OK; | |
5817 | } | |
1da177e4 LT |
5818 | |
5819 | void __init page_alloc_init(void) | |
5820 | { | |
5821 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
5822 | } | |
5823 | ||
cb45b0e9 HA |
5824 | /* |
5825 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
5826 | * or min_free_kbytes changes. | |
5827 | */ | |
5828 | static void calculate_totalreserve_pages(void) | |
5829 | { | |
5830 | struct pglist_data *pgdat; | |
5831 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5832 | enum zone_type i, j; |
cb45b0e9 HA |
5833 | |
5834 | for_each_online_pgdat(pgdat) { | |
5835 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
5836 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 5837 | long max = 0; |
cb45b0e9 HA |
5838 | |
5839 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5840 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5841 | if (zone->lowmem_reserve[j] > max) | |
5842 | max = zone->lowmem_reserve[j]; | |
5843 | } | |
5844 | ||
41858966 MG |
5845 | /* we treat the high watermark as reserved pages. */ |
5846 | max += high_wmark_pages(zone); | |
cb45b0e9 | 5847 | |
b40da049 JL |
5848 | if (max > zone->managed_pages) |
5849 | max = zone->managed_pages; | |
cb45b0e9 | 5850 | reserve_pages += max; |
ab8fabd4 JW |
5851 | /* |
5852 | * Lowmem reserves are not available to | |
5853 | * GFP_HIGHUSER page cache allocations and | |
5854 | * kswapd tries to balance zones to their high | |
5855 | * watermark. As a result, neither should be | |
5856 | * regarded as dirtyable memory, to prevent a | |
5857 | * situation where reclaim has to clean pages | |
5858 | * in order to balance the zones. | |
5859 | */ | |
5860 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
5861 | } |
5862 | } | |
ab8fabd4 | 5863 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
5864 | totalreserve_pages = reserve_pages; |
5865 | } | |
5866 | ||
1da177e4 LT |
5867 | /* |
5868 | * setup_per_zone_lowmem_reserve - called whenever | |
5869 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
5870 | * has a correct pages reserved value, so an adequate number of | |
5871 | * pages are left in the zone after a successful __alloc_pages(). | |
5872 | */ | |
5873 | static void setup_per_zone_lowmem_reserve(void) | |
5874 | { | |
5875 | struct pglist_data *pgdat; | |
2f6726e5 | 5876 | enum zone_type j, idx; |
1da177e4 | 5877 | |
ec936fc5 | 5878 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
5879 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5880 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 5881 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
5882 | |
5883 | zone->lowmem_reserve[j] = 0; | |
5884 | ||
2f6726e5 CL |
5885 | idx = j; |
5886 | while (idx) { | |
1da177e4 LT |
5887 | struct zone *lower_zone; |
5888 | ||
2f6726e5 CL |
5889 | idx--; |
5890 | ||
1da177e4 LT |
5891 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
5892 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
5893 | ||
5894 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 5895 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 5896 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 5897 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
5898 | } |
5899 | } | |
5900 | } | |
cb45b0e9 HA |
5901 | |
5902 | /* update totalreserve_pages */ | |
5903 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5904 | } |
5905 | ||
cfd3da1e | 5906 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5907 | { |
5908 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5909 | unsigned long lowmem_pages = 0; | |
5910 | struct zone *zone; | |
5911 | unsigned long flags; | |
5912 | ||
5913 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5914 | for_each_zone(zone) { | |
5915 | if (!is_highmem(zone)) | |
b40da049 | 5916 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
5917 | } |
5918 | ||
5919 | for_each_zone(zone) { | |
ac924c60 AM |
5920 | u64 tmp; |
5921 | ||
1125b4e3 | 5922 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 5923 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 5924 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
5925 | if (is_highmem(zone)) { |
5926 | /* | |
669ed175 NP |
5927 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5928 | * need highmem pages, so cap pages_min to a small | |
5929 | * value here. | |
5930 | * | |
41858966 | 5931 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 5932 | * deltas control asynch page reclaim, and so should |
669ed175 | 5933 | * not be capped for highmem. |
1da177e4 | 5934 | */ |
90ae8d67 | 5935 | unsigned long min_pages; |
1da177e4 | 5936 | |
b40da049 | 5937 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 5938 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 5939 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5940 | } else { |
669ed175 NP |
5941 | /* |
5942 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5943 | * proportionate to the zone's size. |
5944 | */ | |
41858966 | 5945 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5946 | } |
5947 | ||
41858966 MG |
5948 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
5949 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
49f223a9 | 5950 | |
81c0a2bb | 5951 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
abe5f972 JW |
5952 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
5953 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
81c0a2bb | 5954 | |
56fd56b8 | 5955 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 5956 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5957 | } |
cb45b0e9 HA |
5958 | |
5959 | /* update totalreserve_pages */ | |
5960 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5961 | } |
5962 | ||
cfd3da1e MG |
5963 | /** |
5964 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5965 | * or when memory is hot-{added|removed} | |
5966 | * | |
5967 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5968 | * correctly with respect to min_free_kbytes. | |
5969 | */ | |
5970 | void setup_per_zone_wmarks(void) | |
5971 | { | |
5972 | mutex_lock(&zonelists_mutex); | |
5973 | __setup_per_zone_wmarks(); | |
5974 | mutex_unlock(&zonelists_mutex); | |
5975 | } | |
5976 | ||
55a4462a | 5977 | /* |
556adecb RR |
5978 | * The inactive anon list should be small enough that the VM never has to |
5979 | * do too much work, but large enough that each inactive page has a chance | |
5980 | * to be referenced again before it is swapped out. | |
5981 | * | |
5982 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
5983 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
5984 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
5985 | * the anonymous pages are kept on the inactive list. | |
5986 | * | |
5987 | * total target max | |
5988 | * memory ratio inactive anon | |
5989 | * ------------------------------------- | |
5990 | * 10MB 1 5MB | |
5991 | * 100MB 1 50MB | |
5992 | * 1GB 3 250MB | |
5993 | * 10GB 10 0.9GB | |
5994 | * 100GB 31 3GB | |
5995 | * 1TB 101 10GB | |
5996 | * 10TB 320 32GB | |
5997 | */ | |
1b79acc9 | 5998 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5999 | { |
96cb4df5 | 6000 | unsigned int gb, ratio; |
556adecb | 6001 | |
96cb4df5 | 6002 | /* Zone size in gigabytes */ |
b40da049 | 6003 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 6004 | if (gb) |
556adecb | 6005 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
6006 | else |
6007 | ratio = 1; | |
556adecb | 6008 | |
96cb4df5 MK |
6009 | zone->inactive_ratio = ratio; |
6010 | } | |
556adecb | 6011 | |
839a4fcc | 6012 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
6013 | { |
6014 | struct zone *zone; | |
6015 | ||
6016 | for_each_zone(zone) | |
6017 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
6018 | } |
6019 | ||
1da177e4 LT |
6020 | /* |
6021 | * Initialise min_free_kbytes. | |
6022 | * | |
6023 | * For small machines we want it small (128k min). For large machines | |
6024 | * we want it large (64MB max). But it is not linear, because network | |
6025 | * bandwidth does not increase linearly with machine size. We use | |
6026 | * | |
b8af2941 | 6027 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6028 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6029 | * | |
6030 | * which yields | |
6031 | * | |
6032 | * 16MB: 512k | |
6033 | * 32MB: 724k | |
6034 | * 64MB: 1024k | |
6035 | * 128MB: 1448k | |
6036 | * 256MB: 2048k | |
6037 | * 512MB: 2896k | |
6038 | * 1024MB: 4096k | |
6039 | * 2048MB: 5792k | |
6040 | * 4096MB: 8192k | |
6041 | * 8192MB: 11584k | |
6042 | * 16384MB: 16384k | |
6043 | */ | |
1b79acc9 | 6044 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6045 | { |
6046 | unsigned long lowmem_kbytes; | |
5f12733e | 6047 | int new_min_free_kbytes; |
1da177e4 LT |
6048 | |
6049 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6050 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6051 | ||
6052 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6053 | min_free_kbytes = new_min_free_kbytes; | |
6054 | if (min_free_kbytes < 128) | |
6055 | min_free_kbytes = 128; | |
6056 | if (min_free_kbytes > 65536) | |
6057 | min_free_kbytes = 65536; | |
6058 | } else { | |
6059 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6060 | new_min_free_kbytes, user_min_free_kbytes); | |
6061 | } | |
bc75d33f | 6062 | setup_per_zone_wmarks(); |
a6cccdc3 | 6063 | refresh_zone_stat_thresholds(); |
1da177e4 | 6064 | setup_per_zone_lowmem_reserve(); |
556adecb | 6065 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
6066 | return 0; |
6067 | } | |
bc75d33f | 6068 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
6069 | |
6070 | /* | |
b8af2941 | 6071 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6072 | * that we can call two helper functions whenever min_free_kbytes |
6073 | * changes. | |
6074 | */ | |
cccad5b9 | 6075 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6076 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6077 | { |
da8c757b HP |
6078 | int rc; |
6079 | ||
6080 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6081 | if (rc) | |
6082 | return rc; | |
6083 | ||
5f12733e MH |
6084 | if (write) { |
6085 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6086 | setup_per_zone_wmarks(); |
5f12733e | 6087 | } |
1da177e4 LT |
6088 | return 0; |
6089 | } | |
6090 | ||
9614634f | 6091 | #ifdef CONFIG_NUMA |
cccad5b9 | 6092 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6093 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
6094 | { |
6095 | struct zone *zone; | |
6096 | int rc; | |
6097 | ||
8d65af78 | 6098 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
6099 | if (rc) |
6100 | return rc; | |
6101 | ||
6102 | for_each_zone(zone) | |
b40da049 | 6103 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
6104 | sysctl_min_unmapped_ratio) / 100; |
6105 | return 0; | |
6106 | } | |
0ff38490 | 6107 | |
cccad5b9 | 6108 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6109 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
6110 | { |
6111 | struct zone *zone; | |
6112 | int rc; | |
6113 | ||
8d65af78 | 6114 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
6115 | if (rc) |
6116 | return rc; | |
6117 | ||
6118 | for_each_zone(zone) | |
b40da049 | 6119 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
6120 | sysctl_min_slab_ratio) / 100; |
6121 | return 0; | |
6122 | } | |
9614634f CL |
6123 | #endif |
6124 | ||
1da177e4 LT |
6125 | /* |
6126 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
6127 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
6128 | * whenever sysctl_lowmem_reserve_ratio changes. | |
6129 | * | |
6130 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 6131 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
6132 | * if in function of the boot time zone sizes. |
6133 | */ | |
cccad5b9 | 6134 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6135 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6136 | { |
8d65af78 | 6137 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
6138 | setup_per_zone_lowmem_reserve(); |
6139 | return 0; | |
6140 | } | |
6141 | ||
8ad4b1fb RS |
6142 | /* |
6143 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
6144 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
6145 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 6146 | */ |
cccad5b9 | 6147 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6148 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
6149 | { |
6150 | struct zone *zone; | |
7cd2b0a3 | 6151 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
6152 | int ret; |
6153 | ||
7cd2b0a3 DR |
6154 | mutex_lock(&pcp_batch_high_lock); |
6155 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
6156 | ||
8d65af78 | 6157 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
6158 | if (!write || ret < 0) |
6159 | goto out; | |
6160 | ||
6161 | /* Sanity checking to avoid pcp imbalance */ | |
6162 | if (percpu_pagelist_fraction && | |
6163 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
6164 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
6165 | ret = -EINVAL; | |
6166 | goto out; | |
6167 | } | |
6168 | ||
6169 | /* No change? */ | |
6170 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
6171 | goto out; | |
c8e251fa | 6172 | |
364df0eb | 6173 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
6174 | unsigned int cpu; |
6175 | ||
22a7f12b | 6176 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
6177 | pageset_set_high_and_batch(zone, |
6178 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 6179 | } |
7cd2b0a3 | 6180 | out: |
c8e251fa | 6181 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6182 | return ret; |
8ad4b1fb RS |
6183 | } |
6184 | ||
a9919c79 | 6185 | #ifdef CONFIG_NUMA |
f034b5d4 | 6186 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 6187 | |
1da177e4 LT |
6188 | static int __init set_hashdist(char *str) |
6189 | { | |
6190 | if (!str) | |
6191 | return 0; | |
6192 | hashdist = simple_strtoul(str, &str, 0); | |
6193 | return 1; | |
6194 | } | |
6195 | __setup("hashdist=", set_hashdist); | |
6196 | #endif | |
6197 | ||
6198 | /* | |
6199 | * allocate a large system hash table from bootmem | |
6200 | * - it is assumed that the hash table must contain an exact power-of-2 | |
6201 | * quantity of entries | |
6202 | * - limit is the number of hash buckets, not the total allocation size | |
6203 | */ | |
6204 | void *__init alloc_large_system_hash(const char *tablename, | |
6205 | unsigned long bucketsize, | |
6206 | unsigned long numentries, | |
6207 | int scale, | |
6208 | int flags, | |
6209 | unsigned int *_hash_shift, | |
6210 | unsigned int *_hash_mask, | |
31fe62b9 TB |
6211 | unsigned long low_limit, |
6212 | unsigned long high_limit) | |
1da177e4 | 6213 | { |
31fe62b9 | 6214 | unsigned long long max = high_limit; |
1da177e4 LT |
6215 | unsigned long log2qty, size; |
6216 | void *table = NULL; | |
6217 | ||
6218 | /* allow the kernel cmdline to have a say */ | |
6219 | if (!numentries) { | |
6220 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 6221 | numentries = nr_kernel_pages; |
a7e83318 JZ |
6222 | |
6223 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
6224 | if (PAGE_SHIFT < 20) | |
6225 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
6226 | |
6227 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
6228 | if (scale > PAGE_SHIFT) | |
6229 | numentries >>= (scale - PAGE_SHIFT); | |
6230 | else | |
6231 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
6232 | |
6233 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
6234 | if (unlikely(flags & HASH_SMALL)) { |
6235 | /* Makes no sense without HASH_EARLY */ | |
6236 | WARN_ON(!(flags & HASH_EARLY)); | |
6237 | if (!(numentries >> *_hash_shift)) { | |
6238 | numentries = 1UL << *_hash_shift; | |
6239 | BUG_ON(!numentries); | |
6240 | } | |
6241 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 6242 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 6243 | } |
6e692ed3 | 6244 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
6245 | |
6246 | /* limit allocation size to 1/16 total memory by default */ | |
6247 | if (max == 0) { | |
6248 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
6249 | do_div(max, bucketsize); | |
6250 | } | |
074b8517 | 6251 | max = min(max, 0x80000000ULL); |
1da177e4 | 6252 | |
31fe62b9 TB |
6253 | if (numentries < low_limit) |
6254 | numentries = low_limit; | |
1da177e4 LT |
6255 | if (numentries > max) |
6256 | numentries = max; | |
6257 | ||
f0d1b0b3 | 6258 | log2qty = ilog2(numentries); |
1da177e4 LT |
6259 | |
6260 | do { | |
6261 | size = bucketsize << log2qty; | |
6262 | if (flags & HASH_EARLY) | |
6782832e | 6263 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
6264 | else if (hashdist) |
6265 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
6266 | else { | |
1037b83b ED |
6267 | /* |
6268 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
6269 | * some pages at the end of hash table which |
6270 | * alloc_pages_exact() automatically does | |
1037b83b | 6271 | */ |
264ef8a9 | 6272 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 6273 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
6274 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
6275 | } | |
1da177e4 LT |
6276 | } |
6277 | } while (!table && size > PAGE_SIZE && --log2qty); | |
6278 | ||
6279 | if (!table) | |
6280 | panic("Failed to allocate %s hash table\n", tablename); | |
6281 | ||
f241e660 | 6282 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 6283 | tablename, |
f241e660 | 6284 | (1UL << log2qty), |
f0d1b0b3 | 6285 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
6286 | size); |
6287 | ||
6288 | if (_hash_shift) | |
6289 | *_hash_shift = log2qty; | |
6290 | if (_hash_mask) | |
6291 | *_hash_mask = (1 << log2qty) - 1; | |
6292 | ||
6293 | return table; | |
6294 | } | |
a117e66e | 6295 | |
835c134e MG |
6296 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
6297 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
6298 | unsigned long pfn) | |
6299 | { | |
6300 | #ifdef CONFIG_SPARSEMEM | |
6301 | return __pfn_to_section(pfn)->pageblock_flags; | |
6302 | #else | |
6303 | return zone->pageblock_flags; | |
6304 | #endif /* CONFIG_SPARSEMEM */ | |
6305 | } | |
6306 | ||
6307 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
6308 | { | |
6309 | #ifdef CONFIG_SPARSEMEM | |
6310 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 6311 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 6312 | #else |
c060f943 | 6313 | pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 6314 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
6315 | #endif /* CONFIG_SPARSEMEM */ |
6316 | } | |
6317 | ||
6318 | /** | |
1aab4d77 | 6319 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e | 6320 | * @page: The page within the block of interest |
1aab4d77 RD |
6321 | * @pfn: The target page frame number |
6322 | * @end_bitidx: The last bit of interest to retrieve | |
6323 | * @mask: mask of bits that the caller is interested in | |
6324 | * | |
6325 | * Return: pageblock_bits flags | |
835c134e | 6326 | */ |
dc4b0caf | 6327 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
e58469ba MG |
6328 | unsigned long end_bitidx, |
6329 | unsigned long mask) | |
835c134e MG |
6330 | { |
6331 | struct zone *zone; | |
6332 | unsigned long *bitmap; | |
dc4b0caf | 6333 | unsigned long bitidx, word_bitidx; |
e58469ba | 6334 | unsigned long word; |
835c134e MG |
6335 | |
6336 | zone = page_zone(page); | |
835c134e MG |
6337 | bitmap = get_pageblock_bitmap(zone, pfn); |
6338 | bitidx = pfn_to_bitidx(zone, pfn); | |
e58469ba MG |
6339 | word_bitidx = bitidx / BITS_PER_LONG; |
6340 | bitidx &= (BITS_PER_LONG-1); | |
835c134e | 6341 | |
e58469ba MG |
6342 | word = bitmap[word_bitidx]; |
6343 | bitidx += end_bitidx; | |
6344 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
835c134e MG |
6345 | } |
6346 | ||
6347 | /** | |
dc4b0caf | 6348 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e | 6349 | * @page: The page within the block of interest |
835c134e | 6350 | * @flags: The flags to set |
1aab4d77 RD |
6351 | * @pfn: The target page frame number |
6352 | * @end_bitidx: The last bit of interest | |
6353 | * @mask: mask of bits that the caller is interested in | |
835c134e | 6354 | */ |
dc4b0caf MG |
6355 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
6356 | unsigned long pfn, | |
e58469ba MG |
6357 | unsigned long end_bitidx, |
6358 | unsigned long mask) | |
835c134e MG |
6359 | { |
6360 | struct zone *zone; | |
6361 | unsigned long *bitmap; | |
dc4b0caf | 6362 | unsigned long bitidx, word_bitidx; |
e58469ba MG |
6363 | unsigned long old_word, word; |
6364 | ||
6365 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
835c134e MG |
6366 | |
6367 | zone = page_zone(page); | |
835c134e MG |
6368 | bitmap = get_pageblock_bitmap(zone, pfn); |
6369 | bitidx = pfn_to_bitidx(zone, pfn); | |
e58469ba MG |
6370 | word_bitidx = bitidx / BITS_PER_LONG; |
6371 | bitidx &= (BITS_PER_LONG-1); | |
6372 | ||
309381fe | 6373 | VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page); |
835c134e | 6374 | |
e58469ba MG |
6375 | bitidx += end_bitidx; |
6376 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
6377 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
6378 | ||
4db0c3c2 | 6379 | word = READ_ONCE(bitmap[word_bitidx]); |
e58469ba MG |
6380 | for (;;) { |
6381 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
6382 | if (word == old_word) | |
6383 | break; | |
6384 | word = old_word; | |
6385 | } | |
835c134e | 6386 | } |
a5d76b54 KH |
6387 | |
6388 | /* | |
80934513 MK |
6389 | * This function checks whether pageblock includes unmovable pages or not. |
6390 | * If @count is not zero, it is okay to include less @count unmovable pages | |
6391 | * | |
b8af2941 | 6392 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
6393 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
6394 | * expect this function should be exact. | |
a5d76b54 | 6395 | */ |
b023f468 WC |
6396 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
6397 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
6398 | { |
6399 | unsigned long pfn, iter, found; | |
47118af0 MN |
6400 | int mt; |
6401 | ||
49ac8255 KH |
6402 | /* |
6403 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 6404 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
6405 | */ |
6406 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 6407 | return false; |
47118af0 MN |
6408 | mt = get_pageblock_migratetype(page); |
6409 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 6410 | return false; |
49ac8255 KH |
6411 | |
6412 | pfn = page_to_pfn(page); | |
6413 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
6414 | unsigned long check = pfn + iter; | |
6415 | ||
29723fcc | 6416 | if (!pfn_valid_within(check)) |
49ac8255 | 6417 | continue; |
29723fcc | 6418 | |
49ac8255 | 6419 | page = pfn_to_page(check); |
c8721bbb NH |
6420 | |
6421 | /* | |
6422 | * Hugepages are not in LRU lists, but they're movable. | |
6423 | * We need not scan over tail pages bacause we don't | |
6424 | * handle each tail page individually in migration. | |
6425 | */ | |
6426 | if (PageHuge(page)) { | |
6427 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
6428 | continue; | |
6429 | } | |
6430 | ||
97d255c8 MK |
6431 | /* |
6432 | * We can't use page_count without pin a page | |
6433 | * because another CPU can free compound page. | |
6434 | * This check already skips compound tails of THP | |
6435 | * because their page->_count is zero at all time. | |
6436 | */ | |
6437 | if (!atomic_read(&page->_count)) { | |
49ac8255 KH |
6438 | if (PageBuddy(page)) |
6439 | iter += (1 << page_order(page)) - 1; | |
6440 | continue; | |
6441 | } | |
97d255c8 | 6442 | |
b023f468 WC |
6443 | /* |
6444 | * The HWPoisoned page may be not in buddy system, and | |
6445 | * page_count() is not 0. | |
6446 | */ | |
6447 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6448 | continue; | |
6449 | ||
49ac8255 KH |
6450 | if (!PageLRU(page)) |
6451 | found++; | |
6452 | /* | |
6b4f7799 JW |
6453 | * If there are RECLAIMABLE pages, we need to check |
6454 | * it. But now, memory offline itself doesn't call | |
6455 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
6456 | */ |
6457 | /* | |
6458 | * If the page is not RAM, page_count()should be 0. | |
6459 | * we don't need more check. This is an _used_ not-movable page. | |
6460 | * | |
6461 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6462 | * is set to both of a memory hole page and a _used_ kernel | |
6463 | * page at boot. | |
6464 | */ | |
6465 | if (found > count) | |
80934513 | 6466 | return true; |
49ac8255 | 6467 | } |
80934513 | 6468 | return false; |
49ac8255 KH |
6469 | } |
6470 | ||
6471 | bool is_pageblock_removable_nolock(struct page *page) | |
6472 | { | |
656a0706 MH |
6473 | struct zone *zone; |
6474 | unsigned long pfn; | |
687875fb MH |
6475 | |
6476 | /* | |
6477 | * We have to be careful here because we are iterating over memory | |
6478 | * sections which are not zone aware so we might end up outside of | |
6479 | * the zone but still within the section. | |
656a0706 MH |
6480 | * We have to take care about the node as well. If the node is offline |
6481 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 6482 | */ |
656a0706 MH |
6483 | if (!node_online(page_to_nid(page))) |
6484 | return false; | |
6485 | ||
6486 | zone = page_zone(page); | |
6487 | pfn = page_to_pfn(page); | |
108bcc96 | 6488 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
6489 | return false; |
6490 | ||
b023f468 | 6491 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6492 | } |
0c0e6195 | 6493 | |
041d3a8c MN |
6494 | #ifdef CONFIG_CMA |
6495 | ||
6496 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6497 | { | |
6498 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6499 | pageblock_nr_pages) - 1); | |
6500 | } | |
6501 | ||
6502 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6503 | { | |
6504 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6505 | pageblock_nr_pages)); | |
6506 | } | |
6507 | ||
041d3a8c | 6508 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6509 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6510 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6511 | { |
6512 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6513 | unsigned long nr_reclaimed; |
041d3a8c MN |
6514 | unsigned long pfn = start; |
6515 | unsigned int tries = 0; | |
6516 | int ret = 0; | |
6517 | ||
be49a6e1 | 6518 | migrate_prep(); |
041d3a8c | 6519 | |
bb13ffeb | 6520 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6521 | if (fatal_signal_pending(current)) { |
6522 | ret = -EINTR; | |
6523 | break; | |
6524 | } | |
6525 | ||
bb13ffeb MG |
6526 | if (list_empty(&cc->migratepages)) { |
6527 | cc->nr_migratepages = 0; | |
edc2ca61 | 6528 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
6529 | if (!pfn) { |
6530 | ret = -EINTR; | |
6531 | break; | |
6532 | } | |
6533 | tries = 0; | |
6534 | } else if (++tries == 5) { | |
6535 | ret = ret < 0 ? ret : -EBUSY; | |
6536 | break; | |
6537 | } | |
6538 | ||
beb51eaa MK |
6539 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6540 | &cc->migratepages); | |
6541 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6542 | |
9c620e2b | 6543 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 6544 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 6545 | } |
2a6f5124 SP |
6546 | if (ret < 0) { |
6547 | putback_movable_pages(&cc->migratepages); | |
6548 | return ret; | |
6549 | } | |
6550 | return 0; | |
041d3a8c MN |
6551 | } |
6552 | ||
6553 | /** | |
6554 | * alloc_contig_range() -- tries to allocate given range of pages | |
6555 | * @start: start PFN to allocate | |
6556 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
6557 | * @migratetype: migratetype of the underlaying pageblocks (either |
6558 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
6559 | * in range must have the same migratetype and it must | |
6560 | * be either of the two. | |
041d3a8c MN |
6561 | * |
6562 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
6563 | * aligned, however it's the caller's responsibility to guarantee that | |
6564 | * we are the only thread that changes migrate type of pageblocks the | |
6565 | * pages fall in. | |
6566 | * | |
6567 | * The PFN range must belong to a single zone. | |
6568 | * | |
6569 | * Returns zero on success or negative error code. On success all | |
6570 | * pages which PFN is in [start, end) are allocated for the caller and | |
6571 | * need to be freed with free_contig_range(). | |
6572 | */ | |
0815f3d8 MN |
6573 | int alloc_contig_range(unsigned long start, unsigned long end, |
6574 | unsigned migratetype) | |
041d3a8c | 6575 | { |
041d3a8c MN |
6576 | unsigned long outer_start, outer_end; |
6577 | int ret = 0, order; | |
6578 | ||
bb13ffeb MG |
6579 | struct compact_control cc = { |
6580 | .nr_migratepages = 0, | |
6581 | .order = -1, | |
6582 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 6583 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
6584 | .ignore_skip_hint = true, |
6585 | }; | |
6586 | INIT_LIST_HEAD(&cc.migratepages); | |
6587 | ||
041d3a8c MN |
6588 | /* |
6589 | * What we do here is we mark all pageblocks in range as | |
6590 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6591 | * have different sizes, and due to the way page allocator | |
6592 | * work, we align the range to biggest of the two pages so | |
6593 | * that page allocator won't try to merge buddies from | |
6594 | * different pageblocks and change MIGRATE_ISOLATE to some | |
6595 | * other migration type. | |
6596 | * | |
6597 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6598 | * migrate the pages from an unaligned range (ie. pages that | |
6599 | * we are interested in). This will put all the pages in | |
6600 | * range back to page allocator as MIGRATE_ISOLATE. | |
6601 | * | |
6602 | * When this is done, we take the pages in range from page | |
6603 | * allocator removing them from the buddy system. This way | |
6604 | * page allocator will never consider using them. | |
6605 | * | |
6606 | * This lets us mark the pageblocks back as | |
6607 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6608 | * aligned range but not in the unaligned, original range are | |
6609 | * put back to page allocator so that buddy can use them. | |
6610 | */ | |
6611 | ||
6612 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
6613 | pfn_max_align_up(end), migratetype, |
6614 | false); | |
041d3a8c | 6615 | if (ret) |
86a595f9 | 6616 | return ret; |
041d3a8c | 6617 | |
bb13ffeb | 6618 | ret = __alloc_contig_migrate_range(&cc, start, end); |
041d3a8c MN |
6619 | if (ret) |
6620 | goto done; | |
6621 | ||
6622 | /* | |
6623 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
6624 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
6625 | * more, all pages in [start, end) are free in page allocator. | |
6626 | * What we are going to do is to allocate all pages from | |
6627 | * [start, end) (that is remove them from page allocator). | |
6628 | * | |
6629 | * The only problem is that pages at the beginning and at the | |
6630 | * end of interesting range may be not aligned with pages that | |
6631 | * page allocator holds, ie. they can be part of higher order | |
6632 | * pages. Because of this, we reserve the bigger range and | |
6633 | * once this is done free the pages we are not interested in. | |
6634 | * | |
6635 | * We don't have to hold zone->lock here because the pages are | |
6636 | * isolated thus they won't get removed from buddy. | |
6637 | */ | |
6638 | ||
6639 | lru_add_drain_all(); | |
510f5507 | 6640 | drain_all_pages(cc.zone); |
041d3a8c MN |
6641 | |
6642 | order = 0; | |
6643 | outer_start = start; | |
6644 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
6645 | if (++order >= MAX_ORDER) { | |
6646 | ret = -EBUSY; | |
6647 | goto done; | |
6648 | } | |
6649 | outer_start &= ~0UL << order; | |
6650 | } | |
6651 | ||
6652 | /* Make sure the range is really isolated. */ | |
b023f468 | 6653 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
6654 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
6655 | __func__, outer_start, end); | |
041d3a8c MN |
6656 | ret = -EBUSY; |
6657 | goto done; | |
6658 | } | |
6659 | ||
49f223a9 | 6660 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 6661 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6662 | if (!outer_end) { |
6663 | ret = -EBUSY; | |
6664 | goto done; | |
6665 | } | |
6666 | ||
6667 | /* Free head and tail (if any) */ | |
6668 | if (start != outer_start) | |
6669 | free_contig_range(outer_start, start - outer_start); | |
6670 | if (end != outer_end) | |
6671 | free_contig_range(end, outer_end - end); | |
6672 | ||
6673 | done: | |
6674 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 6675 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
6676 | return ret; |
6677 | } | |
6678 | ||
6679 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
6680 | { | |
bcc2b02f MS |
6681 | unsigned int count = 0; |
6682 | ||
6683 | for (; nr_pages--; pfn++) { | |
6684 | struct page *page = pfn_to_page(pfn); | |
6685 | ||
6686 | count += page_count(page) != 1; | |
6687 | __free_page(page); | |
6688 | } | |
6689 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
6690 | } |
6691 | #endif | |
6692 | ||
4ed7e022 | 6693 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
6694 | /* |
6695 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
6696 | * page high values need to be recalulated. | |
6697 | */ | |
4ed7e022 JL |
6698 | void __meminit zone_pcp_update(struct zone *zone) |
6699 | { | |
0a647f38 | 6700 | unsigned cpu; |
c8e251fa | 6701 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 6702 | for_each_possible_cpu(cpu) |
169f6c19 CS |
6703 | pageset_set_high_and_batch(zone, |
6704 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 6705 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
6706 | } |
6707 | #endif | |
6708 | ||
340175b7 JL |
6709 | void zone_pcp_reset(struct zone *zone) |
6710 | { | |
6711 | unsigned long flags; | |
5a883813 MK |
6712 | int cpu; |
6713 | struct per_cpu_pageset *pset; | |
340175b7 JL |
6714 | |
6715 | /* avoid races with drain_pages() */ | |
6716 | local_irq_save(flags); | |
6717 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
6718 | for_each_online_cpu(cpu) { |
6719 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6720 | drain_zonestat(zone, pset); | |
6721 | } | |
340175b7 JL |
6722 | free_percpu(zone->pageset); |
6723 | zone->pageset = &boot_pageset; | |
6724 | } | |
6725 | local_irq_restore(flags); | |
6726 | } | |
6727 | ||
6dcd73d7 | 6728 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
6729 | /* |
6730 | * All pages in the range must be isolated before calling this. | |
6731 | */ | |
6732 | void | |
6733 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
6734 | { | |
6735 | struct page *page; | |
6736 | struct zone *zone; | |
7aeb09f9 | 6737 | unsigned int order, i; |
0c0e6195 KH |
6738 | unsigned long pfn; |
6739 | unsigned long flags; | |
6740 | /* find the first valid pfn */ | |
6741 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
6742 | if (pfn_valid(pfn)) | |
6743 | break; | |
6744 | if (pfn == end_pfn) | |
6745 | return; | |
6746 | zone = page_zone(pfn_to_page(pfn)); | |
6747 | spin_lock_irqsave(&zone->lock, flags); | |
6748 | pfn = start_pfn; | |
6749 | while (pfn < end_pfn) { | |
6750 | if (!pfn_valid(pfn)) { | |
6751 | pfn++; | |
6752 | continue; | |
6753 | } | |
6754 | page = pfn_to_page(pfn); | |
b023f468 WC |
6755 | /* |
6756 | * The HWPoisoned page may be not in buddy system, and | |
6757 | * page_count() is not 0. | |
6758 | */ | |
6759 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6760 | pfn++; | |
6761 | SetPageReserved(page); | |
6762 | continue; | |
6763 | } | |
6764 | ||
0c0e6195 KH |
6765 | BUG_ON(page_count(page)); |
6766 | BUG_ON(!PageBuddy(page)); | |
6767 | order = page_order(page); | |
6768 | #ifdef CONFIG_DEBUG_VM | |
6769 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
6770 | pfn, 1 << order, end_pfn); | |
6771 | #endif | |
6772 | list_del(&page->lru); | |
6773 | rmv_page_order(page); | |
6774 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
6775 | for (i = 0; i < (1 << order); i++) |
6776 | SetPageReserved((page+i)); | |
6777 | pfn += (1 << order); | |
6778 | } | |
6779 | spin_unlock_irqrestore(&zone->lock, flags); | |
6780 | } | |
6781 | #endif | |
8d22ba1b WF |
6782 | |
6783 | #ifdef CONFIG_MEMORY_FAILURE | |
6784 | bool is_free_buddy_page(struct page *page) | |
6785 | { | |
6786 | struct zone *zone = page_zone(page); | |
6787 | unsigned long pfn = page_to_pfn(page); | |
6788 | unsigned long flags; | |
7aeb09f9 | 6789 | unsigned int order; |
8d22ba1b WF |
6790 | |
6791 | spin_lock_irqsave(&zone->lock, flags); | |
6792 | for (order = 0; order < MAX_ORDER; order++) { | |
6793 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
6794 | ||
6795 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
6796 | break; | |
6797 | } | |
6798 | spin_unlock_irqrestore(&zone->lock, flags); | |
6799 | ||
6800 | return order < MAX_ORDER; | |
6801 | } | |
6802 | #endif |