]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/stddef.h> |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
10ed273f | 22 | #include <linux/jiffies.h> |
1da177e4 | 23 | #include <linux/bootmem.h> |
edbe7d23 | 24 | #include <linux/memblock.h> |
1da177e4 | 25 | #include <linux/compiler.h> |
9f158333 | 26 | #include <linux/kernel.h> |
b1eeab67 | 27 | #include <linux/kmemcheck.h> |
b8c73fc2 | 28 | #include <linux/kasan.h> |
1da177e4 LT |
29 | #include <linux/module.h> |
30 | #include <linux/suspend.h> | |
31 | #include <linux/pagevec.h> | |
32 | #include <linux/blkdev.h> | |
33 | #include <linux/slab.h> | |
a238ab5b | 34 | #include <linux/ratelimit.h> |
5a3135c2 | 35 | #include <linux/oom.h> |
1da177e4 LT |
36 | #include <linux/notifier.h> |
37 | #include <linux/topology.h> | |
38 | #include <linux/sysctl.h> | |
39 | #include <linux/cpu.h> | |
40 | #include <linux/cpuset.h> | |
bdc8cb98 | 41 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
42 | #include <linux/nodemask.h> |
43 | #include <linux/vmalloc.h> | |
a6cccdc3 | 44 | #include <linux/vmstat.h> |
4be38e35 | 45 | #include <linux/mempolicy.h> |
4b94ffdc | 46 | #include <linux/memremap.h> |
6811378e | 47 | #include <linux/stop_machine.h> |
c713216d MG |
48 | #include <linux/sort.h> |
49 | #include <linux/pfn.h> | |
3fcfab16 | 50 | #include <linux/backing-dev.h> |
933e312e | 51 | #include <linux/fault-inject.h> |
a5d76b54 | 52 | #include <linux/page-isolation.h> |
eefa864b | 53 | #include <linux/page_ext.h> |
3ac7fe5a | 54 | #include <linux/debugobjects.h> |
dbb1f81c | 55 | #include <linux/kmemleak.h> |
56de7263 | 56 | #include <linux/compaction.h> |
0d3d062a | 57 | #include <trace/events/kmem.h> |
d379f01d | 58 | #include <trace/events/oom.h> |
268bb0ce | 59 | #include <linux/prefetch.h> |
6e543d57 | 60 | #include <linux/mm_inline.h> |
041d3a8c | 61 | #include <linux/migrate.h> |
949f7ec5 | 62 | #include <linux/hugetlb.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
5b3cc15a | 64 | #include <linux/sched/mm.h> |
48c96a36 | 65 | #include <linux/page_owner.h> |
0e1cc95b | 66 | #include <linux/kthread.h> |
4949148a | 67 | #include <linux/memcontrol.h> |
42c269c8 | 68 | #include <linux/ftrace.h> |
1da177e4 | 69 | |
7ee3d4e8 | 70 | #include <asm/sections.h> |
1da177e4 | 71 | #include <asm/tlbflush.h> |
ac924c60 | 72 | #include <asm/div64.h> |
1da177e4 LT |
73 | #include "internal.h" |
74 | ||
c8e251fa CS |
75 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
76 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 77 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 78 | |
72812019 LS |
79 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
80 | DEFINE_PER_CPU(int, numa_node); | |
81 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
82 | #endif | |
83 | ||
7aac7898 LS |
84 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
85 | /* | |
86 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
87 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
88 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
89 | * defined in <linux/topology.h>. | |
90 | */ | |
91 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
92 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 93 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
94 | #endif |
95 | ||
bd233f53 MG |
96 | /* work_structs for global per-cpu drains */ |
97 | DEFINE_MUTEX(pcpu_drain_mutex); | |
98 | DEFINE_PER_CPU(struct work_struct, pcpu_drain); | |
99 | ||
38addce8 | 100 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 101 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
102 | EXPORT_SYMBOL(latent_entropy); |
103 | #endif | |
104 | ||
1da177e4 | 105 | /* |
13808910 | 106 | * Array of node states. |
1da177e4 | 107 | */ |
13808910 CL |
108 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
109 | [N_POSSIBLE] = NODE_MASK_ALL, | |
110 | [N_ONLINE] = { { [0] = 1UL } }, | |
111 | #ifndef CONFIG_NUMA | |
112 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
113 | #ifdef CONFIG_HIGHMEM | |
114 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
115 | #endif |
116 | #ifdef CONFIG_MOVABLE_NODE | |
117 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
118 | #endif |
119 | [N_CPU] = { { [0] = 1UL } }, | |
120 | #endif /* NUMA */ | |
121 | }; | |
122 | EXPORT_SYMBOL(node_states); | |
123 | ||
c3d5f5f0 JL |
124 | /* Protect totalram_pages and zone->managed_pages */ |
125 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
126 | ||
6c231b7b | 127 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 128 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 129 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 130 | |
1b76b02f | 131 | int percpu_pagelist_fraction; |
dcce284a | 132 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 133 | |
bb14c2c7 VB |
134 | /* |
135 | * A cached value of the page's pageblock's migratetype, used when the page is | |
136 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
137 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
138 | * Also the migratetype set in the page does not necessarily match the pcplist | |
139 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
140 | * other index - this ensures that it will be put on the correct CMA freelist. | |
141 | */ | |
142 | static inline int get_pcppage_migratetype(struct page *page) | |
143 | { | |
144 | return page->index; | |
145 | } | |
146 | ||
147 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
148 | { | |
149 | page->index = migratetype; | |
150 | } | |
151 | ||
452aa699 RW |
152 | #ifdef CONFIG_PM_SLEEP |
153 | /* | |
154 | * The following functions are used by the suspend/hibernate code to temporarily | |
155 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
156 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
157 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
158 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
159 | * guaranteed not to run in parallel with that modification). | |
160 | */ | |
c9e664f1 RW |
161 | |
162 | static gfp_t saved_gfp_mask; | |
163 | ||
164 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
165 | { |
166 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
167 | if (saved_gfp_mask) { |
168 | gfp_allowed_mask = saved_gfp_mask; | |
169 | saved_gfp_mask = 0; | |
170 | } | |
452aa699 RW |
171 | } |
172 | ||
c9e664f1 | 173 | void pm_restrict_gfp_mask(void) |
452aa699 | 174 | { |
452aa699 | 175 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
176 | WARN_ON(saved_gfp_mask); |
177 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 178 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 179 | } |
f90ac398 MG |
180 | |
181 | bool pm_suspended_storage(void) | |
182 | { | |
d0164adc | 183 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
184 | return false; |
185 | return true; | |
186 | } | |
452aa699 RW |
187 | #endif /* CONFIG_PM_SLEEP */ |
188 | ||
d9c23400 | 189 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 190 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
191 | #endif |
192 | ||
d98c7a09 | 193 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 194 | |
1da177e4 LT |
195 | /* |
196 | * results with 256, 32 in the lowmem_reserve sysctl: | |
197 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
198 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
199 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
200 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 201 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
202 | * |
203 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
204 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 205 | */ |
2f1b6248 | 206 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 207 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 208 | 256, |
4b51d669 | 209 | #endif |
fb0e7942 | 210 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 211 | 256, |
fb0e7942 | 212 | #endif |
e53ef38d | 213 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 214 | 32, |
e53ef38d | 215 | #endif |
2a1e274a | 216 | 32, |
2f1b6248 | 217 | }; |
1da177e4 LT |
218 | |
219 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 220 | |
15ad7cdc | 221 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 222 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 223 | "DMA", |
4b51d669 | 224 | #endif |
fb0e7942 | 225 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 226 | "DMA32", |
fb0e7942 | 227 | #endif |
2f1b6248 | 228 | "Normal", |
e53ef38d | 229 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 230 | "HighMem", |
e53ef38d | 231 | #endif |
2a1e274a | 232 | "Movable", |
033fbae9 DW |
233 | #ifdef CONFIG_ZONE_DEVICE |
234 | "Device", | |
235 | #endif | |
2f1b6248 CL |
236 | }; |
237 | ||
60f30350 VB |
238 | char * const migratetype_names[MIGRATE_TYPES] = { |
239 | "Unmovable", | |
240 | "Movable", | |
241 | "Reclaimable", | |
242 | "HighAtomic", | |
243 | #ifdef CONFIG_CMA | |
244 | "CMA", | |
245 | #endif | |
246 | #ifdef CONFIG_MEMORY_ISOLATION | |
247 | "Isolate", | |
248 | #endif | |
249 | }; | |
250 | ||
f1e61557 KS |
251 | compound_page_dtor * const compound_page_dtors[] = { |
252 | NULL, | |
253 | free_compound_page, | |
254 | #ifdef CONFIG_HUGETLB_PAGE | |
255 | free_huge_page, | |
256 | #endif | |
9a982250 KS |
257 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
258 | free_transhuge_page, | |
259 | #endif | |
f1e61557 KS |
260 | }; |
261 | ||
1da177e4 | 262 | int min_free_kbytes = 1024; |
42aa83cb | 263 | int user_min_free_kbytes = -1; |
795ae7a0 | 264 | int watermark_scale_factor = 10; |
1da177e4 | 265 | |
2c85f51d JB |
266 | static unsigned long __meminitdata nr_kernel_pages; |
267 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 268 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 269 | |
0ee332c1 TH |
270 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
271 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
272 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
273 | static unsigned long __initdata required_kernelcore; | |
274 | static unsigned long __initdata required_movablecore; | |
275 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
342332e6 | 276 | static bool mirrored_kernelcore; |
0ee332c1 TH |
277 | |
278 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
279 | int movable_zone; | |
280 | EXPORT_SYMBOL(movable_zone); | |
281 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 282 | |
418508c1 MS |
283 | #if MAX_NUMNODES > 1 |
284 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 285 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 286 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 287 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
288 | #endif |
289 | ||
9ef9acb0 MG |
290 | int page_group_by_mobility_disabled __read_mostly; |
291 | ||
3a80a7fa MG |
292 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
293 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
294 | { | |
864b9a39 MH |
295 | unsigned long max_initialise; |
296 | unsigned long reserved_lowmem; | |
297 | ||
298 | /* | |
299 | * Initialise at least 2G of a node but also take into account that | |
300 | * two large system hashes that can take up 1GB for 0.25TB/node. | |
301 | */ | |
302 | max_initialise = max(2UL << (30 - PAGE_SHIFT), | |
303 | (pgdat->node_spanned_pages >> 8)); | |
304 | ||
305 | /* | |
306 | * Compensate the all the memblock reservations (e.g. crash kernel) | |
307 | * from the initial estimation to make sure we will initialize enough | |
308 | * memory to boot. | |
309 | */ | |
310 | reserved_lowmem = memblock_reserved_memory_within(pgdat->node_start_pfn, | |
311 | pgdat->node_start_pfn + max_initialise); | |
312 | max_initialise += reserved_lowmem; | |
313 | ||
314 | pgdat->static_init_size = min(max_initialise, pgdat->node_spanned_pages); | |
3a80a7fa MG |
315 | pgdat->first_deferred_pfn = ULONG_MAX; |
316 | } | |
317 | ||
318 | /* Returns true if the struct page for the pfn is uninitialised */ | |
0e1cc95b | 319 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 320 | { |
ef70b6f4 MG |
321 | int nid = early_pfn_to_nid(pfn); |
322 | ||
323 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
324 | return true; |
325 | ||
326 | return false; | |
327 | } | |
328 | ||
329 | /* | |
330 | * Returns false when the remaining initialisation should be deferred until | |
331 | * later in the boot cycle when it can be parallelised. | |
332 | */ | |
333 | static inline bool update_defer_init(pg_data_t *pgdat, | |
334 | unsigned long pfn, unsigned long zone_end, | |
335 | unsigned long *nr_initialised) | |
336 | { | |
337 | /* Always populate low zones for address-contrained allocations */ | |
338 | if (zone_end < pgdat_end_pfn(pgdat)) | |
339 | return true; | |
3a80a7fa | 340 | (*nr_initialised)++; |
864b9a39 | 341 | if ((*nr_initialised > pgdat->static_init_size) && |
3a80a7fa MG |
342 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
343 | pgdat->first_deferred_pfn = pfn; | |
344 | return false; | |
345 | } | |
346 | ||
347 | return true; | |
348 | } | |
349 | #else | |
350 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
351 | { | |
352 | } | |
353 | ||
354 | static inline bool early_page_uninitialised(unsigned long pfn) | |
355 | { | |
356 | return false; | |
357 | } | |
358 | ||
359 | static inline bool update_defer_init(pg_data_t *pgdat, | |
360 | unsigned long pfn, unsigned long zone_end, | |
361 | unsigned long *nr_initialised) | |
362 | { | |
363 | return true; | |
364 | } | |
365 | #endif | |
366 | ||
0b423ca2 MG |
367 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
368 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
369 | unsigned long pfn) | |
370 | { | |
371 | #ifdef CONFIG_SPARSEMEM | |
372 | return __pfn_to_section(pfn)->pageblock_flags; | |
373 | #else | |
374 | return page_zone(page)->pageblock_flags; | |
375 | #endif /* CONFIG_SPARSEMEM */ | |
376 | } | |
377 | ||
378 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
379 | { | |
380 | #ifdef CONFIG_SPARSEMEM | |
381 | pfn &= (PAGES_PER_SECTION-1); | |
382 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
383 | #else | |
384 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
385 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
386 | #endif /* CONFIG_SPARSEMEM */ | |
387 | } | |
388 | ||
389 | /** | |
390 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
391 | * @page: The page within the block of interest | |
392 | * @pfn: The target page frame number | |
393 | * @end_bitidx: The last bit of interest to retrieve | |
394 | * @mask: mask of bits that the caller is interested in | |
395 | * | |
396 | * Return: pageblock_bits flags | |
397 | */ | |
398 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
399 | unsigned long pfn, | |
400 | unsigned long end_bitidx, | |
401 | unsigned long mask) | |
402 | { | |
403 | unsigned long *bitmap; | |
404 | unsigned long bitidx, word_bitidx; | |
405 | unsigned long word; | |
406 | ||
407 | bitmap = get_pageblock_bitmap(page, pfn); | |
408 | bitidx = pfn_to_bitidx(page, pfn); | |
409 | word_bitidx = bitidx / BITS_PER_LONG; | |
410 | bitidx &= (BITS_PER_LONG-1); | |
411 | ||
412 | word = bitmap[word_bitidx]; | |
413 | bitidx += end_bitidx; | |
414 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
415 | } | |
416 | ||
417 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
418 | unsigned long end_bitidx, | |
419 | unsigned long mask) | |
420 | { | |
421 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
422 | } | |
423 | ||
424 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
425 | { | |
426 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
427 | } | |
428 | ||
429 | /** | |
430 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
431 | * @page: The page within the block of interest | |
432 | * @flags: The flags to set | |
433 | * @pfn: The target page frame number | |
434 | * @end_bitidx: The last bit of interest | |
435 | * @mask: mask of bits that the caller is interested in | |
436 | */ | |
437 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
438 | unsigned long pfn, | |
439 | unsigned long end_bitidx, | |
440 | unsigned long mask) | |
441 | { | |
442 | unsigned long *bitmap; | |
443 | unsigned long bitidx, word_bitidx; | |
444 | unsigned long old_word, word; | |
445 | ||
446 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
447 | ||
448 | bitmap = get_pageblock_bitmap(page, pfn); | |
449 | bitidx = pfn_to_bitidx(page, pfn); | |
450 | word_bitidx = bitidx / BITS_PER_LONG; | |
451 | bitidx &= (BITS_PER_LONG-1); | |
452 | ||
453 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
454 | ||
455 | bitidx += end_bitidx; | |
456 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
457 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
458 | ||
459 | word = READ_ONCE(bitmap[word_bitidx]); | |
460 | for (;;) { | |
461 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
462 | if (word == old_word) | |
463 | break; | |
464 | word = old_word; | |
465 | } | |
466 | } | |
3a80a7fa | 467 | |
ee6f509c | 468 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 469 | { |
5d0f3f72 KM |
470 | if (unlikely(page_group_by_mobility_disabled && |
471 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
472 | migratetype = MIGRATE_UNMOVABLE; |
473 | ||
b2a0ac88 MG |
474 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
475 | PB_migrate, PB_migrate_end); | |
476 | } | |
477 | ||
13e7444b | 478 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 479 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 480 | { |
bdc8cb98 DH |
481 | int ret = 0; |
482 | unsigned seq; | |
483 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 484 | unsigned long sp, start_pfn; |
c6a57e19 | 485 | |
bdc8cb98 DH |
486 | do { |
487 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
488 | start_pfn = zone->zone_start_pfn; |
489 | sp = zone->spanned_pages; | |
108bcc96 | 490 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
491 | ret = 1; |
492 | } while (zone_span_seqretry(zone, seq)); | |
493 | ||
b5e6a5a2 | 494 | if (ret) |
613813e8 DH |
495 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
496 | pfn, zone_to_nid(zone), zone->name, | |
497 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 498 | |
bdc8cb98 | 499 | return ret; |
c6a57e19 DH |
500 | } |
501 | ||
502 | static int page_is_consistent(struct zone *zone, struct page *page) | |
503 | { | |
14e07298 | 504 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 505 | return 0; |
1da177e4 | 506 | if (zone != page_zone(page)) |
c6a57e19 DH |
507 | return 0; |
508 | ||
509 | return 1; | |
510 | } | |
511 | /* | |
512 | * Temporary debugging check for pages not lying within a given zone. | |
513 | */ | |
d73d3c9f | 514 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
515 | { |
516 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 517 | return 1; |
c6a57e19 DH |
518 | if (!page_is_consistent(zone, page)) |
519 | return 1; | |
520 | ||
1da177e4 LT |
521 | return 0; |
522 | } | |
13e7444b | 523 | #else |
d73d3c9f | 524 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
525 | { |
526 | return 0; | |
527 | } | |
528 | #endif | |
529 | ||
d230dec1 KS |
530 | static void bad_page(struct page *page, const char *reason, |
531 | unsigned long bad_flags) | |
1da177e4 | 532 | { |
d936cf9b HD |
533 | static unsigned long resume; |
534 | static unsigned long nr_shown; | |
535 | static unsigned long nr_unshown; | |
536 | ||
537 | /* | |
538 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
539 | * or allow a steady drip of one report per second. | |
540 | */ | |
541 | if (nr_shown == 60) { | |
542 | if (time_before(jiffies, resume)) { | |
543 | nr_unshown++; | |
544 | goto out; | |
545 | } | |
546 | if (nr_unshown) { | |
ff8e8116 | 547 | pr_alert( |
1e9e6365 | 548 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
549 | nr_unshown); |
550 | nr_unshown = 0; | |
551 | } | |
552 | nr_shown = 0; | |
553 | } | |
554 | if (nr_shown++ == 0) | |
555 | resume = jiffies + 60 * HZ; | |
556 | ||
ff8e8116 | 557 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 558 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
559 | __dump_page(page, reason); |
560 | bad_flags &= page->flags; | |
561 | if (bad_flags) | |
562 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
563 | bad_flags, &bad_flags); | |
4e462112 | 564 | dump_page_owner(page); |
3dc14741 | 565 | |
4f31888c | 566 | print_modules(); |
1da177e4 | 567 | dump_stack(); |
d936cf9b | 568 | out: |
8cc3b392 | 569 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 570 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 571 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
572 | } |
573 | ||
1da177e4 LT |
574 | /* |
575 | * Higher-order pages are called "compound pages". They are structured thusly: | |
576 | * | |
1d798ca3 | 577 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 578 | * |
1d798ca3 KS |
579 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
580 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 581 | * |
1d798ca3 KS |
582 | * The first tail page's ->compound_dtor holds the offset in array of compound |
583 | * page destructors. See compound_page_dtors. | |
1da177e4 | 584 | * |
1d798ca3 | 585 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 586 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 587 | */ |
d98c7a09 | 588 | |
9a982250 | 589 | void free_compound_page(struct page *page) |
d98c7a09 | 590 | { |
d85f3385 | 591 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
592 | } |
593 | ||
d00181b9 | 594 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
595 | { |
596 | int i; | |
597 | int nr_pages = 1 << order; | |
598 | ||
f1e61557 | 599 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
600 | set_compound_order(page, order); |
601 | __SetPageHead(page); | |
602 | for (i = 1; i < nr_pages; i++) { | |
603 | struct page *p = page + i; | |
58a84aa9 | 604 | set_page_count(p, 0); |
1c290f64 | 605 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 606 | set_compound_head(p, page); |
18229df5 | 607 | } |
53f9263b | 608 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
609 | } |
610 | ||
c0a32fc5 SG |
611 | #ifdef CONFIG_DEBUG_PAGEALLOC |
612 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
613 | bool _debug_pagealloc_enabled __read_mostly |
614 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 615 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
616 | bool _debug_guardpage_enabled __read_mostly; |
617 | ||
031bc574 JK |
618 | static int __init early_debug_pagealloc(char *buf) |
619 | { | |
620 | if (!buf) | |
621 | return -EINVAL; | |
2a138dc7 | 622 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
031bc574 JK |
623 | } |
624 | early_param("debug_pagealloc", early_debug_pagealloc); | |
625 | ||
e30825f1 JK |
626 | static bool need_debug_guardpage(void) |
627 | { | |
031bc574 JK |
628 | /* If we don't use debug_pagealloc, we don't need guard page */ |
629 | if (!debug_pagealloc_enabled()) | |
630 | return false; | |
631 | ||
f1c1e9f7 JK |
632 | if (!debug_guardpage_minorder()) |
633 | return false; | |
634 | ||
e30825f1 JK |
635 | return true; |
636 | } | |
637 | ||
638 | static void init_debug_guardpage(void) | |
639 | { | |
031bc574 JK |
640 | if (!debug_pagealloc_enabled()) |
641 | return; | |
642 | ||
f1c1e9f7 JK |
643 | if (!debug_guardpage_minorder()) |
644 | return; | |
645 | ||
e30825f1 JK |
646 | _debug_guardpage_enabled = true; |
647 | } | |
648 | ||
649 | struct page_ext_operations debug_guardpage_ops = { | |
650 | .need = need_debug_guardpage, | |
651 | .init = init_debug_guardpage, | |
652 | }; | |
c0a32fc5 SG |
653 | |
654 | static int __init debug_guardpage_minorder_setup(char *buf) | |
655 | { | |
656 | unsigned long res; | |
657 | ||
658 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 659 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
660 | return 0; |
661 | } | |
662 | _debug_guardpage_minorder = res; | |
1170532b | 663 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
664 | return 0; |
665 | } | |
f1c1e9f7 | 666 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 667 | |
acbc15a4 | 668 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 669 | unsigned int order, int migratetype) |
c0a32fc5 | 670 | { |
e30825f1 JK |
671 | struct page_ext *page_ext; |
672 | ||
673 | if (!debug_guardpage_enabled()) | |
acbc15a4 JK |
674 | return false; |
675 | ||
676 | if (order >= debug_guardpage_minorder()) | |
677 | return false; | |
e30825f1 JK |
678 | |
679 | page_ext = lookup_page_ext(page); | |
f86e4271 | 680 | if (unlikely(!page_ext)) |
acbc15a4 | 681 | return false; |
f86e4271 | 682 | |
e30825f1 JK |
683 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
684 | ||
2847cf95 JK |
685 | INIT_LIST_HEAD(&page->lru); |
686 | set_page_private(page, order); | |
687 | /* Guard pages are not available for any usage */ | |
688 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
689 | |
690 | return true; | |
c0a32fc5 SG |
691 | } |
692 | ||
2847cf95 JK |
693 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
694 | unsigned int order, int migratetype) | |
c0a32fc5 | 695 | { |
e30825f1 JK |
696 | struct page_ext *page_ext; |
697 | ||
698 | if (!debug_guardpage_enabled()) | |
699 | return; | |
700 | ||
701 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
702 | if (unlikely(!page_ext)) |
703 | return; | |
704 | ||
e30825f1 JK |
705 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
706 | ||
2847cf95 JK |
707 | set_page_private(page, 0); |
708 | if (!is_migrate_isolate(migratetype)) | |
709 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
710 | } |
711 | #else | |
980ac167 | 712 | struct page_ext_operations debug_guardpage_ops; |
acbc15a4 JK |
713 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
714 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
715 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
716 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
717 | #endif |
718 | ||
7aeb09f9 | 719 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 720 | { |
4c21e2f2 | 721 | set_page_private(page, order); |
676165a8 | 722 | __SetPageBuddy(page); |
1da177e4 LT |
723 | } |
724 | ||
725 | static inline void rmv_page_order(struct page *page) | |
726 | { | |
676165a8 | 727 | __ClearPageBuddy(page); |
4c21e2f2 | 728 | set_page_private(page, 0); |
1da177e4 LT |
729 | } |
730 | ||
1da177e4 LT |
731 | /* |
732 | * This function checks whether a page is free && is the buddy | |
733 | * we can do coalesce a page and its buddy if | |
13ad59df | 734 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 735 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
736 | * (c) a page and its buddy have the same order && |
737 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 738 | * |
cf6fe945 WSH |
739 | * For recording whether a page is in the buddy system, we set ->_mapcount |
740 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
741 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
742 | * serialized by zone->lock. | |
1da177e4 | 743 | * |
676165a8 | 744 | * For recording page's order, we use page_private(page). |
1da177e4 | 745 | */ |
cb2b95e1 | 746 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 747 | unsigned int order) |
1da177e4 | 748 | { |
c0a32fc5 | 749 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
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 | ||
c0a32fc5 SG |
755 | return 1; |
756 | } | |
757 | ||
cb2b95e1 | 758 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
759 | /* |
760 | * zone check is done late to avoid uselessly | |
761 | * calculating zone/node ids for pages that could | |
762 | * never merge. | |
763 | */ | |
764 | if (page_zone_id(page) != page_zone_id(buddy)) | |
765 | return 0; | |
766 | ||
4c5018ce WY |
767 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
768 | ||
6aa3001b | 769 | return 1; |
676165a8 | 770 | } |
6aa3001b | 771 | return 0; |
1da177e4 LT |
772 | } |
773 | ||
774 | /* | |
775 | * Freeing function for a buddy system allocator. | |
776 | * | |
777 | * The concept of a buddy system is to maintain direct-mapped table | |
778 | * (containing bit values) for memory blocks of various "orders". | |
779 | * The bottom level table contains the map for the smallest allocatable | |
780 | * units of memory (here, pages), and each level above it describes | |
781 | * pairs of units from the levels below, hence, "buddies". | |
782 | * At a high level, all that happens here is marking the table entry | |
783 | * at the bottom level available, and propagating the changes upward | |
784 | * as necessary, plus some accounting needed to play nicely with other | |
785 | * parts of the VM system. | |
786 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
787 | * free pages of length of (1 << order) and marked with _mapcount |
788 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
789 | * field. | |
1da177e4 | 790 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
791 | * other. That is, if we allocate a small block, and both were |
792 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 793 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 794 | * triggers coalescing into a block of larger size. |
1da177e4 | 795 | * |
6d49e352 | 796 | * -- nyc |
1da177e4 LT |
797 | */ |
798 | ||
48db57f8 | 799 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 800 | unsigned long pfn, |
ed0ae21d MG |
801 | struct zone *zone, unsigned int order, |
802 | int migratetype) | |
1da177e4 | 803 | { |
76741e77 VB |
804 | unsigned long combined_pfn; |
805 | unsigned long uninitialized_var(buddy_pfn); | |
6dda9d55 | 806 | struct page *buddy; |
d9dddbf5 VB |
807 | unsigned int max_order; |
808 | ||
809 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 810 | |
d29bb978 | 811 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 812 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 813 | |
ed0ae21d | 814 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 815 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 816 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 817 | |
76741e77 | 818 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 819 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 820 | |
d9dddbf5 | 821 | continue_merging: |
3c605096 | 822 | while (order < max_order - 1) { |
76741e77 VB |
823 | buddy_pfn = __find_buddy_pfn(pfn, order); |
824 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
825 | |
826 | if (!pfn_valid_within(buddy_pfn)) | |
827 | goto done_merging; | |
cb2b95e1 | 828 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 829 | goto done_merging; |
c0a32fc5 SG |
830 | /* |
831 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
832 | * merge with it and move up one order. | |
833 | */ | |
834 | if (page_is_guard(buddy)) { | |
2847cf95 | 835 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
836 | } else { |
837 | list_del(&buddy->lru); | |
838 | zone->free_area[order].nr_free--; | |
839 | rmv_page_order(buddy); | |
840 | } | |
76741e77 VB |
841 | combined_pfn = buddy_pfn & pfn; |
842 | page = page + (combined_pfn - pfn); | |
843 | pfn = combined_pfn; | |
1da177e4 LT |
844 | order++; |
845 | } | |
d9dddbf5 VB |
846 | if (max_order < MAX_ORDER) { |
847 | /* If we are here, it means order is >= pageblock_order. | |
848 | * We want to prevent merge between freepages on isolate | |
849 | * pageblock and normal pageblock. Without this, pageblock | |
850 | * isolation could cause incorrect freepage or CMA accounting. | |
851 | * | |
852 | * We don't want to hit this code for the more frequent | |
853 | * low-order merging. | |
854 | */ | |
855 | if (unlikely(has_isolate_pageblock(zone))) { | |
856 | int buddy_mt; | |
857 | ||
76741e77 VB |
858 | buddy_pfn = __find_buddy_pfn(pfn, order); |
859 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
860 | buddy_mt = get_pageblock_migratetype(buddy); |
861 | ||
862 | if (migratetype != buddy_mt | |
863 | && (is_migrate_isolate(migratetype) || | |
864 | is_migrate_isolate(buddy_mt))) | |
865 | goto done_merging; | |
866 | } | |
867 | max_order++; | |
868 | goto continue_merging; | |
869 | } | |
870 | ||
871 | done_merging: | |
1da177e4 | 872 | set_page_order(page, order); |
6dda9d55 CZ |
873 | |
874 | /* | |
875 | * If this is not the largest possible page, check if the buddy | |
876 | * of the next-highest order is free. If it is, it's possible | |
877 | * that pages are being freed that will coalesce soon. In case, | |
878 | * that is happening, add the free page to the tail of the list | |
879 | * so it's less likely to be used soon and more likely to be merged | |
880 | * as a higher order page | |
881 | */ | |
13ad59df | 882 | if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) { |
6dda9d55 | 883 | struct page *higher_page, *higher_buddy; |
76741e77 VB |
884 | combined_pfn = buddy_pfn & pfn; |
885 | higher_page = page + (combined_pfn - pfn); | |
886 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
887 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
b4fb8f66 TL |
888 | if (pfn_valid_within(buddy_pfn) && |
889 | page_is_buddy(higher_page, higher_buddy, order + 1)) { | |
6dda9d55 CZ |
890 | list_add_tail(&page->lru, |
891 | &zone->free_area[order].free_list[migratetype]); | |
892 | goto out; | |
893 | } | |
894 | } | |
895 | ||
896 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
897 | out: | |
1da177e4 LT |
898 | zone->free_area[order].nr_free++; |
899 | } | |
900 | ||
7bfec6f4 MG |
901 | /* |
902 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
903 | * try and check multiple fields with one check. The caller must do a detailed | |
904 | * check if necessary. | |
905 | */ | |
906 | static inline bool page_expected_state(struct page *page, | |
907 | unsigned long check_flags) | |
908 | { | |
909 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
910 | return false; | |
911 | ||
912 | if (unlikely((unsigned long)page->mapping | | |
913 | page_ref_count(page) | | |
914 | #ifdef CONFIG_MEMCG | |
915 | (unsigned long)page->mem_cgroup | | |
916 | #endif | |
917 | (page->flags & check_flags))) | |
918 | return false; | |
919 | ||
920 | return true; | |
921 | } | |
922 | ||
bb552ac6 | 923 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 924 | { |
7bfec6f4 MG |
925 | const char *bad_reason; |
926 | unsigned long bad_flags; | |
927 | ||
7bfec6f4 MG |
928 | bad_reason = NULL; |
929 | bad_flags = 0; | |
f0b791a3 | 930 | |
53f9263b | 931 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
932 | bad_reason = "nonzero mapcount"; |
933 | if (unlikely(page->mapping != NULL)) | |
934 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 935 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 936 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
937 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
938 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
939 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
940 | } | |
9edad6ea JW |
941 | #ifdef CONFIG_MEMCG |
942 | if (unlikely(page->mem_cgroup)) | |
943 | bad_reason = "page still charged to cgroup"; | |
944 | #endif | |
7bfec6f4 | 945 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
946 | } |
947 | ||
948 | static inline int free_pages_check(struct page *page) | |
949 | { | |
da838d4f | 950 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 951 | return 0; |
bb552ac6 MG |
952 | |
953 | /* Something has gone sideways, find it */ | |
954 | free_pages_check_bad(page); | |
7bfec6f4 | 955 | return 1; |
1da177e4 LT |
956 | } |
957 | ||
4db7548c MG |
958 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
959 | { | |
960 | int ret = 1; | |
961 | ||
962 | /* | |
963 | * We rely page->lru.next never has bit 0 set, unless the page | |
964 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
965 | */ | |
966 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
967 | ||
968 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
969 | ret = 0; | |
970 | goto out; | |
971 | } | |
972 | switch (page - head_page) { | |
973 | case 1: | |
974 | /* the first tail page: ->mapping is compound_mapcount() */ | |
975 | if (unlikely(compound_mapcount(page))) { | |
976 | bad_page(page, "nonzero compound_mapcount", 0); | |
977 | goto out; | |
978 | } | |
979 | break; | |
980 | case 2: | |
981 | /* | |
982 | * the second tail page: ->mapping is | |
983 | * page_deferred_list().next -- ignore value. | |
984 | */ | |
985 | break; | |
986 | default: | |
987 | if (page->mapping != TAIL_MAPPING) { | |
988 | bad_page(page, "corrupted mapping in tail page", 0); | |
989 | goto out; | |
990 | } | |
991 | break; | |
992 | } | |
993 | if (unlikely(!PageTail(page))) { | |
994 | bad_page(page, "PageTail not set", 0); | |
995 | goto out; | |
996 | } | |
997 | if (unlikely(compound_head(page) != head_page)) { | |
998 | bad_page(page, "compound_head not consistent", 0); | |
999 | goto out; | |
1000 | } | |
1001 | ret = 0; | |
1002 | out: | |
1003 | page->mapping = NULL; | |
1004 | clear_compound_head(page); | |
1005 | return ret; | |
1006 | } | |
1007 | ||
e2769dbd MG |
1008 | static __always_inline bool free_pages_prepare(struct page *page, |
1009 | unsigned int order, bool check_free) | |
4db7548c | 1010 | { |
e2769dbd | 1011 | int bad = 0; |
4db7548c | 1012 | |
4db7548c MG |
1013 | VM_BUG_ON_PAGE(PageTail(page), page); |
1014 | ||
e2769dbd MG |
1015 | trace_mm_page_free(page, order); |
1016 | kmemcheck_free_shadow(page, order); | |
e2769dbd MG |
1017 | |
1018 | /* | |
1019 | * Check tail pages before head page information is cleared to | |
1020 | * avoid checking PageCompound for order-0 pages. | |
1021 | */ | |
1022 | if (unlikely(order)) { | |
1023 | bool compound = PageCompound(page); | |
1024 | int i; | |
1025 | ||
1026 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1027 | |
9a73f61b KS |
1028 | if (compound) |
1029 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1030 | for (i = 1; i < (1 << order); i++) { |
1031 | if (compound) | |
1032 | bad += free_tail_pages_check(page, page + i); | |
1033 | if (unlikely(free_pages_check(page + i))) { | |
1034 | bad++; | |
1035 | continue; | |
1036 | } | |
1037 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1038 | } | |
1039 | } | |
bda807d4 | 1040 | if (PageMappingFlags(page)) |
4db7548c | 1041 | page->mapping = NULL; |
c4159a75 | 1042 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
4949148a | 1043 | memcg_kmem_uncharge(page, order); |
e2769dbd MG |
1044 | if (check_free) |
1045 | bad += free_pages_check(page); | |
1046 | if (bad) | |
1047 | return false; | |
4db7548c | 1048 | |
e2769dbd MG |
1049 | page_cpupid_reset_last(page); |
1050 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1051 | reset_page_owner(page, order); | |
4db7548c MG |
1052 | |
1053 | if (!PageHighMem(page)) { | |
1054 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1055 | PAGE_SIZE << order); |
4db7548c | 1056 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1057 | PAGE_SIZE << order); |
4db7548c | 1058 | } |
e2769dbd MG |
1059 | arch_free_page(page, order); |
1060 | kernel_poison_pages(page, 1 << order, 0); | |
1061 | kernel_map_pages(page, 1 << order, 0); | |
29b52de1 | 1062 | kasan_free_pages(page, order); |
4db7548c | 1063 | |
4db7548c MG |
1064 | return true; |
1065 | } | |
1066 | ||
e2769dbd MG |
1067 | #ifdef CONFIG_DEBUG_VM |
1068 | static inline bool free_pcp_prepare(struct page *page) | |
1069 | { | |
1070 | return free_pages_prepare(page, 0, true); | |
1071 | } | |
1072 | ||
1073 | static inline bool bulkfree_pcp_prepare(struct page *page) | |
1074 | { | |
1075 | return false; | |
1076 | } | |
1077 | #else | |
1078 | static bool free_pcp_prepare(struct page *page) | |
1079 | { | |
1080 | return free_pages_prepare(page, 0, false); | |
1081 | } | |
1082 | ||
4db7548c MG |
1083 | static bool bulkfree_pcp_prepare(struct page *page) |
1084 | { | |
1085 | return free_pages_check(page); | |
1086 | } | |
1087 | #endif /* CONFIG_DEBUG_VM */ | |
1088 | ||
1da177e4 | 1089 | /* |
5f8dcc21 | 1090 | * Frees a number of pages from the PCP lists |
1da177e4 | 1091 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1092 | * count is the number of pages to free. |
1da177e4 LT |
1093 | * |
1094 | * If the zone was previously in an "all pages pinned" state then look to | |
1095 | * see if this freeing clears that state. | |
1096 | * | |
1097 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1098 | * pinned" detection logic. | |
1099 | */ | |
5f8dcc21 MG |
1100 | static void free_pcppages_bulk(struct zone *zone, int count, |
1101 | struct per_cpu_pages *pcp) | |
1da177e4 | 1102 | { |
5f8dcc21 | 1103 | int migratetype = 0; |
a6f9edd6 | 1104 | int batch_free = 0; |
3777999d | 1105 | bool isolated_pageblocks; |
5f8dcc21 | 1106 | |
d34b0733 | 1107 | spin_lock(&zone->lock); |
3777999d | 1108 | isolated_pageblocks = has_isolate_pageblock(zone); |
f2260e6b | 1109 | |
e5b31ac2 | 1110 | while (count) { |
48db57f8 | 1111 | struct page *page; |
5f8dcc21 MG |
1112 | struct list_head *list; |
1113 | ||
1114 | /* | |
a6f9edd6 MG |
1115 | * Remove pages from lists in a round-robin fashion. A |
1116 | * batch_free count is maintained that is incremented when an | |
1117 | * empty list is encountered. This is so more pages are freed | |
1118 | * off fuller lists instead of spinning excessively around empty | |
1119 | * lists | |
5f8dcc21 MG |
1120 | */ |
1121 | do { | |
a6f9edd6 | 1122 | batch_free++; |
5f8dcc21 MG |
1123 | if (++migratetype == MIGRATE_PCPTYPES) |
1124 | migratetype = 0; | |
1125 | list = &pcp->lists[migratetype]; | |
1126 | } while (list_empty(list)); | |
48db57f8 | 1127 | |
1d16871d NK |
1128 | /* This is the only non-empty list. Free them all. */ |
1129 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1130 | batch_free = count; |
1d16871d | 1131 | |
a6f9edd6 | 1132 | do { |
770c8aaa BZ |
1133 | int mt; /* migratetype of the to-be-freed page */ |
1134 | ||
a16601c5 | 1135 | page = list_last_entry(list, struct page, lru); |
a6f9edd6 MG |
1136 | /* must delete as __free_one_page list manipulates */ |
1137 | list_del(&page->lru); | |
aa016d14 | 1138 | |
bb14c2c7 | 1139 | mt = get_pcppage_migratetype(page); |
aa016d14 VB |
1140 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1141 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1142 | /* Pageblock could have been isolated meanwhile */ | |
3777999d | 1143 | if (unlikely(isolated_pageblocks)) |
51bb1a40 | 1144 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 1145 | |
4db7548c MG |
1146 | if (bulkfree_pcp_prepare(page)) |
1147 | continue; | |
1148 | ||
dc4b0caf | 1149 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 1150 | trace_mm_page_pcpu_drain(page, 0, mt); |
e5b31ac2 | 1151 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1152 | } |
d34b0733 | 1153 | spin_unlock(&zone->lock); |
1da177e4 LT |
1154 | } |
1155 | ||
dc4b0caf MG |
1156 | static void free_one_page(struct zone *zone, |
1157 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1158 | unsigned int order, |
ed0ae21d | 1159 | int migratetype) |
1da177e4 | 1160 | { |
d34b0733 | 1161 | spin_lock(&zone->lock); |
ad53f92e JK |
1162 | if (unlikely(has_isolate_pageblock(zone) || |
1163 | is_migrate_isolate(migratetype))) { | |
1164 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1165 | } |
dc4b0caf | 1166 | __free_one_page(page, pfn, zone, order, migratetype); |
d34b0733 | 1167 | spin_unlock(&zone->lock); |
48db57f8 NP |
1168 | } |
1169 | ||
1e8ce83c RH |
1170 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
1171 | unsigned long zone, int nid) | |
1172 | { | |
1e8ce83c | 1173 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1174 | init_page_count(page); |
1175 | page_mapcount_reset(page); | |
1176 | page_cpupid_reset_last(page); | |
1e8ce83c | 1177 | |
1e8ce83c RH |
1178 | INIT_LIST_HEAD(&page->lru); |
1179 | #ifdef WANT_PAGE_VIRTUAL | |
1180 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1181 | if (!is_highmem_idx(zone)) | |
1182 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1183 | #endif | |
1184 | } | |
1185 | ||
1186 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
1187 | int nid) | |
1188 | { | |
1189 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
1190 | } | |
1191 | ||
7e18adb4 MG |
1192 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1193 | static void init_reserved_page(unsigned long pfn) | |
1194 | { | |
1195 | pg_data_t *pgdat; | |
1196 | int nid, zid; | |
1197 | ||
1198 | if (!early_page_uninitialised(pfn)) | |
1199 | return; | |
1200 | ||
1201 | nid = early_pfn_to_nid(pfn); | |
1202 | pgdat = NODE_DATA(nid); | |
1203 | ||
1204 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1205 | struct zone *zone = &pgdat->node_zones[zid]; | |
1206 | ||
1207 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1208 | break; | |
1209 | } | |
1210 | __init_single_pfn(pfn, zid, nid); | |
1211 | } | |
1212 | #else | |
1213 | static inline void init_reserved_page(unsigned long pfn) | |
1214 | { | |
1215 | } | |
1216 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1217 | ||
92923ca3 NZ |
1218 | /* |
1219 | * Initialised pages do not have PageReserved set. This function is | |
1220 | * called for each range allocated by the bootmem allocator and | |
1221 | * marks the pages PageReserved. The remaining valid pages are later | |
1222 | * sent to the buddy page allocator. | |
1223 | */ | |
4b50bcc7 | 1224 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1225 | { |
1226 | unsigned long start_pfn = PFN_DOWN(start); | |
1227 | unsigned long end_pfn = PFN_UP(end); | |
1228 | ||
7e18adb4 MG |
1229 | for (; start_pfn < end_pfn; start_pfn++) { |
1230 | if (pfn_valid(start_pfn)) { | |
1231 | struct page *page = pfn_to_page(start_pfn); | |
1232 | ||
1233 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1234 | |
1235 | /* Avoid false-positive PageTail() */ | |
1236 | INIT_LIST_HEAD(&page->lru); | |
1237 | ||
7e18adb4 MG |
1238 | SetPageReserved(page); |
1239 | } | |
1240 | } | |
92923ca3 NZ |
1241 | } |
1242 | ||
ec95f53a KM |
1243 | static void __free_pages_ok(struct page *page, unsigned int order) |
1244 | { | |
d34b0733 | 1245 | unsigned long flags; |
95e34412 | 1246 | int migratetype; |
dc4b0caf | 1247 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1248 | |
e2769dbd | 1249 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1250 | return; |
1251 | ||
cfc47a28 | 1252 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1253 | local_irq_save(flags); |
1254 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1255 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1256 | local_irq_restore(flags); |
1da177e4 LT |
1257 | } |
1258 | ||
949698a3 | 1259 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1260 | { |
c3993076 | 1261 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1262 | struct page *p = page; |
c3993076 | 1263 | unsigned int loop; |
a226f6c8 | 1264 | |
e2d0bd2b YL |
1265 | prefetchw(p); |
1266 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1267 | prefetchw(p + 1); | |
c3993076 JW |
1268 | __ClearPageReserved(p); |
1269 | set_page_count(p, 0); | |
a226f6c8 | 1270 | } |
e2d0bd2b YL |
1271 | __ClearPageReserved(p); |
1272 | set_page_count(p, 0); | |
c3993076 | 1273 | |
e2d0bd2b | 1274 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1275 | set_page_refcounted(page); |
1276 | __free_pages(page, order); | |
a226f6c8 DH |
1277 | } |
1278 | ||
75a592a4 MG |
1279 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1280 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1281 | |
75a592a4 MG |
1282 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1283 | ||
1284 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1285 | { | |
7ace9917 | 1286 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1287 | int nid; |
1288 | ||
7ace9917 | 1289 | spin_lock(&early_pfn_lock); |
75a592a4 | 1290 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1291 | if (nid < 0) |
e4568d38 | 1292 | nid = first_online_node; |
7ace9917 MG |
1293 | spin_unlock(&early_pfn_lock); |
1294 | ||
1295 | return nid; | |
75a592a4 MG |
1296 | } |
1297 | #endif | |
1298 | ||
1299 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
d73d3c9f MK |
1300 | static inline bool __meminit __maybe_unused |
1301 | meminit_pfn_in_nid(unsigned long pfn, int node, | |
1302 | struct mminit_pfnnid_cache *state) | |
75a592a4 MG |
1303 | { |
1304 | int nid; | |
1305 | ||
1306 | nid = __early_pfn_to_nid(pfn, state); | |
1307 | if (nid >= 0 && nid != node) | |
1308 | return false; | |
1309 | return true; | |
1310 | } | |
1311 | ||
1312 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1313 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1314 | { | |
1315 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1316 | } | |
1317 | ||
1318 | #else | |
1319 | ||
1320 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1321 | { | |
1322 | return true; | |
1323 | } | |
d73d3c9f MK |
1324 | static inline bool __meminit __maybe_unused |
1325 | meminit_pfn_in_nid(unsigned long pfn, int node, | |
1326 | struct mminit_pfnnid_cache *state) | |
75a592a4 MG |
1327 | { |
1328 | return true; | |
1329 | } | |
1330 | #endif | |
1331 | ||
1332 | ||
0e1cc95b | 1333 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1334 | unsigned int order) |
1335 | { | |
1336 | if (early_page_uninitialised(pfn)) | |
1337 | return; | |
949698a3 | 1338 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1339 | } |
1340 | ||
7cf91a98 JK |
1341 | /* |
1342 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1343 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1344 | * with the migration of free compaction scanner. The scanners then need to | |
1345 | * use only pfn_valid_within() check for arches that allow holes within | |
1346 | * pageblocks. | |
1347 | * | |
1348 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1349 | * | |
1350 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1351 | * i.e. it's possible that all pages within a zones range of pages do not | |
1352 | * belong to a single zone. We assume that a border between node0 and node1 | |
1353 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1354 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1355 | * the first and last page of a pageblock and avoid checking each individual | |
1356 | * page in a pageblock. | |
1357 | */ | |
1358 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1359 | unsigned long end_pfn, struct zone *zone) | |
1360 | { | |
1361 | struct page *start_page; | |
1362 | struct page *end_page; | |
1363 | ||
1364 | /* end_pfn is one past the range we are checking */ | |
1365 | end_pfn--; | |
1366 | ||
1367 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1368 | return NULL; | |
1369 | ||
2d070eab MH |
1370 | start_page = pfn_to_online_page(start_pfn); |
1371 | if (!start_page) | |
1372 | return NULL; | |
7cf91a98 JK |
1373 | |
1374 | if (page_zone(start_page) != zone) | |
1375 | return NULL; | |
1376 | ||
1377 | end_page = pfn_to_page(end_pfn); | |
1378 | ||
1379 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1380 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1381 | return NULL; | |
1382 | ||
1383 | return start_page; | |
1384 | } | |
1385 | ||
1386 | void set_zone_contiguous(struct zone *zone) | |
1387 | { | |
1388 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1389 | unsigned long block_end_pfn; | |
1390 | ||
1391 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1392 | for (; block_start_pfn < zone_end_pfn(zone); | |
1393 | block_start_pfn = block_end_pfn, | |
1394 | block_end_pfn += pageblock_nr_pages) { | |
1395 | ||
1396 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1397 | ||
1398 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1399 | block_end_pfn, zone)) | |
1400 | return; | |
1401 | } | |
1402 | ||
1403 | /* We confirm that there is no hole */ | |
1404 | zone->contiguous = true; | |
1405 | } | |
1406 | ||
1407 | void clear_zone_contiguous(struct zone *zone) | |
1408 | { | |
1409 | zone->contiguous = false; | |
1410 | } | |
1411 | ||
7e18adb4 | 1412 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
0e1cc95b | 1413 | static void __init deferred_free_range(struct page *page, |
a4de83dd MG |
1414 | unsigned long pfn, int nr_pages) |
1415 | { | |
1416 | int i; | |
1417 | ||
1418 | if (!page) | |
1419 | return; | |
1420 | ||
1421 | /* Free a large naturally-aligned chunk if possible */ | |
e780149b XQ |
1422 | if (nr_pages == pageblock_nr_pages && |
1423 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1424 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
e780149b | 1425 | __free_pages_boot_core(page, pageblock_order); |
a4de83dd MG |
1426 | return; |
1427 | } | |
1428 | ||
e780149b XQ |
1429 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1430 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1431 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
949698a3 | 1432 | __free_pages_boot_core(page, 0); |
e780149b | 1433 | } |
a4de83dd MG |
1434 | } |
1435 | ||
d3cd131d NS |
1436 | /* Completion tracking for deferred_init_memmap() threads */ |
1437 | static atomic_t pgdat_init_n_undone __initdata; | |
1438 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1439 | ||
1440 | static inline void __init pgdat_init_report_one_done(void) | |
1441 | { | |
1442 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1443 | complete(&pgdat_init_all_done_comp); | |
1444 | } | |
0e1cc95b | 1445 | |
7e18adb4 | 1446 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1447 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1448 | { |
0e1cc95b MG |
1449 | pg_data_t *pgdat = data; |
1450 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1451 | struct mminit_pfnnid_cache nid_init_state = { }; |
1452 | unsigned long start = jiffies; | |
1453 | unsigned long nr_pages = 0; | |
1454 | unsigned long walk_start, walk_end; | |
1455 | int i, zid; | |
1456 | struct zone *zone; | |
7e18adb4 | 1457 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
0e1cc95b | 1458 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
7e18adb4 | 1459 | |
0e1cc95b | 1460 | if (first_init_pfn == ULONG_MAX) { |
d3cd131d | 1461 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1462 | return 0; |
1463 | } | |
1464 | ||
1465 | /* Bind memory initialisation thread to a local node if possible */ | |
1466 | if (!cpumask_empty(cpumask)) | |
1467 | set_cpus_allowed_ptr(current, cpumask); | |
7e18adb4 MG |
1468 | |
1469 | /* Sanity check boundaries */ | |
1470 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1471 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1472 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1473 | ||
1474 | /* Only the highest zone is deferred so find it */ | |
1475 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1476 | zone = pgdat->node_zones + zid; | |
1477 | if (first_init_pfn < zone_end_pfn(zone)) | |
1478 | break; | |
1479 | } | |
1480 | ||
1481 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { | |
1482 | unsigned long pfn, end_pfn; | |
54608c3f | 1483 | struct page *page = NULL; |
a4de83dd MG |
1484 | struct page *free_base_page = NULL; |
1485 | unsigned long free_base_pfn = 0; | |
1486 | int nr_to_free = 0; | |
7e18adb4 MG |
1487 | |
1488 | end_pfn = min(walk_end, zone_end_pfn(zone)); | |
1489 | pfn = first_init_pfn; | |
1490 | if (pfn < walk_start) | |
1491 | pfn = walk_start; | |
1492 | if (pfn < zone->zone_start_pfn) | |
1493 | pfn = zone->zone_start_pfn; | |
1494 | ||
1495 | for (; pfn < end_pfn; pfn++) { | |
54608c3f | 1496 | if (!pfn_valid_within(pfn)) |
a4de83dd | 1497 | goto free_range; |
7e18adb4 | 1498 | |
54608c3f MG |
1499 | /* |
1500 | * Ensure pfn_valid is checked every | |
e780149b | 1501 | * pageblock_nr_pages for memory holes |
54608c3f | 1502 | */ |
e780149b | 1503 | if ((pfn & (pageblock_nr_pages - 1)) == 0) { |
54608c3f MG |
1504 | if (!pfn_valid(pfn)) { |
1505 | page = NULL; | |
a4de83dd | 1506 | goto free_range; |
54608c3f MG |
1507 | } |
1508 | } | |
1509 | ||
1510 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { | |
1511 | page = NULL; | |
a4de83dd | 1512 | goto free_range; |
54608c3f MG |
1513 | } |
1514 | ||
1515 | /* Minimise pfn page lookups and scheduler checks */ | |
e780149b | 1516 | if (page && (pfn & (pageblock_nr_pages - 1)) != 0) { |
54608c3f MG |
1517 | page++; |
1518 | } else { | |
a4de83dd MG |
1519 | nr_pages += nr_to_free; |
1520 | deferred_free_range(free_base_page, | |
1521 | free_base_pfn, nr_to_free); | |
1522 | free_base_page = NULL; | |
1523 | free_base_pfn = nr_to_free = 0; | |
1524 | ||
54608c3f MG |
1525 | page = pfn_to_page(pfn); |
1526 | cond_resched(); | |
1527 | } | |
7e18adb4 MG |
1528 | |
1529 | if (page->flags) { | |
1530 | VM_BUG_ON(page_zone(page) != zone); | |
a4de83dd | 1531 | goto free_range; |
7e18adb4 MG |
1532 | } |
1533 | ||
1534 | __init_single_page(page, pfn, zid, nid); | |
a4de83dd MG |
1535 | if (!free_base_page) { |
1536 | free_base_page = page; | |
1537 | free_base_pfn = pfn; | |
1538 | nr_to_free = 0; | |
1539 | } | |
1540 | nr_to_free++; | |
1541 | ||
1542 | /* Where possible, batch up pages for a single free */ | |
1543 | continue; | |
1544 | free_range: | |
1545 | /* Free the current block of pages to allocator */ | |
1546 | nr_pages += nr_to_free; | |
1547 | deferred_free_range(free_base_page, free_base_pfn, | |
1548 | nr_to_free); | |
1549 | free_base_page = NULL; | |
1550 | free_base_pfn = nr_to_free = 0; | |
7e18adb4 | 1551 | } |
e780149b XQ |
1552 | /* Free the last block of pages to allocator */ |
1553 | nr_pages += nr_to_free; | |
1554 | deferred_free_range(free_base_page, free_base_pfn, nr_to_free); | |
a4de83dd | 1555 | |
7e18adb4 MG |
1556 | first_init_pfn = max(end_pfn, first_init_pfn); |
1557 | } | |
1558 | ||
1559 | /* Sanity check that the next zone really is unpopulated */ | |
1560 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1561 | ||
0e1cc95b | 1562 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1563 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1564 | |
1565 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1566 | return 0; |
1567 | } | |
7cf91a98 | 1568 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1569 | |
1570 | void __init page_alloc_init_late(void) | |
1571 | { | |
7cf91a98 JK |
1572 | struct zone *zone; |
1573 | ||
1574 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1575 | int nid; |
1576 | ||
d3cd131d NS |
1577 | /* There will be num_node_state(N_MEMORY) threads */ |
1578 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1579 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1580 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1581 | } | |
1582 | ||
1583 | /* Block until all are initialised */ | |
d3cd131d | 1584 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da MG |
1585 | |
1586 | /* Reinit limits that are based on free pages after the kernel is up */ | |
1587 | files_maxfiles_init(); | |
7cf91a98 JK |
1588 | #endif |
1589 | ||
1590 | for_each_populated_zone(zone) | |
1591 | set_zone_contiguous(zone); | |
7e18adb4 | 1592 | } |
7e18adb4 | 1593 | |
47118af0 | 1594 | #ifdef CONFIG_CMA |
9cf510a5 | 1595 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1596 | void __init init_cma_reserved_pageblock(struct page *page) |
1597 | { | |
1598 | unsigned i = pageblock_nr_pages; | |
1599 | struct page *p = page; | |
1600 | ||
1601 | do { | |
1602 | __ClearPageReserved(p); | |
1603 | set_page_count(p, 0); | |
1604 | } while (++p, --i); | |
1605 | ||
47118af0 | 1606 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1607 | |
1608 | if (pageblock_order >= MAX_ORDER) { | |
1609 | i = pageblock_nr_pages; | |
1610 | p = page; | |
1611 | do { | |
1612 | set_page_refcounted(p); | |
1613 | __free_pages(p, MAX_ORDER - 1); | |
1614 | p += MAX_ORDER_NR_PAGES; | |
1615 | } while (i -= MAX_ORDER_NR_PAGES); | |
1616 | } else { | |
1617 | set_page_refcounted(page); | |
1618 | __free_pages(page, pageblock_order); | |
1619 | } | |
1620 | ||
3dcc0571 | 1621 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1622 | } |
1623 | #endif | |
1da177e4 LT |
1624 | |
1625 | /* | |
1626 | * The order of subdivision here is critical for the IO subsystem. | |
1627 | * Please do not alter this order without good reasons and regression | |
1628 | * testing. Specifically, as large blocks of memory are subdivided, | |
1629 | * the order in which smaller blocks are delivered depends on the order | |
1630 | * they're subdivided in this function. This is the primary factor | |
1631 | * influencing the order in which pages are delivered to the IO | |
1632 | * subsystem according to empirical testing, and this is also justified | |
1633 | * by considering the behavior of a buddy system containing a single | |
1634 | * large block of memory acted on by a series of small allocations. | |
1635 | * This behavior is a critical factor in sglist merging's success. | |
1636 | * | |
6d49e352 | 1637 | * -- nyc |
1da177e4 | 1638 | */ |
085cc7d5 | 1639 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1640 | int low, int high, struct free_area *area, |
1641 | int migratetype) | |
1da177e4 LT |
1642 | { |
1643 | unsigned long size = 1 << high; | |
1644 | ||
1645 | while (high > low) { | |
1646 | area--; | |
1647 | high--; | |
1648 | size >>= 1; | |
309381fe | 1649 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1650 | |
acbc15a4 JK |
1651 | /* |
1652 | * Mark as guard pages (or page), that will allow to | |
1653 | * merge back to allocator when buddy will be freed. | |
1654 | * Corresponding page table entries will not be touched, | |
1655 | * pages will stay not present in virtual address space | |
1656 | */ | |
1657 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 1658 | continue; |
acbc15a4 | 1659 | |
b2a0ac88 | 1660 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1661 | area->nr_free++; |
1662 | set_page_order(&page[size], high); | |
1663 | } | |
1da177e4 LT |
1664 | } |
1665 | ||
4e611801 | 1666 | static void check_new_page_bad(struct page *page) |
1da177e4 | 1667 | { |
4e611801 VB |
1668 | const char *bad_reason = NULL; |
1669 | unsigned long bad_flags = 0; | |
7bfec6f4 | 1670 | |
53f9263b | 1671 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1672 | bad_reason = "nonzero mapcount"; |
1673 | if (unlikely(page->mapping != NULL)) | |
1674 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1675 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1676 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1677 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1678 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1679 | bad_flags = __PG_HWPOISON; | |
e570f56c NH |
1680 | /* Don't complain about hwpoisoned pages */ |
1681 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1682 | return; | |
f4c18e6f | 1683 | } |
f0b791a3 DH |
1684 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1685 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1686 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1687 | } | |
9edad6ea JW |
1688 | #ifdef CONFIG_MEMCG |
1689 | if (unlikely(page->mem_cgroup)) | |
1690 | bad_reason = "page still charged to cgroup"; | |
1691 | #endif | |
4e611801 VB |
1692 | bad_page(page, bad_reason, bad_flags); |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * This page is about to be returned from the page allocator | |
1697 | */ | |
1698 | static inline int check_new_page(struct page *page) | |
1699 | { | |
1700 | if (likely(page_expected_state(page, | |
1701 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1702 | return 0; | |
1703 | ||
1704 | check_new_page_bad(page); | |
1705 | return 1; | |
2a7684a2 WF |
1706 | } |
1707 | ||
bd33ef36 | 1708 | static inline bool free_pages_prezeroed(void) |
1414c7f4 LA |
1709 | { |
1710 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
bd33ef36 | 1711 | page_poisoning_enabled(); |
1414c7f4 LA |
1712 | } |
1713 | ||
479f854a MG |
1714 | #ifdef CONFIG_DEBUG_VM |
1715 | static bool check_pcp_refill(struct page *page) | |
1716 | { | |
1717 | return false; | |
1718 | } | |
1719 | ||
1720 | static bool check_new_pcp(struct page *page) | |
1721 | { | |
1722 | return check_new_page(page); | |
1723 | } | |
1724 | #else | |
1725 | static bool check_pcp_refill(struct page *page) | |
1726 | { | |
1727 | return check_new_page(page); | |
1728 | } | |
1729 | static bool check_new_pcp(struct page *page) | |
1730 | { | |
1731 | return false; | |
1732 | } | |
1733 | #endif /* CONFIG_DEBUG_VM */ | |
1734 | ||
1735 | static bool check_new_pages(struct page *page, unsigned int order) | |
1736 | { | |
1737 | int i; | |
1738 | for (i = 0; i < (1 << order); i++) { | |
1739 | struct page *p = page + i; | |
1740 | ||
1741 | if (unlikely(check_new_page(p))) | |
1742 | return true; | |
1743 | } | |
1744 | ||
1745 | return false; | |
1746 | } | |
1747 | ||
46f24fd8 JK |
1748 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1749 | gfp_t gfp_flags) | |
1750 | { | |
1751 | set_page_private(page, 0); | |
1752 | set_page_refcounted(page); | |
1753 | ||
1754 | arch_alloc_page(page, order); | |
1755 | kernel_map_pages(page, 1 << order, 1); | |
1756 | kernel_poison_pages(page, 1 << order, 1); | |
1757 | kasan_alloc_pages(page, order); | |
1758 | set_page_owner(page, order, gfp_flags); | |
1759 | } | |
1760 | ||
479f854a | 1761 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1762 | unsigned int alloc_flags) |
2a7684a2 WF |
1763 | { |
1764 | int i; | |
689bcebf | 1765 | |
46f24fd8 | 1766 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1767 | |
bd33ef36 | 1768 | if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1769 | for (i = 0; i < (1 << order); i++) |
1770 | clear_highpage(page + i); | |
17cf4406 NP |
1771 | |
1772 | if (order && (gfp_flags & __GFP_COMP)) | |
1773 | prep_compound_page(page, order); | |
1774 | ||
75379191 | 1775 | /* |
2f064f34 | 1776 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1777 | * allocate the page. The expectation is that the caller is taking |
1778 | * steps that will free more memory. The caller should avoid the page | |
1779 | * being used for !PFMEMALLOC purposes. | |
1780 | */ | |
2f064f34 MH |
1781 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1782 | set_page_pfmemalloc(page); | |
1783 | else | |
1784 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1785 | } |
1786 | ||
56fd56b8 MG |
1787 | /* |
1788 | * Go through the free lists for the given migratetype and remove | |
1789 | * the smallest available page from the freelists | |
1790 | */ | |
728ec980 MG |
1791 | static inline |
1792 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1793 | int migratetype) |
1794 | { | |
1795 | unsigned int current_order; | |
b8af2941 | 1796 | struct free_area *area; |
56fd56b8 MG |
1797 | struct page *page; |
1798 | ||
1799 | /* Find a page of the appropriate size in the preferred list */ | |
1800 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1801 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1802 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1803 | struct page, lru); |
a16601c5 GT |
1804 | if (!page) |
1805 | continue; | |
56fd56b8 MG |
1806 | list_del(&page->lru); |
1807 | rmv_page_order(page); | |
1808 | area->nr_free--; | |
56fd56b8 | 1809 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1810 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1811 | return page; |
1812 | } | |
1813 | ||
1814 | return NULL; | |
1815 | } | |
1816 | ||
1817 | ||
b2a0ac88 MG |
1818 | /* |
1819 | * This array describes the order lists are fallen back to when | |
1820 | * the free lists for the desirable migrate type are depleted | |
1821 | */ | |
47118af0 | 1822 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1823 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1824 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1825 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1826 | #ifdef CONFIG_CMA |
974a786e | 1827 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1828 | #endif |
194159fb | 1829 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1830 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1831 | #endif |
b2a0ac88 MG |
1832 | }; |
1833 | ||
dc67647b JK |
1834 | #ifdef CONFIG_CMA |
1835 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1836 | unsigned int order) | |
1837 | { | |
1838 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1839 | } | |
1840 | #else | |
1841 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1842 | unsigned int order) { return NULL; } | |
1843 | #endif | |
1844 | ||
c361be55 MG |
1845 | /* |
1846 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1847 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1848 | * boundary. If alignment is required, use move_freepages_block() |
1849 | */ | |
02aa0cdd | 1850 | static int move_freepages(struct zone *zone, |
b69a7288 | 1851 | struct page *start_page, struct page *end_page, |
02aa0cdd | 1852 | int migratetype, int *num_movable) |
c361be55 MG |
1853 | { |
1854 | struct page *page; | |
d00181b9 | 1855 | unsigned int order; |
d100313f | 1856 | int pages_moved = 0; |
c361be55 MG |
1857 | |
1858 | #ifndef CONFIG_HOLES_IN_ZONE | |
1859 | /* | |
1860 | * page_zone is not safe to call in this context when | |
1861 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1862 | * anyway as we check zone boundaries in move_freepages_block(). | |
1863 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1864 | * grouping pages by mobility |
c361be55 | 1865 | */ |
97ee4ba7 | 1866 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1867 | #endif |
1868 | ||
02aa0cdd VB |
1869 | if (num_movable) |
1870 | *num_movable = 0; | |
1871 | ||
c361be55 MG |
1872 | for (page = start_page; page <= end_page;) { |
1873 | if (!pfn_valid_within(page_to_pfn(page))) { | |
1874 | page++; | |
1875 | continue; | |
1876 | } | |
1877 | ||
f073bdc5 AB |
1878 | /* Make sure we are not inadvertently changing nodes */ |
1879 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
1880 | ||
c361be55 | 1881 | if (!PageBuddy(page)) { |
02aa0cdd VB |
1882 | /* |
1883 | * We assume that pages that could be isolated for | |
1884 | * migration are movable. But we don't actually try | |
1885 | * isolating, as that would be expensive. | |
1886 | */ | |
1887 | if (num_movable && | |
1888 | (PageLRU(page) || __PageMovable(page))) | |
1889 | (*num_movable)++; | |
1890 | ||
c361be55 MG |
1891 | page++; |
1892 | continue; | |
1893 | } | |
1894 | ||
1895 | order = page_order(page); | |
84be48d8 KS |
1896 | list_move(&page->lru, |
1897 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 1898 | page += 1 << order; |
d100313f | 1899 | pages_moved += 1 << order; |
c361be55 MG |
1900 | } |
1901 | ||
d100313f | 1902 | return pages_moved; |
c361be55 MG |
1903 | } |
1904 | ||
ee6f509c | 1905 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 1906 | int migratetype, int *num_movable) |
c361be55 MG |
1907 | { |
1908 | unsigned long start_pfn, end_pfn; | |
1909 | struct page *start_page, *end_page; | |
1910 | ||
1911 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1912 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1913 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1914 | end_page = start_page + pageblock_nr_pages - 1; |
1915 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1916 | |
1917 | /* Do not cross zone boundaries */ | |
108bcc96 | 1918 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1919 | start_page = page; |
108bcc96 | 1920 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1921 | return 0; |
1922 | ||
02aa0cdd VB |
1923 | return move_freepages(zone, start_page, end_page, migratetype, |
1924 | num_movable); | |
c361be55 MG |
1925 | } |
1926 | ||
2f66a68f MG |
1927 | static void change_pageblock_range(struct page *pageblock_page, |
1928 | int start_order, int migratetype) | |
1929 | { | |
1930 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1931 | ||
1932 | while (nr_pageblocks--) { | |
1933 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1934 | pageblock_page += pageblock_nr_pages; | |
1935 | } | |
1936 | } | |
1937 | ||
fef903ef | 1938 | /* |
9c0415eb VB |
1939 | * When we are falling back to another migratetype during allocation, try to |
1940 | * steal extra free pages from the same pageblocks to satisfy further | |
1941 | * allocations, instead of polluting multiple pageblocks. | |
1942 | * | |
1943 | * If we are stealing a relatively large buddy page, it is likely there will | |
1944 | * be more free pages in the pageblock, so try to steal them all. For | |
1945 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1946 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1947 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1948 | * pageblocks. | |
fef903ef | 1949 | */ |
4eb7dce6 JK |
1950 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1951 | { | |
1952 | /* | |
1953 | * Leaving this order check is intended, although there is | |
1954 | * relaxed order check in next check. The reason is that | |
1955 | * we can actually steal whole pageblock if this condition met, | |
1956 | * but, below check doesn't guarantee it and that is just heuristic | |
1957 | * so could be changed anytime. | |
1958 | */ | |
1959 | if (order >= pageblock_order) | |
1960 | return true; | |
1961 | ||
1962 | if (order >= pageblock_order / 2 || | |
1963 | start_mt == MIGRATE_RECLAIMABLE || | |
1964 | start_mt == MIGRATE_UNMOVABLE || | |
1965 | page_group_by_mobility_disabled) | |
1966 | return true; | |
1967 | ||
1968 | return false; | |
1969 | } | |
1970 | ||
1971 | /* | |
1972 | * This function implements actual steal behaviour. If order is large enough, | |
1973 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
1974 | * pageblock to our migratetype and determine how many already-allocated pages |
1975 | * are there in the pageblock with a compatible migratetype. If at least half | |
1976 | * of pages are free or compatible, we can change migratetype of the pageblock | |
1977 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
1978 | */ |
1979 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
3bc48f96 | 1980 | int start_type, bool whole_block) |
fef903ef | 1981 | { |
d00181b9 | 1982 | unsigned int current_order = page_order(page); |
3bc48f96 | 1983 | struct free_area *area; |
02aa0cdd VB |
1984 | int free_pages, movable_pages, alike_pages; |
1985 | int old_block_type; | |
1986 | ||
1987 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 1988 | |
3bc48f96 VB |
1989 | /* |
1990 | * This can happen due to races and we want to prevent broken | |
1991 | * highatomic accounting. | |
1992 | */ | |
02aa0cdd | 1993 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
1994 | goto single_page; |
1995 | ||
fef903ef SB |
1996 | /* Take ownership for orders >= pageblock_order */ |
1997 | if (current_order >= pageblock_order) { | |
1998 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 1999 | goto single_page; |
fef903ef SB |
2000 | } |
2001 | ||
3bc48f96 VB |
2002 | /* We are not allowed to try stealing from the whole block */ |
2003 | if (!whole_block) | |
2004 | goto single_page; | |
2005 | ||
02aa0cdd VB |
2006 | free_pages = move_freepages_block(zone, page, start_type, |
2007 | &movable_pages); | |
2008 | /* | |
2009 | * Determine how many pages are compatible with our allocation. | |
2010 | * For movable allocation, it's the number of movable pages which | |
2011 | * we just obtained. For other types it's a bit more tricky. | |
2012 | */ | |
2013 | if (start_type == MIGRATE_MOVABLE) { | |
2014 | alike_pages = movable_pages; | |
2015 | } else { | |
2016 | /* | |
2017 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2018 | * to MOVABLE pageblock, consider all non-movable pages as | |
2019 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2020 | * vice versa, be conservative since we can't distinguish the | |
2021 | * exact migratetype of non-movable pages. | |
2022 | */ | |
2023 | if (old_block_type == MIGRATE_MOVABLE) | |
2024 | alike_pages = pageblock_nr_pages | |
2025 | - (free_pages + movable_pages); | |
2026 | else | |
2027 | alike_pages = 0; | |
2028 | } | |
2029 | ||
3bc48f96 | 2030 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2031 | if (!free_pages) |
3bc48f96 | 2032 | goto single_page; |
fef903ef | 2033 | |
02aa0cdd VB |
2034 | /* |
2035 | * If a sufficient number of pages in the block are either free or of | |
2036 | * comparable migratability as our allocation, claim the whole block. | |
2037 | */ | |
2038 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2039 | page_group_by_mobility_disabled) |
2040 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2041 | |
2042 | return; | |
2043 | ||
2044 | single_page: | |
2045 | area = &zone->free_area[current_order]; | |
2046 | list_move(&page->lru, &area->free_list[start_type]); | |
4eb7dce6 JK |
2047 | } |
2048 | ||
2149cdae JK |
2049 | /* |
2050 | * Check whether there is a suitable fallback freepage with requested order. | |
2051 | * If only_stealable is true, this function returns fallback_mt only if | |
2052 | * we can steal other freepages all together. This would help to reduce | |
2053 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2054 | */ | |
2055 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2056 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2057 | { |
2058 | int i; | |
2059 | int fallback_mt; | |
2060 | ||
2061 | if (area->nr_free == 0) | |
2062 | return -1; | |
2063 | ||
2064 | *can_steal = false; | |
2065 | for (i = 0;; i++) { | |
2066 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2067 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2068 | break; |
2069 | ||
2070 | if (list_empty(&area->free_list[fallback_mt])) | |
2071 | continue; | |
fef903ef | 2072 | |
4eb7dce6 JK |
2073 | if (can_steal_fallback(order, migratetype)) |
2074 | *can_steal = true; | |
2075 | ||
2149cdae JK |
2076 | if (!only_stealable) |
2077 | return fallback_mt; | |
2078 | ||
2079 | if (*can_steal) | |
2080 | return fallback_mt; | |
fef903ef | 2081 | } |
4eb7dce6 JK |
2082 | |
2083 | return -1; | |
fef903ef SB |
2084 | } |
2085 | ||
0aaa29a5 MG |
2086 | /* |
2087 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2088 | * there are no empty page blocks that contain a page with a suitable order | |
2089 | */ | |
2090 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2091 | unsigned int alloc_order) | |
2092 | { | |
2093 | int mt; | |
2094 | unsigned long max_managed, flags; | |
2095 | ||
2096 | /* | |
2097 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2098 | * Check is race-prone but harmless. | |
2099 | */ | |
2100 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
2101 | if (zone->nr_reserved_highatomic >= max_managed) | |
2102 | return; | |
2103 | ||
2104 | spin_lock_irqsave(&zone->lock, flags); | |
2105 | ||
2106 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2107 | if (zone->nr_reserved_highatomic >= max_managed) | |
2108 | goto out_unlock; | |
2109 | ||
2110 | /* Yoink! */ | |
2111 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2112 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2113 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2114 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2115 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2116 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2117 | } |
2118 | ||
2119 | out_unlock: | |
2120 | spin_unlock_irqrestore(&zone->lock, flags); | |
2121 | } | |
2122 | ||
2123 | /* | |
2124 | * Used when an allocation is about to fail under memory pressure. This | |
2125 | * potentially hurts the reliability of high-order allocations when under | |
2126 | * intense memory pressure but failed atomic allocations should be easier | |
2127 | * to recover from than an OOM. | |
29fac03b MK |
2128 | * |
2129 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2130 | * pageblock is exhausted. | |
0aaa29a5 | 2131 | */ |
29fac03b MK |
2132 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2133 | bool force) | |
0aaa29a5 MG |
2134 | { |
2135 | struct zonelist *zonelist = ac->zonelist; | |
2136 | unsigned long flags; | |
2137 | struct zoneref *z; | |
2138 | struct zone *zone; | |
2139 | struct page *page; | |
2140 | int order; | |
04c8716f | 2141 | bool ret; |
0aaa29a5 MG |
2142 | |
2143 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2144 | ac->nodemask) { | |
29fac03b MK |
2145 | /* |
2146 | * Preserve at least one pageblock unless memory pressure | |
2147 | * is really high. | |
2148 | */ | |
2149 | if (!force && zone->nr_reserved_highatomic <= | |
2150 | pageblock_nr_pages) | |
0aaa29a5 MG |
2151 | continue; |
2152 | ||
2153 | spin_lock_irqsave(&zone->lock, flags); | |
2154 | for (order = 0; order < MAX_ORDER; order++) { | |
2155 | struct free_area *area = &(zone->free_area[order]); | |
2156 | ||
a16601c5 GT |
2157 | page = list_first_entry_or_null( |
2158 | &area->free_list[MIGRATE_HIGHATOMIC], | |
2159 | struct page, lru); | |
2160 | if (!page) | |
0aaa29a5 MG |
2161 | continue; |
2162 | ||
0aaa29a5 | 2163 | /* |
4855e4a7 MK |
2164 | * In page freeing path, migratetype change is racy so |
2165 | * we can counter several free pages in a pageblock | |
2166 | * in this loop althoug we changed the pageblock type | |
2167 | * from highatomic to ac->migratetype. So we should | |
2168 | * adjust the count once. | |
0aaa29a5 | 2169 | */ |
a6ffdc07 | 2170 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2171 | /* |
2172 | * It should never happen but changes to | |
2173 | * locking could inadvertently allow a per-cpu | |
2174 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2175 | * while unreserving so be safe and watch for | |
2176 | * underflows. | |
2177 | */ | |
2178 | zone->nr_reserved_highatomic -= min( | |
2179 | pageblock_nr_pages, | |
2180 | zone->nr_reserved_highatomic); | |
2181 | } | |
0aaa29a5 MG |
2182 | |
2183 | /* | |
2184 | * Convert to ac->migratetype and avoid the normal | |
2185 | * pageblock stealing heuristics. Minimally, the caller | |
2186 | * is doing the work and needs the pages. More | |
2187 | * importantly, if the block was always converted to | |
2188 | * MIGRATE_UNMOVABLE or another type then the number | |
2189 | * of pageblocks that cannot be completely freed | |
2190 | * may increase. | |
2191 | */ | |
2192 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2193 | ret = move_freepages_block(zone, page, ac->migratetype, |
2194 | NULL); | |
29fac03b MK |
2195 | if (ret) { |
2196 | spin_unlock_irqrestore(&zone->lock, flags); | |
2197 | return ret; | |
2198 | } | |
0aaa29a5 MG |
2199 | } |
2200 | spin_unlock_irqrestore(&zone->lock, flags); | |
2201 | } | |
04c8716f MK |
2202 | |
2203 | return false; | |
0aaa29a5 MG |
2204 | } |
2205 | ||
3bc48f96 VB |
2206 | /* |
2207 | * Try finding a free buddy page on the fallback list and put it on the free | |
2208 | * list of requested migratetype, possibly along with other pages from the same | |
2209 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2210 | * fallback was found so that __rmqueue_smallest() can grab it. | |
2211 | */ | |
2212 | static inline bool | |
7aeb09f9 | 2213 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 2214 | { |
b8af2941 | 2215 | struct free_area *area; |
7aeb09f9 | 2216 | unsigned int current_order; |
b2a0ac88 | 2217 | struct page *page; |
4eb7dce6 JK |
2218 | int fallback_mt; |
2219 | bool can_steal; | |
b2a0ac88 MG |
2220 | |
2221 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
2222 | for (current_order = MAX_ORDER-1; |
2223 | current_order >= order && current_order <= MAX_ORDER-1; | |
2224 | --current_order) { | |
4eb7dce6 JK |
2225 | area = &(zone->free_area[current_order]); |
2226 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2227 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2228 | if (fallback_mt == -1) |
2229 | continue; | |
b2a0ac88 | 2230 | |
a16601c5 | 2231 | page = list_first_entry(&area->free_list[fallback_mt], |
4eb7dce6 | 2232 | struct page, lru); |
b2a0ac88 | 2233 | |
3bc48f96 VB |
2234 | steal_suitable_fallback(zone, page, start_migratetype, |
2235 | can_steal); | |
e0fff1bd | 2236 | |
4eb7dce6 JK |
2237 | trace_mm_page_alloc_extfrag(page, order, current_order, |
2238 | start_migratetype, fallback_mt); | |
e0fff1bd | 2239 | |
3bc48f96 | 2240 | return true; |
b2a0ac88 MG |
2241 | } |
2242 | ||
3bc48f96 | 2243 | return false; |
b2a0ac88 MG |
2244 | } |
2245 | ||
56fd56b8 | 2246 | /* |
1da177e4 LT |
2247 | * Do the hard work of removing an element from the buddy allocator. |
2248 | * Call me with the zone->lock already held. | |
2249 | */ | |
b2a0ac88 | 2250 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
6ac0206b | 2251 | int migratetype) |
1da177e4 | 2252 | { |
1da177e4 LT |
2253 | struct page *page; |
2254 | ||
3bc48f96 | 2255 | retry: |
56fd56b8 | 2256 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2257 | if (unlikely(!page)) { |
dc67647b JK |
2258 | if (migratetype == MIGRATE_MOVABLE) |
2259 | page = __rmqueue_cma_fallback(zone, order); | |
2260 | ||
3bc48f96 VB |
2261 | if (!page && __rmqueue_fallback(zone, order, migratetype)) |
2262 | goto retry; | |
728ec980 MG |
2263 | } |
2264 | ||
0d3d062a | 2265 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2266 | return page; |
1da177e4 LT |
2267 | } |
2268 | ||
5f63b720 | 2269 | /* |
1da177e4 LT |
2270 | * Obtain a specified number of elements from the buddy allocator, all under |
2271 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2272 | * Returns the number of new pages which were placed at *list. | |
2273 | */ | |
5f63b720 | 2274 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2275 | unsigned long count, struct list_head *list, |
b745bc85 | 2276 | int migratetype, bool cold) |
1da177e4 | 2277 | { |
a6de734b | 2278 | int i, alloced = 0; |
5f63b720 | 2279 | |
d34b0733 | 2280 | spin_lock(&zone->lock); |
1da177e4 | 2281 | for (i = 0; i < count; ++i) { |
6ac0206b | 2282 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 2283 | if (unlikely(page == NULL)) |
1da177e4 | 2284 | break; |
81eabcbe | 2285 | |
479f854a MG |
2286 | if (unlikely(check_pcp_refill(page))) |
2287 | continue; | |
2288 | ||
81eabcbe MG |
2289 | /* |
2290 | * Split buddy pages returned by expand() are received here | |
2291 | * in physical page order. The page is added to the callers and | |
2292 | * list and the list head then moves forward. From the callers | |
2293 | * perspective, the linked list is ordered by page number in | |
2294 | * some conditions. This is useful for IO devices that can | |
2295 | * merge IO requests if the physical pages are ordered | |
2296 | * properly. | |
2297 | */ | |
b745bc85 | 2298 | if (likely(!cold)) |
e084b2d9 MG |
2299 | list_add(&page->lru, list); |
2300 | else | |
2301 | list_add_tail(&page->lru, list); | |
81eabcbe | 2302 | list = &page->lru; |
a6de734b | 2303 | alloced++; |
bb14c2c7 | 2304 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2305 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2306 | -(1 << order)); | |
1da177e4 | 2307 | } |
a6de734b MG |
2308 | |
2309 | /* | |
2310 | * i pages were removed from the buddy list even if some leak due | |
2311 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2312 | * on i. Do not confuse with 'alloced' which is the number of | |
2313 | * pages added to the pcp list. | |
2314 | */ | |
f2260e6b | 2315 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2316 | spin_unlock(&zone->lock); |
a6de734b | 2317 | return alloced; |
1da177e4 LT |
2318 | } |
2319 | ||
4ae7c039 | 2320 | #ifdef CONFIG_NUMA |
8fce4d8e | 2321 | /* |
4037d452 CL |
2322 | * Called from the vmstat counter updater to drain pagesets of this |
2323 | * currently executing processor on remote nodes after they have | |
2324 | * expired. | |
2325 | * | |
879336c3 CL |
2326 | * Note that this function must be called with the thread pinned to |
2327 | * a single processor. | |
8fce4d8e | 2328 | */ |
4037d452 | 2329 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2330 | { |
4ae7c039 | 2331 | unsigned long flags; |
7be12fc9 | 2332 | int to_drain, batch; |
4ae7c039 | 2333 | |
4037d452 | 2334 | local_irq_save(flags); |
4db0c3c2 | 2335 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2336 | to_drain = min(pcp->count, batch); |
2a13515c KM |
2337 | if (to_drain > 0) { |
2338 | free_pcppages_bulk(zone, to_drain, pcp); | |
2339 | pcp->count -= to_drain; | |
2340 | } | |
4037d452 | 2341 | local_irq_restore(flags); |
4ae7c039 CL |
2342 | } |
2343 | #endif | |
2344 | ||
9f8f2172 | 2345 | /* |
93481ff0 | 2346 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2347 | * |
2348 | * The processor must either be the current processor and the | |
2349 | * thread pinned to the current processor or a processor that | |
2350 | * is not online. | |
2351 | */ | |
93481ff0 | 2352 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2353 | { |
c54ad30c | 2354 | unsigned long flags; |
93481ff0 VB |
2355 | struct per_cpu_pageset *pset; |
2356 | struct per_cpu_pages *pcp; | |
1da177e4 | 2357 | |
93481ff0 VB |
2358 | local_irq_save(flags); |
2359 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2360 | |
93481ff0 VB |
2361 | pcp = &pset->pcp; |
2362 | if (pcp->count) { | |
2363 | free_pcppages_bulk(zone, pcp->count, pcp); | |
2364 | pcp->count = 0; | |
2365 | } | |
2366 | local_irq_restore(flags); | |
2367 | } | |
3dfa5721 | 2368 | |
93481ff0 VB |
2369 | /* |
2370 | * Drain pcplists of all zones on the indicated processor. | |
2371 | * | |
2372 | * The processor must either be the current processor and the | |
2373 | * thread pinned to the current processor or a processor that | |
2374 | * is not online. | |
2375 | */ | |
2376 | static void drain_pages(unsigned int cpu) | |
2377 | { | |
2378 | struct zone *zone; | |
2379 | ||
2380 | for_each_populated_zone(zone) { | |
2381 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2382 | } |
2383 | } | |
1da177e4 | 2384 | |
9f8f2172 CL |
2385 | /* |
2386 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2387 | * |
2388 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2389 | * the single zone's pages. | |
9f8f2172 | 2390 | */ |
93481ff0 | 2391 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2392 | { |
93481ff0 VB |
2393 | int cpu = smp_processor_id(); |
2394 | ||
2395 | if (zone) | |
2396 | drain_pages_zone(cpu, zone); | |
2397 | else | |
2398 | drain_pages(cpu); | |
9f8f2172 CL |
2399 | } |
2400 | ||
0ccce3b9 MG |
2401 | static void drain_local_pages_wq(struct work_struct *work) |
2402 | { | |
a459eeb7 MH |
2403 | /* |
2404 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2405 | * we can race with cpu offline when the WQ can move this from | |
2406 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2407 | * cpu which is allright but we also have to make sure to not move to | |
2408 | * a different one. | |
2409 | */ | |
2410 | preempt_disable(); | |
0ccce3b9 | 2411 | drain_local_pages(NULL); |
a459eeb7 | 2412 | preempt_enable(); |
0ccce3b9 MG |
2413 | } |
2414 | ||
9f8f2172 | 2415 | /* |
74046494 GBY |
2416 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2417 | * | |
93481ff0 VB |
2418 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2419 | * | |
0ccce3b9 | 2420 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2421 | */ |
93481ff0 | 2422 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2423 | { |
74046494 | 2424 | int cpu; |
74046494 GBY |
2425 | |
2426 | /* | |
2427 | * Allocate in the BSS so we wont require allocation in | |
2428 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2429 | */ | |
2430 | static cpumask_t cpus_with_pcps; | |
2431 | ||
ce612879 MH |
2432 | /* |
2433 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2434 | * initialized. | |
2435 | */ | |
2436 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2437 | return; | |
2438 | ||
0ccce3b9 MG |
2439 | /* Workqueues cannot recurse */ |
2440 | if (current->flags & PF_WQ_WORKER) | |
2441 | return; | |
2442 | ||
bd233f53 MG |
2443 | /* |
2444 | * Do not drain if one is already in progress unless it's specific to | |
2445 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2446 | * the drain to be complete when the call returns. | |
2447 | */ | |
2448 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2449 | if (!zone) | |
2450 | return; | |
2451 | mutex_lock(&pcpu_drain_mutex); | |
2452 | } | |
0ccce3b9 | 2453 | |
74046494 GBY |
2454 | /* |
2455 | * We don't care about racing with CPU hotplug event | |
2456 | * as offline notification will cause the notified | |
2457 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2458 | * disables preemption as part of its processing | |
2459 | */ | |
2460 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2461 | struct per_cpu_pageset *pcp; |
2462 | struct zone *z; | |
74046494 | 2463 | bool has_pcps = false; |
93481ff0 VB |
2464 | |
2465 | if (zone) { | |
74046494 | 2466 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2467 | if (pcp->pcp.count) |
74046494 | 2468 | has_pcps = true; |
93481ff0 VB |
2469 | } else { |
2470 | for_each_populated_zone(z) { | |
2471 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2472 | if (pcp->pcp.count) { | |
2473 | has_pcps = true; | |
2474 | break; | |
2475 | } | |
74046494 GBY |
2476 | } |
2477 | } | |
93481ff0 | 2478 | |
74046494 GBY |
2479 | if (has_pcps) |
2480 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2481 | else | |
2482 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2483 | } | |
0ccce3b9 | 2484 | |
bd233f53 MG |
2485 | for_each_cpu(cpu, &cpus_with_pcps) { |
2486 | struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu); | |
2487 | INIT_WORK(work, drain_local_pages_wq); | |
ce612879 | 2488 | queue_work_on(cpu, mm_percpu_wq, work); |
0ccce3b9 | 2489 | } |
bd233f53 MG |
2490 | for_each_cpu(cpu, &cpus_with_pcps) |
2491 | flush_work(per_cpu_ptr(&pcpu_drain, cpu)); | |
2492 | ||
2493 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2494 | } |
2495 | ||
296699de | 2496 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
2497 | |
2498 | void mark_free_pages(struct zone *zone) | |
2499 | { | |
f623f0db RW |
2500 | unsigned long pfn, max_zone_pfn; |
2501 | unsigned long flags; | |
7aeb09f9 | 2502 | unsigned int order, t; |
86760a2c | 2503 | struct page *page; |
1da177e4 | 2504 | |
8080fc03 | 2505 | if (zone_is_empty(zone)) |
1da177e4 LT |
2506 | return; |
2507 | ||
2508 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2509 | |
108bcc96 | 2510 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2511 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2512 | if (pfn_valid(pfn)) { | |
86760a2c | 2513 | page = pfn_to_page(pfn); |
ba6b0979 JK |
2514 | |
2515 | if (page_zone(page) != zone) | |
2516 | continue; | |
2517 | ||
7be98234 RW |
2518 | if (!swsusp_page_is_forbidden(page)) |
2519 | swsusp_unset_page_free(page); | |
f623f0db | 2520 | } |
1da177e4 | 2521 | |
b2a0ac88 | 2522 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2523 | list_for_each_entry(page, |
2524 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2525 | unsigned long i; |
1da177e4 | 2526 | |
86760a2c | 2527 | pfn = page_to_pfn(page); |
f623f0db | 2528 | for (i = 0; i < (1UL << order); i++) |
7be98234 | 2529 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 2530 | } |
b2a0ac88 | 2531 | } |
1da177e4 LT |
2532 | spin_unlock_irqrestore(&zone->lock, flags); |
2533 | } | |
e2c55dc8 | 2534 | #endif /* CONFIG_PM */ |
1da177e4 | 2535 | |
1da177e4 LT |
2536 | /* |
2537 | * Free a 0-order page | |
b745bc85 | 2538 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 2539 | */ |
b745bc85 | 2540 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
2541 | { |
2542 | struct zone *zone = page_zone(page); | |
2543 | struct per_cpu_pages *pcp; | |
d34b0733 | 2544 | unsigned long flags; |
dc4b0caf | 2545 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 2546 | int migratetype; |
1da177e4 | 2547 | |
4db7548c | 2548 | if (!free_pcp_prepare(page)) |
689bcebf HD |
2549 | return; |
2550 | ||
dc4b0caf | 2551 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2552 | set_pcppage_migratetype(page, migratetype); |
d34b0733 MG |
2553 | local_irq_save(flags); |
2554 | __count_vm_event(PGFREE); | |
da456f14 | 2555 | |
5f8dcc21 MG |
2556 | /* |
2557 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2558 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 2559 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
2560 | * areas back if necessary. Otherwise, we may have to free |
2561 | * excessively into the page allocator | |
2562 | */ | |
2563 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2564 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2565 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
2566 | goto out; |
2567 | } | |
2568 | migratetype = MIGRATE_MOVABLE; | |
2569 | } | |
2570 | ||
99dcc3e5 | 2571 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 2572 | if (!cold) |
5f8dcc21 | 2573 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
2574 | else |
2575 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 2576 | pcp->count++; |
48db57f8 | 2577 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2578 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
2579 | free_pcppages_bulk(zone, batch, pcp); |
2580 | pcp->count -= batch; | |
48db57f8 | 2581 | } |
5f8dcc21 MG |
2582 | |
2583 | out: | |
d34b0733 | 2584 | local_irq_restore(flags); |
1da177e4 LT |
2585 | } |
2586 | ||
cc59850e KK |
2587 | /* |
2588 | * Free a list of 0-order pages | |
2589 | */ | |
b745bc85 | 2590 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
2591 | { |
2592 | struct page *page, *next; | |
2593 | ||
2594 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 2595 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
2596 | free_hot_cold_page(page, cold); |
2597 | } | |
2598 | } | |
2599 | ||
8dfcc9ba NP |
2600 | /* |
2601 | * split_page takes a non-compound higher-order page, and splits it into | |
2602 | * n (1<<order) sub-pages: page[0..n] | |
2603 | * Each sub-page must be freed individually. | |
2604 | * | |
2605 | * Note: this is probably too low level an operation for use in drivers. | |
2606 | * Please consult with lkml before using this in your driver. | |
2607 | */ | |
2608 | void split_page(struct page *page, unsigned int order) | |
2609 | { | |
2610 | int i; | |
2611 | ||
309381fe SL |
2612 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2613 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
2614 | |
2615 | #ifdef CONFIG_KMEMCHECK | |
2616 | /* | |
2617 | * Split shadow pages too, because free(page[0]) would | |
2618 | * otherwise free the whole shadow. | |
2619 | */ | |
2620 | if (kmemcheck_page_is_tracked(page)) | |
2621 | split_page(virt_to_page(page[0].shadow), order); | |
2622 | #endif | |
2623 | ||
a9627bc5 | 2624 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2625 | set_page_refcounted(page + i); |
a9627bc5 | 2626 | split_page_owner(page, order); |
8dfcc9ba | 2627 | } |
5853ff23 | 2628 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2629 | |
3c605096 | 2630 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2631 | { |
748446bb MG |
2632 | unsigned long watermark; |
2633 | struct zone *zone; | |
2139cbe6 | 2634 | int mt; |
748446bb MG |
2635 | |
2636 | BUG_ON(!PageBuddy(page)); | |
2637 | ||
2638 | zone = page_zone(page); | |
2e30abd1 | 2639 | mt = get_pageblock_migratetype(page); |
748446bb | 2640 | |
194159fb | 2641 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
2642 | /* |
2643 | * Obey watermarks as if the page was being allocated. We can | |
2644 | * emulate a high-order watermark check with a raised order-0 | |
2645 | * watermark, because we already know our high-order page | |
2646 | * exists. | |
2647 | */ | |
2648 | watermark = min_wmark_pages(zone) + (1UL << order); | |
984fdba6 | 2649 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
2650 | return 0; |
2651 | ||
8fb74b9f | 2652 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2653 | } |
748446bb MG |
2654 | |
2655 | /* Remove page from free list */ | |
2656 | list_del(&page->lru); | |
2657 | zone->free_area[order].nr_free--; | |
2658 | rmv_page_order(page); | |
2139cbe6 | 2659 | |
400bc7fd | 2660 | /* |
2661 | * Set the pageblock if the isolated page is at least half of a | |
2662 | * pageblock | |
2663 | */ | |
748446bb MG |
2664 | if (order >= pageblock_order - 1) { |
2665 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2666 | for (; page < endpage; page += pageblock_nr_pages) { |
2667 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 2668 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 2669 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
2670 | set_pageblock_migratetype(page, |
2671 | MIGRATE_MOVABLE); | |
2672 | } | |
748446bb MG |
2673 | } |
2674 | ||
f3a14ced | 2675 | |
8fb74b9f | 2676 | return 1UL << order; |
1fb3f8ca MG |
2677 | } |
2678 | ||
060e7417 MG |
2679 | /* |
2680 | * Update NUMA hit/miss statistics | |
2681 | * | |
2682 | * Must be called with interrupts disabled. | |
060e7417 | 2683 | */ |
41b6167e | 2684 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
2685 | { |
2686 | #ifdef CONFIG_NUMA | |
060e7417 MG |
2687 | enum zone_stat_item local_stat = NUMA_LOCAL; |
2688 | ||
2df26639 | 2689 | if (z->node != numa_node_id()) |
060e7417 | 2690 | local_stat = NUMA_OTHER; |
060e7417 | 2691 | |
2df26639 | 2692 | if (z->node == preferred_zone->node) |
060e7417 | 2693 | __inc_zone_state(z, NUMA_HIT); |
2df26639 | 2694 | else { |
060e7417 MG |
2695 | __inc_zone_state(z, NUMA_MISS); |
2696 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); | |
2697 | } | |
2df26639 | 2698 | __inc_zone_state(z, local_stat); |
060e7417 MG |
2699 | #endif |
2700 | } | |
2701 | ||
066b2393 MG |
2702 | /* Remove page from the per-cpu list, caller must protect the list */ |
2703 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
2704 | bool cold, struct per_cpu_pages *pcp, | |
2705 | struct list_head *list) | |
2706 | { | |
2707 | struct page *page; | |
2708 | ||
2709 | do { | |
2710 | if (list_empty(list)) { | |
2711 | pcp->count += rmqueue_bulk(zone, 0, | |
2712 | pcp->batch, list, | |
2713 | migratetype, cold); | |
2714 | if (unlikely(list_empty(list))) | |
2715 | return NULL; | |
2716 | } | |
2717 | ||
2718 | if (cold) | |
2719 | page = list_last_entry(list, struct page, lru); | |
2720 | else | |
2721 | page = list_first_entry(list, struct page, lru); | |
2722 | ||
2723 | list_del(&page->lru); | |
2724 | pcp->count--; | |
2725 | } while (check_new_pcp(page)); | |
2726 | ||
2727 | return page; | |
2728 | } | |
2729 | ||
2730 | /* Lock and remove page from the per-cpu list */ | |
2731 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
2732 | struct zone *zone, unsigned int order, | |
2733 | gfp_t gfp_flags, int migratetype) | |
2734 | { | |
2735 | struct per_cpu_pages *pcp; | |
2736 | struct list_head *list; | |
2737 | bool cold = ((gfp_flags & __GFP_COLD) != 0); | |
2738 | struct page *page; | |
d34b0733 | 2739 | unsigned long flags; |
066b2393 | 2740 | |
d34b0733 | 2741 | local_irq_save(flags); |
066b2393 MG |
2742 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2743 | list = &pcp->lists[migratetype]; | |
2744 | page = __rmqueue_pcplist(zone, migratetype, cold, pcp, list); | |
2745 | if (page) { | |
2746 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
2747 | zone_statistics(preferred_zone, zone); | |
2748 | } | |
d34b0733 | 2749 | local_irq_restore(flags); |
066b2393 MG |
2750 | return page; |
2751 | } | |
2752 | ||
1da177e4 | 2753 | /* |
75379191 | 2754 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2755 | */ |
0a15c3e9 | 2756 | static inline |
066b2393 | 2757 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 2758 | struct zone *zone, unsigned int order, |
c603844b MG |
2759 | gfp_t gfp_flags, unsigned int alloc_flags, |
2760 | int migratetype) | |
1da177e4 LT |
2761 | { |
2762 | unsigned long flags; | |
689bcebf | 2763 | struct page *page; |
1da177e4 | 2764 | |
d34b0733 | 2765 | if (likely(order == 0)) { |
066b2393 MG |
2766 | page = rmqueue_pcplist(preferred_zone, zone, order, |
2767 | gfp_flags, migratetype); | |
2768 | goto out; | |
2769 | } | |
83b9355b | 2770 | |
066b2393 MG |
2771 | /* |
2772 | * We most definitely don't want callers attempting to | |
2773 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2774 | */ | |
2775 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
2776 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 2777 | |
066b2393 MG |
2778 | do { |
2779 | page = NULL; | |
2780 | if (alloc_flags & ALLOC_HARDER) { | |
2781 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2782 | if (page) | |
2783 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
2784 | } | |
a74609fa | 2785 | if (!page) |
066b2393 MG |
2786 | page = __rmqueue(zone, order, migratetype); |
2787 | } while (page && check_new_pages(page, order)); | |
2788 | spin_unlock(&zone->lock); | |
2789 | if (!page) | |
2790 | goto failed; | |
2791 | __mod_zone_freepage_state(zone, -(1 << order), | |
2792 | get_pcppage_migratetype(page)); | |
1da177e4 | 2793 | |
16709d1d | 2794 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 2795 | zone_statistics(preferred_zone, zone); |
a74609fa | 2796 | local_irq_restore(flags); |
1da177e4 | 2797 | |
066b2393 MG |
2798 | out: |
2799 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); | |
1da177e4 | 2800 | return page; |
a74609fa NP |
2801 | |
2802 | failed: | |
2803 | local_irq_restore(flags); | |
a74609fa | 2804 | return NULL; |
1da177e4 LT |
2805 | } |
2806 | ||
933e312e AM |
2807 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2808 | ||
b2588c4b | 2809 | static struct { |
933e312e AM |
2810 | struct fault_attr attr; |
2811 | ||
621a5f7a | 2812 | bool ignore_gfp_highmem; |
71baba4b | 2813 | bool ignore_gfp_reclaim; |
54114994 | 2814 | u32 min_order; |
933e312e AM |
2815 | } fail_page_alloc = { |
2816 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 2817 | .ignore_gfp_reclaim = true, |
621a5f7a | 2818 | .ignore_gfp_highmem = true, |
54114994 | 2819 | .min_order = 1, |
933e312e AM |
2820 | }; |
2821 | ||
2822 | static int __init setup_fail_page_alloc(char *str) | |
2823 | { | |
2824 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
2825 | } | |
2826 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
2827 | ||
deaf386e | 2828 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2829 | { |
54114994 | 2830 | if (order < fail_page_alloc.min_order) |
deaf386e | 2831 | return false; |
933e312e | 2832 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 2833 | return false; |
933e312e | 2834 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 2835 | return false; |
71baba4b MG |
2836 | if (fail_page_alloc.ignore_gfp_reclaim && |
2837 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 2838 | return false; |
933e312e AM |
2839 | |
2840 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
2841 | } | |
2842 | ||
2843 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
2844 | ||
2845 | static int __init fail_page_alloc_debugfs(void) | |
2846 | { | |
f4ae40a6 | 2847 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 2848 | struct dentry *dir; |
933e312e | 2849 | |
dd48c085 AM |
2850 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2851 | &fail_page_alloc.attr); | |
2852 | if (IS_ERR(dir)) | |
2853 | return PTR_ERR(dir); | |
933e312e | 2854 | |
b2588c4b | 2855 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 2856 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
2857 | goto fail; |
2858 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
2859 | &fail_page_alloc.ignore_gfp_highmem)) | |
2860 | goto fail; | |
2861 | if (!debugfs_create_u32("min-order", mode, dir, | |
2862 | &fail_page_alloc.min_order)) | |
2863 | goto fail; | |
2864 | ||
2865 | return 0; | |
2866 | fail: | |
dd48c085 | 2867 | debugfs_remove_recursive(dir); |
933e312e | 2868 | |
b2588c4b | 2869 | return -ENOMEM; |
933e312e AM |
2870 | } |
2871 | ||
2872 | late_initcall(fail_page_alloc_debugfs); | |
2873 | ||
2874 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2875 | ||
2876 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2877 | ||
deaf386e | 2878 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2879 | { |
deaf386e | 2880 | return false; |
933e312e AM |
2881 | } |
2882 | ||
2883 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2884 | ||
1da177e4 | 2885 | /* |
97a16fc8 MG |
2886 | * Return true if free base pages are above 'mark'. For high-order checks it |
2887 | * will return true of the order-0 watermark is reached and there is at least | |
2888 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
2889 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 2890 | */ |
86a294a8 MH |
2891 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
2892 | int classzone_idx, unsigned int alloc_flags, | |
2893 | long free_pages) | |
1da177e4 | 2894 | { |
d23ad423 | 2895 | long min = mark; |
1da177e4 | 2896 | int o; |
c603844b | 2897 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
1da177e4 | 2898 | |
0aaa29a5 | 2899 | /* free_pages may go negative - that's OK */ |
df0a6daa | 2900 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 2901 | |
7fb1d9fc | 2902 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2903 | min -= min / 2; |
0aaa29a5 MG |
2904 | |
2905 | /* | |
2906 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
2907 | * the high-atomic reserves. This will over-estimate the size of the | |
2908 | * atomic reserve but it avoids a search. | |
2909 | */ | |
97a16fc8 | 2910 | if (likely(!alloc_harder)) |
0aaa29a5 MG |
2911 | free_pages -= z->nr_reserved_highatomic; |
2912 | else | |
1da177e4 | 2913 | min -= min / 4; |
e2b19197 | 2914 | |
d95ea5d1 BZ |
2915 | #ifdef CONFIG_CMA |
2916 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2917 | if (!(alloc_flags & ALLOC_CMA)) | |
97a16fc8 | 2918 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2919 | #endif |
026b0814 | 2920 | |
97a16fc8 MG |
2921 | /* |
2922 | * Check watermarks for an order-0 allocation request. If these | |
2923 | * are not met, then a high-order request also cannot go ahead | |
2924 | * even if a suitable page happened to be free. | |
2925 | */ | |
2926 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 2927 | return false; |
1da177e4 | 2928 | |
97a16fc8 MG |
2929 | /* If this is an order-0 request then the watermark is fine */ |
2930 | if (!order) | |
2931 | return true; | |
2932 | ||
2933 | /* For a high-order request, check at least one suitable page is free */ | |
2934 | for (o = order; o < MAX_ORDER; o++) { | |
2935 | struct free_area *area = &z->free_area[o]; | |
2936 | int mt; | |
2937 | ||
2938 | if (!area->nr_free) | |
2939 | continue; | |
2940 | ||
2941 | if (alloc_harder) | |
2942 | return true; | |
1da177e4 | 2943 | |
97a16fc8 MG |
2944 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
2945 | if (!list_empty(&area->free_list[mt])) | |
2946 | return true; | |
2947 | } | |
2948 | ||
2949 | #ifdef CONFIG_CMA | |
2950 | if ((alloc_flags & ALLOC_CMA) && | |
2951 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
2952 | return true; | |
2953 | } | |
2954 | #endif | |
1da177e4 | 2955 | } |
97a16fc8 | 2956 | return false; |
88f5acf8 MG |
2957 | } |
2958 | ||
7aeb09f9 | 2959 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 2960 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
2961 | { |
2962 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2963 | zone_page_state(z, NR_FREE_PAGES)); | |
2964 | } | |
2965 | ||
48ee5f36 MG |
2966 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
2967 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
2968 | { | |
2969 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2970 | long cma_pages = 0; | |
2971 | ||
2972 | #ifdef CONFIG_CMA | |
2973 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2974 | if (!(alloc_flags & ALLOC_CMA)) | |
2975 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
2976 | #endif | |
2977 | ||
2978 | /* | |
2979 | * Fast check for order-0 only. If this fails then the reserves | |
2980 | * need to be calculated. There is a corner case where the check | |
2981 | * passes but only the high-order atomic reserve are free. If | |
2982 | * the caller is !atomic then it'll uselessly search the free | |
2983 | * list. That corner case is then slower but it is harmless. | |
2984 | */ | |
2985 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) | |
2986 | return true; | |
2987 | ||
2988 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2989 | free_pages); | |
2990 | } | |
2991 | ||
7aeb09f9 | 2992 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 2993 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
2994 | { |
2995 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2996 | ||
2997 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2998 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2999 | ||
e2b19197 | 3000 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 3001 | free_pages); |
1da177e4 LT |
3002 | } |
3003 | ||
9276b1bc | 3004 | #ifdef CONFIG_NUMA |
957f822a DR |
3005 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3006 | { | |
e02dc017 | 3007 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
5f7a75ac | 3008 | RECLAIM_DISTANCE; |
957f822a | 3009 | } |
9276b1bc | 3010 | #else /* CONFIG_NUMA */ |
957f822a DR |
3011 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3012 | { | |
3013 | return true; | |
3014 | } | |
9276b1bc PJ |
3015 | #endif /* CONFIG_NUMA */ |
3016 | ||
7fb1d9fc | 3017 | /* |
0798e519 | 3018 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3019 | * a page. |
3020 | */ | |
3021 | static struct page * | |
a9263751 VB |
3022 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3023 | const struct alloc_context *ac) | |
753ee728 | 3024 | { |
c33d6c06 | 3025 | struct zoneref *z = ac->preferred_zoneref; |
5117f45d | 3026 | struct zone *zone; |
3b8c0be4 MG |
3027 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
3028 | ||
7fb1d9fc | 3029 | /* |
9276b1bc | 3030 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3031 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3032 | */ |
c33d6c06 | 3033 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 3034 | ac->nodemask) { |
be06af00 | 3035 | struct page *page; |
e085dbc5 JW |
3036 | unsigned long mark; |
3037 | ||
664eedde MG |
3038 | if (cpusets_enabled() && |
3039 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3040 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3041 | continue; |
a756cf59 JW |
3042 | /* |
3043 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3044 | * want to get it from a node that is within its dirty |
3045 | * limit, such that no single node holds more than its | |
a756cf59 | 3046 | * proportional share of globally allowed dirty pages. |
281e3726 | 3047 | * The dirty limits take into account the node's |
a756cf59 JW |
3048 | * lowmem reserves and high watermark so that kswapd |
3049 | * should be able to balance it without having to | |
3050 | * write pages from its LRU list. | |
3051 | * | |
a756cf59 | 3052 | * XXX: For now, allow allocations to potentially |
281e3726 | 3053 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3054 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3055 | * which is important when on a NUMA setup the allowed |
281e3726 | 3056 | * nodes are together not big enough to reach the |
a756cf59 | 3057 | * global limit. The proper fix for these situations |
281e3726 | 3058 | * will require awareness of nodes in the |
a756cf59 JW |
3059 | * dirty-throttling and the flusher threads. |
3060 | */ | |
3b8c0be4 MG |
3061 | if (ac->spread_dirty_pages) { |
3062 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3063 | continue; | |
3064 | ||
3065 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3066 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3067 | continue; | |
3068 | } | |
3069 | } | |
7fb1d9fc | 3070 | |
e085dbc5 | 3071 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 3072 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 3073 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
3074 | int ret; |
3075 | ||
5dab2911 MG |
3076 | /* Checked here to keep the fast path fast */ |
3077 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3078 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3079 | goto try_this_zone; | |
3080 | ||
a5f5f91d | 3081 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3082 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3083 | continue; |
3084 | ||
a5f5f91d | 3085 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3086 | switch (ret) { |
a5f5f91d | 3087 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3088 | /* did not scan */ |
cd38b115 | 3089 | continue; |
a5f5f91d | 3090 | case NODE_RECLAIM_FULL: |
fa5e084e | 3091 | /* scanned but unreclaimable */ |
cd38b115 | 3092 | continue; |
fa5e084e MG |
3093 | default: |
3094 | /* did we reclaim enough */ | |
fed2719e | 3095 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3096 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3097 | goto try_this_zone; |
3098 | ||
fed2719e | 3099 | continue; |
0798e519 | 3100 | } |
7fb1d9fc RS |
3101 | } |
3102 | ||
fa5e084e | 3103 | try_this_zone: |
066b2393 | 3104 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3105 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3106 | if (page) { |
479f854a | 3107 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3108 | |
3109 | /* | |
3110 | * If this is a high-order atomic allocation then check | |
3111 | * if the pageblock should be reserved for the future | |
3112 | */ | |
3113 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3114 | reserve_highatomic_pageblock(page, zone, order); | |
3115 | ||
75379191 VB |
3116 | return page; |
3117 | } | |
54a6eb5c | 3118 | } |
9276b1bc | 3119 | |
4ffeaf35 | 3120 | return NULL; |
753ee728 MH |
3121 | } |
3122 | ||
29423e77 DR |
3123 | /* |
3124 | * Large machines with many possible nodes should not always dump per-node | |
3125 | * meminfo in irq context. | |
3126 | */ | |
3127 | static inline bool should_suppress_show_mem(void) | |
3128 | { | |
3129 | bool ret = false; | |
3130 | ||
3131 | #if NODES_SHIFT > 8 | |
3132 | ret = in_interrupt(); | |
3133 | #endif | |
3134 | return ret; | |
3135 | } | |
3136 | ||
9af744d7 | 3137 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3138 | { |
a238ab5b | 3139 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
aa187507 | 3140 | static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1); |
a238ab5b | 3141 | |
aa187507 | 3142 | if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs)) |
a238ab5b DH |
3143 | return; |
3144 | ||
3145 | /* | |
3146 | * This documents exceptions given to allocations in certain | |
3147 | * contexts that are allowed to allocate outside current's set | |
3148 | * of allowed nodes. | |
3149 | */ | |
3150 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
3151 | if (test_thread_flag(TIF_MEMDIE) || | |
3152 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
3153 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3154 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3155 | filter &= ~SHOW_MEM_FILTER_NODES; |
3156 | ||
9af744d7 | 3157 | show_mem(filter, nodemask); |
aa187507 MH |
3158 | } |
3159 | ||
a8e99259 | 3160 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3161 | { |
3162 | struct va_format vaf; | |
3163 | va_list args; | |
3164 | static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL, | |
3165 | DEFAULT_RATELIMIT_BURST); | |
3166 | ||
0f7896f1 | 3167 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3168 | return; |
3169 | ||
7877cdcc | 3170 | pr_warn("%s: ", current->comm); |
3ee9a4f0 | 3171 | |
7877cdcc MH |
3172 | va_start(args, fmt); |
3173 | vaf.fmt = fmt; | |
3174 | vaf.va = &args; | |
3175 | pr_cont("%pV", &vaf); | |
3176 | va_end(args); | |
3ee9a4f0 | 3177 | |
685dbf6f DR |
3178 | pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask, &gfp_mask); |
3179 | if (nodemask) | |
3180 | pr_cont("%*pbl\n", nodemask_pr_args(nodemask)); | |
3181 | else | |
3182 | pr_cont("(null)\n"); | |
3183 | ||
a8e99259 | 3184 | cpuset_print_current_mems_allowed(); |
3ee9a4f0 | 3185 | |
a238ab5b | 3186 | dump_stack(); |
685dbf6f | 3187 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3188 | } |
3189 | ||
6c18ba7a MH |
3190 | static inline struct page * |
3191 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3192 | unsigned int alloc_flags, | |
3193 | const struct alloc_context *ac) | |
3194 | { | |
3195 | struct page *page; | |
3196 | ||
3197 | page = get_page_from_freelist(gfp_mask, order, | |
3198 | alloc_flags|ALLOC_CPUSET, ac); | |
3199 | /* | |
3200 | * fallback to ignore cpuset restriction if our nodes | |
3201 | * are depleted | |
3202 | */ | |
3203 | if (!page) | |
3204 | page = get_page_from_freelist(gfp_mask, order, | |
3205 | alloc_flags, ac); | |
3206 | ||
3207 | return page; | |
3208 | } | |
3209 | ||
11e33f6a MG |
3210 | static inline struct page * |
3211 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3212 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3213 | { |
6e0fc46d DR |
3214 | struct oom_control oc = { |
3215 | .zonelist = ac->zonelist, | |
3216 | .nodemask = ac->nodemask, | |
2a966b77 | 3217 | .memcg = NULL, |
6e0fc46d DR |
3218 | .gfp_mask = gfp_mask, |
3219 | .order = order, | |
6e0fc46d | 3220 | }; |
11e33f6a MG |
3221 | struct page *page; |
3222 | ||
9879de73 JW |
3223 | *did_some_progress = 0; |
3224 | ||
9879de73 | 3225 | /* |
dc56401f JW |
3226 | * Acquire the oom lock. If that fails, somebody else is |
3227 | * making progress for us. | |
9879de73 | 3228 | */ |
dc56401f | 3229 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3230 | *did_some_progress = 1; |
11e33f6a | 3231 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3232 | return NULL; |
3233 | } | |
6b1de916 | 3234 | |
11e33f6a MG |
3235 | /* |
3236 | * Go through the zonelist yet one more time, keep very high watermark | |
3237 | * here, this is only to catch a parallel oom killing, we must fail if | |
3238 | * we're still under heavy pressure. | |
3239 | */ | |
a9263751 VB |
3240 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
3241 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3242 | if (page) |
11e33f6a MG |
3243 | goto out; |
3244 | ||
06ad276a MH |
3245 | /* Coredumps can quickly deplete all memory reserves */ |
3246 | if (current->flags & PF_DUMPCORE) | |
3247 | goto out; | |
3248 | /* The OOM killer will not help higher order allocs */ | |
3249 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3250 | goto out; | |
3251 | /* The OOM killer does not needlessly kill tasks for lowmem */ | |
3252 | if (ac->high_zoneidx < ZONE_NORMAL) | |
3253 | goto out; | |
3254 | if (pm_suspended_storage()) | |
3255 | goto out; | |
3256 | /* | |
3257 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3258 | * other request to make a forward progress. | |
3259 | * We are in an unfortunate situation where out_of_memory cannot | |
3260 | * do much for this context but let's try it to at least get | |
3261 | * access to memory reserved if the current task is killed (see | |
3262 | * out_of_memory). Once filesystems are ready to handle allocation | |
3263 | * failures more gracefully we should just bail out here. | |
3264 | */ | |
3265 | ||
3266 | /* The OOM killer may not free memory on a specific node */ | |
3267 | if (gfp_mask & __GFP_THISNODE) | |
3268 | goto out; | |
3da88fb3 | 3269 | |
11e33f6a | 3270 | /* Exhausted what can be done so it's blamo time */ |
5020e285 | 3271 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3272 | *did_some_progress = 1; |
5020e285 | 3273 | |
6c18ba7a MH |
3274 | /* |
3275 | * Help non-failing allocations by giving them access to memory | |
3276 | * reserves | |
3277 | */ | |
3278 | if (gfp_mask & __GFP_NOFAIL) | |
3279 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3280 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3281 | } |
11e33f6a | 3282 | out: |
dc56401f | 3283 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3284 | return page; |
3285 | } | |
3286 | ||
33c2d214 MH |
3287 | /* |
3288 | * Maximum number of compaction retries wit a progress before OOM | |
3289 | * killer is consider as the only way to move forward. | |
3290 | */ | |
3291 | #define MAX_COMPACT_RETRIES 16 | |
3292 | ||
56de7263 MG |
3293 | #ifdef CONFIG_COMPACTION |
3294 | /* Try memory compaction for high-order allocations before reclaim */ | |
3295 | static struct page * | |
3296 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3297 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3298 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3299 | { |
98dd3b48 | 3300 | struct page *page; |
499118e9 | 3301 | unsigned int noreclaim_flag; |
53853e2d VB |
3302 | |
3303 | if (!order) | |
66199712 | 3304 | return NULL; |
66199712 | 3305 | |
499118e9 | 3306 | noreclaim_flag = memalloc_noreclaim_save(); |
c5d01d0d | 3307 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
c3486f53 | 3308 | prio); |
499118e9 | 3309 | memalloc_noreclaim_restore(noreclaim_flag); |
56de7263 | 3310 | |
c5d01d0d | 3311 | if (*compact_result <= COMPACT_INACTIVE) |
98dd3b48 | 3312 | return NULL; |
53853e2d | 3313 | |
98dd3b48 VB |
3314 | /* |
3315 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3316 | * count a compaction stall | |
3317 | */ | |
3318 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3319 | |
31a6c190 | 3320 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
53853e2d | 3321 | |
98dd3b48 VB |
3322 | if (page) { |
3323 | struct zone *zone = page_zone(page); | |
53853e2d | 3324 | |
98dd3b48 VB |
3325 | zone->compact_blockskip_flush = false; |
3326 | compaction_defer_reset(zone, order, true); | |
3327 | count_vm_event(COMPACTSUCCESS); | |
3328 | return page; | |
3329 | } | |
56de7263 | 3330 | |
98dd3b48 VB |
3331 | /* |
3332 | * It's bad if compaction run occurs and fails. The most likely reason | |
3333 | * is that pages exist, but not enough to satisfy watermarks. | |
3334 | */ | |
3335 | count_vm_event(COMPACTFAIL); | |
66199712 | 3336 | |
98dd3b48 | 3337 | cond_resched(); |
56de7263 MG |
3338 | |
3339 | return NULL; | |
3340 | } | |
33c2d214 | 3341 | |
3250845d VB |
3342 | static inline bool |
3343 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3344 | enum compact_result compact_result, | |
3345 | enum compact_priority *compact_priority, | |
d9436498 | 3346 | int *compaction_retries) |
3250845d VB |
3347 | { |
3348 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3349 | int min_priority; |
65190cff MH |
3350 | bool ret = false; |
3351 | int retries = *compaction_retries; | |
3352 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3353 | |
3354 | if (!order) | |
3355 | return false; | |
3356 | ||
d9436498 VB |
3357 | if (compaction_made_progress(compact_result)) |
3358 | (*compaction_retries)++; | |
3359 | ||
3250845d VB |
3360 | /* |
3361 | * compaction considers all the zone as desperately out of memory | |
3362 | * so it doesn't really make much sense to retry except when the | |
3363 | * failure could be caused by insufficient priority | |
3364 | */ | |
d9436498 VB |
3365 | if (compaction_failed(compact_result)) |
3366 | goto check_priority; | |
3250845d VB |
3367 | |
3368 | /* | |
3369 | * make sure the compaction wasn't deferred or didn't bail out early | |
3370 | * due to locks contention before we declare that we should give up. | |
3371 | * But do not retry if the given zonelist is not suitable for | |
3372 | * compaction. | |
3373 | */ | |
65190cff MH |
3374 | if (compaction_withdrawn(compact_result)) { |
3375 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
3376 | goto out; | |
3377 | } | |
3250845d VB |
3378 | |
3379 | /* | |
3380 | * !costly requests are much more important than __GFP_REPEAT | |
3381 | * costly ones because they are de facto nofail and invoke OOM | |
3382 | * killer to move on while costly can fail and users are ready | |
3383 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3384 | * would need much more detailed feedback from compaction to | |
3385 | * make a better decision. | |
3386 | */ | |
3387 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3388 | max_retries /= 4; | |
65190cff MH |
3389 | if (*compaction_retries <= max_retries) { |
3390 | ret = true; | |
3391 | goto out; | |
3392 | } | |
3250845d | 3393 | |
d9436498 VB |
3394 | /* |
3395 | * Make sure there are attempts at the highest priority if we exhausted | |
3396 | * all retries or failed at the lower priorities. | |
3397 | */ | |
3398 | check_priority: | |
c2033b00 VB |
3399 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3400 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3401 | |
c2033b00 | 3402 | if (*compact_priority > min_priority) { |
d9436498 VB |
3403 | (*compact_priority)--; |
3404 | *compaction_retries = 0; | |
65190cff | 3405 | ret = true; |
d9436498 | 3406 | } |
65190cff MH |
3407 | out: |
3408 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
3409 | return ret; | |
3250845d | 3410 | } |
56de7263 MG |
3411 | #else |
3412 | static inline struct page * | |
3413 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3414 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3415 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3416 | { |
33c2d214 | 3417 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3418 | return NULL; |
3419 | } | |
33c2d214 MH |
3420 | |
3421 | static inline bool | |
86a294a8 MH |
3422 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3423 | enum compact_result compact_result, | |
a5508cd8 | 3424 | enum compact_priority *compact_priority, |
d9436498 | 3425 | int *compaction_retries) |
33c2d214 | 3426 | { |
31e49bfd MH |
3427 | struct zone *zone; |
3428 | struct zoneref *z; | |
3429 | ||
3430 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3431 | return false; | |
3432 | ||
3433 | /* | |
3434 | * There are setups with compaction disabled which would prefer to loop | |
3435 | * inside the allocator rather than hit the oom killer prematurely. | |
3436 | * Let's give them a good hope and keep retrying while the order-0 | |
3437 | * watermarks are OK. | |
3438 | */ | |
3439 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3440 | ac->nodemask) { | |
3441 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
3442 | ac_classzone_idx(ac), alloc_flags)) | |
3443 | return true; | |
3444 | } | |
33c2d214 MH |
3445 | return false; |
3446 | } | |
3250845d | 3447 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3448 | |
bba90710 MS |
3449 | /* Perform direct synchronous page reclaim */ |
3450 | static int | |
a9263751 VB |
3451 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3452 | const struct alloc_context *ac) | |
11e33f6a | 3453 | { |
11e33f6a | 3454 | struct reclaim_state reclaim_state; |
bba90710 | 3455 | int progress; |
499118e9 | 3456 | unsigned int noreclaim_flag; |
11e33f6a MG |
3457 | |
3458 | cond_resched(); | |
3459 | ||
3460 | /* We now go into synchronous reclaim */ | |
3461 | cpuset_memory_pressure_bump(); | |
499118e9 | 3462 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a MG |
3463 | lockdep_set_current_reclaim_state(gfp_mask); |
3464 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 3465 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3466 | |
a9263751 VB |
3467 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3468 | ac->nodemask); | |
11e33f6a | 3469 | |
c06b1fca | 3470 | current->reclaim_state = NULL; |
11e33f6a | 3471 | lockdep_clear_current_reclaim_state(); |
499118e9 | 3472 | memalloc_noreclaim_restore(noreclaim_flag); |
11e33f6a MG |
3473 | |
3474 | cond_resched(); | |
3475 | ||
bba90710 MS |
3476 | return progress; |
3477 | } | |
3478 | ||
3479 | /* The really slow allocator path where we enter direct reclaim */ | |
3480 | static inline struct page * | |
3481 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3482 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3483 | unsigned long *did_some_progress) |
bba90710 MS |
3484 | { |
3485 | struct page *page = NULL; | |
3486 | bool drained = false; | |
3487 | ||
a9263751 | 3488 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3489 | if (unlikely(!(*did_some_progress))) |
3490 | return NULL; | |
11e33f6a | 3491 | |
9ee493ce | 3492 | retry: |
31a6c190 | 3493 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3494 | |
3495 | /* | |
3496 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3497 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3498 | * Shrink them them and try again | |
9ee493ce MG |
3499 | */ |
3500 | if (!page && !drained) { | |
29fac03b | 3501 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3502 | drain_all_pages(NULL); |
9ee493ce MG |
3503 | drained = true; |
3504 | goto retry; | |
3505 | } | |
3506 | ||
11e33f6a MG |
3507 | return page; |
3508 | } | |
3509 | ||
a9263751 | 3510 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
3511 | { |
3512 | struct zoneref *z; | |
3513 | struct zone *zone; | |
e1a55637 | 3514 | pg_data_t *last_pgdat = NULL; |
3a025760 | 3515 | |
a9263751 | 3516 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
e1a55637 MG |
3517 | ac->high_zoneidx, ac->nodemask) { |
3518 | if (last_pgdat != zone->zone_pgdat) | |
52e9f87a | 3519 | wakeup_kswapd(zone, order, ac->high_zoneidx); |
e1a55637 MG |
3520 | last_pgdat = zone->zone_pgdat; |
3521 | } | |
3a025760 JW |
3522 | } |
3523 | ||
c603844b | 3524 | static inline unsigned int |
341ce06f PZ |
3525 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3526 | { | |
c603844b | 3527 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3528 | |
a56f57ff | 3529 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3530 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3531 | |
341ce06f PZ |
3532 | /* |
3533 | * The caller may dip into page reserves a bit more if the caller | |
3534 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3535 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3536 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3537 | */ |
e6223a3b | 3538 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3539 | |
d0164adc | 3540 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3541 | /* |
b104a35d DR |
3542 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3543 | * if it can't schedule. | |
5c3240d9 | 3544 | */ |
b104a35d | 3545 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3546 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3547 | /* |
b104a35d | 3548 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3549 | * comment for __cpuset_node_allowed(). |
523b9458 | 3550 | */ |
341ce06f | 3551 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3552 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3553 | alloc_flags |= ALLOC_HARDER; |
3554 | ||
d95ea5d1 | 3555 | #ifdef CONFIG_CMA |
43e7a34d | 3556 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
3557 | alloc_flags |= ALLOC_CMA; |
3558 | #endif | |
341ce06f PZ |
3559 | return alloc_flags; |
3560 | } | |
3561 | ||
072bb0aa MG |
3562 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3563 | { | |
31a6c190 VB |
3564 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) |
3565 | return false; | |
3566 | ||
3567 | if (gfp_mask & __GFP_MEMALLOC) | |
3568 | return true; | |
3569 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) | |
3570 | return true; | |
3571 | if (!in_interrupt() && | |
3572 | ((current->flags & PF_MEMALLOC) || | |
3573 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
3574 | return true; | |
3575 | ||
3576 | return false; | |
072bb0aa MG |
3577 | } |
3578 | ||
0a0337e0 MH |
3579 | /* |
3580 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
3581 | * for the given allocation request. | |
491d79ae JW |
3582 | * |
3583 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
3584 | * without success, or when we couldn't even meet the watermark if we | |
3585 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
3586 | * |
3587 | * Returns true if a retry is viable or false to enter the oom path. | |
3588 | */ | |
3589 | static inline bool | |
3590 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
3591 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 3592 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
3593 | { |
3594 | struct zone *zone; | |
3595 | struct zoneref *z; | |
3596 | ||
423b452e VB |
3597 | /* |
3598 | * Costly allocations might have made a progress but this doesn't mean | |
3599 | * their order will become available due to high fragmentation so | |
3600 | * always increment the no progress counter for them | |
3601 | */ | |
3602 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
3603 | *no_progress_loops = 0; | |
3604 | else | |
3605 | (*no_progress_loops)++; | |
3606 | ||
0a0337e0 MH |
3607 | /* |
3608 | * Make sure we converge to OOM if we cannot make any progress | |
3609 | * several times in the row. | |
3610 | */ | |
04c8716f MK |
3611 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
3612 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 3613 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 3614 | } |
0a0337e0 | 3615 | |
bca67592 MG |
3616 | /* |
3617 | * Keep reclaiming pages while there is a chance this will lead | |
3618 | * somewhere. If none of the target zones can satisfy our allocation | |
3619 | * request even if all reclaimable pages are considered then we are | |
3620 | * screwed and have to go OOM. | |
0a0337e0 MH |
3621 | */ |
3622 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3623 | ac->nodemask) { | |
3624 | unsigned long available; | |
ede37713 | 3625 | unsigned long reclaimable; |
d379f01d MH |
3626 | unsigned long min_wmark = min_wmark_pages(zone); |
3627 | bool wmark; | |
0a0337e0 | 3628 | |
5a1c84b4 | 3629 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 3630 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
3631 | |
3632 | /* | |
491d79ae JW |
3633 | * Would the allocation succeed if we reclaimed all |
3634 | * reclaimable pages? | |
0a0337e0 | 3635 | */ |
d379f01d MH |
3636 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
3637 | ac_classzone_idx(ac), alloc_flags, available); | |
3638 | trace_reclaim_retry_zone(z, order, reclaimable, | |
3639 | available, min_wmark, *no_progress_loops, wmark); | |
3640 | if (wmark) { | |
ede37713 MH |
3641 | /* |
3642 | * If we didn't make any progress and have a lot of | |
3643 | * dirty + writeback pages then we should wait for | |
3644 | * an IO to complete to slow down the reclaim and | |
3645 | * prevent from pre mature OOM | |
3646 | */ | |
3647 | if (!did_some_progress) { | |
11fb9989 | 3648 | unsigned long write_pending; |
ede37713 | 3649 | |
5a1c84b4 MG |
3650 | write_pending = zone_page_state_snapshot(zone, |
3651 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 3652 | |
11fb9989 | 3653 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
3654 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
3655 | return true; | |
3656 | } | |
3657 | } | |
5a1c84b4 | 3658 | |
ede37713 MH |
3659 | /* |
3660 | * Memory allocation/reclaim might be called from a WQ | |
3661 | * context and the current implementation of the WQ | |
3662 | * concurrency control doesn't recognize that | |
3663 | * a particular WQ is congested if the worker thread is | |
3664 | * looping without ever sleeping. Therefore we have to | |
3665 | * do a short sleep here rather than calling | |
3666 | * cond_resched(). | |
3667 | */ | |
3668 | if (current->flags & PF_WQ_WORKER) | |
3669 | schedule_timeout_uninterruptible(1); | |
3670 | else | |
3671 | cond_resched(); | |
3672 | ||
0a0337e0 MH |
3673 | return true; |
3674 | } | |
3675 | } | |
3676 | ||
3677 | return false; | |
3678 | } | |
3679 | ||
11e33f6a MG |
3680 | static inline struct page * |
3681 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3682 | struct alloc_context *ac) |
11e33f6a | 3683 | { |
d0164adc | 3684 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 3685 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 3686 | struct page *page = NULL; |
c603844b | 3687 | unsigned int alloc_flags; |
11e33f6a | 3688 | unsigned long did_some_progress; |
5ce9bfef | 3689 | enum compact_priority compact_priority; |
c5d01d0d | 3690 | enum compact_result compact_result; |
5ce9bfef VB |
3691 | int compaction_retries; |
3692 | int no_progress_loops; | |
63f53dea MH |
3693 | unsigned long alloc_start = jiffies; |
3694 | unsigned int stall_timeout = 10 * HZ; | |
5ce9bfef | 3695 | unsigned int cpuset_mems_cookie; |
1da177e4 | 3696 | |
72807a74 MG |
3697 | /* |
3698 | * In the slowpath, we sanity check order to avoid ever trying to | |
3699 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
3700 | * be using allocators in order of preference for an area that is | |
3701 | * too large. | |
3702 | */ | |
1fc28b70 MG |
3703 | if (order >= MAX_ORDER) { |
3704 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 3705 | return NULL; |
1fc28b70 | 3706 | } |
1da177e4 | 3707 | |
d0164adc MG |
3708 | /* |
3709 | * We also sanity check to catch abuse of atomic reserves being used by | |
3710 | * callers that are not in atomic context. | |
3711 | */ | |
3712 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
3713 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
3714 | gfp_mask &= ~__GFP_ATOMIC; | |
3715 | ||
5ce9bfef VB |
3716 | retry_cpuset: |
3717 | compaction_retries = 0; | |
3718 | no_progress_loops = 0; | |
3719 | compact_priority = DEF_COMPACT_PRIORITY; | |
3720 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
3721 | |
3722 | /* | |
3723 | * The fast path uses conservative alloc_flags to succeed only until | |
3724 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
3725 | * alloc_flags precisely. So we do that now. | |
3726 | */ | |
3727 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
3728 | ||
e47483bc VB |
3729 | /* |
3730 | * We need to recalculate the starting point for the zonelist iterator | |
3731 | * because we might have used different nodemask in the fast path, or | |
3732 | * there was a cpuset modification and we are retrying - otherwise we | |
3733 | * could end up iterating over non-eligible zones endlessly. | |
3734 | */ | |
3735 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
3736 | ac->high_zoneidx, ac->nodemask); | |
3737 | if (!ac->preferred_zoneref->zone) | |
3738 | goto nopage; | |
3739 | ||
23771235 VB |
3740 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3741 | wake_all_kswapds(order, ac); | |
3742 | ||
3743 | /* | |
3744 | * The adjusted alloc_flags might result in immediate success, so try | |
3745 | * that first | |
3746 | */ | |
3747 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
3748 | if (page) | |
3749 | goto got_pg; | |
3750 | ||
a8161d1e VB |
3751 | /* |
3752 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
3753 | * that we have enough base pages and don't need to reclaim. For non- |
3754 | * movable high-order allocations, do that as well, as compaction will | |
3755 | * try prevent permanent fragmentation by migrating from blocks of the | |
3756 | * same migratetype. | |
3757 | * Don't try this for allocations that are allowed to ignore | |
3758 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 3759 | */ |
282722b0 VB |
3760 | if (can_direct_reclaim && |
3761 | (costly_order || | |
3762 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
3763 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
3764 | page = __alloc_pages_direct_compact(gfp_mask, order, |
3765 | alloc_flags, ac, | |
a5508cd8 | 3766 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
3767 | &compact_result); |
3768 | if (page) | |
3769 | goto got_pg; | |
3770 | ||
3eb2771b VB |
3771 | /* |
3772 | * Checks for costly allocations with __GFP_NORETRY, which | |
3773 | * includes THP page fault allocations | |
3774 | */ | |
282722b0 | 3775 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { |
a8161d1e VB |
3776 | /* |
3777 | * If compaction is deferred for high-order allocations, | |
3778 | * it is because sync compaction recently failed. If | |
3779 | * this is the case and the caller requested a THP | |
3780 | * allocation, we do not want to heavily disrupt the | |
3781 | * system, so we fail the allocation instead of entering | |
3782 | * direct reclaim. | |
3783 | */ | |
3784 | if (compact_result == COMPACT_DEFERRED) | |
3785 | goto nopage; | |
3786 | ||
a8161d1e | 3787 | /* |
3eb2771b VB |
3788 | * Looks like reclaim/compaction is worth trying, but |
3789 | * sync compaction could be very expensive, so keep | |
25160354 | 3790 | * using async compaction. |
a8161d1e | 3791 | */ |
a5508cd8 | 3792 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
3793 | } |
3794 | } | |
23771235 | 3795 | |
31a6c190 | 3796 | retry: |
23771235 | 3797 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
31a6c190 VB |
3798 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3799 | wake_all_kswapds(order, ac); | |
3800 | ||
23771235 VB |
3801 | if (gfp_pfmemalloc_allowed(gfp_mask)) |
3802 | alloc_flags = ALLOC_NO_WATERMARKS; | |
3803 | ||
e46e7b77 MG |
3804 | /* |
3805 | * Reset the zonelist iterators if memory policies can be ignored. | |
3806 | * These allocations are high priority and system rather than user | |
3807 | * orientated. | |
3808 | */ | |
23771235 | 3809 | if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) { |
e46e7b77 MG |
3810 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
3811 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
3812 | ac->high_zoneidx, ac->nodemask); | |
3813 | } | |
3814 | ||
23771235 | 3815 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 3816 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
3817 | if (page) |
3818 | goto got_pg; | |
1da177e4 | 3819 | |
d0164adc | 3820 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 3821 | if (!can_direct_reclaim) |
1da177e4 LT |
3822 | goto nopage; |
3823 | ||
9a67f648 MH |
3824 | /* Make sure we know about allocations which stall for too long */ |
3825 | if (time_after(jiffies, alloc_start + stall_timeout)) { | |
82251963 | 3826 | warn_alloc(gfp_mask & ~__GFP_NOWARN, ac->nodemask, |
9a67f648 MH |
3827 | "page allocation stalls for %ums, order:%u", |
3828 | jiffies_to_msecs(jiffies-alloc_start), order); | |
3829 | stall_timeout += 10 * HZ; | |
33d53103 | 3830 | } |
341ce06f | 3831 | |
9a67f648 MH |
3832 | /* Avoid recursion of direct reclaim */ |
3833 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
3834 | goto nopage; |
3835 | ||
a8161d1e VB |
3836 | /* Try direct reclaim and then allocating */ |
3837 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
3838 | &did_some_progress); | |
3839 | if (page) | |
3840 | goto got_pg; | |
3841 | ||
3842 | /* Try direct compaction and then allocating */ | |
a9263751 | 3843 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 3844 | compact_priority, &compact_result); |
56de7263 MG |
3845 | if (page) |
3846 | goto got_pg; | |
75f30861 | 3847 | |
9083905a JW |
3848 | /* Do not loop if specifically requested */ |
3849 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 3850 | goto nopage; |
9083905a | 3851 | |
0a0337e0 MH |
3852 | /* |
3853 | * Do not retry costly high order allocations unless they are | |
3854 | * __GFP_REPEAT | |
3855 | */ | |
282722b0 | 3856 | if (costly_order && !(gfp_mask & __GFP_REPEAT)) |
a8161d1e | 3857 | goto nopage; |
0a0337e0 | 3858 | |
0a0337e0 | 3859 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 3860 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
3861 | goto retry; |
3862 | ||
33c2d214 MH |
3863 | /* |
3864 | * It doesn't make any sense to retry for the compaction if the order-0 | |
3865 | * reclaim is not able to make any progress because the current | |
3866 | * implementation of the compaction depends on the sufficient amount | |
3867 | * of free memory (see __compaction_suitable) | |
3868 | */ | |
3869 | if (did_some_progress > 0 && | |
86a294a8 | 3870 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 3871 | compact_result, &compact_priority, |
d9436498 | 3872 | &compaction_retries)) |
33c2d214 MH |
3873 | goto retry; |
3874 | ||
e47483bc VB |
3875 | /* |
3876 | * It's possible we raced with cpuset update so the OOM would be | |
3877 | * premature (see below the nopage: label for full explanation). | |
3878 | */ | |
3879 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
3880 | goto retry_cpuset; | |
3881 | ||
9083905a JW |
3882 | /* Reclaim has failed us, start killing things */ |
3883 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
3884 | if (page) | |
3885 | goto got_pg; | |
3886 | ||
9a67f648 | 3887 | /* Avoid allocations with no watermarks from looping endlessly */ |
c288983d TH |
3888 | if (test_thread_flag(TIF_MEMDIE) && |
3889 | (alloc_flags == ALLOC_NO_WATERMARKS || | |
3890 | (gfp_mask & __GFP_NOMEMALLOC))) | |
9a67f648 MH |
3891 | goto nopage; |
3892 | ||
9083905a | 3893 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
3894 | if (did_some_progress) { |
3895 | no_progress_loops = 0; | |
9083905a | 3896 | goto retry; |
0a0337e0 | 3897 | } |
9083905a | 3898 | |
1da177e4 | 3899 | nopage: |
5ce9bfef | 3900 | /* |
e47483bc VB |
3901 | * When updating a task's mems_allowed or mempolicy nodemask, it is |
3902 | * possible to race with parallel threads in such a way that our | |
3903 | * allocation can fail while the mask is being updated. If we are about | |
3904 | * to fail, check if the cpuset changed during allocation and if so, | |
3905 | * retry. | |
5ce9bfef VB |
3906 | */ |
3907 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
3908 | goto retry_cpuset; | |
3909 | ||
9a67f648 MH |
3910 | /* |
3911 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
3912 | * we always retry | |
3913 | */ | |
3914 | if (gfp_mask & __GFP_NOFAIL) { | |
3915 | /* | |
3916 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
3917 | * of any new users that actually require GFP_NOWAIT | |
3918 | */ | |
3919 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
3920 | goto fail; | |
3921 | ||
3922 | /* | |
3923 | * PF_MEMALLOC request from this context is rather bizarre | |
3924 | * because we cannot reclaim anything and only can loop waiting | |
3925 | * for somebody to do a work for us | |
3926 | */ | |
3927 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
3928 | ||
3929 | /* | |
3930 | * non failing costly orders are a hard requirement which we | |
3931 | * are not prepared for much so let's warn about these users | |
3932 | * so that we can identify them and convert them to something | |
3933 | * else. | |
3934 | */ | |
3935 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
3936 | ||
6c18ba7a MH |
3937 | /* |
3938 | * Help non-failing allocations by giving them access to memory | |
3939 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
3940 | * could deplete whole memory reserves which would just make | |
3941 | * the situation worse | |
3942 | */ | |
3943 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
3944 | if (page) | |
3945 | goto got_pg; | |
3946 | ||
9a67f648 MH |
3947 | cond_resched(); |
3948 | goto retry; | |
3949 | } | |
3950 | fail: | |
a8e99259 | 3951 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 3952 | "page allocation failure: order:%u", order); |
1da177e4 | 3953 | got_pg: |
072bb0aa | 3954 | return page; |
1da177e4 | 3955 | } |
11e33f6a | 3956 | |
9cd75558 MG |
3957 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
3958 | struct zonelist *zonelist, nodemask_t *nodemask, | |
3959 | struct alloc_context *ac, gfp_t *alloc_mask, | |
3960 | unsigned int *alloc_flags) | |
11e33f6a | 3961 | { |
9cd75558 MG |
3962 | ac->high_zoneidx = gfp_zone(gfp_mask); |
3963 | ac->zonelist = zonelist; | |
3964 | ac->nodemask = nodemask; | |
3965 | ac->migratetype = gfpflags_to_migratetype(gfp_mask); | |
11e33f6a | 3966 | |
682a3385 | 3967 | if (cpusets_enabled()) { |
9cd75558 | 3968 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
3969 | if (!ac->nodemask) |
3970 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
3971 | else |
3972 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
3973 | } |
3974 | ||
11e33f6a MG |
3975 | lockdep_trace_alloc(gfp_mask); |
3976 | ||
d0164adc | 3977 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
3978 | |
3979 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 3980 | return false; |
11e33f6a | 3981 | |
9cd75558 MG |
3982 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
3983 | *alloc_flags |= ALLOC_CMA; | |
3984 | ||
3985 | return true; | |
3986 | } | |
21bb9bd1 | 3987 | |
9cd75558 MG |
3988 | /* Determine whether to spread dirty pages and what the first usable zone */ |
3989 | static inline void finalise_ac(gfp_t gfp_mask, | |
3990 | unsigned int order, struct alloc_context *ac) | |
3991 | { | |
c9ab0c4f | 3992 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 3993 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 3994 | |
e46e7b77 MG |
3995 | /* |
3996 | * The preferred zone is used for statistics but crucially it is | |
3997 | * also used as the starting point for the zonelist iterator. It | |
3998 | * may get reset for allocations that ignore memory policies. | |
3999 | */ | |
9cd75558 MG |
4000 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4001 | ac->high_zoneidx, ac->nodemask); | |
4002 | } | |
4003 | ||
4004 | /* | |
4005 | * This is the 'heart' of the zoned buddy allocator. | |
4006 | */ | |
4007 | struct page * | |
4008 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
4009 | struct zonelist *zonelist, nodemask_t *nodemask) | |
4010 | { | |
4011 | struct page *page; | |
4012 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
4013 | gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */ | |
4014 | struct alloc_context ac = { }; | |
4015 | ||
4016 | gfp_mask &= gfp_allowed_mask; | |
4017 | if (!prepare_alloc_pages(gfp_mask, order, zonelist, nodemask, &ac, &alloc_mask, &alloc_flags)) | |
4018 | return NULL; | |
4019 | ||
4020 | finalise_ac(gfp_mask, order, &ac); | |
5bb1b169 | 4021 | |
5117f45d | 4022 | /* First allocation attempt */ |
a9263751 | 4023 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4024 | if (likely(page)) |
4025 | goto out; | |
11e33f6a | 4026 | |
4fcb0971 | 4027 | /* |
7dea19f9 MH |
4028 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4029 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4030 | * from a particular context which has been marked by | |
4031 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4032 | */ |
7dea19f9 | 4033 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4034 | ac.spread_dirty_pages = false; |
23f086f9 | 4035 | |
4741526b MG |
4036 | /* |
4037 | * Restore the original nodemask if it was potentially replaced with | |
4038 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4039 | */ | |
e47483bc | 4040 | if (unlikely(ac.nodemask != nodemask)) |
4741526b | 4041 | ac.nodemask = nodemask; |
16096c25 | 4042 | |
4fcb0971 | 4043 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4044 | |
4fcb0971 | 4045 | out: |
c4159a75 VD |
4046 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
4047 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { | |
4048 | __free_pages(page, order); | |
4049 | page = NULL; | |
4949148a VD |
4050 | } |
4051 | ||
4fcb0971 MG |
4052 | if (kmemcheck_enabled && page) |
4053 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
4054 | ||
4055 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); | |
4056 | ||
11e33f6a | 4057 | return page; |
1da177e4 | 4058 | } |
d239171e | 4059 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4060 | |
4061 | /* | |
4062 | * Common helper functions. | |
4063 | */ | |
920c7a5d | 4064 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4065 | { |
945a1113 AM |
4066 | struct page *page; |
4067 | ||
4068 | /* | |
4069 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
4070 | * a highmem page | |
4071 | */ | |
4072 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
4073 | ||
1da177e4 LT |
4074 | page = alloc_pages(gfp_mask, order); |
4075 | if (!page) | |
4076 | return 0; | |
4077 | return (unsigned long) page_address(page); | |
4078 | } | |
1da177e4 LT |
4079 | EXPORT_SYMBOL(__get_free_pages); |
4080 | ||
920c7a5d | 4081 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4082 | { |
945a1113 | 4083 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4084 | } |
1da177e4 LT |
4085 | EXPORT_SYMBOL(get_zeroed_page); |
4086 | ||
920c7a5d | 4087 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 4088 | { |
b5810039 | 4089 | if (put_page_testzero(page)) { |
1da177e4 | 4090 | if (order == 0) |
b745bc85 | 4091 | free_hot_cold_page(page, false); |
1da177e4 LT |
4092 | else |
4093 | __free_pages_ok(page, order); | |
4094 | } | |
4095 | } | |
4096 | ||
4097 | EXPORT_SYMBOL(__free_pages); | |
4098 | ||
920c7a5d | 4099 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4100 | { |
4101 | if (addr != 0) { | |
725d704e | 4102 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4103 | __free_pages(virt_to_page((void *)addr), order); |
4104 | } | |
4105 | } | |
4106 | ||
4107 | EXPORT_SYMBOL(free_pages); | |
4108 | ||
b63ae8ca AD |
4109 | /* |
4110 | * Page Fragment: | |
4111 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4112 | * within a 0 or higher order page. Multiple fragments within that page | |
4113 | * are individually refcounted, in the page's reference counter. | |
4114 | * | |
4115 | * The page_frag functions below provide a simple allocation framework for | |
4116 | * page fragments. This is used by the network stack and network device | |
4117 | * drivers to provide a backing region of memory for use as either an | |
4118 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4119 | */ | |
2976db80 AD |
4120 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4121 | gfp_t gfp_mask) | |
b63ae8ca AD |
4122 | { |
4123 | struct page *page = NULL; | |
4124 | gfp_t gfp = gfp_mask; | |
4125 | ||
4126 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4127 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4128 | __GFP_NOMEMALLOC; | |
4129 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4130 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4131 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4132 | #endif | |
4133 | if (unlikely(!page)) | |
4134 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4135 | ||
4136 | nc->va = page ? page_address(page) : NULL; | |
4137 | ||
4138 | return page; | |
4139 | } | |
4140 | ||
2976db80 | 4141 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4142 | { |
4143 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4144 | ||
4145 | if (page_ref_sub_and_test(page, count)) { | |
2976db80 AD |
4146 | unsigned int order = compound_order(page); |
4147 | ||
44fdffd7 AD |
4148 | if (order == 0) |
4149 | free_hot_cold_page(page, false); | |
4150 | else | |
4151 | __free_pages_ok(page, order); | |
4152 | } | |
4153 | } | |
2976db80 | 4154 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4155 | |
8c2dd3e4 AD |
4156 | void *page_frag_alloc(struct page_frag_cache *nc, |
4157 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4158 | { |
4159 | unsigned int size = PAGE_SIZE; | |
4160 | struct page *page; | |
4161 | int offset; | |
4162 | ||
4163 | if (unlikely(!nc->va)) { | |
4164 | refill: | |
2976db80 | 4165 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4166 | if (!page) |
4167 | return NULL; | |
4168 | ||
4169 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4170 | /* if size can vary use size else just use PAGE_SIZE */ | |
4171 | size = nc->size; | |
4172 | #endif | |
4173 | /* Even if we own the page, we do not use atomic_set(). | |
4174 | * This would break get_page_unless_zero() users. | |
4175 | */ | |
fe896d18 | 4176 | page_ref_add(page, size - 1); |
b63ae8ca AD |
4177 | |
4178 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4179 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
4180 | nc->pagecnt_bias = size; |
4181 | nc->offset = size; | |
4182 | } | |
4183 | ||
4184 | offset = nc->offset - fragsz; | |
4185 | if (unlikely(offset < 0)) { | |
4186 | page = virt_to_page(nc->va); | |
4187 | ||
fe896d18 | 4188 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4189 | goto refill; |
4190 | ||
4191 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4192 | /* if size can vary use size else just use PAGE_SIZE */ | |
4193 | size = nc->size; | |
4194 | #endif | |
4195 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 4196 | set_page_count(page, size); |
b63ae8ca AD |
4197 | |
4198 | /* reset page count bias and offset to start of new frag */ | |
4199 | nc->pagecnt_bias = size; | |
4200 | offset = size - fragsz; | |
4201 | } | |
4202 | ||
4203 | nc->pagecnt_bias--; | |
4204 | nc->offset = offset; | |
4205 | ||
4206 | return nc->va + offset; | |
4207 | } | |
8c2dd3e4 | 4208 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
4209 | |
4210 | /* | |
4211 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4212 | */ | |
8c2dd3e4 | 4213 | void page_frag_free(void *addr) |
b63ae8ca AD |
4214 | { |
4215 | struct page *page = virt_to_head_page(addr); | |
4216 | ||
4217 | if (unlikely(put_page_testzero(page))) | |
4218 | __free_pages_ok(page, compound_order(page)); | |
4219 | } | |
8c2dd3e4 | 4220 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4221 | |
d00181b9 KS |
4222 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4223 | size_t size) | |
ee85c2e1 AK |
4224 | { |
4225 | if (addr) { | |
4226 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
4227 | unsigned long used = addr + PAGE_ALIGN(size); | |
4228 | ||
4229 | split_page(virt_to_page((void *)addr), order); | |
4230 | while (used < alloc_end) { | |
4231 | free_page(used); | |
4232 | used += PAGE_SIZE; | |
4233 | } | |
4234 | } | |
4235 | return (void *)addr; | |
4236 | } | |
4237 | ||
2be0ffe2 TT |
4238 | /** |
4239 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4240 | * @size: the number of bytes to allocate | |
4241 | * @gfp_mask: GFP flags for the allocation | |
4242 | * | |
4243 | * This function is similar to alloc_pages(), except that it allocates the | |
4244 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4245 | * allocate memory in power-of-two pages. | |
4246 | * | |
4247 | * This function is also limited by MAX_ORDER. | |
4248 | * | |
4249 | * Memory allocated by this function must be released by free_pages_exact(). | |
4250 | */ | |
4251 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4252 | { | |
4253 | unsigned int order = get_order(size); | |
4254 | unsigned long addr; | |
4255 | ||
4256 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 4257 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4258 | } |
4259 | EXPORT_SYMBOL(alloc_pages_exact); | |
4260 | ||
ee85c2e1 AK |
4261 | /** |
4262 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4263 | * pages on a node. | |
b5e6ab58 | 4264 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
4265 | * @size: the number of bytes to allocate |
4266 | * @gfp_mask: GFP flags for the allocation | |
4267 | * | |
4268 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4269 | * back. | |
ee85c2e1 | 4270 | */ |
e1931811 | 4271 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4272 | { |
d00181b9 | 4273 | unsigned int order = get_order(size); |
ee85c2e1 AK |
4274 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
4275 | if (!p) | |
4276 | return NULL; | |
4277 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4278 | } | |
ee85c2e1 | 4279 | |
2be0ffe2 TT |
4280 | /** |
4281 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4282 | * @virt: the value returned by alloc_pages_exact. | |
4283 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4284 | * | |
4285 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4286 | */ | |
4287 | void free_pages_exact(void *virt, size_t size) | |
4288 | { | |
4289 | unsigned long addr = (unsigned long)virt; | |
4290 | unsigned long end = addr + PAGE_ALIGN(size); | |
4291 | ||
4292 | while (addr < end) { | |
4293 | free_page(addr); | |
4294 | addr += PAGE_SIZE; | |
4295 | } | |
4296 | } | |
4297 | EXPORT_SYMBOL(free_pages_exact); | |
4298 | ||
e0fb5815 ZY |
4299 | /** |
4300 | * nr_free_zone_pages - count number of pages beyond high watermark | |
4301 | * @offset: The zone index of the highest zone | |
4302 | * | |
4303 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
4304 | * high watermark within all zones at or below a given zone index. For each | |
4305 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
4306 | * |
4307 | * nr_free_zone_pages = managed_pages - high_pages | |
e0fb5815 | 4308 | */ |
ebec3862 | 4309 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 4310 | { |
dd1a239f | 4311 | struct zoneref *z; |
54a6eb5c MG |
4312 | struct zone *zone; |
4313 | ||
e310fd43 | 4314 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 4315 | unsigned long sum = 0; |
1da177e4 | 4316 | |
0e88460d | 4317 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 4318 | |
54a6eb5c | 4319 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 4320 | unsigned long size = zone->managed_pages; |
41858966 | 4321 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
4322 | if (size > high) |
4323 | sum += size - high; | |
1da177e4 LT |
4324 | } |
4325 | ||
4326 | return sum; | |
4327 | } | |
4328 | ||
e0fb5815 ZY |
4329 | /** |
4330 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
4331 | * | |
4332 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
4333 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 4334 | */ |
ebec3862 | 4335 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 4336 | { |
af4ca457 | 4337 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 4338 | } |
c2f1a551 | 4339 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 4340 | |
e0fb5815 ZY |
4341 | /** |
4342 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
4343 | * | |
4344 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
4345 | * high watermark within all zones. | |
1da177e4 | 4346 | */ |
ebec3862 | 4347 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 4348 | { |
2a1e274a | 4349 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 4350 | } |
08e0f6a9 CL |
4351 | |
4352 | static inline void show_node(struct zone *zone) | |
1da177e4 | 4353 | { |
e5adfffc | 4354 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 4355 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 4356 | } |
1da177e4 | 4357 | |
d02bd27b IR |
4358 | long si_mem_available(void) |
4359 | { | |
4360 | long available; | |
4361 | unsigned long pagecache; | |
4362 | unsigned long wmark_low = 0; | |
4363 | unsigned long pages[NR_LRU_LISTS]; | |
4364 | struct zone *zone; | |
4365 | int lru; | |
4366 | ||
4367 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 4368 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
4369 | |
4370 | for_each_zone(zone) | |
4371 | wmark_low += zone->watermark[WMARK_LOW]; | |
4372 | ||
4373 | /* | |
4374 | * Estimate the amount of memory available for userspace allocations, | |
4375 | * without causing swapping. | |
4376 | */ | |
4377 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; | |
4378 | ||
4379 | /* | |
4380 | * Not all the page cache can be freed, otherwise the system will | |
4381 | * start swapping. Assume at least half of the page cache, or the | |
4382 | * low watermark worth of cache, needs to stay. | |
4383 | */ | |
4384 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
4385 | pagecache -= min(pagecache / 2, wmark_low); | |
4386 | available += pagecache; | |
4387 | ||
4388 | /* | |
4389 | * Part of the reclaimable slab consists of items that are in use, | |
4390 | * and cannot be freed. Cap this estimate at the low watermark. | |
4391 | */ | |
4392 | available += global_page_state(NR_SLAB_RECLAIMABLE) - | |
4393 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); | |
4394 | ||
4395 | if (available < 0) | |
4396 | available = 0; | |
4397 | return available; | |
4398 | } | |
4399 | EXPORT_SYMBOL_GPL(si_mem_available); | |
4400 | ||
1da177e4 LT |
4401 | void si_meminfo(struct sysinfo *val) |
4402 | { | |
4403 | val->totalram = totalram_pages; | |
11fb9989 | 4404 | val->sharedram = global_node_page_state(NR_SHMEM); |
d23ad423 | 4405 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 4406 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
4407 | val->totalhigh = totalhigh_pages; |
4408 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
4409 | val->mem_unit = PAGE_SIZE; |
4410 | } | |
4411 | ||
4412 | EXPORT_SYMBOL(si_meminfo); | |
4413 | ||
4414 | #ifdef CONFIG_NUMA | |
4415 | void si_meminfo_node(struct sysinfo *val, int nid) | |
4416 | { | |
cdd91a77 JL |
4417 | int zone_type; /* needs to be signed */ |
4418 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
4419 | unsigned long managed_highpages = 0; |
4420 | unsigned long free_highpages = 0; | |
1da177e4 LT |
4421 | pg_data_t *pgdat = NODE_DATA(nid); |
4422 | ||
cdd91a77 JL |
4423 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
4424 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
4425 | val->totalram = managed_pages; | |
11fb9989 | 4426 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 4427 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 4428 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
4429 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4430 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4431 | ||
4432 | if (is_highmem(zone)) { | |
4433 | managed_highpages += zone->managed_pages; | |
4434 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
4435 | } | |
4436 | } | |
4437 | val->totalhigh = managed_highpages; | |
4438 | val->freehigh = free_highpages; | |
98d2b0eb | 4439 | #else |
fc2bd799 JK |
4440 | val->totalhigh = managed_highpages; |
4441 | val->freehigh = free_highpages; | |
98d2b0eb | 4442 | #endif |
1da177e4 LT |
4443 | val->mem_unit = PAGE_SIZE; |
4444 | } | |
4445 | #endif | |
4446 | ||
ddd588b5 | 4447 | /* |
7bf02ea2 DR |
4448 | * Determine whether the node should be displayed or not, depending on whether |
4449 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 4450 | */ |
9af744d7 | 4451 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 4452 | { |
ddd588b5 | 4453 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 4454 | return false; |
ddd588b5 | 4455 | |
9af744d7 MH |
4456 | /* |
4457 | * no node mask - aka implicit memory numa policy. Do not bother with | |
4458 | * the synchronization - read_mems_allowed_begin - because we do not | |
4459 | * have to be precise here. | |
4460 | */ | |
4461 | if (!nodemask) | |
4462 | nodemask = &cpuset_current_mems_allowed; | |
4463 | ||
4464 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
4465 | } |
4466 | ||
1da177e4 LT |
4467 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4468 | ||
377e4f16 RV |
4469 | static void show_migration_types(unsigned char type) |
4470 | { | |
4471 | static const char types[MIGRATE_TYPES] = { | |
4472 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 4473 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
4474 | [MIGRATE_RECLAIMABLE] = 'E', |
4475 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
4476 | #ifdef CONFIG_CMA |
4477 | [MIGRATE_CMA] = 'C', | |
4478 | #endif | |
194159fb | 4479 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 4480 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 4481 | #endif |
377e4f16 RV |
4482 | }; |
4483 | char tmp[MIGRATE_TYPES + 1]; | |
4484 | char *p = tmp; | |
4485 | int i; | |
4486 | ||
4487 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
4488 | if (type & (1 << i)) | |
4489 | *p++ = types[i]; | |
4490 | } | |
4491 | ||
4492 | *p = '\0'; | |
1f84a18f | 4493 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
4494 | } |
4495 | ||
1da177e4 LT |
4496 | /* |
4497 | * Show free area list (used inside shift_scroll-lock stuff) | |
4498 | * We also calculate the percentage fragmentation. We do this by counting the | |
4499 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
4500 | * |
4501 | * Bits in @filter: | |
4502 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
4503 | * cpuset. | |
1da177e4 | 4504 | */ |
9af744d7 | 4505 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 4506 | { |
d1bfcdb8 | 4507 | unsigned long free_pcp = 0; |
c7241913 | 4508 | int cpu; |
1da177e4 | 4509 | struct zone *zone; |
599d0c95 | 4510 | pg_data_t *pgdat; |
1da177e4 | 4511 | |
ee99c71c | 4512 | for_each_populated_zone(zone) { |
9af744d7 | 4513 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4514 | continue; |
d1bfcdb8 | 4515 | |
761b0677 KK |
4516 | for_each_online_cpu(cpu) |
4517 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
4518 | } |
4519 | ||
a731286d KM |
4520 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4521 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
4522 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4523 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 4524 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 4525 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
4526 | global_node_page_state(NR_ACTIVE_ANON), |
4527 | global_node_page_state(NR_INACTIVE_ANON), | |
4528 | global_node_page_state(NR_ISOLATED_ANON), | |
4529 | global_node_page_state(NR_ACTIVE_FILE), | |
4530 | global_node_page_state(NR_INACTIVE_FILE), | |
4531 | global_node_page_state(NR_ISOLATED_FILE), | |
4532 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
4533 | global_node_page_state(NR_FILE_DIRTY), |
4534 | global_node_page_state(NR_WRITEBACK), | |
4535 | global_node_page_state(NR_UNSTABLE_NFS), | |
3701b033 KM |
4536 | global_page_state(NR_SLAB_RECLAIMABLE), |
4537 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 4538 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 4539 | global_node_page_state(NR_SHMEM), |
a25700a5 | 4540 | global_page_state(NR_PAGETABLE), |
d1ce749a | 4541 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
4542 | global_page_state(NR_FREE_PAGES), |
4543 | free_pcp, | |
d1ce749a | 4544 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4545 | |
599d0c95 | 4546 | for_each_online_pgdat(pgdat) { |
9af744d7 | 4547 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
4548 | continue; |
4549 | ||
599d0c95 MG |
4550 | printk("Node %d" |
4551 | " active_anon:%lukB" | |
4552 | " inactive_anon:%lukB" | |
4553 | " active_file:%lukB" | |
4554 | " inactive_file:%lukB" | |
4555 | " unevictable:%lukB" | |
4556 | " isolated(anon):%lukB" | |
4557 | " isolated(file):%lukB" | |
50658e2e | 4558 | " mapped:%lukB" |
11fb9989 MG |
4559 | " dirty:%lukB" |
4560 | " writeback:%lukB" | |
4561 | " shmem:%lukB" | |
4562 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4563 | " shmem_thp: %lukB" | |
4564 | " shmem_pmdmapped: %lukB" | |
4565 | " anon_thp: %lukB" | |
4566 | #endif | |
4567 | " writeback_tmp:%lukB" | |
4568 | " unstable:%lukB" | |
599d0c95 MG |
4569 | " all_unreclaimable? %s" |
4570 | "\n", | |
4571 | pgdat->node_id, | |
4572 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
4573 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
4574 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
4575 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
4576 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
4577 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
4578 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 4579 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
4580 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
4581 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 4582 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
4583 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4584 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
4585 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
4586 | * HPAGE_PMD_NR), | |
4587 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
4588 | #endif | |
11fb9989 MG |
4589 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
4590 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), | |
c73322d0 JW |
4591 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
4592 | "yes" : "no"); | |
599d0c95 MG |
4593 | } |
4594 | ||
ee99c71c | 4595 | for_each_populated_zone(zone) { |
1da177e4 LT |
4596 | int i; |
4597 | ||
9af744d7 | 4598 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4599 | continue; |
d1bfcdb8 KK |
4600 | |
4601 | free_pcp = 0; | |
4602 | for_each_online_cpu(cpu) | |
4603 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
4604 | ||
1da177e4 | 4605 | show_node(zone); |
1f84a18f JP |
4606 | printk(KERN_CONT |
4607 | "%s" | |
1da177e4 LT |
4608 | " free:%lukB" |
4609 | " min:%lukB" | |
4610 | " low:%lukB" | |
4611 | " high:%lukB" | |
71c799f4 MK |
4612 | " active_anon:%lukB" |
4613 | " inactive_anon:%lukB" | |
4614 | " active_file:%lukB" | |
4615 | " inactive_file:%lukB" | |
4616 | " unevictable:%lukB" | |
5a1c84b4 | 4617 | " writepending:%lukB" |
1da177e4 | 4618 | " present:%lukB" |
9feedc9d | 4619 | " managed:%lukB" |
4a0aa73f | 4620 | " mlocked:%lukB" |
4a0aa73f KM |
4621 | " slab_reclaimable:%lukB" |
4622 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 4623 | " kernel_stack:%lukB" |
4a0aa73f | 4624 | " pagetables:%lukB" |
4a0aa73f | 4625 | " bounce:%lukB" |
d1bfcdb8 KK |
4626 | " free_pcp:%lukB" |
4627 | " local_pcp:%ukB" | |
d1ce749a | 4628 | " free_cma:%lukB" |
1da177e4 LT |
4629 | "\n", |
4630 | zone->name, | |
88f5acf8 | 4631 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
4632 | K(min_wmark_pages(zone)), |
4633 | K(low_wmark_pages(zone)), | |
4634 | K(high_wmark_pages(zone)), | |
71c799f4 MK |
4635 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
4636 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
4637 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
4638 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
4639 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 4640 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 4641 | K(zone->present_pages), |
9feedc9d | 4642 | K(zone->managed_pages), |
4a0aa73f | 4643 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f KM |
4644 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
4645 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
d30dd8be | 4646 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4a0aa73f | 4647 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 4648 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
4649 | K(free_pcp), |
4650 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 4651 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
4652 | printk("lowmem_reserve[]:"); |
4653 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
4654 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
4655 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
4656 | } |
4657 | ||
ee99c71c | 4658 | for_each_populated_zone(zone) { |
d00181b9 KS |
4659 | unsigned int order; |
4660 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 4661 | unsigned char types[MAX_ORDER]; |
1da177e4 | 4662 | |
9af744d7 | 4663 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4664 | continue; |
1da177e4 | 4665 | show_node(zone); |
1f84a18f | 4666 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
4667 | |
4668 | spin_lock_irqsave(&zone->lock, flags); | |
4669 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
4670 | struct free_area *area = &zone->free_area[order]; |
4671 | int type; | |
4672 | ||
4673 | nr[order] = area->nr_free; | |
8f9de51a | 4674 | total += nr[order] << order; |
377e4f16 RV |
4675 | |
4676 | types[order] = 0; | |
4677 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
4678 | if (!list_empty(&area->free_list[type])) | |
4679 | types[order] |= 1 << type; | |
4680 | } | |
1da177e4 LT |
4681 | } |
4682 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 4683 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
4684 | printk(KERN_CONT "%lu*%lukB ", |
4685 | nr[order], K(1UL) << order); | |
377e4f16 RV |
4686 | if (nr[order]) |
4687 | show_migration_types(types[order]); | |
4688 | } | |
1f84a18f | 4689 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
4690 | } |
4691 | ||
949f7ec5 DR |
4692 | hugetlb_show_meminfo(); |
4693 | ||
11fb9989 | 4694 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 4695 | |
1da177e4 LT |
4696 | show_swap_cache_info(); |
4697 | } | |
4698 | ||
19770b32 MG |
4699 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4700 | { | |
4701 | zoneref->zone = zone; | |
4702 | zoneref->zone_idx = zone_idx(zone); | |
4703 | } | |
4704 | ||
1da177e4 LT |
4705 | /* |
4706 | * Builds allocation fallback zone lists. | |
1a93205b CL |
4707 | * |
4708 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 4709 | */ |
f0c0b2b8 | 4710 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 4711 | int nr_zones) |
1da177e4 | 4712 | { |
1a93205b | 4713 | struct zone *zone; |
bc732f1d | 4714 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
4715 | |
4716 | do { | |
2f6726e5 | 4717 | zone_type--; |
070f8032 | 4718 | zone = pgdat->node_zones + zone_type; |
6aa303de | 4719 | if (managed_zone(zone)) { |
dd1a239f MG |
4720 | zoneref_set_zone(zone, |
4721 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 4722 | check_highest_zone(zone_type); |
1da177e4 | 4723 | } |
2f6726e5 | 4724 | } while (zone_type); |
bc732f1d | 4725 | |
070f8032 | 4726 | return nr_zones; |
1da177e4 LT |
4727 | } |
4728 | ||
f0c0b2b8 KH |
4729 | |
4730 | /* | |
4731 | * zonelist_order: | |
4732 | * 0 = automatic detection of better ordering. | |
4733 | * 1 = order by ([node] distance, -zonetype) | |
4734 | * 2 = order by (-zonetype, [node] distance) | |
4735 | * | |
4736 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
4737 | * the same zonelist. So only NUMA can configure this param. | |
4738 | */ | |
4739 | #define ZONELIST_ORDER_DEFAULT 0 | |
4740 | #define ZONELIST_ORDER_NODE 1 | |
4741 | #define ZONELIST_ORDER_ZONE 2 | |
4742 | ||
4743 | /* zonelist order in the kernel. | |
4744 | * set_zonelist_order() will set this to NODE or ZONE. | |
4745 | */ | |
4746 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4747 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
4748 | ||
4749 | ||
1da177e4 | 4750 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
4751 | /* The value user specified ....changed by config */ |
4752 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4753 | /* string for sysctl */ | |
4754 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
4755 | char numa_zonelist_order[16] = "default"; | |
4756 | ||
4757 | /* | |
4758 | * interface for configure zonelist ordering. | |
4759 | * command line option "numa_zonelist_order" | |
4760 | * = "[dD]efault - default, automatic configuration. | |
4761 | * = "[nN]ode - order by node locality, then by zone within node | |
4762 | * = "[zZ]one - order by zone, then by locality within zone | |
4763 | */ | |
4764 | ||
4765 | static int __parse_numa_zonelist_order(char *s) | |
4766 | { | |
4767 | if (*s == 'd' || *s == 'D') { | |
4768 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4769 | } else if (*s == 'n' || *s == 'N') { | |
4770 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
4771 | } else if (*s == 'z' || *s == 'Z') { | |
4772 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
4773 | } else { | |
1170532b | 4774 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
f0c0b2b8 KH |
4775 | return -EINVAL; |
4776 | } | |
4777 | return 0; | |
4778 | } | |
4779 | ||
4780 | static __init int setup_numa_zonelist_order(char *s) | |
4781 | { | |
ecb256f8 VL |
4782 | int ret; |
4783 | ||
4784 | if (!s) | |
4785 | return 0; | |
4786 | ||
4787 | ret = __parse_numa_zonelist_order(s); | |
4788 | if (ret == 0) | |
4789 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
4790 | ||
4791 | return ret; | |
f0c0b2b8 KH |
4792 | } |
4793 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
4794 | ||
4795 | /* | |
4796 | * sysctl handler for numa_zonelist_order | |
4797 | */ | |
cccad5b9 | 4798 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 4799 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
4800 | loff_t *ppos) |
4801 | { | |
4802 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
4803 | int ret; | |
443c6f14 | 4804 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 4805 | |
443c6f14 | 4806 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
4807 | if (write) { |
4808 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
4809 | ret = -EINVAL; | |
4810 | goto out; | |
4811 | } | |
4812 | strcpy(saved_string, (char *)table->data); | |
4813 | } | |
8d65af78 | 4814 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 4815 | if (ret) |
443c6f14 | 4816 | goto out; |
f0c0b2b8 KH |
4817 | if (write) { |
4818 | int oldval = user_zonelist_order; | |
dacbde09 CG |
4819 | |
4820 | ret = __parse_numa_zonelist_order((char *)table->data); | |
4821 | if (ret) { | |
f0c0b2b8 KH |
4822 | /* |
4823 | * bogus value. restore saved string | |
4824 | */ | |
dacbde09 | 4825 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
4826 | NUMA_ZONELIST_ORDER_LEN); |
4827 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
4828 | } else if (oldval != user_zonelist_order) { |
4829 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 4830 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
4831 | mutex_unlock(&zonelists_mutex); |
4832 | } | |
f0c0b2b8 | 4833 | } |
443c6f14 AK |
4834 | out: |
4835 | mutex_unlock(&zl_order_mutex); | |
4836 | return ret; | |
f0c0b2b8 KH |
4837 | } |
4838 | ||
4839 | ||
62bc62a8 | 4840 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
4841 | static int node_load[MAX_NUMNODES]; |
4842 | ||
1da177e4 | 4843 | /** |
4dc3b16b | 4844 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
4845 | * @node: node whose fallback list we're appending |
4846 | * @used_node_mask: nodemask_t of already used nodes | |
4847 | * | |
4848 | * We use a number of factors to determine which is the next node that should | |
4849 | * appear on a given node's fallback list. The node should not have appeared | |
4850 | * already in @node's fallback list, and it should be the next closest node | |
4851 | * according to the distance array (which contains arbitrary distance values | |
4852 | * from each node to each node in the system), and should also prefer nodes | |
4853 | * with no CPUs, since presumably they'll have very little allocation pressure | |
4854 | * on them otherwise. | |
4855 | * It returns -1 if no node is found. | |
4856 | */ | |
f0c0b2b8 | 4857 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 4858 | { |
4cf808eb | 4859 | int n, val; |
1da177e4 | 4860 | int min_val = INT_MAX; |
00ef2d2f | 4861 | int best_node = NUMA_NO_NODE; |
a70f7302 | 4862 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 4863 | |
4cf808eb LT |
4864 | /* Use the local node if we haven't already */ |
4865 | if (!node_isset(node, *used_node_mask)) { | |
4866 | node_set(node, *used_node_mask); | |
4867 | return node; | |
4868 | } | |
1da177e4 | 4869 | |
4b0ef1fe | 4870 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
4871 | |
4872 | /* Don't want a node to appear more than once */ | |
4873 | if (node_isset(n, *used_node_mask)) | |
4874 | continue; | |
4875 | ||
1da177e4 LT |
4876 | /* Use the distance array to find the distance */ |
4877 | val = node_distance(node, n); | |
4878 | ||
4cf808eb LT |
4879 | /* Penalize nodes under us ("prefer the next node") */ |
4880 | val += (n < node); | |
4881 | ||
1da177e4 | 4882 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
4883 | tmp = cpumask_of_node(n); |
4884 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
4885 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4886 | ||
4887 | /* Slight preference for less loaded node */ | |
4888 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
4889 | val += node_load[n]; | |
4890 | ||
4891 | if (val < min_val) { | |
4892 | min_val = val; | |
4893 | best_node = n; | |
4894 | } | |
4895 | } | |
4896 | ||
4897 | if (best_node >= 0) | |
4898 | node_set(best_node, *used_node_mask); | |
4899 | ||
4900 | return best_node; | |
4901 | } | |
4902 | ||
f0c0b2b8 KH |
4903 | |
4904 | /* | |
4905 | * Build zonelists ordered by node and zones within node. | |
4906 | * This results in maximum locality--normal zone overflows into local | |
4907 | * DMA zone, if any--but risks exhausting DMA zone. | |
4908 | */ | |
4909 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 4910 | { |
f0c0b2b8 | 4911 | int j; |
1da177e4 | 4912 | struct zonelist *zonelist; |
f0c0b2b8 | 4913 | |
c9634cf0 | 4914 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
dd1a239f | 4915 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 4916 | ; |
bc732f1d | 4917 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
4918 | zonelist->_zonerefs[j].zone = NULL; |
4919 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
4920 | } |
4921 | ||
523b9458 CL |
4922 | /* |
4923 | * Build gfp_thisnode zonelists | |
4924 | */ | |
4925 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
4926 | { | |
523b9458 CL |
4927 | int j; |
4928 | struct zonelist *zonelist; | |
4929 | ||
c9634cf0 | 4930 | zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK]; |
bc732f1d | 4931 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
4932 | zonelist->_zonerefs[j].zone = NULL; |
4933 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
4934 | } |
4935 | ||
f0c0b2b8 KH |
4936 | /* |
4937 | * Build zonelists ordered by zone and nodes within zones. | |
4938 | * This results in conserving DMA zone[s] until all Normal memory is | |
4939 | * exhausted, but results in overflowing to remote node while memory | |
4940 | * may still exist in local DMA zone. | |
4941 | */ | |
4942 | static int node_order[MAX_NUMNODES]; | |
4943 | ||
4944 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
4945 | { | |
f0c0b2b8 KH |
4946 | int pos, j, node; |
4947 | int zone_type; /* needs to be signed */ | |
4948 | struct zone *z; | |
4949 | struct zonelist *zonelist; | |
4950 | ||
c9634cf0 | 4951 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
54a6eb5c MG |
4952 | pos = 0; |
4953 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
4954 | for (j = 0; j < nr_nodes; j++) { | |
4955 | node = node_order[j]; | |
4956 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
6aa303de | 4957 | if (managed_zone(z)) { |
dd1a239f MG |
4958 | zoneref_set_zone(z, |
4959 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 4960 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
4961 | } |
4962 | } | |
f0c0b2b8 | 4963 | } |
dd1a239f MG |
4964 | zonelist->_zonerefs[pos].zone = NULL; |
4965 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
4966 | } |
4967 | ||
3193913c MG |
4968 | #if defined(CONFIG_64BIT) |
4969 | /* | |
4970 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
4971 | * penalty to every machine in case the specialised case applies. Default | |
4972 | * to Node-ordering on 64-bit NUMA machines | |
4973 | */ | |
4974 | static int default_zonelist_order(void) | |
4975 | { | |
4976 | return ZONELIST_ORDER_NODE; | |
4977 | } | |
4978 | #else | |
4979 | /* | |
4980 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
4981 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
4982 | * then reclaim will occur more frequency increasing stalls and potentially | |
4983 | * be easier to OOM if a large percentage of the zone is under writeback or | |
4984 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
4985 | * Hence, default to zone ordering on 32-bit. | |
4986 | */ | |
f0c0b2b8 KH |
4987 | static int default_zonelist_order(void) |
4988 | { | |
f0c0b2b8 KH |
4989 | return ZONELIST_ORDER_ZONE; |
4990 | } | |
3193913c | 4991 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
4992 | |
4993 | static void set_zonelist_order(void) | |
4994 | { | |
4995 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
4996 | current_zonelist_order = default_zonelist_order(); | |
4997 | else | |
4998 | current_zonelist_order = user_zonelist_order; | |
4999 | } | |
5000 | ||
5001 | static void build_zonelists(pg_data_t *pgdat) | |
5002 | { | |
c00eb15a | 5003 | int i, node, load; |
1da177e4 | 5004 | nodemask_t used_mask; |
f0c0b2b8 KH |
5005 | int local_node, prev_node; |
5006 | struct zonelist *zonelist; | |
d00181b9 | 5007 | unsigned int order = current_zonelist_order; |
1da177e4 LT |
5008 | |
5009 | /* initialize zonelists */ | |
523b9458 | 5010 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 5011 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
5012 | zonelist->_zonerefs[0].zone = NULL; |
5013 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
5014 | } |
5015 | ||
5016 | /* NUMA-aware ordering of nodes */ | |
5017 | local_node = pgdat->node_id; | |
62bc62a8 | 5018 | load = nr_online_nodes; |
1da177e4 LT |
5019 | prev_node = local_node; |
5020 | nodes_clear(used_mask); | |
f0c0b2b8 | 5021 | |
f0c0b2b8 | 5022 | memset(node_order, 0, sizeof(node_order)); |
c00eb15a | 5023 | i = 0; |
f0c0b2b8 | 5024 | |
1da177e4 LT |
5025 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5026 | /* | |
5027 | * We don't want to pressure a particular node. | |
5028 | * So adding penalty to the first node in same | |
5029 | * distance group to make it round-robin. | |
5030 | */ | |
957f822a DR |
5031 | if (node_distance(local_node, node) != |
5032 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5033 | node_load[node] = load; |
5034 | ||
1da177e4 LT |
5035 | prev_node = node; |
5036 | load--; | |
f0c0b2b8 KH |
5037 | if (order == ZONELIST_ORDER_NODE) |
5038 | build_zonelists_in_node_order(pgdat, node); | |
5039 | else | |
c00eb15a | 5040 | node_order[i++] = node; /* remember order */ |
f0c0b2b8 | 5041 | } |
1da177e4 | 5042 | |
f0c0b2b8 KH |
5043 | if (order == ZONELIST_ORDER_ZONE) { |
5044 | /* calculate node order -- i.e., DMA last! */ | |
c00eb15a | 5045 | build_zonelists_in_zone_order(pgdat, i); |
1da177e4 | 5046 | } |
523b9458 CL |
5047 | |
5048 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
5049 | } |
5050 | ||
7aac7898 LS |
5051 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5052 | /* | |
5053 | * Return node id of node used for "local" allocations. | |
5054 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5055 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5056 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5057 | */ | |
5058 | int local_memory_node(int node) | |
5059 | { | |
c33d6c06 | 5060 | struct zoneref *z; |
7aac7898 | 5061 | |
c33d6c06 | 5062 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5063 | gfp_zone(GFP_KERNEL), |
c33d6c06 MG |
5064 | NULL); |
5065 | return z->zone->node; | |
7aac7898 LS |
5066 | } |
5067 | #endif | |
f0c0b2b8 | 5068 | |
6423aa81 JK |
5069 | static void setup_min_unmapped_ratio(void); |
5070 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5071 | #else /* CONFIG_NUMA */ |
5072 | ||
f0c0b2b8 KH |
5073 | static void set_zonelist_order(void) |
5074 | { | |
5075 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
5076 | } | |
5077 | ||
5078 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 5079 | { |
19655d34 | 5080 | int node, local_node; |
54a6eb5c MG |
5081 | enum zone_type j; |
5082 | struct zonelist *zonelist; | |
1da177e4 LT |
5083 | |
5084 | local_node = pgdat->node_id; | |
1da177e4 | 5085 | |
c9634cf0 | 5086 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
bc732f1d | 5087 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 5088 | |
54a6eb5c MG |
5089 | /* |
5090 | * Now we build the zonelist so that it contains the zones | |
5091 | * of all the other nodes. | |
5092 | * We don't want to pressure a particular node, so when | |
5093 | * building the zones for node N, we make sure that the | |
5094 | * zones coming right after the local ones are those from | |
5095 | * node N+1 (modulo N) | |
5096 | */ | |
5097 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5098 | if (!node_online(node)) | |
5099 | continue; | |
bc732f1d | 5100 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 5101 | } |
54a6eb5c MG |
5102 | for (node = 0; node < local_node; node++) { |
5103 | if (!node_online(node)) | |
5104 | continue; | |
bc732f1d | 5105 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
5106 | } |
5107 | ||
dd1a239f MG |
5108 | zonelist->_zonerefs[j].zone = NULL; |
5109 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
5110 | } |
5111 | ||
5112 | #endif /* CONFIG_NUMA */ | |
5113 | ||
99dcc3e5 CL |
5114 | /* |
5115 | * Boot pageset table. One per cpu which is going to be used for all | |
5116 | * zones and all nodes. The parameters will be set in such a way | |
5117 | * that an item put on a list will immediately be handed over to | |
5118 | * the buddy list. This is safe since pageset manipulation is done | |
5119 | * with interrupts disabled. | |
5120 | * | |
5121 | * The boot_pagesets must be kept even after bootup is complete for | |
5122 | * unused processors and/or zones. They do play a role for bootstrapping | |
5123 | * hotplugged processors. | |
5124 | * | |
5125 | * zoneinfo_show() and maybe other functions do | |
5126 | * not check if the processor is online before following the pageset pointer. | |
5127 | * Other parts of the kernel may not check if the zone is available. | |
5128 | */ | |
5129 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5130 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 5131 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 5132 | |
4eaf3f64 HL |
5133 | /* |
5134 | * Global mutex to protect against size modification of zonelists | |
5135 | * as well as to serialize pageset setup for the new populated zone. | |
5136 | */ | |
5137 | DEFINE_MUTEX(zonelists_mutex); | |
5138 | ||
9b1a4d38 | 5139 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 5140 | static int __build_all_zonelists(void *data) |
1da177e4 | 5141 | { |
6811378e | 5142 | int nid; |
99dcc3e5 | 5143 | int cpu; |
9adb62a5 | 5144 | pg_data_t *self = data; |
9276b1bc | 5145 | |
7f9cfb31 BL |
5146 | #ifdef CONFIG_NUMA |
5147 | memset(node_load, 0, sizeof(node_load)); | |
5148 | #endif | |
9adb62a5 JL |
5149 | |
5150 | if (self && !node_online(self->node_id)) { | |
5151 | build_zonelists(self); | |
9adb62a5 JL |
5152 | } |
5153 | ||
9276b1bc | 5154 | for_each_online_node(nid) { |
7ea1530a CL |
5155 | pg_data_t *pgdat = NODE_DATA(nid); |
5156 | ||
5157 | build_zonelists(pgdat); | |
9276b1bc | 5158 | } |
99dcc3e5 CL |
5159 | |
5160 | /* | |
5161 | * Initialize the boot_pagesets that are going to be used | |
5162 | * for bootstrapping processors. The real pagesets for | |
5163 | * each zone will be allocated later when the per cpu | |
5164 | * allocator is available. | |
5165 | * | |
5166 | * boot_pagesets are used also for bootstrapping offline | |
5167 | * cpus if the system is already booted because the pagesets | |
5168 | * are needed to initialize allocators on a specific cpu too. | |
5169 | * F.e. the percpu allocator needs the page allocator which | |
5170 | * needs the percpu allocator in order to allocate its pagesets | |
5171 | * (a chicken-egg dilemma). | |
5172 | */ | |
7aac7898 | 5173 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
5174 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
5175 | ||
7aac7898 LS |
5176 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5177 | /* | |
5178 | * We now know the "local memory node" for each node-- | |
5179 | * i.e., the node of the first zone in the generic zonelist. | |
5180 | * Set up numa_mem percpu variable for on-line cpus. During | |
5181 | * boot, only the boot cpu should be on-line; we'll init the | |
5182 | * secondary cpus' numa_mem as they come on-line. During | |
5183 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5184 | */ | |
5185 | if (cpu_online(cpu)) | |
5186 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
5187 | #endif | |
5188 | } | |
5189 | ||
6811378e YG |
5190 | return 0; |
5191 | } | |
5192 | ||
061f67bc RV |
5193 | static noinline void __init |
5194 | build_all_zonelists_init(void) | |
5195 | { | |
5196 | __build_all_zonelists(NULL); | |
5197 | mminit_verify_zonelist(); | |
5198 | cpuset_init_current_mems_allowed(); | |
5199 | } | |
5200 | ||
4eaf3f64 HL |
5201 | /* |
5202 | * Called with zonelists_mutex held always | |
5203 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
5204 | * |
5205 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
5206 | * [we're only called with non-NULL zone through __meminit paths] and | |
5207 | * (2) call of __init annotated helper build_all_zonelists_init | |
5208 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 5209 | */ |
9adb62a5 | 5210 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 5211 | { |
f0c0b2b8 KH |
5212 | set_zonelist_order(); |
5213 | ||
6811378e | 5214 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 5215 | build_all_zonelists_init(); |
6811378e | 5216 | } else { |
e9959f0f | 5217 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
5218 | if (zone) |
5219 | setup_zone_pageset(zone); | |
e9959f0f | 5220 | #endif |
dd1895e2 CS |
5221 | /* we have to stop all cpus to guarantee there is no user |
5222 | of zonelist */ | |
9adb62a5 | 5223 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
5224 | /* cpuset refresh routine should be here */ |
5225 | } | |
bd1e22b8 | 5226 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5227 | /* |
5228 | * Disable grouping by mobility if the number of pages in the | |
5229 | * system is too low to allow the mechanism to work. It would be | |
5230 | * more accurate, but expensive to check per-zone. This check is | |
5231 | * made on memory-hotadd so a system can start with mobility | |
5232 | * disabled and enable it later | |
5233 | */ | |
d9c23400 | 5234 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5235 | page_group_by_mobility_disabled = 1; |
5236 | else | |
5237 | page_group_by_mobility_disabled = 0; | |
5238 | ||
756a025f JP |
5239 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
5240 | nr_online_nodes, | |
5241 | zonelist_order_name[current_zonelist_order], | |
5242 | page_group_by_mobility_disabled ? "off" : "on", | |
5243 | vm_total_pages); | |
f0c0b2b8 | 5244 | #ifdef CONFIG_NUMA |
f88dfff5 | 5245 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5246 | #endif |
1da177e4 LT |
5247 | } |
5248 | ||
1da177e4 LT |
5249 | /* |
5250 | * Initially all pages are reserved - free ones are freed | |
5251 | * up by free_all_bootmem() once the early boot process is | |
5252 | * done. Non-atomic initialization, single-pass. | |
5253 | */ | |
c09b4240 | 5254 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 5255 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 5256 | { |
4b94ffdc | 5257 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
29751f69 | 5258 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 5259 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 5260 | unsigned long pfn; |
3a80a7fa | 5261 | unsigned long nr_initialised = 0; |
342332e6 TI |
5262 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5263 | struct memblock_region *r = NULL, *tmp; | |
5264 | #endif | |
1da177e4 | 5265 | |
22b31eec HD |
5266 | if (highest_memmap_pfn < end_pfn - 1) |
5267 | highest_memmap_pfn = end_pfn - 1; | |
5268 | ||
4b94ffdc DW |
5269 | /* |
5270 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
5271 | * memory | |
5272 | */ | |
5273 | if (altmap && start_pfn == altmap->base_pfn) | |
5274 | start_pfn += altmap->reserve; | |
5275 | ||
cbe8dd4a | 5276 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 5277 | /* |
b72d0ffb AM |
5278 | * There can be holes in boot-time mem_map[]s handed to this |
5279 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5280 | */ |
b72d0ffb AM |
5281 | if (context != MEMMAP_EARLY) |
5282 | goto not_early; | |
5283 | ||
b92df1de PB |
5284 | if (!early_pfn_valid(pfn)) { |
5285 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
5286 | /* | |
5287 | * Skip to the pfn preceding the next valid one (or | |
5288 | * end_pfn), such that we hit a valid pfn (or end_pfn) | |
5289 | * on our next iteration of the loop. | |
5290 | */ | |
5291 | pfn = memblock_next_valid_pfn(pfn, end_pfn) - 1; | |
5292 | #endif | |
b72d0ffb | 5293 | continue; |
b92df1de | 5294 | } |
b72d0ffb AM |
5295 | if (!early_pfn_in_nid(pfn, nid)) |
5296 | continue; | |
5297 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
5298 | break; | |
342332e6 TI |
5299 | |
5300 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
5301 | /* |
5302 | * Check given memblock attribute by firmware which can affect | |
5303 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
5304 | * mirrored, it's an overlapped memmap init. skip it. | |
5305 | */ | |
5306 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5307 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
5308 | for_each_memblock(memory, tmp) | |
5309 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
5310 | break; | |
5311 | r = tmp; | |
5312 | } | |
5313 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
5314 | memblock_is_mirror(r)) { | |
5315 | /* already initialized as NORMAL */ | |
5316 | pfn = memblock_region_memory_end_pfn(r); | |
5317 | continue; | |
342332e6 | 5318 | } |
a2f3aa02 | 5319 | } |
b72d0ffb | 5320 | #endif |
ac5d2539 | 5321 | |
b72d0ffb | 5322 | not_early: |
ac5d2539 MG |
5323 | /* |
5324 | * Mark the block movable so that blocks are reserved for | |
5325 | * movable at startup. This will force kernel allocations | |
5326 | * to reserve their blocks rather than leaking throughout | |
5327 | * the address space during boot when many long-lived | |
974a786e | 5328 | * kernel allocations are made. |
ac5d2539 MG |
5329 | * |
5330 | * bitmap is created for zone's valid pfn range. but memmap | |
5331 | * can be created for invalid pages (for alignment) | |
5332 | * check here not to call set_pageblock_migratetype() against | |
5333 | * pfn out of zone. | |
5334 | */ | |
5335 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
5336 | struct page *page = pfn_to_page(pfn); | |
5337 | ||
5338 | __init_single_page(page, pfn, zone, nid); | |
5339 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5340 | } else { | |
5341 | __init_single_pfn(pfn, zone, nid); | |
5342 | } | |
1da177e4 LT |
5343 | } |
5344 | } | |
5345 | ||
1e548deb | 5346 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 5347 | { |
7aeb09f9 | 5348 | unsigned int order, t; |
b2a0ac88 MG |
5349 | for_each_migratetype_order(order, t) { |
5350 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
5351 | zone->free_area[order].nr_free = 0; |
5352 | } | |
5353 | } | |
5354 | ||
5355 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
5356 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 5357 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
5358 | #endif |
5359 | ||
7cd2b0a3 | 5360 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 5361 | { |
3a6be87f | 5362 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
5363 | int batch; |
5364 | ||
5365 | /* | |
5366 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 5367 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
5368 | * |
5369 | * OK, so we don't know how big the cache is. So guess. | |
5370 | */ | |
b40da049 | 5371 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
5372 | if (batch * PAGE_SIZE > 512 * 1024) |
5373 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
5374 | batch /= 4; /* We effectively *= 4 below */ |
5375 | if (batch < 1) | |
5376 | batch = 1; | |
5377 | ||
5378 | /* | |
0ceaacc9 NP |
5379 | * Clamp the batch to a 2^n - 1 value. Having a power |
5380 | * of 2 value was found to be more likely to have | |
5381 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 5382 | * |
0ceaacc9 NP |
5383 | * For example if 2 tasks are alternately allocating |
5384 | * batches of pages, one task can end up with a lot | |
5385 | * of pages of one half of the possible page colors | |
5386 | * and the other with pages of the other colors. | |
e7c8d5c9 | 5387 | */ |
9155203a | 5388 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 5389 | |
e7c8d5c9 | 5390 | return batch; |
3a6be87f DH |
5391 | |
5392 | #else | |
5393 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5394 | * conditions. | |
5395 | * | |
5396 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5397 | * of contiguous memory as there's no hardware page translation to | |
5398 | * assemble apparent contiguous memory from discontiguous pages. | |
5399 | * | |
5400 | * Queueing large contiguous runs of pages for batching, however, | |
5401 | * causes the pages to actually be freed in smaller chunks. As there | |
5402 | * can be a significant delay between the individual batches being | |
5403 | * recycled, this leads to the once large chunks of space being | |
5404 | * fragmented and becoming unavailable for high-order allocations. | |
5405 | */ | |
5406 | return 0; | |
5407 | #endif | |
e7c8d5c9 CL |
5408 | } |
5409 | ||
8d7a8fa9 CS |
5410 | /* |
5411 | * pcp->high and pcp->batch values are related and dependent on one another: | |
5412 | * ->batch must never be higher then ->high. | |
5413 | * The following function updates them in a safe manner without read side | |
5414 | * locking. | |
5415 | * | |
5416 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
5417 | * those fields changing asynchronously (acording the the above rule). | |
5418 | * | |
5419 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5420 | * outside of boot time (or some other assurance that no concurrent updaters | |
5421 | * exist). | |
5422 | */ | |
5423 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
5424 | unsigned long batch) | |
5425 | { | |
5426 | /* start with a fail safe value for batch */ | |
5427 | pcp->batch = 1; | |
5428 | smp_wmb(); | |
5429 | ||
5430 | /* Update high, then batch, in order */ | |
5431 | pcp->high = high; | |
5432 | smp_wmb(); | |
5433 | ||
5434 | pcp->batch = batch; | |
5435 | } | |
5436 | ||
3664033c | 5437 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
5438 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5439 | { | |
8d7a8fa9 | 5440 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
5441 | } |
5442 | ||
88c90dbc | 5443 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
5444 | { |
5445 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 5446 | int migratetype; |
2caaad41 | 5447 | |
1c6fe946 MD |
5448 | memset(p, 0, sizeof(*p)); |
5449 | ||
3dfa5721 | 5450 | pcp = &p->pcp; |
2caaad41 | 5451 | pcp->count = 0; |
5f8dcc21 MG |
5452 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5453 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
5454 | } |
5455 | ||
88c90dbc CS |
5456 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5457 | { | |
5458 | pageset_init(p); | |
5459 | pageset_set_batch(p, batch); | |
5460 | } | |
5461 | ||
8ad4b1fb | 5462 | /* |
3664033c | 5463 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
5464 | * to the value high for the pageset p. |
5465 | */ | |
3664033c | 5466 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
5467 | unsigned long high) |
5468 | { | |
8d7a8fa9 CS |
5469 | unsigned long batch = max(1UL, high / 4); |
5470 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
5471 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 5472 | |
8d7a8fa9 | 5473 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
5474 | } |
5475 | ||
7cd2b0a3 DR |
5476 | static void pageset_set_high_and_batch(struct zone *zone, |
5477 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 5478 | { |
56cef2b8 | 5479 | if (percpu_pagelist_fraction) |
3664033c | 5480 | pageset_set_high(pcp, |
56cef2b8 CS |
5481 | (zone->managed_pages / |
5482 | percpu_pagelist_fraction)); | |
5483 | else | |
5484 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
5485 | } | |
5486 | ||
169f6c19 CS |
5487 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5488 | { | |
5489 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
5490 | ||
5491 | pageset_init(pcp); | |
5492 | pageset_set_high_and_batch(zone, pcp); | |
5493 | } | |
5494 | ||
4ed7e022 | 5495 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
5496 | { |
5497 | int cpu; | |
319774e2 | 5498 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5499 | for_each_possible_cpu(cpu) |
5500 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5501 | } |
5502 | ||
2caaad41 | 5503 | /* |
99dcc3e5 CL |
5504 | * Allocate per cpu pagesets and initialize them. |
5505 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5506 | */ |
99dcc3e5 | 5507 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5508 | { |
b4911ea2 | 5509 | struct pglist_data *pgdat; |
99dcc3e5 | 5510 | struct zone *zone; |
e7c8d5c9 | 5511 | |
319774e2 WF |
5512 | for_each_populated_zone(zone) |
5513 | setup_zone_pageset(zone); | |
b4911ea2 MG |
5514 | |
5515 | for_each_online_pgdat(pgdat) | |
5516 | pgdat->per_cpu_nodestats = | |
5517 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
5518 | } |
5519 | ||
c09b4240 | 5520 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5521 | { |
99dcc3e5 CL |
5522 | /* |
5523 | * per cpu subsystem is not up at this point. The following code | |
5524 | * relies on the ability of the linker to provide the | |
5525 | * offset of a (static) per cpu variable into the per cpu area. | |
5526 | */ | |
5527 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5528 | |
b38a8725 | 5529 | if (populated_zone(zone)) |
99dcc3e5 CL |
5530 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5531 | zone->name, zone->present_pages, | |
5532 | zone_batchsize(zone)); | |
ed8ece2e DH |
5533 | } |
5534 | ||
dc0bbf3b | 5535 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5536 | unsigned long zone_start_pfn, |
b171e409 | 5537 | unsigned long size) |
ed8ece2e DH |
5538 | { |
5539 | struct pglist_data *pgdat = zone->zone_pgdat; | |
9dcb8b68 | 5540 | |
ed8ece2e DH |
5541 | pgdat->nr_zones = zone_idx(zone) + 1; |
5542 | ||
ed8ece2e DH |
5543 | zone->zone_start_pfn = zone_start_pfn; |
5544 | ||
708614e6 MG |
5545 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5546 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5547 | pgdat->node_id, | |
5548 | (unsigned long)zone_idx(zone), | |
5549 | zone_start_pfn, (zone_start_pfn + size)); | |
5550 | ||
1e548deb | 5551 | zone_init_free_lists(zone); |
9dcb8b68 | 5552 | zone->initialized = 1; |
ed8ece2e DH |
5553 | } |
5554 | ||
0ee332c1 | 5555 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5556 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5557 | |
c713216d MG |
5558 | /* |
5559 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5560 | */ |
8a942fde MG |
5561 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5562 | struct mminit_pfnnid_cache *state) | |
c713216d | 5563 | { |
c13291a5 | 5564 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5565 | int nid; |
7c243c71 | 5566 | |
8a942fde MG |
5567 | if (state->last_start <= pfn && pfn < state->last_end) |
5568 | return state->last_nid; | |
c713216d | 5569 | |
e76b63f8 YL |
5570 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5571 | if (nid != -1) { | |
8a942fde MG |
5572 | state->last_start = start_pfn; |
5573 | state->last_end = end_pfn; | |
5574 | state->last_nid = nid; | |
e76b63f8 YL |
5575 | } |
5576 | ||
5577 | return nid; | |
c713216d MG |
5578 | } |
5579 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5580 | ||
c713216d | 5581 | /** |
6782832e | 5582 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5583 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5584 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5585 | * |
7d018176 ZZ |
5586 | * If an architecture guarantees that all ranges registered contain no holes |
5587 | * and may be freed, this this function may be used instead of calling | |
5588 | * memblock_free_early_nid() manually. | |
c713216d | 5589 | */ |
c13291a5 | 5590 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5591 | { |
c13291a5 TH |
5592 | unsigned long start_pfn, end_pfn; |
5593 | int i, this_nid; | |
edbe7d23 | 5594 | |
c13291a5 TH |
5595 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5596 | start_pfn = min(start_pfn, max_low_pfn); | |
5597 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5598 | |
c13291a5 | 5599 | if (start_pfn < end_pfn) |
6782832e SS |
5600 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5601 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5602 | this_nid); | |
edbe7d23 | 5603 | } |
edbe7d23 | 5604 | } |
edbe7d23 | 5605 | |
c713216d MG |
5606 | /** |
5607 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5608 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5609 | * |
7d018176 ZZ |
5610 | * If an architecture guarantees that all ranges registered contain no holes and may |
5611 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5612 | */ |
5613 | void __init sparse_memory_present_with_active_regions(int nid) | |
5614 | { | |
c13291a5 TH |
5615 | unsigned long start_pfn, end_pfn; |
5616 | int i, this_nid; | |
c713216d | 5617 | |
c13291a5 TH |
5618 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5619 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5620 | } |
5621 | ||
5622 | /** | |
5623 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5624 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5625 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5626 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5627 | * |
5628 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5629 | * provided by memblock_set_node(). If called for a node |
c713216d | 5630 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5631 | * PFNs will be 0. |
c713216d | 5632 | */ |
a3142c8e | 5633 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5634 | unsigned long *start_pfn, unsigned long *end_pfn) |
5635 | { | |
c13291a5 | 5636 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5637 | int i; |
c13291a5 | 5638 | |
c713216d MG |
5639 | *start_pfn = -1UL; |
5640 | *end_pfn = 0; | |
5641 | ||
c13291a5 TH |
5642 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5643 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5644 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5645 | } |
5646 | ||
633c0666 | 5647 | if (*start_pfn == -1UL) |
c713216d | 5648 | *start_pfn = 0; |
c713216d MG |
5649 | } |
5650 | ||
2a1e274a MG |
5651 | /* |
5652 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5653 | * assumption is made that zones within a node are ordered in monotonic | |
5654 | * increasing memory addresses so that the "highest" populated zone is used | |
5655 | */ | |
b69a7288 | 5656 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5657 | { |
5658 | int zone_index; | |
5659 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5660 | if (zone_index == ZONE_MOVABLE) | |
5661 | continue; | |
5662 | ||
5663 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5664 | arch_zone_lowest_possible_pfn[zone_index]) | |
5665 | break; | |
5666 | } | |
5667 | ||
5668 | VM_BUG_ON(zone_index == -1); | |
5669 | movable_zone = zone_index; | |
5670 | } | |
5671 | ||
5672 | /* | |
5673 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5674 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5675 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5676 | * in each node depending on the size of each node and how evenly kernelcore | |
5677 | * is distributed. This helper function adjusts the zone ranges | |
5678 | * provided by the architecture for a given node by using the end of the | |
5679 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5680 | * zones within a node are in order of monotonic increases memory addresses | |
5681 | */ | |
b69a7288 | 5682 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5683 | unsigned long zone_type, |
5684 | unsigned long node_start_pfn, | |
5685 | unsigned long node_end_pfn, | |
5686 | unsigned long *zone_start_pfn, | |
5687 | unsigned long *zone_end_pfn) | |
5688 | { | |
5689 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5690 | if (zone_movable_pfn[nid]) { | |
5691 | /* Size ZONE_MOVABLE */ | |
5692 | if (zone_type == ZONE_MOVABLE) { | |
5693 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5694 | *zone_end_pfn = min(node_end_pfn, | |
5695 | arch_zone_highest_possible_pfn[movable_zone]); | |
5696 | ||
e506b996 XQ |
5697 | /* Adjust for ZONE_MOVABLE starting within this range */ |
5698 | } else if (!mirrored_kernelcore && | |
5699 | *zone_start_pfn < zone_movable_pfn[nid] && | |
5700 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
5701 | *zone_end_pfn = zone_movable_pfn[nid]; | |
5702 | ||
2a1e274a MG |
5703 | /* Check if this whole range is within ZONE_MOVABLE */ |
5704 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5705 | *zone_start_pfn = *zone_end_pfn; | |
5706 | } | |
5707 | } | |
5708 | ||
c713216d MG |
5709 | /* |
5710 | * Return the number of pages a zone spans in a node, including holes | |
5711 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5712 | */ | |
6ea6e688 | 5713 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5714 | unsigned long zone_type, |
7960aedd ZY |
5715 | unsigned long node_start_pfn, |
5716 | unsigned long node_end_pfn, | |
d91749c1 TI |
5717 | unsigned long *zone_start_pfn, |
5718 | unsigned long *zone_end_pfn, | |
c713216d MG |
5719 | unsigned long *ignored) |
5720 | { | |
b5685e92 | 5721 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5722 | if (!node_start_pfn && !node_end_pfn) |
5723 | return 0; | |
5724 | ||
7960aedd | 5725 | /* Get the start and end of the zone */ |
d91749c1 TI |
5726 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5727 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5728 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5729 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5730 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5731 | |
5732 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5733 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5734 | return 0; |
5735 | ||
5736 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5737 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5738 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5739 | |
5740 | /* Return the spanned pages */ | |
d91749c1 | 5741 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5742 | } |
5743 | ||
5744 | /* | |
5745 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5746 | * then all holes in the requested range will be accounted for. |
c713216d | 5747 | */ |
32996250 | 5748 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5749 | unsigned long range_start_pfn, |
5750 | unsigned long range_end_pfn) | |
5751 | { | |
96e907d1 TH |
5752 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5753 | unsigned long start_pfn, end_pfn; | |
5754 | int i; | |
c713216d | 5755 | |
96e907d1 TH |
5756 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5757 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5758 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5759 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5760 | } |
96e907d1 | 5761 | return nr_absent; |
c713216d MG |
5762 | } |
5763 | ||
5764 | /** | |
5765 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5766 | * @start_pfn: The start PFN to start searching for holes | |
5767 | * @end_pfn: The end PFN to stop searching for holes | |
5768 | * | |
88ca3b94 | 5769 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5770 | */ |
5771 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5772 | unsigned long end_pfn) | |
5773 | { | |
5774 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5775 | } | |
5776 | ||
5777 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5778 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5779 | unsigned long zone_type, |
7960aedd ZY |
5780 | unsigned long node_start_pfn, |
5781 | unsigned long node_end_pfn, | |
c713216d MG |
5782 | unsigned long *ignored) |
5783 | { | |
96e907d1 TH |
5784 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5785 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5786 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5787 | unsigned long nr_absent; |
9c7cd687 | 5788 | |
b5685e92 | 5789 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5790 | if (!node_start_pfn && !node_end_pfn) |
5791 | return 0; | |
5792 | ||
96e907d1 TH |
5793 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5794 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5795 | |
2a1e274a MG |
5796 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5797 | node_start_pfn, node_end_pfn, | |
5798 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5799 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5800 | ||
5801 | /* | |
5802 | * ZONE_MOVABLE handling. | |
5803 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
5804 | * and vice versa. | |
5805 | */ | |
e506b996 XQ |
5806 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
5807 | unsigned long start_pfn, end_pfn; | |
5808 | struct memblock_region *r; | |
5809 | ||
5810 | for_each_memblock(memory, r) { | |
5811 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
5812 | zone_start_pfn, zone_end_pfn); | |
5813 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
5814 | zone_start_pfn, zone_end_pfn); | |
5815 | ||
5816 | if (zone_type == ZONE_MOVABLE && | |
5817 | memblock_is_mirror(r)) | |
5818 | nr_absent += end_pfn - start_pfn; | |
5819 | ||
5820 | if (zone_type == ZONE_NORMAL && | |
5821 | !memblock_is_mirror(r)) | |
5822 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
5823 | } |
5824 | } | |
5825 | ||
5826 | return nr_absent; | |
c713216d | 5827 | } |
0e0b864e | 5828 | |
0ee332c1 | 5829 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 5830 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5831 | unsigned long zone_type, |
7960aedd ZY |
5832 | unsigned long node_start_pfn, |
5833 | unsigned long node_end_pfn, | |
d91749c1 TI |
5834 | unsigned long *zone_start_pfn, |
5835 | unsigned long *zone_end_pfn, | |
c713216d MG |
5836 | unsigned long *zones_size) |
5837 | { | |
d91749c1 TI |
5838 | unsigned int zone; |
5839 | ||
5840 | *zone_start_pfn = node_start_pfn; | |
5841 | for (zone = 0; zone < zone_type; zone++) | |
5842 | *zone_start_pfn += zones_size[zone]; | |
5843 | ||
5844 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
5845 | ||
c713216d MG |
5846 | return zones_size[zone_type]; |
5847 | } | |
5848 | ||
6ea6e688 | 5849 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5850 | unsigned long zone_type, |
7960aedd ZY |
5851 | unsigned long node_start_pfn, |
5852 | unsigned long node_end_pfn, | |
c713216d MG |
5853 | unsigned long *zholes_size) |
5854 | { | |
5855 | if (!zholes_size) | |
5856 | return 0; | |
5857 | ||
5858 | return zholes_size[zone_type]; | |
5859 | } | |
20e6926d | 5860 | |
0ee332c1 | 5861 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5862 | |
a3142c8e | 5863 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
5864 | unsigned long node_start_pfn, |
5865 | unsigned long node_end_pfn, | |
5866 | unsigned long *zones_size, | |
5867 | unsigned long *zholes_size) | |
c713216d | 5868 | { |
febd5949 | 5869 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
5870 | enum zone_type i; |
5871 | ||
febd5949 GZ |
5872 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5873 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 5874 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 5875 | unsigned long size, real_size; |
c713216d | 5876 | |
febd5949 GZ |
5877 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5878 | node_start_pfn, | |
5879 | node_end_pfn, | |
d91749c1 TI |
5880 | &zone_start_pfn, |
5881 | &zone_end_pfn, | |
febd5949 GZ |
5882 | zones_size); |
5883 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
5884 | node_start_pfn, node_end_pfn, |
5885 | zholes_size); | |
d91749c1 TI |
5886 | if (size) |
5887 | zone->zone_start_pfn = zone_start_pfn; | |
5888 | else | |
5889 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
5890 | zone->spanned_pages = size; |
5891 | zone->present_pages = real_size; | |
5892 | ||
5893 | totalpages += size; | |
5894 | realtotalpages += real_size; | |
5895 | } | |
5896 | ||
5897 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
5898 | pgdat->node_present_pages = realtotalpages; |
5899 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
5900 | realtotalpages); | |
5901 | } | |
5902 | ||
835c134e MG |
5903 | #ifndef CONFIG_SPARSEMEM |
5904 | /* | |
5905 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
5906 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5907 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
5908 | * round what is now in bits to nearest long in bits, then return it in |
5909 | * bytes. | |
5910 | */ | |
7c45512d | 5911 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
5912 | { |
5913 | unsigned long usemapsize; | |
5914 | ||
7c45512d | 5915 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
5916 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
5917 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
5918 | usemapsize *= NR_PAGEBLOCK_BITS; |
5919 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
5920 | ||
5921 | return usemapsize / 8; | |
5922 | } | |
5923 | ||
5924 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
5925 | struct zone *zone, |
5926 | unsigned long zone_start_pfn, | |
5927 | unsigned long zonesize) | |
835c134e | 5928 | { |
7c45512d | 5929 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 5930 | zone->pageblock_flags = NULL; |
58a01a45 | 5931 | if (usemapsize) |
6782832e SS |
5932 | zone->pageblock_flags = |
5933 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5934 | pgdat->node_id); | |
835c134e MG |
5935 | } |
5936 | #else | |
7c45512d LT |
5937 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5938 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5939 | #endif /* CONFIG_SPARSEMEM */ |
5940 | ||
d9c23400 | 5941 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5942 | |
d9c23400 | 5943 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5944 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5945 | { |
955c1cd7 AM |
5946 | unsigned int order; |
5947 | ||
d9c23400 MG |
5948 | /* Check that pageblock_nr_pages has not already been setup */ |
5949 | if (pageblock_order) | |
5950 | return; | |
5951 | ||
955c1cd7 AM |
5952 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5953 | order = HUGETLB_PAGE_ORDER; | |
5954 | else | |
5955 | order = MAX_ORDER - 1; | |
5956 | ||
d9c23400 MG |
5957 | /* |
5958 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5959 | * This value may be variable depending on boot parameters on IA64 and |
5960 | * powerpc. | |
d9c23400 MG |
5961 | */ |
5962 | pageblock_order = order; | |
5963 | } | |
5964 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5965 | ||
ba72cb8c MG |
5966 | /* |
5967 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5968 | * is unused as pageblock_order is set at compile-time. See |
5969 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5970 | * the kernel config | |
ba72cb8c | 5971 | */ |
15ca220e | 5972 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5973 | { |
ba72cb8c | 5974 | } |
d9c23400 MG |
5975 | |
5976 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5977 | ||
01cefaef JL |
5978 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5979 | unsigned long present_pages) | |
5980 | { | |
5981 | unsigned long pages = spanned_pages; | |
5982 | ||
5983 | /* | |
5984 | * Provide a more accurate estimation if there are holes within | |
5985 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5986 | * zone, each populated memory region may cost us one or two extra | |
5987 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 5988 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
5989 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
5990 | */ | |
5991 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5992 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5993 | pages = present_pages; | |
5994 | ||
5995 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5996 | } | |
5997 | ||
1da177e4 LT |
5998 | /* |
5999 | * Set up the zone data structures: | |
6000 | * - mark all pages reserved | |
6001 | * - mark all memory queues empty | |
6002 | * - clear the memory bitmaps | |
6527af5d MK |
6003 | * |
6004 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 6005 | */ |
7f3eb55b | 6006 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 6007 | { |
2f1b6248 | 6008 | enum zone_type j; |
ed8ece2e | 6009 | int nid = pgdat->node_id; |
1da177e4 | 6010 | |
208d54e5 | 6011 | pgdat_resize_init(pgdat); |
8177a420 AA |
6012 | #ifdef CONFIG_NUMA_BALANCING |
6013 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
6014 | pgdat->numabalancing_migrate_nr_pages = 0; | |
6015 | pgdat->numabalancing_migrate_next_window = jiffies; | |
a3d0a918 KS |
6016 | #endif |
6017 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6018 | spin_lock_init(&pgdat->split_queue_lock); | |
6019 | INIT_LIST_HEAD(&pgdat->split_queue); | |
6020 | pgdat->split_queue_len = 0; | |
8177a420 | 6021 | #endif |
1da177e4 | 6022 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6023 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
698b1b30 VB |
6024 | #ifdef CONFIG_COMPACTION |
6025 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6026 | #endif | |
eefa864b | 6027 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6028 | spin_lock_init(&pgdat->lru_lock); |
a9dd0a83 | 6029 | lruvec_init(node_lruvec(pgdat)); |
5f63b720 | 6030 | |
1da177e4 LT |
6031 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6032 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 6033 | unsigned long size, realsize, freesize, memmap_pages; |
d91749c1 | 6034 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6035 | |
febd5949 GZ |
6036 | size = zone->spanned_pages; |
6037 | realsize = freesize = zone->present_pages; | |
1da177e4 | 6038 | |
0e0b864e | 6039 | /* |
9feedc9d | 6040 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6041 | * is used by this zone for memmap. This affects the watermark |
6042 | * and per-cpu initialisations | |
6043 | */ | |
01cefaef | 6044 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
6045 | if (!is_highmem_idx(j)) { |
6046 | if (freesize >= memmap_pages) { | |
6047 | freesize -= memmap_pages; | |
6048 | if (memmap_pages) | |
6049 | printk(KERN_DEBUG | |
6050 | " %s zone: %lu pages used for memmap\n", | |
6051 | zone_names[j], memmap_pages); | |
6052 | } else | |
1170532b | 6053 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6054 | zone_names[j], memmap_pages, freesize); |
6055 | } | |
0e0b864e | 6056 | |
6267276f | 6057 | /* Account for reserved pages */ |
9feedc9d JL |
6058 | if (j == 0 && freesize > dma_reserve) { |
6059 | freesize -= dma_reserve; | |
d903ef9f | 6060 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6061 | zone_names[0], dma_reserve); |
0e0b864e MG |
6062 | } |
6063 | ||
98d2b0eb | 6064 | if (!is_highmem_idx(j)) |
9feedc9d | 6065 | nr_kernel_pages += freesize; |
01cefaef JL |
6066 | /* Charge for highmem memmap if there are enough kernel pages */ |
6067 | else if (nr_kernel_pages > memmap_pages * 2) | |
6068 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6069 | nr_all_pages += freesize; |
1da177e4 | 6070 | |
9feedc9d JL |
6071 | /* |
6072 | * Set an approximate value for lowmem here, it will be adjusted | |
6073 | * when the bootmem allocator frees pages into the buddy system. | |
6074 | * And all highmem pages will be managed by the buddy system. | |
6075 | */ | |
6076 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 6077 | #ifdef CONFIG_NUMA |
d5f541ed | 6078 | zone->node = nid; |
9614634f | 6079 | #endif |
1da177e4 | 6080 | zone->name = zone_names[j]; |
a52633d8 | 6081 | zone->zone_pgdat = pgdat; |
1da177e4 | 6082 | spin_lock_init(&zone->lock); |
bdc8cb98 | 6083 | zone_seqlock_init(zone); |
ed8ece2e | 6084 | zone_pcp_init(zone); |
81c0a2bb | 6085 | |
1da177e4 LT |
6086 | if (!size) |
6087 | continue; | |
6088 | ||
955c1cd7 | 6089 | set_pageblock_order(); |
7c45512d | 6090 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
dc0bbf3b | 6091 | init_currently_empty_zone(zone, zone_start_pfn, size); |
76cdd58e | 6092 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6093 | } |
6094 | } | |
6095 | ||
bd721ea7 | 6096 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6097 | { |
b0aeba74 | 6098 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6099 | unsigned long __maybe_unused offset = 0; |
6100 | ||
1da177e4 LT |
6101 | /* Skip empty nodes */ |
6102 | if (!pgdat->node_spanned_pages) | |
6103 | return; | |
6104 | ||
d41dee36 | 6105 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
b0aeba74 TL |
6106 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6107 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6108 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6109 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6110 | unsigned long size, end; |
d41dee36 AW |
6111 | struct page *map; |
6112 | ||
e984bb43 BP |
6113 | /* |
6114 | * The zone's endpoints aren't required to be MAX_ORDER | |
6115 | * aligned but the node_mem_map endpoints must be in order | |
6116 | * for the buddy allocator to function correctly. | |
6117 | */ | |
108bcc96 | 6118 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6119 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6120 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
6121 | map = alloc_remap(pgdat->node_id, size); |
6122 | if (!map) | |
6782832e SS |
6123 | map = memblock_virt_alloc_node_nopanic(size, |
6124 | pgdat->node_id); | |
a1c34a3b | 6125 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6126 | } |
12d810c1 | 6127 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6128 | /* |
6129 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6130 | */ | |
c713216d | 6131 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6132 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 6133 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 6134 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6135 | mem_map -= offset; |
0ee332c1 | 6136 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6137 | } |
1da177e4 | 6138 | #endif |
d41dee36 | 6139 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
6140 | } |
6141 | ||
9109fb7b JW |
6142 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
6143 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 6144 | { |
9109fb7b | 6145 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6146 | unsigned long start_pfn = 0; |
6147 | unsigned long end_pfn = 0; | |
9109fb7b | 6148 | |
88fdf75d | 6149 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 6150 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 6151 | |
1da177e4 LT |
6152 | pgdat->node_id = nid; |
6153 | pgdat->node_start_pfn = node_start_pfn; | |
75ef7184 | 6154 | pgdat->per_cpu_nodestats = NULL; |
7960aedd ZY |
6155 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6156 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 6157 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6158 | (u64)start_pfn << PAGE_SHIFT, |
6159 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
6160 | #else |
6161 | start_pfn = node_start_pfn; | |
7960aedd ZY |
6162 | #endif |
6163 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
6164 | zones_size, zholes_size); | |
1da177e4 LT |
6165 | |
6166 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
6167 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
6168 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
6169 | nid, (unsigned long)pgdat, | |
6170 | (unsigned long)pgdat->node_mem_map); | |
6171 | #endif | |
1da177e4 | 6172 | |
864b9a39 | 6173 | reset_deferred_meminit(pgdat); |
7f3eb55b | 6174 | free_area_init_core(pgdat); |
1da177e4 LT |
6175 | } |
6176 | ||
0ee332c1 | 6177 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
6178 | |
6179 | #if MAX_NUMNODES > 1 | |
6180 | /* | |
6181 | * Figure out the number of possible node ids. | |
6182 | */ | |
f9872caf | 6183 | void __init setup_nr_node_ids(void) |
418508c1 | 6184 | { |
904a9553 | 6185 | unsigned int highest; |
418508c1 | 6186 | |
904a9553 | 6187 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
6188 | nr_node_ids = highest + 1; |
6189 | } | |
418508c1 MS |
6190 | #endif |
6191 | ||
1e01979c TH |
6192 | /** |
6193 | * node_map_pfn_alignment - determine the maximum internode alignment | |
6194 | * | |
6195 | * This function should be called after node map is populated and sorted. | |
6196 | * It calculates the maximum power of two alignment which can distinguish | |
6197 | * all the nodes. | |
6198 | * | |
6199 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
6200 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
6201 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
6202 | * shifted, 1GiB is enough and this function will indicate so. | |
6203 | * | |
6204 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
6205 | * model has fine enough granularity to avoid incorrect mapping for the | |
6206 | * populated node map. | |
6207 | * | |
6208 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
6209 | * requirement (single node). | |
6210 | */ | |
6211 | unsigned long __init node_map_pfn_alignment(void) | |
6212 | { | |
6213 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 6214 | unsigned long start, end, mask; |
1e01979c | 6215 | int last_nid = -1; |
c13291a5 | 6216 | int i, nid; |
1e01979c | 6217 | |
c13291a5 | 6218 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
6219 | if (!start || last_nid < 0 || last_nid == nid) { |
6220 | last_nid = nid; | |
6221 | last_end = end; | |
6222 | continue; | |
6223 | } | |
6224 | ||
6225 | /* | |
6226 | * Start with a mask granular enough to pin-point to the | |
6227 | * start pfn and tick off bits one-by-one until it becomes | |
6228 | * too coarse to separate the current node from the last. | |
6229 | */ | |
6230 | mask = ~((1 << __ffs(start)) - 1); | |
6231 | while (mask && last_end <= (start & (mask << 1))) | |
6232 | mask <<= 1; | |
6233 | ||
6234 | /* accumulate all internode masks */ | |
6235 | accl_mask |= mask; | |
6236 | } | |
6237 | ||
6238 | /* convert mask to number of pages */ | |
6239 | return ~accl_mask + 1; | |
6240 | } | |
6241 | ||
a6af2bc3 | 6242 | /* Find the lowest pfn for a node */ |
b69a7288 | 6243 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 6244 | { |
a6af2bc3 | 6245 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
6246 | unsigned long start_pfn; |
6247 | int i; | |
1abbfb41 | 6248 | |
c13291a5 TH |
6249 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
6250 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 6251 | |
a6af2bc3 | 6252 | if (min_pfn == ULONG_MAX) { |
1170532b | 6253 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
6254 | return 0; |
6255 | } | |
6256 | ||
6257 | return min_pfn; | |
c713216d MG |
6258 | } |
6259 | ||
6260 | /** | |
6261 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
6262 | * | |
6263 | * It returns the minimum PFN based on information provided via | |
7d018176 | 6264 | * memblock_set_node(). |
c713216d MG |
6265 | */ |
6266 | unsigned long __init find_min_pfn_with_active_regions(void) | |
6267 | { | |
6268 | return find_min_pfn_for_node(MAX_NUMNODES); | |
6269 | } | |
6270 | ||
37b07e41 LS |
6271 | /* |
6272 | * early_calculate_totalpages() | |
6273 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 6274 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 6275 | */ |
484f51f8 | 6276 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 6277 | { |
7e63efef | 6278 | unsigned long totalpages = 0; |
c13291a5 TH |
6279 | unsigned long start_pfn, end_pfn; |
6280 | int i, nid; | |
6281 | ||
6282 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
6283 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 6284 | |
37b07e41 LS |
6285 | totalpages += pages; |
6286 | if (pages) | |
4b0ef1fe | 6287 | node_set_state(nid, N_MEMORY); |
37b07e41 | 6288 | } |
b8af2941 | 6289 | return totalpages; |
7e63efef MG |
6290 | } |
6291 | ||
2a1e274a MG |
6292 | /* |
6293 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
6294 | * is spread evenly between nodes as long as the nodes have enough | |
6295 | * memory. When they don't, some nodes will have more kernelcore than | |
6296 | * others | |
6297 | */ | |
b224ef85 | 6298 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
6299 | { |
6300 | int i, nid; | |
6301 | unsigned long usable_startpfn; | |
6302 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 6303 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 6304 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 6305 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 6306 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 6307 | struct memblock_region *r; |
b2f3eebe TC |
6308 | |
6309 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
6310 | find_usable_zone_for_movable(); | |
6311 | ||
6312 | /* | |
6313 | * If movable_node is specified, ignore kernelcore and movablecore | |
6314 | * options. | |
6315 | */ | |
6316 | if (movable_node_is_enabled()) { | |
136199f0 EM |
6317 | for_each_memblock(memory, r) { |
6318 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
6319 | continue; |
6320 | ||
136199f0 | 6321 | nid = r->nid; |
b2f3eebe | 6322 | |
136199f0 | 6323 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
6324 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6325 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6326 | usable_startpfn; | |
6327 | } | |
6328 | ||
6329 | goto out2; | |
6330 | } | |
2a1e274a | 6331 | |
342332e6 TI |
6332 | /* |
6333 | * If kernelcore=mirror is specified, ignore movablecore option | |
6334 | */ | |
6335 | if (mirrored_kernelcore) { | |
6336 | bool mem_below_4gb_not_mirrored = false; | |
6337 | ||
6338 | for_each_memblock(memory, r) { | |
6339 | if (memblock_is_mirror(r)) | |
6340 | continue; | |
6341 | ||
6342 | nid = r->nid; | |
6343 | ||
6344 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
6345 | ||
6346 | if (usable_startpfn < 0x100000) { | |
6347 | mem_below_4gb_not_mirrored = true; | |
6348 | continue; | |
6349 | } | |
6350 | ||
6351 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
6352 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6353 | usable_startpfn; | |
6354 | } | |
6355 | ||
6356 | if (mem_below_4gb_not_mirrored) | |
6357 | pr_warn("This configuration results in unmirrored kernel memory."); | |
6358 | ||
6359 | goto out2; | |
6360 | } | |
6361 | ||
7e63efef | 6362 | /* |
b2f3eebe | 6363 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
6364 | * kernelcore that corresponds so that memory usable for |
6365 | * any allocation type is evenly spread. If both kernelcore | |
6366 | * and movablecore are specified, then the value of kernelcore | |
6367 | * will be used for required_kernelcore if it's greater than | |
6368 | * what movablecore would have allowed. | |
6369 | */ | |
6370 | if (required_movablecore) { | |
7e63efef MG |
6371 | unsigned long corepages; |
6372 | ||
6373 | /* | |
6374 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
6375 | * was requested by the user | |
6376 | */ | |
6377 | required_movablecore = | |
6378 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 6379 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
6380 | corepages = totalpages - required_movablecore; |
6381 | ||
6382 | required_kernelcore = max(required_kernelcore, corepages); | |
6383 | } | |
6384 | ||
bde304bd XQ |
6385 | /* |
6386 | * If kernelcore was not specified or kernelcore size is larger | |
6387 | * than totalpages, there is no ZONE_MOVABLE. | |
6388 | */ | |
6389 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 6390 | goto out; |
2a1e274a MG |
6391 | |
6392 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
6393 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6394 | ||
6395 | restart: | |
6396 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
6397 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 6398 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
6399 | unsigned long start_pfn, end_pfn; |
6400 | ||
2a1e274a MG |
6401 | /* |
6402 | * Recalculate kernelcore_node if the division per node | |
6403 | * now exceeds what is necessary to satisfy the requested | |
6404 | * amount of memory for the kernel | |
6405 | */ | |
6406 | if (required_kernelcore < kernelcore_node) | |
6407 | kernelcore_node = required_kernelcore / usable_nodes; | |
6408 | ||
6409 | /* | |
6410 | * As the map is walked, we track how much memory is usable | |
6411 | * by the kernel using kernelcore_remaining. When it is | |
6412 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
6413 | */ | |
6414 | kernelcore_remaining = kernelcore_node; | |
6415 | ||
6416 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 6417 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
6418 | unsigned long size_pages; |
6419 | ||
c13291a5 | 6420 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
6421 | if (start_pfn >= end_pfn) |
6422 | continue; | |
6423 | ||
6424 | /* Account for what is only usable for kernelcore */ | |
6425 | if (start_pfn < usable_startpfn) { | |
6426 | unsigned long kernel_pages; | |
6427 | kernel_pages = min(end_pfn, usable_startpfn) | |
6428 | - start_pfn; | |
6429 | ||
6430 | kernelcore_remaining -= min(kernel_pages, | |
6431 | kernelcore_remaining); | |
6432 | required_kernelcore -= min(kernel_pages, | |
6433 | required_kernelcore); | |
6434 | ||
6435 | /* Continue if range is now fully accounted */ | |
6436 | if (end_pfn <= usable_startpfn) { | |
6437 | ||
6438 | /* | |
6439 | * Push zone_movable_pfn to the end so | |
6440 | * that if we have to rebalance | |
6441 | * kernelcore across nodes, we will | |
6442 | * not double account here | |
6443 | */ | |
6444 | zone_movable_pfn[nid] = end_pfn; | |
6445 | continue; | |
6446 | } | |
6447 | start_pfn = usable_startpfn; | |
6448 | } | |
6449 | ||
6450 | /* | |
6451 | * The usable PFN range for ZONE_MOVABLE is from | |
6452 | * start_pfn->end_pfn. Calculate size_pages as the | |
6453 | * number of pages used as kernelcore | |
6454 | */ | |
6455 | size_pages = end_pfn - start_pfn; | |
6456 | if (size_pages > kernelcore_remaining) | |
6457 | size_pages = kernelcore_remaining; | |
6458 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6459 | ||
6460 | /* | |
6461 | * Some kernelcore has been met, update counts and | |
6462 | * break if the kernelcore for this node has been | |
b8af2941 | 6463 | * satisfied |
2a1e274a MG |
6464 | */ |
6465 | required_kernelcore -= min(required_kernelcore, | |
6466 | size_pages); | |
6467 | kernelcore_remaining -= size_pages; | |
6468 | if (!kernelcore_remaining) | |
6469 | break; | |
6470 | } | |
6471 | } | |
6472 | ||
6473 | /* | |
6474 | * If there is still required_kernelcore, we do another pass with one | |
6475 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6476 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6477 | * satisfied |
2a1e274a MG |
6478 | */ |
6479 | usable_nodes--; | |
6480 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6481 | goto restart; | |
6482 | ||
b2f3eebe | 6483 | out2: |
2a1e274a MG |
6484 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6485 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6486 | zone_movable_pfn[nid] = | |
6487 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6488 | |
20e6926d | 6489 | out: |
66918dcd | 6490 | /* restore the node_state */ |
4b0ef1fe | 6491 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6492 | } |
6493 | ||
4b0ef1fe LJ |
6494 | /* Any regular or high memory on that node ? */ |
6495 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6496 | { |
37b07e41 LS |
6497 | enum zone_type zone_type; |
6498 | ||
4b0ef1fe LJ |
6499 | if (N_MEMORY == N_NORMAL_MEMORY) |
6500 | return; | |
6501 | ||
6502 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 6503 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6504 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
6505 | node_set_state(nid, N_HIGH_MEMORY); |
6506 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
6507 | zone_type <= ZONE_NORMAL) | |
6508 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
6509 | break; |
6510 | } | |
37b07e41 | 6511 | } |
37b07e41 LS |
6512 | } |
6513 | ||
c713216d MG |
6514 | /** |
6515 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6516 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6517 | * |
6518 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6519 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6520 | * zone in each node and their holes is calculated. If the maximum PFN |
6521 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6522 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6523 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6524 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6525 | * at arch_max_dma_pfn. | |
6526 | */ | |
6527 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6528 | { | |
c13291a5 TH |
6529 | unsigned long start_pfn, end_pfn; |
6530 | int i, nid; | |
a6af2bc3 | 6531 | |
c713216d MG |
6532 | /* Record where the zone boundaries are */ |
6533 | memset(arch_zone_lowest_possible_pfn, 0, | |
6534 | sizeof(arch_zone_lowest_possible_pfn)); | |
6535 | memset(arch_zone_highest_possible_pfn, 0, | |
6536 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
6537 | |
6538 | start_pfn = find_min_pfn_with_active_regions(); | |
6539 | ||
6540 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6541 | if (i == ZONE_MOVABLE) |
6542 | continue; | |
90cae1fe OH |
6543 | |
6544 | end_pfn = max(max_zone_pfn[i], start_pfn); | |
6545 | arch_zone_lowest_possible_pfn[i] = start_pfn; | |
6546 | arch_zone_highest_possible_pfn[i] = end_pfn; | |
6547 | ||
6548 | start_pfn = end_pfn; | |
c713216d | 6549 | } |
2a1e274a MG |
6550 | |
6551 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6552 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6553 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6554 | |
c713216d | 6555 | /* Print out the zone ranges */ |
f88dfff5 | 6556 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6557 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6558 | if (i == ZONE_MOVABLE) | |
6559 | continue; | |
f88dfff5 | 6560 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6561 | if (arch_zone_lowest_possible_pfn[i] == |
6562 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6563 | pr_cont("empty\n"); |
72f0ba02 | 6564 | else |
8d29e18a JG |
6565 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6566 | (u64)arch_zone_lowest_possible_pfn[i] | |
6567 | << PAGE_SHIFT, | |
6568 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6569 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6570 | } |
6571 | ||
6572 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6573 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6574 | for (i = 0; i < MAX_NUMNODES; i++) { |
6575 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6576 | pr_info(" Node %d: %#018Lx\n", i, |
6577 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6578 | } |
c713216d | 6579 | |
f2d52fe5 | 6580 | /* Print out the early node map */ |
f88dfff5 | 6581 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6582 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6583 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6584 | (u64)start_pfn << PAGE_SHIFT, | |
6585 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6586 | |
6587 | /* Initialise every node */ | |
708614e6 | 6588 | mminit_verify_pageflags_layout(); |
8ef82866 | 6589 | setup_nr_node_ids(); |
c713216d MG |
6590 | for_each_online_node(nid) { |
6591 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6592 | free_area_init_node(nid, NULL, |
c713216d | 6593 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6594 | |
6595 | /* Any memory on that node */ | |
6596 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6597 | node_set_state(nid, N_MEMORY); |
6598 | check_for_memory(pgdat, nid); | |
c713216d MG |
6599 | } |
6600 | } | |
2a1e274a | 6601 | |
7e63efef | 6602 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
6603 | { |
6604 | unsigned long long coremem; | |
6605 | if (!p) | |
6606 | return -EINVAL; | |
6607 | ||
6608 | coremem = memparse(p, &p); | |
7e63efef | 6609 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 6610 | |
7e63efef | 6611 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
6612 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6613 | ||
6614 | return 0; | |
6615 | } | |
ed7ed365 | 6616 | |
7e63efef MG |
6617 | /* |
6618 | * kernelcore=size sets the amount of memory for use for allocations that | |
6619 | * cannot be reclaimed or migrated. | |
6620 | */ | |
6621 | static int __init cmdline_parse_kernelcore(char *p) | |
6622 | { | |
342332e6 TI |
6623 | /* parse kernelcore=mirror */ |
6624 | if (parse_option_str(p, "mirror")) { | |
6625 | mirrored_kernelcore = true; | |
6626 | return 0; | |
6627 | } | |
6628 | ||
7e63efef MG |
6629 | return cmdline_parse_core(p, &required_kernelcore); |
6630 | } | |
6631 | ||
6632 | /* | |
6633 | * movablecore=size sets the amount of memory for use for allocations that | |
6634 | * can be reclaimed or migrated. | |
6635 | */ | |
6636 | static int __init cmdline_parse_movablecore(char *p) | |
6637 | { | |
6638 | return cmdline_parse_core(p, &required_movablecore); | |
6639 | } | |
6640 | ||
ed7ed365 | 6641 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6642 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6643 | |
0ee332c1 | 6644 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6645 | |
c3d5f5f0 JL |
6646 | void adjust_managed_page_count(struct page *page, long count) |
6647 | { | |
6648 | spin_lock(&managed_page_count_lock); | |
6649 | page_zone(page)->managed_pages += count; | |
6650 | totalram_pages += count; | |
3dcc0571 JL |
6651 | #ifdef CONFIG_HIGHMEM |
6652 | if (PageHighMem(page)) | |
6653 | totalhigh_pages += count; | |
6654 | #endif | |
c3d5f5f0 JL |
6655 | spin_unlock(&managed_page_count_lock); |
6656 | } | |
3dcc0571 | 6657 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6658 | |
11199692 | 6659 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6660 | { |
11199692 JL |
6661 | void *pos; |
6662 | unsigned long pages = 0; | |
69afade7 | 6663 | |
11199692 JL |
6664 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6665 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6666 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 6667 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
6668 | memset(pos, poison, PAGE_SIZE); |
6669 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
6670 | } |
6671 | ||
6672 | if (pages && s) | |
adb1fe9a JP |
6673 | pr_info("Freeing %s memory: %ldK\n", |
6674 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
6675 | |
6676 | return pages; | |
6677 | } | |
11199692 | 6678 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6679 | |
cfa11e08 JL |
6680 | #ifdef CONFIG_HIGHMEM |
6681 | void free_highmem_page(struct page *page) | |
6682 | { | |
6683 | __free_reserved_page(page); | |
6684 | totalram_pages++; | |
7b4b2a0d | 6685 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6686 | totalhigh_pages++; |
6687 | } | |
6688 | #endif | |
6689 | ||
7ee3d4e8 JL |
6690 | |
6691 | void __init mem_init_print_info(const char *str) | |
6692 | { | |
6693 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
6694 | unsigned long init_code_size, init_data_size; | |
6695 | ||
6696 | physpages = get_num_physpages(); | |
6697 | codesize = _etext - _stext; | |
6698 | datasize = _edata - _sdata; | |
6699 | rosize = __end_rodata - __start_rodata; | |
6700 | bss_size = __bss_stop - __bss_start; | |
6701 | init_data_size = __init_end - __init_begin; | |
6702 | init_code_size = _einittext - _sinittext; | |
6703 | ||
6704 | /* | |
6705 | * Detect special cases and adjust section sizes accordingly: | |
6706 | * 1) .init.* may be embedded into .data sections | |
6707 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
6708 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
6709 | * 3) .rodata.* may be embedded into .text or .data sections. | |
6710 | */ | |
6711 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
6712 | do { \ |
6713 | if (start <= pos && pos < end && size > adj) \ | |
6714 | size -= adj; \ | |
6715 | } while (0) | |
7ee3d4e8 JL |
6716 | |
6717 | adj_init_size(__init_begin, __init_end, init_data_size, | |
6718 | _sinittext, init_code_size); | |
6719 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
6720 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
6721 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
6722 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
6723 | ||
6724 | #undef adj_init_size | |
6725 | ||
756a025f | 6726 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 6727 | #ifdef CONFIG_HIGHMEM |
756a025f | 6728 | ", %luK highmem" |
7ee3d4e8 | 6729 | #endif |
756a025f JP |
6730 | "%s%s)\n", |
6731 | nr_free_pages() << (PAGE_SHIFT - 10), | |
6732 | physpages << (PAGE_SHIFT - 10), | |
6733 | codesize >> 10, datasize >> 10, rosize >> 10, | |
6734 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
6735 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
6736 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 6737 | #ifdef CONFIG_HIGHMEM |
756a025f | 6738 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 6739 | #endif |
756a025f | 6740 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
6741 | } |
6742 | ||
0e0b864e | 6743 | /** |
88ca3b94 RD |
6744 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6745 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 6746 | * |
013110a7 | 6747 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
6748 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6749 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
6750 | * function may optionally be used to account for unfreeable pages in the |
6751 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
6752 | * smaller per-cpu batchsize. | |
0e0b864e MG |
6753 | */ |
6754 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
6755 | { | |
6756 | dma_reserve = new_dma_reserve; | |
6757 | } | |
6758 | ||
1da177e4 LT |
6759 | void __init free_area_init(unsigned long *zones_size) |
6760 | { | |
9109fb7b | 6761 | free_area_init_node(0, zones_size, |
1da177e4 LT |
6762 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6763 | } | |
1da177e4 | 6764 | |
005fd4bb | 6765 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 6766 | { |
1da177e4 | 6767 | |
005fd4bb SAS |
6768 | lru_add_drain_cpu(cpu); |
6769 | drain_pages(cpu); | |
9f8f2172 | 6770 | |
005fd4bb SAS |
6771 | /* |
6772 | * Spill the event counters of the dead processor | |
6773 | * into the current processors event counters. | |
6774 | * This artificially elevates the count of the current | |
6775 | * processor. | |
6776 | */ | |
6777 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 6778 | |
005fd4bb SAS |
6779 | /* |
6780 | * Zero the differential counters of the dead processor | |
6781 | * so that the vm statistics are consistent. | |
6782 | * | |
6783 | * This is only okay since the processor is dead and cannot | |
6784 | * race with what we are doing. | |
6785 | */ | |
6786 | cpu_vm_stats_fold(cpu); | |
6787 | return 0; | |
1da177e4 | 6788 | } |
1da177e4 LT |
6789 | |
6790 | void __init page_alloc_init(void) | |
6791 | { | |
005fd4bb SAS |
6792 | int ret; |
6793 | ||
6794 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, | |
6795 | "mm/page_alloc:dead", NULL, | |
6796 | page_alloc_cpu_dead); | |
6797 | WARN_ON(ret < 0); | |
1da177e4 LT |
6798 | } |
6799 | ||
cb45b0e9 | 6800 | /* |
34b10060 | 6801 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
6802 | * or min_free_kbytes changes. |
6803 | */ | |
6804 | static void calculate_totalreserve_pages(void) | |
6805 | { | |
6806 | struct pglist_data *pgdat; | |
6807 | unsigned long reserve_pages = 0; | |
2f6726e5 | 6808 | enum zone_type i, j; |
cb45b0e9 HA |
6809 | |
6810 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
6811 | |
6812 | pgdat->totalreserve_pages = 0; | |
6813 | ||
cb45b0e9 HA |
6814 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6815 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 6816 | long max = 0; |
cb45b0e9 HA |
6817 | |
6818 | /* Find valid and maximum lowmem_reserve in the zone */ | |
6819 | for (j = i; j < MAX_NR_ZONES; j++) { | |
6820 | if (zone->lowmem_reserve[j] > max) | |
6821 | max = zone->lowmem_reserve[j]; | |
6822 | } | |
6823 | ||
41858966 MG |
6824 | /* we treat the high watermark as reserved pages. */ |
6825 | max += high_wmark_pages(zone); | |
cb45b0e9 | 6826 | |
b40da049 JL |
6827 | if (max > zone->managed_pages) |
6828 | max = zone->managed_pages; | |
a8d01437 | 6829 | |
281e3726 | 6830 | pgdat->totalreserve_pages += max; |
a8d01437 | 6831 | |
cb45b0e9 HA |
6832 | reserve_pages += max; |
6833 | } | |
6834 | } | |
6835 | totalreserve_pages = reserve_pages; | |
6836 | } | |
6837 | ||
1da177e4 LT |
6838 | /* |
6839 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 6840 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
6841 | * has a correct pages reserved value, so an adequate number of |
6842 | * pages are left in the zone after a successful __alloc_pages(). | |
6843 | */ | |
6844 | static void setup_per_zone_lowmem_reserve(void) | |
6845 | { | |
6846 | struct pglist_data *pgdat; | |
2f6726e5 | 6847 | enum zone_type j, idx; |
1da177e4 | 6848 | |
ec936fc5 | 6849 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
6850 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6851 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 6852 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
6853 | |
6854 | zone->lowmem_reserve[j] = 0; | |
6855 | ||
2f6726e5 CL |
6856 | idx = j; |
6857 | while (idx) { | |
1da177e4 LT |
6858 | struct zone *lower_zone; |
6859 | ||
2f6726e5 CL |
6860 | idx--; |
6861 | ||
1da177e4 LT |
6862 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6863 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
6864 | ||
6865 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 6866 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 6867 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 6868 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
6869 | } |
6870 | } | |
6871 | } | |
cb45b0e9 HA |
6872 | |
6873 | /* update totalreserve_pages */ | |
6874 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6875 | } |
6876 | ||
cfd3da1e | 6877 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
6878 | { |
6879 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
6880 | unsigned long lowmem_pages = 0; | |
6881 | struct zone *zone; | |
6882 | unsigned long flags; | |
6883 | ||
6884 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
6885 | for_each_zone(zone) { | |
6886 | if (!is_highmem(zone)) | |
b40da049 | 6887 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
6888 | } |
6889 | ||
6890 | for_each_zone(zone) { | |
ac924c60 AM |
6891 | u64 tmp; |
6892 | ||
1125b4e3 | 6893 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 6894 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 6895 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
6896 | if (is_highmem(zone)) { |
6897 | /* | |
669ed175 NP |
6898 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6899 | * need highmem pages, so cap pages_min to a small | |
6900 | * value here. | |
6901 | * | |
41858966 | 6902 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 6903 | * deltas control asynch page reclaim, and so should |
669ed175 | 6904 | * not be capped for highmem. |
1da177e4 | 6905 | */ |
90ae8d67 | 6906 | unsigned long min_pages; |
1da177e4 | 6907 | |
b40da049 | 6908 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 6909 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 6910 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 6911 | } else { |
669ed175 NP |
6912 | /* |
6913 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
6914 | * proportionate to the zone's size. |
6915 | */ | |
41858966 | 6916 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
6917 | } |
6918 | ||
795ae7a0 JW |
6919 | /* |
6920 | * Set the kswapd watermarks distance according to the | |
6921 | * scale factor in proportion to available memory, but | |
6922 | * ensure a minimum size on small systems. | |
6923 | */ | |
6924 | tmp = max_t(u64, tmp >> 2, | |
6925 | mult_frac(zone->managed_pages, | |
6926 | watermark_scale_factor, 10000)); | |
6927 | ||
6928 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
6929 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 6930 | |
1125b4e3 | 6931 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 6932 | } |
cb45b0e9 HA |
6933 | |
6934 | /* update totalreserve_pages */ | |
6935 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6936 | } |
6937 | ||
cfd3da1e MG |
6938 | /** |
6939 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
6940 | * or when memory is hot-{added|removed} | |
6941 | * | |
6942 | * Ensures that the watermark[min,low,high] values for each zone are set | |
6943 | * correctly with respect to min_free_kbytes. | |
6944 | */ | |
6945 | void setup_per_zone_wmarks(void) | |
6946 | { | |
6947 | mutex_lock(&zonelists_mutex); | |
6948 | __setup_per_zone_wmarks(); | |
6949 | mutex_unlock(&zonelists_mutex); | |
6950 | } | |
6951 | ||
1da177e4 LT |
6952 | /* |
6953 | * Initialise min_free_kbytes. | |
6954 | * | |
6955 | * For small machines we want it small (128k min). For large machines | |
6956 | * we want it large (64MB max). But it is not linear, because network | |
6957 | * bandwidth does not increase linearly with machine size. We use | |
6958 | * | |
b8af2941 | 6959 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6960 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6961 | * | |
6962 | * which yields | |
6963 | * | |
6964 | * 16MB: 512k | |
6965 | * 32MB: 724k | |
6966 | * 64MB: 1024k | |
6967 | * 128MB: 1448k | |
6968 | * 256MB: 2048k | |
6969 | * 512MB: 2896k | |
6970 | * 1024MB: 4096k | |
6971 | * 2048MB: 5792k | |
6972 | * 4096MB: 8192k | |
6973 | * 8192MB: 11584k | |
6974 | * 16384MB: 16384k | |
6975 | */ | |
1b79acc9 | 6976 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6977 | { |
6978 | unsigned long lowmem_kbytes; | |
5f12733e | 6979 | int new_min_free_kbytes; |
1da177e4 LT |
6980 | |
6981 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6982 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6983 | ||
6984 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6985 | min_free_kbytes = new_min_free_kbytes; | |
6986 | if (min_free_kbytes < 128) | |
6987 | min_free_kbytes = 128; | |
6988 | if (min_free_kbytes > 65536) | |
6989 | min_free_kbytes = 65536; | |
6990 | } else { | |
6991 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6992 | new_min_free_kbytes, user_min_free_kbytes); | |
6993 | } | |
bc75d33f | 6994 | setup_per_zone_wmarks(); |
a6cccdc3 | 6995 | refresh_zone_stat_thresholds(); |
1da177e4 | 6996 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
6997 | |
6998 | #ifdef CONFIG_NUMA | |
6999 | setup_min_unmapped_ratio(); | |
7000 | setup_min_slab_ratio(); | |
7001 | #endif | |
7002 | ||
1da177e4 LT |
7003 | return 0; |
7004 | } | |
bc22af74 | 7005 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
7006 | |
7007 | /* | |
b8af2941 | 7008 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
7009 | * that we can call two helper functions whenever min_free_kbytes |
7010 | * changes. | |
7011 | */ | |
cccad5b9 | 7012 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7013 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7014 | { |
da8c757b HP |
7015 | int rc; |
7016 | ||
7017 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7018 | if (rc) | |
7019 | return rc; | |
7020 | ||
5f12733e MH |
7021 | if (write) { |
7022 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7023 | setup_per_zone_wmarks(); |
5f12733e | 7024 | } |
1da177e4 LT |
7025 | return 0; |
7026 | } | |
7027 | ||
795ae7a0 JW |
7028 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
7029 | void __user *buffer, size_t *length, loff_t *ppos) | |
7030 | { | |
7031 | int rc; | |
7032 | ||
7033 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7034 | if (rc) | |
7035 | return rc; | |
7036 | ||
7037 | if (write) | |
7038 | setup_per_zone_wmarks(); | |
7039 | ||
7040 | return 0; | |
7041 | } | |
7042 | ||
9614634f | 7043 | #ifdef CONFIG_NUMA |
6423aa81 | 7044 | static void setup_min_unmapped_ratio(void) |
9614634f | 7045 | { |
6423aa81 | 7046 | pg_data_t *pgdat; |
9614634f | 7047 | struct zone *zone; |
9614634f | 7048 | |
a5f5f91d | 7049 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7050 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7051 | |
9614634f | 7052 | for_each_zone(zone) |
a5f5f91d | 7053 | zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages * |
9614634f | 7054 | sysctl_min_unmapped_ratio) / 100; |
9614634f | 7055 | } |
0ff38490 | 7056 | |
6423aa81 JK |
7057 | |
7058 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 7059 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7060 | { |
0ff38490 CL |
7061 | int rc; |
7062 | ||
8d65af78 | 7063 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7064 | if (rc) |
7065 | return rc; | |
7066 | ||
6423aa81 JK |
7067 | setup_min_unmapped_ratio(); |
7068 | ||
7069 | return 0; | |
7070 | } | |
7071 | ||
7072 | static void setup_min_slab_ratio(void) | |
7073 | { | |
7074 | pg_data_t *pgdat; | |
7075 | struct zone *zone; | |
7076 | ||
a5f5f91d MG |
7077 | for_each_online_pgdat(pgdat) |
7078 | pgdat->min_slab_pages = 0; | |
7079 | ||
0ff38490 | 7080 | for_each_zone(zone) |
a5f5f91d | 7081 | zone->zone_pgdat->min_slab_pages += (zone->managed_pages * |
0ff38490 | 7082 | sysctl_min_slab_ratio) / 100; |
6423aa81 JK |
7083 | } |
7084 | ||
7085 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
7086 | void __user *buffer, size_t *length, loff_t *ppos) | |
7087 | { | |
7088 | int rc; | |
7089 | ||
7090 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7091 | if (rc) | |
7092 | return rc; | |
7093 | ||
7094 | setup_min_slab_ratio(); | |
7095 | ||
0ff38490 CL |
7096 | return 0; |
7097 | } | |
9614634f CL |
7098 | #endif |
7099 | ||
1da177e4 LT |
7100 | /* |
7101 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7102 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7103 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7104 | * | |
7105 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7106 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7107 | * if in function of the boot time zone sizes. |
7108 | */ | |
cccad5b9 | 7109 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7110 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7111 | { |
8d65af78 | 7112 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
7113 | setup_per_zone_lowmem_reserve(); |
7114 | return 0; | |
7115 | } | |
7116 | ||
8ad4b1fb RS |
7117 | /* |
7118 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
7119 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
7120 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 7121 | */ |
cccad5b9 | 7122 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7123 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
7124 | { |
7125 | struct zone *zone; | |
7cd2b0a3 | 7126 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
7127 | int ret; |
7128 | ||
7cd2b0a3 DR |
7129 | mutex_lock(&pcp_batch_high_lock); |
7130 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
7131 | ||
8d65af78 | 7132 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
7133 | if (!write || ret < 0) |
7134 | goto out; | |
7135 | ||
7136 | /* Sanity checking to avoid pcp imbalance */ | |
7137 | if (percpu_pagelist_fraction && | |
7138 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
7139 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
7140 | ret = -EINVAL; | |
7141 | goto out; | |
7142 | } | |
7143 | ||
7144 | /* No change? */ | |
7145 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
7146 | goto out; | |
c8e251fa | 7147 | |
364df0eb | 7148 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
7149 | unsigned int cpu; |
7150 | ||
22a7f12b | 7151 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
7152 | pageset_set_high_and_batch(zone, |
7153 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 7154 | } |
7cd2b0a3 | 7155 | out: |
c8e251fa | 7156 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 7157 | return ret; |
8ad4b1fb RS |
7158 | } |
7159 | ||
a9919c79 | 7160 | #ifdef CONFIG_NUMA |
f034b5d4 | 7161 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 7162 | |
1da177e4 LT |
7163 | static int __init set_hashdist(char *str) |
7164 | { | |
7165 | if (!str) | |
7166 | return 0; | |
7167 | hashdist = simple_strtoul(str, &str, 0); | |
7168 | return 1; | |
7169 | } | |
7170 | __setup("hashdist=", set_hashdist); | |
7171 | #endif | |
7172 | ||
f6f34b43 SD |
7173 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
7174 | /* | |
7175 | * Returns the number of pages that arch has reserved but | |
7176 | * is not known to alloc_large_system_hash(). | |
7177 | */ | |
7178 | static unsigned long __init arch_reserved_kernel_pages(void) | |
7179 | { | |
7180 | return 0; | |
7181 | } | |
7182 | #endif | |
7183 | ||
9017217b PT |
7184 | /* |
7185 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
7186 | * machines. As memory size is increased the scale is also increased but at | |
7187 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
7188 | * quadruples the scale is increased by one, which means the size of hash table | |
7189 | * only doubles, instead of quadrupling as well. | |
7190 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
7191 | * makes sense, it is disabled on such platforms. | |
7192 | */ | |
7193 | #if __BITS_PER_LONG > 32 | |
7194 | #define ADAPT_SCALE_BASE (64ul << 30) | |
7195 | #define ADAPT_SCALE_SHIFT 2 | |
7196 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
7197 | #endif | |
7198 | ||
1da177e4 LT |
7199 | /* |
7200 | * allocate a large system hash table from bootmem | |
7201 | * - it is assumed that the hash table must contain an exact power-of-2 | |
7202 | * quantity of entries | |
7203 | * - limit is the number of hash buckets, not the total allocation size | |
7204 | */ | |
7205 | void *__init alloc_large_system_hash(const char *tablename, | |
7206 | unsigned long bucketsize, | |
7207 | unsigned long numentries, | |
7208 | int scale, | |
7209 | int flags, | |
7210 | unsigned int *_hash_shift, | |
7211 | unsigned int *_hash_mask, | |
31fe62b9 TB |
7212 | unsigned long low_limit, |
7213 | unsigned long high_limit) | |
1da177e4 | 7214 | { |
31fe62b9 | 7215 | unsigned long long max = high_limit; |
1da177e4 LT |
7216 | unsigned long log2qty, size; |
7217 | void *table = NULL; | |
3749a8f0 | 7218 | gfp_t gfp_flags; |
1da177e4 LT |
7219 | |
7220 | /* allow the kernel cmdline to have a say */ | |
7221 | if (!numentries) { | |
7222 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 7223 | numentries = nr_kernel_pages; |
f6f34b43 | 7224 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
7225 | |
7226 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
7227 | if (PAGE_SHIFT < 20) | |
7228 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 7229 | |
9017217b PT |
7230 | #if __BITS_PER_LONG > 32 |
7231 | if (!high_limit) { | |
7232 | unsigned long adapt; | |
7233 | ||
7234 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
7235 | adapt <<= ADAPT_SCALE_SHIFT) | |
7236 | scale++; | |
7237 | } | |
7238 | #endif | |
7239 | ||
1da177e4 LT |
7240 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
7241 | if (scale > PAGE_SHIFT) | |
7242 | numentries >>= (scale - PAGE_SHIFT); | |
7243 | else | |
7244 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
7245 | |
7246 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
7247 | if (unlikely(flags & HASH_SMALL)) { |
7248 | /* Makes no sense without HASH_EARLY */ | |
7249 | WARN_ON(!(flags & HASH_EARLY)); | |
7250 | if (!(numentries >> *_hash_shift)) { | |
7251 | numentries = 1UL << *_hash_shift; | |
7252 | BUG_ON(!numentries); | |
7253 | } | |
7254 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 7255 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 7256 | } |
6e692ed3 | 7257 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
7258 | |
7259 | /* limit allocation size to 1/16 total memory by default */ | |
7260 | if (max == 0) { | |
7261 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
7262 | do_div(max, bucketsize); | |
7263 | } | |
074b8517 | 7264 | max = min(max, 0x80000000ULL); |
1da177e4 | 7265 | |
31fe62b9 TB |
7266 | if (numentries < low_limit) |
7267 | numentries = low_limit; | |
1da177e4 LT |
7268 | if (numentries > max) |
7269 | numentries = max; | |
7270 | ||
f0d1b0b3 | 7271 | log2qty = ilog2(numentries); |
1da177e4 | 7272 | |
3749a8f0 PT |
7273 | /* |
7274 | * memblock allocator returns zeroed memory already, so HASH_ZERO is | |
7275 | * currently not used when HASH_EARLY is specified. | |
7276 | */ | |
7277 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; | |
1da177e4 LT |
7278 | do { |
7279 | size = bucketsize << log2qty; | |
7280 | if (flags & HASH_EARLY) | |
6782832e | 7281 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 | 7282 | else if (hashdist) |
3749a8f0 | 7283 | table = __vmalloc(size, gfp_flags, PAGE_KERNEL); |
1da177e4 | 7284 | else { |
1037b83b ED |
7285 | /* |
7286 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
7287 | * some pages at the end of hash table which |
7288 | * alloc_pages_exact() automatically does | |
1037b83b | 7289 | */ |
264ef8a9 | 7290 | if (get_order(size) < MAX_ORDER) { |
3749a8f0 PT |
7291 | table = alloc_pages_exact(size, gfp_flags); |
7292 | kmemleak_alloc(table, size, 1, gfp_flags); | |
264ef8a9 | 7293 | } |
1da177e4 LT |
7294 | } |
7295 | } while (!table && size > PAGE_SIZE && --log2qty); | |
7296 | ||
7297 | if (!table) | |
7298 | panic("Failed to allocate %s hash table\n", tablename); | |
7299 | ||
1170532b JP |
7300 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
7301 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
7302 | |
7303 | if (_hash_shift) | |
7304 | *_hash_shift = log2qty; | |
7305 | if (_hash_mask) | |
7306 | *_hash_mask = (1 << log2qty) - 1; | |
7307 | ||
7308 | return table; | |
7309 | } | |
a117e66e | 7310 | |
a5d76b54 | 7311 | /* |
80934513 MK |
7312 | * This function checks whether pageblock includes unmovable pages or not. |
7313 | * If @count is not zero, it is okay to include less @count unmovable pages | |
7314 | * | |
b8af2941 | 7315 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
7316 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
7317 | * check without lock_page also may miss some movable non-lru pages at | |
7318 | * race condition. So you can't expect this function should be exact. | |
a5d76b54 | 7319 | */ |
b023f468 WC |
7320 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
7321 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
7322 | { |
7323 | unsigned long pfn, iter, found; | |
47118af0 MN |
7324 | int mt; |
7325 | ||
49ac8255 KH |
7326 | /* |
7327 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 7328 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
7329 | */ |
7330 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 7331 | return false; |
47118af0 MN |
7332 | mt = get_pageblock_migratetype(page); |
7333 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 7334 | return false; |
49ac8255 KH |
7335 | |
7336 | pfn = page_to_pfn(page); | |
7337 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
7338 | unsigned long check = pfn + iter; | |
7339 | ||
29723fcc | 7340 | if (!pfn_valid_within(check)) |
49ac8255 | 7341 | continue; |
29723fcc | 7342 | |
49ac8255 | 7343 | page = pfn_to_page(check); |
c8721bbb NH |
7344 | |
7345 | /* | |
7346 | * Hugepages are not in LRU lists, but they're movable. | |
7347 | * We need not scan over tail pages bacause we don't | |
7348 | * handle each tail page individually in migration. | |
7349 | */ | |
7350 | if (PageHuge(page)) { | |
7351 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
7352 | continue; | |
7353 | } | |
7354 | ||
97d255c8 MK |
7355 | /* |
7356 | * We can't use page_count without pin a page | |
7357 | * because another CPU can free compound page. | |
7358 | * This check already skips compound tails of THP | |
0139aa7b | 7359 | * because their page->_refcount is zero at all time. |
97d255c8 | 7360 | */ |
fe896d18 | 7361 | if (!page_ref_count(page)) { |
49ac8255 KH |
7362 | if (PageBuddy(page)) |
7363 | iter += (1 << page_order(page)) - 1; | |
7364 | continue; | |
7365 | } | |
97d255c8 | 7366 | |
b023f468 WC |
7367 | /* |
7368 | * The HWPoisoned page may be not in buddy system, and | |
7369 | * page_count() is not 0. | |
7370 | */ | |
7371 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
7372 | continue; | |
7373 | ||
0efadf48 YX |
7374 | if (__PageMovable(page)) |
7375 | continue; | |
7376 | ||
49ac8255 KH |
7377 | if (!PageLRU(page)) |
7378 | found++; | |
7379 | /* | |
6b4f7799 JW |
7380 | * If there are RECLAIMABLE pages, we need to check |
7381 | * it. But now, memory offline itself doesn't call | |
7382 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
7383 | */ |
7384 | /* | |
7385 | * If the page is not RAM, page_count()should be 0. | |
7386 | * we don't need more check. This is an _used_ not-movable page. | |
7387 | * | |
7388 | * The problematic thing here is PG_reserved pages. PG_reserved | |
7389 | * is set to both of a memory hole page and a _used_ kernel | |
7390 | * page at boot. | |
7391 | */ | |
7392 | if (found > count) | |
80934513 | 7393 | return true; |
49ac8255 | 7394 | } |
80934513 | 7395 | return false; |
49ac8255 KH |
7396 | } |
7397 | ||
7398 | bool is_pageblock_removable_nolock(struct page *page) | |
7399 | { | |
656a0706 MH |
7400 | struct zone *zone; |
7401 | unsigned long pfn; | |
687875fb MH |
7402 | |
7403 | /* | |
7404 | * We have to be careful here because we are iterating over memory | |
7405 | * sections which are not zone aware so we might end up outside of | |
7406 | * the zone but still within the section. | |
656a0706 MH |
7407 | * We have to take care about the node as well. If the node is offline |
7408 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 7409 | */ |
656a0706 MH |
7410 | if (!node_online(page_to_nid(page))) |
7411 | return false; | |
7412 | ||
7413 | zone = page_zone(page); | |
7414 | pfn = page_to_pfn(page); | |
108bcc96 | 7415 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
7416 | return false; |
7417 | ||
b023f468 | 7418 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 7419 | } |
0c0e6195 | 7420 | |
080fe206 | 7421 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7422 | |
7423 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7424 | { | |
7425 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7426 | pageblock_nr_pages) - 1); | |
7427 | } | |
7428 | ||
7429 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7430 | { | |
7431 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7432 | pageblock_nr_pages)); | |
7433 | } | |
7434 | ||
041d3a8c | 7435 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7436 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7437 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7438 | { |
7439 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7440 | unsigned long nr_reclaimed; |
041d3a8c MN |
7441 | unsigned long pfn = start; |
7442 | unsigned int tries = 0; | |
7443 | int ret = 0; | |
7444 | ||
be49a6e1 | 7445 | migrate_prep(); |
041d3a8c | 7446 | |
bb13ffeb | 7447 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7448 | if (fatal_signal_pending(current)) { |
7449 | ret = -EINTR; | |
7450 | break; | |
7451 | } | |
7452 | ||
bb13ffeb MG |
7453 | if (list_empty(&cc->migratepages)) { |
7454 | cc->nr_migratepages = 0; | |
edc2ca61 | 7455 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7456 | if (!pfn) { |
7457 | ret = -EINTR; | |
7458 | break; | |
7459 | } | |
7460 | tries = 0; | |
7461 | } else if (++tries == 5) { | |
7462 | ret = ret < 0 ? ret : -EBUSY; | |
7463 | break; | |
7464 | } | |
7465 | ||
beb51eaa MK |
7466 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7467 | &cc->migratepages); | |
7468 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7469 | |
9c620e2b | 7470 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 7471 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 7472 | } |
2a6f5124 SP |
7473 | if (ret < 0) { |
7474 | putback_movable_pages(&cc->migratepages); | |
7475 | return ret; | |
7476 | } | |
7477 | return 0; | |
041d3a8c MN |
7478 | } |
7479 | ||
7480 | /** | |
7481 | * alloc_contig_range() -- tries to allocate given range of pages | |
7482 | * @start: start PFN to allocate | |
7483 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7484 | * @migratetype: migratetype of the underlaying pageblocks (either |
7485 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7486 | * in range must have the same migratetype and it must | |
7487 | * be either of the two. | |
ca96b625 | 7488 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
7489 | * |
7490 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
7491 | * aligned, however it's the caller's responsibility to guarantee that | |
7492 | * we are the only thread that changes migrate type of pageblocks the | |
7493 | * pages fall in. | |
7494 | * | |
7495 | * The PFN range must belong to a single zone. | |
7496 | * | |
7497 | * Returns zero on success or negative error code. On success all | |
7498 | * pages which PFN is in [start, end) are allocated for the caller and | |
7499 | * need to be freed with free_contig_range(). | |
7500 | */ | |
0815f3d8 | 7501 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 7502 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 7503 | { |
041d3a8c | 7504 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7505 | unsigned int order; |
7506 | int ret = 0; | |
041d3a8c | 7507 | |
bb13ffeb MG |
7508 | struct compact_control cc = { |
7509 | .nr_migratepages = 0, | |
7510 | .order = -1, | |
7511 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7512 | .mode = MIGRATE_SYNC, |
bb13ffeb | 7513 | .ignore_skip_hint = true, |
7dea19f9 | 7514 | .gfp_mask = current_gfp_context(gfp_mask), |
bb13ffeb MG |
7515 | }; |
7516 | INIT_LIST_HEAD(&cc.migratepages); | |
7517 | ||
041d3a8c MN |
7518 | /* |
7519 | * What we do here is we mark all pageblocks in range as | |
7520 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7521 | * have different sizes, and due to the way page allocator | |
7522 | * work, we align the range to biggest of the two pages so | |
7523 | * that page allocator won't try to merge buddies from | |
7524 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7525 | * other migration type. | |
7526 | * | |
7527 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7528 | * migrate the pages from an unaligned range (ie. pages that | |
7529 | * we are interested in). This will put all the pages in | |
7530 | * range back to page allocator as MIGRATE_ISOLATE. | |
7531 | * | |
7532 | * When this is done, we take the pages in range from page | |
7533 | * allocator removing them from the buddy system. This way | |
7534 | * page allocator will never consider using them. | |
7535 | * | |
7536 | * This lets us mark the pageblocks back as | |
7537 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7538 | * aligned range but not in the unaligned, original range are | |
7539 | * put back to page allocator so that buddy can use them. | |
7540 | */ | |
7541 | ||
7542 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7543 | pfn_max_align_up(end), migratetype, |
7544 | false); | |
041d3a8c | 7545 | if (ret) |
86a595f9 | 7546 | return ret; |
041d3a8c | 7547 | |
8ef5849f JK |
7548 | /* |
7549 | * In case of -EBUSY, we'd like to know which page causes problem. | |
7550 | * So, just fall through. We will check it in test_pages_isolated(). | |
7551 | */ | |
bb13ffeb | 7552 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7553 | if (ret && ret != -EBUSY) |
041d3a8c MN |
7554 | goto done; |
7555 | ||
7556 | /* | |
7557 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7558 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7559 | * more, all pages in [start, end) are free in page allocator. | |
7560 | * What we are going to do is to allocate all pages from | |
7561 | * [start, end) (that is remove them from page allocator). | |
7562 | * | |
7563 | * The only problem is that pages at the beginning and at the | |
7564 | * end of interesting range may be not aligned with pages that | |
7565 | * page allocator holds, ie. they can be part of higher order | |
7566 | * pages. Because of this, we reserve the bigger range and | |
7567 | * once this is done free the pages we are not interested in. | |
7568 | * | |
7569 | * We don't have to hold zone->lock here because the pages are | |
7570 | * isolated thus they won't get removed from buddy. | |
7571 | */ | |
7572 | ||
7573 | lru_add_drain_all(); | |
510f5507 | 7574 | drain_all_pages(cc.zone); |
041d3a8c MN |
7575 | |
7576 | order = 0; | |
7577 | outer_start = start; | |
7578 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7579 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7580 | outer_start = start; |
7581 | break; | |
041d3a8c MN |
7582 | } |
7583 | outer_start &= ~0UL << order; | |
7584 | } | |
7585 | ||
8ef5849f JK |
7586 | if (outer_start != start) { |
7587 | order = page_order(pfn_to_page(outer_start)); | |
7588 | ||
7589 | /* | |
7590 | * outer_start page could be small order buddy page and | |
7591 | * it doesn't include start page. Adjust outer_start | |
7592 | * in this case to report failed page properly | |
7593 | * on tracepoint in test_pages_isolated() | |
7594 | */ | |
7595 | if (outer_start + (1UL << order) <= start) | |
7596 | outer_start = start; | |
7597 | } | |
7598 | ||
041d3a8c | 7599 | /* Make sure the range is really isolated. */ |
b023f468 | 7600 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
7601 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
7602 | __func__, outer_start, end); | |
041d3a8c MN |
7603 | ret = -EBUSY; |
7604 | goto done; | |
7605 | } | |
7606 | ||
49f223a9 | 7607 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7608 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7609 | if (!outer_end) { |
7610 | ret = -EBUSY; | |
7611 | goto done; | |
7612 | } | |
7613 | ||
7614 | /* Free head and tail (if any) */ | |
7615 | if (start != outer_start) | |
7616 | free_contig_range(outer_start, start - outer_start); | |
7617 | if (end != outer_end) | |
7618 | free_contig_range(end, outer_end - end); | |
7619 | ||
7620 | done: | |
7621 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7622 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7623 | return ret; |
7624 | } | |
7625 | ||
7626 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7627 | { | |
bcc2b02f MS |
7628 | unsigned int count = 0; |
7629 | ||
7630 | for (; nr_pages--; pfn++) { | |
7631 | struct page *page = pfn_to_page(pfn); | |
7632 | ||
7633 | count += page_count(page) != 1; | |
7634 | __free_page(page); | |
7635 | } | |
7636 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7637 | } |
7638 | #endif | |
7639 | ||
4ed7e022 | 7640 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7641 | /* |
7642 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7643 | * page high values need to be recalulated. | |
7644 | */ | |
4ed7e022 JL |
7645 | void __meminit zone_pcp_update(struct zone *zone) |
7646 | { | |
0a647f38 | 7647 | unsigned cpu; |
c8e251fa | 7648 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7649 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7650 | pageset_set_high_and_batch(zone, |
7651 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7652 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7653 | } |
7654 | #endif | |
7655 | ||
340175b7 JL |
7656 | void zone_pcp_reset(struct zone *zone) |
7657 | { | |
7658 | unsigned long flags; | |
5a883813 MK |
7659 | int cpu; |
7660 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7661 | |
7662 | /* avoid races with drain_pages() */ | |
7663 | local_irq_save(flags); | |
7664 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7665 | for_each_online_cpu(cpu) { |
7666 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7667 | drain_zonestat(zone, pset); | |
7668 | } | |
340175b7 JL |
7669 | free_percpu(zone->pageset); |
7670 | zone->pageset = &boot_pageset; | |
7671 | } | |
7672 | local_irq_restore(flags); | |
7673 | } | |
7674 | ||
6dcd73d7 | 7675 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 7676 | /* |
b9eb6319 JK |
7677 | * All pages in the range must be in a single zone and isolated |
7678 | * before calling this. | |
0c0e6195 KH |
7679 | */ |
7680 | void | |
7681 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7682 | { | |
7683 | struct page *page; | |
7684 | struct zone *zone; | |
7aeb09f9 | 7685 | unsigned int order, i; |
0c0e6195 KH |
7686 | unsigned long pfn; |
7687 | unsigned long flags; | |
7688 | /* find the first valid pfn */ | |
7689 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
7690 | if (pfn_valid(pfn)) | |
7691 | break; | |
7692 | if (pfn == end_pfn) | |
7693 | return; | |
2d070eab | 7694 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
7695 | zone = page_zone(pfn_to_page(pfn)); |
7696 | spin_lock_irqsave(&zone->lock, flags); | |
7697 | pfn = start_pfn; | |
7698 | while (pfn < end_pfn) { | |
7699 | if (!pfn_valid(pfn)) { | |
7700 | pfn++; | |
7701 | continue; | |
7702 | } | |
7703 | page = pfn_to_page(pfn); | |
b023f468 WC |
7704 | /* |
7705 | * The HWPoisoned page may be not in buddy system, and | |
7706 | * page_count() is not 0. | |
7707 | */ | |
7708 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
7709 | pfn++; | |
7710 | SetPageReserved(page); | |
7711 | continue; | |
7712 | } | |
7713 | ||
0c0e6195 KH |
7714 | BUG_ON(page_count(page)); |
7715 | BUG_ON(!PageBuddy(page)); | |
7716 | order = page_order(page); | |
7717 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
7718 | pr_info("remove from free list %lx %d %lx\n", |
7719 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
7720 | #endif |
7721 | list_del(&page->lru); | |
7722 | rmv_page_order(page); | |
7723 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
7724 | for (i = 0; i < (1 << order); i++) |
7725 | SetPageReserved((page+i)); | |
7726 | pfn += (1 << order); | |
7727 | } | |
7728 | spin_unlock_irqrestore(&zone->lock, flags); | |
7729 | } | |
7730 | #endif | |
8d22ba1b | 7731 | |
8d22ba1b WF |
7732 | bool is_free_buddy_page(struct page *page) |
7733 | { | |
7734 | struct zone *zone = page_zone(page); | |
7735 | unsigned long pfn = page_to_pfn(page); | |
7736 | unsigned long flags; | |
7aeb09f9 | 7737 | unsigned int order; |
8d22ba1b WF |
7738 | |
7739 | spin_lock_irqsave(&zone->lock, flags); | |
7740 | for (order = 0; order < MAX_ORDER; order++) { | |
7741 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
7742 | ||
7743 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
7744 | break; | |
7745 | } | |
7746 | spin_unlock_irqrestore(&zone->lock, flags); | |
7747 | ||
7748 | return order < MAX_ORDER; | |
7749 | } |