Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/stddef.h> |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
10ed273f | 22 | #include <linux/jiffies.h> |
1da177e4 | 23 | #include <linux/bootmem.h> |
edbe7d23 | 24 | #include <linux/memblock.h> |
1da177e4 | 25 | #include <linux/compiler.h> |
9f158333 | 26 | #include <linux/kernel.h> |
b1eeab67 | 27 | #include <linux/kmemcheck.h> |
b8c73fc2 | 28 | #include <linux/kasan.h> |
1da177e4 LT |
29 | #include <linux/module.h> |
30 | #include <linux/suspend.h> | |
31 | #include <linux/pagevec.h> | |
32 | #include <linux/blkdev.h> | |
33 | #include <linux/slab.h> | |
a238ab5b | 34 | #include <linux/ratelimit.h> |
5a3135c2 | 35 | #include <linux/oom.h> |
1da177e4 LT |
36 | #include <linux/notifier.h> |
37 | #include <linux/topology.h> | |
38 | #include <linux/sysctl.h> | |
39 | #include <linux/cpu.h> | |
40 | #include <linux/cpuset.h> | |
bdc8cb98 | 41 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
42 | #include <linux/nodemask.h> |
43 | #include <linux/vmalloc.h> | |
a6cccdc3 | 44 | #include <linux/vmstat.h> |
4be38e35 | 45 | #include <linux/mempolicy.h> |
4b94ffdc | 46 | #include <linux/memremap.h> |
6811378e | 47 | #include <linux/stop_machine.h> |
c713216d MG |
48 | #include <linux/sort.h> |
49 | #include <linux/pfn.h> | |
3fcfab16 | 50 | #include <linux/backing-dev.h> |
933e312e | 51 | #include <linux/fault-inject.h> |
a5d76b54 | 52 | #include <linux/page-isolation.h> |
eefa864b | 53 | #include <linux/page_ext.h> |
3ac7fe5a | 54 | #include <linux/debugobjects.h> |
dbb1f81c | 55 | #include <linux/kmemleak.h> |
56de7263 | 56 | #include <linux/compaction.h> |
0d3d062a | 57 | #include <trace/events/kmem.h> |
d379f01d | 58 | #include <trace/events/oom.h> |
268bb0ce | 59 | #include <linux/prefetch.h> |
6e543d57 | 60 | #include <linux/mm_inline.h> |
041d3a8c | 61 | #include <linux/migrate.h> |
949f7ec5 | 62 | #include <linux/hugetlb.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
5b3cc15a | 64 | #include <linux/sched/mm.h> |
48c96a36 | 65 | #include <linux/page_owner.h> |
0e1cc95b | 66 | #include <linux/kthread.h> |
4949148a | 67 | #include <linux/memcontrol.h> |
1da177e4 | 68 | |
7ee3d4e8 | 69 | #include <asm/sections.h> |
1da177e4 | 70 | #include <asm/tlbflush.h> |
ac924c60 | 71 | #include <asm/div64.h> |
1da177e4 LT |
72 | #include "internal.h" |
73 | ||
c8e251fa CS |
74 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
75 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 76 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 77 | |
72812019 LS |
78 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
79 | DEFINE_PER_CPU(int, numa_node); | |
80 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
81 | #endif | |
82 | ||
7aac7898 LS |
83 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
84 | /* | |
85 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
86 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
87 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
88 | * defined in <linux/topology.h>. | |
89 | */ | |
90 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
91 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 92 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
93 | #endif |
94 | ||
bd233f53 MG |
95 | /* work_structs for global per-cpu drains */ |
96 | DEFINE_MUTEX(pcpu_drain_mutex); | |
97 | DEFINE_PER_CPU(struct work_struct, pcpu_drain); | |
98 | ||
38addce8 | 99 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 100 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
101 | EXPORT_SYMBOL(latent_entropy); |
102 | #endif | |
103 | ||
1da177e4 | 104 | /* |
13808910 | 105 | * Array of node states. |
1da177e4 | 106 | */ |
13808910 CL |
107 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
108 | [N_POSSIBLE] = NODE_MASK_ALL, | |
109 | [N_ONLINE] = { { [0] = 1UL } }, | |
110 | #ifndef CONFIG_NUMA | |
111 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
112 | #ifdef CONFIG_HIGHMEM | |
113 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
114 | #endif |
115 | #ifdef CONFIG_MOVABLE_NODE | |
116 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
117 | #endif |
118 | [N_CPU] = { { [0] = 1UL } }, | |
119 | #endif /* NUMA */ | |
120 | }; | |
121 | EXPORT_SYMBOL(node_states); | |
122 | ||
c3d5f5f0 JL |
123 | /* Protect totalram_pages and zone->managed_pages */ |
124 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
125 | ||
6c231b7b | 126 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 127 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 128 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 129 | |
1b76b02f | 130 | int percpu_pagelist_fraction; |
dcce284a | 131 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 132 | |
bb14c2c7 VB |
133 | /* |
134 | * A cached value of the page's pageblock's migratetype, used when the page is | |
135 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
136 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
137 | * Also the migratetype set in the page does not necessarily match the pcplist | |
138 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
139 | * other index - this ensures that it will be put on the correct CMA freelist. | |
140 | */ | |
141 | static inline int get_pcppage_migratetype(struct page *page) | |
142 | { | |
143 | return page->index; | |
144 | } | |
145 | ||
146 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
147 | { | |
148 | page->index = migratetype; | |
149 | } | |
150 | ||
452aa699 RW |
151 | #ifdef CONFIG_PM_SLEEP |
152 | /* | |
153 | * The following functions are used by the suspend/hibernate code to temporarily | |
154 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
155 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
156 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
157 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
158 | * guaranteed not to run in parallel with that modification). | |
159 | */ | |
c9e664f1 RW |
160 | |
161 | static gfp_t saved_gfp_mask; | |
162 | ||
163 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
164 | { |
165 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
166 | if (saved_gfp_mask) { |
167 | gfp_allowed_mask = saved_gfp_mask; | |
168 | saved_gfp_mask = 0; | |
169 | } | |
452aa699 RW |
170 | } |
171 | ||
c9e664f1 | 172 | void pm_restrict_gfp_mask(void) |
452aa699 | 173 | { |
452aa699 | 174 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
175 | WARN_ON(saved_gfp_mask); |
176 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 177 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 178 | } |
f90ac398 MG |
179 | |
180 | bool pm_suspended_storage(void) | |
181 | { | |
d0164adc | 182 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
183 | return false; |
184 | return true; | |
185 | } | |
452aa699 RW |
186 | #endif /* CONFIG_PM_SLEEP */ |
187 | ||
d9c23400 | 188 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 189 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
190 | #endif |
191 | ||
d98c7a09 | 192 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 193 | |
1da177e4 LT |
194 | /* |
195 | * results with 256, 32 in the lowmem_reserve sysctl: | |
196 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
197 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
198 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
199 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 200 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
201 | * |
202 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
203 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 204 | */ |
2f1b6248 | 205 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 206 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 207 | 256, |
4b51d669 | 208 | #endif |
fb0e7942 | 209 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 210 | 256, |
fb0e7942 | 211 | #endif |
e53ef38d | 212 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 213 | 32, |
e53ef38d | 214 | #endif |
2a1e274a | 215 | 32, |
2f1b6248 | 216 | }; |
1da177e4 LT |
217 | |
218 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 219 | |
15ad7cdc | 220 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 221 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 222 | "DMA", |
4b51d669 | 223 | #endif |
fb0e7942 | 224 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 225 | "DMA32", |
fb0e7942 | 226 | #endif |
2f1b6248 | 227 | "Normal", |
e53ef38d | 228 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 229 | "HighMem", |
e53ef38d | 230 | #endif |
2a1e274a | 231 | "Movable", |
033fbae9 DW |
232 | #ifdef CONFIG_ZONE_DEVICE |
233 | "Device", | |
234 | #endif | |
2f1b6248 CL |
235 | }; |
236 | ||
60f30350 VB |
237 | char * const migratetype_names[MIGRATE_TYPES] = { |
238 | "Unmovable", | |
239 | "Movable", | |
240 | "Reclaimable", | |
241 | "HighAtomic", | |
242 | #ifdef CONFIG_CMA | |
243 | "CMA", | |
244 | #endif | |
245 | #ifdef CONFIG_MEMORY_ISOLATION | |
246 | "Isolate", | |
247 | #endif | |
248 | }; | |
249 | ||
f1e61557 KS |
250 | compound_page_dtor * const compound_page_dtors[] = { |
251 | NULL, | |
252 | free_compound_page, | |
253 | #ifdef CONFIG_HUGETLB_PAGE | |
254 | free_huge_page, | |
255 | #endif | |
9a982250 KS |
256 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
257 | free_transhuge_page, | |
258 | #endif | |
f1e61557 KS |
259 | }; |
260 | ||
1da177e4 | 261 | int min_free_kbytes = 1024; |
42aa83cb | 262 | int user_min_free_kbytes = -1; |
795ae7a0 | 263 | int watermark_scale_factor = 10; |
1da177e4 | 264 | |
2c85f51d JB |
265 | static unsigned long __meminitdata nr_kernel_pages; |
266 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 267 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 268 | |
0ee332c1 TH |
269 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
270 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
271 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
272 | static unsigned long __initdata required_kernelcore; | |
273 | static unsigned long __initdata required_movablecore; | |
274 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
342332e6 | 275 | static bool mirrored_kernelcore; |
0ee332c1 TH |
276 | |
277 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
278 | int movable_zone; | |
279 | EXPORT_SYMBOL(movable_zone); | |
280 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 281 | |
418508c1 MS |
282 | #if MAX_NUMNODES > 1 |
283 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 284 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 285 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 286 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
287 | #endif |
288 | ||
9ef9acb0 MG |
289 | int page_group_by_mobility_disabled __read_mostly; |
290 | ||
3a80a7fa MG |
291 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
292 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
293 | { | |
294 | pgdat->first_deferred_pfn = ULONG_MAX; | |
295 | } | |
296 | ||
297 | /* Returns true if the struct page for the pfn is uninitialised */ | |
0e1cc95b | 298 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 299 | { |
ef70b6f4 MG |
300 | int nid = early_pfn_to_nid(pfn); |
301 | ||
302 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
303 | return true; |
304 | ||
305 | return false; | |
306 | } | |
307 | ||
308 | /* | |
309 | * Returns false when the remaining initialisation should be deferred until | |
310 | * later in the boot cycle when it can be parallelised. | |
311 | */ | |
312 | static inline bool update_defer_init(pg_data_t *pgdat, | |
313 | unsigned long pfn, unsigned long zone_end, | |
314 | unsigned long *nr_initialised) | |
315 | { | |
987b3095 LZ |
316 | unsigned long max_initialise; |
317 | ||
3a80a7fa MG |
318 | /* Always populate low zones for address-contrained allocations */ |
319 | if (zone_end < pgdat_end_pfn(pgdat)) | |
320 | return true; | |
987b3095 LZ |
321 | /* |
322 | * Initialise at least 2G of a node but also take into account that | |
323 | * two large system hashes that can take up 1GB for 0.25TB/node. | |
324 | */ | |
325 | max_initialise = max(2UL << (30 - PAGE_SHIFT), | |
326 | (pgdat->node_spanned_pages >> 8)); | |
3a80a7fa | 327 | |
3a80a7fa | 328 | (*nr_initialised)++; |
987b3095 | 329 | if ((*nr_initialised > max_initialise) && |
3a80a7fa MG |
330 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
331 | pgdat->first_deferred_pfn = pfn; | |
332 | return false; | |
333 | } | |
334 | ||
335 | return true; | |
336 | } | |
337 | #else | |
338 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
339 | { | |
340 | } | |
341 | ||
342 | static inline bool early_page_uninitialised(unsigned long pfn) | |
343 | { | |
344 | return false; | |
345 | } | |
346 | ||
347 | static inline bool update_defer_init(pg_data_t *pgdat, | |
348 | unsigned long pfn, unsigned long zone_end, | |
349 | unsigned long *nr_initialised) | |
350 | { | |
351 | return true; | |
352 | } | |
353 | #endif | |
354 | ||
0b423ca2 MG |
355 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
356 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
357 | unsigned long pfn) | |
358 | { | |
359 | #ifdef CONFIG_SPARSEMEM | |
360 | return __pfn_to_section(pfn)->pageblock_flags; | |
361 | #else | |
362 | return page_zone(page)->pageblock_flags; | |
363 | #endif /* CONFIG_SPARSEMEM */ | |
364 | } | |
365 | ||
366 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
367 | { | |
368 | #ifdef CONFIG_SPARSEMEM | |
369 | pfn &= (PAGES_PER_SECTION-1); | |
370 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
371 | #else | |
372 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
373 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
374 | #endif /* CONFIG_SPARSEMEM */ | |
375 | } | |
376 | ||
377 | /** | |
378 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
379 | * @page: The page within the block of interest | |
380 | * @pfn: The target page frame number | |
381 | * @end_bitidx: The last bit of interest to retrieve | |
382 | * @mask: mask of bits that the caller is interested in | |
383 | * | |
384 | * Return: pageblock_bits flags | |
385 | */ | |
386 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
387 | unsigned long pfn, | |
388 | unsigned long end_bitidx, | |
389 | unsigned long mask) | |
390 | { | |
391 | unsigned long *bitmap; | |
392 | unsigned long bitidx, word_bitidx; | |
393 | unsigned long word; | |
394 | ||
395 | bitmap = get_pageblock_bitmap(page, pfn); | |
396 | bitidx = pfn_to_bitidx(page, pfn); | |
397 | word_bitidx = bitidx / BITS_PER_LONG; | |
398 | bitidx &= (BITS_PER_LONG-1); | |
399 | ||
400 | word = bitmap[word_bitidx]; | |
401 | bitidx += end_bitidx; | |
402 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
403 | } | |
404 | ||
405 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
406 | unsigned long end_bitidx, | |
407 | unsigned long mask) | |
408 | { | |
409 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
410 | } | |
411 | ||
412 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
413 | { | |
414 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
415 | } | |
416 | ||
417 | /** | |
418 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
419 | * @page: The page within the block of interest | |
420 | * @flags: The flags to set | |
421 | * @pfn: The target page frame number | |
422 | * @end_bitidx: The last bit of interest | |
423 | * @mask: mask of bits that the caller is interested in | |
424 | */ | |
425 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
426 | unsigned long pfn, | |
427 | unsigned long end_bitidx, | |
428 | unsigned long mask) | |
429 | { | |
430 | unsigned long *bitmap; | |
431 | unsigned long bitidx, word_bitidx; | |
432 | unsigned long old_word, word; | |
433 | ||
434 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
435 | ||
436 | bitmap = get_pageblock_bitmap(page, pfn); | |
437 | bitidx = pfn_to_bitidx(page, pfn); | |
438 | word_bitidx = bitidx / BITS_PER_LONG; | |
439 | bitidx &= (BITS_PER_LONG-1); | |
440 | ||
441 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
442 | ||
443 | bitidx += end_bitidx; | |
444 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
445 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
446 | ||
447 | word = READ_ONCE(bitmap[word_bitidx]); | |
448 | for (;;) { | |
449 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
450 | if (word == old_word) | |
451 | break; | |
452 | word = old_word; | |
453 | } | |
454 | } | |
3a80a7fa | 455 | |
ee6f509c | 456 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 457 | { |
5d0f3f72 KM |
458 | if (unlikely(page_group_by_mobility_disabled && |
459 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
460 | migratetype = MIGRATE_UNMOVABLE; |
461 | ||
b2a0ac88 MG |
462 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
463 | PB_migrate, PB_migrate_end); | |
464 | } | |
465 | ||
13e7444b | 466 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 467 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 468 | { |
bdc8cb98 DH |
469 | int ret = 0; |
470 | unsigned seq; | |
471 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 472 | unsigned long sp, start_pfn; |
c6a57e19 | 473 | |
bdc8cb98 DH |
474 | do { |
475 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
476 | start_pfn = zone->zone_start_pfn; |
477 | sp = zone->spanned_pages; | |
108bcc96 | 478 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
479 | ret = 1; |
480 | } while (zone_span_seqretry(zone, seq)); | |
481 | ||
b5e6a5a2 | 482 | if (ret) |
613813e8 DH |
483 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
484 | pfn, zone_to_nid(zone), zone->name, | |
485 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 486 | |
bdc8cb98 | 487 | return ret; |
c6a57e19 DH |
488 | } |
489 | ||
490 | static int page_is_consistent(struct zone *zone, struct page *page) | |
491 | { | |
14e07298 | 492 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 493 | return 0; |
1da177e4 | 494 | if (zone != page_zone(page)) |
c6a57e19 DH |
495 | return 0; |
496 | ||
497 | return 1; | |
498 | } | |
499 | /* | |
500 | * Temporary debugging check for pages not lying within a given zone. | |
501 | */ | |
502 | static int bad_range(struct zone *zone, struct page *page) | |
503 | { | |
504 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 505 | return 1; |
c6a57e19 DH |
506 | if (!page_is_consistent(zone, page)) |
507 | return 1; | |
508 | ||
1da177e4 LT |
509 | return 0; |
510 | } | |
13e7444b NP |
511 | #else |
512 | static inline int bad_range(struct zone *zone, struct page *page) | |
513 | { | |
514 | return 0; | |
515 | } | |
516 | #endif | |
517 | ||
d230dec1 KS |
518 | static void bad_page(struct page *page, const char *reason, |
519 | unsigned long bad_flags) | |
1da177e4 | 520 | { |
d936cf9b HD |
521 | static unsigned long resume; |
522 | static unsigned long nr_shown; | |
523 | static unsigned long nr_unshown; | |
524 | ||
525 | /* | |
526 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
527 | * or allow a steady drip of one report per second. | |
528 | */ | |
529 | if (nr_shown == 60) { | |
530 | if (time_before(jiffies, resume)) { | |
531 | nr_unshown++; | |
532 | goto out; | |
533 | } | |
534 | if (nr_unshown) { | |
ff8e8116 | 535 | pr_alert( |
1e9e6365 | 536 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
537 | nr_unshown); |
538 | nr_unshown = 0; | |
539 | } | |
540 | nr_shown = 0; | |
541 | } | |
542 | if (nr_shown++ == 0) | |
543 | resume = jiffies + 60 * HZ; | |
544 | ||
ff8e8116 | 545 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 546 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
547 | __dump_page(page, reason); |
548 | bad_flags &= page->flags; | |
549 | if (bad_flags) | |
550 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
551 | bad_flags, &bad_flags); | |
4e462112 | 552 | dump_page_owner(page); |
3dc14741 | 553 | |
4f31888c | 554 | print_modules(); |
1da177e4 | 555 | dump_stack(); |
d936cf9b | 556 | out: |
8cc3b392 | 557 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 558 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 559 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
560 | } |
561 | ||
1da177e4 LT |
562 | /* |
563 | * Higher-order pages are called "compound pages". They are structured thusly: | |
564 | * | |
1d798ca3 | 565 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 566 | * |
1d798ca3 KS |
567 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
568 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 569 | * |
1d798ca3 KS |
570 | * The first tail page's ->compound_dtor holds the offset in array of compound |
571 | * page destructors. See compound_page_dtors. | |
1da177e4 | 572 | * |
1d798ca3 | 573 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 574 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 575 | */ |
d98c7a09 | 576 | |
9a982250 | 577 | void free_compound_page(struct page *page) |
d98c7a09 | 578 | { |
d85f3385 | 579 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
580 | } |
581 | ||
d00181b9 | 582 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
583 | { |
584 | int i; | |
585 | int nr_pages = 1 << order; | |
586 | ||
f1e61557 | 587 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
588 | set_compound_order(page, order); |
589 | __SetPageHead(page); | |
590 | for (i = 1; i < nr_pages; i++) { | |
591 | struct page *p = page + i; | |
58a84aa9 | 592 | set_page_count(p, 0); |
1c290f64 | 593 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 594 | set_compound_head(p, page); |
18229df5 | 595 | } |
53f9263b | 596 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
597 | } |
598 | ||
c0a32fc5 SG |
599 | #ifdef CONFIG_DEBUG_PAGEALLOC |
600 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
601 | bool _debug_pagealloc_enabled __read_mostly |
602 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 603 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
604 | bool _debug_guardpage_enabled __read_mostly; |
605 | ||
031bc574 JK |
606 | static int __init early_debug_pagealloc(char *buf) |
607 | { | |
608 | if (!buf) | |
609 | return -EINVAL; | |
2a138dc7 | 610 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
031bc574 JK |
611 | } |
612 | early_param("debug_pagealloc", early_debug_pagealloc); | |
613 | ||
e30825f1 JK |
614 | static bool need_debug_guardpage(void) |
615 | { | |
031bc574 JK |
616 | /* If we don't use debug_pagealloc, we don't need guard page */ |
617 | if (!debug_pagealloc_enabled()) | |
618 | return false; | |
619 | ||
f1c1e9f7 JK |
620 | if (!debug_guardpage_minorder()) |
621 | return false; | |
622 | ||
e30825f1 JK |
623 | return true; |
624 | } | |
625 | ||
626 | static void init_debug_guardpage(void) | |
627 | { | |
031bc574 JK |
628 | if (!debug_pagealloc_enabled()) |
629 | return; | |
630 | ||
f1c1e9f7 JK |
631 | if (!debug_guardpage_minorder()) |
632 | return; | |
633 | ||
e30825f1 JK |
634 | _debug_guardpage_enabled = true; |
635 | } | |
636 | ||
637 | struct page_ext_operations debug_guardpage_ops = { | |
638 | .need = need_debug_guardpage, | |
639 | .init = init_debug_guardpage, | |
640 | }; | |
c0a32fc5 SG |
641 | |
642 | static int __init debug_guardpage_minorder_setup(char *buf) | |
643 | { | |
644 | unsigned long res; | |
645 | ||
646 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 647 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
648 | return 0; |
649 | } | |
650 | _debug_guardpage_minorder = res; | |
1170532b | 651 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
652 | return 0; |
653 | } | |
f1c1e9f7 | 654 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 655 | |
acbc15a4 | 656 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 657 | unsigned int order, int migratetype) |
c0a32fc5 | 658 | { |
e30825f1 JK |
659 | struct page_ext *page_ext; |
660 | ||
661 | if (!debug_guardpage_enabled()) | |
acbc15a4 JK |
662 | return false; |
663 | ||
664 | if (order >= debug_guardpage_minorder()) | |
665 | return false; | |
e30825f1 JK |
666 | |
667 | page_ext = lookup_page_ext(page); | |
f86e4271 | 668 | if (unlikely(!page_ext)) |
acbc15a4 | 669 | return false; |
f86e4271 | 670 | |
e30825f1 JK |
671 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
672 | ||
2847cf95 JK |
673 | INIT_LIST_HEAD(&page->lru); |
674 | set_page_private(page, order); | |
675 | /* Guard pages are not available for any usage */ | |
676 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
677 | |
678 | return true; | |
c0a32fc5 SG |
679 | } |
680 | ||
2847cf95 JK |
681 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
682 | unsigned int order, int migratetype) | |
c0a32fc5 | 683 | { |
e30825f1 JK |
684 | struct page_ext *page_ext; |
685 | ||
686 | if (!debug_guardpage_enabled()) | |
687 | return; | |
688 | ||
689 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
690 | if (unlikely(!page_ext)) |
691 | return; | |
692 | ||
e30825f1 JK |
693 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
694 | ||
2847cf95 JK |
695 | set_page_private(page, 0); |
696 | if (!is_migrate_isolate(migratetype)) | |
697 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
698 | } |
699 | #else | |
980ac167 | 700 | struct page_ext_operations debug_guardpage_ops; |
acbc15a4 JK |
701 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
702 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
703 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
704 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
705 | #endif |
706 | ||
7aeb09f9 | 707 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 708 | { |
4c21e2f2 | 709 | set_page_private(page, order); |
676165a8 | 710 | __SetPageBuddy(page); |
1da177e4 LT |
711 | } |
712 | ||
713 | static inline void rmv_page_order(struct page *page) | |
714 | { | |
676165a8 | 715 | __ClearPageBuddy(page); |
4c21e2f2 | 716 | set_page_private(page, 0); |
1da177e4 LT |
717 | } |
718 | ||
1da177e4 LT |
719 | /* |
720 | * This function checks whether a page is free && is the buddy | |
721 | * we can do coalesce a page and its buddy if | |
13ad59df | 722 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 723 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
724 | * (c) a page and its buddy have the same order && |
725 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 726 | * |
cf6fe945 WSH |
727 | * For recording whether a page is in the buddy system, we set ->_mapcount |
728 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
729 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
730 | * serialized by zone->lock. | |
1da177e4 | 731 | * |
676165a8 | 732 | * For recording page's order, we use page_private(page). |
1da177e4 | 733 | */ |
cb2b95e1 | 734 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 735 | unsigned int order) |
1da177e4 | 736 | { |
c0a32fc5 | 737 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
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 | ||
c0a32fc5 SG |
743 | return 1; |
744 | } | |
745 | ||
cb2b95e1 | 746 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
747 | /* |
748 | * zone check is done late to avoid uselessly | |
749 | * calculating zone/node ids for pages that could | |
750 | * never merge. | |
751 | */ | |
752 | if (page_zone_id(page) != page_zone_id(buddy)) | |
753 | return 0; | |
754 | ||
4c5018ce WY |
755 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
756 | ||
6aa3001b | 757 | return 1; |
676165a8 | 758 | } |
6aa3001b | 759 | return 0; |
1da177e4 LT |
760 | } |
761 | ||
762 | /* | |
763 | * Freeing function for a buddy system allocator. | |
764 | * | |
765 | * The concept of a buddy system is to maintain direct-mapped table | |
766 | * (containing bit values) for memory blocks of various "orders". | |
767 | * The bottom level table contains the map for the smallest allocatable | |
768 | * units of memory (here, pages), and each level above it describes | |
769 | * pairs of units from the levels below, hence, "buddies". | |
770 | * At a high level, all that happens here is marking the table entry | |
771 | * at the bottom level available, and propagating the changes upward | |
772 | * as necessary, plus some accounting needed to play nicely with other | |
773 | * parts of the VM system. | |
774 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
775 | * free pages of length of (1 << order) and marked with _mapcount |
776 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
777 | * field. | |
1da177e4 | 778 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
779 | * other. That is, if we allocate a small block, and both were |
780 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 781 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 782 | * triggers coalescing into a block of larger size. |
1da177e4 | 783 | * |
6d49e352 | 784 | * -- nyc |
1da177e4 LT |
785 | */ |
786 | ||
48db57f8 | 787 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 788 | unsigned long pfn, |
ed0ae21d MG |
789 | struct zone *zone, unsigned int order, |
790 | int migratetype) | |
1da177e4 | 791 | { |
76741e77 VB |
792 | unsigned long combined_pfn; |
793 | unsigned long uninitialized_var(buddy_pfn); | |
6dda9d55 | 794 | struct page *buddy; |
d9dddbf5 VB |
795 | unsigned int max_order; |
796 | ||
797 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 798 | |
d29bb978 | 799 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 800 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 801 | |
ed0ae21d | 802 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 803 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 804 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 805 | |
76741e77 | 806 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 807 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 808 | |
d9dddbf5 | 809 | continue_merging: |
3c605096 | 810 | while (order < max_order - 1) { |
76741e77 VB |
811 | buddy_pfn = __find_buddy_pfn(pfn, order); |
812 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
813 | |
814 | if (!pfn_valid_within(buddy_pfn)) | |
815 | goto done_merging; | |
cb2b95e1 | 816 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 817 | goto done_merging; |
c0a32fc5 SG |
818 | /* |
819 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
820 | * merge with it and move up one order. | |
821 | */ | |
822 | if (page_is_guard(buddy)) { | |
2847cf95 | 823 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
824 | } else { |
825 | list_del(&buddy->lru); | |
826 | zone->free_area[order].nr_free--; | |
827 | rmv_page_order(buddy); | |
828 | } | |
76741e77 VB |
829 | combined_pfn = buddy_pfn & pfn; |
830 | page = page + (combined_pfn - pfn); | |
831 | pfn = combined_pfn; | |
1da177e4 LT |
832 | order++; |
833 | } | |
d9dddbf5 VB |
834 | if (max_order < MAX_ORDER) { |
835 | /* If we are here, it means order is >= pageblock_order. | |
836 | * We want to prevent merge between freepages on isolate | |
837 | * pageblock and normal pageblock. Without this, pageblock | |
838 | * isolation could cause incorrect freepage or CMA accounting. | |
839 | * | |
840 | * We don't want to hit this code for the more frequent | |
841 | * low-order merging. | |
842 | */ | |
843 | if (unlikely(has_isolate_pageblock(zone))) { | |
844 | int buddy_mt; | |
845 | ||
76741e77 VB |
846 | buddy_pfn = __find_buddy_pfn(pfn, order); |
847 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
848 | buddy_mt = get_pageblock_migratetype(buddy); |
849 | ||
850 | if (migratetype != buddy_mt | |
851 | && (is_migrate_isolate(migratetype) || | |
852 | is_migrate_isolate(buddy_mt))) | |
853 | goto done_merging; | |
854 | } | |
855 | max_order++; | |
856 | goto continue_merging; | |
857 | } | |
858 | ||
859 | done_merging: | |
1da177e4 | 860 | set_page_order(page, order); |
6dda9d55 CZ |
861 | |
862 | /* | |
863 | * If this is not the largest possible page, check if the buddy | |
864 | * of the next-highest order is free. If it is, it's possible | |
865 | * that pages are being freed that will coalesce soon. In case, | |
866 | * that is happening, add the free page to the tail of the list | |
867 | * so it's less likely to be used soon and more likely to be merged | |
868 | * as a higher order page | |
869 | */ | |
13ad59df | 870 | if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) { |
6dda9d55 | 871 | struct page *higher_page, *higher_buddy; |
76741e77 VB |
872 | combined_pfn = buddy_pfn & pfn; |
873 | higher_page = page + (combined_pfn - pfn); | |
874 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
875 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
6dda9d55 CZ |
876 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
877 | list_add_tail(&page->lru, | |
878 | &zone->free_area[order].free_list[migratetype]); | |
879 | goto out; | |
880 | } | |
881 | } | |
882 | ||
883 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
884 | out: | |
1da177e4 LT |
885 | zone->free_area[order].nr_free++; |
886 | } | |
887 | ||
7bfec6f4 MG |
888 | /* |
889 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
890 | * try and check multiple fields with one check. The caller must do a detailed | |
891 | * check if necessary. | |
892 | */ | |
893 | static inline bool page_expected_state(struct page *page, | |
894 | unsigned long check_flags) | |
895 | { | |
896 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
897 | return false; | |
898 | ||
899 | if (unlikely((unsigned long)page->mapping | | |
900 | page_ref_count(page) | | |
901 | #ifdef CONFIG_MEMCG | |
902 | (unsigned long)page->mem_cgroup | | |
903 | #endif | |
904 | (page->flags & check_flags))) | |
905 | return false; | |
906 | ||
907 | return true; | |
908 | } | |
909 | ||
bb552ac6 | 910 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 911 | { |
7bfec6f4 MG |
912 | const char *bad_reason; |
913 | unsigned long bad_flags; | |
914 | ||
7bfec6f4 MG |
915 | bad_reason = NULL; |
916 | bad_flags = 0; | |
f0b791a3 | 917 | |
53f9263b | 918 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
919 | bad_reason = "nonzero mapcount"; |
920 | if (unlikely(page->mapping != NULL)) | |
921 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 922 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 923 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
924 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
925 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
926 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
927 | } | |
9edad6ea JW |
928 | #ifdef CONFIG_MEMCG |
929 | if (unlikely(page->mem_cgroup)) | |
930 | bad_reason = "page still charged to cgroup"; | |
931 | #endif | |
7bfec6f4 | 932 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
933 | } |
934 | ||
935 | static inline int free_pages_check(struct page *page) | |
936 | { | |
da838d4f | 937 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 938 | return 0; |
bb552ac6 MG |
939 | |
940 | /* Something has gone sideways, find it */ | |
941 | free_pages_check_bad(page); | |
7bfec6f4 | 942 | return 1; |
1da177e4 LT |
943 | } |
944 | ||
4db7548c MG |
945 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
946 | { | |
947 | int ret = 1; | |
948 | ||
949 | /* | |
950 | * We rely page->lru.next never has bit 0 set, unless the page | |
951 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
952 | */ | |
953 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
954 | ||
955 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
956 | ret = 0; | |
957 | goto out; | |
958 | } | |
959 | switch (page - head_page) { | |
960 | case 1: | |
961 | /* the first tail page: ->mapping is compound_mapcount() */ | |
962 | if (unlikely(compound_mapcount(page))) { | |
963 | bad_page(page, "nonzero compound_mapcount", 0); | |
964 | goto out; | |
965 | } | |
966 | break; | |
967 | case 2: | |
968 | /* | |
969 | * the second tail page: ->mapping is | |
970 | * page_deferred_list().next -- ignore value. | |
971 | */ | |
972 | break; | |
973 | default: | |
974 | if (page->mapping != TAIL_MAPPING) { | |
975 | bad_page(page, "corrupted mapping in tail page", 0); | |
976 | goto out; | |
977 | } | |
978 | break; | |
979 | } | |
980 | if (unlikely(!PageTail(page))) { | |
981 | bad_page(page, "PageTail not set", 0); | |
982 | goto out; | |
983 | } | |
984 | if (unlikely(compound_head(page) != head_page)) { | |
985 | bad_page(page, "compound_head not consistent", 0); | |
986 | goto out; | |
987 | } | |
988 | ret = 0; | |
989 | out: | |
990 | page->mapping = NULL; | |
991 | clear_compound_head(page); | |
992 | return ret; | |
993 | } | |
994 | ||
e2769dbd MG |
995 | static __always_inline bool free_pages_prepare(struct page *page, |
996 | unsigned int order, bool check_free) | |
4db7548c | 997 | { |
e2769dbd | 998 | int bad = 0; |
4db7548c | 999 | |
4db7548c MG |
1000 | VM_BUG_ON_PAGE(PageTail(page), page); |
1001 | ||
e2769dbd MG |
1002 | trace_mm_page_free(page, order); |
1003 | kmemcheck_free_shadow(page, order); | |
e2769dbd MG |
1004 | |
1005 | /* | |
1006 | * Check tail pages before head page information is cleared to | |
1007 | * avoid checking PageCompound for order-0 pages. | |
1008 | */ | |
1009 | if (unlikely(order)) { | |
1010 | bool compound = PageCompound(page); | |
1011 | int i; | |
1012 | ||
1013 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1014 | |
9a73f61b KS |
1015 | if (compound) |
1016 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1017 | for (i = 1; i < (1 << order); i++) { |
1018 | if (compound) | |
1019 | bad += free_tail_pages_check(page, page + i); | |
1020 | if (unlikely(free_pages_check(page + i))) { | |
1021 | bad++; | |
1022 | continue; | |
1023 | } | |
1024 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1025 | } | |
1026 | } | |
bda807d4 | 1027 | if (PageMappingFlags(page)) |
4db7548c | 1028 | page->mapping = NULL; |
c4159a75 | 1029 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
4949148a | 1030 | memcg_kmem_uncharge(page, order); |
e2769dbd MG |
1031 | if (check_free) |
1032 | bad += free_pages_check(page); | |
1033 | if (bad) | |
1034 | return false; | |
4db7548c | 1035 | |
e2769dbd MG |
1036 | page_cpupid_reset_last(page); |
1037 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1038 | reset_page_owner(page, order); | |
4db7548c MG |
1039 | |
1040 | if (!PageHighMem(page)) { | |
1041 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1042 | PAGE_SIZE << order); |
4db7548c | 1043 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1044 | PAGE_SIZE << order); |
4db7548c | 1045 | } |
e2769dbd MG |
1046 | arch_free_page(page, order); |
1047 | kernel_poison_pages(page, 1 << order, 0); | |
1048 | kernel_map_pages(page, 1 << order, 0); | |
29b52de1 | 1049 | kasan_free_pages(page, order); |
4db7548c | 1050 | |
4db7548c MG |
1051 | return true; |
1052 | } | |
1053 | ||
e2769dbd MG |
1054 | #ifdef CONFIG_DEBUG_VM |
1055 | static inline bool free_pcp_prepare(struct page *page) | |
1056 | { | |
1057 | return free_pages_prepare(page, 0, true); | |
1058 | } | |
1059 | ||
1060 | static inline bool bulkfree_pcp_prepare(struct page *page) | |
1061 | { | |
1062 | return false; | |
1063 | } | |
1064 | #else | |
1065 | static bool free_pcp_prepare(struct page *page) | |
1066 | { | |
1067 | return free_pages_prepare(page, 0, false); | |
1068 | } | |
1069 | ||
4db7548c MG |
1070 | static bool bulkfree_pcp_prepare(struct page *page) |
1071 | { | |
1072 | return free_pages_check(page); | |
1073 | } | |
1074 | #endif /* CONFIG_DEBUG_VM */ | |
1075 | ||
1da177e4 | 1076 | /* |
5f8dcc21 | 1077 | * Frees a number of pages from the PCP lists |
1da177e4 | 1078 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1079 | * count is the number of pages to free. |
1da177e4 LT |
1080 | * |
1081 | * If the zone was previously in an "all pages pinned" state then look to | |
1082 | * see if this freeing clears that state. | |
1083 | * | |
1084 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1085 | * pinned" detection logic. | |
1086 | */ | |
5f8dcc21 MG |
1087 | static void free_pcppages_bulk(struct zone *zone, int count, |
1088 | struct per_cpu_pages *pcp) | |
1da177e4 | 1089 | { |
5f8dcc21 | 1090 | int migratetype = 0; |
a6f9edd6 | 1091 | int batch_free = 0; |
374ad05a | 1092 | unsigned long nr_scanned, flags; |
3777999d | 1093 | bool isolated_pageblocks; |
5f8dcc21 | 1094 | |
374ad05a | 1095 | spin_lock_irqsave(&zone->lock, flags); |
3777999d | 1096 | isolated_pageblocks = has_isolate_pageblock(zone); |
599d0c95 | 1097 | nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED); |
0d5d823a | 1098 | if (nr_scanned) |
599d0c95 | 1099 | __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned); |
f2260e6b | 1100 | |
e5b31ac2 | 1101 | while (count) { |
48db57f8 | 1102 | struct page *page; |
5f8dcc21 MG |
1103 | struct list_head *list; |
1104 | ||
1105 | /* | |
a6f9edd6 MG |
1106 | * Remove pages from lists in a round-robin fashion. A |
1107 | * batch_free count is maintained that is incremented when an | |
1108 | * empty list is encountered. This is so more pages are freed | |
1109 | * off fuller lists instead of spinning excessively around empty | |
1110 | * lists | |
5f8dcc21 MG |
1111 | */ |
1112 | do { | |
a6f9edd6 | 1113 | batch_free++; |
5f8dcc21 MG |
1114 | if (++migratetype == MIGRATE_PCPTYPES) |
1115 | migratetype = 0; | |
1116 | list = &pcp->lists[migratetype]; | |
1117 | } while (list_empty(list)); | |
48db57f8 | 1118 | |
1d16871d NK |
1119 | /* This is the only non-empty list. Free them all. */ |
1120 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1121 | batch_free = count; |
1d16871d | 1122 | |
a6f9edd6 | 1123 | do { |
770c8aaa BZ |
1124 | int mt; /* migratetype of the to-be-freed page */ |
1125 | ||
a16601c5 | 1126 | page = list_last_entry(list, struct page, lru); |
a6f9edd6 MG |
1127 | /* must delete as __free_one_page list manipulates */ |
1128 | list_del(&page->lru); | |
aa016d14 | 1129 | |
bb14c2c7 | 1130 | mt = get_pcppage_migratetype(page); |
aa016d14 VB |
1131 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1132 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1133 | /* Pageblock could have been isolated meanwhile */ | |
3777999d | 1134 | if (unlikely(isolated_pageblocks)) |
51bb1a40 | 1135 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 1136 | |
4db7548c MG |
1137 | if (bulkfree_pcp_prepare(page)) |
1138 | continue; | |
1139 | ||
dc4b0caf | 1140 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 1141 | trace_mm_page_pcpu_drain(page, 0, mt); |
e5b31ac2 | 1142 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1143 | } |
374ad05a | 1144 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 LT |
1145 | } |
1146 | ||
dc4b0caf MG |
1147 | static void free_one_page(struct zone *zone, |
1148 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1149 | unsigned int order, |
ed0ae21d | 1150 | int migratetype) |
1da177e4 | 1151 | { |
374ad05a MG |
1152 | unsigned long nr_scanned, flags; |
1153 | spin_lock_irqsave(&zone->lock, flags); | |
1154 | __count_vm_events(PGFREE, 1 << order); | |
599d0c95 | 1155 | nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED); |
0d5d823a | 1156 | if (nr_scanned) |
599d0c95 | 1157 | __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned); |
f2260e6b | 1158 | |
ad53f92e JK |
1159 | if (unlikely(has_isolate_pageblock(zone) || |
1160 | is_migrate_isolate(migratetype))) { | |
1161 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1162 | } |
dc4b0caf | 1163 | __free_one_page(page, pfn, zone, order, migratetype); |
374ad05a | 1164 | spin_unlock_irqrestore(&zone->lock, flags); |
48db57f8 NP |
1165 | } |
1166 | ||
1e8ce83c RH |
1167 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
1168 | unsigned long zone, int nid) | |
1169 | { | |
1e8ce83c | 1170 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1171 | init_page_count(page); |
1172 | page_mapcount_reset(page); | |
1173 | page_cpupid_reset_last(page); | |
1e8ce83c | 1174 | |
1e8ce83c RH |
1175 | INIT_LIST_HEAD(&page->lru); |
1176 | #ifdef WANT_PAGE_VIRTUAL | |
1177 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1178 | if (!is_highmem_idx(zone)) | |
1179 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1180 | #endif | |
1181 | } | |
1182 | ||
1183 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
1184 | int nid) | |
1185 | { | |
1186 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
1187 | } | |
1188 | ||
7e18adb4 MG |
1189 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1190 | static void init_reserved_page(unsigned long pfn) | |
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 | } | |
1207 | __init_single_pfn(pfn, zid, nid); | |
1208 | } | |
1209 | #else | |
1210 | static inline void init_reserved_page(unsigned long pfn) | |
1211 | { | |
1212 | } | |
1213 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1214 | ||
92923ca3 NZ |
1215 | /* |
1216 | * Initialised pages do not have PageReserved set. This function is | |
1217 | * called for each range allocated by the bootmem allocator and | |
1218 | * marks the pages PageReserved. The remaining valid pages are later | |
1219 | * sent to the buddy page allocator. | |
1220 | */ | |
4b50bcc7 | 1221 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1222 | { |
1223 | unsigned long start_pfn = PFN_DOWN(start); | |
1224 | unsigned long end_pfn = PFN_UP(end); | |
1225 | ||
7e18adb4 MG |
1226 | for (; start_pfn < end_pfn; start_pfn++) { |
1227 | if (pfn_valid(start_pfn)) { | |
1228 | struct page *page = pfn_to_page(start_pfn); | |
1229 | ||
1230 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1231 | |
1232 | /* Avoid false-positive PageTail() */ | |
1233 | INIT_LIST_HEAD(&page->lru); | |
1234 | ||
7e18adb4 MG |
1235 | SetPageReserved(page); |
1236 | } | |
1237 | } | |
92923ca3 NZ |
1238 | } |
1239 | ||
ec95f53a KM |
1240 | static void __free_pages_ok(struct page *page, unsigned int order) |
1241 | { | |
95e34412 | 1242 | int migratetype; |
dc4b0caf | 1243 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1244 | |
e2769dbd | 1245 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1246 | return; |
1247 | ||
cfc47a28 | 1248 | migratetype = get_pfnblock_migratetype(page, pfn); |
dc4b0caf | 1249 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
1da177e4 LT |
1250 | } |
1251 | ||
949698a3 | 1252 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1253 | { |
c3993076 | 1254 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1255 | struct page *p = page; |
c3993076 | 1256 | unsigned int loop; |
a226f6c8 | 1257 | |
e2d0bd2b YL |
1258 | prefetchw(p); |
1259 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1260 | prefetchw(p + 1); | |
c3993076 JW |
1261 | __ClearPageReserved(p); |
1262 | set_page_count(p, 0); | |
a226f6c8 | 1263 | } |
e2d0bd2b YL |
1264 | __ClearPageReserved(p); |
1265 | set_page_count(p, 0); | |
c3993076 | 1266 | |
e2d0bd2b | 1267 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1268 | set_page_refcounted(page); |
1269 | __free_pages(page, order); | |
a226f6c8 DH |
1270 | } |
1271 | ||
75a592a4 MG |
1272 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1273 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1274 | |
75a592a4 MG |
1275 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1276 | ||
1277 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1278 | { | |
7ace9917 | 1279 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1280 | int nid; |
1281 | ||
7ace9917 | 1282 | spin_lock(&early_pfn_lock); |
75a592a4 | 1283 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1284 | if (nid < 0) |
e4568d38 | 1285 | nid = first_online_node; |
7ace9917 MG |
1286 | spin_unlock(&early_pfn_lock); |
1287 | ||
1288 | return nid; | |
75a592a4 MG |
1289 | } |
1290 | #endif | |
1291 | ||
1292 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
1293 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1294 | struct mminit_pfnnid_cache *state) | |
1295 | { | |
1296 | int nid; | |
1297 | ||
1298 | nid = __early_pfn_to_nid(pfn, state); | |
1299 | if (nid >= 0 && nid != node) | |
1300 | return false; | |
1301 | return true; | |
1302 | } | |
1303 | ||
1304 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1305 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1306 | { | |
1307 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1308 | } | |
1309 | ||
1310 | #else | |
1311 | ||
1312 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1313 | { | |
1314 | return true; | |
1315 | } | |
1316 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1317 | struct mminit_pfnnid_cache *state) | |
1318 | { | |
1319 | return true; | |
1320 | } | |
1321 | #endif | |
1322 | ||
1323 | ||
0e1cc95b | 1324 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1325 | unsigned int order) |
1326 | { | |
1327 | if (early_page_uninitialised(pfn)) | |
1328 | return; | |
949698a3 | 1329 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1330 | } |
1331 | ||
7cf91a98 JK |
1332 | /* |
1333 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1334 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1335 | * with the migration of free compaction scanner. The scanners then need to | |
1336 | * use only pfn_valid_within() check for arches that allow holes within | |
1337 | * pageblocks. | |
1338 | * | |
1339 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1340 | * | |
1341 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1342 | * i.e. it's possible that all pages within a zones range of pages do not | |
1343 | * belong to a single zone. We assume that a border between node0 and node1 | |
1344 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1345 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1346 | * the first and last page of a pageblock and avoid checking each individual | |
1347 | * page in a pageblock. | |
1348 | */ | |
1349 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1350 | unsigned long end_pfn, struct zone *zone) | |
1351 | { | |
1352 | struct page *start_page; | |
1353 | struct page *end_page; | |
1354 | ||
1355 | /* end_pfn is one past the range we are checking */ | |
1356 | end_pfn--; | |
1357 | ||
1358 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1359 | return NULL; | |
1360 | ||
1361 | start_page = pfn_to_page(start_pfn); | |
1362 | ||
1363 | if (page_zone(start_page) != zone) | |
1364 | return NULL; | |
1365 | ||
1366 | end_page = pfn_to_page(end_pfn); | |
1367 | ||
1368 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1369 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1370 | return NULL; | |
1371 | ||
1372 | return start_page; | |
1373 | } | |
1374 | ||
1375 | void set_zone_contiguous(struct zone *zone) | |
1376 | { | |
1377 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1378 | unsigned long block_end_pfn; | |
1379 | ||
1380 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1381 | for (; block_start_pfn < zone_end_pfn(zone); | |
1382 | block_start_pfn = block_end_pfn, | |
1383 | block_end_pfn += pageblock_nr_pages) { | |
1384 | ||
1385 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1386 | ||
1387 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1388 | block_end_pfn, zone)) | |
1389 | return; | |
1390 | } | |
1391 | ||
1392 | /* We confirm that there is no hole */ | |
1393 | zone->contiguous = true; | |
1394 | } | |
1395 | ||
1396 | void clear_zone_contiguous(struct zone *zone) | |
1397 | { | |
1398 | zone->contiguous = false; | |
1399 | } | |
1400 | ||
7e18adb4 | 1401 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
0e1cc95b | 1402 | static void __init deferred_free_range(struct page *page, |
a4de83dd MG |
1403 | unsigned long pfn, int nr_pages) |
1404 | { | |
1405 | int i; | |
1406 | ||
1407 | if (!page) | |
1408 | return; | |
1409 | ||
1410 | /* Free a large naturally-aligned chunk if possible */ | |
e780149b XQ |
1411 | if (nr_pages == pageblock_nr_pages && |
1412 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1413 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
e780149b | 1414 | __free_pages_boot_core(page, pageblock_order); |
a4de83dd MG |
1415 | return; |
1416 | } | |
1417 | ||
e780149b XQ |
1418 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1419 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1420 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
949698a3 | 1421 | __free_pages_boot_core(page, 0); |
e780149b | 1422 | } |
a4de83dd MG |
1423 | } |
1424 | ||
d3cd131d NS |
1425 | /* Completion tracking for deferred_init_memmap() threads */ |
1426 | static atomic_t pgdat_init_n_undone __initdata; | |
1427 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1428 | ||
1429 | static inline void __init pgdat_init_report_one_done(void) | |
1430 | { | |
1431 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1432 | complete(&pgdat_init_all_done_comp); | |
1433 | } | |
0e1cc95b | 1434 | |
7e18adb4 | 1435 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1436 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1437 | { |
0e1cc95b MG |
1438 | pg_data_t *pgdat = data; |
1439 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1440 | struct mminit_pfnnid_cache nid_init_state = { }; |
1441 | unsigned long start = jiffies; | |
1442 | unsigned long nr_pages = 0; | |
1443 | unsigned long walk_start, walk_end; | |
1444 | int i, zid; | |
1445 | struct zone *zone; | |
7e18adb4 | 1446 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
0e1cc95b | 1447 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
7e18adb4 | 1448 | |
0e1cc95b | 1449 | if (first_init_pfn == ULONG_MAX) { |
d3cd131d | 1450 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1451 | return 0; |
1452 | } | |
1453 | ||
1454 | /* Bind memory initialisation thread to a local node if possible */ | |
1455 | if (!cpumask_empty(cpumask)) | |
1456 | set_cpus_allowed_ptr(current, cpumask); | |
7e18adb4 MG |
1457 | |
1458 | /* Sanity check boundaries */ | |
1459 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1460 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1461 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1462 | ||
1463 | /* Only the highest zone is deferred so find it */ | |
1464 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1465 | zone = pgdat->node_zones + zid; | |
1466 | if (first_init_pfn < zone_end_pfn(zone)) | |
1467 | break; | |
1468 | } | |
1469 | ||
1470 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { | |
1471 | unsigned long pfn, end_pfn; | |
54608c3f | 1472 | struct page *page = NULL; |
a4de83dd MG |
1473 | struct page *free_base_page = NULL; |
1474 | unsigned long free_base_pfn = 0; | |
1475 | int nr_to_free = 0; | |
7e18adb4 MG |
1476 | |
1477 | end_pfn = min(walk_end, zone_end_pfn(zone)); | |
1478 | pfn = first_init_pfn; | |
1479 | if (pfn < walk_start) | |
1480 | pfn = walk_start; | |
1481 | if (pfn < zone->zone_start_pfn) | |
1482 | pfn = zone->zone_start_pfn; | |
1483 | ||
1484 | for (; pfn < end_pfn; pfn++) { | |
54608c3f | 1485 | if (!pfn_valid_within(pfn)) |
a4de83dd | 1486 | goto free_range; |
7e18adb4 | 1487 | |
54608c3f MG |
1488 | /* |
1489 | * Ensure pfn_valid is checked every | |
e780149b | 1490 | * pageblock_nr_pages for memory holes |
54608c3f | 1491 | */ |
e780149b | 1492 | if ((pfn & (pageblock_nr_pages - 1)) == 0) { |
54608c3f MG |
1493 | if (!pfn_valid(pfn)) { |
1494 | page = NULL; | |
a4de83dd | 1495 | goto free_range; |
54608c3f MG |
1496 | } |
1497 | } | |
1498 | ||
1499 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { | |
1500 | page = NULL; | |
a4de83dd | 1501 | goto free_range; |
54608c3f MG |
1502 | } |
1503 | ||
1504 | /* Minimise pfn page lookups and scheduler checks */ | |
e780149b | 1505 | if (page && (pfn & (pageblock_nr_pages - 1)) != 0) { |
54608c3f MG |
1506 | page++; |
1507 | } else { | |
a4de83dd MG |
1508 | nr_pages += nr_to_free; |
1509 | deferred_free_range(free_base_page, | |
1510 | free_base_pfn, nr_to_free); | |
1511 | free_base_page = NULL; | |
1512 | free_base_pfn = nr_to_free = 0; | |
1513 | ||
54608c3f MG |
1514 | page = pfn_to_page(pfn); |
1515 | cond_resched(); | |
1516 | } | |
7e18adb4 MG |
1517 | |
1518 | if (page->flags) { | |
1519 | VM_BUG_ON(page_zone(page) != zone); | |
a4de83dd | 1520 | goto free_range; |
7e18adb4 MG |
1521 | } |
1522 | ||
1523 | __init_single_page(page, pfn, zid, nid); | |
a4de83dd MG |
1524 | if (!free_base_page) { |
1525 | free_base_page = page; | |
1526 | free_base_pfn = pfn; | |
1527 | nr_to_free = 0; | |
1528 | } | |
1529 | nr_to_free++; | |
1530 | ||
1531 | /* Where possible, batch up pages for a single free */ | |
1532 | continue; | |
1533 | free_range: | |
1534 | /* Free the current block of pages to allocator */ | |
1535 | nr_pages += nr_to_free; | |
1536 | deferred_free_range(free_base_page, free_base_pfn, | |
1537 | nr_to_free); | |
1538 | free_base_page = NULL; | |
1539 | free_base_pfn = nr_to_free = 0; | |
7e18adb4 | 1540 | } |
e780149b XQ |
1541 | /* Free the last block of pages to allocator */ |
1542 | nr_pages += nr_to_free; | |
1543 | deferred_free_range(free_base_page, free_base_pfn, nr_to_free); | |
a4de83dd | 1544 | |
7e18adb4 MG |
1545 | first_init_pfn = max(end_pfn, first_init_pfn); |
1546 | } | |
1547 | ||
1548 | /* Sanity check that the next zone really is unpopulated */ | |
1549 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1550 | ||
0e1cc95b | 1551 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1552 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1553 | |
1554 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1555 | return 0; |
1556 | } | |
7cf91a98 | 1557 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1558 | |
1559 | void __init page_alloc_init_late(void) | |
1560 | { | |
7cf91a98 JK |
1561 | struct zone *zone; |
1562 | ||
1563 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1564 | int nid; |
1565 | ||
d3cd131d NS |
1566 | /* There will be num_node_state(N_MEMORY) threads */ |
1567 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1568 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1569 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1570 | } | |
1571 | ||
1572 | /* Block until all are initialised */ | |
d3cd131d | 1573 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da MG |
1574 | |
1575 | /* Reinit limits that are based on free pages after the kernel is up */ | |
1576 | files_maxfiles_init(); | |
7cf91a98 JK |
1577 | #endif |
1578 | ||
1579 | for_each_populated_zone(zone) | |
1580 | set_zone_contiguous(zone); | |
7e18adb4 | 1581 | } |
7e18adb4 | 1582 | |
47118af0 | 1583 | #ifdef CONFIG_CMA |
9cf510a5 | 1584 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1585 | void __init init_cma_reserved_pageblock(struct page *page) |
1586 | { | |
1587 | unsigned i = pageblock_nr_pages; | |
1588 | struct page *p = page; | |
1589 | ||
1590 | do { | |
1591 | __ClearPageReserved(p); | |
1592 | set_page_count(p, 0); | |
1593 | } while (++p, --i); | |
1594 | ||
47118af0 | 1595 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1596 | |
1597 | if (pageblock_order >= MAX_ORDER) { | |
1598 | i = pageblock_nr_pages; | |
1599 | p = page; | |
1600 | do { | |
1601 | set_page_refcounted(p); | |
1602 | __free_pages(p, MAX_ORDER - 1); | |
1603 | p += MAX_ORDER_NR_PAGES; | |
1604 | } while (i -= MAX_ORDER_NR_PAGES); | |
1605 | } else { | |
1606 | set_page_refcounted(page); | |
1607 | __free_pages(page, pageblock_order); | |
1608 | } | |
1609 | ||
3dcc0571 | 1610 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1611 | } |
1612 | #endif | |
1da177e4 LT |
1613 | |
1614 | /* | |
1615 | * The order of subdivision here is critical for the IO subsystem. | |
1616 | * Please do not alter this order without good reasons and regression | |
1617 | * testing. Specifically, as large blocks of memory are subdivided, | |
1618 | * the order in which smaller blocks are delivered depends on the order | |
1619 | * they're subdivided in this function. This is the primary factor | |
1620 | * influencing the order in which pages are delivered to the IO | |
1621 | * subsystem according to empirical testing, and this is also justified | |
1622 | * by considering the behavior of a buddy system containing a single | |
1623 | * large block of memory acted on by a series of small allocations. | |
1624 | * This behavior is a critical factor in sglist merging's success. | |
1625 | * | |
6d49e352 | 1626 | * -- nyc |
1da177e4 | 1627 | */ |
085cc7d5 | 1628 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1629 | int low, int high, struct free_area *area, |
1630 | int migratetype) | |
1da177e4 LT |
1631 | { |
1632 | unsigned long size = 1 << high; | |
1633 | ||
1634 | while (high > low) { | |
1635 | area--; | |
1636 | high--; | |
1637 | size >>= 1; | |
309381fe | 1638 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1639 | |
acbc15a4 JK |
1640 | /* |
1641 | * Mark as guard pages (or page), that will allow to | |
1642 | * merge back to allocator when buddy will be freed. | |
1643 | * Corresponding page table entries will not be touched, | |
1644 | * pages will stay not present in virtual address space | |
1645 | */ | |
1646 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 1647 | continue; |
acbc15a4 | 1648 | |
b2a0ac88 | 1649 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1650 | area->nr_free++; |
1651 | set_page_order(&page[size], high); | |
1652 | } | |
1da177e4 LT |
1653 | } |
1654 | ||
4e611801 | 1655 | static void check_new_page_bad(struct page *page) |
1da177e4 | 1656 | { |
4e611801 VB |
1657 | const char *bad_reason = NULL; |
1658 | unsigned long bad_flags = 0; | |
7bfec6f4 | 1659 | |
53f9263b | 1660 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1661 | bad_reason = "nonzero mapcount"; |
1662 | if (unlikely(page->mapping != NULL)) | |
1663 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1664 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1665 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1666 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1667 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1668 | bad_flags = __PG_HWPOISON; | |
e570f56c NH |
1669 | /* Don't complain about hwpoisoned pages */ |
1670 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1671 | return; | |
f4c18e6f | 1672 | } |
f0b791a3 DH |
1673 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1674 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1675 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1676 | } | |
9edad6ea JW |
1677 | #ifdef CONFIG_MEMCG |
1678 | if (unlikely(page->mem_cgroup)) | |
1679 | bad_reason = "page still charged to cgroup"; | |
1680 | #endif | |
4e611801 VB |
1681 | bad_page(page, bad_reason, bad_flags); |
1682 | } | |
1683 | ||
1684 | /* | |
1685 | * This page is about to be returned from the page allocator | |
1686 | */ | |
1687 | static inline int check_new_page(struct page *page) | |
1688 | { | |
1689 | if (likely(page_expected_state(page, | |
1690 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1691 | return 0; | |
1692 | ||
1693 | check_new_page_bad(page); | |
1694 | return 1; | |
2a7684a2 WF |
1695 | } |
1696 | ||
1414c7f4 LA |
1697 | static inline bool free_pages_prezeroed(bool poisoned) |
1698 | { | |
1699 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
1700 | page_poisoning_enabled() && poisoned; | |
1701 | } | |
1702 | ||
479f854a MG |
1703 | #ifdef CONFIG_DEBUG_VM |
1704 | static bool check_pcp_refill(struct page *page) | |
1705 | { | |
1706 | return false; | |
1707 | } | |
1708 | ||
1709 | static bool check_new_pcp(struct page *page) | |
1710 | { | |
1711 | return check_new_page(page); | |
1712 | } | |
1713 | #else | |
1714 | static bool check_pcp_refill(struct page *page) | |
1715 | { | |
1716 | return check_new_page(page); | |
1717 | } | |
1718 | static bool check_new_pcp(struct page *page) | |
1719 | { | |
1720 | return false; | |
1721 | } | |
1722 | #endif /* CONFIG_DEBUG_VM */ | |
1723 | ||
1724 | static bool check_new_pages(struct page *page, unsigned int order) | |
1725 | { | |
1726 | int i; | |
1727 | for (i = 0; i < (1 << order); i++) { | |
1728 | struct page *p = page + i; | |
1729 | ||
1730 | if (unlikely(check_new_page(p))) | |
1731 | return true; | |
1732 | } | |
1733 | ||
1734 | return false; | |
1735 | } | |
1736 | ||
46f24fd8 JK |
1737 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1738 | gfp_t gfp_flags) | |
1739 | { | |
1740 | set_page_private(page, 0); | |
1741 | set_page_refcounted(page); | |
1742 | ||
1743 | arch_alloc_page(page, order); | |
1744 | kernel_map_pages(page, 1 << order, 1); | |
1745 | kernel_poison_pages(page, 1 << order, 1); | |
1746 | kasan_alloc_pages(page, order); | |
1747 | set_page_owner(page, order, gfp_flags); | |
1748 | } | |
1749 | ||
479f854a | 1750 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1751 | unsigned int alloc_flags) |
2a7684a2 WF |
1752 | { |
1753 | int i; | |
1414c7f4 | 1754 | bool poisoned = true; |
2a7684a2 WF |
1755 | |
1756 | for (i = 0; i < (1 << order); i++) { | |
1757 | struct page *p = page + i; | |
1414c7f4 LA |
1758 | if (poisoned) |
1759 | poisoned &= page_is_poisoned(p); | |
2a7684a2 | 1760 | } |
689bcebf | 1761 | |
46f24fd8 | 1762 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1763 | |
1414c7f4 | 1764 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1765 | for (i = 0; i < (1 << order); i++) |
1766 | clear_highpage(page + i); | |
17cf4406 NP |
1767 | |
1768 | if (order && (gfp_flags & __GFP_COMP)) | |
1769 | prep_compound_page(page, order); | |
1770 | ||
75379191 | 1771 | /* |
2f064f34 | 1772 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1773 | * allocate the page. The expectation is that the caller is taking |
1774 | * steps that will free more memory. The caller should avoid the page | |
1775 | * being used for !PFMEMALLOC purposes. | |
1776 | */ | |
2f064f34 MH |
1777 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1778 | set_page_pfmemalloc(page); | |
1779 | else | |
1780 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1781 | } |
1782 | ||
56fd56b8 MG |
1783 | /* |
1784 | * Go through the free lists for the given migratetype and remove | |
1785 | * the smallest available page from the freelists | |
1786 | */ | |
728ec980 MG |
1787 | static inline |
1788 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1789 | int migratetype) |
1790 | { | |
1791 | unsigned int current_order; | |
b8af2941 | 1792 | struct free_area *area; |
56fd56b8 MG |
1793 | struct page *page; |
1794 | ||
1795 | /* Find a page of the appropriate size in the preferred list */ | |
1796 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1797 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1798 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1799 | struct page, lru); |
a16601c5 GT |
1800 | if (!page) |
1801 | continue; | |
56fd56b8 MG |
1802 | list_del(&page->lru); |
1803 | rmv_page_order(page); | |
1804 | area->nr_free--; | |
56fd56b8 | 1805 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1806 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1807 | return page; |
1808 | } | |
1809 | ||
1810 | return NULL; | |
1811 | } | |
1812 | ||
1813 | ||
b2a0ac88 MG |
1814 | /* |
1815 | * This array describes the order lists are fallen back to when | |
1816 | * the free lists for the desirable migrate type are depleted | |
1817 | */ | |
47118af0 | 1818 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1819 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1820 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1821 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1822 | #ifdef CONFIG_CMA |
974a786e | 1823 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1824 | #endif |
194159fb | 1825 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1826 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1827 | #endif |
b2a0ac88 MG |
1828 | }; |
1829 | ||
dc67647b JK |
1830 | #ifdef CONFIG_CMA |
1831 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1832 | unsigned int order) | |
1833 | { | |
1834 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1835 | } | |
1836 | #else | |
1837 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1838 | unsigned int order) { return NULL; } | |
1839 | #endif | |
1840 | ||
c361be55 MG |
1841 | /* |
1842 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1843 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1844 | * boundary. If alignment is required, use move_freepages_block() |
1845 | */ | |
435b405c | 1846 | int move_freepages(struct zone *zone, |
b69a7288 AB |
1847 | struct page *start_page, struct page *end_page, |
1848 | int migratetype) | |
c361be55 MG |
1849 | { |
1850 | struct page *page; | |
d00181b9 | 1851 | unsigned int order; |
d100313f | 1852 | int pages_moved = 0; |
c361be55 MG |
1853 | |
1854 | #ifndef CONFIG_HOLES_IN_ZONE | |
1855 | /* | |
1856 | * page_zone is not safe to call in this context when | |
1857 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1858 | * anyway as we check zone boundaries in move_freepages_block(). | |
1859 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1860 | * grouping pages by mobility |
c361be55 | 1861 | */ |
97ee4ba7 | 1862 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1863 | #endif |
1864 | ||
1865 | for (page = start_page; page <= end_page;) { | |
1866 | if (!pfn_valid_within(page_to_pfn(page))) { | |
1867 | page++; | |
1868 | continue; | |
1869 | } | |
1870 | ||
f073bdc5 AB |
1871 | /* Make sure we are not inadvertently changing nodes */ |
1872 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
1873 | ||
c361be55 MG |
1874 | if (!PageBuddy(page)) { |
1875 | page++; | |
1876 | continue; | |
1877 | } | |
1878 | ||
1879 | order = page_order(page); | |
84be48d8 KS |
1880 | list_move(&page->lru, |
1881 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 1882 | page += 1 << order; |
d100313f | 1883 | pages_moved += 1 << order; |
c361be55 MG |
1884 | } |
1885 | ||
d100313f | 1886 | return pages_moved; |
c361be55 MG |
1887 | } |
1888 | ||
ee6f509c | 1889 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1890 | int migratetype) |
c361be55 MG |
1891 | { |
1892 | unsigned long start_pfn, end_pfn; | |
1893 | struct page *start_page, *end_page; | |
1894 | ||
1895 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1896 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1897 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1898 | end_page = start_page + pageblock_nr_pages - 1; |
1899 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1900 | |
1901 | /* Do not cross zone boundaries */ | |
108bcc96 | 1902 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1903 | start_page = page; |
108bcc96 | 1904 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1905 | return 0; |
1906 | ||
1907 | return move_freepages(zone, start_page, end_page, migratetype); | |
1908 | } | |
1909 | ||
2f66a68f MG |
1910 | static void change_pageblock_range(struct page *pageblock_page, |
1911 | int start_order, int migratetype) | |
1912 | { | |
1913 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1914 | ||
1915 | while (nr_pageblocks--) { | |
1916 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1917 | pageblock_page += pageblock_nr_pages; | |
1918 | } | |
1919 | } | |
1920 | ||
fef903ef | 1921 | /* |
9c0415eb VB |
1922 | * When we are falling back to another migratetype during allocation, try to |
1923 | * steal extra free pages from the same pageblocks to satisfy further | |
1924 | * allocations, instead of polluting multiple pageblocks. | |
1925 | * | |
1926 | * If we are stealing a relatively large buddy page, it is likely there will | |
1927 | * be more free pages in the pageblock, so try to steal them all. For | |
1928 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1929 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1930 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1931 | * pageblocks. | |
fef903ef | 1932 | */ |
4eb7dce6 JK |
1933 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1934 | { | |
1935 | /* | |
1936 | * Leaving this order check is intended, although there is | |
1937 | * relaxed order check in next check. The reason is that | |
1938 | * we can actually steal whole pageblock if this condition met, | |
1939 | * but, below check doesn't guarantee it and that is just heuristic | |
1940 | * so could be changed anytime. | |
1941 | */ | |
1942 | if (order >= pageblock_order) | |
1943 | return true; | |
1944 | ||
1945 | if (order >= pageblock_order / 2 || | |
1946 | start_mt == MIGRATE_RECLAIMABLE || | |
1947 | start_mt == MIGRATE_UNMOVABLE || | |
1948 | page_group_by_mobility_disabled) | |
1949 | return true; | |
1950 | ||
1951 | return false; | |
1952 | } | |
1953 | ||
1954 | /* | |
1955 | * This function implements actual steal behaviour. If order is large enough, | |
1956 | * we can steal whole pageblock. If not, we first move freepages in this | |
1957 | * pageblock and check whether half of pages are moved or not. If half of | |
1958 | * pages are moved, we can change migratetype of pageblock and permanently | |
1959 | * use it's pages as requested migratetype in the future. | |
1960 | */ | |
1961 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1962 | int start_type) | |
fef903ef | 1963 | { |
d00181b9 | 1964 | unsigned int current_order = page_order(page); |
4eb7dce6 | 1965 | int pages; |
fef903ef | 1966 | |
fef903ef SB |
1967 | /* Take ownership for orders >= pageblock_order */ |
1968 | if (current_order >= pageblock_order) { | |
1969 | change_pageblock_range(page, current_order, start_type); | |
3a1086fb | 1970 | return; |
fef903ef SB |
1971 | } |
1972 | ||
4eb7dce6 | 1973 | pages = move_freepages_block(zone, page, start_type); |
fef903ef | 1974 | |
4eb7dce6 JK |
1975 | /* Claim the whole block if over half of it is free */ |
1976 | if (pages >= (1 << (pageblock_order-1)) || | |
1977 | page_group_by_mobility_disabled) | |
1978 | set_pageblock_migratetype(page, start_type); | |
1979 | } | |
1980 | ||
2149cdae JK |
1981 | /* |
1982 | * Check whether there is a suitable fallback freepage with requested order. | |
1983 | * If only_stealable is true, this function returns fallback_mt only if | |
1984 | * we can steal other freepages all together. This would help to reduce | |
1985 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1986 | */ | |
1987 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1988 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1989 | { |
1990 | int i; | |
1991 | int fallback_mt; | |
1992 | ||
1993 | if (area->nr_free == 0) | |
1994 | return -1; | |
1995 | ||
1996 | *can_steal = false; | |
1997 | for (i = 0;; i++) { | |
1998 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 1999 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2000 | break; |
2001 | ||
2002 | if (list_empty(&area->free_list[fallback_mt])) | |
2003 | continue; | |
fef903ef | 2004 | |
4eb7dce6 JK |
2005 | if (can_steal_fallback(order, migratetype)) |
2006 | *can_steal = true; | |
2007 | ||
2149cdae JK |
2008 | if (!only_stealable) |
2009 | return fallback_mt; | |
2010 | ||
2011 | if (*can_steal) | |
2012 | return fallback_mt; | |
fef903ef | 2013 | } |
4eb7dce6 JK |
2014 | |
2015 | return -1; | |
fef903ef SB |
2016 | } |
2017 | ||
0aaa29a5 MG |
2018 | /* |
2019 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2020 | * there are no empty page blocks that contain a page with a suitable order | |
2021 | */ | |
2022 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2023 | unsigned int alloc_order) | |
2024 | { | |
2025 | int mt; | |
2026 | unsigned long max_managed, flags; | |
2027 | ||
2028 | /* | |
2029 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2030 | * Check is race-prone but harmless. | |
2031 | */ | |
2032 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
2033 | if (zone->nr_reserved_highatomic >= max_managed) | |
2034 | return; | |
2035 | ||
2036 | spin_lock_irqsave(&zone->lock, flags); | |
2037 | ||
2038 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2039 | if (zone->nr_reserved_highatomic >= max_managed) | |
2040 | goto out_unlock; | |
2041 | ||
2042 | /* Yoink! */ | |
2043 | mt = get_pageblock_migratetype(page); | |
2044 | if (mt != MIGRATE_HIGHATOMIC && | |
2045 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { | |
2046 | zone->nr_reserved_highatomic += pageblock_nr_pages; | |
2047 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
2048 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); | |
2049 | } | |
2050 | ||
2051 | out_unlock: | |
2052 | spin_unlock_irqrestore(&zone->lock, flags); | |
2053 | } | |
2054 | ||
2055 | /* | |
2056 | * Used when an allocation is about to fail under memory pressure. This | |
2057 | * potentially hurts the reliability of high-order allocations when under | |
2058 | * intense memory pressure but failed atomic allocations should be easier | |
2059 | * to recover from than an OOM. | |
29fac03b MK |
2060 | * |
2061 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2062 | * pageblock is exhausted. | |
0aaa29a5 | 2063 | */ |
29fac03b MK |
2064 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2065 | bool force) | |
0aaa29a5 MG |
2066 | { |
2067 | struct zonelist *zonelist = ac->zonelist; | |
2068 | unsigned long flags; | |
2069 | struct zoneref *z; | |
2070 | struct zone *zone; | |
2071 | struct page *page; | |
2072 | int order; | |
04c8716f | 2073 | bool ret; |
0aaa29a5 MG |
2074 | |
2075 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2076 | ac->nodemask) { | |
29fac03b MK |
2077 | /* |
2078 | * Preserve at least one pageblock unless memory pressure | |
2079 | * is really high. | |
2080 | */ | |
2081 | if (!force && zone->nr_reserved_highatomic <= | |
2082 | pageblock_nr_pages) | |
0aaa29a5 MG |
2083 | continue; |
2084 | ||
2085 | spin_lock_irqsave(&zone->lock, flags); | |
2086 | for (order = 0; order < MAX_ORDER; order++) { | |
2087 | struct free_area *area = &(zone->free_area[order]); | |
2088 | ||
a16601c5 GT |
2089 | page = list_first_entry_or_null( |
2090 | &area->free_list[MIGRATE_HIGHATOMIC], | |
2091 | struct page, lru); | |
2092 | if (!page) | |
0aaa29a5 MG |
2093 | continue; |
2094 | ||
0aaa29a5 | 2095 | /* |
4855e4a7 MK |
2096 | * In page freeing path, migratetype change is racy so |
2097 | * we can counter several free pages in a pageblock | |
2098 | * in this loop althoug we changed the pageblock type | |
2099 | * from highatomic to ac->migratetype. So we should | |
2100 | * adjust the count once. | |
0aaa29a5 | 2101 | */ |
4855e4a7 MK |
2102 | if (get_pageblock_migratetype(page) == |
2103 | MIGRATE_HIGHATOMIC) { | |
2104 | /* | |
2105 | * It should never happen but changes to | |
2106 | * locking could inadvertently allow a per-cpu | |
2107 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2108 | * while unreserving so be safe and watch for | |
2109 | * underflows. | |
2110 | */ | |
2111 | zone->nr_reserved_highatomic -= min( | |
2112 | pageblock_nr_pages, | |
2113 | zone->nr_reserved_highatomic); | |
2114 | } | |
0aaa29a5 MG |
2115 | |
2116 | /* | |
2117 | * Convert to ac->migratetype and avoid the normal | |
2118 | * pageblock stealing heuristics. Minimally, the caller | |
2119 | * is doing the work and needs the pages. More | |
2120 | * importantly, if the block was always converted to | |
2121 | * MIGRATE_UNMOVABLE or another type then the number | |
2122 | * of pageblocks that cannot be completely freed | |
2123 | * may increase. | |
2124 | */ | |
2125 | set_pageblock_migratetype(page, ac->migratetype); | |
04c8716f | 2126 | ret = move_freepages_block(zone, page, ac->migratetype); |
29fac03b MK |
2127 | if (ret) { |
2128 | spin_unlock_irqrestore(&zone->lock, flags); | |
2129 | return ret; | |
2130 | } | |
0aaa29a5 MG |
2131 | } |
2132 | spin_unlock_irqrestore(&zone->lock, flags); | |
2133 | } | |
04c8716f MK |
2134 | |
2135 | return false; | |
0aaa29a5 MG |
2136 | } |
2137 | ||
b2a0ac88 | 2138 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 2139 | static inline struct page * |
7aeb09f9 | 2140 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 2141 | { |
b8af2941 | 2142 | struct free_area *area; |
7aeb09f9 | 2143 | unsigned int current_order; |
b2a0ac88 | 2144 | struct page *page; |
4eb7dce6 JK |
2145 | int fallback_mt; |
2146 | bool can_steal; | |
b2a0ac88 MG |
2147 | |
2148 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
2149 | for (current_order = MAX_ORDER-1; |
2150 | current_order >= order && current_order <= MAX_ORDER-1; | |
2151 | --current_order) { | |
4eb7dce6 JK |
2152 | area = &(zone->free_area[current_order]); |
2153 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2154 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2155 | if (fallback_mt == -1) |
2156 | continue; | |
b2a0ac88 | 2157 | |
a16601c5 | 2158 | page = list_first_entry(&area->free_list[fallback_mt], |
4eb7dce6 | 2159 | struct page, lru); |
88ed365e MK |
2160 | if (can_steal && |
2161 | get_pageblock_migratetype(page) != MIGRATE_HIGHATOMIC) | |
4eb7dce6 | 2162 | steal_suitable_fallback(zone, page, start_migratetype); |
b2a0ac88 | 2163 | |
4eb7dce6 JK |
2164 | /* Remove the page from the freelists */ |
2165 | area->nr_free--; | |
2166 | list_del(&page->lru); | |
2167 | rmv_page_order(page); | |
3a1086fb | 2168 | |
4eb7dce6 JK |
2169 | expand(zone, page, order, current_order, area, |
2170 | start_migratetype); | |
2171 | /* | |
bb14c2c7 | 2172 | * The pcppage_migratetype may differ from pageblock's |
4eb7dce6 | 2173 | * migratetype depending on the decisions in |
bb14c2c7 VB |
2174 | * find_suitable_fallback(). This is OK as long as it does not |
2175 | * differ for MIGRATE_CMA pageblocks. Those can be used as | |
2176 | * fallback only via special __rmqueue_cma_fallback() function | |
4eb7dce6 | 2177 | */ |
bb14c2c7 | 2178 | set_pcppage_migratetype(page, start_migratetype); |
e0fff1bd | 2179 | |
4eb7dce6 JK |
2180 | trace_mm_page_alloc_extfrag(page, order, current_order, |
2181 | start_migratetype, fallback_mt); | |
e0fff1bd | 2182 | |
4eb7dce6 | 2183 | return page; |
b2a0ac88 MG |
2184 | } |
2185 | ||
728ec980 | 2186 | return NULL; |
b2a0ac88 MG |
2187 | } |
2188 | ||
56fd56b8 | 2189 | /* |
1da177e4 LT |
2190 | * Do the hard work of removing an element from the buddy allocator. |
2191 | * Call me with the zone->lock already held. | |
2192 | */ | |
b2a0ac88 | 2193 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
6ac0206b | 2194 | int migratetype) |
1da177e4 | 2195 | { |
1da177e4 LT |
2196 | struct page *page; |
2197 | ||
56fd56b8 | 2198 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2199 | if (unlikely(!page)) { |
dc67647b JK |
2200 | if (migratetype == MIGRATE_MOVABLE) |
2201 | page = __rmqueue_cma_fallback(zone, order); | |
2202 | ||
2203 | if (!page) | |
2204 | page = __rmqueue_fallback(zone, order, migratetype); | |
728ec980 MG |
2205 | } |
2206 | ||
0d3d062a | 2207 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2208 | return page; |
1da177e4 LT |
2209 | } |
2210 | ||
5f63b720 | 2211 | /* |
1da177e4 LT |
2212 | * Obtain a specified number of elements from the buddy allocator, all under |
2213 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2214 | * Returns the number of new pages which were placed at *list. | |
2215 | */ | |
5f63b720 | 2216 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2217 | unsigned long count, struct list_head *list, |
b745bc85 | 2218 | int migratetype, bool cold) |
1da177e4 | 2219 | { |
a6de734b | 2220 | int i, alloced = 0; |
374ad05a | 2221 | unsigned long flags; |
5f63b720 | 2222 | |
374ad05a | 2223 | spin_lock_irqsave(&zone->lock, flags); |
1da177e4 | 2224 | for (i = 0; i < count; ++i) { |
6ac0206b | 2225 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 2226 | if (unlikely(page == NULL)) |
1da177e4 | 2227 | break; |
81eabcbe | 2228 | |
479f854a MG |
2229 | if (unlikely(check_pcp_refill(page))) |
2230 | continue; | |
2231 | ||
81eabcbe MG |
2232 | /* |
2233 | * Split buddy pages returned by expand() are received here | |
2234 | * in physical page order. The page is added to the callers and | |
2235 | * list and the list head then moves forward. From the callers | |
2236 | * perspective, the linked list is ordered by page number in | |
2237 | * some conditions. This is useful for IO devices that can | |
2238 | * merge IO requests if the physical pages are ordered | |
2239 | * properly. | |
2240 | */ | |
b745bc85 | 2241 | if (likely(!cold)) |
e084b2d9 MG |
2242 | list_add(&page->lru, list); |
2243 | else | |
2244 | list_add_tail(&page->lru, list); | |
81eabcbe | 2245 | list = &page->lru; |
a6de734b | 2246 | alloced++; |
bb14c2c7 | 2247 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2248 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2249 | -(1 << order)); | |
1da177e4 | 2250 | } |
a6de734b MG |
2251 | |
2252 | /* | |
2253 | * i pages were removed from the buddy list even if some leak due | |
2254 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2255 | * on i. Do not confuse with 'alloced' which is the number of | |
2256 | * pages added to the pcp list. | |
2257 | */ | |
f2260e6b | 2258 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
374ad05a | 2259 | spin_unlock_irqrestore(&zone->lock, flags); |
a6de734b | 2260 | return alloced; |
1da177e4 LT |
2261 | } |
2262 | ||
4ae7c039 | 2263 | #ifdef CONFIG_NUMA |
8fce4d8e | 2264 | /* |
4037d452 CL |
2265 | * Called from the vmstat counter updater to drain pagesets of this |
2266 | * currently executing processor on remote nodes after they have | |
2267 | * expired. | |
2268 | * | |
879336c3 CL |
2269 | * Note that this function must be called with the thread pinned to |
2270 | * a single processor. | |
8fce4d8e | 2271 | */ |
4037d452 | 2272 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2273 | { |
4ae7c039 | 2274 | unsigned long flags; |
7be12fc9 | 2275 | int to_drain, batch; |
4ae7c039 | 2276 | |
4037d452 | 2277 | local_irq_save(flags); |
4db0c3c2 | 2278 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2279 | to_drain = min(pcp->count, batch); |
2a13515c KM |
2280 | if (to_drain > 0) { |
2281 | free_pcppages_bulk(zone, to_drain, pcp); | |
2282 | pcp->count -= to_drain; | |
2283 | } | |
4037d452 | 2284 | local_irq_restore(flags); |
4ae7c039 CL |
2285 | } |
2286 | #endif | |
2287 | ||
9f8f2172 | 2288 | /* |
93481ff0 | 2289 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2290 | * |
2291 | * The processor must either be the current processor and the | |
2292 | * thread pinned to the current processor or a processor that | |
2293 | * is not online. | |
2294 | */ | |
93481ff0 | 2295 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2296 | { |
c54ad30c | 2297 | unsigned long flags; |
93481ff0 VB |
2298 | struct per_cpu_pageset *pset; |
2299 | struct per_cpu_pages *pcp; | |
1da177e4 | 2300 | |
93481ff0 VB |
2301 | local_irq_save(flags); |
2302 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2303 | |
93481ff0 VB |
2304 | pcp = &pset->pcp; |
2305 | if (pcp->count) { | |
2306 | free_pcppages_bulk(zone, pcp->count, pcp); | |
2307 | pcp->count = 0; | |
2308 | } | |
2309 | local_irq_restore(flags); | |
2310 | } | |
3dfa5721 | 2311 | |
93481ff0 VB |
2312 | /* |
2313 | * Drain pcplists of all zones on the indicated processor. | |
2314 | * | |
2315 | * The processor must either be the current processor and the | |
2316 | * thread pinned to the current processor or a processor that | |
2317 | * is not online. | |
2318 | */ | |
2319 | static void drain_pages(unsigned int cpu) | |
2320 | { | |
2321 | struct zone *zone; | |
2322 | ||
2323 | for_each_populated_zone(zone) { | |
2324 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2325 | } |
2326 | } | |
1da177e4 | 2327 | |
9f8f2172 CL |
2328 | /* |
2329 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2330 | * |
2331 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2332 | * the single zone's pages. | |
9f8f2172 | 2333 | */ |
93481ff0 | 2334 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2335 | { |
93481ff0 VB |
2336 | int cpu = smp_processor_id(); |
2337 | ||
2338 | if (zone) | |
2339 | drain_pages_zone(cpu, zone); | |
2340 | else | |
2341 | drain_pages(cpu); | |
9f8f2172 CL |
2342 | } |
2343 | ||
0ccce3b9 MG |
2344 | static void drain_local_pages_wq(struct work_struct *work) |
2345 | { | |
a459eeb7 MH |
2346 | /* |
2347 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2348 | * we can race with cpu offline when the WQ can move this from | |
2349 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2350 | * cpu which is allright but we also have to make sure to not move to | |
2351 | * a different one. | |
2352 | */ | |
2353 | preempt_disable(); | |
0ccce3b9 | 2354 | drain_local_pages(NULL); |
a459eeb7 | 2355 | preempt_enable(); |
0ccce3b9 MG |
2356 | } |
2357 | ||
9f8f2172 | 2358 | /* |
74046494 GBY |
2359 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2360 | * | |
93481ff0 VB |
2361 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2362 | * | |
0ccce3b9 | 2363 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2364 | */ |
93481ff0 | 2365 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2366 | { |
74046494 | 2367 | int cpu; |
74046494 GBY |
2368 | |
2369 | /* | |
2370 | * Allocate in the BSS so we wont require allocation in | |
2371 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2372 | */ | |
2373 | static cpumask_t cpus_with_pcps; | |
2374 | ||
0ccce3b9 MG |
2375 | /* Workqueues cannot recurse */ |
2376 | if (current->flags & PF_WQ_WORKER) | |
2377 | return; | |
2378 | ||
bd233f53 MG |
2379 | /* |
2380 | * Do not drain if one is already in progress unless it's specific to | |
2381 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2382 | * the drain to be complete when the call returns. | |
2383 | */ | |
2384 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2385 | if (!zone) | |
2386 | return; | |
2387 | mutex_lock(&pcpu_drain_mutex); | |
2388 | } | |
0ccce3b9 | 2389 | |
74046494 GBY |
2390 | /* |
2391 | * We don't care about racing with CPU hotplug event | |
2392 | * as offline notification will cause the notified | |
2393 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2394 | * disables preemption as part of its processing | |
2395 | */ | |
2396 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2397 | struct per_cpu_pageset *pcp; |
2398 | struct zone *z; | |
74046494 | 2399 | bool has_pcps = false; |
93481ff0 VB |
2400 | |
2401 | if (zone) { | |
74046494 | 2402 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2403 | if (pcp->pcp.count) |
74046494 | 2404 | has_pcps = true; |
93481ff0 VB |
2405 | } else { |
2406 | for_each_populated_zone(z) { | |
2407 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2408 | if (pcp->pcp.count) { | |
2409 | has_pcps = true; | |
2410 | break; | |
2411 | } | |
74046494 GBY |
2412 | } |
2413 | } | |
93481ff0 | 2414 | |
74046494 GBY |
2415 | if (has_pcps) |
2416 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2417 | else | |
2418 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2419 | } | |
0ccce3b9 | 2420 | |
bd233f53 MG |
2421 | for_each_cpu(cpu, &cpus_with_pcps) { |
2422 | struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu); | |
2423 | INIT_WORK(work, drain_local_pages_wq); | |
2424 | schedule_work_on(cpu, work); | |
0ccce3b9 | 2425 | } |
bd233f53 MG |
2426 | for_each_cpu(cpu, &cpus_with_pcps) |
2427 | flush_work(per_cpu_ptr(&pcpu_drain, cpu)); | |
2428 | ||
2429 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2430 | } |
2431 | ||
296699de | 2432 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
2433 | |
2434 | void mark_free_pages(struct zone *zone) | |
2435 | { | |
f623f0db RW |
2436 | unsigned long pfn, max_zone_pfn; |
2437 | unsigned long flags; | |
7aeb09f9 | 2438 | unsigned int order, t; |
86760a2c | 2439 | struct page *page; |
1da177e4 | 2440 | |
8080fc03 | 2441 | if (zone_is_empty(zone)) |
1da177e4 LT |
2442 | return; |
2443 | ||
2444 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2445 | |
108bcc96 | 2446 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2447 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2448 | if (pfn_valid(pfn)) { | |
86760a2c | 2449 | page = pfn_to_page(pfn); |
ba6b0979 JK |
2450 | |
2451 | if (page_zone(page) != zone) | |
2452 | continue; | |
2453 | ||
7be98234 RW |
2454 | if (!swsusp_page_is_forbidden(page)) |
2455 | swsusp_unset_page_free(page); | |
f623f0db | 2456 | } |
1da177e4 | 2457 | |
b2a0ac88 | 2458 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2459 | list_for_each_entry(page, |
2460 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2461 | unsigned long i; |
1da177e4 | 2462 | |
86760a2c | 2463 | pfn = page_to_pfn(page); |
f623f0db | 2464 | for (i = 0; i < (1UL << order); i++) |
7be98234 | 2465 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 2466 | } |
b2a0ac88 | 2467 | } |
1da177e4 LT |
2468 | spin_unlock_irqrestore(&zone->lock, flags); |
2469 | } | |
e2c55dc8 | 2470 | #endif /* CONFIG_PM */ |
1da177e4 | 2471 | |
1da177e4 LT |
2472 | /* |
2473 | * Free a 0-order page | |
b745bc85 | 2474 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 2475 | */ |
b745bc85 | 2476 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
2477 | { |
2478 | struct zone *zone = page_zone(page); | |
2479 | struct per_cpu_pages *pcp; | |
dc4b0caf | 2480 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 2481 | int migratetype; |
1da177e4 | 2482 | |
374ad05a MG |
2483 | if (in_interrupt()) { |
2484 | __free_pages_ok(page, 0); | |
2485 | return; | |
2486 | } | |
2487 | ||
4db7548c | 2488 | if (!free_pcp_prepare(page)) |
689bcebf HD |
2489 | return; |
2490 | ||
dc4b0caf | 2491 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2492 | set_pcppage_migratetype(page, migratetype); |
374ad05a | 2493 | preempt_disable(); |
da456f14 | 2494 | |
5f8dcc21 MG |
2495 | /* |
2496 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2497 | * Free ISOLATE pages back to the allocator because they are being | |
2498 | * offlined but treat RESERVE as movable pages so we can get those | |
2499 | * areas back if necessary. Otherwise, we may have to free | |
2500 | * excessively into the page allocator | |
2501 | */ | |
2502 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2503 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2504 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
2505 | goto out; |
2506 | } | |
2507 | migratetype = MIGRATE_MOVABLE; | |
2508 | } | |
2509 | ||
374ad05a | 2510 | __count_vm_event(PGFREE); |
99dcc3e5 | 2511 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 2512 | if (!cold) |
5f8dcc21 | 2513 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
2514 | else |
2515 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 2516 | pcp->count++; |
48db57f8 | 2517 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2518 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
2519 | free_pcppages_bulk(zone, batch, pcp); |
2520 | pcp->count -= batch; | |
48db57f8 | 2521 | } |
5f8dcc21 MG |
2522 | |
2523 | out: | |
374ad05a | 2524 | preempt_enable(); |
1da177e4 LT |
2525 | } |
2526 | ||
cc59850e KK |
2527 | /* |
2528 | * Free a list of 0-order pages | |
2529 | */ | |
b745bc85 | 2530 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
2531 | { |
2532 | struct page *page, *next; | |
2533 | ||
2534 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 2535 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
2536 | free_hot_cold_page(page, cold); |
2537 | } | |
2538 | } | |
2539 | ||
8dfcc9ba NP |
2540 | /* |
2541 | * split_page takes a non-compound higher-order page, and splits it into | |
2542 | * n (1<<order) sub-pages: page[0..n] | |
2543 | * Each sub-page must be freed individually. | |
2544 | * | |
2545 | * Note: this is probably too low level an operation for use in drivers. | |
2546 | * Please consult with lkml before using this in your driver. | |
2547 | */ | |
2548 | void split_page(struct page *page, unsigned int order) | |
2549 | { | |
2550 | int i; | |
2551 | ||
309381fe SL |
2552 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2553 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
2554 | |
2555 | #ifdef CONFIG_KMEMCHECK | |
2556 | /* | |
2557 | * Split shadow pages too, because free(page[0]) would | |
2558 | * otherwise free the whole shadow. | |
2559 | */ | |
2560 | if (kmemcheck_page_is_tracked(page)) | |
2561 | split_page(virt_to_page(page[0].shadow), order); | |
2562 | #endif | |
2563 | ||
a9627bc5 | 2564 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2565 | set_page_refcounted(page + i); |
a9627bc5 | 2566 | split_page_owner(page, order); |
8dfcc9ba | 2567 | } |
5853ff23 | 2568 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2569 | |
3c605096 | 2570 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2571 | { |
748446bb MG |
2572 | unsigned long watermark; |
2573 | struct zone *zone; | |
2139cbe6 | 2574 | int mt; |
748446bb MG |
2575 | |
2576 | BUG_ON(!PageBuddy(page)); | |
2577 | ||
2578 | zone = page_zone(page); | |
2e30abd1 | 2579 | mt = get_pageblock_migratetype(page); |
748446bb | 2580 | |
194159fb | 2581 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
2582 | /* |
2583 | * Obey watermarks as if the page was being allocated. We can | |
2584 | * emulate a high-order watermark check with a raised order-0 | |
2585 | * watermark, because we already know our high-order page | |
2586 | * exists. | |
2587 | */ | |
2588 | watermark = min_wmark_pages(zone) + (1UL << order); | |
984fdba6 | 2589 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
2590 | return 0; |
2591 | ||
8fb74b9f | 2592 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2593 | } |
748446bb MG |
2594 | |
2595 | /* Remove page from free list */ | |
2596 | list_del(&page->lru); | |
2597 | zone->free_area[order].nr_free--; | |
2598 | rmv_page_order(page); | |
2139cbe6 | 2599 | |
400bc7fd | 2600 | /* |
2601 | * Set the pageblock if the isolated page is at least half of a | |
2602 | * pageblock | |
2603 | */ | |
748446bb MG |
2604 | if (order >= pageblock_order - 1) { |
2605 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2606 | for (; page < endpage; page += pageblock_nr_pages) { |
2607 | int mt = get_pageblock_migratetype(page); | |
88ed365e MK |
2608 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
2609 | && mt != MIGRATE_HIGHATOMIC) | |
47118af0 MN |
2610 | set_pageblock_migratetype(page, |
2611 | MIGRATE_MOVABLE); | |
2612 | } | |
748446bb MG |
2613 | } |
2614 | ||
f3a14ced | 2615 | |
8fb74b9f | 2616 | return 1UL << order; |
1fb3f8ca MG |
2617 | } |
2618 | ||
060e7417 MG |
2619 | /* |
2620 | * Update NUMA hit/miss statistics | |
2621 | * | |
2622 | * Must be called with interrupts disabled. | |
060e7417 | 2623 | */ |
41b6167e | 2624 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
2625 | { |
2626 | #ifdef CONFIG_NUMA | |
060e7417 MG |
2627 | enum zone_stat_item local_stat = NUMA_LOCAL; |
2628 | ||
2df26639 | 2629 | if (z->node != numa_node_id()) |
060e7417 | 2630 | local_stat = NUMA_OTHER; |
060e7417 | 2631 | |
2df26639 | 2632 | if (z->node == preferred_zone->node) |
060e7417 | 2633 | __inc_zone_state(z, NUMA_HIT); |
2df26639 | 2634 | else { |
060e7417 MG |
2635 | __inc_zone_state(z, NUMA_MISS); |
2636 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); | |
2637 | } | |
2df26639 | 2638 | __inc_zone_state(z, local_stat); |
060e7417 MG |
2639 | #endif |
2640 | } | |
2641 | ||
066b2393 MG |
2642 | /* Remove page from the per-cpu list, caller must protect the list */ |
2643 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
2644 | bool cold, struct per_cpu_pages *pcp, | |
2645 | struct list_head *list) | |
2646 | { | |
2647 | struct page *page; | |
2648 | ||
374ad05a MG |
2649 | VM_BUG_ON(in_interrupt()); |
2650 | ||
066b2393 MG |
2651 | do { |
2652 | if (list_empty(list)) { | |
2653 | pcp->count += rmqueue_bulk(zone, 0, | |
2654 | pcp->batch, list, | |
2655 | migratetype, cold); | |
2656 | if (unlikely(list_empty(list))) | |
2657 | return NULL; | |
2658 | } | |
2659 | ||
2660 | if (cold) | |
2661 | page = list_last_entry(list, struct page, lru); | |
2662 | else | |
2663 | page = list_first_entry(list, struct page, lru); | |
2664 | ||
2665 | list_del(&page->lru); | |
2666 | pcp->count--; | |
2667 | } while (check_new_pcp(page)); | |
2668 | ||
2669 | return page; | |
2670 | } | |
2671 | ||
2672 | /* Lock and remove page from the per-cpu list */ | |
2673 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
2674 | struct zone *zone, unsigned int order, | |
2675 | gfp_t gfp_flags, int migratetype) | |
2676 | { | |
2677 | struct per_cpu_pages *pcp; | |
2678 | struct list_head *list; | |
2679 | bool cold = ((gfp_flags & __GFP_COLD) != 0); | |
2680 | struct page *page; | |
066b2393 | 2681 | |
374ad05a | 2682 | preempt_disable(); |
066b2393 MG |
2683 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2684 | list = &pcp->lists[migratetype]; | |
2685 | page = __rmqueue_pcplist(zone, migratetype, cold, pcp, list); | |
2686 | if (page) { | |
2687 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
2688 | zone_statistics(preferred_zone, zone); | |
2689 | } | |
374ad05a | 2690 | preempt_enable(); |
066b2393 MG |
2691 | return page; |
2692 | } | |
2693 | ||
1da177e4 | 2694 | /* |
75379191 | 2695 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2696 | */ |
0a15c3e9 | 2697 | static inline |
066b2393 | 2698 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 2699 | struct zone *zone, unsigned int order, |
c603844b MG |
2700 | gfp_t gfp_flags, unsigned int alloc_flags, |
2701 | int migratetype) | |
1da177e4 LT |
2702 | { |
2703 | unsigned long flags; | |
689bcebf | 2704 | struct page *page; |
1da177e4 | 2705 | |
374ad05a | 2706 | if (likely(order == 0) && !in_interrupt()) { |
066b2393 MG |
2707 | page = rmqueue_pcplist(preferred_zone, zone, order, |
2708 | gfp_flags, migratetype); | |
2709 | goto out; | |
2710 | } | |
83b9355b | 2711 | |
066b2393 MG |
2712 | /* |
2713 | * We most definitely don't want callers attempting to | |
2714 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2715 | */ | |
2716 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
2717 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 2718 | |
066b2393 MG |
2719 | do { |
2720 | page = NULL; | |
2721 | if (alloc_flags & ALLOC_HARDER) { | |
2722 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2723 | if (page) | |
2724 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
2725 | } | |
a74609fa | 2726 | if (!page) |
066b2393 MG |
2727 | page = __rmqueue(zone, order, migratetype); |
2728 | } while (page && check_new_pages(page, order)); | |
2729 | spin_unlock(&zone->lock); | |
2730 | if (!page) | |
2731 | goto failed; | |
2732 | __mod_zone_freepage_state(zone, -(1 << order), | |
2733 | get_pcppage_migratetype(page)); | |
1da177e4 | 2734 | |
16709d1d | 2735 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 2736 | zone_statistics(preferred_zone, zone); |
a74609fa | 2737 | local_irq_restore(flags); |
1da177e4 | 2738 | |
066b2393 MG |
2739 | out: |
2740 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); | |
1da177e4 | 2741 | return page; |
a74609fa NP |
2742 | |
2743 | failed: | |
2744 | local_irq_restore(flags); | |
a74609fa | 2745 | return NULL; |
1da177e4 LT |
2746 | } |
2747 | ||
933e312e AM |
2748 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2749 | ||
b2588c4b | 2750 | static struct { |
933e312e AM |
2751 | struct fault_attr attr; |
2752 | ||
621a5f7a | 2753 | bool ignore_gfp_highmem; |
71baba4b | 2754 | bool ignore_gfp_reclaim; |
54114994 | 2755 | u32 min_order; |
933e312e AM |
2756 | } fail_page_alloc = { |
2757 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 2758 | .ignore_gfp_reclaim = true, |
621a5f7a | 2759 | .ignore_gfp_highmem = true, |
54114994 | 2760 | .min_order = 1, |
933e312e AM |
2761 | }; |
2762 | ||
2763 | static int __init setup_fail_page_alloc(char *str) | |
2764 | { | |
2765 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
2766 | } | |
2767 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
2768 | ||
deaf386e | 2769 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2770 | { |
54114994 | 2771 | if (order < fail_page_alloc.min_order) |
deaf386e | 2772 | return false; |
933e312e | 2773 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 2774 | return false; |
933e312e | 2775 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 2776 | return false; |
71baba4b MG |
2777 | if (fail_page_alloc.ignore_gfp_reclaim && |
2778 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 2779 | return false; |
933e312e AM |
2780 | |
2781 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
2782 | } | |
2783 | ||
2784 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
2785 | ||
2786 | static int __init fail_page_alloc_debugfs(void) | |
2787 | { | |
f4ae40a6 | 2788 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 2789 | struct dentry *dir; |
933e312e | 2790 | |
dd48c085 AM |
2791 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2792 | &fail_page_alloc.attr); | |
2793 | if (IS_ERR(dir)) | |
2794 | return PTR_ERR(dir); | |
933e312e | 2795 | |
b2588c4b | 2796 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 2797 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
2798 | goto fail; |
2799 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
2800 | &fail_page_alloc.ignore_gfp_highmem)) | |
2801 | goto fail; | |
2802 | if (!debugfs_create_u32("min-order", mode, dir, | |
2803 | &fail_page_alloc.min_order)) | |
2804 | goto fail; | |
2805 | ||
2806 | return 0; | |
2807 | fail: | |
dd48c085 | 2808 | debugfs_remove_recursive(dir); |
933e312e | 2809 | |
b2588c4b | 2810 | return -ENOMEM; |
933e312e AM |
2811 | } |
2812 | ||
2813 | late_initcall(fail_page_alloc_debugfs); | |
2814 | ||
2815 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2816 | ||
2817 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2818 | ||
deaf386e | 2819 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2820 | { |
deaf386e | 2821 | return false; |
933e312e AM |
2822 | } |
2823 | ||
2824 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2825 | ||
1da177e4 | 2826 | /* |
97a16fc8 MG |
2827 | * Return true if free base pages are above 'mark'. For high-order checks it |
2828 | * will return true of the order-0 watermark is reached and there is at least | |
2829 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
2830 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 2831 | */ |
86a294a8 MH |
2832 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
2833 | int classzone_idx, unsigned int alloc_flags, | |
2834 | long free_pages) | |
1da177e4 | 2835 | { |
d23ad423 | 2836 | long min = mark; |
1da177e4 | 2837 | int o; |
c603844b | 2838 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
1da177e4 | 2839 | |
0aaa29a5 | 2840 | /* free_pages may go negative - that's OK */ |
df0a6daa | 2841 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 2842 | |
7fb1d9fc | 2843 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2844 | min -= min / 2; |
0aaa29a5 MG |
2845 | |
2846 | /* | |
2847 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
2848 | * the high-atomic reserves. This will over-estimate the size of the | |
2849 | * atomic reserve but it avoids a search. | |
2850 | */ | |
97a16fc8 | 2851 | if (likely(!alloc_harder)) |
0aaa29a5 MG |
2852 | free_pages -= z->nr_reserved_highatomic; |
2853 | else | |
1da177e4 | 2854 | min -= min / 4; |
e2b19197 | 2855 | |
d95ea5d1 BZ |
2856 | #ifdef CONFIG_CMA |
2857 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2858 | if (!(alloc_flags & ALLOC_CMA)) | |
97a16fc8 | 2859 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2860 | #endif |
026b0814 | 2861 | |
97a16fc8 MG |
2862 | /* |
2863 | * Check watermarks for an order-0 allocation request. If these | |
2864 | * are not met, then a high-order request also cannot go ahead | |
2865 | * even if a suitable page happened to be free. | |
2866 | */ | |
2867 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 2868 | return false; |
1da177e4 | 2869 | |
97a16fc8 MG |
2870 | /* If this is an order-0 request then the watermark is fine */ |
2871 | if (!order) | |
2872 | return true; | |
2873 | ||
2874 | /* For a high-order request, check at least one suitable page is free */ | |
2875 | for (o = order; o < MAX_ORDER; o++) { | |
2876 | struct free_area *area = &z->free_area[o]; | |
2877 | int mt; | |
2878 | ||
2879 | if (!area->nr_free) | |
2880 | continue; | |
2881 | ||
2882 | if (alloc_harder) | |
2883 | return true; | |
1da177e4 | 2884 | |
97a16fc8 MG |
2885 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
2886 | if (!list_empty(&area->free_list[mt])) | |
2887 | return true; | |
2888 | } | |
2889 | ||
2890 | #ifdef CONFIG_CMA | |
2891 | if ((alloc_flags & ALLOC_CMA) && | |
2892 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
2893 | return true; | |
2894 | } | |
2895 | #endif | |
1da177e4 | 2896 | } |
97a16fc8 | 2897 | return false; |
88f5acf8 MG |
2898 | } |
2899 | ||
7aeb09f9 | 2900 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 2901 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
2902 | { |
2903 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2904 | zone_page_state(z, NR_FREE_PAGES)); | |
2905 | } | |
2906 | ||
48ee5f36 MG |
2907 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
2908 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
2909 | { | |
2910 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2911 | long cma_pages = 0; | |
2912 | ||
2913 | #ifdef CONFIG_CMA | |
2914 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2915 | if (!(alloc_flags & ALLOC_CMA)) | |
2916 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
2917 | #endif | |
2918 | ||
2919 | /* | |
2920 | * Fast check for order-0 only. If this fails then the reserves | |
2921 | * need to be calculated. There is a corner case where the check | |
2922 | * passes but only the high-order atomic reserve are free. If | |
2923 | * the caller is !atomic then it'll uselessly search the free | |
2924 | * list. That corner case is then slower but it is harmless. | |
2925 | */ | |
2926 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) | |
2927 | return true; | |
2928 | ||
2929 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2930 | free_pages); | |
2931 | } | |
2932 | ||
7aeb09f9 | 2933 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 2934 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
2935 | { |
2936 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2937 | ||
2938 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2939 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2940 | ||
e2b19197 | 2941 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 2942 | free_pages); |
1da177e4 LT |
2943 | } |
2944 | ||
9276b1bc | 2945 | #ifdef CONFIG_NUMA |
957f822a DR |
2946 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2947 | { | |
e02dc017 | 2948 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
5f7a75ac | 2949 | RECLAIM_DISTANCE; |
957f822a | 2950 | } |
9276b1bc | 2951 | #else /* CONFIG_NUMA */ |
957f822a DR |
2952 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2953 | { | |
2954 | return true; | |
2955 | } | |
9276b1bc PJ |
2956 | #endif /* CONFIG_NUMA */ |
2957 | ||
7fb1d9fc | 2958 | /* |
0798e519 | 2959 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
2960 | * a page. |
2961 | */ | |
2962 | static struct page * | |
a9263751 VB |
2963 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
2964 | const struct alloc_context *ac) | |
753ee728 | 2965 | { |
c33d6c06 | 2966 | struct zoneref *z = ac->preferred_zoneref; |
5117f45d | 2967 | struct zone *zone; |
3b8c0be4 MG |
2968 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
2969 | ||
7fb1d9fc | 2970 | /* |
9276b1bc | 2971 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 2972 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 2973 | */ |
c33d6c06 | 2974 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 2975 | ac->nodemask) { |
be06af00 | 2976 | struct page *page; |
e085dbc5 JW |
2977 | unsigned long mark; |
2978 | ||
664eedde MG |
2979 | if (cpusets_enabled() && |
2980 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 2981 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 2982 | continue; |
a756cf59 JW |
2983 | /* |
2984 | * When allocating a page cache page for writing, we | |
281e3726 MG |
2985 | * want to get it from a node that is within its dirty |
2986 | * limit, such that no single node holds more than its | |
a756cf59 | 2987 | * proportional share of globally allowed dirty pages. |
281e3726 | 2988 | * The dirty limits take into account the node's |
a756cf59 JW |
2989 | * lowmem reserves and high watermark so that kswapd |
2990 | * should be able to balance it without having to | |
2991 | * write pages from its LRU list. | |
2992 | * | |
a756cf59 | 2993 | * XXX: For now, allow allocations to potentially |
281e3726 | 2994 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 2995 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 2996 | * which is important when on a NUMA setup the allowed |
281e3726 | 2997 | * nodes are together not big enough to reach the |
a756cf59 | 2998 | * global limit. The proper fix for these situations |
281e3726 | 2999 | * will require awareness of nodes in the |
a756cf59 JW |
3000 | * dirty-throttling and the flusher threads. |
3001 | */ | |
3b8c0be4 MG |
3002 | if (ac->spread_dirty_pages) { |
3003 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3004 | continue; | |
3005 | ||
3006 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3007 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3008 | continue; | |
3009 | } | |
3010 | } | |
7fb1d9fc | 3011 | |
e085dbc5 | 3012 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 3013 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 3014 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
3015 | int ret; |
3016 | ||
5dab2911 MG |
3017 | /* Checked here to keep the fast path fast */ |
3018 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3019 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3020 | goto try_this_zone; | |
3021 | ||
a5f5f91d | 3022 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3023 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3024 | continue; |
3025 | ||
a5f5f91d | 3026 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3027 | switch (ret) { |
a5f5f91d | 3028 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3029 | /* did not scan */ |
cd38b115 | 3030 | continue; |
a5f5f91d | 3031 | case NODE_RECLAIM_FULL: |
fa5e084e | 3032 | /* scanned but unreclaimable */ |
cd38b115 | 3033 | continue; |
fa5e084e MG |
3034 | default: |
3035 | /* did we reclaim enough */ | |
fed2719e | 3036 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3037 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3038 | goto try_this_zone; |
3039 | ||
fed2719e | 3040 | continue; |
0798e519 | 3041 | } |
7fb1d9fc RS |
3042 | } |
3043 | ||
fa5e084e | 3044 | try_this_zone: |
066b2393 | 3045 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3046 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3047 | if (page) { |
479f854a | 3048 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3049 | |
3050 | /* | |
3051 | * If this is a high-order atomic allocation then check | |
3052 | * if the pageblock should be reserved for the future | |
3053 | */ | |
3054 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3055 | reserve_highatomic_pageblock(page, zone, order); | |
3056 | ||
75379191 VB |
3057 | return page; |
3058 | } | |
54a6eb5c | 3059 | } |
9276b1bc | 3060 | |
4ffeaf35 | 3061 | return NULL; |
753ee728 MH |
3062 | } |
3063 | ||
29423e77 DR |
3064 | /* |
3065 | * Large machines with many possible nodes should not always dump per-node | |
3066 | * meminfo in irq context. | |
3067 | */ | |
3068 | static inline bool should_suppress_show_mem(void) | |
3069 | { | |
3070 | bool ret = false; | |
3071 | ||
3072 | #if NODES_SHIFT > 8 | |
3073 | ret = in_interrupt(); | |
3074 | #endif | |
3075 | return ret; | |
3076 | } | |
3077 | ||
9af744d7 | 3078 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3079 | { |
a238ab5b | 3080 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
aa187507 | 3081 | static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1); |
a238ab5b | 3082 | |
aa187507 | 3083 | if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs)) |
a238ab5b DH |
3084 | return; |
3085 | ||
3086 | /* | |
3087 | * This documents exceptions given to allocations in certain | |
3088 | * contexts that are allowed to allocate outside current's set | |
3089 | * of allowed nodes. | |
3090 | */ | |
3091 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
3092 | if (test_thread_flag(TIF_MEMDIE) || | |
3093 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
3094 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3095 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3096 | filter &= ~SHOW_MEM_FILTER_NODES; |
3097 | ||
9af744d7 | 3098 | show_mem(filter, nodemask); |
aa187507 MH |
3099 | } |
3100 | ||
a8e99259 | 3101 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3102 | { |
3103 | struct va_format vaf; | |
3104 | va_list args; | |
3105 | static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL, | |
3106 | DEFAULT_RATELIMIT_BURST); | |
3107 | ||
3108 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || | |
3109 | debug_guardpage_minorder() > 0) | |
3110 | return; | |
3111 | ||
7877cdcc | 3112 | pr_warn("%s: ", current->comm); |
3ee9a4f0 | 3113 | |
7877cdcc MH |
3114 | va_start(args, fmt); |
3115 | vaf.fmt = fmt; | |
3116 | vaf.va = &args; | |
3117 | pr_cont("%pV", &vaf); | |
3118 | va_end(args); | |
3ee9a4f0 | 3119 | |
685dbf6f DR |
3120 | pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask, &gfp_mask); |
3121 | if (nodemask) | |
3122 | pr_cont("%*pbl\n", nodemask_pr_args(nodemask)); | |
3123 | else | |
3124 | pr_cont("(null)\n"); | |
3125 | ||
a8e99259 | 3126 | cpuset_print_current_mems_allowed(); |
3ee9a4f0 | 3127 | |
a238ab5b | 3128 | dump_stack(); |
685dbf6f | 3129 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3130 | } |
3131 | ||
6c18ba7a MH |
3132 | static inline struct page * |
3133 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3134 | unsigned int alloc_flags, | |
3135 | const struct alloc_context *ac) | |
3136 | { | |
3137 | struct page *page; | |
3138 | ||
3139 | page = get_page_from_freelist(gfp_mask, order, | |
3140 | alloc_flags|ALLOC_CPUSET, ac); | |
3141 | /* | |
3142 | * fallback to ignore cpuset restriction if our nodes | |
3143 | * are depleted | |
3144 | */ | |
3145 | if (!page) | |
3146 | page = get_page_from_freelist(gfp_mask, order, | |
3147 | alloc_flags, ac); | |
3148 | ||
3149 | return page; | |
3150 | } | |
3151 | ||
11e33f6a MG |
3152 | static inline struct page * |
3153 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3154 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3155 | { |
6e0fc46d DR |
3156 | struct oom_control oc = { |
3157 | .zonelist = ac->zonelist, | |
3158 | .nodemask = ac->nodemask, | |
2a966b77 | 3159 | .memcg = NULL, |
6e0fc46d DR |
3160 | .gfp_mask = gfp_mask, |
3161 | .order = order, | |
6e0fc46d | 3162 | }; |
11e33f6a MG |
3163 | struct page *page; |
3164 | ||
9879de73 JW |
3165 | *did_some_progress = 0; |
3166 | ||
9879de73 | 3167 | /* |
dc56401f JW |
3168 | * Acquire the oom lock. If that fails, somebody else is |
3169 | * making progress for us. | |
9879de73 | 3170 | */ |
dc56401f | 3171 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3172 | *did_some_progress = 1; |
11e33f6a | 3173 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3174 | return NULL; |
3175 | } | |
6b1de916 | 3176 | |
11e33f6a MG |
3177 | /* |
3178 | * Go through the zonelist yet one more time, keep very high watermark | |
3179 | * here, this is only to catch a parallel oom killing, we must fail if | |
3180 | * we're still under heavy pressure. | |
3181 | */ | |
a9263751 VB |
3182 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
3183 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3184 | if (page) |
11e33f6a MG |
3185 | goto out; |
3186 | ||
06ad276a MH |
3187 | /* Coredumps can quickly deplete all memory reserves */ |
3188 | if (current->flags & PF_DUMPCORE) | |
3189 | goto out; | |
3190 | /* The OOM killer will not help higher order allocs */ | |
3191 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3192 | goto out; | |
3193 | /* The OOM killer does not needlessly kill tasks for lowmem */ | |
3194 | if (ac->high_zoneidx < ZONE_NORMAL) | |
3195 | goto out; | |
3196 | if (pm_suspended_storage()) | |
3197 | goto out; | |
3198 | /* | |
3199 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3200 | * other request to make a forward progress. | |
3201 | * We are in an unfortunate situation where out_of_memory cannot | |
3202 | * do much for this context but let's try it to at least get | |
3203 | * access to memory reserved if the current task is killed (see | |
3204 | * out_of_memory). Once filesystems are ready to handle allocation | |
3205 | * failures more gracefully we should just bail out here. | |
3206 | */ | |
3207 | ||
3208 | /* The OOM killer may not free memory on a specific node */ | |
3209 | if (gfp_mask & __GFP_THISNODE) | |
3210 | goto out; | |
3da88fb3 | 3211 | |
11e33f6a | 3212 | /* Exhausted what can be done so it's blamo time */ |
5020e285 | 3213 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3214 | *did_some_progress = 1; |
5020e285 | 3215 | |
6c18ba7a MH |
3216 | /* |
3217 | * Help non-failing allocations by giving them access to memory | |
3218 | * reserves | |
3219 | */ | |
3220 | if (gfp_mask & __GFP_NOFAIL) | |
3221 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3222 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3223 | } |
11e33f6a | 3224 | out: |
dc56401f | 3225 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3226 | return page; |
3227 | } | |
3228 | ||
33c2d214 MH |
3229 | /* |
3230 | * Maximum number of compaction retries wit a progress before OOM | |
3231 | * killer is consider as the only way to move forward. | |
3232 | */ | |
3233 | #define MAX_COMPACT_RETRIES 16 | |
3234 | ||
56de7263 MG |
3235 | #ifdef CONFIG_COMPACTION |
3236 | /* Try memory compaction for high-order allocations before reclaim */ | |
3237 | static struct page * | |
3238 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3239 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3240 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3241 | { |
98dd3b48 | 3242 | struct page *page; |
53853e2d VB |
3243 | |
3244 | if (!order) | |
66199712 | 3245 | return NULL; |
66199712 | 3246 | |
c06b1fca | 3247 | current->flags |= PF_MEMALLOC; |
c5d01d0d | 3248 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
c3486f53 | 3249 | prio); |
c06b1fca | 3250 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 3251 | |
c5d01d0d | 3252 | if (*compact_result <= COMPACT_INACTIVE) |
98dd3b48 | 3253 | return NULL; |
53853e2d | 3254 | |
98dd3b48 VB |
3255 | /* |
3256 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3257 | * count a compaction stall | |
3258 | */ | |
3259 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3260 | |
31a6c190 | 3261 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
53853e2d | 3262 | |
98dd3b48 VB |
3263 | if (page) { |
3264 | struct zone *zone = page_zone(page); | |
53853e2d | 3265 | |
98dd3b48 VB |
3266 | zone->compact_blockskip_flush = false; |
3267 | compaction_defer_reset(zone, order, true); | |
3268 | count_vm_event(COMPACTSUCCESS); | |
3269 | return page; | |
3270 | } | |
56de7263 | 3271 | |
98dd3b48 VB |
3272 | /* |
3273 | * It's bad if compaction run occurs and fails. The most likely reason | |
3274 | * is that pages exist, but not enough to satisfy watermarks. | |
3275 | */ | |
3276 | count_vm_event(COMPACTFAIL); | |
66199712 | 3277 | |
98dd3b48 | 3278 | cond_resched(); |
56de7263 MG |
3279 | |
3280 | return NULL; | |
3281 | } | |
33c2d214 | 3282 | |
3250845d VB |
3283 | static inline bool |
3284 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3285 | enum compact_result compact_result, | |
3286 | enum compact_priority *compact_priority, | |
d9436498 | 3287 | int *compaction_retries) |
3250845d VB |
3288 | { |
3289 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3290 | int min_priority; |
65190cff MH |
3291 | bool ret = false; |
3292 | int retries = *compaction_retries; | |
3293 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3294 | |
3295 | if (!order) | |
3296 | return false; | |
3297 | ||
d9436498 VB |
3298 | if (compaction_made_progress(compact_result)) |
3299 | (*compaction_retries)++; | |
3300 | ||
3250845d VB |
3301 | /* |
3302 | * compaction considers all the zone as desperately out of memory | |
3303 | * so it doesn't really make much sense to retry except when the | |
3304 | * failure could be caused by insufficient priority | |
3305 | */ | |
d9436498 VB |
3306 | if (compaction_failed(compact_result)) |
3307 | goto check_priority; | |
3250845d VB |
3308 | |
3309 | /* | |
3310 | * make sure the compaction wasn't deferred or didn't bail out early | |
3311 | * due to locks contention before we declare that we should give up. | |
3312 | * But do not retry if the given zonelist is not suitable for | |
3313 | * compaction. | |
3314 | */ | |
65190cff MH |
3315 | if (compaction_withdrawn(compact_result)) { |
3316 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
3317 | goto out; | |
3318 | } | |
3250845d VB |
3319 | |
3320 | /* | |
3321 | * !costly requests are much more important than __GFP_REPEAT | |
3322 | * costly ones because they are de facto nofail and invoke OOM | |
3323 | * killer to move on while costly can fail and users are ready | |
3324 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3325 | * would need much more detailed feedback from compaction to | |
3326 | * make a better decision. | |
3327 | */ | |
3328 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3329 | max_retries /= 4; | |
65190cff MH |
3330 | if (*compaction_retries <= max_retries) { |
3331 | ret = true; | |
3332 | goto out; | |
3333 | } | |
3250845d | 3334 | |
d9436498 VB |
3335 | /* |
3336 | * Make sure there are attempts at the highest priority if we exhausted | |
3337 | * all retries or failed at the lower priorities. | |
3338 | */ | |
3339 | check_priority: | |
c2033b00 VB |
3340 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3341 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3342 | |
c2033b00 | 3343 | if (*compact_priority > min_priority) { |
d9436498 VB |
3344 | (*compact_priority)--; |
3345 | *compaction_retries = 0; | |
65190cff | 3346 | ret = true; |
d9436498 | 3347 | } |
65190cff MH |
3348 | out: |
3349 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
3350 | return ret; | |
3250845d | 3351 | } |
56de7263 MG |
3352 | #else |
3353 | static inline struct page * | |
3354 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3355 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3356 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3357 | { |
33c2d214 | 3358 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3359 | return NULL; |
3360 | } | |
33c2d214 MH |
3361 | |
3362 | static inline bool | |
86a294a8 MH |
3363 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3364 | enum compact_result compact_result, | |
a5508cd8 | 3365 | enum compact_priority *compact_priority, |
d9436498 | 3366 | int *compaction_retries) |
33c2d214 | 3367 | { |
31e49bfd MH |
3368 | struct zone *zone; |
3369 | struct zoneref *z; | |
3370 | ||
3371 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3372 | return false; | |
3373 | ||
3374 | /* | |
3375 | * There are setups with compaction disabled which would prefer to loop | |
3376 | * inside the allocator rather than hit the oom killer prematurely. | |
3377 | * Let's give them a good hope and keep retrying while the order-0 | |
3378 | * watermarks are OK. | |
3379 | */ | |
3380 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3381 | ac->nodemask) { | |
3382 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
3383 | ac_classzone_idx(ac), alloc_flags)) | |
3384 | return true; | |
3385 | } | |
33c2d214 MH |
3386 | return false; |
3387 | } | |
3250845d | 3388 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3389 | |
bba90710 MS |
3390 | /* Perform direct synchronous page reclaim */ |
3391 | static int | |
a9263751 VB |
3392 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3393 | const struct alloc_context *ac) | |
11e33f6a | 3394 | { |
11e33f6a | 3395 | struct reclaim_state reclaim_state; |
bba90710 | 3396 | int progress; |
11e33f6a MG |
3397 | |
3398 | cond_resched(); | |
3399 | ||
3400 | /* We now go into synchronous reclaim */ | |
3401 | cpuset_memory_pressure_bump(); | |
c06b1fca | 3402 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
3403 | lockdep_set_current_reclaim_state(gfp_mask); |
3404 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 3405 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3406 | |
a9263751 VB |
3407 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3408 | ac->nodemask); | |
11e33f6a | 3409 | |
c06b1fca | 3410 | current->reclaim_state = NULL; |
11e33f6a | 3411 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 3412 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
3413 | |
3414 | cond_resched(); | |
3415 | ||
bba90710 MS |
3416 | return progress; |
3417 | } | |
3418 | ||
3419 | /* The really slow allocator path where we enter direct reclaim */ | |
3420 | static inline struct page * | |
3421 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3422 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3423 | unsigned long *did_some_progress) |
bba90710 MS |
3424 | { |
3425 | struct page *page = NULL; | |
3426 | bool drained = false; | |
3427 | ||
a9263751 | 3428 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3429 | if (unlikely(!(*did_some_progress))) |
3430 | return NULL; | |
11e33f6a | 3431 | |
9ee493ce | 3432 | retry: |
31a6c190 | 3433 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3434 | |
3435 | /* | |
3436 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3437 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3438 | * Shrink them them and try again | |
9ee493ce MG |
3439 | */ |
3440 | if (!page && !drained) { | |
29fac03b | 3441 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3442 | drain_all_pages(NULL); |
9ee493ce MG |
3443 | drained = true; |
3444 | goto retry; | |
3445 | } | |
3446 | ||
11e33f6a MG |
3447 | return page; |
3448 | } | |
3449 | ||
a9263751 | 3450 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
3451 | { |
3452 | struct zoneref *z; | |
3453 | struct zone *zone; | |
e1a55637 | 3454 | pg_data_t *last_pgdat = NULL; |
3a025760 | 3455 | |
a9263751 | 3456 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
e1a55637 MG |
3457 | ac->high_zoneidx, ac->nodemask) { |
3458 | if (last_pgdat != zone->zone_pgdat) | |
52e9f87a | 3459 | wakeup_kswapd(zone, order, ac->high_zoneidx); |
e1a55637 MG |
3460 | last_pgdat = zone->zone_pgdat; |
3461 | } | |
3a025760 JW |
3462 | } |
3463 | ||
c603844b | 3464 | static inline unsigned int |
341ce06f PZ |
3465 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3466 | { | |
c603844b | 3467 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3468 | |
a56f57ff | 3469 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3470 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3471 | |
341ce06f PZ |
3472 | /* |
3473 | * The caller may dip into page reserves a bit more if the caller | |
3474 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3475 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3476 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3477 | */ |
e6223a3b | 3478 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3479 | |
d0164adc | 3480 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3481 | /* |
b104a35d DR |
3482 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3483 | * if it can't schedule. | |
5c3240d9 | 3484 | */ |
b104a35d | 3485 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3486 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3487 | /* |
b104a35d | 3488 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3489 | * comment for __cpuset_node_allowed(). |
523b9458 | 3490 | */ |
341ce06f | 3491 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3492 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3493 | alloc_flags |= ALLOC_HARDER; |
3494 | ||
d95ea5d1 | 3495 | #ifdef CONFIG_CMA |
43e7a34d | 3496 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
3497 | alloc_flags |= ALLOC_CMA; |
3498 | #endif | |
341ce06f PZ |
3499 | return alloc_flags; |
3500 | } | |
3501 | ||
072bb0aa MG |
3502 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3503 | { | |
31a6c190 VB |
3504 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) |
3505 | return false; | |
3506 | ||
3507 | if (gfp_mask & __GFP_MEMALLOC) | |
3508 | return true; | |
3509 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) | |
3510 | return true; | |
3511 | if (!in_interrupt() && | |
3512 | ((current->flags & PF_MEMALLOC) || | |
3513 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
3514 | return true; | |
3515 | ||
3516 | return false; | |
072bb0aa MG |
3517 | } |
3518 | ||
0a0337e0 MH |
3519 | /* |
3520 | * Maximum number of reclaim retries without any progress before OOM killer | |
3521 | * is consider as the only way to move forward. | |
3522 | */ | |
3523 | #define MAX_RECLAIM_RETRIES 16 | |
3524 | ||
3525 | /* | |
3526 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
3527 | * for the given allocation request. | |
3528 | * The reclaim feedback represented by did_some_progress (any progress during | |
7854ea6c MH |
3529 | * the last reclaim round) and no_progress_loops (number of reclaim rounds without |
3530 | * any progress in a row) is considered as well as the reclaimable pages on the | |
3531 | * applicable zone list (with a backoff mechanism which is a function of | |
3532 | * no_progress_loops). | |
0a0337e0 MH |
3533 | * |
3534 | * Returns true if a retry is viable or false to enter the oom path. | |
3535 | */ | |
3536 | static inline bool | |
3537 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
3538 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 3539 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
3540 | { |
3541 | struct zone *zone; | |
3542 | struct zoneref *z; | |
3543 | ||
423b452e VB |
3544 | /* |
3545 | * Costly allocations might have made a progress but this doesn't mean | |
3546 | * their order will become available due to high fragmentation so | |
3547 | * always increment the no progress counter for them | |
3548 | */ | |
3549 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
3550 | *no_progress_loops = 0; | |
3551 | else | |
3552 | (*no_progress_loops)++; | |
3553 | ||
0a0337e0 MH |
3554 | /* |
3555 | * Make sure we converge to OOM if we cannot make any progress | |
3556 | * several times in the row. | |
3557 | */ | |
04c8716f MK |
3558 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
3559 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 3560 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 3561 | } |
0a0337e0 | 3562 | |
bca67592 MG |
3563 | /* |
3564 | * Keep reclaiming pages while there is a chance this will lead | |
3565 | * somewhere. If none of the target zones can satisfy our allocation | |
3566 | * request even if all reclaimable pages are considered then we are | |
3567 | * screwed and have to go OOM. | |
0a0337e0 MH |
3568 | */ |
3569 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3570 | ac->nodemask) { | |
3571 | unsigned long available; | |
ede37713 | 3572 | unsigned long reclaimable; |
d379f01d MH |
3573 | unsigned long min_wmark = min_wmark_pages(zone); |
3574 | bool wmark; | |
0a0337e0 | 3575 | |
5a1c84b4 | 3576 | available = reclaimable = zone_reclaimable_pages(zone); |
423b452e | 3577 | available -= DIV_ROUND_UP((*no_progress_loops) * available, |
0a0337e0 | 3578 | MAX_RECLAIM_RETRIES); |
5a1c84b4 | 3579 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
3580 | |
3581 | /* | |
3582 | * Would the allocation succeed if we reclaimed the whole | |
5a1c84b4 | 3583 | * available? |
0a0337e0 | 3584 | */ |
d379f01d MH |
3585 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
3586 | ac_classzone_idx(ac), alloc_flags, available); | |
3587 | trace_reclaim_retry_zone(z, order, reclaimable, | |
3588 | available, min_wmark, *no_progress_loops, wmark); | |
3589 | if (wmark) { | |
ede37713 MH |
3590 | /* |
3591 | * If we didn't make any progress and have a lot of | |
3592 | * dirty + writeback pages then we should wait for | |
3593 | * an IO to complete to slow down the reclaim and | |
3594 | * prevent from pre mature OOM | |
3595 | */ | |
3596 | if (!did_some_progress) { | |
11fb9989 | 3597 | unsigned long write_pending; |
ede37713 | 3598 | |
5a1c84b4 MG |
3599 | write_pending = zone_page_state_snapshot(zone, |
3600 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 3601 | |
11fb9989 | 3602 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
3603 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
3604 | return true; | |
3605 | } | |
3606 | } | |
5a1c84b4 | 3607 | |
ede37713 MH |
3608 | /* |
3609 | * Memory allocation/reclaim might be called from a WQ | |
3610 | * context and the current implementation of the WQ | |
3611 | * concurrency control doesn't recognize that | |
3612 | * a particular WQ is congested if the worker thread is | |
3613 | * looping without ever sleeping. Therefore we have to | |
3614 | * do a short sleep here rather than calling | |
3615 | * cond_resched(). | |
3616 | */ | |
3617 | if (current->flags & PF_WQ_WORKER) | |
3618 | schedule_timeout_uninterruptible(1); | |
3619 | else | |
3620 | cond_resched(); | |
3621 | ||
0a0337e0 MH |
3622 | return true; |
3623 | } | |
3624 | } | |
3625 | ||
3626 | return false; | |
3627 | } | |
3628 | ||
11e33f6a MG |
3629 | static inline struct page * |
3630 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3631 | struct alloc_context *ac) |
11e33f6a | 3632 | { |
d0164adc | 3633 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
11e33f6a | 3634 | struct page *page = NULL; |
c603844b | 3635 | unsigned int alloc_flags; |
11e33f6a | 3636 | unsigned long did_some_progress; |
5ce9bfef | 3637 | enum compact_priority compact_priority; |
c5d01d0d | 3638 | enum compact_result compact_result; |
5ce9bfef VB |
3639 | int compaction_retries; |
3640 | int no_progress_loops; | |
63f53dea MH |
3641 | unsigned long alloc_start = jiffies; |
3642 | unsigned int stall_timeout = 10 * HZ; | |
5ce9bfef | 3643 | unsigned int cpuset_mems_cookie; |
1da177e4 | 3644 | |
72807a74 MG |
3645 | /* |
3646 | * In the slowpath, we sanity check order to avoid ever trying to | |
3647 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
3648 | * be using allocators in order of preference for an area that is | |
3649 | * too large. | |
3650 | */ | |
1fc28b70 MG |
3651 | if (order >= MAX_ORDER) { |
3652 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 3653 | return NULL; |
1fc28b70 | 3654 | } |
1da177e4 | 3655 | |
d0164adc MG |
3656 | /* |
3657 | * We also sanity check to catch abuse of atomic reserves being used by | |
3658 | * callers that are not in atomic context. | |
3659 | */ | |
3660 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
3661 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
3662 | gfp_mask &= ~__GFP_ATOMIC; | |
3663 | ||
5ce9bfef VB |
3664 | retry_cpuset: |
3665 | compaction_retries = 0; | |
3666 | no_progress_loops = 0; | |
3667 | compact_priority = DEF_COMPACT_PRIORITY; | |
3668 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
3669 | |
3670 | /* | |
3671 | * The fast path uses conservative alloc_flags to succeed only until | |
3672 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
3673 | * alloc_flags precisely. So we do that now. | |
3674 | */ | |
3675 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
3676 | ||
e47483bc VB |
3677 | /* |
3678 | * We need to recalculate the starting point for the zonelist iterator | |
3679 | * because we might have used different nodemask in the fast path, or | |
3680 | * there was a cpuset modification and we are retrying - otherwise we | |
3681 | * could end up iterating over non-eligible zones endlessly. | |
3682 | */ | |
3683 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
3684 | ac->high_zoneidx, ac->nodemask); | |
3685 | if (!ac->preferred_zoneref->zone) | |
3686 | goto nopage; | |
3687 | ||
23771235 VB |
3688 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3689 | wake_all_kswapds(order, ac); | |
3690 | ||
3691 | /* | |
3692 | * The adjusted alloc_flags might result in immediate success, so try | |
3693 | * that first | |
3694 | */ | |
3695 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
3696 | if (page) | |
3697 | goto got_pg; | |
3698 | ||
a8161d1e VB |
3699 | /* |
3700 | * For costly allocations, try direct compaction first, as it's likely | |
3701 | * that we have enough base pages and don't need to reclaim. Don't try | |
3702 | * that for allocations that are allowed to ignore watermarks, as the | |
3703 | * ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
3704 | */ | |
3705 | if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER && | |
3706 | !gfp_pfmemalloc_allowed(gfp_mask)) { | |
3707 | page = __alloc_pages_direct_compact(gfp_mask, order, | |
3708 | alloc_flags, ac, | |
a5508cd8 | 3709 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
3710 | &compact_result); |
3711 | if (page) | |
3712 | goto got_pg; | |
3713 | ||
3eb2771b VB |
3714 | /* |
3715 | * Checks for costly allocations with __GFP_NORETRY, which | |
3716 | * includes THP page fault allocations | |
3717 | */ | |
3718 | if (gfp_mask & __GFP_NORETRY) { | |
a8161d1e VB |
3719 | /* |
3720 | * If compaction is deferred for high-order allocations, | |
3721 | * it is because sync compaction recently failed. If | |
3722 | * this is the case and the caller requested a THP | |
3723 | * allocation, we do not want to heavily disrupt the | |
3724 | * system, so we fail the allocation instead of entering | |
3725 | * direct reclaim. | |
3726 | */ | |
3727 | if (compact_result == COMPACT_DEFERRED) | |
3728 | goto nopage; | |
3729 | ||
a8161d1e | 3730 | /* |
3eb2771b VB |
3731 | * Looks like reclaim/compaction is worth trying, but |
3732 | * sync compaction could be very expensive, so keep | |
25160354 | 3733 | * using async compaction. |
a8161d1e | 3734 | */ |
a5508cd8 | 3735 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
3736 | } |
3737 | } | |
23771235 | 3738 | |
31a6c190 | 3739 | retry: |
23771235 | 3740 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
31a6c190 VB |
3741 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3742 | wake_all_kswapds(order, ac); | |
3743 | ||
23771235 VB |
3744 | if (gfp_pfmemalloc_allowed(gfp_mask)) |
3745 | alloc_flags = ALLOC_NO_WATERMARKS; | |
3746 | ||
e46e7b77 MG |
3747 | /* |
3748 | * Reset the zonelist iterators if memory policies can be ignored. | |
3749 | * These allocations are high priority and system rather than user | |
3750 | * orientated. | |
3751 | */ | |
23771235 | 3752 | if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) { |
e46e7b77 MG |
3753 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
3754 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
3755 | ac->high_zoneidx, ac->nodemask); | |
3756 | } | |
3757 | ||
23771235 | 3758 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 3759 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
3760 | if (page) |
3761 | goto got_pg; | |
1da177e4 | 3762 | |
d0164adc | 3763 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 3764 | if (!can_direct_reclaim) |
1da177e4 LT |
3765 | goto nopage; |
3766 | ||
9a67f648 MH |
3767 | /* Make sure we know about allocations which stall for too long */ |
3768 | if (time_after(jiffies, alloc_start + stall_timeout)) { | |
3769 | warn_alloc(gfp_mask, ac->nodemask, | |
3770 | "page allocation stalls for %ums, order:%u", | |
3771 | jiffies_to_msecs(jiffies-alloc_start), order); | |
3772 | stall_timeout += 10 * HZ; | |
33d53103 | 3773 | } |
341ce06f | 3774 | |
9a67f648 MH |
3775 | /* Avoid recursion of direct reclaim */ |
3776 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
3777 | goto nopage; |
3778 | ||
a8161d1e VB |
3779 | /* Try direct reclaim and then allocating */ |
3780 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
3781 | &did_some_progress); | |
3782 | if (page) | |
3783 | goto got_pg; | |
3784 | ||
3785 | /* Try direct compaction and then allocating */ | |
a9263751 | 3786 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 3787 | compact_priority, &compact_result); |
56de7263 MG |
3788 | if (page) |
3789 | goto got_pg; | |
75f30861 | 3790 | |
9083905a JW |
3791 | /* Do not loop if specifically requested */ |
3792 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 3793 | goto nopage; |
9083905a | 3794 | |
0a0337e0 MH |
3795 | /* |
3796 | * Do not retry costly high order allocations unless they are | |
3797 | * __GFP_REPEAT | |
3798 | */ | |
3799 | if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT)) | |
a8161d1e | 3800 | goto nopage; |
0a0337e0 | 3801 | |
0a0337e0 | 3802 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 3803 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
3804 | goto retry; |
3805 | ||
33c2d214 MH |
3806 | /* |
3807 | * It doesn't make any sense to retry for the compaction if the order-0 | |
3808 | * reclaim is not able to make any progress because the current | |
3809 | * implementation of the compaction depends on the sufficient amount | |
3810 | * of free memory (see __compaction_suitable) | |
3811 | */ | |
3812 | if (did_some_progress > 0 && | |
86a294a8 | 3813 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 3814 | compact_result, &compact_priority, |
d9436498 | 3815 | &compaction_retries)) |
33c2d214 MH |
3816 | goto retry; |
3817 | ||
e47483bc VB |
3818 | /* |
3819 | * It's possible we raced with cpuset update so the OOM would be | |
3820 | * premature (see below the nopage: label for full explanation). | |
3821 | */ | |
3822 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
3823 | goto retry_cpuset; | |
3824 | ||
9083905a JW |
3825 | /* Reclaim has failed us, start killing things */ |
3826 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
3827 | if (page) | |
3828 | goto got_pg; | |
3829 | ||
9a67f648 MH |
3830 | /* Avoid allocations with no watermarks from looping endlessly */ |
3831 | if (test_thread_flag(TIF_MEMDIE)) | |
3832 | goto nopage; | |
3833 | ||
9083905a | 3834 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
3835 | if (did_some_progress) { |
3836 | no_progress_loops = 0; | |
9083905a | 3837 | goto retry; |
0a0337e0 | 3838 | } |
9083905a | 3839 | |
1da177e4 | 3840 | nopage: |
5ce9bfef | 3841 | /* |
e47483bc VB |
3842 | * When updating a task's mems_allowed or mempolicy nodemask, it is |
3843 | * possible to race with parallel threads in such a way that our | |
3844 | * allocation can fail while the mask is being updated. If we are about | |
3845 | * to fail, check if the cpuset changed during allocation and if so, | |
3846 | * retry. | |
5ce9bfef VB |
3847 | */ |
3848 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
3849 | goto retry_cpuset; | |
3850 | ||
9a67f648 MH |
3851 | /* |
3852 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
3853 | * we always retry | |
3854 | */ | |
3855 | if (gfp_mask & __GFP_NOFAIL) { | |
3856 | /* | |
3857 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
3858 | * of any new users that actually require GFP_NOWAIT | |
3859 | */ | |
3860 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
3861 | goto fail; | |
3862 | ||
3863 | /* | |
3864 | * PF_MEMALLOC request from this context is rather bizarre | |
3865 | * because we cannot reclaim anything and only can loop waiting | |
3866 | * for somebody to do a work for us | |
3867 | */ | |
3868 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
3869 | ||
3870 | /* | |
3871 | * non failing costly orders are a hard requirement which we | |
3872 | * are not prepared for much so let's warn about these users | |
3873 | * so that we can identify them and convert them to something | |
3874 | * else. | |
3875 | */ | |
3876 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
3877 | ||
6c18ba7a MH |
3878 | /* |
3879 | * Help non-failing allocations by giving them access to memory | |
3880 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
3881 | * could deplete whole memory reserves which would just make | |
3882 | * the situation worse | |
3883 | */ | |
3884 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
3885 | if (page) | |
3886 | goto got_pg; | |
3887 | ||
9a67f648 MH |
3888 | cond_resched(); |
3889 | goto retry; | |
3890 | } | |
3891 | fail: | |
a8e99259 | 3892 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 3893 | "page allocation failure: order:%u", order); |
1da177e4 | 3894 | got_pg: |
072bb0aa | 3895 | return page; |
1da177e4 | 3896 | } |
11e33f6a | 3897 | |
9cd75558 MG |
3898 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
3899 | struct zonelist *zonelist, nodemask_t *nodemask, | |
3900 | struct alloc_context *ac, gfp_t *alloc_mask, | |
3901 | unsigned int *alloc_flags) | |
11e33f6a | 3902 | { |
9cd75558 MG |
3903 | ac->high_zoneidx = gfp_zone(gfp_mask); |
3904 | ac->zonelist = zonelist; | |
3905 | ac->nodemask = nodemask; | |
3906 | ac->migratetype = gfpflags_to_migratetype(gfp_mask); | |
11e33f6a | 3907 | |
682a3385 | 3908 | if (cpusets_enabled()) { |
9cd75558 | 3909 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
3910 | if (!ac->nodemask) |
3911 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
3912 | else |
3913 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
3914 | } |
3915 | ||
11e33f6a MG |
3916 | lockdep_trace_alloc(gfp_mask); |
3917 | ||
d0164adc | 3918 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
3919 | |
3920 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 3921 | return false; |
11e33f6a | 3922 | |
9cd75558 MG |
3923 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
3924 | *alloc_flags |= ALLOC_CMA; | |
3925 | ||
3926 | return true; | |
3927 | } | |
21bb9bd1 | 3928 | |
9cd75558 MG |
3929 | /* Determine whether to spread dirty pages and what the first usable zone */ |
3930 | static inline void finalise_ac(gfp_t gfp_mask, | |
3931 | unsigned int order, struct alloc_context *ac) | |
3932 | { | |
c9ab0c4f | 3933 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 3934 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 3935 | |
e46e7b77 MG |
3936 | /* |
3937 | * The preferred zone is used for statistics but crucially it is | |
3938 | * also used as the starting point for the zonelist iterator. It | |
3939 | * may get reset for allocations that ignore memory policies. | |
3940 | */ | |
9cd75558 MG |
3941 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
3942 | ac->high_zoneidx, ac->nodemask); | |
3943 | } | |
3944 | ||
3945 | /* | |
3946 | * This is the 'heart' of the zoned buddy allocator. | |
3947 | */ | |
3948 | struct page * | |
3949 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
3950 | struct zonelist *zonelist, nodemask_t *nodemask) | |
3951 | { | |
3952 | struct page *page; | |
3953 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
3954 | gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */ | |
3955 | struct alloc_context ac = { }; | |
3956 | ||
3957 | gfp_mask &= gfp_allowed_mask; | |
3958 | if (!prepare_alloc_pages(gfp_mask, order, zonelist, nodemask, &ac, &alloc_mask, &alloc_flags)) | |
3959 | return NULL; | |
3960 | ||
3961 | finalise_ac(gfp_mask, order, &ac); | |
5bb1b169 | 3962 | |
5117f45d | 3963 | /* First allocation attempt */ |
a9263751 | 3964 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
3965 | if (likely(page)) |
3966 | goto out; | |
11e33f6a | 3967 | |
4fcb0971 MG |
3968 | /* |
3969 | * Runtime PM, block IO and its error handling path can deadlock | |
3970 | * because I/O on the device might not complete. | |
3971 | */ | |
3972 | alloc_mask = memalloc_noio_flags(gfp_mask); | |
3973 | ac.spread_dirty_pages = false; | |
23f086f9 | 3974 | |
4741526b MG |
3975 | /* |
3976 | * Restore the original nodemask if it was potentially replaced with | |
3977 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
3978 | */ | |
e47483bc | 3979 | if (unlikely(ac.nodemask != nodemask)) |
4741526b | 3980 | ac.nodemask = nodemask; |
16096c25 | 3981 | |
4fcb0971 | 3982 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 3983 | |
4fcb0971 | 3984 | out: |
c4159a75 VD |
3985 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
3986 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { | |
3987 | __free_pages(page, order); | |
3988 | page = NULL; | |
4949148a VD |
3989 | } |
3990 | ||
4fcb0971 MG |
3991 | if (kmemcheck_enabled && page) |
3992 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
3993 | ||
3994 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); | |
3995 | ||
11e33f6a | 3996 | return page; |
1da177e4 | 3997 | } |
d239171e | 3998 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
3999 | |
4000 | /* | |
4001 | * Common helper functions. | |
4002 | */ | |
920c7a5d | 4003 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4004 | { |
945a1113 AM |
4005 | struct page *page; |
4006 | ||
4007 | /* | |
4008 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
4009 | * a highmem page | |
4010 | */ | |
4011 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
4012 | ||
1da177e4 LT |
4013 | page = alloc_pages(gfp_mask, order); |
4014 | if (!page) | |
4015 | return 0; | |
4016 | return (unsigned long) page_address(page); | |
4017 | } | |
1da177e4 LT |
4018 | EXPORT_SYMBOL(__get_free_pages); |
4019 | ||
920c7a5d | 4020 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4021 | { |
945a1113 | 4022 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4023 | } |
1da177e4 LT |
4024 | EXPORT_SYMBOL(get_zeroed_page); |
4025 | ||
920c7a5d | 4026 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 4027 | { |
b5810039 | 4028 | if (put_page_testzero(page)) { |
1da177e4 | 4029 | if (order == 0) |
b745bc85 | 4030 | free_hot_cold_page(page, false); |
1da177e4 LT |
4031 | else |
4032 | __free_pages_ok(page, order); | |
4033 | } | |
4034 | } | |
4035 | ||
4036 | EXPORT_SYMBOL(__free_pages); | |
4037 | ||
920c7a5d | 4038 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4039 | { |
4040 | if (addr != 0) { | |
725d704e | 4041 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4042 | __free_pages(virt_to_page((void *)addr), order); |
4043 | } | |
4044 | } | |
4045 | ||
4046 | EXPORT_SYMBOL(free_pages); | |
4047 | ||
b63ae8ca AD |
4048 | /* |
4049 | * Page Fragment: | |
4050 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4051 | * within a 0 or higher order page. Multiple fragments within that page | |
4052 | * are individually refcounted, in the page's reference counter. | |
4053 | * | |
4054 | * The page_frag functions below provide a simple allocation framework for | |
4055 | * page fragments. This is used by the network stack and network device | |
4056 | * drivers to provide a backing region of memory for use as either an | |
4057 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4058 | */ | |
2976db80 AD |
4059 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4060 | gfp_t gfp_mask) | |
b63ae8ca AD |
4061 | { |
4062 | struct page *page = NULL; | |
4063 | gfp_t gfp = gfp_mask; | |
4064 | ||
4065 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4066 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4067 | __GFP_NOMEMALLOC; | |
4068 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4069 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4070 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4071 | #endif | |
4072 | if (unlikely(!page)) | |
4073 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4074 | ||
4075 | nc->va = page ? page_address(page) : NULL; | |
4076 | ||
4077 | return page; | |
4078 | } | |
4079 | ||
2976db80 | 4080 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4081 | { |
4082 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4083 | ||
4084 | if (page_ref_sub_and_test(page, count)) { | |
2976db80 AD |
4085 | unsigned int order = compound_order(page); |
4086 | ||
44fdffd7 AD |
4087 | if (order == 0) |
4088 | free_hot_cold_page(page, false); | |
4089 | else | |
4090 | __free_pages_ok(page, order); | |
4091 | } | |
4092 | } | |
2976db80 | 4093 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4094 | |
8c2dd3e4 AD |
4095 | void *page_frag_alloc(struct page_frag_cache *nc, |
4096 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4097 | { |
4098 | unsigned int size = PAGE_SIZE; | |
4099 | struct page *page; | |
4100 | int offset; | |
4101 | ||
4102 | if (unlikely(!nc->va)) { | |
4103 | refill: | |
2976db80 | 4104 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4105 | if (!page) |
4106 | return NULL; | |
4107 | ||
4108 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4109 | /* if size can vary use size else just use PAGE_SIZE */ | |
4110 | size = nc->size; | |
4111 | #endif | |
4112 | /* Even if we own the page, we do not use atomic_set(). | |
4113 | * This would break get_page_unless_zero() users. | |
4114 | */ | |
fe896d18 | 4115 | page_ref_add(page, size - 1); |
b63ae8ca AD |
4116 | |
4117 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4118 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
4119 | nc->pagecnt_bias = size; |
4120 | nc->offset = size; | |
4121 | } | |
4122 | ||
4123 | offset = nc->offset - fragsz; | |
4124 | if (unlikely(offset < 0)) { | |
4125 | page = virt_to_page(nc->va); | |
4126 | ||
fe896d18 | 4127 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4128 | goto refill; |
4129 | ||
4130 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4131 | /* if size can vary use size else just use PAGE_SIZE */ | |
4132 | size = nc->size; | |
4133 | #endif | |
4134 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 4135 | set_page_count(page, size); |
b63ae8ca AD |
4136 | |
4137 | /* reset page count bias and offset to start of new frag */ | |
4138 | nc->pagecnt_bias = size; | |
4139 | offset = size - fragsz; | |
4140 | } | |
4141 | ||
4142 | nc->pagecnt_bias--; | |
4143 | nc->offset = offset; | |
4144 | ||
4145 | return nc->va + offset; | |
4146 | } | |
8c2dd3e4 | 4147 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
4148 | |
4149 | /* | |
4150 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4151 | */ | |
8c2dd3e4 | 4152 | void page_frag_free(void *addr) |
b63ae8ca AD |
4153 | { |
4154 | struct page *page = virt_to_head_page(addr); | |
4155 | ||
4156 | if (unlikely(put_page_testzero(page))) | |
4157 | __free_pages_ok(page, compound_order(page)); | |
4158 | } | |
8c2dd3e4 | 4159 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4160 | |
d00181b9 KS |
4161 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4162 | size_t size) | |
ee85c2e1 AK |
4163 | { |
4164 | if (addr) { | |
4165 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
4166 | unsigned long used = addr + PAGE_ALIGN(size); | |
4167 | ||
4168 | split_page(virt_to_page((void *)addr), order); | |
4169 | while (used < alloc_end) { | |
4170 | free_page(used); | |
4171 | used += PAGE_SIZE; | |
4172 | } | |
4173 | } | |
4174 | return (void *)addr; | |
4175 | } | |
4176 | ||
2be0ffe2 TT |
4177 | /** |
4178 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4179 | * @size: the number of bytes to allocate | |
4180 | * @gfp_mask: GFP flags for the allocation | |
4181 | * | |
4182 | * This function is similar to alloc_pages(), except that it allocates the | |
4183 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4184 | * allocate memory in power-of-two pages. | |
4185 | * | |
4186 | * This function is also limited by MAX_ORDER. | |
4187 | * | |
4188 | * Memory allocated by this function must be released by free_pages_exact(). | |
4189 | */ | |
4190 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4191 | { | |
4192 | unsigned int order = get_order(size); | |
4193 | unsigned long addr; | |
4194 | ||
4195 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 4196 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4197 | } |
4198 | EXPORT_SYMBOL(alloc_pages_exact); | |
4199 | ||
ee85c2e1 AK |
4200 | /** |
4201 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4202 | * pages on a node. | |
b5e6ab58 | 4203 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
4204 | * @size: the number of bytes to allocate |
4205 | * @gfp_mask: GFP flags for the allocation | |
4206 | * | |
4207 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4208 | * back. | |
ee85c2e1 | 4209 | */ |
e1931811 | 4210 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4211 | { |
d00181b9 | 4212 | unsigned int order = get_order(size); |
ee85c2e1 AK |
4213 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
4214 | if (!p) | |
4215 | return NULL; | |
4216 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4217 | } | |
ee85c2e1 | 4218 | |
2be0ffe2 TT |
4219 | /** |
4220 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4221 | * @virt: the value returned by alloc_pages_exact. | |
4222 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4223 | * | |
4224 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4225 | */ | |
4226 | void free_pages_exact(void *virt, size_t size) | |
4227 | { | |
4228 | unsigned long addr = (unsigned long)virt; | |
4229 | unsigned long end = addr + PAGE_ALIGN(size); | |
4230 | ||
4231 | while (addr < end) { | |
4232 | free_page(addr); | |
4233 | addr += PAGE_SIZE; | |
4234 | } | |
4235 | } | |
4236 | EXPORT_SYMBOL(free_pages_exact); | |
4237 | ||
e0fb5815 ZY |
4238 | /** |
4239 | * nr_free_zone_pages - count number of pages beyond high watermark | |
4240 | * @offset: The zone index of the highest zone | |
4241 | * | |
4242 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
4243 | * high watermark within all zones at or below a given zone index. For each | |
4244 | * zone, the number of pages is calculated as: | |
834405c3 | 4245 | * managed_pages - high_pages |
e0fb5815 | 4246 | */ |
ebec3862 | 4247 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 4248 | { |
dd1a239f | 4249 | struct zoneref *z; |
54a6eb5c MG |
4250 | struct zone *zone; |
4251 | ||
e310fd43 | 4252 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 4253 | unsigned long sum = 0; |
1da177e4 | 4254 | |
0e88460d | 4255 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 4256 | |
54a6eb5c | 4257 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 4258 | unsigned long size = zone->managed_pages; |
41858966 | 4259 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
4260 | if (size > high) |
4261 | sum += size - high; | |
1da177e4 LT |
4262 | } |
4263 | ||
4264 | return sum; | |
4265 | } | |
4266 | ||
e0fb5815 ZY |
4267 | /** |
4268 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
4269 | * | |
4270 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
4271 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 4272 | */ |
ebec3862 | 4273 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 4274 | { |
af4ca457 | 4275 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 4276 | } |
c2f1a551 | 4277 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 4278 | |
e0fb5815 ZY |
4279 | /** |
4280 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
4281 | * | |
4282 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
4283 | * high watermark within all zones. | |
1da177e4 | 4284 | */ |
ebec3862 | 4285 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 4286 | { |
2a1e274a | 4287 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 4288 | } |
08e0f6a9 CL |
4289 | |
4290 | static inline void show_node(struct zone *zone) | |
1da177e4 | 4291 | { |
e5adfffc | 4292 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 4293 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 4294 | } |
1da177e4 | 4295 | |
d02bd27b IR |
4296 | long si_mem_available(void) |
4297 | { | |
4298 | long available; | |
4299 | unsigned long pagecache; | |
4300 | unsigned long wmark_low = 0; | |
4301 | unsigned long pages[NR_LRU_LISTS]; | |
4302 | struct zone *zone; | |
4303 | int lru; | |
4304 | ||
4305 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 4306 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
4307 | |
4308 | for_each_zone(zone) | |
4309 | wmark_low += zone->watermark[WMARK_LOW]; | |
4310 | ||
4311 | /* | |
4312 | * Estimate the amount of memory available for userspace allocations, | |
4313 | * without causing swapping. | |
4314 | */ | |
4315 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; | |
4316 | ||
4317 | /* | |
4318 | * Not all the page cache can be freed, otherwise the system will | |
4319 | * start swapping. Assume at least half of the page cache, or the | |
4320 | * low watermark worth of cache, needs to stay. | |
4321 | */ | |
4322 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
4323 | pagecache -= min(pagecache / 2, wmark_low); | |
4324 | available += pagecache; | |
4325 | ||
4326 | /* | |
4327 | * Part of the reclaimable slab consists of items that are in use, | |
4328 | * and cannot be freed. Cap this estimate at the low watermark. | |
4329 | */ | |
4330 | available += global_page_state(NR_SLAB_RECLAIMABLE) - | |
4331 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); | |
4332 | ||
4333 | if (available < 0) | |
4334 | available = 0; | |
4335 | return available; | |
4336 | } | |
4337 | EXPORT_SYMBOL_GPL(si_mem_available); | |
4338 | ||
1da177e4 LT |
4339 | void si_meminfo(struct sysinfo *val) |
4340 | { | |
4341 | val->totalram = totalram_pages; | |
11fb9989 | 4342 | val->sharedram = global_node_page_state(NR_SHMEM); |
d23ad423 | 4343 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 4344 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
4345 | val->totalhigh = totalhigh_pages; |
4346 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
4347 | val->mem_unit = PAGE_SIZE; |
4348 | } | |
4349 | ||
4350 | EXPORT_SYMBOL(si_meminfo); | |
4351 | ||
4352 | #ifdef CONFIG_NUMA | |
4353 | void si_meminfo_node(struct sysinfo *val, int nid) | |
4354 | { | |
cdd91a77 JL |
4355 | int zone_type; /* needs to be signed */ |
4356 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
4357 | unsigned long managed_highpages = 0; |
4358 | unsigned long free_highpages = 0; | |
1da177e4 LT |
4359 | pg_data_t *pgdat = NODE_DATA(nid); |
4360 | ||
cdd91a77 JL |
4361 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
4362 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
4363 | val->totalram = managed_pages; | |
11fb9989 | 4364 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 4365 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 4366 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
4367 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4368 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4369 | ||
4370 | if (is_highmem(zone)) { | |
4371 | managed_highpages += zone->managed_pages; | |
4372 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
4373 | } | |
4374 | } | |
4375 | val->totalhigh = managed_highpages; | |
4376 | val->freehigh = free_highpages; | |
98d2b0eb | 4377 | #else |
fc2bd799 JK |
4378 | val->totalhigh = managed_highpages; |
4379 | val->freehigh = free_highpages; | |
98d2b0eb | 4380 | #endif |
1da177e4 LT |
4381 | val->mem_unit = PAGE_SIZE; |
4382 | } | |
4383 | #endif | |
4384 | ||
ddd588b5 | 4385 | /* |
7bf02ea2 DR |
4386 | * Determine whether the node should be displayed or not, depending on whether |
4387 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 4388 | */ |
9af744d7 | 4389 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 4390 | { |
ddd588b5 | 4391 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 4392 | return false; |
ddd588b5 | 4393 | |
9af744d7 MH |
4394 | /* |
4395 | * no node mask - aka implicit memory numa policy. Do not bother with | |
4396 | * the synchronization - read_mems_allowed_begin - because we do not | |
4397 | * have to be precise here. | |
4398 | */ | |
4399 | if (!nodemask) | |
4400 | nodemask = &cpuset_current_mems_allowed; | |
4401 | ||
4402 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
4403 | } |
4404 | ||
1da177e4 LT |
4405 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4406 | ||
377e4f16 RV |
4407 | static void show_migration_types(unsigned char type) |
4408 | { | |
4409 | static const char types[MIGRATE_TYPES] = { | |
4410 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 4411 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
4412 | [MIGRATE_RECLAIMABLE] = 'E', |
4413 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
4414 | #ifdef CONFIG_CMA |
4415 | [MIGRATE_CMA] = 'C', | |
4416 | #endif | |
194159fb | 4417 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 4418 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 4419 | #endif |
377e4f16 RV |
4420 | }; |
4421 | char tmp[MIGRATE_TYPES + 1]; | |
4422 | char *p = tmp; | |
4423 | int i; | |
4424 | ||
4425 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
4426 | if (type & (1 << i)) | |
4427 | *p++ = types[i]; | |
4428 | } | |
4429 | ||
4430 | *p = '\0'; | |
1f84a18f | 4431 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
4432 | } |
4433 | ||
1da177e4 LT |
4434 | /* |
4435 | * Show free area list (used inside shift_scroll-lock stuff) | |
4436 | * We also calculate the percentage fragmentation. We do this by counting the | |
4437 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
4438 | * |
4439 | * Bits in @filter: | |
4440 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
4441 | * cpuset. | |
1da177e4 | 4442 | */ |
9af744d7 | 4443 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 4444 | { |
d1bfcdb8 | 4445 | unsigned long free_pcp = 0; |
c7241913 | 4446 | int cpu; |
1da177e4 | 4447 | struct zone *zone; |
599d0c95 | 4448 | pg_data_t *pgdat; |
1da177e4 | 4449 | |
ee99c71c | 4450 | for_each_populated_zone(zone) { |
9af744d7 | 4451 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4452 | continue; |
d1bfcdb8 | 4453 | |
761b0677 KK |
4454 | for_each_online_cpu(cpu) |
4455 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
4456 | } |
4457 | ||
a731286d KM |
4458 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4459 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
4460 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4461 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 4462 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 4463 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
4464 | global_node_page_state(NR_ACTIVE_ANON), |
4465 | global_node_page_state(NR_INACTIVE_ANON), | |
4466 | global_node_page_state(NR_ISOLATED_ANON), | |
4467 | global_node_page_state(NR_ACTIVE_FILE), | |
4468 | global_node_page_state(NR_INACTIVE_FILE), | |
4469 | global_node_page_state(NR_ISOLATED_FILE), | |
4470 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
4471 | global_node_page_state(NR_FILE_DIRTY), |
4472 | global_node_page_state(NR_WRITEBACK), | |
4473 | global_node_page_state(NR_UNSTABLE_NFS), | |
3701b033 KM |
4474 | global_page_state(NR_SLAB_RECLAIMABLE), |
4475 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 4476 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 4477 | global_node_page_state(NR_SHMEM), |
a25700a5 | 4478 | global_page_state(NR_PAGETABLE), |
d1ce749a | 4479 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
4480 | global_page_state(NR_FREE_PAGES), |
4481 | free_pcp, | |
d1ce749a | 4482 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4483 | |
599d0c95 | 4484 | for_each_online_pgdat(pgdat) { |
9af744d7 | 4485 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
4486 | continue; |
4487 | ||
599d0c95 MG |
4488 | printk("Node %d" |
4489 | " active_anon:%lukB" | |
4490 | " inactive_anon:%lukB" | |
4491 | " active_file:%lukB" | |
4492 | " inactive_file:%lukB" | |
4493 | " unevictable:%lukB" | |
4494 | " isolated(anon):%lukB" | |
4495 | " isolated(file):%lukB" | |
50658e2e | 4496 | " mapped:%lukB" |
11fb9989 MG |
4497 | " dirty:%lukB" |
4498 | " writeback:%lukB" | |
4499 | " shmem:%lukB" | |
4500 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4501 | " shmem_thp: %lukB" | |
4502 | " shmem_pmdmapped: %lukB" | |
4503 | " anon_thp: %lukB" | |
4504 | #endif | |
4505 | " writeback_tmp:%lukB" | |
4506 | " unstable:%lukB" | |
33e077bd | 4507 | " pages_scanned:%lu" |
599d0c95 MG |
4508 | " all_unreclaimable? %s" |
4509 | "\n", | |
4510 | pgdat->node_id, | |
4511 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
4512 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
4513 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
4514 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
4515 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
4516 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
4517 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 4518 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
4519 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
4520 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
4521 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4522 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
4523 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
4524 | * HPAGE_PMD_NR), | |
4525 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
4526 | #endif | |
4527 | K(node_page_state(pgdat, NR_SHMEM)), | |
4528 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), | |
4529 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), | |
33e077bd | 4530 | node_page_state(pgdat, NR_PAGES_SCANNED), |
599d0c95 MG |
4531 | !pgdat_reclaimable(pgdat) ? "yes" : "no"); |
4532 | } | |
4533 | ||
ee99c71c | 4534 | for_each_populated_zone(zone) { |
1da177e4 LT |
4535 | int i; |
4536 | ||
9af744d7 | 4537 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4538 | continue; |
d1bfcdb8 KK |
4539 | |
4540 | free_pcp = 0; | |
4541 | for_each_online_cpu(cpu) | |
4542 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
4543 | ||
1da177e4 | 4544 | show_node(zone); |
1f84a18f JP |
4545 | printk(KERN_CONT |
4546 | "%s" | |
1da177e4 LT |
4547 | " free:%lukB" |
4548 | " min:%lukB" | |
4549 | " low:%lukB" | |
4550 | " high:%lukB" | |
71c799f4 MK |
4551 | " active_anon:%lukB" |
4552 | " inactive_anon:%lukB" | |
4553 | " active_file:%lukB" | |
4554 | " inactive_file:%lukB" | |
4555 | " unevictable:%lukB" | |
5a1c84b4 | 4556 | " writepending:%lukB" |
1da177e4 | 4557 | " present:%lukB" |
9feedc9d | 4558 | " managed:%lukB" |
4a0aa73f | 4559 | " mlocked:%lukB" |
4a0aa73f KM |
4560 | " slab_reclaimable:%lukB" |
4561 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 4562 | " kernel_stack:%lukB" |
4a0aa73f | 4563 | " pagetables:%lukB" |
4a0aa73f | 4564 | " bounce:%lukB" |
d1bfcdb8 KK |
4565 | " free_pcp:%lukB" |
4566 | " local_pcp:%ukB" | |
d1ce749a | 4567 | " free_cma:%lukB" |
1da177e4 LT |
4568 | "\n", |
4569 | zone->name, | |
88f5acf8 | 4570 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
4571 | K(min_wmark_pages(zone)), |
4572 | K(low_wmark_pages(zone)), | |
4573 | K(high_wmark_pages(zone)), | |
71c799f4 MK |
4574 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
4575 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
4576 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
4577 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
4578 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 4579 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 4580 | K(zone->present_pages), |
9feedc9d | 4581 | K(zone->managed_pages), |
4a0aa73f | 4582 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f KM |
4583 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
4584 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
d30dd8be | 4585 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4a0aa73f | 4586 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 4587 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
4588 | K(free_pcp), |
4589 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 4590 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
4591 | printk("lowmem_reserve[]:"); |
4592 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
4593 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
4594 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
4595 | } |
4596 | ||
ee99c71c | 4597 | for_each_populated_zone(zone) { |
d00181b9 KS |
4598 | unsigned int order; |
4599 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 4600 | unsigned char types[MAX_ORDER]; |
1da177e4 | 4601 | |
9af744d7 | 4602 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4603 | continue; |
1da177e4 | 4604 | show_node(zone); |
1f84a18f | 4605 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
4606 | |
4607 | spin_lock_irqsave(&zone->lock, flags); | |
4608 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
4609 | struct free_area *area = &zone->free_area[order]; |
4610 | int type; | |
4611 | ||
4612 | nr[order] = area->nr_free; | |
8f9de51a | 4613 | total += nr[order] << order; |
377e4f16 RV |
4614 | |
4615 | types[order] = 0; | |
4616 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
4617 | if (!list_empty(&area->free_list[type])) | |
4618 | types[order] |= 1 << type; | |
4619 | } | |
1da177e4 LT |
4620 | } |
4621 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 4622 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
4623 | printk(KERN_CONT "%lu*%lukB ", |
4624 | nr[order], K(1UL) << order); | |
377e4f16 RV |
4625 | if (nr[order]) |
4626 | show_migration_types(types[order]); | |
4627 | } | |
1f84a18f | 4628 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
4629 | } |
4630 | ||
949f7ec5 DR |
4631 | hugetlb_show_meminfo(); |
4632 | ||
11fb9989 | 4633 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 4634 | |
1da177e4 LT |
4635 | show_swap_cache_info(); |
4636 | } | |
4637 | ||
19770b32 MG |
4638 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4639 | { | |
4640 | zoneref->zone = zone; | |
4641 | zoneref->zone_idx = zone_idx(zone); | |
4642 | } | |
4643 | ||
1da177e4 LT |
4644 | /* |
4645 | * Builds allocation fallback zone lists. | |
1a93205b CL |
4646 | * |
4647 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 4648 | */ |
f0c0b2b8 | 4649 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 4650 | int nr_zones) |
1da177e4 | 4651 | { |
1a93205b | 4652 | struct zone *zone; |
bc732f1d | 4653 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
4654 | |
4655 | do { | |
2f6726e5 | 4656 | zone_type--; |
070f8032 | 4657 | zone = pgdat->node_zones + zone_type; |
6aa303de | 4658 | if (managed_zone(zone)) { |
dd1a239f MG |
4659 | zoneref_set_zone(zone, |
4660 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 4661 | check_highest_zone(zone_type); |
1da177e4 | 4662 | } |
2f6726e5 | 4663 | } while (zone_type); |
bc732f1d | 4664 | |
070f8032 | 4665 | return nr_zones; |
1da177e4 LT |
4666 | } |
4667 | ||
f0c0b2b8 KH |
4668 | |
4669 | /* | |
4670 | * zonelist_order: | |
4671 | * 0 = automatic detection of better ordering. | |
4672 | * 1 = order by ([node] distance, -zonetype) | |
4673 | * 2 = order by (-zonetype, [node] distance) | |
4674 | * | |
4675 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
4676 | * the same zonelist. So only NUMA can configure this param. | |
4677 | */ | |
4678 | #define ZONELIST_ORDER_DEFAULT 0 | |
4679 | #define ZONELIST_ORDER_NODE 1 | |
4680 | #define ZONELIST_ORDER_ZONE 2 | |
4681 | ||
4682 | /* zonelist order in the kernel. | |
4683 | * set_zonelist_order() will set this to NODE or ZONE. | |
4684 | */ | |
4685 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4686 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
4687 | ||
4688 | ||
1da177e4 | 4689 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
4690 | /* The value user specified ....changed by config */ |
4691 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4692 | /* string for sysctl */ | |
4693 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
4694 | char numa_zonelist_order[16] = "default"; | |
4695 | ||
4696 | /* | |
4697 | * interface for configure zonelist ordering. | |
4698 | * command line option "numa_zonelist_order" | |
4699 | * = "[dD]efault - default, automatic configuration. | |
4700 | * = "[nN]ode - order by node locality, then by zone within node | |
4701 | * = "[zZ]one - order by zone, then by locality within zone | |
4702 | */ | |
4703 | ||
4704 | static int __parse_numa_zonelist_order(char *s) | |
4705 | { | |
4706 | if (*s == 'd' || *s == 'D') { | |
4707 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4708 | } else if (*s == 'n' || *s == 'N') { | |
4709 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
4710 | } else if (*s == 'z' || *s == 'Z') { | |
4711 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
4712 | } else { | |
1170532b | 4713 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
f0c0b2b8 KH |
4714 | return -EINVAL; |
4715 | } | |
4716 | return 0; | |
4717 | } | |
4718 | ||
4719 | static __init int setup_numa_zonelist_order(char *s) | |
4720 | { | |
ecb256f8 VL |
4721 | int ret; |
4722 | ||
4723 | if (!s) | |
4724 | return 0; | |
4725 | ||
4726 | ret = __parse_numa_zonelist_order(s); | |
4727 | if (ret == 0) | |
4728 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
4729 | ||
4730 | return ret; | |
f0c0b2b8 KH |
4731 | } |
4732 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
4733 | ||
4734 | /* | |
4735 | * sysctl handler for numa_zonelist_order | |
4736 | */ | |
cccad5b9 | 4737 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 4738 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
4739 | loff_t *ppos) |
4740 | { | |
4741 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
4742 | int ret; | |
443c6f14 | 4743 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 4744 | |
443c6f14 | 4745 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
4746 | if (write) { |
4747 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
4748 | ret = -EINVAL; | |
4749 | goto out; | |
4750 | } | |
4751 | strcpy(saved_string, (char *)table->data); | |
4752 | } | |
8d65af78 | 4753 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 4754 | if (ret) |
443c6f14 | 4755 | goto out; |
f0c0b2b8 KH |
4756 | if (write) { |
4757 | int oldval = user_zonelist_order; | |
dacbde09 CG |
4758 | |
4759 | ret = __parse_numa_zonelist_order((char *)table->data); | |
4760 | if (ret) { | |
f0c0b2b8 KH |
4761 | /* |
4762 | * bogus value. restore saved string | |
4763 | */ | |
dacbde09 | 4764 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
4765 | NUMA_ZONELIST_ORDER_LEN); |
4766 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
4767 | } else if (oldval != user_zonelist_order) { |
4768 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 4769 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
4770 | mutex_unlock(&zonelists_mutex); |
4771 | } | |
f0c0b2b8 | 4772 | } |
443c6f14 AK |
4773 | out: |
4774 | mutex_unlock(&zl_order_mutex); | |
4775 | return ret; | |
f0c0b2b8 KH |
4776 | } |
4777 | ||
4778 | ||
62bc62a8 | 4779 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
4780 | static int node_load[MAX_NUMNODES]; |
4781 | ||
1da177e4 | 4782 | /** |
4dc3b16b | 4783 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
4784 | * @node: node whose fallback list we're appending |
4785 | * @used_node_mask: nodemask_t of already used nodes | |
4786 | * | |
4787 | * We use a number of factors to determine which is the next node that should | |
4788 | * appear on a given node's fallback list. The node should not have appeared | |
4789 | * already in @node's fallback list, and it should be the next closest node | |
4790 | * according to the distance array (which contains arbitrary distance values | |
4791 | * from each node to each node in the system), and should also prefer nodes | |
4792 | * with no CPUs, since presumably they'll have very little allocation pressure | |
4793 | * on them otherwise. | |
4794 | * It returns -1 if no node is found. | |
4795 | */ | |
f0c0b2b8 | 4796 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 4797 | { |
4cf808eb | 4798 | int n, val; |
1da177e4 | 4799 | int min_val = INT_MAX; |
00ef2d2f | 4800 | int best_node = NUMA_NO_NODE; |
a70f7302 | 4801 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 4802 | |
4cf808eb LT |
4803 | /* Use the local node if we haven't already */ |
4804 | if (!node_isset(node, *used_node_mask)) { | |
4805 | node_set(node, *used_node_mask); | |
4806 | return node; | |
4807 | } | |
1da177e4 | 4808 | |
4b0ef1fe | 4809 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
4810 | |
4811 | /* Don't want a node to appear more than once */ | |
4812 | if (node_isset(n, *used_node_mask)) | |
4813 | continue; | |
4814 | ||
1da177e4 LT |
4815 | /* Use the distance array to find the distance */ |
4816 | val = node_distance(node, n); | |
4817 | ||
4cf808eb LT |
4818 | /* Penalize nodes under us ("prefer the next node") */ |
4819 | val += (n < node); | |
4820 | ||
1da177e4 | 4821 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
4822 | tmp = cpumask_of_node(n); |
4823 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
4824 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4825 | ||
4826 | /* Slight preference for less loaded node */ | |
4827 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
4828 | val += node_load[n]; | |
4829 | ||
4830 | if (val < min_val) { | |
4831 | min_val = val; | |
4832 | best_node = n; | |
4833 | } | |
4834 | } | |
4835 | ||
4836 | if (best_node >= 0) | |
4837 | node_set(best_node, *used_node_mask); | |
4838 | ||
4839 | return best_node; | |
4840 | } | |
4841 | ||
f0c0b2b8 KH |
4842 | |
4843 | /* | |
4844 | * Build zonelists ordered by node and zones within node. | |
4845 | * This results in maximum locality--normal zone overflows into local | |
4846 | * DMA zone, if any--but risks exhausting DMA zone. | |
4847 | */ | |
4848 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 4849 | { |
f0c0b2b8 | 4850 | int j; |
1da177e4 | 4851 | struct zonelist *zonelist; |
f0c0b2b8 | 4852 | |
c9634cf0 | 4853 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
dd1a239f | 4854 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 4855 | ; |
bc732f1d | 4856 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
4857 | zonelist->_zonerefs[j].zone = NULL; |
4858 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
4859 | } |
4860 | ||
523b9458 CL |
4861 | /* |
4862 | * Build gfp_thisnode zonelists | |
4863 | */ | |
4864 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
4865 | { | |
523b9458 CL |
4866 | int j; |
4867 | struct zonelist *zonelist; | |
4868 | ||
c9634cf0 | 4869 | zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK]; |
bc732f1d | 4870 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
4871 | zonelist->_zonerefs[j].zone = NULL; |
4872 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
4873 | } |
4874 | ||
f0c0b2b8 KH |
4875 | /* |
4876 | * Build zonelists ordered by zone and nodes within zones. | |
4877 | * This results in conserving DMA zone[s] until all Normal memory is | |
4878 | * exhausted, but results in overflowing to remote node while memory | |
4879 | * may still exist in local DMA zone. | |
4880 | */ | |
4881 | static int node_order[MAX_NUMNODES]; | |
4882 | ||
4883 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
4884 | { | |
f0c0b2b8 KH |
4885 | int pos, j, node; |
4886 | int zone_type; /* needs to be signed */ | |
4887 | struct zone *z; | |
4888 | struct zonelist *zonelist; | |
4889 | ||
c9634cf0 | 4890 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
54a6eb5c MG |
4891 | pos = 0; |
4892 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
4893 | for (j = 0; j < nr_nodes; j++) { | |
4894 | node = node_order[j]; | |
4895 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
6aa303de | 4896 | if (managed_zone(z)) { |
dd1a239f MG |
4897 | zoneref_set_zone(z, |
4898 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 4899 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
4900 | } |
4901 | } | |
f0c0b2b8 | 4902 | } |
dd1a239f MG |
4903 | zonelist->_zonerefs[pos].zone = NULL; |
4904 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
4905 | } |
4906 | ||
3193913c MG |
4907 | #if defined(CONFIG_64BIT) |
4908 | /* | |
4909 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
4910 | * penalty to every machine in case the specialised case applies. Default | |
4911 | * to Node-ordering on 64-bit NUMA machines | |
4912 | */ | |
4913 | static int default_zonelist_order(void) | |
4914 | { | |
4915 | return ZONELIST_ORDER_NODE; | |
4916 | } | |
4917 | #else | |
4918 | /* | |
4919 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
4920 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
4921 | * then reclaim will occur more frequency increasing stalls and potentially | |
4922 | * be easier to OOM if a large percentage of the zone is under writeback or | |
4923 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
4924 | * Hence, default to zone ordering on 32-bit. | |
4925 | */ | |
f0c0b2b8 KH |
4926 | static int default_zonelist_order(void) |
4927 | { | |
f0c0b2b8 KH |
4928 | return ZONELIST_ORDER_ZONE; |
4929 | } | |
3193913c | 4930 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
4931 | |
4932 | static void set_zonelist_order(void) | |
4933 | { | |
4934 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
4935 | current_zonelist_order = default_zonelist_order(); | |
4936 | else | |
4937 | current_zonelist_order = user_zonelist_order; | |
4938 | } | |
4939 | ||
4940 | static void build_zonelists(pg_data_t *pgdat) | |
4941 | { | |
c00eb15a | 4942 | int i, node, load; |
1da177e4 | 4943 | nodemask_t used_mask; |
f0c0b2b8 KH |
4944 | int local_node, prev_node; |
4945 | struct zonelist *zonelist; | |
d00181b9 | 4946 | unsigned int order = current_zonelist_order; |
1da177e4 LT |
4947 | |
4948 | /* initialize zonelists */ | |
523b9458 | 4949 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 4950 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
4951 | zonelist->_zonerefs[0].zone = NULL; |
4952 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
4953 | } |
4954 | ||
4955 | /* NUMA-aware ordering of nodes */ | |
4956 | local_node = pgdat->node_id; | |
62bc62a8 | 4957 | load = nr_online_nodes; |
1da177e4 LT |
4958 | prev_node = local_node; |
4959 | nodes_clear(used_mask); | |
f0c0b2b8 | 4960 | |
f0c0b2b8 | 4961 | memset(node_order, 0, sizeof(node_order)); |
c00eb15a | 4962 | i = 0; |
f0c0b2b8 | 4963 | |
1da177e4 LT |
4964 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
4965 | /* | |
4966 | * We don't want to pressure a particular node. | |
4967 | * So adding penalty to the first node in same | |
4968 | * distance group to make it round-robin. | |
4969 | */ | |
957f822a DR |
4970 | if (node_distance(local_node, node) != |
4971 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
4972 | node_load[node] = load; |
4973 | ||
1da177e4 LT |
4974 | prev_node = node; |
4975 | load--; | |
f0c0b2b8 KH |
4976 | if (order == ZONELIST_ORDER_NODE) |
4977 | build_zonelists_in_node_order(pgdat, node); | |
4978 | else | |
c00eb15a | 4979 | node_order[i++] = node; /* remember order */ |
f0c0b2b8 | 4980 | } |
1da177e4 | 4981 | |
f0c0b2b8 KH |
4982 | if (order == ZONELIST_ORDER_ZONE) { |
4983 | /* calculate node order -- i.e., DMA last! */ | |
c00eb15a | 4984 | build_zonelists_in_zone_order(pgdat, i); |
1da177e4 | 4985 | } |
523b9458 CL |
4986 | |
4987 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
4988 | } |
4989 | ||
7aac7898 LS |
4990 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4991 | /* | |
4992 | * Return node id of node used for "local" allocations. | |
4993 | * I.e., first node id of first zone in arg node's generic zonelist. | |
4994 | * Used for initializing percpu 'numa_mem', which is used primarily | |
4995 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
4996 | */ | |
4997 | int local_memory_node(int node) | |
4998 | { | |
c33d6c06 | 4999 | struct zoneref *z; |
7aac7898 | 5000 | |
c33d6c06 | 5001 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5002 | gfp_zone(GFP_KERNEL), |
c33d6c06 MG |
5003 | NULL); |
5004 | return z->zone->node; | |
7aac7898 LS |
5005 | } |
5006 | #endif | |
f0c0b2b8 | 5007 | |
6423aa81 JK |
5008 | static void setup_min_unmapped_ratio(void); |
5009 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5010 | #else /* CONFIG_NUMA */ |
5011 | ||
f0c0b2b8 KH |
5012 | static void set_zonelist_order(void) |
5013 | { | |
5014 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
5015 | } | |
5016 | ||
5017 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 5018 | { |
19655d34 | 5019 | int node, local_node; |
54a6eb5c MG |
5020 | enum zone_type j; |
5021 | struct zonelist *zonelist; | |
1da177e4 LT |
5022 | |
5023 | local_node = pgdat->node_id; | |
1da177e4 | 5024 | |
c9634cf0 | 5025 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
bc732f1d | 5026 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 5027 | |
54a6eb5c MG |
5028 | /* |
5029 | * Now we build the zonelist so that it contains the zones | |
5030 | * of all the other nodes. | |
5031 | * We don't want to pressure a particular node, so when | |
5032 | * building the zones for node N, we make sure that the | |
5033 | * zones coming right after the local ones are those from | |
5034 | * node N+1 (modulo N) | |
5035 | */ | |
5036 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5037 | if (!node_online(node)) | |
5038 | continue; | |
bc732f1d | 5039 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 5040 | } |
54a6eb5c MG |
5041 | for (node = 0; node < local_node; node++) { |
5042 | if (!node_online(node)) | |
5043 | continue; | |
bc732f1d | 5044 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
5045 | } |
5046 | ||
dd1a239f MG |
5047 | zonelist->_zonerefs[j].zone = NULL; |
5048 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
5049 | } |
5050 | ||
5051 | #endif /* CONFIG_NUMA */ | |
5052 | ||
99dcc3e5 CL |
5053 | /* |
5054 | * Boot pageset table. One per cpu which is going to be used for all | |
5055 | * zones and all nodes. The parameters will be set in such a way | |
5056 | * that an item put on a list will immediately be handed over to | |
5057 | * the buddy list. This is safe since pageset manipulation is done | |
5058 | * with interrupts disabled. | |
5059 | * | |
5060 | * The boot_pagesets must be kept even after bootup is complete for | |
5061 | * unused processors and/or zones. They do play a role for bootstrapping | |
5062 | * hotplugged processors. | |
5063 | * | |
5064 | * zoneinfo_show() and maybe other functions do | |
5065 | * not check if the processor is online before following the pageset pointer. | |
5066 | * Other parts of the kernel may not check if the zone is available. | |
5067 | */ | |
5068 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5069 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 5070 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 5071 | |
4eaf3f64 HL |
5072 | /* |
5073 | * Global mutex to protect against size modification of zonelists | |
5074 | * as well as to serialize pageset setup for the new populated zone. | |
5075 | */ | |
5076 | DEFINE_MUTEX(zonelists_mutex); | |
5077 | ||
9b1a4d38 | 5078 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 5079 | static int __build_all_zonelists(void *data) |
1da177e4 | 5080 | { |
6811378e | 5081 | int nid; |
99dcc3e5 | 5082 | int cpu; |
9adb62a5 | 5083 | pg_data_t *self = data; |
9276b1bc | 5084 | |
7f9cfb31 BL |
5085 | #ifdef CONFIG_NUMA |
5086 | memset(node_load, 0, sizeof(node_load)); | |
5087 | #endif | |
9adb62a5 JL |
5088 | |
5089 | if (self && !node_online(self->node_id)) { | |
5090 | build_zonelists(self); | |
9adb62a5 JL |
5091 | } |
5092 | ||
9276b1bc | 5093 | for_each_online_node(nid) { |
7ea1530a CL |
5094 | pg_data_t *pgdat = NODE_DATA(nid); |
5095 | ||
5096 | build_zonelists(pgdat); | |
9276b1bc | 5097 | } |
99dcc3e5 CL |
5098 | |
5099 | /* | |
5100 | * Initialize the boot_pagesets that are going to be used | |
5101 | * for bootstrapping processors. The real pagesets for | |
5102 | * each zone will be allocated later when the per cpu | |
5103 | * allocator is available. | |
5104 | * | |
5105 | * boot_pagesets are used also for bootstrapping offline | |
5106 | * cpus if the system is already booted because the pagesets | |
5107 | * are needed to initialize allocators on a specific cpu too. | |
5108 | * F.e. the percpu allocator needs the page allocator which | |
5109 | * needs the percpu allocator in order to allocate its pagesets | |
5110 | * (a chicken-egg dilemma). | |
5111 | */ | |
7aac7898 | 5112 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
5113 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
5114 | ||
7aac7898 LS |
5115 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5116 | /* | |
5117 | * We now know the "local memory node" for each node-- | |
5118 | * i.e., the node of the first zone in the generic zonelist. | |
5119 | * Set up numa_mem percpu variable for on-line cpus. During | |
5120 | * boot, only the boot cpu should be on-line; we'll init the | |
5121 | * secondary cpus' numa_mem as they come on-line. During | |
5122 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5123 | */ | |
5124 | if (cpu_online(cpu)) | |
5125 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
5126 | #endif | |
5127 | } | |
5128 | ||
6811378e YG |
5129 | return 0; |
5130 | } | |
5131 | ||
061f67bc RV |
5132 | static noinline void __init |
5133 | build_all_zonelists_init(void) | |
5134 | { | |
5135 | __build_all_zonelists(NULL); | |
5136 | mminit_verify_zonelist(); | |
5137 | cpuset_init_current_mems_allowed(); | |
5138 | } | |
5139 | ||
4eaf3f64 HL |
5140 | /* |
5141 | * Called with zonelists_mutex held always | |
5142 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
5143 | * |
5144 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
5145 | * [we're only called with non-NULL zone through __meminit paths] and | |
5146 | * (2) call of __init annotated helper build_all_zonelists_init | |
5147 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 5148 | */ |
9adb62a5 | 5149 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 5150 | { |
f0c0b2b8 KH |
5151 | set_zonelist_order(); |
5152 | ||
6811378e | 5153 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 5154 | build_all_zonelists_init(); |
6811378e | 5155 | } else { |
e9959f0f | 5156 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
5157 | if (zone) |
5158 | setup_zone_pageset(zone); | |
e9959f0f | 5159 | #endif |
dd1895e2 CS |
5160 | /* we have to stop all cpus to guarantee there is no user |
5161 | of zonelist */ | |
9adb62a5 | 5162 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
5163 | /* cpuset refresh routine should be here */ |
5164 | } | |
bd1e22b8 | 5165 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5166 | /* |
5167 | * Disable grouping by mobility if the number of pages in the | |
5168 | * system is too low to allow the mechanism to work. It would be | |
5169 | * more accurate, but expensive to check per-zone. This check is | |
5170 | * made on memory-hotadd so a system can start with mobility | |
5171 | * disabled and enable it later | |
5172 | */ | |
d9c23400 | 5173 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5174 | page_group_by_mobility_disabled = 1; |
5175 | else | |
5176 | page_group_by_mobility_disabled = 0; | |
5177 | ||
756a025f JP |
5178 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
5179 | nr_online_nodes, | |
5180 | zonelist_order_name[current_zonelist_order], | |
5181 | page_group_by_mobility_disabled ? "off" : "on", | |
5182 | vm_total_pages); | |
f0c0b2b8 | 5183 | #ifdef CONFIG_NUMA |
f88dfff5 | 5184 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5185 | #endif |
1da177e4 LT |
5186 | } |
5187 | ||
1da177e4 LT |
5188 | /* |
5189 | * Initially all pages are reserved - free ones are freed | |
5190 | * up by free_all_bootmem() once the early boot process is | |
5191 | * done. Non-atomic initialization, single-pass. | |
5192 | */ | |
c09b4240 | 5193 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 5194 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 5195 | { |
4b94ffdc | 5196 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
29751f69 | 5197 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 5198 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 5199 | unsigned long pfn; |
3a80a7fa | 5200 | unsigned long nr_initialised = 0; |
342332e6 TI |
5201 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5202 | struct memblock_region *r = NULL, *tmp; | |
5203 | #endif | |
1da177e4 | 5204 | |
22b31eec HD |
5205 | if (highest_memmap_pfn < end_pfn - 1) |
5206 | highest_memmap_pfn = end_pfn - 1; | |
5207 | ||
4b94ffdc DW |
5208 | /* |
5209 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
5210 | * memory | |
5211 | */ | |
5212 | if (altmap && start_pfn == altmap->base_pfn) | |
5213 | start_pfn += altmap->reserve; | |
5214 | ||
cbe8dd4a | 5215 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 5216 | /* |
b72d0ffb AM |
5217 | * There can be holes in boot-time mem_map[]s handed to this |
5218 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5219 | */ |
b72d0ffb AM |
5220 | if (context != MEMMAP_EARLY) |
5221 | goto not_early; | |
5222 | ||
b92df1de PB |
5223 | if (!early_pfn_valid(pfn)) { |
5224 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
5225 | /* | |
5226 | * Skip to the pfn preceding the next valid one (or | |
5227 | * end_pfn), such that we hit a valid pfn (or end_pfn) | |
5228 | * on our next iteration of the loop. | |
5229 | */ | |
5230 | pfn = memblock_next_valid_pfn(pfn, end_pfn) - 1; | |
5231 | #endif | |
b72d0ffb | 5232 | continue; |
b92df1de | 5233 | } |
b72d0ffb AM |
5234 | if (!early_pfn_in_nid(pfn, nid)) |
5235 | continue; | |
5236 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
5237 | break; | |
342332e6 TI |
5238 | |
5239 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
5240 | /* |
5241 | * Check given memblock attribute by firmware which can affect | |
5242 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
5243 | * mirrored, it's an overlapped memmap init. skip it. | |
5244 | */ | |
5245 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5246 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
5247 | for_each_memblock(memory, tmp) | |
5248 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
5249 | break; | |
5250 | r = tmp; | |
5251 | } | |
5252 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
5253 | memblock_is_mirror(r)) { | |
5254 | /* already initialized as NORMAL */ | |
5255 | pfn = memblock_region_memory_end_pfn(r); | |
5256 | continue; | |
342332e6 | 5257 | } |
a2f3aa02 | 5258 | } |
b72d0ffb | 5259 | #endif |
ac5d2539 | 5260 | |
b72d0ffb | 5261 | not_early: |
ac5d2539 MG |
5262 | /* |
5263 | * Mark the block movable so that blocks are reserved for | |
5264 | * movable at startup. This will force kernel allocations | |
5265 | * to reserve their blocks rather than leaking throughout | |
5266 | * the address space during boot when many long-lived | |
974a786e | 5267 | * kernel allocations are made. |
ac5d2539 MG |
5268 | * |
5269 | * bitmap is created for zone's valid pfn range. but memmap | |
5270 | * can be created for invalid pages (for alignment) | |
5271 | * check here not to call set_pageblock_migratetype() against | |
5272 | * pfn out of zone. | |
5273 | */ | |
5274 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
5275 | struct page *page = pfn_to_page(pfn); | |
5276 | ||
5277 | __init_single_page(page, pfn, zone, nid); | |
5278 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5279 | } else { | |
5280 | __init_single_pfn(pfn, zone, nid); | |
5281 | } | |
1da177e4 LT |
5282 | } |
5283 | } | |
5284 | ||
1e548deb | 5285 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 5286 | { |
7aeb09f9 | 5287 | unsigned int order, t; |
b2a0ac88 MG |
5288 | for_each_migratetype_order(order, t) { |
5289 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
5290 | zone->free_area[order].nr_free = 0; |
5291 | } | |
5292 | } | |
5293 | ||
5294 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
5295 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 5296 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
5297 | #endif |
5298 | ||
7cd2b0a3 | 5299 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 5300 | { |
3a6be87f | 5301 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
5302 | int batch; |
5303 | ||
5304 | /* | |
5305 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 5306 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
5307 | * |
5308 | * OK, so we don't know how big the cache is. So guess. | |
5309 | */ | |
b40da049 | 5310 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
5311 | if (batch * PAGE_SIZE > 512 * 1024) |
5312 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
5313 | batch /= 4; /* We effectively *= 4 below */ |
5314 | if (batch < 1) | |
5315 | batch = 1; | |
5316 | ||
5317 | /* | |
0ceaacc9 NP |
5318 | * Clamp the batch to a 2^n - 1 value. Having a power |
5319 | * of 2 value was found to be more likely to have | |
5320 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 5321 | * |
0ceaacc9 NP |
5322 | * For example if 2 tasks are alternately allocating |
5323 | * batches of pages, one task can end up with a lot | |
5324 | * of pages of one half of the possible page colors | |
5325 | * and the other with pages of the other colors. | |
e7c8d5c9 | 5326 | */ |
9155203a | 5327 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 5328 | |
e7c8d5c9 | 5329 | return batch; |
3a6be87f DH |
5330 | |
5331 | #else | |
5332 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5333 | * conditions. | |
5334 | * | |
5335 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5336 | * of contiguous memory as there's no hardware page translation to | |
5337 | * assemble apparent contiguous memory from discontiguous pages. | |
5338 | * | |
5339 | * Queueing large contiguous runs of pages for batching, however, | |
5340 | * causes the pages to actually be freed in smaller chunks. As there | |
5341 | * can be a significant delay between the individual batches being | |
5342 | * recycled, this leads to the once large chunks of space being | |
5343 | * fragmented and becoming unavailable for high-order allocations. | |
5344 | */ | |
5345 | return 0; | |
5346 | #endif | |
e7c8d5c9 CL |
5347 | } |
5348 | ||
8d7a8fa9 CS |
5349 | /* |
5350 | * pcp->high and pcp->batch values are related and dependent on one another: | |
5351 | * ->batch must never be higher then ->high. | |
5352 | * The following function updates them in a safe manner without read side | |
5353 | * locking. | |
5354 | * | |
5355 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
5356 | * those fields changing asynchronously (acording the the above rule). | |
5357 | * | |
5358 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5359 | * outside of boot time (or some other assurance that no concurrent updaters | |
5360 | * exist). | |
5361 | */ | |
5362 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
5363 | unsigned long batch) | |
5364 | { | |
5365 | /* start with a fail safe value for batch */ | |
5366 | pcp->batch = 1; | |
5367 | smp_wmb(); | |
5368 | ||
5369 | /* Update high, then batch, in order */ | |
5370 | pcp->high = high; | |
5371 | smp_wmb(); | |
5372 | ||
5373 | pcp->batch = batch; | |
5374 | } | |
5375 | ||
3664033c | 5376 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
5377 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5378 | { | |
8d7a8fa9 | 5379 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
5380 | } |
5381 | ||
88c90dbc | 5382 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
5383 | { |
5384 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 5385 | int migratetype; |
2caaad41 | 5386 | |
1c6fe946 MD |
5387 | memset(p, 0, sizeof(*p)); |
5388 | ||
3dfa5721 | 5389 | pcp = &p->pcp; |
2caaad41 | 5390 | pcp->count = 0; |
5f8dcc21 MG |
5391 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5392 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
5393 | } |
5394 | ||
88c90dbc CS |
5395 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5396 | { | |
5397 | pageset_init(p); | |
5398 | pageset_set_batch(p, batch); | |
5399 | } | |
5400 | ||
8ad4b1fb | 5401 | /* |
3664033c | 5402 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
5403 | * to the value high for the pageset p. |
5404 | */ | |
3664033c | 5405 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
5406 | unsigned long high) |
5407 | { | |
8d7a8fa9 CS |
5408 | unsigned long batch = max(1UL, high / 4); |
5409 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
5410 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 5411 | |
8d7a8fa9 | 5412 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
5413 | } |
5414 | ||
7cd2b0a3 DR |
5415 | static void pageset_set_high_and_batch(struct zone *zone, |
5416 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 5417 | { |
56cef2b8 | 5418 | if (percpu_pagelist_fraction) |
3664033c | 5419 | pageset_set_high(pcp, |
56cef2b8 CS |
5420 | (zone->managed_pages / |
5421 | percpu_pagelist_fraction)); | |
5422 | else | |
5423 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
5424 | } | |
5425 | ||
169f6c19 CS |
5426 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5427 | { | |
5428 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
5429 | ||
5430 | pageset_init(pcp); | |
5431 | pageset_set_high_and_batch(zone, pcp); | |
5432 | } | |
5433 | ||
4ed7e022 | 5434 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
5435 | { |
5436 | int cpu; | |
319774e2 | 5437 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5438 | for_each_possible_cpu(cpu) |
5439 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5440 | } |
5441 | ||
2caaad41 | 5442 | /* |
99dcc3e5 CL |
5443 | * Allocate per cpu pagesets and initialize them. |
5444 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5445 | */ |
99dcc3e5 | 5446 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5447 | { |
b4911ea2 | 5448 | struct pglist_data *pgdat; |
99dcc3e5 | 5449 | struct zone *zone; |
e7c8d5c9 | 5450 | |
319774e2 WF |
5451 | for_each_populated_zone(zone) |
5452 | setup_zone_pageset(zone); | |
b4911ea2 MG |
5453 | |
5454 | for_each_online_pgdat(pgdat) | |
5455 | pgdat->per_cpu_nodestats = | |
5456 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
5457 | } |
5458 | ||
c09b4240 | 5459 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5460 | { |
99dcc3e5 CL |
5461 | /* |
5462 | * per cpu subsystem is not up at this point. The following code | |
5463 | * relies on the ability of the linker to provide the | |
5464 | * offset of a (static) per cpu variable into the per cpu area. | |
5465 | */ | |
5466 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5467 | |
b38a8725 | 5468 | if (populated_zone(zone)) |
99dcc3e5 CL |
5469 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5470 | zone->name, zone->present_pages, | |
5471 | zone_batchsize(zone)); | |
ed8ece2e DH |
5472 | } |
5473 | ||
4ed7e022 | 5474 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5475 | unsigned long zone_start_pfn, |
b171e409 | 5476 | unsigned long size) |
ed8ece2e DH |
5477 | { |
5478 | struct pglist_data *pgdat = zone->zone_pgdat; | |
9dcb8b68 | 5479 | |
ed8ece2e DH |
5480 | pgdat->nr_zones = zone_idx(zone) + 1; |
5481 | ||
ed8ece2e DH |
5482 | zone->zone_start_pfn = zone_start_pfn; |
5483 | ||
708614e6 MG |
5484 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5485 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5486 | pgdat->node_id, | |
5487 | (unsigned long)zone_idx(zone), | |
5488 | zone_start_pfn, (zone_start_pfn + size)); | |
5489 | ||
1e548deb | 5490 | zone_init_free_lists(zone); |
9dcb8b68 | 5491 | zone->initialized = 1; |
718127cc YG |
5492 | |
5493 | return 0; | |
ed8ece2e DH |
5494 | } |
5495 | ||
0ee332c1 | 5496 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5497 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5498 | |
c713216d MG |
5499 | /* |
5500 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5501 | */ |
8a942fde MG |
5502 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5503 | struct mminit_pfnnid_cache *state) | |
c713216d | 5504 | { |
c13291a5 | 5505 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5506 | int nid; |
7c243c71 | 5507 | |
8a942fde MG |
5508 | if (state->last_start <= pfn && pfn < state->last_end) |
5509 | return state->last_nid; | |
c713216d | 5510 | |
e76b63f8 YL |
5511 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5512 | if (nid != -1) { | |
8a942fde MG |
5513 | state->last_start = start_pfn; |
5514 | state->last_end = end_pfn; | |
5515 | state->last_nid = nid; | |
e76b63f8 YL |
5516 | } |
5517 | ||
5518 | return nid; | |
c713216d MG |
5519 | } |
5520 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5521 | ||
c713216d | 5522 | /** |
6782832e | 5523 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5524 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5525 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5526 | * |
7d018176 ZZ |
5527 | * If an architecture guarantees that all ranges registered contain no holes |
5528 | * and may be freed, this this function may be used instead of calling | |
5529 | * memblock_free_early_nid() manually. | |
c713216d | 5530 | */ |
c13291a5 | 5531 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5532 | { |
c13291a5 TH |
5533 | unsigned long start_pfn, end_pfn; |
5534 | int i, this_nid; | |
edbe7d23 | 5535 | |
c13291a5 TH |
5536 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5537 | start_pfn = min(start_pfn, max_low_pfn); | |
5538 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5539 | |
c13291a5 | 5540 | if (start_pfn < end_pfn) |
6782832e SS |
5541 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5542 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5543 | this_nid); | |
edbe7d23 | 5544 | } |
edbe7d23 | 5545 | } |
edbe7d23 | 5546 | |
c713216d MG |
5547 | /** |
5548 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5549 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5550 | * |
7d018176 ZZ |
5551 | * If an architecture guarantees that all ranges registered contain no holes and may |
5552 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5553 | */ |
5554 | void __init sparse_memory_present_with_active_regions(int nid) | |
5555 | { | |
c13291a5 TH |
5556 | unsigned long start_pfn, end_pfn; |
5557 | int i, this_nid; | |
c713216d | 5558 | |
c13291a5 TH |
5559 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5560 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5561 | } |
5562 | ||
5563 | /** | |
5564 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5565 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5566 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5567 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5568 | * |
5569 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5570 | * provided by memblock_set_node(). If called for a node |
c713216d | 5571 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5572 | * PFNs will be 0. |
c713216d | 5573 | */ |
a3142c8e | 5574 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5575 | unsigned long *start_pfn, unsigned long *end_pfn) |
5576 | { | |
c13291a5 | 5577 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5578 | int i; |
c13291a5 | 5579 | |
c713216d MG |
5580 | *start_pfn = -1UL; |
5581 | *end_pfn = 0; | |
5582 | ||
c13291a5 TH |
5583 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5584 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5585 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5586 | } |
5587 | ||
633c0666 | 5588 | if (*start_pfn == -1UL) |
c713216d | 5589 | *start_pfn = 0; |
c713216d MG |
5590 | } |
5591 | ||
2a1e274a MG |
5592 | /* |
5593 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5594 | * assumption is made that zones within a node are ordered in monotonic | |
5595 | * increasing memory addresses so that the "highest" populated zone is used | |
5596 | */ | |
b69a7288 | 5597 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5598 | { |
5599 | int zone_index; | |
5600 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5601 | if (zone_index == ZONE_MOVABLE) | |
5602 | continue; | |
5603 | ||
5604 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5605 | arch_zone_lowest_possible_pfn[zone_index]) | |
5606 | break; | |
5607 | } | |
5608 | ||
5609 | VM_BUG_ON(zone_index == -1); | |
5610 | movable_zone = zone_index; | |
5611 | } | |
5612 | ||
5613 | /* | |
5614 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5615 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5616 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5617 | * in each node depending on the size of each node and how evenly kernelcore | |
5618 | * is distributed. This helper function adjusts the zone ranges | |
5619 | * provided by the architecture for a given node by using the end of the | |
5620 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5621 | * zones within a node are in order of monotonic increases memory addresses | |
5622 | */ | |
b69a7288 | 5623 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5624 | unsigned long zone_type, |
5625 | unsigned long node_start_pfn, | |
5626 | unsigned long node_end_pfn, | |
5627 | unsigned long *zone_start_pfn, | |
5628 | unsigned long *zone_end_pfn) | |
5629 | { | |
5630 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5631 | if (zone_movable_pfn[nid]) { | |
5632 | /* Size ZONE_MOVABLE */ | |
5633 | if (zone_type == ZONE_MOVABLE) { | |
5634 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5635 | *zone_end_pfn = min(node_end_pfn, | |
5636 | arch_zone_highest_possible_pfn[movable_zone]); | |
5637 | ||
e506b996 XQ |
5638 | /* Adjust for ZONE_MOVABLE starting within this range */ |
5639 | } else if (!mirrored_kernelcore && | |
5640 | *zone_start_pfn < zone_movable_pfn[nid] && | |
5641 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
5642 | *zone_end_pfn = zone_movable_pfn[nid]; | |
5643 | ||
2a1e274a MG |
5644 | /* Check if this whole range is within ZONE_MOVABLE */ |
5645 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5646 | *zone_start_pfn = *zone_end_pfn; | |
5647 | } | |
5648 | } | |
5649 | ||
c713216d MG |
5650 | /* |
5651 | * Return the number of pages a zone spans in a node, including holes | |
5652 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5653 | */ | |
6ea6e688 | 5654 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5655 | unsigned long zone_type, |
7960aedd ZY |
5656 | unsigned long node_start_pfn, |
5657 | unsigned long node_end_pfn, | |
d91749c1 TI |
5658 | unsigned long *zone_start_pfn, |
5659 | unsigned long *zone_end_pfn, | |
c713216d MG |
5660 | unsigned long *ignored) |
5661 | { | |
b5685e92 | 5662 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5663 | if (!node_start_pfn && !node_end_pfn) |
5664 | return 0; | |
5665 | ||
7960aedd | 5666 | /* Get the start and end of the zone */ |
d91749c1 TI |
5667 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5668 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5669 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5670 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5671 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5672 | |
5673 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5674 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5675 | return 0; |
5676 | ||
5677 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5678 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5679 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5680 | |
5681 | /* Return the spanned pages */ | |
d91749c1 | 5682 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5683 | } |
5684 | ||
5685 | /* | |
5686 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5687 | * then all holes in the requested range will be accounted for. |
c713216d | 5688 | */ |
32996250 | 5689 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5690 | unsigned long range_start_pfn, |
5691 | unsigned long range_end_pfn) | |
5692 | { | |
96e907d1 TH |
5693 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5694 | unsigned long start_pfn, end_pfn; | |
5695 | int i; | |
c713216d | 5696 | |
96e907d1 TH |
5697 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5698 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5699 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5700 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5701 | } |
96e907d1 | 5702 | return nr_absent; |
c713216d MG |
5703 | } |
5704 | ||
5705 | /** | |
5706 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5707 | * @start_pfn: The start PFN to start searching for holes | |
5708 | * @end_pfn: The end PFN to stop searching for holes | |
5709 | * | |
88ca3b94 | 5710 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5711 | */ |
5712 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5713 | unsigned long end_pfn) | |
5714 | { | |
5715 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5716 | } | |
5717 | ||
5718 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5719 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5720 | unsigned long zone_type, |
7960aedd ZY |
5721 | unsigned long node_start_pfn, |
5722 | unsigned long node_end_pfn, | |
c713216d MG |
5723 | unsigned long *ignored) |
5724 | { | |
96e907d1 TH |
5725 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5726 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5727 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5728 | unsigned long nr_absent; |
9c7cd687 | 5729 | |
b5685e92 | 5730 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5731 | if (!node_start_pfn && !node_end_pfn) |
5732 | return 0; | |
5733 | ||
96e907d1 TH |
5734 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5735 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5736 | |
2a1e274a MG |
5737 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5738 | node_start_pfn, node_end_pfn, | |
5739 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5740 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5741 | ||
5742 | /* | |
5743 | * ZONE_MOVABLE handling. | |
5744 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
5745 | * and vice versa. | |
5746 | */ | |
e506b996 XQ |
5747 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
5748 | unsigned long start_pfn, end_pfn; | |
5749 | struct memblock_region *r; | |
5750 | ||
5751 | for_each_memblock(memory, r) { | |
5752 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
5753 | zone_start_pfn, zone_end_pfn); | |
5754 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
5755 | zone_start_pfn, zone_end_pfn); | |
5756 | ||
5757 | if (zone_type == ZONE_MOVABLE && | |
5758 | memblock_is_mirror(r)) | |
5759 | nr_absent += end_pfn - start_pfn; | |
5760 | ||
5761 | if (zone_type == ZONE_NORMAL && | |
5762 | !memblock_is_mirror(r)) | |
5763 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
5764 | } |
5765 | } | |
5766 | ||
5767 | return nr_absent; | |
c713216d | 5768 | } |
0e0b864e | 5769 | |
0ee332c1 | 5770 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 5771 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5772 | unsigned long zone_type, |
7960aedd ZY |
5773 | unsigned long node_start_pfn, |
5774 | unsigned long node_end_pfn, | |
d91749c1 TI |
5775 | unsigned long *zone_start_pfn, |
5776 | unsigned long *zone_end_pfn, | |
c713216d MG |
5777 | unsigned long *zones_size) |
5778 | { | |
d91749c1 TI |
5779 | unsigned int zone; |
5780 | ||
5781 | *zone_start_pfn = node_start_pfn; | |
5782 | for (zone = 0; zone < zone_type; zone++) | |
5783 | *zone_start_pfn += zones_size[zone]; | |
5784 | ||
5785 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
5786 | ||
c713216d MG |
5787 | return zones_size[zone_type]; |
5788 | } | |
5789 | ||
6ea6e688 | 5790 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5791 | unsigned long zone_type, |
7960aedd ZY |
5792 | unsigned long node_start_pfn, |
5793 | unsigned long node_end_pfn, | |
c713216d MG |
5794 | unsigned long *zholes_size) |
5795 | { | |
5796 | if (!zholes_size) | |
5797 | return 0; | |
5798 | ||
5799 | return zholes_size[zone_type]; | |
5800 | } | |
20e6926d | 5801 | |
0ee332c1 | 5802 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5803 | |
a3142c8e | 5804 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
5805 | unsigned long node_start_pfn, |
5806 | unsigned long node_end_pfn, | |
5807 | unsigned long *zones_size, | |
5808 | unsigned long *zholes_size) | |
c713216d | 5809 | { |
febd5949 | 5810 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
5811 | enum zone_type i; |
5812 | ||
febd5949 GZ |
5813 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5814 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 5815 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 5816 | unsigned long size, real_size; |
c713216d | 5817 | |
febd5949 GZ |
5818 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5819 | node_start_pfn, | |
5820 | node_end_pfn, | |
d91749c1 TI |
5821 | &zone_start_pfn, |
5822 | &zone_end_pfn, | |
febd5949 GZ |
5823 | zones_size); |
5824 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
5825 | node_start_pfn, node_end_pfn, |
5826 | zholes_size); | |
d91749c1 TI |
5827 | if (size) |
5828 | zone->zone_start_pfn = zone_start_pfn; | |
5829 | else | |
5830 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
5831 | zone->spanned_pages = size; |
5832 | zone->present_pages = real_size; | |
5833 | ||
5834 | totalpages += size; | |
5835 | realtotalpages += real_size; | |
5836 | } | |
5837 | ||
5838 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
5839 | pgdat->node_present_pages = realtotalpages; |
5840 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
5841 | realtotalpages); | |
5842 | } | |
5843 | ||
835c134e MG |
5844 | #ifndef CONFIG_SPARSEMEM |
5845 | /* | |
5846 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
5847 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5848 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
5849 | * round what is now in bits to nearest long in bits, then return it in |
5850 | * bytes. | |
5851 | */ | |
7c45512d | 5852 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
5853 | { |
5854 | unsigned long usemapsize; | |
5855 | ||
7c45512d | 5856 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
5857 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
5858 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
5859 | usemapsize *= NR_PAGEBLOCK_BITS; |
5860 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
5861 | ||
5862 | return usemapsize / 8; | |
5863 | } | |
5864 | ||
5865 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
5866 | struct zone *zone, |
5867 | unsigned long zone_start_pfn, | |
5868 | unsigned long zonesize) | |
835c134e | 5869 | { |
7c45512d | 5870 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 5871 | zone->pageblock_flags = NULL; |
58a01a45 | 5872 | if (usemapsize) |
6782832e SS |
5873 | zone->pageblock_flags = |
5874 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5875 | pgdat->node_id); | |
835c134e MG |
5876 | } |
5877 | #else | |
7c45512d LT |
5878 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5879 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5880 | #endif /* CONFIG_SPARSEMEM */ |
5881 | ||
d9c23400 | 5882 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5883 | |
d9c23400 | 5884 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5885 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5886 | { |
955c1cd7 AM |
5887 | unsigned int order; |
5888 | ||
d9c23400 MG |
5889 | /* Check that pageblock_nr_pages has not already been setup */ |
5890 | if (pageblock_order) | |
5891 | return; | |
5892 | ||
955c1cd7 AM |
5893 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5894 | order = HUGETLB_PAGE_ORDER; | |
5895 | else | |
5896 | order = MAX_ORDER - 1; | |
5897 | ||
d9c23400 MG |
5898 | /* |
5899 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5900 | * This value may be variable depending on boot parameters on IA64 and |
5901 | * powerpc. | |
d9c23400 MG |
5902 | */ |
5903 | pageblock_order = order; | |
5904 | } | |
5905 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5906 | ||
ba72cb8c MG |
5907 | /* |
5908 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5909 | * is unused as pageblock_order is set at compile-time. See |
5910 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5911 | * the kernel config | |
ba72cb8c | 5912 | */ |
15ca220e | 5913 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5914 | { |
ba72cb8c | 5915 | } |
d9c23400 MG |
5916 | |
5917 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5918 | ||
01cefaef JL |
5919 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5920 | unsigned long present_pages) | |
5921 | { | |
5922 | unsigned long pages = spanned_pages; | |
5923 | ||
5924 | /* | |
5925 | * Provide a more accurate estimation if there are holes within | |
5926 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5927 | * zone, each populated memory region may cost us one or two extra | |
5928 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 5929 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
5930 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
5931 | */ | |
5932 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5933 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5934 | pages = present_pages; | |
5935 | ||
5936 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5937 | } | |
5938 | ||
1da177e4 LT |
5939 | /* |
5940 | * Set up the zone data structures: | |
5941 | * - mark all pages reserved | |
5942 | * - mark all memory queues empty | |
5943 | * - clear the memory bitmaps | |
6527af5d MK |
5944 | * |
5945 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 5946 | */ |
7f3eb55b | 5947 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 5948 | { |
2f1b6248 | 5949 | enum zone_type j; |
ed8ece2e | 5950 | int nid = pgdat->node_id; |
718127cc | 5951 | int ret; |
1da177e4 | 5952 | |
208d54e5 | 5953 | pgdat_resize_init(pgdat); |
8177a420 AA |
5954 | #ifdef CONFIG_NUMA_BALANCING |
5955 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
5956 | pgdat->numabalancing_migrate_nr_pages = 0; | |
5957 | pgdat->numabalancing_migrate_next_window = jiffies; | |
a3d0a918 KS |
5958 | #endif |
5959 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5960 | spin_lock_init(&pgdat->split_queue_lock); | |
5961 | INIT_LIST_HEAD(&pgdat->split_queue); | |
5962 | pgdat->split_queue_len = 0; | |
8177a420 | 5963 | #endif |
1da177e4 | 5964 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 5965 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
698b1b30 VB |
5966 | #ifdef CONFIG_COMPACTION |
5967 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
5968 | #endif | |
eefa864b | 5969 | pgdat_page_ext_init(pgdat); |
a52633d8 | 5970 | spin_lock_init(&pgdat->lru_lock); |
a9dd0a83 | 5971 | lruvec_init(node_lruvec(pgdat)); |
5f63b720 | 5972 | |
1da177e4 LT |
5973 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5974 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 5975 | unsigned long size, realsize, freesize, memmap_pages; |
d91749c1 | 5976 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 5977 | |
febd5949 GZ |
5978 | size = zone->spanned_pages; |
5979 | realsize = freesize = zone->present_pages; | |
1da177e4 | 5980 | |
0e0b864e | 5981 | /* |
9feedc9d | 5982 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
5983 | * is used by this zone for memmap. This affects the watermark |
5984 | * and per-cpu initialisations | |
5985 | */ | |
01cefaef | 5986 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
5987 | if (!is_highmem_idx(j)) { |
5988 | if (freesize >= memmap_pages) { | |
5989 | freesize -= memmap_pages; | |
5990 | if (memmap_pages) | |
5991 | printk(KERN_DEBUG | |
5992 | " %s zone: %lu pages used for memmap\n", | |
5993 | zone_names[j], memmap_pages); | |
5994 | } else | |
1170532b | 5995 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
5996 | zone_names[j], memmap_pages, freesize); |
5997 | } | |
0e0b864e | 5998 | |
6267276f | 5999 | /* Account for reserved pages */ |
9feedc9d JL |
6000 | if (j == 0 && freesize > dma_reserve) { |
6001 | freesize -= dma_reserve; | |
d903ef9f | 6002 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6003 | zone_names[0], dma_reserve); |
0e0b864e MG |
6004 | } |
6005 | ||
98d2b0eb | 6006 | if (!is_highmem_idx(j)) |
9feedc9d | 6007 | nr_kernel_pages += freesize; |
01cefaef JL |
6008 | /* Charge for highmem memmap if there are enough kernel pages */ |
6009 | else if (nr_kernel_pages > memmap_pages * 2) | |
6010 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6011 | nr_all_pages += freesize; |
1da177e4 | 6012 | |
9feedc9d JL |
6013 | /* |
6014 | * Set an approximate value for lowmem here, it will be adjusted | |
6015 | * when the bootmem allocator frees pages into the buddy system. | |
6016 | * And all highmem pages will be managed by the buddy system. | |
6017 | */ | |
6018 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 6019 | #ifdef CONFIG_NUMA |
d5f541ed | 6020 | zone->node = nid; |
9614634f | 6021 | #endif |
1da177e4 | 6022 | zone->name = zone_names[j]; |
a52633d8 | 6023 | zone->zone_pgdat = pgdat; |
1da177e4 | 6024 | spin_lock_init(&zone->lock); |
bdc8cb98 | 6025 | zone_seqlock_init(zone); |
ed8ece2e | 6026 | zone_pcp_init(zone); |
81c0a2bb | 6027 | |
1da177e4 LT |
6028 | if (!size) |
6029 | continue; | |
6030 | ||
955c1cd7 | 6031 | set_pageblock_order(); |
7c45512d | 6032 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
b171e409 | 6033 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
718127cc | 6034 | BUG_ON(ret); |
76cdd58e | 6035 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6036 | } |
6037 | } | |
6038 | ||
bd721ea7 | 6039 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6040 | { |
b0aeba74 | 6041 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6042 | unsigned long __maybe_unused offset = 0; |
6043 | ||
1da177e4 LT |
6044 | /* Skip empty nodes */ |
6045 | if (!pgdat->node_spanned_pages) | |
6046 | return; | |
6047 | ||
d41dee36 | 6048 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
b0aeba74 TL |
6049 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6050 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6051 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6052 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6053 | unsigned long size, end; |
d41dee36 AW |
6054 | struct page *map; |
6055 | ||
e984bb43 BP |
6056 | /* |
6057 | * The zone's endpoints aren't required to be MAX_ORDER | |
6058 | * aligned but the node_mem_map endpoints must be in order | |
6059 | * for the buddy allocator to function correctly. | |
6060 | */ | |
108bcc96 | 6061 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6062 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6063 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
6064 | map = alloc_remap(pgdat->node_id, size); |
6065 | if (!map) | |
6782832e SS |
6066 | map = memblock_virt_alloc_node_nopanic(size, |
6067 | pgdat->node_id); | |
a1c34a3b | 6068 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6069 | } |
12d810c1 | 6070 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6071 | /* |
6072 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6073 | */ | |
c713216d | 6074 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6075 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 6076 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 6077 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6078 | mem_map -= offset; |
0ee332c1 | 6079 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6080 | } |
1da177e4 | 6081 | #endif |
d41dee36 | 6082 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
6083 | } |
6084 | ||
9109fb7b JW |
6085 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
6086 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 6087 | { |
9109fb7b | 6088 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6089 | unsigned long start_pfn = 0; |
6090 | unsigned long end_pfn = 0; | |
9109fb7b | 6091 | |
88fdf75d | 6092 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 6093 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 6094 | |
3a80a7fa | 6095 | reset_deferred_meminit(pgdat); |
1da177e4 LT |
6096 | pgdat->node_id = nid; |
6097 | pgdat->node_start_pfn = node_start_pfn; | |
75ef7184 | 6098 | pgdat->per_cpu_nodestats = NULL; |
7960aedd ZY |
6099 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6100 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 6101 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6102 | (u64)start_pfn << PAGE_SHIFT, |
6103 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
6104 | #else |
6105 | start_pfn = node_start_pfn; | |
7960aedd ZY |
6106 | #endif |
6107 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
6108 | zones_size, zholes_size); | |
1da177e4 LT |
6109 | |
6110 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
6111 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
6112 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
6113 | nid, (unsigned long)pgdat, | |
6114 | (unsigned long)pgdat->node_mem_map); | |
6115 | #endif | |
1da177e4 | 6116 | |
7f3eb55b | 6117 | free_area_init_core(pgdat); |
1da177e4 LT |
6118 | } |
6119 | ||
0ee332c1 | 6120 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
6121 | |
6122 | #if MAX_NUMNODES > 1 | |
6123 | /* | |
6124 | * Figure out the number of possible node ids. | |
6125 | */ | |
f9872caf | 6126 | void __init setup_nr_node_ids(void) |
418508c1 | 6127 | { |
904a9553 | 6128 | unsigned int highest; |
418508c1 | 6129 | |
904a9553 | 6130 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
6131 | nr_node_ids = highest + 1; |
6132 | } | |
418508c1 MS |
6133 | #endif |
6134 | ||
1e01979c TH |
6135 | /** |
6136 | * node_map_pfn_alignment - determine the maximum internode alignment | |
6137 | * | |
6138 | * This function should be called after node map is populated and sorted. | |
6139 | * It calculates the maximum power of two alignment which can distinguish | |
6140 | * all the nodes. | |
6141 | * | |
6142 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
6143 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
6144 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
6145 | * shifted, 1GiB is enough and this function will indicate so. | |
6146 | * | |
6147 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
6148 | * model has fine enough granularity to avoid incorrect mapping for the | |
6149 | * populated node map. | |
6150 | * | |
6151 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
6152 | * requirement (single node). | |
6153 | */ | |
6154 | unsigned long __init node_map_pfn_alignment(void) | |
6155 | { | |
6156 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 6157 | unsigned long start, end, mask; |
1e01979c | 6158 | int last_nid = -1; |
c13291a5 | 6159 | int i, nid; |
1e01979c | 6160 | |
c13291a5 | 6161 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
6162 | if (!start || last_nid < 0 || last_nid == nid) { |
6163 | last_nid = nid; | |
6164 | last_end = end; | |
6165 | continue; | |
6166 | } | |
6167 | ||
6168 | /* | |
6169 | * Start with a mask granular enough to pin-point to the | |
6170 | * start pfn and tick off bits one-by-one until it becomes | |
6171 | * too coarse to separate the current node from the last. | |
6172 | */ | |
6173 | mask = ~((1 << __ffs(start)) - 1); | |
6174 | while (mask && last_end <= (start & (mask << 1))) | |
6175 | mask <<= 1; | |
6176 | ||
6177 | /* accumulate all internode masks */ | |
6178 | accl_mask |= mask; | |
6179 | } | |
6180 | ||
6181 | /* convert mask to number of pages */ | |
6182 | return ~accl_mask + 1; | |
6183 | } | |
6184 | ||
a6af2bc3 | 6185 | /* Find the lowest pfn for a node */ |
b69a7288 | 6186 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 6187 | { |
a6af2bc3 | 6188 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
6189 | unsigned long start_pfn; |
6190 | int i; | |
1abbfb41 | 6191 | |
c13291a5 TH |
6192 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
6193 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 6194 | |
a6af2bc3 | 6195 | if (min_pfn == ULONG_MAX) { |
1170532b | 6196 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
6197 | return 0; |
6198 | } | |
6199 | ||
6200 | return min_pfn; | |
c713216d MG |
6201 | } |
6202 | ||
6203 | /** | |
6204 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
6205 | * | |
6206 | * It returns the minimum PFN based on information provided via | |
7d018176 | 6207 | * memblock_set_node(). |
c713216d MG |
6208 | */ |
6209 | unsigned long __init find_min_pfn_with_active_regions(void) | |
6210 | { | |
6211 | return find_min_pfn_for_node(MAX_NUMNODES); | |
6212 | } | |
6213 | ||
37b07e41 LS |
6214 | /* |
6215 | * early_calculate_totalpages() | |
6216 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 6217 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 6218 | */ |
484f51f8 | 6219 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 6220 | { |
7e63efef | 6221 | unsigned long totalpages = 0; |
c13291a5 TH |
6222 | unsigned long start_pfn, end_pfn; |
6223 | int i, nid; | |
6224 | ||
6225 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
6226 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 6227 | |
37b07e41 LS |
6228 | totalpages += pages; |
6229 | if (pages) | |
4b0ef1fe | 6230 | node_set_state(nid, N_MEMORY); |
37b07e41 | 6231 | } |
b8af2941 | 6232 | return totalpages; |
7e63efef MG |
6233 | } |
6234 | ||
2a1e274a MG |
6235 | /* |
6236 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
6237 | * is spread evenly between nodes as long as the nodes have enough | |
6238 | * memory. When they don't, some nodes will have more kernelcore than | |
6239 | * others | |
6240 | */ | |
b224ef85 | 6241 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
6242 | { |
6243 | int i, nid; | |
6244 | unsigned long usable_startpfn; | |
6245 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 6246 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 6247 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 6248 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 6249 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 6250 | struct memblock_region *r; |
b2f3eebe TC |
6251 | |
6252 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
6253 | find_usable_zone_for_movable(); | |
6254 | ||
6255 | /* | |
6256 | * If movable_node is specified, ignore kernelcore and movablecore | |
6257 | * options. | |
6258 | */ | |
6259 | if (movable_node_is_enabled()) { | |
136199f0 EM |
6260 | for_each_memblock(memory, r) { |
6261 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
6262 | continue; |
6263 | ||
136199f0 | 6264 | nid = r->nid; |
b2f3eebe | 6265 | |
136199f0 | 6266 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
6267 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6268 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6269 | usable_startpfn; | |
6270 | } | |
6271 | ||
6272 | goto out2; | |
6273 | } | |
2a1e274a | 6274 | |
342332e6 TI |
6275 | /* |
6276 | * If kernelcore=mirror is specified, ignore movablecore option | |
6277 | */ | |
6278 | if (mirrored_kernelcore) { | |
6279 | bool mem_below_4gb_not_mirrored = false; | |
6280 | ||
6281 | for_each_memblock(memory, r) { | |
6282 | if (memblock_is_mirror(r)) | |
6283 | continue; | |
6284 | ||
6285 | nid = r->nid; | |
6286 | ||
6287 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
6288 | ||
6289 | if (usable_startpfn < 0x100000) { | |
6290 | mem_below_4gb_not_mirrored = true; | |
6291 | continue; | |
6292 | } | |
6293 | ||
6294 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
6295 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6296 | usable_startpfn; | |
6297 | } | |
6298 | ||
6299 | if (mem_below_4gb_not_mirrored) | |
6300 | pr_warn("This configuration results in unmirrored kernel memory."); | |
6301 | ||
6302 | goto out2; | |
6303 | } | |
6304 | ||
7e63efef | 6305 | /* |
b2f3eebe | 6306 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
6307 | * kernelcore that corresponds so that memory usable for |
6308 | * any allocation type is evenly spread. If both kernelcore | |
6309 | * and movablecore are specified, then the value of kernelcore | |
6310 | * will be used for required_kernelcore if it's greater than | |
6311 | * what movablecore would have allowed. | |
6312 | */ | |
6313 | if (required_movablecore) { | |
7e63efef MG |
6314 | unsigned long corepages; |
6315 | ||
6316 | /* | |
6317 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
6318 | * was requested by the user | |
6319 | */ | |
6320 | required_movablecore = | |
6321 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 6322 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
6323 | corepages = totalpages - required_movablecore; |
6324 | ||
6325 | required_kernelcore = max(required_kernelcore, corepages); | |
6326 | } | |
6327 | ||
bde304bd XQ |
6328 | /* |
6329 | * If kernelcore was not specified or kernelcore size is larger | |
6330 | * than totalpages, there is no ZONE_MOVABLE. | |
6331 | */ | |
6332 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 6333 | goto out; |
2a1e274a MG |
6334 | |
6335 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
6336 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6337 | ||
6338 | restart: | |
6339 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
6340 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 6341 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
6342 | unsigned long start_pfn, end_pfn; |
6343 | ||
2a1e274a MG |
6344 | /* |
6345 | * Recalculate kernelcore_node if the division per node | |
6346 | * now exceeds what is necessary to satisfy the requested | |
6347 | * amount of memory for the kernel | |
6348 | */ | |
6349 | if (required_kernelcore < kernelcore_node) | |
6350 | kernelcore_node = required_kernelcore / usable_nodes; | |
6351 | ||
6352 | /* | |
6353 | * As the map is walked, we track how much memory is usable | |
6354 | * by the kernel using kernelcore_remaining. When it is | |
6355 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
6356 | */ | |
6357 | kernelcore_remaining = kernelcore_node; | |
6358 | ||
6359 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 6360 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
6361 | unsigned long size_pages; |
6362 | ||
c13291a5 | 6363 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
6364 | if (start_pfn >= end_pfn) |
6365 | continue; | |
6366 | ||
6367 | /* Account for what is only usable for kernelcore */ | |
6368 | if (start_pfn < usable_startpfn) { | |
6369 | unsigned long kernel_pages; | |
6370 | kernel_pages = min(end_pfn, usable_startpfn) | |
6371 | - start_pfn; | |
6372 | ||
6373 | kernelcore_remaining -= min(kernel_pages, | |
6374 | kernelcore_remaining); | |
6375 | required_kernelcore -= min(kernel_pages, | |
6376 | required_kernelcore); | |
6377 | ||
6378 | /* Continue if range is now fully accounted */ | |
6379 | if (end_pfn <= usable_startpfn) { | |
6380 | ||
6381 | /* | |
6382 | * Push zone_movable_pfn to the end so | |
6383 | * that if we have to rebalance | |
6384 | * kernelcore across nodes, we will | |
6385 | * not double account here | |
6386 | */ | |
6387 | zone_movable_pfn[nid] = end_pfn; | |
6388 | continue; | |
6389 | } | |
6390 | start_pfn = usable_startpfn; | |
6391 | } | |
6392 | ||
6393 | /* | |
6394 | * The usable PFN range for ZONE_MOVABLE is from | |
6395 | * start_pfn->end_pfn. Calculate size_pages as the | |
6396 | * number of pages used as kernelcore | |
6397 | */ | |
6398 | size_pages = end_pfn - start_pfn; | |
6399 | if (size_pages > kernelcore_remaining) | |
6400 | size_pages = kernelcore_remaining; | |
6401 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6402 | ||
6403 | /* | |
6404 | * Some kernelcore has been met, update counts and | |
6405 | * break if the kernelcore for this node has been | |
b8af2941 | 6406 | * satisfied |
2a1e274a MG |
6407 | */ |
6408 | required_kernelcore -= min(required_kernelcore, | |
6409 | size_pages); | |
6410 | kernelcore_remaining -= size_pages; | |
6411 | if (!kernelcore_remaining) | |
6412 | break; | |
6413 | } | |
6414 | } | |
6415 | ||
6416 | /* | |
6417 | * If there is still required_kernelcore, we do another pass with one | |
6418 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6419 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6420 | * satisfied |
2a1e274a MG |
6421 | */ |
6422 | usable_nodes--; | |
6423 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6424 | goto restart; | |
6425 | ||
b2f3eebe | 6426 | out2: |
2a1e274a MG |
6427 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6428 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6429 | zone_movable_pfn[nid] = | |
6430 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6431 | |
20e6926d | 6432 | out: |
66918dcd | 6433 | /* restore the node_state */ |
4b0ef1fe | 6434 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6435 | } |
6436 | ||
4b0ef1fe LJ |
6437 | /* Any regular or high memory on that node ? */ |
6438 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6439 | { |
37b07e41 LS |
6440 | enum zone_type zone_type; |
6441 | ||
4b0ef1fe LJ |
6442 | if (N_MEMORY == N_NORMAL_MEMORY) |
6443 | return; | |
6444 | ||
6445 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 6446 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6447 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
6448 | node_set_state(nid, N_HIGH_MEMORY); |
6449 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
6450 | zone_type <= ZONE_NORMAL) | |
6451 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
6452 | break; |
6453 | } | |
37b07e41 | 6454 | } |
37b07e41 LS |
6455 | } |
6456 | ||
c713216d MG |
6457 | /** |
6458 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6459 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6460 | * |
6461 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6462 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6463 | * zone in each node and their holes is calculated. If the maximum PFN |
6464 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6465 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6466 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6467 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6468 | * at arch_max_dma_pfn. | |
6469 | */ | |
6470 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6471 | { | |
c13291a5 TH |
6472 | unsigned long start_pfn, end_pfn; |
6473 | int i, nid; | |
a6af2bc3 | 6474 | |
c713216d MG |
6475 | /* Record where the zone boundaries are */ |
6476 | memset(arch_zone_lowest_possible_pfn, 0, | |
6477 | sizeof(arch_zone_lowest_possible_pfn)); | |
6478 | memset(arch_zone_highest_possible_pfn, 0, | |
6479 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
6480 | |
6481 | start_pfn = find_min_pfn_with_active_regions(); | |
6482 | ||
6483 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6484 | if (i == ZONE_MOVABLE) |
6485 | continue; | |
90cae1fe OH |
6486 | |
6487 | end_pfn = max(max_zone_pfn[i], start_pfn); | |
6488 | arch_zone_lowest_possible_pfn[i] = start_pfn; | |
6489 | arch_zone_highest_possible_pfn[i] = end_pfn; | |
6490 | ||
6491 | start_pfn = end_pfn; | |
c713216d | 6492 | } |
2a1e274a MG |
6493 | |
6494 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6495 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6496 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6497 | |
c713216d | 6498 | /* Print out the zone ranges */ |
f88dfff5 | 6499 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6500 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6501 | if (i == ZONE_MOVABLE) | |
6502 | continue; | |
f88dfff5 | 6503 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6504 | if (arch_zone_lowest_possible_pfn[i] == |
6505 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6506 | pr_cont("empty\n"); |
72f0ba02 | 6507 | else |
8d29e18a JG |
6508 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6509 | (u64)arch_zone_lowest_possible_pfn[i] | |
6510 | << PAGE_SHIFT, | |
6511 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6512 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6513 | } |
6514 | ||
6515 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6516 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6517 | for (i = 0; i < MAX_NUMNODES; i++) { |
6518 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6519 | pr_info(" Node %d: %#018Lx\n", i, |
6520 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6521 | } |
c713216d | 6522 | |
f2d52fe5 | 6523 | /* Print out the early node map */ |
f88dfff5 | 6524 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6525 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6526 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6527 | (u64)start_pfn << PAGE_SHIFT, | |
6528 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6529 | |
6530 | /* Initialise every node */ | |
708614e6 | 6531 | mminit_verify_pageflags_layout(); |
8ef82866 | 6532 | setup_nr_node_ids(); |
c713216d MG |
6533 | for_each_online_node(nid) { |
6534 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6535 | free_area_init_node(nid, NULL, |
c713216d | 6536 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6537 | |
6538 | /* Any memory on that node */ | |
6539 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6540 | node_set_state(nid, N_MEMORY); |
6541 | check_for_memory(pgdat, nid); | |
c713216d MG |
6542 | } |
6543 | } | |
2a1e274a | 6544 | |
7e63efef | 6545 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
6546 | { |
6547 | unsigned long long coremem; | |
6548 | if (!p) | |
6549 | return -EINVAL; | |
6550 | ||
6551 | coremem = memparse(p, &p); | |
7e63efef | 6552 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 6553 | |
7e63efef | 6554 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
6555 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6556 | ||
6557 | return 0; | |
6558 | } | |
ed7ed365 | 6559 | |
7e63efef MG |
6560 | /* |
6561 | * kernelcore=size sets the amount of memory for use for allocations that | |
6562 | * cannot be reclaimed or migrated. | |
6563 | */ | |
6564 | static int __init cmdline_parse_kernelcore(char *p) | |
6565 | { | |
342332e6 TI |
6566 | /* parse kernelcore=mirror */ |
6567 | if (parse_option_str(p, "mirror")) { | |
6568 | mirrored_kernelcore = true; | |
6569 | return 0; | |
6570 | } | |
6571 | ||
7e63efef MG |
6572 | return cmdline_parse_core(p, &required_kernelcore); |
6573 | } | |
6574 | ||
6575 | /* | |
6576 | * movablecore=size sets the amount of memory for use for allocations that | |
6577 | * can be reclaimed or migrated. | |
6578 | */ | |
6579 | static int __init cmdline_parse_movablecore(char *p) | |
6580 | { | |
6581 | return cmdline_parse_core(p, &required_movablecore); | |
6582 | } | |
6583 | ||
ed7ed365 | 6584 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6585 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6586 | |
0ee332c1 | 6587 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6588 | |
c3d5f5f0 JL |
6589 | void adjust_managed_page_count(struct page *page, long count) |
6590 | { | |
6591 | spin_lock(&managed_page_count_lock); | |
6592 | page_zone(page)->managed_pages += count; | |
6593 | totalram_pages += count; | |
3dcc0571 JL |
6594 | #ifdef CONFIG_HIGHMEM |
6595 | if (PageHighMem(page)) | |
6596 | totalhigh_pages += count; | |
6597 | #endif | |
c3d5f5f0 JL |
6598 | spin_unlock(&managed_page_count_lock); |
6599 | } | |
3dcc0571 | 6600 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6601 | |
11199692 | 6602 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6603 | { |
11199692 JL |
6604 | void *pos; |
6605 | unsigned long pages = 0; | |
69afade7 | 6606 | |
11199692 JL |
6607 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6608 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6609 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 6610 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
6611 | memset(pos, poison, PAGE_SIZE); |
6612 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
6613 | } |
6614 | ||
6615 | if (pages && s) | |
adb1fe9a JP |
6616 | pr_info("Freeing %s memory: %ldK\n", |
6617 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
6618 | |
6619 | return pages; | |
6620 | } | |
11199692 | 6621 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6622 | |
cfa11e08 JL |
6623 | #ifdef CONFIG_HIGHMEM |
6624 | void free_highmem_page(struct page *page) | |
6625 | { | |
6626 | __free_reserved_page(page); | |
6627 | totalram_pages++; | |
7b4b2a0d | 6628 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6629 | totalhigh_pages++; |
6630 | } | |
6631 | #endif | |
6632 | ||
7ee3d4e8 JL |
6633 | |
6634 | void __init mem_init_print_info(const char *str) | |
6635 | { | |
6636 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
6637 | unsigned long init_code_size, init_data_size; | |
6638 | ||
6639 | physpages = get_num_physpages(); | |
6640 | codesize = _etext - _stext; | |
6641 | datasize = _edata - _sdata; | |
6642 | rosize = __end_rodata - __start_rodata; | |
6643 | bss_size = __bss_stop - __bss_start; | |
6644 | init_data_size = __init_end - __init_begin; | |
6645 | init_code_size = _einittext - _sinittext; | |
6646 | ||
6647 | /* | |
6648 | * Detect special cases and adjust section sizes accordingly: | |
6649 | * 1) .init.* may be embedded into .data sections | |
6650 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
6651 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
6652 | * 3) .rodata.* may be embedded into .text or .data sections. | |
6653 | */ | |
6654 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
6655 | do { \ |
6656 | if (start <= pos && pos < end && size > adj) \ | |
6657 | size -= adj; \ | |
6658 | } while (0) | |
7ee3d4e8 JL |
6659 | |
6660 | adj_init_size(__init_begin, __init_end, init_data_size, | |
6661 | _sinittext, init_code_size); | |
6662 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
6663 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
6664 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
6665 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
6666 | ||
6667 | #undef adj_init_size | |
6668 | ||
756a025f | 6669 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 6670 | #ifdef CONFIG_HIGHMEM |
756a025f | 6671 | ", %luK highmem" |
7ee3d4e8 | 6672 | #endif |
756a025f JP |
6673 | "%s%s)\n", |
6674 | nr_free_pages() << (PAGE_SHIFT - 10), | |
6675 | physpages << (PAGE_SHIFT - 10), | |
6676 | codesize >> 10, datasize >> 10, rosize >> 10, | |
6677 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
6678 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
6679 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 6680 | #ifdef CONFIG_HIGHMEM |
756a025f | 6681 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 6682 | #endif |
756a025f | 6683 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
6684 | } |
6685 | ||
0e0b864e | 6686 | /** |
88ca3b94 RD |
6687 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6688 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 6689 | * |
013110a7 | 6690 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
6691 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6692 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
6693 | * function may optionally be used to account for unfreeable pages in the |
6694 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
6695 | * smaller per-cpu batchsize. | |
0e0b864e MG |
6696 | */ |
6697 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
6698 | { | |
6699 | dma_reserve = new_dma_reserve; | |
6700 | } | |
6701 | ||
1da177e4 LT |
6702 | void __init free_area_init(unsigned long *zones_size) |
6703 | { | |
9109fb7b | 6704 | free_area_init_node(0, zones_size, |
1da177e4 LT |
6705 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6706 | } | |
1da177e4 | 6707 | |
005fd4bb | 6708 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 6709 | { |
1da177e4 | 6710 | |
005fd4bb SAS |
6711 | lru_add_drain_cpu(cpu); |
6712 | drain_pages(cpu); | |
9f8f2172 | 6713 | |
005fd4bb SAS |
6714 | /* |
6715 | * Spill the event counters of the dead processor | |
6716 | * into the current processors event counters. | |
6717 | * This artificially elevates the count of the current | |
6718 | * processor. | |
6719 | */ | |
6720 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 6721 | |
005fd4bb SAS |
6722 | /* |
6723 | * Zero the differential counters of the dead processor | |
6724 | * so that the vm statistics are consistent. | |
6725 | * | |
6726 | * This is only okay since the processor is dead and cannot | |
6727 | * race with what we are doing. | |
6728 | */ | |
6729 | cpu_vm_stats_fold(cpu); | |
6730 | return 0; | |
1da177e4 | 6731 | } |
1da177e4 LT |
6732 | |
6733 | void __init page_alloc_init(void) | |
6734 | { | |
005fd4bb SAS |
6735 | int ret; |
6736 | ||
6737 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, | |
6738 | "mm/page_alloc:dead", NULL, | |
6739 | page_alloc_cpu_dead); | |
6740 | WARN_ON(ret < 0); | |
1da177e4 LT |
6741 | } |
6742 | ||
cb45b0e9 | 6743 | /* |
34b10060 | 6744 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
6745 | * or min_free_kbytes changes. |
6746 | */ | |
6747 | static void calculate_totalreserve_pages(void) | |
6748 | { | |
6749 | struct pglist_data *pgdat; | |
6750 | unsigned long reserve_pages = 0; | |
2f6726e5 | 6751 | enum zone_type i, j; |
cb45b0e9 HA |
6752 | |
6753 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
6754 | |
6755 | pgdat->totalreserve_pages = 0; | |
6756 | ||
cb45b0e9 HA |
6757 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6758 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 6759 | long max = 0; |
cb45b0e9 HA |
6760 | |
6761 | /* Find valid and maximum lowmem_reserve in the zone */ | |
6762 | for (j = i; j < MAX_NR_ZONES; j++) { | |
6763 | if (zone->lowmem_reserve[j] > max) | |
6764 | max = zone->lowmem_reserve[j]; | |
6765 | } | |
6766 | ||
41858966 MG |
6767 | /* we treat the high watermark as reserved pages. */ |
6768 | max += high_wmark_pages(zone); | |
cb45b0e9 | 6769 | |
b40da049 JL |
6770 | if (max > zone->managed_pages) |
6771 | max = zone->managed_pages; | |
a8d01437 | 6772 | |
281e3726 | 6773 | pgdat->totalreserve_pages += max; |
a8d01437 | 6774 | |
cb45b0e9 HA |
6775 | reserve_pages += max; |
6776 | } | |
6777 | } | |
6778 | totalreserve_pages = reserve_pages; | |
6779 | } | |
6780 | ||
1da177e4 LT |
6781 | /* |
6782 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 6783 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
6784 | * has a correct pages reserved value, so an adequate number of |
6785 | * pages are left in the zone after a successful __alloc_pages(). | |
6786 | */ | |
6787 | static void setup_per_zone_lowmem_reserve(void) | |
6788 | { | |
6789 | struct pglist_data *pgdat; | |
2f6726e5 | 6790 | enum zone_type j, idx; |
1da177e4 | 6791 | |
ec936fc5 | 6792 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
6793 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6794 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 6795 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
6796 | |
6797 | zone->lowmem_reserve[j] = 0; | |
6798 | ||
2f6726e5 CL |
6799 | idx = j; |
6800 | while (idx) { | |
1da177e4 LT |
6801 | struct zone *lower_zone; |
6802 | ||
2f6726e5 CL |
6803 | idx--; |
6804 | ||
1da177e4 LT |
6805 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6806 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
6807 | ||
6808 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 6809 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 6810 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 6811 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
6812 | } |
6813 | } | |
6814 | } | |
cb45b0e9 HA |
6815 | |
6816 | /* update totalreserve_pages */ | |
6817 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6818 | } |
6819 | ||
cfd3da1e | 6820 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
6821 | { |
6822 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
6823 | unsigned long lowmem_pages = 0; | |
6824 | struct zone *zone; | |
6825 | unsigned long flags; | |
6826 | ||
6827 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
6828 | for_each_zone(zone) { | |
6829 | if (!is_highmem(zone)) | |
b40da049 | 6830 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
6831 | } |
6832 | ||
6833 | for_each_zone(zone) { | |
ac924c60 AM |
6834 | u64 tmp; |
6835 | ||
1125b4e3 | 6836 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 6837 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 6838 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
6839 | if (is_highmem(zone)) { |
6840 | /* | |
669ed175 NP |
6841 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6842 | * need highmem pages, so cap pages_min to a small | |
6843 | * value here. | |
6844 | * | |
41858966 | 6845 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 6846 | * deltas control asynch page reclaim, and so should |
669ed175 | 6847 | * not be capped for highmem. |
1da177e4 | 6848 | */ |
90ae8d67 | 6849 | unsigned long min_pages; |
1da177e4 | 6850 | |
b40da049 | 6851 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 6852 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 6853 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 6854 | } else { |
669ed175 NP |
6855 | /* |
6856 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
6857 | * proportionate to the zone's size. |
6858 | */ | |
41858966 | 6859 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
6860 | } |
6861 | ||
795ae7a0 JW |
6862 | /* |
6863 | * Set the kswapd watermarks distance according to the | |
6864 | * scale factor in proportion to available memory, but | |
6865 | * ensure a minimum size on small systems. | |
6866 | */ | |
6867 | tmp = max_t(u64, tmp >> 2, | |
6868 | mult_frac(zone->managed_pages, | |
6869 | watermark_scale_factor, 10000)); | |
6870 | ||
6871 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
6872 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 6873 | |
1125b4e3 | 6874 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 6875 | } |
cb45b0e9 HA |
6876 | |
6877 | /* update totalreserve_pages */ | |
6878 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6879 | } |
6880 | ||
cfd3da1e MG |
6881 | /** |
6882 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
6883 | * or when memory is hot-{added|removed} | |
6884 | * | |
6885 | * Ensures that the watermark[min,low,high] values for each zone are set | |
6886 | * correctly with respect to min_free_kbytes. | |
6887 | */ | |
6888 | void setup_per_zone_wmarks(void) | |
6889 | { | |
6890 | mutex_lock(&zonelists_mutex); | |
6891 | __setup_per_zone_wmarks(); | |
6892 | mutex_unlock(&zonelists_mutex); | |
6893 | } | |
6894 | ||
1da177e4 LT |
6895 | /* |
6896 | * Initialise min_free_kbytes. | |
6897 | * | |
6898 | * For small machines we want it small (128k min). For large machines | |
6899 | * we want it large (64MB max). But it is not linear, because network | |
6900 | * bandwidth does not increase linearly with machine size. We use | |
6901 | * | |
b8af2941 | 6902 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6903 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6904 | * | |
6905 | * which yields | |
6906 | * | |
6907 | * 16MB: 512k | |
6908 | * 32MB: 724k | |
6909 | * 64MB: 1024k | |
6910 | * 128MB: 1448k | |
6911 | * 256MB: 2048k | |
6912 | * 512MB: 2896k | |
6913 | * 1024MB: 4096k | |
6914 | * 2048MB: 5792k | |
6915 | * 4096MB: 8192k | |
6916 | * 8192MB: 11584k | |
6917 | * 16384MB: 16384k | |
6918 | */ | |
1b79acc9 | 6919 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6920 | { |
6921 | unsigned long lowmem_kbytes; | |
5f12733e | 6922 | int new_min_free_kbytes; |
1da177e4 LT |
6923 | |
6924 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6925 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6926 | ||
6927 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6928 | min_free_kbytes = new_min_free_kbytes; | |
6929 | if (min_free_kbytes < 128) | |
6930 | min_free_kbytes = 128; | |
6931 | if (min_free_kbytes > 65536) | |
6932 | min_free_kbytes = 65536; | |
6933 | } else { | |
6934 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6935 | new_min_free_kbytes, user_min_free_kbytes); | |
6936 | } | |
bc75d33f | 6937 | setup_per_zone_wmarks(); |
a6cccdc3 | 6938 | refresh_zone_stat_thresholds(); |
1da177e4 | 6939 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
6940 | |
6941 | #ifdef CONFIG_NUMA | |
6942 | setup_min_unmapped_ratio(); | |
6943 | setup_min_slab_ratio(); | |
6944 | #endif | |
6945 | ||
1da177e4 LT |
6946 | return 0; |
6947 | } | |
bc22af74 | 6948 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
6949 | |
6950 | /* | |
b8af2941 | 6951 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6952 | * that we can call two helper functions whenever min_free_kbytes |
6953 | * changes. | |
6954 | */ | |
cccad5b9 | 6955 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6956 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6957 | { |
da8c757b HP |
6958 | int rc; |
6959 | ||
6960 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6961 | if (rc) | |
6962 | return rc; | |
6963 | ||
5f12733e MH |
6964 | if (write) { |
6965 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6966 | setup_per_zone_wmarks(); |
5f12733e | 6967 | } |
1da177e4 LT |
6968 | return 0; |
6969 | } | |
6970 | ||
795ae7a0 JW |
6971 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
6972 | void __user *buffer, size_t *length, loff_t *ppos) | |
6973 | { | |
6974 | int rc; | |
6975 | ||
6976 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6977 | if (rc) | |
6978 | return rc; | |
6979 | ||
6980 | if (write) | |
6981 | setup_per_zone_wmarks(); | |
6982 | ||
6983 | return 0; | |
6984 | } | |
6985 | ||
9614634f | 6986 | #ifdef CONFIG_NUMA |
6423aa81 | 6987 | static void setup_min_unmapped_ratio(void) |
9614634f | 6988 | { |
6423aa81 | 6989 | pg_data_t *pgdat; |
9614634f | 6990 | struct zone *zone; |
9614634f | 6991 | |
a5f5f91d | 6992 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 6993 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 6994 | |
9614634f | 6995 | for_each_zone(zone) |
a5f5f91d | 6996 | zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages * |
9614634f | 6997 | sysctl_min_unmapped_ratio) / 100; |
9614634f | 6998 | } |
0ff38490 | 6999 | |
6423aa81 JK |
7000 | |
7001 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 7002 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7003 | { |
0ff38490 CL |
7004 | int rc; |
7005 | ||
8d65af78 | 7006 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7007 | if (rc) |
7008 | return rc; | |
7009 | ||
6423aa81 JK |
7010 | setup_min_unmapped_ratio(); |
7011 | ||
7012 | return 0; | |
7013 | } | |
7014 | ||
7015 | static void setup_min_slab_ratio(void) | |
7016 | { | |
7017 | pg_data_t *pgdat; | |
7018 | struct zone *zone; | |
7019 | ||
a5f5f91d MG |
7020 | for_each_online_pgdat(pgdat) |
7021 | pgdat->min_slab_pages = 0; | |
7022 | ||
0ff38490 | 7023 | for_each_zone(zone) |
a5f5f91d | 7024 | zone->zone_pgdat->min_slab_pages += (zone->managed_pages * |
0ff38490 | 7025 | sysctl_min_slab_ratio) / 100; |
6423aa81 JK |
7026 | } |
7027 | ||
7028 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
7029 | void __user *buffer, size_t *length, loff_t *ppos) | |
7030 | { | |
7031 | int rc; | |
7032 | ||
7033 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7034 | if (rc) | |
7035 | return rc; | |
7036 | ||
7037 | setup_min_slab_ratio(); | |
7038 | ||
0ff38490 CL |
7039 | return 0; |
7040 | } | |
9614634f CL |
7041 | #endif |
7042 | ||
1da177e4 LT |
7043 | /* |
7044 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7045 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7046 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7047 | * | |
7048 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7049 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7050 | * if in function of the boot time zone sizes. |
7051 | */ | |
cccad5b9 | 7052 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7053 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7054 | { |
8d65af78 | 7055 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
7056 | setup_per_zone_lowmem_reserve(); |
7057 | return 0; | |
7058 | } | |
7059 | ||
8ad4b1fb RS |
7060 | /* |
7061 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
7062 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
7063 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 7064 | */ |
cccad5b9 | 7065 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7066 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
7067 | { |
7068 | struct zone *zone; | |
7cd2b0a3 | 7069 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
7070 | int ret; |
7071 | ||
7cd2b0a3 DR |
7072 | mutex_lock(&pcp_batch_high_lock); |
7073 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
7074 | ||
8d65af78 | 7075 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
7076 | if (!write || ret < 0) |
7077 | goto out; | |
7078 | ||
7079 | /* Sanity checking to avoid pcp imbalance */ | |
7080 | if (percpu_pagelist_fraction && | |
7081 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
7082 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
7083 | ret = -EINVAL; | |
7084 | goto out; | |
7085 | } | |
7086 | ||
7087 | /* No change? */ | |
7088 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
7089 | goto out; | |
c8e251fa | 7090 | |
364df0eb | 7091 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
7092 | unsigned int cpu; |
7093 | ||
22a7f12b | 7094 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
7095 | pageset_set_high_and_batch(zone, |
7096 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 7097 | } |
7cd2b0a3 | 7098 | out: |
c8e251fa | 7099 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 7100 | return ret; |
8ad4b1fb RS |
7101 | } |
7102 | ||
a9919c79 | 7103 | #ifdef CONFIG_NUMA |
f034b5d4 | 7104 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 7105 | |
1da177e4 LT |
7106 | static int __init set_hashdist(char *str) |
7107 | { | |
7108 | if (!str) | |
7109 | return 0; | |
7110 | hashdist = simple_strtoul(str, &str, 0); | |
7111 | return 1; | |
7112 | } | |
7113 | __setup("hashdist=", set_hashdist); | |
7114 | #endif | |
7115 | ||
f6f34b43 SD |
7116 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
7117 | /* | |
7118 | * Returns the number of pages that arch has reserved but | |
7119 | * is not known to alloc_large_system_hash(). | |
7120 | */ | |
7121 | static unsigned long __init arch_reserved_kernel_pages(void) | |
7122 | { | |
7123 | return 0; | |
7124 | } | |
7125 | #endif | |
7126 | ||
1da177e4 LT |
7127 | /* |
7128 | * allocate a large system hash table from bootmem | |
7129 | * - it is assumed that the hash table must contain an exact power-of-2 | |
7130 | * quantity of entries | |
7131 | * - limit is the number of hash buckets, not the total allocation size | |
7132 | */ | |
7133 | void *__init alloc_large_system_hash(const char *tablename, | |
7134 | unsigned long bucketsize, | |
7135 | unsigned long numentries, | |
7136 | int scale, | |
7137 | int flags, | |
7138 | unsigned int *_hash_shift, | |
7139 | unsigned int *_hash_mask, | |
31fe62b9 TB |
7140 | unsigned long low_limit, |
7141 | unsigned long high_limit) | |
1da177e4 | 7142 | { |
31fe62b9 | 7143 | unsigned long long max = high_limit; |
1da177e4 LT |
7144 | unsigned long log2qty, size; |
7145 | void *table = NULL; | |
7146 | ||
7147 | /* allow the kernel cmdline to have a say */ | |
7148 | if (!numentries) { | |
7149 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 7150 | numentries = nr_kernel_pages; |
f6f34b43 | 7151 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
7152 | |
7153 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
7154 | if (PAGE_SHIFT < 20) | |
7155 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
7156 | |
7157 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
7158 | if (scale > PAGE_SHIFT) | |
7159 | numentries >>= (scale - PAGE_SHIFT); | |
7160 | else | |
7161 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
7162 | |
7163 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
7164 | if (unlikely(flags & HASH_SMALL)) { |
7165 | /* Makes no sense without HASH_EARLY */ | |
7166 | WARN_ON(!(flags & HASH_EARLY)); | |
7167 | if (!(numentries >> *_hash_shift)) { | |
7168 | numentries = 1UL << *_hash_shift; | |
7169 | BUG_ON(!numentries); | |
7170 | } | |
7171 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 7172 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 7173 | } |
6e692ed3 | 7174 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
7175 | |
7176 | /* limit allocation size to 1/16 total memory by default */ | |
7177 | if (max == 0) { | |
7178 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
7179 | do_div(max, bucketsize); | |
7180 | } | |
074b8517 | 7181 | max = min(max, 0x80000000ULL); |
1da177e4 | 7182 | |
31fe62b9 TB |
7183 | if (numentries < low_limit) |
7184 | numentries = low_limit; | |
1da177e4 LT |
7185 | if (numentries > max) |
7186 | numentries = max; | |
7187 | ||
f0d1b0b3 | 7188 | log2qty = ilog2(numentries); |
1da177e4 LT |
7189 | |
7190 | do { | |
7191 | size = bucketsize << log2qty; | |
7192 | if (flags & HASH_EARLY) | |
6782832e | 7193 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
7194 | else if (hashdist) |
7195 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
7196 | else { | |
1037b83b ED |
7197 | /* |
7198 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
7199 | * some pages at the end of hash table which |
7200 | * alloc_pages_exact() automatically does | |
1037b83b | 7201 | */ |
264ef8a9 | 7202 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 7203 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
7204 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
7205 | } | |
1da177e4 LT |
7206 | } |
7207 | } while (!table && size > PAGE_SIZE && --log2qty); | |
7208 | ||
7209 | if (!table) | |
7210 | panic("Failed to allocate %s hash table\n", tablename); | |
7211 | ||
1170532b JP |
7212 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
7213 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
7214 | |
7215 | if (_hash_shift) | |
7216 | *_hash_shift = log2qty; | |
7217 | if (_hash_mask) | |
7218 | *_hash_mask = (1 << log2qty) - 1; | |
7219 | ||
7220 | return table; | |
7221 | } | |
a117e66e | 7222 | |
a5d76b54 | 7223 | /* |
80934513 MK |
7224 | * This function checks whether pageblock includes unmovable pages or not. |
7225 | * If @count is not zero, it is okay to include less @count unmovable pages | |
7226 | * | |
b8af2941 | 7227 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
7228 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
7229 | * check without lock_page also may miss some movable non-lru pages at | |
7230 | * race condition. So you can't expect this function should be exact. | |
a5d76b54 | 7231 | */ |
b023f468 WC |
7232 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
7233 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
7234 | { |
7235 | unsigned long pfn, iter, found; | |
47118af0 MN |
7236 | int mt; |
7237 | ||
49ac8255 KH |
7238 | /* |
7239 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 7240 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
7241 | */ |
7242 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 7243 | return false; |
47118af0 MN |
7244 | mt = get_pageblock_migratetype(page); |
7245 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 7246 | return false; |
49ac8255 KH |
7247 | |
7248 | pfn = page_to_pfn(page); | |
7249 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
7250 | unsigned long check = pfn + iter; | |
7251 | ||
29723fcc | 7252 | if (!pfn_valid_within(check)) |
49ac8255 | 7253 | continue; |
29723fcc | 7254 | |
49ac8255 | 7255 | page = pfn_to_page(check); |
c8721bbb NH |
7256 | |
7257 | /* | |
7258 | * Hugepages are not in LRU lists, but they're movable. | |
7259 | * We need not scan over tail pages bacause we don't | |
7260 | * handle each tail page individually in migration. | |
7261 | */ | |
7262 | if (PageHuge(page)) { | |
7263 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
7264 | continue; | |
7265 | } | |
7266 | ||
97d255c8 MK |
7267 | /* |
7268 | * We can't use page_count without pin a page | |
7269 | * because another CPU can free compound page. | |
7270 | * This check already skips compound tails of THP | |
0139aa7b | 7271 | * because their page->_refcount is zero at all time. |
97d255c8 | 7272 | */ |
fe896d18 | 7273 | if (!page_ref_count(page)) { |
49ac8255 KH |
7274 | if (PageBuddy(page)) |
7275 | iter += (1 << page_order(page)) - 1; | |
7276 | continue; | |
7277 | } | |
97d255c8 | 7278 | |
b023f468 WC |
7279 | /* |
7280 | * The HWPoisoned page may be not in buddy system, and | |
7281 | * page_count() is not 0. | |
7282 | */ | |
7283 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
7284 | continue; | |
7285 | ||
0efadf48 YX |
7286 | if (__PageMovable(page)) |
7287 | continue; | |
7288 | ||
49ac8255 KH |
7289 | if (!PageLRU(page)) |
7290 | found++; | |
7291 | /* | |
6b4f7799 JW |
7292 | * If there are RECLAIMABLE pages, we need to check |
7293 | * it. But now, memory offline itself doesn't call | |
7294 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
7295 | */ |
7296 | /* | |
7297 | * If the page is not RAM, page_count()should be 0. | |
7298 | * we don't need more check. This is an _used_ not-movable page. | |
7299 | * | |
7300 | * The problematic thing here is PG_reserved pages. PG_reserved | |
7301 | * is set to both of a memory hole page and a _used_ kernel | |
7302 | * page at boot. | |
7303 | */ | |
7304 | if (found > count) | |
80934513 | 7305 | return true; |
49ac8255 | 7306 | } |
80934513 | 7307 | return false; |
49ac8255 KH |
7308 | } |
7309 | ||
7310 | bool is_pageblock_removable_nolock(struct page *page) | |
7311 | { | |
656a0706 MH |
7312 | struct zone *zone; |
7313 | unsigned long pfn; | |
687875fb MH |
7314 | |
7315 | /* | |
7316 | * We have to be careful here because we are iterating over memory | |
7317 | * sections which are not zone aware so we might end up outside of | |
7318 | * the zone but still within the section. | |
656a0706 MH |
7319 | * We have to take care about the node as well. If the node is offline |
7320 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 7321 | */ |
656a0706 MH |
7322 | if (!node_online(page_to_nid(page))) |
7323 | return false; | |
7324 | ||
7325 | zone = page_zone(page); | |
7326 | pfn = page_to_pfn(page); | |
108bcc96 | 7327 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
7328 | return false; |
7329 | ||
b023f468 | 7330 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 7331 | } |
0c0e6195 | 7332 | |
080fe206 | 7333 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7334 | |
7335 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7336 | { | |
7337 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7338 | pageblock_nr_pages) - 1); | |
7339 | } | |
7340 | ||
7341 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7342 | { | |
7343 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7344 | pageblock_nr_pages)); | |
7345 | } | |
7346 | ||
041d3a8c | 7347 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7348 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7349 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7350 | { |
7351 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7352 | unsigned long nr_reclaimed; |
041d3a8c MN |
7353 | unsigned long pfn = start; |
7354 | unsigned int tries = 0; | |
7355 | int ret = 0; | |
7356 | ||
be49a6e1 | 7357 | migrate_prep(); |
041d3a8c | 7358 | |
bb13ffeb | 7359 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7360 | if (fatal_signal_pending(current)) { |
7361 | ret = -EINTR; | |
7362 | break; | |
7363 | } | |
7364 | ||
bb13ffeb MG |
7365 | if (list_empty(&cc->migratepages)) { |
7366 | cc->nr_migratepages = 0; | |
edc2ca61 | 7367 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7368 | if (!pfn) { |
7369 | ret = -EINTR; | |
7370 | break; | |
7371 | } | |
7372 | tries = 0; | |
7373 | } else if (++tries == 5) { | |
7374 | ret = ret < 0 ? ret : -EBUSY; | |
7375 | break; | |
7376 | } | |
7377 | ||
beb51eaa MK |
7378 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7379 | &cc->migratepages); | |
7380 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7381 | |
9c620e2b | 7382 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 7383 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 7384 | } |
2a6f5124 SP |
7385 | if (ret < 0) { |
7386 | putback_movable_pages(&cc->migratepages); | |
7387 | return ret; | |
7388 | } | |
7389 | return 0; | |
041d3a8c MN |
7390 | } |
7391 | ||
7392 | /** | |
7393 | * alloc_contig_range() -- tries to allocate given range of pages | |
7394 | * @start: start PFN to allocate | |
7395 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7396 | * @migratetype: migratetype of the underlaying pageblocks (either |
7397 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7398 | * in range must have the same migratetype and it must | |
7399 | * be either of the two. | |
ca96b625 | 7400 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
7401 | * |
7402 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
7403 | * aligned, however it's the caller's responsibility to guarantee that | |
7404 | * we are the only thread that changes migrate type of pageblocks the | |
7405 | * pages fall in. | |
7406 | * | |
7407 | * The PFN range must belong to a single zone. | |
7408 | * | |
7409 | * Returns zero on success or negative error code. On success all | |
7410 | * pages which PFN is in [start, end) are allocated for the caller and | |
7411 | * need to be freed with free_contig_range(). | |
7412 | */ | |
0815f3d8 | 7413 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 7414 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 7415 | { |
041d3a8c | 7416 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7417 | unsigned int order; |
7418 | int ret = 0; | |
041d3a8c | 7419 | |
bb13ffeb MG |
7420 | struct compact_control cc = { |
7421 | .nr_migratepages = 0, | |
7422 | .order = -1, | |
7423 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7424 | .mode = MIGRATE_SYNC, |
bb13ffeb | 7425 | .ignore_skip_hint = true, |
ca96b625 | 7426 | .gfp_mask = memalloc_noio_flags(gfp_mask), |
bb13ffeb MG |
7427 | }; |
7428 | INIT_LIST_HEAD(&cc.migratepages); | |
7429 | ||
041d3a8c MN |
7430 | /* |
7431 | * What we do here is we mark all pageblocks in range as | |
7432 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7433 | * have different sizes, and due to the way page allocator | |
7434 | * work, we align the range to biggest of the two pages so | |
7435 | * that page allocator won't try to merge buddies from | |
7436 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7437 | * other migration type. | |
7438 | * | |
7439 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7440 | * migrate the pages from an unaligned range (ie. pages that | |
7441 | * we are interested in). This will put all the pages in | |
7442 | * range back to page allocator as MIGRATE_ISOLATE. | |
7443 | * | |
7444 | * When this is done, we take the pages in range from page | |
7445 | * allocator removing them from the buddy system. This way | |
7446 | * page allocator will never consider using them. | |
7447 | * | |
7448 | * This lets us mark the pageblocks back as | |
7449 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7450 | * aligned range but not in the unaligned, original range are | |
7451 | * put back to page allocator so that buddy can use them. | |
7452 | */ | |
7453 | ||
7454 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7455 | pfn_max_align_up(end), migratetype, |
7456 | false); | |
041d3a8c | 7457 | if (ret) |
86a595f9 | 7458 | return ret; |
041d3a8c | 7459 | |
8ef5849f JK |
7460 | /* |
7461 | * In case of -EBUSY, we'd like to know which page causes problem. | |
7462 | * So, just fall through. We will check it in test_pages_isolated(). | |
7463 | */ | |
bb13ffeb | 7464 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7465 | if (ret && ret != -EBUSY) |
041d3a8c MN |
7466 | goto done; |
7467 | ||
7468 | /* | |
7469 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7470 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7471 | * more, all pages in [start, end) are free in page allocator. | |
7472 | * What we are going to do is to allocate all pages from | |
7473 | * [start, end) (that is remove them from page allocator). | |
7474 | * | |
7475 | * The only problem is that pages at the beginning and at the | |
7476 | * end of interesting range may be not aligned with pages that | |
7477 | * page allocator holds, ie. they can be part of higher order | |
7478 | * pages. Because of this, we reserve the bigger range and | |
7479 | * once this is done free the pages we are not interested in. | |
7480 | * | |
7481 | * We don't have to hold zone->lock here because the pages are | |
7482 | * isolated thus they won't get removed from buddy. | |
7483 | */ | |
7484 | ||
7485 | lru_add_drain_all(); | |
510f5507 | 7486 | drain_all_pages(cc.zone); |
041d3a8c MN |
7487 | |
7488 | order = 0; | |
7489 | outer_start = start; | |
7490 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7491 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7492 | outer_start = start; |
7493 | break; | |
041d3a8c MN |
7494 | } |
7495 | outer_start &= ~0UL << order; | |
7496 | } | |
7497 | ||
8ef5849f JK |
7498 | if (outer_start != start) { |
7499 | order = page_order(pfn_to_page(outer_start)); | |
7500 | ||
7501 | /* | |
7502 | * outer_start page could be small order buddy page and | |
7503 | * it doesn't include start page. Adjust outer_start | |
7504 | * in this case to report failed page properly | |
7505 | * on tracepoint in test_pages_isolated() | |
7506 | */ | |
7507 | if (outer_start + (1UL << order) <= start) | |
7508 | outer_start = start; | |
7509 | } | |
7510 | ||
041d3a8c | 7511 | /* Make sure the range is really isolated. */ |
b023f468 | 7512 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
7513 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
7514 | __func__, outer_start, end); | |
041d3a8c MN |
7515 | ret = -EBUSY; |
7516 | goto done; | |
7517 | } | |
7518 | ||
49f223a9 | 7519 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7520 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7521 | if (!outer_end) { |
7522 | ret = -EBUSY; | |
7523 | goto done; | |
7524 | } | |
7525 | ||
7526 | /* Free head and tail (if any) */ | |
7527 | if (start != outer_start) | |
7528 | free_contig_range(outer_start, start - outer_start); | |
7529 | if (end != outer_end) | |
7530 | free_contig_range(end, outer_end - end); | |
7531 | ||
7532 | done: | |
7533 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7534 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7535 | return ret; |
7536 | } | |
7537 | ||
7538 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7539 | { | |
bcc2b02f MS |
7540 | unsigned int count = 0; |
7541 | ||
7542 | for (; nr_pages--; pfn++) { | |
7543 | struct page *page = pfn_to_page(pfn); | |
7544 | ||
7545 | count += page_count(page) != 1; | |
7546 | __free_page(page); | |
7547 | } | |
7548 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7549 | } |
7550 | #endif | |
7551 | ||
4ed7e022 | 7552 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7553 | /* |
7554 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7555 | * page high values need to be recalulated. | |
7556 | */ | |
4ed7e022 JL |
7557 | void __meminit zone_pcp_update(struct zone *zone) |
7558 | { | |
0a647f38 | 7559 | unsigned cpu; |
c8e251fa | 7560 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7561 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7562 | pageset_set_high_and_batch(zone, |
7563 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7564 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7565 | } |
7566 | #endif | |
7567 | ||
340175b7 JL |
7568 | void zone_pcp_reset(struct zone *zone) |
7569 | { | |
7570 | unsigned long flags; | |
5a883813 MK |
7571 | int cpu; |
7572 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7573 | |
7574 | /* avoid races with drain_pages() */ | |
7575 | local_irq_save(flags); | |
7576 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7577 | for_each_online_cpu(cpu) { |
7578 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7579 | drain_zonestat(zone, pset); | |
7580 | } | |
340175b7 JL |
7581 | free_percpu(zone->pageset); |
7582 | zone->pageset = &boot_pageset; | |
7583 | } | |
7584 | local_irq_restore(flags); | |
7585 | } | |
7586 | ||
6dcd73d7 | 7587 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 7588 | /* |
b9eb6319 JK |
7589 | * All pages in the range must be in a single zone and isolated |
7590 | * before calling this. | |
0c0e6195 KH |
7591 | */ |
7592 | void | |
7593 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7594 | { | |
7595 | struct page *page; | |
7596 | struct zone *zone; | |
7aeb09f9 | 7597 | unsigned int order, i; |
0c0e6195 KH |
7598 | unsigned long pfn; |
7599 | unsigned long flags; | |
7600 | /* find the first valid pfn */ | |
7601 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
7602 | if (pfn_valid(pfn)) | |
7603 | break; | |
7604 | if (pfn == end_pfn) | |
7605 | return; | |
7606 | zone = page_zone(pfn_to_page(pfn)); | |
7607 | spin_lock_irqsave(&zone->lock, flags); | |
7608 | pfn = start_pfn; | |
7609 | while (pfn < end_pfn) { | |
7610 | if (!pfn_valid(pfn)) { | |
7611 | pfn++; | |
7612 | continue; | |
7613 | } | |
7614 | page = pfn_to_page(pfn); | |
b023f468 WC |
7615 | /* |
7616 | * The HWPoisoned page may be not in buddy system, and | |
7617 | * page_count() is not 0. | |
7618 | */ | |
7619 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
7620 | pfn++; | |
7621 | SetPageReserved(page); | |
7622 | continue; | |
7623 | } | |
7624 | ||
0c0e6195 KH |
7625 | BUG_ON(page_count(page)); |
7626 | BUG_ON(!PageBuddy(page)); | |
7627 | order = page_order(page); | |
7628 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
7629 | pr_info("remove from free list %lx %d %lx\n", |
7630 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
7631 | #endif |
7632 | list_del(&page->lru); | |
7633 | rmv_page_order(page); | |
7634 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
7635 | for (i = 0; i < (1 << order); i++) |
7636 | SetPageReserved((page+i)); | |
7637 | pfn += (1 << order); | |
7638 | } | |
7639 | spin_unlock_irqrestore(&zone->lock, flags); | |
7640 | } | |
7641 | #endif | |
8d22ba1b | 7642 | |
8d22ba1b WF |
7643 | bool is_free_buddy_page(struct page *page) |
7644 | { | |
7645 | struct zone *zone = page_zone(page); | |
7646 | unsigned long pfn = page_to_pfn(page); | |
7647 | unsigned long flags; | |
7aeb09f9 | 7648 | unsigned int order; |
8d22ba1b WF |
7649 | |
7650 | spin_lock_irqsave(&zone->lock, flags); | |
7651 | for (order = 0; order < MAX_ORDER; order++) { | |
7652 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
7653 | ||
7654 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
7655 | break; | |
7656 | } | |
7657 | spin_unlock_irqrestore(&zone->lock, flags); | |
7658 | ||
7659 | return order < MAX_ORDER; | |
7660 | } |