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> |
268bb0ce | 58 | #include <linux/prefetch.h> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
041d3a8c | 60 | #include <linux/migrate.h> |
e30825f1 | 61 | #include <linux/page_ext.h> |
949f7ec5 | 62 | #include <linux/hugetlb.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
48c96a36 | 64 | #include <linux/page_owner.h> |
0e1cc95b | 65 | #include <linux/kthread.h> |
4949148a | 66 | #include <linux/memcontrol.h> |
1da177e4 | 67 | |
7ee3d4e8 | 68 | #include <asm/sections.h> |
1da177e4 | 69 | #include <asm/tlbflush.h> |
ac924c60 | 70 | #include <asm/div64.h> |
1da177e4 LT |
71 | #include "internal.h" |
72 | ||
c8e251fa CS |
73 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
74 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 75 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 76 | |
72812019 LS |
77 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
78 | DEFINE_PER_CPU(int, numa_node); | |
79 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
80 | #endif | |
81 | ||
7aac7898 LS |
82 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
83 | /* | |
84 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
85 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
86 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
87 | * defined in <linux/topology.h>. | |
88 | */ | |
89 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
90 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 91 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
92 | #endif |
93 | ||
38addce8 | 94 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 95 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
96 | EXPORT_SYMBOL(latent_entropy); |
97 | #endif | |
98 | ||
1da177e4 | 99 | /* |
13808910 | 100 | * Array of node states. |
1da177e4 | 101 | */ |
13808910 CL |
102 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
103 | [N_POSSIBLE] = NODE_MASK_ALL, | |
104 | [N_ONLINE] = { { [0] = 1UL } }, | |
105 | #ifndef CONFIG_NUMA | |
106 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
107 | #ifdef CONFIG_HIGHMEM | |
108 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
109 | #endif |
110 | #ifdef CONFIG_MOVABLE_NODE | |
111 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
112 | #endif |
113 | [N_CPU] = { { [0] = 1UL } }, | |
114 | #endif /* NUMA */ | |
115 | }; | |
116 | EXPORT_SYMBOL(node_states); | |
117 | ||
c3d5f5f0 JL |
118 | /* Protect totalram_pages and zone->managed_pages */ |
119 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
120 | ||
6c231b7b | 121 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 122 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 123 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 124 | |
1b76b02f | 125 | int percpu_pagelist_fraction; |
dcce284a | 126 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 127 | |
bb14c2c7 VB |
128 | /* |
129 | * A cached value of the page's pageblock's migratetype, used when the page is | |
130 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
131 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
132 | * Also the migratetype set in the page does not necessarily match the pcplist | |
133 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
134 | * other index - this ensures that it will be put on the correct CMA freelist. | |
135 | */ | |
136 | static inline int get_pcppage_migratetype(struct page *page) | |
137 | { | |
138 | return page->index; | |
139 | } | |
140 | ||
141 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
142 | { | |
143 | page->index = migratetype; | |
144 | } | |
145 | ||
452aa699 RW |
146 | #ifdef CONFIG_PM_SLEEP |
147 | /* | |
148 | * The following functions are used by the suspend/hibernate code to temporarily | |
149 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
150 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
151 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
152 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
153 | * guaranteed not to run in parallel with that modification). | |
154 | */ | |
c9e664f1 RW |
155 | |
156 | static gfp_t saved_gfp_mask; | |
157 | ||
158 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
159 | { |
160 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
161 | if (saved_gfp_mask) { |
162 | gfp_allowed_mask = saved_gfp_mask; | |
163 | saved_gfp_mask = 0; | |
164 | } | |
452aa699 RW |
165 | } |
166 | ||
c9e664f1 | 167 | void pm_restrict_gfp_mask(void) |
452aa699 | 168 | { |
452aa699 | 169 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
170 | WARN_ON(saved_gfp_mask); |
171 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 172 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 173 | } |
f90ac398 MG |
174 | |
175 | bool pm_suspended_storage(void) | |
176 | { | |
d0164adc | 177 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
178 | return false; |
179 | return true; | |
180 | } | |
452aa699 RW |
181 | #endif /* CONFIG_PM_SLEEP */ |
182 | ||
d9c23400 | 183 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 184 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
185 | #endif |
186 | ||
d98c7a09 | 187 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 188 | |
1da177e4 LT |
189 | /* |
190 | * results with 256, 32 in the lowmem_reserve sysctl: | |
191 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
192 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
193 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
194 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 195 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
196 | * |
197 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
198 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 199 | */ |
2f1b6248 | 200 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 201 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 202 | 256, |
4b51d669 | 203 | #endif |
fb0e7942 | 204 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 205 | 256, |
fb0e7942 | 206 | #endif |
e53ef38d | 207 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 208 | 32, |
e53ef38d | 209 | #endif |
2a1e274a | 210 | 32, |
2f1b6248 | 211 | }; |
1da177e4 LT |
212 | |
213 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 214 | |
15ad7cdc | 215 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 216 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 217 | "DMA", |
4b51d669 | 218 | #endif |
fb0e7942 | 219 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 220 | "DMA32", |
fb0e7942 | 221 | #endif |
2f1b6248 | 222 | "Normal", |
e53ef38d | 223 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 224 | "HighMem", |
e53ef38d | 225 | #endif |
2a1e274a | 226 | "Movable", |
033fbae9 DW |
227 | #ifdef CONFIG_ZONE_DEVICE |
228 | "Device", | |
229 | #endif | |
2f1b6248 CL |
230 | }; |
231 | ||
60f30350 VB |
232 | char * const migratetype_names[MIGRATE_TYPES] = { |
233 | "Unmovable", | |
234 | "Movable", | |
235 | "Reclaimable", | |
236 | "HighAtomic", | |
237 | #ifdef CONFIG_CMA | |
238 | "CMA", | |
239 | #endif | |
240 | #ifdef CONFIG_MEMORY_ISOLATION | |
241 | "Isolate", | |
242 | #endif | |
243 | }; | |
244 | ||
f1e61557 KS |
245 | compound_page_dtor * const compound_page_dtors[] = { |
246 | NULL, | |
247 | free_compound_page, | |
248 | #ifdef CONFIG_HUGETLB_PAGE | |
249 | free_huge_page, | |
250 | #endif | |
9a982250 KS |
251 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
252 | free_transhuge_page, | |
253 | #endif | |
f1e61557 KS |
254 | }; |
255 | ||
1da177e4 | 256 | int min_free_kbytes = 1024; |
42aa83cb | 257 | int user_min_free_kbytes = -1; |
795ae7a0 | 258 | int watermark_scale_factor = 10; |
1da177e4 | 259 | |
2c85f51d JB |
260 | static unsigned long __meminitdata nr_kernel_pages; |
261 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 262 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 263 | |
0ee332c1 TH |
264 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
265 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
266 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
267 | static unsigned long __initdata required_kernelcore; | |
268 | static unsigned long __initdata required_movablecore; | |
269 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
342332e6 | 270 | static bool mirrored_kernelcore; |
0ee332c1 TH |
271 | |
272 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
273 | int movable_zone; | |
274 | EXPORT_SYMBOL(movable_zone); | |
275 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 276 | |
418508c1 MS |
277 | #if MAX_NUMNODES > 1 |
278 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 279 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 280 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 281 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
282 | #endif |
283 | ||
9ef9acb0 MG |
284 | int page_group_by_mobility_disabled __read_mostly; |
285 | ||
3a80a7fa MG |
286 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
287 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
288 | { | |
289 | pgdat->first_deferred_pfn = ULONG_MAX; | |
290 | } | |
291 | ||
292 | /* Returns true if the struct page for the pfn is uninitialised */ | |
0e1cc95b | 293 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 294 | { |
ef70b6f4 MG |
295 | int nid = early_pfn_to_nid(pfn); |
296 | ||
297 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
298 | return true; |
299 | ||
300 | return false; | |
301 | } | |
302 | ||
303 | /* | |
304 | * Returns false when the remaining initialisation should be deferred until | |
305 | * later in the boot cycle when it can be parallelised. | |
306 | */ | |
307 | static inline bool update_defer_init(pg_data_t *pgdat, | |
308 | unsigned long pfn, unsigned long zone_end, | |
309 | unsigned long *nr_initialised) | |
310 | { | |
987b3095 LZ |
311 | unsigned long max_initialise; |
312 | ||
3a80a7fa MG |
313 | /* Always populate low zones for address-contrained allocations */ |
314 | if (zone_end < pgdat_end_pfn(pgdat)) | |
315 | return true; | |
987b3095 LZ |
316 | /* |
317 | * Initialise at least 2G of a node but also take into account that | |
318 | * two large system hashes that can take up 1GB for 0.25TB/node. | |
319 | */ | |
320 | max_initialise = max(2UL << (30 - PAGE_SHIFT), | |
321 | (pgdat->node_spanned_pages >> 8)); | |
3a80a7fa | 322 | |
3a80a7fa | 323 | (*nr_initialised)++; |
987b3095 | 324 | if ((*nr_initialised > max_initialise) && |
3a80a7fa MG |
325 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
326 | pgdat->first_deferred_pfn = pfn; | |
327 | return false; | |
328 | } | |
329 | ||
330 | return true; | |
331 | } | |
332 | #else | |
333 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
334 | { | |
335 | } | |
336 | ||
337 | static inline bool early_page_uninitialised(unsigned long pfn) | |
338 | { | |
339 | return false; | |
340 | } | |
341 | ||
342 | static inline bool update_defer_init(pg_data_t *pgdat, | |
343 | unsigned long pfn, unsigned long zone_end, | |
344 | unsigned long *nr_initialised) | |
345 | { | |
346 | return true; | |
347 | } | |
348 | #endif | |
349 | ||
0b423ca2 MG |
350 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
351 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
352 | unsigned long pfn) | |
353 | { | |
354 | #ifdef CONFIG_SPARSEMEM | |
355 | return __pfn_to_section(pfn)->pageblock_flags; | |
356 | #else | |
357 | return page_zone(page)->pageblock_flags; | |
358 | #endif /* CONFIG_SPARSEMEM */ | |
359 | } | |
360 | ||
361 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
362 | { | |
363 | #ifdef CONFIG_SPARSEMEM | |
364 | pfn &= (PAGES_PER_SECTION-1); | |
365 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
366 | #else | |
367 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
368 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
369 | #endif /* CONFIG_SPARSEMEM */ | |
370 | } | |
371 | ||
372 | /** | |
373 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
374 | * @page: The page within the block of interest | |
375 | * @pfn: The target page frame number | |
376 | * @end_bitidx: The last bit of interest to retrieve | |
377 | * @mask: mask of bits that the caller is interested in | |
378 | * | |
379 | * Return: pageblock_bits flags | |
380 | */ | |
381 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
382 | unsigned long pfn, | |
383 | unsigned long end_bitidx, | |
384 | unsigned long mask) | |
385 | { | |
386 | unsigned long *bitmap; | |
387 | unsigned long bitidx, word_bitidx; | |
388 | unsigned long word; | |
389 | ||
390 | bitmap = get_pageblock_bitmap(page, pfn); | |
391 | bitidx = pfn_to_bitidx(page, pfn); | |
392 | word_bitidx = bitidx / BITS_PER_LONG; | |
393 | bitidx &= (BITS_PER_LONG-1); | |
394 | ||
395 | word = bitmap[word_bitidx]; | |
396 | bitidx += end_bitidx; | |
397 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
398 | } | |
399 | ||
400 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
401 | unsigned long end_bitidx, | |
402 | unsigned long mask) | |
403 | { | |
404 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
405 | } | |
406 | ||
407 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
408 | { | |
409 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
410 | } | |
411 | ||
412 | /** | |
413 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
414 | * @page: The page within the block of interest | |
415 | * @flags: The flags to set | |
416 | * @pfn: The target page frame number | |
417 | * @end_bitidx: The last bit of interest | |
418 | * @mask: mask of bits that the caller is interested in | |
419 | */ | |
420 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
421 | unsigned long pfn, | |
422 | unsigned long end_bitidx, | |
423 | unsigned long mask) | |
424 | { | |
425 | unsigned long *bitmap; | |
426 | unsigned long bitidx, word_bitidx; | |
427 | unsigned long old_word, word; | |
428 | ||
429 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
430 | ||
431 | bitmap = get_pageblock_bitmap(page, pfn); | |
432 | bitidx = pfn_to_bitidx(page, pfn); | |
433 | word_bitidx = bitidx / BITS_PER_LONG; | |
434 | bitidx &= (BITS_PER_LONG-1); | |
435 | ||
436 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
437 | ||
438 | bitidx += end_bitidx; | |
439 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
440 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
441 | ||
442 | word = READ_ONCE(bitmap[word_bitidx]); | |
443 | for (;;) { | |
444 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
445 | if (word == old_word) | |
446 | break; | |
447 | word = old_word; | |
448 | } | |
449 | } | |
3a80a7fa | 450 | |
ee6f509c | 451 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 452 | { |
5d0f3f72 KM |
453 | if (unlikely(page_group_by_mobility_disabled && |
454 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
455 | migratetype = MIGRATE_UNMOVABLE; |
456 | ||
b2a0ac88 MG |
457 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
458 | PB_migrate, PB_migrate_end); | |
459 | } | |
460 | ||
13e7444b | 461 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 462 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 463 | { |
bdc8cb98 DH |
464 | int ret = 0; |
465 | unsigned seq; | |
466 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 467 | unsigned long sp, start_pfn; |
c6a57e19 | 468 | |
bdc8cb98 DH |
469 | do { |
470 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
471 | start_pfn = zone->zone_start_pfn; |
472 | sp = zone->spanned_pages; | |
108bcc96 | 473 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
474 | ret = 1; |
475 | } while (zone_span_seqretry(zone, seq)); | |
476 | ||
b5e6a5a2 | 477 | if (ret) |
613813e8 DH |
478 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
479 | pfn, zone_to_nid(zone), zone->name, | |
480 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 481 | |
bdc8cb98 | 482 | return ret; |
c6a57e19 DH |
483 | } |
484 | ||
485 | static int page_is_consistent(struct zone *zone, struct page *page) | |
486 | { | |
14e07298 | 487 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 488 | return 0; |
1da177e4 | 489 | if (zone != page_zone(page)) |
c6a57e19 DH |
490 | return 0; |
491 | ||
492 | return 1; | |
493 | } | |
494 | /* | |
495 | * Temporary debugging check for pages not lying within a given zone. | |
496 | */ | |
497 | static int bad_range(struct zone *zone, struct page *page) | |
498 | { | |
499 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 500 | return 1; |
c6a57e19 DH |
501 | if (!page_is_consistent(zone, page)) |
502 | return 1; | |
503 | ||
1da177e4 LT |
504 | return 0; |
505 | } | |
13e7444b NP |
506 | #else |
507 | static inline int bad_range(struct zone *zone, struct page *page) | |
508 | { | |
509 | return 0; | |
510 | } | |
511 | #endif | |
512 | ||
d230dec1 KS |
513 | static void bad_page(struct page *page, const char *reason, |
514 | unsigned long bad_flags) | |
1da177e4 | 515 | { |
d936cf9b HD |
516 | static unsigned long resume; |
517 | static unsigned long nr_shown; | |
518 | static unsigned long nr_unshown; | |
519 | ||
520 | /* | |
521 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
522 | * or allow a steady drip of one report per second. | |
523 | */ | |
524 | if (nr_shown == 60) { | |
525 | if (time_before(jiffies, resume)) { | |
526 | nr_unshown++; | |
527 | goto out; | |
528 | } | |
529 | if (nr_unshown) { | |
ff8e8116 | 530 | pr_alert( |
1e9e6365 | 531 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
532 | nr_unshown); |
533 | nr_unshown = 0; | |
534 | } | |
535 | nr_shown = 0; | |
536 | } | |
537 | if (nr_shown++ == 0) | |
538 | resume = jiffies + 60 * HZ; | |
539 | ||
ff8e8116 | 540 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 541 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
542 | __dump_page(page, reason); |
543 | bad_flags &= page->flags; | |
544 | if (bad_flags) | |
545 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
546 | bad_flags, &bad_flags); | |
4e462112 | 547 | dump_page_owner(page); |
3dc14741 | 548 | |
4f31888c | 549 | print_modules(); |
1da177e4 | 550 | dump_stack(); |
d936cf9b | 551 | out: |
8cc3b392 | 552 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 553 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 554 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
555 | } |
556 | ||
1da177e4 LT |
557 | /* |
558 | * Higher-order pages are called "compound pages". They are structured thusly: | |
559 | * | |
1d798ca3 | 560 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 561 | * |
1d798ca3 KS |
562 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
563 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 564 | * |
1d798ca3 KS |
565 | * The first tail page's ->compound_dtor holds the offset in array of compound |
566 | * page destructors. See compound_page_dtors. | |
1da177e4 | 567 | * |
1d798ca3 | 568 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 569 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 570 | */ |
d98c7a09 | 571 | |
9a982250 | 572 | void free_compound_page(struct page *page) |
d98c7a09 | 573 | { |
d85f3385 | 574 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
575 | } |
576 | ||
d00181b9 | 577 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
578 | { |
579 | int i; | |
580 | int nr_pages = 1 << order; | |
581 | ||
f1e61557 | 582 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
583 | set_compound_order(page, order); |
584 | __SetPageHead(page); | |
585 | for (i = 1; i < nr_pages; i++) { | |
586 | struct page *p = page + i; | |
58a84aa9 | 587 | set_page_count(p, 0); |
1c290f64 | 588 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 589 | set_compound_head(p, page); |
18229df5 | 590 | } |
53f9263b | 591 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
592 | } |
593 | ||
c0a32fc5 SG |
594 | #ifdef CONFIG_DEBUG_PAGEALLOC |
595 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
596 | bool _debug_pagealloc_enabled __read_mostly |
597 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 598 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
599 | bool _debug_guardpage_enabled __read_mostly; |
600 | ||
031bc574 JK |
601 | static int __init early_debug_pagealloc(char *buf) |
602 | { | |
603 | if (!buf) | |
604 | return -EINVAL; | |
2a138dc7 | 605 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
031bc574 JK |
606 | } |
607 | early_param("debug_pagealloc", early_debug_pagealloc); | |
608 | ||
e30825f1 JK |
609 | static bool need_debug_guardpage(void) |
610 | { | |
031bc574 JK |
611 | /* If we don't use debug_pagealloc, we don't need guard page */ |
612 | if (!debug_pagealloc_enabled()) | |
613 | return false; | |
614 | ||
f1c1e9f7 JK |
615 | if (!debug_guardpage_minorder()) |
616 | return false; | |
617 | ||
e30825f1 JK |
618 | return true; |
619 | } | |
620 | ||
621 | static void init_debug_guardpage(void) | |
622 | { | |
031bc574 JK |
623 | if (!debug_pagealloc_enabled()) |
624 | return; | |
625 | ||
f1c1e9f7 JK |
626 | if (!debug_guardpage_minorder()) |
627 | return; | |
628 | ||
e30825f1 JK |
629 | _debug_guardpage_enabled = true; |
630 | } | |
631 | ||
632 | struct page_ext_operations debug_guardpage_ops = { | |
633 | .need = need_debug_guardpage, | |
634 | .init = init_debug_guardpage, | |
635 | }; | |
c0a32fc5 SG |
636 | |
637 | static int __init debug_guardpage_minorder_setup(char *buf) | |
638 | { | |
639 | unsigned long res; | |
640 | ||
641 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 642 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
643 | return 0; |
644 | } | |
645 | _debug_guardpage_minorder = res; | |
1170532b | 646 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
647 | return 0; |
648 | } | |
f1c1e9f7 | 649 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 650 | |
acbc15a4 | 651 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 652 | unsigned int order, int migratetype) |
c0a32fc5 | 653 | { |
e30825f1 JK |
654 | struct page_ext *page_ext; |
655 | ||
656 | if (!debug_guardpage_enabled()) | |
acbc15a4 JK |
657 | return false; |
658 | ||
659 | if (order >= debug_guardpage_minorder()) | |
660 | return false; | |
e30825f1 JK |
661 | |
662 | page_ext = lookup_page_ext(page); | |
f86e4271 | 663 | if (unlikely(!page_ext)) |
acbc15a4 | 664 | return false; |
f86e4271 | 665 | |
e30825f1 JK |
666 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
667 | ||
2847cf95 JK |
668 | INIT_LIST_HEAD(&page->lru); |
669 | set_page_private(page, order); | |
670 | /* Guard pages are not available for any usage */ | |
671 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
672 | |
673 | return true; | |
c0a32fc5 SG |
674 | } |
675 | ||
2847cf95 JK |
676 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
677 | unsigned int order, int migratetype) | |
c0a32fc5 | 678 | { |
e30825f1 JK |
679 | struct page_ext *page_ext; |
680 | ||
681 | if (!debug_guardpage_enabled()) | |
682 | return; | |
683 | ||
684 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
685 | if (unlikely(!page_ext)) |
686 | return; | |
687 | ||
e30825f1 JK |
688 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
689 | ||
2847cf95 JK |
690 | set_page_private(page, 0); |
691 | if (!is_migrate_isolate(migratetype)) | |
692 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
693 | } |
694 | #else | |
980ac167 | 695 | struct page_ext_operations debug_guardpage_ops; |
acbc15a4 JK |
696 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
697 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
698 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
699 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
700 | #endif |
701 | ||
7aeb09f9 | 702 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 703 | { |
4c21e2f2 | 704 | set_page_private(page, order); |
676165a8 | 705 | __SetPageBuddy(page); |
1da177e4 LT |
706 | } |
707 | ||
708 | static inline void rmv_page_order(struct page *page) | |
709 | { | |
676165a8 | 710 | __ClearPageBuddy(page); |
4c21e2f2 | 711 | set_page_private(page, 0); |
1da177e4 LT |
712 | } |
713 | ||
1da177e4 LT |
714 | /* |
715 | * This function checks whether a page is free && is the buddy | |
716 | * we can do coalesce a page and its buddy if | |
13e7444b | 717 | * (a) the buddy is not in a hole && |
676165a8 | 718 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
719 | * (c) a page and its buddy have the same order && |
720 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 721 | * |
cf6fe945 WSH |
722 | * For recording whether a page is in the buddy system, we set ->_mapcount |
723 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
724 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
725 | * serialized by zone->lock. | |
1da177e4 | 726 | * |
676165a8 | 727 | * For recording page's order, we use page_private(page). |
1da177e4 | 728 | */ |
cb2b95e1 | 729 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 730 | unsigned int order) |
1da177e4 | 731 | { |
14e07298 | 732 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 733 | return 0; |
13e7444b | 734 | |
c0a32fc5 | 735 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
736 | if (page_zone_id(page) != page_zone_id(buddy)) |
737 | return 0; | |
738 | ||
4c5018ce WY |
739 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
740 | ||
c0a32fc5 SG |
741 | return 1; |
742 | } | |
743 | ||
cb2b95e1 | 744 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
745 | /* |
746 | * zone check is done late to avoid uselessly | |
747 | * calculating zone/node ids for pages that could | |
748 | * never merge. | |
749 | */ | |
750 | if (page_zone_id(page) != page_zone_id(buddy)) | |
751 | return 0; | |
752 | ||
4c5018ce WY |
753 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
754 | ||
6aa3001b | 755 | return 1; |
676165a8 | 756 | } |
6aa3001b | 757 | return 0; |
1da177e4 LT |
758 | } |
759 | ||
760 | /* | |
761 | * Freeing function for a buddy system allocator. | |
762 | * | |
763 | * The concept of a buddy system is to maintain direct-mapped table | |
764 | * (containing bit values) for memory blocks of various "orders". | |
765 | * The bottom level table contains the map for the smallest allocatable | |
766 | * units of memory (here, pages), and each level above it describes | |
767 | * pairs of units from the levels below, hence, "buddies". | |
768 | * At a high level, all that happens here is marking the table entry | |
769 | * at the bottom level available, and propagating the changes upward | |
770 | * as necessary, plus some accounting needed to play nicely with other | |
771 | * parts of the VM system. | |
772 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
773 | * free pages of length of (1 << order) and marked with _mapcount |
774 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
775 | * field. | |
1da177e4 | 776 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
777 | * other. That is, if we allocate a small block, and both were |
778 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 779 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 780 | * triggers coalescing into a block of larger size. |
1da177e4 | 781 | * |
6d49e352 | 782 | * -- nyc |
1da177e4 LT |
783 | */ |
784 | ||
48db57f8 | 785 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 786 | unsigned long pfn, |
ed0ae21d MG |
787 | struct zone *zone, unsigned int order, |
788 | int migratetype) | |
1da177e4 LT |
789 | { |
790 | unsigned long page_idx; | |
6dda9d55 | 791 | unsigned long combined_idx; |
43506fad | 792 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 793 | struct page *buddy; |
d9dddbf5 VB |
794 | unsigned int max_order; |
795 | ||
796 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 797 | |
d29bb978 | 798 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 799 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 800 | |
ed0ae21d | 801 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 802 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 803 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 804 | |
d9dddbf5 | 805 | page_idx = pfn & ((1 << MAX_ORDER) - 1); |
1da177e4 | 806 | |
309381fe SL |
807 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
808 | VM_BUG_ON_PAGE(bad_range(zone, page), page); | |
1da177e4 | 809 | |
d9dddbf5 | 810 | continue_merging: |
3c605096 | 811 | while (order < max_order - 1) { |
43506fad KC |
812 | buddy_idx = __find_buddy_index(page_idx, order); |
813 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 814 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 815 | goto done_merging; |
c0a32fc5 SG |
816 | /* |
817 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
818 | * merge with it and move up one order. | |
819 | */ | |
820 | if (page_is_guard(buddy)) { | |
2847cf95 | 821 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
822 | } else { |
823 | list_del(&buddy->lru); | |
824 | zone->free_area[order].nr_free--; | |
825 | rmv_page_order(buddy); | |
826 | } | |
43506fad | 827 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
828 | page = page + (combined_idx - page_idx); |
829 | page_idx = combined_idx; | |
830 | order++; | |
831 | } | |
d9dddbf5 VB |
832 | if (max_order < MAX_ORDER) { |
833 | /* If we are here, it means order is >= pageblock_order. | |
834 | * We want to prevent merge between freepages on isolate | |
835 | * pageblock and normal pageblock. Without this, pageblock | |
836 | * isolation could cause incorrect freepage or CMA accounting. | |
837 | * | |
838 | * We don't want to hit this code for the more frequent | |
839 | * low-order merging. | |
840 | */ | |
841 | if (unlikely(has_isolate_pageblock(zone))) { | |
842 | int buddy_mt; | |
843 | ||
844 | buddy_idx = __find_buddy_index(page_idx, order); | |
845 | buddy = page + (buddy_idx - page_idx); | |
846 | buddy_mt = get_pageblock_migratetype(buddy); | |
847 | ||
848 | if (migratetype != buddy_mt | |
849 | && (is_migrate_isolate(migratetype) || | |
850 | is_migrate_isolate(buddy_mt))) | |
851 | goto done_merging; | |
852 | } | |
853 | max_order++; | |
854 | goto continue_merging; | |
855 | } | |
856 | ||
857 | done_merging: | |
1da177e4 | 858 | set_page_order(page, order); |
6dda9d55 CZ |
859 | |
860 | /* | |
861 | * If this is not the largest possible page, check if the buddy | |
862 | * of the next-highest order is free. If it is, it's possible | |
863 | * that pages are being freed that will coalesce soon. In case, | |
864 | * that is happening, add the free page to the tail of the list | |
865 | * so it's less likely to be used soon and more likely to be merged | |
866 | * as a higher order page | |
867 | */ | |
b7f50cfa | 868 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 869 | struct page *higher_page, *higher_buddy; |
43506fad KC |
870 | combined_idx = buddy_idx & page_idx; |
871 | higher_page = page + (combined_idx - page_idx); | |
872 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 873 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
874 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
875 | list_add_tail(&page->lru, | |
876 | &zone->free_area[order].free_list[migratetype]); | |
877 | goto out; | |
878 | } | |
879 | } | |
880 | ||
881 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
882 | out: | |
1da177e4 LT |
883 | zone->free_area[order].nr_free++; |
884 | } | |
885 | ||
7bfec6f4 MG |
886 | /* |
887 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
888 | * try and check multiple fields with one check. The caller must do a detailed | |
889 | * check if necessary. | |
890 | */ | |
891 | static inline bool page_expected_state(struct page *page, | |
892 | unsigned long check_flags) | |
893 | { | |
894 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
895 | return false; | |
896 | ||
897 | if (unlikely((unsigned long)page->mapping | | |
898 | page_ref_count(page) | | |
899 | #ifdef CONFIG_MEMCG | |
900 | (unsigned long)page->mem_cgroup | | |
901 | #endif | |
902 | (page->flags & check_flags))) | |
903 | return false; | |
904 | ||
905 | return true; | |
906 | } | |
907 | ||
bb552ac6 | 908 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 909 | { |
7bfec6f4 MG |
910 | const char *bad_reason; |
911 | unsigned long bad_flags; | |
912 | ||
7bfec6f4 MG |
913 | bad_reason = NULL; |
914 | bad_flags = 0; | |
f0b791a3 | 915 | |
53f9263b | 916 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
917 | bad_reason = "nonzero mapcount"; |
918 | if (unlikely(page->mapping != NULL)) | |
919 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 920 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 921 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
922 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
923 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
924 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
925 | } | |
9edad6ea JW |
926 | #ifdef CONFIG_MEMCG |
927 | if (unlikely(page->mem_cgroup)) | |
928 | bad_reason = "page still charged to cgroup"; | |
929 | #endif | |
7bfec6f4 | 930 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
931 | } |
932 | ||
933 | static inline int free_pages_check(struct page *page) | |
934 | { | |
da838d4f | 935 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 936 | return 0; |
bb552ac6 MG |
937 | |
938 | /* Something has gone sideways, find it */ | |
939 | free_pages_check_bad(page); | |
7bfec6f4 | 940 | return 1; |
1da177e4 LT |
941 | } |
942 | ||
4db7548c MG |
943 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
944 | { | |
945 | int ret = 1; | |
946 | ||
947 | /* | |
948 | * We rely page->lru.next never has bit 0 set, unless the page | |
949 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
950 | */ | |
951 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
952 | ||
953 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
954 | ret = 0; | |
955 | goto out; | |
956 | } | |
957 | switch (page - head_page) { | |
958 | case 1: | |
959 | /* the first tail page: ->mapping is compound_mapcount() */ | |
960 | if (unlikely(compound_mapcount(page))) { | |
961 | bad_page(page, "nonzero compound_mapcount", 0); | |
962 | goto out; | |
963 | } | |
964 | break; | |
965 | case 2: | |
966 | /* | |
967 | * the second tail page: ->mapping is | |
968 | * page_deferred_list().next -- ignore value. | |
969 | */ | |
970 | break; | |
971 | default: | |
972 | if (page->mapping != TAIL_MAPPING) { | |
973 | bad_page(page, "corrupted mapping in tail page", 0); | |
974 | goto out; | |
975 | } | |
976 | break; | |
977 | } | |
978 | if (unlikely(!PageTail(page))) { | |
979 | bad_page(page, "PageTail not set", 0); | |
980 | goto out; | |
981 | } | |
982 | if (unlikely(compound_head(page) != head_page)) { | |
983 | bad_page(page, "compound_head not consistent", 0); | |
984 | goto out; | |
985 | } | |
986 | ret = 0; | |
987 | out: | |
988 | page->mapping = NULL; | |
989 | clear_compound_head(page); | |
990 | return ret; | |
991 | } | |
992 | ||
e2769dbd MG |
993 | static __always_inline bool free_pages_prepare(struct page *page, |
994 | unsigned int order, bool check_free) | |
4db7548c | 995 | { |
e2769dbd | 996 | int bad = 0; |
4db7548c | 997 | |
4db7548c MG |
998 | VM_BUG_ON_PAGE(PageTail(page), page); |
999 | ||
e2769dbd MG |
1000 | trace_mm_page_free(page, order); |
1001 | kmemcheck_free_shadow(page, order); | |
e2769dbd MG |
1002 | |
1003 | /* | |
1004 | * Check tail pages before head page information is cleared to | |
1005 | * avoid checking PageCompound for order-0 pages. | |
1006 | */ | |
1007 | if (unlikely(order)) { | |
1008 | bool compound = PageCompound(page); | |
1009 | int i; | |
1010 | ||
1011 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1012 | |
9a73f61b KS |
1013 | if (compound) |
1014 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1015 | for (i = 1; i < (1 << order); i++) { |
1016 | if (compound) | |
1017 | bad += free_tail_pages_check(page, page + i); | |
1018 | if (unlikely(free_pages_check(page + i))) { | |
1019 | bad++; | |
1020 | continue; | |
1021 | } | |
1022 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1023 | } | |
1024 | } | |
bda807d4 | 1025 | if (PageMappingFlags(page)) |
4db7548c | 1026 | page->mapping = NULL; |
c4159a75 | 1027 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
4949148a | 1028 | memcg_kmem_uncharge(page, order); |
e2769dbd MG |
1029 | if (check_free) |
1030 | bad += free_pages_check(page); | |
1031 | if (bad) | |
1032 | return false; | |
4db7548c | 1033 | |
e2769dbd MG |
1034 | page_cpupid_reset_last(page); |
1035 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1036 | reset_page_owner(page, order); | |
4db7548c MG |
1037 | |
1038 | if (!PageHighMem(page)) { | |
1039 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1040 | PAGE_SIZE << order); |
4db7548c | 1041 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1042 | PAGE_SIZE << order); |
4db7548c | 1043 | } |
e2769dbd MG |
1044 | arch_free_page(page, order); |
1045 | kernel_poison_pages(page, 1 << order, 0); | |
1046 | kernel_map_pages(page, 1 << order, 0); | |
29b52de1 | 1047 | kasan_free_pages(page, order); |
4db7548c | 1048 | |
4db7548c MG |
1049 | return true; |
1050 | } | |
1051 | ||
e2769dbd MG |
1052 | #ifdef CONFIG_DEBUG_VM |
1053 | static inline bool free_pcp_prepare(struct page *page) | |
1054 | { | |
1055 | return free_pages_prepare(page, 0, true); | |
1056 | } | |
1057 | ||
1058 | static inline bool bulkfree_pcp_prepare(struct page *page) | |
1059 | { | |
1060 | return false; | |
1061 | } | |
1062 | #else | |
1063 | static bool free_pcp_prepare(struct page *page) | |
1064 | { | |
1065 | return free_pages_prepare(page, 0, false); | |
1066 | } | |
1067 | ||
4db7548c MG |
1068 | static bool bulkfree_pcp_prepare(struct page *page) |
1069 | { | |
1070 | return free_pages_check(page); | |
1071 | } | |
1072 | #endif /* CONFIG_DEBUG_VM */ | |
1073 | ||
1da177e4 | 1074 | /* |
5f8dcc21 | 1075 | * Frees a number of pages from the PCP lists |
1da177e4 | 1076 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1077 | * count is the number of pages to free. |
1da177e4 LT |
1078 | * |
1079 | * If the zone was previously in an "all pages pinned" state then look to | |
1080 | * see if this freeing clears that state. | |
1081 | * | |
1082 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1083 | * pinned" detection logic. | |
1084 | */ | |
5f8dcc21 MG |
1085 | static void free_pcppages_bulk(struct zone *zone, int count, |
1086 | struct per_cpu_pages *pcp) | |
1da177e4 | 1087 | { |
5f8dcc21 | 1088 | int migratetype = 0; |
a6f9edd6 | 1089 | int batch_free = 0; |
0d5d823a | 1090 | unsigned long nr_scanned; |
3777999d | 1091 | bool isolated_pageblocks; |
5f8dcc21 | 1092 | |
c54ad30c | 1093 | spin_lock(&zone->lock); |
3777999d | 1094 | isolated_pageblocks = has_isolate_pageblock(zone); |
599d0c95 | 1095 | nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED); |
0d5d823a | 1096 | if (nr_scanned) |
599d0c95 | 1097 | __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned); |
f2260e6b | 1098 | |
e5b31ac2 | 1099 | while (count) { |
48db57f8 | 1100 | struct page *page; |
5f8dcc21 MG |
1101 | struct list_head *list; |
1102 | ||
1103 | /* | |
a6f9edd6 MG |
1104 | * Remove pages from lists in a round-robin fashion. A |
1105 | * batch_free count is maintained that is incremented when an | |
1106 | * empty list is encountered. This is so more pages are freed | |
1107 | * off fuller lists instead of spinning excessively around empty | |
1108 | * lists | |
5f8dcc21 MG |
1109 | */ |
1110 | do { | |
a6f9edd6 | 1111 | batch_free++; |
5f8dcc21 MG |
1112 | if (++migratetype == MIGRATE_PCPTYPES) |
1113 | migratetype = 0; | |
1114 | list = &pcp->lists[migratetype]; | |
1115 | } while (list_empty(list)); | |
48db57f8 | 1116 | |
1d16871d NK |
1117 | /* This is the only non-empty list. Free them all. */ |
1118 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1119 | batch_free = count; |
1d16871d | 1120 | |
a6f9edd6 | 1121 | do { |
770c8aaa BZ |
1122 | int mt; /* migratetype of the to-be-freed page */ |
1123 | ||
a16601c5 | 1124 | page = list_last_entry(list, struct page, lru); |
a6f9edd6 MG |
1125 | /* must delete as __free_one_page list manipulates */ |
1126 | list_del(&page->lru); | |
aa016d14 | 1127 | |
bb14c2c7 | 1128 | mt = get_pcppage_migratetype(page); |
aa016d14 VB |
1129 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1130 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1131 | /* Pageblock could have been isolated meanwhile */ | |
3777999d | 1132 | if (unlikely(isolated_pageblocks)) |
51bb1a40 | 1133 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 1134 | |
4db7548c MG |
1135 | if (bulkfree_pcp_prepare(page)) |
1136 | continue; | |
1137 | ||
dc4b0caf | 1138 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 1139 | trace_mm_page_pcpu_drain(page, 0, mt); |
e5b31ac2 | 1140 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1141 | } |
c54ad30c | 1142 | spin_unlock(&zone->lock); |
1da177e4 LT |
1143 | } |
1144 | ||
dc4b0caf MG |
1145 | static void free_one_page(struct zone *zone, |
1146 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1147 | unsigned int order, |
ed0ae21d | 1148 | int migratetype) |
1da177e4 | 1149 | { |
0d5d823a | 1150 | unsigned long nr_scanned; |
006d22d9 | 1151 | spin_lock(&zone->lock); |
599d0c95 | 1152 | nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED); |
0d5d823a | 1153 | if (nr_scanned) |
599d0c95 | 1154 | __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned); |
f2260e6b | 1155 | |
ad53f92e JK |
1156 | if (unlikely(has_isolate_pageblock(zone) || |
1157 | is_migrate_isolate(migratetype))) { | |
1158 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1159 | } |
dc4b0caf | 1160 | __free_one_page(page, pfn, zone, order, migratetype); |
006d22d9 | 1161 | spin_unlock(&zone->lock); |
48db57f8 NP |
1162 | } |
1163 | ||
1e8ce83c RH |
1164 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
1165 | unsigned long zone, int nid) | |
1166 | { | |
1e8ce83c | 1167 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1168 | init_page_count(page); |
1169 | page_mapcount_reset(page); | |
1170 | page_cpupid_reset_last(page); | |
1e8ce83c | 1171 | |
1e8ce83c RH |
1172 | INIT_LIST_HEAD(&page->lru); |
1173 | #ifdef WANT_PAGE_VIRTUAL | |
1174 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1175 | if (!is_highmem_idx(zone)) | |
1176 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1177 | #endif | |
1178 | } | |
1179 | ||
1180 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
1181 | int nid) | |
1182 | { | |
1183 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
1184 | } | |
1185 | ||
7e18adb4 MG |
1186 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1187 | static void init_reserved_page(unsigned long pfn) | |
1188 | { | |
1189 | pg_data_t *pgdat; | |
1190 | int nid, zid; | |
1191 | ||
1192 | if (!early_page_uninitialised(pfn)) | |
1193 | return; | |
1194 | ||
1195 | nid = early_pfn_to_nid(pfn); | |
1196 | pgdat = NODE_DATA(nid); | |
1197 | ||
1198 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1199 | struct zone *zone = &pgdat->node_zones[zid]; | |
1200 | ||
1201 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1202 | break; | |
1203 | } | |
1204 | __init_single_pfn(pfn, zid, nid); | |
1205 | } | |
1206 | #else | |
1207 | static inline void init_reserved_page(unsigned long pfn) | |
1208 | { | |
1209 | } | |
1210 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1211 | ||
92923ca3 NZ |
1212 | /* |
1213 | * Initialised pages do not have PageReserved set. This function is | |
1214 | * called for each range allocated by the bootmem allocator and | |
1215 | * marks the pages PageReserved. The remaining valid pages are later | |
1216 | * sent to the buddy page allocator. | |
1217 | */ | |
4b50bcc7 | 1218 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1219 | { |
1220 | unsigned long start_pfn = PFN_DOWN(start); | |
1221 | unsigned long end_pfn = PFN_UP(end); | |
1222 | ||
7e18adb4 MG |
1223 | for (; start_pfn < end_pfn; start_pfn++) { |
1224 | if (pfn_valid(start_pfn)) { | |
1225 | struct page *page = pfn_to_page(start_pfn); | |
1226 | ||
1227 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1228 | |
1229 | /* Avoid false-positive PageTail() */ | |
1230 | INIT_LIST_HEAD(&page->lru); | |
1231 | ||
7e18adb4 MG |
1232 | SetPageReserved(page); |
1233 | } | |
1234 | } | |
92923ca3 NZ |
1235 | } |
1236 | ||
ec95f53a KM |
1237 | static void __free_pages_ok(struct page *page, unsigned int order) |
1238 | { | |
1239 | unsigned long flags; | |
95e34412 | 1240 | int migratetype; |
dc4b0caf | 1241 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1242 | |
e2769dbd | 1243 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1244 | return; |
1245 | ||
cfc47a28 | 1246 | migratetype = get_pfnblock_migratetype(page, pfn); |
c54ad30c | 1247 | local_irq_save(flags); |
f8891e5e | 1248 | __count_vm_events(PGFREE, 1 << order); |
dc4b0caf | 1249 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
c54ad30c | 1250 | local_irq_restore(flags); |
1da177e4 LT |
1251 | } |
1252 | ||
949698a3 | 1253 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1254 | { |
c3993076 | 1255 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1256 | struct page *p = page; |
c3993076 | 1257 | unsigned int loop; |
a226f6c8 | 1258 | |
e2d0bd2b YL |
1259 | prefetchw(p); |
1260 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1261 | prefetchw(p + 1); | |
c3993076 JW |
1262 | __ClearPageReserved(p); |
1263 | set_page_count(p, 0); | |
a226f6c8 | 1264 | } |
e2d0bd2b YL |
1265 | __ClearPageReserved(p); |
1266 | set_page_count(p, 0); | |
c3993076 | 1267 | |
e2d0bd2b | 1268 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1269 | set_page_refcounted(page); |
1270 | __free_pages(page, order); | |
a226f6c8 DH |
1271 | } |
1272 | ||
75a592a4 MG |
1273 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1274 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1275 | |
75a592a4 MG |
1276 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1277 | ||
1278 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1279 | { | |
7ace9917 | 1280 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1281 | int nid; |
1282 | ||
7ace9917 | 1283 | spin_lock(&early_pfn_lock); |
75a592a4 | 1284 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1285 | if (nid < 0) |
e4568d38 | 1286 | nid = first_online_node; |
7ace9917 MG |
1287 | spin_unlock(&early_pfn_lock); |
1288 | ||
1289 | return nid; | |
75a592a4 MG |
1290 | } |
1291 | #endif | |
1292 | ||
1293 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
1294 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1295 | struct mminit_pfnnid_cache *state) | |
1296 | { | |
1297 | int nid; | |
1298 | ||
1299 | nid = __early_pfn_to_nid(pfn, state); | |
1300 | if (nid >= 0 && nid != node) | |
1301 | return false; | |
1302 | return true; | |
1303 | } | |
1304 | ||
1305 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1306 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1307 | { | |
1308 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1309 | } | |
1310 | ||
1311 | #else | |
1312 | ||
1313 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1314 | { | |
1315 | return true; | |
1316 | } | |
1317 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1318 | struct mminit_pfnnid_cache *state) | |
1319 | { | |
1320 | return true; | |
1321 | } | |
1322 | #endif | |
1323 | ||
1324 | ||
0e1cc95b | 1325 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1326 | unsigned int order) |
1327 | { | |
1328 | if (early_page_uninitialised(pfn)) | |
1329 | return; | |
949698a3 | 1330 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1331 | } |
1332 | ||
7cf91a98 JK |
1333 | /* |
1334 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1335 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1336 | * with the migration of free compaction scanner. The scanners then need to | |
1337 | * use only pfn_valid_within() check for arches that allow holes within | |
1338 | * pageblocks. | |
1339 | * | |
1340 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1341 | * | |
1342 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1343 | * i.e. it's possible that all pages within a zones range of pages do not | |
1344 | * belong to a single zone. We assume that a border between node0 and node1 | |
1345 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1346 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1347 | * the first and last page of a pageblock and avoid checking each individual | |
1348 | * page in a pageblock. | |
1349 | */ | |
1350 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1351 | unsigned long end_pfn, struct zone *zone) | |
1352 | { | |
1353 | struct page *start_page; | |
1354 | struct page *end_page; | |
1355 | ||
1356 | /* end_pfn is one past the range we are checking */ | |
1357 | end_pfn--; | |
1358 | ||
1359 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1360 | return NULL; | |
1361 | ||
1362 | start_page = pfn_to_page(start_pfn); | |
1363 | ||
1364 | if (page_zone(start_page) != zone) | |
1365 | return NULL; | |
1366 | ||
1367 | end_page = pfn_to_page(end_pfn); | |
1368 | ||
1369 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1370 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1371 | return NULL; | |
1372 | ||
1373 | return start_page; | |
1374 | } | |
1375 | ||
1376 | void set_zone_contiguous(struct zone *zone) | |
1377 | { | |
1378 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1379 | unsigned long block_end_pfn; | |
1380 | ||
1381 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1382 | for (; block_start_pfn < zone_end_pfn(zone); | |
1383 | block_start_pfn = block_end_pfn, | |
1384 | block_end_pfn += pageblock_nr_pages) { | |
1385 | ||
1386 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1387 | ||
1388 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1389 | block_end_pfn, zone)) | |
1390 | return; | |
1391 | } | |
1392 | ||
1393 | /* We confirm that there is no hole */ | |
1394 | zone->contiguous = true; | |
1395 | } | |
1396 | ||
1397 | void clear_zone_contiguous(struct zone *zone) | |
1398 | { | |
1399 | zone->contiguous = false; | |
1400 | } | |
1401 | ||
7e18adb4 | 1402 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
0e1cc95b | 1403 | static void __init deferred_free_range(struct page *page, |
a4de83dd MG |
1404 | unsigned long pfn, int nr_pages) |
1405 | { | |
1406 | int i; | |
1407 | ||
1408 | if (!page) | |
1409 | return; | |
1410 | ||
1411 | /* Free a large naturally-aligned chunk if possible */ | |
e780149b XQ |
1412 | if (nr_pages == pageblock_nr_pages && |
1413 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1414 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
e780149b | 1415 | __free_pages_boot_core(page, pageblock_order); |
a4de83dd MG |
1416 | return; |
1417 | } | |
1418 | ||
e780149b XQ |
1419 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1420 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1421 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
949698a3 | 1422 | __free_pages_boot_core(page, 0); |
e780149b | 1423 | } |
a4de83dd MG |
1424 | } |
1425 | ||
d3cd131d NS |
1426 | /* Completion tracking for deferred_init_memmap() threads */ |
1427 | static atomic_t pgdat_init_n_undone __initdata; | |
1428 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1429 | ||
1430 | static inline void __init pgdat_init_report_one_done(void) | |
1431 | { | |
1432 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1433 | complete(&pgdat_init_all_done_comp); | |
1434 | } | |
0e1cc95b | 1435 | |
7e18adb4 | 1436 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1437 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1438 | { |
0e1cc95b MG |
1439 | pg_data_t *pgdat = data; |
1440 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1441 | struct mminit_pfnnid_cache nid_init_state = { }; |
1442 | unsigned long start = jiffies; | |
1443 | unsigned long nr_pages = 0; | |
1444 | unsigned long walk_start, walk_end; | |
1445 | int i, zid; | |
1446 | struct zone *zone; | |
7e18adb4 | 1447 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
0e1cc95b | 1448 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
7e18adb4 | 1449 | |
0e1cc95b | 1450 | if (first_init_pfn == ULONG_MAX) { |
d3cd131d | 1451 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1452 | return 0; |
1453 | } | |
1454 | ||
1455 | /* Bind memory initialisation thread to a local node if possible */ | |
1456 | if (!cpumask_empty(cpumask)) | |
1457 | set_cpus_allowed_ptr(current, cpumask); | |
7e18adb4 MG |
1458 | |
1459 | /* Sanity check boundaries */ | |
1460 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1461 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1462 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1463 | ||
1464 | /* Only the highest zone is deferred so find it */ | |
1465 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1466 | zone = pgdat->node_zones + zid; | |
1467 | if (first_init_pfn < zone_end_pfn(zone)) | |
1468 | break; | |
1469 | } | |
1470 | ||
1471 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { | |
1472 | unsigned long pfn, end_pfn; | |
54608c3f | 1473 | struct page *page = NULL; |
a4de83dd MG |
1474 | struct page *free_base_page = NULL; |
1475 | unsigned long free_base_pfn = 0; | |
1476 | int nr_to_free = 0; | |
7e18adb4 MG |
1477 | |
1478 | end_pfn = min(walk_end, zone_end_pfn(zone)); | |
1479 | pfn = first_init_pfn; | |
1480 | if (pfn < walk_start) | |
1481 | pfn = walk_start; | |
1482 | if (pfn < zone->zone_start_pfn) | |
1483 | pfn = zone->zone_start_pfn; | |
1484 | ||
1485 | for (; pfn < end_pfn; pfn++) { | |
54608c3f | 1486 | if (!pfn_valid_within(pfn)) |
a4de83dd | 1487 | goto free_range; |
7e18adb4 | 1488 | |
54608c3f MG |
1489 | /* |
1490 | * Ensure pfn_valid is checked every | |
e780149b | 1491 | * pageblock_nr_pages for memory holes |
54608c3f | 1492 | */ |
e780149b | 1493 | if ((pfn & (pageblock_nr_pages - 1)) == 0) { |
54608c3f MG |
1494 | if (!pfn_valid(pfn)) { |
1495 | page = NULL; | |
a4de83dd | 1496 | goto free_range; |
54608c3f MG |
1497 | } |
1498 | } | |
1499 | ||
1500 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { | |
1501 | page = NULL; | |
a4de83dd | 1502 | goto free_range; |
54608c3f MG |
1503 | } |
1504 | ||
1505 | /* Minimise pfn page lookups and scheduler checks */ | |
e780149b | 1506 | if (page && (pfn & (pageblock_nr_pages - 1)) != 0) { |
54608c3f MG |
1507 | page++; |
1508 | } else { | |
a4de83dd MG |
1509 | nr_pages += nr_to_free; |
1510 | deferred_free_range(free_base_page, | |
1511 | free_base_pfn, nr_to_free); | |
1512 | free_base_page = NULL; | |
1513 | free_base_pfn = nr_to_free = 0; | |
1514 | ||
54608c3f MG |
1515 | page = pfn_to_page(pfn); |
1516 | cond_resched(); | |
1517 | } | |
7e18adb4 MG |
1518 | |
1519 | if (page->flags) { | |
1520 | VM_BUG_ON(page_zone(page) != zone); | |
a4de83dd | 1521 | goto free_range; |
7e18adb4 MG |
1522 | } |
1523 | ||
1524 | __init_single_page(page, pfn, zid, nid); | |
a4de83dd MG |
1525 | if (!free_base_page) { |
1526 | free_base_page = page; | |
1527 | free_base_pfn = pfn; | |
1528 | nr_to_free = 0; | |
1529 | } | |
1530 | nr_to_free++; | |
1531 | ||
1532 | /* Where possible, batch up pages for a single free */ | |
1533 | continue; | |
1534 | free_range: | |
1535 | /* Free the current block of pages to allocator */ | |
1536 | nr_pages += nr_to_free; | |
1537 | deferred_free_range(free_base_page, free_base_pfn, | |
1538 | nr_to_free); | |
1539 | free_base_page = NULL; | |
1540 | free_base_pfn = nr_to_free = 0; | |
7e18adb4 | 1541 | } |
e780149b XQ |
1542 | /* Free the last block of pages to allocator */ |
1543 | nr_pages += nr_to_free; | |
1544 | deferred_free_range(free_base_page, free_base_pfn, nr_to_free); | |
a4de83dd | 1545 | |
7e18adb4 MG |
1546 | first_init_pfn = max(end_pfn, first_init_pfn); |
1547 | } | |
1548 | ||
1549 | /* Sanity check that the next zone really is unpopulated */ | |
1550 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1551 | ||
0e1cc95b | 1552 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1553 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1554 | |
1555 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1556 | return 0; |
1557 | } | |
7cf91a98 | 1558 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1559 | |
1560 | void __init page_alloc_init_late(void) | |
1561 | { | |
7cf91a98 JK |
1562 | struct zone *zone; |
1563 | ||
1564 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1565 | int nid; |
1566 | ||
d3cd131d NS |
1567 | /* There will be num_node_state(N_MEMORY) threads */ |
1568 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1569 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1570 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1571 | } | |
1572 | ||
1573 | /* Block until all are initialised */ | |
d3cd131d | 1574 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da MG |
1575 | |
1576 | /* Reinit limits that are based on free pages after the kernel is up */ | |
1577 | files_maxfiles_init(); | |
7cf91a98 JK |
1578 | #endif |
1579 | ||
1580 | for_each_populated_zone(zone) | |
1581 | set_zone_contiguous(zone); | |
7e18adb4 | 1582 | } |
7e18adb4 | 1583 | |
47118af0 | 1584 | #ifdef CONFIG_CMA |
9cf510a5 | 1585 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1586 | void __init init_cma_reserved_pageblock(struct page *page) |
1587 | { | |
1588 | unsigned i = pageblock_nr_pages; | |
1589 | struct page *p = page; | |
1590 | ||
1591 | do { | |
1592 | __ClearPageReserved(p); | |
1593 | set_page_count(p, 0); | |
1594 | } while (++p, --i); | |
1595 | ||
47118af0 | 1596 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1597 | |
1598 | if (pageblock_order >= MAX_ORDER) { | |
1599 | i = pageblock_nr_pages; | |
1600 | p = page; | |
1601 | do { | |
1602 | set_page_refcounted(p); | |
1603 | __free_pages(p, MAX_ORDER - 1); | |
1604 | p += MAX_ORDER_NR_PAGES; | |
1605 | } while (i -= MAX_ORDER_NR_PAGES); | |
1606 | } else { | |
1607 | set_page_refcounted(page); | |
1608 | __free_pages(page, pageblock_order); | |
1609 | } | |
1610 | ||
3dcc0571 | 1611 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1612 | } |
1613 | #endif | |
1da177e4 LT |
1614 | |
1615 | /* | |
1616 | * The order of subdivision here is critical for the IO subsystem. | |
1617 | * Please do not alter this order without good reasons and regression | |
1618 | * testing. Specifically, as large blocks of memory are subdivided, | |
1619 | * the order in which smaller blocks are delivered depends on the order | |
1620 | * they're subdivided in this function. This is the primary factor | |
1621 | * influencing the order in which pages are delivered to the IO | |
1622 | * subsystem according to empirical testing, and this is also justified | |
1623 | * by considering the behavior of a buddy system containing a single | |
1624 | * large block of memory acted on by a series of small allocations. | |
1625 | * This behavior is a critical factor in sglist merging's success. | |
1626 | * | |
6d49e352 | 1627 | * -- nyc |
1da177e4 | 1628 | */ |
085cc7d5 | 1629 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1630 | int low, int high, struct free_area *area, |
1631 | int migratetype) | |
1da177e4 LT |
1632 | { |
1633 | unsigned long size = 1 << high; | |
1634 | ||
1635 | while (high > low) { | |
1636 | area--; | |
1637 | high--; | |
1638 | size >>= 1; | |
309381fe | 1639 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1640 | |
acbc15a4 JK |
1641 | /* |
1642 | * Mark as guard pages (or page), that will allow to | |
1643 | * merge back to allocator when buddy will be freed. | |
1644 | * Corresponding page table entries will not be touched, | |
1645 | * pages will stay not present in virtual address space | |
1646 | */ | |
1647 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 1648 | continue; |
acbc15a4 | 1649 | |
b2a0ac88 | 1650 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1651 | area->nr_free++; |
1652 | set_page_order(&page[size], high); | |
1653 | } | |
1da177e4 LT |
1654 | } |
1655 | ||
4e611801 | 1656 | static void check_new_page_bad(struct page *page) |
1da177e4 | 1657 | { |
4e611801 VB |
1658 | const char *bad_reason = NULL; |
1659 | unsigned long bad_flags = 0; | |
7bfec6f4 | 1660 | |
53f9263b | 1661 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1662 | bad_reason = "nonzero mapcount"; |
1663 | if (unlikely(page->mapping != NULL)) | |
1664 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1665 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1666 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1667 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1668 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1669 | bad_flags = __PG_HWPOISON; | |
e570f56c NH |
1670 | /* Don't complain about hwpoisoned pages */ |
1671 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1672 | return; | |
f4c18e6f | 1673 | } |
f0b791a3 DH |
1674 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1675 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1676 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1677 | } | |
9edad6ea JW |
1678 | #ifdef CONFIG_MEMCG |
1679 | if (unlikely(page->mem_cgroup)) | |
1680 | bad_reason = "page still charged to cgroup"; | |
1681 | #endif | |
4e611801 VB |
1682 | bad_page(page, bad_reason, bad_flags); |
1683 | } | |
1684 | ||
1685 | /* | |
1686 | * This page is about to be returned from the page allocator | |
1687 | */ | |
1688 | static inline int check_new_page(struct page *page) | |
1689 | { | |
1690 | if (likely(page_expected_state(page, | |
1691 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1692 | return 0; | |
1693 | ||
1694 | check_new_page_bad(page); | |
1695 | return 1; | |
2a7684a2 WF |
1696 | } |
1697 | ||
1414c7f4 LA |
1698 | static inline bool free_pages_prezeroed(bool poisoned) |
1699 | { | |
1700 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
1701 | page_poisoning_enabled() && poisoned; | |
1702 | } | |
1703 | ||
479f854a MG |
1704 | #ifdef CONFIG_DEBUG_VM |
1705 | static bool check_pcp_refill(struct page *page) | |
1706 | { | |
1707 | return false; | |
1708 | } | |
1709 | ||
1710 | static bool check_new_pcp(struct page *page) | |
1711 | { | |
1712 | return check_new_page(page); | |
1713 | } | |
1714 | #else | |
1715 | static bool check_pcp_refill(struct page *page) | |
1716 | { | |
1717 | return check_new_page(page); | |
1718 | } | |
1719 | static bool check_new_pcp(struct page *page) | |
1720 | { | |
1721 | return false; | |
1722 | } | |
1723 | #endif /* CONFIG_DEBUG_VM */ | |
1724 | ||
1725 | static bool check_new_pages(struct page *page, unsigned int order) | |
1726 | { | |
1727 | int i; | |
1728 | for (i = 0; i < (1 << order); i++) { | |
1729 | struct page *p = page + i; | |
1730 | ||
1731 | if (unlikely(check_new_page(p))) | |
1732 | return true; | |
1733 | } | |
1734 | ||
1735 | return false; | |
1736 | } | |
1737 | ||
46f24fd8 JK |
1738 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1739 | gfp_t gfp_flags) | |
1740 | { | |
1741 | set_page_private(page, 0); | |
1742 | set_page_refcounted(page); | |
1743 | ||
1744 | arch_alloc_page(page, order); | |
1745 | kernel_map_pages(page, 1 << order, 1); | |
1746 | kernel_poison_pages(page, 1 << order, 1); | |
1747 | kasan_alloc_pages(page, order); | |
1748 | set_page_owner(page, order, gfp_flags); | |
1749 | } | |
1750 | ||
479f854a | 1751 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1752 | unsigned int alloc_flags) |
2a7684a2 WF |
1753 | { |
1754 | int i; | |
1414c7f4 | 1755 | bool poisoned = true; |
2a7684a2 WF |
1756 | |
1757 | for (i = 0; i < (1 << order); i++) { | |
1758 | struct page *p = page + i; | |
1414c7f4 LA |
1759 | if (poisoned) |
1760 | poisoned &= page_is_poisoned(p); | |
2a7684a2 | 1761 | } |
689bcebf | 1762 | |
46f24fd8 | 1763 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1764 | |
1414c7f4 | 1765 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1766 | for (i = 0; i < (1 << order); i++) |
1767 | clear_highpage(page + i); | |
17cf4406 NP |
1768 | |
1769 | if (order && (gfp_flags & __GFP_COMP)) | |
1770 | prep_compound_page(page, order); | |
1771 | ||
75379191 | 1772 | /* |
2f064f34 | 1773 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1774 | * allocate the page. The expectation is that the caller is taking |
1775 | * steps that will free more memory. The caller should avoid the page | |
1776 | * being used for !PFMEMALLOC purposes. | |
1777 | */ | |
2f064f34 MH |
1778 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1779 | set_page_pfmemalloc(page); | |
1780 | else | |
1781 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1782 | } |
1783 | ||
56fd56b8 MG |
1784 | /* |
1785 | * Go through the free lists for the given migratetype and remove | |
1786 | * the smallest available page from the freelists | |
1787 | */ | |
728ec980 MG |
1788 | static inline |
1789 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1790 | int migratetype) |
1791 | { | |
1792 | unsigned int current_order; | |
b8af2941 | 1793 | struct free_area *area; |
56fd56b8 MG |
1794 | struct page *page; |
1795 | ||
1796 | /* Find a page of the appropriate size in the preferred list */ | |
1797 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1798 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1799 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1800 | struct page, lru); |
a16601c5 GT |
1801 | if (!page) |
1802 | continue; | |
56fd56b8 MG |
1803 | list_del(&page->lru); |
1804 | rmv_page_order(page); | |
1805 | area->nr_free--; | |
56fd56b8 | 1806 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1807 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1808 | return page; |
1809 | } | |
1810 | ||
1811 | return NULL; | |
1812 | } | |
1813 | ||
1814 | ||
b2a0ac88 MG |
1815 | /* |
1816 | * This array describes the order lists are fallen back to when | |
1817 | * the free lists for the desirable migrate type are depleted | |
1818 | */ | |
47118af0 | 1819 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1820 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1821 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1822 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1823 | #ifdef CONFIG_CMA |
974a786e | 1824 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1825 | #endif |
194159fb | 1826 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1827 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1828 | #endif |
b2a0ac88 MG |
1829 | }; |
1830 | ||
dc67647b JK |
1831 | #ifdef CONFIG_CMA |
1832 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1833 | unsigned int order) | |
1834 | { | |
1835 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1836 | } | |
1837 | #else | |
1838 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1839 | unsigned int order) { return NULL; } | |
1840 | #endif | |
1841 | ||
c361be55 MG |
1842 | /* |
1843 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1844 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1845 | * boundary. If alignment is required, use move_freepages_block() |
1846 | */ | |
435b405c | 1847 | int move_freepages(struct zone *zone, |
b69a7288 AB |
1848 | struct page *start_page, struct page *end_page, |
1849 | int migratetype) | |
c361be55 MG |
1850 | { |
1851 | struct page *page; | |
d00181b9 | 1852 | unsigned int order; |
d100313f | 1853 | int pages_moved = 0; |
c361be55 MG |
1854 | |
1855 | #ifndef CONFIG_HOLES_IN_ZONE | |
1856 | /* | |
1857 | * page_zone is not safe to call in this context when | |
1858 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1859 | * anyway as we check zone boundaries in move_freepages_block(). | |
1860 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1861 | * grouping pages by mobility |
c361be55 | 1862 | */ |
97ee4ba7 | 1863 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1864 | #endif |
1865 | ||
1866 | for (page = start_page; page <= end_page;) { | |
1867 | if (!pfn_valid_within(page_to_pfn(page))) { | |
1868 | page++; | |
1869 | continue; | |
1870 | } | |
1871 | ||
f073bdc5 AB |
1872 | /* Make sure we are not inadvertently changing nodes */ |
1873 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
1874 | ||
c361be55 MG |
1875 | if (!PageBuddy(page)) { |
1876 | page++; | |
1877 | continue; | |
1878 | } | |
1879 | ||
1880 | order = page_order(page); | |
84be48d8 KS |
1881 | list_move(&page->lru, |
1882 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 1883 | page += 1 << order; |
d100313f | 1884 | pages_moved += 1 << order; |
c361be55 MG |
1885 | } |
1886 | ||
d100313f | 1887 | return pages_moved; |
c361be55 MG |
1888 | } |
1889 | ||
ee6f509c | 1890 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1891 | int migratetype) |
c361be55 MG |
1892 | { |
1893 | unsigned long start_pfn, end_pfn; | |
1894 | struct page *start_page, *end_page; | |
1895 | ||
1896 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1897 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1898 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1899 | end_page = start_page + pageblock_nr_pages - 1; |
1900 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1901 | |
1902 | /* Do not cross zone boundaries */ | |
108bcc96 | 1903 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1904 | start_page = page; |
108bcc96 | 1905 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1906 | return 0; |
1907 | ||
1908 | return move_freepages(zone, start_page, end_page, migratetype); | |
1909 | } | |
1910 | ||
2f66a68f MG |
1911 | static void change_pageblock_range(struct page *pageblock_page, |
1912 | int start_order, int migratetype) | |
1913 | { | |
1914 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1915 | ||
1916 | while (nr_pageblocks--) { | |
1917 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1918 | pageblock_page += pageblock_nr_pages; | |
1919 | } | |
1920 | } | |
1921 | ||
fef903ef | 1922 | /* |
9c0415eb VB |
1923 | * When we are falling back to another migratetype during allocation, try to |
1924 | * steal extra free pages from the same pageblocks to satisfy further | |
1925 | * allocations, instead of polluting multiple pageblocks. | |
1926 | * | |
1927 | * If we are stealing a relatively large buddy page, it is likely there will | |
1928 | * be more free pages in the pageblock, so try to steal them all. For | |
1929 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1930 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1931 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1932 | * pageblocks. | |
fef903ef | 1933 | */ |
4eb7dce6 JK |
1934 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1935 | { | |
1936 | /* | |
1937 | * Leaving this order check is intended, although there is | |
1938 | * relaxed order check in next check. The reason is that | |
1939 | * we can actually steal whole pageblock if this condition met, | |
1940 | * but, below check doesn't guarantee it and that is just heuristic | |
1941 | * so could be changed anytime. | |
1942 | */ | |
1943 | if (order >= pageblock_order) | |
1944 | return true; | |
1945 | ||
1946 | if (order >= pageblock_order / 2 || | |
1947 | start_mt == MIGRATE_RECLAIMABLE || | |
1948 | start_mt == MIGRATE_UNMOVABLE || | |
1949 | page_group_by_mobility_disabled) | |
1950 | return true; | |
1951 | ||
1952 | return false; | |
1953 | } | |
1954 | ||
1955 | /* | |
1956 | * This function implements actual steal behaviour. If order is large enough, | |
1957 | * we can steal whole pageblock. If not, we first move freepages in this | |
1958 | * pageblock and check whether half of pages are moved or not. If half of | |
1959 | * pages are moved, we can change migratetype of pageblock and permanently | |
1960 | * use it's pages as requested migratetype in the future. | |
1961 | */ | |
1962 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1963 | int start_type) | |
fef903ef | 1964 | { |
d00181b9 | 1965 | unsigned int current_order = page_order(page); |
4eb7dce6 | 1966 | int pages; |
fef903ef | 1967 | |
fef903ef SB |
1968 | /* Take ownership for orders >= pageblock_order */ |
1969 | if (current_order >= pageblock_order) { | |
1970 | change_pageblock_range(page, current_order, start_type); | |
3a1086fb | 1971 | return; |
fef903ef SB |
1972 | } |
1973 | ||
4eb7dce6 | 1974 | pages = move_freepages_block(zone, page, start_type); |
fef903ef | 1975 | |
4eb7dce6 JK |
1976 | /* Claim the whole block if over half of it is free */ |
1977 | if (pages >= (1 << (pageblock_order-1)) || | |
1978 | page_group_by_mobility_disabled) | |
1979 | set_pageblock_migratetype(page, start_type); | |
1980 | } | |
1981 | ||
2149cdae JK |
1982 | /* |
1983 | * Check whether there is a suitable fallback freepage with requested order. | |
1984 | * If only_stealable is true, this function returns fallback_mt only if | |
1985 | * we can steal other freepages all together. This would help to reduce | |
1986 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1987 | */ | |
1988 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1989 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1990 | { |
1991 | int i; | |
1992 | int fallback_mt; | |
1993 | ||
1994 | if (area->nr_free == 0) | |
1995 | return -1; | |
1996 | ||
1997 | *can_steal = false; | |
1998 | for (i = 0;; i++) { | |
1999 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2000 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2001 | break; |
2002 | ||
2003 | if (list_empty(&area->free_list[fallback_mt])) | |
2004 | continue; | |
fef903ef | 2005 | |
4eb7dce6 JK |
2006 | if (can_steal_fallback(order, migratetype)) |
2007 | *can_steal = true; | |
2008 | ||
2149cdae JK |
2009 | if (!only_stealable) |
2010 | return fallback_mt; | |
2011 | ||
2012 | if (*can_steal) | |
2013 | return fallback_mt; | |
fef903ef | 2014 | } |
4eb7dce6 JK |
2015 | |
2016 | return -1; | |
fef903ef SB |
2017 | } |
2018 | ||
0aaa29a5 MG |
2019 | /* |
2020 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2021 | * there are no empty page blocks that contain a page with a suitable order | |
2022 | */ | |
2023 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2024 | unsigned int alloc_order) | |
2025 | { | |
2026 | int mt; | |
2027 | unsigned long max_managed, flags; | |
2028 | ||
2029 | /* | |
2030 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2031 | * Check is race-prone but harmless. | |
2032 | */ | |
2033 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
2034 | if (zone->nr_reserved_highatomic >= max_managed) | |
2035 | return; | |
2036 | ||
2037 | spin_lock_irqsave(&zone->lock, flags); | |
2038 | ||
2039 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2040 | if (zone->nr_reserved_highatomic >= max_managed) | |
2041 | goto out_unlock; | |
2042 | ||
2043 | /* Yoink! */ | |
2044 | mt = get_pageblock_migratetype(page); | |
2045 | if (mt != MIGRATE_HIGHATOMIC && | |
2046 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { | |
2047 | zone->nr_reserved_highatomic += pageblock_nr_pages; | |
2048 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
2049 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); | |
2050 | } | |
2051 | ||
2052 | out_unlock: | |
2053 | spin_unlock_irqrestore(&zone->lock, flags); | |
2054 | } | |
2055 | ||
2056 | /* | |
2057 | * Used when an allocation is about to fail under memory pressure. This | |
2058 | * potentially hurts the reliability of high-order allocations when under | |
2059 | * intense memory pressure but failed atomic allocations should be easier | |
2060 | * to recover from than an OOM. | |
29fac03b MK |
2061 | * |
2062 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2063 | * pageblock is exhausted. | |
0aaa29a5 | 2064 | */ |
29fac03b MK |
2065 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2066 | bool force) | |
0aaa29a5 MG |
2067 | { |
2068 | struct zonelist *zonelist = ac->zonelist; | |
2069 | unsigned long flags; | |
2070 | struct zoneref *z; | |
2071 | struct zone *zone; | |
2072 | struct page *page; | |
2073 | int order; | |
04c8716f | 2074 | bool ret; |
0aaa29a5 MG |
2075 | |
2076 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2077 | ac->nodemask) { | |
29fac03b MK |
2078 | /* |
2079 | * Preserve at least one pageblock unless memory pressure | |
2080 | * is really high. | |
2081 | */ | |
2082 | if (!force && zone->nr_reserved_highatomic <= | |
2083 | pageblock_nr_pages) | |
0aaa29a5 MG |
2084 | continue; |
2085 | ||
2086 | spin_lock_irqsave(&zone->lock, flags); | |
2087 | for (order = 0; order < MAX_ORDER; order++) { | |
2088 | struct free_area *area = &(zone->free_area[order]); | |
2089 | ||
a16601c5 GT |
2090 | page = list_first_entry_or_null( |
2091 | &area->free_list[MIGRATE_HIGHATOMIC], | |
2092 | struct page, lru); | |
2093 | if (!page) | |
0aaa29a5 MG |
2094 | continue; |
2095 | ||
0aaa29a5 | 2096 | /* |
4855e4a7 MK |
2097 | * In page freeing path, migratetype change is racy so |
2098 | * we can counter several free pages in a pageblock | |
2099 | * in this loop althoug we changed the pageblock type | |
2100 | * from highatomic to ac->migratetype. So we should | |
2101 | * adjust the count once. | |
0aaa29a5 | 2102 | */ |
4855e4a7 MK |
2103 | if (get_pageblock_migratetype(page) == |
2104 | MIGRATE_HIGHATOMIC) { | |
2105 | /* | |
2106 | * It should never happen but changes to | |
2107 | * locking could inadvertently allow a per-cpu | |
2108 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2109 | * while unreserving so be safe and watch for | |
2110 | * underflows. | |
2111 | */ | |
2112 | zone->nr_reserved_highatomic -= min( | |
2113 | pageblock_nr_pages, | |
2114 | zone->nr_reserved_highatomic); | |
2115 | } | |
0aaa29a5 MG |
2116 | |
2117 | /* | |
2118 | * Convert to ac->migratetype and avoid the normal | |
2119 | * pageblock stealing heuristics. Minimally, the caller | |
2120 | * is doing the work and needs the pages. More | |
2121 | * importantly, if the block was always converted to | |
2122 | * MIGRATE_UNMOVABLE or another type then the number | |
2123 | * of pageblocks that cannot be completely freed | |
2124 | * may increase. | |
2125 | */ | |
2126 | set_pageblock_migratetype(page, ac->migratetype); | |
04c8716f | 2127 | ret = move_freepages_block(zone, page, ac->migratetype); |
29fac03b MK |
2128 | if (ret) { |
2129 | spin_unlock_irqrestore(&zone->lock, flags); | |
2130 | return ret; | |
2131 | } | |
0aaa29a5 MG |
2132 | } |
2133 | spin_unlock_irqrestore(&zone->lock, flags); | |
2134 | } | |
04c8716f MK |
2135 | |
2136 | return false; | |
0aaa29a5 MG |
2137 | } |
2138 | ||
b2a0ac88 | 2139 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 2140 | static inline struct page * |
7aeb09f9 | 2141 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 2142 | { |
b8af2941 | 2143 | struct free_area *area; |
7aeb09f9 | 2144 | unsigned int current_order; |
b2a0ac88 | 2145 | struct page *page; |
4eb7dce6 JK |
2146 | int fallback_mt; |
2147 | bool can_steal; | |
b2a0ac88 MG |
2148 | |
2149 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
2150 | for (current_order = MAX_ORDER-1; |
2151 | current_order >= order && current_order <= MAX_ORDER-1; | |
2152 | --current_order) { | |
4eb7dce6 JK |
2153 | area = &(zone->free_area[current_order]); |
2154 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2155 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2156 | if (fallback_mt == -1) |
2157 | continue; | |
b2a0ac88 | 2158 | |
a16601c5 | 2159 | page = list_first_entry(&area->free_list[fallback_mt], |
4eb7dce6 | 2160 | struct page, lru); |
88ed365e MK |
2161 | if (can_steal && |
2162 | get_pageblock_migratetype(page) != MIGRATE_HIGHATOMIC) | |
4eb7dce6 | 2163 | steal_suitable_fallback(zone, page, start_migratetype); |
b2a0ac88 | 2164 | |
4eb7dce6 JK |
2165 | /* Remove the page from the freelists */ |
2166 | area->nr_free--; | |
2167 | list_del(&page->lru); | |
2168 | rmv_page_order(page); | |
3a1086fb | 2169 | |
4eb7dce6 JK |
2170 | expand(zone, page, order, current_order, area, |
2171 | start_migratetype); | |
2172 | /* | |
bb14c2c7 | 2173 | * The pcppage_migratetype may differ from pageblock's |
4eb7dce6 | 2174 | * migratetype depending on the decisions in |
bb14c2c7 VB |
2175 | * find_suitable_fallback(). This is OK as long as it does not |
2176 | * differ for MIGRATE_CMA pageblocks. Those can be used as | |
2177 | * fallback only via special __rmqueue_cma_fallback() function | |
4eb7dce6 | 2178 | */ |
bb14c2c7 | 2179 | set_pcppage_migratetype(page, start_migratetype); |
e0fff1bd | 2180 | |
4eb7dce6 JK |
2181 | trace_mm_page_alloc_extfrag(page, order, current_order, |
2182 | start_migratetype, fallback_mt); | |
e0fff1bd | 2183 | |
4eb7dce6 | 2184 | return page; |
b2a0ac88 MG |
2185 | } |
2186 | ||
728ec980 | 2187 | return NULL; |
b2a0ac88 MG |
2188 | } |
2189 | ||
56fd56b8 | 2190 | /* |
1da177e4 LT |
2191 | * Do the hard work of removing an element from the buddy allocator. |
2192 | * Call me with the zone->lock already held. | |
2193 | */ | |
b2a0ac88 | 2194 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
6ac0206b | 2195 | int migratetype) |
1da177e4 | 2196 | { |
1da177e4 LT |
2197 | struct page *page; |
2198 | ||
56fd56b8 | 2199 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2200 | if (unlikely(!page)) { |
dc67647b JK |
2201 | if (migratetype == MIGRATE_MOVABLE) |
2202 | page = __rmqueue_cma_fallback(zone, order); | |
2203 | ||
2204 | if (!page) | |
2205 | page = __rmqueue_fallback(zone, order, migratetype); | |
728ec980 MG |
2206 | } |
2207 | ||
0d3d062a | 2208 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2209 | return page; |
1da177e4 LT |
2210 | } |
2211 | ||
5f63b720 | 2212 | /* |
1da177e4 LT |
2213 | * Obtain a specified number of elements from the buddy allocator, all under |
2214 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2215 | * Returns the number of new pages which were placed at *list. | |
2216 | */ | |
5f63b720 | 2217 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2218 | unsigned long count, struct list_head *list, |
b745bc85 | 2219 | int migratetype, bool cold) |
1da177e4 | 2220 | { |
a6de734b | 2221 | int i, alloced = 0; |
5f63b720 | 2222 | |
c54ad30c | 2223 | spin_lock(&zone->lock); |
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)); |
c54ad30c | 2259 | spin_unlock(&zone->lock); |
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 | ||
2344 | /* | |
74046494 GBY |
2345 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2346 | * | |
93481ff0 VB |
2347 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2348 | * | |
74046494 GBY |
2349 | * Note that this code is protected against sending an IPI to an offline |
2350 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
2351 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
2352 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
2353 | * before the call to on_each_cpu_mask(). | |
9f8f2172 | 2354 | */ |
93481ff0 | 2355 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2356 | { |
74046494 | 2357 | int cpu; |
74046494 GBY |
2358 | |
2359 | /* | |
2360 | * Allocate in the BSS so we wont require allocation in | |
2361 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2362 | */ | |
2363 | static cpumask_t cpus_with_pcps; | |
2364 | ||
2365 | /* | |
2366 | * We don't care about racing with CPU hotplug event | |
2367 | * as offline notification will cause the notified | |
2368 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2369 | * disables preemption as part of its processing | |
2370 | */ | |
2371 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2372 | struct per_cpu_pageset *pcp; |
2373 | struct zone *z; | |
74046494 | 2374 | bool has_pcps = false; |
93481ff0 VB |
2375 | |
2376 | if (zone) { | |
74046494 | 2377 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2378 | if (pcp->pcp.count) |
74046494 | 2379 | has_pcps = true; |
93481ff0 VB |
2380 | } else { |
2381 | for_each_populated_zone(z) { | |
2382 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2383 | if (pcp->pcp.count) { | |
2384 | has_pcps = true; | |
2385 | break; | |
2386 | } | |
74046494 GBY |
2387 | } |
2388 | } | |
93481ff0 | 2389 | |
74046494 GBY |
2390 | if (has_pcps) |
2391 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2392 | else | |
2393 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2394 | } | |
93481ff0 VB |
2395 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
2396 | zone, 1); | |
9f8f2172 CL |
2397 | } |
2398 | ||
296699de | 2399 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
2400 | |
2401 | void mark_free_pages(struct zone *zone) | |
2402 | { | |
f623f0db RW |
2403 | unsigned long pfn, max_zone_pfn; |
2404 | unsigned long flags; | |
7aeb09f9 | 2405 | unsigned int order, t; |
86760a2c | 2406 | struct page *page; |
1da177e4 | 2407 | |
8080fc03 | 2408 | if (zone_is_empty(zone)) |
1da177e4 LT |
2409 | return; |
2410 | ||
2411 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2412 | |
108bcc96 | 2413 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2414 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2415 | if (pfn_valid(pfn)) { | |
86760a2c | 2416 | page = pfn_to_page(pfn); |
ba6b0979 JK |
2417 | |
2418 | if (page_zone(page) != zone) | |
2419 | continue; | |
2420 | ||
7be98234 RW |
2421 | if (!swsusp_page_is_forbidden(page)) |
2422 | swsusp_unset_page_free(page); | |
f623f0db | 2423 | } |
1da177e4 | 2424 | |
b2a0ac88 | 2425 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2426 | list_for_each_entry(page, |
2427 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2428 | unsigned long i; |
1da177e4 | 2429 | |
86760a2c | 2430 | pfn = page_to_pfn(page); |
f623f0db | 2431 | for (i = 0; i < (1UL << order); i++) |
7be98234 | 2432 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 2433 | } |
b2a0ac88 | 2434 | } |
1da177e4 LT |
2435 | spin_unlock_irqrestore(&zone->lock, flags); |
2436 | } | |
e2c55dc8 | 2437 | #endif /* CONFIG_PM */ |
1da177e4 | 2438 | |
1da177e4 LT |
2439 | /* |
2440 | * Free a 0-order page | |
b745bc85 | 2441 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 2442 | */ |
b745bc85 | 2443 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
2444 | { |
2445 | struct zone *zone = page_zone(page); | |
2446 | struct per_cpu_pages *pcp; | |
2447 | unsigned long flags; | |
dc4b0caf | 2448 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 2449 | int migratetype; |
1da177e4 | 2450 | |
4db7548c | 2451 | if (!free_pcp_prepare(page)) |
689bcebf HD |
2452 | return; |
2453 | ||
dc4b0caf | 2454 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2455 | set_pcppage_migratetype(page, migratetype); |
1da177e4 | 2456 | local_irq_save(flags); |
f8891e5e | 2457 | __count_vm_event(PGFREE); |
da456f14 | 2458 | |
5f8dcc21 MG |
2459 | /* |
2460 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2461 | * Free ISOLATE pages back to the allocator because they are being | |
2462 | * offlined but treat RESERVE as movable pages so we can get those | |
2463 | * areas back if necessary. Otherwise, we may have to free | |
2464 | * excessively into the page allocator | |
2465 | */ | |
2466 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2467 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2468 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
2469 | goto out; |
2470 | } | |
2471 | migratetype = MIGRATE_MOVABLE; | |
2472 | } | |
2473 | ||
99dcc3e5 | 2474 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 2475 | if (!cold) |
5f8dcc21 | 2476 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
2477 | else |
2478 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 2479 | pcp->count++; |
48db57f8 | 2480 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2481 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
2482 | free_pcppages_bulk(zone, batch, pcp); |
2483 | pcp->count -= batch; | |
48db57f8 | 2484 | } |
5f8dcc21 MG |
2485 | |
2486 | out: | |
1da177e4 | 2487 | local_irq_restore(flags); |
1da177e4 LT |
2488 | } |
2489 | ||
cc59850e KK |
2490 | /* |
2491 | * Free a list of 0-order pages | |
2492 | */ | |
b745bc85 | 2493 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
2494 | { |
2495 | struct page *page, *next; | |
2496 | ||
2497 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 2498 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
2499 | free_hot_cold_page(page, cold); |
2500 | } | |
2501 | } | |
2502 | ||
8dfcc9ba NP |
2503 | /* |
2504 | * split_page takes a non-compound higher-order page, and splits it into | |
2505 | * n (1<<order) sub-pages: page[0..n] | |
2506 | * Each sub-page must be freed individually. | |
2507 | * | |
2508 | * Note: this is probably too low level an operation for use in drivers. | |
2509 | * Please consult with lkml before using this in your driver. | |
2510 | */ | |
2511 | void split_page(struct page *page, unsigned int order) | |
2512 | { | |
2513 | int i; | |
2514 | ||
309381fe SL |
2515 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2516 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
2517 | |
2518 | #ifdef CONFIG_KMEMCHECK | |
2519 | /* | |
2520 | * Split shadow pages too, because free(page[0]) would | |
2521 | * otherwise free the whole shadow. | |
2522 | */ | |
2523 | if (kmemcheck_page_is_tracked(page)) | |
2524 | split_page(virt_to_page(page[0].shadow), order); | |
2525 | #endif | |
2526 | ||
a9627bc5 | 2527 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2528 | set_page_refcounted(page + i); |
a9627bc5 | 2529 | split_page_owner(page, order); |
8dfcc9ba | 2530 | } |
5853ff23 | 2531 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2532 | |
3c605096 | 2533 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2534 | { |
748446bb MG |
2535 | unsigned long watermark; |
2536 | struct zone *zone; | |
2139cbe6 | 2537 | int mt; |
748446bb MG |
2538 | |
2539 | BUG_ON(!PageBuddy(page)); | |
2540 | ||
2541 | zone = page_zone(page); | |
2e30abd1 | 2542 | mt = get_pageblock_migratetype(page); |
748446bb | 2543 | |
194159fb | 2544 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
2545 | /* |
2546 | * Obey watermarks as if the page was being allocated. We can | |
2547 | * emulate a high-order watermark check with a raised order-0 | |
2548 | * watermark, because we already know our high-order page | |
2549 | * exists. | |
2550 | */ | |
2551 | watermark = min_wmark_pages(zone) + (1UL << order); | |
984fdba6 | 2552 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
2553 | return 0; |
2554 | ||
8fb74b9f | 2555 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2556 | } |
748446bb MG |
2557 | |
2558 | /* Remove page from free list */ | |
2559 | list_del(&page->lru); | |
2560 | zone->free_area[order].nr_free--; | |
2561 | rmv_page_order(page); | |
2139cbe6 | 2562 | |
400bc7fd | 2563 | /* |
2564 | * Set the pageblock if the isolated page is at least half of a | |
2565 | * pageblock | |
2566 | */ | |
748446bb MG |
2567 | if (order >= pageblock_order - 1) { |
2568 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2569 | for (; page < endpage; page += pageblock_nr_pages) { |
2570 | int mt = get_pageblock_migratetype(page); | |
88ed365e MK |
2571 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
2572 | && mt != MIGRATE_HIGHATOMIC) | |
47118af0 MN |
2573 | set_pageblock_migratetype(page, |
2574 | MIGRATE_MOVABLE); | |
2575 | } | |
748446bb MG |
2576 | } |
2577 | ||
f3a14ced | 2578 | |
8fb74b9f | 2579 | return 1UL << order; |
1fb3f8ca MG |
2580 | } |
2581 | ||
060e7417 MG |
2582 | /* |
2583 | * Update NUMA hit/miss statistics | |
2584 | * | |
2585 | * Must be called with interrupts disabled. | |
060e7417 | 2586 | */ |
41b6167e | 2587 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
2588 | { |
2589 | #ifdef CONFIG_NUMA | |
060e7417 MG |
2590 | enum zone_stat_item local_stat = NUMA_LOCAL; |
2591 | ||
2df26639 | 2592 | if (z->node != numa_node_id()) |
060e7417 | 2593 | local_stat = NUMA_OTHER; |
060e7417 | 2594 | |
2df26639 | 2595 | if (z->node == preferred_zone->node) |
060e7417 | 2596 | __inc_zone_state(z, NUMA_HIT); |
2df26639 | 2597 | else { |
060e7417 MG |
2598 | __inc_zone_state(z, NUMA_MISS); |
2599 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); | |
2600 | } | |
2df26639 | 2601 | __inc_zone_state(z, local_stat); |
060e7417 MG |
2602 | #endif |
2603 | } | |
2604 | ||
1da177e4 | 2605 | /* |
75379191 | 2606 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2607 | */ |
0a15c3e9 MG |
2608 | static inline |
2609 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
7aeb09f9 | 2610 | struct zone *zone, unsigned int order, |
c603844b MG |
2611 | gfp_t gfp_flags, unsigned int alloc_flags, |
2612 | int migratetype) | |
1da177e4 LT |
2613 | { |
2614 | unsigned long flags; | |
689bcebf | 2615 | struct page *page; |
b745bc85 | 2616 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
1da177e4 | 2617 | |
48db57f8 | 2618 | if (likely(order == 0)) { |
1da177e4 | 2619 | struct per_cpu_pages *pcp; |
5f8dcc21 | 2620 | struct list_head *list; |
1da177e4 | 2621 | |
1da177e4 | 2622 | local_irq_save(flags); |
479f854a MG |
2623 | do { |
2624 | pcp = &this_cpu_ptr(zone->pageset)->pcp; | |
2625 | list = &pcp->lists[migratetype]; | |
2626 | if (list_empty(list)) { | |
2627 | pcp->count += rmqueue_bulk(zone, 0, | |
2628 | pcp->batch, list, | |
2629 | migratetype, cold); | |
2630 | if (unlikely(list_empty(list))) | |
2631 | goto failed; | |
2632 | } | |
b92a6edd | 2633 | |
479f854a MG |
2634 | if (cold) |
2635 | page = list_last_entry(list, struct page, lru); | |
2636 | else | |
2637 | page = list_first_entry(list, struct page, lru); | |
5f8dcc21 | 2638 | |
83b9355b VB |
2639 | list_del(&page->lru); |
2640 | pcp->count--; | |
2641 | ||
2642 | } while (check_new_pcp(page)); | |
7fb1d9fc | 2643 | } else { |
0f352e53 MH |
2644 | /* |
2645 | * We most definitely don't want callers attempting to | |
2646 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2647 | */ | |
2648 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
1da177e4 | 2649 | spin_lock_irqsave(&zone->lock, flags); |
0aaa29a5 | 2650 | |
479f854a MG |
2651 | do { |
2652 | page = NULL; | |
2653 | if (alloc_flags & ALLOC_HARDER) { | |
2654 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2655 | if (page) | |
2656 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
2657 | } | |
2658 | if (!page) | |
2659 | page = __rmqueue(zone, order, migratetype); | |
2660 | } while (page && check_new_pages(page, order)); | |
a74609fa NP |
2661 | spin_unlock(&zone->lock); |
2662 | if (!page) | |
2663 | goto failed; | |
d1ce749a | 2664 | __mod_zone_freepage_state(zone, -(1 << order), |
bb14c2c7 | 2665 | get_pcppage_migratetype(page)); |
1da177e4 LT |
2666 | } |
2667 | ||
16709d1d | 2668 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 2669 | zone_statistics(preferred_zone, zone); |
a74609fa | 2670 | local_irq_restore(flags); |
1da177e4 | 2671 | |
309381fe | 2672 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 2673 | return page; |
a74609fa NP |
2674 | |
2675 | failed: | |
2676 | local_irq_restore(flags); | |
a74609fa | 2677 | return NULL; |
1da177e4 LT |
2678 | } |
2679 | ||
933e312e AM |
2680 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2681 | ||
b2588c4b | 2682 | static struct { |
933e312e AM |
2683 | struct fault_attr attr; |
2684 | ||
621a5f7a | 2685 | bool ignore_gfp_highmem; |
71baba4b | 2686 | bool ignore_gfp_reclaim; |
54114994 | 2687 | u32 min_order; |
933e312e AM |
2688 | } fail_page_alloc = { |
2689 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 2690 | .ignore_gfp_reclaim = true, |
621a5f7a | 2691 | .ignore_gfp_highmem = true, |
54114994 | 2692 | .min_order = 1, |
933e312e AM |
2693 | }; |
2694 | ||
2695 | static int __init setup_fail_page_alloc(char *str) | |
2696 | { | |
2697 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
2698 | } | |
2699 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
2700 | ||
deaf386e | 2701 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2702 | { |
54114994 | 2703 | if (order < fail_page_alloc.min_order) |
deaf386e | 2704 | return false; |
933e312e | 2705 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 2706 | return false; |
933e312e | 2707 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 2708 | return false; |
71baba4b MG |
2709 | if (fail_page_alloc.ignore_gfp_reclaim && |
2710 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 2711 | return false; |
933e312e AM |
2712 | |
2713 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
2714 | } | |
2715 | ||
2716 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
2717 | ||
2718 | static int __init fail_page_alloc_debugfs(void) | |
2719 | { | |
f4ae40a6 | 2720 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 2721 | struct dentry *dir; |
933e312e | 2722 | |
dd48c085 AM |
2723 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2724 | &fail_page_alloc.attr); | |
2725 | if (IS_ERR(dir)) | |
2726 | return PTR_ERR(dir); | |
933e312e | 2727 | |
b2588c4b | 2728 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 2729 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
2730 | goto fail; |
2731 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
2732 | &fail_page_alloc.ignore_gfp_highmem)) | |
2733 | goto fail; | |
2734 | if (!debugfs_create_u32("min-order", mode, dir, | |
2735 | &fail_page_alloc.min_order)) | |
2736 | goto fail; | |
2737 | ||
2738 | return 0; | |
2739 | fail: | |
dd48c085 | 2740 | debugfs_remove_recursive(dir); |
933e312e | 2741 | |
b2588c4b | 2742 | return -ENOMEM; |
933e312e AM |
2743 | } |
2744 | ||
2745 | late_initcall(fail_page_alloc_debugfs); | |
2746 | ||
2747 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2748 | ||
2749 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2750 | ||
deaf386e | 2751 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2752 | { |
deaf386e | 2753 | return false; |
933e312e AM |
2754 | } |
2755 | ||
2756 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2757 | ||
1da177e4 | 2758 | /* |
97a16fc8 MG |
2759 | * Return true if free base pages are above 'mark'. For high-order checks it |
2760 | * will return true of the order-0 watermark is reached and there is at least | |
2761 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
2762 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 2763 | */ |
86a294a8 MH |
2764 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
2765 | int classzone_idx, unsigned int alloc_flags, | |
2766 | long free_pages) | |
1da177e4 | 2767 | { |
d23ad423 | 2768 | long min = mark; |
1da177e4 | 2769 | int o; |
c603844b | 2770 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
1da177e4 | 2771 | |
0aaa29a5 | 2772 | /* free_pages may go negative - that's OK */ |
df0a6daa | 2773 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 2774 | |
7fb1d9fc | 2775 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2776 | min -= min / 2; |
0aaa29a5 MG |
2777 | |
2778 | /* | |
2779 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
2780 | * the high-atomic reserves. This will over-estimate the size of the | |
2781 | * atomic reserve but it avoids a search. | |
2782 | */ | |
97a16fc8 | 2783 | if (likely(!alloc_harder)) |
0aaa29a5 MG |
2784 | free_pages -= z->nr_reserved_highatomic; |
2785 | else | |
1da177e4 | 2786 | min -= min / 4; |
e2b19197 | 2787 | |
d95ea5d1 BZ |
2788 | #ifdef CONFIG_CMA |
2789 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2790 | if (!(alloc_flags & ALLOC_CMA)) | |
97a16fc8 | 2791 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2792 | #endif |
026b0814 | 2793 | |
97a16fc8 MG |
2794 | /* |
2795 | * Check watermarks for an order-0 allocation request. If these | |
2796 | * are not met, then a high-order request also cannot go ahead | |
2797 | * even if a suitable page happened to be free. | |
2798 | */ | |
2799 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 2800 | return false; |
1da177e4 | 2801 | |
97a16fc8 MG |
2802 | /* If this is an order-0 request then the watermark is fine */ |
2803 | if (!order) | |
2804 | return true; | |
2805 | ||
2806 | /* For a high-order request, check at least one suitable page is free */ | |
2807 | for (o = order; o < MAX_ORDER; o++) { | |
2808 | struct free_area *area = &z->free_area[o]; | |
2809 | int mt; | |
2810 | ||
2811 | if (!area->nr_free) | |
2812 | continue; | |
2813 | ||
2814 | if (alloc_harder) | |
2815 | return true; | |
1da177e4 | 2816 | |
97a16fc8 MG |
2817 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
2818 | if (!list_empty(&area->free_list[mt])) | |
2819 | return true; | |
2820 | } | |
2821 | ||
2822 | #ifdef CONFIG_CMA | |
2823 | if ((alloc_flags & ALLOC_CMA) && | |
2824 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
2825 | return true; | |
2826 | } | |
2827 | #endif | |
1da177e4 | 2828 | } |
97a16fc8 | 2829 | return false; |
88f5acf8 MG |
2830 | } |
2831 | ||
7aeb09f9 | 2832 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 2833 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
2834 | { |
2835 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2836 | zone_page_state(z, NR_FREE_PAGES)); | |
2837 | } | |
2838 | ||
48ee5f36 MG |
2839 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
2840 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
2841 | { | |
2842 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2843 | long cma_pages = 0; | |
2844 | ||
2845 | #ifdef CONFIG_CMA | |
2846 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2847 | if (!(alloc_flags & ALLOC_CMA)) | |
2848 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
2849 | #endif | |
2850 | ||
2851 | /* | |
2852 | * Fast check for order-0 only. If this fails then the reserves | |
2853 | * need to be calculated. There is a corner case where the check | |
2854 | * passes but only the high-order atomic reserve are free. If | |
2855 | * the caller is !atomic then it'll uselessly search the free | |
2856 | * list. That corner case is then slower but it is harmless. | |
2857 | */ | |
2858 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) | |
2859 | return true; | |
2860 | ||
2861 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2862 | free_pages); | |
2863 | } | |
2864 | ||
7aeb09f9 | 2865 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 2866 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
2867 | { |
2868 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2869 | ||
2870 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2871 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2872 | ||
e2b19197 | 2873 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 2874 | free_pages); |
1da177e4 LT |
2875 | } |
2876 | ||
9276b1bc | 2877 | #ifdef CONFIG_NUMA |
957f822a DR |
2878 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2879 | { | |
5f7a75ac MG |
2880 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
2881 | RECLAIM_DISTANCE; | |
957f822a | 2882 | } |
9276b1bc | 2883 | #else /* CONFIG_NUMA */ |
957f822a DR |
2884 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2885 | { | |
2886 | return true; | |
2887 | } | |
9276b1bc PJ |
2888 | #endif /* CONFIG_NUMA */ |
2889 | ||
7fb1d9fc | 2890 | /* |
0798e519 | 2891 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
2892 | * a page. |
2893 | */ | |
2894 | static struct page * | |
a9263751 VB |
2895 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
2896 | const struct alloc_context *ac) | |
753ee728 | 2897 | { |
c33d6c06 | 2898 | struct zoneref *z = ac->preferred_zoneref; |
5117f45d | 2899 | struct zone *zone; |
3b8c0be4 MG |
2900 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
2901 | ||
7fb1d9fc | 2902 | /* |
9276b1bc | 2903 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 2904 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 2905 | */ |
c33d6c06 | 2906 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 2907 | ac->nodemask) { |
be06af00 | 2908 | struct page *page; |
e085dbc5 JW |
2909 | unsigned long mark; |
2910 | ||
664eedde MG |
2911 | if (cpusets_enabled() && |
2912 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 2913 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 2914 | continue; |
a756cf59 JW |
2915 | /* |
2916 | * When allocating a page cache page for writing, we | |
281e3726 MG |
2917 | * want to get it from a node that is within its dirty |
2918 | * limit, such that no single node holds more than its | |
a756cf59 | 2919 | * proportional share of globally allowed dirty pages. |
281e3726 | 2920 | * The dirty limits take into account the node's |
a756cf59 JW |
2921 | * lowmem reserves and high watermark so that kswapd |
2922 | * should be able to balance it without having to | |
2923 | * write pages from its LRU list. | |
2924 | * | |
a756cf59 | 2925 | * XXX: For now, allow allocations to potentially |
281e3726 | 2926 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 2927 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 2928 | * which is important when on a NUMA setup the allowed |
281e3726 | 2929 | * nodes are together not big enough to reach the |
a756cf59 | 2930 | * global limit. The proper fix for these situations |
281e3726 | 2931 | * will require awareness of nodes in the |
a756cf59 JW |
2932 | * dirty-throttling and the flusher threads. |
2933 | */ | |
3b8c0be4 MG |
2934 | if (ac->spread_dirty_pages) { |
2935 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
2936 | continue; | |
2937 | ||
2938 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
2939 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
2940 | continue; | |
2941 | } | |
2942 | } | |
7fb1d9fc | 2943 | |
e085dbc5 | 2944 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 2945 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 2946 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
2947 | int ret; |
2948 | ||
5dab2911 MG |
2949 | /* Checked here to keep the fast path fast */ |
2950 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
2951 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
2952 | goto try_this_zone; | |
2953 | ||
a5f5f91d | 2954 | if (node_reclaim_mode == 0 || |
c33d6c06 | 2955 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
2956 | continue; |
2957 | ||
a5f5f91d | 2958 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 2959 | switch (ret) { |
a5f5f91d | 2960 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 2961 | /* did not scan */ |
cd38b115 | 2962 | continue; |
a5f5f91d | 2963 | case NODE_RECLAIM_FULL: |
fa5e084e | 2964 | /* scanned but unreclaimable */ |
cd38b115 | 2965 | continue; |
fa5e084e MG |
2966 | default: |
2967 | /* did we reclaim enough */ | |
fed2719e | 2968 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 2969 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
2970 | goto try_this_zone; |
2971 | ||
fed2719e | 2972 | continue; |
0798e519 | 2973 | } |
7fb1d9fc RS |
2974 | } |
2975 | ||
fa5e084e | 2976 | try_this_zone: |
c33d6c06 | 2977 | page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 2978 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 2979 | if (page) { |
479f854a | 2980 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
2981 | |
2982 | /* | |
2983 | * If this is a high-order atomic allocation then check | |
2984 | * if the pageblock should be reserved for the future | |
2985 | */ | |
2986 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
2987 | reserve_highatomic_pageblock(page, zone, order); | |
2988 | ||
75379191 VB |
2989 | return page; |
2990 | } | |
54a6eb5c | 2991 | } |
9276b1bc | 2992 | |
4ffeaf35 | 2993 | return NULL; |
753ee728 MH |
2994 | } |
2995 | ||
29423e77 DR |
2996 | /* |
2997 | * Large machines with many possible nodes should not always dump per-node | |
2998 | * meminfo in irq context. | |
2999 | */ | |
3000 | static inline bool should_suppress_show_mem(void) | |
3001 | { | |
3002 | bool ret = false; | |
3003 | ||
3004 | #if NODES_SHIFT > 8 | |
3005 | ret = in_interrupt(); | |
3006 | #endif | |
3007 | return ret; | |
3008 | } | |
3009 | ||
a238ab5b DH |
3010 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
3011 | DEFAULT_RATELIMIT_INTERVAL, | |
3012 | DEFAULT_RATELIMIT_BURST); | |
3013 | ||
7877cdcc | 3014 | void warn_alloc(gfp_t gfp_mask, const char *fmt, ...) |
a238ab5b | 3015 | { |
a238ab5b | 3016 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
7877cdcc MH |
3017 | struct va_format vaf; |
3018 | va_list args; | |
a238ab5b | 3019 | |
c0a32fc5 SG |
3020 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
3021 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
3022 | return; |
3023 | ||
3024 | /* | |
3025 | * This documents exceptions given to allocations in certain | |
3026 | * contexts that are allowed to allocate outside current's set | |
3027 | * of allowed nodes. | |
3028 | */ | |
3029 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
3030 | if (test_thread_flag(TIF_MEMDIE) || | |
3031 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
3032 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3033 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3034 | filter &= ~SHOW_MEM_FILTER_NODES; |
3035 | ||
7877cdcc | 3036 | pr_warn("%s: ", current->comm); |
3ee9a4f0 | 3037 | |
7877cdcc MH |
3038 | va_start(args, fmt); |
3039 | vaf.fmt = fmt; | |
3040 | vaf.va = &args; | |
3041 | pr_cont("%pV", &vaf); | |
3042 | va_end(args); | |
3ee9a4f0 | 3043 | |
7877cdcc | 3044 | pr_cont(", mode:%#x(%pGg)\n", gfp_mask, &gfp_mask); |
3ee9a4f0 | 3045 | |
a238ab5b DH |
3046 | dump_stack(); |
3047 | if (!should_suppress_show_mem()) | |
3048 | show_mem(filter); | |
3049 | } | |
3050 | ||
11e33f6a MG |
3051 | static inline struct page * |
3052 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3053 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3054 | { |
6e0fc46d DR |
3055 | struct oom_control oc = { |
3056 | .zonelist = ac->zonelist, | |
3057 | .nodemask = ac->nodemask, | |
2a966b77 | 3058 | .memcg = NULL, |
6e0fc46d DR |
3059 | .gfp_mask = gfp_mask, |
3060 | .order = order, | |
6e0fc46d | 3061 | }; |
11e33f6a MG |
3062 | struct page *page; |
3063 | ||
9879de73 JW |
3064 | *did_some_progress = 0; |
3065 | ||
9879de73 | 3066 | /* |
dc56401f JW |
3067 | * Acquire the oom lock. If that fails, somebody else is |
3068 | * making progress for us. | |
9879de73 | 3069 | */ |
dc56401f | 3070 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3071 | *did_some_progress = 1; |
11e33f6a | 3072 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3073 | return NULL; |
3074 | } | |
6b1de916 | 3075 | |
11e33f6a MG |
3076 | /* |
3077 | * Go through the zonelist yet one more time, keep very high watermark | |
3078 | * here, this is only to catch a parallel oom killing, we must fail if | |
3079 | * we're still under heavy pressure. | |
3080 | */ | |
a9263751 VB |
3081 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
3082 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3083 | if (page) |
11e33f6a MG |
3084 | goto out; |
3085 | ||
4365a567 | 3086 | if (!(gfp_mask & __GFP_NOFAIL)) { |
9879de73 JW |
3087 | /* Coredumps can quickly deplete all memory reserves */ |
3088 | if (current->flags & PF_DUMPCORE) | |
3089 | goto out; | |
4365a567 KH |
3090 | /* The OOM killer will not help higher order allocs */ |
3091 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3092 | goto out; | |
03668b3c | 3093 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
a9263751 | 3094 | if (ac->high_zoneidx < ZONE_NORMAL) |
03668b3c | 3095 | goto out; |
9083905a JW |
3096 | if (pm_suspended_storage()) |
3097 | goto out; | |
3da88fb3 MH |
3098 | /* |
3099 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3100 | * other request to make a forward progress. | |
3101 | * We are in an unfortunate situation where out_of_memory cannot | |
3102 | * do much for this context but let's try it to at least get | |
3103 | * access to memory reserved if the current task is killed (see | |
3104 | * out_of_memory). Once filesystems are ready to handle allocation | |
3105 | * failures more gracefully we should just bail out here. | |
3106 | */ | |
3107 | ||
4167e9b2 | 3108 | /* The OOM killer may not free memory on a specific node */ |
4365a567 KH |
3109 | if (gfp_mask & __GFP_THISNODE) |
3110 | goto out; | |
3111 | } | |
11e33f6a | 3112 | /* Exhausted what can be done so it's blamo time */ |
5020e285 | 3113 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3114 | *did_some_progress = 1; |
5020e285 MH |
3115 | |
3116 | if (gfp_mask & __GFP_NOFAIL) { | |
3117 | page = get_page_from_freelist(gfp_mask, order, | |
3118 | ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac); | |
3119 | /* | |
3120 | * fallback to ignore cpuset restriction if our nodes | |
3121 | * are depleted | |
3122 | */ | |
3123 | if (!page) | |
3124 | page = get_page_from_freelist(gfp_mask, order, | |
3125 | ALLOC_NO_WATERMARKS, ac); | |
3126 | } | |
3127 | } | |
11e33f6a | 3128 | out: |
dc56401f | 3129 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3130 | return page; |
3131 | } | |
3132 | ||
33c2d214 MH |
3133 | /* |
3134 | * Maximum number of compaction retries wit a progress before OOM | |
3135 | * killer is consider as the only way to move forward. | |
3136 | */ | |
3137 | #define MAX_COMPACT_RETRIES 16 | |
3138 | ||
56de7263 MG |
3139 | #ifdef CONFIG_COMPACTION |
3140 | /* Try memory compaction for high-order allocations before reclaim */ | |
3141 | static struct page * | |
3142 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3143 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3144 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3145 | { |
98dd3b48 | 3146 | struct page *page; |
53853e2d VB |
3147 | |
3148 | if (!order) | |
66199712 | 3149 | return NULL; |
66199712 | 3150 | |
c06b1fca | 3151 | current->flags |= PF_MEMALLOC; |
c5d01d0d | 3152 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
c3486f53 | 3153 | prio); |
c06b1fca | 3154 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 3155 | |
c5d01d0d | 3156 | if (*compact_result <= COMPACT_INACTIVE) |
98dd3b48 | 3157 | return NULL; |
53853e2d | 3158 | |
98dd3b48 VB |
3159 | /* |
3160 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3161 | * count a compaction stall | |
3162 | */ | |
3163 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3164 | |
31a6c190 | 3165 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
53853e2d | 3166 | |
98dd3b48 VB |
3167 | if (page) { |
3168 | struct zone *zone = page_zone(page); | |
53853e2d | 3169 | |
98dd3b48 VB |
3170 | zone->compact_blockskip_flush = false; |
3171 | compaction_defer_reset(zone, order, true); | |
3172 | count_vm_event(COMPACTSUCCESS); | |
3173 | return page; | |
3174 | } | |
56de7263 | 3175 | |
98dd3b48 VB |
3176 | /* |
3177 | * It's bad if compaction run occurs and fails. The most likely reason | |
3178 | * is that pages exist, but not enough to satisfy watermarks. | |
3179 | */ | |
3180 | count_vm_event(COMPACTFAIL); | |
66199712 | 3181 | |
98dd3b48 | 3182 | cond_resched(); |
56de7263 MG |
3183 | |
3184 | return NULL; | |
3185 | } | |
33c2d214 | 3186 | |
3250845d VB |
3187 | static inline bool |
3188 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3189 | enum compact_result compact_result, | |
3190 | enum compact_priority *compact_priority, | |
d9436498 | 3191 | int *compaction_retries) |
3250845d VB |
3192 | { |
3193 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3194 | int min_priority; |
3250845d VB |
3195 | |
3196 | if (!order) | |
3197 | return false; | |
3198 | ||
d9436498 VB |
3199 | if (compaction_made_progress(compact_result)) |
3200 | (*compaction_retries)++; | |
3201 | ||
3250845d VB |
3202 | /* |
3203 | * compaction considers all the zone as desperately out of memory | |
3204 | * so it doesn't really make much sense to retry except when the | |
3205 | * failure could be caused by insufficient priority | |
3206 | */ | |
d9436498 VB |
3207 | if (compaction_failed(compact_result)) |
3208 | goto check_priority; | |
3250845d VB |
3209 | |
3210 | /* | |
3211 | * make sure the compaction wasn't deferred or didn't bail out early | |
3212 | * due to locks contention before we declare that we should give up. | |
3213 | * But do not retry if the given zonelist is not suitable for | |
3214 | * compaction. | |
3215 | */ | |
3216 | if (compaction_withdrawn(compact_result)) | |
3217 | return compaction_zonelist_suitable(ac, order, alloc_flags); | |
3218 | ||
3219 | /* | |
3220 | * !costly requests are much more important than __GFP_REPEAT | |
3221 | * costly ones because they are de facto nofail and invoke OOM | |
3222 | * killer to move on while costly can fail and users are ready | |
3223 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3224 | * would need much more detailed feedback from compaction to | |
3225 | * make a better decision. | |
3226 | */ | |
3227 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3228 | max_retries /= 4; | |
d9436498 | 3229 | if (*compaction_retries <= max_retries) |
3250845d VB |
3230 | return true; |
3231 | ||
d9436498 VB |
3232 | /* |
3233 | * Make sure there are attempts at the highest priority if we exhausted | |
3234 | * all retries or failed at the lower priorities. | |
3235 | */ | |
3236 | check_priority: | |
c2033b00 VB |
3237 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3238 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
3239 | if (*compact_priority > min_priority) { | |
d9436498 VB |
3240 | (*compact_priority)--; |
3241 | *compaction_retries = 0; | |
3242 | return true; | |
3243 | } | |
3250845d VB |
3244 | return false; |
3245 | } | |
56de7263 MG |
3246 | #else |
3247 | static inline struct page * | |
3248 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3249 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3250 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3251 | { |
33c2d214 | 3252 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3253 | return NULL; |
3254 | } | |
33c2d214 MH |
3255 | |
3256 | static inline bool | |
86a294a8 MH |
3257 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3258 | enum compact_result compact_result, | |
a5508cd8 | 3259 | enum compact_priority *compact_priority, |
d9436498 | 3260 | int *compaction_retries) |
33c2d214 | 3261 | { |
31e49bfd MH |
3262 | struct zone *zone; |
3263 | struct zoneref *z; | |
3264 | ||
3265 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3266 | return false; | |
3267 | ||
3268 | /* | |
3269 | * There are setups with compaction disabled which would prefer to loop | |
3270 | * inside the allocator rather than hit the oom killer prematurely. | |
3271 | * Let's give them a good hope and keep retrying while the order-0 | |
3272 | * watermarks are OK. | |
3273 | */ | |
3274 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3275 | ac->nodemask) { | |
3276 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
3277 | ac_classzone_idx(ac), alloc_flags)) | |
3278 | return true; | |
3279 | } | |
33c2d214 MH |
3280 | return false; |
3281 | } | |
3250845d | 3282 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3283 | |
bba90710 MS |
3284 | /* Perform direct synchronous page reclaim */ |
3285 | static int | |
a9263751 VB |
3286 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3287 | const struct alloc_context *ac) | |
11e33f6a | 3288 | { |
11e33f6a | 3289 | struct reclaim_state reclaim_state; |
bba90710 | 3290 | int progress; |
11e33f6a MG |
3291 | |
3292 | cond_resched(); | |
3293 | ||
3294 | /* We now go into synchronous reclaim */ | |
3295 | cpuset_memory_pressure_bump(); | |
c06b1fca | 3296 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
3297 | lockdep_set_current_reclaim_state(gfp_mask); |
3298 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 3299 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3300 | |
a9263751 VB |
3301 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3302 | ac->nodemask); | |
11e33f6a | 3303 | |
c06b1fca | 3304 | current->reclaim_state = NULL; |
11e33f6a | 3305 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 3306 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
3307 | |
3308 | cond_resched(); | |
3309 | ||
bba90710 MS |
3310 | return progress; |
3311 | } | |
3312 | ||
3313 | /* The really slow allocator path where we enter direct reclaim */ | |
3314 | static inline struct page * | |
3315 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3316 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3317 | unsigned long *did_some_progress) |
bba90710 MS |
3318 | { |
3319 | struct page *page = NULL; | |
3320 | bool drained = false; | |
3321 | ||
a9263751 | 3322 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3323 | if (unlikely(!(*did_some_progress))) |
3324 | return NULL; | |
11e33f6a | 3325 | |
9ee493ce | 3326 | retry: |
31a6c190 | 3327 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3328 | |
3329 | /* | |
3330 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3331 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3332 | * Shrink them them and try again | |
9ee493ce MG |
3333 | */ |
3334 | if (!page && !drained) { | |
29fac03b | 3335 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3336 | drain_all_pages(NULL); |
9ee493ce MG |
3337 | drained = true; |
3338 | goto retry; | |
3339 | } | |
3340 | ||
11e33f6a MG |
3341 | return page; |
3342 | } | |
3343 | ||
a9263751 | 3344 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
3345 | { |
3346 | struct zoneref *z; | |
3347 | struct zone *zone; | |
e1a55637 | 3348 | pg_data_t *last_pgdat = NULL; |
3a025760 | 3349 | |
a9263751 | 3350 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
e1a55637 MG |
3351 | ac->high_zoneidx, ac->nodemask) { |
3352 | if (last_pgdat != zone->zone_pgdat) | |
52e9f87a | 3353 | wakeup_kswapd(zone, order, ac->high_zoneidx); |
e1a55637 MG |
3354 | last_pgdat = zone->zone_pgdat; |
3355 | } | |
3a025760 JW |
3356 | } |
3357 | ||
c603844b | 3358 | static inline unsigned int |
341ce06f PZ |
3359 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3360 | { | |
c603844b | 3361 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3362 | |
a56f57ff | 3363 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3364 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3365 | |
341ce06f PZ |
3366 | /* |
3367 | * The caller may dip into page reserves a bit more if the caller | |
3368 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3369 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3370 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3371 | */ |
e6223a3b | 3372 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3373 | |
d0164adc | 3374 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3375 | /* |
b104a35d DR |
3376 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3377 | * if it can't schedule. | |
5c3240d9 | 3378 | */ |
b104a35d | 3379 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3380 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3381 | /* |
b104a35d | 3382 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3383 | * comment for __cpuset_node_allowed(). |
523b9458 | 3384 | */ |
341ce06f | 3385 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3386 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3387 | alloc_flags |= ALLOC_HARDER; |
3388 | ||
d95ea5d1 | 3389 | #ifdef CONFIG_CMA |
43e7a34d | 3390 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
3391 | alloc_flags |= ALLOC_CMA; |
3392 | #endif | |
341ce06f PZ |
3393 | return alloc_flags; |
3394 | } | |
3395 | ||
072bb0aa MG |
3396 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3397 | { | |
31a6c190 VB |
3398 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) |
3399 | return false; | |
3400 | ||
3401 | if (gfp_mask & __GFP_MEMALLOC) | |
3402 | return true; | |
3403 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) | |
3404 | return true; | |
3405 | if (!in_interrupt() && | |
3406 | ((current->flags & PF_MEMALLOC) || | |
3407 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
3408 | return true; | |
3409 | ||
3410 | return false; | |
072bb0aa MG |
3411 | } |
3412 | ||
0a0337e0 MH |
3413 | /* |
3414 | * Maximum number of reclaim retries without any progress before OOM killer | |
3415 | * is consider as the only way to move forward. | |
3416 | */ | |
3417 | #define MAX_RECLAIM_RETRIES 16 | |
3418 | ||
3419 | /* | |
3420 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
3421 | * for the given allocation request. | |
3422 | * The reclaim feedback represented by did_some_progress (any progress during | |
7854ea6c MH |
3423 | * the last reclaim round) and no_progress_loops (number of reclaim rounds without |
3424 | * any progress in a row) is considered as well as the reclaimable pages on the | |
3425 | * applicable zone list (with a backoff mechanism which is a function of | |
3426 | * no_progress_loops). | |
0a0337e0 MH |
3427 | * |
3428 | * Returns true if a retry is viable or false to enter the oom path. | |
3429 | */ | |
3430 | static inline bool | |
3431 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
3432 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 3433 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
3434 | { |
3435 | struct zone *zone; | |
3436 | struct zoneref *z; | |
3437 | ||
423b452e VB |
3438 | /* |
3439 | * Costly allocations might have made a progress but this doesn't mean | |
3440 | * their order will become available due to high fragmentation so | |
3441 | * always increment the no progress counter for them | |
3442 | */ | |
3443 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
3444 | *no_progress_loops = 0; | |
3445 | else | |
3446 | (*no_progress_loops)++; | |
3447 | ||
0a0337e0 MH |
3448 | /* |
3449 | * Make sure we converge to OOM if we cannot make any progress | |
3450 | * several times in the row. | |
3451 | */ | |
04c8716f MK |
3452 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
3453 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 3454 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 3455 | } |
0a0337e0 | 3456 | |
bca67592 MG |
3457 | /* |
3458 | * Keep reclaiming pages while there is a chance this will lead | |
3459 | * somewhere. If none of the target zones can satisfy our allocation | |
3460 | * request even if all reclaimable pages are considered then we are | |
3461 | * screwed and have to go OOM. | |
0a0337e0 MH |
3462 | */ |
3463 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3464 | ac->nodemask) { | |
3465 | unsigned long available; | |
ede37713 | 3466 | unsigned long reclaimable; |
0a0337e0 | 3467 | |
5a1c84b4 | 3468 | available = reclaimable = zone_reclaimable_pages(zone); |
423b452e | 3469 | available -= DIV_ROUND_UP((*no_progress_loops) * available, |
0a0337e0 | 3470 | MAX_RECLAIM_RETRIES); |
5a1c84b4 | 3471 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
3472 | |
3473 | /* | |
3474 | * Would the allocation succeed if we reclaimed the whole | |
5a1c84b4 | 3475 | * available? |
0a0337e0 | 3476 | */ |
5a1c84b4 MG |
3477 | if (__zone_watermark_ok(zone, order, min_wmark_pages(zone), |
3478 | ac_classzone_idx(ac), alloc_flags, available)) { | |
ede37713 MH |
3479 | /* |
3480 | * If we didn't make any progress and have a lot of | |
3481 | * dirty + writeback pages then we should wait for | |
3482 | * an IO to complete to slow down the reclaim and | |
3483 | * prevent from pre mature OOM | |
3484 | */ | |
3485 | if (!did_some_progress) { | |
11fb9989 | 3486 | unsigned long write_pending; |
ede37713 | 3487 | |
5a1c84b4 MG |
3488 | write_pending = zone_page_state_snapshot(zone, |
3489 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 3490 | |
11fb9989 | 3491 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
3492 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
3493 | return true; | |
3494 | } | |
3495 | } | |
5a1c84b4 | 3496 | |
ede37713 MH |
3497 | /* |
3498 | * Memory allocation/reclaim might be called from a WQ | |
3499 | * context and the current implementation of the WQ | |
3500 | * concurrency control doesn't recognize that | |
3501 | * a particular WQ is congested if the worker thread is | |
3502 | * looping without ever sleeping. Therefore we have to | |
3503 | * do a short sleep here rather than calling | |
3504 | * cond_resched(). | |
3505 | */ | |
3506 | if (current->flags & PF_WQ_WORKER) | |
3507 | schedule_timeout_uninterruptible(1); | |
3508 | else | |
3509 | cond_resched(); | |
3510 | ||
0a0337e0 MH |
3511 | return true; |
3512 | } | |
3513 | } | |
3514 | ||
3515 | return false; | |
3516 | } | |
3517 | ||
11e33f6a MG |
3518 | static inline struct page * |
3519 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3520 | struct alloc_context *ac) |
11e33f6a | 3521 | { |
d0164adc | 3522 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
11e33f6a | 3523 | struct page *page = NULL; |
c603844b | 3524 | unsigned int alloc_flags; |
11e33f6a | 3525 | unsigned long did_some_progress; |
a5508cd8 | 3526 | enum compact_priority compact_priority = DEF_COMPACT_PRIORITY; |
c5d01d0d | 3527 | enum compact_result compact_result; |
33c2d214 | 3528 | int compaction_retries = 0; |
0a0337e0 | 3529 | int no_progress_loops = 0; |
63f53dea MH |
3530 | unsigned long alloc_start = jiffies; |
3531 | unsigned int stall_timeout = 10 * HZ; | |
1da177e4 | 3532 | |
72807a74 MG |
3533 | /* |
3534 | * In the slowpath, we sanity check order to avoid ever trying to | |
3535 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
3536 | * be using allocators in order of preference for an area that is | |
3537 | * too large. | |
3538 | */ | |
1fc28b70 MG |
3539 | if (order >= MAX_ORDER) { |
3540 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 3541 | return NULL; |
1fc28b70 | 3542 | } |
1da177e4 | 3543 | |
d0164adc MG |
3544 | /* |
3545 | * We also sanity check to catch abuse of atomic reserves being used by | |
3546 | * callers that are not in atomic context. | |
3547 | */ | |
3548 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
3549 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
3550 | gfp_mask &= ~__GFP_ATOMIC; | |
3551 | ||
9bf2229f | 3552 | /* |
31a6c190 VB |
3553 | * The fast path uses conservative alloc_flags to succeed only until |
3554 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
3555 | * alloc_flags precisely. So we do that now. | |
9bf2229f | 3556 | */ |
341ce06f | 3557 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 3558 | |
23771235 VB |
3559 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3560 | wake_all_kswapds(order, ac); | |
3561 | ||
3562 | /* | |
3563 | * The adjusted alloc_flags might result in immediate success, so try | |
3564 | * that first | |
3565 | */ | |
3566 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
3567 | if (page) | |
3568 | goto got_pg; | |
3569 | ||
a8161d1e VB |
3570 | /* |
3571 | * For costly allocations, try direct compaction first, as it's likely | |
3572 | * that we have enough base pages and don't need to reclaim. Don't try | |
3573 | * that for allocations that are allowed to ignore watermarks, as the | |
3574 | * ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
3575 | */ | |
3576 | if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER && | |
3577 | !gfp_pfmemalloc_allowed(gfp_mask)) { | |
3578 | page = __alloc_pages_direct_compact(gfp_mask, order, | |
3579 | alloc_flags, ac, | |
a5508cd8 | 3580 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
3581 | &compact_result); |
3582 | if (page) | |
3583 | goto got_pg; | |
3584 | ||
3eb2771b VB |
3585 | /* |
3586 | * Checks for costly allocations with __GFP_NORETRY, which | |
3587 | * includes THP page fault allocations | |
3588 | */ | |
3589 | if (gfp_mask & __GFP_NORETRY) { | |
a8161d1e VB |
3590 | /* |
3591 | * If compaction is deferred for high-order allocations, | |
3592 | * it is because sync compaction recently failed. If | |
3593 | * this is the case and the caller requested a THP | |
3594 | * allocation, we do not want to heavily disrupt the | |
3595 | * system, so we fail the allocation instead of entering | |
3596 | * direct reclaim. | |
3597 | */ | |
3598 | if (compact_result == COMPACT_DEFERRED) | |
3599 | goto nopage; | |
3600 | ||
a8161d1e | 3601 | /* |
3eb2771b VB |
3602 | * Looks like reclaim/compaction is worth trying, but |
3603 | * sync compaction could be very expensive, so keep | |
25160354 | 3604 | * using async compaction. |
a8161d1e | 3605 | */ |
a5508cd8 | 3606 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
3607 | } |
3608 | } | |
23771235 | 3609 | |
31a6c190 | 3610 | retry: |
23771235 | 3611 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
31a6c190 VB |
3612 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3613 | wake_all_kswapds(order, ac); | |
3614 | ||
23771235 VB |
3615 | if (gfp_pfmemalloc_allowed(gfp_mask)) |
3616 | alloc_flags = ALLOC_NO_WATERMARKS; | |
3617 | ||
e46e7b77 MG |
3618 | /* |
3619 | * Reset the zonelist iterators if memory policies can be ignored. | |
3620 | * These allocations are high priority and system rather than user | |
3621 | * orientated. | |
3622 | */ | |
23771235 | 3623 | if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) { |
e46e7b77 MG |
3624 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
3625 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
3626 | ac->high_zoneidx, ac->nodemask); | |
3627 | } | |
3628 | ||
23771235 | 3629 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 3630 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
3631 | if (page) |
3632 | goto got_pg; | |
1da177e4 | 3633 | |
d0164adc MG |
3634 | /* Caller is not willing to reclaim, we can't balance anything */ |
3635 | if (!can_direct_reclaim) { | |
aed0a0e3 | 3636 | /* |
33d53103 MH |
3637 | * All existing users of the __GFP_NOFAIL are blockable, so warn |
3638 | * of any new users that actually allow this type of allocation | |
3639 | * to fail. | |
aed0a0e3 DR |
3640 | */ |
3641 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 3642 | goto nopage; |
aed0a0e3 | 3643 | } |
1da177e4 | 3644 | |
341ce06f | 3645 | /* Avoid recursion of direct reclaim */ |
33d53103 MH |
3646 | if (current->flags & PF_MEMALLOC) { |
3647 | /* | |
3648 | * __GFP_NOFAIL request from this context is rather bizarre | |
3649 | * because we cannot reclaim anything and only can loop waiting | |
3650 | * for somebody to do a work for us. | |
3651 | */ | |
3652 | if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { | |
3653 | cond_resched(); | |
3654 | goto retry; | |
3655 | } | |
341ce06f | 3656 | goto nopage; |
33d53103 | 3657 | } |
341ce06f | 3658 | |
6583bb64 DR |
3659 | /* Avoid allocations with no watermarks from looping endlessly */ |
3660 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
3661 | goto nopage; | |
3662 | ||
a8161d1e VB |
3663 | |
3664 | /* Try direct reclaim and then allocating */ | |
3665 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
3666 | &did_some_progress); | |
3667 | if (page) | |
3668 | goto got_pg; | |
3669 | ||
3670 | /* Try direct compaction and then allocating */ | |
a9263751 | 3671 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 3672 | compact_priority, &compact_result); |
56de7263 MG |
3673 | if (page) |
3674 | goto got_pg; | |
75f30861 | 3675 | |
9083905a JW |
3676 | /* Do not loop if specifically requested */ |
3677 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 3678 | goto nopage; |
9083905a | 3679 | |
0a0337e0 MH |
3680 | /* |
3681 | * Do not retry costly high order allocations unless they are | |
3682 | * __GFP_REPEAT | |
3683 | */ | |
3684 | if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT)) | |
a8161d1e | 3685 | goto nopage; |
0a0337e0 | 3686 | |
63f53dea MH |
3687 | /* Make sure we know about allocations which stall for too long */ |
3688 | if (time_after(jiffies, alloc_start + stall_timeout)) { | |
3689 | warn_alloc(gfp_mask, | |
9e80c719 | 3690 | "page allocation stalls for %ums, order:%u", |
63f53dea MH |
3691 | jiffies_to_msecs(jiffies-alloc_start), order); |
3692 | stall_timeout += 10 * HZ; | |
3693 | } | |
3694 | ||
0a0337e0 | 3695 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 3696 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
3697 | goto retry; |
3698 | ||
33c2d214 MH |
3699 | /* |
3700 | * It doesn't make any sense to retry for the compaction if the order-0 | |
3701 | * reclaim is not able to make any progress because the current | |
3702 | * implementation of the compaction depends on the sufficient amount | |
3703 | * of free memory (see __compaction_suitable) | |
3704 | */ | |
3705 | if (did_some_progress > 0 && | |
86a294a8 | 3706 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 3707 | compact_result, &compact_priority, |
d9436498 | 3708 | &compaction_retries)) |
33c2d214 MH |
3709 | goto retry; |
3710 | ||
9083905a JW |
3711 | /* Reclaim has failed us, start killing things */ |
3712 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
3713 | if (page) | |
3714 | goto got_pg; | |
3715 | ||
3716 | /* Retry as long as the OOM killer is making progress */ | |
0a0337e0 MH |
3717 | if (did_some_progress) { |
3718 | no_progress_loops = 0; | |
9083905a | 3719 | goto retry; |
0a0337e0 | 3720 | } |
9083905a | 3721 | |
1da177e4 | 3722 | nopage: |
7877cdcc MH |
3723 | warn_alloc(gfp_mask, |
3724 | "page allocation failure: order:%u", order); | |
1da177e4 | 3725 | got_pg: |
072bb0aa | 3726 | return page; |
1da177e4 | 3727 | } |
11e33f6a MG |
3728 | |
3729 | /* | |
3730 | * This is the 'heart' of the zoned buddy allocator. | |
3731 | */ | |
3732 | struct page * | |
3733 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
3734 | struct zonelist *zonelist, nodemask_t *nodemask) | |
3735 | { | |
5bb1b169 | 3736 | struct page *page; |
cc9a6c87 | 3737 | unsigned int cpuset_mems_cookie; |
e6cbd7f2 | 3738 | unsigned int alloc_flags = ALLOC_WMARK_LOW; |
83d4ca81 | 3739 | gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */ |
a9263751 VB |
3740 | struct alloc_context ac = { |
3741 | .high_zoneidx = gfp_zone(gfp_mask), | |
682a3385 | 3742 | .zonelist = zonelist, |
a9263751 VB |
3743 | .nodemask = nodemask, |
3744 | .migratetype = gfpflags_to_migratetype(gfp_mask), | |
3745 | }; | |
11e33f6a | 3746 | |
682a3385 | 3747 | if (cpusets_enabled()) { |
83d4ca81 | 3748 | alloc_mask |= __GFP_HARDWALL; |
682a3385 MG |
3749 | alloc_flags |= ALLOC_CPUSET; |
3750 | if (!ac.nodemask) | |
3751 | ac.nodemask = &cpuset_current_mems_allowed; | |
3752 | } | |
3753 | ||
dcce284a BH |
3754 | gfp_mask &= gfp_allowed_mask; |
3755 | ||
11e33f6a MG |
3756 | lockdep_trace_alloc(gfp_mask); |
3757 | ||
d0164adc | 3758 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
3759 | |
3760 | if (should_fail_alloc_page(gfp_mask, order)) | |
3761 | return NULL; | |
3762 | ||
3763 | /* | |
3764 | * Check the zones suitable for the gfp_mask contain at least one | |
3765 | * valid zone. It's possible to have an empty zonelist as a result | |
4167e9b2 | 3766 | * of __GFP_THISNODE and a memoryless node |
11e33f6a MG |
3767 | */ |
3768 | if (unlikely(!zonelist->_zonerefs->zone)) | |
3769 | return NULL; | |
3770 | ||
a9263751 | 3771 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
21bb9bd1 VB |
3772 | alloc_flags |= ALLOC_CMA; |
3773 | ||
cc9a6c87 | 3774 | retry_cpuset: |
d26914d1 | 3775 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3776 | |
c9ab0c4f MG |
3777 | /* Dirty zone balancing only done in the fast path */ |
3778 | ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE); | |
3779 | ||
e46e7b77 MG |
3780 | /* |
3781 | * The preferred zone is used for statistics but crucially it is | |
3782 | * also used as the starting point for the zonelist iterator. It | |
3783 | * may get reset for allocations that ignore memory policies. | |
3784 | */ | |
c33d6c06 MG |
3785 | ac.preferred_zoneref = first_zones_zonelist(ac.zonelist, |
3786 | ac.high_zoneidx, ac.nodemask); | |
3787 | if (!ac.preferred_zoneref) { | |
5bb1b169 | 3788 | page = NULL; |
4fcb0971 | 3789 | goto no_zone; |
5bb1b169 MG |
3790 | } |
3791 | ||
5117f45d | 3792 | /* First allocation attempt */ |
a9263751 | 3793 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
3794 | if (likely(page)) |
3795 | goto out; | |
11e33f6a | 3796 | |
4fcb0971 MG |
3797 | /* |
3798 | * Runtime PM, block IO and its error handling path can deadlock | |
3799 | * because I/O on the device might not complete. | |
3800 | */ | |
3801 | alloc_mask = memalloc_noio_flags(gfp_mask); | |
3802 | ac.spread_dirty_pages = false; | |
23f086f9 | 3803 | |
4741526b MG |
3804 | /* |
3805 | * Restore the original nodemask if it was potentially replaced with | |
3806 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
3807 | */ | |
3808 | if (cpusets_enabled()) | |
3809 | ac.nodemask = nodemask; | |
4fcb0971 | 3810 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 3811 | |
4fcb0971 | 3812 | no_zone: |
cc9a6c87 MG |
3813 | /* |
3814 | * When updating a task's mems_allowed, it is possible to race with | |
3815 | * parallel threads in such a way that an allocation can fail while | |
3816 | * the mask is being updated. If a page allocation is about to fail, | |
3817 | * check if the cpuset changed during allocation and if so, retry. | |
3818 | */ | |
83d4ca81 MG |
3819 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) { |
3820 | alloc_mask = gfp_mask; | |
cc9a6c87 | 3821 | goto retry_cpuset; |
83d4ca81 | 3822 | } |
cc9a6c87 | 3823 | |
4fcb0971 | 3824 | out: |
c4159a75 VD |
3825 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
3826 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { | |
3827 | __free_pages(page, order); | |
3828 | page = NULL; | |
4949148a VD |
3829 | } |
3830 | ||
4fcb0971 MG |
3831 | if (kmemcheck_enabled && page) |
3832 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
3833 | ||
3834 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); | |
3835 | ||
11e33f6a | 3836 | return page; |
1da177e4 | 3837 | } |
d239171e | 3838 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
3839 | |
3840 | /* | |
3841 | * Common helper functions. | |
3842 | */ | |
920c7a5d | 3843 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 3844 | { |
945a1113 AM |
3845 | struct page *page; |
3846 | ||
3847 | /* | |
3848 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
3849 | * a highmem page | |
3850 | */ | |
3851 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
3852 | ||
1da177e4 LT |
3853 | page = alloc_pages(gfp_mask, order); |
3854 | if (!page) | |
3855 | return 0; | |
3856 | return (unsigned long) page_address(page); | |
3857 | } | |
1da177e4 LT |
3858 | EXPORT_SYMBOL(__get_free_pages); |
3859 | ||
920c7a5d | 3860 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 3861 | { |
945a1113 | 3862 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 3863 | } |
1da177e4 LT |
3864 | EXPORT_SYMBOL(get_zeroed_page); |
3865 | ||
920c7a5d | 3866 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 3867 | { |
b5810039 | 3868 | if (put_page_testzero(page)) { |
1da177e4 | 3869 | if (order == 0) |
b745bc85 | 3870 | free_hot_cold_page(page, false); |
1da177e4 LT |
3871 | else |
3872 | __free_pages_ok(page, order); | |
3873 | } | |
3874 | } | |
3875 | ||
3876 | EXPORT_SYMBOL(__free_pages); | |
3877 | ||
920c7a5d | 3878 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
3879 | { |
3880 | if (addr != 0) { | |
725d704e | 3881 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
3882 | __free_pages(virt_to_page((void *)addr), order); |
3883 | } | |
3884 | } | |
3885 | ||
3886 | EXPORT_SYMBOL(free_pages); | |
3887 | ||
b63ae8ca AD |
3888 | /* |
3889 | * Page Fragment: | |
3890 | * An arbitrary-length arbitrary-offset area of memory which resides | |
3891 | * within a 0 or higher order page. Multiple fragments within that page | |
3892 | * are individually refcounted, in the page's reference counter. | |
3893 | * | |
3894 | * The page_frag functions below provide a simple allocation framework for | |
3895 | * page fragments. This is used by the network stack and network device | |
3896 | * drivers to provide a backing region of memory for use as either an | |
3897 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
3898 | */ | |
2976db80 AD |
3899 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
3900 | gfp_t gfp_mask) | |
b63ae8ca AD |
3901 | { |
3902 | struct page *page = NULL; | |
3903 | gfp_t gfp = gfp_mask; | |
3904 | ||
3905 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3906 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
3907 | __GFP_NOMEMALLOC; | |
3908 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
3909 | PAGE_FRAG_CACHE_MAX_ORDER); | |
3910 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
3911 | #endif | |
3912 | if (unlikely(!page)) | |
3913 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
3914 | ||
3915 | nc->va = page ? page_address(page) : NULL; | |
3916 | ||
3917 | return page; | |
3918 | } | |
3919 | ||
2976db80 | 3920 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
3921 | { |
3922 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
3923 | ||
3924 | if (page_ref_sub_and_test(page, count)) { | |
2976db80 AD |
3925 | unsigned int order = compound_order(page); |
3926 | ||
44fdffd7 AD |
3927 | if (order == 0) |
3928 | free_hot_cold_page(page, false); | |
3929 | else | |
3930 | __free_pages_ok(page, order); | |
3931 | } | |
3932 | } | |
2976db80 | 3933 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 3934 | |
8c2dd3e4 AD |
3935 | void *page_frag_alloc(struct page_frag_cache *nc, |
3936 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
3937 | { |
3938 | unsigned int size = PAGE_SIZE; | |
3939 | struct page *page; | |
3940 | int offset; | |
3941 | ||
3942 | if (unlikely(!nc->va)) { | |
3943 | refill: | |
2976db80 | 3944 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
3945 | if (!page) |
3946 | return NULL; | |
3947 | ||
3948 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3949 | /* if size can vary use size else just use PAGE_SIZE */ | |
3950 | size = nc->size; | |
3951 | #endif | |
3952 | /* Even if we own the page, we do not use atomic_set(). | |
3953 | * This would break get_page_unless_zero() users. | |
3954 | */ | |
fe896d18 | 3955 | page_ref_add(page, size - 1); |
b63ae8ca AD |
3956 | |
3957 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 3958 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
3959 | nc->pagecnt_bias = size; |
3960 | nc->offset = size; | |
3961 | } | |
3962 | ||
3963 | offset = nc->offset - fragsz; | |
3964 | if (unlikely(offset < 0)) { | |
3965 | page = virt_to_page(nc->va); | |
3966 | ||
fe896d18 | 3967 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
3968 | goto refill; |
3969 | ||
3970 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3971 | /* if size can vary use size else just use PAGE_SIZE */ | |
3972 | size = nc->size; | |
3973 | #endif | |
3974 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 3975 | set_page_count(page, size); |
b63ae8ca AD |
3976 | |
3977 | /* reset page count bias and offset to start of new frag */ | |
3978 | nc->pagecnt_bias = size; | |
3979 | offset = size - fragsz; | |
3980 | } | |
3981 | ||
3982 | nc->pagecnt_bias--; | |
3983 | nc->offset = offset; | |
3984 | ||
3985 | return nc->va + offset; | |
3986 | } | |
8c2dd3e4 | 3987 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
3988 | |
3989 | /* | |
3990 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
3991 | */ | |
8c2dd3e4 | 3992 | void page_frag_free(void *addr) |
b63ae8ca AD |
3993 | { |
3994 | struct page *page = virt_to_head_page(addr); | |
3995 | ||
3996 | if (unlikely(put_page_testzero(page))) | |
3997 | __free_pages_ok(page, compound_order(page)); | |
3998 | } | |
8c2dd3e4 | 3999 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4000 | |
d00181b9 KS |
4001 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4002 | size_t size) | |
ee85c2e1 AK |
4003 | { |
4004 | if (addr) { | |
4005 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
4006 | unsigned long used = addr + PAGE_ALIGN(size); | |
4007 | ||
4008 | split_page(virt_to_page((void *)addr), order); | |
4009 | while (used < alloc_end) { | |
4010 | free_page(used); | |
4011 | used += PAGE_SIZE; | |
4012 | } | |
4013 | } | |
4014 | return (void *)addr; | |
4015 | } | |
4016 | ||
2be0ffe2 TT |
4017 | /** |
4018 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4019 | * @size: the number of bytes to allocate | |
4020 | * @gfp_mask: GFP flags for the allocation | |
4021 | * | |
4022 | * This function is similar to alloc_pages(), except that it allocates the | |
4023 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4024 | * allocate memory in power-of-two pages. | |
4025 | * | |
4026 | * This function is also limited by MAX_ORDER. | |
4027 | * | |
4028 | * Memory allocated by this function must be released by free_pages_exact(). | |
4029 | */ | |
4030 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4031 | { | |
4032 | unsigned int order = get_order(size); | |
4033 | unsigned long addr; | |
4034 | ||
4035 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 4036 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4037 | } |
4038 | EXPORT_SYMBOL(alloc_pages_exact); | |
4039 | ||
ee85c2e1 AK |
4040 | /** |
4041 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4042 | * pages on a node. | |
b5e6ab58 | 4043 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
4044 | * @size: the number of bytes to allocate |
4045 | * @gfp_mask: GFP flags for the allocation | |
4046 | * | |
4047 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4048 | * back. | |
ee85c2e1 | 4049 | */ |
e1931811 | 4050 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4051 | { |
d00181b9 | 4052 | unsigned int order = get_order(size); |
ee85c2e1 AK |
4053 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
4054 | if (!p) | |
4055 | return NULL; | |
4056 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4057 | } | |
ee85c2e1 | 4058 | |
2be0ffe2 TT |
4059 | /** |
4060 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4061 | * @virt: the value returned by alloc_pages_exact. | |
4062 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4063 | * | |
4064 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4065 | */ | |
4066 | void free_pages_exact(void *virt, size_t size) | |
4067 | { | |
4068 | unsigned long addr = (unsigned long)virt; | |
4069 | unsigned long end = addr + PAGE_ALIGN(size); | |
4070 | ||
4071 | while (addr < end) { | |
4072 | free_page(addr); | |
4073 | addr += PAGE_SIZE; | |
4074 | } | |
4075 | } | |
4076 | EXPORT_SYMBOL(free_pages_exact); | |
4077 | ||
e0fb5815 ZY |
4078 | /** |
4079 | * nr_free_zone_pages - count number of pages beyond high watermark | |
4080 | * @offset: The zone index of the highest zone | |
4081 | * | |
4082 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
4083 | * high watermark within all zones at or below a given zone index. For each | |
4084 | * zone, the number of pages is calculated as: | |
834405c3 | 4085 | * managed_pages - high_pages |
e0fb5815 | 4086 | */ |
ebec3862 | 4087 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 4088 | { |
dd1a239f | 4089 | struct zoneref *z; |
54a6eb5c MG |
4090 | struct zone *zone; |
4091 | ||
e310fd43 | 4092 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 4093 | unsigned long sum = 0; |
1da177e4 | 4094 | |
0e88460d | 4095 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 4096 | |
54a6eb5c | 4097 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 4098 | unsigned long size = zone->managed_pages; |
41858966 | 4099 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
4100 | if (size > high) |
4101 | sum += size - high; | |
1da177e4 LT |
4102 | } |
4103 | ||
4104 | return sum; | |
4105 | } | |
4106 | ||
e0fb5815 ZY |
4107 | /** |
4108 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
4109 | * | |
4110 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
4111 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 4112 | */ |
ebec3862 | 4113 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 4114 | { |
af4ca457 | 4115 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 4116 | } |
c2f1a551 | 4117 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 4118 | |
e0fb5815 ZY |
4119 | /** |
4120 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
4121 | * | |
4122 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
4123 | * high watermark within all zones. | |
1da177e4 | 4124 | */ |
ebec3862 | 4125 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 4126 | { |
2a1e274a | 4127 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 4128 | } |
08e0f6a9 CL |
4129 | |
4130 | static inline void show_node(struct zone *zone) | |
1da177e4 | 4131 | { |
e5adfffc | 4132 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 4133 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 4134 | } |
1da177e4 | 4135 | |
d02bd27b IR |
4136 | long si_mem_available(void) |
4137 | { | |
4138 | long available; | |
4139 | unsigned long pagecache; | |
4140 | unsigned long wmark_low = 0; | |
4141 | unsigned long pages[NR_LRU_LISTS]; | |
4142 | struct zone *zone; | |
4143 | int lru; | |
4144 | ||
4145 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 4146 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
4147 | |
4148 | for_each_zone(zone) | |
4149 | wmark_low += zone->watermark[WMARK_LOW]; | |
4150 | ||
4151 | /* | |
4152 | * Estimate the amount of memory available for userspace allocations, | |
4153 | * without causing swapping. | |
4154 | */ | |
4155 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; | |
4156 | ||
4157 | /* | |
4158 | * Not all the page cache can be freed, otherwise the system will | |
4159 | * start swapping. Assume at least half of the page cache, or the | |
4160 | * low watermark worth of cache, needs to stay. | |
4161 | */ | |
4162 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
4163 | pagecache -= min(pagecache / 2, wmark_low); | |
4164 | available += pagecache; | |
4165 | ||
4166 | /* | |
4167 | * Part of the reclaimable slab consists of items that are in use, | |
4168 | * and cannot be freed. Cap this estimate at the low watermark. | |
4169 | */ | |
4170 | available += global_page_state(NR_SLAB_RECLAIMABLE) - | |
4171 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); | |
4172 | ||
4173 | if (available < 0) | |
4174 | available = 0; | |
4175 | return available; | |
4176 | } | |
4177 | EXPORT_SYMBOL_GPL(si_mem_available); | |
4178 | ||
1da177e4 LT |
4179 | void si_meminfo(struct sysinfo *val) |
4180 | { | |
4181 | val->totalram = totalram_pages; | |
11fb9989 | 4182 | val->sharedram = global_node_page_state(NR_SHMEM); |
d23ad423 | 4183 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 4184 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
4185 | val->totalhigh = totalhigh_pages; |
4186 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
4187 | val->mem_unit = PAGE_SIZE; |
4188 | } | |
4189 | ||
4190 | EXPORT_SYMBOL(si_meminfo); | |
4191 | ||
4192 | #ifdef CONFIG_NUMA | |
4193 | void si_meminfo_node(struct sysinfo *val, int nid) | |
4194 | { | |
cdd91a77 JL |
4195 | int zone_type; /* needs to be signed */ |
4196 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
4197 | unsigned long managed_highpages = 0; |
4198 | unsigned long free_highpages = 0; | |
1da177e4 LT |
4199 | pg_data_t *pgdat = NODE_DATA(nid); |
4200 | ||
cdd91a77 JL |
4201 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
4202 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
4203 | val->totalram = managed_pages; | |
11fb9989 | 4204 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 4205 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 4206 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
4207 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4208 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4209 | ||
4210 | if (is_highmem(zone)) { | |
4211 | managed_highpages += zone->managed_pages; | |
4212 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
4213 | } | |
4214 | } | |
4215 | val->totalhigh = managed_highpages; | |
4216 | val->freehigh = free_highpages; | |
98d2b0eb | 4217 | #else |
fc2bd799 JK |
4218 | val->totalhigh = managed_highpages; |
4219 | val->freehigh = free_highpages; | |
98d2b0eb | 4220 | #endif |
1da177e4 LT |
4221 | val->mem_unit = PAGE_SIZE; |
4222 | } | |
4223 | #endif | |
4224 | ||
ddd588b5 | 4225 | /* |
7bf02ea2 DR |
4226 | * Determine whether the node should be displayed or not, depending on whether |
4227 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 4228 | */ |
7bf02ea2 | 4229 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
4230 | { |
4231 | bool ret = false; | |
cc9a6c87 | 4232 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
4233 | |
4234 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
4235 | goto out; | |
4236 | ||
cc9a6c87 | 4237 | do { |
d26914d1 | 4238 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 4239 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 4240 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
4241 | out: |
4242 | return ret; | |
4243 | } | |
4244 | ||
1da177e4 LT |
4245 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4246 | ||
377e4f16 RV |
4247 | static void show_migration_types(unsigned char type) |
4248 | { | |
4249 | static const char types[MIGRATE_TYPES] = { | |
4250 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 4251 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
4252 | [MIGRATE_RECLAIMABLE] = 'E', |
4253 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
4254 | #ifdef CONFIG_CMA |
4255 | [MIGRATE_CMA] = 'C', | |
4256 | #endif | |
194159fb | 4257 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 4258 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 4259 | #endif |
377e4f16 RV |
4260 | }; |
4261 | char tmp[MIGRATE_TYPES + 1]; | |
4262 | char *p = tmp; | |
4263 | int i; | |
4264 | ||
4265 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
4266 | if (type & (1 << i)) | |
4267 | *p++ = types[i]; | |
4268 | } | |
4269 | ||
4270 | *p = '\0'; | |
1f84a18f | 4271 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
4272 | } |
4273 | ||
1da177e4 LT |
4274 | /* |
4275 | * Show free area list (used inside shift_scroll-lock stuff) | |
4276 | * We also calculate the percentage fragmentation. We do this by counting the | |
4277 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
4278 | * |
4279 | * Bits in @filter: | |
4280 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
4281 | * cpuset. | |
1da177e4 | 4282 | */ |
7bf02ea2 | 4283 | void show_free_areas(unsigned int filter) |
1da177e4 | 4284 | { |
d1bfcdb8 | 4285 | unsigned long free_pcp = 0; |
c7241913 | 4286 | int cpu; |
1da177e4 | 4287 | struct zone *zone; |
599d0c95 | 4288 | pg_data_t *pgdat; |
1da177e4 | 4289 | |
ee99c71c | 4290 | for_each_populated_zone(zone) { |
7bf02ea2 | 4291 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4292 | continue; |
d1bfcdb8 | 4293 | |
761b0677 KK |
4294 | for_each_online_cpu(cpu) |
4295 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
4296 | } |
4297 | ||
a731286d KM |
4298 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4299 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
4300 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4301 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 4302 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 4303 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
4304 | global_node_page_state(NR_ACTIVE_ANON), |
4305 | global_node_page_state(NR_INACTIVE_ANON), | |
4306 | global_node_page_state(NR_ISOLATED_ANON), | |
4307 | global_node_page_state(NR_ACTIVE_FILE), | |
4308 | global_node_page_state(NR_INACTIVE_FILE), | |
4309 | global_node_page_state(NR_ISOLATED_FILE), | |
4310 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
4311 | global_node_page_state(NR_FILE_DIRTY), |
4312 | global_node_page_state(NR_WRITEBACK), | |
4313 | global_node_page_state(NR_UNSTABLE_NFS), | |
3701b033 KM |
4314 | global_page_state(NR_SLAB_RECLAIMABLE), |
4315 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 4316 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 4317 | global_node_page_state(NR_SHMEM), |
a25700a5 | 4318 | global_page_state(NR_PAGETABLE), |
d1ce749a | 4319 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
4320 | global_page_state(NR_FREE_PAGES), |
4321 | free_pcp, | |
d1ce749a | 4322 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4323 | |
599d0c95 MG |
4324 | for_each_online_pgdat(pgdat) { |
4325 | printk("Node %d" | |
4326 | " active_anon:%lukB" | |
4327 | " inactive_anon:%lukB" | |
4328 | " active_file:%lukB" | |
4329 | " inactive_file:%lukB" | |
4330 | " unevictable:%lukB" | |
4331 | " isolated(anon):%lukB" | |
4332 | " isolated(file):%lukB" | |
50658e2e | 4333 | " mapped:%lukB" |
11fb9989 MG |
4334 | " dirty:%lukB" |
4335 | " writeback:%lukB" | |
4336 | " shmem:%lukB" | |
4337 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4338 | " shmem_thp: %lukB" | |
4339 | " shmem_pmdmapped: %lukB" | |
4340 | " anon_thp: %lukB" | |
4341 | #endif | |
4342 | " writeback_tmp:%lukB" | |
4343 | " unstable:%lukB" | |
33e077bd | 4344 | " pages_scanned:%lu" |
599d0c95 MG |
4345 | " all_unreclaimable? %s" |
4346 | "\n", | |
4347 | pgdat->node_id, | |
4348 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
4349 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
4350 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
4351 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
4352 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
4353 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
4354 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 4355 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
4356 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
4357 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
4358 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4359 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
4360 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
4361 | * HPAGE_PMD_NR), | |
4362 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
4363 | #endif | |
4364 | K(node_page_state(pgdat, NR_SHMEM)), | |
4365 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), | |
4366 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), | |
33e077bd | 4367 | node_page_state(pgdat, NR_PAGES_SCANNED), |
599d0c95 MG |
4368 | !pgdat_reclaimable(pgdat) ? "yes" : "no"); |
4369 | } | |
4370 | ||
ee99c71c | 4371 | for_each_populated_zone(zone) { |
1da177e4 LT |
4372 | int i; |
4373 | ||
7bf02ea2 | 4374 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4375 | continue; |
d1bfcdb8 KK |
4376 | |
4377 | free_pcp = 0; | |
4378 | for_each_online_cpu(cpu) | |
4379 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
4380 | ||
1da177e4 | 4381 | show_node(zone); |
1f84a18f JP |
4382 | printk(KERN_CONT |
4383 | "%s" | |
1da177e4 LT |
4384 | " free:%lukB" |
4385 | " min:%lukB" | |
4386 | " low:%lukB" | |
4387 | " high:%lukB" | |
71c799f4 MK |
4388 | " active_anon:%lukB" |
4389 | " inactive_anon:%lukB" | |
4390 | " active_file:%lukB" | |
4391 | " inactive_file:%lukB" | |
4392 | " unevictable:%lukB" | |
5a1c84b4 | 4393 | " writepending:%lukB" |
1da177e4 | 4394 | " present:%lukB" |
9feedc9d | 4395 | " managed:%lukB" |
4a0aa73f | 4396 | " mlocked:%lukB" |
4a0aa73f KM |
4397 | " slab_reclaimable:%lukB" |
4398 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 4399 | " kernel_stack:%lukB" |
4a0aa73f | 4400 | " pagetables:%lukB" |
4a0aa73f | 4401 | " bounce:%lukB" |
d1bfcdb8 KK |
4402 | " free_pcp:%lukB" |
4403 | " local_pcp:%ukB" | |
d1ce749a | 4404 | " free_cma:%lukB" |
1da177e4 LT |
4405 | "\n", |
4406 | zone->name, | |
88f5acf8 | 4407 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
4408 | K(min_wmark_pages(zone)), |
4409 | K(low_wmark_pages(zone)), | |
4410 | K(high_wmark_pages(zone)), | |
71c799f4 MK |
4411 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
4412 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
4413 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
4414 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
4415 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 4416 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 4417 | K(zone->present_pages), |
9feedc9d | 4418 | K(zone->managed_pages), |
4a0aa73f | 4419 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f KM |
4420 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
4421 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
d30dd8be | 4422 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4a0aa73f | 4423 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 4424 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
4425 | K(free_pcp), |
4426 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 4427 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
4428 | printk("lowmem_reserve[]:"); |
4429 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
4430 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
4431 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
4432 | } |
4433 | ||
ee99c71c | 4434 | for_each_populated_zone(zone) { |
d00181b9 KS |
4435 | unsigned int order; |
4436 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 4437 | unsigned char types[MAX_ORDER]; |
1da177e4 | 4438 | |
7bf02ea2 | 4439 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4440 | continue; |
1da177e4 | 4441 | show_node(zone); |
1f84a18f | 4442 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
4443 | |
4444 | spin_lock_irqsave(&zone->lock, flags); | |
4445 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
4446 | struct free_area *area = &zone->free_area[order]; |
4447 | int type; | |
4448 | ||
4449 | nr[order] = area->nr_free; | |
8f9de51a | 4450 | total += nr[order] << order; |
377e4f16 RV |
4451 | |
4452 | types[order] = 0; | |
4453 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
4454 | if (!list_empty(&area->free_list[type])) | |
4455 | types[order] |= 1 << type; | |
4456 | } | |
1da177e4 LT |
4457 | } |
4458 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 4459 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
4460 | printk(KERN_CONT "%lu*%lukB ", |
4461 | nr[order], K(1UL) << order); | |
377e4f16 RV |
4462 | if (nr[order]) |
4463 | show_migration_types(types[order]); | |
4464 | } | |
1f84a18f | 4465 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
4466 | } |
4467 | ||
949f7ec5 DR |
4468 | hugetlb_show_meminfo(); |
4469 | ||
11fb9989 | 4470 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 4471 | |
1da177e4 LT |
4472 | show_swap_cache_info(); |
4473 | } | |
4474 | ||
19770b32 MG |
4475 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4476 | { | |
4477 | zoneref->zone = zone; | |
4478 | zoneref->zone_idx = zone_idx(zone); | |
4479 | } | |
4480 | ||
1da177e4 LT |
4481 | /* |
4482 | * Builds allocation fallback zone lists. | |
1a93205b CL |
4483 | * |
4484 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 4485 | */ |
f0c0b2b8 | 4486 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 4487 | int nr_zones) |
1da177e4 | 4488 | { |
1a93205b | 4489 | struct zone *zone; |
bc732f1d | 4490 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
4491 | |
4492 | do { | |
2f6726e5 | 4493 | zone_type--; |
070f8032 | 4494 | zone = pgdat->node_zones + zone_type; |
6aa303de | 4495 | if (managed_zone(zone)) { |
dd1a239f MG |
4496 | zoneref_set_zone(zone, |
4497 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 4498 | check_highest_zone(zone_type); |
1da177e4 | 4499 | } |
2f6726e5 | 4500 | } while (zone_type); |
bc732f1d | 4501 | |
070f8032 | 4502 | return nr_zones; |
1da177e4 LT |
4503 | } |
4504 | ||
f0c0b2b8 KH |
4505 | |
4506 | /* | |
4507 | * zonelist_order: | |
4508 | * 0 = automatic detection of better ordering. | |
4509 | * 1 = order by ([node] distance, -zonetype) | |
4510 | * 2 = order by (-zonetype, [node] distance) | |
4511 | * | |
4512 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
4513 | * the same zonelist. So only NUMA can configure this param. | |
4514 | */ | |
4515 | #define ZONELIST_ORDER_DEFAULT 0 | |
4516 | #define ZONELIST_ORDER_NODE 1 | |
4517 | #define ZONELIST_ORDER_ZONE 2 | |
4518 | ||
4519 | /* zonelist order in the kernel. | |
4520 | * set_zonelist_order() will set this to NODE or ZONE. | |
4521 | */ | |
4522 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4523 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
4524 | ||
4525 | ||
1da177e4 | 4526 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
4527 | /* The value user specified ....changed by config */ |
4528 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4529 | /* string for sysctl */ | |
4530 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
4531 | char numa_zonelist_order[16] = "default"; | |
4532 | ||
4533 | /* | |
4534 | * interface for configure zonelist ordering. | |
4535 | * command line option "numa_zonelist_order" | |
4536 | * = "[dD]efault - default, automatic configuration. | |
4537 | * = "[nN]ode - order by node locality, then by zone within node | |
4538 | * = "[zZ]one - order by zone, then by locality within zone | |
4539 | */ | |
4540 | ||
4541 | static int __parse_numa_zonelist_order(char *s) | |
4542 | { | |
4543 | if (*s == 'd' || *s == 'D') { | |
4544 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4545 | } else if (*s == 'n' || *s == 'N') { | |
4546 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
4547 | } else if (*s == 'z' || *s == 'Z') { | |
4548 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
4549 | } else { | |
1170532b | 4550 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
f0c0b2b8 KH |
4551 | return -EINVAL; |
4552 | } | |
4553 | return 0; | |
4554 | } | |
4555 | ||
4556 | static __init int setup_numa_zonelist_order(char *s) | |
4557 | { | |
ecb256f8 VL |
4558 | int ret; |
4559 | ||
4560 | if (!s) | |
4561 | return 0; | |
4562 | ||
4563 | ret = __parse_numa_zonelist_order(s); | |
4564 | if (ret == 0) | |
4565 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
4566 | ||
4567 | return ret; | |
f0c0b2b8 KH |
4568 | } |
4569 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
4570 | ||
4571 | /* | |
4572 | * sysctl handler for numa_zonelist_order | |
4573 | */ | |
cccad5b9 | 4574 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 4575 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
4576 | loff_t *ppos) |
4577 | { | |
4578 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
4579 | int ret; | |
443c6f14 | 4580 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 4581 | |
443c6f14 | 4582 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
4583 | if (write) { |
4584 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
4585 | ret = -EINVAL; | |
4586 | goto out; | |
4587 | } | |
4588 | strcpy(saved_string, (char *)table->data); | |
4589 | } | |
8d65af78 | 4590 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 4591 | if (ret) |
443c6f14 | 4592 | goto out; |
f0c0b2b8 KH |
4593 | if (write) { |
4594 | int oldval = user_zonelist_order; | |
dacbde09 CG |
4595 | |
4596 | ret = __parse_numa_zonelist_order((char *)table->data); | |
4597 | if (ret) { | |
f0c0b2b8 KH |
4598 | /* |
4599 | * bogus value. restore saved string | |
4600 | */ | |
dacbde09 | 4601 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
4602 | NUMA_ZONELIST_ORDER_LEN); |
4603 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
4604 | } else if (oldval != user_zonelist_order) { |
4605 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 4606 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
4607 | mutex_unlock(&zonelists_mutex); |
4608 | } | |
f0c0b2b8 | 4609 | } |
443c6f14 AK |
4610 | out: |
4611 | mutex_unlock(&zl_order_mutex); | |
4612 | return ret; | |
f0c0b2b8 KH |
4613 | } |
4614 | ||
4615 | ||
62bc62a8 | 4616 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
4617 | static int node_load[MAX_NUMNODES]; |
4618 | ||
1da177e4 | 4619 | /** |
4dc3b16b | 4620 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
4621 | * @node: node whose fallback list we're appending |
4622 | * @used_node_mask: nodemask_t of already used nodes | |
4623 | * | |
4624 | * We use a number of factors to determine which is the next node that should | |
4625 | * appear on a given node's fallback list. The node should not have appeared | |
4626 | * already in @node's fallback list, and it should be the next closest node | |
4627 | * according to the distance array (which contains arbitrary distance values | |
4628 | * from each node to each node in the system), and should also prefer nodes | |
4629 | * with no CPUs, since presumably they'll have very little allocation pressure | |
4630 | * on them otherwise. | |
4631 | * It returns -1 if no node is found. | |
4632 | */ | |
f0c0b2b8 | 4633 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 4634 | { |
4cf808eb | 4635 | int n, val; |
1da177e4 | 4636 | int min_val = INT_MAX; |
00ef2d2f | 4637 | int best_node = NUMA_NO_NODE; |
a70f7302 | 4638 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 4639 | |
4cf808eb LT |
4640 | /* Use the local node if we haven't already */ |
4641 | if (!node_isset(node, *used_node_mask)) { | |
4642 | node_set(node, *used_node_mask); | |
4643 | return node; | |
4644 | } | |
1da177e4 | 4645 | |
4b0ef1fe | 4646 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
4647 | |
4648 | /* Don't want a node to appear more than once */ | |
4649 | if (node_isset(n, *used_node_mask)) | |
4650 | continue; | |
4651 | ||
1da177e4 LT |
4652 | /* Use the distance array to find the distance */ |
4653 | val = node_distance(node, n); | |
4654 | ||
4cf808eb LT |
4655 | /* Penalize nodes under us ("prefer the next node") */ |
4656 | val += (n < node); | |
4657 | ||
1da177e4 | 4658 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
4659 | tmp = cpumask_of_node(n); |
4660 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
4661 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4662 | ||
4663 | /* Slight preference for less loaded node */ | |
4664 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
4665 | val += node_load[n]; | |
4666 | ||
4667 | if (val < min_val) { | |
4668 | min_val = val; | |
4669 | best_node = n; | |
4670 | } | |
4671 | } | |
4672 | ||
4673 | if (best_node >= 0) | |
4674 | node_set(best_node, *used_node_mask); | |
4675 | ||
4676 | return best_node; | |
4677 | } | |
4678 | ||
f0c0b2b8 KH |
4679 | |
4680 | /* | |
4681 | * Build zonelists ordered by node and zones within node. | |
4682 | * This results in maximum locality--normal zone overflows into local | |
4683 | * DMA zone, if any--but risks exhausting DMA zone. | |
4684 | */ | |
4685 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 4686 | { |
f0c0b2b8 | 4687 | int j; |
1da177e4 | 4688 | struct zonelist *zonelist; |
f0c0b2b8 | 4689 | |
c9634cf0 | 4690 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
dd1a239f | 4691 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 4692 | ; |
bc732f1d | 4693 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
4694 | zonelist->_zonerefs[j].zone = NULL; |
4695 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
4696 | } |
4697 | ||
523b9458 CL |
4698 | /* |
4699 | * Build gfp_thisnode zonelists | |
4700 | */ | |
4701 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
4702 | { | |
523b9458 CL |
4703 | int j; |
4704 | struct zonelist *zonelist; | |
4705 | ||
c9634cf0 | 4706 | zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK]; |
bc732f1d | 4707 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
4708 | zonelist->_zonerefs[j].zone = NULL; |
4709 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
4710 | } |
4711 | ||
f0c0b2b8 KH |
4712 | /* |
4713 | * Build zonelists ordered by zone and nodes within zones. | |
4714 | * This results in conserving DMA zone[s] until all Normal memory is | |
4715 | * exhausted, but results in overflowing to remote node while memory | |
4716 | * may still exist in local DMA zone. | |
4717 | */ | |
4718 | static int node_order[MAX_NUMNODES]; | |
4719 | ||
4720 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
4721 | { | |
f0c0b2b8 KH |
4722 | int pos, j, node; |
4723 | int zone_type; /* needs to be signed */ | |
4724 | struct zone *z; | |
4725 | struct zonelist *zonelist; | |
4726 | ||
c9634cf0 | 4727 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
54a6eb5c MG |
4728 | pos = 0; |
4729 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
4730 | for (j = 0; j < nr_nodes; j++) { | |
4731 | node = node_order[j]; | |
4732 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
6aa303de | 4733 | if (managed_zone(z)) { |
dd1a239f MG |
4734 | zoneref_set_zone(z, |
4735 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 4736 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
4737 | } |
4738 | } | |
f0c0b2b8 | 4739 | } |
dd1a239f MG |
4740 | zonelist->_zonerefs[pos].zone = NULL; |
4741 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
4742 | } |
4743 | ||
3193913c MG |
4744 | #if defined(CONFIG_64BIT) |
4745 | /* | |
4746 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
4747 | * penalty to every machine in case the specialised case applies. Default | |
4748 | * to Node-ordering on 64-bit NUMA machines | |
4749 | */ | |
4750 | static int default_zonelist_order(void) | |
4751 | { | |
4752 | return ZONELIST_ORDER_NODE; | |
4753 | } | |
4754 | #else | |
4755 | /* | |
4756 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
4757 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
4758 | * then reclaim will occur more frequency increasing stalls and potentially | |
4759 | * be easier to OOM if a large percentage of the zone is under writeback or | |
4760 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
4761 | * Hence, default to zone ordering on 32-bit. | |
4762 | */ | |
f0c0b2b8 KH |
4763 | static int default_zonelist_order(void) |
4764 | { | |
f0c0b2b8 KH |
4765 | return ZONELIST_ORDER_ZONE; |
4766 | } | |
3193913c | 4767 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
4768 | |
4769 | static void set_zonelist_order(void) | |
4770 | { | |
4771 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
4772 | current_zonelist_order = default_zonelist_order(); | |
4773 | else | |
4774 | current_zonelist_order = user_zonelist_order; | |
4775 | } | |
4776 | ||
4777 | static void build_zonelists(pg_data_t *pgdat) | |
4778 | { | |
c00eb15a | 4779 | int i, node, load; |
1da177e4 | 4780 | nodemask_t used_mask; |
f0c0b2b8 KH |
4781 | int local_node, prev_node; |
4782 | struct zonelist *zonelist; | |
d00181b9 | 4783 | unsigned int order = current_zonelist_order; |
1da177e4 LT |
4784 | |
4785 | /* initialize zonelists */ | |
523b9458 | 4786 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 4787 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
4788 | zonelist->_zonerefs[0].zone = NULL; |
4789 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
4790 | } |
4791 | ||
4792 | /* NUMA-aware ordering of nodes */ | |
4793 | local_node = pgdat->node_id; | |
62bc62a8 | 4794 | load = nr_online_nodes; |
1da177e4 LT |
4795 | prev_node = local_node; |
4796 | nodes_clear(used_mask); | |
f0c0b2b8 | 4797 | |
f0c0b2b8 | 4798 | memset(node_order, 0, sizeof(node_order)); |
c00eb15a | 4799 | i = 0; |
f0c0b2b8 | 4800 | |
1da177e4 LT |
4801 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
4802 | /* | |
4803 | * We don't want to pressure a particular node. | |
4804 | * So adding penalty to the first node in same | |
4805 | * distance group to make it round-robin. | |
4806 | */ | |
957f822a DR |
4807 | if (node_distance(local_node, node) != |
4808 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
4809 | node_load[node] = load; |
4810 | ||
1da177e4 LT |
4811 | prev_node = node; |
4812 | load--; | |
f0c0b2b8 KH |
4813 | if (order == ZONELIST_ORDER_NODE) |
4814 | build_zonelists_in_node_order(pgdat, node); | |
4815 | else | |
c00eb15a | 4816 | node_order[i++] = node; /* remember order */ |
f0c0b2b8 | 4817 | } |
1da177e4 | 4818 | |
f0c0b2b8 KH |
4819 | if (order == ZONELIST_ORDER_ZONE) { |
4820 | /* calculate node order -- i.e., DMA last! */ | |
c00eb15a | 4821 | build_zonelists_in_zone_order(pgdat, i); |
1da177e4 | 4822 | } |
523b9458 CL |
4823 | |
4824 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
4825 | } |
4826 | ||
7aac7898 LS |
4827 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4828 | /* | |
4829 | * Return node id of node used for "local" allocations. | |
4830 | * I.e., first node id of first zone in arg node's generic zonelist. | |
4831 | * Used for initializing percpu 'numa_mem', which is used primarily | |
4832 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
4833 | */ | |
4834 | int local_memory_node(int node) | |
4835 | { | |
c33d6c06 | 4836 | struct zoneref *z; |
7aac7898 | 4837 | |
c33d6c06 | 4838 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 4839 | gfp_zone(GFP_KERNEL), |
c33d6c06 MG |
4840 | NULL); |
4841 | return z->zone->node; | |
7aac7898 LS |
4842 | } |
4843 | #endif | |
f0c0b2b8 | 4844 | |
6423aa81 JK |
4845 | static void setup_min_unmapped_ratio(void); |
4846 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
4847 | #else /* CONFIG_NUMA */ |
4848 | ||
f0c0b2b8 KH |
4849 | static void set_zonelist_order(void) |
4850 | { | |
4851 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
4852 | } | |
4853 | ||
4854 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 4855 | { |
19655d34 | 4856 | int node, local_node; |
54a6eb5c MG |
4857 | enum zone_type j; |
4858 | struct zonelist *zonelist; | |
1da177e4 LT |
4859 | |
4860 | local_node = pgdat->node_id; | |
1da177e4 | 4861 | |
c9634cf0 | 4862 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
bc732f1d | 4863 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 4864 | |
54a6eb5c MG |
4865 | /* |
4866 | * Now we build the zonelist so that it contains the zones | |
4867 | * of all the other nodes. | |
4868 | * We don't want to pressure a particular node, so when | |
4869 | * building the zones for node N, we make sure that the | |
4870 | * zones coming right after the local ones are those from | |
4871 | * node N+1 (modulo N) | |
4872 | */ | |
4873 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
4874 | if (!node_online(node)) | |
4875 | continue; | |
bc732f1d | 4876 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 4877 | } |
54a6eb5c MG |
4878 | for (node = 0; node < local_node; node++) { |
4879 | if (!node_online(node)) | |
4880 | continue; | |
bc732f1d | 4881 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
4882 | } |
4883 | ||
dd1a239f MG |
4884 | zonelist->_zonerefs[j].zone = NULL; |
4885 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
4886 | } |
4887 | ||
4888 | #endif /* CONFIG_NUMA */ | |
4889 | ||
99dcc3e5 CL |
4890 | /* |
4891 | * Boot pageset table. One per cpu which is going to be used for all | |
4892 | * zones and all nodes. The parameters will be set in such a way | |
4893 | * that an item put on a list will immediately be handed over to | |
4894 | * the buddy list. This is safe since pageset manipulation is done | |
4895 | * with interrupts disabled. | |
4896 | * | |
4897 | * The boot_pagesets must be kept even after bootup is complete for | |
4898 | * unused processors and/or zones. They do play a role for bootstrapping | |
4899 | * hotplugged processors. | |
4900 | * | |
4901 | * zoneinfo_show() and maybe other functions do | |
4902 | * not check if the processor is online before following the pageset pointer. | |
4903 | * Other parts of the kernel may not check if the zone is available. | |
4904 | */ | |
4905 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
4906 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 4907 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 4908 | |
4eaf3f64 HL |
4909 | /* |
4910 | * Global mutex to protect against size modification of zonelists | |
4911 | * as well as to serialize pageset setup for the new populated zone. | |
4912 | */ | |
4913 | DEFINE_MUTEX(zonelists_mutex); | |
4914 | ||
9b1a4d38 | 4915 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 4916 | static int __build_all_zonelists(void *data) |
1da177e4 | 4917 | { |
6811378e | 4918 | int nid; |
99dcc3e5 | 4919 | int cpu; |
9adb62a5 | 4920 | pg_data_t *self = data; |
9276b1bc | 4921 | |
7f9cfb31 BL |
4922 | #ifdef CONFIG_NUMA |
4923 | memset(node_load, 0, sizeof(node_load)); | |
4924 | #endif | |
9adb62a5 JL |
4925 | |
4926 | if (self && !node_online(self->node_id)) { | |
4927 | build_zonelists(self); | |
9adb62a5 JL |
4928 | } |
4929 | ||
9276b1bc | 4930 | for_each_online_node(nid) { |
7ea1530a CL |
4931 | pg_data_t *pgdat = NODE_DATA(nid); |
4932 | ||
4933 | build_zonelists(pgdat); | |
9276b1bc | 4934 | } |
99dcc3e5 CL |
4935 | |
4936 | /* | |
4937 | * Initialize the boot_pagesets that are going to be used | |
4938 | * for bootstrapping processors. The real pagesets for | |
4939 | * each zone will be allocated later when the per cpu | |
4940 | * allocator is available. | |
4941 | * | |
4942 | * boot_pagesets are used also for bootstrapping offline | |
4943 | * cpus if the system is already booted because the pagesets | |
4944 | * are needed to initialize allocators on a specific cpu too. | |
4945 | * F.e. the percpu allocator needs the page allocator which | |
4946 | * needs the percpu allocator in order to allocate its pagesets | |
4947 | * (a chicken-egg dilemma). | |
4948 | */ | |
7aac7898 | 4949 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
4950 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
4951 | ||
7aac7898 LS |
4952 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4953 | /* | |
4954 | * We now know the "local memory node" for each node-- | |
4955 | * i.e., the node of the first zone in the generic zonelist. | |
4956 | * Set up numa_mem percpu variable for on-line cpus. During | |
4957 | * boot, only the boot cpu should be on-line; we'll init the | |
4958 | * secondary cpus' numa_mem as they come on-line. During | |
4959 | * node/memory hotplug, we'll fixup all on-line cpus. | |
4960 | */ | |
4961 | if (cpu_online(cpu)) | |
4962 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
4963 | #endif | |
4964 | } | |
4965 | ||
6811378e YG |
4966 | return 0; |
4967 | } | |
4968 | ||
061f67bc RV |
4969 | static noinline void __init |
4970 | build_all_zonelists_init(void) | |
4971 | { | |
4972 | __build_all_zonelists(NULL); | |
4973 | mminit_verify_zonelist(); | |
4974 | cpuset_init_current_mems_allowed(); | |
4975 | } | |
4976 | ||
4eaf3f64 HL |
4977 | /* |
4978 | * Called with zonelists_mutex held always | |
4979 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
4980 | * |
4981 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
4982 | * [we're only called with non-NULL zone through __meminit paths] and | |
4983 | * (2) call of __init annotated helper build_all_zonelists_init | |
4984 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 4985 | */ |
9adb62a5 | 4986 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 4987 | { |
f0c0b2b8 KH |
4988 | set_zonelist_order(); |
4989 | ||
6811378e | 4990 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 4991 | build_all_zonelists_init(); |
6811378e | 4992 | } else { |
e9959f0f | 4993 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
4994 | if (zone) |
4995 | setup_zone_pageset(zone); | |
e9959f0f | 4996 | #endif |
dd1895e2 CS |
4997 | /* we have to stop all cpus to guarantee there is no user |
4998 | of zonelist */ | |
9adb62a5 | 4999 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
5000 | /* cpuset refresh routine should be here */ |
5001 | } | |
bd1e22b8 | 5002 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5003 | /* |
5004 | * Disable grouping by mobility if the number of pages in the | |
5005 | * system is too low to allow the mechanism to work. It would be | |
5006 | * more accurate, but expensive to check per-zone. This check is | |
5007 | * made on memory-hotadd so a system can start with mobility | |
5008 | * disabled and enable it later | |
5009 | */ | |
d9c23400 | 5010 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5011 | page_group_by_mobility_disabled = 1; |
5012 | else | |
5013 | page_group_by_mobility_disabled = 0; | |
5014 | ||
756a025f JP |
5015 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
5016 | nr_online_nodes, | |
5017 | zonelist_order_name[current_zonelist_order], | |
5018 | page_group_by_mobility_disabled ? "off" : "on", | |
5019 | vm_total_pages); | |
f0c0b2b8 | 5020 | #ifdef CONFIG_NUMA |
f88dfff5 | 5021 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5022 | #endif |
1da177e4 LT |
5023 | } |
5024 | ||
1da177e4 LT |
5025 | /* |
5026 | * Initially all pages are reserved - free ones are freed | |
5027 | * up by free_all_bootmem() once the early boot process is | |
5028 | * done. Non-atomic initialization, single-pass. | |
5029 | */ | |
c09b4240 | 5030 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 5031 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 5032 | { |
4b94ffdc | 5033 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
29751f69 | 5034 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 5035 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 5036 | unsigned long pfn; |
3a80a7fa | 5037 | unsigned long nr_initialised = 0; |
342332e6 TI |
5038 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5039 | struct memblock_region *r = NULL, *tmp; | |
5040 | #endif | |
1da177e4 | 5041 | |
22b31eec HD |
5042 | if (highest_memmap_pfn < end_pfn - 1) |
5043 | highest_memmap_pfn = end_pfn - 1; | |
5044 | ||
4b94ffdc DW |
5045 | /* |
5046 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
5047 | * memory | |
5048 | */ | |
5049 | if (altmap && start_pfn == altmap->base_pfn) | |
5050 | start_pfn += altmap->reserve; | |
5051 | ||
cbe8dd4a | 5052 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 5053 | /* |
b72d0ffb AM |
5054 | * There can be holes in boot-time mem_map[]s handed to this |
5055 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5056 | */ |
b72d0ffb AM |
5057 | if (context != MEMMAP_EARLY) |
5058 | goto not_early; | |
5059 | ||
5060 | if (!early_pfn_valid(pfn)) | |
5061 | continue; | |
5062 | if (!early_pfn_in_nid(pfn, nid)) | |
5063 | continue; | |
5064 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
5065 | break; | |
342332e6 TI |
5066 | |
5067 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
5068 | /* |
5069 | * Check given memblock attribute by firmware which can affect | |
5070 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
5071 | * mirrored, it's an overlapped memmap init. skip it. | |
5072 | */ | |
5073 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5074 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
5075 | for_each_memblock(memory, tmp) | |
5076 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
5077 | break; | |
5078 | r = tmp; | |
5079 | } | |
5080 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
5081 | memblock_is_mirror(r)) { | |
5082 | /* already initialized as NORMAL */ | |
5083 | pfn = memblock_region_memory_end_pfn(r); | |
5084 | continue; | |
342332e6 | 5085 | } |
a2f3aa02 | 5086 | } |
b72d0ffb | 5087 | #endif |
ac5d2539 | 5088 | |
b72d0ffb | 5089 | not_early: |
ac5d2539 MG |
5090 | /* |
5091 | * Mark the block movable so that blocks are reserved for | |
5092 | * movable at startup. This will force kernel allocations | |
5093 | * to reserve their blocks rather than leaking throughout | |
5094 | * the address space during boot when many long-lived | |
974a786e | 5095 | * kernel allocations are made. |
ac5d2539 MG |
5096 | * |
5097 | * bitmap is created for zone's valid pfn range. but memmap | |
5098 | * can be created for invalid pages (for alignment) | |
5099 | * check here not to call set_pageblock_migratetype() against | |
5100 | * pfn out of zone. | |
5101 | */ | |
5102 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
5103 | struct page *page = pfn_to_page(pfn); | |
5104 | ||
5105 | __init_single_page(page, pfn, zone, nid); | |
5106 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5107 | } else { | |
5108 | __init_single_pfn(pfn, zone, nid); | |
5109 | } | |
1da177e4 LT |
5110 | } |
5111 | } | |
5112 | ||
1e548deb | 5113 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 5114 | { |
7aeb09f9 | 5115 | unsigned int order, t; |
b2a0ac88 MG |
5116 | for_each_migratetype_order(order, t) { |
5117 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
5118 | zone->free_area[order].nr_free = 0; |
5119 | } | |
5120 | } | |
5121 | ||
5122 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
5123 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 5124 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
5125 | #endif |
5126 | ||
7cd2b0a3 | 5127 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 5128 | { |
3a6be87f | 5129 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
5130 | int batch; |
5131 | ||
5132 | /* | |
5133 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 5134 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
5135 | * |
5136 | * OK, so we don't know how big the cache is. So guess. | |
5137 | */ | |
b40da049 | 5138 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
5139 | if (batch * PAGE_SIZE > 512 * 1024) |
5140 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
5141 | batch /= 4; /* We effectively *= 4 below */ |
5142 | if (batch < 1) | |
5143 | batch = 1; | |
5144 | ||
5145 | /* | |
0ceaacc9 NP |
5146 | * Clamp the batch to a 2^n - 1 value. Having a power |
5147 | * of 2 value was found to be more likely to have | |
5148 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 5149 | * |
0ceaacc9 NP |
5150 | * For example if 2 tasks are alternately allocating |
5151 | * batches of pages, one task can end up with a lot | |
5152 | * of pages of one half of the possible page colors | |
5153 | * and the other with pages of the other colors. | |
e7c8d5c9 | 5154 | */ |
9155203a | 5155 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 5156 | |
e7c8d5c9 | 5157 | return batch; |
3a6be87f DH |
5158 | |
5159 | #else | |
5160 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5161 | * conditions. | |
5162 | * | |
5163 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5164 | * of contiguous memory as there's no hardware page translation to | |
5165 | * assemble apparent contiguous memory from discontiguous pages. | |
5166 | * | |
5167 | * Queueing large contiguous runs of pages for batching, however, | |
5168 | * causes the pages to actually be freed in smaller chunks. As there | |
5169 | * can be a significant delay between the individual batches being | |
5170 | * recycled, this leads to the once large chunks of space being | |
5171 | * fragmented and becoming unavailable for high-order allocations. | |
5172 | */ | |
5173 | return 0; | |
5174 | #endif | |
e7c8d5c9 CL |
5175 | } |
5176 | ||
8d7a8fa9 CS |
5177 | /* |
5178 | * pcp->high and pcp->batch values are related and dependent on one another: | |
5179 | * ->batch must never be higher then ->high. | |
5180 | * The following function updates them in a safe manner without read side | |
5181 | * locking. | |
5182 | * | |
5183 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
5184 | * those fields changing asynchronously (acording the the above rule). | |
5185 | * | |
5186 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5187 | * outside of boot time (or some other assurance that no concurrent updaters | |
5188 | * exist). | |
5189 | */ | |
5190 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
5191 | unsigned long batch) | |
5192 | { | |
5193 | /* start with a fail safe value for batch */ | |
5194 | pcp->batch = 1; | |
5195 | smp_wmb(); | |
5196 | ||
5197 | /* Update high, then batch, in order */ | |
5198 | pcp->high = high; | |
5199 | smp_wmb(); | |
5200 | ||
5201 | pcp->batch = batch; | |
5202 | } | |
5203 | ||
3664033c | 5204 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
5205 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5206 | { | |
8d7a8fa9 | 5207 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
5208 | } |
5209 | ||
88c90dbc | 5210 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
5211 | { |
5212 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 5213 | int migratetype; |
2caaad41 | 5214 | |
1c6fe946 MD |
5215 | memset(p, 0, sizeof(*p)); |
5216 | ||
3dfa5721 | 5217 | pcp = &p->pcp; |
2caaad41 | 5218 | pcp->count = 0; |
5f8dcc21 MG |
5219 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5220 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
5221 | } |
5222 | ||
88c90dbc CS |
5223 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5224 | { | |
5225 | pageset_init(p); | |
5226 | pageset_set_batch(p, batch); | |
5227 | } | |
5228 | ||
8ad4b1fb | 5229 | /* |
3664033c | 5230 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
5231 | * to the value high for the pageset p. |
5232 | */ | |
3664033c | 5233 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
5234 | unsigned long high) |
5235 | { | |
8d7a8fa9 CS |
5236 | unsigned long batch = max(1UL, high / 4); |
5237 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
5238 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 5239 | |
8d7a8fa9 | 5240 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
5241 | } |
5242 | ||
7cd2b0a3 DR |
5243 | static void pageset_set_high_and_batch(struct zone *zone, |
5244 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 5245 | { |
56cef2b8 | 5246 | if (percpu_pagelist_fraction) |
3664033c | 5247 | pageset_set_high(pcp, |
56cef2b8 CS |
5248 | (zone->managed_pages / |
5249 | percpu_pagelist_fraction)); | |
5250 | else | |
5251 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
5252 | } | |
5253 | ||
169f6c19 CS |
5254 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5255 | { | |
5256 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
5257 | ||
5258 | pageset_init(pcp); | |
5259 | pageset_set_high_and_batch(zone, pcp); | |
5260 | } | |
5261 | ||
4ed7e022 | 5262 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
5263 | { |
5264 | int cpu; | |
319774e2 | 5265 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5266 | for_each_possible_cpu(cpu) |
5267 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5268 | } |
5269 | ||
2caaad41 | 5270 | /* |
99dcc3e5 CL |
5271 | * Allocate per cpu pagesets and initialize them. |
5272 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5273 | */ |
99dcc3e5 | 5274 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5275 | { |
b4911ea2 | 5276 | struct pglist_data *pgdat; |
99dcc3e5 | 5277 | struct zone *zone; |
e7c8d5c9 | 5278 | |
319774e2 WF |
5279 | for_each_populated_zone(zone) |
5280 | setup_zone_pageset(zone); | |
b4911ea2 MG |
5281 | |
5282 | for_each_online_pgdat(pgdat) | |
5283 | pgdat->per_cpu_nodestats = | |
5284 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
5285 | } |
5286 | ||
c09b4240 | 5287 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5288 | { |
99dcc3e5 CL |
5289 | /* |
5290 | * per cpu subsystem is not up at this point. The following code | |
5291 | * relies on the ability of the linker to provide the | |
5292 | * offset of a (static) per cpu variable into the per cpu area. | |
5293 | */ | |
5294 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5295 | |
b38a8725 | 5296 | if (populated_zone(zone)) |
99dcc3e5 CL |
5297 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5298 | zone->name, zone->present_pages, | |
5299 | zone_batchsize(zone)); | |
ed8ece2e DH |
5300 | } |
5301 | ||
4ed7e022 | 5302 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5303 | unsigned long zone_start_pfn, |
b171e409 | 5304 | unsigned long size) |
ed8ece2e DH |
5305 | { |
5306 | struct pglist_data *pgdat = zone->zone_pgdat; | |
9dcb8b68 | 5307 | |
ed8ece2e DH |
5308 | pgdat->nr_zones = zone_idx(zone) + 1; |
5309 | ||
ed8ece2e DH |
5310 | zone->zone_start_pfn = zone_start_pfn; |
5311 | ||
708614e6 MG |
5312 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5313 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5314 | pgdat->node_id, | |
5315 | (unsigned long)zone_idx(zone), | |
5316 | zone_start_pfn, (zone_start_pfn + size)); | |
5317 | ||
1e548deb | 5318 | zone_init_free_lists(zone); |
9dcb8b68 | 5319 | zone->initialized = 1; |
718127cc YG |
5320 | |
5321 | return 0; | |
ed8ece2e DH |
5322 | } |
5323 | ||
0ee332c1 | 5324 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5325 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5326 | |
c713216d MG |
5327 | /* |
5328 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5329 | */ |
8a942fde MG |
5330 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5331 | struct mminit_pfnnid_cache *state) | |
c713216d | 5332 | { |
c13291a5 | 5333 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5334 | int nid; |
7c243c71 | 5335 | |
8a942fde MG |
5336 | if (state->last_start <= pfn && pfn < state->last_end) |
5337 | return state->last_nid; | |
c713216d | 5338 | |
e76b63f8 YL |
5339 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5340 | if (nid != -1) { | |
8a942fde MG |
5341 | state->last_start = start_pfn; |
5342 | state->last_end = end_pfn; | |
5343 | state->last_nid = nid; | |
e76b63f8 YL |
5344 | } |
5345 | ||
5346 | return nid; | |
c713216d MG |
5347 | } |
5348 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5349 | ||
c713216d | 5350 | /** |
6782832e | 5351 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5352 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5353 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5354 | * |
7d018176 ZZ |
5355 | * If an architecture guarantees that all ranges registered contain no holes |
5356 | * and may be freed, this this function may be used instead of calling | |
5357 | * memblock_free_early_nid() manually. | |
c713216d | 5358 | */ |
c13291a5 | 5359 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5360 | { |
c13291a5 TH |
5361 | unsigned long start_pfn, end_pfn; |
5362 | int i, this_nid; | |
edbe7d23 | 5363 | |
c13291a5 TH |
5364 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5365 | start_pfn = min(start_pfn, max_low_pfn); | |
5366 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5367 | |
c13291a5 | 5368 | if (start_pfn < end_pfn) |
6782832e SS |
5369 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5370 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5371 | this_nid); | |
edbe7d23 | 5372 | } |
edbe7d23 | 5373 | } |
edbe7d23 | 5374 | |
c713216d MG |
5375 | /** |
5376 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5377 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5378 | * |
7d018176 ZZ |
5379 | * If an architecture guarantees that all ranges registered contain no holes and may |
5380 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5381 | */ |
5382 | void __init sparse_memory_present_with_active_regions(int nid) | |
5383 | { | |
c13291a5 TH |
5384 | unsigned long start_pfn, end_pfn; |
5385 | int i, this_nid; | |
c713216d | 5386 | |
c13291a5 TH |
5387 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5388 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5389 | } |
5390 | ||
5391 | /** | |
5392 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5393 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5394 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5395 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5396 | * |
5397 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5398 | * provided by memblock_set_node(). If called for a node |
c713216d | 5399 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5400 | * PFNs will be 0. |
c713216d | 5401 | */ |
a3142c8e | 5402 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5403 | unsigned long *start_pfn, unsigned long *end_pfn) |
5404 | { | |
c13291a5 | 5405 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5406 | int i; |
c13291a5 | 5407 | |
c713216d MG |
5408 | *start_pfn = -1UL; |
5409 | *end_pfn = 0; | |
5410 | ||
c13291a5 TH |
5411 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5412 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5413 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5414 | } |
5415 | ||
633c0666 | 5416 | if (*start_pfn == -1UL) |
c713216d | 5417 | *start_pfn = 0; |
c713216d MG |
5418 | } |
5419 | ||
2a1e274a MG |
5420 | /* |
5421 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5422 | * assumption is made that zones within a node are ordered in monotonic | |
5423 | * increasing memory addresses so that the "highest" populated zone is used | |
5424 | */ | |
b69a7288 | 5425 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5426 | { |
5427 | int zone_index; | |
5428 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5429 | if (zone_index == ZONE_MOVABLE) | |
5430 | continue; | |
5431 | ||
5432 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5433 | arch_zone_lowest_possible_pfn[zone_index]) | |
5434 | break; | |
5435 | } | |
5436 | ||
5437 | VM_BUG_ON(zone_index == -1); | |
5438 | movable_zone = zone_index; | |
5439 | } | |
5440 | ||
5441 | /* | |
5442 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5443 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5444 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5445 | * in each node depending on the size of each node and how evenly kernelcore | |
5446 | * is distributed. This helper function adjusts the zone ranges | |
5447 | * provided by the architecture for a given node by using the end of the | |
5448 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5449 | * zones within a node are in order of monotonic increases memory addresses | |
5450 | */ | |
b69a7288 | 5451 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5452 | unsigned long zone_type, |
5453 | unsigned long node_start_pfn, | |
5454 | unsigned long node_end_pfn, | |
5455 | unsigned long *zone_start_pfn, | |
5456 | unsigned long *zone_end_pfn) | |
5457 | { | |
5458 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5459 | if (zone_movable_pfn[nid]) { | |
5460 | /* Size ZONE_MOVABLE */ | |
5461 | if (zone_type == ZONE_MOVABLE) { | |
5462 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5463 | *zone_end_pfn = min(node_end_pfn, | |
5464 | arch_zone_highest_possible_pfn[movable_zone]); | |
5465 | ||
e506b996 XQ |
5466 | /* Adjust for ZONE_MOVABLE starting within this range */ |
5467 | } else if (!mirrored_kernelcore && | |
5468 | *zone_start_pfn < zone_movable_pfn[nid] && | |
5469 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
5470 | *zone_end_pfn = zone_movable_pfn[nid]; | |
5471 | ||
2a1e274a MG |
5472 | /* Check if this whole range is within ZONE_MOVABLE */ |
5473 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5474 | *zone_start_pfn = *zone_end_pfn; | |
5475 | } | |
5476 | } | |
5477 | ||
c713216d MG |
5478 | /* |
5479 | * Return the number of pages a zone spans in a node, including holes | |
5480 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5481 | */ | |
6ea6e688 | 5482 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5483 | unsigned long zone_type, |
7960aedd ZY |
5484 | unsigned long node_start_pfn, |
5485 | unsigned long node_end_pfn, | |
d91749c1 TI |
5486 | unsigned long *zone_start_pfn, |
5487 | unsigned long *zone_end_pfn, | |
c713216d MG |
5488 | unsigned long *ignored) |
5489 | { | |
b5685e92 | 5490 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5491 | if (!node_start_pfn && !node_end_pfn) |
5492 | return 0; | |
5493 | ||
7960aedd | 5494 | /* Get the start and end of the zone */ |
d91749c1 TI |
5495 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5496 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5497 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5498 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5499 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5500 | |
5501 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5502 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5503 | return 0; |
5504 | ||
5505 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5506 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5507 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5508 | |
5509 | /* Return the spanned pages */ | |
d91749c1 | 5510 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5511 | } |
5512 | ||
5513 | /* | |
5514 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5515 | * then all holes in the requested range will be accounted for. |
c713216d | 5516 | */ |
32996250 | 5517 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5518 | unsigned long range_start_pfn, |
5519 | unsigned long range_end_pfn) | |
5520 | { | |
96e907d1 TH |
5521 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5522 | unsigned long start_pfn, end_pfn; | |
5523 | int i; | |
c713216d | 5524 | |
96e907d1 TH |
5525 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5526 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5527 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5528 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5529 | } |
96e907d1 | 5530 | return nr_absent; |
c713216d MG |
5531 | } |
5532 | ||
5533 | /** | |
5534 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5535 | * @start_pfn: The start PFN to start searching for holes | |
5536 | * @end_pfn: The end PFN to stop searching for holes | |
5537 | * | |
88ca3b94 | 5538 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5539 | */ |
5540 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5541 | unsigned long end_pfn) | |
5542 | { | |
5543 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5544 | } | |
5545 | ||
5546 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5547 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5548 | unsigned long zone_type, |
7960aedd ZY |
5549 | unsigned long node_start_pfn, |
5550 | unsigned long node_end_pfn, | |
c713216d MG |
5551 | unsigned long *ignored) |
5552 | { | |
96e907d1 TH |
5553 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5554 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5555 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5556 | unsigned long nr_absent; |
9c7cd687 | 5557 | |
b5685e92 | 5558 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5559 | if (!node_start_pfn && !node_end_pfn) |
5560 | return 0; | |
5561 | ||
96e907d1 TH |
5562 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5563 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5564 | |
2a1e274a MG |
5565 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5566 | node_start_pfn, node_end_pfn, | |
5567 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5568 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5569 | ||
5570 | /* | |
5571 | * ZONE_MOVABLE handling. | |
5572 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
5573 | * and vice versa. | |
5574 | */ | |
e506b996 XQ |
5575 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
5576 | unsigned long start_pfn, end_pfn; | |
5577 | struct memblock_region *r; | |
5578 | ||
5579 | for_each_memblock(memory, r) { | |
5580 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
5581 | zone_start_pfn, zone_end_pfn); | |
5582 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
5583 | zone_start_pfn, zone_end_pfn); | |
5584 | ||
5585 | if (zone_type == ZONE_MOVABLE && | |
5586 | memblock_is_mirror(r)) | |
5587 | nr_absent += end_pfn - start_pfn; | |
5588 | ||
5589 | if (zone_type == ZONE_NORMAL && | |
5590 | !memblock_is_mirror(r)) | |
5591 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
5592 | } |
5593 | } | |
5594 | ||
5595 | return nr_absent; | |
c713216d | 5596 | } |
0e0b864e | 5597 | |
0ee332c1 | 5598 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 5599 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5600 | unsigned long zone_type, |
7960aedd ZY |
5601 | unsigned long node_start_pfn, |
5602 | unsigned long node_end_pfn, | |
d91749c1 TI |
5603 | unsigned long *zone_start_pfn, |
5604 | unsigned long *zone_end_pfn, | |
c713216d MG |
5605 | unsigned long *zones_size) |
5606 | { | |
d91749c1 TI |
5607 | unsigned int zone; |
5608 | ||
5609 | *zone_start_pfn = node_start_pfn; | |
5610 | for (zone = 0; zone < zone_type; zone++) | |
5611 | *zone_start_pfn += zones_size[zone]; | |
5612 | ||
5613 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
5614 | ||
c713216d MG |
5615 | return zones_size[zone_type]; |
5616 | } | |
5617 | ||
6ea6e688 | 5618 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5619 | unsigned long zone_type, |
7960aedd ZY |
5620 | unsigned long node_start_pfn, |
5621 | unsigned long node_end_pfn, | |
c713216d MG |
5622 | unsigned long *zholes_size) |
5623 | { | |
5624 | if (!zholes_size) | |
5625 | return 0; | |
5626 | ||
5627 | return zholes_size[zone_type]; | |
5628 | } | |
20e6926d | 5629 | |
0ee332c1 | 5630 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5631 | |
a3142c8e | 5632 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
5633 | unsigned long node_start_pfn, |
5634 | unsigned long node_end_pfn, | |
5635 | unsigned long *zones_size, | |
5636 | unsigned long *zholes_size) | |
c713216d | 5637 | { |
febd5949 | 5638 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
5639 | enum zone_type i; |
5640 | ||
febd5949 GZ |
5641 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5642 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 5643 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 5644 | unsigned long size, real_size; |
c713216d | 5645 | |
febd5949 GZ |
5646 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5647 | node_start_pfn, | |
5648 | node_end_pfn, | |
d91749c1 TI |
5649 | &zone_start_pfn, |
5650 | &zone_end_pfn, | |
febd5949 GZ |
5651 | zones_size); |
5652 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
5653 | node_start_pfn, node_end_pfn, |
5654 | zholes_size); | |
d91749c1 TI |
5655 | if (size) |
5656 | zone->zone_start_pfn = zone_start_pfn; | |
5657 | else | |
5658 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
5659 | zone->spanned_pages = size; |
5660 | zone->present_pages = real_size; | |
5661 | ||
5662 | totalpages += size; | |
5663 | realtotalpages += real_size; | |
5664 | } | |
5665 | ||
5666 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
5667 | pgdat->node_present_pages = realtotalpages; |
5668 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
5669 | realtotalpages); | |
5670 | } | |
5671 | ||
835c134e MG |
5672 | #ifndef CONFIG_SPARSEMEM |
5673 | /* | |
5674 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
5675 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5676 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
5677 | * round what is now in bits to nearest long in bits, then return it in |
5678 | * bytes. | |
5679 | */ | |
7c45512d | 5680 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
5681 | { |
5682 | unsigned long usemapsize; | |
5683 | ||
7c45512d | 5684 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
5685 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
5686 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
5687 | usemapsize *= NR_PAGEBLOCK_BITS; |
5688 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
5689 | ||
5690 | return usemapsize / 8; | |
5691 | } | |
5692 | ||
5693 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
5694 | struct zone *zone, |
5695 | unsigned long zone_start_pfn, | |
5696 | unsigned long zonesize) | |
835c134e | 5697 | { |
7c45512d | 5698 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 5699 | zone->pageblock_flags = NULL; |
58a01a45 | 5700 | if (usemapsize) |
6782832e SS |
5701 | zone->pageblock_flags = |
5702 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5703 | pgdat->node_id); | |
835c134e MG |
5704 | } |
5705 | #else | |
7c45512d LT |
5706 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5707 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5708 | #endif /* CONFIG_SPARSEMEM */ |
5709 | ||
d9c23400 | 5710 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5711 | |
d9c23400 | 5712 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5713 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5714 | { |
955c1cd7 AM |
5715 | unsigned int order; |
5716 | ||
d9c23400 MG |
5717 | /* Check that pageblock_nr_pages has not already been setup */ |
5718 | if (pageblock_order) | |
5719 | return; | |
5720 | ||
955c1cd7 AM |
5721 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5722 | order = HUGETLB_PAGE_ORDER; | |
5723 | else | |
5724 | order = MAX_ORDER - 1; | |
5725 | ||
d9c23400 MG |
5726 | /* |
5727 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5728 | * This value may be variable depending on boot parameters on IA64 and |
5729 | * powerpc. | |
d9c23400 MG |
5730 | */ |
5731 | pageblock_order = order; | |
5732 | } | |
5733 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5734 | ||
ba72cb8c MG |
5735 | /* |
5736 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5737 | * is unused as pageblock_order is set at compile-time. See |
5738 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5739 | * the kernel config | |
ba72cb8c | 5740 | */ |
15ca220e | 5741 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5742 | { |
ba72cb8c | 5743 | } |
d9c23400 MG |
5744 | |
5745 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5746 | ||
01cefaef JL |
5747 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5748 | unsigned long present_pages) | |
5749 | { | |
5750 | unsigned long pages = spanned_pages; | |
5751 | ||
5752 | /* | |
5753 | * Provide a more accurate estimation if there are holes within | |
5754 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5755 | * zone, each populated memory region may cost us one or two extra | |
5756 | * memmap pages due to alignment because memmap pages for each | |
5757 | * populated regions may not naturally algined on page boundary. | |
5758 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
5759 | */ | |
5760 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5761 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5762 | pages = present_pages; | |
5763 | ||
5764 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5765 | } | |
5766 | ||
1da177e4 LT |
5767 | /* |
5768 | * Set up the zone data structures: | |
5769 | * - mark all pages reserved | |
5770 | * - mark all memory queues empty | |
5771 | * - clear the memory bitmaps | |
6527af5d MK |
5772 | * |
5773 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 5774 | */ |
7f3eb55b | 5775 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 5776 | { |
2f1b6248 | 5777 | enum zone_type j; |
ed8ece2e | 5778 | int nid = pgdat->node_id; |
718127cc | 5779 | int ret; |
1da177e4 | 5780 | |
208d54e5 | 5781 | pgdat_resize_init(pgdat); |
8177a420 AA |
5782 | #ifdef CONFIG_NUMA_BALANCING |
5783 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
5784 | pgdat->numabalancing_migrate_nr_pages = 0; | |
5785 | pgdat->numabalancing_migrate_next_window = jiffies; | |
a3d0a918 KS |
5786 | #endif |
5787 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5788 | spin_lock_init(&pgdat->split_queue_lock); | |
5789 | INIT_LIST_HEAD(&pgdat->split_queue); | |
5790 | pgdat->split_queue_len = 0; | |
8177a420 | 5791 | #endif |
1da177e4 | 5792 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 5793 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
698b1b30 VB |
5794 | #ifdef CONFIG_COMPACTION |
5795 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
5796 | #endif | |
eefa864b | 5797 | pgdat_page_ext_init(pgdat); |
a52633d8 | 5798 | spin_lock_init(&pgdat->lru_lock); |
a9dd0a83 | 5799 | lruvec_init(node_lruvec(pgdat)); |
5f63b720 | 5800 | |
1da177e4 LT |
5801 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5802 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 5803 | unsigned long size, realsize, freesize, memmap_pages; |
d91749c1 | 5804 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 5805 | |
febd5949 GZ |
5806 | size = zone->spanned_pages; |
5807 | realsize = freesize = zone->present_pages; | |
1da177e4 | 5808 | |
0e0b864e | 5809 | /* |
9feedc9d | 5810 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
5811 | * is used by this zone for memmap. This affects the watermark |
5812 | * and per-cpu initialisations | |
5813 | */ | |
01cefaef | 5814 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
5815 | if (!is_highmem_idx(j)) { |
5816 | if (freesize >= memmap_pages) { | |
5817 | freesize -= memmap_pages; | |
5818 | if (memmap_pages) | |
5819 | printk(KERN_DEBUG | |
5820 | " %s zone: %lu pages used for memmap\n", | |
5821 | zone_names[j], memmap_pages); | |
5822 | } else | |
1170532b | 5823 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
5824 | zone_names[j], memmap_pages, freesize); |
5825 | } | |
0e0b864e | 5826 | |
6267276f | 5827 | /* Account for reserved pages */ |
9feedc9d JL |
5828 | if (j == 0 && freesize > dma_reserve) { |
5829 | freesize -= dma_reserve; | |
d903ef9f | 5830 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 5831 | zone_names[0], dma_reserve); |
0e0b864e MG |
5832 | } |
5833 | ||
98d2b0eb | 5834 | if (!is_highmem_idx(j)) |
9feedc9d | 5835 | nr_kernel_pages += freesize; |
01cefaef JL |
5836 | /* Charge for highmem memmap if there are enough kernel pages */ |
5837 | else if (nr_kernel_pages > memmap_pages * 2) | |
5838 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 5839 | nr_all_pages += freesize; |
1da177e4 | 5840 | |
9feedc9d JL |
5841 | /* |
5842 | * Set an approximate value for lowmem here, it will be adjusted | |
5843 | * when the bootmem allocator frees pages into the buddy system. | |
5844 | * And all highmem pages will be managed by the buddy system. | |
5845 | */ | |
5846 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 5847 | #ifdef CONFIG_NUMA |
d5f541ed | 5848 | zone->node = nid; |
9614634f | 5849 | #endif |
1da177e4 | 5850 | zone->name = zone_names[j]; |
a52633d8 | 5851 | zone->zone_pgdat = pgdat; |
1da177e4 | 5852 | spin_lock_init(&zone->lock); |
bdc8cb98 | 5853 | zone_seqlock_init(zone); |
ed8ece2e | 5854 | zone_pcp_init(zone); |
81c0a2bb | 5855 | |
1da177e4 LT |
5856 | if (!size) |
5857 | continue; | |
5858 | ||
955c1cd7 | 5859 | set_pageblock_order(); |
7c45512d | 5860 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
b171e409 | 5861 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
718127cc | 5862 | BUG_ON(ret); |
76cdd58e | 5863 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
5864 | } |
5865 | } | |
5866 | ||
bd721ea7 | 5867 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 5868 | { |
b0aeba74 | 5869 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
5870 | unsigned long __maybe_unused offset = 0; |
5871 | ||
1da177e4 LT |
5872 | /* Skip empty nodes */ |
5873 | if (!pgdat->node_spanned_pages) | |
5874 | return; | |
5875 | ||
d41dee36 | 5876 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
b0aeba74 TL |
5877 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
5878 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
5879 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
5880 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 5881 | unsigned long size, end; |
d41dee36 AW |
5882 | struct page *map; |
5883 | ||
e984bb43 BP |
5884 | /* |
5885 | * The zone's endpoints aren't required to be MAX_ORDER | |
5886 | * aligned but the node_mem_map endpoints must be in order | |
5887 | * for the buddy allocator to function correctly. | |
5888 | */ | |
108bcc96 | 5889 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
5890 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
5891 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
5892 | map = alloc_remap(pgdat->node_id, size); |
5893 | if (!map) | |
6782832e SS |
5894 | map = memblock_virt_alloc_node_nopanic(size, |
5895 | pgdat->node_id); | |
a1c34a3b | 5896 | pgdat->node_mem_map = map + offset; |
1da177e4 | 5897 | } |
12d810c1 | 5898 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
5899 | /* |
5900 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
5901 | */ | |
c713216d | 5902 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 5903 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 5904 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 5905 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 5906 | mem_map -= offset; |
0ee332c1 | 5907 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5908 | } |
1da177e4 | 5909 | #endif |
d41dee36 | 5910 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
5911 | } |
5912 | ||
9109fb7b JW |
5913 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
5914 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 5915 | { |
9109fb7b | 5916 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
5917 | unsigned long start_pfn = 0; |
5918 | unsigned long end_pfn = 0; | |
9109fb7b | 5919 | |
88fdf75d | 5920 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 5921 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 5922 | |
3a80a7fa | 5923 | reset_deferred_meminit(pgdat); |
1da177e4 LT |
5924 | pgdat->node_id = nid; |
5925 | pgdat->node_start_pfn = node_start_pfn; | |
75ef7184 | 5926 | pgdat->per_cpu_nodestats = NULL; |
7960aedd ZY |
5927 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5928 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 5929 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
5930 | (u64)start_pfn << PAGE_SHIFT, |
5931 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
5932 | #else |
5933 | start_pfn = node_start_pfn; | |
7960aedd ZY |
5934 | #endif |
5935 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
5936 | zones_size, zholes_size); | |
1da177e4 LT |
5937 | |
5938 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
5939 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
5940 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
5941 | nid, (unsigned long)pgdat, | |
5942 | (unsigned long)pgdat->node_mem_map); | |
5943 | #endif | |
1da177e4 | 5944 | |
7f3eb55b | 5945 | free_area_init_core(pgdat); |
1da177e4 LT |
5946 | } |
5947 | ||
0ee332c1 | 5948 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
5949 | |
5950 | #if MAX_NUMNODES > 1 | |
5951 | /* | |
5952 | * Figure out the number of possible node ids. | |
5953 | */ | |
f9872caf | 5954 | void __init setup_nr_node_ids(void) |
418508c1 | 5955 | { |
904a9553 | 5956 | unsigned int highest; |
418508c1 | 5957 | |
904a9553 | 5958 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
5959 | nr_node_ids = highest + 1; |
5960 | } | |
418508c1 MS |
5961 | #endif |
5962 | ||
1e01979c TH |
5963 | /** |
5964 | * node_map_pfn_alignment - determine the maximum internode alignment | |
5965 | * | |
5966 | * This function should be called after node map is populated and sorted. | |
5967 | * It calculates the maximum power of two alignment which can distinguish | |
5968 | * all the nodes. | |
5969 | * | |
5970 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
5971 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
5972 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
5973 | * shifted, 1GiB is enough and this function will indicate so. | |
5974 | * | |
5975 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
5976 | * model has fine enough granularity to avoid incorrect mapping for the | |
5977 | * populated node map. | |
5978 | * | |
5979 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
5980 | * requirement (single node). | |
5981 | */ | |
5982 | unsigned long __init node_map_pfn_alignment(void) | |
5983 | { | |
5984 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 5985 | unsigned long start, end, mask; |
1e01979c | 5986 | int last_nid = -1; |
c13291a5 | 5987 | int i, nid; |
1e01979c | 5988 | |
c13291a5 | 5989 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5990 | if (!start || last_nid < 0 || last_nid == nid) { |
5991 | last_nid = nid; | |
5992 | last_end = end; | |
5993 | continue; | |
5994 | } | |
5995 | ||
5996 | /* | |
5997 | * Start with a mask granular enough to pin-point to the | |
5998 | * start pfn and tick off bits one-by-one until it becomes | |
5999 | * too coarse to separate the current node from the last. | |
6000 | */ | |
6001 | mask = ~((1 << __ffs(start)) - 1); | |
6002 | while (mask && last_end <= (start & (mask << 1))) | |
6003 | mask <<= 1; | |
6004 | ||
6005 | /* accumulate all internode masks */ | |
6006 | accl_mask |= mask; | |
6007 | } | |
6008 | ||
6009 | /* convert mask to number of pages */ | |
6010 | return ~accl_mask + 1; | |
6011 | } | |
6012 | ||
a6af2bc3 | 6013 | /* Find the lowest pfn for a node */ |
b69a7288 | 6014 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 6015 | { |
a6af2bc3 | 6016 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
6017 | unsigned long start_pfn; |
6018 | int i; | |
1abbfb41 | 6019 | |
c13291a5 TH |
6020 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
6021 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 6022 | |
a6af2bc3 | 6023 | if (min_pfn == ULONG_MAX) { |
1170532b | 6024 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
6025 | return 0; |
6026 | } | |
6027 | ||
6028 | return min_pfn; | |
c713216d MG |
6029 | } |
6030 | ||
6031 | /** | |
6032 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
6033 | * | |
6034 | * It returns the minimum PFN based on information provided via | |
7d018176 | 6035 | * memblock_set_node(). |
c713216d MG |
6036 | */ |
6037 | unsigned long __init find_min_pfn_with_active_regions(void) | |
6038 | { | |
6039 | return find_min_pfn_for_node(MAX_NUMNODES); | |
6040 | } | |
6041 | ||
37b07e41 LS |
6042 | /* |
6043 | * early_calculate_totalpages() | |
6044 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 6045 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 6046 | */ |
484f51f8 | 6047 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 6048 | { |
7e63efef | 6049 | unsigned long totalpages = 0; |
c13291a5 TH |
6050 | unsigned long start_pfn, end_pfn; |
6051 | int i, nid; | |
6052 | ||
6053 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
6054 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 6055 | |
37b07e41 LS |
6056 | totalpages += pages; |
6057 | if (pages) | |
4b0ef1fe | 6058 | node_set_state(nid, N_MEMORY); |
37b07e41 | 6059 | } |
b8af2941 | 6060 | return totalpages; |
7e63efef MG |
6061 | } |
6062 | ||
2a1e274a MG |
6063 | /* |
6064 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
6065 | * is spread evenly between nodes as long as the nodes have enough | |
6066 | * memory. When they don't, some nodes will have more kernelcore than | |
6067 | * others | |
6068 | */ | |
b224ef85 | 6069 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
6070 | { |
6071 | int i, nid; | |
6072 | unsigned long usable_startpfn; | |
6073 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 6074 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 6075 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 6076 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 6077 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 6078 | struct memblock_region *r; |
b2f3eebe TC |
6079 | |
6080 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
6081 | find_usable_zone_for_movable(); | |
6082 | ||
6083 | /* | |
6084 | * If movable_node is specified, ignore kernelcore and movablecore | |
6085 | * options. | |
6086 | */ | |
6087 | if (movable_node_is_enabled()) { | |
136199f0 EM |
6088 | for_each_memblock(memory, r) { |
6089 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
6090 | continue; |
6091 | ||
136199f0 | 6092 | nid = r->nid; |
b2f3eebe | 6093 | |
136199f0 | 6094 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
6095 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6096 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6097 | usable_startpfn; | |
6098 | } | |
6099 | ||
6100 | goto out2; | |
6101 | } | |
2a1e274a | 6102 | |
342332e6 TI |
6103 | /* |
6104 | * If kernelcore=mirror is specified, ignore movablecore option | |
6105 | */ | |
6106 | if (mirrored_kernelcore) { | |
6107 | bool mem_below_4gb_not_mirrored = false; | |
6108 | ||
6109 | for_each_memblock(memory, r) { | |
6110 | if (memblock_is_mirror(r)) | |
6111 | continue; | |
6112 | ||
6113 | nid = r->nid; | |
6114 | ||
6115 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
6116 | ||
6117 | if (usable_startpfn < 0x100000) { | |
6118 | mem_below_4gb_not_mirrored = true; | |
6119 | continue; | |
6120 | } | |
6121 | ||
6122 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
6123 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6124 | usable_startpfn; | |
6125 | } | |
6126 | ||
6127 | if (mem_below_4gb_not_mirrored) | |
6128 | pr_warn("This configuration results in unmirrored kernel memory."); | |
6129 | ||
6130 | goto out2; | |
6131 | } | |
6132 | ||
7e63efef | 6133 | /* |
b2f3eebe | 6134 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
6135 | * kernelcore that corresponds so that memory usable for |
6136 | * any allocation type is evenly spread. If both kernelcore | |
6137 | * and movablecore are specified, then the value of kernelcore | |
6138 | * will be used for required_kernelcore if it's greater than | |
6139 | * what movablecore would have allowed. | |
6140 | */ | |
6141 | if (required_movablecore) { | |
7e63efef MG |
6142 | unsigned long corepages; |
6143 | ||
6144 | /* | |
6145 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
6146 | * was requested by the user | |
6147 | */ | |
6148 | required_movablecore = | |
6149 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 6150 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
6151 | corepages = totalpages - required_movablecore; |
6152 | ||
6153 | required_kernelcore = max(required_kernelcore, corepages); | |
6154 | } | |
6155 | ||
bde304bd XQ |
6156 | /* |
6157 | * If kernelcore was not specified or kernelcore size is larger | |
6158 | * than totalpages, there is no ZONE_MOVABLE. | |
6159 | */ | |
6160 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 6161 | goto out; |
2a1e274a MG |
6162 | |
6163 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
6164 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6165 | ||
6166 | restart: | |
6167 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
6168 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 6169 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
6170 | unsigned long start_pfn, end_pfn; |
6171 | ||
2a1e274a MG |
6172 | /* |
6173 | * Recalculate kernelcore_node if the division per node | |
6174 | * now exceeds what is necessary to satisfy the requested | |
6175 | * amount of memory for the kernel | |
6176 | */ | |
6177 | if (required_kernelcore < kernelcore_node) | |
6178 | kernelcore_node = required_kernelcore / usable_nodes; | |
6179 | ||
6180 | /* | |
6181 | * As the map is walked, we track how much memory is usable | |
6182 | * by the kernel using kernelcore_remaining. When it is | |
6183 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
6184 | */ | |
6185 | kernelcore_remaining = kernelcore_node; | |
6186 | ||
6187 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 6188 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
6189 | unsigned long size_pages; |
6190 | ||
c13291a5 | 6191 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
6192 | if (start_pfn >= end_pfn) |
6193 | continue; | |
6194 | ||
6195 | /* Account for what is only usable for kernelcore */ | |
6196 | if (start_pfn < usable_startpfn) { | |
6197 | unsigned long kernel_pages; | |
6198 | kernel_pages = min(end_pfn, usable_startpfn) | |
6199 | - start_pfn; | |
6200 | ||
6201 | kernelcore_remaining -= min(kernel_pages, | |
6202 | kernelcore_remaining); | |
6203 | required_kernelcore -= min(kernel_pages, | |
6204 | required_kernelcore); | |
6205 | ||
6206 | /* Continue if range is now fully accounted */ | |
6207 | if (end_pfn <= usable_startpfn) { | |
6208 | ||
6209 | /* | |
6210 | * Push zone_movable_pfn to the end so | |
6211 | * that if we have to rebalance | |
6212 | * kernelcore across nodes, we will | |
6213 | * not double account here | |
6214 | */ | |
6215 | zone_movable_pfn[nid] = end_pfn; | |
6216 | continue; | |
6217 | } | |
6218 | start_pfn = usable_startpfn; | |
6219 | } | |
6220 | ||
6221 | /* | |
6222 | * The usable PFN range for ZONE_MOVABLE is from | |
6223 | * start_pfn->end_pfn. Calculate size_pages as the | |
6224 | * number of pages used as kernelcore | |
6225 | */ | |
6226 | size_pages = end_pfn - start_pfn; | |
6227 | if (size_pages > kernelcore_remaining) | |
6228 | size_pages = kernelcore_remaining; | |
6229 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6230 | ||
6231 | /* | |
6232 | * Some kernelcore has been met, update counts and | |
6233 | * break if the kernelcore for this node has been | |
b8af2941 | 6234 | * satisfied |
2a1e274a MG |
6235 | */ |
6236 | required_kernelcore -= min(required_kernelcore, | |
6237 | size_pages); | |
6238 | kernelcore_remaining -= size_pages; | |
6239 | if (!kernelcore_remaining) | |
6240 | break; | |
6241 | } | |
6242 | } | |
6243 | ||
6244 | /* | |
6245 | * If there is still required_kernelcore, we do another pass with one | |
6246 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6247 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6248 | * satisfied |
2a1e274a MG |
6249 | */ |
6250 | usable_nodes--; | |
6251 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6252 | goto restart; | |
6253 | ||
b2f3eebe | 6254 | out2: |
2a1e274a MG |
6255 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6256 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6257 | zone_movable_pfn[nid] = | |
6258 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6259 | |
20e6926d | 6260 | out: |
66918dcd | 6261 | /* restore the node_state */ |
4b0ef1fe | 6262 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6263 | } |
6264 | ||
4b0ef1fe LJ |
6265 | /* Any regular or high memory on that node ? */ |
6266 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6267 | { |
37b07e41 LS |
6268 | enum zone_type zone_type; |
6269 | ||
4b0ef1fe LJ |
6270 | if (N_MEMORY == N_NORMAL_MEMORY) |
6271 | return; | |
6272 | ||
6273 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 6274 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6275 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
6276 | node_set_state(nid, N_HIGH_MEMORY); |
6277 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
6278 | zone_type <= ZONE_NORMAL) | |
6279 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
6280 | break; |
6281 | } | |
37b07e41 | 6282 | } |
37b07e41 LS |
6283 | } |
6284 | ||
c713216d MG |
6285 | /** |
6286 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6287 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6288 | * |
6289 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6290 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6291 | * zone in each node and their holes is calculated. If the maximum PFN |
6292 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6293 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6294 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6295 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6296 | * at arch_max_dma_pfn. | |
6297 | */ | |
6298 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6299 | { | |
c13291a5 TH |
6300 | unsigned long start_pfn, end_pfn; |
6301 | int i, nid; | |
a6af2bc3 | 6302 | |
c713216d MG |
6303 | /* Record where the zone boundaries are */ |
6304 | memset(arch_zone_lowest_possible_pfn, 0, | |
6305 | sizeof(arch_zone_lowest_possible_pfn)); | |
6306 | memset(arch_zone_highest_possible_pfn, 0, | |
6307 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
6308 | |
6309 | start_pfn = find_min_pfn_with_active_regions(); | |
6310 | ||
6311 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6312 | if (i == ZONE_MOVABLE) |
6313 | continue; | |
90cae1fe OH |
6314 | |
6315 | end_pfn = max(max_zone_pfn[i], start_pfn); | |
6316 | arch_zone_lowest_possible_pfn[i] = start_pfn; | |
6317 | arch_zone_highest_possible_pfn[i] = end_pfn; | |
6318 | ||
6319 | start_pfn = end_pfn; | |
c713216d | 6320 | } |
2a1e274a MG |
6321 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
6322 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
6323 | ||
6324 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6325 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6326 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6327 | |
c713216d | 6328 | /* Print out the zone ranges */ |
f88dfff5 | 6329 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6330 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6331 | if (i == ZONE_MOVABLE) | |
6332 | continue; | |
f88dfff5 | 6333 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6334 | if (arch_zone_lowest_possible_pfn[i] == |
6335 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6336 | pr_cont("empty\n"); |
72f0ba02 | 6337 | else |
8d29e18a JG |
6338 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6339 | (u64)arch_zone_lowest_possible_pfn[i] | |
6340 | << PAGE_SHIFT, | |
6341 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6342 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6343 | } |
6344 | ||
6345 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6346 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6347 | for (i = 0; i < MAX_NUMNODES; i++) { |
6348 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6349 | pr_info(" Node %d: %#018Lx\n", i, |
6350 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6351 | } |
c713216d | 6352 | |
f2d52fe5 | 6353 | /* Print out the early node map */ |
f88dfff5 | 6354 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6355 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6356 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6357 | (u64)start_pfn << PAGE_SHIFT, | |
6358 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6359 | |
6360 | /* Initialise every node */ | |
708614e6 | 6361 | mminit_verify_pageflags_layout(); |
8ef82866 | 6362 | setup_nr_node_ids(); |
c713216d MG |
6363 | for_each_online_node(nid) { |
6364 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6365 | free_area_init_node(nid, NULL, |
c713216d | 6366 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6367 | |
6368 | /* Any memory on that node */ | |
6369 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6370 | node_set_state(nid, N_MEMORY); |
6371 | check_for_memory(pgdat, nid); | |
c713216d MG |
6372 | } |
6373 | } | |
2a1e274a | 6374 | |
7e63efef | 6375 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
6376 | { |
6377 | unsigned long long coremem; | |
6378 | if (!p) | |
6379 | return -EINVAL; | |
6380 | ||
6381 | coremem = memparse(p, &p); | |
7e63efef | 6382 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 6383 | |
7e63efef | 6384 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
6385 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6386 | ||
6387 | return 0; | |
6388 | } | |
ed7ed365 | 6389 | |
7e63efef MG |
6390 | /* |
6391 | * kernelcore=size sets the amount of memory for use for allocations that | |
6392 | * cannot be reclaimed or migrated. | |
6393 | */ | |
6394 | static int __init cmdline_parse_kernelcore(char *p) | |
6395 | { | |
342332e6 TI |
6396 | /* parse kernelcore=mirror */ |
6397 | if (parse_option_str(p, "mirror")) { | |
6398 | mirrored_kernelcore = true; | |
6399 | return 0; | |
6400 | } | |
6401 | ||
7e63efef MG |
6402 | return cmdline_parse_core(p, &required_kernelcore); |
6403 | } | |
6404 | ||
6405 | /* | |
6406 | * movablecore=size sets the amount of memory for use for allocations that | |
6407 | * can be reclaimed or migrated. | |
6408 | */ | |
6409 | static int __init cmdline_parse_movablecore(char *p) | |
6410 | { | |
6411 | return cmdline_parse_core(p, &required_movablecore); | |
6412 | } | |
6413 | ||
ed7ed365 | 6414 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6415 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6416 | |
0ee332c1 | 6417 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6418 | |
c3d5f5f0 JL |
6419 | void adjust_managed_page_count(struct page *page, long count) |
6420 | { | |
6421 | spin_lock(&managed_page_count_lock); | |
6422 | page_zone(page)->managed_pages += count; | |
6423 | totalram_pages += count; | |
3dcc0571 JL |
6424 | #ifdef CONFIG_HIGHMEM |
6425 | if (PageHighMem(page)) | |
6426 | totalhigh_pages += count; | |
6427 | #endif | |
c3d5f5f0 JL |
6428 | spin_unlock(&managed_page_count_lock); |
6429 | } | |
3dcc0571 | 6430 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6431 | |
11199692 | 6432 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6433 | { |
11199692 JL |
6434 | void *pos; |
6435 | unsigned long pages = 0; | |
69afade7 | 6436 | |
11199692 JL |
6437 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6438 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6439 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 6440 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
6441 | memset(pos, poison, PAGE_SIZE); |
6442 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
6443 | } |
6444 | ||
6445 | if (pages && s) | |
adb1fe9a JP |
6446 | pr_info("Freeing %s memory: %ldK\n", |
6447 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
6448 | |
6449 | return pages; | |
6450 | } | |
11199692 | 6451 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6452 | |
cfa11e08 JL |
6453 | #ifdef CONFIG_HIGHMEM |
6454 | void free_highmem_page(struct page *page) | |
6455 | { | |
6456 | __free_reserved_page(page); | |
6457 | totalram_pages++; | |
7b4b2a0d | 6458 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6459 | totalhigh_pages++; |
6460 | } | |
6461 | #endif | |
6462 | ||
7ee3d4e8 JL |
6463 | |
6464 | void __init mem_init_print_info(const char *str) | |
6465 | { | |
6466 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
6467 | unsigned long init_code_size, init_data_size; | |
6468 | ||
6469 | physpages = get_num_physpages(); | |
6470 | codesize = _etext - _stext; | |
6471 | datasize = _edata - _sdata; | |
6472 | rosize = __end_rodata - __start_rodata; | |
6473 | bss_size = __bss_stop - __bss_start; | |
6474 | init_data_size = __init_end - __init_begin; | |
6475 | init_code_size = _einittext - _sinittext; | |
6476 | ||
6477 | /* | |
6478 | * Detect special cases and adjust section sizes accordingly: | |
6479 | * 1) .init.* may be embedded into .data sections | |
6480 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
6481 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
6482 | * 3) .rodata.* may be embedded into .text or .data sections. | |
6483 | */ | |
6484 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
6485 | do { \ |
6486 | if (start <= pos && pos < end && size > adj) \ | |
6487 | size -= adj; \ | |
6488 | } while (0) | |
7ee3d4e8 JL |
6489 | |
6490 | adj_init_size(__init_begin, __init_end, init_data_size, | |
6491 | _sinittext, init_code_size); | |
6492 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
6493 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
6494 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
6495 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
6496 | ||
6497 | #undef adj_init_size | |
6498 | ||
756a025f | 6499 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 6500 | #ifdef CONFIG_HIGHMEM |
756a025f | 6501 | ", %luK highmem" |
7ee3d4e8 | 6502 | #endif |
756a025f JP |
6503 | "%s%s)\n", |
6504 | nr_free_pages() << (PAGE_SHIFT - 10), | |
6505 | physpages << (PAGE_SHIFT - 10), | |
6506 | codesize >> 10, datasize >> 10, rosize >> 10, | |
6507 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
6508 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
6509 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 6510 | #ifdef CONFIG_HIGHMEM |
756a025f | 6511 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 6512 | #endif |
756a025f | 6513 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
6514 | } |
6515 | ||
0e0b864e | 6516 | /** |
88ca3b94 RD |
6517 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6518 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 6519 | * |
013110a7 | 6520 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
6521 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6522 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
6523 | * function may optionally be used to account for unfreeable pages in the |
6524 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
6525 | * smaller per-cpu batchsize. | |
0e0b864e MG |
6526 | */ |
6527 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
6528 | { | |
6529 | dma_reserve = new_dma_reserve; | |
6530 | } | |
6531 | ||
1da177e4 LT |
6532 | void __init free_area_init(unsigned long *zones_size) |
6533 | { | |
9109fb7b | 6534 | free_area_init_node(0, zones_size, |
1da177e4 LT |
6535 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6536 | } | |
1da177e4 | 6537 | |
005fd4bb | 6538 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 6539 | { |
1da177e4 | 6540 | |
005fd4bb SAS |
6541 | lru_add_drain_cpu(cpu); |
6542 | drain_pages(cpu); | |
9f8f2172 | 6543 | |
005fd4bb SAS |
6544 | /* |
6545 | * Spill the event counters of the dead processor | |
6546 | * into the current processors event counters. | |
6547 | * This artificially elevates the count of the current | |
6548 | * processor. | |
6549 | */ | |
6550 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 6551 | |
005fd4bb SAS |
6552 | /* |
6553 | * Zero the differential counters of the dead processor | |
6554 | * so that the vm statistics are consistent. | |
6555 | * | |
6556 | * This is only okay since the processor is dead and cannot | |
6557 | * race with what we are doing. | |
6558 | */ | |
6559 | cpu_vm_stats_fold(cpu); | |
6560 | return 0; | |
1da177e4 | 6561 | } |
1da177e4 LT |
6562 | |
6563 | void __init page_alloc_init(void) | |
6564 | { | |
005fd4bb SAS |
6565 | int ret; |
6566 | ||
6567 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, | |
6568 | "mm/page_alloc:dead", NULL, | |
6569 | page_alloc_cpu_dead); | |
6570 | WARN_ON(ret < 0); | |
1da177e4 LT |
6571 | } |
6572 | ||
cb45b0e9 | 6573 | /* |
34b10060 | 6574 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
6575 | * or min_free_kbytes changes. |
6576 | */ | |
6577 | static void calculate_totalreserve_pages(void) | |
6578 | { | |
6579 | struct pglist_data *pgdat; | |
6580 | unsigned long reserve_pages = 0; | |
2f6726e5 | 6581 | enum zone_type i, j; |
cb45b0e9 HA |
6582 | |
6583 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
6584 | |
6585 | pgdat->totalreserve_pages = 0; | |
6586 | ||
cb45b0e9 HA |
6587 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6588 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 6589 | long max = 0; |
cb45b0e9 HA |
6590 | |
6591 | /* Find valid and maximum lowmem_reserve in the zone */ | |
6592 | for (j = i; j < MAX_NR_ZONES; j++) { | |
6593 | if (zone->lowmem_reserve[j] > max) | |
6594 | max = zone->lowmem_reserve[j]; | |
6595 | } | |
6596 | ||
41858966 MG |
6597 | /* we treat the high watermark as reserved pages. */ |
6598 | max += high_wmark_pages(zone); | |
cb45b0e9 | 6599 | |
b40da049 JL |
6600 | if (max > zone->managed_pages) |
6601 | max = zone->managed_pages; | |
a8d01437 | 6602 | |
281e3726 | 6603 | pgdat->totalreserve_pages += max; |
a8d01437 | 6604 | |
cb45b0e9 HA |
6605 | reserve_pages += max; |
6606 | } | |
6607 | } | |
6608 | totalreserve_pages = reserve_pages; | |
6609 | } | |
6610 | ||
1da177e4 LT |
6611 | /* |
6612 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 6613 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
6614 | * has a correct pages reserved value, so an adequate number of |
6615 | * pages are left in the zone after a successful __alloc_pages(). | |
6616 | */ | |
6617 | static void setup_per_zone_lowmem_reserve(void) | |
6618 | { | |
6619 | struct pglist_data *pgdat; | |
2f6726e5 | 6620 | enum zone_type j, idx; |
1da177e4 | 6621 | |
ec936fc5 | 6622 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
6623 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6624 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 6625 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
6626 | |
6627 | zone->lowmem_reserve[j] = 0; | |
6628 | ||
2f6726e5 CL |
6629 | idx = j; |
6630 | while (idx) { | |
1da177e4 LT |
6631 | struct zone *lower_zone; |
6632 | ||
2f6726e5 CL |
6633 | idx--; |
6634 | ||
1da177e4 LT |
6635 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6636 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
6637 | ||
6638 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 6639 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 6640 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 6641 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
6642 | } |
6643 | } | |
6644 | } | |
cb45b0e9 HA |
6645 | |
6646 | /* update totalreserve_pages */ | |
6647 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6648 | } |
6649 | ||
cfd3da1e | 6650 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
6651 | { |
6652 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
6653 | unsigned long lowmem_pages = 0; | |
6654 | struct zone *zone; | |
6655 | unsigned long flags; | |
6656 | ||
6657 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
6658 | for_each_zone(zone) { | |
6659 | if (!is_highmem(zone)) | |
b40da049 | 6660 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
6661 | } |
6662 | ||
6663 | for_each_zone(zone) { | |
ac924c60 AM |
6664 | u64 tmp; |
6665 | ||
1125b4e3 | 6666 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 6667 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 6668 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
6669 | if (is_highmem(zone)) { |
6670 | /* | |
669ed175 NP |
6671 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6672 | * need highmem pages, so cap pages_min to a small | |
6673 | * value here. | |
6674 | * | |
41858966 | 6675 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 6676 | * deltas control asynch page reclaim, and so should |
669ed175 | 6677 | * not be capped for highmem. |
1da177e4 | 6678 | */ |
90ae8d67 | 6679 | unsigned long min_pages; |
1da177e4 | 6680 | |
b40da049 | 6681 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 6682 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 6683 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 6684 | } else { |
669ed175 NP |
6685 | /* |
6686 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
6687 | * proportionate to the zone's size. |
6688 | */ | |
41858966 | 6689 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
6690 | } |
6691 | ||
795ae7a0 JW |
6692 | /* |
6693 | * Set the kswapd watermarks distance according to the | |
6694 | * scale factor in proportion to available memory, but | |
6695 | * ensure a minimum size on small systems. | |
6696 | */ | |
6697 | tmp = max_t(u64, tmp >> 2, | |
6698 | mult_frac(zone->managed_pages, | |
6699 | watermark_scale_factor, 10000)); | |
6700 | ||
6701 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
6702 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 6703 | |
1125b4e3 | 6704 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 6705 | } |
cb45b0e9 HA |
6706 | |
6707 | /* update totalreserve_pages */ | |
6708 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6709 | } |
6710 | ||
cfd3da1e MG |
6711 | /** |
6712 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
6713 | * or when memory is hot-{added|removed} | |
6714 | * | |
6715 | * Ensures that the watermark[min,low,high] values for each zone are set | |
6716 | * correctly with respect to min_free_kbytes. | |
6717 | */ | |
6718 | void setup_per_zone_wmarks(void) | |
6719 | { | |
6720 | mutex_lock(&zonelists_mutex); | |
6721 | __setup_per_zone_wmarks(); | |
6722 | mutex_unlock(&zonelists_mutex); | |
6723 | } | |
6724 | ||
1da177e4 LT |
6725 | /* |
6726 | * Initialise min_free_kbytes. | |
6727 | * | |
6728 | * For small machines we want it small (128k min). For large machines | |
6729 | * we want it large (64MB max). But it is not linear, because network | |
6730 | * bandwidth does not increase linearly with machine size. We use | |
6731 | * | |
b8af2941 | 6732 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6733 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6734 | * | |
6735 | * which yields | |
6736 | * | |
6737 | * 16MB: 512k | |
6738 | * 32MB: 724k | |
6739 | * 64MB: 1024k | |
6740 | * 128MB: 1448k | |
6741 | * 256MB: 2048k | |
6742 | * 512MB: 2896k | |
6743 | * 1024MB: 4096k | |
6744 | * 2048MB: 5792k | |
6745 | * 4096MB: 8192k | |
6746 | * 8192MB: 11584k | |
6747 | * 16384MB: 16384k | |
6748 | */ | |
1b79acc9 | 6749 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6750 | { |
6751 | unsigned long lowmem_kbytes; | |
5f12733e | 6752 | int new_min_free_kbytes; |
1da177e4 LT |
6753 | |
6754 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6755 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6756 | ||
6757 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6758 | min_free_kbytes = new_min_free_kbytes; | |
6759 | if (min_free_kbytes < 128) | |
6760 | min_free_kbytes = 128; | |
6761 | if (min_free_kbytes > 65536) | |
6762 | min_free_kbytes = 65536; | |
6763 | } else { | |
6764 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6765 | new_min_free_kbytes, user_min_free_kbytes); | |
6766 | } | |
bc75d33f | 6767 | setup_per_zone_wmarks(); |
a6cccdc3 | 6768 | refresh_zone_stat_thresholds(); |
1da177e4 | 6769 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
6770 | |
6771 | #ifdef CONFIG_NUMA | |
6772 | setup_min_unmapped_ratio(); | |
6773 | setup_min_slab_ratio(); | |
6774 | #endif | |
6775 | ||
1da177e4 LT |
6776 | return 0; |
6777 | } | |
bc22af74 | 6778 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
6779 | |
6780 | /* | |
b8af2941 | 6781 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6782 | * that we can call two helper functions whenever min_free_kbytes |
6783 | * changes. | |
6784 | */ | |
cccad5b9 | 6785 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6786 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6787 | { |
da8c757b HP |
6788 | int rc; |
6789 | ||
6790 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6791 | if (rc) | |
6792 | return rc; | |
6793 | ||
5f12733e MH |
6794 | if (write) { |
6795 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6796 | setup_per_zone_wmarks(); |
5f12733e | 6797 | } |
1da177e4 LT |
6798 | return 0; |
6799 | } | |
6800 | ||
795ae7a0 JW |
6801 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
6802 | void __user *buffer, size_t *length, loff_t *ppos) | |
6803 | { | |
6804 | int rc; | |
6805 | ||
6806 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6807 | if (rc) | |
6808 | return rc; | |
6809 | ||
6810 | if (write) | |
6811 | setup_per_zone_wmarks(); | |
6812 | ||
6813 | return 0; | |
6814 | } | |
6815 | ||
9614634f | 6816 | #ifdef CONFIG_NUMA |
6423aa81 | 6817 | static void setup_min_unmapped_ratio(void) |
9614634f | 6818 | { |
6423aa81 | 6819 | pg_data_t *pgdat; |
9614634f | 6820 | struct zone *zone; |
9614634f | 6821 | |
a5f5f91d | 6822 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 6823 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 6824 | |
9614634f | 6825 | for_each_zone(zone) |
a5f5f91d | 6826 | zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages * |
9614634f | 6827 | sysctl_min_unmapped_ratio) / 100; |
9614634f | 6828 | } |
0ff38490 | 6829 | |
6423aa81 JK |
6830 | |
6831 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 6832 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 6833 | { |
0ff38490 CL |
6834 | int rc; |
6835 | ||
8d65af78 | 6836 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
6837 | if (rc) |
6838 | return rc; | |
6839 | ||
6423aa81 JK |
6840 | setup_min_unmapped_ratio(); |
6841 | ||
6842 | return 0; | |
6843 | } | |
6844 | ||
6845 | static void setup_min_slab_ratio(void) | |
6846 | { | |
6847 | pg_data_t *pgdat; | |
6848 | struct zone *zone; | |
6849 | ||
a5f5f91d MG |
6850 | for_each_online_pgdat(pgdat) |
6851 | pgdat->min_slab_pages = 0; | |
6852 | ||
0ff38490 | 6853 | for_each_zone(zone) |
a5f5f91d | 6854 | zone->zone_pgdat->min_slab_pages += (zone->managed_pages * |
0ff38490 | 6855 | sysctl_min_slab_ratio) / 100; |
6423aa81 JK |
6856 | } |
6857 | ||
6858 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
6859 | void __user *buffer, size_t *length, loff_t *ppos) | |
6860 | { | |
6861 | int rc; | |
6862 | ||
6863 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6864 | if (rc) | |
6865 | return rc; | |
6866 | ||
6867 | setup_min_slab_ratio(); | |
6868 | ||
0ff38490 CL |
6869 | return 0; |
6870 | } | |
9614634f CL |
6871 | #endif |
6872 | ||
1da177e4 LT |
6873 | /* |
6874 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
6875 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
6876 | * whenever sysctl_lowmem_reserve_ratio changes. | |
6877 | * | |
6878 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 6879 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
6880 | * if in function of the boot time zone sizes. |
6881 | */ | |
cccad5b9 | 6882 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6883 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6884 | { |
8d65af78 | 6885 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
6886 | setup_per_zone_lowmem_reserve(); |
6887 | return 0; | |
6888 | } | |
6889 | ||
8ad4b1fb RS |
6890 | /* |
6891 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
6892 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
6893 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 6894 | */ |
cccad5b9 | 6895 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6896 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
6897 | { |
6898 | struct zone *zone; | |
7cd2b0a3 | 6899 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
6900 | int ret; |
6901 | ||
7cd2b0a3 DR |
6902 | mutex_lock(&pcp_batch_high_lock); |
6903 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
6904 | ||
8d65af78 | 6905 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
6906 | if (!write || ret < 0) |
6907 | goto out; | |
6908 | ||
6909 | /* Sanity checking to avoid pcp imbalance */ | |
6910 | if (percpu_pagelist_fraction && | |
6911 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
6912 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
6913 | ret = -EINVAL; | |
6914 | goto out; | |
6915 | } | |
6916 | ||
6917 | /* No change? */ | |
6918 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
6919 | goto out; | |
c8e251fa | 6920 | |
364df0eb | 6921 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
6922 | unsigned int cpu; |
6923 | ||
22a7f12b | 6924 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
6925 | pageset_set_high_and_batch(zone, |
6926 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 6927 | } |
7cd2b0a3 | 6928 | out: |
c8e251fa | 6929 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6930 | return ret; |
8ad4b1fb RS |
6931 | } |
6932 | ||
a9919c79 | 6933 | #ifdef CONFIG_NUMA |
f034b5d4 | 6934 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 6935 | |
1da177e4 LT |
6936 | static int __init set_hashdist(char *str) |
6937 | { | |
6938 | if (!str) | |
6939 | return 0; | |
6940 | hashdist = simple_strtoul(str, &str, 0); | |
6941 | return 1; | |
6942 | } | |
6943 | __setup("hashdist=", set_hashdist); | |
6944 | #endif | |
6945 | ||
f6f34b43 SD |
6946 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
6947 | /* | |
6948 | * Returns the number of pages that arch has reserved but | |
6949 | * is not known to alloc_large_system_hash(). | |
6950 | */ | |
6951 | static unsigned long __init arch_reserved_kernel_pages(void) | |
6952 | { | |
6953 | return 0; | |
6954 | } | |
6955 | #endif | |
6956 | ||
1da177e4 LT |
6957 | /* |
6958 | * allocate a large system hash table from bootmem | |
6959 | * - it is assumed that the hash table must contain an exact power-of-2 | |
6960 | * quantity of entries | |
6961 | * - limit is the number of hash buckets, not the total allocation size | |
6962 | */ | |
6963 | void *__init alloc_large_system_hash(const char *tablename, | |
6964 | unsigned long bucketsize, | |
6965 | unsigned long numentries, | |
6966 | int scale, | |
6967 | int flags, | |
6968 | unsigned int *_hash_shift, | |
6969 | unsigned int *_hash_mask, | |
31fe62b9 TB |
6970 | unsigned long low_limit, |
6971 | unsigned long high_limit) | |
1da177e4 | 6972 | { |
31fe62b9 | 6973 | unsigned long long max = high_limit; |
1da177e4 LT |
6974 | unsigned long log2qty, size; |
6975 | void *table = NULL; | |
6976 | ||
6977 | /* allow the kernel cmdline to have a say */ | |
6978 | if (!numentries) { | |
6979 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 6980 | numentries = nr_kernel_pages; |
f6f34b43 | 6981 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
6982 | |
6983 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
6984 | if (PAGE_SHIFT < 20) | |
6985 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
6986 | |
6987 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
6988 | if (scale > PAGE_SHIFT) | |
6989 | numentries >>= (scale - PAGE_SHIFT); | |
6990 | else | |
6991 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
6992 | |
6993 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
6994 | if (unlikely(flags & HASH_SMALL)) { |
6995 | /* Makes no sense without HASH_EARLY */ | |
6996 | WARN_ON(!(flags & HASH_EARLY)); | |
6997 | if (!(numentries >> *_hash_shift)) { | |
6998 | numentries = 1UL << *_hash_shift; | |
6999 | BUG_ON(!numentries); | |
7000 | } | |
7001 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 7002 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 7003 | } |
6e692ed3 | 7004 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
7005 | |
7006 | /* limit allocation size to 1/16 total memory by default */ | |
7007 | if (max == 0) { | |
7008 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
7009 | do_div(max, bucketsize); | |
7010 | } | |
074b8517 | 7011 | max = min(max, 0x80000000ULL); |
1da177e4 | 7012 | |
31fe62b9 TB |
7013 | if (numentries < low_limit) |
7014 | numentries = low_limit; | |
1da177e4 LT |
7015 | if (numentries > max) |
7016 | numentries = max; | |
7017 | ||
f0d1b0b3 | 7018 | log2qty = ilog2(numentries); |
1da177e4 LT |
7019 | |
7020 | do { | |
7021 | size = bucketsize << log2qty; | |
7022 | if (flags & HASH_EARLY) | |
6782832e | 7023 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
7024 | else if (hashdist) |
7025 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
7026 | else { | |
1037b83b ED |
7027 | /* |
7028 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
7029 | * some pages at the end of hash table which |
7030 | * alloc_pages_exact() automatically does | |
1037b83b | 7031 | */ |
264ef8a9 | 7032 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 7033 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
7034 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
7035 | } | |
1da177e4 LT |
7036 | } |
7037 | } while (!table && size > PAGE_SIZE && --log2qty); | |
7038 | ||
7039 | if (!table) | |
7040 | panic("Failed to allocate %s hash table\n", tablename); | |
7041 | ||
1170532b JP |
7042 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
7043 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
7044 | |
7045 | if (_hash_shift) | |
7046 | *_hash_shift = log2qty; | |
7047 | if (_hash_mask) | |
7048 | *_hash_mask = (1 << log2qty) - 1; | |
7049 | ||
7050 | return table; | |
7051 | } | |
a117e66e | 7052 | |
a5d76b54 | 7053 | /* |
80934513 MK |
7054 | * This function checks whether pageblock includes unmovable pages or not. |
7055 | * If @count is not zero, it is okay to include less @count unmovable pages | |
7056 | * | |
b8af2941 | 7057 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
7058 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
7059 | * expect this function should be exact. | |
a5d76b54 | 7060 | */ |
b023f468 WC |
7061 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
7062 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
7063 | { |
7064 | unsigned long pfn, iter, found; | |
47118af0 MN |
7065 | int mt; |
7066 | ||
49ac8255 KH |
7067 | /* |
7068 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 7069 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
7070 | */ |
7071 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 7072 | return false; |
47118af0 MN |
7073 | mt = get_pageblock_migratetype(page); |
7074 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 7075 | return false; |
49ac8255 KH |
7076 | |
7077 | pfn = page_to_pfn(page); | |
7078 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
7079 | unsigned long check = pfn + iter; | |
7080 | ||
29723fcc | 7081 | if (!pfn_valid_within(check)) |
49ac8255 | 7082 | continue; |
29723fcc | 7083 | |
49ac8255 | 7084 | page = pfn_to_page(check); |
c8721bbb NH |
7085 | |
7086 | /* | |
7087 | * Hugepages are not in LRU lists, but they're movable. | |
7088 | * We need not scan over tail pages bacause we don't | |
7089 | * handle each tail page individually in migration. | |
7090 | */ | |
7091 | if (PageHuge(page)) { | |
7092 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
7093 | continue; | |
7094 | } | |
7095 | ||
97d255c8 MK |
7096 | /* |
7097 | * We can't use page_count without pin a page | |
7098 | * because another CPU can free compound page. | |
7099 | * This check already skips compound tails of THP | |
0139aa7b | 7100 | * because their page->_refcount is zero at all time. |
97d255c8 | 7101 | */ |
fe896d18 | 7102 | if (!page_ref_count(page)) { |
49ac8255 KH |
7103 | if (PageBuddy(page)) |
7104 | iter += (1 << page_order(page)) - 1; | |
7105 | continue; | |
7106 | } | |
97d255c8 | 7107 | |
b023f468 WC |
7108 | /* |
7109 | * The HWPoisoned page may be not in buddy system, and | |
7110 | * page_count() is not 0. | |
7111 | */ | |
7112 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
7113 | continue; | |
7114 | ||
49ac8255 KH |
7115 | if (!PageLRU(page)) |
7116 | found++; | |
7117 | /* | |
6b4f7799 JW |
7118 | * If there are RECLAIMABLE pages, we need to check |
7119 | * it. But now, memory offline itself doesn't call | |
7120 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
7121 | */ |
7122 | /* | |
7123 | * If the page is not RAM, page_count()should be 0. | |
7124 | * we don't need more check. This is an _used_ not-movable page. | |
7125 | * | |
7126 | * The problematic thing here is PG_reserved pages. PG_reserved | |
7127 | * is set to both of a memory hole page and a _used_ kernel | |
7128 | * page at boot. | |
7129 | */ | |
7130 | if (found > count) | |
80934513 | 7131 | return true; |
49ac8255 | 7132 | } |
80934513 | 7133 | return false; |
49ac8255 KH |
7134 | } |
7135 | ||
7136 | bool is_pageblock_removable_nolock(struct page *page) | |
7137 | { | |
656a0706 MH |
7138 | struct zone *zone; |
7139 | unsigned long pfn; | |
687875fb MH |
7140 | |
7141 | /* | |
7142 | * We have to be careful here because we are iterating over memory | |
7143 | * sections which are not zone aware so we might end up outside of | |
7144 | * the zone but still within the section. | |
656a0706 MH |
7145 | * We have to take care about the node as well. If the node is offline |
7146 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 7147 | */ |
656a0706 MH |
7148 | if (!node_online(page_to_nid(page))) |
7149 | return false; | |
7150 | ||
7151 | zone = page_zone(page); | |
7152 | pfn = page_to_pfn(page); | |
108bcc96 | 7153 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
7154 | return false; |
7155 | ||
b023f468 | 7156 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 7157 | } |
0c0e6195 | 7158 | |
080fe206 | 7159 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7160 | |
7161 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7162 | { | |
7163 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7164 | pageblock_nr_pages) - 1); | |
7165 | } | |
7166 | ||
7167 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7168 | { | |
7169 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7170 | pageblock_nr_pages)); | |
7171 | } | |
7172 | ||
041d3a8c | 7173 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7174 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7175 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7176 | { |
7177 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7178 | unsigned long nr_reclaimed; |
041d3a8c MN |
7179 | unsigned long pfn = start; |
7180 | unsigned int tries = 0; | |
7181 | int ret = 0; | |
7182 | ||
be49a6e1 | 7183 | migrate_prep(); |
041d3a8c | 7184 | |
bb13ffeb | 7185 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7186 | if (fatal_signal_pending(current)) { |
7187 | ret = -EINTR; | |
7188 | break; | |
7189 | } | |
7190 | ||
bb13ffeb MG |
7191 | if (list_empty(&cc->migratepages)) { |
7192 | cc->nr_migratepages = 0; | |
edc2ca61 | 7193 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7194 | if (!pfn) { |
7195 | ret = -EINTR; | |
7196 | break; | |
7197 | } | |
7198 | tries = 0; | |
7199 | } else if (++tries == 5) { | |
7200 | ret = ret < 0 ? ret : -EBUSY; | |
7201 | break; | |
7202 | } | |
7203 | ||
beb51eaa MK |
7204 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7205 | &cc->migratepages); | |
7206 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7207 | |
9c620e2b | 7208 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 7209 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 7210 | } |
2a6f5124 SP |
7211 | if (ret < 0) { |
7212 | putback_movable_pages(&cc->migratepages); | |
7213 | return ret; | |
7214 | } | |
7215 | return 0; | |
041d3a8c MN |
7216 | } |
7217 | ||
7218 | /** | |
7219 | * alloc_contig_range() -- tries to allocate given range of pages | |
7220 | * @start: start PFN to allocate | |
7221 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7222 | * @migratetype: migratetype of the underlaying pageblocks (either |
7223 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7224 | * in range must have the same migratetype and it must | |
7225 | * be either of the two. | |
041d3a8c MN |
7226 | * |
7227 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
7228 | * aligned, however it's the caller's responsibility to guarantee that | |
7229 | * we are the only thread that changes migrate type of pageblocks the | |
7230 | * pages fall in. | |
7231 | * | |
7232 | * The PFN range must belong to a single zone. | |
7233 | * | |
7234 | * Returns zero on success or negative error code. On success all | |
7235 | * pages which PFN is in [start, end) are allocated for the caller and | |
7236 | * need to be freed with free_contig_range(). | |
7237 | */ | |
0815f3d8 MN |
7238 | int alloc_contig_range(unsigned long start, unsigned long end, |
7239 | unsigned migratetype) | |
041d3a8c | 7240 | { |
041d3a8c | 7241 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7242 | unsigned int order; |
7243 | int ret = 0; | |
041d3a8c | 7244 | |
bb13ffeb MG |
7245 | struct compact_control cc = { |
7246 | .nr_migratepages = 0, | |
7247 | .order = -1, | |
7248 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7249 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
7250 | .ignore_skip_hint = true, |
7251 | }; | |
7252 | INIT_LIST_HEAD(&cc.migratepages); | |
7253 | ||
041d3a8c MN |
7254 | /* |
7255 | * What we do here is we mark all pageblocks in range as | |
7256 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7257 | * have different sizes, and due to the way page allocator | |
7258 | * work, we align the range to biggest of the two pages so | |
7259 | * that page allocator won't try to merge buddies from | |
7260 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7261 | * other migration type. | |
7262 | * | |
7263 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7264 | * migrate the pages from an unaligned range (ie. pages that | |
7265 | * we are interested in). This will put all the pages in | |
7266 | * range back to page allocator as MIGRATE_ISOLATE. | |
7267 | * | |
7268 | * When this is done, we take the pages in range from page | |
7269 | * allocator removing them from the buddy system. This way | |
7270 | * page allocator will never consider using them. | |
7271 | * | |
7272 | * This lets us mark the pageblocks back as | |
7273 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7274 | * aligned range but not in the unaligned, original range are | |
7275 | * put back to page allocator so that buddy can use them. | |
7276 | */ | |
7277 | ||
7278 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7279 | pfn_max_align_up(end), migratetype, |
7280 | false); | |
041d3a8c | 7281 | if (ret) |
86a595f9 | 7282 | return ret; |
041d3a8c | 7283 | |
8ef5849f JK |
7284 | /* |
7285 | * In case of -EBUSY, we'd like to know which page causes problem. | |
7286 | * So, just fall through. We will check it in test_pages_isolated(). | |
7287 | */ | |
bb13ffeb | 7288 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7289 | if (ret && ret != -EBUSY) |
041d3a8c MN |
7290 | goto done; |
7291 | ||
7292 | /* | |
7293 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7294 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7295 | * more, all pages in [start, end) are free in page allocator. | |
7296 | * What we are going to do is to allocate all pages from | |
7297 | * [start, end) (that is remove them from page allocator). | |
7298 | * | |
7299 | * The only problem is that pages at the beginning and at the | |
7300 | * end of interesting range may be not aligned with pages that | |
7301 | * page allocator holds, ie. they can be part of higher order | |
7302 | * pages. Because of this, we reserve the bigger range and | |
7303 | * once this is done free the pages we are not interested in. | |
7304 | * | |
7305 | * We don't have to hold zone->lock here because the pages are | |
7306 | * isolated thus they won't get removed from buddy. | |
7307 | */ | |
7308 | ||
7309 | lru_add_drain_all(); | |
510f5507 | 7310 | drain_all_pages(cc.zone); |
041d3a8c MN |
7311 | |
7312 | order = 0; | |
7313 | outer_start = start; | |
7314 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7315 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7316 | outer_start = start; |
7317 | break; | |
041d3a8c MN |
7318 | } |
7319 | outer_start &= ~0UL << order; | |
7320 | } | |
7321 | ||
8ef5849f JK |
7322 | if (outer_start != start) { |
7323 | order = page_order(pfn_to_page(outer_start)); | |
7324 | ||
7325 | /* | |
7326 | * outer_start page could be small order buddy page and | |
7327 | * it doesn't include start page. Adjust outer_start | |
7328 | * in this case to report failed page properly | |
7329 | * on tracepoint in test_pages_isolated() | |
7330 | */ | |
7331 | if (outer_start + (1UL << order) <= start) | |
7332 | outer_start = start; | |
7333 | } | |
7334 | ||
041d3a8c | 7335 | /* Make sure the range is really isolated. */ |
b023f468 | 7336 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
7337 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
7338 | __func__, outer_start, end); | |
041d3a8c MN |
7339 | ret = -EBUSY; |
7340 | goto done; | |
7341 | } | |
7342 | ||
49f223a9 | 7343 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7344 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7345 | if (!outer_end) { |
7346 | ret = -EBUSY; | |
7347 | goto done; | |
7348 | } | |
7349 | ||
7350 | /* Free head and tail (if any) */ | |
7351 | if (start != outer_start) | |
7352 | free_contig_range(outer_start, start - outer_start); | |
7353 | if (end != outer_end) | |
7354 | free_contig_range(end, outer_end - end); | |
7355 | ||
7356 | done: | |
7357 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7358 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7359 | return ret; |
7360 | } | |
7361 | ||
7362 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7363 | { | |
bcc2b02f MS |
7364 | unsigned int count = 0; |
7365 | ||
7366 | for (; nr_pages--; pfn++) { | |
7367 | struct page *page = pfn_to_page(pfn); | |
7368 | ||
7369 | count += page_count(page) != 1; | |
7370 | __free_page(page); | |
7371 | } | |
7372 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7373 | } |
7374 | #endif | |
7375 | ||
4ed7e022 | 7376 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7377 | /* |
7378 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7379 | * page high values need to be recalulated. | |
7380 | */ | |
4ed7e022 JL |
7381 | void __meminit zone_pcp_update(struct zone *zone) |
7382 | { | |
0a647f38 | 7383 | unsigned cpu; |
c8e251fa | 7384 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7385 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7386 | pageset_set_high_and_batch(zone, |
7387 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7388 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7389 | } |
7390 | #endif | |
7391 | ||
340175b7 JL |
7392 | void zone_pcp_reset(struct zone *zone) |
7393 | { | |
7394 | unsigned long flags; | |
5a883813 MK |
7395 | int cpu; |
7396 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7397 | |
7398 | /* avoid races with drain_pages() */ | |
7399 | local_irq_save(flags); | |
7400 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7401 | for_each_online_cpu(cpu) { |
7402 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7403 | drain_zonestat(zone, pset); | |
7404 | } | |
340175b7 JL |
7405 | free_percpu(zone->pageset); |
7406 | zone->pageset = &boot_pageset; | |
7407 | } | |
7408 | local_irq_restore(flags); | |
7409 | } | |
7410 | ||
6dcd73d7 | 7411 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 7412 | /* |
b9eb6319 JK |
7413 | * All pages in the range must be in a single zone and isolated |
7414 | * before calling this. | |
0c0e6195 KH |
7415 | */ |
7416 | void | |
7417 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7418 | { | |
7419 | struct page *page; | |
7420 | struct zone *zone; | |
7aeb09f9 | 7421 | unsigned int order, i; |
0c0e6195 KH |
7422 | unsigned long pfn; |
7423 | unsigned long flags; | |
7424 | /* find the first valid pfn */ | |
7425 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
7426 | if (pfn_valid(pfn)) | |
7427 | break; | |
7428 | if (pfn == end_pfn) | |
7429 | return; | |
7430 | zone = page_zone(pfn_to_page(pfn)); | |
7431 | spin_lock_irqsave(&zone->lock, flags); | |
7432 | pfn = start_pfn; | |
7433 | while (pfn < end_pfn) { | |
7434 | if (!pfn_valid(pfn)) { | |
7435 | pfn++; | |
7436 | continue; | |
7437 | } | |
7438 | page = pfn_to_page(pfn); | |
b023f468 WC |
7439 | /* |
7440 | * The HWPoisoned page may be not in buddy system, and | |
7441 | * page_count() is not 0. | |
7442 | */ | |
7443 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
7444 | pfn++; | |
7445 | SetPageReserved(page); | |
7446 | continue; | |
7447 | } | |
7448 | ||
0c0e6195 KH |
7449 | BUG_ON(page_count(page)); |
7450 | BUG_ON(!PageBuddy(page)); | |
7451 | order = page_order(page); | |
7452 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
7453 | pr_info("remove from free list %lx %d %lx\n", |
7454 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
7455 | #endif |
7456 | list_del(&page->lru); | |
7457 | rmv_page_order(page); | |
7458 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
7459 | for (i = 0; i < (1 << order); i++) |
7460 | SetPageReserved((page+i)); | |
7461 | pfn += (1 << order); | |
7462 | } | |
7463 | spin_unlock_irqrestore(&zone->lock, flags); | |
7464 | } | |
7465 | #endif | |
8d22ba1b | 7466 | |
8d22ba1b WF |
7467 | bool is_free_buddy_page(struct page *page) |
7468 | { | |
7469 | struct zone *zone = page_zone(page); | |
7470 | unsigned long pfn = page_to_pfn(page); | |
7471 | unsigned long flags; | |
7aeb09f9 | 7472 | unsigned int order; |
8d22ba1b WF |
7473 | |
7474 | spin_lock_irqsave(&zone->lock, flags); | |
7475 | for (order = 0; order < MAX_ORDER; order++) { | |
7476 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
7477 | ||
7478 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
7479 | break; | |
7480 | } | |
7481 | spin_unlock_irqrestore(&zone->lock, flags); | |
7482 | ||
7483 | return order < MAX_ORDER; | |
7484 | } |