]>
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> |
b8c73fc2 | 27 | #include <linux/kasan.h> |
1da177e4 LT |
28 | #include <linux/module.h> |
29 | #include <linux/suspend.h> | |
30 | #include <linux/pagevec.h> | |
31 | #include <linux/blkdev.h> | |
32 | #include <linux/slab.h> | |
a238ab5b | 33 | #include <linux/ratelimit.h> |
5a3135c2 | 34 | #include <linux/oom.h> |
1da177e4 LT |
35 | #include <linux/notifier.h> |
36 | #include <linux/topology.h> | |
37 | #include <linux/sysctl.h> | |
38 | #include <linux/cpu.h> | |
39 | #include <linux/cpuset.h> | |
bdc8cb98 | 40 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
41 | #include <linux/nodemask.h> |
42 | #include <linux/vmalloc.h> | |
a6cccdc3 | 43 | #include <linux/vmstat.h> |
4be38e35 | 44 | #include <linux/mempolicy.h> |
4b94ffdc | 45 | #include <linux/memremap.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> |
d379f01d | 57 | #include <trace/events/oom.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
041d3a8c | 60 | #include <linux/migrate.h> |
949f7ec5 | 61 | #include <linux/hugetlb.h> |
8bd75c77 | 62 | #include <linux/sched/rt.h> |
5b3cc15a | 63 | #include <linux/sched/mm.h> |
48c96a36 | 64 | #include <linux/page_owner.h> |
0e1cc95b | 65 | #include <linux/kthread.h> |
4949148a | 66 | #include <linux/memcontrol.h> |
42c269c8 | 67 | #include <linux/ftrace.h> |
d92a8cfc | 68 | #include <linux/lockdep.h> |
556b969a | 69 | #include <linux/nmi.h> |
1da177e4 | 70 | |
7ee3d4e8 | 71 | #include <asm/sections.h> |
1da177e4 | 72 | #include <asm/tlbflush.h> |
ac924c60 | 73 | #include <asm/div64.h> |
1da177e4 LT |
74 | #include "internal.h" |
75 | ||
c8e251fa CS |
76 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
77 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 78 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 79 | |
72812019 LS |
80 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
81 | DEFINE_PER_CPU(int, numa_node); | |
82 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
83 | #endif | |
84 | ||
4518085e KW |
85 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
86 | ||
7aac7898 LS |
87 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
88 | /* | |
89 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
90 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
91 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
92 | * defined in <linux/topology.h>. | |
93 | */ | |
94 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
95 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 96 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
97 | #endif |
98 | ||
bd233f53 MG |
99 | /* work_structs for global per-cpu drains */ |
100 | DEFINE_MUTEX(pcpu_drain_mutex); | |
101 | DEFINE_PER_CPU(struct work_struct, pcpu_drain); | |
102 | ||
38addce8 | 103 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 104 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
105 | EXPORT_SYMBOL(latent_entropy); |
106 | #endif | |
107 | ||
1da177e4 | 108 | /* |
13808910 | 109 | * Array of node states. |
1da177e4 | 110 | */ |
13808910 CL |
111 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
112 | [N_POSSIBLE] = NODE_MASK_ALL, | |
113 | [N_ONLINE] = { { [0] = 1UL } }, | |
114 | #ifndef CONFIG_NUMA | |
115 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
116 | #ifdef CONFIG_HIGHMEM | |
117 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 118 | #endif |
20b2f52b | 119 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
120 | [N_CPU] = { { [0] = 1UL } }, |
121 | #endif /* NUMA */ | |
122 | }; | |
123 | EXPORT_SYMBOL(node_states); | |
124 | ||
c3d5f5f0 JL |
125 | /* Protect totalram_pages and zone->managed_pages */ |
126 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
127 | ||
6c231b7b | 128 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 129 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 130 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 131 | |
1b76b02f | 132 | int percpu_pagelist_fraction; |
dcce284a | 133 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 134 | |
bb14c2c7 VB |
135 | /* |
136 | * A cached value of the page's pageblock's migratetype, used when the page is | |
137 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
138 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
139 | * Also the migratetype set in the page does not necessarily match the pcplist | |
140 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
141 | * other index - this ensures that it will be put on the correct CMA freelist. | |
142 | */ | |
143 | static inline int get_pcppage_migratetype(struct page *page) | |
144 | { | |
145 | return page->index; | |
146 | } | |
147 | ||
148 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
149 | { | |
150 | page->index = migratetype; | |
151 | } | |
152 | ||
452aa699 RW |
153 | #ifdef CONFIG_PM_SLEEP |
154 | /* | |
155 | * The following functions are used by the suspend/hibernate code to temporarily | |
156 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
157 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
158 | * they should always be called with system_transition_mutex held |
159 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
160 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
161 | * with that modification). | |
452aa699 | 162 | */ |
c9e664f1 RW |
163 | |
164 | static gfp_t saved_gfp_mask; | |
165 | ||
166 | void pm_restore_gfp_mask(void) | |
452aa699 | 167 | { |
55f2503c | 168 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
169 | if (saved_gfp_mask) { |
170 | gfp_allowed_mask = saved_gfp_mask; | |
171 | saved_gfp_mask = 0; | |
172 | } | |
452aa699 RW |
173 | } |
174 | ||
c9e664f1 | 175 | void pm_restrict_gfp_mask(void) |
452aa699 | 176 | { |
55f2503c | 177 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
178 | WARN_ON(saved_gfp_mask); |
179 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 180 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 181 | } |
f90ac398 MG |
182 | |
183 | bool pm_suspended_storage(void) | |
184 | { | |
d0164adc | 185 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
186 | return false; |
187 | return true; | |
188 | } | |
452aa699 RW |
189 | #endif /* CONFIG_PM_SLEEP */ |
190 | ||
d9c23400 | 191 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 192 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
193 | #endif |
194 | ||
d98c7a09 | 195 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 196 | |
1da177e4 LT |
197 | /* |
198 | * results with 256, 32 in the lowmem_reserve sysctl: | |
199 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
200 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
201 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
202 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 203 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
204 | * |
205 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
206 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 207 | */ |
d3cda233 | 208 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 209 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 210 | [ZONE_DMA] = 256, |
4b51d669 | 211 | #endif |
fb0e7942 | 212 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 213 | [ZONE_DMA32] = 256, |
fb0e7942 | 214 | #endif |
d3cda233 | 215 | [ZONE_NORMAL] = 32, |
e53ef38d | 216 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 217 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 218 | #endif |
d3cda233 | 219 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 220 | }; |
1da177e4 LT |
221 | |
222 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 223 | |
15ad7cdc | 224 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 225 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 226 | "DMA", |
4b51d669 | 227 | #endif |
fb0e7942 | 228 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 229 | "DMA32", |
fb0e7942 | 230 | #endif |
2f1b6248 | 231 | "Normal", |
e53ef38d | 232 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 233 | "HighMem", |
e53ef38d | 234 | #endif |
2a1e274a | 235 | "Movable", |
033fbae9 DW |
236 | #ifdef CONFIG_ZONE_DEVICE |
237 | "Device", | |
238 | #endif | |
2f1b6248 CL |
239 | }; |
240 | ||
60f30350 VB |
241 | char * const migratetype_names[MIGRATE_TYPES] = { |
242 | "Unmovable", | |
243 | "Movable", | |
244 | "Reclaimable", | |
245 | "HighAtomic", | |
246 | #ifdef CONFIG_CMA | |
247 | "CMA", | |
248 | #endif | |
249 | #ifdef CONFIG_MEMORY_ISOLATION | |
250 | "Isolate", | |
251 | #endif | |
252 | }; | |
253 | ||
f1e61557 KS |
254 | compound_page_dtor * const compound_page_dtors[] = { |
255 | NULL, | |
256 | free_compound_page, | |
257 | #ifdef CONFIG_HUGETLB_PAGE | |
258 | free_huge_page, | |
259 | #endif | |
9a982250 KS |
260 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
261 | free_transhuge_page, | |
262 | #endif | |
f1e61557 KS |
263 | }; |
264 | ||
1da177e4 | 265 | int min_free_kbytes = 1024; |
42aa83cb | 266 | int user_min_free_kbytes = -1; |
795ae7a0 | 267 | int watermark_scale_factor = 10; |
1da177e4 | 268 | |
7f16f91f DR |
269 | static unsigned long nr_kernel_pages __meminitdata; |
270 | static unsigned long nr_all_pages __meminitdata; | |
271 | static unsigned long dma_reserve __meminitdata; | |
1da177e4 | 272 | |
0ee332c1 | 273 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
7f16f91f DR |
274 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __meminitdata; |
275 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __meminitdata; | |
276 | static unsigned long required_kernelcore __initdata; | |
a5c6d650 | 277 | static unsigned long required_kernelcore_percent __initdata; |
7f16f91f | 278 | static unsigned long required_movablecore __initdata; |
a5c6d650 | 279 | static unsigned long required_movablecore_percent __initdata; |
7f16f91f DR |
280 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __meminitdata; |
281 | static bool mirrored_kernelcore __meminitdata; | |
0ee332c1 TH |
282 | |
283 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
284 | int movable_zone; | |
285 | EXPORT_SYMBOL(movable_zone); | |
286 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 287 | |
418508c1 MS |
288 | #if MAX_NUMNODES > 1 |
289 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 290 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 291 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 292 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
293 | #endif |
294 | ||
9ef9acb0 MG |
295 | int page_group_by_mobility_disabled __read_mostly; |
296 | ||
3a80a7fa | 297 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3a80a7fa | 298 | /* Returns true if the struct page for the pfn is uninitialised */ |
0e1cc95b | 299 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 300 | { |
ef70b6f4 MG |
301 | int nid = early_pfn_to_nid(pfn); |
302 | ||
303 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
304 | return true; |
305 | ||
306 | return false; | |
307 | } | |
308 | ||
309 | /* | |
310 | * Returns false when the remaining initialisation should be deferred until | |
311 | * later in the boot cycle when it can be parallelised. | |
312 | */ | |
313 | static inline bool update_defer_init(pg_data_t *pgdat, | |
314 | unsigned long pfn, unsigned long zone_end, | |
315 | unsigned long *nr_initialised) | |
316 | { | |
3c2c6488 | 317 | /* Always populate low zones for address-constrained allocations */ |
3a80a7fa MG |
318 | if (zone_end < pgdat_end_pfn(pgdat)) |
319 | return true; | |
3a80a7fa | 320 | (*nr_initialised)++; |
d135e575 | 321 | if ((*nr_initialised > pgdat->static_init_pgcnt) && |
3a80a7fa MG |
322 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
323 | pgdat->first_deferred_pfn = pfn; | |
324 | return false; | |
325 | } | |
326 | ||
327 | return true; | |
328 | } | |
329 | #else | |
3a80a7fa MG |
330 | static inline bool early_page_uninitialised(unsigned long pfn) |
331 | { | |
332 | return false; | |
333 | } | |
334 | ||
335 | static inline bool update_defer_init(pg_data_t *pgdat, | |
336 | unsigned long pfn, unsigned long zone_end, | |
337 | unsigned long *nr_initialised) | |
338 | { | |
339 | return true; | |
340 | } | |
341 | #endif | |
342 | ||
0b423ca2 MG |
343 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
344 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
345 | unsigned long pfn) | |
346 | { | |
347 | #ifdef CONFIG_SPARSEMEM | |
348 | return __pfn_to_section(pfn)->pageblock_flags; | |
349 | #else | |
350 | return page_zone(page)->pageblock_flags; | |
351 | #endif /* CONFIG_SPARSEMEM */ | |
352 | } | |
353 | ||
354 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
355 | { | |
356 | #ifdef CONFIG_SPARSEMEM | |
357 | pfn &= (PAGES_PER_SECTION-1); | |
358 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
359 | #else | |
360 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
361 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
362 | #endif /* CONFIG_SPARSEMEM */ | |
363 | } | |
364 | ||
365 | /** | |
366 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
367 | * @page: The page within the block of interest | |
368 | * @pfn: The target page frame number | |
369 | * @end_bitidx: The last bit of interest to retrieve | |
370 | * @mask: mask of bits that the caller is interested in | |
371 | * | |
372 | * Return: pageblock_bits flags | |
373 | */ | |
374 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
375 | unsigned long pfn, | |
376 | unsigned long end_bitidx, | |
377 | unsigned long mask) | |
378 | { | |
379 | unsigned long *bitmap; | |
380 | unsigned long bitidx, word_bitidx; | |
381 | unsigned long word; | |
382 | ||
383 | bitmap = get_pageblock_bitmap(page, pfn); | |
384 | bitidx = pfn_to_bitidx(page, pfn); | |
385 | word_bitidx = bitidx / BITS_PER_LONG; | |
386 | bitidx &= (BITS_PER_LONG-1); | |
387 | ||
388 | word = bitmap[word_bitidx]; | |
389 | bitidx += end_bitidx; | |
390 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
391 | } | |
392 | ||
393 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
394 | unsigned long end_bitidx, | |
395 | unsigned long mask) | |
396 | { | |
397 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
398 | } | |
399 | ||
400 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
401 | { | |
402 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
403 | } | |
404 | ||
405 | /** | |
406 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
407 | * @page: The page within the block of interest | |
408 | * @flags: The flags to set | |
409 | * @pfn: The target page frame number | |
410 | * @end_bitidx: The last bit of interest | |
411 | * @mask: mask of bits that the caller is interested in | |
412 | */ | |
413 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
414 | unsigned long pfn, | |
415 | unsigned long end_bitidx, | |
416 | unsigned long mask) | |
417 | { | |
418 | unsigned long *bitmap; | |
419 | unsigned long bitidx, word_bitidx; | |
420 | unsigned long old_word, word; | |
421 | ||
422 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
423 | ||
424 | bitmap = get_pageblock_bitmap(page, pfn); | |
425 | bitidx = pfn_to_bitidx(page, pfn); | |
426 | word_bitidx = bitidx / BITS_PER_LONG; | |
427 | bitidx &= (BITS_PER_LONG-1); | |
428 | ||
429 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
430 | ||
431 | bitidx += end_bitidx; | |
432 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
433 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
434 | ||
435 | word = READ_ONCE(bitmap[word_bitidx]); | |
436 | for (;;) { | |
437 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
438 | if (word == old_word) | |
439 | break; | |
440 | word = old_word; | |
441 | } | |
442 | } | |
3a80a7fa | 443 | |
ee6f509c | 444 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 445 | { |
5d0f3f72 KM |
446 | if (unlikely(page_group_by_mobility_disabled && |
447 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
448 | migratetype = MIGRATE_UNMOVABLE; |
449 | ||
b2a0ac88 MG |
450 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
451 | PB_migrate, PB_migrate_end); | |
452 | } | |
453 | ||
13e7444b | 454 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 455 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 456 | { |
bdc8cb98 DH |
457 | int ret = 0; |
458 | unsigned seq; | |
459 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 460 | unsigned long sp, start_pfn; |
c6a57e19 | 461 | |
bdc8cb98 DH |
462 | do { |
463 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
464 | start_pfn = zone->zone_start_pfn; |
465 | sp = zone->spanned_pages; | |
108bcc96 | 466 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
467 | ret = 1; |
468 | } while (zone_span_seqretry(zone, seq)); | |
469 | ||
b5e6a5a2 | 470 | if (ret) |
613813e8 DH |
471 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
472 | pfn, zone_to_nid(zone), zone->name, | |
473 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 474 | |
bdc8cb98 | 475 | return ret; |
c6a57e19 DH |
476 | } |
477 | ||
478 | static int page_is_consistent(struct zone *zone, struct page *page) | |
479 | { | |
14e07298 | 480 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 481 | return 0; |
1da177e4 | 482 | if (zone != page_zone(page)) |
c6a57e19 DH |
483 | return 0; |
484 | ||
485 | return 1; | |
486 | } | |
487 | /* | |
488 | * Temporary debugging check for pages not lying within a given zone. | |
489 | */ | |
d73d3c9f | 490 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
491 | { |
492 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 493 | return 1; |
c6a57e19 DH |
494 | if (!page_is_consistent(zone, page)) |
495 | return 1; | |
496 | ||
1da177e4 LT |
497 | return 0; |
498 | } | |
13e7444b | 499 | #else |
d73d3c9f | 500 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
501 | { |
502 | return 0; | |
503 | } | |
504 | #endif | |
505 | ||
d230dec1 KS |
506 | static void bad_page(struct page *page, const char *reason, |
507 | unsigned long bad_flags) | |
1da177e4 | 508 | { |
d936cf9b HD |
509 | static unsigned long resume; |
510 | static unsigned long nr_shown; | |
511 | static unsigned long nr_unshown; | |
512 | ||
513 | /* | |
514 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
515 | * or allow a steady drip of one report per second. | |
516 | */ | |
517 | if (nr_shown == 60) { | |
518 | if (time_before(jiffies, resume)) { | |
519 | nr_unshown++; | |
520 | goto out; | |
521 | } | |
522 | if (nr_unshown) { | |
ff8e8116 | 523 | pr_alert( |
1e9e6365 | 524 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
525 | nr_unshown); |
526 | nr_unshown = 0; | |
527 | } | |
528 | nr_shown = 0; | |
529 | } | |
530 | if (nr_shown++ == 0) | |
531 | resume = jiffies + 60 * HZ; | |
532 | ||
ff8e8116 | 533 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 534 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
535 | __dump_page(page, reason); |
536 | bad_flags &= page->flags; | |
537 | if (bad_flags) | |
538 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
539 | bad_flags, &bad_flags); | |
4e462112 | 540 | dump_page_owner(page); |
3dc14741 | 541 | |
4f31888c | 542 | print_modules(); |
1da177e4 | 543 | dump_stack(); |
d936cf9b | 544 | out: |
8cc3b392 | 545 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 546 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 547 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
548 | } |
549 | ||
1da177e4 LT |
550 | /* |
551 | * Higher-order pages are called "compound pages". They are structured thusly: | |
552 | * | |
1d798ca3 | 553 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 554 | * |
1d798ca3 KS |
555 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
556 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 557 | * |
1d798ca3 KS |
558 | * The first tail page's ->compound_dtor holds the offset in array of compound |
559 | * page destructors. See compound_page_dtors. | |
1da177e4 | 560 | * |
1d798ca3 | 561 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 562 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 563 | */ |
d98c7a09 | 564 | |
9a982250 | 565 | void free_compound_page(struct page *page) |
d98c7a09 | 566 | { |
d85f3385 | 567 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
568 | } |
569 | ||
d00181b9 | 570 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
571 | { |
572 | int i; | |
573 | int nr_pages = 1 << order; | |
574 | ||
f1e61557 | 575 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
576 | set_compound_order(page, order); |
577 | __SetPageHead(page); | |
578 | for (i = 1; i < nr_pages; i++) { | |
579 | struct page *p = page + i; | |
58a84aa9 | 580 | set_page_count(p, 0); |
1c290f64 | 581 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 582 | set_compound_head(p, page); |
18229df5 | 583 | } |
53f9263b | 584 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
585 | } |
586 | ||
c0a32fc5 SG |
587 | #ifdef CONFIG_DEBUG_PAGEALLOC |
588 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
589 | bool _debug_pagealloc_enabled __read_mostly |
590 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 591 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
592 | bool _debug_guardpage_enabled __read_mostly; |
593 | ||
031bc574 JK |
594 | static int __init early_debug_pagealloc(char *buf) |
595 | { | |
596 | if (!buf) | |
597 | return -EINVAL; | |
2a138dc7 | 598 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
031bc574 JK |
599 | } |
600 | early_param("debug_pagealloc", early_debug_pagealloc); | |
601 | ||
e30825f1 JK |
602 | static bool need_debug_guardpage(void) |
603 | { | |
031bc574 JK |
604 | /* If we don't use debug_pagealloc, we don't need guard page */ |
605 | if (!debug_pagealloc_enabled()) | |
606 | return false; | |
607 | ||
f1c1e9f7 JK |
608 | if (!debug_guardpage_minorder()) |
609 | return false; | |
610 | ||
e30825f1 JK |
611 | return true; |
612 | } | |
613 | ||
614 | static void init_debug_guardpage(void) | |
615 | { | |
031bc574 JK |
616 | if (!debug_pagealloc_enabled()) |
617 | return; | |
618 | ||
f1c1e9f7 JK |
619 | if (!debug_guardpage_minorder()) |
620 | return; | |
621 | ||
e30825f1 JK |
622 | _debug_guardpage_enabled = true; |
623 | } | |
624 | ||
625 | struct page_ext_operations debug_guardpage_ops = { | |
626 | .need = need_debug_guardpage, | |
627 | .init = init_debug_guardpage, | |
628 | }; | |
c0a32fc5 SG |
629 | |
630 | static int __init debug_guardpage_minorder_setup(char *buf) | |
631 | { | |
632 | unsigned long res; | |
633 | ||
634 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 635 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
636 | return 0; |
637 | } | |
638 | _debug_guardpage_minorder = res; | |
1170532b | 639 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
640 | return 0; |
641 | } | |
f1c1e9f7 | 642 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 643 | |
acbc15a4 | 644 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 645 | unsigned int order, int migratetype) |
c0a32fc5 | 646 | { |
e30825f1 JK |
647 | struct page_ext *page_ext; |
648 | ||
649 | if (!debug_guardpage_enabled()) | |
acbc15a4 JK |
650 | return false; |
651 | ||
652 | if (order >= debug_guardpage_minorder()) | |
653 | return false; | |
e30825f1 JK |
654 | |
655 | page_ext = lookup_page_ext(page); | |
f86e4271 | 656 | if (unlikely(!page_ext)) |
acbc15a4 | 657 | return false; |
f86e4271 | 658 | |
e30825f1 JK |
659 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
660 | ||
2847cf95 JK |
661 | INIT_LIST_HEAD(&page->lru); |
662 | set_page_private(page, order); | |
663 | /* Guard pages are not available for any usage */ | |
664 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
665 | |
666 | return true; | |
c0a32fc5 SG |
667 | } |
668 | ||
2847cf95 JK |
669 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
670 | unsigned int order, int migratetype) | |
c0a32fc5 | 671 | { |
e30825f1 JK |
672 | struct page_ext *page_ext; |
673 | ||
674 | if (!debug_guardpage_enabled()) | |
675 | return; | |
676 | ||
677 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
678 | if (unlikely(!page_ext)) |
679 | return; | |
680 | ||
e30825f1 JK |
681 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
682 | ||
2847cf95 JK |
683 | set_page_private(page, 0); |
684 | if (!is_migrate_isolate(migratetype)) | |
685 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
686 | } |
687 | #else | |
980ac167 | 688 | struct page_ext_operations debug_guardpage_ops; |
acbc15a4 JK |
689 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
690 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
691 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
692 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
693 | #endif |
694 | ||
7aeb09f9 | 695 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 696 | { |
4c21e2f2 | 697 | set_page_private(page, order); |
676165a8 | 698 | __SetPageBuddy(page); |
1da177e4 LT |
699 | } |
700 | ||
701 | static inline void rmv_page_order(struct page *page) | |
702 | { | |
676165a8 | 703 | __ClearPageBuddy(page); |
4c21e2f2 | 704 | set_page_private(page, 0); |
1da177e4 LT |
705 | } |
706 | ||
1da177e4 LT |
707 | /* |
708 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 709 | * we can coalesce a page and its buddy if |
13ad59df | 710 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 711 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
712 | * (c) a page and its buddy have the same order && |
713 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 714 | * |
6e292b9b MW |
715 | * For recording whether a page is in the buddy system, we set PageBuddy. |
716 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 717 | * |
676165a8 | 718 | * For recording page's order, we use page_private(page). |
1da177e4 | 719 | */ |
cb2b95e1 | 720 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 721 | unsigned int order) |
1da177e4 | 722 | { |
c0a32fc5 | 723 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
724 | if (page_zone_id(page) != page_zone_id(buddy)) |
725 | return 0; | |
726 | ||
4c5018ce WY |
727 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
728 | ||
c0a32fc5 SG |
729 | return 1; |
730 | } | |
731 | ||
cb2b95e1 | 732 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
733 | /* |
734 | * zone check is done late to avoid uselessly | |
735 | * calculating zone/node ids for pages that could | |
736 | * never merge. | |
737 | */ | |
738 | if (page_zone_id(page) != page_zone_id(buddy)) | |
739 | return 0; | |
740 | ||
4c5018ce WY |
741 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
742 | ||
6aa3001b | 743 | return 1; |
676165a8 | 744 | } |
6aa3001b | 745 | return 0; |
1da177e4 LT |
746 | } |
747 | ||
748 | /* | |
749 | * Freeing function for a buddy system allocator. | |
750 | * | |
751 | * The concept of a buddy system is to maintain direct-mapped table | |
752 | * (containing bit values) for memory blocks of various "orders". | |
753 | * The bottom level table contains the map for the smallest allocatable | |
754 | * units of memory (here, pages), and each level above it describes | |
755 | * pairs of units from the levels below, hence, "buddies". | |
756 | * At a high level, all that happens here is marking the table entry | |
757 | * at the bottom level available, and propagating the changes upward | |
758 | * as necessary, plus some accounting needed to play nicely with other | |
759 | * parts of the VM system. | |
760 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
761 | * free pages of length of (1 << order) and marked with PageBuddy. |
762 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 763 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
764 | * other. That is, if we allocate a small block, and both were |
765 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 766 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 767 | * triggers coalescing into a block of larger size. |
1da177e4 | 768 | * |
6d49e352 | 769 | * -- nyc |
1da177e4 LT |
770 | */ |
771 | ||
48db57f8 | 772 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 773 | unsigned long pfn, |
ed0ae21d MG |
774 | struct zone *zone, unsigned int order, |
775 | int migratetype) | |
1da177e4 | 776 | { |
76741e77 VB |
777 | unsigned long combined_pfn; |
778 | unsigned long uninitialized_var(buddy_pfn); | |
6dda9d55 | 779 | struct page *buddy; |
d9dddbf5 VB |
780 | unsigned int max_order; |
781 | ||
782 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 783 | |
d29bb978 | 784 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 785 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 786 | |
ed0ae21d | 787 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 788 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 789 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 790 | |
76741e77 | 791 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 792 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 793 | |
d9dddbf5 | 794 | continue_merging: |
3c605096 | 795 | while (order < max_order - 1) { |
76741e77 VB |
796 | buddy_pfn = __find_buddy_pfn(pfn, order); |
797 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
798 | |
799 | if (!pfn_valid_within(buddy_pfn)) | |
800 | goto done_merging; | |
cb2b95e1 | 801 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 802 | goto done_merging; |
c0a32fc5 SG |
803 | /* |
804 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
805 | * merge with it and move up one order. | |
806 | */ | |
807 | if (page_is_guard(buddy)) { | |
2847cf95 | 808 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
809 | } else { |
810 | list_del(&buddy->lru); | |
811 | zone->free_area[order].nr_free--; | |
812 | rmv_page_order(buddy); | |
813 | } | |
76741e77 VB |
814 | combined_pfn = buddy_pfn & pfn; |
815 | page = page + (combined_pfn - pfn); | |
816 | pfn = combined_pfn; | |
1da177e4 LT |
817 | order++; |
818 | } | |
d9dddbf5 VB |
819 | if (max_order < MAX_ORDER) { |
820 | /* If we are here, it means order is >= pageblock_order. | |
821 | * We want to prevent merge between freepages on isolate | |
822 | * pageblock and normal pageblock. Without this, pageblock | |
823 | * isolation could cause incorrect freepage or CMA accounting. | |
824 | * | |
825 | * We don't want to hit this code for the more frequent | |
826 | * low-order merging. | |
827 | */ | |
828 | if (unlikely(has_isolate_pageblock(zone))) { | |
829 | int buddy_mt; | |
830 | ||
76741e77 VB |
831 | buddy_pfn = __find_buddy_pfn(pfn, order); |
832 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
833 | buddy_mt = get_pageblock_migratetype(buddy); |
834 | ||
835 | if (migratetype != buddy_mt | |
836 | && (is_migrate_isolate(migratetype) || | |
837 | is_migrate_isolate(buddy_mt))) | |
838 | goto done_merging; | |
839 | } | |
840 | max_order++; | |
841 | goto continue_merging; | |
842 | } | |
843 | ||
844 | done_merging: | |
1da177e4 | 845 | set_page_order(page, order); |
6dda9d55 CZ |
846 | |
847 | /* | |
848 | * If this is not the largest possible page, check if the buddy | |
849 | * of the next-highest order is free. If it is, it's possible | |
850 | * that pages are being freed that will coalesce soon. In case, | |
851 | * that is happening, add the free page to the tail of the list | |
852 | * so it's less likely to be used soon and more likely to be merged | |
853 | * as a higher order page | |
854 | */ | |
13ad59df | 855 | if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) { |
6dda9d55 | 856 | struct page *higher_page, *higher_buddy; |
76741e77 VB |
857 | combined_pfn = buddy_pfn & pfn; |
858 | higher_page = page + (combined_pfn - pfn); | |
859 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
860 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
b4fb8f66 TL |
861 | if (pfn_valid_within(buddy_pfn) && |
862 | page_is_buddy(higher_page, higher_buddy, order + 1)) { | |
6dda9d55 CZ |
863 | list_add_tail(&page->lru, |
864 | &zone->free_area[order].free_list[migratetype]); | |
865 | goto out; | |
866 | } | |
867 | } | |
868 | ||
869 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
870 | out: | |
1da177e4 LT |
871 | zone->free_area[order].nr_free++; |
872 | } | |
873 | ||
7bfec6f4 MG |
874 | /* |
875 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
876 | * try and check multiple fields with one check. The caller must do a detailed | |
877 | * check if necessary. | |
878 | */ | |
879 | static inline bool page_expected_state(struct page *page, | |
880 | unsigned long check_flags) | |
881 | { | |
882 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
883 | return false; | |
884 | ||
885 | if (unlikely((unsigned long)page->mapping | | |
886 | page_ref_count(page) | | |
887 | #ifdef CONFIG_MEMCG | |
888 | (unsigned long)page->mem_cgroup | | |
889 | #endif | |
890 | (page->flags & check_flags))) | |
891 | return false; | |
892 | ||
893 | return true; | |
894 | } | |
895 | ||
bb552ac6 | 896 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 897 | { |
7bfec6f4 MG |
898 | const char *bad_reason; |
899 | unsigned long bad_flags; | |
900 | ||
7bfec6f4 MG |
901 | bad_reason = NULL; |
902 | bad_flags = 0; | |
f0b791a3 | 903 | |
53f9263b | 904 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
905 | bad_reason = "nonzero mapcount"; |
906 | if (unlikely(page->mapping != NULL)) | |
907 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 908 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 909 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
910 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
911 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
912 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
913 | } | |
9edad6ea JW |
914 | #ifdef CONFIG_MEMCG |
915 | if (unlikely(page->mem_cgroup)) | |
916 | bad_reason = "page still charged to cgroup"; | |
917 | #endif | |
7bfec6f4 | 918 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
919 | } |
920 | ||
921 | static inline int free_pages_check(struct page *page) | |
922 | { | |
da838d4f | 923 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 924 | return 0; |
bb552ac6 MG |
925 | |
926 | /* Something has gone sideways, find it */ | |
927 | free_pages_check_bad(page); | |
7bfec6f4 | 928 | return 1; |
1da177e4 LT |
929 | } |
930 | ||
4db7548c MG |
931 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
932 | { | |
933 | int ret = 1; | |
934 | ||
935 | /* | |
936 | * We rely page->lru.next never has bit 0 set, unless the page | |
937 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
938 | */ | |
939 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
940 | ||
941 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
942 | ret = 0; | |
943 | goto out; | |
944 | } | |
945 | switch (page - head_page) { | |
946 | case 1: | |
4da1984e | 947 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c MG |
948 | if (unlikely(compound_mapcount(page))) { |
949 | bad_page(page, "nonzero compound_mapcount", 0); | |
950 | goto out; | |
951 | } | |
952 | break; | |
953 | case 2: | |
954 | /* | |
955 | * the second tail page: ->mapping is | |
fa3015b7 | 956 | * deferred_list.next -- ignore value. |
4db7548c MG |
957 | */ |
958 | break; | |
959 | default: | |
960 | if (page->mapping != TAIL_MAPPING) { | |
961 | bad_page(page, "corrupted mapping in tail page", 0); | |
962 | goto out; | |
963 | } | |
964 | break; | |
965 | } | |
966 | if (unlikely(!PageTail(page))) { | |
967 | bad_page(page, "PageTail not set", 0); | |
968 | goto out; | |
969 | } | |
970 | if (unlikely(compound_head(page) != head_page)) { | |
971 | bad_page(page, "compound_head not consistent", 0); | |
972 | goto out; | |
973 | } | |
974 | ret = 0; | |
975 | out: | |
976 | page->mapping = NULL; | |
977 | clear_compound_head(page); | |
978 | return ret; | |
979 | } | |
980 | ||
e2769dbd MG |
981 | static __always_inline bool free_pages_prepare(struct page *page, |
982 | unsigned int order, bool check_free) | |
4db7548c | 983 | { |
e2769dbd | 984 | int bad = 0; |
4db7548c | 985 | |
4db7548c MG |
986 | VM_BUG_ON_PAGE(PageTail(page), page); |
987 | ||
e2769dbd | 988 | trace_mm_page_free(page, order); |
e2769dbd MG |
989 | |
990 | /* | |
991 | * Check tail pages before head page information is cleared to | |
992 | * avoid checking PageCompound for order-0 pages. | |
993 | */ | |
994 | if (unlikely(order)) { | |
995 | bool compound = PageCompound(page); | |
996 | int i; | |
997 | ||
998 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 999 | |
9a73f61b KS |
1000 | if (compound) |
1001 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1002 | for (i = 1; i < (1 << order); i++) { |
1003 | if (compound) | |
1004 | bad += free_tail_pages_check(page, page + i); | |
1005 | if (unlikely(free_pages_check(page + i))) { | |
1006 | bad++; | |
1007 | continue; | |
1008 | } | |
1009 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1010 | } | |
1011 | } | |
bda807d4 | 1012 | if (PageMappingFlags(page)) |
4db7548c | 1013 | page->mapping = NULL; |
c4159a75 | 1014 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
4949148a | 1015 | memcg_kmem_uncharge(page, order); |
e2769dbd MG |
1016 | if (check_free) |
1017 | bad += free_pages_check(page); | |
1018 | if (bad) | |
1019 | return false; | |
4db7548c | 1020 | |
e2769dbd MG |
1021 | page_cpupid_reset_last(page); |
1022 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1023 | reset_page_owner(page, order); | |
4db7548c MG |
1024 | |
1025 | if (!PageHighMem(page)) { | |
1026 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1027 | PAGE_SIZE << order); |
4db7548c | 1028 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1029 | PAGE_SIZE << order); |
4db7548c | 1030 | } |
e2769dbd MG |
1031 | arch_free_page(page, order); |
1032 | kernel_poison_pages(page, 1 << order, 0); | |
1033 | kernel_map_pages(page, 1 << order, 0); | |
29b52de1 | 1034 | kasan_free_pages(page, order); |
4db7548c | 1035 | |
4db7548c MG |
1036 | return true; |
1037 | } | |
1038 | ||
e2769dbd MG |
1039 | #ifdef CONFIG_DEBUG_VM |
1040 | static inline bool free_pcp_prepare(struct page *page) | |
1041 | { | |
1042 | return free_pages_prepare(page, 0, true); | |
1043 | } | |
1044 | ||
1045 | static inline bool bulkfree_pcp_prepare(struct page *page) | |
1046 | { | |
1047 | return false; | |
1048 | } | |
1049 | #else | |
1050 | static bool free_pcp_prepare(struct page *page) | |
1051 | { | |
1052 | return free_pages_prepare(page, 0, false); | |
1053 | } | |
1054 | ||
4db7548c MG |
1055 | static bool bulkfree_pcp_prepare(struct page *page) |
1056 | { | |
1057 | return free_pages_check(page); | |
1058 | } | |
1059 | #endif /* CONFIG_DEBUG_VM */ | |
1060 | ||
97334162 AL |
1061 | static inline void prefetch_buddy(struct page *page) |
1062 | { | |
1063 | unsigned long pfn = page_to_pfn(page); | |
1064 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1065 | struct page *buddy = page + (buddy_pfn - pfn); | |
1066 | ||
1067 | prefetch(buddy); | |
1068 | } | |
1069 | ||
1da177e4 | 1070 | /* |
5f8dcc21 | 1071 | * Frees a number of pages from the PCP lists |
1da177e4 | 1072 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1073 | * count is the number of pages to free. |
1da177e4 LT |
1074 | * |
1075 | * If the zone was previously in an "all pages pinned" state then look to | |
1076 | * see if this freeing clears that state. | |
1077 | * | |
1078 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1079 | * pinned" detection logic. | |
1080 | */ | |
5f8dcc21 MG |
1081 | static void free_pcppages_bulk(struct zone *zone, int count, |
1082 | struct per_cpu_pages *pcp) | |
1da177e4 | 1083 | { |
5f8dcc21 | 1084 | int migratetype = 0; |
a6f9edd6 | 1085 | int batch_free = 0; |
97334162 | 1086 | int prefetch_nr = 0; |
3777999d | 1087 | bool isolated_pageblocks; |
0a5f4e5b AL |
1088 | struct page *page, *tmp; |
1089 | LIST_HEAD(head); | |
f2260e6b | 1090 | |
e5b31ac2 | 1091 | while (count) { |
5f8dcc21 MG |
1092 | struct list_head *list; |
1093 | ||
1094 | /* | |
a6f9edd6 MG |
1095 | * Remove pages from lists in a round-robin fashion. A |
1096 | * batch_free count is maintained that is incremented when an | |
1097 | * empty list is encountered. This is so more pages are freed | |
1098 | * off fuller lists instead of spinning excessively around empty | |
1099 | * lists | |
5f8dcc21 MG |
1100 | */ |
1101 | do { | |
a6f9edd6 | 1102 | batch_free++; |
5f8dcc21 MG |
1103 | if (++migratetype == MIGRATE_PCPTYPES) |
1104 | migratetype = 0; | |
1105 | list = &pcp->lists[migratetype]; | |
1106 | } while (list_empty(list)); | |
48db57f8 | 1107 | |
1d16871d NK |
1108 | /* This is the only non-empty list. Free them all. */ |
1109 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1110 | batch_free = count; |
1d16871d | 1111 | |
a6f9edd6 | 1112 | do { |
a16601c5 | 1113 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1114 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1115 | list_del(&page->lru); |
77ba9062 | 1116 | pcp->count--; |
aa016d14 | 1117 | |
4db7548c MG |
1118 | if (bulkfree_pcp_prepare(page)) |
1119 | continue; | |
1120 | ||
0a5f4e5b | 1121 | list_add_tail(&page->lru, &head); |
97334162 AL |
1122 | |
1123 | /* | |
1124 | * We are going to put the page back to the global | |
1125 | * pool, prefetch its buddy to speed up later access | |
1126 | * under zone->lock. It is believed the overhead of | |
1127 | * an additional test and calculating buddy_pfn here | |
1128 | * can be offset by reduced memory latency later. To | |
1129 | * avoid excessive prefetching due to large count, only | |
1130 | * prefetch buddy for the first pcp->batch nr of pages. | |
1131 | */ | |
1132 | if (prefetch_nr++ < pcp->batch) | |
1133 | prefetch_buddy(page); | |
e5b31ac2 | 1134 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1135 | } |
0a5f4e5b AL |
1136 | |
1137 | spin_lock(&zone->lock); | |
1138 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1139 | ||
1140 | /* | |
1141 | * Use safe version since after __free_one_page(), | |
1142 | * page->lru.next will not point to original list. | |
1143 | */ | |
1144 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1145 | int mt = get_pcppage_migratetype(page); | |
1146 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1147 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1148 | /* Pageblock could have been isolated meanwhile */ | |
1149 | if (unlikely(isolated_pageblocks)) | |
1150 | mt = get_pageblock_migratetype(page); | |
1151 | ||
1152 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); | |
1153 | trace_mm_page_pcpu_drain(page, 0, mt); | |
1154 | } | |
d34b0733 | 1155 | spin_unlock(&zone->lock); |
1da177e4 LT |
1156 | } |
1157 | ||
dc4b0caf MG |
1158 | static void free_one_page(struct zone *zone, |
1159 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1160 | unsigned int order, |
ed0ae21d | 1161 | int migratetype) |
1da177e4 | 1162 | { |
d34b0733 | 1163 | spin_lock(&zone->lock); |
ad53f92e JK |
1164 | if (unlikely(has_isolate_pageblock(zone) || |
1165 | is_migrate_isolate(migratetype))) { | |
1166 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1167 | } |
dc4b0caf | 1168 | __free_one_page(page, pfn, zone, order, migratetype); |
d34b0733 | 1169 | spin_unlock(&zone->lock); |
48db57f8 NP |
1170 | } |
1171 | ||
1e8ce83c | 1172 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1173 | unsigned long zone, int nid) |
1e8ce83c | 1174 | { |
d0dc12e8 | 1175 | mm_zero_struct_page(page); |
1e8ce83c | 1176 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1177 | init_page_count(page); |
1178 | page_mapcount_reset(page); | |
1179 | page_cpupid_reset_last(page); | |
1e8ce83c | 1180 | |
1e8ce83c RH |
1181 | INIT_LIST_HEAD(&page->lru); |
1182 | #ifdef WANT_PAGE_VIRTUAL | |
1183 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1184 | if (!is_highmem_idx(zone)) | |
1185 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1186 | #endif | |
1187 | } | |
1188 | ||
7e18adb4 | 1189 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1190 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1191 | { |
1192 | pg_data_t *pgdat; | |
1193 | int nid, zid; | |
1194 | ||
1195 | if (!early_page_uninitialised(pfn)) | |
1196 | return; | |
1197 | ||
1198 | nid = early_pfn_to_nid(pfn); | |
1199 | pgdat = NODE_DATA(nid); | |
1200 | ||
1201 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1202 | struct zone *zone = &pgdat->node_zones[zid]; | |
1203 | ||
1204 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1205 | break; | |
1206 | } | |
d0dc12e8 | 1207 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1208 | } |
1209 | #else | |
1210 | static inline void init_reserved_page(unsigned long pfn) | |
1211 | { | |
1212 | } | |
1213 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1214 | ||
92923ca3 NZ |
1215 | /* |
1216 | * Initialised pages do not have PageReserved set. This function is | |
1217 | * called for each range allocated by the bootmem allocator and | |
1218 | * marks the pages PageReserved. The remaining valid pages are later | |
1219 | * sent to the buddy page allocator. | |
1220 | */ | |
4b50bcc7 | 1221 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1222 | { |
1223 | unsigned long start_pfn = PFN_DOWN(start); | |
1224 | unsigned long end_pfn = PFN_UP(end); | |
1225 | ||
7e18adb4 MG |
1226 | for (; start_pfn < end_pfn; start_pfn++) { |
1227 | if (pfn_valid(start_pfn)) { | |
1228 | struct page *page = pfn_to_page(start_pfn); | |
1229 | ||
1230 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1231 | |
1232 | /* Avoid false-positive PageTail() */ | |
1233 | INIT_LIST_HEAD(&page->lru); | |
1234 | ||
7e18adb4 MG |
1235 | SetPageReserved(page); |
1236 | } | |
1237 | } | |
92923ca3 NZ |
1238 | } |
1239 | ||
ec95f53a KM |
1240 | static void __free_pages_ok(struct page *page, unsigned int order) |
1241 | { | |
d34b0733 | 1242 | unsigned long flags; |
95e34412 | 1243 | int migratetype; |
dc4b0caf | 1244 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1245 | |
e2769dbd | 1246 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1247 | return; |
1248 | ||
cfc47a28 | 1249 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1250 | local_irq_save(flags); |
1251 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1252 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1253 | local_irq_restore(flags); |
1da177e4 LT |
1254 | } |
1255 | ||
949698a3 | 1256 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1257 | { |
c3993076 | 1258 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1259 | struct page *p = page; |
c3993076 | 1260 | unsigned int loop; |
a226f6c8 | 1261 | |
e2d0bd2b YL |
1262 | prefetchw(p); |
1263 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1264 | prefetchw(p + 1); | |
c3993076 JW |
1265 | __ClearPageReserved(p); |
1266 | set_page_count(p, 0); | |
a226f6c8 | 1267 | } |
e2d0bd2b YL |
1268 | __ClearPageReserved(p); |
1269 | set_page_count(p, 0); | |
c3993076 | 1270 | |
e2d0bd2b | 1271 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1272 | set_page_refcounted(page); |
1273 | __free_pages(page, order); | |
a226f6c8 DH |
1274 | } |
1275 | ||
75a592a4 MG |
1276 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1277 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1278 | |
75a592a4 MG |
1279 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1280 | ||
1281 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1282 | { | |
7ace9917 | 1283 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1284 | int nid; |
1285 | ||
7ace9917 | 1286 | spin_lock(&early_pfn_lock); |
75a592a4 | 1287 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1288 | if (nid < 0) |
e4568d38 | 1289 | nid = first_online_node; |
7ace9917 MG |
1290 | spin_unlock(&early_pfn_lock); |
1291 | ||
1292 | return nid; | |
75a592a4 MG |
1293 | } |
1294 | #endif | |
1295 | ||
1296 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
d73d3c9f MK |
1297 | static inline bool __meminit __maybe_unused |
1298 | meminit_pfn_in_nid(unsigned long pfn, int node, | |
1299 | struct mminit_pfnnid_cache *state) | |
75a592a4 MG |
1300 | { |
1301 | int nid; | |
1302 | ||
1303 | nid = __early_pfn_to_nid(pfn, state); | |
1304 | if (nid >= 0 && nid != node) | |
1305 | return false; | |
1306 | return true; | |
1307 | } | |
1308 | ||
1309 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1310 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1311 | { | |
1312 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1313 | } | |
1314 | ||
1315 | #else | |
1316 | ||
1317 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1318 | { | |
1319 | return true; | |
1320 | } | |
d73d3c9f MK |
1321 | static inline bool __meminit __maybe_unused |
1322 | meminit_pfn_in_nid(unsigned long pfn, int node, | |
1323 | struct mminit_pfnnid_cache *state) | |
75a592a4 MG |
1324 | { |
1325 | return true; | |
1326 | } | |
1327 | #endif | |
1328 | ||
1329 | ||
0e1cc95b | 1330 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1331 | unsigned int order) |
1332 | { | |
1333 | if (early_page_uninitialised(pfn)) | |
1334 | return; | |
949698a3 | 1335 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1336 | } |
1337 | ||
7cf91a98 JK |
1338 | /* |
1339 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1340 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1341 | * with the migration of free compaction scanner. The scanners then need to | |
1342 | * use only pfn_valid_within() check for arches that allow holes within | |
1343 | * pageblocks. | |
1344 | * | |
1345 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1346 | * | |
1347 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1348 | * i.e. it's possible that all pages within a zones range of pages do not | |
1349 | * belong to a single zone. We assume that a border between node0 and node1 | |
1350 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1351 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1352 | * the first and last page of a pageblock and avoid checking each individual | |
1353 | * page in a pageblock. | |
1354 | */ | |
1355 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1356 | unsigned long end_pfn, struct zone *zone) | |
1357 | { | |
1358 | struct page *start_page; | |
1359 | struct page *end_page; | |
1360 | ||
1361 | /* end_pfn is one past the range we are checking */ | |
1362 | end_pfn--; | |
1363 | ||
1364 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1365 | return NULL; | |
1366 | ||
2d070eab MH |
1367 | start_page = pfn_to_online_page(start_pfn); |
1368 | if (!start_page) | |
1369 | return NULL; | |
7cf91a98 JK |
1370 | |
1371 | if (page_zone(start_page) != zone) | |
1372 | return NULL; | |
1373 | ||
1374 | end_page = pfn_to_page(end_pfn); | |
1375 | ||
1376 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1377 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1378 | return NULL; | |
1379 | ||
1380 | return start_page; | |
1381 | } | |
1382 | ||
1383 | void set_zone_contiguous(struct zone *zone) | |
1384 | { | |
1385 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1386 | unsigned long block_end_pfn; | |
1387 | ||
1388 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1389 | for (; block_start_pfn < zone_end_pfn(zone); | |
1390 | block_start_pfn = block_end_pfn, | |
1391 | block_end_pfn += pageblock_nr_pages) { | |
1392 | ||
1393 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1394 | ||
1395 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1396 | block_end_pfn, zone)) | |
1397 | return; | |
1398 | } | |
1399 | ||
1400 | /* We confirm that there is no hole */ | |
1401 | zone->contiguous = true; | |
1402 | } | |
1403 | ||
1404 | void clear_zone_contiguous(struct zone *zone) | |
1405 | { | |
1406 | zone->contiguous = false; | |
1407 | } | |
1408 | ||
7e18adb4 | 1409 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1410 | static void __init deferred_free_range(unsigned long pfn, |
1411 | unsigned long nr_pages) | |
a4de83dd | 1412 | { |
2f47a91f PT |
1413 | struct page *page; |
1414 | unsigned long i; | |
a4de83dd | 1415 | |
2f47a91f | 1416 | if (!nr_pages) |
a4de83dd MG |
1417 | return; |
1418 | ||
2f47a91f PT |
1419 | page = pfn_to_page(pfn); |
1420 | ||
a4de83dd | 1421 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1422 | if (nr_pages == pageblock_nr_pages && |
1423 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1424 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
e780149b | 1425 | __free_pages_boot_core(page, pageblock_order); |
a4de83dd MG |
1426 | return; |
1427 | } | |
1428 | ||
e780149b XQ |
1429 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1430 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1431 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
949698a3 | 1432 | __free_pages_boot_core(page, 0); |
e780149b | 1433 | } |
a4de83dd MG |
1434 | } |
1435 | ||
d3cd131d NS |
1436 | /* Completion tracking for deferred_init_memmap() threads */ |
1437 | static atomic_t pgdat_init_n_undone __initdata; | |
1438 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1439 | ||
1440 | static inline void __init pgdat_init_report_one_done(void) | |
1441 | { | |
1442 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1443 | complete(&pgdat_init_all_done_comp); | |
1444 | } | |
0e1cc95b | 1445 | |
2f47a91f | 1446 | /* |
80b1f41c PT |
1447 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1448 | * | |
1449 | * First we check if pfn is valid on architectures where it is possible to have | |
1450 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1451 | * function is optimized out. | |
1452 | * | |
1453 | * Then, we check if a current large page is valid by only checking the validity | |
1454 | * of the head pfn. | |
1455 | * | |
1456 | * Finally, meminit_pfn_in_nid is checked on systems where pfns can interleave | |
1457 | * within a node: a pfn is between start and end of a node, but does not belong | |
1458 | * to this memory node. | |
2f47a91f | 1459 | */ |
80b1f41c PT |
1460 | static inline bool __init |
1461 | deferred_pfn_valid(int nid, unsigned long pfn, | |
1462 | struct mminit_pfnnid_cache *nid_init_state) | |
2f47a91f | 1463 | { |
80b1f41c PT |
1464 | if (!pfn_valid_within(pfn)) |
1465 | return false; | |
1466 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1467 | return false; | |
1468 | if (!meminit_pfn_in_nid(pfn, nid, nid_init_state)) | |
1469 | return false; | |
1470 | return true; | |
1471 | } | |
2f47a91f | 1472 | |
80b1f41c PT |
1473 | /* |
1474 | * Free pages to buddy allocator. Try to free aligned pages in | |
1475 | * pageblock_nr_pages sizes. | |
1476 | */ | |
1477 | static void __init deferred_free_pages(int nid, int zid, unsigned long pfn, | |
1478 | unsigned long end_pfn) | |
1479 | { | |
1480 | struct mminit_pfnnid_cache nid_init_state = { }; | |
1481 | unsigned long nr_pgmask = pageblock_nr_pages - 1; | |
1482 | unsigned long nr_free = 0; | |
2f47a91f | 1483 | |
80b1f41c PT |
1484 | for (; pfn < end_pfn; pfn++) { |
1485 | if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) { | |
1486 | deferred_free_range(pfn - nr_free, nr_free); | |
1487 | nr_free = 0; | |
1488 | } else if (!(pfn & nr_pgmask)) { | |
1489 | deferred_free_range(pfn - nr_free, nr_free); | |
1490 | nr_free = 1; | |
3a2d7fa8 | 1491 | touch_nmi_watchdog(); |
80b1f41c PT |
1492 | } else { |
1493 | nr_free++; | |
1494 | } | |
1495 | } | |
1496 | /* Free the last block of pages to allocator */ | |
1497 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1498 | } |
1499 | ||
80b1f41c PT |
1500 | /* |
1501 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1502 | * by performing it only once every pageblock_nr_pages. | |
1503 | * Return number of pages initialized. | |
1504 | */ | |
1505 | static unsigned long __init deferred_init_pages(int nid, int zid, | |
1506 | unsigned long pfn, | |
1507 | unsigned long end_pfn) | |
2f47a91f PT |
1508 | { |
1509 | struct mminit_pfnnid_cache nid_init_state = { }; | |
1510 | unsigned long nr_pgmask = pageblock_nr_pages - 1; | |
2f47a91f | 1511 | unsigned long nr_pages = 0; |
2f47a91f | 1512 | struct page *page = NULL; |
2f47a91f | 1513 | |
80b1f41c PT |
1514 | for (; pfn < end_pfn; pfn++) { |
1515 | if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) { | |
1516 | page = NULL; | |
2f47a91f | 1517 | continue; |
80b1f41c | 1518 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1519 | page = pfn_to_page(pfn); |
3a2d7fa8 | 1520 | touch_nmi_watchdog(); |
80b1f41c PT |
1521 | } else { |
1522 | page++; | |
2f47a91f | 1523 | } |
d0dc12e8 | 1524 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1525 | nr_pages++; |
2f47a91f | 1526 | } |
80b1f41c | 1527 | return (nr_pages); |
2f47a91f PT |
1528 | } |
1529 | ||
7e18adb4 | 1530 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1531 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1532 | { |
0e1cc95b MG |
1533 | pg_data_t *pgdat = data; |
1534 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1535 | unsigned long start = jiffies; |
1536 | unsigned long nr_pages = 0; | |
3a2d7fa8 | 1537 | unsigned long spfn, epfn, first_init_pfn, flags; |
2f47a91f PT |
1538 | phys_addr_t spa, epa; |
1539 | int zid; | |
7e18adb4 | 1540 | struct zone *zone; |
0e1cc95b | 1541 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
2f47a91f | 1542 | u64 i; |
7e18adb4 | 1543 | |
3a2d7fa8 PT |
1544 | /* Bind memory initialisation thread to a local node if possible */ |
1545 | if (!cpumask_empty(cpumask)) | |
1546 | set_cpus_allowed_ptr(current, cpumask); | |
1547 | ||
1548 | pgdat_resize_lock(pgdat, &flags); | |
1549 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1550 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1551 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1552 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1553 | return 0; |
1554 | } | |
1555 | ||
7e18adb4 MG |
1556 | /* Sanity check boundaries */ |
1557 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1558 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1559 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1560 | ||
1561 | /* Only the highest zone is deferred so find it */ | |
1562 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1563 | zone = pgdat->node_zones + zid; | |
1564 | if (first_init_pfn < zone_end_pfn(zone)) | |
1565 | break; | |
1566 | } | |
2f47a91f | 1567 | first_init_pfn = max(zone->zone_start_pfn, first_init_pfn); |
7e18adb4 | 1568 | |
80b1f41c PT |
1569 | /* |
1570 | * Initialize and free pages. We do it in two loops: first we initialize | |
1571 | * struct page, than free to buddy allocator, because while we are | |
1572 | * freeing pages we can access pages that are ahead (computing buddy | |
1573 | * page in __free_one_page()). | |
1574 | */ | |
1575 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { | |
1576 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1577 | epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa)); | |
1578 | nr_pages += deferred_init_pages(nid, zid, spfn, epfn); | |
1579 | } | |
2f47a91f PT |
1580 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { |
1581 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1582 | epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa)); | |
80b1f41c | 1583 | deferred_free_pages(nid, zid, spfn, epfn); |
7e18adb4 | 1584 | } |
3a2d7fa8 | 1585 | pgdat_resize_unlock(pgdat, &flags); |
7e18adb4 MG |
1586 | |
1587 | /* Sanity check that the next zone really is unpopulated */ | |
1588 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1589 | ||
0e1cc95b | 1590 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1591 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1592 | |
1593 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1594 | return 0; |
1595 | } | |
c9e97a19 PT |
1596 | |
1597 | /* | |
1598 | * During boot we initialize deferred pages on-demand, as needed, but once | |
1599 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
1600 | * and we can permanently disable that path. | |
1601 | */ | |
1602 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
1603 | ||
1604 | /* | |
1605 | * If this zone has deferred pages, try to grow it by initializing enough | |
1606 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
1607 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
1608 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
1609 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
1610 | * | |
1611 | * Return true when zone was grown, otherwise return false. We return true even | |
1612 | * when we grow less than requested, to let the caller decide if there are | |
1613 | * enough pages to satisfy the allocation. | |
1614 | * | |
1615 | * Note: We use noinline because this function is needed only during boot, and | |
1616 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
1617 | * making sure that it is not inlined into permanent text section. | |
1618 | */ | |
1619 | static noinline bool __init | |
1620 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
1621 | { | |
1622 | int zid = zone_idx(zone); | |
1623 | int nid = zone_to_nid(zone); | |
1624 | pg_data_t *pgdat = NODE_DATA(nid); | |
1625 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); | |
1626 | unsigned long nr_pages = 0; | |
1627 | unsigned long first_init_pfn, spfn, epfn, t, flags; | |
1628 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; | |
1629 | phys_addr_t spa, epa; | |
1630 | u64 i; | |
1631 | ||
1632 | /* Only the last zone may have deferred pages */ | |
1633 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
1634 | return false; | |
1635 | ||
1636 | pgdat_resize_lock(pgdat, &flags); | |
1637 | ||
1638 | /* | |
1639 | * If deferred pages have been initialized while we were waiting for | |
1640 | * the lock, return true, as the zone was grown. The caller will retry | |
1641 | * this zone. We won't return to this function since the caller also | |
1642 | * has this static branch. | |
1643 | */ | |
1644 | if (!static_branch_unlikely(&deferred_pages)) { | |
1645 | pgdat_resize_unlock(pgdat, &flags); | |
1646 | return true; | |
1647 | } | |
1648 | ||
1649 | /* | |
1650 | * If someone grew this zone while we were waiting for spinlock, return | |
1651 | * true, as there might be enough pages already. | |
1652 | */ | |
1653 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
1654 | pgdat_resize_unlock(pgdat, &flags); | |
1655 | return true; | |
1656 | } | |
1657 | ||
1658 | first_init_pfn = max(zone->zone_start_pfn, first_deferred_pfn); | |
1659 | ||
1660 | if (first_init_pfn >= pgdat_end_pfn(pgdat)) { | |
1661 | pgdat_resize_unlock(pgdat, &flags); | |
1662 | return false; | |
1663 | } | |
1664 | ||
1665 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { | |
1666 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1667 | epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa)); | |
1668 | ||
1669 | while (spfn < epfn && nr_pages < nr_pages_needed) { | |
1670 | t = ALIGN(spfn + PAGES_PER_SECTION, PAGES_PER_SECTION); | |
1671 | first_deferred_pfn = min(t, epfn); | |
1672 | nr_pages += deferred_init_pages(nid, zid, spfn, | |
1673 | first_deferred_pfn); | |
1674 | spfn = first_deferred_pfn; | |
1675 | } | |
1676 | ||
1677 | if (nr_pages >= nr_pages_needed) | |
1678 | break; | |
1679 | } | |
1680 | ||
1681 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { | |
1682 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1683 | epfn = min_t(unsigned long, first_deferred_pfn, PFN_DOWN(epa)); | |
1684 | deferred_free_pages(nid, zid, spfn, epfn); | |
1685 | ||
1686 | if (first_deferred_pfn == epfn) | |
1687 | break; | |
1688 | } | |
1689 | pgdat->first_deferred_pfn = first_deferred_pfn; | |
1690 | pgdat_resize_unlock(pgdat, &flags); | |
1691 | ||
1692 | return nr_pages > 0; | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * deferred_grow_zone() is __init, but it is called from | |
1697 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
1698 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
1699 | * and to ensure that the function body gets unloaded. | |
1700 | */ | |
1701 | static bool __ref | |
1702 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
1703 | { | |
1704 | return deferred_grow_zone(zone, order); | |
1705 | } | |
1706 | ||
7cf91a98 | 1707 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1708 | |
1709 | void __init page_alloc_init_late(void) | |
1710 | { | |
7cf91a98 JK |
1711 | struct zone *zone; |
1712 | ||
1713 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1714 | int nid; |
1715 | ||
d3cd131d NS |
1716 | /* There will be num_node_state(N_MEMORY) threads */ |
1717 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1718 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1719 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1720 | } | |
1721 | ||
1722 | /* Block until all are initialised */ | |
d3cd131d | 1723 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 1724 | |
c9e97a19 PT |
1725 | /* |
1726 | * We initialized the rest of the deferred pages. Permanently disable | |
1727 | * on-demand struct page initialization. | |
1728 | */ | |
1729 | static_branch_disable(&deferred_pages); | |
1730 | ||
4248b0da MG |
1731 | /* Reinit limits that are based on free pages after the kernel is up */ |
1732 | files_maxfiles_init(); | |
7cf91a98 | 1733 | #endif |
3010f876 PT |
1734 | #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK |
1735 | /* Discard memblock private memory */ | |
1736 | memblock_discard(); | |
1737 | #endif | |
7cf91a98 JK |
1738 | |
1739 | for_each_populated_zone(zone) | |
1740 | set_zone_contiguous(zone); | |
7e18adb4 | 1741 | } |
7e18adb4 | 1742 | |
47118af0 | 1743 | #ifdef CONFIG_CMA |
9cf510a5 | 1744 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1745 | void __init init_cma_reserved_pageblock(struct page *page) |
1746 | { | |
1747 | unsigned i = pageblock_nr_pages; | |
1748 | struct page *p = page; | |
1749 | ||
1750 | do { | |
1751 | __ClearPageReserved(p); | |
1752 | set_page_count(p, 0); | |
d883c6cf | 1753 | } while (++p, --i); |
47118af0 | 1754 | |
47118af0 | 1755 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1756 | |
1757 | if (pageblock_order >= MAX_ORDER) { | |
1758 | i = pageblock_nr_pages; | |
1759 | p = page; | |
1760 | do { | |
1761 | set_page_refcounted(p); | |
1762 | __free_pages(p, MAX_ORDER - 1); | |
1763 | p += MAX_ORDER_NR_PAGES; | |
1764 | } while (i -= MAX_ORDER_NR_PAGES); | |
1765 | } else { | |
1766 | set_page_refcounted(page); | |
1767 | __free_pages(page, pageblock_order); | |
1768 | } | |
1769 | ||
3dcc0571 | 1770 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1771 | } |
1772 | #endif | |
1da177e4 LT |
1773 | |
1774 | /* | |
1775 | * The order of subdivision here is critical for the IO subsystem. | |
1776 | * Please do not alter this order without good reasons and regression | |
1777 | * testing. Specifically, as large blocks of memory are subdivided, | |
1778 | * the order in which smaller blocks are delivered depends on the order | |
1779 | * they're subdivided in this function. This is the primary factor | |
1780 | * influencing the order in which pages are delivered to the IO | |
1781 | * subsystem according to empirical testing, and this is also justified | |
1782 | * by considering the behavior of a buddy system containing a single | |
1783 | * large block of memory acted on by a series of small allocations. | |
1784 | * This behavior is a critical factor in sglist merging's success. | |
1785 | * | |
6d49e352 | 1786 | * -- nyc |
1da177e4 | 1787 | */ |
085cc7d5 | 1788 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1789 | int low, int high, struct free_area *area, |
1790 | int migratetype) | |
1da177e4 LT |
1791 | { |
1792 | unsigned long size = 1 << high; | |
1793 | ||
1794 | while (high > low) { | |
1795 | area--; | |
1796 | high--; | |
1797 | size >>= 1; | |
309381fe | 1798 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1799 | |
acbc15a4 JK |
1800 | /* |
1801 | * Mark as guard pages (or page), that will allow to | |
1802 | * merge back to allocator when buddy will be freed. | |
1803 | * Corresponding page table entries will not be touched, | |
1804 | * pages will stay not present in virtual address space | |
1805 | */ | |
1806 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 1807 | continue; |
acbc15a4 | 1808 | |
b2a0ac88 | 1809 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1810 | area->nr_free++; |
1811 | set_page_order(&page[size], high); | |
1812 | } | |
1da177e4 LT |
1813 | } |
1814 | ||
4e611801 | 1815 | static void check_new_page_bad(struct page *page) |
1da177e4 | 1816 | { |
4e611801 VB |
1817 | const char *bad_reason = NULL; |
1818 | unsigned long bad_flags = 0; | |
7bfec6f4 | 1819 | |
53f9263b | 1820 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1821 | bad_reason = "nonzero mapcount"; |
1822 | if (unlikely(page->mapping != NULL)) | |
1823 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1824 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1825 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1826 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1827 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1828 | bad_flags = __PG_HWPOISON; | |
e570f56c NH |
1829 | /* Don't complain about hwpoisoned pages */ |
1830 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1831 | return; | |
f4c18e6f | 1832 | } |
f0b791a3 DH |
1833 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1834 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1835 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1836 | } | |
9edad6ea JW |
1837 | #ifdef CONFIG_MEMCG |
1838 | if (unlikely(page->mem_cgroup)) | |
1839 | bad_reason = "page still charged to cgroup"; | |
1840 | #endif | |
4e611801 VB |
1841 | bad_page(page, bad_reason, bad_flags); |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * This page is about to be returned from the page allocator | |
1846 | */ | |
1847 | static inline int check_new_page(struct page *page) | |
1848 | { | |
1849 | if (likely(page_expected_state(page, | |
1850 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1851 | return 0; | |
1852 | ||
1853 | check_new_page_bad(page); | |
1854 | return 1; | |
2a7684a2 WF |
1855 | } |
1856 | ||
bd33ef36 | 1857 | static inline bool free_pages_prezeroed(void) |
1414c7f4 LA |
1858 | { |
1859 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
bd33ef36 | 1860 | page_poisoning_enabled(); |
1414c7f4 LA |
1861 | } |
1862 | ||
479f854a MG |
1863 | #ifdef CONFIG_DEBUG_VM |
1864 | static bool check_pcp_refill(struct page *page) | |
1865 | { | |
1866 | return false; | |
1867 | } | |
1868 | ||
1869 | static bool check_new_pcp(struct page *page) | |
1870 | { | |
1871 | return check_new_page(page); | |
1872 | } | |
1873 | #else | |
1874 | static bool check_pcp_refill(struct page *page) | |
1875 | { | |
1876 | return check_new_page(page); | |
1877 | } | |
1878 | static bool check_new_pcp(struct page *page) | |
1879 | { | |
1880 | return false; | |
1881 | } | |
1882 | #endif /* CONFIG_DEBUG_VM */ | |
1883 | ||
1884 | static bool check_new_pages(struct page *page, unsigned int order) | |
1885 | { | |
1886 | int i; | |
1887 | for (i = 0; i < (1 << order); i++) { | |
1888 | struct page *p = page + i; | |
1889 | ||
1890 | if (unlikely(check_new_page(p))) | |
1891 | return true; | |
1892 | } | |
1893 | ||
1894 | return false; | |
1895 | } | |
1896 | ||
46f24fd8 JK |
1897 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1898 | gfp_t gfp_flags) | |
1899 | { | |
1900 | set_page_private(page, 0); | |
1901 | set_page_refcounted(page); | |
1902 | ||
1903 | arch_alloc_page(page, order); | |
1904 | kernel_map_pages(page, 1 << order, 1); | |
1905 | kernel_poison_pages(page, 1 << order, 1); | |
1906 | kasan_alloc_pages(page, order); | |
1907 | set_page_owner(page, order, gfp_flags); | |
1908 | } | |
1909 | ||
479f854a | 1910 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1911 | unsigned int alloc_flags) |
2a7684a2 WF |
1912 | { |
1913 | int i; | |
689bcebf | 1914 | |
46f24fd8 | 1915 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1916 | |
bd33ef36 | 1917 | if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1918 | for (i = 0; i < (1 << order); i++) |
1919 | clear_highpage(page + i); | |
17cf4406 NP |
1920 | |
1921 | if (order && (gfp_flags & __GFP_COMP)) | |
1922 | prep_compound_page(page, order); | |
1923 | ||
75379191 | 1924 | /* |
2f064f34 | 1925 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1926 | * allocate the page. The expectation is that the caller is taking |
1927 | * steps that will free more memory. The caller should avoid the page | |
1928 | * being used for !PFMEMALLOC purposes. | |
1929 | */ | |
2f064f34 MH |
1930 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1931 | set_page_pfmemalloc(page); | |
1932 | else | |
1933 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1934 | } |
1935 | ||
56fd56b8 MG |
1936 | /* |
1937 | * Go through the free lists for the given migratetype and remove | |
1938 | * the smallest available page from the freelists | |
1939 | */ | |
85ccc8fa | 1940 | static __always_inline |
728ec980 | 1941 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
1942 | int migratetype) |
1943 | { | |
1944 | unsigned int current_order; | |
b8af2941 | 1945 | struct free_area *area; |
56fd56b8 MG |
1946 | struct page *page; |
1947 | ||
1948 | /* Find a page of the appropriate size in the preferred list */ | |
1949 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1950 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1951 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1952 | struct page, lru); |
a16601c5 GT |
1953 | if (!page) |
1954 | continue; | |
56fd56b8 MG |
1955 | list_del(&page->lru); |
1956 | rmv_page_order(page); | |
1957 | area->nr_free--; | |
56fd56b8 | 1958 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1959 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1960 | return page; |
1961 | } | |
1962 | ||
1963 | return NULL; | |
1964 | } | |
1965 | ||
1966 | ||
b2a0ac88 MG |
1967 | /* |
1968 | * This array describes the order lists are fallen back to when | |
1969 | * the free lists for the desirable migrate type are depleted | |
1970 | */ | |
47118af0 | 1971 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1972 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1973 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1974 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1975 | #ifdef CONFIG_CMA |
974a786e | 1976 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1977 | #endif |
194159fb | 1978 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1979 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1980 | #endif |
b2a0ac88 MG |
1981 | }; |
1982 | ||
dc67647b | 1983 | #ifdef CONFIG_CMA |
85ccc8fa | 1984 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
1985 | unsigned int order) |
1986 | { | |
1987 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1988 | } | |
1989 | #else | |
1990 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1991 | unsigned int order) { return NULL; } | |
1992 | #endif | |
1993 | ||
c361be55 MG |
1994 | /* |
1995 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1996 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1997 | * boundary. If alignment is required, use move_freepages_block() |
1998 | */ | |
02aa0cdd | 1999 | static int move_freepages(struct zone *zone, |
b69a7288 | 2000 | struct page *start_page, struct page *end_page, |
02aa0cdd | 2001 | int migratetype, int *num_movable) |
c361be55 MG |
2002 | { |
2003 | struct page *page; | |
d00181b9 | 2004 | unsigned int order; |
d100313f | 2005 | int pages_moved = 0; |
c361be55 MG |
2006 | |
2007 | #ifndef CONFIG_HOLES_IN_ZONE | |
2008 | /* | |
2009 | * page_zone is not safe to call in this context when | |
2010 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
2011 | * anyway as we check zone boundaries in move_freepages_block(). | |
2012 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 2013 | * grouping pages by mobility |
c361be55 | 2014 | */ |
3e04040d AB |
2015 | VM_BUG_ON(pfn_valid(page_to_pfn(start_page)) && |
2016 | pfn_valid(page_to_pfn(end_page)) && | |
2017 | page_zone(start_page) != page_zone(end_page)); | |
c361be55 MG |
2018 | #endif |
2019 | ||
02aa0cdd VB |
2020 | if (num_movable) |
2021 | *num_movable = 0; | |
2022 | ||
c361be55 MG |
2023 | for (page = start_page; page <= end_page;) { |
2024 | if (!pfn_valid_within(page_to_pfn(page))) { | |
2025 | page++; | |
2026 | continue; | |
2027 | } | |
2028 | ||
f073bdc5 AB |
2029 | /* Make sure we are not inadvertently changing nodes */ |
2030 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2031 | ||
c361be55 | 2032 | if (!PageBuddy(page)) { |
02aa0cdd VB |
2033 | /* |
2034 | * We assume that pages that could be isolated for | |
2035 | * migration are movable. But we don't actually try | |
2036 | * isolating, as that would be expensive. | |
2037 | */ | |
2038 | if (num_movable && | |
2039 | (PageLRU(page) || __PageMovable(page))) | |
2040 | (*num_movable)++; | |
2041 | ||
c361be55 MG |
2042 | page++; |
2043 | continue; | |
2044 | } | |
2045 | ||
2046 | order = page_order(page); | |
84be48d8 KS |
2047 | list_move(&page->lru, |
2048 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 2049 | page += 1 << order; |
d100313f | 2050 | pages_moved += 1 << order; |
c361be55 MG |
2051 | } |
2052 | ||
d100313f | 2053 | return pages_moved; |
c361be55 MG |
2054 | } |
2055 | ||
ee6f509c | 2056 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2057 | int migratetype, int *num_movable) |
c361be55 MG |
2058 | { |
2059 | unsigned long start_pfn, end_pfn; | |
2060 | struct page *start_page, *end_page; | |
2061 | ||
2062 | start_pfn = page_to_pfn(page); | |
d9c23400 | 2063 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 2064 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
2065 | end_page = start_page + pageblock_nr_pages - 1; |
2066 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
2067 | |
2068 | /* Do not cross zone boundaries */ | |
108bcc96 | 2069 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 2070 | start_page = page; |
108bcc96 | 2071 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2072 | return 0; |
2073 | ||
02aa0cdd VB |
2074 | return move_freepages(zone, start_page, end_page, migratetype, |
2075 | num_movable); | |
c361be55 MG |
2076 | } |
2077 | ||
2f66a68f MG |
2078 | static void change_pageblock_range(struct page *pageblock_page, |
2079 | int start_order, int migratetype) | |
2080 | { | |
2081 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2082 | ||
2083 | while (nr_pageblocks--) { | |
2084 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2085 | pageblock_page += pageblock_nr_pages; | |
2086 | } | |
2087 | } | |
2088 | ||
fef903ef | 2089 | /* |
9c0415eb VB |
2090 | * When we are falling back to another migratetype during allocation, try to |
2091 | * steal extra free pages from the same pageblocks to satisfy further | |
2092 | * allocations, instead of polluting multiple pageblocks. | |
2093 | * | |
2094 | * If we are stealing a relatively large buddy page, it is likely there will | |
2095 | * be more free pages in the pageblock, so try to steal them all. For | |
2096 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2097 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2098 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2099 | * pageblocks. | |
fef903ef | 2100 | */ |
4eb7dce6 JK |
2101 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2102 | { | |
2103 | /* | |
2104 | * Leaving this order check is intended, although there is | |
2105 | * relaxed order check in next check. The reason is that | |
2106 | * we can actually steal whole pageblock if this condition met, | |
2107 | * but, below check doesn't guarantee it and that is just heuristic | |
2108 | * so could be changed anytime. | |
2109 | */ | |
2110 | if (order >= pageblock_order) | |
2111 | return true; | |
2112 | ||
2113 | if (order >= pageblock_order / 2 || | |
2114 | start_mt == MIGRATE_RECLAIMABLE || | |
2115 | start_mt == MIGRATE_UNMOVABLE || | |
2116 | page_group_by_mobility_disabled) | |
2117 | return true; | |
2118 | ||
2119 | return false; | |
2120 | } | |
2121 | ||
2122 | /* | |
2123 | * This function implements actual steal behaviour. If order is large enough, | |
2124 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2125 | * pageblock to our migratetype and determine how many already-allocated pages |
2126 | * are there in the pageblock with a compatible migratetype. If at least half | |
2127 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2128 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2129 | */ |
2130 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
3bc48f96 | 2131 | int start_type, bool whole_block) |
fef903ef | 2132 | { |
d00181b9 | 2133 | unsigned int current_order = page_order(page); |
3bc48f96 | 2134 | struct free_area *area; |
02aa0cdd VB |
2135 | int free_pages, movable_pages, alike_pages; |
2136 | int old_block_type; | |
2137 | ||
2138 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2139 | |
3bc48f96 VB |
2140 | /* |
2141 | * This can happen due to races and we want to prevent broken | |
2142 | * highatomic accounting. | |
2143 | */ | |
02aa0cdd | 2144 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2145 | goto single_page; |
2146 | ||
fef903ef SB |
2147 | /* Take ownership for orders >= pageblock_order */ |
2148 | if (current_order >= pageblock_order) { | |
2149 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2150 | goto single_page; |
fef903ef SB |
2151 | } |
2152 | ||
3bc48f96 VB |
2153 | /* We are not allowed to try stealing from the whole block */ |
2154 | if (!whole_block) | |
2155 | goto single_page; | |
2156 | ||
02aa0cdd VB |
2157 | free_pages = move_freepages_block(zone, page, start_type, |
2158 | &movable_pages); | |
2159 | /* | |
2160 | * Determine how many pages are compatible with our allocation. | |
2161 | * For movable allocation, it's the number of movable pages which | |
2162 | * we just obtained. For other types it's a bit more tricky. | |
2163 | */ | |
2164 | if (start_type == MIGRATE_MOVABLE) { | |
2165 | alike_pages = movable_pages; | |
2166 | } else { | |
2167 | /* | |
2168 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2169 | * to MOVABLE pageblock, consider all non-movable pages as | |
2170 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2171 | * vice versa, be conservative since we can't distinguish the | |
2172 | * exact migratetype of non-movable pages. | |
2173 | */ | |
2174 | if (old_block_type == MIGRATE_MOVABLE) | |
2175 | alike_pages = pageblock_nr_pages | |
2176 | - (free_pages + movable_pages); | |
2177 | else | |
2178 | alike_pages = 0; | |
2179 | } | |
2180 | ||
3bc48f96 | 2181 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2182 | if (!free_pages) |
3bc48f96 | 2183 | goto single_page; |
fef903ef | 2184 | |
02aa0cdd VB |
2185 | /* |
2186 | * If a sufficient number of pages in the block are either free or of | |
2187 | * comparable migratability as our allocation, claim the whole block. | |
2188 | */ | |
2189 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2190 | page_group_by_mobility_disabled) |
2191 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2192 | |
2193 | return; | |
2194 | ||
2195 | single_page: | |
2196 | area = &zone->free_area[current_order]; | |
2197 | list_move(&page->lru, &area->free_list[start_type]); | |
4eb7dce6 JK |
2198 | } |
2199 | ||
2149cdae JK |
2200 | /* |
2201 | * Check whether there is a suitable fallback freepage with requested order. | |
2202 | * If only_stealable is true, this function returns fallback_mt only if | |
2203 | * we can steal other freepages all together. This would help to reduce | |
2204 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2205 | */ | |
2206 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2207 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2208 | { |
2209 | int i; | |
2210 | int fallback_mt; | |
2211 | ||
2212 | if (area->nr_free == 0) | |
2213 | return -1; | |
2214 | ||
2215 | *can_steal = false; | |
2216 | for (i = 0;; i++) { | |
2217 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2218 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2219 | break; |
2220 | ||
2221 | if (list_empty(&area->free_list[fallback_mt])) | |
2222 | continue; | |
fef903ef | 2223 | |
4eb7dce6 JK |
2224 | if (can_steal_fallback(order, migratetype)) |
2225 | *can_steal = true; | |
2226 | ||
2149cdae JK |
2227 | if (!only_stealable) |
2228 | return fallback_mt; | |
2229 | ||
2230 | if (*can_steal) | |
2231 | return fallback_mt; | |
fef903ef | 2232 | } |
4eb7dce6 JK |
2233 | |
2234 | return -1; | |
fef903ef SB |
2235 | } |
2236 | ||
0aaa29a5 MG |
2237 | /* |
2238 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2239 | * there are no empty page blocks that contain a page with a suitable order | |
2240 | */ | |
2241 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2242 | unsigned int alloc_order) | |
2243 | { | |
2244 | int mt; | |
2245 | unsigned long max_managed, flags; | |
2246 | ||
2247 | /* | |
2248 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2249 | * Check is race-prone but harmless. | |
2250 | */ | |
2251 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
2252 | if (zone->nr_reserved_highatomic >= max_managed) | |
2253 | return; | |
2254 | ||
2255 | spin_lock_irqsave(&zone->lock, flags); | |
2256 | ||
2257 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2258 | if (zone->nr_reserved_highatomic >= max_managed) | |
2259 | goto out_unlock; | |
2260 | ||
2261 | /* Yoink! */ | |
2262 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2263 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2264 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2265 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2266 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2267 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2268 | } |
2269 | ||
2270 | out_unlock: | |
2271 | spin_unlock_irqrestore(&zone->lock, flags); | |
2272 | } | |
2273 | ||
2274 | /* | |
2275 | * Used when an allocation is about to fail under memory pressure. This | |
2276 | * potentially hurts the reliability of high-order allocations when under | |
2277 | * intense memory pressure but failed atomic allocations should be easier | |
2278 | * to recover from than an OOM. | |
29fac03b MK |
2279 | * |
2280 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2281 | * pageblock is exhausted. | |
0aaa29a5 | 2282 | */ |
29fac03b MK |
2283 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2284 | bool force) | |
0aaa29a5 MG |
2285 | { |
2286 | struct zonelist *zonelist = ac->zonelist; | |
2287 | unsigned long flags; | |
2288 | struct zoneref *z; | |
2289 | struct zone *zone; | |
2290 | struct page *page; | |
2291 | int order; | |
04c8716f | 2292 | bool ret; |
0aaa29a5 MG |
2293 | |
2294 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2295 | ac->nodemask) { | |
29fac03b MK |
2296 | /* |
2297 | * Preserve at least one pageblock unless memory pressure | |
2298 | * is really high. | |
2299 | */ | |
2300 | if (!force && zone->nr_reserved_highatomic <= | |
2301 | pageblock_nr_pages) | |
0aaa29a5 MG |
2302 | continue; |
2303 | ||
2304 | spin_lock_irqsave(&zone->lock, flags); | |
2305 | for (order = 0; order < MAX_ORDER; order++) { | |
2306 | struct free_area *area = &(zone->free_area[order]); | |
2307 | ||
a16601c5 GT |
2308 | page = list_first_entry_or_null( |
2309 | &area->free_list[MIGRATE_HIGHATOMIC], | |
2310 | struct page, lru); | |
2311 | if (!page) | |
0aaa29a5 MG |
2312 | continue; |
2313 | ||
0aaa29a5 | 2314 | /* |
4855e4a7 MK |
2315 | * In page freeing path, migratetype change is racy so |
2316 | * we can counter several free pages in a pageblock | |
2317 | * in this loop althoug we changed the pageblock type | |
2318 | * from highatomic to ac->migratetype. So we should | |
2319 | * adjust the count once. | |
0aaa29a5 | 2320 | */ |
a6ffdc07 | 2321 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2322 | /* |
2323 | * It should never happen but changes to | |
2324 | * locking could inadvertently allow a per-cpu | |
2325 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2326 | * while unreserving so be safe and watch for | |
2327 | * underflows. | |
2328 | */ | |
2329 | zone->nr_reserved_highatomic -= min( | |
2330 | pageblock_nr_pages, | |
2331 | zone->nr_reserved_highatomic); | |
2332 | } | |
0aaa29a5 MG |
2333 | |
2334 | /* | |
2335 | * Convert to ac->migratetype and avoid the normal | |
2336 | * pageblock stealing heuristics. Minimally, the caller | |
2337 | * is doing the work and needs the pages. More | |
2338 | * importantly, if the block was always converted to | |
2339 | * MIGRATE_UNMOVABLE or another type then the number | |
2340 | * of pageblocks that cannot be completely freed | |
2341 | * may increase. | |
2342 | */ | |
2343 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2344 | ret = move_freepages_block(zone, page, ac->migratetype, |
2345 | NULL); | |
29fac03b MK |
2346 | if (ret) { |
2347 | spin_unlock_irqrestore(&zone->lock, flags); | |
2348 | return ret; | |
2349 | } | |
0aaa29a5 MG |
2350 | } |
2351 | spin_unlock_irqrestore(&zone->lock, flags); | |
2352 | } | |
04c8716f MK |
2353 | |
2354 | return false; | |
0aaa29a5 MG |
2355 | } |
2356 | ||
3bc48f96 VB |
2357 | /* |
2358 | * Try finding a free buddy page on the fallback list and put it on the free | |
2359 | * list of requested migratetype, possibly along with other pages from the same | |
2360 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2361 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2362 | * |
2363 | * The use of signed ints for order and current_order is a deliberate | |
2364 | * deviation from the rest of this file, to make the for loop | |
2365 | * condition simpler. | |
3bc48f96 | 2366 | */ |
85ccc8fa | 2367 | static __always_inline bool |
b002529d | 2368 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) |
b2a0ac88 | 2369 | { |
b8af2941 | 2370 | struct free_area *area; |
b002529d | 2371 | int current_order; |
b2a0ac88 | 2372 | struct page *page; |
4eb7dce6 JK |
2373 | int fallback_mt; |
2374 | bool can_steal; | |
b2a0ac88 | 2375 | |
7a8f58f3 VB |
2376 | /* |
2377 | * Find the largest available free page in the other list. This roughly | |
2378 | * approximates finding the pageblock with the most free pages, which | |
2379 | * would be too costly to do exactly. | |
2380 | */ | |
b002529d | 2381 | for (current_order = MAX_ORDER - 1; current_order >= order; |
7aeb09f9 | 2382 | --current_order) { |
4eb7dce6 JK |
2383 | area = &(zone->free_area[current_order]); |
2384 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2385 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2386 | if (fallback_mt == -1) |
2387 | continue; | |
b2a0ac88 | 2388 | |
7a8f58f3 VB |
2389 | /* |
2390 | * We cannot steal all free pages from the pageblock and the | |
2391 | * requested migratetype is movable. In that case it's better to | |
2392 | * steal and split the smallest available page instead of the | |
2393 | * largest available page, because even if the next movable | |
2394 | * allocation falls back into a different pageblock than this | |
2395 | * one, it won't cause permanent fragmentation. | |
2396 | */ | |
2397 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2398 | && current_order > order) | |
2399 | goto find_smallest; | |
b2a0ac88 | 2400 | |
7a8f58f3 VB |
2401 | goto do_steal; |
2402 | } | |
e0fff1bd | 2403 | |
7a8f58f3 | 2404 | return false; |
e0fff1bd | 2405 | |
7a8f58f3 VB |
2406 | find_smallest: |
2407 | for (current_order = order; current_order < MAX_ORDER; | |
2408 | current_order++) { | |
2409 | area = &(zone->free_area[current_order]); | |
2410 | fallback_mt = find_suitable_fallback(area, current_order, | |
2411 | start_migratetype, false, &can_steal); | |
2412 | if (fallback_mt != -1) | |
2413 | break; | |
b2a0ac88 MG |
2414 | } |
2415 | ||
7a8f58f3 VB |
2416 | /* |
2417 | * This should not happen - we already found a suitable fallback | |
2418 | * when looking for the largest page. | |
2419 | */ | |
2420 | VM_BUG_ON(current_order == MAX_ORDER); | |
2421 | ||
2422 | do_steal: | |
2423 | page = list_first_entry(&area->free_list[fallback_mt], | |
2424 | struct page, lru); | |
2425 | ||
2426 | steal_suitable_fallback(zone, page, start_migratetype, can_steal); | |
2427 | ||
2428 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2429 | start_migratetype, fallback_mt); | |
2430 | ||
2431 | return true; | |
2432 | ||
b2a0ac88 MG |
2433 | } |
2434 | ||
56fd56b8 | 2435 | /* |
1da177e4 LT |
2436 | * Do the hard work of removing an element from the buddy allocator. |
2437 | * Call me with the zone->lock already held. | |
2438 | */ | |
85ccc8fa AL |
2439 | static __always_inline struct page * |
2440 | __rmqueue(struct zone *zone, unsigned int order, int migratetype) | |
1da177e4 | 2441 | { |
1da177e4 LT |
2442 | struct page *page; |
2443 | ||
3bc48f96 | 2444 | retry: |
56fd56b8 | 2445 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2446 | if (unlikely(!page)) { |
dc67647b JK |
2447 | if (migratetype == MIGRATE_MOVABLE) |
2448 | page = __rmqueue_cma_fallback(zone, order); | |
2449 | ||
3bc48f96 VB |
2450 | if (!page && __rmqueue_fallback(zone, order, migratetype)) |
2451 | goto retry; | |
728ec980 MG |
2452 | } |
2453 | ||
0d3d062a | 2454 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2455 | return page; |
1da177e4 LT |
2456 | } |
2457 | ||
5f63b720 | 2458 | /* |
1da177e4 LT |
2459 | * Obtain a specified number of elements from the buddy allocator, all under |
2460 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2461 | * Returns the number of new pages which were placed at *list. | |
2462 | */ | |
5f63b720 | 2463 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2464 | unsigned long count, struct list_head *list, |
453f85d4 | 2465 | int migratetype) |
1da177e4 | 2466 | { |
a6de734b | 2467 | int i, alloced = 0; |
5f63b720 | 2468 | |
d34b0733 | 2469 | spin_lock(&zone->lock); |
1da177e4 | 2470 | for (i = 0; i < count; ++i) { |
6ac0206b | 2471 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 2472 | if (unlikely(page == NULL)) |
1da177e4 | 2473 | break; |
81eabcbe | 2474 | |
479f854a MG |
2475 | if (unlikely(check_pcp_refill(page))) |
2476 | continue; | |
2477 | ||
81eabcbe | 2478 | /* |
0fac3ba5 VB |
2479 | * Split buddy pages returned by expand() are received here in |
2480 | * physical page order. The page is added to the tail of | |
2481 | * caller's list. From the callers perspective, the linked list | |
2482 | * is ordered by page number under some conditions. This is | |
2483 | * useful for IO devices that can forward direction from the | |
2484 | * head, thus also in the physical page order. This is useful | |
2485 | * for IO devices that can merge IO requests if the physical | |
2486 | * pages are ordered properly. | |
81eabcbe | 2487 | */ |
0fac3ba5 | 2488 | list_add_tail(&page->lru, list); |
a6de734b | 2489 | alloced++; |
bb14c2c7 | 2490 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2491 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2492 | -(1 << order)); | |
1da177e4 | 2493 | } |
a6de734b MG |
2494 | |
2495 | /* | |
2496 | * i pages were removed from the buddy list even if some leak due | |
2497 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2498 | * on i. Do not confuse with 'alloced' which is the number of | |
2499 | * pages added to the pcp list. | |
2500 | */ | |
f2260e6b | 2501 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2502 | spin_unlock(&zone->lock); |
a6de734b | 2503 | return alloced; |
1da177e4 LT |
2504 | } |
2505 | ||
4ae7c039 | 2506 | #ifdef CONFIG_NUMA |
8fce4d8e | 2507 | /* |
4037d452 CL |
2508 | * Called from the vmstat counter updater to drain pagesets of this |
2509 | * currently executing processor on remote nodes after they have | |
2510 | * expired. | |
2511 | * | |
879336c3 CL |
2512 | * Note that this function must be called with the thread pinned to |
2513 | * a single processor. | |
8fce4d8e | 2514 | */ |
4037d452 | 2515 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2516 | { |
4ae7c039 | 2517 | unsigned long flags; |
7be12fc9 | 2518 | int to_drain, batch; |
4ae7c039 | 2519 | |
4037d452 | 2520 | local_irq_save(flags); |
4db0c3c2 | 2521 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2522 | to_drain = min(pcp->count, batch); |
77ba9062 | 2523 | if (to_drain > 0) |
2a13515c | 2524 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 2525 | local_irq_restore(flags); |
4ae7c039 CL |
2526 | } |
2527 | #endif | |
2528 | ||
9f8f2172 | 2529 | /* |
93481ff0 | 2530 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2531 | * |
2532 | * The processor must either be the current processor and the | |
2533 | * thread pinned to the current processor or a processor that | |
2534 | * is not online. | |
2535 | */ | |
93481ff0 | 2536 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2537 | { |
c54ad30c | 2538 | unsigned long flags; |
93481ff0 VB |
2539 | struct per_cpu_pageset *pset; |
2540 | struct per_cpu_pages *pcp; | |
1da177e4 | 2541 | |
93481ff0 VB |
2542 | local_irq_save(flags); |
2543 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2544 | |
93481ff0 | 2545 | pcp = &pset->pcp; |
77ba9062 | 2546 | if (pcp->count) |
93481ff0 | 2547 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
2548 | local_irq_restore(flags); |
2549 | } | |
3dfa5721 | 2550 | |
93481ff0 VB |
2551 | /* |
2552 | * Drain pcplists of all zones on the indicated processor. | |
2553 | * | |
2554 | * The processor must either be the current processor and the | |
2555 | * thread pinned to the current processor or a processor that | |
2556 | * is not online. | |
2557 | */ | |
2558 | static void drain_pages(unsigned int cpu) | |
2559 | { | |
2560 | struct zone *zone; | |
2561 | ||
2562 | for_each_populated_zone(zone) { | |
2563 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2564 | } |
2565 | } | |
1da177e4 | 2566 | |
9f8f2172 CL |
2567 | /* |
2568 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2569 | * |
2570 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2571 | * the single zone's pages. | |
9f8f2172 | 2572 | */ |
93481ff0 | 2573 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2574 | { |
93481ff0 VB |
2575 | int cpu = smp_processor_id(); |
2576 | ||
2577 | if (zone) | |
2578 | drain_pages_zone(cpu, zone); | |
2579 | else | |
2580 | drain_pages(cpu); | |
9f8f2172 CL |
2581 | } |
2582 | ||
0ccce3b9 MG |
2583 | static void drain_local_pages_wq(struct work_struct *work) |
2584 | { | |
a459eeb7 MH |
2585 | /* |
2586 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2587 | * we can race with cpu offline when the WQ can move this from | |
2588 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2589 | * cpu which is allright but we also have to make sure to not move to | |
2590 | * a different one. | |
2591 | */ | |
2592 | preempt_disable(); | |
0ccce3b9 | 2593 | drain_local_pages(NULL); |
a459eeb7 | 2594 | preempt_enable(); |
0ccce3b9 MG |
2595 | } |
2596 | ||
9f8f2172 | 2597 | /* |
74046494 GBY |
2598 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2599 | * | |
93481ff0 VB |
2600 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2601 | * | |
0ccce3b9 | 2602 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2603 | */ |
93481ff0 | 2604 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2605 | { |
74046494 | 2606 | int cpu; |
74046494 GBY |
2607 | |
2608 | /* | |
2609 | * Allocate in the BSS so we wont require allocation in | |
2610 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2611 | */ | |
2612 | static cpumask_t cpus_with_pcps; | |
2613 | ||
ce612879 MH |
2614 | /* |
2615 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2616 | * initialized. | |
2617 | */ | |
2618 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2619 | return; | |
2620 | ||
bd233f53 MG |
2621 | /* |
2622 | * Do not drain if one is already in progress unless it's specific to | |
2623 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2624 | * the drain to be complete when the call returns. | |
2625 | */ | |
2626 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2627 | if (!zone) | |
2628 | return; | |
2629 | mutex_lock(&pcpu_drain_mutex); | |
2630 | } | |
0ccce3b9 | 2631 | |
74046494 GBY |
2632 | /* |
2633 | * We don't care about racing with CPU hotplug event | |
2634 | * as offline notification will cause the notified | |
2635 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2636 | * disables preemption as part of its processing | |
2637 | */ | |
2638 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2639 | struct per_cpu_pageset *pcp; |
2640 | struct zone *z; | |
74046494 | 2641 | bool has_pcps = false; |
93481ff0 VB |
2642 | |
2643 | if (zone) { | |
74046494 | 2644 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2645 | if (pcp->pcp.count) |
74046494 | 2646 | has_pcps = true; |
93481ff0 VB |
2647 | } else { |
2648 | for_each_populated_zone(z) { | |
2649 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2650 | if (pcp->pcp.count) { | |
2651 | has_pcps = true; | |
2652 | break; | |
2653 | } | |
74046494 GBY |
2654 | } |
2655 | } | |
93481ff0 | 2656 | |
74046494 GBY |
2657 | if (has_pcps) |
2658 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2659 | else | |
2660 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2661 | } | |
0ccce3b9 | 2662 | |
bd233f53 MG |
2663 | for_each_cpu(cpu, &cpus_with_pcps) { |
2664 | struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu); | |
2665 | INIT_WORK(work, drain_local_pages_wq); | |
ce612879 | 2666 | queue_work_on(cpu, mm_percpu_wq, work); |
0ccce3b9 | 2667 | } |
bd233f53 MG |
2668 | for_each_cpu(cpu, &cpus_with_pcps) |
2669 | flush_work(per_cpu_ptr(&pcpu_drain, cpu)); | |
2670 | ||
2671 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2672 | } |
2673 | ||
296699de | 2674 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 2675 | |
556b969a CY |
2676 | /* |
2677 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
2678 | */ | |
2679 | #define WD_PAGE_COUNT (128*1024) | |
2680 | ||
1da177e4 LT |
2681 | void mark_free_pages(struct zone *zone) |
2682 | { | |
556b969a | 2683 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 2684 | unsigned long flags; |
7aeb09f9 | 2685 | unsigned int order, t; |
86760a2c | 2686 | struct page *page; |
1da177e4 | 2687 | |
8080fc03 | 2688 | if (zone_is_empty(zone)) |
1da177e4 LT |
2689 | return; |
2690 | ||
2691 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2692 | |
108bcc96 | 2693 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2694 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2695 | if (pfn_valid(pfn)) { | |
86760a2c | 2696 | page = pfn_to_page(pfn); |
ba6b0979 | 2697 | |
556b969a CY |
2698 | if (!--page_count) { |
2699 | touch_nmi_watchdog(); | |
2700 | page_count = WD_PAGE_COUNT; | |
2701 | } | |
2702 | ||
ba6b0979 JK |
2703 | if (page_zone(page) != zone) |
2704 | continue; | |
2705 | ||
7be98234 RW |
2706 | if (!swsusp_page_is_forbidden(page)) |
2707 | swsusp_unset_page_free(page); | |
f623f0db | 2708 | } |
1da177e4 | 2709 | |
b2a0ac88 | 2710 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2711 | list_for_each_entry(page, |
2712 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2713 | unsigned long i; |
1da177e4 | 2714 | |
86760a2c | 2715 | pfn = page_to_pfn(page); |
556b969a CY |
2716 | for (i = 0; i < (1UL << order); i++) { |
2717 | if (!--page_count) { | |
2718 | touch_nmi_watchdog(); | |
2719 | page_count = WD_PAGE_COUNT; | |
2720 | } | |
7be98234 | 2721 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 2722 | } |
f623f0db | 2723 | } |
b2a0ac88 | 2724 | } |
1da177e4 LT |
2725 | spin_unlock_irqrestore(&zone->lock, flags); |
2726 | } | |
e2c55dc8 | 2727 | #endif /* CONFIG_PM */ |
1da177e4 | 2728 | |
2d4894b5 | 2729 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 2730 | { |
5f8dcc21 | 2731 | int migratetype; |
1da177e4 | 2732 | |
4db7548c | 2733 | if (!free_pcp_prepare(page)) |
9cca35d4 | 2734 | return false; |
689bcebf | 2735 | |
dc4b0caf | 2736 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2737 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
2738 | return true; |
2739 | } | |
2740 | ||
2d4894b5 | 2741 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
2742 | { |
2743 | struct zone *zone = page_zone(page); | |
2744 | struct per_cpu_pages *pcp; | |
2745 | int migratetype; | |
2746 | ||
2747 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 2748 | __count_vm_event(PGFREE); |
da456f14 | 2749 | |
5f8dcc21 MG |
2750 | /* |
2751 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2752 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 2753 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
2754 | * areas back if necessary. Otherwise, we may have to free |
2755 | * excessively into the page allocator | |
2756 | */ | |
2757 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2758 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2759 | free_one_page(zone, page, pfn, 0, migratetype); |
9cca35d4 | 2760 | return; |
5f8dcc21 MG |
2761 | } |
2762 | migratetype = MIGRATE_MOVABLE; | |
2763 | } | |
2764 | ||
99dcc3e5 | 2765 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 2766 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 2767 | pcp->count++; |
48db57f8 | 2768 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2769 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb | 2770 | free_pcppages_bulk(zone, batch, pcp); |
48db57f8 | 2771 | } |
9cca35d4 | 2772 | } |
5f8dcc21 | 2773 | |
9cca35d4 MG |
2774 | /* |
2775 | * Free a 0-order page | |
9cca35d4 | 2776 | */ |
2d4894b5 | 2777 | void free_unref_page(struct page *page) |
9cca35d4 MG |
2778 | { |
2779 | unsigned long flags; | |
2780 | unsigned long pfn = page_to_pfn(page); | |
2781 | ||
2d4894b5 | 2782 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
2783 | return; |
2784 | ||
2785 | local_irq_save(flags); | |
2d4894b5 | 2786 | free_unref_page_commit(page, pfn); |
d34b0733 | 2787 | local_irq_restore(flags); |
1da177e4 LT |
2788 | } |
2789 | ||
cc59850e KK |
2790 | /* |
2791 | * Free a list of 0-order pages | |
2792 | */ | |
2d4894b5 | 2793 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
2794 | { |
2795 | struct page *page, *next; | |
9cca35d4 | 2796 | unsigned long flags, pfn; |
c24ad77d | 2797 | int batch_count = 0; |
9cca35d4 MG |
2798 | |
2799 | /* Prepare pages for freeing */ | |
2800 | list_for_each_entry_safe(page, next, list, lru) { | |
2801 | pfn = page_to_pfn(page); | |
2d4894b5 | 2802 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
2803 | list_del(&page->lru); |
2804 | set_page_private(page, pfn); | |
2805 | } | |
cc59850e | 2806 | |
9cca35d4 | 2807 | local_irq_save(flags); |
cc59850e | 2808 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
2809 | unsigned long pfn = page_private(page); |
2810 | ||
2811 | set_page_private(page, 0); | |
2d4894b5 MG |
2812 | trace_mm_page_free_batched(page); |
2813 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
2814 | |
2815 | /* | |
2816 | * Guard against excessive IRQ disabled times when we get | |
2817 | * a large list of pages to free. | |
2818 | */ | |
2819 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
2820 | local_irq_restore(flags); | |
2821 | batch_count = 0; | |
2822 | local_irq_save(flags); | |
2823 | } | |
cc59850e | 2824 | } |
9cca35d4 | 2825 | local_irq_restore(flags); |
cc59850e KK |
2826 | } |
2827 | ||
8dfcc9ba NP |
2828 | /* |
2829 | * split_page takes a non-compound higher-order page, and splits it into | |
2830 | * n (1<<order) sub-pages: page[0..n] | |
2831 | * Each sub-page must be freed individually. | |
2832 | * | |
2833 | * Note: this is probably too low level an operation for use in drivers. | |
2834 | * Please consult with lkml before using this in your driver. | |
2835 | */ | |
2836 | void split_page(struct page *page, unsigned int order) | |
2837 | { | |
2838 | int i; | |
2839 | ||
309381fe SL |
2840 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2841 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 2842 | |
a9627bc5 | 2843 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2844 | set_page_refcounted(page + i); |
a9627bc5 | 2845 | split_page_owner(page, order); |
8dfcc9ba | 2846 | } |
5853ff23 | 2847 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2848 | |
3c605096 | 2849 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2850 | { |
748446bb MG |
2851 | unsigned long watermark; |
2852 | struct zone *zone; | |
2139cbe6 | 2853 | int mt; |
748446bb MG |
2854 | |
2855 | BUG_ON(!PageBuddy(page)); | |
2856 | ||
2857 | zone = page_zone(page); | |
2e30abd1 | 2858 | mt = get_pageblock_migratetype(page); |
748446bb | 2859 | |
194159fb | 2860 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
2861 | /* |
2862 | * Obey watermarks as if the page was being allocated. We can | |
2863 | * emulate a high-order watermark check with a raised order-0 | |
2864 | * watermark, because we already know our high-order page | |
2865 | * exists. | |
2866 | */ | |
2867 | watermark = min_wmark_pages(zone) + (1UL << order); | |
d883c6cf | 2868 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
2869 | return 0; |
2870 | ||
8fb74b9f | 2871 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2872 | } |
748446bb MG |
2873 | |
2874 | /* Remove page from free list */ | |
2875 | list_del(&page->lru); | |
2876 | zone->free_area[order].nr_free--; | |
2877 | rmv_page_order(page); | |
2139cbe6 | 2878 | |
400bc7fd | 2879 | /* |
2880 | * Set the pageblock if the isolated page is at least half of a | |
2881 | * pageblock | |
2882 | */ | |
748446bb MG |
2883 | if (order >= pageblock_order - 1) { |
2884 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2885 | for (; page < endpage; page += pageblock_nr_pages) { |
2886 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 2887 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 2888 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
2889 | set_pageblock_migratetype(page, |
2890 | MIGRATE_MOVABLE); | |
2891 | } | |
748446bb MG |
2892 | } |
2893 | ||
f3a14ced | 2894 | |
8fb74b9f | 2895 | return 1UL << order; |
1fb3f8ca MG |
2896 | } |
2897 | ||
060e7417 MG |
2898 | /* |
2899 | * Update NUMA hit/miss statistics | |
2900 | * | |
2901 | * Must be called with interrupts disabled. | |
060e7417 | 2902 | */ |
41b6167e | 2903 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
2904 | { |
2905 | #ifdef CONFIG_NUMA | |
3a321d2a | 2906 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 2907 | |
4518085e KW |
2908 | /* skip numa counters update if numa stats is disabled */ |
2909 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2910 | return; | |
2911 | ||
2df26639 | 2912 | if (z->node != numa_node_id()) |
060e7417 | 2913 | local_stat = NUMA_OTHER; |
060e7417 | 2914 | |
2df26639 | 2915 | if (z->node == preferred_zone->node) |
3a321d2a | 2916 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 2917 | else { |
3a321d2a KW |
2918 | __inc_numa_state(z, NUMA_MISS); |
2919 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 2920 | } |
3a321d2a | 2921 | __inc_numa_state(z, local_stat); |
060e7417 MG |
2922 | #endif |
2923 | } | |
2924 | ||
066b2393 MG |
2925 | /* Remove page from the per-cpu list, caller must protect the list */ |
2926 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
453f85d4 | 2927 | struct per_cpu_pages *pcp, |
066b2393 MG |
2928 | struct list_head *list) |
2929 | { | |
2930 | struct page *page; | |
2931 | ||
2932 | do { | |
2933 | if (list_empty(list)) { | |
2934 | pcp->count += rmqueue_bulk(zone, 0, | |
2935 | pcp->batch, list, | |
453f85d4 | 2936 | migratetype); |
066b2393 MG |
2937 | if (unlikely(list_empty(list))) |
2938 | return NULL; | |
2939 | } | |
2940 | ||
453f85d4 | 2941 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
2942 | list_del(&page->lru); |
2943 | pcp->count--; | |
2944 | } while (check_new_pcp(page)); | |
2945 | ||
2946 | return page; | |
2947 | } | |
2948 | ||
2949 | /* Lock and remove page from the per-cpu list */ | |
2950 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
2951 | struct zone *zone, unsigned int order, | |
2952 | gfp_t gfp_flags, int migratetype) | |
2953 | { | |
2954 | struct per_cpu_pages *pcp; | |
2955 | struct list_head *list; | |
066b2393 | 2956 | struct page *page; |
d34b0733 | 2957 | unsigned long flags; |
066b2393 | 2958 | |
d34b0733 | 2959 | local_irq_save(flags); |
066b2393 MG |
2960 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2961 | list = &pcp->lists[migratetype]; | |
453f85d4 | 2962 | page = __rmqueue_pcplist(zone, migratetype, pcp, list); |
066b2393 MG |
2963 | if (page) { |
2964 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
2965 | zone_statistics(preferred_zone, zone); | |
2966 | } | |
d34b0733 | 2967 | local_irq_restore(flags); |
066b2393 MG |
2968 | return page; |
2969 | } | |
2970 | ||
1da177e4 | 2971 | /* |
75379191 | 2972 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2973 | */ |
0a15c3e9 | 2974 | static inline |
066b2393 | 2975 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 2976 | struct zone *zone, unsigned int order, |
c603844b MG |
2977 | gfp_t gfp_flags, unsigned int alloc_flags, |
2978 | int migratetype) | |
1da177e4 LT |
2979 | { |
2980 | unsigned long flags; | |
689bcebf | 2981 | struct page *page; |
1da177e4 | 2982 | |
d34b0733 | 2983 | if (likely(order == 0)) { |
066b2393 MG |
2984 | page = rmqueue_pcplist(preferred_zone, zone, order, |
2985 | gfp_flags, migratetype); | |
2986 | goto out; | |
2987 | } | |
83b9355b | 2988 | |
066b2393 MG |
2989 | /* |
2990 | * We most definitely don't want callers attempting to | |
2991 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2992 | */ | |
2993 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
2994 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 2995 | |
066b2393 MG |
2996 | do { |
2997 | page = NULL; | |
2998 | if (alloc_flags & ALLOC_HARDER) { | |
2999 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3000 | if (page) | |
3001 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3002 | } | |
a74609fa | 3003 | if (!page) |
066b2393 MG |
3004 | page = __rmqueue(zone, order, migratetype); |
3005 | } while (page && check_new_pages(page, order)); | |
3006 | spin_unlock(&zone->lock); | |
3007 | if (!page) | |
3008 | goto failed; | |
3009 | __mod_zone_freepage_state(zone, -(1 << order), | |
3010 | get_pcppage_migratetype(page)); | |
1da177e4 | 3011 | |
16709d1d | 3012 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3013 | zone_statistics(preferred_zone, zone); |
a74609fa | 3014 | local_irq_restore(flags); |
1da177e4 | 3015 | |
066b2393 MG |
3016 | out: |
3017 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); | |
1da177e4 | 3018 | return page; |
a74609fa NP |
3019 | |
3020 | failed: | |
3021 | local_irq_restore(flags); | |
a74609fa | 3022 | return NULL; |
1da177e4 LT |
3023 | } |
3024 | ||
933e312e AM |
3025 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3026 | ||
b2588c4b | 3027 | static struct { |
933e312e AM |
3028 | struct fault_attr attr; |
3029 | ||
621a5f7a | 3030 | bool ignore_gfp_highmem; |
71baba4b | 3031 | bool ignore_gfp_reclaim; |
54114994 | 3032 | u32 min_order; |
933e312e AM |
3033 | } fail_page_alloc = { |
3034 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3035 | .ignore_gfp_reclaim = true, |
621a5f7a | 3036 | .ignore_gfp_highmem = true, |
54114994 | 3037 | .min_order = 1, |
933e312e AM |
3038 | }; |
3039 | ||
3040 | static int __init setup_fail_page_alloc(char *str) | |
3041 | { | |
3042 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3043 | } | |
3044 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3045 | ||
deaf386e | 3046 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3047 | { |
54114994 | 3048 | if (order < fail_page_alloc.min_order) |
deaf386e | 3049 | return false; |
933e312e | 3050 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3051 | return false; |
933e312e | 3052 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3053 | return false; |
71baba4b MG |
3054 | if (fail_page_alloc.ignore_gfp_reclaim && |
3055 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3056 | return false; |
933e312e AM |
3057 | |
3058 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3059 | } | |
3060 | ||
3061 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3062 | ||
3063 | static int __init fail_page_alloc_debugfs(void) | |
3064 | { | |
0825a6f9 | 3065 | umode_t mode = S_IFREG | 0600; |
933e312e | 3066 | struct dentry *dir; |
933e312e | 3067 | |
dd48c085 AM |
3068 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3069 | &fail_page_alloc.attr); | |
3070 | if (IS_ERR(dir)) | |
3071 | return PTR_ERR(dir); | |
933e312e | 3072 | |
b2588c4b | 3073 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 3074 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
3075 | goto fail; |
3076 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3077 | &fail_page_alloc.ignore_gfp_highmem)) | |
3078 | goto fail; | |
3079 | if (!debugfs_create_u32("min-order", mode, dir, | |
3080 | &fail_page_alloc.min_order)) | |
3081 | goto fail; | |
3082 | ||
3083 | return 0; | |
3084 | fail: | |
dd48c085 | 3085 | debugfs_remove_recursive(dir); |
933e312e | 3086 | |
b2588c4b | 3087 | return -ENOMEM; |
933e312e AM |
3088 | } |
3089 | ||
3090 | late_initcall(fail_page_alloc_debugfs); | |
3091 | ||
3092 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3093 | ||
3094 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3095 | ||
deaf386e | 3096 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3097 | { |
deaf386e | 3098 | return false; |
933e312e AM |
3099 | } |
3100 | ||
3101 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3102 | ||
1da177e4 | 3103 | /* |
97a16fc8 MG |
3104 | * Return true if free base pages are above 'mark'. For high-order checks it |
3105 | * will return true of the order-0 watermark is reached and there is at least | |
3106 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3107 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3108 | */ |
86a294a8 MH |
3109 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
3110 | int classzone_idx, unsigned int alloc_flags, | |
3111 | long free_pages) | |
1da177e4 | 3112 | { |
d23ad423 | 3113 | long min = mark; |
1da177e4 | 3114 | int o; |
cd04ae1e | 3115 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3116 | |
0aaa29a5 | 3117 | /* free_pages may go negative - that's OK */ |
df0a6daa | 3118 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 3119 | |
7fb1d9fc | 3120 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3121 | min -= min / 2; |
0aaa29a5 MG |
3122 | |
3123 | /* | |
3124 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3125 | * the high-atomic reserves. This will over-estimate the size of the | |
3126 | * atomic reserve but it avoids a search. | |
3127 | */ | |
cd04ae1e | 3128 | if (likely(!alloc_harder)) { |
0aaa29a5 | 3129 | free_pages -= z->nr_reserved_highatomic; |
cd04ae1e MH |
3130 | } else { |
3131 | /* | |
3132 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3133 | * users on the grounds that it's definitely going to be in | |
3134 | * the exit path shortly and free memory. Any allocation it | |
3135 | * makes during the free path will be small and short-lived. | |
3136 | */ | |
3137 | if (alloc_flags & ALLOC_OOM) | |
3138 | min -= min / 2; | |
3139 | else | |
3140 | min -= min / 4; | |
3141 | } | |
3142 | ||
e2b19197 | 3143 | |
d883c6cf JK |
3144 | #ifdef CONFIG_CMA |
3145 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3146 | if (!(alloc_flags & ALLOC_CMA)) | |
3147 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); | |
3148 | #endif | |
3149 | ||
97a16fc8 MG |
3150 | /* |
3151 | * Check watermarks for an order-0 allocation request. If these | |
3152 | * are not met, then a high-order request also cannot go ahead | |
3153 | * even if a suitable page happened to be free. | |
3154 | */ | |
3155 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 3156 | return false; |
1da177e4 | 3157 | |
97a16fc8 MG |
3158 | /* If this is an order-0 request then the watermark is fine */ |
3159 | if (!order) | |
3160 | return true; | |
3161 | ||
3162 | /* For a high-order request, check at least one suitable page is free */ | |
3163 | for (o = order; o < MAX_ORDER; o++) { | |
3164 | struct free_area *area = &z->free_area[o]; | |
3165 | int mt; | |
3166 | ||
3167 | if (!area->nr_free) | |
3168 | continue; | |
3169 | ||
97a16fc8 MG |
3170 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
3171 | if (!list_empty(&area->free_list[mt])) | |
3172 | return true; | |
3173 | } | |
3174 | ||
3175 | #ifdef CONFIG_CMA | |
d883c6cf JK |
3176 | if ((alloc_flags & ALLOC_CMA) && |
3177 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
97a16fc8 | 3178 | return true; |
d883c6cf | 3179 | } |
97a16fc8 | 3180 | #endif |
b050e376 VB |
3181 | if (alloc_harder && |
3182 | !list_empty(&area->free_list[MIGRATE_HIGHATOMIC])) | |
3183 | return true; | |
1da177e4 | 3184 | } |
97a16fc8 | 3185 | return false; |
88f5acf8 MG |
3186 | } |
3187 | ||
7aeb09f9 | 3188 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 3189 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
3190 | { |
3191 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3192 | zone_page_state(z, NR_FREE_PAGES)); | |
3193 | } | |
3194 | ||
48ee5f36 MG |
3195 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
3196 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
3197 | { | |
3198 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
d883c6cf JK |
3199 | long cma_pages = 0; |
3200 | ||
3201 | #ifdef CONFIG_CMA | |
3202 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3203 | if (!(alloc_flags & ALLOC_CMA)) | |
3204 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
3205 | #endif | |
48ee5f36 MG |
3206 | |
3207 | /* | |
3208 | * Fast check for order-0 only. If this fails then the reserves | |
3209 | * need to be calculated. There is a corner case where the check | |
3210 | * passes but only the high-order atomic reserve are free. If | |
3211 | * the caller is !atomic then it'll uselessly search the free | |
3212 | * list. That corner case is then slower but it is harmless. | |
3213 | */ | |
d883c6cf | 3214 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) |
48ee5f36 MG |
3215 | return true; |
3216 | ||
3217 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3218 | free_pages); | |
3219 | } | |
3220 | ||
7aeb09f9 | 3221 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 3222 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
3223 | { |
3224 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3225 | ||
3226 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3227 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3228 | ||
e2b19197 | 3229 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 3230 | free_pages); |
1da177e4 LT |
3231 | } |
3232 | ||
9276b1bc | 3233 | #ifdef CONFIG_NUMA |
957f822a DR |
3234 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3235 | { | |
e02dc017 | 3236 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
5f7a75ac | 3237 | RECLAIM_DISTANCE; |
957f822a | 3238 | } |
9276b1bc | 3239 | #else /* CONFIG_NUMA */ |
957f822a DR |
3240 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3241 | { | |
3242 | return true; | |
3243 | } | |
9276b1bc PJ |
3244 | #endif /* CONFIG_NUMA */ |
3245 | ||
7fb1d9fc | 3246 | /* |
0798e519 | 3247 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3248 | * a page. |
3249 | */ | |
3250 | static struct page * | |
a9263751 VB |
3251 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3252 | const struct alloc_context *ac) | |
753ee728 | 3253 | { |
c33d6c06 | 3254 | struct zoneref *z = ac->preferred_zoneref; |
5117f45d | 3255 | struct zone *zone; |
3b8c0be4 MG |
3256 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
3257 | ||
7fb1d9fc | 3258 | /* |
9276b1bc | 3259 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3260 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3261 | */ |
c33d6c06 | 3262 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 3263 | ac->nodemask) { |
be06af00 | 3264 | struct page *page; |
e085dbc5 JW |
3265 | unsigned long mark; |
3266 | ||
664eedde MG |
3267 | if (cpusets_enabled() && |
3268 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3269 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3270 | continue; |
a756cf59 JW |
3271 | /* |
3272 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3273 | * want to get it from a node that is within its dirty |
3274 | * limit, such that no single node holds more than its | |
a756cf59 | 3275 | * proportional share of globally allowed dirty pages. |
281e3726 | 3276 | * The dirty limits take into account the node's |
a756cf59 JW |
3277 | * lowmem reserves and high watermark so that kswapd |
3278 | * should be able to balance it without having to | |
3279 | * write pages from its LRU list. | |
3280 | * | |
a756cf59 | 3281 | * XXX: For now, allow allocations to potentially |
281e3726 | 3282 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3283 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3284 | * which is important when on a NUMA setup the allowed |
281e3726 | 3285 | * nodes are together not big enough to reach the |
a756cf59 | 3286 | * global limit. The proper fix for these situations |
281e3726 | 3287 | * will require awareness of nodes in the |
a756cf59 JW |
3288 | * dirty-throttling and the flusher threads. |
3289 | */ | |
3b8c0be4 MG |
3290 | if (ac->spread_dirty_pages) { |
3291 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3292 | continue; | |
3293 | ||
3294 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3295 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3296 | continue; | |
3297 | } | |
3298 | } | |
7fb1d9fc | 3299 | |
e085dbc5 | 3300 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 3301 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 3302 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
3303 | int ret; |
3304 | ||
c9e97a19 PT |
3305 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3306 | /* | |
3307 | * Watermark failed for this zone, but see if we can | |
3308 | * grow this zone if it contains deferred pages. | |
3309 | */ | |
3310 | if (static_branch_unlikely(&deferred_pages)) { | |
3311 | if (_deferred_grow_zone(zone, order)) | |
3312 | goto try_this_zone; | |
3313 | } | |
3314 | #endif | |
5dab2911 MG |
3315 | /* Checked here to keep the fast path fast */ |
3316 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3317 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3318 | goto try_this_zone; | |
3319 | ||
a5f5f91d | 3320 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3321 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3322 | continue; |
3323 | ||
a5f5f91d | 3324 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3325 | switch (ret) { |
a5f5f91d | 3326 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3327 | /* did not scan */ |
cd38b115 | 3328 | continue; |
a5f5f91d | 3329 | case NODE_RECLAIM_FULL: |
fa5e084e | 3330 | /* scanned but unreclaimable */ |
cd38b115 | 3331 | continue; |
fa5e084e MG |
3332 | default: |
3333 | /* did we reclaim enough */ | |
fed2719e | 3334 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3335 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3336 | goto try_this_zone; |
3337 | ||
fed2719e | 3338 | continue; |
0798e519 | 3339 | } |
7fb1d9fc RS |
3340 | } |
3341 | ||
fa5e084e | 3342 | try_this_zone: |
066b2393 | 3343 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3344 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3345 | if (page) { |
479f854a | 3346 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3347 | |
3348 | /* | |
3349 | * If this is a high-order atomic allocation then check | |
3350 | * if the pageblock should be reserved for the future | |
3351 | */ | |
3352 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3353 | reserve_highatomic_pageblock(page, zone, order); | |
3354 | ||
75379191 | 3355 | return page; |
c9e97a19 PT |
3356 | } else { |
3357 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
3358 | /* Try again if zone has deferred pages */ | |
3359 | if (static_branch_unlikely(&deferred_pages)) { | |
3360 | if (_deferred_grow_zone(zone, order)) | |
3361 | goto try_this_zone; | |
3362 | } | |
3363 | #endif | |
75379191 | 3364 | } |
54a6eb5c | 3365 | } |
9276b1bc | 3366 | |
4ffeaf35 | 3367 | return NULL; |
753ee728 MH |
3368 | } |
3369 | ||
29423e77 DR |
3370 | /* |
3371 | * Large machines with many possible nodes should not always dump per-node | |
3372 | * meminfo in irq context. | |
3373 | */ | |
3374 | static inline bool should_suppress_show_mem(void) | |
3375 | { | |
3376 | bool ret = false; | |
3377 | ||
3378 | #if NODES_SHIFT > 8 | |
3379 | ret = in_interrupt(); | |
3380 | #endif | |
3381 | return ret; | |
3382 | } | |
3383 | ||
9af744d7 | 3384 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3385 | { |
a238ab5b | 3386 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
aa187507 | 3387 | static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1); |
a238ab5b | 3388 | |
aa187507 | 3389 | if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs)) |
a238ab5b DH |
3390 | return; |
3391 | ||
3392 | /* | |
3393 | * This documents exceptions given to allocations in certain | |
3394 | * contexts that are allowed to allocate outside current's set | |
3395 | * of allowed nodes. | |
3396 | */ | |
3397 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3398 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3399 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3400 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3401 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3402 | filter &= ~SHOW_MEM_FILTER_NODES; |
3403 | ||
9af744d7 | 3404 | show_mem(filter, nodemask); |
aa187507 MH |
3405 | } |
3406 | ||
a8e99259 | 3407 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3408 | { |
3409 | struct va_format vaf; | |
3410 | va_list args; | |
3411 | static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL, | |
3412 | DEFAULT_RATELIMIT_BURST); | |
3413 | ||
0f7896f1 | 3414 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3415 | return; |
3416 | ||
7877cdcc MH |
3417 | va_start(args, fmt); |
3418 | vaf.fmt = fmt; | |
3419 | vaf.va = &args; | |
0205f755 MH |
3420 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl\n", |
3421 | current->comm, &vaf, gfp_mask, &gfp_mask, | |
3422 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3423 | va_end(args); |
3ee9a4f0 | 3424 | |
a8e99259 | 3425 | cpuset_print_current_mems_allowed(); |
3ee9a4f0 | 3426 | |
a238ab5b | 3427 | dump_stack(); |
685dbf6f | 3428 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3429 | } |
3430 | ||
6c18ba7a MH |
3431 | static inline struct page * |
3432 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3433 | unsigned int alloc_flags, | |
3434 | const struct alloc_context *ac) | |
3435 | { | |
3436 | struct page *page; | |
3437 | ||
3438 | page = get_page_from_freelist(gfp_mask, order, | |
3439 | alloc_flags|ALLOC_CPUSET, ac); | |
3440 | /* | |
3441 | * fallback to ignore cpuset restriction if our nodes | |
3442 | * are depleted | |
3443 | */ | |
3444 | if (!page) | |
3445 | page = get_page_from_freelist(gfp_mask, order, | |
3446 | alloc_flags, ac); | |
3447 | ||
3448 | return page; | |
3449 | } | |
3450 | ||
11e33f6a MG |
3451 | static inline struct page * |
3452 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3453 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3454 | { |
6e0fc46d DR |
3455 | struct oom_control oc = { |
3456 | .zonelist = ac->zonelist, | |
3457 | .nodemask = ac->nodemask, | |
2a966b77 | 3458 | .memcg = NULL, |
6e0fc46d DR |
3459 | .gfp_mask = gfp_mask, |
3460 | .order = order, | |
6e0fc46d | 3461 | }; |
11e33f6a MG |
3462 | struct page *page; |
3463 | ||
9879de73 JW |
3464 | *did_some_progress = 0; |
3465 | ||
9879de73 | 3466 | /* |
dc56401f JW |
3467 | * Acquire the oom lock. If that fails, somebody else is |
3468 | * making progress for us. | |
9879de73 | 3469 | */ |
dc56401f | 3470 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3471 | *did_some_progress = 1; |
11e33f6a | 3472 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3473 | return NULL; |
3474 | } | |
6b1de916 | 3475 | |
11e33f6a MG |
3476 | /* |
3477 | * Go through the zonelist yet one more time, keep very high watermark | |
3478 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3479 | * we're still under heavy pressure. But make sure that this reclaim |
3480 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3481 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3482 | */ |
e746bf73 TH |
3483 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3484 | ~__GFP_DIRECT_RECLAIM, order, | |
3485 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3486 | if (page) |
11e33f6a MG |
3487 | goto out; |
3488 | ||
06ad276a MH |
3489 | /* Coredumps can quickly deplete all memory reserves */ |
3490 | if (current->flags & PF_DUMPCORE) | |
3491 | goto out; | |
3492 | /* The OOM killer will not help higher order allocs */ | |
3493 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3494 | goto out; | |
dcda9b04 MH |
3495 | /* |
3496 | * We have already exhausted all our reclaim opportunities without any | |
3497 | * success so it is time to admit defeat. We will skip the OOM killer | |
3498 | * because it is very likely that the caller has a more reasonable | |
3499 | * fallback than shooting a random task. | |
3500 | */ | |
3501 | if (gfp_mask & __GFP_RETRY_MAYFAIL) | |
3502 | goto out; | |
06ad276a MH |
3503 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
3504 | if (ac->high_zoneidx < ZONE_NORMAL) | |
3505 | goto out; | |
3506 | if (pm_suspended_storage()) | |
3507 | goto out; | |
3508 | /* | |
3509 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3510 | * other request to make a forward progress. | |
3511 | * We are in an unfortunate situation where out_of_memory cannot | |
3512 | * do much for this context but let's try it to at least get | |
3513 | * access to memory reserved if the current task is killed (see | |
3514 | * out_of_memory). Once filesystems are ready to handle allocation | |
3515 | * failures more gracefully we should just bail out here. | |
3516 | */ | |
3517 | ||
3518 | /* The OOM killer may not free memory on a specific node */ | |
3519 | if (gfp_mask & __GFP_THISNODE) | |
3520 | goto out; | |
3da88fb3 | 3521 | |
3c2c6488 | 3522 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 3523 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3524 | *did_some_progress = 1; |
5020e285 | 3525 | |
6c18ba7a MH |
3526 | /* |
3527 | * Help non-failing allocations by giving them access to memory | |
3528 | * reserves | |
3529 | */ | |
3530 | if (gfp_mask & __GFP_NOFAIL) | |
3531 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3532 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3533 | } |
11e33f6a | 3534 | out: |
dc56401f | 3535 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3536 | return page; |
3537 | } | |
3538 | ||
33c2d214 MH |
3539 | /* |
3540 | * Maximum number of compaction retries wit a progress before OOM | |
3541 | * killer is consider as the only way to move forward. | |
3542 | */ | |
3543 | #define MAX_COMPACT_RETRIES 16 | |
3544 | ||
56de7263 MG |
3545 | #ifdef CONFIG_COMPACTION |
3546 | /* Try memory compaction for high-order allocations before reclaim */ | |
3547 | static struct page * | |
3548 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3549 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3550 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3551 | { |
98dd3b48 | 3552 | struct page *page; |
499118e9 | 3553 | unsigned int noreclaim_flag; |
53853e2d VB |
3554 | |
3555 | if (!order) | |
66199712 | 3556 | return NULL; |
66199712 | 3557 | |
499118e9 | 3558 | noreclaim_flag = memalloc_noreclaim_save(); |
c5d01d0d | 3559 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
c3486f53 | 3560 | prio); |
499118e9 | 3561 | memalloc_noreclaim_restore(noreclaim_flag); |
56de7263 | 3562 | |
c5d01d0d | 3563 | if (*compact_result <= COMPACT_INACTIVE) |
98dd3b48 | 3564 | return NULL; |
53853e2d | 3565 | |
98dd3b48 VB |
3566 | /* |
3567 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3568 | * count a compaction stall | |
3569 | */ | |
3570 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3571 | |
31a6c190 | 3572 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
53853e2d | 3573 | |
98dd3b48 VB |
3574 | if (page) { |
3575 | struct zone *zone = page_zone(page); | |
53853e2d | 3576 | |
98dd3b48 VB |
3577 | zone->compact_blockskip_flush = false; |
3578 | compaction_defer_reset(zone, order, true); | |
3579 | count_vm_event(COMPACTSUCCESS); | |
3580 | return page; | |
3581 | } | |
56de7263 | 3582 | |
98dd3b48 VB |
3583 | /* |
3584 | * It's bad if compaction run occurs and fails. The most likely reason | |
3585 | * is that pages exist, but not enough to satisfy watermarks. | |
3586 | */ | |
3587 | count_vm_event(COMPACTFAIL); | |
66199712 | 3588 | |
98dd3b48 | 3589 | cond_resched(); |
56de7263 MG |
3590 | |
3591 | return NULL; | |
3592 | } | |
33c2d214 | 3593 | |
3250845d VB |
3594 | static inline bool |
3595 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3596 | enum compact_result compact_result, | |
3597 | enum compact_priority *compact_priority, | |
d9436498 | 3598 | int *compaction_retries) |
3250845d VB |
3599 | { |
3600 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3601 | int min_priority; |
65190cff MH |
3602 | bool ret = false; |
3603 | int retries = *compaction_retries; | |
3604 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3605 | |
3606 | if (!order) | |
3607 | return false; | |
3608 | ||
d9436498 VB |
3609 | if (compaction_made_progress(compact_result)) |
3610 | (*compaction_retries)++; | |
3611 | ||
3250845d VB |
3612 | /* |
3613 | * compaction considers all the zone as desperately out of memory | |
3614 | * so it doesn't really make much sense to retry except when the | |
3615 | * failure could be caused by insufficient priority | |
3616 | */ | |
d9436498 VB |
3617 | if (compaction_failed(compact_result)) |
3618 | goto check_priority; | |
3250845d VB |
3619 | |
3620 | /* | |
3621 | * make sure the compaction wasn't deferred or didn't bail out early | |
3622 | * due to locks contention before we declare that we should give up. | |
3623 | * But do not retry if the given zonelist is not suitable for | |
3624 | * compaction. | |
3625 | */ | |
65190cff MH |
3626 | if (compaction_withdrawn(compact_result)) { |
3627 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
3628 | goto out; | |
3629 | } | |
3250845d VB |
3630 | |
3631 | /* | |
dcda9b04 | 3632 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
3633 | * costly ones because they are de facto nofail and invoke OOM |
3634 | * killer to move on while costly can fail and users are ready | |
3635 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3636 | * would need much more detailed feedback from compaction to | |
3637 | * make a better decision. | |
3638 | */ | |
3639 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3640 | max_retries /= 4; | |
65190cff MH |
3641 | if (*compaction_retries <= max_retries) { |
3642 | ret = true; | |
3643 | goto out; | |
3644 | } | |
3250845d | 3645 | |
d9436498 VB |
3646 | /* |
3647 | * Make sure there are attempts at the highest priority if we exhausted | |
3648 | * all retries or failed at the lower priorities. | |
3649 | */ | |
3650 | check_priority: | |
c2033b00 VB |
3651 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3652 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3653 | |
c2033b00 | 3654 | if (*compact_priority > min_priority) { |
d9436498 VB |
3655 | (*compact_priority)--; |
3656 | *compaction_retries = 0; | |
65190cff | 3657 | ret = true; |
d9436498 | 3658 | } |
65190cff MH |
3659 | out: |
3660 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
3661 | return ret; | |
3250845d | 3662 | } |
56de7263 MG |
3663 | #else |
3664 | static inline struct page * | |
3665 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3666 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3667 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3668 | { |
33c2d214 | 3669 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3670 | return NULL; |
3671 | } | |
33c2d214 MH |
3672 | |
3673 | static inline bool | |
86a294a8 MH |
3674 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3675 | enum compact_result compact_result, | |
a5508cd8 | 3676 | enum compact_priority *compact_priority, |
d9436498 | 3677 | int *compaction_retries) |
33c2d214 | 3678 | { |
31e49bfd MH |
3679 | struct zone *zone; |
3680 | struct zoneref *z; | |
3681 | ||
3682 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3683 | return false; | |
3684 | ||
3685 | /* | |
3686 | * There are setups with compaction disabled which would prefer to loop | |
3687 | * inside the allocator rather than hit the oom killer prematurely. | |
3688 | * Let's give them a good hope and keep retrying while the order-0 | |
3689 | * watermarks are OK. | |
3690 | */ | |
3691 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3692 | ac->nodemask) { | |
3693 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
3694 | ac_classzone_idx(ac), alloc_flags)) | |
3695 | return true; | |
3696 | } | |
33c2d214 MH |
3697 | return false; |
3698 | } | |
3250845d | 3699 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3700 | |
d92a8cfc | 3701 | #ifdef CONFIG_LOCKDEP |
93781325 | 3702 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
3703 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
3704 | ||
3705 | static bool __need_fs_reclaim(gfp_t gfp_mask) | |
3706 | { | |
3707 | gfp_mask = current_gfp_context(gfp_mask); | |
3708 | ||
3709 | /* no reclaim without waiting on it */ | |
3710 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
3711 | return false; | |
3712 | ||
3713 | /* this guy won't enter reclaim */ | |
2e517d68 | 3714 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
3715 | return false; |
3716 | ||
3717 | /* We're only interested __GFP_FS allocations for now */ | |
3718 | if (!(gfp_mask & __GFP_FS)) | |
3719 | return false; | |
3720 | ||
3721 | if (gfp_mask & __GFP_NOLOCKDEP) | |
3722 | return false; | |
3723 | ||
3724 | return true; | |
3725 | } | |
3726 | ||
93781325 OS |
3727 | void __fs_reclaim_acquire(void) |
3728 | { | |
3729 | lock_map_acquire(&__fs_reclaim_map); | |
3730 | } | |
3731 | ||
3732 | void __fs_reclaim_release(void) | |
3733 | { | |
3734 | lock_map_release(&__fs_reclaim_map); | |
3735 | } | |
3736 | ||
d92a8cfc PZ |
3737 | void fs_reclaim_acquire(gfp_t gfp_mask) |
3738 | { | |
3739 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 3740 | __fs_reclaim_acquire(); |
d92a8cfc PZ |
3741 | } |
3742 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
3743 | ||
3744 | void fs_reclaim_release(gfp_t gfp_mask) | |
3745 | { | |
3746 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 3747 | __fs_reclaim_release(); |
d92a8cfc PZ |
3748 | } |
3749 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
3750 | #endif | |
3751 | ||
bba90710 MS |
3752 | /* Perform direct synchronous page reclaim */ |
3753 | static int | |
a9263751 VB |
3754 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3755 | const struct alloc_context *ac) | |
11e33f6a | 3756 | { |
11e33f6a | 3757 | struct reclaim_state reclaim_state; |
bba90710 | 3758 | int progress; |
499118e9 | 3759 | unsigned int noreclaim_flag; |
11e33f6a MG |
3760 | |
3761 | cond_resched(); | |
3762 | ||
3763 | /* We now go into synchronous reclaim */ | |
3764 | cpuset_memory_pressure_bump(); | |
d92a8cfc | 3765 | fs_reclaim_acquire(gfp_mask); |
93781325 | 3766 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 3767 | reclaim_state.reclaimed_slab = 0; |
c06b1fca | 3768 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3769 | |
a9263751 VB |
3770 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3771 | ac->nodemask); | |
11e33f6a | 3772 | |
c06b1fca | 3773 | current->reclaim_state = NULL; |
499118e9 | 3774 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 3775 | fs_reclaim_release(gfp_mask); |
11e33f6a MG |
3776 | |
3777 | cond_resched(); | |
3778 | ||
bba90710 MS |
3779 | return progress; |
3780 | } | |
3781 | ||
3782 | /* The really slow allocator path where we enter direct reclaim */ | |
3783 | static inline struct page * | |
3784 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3785 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3786 | unsigned long *did_some_progress) |
bba90710 MS |
3787 | { |
3788 | struct page *page = NULL; | |
3789 | bool drained = false; | |
3790 | ||
a9263751 | 3791 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3792 | if (unlikely(!(*did_some_progress))) |
3793 | return NULL; | |
11e33f6a | 3794 | |
9ee493ce | 3795 | retry: |
31a6c190 | 3796 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3797 | |
3798 | /* | |
3799 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3800 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3801 | * Shrink them them and try again | |
9ee493ce MG |
3802 | */ |
3803 | if (!page && !drained) { | |
29fac03b | 3804 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3805 | drain_all_pages(NULL); |
9ee493ce MG |
3806 | drained = true; |
3807 | goto retry; | |
3808 | } | |
3809 | ||
11e33f6a MG |
3810 | return page; |
3811 | } | |
3812 | ||
5ecd9d40 DR |
3813 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
3814 | const struct alloc_context *ac) | |
3a025760 JW |
3815 | { |
3816 | struct zoneref *z; | |
3817 | struct zone *zone; | |
e1a55637 | 3818 | pg_data_t *last_pgdat = NULL; |
5ecd9d40 | 3819 | enum zone_type high_zoneidx = ac->high_zoneidx; |
3a025760 | 3820 | |
5ecd9d40 DR |
3821 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx, |
3822 | ac->nodemask) { | |
e1a55637 | 3823 | if (last_pgdat != zone->zone_pgdat) |
5ecd9d40 | 3824 | wakeup_kswapd(zone, gfp_mask, order, high_zoneidx); |
e1a55637 MG |
3825 | last_pgdat = zone->zone_pgdat; |
3826 | } | |
3a025760 JW |
3827 | } |
3828 | ||
c603844b | 3829 | static inline unsigned int |
341ce06f PZ |
3830 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3831 | { | |
c603844b | 3832 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3833 | |
a56f57ff | 3834 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3835 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3836 | |
341ce06f PZ |
3837 | /* |
3838 | * The caller may dip into page reserves a bit more if the caller | |
3839 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3840 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3841 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3842 | */ |
e6223a3b | 3843 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3844 | |
d0164adc | 3845 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3846 | /* |
b104a35d DR |
3847 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3848 | * if it can't schedule. | |
5c3240d9 | 3849 | */ |
b104a35d | 3850 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3851 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3852 | /* |
b104a35d | 3853 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3854 | * comment for __cpuset_node_allowed(). |
523b9458 | 3855 | */ |
341ce06f | 3856 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3857 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3858 | alloc_flags |= ALLOC_HARDER; |
3859 | ||
d883c6cf JK |
3860 | #ifdef CONFIG_CMA |
3861 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
3862 | alloc_flags |= ALLOC_CMA; | |
3863 | #endif | |
341ce06f PZ |
3864 | return alloc_flags; |
3865 | } | |
3866 | ||
cd04ae1e | 3867 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 3868 | { |
cd04ae1e MH |
3869 | if (!tsk_is_oom_victim(tsk)) |
3870 | return false; | |
3871 | ||
3872 | /* | |
3873 | * !MMU doesn't have oom reaper so give access to memory reserves | |
3874 | * only to the thread with TIF_MEMDIE set | |
3875 | */ | |
3876 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
3877 | return false; |
3878 | ||
cd04ae1e MH |
3879 | return true; |
3880 | } | |
3881 | ||
3882 | /* | |
3883 | * Distinguish requests which really need access to full memory | |
3884 | * reserves from oom victims which can live with a portion of it | |
3885 | */ | |
3886 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
3887 | { | |
3888 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
3889 | return 0; | |
31a6c190 | 3890 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 3891 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 3892 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
3893 | return ALLOC_NO_WATERMARKS; |
3894 | if (!in_interrupt()) { | |
3895 | if (current->flags & PF_MEMALLOC) | |
3896 | return ALLOC_NO_WATERMARKS; | |
3897 | else if (oom_reserves_allowed(current)) | |
3898 | return ALLOC_OOM; | |
3899 | } | |
31a6c190 | 3900 | |
cd04ae1e MH |
3901 | return 0; |
3902 | } | |
3903 | ||
3904 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
3905 | { | |
3906 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
3907 | } |
3908 | ||
0a0337e0 MH |
3909 | /* |
3910 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
3911 | * for the given allocation request. | |
491d79ae JW |
3912 | * |
3913 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
3914 | * without success, or when we couldn't even meet the watermark if we | |
3915 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
3916 | * |
3917 | * Returns true if a retry is viable or false to enter the oom path. | |
3918 | */ | |
3919 | static inline bool | |
3920 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
3921 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 3922 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
3923 | { |
3924 | struct zone *zone; | |
3925 | struct zoneref *z; | |
3926 | ||
423b452e VB |
3927 | /* |
3928 | * Costly allocations might have made a progress but this doesn't mean | |
3929 | * their order will become available due to high fragmentation so | |
3930 | * always increment the no progress counter for them | |
3931 | */ | |
3932 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
3933 | *no_progress_loops = 0; | |
3934 | else | |
3935 | (*no_progress_loops)++; | |
3936 | ||
0a0337e0 MH |
3937 | /* |
3938 | * Make sure we converge to OOM if we cannot make any progress | |
3939 | * several times in the row. | |
3940 | */ | |
04c8716f MK |
3941 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
3942 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 3943 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 3944 | } |
0a0337e0 | 3945 | |
bca67592 MG |
3946 | /* |
3947 | * Keep reclaiming pages while there is a chance this will lead | |
3948 | * somewhere. If none of the target zones can satisfy our allocation | |
3949 | * request even if all reclaimable pages are considered then we are | |
3950 | * screwed and have to go OOM. | |
0a0337e0 MH |
3951 | */ |
3952 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3953 | ac->nodemask) { | |
3954 | unsigned long available; | |
ede37713 | 3955 | unsigned long reclaimable; |
d379f01d MH |
3956 | unsigned long min_wmark = min_wmark_pages(zone); |
3957 | bool wmark; | |
0a0337e0 | 3958 | |
5a1c84b4 | 3959 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 3960 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
3961 | |
3962 | /* | |
491d79ae JW |
3963 | * Would the allocation succeed if we reclaimed all |
3964 | * reclaimable pages? | |
0a0337e0 | 3965 | */ |
d379f01d MH |
3966 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
3967 | ac_classzone_idx(ac), alloc_flags, available); | |
3968 | trace_reclaim_retry_zone(z, order, reclaimable, | |
3969 | available, min_wmark, *no_progress_loops, wmark); | |
3970 | if (wmark) { | |
ede37713 MH |
3971 | /* |
3972 | * If we didn't make any progress and have a lot of | |
3973 | * dirty + writeback pages then we should wait for | |
3974 | * an IO to complete to slow down the reclaim and | |
3975 | * prevent from pre mature OOM | |
3976 | */ | |
3977 | if (!did_some_progress) { | |
11fb9989 | 3978 | unsigned long write_pending; |
ede37713 | 3979 | |
5a1c84b4 MG |
3980 | write_pending = zone_page_state_snapshot(zone, |
3981 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 3982 | |
11fb9989 | 3983 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
3984 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
3985 | return true; | |
3986 | } | |
3987 | } | |
5a1c84b4 | 3988 | |
ede37713 MH |
3989 | /* |
3990 | * Memory allocation/reclaim might be called from a WQ | |
3991 | * context and the current implementation of the WQ | |
3992 | * concurrency control doesn't recognize that | |
3993 | * a particular WQ is congested if the worker thread is | |
3994 | * looping without ever sleeping. Therefore we have to | |
3995 | * do a short sleep here rather than calling | |
3996 | * cond_resched(). | |
3997 | */ | |
3998 | if (current->flags & PF_WQ_WORKER) | |
3999 | schedule_timeout_uninterruptible(1); | |
4000 | else | |
4001 | cond_resched(); | |
4002 | ||
0a0337e0 MH |
4003 | return true; |
4004 | } | |
4005 | } | |
4006 | ||
4007 | return false; | |
4008 | } | |
4009 | ||
902b6281 VB |
4010 | static inline bool |
4011 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4012 | { | |
4013 | /* | |
4014 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4015 | * mempolicy don't intersect. This should be normally dealt with by | |
4016 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4017 | * such a way the check therein was true, and then it became false | |
4018 | * before we got our cpuset_mems_cookie here. | |
4019 | * This assumes that for all allocations, ac->nodemask can come only | |
4020 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4021 | * when it does not intersect with the cpuset restrictions) or the | |
4022 | * caller can deal with a violated nodemask. | |
4023 | */ | |
4024 | if (cpusets_enabled() && ac->nodemask && | |
4025 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4026 | ac->nodemask = NULL; | |
4027 | return true; | |
4028 | } | |
4029 | ||
4030 | /* | |
4031 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4032 | * possible to race with parallel threads in such a way that our | |
4033 | * allocation can fail while the mask is being updated. If we are about | |
4034 | * to fail, check if the cpuset changed during allocation and if so, | |
4035 | * retry. | |
4036 | */ | |
4037 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4038 | return true; | |
4039 | ||
4040 | return false; | |
4041 | } | |
4042 | ||
11e33f6a MG |
4043 | static inline struct page * |
4044 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4045 | struct alloc_context *ac) |
11e33f6a | 4046 | { |
d0164adc | 4047 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4048 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4049 | struct page *page = NULL; |
c603844b | 4050 | unsigned int alloc_flags; |
11e33f6a | 4051 | unsigned long did_some_progress; |
5ce9bfef | 4052 | enum compact_priority compact_priority; |
c5d01d0d | 4053 | enum compact_result compact_result; |
5ce9bfef VB |
4054 | int compaction_retries; |
4055 | int no_progress_loops; | |
5ce9bfef | 4056 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4057 | int reserve_flags; |
1da177e4 | 4058 | |
72807a74 MG |
4059 | /* |
4060 | * In the slowpath, we sanity check order to avoid ever trying to | |
4061 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
4062 | * be using allocators in order of preference for an area that is | |
4063 | * too large. | |
4064 | */ | |
1fc28b70 MG |
4065 | if (order >= MAX_ORDER) { |
4066 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 4067 | return NULL; |
1fc28b70 | 4068 | } |
1da177e4 | 4069 | |
d0164adc MG |
4070 | /* |
4071 | * We also sanity check to catch abuse of atomic reserves being used by | |
4072 | * callers that are not in atomic context. | |
4073 | */ | |
4074 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4075 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4076 | gfp_mask &= ~__GFP_ATOMIC; | |
4077 | ||
5ce9bfef VB |
4078 | retry_cpuset: |
4079 | compaction_retries = 0; | |
4080 | no_progress_loops = 0; | |
4081 | compact_priority = DEF_COMPACT_PRIORITY; | |
4082 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4083 | |
4084 | /* | |
4085 | * The fast path uses conservative alloc_flags to succeed only until | |
4086 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4087 | * alloc_flags precisely. So we do that now. | |
4088 | */ | |
4089 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4090 | ||
e47483bc VB |
4091 | /* |
4092 | * We need to recalculate the starting point for the zonelist iterator | |
4093 | * because we might have used different nodemask in the fast path, or | |
4094 | * there was a cpuset modification and we are retrying - otherwise we | |
4095 | * could end up iterating over non-eligible zones endlessly. | |
4096 | */ | |
4097 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
4098 | ac->high_zoneidx, ac->nodemask); | |
4099 | if (!ac->preferred_zoneref->zone) | |
4100 | goto nopage; | |
4101 | ||
23771235 | 4102 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
5ecd9d40 | 4103 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4104 | |
4105 | /* | |
4106 | * The adjusted alloc_flags might result in immediate success, so try | |
4107 | * that first | |
4108 | */ | |
4109 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4110 | if (page) | |
4111 | goto got_pg; | |
4112 | ||
a8161d1e VB |
4113 | /* |
4114 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4115 | * that we have enough base pages and don't need to reclaim. For non- |
4116 | * movable high-order allocations, do that as well, as compaction will | |
4117 | * try prevent permanent fragmentation by migrating from blocks of the | |
4118 | * same migratetype. | |
4119 | * Don't try this for allocations that are allowed to ignore | |
4120 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4121 | */ |
282722b0 VB |
4122 | if (can_direct_reclaim && |
4123 | (costly_order || | |
4124 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4125 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4126 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4127 | alloc_flags, ac, | |
a5508cd8 | 4128 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4129 | &compact_result); |
4130 | if (page) | |
4131 | goto got_pg; | |
4132 | ||
3eb2771b VB |
4133 | /* |
4134 | * Checks for costly allocations with __GFP_NORETRY, which | |
4135 | * includes THP page fault allocations | |
4136 | */ | |
282722b0 | 4137 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { |
a8161d1e VB |
4138 | /* |
4139 | * If compaction is deferred for high-order allocations, | |
4140 | * it is because sync compaction recently failed. If | |
4141 | * this is the case and the caller requested a THP | |
4142 | * allocation, we do not want to heavily disrupt the | |
4143 | * system, so we fail the allocation instead of entering | |
4144 | * direct reclaim. | |
4145 | */ | |
4146 | if (compact_result == COMPACT_DEFERRED) | |
4147 | goto nopage; | |
4148 | ||
a8161d1e | 4149 | /* |
3eb2771b VB |
4150 | * Looks like reclaim/compaction is worth trying, but |
4151 | * sync compaction could be very expensive, so keep | |
25160354 | 4152 | * using async compaction. |
a8161d1e | 4153 | */ |
a5508cd8 | 4154 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4155 | } |
4156 | } | |
23771235 | 4157 | |
31a6c190 | 4158 | retry: |
23771235 | 4159 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
31a6c190 | 4160 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
5ecd9d40 | 4161 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4162 | |
cd04ae1e MH |
4163 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4164 | if (reserve_flags) | |
4165 | alloc_flags = reserve_flags; | |
23771235 | 4166 | |
e46e7b77 | 4167 | /* |
d6a24df0 VB |
4168 | * Reset the nodemask and zonelist iterators if memory policies can be |
4169 | * ignored. These allocations are high priority and system rather than | |
4170 | * user oriented. | |
e46e7b77 | 4171 | */ |
cd04ae1e | 4172 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4173 | ac->nodemask = NULL; |
e46e7b77 MG |
4174 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4175 | ac->high_zoneidx, ac->nodemask); | |
4176 | } | |
4177 | ||
23771235 | 4178 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4179 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4180 | if (page) |
4181 | goto got_pg; | |
1da177e4 | 4182 | |
d0164adc | 4183 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4184 | if (!can_direct_reclaim) |
1da177e4 LT |
4185 | goto nopage; |
4186 | ||
9a67f648 MH |
4187 | /* Avoid recursion of direct reclaim */ |
4188 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4189 | goto nopage; |
4190 | ||
a8161d1e VB |
4191 | /* Try direct reclaim and then allocating */ |
4192 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4193 | &did_some_progress); | |
4194 | if (page) | |
4195 | goto got_pg; | |
4196 | ||
4197 | /* Try direct compaction and then allocating */ | |
a9263751 | 4198 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4199 | compact_priority, &compact_result); |
56de7263 MG |
4200 | if (page) |
4201 | goto got_pg; | |
75f30861 | 4202 | |
9083905a JW |
4203 | /* Do not loop if specifically requested */ |
4204 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4205 | goto nopage; |
9083905a | 4206 | |
0a0337e0 MH |
4207 | /* |
4208 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4209 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4210 | */ |
dcda9b04 | 4211 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4212 | goto nopage; |
0a0337e0 | 4213 | |
0a0337e0 | 4214 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4215 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4216 | goto retry; |
4217 | ||
33c2d214 MH |
4218 | /* |
4219 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4220 | * reclaim is not able to make any progress because the current | |
4221 | * implementation of the compaction depends on the sufficient amount | |
4222 | * of free memory (see __compaction_suitable) | |
4223 | */ | |
4224 | if (did_some_progress > 0 && | |
86a294a8 | 4225 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4226 | compact_result, &compact_priority, |
d9436498 | 4227 | &compaction_retries)) |
33c2d214 MH |
4228 | goto retry; |
4229 | ||
902b6281 VB |
4230 | |
4231 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4232 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4233 | goto retry_cpuset; |
4234 | ||
9083905a JW |
4235 | /* Reclaim has failed us, start killing things */ |
4236 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4237 | if (page) | |
4238 | goto got_pg; | |
4239 | ||
9a67f648 | 4240 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e MH |
4241 | if (tsk_is_oom_victim(current) && |
4242 | (alloc_flags == ALLOC_OOM || | |
c288983d | 4243 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4244 | goto nopage; |
4245 | ||
9083905a | 4246 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4247 | if (did_some_progress) { |
4248 | no_progress_loops = 0; | |
9083905a | 4249 | goto retry; |
0a0337e0 | 4250 | } |
9083905a | 4251 | |
1da177e4 | 4252 | nopage: |
902b6281 VB |
4253 | /* Deal with possible cpuset update races before we fail */ |
4254 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4255 | goto retry_cpuset; |
4256 | ||
9a67f648 MH |
4257 | /* |
4258 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4259 | * we always retry | |
4260 | */ | |
4261 | if (gfp_mask & __GFP_NOFAIL) { | |
4262 | /* | |
4263 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4264 | * of any new users that actually require GFP_NOWAIT | |
4265 | */ | |
4266 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4267 | goto fail; | |
4268 | ||
4269 | /* | |
4270 | * PF_MEMALLOC request from this context is rather bizarre | |
4271 | * because we cannot reclaim anything and only can loop waiting | |
4272 | * for somebody to do a work for us | |
4273 | */ | |
4274 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4275 | ||
4276 | /* | |
4277 | * non failing costly orders are a hard requirement which we | |
4278 | * are not prepared for much so let's warn about these users | |
4279 | * so that we can identify them and convert them to something | |
4280 | * else. | |
4281 | */ | |
4282 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4283 | ||
6c18ba7a MH |
4284 | /* |
4285 | * Help non-failing allocations by giving them access to memory | |
4286 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4287 | * could deplete whole memory reserves which would just make | |
4288 | * the situation worse | |
4289 | */ | |
4290 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4291 | if (page) | |
4292 | goto got_pg; | |
4293 | ||
9a67f648 MH |
4294 | cond_resched(); |
4295 | goto retry; | |
4296 | } | |
4297 | fail: | |
a8e99259 | 4298 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4299 | "page allocation failure: order:%u", order); |
1da177e4 | 4300 | got_pg: |
072bb0aa | 4301 | return page; |
1da177e4 | 4302 | } |
11e33f6a | 4303 | |
9cd75558 | 4304 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4305 | int preferred_nid, nodemask_t *nodemask, |
9cd75558 MG |
4306 | struct alloc_context *ac, gfp_t *alloc_mask, |
4307 | unsigned int *alloc_flags) | |
11e33f6a | 4308 | { |
9cd75558 | 4309 | ac->high_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4310 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 MG |
4311 | ac->nodemask = nodemask; |
4312 | ac->migratetype = gfpflags_to_migratetype(gfp_mask); | |
11e33f6a | 4313 | |
682a3385 | 4314 | if (cpusets_enabled()) { |
9cd75558 | 4315 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
4316 | if (!ac->nodemask) |
4317 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
4318 | else |
4319 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4320 | } |
4321 | ||
d92a8cfc PZ |
4322 | fs_reclaim_acquire(gfp_mask); |
4323 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4324 | |
d0164adc | 4325 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4326 | |
4327 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4328 | return false; |
11e33f6a | 4329 | |
d883c6cf JK |
4330 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
4331 | *alloc_flags |= ALLOC_CMA; | |
4332 | ||
9cd75558 MG |
4333 | return true; |
4334 | } | |
21bb9bd1 | 4335 | |
9cd75558 | 4336 | /* Determine whether to spread dirty pages and what the first usable zone */ |
a380b40a | 4337 | static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac) |
9cd75558 | 4338 | { |
c9ab0c4f | 4339 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4340 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4341 | |
e46e7b77 MG |
4342 | /* |
4343 | * The preferred zone is used for statistics but crucially it is | |
4344 | * also used as the starting point for the zonelist iterator. It | |
4345 | * may get reset for allocations that ignore memory policies. | |
4346 | */ | |
9cd75558 MG |
4347 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4348 | ac->high_zoneidx, ac->nodemask); | |
4349 | } | |
4350 | ||
4351 | /* | |
4352 | * This is the 'heart' of the zoned buddy allocator. | |
4353 | */ | |
4354 | struct page * | |
04ec6264 VB |
4355 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
4356 | nodemask_t *nodemask) | |
9cd75558 MG |
4357 | { |
4358 | struct page *page; | |
4359 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
f19360f0 | 4360 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4361 | struct alloc_context ac = { }; |
4362 | ||
4363 | gfp_mask &= gfp_allowed_mask; | |
f19360f0 | 4364 | alloc_mask = gfp_mask; |
04ec6264 | 4365 | if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags)) |
9cd75558 MG |
4366 | return NULL; |
4367 | ||
a380b40a | 4368 | finalise_ac(gfp_mask, &ac); |
5bb1b169 | 4369 | |
5117f45d | 4370 | /* First allocation attempt */ |
a9263751 | 4371 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4372 | if (likely(page)) |
4373 | goto out; | |
11e33f6a | 4374 | |
4fcb0971 | 4375 | /* |
7dea19f9 MH |
4376 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4377 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4378 | * from a particular context which has been marked by | |
4379 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4380 | */ |
7dea19f9 | 4381 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4382 | ac.spread_dirty_pages = false; |
23f086f9 | 4383 | |
4741526b MG |
4384 | /* |
4385 | * Restore the original nodemask if it was potentially replaced with | |
4386 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4387 | */ | |
e47483bc | 4388 | if (unlikely(ac.nodemask != nodemask)) |
4741526b | 4389 | ac.nodemask = nodemask; |
16096c25 | 4390 | |
4fcb0971 | 4391 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4392 | |
4fcb0971 | 4393 | out: |
c4159a75 VD |
4394 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
4395 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { | |
4396 | __free_pages(page, order); | |
4397 | page = NULL; | |
4949148a VD |
4398 | } |
4399 | ||
4fcb0971 MG |
4400 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
4401 | ||
11e33f6a | 4402 | return page; |
1da177e4 | 4403 | } |
d239171e | 4404 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4405 | |
4406 | /* | |
9ea9a680 MH |
4407 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4408 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4409 | * you need to access high mem. | |
1da177e4 | 4410 | */ |
920c7a5d | 4411 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4412 | { |
945a1113 AM |
4413 | struct page *page; |
4414 | ||
9ea9a680 | 4415 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4416 | if (!page) |
4417 | return 0; | |
4418 | return (unsigned long) page_address(page); | |
4419 | } | |
1da177e4 LT |
4420 | EXPORT_SYMBOL(__get_free_pages); |
4421 | ||
920c7a5d | 4422 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4423 | { |
945a1113 | 4424 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4425 | } |
1da177e4 LT |
4426 | EXPORT_SYMBOL(get_zeroed_page); |
4427 | ||
920c7a5d | 4428 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 4429 | { |
b5810039 | 4430 | if (put_page_testzero(page)) { |
1da177e4 | 4431 | if (order == 0) |
2d4894b5 | 4432 | free_unref_page(page); |
1da177e4 LT |
4433 | else |
4434 | __free_pages_ok(page, order); | |
4435 | } | |
4436 | } | |
4437 | ||
4438 | EXPORT_SYMBOL(__free_pages); | |
4439 | ||
920c7a5d | 4440 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4441 | { |
4442 | if (addr != 0) { | |
725d704e | 4443 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4444 | __free_pages(virt_to_page((void *)addr), order); |
4445 | } | |
4446 | } | |
4447 | ||
4448 | EXPORT_SYMBOL(free_pages); | |
4449 | ||
b63ae8ca AD |
4450 | /* |
4451 | * Page Fragment: | |
4452 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4453 | * within a 0 or higher order page. Multiple fragments within that page | |
4454 | * are individually refcounted, in the page's reference counter. | |
4455 | * | |
4456 | * The page_frag functions below provide a simple allocation framework for | |
4457 | * page fragments. This is used by the network stack and network device | |
4458 | * drivers to provide a backing region of memory for use as either an | |
4459 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4460 | */ | |
2976db80 AD |
4461 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4462 | gfp_t gfp_mask) | |
b63ae8ca AD |
4463 | { |
4464 | struct page *page = NULL; | |
4465 | gfp_t gfp = gfp_mask; | |
4466 | ||
4467 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4468 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4469 | __GFP_NOMEMALLOC; | |
4470 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4471 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4472 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4473 | #endif | |
4474 | if (unlikely(!page)) | |
4475 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4476 | ||
4477 | nc->va = page ? page_address(page) : NULL; | |
4478 | ||
4479 | return page; | |
4480 | } | |
4481 | ||
2976db80 | 4482 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4483 | { |
4484 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4485 | ||
4486 | if (page_ref_sub_and_test(page, count)) { | |
2976db80 AD |
4487 | unsigned int order = compound_order(page); |
4488 | ||
44fdffd7 | 4489 | if (order == 0) |
2d4894b5 | 4490 | free_unref_page(page); |
44fdffd7 AD |
4491 | else |
4492 | __free_pages_ok(page, order); | |
4493 | } | |
4494 | } | |
2976db80 | 4495 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4496 | |
8c2dd3e4 AD |
4497 | void *page_frag_alloc(struct page_frag_cache *nc, |
4498 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4499 | { |
4500 | unsigned int size = PAGE_SIZE; | |
4501 | struct page *page; | |
4502 | int offset; | |
4503 | ||
4504 | if (unlikely(!nc->va)) { | |
4505 | refill: | |
2976db80 | 4506 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4507 | if (!page) |
4508 | return NULL; | |
4509 | ||
4510 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4511 | /* if size can vary use size else just use PAGE_SIZE */ | |
4512 | size = nc->size; | |
4513 | #endif | |
4514 | /* Even if we own the page, we do not use atomic_set(). | |
4515 | * This would break get_page_unless_zero() users. | |
4516 | */ | |
fe896d18 | 4517 | page_ref_add(page, size - 1); |
b63ae8ca AD |
4518 | |
4519 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4520 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
4521 | nc->pagecnt_bias = size; |
4522 | nc->offset = size; | |
4523 | } | |
4524 | ||
4525 | offset = nc->offset - fragsz; | |
4526 | if (unlikely(offset < 0)) { | |
4527 | page = virt_to_page(nc->va); | |
4528 | ||
fe896d18 | 4529 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4530 | goto refill; |
4531 | ||
4532 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4533 | /* if size can vary use size else just use PAGE_SIZE */ | |
4534 | size = nc->size; | |
4535 | #endif | |
4536 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 4537 | set_page_count(page, size); |
b63ae8ca AD |
4538 | |
4539 | /* reset page count bias and offset to start of new frag */ | |
4540 | nc->pagecnt_bias = size; | |
4541 | offset = size - fragsz; | |
4542 | } | |
4543 | ||
4544 | nc->pagecnt_bias--; | |
4545 | nc->offset = offset; | |
4546 | ||
4547 | return nc->va + offset; | |
4548 | } | |
8c2dd3e4 | 4549 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
4550 | |
4551 | /* | |
4552 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4553 | */ | |
8c2dd3e4 | 4554 | void page_frag_free(void *addr) |
b63ae8ca AD |
4555 | { |
4556 | struct page *page = virt_to_head_page(addr); | |
4557 | ||
4558 | if (unlikely(put_page_testzero(page))) | |
4559 | __free_pages_ok(page, compound_order(page)); | |
4560 | } | |
8c2dd3e4 | 4561 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4562 | |
d00181b9 KS |
4563 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4564 | size_t size) | |
ee85c2e1 AK |
4565 | { |
4566 | if (addr) { | |
4567 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
4568 | unsigned long used = addr + PAGE_ALIGN(size); | |
4569 | ||
4570 | split_page(virt_to_page((void *)addr), order); | |
4571 | while (used < alloc_end) { | |
4572 | free_page(used); | |
4573 | used += PAGE_SIZE; | |
4574 | } | |
4575 | } | |
4576 | return (void *)addr; | |
4577 | } | |
4578 | ||
2be0ffe2 TT |
4579 | /** |
4580 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4581 | * @size: the number of bytes to allocate | |
4582 | * @gfp_mask: GFP flags for the allocation | |
4583 | * | |
4584 | * This function is similar to alloc_pages(), except that it allocates the | |
4585 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4586 | * allocate memory in power-of-two pages. | |
4587 | * | |
4588 | * This function is also limited by MAX_ORDER. | |
4589 | * | |
4590 | * Memory allocated by this function must be released by free_pages_exact(). | |
4591 | */ | |
4592 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4593 | { | |
4594 | unsigned int order = get_order(size); | |
4595 | unsigned long addr; | |
4596 | ||
4597 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 4598 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4599 | } |
4600 | EXPORT_SYMBOL(alloc_pages_exact); | |
4601 | ||
ee85c2e1 AK |
4602 | /** |
4603 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4604 | * pages on a node. | |
b5e6ab58 | 4605 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
4606 | * @size: the number of bytes to allocate |
4607 | * @gfp_mask: GFP flags for the allocation | |
4608 | * | |
4609 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4610 | * back. | |
ee85c2e1 | 4611 | */ |
e1931811 | 4612 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4613 | { |
d00181b9 | 4614 | unsigned int order = get_order(size); |
ee85c2e1 AK |
4615 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
4616 | if (!p) | |
4617 | return NULL; | |
4618 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4619 | } | |
ee85c2e1 | 4620 | |
2be0ffe2 TT |
4621 | /** |
4622 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4623 | * @virt: the value returned by alloc_pages_exact. | |
4624 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4625 | * | |
4626 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4627 | */ | |
4628 | void free_pages_exact(void *virt, size_t size) | |
4629 | { | |
4630 | unsigned long addr = (unsigned long)virt; | |
4631 | unsigned long end = addr + PAGE_ALIGN(size); | |
4632 | ||
4633 | while (addr < end) { | |
4634 | free_page(addr); | |
4635 | addr += PAGE_SIZE; | |
4636 | } | |
4637 | } | |
4638 | EXPORT_SYMBOL(free_pages_exact); | |
4639 | ||
e0fb5815 ZY |
4640 | /** |
4641 | * nr_free_zone_pages - count number of pages beyond high watermark | |
4642 | * @offset: The zone index of the highest zone | |
4643 | * | |
4644 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
4645 | * high watermark within all zones at or below a given zone index. For each | |
4646 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
4647 | * |
4648 | * nr_free_zone_pages = managed_pages - high_pages | |
e0fb5815 | 4649 | */ |
ebec3862 | 4650 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 4651 | { |
dd1a239f | 4652 | struct zoneref *z; |
54a6eb5c MG |
4653 | struct zone *zone; |
4654 | ||
e310fd43 | 4655 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 4656 | unsigned long sum = 0; |
1da177e4 | 4657 | |
0e88460d | 4658 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 4659 | |
54a6eb5c | 4660 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 4661 | unsigned long size = zone->managed_pages; |
41858966 | 4662 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
4663 | if (size > high) |
4664 | sum += size - high; | |
1da177e4 LT |
4665 | } |
4666 | ||
4667 | return sum; | |
4668 | } | |
4669 | ||
e0fb5815 ZY |
4670 | /** |
4671 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
4672 | * | |
4673 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
4674 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 4675 | */ |
ebec3862 | 4676 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 4677 | { |
af4ca457 | 4678 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 4679 | } |
c2f1a551 | 4680 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 4681 | |
e0fb5815 ZY |
4682 | /** |
4683 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
4684 | * | |
4685 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
4686 | * high watermark within all zones. | |
1da177e4 | 4687 | */ |
ebec3862 | 4688 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 4689 | { |
2a1e274a | 4690 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 4691 | } |
08e0f6a9 CL |
4692 | |
4693 | static inline void show_node(struct zone *zone) | |
1da177e4 | 4694 | { |
e5adfffc | 4695 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 4696 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 4697 | } |
1da177e4 | 4698 | |
d02bd27b IR |
4699 | long si_mem_available(void) |
4700 | { | |
4701 | long available; | |
4702 | unsigned long pagecache; | |
4703 | unsigned long wmark_low = 0; | |
4704 | unsigned long pages[NR_LRU_LISTS]; | |
4705 | struct zone *zone; | |
4706 | int lru; | |
4707 | ||
4708 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 4709 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
4710 | |
4711 | for_each_zone(zone) | |
4712 | wmark_low += zone->watermark[WMARK_LOW]; | |
4713 | ||
4714 | /* | |
4715 | * Estimate the amount of memory available for userspace allocations, | |
4716 | * without causing swapping. | |
4717 | */ | |
c41f012a | 4718 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
4719 | |
4720 | /* | |
4721 | * Not all the page cache can be freed, otherwise the system will | |
4722 | * start swapping. Assume at least half of the page cache, or the | |
4723 | * low watermark worth of cache, needs to stay. | |
4724 | */ | |
4725 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
4726 | pagecache -= min(pagecache / 2, wmark_low); | |
4727 | available += pagecache; | |
4728 | ||
4729 | /* | |
4730 | * Part of the reclaimable slab consists of items that are in use, | |
4731 | * and cannot be freed. Cap this estimate at the low watermark. | |
4732 | */ | |
d507e2eb JW |
4733 | available += global_node_page_state(NR_SLAB_RECLAIMABLE) - |
4734 | min(global_node_page_state(NR_SLAB_RECLAIMABLE) / 2, | |
4735 | wmark_low); | |
d02bd27b | 4736 | |
034ebf65 RG |
4737 | /* |
4738 | * Part of the kernel memory, which can be released under memory | |
4739 | * pressure. | |
4740 | */ | |
4741 | available += global_node_page_state(NR_INDIRECTLY_RECLAIMABLE_BYTES) >> | |
4742 | PAGE_SHIFT; | |
4743 | ||
d02bd27b IR |
4744 | if (available < 0) |
4745 | available = 0; | |
4746 | return available; | |
4747 | } | |
4748 | EXPORT_SYMBOL_GPL(si_mem_available); | |
4749 | ||
1da177e4 LT |
4750 | void si_meminfo(struct sysinfo *val) |
4751 | { | |
4752 | val->totalram = totalram_pages; | |
11fb9989 | 4753 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 4754 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 4755 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
4756 | val->totalhigh = totalhigh_pages; |
4757 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
4758 | val->mem_unit = PAGE_SIZE; |
4759 | } | |
4760 | ||
4761 | EXPORT_SYMBOL(si_meminfo); | |
4762 | ||
4763 | #ifdef CONFIG_NUMA | |
4764 | void si_meminfo_node(struct sysinfo *val, int nid) | |
4765 | { | |
cdd91a77 JL |
4766 | int zone_type; /* needs to be signed */ |
4767 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
4768 | unsigned long managed_highpages = 0; |
4769 | unsigned long free_highpages = 0; | |
1da177e4 LT |
4770 | pg_data_t *pgdat = NODE_DATA(nid); |
4771 | ||
cdd91a77 JL |
4772 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
4773 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
4774 | val->totalram = managed_pages; | |
11fb9989 | 4775 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 4776 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 4777 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
4778 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4779 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4780 | ||
4781 | if (is_highmem(zone)) { | |
4782 | managed_highpages += zone->managed_pages; | |
4783 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
4784 | } | |
4785 | } | |
4786 | val->totalhigh = managed_highpages; | |
4787 | val->freehigh = free_highpages; | |
98d2b0eb | 4788 | #else |
fc2bd799 JK |
4789 | val->totalhigh = managed_highpages; |
4790 | val->freehigh = free_highpages; | |
98d2b0eb | 4791 | #endif |
1da177e4 LT |
4792 | val->mem_unit = PAGE_SIZE; |
4793 | } | |
4794 | #endif | |
4795 | ||
ddd588b5 | 4796 | /* |
7bf02ea2 DR |
4797 | * Determine whether the node should be displayed or not, depending on whether |
4798 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 4799 | */ |
9af744d7 | 4800 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 4801 | { |
ddd588b5 | 4802 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 4803 | return false; |
ddd588b5 | 4804 | |
9af744d7 MH |
4805 | /* |
4806 | * no node mask - aka implicit memory numa policy. Do not bother with | |
4807 | * the synchronization - read_mems_allowed_begin - because we do not | |
4808 | * have to be precise here. | |
4809 | */ | |
4810 | if (!nodemask) | |
4811 | nodemask = &cpuset_current_mems_allowed; | |
4812 | ||
4813 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
4814 | } |
4815 | ||
1da177e4 LT |
4816 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4817 | ||
377e4f16 RV |
4818 | static void show_migration_types(unsigned char type) |
4819 | { | |
4820 | static const char types[MIGRATE_TYPES] = { | |
4821 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 4822 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
4823 | [MIGRATE_RECLAIMABLE] = 'E', |
4824 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
4825 | #ifdef CONFIG_CMA |
4826 | [MIGRATE_CMA] = 'C', | |
4827 | #endif | |
194159fb | 4828 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 4829 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 4830 | #endif |
377e4f16 RV |
4831 | }; |
4832 | char tmp[MIGRATE_TYPES + 1]; | |
4833 | char *p = tmp; | |
4834 | int i; | |
4835 | ||
4836 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
4837 | if (type & (1 << i)) | |
4838 | *p++ = types[i]; | |
4839 | } | |
4840 | ||
4841 | *p = '\0'; | |
1f84a18f | 4842 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
4843 | } |
4844 | ||
1da177e4 LT |
4845 | /* |
4846 | * Show free area list (used inside shift_scroll-lock stuff) | |
4847 | * We also calculate the percentage fragmentation. We do this by counting the | |
4848 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
4849 | * |
4850 | * Bits in @filter: | |
4851 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
4852 | * cpuset. | |
1da177e4 | 4853 | */ |
9af744d7 | 4854 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 4855 | { |
d1bfcdb8 | 4856 | unsigned long free_pcp = 0; |
c7241913 | 4857 | int cpu; |
1da177e4 | 4858 | struct zone *zone; |
599d0c95 | 4859 | pg_data_t *pgdat; |
1da177e4 | 4860 | |
ee99c71c | 4861 | for_each_populated_zone(zone) { |
9af744d7 | 4862 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4863 | continue; |
d1bfcdb8 | 4864 | |
761b0677 KK |
4865 | for_each_online_cpu(cpu) |
4866 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
4867 | } |
4868 | ||
a731286d KM |
4869 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4870 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
4871 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4872 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 4873 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 4874 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
4875 | global_node_page_state(NR_ACTIVE_ANON), |
4876 | global_node_page_state(NR_INACTIVE_ANON), | |
4877 | global_node_page_state(NR_ISOLATED_ANON), | |
4878 | global_node_page_state(NR_ACTIVE_FILE), | |
4879 | global_node_page_state(NR_INACTIVE_FILE), | |
4880 | global_node_page_state(NR_ISOLATED_FILE), | |
4881 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
4882 | global_node_page_state(NR_FILE_DIRTY), |
4883 | global_node_page_state(NR_WRITEBACK), | |
4884 | global_node_page_state(NR_UNSTABLE_NFS), | |
d507e2eb JW |
4885 | global_node_page_state(NR_SLAB_RECLAIMABLE), |
4886 | global_node_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 4887 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 4888 | global_node_page_state(NR_SHMEM), |
c41f012a MH |
4889 | global_zone_page_state(NR_PAGETABLE), |
4890 | global_zone_page_state(NR_BOUNCE), | |
4891 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 4892 | free_pcp, |
c41f012a | 4893 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4894 | |
599d0c95 | 4895 | for_each_online_pgdat(pgdat) { |
9af744d7 | 4896 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
4897 | continue; |
4898 | ||
599d0c95 MG |
4899 | printk("Node %d" |
4900 | " active_anon:%lukB" | |
4901 | " inactive_anon:%lukB" | |
4902 | " active_file:%lukB" | |
4903 | " inactive_file:%lukB" | |
4904 | " unevictable:%lukB" | |
4905 | " isolated(anon):%lukB" | |
4906 | " isolated(file):%lukB" | |
50658e2e | 4907 | " mapped:%lukB" |
11fb9989 MG |
4908 | " dirty:%lukB" |
4909 | " writeback:%lukB" | |
4910 | " shmem:%lukB" | |
4911 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4912 | " shmem_thp: %lukB" | |
4913 | " shmem_pmdmapped: %lukB" | |
4914 | " anon_thp: %lukB" | |
4915 | #endif | |
4916 | " writeback_tmp:%lukB" | |
4917 | " unstable:%lukB" | |
599d0c95 MG |
4918 | " all_unreclaimable? %s" |
4919 | "\n", | |
4920 | pgdat->node_id, | |
4921 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
4922 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
4923 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
4924 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
4925 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
4926 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
4927 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 4928 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
4929 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
4930 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 4931 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
4932 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4933 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
4934 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
4935 | * HPAGE_PMD_NR), | |
4936 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
4937 | #endif | |
11fb9989 MG |
4938 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
4939 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), | |
c73322d0 JW |
4940 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
4941 | "yes" : "no"); | |
599d0c95 MG |
4942 | } |
4943 | ||
ee99c71c | 4944 | for_each_populated_zone(zone) { |
1da177e4 LT |
4945 | int i; |
4946 | ||
9af744d7 | 4947 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4948 | continue; |
d1bfcdb8 KK |
4949 | |
4950 | free_pcp = 0; | |
4951 | for_each_online_cpu(cpu) | |
4952 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
4953 | ||
1da177e4 | 4954 | show_node(zone); |
1f84a18f JP |
4955 | printk(KERN_CONT |
4956 | "%s" | |
1da177e4 LT |
4957 | " free:%lukB" |
4958 | " min:%lukB" | |
4959 | " low:%lukB" | |
4960 | " high:%lukB" | |
71c799f4 MK |
4961 | " active_anon:%lukB" |
4962 | " inactive_anon:%lukB" | |
4963 | " active_file:%lukB" | |
4964 | " inactive_file:%lukB" | |
4965 | " unevictable:%lukB" | |
5a1c84b4 | 4966 | " writepending:%lukB" |
1da177e4 | 4967 | " present:%lukB" |
9feedc9d | 4968 | " managed:%lukB" |
4a0aa73f | 4969 | " mlocked:%lukB" |
c6a7f572 | 4970 | " kernel_stack:%lukB" |
4a0aa73f | 4971 | " pagetables:%lukB" |
4a0aa73f | 4972 | " bounce:%lukB" |
d1bfcdb8 KK |
4973 | " free_pcp:%lukB" |
4974 | " local_pcp:%ukB" | |
d1ce749a | 4975 | " free_cma:%lukB" |
1da177e4 LT |
4976 | "\n", |
4977 | zone->name, | |
88f5acf8 | 4978 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
4979 | K(min_wmark_pages(zone)), |
4980 | K(low_wmark_pages(zone)), | |
4981 | K(high_wmark_pages(zone)), | |
71c799f4 MK |
4982 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
4983 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
4984 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
4985 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
4986 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 4987 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 4988 | K(zone->present_pages), |
9feedc9d | 4989 | K(zone->managed_pages), |
4a0aa73f | 4990 | K(zone_page_state(zone, NR_MLOCK)), |
d30dd8be | 4991 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4a0aa73f | 4992 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 4993 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
4994 | K(free_pcp), |
4995 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 4996 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
4997 | printk("lowmem_reserve[]:"); |
4998 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
4999 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5000 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5001 | } |
5002 | ||
ee99c71c | 5003 | for_each_populated_zone(zone) { |
d00181b9 KS |
5004 | unsigned int order; |
5005 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5006 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5007 | |
9af744d7 | 5008 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5009 | continue; |
1da177e4 | 5010 | show_node(zone); |
1f84a18f | 5011 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5012 | |
5013 | spin_lock_irqsave(&zone->lock, flags); | |
5014 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5015 | struct free_area *area = &zone->free_area[order]; |
5016 | int type; | |
5017 | ||
5018 | nr[order] = area->nr_free; | |
8f9de51a | 5019 | total += nr[order] << order; |
377e4f16 RV |
5020 | |
5021 | types[order] = 0; | |
5022 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
5023 | if (!list_empty(&area->free_list[type])) | |
5024 | types[order] |= 1 << type; | |
5025 | } | |
1da177e4 LT |
5026 | } |
5027 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5028 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5029 | printk(KERN_CONT "%lu*%lukB ", |
5030 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5031 | if (nr[order]) |
5032 | show_migration_types(types[order]); | |
5033 | } | |
1f84a18f | 5034 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5035 | } |
5036 | ||
949f7ec5 DR |
5037 | hugetlb_show_meminfo(); |
5038 | ||
11fb9989 | 5039 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5040 | |
1da177e4 LT |
5041 | show_swap_cache_info(); |
5042 | } | |
5043 | ||
19770b32 MG |
5044 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5045 | { | |
5046 | zoneref->zone = zone; | |
5047 | zoneref->zone_idx = zone_idx(zone); | |
5048 | } | |
5049 | ||
1da177e4 LT |
5050 | /* |
5051 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5052 | * |
5053 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5054 | */ |
9d3be21b | 5055 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5056 | { |
1a93205b | 5057 | struct zone *zone; |
bc732f1d | 5058 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5059 | int nr_zones = 0; |
02a68a5e CL |
5060 | |
5061 | do { | |
2f6726e5 | 5062 | zone_type--; |
070f8032 | 5063 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5064 | if (managed_zone(zone)) { |
9d3be21b | 5065 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5066 | check_highest_zone(zone_type); |
1da177e4 | 5067 | } |
2f6726e5 | 5068 | } while (zone_type); |
bc732f1d | 5069 | |
070f8032 | 5070 | return nr_zones; |
1da177e4 LT |
5071 | } |
5072 | ||
5073 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5074 | |
5075 | static int __parse_numa_zonelist_order(char *s) | |
5076 | { | |
c9bff3ee MH |
5077 | /* |
5078 | * We used to support different zonlists modes but they turned | |
5079 | * out to be just not useful. Let's keep the warning in place | |
5080 | * if somebody still use the cmd line parameter so that we do | |
5081 | * not fail it silently | |
5082 | */ | |
5083 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5084 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5085 | return -EINVAL; |
5086 | } | |
5087 | return 0; | |
5088 | } | |
5089 | ||
5090 | static __init int setup_numa_zonelist_order(char *s) | |
5091 | { | |
ecb256f8 VL |
5092 | if (!s) |
5093 | return 0; | |
5094 | ||
c9bff3ee | 5095 | return __parse_numa_zonelist_order(s); |
f0c0b2b8 KH |
5096 | } |
5097 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
5098 | ||
c9bff3ee MH |
5099 | char numa_zonelist_order[] = "Node"; |
5100 | ||
f0c0b2b8 KH |
5101 | /* |
5102 | * sysctl handler for numa_zonelist_order | |
5103 | */ | |
cccad5b9 | 5104 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 5105 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
5106 | loff_t *ppos) |
5107 | { | |
c9bff3ee | 5108 | char *str; |
f0c0b2b8 KH |
5109 | int ret; |
5110 | ||
c9bff3ee MH |
5111 | if (!write) |
5112 | return proc_dostring(table, write, buffer, length, ppos); | |
5113 | str = memdup_user_nul(buffer, 16); | |
5114 | if (IS_ERR(str)) | |
5115 | return PTR_ERR(str); | |
dacbde09 | 5116 | |
c9bff3ee MH |
5117 | ret = __parse_numa_zonelist_order(str); |
5118 | kfree(str); | |
443c6f14 | 5119 | return ret; |
f0c0b2b8 KH |
5120 | } |
5121 | ||
5122 | ||
62bc62a8 | 5123 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5124 | static int node_load[MAX_NUMNODES]; |
5125 | ||
1da177e4 | 5126 | /** |
4dc3b16b | 5127 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5128 | * @node: node whose fallback list we're appending |
5129 | * @used_node_mask: nodemask_t of already used nodes | |
5130 | * | |
5131 | * We use a number of factors to determine which is the next node that should | |
5132 | * appear on a given node's fallback list. The node should not have appeared | |
5133 | * already in @node's fallback list, and it should be the next closest node | |
5134 | * according to the distance array (which contains arbitrary distance values | |
5135 | * from each node to each node in the system), and should also prefer nodes | |
5136 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5137 | * on them otherwise. | |
5138 | * It returns -1 if no node is found. | |
5139 | */ | |
f0c0b2b8 | 5140 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5141 | { |
4cf808eb | 5142 | int n, val; |
1da177e4 | 5143 | int min_val = INT_MAX; |
00ef2d2f | 5144 | int best_node = NUMA_NO_NODE; |
a70f7302 | 5145 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 5146 | |
4cf808eb LT |
5147 | /* Use the local node if we haven't already */ |
5148 | if (!node_isset(node, *used_node_mask)) { | |
5149 | node_set(node, *used_node_mask); | |
5150 | return node; | |
5151 | } | |
1da177e4 | 5152 | |
4b0ef1fe | 5153 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5154 | |
5155 | /* Don't want a node to appear more than once */ | |
5156 | if (node_isset(n, *used_node_mask)) | |
5157 | continue; | |
5158 | ||
1da177e4 LT |
5159 | /* Use the distance array to find the distance */ |
5160 | val = node_distance(node, n); | |
5161 | ||
4cf808eb LT |
5162 | /* Penalize nodes under us ("prefer the next node") */ |
5163 | val += (n < node); | |
5164 | ||
1da177e4 | 5165 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
5166 | tmp = cpumask_of_node(n); |
5167 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
5168 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5169 | ||
5170 | /* Slight preference for less loaded node */ | |
5171 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
5172 | val += node_load[n]; | |
5173 | ||
5174 | if (val < min_val) { | |
5175 | min_val = val; | |
5176 | best_node = n; | |
5177 | } | |
5178 | } | |
5179 | ||
5180 | if (best_node >= 0) | |
5181 | node_set(best_node, *used_node_mask); | |
5182 | ||
5183 | return best_node; | |
5184 | } | |
5185 | ||
f0c0b2b8 KH |
5186 | |
5187 | /* | |
5188 | * Build zonelists ordered by node and zones within node. | |
5189 | * This results in maximum locality--normal zone overflows into local | |
5190 | * DMA zone, if any--but risks exhausting DMA zone. | |
5191 | */ | |
9d3be21b MH |
5192 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5193 | unsigned nr_nodes) | |
1da177e4 | 5194 | { |
9d3be21b MH |
5195 | struct zoneref *zonerefs; |
5196 | int i; | |
5197 | ||
5198 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5199 | ||
5200 | for (i = 0; i < nr_nodes; i++) { | |
5201 | int nr_zones; | |
5202 | ||
5203 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5204 | |
9d3be21b MH |
5205 | nr_zones = build_zonerefs_node(node, zonerefs); |
5206 | zonerefs += nr_zones; | |
5207 | } | |
5208 | zonerefs->zone = NULL; | |
5209 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5210 | } |
5211 | ||
523b9458 CL |
5212 | /* |
5213 | * Build gfp_thisnode zonelists | |
5214 | */ | |
5215 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5216 | { | |
9d3be21b MH |
5217 | struct zoneref *zonerefs; |
5218 | int nr_zones; | |
523b9458 | 5219 | |
9d3be21b MH |
5220 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5221 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5222 | zonerefs += nr_zones; | |
5223 | zonerefs->zone = NULL; | |
5224 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5225 | } |
5226 | ||
f0c0b2b8 KH |
5227 | /* |
5228 | * Build zonelists ordered by zone and nodes within zones. | |
5229 | * This results in conserving DMA zone[s] until all Normal memory is | |
5230 | * exhausted, but results in overflowing to remote node while memory | |
5231 | * may still exist in local DMA zone. | |
5232 | */ | |
f0c0b2b8 | 5233 | |
f0c0b2b8 KH |
5234 | static void build_zonelists(pg_data_t *pgdat) |
5235 | { | |
9d3be21b MH |
5236 | static int node_order[MAX_NUMNODES]; |
5237 | int node, load, nr_nodes = 0; | |
1da177e4 | 5238 | nodemask_t used_mask; |
f0c0b2b8 | 5239 | int local_node, prev_node; |
1da177e4 LT |
5240 | |
5241 | /* NUMA-aware ordering of nodes */ | |
5242 | local_node = pgdat->node_id; | |
62bc62a8 | 5243 | load = nr_online_nodes; |
1da177e4 LT |
5244 | prev_node = local_node; |
5245 | nodes_clear(used_mask); | |
f0c0b2b8 | 5246 | |
f0c0b2b8 | 5247 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5248 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5249 | /* | |
5250 | * We don't want to pressure a particular node. | |
5251 | * So adding penalty to the first node in same | |
5252 | * distance group to make it round-robin. | |
5253 | */ | |
957f822a DR |
5254 | if (node_distance(local_node, node) != |
5255 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5256 | node_load[node] = load; |
5257 | ||
9d3be21b | 5258 | node_order[nr_nodes++] = node; |
1da177e4 LT |
5259 | prev_node = node; |
5260 | load--; | |
1da177e4 | 5261 | } |
523b9458 | 5262 | |
9d3be21b | 5263 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5264 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
5265 | } |
5266 | ||
7aac7898 LS |
5267 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5268 | /* | |
5269 | * Return node id of node used for "local" allocations. | |
5270 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5271 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5272 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5273 | */ | |
5274 | int local_memory_node(int node) | |
5275 | { | |
c33d6c06 | 5276 | struct zoneref *z; |
7aac7898 | 5277 | |
c33d6c06 | 5278 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5279 | gfp_zone(GFP_KERNEL), |
c33d6c06 MG |
5280 | NULL); |
5281 | return z->zone->node; | |
7aac7898 LS |
5282 | } |
5283 | #endif | |
f0c0b2b8 | 5284 | |
6423aa81 JK |
5285 | static void setup_min_unmapped_ratio(void); |
5286 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5287 | #else /* CONFIG_NUMA */ |
5288 | ||
f0c0b2b8 | 5289 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5290 | { |
19655d34 | 5291 | int node, local_node; |
9d3be21b MH |
5292 | struct zoneref *zonerefs; |
5293 | int nr_zones; | |
1da177e4 LT |
5294 | |
5295 | local_node = pgdat->node_id; | |
1da177e4 | 5296 | |
9d3be21b MH |
5297 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5298 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5299 | zonerefs += nr_zones; | |
1da177e4 | 5300 | |
54a6eb5c MG |
5301 | /* |
5302 | * Now we build the zonelist so that it contains the zones | |
5303 | * of all the other nodes. | |
5304 | * We don't want to pressure a particular node, so when | |
5305 | * building the zones for node N, we make sure that the | |
5306 | * zones coming right after the local ones are those from | |
5307 | * node N+1 (modulo N) | |
5308 | */ | |
5309 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5310 | if (!node_online(node)) | |
5311 | continue; | |
9d3be21b MH |
5312 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5313 | zonerefs += nr_zones; | |
1da177e4 | 5314 | } |
54a6eb5c MG |
5315 | for (node = 0; node < local_node; node++) { |
5316 | if (!node_online(node)) | |
5317 | continue; | |
9d3be21b MH |
5318 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5319 | zonerefs += nr_zones; | |
54a6eb5c MG |
5320 | } |
5321 | ||
9d3be21b MH |
5322 | zonerefs->zone = NULL; |
5323 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5324 | } |
5325 | ||
5326 | #endif /* CONFIG_NUMA */ | |
5327 | ||
99dcc3e5 CL |
5328 | /* |
5329 | * Boot pageset table. One per cpu which is going to be used for all | |
5330 | * zones and all nodes. The parameters will be set in such a way | |
5331 | * that an item put on a list will immediately be handed over to | |
5332 | * the buddy list. This is safe since pageset manipulation is done | |
5333 | * with interrupts disabled. | |
5334 | * | |
5335 | * The boot_pagesets must be kept even after bootup is complete for | |
5336 | * unused processors and/or zones. They do play a role for bootstrapping | |
5337 | * hotplugged processors. | |
5338 | * | |
5339 | * zoneinfo_show() and maybe other functions do | |
5340 | * not check if the processor is online before following the pageset pointer. | |
5341 | * Other parts of the kernel may not check if the zone is available. | |
5342 | */ | |
5343 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5344 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
385386cf | 5345 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 5346 | |
11cd8638 | 5347 | static void __build_all_zonelists(void *data) |
1da177e4 | 5348 | { |
6811378e | 5349 | int nid; |
afb6ebb3 | 5350 | int __maybe_unused cpu; |
9adb62a5 | 5351 | pg_data_t *self = data; |
b93e0f32 MH |
5352 | static DEFINE_SPINLOCK(lock); |
5353 | ||
5354 | spin_lock(&lock); | |
9276b1bc | 5355 | |
7f9cfb31 BL |
5356 | #ifdef CONFIG_NUMA |
5357 | memset(node_load, 0, sizeof(node_load)); | |
5358 | #endif | |
9adb62a5 | 5359 | |
c1152583 WY |
5360 | /* |
5361 | * This node is hotadded and no memory is yet present. So just | |
5362 | * building zonelists is fine - no need to touch other nodes. | |
5363 | */ | |
9adb62a5 JL |
5364 | if (self && !node_online(self->node_id)) { |
5365 | build_zonelists(self); | |
c1152583 WY |
5366 | } else { |
5367 | for_each_online_node(nid) { | |
5368 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 5369 | |
c1152583 WY |
5370 | build_zonelists(pgdat); |
5371 | } | |
99dcc3e5 | 5372 | |
7aac7898 LS |
5373 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5374 | /* | |
5375 | * We now know the "local memory node" for each node-- | |
5376 | * i.e., the node of the first zone in the generic zonelist. | |
5377 | * Set up numa_mem percpu variable for on-line cpus. During | |
5378 | * boot, only the boot cpu should be on-line; we'll init the | |
5379 | * secondary cpus' numa_mem as they come on-line. During | |
5380 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5381 | */ | |
d9c9a0b9 | 5382 | for_each_online_cpu(cpu) |
7aac7898 | 5383 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5384 | #endif |
d9c9a0b9 | 5385 | } |
b93e0f32 MH |
5386 | |
5387 | spin_unlock(&lock); | |
6811378e YG |
5388 | } |
5389 | ||
061f67bc RV |
5390 | static noinline void __init |
5391 | build_all_zonelists_init(void) | |
5392 | { | |
afb6ebb3 MH |
5393 | int cpu; |
5394 | ||
061f67bc | 5395 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5396 | |
5397 | /* | |
5398 | * Initialize the boot_pagesets that are going to be used | |
5399 | * for bootstrapping processors. The real pagesets for | |
5400 | * each zone will be allocated later when the per cpu | |
5401 | * allocator is available. | |
5402 | * | |
5403 | * boot_pagesets are used also for bootstrapping offline | |
5404 | * cpus if the system is already booted because the pagesets | |
5405 | * are needed to initialize allocators on a specific cpu too. | |
5406 | * F.e. the percpu allocator needs the page allocator which | |
5407 | * needs the percpu allocator in order to allocate its pagesets | |
5408 | * (a chicken-egg dilemma). | |
5409 | */ | |
5410 | for_each_possible_cpu(cpu) | |
5411 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); | |
5412 | ||
061f67bc RV |
5413 | mminit_verify_zonelist(); |
5414 | cpuset_init_current_mems_allowed(); | |
5415 | } | |
5416 | ||
4eaf3f64 | 5417 | /* |
4eaf3f64 | 5418 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5419 | * |
72675e13 | 5420 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5421 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5422 | */ |
72675e13 | 5423 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e YG |
5424 | { |
5425 | if (system_state == SYSTEM_BOOTING) { | |
061f67bc | 5426 | build_all_zonelists_init(); |
6811378e | 5427 | } else { |
11cd8638 | 5428 | __build_all_zonelists(pgdat); |
6811378e YG |
5429 | /* cpuset refresh routine should be here */ |
5430 | } | |
bd1e22b8 | 5431 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5432 | /* |
5433 | * Disable grouping by mobility if the number of pages in the | |
5434 | * system is too low to allow the mechanism to work. It would be | |
5435 | * more accurate, but expensive to check per-zone. This check is | |
5436 | * made on memory-hotadd so a system can start with mobility | |
5437 | * disabled and enable it later | |
5438 | */ | |
d9c23400 | 5439 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5440 | page_group_by_mobility_disabled = 1; |
5441 | else | |
5442 | page_group_by_mobility_disabled = 0; | |
5443 | ||
c9bff3ee | 5444 | pr_info("Built %i zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5445 | nr_online_nodes, |
756a025f JP |
5446 | page_group_by_mobility_disabled ? "off" : "on", |
5447 | vm_total_pages); | |
f0c0b2b8 | 5448 | #ifdef CONFIG_NUMA |
f88dfff5 | 5449 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5450 | #endif |
1da177e4 LT |
5451 | } |
5452 | ||
1da177e4 LT |
5453 | /* |
5454 | * Initially all pages are reserved - free ones are freed | |
5455 | * up by free_all_bootmem() once the early boot process is | |
5456 | * done. Non-atomic initialization, single-pass. | |
5457 | */ | |
c09b4240 | 5458 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a99583e7 CH |
5459 | unsigned long start_pfn, enum memmap_context context, |
5460 | struct vmem_altmap *altmap) | |
1da177e4 | 5461 | { |
29751f69 | 5462 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 5463 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 5464 | unsigned long pfn; |
3a80a7fa | 5465 | unsigned long nr_initialised = 0; |
d0dc12e8 | 5466 | struct page *page; |
342332e6 TI |
5467 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5468 | struct memblock_region *r = NULL, *tmp; | |
5469 | #endif | |
1da177e4 | 5470 | |
22b31eec HD |
5471 | if (highest_memmap_pfn < end_pfn - 1) |
5472 | highest_memmap_pfn = end_pfn - 1; | |
5473 | ||
4b94ffdc DW |
5474 | /* |
5475 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
5476 | * memory | |
5477 | */ | |
5478 | if (altmap && start_pfn == altmap->base_pfn) | |
5479 | start_pfn += altmap->reserve; | |
5480 | ||
cbe8dd4a | 5481 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 5482 | /* |
b72d0ffb AM |
5483 | * There can be holes in boot-time mem_map[]s handed to this |
5484 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5485 | */ |
b72d0ffb AM |
5486 | if (context != MEMMAP_EARLY) |
5487 | goto not_early; | |
5488 | ||
f59f1caf | 5489 | if (!early_pfn_valid(pfn)) |
b72d0ffb AM |
5490 | continue; |
5491 | if (!early_pfn_in_nid(pfn, nid)) | |
5492 | continue; | |
5493 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
5494 | break; | |
342332e6 TI |
5495 | |
5496 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
5497 | /* |
5498 | * Check given memblock attribute by firmware which can affect | |
5499 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
5500 | * mirrored, it's an overlapped memmap init. skip it. | |
5501 | */ | |
5502 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5503 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
5504 | for_each_memblock(memory, tmp) | |
5505 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
5506 | break; | |
5507 | r = tmp; | |
5508 | } | |
5509 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
5510 | memblock_is_mirror(r)) { | |
5511 | /* already initialized as NORMAL */ | |
5512 | pfn = memblock_region_memory_end_pfn(r); | |
5513 | continue; | |
342332e6 | 5514 | } |
a2f3aa02 | 5515 | } |
b72d0ffb | 5516 | #endif |
ac5d2539 | 5517 | |
b72d0ffb | 5518 | not_early: |
d0dc12e8 PT |
5519 | page = pfn_to_page(pfn); |
5520 | __init_single_page(page, pfn, zone, nid); | |
5521 | if (context == MEMMAP_HOTPLUG) | |
5522 | SetPageReserved(page); | |
5523 | ||
ac5d2539 MG |
5524 | /* |
5525 | * Mark the block movable so that blocks are reserved for | |
5526 | * movable at startup. This will force kernel allocations | |
5527 | * to reserve their blocks rather than leaking throughout | |
5528 | * the address space during boot when many long-lived | |
974a786e | 5529 | * kernel allocations are made. |
ac5d2539 MG |
5530 | * |
5531 | * bitmap is created for zone's valid pfn range. but memmap | |
5532 | * can be created for invalid pages (for alignment) | |
5533 | * check here not to call set_pageblock_migratetype() against | |
5534 | * pfn out of zone. | |
9bb5a391 MH |
5535 | * |
5536 | * Please note that MEMMAP_HOTPLUG path doesn't clear memmap | |
5537 | * because this is done early in sparse_add_one_section | |
ac5d2539 MG |
5538 | */ |
5539 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
ac5d2539 | 5540 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
9b6e63cb | 5541 | cond_resched(); |
ac5d2539 | 5542 | } |
1da177e4 LT |
5543 | } |
5544 | } | |
5545 | ||
1e548deb | 5546 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 5547 | { |
7aeb09f9 | 5548 | unsigned int order, t; |
b2a0ac88 MG |
5549 | for_each_migratetype_order(order, t) { |
5550 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
5551 | zone->free_area[order].nr_free = 0; |
5552 | } | |
5553 | } | |
5554 | ||
5555 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
5556 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a99583e7 | 5557 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY, NULL) |
1da177e4 LT |
5558 | #endif |
5559 | ||
7cd2b0a3 | 5560 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 5561 | { |
3a6be87f | 5562 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
5563 | int batch; |
5564 | ||
5565 | /* | |
5566 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 5567 | * size of the zone. |
e7c8d5c9 | 5568 | */ |
b40da049 | 5569 | batch = zone->managed_pages / 1024; |
d8a759b5 AL |
5570 | /* But no more than a meg. */ |
5571 | if (batch * PAGE_SIZE > 1024 * 1024) | |
5572 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
5573 | batch /= 4; /* We effectively *= 4 below */ |
5574 | if (batch < 1) | |
5575 | batch = 1; | |
5576 | ||
5577 | /* | |
0ceaacc9 NP |
5578 | * Clamp the batch to a 2^n - 1 value. Having a power |
5579 | * of 2 value was found to be more likely to have | |
5580 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 5581 | * |
0ceaacc9 NP |
5582 | * For example if 2 tasks are alternately allocating |
5583 | * batches of pages, one task can end up with a lot | |
5584 | * of pages of one half of the possible page colors | |
5585 | * and the other with pages of the other colors. | |
e7c8d5c9 | 5586 | */ |
9155203a | 5587 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 5588 | |
e7c8d5c9 | 5589 | return batch; |
3a6be87f DH |
5590 | |
5591 | #else | |
5592 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5593 | * conditions. | |
5594 | * | |
5595 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5596 | * of contiguous memory as there's no hardware page translation to | |
5597 | * assemble apparent contiguous memory from discontiguous pages. | |
5598 | * | |
5599 | * Queueing large contiguous runs of pages for batching, however, | |
5600 | * causes the pages to actually be freed in smaller chunks. As there | |
5601 | * can be a significant delay between the individual batches being | |
5602 | * recycled, this leads to the once large chunks of space being | |
5603 | * fragmented and becoming unavailable for high-order allocations. | |
5604 | */ | |
5605 | return 0; | |
5606 | #endif | |
e7c8d5c9 CL |
5607 | } |
5608 | ||
8d7a8fa9 CS |
5609 | /* |
5610 | * pcp->high and pcp->batch values are related and dependent on one another: | |
5611 | * ->batch must never be higher then ->high. | |
5612 | * The following function updates them in a safe manner without read side | |
5613 | * locking. | |
5614 | * | |
5615 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
5616 | * those fields changing asynchronously (acording the the above rule). | |
5617 | * | |
5618 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5619 | * outside of boot time (or some other assurance that no concurrent updaters | |
5620 | * exist). | |
5621 | */ | |
5622 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
5623 | unsigned long batch) | |
5624 | { | |
5625 | /* start with a fail safe value for batch */ | |
5626 | pcp->batch = 1; | |
5627 | smp_wmb(); | |
5628 | ||
5629 | /* Update high, then batch, in order */ | |
5630 | pcp->high = high; | |
5631 | smp_wmb(); | |
5632 | ||
5633 | pcp->batch = batch; | |
5634 | } | |
5635 | ||
3664033c | 5636 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
5637 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5638 | { | |
8d7a8fa9 | 5639 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
5640 | } |
5641 | ||
88c90dbc | 5642 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
5643 | { |
5644 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 5645 | int migratetype; |
2caaad41 | 5646 | |
1c6fe946 MD |
5647 | memset(p, 0, sizeof(*p)); |
5648 | ||
3dfa5721 | 5649 | pcp = &p->pcp; |
2caaad41 | 5650 | pcp->count = 0; |
5f8dcc21 MG |
5651 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5652 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
5653 | } |
5654 | ||
88c90dbc CS |
5655 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5656 | { | |
5657 | pageset_init(p); | |
5658 | pageset_set_batch(p, batch); | |
5659 | } | |
5660 | ||
8ad4b1fb | 5661 | /* |
3664033c | 5662 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
5663 | * to the value high for the pageset p. |
5664 | */ | |
3664033c | 5665 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
5666 | unsigned long high) |
5667 | { | |
8d7a8fa9 CS |
5668 | unsigned long batch = max(1UL, high / 4); |
5669 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
5670 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 5671 | |
8d7a8fa9 | 5672 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
5673 | } |
5674 | ||
7cd2b0a3 DR |
5675 | static void pageset_set_high_and_batch(struct zone *zone, |
5676 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 5677 | { |
56cef2b8 | 5678 | if (percpu_pagelist_fraction) |
3664033c | 5679 | pageset_set_high(pcp, |
56cef2b8 CS |
5680 | (zone->managed_pages / |
5681 | percpu_pagelist_fraction)); | |
5682 | else | |
5683 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
5684 | } | |
5685 | ||
169f6c19 CS |
5686 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5687 | { | |
5688 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
5689 | ||
5690 | pageset_init(pcp); | |
5691 | pageset_set_high_and_batch(zone, pcp); | |
5692 | } | |
5693 | ||
72675e13 | 5694 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
5695 | { |
5696 | int cpu; | |
319774e2 | 5697 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5698 | for_each_possible_cpu(cpu) |
5699 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5700 | } |
5701 | ||
2caaad41 | 5702 | /* |
99dcc3e5 CL |
5703 | * Allocate per cpu pagesets and initialize them. |
5704 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5705 | */ |
99dcc3e5 | 5706 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5707 | { |
b4911ea2 | 5708 | struct pglist_data *pgdat; |
99dcc3e5 | 5709 | struct zone *zone; |
e7c8d5c9 | 5710 | |
319774e2 WF |
5711 | for_each_populated_zone(zone) |
5712 | setup_zone_pageset(zone); | |
b4911ea2 MG |
5713 | |
5714 | for_each_online_pgdat(pgdat) | |
5715 | pgdat->per_cpu_nodestats = | |
5716 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
5717 | } |
5718 | ||
c09b4240 | 5719 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5720 | { |
99dcc3e5 CL |
5721 | /* |
5722 | * per cpu subsystem is not up at this point. The following code | |
5723 | * relies on the ability of the linker to provide the | |
5724 | * offset of a (static) per cpu variable into the per cpu area. | |
5725 | */ | |
5726 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5727 | |
b38a8725 | 5728 | if (populated_zone(zone)) |
99dcc3e5 CL |
5729 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5730 | zone->name, zone->present_pages, | |
5731 | zone_batchsize(zone)); | |
ed8ece2e DH |
5732 | } |
5733 | ||
dc0bbf3b | 5734 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5735 | unsigned long zone_start_pfn, |
b171e409 | 5736 | unsigned long size) |
ed8ece2e DH |
5737 | { |
5738 | struct pglist_data *pgdat = zone->zone_pgdat; | |
9dcb8b68 | 5739 | |
ed8ece2e DH |
5740 | pgdat->nr_zones = zone_idx(zone) + 1; |
5741 | ||
ed8ece2e DH |
5742 | zone->zone_start_pfn = zone_start_pfn; |
5743 | ||
708614e6 MG |
5744 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5745 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5746 | pgdat->node_id, | |
5747 | (unsigned long)zone_idx(zone), | |
5748 | zone_start_pfn, (zone_start_pfn + size)); | |
5749 | ||
1e548deb | 5750 | zone_init_free_lists(zone); |
9dcb8b68 | 5751 | zone->initialized = 1; |
ed8ece2e DH |
5752 | } |
5753 | ||
0ee332c1 | 5754 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5755 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5756 | |
c713216d MG |
5757 | /* |
5758 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5759 | */ |
8a942fde MG |
5760 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5761 | struct mminit_pfnnid_cache *state) | |
c713216d | 5762 | { |
c13291a5 | 5763 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5764 | int nid; |
7c243c71 | 5765 | |
8a942fde MG |
5766 | if (state->last_start <= pfn && pfn < state->last_end) |
5767 | return state->last_nid; | |
c713216d | 5768 | |
e76b63f8 YL |
5769 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5770 | if (nid != -1) { | |
8a942fde MG |
5771 | state->last_start = start_pfn; |
5772 | state->last_end = end_pfn; | |
5773 | state->last_nid = nid; | |
e76b63f8 YL |
5774 | } |
5775 | ||
5776 | return nid; | |
c713216d MG |
5777 | } |
5778 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5779 | ||
c713216d | 5780 | /** |
6782832e | 5781 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5782 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5783 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5784 | * |
7d018176 ZZ |
5785 | * If an architecture guarantees that all ranges registered contain no holes |
5786 | * and may be freed, this this function may be used instead of calling | |
5787 | * memblock_free_early_nid() manually. | |
c713216d | 5788 | */ |
c13291a5 | 5789 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5790 | { |
c13291a5 TH |
5791 | unsigned long start_pfn, end_pfn; |
5792 | int i, this_nid; | |
edbe7d23 | 5793 | |
c13291a5 TH |
5794 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5795 | start_pfn = min(start_pfn, max_low_pfn); | |
5796 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5797 | |
c13291a5 | 5798 | if (start_pfn < end_pfn) |
6782832e SS |
5799 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5800 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5801 | this_nid); | |
edbe7d23 | 5802 | } |
edbe7d23 | 5803 | } |
edbe7d23 | 5804 | |
c713216d MG |
5805 | /** |
5806 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5807 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5808 | * |
7d018176 ZZ |
5809 | * If an architecture guarantees that all ranges registered contain no holes and may |
5810 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5811 | */ |
5812 | void __init sparse_memory_present_with_active_regions(int nid) | |
5813 | { | |
c13291a5 TH |
5814 | unsigned long start_pfn, end_pfn; |
5815 | int i, this_nid; | |
c713216d | 5816 | |
c13291a5 TH |
5817 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5818 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5819 | } |
5820 | ||
5821 | /** | |
5822 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5823 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5824 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5825 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5826 | * |
5827 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5828 | * provided by memblock_set_node(). If called for a node |
c713216d | 5829 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5830 | * PFNs will be 0. |
c713216d | 5831 | */ |
a3142c8e | 5832 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5833 | unsigned long *start_pfn, unsigned long *end_pfn) |
5834 | { | |
c13291a5 | 5835 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5836 | int i; |
c13291a5 | 5837 | |
c713216d MG |
5838 | *start_pfn = -1UL; |
5839 | *end_pfn = 0; | |
5840 | ||
c13291a5 TH |
5841 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5842 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5843 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5844 | } |
5845 | ||
633c0666 | 5846 | if (*start_pfn == -1UL) |
c713216d | 5847 | *start_pfn = 0; |
c713216d MG |
5848 | } |
5849 | ||
2a1e274a MG |
5850 | /* |
5851 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5852 | * assumption is made that zones within a node are ordered in monotonic | |
5853 | * increasing memory addresses so that the "highest" populated zone is used | |
5854 | */ | |
b69a7288 | 5855 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5856 | { |
5857 | int zone_index; | |
5858 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5859 | if (zone_index == ZONE_MOVABLE) | |
5860 | continue; | |
5861 | ||
5862 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5863 | arch_zone_lowest_possible_pfn[zone_index]) | |
5864 | break; | |
5865 | } | |
5866 | ||
5867 | VM_BUG_ON(zone_index == -1); | |
5868 | movable_zone = zone_index; | |
5869 | } | |
5870 | ||
5871 | /* | |
5872 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5873 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5874 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5875 | * in each node depending on the size of each node and how evenly kernelcore | |
5876 | * is distributed. This helper function adjusts the zone ranges | |
5877 | * provided by the architecture for a given node by using the end of the | |
5878 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5879 | * zones within a node are in order of monotonic increases memory addresses | |
5880 | */ | |
b69a7288 | 5881 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5882 | unsigned long zone_type, |
5883 | unsigned long node_start_pfn, | |
5884 | unsigned long node_end_pfn, | |
5885 | unsigned long *zone_start_pfn, | |
5886 | unsigned long *zone_end_pfn) | |
5887 | { | |
5888 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5889 | if (zone_movable_pfn[nid]) { | |
5890 | /* Size ZONE_MOVABLE */ | |
5891 | if (zone_type == ZONE_MOVABLE) { | |
5892 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5893 | *zone_end_pfn = min(node_end_pfn, | |
5894 | arch_zone_highest_possible_pfn[movable_zone]); | |
5895 | ||
e506b996 XQ |
5896 | /* Adjust for ZONE_MOVABLE starting within this range */ |
5897 | } else if (!mirrored_kernelcore && | |
5898 | *zone_start_pfn < zone_movable_pfn[nid] && | |
5899 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
5900 | *zone_end_pfn = zone_movable_pfn[nid]; | |
5901 | ||
2a1e274a MG |
5902 | /* Check if this whole range is within ZONE_MOVABLE */ |
5903 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5904 | *zone_start_pfn = *zone_end_pfn; | |
5905 | } | |
5906 | } | |
5907 | ||
c713216d MG |
5908 | /* |
5909 | * Return the number of pages a zone spans in a node, including holes | |
5910 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5911 | */ | |
6ea6e688 | 5912 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5913 | unsigned long zone_type, |
7960aedd ZY |
5914 | unsigned long node_start_pfn, |
5915 | unsigned long node_end_pfn, | |
d91749c1 TI |
5916 | unsigned long *zone_start_pfn, |
5917 | unsigned long *zone_end_pfn, | |
c713216d MG |
5918 | unsigned long *ignored) |
5919 | { | |
b5685e92 | 5920 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5921 | if (!node_start_pfn && !node_end_pfn) |
5922 | return 0; | |
5923 | ||
7960aedd | 5924 | /* Get the start and end of the zone */ |
d91749c1 TI |
5925 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5926 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5927 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5928 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5929 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5930 | |
5931 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5932 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5933 | return 0; |
5934 | ||
5935 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5936 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5937 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5938 | |
5939 | /* Return the spanned pages */ | |
d91749c1 | 5940 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5941 | } |
5942 | ||
5943 | /* | |
5944 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5945 | * then all holes in the requested range will be accounted for. |
c713216d | 5946 | */ |
32996250 | 5947 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5948 | unsigned long range_start_pfn, |
5949 | unsigned long range_end_pfn) | |
5950 | { | |
96e907d1 TH |
5951 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5952 | unsigned long start_pfn, end_pfn; | |
5953 | int i; | |
c713216d | 5954 | |
96e907d1 TH |
5955 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5956 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5957 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5958 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5959 | } |
96e907d1 | 5960 | return nr_absent; |
c713216d MG |
5961 | } |
5962 | ||
5963 | /** | |
5964 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5965 | * @start_pfn: The start PFN to start searching for holes | |
5966 | * @end_pfn: The end PFN to stop searching for holes | |
5967 | * | |
88ca3b94 | 5968 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5969 | */ |
5970 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5971 | unsigned long end_pfn) | |
5972 | { | |
5973 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5974 | } | |
5975 | ||
5976 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5977 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5978 | unsigned long zone_type, |
7960aedd ZY |
5979 | unsigned long node_start_pfn, |
5980 | unsigned long node_end_pfn, | |
c713216d MG |
5981 | unsigned long *ignored) |
5982 | { | |
96e907d1 TH |
5983 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5984 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5985 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5986 | unsigned long nr_absent; |
9c7cd687 | 5987 | |
b5685e92 | 5988 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5989 | if (!node_start_pfn && !node_end_pfn) |
5990 | return 0; | |
5991 | ||
96e907d1 TH |
5992 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5993 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5994 | |
2a1e274a MG |
5995 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5996 | node_start_pfn, node_end_pfn, | |
5997 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5998 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5999 | ||
6000 | /* | |
6001 | * ZONE_MOVABLE handling. | |
6002 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6003 | * and vice versa. | |
6004 | */ | |
e506b996 XQ |
6005 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6006 | unsigned long start_pfn, end_pfn; | |
6007 | struct memblock_region *r; | |
6008 | ||
6009 | for_each_memblock(memory, r) { | |
6010 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
6011 | zone_start_pfn, zone_end_pfn); | |
6012 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6013 | zone_start_pfn, zone_end_pfn); | |
6014 | ||
6015 | if (zone_type == ZONE_MOVABLE && | |
6016 | memblock_is_mirror(r)) | |
6017 | nr_absent += end_pfn - start_pfn; | |
6018 | ||
6019 | if (zone_type == ZONE_NORMAL && | |
6020 | !memblock_is_mirror(r)) | |
6021 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6022 | } |
6023 | } | |
6024 | ||
6025 | return nr_absent; | |
c713216d | 6026 | } |
0e0b864e | 6027 | |
0ee332c1 | 6028 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 6029 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 6030 | unsigned long zone_type, |
7960aedd ZY |
6031 | unsigned long node_start_pfn, |
6032 | unsigned long node_end_pfn, | |
d91749c1 TI |
6033 | unsigned long *zone_start_pfn, |
6034 | unsigned long *zone_end_pfn, | |
c713216d MG |
6035 | unsigned long *zones_size) |
6036 | { | |
d91749c1 TI |
6037 | unsigned int zone; |
6038 | ||
6039 | *zone_start_pfn = node_start_pfn; | |
6040 | for (zone = 0; zone < zone_type; zone++) | |
6041 | *zone_start_pfn += zones_size[zone]; | |
6042 | ||
6043 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
6044 | ||
c713216d MG |
6045 | return zones_size[zone_type]; |
6046 | } | |
6047 | ||
6ea6e688 | 6048 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 6049 | unsigned long zone_type, |
7960aedd ZY |
6050 | unsigned long node_start_pfn, |
6051 | unsigned long node_end_pfn, | |
c713216d MG |
6052 | unsigned long *zholes_size) |
6053 | { | |
6054 | if (!zholes_size) | |
6055 | return 0; | |
6056 | ||
6057 | return zholes_size[zone_type]; | |
6058 | } | |
20e6926d | 6059 | |
0ee332c1 | 6060 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6061 | |
a3142c8e | 6062 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
6063 | unsigned long node_start_pfn, |
6064 | unsigned long node_end_pfn, | |
6065 | unsigned long *zones_size, | |
6066 | unsigned long *zholes_size) | |
c713216d | 6067 | { |
febd5949 | 6068 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6069 | enum zone_type i; |
6070 | ||
febd5949 GZ |
6071 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6072 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6073 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 6074 | unsigned long size, real_size; |
c713216d | 6075 | |
febd5949 GZ |
6076 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
6077 | node_start_pfn, | |
6078 | node_end_pfn, | |
d91749c1 TI |
6079 | &zone_start_pfn, |
6080 | &zone_end_pfn, | |
febd5949 GZ |
6081 | zones_size); |
6082 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
6083 | node_start_pfn, node_end_pfn, |
6084 | zholes_size); | |
d91749c1 TI |
6085 | if (size) |
6086 | zone->zone_start_pfn = zone_start_pfn; | |
6087 | else | |
6088 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
6089 | zone->spanned_pages = size; |
6090 | zone->present_pages = real_size; | |
6091 | ||
6092 | totalpages += size; | |
6093 | realtotalpages += real_size; | |
6094 | } | |
6095 | ||
6096 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
6097 | pgdat->node_present_pages = realtotalpages; |
6098 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
6099 | realtotalpages); | |
6100 | } | |
6101 | ||
835c134e MG |
6102 | #ifndef CONFIG_SPARSEMEM |
6103 | /* | |
6104 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
6105 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
6106 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
6107 | * round what is now in bits to nearest long in bits, then return it in |
6108 | * bytes. | |
6109 | */ | |
7c45512d | 6110 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
6111 | { |
6112 | unsigned long usemapsize; | |
6113 | ||
7c45512d | 6114 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
6115 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
6116 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
6117 | usemapsize *= NR_PAGEBLOCK_BITS; |
6118 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
6119 | ||
6120 | return usemapsize / 8; | |
6121 | } | |
6122 | ||
6123 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
6124 | struct zone *zone, |
6125 | unsigned long zone_start_pfn, | |
6126 | unsigned long zonesize) | |
835c134e | 6127 | { |
7c45512d | 6128 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 6129 | zone->pageblock_flags = NULL; |
58a01a45 | 6130 | if (usemapsize) |
6782832e SS |
6131 | zone->pageblock_flags = |
6132 | memblock_virt_alloc_node_nopanic(usemapsize, | |
6133 | pgdat->node_id); | |
835c134e MG |
6134 | } |
6135 | #else | |
7c45512d LT |
6136 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
6137 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
6138 | #endif /* CONFIG_SPARSEMEM */ |
6139 | ||
d9c23400 | 6140 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 6141 | |
d9c23400 | 6142 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 6143 | void __paginginit set_pageblock_order(void) |
d9c23400 | 6144 | { |
955c1cd7 AM |
6145 | unsigned int order; |
6146 | ||
d9c23400 MG |
6147 | /* Check that pageblock_nr_pages has not already been setup */ |
6148 | if (pageblock_order) | |
6149 | return; | |
6150 | ||
955c1cd7 AM |
6151 | if (HPAGE_SHIFT > PAGE_SHIFT) |
6152 | order = HUGETLB_PAGE_ORDER; | |
6153 | else | |
6154 | order = MAX_ORDER - 1; | |
6155 | ||
d9c23400 MG |
6156 | /* |
6157 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
6158 | * This value may be variable depending on boot parameters on IA64 and |
6159 | * powerpc. | |
d9c23400 MG |
6160 | */ |
6161 | pageblock_order = order; | |
6162 | } | |
6163 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6164 | ||
ba72cb8c MG |
6165 | /* |
6166 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
6167 | * is unused as pageblock_order is set at compile-time. See |
6168 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
6169 | * the kernel config | |
ba72cb8c | 6170 | */ |
15ca220e | 6171 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 6172 | { |
ba72cb8c | 6173 | } |
d9c23400 MG |
6174 | |
6175 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6176 | ||
01cefaef JL |
6177 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
6178 | unsigned long present_pages) | |
6179 | { | |
6180 | unsigned long pages = spanned_pages; | |
6181 | ||
6182 | /* | |
6183 | * Provide a more accurate estimation if there are holes within | |
6184 | * the zone and SPARSEMEM is in use. If there are holes within the | |
6185 | * zone, each populated memory region may cost us one or two extra | |
6186 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 6187 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
6188 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
6189 | */ | |
6190 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
6191 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
6192 | pages = present_pages; | |
6193 | ||
6194 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
6195 | } | |
6196 | ||
ace1db39 OS |
6197 | #ifdef CONFIG_NUMA_BALANCING |
6198 | static void pgdat_init_numabalancing(struct pglist_data *pgdat) | |
6199 | { | |
6200 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
6201 | pgdat->numabalancing_migrate_nr_pages = 0; | |
6202 | pgdat->numabalancing_migrate_next_window = jiffies; | |
6203 | } | |
6204 | #else | |
6205 | static void pgdat_init_numabalancing(struct pglist_data *pgdat) {} | |
6206 | #endif | |
6207 | ||
6208 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6209 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
6210 | { | |
6211 | spin_lock_init(&pgdat->split_queue_lock); | |
6212 | INIT_LIST_HEAD(&pgdat->split_queue); | |
6213 | pgdat->split_queue_len = 0; | |
6214 | } | |
6215 | #else | |
6216 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
6217 | #endif | |
6218 | ||
6219 | #ifdef CONFIG_COMPACTION | |
6220 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
6221 | { | |
6222 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6223 | } | |
6224 | #else | |
6225 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
6226 | #endif | |
6227 | ||
1da177e4 LT |
6228 | /* |
6229 | * Set up the zone data structures: | |
6230 | * - mark all pages reserved | |
6231 | * - mark all memory queues empty | |
6232 | * - clear the memory bitmaps | |
6527af5d MK |
6233 | * |
6234 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 6235 | */ |
7f3eb55b | 6236 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 6237 | { |
2f1b6248 | 6238 | enum zone_type j; |
ed8ece2e | 6239 | int nid = pgdat->node_id; |
1da177e4 | 6240 | |
208d54e5 | 6241 | pgdat_resize_init(pgdat); |
ace1db39 OS |
6242 | |
6243 | pgdat_init_numabalancing(pgdat); | |
6244 | pgdat_init_split_queue(pgdat); | |
6245 | pgdat_init_kcompactd(pgdat); | |
6246 | ||
1da177e4 | 6247 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6248 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 6249 | |
eefa864b | 6250 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6251 | spin_lock_init(&pgdat->lru_lock); |
a9dd0a83 | 6252 | lruvec_init(node_lruvec(pgdat)); |
5f63b720 | 6253 | |
385386cf JW |
6254 | pgdat->per_cpu_nodestats = &boot_nodestats; |
6255 | ||
1da177e4 LT |
6256 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6257 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 6258 | unsigned long size, freesize, memmap_pages; |
d91749c1 | 6259 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6260 | |
febd5949 | 6261 | size = zone->spanned_pages; |
e6943859 | 6262 | freesize = zone->present_pages; |
1da177e4 | 6263 | |
0e0b864e | 6264 | /* |
9feedc9d | 6265 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6266 | * is used by this zone for memmap. This affects the watermark |
6267 | * and per-cpu initialisations | |
6268 | */ | |
e6943859 | 6269 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
6270 | if (!is_highmem_idx(j)) { |
6271 | if (freesize >= memmap_pages) { | |
6272 | freesize -= memmap_pages; | |
6273 | if (memmap_pages) | |
6274 | printk(KERN_DEBUG | |
6275 | " %s zone: %lu pages used for memmap\n", | |
6276 | zone_names[j], memmap_pages); | |
6277 | } else | |
1170532b | 6278 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6279 | zone_names[j], memmap_pages, freesize); |
6280 | } | |
0e0b864e | 6281 | |
6267276f | 6282 | /* Account for reserved pages */ |
9feedc9d JL |
6283 | if (j == 0 && freesize > dma_reserve) { |
6284 | freesize -= dma_reserve; | |
d903ef9f | 6285 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6286 | zone_names[0], dma_reserve); |
0e0b864e MG |
6287 | } |
6288 | ||
98d2b0eb | 6289 | if (!is_highmem_idx(j)) |
9feedc9d | 6290 | nr_kernel_pages += freesize; |
01cefaef JL |
6291 | /* Charge for highmem memmap if there are enough kernel pages */ |
6292 | else if (nr_kernel_pages > memmap_pages * 2) | |
6293 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6294 | nr_all_pages += freesize; |
1da177e4 | 6295 | |
9feedc9d JL |
6296 | /* |
6297 | * Set an approximate value for lowmem here, it will be adjusted | |
6298 | * when the bootmem allocator frees pages into the buddy system. | |
6299 | * And all highmem pages will be managed by the buddy system. | |
6300 | */ | |
e6943859 | 6301 | zone->managed_pages = freesize; |
9614634f | 6302 | #ifdef CONFIG_NUMA |
d5f541ed | 6303 | zone->node = nid; |
9614634f | 6304 | #endif |
1da177e4 | 6305 | zone->name = zone_names[j]; |
a52633d8 | 6306 | zone->zone_pgdat = pgdat; |
1da177e4 | 6307 | spin_lock_init(&zone->lock); |
bdc8cb98 | 6308 | zone_seqlock_init(zone); |
ed8ece2e | 6309 | zone_pcp_init(zone); |
81c0a2bb | 6310 | |
d883c6cf | 6311 | if (!size) |
1da177e4 LT |
6312 | continue; |
6313 | ||
955c1cd7 | 6314 | set_pageblock_order(); |
d883c6cf JK |
6315 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
6316 | init_currently_empty_zone(zone, zone_start_pfn, size); | |
76cdd58e | 6317 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6318 | } |
6319 | } | |
6320 | ||
0cd842f9 | 6321 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 6322 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6323 | { |
b0aeba74 | 6324 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6325 | unsigned long __maybe_unused offset = 0; |
6326 | ||
1da177e4 LT |
6327 | /* Skip empty nodes */ |
6328 | if (!pgdat->node_spanned_pages) | |
6329 | return; | |
6330 | ||
b0aeba74 TL |
6331 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6332 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6333 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6334 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6335 | unsigned long size, end; |
d41dee36 AW |
6336 | struct page *map; |
6337 | ||
e984bb43 BP |
6338 | /* |
6339 | * The zone's endpoints aren't required to be MAX_ORDER | |
6340 | * aligned but the node_mem_map endpoints must be in order | |
6341 | * for the buddy allocator to function correctly. | |
6342 | */ | |
108bcc96 | 6343 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6344 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6345 | size = (end - start) * sizeof(struct page); | |
79375ea3 | 6346 | map = memblock_virt_alloc_node_nopanic(size, pgdat->node_id); |
a1c34a3b | 6347 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6348 | } |
0cd842f9 OS |
6349 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
6350 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
6351 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 6352 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6353 | /* |
6354 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6355 | */ | |
c713216d | 6356 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6357 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 6358 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 6359 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6360 | mem_map -= offset; |
0ee332c1 | 6361 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6362 | } |
1da177e4 LT |
6363 | #endif |
6364 | } | |
0cd842f9 OS |
6365 | #else |
6366 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
6367 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 6368 | |
9109fb7b JW |
6369 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
6370 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 6371 | { |
9109fb7b | 6372 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6373 | unsigned long start_pfn = 0; |
6374 | unsigned long end_pfn = 0; | |
9109fb7b | 6375 | |
88fdf75d | 6376 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 6377 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 6378 | |
1da177e4 LT |
6379 | pgdat->node_id = nid; |
6380 | pgdat->node_start_pfn = node_start_pfn; | |
75ef7184 | 6381 | pgdat->per_cpu_nodestats = NULL; |
7960aedd ZY |
6382 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6383 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 6384 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6385 | (u64)start_pfn << PAGE_SHIFT, |
6386 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
6387 | #else |
6388 | start_pfn = node_start_pfn; | |
7960aedd ZY |
6389 | #endif |
6390 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
6391 | zones_size, zholes_size); | |
1da177e4 LT |
6392 | |
6393 | alloc_node_mem_map(pgdat); | |
6394 | ||
c9e97a19 PT |
6395 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
6396 | /* | |
6397 | * We start only with one section of pages, more pages are added as | |
6398 | * needed until the rest of deferred pages are initialized. | |
6399 | */ | |
6400 | pgdat->static_init_pgcnt = min_t(unsigned long, PAGES_PER_SECTION, | |
6401 | pgdat->node_spanned_pages); | |
6402 | pgdat->first_deferred_pfn = ULONG_MAX; | |
6403 | #endif | |
7f3eb55b | 6404 | free_area_init_core(pgdat); |
1da177e4 LT |
6405 | } |
6406 | ||
d1b47a7c | 6407 | #if defined(CONFIG_HAVE_MEMBLOCK) && !defined(CONFIG_FLAT_NODE_MEM_MAP) |
a4a3ede2 PT |
6408 | /* |
6409 | * Only struct pages that are backed by physical memory are zeroed and | |
6410 | * initialized by going through __init_single_page(). But, there are some | |
6411 | * struct pages which are reserved in memblock allocator and their fields | |
6412 | * may be accessed (for example page_to_pfn() on some configuration accesses | |
6413 | * flags). We must explicitly zero those struct pages. | |
6414 | */ | |
6415 | void __paginginit zero_resv_unavail(void) | |
6416 | { | |
6417 | phys_addr_t start, end; | |
6418 | unsigned long pfn; | |
6419 | u64 i, pgcnt; | |
6420 | ||
6421 | /* | |
6422 | * Loop through ranges that are reserved, but do not have reported | |
6423 | * physical memory backing. | |
6424 | */ | |
6425 | pgcnt = 0; | |
6426 | for_each_resv_unavail_range(i, &start, &end) { | |
6427 | for (pfn = PFN_DOWN(start); pfn < PFN_UP(end); pfn++) { | |
720e14eb PT |
6428 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { |
6429 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6430 | + pageblock_nr_pages - 1; | |
e8c24773 | 6431 | continue; |
720e14eb | 6432 | } |
a4a3ede2 PT |
6433 | mm_zero_struct_page(pfn_to_page(pfn)); |
6434 | pgcnt++; | |
6435 | } | |
6436 | } | |
6437 | ||
6438 | /* | |
6439 | * Struct pages that do not have backing memory. This could be because | |
6440 | * firmware is using some of this memory, or for some other reasons. | |
6441 | * Once memblock is changed so such behaviour is not allowed: i.e. | |
6442 | * list of "reserved" memory must be a subset of list of "memory", then | |
6443 | * this code can be removed. | |
6444 | */ | |
6445 | if (pgcnt) | |
6446 | pr_info("Reserved but unavailable: %lld pages", pgcnt); | |
6447 | } | |
d1b47a7c | 6448 | #endif /* CONFIG_HAVE_MEMBLOCK && !CONFIG_FLAT_NODE_MEM_MAP */ |
a4a3ede2 | 6449 | |
0ee332c1 | 6450 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
6451 | |
6452 | #if MAX_NUMNODES > 1 | |
6453 | /* | |
6454 | * Figure out the number of possible node ids. | |
6455 | */ | |
f9872caf | 6456 | void __init setup_nr_node_ids(void) |
418508c1 | 6457 | { |
904a9553 | 6458 | unsigned int highest; |
418508c1 | 6459 | |
904a9553 | 6460 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
6461 | nr_node_ids = highest + 1; |
6462 | } | |
418508c1 MS |
6463 | #endif |
6464 | ||
1e01979c TH |
6465 | /** |
6466 | * node_map_pfn_alignment - determine the maximum internode alignment | |
6467 | * | |
6468 | * This function should be called after node map is populated and sorted. | |
6469 | * It calculates the maximum power of two alignment which can distinguish | |
6470 | * all the nodes. | |
6471 | * | |
6472 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
6473 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
6474 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
6475 | * shifted, 1GiB is enough and this function will indicate so. | |
6476 | * | |
6477 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
6478 | * model has fine enough granularity to avoid incorrect mapping for the | |
6479 | * populated node map. | |
6480 | * | |
6481 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
6482 | * requirement (single node). | |
6483 | */ | |
6484 | unsigned long __init node_map_pfn_alignment(void) | |
6485 | { | |
6486 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 6487 | unsigned long start, end, mask; |
1e01979c | 6488 | int last_nid = -1; |
c13291a5 | 6489 | int i, nid; |
1e01979c | 6490 | |
c13291a5 | 6491 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
6492 | if (!start || last_nid < 0 || last_nid == nid) { |
6493 | last_nid = nid; | |
6494 | last_end = end; | |
6495 | continue; | |
6496 | } | |
6497 | ||
6498 | /* | |
6499 | * Start with a mask granular enough to pin-point to the | |
6500 | * start pfn and tick off bits one-by-one until it becomes | |
6501 | * too coarse to separate the current node from the last. | |
6502 | */ | |
6503 | mask = ~((1 << __ffs(start)) - 1); | |
6504 | while (mask && last_end <= (start & (mask << 1))) | |
6505 | mask <<= 1; | |
6506 | ||
6507 | /* accumulate all internode masks */ | |
6508 | accl_mask |= mask; | |
6509 | } | |
6510 | ||
6511 | /* convert mask to number of pages */ | |
6512 | return ~accl_mask + 1; | |
6513 | } | |
6514 | ||
a6af2bc3 | 6515 | /* Find the lowest pfn for a node */ |
b69a7288 | 6516 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 6517 | { |
a6af2bc3 | 6518 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
6519 | unsigned long start_pfn; |
6520 | int i; | |
1abbfb41 | 6521 | |
c13291a5 TH |
6522 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
6523 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 6524 | |
a6af2bc3 | 6525 | if (min_pfn == ULONG_MAX) { |
1170532b | 6526 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
6527 | return 0; |
6528 | } | |
6529 | ||
6530 | return min_pfn; | |
c713216d MG |
6531 | } |
6532 | ||
6533 | /** | |
6534 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
6535 | * | |
6536 | * It returns the minimum PFN based on information provided via | |
7d018176 | 6537 | * memblock_set_node(). |
c713216d MG |
6538 | */ |
6539 | unsigned long __init find_min_pfn_with_active_regions(void) | |
6540 | { | |
6541 | return find_min_pfn_for_node(MAX_NUMNODES); | |
6542 | } | |
6543 | ||
37b07e41 LS |
6544 | /* |
6545 | * early_calculate_totalpages() | |
6546 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 6547 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 6548 | */ |
484f51f8 | 6549 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 6550 | { |
7e63efef | 6551 | unsigned long totalpages = 0; |
c13291a5 TH |
6552 | unsigned long start_pfn, end_pfn; |
6553 | int i, nid; | |
6554 | ||
6555 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
6556 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 6557 | |
37b07e41 LS |
6558 | totalpages += pages; |
6559 | if (pages) | |
4b0ef1fe | 6560 | node_set_state(nid, N_MEMORY); |
37b07e41 | 6561 | } |
b8af2941 | 6562 | return totalpages; |
7e63efef MG |
6563 | } |
6564 | ||
2a1e274a MG |
6565 | /* |
6566 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
6567 | * is spread evenly between nodes as long as the nodes have enough | |
6568 | * memory. When they don't, some nodes will have more kernelcore than | |
6569 | * others | |
6570 | */ | |
b224ef85 | 6571 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
6572 | { |
6573 | int i, nid; | |
6574 | unsigned long usable_startpfn; | |
6575 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 6576 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 6577 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 6578 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 6579 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 6580 | struct memblock_region *r; |
b2f3eebe TC |
6581 | |
6582 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
6583 | find_usable_zone_for_movable(); | |
6584 | ||
6585 | /* | |
6586 | * If movable_node is specified, ignore kernelcore and movablecore | |
6587 | * options. | |
6588 | */ | |
6589 | if (movable_node_is_enabled()) { | |
136199f0 EM |
6590 | for_each_memblock(memory, r) { |
6591 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
6592 | continue; |
6593 | ||
136199f0 | 6594 | nid = r->nid; |
b2f3eebe | 6595 | |
136199f0 | 6596 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
6597 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6598 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6599 | usable_startpfn; | |
6600 | } | |
6601 | ||
6602 | goto out2; | |
6603 | } | |
2a1e274a | 6604 | |
342332e6 TI |
6605 | /* |
6606 | * If kernelcore=mirror is specified, ignore movablecore option | |
6607 | */ | |
6608 | if (mirrored_kernelcore) { | |
6609 | bool mem_below_4gb_not_mirrored = false; | |
6610 | ||
6611 | for_each_memblock(memory, r) { | |
6612 | if (memblock_is_mirror(r)) | |
6613 | continue; | |
6614 | ||
6615 | nid = r->nid; | |
6616 | ||
6617 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
6618 | ||
6619 | if (usable_startpfn < 0x100000) { | |
6620 | mem_below_4gb_not_mirrored = true; | |
6621 | continue; | |
6622 | } | |
6623 | ||
6624 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
6625 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6626 | usable_startpfn; | |
6627 | } | |
6628 | ||
6629 | if (mem_below_4gb_not_mirrored) | |
6630 | pr_warn("This configuration results in unmirrored kernel memory."); | |
6631 | ||
6632 | goto out2; | |
6633 | } | |
6634 | ||
7e63efef | 6635 | /* |
a5c6d650 DR |
6636 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
6637 | * amount of necessary memory. | |
6638 | */ | |
6639 | if (required_kernelcore_percent) | |
6640 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
6641 | 10000UL; | |
6642 | if (required_movablecore_percent) | |
6643 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
6644 | 10000UL; | |
6645 | ||
6646 | /* | |
6647 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
6648 | * kernelcore that corresponds so that memory usable for |
6649 | * any allocation type is evenly spread. If both kernelcore | |
6650 | * and movablecore are specified, then the value of kernelcore | |
6651 | * will be used for required_kernelcore if it's greater than | |
6652 | * what movablecore would have allowed. | |
6653 | */ | |
6654 | if (required_movablecore) { | |
7e63efef MG |
6655 | unsigned long corepages; |
6656 | ||
6657 | /* | |
6658 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
6659 | * was requested by the user | |
6660 | */ | |
6661 | required_movablecore = | |
6662 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 6663 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
6664 | corepages = totalpages - required_movablecore; |
6665 | ||
6666 | required_kernelcore = max(required_kernelcore, corepages); | |
6667 | } | |
6668 | ||
bde304bd XQ |
6669 | /* |
6670 | * If kernelcore was not specified or kernelcore size is larger | |
6671 | * than totalpages, there is no ZONE_MOVABLE. | |
6672 | */ | |
6673 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 6674 | goto out; |
2a1e274a MG |
6675 | |
6676 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
6677 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6678 | ||
6679 | restart: | |
6680 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
6681 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 6682 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
6683 | unsigned long start_pfn, end_pfn; |
6684 | ||
2a1e274a MG |
6685 | /* |
6686 | * Recalculate kernelcore_node if the division per node | |
6687 | * now exceeds what is necessary to satisfy the requested | |
6688 | * amount of memory for the kernel | |
6689 | */ | |
6690 | if (required_kernelcore < kernelcore_node) | |
6691 | kernelcore_node = required_kernelcore / usable_nodes; | |
6692 | ||
6693 | /* | |
6694 | * As the map is walked, we track how much memory is usable | |
6695 | * by the kernel using kernelcore_remaining. When it is | |
6696 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
6697 | */ | |
6698 | kernelcore_remaining = kernelcore_node; | |
6699 | ||
6700 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 6701 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
6702 | unsigned long size_pages; |
6703 | ||
c13291a5 | 6704 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
6705 | if (start_pfn >= end_pfn) |
6706 | continue; | |
6707 | ||
6708 | /* Account for what is only usable for kernelcore */ | |
6709 | if (start_pfn < usable_startpfn) { | |
6710 | unsigned long kernel_pages; | |
6711 | kernel_pages = min(end_pfn, usable_startpfn) | |
6712 | - start_pfn; | |
6713 | ||
6714 | kernelcore_remaining -= min(kernel_pages, | |
6715 | kernelcore_remaining); | |
6716 | required_kernelcore -= min(kernel_pages, | |
6717 | required_kernelcore); | |
6718 | ||
6719 | /* Continue if range is now fully accounted */ | |
6720 | if (end_pfn <= usable_startpfn) { | |
6721 | ||
6722 | /* | |
6723 | * Push zone_movable_pfn to the end so | |
6724 | * that if we have to rebalance | |
6725 | * kernelcore across nodes, we will | |
6726 | * not double account here | |
6727 | */ | |
6728 | zone_movable_pfn[nid] = end_pfn; | |
6729 | continue; | |
6730 | } | |
6731 | start_pfn = usable_startpfn; | |
6732 | } | |
6733 | ||
6734 | /* | |
6735 | * The usable PFN range for ZONE_MOVABLE is from | |
6736 | * start_pfn->end_pfn. Calculate size_pages as the | |
6737 | * number of pages used as kernelcore | |
6738 | */ | |
6739 | size_pages = end_pfn - start_pfn; | |
6740 | if (size_pages > kernelcore_remaining) | |
6741 | size_pages = kernelcore_remaining; | |
6742 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6743 | ||
6744 | /* | |
6745 | * Some kernelcore has been met, update counts and | |
6746 | * break if the kernelcore for this node has been | |
b8af2941 | 6747 | * satisfied |
2a1e274a MG |
6748 | */ |
6749 | required_kernelcore -= min(required_kernelcore, | |
6750 | size_pages); | |
6751 | kernelcore_remaining -= size_pages; | |
6752 | if (!kernelcore_remaining) | |
6753 | break; | |
6754 | } | |
6755 | } | |
6756 | ||
6757 | /* | |
6758 | * If there is still required_kernelcore, we do another pass with one | |
6759 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6760 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6761 | * satisfied |
2a1e274a MG |
6762 | */ |
6763 | usable_nodes--; | |
6764 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6765 | goto restart; | |
6766 | ||
b2f3eebe | 6767 | out2: |
2a1e274a MG |
6768 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6769 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6770 | zone_movable_pfn[nid] = | |
6771 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6772 | |
20e6926d | 6773 | out: |
66918dcd | 6774 | /* restore the node_state */ |
4b0ef1fe | 6775 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6776 | } |
6777 | ||
4b0ef1fe LJ |
6778 | /* Any regular or high memory on that node ? */ |
6779 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6780 | { |
37b07e41 LS |
6781 | enum zone_type zone_type; |
6782 | ||
4b0ef1fe LJ |
6783 | if (N_MEMORY == N_NORMAL_MEMORY) |
6784 | return; | |
6785 | ||
6786 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 6787 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6788 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
6789 | node_set_state(nid, N_HIGH_MEMORY); |
6790 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
6791 | zone_type <= ZONE_NORMAL) | |
6792 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
6793 | break; |
6794 | } | |
37b07e41 | 6795 | } |
37b07e41 LS |
6796 | } |
6797 | ||
c713216d MG |
6798 | /** |
6799 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6800 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6801 | * |
6802 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6803 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6804 | * zone in each node and their holes is calculated. If the maximum PFN |
6805 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6806 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6807 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6808 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6809 | * at arch_max_dma_pfn. | |
6810 | */ | |
6811 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6812 | { | |
c13291a5 TH |
6813 | unsigned long start_pfn, end_pfn; |
6814 | int i, nid; | |
a6af2bc3 | 6815 | |
c713216d MG |
6816 | /* Record where the zone boundaries are */ |
6817 | memset(arch_zone_lowest_possible_pfn, 0, | |
6818 | sizeof(arch_zone_lowest_possible_pfn)); | |
6819 | memset(arch_zone_highest_possible_pfn, 0, | |
6820 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
6821 | |
6822 | start_pfn = find_min_pfn_with_active_regions(); | |
6823 | ||
6824 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6825 | if (i == ZONE_MOVABLE) |
6826 | continue; | |
90cae1fe OH |
6827 | |
6828 | end_pfn = max(max_zone_pfn[i], start_pfn); | |
6829 | arch_zone_lowest_possible_pfn[i] = start_pfn; | |
6830 | arch_zone_highest_possible_pfn[i] = end_pfn; | |
6831 | ||
6832 | start_pfn = end_pfn; | |
c713216d | 6833 | } |
2a1e274a MG |
6834 | |
6835 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6836 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6837 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6838 | |
c713216d | 6839 | /* Print out the zone ranges */ |
f88dfff5 | 6840 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6841 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6842 | if (i == ZONE_MOVABLE) | |
6843 | continue; | |
f88dfff5 | 6844 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6845 | if (arch_zone_lowest_possible_pfn[i] == |
6846 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6847 | pr_cont("empty\n"); |
72f0ba02 | 6848 | else |
8d29e18a JG |
6849 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6850 | (u64)arch_zone_lowest_possible_pfn[i] | |
6851 | << PAGE_SHIFT, | |
6852 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6853 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6854 | } |
6855 | ||
6856 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6857 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6858 | for (i = 0; i < MAX_NUMNODES; i++) { |
6859 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6860 | pr_info(" Node %d: %#018Lx\n", i, |
6861 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6862 | } |
c713216d | 6863 | |
f2d52fe5 | 6864 | /* Print out the early node map */ |
f88dfff5 | 6865 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6866 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6867 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6868 | (u64)start_pfn << PAGE_SHIFT, | |
6869 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6870 | |
6871 | /* Initialise every node */ | |
708614e6 | 6872 | mminit_verify_pageflags_layout(); |
8ef82866 | 6873 | setup_nr_node_ids(); |
e181ae0c | 6874 | zero_resv_unavail(); |
c713216d MG |
6875 | for_each_online_node(nid) { |
6876 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6877 | free_area_init_node(nid, NULL, |
c713216d | 6878 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6879 | |
6880 | /* Any memory on that node */ | |
6881 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6882 | node_set_state(nid, N_MEMORY); |
6883 | check_for_memory(pgdat, nid); | |
c713216d MG |
6884 | } |
6885 | } | |
2a1e274a | 6886 | |
a5c6d650 DR |
6887 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
6888 | unsigned long *percent) | |
2a1e274a MG |
6889 | { |
6890 | unsigned long long coremem; | |
a5c6d650 DR |
6891 | char *endptr; |
6892 | ||
2a1e274a MG |
6893 | if (!p) |
6894 | return -EINVAL; | |
6895 | ||
a5c6d650 DR |
6896 | /* Value may be a percentage of total memory, otherwise bytes */ |
6897 | coremem = simple_strtoull(p, &endptr, 0); | |
6898 | if (*endptr == '%') { | |
6899 | /* Paranoid check for percent values greater than 100 */ | |
6900 | WARN_ON(coremem > 100); | |
2a1e274a | 6901 | |
a5c6d650 DR |
6902 | *percent = coremem; |
6903 | } else { | |
6904 | coremem = memparse(p, &p); | |
6905 | /* Paranoid check that UL is enough for the coremem value */ | |
6906 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 6907 | |
a5c6d650 DR |
6908 | *core = coremem >> PAGE_SHIFT; |
6909 | *percent = 0UL; | |
6910 | } | |
2a1e274a MG |
6911 | return 0; |
6912 | } | |
ed7ed365 | 6913 | |
7e63efef MG |
6914 | /* |
6915 | * kernelcore=size sets the amount of memory for use for allocations that | |
6916 | * cannot be reclaimed or migrated. | |
6917 | */ | |
6918 | static int __init cmdline_parse_kernelcore(char *p) | |
6919 | { | |
342332e6 TI |
6920 | /* parse kernelcore=mirror */ |
6921 | if (parse_option_str(p, "mirror")) { | |
6922 | mirrored_kernelcore = true; | |
6923 | return 0; | |
6924 | } | |
6925 | ||
a5c6d650 DR |
6926 | return cmdline_parse_core(p, &required_kernelcore, |
6927 | &required_kernelcore_percent); | |
7e63efef MG |
6928 | } |
6929 | ||
6930 | /* | |
6931 | * movablecore=size sets the amount of memory for use for allocations that | |
6932 | * can be reclaimed or migrated. | |
6933 | */ | |
6934 | static int __init cmdline_parse_movablecore(char *p) | |
6935 | { | |
a5c6d650 DR |
6936 | return cmdline_parse_core(p, &required_movablecore, |
6937 | &required_movablecore_percent); | |
7e63efef MG |
6938 | } |
6939 | ||
ed7ed365 | 6940 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6941 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6942 | |
0ee332c1 | 6943 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6944 | |
c3d5f5f0 JL |
6945 | void adjust_managed_page_count(struct page *page, long count) |
6946 | { | |
6947 | spin_lock(&managed_page_count_lock); | |
6948 | page_zone(page)->managed_pages += count; | |
6949 | totalram_pages += count; | |
3dcc0571 JL |
6950 | #ifdef CONFIG_HIGHMEM |
6951 | if (PageHighMem(page)) | |
6952 | totalhigh_pages += count; | |
6953 | #endif | |
c3d5f5f0 JL |
6954 | spin_unlock(&managed_page_count_lock); |
6955 | } | |
3dcc0571 | 6956 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6957 | |
11199692 | 6958 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6959 | { |
11199692 JL |
6960 | void *pos; |
6961 | unsigned long pages = 0; | |
69afade7 | 6962 | |
11199692 JL |
6963 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6964 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6965 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
6966 | struct page *page = virt_to_page(pos); |
6967 | void *direct_map_addr; | |
6968 | ||
6969 | /* | |
6970 | * 'direct_map_addr' might be different from 'pos' | |
6971 | * because some architectures' virt_to_page() | |
6972 | * work with aliases. Getting the direct map | |
6973 | * address ensures that we get a _writeable_ | |
6974 | * alias for the memset(). | |
6975 | */ | |
6976 | direct_map_addr = page_address(page); | |
dbe67df4 | 6977 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
6978 | memset(direct_map_addr, poison, PAGE_SIZE); |
6979 | ||
6980 | free_reserved_page(page); | |
69afade7 JL |
6981 | } |
6982 | ||
6983 | if (pages && s) | |
adb1fe9a JP |
6984 | pr_info("Freeing %s memory: %ldK\n", |
6985 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
6986 | |
6987 | return pages; | |
6988 | } | |
11199692 | 6989 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6990 | |
cfa11e08 JL |
6991 | #ifdef CONFIG_HIGHMEM |
6992 | void free_highmem_page(struct page *page) | |
6993 | { | |
6994 | __free_reserved_page(page); | |
6995 | totalram_pages++; | |
7b4b2a0d | 6996 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6997 | totalhigh_pages++; |
6998 | } | |
6999 | #endif | |
7000 | ||
7ee3d4e8 JL |
7001 | |
7002 | void __init mem_init_print_info(const char *str) | |
7003 | { | |
7004 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7005 | unsigned long init_code_size, init_data_size; | |
7006 | ||
7007 | physpages = get_num_physpages(); | |
7008 | codesize = _etext - _stext; | |
7009 | datasize = _edata - _sdata; | |
7010 | rosize = __end_rodata - __start_rodata; | |
7011 | bss_size = __bss_stop - __bss_start; | |
7012 | init_data_size = __init_end - __init_begin; | |
7013 | init_code_size = _einittext - _sinittext; | |
7014 | ||
7015 | /* | |
7016 | * Detect special cases and adjust section sizes accordingly: | |
7017 | * 1) .init.* may be embedded into .data sections | |
7018 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7019 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7020 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7021 | */ | |
7022 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7023 | do { \ |
7024 | if (start <= pos && pos < end && size > adj) \ | |
7025 | size -= adj; \ | |
7026 | } while (0) | |
7ee3d4e8 JL |
7027 | |
7028 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7029 | _sinittext, init_code_size); | |
7030 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7031 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7032 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7033 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7034 | ||
7035 | #undef adj_init_size | |
7036 | ||
756a025f | 7037 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7038 | #ifdef CONFIG_HIGHMEM |
756a025f | 7039 | ", %luK highmem" |
7ee3d4e8 | 7040 | #endif |
756a025f JP |
7041 | "%s%s)\n", |
7042 | nr_free_pages() << (PAGE_SHIFT - 10), | |
7043 | physpages << (PAGE_SHIFT - 10), | |
7044 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7045 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
7046 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
7047 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 7048 | #ifdef CONFIG_HIGHMEM |
756a025f | 7049 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7050 | #endif |
756a025f | 7051 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
7052 | } |
7053 | ||
0e0b864e | 7054 | /** |
88ca3b94 RD |
7055 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7056 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7057 | * |
013110a7 | 7058 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7059 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7060 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7061 | * function may optionally be used to account for unfreeable pages in the |
7062 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7063 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7064 | */ |
7065 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7066 | { | |
7067 | dma_reserve = new_dma_reserve; | |
7068 | } | |
7069 | ||
1da177e4 LT |
7070 | void __init free_area_init(unsigned long *zones_size) |
7071 | { | |
e181ae0c | 7072 | zero_resv_unavail(); |
9109fb7b | 7073 | free_area_init_node(0, zones_size, |
1da177e4 LT |
7074 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
7075 | } | |
1da177e4 | 7076 | |
005fd4bb | 7077 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7078 | { |
1da177e4 | 7079 | |
005fd4bb SAS |
7080 | lru_add_drain_cpu(cpu); |
7081 | drain_pages(cpu); | |
9f8f2172 | 7082 | |
005fd4bb SAS |
7083 | /* |
7084 | * Spill the event counters of the dead processor | |
7085 | * into the current processors event counters. | |
7086 | * This artificially elevates the count of the current | |
7087 | * processor. | |
7088 | */ | |
7089 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7090 | |
005fd4bb SAS |
7091 | /* |
7092 | * Zero the differential counters of the dead processor | |
7093 | * so that the vm statistics are consistent. | |
7094 | * | |
7095 | * This is only okay since the processor is dead and cannot | |
7096 | * race with what we are doing. | |
7097 | */ | |
7098 | cpu_vm_stats_fold(cpu); | |
7099 | return 0; | |
1da177e4 | 7100 | } |
1da177e4 LT |
7101 | |
7102 | void __init page_alloc_init(void) | |
7103 | { | |
005fd4bb SAS |
7104 | int ret; |
7105 | ||
7106 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, | |
7107 | "mm/page_alloc:dead", NULL, | |
7108 | page_alloc_cpu_dead); | |
7109 | WARN_ON(ret < 0); | |
1da177e4 LT |
7110 | } |
7111 | ||
cb45b0e9 | 7112 | /* |
34b10060 | 7113 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
7114 | * or min_free_kbytes changes. |
7115 | */ | |
7116 | static void calculate_totalreserve_pages(void) | |
7117 | { | |
7118 | struct pglist_data *pgdat; | |
7119 | unsigned long reserve_pages = 0; | |
2f6726e5 | 7120 | enum zone_type i, j; |
cb45b0e9 HA |
7121 | |
7122 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
7123 | |
7124 | pgdat->totalreserve_pages = 0; | |
7125 | ||
cb45b0e9 HA |
7126 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7127 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 7128 | long max = 0; |
cb45b0e9 HA |
7129 | |
7130 | /* Find valid and maximum lowmem_reserve in the zone */ | |
7131 | for (j = i; j < MAX_NR_ZONES; j++) { | |
7132 | if (zone->lowmem_reserve[j] > max) | |
7133 | max = zone->lowmem_reserve[j]; | |
7134 | } | |
7135 | ||
41858966 MG |
7136 | /* we treat the high watermark as reserved pages. */ |
7137 | max += high_wmark_pages(zone); | |
cb45b0e9 | 7138 | |
b40da049 JL |
7139 | if (max > zone->managed_pages) |
7140 | max = zone->managed_pages; | |
a8d01437 | 7141 | |
281e3726 | 7142 | pgdat->totalreserve_pages += max; |
a8d01437 | 7143 | |
cb45b0e9 HA |
7144 | reserve_pages += max; |
7145 | } | |
7146 | } | |
7147 | totalreserve_pages = reserve_pages; | |
7148 | } | |
7149 | ||
1da177e4 LT |
7150 | /* |
7151 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 7152 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
7153 | * has a correct pages reserved value, so an adequate number of |
7154 | * pages are left in the zone after a successful __alloc_pages(). | |
7155 | */ | |
7156 | static void setup_per_zone_lowmem_reserve(void) | |
7157 | { | |
7158 | struct pglist_data *pgdat; | |
2f6726e5 | 7159 | enum zone_type j, idx; |
1da177e4 | 7160 | |
ec936fc5 | 7161 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
7162 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7163 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 7164 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
7165 | |
7166 | zone->lowmem_reserve[j] = 0; | |
7167 | ||
2f6726e5 CL |
7168 | idx = j; |
7169 | while (idx) { | |
1da177e4 LT |
7170 | struct zone *lower_zone; |
7171 | ||
2f6726e5 | 7172 | idx--; |
1da177e4 | 7173 | lower_zone = pgdat->node_zones + idx; |
d3cda233 JK |
7174 | |
7175 | if (sysctl_lowmem_reserve_ratio[idx] < 1) { | |
7176 | sysctl_lowmem_reserve_ratio[idx] = 0; | |
7177 | lower_zone->lowmem_reserve[j] = 0; | |
7178 | } else { | |
7179 | lower_zone->lowmem_reserve[j] = | |
7180 | managed_pages / sysctl_lowmem_reserve_ratio[idx]; | |
7181 | } | |
b40da049 | 7182 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
7183 | } |
7184 | } | |
7185 | } | |
cb45b0e9 HA |
7186 | |
7187 | /* update totalreserve_pages */ | |
7188 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7189 | } |
7190 | ||
cfd3da1e | 7191 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
7192 | { |
7193 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
7194 | unsigned long lowmem_pages = 0; | |
7195 | struct zone *zone; | |
7196 | unsigned long flags; | |
7197 | ||
7198 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
7199 | for_each_zone(zone) { | |
7200 | if (!is_highmem(zone)) | |
b40da049 | 7201 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
7202 | } |
7203 | ||
7204 | for_each_zone(zone) { | |
ac924c60 AM |
7205 | u64 tmp; |
7206 | ||
1125b4e3 | 7207 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 7208 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 7209 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
7210 | if (is_highmem(zone)) { |
7211 | /* | |
669ed175 NP |
7212 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
7213 | * need highmem pages, so cap pages_min to a small | |
7214 | * value here. | |
7215 | * | |
41858966 | 7216 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 7217 | * deltas control asynch page reclaim, and so should |
669ed175 | 7218 | * not be capped for highmem. |
1da177e4 | 7219 | */ |
90ae8d67 | 7220 | unsigned long min_pages; |
1da177e4 | 7221 | |
b40da049 | 7222 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 7223 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 7224 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 7225 | } else { |
669ed175 NP |
7226 | /* |
7227 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
7228 | * proportionate to the zone's size. |
7229 | */ | |
41858966 | 7230 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
7231 | } |
7232 | ||
795ae7a0 JW |
7233 | /* |
7234 | * Set the kswapd watermarks distance according to the | |
7235 | * scale factor in proportion to available memory, but | |
7236 | * ensure a minimum size on small systems. | |
7237 | */ | |
7238 | tmp = max_t(u64, tmp >> 2, | |
7239 | mult_frac(zone->managed_pages, | |
7240 | watermark_scale_factor, 10000)); | |
7241 | ||
7242 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
7243 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 7244 | |
1125b4e3 | 7245 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 7246 | } |
cb45b0e9 HA |
7247 | |
7248 | /* update totalreserve_pages */ | |
7249 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7250 | } |
7251 | ||
cfd3da1e MG |
7252 | /** |
7253 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
7254 | * or when memory is hot-{added|removed} | |
7255 | * | |
7256 | * Ensures that the watermark[min,low,high] values for each zone are set | |
7257 | * correctly with respect to min_free_kbytes. | |
7258 | */ | |
7259 | void setup_per_zone_wmarks(void) | |
7260 | { | |
b93e0f32 MH |
7261 | static DEFINE_SPINLOCK(lock); |
7262 | ||
7263 | spin_lock(&lock); | |
cfd3da1e | 7264 | __setup_per_zone_wmarks(); |
b93e0f32 | 7265 | spin_unlock(&lock); |
cfd3da1e MG |
7266 | } |
7267 | ||
1da177e4 LT |
7268 | /* |
7269 | * Initialise min_free_kbytes. | |
7270 | * | |
7271 | * For small machines we want it small (128k min). For large machines | |
7272 | * we want it large (64MB max). But it is not linear, because network | |
7273 | * bandwidth does not increase linearly with machine size. We use | |
7274 | * | |
b8af2941 | 7275 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
7276 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
7277 | * | |
7278 | * which yields | |
7279 | * | |
7280 | * 16MB: 512k | |
7281 | * 32MB: 724k | |
7282 | * 64MB: 1024k | |
7283 | * 128MB: 1448k | |
7284 | * 256MB: 2048k | |
7285 | * 512MB: 2896k | |
7286 | * 1024MB: 4096k | |
7287 | * 2048MB: 5792k | |
7288 | * 4096MB: 8192k | |
7289 | * 8192MB: 11584k | |
7290 | * 16384MB: 16384k | |
7291 | */ | |
1b79acc9 | 7292 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
7293 | { |
7294 | unsigned long lowmem_kbytes; | |
5f12733e | 7295 | int new_min_free_kbytes; |
1da177e4 LT |
7296 | |
7297 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
7298 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
7299 | ||
7300 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
7301 | min_free_kbytes = new_min_free_kbytes; | |
7302 | if (min_free_kbytes < 128) | |
7303 | min_free_kbytes = 128; | |
7304 | if (min_free_kbytes > 65536) | |
7305 | min_free_kbytes = 65536; | |
7306 | } else { | |
7307 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
7308 | new_min_free_kbytes, user_min_free_kbytes); | |
7309 | } | |
bc75d33f | 7310 | setup_per_zone_wmarks(); |
a6cccdc3 | 7311 | refresh_zone_stat_thresholds(); |
1da177e4 | 7312 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
7313 | |
7314 | #ifdef CONFIG_NUMA | |
7315 | setup_min_unmapped_ratio(); | |
7316 | setup_min_slab_ratio(); | |
7317 | #endif | |
7318 | ||
1da177e4 LT |
7319 | return 0; |
7320 | } | |
bc22af74 | 7321 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
7322 | |
7323 | /* | |
b8af2941 | 7324 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
7325 | * that we can call two helper functions whenever min_free_kbytes |
7326 | * changes. | |
7327 | */ | |
cccad5b9 | 7328 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7329 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7330 | { |
da8c757b HP |
7331 | int rc; |
7332 | ||
7333 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7334 | if (rc) | |
7335 | return rc; | |
7336 | ||
5f12733e MH |
7337 | if (write) { |
7338 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7339 | setup_per_zone_wmarks(); |
5f12733e | 7340 | } |
1da177e4 LT |
7341 | return 0; |
7342 | } | |
7343 | ||
795ae7a0 JW |
7344 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
7345 | void __user *buffer, size_t *length, loff_t *ppos) | |
7346 | { | |
7347 | int rc; | |
7348 | ||
7349 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7350 | if (rc) | |
7351 | return rc; | |
7352 | ||
7353 | if (write) | |
7354 | setup_per_zone_wmarks(); | |
7355 | ||
7356 | return 0; | |
7357 | } | |
7358 | ||
9614634f | 7359 | #ifdef CONFIG_NUMA |
6423aa81 | 7360 | static void setup_min_unmapped_ratio(void) |
9614634f | 7361 | { |
6423aa81 | 7362 | pg_data_t *pgdat; |
9614634f | 7363 | struct zone *zone; |
9614634f | 7364 | |
a5f5f91d | 7365 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7366 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7367 | |
9614634f | 7368 | for_each_zone(zone) |
a5f5f91d | 7369 | zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages * |
9614634f | 7370 | sysctl_min_unmapped_ratio) / 100; |
9614634f | 7371 | } |
0ff38490 | 7372 | |
6423aa81 JK |
7373 | |
7374 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 7375 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7376 | { |
0ff38490 CL |
7377 | int rc; |
7378 | ||
8d65af78 | 7379 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7380 | if (rc) |
7381 | return rc; | |
7382 | ||
6423aa81 JK |
7383 | setup_min_unmapped_ratio(); |
7384 | ||
7385 | return 0; | |
7386 | } | |
7387 | ||
7388 | static void setup_min_slab_ratio(void) | |
7389 | { | |
7390 | pg_data_t *pgdat; | |
7391 | struct zone *zone; | |
7392 | ||
a5f5f91d MG |
7393 | for_each_online_pgdat(pgdat) |
7394 | pgdat->min_slab_pages = 0; | |
7395 | ||
0ff38490 | 7396 | for_each_zone(zone) |
a5f5f91d | 7397 | zone->zone_pgdat->min_slab_pages += (zone->managed_pages * |
0ff38490 | 7398 | sysctl_min_slab_ratio) / 100; |
6423aa81 JK |
7399 | } |
7400 | ||
7401 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
7402 | void __user *buffer, size_t *length, loff_t *ppos) | |
7403 | { | |
7404 | int rc; | |
7405 | ||
7406 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7407 | if (rc) | |
7408 | return rc; | |
7409 | ||
7410 | setup_min_slab_ratio(); | |
7411 | ||
0ff38490 CL |
7412 | return 0; |
7413 | } | |
9614634f CL |
7414 | #endif |
7415 | ||
1da177e4 LT |
7416 | /* |
7417 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7418 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7419 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7420 | * | |
7421 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7422 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7423 | * if in function of the boot time zone sizes. |
7424 | */ | |
cccad5b9 | 7425 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7426 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7427 | { |
8d65af78 | 7428 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
7429 | setup_per_zone_lowmem_reserve(); |
7430 | return 0; | |
7431 | } | |
7432 | ||
8ad4b1fb RS |
7433 | /* |
7434 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
7435 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
7436 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 7437 | */ |
cccad5b9 | 7438 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7439 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
7440 | { |
7441 | struct zone *zone; | |
7cd2b0a3 | 7442 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
7443 | int ret; |
7444 | ||
7cd2b0a3 DR |
7445 | mutex_lock(&pcp_batch_high_lock); |
7446 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
7447 | ||
8d65af78 | 7448 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
7449 | if (!write || ret < 0) |
7450 | goto out; | |
7451 | ||
7452 | /* Sanity checking to avoid pcp imbalance */ | |
7453 | if (percpu_pagelist_fraction && | |
7454 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
7455 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
7456 | ret = -EINVAL; | |
7457 | goto out; | |
7458 | } | |
7459 | ||
7460 | /* No change? */ | |
7461 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
7462 | goto out; | |
c8e251fa | 7463 | |
364df0eb | 7464 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
7465 | unsigned int cpu; |
7466 | ||
22a7f12b | 7467 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
7468 | pageset_set_high_and_batch(zone, |
7469 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 7470 | } |
7cd2b0a3 | 7471 | out: |
c8e251fa | 7472 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 7473 | return ret; |
8ad4b1fb RS |
7474 | } |
7475 | ||
a9919c79 | 7476 | #ifdef CONFIG_NUMA |
f034b5d4 | 7477 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 7478 | |
1da177e4 LT |
7479 | static int __init set_hashdist(char *str) |
7480 | { | |
7481 | if (!str) | |
7482 | return 0; | |
7483 | hashdist = simple_strtoul(str, &str, 0); | |
7484 | return 1; | |
7485 | } | |
7486 | __setup("hashdist=", set_hashdist); | |
7487 | #endif | |
7488 | ||
f6f34b43 SD |
7489 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
7490 | /* | |
7491 | * Returns the number of pages that arch has reserved but | |
7492 | * is not known to alloc_large_system_hash(). | |
7493 | */ | |
7494 | static unsigned long __init arch_reserved_kernel_pages(void) | |
7495 | { | |
7496 | return 0; | |
7497 | } | |
7498 | #endif | |
7499 | ||
9017217b PT |
7500 | /* |
7501 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
7502 | * machines. As memory size is increased the scale is also increased but at | |
7503 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
7504 | * quadruples the scale is increased by one, which means the size of hash table | |
7505 | * only doubles, instead of quadrupling as well. | |
7506 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
7507 | * makes sense, it is disabled on such platforms. | |
7508 | */ | |
7509 | #if __BITS_PER_LONG > 32 | |
7510 | #define ADAPT_SCALE_BASE (64ul << 30) | |
7511 | #define ADAPT_SCALE_SHIFT 2 | |
7512 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
7513 | #endif | |
7514 | ||
1da177e4 LT |
7515 | /* |
7516 | * allocate a large system hash table from bootmem | |
7517 | * - it is assumed that the hash table must contain an exact power-of-2 | |
7518 | * quantity of entries | |
7519 | * - limit is the number of hash buckets, not the total allocation size | |
7520 | */ | |
7521 | void *__init alloc_large_system_hash(const char *tablename, | |
7522 | unsigned long bucketsize, | |
7523 | unsigned long numentries, | |
7524 | int scale, | |
7525 | int flags, | |
7526 | unsigned int *_hash_shift, | |
7527 | unsigned int *_hash_mask, | |
31fe62b9 TB |
7528 | unsigned long low_limit, |
7529 | unsigned long high_limit) | |
1da177e4 | 7530 | { |
31fe62b9 | 7531 | unsigned long long max = high_limit; |
1da177e4 LT |
7532 | unsigned long log2qty, size; |
7533 | void *table = NULL; | |
3749a8f0 | 7534 | gfp_t gfp_flags; |
1da177e4 LT |
7535 | |
7536 | /* allow the kernel cmdline to have a say */ | |
7537 | if (!numentries) { | |
7538 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 7539 | numentries = nr_kernel_pages; |
f6f34b43 | 7540 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
7541 | |
7542 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
7543 | if (PAGE_SHIFT < 20) | |
7544 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 7545 | |
9017217b PT |
7546 | #if __BITS_PER_LONG > 32 |
7547 | if (!high_limit) { | |
7548 | unsigned long adapt; | |
7549 | ||
7550 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
7551 | adapt <<= ADAPT_SCALE_SHIFT) | |
7552 | scale++; | |
7553 | } | |
7554 | #endif | |
7555 | ||
1da177e4 LT |
7556 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
7557 | if (scale > PAGE_SHIFT) | |
7558 | numentries >>= (scale - PAGE_SHIFT); | |
7559 | else | |
7560 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
7561 | |
7562 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
7563 | if (unlikely(flags & HASH_SMALL)) { |
7564 | /* Makes no sense without HASH_EARLY */ | |
7565 | WARN_ON(!(flags & HASH_EARLY)); | |
7566 | if (!(numentries >> *_hash_shift)) { | |
7567 | numentries = 1UL << *_hash_shift; | |
7568 | BUG_ON(!numentries); | |
7569 | } | |
7570 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 7571 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 7572 | } |
6e692ed3 | 7573 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
7574 | |
7575 | /* limit allocation size to 1/16 total memory by default */ | |
7576 | if (max == 0) { | |
7577 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
7578 | do_div(max, bucketsize); | |
7579 | } | |
074b8517 | 7580 | max = min(max, 0x80000000ULL); |
1da177e4 | 7581 | |
31fe62b9 TB |
7582 | if (numentries < low_limit) |
7583 | numentries = low_limit; | |
1da177e4 LT |
7584 | if (numentries > max) |
7585 | numentries = max; | |
7586 | ||
f0d1b0b3 | 7587 | log2qty = ilog2(numentries); |
1da177e4 | 7588 | |
3749a8f0 | 7589 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 LT |
7590 | do { |
7591 | size = bucketsize << log2qty; | |
ea1f5f37 PT |
7592 | if (flags & HASH_EARLY) { |
7593 | if (flags & HASH_ZERO) | |
7594 | table = memblock_virt_alloc_nopanic(size, 0); | |
7595 | else | |
7596 | table = memblock_virt_alloc_raw(size, 0); | |
7597 | } else if (hashdist) { | |
3749a8f0 | 7598 | table = __vmalloc(size, gfp_flags, PAGE_KERNEL); |
ea1f5f37 | 7599 | } else { |
1037b83b ED |
7600 | /* |
7601 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
7602 | * some pages at the end of hash table which |
7603 | * alloc_pages_exact() automatically does | |
1037b83b | 7604 | */ |
264ef8a9 | 7605 | if (get_order(size) < MAX_ORDER) { |
3749a8f0 PT |
7606 | table = alloc_pages_exact(size, gfp_flags); |
7607 | kmemleak_alloc(table, size, 1, gfp_flags); | |
264ef8a9 | 7608 | } |
1da177e4 LT |
7609 | } |
7610 | } while (!table && size > PAGE_SIZE && --log2qty); | |
7611 | ||
7612 | if (!table) | |
7613 | panic("Failed to allocate %s hash table\n", tablename); | |
7614 | ||
1170532b JP |
7615 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
7616 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
7617 | |
7618 | if (_hash_shift) | |
7619 | *_hash_shift = log2qty; | |
7620 | if (_hash_mask) | |
7621 | *_hash_mask = (1 << log2qty) - 1; | |
7622 | ||
7623 | return table; | |
7624 | } | |
a117e66e | 7625 | |
a5d76b54 | 7626 | /* |
80934513 MK |
7627 | * This function checks whether pageblock includes unmovable pages or not. |
7628 | * If @count is not zero, it is okay to include less @count unmovable pages | |
7629 | * | |
b8af2941 | 7630 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
7631 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
7632 | * check without lock_page also may miss some movable non-lru pages at | |
7633 | * race condition. So you can't expect this function should be exact. | |
a5d76b54 | 7634 | */ |
b023f468 | 7635 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
4da2ce25 | 7636 | int migratetype, |
b023f468 | 7637 | bool skip_hwpoisoned_pages) |
49ac8255 KH |
7638 | { |
7639 | unsigned long pfn, iter, found; | |
47118af0 | 7640 | |
49ac8255 | 7641 | /* |
15c30bc0 MH |
7642 | * TODO we could make this much more efficient by not checking every |
7643 | * page in the range if we know all of them are in MOVABLE_ZONE and | |
7644 | * that the movable zone guarantees that pages are migratable but | |
7645 | * the later is not the case right now unfortunatelly. E.g. movablecore | |
7646 | * can still lead to having bootmem allocations in zone_movable. | |
49ac8255 | 7647 | */ |
49ac8255 | 7648 | |
4da2ce25 MH |
7649 | /* |
7650 | * CMA allocations (alloc_contig_range) really need to mark isolate | |
7651 | * CMA pageblocks even when they are not movable in fact so consider | |
7652 | * them movable here. | |
7653 | */ | |
7654 | if (is_migrate_cma(migratetype) && | |
7655 | is_migrate_cma(get_pageblock_migratetype(page))) | |
7656 | return false; | |
7657 | ||
49ac8255 KH |
7658 | pfn = page_to_pfn(page); |
7659 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
7660 | unsigned long check = pfn + iter; | |
7661 | ||
29723fcc | 7662 | if (!pfn_valid_within(check)) |
49ac8255 | 7663 | continue; |
29723fcc | 7664 | |
49ac8255 | 7665 | page = pfn_to_page(check); |
c8721bbb | 7666 | |
d7ab3672 | 7667 | if (PageReserved(page)) |
15c30bc0 | 7668 | goto unmovable; |
d7ab3672 | 7669 | |
c8721bbb NH |
7670 | /* |
7671 | * Hugepages are not in LRU lists, but they're movable. | |
7672 | * We need not scan over tail pages bacause we don't | |
7673 | * handle each tail page individually in migration. | |
7674 | */ | |
7675 | if (PageHuge(page)) { | |
7676 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
7677 | continue; | |
7678 | } | |
7679 | ||
97d255c8 MK |
7680 | /* |
7681 | * We can't use page_count without pin a page | |
7682 | * because another CPU can free compound page. | |
7683 | * This check already skips compound tails of THP | |
0139aa7b | 7684 | * because their page->_refcount is zero at all time. |
97d255c8 | 7685 | */ |
fe896d18 | 7686 | if (!page_ref_count(page)) { |
49ac8255 KH |
7687 | if (PageBuddy(page)) |
7688 | iter += (1 << page_order(page)) - 1; | |
7689 | continue; | |
7690 | } | |
97d255c8 | 7691 | |
b023f468 WC |
7692 | /* |
7693 | * The HWPoisoned page may be not in buddy system, and | |
7694 | * page_count() is not 0. | |
7695 | */ | |
7696 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
7697 | continue; | |
7698 | ||
0efadf48 YX |
7699 | if (__PageMovable(page)) |
7700 | continue; | |
7701 | ||
49ac8255 KH |
7702 | if (!PageLRU(page)) |
7703 | found++; | |
7704 | /* | |
6b4f7799 JW |
7705 | * If there are RECLAIMABLE pages, we need to check |
7706 | * it. But now, memory offline itself doesn't call | |
7707 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
7708 | */ |
7709 | /* | |
7710 | * If the page is not RAM, page_count()should be 0. | |
7711 | * we don't need more check. This is an _used_ not-movable page. | |
7712 | * | |
7713 | * The problematic thing here is PG_reserved pages. PG_reserved | |
7714 | * is set to both of a memory hole page and a _used_ kernel | |
7715 | * page at boot. | |
7716 | */ | |
7717 | if (found > count) | |
15c30bc0 | 7718 | goto unmovable; |
49ac8255 | 7719 | } |
80934513 | 7720 | return false; |
15c30bc0 MH |
7721 | unmovable: |
7722 | WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE); | |
7723 | return true; | |
49ac8255 KH |
7724 | } |
7725 | ||
080fe206 | 7726 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7727 | |
7728 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7729 | { | |
7730 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7731 | pageblock_nr_pages) - 1); | |
7732 | } | |
7733 | ||
7734 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7735 | { | |
7736 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7737 | pageblock_nr_pages)); | |
7738 | } | |
7739 | ||
041d3a8c | 7740 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7741 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7742 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7743 | { |
7744 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7745 | unsigned long nr_reclaimed; |
041d3a8c MN |
7746 | unsigned long pfn = start; |
7747 | unsigned int tries = 0; | |
7748 | int ret = 0; | |
7749 | ||
be49a6e1 | 7750 | migrate_prep(); |
041d3a8c | 7751 | |
bb13ffeb | 7752 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7753 | if (fatal_signal_pending(current)) { |
7754 | ret = -EINTR; | |
7755 | break; | |
7756 | } | |
7757 | ||
bb13ffeb MG |
7758 | if (list_empty(&cc->migratepages)) { |
7759 | cc->nr_migratepages = 0; | |
edc2ca61 | 7760 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7761 | if (!pfn) { |
7762 | ret = -EINTR; | |
7763 | break; | |
7764 | } | |
7765 | tries = 0; | |
7766 | } else if (++tries == 5) { | |
7767 | ret = ret < 0 ? ret : -EBUSY; | |
7768 | break; | |
7769 | } | |
7770 | ||
beb51eaa MK |
7771 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7772 | &cc->migratepages); | |
7773 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7774 | |
9c620e2b | 7775 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
31025351 | 7776 | NULL, 0, cc->mode, MR_CONTIG_RANGE); |
041d3a8c | 7777 | } |
2a6f5124 SP |
7778 | if (ret < 0) { |
7779 | putback_movable_pages(&cc->migratepages); | |
7780 | return ret; | |
7781 | } | |
7782 | return 0; | |
041d3a8c MN |
7783 | } |
7784 | ||
7785 | /** | |
7786 | * alloc_contig_range() -- tries to allocate given range of pages | |
7787 | * @start: start PFN to allocate | |
7788 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7789 | * @migratetype: migratetype of the underlaying pageblocks (either |
7790 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7791 | * in range must have the same migratetype and it must | |
7792 | * be either of the two. | |
ca96b625 | 7793 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
7794 | * |
7795 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 7796 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 7797 | * |
2c7452a0 MK |
7798 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
7799 | * pageblocks in the range. Once isolated, the pageblocks should not | |
7800 | * be modified by others. | |
041d3a8c MN |
7801 | * |
7802 | * Returns zero on success or negative error code. On success all | |
7803 | * pages which PFN is in [start, end) are allocated for the caller and | |
7804 | * need to be freed with free_contig_range(). | |
7805 | */ | |
0815f3d8 | 7806 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 7807 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 7808 | { |
041d3a8c | 7809 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7810 | unsigned int order; |
7811 | int ret = 0; | |
041d3a8c | 7812 | |
bb13ffeb MG |
7813 | struct compact_control cc = { |
7814 | .nr_migratepages = 0, | |
7815 | .order = -1, | |
7816 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7817 | .mode = MIGRATE_SYNC, |
bb13ffeb | 7818 | .ignore_skip_hint = true, |
2583d671 | 7819 | .no_set_skip_hint = true, |
7dea19f9 | 7820 | .gfp_mask = current_gfp_context(gfp_mask), |
bb13ffeb MG |
7821 | }; |
7822 | INIT_LIST_HEAD(&cc.migratepages); | |
7823 | ||
041d3a8c MN |
7824 | /* |
7825 | * What we do here is we mark all pageblocks in range as | |
7826 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7827 | * have different sizes, and due to the way page allocator | |
7828 | * work, we align the range to biggest of the two pages so | |
7829 | * that page allocator won't try to merge buddies from | |
7830 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7831 | * other migration type. | |
7832 | * | |
7833 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7834 | * migrate the pages from an unaligned range (ie. pages that | |
7835 | * we are interested in). This will put all the pages in | |
7836 | * range back to page allocator as MIGRATE_ISOLATE. | |
7837 | * | |
7838 | * When this is done, we take the pages in range from page | |
7839 | * allocator removing them from the buddy system. This way | |
7840 | * page allocator will never consider using them. | |
7841 | * | |
7842 | * This lets us mark the pageblocks back as | |
7843 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7844 | * aligned range but not in the unaligned, original range are | |
7845 | * put back to page allocator so that buddy can use them. | |
7846 | */ | |
7847 | ||
7848 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7849 | pfn_max_align_up(end), migratetype, |
7850 | false); | |
041d3a8c | 7851 | if (ret) |
86a595f9 | 7852 | return ret; |
041d3a8c | 7853 | |
8ef5849f JK |
7854 | /* |
7855 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
7856 | * So, just fall through. test_pages_isolated() has a tracepoint |
7857 | * which will report the busy page. | |
7858 | * | |
7859 | * It is possible that busy pages could become available before | |
7860 | * the call to test_pages_isolated, and the range will actually be | |
7861 | * allocated. So, if we fall through be sure to clear ret so that | |
7862 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 7863 | */ |
bb13ffeb | 7864 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7865 | if (ret && ret != -EBUSY) |
041d3a8c | 7866 | goto done; |
63cd4489 | 7867 | ret =0; |
041d3a8c MN |
7868 | |
7869 | /* | |
7870 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7871 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7872 | * more, all pages in [start, end) are free in page allocator. | |
7873 | * What we are going to do is to allocate all pages from | |
7874 | * [start, end) (that is remove them from page allocator). | |
7875 | * | |
7876 | * The only problem is that pages at the beginning and at the | |
7877 | * end of interesting range may be not aligned with pages that | |
7878 | * page allocator holds, ie. they can be part of higher order | |
7879 | * pages. Because of this, we reserve the bigger range and | |
7880 | * once this is done free the pages we are not interested in. | |
7881 | * | |
7882 | * We don't have to hold zone->lock here because the pages are | |
7883 | * isolated thus they won't get removed from buddy. | |
7884 | */ | |
7885 | ||
7886 | lru_add_drain_all(); | |
510f5507 | 7887 | drain_all_pages(cc.zone); |
041d3a8c MN |
7888 | |
7889 | order = 0; | |
7890 | outer_start = start; | |
7891 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7892 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7893 | outer_start = start; |
7894 | break; | |
041d3a8c MN |
7895 | } |
7896 | outer_start &= ~0UL << order; | |
7897 | } | |
7898 | ||
8ef5849f JK |
7899 | if (outer_start != start) { |
7900 | order = page_order(pfn_to_page(outer_start)); | |
7901 | ||
7902 | /* | |
7903 | * outer_start page could be small order buddy page and | |
7904 | * it doesn't include start page. Adjust outer_start | |
7905 | * in this case to report failed page properly | |
7906 | * on tracepoint in test_pages_isolated() | |
7907 | */ | |
7908 | if (outer_start + (1UL << order) <= start) | |
7909 | outer_start = start; | |
7910 | } | |
7911 | ||
041d3a8c | 7912 | /* Make sure the range is really isolated. */ |
b023f468 | 7913 | if (test_pages_isolated(outer_start, end, false)) { |
75dddef3 | 7914 | pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n", |
dae803e1 | 7915 | __func__, outer_start, end); |
041d3a8c MN |
7916 | ret = -EBUSY; |
7917 | goto done; | |
7918 | } | |
7919 | ||
49f223a9 | 7920 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7921 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7922 | if (!outer_end) { |
7923 | ret = -EBUSY; | |
7924 | goto done; | |
7925 | } | |
7926 | ||
7927 | /* Free head and tail (if any) */ | |
7928 | if (start != outer_start) | |
7929 | free_contig_range(outer_start, start - outer_start); | |
7930 | if (end != outer_end) | |
7931 | free_contig_range(end, outer_end - end); | |
7932 | ||
7933 | done: | |
7934 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7935 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7936 | return ret; |
7937 | } | |
7938 | ||
7939 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7940 | { | |
bcc2b02f MS |
7941 | unsigned int count = 0; |
7942 | ||
7943 | for (; nr_pages--; pfn++) { | |
7944 | struct page *page = pfn_to_page(pfn); | |
7945 | ||
7946 | count += page_count(page) != 1; | |
7947 | __free_page(page); | |
7948 | } | |
7949 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7950 | } |
7951 | #endif | |
7952 | ||
d883c6cf | 7953 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7954 | /* |
7955 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7956 | * page high values need to be recalulated. | |
7957 | */ | |
4ed7e022 JL |
7958 | void __meminit zone_pcp_update(struct zone *zone) |
7959 | { | |
0a647f38 | 7960 | unsigned cpu; |
c8e251fa | 7961 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7962 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7963 | pageset_set_high_and_batch(zone, |
7964 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7965 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7966 | } |
7967 | #endif | |
7968 | ||
340175b7 JL |
7969 | void zone_pcp_reset(struct zone *zone) |
7970 | { | |
7971 | unsigned long flags; | |
5a883813 MK |
7972 | int cpu; |
7973 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7974 | |
7975 | /* avoid races with drain_pages() */ | |
7976 | local_irq_save(flags); | |
7977 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7978 | for_each_online_cpu(cpu) { |
7979 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7980 | drain_zonestat(zone, pset); | |
7981 | } | |
340175b7 JL |
7982 | free_percpu(zone->pageset); |
7983 | zone->pageset = &boot_pageset; | |
7984 | } | |
7985 | local_irq_restore(flags); | |
7986 | } | |
7987 | ||
6dcd73d7 | 7988 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 7989 | /* |
b9eb6319 JK |
7990 | * All pages in the range must be in a single zone and isolated |
7991 | * before calling this. | |
0c0e6195 KH |
7992 | */ |
7993 | void | |
7994 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7995 | { | |
7996 | struct page *page; | |
7997 | struct zone *zone; | |
7aeb09f9 | 7998 | unsigned int order, i; |
0c0e6195 KH |
7999 | unsigned long pfn; |
8000 | unsigned long flags; | |
8001 | /* find the first valid pfn */ | |
8002 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
8003 | if (pfn_valid(pfn)) | |
8004 | break; | |
8005 | if (pfn == end_pfn) | |
8006 | return; | |
2d070eab | 8007 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
8008 | zone = page_zone(pfn_to_page(pfn)); |
8009 | spin_lock_irqsave(&zone->lock, flags); | |
8010 | pfn = start_pfn; | |
8011 | while (pfn < end_pfn) { | |
8012 | if (!pfn_valid(pfn)) { | |
8013 | pfn++; | |
8014 | continue; | |
8015 | } | |
8016 | page = pfn_to_page(pfn); | |
b023f468 WC |
8017 | /* |
8018 | * The HWPoisoned page may be not in buddy system, and | |
8019 | * page_count() is not 0. | |
8020 | */ | |
8021 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
8022 | pfn++; | |
8023 | SetPageReserved(page); | |
8024 | continue; | |
8025 | } | |
8026 | ||
0c0e6195 KH |
8027 | BUG_ON(page_count(page)); |
8028 | BUG_ON(!PageBuddy(page)); | |
8029 | order = page_order(page); | |
8030 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
8031 | pr_info("remove from free list %lx %d %lx\n", |
8032 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
8033 | #endif |
8034 | list_del(&page->lru); | |
8035 | rmv_page_order(page); | |
8036 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
8037 | for (i = 0; i < (1 << order); i++) |
8038 | SetPageReserved((page+i)); | |
8039 | pfn += (1 << order); | |
8040 | } | |
8041 | spin_unlock_irqrestore(&zone->lock, flags); | |
8042 | } | |
8043 | #endif | |
8d22ba1b | 8044 | |
8d22ba1b WF |
8045 | bool is_free_buddy_page(struct page *page) |
8046 | { | |
8047 | struct zone *zone = page_zone(page); | |
8048 | unsigned long pfn = page_to_pfn(page); | |
8049 | unsigned long flags; | |
7aeb09f9 | 8050 | unsigned int order; |
8d22ba1b WF |
8051 | |
8052 | spin_lock_irqsave(&zone->lock, flags); | |
8053 | for (order = 0; order < MAX_ORDER; order++) { | |
8054 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8055 | ||
8056 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
8057 | break; | |
8058 | } | |
8059 | spin_unlock_irqrestore(&zone->lock, flags); | |
8060 | ||
8061 | return order < MAX_ORDER; | |
8062 | } |