]>
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/topology.h> |
36 | #include <linux/sysctl.h> | |
37 | #include <linux/cpu.h> | |
38 | #include <linux/cpuset.h> | |
bdc8cb98 | 39 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
40 | #include <linux/nodemask.h> |
41 | #include <linux/vmalloc.h> | |
a6cccdc3 | 42 | #include <linux/vmstat.h> |
4be38e35 | 43 | #include <linux/mempolicy.h> |
4b94ffdc | 44 | #include <linux/memremap.h> |
6811378e | 45 | #include <linux/stop_machine.h> |
c713216d MG |
46 | #include <linux/sort.h> |
47 | #include <linux/pfn.h> | |
3fcfab16 | 48 | #include <linux/backing-dev.h> |
933e312e | 49 | #include <linux/fault-inject.h> |
a5d76b54 | 50 | #include <linux/page-isolation.h> |
eefa864b | 51 | #include <linux/page_ext.h> |
3ac7fe5a | 52 | #include <linux/debugobjects.h> |
dbb1f81c | 53 | #include <linux/kmemleak.h> |
56de7263 | 54 | #include <linux/compaction.h> |
0d3d062a | 55 | #include <trace/events/kmem.h> |
d379f01d | 56 | #include <trace/events/oom.h> |
268bb0ce | 57 | #include <linux/prefetch.h> |
6e543d57 | 58 | #include <linux/mm_inline.h> |
041d3a8c | 59 | #include <linux/migrate.h> |
949f7ec5 | 60 | #include <linux/hugetlb.h> |
8bd75c77 | 61 | #include <linux/sched/rt.h> |
5b3cc15a | 62 | #include <linux/sched/mm.h> |
48c96a36 | 63 | #include <linux/page_owner.h> |
0e1cc95b | 64 | #include <linux/kthread.h> |
4949148a | 65 | #include <linux/memcontrol.h> |
42c269c8 | 66 | #include <linux/ftrace.h> |
d92a8cfc | 67 | #include <linux/lockdep.h> |
556b969a | 68 | #include <linux/nmi.h> |
eb414681 | 69 | #include <linux/psi.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 | ||
d483da5b AD |
1235 | /* |
1236 | * no need for atomic set_bit because the struct | |
1237 | * page is not visible yet so nobody should | |
1238 | * access it yet. | |
1239 | */ | |
1240 | __SetPageReserved(page); | |
7e18adb4 MG |
1241 | } |
1242 | } | |
92923ca3 NZ |
1243 | } |
1244 | ||
ec95f53a KM |
1245 | static void __free_pages_ok(struct page *page, unsigned int order) |
1246 | { | |
d34b0733 | 1247 | unsigned long flags; |
95e34412 | 1248 | int migratetype; |
dc4b0caf | 1249 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1250 | |
e2769dbd | 1251 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1252 | return; |
1253 | ||
cfc47a28 | 1254 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1255 | local_irq_save(flags); |
1256 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1257 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1258 | local_irq_restore(flags); |
1da177e4 LT |
1259 | } |
1260 | ||
949698a3 | 1261 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1262 | { |
c3993076 | 1263 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1264 | struct page *p = page; |
c3993076 | 1265 | unsigned int loop; |
a226f6c8 | 1266 | |
e2d0bd2b YL |
1267 | prefetchw(p); |
1268 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1269 | prefetchw(p + 1); | |
c3993076 JW |
1270 | __ClearPageReserved(p); |
1271 | set_page_count(p, 0); | |
a226f6c8 | 1272 | } |
e2d0bd2b YL |
1273 | __ClearPageReserved(p); |
1274 | set_page_count(p, 0); | |
c3993076 | 1275 | |
e2d0bd2b | 1276 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1277 | set_page_refcounted(page); |
1278 | __free_pages(page, order); | |
a226f6c8 DH |
1279 | } |
1280 | ||
75a592a4 MG |
1281 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1282 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1283 | |
75a592a4 MG |
1284 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1285 | ||
1286 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1287 | { | |
7ace9917 | 1288 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1289 | int nid; |
1290 | ||
7ace9917 | 1291 | spin_lock(&early_pfn_lock); |
75a592a4 | 1292 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1293 | if (nid < 0) |
e4568d38 | 1294 | nid = first_online_node; |
7ace9917 MG |
1295 | spin_unlock(&early_pfn_lock); |
1296 | ||
1297 | return nid; | |
75a592a4 MG |
1298 | } |
1299 | #endif | |
1300 | ||
1301 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
d73d3c9f MK |
1302 | static inline bool __meminit __maybe_unused |
1303 | meminit_pfn_in_nid(unsigned long pfn, int node, | |
1304 | struct mminit_pfnnid_cache *state) | |
75a592a4 MG |
1305 | { |
1306 | int nid; | |
1307 | ||
1308 | nid = __early_pfn_to_nid(pfn, state); | |
1309 | if (nid >= 0 && nid != node) | |
1310 | return false; | |
1311 | return true; | |
1312 | } | |
1313 | ||
1314 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1315 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1316 | { | |
1317 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1318 | } | |
1319 | ||
1320 | #else | |
1321 | ||
1322 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1323 | { | |
1324 | return true; | |
1325 | } | |
d73d3c9f MK |
1326 | static inline bool __meminit __maybe_unused |
1327 | meminit_pfn_in_nid(unsigned long pfn, int node, | |
1328 | struct mminit_pfnnid_cache *state) | |
75a592a4 MG |
1329 | { |
1330 | return true; | |
1331 | } | |
1332 | #endif | |
1333 | ||
1334 | ||
0e1cc95b | 1335 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1336 | unsigned int order) |
1337 | { | |
1338 | if (early_page_uninitialised(pfn)) | |
1339 | return; | |
949698a3 | 1340 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1341 | } |
1342 | ||
7cf91a98 JK |
1343 | /* |
1344 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1345 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1346 | * with the migration of free compaction scanner. The scanners then need to | |
1347 | * use only pfn_valid_within() check for arches that allow holes within | |
1348 | * pageblocks. | |
1349 | * | |
1350 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1351 | * | |
1352 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1353 | * i.e. it's possible that all pages within a zones range of pages do not | |
1354 | * belong to a single zone. We assume that a border between node0 and node1 | |
1355 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1356 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1357 | * the first and last page of a pageblock and avoid checking each individual | |
1358 | * page in a pageblock. | |
1359 | */ | |
1360 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1361 | unsigned long end_pfn, struct zone *zone) | |
1362 | { | |
1363 | struct page *start_page; | |
1364 | struct page *end_page; | |
1365 | ||
1366 | /* end_pfn is one past the range we are checking */ | |
1367 | end_pfn--; | |
1368 | ||
1369 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1370 | return NULL; | |
1371 | ||
2d070eab MH |
1372 | start_page = pfn_to_online_page(start_pfn); |
1373 | if (!start_page) | |
1374 | return NULL; | |
7cf91a98 JK |
1375 | |
1376 | if (page_zone(start_page) != zone) | |
1377 | return NULL; | |
1378 | ||
1379 | end_page = pfn_to_page(end_pfn); | |
1380 | ||
1381 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1382 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1383 | return NULL; | |
1384 | ||
1385 | return start_page; | |
1386 | } | |
1387 | ||
1388 | void set_zone_contiguous(struct zone *zone) | |
1389 | { | |
1390 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1391 | unsigned long block_end_pfn; | |
1392 | ||
1393 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1394 | for (; block_start_pfn < zone_end_pfn(zone); | |
1395 | block_start_pfn = block_end_pfn, | |
1396 | block_end_pfn += pageblock_nr_pages) { | |
1397 | ||
1398 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1399 | ||
1400 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1401 | block_end_pfn, zone)) | |
1402 | return; | |
1403 | } | |
1404 | ||
1405 | /* We confirm that there is no hole */ | |
1406 | zone->contiguous = true; | |
1407 | } | |
1408 | ||
1409 | void clear_zone_contiguous(struct zone *zone) | |
1410 | { | |
1411 | zone->contiguous = false; | |
1412 | } | |
1413 | ||
7e18adb4 | 1414 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1415 | static void __init deferred_free_range(unsigned long pfn, |
1416 | unsigned long nr_pages) | |
a4de83dd | 1417 | { |
2f47a91f PT |
1418 | struct page *page; |
1419 | unsigned long i; | |
a4de83dd | 1420 | |
2f47a91f | 1421 | if (!nr_pages) |
a4de83dd MG |
1422 | return; |
1423 | ||
2f47a91f PT |
1424 | page = pfn_to_page(pfn); |
1425 | ||
a4de83dd | 1426 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1427 | if (nr_pages == pageblock_nr_pages && |
1428 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1429 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
e780149b | 1430 | __free_pages_boot_core(page, pageblock_order); |
a4de83dd MG |
1431 | return; |
1432 | } | |
1433 | ||
e780149b XQ |
1434 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1435 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1436 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
949698a3 | 1437 | __free_pages_boot_core(page, 0); |
e780149b | 1438 | } |
a4de83dd MG |
1439 | } |
1440 | ||
d3cd131d NS |
1441 | /* Completion tracking for deferred_init_memmap() threads */ |
1442 | static atomic_t pgdat_init_n_undone __initdata; | |
1443 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1444 | ||
1445 | static inline void __init pgdat_init_report_one_done(void) | |
1446 | { | |
1447 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1448 | complete(&pgdat_init_all_done_comp); | |
1449 | } | |
0e1cc95b | 1450 | |
2f47a91f | 1451 | /* |
80b1f41c PT |
1452 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1453 | * | |
1454 | * First we check if pfn is valid on architectures where it is possible to have | |
1455 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1456 | * function is optimized out. | |
1457 | * | |
1458 | * Then, we check if a current large page is valid by only checking the validity | |
1459 | * of the head pfn. | |
1460 | * | |
1461 | * Finally, meminit_pfn_in_nid is checked on systems where pfns can interleave | |
1462 | * within a node: a pfn is between start and end of a node, but does not belong | |
1463 | * to this memory node. | |
2f47a91f | 1464 | */ |
80b1f41c PT |
1465 | static inline bool __init |
1466 | deferred_pfn_valid(int nid, unsigned long pfn, | |
1467 | struct mminit_pfnnid_cache *nid_init_state) | |
2f47a91f | 1468 | { |
80b1f41c PT |
1469 | if (!pfn_valid_within(pfn)) |
1470 | return false; | |
1471 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1472 | return false; | |
1473 | if (!meminit_pfn_in_nid(pfn, nid, nid_init_state)) | |
1474 | return false; | |
1475 | return true; | |
1476 | } | |
2f47a91f | 1477 | |
80b1f41c PT |
1478 | /* |
1479 | * Free pages to buddy allocator. Try to free aligned pages in | |
1480 | * pageblock_nr_pages sizes. | |
1481 | */ | |
1482 | static void __init deferred_free_pages(int nid, int zid, unsigned long pfn, | |
1483 | unsigned long end_pfn) | |
1484 | { | |
1485 | struct mminit_pfnnid_cache nid_init_state = { }; | |
1486 | unsigned long nr_pgmask = pageblock_nr_pages - 1; | |
1487 | unsigned long nr_free = 0; | |
2f47a91f | 1488 | |
80b1f41c PT |
1489 | for (; pfn < end_pfn; pfn++) { |
1490 | if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) { | |
1491 | deferred_free_range(pfn - nr_free, nr_free); | |
1492 | nr_free = 0; | |
1493 | } else if (!(pfn & nr_pgmask)) { | |
1494 | deferred_free_range(pfn - nr_free, nr_free); | |
1495 | nr_free = 1; | |
3a2d7fa8 | 1496 | touch_nmi_watchdog(); |
80b1f41c PT |
1497 | } else { |
1498 | nr_free++; | |
1499 | } | |
1500 | } | |
1501 | /* Free the last block of pages to allocator */ | |
1502 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1503 | } |
1504 | ||
80b1f41c PT |
1505 | /* |
1506 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1507 | * by performing it only once every pageblock_nr_pages. | |
1508 | * Return number of pages initialized. | |
1509 | */ | |
1510 | static unsigned long __init deferred_init_pages(int nid, int zid, | |
1511 | unsigned long pfn, | |
1512 | unsigned long end_pfn) | |
2f47a91f PT |
1513 | { |
1514 | struct mminit_pfnnid_cache nid_init_state = { }; | |
1515 | unsigned long nr_pgmask = pageblock_nr_pages - 1; | |
2f47a91f | 1516 | unsigned long nr_pages = 0; |
2f47a91f | 1517 | struct page *page = NULL; |
2f47a91f | 1518 | |
80b1f41c PT |
1519 | for (; pfn < end_pfn; pfn++) { |
1520 | if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) { | |
1521 | page = NULL; | |
2f47a91f | 1522 | continue; |
80b1f41c | 1523 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1524 | page = pfn_to_page(pfn); |
3a2d7fa8 | 1525 | touch_nmi_watchdog(); |
80b1f41c PT |
1526 | } else { |
1527 | page++; | |
2f47a91f | 1528 | } |
d0dc12e8 | 1529 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1530 | nr_pages++; |
2f47a91f | 1531 | } |
80b1f41c | 1532 | return (nr_pages); |
2f47a91f PT |
1533 | } |
1534 | ||
7e18adb4 | 1535 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1536 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1537 | { |
0e1cc95b MG |
1538 | pg_data_t *pgdat = data; |
1539 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1540 | unsigned long start = jiffies; |
1541 | unsigned long nr_pages = 0; | |
3a2d7fa8 | 1542 | unsigned long spfn, epfn, first_init_pfn, flags; |
2f47a91f PT |
1543 | phys_addr_t spa, epa; |
1544 | int zid; | |
7e18adb4 | 1545 | struct zone *zone; |
0e1cc95b | 1546 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
2f47a91f | 1547 | u64 i; |
7e18adb4 | 1548 | |
3a2d7fa8 PT |
1549 | /* Bind memory initialisation thread to a local node if possible */ |
1550 | if (!cpumask_empty(cpumask)) | |
1551 | set_cpus_allowed_ptr(current, cpumask); | |
1552 | ||
1553 | pgdat_resize_lock(pgdat, &flags); | |
1554 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1555 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1556 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1557 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1558 | return 0; |
1559 | } | |
1560 | ||
7e18adb4 MG |
1561 | /* Sanity check boundaries */ |
1562 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1563 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1564 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1565 | ||
1566 | /* Only the highest zone is deferred so find it */ | |
1567 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1568 | zone = pgdat->node_zones + zid; | |
1569 | if (first_init_pfn < zone_end_pfn(zone)) | |
1570 | break; | |
1571 | } | |
2f47a91f | 1572 | first_init_pfn = max(zone->zone_start_pfn, first_init_pfn); |
7e18adb4 | 1573 | |
80b1f41c PT |
1574 | /* |
1575 | * Initialize and free pages. We do it in two loops: first we initialize | |
1576 | * struct page, than free to buddy allocator, because while we are | |
1577 | * freeing pages we can access pages that are ahead (computing buddy | |
1578 | * page in __free_one_page()). | |
1579 | */ | |
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)); | |
1583 | nr_pages += deferred_init_pages(nid, zid, spfn, epfn); | |
1584 | } | |
2f47a91f PT |
1585 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { |
1586 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1587 | epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa)); | |
80b1f41c | 1588 | deferred_free_pages(nid, zid, spfn, epfn); |
7e18adb4 | 1589 | } |
3a2d7fa8 | 1590 | pgdat_resize_unlock(pgdat, &flags); |
7e18adb4 MG |
1591 | |
1592 | /* Sanity check that the next zone really is unpopulated */ | |
1593 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1594 | ||
0e1cc95b | 1595 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1596 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1597 | |
1598 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1599 | return 0; |
1600 | } | |
c9e97a19 PT |
1601 | |
1602 | /* | |
1603 | * During boot we initialize deferred pages on-demand, as needed, but once | |
1604 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
1605 | * and we can permanently disable that path. | |
1606 | */ | |
1607 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
1608 | ||
1609 | /* | |
1610 | * If this zone has deferred pages, try to grow it by initializing enough | |
1611 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
1612 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
1613 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
1614 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
1615 | * | |
1616 | * Return true when zone was grown, otherwise return false. We return true even | |
1617 | * when we grow less than requested, to let the caller decide if there are | |
1618 | * enough pages to satisfy the allocation. | |
1619 | * | |
1620 | * Note: We use noinline because this function is needed only during boot, and | |
1621 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
1622 | * making sure that it is not inlined into permanent text section. | |
1623 | */ | |
1624 | static noinline bool __init | |
1625 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
1626 | { | |
1627 | int zid = zone_idx(zone); | |
1628 | int nid = zone_to_nid(zone); | |
1629 | pg_data_t *pgdat = NODE_DATA(nid); | |
1630 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); | |
1631 | unsigned long nr_pages = 0; | |
1632 | unsigned long first_init_pfn, spfn, epfn, t, flags; | |
1633 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; | |
1634 | phys_addr_t spa, epa; | |
1635 | u64 i; | |
1636 | ||
1637 | /* Only the last zone may have deferred pages */ | |
1638 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
1639 | return false; | |
1640 | ||
1641 | pgdat_resize_lock(pgdat, &flags); | |
1642 | ||
1643 | /* | |
1644 | * If deferred pages have been initialized while we were waiting for | |
1645 | * the lock, return true, as the zone was grown. The caller will retry | |
1646 | * this zone. We won't return to this function since the caller also | |
1647 | * has this static branch. | |
1648 | */ | |
1649 | if (!static_branch_unlikely(&deferred_pages)) { | |
1650 | pgdat_resize_unlock(pgdat, &flags); | |
1651 | return true; | |
1652 | } | |
1653 | ||
1654 | /* | |
1655 | * If someone grew this zone while we were waiting for spinlock, return | |
1656 | * true, as there might be enough pages already. | |
1657 | */ | |
1658 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
1659 | pgdat_resize_unlock(pgdat, &flags); | |
1660 | return true; | |
1661 | } | |
1662 | ||
1663 | first_init_pfn = max(zone->zone_start_pfn, first_deferred_pfn); | |
1664 | ||
1665 | if (first_init_pfn >= pgdat_end_pfn(pgdat)) { | |
1666 | pgdat_resize_unlock(pgdat, &flags); | |
1667 | return false; | |
1668 | } | |
1669 | ||
1670 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { | |
1671 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1672 | epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa)); | |
1673 | ||
1674 | while (spfn < epfn && nr_pages < nr_pages_needed) { | |
1675 | t = ALIGN(spfn + PAGES_PER_SECTION, PAGES_PER_SECTION); | |
1676 | first_deferred_pfn = min(t, epfn); | |
1677 | nr_pages += deferred_init_pages(nid, zid, spfn, | |
1678 | first_deferred_pfn); | |
1679 | spfn = first_deferred_pfn; | |
1680 | } | |
1681 | ||
1682 | if (nr_pages >= nr_pages_needed) | |
1683 | break; | |
1684 | } | |
1685 | ||
1686 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { | |
1687 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1688 | epfn = min_t(unsigned long, first_deferred_pfn, PFN_DOWN(epa)); | |
1689 | deferred_free_pages(nid, zid, spfn, epfn); | |
1690 | ||
1691 | if (first_deferred_pfn == epfn) | |
1692 | break; | |
1693 | } | |
1694 | pgdat->first_deferred_pfn = first_deferred_pfn; | |
1695 | pgdat_resize_unlock(pgdat, &flags); | |
1696 | ||
1697 | return nr_pages > 0; | |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * deferred_grow_zone() is __init, but it is called from | |
1702 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
1703 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
1704 | * and to ensure that the function body gets unloaded. | |
1705 | */ | |
1706 | static bool __ref | |
1707 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
1708 | { | |
1709 | return deferred_grow_zone(zone, order); | |
1710 | } | |
1711 | ||
7cf91a98 | 1712 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1713 | |
1714 | void __init page_alloc_init_late(void) | |
1715 | { | |
7cf91a98 JK |
1716 | struct zone *zone; |
1717 | ||
1718 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1719 | int nid; |
1720 | ||
d3cd131d NS |
1721 | /* There will be num_node_state(N_MEMORY) threads */ |
1722 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1723 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1724 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1725 | } | |
1726 | ||
1727 | /* Block until all are initialised */ | |
d3cd131d | 1728 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 1729 | |
c9e97a19 PT |
1730 | /* |
1731 | * We initialized the rest of the deferred pages. Permanently disable | |
1732 | * on-demand struct page initialization. | |
1733 | */ | |
1734 | static_branch_disable(&deferred_pages); | |
1735 | ||
4248b0da MG |
1736 | /* Reinit limits that are based on free pages after the kernel is up */ |
1737 | files_maxfiles_init(); | |
7cf91a98 | 1738 | #endif |
3010f876 PT |
1739 | #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK |
1740 | /* Discard memblock private memory */ | |
1741 | memblock_discard(); | |
1742 | #endif | |
7cf91a98 JK |
1743 | |
1744 | for_each_populated_zone(zone) | |
1745 | set_zone_contiguous(zone); | |
7e18adb4 | 1746 | } |
7e18adb4 | 1747 | |
47118af0 | 1748 | #ifdef CONFIG_CMA |
9cf510a5 | 1749 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1750 | void __init init_cma_reserved_pageblock(struct page *page) |
1751 | { | |
1752 | unsigned i = pageblock_nr_pages; | |
1753 | struct page *p = page; | |
1754 | ||
1755 | do { | |
1756 | __ClearPageReserved(p); | |
1757 | set_page_count(p, 0); | |
d883c6cf | 1758 | } while (++p, --i); |
47118af0 | 1759 | |
47118af0 | 1760 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1761 | |
1762 | if (pageblock_order >= MAX_ORDER) { | |
1763 | i = pageblock_nr_pages; | |
1764 | p = page; | |
1765 | do { | |
1766 | set_page_refcounted(p); | |
1767 | __free_pages(p, MAX_ORDER - 1); | |
1768 | p += MAX_ORDER_NR_PAGES; | |
1769 | } while (i -= MAX_ORDER_NR_PAGES); | |
1770 | } else { | |
1771 | set_page_refcounted(page); | |
1772 | __free_pages(page, pageblock_order); | |
1773 | } | |
1774 | ||
3dcc0571 | 1775 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1776 | } |
1777 | #endif | |
1da177e4 LT |
1778 | |
1779 | /* | |
1780 | * The order of subdivision here is critical for the IO subsystem. | |
1781 | * Please do not alter this order without good reasons and regression | |
1782 | * testing. Specifically, as large blocks of memory are subdivided, | |
1783 | * the order in which smaller blocks are delivered depends on the order | |
1784 | * they're subdivided in this function. This is the primary factor | |
1785 | * influencing the order in which pages are delivered to the IO | |
1786 | * subsystem according to empirical testing, and this is also justified | |
1787 | * by considering the behavior of a buddy system containing a single | |
1788 | * large block of memory acted on by a series of small allocations. | |
1789 | * This behavior is a critical factor in sglist merging's success. | |
1790 | * | |
6d49e352 | 1791 | * -- nyc |
1da177e4 | 1792 | */ |
085cc7d5 | 1793 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1794 | int low, int high, struct free_area *area, |
1795 | int migratetype) | |
1da177e4 LT |
1796 | { |
1797 | unsigned long size = 1 << high; | |
1798 | ||
1799 | while (high > low) { | |
1800 | area--; | |
1801 | high--; | |
1802 | size >>= 1; | |
309381fe | 1803 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1804 | |
acbc15a4 JK |
1805 | /* |
1806 | * Mark as guard pages (or page), that will allow to | |
1807 | * merge back to allocator when buddy will be freed. | |
1808 | * Corresponding page table entries will not be touched, | |
1809 | * pages will stay not present in virtual address space | |
1810 | */ | |
1811 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 1812 | continue; |
acbc15a4 | 1813 | |
b2a0ac88 | 1814 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1815 | area->nr_free++; |
1816 | set_page_order(&page[size], high); | |
1817 | } | |
1da177e4 LT |
1818 | } |
1819 | ||
4e611801 | 1820 | static void check_new_page_bad(struct page *page) |
1da177e4 | 1821 | { |
4e611801 VB |
1822 | const char *bad_reason = NULL; |
1823 | unsigned long bad_flags = 0; | |
7bfec6f4 | 1824 | |
53f9263b | 1825 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1826 | bad_reason = "nonzero mapcount"; |
1827 | if (unlikely(page->mapping != NULL)) | |
1828 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1829 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1830 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1831 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1832 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1833 | bad_flags = __PG_HWPOISON; | |
e570f56c NH |
1834 | /* Don't complain about hwpoisoned pages */ |
1835 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1836 | return; | |
f4c18e6f | 1837 | } |
f0b791a3 DH |
1838 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1839 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1840 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1841 | } | |
9edad6ea JW |
1842 | #ifdef CONFIG_MEMCG |
1843 | if (unlikely(page->mem_cgroup)) | |
1844 | bad_reason = "page still charged to cgroup"; | |
1845 | #endif | |
4e611801 VB |
1846 | bad_page(page, bad_reason, bad_flags); |
1847 | } | |
1848 | ||
1849 | /* | |
1850 | * This page is about to be returned from the page allocator | |
1851 | */ | |
1852 | static inline int check_new_page(struct page *page) | |
1853 | { | |
1854 | if (likely(page_expected_state(page, | |
1855 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1856 | return 0; | |
1857 | ||
1858 | check_new_page_bad(page); | |
1859 | return 1; | |
2a7684a2 WF |
1860 | } |
1861 | ||
bd33ef36 | 1862 | static inline bool free_pages_prezeroed(void) |
1414c7f4 LA |
1863 | { |
1864 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
bd33ef36 | 1865 | page_poisoning_enabled(); |
1414c7f4 LA |
1866 | } |
1867 | ||
479f854a MG |
1868 | #ifdef CONFIG_DEBUG_VM |
1869 | static bool check_pcp_refill(struct page *page) | |
1870 | { | |
1871 | return false; | |
1872 | } | |
1873 | ||
1874 | static bool check_new_pcp(struct page *page) | |
1875 | { | |
1876 | return check_new_page(page); | |
1877 | } | |
1878 | #else | |
1879 | static bool check_pcp_refill(struct page *page) | |
1880 | { | |
1881 | return check_new_page(page); | |
1882 | } | |
1883 | static bool check_new_pcp(struct page *page) | |
1884 | { | |
1885 | return false; | |
1886 | } | |
1887 | #endif /* CONFIG_DEBUG_VM */ | |
1888 | ||
1889 | static bool check_new_pages(struct page *page, unsigned int order) | |
1890 | { | |
1891 | int i; | |
1892 | for (i = 0; i < (1 << order); i++) { | |
1893 | struct page *p = page + i; | |
1894 | ||
1895 | if (unlikely(check_new_page(p))) | |
1896 | return true; | |
1897 | } | |
1898 | ||
1899 | return false; | |
1900 | } | |
1901 | ||
46f24fd8 JK |
1902 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1903 | gfp_t gfp_flags) | |
1904 | { | |
1905 | set_page_private(page, 0); | |
1906 | set_page_refcounted(page); | |
1907 | ||
1908 | arch_alloc_page(page, order); | |
1909 | kernel_map_pages(page, 1 << order, 1); | |
1910 | kernel_poison_pages(page, 1 << order, 1); | |
1911 | kasan_alloc_pages(page, order); | |
1912 | set_page_owner(page, order, gfp_flags); | |
1913 | } | |
1914 | ||
479f854a | 1915 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1916 | unsigned int alloc_flags) |
2a7684a2 WF |
1917 | { |
1918 | int i; | |
689bcebf | 1919 | |
46f24fd8 | 1920 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1921 | |
bd33ef36 | 1922 | if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1923 | for (i = 0; i < (1 << order); i++) |
1924 | clear_highpage(page + i); | |
17cf4406 NP |
1925 | |
1926 | if (order && (gfp_flags & __GFP_COMP)) | |
1927 | prep_compound_page(page, order); | |
1928 | ||
75379191 | 1929 | /* |
2f064f34 | 1930 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1931 | * allocate the page. The expectation is that the caller is taking |
1932 | * steps that will free more memory. The caller should avoid the page | |
1933 | * being used for !PFMEMALLOC purposes. | |
1934 | */ | |
2f064f34 MH |
1935 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1936 | set_page_pfmemalloc(page); | |
1937 | else | |
1938 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1939 | } |
1940 | ||
56fd56b8 MG |
1941 | /* |
1942 | * Go through the free lists for the given migratetype and remove | |
1943 | * the smallest available page from the freelists | |
1944 | */ | |
85ccc8fa | 1945 | static __always_inline |
728ec980 | 1946 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
1947 | int migratetype) |
1948 | { | |
1949 | unsigned int current_order; | |
b8af2941 | 1950 | struct free_area *area; |
56fd56b8 MG |
1951 | struct page *page; |
1952 | ||
1953 | /* Find a page of the appropriate size in the preferred list */ | |
1954 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1955 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1956 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1957 | struct page, lru); |
a16601c5 GT |
1958 | if (!page) |
1959 | continue; | |
56fd56b8 MG |
1960 | list_del(&page->lru); |
1961 | rmv_page_order(page); | |
1962 | area->nr_free--; | |
56fd56b8 | 1963 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1964 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1965 | return page; |
1966 | } | |
1967 | ||
1968 | return NULL; | |
1969 | } | |
1970 | ||
1971 | ||
b2a0ac88 MG |
1972 | /* |
1973 | * This array describes the order lists are fallen back to when | |
1974 | * the free lists for the desirable migrate type are depleted | |
1975 | */ | |
47118af0 | 1976 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1977 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1978 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1979 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1980 | #ifdef CONFIG_CMA |
974a786e | 1981 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1982 | #endif |
194159fb | 1983 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1984 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1985 | #endif |
b2a0ac88 MG |
1986 | }; |
1987 | ||
dc67647b | 1988 | #ifdef CONFIG_CMA |
85ccc8fa | 1989 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
1990 | unsigned int order) |
1991 | { | |
1992 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1993 | } | |
1994 | #else | |
1995 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1996 | unsigned int order) { return NULL; } | |
1997 | #endif | |
1998 | ||
c361be55 MG |
1999 | /* |
2000 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 2001 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2002 | * boundary. If alignment is required, use move_freepages_block() |
2003 | */ | |
02aa0cdd | 2004 | static int move_freepages(struct zone *zone, |
b69a7288 | 2005 | struct page *start_page, struct page *end_page, |
02aa0cdd | 2006 | int migratetype, int *num_movable) |
c361be55 MG |
2007 | { |
2008 | struct page *page; | |
d00181b9 | 2009 | unsigned int order; |
d100313f | 2010 | int pages_moved = 0; |
c361be55 MG |
2011 | |
2012 | #ifndef CONFIG_HOLES_IN_ZONE | |
2013 | /* | |
2014 | * page_zone is not safe to call in this context when | |
2015 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
2016 | * anyway as we check zone boundaries in move_freepages_block(). | |
2017 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 2018 | * grouping pages by mobility |
c361be55 | 2019 | */ |
3e04040d AB |
2020 | VM_BUG_ON(pfn_valid(page_to_pfn(start_page)) && |
2021 | pfn_valid(page_to_pfn(end_page)) && | |
2022 | page_zone(start_page) != page_zone(end_page)); | |
c361be55 MG |
2023 | #endif |
2024 | ||
02aa0cdd VB |
2025 | if (num_movable) |
2026 | *num_movable = 0; | |
2027 | ||
c361be55 MG |
2028 | for (page = start_page; page <= end_page;) { |
2029 | if (!pfn_valid_within(page_to_pfn(page))) { | |
2030 | page++; | |
2031 | continue; | |
2032 | } | |
2033 | ||
f073bdc5 AB |
2034 | /* Make sure we are not inadvertently changing nodes */ |
2035 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2036 | ||
c361be55 | 2037 | if (!PageBuddy(page)) { |
02aa0cdd VB |
2038 | /* |
2039 | * We assume that pages that could be isolated for | |
2040 | * migration are movable. But we don't actually try | |
2041 | * isolating, as that would be expensive. | |
2042 | */ | |
2043 | if (num_movable && | |
2044 | (PageLRU(page) || __PageMovable(page))) | |
2045 | (*num_movable)++; | |
2046 | ||
c361be55 MG |
2047 | page++; |
2048 | continue; | |
2049 | } | |
2050 | ||
2051 | order = page_order(page); | |
84be48d8 KS |
2052 | list_move(&page->lru, |
2053 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 2054 | page += 1 << order; |
d100313f | 2055 | pages_moved += 1 << order; |
c361be55 MG |
2056 | } |
2057 | ||
d100313f | 2058 | return pages_moved; |
c361be55 MG |
2059 | } |
2060 | ||
ee6f509c | 2061 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2062 | int migratetype, int *num_movable) |
c361be55 MG |
2063 | { |
2064 | unsigned long start_pfn, end_pfn; | |
2065 | struct page *start_page, *end_page; | |
2066 | ||
2067 | start_pfn = page_to_pfn(page); | |
d9c23400 | 2068 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 2069 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
2070 | end_page = start_page + pageblock_nr_pages - 1; |
2071 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
2072 | |
2073 | /* Do not cross zone boundaries */ | |
108bcc96 | 2074 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 2075 | start_page = page; |
108bcc96 | 2076 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2077 | return 0; |
2078 | ||
02aa0cdd VB |
2079 | return move_freepages(zone, start_page, end_page, migratetype, |
2080 | num_movable); | |
c361be55 MG |
2081 | } |
2082 | ||
2f66a68f MG |
2083 | static void change_pageblock_range(struct page *pageblock_page, |
2084 | int start_order, int migratetype) | |
2085 | { | |
2086 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2087 | ||
2088 | while (nr_pageblocks--) { | |
2089 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2090 | pageblock_page += pageblock_nr_pages; | |
2091 | } | |
2092 | } | |
2093 | ||
fef903ef | 2094 | /* |
9c0415eb VB |
2095 | * When we are falling back to another migratetype during allocation, try to |
2096 | * steal extra free pages from the same pageblocks to satisfy further | |
2097 | * allocations, instead of polluting multiple pageblocks. | |
2098 | * | |
2099 | * If we are stealing a relatively large buddy page, it is likely there will | |
2100 | * be more free pages in the pageblock, so try to steal them all. For | |
2101 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2102 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2103 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2104 | * pageblocks. | |
fef903ef | 2105 | */ |
4eb7dce6 JK |
2106 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2107 | { | |
2108 | /* | |
2109 | * Leaving this order check is intended, although there is | |
2110 | * relaxed order check in next check. The reason is that | |
2111 | * we can actually steal whole pageblock if this condition met, | |
2112 | * but, below check doesn't guarantee it and that is just heuristic | |
2113 | * so could be changed anytime. | |
2114 | */ | |
2115 | if (order >= pageblock_order) | |
2116 | return true; | |
2117 | ||
2118 | if (order >= pageblock_order / 2 || | |
2119 | start_mt == MIGRATE_RECLAIMABLE || | |
2120 | start_mt == MIGRATE_UNMOVABLE || | |
2121 | page_group_by_mobility_disabled) | |
2122 | return true; | |
2123 | ||
2124 | return false; | |
2125 | } | |
2126 | ||
2127 | /* | |
2128 | * This function implements actual steal behaviour. If order is large enough, | |
2129 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2130 | * pageblock to our migratetype and determine how many already-allocated pages |
2131 | * are there in the pageblock with a compatible migratetype. If at least half | |
2132 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2133 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2134 | */ |
2135 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
3bc48f96 | 2136 | int start_type, bool whole_block) |
fef903ef | 2137 | { |
d00181b9 | 2138 | unsigned int current_order = page_order(page); |
3bc48f96 | 2139 | struct free_area *area; |
02aa0cdd VB |
2140 | int free_pages, movable_pages, alike_pages; |
2141 | int old_block_type; | |
2142 | ||
2143 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2144 | |
3bc48f96 VB |
2145 | /* |
2146 | * This can happen due to races and we want to prevent broken | |
2147 | * highatomic accounting. | |
2148 | */ | |
02aa0cdd | 2149 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2150 | goto single_page; |
2151 | ||
fef903ef SB |
2152 | /* Take ownership for orders >= pageblock_order */ |
2153 | if (current_order >= pageblock_order) { | |
2154 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2155 | goto single_page; |
fef903ef SB |
2156 | } |
2157 | ||
3bc48f96 VB |
2158 | /* We are not allowed to try stealing from the whole block */ |
2159 | if (!whole_block) | |
2160 | goto single_page; | |
2161 | ||
02aa0cdd VB |
2162 | free_pages = move_freepages_block(zone, page, start_type, |
2163 | &movable_pages); | |
2164 | /* | |
2165 | * Determine how many pages are compatible with our allocation. | |
2166 | * For movable allocation, it's the number of movable pages which | |
2167 | * we just obtained. For other types it's a bit more tricky. | |
2168 | */ | |
2169 | if (start_type == MIGRATE_MOVABLE) { | |
2170 | alike_pages = movable_pages; | |
2171 | } else { | |
2172 | /* | |
2173 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2174 | * to MOVABLE pageblock, consider all non-movable pages as | |
2175 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2176 | * vice versa, be conservative since we can't distinguish the | |
2177 | * exact migratetype of non-movable pages. | |
2178 | */ | |
2179 | if (old_block_type == MIGRATE_MOVABLE) | |
2180 | alike_pages = pageblock_nr_pages | |
2181 | - (free_pages + movable_pages); | |
2182 | else | |
2183 | alike_pages = 0; | |
2184 | } | |
2185 | ||
3bc48f96 | 2186 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2187 | if (!free_pages) |
3bc48f96 | 2188 | goto single_page; |
fef903ef | 2189 | |
02aa0cdd VB |
2190 | /* |
2191 | * If a sufficient number of pages in the block are either free or of | |
2192 | * comparable migratability as our allocation, claim the whole block. | |
2193 | */ | |
2194 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2195 | page_group_by_mobility_disabled) |
2196 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2197 | |
2198 | return; | |
2199 | ||
2200 | single_page: | |
2201 | area = &zone->free_area[current_order]; | |
2202 | list_move(&page->lru, &area->free_list[start_type]); | |
4eb7dce6 JK |
2203 | } |
2204 | ||
2149cdae JK |
2205 | /* |
2206 | * Check whether there is a suitable fallback freepage with requested order. | |
2207 | * If only_stealable is true, this function returns fallback_mt only if | |
2208 | * we can steal other freepages all together. This would help to reduce | |
2209 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2210 | */ | |
2211 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2212 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2213 | { |
2214 | int i; | |
2215 | int fallback_mt; | |
2216 | ||
2217 | if (area->nr_free == 0) | |
2218 | return -1; | |
2219 | ||
2220 | *can_steal = false; | |
2221 | for (i = 0;; i++) { | |
2222 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2223 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2224 | break; |
2225 | ||
2226 | if (list_empty(&area->free_list[fallback_mt])) | |
2227 | continue; | |
fef903ef | 2228 | |
4eb7dce6 JK |
2229 | if (can_steal_fallback(order, migratetype)) |
2230 | *can_steal = true; | |
2231 | ||
2149cdae JK |
2232 | if (!only_stealable) |
2233 | return fallback_mt; | |
2234 | ||
2235 | if (*can_steal) | |
2236 | return fallback_mt; | |
fef903ef | 2237 | } |
4eb7dce6 JK |
2238 | |
2239 | return -1; | |
fef903ef SB |
2240 | } |
2241 | ||
0aaa29a5 MG |
2242 | /* |
2243 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2244 | * there are no empty page blocks that contain a page with a suitable order | |
2245 | */ | |
2246 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2247 | unsigned int alloc_order) | |
2248 | { | |
2249 | int mt; | |
2250 | unsigned long max_managed, flags; | |
2251 | ||
2252 | /* | |
2253 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2254 | * Check is race-prone but harmless. | |
2255 | */ | |
2256 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
2257 | if (zone->nr_reserved_highatomic >= max_managed) | |
2258 | return; | |
2259 | ||
2260 | spin_lock_irqsave(&zone->lock, flags); | |
2261 | ||
2262 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2263 | if (zone->nr_reserved_highatomic >= max_managed) | |
2264 | goto out_unlock; | |
2265 | ||
2266 | /* Yoink! */ | |
2267 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2268 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2269 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2270 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2271 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2272 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2273 | } |
2274 | ||
2275 | out_unlock: | |
2276 | spin_unlock_irqrestore(&zone->lock, flags); | |
2277 | } | |
2278 | ||
2279 | /* | |
2280 | * Used when an allocation is about to fail under memory pressure. This | |
2281 | * potentially hurts the reliability of high-order allocations when under | |
2282 | * intense memory pressure but failed atomic allocations should be easier | |
2283 | * to recover from than an OOM. | |
29fac03b MK |
2284 | * |
2285 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2286 | * pageblock is exhausted. | |
0aaa29a5 | 2287 | */ |
29fac03b MK |
2288 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2289 | bool force) | |
0aaa29a5 MG |
2290 | { |
2291 | struct zonelist *zonelist = ac->zonelist; | |
2292 | unsigned long flags; | |
2293 | struct zoneref *z; | |
2294 | struct zone *zone; | |
2295 | struct page *page; | |
2296 | int order; | |
04c8716f | 2297 | bool ret; |
0aaa29a5 MG |
2298 | |
2299 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2300 | ac->nodemask) { | |
29fac03b MK |
2301 | /* |
2302 | * Preserve at least one pageblock unless memory pressure | |
2303 | * is really high. | |
2304 | */ | |
2305 | if (!force && zone->nr_reserved_highatomic <= | |
2306 | pageblock_nr_pages) | |
0aaa29a5 MG |
2307 | continue; |
2308 | ||
2309 | spin_lock_irqsave(&zone->lock, flags); | |
2310 | for (order = 0; order < MAX_ORDER; order++) { | |
2311 | struct free_area *area = &(zone->free_area[order]); | |
2312 | ||
a16601c5 GT |
2313 | page = list_first_entry_or_null( |
2314 | &area->free_list[MIGRATE_HIGHATOMIC], | |
2315 | struct page, lru); | |
2316 | if (!page) | |
0aaa29a5 MG |
2317 | continue; |
2318 | ||
0aaa29a5 | 2319 | /* |
4855e4a7 MK |
2320 | * In page freeing path, migratetype change is racy so |
2321 | * we can counter several free pages in a pageblock | |
2322 | * in this loop althoug we changed the pageblock type | |
2323 | * from highatomic to ac->migratetype. So we should | |
2324 | * adjust the count once. | |
0aaa29a5 | 2325 | */ |
a6ffdc07 | 2326 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2327 | /* |
2328 | * It should never happen but changes to | |
2329 | * locking could inadvertently allow a per-cpu | |
2330 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2331 | * while unreserving so be safe and watch for | |
2332 | * underflows. | |
2333 | */ | |
2334 | zone->nr_reserved_highatomic -= min( | |
2335 | pageblock_nr_pages, | |
2336 | zone->nr_reserved_highatomic); | |
2337 | } | |
0aaa29a5 MG |
2338 | |
2339 | /* | |
2340 | * Convert to ac->migratetype and avoid the normal | |
2341 | * pageblock stealing heuristics. Minimally, the caller | |
2342 | * is doing the work and needs the pages. More | |
2343 | * importantly, if the block was always converted to | |
2344 | * MIGRATE_UNMOVABLE or another type then the number | |
2345 | * of pageblocks that cannot be completely freed | |
2346 | * may increase. | |
2347 | */ | |
2348 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2349 | ret = move_freepages_block(zone, page, ac->migratetype, |
2350 | NULL); | |
29fac03b MK |
2351 | if (ret) { |
2352 | spin_unlock_irqrestore(&zone->lock, flags); | |
2353 | return ret; | |
2354 | } | |
0aaa29a5 MG |
2355 | } |
2356 | spin_unlock_irqrestore(&zone->lock, flags); | |
2357 | } | |
04c8716f MK |
2358 | |
2359 | return false; | |
0aaa29a5 MG |
2360 | } |
2361 | ||
3bc48f96 VB |
2362 | /* |
2363 | * Try finding a free buddy page on the fallback list and put it on the free | |
2364 | * list of requested migratetype, possibly along with other pages from the same | |
2365 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2366 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2367 | * |
2368 | * The use of signed ints for order and current_order is a deliberate | |
2369 | * deviation from the rest of this file, to make the for loop | |
2370 | * condition simpler. | |
3bc48f96 | 2371 | */ |
85ccc8fa | 2372 | static __always_inline bool |
b002529d | 2373 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) |
b2a0ac88 | 2374 | { |
b8af2941 | 2375 | struct free_area *area; |
b002529d | 2376 | int current_order; |
b2a0ac88 | 2377 | struct page *page; |
4eb7dce6 JK |
2378 | int fallback_mt; |
2379 | bool can_steal; | |
b2a0ac88 | 2380 | |
7a8f58f3 VB |
2381 | /* |
2382 | * Find the largest available free page in the other list. This roughly | |
2383 | * approximates finding the pageblock with the most free pages, which | |
2384 | * would be too costly to do exactly. | |
2385 | */ | |
b002529d | 2386 | for (current_order = MAX_ORDER - 1; current_order >= order; |
7aeb09f9 | 2387 | --current_order) { |
4eb7dce6 JK |
2388 | area = &(zone->free_area[current_order]); |
2389 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2390 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2391 | if (fallback_mt == -1) |
2392 | continue; | |
b2a0ac88 | 2393 | |
7a8f58f3 VB |
2394 | /* |
2395 | * We cannot steal all free pages from the pageblock and the | |
2396 | * requested migratetype is movable. In that case it's better to | |
2397 | * steal and split the smallest available page instead of the | |
2398 | * largest available page, because even if the next movable | |
2399 | * allocation falls back into a different pageblock than this | |
2400 | * one, it won't cause permanent fragmentation. | |
2401 | */ | |
2402 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2403 | && current_order > order) | |
2404 | goto find_smallest; | |
b2a0ac88 | 2405 | |
7a8f58f3 VB |
2406 | goto do_steal; |
2407 | } | |
e0fff1bd | 2408 | |
7a8f58f3 | 2409 | return false; |
e0fff1bd | 2410 | |
7a8f58f3 VB |
2411 | find_smallest: |
2412 | for (current_order = order; current_order < MAX_ORDER; | |
2413 | current_order++) { | |
2414 | area = &(zone->free_area[current_order]); | |
2415 | fallback_mt = find_suitable_fallback(area, current_order, | |
2416 | start_migratetype, false, &can_steal); | |
2417 | if (fallback_mt != -1) | |
2418 | break; | |
b2a0ac88 MG |
2419 | } |
2420 | ||
7a8f58f3 VB |
2421 | /* |
2422 | * This should not happen - we already found a suitable fallback | |
2423 | * when looking for the largest page. | |
2424 | */ | |
2425 | VM_BUG_ON(current_order == MAX_ORDER); | |
2426 | ||
2427 | do_steal: | |
2428 | page = list_first_entry(&area->free_list[fallback_mt], | |
2429 | struct page, lru); | |
2430 | ||
2431 | steal_suitable_fallback(zone, page, start_migratetype, can_steal); | |
2432 | ||
2433 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2434 | start_migratetype, fallback_mt); | |
2435 | ||
2436 | return true; | |
2437 | ||
b2a0ac88 MG |
2438 | } |
2439 | ||
56fd56b8 | 2440 | /* |
1da177e4 LT |
2441 | * Do the hard work of removing an element from the buddy allocator. |
2442 | * Call me with the zone->lock already held. | |
2443 | */ | |
85ccc8fa AL |
2444 | static __always_inline struct page * |
2445 | __rmqueue(struct zone *zone, unsigned int order, int migratetype) | |
1da177e4 | 2446 | { |
1da177e4 LT |
2447 | struct page *page; |
2448 | ||
3bc48f96 | 2449 | retry: |
56fd56b8 | 2450 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2451 | if (unlikely(!page)) { |
dc67647b JK |
2452 | if (migratetype == MIGRATE_MOVABLE) |
2453 | page = __rmqueue_cma_fallback(zone, order); | |
2454 | ||
3bc48f96 VB |
2455 | if (!page && __rmqueue_fallback(zone, order, migratetype)) |
2456 | goto retry; | |
728ec980 MG |
2457 | } |
2458 | ||
0d3d062a | 2459 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2460 | return page; |
1da177e4 LT |
2461 | } |
2462 | ||
5f63b720 | 2463 | /* |
1da177e4 LT |
2464 | * Obtain a specified number of elements from the buddy allocator, all under |
2465 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2466 | * Returns the number of new pages which were placed at *list. | |
2467 | */ | |
5f63b720 | 2468 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2469 | unsigned long count, struct list_head *list, |
453f85d4 | 2470 | int migratetype) |
1da177e4 | 2471 | { |
a6de734b | 2472 | int i, alloced = 0; |
5f63b720 | 2473 | |
d34b0733 | 2474 | spin_lock(&zone->lock); |
1da177e4 | 2475 | for (i = 0; i < count; ++i) { |
6ac0206b | 2476 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 2477 | if (unlikely(page == NULL)) |
1da177e4 | 2478 | break; |
81eabcbe | 2479 | |
479f854a MG |
2480 | if (unlikely(check_pcp_refill(page))) |
2481 | continue; | |
2482 | ||
81eabcbe | 2483 | /* |
0fac3ba5 VB |
2484 | * Split buddy pages returned by expand() are received here in |
2485 | * physical page order. The page is added to the tail of | |
2486 | * caller's list. From the callers perspective, the linked list | |
2487 | * is ordered by page number under some conditions. This is | |
2488 | * useful for IO devices that can forward direction from the | |
2489 | * head, thus also in the physical page order. This is useful | |
2490 | * for IO devices that can merge IO requests if the physical | |
2491 | * pages are ordered properly. | |
81eabcbe | 2492 | */ |
0fac3ba5 | 2493 | list_add_tail(&page->lru, list); |
a6de734b | 2494 | alloced++; |
bb14c2c7 | 2495 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2496 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2497 | -(1 << order)); | |
1da177e4 | 2498 | } |
a6de734b MG |
2499 | |
2500 | /* | |
2501 | * i pages were removed from the buddy list even if some leak due | |
2502 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2503 | * on i. Do not confuse with 'alloced' which is the number of | |
2504 | * pages added to the pcp list. | |
2505 | */ | |
f2260e6b | 2506 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2507 | spin_unlock(&zone->lock); |
a6de734b | 2508 | return alloced; |
1da177e4 LT |
2509 | } |
2510 | ||
4ae7c039 | 2511 | #ifdef CONFIG_NUMA |
8fce4d8e | 2512 | /* |
4037d452 CL |
2513 | * Called from the vmstat counter updater to drain pagesets of this |
2514 | * currently executing processor on remote nodes after they have | |
2515 | * expired. | |
2516 | * | |
879336c3 CL |
2517 | * Note that this function must be called with the thread pinned to |
2518 | * a single processor. | |
8fce4d8e | 2519 | */ |
4037d452 | 2520 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2521 | { |
4ae7c039 | 2522 | unsigned long flags; |
7be12fc9 | 2523 | int to_drain, batch; |
4ae7c039 | 2524 | |
4037d452 | 2525 | local_irq_save(flags); |
4db0c3c2 | 2526 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2527 | to_drain = min(pcp->count, batch); |
77ba9062 | 2528 | if (to_drain > 0) |
2a13515c | 2529 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 2530 | local_irq_restore(flags); |
4ae7c039 CL |
2531 | } |
2532 | #endif | |
2533 | ||
9f8f2172 | 2534 | /* |
93481ff0 | 2535 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2536 | * |
2537 | * The processor must either be the current processor and the | |
2538 | * thread pinned to the current processor or a processor that | |
2539 | * is not online. | |
2540 | */ | |
93481ff0 | 2541 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2542 | { |
c54ad30c | 2543 | unsigned long flags; |
93481ff0 VB |
2544 | struct per_cpu_pageset *pset; |
2545 | struct per_cpu_pages *pcp; | |
1da177e4 | 2546 | |
93481ff0 VB |
2547 | local_irq_save(flags); |
2548 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2549 | |
93481ff0 | 2550 | pcp = &pset->pcp; |
77ba9062 | 2551 | if (pcp->count) |
93481ff0 | 2552 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
2553 | local_irq_restore(flags); |
2554 | } | |
3dfa5721 | 2555 | |
93481ff0 VB |
2556 | /* |
2557 | * Drain pcplists of all zones on the indicated processor. | |
2558 | * | |
2559 | * The processor must either be the current processor and the | |
2560 | * thread pinned to the current processor or a processor that | |
2561 | * is not online. | |
2562 | */ | |
2563 | static void drain_pages(unsigned int cpu) | |
2564 | { | |
2565 | struct zone *zone; | |
2566 | ||
2567 | for_each_populated_zone(zone) { | |
2568 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2569 | } |
2570 | } | |
1da177e4 | 2571 | |
9f8f2172 CL |
2572 | /* |
2573 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2574 | * |
2575 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2576 | * the single zone's pages. | |
9f8f2172 | 2577 | */ |
93481ff0 | 2578 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2579 | { |
93481ff0 VB |
2580 | int cpu = smp_processor_id(); |
2581 | ||
2582 | if (zone) | |
2583 | drain_pages_zone(cpu, zone); | |
2584 | else | |
2585 | drain_pages(cpu); | |
9f8f2172 CL |
2586 | } |
2587 | ||
0ccce3b9 MG |
2588 | static void drain_local_pages_wq(struct work_struct *work) |
2589 | { | |
a459eeb7 MH |
2590 | /* |
2591 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2592 | * we can race with cpu offline when the WQ can move this from | |
2593 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2594 | * cpu which is allright but we also have to make sure to not move to | |
2595 | * a different one. | |
2596 | */ | |
2597 | preempt_disable(); | |
0ccce3b9 | 2598 | drain_local_pages(NULL); |
a459eeb7 | 2599 | preempt_enable(); |
0ccce3b9 MG |
2600 | } |
2601 | ||
9f8f2172 | 2602 | /* |
74046494 GBY |
2603 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2604 | * | |
93481ff0 VB |
2605 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2606 | * | |
0ccce3b9 | 2607 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2608 | */ |
93481ff0 | 2609 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2610 | { |
74046494 | 2611 | int cpu; |
74046494 GBY |
2612 | |
2613 | /* | |
2614 | * Allocate in the BSS so we wont require allocation in | |
2615 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2616 | */ | |
2617 | static cpumask_t cpus_with_pcps; | |
2618 | ||
ce612879 MH |
2619 | /* |
2620 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2621 | * initialized. | |
2622 | */ | |
2623 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2624 | return; | |
2625 | ||
bd233f53 MG |
2626 | /* |
2627 | * Do not drain if one is already in progress unless it's specific to | |
2628 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2629 | * the drain to be complete when the call returns. | |
2630 | */ | |
2631 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2632 | if (!zone) | |
2633 | return; | |
2634 | mutex_lock(&pcpu_drain_mutex); | |
2635 | } | |
0ccce3b9 | 2636 | |
74046494 GBY |
2637 | /* |
2638 | * We don't care about racing with CPU hotplug event | |
2639 | * as offline notification will cause the notified | |
2640 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2641 | * disables preemption as part of its processing | |
2642 | */ | |
2643 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2644 | struct per_cpu_pageset *pcp; |
2645 | struct zone *z; | |
74046494 | 2646 | bool has_pcps = false; |
93481ff0 VB |
2647 | |
2648 | if (zone) { | |
74046494 | 2649 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2650 | if (pcp->pcp.count) |
74046494 | 2651 | has_pcps = true; |
93481ff0 VB |
2652 | } else { |
2653 | for_each_populated_zone(z) { | |
2654 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2655 | if (pcp->pcp.count) { | |
2656 | has_pcps = true; | |
2657 | break; | |
2658 | } | |
74046494 GBY |
2659 | } |
2660 | } | |
93481ff0 | 2661 | |
74046494 GBY |
2662 | if (has_pcps) |
2663 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2664 | else | |
2665 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2666 | } | |
0ccce3b9 | 2667 | |
bd233f53 MG |
2668 | for_each_cpu(cpu, &cpus_with_pcps) { |
2669 | struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu); | |
2670 | INIT_WORK(work, drain_local_pages_wq); | |
ce612879 | 2671 | queue_work_on(cpu, mm_percpu_wq, work); |
0ccce3b9 | 2672 | } |
bd233f53 MG |
2673 | for_each_cpu(cpu, &cpus_with_pcps) |
2674 | flush_work(per_cpu_ptr(&pcpu_drain, cpu)); | |
2675 | ||
2676 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2677 | } |
2678 | ||
296699de | 2679 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 2680 | |
556b969a CY |
2681 | /* |
2682 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
2683 | */ | |
2684 | #define WD_PAGE_COUNT (128*1024) | |
2685 | ||
1da177e4 LT |
2686 | void mark_free_pages(struct zone *zone) |
2687 | { | |
556b969a | 2688 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 2689 | unsigned long flags; |
7aeb09f9 | 2690 | unsigned int order, t; |
86760a2c | 2691 | struct page *page; |
1da177e4 | 2692 | |
8080fc03 | 2693 | if (zone_is_empty(zone)) |
1da177e4 LT |
2694 | return; |
2695 | ||
2696 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2697 | |
108bcc96 | 2698 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2699 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2700 | if (pfn_valid(pfn)) { | |
86760a2c | 2701 | page = pfn_to_page(pfn); |
ba6b0979 | 2702 | |
556b969a CY |
2703 | if (!--page_count) { |
2704 | touch_nmi_watchdog(); | |
2705 | page_count = WD_PAGE_COUNT; | |
2706 | } | |
2707 | ||
ba6b0979 JK |
2708 | if (page_zone(page) != zone) |
2709 | continue; | |
2710 | ||
7be98234 RW |
2711 | if (!swsusp_page_is_forbidden(page)) |
2712 | swsusp_unset_page_free(page); | |
f623f0db | 2713 | } |
1da177e4 | 2714 | |
b2a0ac88 | 2715 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2716 | list_for_each_entry(page, |
2717 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2718 | unsigned long i; |
1da177e4 | 2719 | |
86760a2c | 2720 | pfn = page_to_pfn(page); |
556b969a CY |
2721 | for (i = 0; i < (1UL << order); i++) { |
2722 | if (!--page_count) { | |
2723 | touch_nmi_watchdog(); | |
2724 | page_count = WD_PAGE_COUNT; | |
2725 | } | |
7be98234 | 2726 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 2727 | } |
f623f0db | 2728 | } |
b2a0ac88 | 2729 | } |
1da177e4 LT |
2730 | spin_unlock_irqrestore(&zone->lock, flags); |
2731 | } | |
e2c55dc8 | 2732 | #endif /* CONFIG_PM */ |
1da177e4 | 2733 | |
2d4894b5 | 2734 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 2735 | { |
5f8dcc21 | 2736 | int migratetype; |
1da177e4 | 2737 | |
4db7548c | 2738 | if (!free_pcp_prepare(page)) |
9cca35d4 | 2739 | return false; |
689bcebf | 2740 | |
dc4b0caf | 2741 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2742 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
2743 | return true; |
2744 | } | |
2745 | ||
2d4894b5 | 2746 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
2747 | { |
2748 | struct zone *zone = page_zone(page); | |
2749 | struct per_cpu_pages *pcp; | |
2750 | int migratetype; | |
2751 | ||
2752 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 2753 | __count_vm_event(PGFREE); |
da456f14 | 2754 | |
5f8dcc21 MG |
2755 | /* |
2756 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2757 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 2758 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
2759 | * areas back if necessary. Otherwise, we may have to free |
2760 | * excessively into the page allocator | |
2761 | */ | |
2762 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2763 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2764 | free_one_page(zone, page, pfn, 0, migratetype); |
9cca35d4 | 2765 | return; |
5f8dcc21 MG |
2766 | } |
2767 | migratetype = MIGRATE_MOVABLE; | |
2768 | } | |
2769 | ||
99dcc3e5 | 2770 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 2771 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 2772 | pcp->count++; |
48db57f8 | 2773 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2774 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb | 2775 | free_pcppages_bulk(zone, batch, pcp); |
48db57f8 | 2776 | } |
9cca35d4 | 2777 | } |
5f8dcc21 | 2778 | |
9cca35d4 MG |
2779 | /* |
2780 | * Free a 0-order page | |
9cca35d4 | 2781 | */ |
2d4894b5 | 2782 | void free_unref_page(struct page *page) |
9cca35d4 MG |
2783 | { |
2784 | unsigned long flags; | |
2785 | unsigned long pfn = page_to_pfn(page); | |
2786 | ||
2d4894b5 | 2787 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
2788 | return; |
2789 | ||
2790 | local_irq_save(flags); | |
2d4894b5 | 2791 | free_unref_page_commit(page, pfn); |
d34b0733 | 2792 | local_irq_restore(flags); |
1da177e4 LT |
2793 | } |
2794 | ||
cc59850e KK |
2795 | /* |
2796 | * Free a list of 0-order pages | |
2797 | */ | |
2d4894b5 | 2798 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
2799 | { |
2800 | struct page *page, *next; | |
9cca35d4 | 2801 | unsigned long flags, pfn; |
c24ad77d | 2802 | int batch_count = 0; |
9cca35d4 MG |
2803 | |
2804 | /* Prepare pages for freeing */ | |
2805 | list_for_each_entry_safe(page, next, list, lru) { | |
2806 | pfn = page_to_pfn(page); | |
2d4894b5 | 2807 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
2808 | list_del(&page->lru); |
2809 | set_page_private(page, pfn); | |
2810 | } | |
cc59850e | 2811 | |
9cca35d4 | 2812 | local_irq_save(flags); |
cc59850e | 2813 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
2814 | unsigned long pfn = page_private(page); |
2815 | ||
2816 | set_page_private(page, 0); | |
2d4894b5 MG |
2817 | trace_mm_page_free_batched(page); |
2818 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
2819 | |
2820 | /* | |
2821 | * Guard against excessive IRQ disabled times when we get | |
2822 | * a large list of pages to free. | |
2823 | */ | |
2824 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
2825 | local_irq_restore(flags); | |
2826 | batch_count = 0; | |
2827 | local_irq_save(flags); | |
2828 | } | |
cc59850e | 2829 | } |
9cca35d4 | 2830 | local_irq_restore(flags); |
cc59850e KK |
2831 | } |
2832 | ||
8dfcc9ba NP |
2833 | /* |
2834 | * split_page takes a non-compound higher-order page, and splits it into | |
2835 | * n (1<<order) sub-pages: page[0..n] | |
2836 | * Each sub-page must be freed individually. | |
2837 | * | |
2838 | * Note: this is probably too low level an operation for use in drivers. | |
2839 | * Please consult with lkml before using this in your driver. | |
2840 | */ | |
2841 | void split_page(struct page *page, unsigned int order) | |
2842 | { | |
2843 | int i; | |
2844 | ||
309381fe SL |
2845 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2846 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 2847 | |
a9627bc5 | 2848 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2849 | set_page_refcounted(page + i); |
a9627bc5 | 2850 | split_page_owner(page, order); |
8dfcc9ba | 2851 | } |
5853ff23 | 2852 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2853 | |
3c605096 | 2854 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2855 | { |
748446bb MG |
2856 | unsigned long watermark; |
2857 | struct zone *zone; | |
2139cbe6 | 2858 | int mt; |
748446bb MG |
2859 | |
2860 | BUG_ON(!PageBuddy(page)); | |
2861 | ||
2862 | zone = page_zone(page); | |
2e30abd1 | 2863 | mt = get_pageblock_migratetype(page); |
748446bb | 2864 | |
194159fb | 2865 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
2866 | /* |
2867 | * Obey watermarks as if the page was being allocated. We can | |
2868 | * emulate a high-order watermark check with a raised order-0 | |
2869 | * watermark, because we already know our high-order page | |
2870 | * exists. | |
2871 | */ | |
2872 | watermark = min_wmark_pages(zone) + (1UL << order); | |
d883c6cf | 2873 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
2874 | return 0; |
2875 | ||
8fb74b9f | 2876 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2877 | } |
748446bb MG |
2878 | |
2879 | /* Remove page from free list */ | |
2880 | list_del(&page->lru); | |
2881 | zone->free_area[order].nr_free--; | |
2882 | rmv_page_order(page); | |
2139cbe6 | 2883 | |
400bc7fd | 2884 | /* |
2885 | * Set the pageblock if the isolated page is at least half of a | |
2886 | * pageblock | |
2887 | */ | |
748446bb MG |
2888 | if (order >= pageblock_order - 1) { |
2889 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2890 | for (; page < endpage; page += pageblock_nr_pages) { |
2891 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 2892 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 2893 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
2894 | set_pageblock_migratetype(page, |
2895 | MIGRATE_MOVABLE); | |
2896 | } | |
748446bb MG |
2897 | } |
2898 | ||
f3a14ced | 2899 | |
8fb74b9f | 2900 | return 1UL << order; |
1fb3f8ca MG |
2901 | } |
2902 | ||
060e7417 MG |
2903 | /* |
2904 | * Update NUMA hit/miss statistics | |
2905 | * | |
2906 | * Must be called with interrupts disabled. | |
060e7417 | 2907 | */ |
41b6167e | 2908 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
2909 | { |
2910 | #ifdef CONFIG_NUMA | |
3a321d2a | 2911 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 2912 | |
4518085e KW |
2913 | /* skip numa counters update if numa stats is disabled */ |
2914 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2915 | return; | |
2916 | ||
c1093b74 | 2917 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 2918 | local_stat = NUMA_OTHER; |
060e7417 | 2919 | |
c1093b74 | 2920 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3a321d2a | 2921 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 2922 | else { |
3a321d2a KW |
2923 | __inc_numa_state(z, NUMA_MISS); |
2924 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 2925 | } |
3a321d2a | 2926 | __inc_numa_state(z, local_stat); |
060e7417 MG |
2927 | #endif |
2928 | } | |
2929 | ||
066b2393 MG |
2930 | /* Remove page from the per-cpu list, caller must protect the list */ |
2931 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
453f85d4 | 2932 | struct per_cpu_pages *pcp, |
066b2393 MG |
2933 | struct list_head *list) |
2934 | { | |
2935 | struct page *page; | |
2936 | ||
2937 | do { | |
2938 | if (list_empty(list)) { | |
2939 | pcp->count += rmqueue_bulk(zone, 0, | |
2940 | pcp->batch, list, | |
453f85d4 | 2941 | migratetype); |
066b2393 MG |
2942 | if (unlikely(list_empty(list))) |
2943 | return NULL; | |
2944 | } | |
2945 | ||
453f85d4 | 2946 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
2947 | list_del(&page->lru); |
2948 | pcp->count--; | |
2949 | } while (check_new_pcp(page)); | |
2950 | ||
2951 | return page; | |
2952 | } | |
2953 | ||
2954 | /* Lock and remove page from the per-cpu list */ | |
2955 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
2956 | struct zone *zone, unsigned int order, | |
2957 | gfp_t gfp_flags, int migratetype) | |
2958 | { | |
2959 | struct per_cpu_pages *pcp; | |
2960 | struct list_head *list; | |
066b2393 | 2961 | struct page *page; |
d34b0733 | 2962 | unsigned long flags; |
066b2393 | 2963 | |
d34b0733 | 2964 | local_irq_save(flags); |
066b2393 MG |
2965 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2966 | list = &pcp->lists[migratetype]; | |
453f85d4 | 2967 | page = __rmqueue_pcplist(zone, migratetype, pcp, list); |
066b2393 MG |
2968 | if (page) { |
2969 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
2970 | zone_statistics(preferred_zone, zone); | |
2971 | } | |
d34b0733 | 2972 | local_irq_restore(flags); |
066b2393 MG |
2973 | return page; |
2974 | } | |
2975 | ||
1da177e4 | 2976 | /* |
75379191 | 2977 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2978 | */ |
0a15c3e9 | 2979 | static inline |
066b2393 | 2980 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 2981 | struct zone *zone, unsigned int order, |
c603844b MG |
2982 | gfp_t gfp_flags, unsigned int alloc_flags, |
2983 | int migratetype) | |
1da177e4 LT |
2984 | { |
2985 | unsigned long flags; | |
689bcebf | 2986 | struct page *page; |
1da177e4 | 2987 | |
d34b0733 | 2988 | if (likely(order == 0)) { |
066b2393 MG |
2989 | page = rmqueue_pcplist(preferred_zone, zone, order, |
2990 | gfp_flags, migratetype); | |
2991 | goto out; | |
2992 | } | |
83b9355b | 2993 | |
066b2393 MG |
2994 | /* |
2995 | * We most definitely don't want callers attempting to | |
2996 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2997 | */ | |
2998 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
2999 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3000 | |
066b2393 MG |
3001 | do { |
3002 | page = NULL; | |
3003 | if (alloc_flags & ALLOC_HARDER) { | |
3004 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3005 | if (page) | |
3006 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3007 | } | |
a74609fa | 3008 | if (!page) |
066b2393 MG |
3009 | page = __rmqueue(zone, order, migratetype); |
3010 | } while (page && check_new_pages(page, order)); | |
3011 | spin_unlock(&zone->lock); | |
3012 | if (!page) | |
3013 | goto failed; | |
3014 | __mod_zone_freepage_state(zone, -(1 << order), | |
3015 | get_pcppage_migratetype(page)); | |
1da177e4 | 3016 | |
16709d1d | 3017 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3018 | zone_statistics(preferred_zone, zone); |
a74609fa | 3019 | local_irq_restore(flags); |
1da177e4 | 3020 | |
066b2393 MG |
3021 | out: |
3022 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); | |
1da177e4 | 3023 | return page; |
a74609fa NP |
3024 | |
3025 | failed: | |
3026 | local_irq_restore(flags); | |
a74609fa | 3027 | return NULL; |
1da177e4 LT |
3028 | } |
3029 | ||
933e312e AM |
3030 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3031 | ||
b2588c4b | 3032 | static struct { |
933e312e AM |
3033 | struct fault_attr attr; |
3034 | ||
621a5f7a | 3035 | bool ignore_gfp_highmem; |
71baba4b | 3036 | bool ignore_gfp_reclaim; |
54114994 | 3037 | u32 min_order; |
933e312e AM |
3038 | } fail_page_alloc = { |
3039 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3040 | .ignore_gfp_reclaim = true, |
621a5f7a | 3041 | .ignore_gfp_highmem = true, |
54114994 | 3042 | .min_order = 1, |
933e312e AM |
3043 | }; |
3044 | ||
3045 | static int __init setup_fail_page_alloc(char *str) | |
3046 | { | |
3047 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3048 | } | |
3049 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3050 | ||
deaf386e | 3051 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3052 | { |
54114994 | 3053 | if (order < fail_page_alloc.min_order) |
deaf386e | 3054 | return false; |
933e312e | 3055 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3056 | return false; |
933e312e | 3057 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3058 | return false; |
71baba4b MG |
3059 | if (fail_page_alloc.ignore_gfp_reclaim && |
3060 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3061 | return false; |
933e312e AM |
3062 | |
3063 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3064 | } | |
3065 | ||
3066 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3067 | ||
3068 | static int __init fail_page_alloc_debugfs(void) | |
3069 | { | |
0825a6f9 | 3070 | umode_t mode = S_IFREG | 0600; |
933e312e | 3071 | struct dentry *dir; |
933e312e | 3072 | |
dd48c085 AM |
3073 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3074 | &fail_page_alloc.attr); | |
3075 | if (IS_ERR(dir)) | |
3076 | return PTR_ERR(dir); | |
933e312e | 3077 | |
b2588c4b | 3078 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 3079 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
3080 | goto fail; |
3081 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3082 | &fail_page_alloc.ignore_gfp_highmem)) | |
3083 | goto fail; | |
3084 | if (!debugfs_create_u32("min-order", mode, dir, | |
3085 | &fail_page_alloc.min_order)) | |
3086 | goto fail; | |
3087 | ||
3088 | return 0; | |
3089 | fail: | |
dd48c085 | 3090 | debugfs_remove_recursive(dir); |
933e312e | 3091 | |
b2588c4b | 3092 | return -ENOMEM; |
933e312e AM |
3093 | } |
3094 | ||
3095 | late_initcall(fail_page_alloc_debugfs); | |
3096 | ||
3097 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3098 | ||
3099 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3100 | ||
deaf386e | 3101 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3102 | { |
deaf386e | 3103 | return false; |
933e312e AM |
3104 | } |
3105 | ||
3106 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3107 | ||
1da177e4 | 3108 | /* |
97a16fc8 MG |
3109 | * Return true if free base pages are above 'mark'. For high-order checks it |
3110 | * will return true of the order-0 watermark is reached and there is at least | |
3111 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3112 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3113 | */ |
86a294a8 MH |
3114 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
3115 | int classzone_idx, unsigned int alloc_flags, | |
3116 | long free_pages) | |
1da177e4 | 3117 | { |
d23ad423 | 3118 | long min = mark; |
1da177e4 | 3119 | int o; |
cd04ae1e | 3120 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3121 | |
0aaa29a5 | 3122 | /* free_pages may go negative - that's OK */ |
df0a6daa | 3123 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 3124 | |
7fb1d9fc | 3125 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3126 | min -= min / 2; |
0aaa29a5 MG |
3127 | |
3128 | /* | |
3129 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3130 | * the high-atomic reserves. This will over-estimate the size of the | |
3131 | * atomic reserve but it avoids a search. | |
3132 | */ | |
cd04ae1e | 3133 | if (likely(!alloc_harder)) { |
0aaa29a5 | 3134 | free_pages -= z->nr_reserved_highatomic; |
cd04ae1e MH |
3135 | } else { |
3136 | /* | |
3137 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3138 | * users on the grounds that it's definitely going to be in | |
3139 | * the exit path shortly and free memory. Any allocation it | |
3140 | * makes during the free path will be small and short-lived. | |
3141 | */ | |
3142 | if (alloc_flags & ALLOC_OOM) | |
3143 | min -= min / 2; | |
3144 | else | |
3145 | min -= min / 4; | |
3146 | } | |
3147 | ||
e2b19197 | 3148 | |
d883c6cf JK |
3149 | #ifdef CONFIG_CMA |
3150 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3151 | if (!(alloc_flags & ALLOC_CMA)) | |
3152 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); | |
3153 | #endif | |
3154 | ||
97a16fc8 MG |
3155 | /* |
3156 | * Check watermarks for an order-0 allocation request. If these | |
3157 | * are not met, then a high-order request also cannot go ahead | |
3158 | * even if a suitable page happened to be free. | |
3159 | */ | |
3160 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 3161 | return false; |
1da177e4 | 3162 | |
97a16fc8 MG |
3163 | /* If this is an order-0 request then the watermark is fine */ |
3164 | if (!order) | |
3165 | return true; | |
3166 | ||
3167 | /* For a high-order request, check at least one suitable page is free */ | |
3168 | for (o = order; o < MAX_ORDER; o++) { | |
3169 | struct free_area *area = &z->free_area[o]; | |
3170 | int mt; | |
3171 | ||
3172 | if (!area->nr_free) | |
3173 | continue; | |
3174 | ||
97a16fc8 MG |
3175 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
3176 | if (!list_empty(&area->free_list[mt])) | |
3177 | return true; | |
3178 | } | |
3179 | ||
3180 | #ifdef CONFIG_CMA | |
d883c6cf JK |
3181 | if ((alloc_flags & ALLOC_CMA) && |
3182 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
97a16fc8 | 3183 | return true; |
d883c6cf | 3184 | } |
97a16fc8 | 3185 | #endif |
b050e376 VB |
3186 | if (alloc_harder && |
3187 | !list_empty(&area->free_list[MIGRATE_HIGHATOMIC])) | |
3188 | return true; | |
1da177e4 | 3189 | } |
97a16fc8 | 3190 | return false; |
88f5acf8 MG |
3191 | } |
3192 | ||
7aeb09f9 | 3193 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 3194 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
3195 | { |
3196 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3197 | zone_page_state(z, NR_FREE_PAGES)); | |
3198 | } | |
3199 | ||
48ee5f36 MG |
3200 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
3201 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
3202 | { | |
3203 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
d883c6cf JK |
3204 | long cma_pages = 0; |
3205 | ||
3206 | #ifdef CONFIG_CMA | |
3207 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3208 | if (!(alloc_flags & ALLOC_CMA)) | |
3209 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
3210 | #endif | |
48ee5f36 MG |
3211 | |
3212 | /* | |
3213 | * Fast check for order-0 only. If this fails then the reserves | |
3214 | * need to be calculated. There is a corner case where the check | |
3215 | * passes but only the high-order atomic reserve are free. If | |
3216 | * the caller is !atomic then it'll uselessly search the free | |
3217 | * list. That corner case is then slower but it is harmless. | |
3218 | */ | |
d883c6cf | 3219 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) |
48ee5f36 MG |
3220 | return true; |
3221 | ||
3222 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3223 | free_pages); | |
3224 | } | |
3225 | ||
7aeb09f9 | 3226 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 3227 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
3228 | { |
3229 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3230 | ||
3231 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3232 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3233 | ||
e2b19197 | 3234 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 3235 | free_pages); |
1da177e4 LT |
3236 | } |
3237 | ||
9276b1bc | 3238 | #ifdef CONFIG_NUMA |
957f822a DR |
3239 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3240 | { | |
e02dc017 | 3241 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
5f7a75ac | 3242 | RECLAIM_DISTANCE; |
957f822a | 3243 | } |
9276b1bc | 3244 | #else /* CONFIG_NUMA */ |
957f822a DR |
3245 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3246 | { | |
3247 | return true; | |
3248 | } | |
9276b1bc PJ |
3249 | #endif /* CONFIG_NUMA */ |
3250 | ||
7fb1d9fc | 3251 | /* |
0798e519 | 3252 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3253 | * a page. |
3254 | */ | |
3255 | static struct page * | |
a9263751 VB |
3256 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3257 | const struct alloc_context *ac) | |
753ee728 | 3258 | { |
c33d6c06 | 3259 | struct zoneref *z = ac->preferred_zoneref; |
5117f45d | 3260 | struct zone *zone; |
3b8c0be4 MG |
3261 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
3262 | ||
7fb1d9fc | 3263 | /* |
9276b1bc | 3264 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3265 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3266 | */ |
c33d6c06 | 3267 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 3268 | ac->nodemask) { |
be06af00 | 3269 | struct page *page; |
e085dbc5 JW |
3270 | unsigned long mark; |
3271 | ||
664eedde MG |
3272 | if (cpusets_enabled() && |
3273 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3274 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3275 | continue; |
a756cf59 JW |
3276 | /* |
3277 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3278 | * want to get it from a node that is within its dirty |
3279 | * limit, such that no single node holds more than its | |
a756cf59 | 3280 | * proportional share of globally allowed dirty pages. |
281e3726 | 3281 | * The dirty limits take into account the node's |
a756cf59 JW |
3282 | * lowmem reserves and high watermark so that kswapd |
3283 | * should be able to balance it without having to | |
3284 | * write pages from its LRU list. | |
3285 | * | |
a756cf59 | 3286 | * XXX: For now, allow allocations to potentially |
281e3726 | 3287 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3288 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3289 | * which is important when on a NUMA setup the allowed |
281e3726 | 3290 | * nodes are together not big enough to reach the |
a756cf59 | 3291 | * global limit. The proper fix for these situations |
281e3726 | 3292 | * will require awareness of nodes in the |
a756cf59 JW |
3293 | * dirty-throttling and the flusher threads. |
3294 | */ | |
3b8c0be4 MG |
3295 | if (ac->spread_dirty_pages) { |
3296 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3297 | continue; | |
3298 | ||
3299 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3300 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3301 | continue; | |
3302 | } | |
3303 | } | |
7fb1d9fc | 3304 | |
e085dbc5 | 3305 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 3306 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 3307 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
3308 | int ret; |
3309 | ||
c9e97a19 PT |
3310 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3311 | /* | |
3312 | * Watermark failed for this zone, but see if we can | |
3313 | * grow this zone if it contains deferred pages. | |
3314 | */ | |
3315 | if (static_branch_unlikely(&deferred_pages)) { | |
3316 | if (_deferred_grow_zone(zone, order)) | |
3317 | goto try_this_zone; | |
3318 | } | |
3319 | #endif | |
5dab2911 MG |
3320 | /* Checked here to keep the fast path fast */ |
3321 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3322 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3323 | goto try_this_zone; | |
3324 | ||
a5f5f91d | 3325 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3326 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3327 | continue; |
3328 | ||
a5f5f91d | 3329 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3330 | switch (ret) { |
a5f5f91d | 3331 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3332 | /* did not scan */ |
cd38b115 | 3333 | continue; |
a5f5f91d | 3334 | case NODE_RECLAIM_FULL: |
fa5e084e | 3335 | /* scanned but unreclaimable */ |
cd38b115 | 3336 | continue; |
fa5e084e MG |
3337 | default: |
3338 | /* did we reclaim enough */ | |
fed2719e | 3339 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3340 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3341 | goto try_this_zone; |
3342 | ||
fed2719e | 3343 | continue; |
0798e519 | 3344 | } |
7fb1d9fc RS |
3345 | } |
3346 | ||
fa5e084e | 3347 | try_this_zone: |
066b2393 | 3348 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3349 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3350 | if (page) { |
479f854a | 3351 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3352 | |
3353 | /* | |
3354 | * If this is a high-order atomic allocation then check | |
3355 | * if the pageblock should be reserved for the future | |
3356 | */ | |
3357 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3358 | reserve_highatomic_pageblock(page, zone, order); | |
3359 | ||
75379191 | 3360 | return page; |
c9e97a19 PT |
3361 | } else { |
3362 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
3363 | /* Try again if zone has deferred pages */ | |
3364 | if (static_branch_unlikely(&deferred_pages)) { | |
3365 | if (_deferred_grow_zone(zone, order)) | |
3366 | goto try_this_zone; | |
3367 | } | |
3368 | #endif | |
75379191 | 3369 | } |
54a6eb5c | 3370 | } |
9276b1bc | 3371 | |
4ffeaf35 | 3372 | return NULL; |
753ee728 MH |
3373 | } |
3374 | ||
9af744d7 | 3375 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3376 | { |
a238ab5b | 3377 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
aa187507 | 3378 | static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1); |
a238ab5b | 3379 | |
2c029a1e | 3380 | if (!__ratelimit(&show_mem_rs)) |
a238ab5b DH |
3381 | return; |
3382 | ||
3383 | /* | |
3384 | * This documents exceptions given to allocations in certain | |
3385 | * contexts that are allowed to allocate outside current's set | |
3386 | * of allowed nodes. | |
3387 | */ | |
3388 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3389 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3390 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3391 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3392 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3393 | filter &= ~SHOW_MEM_FILTER_NODES; |
3394 | ||
9af744d7 | 3395 | show_mem(filter, nodemask); |
aa187507 MH |
3396 | } |
3397 | ||
a8e99259 | 3398 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3399 | { |
3400 | struct va_format vaf; | |
3401 | va_list args; | |
3402 | static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL, | |
3403 | DEFAULT_RATELIMIT_BURST); | |
3404 | ||
0f7896f1 | 3405 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3406 | return; |
3407 | ||
7877cdcc MH |
3408 | va_start(args, fmt); |
3409 | vaf.fmt = fmt; | |
3410 | vaf.va = &args; | |
0205f755 MH |
3411 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl\n", |
3412 | current->comm, &vaf, gfp_mask, &gfp_mask, | |
3413 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3414 | va_end(args); |
3ee9a4f0 | 3415 | |
a8e99259 | 3416 | cpuset_print_current_mems_allowed(); |
3ee9a4f0 | 3417 | |
a238ab5b | 3418 | dump_stack(); |
685dbf6f | 3419 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3420 | } |
3421 | ||
6c18ba7a MH |
3422 | static inline struct page * |
3423 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3424 | unsigned int alloc_flags, | |
3425 | const struct alloc_context *ac) | |
3426 | { | |
3427 | struct page *page; | |
3428 | ||
3429 | page = get_page_from_freelist(gfp_mask, order, | |
3430 | alloc_flags|ALLOC_CPUSET, ac); | |
3431 | /* | |
3432 | * fallback to ignore cpuset restriction if our nodes | |
3433 | * are depleted | |
3434 | */ | |
3435 | if (!page) | |
3436 | page = get_page_from_freelist(gfp_mask, order, | |
3437 | alloc_flags, ac); | |
3438 | ||
3439 | return page; | |
3440 | } | |
3441 | ||
11e33f6a MG |
3442 | static inline struct page * |
3443 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3444 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3445 | { |
6e0fc46d DR |
3446 | struct oom_control oc = { |
3447 | .zonelist = ac->zonelist, | |
3448 | .nodemask = ac->nodemask, | |
2a966b77 | 3449 | .memcg = NULL, |
6e0fc46d DR |
3450 | .gfp_mask = gfp_mask, |
3451 | .order = order, | |
6e0fc46d | 3452 | }; |
11e33f6a MG |
3453 | struct page *page; |
3454 | ||
9879de73 JW |
3455 | *did_some_progress = 0; |
3456 | ||
9879de73 | 3457 | /* |
dc56401f JW |
3458 | * Acquire the oom lock. If that fails, somebody else is |
3459 | * making progress for us. | |
9879de73 | 3460 | */ |
dc56401f | 3461 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3462 | *did_some_progress = 1; |
11e33f6a | 3463 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3464 | return NULL; |
3465 | } | |
6b1de916 | 3466 | |
11e33f6a MG |
3467 | /* |
3468 | * Go through the zonelist yet one more time, keep very high watermark | |
3469 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3470 | * we're still under heavy pressure. But make sure that this reclaim |
3471 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3472 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3473 | */ |
e746bf73 TH |
3474 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3475 | ~__GFP_DIRECT_RECLAIM, order, | |
3476 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3477 | if (page) |
11e33f6a MG |
3478 | goto out; |
3479 | ||
06ad276a MH |
3480 | /* Coredumps can quickly deplete all memory reserves */ |
3481 | if (current->flags & PF_DUMPCORE) | |
3482 | goto out; | |
3483 | /* The OOM killer will not help higher order allocs */ | |
3484 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3485 | goto out; | |
dcda9b04 MH |
3486 | /* |
3487 | * We have already exhausted all our reclaim opportunities without any | |
3488 | * success so it is time to admit defeat. We will skip the OOM killer | |
3489 | * because it is very likely that the caller has a more reasonable | |
3490 | * fallback than shooting a random task. | |
3491 | */ | |
3492 | if (gfp_mask & __GFP_RETRY_MAYFAIL) | |
3493 | goto out; | |
06ad276a MH |
3494 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
3495 | if (ac->high_zoneidx < ZONE_NORMAL) | |
3496 | goto out; | |
3497 | if (pm_suspended_storage()) | |
3498 | goto out; | |
3499 | /* | |
3500 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3501 | * other request to make a forward progress. | |
3502 | * We are in an unfortunate situation where out_of_memory cannot | |
3503 | * do much for this context but let's try it to at least get | |
3504 | * access to memory reserved if the current task is killed (see | |
3505 | * out_of_memory). Once filesystems are ready to handle allocation | |
3506 | * failures more gracefully we should just bail out here. | |
3507 | */ | |
3508 | ||
3509 | /* The OOM killer may not free memory on a specific node */ | |
3510 | if (gfp_mask & __GFP_THISNODE) | |
3511 | goto out; | |
3da88fb3 | 3512 | |
3c2c6488 | 3513 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 3514 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3515 | *did_some_progress = 1; |
5020e285 | 3516 | |
6c18ba7a MH |
3517 | /* |
3518 | * Help non-failing allocations by giving them access to memory | |
3519 | * reserves | |
3520 | */ | |
3521 | if (gfp_mask & __GFP_NOFAIL) | |
3522 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3523 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3524 | } |
11e33f6a | 3525 | out: |
dc56401f | 3526 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3527 | return page; |
3528 | } | |
3529 | ||
33c2d214 MH |
3530 | /* |
3531 | * Maximum number of compaction retries wit a progress before OOM | |
3532 | * killer is consider as the only way to move forward. | |
3533 | */ | |
3534 | #define MAX_COMPACT_RETRIES 16 | |
3535 | ||
56de7263 MG |
3536 | #ifdef CONFIG_COMPACTION |
3537 | /* Try memory compaction for high-order allocations before reclaim */ | |
3538 | static struct page * | |
3539 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3540 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3541 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3542 | { |
98dd3b48 | 3543 | struct page *page; |
eb414681 | 3544 | unsigned long pflags; |
499118e9 | 3545 | unsigned int noreclaim_flag; |
53853e2d VB |
3546 | |
3547 | if (!order) | |
66199712 | 3548 | return NULL; |
66199712 | 3549 | |
eb414681 | 3550 | psi_memstall_enter(&pflags); |
499118e9 | 3551 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 3552 | |
c5d01d0d | 3553 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
c3486f53 | 3554 | prio); |
eb414681 | 3555 | |
499118e9 | 3556 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 3557 | psi_memstall_leave(&pflags); |
56de7263 | 3558 | |
c5d01d0d | 3559 | if (*compact_result <= COMPACT_INACTIVE) |
98dd3b48 | 3560 | return NULL; |
53853e2d | 3561 | |
98dd3b48 VB |
3562 | /* |
3563 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3564 | * count a compaction stall | |
3565 | */ | |
3566 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3567 | |
31a6c190 | 3568 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
53853e2d | 3569 | |
98dd3b48 VB |
3570 | if (page) { |
3571 | struct zone *zone = page_zone(page); | |
53853e2d | 3572 | |
98dd3b48 VB |
3573 | zone->compact_blockskip_flush = false; |
3574 | compaction_defer_reset(zone, order, true); | |
3575 | count_vm_event(COMPACTSUCCESS); | |
3576 | return page; | |
3577 | } | |
56de7263 | 3578 | |
98dd3b48 VB |
3579 | /* |
3580 | * It's bad if compaction run occurs and fails. The most likely reason | |
3581 | * is that pages exist, but not enough to satisfy watermarks. | |
3582 | */ | |
3583 | count_vm_event(COMPACTFAIL); | |
66199712 | 3584 | |
98dd3b48 | 3585 | cond_resched(); |
56de7263 MG |
3586 | |
3587 | return NULL; | |
3588 | } | |
33c2d214 | 3589 | |
3250845d VB |
3590 | static inline bool |
3591 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3592 | enum compact_result compact_result, | |
3593 | enum compact_priority *compact_priority, | |
d9436498 | 3594 | int *compaction_retries) |
3250845d VB |
3595 | { |
3596 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3597 | int min_priority; |
65190cff MH |
3598 | bool ret = false; |
3599 | int retries = *compaction_retries; | |
3600 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3601 | |
3602 | if (!order) | |
3603 | return false; | |
3604 | ||
d9436498 VB |
3605 | if (compaction_made_progress(compact_result)) |
3606 | (*compaction_retries)++; | |
3607 | ||
3250845d VB |
3608 | /* |
3609 | * compaction considers all the zone as desperately out of memory | |
3610 | * so it doesn't really make much sense to retry except when the | |
3611 | * failure could be caused by insufficient priority | |
3612 | */ | |
d9436498 VB |
3613 | if (compaction_failed(compact_result)) |
3614 | goto check_priority; | |
3250845d VB |
3615 | |
3616 | /* | |
3617 | * make sure the compaction wasn't deferred or didn't bail out early | |
3618 | * due to locks contention before we declare that we should give up. | |
3619 | * But do not retry if the given zonelist is not suitable for | |
3620 | * compaction. | |
3621 | */ | |
65190cff MH |
3622 | if (compaction_withdrawn(compact_result)) { |
3623 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
3624 | goto out; | |
3625 | } | |
3250845d VB |
3626 | |
3627 | /* | |
dcda9b04 | 3628 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
3629 | * costly ones because they are de facto nofail and invoke OOM |
3630 | * killer to move on while costly can fail and users are ready | |
3631 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3632 | * would need much more detailed feedback from compaction to | |
3633 | * make a better decision. | |
3634 | */ | |
3635 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3636 | max_retries /= 4; | |
65190cff MH |
3637 | if (*compaction_retries <= max_retries) { |
3638 | ret = true; | |
3639 | goto out; | |
3640 | } | |
3250845d | 3641 | |
d9436498 VB |
3642 | /* |
3643 | * Make sure there are attempts at the highest priority if we exhausted | |
3644 | * all retries or failed at the lower priorities. | |
3645 | */ | |
3646 | check_priority: | |
c2033b00 VB |
3647 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3648 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3649 | |
c2033b00 | 3650 | if (*compact_priority > min_priority) { |
d9436498 VB |
3651 | (*compact_priority)--; |
3652 | *compaction_retries = 0; | |
65190cff | 3653 | ret = true; |
d9436498 | 3654 | } |
65190cff MH |
3655 | out: |
3656 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
3657 | return ret; | |
3250845d | 3658 | } |
56de7263 MG |
3659 | #else |
3660 | static inline struct page * | |
3661 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3662 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3663 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3664 | { |
33c2d214 | 3665 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3666 | return NULL; |
3667 | } | |
33c2d214 MH |
3668 | |
3669 | static inline bool | |
86a294a8 MH |
3670 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3671 | enum compact_result compact_result, | |
a5508cd8 | 3672 | enum compact_priority *compact_priority, |
d9436498 | 3673 | int *compaction_retries) |
33c2d214 | 3674 | { |
31e49bfd MH |
3675 | struct zone *zone; |
3676 | struct zoneref *z; | |
3677 | ||
3678 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3679 | return false; | |
3680 | ||
3681 | /* | |
3682 | * There are setups with compaction disabled which would prefer to loop | |
3683 | * inside the allocator rather than hit the oom killer prematurely. | |
3684 | * Let's give them a good hope and keep retrying while the order-0 | |
3685 | * watermarks are OK. | |
3686 | */ | |
3687 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3688 | ac->nodemask) { | |
3689 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
3690 | ac_classzone_idx(ac), alloc_flags)) | |
3691 | return true; | |
3692 | } | |
33c2d214 MH |
3693 | return false; |
3694 | } | |
3250845d | 3695 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3696 | |
d92a8cfc | 3697 | #ifdef CONFIG_LOCKDEP |
93781325 | 3698 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
3699 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
3700 | ||
3701 | static bool __need_fs_reclaim(gfp_t gfp_mask) | |
3702 | { | |
3703 | gfp_mask = current_gfp_context(gfp_mask); | |
3704 | ||
3705 | /* no reclaim without waiting on it */ | |
3706 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
3707 | return false; | |
3708 | ||
3709 | /* this guy won't enter reclaim */ | |
2e517d68 | 3710 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
3711 | return false; |
3712 | ||
3713 | /* We're only interested __GFP_FS allocations for now */ | |
3714 | if (!(gfp_mask & __GFP_FS)) | |
3715 | return false; | |
3716 | ||
3717 | if (gfp_mask & __GFP_NOLOCKDEP) | |
3718 | return false; | |
3719 | ||
3720 | return true; | |
3721 | } | |
3722 | ||
93781325 OS |
3723 | void __fs_reclaim_acquire(void) |
3724 | { | |
3725 | lock_map_acquire(&__fs_reclaim_map); | |
3726 | } | |
3727 | ||
3728 | void __fs_reclaim_release(void) | |
3729 | { | |
3730 | lock_map_release(&__fs_reclaim_map); | |
3731 | } | |
3732 | ||
d92a8cfc PZ |
3733 | void fs_reclaim_acquire(gfp_t gfp_mask) |
3734 | { | |
3735 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 3736 | __fs_reclaim_acquire(); |
d92a8cfc PZ |
3737 | } |
3738 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
3739 | ||
3740 | void fs_reclaim_release(gfp_t gfp_mask) | |
3741 | { | |
3742 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 3743 | __fs_reclaim_release(); |
d92a8cfc PZ |
3744 | } |
3745 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
3746 | #endif | |
3747 | ||
bba90710 MS |
3748 | /* Perform direct synchronous page reclaim */ |
3749 | static int | |
a9263751 VB |
3750 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3751 | const struct alloc_context *ac) | |
11e33f6a | 3752 | { |
11e33f6a | 3753 | struct reclaim_state reclaim_state; |
bba90710 | 3754 | int progress; |
499118e9 | 3755 | unsigned int noreclaim_flag; |
eb414681 | 3756 | unsigned long pflags; |
11e33f6a MG |
3757 | |
3758 | cond_resched(); | |
3759 | ||
3760 | /* We now go into synchronous reclaim */ | |
3761 | cpuset_memory_pressure_bump(); | |
eb414681 | 3762 | psi_memstall_enter(&pflags); |
d92a8cfc | 3763 | fs_reclaim_acquire(gfp_mask); |
93781325 | 3764 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 3765 | reclaim_state.reclaimed_slab = 0; |
c06b1fca | 3766 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3767 | |
a9263751 VB |
3768 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3769 | ac->nodemask); | |
11e33f6a | 3770 | |
c06b1fca | 3771 | current->reclaim_state = NULL; |
499118e9 | 3772 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 3773 | fs_reclaim_release(gfp_mask); |
eb414681 | 3774 | psi_memstall_leave(&pflags); |
11e33f6a MG |
3775 | |
3776 | cond_resched(); | |
3777 | ||
bba90710 MS |
3778 | return progress; |
3779 | } | |
3780 | ||
3781 | /* The really slow allocator path where we enter direct reclaim */ | |
3782 | static inline struct page * | |
3783 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3784 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3785 | unsigned long *did_some_progress) |
bba90710 MS |
3786 | { |
3787 | struct page *page = NULL; | |
3788 | bool drained = false; | |
3789 | ||
a9263751 | 3790 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3791 | if (unlikely(!(*did_some_progress))) |
3792 | return NULL; | |
11e33f6a | 3793 | |
9ee493ce | 3794 | retry: |
31a6c190 | 3795 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3796 | |
3797 | /* | |
3798 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3799 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3800 | * Shrink them them and try again | |
9ee493ce MG |
3801 | */ |
3802 | if (!page && !drained) { | |
29fac03b | 3803 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3804 | drain_all_pages(NULL); |
9ee493ce MG |
3805 | drained = true; |
3806 | goto retry; | |
3807 | } | |
3808 | ||
11e33f6a MG |
3809 | return page; |
3810 | } | |
3811 | ||
5ecd9d40 DR |
3812 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
3813 | const struct alloc_context *ac) | |
3a025760 JW |
3814 | { |
3815 | struct zoneref *z; | |
3816 | struct zone *zone; | |
e1a55637 | 3817 | pg_data_t *last_pgdat = NULL; |
5ecd9d40 | 3818 | enum zone_type high_zoneidx = ac->high_zoneidx; |
3a025760 | 3819 | |
5ecd9d40 DR |
3820 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx, |
3821 | ac->nodemask) { | |
e1a55637 | 3822 | if (last_pgdat != zone->zone_pgdat) |
5ecd9d40 | 3823 | wakeup_kswapd(zone, gfp_mask, order, high_zoneidx); |
e1a55637 MG |
3824 | last_pgdat = zone->zone_pgdat; |
3825 | } | |
3a025760 JW |
3826 | } |
3827 | ||
c603844b | 3828 | static inline unsigned int |
341ce06f PZ |
3829 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3830 | { | |
c603844b | 3831 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3832 | |
a56f57ff | 3833 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3834 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3835 | |
341ce06f PZ |
3836 | /* |
3837 | * The caller may dip into page reserves a bit more if the caller | |
3838 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3839 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3840 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3841 | */ |
e6223a3b | 3842 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3843 | |
d0164adc | 3844 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3845 | /* |
b104a35d DR |
3846 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3847 | * if it can't schedule. | |
5c3240d9 | 3848 | */ |
b104a35d | 3849 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3850 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3851 | /* |
b104a35d | 3852 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3853 | * comment for __cpuset_node_allowed(). |
523b9458 | 3854 | */ |
341ce06f | 3855 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3856 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3857 | alloc_flags |= ALLOC_HARDER; |
3858 | ||
d883c6cf JK |
3859 | #ifdef CONFIG_CMA |
3860 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
3861 | alloc_flags |= ALLOC_CMA; | |
3862 | #endif | |
341ce06f PZ |
3863 | return alloc_flags; |
3864 | } | |
3865 | ||
cd04ae1e | 3866 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 3867 | { |
cd04ae1e MH |
3868 | if (!tsk_is_oom_victim(tsk)) |
3869 | return false; | |
3870 | ||
3871 | /* | |
3872 | * !MMU doesn't have oom reaper so give access to memory reserves | |
3873 | * only to the thread with TIF_MEMDIE set | |
3874 | */ | |
3875 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
3876 | return false; |
3877 | ||
cd04ae1e MH |
3878 | return true; |
3879 | } | |
3880 | ||
3881 | /* | |
3882 | * Distinguish requests which really need access to full memory | |
3883 | * reserves from oom victims which can live with a portion of it | |
3884 | */ | |
3885 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
3886 | { | |
3887 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
3888 | return 0; | |
31a6c190 | 3889 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 3890 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 3891 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
3892 | return ALLOC_NO_WATERMARKS; |
3893 | if (!in_interrupt()) { | |
3894 | if (current->flags & PF_MEMALLOC) | |
3895 | return ALLOC_NO_WATERMARKS; | |
3896 | else if (oom_reserves_allowed(current)) | |
3897 | return ALLOC_OOM; | |
3898 | } | |
31a6c190 | 3899 | |
cd04ae1e MH |
3900 | return 0; |
3901 | } | |
3902 | ||
3903 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
3904 | { | |
3905 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
3906 | } |
3907 | ||
0a0337e0 MH |
3908 | /* |
3909 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
3910 | * for the given allocation request. | |
491d79ae JW |
3911 | * |
3912 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
3913 | * without success, or when we couldn't even meet the watermark if we | |
3914 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
3915 | * |
3916 | * Returns true if a retry is viable or false to enter the oom path. | |
3917 | */ | |
3918 | static inline bool | |
3919 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
3920 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 3921 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
3922 | { |
3923 | struct zone *zone; | |
3924 | struct zoneref *z; | |
15f570bf | 3925 | bool ret = false; |
0a0337e0 | 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 | |
15f570bf MH |
3989 | ret = true; |
3990 | goto out; | |
0a0337e0 MH |
3991 | } |
3992 | } | |
3993 | ||
15f570bf MH |
3994 | out: |
3995 | /* | |
3996 | * Memory allocation/reclaim might be called from a WQ context and the | |
3997 | * current implementation of the WQ concurrency control doesn't | |
3998 | * recognize that a particular WQ is congested if the worker thread is | |
3999 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4000 | * here rather than calling cond_resched(). | |
4001 | */ | |
4002 | if (current->flags & PF_WQ_WORKER) | |
4003 | schedule_timeout_uninterruptible(1); | |
4004 | else | |
4005 | cond_resched(); | |
4006 | return ret; | |
0a0337e0 MH |
4007 | } |
4008 | ||
902b6281 VB |
4009 | static inline bool |
4010 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4011 | { | |
4012 | /* | |
4013 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4014 | * mempolicy don't intersect. This should be normally dealt with by | |
4015 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4016 | * such a way the check therein was true, and then it became false | |
4017 | * before we got our cpuset_mems_cookie here. | |
4018 | * This assumes that for all allocations, ac->nodemask can come only | |
4019 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4020 | * when it does not intersect with the cpuset restrictions) or the | |
4021 | * caller can deal with a violated nodemask. | |
4022 | */ | |
4023 | if (cpusets_enabled() && ac->nodemask && | |
4024 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4025 | ac->nodemask = NULL; | |
4026 | return true; | |
4027 | } | |
4028 | ||
4029 | /* | |
4030 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4031 | * possible to race with parallel threads in such a way that our | |
4032 | * allocation can fail while the mask is being updated. If we are about | |
4033 | * to fail, check if the cpuset changed during allocation and if so, | |
4034 | * retry. | |
4035 | */ | |
4036 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4037 | return true; | |
4038 | ||
4039 | return false; | |
4040 | } | |
4041 | ||
11e33f6a MG |
4042 | static inline struct page * |
4043 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4044 | struct alloc_context *ac) |
11e33f6a | 4045 | { |
d0164adc | 4046 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4047 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4048 | struct page *page = NULL; |
c603844b | 4049 | unsigned int alloc_flags; |
11e33f6a | 4050 | unsigned long did_some_progress; |
5ce9bfef | 4051 | enum compact_priority compact_priority; |
c5d01d0d | 4052 | enum compact_result compact_result; |
5ce9bfef VB |
4053 | int compaction_retries; |
4054 | int no_progress_loops; | |
5ce9bfef | 4055 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4056 | int reserve_flags; |
1da177e4 | 4057 | |
72807a74 MG |
4058 | /* |
4059 | * In the slowpath, we sanity check order to avoid ever trying to | |
4060 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
4061 | * be using allocators in order of preference for an area that is | |
4062 | * too large. | |
4063 | */ | |
1fc28b70 MG |
4064 | if (order >= MAX_ORDER) { |
4065 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 4066 | return NULL; |
1fc28b70 | 4067 | } |
1da177e4 | 4068 | |
d0164adc MG |
4069 | /* |
4070 | * We also sanity check to catch abuse of atomic reserves being used by | |
4071 | * callers that are not in atomic context. | |
4072 | */ | |
4073 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4074 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4075 | gfp_mask &= ~__GFP_ATOMIC; | |
4076 | ||
5ce9bfef VB |
4077 | retry_cpuset: |
4078 | compaction_retries = 0; | |
4079 | no_progress_loops = 0; | |
4080 | compact_priority = DEF_COMPACT_PRIORITY; | |
4081 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4082 | |
4083 | /* | |
4084 | * The fast path uses conservative alloc_flags to succeed only until | |
4085 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4086 | * alloc_flags precisely. So we do that now. | |
4087 | */ | |
4088 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4089 | ||
e47483bc VB |
4090 | /* |
4091 | * We need to recalculate the starting point for the zonelist iterator | |
4092 | * because we might have used different nodemask in the fast path, or | |
4093 | * there was a cpuset modification and we are retrying - otherwise we | |
4094 | * could end up iterating over non-eligible zones endlessly. | |
4095 | */ | |
4096 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
4097 | ac->high_zoneidx, ac->nodemask); | |
4098 | if (!ac->preferred_zoneref->zone) | |
4099 | goto nopage; | |
4100 | ||
23771235 | 4101 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
5ecd9d40 | 4102 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4103 | |
4104 | /* | |
4105 | * The adjusted alloc_flags might result in immediate success, so try | |
4106 | * that first | |
4107 | */ | |
4108 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4109 | if (page) | |
4110 | goto got_pg; | |
4111 | ||
a8161d1e VB |
4112 | /* |
4113 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4114 | * that we have enough base pages and don't need to reclaim. For non- |
4115 | * movable high-order allocations, do that as well, as compaction will | |
4116 | * try prevent permanent fragmentation by migrating from blocks of the | |
4117 | * same migratetype. | |
4118 | * Don't try this for allocations that are allowed to ignore | |
4119 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4120 | */ |
282722b0 VB |
4121 | if (can_direct_reclaim && |
4122 | (costly_order || | |
4123 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4124 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4125 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4126 | alloc_flags, ac, | |
a5508cd8 | 4127 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4128 | &compact_result); |
4129 | if (page) | |
4130 | goto got_pg; | |
4131 | ||
3eb2771b VB |
4132 | /* |
4133 | * Checks for costly allocations with __GFP_NORETRY, which | |
4134 | * includes THP page fault allocations | |
4135 | */ | |
282722b0 | 4136 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { |
a8161d1e VB |
4137 | /* |
4138 | * If compaction is deferred for high-order allocations, | |
4139 | * it is because sync compaction recently failed. If | |
4140 | * this is the case and the caller requested a THP | |
4141 | * allocation, we do not want to heavily disrupt the | |
4142 | * system, so we fail the allocation instead of entering | |
4143 | * direct reclaim. | |
4144 | */ | |
4145 | if (compact_result == COMPACT_DEFERRED) | |
4146 | goto nopage; | |
4147 | ||
a8161d1e | 4148 | /* |
3eb2771b VB |
4149 | * Looks like reclaim/compaction is worth trying, but |
4150 | * sync compaction could be very expensive, so keep | |
25160354 | 4151 | * using async compaction. |
a8161d1e | 4152 | */ |
a5508cd8 | 4153 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4154 | } |
4155 | } | |
23771235 | 4156 | |
31a6c190 | 4157 | retry: |
23771235 | 4158 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
31a6c190 | 4159 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
5ecd9d40 | 4160 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4161 | |
cd04ae1e MH |
4162 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4163 | if (reserve_flags) | |
4164 | alloc_flags = reserve_flags; | |
23771235 | 4165 | |
e46e7b77 | 4166 | /* |
d6a24df0 VB |
4167 | * Reset the nodemask and zonelist iterators if memory policies can be |
4168 | * ignored. These allocations are high priority and system rather than | |
4169 | * user oriented. | |
e46e7b77 | 4170 | */ |
cd04ae1e | 4171 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4172 | ac->nodemask = NULL; |
e46e7b77 MG |
4173 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4174 | ac->high_zoneidx, ac->nodemask); | |
4175 | } | |
4176 | ||
23771235 | 4177 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4178 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4179 | if (page) |
4180 | goto got_pg; | |
1da177e4 | 4181 | |
d0164adc | 4182 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4183 | if (!can_direct_reclaim) |
1da177e4 LT |
4184 | goto nopage; |
4185 | ||
9a67f648 MH |
4186 | /* Avoid recursion of direct reclaim */ |
4187 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4188 | goto nopage; |
4189 | ||
a8161d1e VB |
4190 | /* Try direct reclaim and then allocating */ |
4191 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4192 | &did_some_progress); | |
4193 | if (page) | |
4194 | goto got_pg; | |
4195 | ||
4196 | /* Try direct compaction and then allocating */ | |
a9263751 | 4197 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4198 | compact_priority, &compact_result); |
56de7263 MG |
4199 | if (page) |
4200 | goto got_pg; | |
75f30861 | 4201 | |
9083905a JW |
4202 | /* Do not loop if specifically requested */ |
4203 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4204 | goto nopage; |
9083905a | 4205 | |
0a0337e0 MH |
4206 | /* |
4207 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4208 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4209 | */ |
dcda9b04 | 4210 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4211 | goto nopage; |
0a0337e0 | 4212 | |
0a0337e0 | 4213 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4214 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4215 | goto retry; |
4216 | ||
33c2d214 MH |
4217 | /* |
4218 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4219 | * reclaim is not able to make any progress because the current | |
4220 | * implementation of the compaction depends on the sufficient amount | |
4221 | * of free memory (see __compaction_suitable) | |
4222 | */ | |
4223 | if (did_some_progress > 0 && | |
86a294a8 | 4224 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4225 | compact_result, &compact_priority, |
d9436498 | 4226 | &compaction_retries)) |
33c2d214 MH |
4227 | goto retry; |
4228 | ||
902b6281 VB |
4229 | |
4230 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4231 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4232 | goto retry_cpuset; |
4233 | ||
9083905a JW |
4234 | /* Reclaim has failed us, start killing things */ |
4235 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4236 | if (page) | |
4237 | goto got_pg; | |
4238 | ||
9a67f648 | 4239 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e MH |
4240 | if (tsk_is_oom_victim(current) && |
4241 | (alloc_flags == ALLOC_OOM || | |
c288983d | 4242 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4243 | goto nopage; |
4244 | ||
9083905a | 4245 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4246 | if (did_some_progress) { |
4247 | no_progress_loops = 0; | |
9083905a | 4248 | goto retry; |
0a0337e0 | 4249 | } |
9083905a | 4250 | |
1da177e4 | 4251 | nopage: |
902b6281 VB |
4252 | /* Deal with possible cpuset update races before we fail */ |
4253 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4254 | goto retry_cpuset; |
4255 | ||
9a67f648 MH |
4256 | /* |
4257 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4258 | * we always retry | |
4259 | */ | |
4260 | if (gfp_mask & __GFP_NOFAIL) { | |
4261 | /* | |
4262 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4263 | * of any new users that actually require GFP_NOWAIT | |
4264 | */ | |
4265 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4266 | goto fail; | |
4267 | ||
4268 | /* | |
4269 | * PF_MEMALLOC request from this context is rather bizarre | |
4270 | * because we cannot reclaim anything and only can loop waiting | |
4271 | * for somebody to do a work for us | |
4272 | */ | |
4273 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4274 | ||
4275 | /* | |
4276 | * non failing costly orders are a hard requirement which we | |
4277 | * are not prepared for much so let's warn about these users | |
4278 | * so that we can identify them and convert them to something | |
4279 | * else. | |
4280 | */ | |
4281 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4282 | ||
6c18ba7a MH |
4283 | /* |
4284 | * Help non-failing allocations by giving them access to memory | |
4285 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4286 | * could deplete whole memory reserves which would just make | |
4287 | * the situation worse | |
4288 | */ | |
4289 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4290 | if (page) | |
4291 | goto got_pg; | |
4292 | ||
9a67f648 MH |
4293 | cond_resched(); |
4294 | goto retry; | |
4295 | } | |
4296 | fail: | |
a8e99259 | 4297 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4298 | "page allocation failure: order:%u", order); |
1da177e4 | 4299 | got_pg: |
072bb0aa | 4300 | return page; |
1da177e4 | 4301 | } |
11e33f6a | 4302 | |
9cd75558 | 4303 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4304 | int preferred_nid, nodemask_t *nodemask, |
9cd75558 MG |
4305 | struct alloc_context *ac, gfp_t *alloc_mask, |
4306 | unsigned int *alloc_flags) | |
11e33f6a | 4307 | { |
9cd75558 | 4308 | ac->high_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4309 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 MG |
4310 | ac->nodemask = nodemask; |
4311 | ac->migratetype = gfpflags_to_migratetype(gfp_mask); | |
11e33f6a | 4312 | |
682a3385 | 4313 | if (cpusets_enabled()) { |
9cd75558 | 4314 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
4315 | if (!ac->nodemask) |
4316 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
4317 | else |
4318 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4319 | } |
4320 | ||
d92a8cfc PZ |
4321 | fs_reclaim_acquire(gfp_mask); |
4322 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4323 | |
d0164adc | 4324 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4325 | |
4326 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4327 | return false; |
11e33f6a | 4328 | |
d883c6cf JK |
4329 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
4330 | *alloc_flags |= ALLOC_CMA; | |
4331 | ||
9cd75558 MG |
4332 | return true; |
4333 | } | |
21bb9bd1 | 4334 | |
9cd75558 | 4335 | /* Determine whether to spread dirty pages and what the first usable zone */ |
a380b40a | 4336 | static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac) |
9cd75558 | 4337 | { |
c9ab0c4f | 4338 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4339 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4340 | |
e46e7b77 MG |
4341 | /* |
4342 | * The preferred zone is used for statistics but crucially it is | |
4343 | * also used as the starting point for the zonelist iterator. It | |
4344 | * may get reset for allocations that ignore memory policies. | |
4345 | */ | |
9cd75558 MG |
4346 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4347 | ac->high_zoneidx, ac->nodemask); | |
4348 | } | |
4349 | ||
4350 | /* | |
4351 | * This is the 'heart' of the zoned buddy allocator. | |
4352 | */ | |
4353 | struct page * | |
04ec6264 VB |
4354 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
4355 | nodemask_t *nodemask) | |
9cd75558 MG |
4356 | { |
4357 | struct page *page; | |
4358 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
f19360f0 | 4359 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4360 | struct alloc_context ac = { }; |
4361 | ||
4362 | gfp_mask &= gfp_allowed_mask; | |
f19360f0 | 4363 | alloc_mask = gfp_mask; |
04ec6264 | 4364 | if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags)) |
9cd75558 MG |
4365 | return NULL; |
4366 | ||
a380b40a | 4367 | finalise_ac(gfp_mask, &ac); |
5bb1b169 | 4368 | |
5117f45d | 4369 | /* First allocation attempt */ |
a9263751 | 4370 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4371 | if (likely(page)) |
4372 | goto out; | |
11e33f6a | 4373 | |
4fcb0971 | 4374 | /* |
7dea19f9 MH |
4375 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4376 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4377 | * from a particular context which has been marked by | |
4378 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4379 | */ |
7dea19f9 | 4380 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4381 | ac.spread_dirty_pages = false; |
23f086f9 | 4382 | |
4741526b MG |
4383 | /* |
4384 | * Restore the original nodemask if it was potentially replaced with | |
4385 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4386 | */ | |
e47483bc | 4387 | if (unlikely(ac.nodemask != nodemask)) |
4741526b | 4388 | ac.nodemask = nodemask; |
16096c25 | 4389 | |
4fcb0971 | 4390 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4391 | |
4fcb0971 | 4392 | out: |
c4159a75 VD |
4393 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
4394 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { | |
4395 | __free_pages(page, order); | |
4396 | page = NULL; | |
4949148a VD |
4397 | } |
4398 | ||
4fcb0971 MG |
4399 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
4400 | ||
11e33f6a | 4401 | return page; |
1da177e4 | 4402 | } |
d239171e | 4403 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4404 | |
4405 | /* | |
9ea9a680 MH |
4406 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4407 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4408 | * you need to access high mem. | |
1da177e4 | 4409 | */ |
920c7a5d | 4410 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4411 | { |
945a1113 AM |
4412 | struct page *page; |
4413 | ||
9ea9a680 | 4414 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4415 | if (!page) |
4416 | return 0; | |
4417 | return (unsigned long) page_address(page); | |
4418 | } | |
1da177e4 LT |
4419 | EXPORT_SYMBOL(__get_free_pages); |
4420 | ||
920c7a5d | 4421 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4422 | { |
945a1113 | 4423 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4424 | } |
1da177e4 LT |
4425 | EXPORT_SYMBOL(get_zeroed_page); |
4426 | ||
920c7a5d | 4427 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 4428 | { |
b5810039 | 4429 | if (put_page_testzero(page)) { |
1da177e4 | 4430 | if (order == 0) |
2d4894b5 | 4431 | free_unref_page(page); |
1da177e4 LT |
4432 | else |
4433 | __free_pages_ok(page, order); | |
4434 | } | |
4435 | } | |
4436 | ||
4437 | EXPORT_SYMBOL(__free_pages); | |
4438 | ||
920c7a5d | 4439 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4440 | { |
4441 | if (addr != 0) { | |
725d704e | 4442 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4443 | __free_pages(virt_to_page((void *)addr), order); |
4444 | } | |
4445 | } | |
4446 | ||
4447 | EXPORT_SYMBOL(free_pages); | |
4448 | ||
b63ae8ca AD |
4449 | /* |
4450 | * Page Fragment: | |
4451 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4452 | * within a 0 or higher order page. Multiple fragments within that page | |
4453 | * are individually refcounted, in the page's reference counter. | |
4454 | * | |
4455 | * The page_frag functions below provide a simple allocation framework for | |
4456 | * page fragments. This is used by the network stack and network device | |
4457 | * drivers to provide a backing region of memory for use as either an | |
4458 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4459 | */ | |
2976db80 AD |
4460 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4461 | gfp_t gfp_mask) | |
b63ae8ca AD |
4462 | { |
4463 | struct page *page = NULL; | |
4464 | gfp_t gfp = gfp_mask; | |
4465 | ||
4466 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4467 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4468 | __GFP_NOMEMALLOC; | |
4469 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4470 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4471 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4472 | #endif | |
4473 | if (unlikely(!page)) | |
4474 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4475 | ||
4476 | nc->va = page ? page_address(page) : NULL; | |
4477 | ||
4478 | return page; | |
4479 | } | |
4480 | ||
2976db80 | 4481 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4482 | { |
4483 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4484 | ||
4485 | if (page_ref_sub_and_test(page, count)) { | |
2976db80 AD |
4486 | unsigned int order = compound_order(page); |
4487 | ||
44fdffd7 | 4488 | if (order == 0) |
2d4894b5 | 4489 | free_unref_page(page); |
44fdffd7 AD |
4490 | else |
4491 | __free_pages_ok(page, order); | |
4492 | } | |
4493 | } | |
2976db80 | 4494 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4495 | |
8c2dd3e4 AD |
4496 | void *page_frag_alloc(struct page_frag_cache *nc, |
4497 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4498 | { |
4499 | unsigned int size = PAGE_SIZE; | |
4500 | struct page *page; | |
4501 | int offset; | |
4502 | ||
4503 | if (unlikely(!nc->va)) { | |
4504 | refill: | |
2976db80 | 4505 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4506 | if (!page) |
4507 | return NULL; | |
4508 | ||
4509 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4510 | /* if size can vary use size else just use PAGE_SIZE */ | |
4511 | size = nc->size; | |
4512 | #endif | |
4513 | /* Even if we own the page, we do not use atomic_set(). | |
4514 | * This would break get_page_unless_zero() users. | |
4515 | */ | |
fe896d18 | 4516 | page_ref_add(page, size - 1); |
b63ae8ca AD |
4517 | |
4518 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4519 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
4520 | nc->pagecnt_bias = size; |
4521 | nc->offset = size; | |
4522 | } | |
4523 | ||
4524 | offset = nc->offset - fragsz; | |
4525 | if (unlikely(offset < 0)) { | |
4526 | page = virt_to_page(nc->va); | |
4527 | ||
fe896d18 | 4528 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4529 | goto refill; |
4530 | ||
4531 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4532 | /* if size can vary use size else just use PAGE_SIZE */ | |
4533 | size = nc->size; | |
4534 | #endif | |
4535 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 4536 | set_page_count(page, size); |
b63ae8ca AD |
4537 | |
4538 | /* reset page count bias and offset to start of new frag */ | |
4539 | nc->pagecnt_bias = size; | |
4540 | offset = size - fragsz; | |
4541 | } | |
4542 | ||
4543 | nc->pagecnt_bias--; | |
4544 | nc->offset = offset; | |
4545 | ||
4546 | return nc->va + offset; | |
4547 | } | |
8c2dd3e4 | 4548 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
4549 | |
4550 | /* | |
4551 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4552 | */ | |
8c2dd3e4 | 4553 | void page_frag_free(void *addr) |
b63ae8ca AD |
4554 | { |
4555 | struct page *page = virt_to_head_page(addr); | |
4556 | ||
4557 | if (unlikely(put_page_testzero(page))) | |
4558 | __free_pages_ok(page, compound_order(page)); | |
4559 | } | |
8c2dd3e4 | 4560 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4561 | |
d00181b9 KS |
4562 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4563 | size_t size) | |
ee85c2e1 AK |
4564 | { |
4565 | if (addr) { | |
4566 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
4567 | unsigned long used = addr + PAGE_ALIGN(size); | |
4568 | ||
4569 | split_page(virt_to_page((void *)addr), order); | |
4570 | while (used < alloc_end) { | |
4571 | free_page(used); | |
4572 | used += PAGE_SIZE; | |
4573 | } | |
4574 | } | |
4575 | return (void *)addr; | |
4576 | } | |
4577 | ||
2be0ffe2 TT |
4578 | /** |
4579 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4580 | * @size: the number of bytes to allocate | |
4581 | * @gfp_mask: GFP flags for the allocation | |
4582 | * | |
4583 | * This function is similar to alloc_pages(), except that it allocates the | |
4584 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4585 | * allocate memory in power-of-two pages. | |
4586 | * | |
4587 | * This function is also limited by MAX_ORDER. | |
4588 | * | |
4589 | * Memory allocated by this function must be released by free_pages_exact(). | |
4590 | */ | |
4591 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4592 | { | |
4593 | unsigned int order = get_order(size); | |
4594 | unsigned long addr; | |
4595 | ||
4596 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 4597 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4598 | } |
4599 | EXPORT_SYMBOL(alloc_pages_exact); | |
4600 | ||
ee85c2e1 AK |
4601 | /** |
4602 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4603 | * pages on a node. | |
b5e6ab58 | 4604 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
4605 | * @size: the number of bytes to allocate |
4606 | * @gfp_mask: GFP flags for the allocation | |
4607 | * | |
4608 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4609 | * back. | |
ee85c2e1 | 4610 | */ |
e1931811 | 4611 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4612 | { |
d00181b9 | 4613 | unsigned int order = get_order(size); |
ee85c2e1 AK |
4614 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
4615 | if (!p) | |
4616 | return NULL; | |
4617 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4618 | } | |
ee85c2e1 | 4619 | |
2be0ffe2 TT |
4620 | /** |
4621 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4622 | * @virt: the value returned by alloc_pages_exact. | |
4623 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4624 | * | |
4625 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4626 | */ | |
4627 | void free_pages_exact(void *virt, size_t size) | |
4628 | { | |
4629 | unsigned long addr = (unsigned long)virt; | |
4630 | unsigned long end = addr + PAGE_ALIGN(size); | |
4631 | ||
4632 | while (addr < end) { | |
4633 | free_page(addr); | |
4634 | addr += PAGE_SIZE; | |
4635 | } | |
4636 | } | |
4637 | EXPORT_SYMBOL(free_pages_exact); | |
4638 | ||
e0fb5815 ZY |
4639 | /** |
4640 | * nr_free_zone_pages - count number of pages beyond high watermark | |
4641 | * @offset: The zone index of the highest zone | |
4642 | * | |
4643 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
4644 | * high watermark within all zones at or below a given zone index. For each | |
4645 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
4646 | * |
4647 | * nr_free_zone_pages = managed_pages - high_pages | |
e0fb5815 | 4648 | */ |
ebec3862 | 4649 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 4650 | { |
dd1a239f | 4651 | struct zoneref *z; |
54a6eb5c MG |
4652 | struct zone *zone; |
4653 | ||
e310fd43 | 4654 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 4655 | unsigned long sum = 0; |
1da177e4 | 4656 | |
0e88460d | 4657 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 4658 | |
54a6eb5c | 4659 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 4660 | unsigned long size = zone->managed_pages; |
41858966 | 4661 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
4662 | if (size > high) |
4663 | sum += size - high; | |
1da177e4 LT |
4664 | } |
4665 | ||
4666 | return sum; | |
4667 | } | |
4668 | ||
e0fb5815 ZY |
4669 | /** |
4670 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
4671 | * | |
4672 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
4673 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 4674 | */ |
ebec3862 | 4675 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 4676 | { |
af4ca457 | 4677 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 4678 | } |
c2f1a551 | 4679 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 4680 | |
e0fb5815 ZY |
4681 | /** |
4682 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
4683 | * | |
4684 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
4685 | * high watermark within all zones. | |
1da177e4 | 4686 | */ |
ebec3862 | 4687 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 4688 | { |
2a1e274a | 4689 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 4690 | } |
08e0f6a9 CL |
4691 | |
4692 | static inline void show_node(struct zone *zone) | |
1da177e4 | 4693 | { |
e5adfffc | 4694 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 4695 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 4696 | } |
1da177e4 | 4697 | |
d02bd27b IR |
4698 | long si_mem_available(void) |
4699 | { | |
4700 | long available; | |
4701 | unsigned long pagecache; | |
4702 | unsigned long wmark_low = 0; | |
4703 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 4704 | unsigned long reclaimable; |
d02bd27b IR |
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 | /* | |
b29940c1 VB |
4730 | * Part of the reclaimable slab and other kernel memory consists of |
4731 | * items that are in use, and cannot be freed. Cap this estimate at the | |
4732 | * low watermark. | |
d02bd27b | 4733 | */ |
b29940c1 VB |
4734 | reclaimable = global_node_page_state(NR_SLAB_RECLAIMABLE) + |
4735 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
4736 | available += reclaimable - min(reclaimable / 2, wmark_low); | |
034ebf65 | 4737 | |
d02bd27b IR |
4738 | if (available < 0) |
4739 | available = 0; | |
4740 | return available; | |
4741 | } | |
4742 | EXPORT_SYMBOL_GPL(si_mem_available); | |
4743 | ||
1da177e4 LT |
4744 | void si_meminfo(struct sysinfo *val) |
4745 | { | |
4746 | val->totalram = totalram_pages; | |
11fb9989 | 4747 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 4748 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 4749 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
4750 | val->totalhigh = totalhigh_pages; |
4751 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
4752 | val->mem_unit = PAGE_SIZE; |
4753 | } | |
4754 | ||
4755 | EXPORT_SYMBOL(si_meminfo); | |
4756 | ||
4757 | #ifdef CONFIG_NUMA | |
4758 | void si_meminfo_node(struct sysinfo *val, int nid) | |
4759 | { | |
cdd91a77 JL |
4760 | int zone_type; /* needs to be signed */ |
4761 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
4762 | unsigned long managed_highpages = 0; |
4763 | unsigned long free_highpages = 0; | |
1da177e4 LT |
4764 | pg_data_t *pgdat = NODE_DATA(nid); |
4765 | ||
cdd91a77 JL |
4766 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
4767 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
4768 | val->totalram = managed_pages; | |
11fb9989 | 4769 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 4770 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 4771 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
4772 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4773 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4774 | ||
4775 | if (is_highmem(zone)) { | |
4776 | managed_highpages += zone->managed_pages; | |
4777 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
4778 | } | |
4779 | } | |
4780 | val->totalhigh = managed_highpages; | |
4781 | val->freehigh = free_highpages; | |
98d2b0eb | 4782 | #else |
fc2bd799 JK |
4783 | val->totalhigh = managed_highpages; |
4784 | val->freehigh = free_highpages; | |
98d2b0eb | 4785 | #endif |
1da177e4 LT |
4786 | val->mem_unit = PAGE_SIZE; |
4787 | } | |
4788 | #endif | |
4789 | ||
ddd588b5 | 4790 | /* |
7bf02ea2 DR |
4791 | * Determine whether the node should be displayed or not, depending on whether |
4792 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 4793 | */ |
9af744d7 | 4794 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 4795 | { |
ddd588b5 | 4796 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 4797 | return false; |
ddd588b5 | 4798 | |
9af744d7 MH |
4799 | /* |
4800 | * no node mask - aka implicit memory numa policy. Do not bother with | |
4801 | * the synchronization - read_mems_allowed_begin - because we do not | |
4802 | * have to be precise here. | |
4803 | */ | |
4804 | if (!nodemask) | |
4805 | nodemask = &cpuset_current_mems_allowed; | |
4806 | ||
4807 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
4808 | } |
4809 | ||
1da177e4 LT |
4810 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4811 | ||
377e4f16 RV |
4812 | static void show_migration_types(unsigned char type) |
4813 | { | |
4814 | static const char types[MIGRATE_TYPES] = { | |
4815 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 4816 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
4817 | [MIGRATE_RECLAIMABLE] = 'E', |
4818 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
4819 | #ifdef CONFIG_CMA |
4820 | [MIGRATE_CMA] = 'C', | |
4821 | #endif | |
194159fb | 4822 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 4823 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 4824 | #endif |
377e4f16 RV |
4825 | }; |
4826 | char tmp[MIGRATE_TYPES + 1]; | |
4827 | char *p = tmp; | |
4828 | int i; | |
4829 | ||
4830 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
4831 | if (type & (1 << i)) | |
4832 | *p++ = types[i]; | |
4833 | } | |
4834 | ||
4835 | *p = '\0'; | |
1f84a18f | 4836 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
4837 | } |
4838 | ||
1da177e4 LT |
4839 | /* |
4840 | * Show free area list (used inside shift_scroll-lock stuff) | |
4841 | * We also calculate the percentage fragmentation. We do this by counting the | |
4842 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
4843 | * |
4844 | * Bits in @filter: | |
4845 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
4846 | * cpuset. | |
1da177e4 | 4847 | */ |
9af744d7 | 4848 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 4849 | { |
d1bfcdb8 | 4850 | unsigned long free_pcp = 0; |
c7241913 | 4851 | int cpu; |
1da177e4 | 4852 | struct zone *zone; |
599d0c95 | 4853 | pg_data_t *pgdat; |
1da177e4 | 4854 | |
ee99c71c | 4855 | for_each_populated_zone(zone) { |
9af744d7 | 4856 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4857 | continue; |
d1bfcdb8 | 4858 | |
761b0677 KK |
4859 | for_each_online_cpu(cpu) |
4860 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
4861 | } |
4862 | ||
a731286d KM |
4863 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4864 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
4865 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4866 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 4867 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 4868 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
4869 | global_node_page_state(NR_ACTIVE_ANON), |
4870 | global_node_page_state(NR_INACTIVE_ANON), | |
4871 | global_node_page_state(NR_ISOLATED_ANON), | |
4872 | global_node_page_state(NR_ACTIVE_FILE), | |
4873 | global_node_page_state(NR_INACTIVE_FILE), | |
4874 | global_node_page_state(NR_ISOLATED_FILE), | |
4875 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
4876 | global_node_page_state(NR_FILE_DIRTY), |
4877 | global_node_page_state(NR_WRITEBACK), | |
4878 | global_node_page_state(NR_UNSTABLE_NFS), | |
d507e2eb JW |
4879 | global_node_page_state(NR_SLAB_RECLAIMABLE), |
4880 | global_node_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 4881 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 4882 | global_node_page_state(NR_SHMEM), |
c41f012a MH |
4883 | global_zone_page_state(NR_PAGETABLE), |
4884 | global_zone_page_state(NR_BOUNCE), | |
4885 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 4886 | free_pcp, |
c41f012a | 4887 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4888 | |
599d0c95 | 4889 | for_each_online_pgdat(pgdat) { |
9af744d7 | 4890 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
4891 | continue; |
4892 | ||
599d0c95 MG |
4893 | printk("Node %d" |
4894 | " active_anon:%lukB" | |
4895 | " inactive_anon:%lukB" | |
4896 | " active_file:%lukB" | |
4897 | " inactive_file:%lukB" | |
4898 | " unevictable:%lukB" | |
4899 | " isolated(anon):%lukB" | |
4900 | " isolated(file):%lukB" | |
50658e2e | 4901 | " mapped:%lukB" |
11fb9989 MG |
4902 | " dirty:%lukB" |
4903 | " writeback:%lukB" | |
4904 | " shmem:%lukB" | |
4905 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4906 | " shmem_thp: %lukB" | |
4907 | " shmem_pmdmapped: %lukB" | |
4908 | " anon_thp: %lukB" | |
4909 | #endif | |
4910 | " writeback_tmp:%lukB" | |
4911 | " unstable:%lukB" | |
599d0c95 MG |
4912 | " all_unreclaimable? %s" |
4913 | "\n", | |
4914 | pgdat->node_id, | |
4915 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
4916 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
4917 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
4918 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
4919 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
4920 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
4921 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 4922 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
4923 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
4924 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 4925 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
4926 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4927 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
4928 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
4929 | * HPAGE_PMD_NR), | |
4930 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
4931 | #endif | |
11fb9989 MG |
4932 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
4933 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), | |
c73322d0 JW |
4934 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
4935 | "yes" : "no"); | |
599d0c95 MG |
4936 | } |
4937 | ||
ee99c71c | 4938 | for_each_populated_zone(zone) { |
1da177e4 LT |
4939 | int i; |
4940 | ||
9af744d7 | 4941 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4942 | continue; |
d1bfcdb8 KK |
4943 | |
4944 | free_pcp = 0; | |
4945 | for_each_online_cpu(cpu) | |
4946 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
4947 | ||
1da177e4 | 4948 | show_node(zone); |
1f84a18f JP |
4949 | printk(KERN_CONT |
4950 | "%s" | |
1da177e4 LT |
4951 | " free:%lukB" |
4952 | " min:%lukB" | |
4953 | " low:%lukB" | |
4954 | " high:%lukB" | |
71c799f4 MK |
4955 | " active_anon:%lukB" |
4956 | " inactive_anon:%lukB" | |
4957 | " active_file:%lukB" | |
4958 | " inactive_file:%lukB" | |
4959 | " unevictable:%lukB" | |
5a1c84b4 | 4960 | " writepending:%lukB" |
1da177e4 | 4961 | " present:%lukB" |
9feedc9d | 4962 | " managed:%lukB" |
4a0aa73f | 4963 | " mlocked:%lukB" |
c6a7f572 | 4964 | " kernel_stack:%lukB" |
4a0aa73f | 4965 | " pagetables:%lukB" |
4a0aa73f | 4966 | " bounce:%lukB" |
d1bfcdb8 KK |
4967 | " free_pcp:%lukB" |
4968 | " local_pcp:%ukB" | |
d1ce749a | 4969 | " free_cma:%lukB" |
1da177e4 LT |
4970 | "\n", |
4971 | zone->name, | |
88f5acf8 | 4972 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
4973 | K(min_wmark_pages(zone)), |
4974 | K(low_wmark_pages(zone)), | |
4975 | K(high_wmark_pages(zone)), | |
71c799f4 MK |
4976 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
4977 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
4978 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
4979 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
4980 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 4981 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 4982 | K(zone->present_pages), |
9feedc9d | 4983 | K(zone->managed_pages), |
4a0aa73f | 4984 | K(zone_page_state(zone, NR_MLOCK)), |
d30dd8be | 4985 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4a0aa73f | 4986 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 4987 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
4988 | K(free_pcp), |
4989 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 4990 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
4991 | printk("lowmem_reserve[]:"); |
4992 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
4993 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
4994 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
4995 | } |
4996 | ||
ee99c71c | 4997 | for_each_populated_zone(zone) { |
d00181b9 KS |
4998 | unsigned int order; |
4999 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5000 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5001 | |
9af744d7 | 5002 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5003 | continue; |
1da177e4 | 5004 | show_node(zone); |
1f84a18f | 5005 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5006 | |
5007 | spin_lock_irqsave(&zone->lock, flags); | |
5008 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5009 | struct free_area *area = &zone->free_area[order]; |
5010 | int type; | |
5011 | ||
5012 | nr[order] = area->nr_free; | |
8f9de51a | 5013 | total += nr[order] << order; |
377e4f16 RV |
5014 | |
5015 | types[order] = 0; | |
5016 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
5017 | if (!list_empty(&area->free_list[type])) | |
5018 | types[order] |= 1 << type; | |
5019 | } | |
1da177e4 LT |
5020 | } |
5021 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5022 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5023 | printk(KERN_CONT "%lu*%lukB ", |
5024 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5025 | if (nr[order]) |
5026 | show_migration_types(types[order]); | |
5027 | } | |
1f84a18f | 5028 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5029 | } |
5030 | ||
949f7ec5 DR |
5031 | hugetlb_show_meminfo(); |
5032 | ||
11fb9989 | 5033 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5034 | |
1da177e4 LT |
5035 | show_swap_cache_info(); |
5036 | } | |
5037 | ||
19770b32 MG |
5038 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5039 | { | |
5040 | zoneref->zone = zone; | |
5041 | zoneref->zone_idx = zone_idx(zone); | |
5042 | } | |
5043 | ||
1da177e4 LT |
5044 | /* |
5045 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5046 | * |
5047 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5048 | */ |
9d3be21b | 5049 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5050 | { |
1a93205b | 5051 | struct zone *zone; |
bc732f1d | 5052 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5053 | int nr_zones = 0; |
02a68a5e CL |
5054 | |
5055 | do { | |
2f6726e5 | 5056 | zone_type--; |
070f8032 | 5057 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5058 | if (managed_zone(zone)) { |
9d3be21b | 5059 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5060 | check_highest_zone(zone_type); |
1da177e4 | 5061 | } |
2f6726e5 | 5062 | } while (zone_type); |
bc732f1d | 5063 | |
070f8032 | 5064 | return nr_zones; |
1da177e4 LT |
5065 | } |
5066 | ||
5067 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5068 | |
5069 | static int __parse_numa_zonelist_order(char *s) | |
5070 | { | |
c9bff3ee MH |
5071 | /* |
5072 | * We used to support different zonlists modes but they turned | |
5073 | * out to be just not useful. Let's keep the warning in place | |
5074 | * if somebody still use the cmd line parameter so that we do | |
5075 | * not fail it silently | |
5076 | */ | |
5077 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5078 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5079 | return -EINVAL; |
5080 | } | |
5081 | return 0; | |
5082 | } | |
5083 | ||
5084 | static __init int setup_numa_zonelist_order(char *s) | |
5085 | { | |
ecb256f8 VL |
5086 | if (!s) |
5087 | return 0; | |
5088 | ||
c9bff3ee | 5089 | return __parse_numa_zonelist_order(s); |
f0c0b2b8 KH |
5090 | } |
5091 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
5092 | ||
c9bff3ee MH |
5093 | char numa_zonelist_order[] = "Node"; |
5094 | ||
f0c0b2b8 KH |
5095 | /* |
5096 | * sysctl handler for numa_zonelist_order | |
5097 | */ | |
cccad5b9 | 5098 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 5099 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
5100 | loff_t *ppos) |
5101 | { | |
c9bff3ee | 5102 | char *str; |
f0c0b2b8 KH |
5103 | int ret; |
5104 | ||
c9bff3ee MH |
5105 | if (!write) |
5106 | return proc_dostring(table, write, buffer, length, ppos); | |
5107 | str = memdup_user_nul(buffer, 16); | |
5108 | if (IS_ERR(str)) | |
5109 | return PTR_ERR(str); | |
dacbde09 | 5110 | |
c9bff3ee MH |
5111 | ret = __parse_numa_zonelist_order(str); |
5112 | kfree(str); | |
443c6f14 | 5113 | return ret; |
f0c0b2b8 KH |
5114 | } |
5115 | ||
5116 | ||
62bc62a8 | 5117 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5118 | static int node_load[MAX_NUMNODES]; |
5119 | ||
1da177e4 | 5120 | /** |
4dc3b16b | 5121 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5122 | * @node: node whose fallback list we're appending |
5123 | * @used_node_mask: nodemask_t of already used nodes | |
5124 | * | |
5125 | * We use a number of factors to determine which is the next node that should | |
5126 | * appear on a given node's fallback list. The node should not have appeared | |
5127 | * already in @node's fallback list, and it should be the next closest node | |
5128 | * according to the distance array (which contains arbitrary distance values | |
5129 | * from each node to each node in the system), and should also prefer nodes | |
5130 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5131 | * on them otherwise. | |
5132 | * It returns -1 if no node is found. | |
5133 | */ | |
f0c0b2b8 | 5134 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5135 | { |
4cf808eb | 5136 | int n, val; |
1da177e4 | 5137 | int min_val = INT_MAX; |
00ef2d2f | 5138 | int best_node = NUMA_NO_NODE; |
a70f7302 | 5139 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 5140 | |
4cf808eb LT |
5141 | /* Use the local node if we haven't already */ |
5142 | if (!node_isset(node, *used_node_mask)) { | |
5143 | node_set(node, *used_node_mask); | |
5144 | return node; | |
5145 | } | |
1da177e4 | 5146 | |
4b0ef1fe | 5147 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5148 | |
5149 | /* Don't want a node to appear more than once */ | |
5150 | if (node_isset(n, *used_node_mask)) | |
5151 | continue; | |
5152 | ||
1da177e4 LT |
5153 | /* Use the distance array to find the distance */ |
5154 | val = node_distance(node, n); | |
5155 | ||
4cf808eb LT |
5156 | /* Penalize nodes under us ("prefer the next node") */ |
5157 | val += (n < node); | |
5158 | ||
1da177e4 | 5159 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
5160 | tmp = cpumask_of_node(n); |
5161 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
5162 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5163 | ||
5164 | /* Slight preference for less loaded node */ | |
5165 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
5166 | val += node_load[n]; | |
5167 | ||
5168 | if (val < min_val) { | |
5169 | min_val = val; | |
5170 | best_node = n; | |
5171 | } | |
5172 | } | |
5173 | ||
5174 | if (best_node >= 0) | |
5175 | node_set(best_node, *used_node_mask); | |
5176 | ||
5177 | return best_node; | |
5178 | } | |
5179 | ||
f0c0b2b8 KH |
5180 | |
5181 | /* | |
5182 | * Build zonelists ordered by node and zones within node. | |
5183 | * This results in maximum locality--normal zone overflows into local | |
5184 | * DMA zone, if any--but risks exhausting DMA zone. | |
5185 | */ | |
9d3be21b MH |
5186 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5187 | unsigned nr_nodes) | |
1da177e4 | 5188 | { |
9d3be21b MH |
5189 | struct zoneref *zonerefs; |
5190 | int i; | |
5191 | ||
5192 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5193 | ||
5194 | for (i = 0; i < nr_nodes; i++) { | |
5195 | int nr_zones; | |
5196 | ||
5197 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5198 | |
9d3be21b MH |
5199 | nr_zones = build_zonerefs_node(node, zonerefs); |
5200 | zonerefs += nr_zones; | |
5201 | } | |
5202 | zonerefs->zone = NULL; | |
5203 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5204 | } |
5205 | ||
523b9458 CL |
5206 | /* |
5207 | * Build gfp_thisnode zonelists | |
5208 | */ | |
5209 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5210 | { | |
9d3be21b MH |
5211 | struct zoneref *zonerefs; |
5212 | int nr_zones; | |
523b9458 | 5213 | |
9d3be21b MH |
5214 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5215 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5216 | zonerefs += nr_zones; | |
5217 | zonerefs->zone = NULL; | |
5218 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5219 | } |
5220 | ||
f0c0b2b8 KH |
5221 | /* |
5222 | * Build zonelists ordered by zone and nodes within zones. | |
5223 | * This results in conserving DMA zone[s] until all Normal memory is | |
5224 | * exhausted, but results in overflowing to remote node while memory | |
5225 | * may still exist in local DMA zone. | |
5226 | */ | |
f0c0b2b8 | 5227 | |
f0c0b2b8 KH |
5228 | static void build_zonelists(pg_data_t *pgdat) |
5229 | { | |
9d3be21b MH |
5230 | static int node_order[MAX_NUMNODES]; |
5231 | int node, load, nr_nodes = 0; | |
1da177e4 | 5232 | nodemask_t used_mask; |
f0c0b2b8 | 5233 | int local_node, prev_node; |
1da177e4 LT |
5234 | |
5235 | /* NUMA-aware ordering of nodes */ | |
5236 | local_node = pgdat->node_id; | |
62bc62a8 | 5237 | load = nr_online_nodes; |
1da177e4 LT |
5238 | prev_node = local_node; |
5239 | nodes_clear(used_mask); | |
f0c0b2b8 | 5240 | |
f0c0b2b8 | 5241 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5242 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5243 | /* | |
5244 | * We don't want to pressure a particular node. | |
5245 | * So adding penalty to the first node in same | |
5246 | * distance group to make it round-robin. | |
5247 | */ | |
957f822a DR |
5248 | if (node_distance(local_node, node) != |
5249 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5250 | node_load[node] = load; |
5251 | ||
9d3be21b | 5252 | node_order[nr_nodes++] = node; |
1da177e4 LT |
5253 | prev_node = node; |
5254 | load--; | |
1da177e4 | 5255 | } |
523b9458 | 5256 | |
9d3be21b | 5257 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5258 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
5259 | } |
5260 | ||
7aac7898 LS |
5261 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5262 | /* | |
5263 | * Return node id of node used for "local" allocations. | |
5264 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5265 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5266 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5267 | */ | |
5268 | int local_memory_node(int node) | |
5269 | { | |
c33d6c06 | 5270 | struct zoneref *z; |
7aac7898 | 5271 | |
c33d6c06 | 5272 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5273 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 5274 | NULL); |
c1093b74 | 5275 | return zone_to_nid(z->zone); |
7aac7898 LS |
5276 | } |
5277 | #endif | |
f0c0b2b8 | 5278 | |
6423aa81 JK |
5279 | static void setup_min_unmapped_ratio(void); |
5280 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5281 | #else /* CONFIG_NUMA */ |
5282 | ||
f0c0b2b8 | 5283 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5284 | { |
19655d34 | 5285 | int node, local_node; |
9d3be21b MH |
5286 | struct zoneref *zonerefs; |
5287 | int nr_zones; | |
1da177e4 LT |
5288 | |
5289 | local_node = pgdat->node_id; | |
1da177e4 | 5290 | |
9d3be21b MH |
5291 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5292 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5293 | zonerefs += nr_zones; | |
1da177e4 | 5294 | |
54a6eb5c MG |
5295 | /* |
5296 | * Now we build the zonelist so that it contains the zones | |
5297 | * of all the other nodes. | |
5298 | * We don't want to pressure a particular node, so when | |
5299 | * building the zones for node N, we make sure that the | |
5300 | * zones coming right after the local ones are those from | |
5301 | * node N+1 (modulo N) | |
5302 | */ | |
5303 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5304 | if (!node_online(node)) | |
5305 | continue; | |
9d3be21b MH |
5306 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5307 | zonerefs += nr_zones; | |
1da177e4 | 5308 | } |
54a6eb5c MG |
5309 | for (node = 0; node < local_node; node++) { |
5310 | if (!node_online(node)) | |
5311 | continue; | |
9d3be21b MH |
5312 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5313 | zonerefs += nr_zones; | |
54a6eb5c MG |
5314 | } |
5315 | ||
9d3be21b MH |
5316 | zonerefs->zone = NULL; |
5317 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5318 | } |
5319 | ||
5320 | #endif /* CONFIG_NUMA */ | |
5321 | ||
99dcc3e5 CL |
5322 | /* |
5323 | * Boot pageset table. One per cpu which is going to be used for all | |
5324 | * zones and all nodes. The parameters will be set in such a way | |
5325 | * that an item put on a list will immediately be handed over to | |
5326 | * the buddy list. This is safe since pageset manipulation is done | |
5327 | * with interrupts disabled. | |
5328 | * | |
5329 | * The boot_pagesets must be kept even after bootup is complete for | |
5330 | * unused processors and/or zones. They do play a role for bootstrapping | |
5331 | * hotplugged processors. | |
5332 | * | |
5333 | * zoneinfo_show() and maybe other functions do | |
5334 | * not check if the processor is online before following the pageset pointer. | |
5335 | * Other parts of the kernel may not check if the zone is available. | |
5336 | */ | |
5337 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5338 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
385386cf | 5339 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 5340 | |
11cd8638 | 5341 | static void __build_all_zonelists(void *data) |
1da177e4 | 5342 | { |
6811378e | 5343 | int nid; |
afb6ebb3 | 5344 | int __maybe_unused cpu; |
9adb62a5 | 5345 | pg_data_t *self = data; |
b93e0f32 MH |
5346 | static DEFINE_SPINLOCK(lock); |
5347 | ||
5348 | spin_lock(&lock); | |
9276b1bc | 5349 | |
7f9cfb31 BL |
5350 | #ifdef CONFIG_NUMA |
5351 | memset(node_load, 0, sizeof(node_load)); | |
5352 | #endif | |
9adb62a5 | 5353 | |
c1152583 WY |
5354 | /* |
5355 | * This node is hotadded and no memory is yet present. So just | |
5356 | * building zonelists is fine - no need to touch other nodes. | |
5357 | */ | |
9adb62a5 JL |
5358 | if (self && !node_online(self->node_id)) { |
5359 | build_zonelists(self); | |
c1152583 WY |
5360 | } else { |
5361 | for_each_online_node(nid) { | |
5362 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 5363 | |
c1152583 WY |
5364 | build_zonelists(pgdat); |
5365 | } | |
99dcc3e5 | 5366 | |
7aac7898 LS |
5367 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5368 | /* | |
5369 | * We now know the "local memory node" for each node-- | |
5370 | * i.e., the node of the first zone in the generic zonelist. | |
5371 | * Set up numa_mem percpu variable for on-line cpus. During | |
5372 | * boot, only the boot cpu should be on-line; we'll init the | |
5373 | * secondary cpus' numa_mem as they come on-line. During | |
5374 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5375 | */ | |
d9c9a0b9 | 5376 | for_each_online_cpu(cpu) |
7aac7898 | 5377 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5378 | #endif |
d9c9a0b9 | 5379 | } |
b93e0f32 MH |
5380 | |
5381 | spin_unlock(&lock); | |
6811378e YG |
5382 | } |
5383 | ||
061f67bc RV |
5384 | static noinline void __init |
5385 | build_all_zonelists_init(void) | |
5386 | { | |
afb6ebb3 MH |
5387 | int cpu; |
5388 | ||
061f67bc | 5389 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5390 | |
5391 | /* | |
5392 | * Initialize the boot_pagesets that are going to be used | |
5393 | * for bootstrapping processors. The real pagesets for | |
5394 | * each zone will be allocated later when the per cpu | |
5395 | * allocator is available. | |
5396 | * | |
5397 | * boot_pagesets are used also for bootstrapping offline | |
5398 | * cpus if the system is already booted because the pagesets | |
5399 | * are needed to initialize allocators on a specific cpu too. | |
5400 | * F.e. the percpu allocator needs the page allocator which | |
5401 | * needs the percpu allocator in order to allocate its pagesets | |
5402 | * (a chicken-egg dilemma). | |
5403 | */ | |
5404 | for_each_possible_cpu(cpu) | |
5405 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); | |
5406 | ||
061f67bc RV |
5407 | mminit_verify_zonelist(); |
5408 | cpuset_init_current_mems_allowed(); | |
5409 | } | |
5410 | ||
4eaf3f64 | 5411 | /* |
4eaf3f64 | 5412 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5413 | * |
72675e13 | 5414 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5415 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5416 | */ |
72675e13 | 5417 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e YG |
5418 | { |
5419 | if (system_state == SYSTEM_BOOTING) { | |
061f67bc | 5420 | build_all_zonelists_init(); |
6811378e | 5421 | } else { |
11cd8638 | 5422 | __build_all_zonelists(pgdat); |
6811378e YG |
5423 | /* cpuset refresh routine should be here */ |
5424 | } | |
bd1e22b8 | 5425 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5426 | /* |
5427 | * Disable grouping by mobility if the number of pages in the | |
5428 | * system is too low to allow the mechanism to work. It would be | |
5429 | * more accurate, but expensive to check per-zone. This check is | |
5430 | * made on memory-hotadd so a system can start with mobility | |
5431 | * disabled and enable it later | |
5432 | */ | |
d9c23400 | 5433 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5434 | page_group_by_mobility_disabled = 1; |
5435 | else | |
5436 | page_group_by_mobility_disabled = 0; | |
5437 | ||
c9bff3ee | 5438 | pr_info("Built %i zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5439 | nr_online_nodes, |
756a025f JP |
5440 | page_group_by_mobility_disabled ? "off" : "on", |
5441 | vm_total_pages); | |
f0c0b2b8 | 5442 | #ifdef CONFIG_NUMA |
f88dfff5 | 5443 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5444 | #endif |
1da177e4 LT |
5445 | } |
5446 | ||
1da177e4 LT |
5447 | /* |
5448 | * Initially all pages are reserved - free ones are freed | |
5449 | * up by free_all_bootmem() once the early boot process is | |
5450 | * done. Non-atomic initialization, single-pass. | |
5451 | */ | |
c09b4240 | 5452 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a99583e7 CH |
5453 | unsigned long start_pfn, enum memmap_context context, |
5454 | struct vmem_altmap *altmap) | |
1da177e4 | 5455 | { |
29751f69 | 5456 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 5457 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 5458 | unsigned long pfn; |
3a80a7fa | 5459 | unsigned long nr_initialised = 0; |
d0dc12e8 | 5460 | struct page *page; |
342332e6 TI |
5461 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5462 | struct memblock_region *r = NULL, *tmp; | |
5463 | #endif | |
1da177e4 | 5464 | |
22b31eec HD |
5465 | if (highest_memmap_pfn < end_pfn - 1) |
5466 | highest_memmap_pfn = end_pfn - 1; | |
5467 | ||
966cf44f | 5468 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
5469 | /* |
5470 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
5471 | * memory. We limit the total number of pages to initialize to just |
5472 | * those that might contain the memory mapping. We will defer the | |
5473 | * ZONE_DEVICE page initialization until after we have released | |
5474 | * the hotplug lock. | |
4b94ffdc | 5475 | */ |
966cf44f AD |
5476 | if (zone == ZONE_DEVICE) { |
5477 | if (!altmap) | |
5478 | return; | |
5479 | ||
5480 | if (start_pfn == altmap->base_pfn) | |
5481 | start_pfn += altmap->reserve; | |
5482 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
5483 | } | |
5484 | #endif | |
4b94ffdc | 5485 | |
cbe8dd4a | 5486 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 5487 | /* |
b72d0ffb AM |
5488 | * There can be holes in boot-time mem_map[]s handed to this |
5489 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5490 | */ |
b72d0ffb AM |
5491 | if (context != MEMMAP_EARLY) |
5492 | goto not_early; | |
5493 | ||
f59f1caf | 5494 | if (!early_pfn_valid(pfn)) |
b72d0ffb AM |
5495 | continue; |
5496 | if (!early_pfn_in_nid(pfn, nid)) | |
5497 | continue; | |
5498 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
5499 | break; | |
342332e6 TI |
5500 | |
5501 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
5502 | /* |
5503 | * Check given memblock attribute by firmware which can affect | |
5504 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
5505 | * mirrored, it's an overlapped memmap init. skip it. | |
5506 | */ | |
5507 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5508 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
5509 | for_each_memblock(memory, tmp) | |
5510 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
5511 | break; | |
5512 | r = tmp; | |
5513 | } | |
5514 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
5515 | memblock_is_mirror(r)) { | |
5516 | /* already initialized as NORMAL */ | |
5517 | pfn = memblock_region_memory_end_pfn(r); | |
5518 | continue; | |
342332e6 | 5519 | } |
a2f3aa02 | 5520 | } |
b72d0ffb | 5521 | #endif |
ac5d2539 | 5522 | |
b72d0ffb | 5523 | not_early: |
d0dc12e8 PT |
5524 | page = pfn_to_page(pfn); |
5525 | __init_single_page(page, pfn, zone, nid); | |
5526 | if (context == MEMMAP_HOTPLUG) | |
d483da5b | 5527 | __SetPageReserved(page); |
d0dc12e8 | 5528 | |
ac5d2539 MG |
5529 | /* |
5530 | * Mark the block movable so that blocks are reserved for | |
5531 | * movable at startup. This will force kernel allocations | |
5532 | * to reserve their blocks rather than leaking throughout | |
5533 | * the address space during boot when many long-lived | |
974a786e | 5534 | * kernel allocations are made. |
ac5d2539 MG |
5535 | * |
5536 | * bitmap is created for zone's valid pfn range. but memmap | |
5537 | * can be created for invalid pages (for alignment) | |
5538 | * check here not to call set_pageblock_migratetype() against | |
5539 | * pfn out of zone. | |
9bb5a391 MH |
5540 | * |
5541 | * Please note that MEMMAP_HOTPLUG path doesn't clear memmap | |
5542 | * because this is done early in sparse_add_one_section | |
ac5d2539 MG |
5543 | */ |
5544 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
ac5d2539 | 5545 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
9b6e63cb | 5546 | cond_resched(); |
ac5d2539 | 5547 | } |
1da177e4 LT |
5548 | } |
5549 | } | |
5550 | ||
966cf44f AD |
5551 | #ifdef CONFIG_ZONE_DEVICE |
5552 | void __ref memmap_init_zone_device(struct zone *zone, | |
5553 | unsigned long start_pfn, | |
5554 | unsigned long size, | |
5555 | struct dev_pagemap *pgmap) | |
5556 | { | |
5557 | unsigned long pfn, end_pfn = start_pfn + size; | |
5558 | struct pglist_data *pgdat = zone->zone_pgdat; | |
5559 | unsigned long zone_idx = zone_idx(zone); | |
5560 | unsigned long start = jiffies; | |
5561 | int nid = pgdat->node_id; | |
5562 | ||
5563 | if (WARN_ON_ONCE(!pgmap || !is_dev_zone(zone))) | |
5564 | return; | |
5565 | ||
5566 | /* | |
5567 | * The call to memmap_init_zone should have already taken care | |
5568 | * of the pages reserved for the memmap, so we can just jump to | |
5569 | * the end of that region and start processing the device pages. | |
5570 | */ | |
5571 | if (pgmap->altmap_valid) { | |
5572 | struct vmem_altmap *altmap = &pgmap->altmap; | |
5573 | ||
5574 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
5575 | size = end_pfn - start_pfn; | |
5576 | } | |
5577 | ||
5578 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
5579 | struct page *page = pfn_to_page(pfn); | |
5580 | ||
5581 | __init_single_page(page, pfn, zone_idx, nid); | |
5582 | ||
5583 | /* | |
5584 | * Mark page reserved as it will need to wait for onlining | |
5585 | * phase for it to be fully associated with a zone. | |
5586 | * | |
5587 | * We can use the non-atomic __set_bit operation for setting | |
5588 | * the flag as we are still initializing the pages. | |
5589 | */ | |
5590 | __SetPageReserved(page); | |
5591 | ||
5592 | /* | |
5593 | * ZONE_DEVICE pages union ->lru with a ->pgmap back | |
5594 | * pointer and hmm_data. It is a bug if a ZONE_DEVICE | |
5595 | * page is ever freed or placed on a driver-private list. | |
5596 | */ | |
5597 | page->pgmap = pgmap; | |
5598 | page->hmm_data = 0; | |
5599 | ||
5600 | /* | |
5601 | * Mark the block movable so that blocks are reserved for | |
5602 | * movable at startup. This will force kernel allocations | |
5603 | * to reserve their blocks rather than leaking throughout | |
5604 | * the address space during boot when many long-lived | |
5605 | * kernel allocations are made. | |
5606 | * | |
5607 | * bitmap is created for zone's valid pfn range. but memmap | |
5608 | * can be created for invalid pages (for alignment) | |
5609 | * check here not to call set_pageblock_migratetype() against | |
5610 | * pfn out of zone. | |
5611 | * | |
5612 | * Please note that MEMMAP_HOTPLUG path doesn't clear memmap | |
5613 | * because this is done early in sparse_add_one_section | |
5614 | */ | |
5615 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
5616 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5617 | cond_resched(); | |
5618 | } | |
5619 | } | |
5620 | ||
5621 | pr_info("%s initialised, %lu pages in %ums\n", dev_name(pgmap->dev), | |
5622 | size, jiffies_to_msecs(jiffies - start)); | |
5623 | } | |
5624 | ||
5625 | #endif | |
1e548deb | 5626 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 5627 | { |
7aeb09f9 | 5628 | unsigned int order, t; |
b2a0ac88 MG |
5629 | for_each_migratetype_order(order, t) { |
5630 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
5631 | zone->free_area[order].nr_free = 0; |
5632 | } | |
5633 | } | |
5634 | ||
5635 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
5636 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a99583e7 | 5637 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY, NULL) |
1da177e4 LT |
5638 | #endif |
5639 | ||
7cd2b0a3 | 5640 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 5641 | { |
3a6be87f | 5642 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
5643 | int batch; |
5644 | ||
5645 | /* | |
5646 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 5647 | * size of the zone. |
e7c8d5c9 | 5648 | */ |
b40da049 | 5649 | batch = zone->managed_pages / 1024; |
d8a759b5 AL |
5650 | /* But no more than a meg. */ |
5651 | if (batch * PAGE_SIZE > 1024 * 1024) | |
5652 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
5653 | batch /= 4; /* We effectively *= 4 below */ |
5654 | if (batch < 1) | |
5655 | batch = 1; | |
5656 | ||
5657 | /* | |
0ceaacc9 NP |
5658 | * Clamp the batch to a 2^n - 1 value. Having a power |
5659 | * of 2 value was found to be more likely to have | |
5660 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 5661 | * |
0ceaacc9 NP |
5662 | * For example if 2 tasks are alternately allocating |
5663 | * batches of pages, one task can end up with a lot | |
5664 | * of pages of one half of the possible page colors | |
5665 | * and the other with pages of the other colors. | |
e7c8d5c9 | 5666 | */ |
9155203a | 5667 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 5668 | |
e7c8d5c9 | 5669 | return batch; |
3a6be87f DH |
5670 | |
5671 | #else | |
5672 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5673 | * conditions. | |
5674 | * | |
5675 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5676 | * of contiguous memory as there's no hardware page translation to | |
5677 | * assemble apparent contiguous memory from discontiguous pages. | |
5678 | * | |
5679 | * Queueing large contiguous runs of pages for batching, however, | |
5680 | * causes the pages to actually be freed in smaller chunks. As there | |
5681 | * can be a significant delay between the individual batches being | |
5682 | * recycled, this leads to the once large chunks of space being | |
5683 | * fragmented and becoming unavailable for high-order allocations. | |
5684 | */ | |
5685 | return 0; | |
5686 | #endif | |
e7c8d5c9 CL |
5687 | } |
5688 | ||
8d7a8fa9 CS |
5689 | /* |
5690 | * pcp->high and pcp->batch values are related and dependent on one another: | |
5691 | * ->batch must never be higher then ->high. | |
5692 | * The following function updates them in a safe manner without read side | |
5693 | * locking. | |
5694 | * | |
5695 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
5696 | * those fields changing asynchronously (acording the the above rule). | |
5697 | * | |
5698 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5699 | * outside of boot time (or some other assurance that no concurrent updaters | |
5700 | * exist). | |
5701 | */ | |
5702 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
5703 | unsigned long batch) | |
5704 | { | |
5705 | /* start with a fail safe value for batch */ | |
5706 | pcp->batch = 1; | |
5707 | smp_wmb(); | |
5708 | ||
5709 | /* Update high, then batch, in order */ | |
5710 | pcp->high = high; | |
5711 | smp_wmb(); | |
5712 | ||
5713 | pcp->batch = batch; | |
5714 | } | |
5715 | ||
3664033c | 5716 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
5717 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5718 | { | |
8d7a8fa9 | 5719 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
5720 | } |
5721 | ||
88c90dbc | 5722 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
5723 | { |
5724 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 5725 | int migratetype; |
2caaad41 | 5726 | |
1c6fe946 MD |
5727 | memset(p, 0, sizeof(*p)); |
5728 | ||
3dfa5721 | 5729 | pcp = &p->pcp; |
2caaad41 | 5730 | pcp->count = 0; |
5f8dcc21 MG |
5731 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5732 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
5733 | } |
5734 | ||
88c90dbc CS |
5735 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5736 | { | |
5737 | pageset_init(p); | |
5738 | pageset_set_batch(p, batch); | |
5739 | } | |
5740 | ||
8ad4b1fb | 5741 | /* |
3664033c | 5742 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
5743 | * to the value high for the pageset p. |
5744 | */ | |
3664033c | 5745 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
5746 | unsigned long high) |
5747 | { | |
8d7a8fa9 CS |
5748 | unsigned long batch = max(1UL, high / 4); |
5749 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
5750 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 5751 | |
8d7a8fa9 | 5752 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
5753 | } |
5754 | ||
7cd2b0a3 DR |
5755 | static void pageset_set_high_and_batch(struct zone *zone, |
5756 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 5757 | { |
56cef2b8 | 5758 | if (percpu_pagelist_fraction) |
3664033c | 5759 | pageset_set_high(pcp, |
56cef2b8 CS |
5760 | (zone->managed_pages / |
5761 | percpu_pagelist_fraction)); | |
5762 | else | |
5763 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
5764 | } | |
5765 | ||
169f6c19 CS |
5766 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5767 | { | |
5768 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
5769 | ||
5770 | pageset_init(pcp); | |
5771 | pageset_set_high_and_batch(zone, pcp); | |
5772 | } | |
5773 | ||
72675e13 | 5774 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
5775 | { |
5776 | int cpu; | |
319774e2 | 5777 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5778 | for_each_possible_cpu(cpu) |
5779 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5780 | } |
5781 | ||
2caaad41 | 5782 | /* |
99dcc3e5 CL |
5783 | * Allocate per cpu pagesets and initialize them. |
5784 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5785 | */ |
99dcc3e5 | 5786 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5787 | { |
b4911ea2 | 5788 | struct pglist_data *pgdat; |
99dcc3e5 | 5789 | struct zone *zone; |
e7c8d5c9 | 5790 | |
319774e2 WF |
5791 | for_each_populated_zone(zone) |
5792 | setup_zone_pageset(zone); | |
b4911ea2 MG |
5793 | |
5794 | for_each_online_pgdat(pgdat) | |
5795 | pgdat->per_cpu_nodestats = | |
5796 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
5797 | } |
5798 | ||
c09b4240 | 5799 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5800 | { |
99dcc3e5 CL |
5801 | /* |
5802 | * per cpu subsystem is not up at this point. The following code | |
5803 | * relies on the ability of the linker to provide the | |
5804 | * offset of a (static) per cpu variable into the per cpu area. | |
5805 | */ | |
5806 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5807 | |
b38a8725 | 5808 | if (populated_zone(zone)) |
99dcc3e5 CL |
5809 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5810 | zone->name, zone->present_pages, | |
5811 | zone_batchsize(zone)); | |
ed8ece2e DH |
5812 | } |
5813 | ||
dc0bbf3b | 5814 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5815 | unsigned long zone_start_pfn, |
b171e409 | 5816 | unsigned long size) |
ed8ece2e DH |
5817 | { |
5818 | struct pglist_data *pgdat = zone->zone_pgdat; | |
9dcb8b68 | 5819 | |
ed8ece2e DH |
5820 | pgdat->nr_zones = zone_idx(zone) + 1; |
5821 | ||
ed8ece2e DH |
5822 | zone->zone_start_pfn = zone_start_pfn; |
5823 | ||
708614e6 MG |
5824 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5825 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5826 | pgdat->node_id, | |
5827 | (unsigned long)zone_idx(zone), | |
5828 | zone_start_pfn, (zone_start_pfn + size)); | |
5829 | ||
1e548deb | 5830 | zone_init_free_lists(zone); |
9dcb8b68 | 5831 | zone->initialized = 1; |
ed8ece2e DH |
5832 | } |
5833 | ||
0ee332c1 | 5834 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5835 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5836 | |
c713216d MG |
5837 | /* |
5838 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5839 | */ |
8a942fde MG |
5840 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5841 | struct mminit_pfnnid_cache *state) | |
c713216d | 5842 | { |
c13291a5 | 5843 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5844 | int nid; |
7c243c71 | 5845 | |
8a942fde MG |
5846 | if (state->last_start <= pfn && pfn < state->last_end) |
5847 | return state->last_nid; | |
c713216d | 5848 | |
e76b63f8 YL |
5849 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5850 | if (nid != -1) { | |
8a942fde MG |
5851 | state->last_start = start_pfn; |
5852 | state->last_end = end_pfn; | |
5853 | state->last_nid = nid; | |
e76b63f8 YL |
5854 | } |
5855 | ||
5856 | return nid; | |
c713216d MG |
5857 | } |
5858 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5859 | ||
c713216d | 5860 | /** |
6782832e | 5861 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5862 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5863 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5864 | * |
7d018176 ZZ |
5865 | * If an architecture guarantees that all ranges registered contain no holes |
5866 | * and may be freed, this this function may be used instead of calling | |
5867 | * memblock_free_early_nid() manually. | |
c713216d | 5868 | */ |
c13291a5 | 5869 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5870 | { |
c13291a5 TH |
5871 | unsigned long start_pfn, end_pfn; |
5872 | int i, this_nid; | |
edbe7d23 | 5873 | |
c13291a5 TH |
5874 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5875 | start_pfn = min(start_pfn, max_low_pfn); | |
5876 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5877 | |
c13291a5 | 5878 | if (start_pfn < end_pfn) |
6782832e SS |
5879 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5880 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5881 | this_nid); | |
edbe7d23 | 5882 | } |
edbe7d23 | 5883 | } |
edbe7d23 | 5884 | |
c713216d MG |
5885 | /** |
5886 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5887 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5888 | * |
7d018176 ZZ |
5889 | * If an architecture guarantees that all ranges registered contain no holes and may |
5890 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5891 | */ |
5892 | void __init sparse_memory_present_with_active_regions(int nid) | |
5893 | { | |
c13291a5 TH |
5894 | unsigned long start_pfn, end_pfn; |
5895 | int i, this_nid; | |
c713216d | 5896 | |
c13291a5 TH |
5897 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5898 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5899 | } |
5900 | ||
5901 | /** | |
5902 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5903 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5904 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5905 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5906 | * |
5907 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5908 | * provided by memblock_set_node(). If called for a node |
c713216d | 5909 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5910 | * PFNs will be 0. |
c713216d | 5911 | */ |
a3142c8e | 5912 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5913 | unsigned long *start_pfn, unsigned long *end_pfn) |
5914 | { | |
c13291a5 | 5915 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5916 | int i; |
c13291a5 | 5917 | |
c713216d MG |
5918 | *start_pfn = -1UL; |
5919 | *end_pfn = 0; | |
5920 | ||
c13291a5 TH |
5921 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5922 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5923 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5924 | } |
5925 | ||
633c0666 | 5926 | if (*start_pfn == -1UL) |
c713216d | 5927 | *start_pfn = 0; |
c713216d MG |
5928 | } |
5929 | ||
2a1e274a MG |
5930 | /* |
5931 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5932 | * assumption is made that zones within a node are ordered in monotonic | |
5933 | * increasing memory addresses so that the "highest" populated zone is used | |
5934 | */ | |
b69a7288 | 5935 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5936 | { |
5937 | int zone_index; | |
5938 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5939 | if (zone_index == ZONE_MOVABLE) | |
5940 | continue; | |
5941 | ||
5942 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5943 | arch_zone_lowest_possible_pfn[zone_index]) | |
5944 | break; | |
5945 | } | |
5946 | ||
5947 | VM_BUG_ON(zone_index == -1); | |
5948 | movable_zone = zone_index; | |
5949 | } | |
5950 | ||
5951 | /* | |
5952 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5953 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5954 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5955 | * in each node depending on the size of each node and how evenly kernelcore | |
5956 | * is distributed. This helper function adjusts the zone ranges | |
5957 | * provided by the architecture for a given node by using the end of the | |
5958 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5959 | * zones within a node are in order of monotonic increases memory addresses | |
5960 | */ | |
b69a7288 | 5961 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5962 | unsigned long zone_type, |
5963 | unsigned long node_start_pfn, | |
5964 | unsigned long node_end_pfn, | |
5965 | unsigned long *zone_start_pfn, | |
5966 | unsigned long *zone_end_pfn) | |
5967 | { | |
5968 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5969 | if (zone_movable_pfn[nid]) { | |
5970 | /* Size ZONE_MOVABLE */ | |
5971 | if (zone_type == ZONE_MOVABLE) { | |
5972 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5973 | *zone_end_pfn = min(node_end_pfn, | |
5974 | arch_zone_highest_possible_pfn[movable_zone]); | |
5975 | ||
e506b996 XQ |
5976 | /* Adjust for ZONE_MOVABLE starting within this range */ |
5977 | } else if (!mirrored_kernelcore && | |
5978 | *zone_start_pfn < zone_movable_pfn[nid] && | |
5979 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
5980 | *zone_end_pfn = zone_movable_pfn[nid]; | |
5981 | ||
2a1e274a MG |
5982 | /* Check if this whole range is within ZONE_MOVABLE */ |
5983 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5984 | *zone_start_pfn = *zone_end_pfn; | |
5985 | } | |
5986 | } | |
5987 | ||
c713216d MG |
5988 | /* |
5989 | * Return the number of pages a zone spans in a node, including holes | |
5990 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5991 | */ | |
6ea6e688 | 5992 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5993 | unsigned long zone_type, |
7960aedd ZY |
5994 | unsigned long node_start_pfn, |
5995 | unsigned long node_end_pfn, | |
d91749c1 TI |
5996 | unsigned long *zone_start_pfn, |
5997 | unsigned long *zone_end_pfn, | |
c713216d MG |
5998 | unsigned long *ignored) |
5999 | { | |
b5685e92 | 6000 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6001 | if (!node_start_pfn && !node_end_pfn) |
6002 | return 0; | |
6003 | ||
7960aedd | 6004 | /* Get the start and end of the zone */ |
d91749c1 TI |
6005 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
6006 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
6007 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6008 | node_start_pfn, node_end_pfn, | |
d91749c1 | 6009 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
6010 | |
6011 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 6012 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
6013 | return 0; |
6014 | ||
6015 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
6016 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
6017 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
6018 | |
6019 | /* Return the spanned pages */ | |
d91749c1 | 6020 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
6021 | } |
6022 | ||
6023 | /* | |
6024 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 6025 | * then all holes in the requested range will be accounted for. |
c713216d | 6026 | */ |
32996250 | 6027 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
6028 | unsigned long range_start_pfn, |
6029 | unsigned long range_end_pfn) | |
6030 | { | |
96e907d1 TH |
6031 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
6032 | unsigned long start_pfn, end_pfn; | |
6033 | int i; | |
c713216d | 6034 | |
96e907d1 TH |
6035 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6036 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6037 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6038 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 6039 | } |
96e907d1 | 6040 | return nr_absent; |
c713216d MG |
6041 | } |
6042 | ||
6043 | /** | |
6044 | * absent_pages_in_range - Return number of page frames in holes within a range | |
6045 | * @start_pfn: The start PFN to start searching for holes | |
6046 | * @end_pfn: The end PFN to stop searching for holes | |
6047 | * | |
88ca3b94 | 6048 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
6049 | */ |
6050 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
6051 | unsigned long end_pfn) | |
6052 | { | |
6053 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
6054 | } | |
6055 | ||
6056 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 6057 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 6058 | unsigned long zone_type, |
7960aedd ZY |
6059 | unsigned long node_start_pfn, |
6060 | unsigned long node_end_pfn, | |
c713216d MG |
6061 | unsigned long *ignored) |
6062 | { | |
96e907d1 TH |
6063 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6064 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 6065 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 6066 | unsigned long nr_absent; |
9c7cd687 | 6067 | |
b5685e92 | 6068 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6069 | if (!node_start_pfn && !node_end_pfn) |
6070 | return 0; | |
6071 | ||
96e907d1 TH |
6072 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6073 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 6074 | |
2a1e274a MG |
6075 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6076 | node_start_pfn, node_end_pfn, | |
6077 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
6078 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
6079 | ||
6080 | /* | |
6081 | * ZONE_MOVABLE handling. | |
6082 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6083 | * and vice versa. | |
6084 | */ | |
e506b996 XQ |
6085 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6086 | unsigned long start_pfn, end_pfn; | |
6087 | struct memblock_region *r; | |
6088 | ||
6089 | for_each_memblock(memory, r) { | |
6090 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
6091 | zone_start_pfn, zone_end_pfn); | |
6092 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6093 | zone_start_pfn, zone_end_pfn); | |
6094 | ||
6095 | if (zone_type == ZONE_MOVABLE && | |
6096 | memblock_is_mirror(r)) | |
6097 | nr_absent += end_pfn - start_pfn; | |
6098 | ||
6099 | if (zone_type == ZONE_NORMAL && | |
6100 | !memblock_is_mirror(r)) | |
6101 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6102 | } |
6103 | } | |
6104 | ||
6105 | return nr_absent; | |
c713216d | 6106 | } |
0e0b864e | 6107 | |
0ee332c1 | 6108 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 6109 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 6110 | unsigned long zone_type, |
7960aedd ZY |
6111 | unsigned long node_start_pfn, |
6112 | unsigned long node_end_pfn, | |
d91749c1 TI |
6113 | unsigned long *zone_start_pfn, |
6114 | unsigned long *zone_end_pfn, | |
c713216d MG |
6115 | unsigned long *zones_size) |
6116 | { | |
d91749c1 TI |
6117 | unsigned int zone; |
6118 | ||
6119 | *zone_start_pfn = node_start_pfn; | |
6120 | for (zone = 0; zone < zone_type; zone++) | |
6121 | *zone_start_pfn += zones_size[zone]; | |
6122 | ||
6123 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
6124 | ||
c713216d MG |
6125 | return zones_size[zone_type]; |
6126 | } | |
6127 | ||
6ea6e688 | 6128 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 6129 | unsigned long zone_type, |
7960aedd ZY |
6130 | unsigned long node_start_pfn, |
6131 | unsigned long node_end_pfn, | |
c713216d MG |
6132 | unsigned long *zholes_size) |
6133 | { | |
6134 | if (!zholes_size) | |
6135 | return 0; | |
6136 | ||
6137 | return zholes_size[zone_type]; | |
6138 | } | |
20e6926d | 6139 | |
0ee332c1 | 6140 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6141 | |
a3142c8e | 6142 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
6143 | unsigned long node_start_pfn, |
6144 | unsigned long node_end_pfn, | |
6145 | unsigned long *zones_size, | |
6146 | unsigned long *zholes_size) | |
c713216d | 6147 | { |
febd5949 | 6148 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6149 | enum zone_type i; |
6150 | ||
febd5949 GZ |
6151 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6152 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6153 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 6154 | unsigned long size, real_size; |
c713216d | 6155 | |
febd5949 GZ |
6156 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
6157 | node_start_pfn, | |
6158 | node_end_pfn, | |
d91749c1 TI |
6159 | &zone_start_pfn, |
6160 | &zone_end_pfn, | |
febd5949 GZ |
6161 | zones_size); |
6162 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
6163 | node_start_pfn, node_end_pfn, |
6164 | zholes_size); | |
d91749c1 TI |
6165 | if (size) |
6166 | zone->zone_start_pfn = zone_start_pfn; | |
6167 | else | |
6168 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
6169 | zone->spanned_pages = size; |
6170 | zone->present_pages = real_size; | |
6171 | ||
6172 | totalpages += size; | |
6173 | realtotalpages += real_size; | |
6174 | } | |
6175 | ||
6176 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
6177 | pgdat->node_present_pages = realtotalpages; |
6178 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
6179 | realtotalpages); | |
6180 | } | |
6181 | ||
835c134e MG |
6182 | #ifndef CONFIG_SPARSEMEM |
6183 | /* | |
6184 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
6185 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
6186 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
6187 | * round what is now in bits to nearest long in bits, then return it in |
6188 | * bytes. | |
6189 | */ | |
7c45512d | 6190 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
6191 | { |
6192 | unsigned long usemapsize; | |
6193 | ||
7c45512d | 6194 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
6195 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
6196 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
6197 | usemapsize *= NR_PAGEBLOCK_BITS; |
6198 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
6199 | ||
6200 | return usemapsize / 8; | |
6201 | } | |
6202 | ||
7cc2a959 | 6203 | static void __ref setup_usemap(struct pglist_data *pgdat, |
7c45512d LT |
6204 | struct zone *zone, |
6205 | unsigned long zone_start_pfn, | |
6206 | unsigned long zonesize) | |
835c134e | 6207 | { |
7c45512d | 6208 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 6209 | zone->pageblock_flags = NULL; |
58a01a45 | 6210 | if (usemapsize) |
6782832e SS |
6211 | zone->pageblock_flags = |
6212 | memblock_virt_alloc_node_nopanic(usemapsize, | |
6213 | pgdat->node_id); | |
835c134e MG |
6214 | } |
6215 | #else | |
7c45512d LT |
6216 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
6217 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
6218 | #endif /* CONFIG_SPARSEMEM */ |
6219 | ||
d9c23400 | 6220 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 6221 | |
d9c23400 | 6222 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 6223 | void __init set_pageblock_order(void) |
d9c23400 | 6224 | { |
955c1cd7 AM |
6225 | unsigned int order; |
6226 | ||
d9c23400 MG |
6227 | /* Check that pageblock_nr_pages has not already been setup */ |
6228 | if (pageblock_order) | |
6229 | return; | |
6230 | ||
955c1cd7 AM |
6231 | if (HPAGE_SHIFT > PAGE_SHIFT) |
6232 | order = HUGETLB_PAGE_ORDER; | |
6233 | else | |
6234 | order = MAX_ORDER - 1; | |
6235 | ||
d9c23400 MG |
6236 | /* |
6237 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
6238 | * This value may be variable depending on boot parameters on IA64 and |
6239 | * powerpc. | |
d9c23400 MG |
6240 | */ |
6241 | pageblock_order = order; | |
6242 | } | |
6243 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6244 | ||
ba72cb8c MG |
6245 | /* |
6246 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
6247 | * is unused as pageblock_order is set at compile-time. See |
6248 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
6249 | * the kernel config | |
ba72cb8c | 6250 | */ |
03e85f9d | 6251 | void __init set_pageblock_order(void) |
ba72cb8c | 6252 | { |
ba72cb8c | 6253 | } |
d9c23400 MG |
6254 | |
6255 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6256 | ||
03e85f9d | 6257 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 6258 | unsigned long present_pages) |
01cefaef JL |
6259 | { |
6260 | unsigned long pages = spanned_pages; | |
6261 | ||
6262 | /* | |
6263 | * Provide a more accurate estimation if there are holes within | |
6264 | * the zone and SPARSEMEM is in use. If there are holes within the | |
6265 | * zone, each populated memory region may cost us one or two extra | |
6266 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 6267 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
6268 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
6269 | */ | |
6270 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
6271 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
6272 | pages = present_pages; | |
6273 | ||
6274 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
6275 | } | |
6276 | ||
ace1db39 OS |
6277 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6278 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
6279 | { | |
6280 | spin_lock_init(&pgdat->split_queue_lock); | |
6281 | INIT_LIST_HEAD(&pgdat->split_queue); | |
6282 | pgdat->split_queue_len = 0; | |
6283 | } | |
6284 | #else | |
6285 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
6286 | #endif | |
6287 | ||
6288 | #ifdef CONFIG_COMPACTION | |
6289 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
6290 | { | |
6291 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6292 | } | |
6293 | #else | |
6294 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
6295 | #endif | |
6296 | ||
03e85f9d | 6297 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 6298 | { |
208d54e5 | 6299 | pgdat_resize_init(pgdat); |
ace1db39 | 6300 | |
ace1db39 OS |
6301 | pgdat_init_split_queue(pgdat); |
6302 | pgdat_init_kcompactd(pgdat); | |
6303 | ||
1da177e4 | 6304 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6305 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 6306 | |
eefa864b | 6307 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6308 | spin_lock_init(&pgdat->lru_lock); |
a9dd0a83 | 6309 | lruvec_init(node_lruvec(pgdat)); |
03e85f9d OS |
6310 | } |
6311 | ||
6312 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
6313 | unsigned long remaining_pages) | |
6314 | { | |
6315 | zone->managed_pages = remaining_pages; | |
6316 | zone_set_nid(zone, nid); | |
6317 | zone->name = zone_names[idx]; | |
6318 | zone->zone_pgdat = NODE_DATA(nid); | |
6319 | spin_lock_init(&zone->lock); | |
6320 | zone_seqlock_init(zone); | |
6321 | zone_pcp_init(zone); | |
6322 | } | |
6323 | ||
6324 | /* | |
6325 | * Set up the zone data structures | |
6326 | * - init pgdat internals | |
6327 | * - init all zones belonging to this node | |
6328 | * | |
6329 | * NOTE: this function is only called during memory hotplug | |
6330 | */ | |
6331 | #ifdef CONFIG_MEMORY_HOTPLUG | |
6332 | void __ref free_area_init_core_hotplug(int nid) | |
6333 | { | |
6334 | enum zone_type z; | |
6335 | pg_data_t *pgdat = NODE_DATA(nid); | |
6336 | ||
6337 | pgdat_init_internals(pgdat); | |
6338 | for (z = 0; z < MAX_NR_ZONES; z++) | |
6339 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
6340 | } | |
6341 | #endif | |
6342 | ||
6343 | /* | |
6344 | * Set up the zone data structures: | |
6345 | * - mark all pages reserved | |
6346 | * - mark all memory queues empty | |
6347 | * - clear the memory bitmaps | |
6348 | * | |
6349 | * NOTE: pgdat should get zeroed by caller. | |
6350 | * NOTE: this function is only called during early init. | |
6351 | */ | |
6352 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
6353 | { | |
6354 | enum zone_type j; | |
6355 | int nid = pgdat->node_id; | |
5f63b720 | 6356 | |
03e85f9d | 6357 | pgdat_init_internals(pgdat); |
385386cf JW |
6358 | pgdat->per_cpu_nodestats = &boot_nodestats; |
6359 | ||
1da177e4 LT |
6360 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6361 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 6362 | unsigned long size, freesize, memmap_pages; |
d91749c1 | 6363 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6364 | |
febd5949 | 6365 | size = zone->spanned_pages; |
e6943859 | 6366 | freesize = zone->present_pages; |
1da177e4 | 6367 | |
0e0b864e | 6368 | /* |
9feedc9d | 6369 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6370 | * is used by this zone for memmap. This affects the watermark |
6371 | * and per-cpu initialisations | |
6372 | */ | |
e6943859 | 6373 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
6374 | if (!is_highmem_idx(j)) { |
6375 | if (freesize >= memmap_pages) { | |
6376 | freesize -= memmap_pages; | |
6377 | if (memmap_pages) | |
6378 | printk(KERN_DEBUG | |
6379 | " %s zone: %lu pages used for memmap\n", | |
6380 | zone_names[j], memmap_pages); | |
6381 | } else | |
1170532b | 6382 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6383 | zone_names[j], memmap_pages, freesize); |
6384 | } | |
0e0b864e | 6385 | |
6267276f | 6386 | /* Account for reserved pages */ |
9feedc9d JL |
6387 | if (j == 0 && freesize > dma_reserve) { |
6388 | freesize -= dma_reserve; | |
d903ef9f | 6389 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6390 | zone_names[0], dma_reserve); |
0e0b864e MG |
6391 | } |
6392 | ||
98d2b0eb | 6393 | if (!is_highmem_idx(j)) |
9feedc9d | 6394 | nr_kernel_pages += freesize; |
01cefaef JL |
6395 | /* Charge for highmem memmap if there are enough kernel pages */ |
6396 | else if (nr_kernel_pages > memmap_pages * 2) | |
6397 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6398 | nr_all_pages += freesize; |
1da177e4 | 6399 | |
9feedc9d JL |
6400 | /* |
6401 | * Set an approximate value for lowmem here, it will be adjusted | |
6402 | * when the bootmem allocator frees pages into the buddy system. | |
6403 | * And all highmem pages will be managed by the buddy system. | |
6404 | */ | |
03e85f9d | 6405 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 6406 | |
d883c6cf | 6407 | if (!size) |
1da177e4 LT |
6408 | continue; |
6409 | ||
955c1cd7 | 6410 | set_pageblock_order(); |
d883c6cf JK |
6411 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
6412 | init_currently_empty_zone(zone, zone_start_pfn, size); | |
76cdd58e | 6413 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6414 | } |
6415 | } | |
6416 | ||
0cd842f9 | 6417 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 6418 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6419 | { |
b0aeba74 | 6420 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6421 | unsigned long __maybe_unused offset = 0; |
6422 | ||
1da177e4 LT |
6423 | /* Skip empty nodes */ |
6424 | if (!pgdat->node_spanned_pages) | |
6425 | return; | |
6426 | ||
b0aeba74 TL |
6427 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6428 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6429 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6430 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6431 | unsigned long size, end; |
d41dee36 AW |
6432 | struct page *map; |
6433 | ||
e984bb43 BP |
6434 | /* |
6435 | * The zone's endpoints aren't required to be MAX_ORDER | |
6436 | * aligned but the node_mem_map endpoints must be in order | |
6437 | * for the buddy allocator to function correctly. | |
6438 | */ | |
108bcc96 | 6439 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6440 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6441 | size = (end - start) * sizeof(struct page); | |
79375ea3 | 6442 | map = memblock_virt_alloc_node_nopanic(size, pgdat->node_id); |
a1c34a3b | 6443 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6444 | } |
0cd842f9 OS |
6445 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
6446 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
6447 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 6448 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6449 | /* |
6450 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6451 | */ | |
c713216d | 6452 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6453 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 6454 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 6455 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6456 | mem_map -= offset; |
0ee332c1 | 6457 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6458 | } |
1da177e4 LT |
6459 | #endif |
6460 | } | |
0cd842f9 OS |
6461 | #else |
6462 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
6463 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 6464 | |
0188dc98 OS |
6465 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
6466 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
6467 | { | |
6468 | /* | |
6469 | * We start only with one section of pages, more pages are added as | |
6470 | * needed until the rest of deferred pages are initialized. | |
6471 | */ | |
6472 | pgdat->static_init_pgcnt = min_t(unsigned long, PAGES_PER_SECTION, | |
6473 | pgdat->node_spanned_pages); | |
6474 | pgdat->first_deferred_pfn = ULONG_MAX; | |
6475 | } | |
6476 | #else | |
6477 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
6478 | #endif | |
6479 | ||
03e85f9d | 6480 | void __init free_area_init_node(int nid, unsigned long *zones_size, |
7cc2a959 PT |
6481 | unsigned long node_start_pfn, |
6482 | unsigned long *zholes_size) | |
1da177e4 | 6483 | { |
9109fb7b | 6484 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6485 | unsigned long start_pfn = 0; |
6486 | unsigned long end_pfn = 0; | |
9109fb7b | 6487 | |
88fdf75d | 6488 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 6489 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 6490 | |
1da177e4 LT |
6491 | pgdat->node_id = nid; |
6492 | pgdat->node_start_pfn = node_start_pfn; | |
75ef7184 | 6493 | pgdat->per_cpu_nodestats = NULL; |
7960aedd ZY |
6494 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6495 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 6496 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6497 | (u64)start_pfn << PAGE_SHIFT, |
6498 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
6499 | #else |
6500 | start_pfn = node_start_pfn; | |
7960aedd ZY |
6501 | #endif |
6502 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
6503 | zones_size, zholes_size); | |
1da177e4 LT |
6504 | |
6505 | alloc_node_mem_map(pgdat); | |
0188dc98 | 6506 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 6507 | |
7f3eb55b | 6508 | free_area_init_core(pgdat); |
1da177e4 LT |
6509 | } |
6510 | ||
d1b47a7c | 6511 | #if defined(CONFIG_HAVE_MEMBLOCK) && !defined(CONFIG_FLAT_NODE_MEM_MAP) |
a4a3ede2 PT |
6512 | /* |
6513 | * Only struct pages that are backed by physical memory are zeroed and | |
6514 | * initialized by going through __init_single_page(). But, there are some | |
6515 | * struct pages which are reserved in memblock allocator and their fields | |
6516 | * may be accessed (for example page_to_pfn() on some configuration accesses | |
6517 | * flags). We must explicitly zero those struct pages. | |
6518 | */ | |
03e85f9d | 6519 | void __init zero_resv_unavail(void) |
a4a3ede2 PT |
6520 | { |
6521 | phys_addr_t start, end; | |
6522 | unsigned long pfn; | |
6523 | u64 i, pgcnt; | |
6524 | ||
6525 | /* | |
6526 | * Loop through ranges that are reserved, but do not have reported | |
6527 | * physical memory backing. | |
6528 | */ | |
6529 | pgcnt = 0; | |
6530 | for_each_resv_unavail_range(i, &start, &end) { | |
6531 | for (pfn = PFN_DOWN(start); pfn < PFN_UP(end); pfn++) { | |
720e14eb PT |
6532 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { |
6533 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6534 | + pageblock_nr_pages - 1; | |
e8c24773 | 6535 | continue; |
720e14eb | 6536 | } |
a4a3ede2 PT |
6537 | mm_zero_struct_page(pfn_to_page(pfn)); |
6538 | pgcnt++; | |
6539 | } | |
6540 | } | |
6541 | ||
6542 | /* | |
6543 | * Struct pages that do not have backing memory. This could be because | |
6544 | * firmware is using some of this memory, or for some other reasons. | |
6545 | * Once memblock is changed so such behaviour is not allowed: i.e. | |
6546 | * list of "reserved" memory must be a subset of list of "memory", then | |
6547 | * this code can be removed. | |
6548 | */ | |
6549 | if (pgcnt) | |
6550 | pr_info("Reserved but unavailable: %lld pages", pgcnt); | |
6551 | } | |
d1b47a7c | 6552 | #endif /* CONFIG_HAVE_MEMBLOCK && !CONFIG_FLAT_NODE_MEM_MAP */ |
a4a3ede2 | 6553 | |
0ee332c1 | 6554 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
6555 | |
6556 | #if MAX_NUMNODES > 1 | |
6557 | /* | |
6558 | * Figure out the number of possible node ids. | |
6559 | */ | |
f9872caf | 6560 | void __init setup_nr_node_ids(void) |
418508c1 | 6561 | { |
904a9553 | 6562 | unsigned int highest; |
418508c1 | 6563 | |
904a9553 | 6564 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
6565 | nr_node_ids = highest + 1; |
6566 | } | |
418508c1 MS |
6567 | #endif |
6568 | ||
1e01979c TH |
6569 | /** |
6570 | * node_map_pfn_alignment - determine the maximum internode alignment | |
6571 | * | |
6572 | * This function should be called after node map is populated and sorted. | |
6573 | * It calculates the maximum power of two alignment which can distinguish | |
6574 | * all the nodes. | |
6575 | * | |
6576 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
6577 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
6578 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
6579 | * shifted, 1GiB is enough and this function will indicate so. | |
6580 | * | |
6581 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
6582 | * model has fine enough granularity to avoid incorrect mapping for the | |
6583 | * populated node map. | |
6584 | * | |
6585 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
6586 | * requirement (single node). | |
6587 | */ | |
6588 | unsigned long __init node_map_pfn_alignment(void) | |
6589 | { | |
6590 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 6591 | unsigned long start, end, mask; |
1e01979c | 6592 | int last_nid = -1; |
c13291a5 | 6593 | int i, nid; |
1e01979c | 6594 | |
c13291a5 | 6595 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
6596 | if (!start || last_nid < 0 || last_nid == nid) { |
6597 | last_nid = nid; | |
6598 | last_end = end; | |
6599 | continue; | |
6600 | } | |
6601 | ||
6602 | /* | |
6603 | * Start with a mask granular enough to pin-point to the | |
6604 | * start pfn and tick off bits one-by-one until it becomes | |
6605 | * too coarse to separate the current node from the last. | |
6606 | */ | |
6607 | mask = ~((1 << __ffs(start)) - 1); | |
6608 | while (mask && last_end <= (start & (mask << 1))) | |
6609 | mask <<= 1; | |
6610 | ||
6611 | /* accumulate all internode masks */ | |
6612 | accl_mask |= mask; | |
6613 | } | |
6614 | ||
6615 | /* convert mask to number of pages */ | |
6616 | return ~accl_mask + 1; | |
6617 | } | |
6618 | ||
a6af2bc3 | 6619 | /* Find the lowest pfn for a node */ |
b69a7288 | 6620 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 6621 | { |
a6af2bc3 | 6622 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
6623 | unsigned long start_pfn; |
6624 | int i; | |
1abbfb41 | 6625 | |
c13291a5 TH |
6626 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
6627 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 6628 | |
a6af2bc3 | 6629 | if (min_pfn == ULONG_MAX) { |
1170532b | 6630 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
6631 | return 0; |
6632 | } | |
6633 | ||
6634 | return min_pfn; | |
c713216d MG |
6635 | } |
6636 | ||
6637 | /** | |
6638 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
6639 | * | |
6640 | * It returns the minimum PFN based on information provided via | |
7d018176 | 6641 | * memblock_set_node(). |
c713216d MG |
6642 | */ |
6643 | unsigned long __init find_min_pfn_with_active_regions(void) | |
6644 | { | |
6645 | return find_min_pfn_for_node(MAX_NUMNODES); | |
6646 | } | |
6647 | ||
37b07e41 LS |
6648 | /* |
6649 | * early_calculate_totalpages() | |
6650 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 6651 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 6652 | */ |
484f51f8 | 6653 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 6654 | { |
7e63efef | 6655 | unsigned long totalpages = 0; |
c13291a5 TH |
6656 | unsigned long start_pfn, end_pfn; |
6657 | int i, nid; | |
6658 | ||
6659 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
6660 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 6661 | |
37b07e41 LS |
6662 | totalpages += pages; |
6663 | if (pages) | |
4b0ef1fe | 6664 | node_set_state(nid, N_MEMORY); |
37b07e41 | 6665 | } |
b8af2941 | 6666 | return totalpages; |
7e63efef MG |
6667 | } |
6668 | ||
2a1e274a MG |
6669 | /* |
6670 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
6671 | * is spread evenly between nodes as long as the nodes have enough | |
6672 | * memory. When they don't, some nodes will have more kernelcore than | |
6673 | * others | |
6674 | */ | |
b224ef85 | 6675 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
6676 | { |
6677 | int i, nid; | |
6678 | unsigned long usable_startpfn; | |
6679 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 6680 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 6681 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 6682 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 6683 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 6684 | struct memblock_region *r; |
b2f3eebe TC |
6685 | |
6686 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
6687 | find_usable_zone_for_movable(); | |
6688 | ||
6689 | /* | |
6690 | * If movable_node is specified, ignore kernelcore and movablecore | |
6691 | * options. | |
6692 | */ | |
6693 | if (movable_node_is_enabled()) { | |
136199f0 EM |
6694 | for_each_memblock(memory, r) { |
6695 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
6696 | continue; |
6697 | ||
136199f0 | 6698 | nid = r->nid; |
b2f3eebe | 6699 | |
136199f0 | 6700 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
6701 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6702 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6703 | usable_startpfn; | |
6704 | } | |
6705 | ||
6706 | goto out2; | |
6707 | } | |
2a1e274a | 6708 | |
342332e6 TI |
6709 | /* |
6710 | * If kernelcore=mirror is specified, ignore movablecore option | |
6711 | */ | |
6712 | if (mirrored_kernelcore) { | |
6713 | bool mem_below_4gb_not_mirrored = false; | |
6714 | ||
6715 | for_each_memblock(memory, r) { | |
6716 | if (memblock_is_mirror(r)) | |
6717 | continue; | |
6718 | ||
6719 | nid = r->nid; | |
6720 | ||
6721 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
6722 | ||
6723 | if (usable_startpfn < 0x100000) { | |
6724 | mem_below_4gb_not_mirrored = true; | |
6725 | continue; | |
6726 | } | |
6727 | ||
6728 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
6729 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6730 | usable_startpfn; | |
6731 | } | |
6732 | ||
6733 | if (mem_below_4gb_not_mirrored) | |
6734 | pr_warn("This configuration results in unmirrored kernel memory."); | |
6735 | ||
6736 | goto out2; | |
6737 | } | |
6738 | ||
7e63efef | 6739 | /* |
a5c6d650 DR |
6740 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
6741 | * amount of necessary memory. | |
6742 | */ | |
6743 | if (required_kernelcore_percent) | |
6744 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
6745 | 10000UL; | |
6746 | if (required_movablecore_percent) | |
6747 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
6748 | 10000UL; | |
6749 | ||
6750 | /* | |
6751 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
6752 | * kernelcore that corresponds so that memory usable for |
6753 | * any allocation type is evenly spread. If both kernelcore | |
6754 | * and movablecore are specified, then the value of kernelcore | |
6755 | * will be used for required_kernelcore if it's greater than | |
6756 | * what movablecore would have allowed. | |
6757 | */ | |
6758 | if (required_movablecore) { | |
7e63efef MG |
6759 | unsigned long corepages; |
6760 | ||
6761 | /* | |
6762 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
6763 | * was requested by the user | |
6764 | */ | |
6765 | required_movablecore = | |
6766 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 6767 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
6768 | corepages = totalpages - required_movablecore; |
6769 | ||
6770 | required_kernelcore = max(required_kernelcore, corepages); | |
6771 | } | |
6772 | ||
bde304bd XQ |
6773 | /* |
6774 | * If kernelcore was not specified or kernelcore size is larger | |
6775 | * than totalpages, there is no ZONE_MOVABLE. | |
6776 | */ | |
6777 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 6778 | goto out; |
2a1e274a MG |
6779 | |
6780 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
6781 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6782 | ||
6783 | restart: | |
6784 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
6785 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 6786 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
6787 | unsigned long start_pfn, end_pfn; |
6788 | ||
2a1e274a MG |
6789 | /* |
6790 | * Recalculate kernelcore_node if the division per node | |
6791 | * now exceeds what is necessary to satisfy the requested | |
6792 | * amount of memory for the kernel | |
6793 | */ | |
6794 | if (required_kernelcore < kernelcore_node) | |
6795 | kernelcore_node = required_kernelcore / usable_nodes; | |
6796 | ||
6797 | /* | |
6798 | * As the map is walked, we track how much memory is usable | |
6799 | * by the kernel using kernelcore_remaining. When it is | |
6800 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
6801 | */ | |
6802 | kernelcore_remaining = kernelcore_node; | |
6803 | ||
6804 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 6805 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
6806 | unsigned long size_pages; |
6807 | ||
c13291a5 | 6808 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
6809 | if (start_pfn >= end_pfn) |
6810 | continue; | |
6811 | ||
6812 | /* Account for what is only usable for kernelcore */ | |
6813 | if (start_pfn < usable_startpfn) { | |
6814 | unsigned long kernel_pages; | |
6815 | kernel_pages = min(end_pfn, usable_startpfn) | |
6816 | - start_pfn; | |
6817 | ||
6818 | kernelcore_remaining -= min(kernel_pages, | |
6819 | kernelcore_remaining); | |
6820 | required_kernelcore -= min(kernel_pages, | |
6821 | required_kernelcore); | |
6822 | ||
6823 | /* Continue if range is now fully accounted */ | |
6824 | if (end_pfn <= usable_startpfn) { | |
6825 | ||
6826 | /* | |
6827 | * Push zone_movable_pfn to the end so | |
6828 | * that if we have to rebalance | |
6829 | * kernelcore across nodes, we will | |
6830 | * not double account here | |
6831 | */ | |
6832 | zone_movable_pfn[nid] = end_pfn; | |
6833 | continue; | |
6834 | } | |
6835 | start_pfn = usable_startpfn; | |
6836 | } | |
6837 | ||
6838 | /* | |
6839 | * The usable PFN range for ZONE_MOVABLE is from | |
6840 | * start_pfn->end_pfn. Calculate size_pages as the | |
6841 | * number of pages used as kernelcore | |
6842 | */ | |
6843 | size_pages = end_pfn - start_pfn; | |
6844 | if (size_pages > kernelcore_remaining) | |
6845 | size_pages = kernelcore_remaining; | |
6846 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6847 | ||
6848 | /* | |
6849 | * Some kernelcore has been met, update counts and | |
6850 | * break if the kernelcore for this node has been | |
b8af2941 | 6851 | * satisfied |
2a1e274a MG |
6852 | */ |
6853 | required_kernelcore -= min(required_kernelcore, | |
6854 | size_pages); | |
6855 | kernelcore_remaining -= size_pages; | |
6856 | if (!kernelcore_remaining) | |
6857 | break; | |
6858 | } | |
6859 | } | |
6860 | ||
6861 | /* | |
6862 | * If there is still required_kernelcore, we do another pass with one | |
6863 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6864 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6865 | * satisfied |
2a1e274a MG |
6866 | */ |
6867 | usable_nodes--; | |
6868 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6869 | goto restart; | |
6870 | ||
b2f3eebe | 6871 | out2: |
2a1e274a MG |
6872 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6873 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6874 | zone_movable_pfn[nid] = | |
6875 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6876 | |
20e6926d | 6877 | out: |
66918dcd | 6878 | /* restore the node_state */ |
4b0ef1fe | 6879 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6880 | } |
6881 | ||
4b0ef1fe LJ |
6882 | /* Any regular or high memory on that node ? */ |
6883 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6884 | { |
37b07e41 LS |
6885 | enum zone_type zone_type; |
6886 | ||
4b0ef1fe | 6887 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 6888 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6889 | if (populated_zone(zone)) { |
7b0e0c0e OS |
6890 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
6891 | node_set_state(nid, N_HIGH_MEMORY); | |
6892 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 6893 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
6894 | break; |
6895 | } | |
37b07e41 | 6896 | } |
37b07e41 LS |
6897 | } |
6898 | ||
c713216d MG |
6899 | /** |
6900 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6901 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6902 | * |
6903 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6904 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6905 | * zone in each node and their holes is calculated. If the maximum PFN |
6906 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6907 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6908 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6909 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6910 | * at arch_max_dma_pfn. | |
6911 | */ | |
6912 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6913 | { | |
c13291a5 TH |
6914 | unsigned long start_pfn, end_pfn; |
6915 | int i, nid; | |
a6af2bc3 | 6916 | |
c713216d MG |
6917 | /* Record where the zone boundaries are */ |
6918 | memset(arch_zone_lowest_possible_pfn, 0, | |
6919 | sizeof(arch_zone_lowest_possible_pfn)); | |
6920 | memset(arch_zone_highest_possible_pfn, 0, | |
6921 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
6922 | |
6923 | start_pfn = find_min_pfn_with_active_regions(); | |
6924 | ||
6925 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6926 | if (i == ZONE_MOVABLE) |
6927 | continue; | |
90cae1fe OH |
6928 | |
6929 | end_pfn = max(max_zone_pfn[i], start_pfn); | |
6930 | arch_zone_lowest_possible_pfn[i] = start_pfn; | |
6931 | arch_zone_highest_possible_pfn[i] = end_pfn; | |
6932 | ||
6933 | start_pfn = end_pfn; | |
c713216d | 6934 | } |
2a1e274a MG |
6935 | |
6936 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6937 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6938 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6939 | |
c713216d | 6940 | /* Print out the zone ranges */ |
f88dfff5 | 6941 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6942 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6943 | if (i == ZONE_MOVABLE) | |
6944 | continue; | |
f88dfff5 | 6945 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6946 | if (arch_zone_lowest_possible_pfn[i] == |
6947 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6948 | pr_cont("empty\n"); |
72f0ba02 | 6949 | else |
8d29e18a JG |
6950 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6951 | (u64)arch_zone_lowest_possible_pfn[i] | |
6952 | << PAGE_SHIFT, | |
6953 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6954 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6955 | } |
6956 | ||
6957 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6958 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6959 | for (i = 0; i < MAX_NUMNODES; i++) { |
6960 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6961 | pr_info(" Node %d: %#018Lx\n", i, |
6962 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6963 | } |
c713216d | 6964 | |
f2d52fe5 | 6965 | /* Print out the early node map */ |
f88dfff5 | 6966 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6967 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6968 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6969 | (u64)start_pfn << PAGE_SHIFT, | |
6970 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6971 | |
6972 | /* Initialise every node */ | |
708614e6 | 6973 | mminit_verify_pageflags_layout(); |
8ef82866 | 6974 | setup_nr_node_ids(); |
e181ae0c | 6975 | zero_resv_unavail(); |
c713216d MG |
6976 | for_each_online_node(nid) { |
6977 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6978 | free_area_init_node(nid, NULL, |
c713216d | 6979 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6980 | |
6981 | /* Any memory on that node */ | |
6982 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6983 | node_set_state(nid, N_MEMORY); |
6984 | check_for_memory(pgdat, nid); | |
c713216d MG |
6985 | } |
6986 | } | |
2a1e274a | 6987 | |
a5c6d650 DR |
6988 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
6989 | unsigned long *percent) | |
2a1e274a MG |
6990 | { |
6991 | unsigned long long coremem; | |
a5c6d650 DR |
6992 | char *endptr; |
6993 | ||
2a1e274a MG |
6994 | if (!p) |
6995 | return -EINVAL; | |
6996 | ||
a5c6d650 DR |
6997 | /* Value may be a percentage of total memory, otherwise bytes */ |
6998 | coremem = simple_strtoull(p, &endptr, 0); | |
6999 | if (*endptr == '%') { | |
7000 | /* Paranoid check for percent values greater than 100 */ | |
7001 | WARN_ON(coremem > 100); | |
2a1e274a | 7002 | |
a5c6d650 DR |
7003 | *percent = coremem; |
7004 | } else { | |
7005 | coremem = memparse(p, &p); | |
7006 | /* Paranoid check that UL is enough for the coremem value */ | |
7007 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7008 | |
a5c6d650 DR |
7009 | *core = coremem >> PAGE_SHIFT; |
7010 | *percent = 0UL; | |
7011 | } | |
2a1e274a MG |
7012 | return 0; |
7013 | } | |
ed7ed365 | 7014 | |
7e63efef MG |
7015 | /* |
7016 | * kernelcore=size sets the amount of memory for use for allocations that | |
7017 | * cannot be reclaimed or migrated. | |
7018 | */ | |
7019 | static int __init cmdline_parse_kernelcore(char *p) | |
7020 | { | |
342332e6 TI |
7021 | /* parse kernelcore=mirror */ |
7022 | if (parse_option_str(p, "mirror")) { | |
7023 | mirrored_kernelcore = true; | |
7024 | return 0; | |
7025 | } | |
7026 | ||
a5c6d650 DR |
7027 | return cmdline_parse_core(p, &required_kernelcore, |
7028 | &required_kernelcore_percent); | |
7e63efef MG |
7029 | } |
7030 | ||
7031 | /* | |
7032 | * movablecore=size sets the amount of memory for use for allocations that | |
7033 | * can be reclaimed or migrated. | |
7034 | */ | |
7035 | static int __init cmdline_parse_movablecore(char *p) | |
7036 | { | |
a5c6d650 DR |
7037 | return cmdline_parse_core(p, &required_movablecore, |
7038 | &required_movablecore_percent); | |
7e63efef MG |
7039 | } |
7040 | ||
ed7ed365 | 7041 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7042 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7043 | |
0ee332c1 | 7044 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 7045 | |
c3d5f5f0 JL |
7046 | void adjust_managed_page_count(struct page *page, long count) |
7047 | { | |
7048 | spin_lock(&managed_page_count_lock); | |
7049 | page_zone(page)->managed_pages += count; | |
7050 | totalram_pages += count; | |
3dcc0571 JL |
7051 | #ifdef CONFIG_HIGHMEM |
7052 | if (PageHighMem(page)) | |
7053 | totalhigh_pages += count; | |
7054 | #endif | |
c3d5f5f0 JL |
7055 | spin_unlock(&managed_page_count_lock); |
7056 | } | |
3dcc0571 | 7057 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7058 | |
11199692 | 7059 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 7060 | { |
11199692 JL |
7061 | void *pos; |
7062 | unsigned long pages = 0; | |
69afade7 | 7063 | |
11199692 JL |
7064 | start = (void *)PAGE_ALIGN((unsigned long)start); |
7065 | end = (void *)((unsigned long)end & PAGE_MASK); | |
7066 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
7067 | struct page *page = virt_to_page(pos); |
7068 | void *direct_map_addr; | |
7069 | ||
7070 | /* | |
7071 | * 'direct_map_addr' might be different from 'pos' | |
7072 | * because some architectures' virt_to_page() | |
7073 | * work with aliases. Getting the direct map | |
7074 | * address ensures that we get a _writeable_ | |
7075 | * alias for the memset(). | |
7076 | */ | |
7077 | direct_map_addr = page_address(page); | |
dbe67df4 | 7078 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
7079 | memset(direct_map_addr, poison, PAGE_SIZE); |
7080 | ||
7081 | free_reserved_page(page); | |
69afade7 JL |
7082 | } |
7083 | ||
7084 | if (pages && s) | |
adb1fe9a JP |
7085 | pr_info("Freeing %s memory: %ldK\n", |
7086 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
7087 | |
7088 | return pages; | |
7089 | } | |
11199692 | 7090 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 7091 | |
cfa11e08 JL |
7092 | #ifdef CONFIG_HIGHMEM |
7093 | void free_highmem_page(struct page *page) | |
7094 | { | |
7095 | __free_reserved_page(page); | |
7096 | totalram_pages++; | |
7b4b2a0d | 7097 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
7098 | totalhigh_pages++; |
7099 | } | |
7100 | #endif | |
7101 | ||
7ee3d4e8 JL |
7102 | |
7103 | void __init mem_init_print_info(const char *str) | |
7104 | { | |
7105 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7106 | unsigned long init_code_size, init_data_size; | |
7107 | ||
7108 | physpages = get_num_physpages(); | |
7109 | codesize = _etext - _stext; | |
7110 | datasize = _edata - _sdata; | |
7111 | rosize = __end_rodata - __start_rodata; | |
7112 | bss_size = __bss_stop - __bss_start; | |
7113 | init_data_size = __init_end - __init_begin; | |
7114 | init_code_size = _einittext - _sinittext; | |
7115 | ||
7116 | /* | |
7117 | * Detect special cases and adjust section sizes accordingly: | |
7118 | * 1) .init.* may be embedded into .data sections | |
7119 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7120 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7121 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7122 | */ | |
7123 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7124 | do { \ |
7125 | if (start <= pos && pos < end && size > adj) \ | |
7126 | size -= adj; \ | |
7127 | } while (0) | |
7ee3d4e8 JL |
7128 | |
7129 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7130 | _sinittext, init_code_size); | |
7131 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7132 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7133 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7134 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7135 | ||
7136 | #undef adj_init_size | |
7137 | ||
756a025f | 7138 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7139 | #ifdef CONFIG_HIGHMEM |
756a025f | 7140 | ", %luK highmem" |
7ee3d4e8 | 7141 | #endif |
756a025f JP |
7142 | "%s%s)\n", |
7143 | nr_free_pages() << (PAGE_SHIFT - 10), | |
7144 | physpages << (PAGE_SHIFT - 10), | |
7145 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7146 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
7147 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
7148 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 7149 | #ifdef CONFIG_HIGHMEM |
756a025f | 7150 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7151 | #endif |
756a025f | 7152 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
7153 | } |
7154 | ||
0e0b864e | 7155 | /** |
88ca3b94 RD |
7156 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7157 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7158 | * |
013110a7 | 7159 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7160 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7161 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7162 | * function may optionally be used to account for unfreeable pages in the |
7163 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7164 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7165 | */ |
7166 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7167 | { | |
7168 | dma_reserve = new_dma_reserve; | |
7169 | } | |
7170 | ||
1da177e4 LT |
7171 | void __init free_area_init(unsigned long *zones_size) |
7172 | { | |
e181ae0c | 7173 | zero_resv_unavail(); |
9109fb7b | 7174 | free_area_init_node(0, zones_size, |
1da177e4 LT |
7175 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
7176 | } | |
1da177e4 | 7177 | |
005fd4bb | 7178 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7179 | { |
1da177e4 | 7180 | |
005fd4bb SAS |
7181 | lru_add_drain_cpu(cpu); |
7182 | drain_pages(cpu); | |
9f8f2172 | 7183 | |
005fd4bb SAS |
7184 | /* |
7185 | * Spill the event counters of the dead processor | |
7186 | * into the current processors event counters. | |
7187 | * This artificially elevates the count of the current | |
7188 | * processor. | |
7189 | */ | |
7190 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7191 | |
005fd4bb SAS |
7192 | /* |
7193 | * Zero the differential counters of the dead processor | |
7194 | * so that the vm statistics are consistent. | |
7195 | * | |
7196 | * This is only okay since the processor is dead and cannot | |
7197 | * race with what we are doing. | |
7198 | */ | |
7199 | cpu_vm_stats_fold(cpu); | |
7200 | return 0; | |
1da177e4 | 7201 | } |
1da177e4 LT |
7202 | |
7203 | void __init page_alloc_init(void) | |
7204 | { | |
005fd4bb SAS |
7205 | int ret; |
7206 | ||
7207 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, | |
7208 | "mm/page_alloc:dead", NULL, | |
7209 | page_alloc_cpu_dead); | |
7210 | WARN_ON(ret < 0); | |
1da177e4 LT |
7211 | } |
7212 | ||
cb45b0e9 | 7213 | /* |
34b10060 | 7214 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
7215 | * or min_free_kbytes changes. |
7216 | */ | |
7217 | static void calculate_totalreserve_pages(void) | |
7218 | { | |
7219 | struct pglist_data *pgdat; | |
7220 | unsigned long reserve_pages = 0; | |
2f6726e5 | 7221 | enum zone_type i, j; |
cb45b0e9 HA |
7222 | |
7223 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
7224 | |
7225 | pgdat->totalreserve_pages = 0; | |
7226 | ||
cb45b0e9 HA |
7227 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7228 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 7229 | long max = 0; |
cb45b0e9 HA |
7230 | |
7231 | /* Find valid and maximum lowmem_reserve in the zone */ | |
7232 | for (j = i; j < MAX_NR_ZONES; j++) { | |
7233 | if (zone->lowmem_reserve[j] > max) | |
7234 | max = zone->lowmem_reserve[j]; | |
7235 | } | |
7236 | ||
41858966 MG |
7237 | /* we treat the high watermark as reserved pages. */ |
7238 | max += high_wmark_pages(zone); | |
cb45b0e9 | 7239 | |
b40da049 JL |
7240 | if (max > zone->managed_pages) |
7241 | max = zone->managed_pages; | |
a8d01437 | 7242 | |
281e3726 | 7243 | pgdat->totalreserve_pages += max; |
a8d01437 | 7244 | |
cb45b0e9 HA |
7245 | reserve_pages += max; |
7246 | } | |
7247 | } | |
7248 | totalreserve_pages = reserve_pages; | |
7249 | } | |
7250 | ||
1da177e4 LT |
7251 | /* |
7252 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 7253 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
7254 | * has a correct pages reserved value, so an adequate number of |
7255 | * pages are left in the zone after a successful __alloc_pages(). | |
7256 | */ | |
7257 | static void setup_per_zone_lowmem_reserve(void) | |
7258 | { | |
7259 | struct pglist_data *pgdat; | |
2f6726e5 | 7260 | enum zone_type j, idx; |
1da177e4 | 7261 | |
ec936fc5 | 7262 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
7263 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7264 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 7265 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
7266 | |
7267 | zone->lowmem_reserve[j] = 0; | |
7268 | ||
2f6726e5 CL |
7269 | idx = j; |
7270 | while (idx) { | |
1da177e4 LT |
7271 | struct zone *lower_zone; |
7272 | ||
2f6726e5 | 7273 | idx--; |
1da177e4 | 7274 | lower_zone = pgdat->node_zones + idx; |
d3cda233 JK |
7275 | |
7276 | if (sysctl_lowmem_reserve_ratio[idx] < 1) { | |
7277 | sysctl_lowmem_reserve_ratio[idx] = 0; | |
7278 | lower_zone->lowmem_reserve[j] = 0; | |
7279 | } else { | |
7280 | lower_zone->lowmem_reserve[j] = | |
7281 | managed_pages / sysctl_lowmem_reserve_ratio[idx]; | |
7282 | } | |
b40da049 | 7283 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
7284 | } |
7285 | } | |
7286 | } | |
cb45b0e9 HA |
7287 | |
7288 | /* update totalreserve_pages */ | |
7289 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7290 | } |
7291 | ||
cfd3da1e | 7292 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
7293 | { |
7294 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
7295 | unsigned long lowmem_pages = 0; | |
7296 | struct zone *zone; | |
7297 | unsigned long flags; | |
7298 | ||
7299 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
7300 | for_each_zone(zone) { | |
7301 | if (!is_highmem(zone)) | |
b40da049 | 7302 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
7303 | } |
7304 | ||
7305 | for_each_zone(zone) { | |
ac924c60 AM |
7306 | u64 tmp; |
7307 | ||
1125b4e3 | 7308 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 7309 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 7310 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
7311 | if (is_highmem(zone)) { |
7312 | /* | |
669ed175 NP |
7313 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
7314 | * need highmem pages, so cap pages_min to a small | |
7315 | * value here. | |
7316 | * | |
41858966 | 7317 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 7318 | * deltas control asynch page reclaim, and so should |
669ed175 | 7319 | * not be capped for highmem. |
1da177e4 | 7320 | */ |
90ae8d67 | 7321 | unsigned long min_pages; |
1da177e4 | 7322 | |
b40da049 | 7323 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 7324 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 7325 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 7326 | } else { |
669ed175 NP |
7327 | /* |
7328 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
7329 | * proportionate to the zone's size. |
7330 | */ | |
41858966 | 7331 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
7332 | } |
7333 | ||
795ae7a0 JW |
7334 | /* |
7335 | * Set the kswapd watermarks distance according to the | |
7336 | * scale factor in proportion to available memory, but | |
7337 | * ensure a minimum size on small systems. | |
7338 | */ | |
7339 | tmp = max_t(u64, tmp >> 2, | |
7340 | mult_frac(zone->managed_pages, | |
7341 | watermark_scale_factor, 10000)); | |
7342 | ||
7343 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
7344 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 7345 | |
1125b4e3 | 7346 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 7347 | } |
cb45b0e9 HA |
7348 | |
7349 | /* update totalreserve_pages */ | |
7350 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7351 | } |
7352 | ||
cfd3da1e MG |
7353 | /** |
7354 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
7355 | * or when memory is hot-{added|removed} | |
7356 | * | |
7357 | * Ensures that the watermark[min,low,high] values for each zone are set | |
7358 | * correctly with respect to min_free_kbytes. | |
7359 | */ | |
7360 | void setup_per_zone_wmarks(void) | |
7361 | { | |
b93e0f32 MH |
7362 | static DEFINE_SPINLOCK(lock); |
7363 | ||
7364 | spin_lock(&lock); | |
cfd3da1e | 7365 | __setup_per_zone_wmarks(); |
b93e0f32 | 7366 | spin_unlock(&lock); |
cfd3da1e MG |
7367 | } |
7368 | ||
1da177e4 LT |
7369 | /* |
7370 | * Initialise min_free_kbytes. | |
7371 | * | |
7372 | * For small machines we want it small (128k min). For large machines | |
7373 | * we want it large (64MB max). But it is not linear, because network | |
7374 | * bandwidth does not increase linearly with machine size. We use | |
7375 | * | |
b8af2941 | 7376 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
7377 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
7378 | * | |
7379 | * which yields | |
7380 | * | |
7381 | * 16MB: 512k | |
7382 | * 32MB: 724k | |
7383 | * 64MB: 1024k | |
7384 | * 128MB: 1448k | |
7385 | * 256MB: 2048k | |
7386 | * 512MB: 2896k | |
7387 | * 1024MB: 4096k | |
7388 | * 2048MB: 5792k | |
7389 | * 4096MB: 8192k | |
7390 | * 8192MB: 11584k | |
7391 | * 16384MB: 16384k | |
7392 | */ | |
1b79acc9 | 7393 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
7394 | { |
7395 | unsigned long lowmem_kbytes; | |
5f12733e | 7396 | int new_min_free_kbytes; |
1da177e4 LT |
7397 | |
7398 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
7399 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
7400 | ||
7401 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
7402 | min_free_kbytes = new_min_free_kbytes; | |
7403 | if (min_free_kbytes < 128) | |
7404 | min_free_kbytes = 128; | |
7405 | if (min_free_kbytes > 65536) | |
7406 | min_free_kbytes = 65536; | |
7407 | } else { | |
7408 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
7409 | new_min_free_kbytes, user_min_free_kbytes); | |
7410 | } | |
bc75d33f | 7411 | setup_per_zone_wmarks(); |
a6cccdc3 | 7412 | refresh_zone_stat_thresholds(); |
1da177e4 | 7413 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
7414 | |
7415 | #ifdef CONFIG_NUMA | |
7416 | setup_min_unmapped_ratio(); | |
7417 | setup_min_slab_ratio(); | |
7418 | #endif | |
7419 | ||
1da177e4 LT |
7420 | return 0; |
7421 | } | |
bc22af74 | 7422 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
7423 | |
7424 | /* | |
b8af2941 | 7425 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
7426 | * that we can call two helper functions whenever min_free_kbytes |
7427 | * changes. | |
7428 | */ | |
cccad5b9 | 7429 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7430 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7431 | { |
da8c757b HP |
7432 | int rc; |
7433 | ||
7434 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7435 | if (rc) | |
7436 | return rc; | |
7437 | ||
5f12733e MH |
7438 | if (write) { |
7439 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7440 | setup_per_zone_wmarks(); |
5f12733e | 7441 | } |
1da177e4 LT |
7442 | return 0; |
7443 | } | |
7444 | ||
795ae7a0 JW |
7445 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
7446 | void __user *buffer, size_t *length, loff_t *ppos) | |
7447 | { | |
7448 | int rc; | |
7449 | ||
7450 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7451 | if (rc) | |
7452 | return rc; | |
7453 | ||
7454 | if (write) | |
7455 | setup_per_zone_wmarks(); | |
7456 | ||
7457 | return 0; | |
7458 | } | |
7459 | ||
9614634f | 7460 | #ifdef CONFIG_NUMA |
6423aa81 | 7461 | static void setup_min_unmapped_ratio(void) |
9614634f | 7462 | { |
6423aa81 | 7463 | pg_data_t *pgdat; |
9614634f | 7464 | struct zone *zone; |
9614634f | 7465 | |
a5f5f91d | 7466 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7467 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7468 | |
9614634f | 7469 | for_each_zone(zone) |
a5f5f91d | 7470 | zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages * |
9614634f | 7471 | sysctl_min_unmapped_ratio) / 100; |
9614634f | 7472 | } |
0ff38490 | 7473 | |
6423aa81 JK |
7474 | |
7475 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 7476 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7477 | { |
0ff38490 CL |
7478 | int rc; |
7479 | ||
8d65af78 | 7480 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7481 | if (rc) |
7482 | return rc; | |
7483 | ||
6423aa81 JK |
7484 | setup_min_unmapped_ratio(); |
7485 | ||
7486 | return 0; | |
7487 | } | |
7488 | ||
7489 | static void setup_min_slab_ratio(void) | |
7490 | { | |
7491 | pg_data_t *pgdat; | |
7492 | struct zone *zone; | |
7493 | ||
a5f5f91d MG |
7494 | for_each_online_pgdat(pgdat) |
7495 | pgdat->min_slab_pages = 0; | |
7496 | ||
0ff38490 | 7497 | for_each_zone(zone) |
a5f5f91d | 7498 | zone->zone_pgdat->min_slab_pages += (zone->managed_pages * |
0ff38490 | 7499 | sysctl_min_slab_ratio) / 100; |
6423aa81 JK |
7500 | } |
7501 | ||
7502 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
7503 | void __user *buffer, size_t *length, loff_t *ppos) | |
7504 | { | |
7505 | int rc; | |
7506 | ||
7507 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7508 | if (rc) | |
7509 | return rc; | |
7510 | ||
7511 | setup_min_slab_ratio(); | |
7512 | ||
0ff38490 CL |
7513 | return 0; |
7514 | } | |
9614634f CL |
7515 | #endif |
7516 | ||
1da177e4 LT |
7517 | /* |
7518 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7519 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7520 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7521 | * | |
7522 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7523 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7524 | * if in function of the boot time zone sizes. |
7525 | */ | |
cccad5b9 | 7526 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7527 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7528 | { |
8d65af78 | 7529 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
7530 | setup_per_zone_lowmem_reserve(); |
7531 | return 0; | |
7532 | } | |
7533 | ||
8ad4b1fb RS |
7534 | /* |
7535 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
7536 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
7537 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 7538 | */ |
cccad5b9 | 7539 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7540 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
7541 | { |
7542 | struct zone *zone; | |
7cd2b0a3 | 7543 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
7544 | int ret; |
7545 | ||
7cd2b0a3 DR |
7546 | mutex_lock(&pcp_batch_high_lock); |
7547 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
7548 | ||
8d65af78 | 7549 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
7550 | if (!write || ret < 0) |
7551 | goto out; | |
7552 | ||
7553 | /* Sanity checking to avoid pcp imbalance */ | |
7554 | if (percpu_pagelist_fraction && | |
7555 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
7556 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
7557 | ret = -EINVAL; | |
7558 | goto out; | |
7559 | } | |
7560 | ||
7561 | /* No change? */ | |
7562 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
7563 | goto out; | |
c8e251fa | 7564 | |
364df0eb | 7565 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
7566 | unsigned int cpu; |
7567 | ||
22a7f12b | 7568 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
7569 | pageset_set_high_and_batch(zone, |
7570 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 7571 | } |
7cd2b0a3 | 7572 | out: |
c8e251fa | 7573 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 7574 | return ret; |
8ad4b1fb RS |
7575 | } |
7576 | ||
a9919c79 | 7577 | #ifdef CONFIG_NUMA |
f034b5d4 | 7578 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 7579 | |
1da177e4 LT |
7580 | static int __init set_hashdist(char *str) |
7581 | { | |
7582 | if (!str) | |
7583 | return 0; | |
7584 | hashdist = simple_strtoul(str, &str, 0); | |
7585 | return 1; | |
7586 | } | |
7587 | __setup("hashdist=", set_hashdist); | |
7588 | #endif | |
7589 | ||
f6f34b43 SD |
7590 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
7591 | /* | |
7592 | * Returns the number of pages that arch has reserved but | |
7593 | * is not known to alloc_large_system_hash(). | |
7594 | */ | |
7595 | static unsigned long __init arch_reserved_kernel_pages(void) | |
7596 | { | |
7597 | return 0; | |
7598 | } | |
7599 | #endif | |
7600 | ||
9017217b PT |
7601 | /* |
7602 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
7603 | * machines. As memory size is increased the scale is also increased but at | |
7604 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
7605 | * quadruples the scale is increased by one, which means the size of hash table | |
7606 | * only doubles, instead of quadrupling as well. | |
7607 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
7608 | * makes sense, it is disabled on such platforms. | |
7609 | */ | |
7610 | #if __BITS_PER_LONG > 32 | |
7611 | #define ADAPT_SCALE_BASE (64ul << 30) | |
7612 | #define ADAPT_SCALE_SHIFT 2 | |
7613 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
7614 | #endif | |
7615 | ||
1da177e4 LT |
7616 | /* |
7617 | * allocate a large system hash table from bootmem | |
7618 | * - it is assumed that the hash table must contain an exact power-of-2 | |
7619 | * quantity of entries | |
7620 | * - limit is the number of hash buckets, not the total allocation size | |
7621 | */ | |
7622 | void *__init alloc_large_system_hash(const char *tablename, | |
7623 | unsigned long bucketsize, | |
7624 | unsigned long numentries, | |
7625 | int scale, | |
7626 | int flags, | |
7627 | unsigned int *_hash_shift, | |
7628 | unsigned int *_hash_mask, | |
31fe62b9 TB |
7629 | unsigned long low_limit, |
7630 | unsigned long high_limit) | |
1da177e4 | 7631 | { |
31fe62b9 | 7632 | unsigned long long max = high_limit; |
1da177e4 LT |
7633 | unsigned long log2qty, size; |
7634 | void *table = NULL; | |
3749a8f0 | 7635 | gfp_t gfp_flags; |
1da177e4 LT |
7636 | |
7637 | /* allow the kernel cmdline to have a say */ | |
7638 | if (!numentries) { | |
7639 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 7640 | numentries = nr_kernel_pages; |
f6f34b43 | 7641 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
7642 | |
7643 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
7644 | if (PAGE_SHIFT < 20) | |
7645 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 7646 | |
9017217b PT |
7647 | #if __BITS_PER_LONG > 32 |
7648 | if (!high_limit) { | |
7649 | unsigned long adapt; | |
7650 | ||
7651 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
7652 | adapt <<= ADAPT_SCALE_SHIFT) | |
7653 | scale++; | |
7654 | } | |
7655 | #endif | |
7656 | ||
1da177e4 LT |
7657 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
7658 | if (scale > PAGE_SHIFT) | |
7659 | numentries >>= (scale - PAGE_SHIFT); | |
7660 | else | |
7661 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
7662 | |
7663 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
7664 | if (unlikely(flags & HASH_SMALL)) { |
7665 | /* Makes no sense without HASH_EARLY */ | |
7666 | WARN_ON(!(flags & HASH_EARLY)); | |
7667 | if (!(numentries >> *_hash_shift)) { | |
7668 | numentries = 1UL << *_hash_shift; | |
7669 | BUG_ON(!numentries); | |
7670 | } | |
7671 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 7672 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 7673 | } |
6e692ed3 | 7674 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
7675 | |
7676 | /* limit allocation size to 1/16 total memory by default */ | |
7677 | if (max == 0) { | |
7678 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
7679 | do_div(max, bucketsize); | |
7680 | } | |
074b8517 | 7681 | max = min(max, 0x80000000ULL); |
1da177e4 | 7682 | |
31fe62b9 TB |
7683 | if (numentries < low_limit) |
7684 | numentries = low_limit; | |
1da177e4 LT |
7685 | if (numentries > max) |
7686 | numentries = max; | |
7687 | ||
f0d1b0b3 | 7688 | log2qty = ilog2(numentries); |
1da177e4 | 7689 | |
3749a8f0 | 7690 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 LT |
7691 | do { |
7692 | size = bucketsize << log2qty; | |
ea1f5f37 PT |
7693 | if (flags & HASH_EARLY) { |
7694 | if (flags & HASH_ZERO) | |
7695 | table = memblock_virt_alloc_nopanic(size, 0); | |
7696 | else | |
7697 | table = memblock_virt_alloc_raw(size, 0); | |
7698 | } else if (hashdist) { | |
3749a8f0 | 7699 | table = __vmalloc(size, gfp_flags, PAGE_KERNEL); |
ea1f5f37 | 7700 | } else { |
1037b83b ED |
7701 | /* |
7702 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
7703 | * some pages at the end of hash table which |
7704 | * alloc_pages_exact() automatically does | |
1037b83b | 7705 | */ |
264ef8a9 | 7706 | if (get_order(size) < MAX_ORDER) { |
3749a8f0 PT |
7707 | table = alloc_pages_exact(size, gfp_flags); |
7708 | kmemleak_alloc(table, size, 1, gfp_flags); | |
264ef8a9 | 7709 | } |
1da177e4 LT |
7710 | } |
7711 | } while (!table && size > PAGE_SIZE && --log2qty); | |
7712 | ||
7713 | if (!table) | |
7714 | panic("Failed to allocate %s hash table\n", tablename); | |
7715 | ||
1170532b JP |
7716 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
7717 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
7718 | |
7719 | if (_hash_shift) | |
7720 | *_hash_shift = log2qty; | |
7721 | if (_hash_mask) | |
7722 | *_hash_mask = (1 << log2qty) - 1; | |
7723 | ||
7724 | return table; | |
7725 | } | |
a117e66e | 7726 | |
a5d76b54 | 7727 | /* |
80934513 MK |
7728 | * This function checks whether pageblock includes unmovable pages or not. |
7729 | * If @count is not zero, it is okay to include less @count unmovable pages | |
7730 | * | |
b8af2941 | 7731 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
7732 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
7733 | * check without lock_page also may miss some movable non-lru pages at | |
7734 | * race condition. So you can't expect this function should be exact. | |
a5d76b54 | 7735 | */ |
b023f468 | 7736 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
4da2ce25 | 7737 | int migratetype, |
b023f468 | 7738 | bool skip_hwpoisoned_pages) |
49ac8255 KH |
7739 | { |
7740 | unsigned long pfn, iter, found; | |
47118af0 | 7741 | |
49ac8255 | 7742 | /* |
15c30bc0 MH |
7743 | * TODO we could make this much more efficient by not checking every |
7744 | * page in the range if we know all of them are in MOVABLE_ZONE and | |
7745 | * that the movable zone guarantees that pages are migratable but | |
7746 | * the later is not the case right now unfortunatelly. E.g. movablecore | |
7747 | * can still lead to having bootmem allocations in zone_movable. | |
49ac8255 | 7748 | */ |
49ac8255 | 7749 | |
4da2ce25 MH |
7750 | /* |
7751 | * CMA allocations (alloc_contig_range) really need to mark isolate | |
7752 | * CMA pageblocks even when they are not movable in fact so consider | |
7753 | * them movable here. | |
7754 | */ | |
7755 | if (is_migrate_cma(migratetype) && | |
7756 | is_migrate_cma(get_pageblock_migratetype(page))) | |
7757 | return false; | |
7758 | ||
49ac8255 KH |
7759 | pfn = page_to_pfn(page); |
7760 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
7761 | unsigned long check = pfn + iter; | |
7762 | ||
29723fcc | 7763 | if (!pfn_valid_within(check)) |
49ac8255 | 7764 | continue; |
29723fcc | 7765 | |
49ac8255 | 7766 | page = pfn_to_page(check); |
c8721bbb | 7767 | |
d7ab3672 | 7768 | if (PageReserved(page)) |
15c30bc0 | 7769 | goto unmovable; |
d7ab3672 | 7770 | |
c8721bbb NH |
7771 | /* |
7772 | * Hugepages are not in LRU lists, but they're movable. | |
7773 | * We need not scan over tail pages bacause we don't | |
7774 | * handle each tail page individually in migration. | |
7775 | */ | |
7776 | if (PageHuge(page)) { | |
464c7ffb AK |
7777 | |
7778 | if (!hugepage_migration_supported(page_hstate(page))) | |
7779 | goto unmovable; | |
7780 | ||
c8721bbb NH |
7781 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; |
7782 | continue; | |
7783 | } | |
7784 | ||
97d255c8 MK |
7785 | /* |
7786 | * We can't use page_count without pin a page | |
7787 | * because another CPU can free compound page. | |
7788 | * This check already skips compound tails of THP | |
0139aa7b | 7789 | * because their page->_refcount is zero at all time. |
97d255c8 | 7790 | */ |
fe896d18 | 7791 | if (!page_ref_count(page)) { |
49ac8255 KH |
7792 | if (PageBuddy(page)) |
7793 | iter += (1 << page_order(page)) - 1; | |
7794 | continue; | |
7795 | } | |
97d255c8 | 7796 | |
b023f468 WC |
7797 | /* |
7798 | * The HWPoisoned page may be not in buddy system, and | |
7799 | * page_count() is not 0. | |
7800 | */ | |
7801 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
7802 | continue; | |
7803 | ||
0efadf48 YX |
7804 | if (__PageMovable(page)) |
7805 | continue; | |
7806 | ||
49ac8255 KH |
7807 | if (!PageLRU(page)) |
7808 | found++; | |
7809 | /* | |
6b4f7799 JW |
7810 | * If there are RECLAIMABLE pages, we need to check |
7811 | * it. But now, memory offline itself doesn't call | |
7812 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
7813 | */ |
7814 | /* | |
7815 | * If the page is not RAM, page_count()should be 0. | |
7816 | * we don't need more check. This is an _used_ not-movable page. | |
7817 | * | |
7818 | * The problematic thing here is PG_reserved pages. PG_reserved | |
7819 | * is set to both of a memory hole page and a _used_ kernel | |
7820 | * page at boot. | |
7821 | */ | |
7822 | if (found > count) | |
15c30bc0 | 7823 | goto unmovable; |
49ac8255 | 7824 | } |
80934513 | 7825 | return false; |
15c30bc0 MH |
7826 | unmovable: |
7827 | WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE); | |
7828 | return true; | |
49ac8255 KH |
7829 | } |
7830 | ||
080fe206 | 7831 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7832 | |
7833 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7834 | { | |
7835 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7836 | pageblock_nr_pages) - 1); | |
7837 | } | |
7838 | ||
7839 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7840 | { | |
7841 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7842 | pageblock_nr_pages)); | |
7843 | } | |
7844 | ||
041d3a8c | 7845 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7846 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7847 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7848 | { |
7849 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7850 | unsigned long nr_reclaimed; |
041d3a8c MN |
7851 | unsigned long pfn = start; |
7852 | unsigned int tries = 0; | |
7853 | int ret = 0; | |
7854 | ||
be49a6e1 | 7855 | migrate_prep(); |
041d3a8c | 7856 | |
bb13ffeb | 7857 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7858 | if (fatal_signal_pending(current)) { |
7859 | ret = -EINTR; | |
7860 | break; | |
7861 | } | |
7862 | ||
bb13ffeb MG |
7863 | if (list_empty(&cc->migratepages)) { |
7864 | cc->nr_migratepages = 0; | |
edc2ca61 | 7865 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7866 | if (!pfn) { |
7867 | ret = -EINTR; | |
7868 | break; | |
7869 | } | |
7870 | tries = 0; | |
7871 | } else if (++tries == 5) { | |
7872 | ret = ret < 0 ? ret : -EBUSY; | |
7873 | break; | |
7874 | } | |
7875 | ||
beb51eaa MK |
7876 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7877 | &cc->migratepages); | |
7878 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7879 | |
9c620e2b | 7880 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
31025351 | 7881 | NULL, 0, cc->mode, MR_CONTIG_RANGE); |
041d3a8c | 7882 | } |
2a6f5124 SP |
7883 | if (ret < 0) { |
7884 | putback_movable_pages(&cc->migratepages); | |
7885 | return ret; | |
7886 | } | |
7887 | return 0; | |
041d3a8c MN |
7888 | } |
7889 | ||
7890 | /** | |
7891 | * alloc_contig_range() -- tries to allocate given range of pages | |
7892 | * @start: start PFN to allocate | |
7893 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7894 | * @migratetype: migratetype of the underlaying pageblocks (either |
7895 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7896 | * in range must have the same migratetype and it must | |
7897 | * be either of the two. | |
ca96b625 | 7898 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
7899 | * |
7900 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 7901 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 7902 | * |
2c7452a0 MK |
7903 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
7904 | * pageblocks in the range. Once isolated, the pageblocks should not | |
7905 | * be modified by others. | |
041d3a8c MN |
7906 | * |
7907 | * Returns zero on success or negative error code. On success all | |
7908 | * pages which PFN is in [start, end) are allocated for the caller and | |
7909 | * need to be freed with free_contig_range(). | |
7910 | */ | |
0815f3d8 | 7911 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 7912 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 7913 | { |
041d3a8c | 7914 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7915 | unsigned int order; |
7916 | int ret = 0; | |
041d3a8c | 7917 | |
bb13ffeb MG |
7918 | struct compact_control cc = { |
7919 | .nr_migratepages = 0, | |
7920 | .order = -1, | |
7921 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7922 | .mode = MIGRATE_SYNC, |
bb13ffeb | 7923 | .ignore_skip_hint = true, |
2583d671 | 7924 | .no_set_skip_hint = true, |
7dea19f9 | 7925 | .gfp_mask = current_gfp_context(gfp_mask), |
bb13ffeb MG |
7926 | }; |
7927 | INIT_LIST_HEAD(&cc.migratepages); | |
7928 | ||
041d3a8c MN |
7929 | /* |
7930 | * What we do here is we mark all pageblocks in range as | |
7931 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7932 | * have different sizes, and due to the way page allocator | |
7933 | * work, we align the range to biggest of the two pages so | |
7934 | * that page allocator won't try to merge buddies from | |
7935 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7936 | * other migration type. | |
7937 | * | |
7938 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7939 | * migrate the pages from an unaligned range (ie. pages that | |
7940 | * we are interested in). This will put all the pages in | |
7941 | * range back to page allocator as MIGRATE_ISOLATE. | |
7942 | * | |
7943 | * When this is done, we take the pages in range from page | |
7944 | * allocator removing them from the buddy system. This way | |
7945 | * page allocator will never consider using them. | |
7946 | * | |
7947 | * This lets us mark the pageblocks back as | |
7948 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7949 | * aligned range but not in the unaligned, original range are | |
7950 | * put back to page allocator so that buddy can use them. | |
7951 | */ | |
7952 | ||
7953 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7954 | pfn_max_align_up(end), migratetype, |
7955 | false); | |
041d3a8c | 7956 | if (ret) |
86a595f9 | 7957 | return ret; |
041d3a8c | 7958 | |
8ef5849f JK |
7959 | /* |
7960 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
7961 | * So, just fall through. test_pages_isolated() has a tracepoint |
7962 | * which will report the busy page. | |
7963 | * | |
7964 | * It is possible that busy pages could become available before | |
7965 | * the call to test_pages_isolated, and the range will actually be | |
7966 | * allocated. So, if we fall through be sure to clear ret so that | |
7967 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 7968 | */ |
bb13ffeb | 7969 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7970 | if (ret && ret != -EBUSY) |
041d3a8c | 7971 | goto done; |
63cd4489 | 7972 | ret =0; |
041d3a8c MN |
7973 | |
7974 | /* | |
7975 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7976 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7977 | * more, all pages in [start, end) are free in page allocator. | |
7978 | * What we are going to do is to allocate all pages from | |
7979 | * [start, end) (that is remove them from page allocator). | |
7980 | * | |
7981 | * The only problem is that pages at the beginning and at the | |
7982 | * end of interesting range may be not aligned with pages that | |
7983 | * page allocator holds, ie. they can be part of higher order | |
7984 | * pages. Because of this, we reserve the bigger range and | |
7985 | * once this is done free the pages we are not interested in. | |
7986 | * | |
7987 | * We don't have to hold zone->lock here because the pages are | |
7988 | * isolated thus they won't get removed from buddy. | |
7989 | */ | |
7990 | ||
7991 | lru_add_drain_all(); | |
510f5507 | 7992 | drain_all_pages(cc.zone); |
041d3a8c MN |
7993 | |
7994 | order = 0; | |
7995 | outer_start = start; | |
7996 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7997 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7998 | outer_start = start; |
7999 | break; | |
041d3a8c MN |
8000 | } |
8001 | outer_start &= ~0UL << order; | |
8002 | } | |
8003 | ||
8ef5849f JK |
8004 | if (outer_start != start) { |
8005 | order = page_order(pfn_to_page(outer_start)); | |
8006 | ||
8007 | /* | |
8008 | * outer_start page could be small order buddy page and | |
8009 | * it doesn't include start page. Adjust outer_start | |
8010 | * in this case to report failed page properly | |
8011 | * on tracepoint in test_pages_isolated() | |
8012 | */ | |
8013 | if (outer_start + (1UL << order) <= start) | |
8014 | outer_start = start; | |
8015 | } | |
8016 | ||
041d3a8c | 8017 | /* Make sure the range is really isolated. */ |
b023f468 | 8018 | if (test_pages_isolated(outer_start, end, false)) { |
75dddef3 | 8019 | pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n", |
dae803e1 | 8020 | __func__, outer_start, end); |
041d3a8c MN |
8021 | ret = -EBUSY; |
8022 | goto done; | |
8023 | } | |
8024 | ||
49f223a9 | 8025 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 8026 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
8027 | if (!outer_end) { |
8028 | ret = -EBUSY; | |
8029 | goto done; | |
8030 | } | |
8031 | ||
8032 | /* Free head and tail (if any) */ | |
8033 | if (start != outer_start) | |
8034 | free_contig_range(outer_start, start - outer_start); | |
8035 | if (end != outer_end) | |
8036 | free_contig_range(end, outer_end - end); | |
8037 | ||
8038 | done: | |
8039 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 8040 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
8041 | return ret; |
8042 | } | |
8043 | ||
8044 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
8045 | { | |
bcc2b02f MS |
8046 | unsigned int count = 0; |
8047 | ||
8048 | for (; nr_pages--; pfn++) { | |
8049 | struct page *page = pfn_to_page(pfn); | |
8050 | ||
8051 | count += page_count(page) != 1; | |
8052 | __free_page(page); | |
8053 | } | |
8054 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
8055 | } |
8056 | #endif | |
8057 | ||
d883c6cf | 8058 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
8059 | /* |
8060 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
8061 | * page high values need to be recalulated. | |
8062 | */ | |
4ed7e022 JL |
8063 | void __meminit zone_pcp_update(struct zone *zone) |
8064 | { | |
0a647f38 | 8065 | unsigned cpu; |
c8e251fa | 8066 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 8067 | for_each_possible_cpu(cpu) |
169f6c19 CS |
8068 | pageset_set_high_and_batch(zone, |
8069 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 8070 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
8071 | } |
8072 | #endif | |
8073 | ||
340175b7 JL |
8074 | void zone_pcp_reset(struct zone *zone) |
8075 | { | |
8076 | unsigned long flags; | |
5a883813 MK |
8077 | int cpu; |
8078 | struct per_cpu_pageset *pset; | |
340175b7 JL |
8079 | |
8080 | /* avoid races with drain_pages() */ | |
8081 | local_irq_save(flags); | |
8082 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
8083 | for_each_online_cpu(cpu) { |
8084 | pset = per_cpu_ptr(zone->pageset, cpu); | |
8085 | drain_zonestat(zone, pset); | |
8086 | } | |
340175b7 JL |
8087 | free_percpu(zone->pageset); |
8088 | zone->pageset = &boot_pageset; | |
8089 | } | |
8090 | local_irq_restore(flags); | |
8091 | } | |
8092 | ||
6dcd73d7 | 8093 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 8094 | /* |
b9eb6319 JK |
8095 | * All pages in the range must be in a single zone and isolated |
8096 | * before calling this. | |
0c0e6195 KH |
8097 | */ |
8098 | void | |
8099 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
8100 | { | |
8101 | struct page *page; | |
8102 | struct zone *zone; | |
7aeb09f9 | 8103 | unsigned int order, i; |
0c0e6195 KH |
8104 | unsigned long pfn; |
8105 | unsigned long flags; | |
8106 | /* find the first valid pfn */ | |
8107 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
8108 | if (pfn_valid(pfn)) | |
8109 | break; | |
8110 | if (pfn == end_pfn) | |
8111 | return; | |
2d070eab | 8112 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
8113 | zone = page_zone(pfn_to_page(pfn)); |
8114 | spin_lock_irqsave(&zone->lock, flags); | |
8115 | pfn = start_pfn; | |
8116 | while (pfn < end_pfn) { | |
8117 | if (!pfn_valid(pfn)) { | |
8118 | pfn++; | |
8119 | continue; | |
8120 | } | |
8121 | page = pfn_to_page(pfn); | |
b023f468 WC |
8122 | /* |
8123 | * The HWPoisoned page may be not in buddy system, and | |
8124 | * page_count() is not 0. | |
8125 | */ | |
8126 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
8127 | pfn++; | |
8128 | SetPageReserved(page); | |
8129 | continue; | |
8130 | } | |
8131 | ||
0c0e6195 KH |
8132 | BUG_ON(page_count(page)); |
8133 | BUG_ON(!PageBuddy(page)); | |
8134 | order = page_order(page); | |
8135 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
8136 | pr_info("remove from free list %lx %d %lx\n", |
8137 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
8138 | #endif |
8139 | list_del(&page->lru); | |
8140 | rmv_page_order(page); | |
8141 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
8142 | for (i = 0; i < (1 << order); i++) |
8143 | SetPageReserved((page+i)); | |
8144 | pfn += (1 << order); | |
8145 | } | |
8146 | spin_unlock_irqrestore(&zone->lock, flags); | |
8147 | } | |
8148 | #endif | |
8d22ba1b | 8149 | |
8d22ba1b WF |
8150 | bool is_free_buddy_page(struct page *page) |
8151 | { | |
8152 | struct zone *zone = page_zone(page); | |
8153 | unsigned long pfn = page_to_pfn(page); | |
8154 | unsigned long flags; | |
7aeb09f9 | 8155 | unsigned int order; |
8d22ba1b WF |
8156 | |
8157 | spin_lock_irqsave(&zone->lock, flags); | |
8158 | for (order = 0; order < MAX_ORDER; order++) { | |
8159 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8160 | ||
8161 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
8162 | break; | |
8163 | } | |
8164 | spin_unlock_irqrestore(&zone->lock, flags); | |
8165 | ||
8166 | return order < MAX_ORDER; | |
8167 | } | |
d4ae9916 NH |
8168 | |
8169 | #ifdef CONFIG_MEMORY_FAILURE | |
8170 | /* | |
8171 | * Set PG_hwpoison flag if a given page is confirmed to be a free page. This | |
8172 | * test is performed under the zone lock to prevent a race against page | |
8173 | * allocation. | |
8174 | */ | |
8175 | bool set_hwpoison_free_buddy_page(struct page *page) | |
8176 | { | |
8177 | struct zone *zone = page_zone(page); | |
8178 | unsigned long pfn = page_to_pfn(page); | |
8179 | unsigned long flags; | |
8180 | unsigned int order; | |
8181 | bool hwpoisoned = false; | |
8182 | ||
8183 | spin_lock_irqsave(&zone->lock, flags); | |
8184 | for (order = 0; order < MAX_ORDER; order++) { | |
8185 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8186 | ||
8187 | if (PageBuddy(page_head) && page_order(page_head) >= order) { | |
8188 | if (!TestSetPageHWPoison(page)) | |
8189 | hwpoisoned = true; | |
8190 | break; | |
8191 | } | |
8192 | } | |
8193 | spin_unlock_irqrestore(&zone->lock, flags); | |
8194 | ||
8195 | return hwpoisoned; | |
8196 | } | |
8197 | #endif |