]>
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 | { | |
295 | pgdat->first_deferred_pfn = ULONG_MAX; | |
296 | } | |
297 | ||
298 | /* Returns true if the struct page for the pfn is uninitialised */ | |
0e1cc95b | 299 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 300 | { |
ef70b6f4 MG |
301 | int nid = early_pfn_to_nid(pfn); |
302 | ||
303 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
304 | return true; |
305 | ||
306 | return false; | |
307 | } | |
308 | ||
309 | /* | |
310 | * Returns false when the remaining initialisation should be deferred until | |
311 | * later in the boot cycle when it can be parallelised. | |
312 | */ | |
313 | static inline bool update_defer_init(pg_data_t *pgdat, | |
314 | unsigned long pfn, unsigned long zone_end, | |
315 | unsigned long *nr_initialised) | |
316 | { | |
987b3095 LZ |
317 | unsigned long max_initialise; |
318 | ||
3a80a7fa MG |
319 | /* Always populate low zones for address-contrained allocations */ |
320 | if (zone_end < pgdat_end_pfn(pgdat)) | |
321 | return true; | |
987b3095 LZ |
322 | /* |
323 | * Initialise at least 2G of a node but also take into account that | |
324 | * two large system hashes that can take up 1GB for 0.25TB/node. | |
325 | */ | |
326 | max_initialise = max(2UL << (30 - PAGE_SHIFT), | |
327 | (pgdat->node_spanned_pages >> 8)); | |
3a80a7fa | 328 | |
3a80a7fa | 329 | (*nr_initialised)++; |
987b3095 | 330 | if ((*nr_initialised > max_initialise) && |
3a80a7fa MG |
331 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
332 | pgdat->first_deferred_pfn = pfn; | |
333 | return false; | |
334 | } | |
335 | ||
336 | return true; | |
337 | } | |
338 | #else | |
339 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
340 | { | |
341 | } | |
342 | ||
343 | static inline bool early_page_uninitialised(unsigned long pfn) | |
344 | { | |
345 | return false; | |
346 | } | |
347 | ||
348 | static inline bool update_defer_init(pg_data_t *pgdat, | |
349 | unsigned long pfn, unsigned long zone_end, | |
350 | unsigned long *nr_initialised) | |
351 | { | |
352 | return true; | |
353 | } | |
354 | #endif | |
355 | ||
0b423ca2 MG |
356 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
357 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
358 | unsigned long pfn) | |
359 | { | |
360 | #ifdef CONFIG_SPARSEMEM | |
361 | return __pfn_to_section(pfn)->pageblock_flags; | |
362 | #else | |
363 | return page_zone(page)->pageblock_flags; | |
364 | #endif /* CONFIG_SPARSEMEM */ | |
365 | } | |
366 | ||
367 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
368 | { | |
369 | #ifdef CONFIG_SPARSEMEM | |
370 | pfn &= (PAGES_PER_SECTION-1); | |
371 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
372 | #else | |
373 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
374 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
375 | #endif /* CONFIG_SPARSEMEM */ | |
376 | } | |
377 | ||
378 | /** | |
379 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
380 | * @page: The page within the block of interest | |
381 | * @pfn: The target page frame number | |
382 | * @end_bitidx: The last bit of interest to retrieve | |
383 | * @mask: mask of bits that the caller is interested in | |
384 | * | |
385 | * Return: pageblock_bits flags | |
386 | */ | |
387 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
388 | unsigned long pfn, | |
389 | unsigned long end_bitidx, | |
390 | unsigned long mask) | |
391 | { | |
392 | unsigned long *bitmap; | |
393 | unsigned long bitidx, word_bitidx; | |
394 | unsigned long word; | |
395 | ||
396 | bitmap = get_pageblock_bitmap(page, pfn); | |
397 | bitidx = pfn_to_bitidx(page, pfn); | |
398 | word_bitidx = bitidx / BITS_PER_LONG; | |
399 | bitidx &= (BITS_PER_LONG-1); | |
400 | ||
401 | word = bitmap[word_bitidx]; | |
402 | bitidx += end_bitidx; | |
403 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
404 | } | |
405 | ||
406 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
407 | unsigned long end_bitidx, | |
408 | unsigned long mask) | |
409 | { | |
410 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
411 | } | |
412 | ||
413 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
414 | { | |
415 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
416 | } | |
417 | ||
418 | /** | |
419 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
420 | * @page: The page within the block of interest | |
421 | * @flags: The flags to set | |
422 | * @pfn: The target page frame number | |
423 | * @end_bitidx: The last bit of interest | |
424 | * @mask: mask of bits that the caller is interested in | |
425 | */ | |
426 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
427 | unsigned long pfn, | |
428 | unsigned long end_bitidx, | |
429 | unsigned long mask) | |
430 | { | |
431 | unsigned long *bitmap; | |
432 | unsigned long bitidx, word_bitidx; | |
433 | unsigned long old_word, word; | |
434 | ||
435 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
436 | ||
437 | bitmap = get_pageblock_bitmap(page, pfn); | |
438 | bitidx = pfn_to_bitidx(page, pfn); | |
439 | word_bitidx = bitidx / BITS_PER_LONG; | |
440 | bitidx &= (BITS_PER_LONG-1); | |
441 | ||
442 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
443 | ||
444 | bitidx += end_bitidx; | |
445 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
446 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
447 | ||
448 | word = READ_ONCE(bitmap[word_bitidx]); | |
449 | for (;;) { | |
450 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
451 | if (word == old_word) | |
452 | break; | |
453 | word = old_word; | |
454 | } | |
455 | } | |
3a80a7fa | 456 | |
ee6f509c | 457 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 458 | { |
5d0f3f72 KM |
459 | if (unlikely(page_group_by_mobility_disabled && |
460 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
461 | migratetype = MIGRATE_UNMOVABLE; |
462 | ||
b2a0ac88 MG |
463 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
464 | PB_migrate, PB_migrate_end); | |
465 | } | |
466 | ||
13e7444b | 467 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 468 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 469 | { |
bdc8cb98 DH |
470 | int ret = 0; |
471 | unsigned seq; | |
472 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 473 | unsigned long sp, start_pfn; |
c6a57e19 | 474 | |
bdc8cb98 DH |
475 | do { |
476 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
477 | start_pfn = zone->zone_start_pfn; |
478 | sp = zone->spanned_pages; | |
108bcc96 | 479 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
480 | ret = 1; |
481 | } while (zone_span_seqretry(zone, seq)); | |
482 | ||
b5e6a5a2 | 483 | if (ret) |
613813e8 DH |
484 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
485 | pfn, zone_to_nid(zone), zone->name, | |
486 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 487 | |
bdc8cb98 | 488 | return ret; |
c6a57e19 DH |
489 | } |
490 | ||
491 | static int page_is_consistent(struct zone *zone, struct page *page) | |
492 | { | |
14e07298 | 493 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 494 | return 0; |
1da177e4 | 495 | if (zone != page_zone(page)) |
c6a57e19 DH |
496 | return 0; |
497 | ||
498 | return 1; | |
499 | } | |
500 | /* | |
501 | * Temporary debugging check for pages not lying within a given zone. | |
502 | */ | |
503 | static int bad_range(struct zone *zone, struct page *page) | |
504 | { | |
505 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 506 | return 1; |
c6a57e19 DH |
507 | if (!page_is_consistent(zone, page)) |
508 | return 1; | |
509 | ||
1da177e4 LT |
510 | return 0; |
511 | } | |
13e7444b NP |
512 | #else |
513 | static inline int bad_range(struct zone *zone, struct page *page) | |
514 | { | |
515 | return 0; | |
516 | } | |
517 | #endif | |
518 | ||
d230dec1 KS |
519 | static void bad_page(struct page *page, const char *reason, |
520 | unsigned long bad_flags) | |
1da177e4 | 521 | { |
d936cf9b HD |
522 | static unsigned long resume; |
523 | static unsigned long nr_shown; | |
524 | static unsigned long nr_unshown; | |
525 | ||
526 | /* | |
527 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
528 | * or allow a steady drip of one report per second. | |
529 | */ | |
530 | if (nr_shown == 60) { | |
531 | if (time_before(jiffies, resume)) { | |
532 | nr_unshown++; | |
533 | goto out; | |
534 | } | |
535 | if (nr_unshown) { | |
ff8e8116 | 536 | pr_alert( |
1e9e6365 | 537 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
538 | nr_unshown); |
539 | nr_unshown = 0; | |
540 | } | |
541 | nr_shown = 0; | |
542 | } | |
543 | if (nr_shown++ == 0) | |
544 | resume = jiffies + 60 * HZ; | |
545 | ||
ff8e8116 | 546 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 547 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
548 | __dump_page(page, reason); |
549 | bad_flags &= page->flags; | |
550 | if (bad_flags) | |
551 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
552 | bad_flags, &bad_flags); | |
4e462112 | 553 | dump_page_owner(page); |
3dc14741 | 554 | |
4f31888c | 555 | print_modules(); |
1da177e4 | 556 | dump_stack(); |
d936cf9b | 557 | out: |
8cc3b392 | 558 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 559 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 560 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
561 | } |
562 | ||
1da177e4 LT |
563 | /* |
564 | * Higher-order pages are called "compound pages". They are structured thusly: | |
565 | * | |
1d798ca3 | 566 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 567 | * |
1d798ca3 KS |
568 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
569 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 570 | * |
1d798ca3 KS |
571 | * The first tail page's ->compound_dtor holds the offset in array of compound |
572 | * page destructors. See compound_page_dtors. | |
1da177e4 | 573 | * |
1d798ca3 | 574 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 575 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 576 | */ |
d98c7a09 | 577 | |
9a982250 | 578 | void free_compound_page(struct page *page) |
d98c7a09 | 579 | { |
d85f3385 | 580 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
581 | } |
582 | ||
d00181b9 | 583 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
584 | { |
585 | int i; | |
586 | int nr_pages = 1 << order; | |
587 | ||
f1e61557 | 588 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
589 | set_compound_order(page, order); |
590 | __SetPageHead(page); | |
591 | for (i = 1; i < nr_pages; i++) { | |
592 | struct page *p = page + i; | |
58a84aa9 | 593 | set_page_count(p, 0); |
1c290f64 | 594 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 595 | set_compound_head(p, page); |
18229df5 | 596 | } |
53f9263b | 597 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
598 | } |
599 | ||
c0a32fc5 SG |
600 | #ifdef CONFIG_DEBUG_PAGEALLOC |
601 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
602 | bool _debug_pagealloc_enabled __read_mostly |
603 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 604 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
605 | bool _debug_guardpage_enabled __read_mostly; |
606 | ||
031bc574 JK |
607 | static int __init early_debug_pagealloc(char *buf) |
608 | { | |
609 | if (!buf) | |
610 | return -EINVAL; | |
2a138dc7 | 611 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
031bc574 JK |
612 | } |
613 | early_param("debug_pagealloc", early_debug_pagealloc); | |
614 | ||
e30825f1 JK |
615 | static bool need_debug_guardpage(void) |
616 | { | |
031bc574 JK |
617 | /* If we don't use debug_pagealloc, we don't need guard page */ |
618 | if (!debug_pagealloc_enabled()) | |
619 | return false; | |
620 | ||
f1c1e9f7 JK |
621 | if (!debug_guardpage_minorder()) |
622 | return false; | |
623 | ||
e30825f1 JK |
624 | return true; |
625 | } | |
626 | ||
627 | static void init_debug_guardpage(void) | |
628 | { | |
031bc574 JK |
629 | if (!debug_pagealloc_enabled()) |
630 | return; | |
631 | ||
f1c1e9f7 JK |
632 | if (!debug_guardpage_minorder()) |
633 | return; | |
634 | ||
e30825f1 JK |
635 | _debug_guardpage_enabled = true; |
636 | } | |
637 | ||
638 | struct page_ext_operations debug_guardpage_ops = { | |
639 | .need = need_debug_guardpage, | |
640 | .init = init_debug_guardpage, | |
641 | }; | |
c0a32fc5 SG |
642 | |
643 | static int __init debug_guardpage_minorder_setup(char *buf) | |
644 | { | |
645 | unsigned long res; | |
646 | ||
647 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 648 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
649 | return 0; |
650 | } | |
651 | _debug_guardpage_minorder = res; | |
1170532b | 652 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
653 | return 0; |
654 | } | |
f1c1e9f7 | 655 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 656 | |
acbc15a4 | 657 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 658 | unsigned int order, int migratetype) |
c0a32fc5 | 659 | { |
e30825f1 JK |
660 | struct page_ext *page_ext; |
661 | ||
662 | if (!debug_guardpage_enabled()) | |
acbc15a4 JK |
663 | return false; |
664 | ||
665 | if (order >= debug_guardpage_minorder()) | |
666 | return false; | |
e30825f1 JK |
667 | |
668 | page_ext = lookup_page_ext(page); | |
f86e4271 | 669 | if (unlikely(!page_ext)) |
acbc15a4 | 670 | return false; |
f86e4271 | 671 | |
e30825f1 JK |
672 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
673 | ||
2847cf95 JK |
674 | INIT_LIST_HEAD(&page->lru); |
675 | set_page_private(page, order); | |
676 | /* Guard pages are not available for any usage */ | |
677 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
678 | |
679 | return true; | |
c0a32fc5 SG |
680 | } |
681 | ||
2847cf95 JK |
682 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
683 | unsigned int order, int migratetype) | |
c0a32fc5 | 684 | { |
e30825f1 JK |
685 | struct page_ext *page_ext; |
686 | ||
687 | if (!debug_guardpage_enabled()) | |
688 | return; | |
689 | ||
690 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
691 | if (unlikely(!page_ext)) |
692 | return; | |
693 | ||
e30825f1 JK |
694 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
695 | ||
2847cf95 JK |
696 | set_page_private(page, 0); |
697 | if (!is_migrate_isolate(migratetype)) | |
698 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
699 | } |
700 | #else | |
980ac167 | 701 | struct page_ext_operations debug_guardpage_ops; |
acbc15a4 JK |
702 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
703 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
704 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
705 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
706 | #endif |
707 | ||
7aeb09f9 | 708 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 709 | { |
4c21e2f2 | 710 | set_page_private(page, order); |
676165a8 | 711 | __SetPageBuddy(page); |
1da177e4 LT |
712 | } |
713 | ||
714 | static inline void rmv_page_order(struct page *page) | |
715 | { | |
676165a8 | 716 | __ClearPageBuddy(page); |
4c21e2f2 | 717 | set_page_private(page, 0); |
1da177e4 LT |
718 | } |
719 | ||
1da177e4 LT |
720 | /* |
721 | * This function checks whether a page is free && is the buddy | |
722 | * we can do coalesce a page and its buddy if | |
13ad59df | 723 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 724 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
725 | * (c) a page and its buddy have the same order && |
726 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 727 | * |
cf6fe945 WSH |
728 | * For recording whether a page is in the buddy system, we set ->_mapcount |
729 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
730 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
731 | * serialized by zone->lock. | |
1da177e4 | 732 | * |
676165a8 | 733 | * For recording page's order, we use page_private(page). |
1da177e4 | 734 | */ |
cb2b95e1 | 735 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 736 | unsigned int order) |
1da177e4 | 737 | { |
c0a32fc5 | 738 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
739 | if (page_zone_id(page) != page_zone_id(buddy)) |
740 | return 0; | |
741 | ||
4c5018ce WY |
742 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
743 | ||
c0a32fc5 SG |
744 | return 1; |
745 | } | |
746 | ||
cb2b95e1 | 747 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
748 | /* |
749 | * zone check is done late to avoid uselessly | |
750 | * calculating zone/node ids for pages that could | |
751 | * never merge. | |
752 | */ | |
753 | if (page_zone_id(page) != page_zone_id(buddy)) | |
754 | return 0; | |
755 | ||
4c5018ce WY |
756 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
757 | ||
6aa3001b | 758 | return 1; |
676165a8 | 759 | } |
6aa3001b | 760 | return 0; |
1da177e4 LT |
761 | } |
762 | ||
763 | /* | |
764 | * Freeing function for a buddy system allocator. | |
765 | * | |
766 | * The concept of a buddy system is to maintain direct-mapped table | |
767 | * (containing bit values) for memory blocks of various "orders". | |
768 | * The bottom level table contains the map for the smallest allocatable | |
769 | * units of memory (here, pages), and each level above it describes | |
770 | * pairs of units from the levels below, hence, "buddies". | |
771 | * At a high level, all that happens here is marking the table entry | |
772 | * at the bottom level available, and propagating the changes upward | |
773 | * as necessary, plus some accounting needed to play nicely with other | |
774 | * parts of the VM system. | |
775 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
776 | * free pages of length of (1 << order) and marked with _mapcount |
777 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
778 | * field. | |
1da177e4 | 779 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
780 | * other. That is, if we allocate a small block, and both were |
781 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 782 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 783 | * triggers coalescing into a block of larger size. |
1da177e4 | 784 | * |
6d49e352 | 785 | * -- nyc |
1da177e4 LT |
786 | */ |
787 | ||
48db57f8 | 788 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 789 | unsigned long pfn, |
ed0ae21d MG |
790 | struct zone *zone, unsigned int order, |
791 | int migratetype) | |
1da177e4 | 792 | { |
76741e77 VB |
793 | unsigned long combined_pfn; |
794 | unsigned long uninitialized_var(buddy_pfn); | |
6dda9d55 | 795 | struct page *buddy; |
d9dddbf5 VB |
796 | unsigned int max_order; |
797 | ||
798 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 799 | |
d29bb978 | 800 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 801 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 802 | |
ed0ae21d | 803 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 804 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 805 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 806 | |
76741e77 | 807 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 808 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 809 | |
d9dddbf5 | 810 | continue_merging: |
3c605096 | 811 | while (order < max_order - 1) { |
76741e77 VB |
812 | buddy_pfn = __find_buddy_pfn(pfn, order); |
813 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
814 | |
815 | if (!pfn_valid_within(buddy_pfn)) | |
816 | goto done_merging; | |
cb2b95e1 | 817 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 818 | goto done_merging; |
c0a32fc5 SG |
819 | /* |
820 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
821 | * merge with it and move up one order. | |
822 | */ | |
823 | if (page_is_guard(buddy)) { | |
2847cf95 | 824 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
825 | } else { |
826 | list_del(&buddy->lru); | |
827 | zone->free_area[order].nr_free--; | |
828 | rmv_page_order(buddy); | |
829 | } | |
76741e77 VB |
830 | combined_pfn = buddy_pfn & pfn; |
831 | page = page + (combined_pfn - pfn); | |
832 | pfn = combined_pfn; | |
1da177e4 LT |
833 | order++; |
834 | } | |
d9dddbf5 VB |
835 | if (max_order < MAX_ORDER) { |
836 | /* If we are here, it means order is >= pageblock_order. | |
837 | * We want to prevent merge between freepages on isolate | |
838 | * pageblock and normal pageblock. Without this, pageblock | |
839 | * isolation could cause incorrect freepage or CMA accounting. | |
840 | * | |
841 | * We don't want to hit this code for the more frequent | |
842 | * low-order merging. | |
843 | */ | |
844 | if (unlikely(has_isolate_pageblock(zone))) { | |
845 | int buddy_mt; | |
846 | ||
76741e77 VB |
847 | buddy_pfn = __find_buddy_pfn(pfn, order); |
848 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
849 | buddy_mt = get_pageblock_migratetype(buddy); |
850 | ||
851 | if (migratetype != buddy_mt | |
852 | && (is_migrate_isolate(migratetype) || | |
853 | is_migrate_isolate(buddy_mt))) | |
854 | goto done_merging; | |
855 | } | |
856 | max_order++; | |
857 | goto continue_merging; | |
858 | } | |
859 | ||
860 | done_merging: | |
1da177e4 | 861 | set_page_order(page, order); |
6dda9d55 CZ |
862 | |
863 | /* | |
864 | * If this is not the largest possible page, check if the buddy | |
865 | * of the next-highest order is free. If it is, it's possible | |
866 | * that pages are being freed that will coalesce soon. In case, | |
867 | * that is happening, add the free page to the tail of the list | |
868 | * so it's less likely to be used soon and more likely to be merged | |
869 | * as a higher order page | |
870 | */ | |
13ad59df | 871 | if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) { |
6dda9d55 | 872 | struct page *higher_page, *higher_buddy; |
76741e77 VB |
873 | combined_pfn = buddy_pfn & pfn; |
874 | higher_page = page + (combined_pfn - pfn); | |
875 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
876 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
b4fb8f66 TL |
877 | if (pfn_valid_within(buddy_pfn) && |
878 | page_is_buddy(higher_page, higher_buddy, order + 1)) { | |
6dda9d55 CZ |
879 | list_add_tail(&page->lru, |
880 | &zone->free_area[order].free_list[migratetype]); | |
881 | goto out; | |
882 | } | |
883 | } | |
884 | ||
885 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
886 | out: | |
1da177e4 LT |
887 | zone->free_area[order].nr_free++; |
888 | } | |
889 | ||
7bfec6f4 MG |
890 | /* |
891 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
892 | * try and check multiple fields with one check. The caller must do a detailed | |
893 | * check if necessary. | |
894 | */ | |
895 | static inline bool page_expected_state(struct page *page, | |
896 | unsigned long check_flags) | |
897 | { | |
898 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
899 | return false; | |
900 | ||
901 | if (unlikely((unsigned long)page->mapping | | |
902 | page_ref_count(page) | | |
903 | #ifdef CONFIG_MEMCG | |
904 | (unsigned long)page->mem_cgroup | | |
905 | #endif | |
906 | (page->flags & check_flags))) | |
907 | return false; | |
908 | ||
909 | return true; | |
910 | } | |
911 | ||
bb552ac6 | 912 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 913 | { |
7bfec6f4 MG |
914 | const char *bad_reason; |
915 | unsigned long bad_flags; | |
916 | ||
7bfec6f4 MG |
917 | bad_reason = NULL; |
918 | bad_flags = 0; | |
f0b791a3 | 919 | |
53f9263b | 920 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
921 | bad_reason = "nonzero mapcount"; |
922 | if (unlikely(page->mapping != NULL)) | |
923 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 924 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 925 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
926 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
927 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
928 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
929 | } | |
9edad6ea JW |
930 | #ifdef CONFIG_MEMCG |
931 | if (unlikely(page->mem_cgroup)) | |
932 | bad_reason = "page still charged to cgroup"; | |
933 | #endif | |
7bfec6f4 | 934 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
935 | } |
936 | ||
937 | static inline int free_pages_check(struct page *page) | |
938 | { | |
da838d4f | 939 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 940 | return 0; |
bb552ac6 MG |
941 | |
942 | /* Something has gone sideways, find it */ | |
943 | free_pages_check_bad(page); | |
7bfec6f4 | 944 | return 1; |
1da177e4 LT |
945 | } |
946 | ||
4db7548c MG |
947 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
948 | { | |
949 | int ret = 1; | |
950 | ||
951 | /* | |
952 | * We rely page->lru.next never has bit 0 set, unless the page | |
953 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
954 | */ | |
955 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
956 | ||
957 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
958 | ret = 0; | |
959 | goto out; | |
960 | } | |
961 | switch (page - head_page) { | |
962 | case 1: | |
963 | /* the first tail page: ->mapping is compound_mapcount() */ | |
964 | if (unlikely(compound_mapcount(page))) { | |
965 | bad_page(page, "nonzero compound_mapcount", 0); | |
966 | goto out; | |
967 | } | |
968 | break; | |
969 | case 2: | |
970 | /* | |
971 | * the second tail page: ->mapping is | |
972 | * page_deferred_list().next -- ignore value. | |
973 | */ | |
974 | break; | |
975 | default: | |
976 | if (page->mapping != TAIL_MAPPING) { | |
977 | bad_page(page, "corrupted mapping in tail page", 0); | |
978 | goto out; | |
979 | } | |
980 | break; | |
981 | } | |
982 | if (unlikely(!PageTail(page))) { | |
983 | bad_page(page, "PageTail not set", 0); | |
984 | goto out; | |
985 | } | |
986 | if (unlikely(compound_head(page) != head_page)) { | |
987 | bad_page(page, "compound_head not consistent", 0); | |
988 | goto out; | |
989 | } | |
990 | ret = 0; | |
991 | out: | |
992 | page->mapping = NULL; | |
993 | clear_compound_head(page); | |
994 | return ret; | |
995 | } | |
996 | ||
e2769dbd MG |
997 | static __always_inline bool free_pages_prepare(struct page *page, |
998 | unsigned int order, bool check_free) | |
4db7548c | 999 | { |
e2769dbd | 1000 | int bad = 0; |
4db7548c | 1001 | |
4db7548c MG |
1002 | VM_BUG_ON_PAGE(PageTail(page), page); |
1003 | ||
e2769dbd MG |
1004 | trace_mm_page_free(page, order); |
1005 | kmemcheck_free_shadow(page, order); | |
e2769dbd MG |
1006 | |
1007 | /* | |
1008 | * Check tail pages before head page information is cleared to | |
1009 | * avoid checking PageCompound for order-0 pages. | |
1010 | */ | |
1011 | if (unlikely(order)) { | |
1012 | bool compound = PageCompound(page); | |
1013 | int i; | |
1014 | ||
1015 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1016 | |
9a73f61b KS |
1017 | if (compound) |
1018 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1019 | for (i = 1; i < (1 << order); i++) { |
1020 | if (compound) | |
1021 | bad += free_tail_pages_check(page, page + i); | |
1022 | if (unlikely(free_pages_check(page + i))) { | |
1023 | bad++; | |
1024 | continue; | |
1025 | } | |
1026 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1027 | } | |
1028 | } | |
bda807d4 | 1029 | if (PageMappingFlags(page)) |
4db7548c | 1030 | page->mapping = NULL; |
c4159a75 | 1031 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
4949148a | 1032 | memcg_kmem_uncharge(page, order); |
e2769dbd MG |
1033 | if (check_free) |
1034 | bad += free_pages_check(page); | |
1035 | if (bad) | |
1036 | return false; | |
4db7548c | 1037 | |
e2769dbd MG |
1038 | page_cpupid_reset_last(page); |
1039 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1040 | reset_page_owner(page, order); | |
4db7548c MG |
1041 | |
1042 | if (!PageHighMem(page)) { | |
1043 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1044 | PAGE_SIZE << order); |
4db7548c | 1045 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1046 | PAGE_SIZE << order); |
4db7548c | 1047 | } |
e2769dbd MG |
1048 | arch_free_page(page, order); |
1049 | kernel_poison_pages(page, 1 << order, 0); | |
1050 | kernel_map_pages(page, 1 << order, 0); | |
29b52de1 | 1051 | kasan_free_pages(page, order); |
4db7548c | 1052 | |
4db7548c MG |
1053 | return true; |
1054 | } | |
1055 | ||
e2769dbd MG |
1056 | #ifdef CONFIG_DEBUG_VM |
1057 | static inline bool free_pcp_prepare(struct page *page) | |
1058 | { | |
1059 | return free_pages_prepare(page, 0, true); | |
1060 | } | |
1061 | ||
1062 | static inline bool bulkfree_pcp_prepare(struct page *page) | |
1063 | { | |
1064 | return false; | |
1065 | } | |
1066 | #else | |
1067 | static bool free_pcp_prepare(struct page *page) | |
1068 | { | |
1069 | return free_pages_prepare(page, 0, false); | |
1070 | } | |
1071 | ||
4db7548c MG |
1072 | static bool bulkfree_pcp_prepare(struct page *page) |
1073 | { | |
1074 | return free_pages_check(page); | |
1075 | } | |
1076 | #endif /* CONFIG_DEBUG_VM */ | |
1077 | ||
1da177e4 | 1078 | /* |
5f8dcc21 | 1079 | * Frees a number of pages from the PCP lists |
1da177e4 | 1080 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1081 | * count is the number of pages to free. |
1da177e4 LT |
1082 | * |
1083 | * If the zone was previously in an "all pages pinned" state then look to | |
1084 | * see if this freeing clears that state. | |
1085 | * | |
1086 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1087 | * pinned" detection logic. | |
1088 | */ | |
5f8dcc21 MG |
1089 | static void free_pcppages_bulk(struct zone *zone, int count, |
1090 | struct per_cpu_pages *pcp) | |
1da177e4 | 1091 | { |
5f8dcc21 | 1092 | int migratetype = 0; |
a6f9edd6 | 1093 | int batch_free = 0; |
3777999d | 1094 | bool isolated_pageblocks; |
5f8dcc21 | 1095 | |
d34b0733 | 1096 | spin_lock(&zone->lock); |
3777999d | 1097 | isolated_pageblocks = has_isolate_pageblock(zone); |
f2260e6b | 1098 | |
e5b31ac2 | 1099 | while (count) { |
48db57f8 | 1100 | struct page *page; |
5f8dcc21 MG |
1101 | struct list_head *list; |
1102 | ||
1103 | /* | |
a6f9edd6 MG |
1104 | * Remove pages from lists in a round-robin fashion. A |
1105 | * batch_free count is maintained that is incremented when an | |
1106 | * empty list is encountered. This is so more pages are freed | |
1107 | * off fuller lists instead of spinning excessively around empty | |
1108 | * lists | |
5f8dcc21 MG |
1109 | */ |
1110 | do { | |
a6f9edd6 | 1111 | batch_free++; |
5f8dcc21 MG |
1112 | if (++migratetype == MIGRATE_PCPTYPES) |
1113 | migratetype = 0; | |
1114 | list = &pcp->lists[migratetype]; | |
1115 | } while (list_empty(list)); | |
48db57f8 | 1116 | |
1d16871d NK |
1117 | /* This is the only non-empty list. Free them all. */ |
1118 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1119 | batch_free = count; |
1d16871d | 1120 | |
a6f9edd6 | 1121 | do { |
770c8aaa BZ |
1122 | int mt; /* migratetype of the to-be-freed page */ |
1123 | ||
a16601c5 | 1124 | page = list_last_entry(list, struct page, lru); |
a6f9edd6 MG |
1125 | /* must delete as __free_one_page list manipulates */ |
1126 | list_del(&page->lru); | |
aa016d14 | 1127 | |
bb14c2c7 | 1128 | mt = get_pcppage_migratetype(page); |
aa016d14 VB |
1129 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1130 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1131 | /* Pageblock could have been isolated meanwhile */ | |
3777999d | 1132 | if (unlikely(isolated_pageblocks)) |
51bb1a40 | 1133 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 1134 | |
4db7548c MG |
1135 | if (bulkfree_pcp_prepare(page)) |
1136 | continue; | |
1137 | ||
dc4b0caf | 1138 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 1139 | trace_mm_page_pcpu_drain(page, 0, mt); |
e5b31ac2 | 1140 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1141 | } |
d34b0733 | 1142 | spin_unlock(&zone->lock); |
1da177e4 LT |
1143 | } |
1144 | ||
dc4b0caf MG |
1145 | static void free_one_page(struct zone *zone, |
1146 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1147 | unsigned int order, |
ed0ae21d | 1148 | int migratetype) |
1da177e4 | 1149 | { |
d34b0733 | 1150 | spin_lock(&zone->lock); |
ad53f92e JK |
1151 | if (unlikely(has_isolate_pageblock(zone) || |
1152 | is_migrate_isolate(migratetype))) { | |
1153 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1154 | } |
dc4b0caf | 1155 | __free_one_page(page, pfn, zone, order, migratetype); |
d34b0733 | 1156 | spin_unlock(&zone->lock); |
48db57f8 NP |
1157 | } |
1158 | ||
1e8ce83c RH |
1159 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
1160 | unsigned long zone, int nid) | |
1161 | { | |
1e8ce83c | 1162 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1163 | init_page_count(page); |
1164 | page_mapcount_reset(page); | |
1165 | page_cpupid_reset_last(page); | |
1e8ce83c | 1166 | |
1e8ce83c RH |
1167 | INIT_LIST_HEAD(&page->lru); |
1168 | #ifdef WANT_PAGE_VIRTUAL | |
1169 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1170 | if (!is_highmem_idx(zone)) | |
1171 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1172 | #endif | |
1173 | } | |
1174 | ||
1175 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
1176 | int nid) | |
1177 | { | |
1178 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
1179 | } | |
1180 | ||
7e18adb4 MG |
1181 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1182 | static void init_reserved_page(unsigned long pfn) | |
1183 | { | |
1184 | pg_data_t *pgdat; | |
1185 | int nid, zid; | |
1186 | ||
1187 | if (!early_page_uninitialised(pfn)) | |
1188 | return; | |
1189 | ||
1190 | nid = early_pfn_to_nid(pfn); | |
1191 | pgdat = NODE_DATA(nid); | |
1192 | ||
1193 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1194 | struct zone *zone = &pgdat->node_zones[zid]; | |
1195 | ||
1196 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1197 | break; | |
1198 | } | |
1199 | __init_single_pfn(pfn, zid, nid); | |
1200 | } | |
1201 | #else | |
1202 | static inline void init_reserved_page(unsigned long pfn) | |
1203 | { | |
1204 | } | |
1205 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1206 | ||
92923ca3 NZ |
1207 | /* |
1208 | * Initialised pages do not have PageReserved set. This function is | |
1209 | * called for each range allocated by the bootmem allocator and | |
1210 | * marks the pages PageReserved. The remaining valid pages are later | |
1211 | * sent to the buddy page allocator. | |
1212 | */ | |
4b50bcc7 | 1213 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1214 | { |
1215 | unsigned long start_pfn = PFN_DOWN(start); | |
1216 | unsigned long end_pfn = PFN_UP(end); | |
1217 | ||
7e18adb4 MG |
1218 | for (; start_pfn < end_pfn; start_pfn++) { |
1219 | if (pfn_valid(start_pfn)) { | |
1220 | struct page *page = pfn_to_page(start_pfn); | |
1221 | ||
1222 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1223 | |
1224 | /* Avoid false-positive PageTail() */ | |
1225 | INIT_LIST_HEAD(&page->lru); | |
1226 | ||
7e18adb4 MG |
1227 | SetPageReserved(page); |
1228 | } | |
1229 | } | |
92923ca3 NZ |
1230 | } |
1231 | ||
ec95f53a KM |
1232 | static void __free_pages_ok(struct page *page, unsigned int order) |
1233 | { | |
d34b0733 | 1234 | unsigned long flags; |
95e34412 | 1235 | int migratetype; |
dc4b0caf | 1236 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1237 | |
e2769dbd | 1238 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1239 | return; |
1240 | ||
cfc47a28 | 1241 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1242 | local_irq_save(flags); |
1243 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1244 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1245 | local_irq_restore(flags); |
1da177e4 LT |
1246 | } |
1247 | ||
949698a3 | 1248 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1249 | { |
c3993076 | 1250 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1251 | struct page *p = page; |
c3993076 | 1252 | unsigned int loop; |
a226f6c8 | 1253 | |
e2d0bd2b YL |
1254 | prefetchw(p); |
1255 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1256 | prefetchw(p + 1); | |
c3993076 JW |
1257 | __ClearPageReserved(p); |
1258 | set_page_count(p, 0); | |
a226f6c8 | 1259 | } |
e2d0bd2b YL |
1260 | __ClearPageReserved(p); |
1261 | set_page_count(p, 0); | |
c3993076 | 1262 | |
e2d0bd2b | 1263 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1264 | set_page_refcounted(page); |
1265 | __free_pages(page, order); | |
a226f6c8 DH |
1266 | } |
1267 | ||
75a592a4 MG |
1268 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1269 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1270 | |
75a592a4 MG |
1271 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1272 | ||
1273 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1274 | { | |
7ace9917 | 1275 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1276 | int nid; |
1277 | ||
7ace9917 | 1278 | spin_lock(&early_pfn_lock); |
75a592a4 | 1279 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1280 | if (nid < 0) |
e4568d38 | 1281 | nid = first_online_node; |
7ace9917 MG |
1282 | spin_unlock(&early_pfn_lock); |
1283 | ||
1284 | return nid; | |
75a592a4 MG |
1285 | } |
1286 | #endif | |
1287 | ||
1288 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
1289 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1290 | struct mminit_pfnnid_cache *state) | |
1291 | { | |
1292 | int nid; | |
1293 | ||
1294 | nid = __early_pfn_to_nid(pfn, state); | |
1295 | if (nid >= 0 && nid != node) | |
1296 | return false; | |
1297 | return true; | |
1298 | } | |
1299 | ||
1300 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1301 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1302 | { | |
1303 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1304 | } | |
1305 | ||
1306 | #else | |
1307 | ||
1308 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1309 | { | |
1310 | return true; | |
1311 | } | |
1312 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1313 | struct mminit_pfnnid_cache *state) | |
1314 | { | |
1315 | return true; | |
1316 | } | |
1317 | #endif | |
1318 | ||
1319 | ||
0e1cc95b | 1320 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1321 | unsigned int order) |
1322 | { | |
1323 | if (early_page_uninitialised(pfn)) | |
1324 | return; | |
949698a3 | 1325 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1326 | } |
1327 | ||
7cf91a98 JK |
1328 | /* |
1329 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1330 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1331 | * with the migration of free compaction scanner. The scanners then need to | |
1332 | * use only pfn_valid_within() check for arches that allow holes within | |
1333 | * pageblocks. | |
1334 | * | |
1335 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1336 | * | |
1337 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1338 | * i.e. it's possible that all pages within a zones range of pages do not | |
1339 | * belong to a single zone. We assume that a border between node0 and node1 | |
1340 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1341 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1342 | * the first and last page of a pageblock and avoid checking each individual | |
1343 | * page in a pageblock. | |
1344 | */ | |
1345 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1346 | unsigned long end_pfn, struct zone *zone) | |
1347 | { | |
1348 | struct page *start_page; | |
1349 | struct page *end_page; | |
1350 | ||
1351 | /* end_pfn is one past the range we are checking */ | |
1352 | end_pfn--; | |
1353 | ||
1354 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1355 | return NULL; | |
1356 | ||
1357 | start_page = pfn_to_page(start_pfn); | |
1358 | ||
1359 | if (page_zone(start_page) != zone) | |
1360 | return NULL; | |
1361 | ||
1362 | end_page = pfn_to_page(end_pfn); | |
1363 | ||
1364 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1365 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1366 | return NULL; | |
1367 | ||
1368 | return start_page; | |
1369 | } | |
1370 | ||
1371 | void set_zone_contiguous(struct zone *zone) | |
1372 | { | |
1373 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1374 | unsigned long block_end_pfn; | |
1375 | ||
1376 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1377 | for (; block_start_pfn < zone_end_pfn(zone); | |
1378 | block_start_pfn = block_end_pfn, | |
1379 | block_end_pfn += pageblock_nr_pages) { | |
1380 | ||
1381 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1382 | ||
1383 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1384 | block_end_pfn, zone)) | |
1385 | return; | |
1386 | } | |
1387 | ||
1388 | /* We confirm that there is no hole */ | |
1389 | zone->contiguous = true; | |
1390 | } | |
1391 | ||
1392 | void clear_zone_contiguous(struct zone *zone) | |
1393 | { | |
1394 | zone->contiguous = false; | |
1395 | } | |
1396 | ||
7e18adb4 | 1397 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
0e1cc95b | 1398 | static void __init deferred_free_range(struct page *page, |
a4de83dd MG |
1399 | unsigned long pfn, int nr_pages) |
1400 | { | |
1401 | int i; | |
1402 | ||
1403 | if (!page) | |
1404 | return; | |
1405 | ||
1406 | /* Free a large naturally-aligned chunk if possible */ | |
e780149b XQ |
1407 | if (nr_pages == pageblock_nr_pages && |
1408 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1409 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
e780149b | 1410 | __free_pages_boot_core(page, pageblock_order); |
a4de83dd MG |
1411 | return; |
1412 | } | |
1413 | ||
e780149b XQ |
1414 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1415 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1416 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
949698a3 | 1417 | __free_pages_boot_core(page, 0); |
e780149b | 1418 | } |
a4de83dd MG |
1419 | } |
1420 | ||
d3cd131d NS |
1421 | /* Completion tracking for deferred_init_memmap() threads */ |
1422 | static atomic_t pgdat_init_n_undone __initdata; | |
1423 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1424 | ||
1425 | static inline void __init pgdat_init_report_one_done(void) | |
1426 | { | |
1427 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1428 | complete(&pgdat_init_all_done_comp); | |
1429 | } | |
0e1cc95b | 1430 | |
7e18adb4 | 1431 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1432 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1433 | { |
0e1cc95b MG |
1434 | pg_data_t *pgdat = data; |
1435 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1436 | struct mminit_pfnnid_cache nid_init_state = { }; |
1437 | unsigned long start = jiffies; | |
1438 | unsigned long nr_pages = 0; | |
1439 | unsigned long walk_start, walk_end; | |
1440 | int i, zid; | |
1441 | struct zone *zone; | |
7e18adb4 | 1442 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
0e1cc95b | 1443 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
7e18adb4 | 1444 | |
0e1cc95b | 1445 | if (first_init_pfn == ULONG_MAX) { |
d3cd131d | 1446 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1447 | return 0; |
1448 | } | |
1449 | ||
1450 | /* Bind memory initialisation thread to a local node if possible */ | |
1451 | if (!cpumask_empty(cpumask)) | |
1452 | set_cpus_allowed_ptr(current, cpumask); | |
7e18adb4 MG |
1453 | |
1454 | /* Sanity check boundaries */ | |
1455 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1456 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1457 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1458 | ||
1459 | /* Only the highest zone is deferred so find it */ | |
1460 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1461 | zone = pgdat->node_zones + zid; | |
1462 | if (first_init_pfn < zone_end_pfn(zone)) | |
1463 | break; | |
1464 | } | |
1465 | ||
1466 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { | |
1467 | unsigned long pfn, end_pfn; | |
54608c3f | 1468 | struct page *page = NULL; |
a4de83dd MG |
1469 | struct page *free_base_page = NULL; |
1470 | unsigned long free_base_pfn = 0; | |
1471 | int nr_to_free = 0; | |
7e18adb4 MG |
1472 | |
1473 | end_pfn = min(walk_end, zone_end_pfn(zone)); | |
1474 | pfn = first_init_pfn; | |
1475 | if (pfn < walk_start) | |
1476 | pfn = walk_start; | |
1477 | if (pfn < zone->zone_start_pfn) | |
1478 | pfn = zone->zone_start_pfn; | |
1479 | ||
1480 | for (; pfn < end_pfn; pfn++) { | |
54608c3f | 1481 | if (!pfn_valid_within(pfn)) |
a4de83dd | 1482 | goto free_range; |
7e18adb4 | 1483 | |
54608c3f MG |
1484 | /* |
1485 | * Ensure pfn_valid is checked every | |
e780149b | 1486 | * pageblock_nr_pages for memory holes |
54608c3f | 1487 | */ |
e780149b | 1488 | if ((pfn & (pageblock_nr_pages - 1)) == 0) { |
54608c3f MG |
1489 | if (!pfn_valid(pfn)) { |
1490 | page = NULL; | |
a4de83dd | 1491 | goto free_range; |
54608c3f MG |
1492 | } |
1493 | } | |
1494 | ||
1495 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { | |
1496 | page = NULL; | |
a4de83dd | 1497 | goto free_range; |
54608c3f MG |
1498 | } |
1499 | ||
1500 | /* Minimise pfn page lookups and scheduler checks */ | |
e780149b | 1501 | if (page && (pfn & (pageblock_nr_pages - 1)) != 0) { |
54608c3f MG |
1502 | page++; |
1503 | } else { | |
a4de83dd MG |
1504 | nr_pages += nr_to_free; |
1505 | deferred_free_range(free_base_page, | |
1506 | free_base_pfn, nr_to_free); | |
1507 | free_base_page = NULL; | |
1508 | free_base_pfn = nr_to_free = 0; | |
1509 | ||
54608c3f MG |
1510 | page = pfn_to_page(pfn); |
1511 | cond_resched(); | |
1512 | } | |
7e18adb4 MG |
1513 | |
1514 | if (page->flags) { | |
1515 | VM_BUG_ON(page_zone(page) != zone); | |
a4de83dd | 1516 | goto free_range; |
7e18adb4 MG |
1517 | } |
1518 | ||
1519 | __init_single_page(page, pfn, zid, nid); | |
a4de83dd MG |
1520 | if (!free_base_page) { |
1521 | free_base_page = page; | |
1522 | free_base_pfn = pfn; | |
1523 | nr_to_free = 0; | |
1524 | } | |
1525 | nr_to_free++; | |
1526 | ||
1527 | /* Where possible, batch up pages for a single free */ | |
1528 | continue; | |
1529 | free_range: | |
1530 | /* Free the current block of pages to allocator */ | |
1531 | nr_pages += nr_to_free; | |
1532 | deferred_free_range(free_base_page, free_base_pfn, | |
1533 | nr_to_free); | |
1534 | free_base_page = NULL; | |
1535 | free_base_pfn = nr_to_free = 0; | |
7e18adb4 | 1536 | } |
e780149b XQ |
1537 | /* Free the last block of pages to allocator */ |
1538 | nr_pages += nr_to_free; | |
1539 | deferred_free_range(free_base_page, free_base_pfn, nr_to_free); | |
a4de83dd | 1540 | |
7e18adb4 MG |
1541 | first_init_pfn = max(end_pfn, first_init_pfn); |
1542 | } | |
1543 | ||
1544 | /* Sanity check that the next zone really is unpopulated */ | |
1545 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1546 | ||
0e1cc95b | 1547 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1548 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1549 | |
1550 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1551 | return 0; |
1552 | } | |
7cf91a98 | 1553 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1554 | |
1555 | void __init page_alloc_init_late(void) | |
1556 | { | |
7cf91a98 JK |
1557 | struct zone *zone; |
1558 | ||
1559 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1560 | int nid; |
1561 | ||
d3cd131d NS |
1562 | /* There will be num_node_state(N_MEMORY) threads */ |
1563 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1564 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1565 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1566 | } | |
1567 | ||
1568 | /* Block until all are initialised */ | |
d3cd131d | 1569 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da MG |
1570 | |
1571 | /* Reinit limits that are based on free pages after the kernel is up */ | |
1572 | files_maxfiles_init(); | |
7cf91a98 JK |
1573 | #endif |
1574 | ||
1575 | for_each_populated_zone(zone) | |
1576 | set_zone_contiguous(zone); | |
7e18adb4 | 1577 | } |
7e18adb4 | 1578 | |
47118af0 | 1579 | #ifdef CONFIG_CMA |
9cf510a5 | 1580 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1581 | void __init init_cma_reserved_pageblock(struct page *page) |
1582 | { | |
1583 | unsigned i = pageblock_nr_pages; | |
1584 | struct page *p = page; | |
1585 | ||
1586 | do { | |
1587 | __ClearPageReserved(p); | |
1588 | set_page_count(p, 0); | |
1589 | } while (++p, --i); | |
1590 | ||
47118af0 | 1591 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1592 | |
1593 | if (pageblock_order >= MAX_ORDER) { | |
1594 | i = pageblock_nr_pages; | |
1595 | p = page; | |
1596 | do { | |
1597 | set_page_refcounted(p); | |
1598 | __free_pages(p, MAX_ORDER - 1); | |
1599 | p += MAX_ORDER_NR_PAGES; | |
1600 | } while (i -= MAX_ORDER_NR_PAGES); | |
1601 | } else { | |
1602 | set_page_refcounted(page); | |
1603 | __free_pages(page, pageblock_order); | |
1604 | } | |
1605 | ||
3dcc0571 | 1606 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1607 | } |
1608 | #endif | |
1da177e4 LT |
1609 | |
1610 | /* | |
1611 | * The order of subdivision here is critical for the IO subsystem. | |
1612 | * Please do not alter this order without good reasons and regression | |
1613 | * testing. Specifically, as large blocks of memory are subdivided, | |
1614 | * the order in which smaller blocks are delivered depends on the order | |
1615 | * they're subdivided in this function. This is the primary factor | |
1616 | * influencing the order in which pages are delivered to the IO | |
1617 | * subsystem according to empirical testing, and this is also justified | |
1618 | * by considering the behavior of a buddy system containing a single | |
1619 | * large block of memory acted on by a series of small allocations. | |
1620 | * This behavior is a critical factor in sglist merging's success. | |
1621 | * | |
6d49e352 | 1622 | * -- nyc |
1da177e4 | 1623 | */ |
085cc7d5 | 1624 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1625 | int low, int high, struct free_area *area, |
1626 | int migratetype) | |
1da177e4 LT |
1627 | { |
1628 | unsigned long size = 1 << high; | |
1629 | ||
1630 | while (high > low) { | |
1631 | area--; | |
1632 | high--; | |
1633 | size >>= 1; | |
309381fe | 1634 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1635 | |
acbc15a4 JK |
1636 | /* |
1637 | * Mark as guard pages (or page), that will allow to | |
1638 | * merge back to allocator when buddy will be freed. | |
1639 | * Corresponding page table entries will not be touched, | |
1640 | * pages will stay not present in virtual address space | |
1641 | */ | |
1642 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 1643 | continue; |
acbc15a4 | 1644 | |
b2a0ac88 | 1645 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1646 | area->nr_free++; |
1647 | set_page_order(&page[size], high); | |
1648 | } | |
1da177e4 LT |
1649 | } |
1650 | ||
4e611801 | 1651 | static void check_new_page_bad(struct page *page) |
1da177e4 | 1652 | { |
4e611801 VB |
1653 | const char *bad_reason = NULL; |
1654 | unsigned long bad_flags = 0; | |
7bfec6f4 | 1655 | |
53f9263b | 1656 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1657 | bad_reason = "nonzero mapcount"; |
1658 | if (unlikely(page->mapping != NULL)) | |
1659 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1660 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1661 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1662 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1663 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1664 | bad_flags = __PG_HWPOISON; | |
e570f56c NH |
1665 | /* Don't complain about hwpoisoned pages */ |
1666 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1667 | return; | |
f4c18e6f | 1668 | } |
f0b791a3 DH |
1669 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1670 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1671 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1672 | } | |
9edad6ea JW |
1673 | #ifdef CONFIG_MEMCG |
1674 | if (unlikely(page->mem_cgroup)) | |
1675 | bad_reason = "page still charged to cgroup"; | |
1676 | #endif | |
4e611801 VB |
1677 | bad_page(page, bad_reason, bad_flags); |
1678 | } | |
1679 | ||
1680 | /* | |
1681 | * This page is about to be returned from the page allocator | |
1682 | */ | |
1683 | static inline int check_new_page(struct page *page) | |
1684 | { | |
1685 | if (likely(page_expected_state(page, | |
1686 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1687 | return 0; | |
1688 | ||
1689 | check_new_page_bad(page); | |
1690 | return 1; | |
2a7684a2 WF |
1691 | } |
1692 | ||
bd33ef36 | 1693 | static inline bool free_pages_prezeroed(void) |
1414c7f4 LA |
1694 | { |
1695 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
bd33ef36 | 1696 | page_poisoning_enabled(); |
1414c7f4 LA |
1697 | } |
1698 | ||
479f854a MG |
1699 | #ifdef CONFIG_DEBUG_VM |
1700 | static bool check_pcp_refill(struct page *page) | |
1701 | { | |
1702 | return false; | |
1703 | } | |
1704 | ||
1705 | static bool check_new_pcp(struct page *page) | |
1706 | { | |
1707 | return check_new_page(page); | |
1708 | } | |
1709 | #else | |
1710 | static bool check_pcp_refill(struct page *page) | |
1711 | { | |
1712 | return check_new_page(page); | |
1713 | } | |
1714 | static bool check_new_pcp(struct page *page) | |
1715 | { | |
1716 | return false; | |
1717 | } | |
1718 | #endif /* CONFIG_DEBUG_VM */ | |
1719 | ||
1720 | static bool check_new_pages(struct page *page, unsigned int order) | |
1721 | { | |
1722 | int i; | |
1723 | for (i = 0; i < (1 << order); i++) { | |
1724 | struct page *p = page + i; | |
1725 | ||
1726 | if (unlikely(check_new_page(p))) | |
1727 | return true; | |
1728 | } | |
1729 | ||
1730 | return false; | |
1731 | } | |
1732 | ||
46f24fd8 JK |
1733 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1734 | gfp_t gfp_flags) | |
1735 | { | |
1736 | set_page_private(page, 0); | |
1737 | set_page_refcounted(page); | |
1738 | ||
1739 | arch_alloc_page(page, order); | |
1740 | kernel_map_pages(page, 1 << order, 1); | |
1741 | kernel_poison_pages(page, 1 << order, 1); | |
1742 | kasan_alloc_pages(page, order); | |
1743 | set_page_owner(page, order, gfp_flags); | |
1744 | } | |
1745 | ||
479f854a | 1746 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1747 | unsigned int alloc_flags) |
2a7684a2 WF |
1748 | { |
1749 | int i; | |
689bcebf | 1750 | |
46f24fd8 | 1751 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1752 | |
bd33ef36 | 1753 | if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1754 | for (i = 0; i < (1 << order); i++) |
1755 | clear_highpage(page + i); | |
17cf4406 NP |
1756 | |
1757 | if (order && (gfp_flags & __GFP_COMP)) | |
1758 | prep_compound_page(page, order); | |
1759 | ||
75379191 | 1760 | /* |
2f064f34 | 1761 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1762 | * allocate the page. The expectation is that the caller is taking |
1763 | * steps that will free more memory. The caller should avoid the page | |
1764 | * being used for !PFMEMALLOC purposes. | |
1765 | */ | |
2f064f34 MH |
1766 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1767 | set_page_pfmemalloc(page); | |
1768 | else | |
1769 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1770 | } |
1771 | ||
56fd56b8 MG |
1772 | /* |
1773 | * Go through the free lists for the given migratetype and remove | |
1774 | * the smallest available page from the freelists | |
1775 | */ | |
728ec980 MG |
1776 | static inline |
1777 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1778 | int migratetype) |
1779 | { | |
1780 | unsigned int current_order; | |
b8af2941 | 1781 | struct free_area *area; |
56fd56b8 MG |
1782 | struct page *page; |
1783 | ||
1784 | /* Find a page of the appropriate size in the preferred list */ | |
1785 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1786 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1787 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1788 | struct page, lru); |
a16601c5 GT |
1789 | if (!page) |
1790 | continue; | |
56fd56b8 MG |
1791 | list_del(&page->lru); |
1792 | rmv_page_order(page); | |
1793 | area->nr_free--; | |
56fd56b8 | 1794 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1795 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1796 | return page; |
1797 | } | |
1798 | ||
1799 | return NULL; | |
1800 | } | |
1801 | ||
1802 | ||
b2a0ac88 MG |
1803 | /* |
1804 | * This array describes the order lists are fallen back to when | |
1805 | * the free lists for the desirable migrate type are depleted | |
1806 | */ | |
47118af0 | 1807 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1808 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1809 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1810 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1811 | #ifdef CONFIG_CMA |
974a786e | 1812 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1813 | #endif |
194159fb | 1814 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1815 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1816 | #endif |
b2a0ac88 MG |
1817 | }; |
1818 | ||
dc67647b JK |
1819 | #ifdef CONFIG_CMA |
1820 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1821 | unsigned int order) | |
1822 | { | |
1823 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1824 | } | |
1825 | #else | |
1826 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1827 | unsigned int order) { return NULL; } | |
1828 | #endif | |
1829 | ||
c361be55 MG |
1830 | /* |
1831 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1832 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1833 | * boundary. If alignment is required, use move_freepages_block() |
1834 | */ | |
02aa0cdd | 1835 | static int move_freepages(struct zone *zone, |
b69a7288 | 1836 | struct page *start_page, struct page *end_page, |
02aa0cdd | 1837 | int migratetype, int *num_movable) |
c361be55 MG |
1838 | { |
1839 | struct page *page; | |
d00181b9 | 1840 | unsigned int order; |
d100313f | 1841 | int pages_moved = 0; |
c361be55 MG |
1842 | |
1843 | #ifndef CONFIG_HOLES_IN_ZONE | |
1844 | /* | |
1845 | * page_zone is not safe to call in this context when | |
1846 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1847 | * anyway as we check zone boundaries in move_freepages_block(). | |
1848 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1849 | * grouping pages by mobility |
c361be55 | 1850 | */ |
97ee4ba7 | 1851 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1852 | #endif |
1853 | ||
02aa0cdd VB |
1854 | if (num_movable) |
1855 | *num_movable = 0; | |
1856 | ||
c361be55 MG |
1857 | for (page = start_page; page <= end_page;) { |
1858 | if (!pfn_valid_within(page_to_pfn(page))) { | |
1859 | page++; | |
1860 | continue; | |
1861 | } | |
1862 | ||
f073bdc5 AB |
1863 | /* Make sure we are not inadvertently changing nodes */ |
1864 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
1865 | ||
c361be55 | 1866 | if (!PageBuddy(page)) { |
02aa0cdd VB |
1867 | /* |
1868 | * We assume that pages that could be isolated for | |
1869 | * migration are movable. But we don't actually try | |
1870 | * isolating, as that would be expensive. | |
1871 | */ | |
1872 | if (num_movable && | |
1873 | (PageLRU(page) || __PageMovable(page))) | |
1874 | (*num_movable)++; | |
1875 | ||
c361be55 MG |
1876 | page++; |
1877 | continue; | |
1878 | } | |
1879 | ||
1880 | order = page_order(page); | |
84be48d8 KS |
1881 | list_move(&page->lru, |
1882 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 1883 | page += 1 << order; |
d100313f | 1884 | pages_moved += 1 << order; |
c361be55 MG |
1885 | } |
1886 | ||
d100313f | 1887 | return pages_moved; |
c361be55 MG |
1888 | } |
1889 | ||
ee6f509c | 1890 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 1891 | int migratetype, int *num_movable) |
c361be55 MG |
1892 | { |
1893 | unsigned long start_pfn, end_pfn; | |
1894 | struct page *start_page, *end_page; | |
1895 | ||
1896 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1897 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1898 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1899 | end_page = start_page + pageblock_nr_pages - 1; |
1900 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1901 | |
1902 | /* Do not cross zone boundaries */ | |
108bcc96 | 1903 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1904 | start_page = page; |
108bcc96 | 1905 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1906 | return 0; |
1907 | ||
02aa0cdd VB |
1908 | return move_freepages(zone, start_page, end_page, migratetype, |
1909 | num_movable); | |
c361be55 MG |
1910 | } |
1911 | ||
2f66a68f MG |
1912 | static void change_pageblock_range(struct page *pageblock_page, |
1913 | int start_order, int migratetype) | |
1914 | { | |
1915 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1916 | ||
1917 | while (nr_pageblocks--) { | |
1918 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1919 | pageblock_page += pageblock_nr_pages; | |
1920 | } | |
1921 | } | |
1922 | ||
fef903ef | 1923 | /* |
9c0415eb VB |
1924 | * When we are falling back to another migratetype during allocation, try to |
1925 | * steal extra free pages from the same pageblocks to satisfy further | |
1926 | * allocations, instead of polluting multiple pageblocks. | |
1927 | * | |
1928 | * If we are stealing a relatively large buddy page, it is likely there will | |
1929 | * be more free pages in the pageblock, so try to steal them all. For | |
1930 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1931 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1932 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1933 | * pageblocks. | |
fef903ef | 1934 | */ |
4eb7dce6 JK |
1935 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1936 | { | |
1937 | /* | |
1938 | * Leaving this order check is intended, although there is | |
1939 | * relaxed order check in next check. The reason is that | |
1940 | * we can actually steal whole pageblock if this condition met, | |
1941 | * but, below check doesn't guarantee it and that is just heuristic | |
1942 | * so could be changed anytime. | |
1943 | */ | |
1944 | if (order >= pageblock_order) | |
1945 | return true; | |
1946 | ||
1947 | if (order >= pageblock_order / 2 || | |
1948 | start_mt == MIGRATE_RECLAIMABLE || | |
1949 | start_mt == MIGRATE_UNMOVABLE || | |
1950 | page_group_by_mobility_disabled) | |
1951 | return true; | |
1952 | ||
1953 | return false; | |
1954 | } | |
1955 | ||
1956 | /* | |
1957 | * This function implements actual steal behaviour. If order is large enough, | |
1958 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
1959 | * pageblock to our migratetype and determine how many already-allocated pages |
1960 | * are there in the pageblock with a compatible migratetype. If at least half | |
1961 | * of pages are free or compatible, we can change migratetype of the pageblock | |
1962 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
1963 | */ |
1964 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
3bc48f96 | 1965 | int start_type, bool whole_block) |
fef903ef | 1966 | { |
d00181b9 | 1967 | unsigned int current_order = page_order(page); |
3bc48f96 | 1968 | struct free_area *area; |
02aa0cdd VB |
1969 | int free_pages, movable_pages, alike_pages; |
1970 | int old_block_type; | |
1971 | ||
1972 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 1973 | |
3bc48f96 VB |
1974 | /* |
1975 | * This can happen due to races and we want to prevent broken | |
1976 | * highatomic accounting. | |
1977 | */ | |
02aa0cdd | 1978 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
1979 | goto single_page; |
1980 | ||
fef903ef SB |
1981 | /* Take ownership for orders >= pageblock_order */ |
1982 | if (current_order >= pageblock_order) { | |
1983 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 1984 | goto single_page; |
fef903ef SB |
1985 | } |
1986 | ||
3bc48f96 VB |
1987 | /* We are not allowed to try stealing from the whole block */ |
1988 | if (!whole_block) | |
1989 | goto single_page; | |
1990 | ||
02aa0cdd VB |
1991 | free_pages = move_freepages_block(zone, page, start_type, |
1992 | &movable_pages); | |
1993 | /* | |
1994 | * Determine how many pages are compatible with our allocation. | |
1995 | * For movable allocation, it's the number of movable pages which | |
1996 | * we just obtained. For other types it's a bit more tricky. | |
1997 | */ | |
1998 | if (start_type == MIGRATE_MOVABLE) { | |
1999 | alike_pages = movable_pages; | |
2000 | } else { | |
2001 | /* | |
2002 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2003 | * to MOVABLE pageblock, consider all non-movable pages as | |
2004 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2005 | * vice versa, be conservative since we can't distinguish the | |
2006 | * exact migratetype of non-movable pages. | |
2007 | */ | |
2008 | if (old_block_type == MIGRATE_MOVABLE) | |
2009 | alike_pages = pageblock_nr_pages | |
2010 | - (free_pages + movable_pages); | |
2011 | else | |
2012 | alike_pages = 0; | |
2013 | } | |
2014 | ||
3bc48f96 | 2015 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2016 | if (!free_pages) |
3bc48f96 | 2017 | goto single_page; |
fef903ef | 2018 | |
02aa0cdd VB |
2019 | /* |
2020 | * If a sufficient number of pages in the block are either free or of | |
2021 | * comparable migratability as our allocation, claim the whole block. | |
2022 | */ | |
2023 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2024 | page_group_by_mobility_disabled) |
2025 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2026 | |
2027 | return; | |
2028 | ||
2029 | single_page: | |
2030 | area = &zone->free_area[current_order]; | |
2031 | list_move(&page->lru, &area->free_list[start_type]); | |
4eb7dce6 JK |
2032 | } |
2033 | ||
2149cdae JK |
2034 | /* |
2035 | * Check whether there is a suitable fallback freepage with requested order. | |
2036 | * If only_stealable is true, this function returns fallback_mt only if | |
2037 | * we can steal other freepages all together. This would help to reduce | |
2038 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2039 | */ | |
2040 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2041 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2042 | { |
2043 | int i; | |
2044 | int fallback_mt; | |
2045 | ||
2046 | if (area->nr_free == 0) | |
2047 | return -1; | |
2048 | ||
2049 | *can_steal = false; | |
2050 | for (i = 0;; i++) { | |
2051 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2052 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2053 | break; |
2054 | ||
2055 | if (list_empty(&area->free_list[fallback_mt])) | |
2056 | continue; | |
fef903ef | 2057 | |
4eb7dce6 JK |
2058 | if (can_steal_fallback(order, migratetype)) |
2059 | *can_steal = true; | |
2060 | ||
2149cdae JK |
2061 | if (!only_stealable) |
2062 | return fallback_mt; | |
2063 | ||
2064 | if (*can_steal) | |
2065 | return fallback_mt; | |
fef903ef | 2066 | } |
4eb7dce6 JK |
2067 | |
2068 | return -1; | |
fef903ef SB |
2069 | } |
2070 | ||
0aaa29a5 MG |
2071 | /* |
2072 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2073 | * there are no empty page blocks that contain a page with a suitable order | |
2074 | */ | |
2075 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2076 | unsigned int alloc_order) | |
2077 | { | |
2078 | int mt; | |
2079 | unsigned long max_managed, flags; | |
2080 | ||
2081 | /* | |
2082 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2083 | * Check is race-prone but harmless. | |
2084 | */ | |
2085 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
2086 | if (zone->nr_reserved_highatomic >= max_managed) | |
2087 | return; | |
2088 | ||
2089 | spin_lock_irqsave(&zone->lock, flags); | |
2090 | ||
2091 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2092 | if (zone->nr_reserved_highatomic >= max_managed) | |
2093 | goto out_unlock; | |
2094 | ||
2095 | /* Yoink! */ | |
2096 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2097 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2098 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2099 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2100 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2101 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2102 | } |
2103 | ||
2104 | out_unlock: | |
2105 | spin_unlock_irqrestore(&zone->lock, flags); | |
2106 | } | |
2107 | ||
2108 | /* | |
2109 | * Used when an allocation is about to fail under memory pressure. This | |
2110 | * potentially hurts the reliability of high-order allocations when under | |
2111 | * intense memory pressure but failed atomic allocations should be easier | |
2112 | * to recover from than an OOM. | |
29fac03b MK |
2113 | * |
2114 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2115 | * pageblock is exhausted. | |
0aaa29a5 | 2116 | */ |
29fac03b MK |
2117 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2118 | bool force) | |
0aaa29a5 MG |
2119 | { |
2120 | struct zonelist *zonelist = ac->zonelist; | |
2121 | unsigned long flags; | |
2122 | struct zoneref *z; | |
2123 | struct zone *zone; | |
2124 | struct page *page; | |
2125 | int order; | |
04c8716f | 2126 | bool ret; |
0aaa29a5 MG |
2127 | |
2128 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2129 | ac->nodemask) { | |
29fac03b MK |
2130 | /* |
2131 | * Preserve at least one pageblock unless memory pressure | |
2132 | * is really high. | |
2133 | */ | |
2134 | if (!force && zone->nr_reserved_highatomic <= | |
2135 | pageblock_nr_pages) | |
0aaa29a5 MG |
2136 | continue; |
2137 | ||
2138 | spin_lock_irqsave(&zone->lock, flags); | |
2139 | for (order = 0; order < MAX_ORDER; order++) { | |
2140 | struct free_area *area = &(zone->free_area[order]); | |
2141 | ||
a16601c5 GT |
2142 | page = list_first_entry_or_null( |
2143 | &area->free_list[MIGRATE_HIGHATOMIC], | |
2144 | struct page, lru); | |
2145 | if (!page) | |
0aaa29a5 MG |
2146 | continue; |
2147 | ||
0aaa29a5 | 2148 | /* |
4855e4a7 MK |
2149 | * In page freeing path, migratetype change is racy so |
2150 | * we can counter several free pages in a pageblock | |
2151 | * in this loop althoug we changed the pageblock type | |
2152 | * from highatomic to ac->migratetype. So we should | |
2153 | * adjust the count once. | |
0aaa29a5 | 2154 | */ |
a6ffdc07 | 2155 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2156 | /* |
2157 | * It should never happen but changes to | |
2158 | * locking could inadvertently allow a per-cpu | |
2159 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2160 | * while unreserving so be safe and watch for | |
2161 | * underflows. | |
2162 | */ | |
2163 | zone->nr_reserved_highatomic -= min( | |
2164 | pageblock_nr_pages, | |
2165 | zone->nr_reserved_highatomic); | |
2166 | } | |
0aaa29a5 MG |
2167 | |
2168 | /* | |
2169 | * Convert to ac->migratetype and avoid the normal | |
2170 | * pageblock stealing heuristics. Minimally, the caller | |
2171 | * is doing the work and needs the pages. More | |
2172 | * importantly, if the block was always converted to | |
2173 | * MIGRATE_UNMOVABLE or another type then the number | |
2174 | * of pageblocks that cannot be completely freed | |
2175 | * may increase. | |
2176 | */ | |
2177 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2178 | ret = move_freepages_block(zone, page, ac->migratetype, |
2179 | NULL); | |
29fac03b MK |
2180 | if (ret) { |
2181 | spin_unlock_irqrestore(&zone->lock, flags); | |
2182 | return ret; | |
2183 | } | |
0aaa29a5 MG |
2184 | } |
2185 | spin_unlock_irqrestore(&zone->lock, flags); | |
2186 | } | |
04c8716f MK |
2187 | |
2188 | return false; | |
0aaa29a5 MG |
2189 | } |
2190 | ||
3bc48f96 VB |
2191 | /* |
2192 | * Try finding a free buddy page on the fallback list and put it on the free | |
2193 | * list of requested migratetype, possibly along with other pages from the same | |
2194 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2195 | * fallback was found so that __rmqueue_smallest() can grab it. | |
2196 | */ | |
2197 | static inline bool | |
7aeb09f9 | 2198 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 2199 | { |
b8af2941 | 2200 | struct free_area *area; |
7aeb09f9 | 2201 | unsigned int current_order; |
b2a0ac88 | 2202 | struct page *page; |
4eb7dce6 JK |
2203 | int fallback_mt; |
2204 | bool can_steal; | |
b2a0ac88 MG |
2205 | |
2206 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
2207 | for (current_order = MAX_ORDER-1; |
2208 | current_order >= order && current_order <= MAX_ORDER-1; | |
2209 | --current_order) { | |
4eb7dce6 JK |
2210 | area = &(zone->free_area[current_order]); |
2211 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2212 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2213 | if (fallback_mt == -1) |
2214 | continue; | |
b2a0ac88 | 2215 | |
a16601c5 | 2216 | page = list_first_entry(&area->free_list[fallback_mt], |
4eb7dce6 | 2217 | struct page, lru); |
b2a0ac88 | 2218 | |
3bc48f96 VB |
2219 | steal_suitable_fallback(zone, page, start_migratetype, |
2220 | can_steal); | |
e0fff1bd | 2221 | |
4eb7dce6 JK |
2222 | trace_mm_page_alloc_extfrag(page, order, current_order, |
2223 | start_migratetype, fallback_mt); | |
e0fff1bd | 2224 | |
3bc48f96 | 2225 | return true; |
b2a0ac88 MG |
2226 | } |
2227 | ||
3bc48f96 | 2228 | return false; |
b2a0ac88 MG |
2229 | } |
2230 | ||
56fd56b8 | 2231 | /* |
1da177e4 LT |
2232 | * Do the hard work of removing an element from the buddy allocator. |
2233 | * Call me with the zone->lock already held. | |
2234 | */ | |
b2a0ac88 | 2235 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
6ac0206b | 2236 | int migratetype) |
1da177e4 | 2237 | { |
1da177e4 LT |
2238 | struct page *page; |
2239 | ||
3bc48f96 | 2240 | retry: |
56fd56b8 | 2241 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2242 | if (unlikely(!page)) { |
dc67647b JK |
2243 | if (migratetype == MIGRATE_MOVABLE) |
2244 | page = __rmqueue_cma_fallback(zone, order); | |
2245 | ||
3bc48f96 VB |
2246 | if (!page && __rmqueue_fallback(zone, order, migratetype)) |
2247 | goto retry; | |
728ec980 MG |
2248 | } |
2249 | ||
0d3d062a | 2250 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2251 | return page; |
1da177e4 LT |
2252 | } |
2253 | ||
5f63b720 | 2254 | /* |
1da177e4 LT |
2255 | * Obtain a specified number of elements from the buddy allocator, all under |
2256 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2257 | * Returns the number of new pages which were placed at *list. | |
2258 | */ | |
5f63b720 | 2259 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2260 | unsigned long count, struct list_head *list, |
b745bc85 | 2261 | int migratetype, bool cold) |
1da177e4 | 2262 | { |
a6de734b | 2263 | int i, alloced = 0; |
5f63b720 | 2264 | |
d34b0733 | 2265 | spin_lock(&zone->lock); |
1da177e4 | 2266 | for (i = 0; i < count; ++i) { |
6ac0206b | 2267 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 2268 | if (unlikely(page == NULL)) |
1da177e4 | 2269 | break; |
81eabcbe | 2270 | |
479f854a MG |
2271 | if (unlikely(check_pcp_refill(page))) |
2272 | continue; | |
2273 | ||
81eabcbe MG |
2274 | /* |
2275 | * Split buddy pages returned by expand() are received here | |
2276 | * in physical page order. The page is added to the callers and | |
2277 | * list and the list head then moves forward. From the callers | |
2278 | * perspective, the linked list is ordered by page number in | |
2279 | * some conditions. This is useful for IO devices that can | |
2280 | * merge IO requests if the physical pages are ordered | |
2281 | * properly. | |
2282 | */ | |
b745bc85 | 2283 | if (likely(!cold)) |
e084b2d9 MG |
2284 | list_add(&page->lru, list); |
2285 | else | |
2286 | list_add_tail(&page->lru, list); | |
81eabcbe | 2287 | list = &page->lru; |
a6de734b | 2288 | alloced++; |
bb14c2c7 | 2289 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2290 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2291 | -(1 << order)); | |
1da177e4 | 2292 | } |
a6de734b MG |
2293 | |
2294 | /* | |
2295 | * i pages were removed from the buddy list even if some leak due | |
2296 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2297 | * on i. Do not confuse with 'alloced' which is the number of | |
2298 | * pages added to the pcp list. | |
2299 | */ | |
f2260e6b | 2300 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2301 | spin_unlock(&zone->lock); |
a6de734b | 2302 | return alloced; |
1da177e4 LT |
2303 | } |
2304 | ||
4ae7c039 | 2305 | #ifdef CONFIG_NUMA |
8fce4d8e | 2306 | /* |
4037d452 CL |
2307 | * Called from the vmstat counter updater to drain pagesets of this |
2308 | * currently executing processor on remote nodes after they have | |
2309 | * expired. | |
2310 | * | |
879336c3 CL |
2311 | * Note that this function must be called with the thread pinned to |
2312 | * a single processor. | |
8fce4d8e | 2313 | */ |
4037d452 | 2314 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2315 | { |
4ae7c039 | 2316 | unsigned long flags; |
7be12fc9 | 2317 | int to_drain, batch; |
4ae7c039 | 2318 | |
4037d452 | 2319 | local_irq_save(flags); |
4db0c3c2 | 2320 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2321 | to_drain = min(pcp->count, batch); |
2a13515c KM |
2322 | if (to_drain > 0) { |
2323 | free_pcppages_bulk(zone, to_drain, pcp); | |
2324 | pcp->count -= to_drain; | |
2325 | } | |
4037d452 | 2326 | local_irq_restore(flags); |
4ae7c039 CL |
2327 | } |
2328 | #endif | |
2329 | ||
9f8f2172 | 2330 | /* |
93481ff0 | 2331 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2332 | * |
2333 | * The processor must either be the current processor and the | |
2334 | * thread pinned to the current processor or a processor that | |
2335 | * is not online. | |
2336 | */ | |
93481ff0 | 2337 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2338 | { |
c54ad30c | 2339 | unsigned long flags; |
93481ff0 VB |
2340 | struct per_cpu_pageset *pset; |
2341 | struct per_cpu_pages *pcp; | |
1da177e4 | 2342 | |
93481ff0 VB |
2343 | local_irq_save(flags); |
2344 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2345 | |
93481ff0 VB |
2346 | pcp = &pset->pcp; |
2347 | if (pcp->count) { | |
2348 | free_pcppages_bulk(zone, pcp->count, pcp); | |
2349 | pcp->count = 0; | |
2350 | } | |
2351 | local_irq_restore(flags); | |
2352 | } | |
3dfa5721 | 2353 | |
93481ff0 VB |
2354 | /* |
2355 | * Drain pcplists of all zones on the indicated processor. | |
2356 | * | |
2357 | * The processor must either be the current processor and the | |
2358 | * thread pinned to the current processor or a processor that | |
2359 | * is not online. | |
2360 | */ | |
2361 | static void drain_pages(unsigned int cpu) | |
2362 | { | |
2363 | struct zone *zone; | |
2364 | ||
2365 | for_each_populated_zone(zone) { | |
2366 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2367 | } |
2368 | } | |
1da177e4 | 2369 | |
9f8f2172 CL |
2370 | /* |
2371 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2372 | * |
2373 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2374 | * the single zone's pages. | |
9f8f2172 | 2375 | */ |
93481ff0 | 2376 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2377 | { |
93481ff0 VB |
2378 | int cpu = smp_processor_id(); |
2379 | ||
2380 | if (zone) | |
2381 | drain_pages_zone(cpu, zone); | |
2382 | else | |
2383 | drain_pages(cpu); | |
9f8f2172 CL |
2384 | } |
2385 | ||
0ccce3b9 MG |
2386 | static void drain_local_pages_wq(struct work_struct *work) |
2387 | { | |
a459eeb7 MH |
2388 | /* |
2389 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2390 | * we can race with cpu offline when the WQ can move this from | |
2391 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2392 | * cpu which is allright but we also have to make sure to not move to | |
2393 | * a different one. | |
2394 | */ | |
2395 | preempt_disable(); | |
0ccce3b9 | 2396 | drain_local_pages(NULL); |
a459eeb7 | 2397 | preempt_enable(); |
0ccce3b9 MG |
2398 | } |
2399 | ||
9f8f2172 | 2400 | /* |
74046494 GBY |
2401 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2402 | * | |
93481ff0 VB |
2403 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2404 | * | |
0ccce3b9 | 2405 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2406 | */ |
93481ff0 | 2407 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2408 | { |
74046494 | 2409 | int cpu; |
74046494 GBY |
2410 | |
2411 | /* | |
2412 | * Allocate in the BSS so we wont require allocation in | |
2413 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2414 | */ | |
2415 | static cpumask_t cpus_with_pcps; | |
2416 | ||
ce612879 MH |
2417 | /* |
2418 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2419 | * initialized. | |
2420 | */ | |
2421 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2422 | return; | |
2423 | ||
0ccce3b9 MG |
2424 | /* Workqueues cannot recurse */ |
2425 | if (current->flags & PF_WQ_WORKER) | |
2426 | return; | |
2427 | ||
bd233f53 MG |
2428 | /* |
2429 | * Do not drain if one is already in progress unless it's specific to | |
2430 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2431 | * the drain to be complete when the call returns. | |
2432 | */ | |
2433 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2434 | if (!zone) | |
2435 | return; | |
2436 | mutex_lock(&pcpu_drain_mutex); | |
2437 | } | |
0ccce3b9 | 2438 | |
74046494 GBY |
2439 | /* |
2440 | * We don't care about racing with CPU hotplug event | |
2441 | * as offline notification will cause the notified | |
2442 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2443 | * disables preemption as part of its processing | |
2444 | */ | |
2445 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2446 | struct per_cpu_pageset *pcp; |
2447 | struct zone *z; | |
74046494 | 2448 | bool has_pcps = false; |
93481ff0 VB |
2449 | |
2450 | if (zone) { | |
74046494 | 2451 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2452 | if (pcp->pcp.count) |
74046494 | 2453 | has_pcps = true; |
93481ff0 VB |
2454 | } else { |
2455 | for_each_populated_zone(z) { | |
2456 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2457 | if (pcp->pcp.count) { | |
2458 | has_pcps = true; | |
2459 | break; | |
2460 | } | |
74046494 GBY |
2461 | } |
2462 | } | |
93481ff0 | 2463 | |
74046494 GBY |
2464 | if (has_pcps) |
2465 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2466 | else | |
2467 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2468 | } | |
0ccce3b9 | 2469 | |
bd233f53 MG |
2470 | for_each_cpu(cpu, &cpus_with_pcps) { |
2471 | struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu); | |
2472 | INIT_WORK(work, drain_local_pages_wq); | |
ce612879 | 2473 | queue_work_on(cpu, mm_percpu_wq, work); |
0ccce3b9 | 2474 | } |
bd233f53 MG |
2475 | for_each_cpu(cpu, &cpus_with_pcps) |
2476 | flush_work(per_cpu_ptr(&pcpu_drain, cpu)); | |
2477 | ||
2478 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2479 | } |
2480 | ||
296699de | 2481 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
2482 | |
2483 | void mark_free_pages(struct zone *zone) | |
2484 | { | |
f623f0db RW |
2485 | unsigned long pfn, max_zone_pfn; |
2486 | unsigned long flags; | |
7aeb09f9 | 2487 | unsigned int order, t; |
86760a2c | 2488 | struct page *page; |
1da177e4 | 2489 | |
8080fc03 | 2490 | if (zone_is_empty(zone)) |
1da177e4 LT |
2491 | return; |
2492 | ||
2493 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2494 | |
108bcc96 | 2495 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2496 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2497 | if (pfn_valid(pfn)) { | |
86760a2c | 2498 | page = pfn_to_page(pfn); |
ba6b0979 JK |
2499 | |
2500 | if (page_zone(page) != zone) | |
2501 | continue; | |
2502 | ||
7be98234 RW |
2503 | if (!swsusp_page_is_forbidden(page)) |
2504 | swsusp_unset_page_free(page); | |
f623f0db | 2505 | } |
1da177e4 | 2506 | |
b2a0ac88 | 2507 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2508 | list_for_each_entry(page, |
2509 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2510 | unsigned long i; |
1da177e4 | 2511 | |
86760a2c | 2512 | pfn = page_to_pfn(page); |
f623f0db | 2513 | for (i = 0; i < (1UL << order); i++) |
7be98234 | 2514 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 2515 | } |
b2a0ac88 | 2516 | } |
1da177e4 LT |
2517 | spin_unlock_irqrestore(&zone->lock, flags); |
2518 | } | |
e2c55dc8 | 2519 | #endif /* CONFIG_PM */ |
1da177e4 | 2520 | |
1da177e4 LT |
2521 | /* |
2522 | * Free a 0-order page | |
b745bc85 | 2523 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 2524 | */ |
b745bc85 | 2525 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
2526 | { |
2527 | struct zone *zone = page_zone(page); | |
2528 | struct per_cpu_pages *pcp; | |
d34b0733 | 2529 | unsigned long flags; |
dc4b0caf | 2530 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 2531 | int migratetype; |
1da177e4 | 2532 | |
4db7548c | 2533 | if (!free_pcp_prepare(page)) |
689bcebf HD |
2534 | return; |
2535 | ||
dc4b0caf | 2536 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2537 | set_pcppage_migratetype(page, migratetype); |
d34b0733 MG |
2538 | local_irq_save(flags); |
2539 | __count_vm_event(PGFREE); | |
da456f14 | 2540 | |
5f8dcc21 MG |
2541 | /* |
2542 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2543 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 2544 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
2545 | * areas back if necessary. Otherwise, we may have to free |
2546 | * excessively into the page allocator | |
2547 | */ | |
2548 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2549 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2550 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
2551 | goto out; |
2552 | } | |
2553 | migratetype = MIGRATE_MOVABLE; | |
2554 | } | |
2555 | ||
99dcc3e5 | 2556 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 2557 | if (!cold) |
5f8dcc21 | 2558 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
2559 | else |
2560 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 2561 | pcp->count++; |
48db57f8 | 2562 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2563 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
2564 | free_pcppages_bulk(zone, batch, pcp); |
2565 | pcp->count -= batch; | |
48db57f8 | 2566 | } |
5f8dcc21 MG |
2567 | |
2568 | out: | |
d34b0733 | 2569 | local_irq_restore(flags); |
1da177e4 LT |
2570 | } |
2571 | ||
cc59850e KK |
2572 | /* |
2573 | * Free a list of 0-order pages | |
2574 | */ | |
b745bc85 | 2575 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
2576 | { |
2577 | struct page *page, *next; | |
2578 | ||
2579 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 2580 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
2581 | free_hot_cold_page(page, cold); |
2582 | } | |
2583 | } | |
2584 | ||
8dfcc9ba NP |
2585 | /* |
2586 | * split_page takes a non-compound higher-order page, and splits it into | |
2587 | * n (1<<order) sub-pages: page[0..n] | |
2588 | * Each sub-page must be freed individually. | |
2589 | * | |
2590 | * Note: this is probably too low level an operation for use in drivers. | |
2591 | * Please consult with lkml before using this in your driver. | |
2592 | */ | |
2593 | void split_page(struct page *page, unsigned int order) | |
2594 | { | |
2595 | int i; | |
2596 | ||
309381fe SL |
2597 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2598 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
2599 | |
2600 | #ifdef CONFIG_KMEMCHECK | |
2601 | /* | |
2602 | * Split shadow pages too, because free(page[0]) would | |
2603 | * otherwise free the whole shadow. | |
2604 | */ | |
2605 | if (kmemcheck_page_is_tracked(page)) | |
2606 | split_page(virt_to_page(page[0].shadow), order); | |
2607 | #endif | |
2608 | ||
a9627bc5 | 2609 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2610 | set_page_refcounted(page + i); |
a9627bc5 | 2611 | split_page_owner(page, order); |
8dfcc9ba | 2612 | } |
5853ff23 | 2613 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2614 | |
3c605096 | 2615 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2616 | { |
748446bb MG |
2617 | unsigned long watermark; |
2618 | struct zone *zone; | |
2139cbe6 | 2619 | int mt; |
748446bb MG |
2620 | |
2621 | BUG_ON(!PageBuddy(page)); | |
2622 | ||
2623 | zone = page_zone(page); | |
2e30abd1 | 2624 | mt = get_pageblock_migratetype(page); |
748446bb | 2625 | |
194159fb | 2626 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
2627 | /* |
2628 | * Obey watermarks as if the page was being allocated. We can | |
2629 | * emulate a high-order watermark check with a raised order-0 | |
2630 | * watermark, because we already know our high-order page | |
2631 | * exists. | |
2632 | */ | |
2633 | watermark = min_wmark_pages(zone) + (1UL << order); | |
984fdba6 | 2634 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
2635 | return 0; |
2636 | ||
8fb74b9f | 2637 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2638 | } |
748446bb MG |
2639 | |
2640 | /* Remove page from free list */ | |
2641 | list_del(&page->lru); | |
2642 | zone->free_area[order].nr_free--; | |
2643 | rmv_page_order(page); | |
2139cbe6 | 2644 | |
400bc7fd | 2645 | /* |
2646 | * Set the pageblock if the isolated page is at least half of a | |
2647 | * pageblock | |
2648 | */ | |
748446bb MG |
2649 | if (order >= pageblock_order - 1) { |
2650 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2651 | for (; page < endpage; page += pageblock_nr_pages) { |
2652 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 2653 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 2654 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
2655 | set_pageblock_migratetype(page, |
2656 | MIGRATE_MOVABLE); | |
2657 | } | |
748446bb MG |
2658 | } |
2659 | ||
f3a14ced | 2660 | |
8fb74b9f | 2661 | return 1UL << order; |
1fb3f8ca MG |
2662 | } |
2663 | ||
060e7417 MG |
2664 | /* |
2665 | * Update NUMA hit/miss statistics | |
2666 | * | |
2667 | * Must be called with interrupts disabled. | |
060e7417 | 2668 | */ |
41b6167e | 2669 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
2670 | { |
2671 | #ifdef CONFIG_NUMA | |
060e7417 MG |
2672 | enum zone_stat_item local_stat = NUMA_LOCAL; |
2673 | ||
2df26639 | 2674 | if (z->node != numa_node_id()) |
060e7417 | 2675 | local_stat = NUMA_OTHER; |
060e7417 | 2676 | |
2df26639 | 2677 | if (z->node == preferred_zone->node) |
060e7417 | 2678 | __inc_zone_state(z, NUMA_HIT); |
2df26639 | 2679 | else { |
060e7417 MG |
2680 | __inc_zone_state(z, NUMA_MISS); |
2681 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); | |
2682 | } | |
2df26639 | 2683 | __inc_zone_state(z, local_stat); |
060e7417 MG |
2684 | #endif |
2685 | } | |
2686 | ||
066b2393 MG |
2687 | /* Remove page from the per-cpu list, caller must protect the list */ |
2688 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
2689 | bool cold, struct per_cpu_pages *pcp, | |
2690 | struct list_head *list) | |
2691 | { | |
2692 | struct page *page; | |
2693 | ||
2694 | do { | |
2695 | if (list_empty(list)) { | |
2696 | pcp->count += rmqueue_bulk(zone, 0, | |
2697 | pcp->batch, list, | |
2698 | migratetype, cold); | |
2699 | if (unlikely(list_empty(list))) | |
2700 | return NULL; | |
2701 | } | |
2702 | ||
2703 | if (cold) | |
2704 | page = list_last_entry(list, struct page, lru); | |
2705 | else | |
2706 | page = list_first_entry(list, struct page, lru); | |
2707 | ||
2708 | list_del(&page->lru); | |
2709 | pcp->count--; | |
2710 | } while (check_new_pcp(page)); | |
2711 | ||
2712 | return page; | |
2713 | } | |
2714 | ||
2715 | /* Lock and remove page from the per-cpu list */ | |
2716 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
2717 | struct zone *zone, unsigned int order, | |
2718 | gfp_t gfp_flags, int migratetype) | |
2719 | { | |
2720 | struct per_cpu_pages *pcp; | |
2721 | struct list_head *list; | |
2722 | bool cold = ((gfp_flags & __GFP_COLD) != 0); | |
2723 | struct page *page; | |
d34b0733 | 2724 | unsigned long flags; |
066b2393 | 2725 | |
d34b0733 | 2726 | local_irq_save(flags); |
066b2393 MG |
2727 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2728 | list = &pcp->lists[migratetype]; | |
2729 | page = __rmqueue_pcplist(zone, migratetype, cold, pcp, list); | |
2730 | if (page) { | |
2731 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
2732 | zone_statistics(preferred_zone, zone); | |
2733 | } | |
d34b0733 | 2734 | local_irq_restore(flags); |
066b2393 MG |
2735 | return page; |
2736 | } | |
2737 | ||
1da177e4 | 2738 | /* |
75379191 | 2739 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2740 | */ |
0a15c3e9 | 2741 | static inline |
066b2393 | 2742 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 2743 | struct zone *zone, unsigned int order, |
c603844b MG |
2744 | gfp_t gfp_flags, unsigned int alloc_flags, |
2745 | int migratetype) | |
1da177e4 LT |
2746 | { |
2747 | unsigned long flags; | |
689bcebf | 2748 | struct page *page; |
1da177e4 | 2749 | |
d34b0733 | 2750 | if (likely(order == 0)) { |
066b2393 MG |
2751 | page = rmqueue_pcplist(preferred_zone, zone, order, |
2752 | gfp_flags, migratetype); | |
2753 | goto out; | |
2754 | } | |
83b9355b | 2755 | |
066b2393 MG |
2756 | /* |
2757 | * We most definitely don't want callers attempting to | |
2758 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2759 | */ | |
2760 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
2761 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 2762 | |
066b2393 MG |
2763 | do { |
2764 | page = NULL; | |
2765 | if (alloc_flags & ALLOC_HARDER) { | |
2766 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2767 | if (page) | |
2768 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
2769 | } | |
a74609fa | 2770 | if (!page) |
066b2393 MG |
2771 | page = __rmqueue(zone, order, migratetype); |
2772 | } while (page && check_new_pages(page, order)); | |
2773 | spin_unlock(&zone->lock); | |
2774 | if (!page) | |
2775 | goto failed; | |
2776 | __mod_zone_freepage_state(zone, -(1 << order), | |
2777 | get_pcppage_migratetype(page)); | |
1da177e4 | 2778 | |
16709d1d | 2779 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 2780 | zone_statistics(preferred_zone, zone); |
a74609fa | 2781 | local_irq_restore(flags); |
1da177e4 | 2782 | |
066b2393 MG |
2783 | out: |
2784 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); | |
1da177e4 | 2785 | return page; |
a74609fa NP |
2786 | |
2787 | failed: | |
2788 | local_irq_restore(flags); | |
a74609fa | 2789 | return NULL; |
1da177e4 LT |
2790 | } |
2791 | ||
933e312e AM |
2792 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2793 | ||
b2588c4b | 2794 | static struct { |
933e312e AM |
2795 | struct fault_attr attr; |
2796 | ||
621a5f7a | 2797 | bool ignore_gfp_highmem; |
71baba4b | 2798 | bool ignore_gfp_reclaim; |
54114994 | 2799 | u32 min_order; |
933e312e AM |
2800 | } fail_page_alloc = { |
2801 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 2802 | .ignore_gfp_reclaim = true, |
621a5f7a | 2803 | .ignore_gfp_highmem = true, |
54114994 | 2804 | .min_order = 1, |
933e312e AM |
2805 | }; |
2806 | ||
2807 | static int __init setup_fail_page_alloc(char *str) | |
2808 | { | |
2809 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
2810 | } | |
2811 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
2812 | ||
deaf386e | 2813 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2814 | { |
54114994 | 2815 | if (order < fail_page_alloc.min_order) |
deaf386e | 2816 | return false; |
933e312e | 2817 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 2818 | return false; |
933e312e | 2819 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 2820 | return false; |
71baba4b MG |
2821 | if (fail_page_alloc.ignore_gfp_reclaim && |
2822 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 2823 | return false; |
933e312e AM |
2824 | |
2825 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
2826 | } | |
2827 | ||
2828 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
2829 | ||
2830 | static int __init fail_page_alloc_debugfs(void) | |
2831 | { | |
f4ae40a6 | 2832 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 2833 | struct dentry *dir; |
933e312e | 2834 | |
dd48c085 AM |
2835 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2836 | &fail_page_alloc.attr); | |
2837 | if (IS_ERR(dir)) | |
2838 | return PTR_ERR(dir); | |
933e312e | 2839 | |
b2588c4b | 2840 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 2841 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
2842 | goto fail; |
2843 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
2844 | &fail_page_alloc.ignore_gfp_highmem)) | |
2845 | goto fail; | |
2846 | if (!debugfs_create_u32("min-order", mode, dir, | |
2847 | &fail_page_alloc.min_order)) | |
2848 | goto fail; | |
2849 | ||
2850 | return 0; | |
2851 | fail: | |
dd48c085 | 2852 | debugfs_remove_recursive(dir); |
933e312e | 2853 | |
b2588c4b | 2854 | return -ENOMEM; |
933e312e AM |
2855 | } |
2856 | ||
2857 | late_initcall(fail_page_alloc_debugfs); | |
2858 | ||
2859 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2860 | ||
2861 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2862 | ||
deaf386e | 2863 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2864 | { |
deaf386e | 2865 | return false; |
933e312e AM |
2866 | } |
2867 | ||
2868 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2869 | ||
1da177e4 | 2870 | /* |
97a16fc8 MG |
2871 | * Return true if free base pages are above 'mark'. For high-order checks it |
2872 | * will return true of the order-0 watermark is reached and there is at least | |
2873 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
2874 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 2875 | */ |
86a294a8 MH |
2876 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
2877 | int classzone_idx, unsigned int alloc_flags, | |
2878 | long free_pages) | |
1da177e4 | 2879 | { |
d23ad423 | 2880 | long min = mark; |
1da177e4 | 2881 | int o; |
c603844b | 2882 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
1da177e4 | 2883 | |
0aaa29a5 | 2884 | /* free_pages may go negative - that's OK */ |
df0a6daa | 2885 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 2886 | |
7fb1d9fc | 2887 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2888 | min -= min / 2; |
0aaa29a5 MG |
2889 | |
2890 | /* | |
2891 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
2892 | * the high-atomic reserves. This will over-estimate the size of the | |
2893 | * atomic reserve but it avoids a search. | |
2894 | */ | |
97a16fc8 | 2895 | if (likely(!alloc_harder)) |
0aaa29a5 MG |
2896 | free_pages -= z->nr_reserved_highatomic; |
2897 | else | |
1da177e4 | 2898 | min -= min / 4; |
e2b19197 | 2899 | |
d95ea5d1 BZ |
2900 | #ifdef CONFIG_CMA |
2901 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2902 | if (!(alloc_flags & ALLOC_CMA)) | |
97a16fc8 | 2903 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2904 | #endif |
026b0814 | 2905 | |
97a16fc8 MG |
2906 | /* |
2907 | * Check watermarks for an order-0 allocation request. If these | |
2908 | * are not met, then a high-order request also cannot go ahead | |
2909 | * even if a suitable page happened to be free. | |
2910 | */ | |
2911 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 2912 | return false; |
1da177e4 | 2913 | |
97a16fc8 MG |
2914 | /* If this is an order-0 request then the watermark is fine */ |
2915 | if (!order) | |
2916 | return true; | |
2917 | ||
2918 | /* For a high-order request, check at least one suitable page is free */ | |
2919 | for (o = order; o < MAX_ORDER; o++) { | |
2920 | struct free_area *area = &z->free_area[o]; | |
2921 | int mt; | |
2922 | ||
2923 | if (!area->nr_free) | |
2924 | continue; | |
2925 | ||
2926 | if (alloc_harder) | |
2927 | return true; | |
1da177e4 | 2928 | |
97a16fc8 MG |
2929 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
2930 | if (!list_empty(&area->free_list[mt])) | |
2931 | return true; | |
2932 | } | |
2933 | ||
2934 | #ifdef CONFIG_CMA | |
2935 | if ((alloc_flags & ALLOC_CMA) && | |
2936 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
2937 | return true; | |
2938 | } | |
2939 | #endif | |
1da177e4 | 2940 | } |
97a16fc8 | 2941 | return false; |
88f5acf8 MG |
2942 | } |
2943 | ||
7aeb09f9 | 2944 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 2945 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
2946 | { |
2947 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2948 | zone_page_state(z, NR_FREE_PAGES)); | |
2949 | } | |
2950 | ||
48ee5f36 MG |
2951 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
2952 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
2953 | { | |
2954 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2955 | long cma_pages = 0; | |
2956 | ||
2957 | #ifdef CONFIG_CMA | |
2958 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2959 | if (!(alloc_flags & ALLOC_CMA)) | |
2960 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
2961 | #endif | |
2962 | ||
2963 | /* | |
2964 | * Fast check for order-0 only. If this fails then the reserves | |
2965 | * need to be calculated. There is a corner case where the check | |
2966 | * passes but only the high-order atomic reserve are free. If | |
2967 | * the caller is !atomic then it'll uselessly search the free | |
2968 | * list. That corner case is then slower but it is harmless. | |
2969 | */ | |
2970 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) | |
2971 | return true; | |
2972 | ||
2973 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2974 | free_pages); | |
2975 | } | |
2976 | ||
7aeb09f9 | 2977 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 2978 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
2979 | { |
2980 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2981 | ||
2982 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2983 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2984 | ||
e2b19197 | 2985 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 2986 | free_pages); |
1da177e4 LT |
2987 | } |
2988 | ||
9276b1bc | 2989 | #ifdef CONFIG_NUMA |
957f822a DR |
2990 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2991 | { | |
e02dc017 | 2992 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
5f7a75ac | 2993 | RECLAIM_DISTANCE; |
957f822a | 2994 | } |
9276b1bc | 2995 | #else /* CONFIG_NUMA */ |
957f822a DR |
2996 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2997 | { | |
2998 | return true; | |
2999 | } | |
9276b1bc PJ |
3000 | #endif /* CONFIG_NUMA */ |
3001 | ||
7fb1d9fc | 3002 | /* |
0798e519 | 3003 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3004 | * a page. |
3005 | */ | |
3006 | static struct page * | |
a9263751 VB |
3007 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3008 | const struct alloc_context *ac) | |
753ee728 | 3009 | { |
c33d6c06 | 3010 | struct zoneref *z = ac->preferred_zoneref; |
5117f45d | 3011 | struct zone *zone; |
3b8c0be4 MG |
3012 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
3013 | ||
7fb1d9fc | 3014 | /* |
9276b1bc | 3015 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3016 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3017 | */ |
c33d6c06 | 3018 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 3019 | ac->nodemask) { |
be06af00 | 3020 | struct page *page; |
e085dbc5 JW |
3021 | unsigned long mark; |
3022 | ||
664eedde MG |
3023 | if (cpusets_enabled() && |
3024 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3025 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3026 | continue; |
a756cf59 JW |
3027 | /* |
3028 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3029 | * want to get it from a node that is within its dirty |
3030 | * limit, such that no single node holds more than its | |
a756cf59 | 3031 | * proportional share of globally allowed dirty pages. |
281e3726 | 3032 | * The dirty limits take into account the node's |
a756cf59 JW |
3033 | * lowmem reserves and high watermark so that kswapd |
3034 | * should be able to balance it without having to | |
3035 | * write pages from its LRU list. | |
3036 | * | |
a756cf59 | 3037 | * XXX: For now, allow allocations to potentially |
281e3726 | 3038 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3039 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3040 | * which is important when on a NUMA setup the allowed |
281e3726 | 3041 | * nodes are together not big enough to reach the |
a756cf59 | 3042 | * global limit. The proper fix for these situations |
281e3726 | 3043 | * will require awareness of nodes in the |
a756cf59 JW |
3044 | * dirty-throttling and the flusher threads. |
3045 | */ | |
3b8c0be4 MG |
3046 | if (ac->spread_dirty_pages) { |
3047 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3048 | continue; | |
3049 | ||
3050 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3051 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3052 | continue; | |
3053 | } | |
3054 | } | |
7fb1d9fc | 3055 | |
e085dbc5 | 3056 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 3057 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 3058 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
3059 | int ret; |
3060 | ||
5dab2911 MG |
3061 | /* Checked here to keep the fast path fast */ |
3062 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3063 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3064 | goto try_this_zone; | |
3065 | ||
a5f5f91d | 3066 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3067 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3068 | continue; |
3069 | ||
a5f5f91d | 3070 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3071 | switch (ret) { |
a5f5f91d | 3072 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3073 | /* did not scan */ |
cd38b115 | 3074 | continue; |
a5f5f91d | 3075 | case NODE_RECLAIM_FULL: |
fa5e084e | 3076 | /* scanned but unreclaimable */ |
cd38b115 | 3077 | continue; |
fa5e084e MG |
3078 | default: |
3079 | /* did we reclaim enough */ | |
fed2719e | 3080 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3081 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3082 | goto try_this_zone; |
3083 | ||
fed2719e | 3084 | continue; |
0798e519 | 3085 | } |
7fb1d9fc RS |
3086 | } |
3087 | ||
fa5e084e | 3088 | try_this_zone: |
066b2393 | 3089 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3090 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3091 | if (page) { |
479f854a | 3092 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3093 | |
3094 | /* | |
3095 | * If this is a high-order atomic allocation then check | |
3096 | * if the pageblock should be reserved for the future | |
3097 | */ | |
3098 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3099 | reserve_highatomic_pageblock(page, zone, order); | |
3100 | ||
75379191 VB |
3101 | return page; |
3102 | } | |
54a6eb5c | 3103 | } |
9276b1bc | 3104 | |
4ffeaf35 | 3105 | return NULL; |
753ee728 MH |
3106 | } |
3107 | ||
29423e77 DR |
3108 | /* |
3109 | * Large machines with many possible nodes should not always dump per-node | |
3110 | * meminfo in irq context. | |
3111 | */ | |
3112 | static inline bool should_suppress_show_mem(void) | |
3113 | { | |
3114 | bool ret = false; | |
3115 | ||
3116 | #if NODES_SHIFT > 8 | |
3117 | ret = in_interrupt(); | |
3118 | #endif | |
3119 | return ret; | |
3120 | } | |
3121 | ||
9af744d7 | 3122 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3123 | { |
a238ab5b | 3124 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
aa187507 | 3125 | static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1); |
a238ab5b | 3126 | |
aa187507 | 3127 | if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs)) |
a238ab5b DH |
3128 | return; |
3129 | ||
3130 | /* | |
3131 | * This documents exceptions given to allocations in certain | |
3132 | * contexts that are allowed to allocate outside current's set | |
3133 | * of allowed nodes. | |
3134 | */ | |
3135 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
3136 | if (test_thread_flag(TIF_MEMDIE) || | |
3137 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
3138 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3139 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3140 | filter &= ~SHOW_MEM_FILTER_NODES; |
3141 | ||
9af744d7 | 3142 | show_mem(filter, nodemask); |
aa187507 MH |
3143 | } |
3144 | ||
a8e99259 | 3145 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3146 | { |
3147 | struct va_format vaf; | |
3148 | va_list args; | |
3149 | static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL, | |
3150 | DEFAULT_RATELIMIT_BURST); | |
3151 | ||
0f7896f1 | 3152 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3153 | return; |
3154 | ||
7877cdcc | 3155 | pr_warn("%s: ", current->comm); |
3ee9a4f0 | 3156 | |
7877cdcc MH |
3157 | va_start(args, fmt); |
3158 | vaf.fmt = fmt; | |
3159 | vaf.va = &args; | |
3160 | pr_cont("%pV", &vaf); | |
3161 | va_end(args); | |
3ee9a4f0 | 3162 | |
685dbf6f DR |
3163 | pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask, &gfp_mask); |
3164 | if (nodemask) | |
3165 | pr_cont("%*pbl\n", nodemask_pr_args(nodemask)); | |
3166 | else | |
3167 | pr_cont("(null)\n"); | |
3168 | ||
a8e99259 | 3169 | cpuset_print_current_mems_allowed(); |
3ee9a4f0 | 3170 | |
a238ab5b | 3171 | dump_stack(); |
685dbf6f | 3172 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3173 | } |
3174 | ||
6c18ba7a MH |
3175 | static inline struct page * |
3176 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3177 | unsigned int alloc_flags, | |
3178 | const struct alloc_context *ac) | |
3179 | { | |
3180 | struct page *page; | |
3181 | ||
3182 | page = get_page_from_freelist(gfp_mask, order, | |
3183 | alloc_flags|ALLOC_CPUSET, ac); | |
3184 | /* | |
3185 | * fallback to ignore cpuset restriction if our nodes | |
3186 | * are depleted | |
3187 | */ | |
3188 | if (!page) | |
3189 | page = get_page_from_freelist(gfp_mask, order, | |
3190 | alloc_flags, ac); | |
3191 | ||
3192 | return page; | |
3193 | } | |
3194 | ||
11e33f6a MG |
3195 | static inline struct page * |
3196 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3197 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3198 | { |
6e0fc46d DR |
3199 | struct oom_control oc = { |
3200 | .zonelist = ac->zonelist, | |
3201 | .nodemask = ac->nodemask, | |
2a966b77 | 3202 | .memcg = NULL, |
6e0fc46d DR |
3203 | .gfp_mask = gfp_mask, |
3204 | .order = order, | |
6e0fc46d | 3205 | }; |
11e33f6a MG |
3206 | struct page *page; |
3207 | ||
9879de73 JW |
3208 | *did_some_progress = 0; |
3209 | ||
9879de73 | 3210 | /* |
dc56401f JW |
3211 | * Acquire the oom lock. If that fails, somebody else is |
3212 | * making progress for us. | |
9879de73 | 3213 | */ |
dc56401f | 3214 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3215 | *did_some_progress = 1; |
11e33f6a | 3216 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3217 | return NULL; |
3218 | } | |
6b1de916 | 3219 | |
11e33f6a MG |
3220 | /* |
3221 | * Go through the zonelist yet one more time, keep very high watermark | |
3222 | * here, this is only to catch a parallel oom killing, we must fail if | |
3223 | * we're still under heavy pressure. | |
3224 | */ | |
a9263751 VB |
3225 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
3226 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3227 | if (page) |
11e33f6a MG |
3228 | goto out; |
3229 | ||
06ad276a MH |
3230 | /* Coredumps can quickly deplete all memory reserves */ |
3231 | if (current->flags & PF_DUMPCORE) | |
3232 | goto out; | |
3233 | /* The OOM killer will not help higher order allocs */ | |
3234 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3235 | goto out; | |
3236 | /* The OOM killer does not needlessly kill tasks for lowmem */ | |
3237 | if (ac->high_zoneidx < ZONE_NORMAL) | |
3238 | goto out; | |
3239 | if (pm_suspended_storage()) | |
3240 | goto out; | |
3241 | /* | |
3242 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3243 | * other request to make a forward progress. | |
3244 | * We are in an unfortunate situation where out_of_memory cannot | |
3245 | * do much for this context but let's try it to at least get | |
3246 | * access to memory reserved if the current task is killed (see | |
3247 | * out_of_memory). Once filesystems are ready to handle allocation | |
3248 | * failures more gracefully we should just bail out here. | |
3249 | */ | |
3250 | ||
3251 | /* The OOM killer may not free memory on a specific node */ | |
3252 | if (gfp_mask & __GFP_THISNODE) | |
3253 | goto out; | |
3da88fb3 | 3254 | |
11e33f6a | 3255 | /* Exhausted what can be done so it's blamo time */ |
5020e285 | 3256 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3257 | *did_some_progress = 1; |
5020e285 | 3258 | |
6c18ba7a MH |
3259 | /* |
3260 | * Help non-failing allocations by giving them access to memory | |
3261 | * reserves | |
3262 | */ | |
3263 | if (gfp_mask & __GFP_NOFAIL) | |
3264 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3265 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3266 | } |
11e33f6a | 3267 | out: |
dc56401f | 3268 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3269 | return page; |
3270 | } | |
3271 | ||
33c2d214 MH |
3272 | /* |
3273 | * Maximum number of compaction retries wit a progress before OOM | |
3274 | * killer is consider as the only way to move forward. | |
3275 | */ | |
3276 | #define MAX_COMPACT_RETRIES 16 | |
3277 | ||
56de7263 MG |
3278 | #ifdef CONFIG_COMPACTION |
3279 | /* Try memory compaction for high-order allocations before reclaim */ | |
3280 | static struct page * | |
3281 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3282 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3283 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3284 | { |
98dd3b48 | 3285 | struct page *page; |
499118e9 | 3286 | unsigned int noreclaim_flag; |
53853e2d VB |
3287 | |
3288 | if (!order) | |
66199712 | 3289 | return NULL; |
66199712 | 3290 | |
499118e9 | 3291 | noreclaim_flag = memalloc_noreclaim_save(); |
c5d01d0d | 3292 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
c3486f53 | 3293 | prio); |
499118e9 | 3294 | memalloc_noreclaim_restore(noreclaim_flag); |
56de7263 | 3295 | |
c5d01d0d | 3296 | if (*compact_result <= COMPACT_INACTIVE) |
98dd3b48 | 3297 | return NULL; |
53853e2d | 3298 | |
98dd3b48 VB |
3299 | /* |
3300 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3301 | * count a compaction stall | |
3302 | */ | |
3303 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3304 | |
31a6c190 | 3305 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
53853e2d | 3306 | |
98dd3b48 VB |
3307 | if (page) { |
3308 | struct zone *zone = page_zone(page); | |
53853e2d | 3309 | |
98dd3b48 VB |
3310 | zone->compact_blockskip_flush = false; |
3311 | compaction_defer_reset(zone, order, true); | |
3312 | count_vm_event(COMPACTSUCCESS); | |
3313 | return page; | |
3314 | } | |
56de7263 | 3315 | |
98dd3b48 VB |
3316 | /* |
3317 | * It's bad if compaction run occurs and fails. The most likely reason | |
3318 | * is that pages exist, but not enough to satisfy watermarks. | |
3319 | */ | |
3320 | count_vm_event(COMPACTFAIL); | |
66199712 | 3321 | |
98dd3b48 | 3322 | cond_resched(); |
56de7263 MG |
3323 | |
3324 | return NULL; | |
3325 | } | |
33c2d214 | 3326 | |
3250845d VB |
3327 | static inline bool |
3328 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3329 | enum compact_result compact_result, | |
3330 | enum compact_priority *compact_priority, | |
d9436498 | 3331 | int *compaction_retries) |
3250845d VB |
3332 | { |
3333 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3334 | int min_priority; |
65190cff MH |
3335 | bool ret = false; |
3336 | int retries = *compaction_retries; | |
3337 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3338 | |
3339 | if (!order) | |
3340 | return false; | |
3341 | ||
d9436498 VB |
3342 | if (compaction_made_progress(compact_result)) |
3343 | (*compaction_retries)++; | |
3344 | ||
3250845d VB |
3345 | /* |
3346 | * compaction considers all the zone as desperately out of memory | |
3347 | * so it doesn't really make much sense to retry except when the | |
3348 | * failure could be caused by insufficient priority | |
3349 | */ | |
d9436498 VB |
3350 | if (compaction_failed(compact_result)) |
3351 | goto check_priority; | |
3250845d VB |
3352 | |
3353 | /* | |
3354 | * make sure the compaction wasn't deferred or didn't bail out early | |
3355 | * due to locks contention before we declare that we should give up. | |
3356 | * But do not retry if the given zonelist is not suitable for | |
3357 | * compaction. | |
3358 | */ | |
65190cff MH |
3359 | if (compaction_withdrawn(compact_result)) { |
3360 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
3361 | goto out; | |
3362 | } | |
3250845d VB |
3363 | |
3364 | /* | |
3365 | * !costly requests are much more important than __GFP_REPEAT | |
3366 | * costly ones because they are de facto nofail and invoke OOM | |
3367 | * killer to move on while costly can fail and users are ready | |
3368 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3369 | * would need much more detailed feedback from compaction to | |
3370 | * make a better decision. | |
3371 | */ | |
3372 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3373 | max_retries /= 4; | |
65190cff MH |
3374 | if (*compaction_retries <= max_retries) { |
3375 | ret = true; | |
3376 | goto out; | |
3377 | } | |
3250845d | 3378 | |
d9436498 VB |
3379 | /* |
3380 | * Make sure there are attempts at the highest priority if we exhausted | |
3381 | * all retries or failed at the lower priorities. | |
3382 | */ | |
3383 | check_priority: | |
c2033b00 VB |
3384 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3385 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3386 | |
c2033b00 | 3387 | if (*compact_priority > min_priority) { |
d9436498 VB |
3388 | (*compact_priority)--; |
3389 | *compaction_retries = 0; | |
65190cff | 3390 | ret = true; |
d9436498 | 3391 | } |
65190cff MH |
3392 | out: |
3393 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
3394 | return ret; | |
3250845d | 3395 | } |
56de7263 MG |
3396 | #else |
3397 | static inline struct page * | |
3398 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3399 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3400 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3401 | { |
33c2d214 | 3402 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3403 | return NULL; |
3404 | } | |
33c2d214 MH |
3405 | |
3406 | static inline bool | |
86a294a8 MH |
3407 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3408 | enum compact_result compact_result, | |
a5508cd8 | 3409 | enum compact_priority *compact_priority, |
d9436498 | 3410 | int *compaction_retries) |
33c2d214 | 3411 | { |
31e49bfd MH |
3412 | struct zone *zone; |
3413 | struct zoneref *z; | |
3414 | ||
3415 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3416 | return false; | |
3417 | ||
3418 | /* | |
3419 | * There are setups with compaction disabled which would prefer to loop | |
3420 | * inside the allocator rather than hit the oom killer prematurely. | |
3421 | * Let's give them a good hope and keep retrying while the order-0 | |
3422 | * watermarks are OK. | |
3423 | */ | |
3424 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3425 | ac->nodemask) { | |
3426 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
3427 | ac_classzone_idx(ac), alloc_flags)) | |
3428 | return true; | |
3429 | } | |
33c2d214 MH |
3430 | return false; |
3431 | } | |
3250845d | 3432 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3433 | |
bba90710 MS |
3434 | /* Perform direct synchronous page reclaim */ |
3435 | static int | |
a9263751 VB |
3436 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3437 | const struct alloc_context *ac) | |
11e33f6a | 3438 | { |
11e33f6a | 3439 | struct reclaim_state reclaim_state; |
bba90710 | 3440 | int progress; |
499118e9 | 3441 | unsigned int noreclaim_flag; |
11e33f6a MG |
3442 | |
3443 | cond_resched(); | |
3444 | ||
3445 | /* We now go into synchronous reclaim */ | |
3446 | cpuset_memory_pressure_bump(); | |
499118e9 | 3447 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a MG |
3448 | lockdep_set_current_reclaim_state(gfp_mask); |
3449 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 3450 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3451 | |
a9263751 VB |
3452 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3453 | ac->nodemask); | |
11e33f6a | 3454 | |
c06b1fca | 3455 | current->reclaim_state = NULL; |
11e33f6a | 3456 | lockdep_clear_current_reclaim_state(); |
499118e9 | 3457 | memalloc_noreclaim_restore(noreclaim_flag); |
11e33f6a MG |
3458 | |
3459 | cond_resched(); | |
3460 | ||
bba90710 MS |
3461 | return progress; |
3462 | } | |
3463 | ||
3464 | /* The really slow allocator path where we enter direct reclaim */ | |
3465 | static inline struct page * | |
3466 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3467 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3468 | unsigned long *did_some_progress) |
bba90710 MS |
3469 | { |
3470 | struct page *page = NULL; | |
3471 | bool drained = false; | |
3472 | ||
a9263751 | 3473 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3474 | if (unlikely(!(*did_some_progress))) |
3475 | return NULL; | |
11e33f6a | 3476 | |
9ee493ce | 3477 | retry: |
31a6c190 | 3478 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3479 | |
3480 | /* | |
3481 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3482 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3483 | * Shrink them them and try again | |
9ee493ce MG |
3484 | */ |
3485 | if (!page && !drained) { | |
29fac03b | 3486 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3487 | drain_all_pages(NULL); |
9ee493ce MG |
3488 | drained = true; |
3489 | goto retry; | |
3490 | } | |
3491 | ||
11e33f6a MG |
3492 | return page; |
3493 | } | |
3494 | ||
a9263751 | 3495 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
3496 | { |
3497 | struct zoneref *z; | |
3498 | struct zone *zone; | |
e1a55637 | 3499 | pg_data_t *last_pgdat = NULL; |
3a025760 | 3500 | |
a9263751 | 3501 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
e1a55637 MG |
3502 | ac->high_zoneidx, ac->nodemask) { |
3503 | if (last_pgdat != zone->zone_pgdat) | |
52e9f87a | 3504 | wakeup_kswapd(zone, order, ac->high_zoneidx); |
e1a55637 MG |
3505 | last_pgdat = zone->zone_pgdat; |
3506 | } | |
3a025760 JW |
3507 | } |
3508 | ||
c603844b | 3509 | static inline unsigned int |
341ce06f PZ |
3510 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3511 | { | |
c603844b | 3512 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3513 | |
a56f57ff | 3514 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3515 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3516 | |
341ce06f PZ |
3517 | /* |
3518 | * The caller may dip into page reserves a bit more if the caller | |
3519 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3520 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3521 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3522 | */ |
e6223a3b | 3523 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3524 | |
d0164adc | 3525 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3526 | /* |
b104a35d DR |
3527 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3528 | * if it can't schedule. | |
5c3240d9 | 3529 | */ |
b104a35d | 3530 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3531 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3532 | /* |
b104a35d | 3533 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3534 | * comment for __cpuset_node_allowed(). |
523b9458 | 3535 | */ |
341ce06f | 3536 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3537 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3538 | alloc_flags |= ALLOC_HARDER; |
3539 | ||
d95ea5d1 | 3540 | #ifdef CONFIG_CMA |
43e7a34d | 3541 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
3542 | alloc_flags |= ALLOC_CMA; |
3543 | #endif | |
341ce06f PZ |
3544 | return alloc_flags; |
3545 | } | |
3546 | ||
072bb0aa MG |
3547 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3548 | { | |
31a6c190 VB |
3549 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) |
3550 | return false; | |
3551 | ||
3552 | if (gfp_mask & __GFP_MEMALLOC) | |
3553 | return true; | |
3554 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) | |
3555 | return true; | |
3556 | if (!in_interrupt() && | |
3557 | ((current->flags & PF_MEMALLOC) || | |
3558 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
3559 | return true; | |
3560 | ||
3561 | return false; | |
072bb0aa MG |
3562 | } |
3563 | ||
0a0337e0 MH |
3564 | /* |
3565 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
3566 | * for the given allocation request. | |
491d79ae JW |
3567 | * |
3568 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
3569 | * without success, or when we couldn't even meet the watermark if we | |
3570 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
3571 | * |
3572 | * Returns true if a retry is viable or false to enter the oom path. | |
3573 | */ | |
3574 | static inline bool | |
3575 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
3576 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 3577 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
3578 | { |
3579 | struct zone *zone; | |
3580 | struct zoneref *z; | |
3581 | ||
423b452e VB |
3582 | /* |
3583 | * Costly allocations might have made a progress but this doesn't mean | |
3584 | * their order will become available due to high fragmentation so | |
3585 | * always increment the no progress counter for them | |
3586 | */ | |
3587 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
3588 | *no_progress_loops = 0; | |
3589 | else | |
3590 | (*no_progress_loops)++; | |
3591 | ||
0a0337e0 MH |
3592 | /* |
3593 | * Make sure we converge to OOM if we cannot make any progress | |
3594 | * several times in the row. | |
3595 | */ | |
04c8716f MK |
3596 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
3597 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 3598 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 3599 | } |
0a0337e0 | 3600 | |
bca67592 MG |
3601 | /* |
3602 | * Keep reclaiming pages while there is a chance this will lead | |
3603 | * somewhere. If none of the target zones can satisfy our allocation | |
3604 | * request even if all reclaimable pages are considered then we are | |
3605 | * screwed and have to go OOM. | |
0a0337e0 MH |
3606 | */ |
3607 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3608 | ac->nodemask) { | |
3609 | unsigned long available; | |
ede37713 | 3610 | unsigned long reclaimable; |
d379f01d MH |
3611 | unsigned long min_wmark = min_wmark_pages(zone); |
3612 | bool wmark; | |
0a0337e0 | 3613 | |
5a1c84b4 | 3614 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 3615 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
3616 | |
3617 | /* | |
491d79ae JW |
3618 | * Would the allocation succeed if we reclaimed all |
3619 | * reclaimable pages? | |
0a0337e0 | 3620 | */ |
d379f01d MH |
3621 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
3622 | ac_classzone_idx(ac), alloc_flags, available); | |
3623 | trace_reclaim_retry_zone(z, order, reclaimable, | |
3624 | available, min_wmark, *no_progress_loops, wmark); | |
3625 | if (wmark) { | |
ede37713 MH |
3626 | /* |
3627 | * If we didn't make any progress and have a lot of | |
3628 | * dirty + writeback pages then we should wait for | |
3629 | * an IO to complete to slow down the reclaim and | |
3630 | * prevent from pre mature OOM | |
3631 | */ | |
3632 | if (!did_some_progress) { | |
11fb9989 | 3633 | unsigned long write_pending; |
ede37713 | 3634 | |
5a1c84b4 MG |
3635 | write_pending = zone_page_state_snapshot(zone, |
3636 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 3637 | |
11fb9989 | 3638 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
3639 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
3640 | return true; | |
3641 | } | |
3642 | } | |
5a1c84b4 | 3643 | |
ede37713 MH |
3644 | /* |
3645 | * Memory allocation/reclaim might be called from a WQ | |
3646 | * context and the current implementation of the WQ | |
3647 | * concurrency control doesn't recognize that | |
3648 | * a particular WQ is congested if the worker thread is | |
3649 | * looping without ever sleeping. Therefore we have to | |
3650 | * do a short sleep here rather than calling | |
3651 | * cond_resched(). | |
3652 | */ | |
3653 | if (current->flags & PF_WQ_WORKER) | |
3654 | schedule_timeout_uninterruptible(1); | |
3655 | else | |
3656 | cond_resched(); | |
3657 | ||
0a0337e0 MH |
3658 | return true; |
3659 | } | |
3660 | } | |
3661 | ||
3662 | return false; | |
3663 | } | |
3664 | ||
11e33f6a MG |
3665 | static inline struct page * |
3666 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3667 | struct alloc_context *ac) |
11e33f6a | 3668 | { |
d0164adc | 3669 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 3670 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 3671 | struct page *page = NULL; |
c603844b | 3672 | unsigned int alloc_flags; |
11e33f6a | 3673 | unsigned long did_some_progress; |
5ce9bfef | 3674 | enum compact_priority compact_priority; |
c5d01d0d | 3675 | enum compact_result compact_result; |
5ce9bfef VB |
3676 | int compaction_retries; |
3677 | int no_progress_loops; | |
63f53dea MH |
3678 | unsigned long alloc_start = jiffies; |
3679 | unsigned int stall_timeout = 10 * HZ; | |
5ce9bfef | 3680 | unsigned int cpuset_mems_cookie; |
1da177e4 | 3681 | |
72807a74 MG |
3682 | /* |
3683 | * In the slowpath, we sanity check order to avoid ever trying to | |
3684 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
3685 | * be using allocators in order of preference for an area that is | |
3686 | * too large. | |
3687 | */ | |
1fc28b70 MG |
3688 | if (order >= MAX_ORDER) { |
3689 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 3690 | return NULL; |
1fc28b70 | 3691 | } |
1da177e4 | 3692 | |
d0164adc MG |
3693 | /* |
3694 | * We also sanity check to catch abuse of atomic reserves being used by | |
3695 | * callers that are not in atomic context. | |
3696 | */ | |
3697 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
3698 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
3699 | gfp_mask &= ~__GFP_ATOMIC; | |
3700 | ||
5ce9bfef VB |
3701 | retry_cpuset: |
3702 | compaction_retries = 0; | |
3703 | no_progress_loops = 0; | |
3704 | compact_priority = DEF_COMPACT_PRIORITY; | |
3705 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
3706 | |
3707 | /* | |
3708 | * The fast path uses conservative alloc_flags to succeed only until | |
3709 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
3710 | * alloc_flags precisely. So we do that now. | |
3711 | */ | |
3712 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
3713 | ||
e47483bc VB |
3714 | /* |
3715 | * We need to recalculate the starting point for the zonelist iterator | |
3716 | * because we might have used different nodemask in the fast path, or | |
3717 | * there was a cpuset modification and we are retrying - otherwise we | |
3718 | * could end up iterating over non-eligible zones endlessly. | |
3719 | */ | |
3720 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
3721 | ac->high_zoneidx, ac->nodemask); | |
3722 | if (!ac->preferred_zoneref->zone) | |
3723 | goto nopage; | |
3724 | ||
23771235 VB |
3725 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3726 | wake_all_kswapds(order, ac); | |
3727 | ||
3728 | /* | |
3729 | * The adjusted alloc_flags might result in immediate success, so try | |
3730 | * that first | |
3731 | */ | |
3732 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
3733 | if (page) | |
3734 | goto got_pg; | |
3735 | ||
a8161d1e VB |
3736 | /* |
3737 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
3738 | * that we have enough base pages and don't need to reclaim. For non- |
3739 | * movable high-order allocations, do that as well, as compaction will | |
3740 | * try prevent permanent fragmentation by migrating from blocks of the | |
3741 | * same migratetype. | |
3742 | * Don't try this for allocations that are allowed to ignore | |
3743 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 3744 | */ |
282722b0 VB |
3745 | if (can_direct_reclaim && |
3746 | (costly_order || | |
3747 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
3748 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
3749 | page = __alloc_pages_direct_compact(gfp_mask, order, |
3750 | alloc_flags, ac, | |
a5508cd8 | 3751 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
3752 | &compact_result); |
3753 | if (page) | |
3754 | goto got_pg; | |
3755 | ||
3eb2771b VB |
3756 | /* |
3757 | * Checks for costly allocations with __GFP_NORETRY, which | |
3758 | * includes THP page fault allocations | |
3759 | */ | |
282722b0 | 3760 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { |
a8161d1e VB |
3761 | /* |
3762 | * If compaction is deferred for high-order allocations, | |
3763 | * it is because sync compaction recently failed. If | |
3764 | * this is the case and the caller requested a THP | |
3765 | * allocation, we do not want to heavily disrupt the | |
3766 | * system, so we fail the allocation instead of entering | |
3767 | * direct reclaim. | |
3768 | */ | |
3769 | if (compact_result == COMPACT_DEFERRED) | |
3770 | goto nopage; | |
3771 | ||
a8161d1e | 3772 | /* |
3eb2771b VB |
3773 | * Looks like reclaim/compaction is worth trying, but |
3774 | * sync compaction could be very expensive, so keep | |
25160354 | 3775 | * using async compaction. |
a8161d1e | 3776 | */ |
a5508cd8 | 3777 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
3778 | } |
3779 | } | |
23771235 | 3780 | |
31a6c190 | 3781 | retry: |
23771235 | 3782 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
31a6c190 VB |
3783 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3784 | wake_all_kswapds(order, ac); | |
3785 | ||
23771235 VB |
3786 | if (gfp_pfmemalloc_allowed(gfp_mask)) |
3787 | alloc_flags = ALLOC_NO_WATERMARKS; | |
3788 | ||
e46e7b77 MG |
3789 | /* |
3790 | * Reset the zonelist iterators if memory policies can be ignored. | |
3791 | * These allocations are high priority and system rather than user | |
3792 | * orientated. | |
3793 | */ | |
23771235 | 3794 | if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) { |
e46e7b77 MG |
3795 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
3796 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
3797 | ac->high_zoneidx, ac->nodemask); | |
3798 | } | |
3799 | ||
23771235 | 3800 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 3801 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
3802 | if (page) |
3803 | goto got_pg; | |
1da177e4 | 3804 | |
d0164adc | 3805 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 3806 | if (!can_direct_reclaim) |
1da177e4 LT |
3807 | goto nopage; |
3808 | ||
9a67f648 MH |
3809 | /* Make sure we know about allocations which stall for too long */ |
3810 | if (time_after(jiffies, alloc_start + stall_timeout)) { | |
82251963 | 3811 | warn_alloc(gfp_mask & ~__GFP_NOWARN, ac->nodemask, |
9a67f648 MH |
3812 | "page allocation stalls for %ums, order:%u", |
3813 | jiffies_to_msecs(jiffies-alloc_start), order); | |
3814 | stall_timeout += 10 * HZ; | |
33d53103 | 3815 | } |
341ce06f | 3816 | |
9a67f648 MH |
3817 | /* Avoid recursion of direct reclaim */ |
3818 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
3819 | goto nopage; |
3820 | ||
a8161d1e VB |
3821 | /* Try direct reclaim and then allocating */ |
3822 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
3823 | &did_some_progress); | |
3824 | if (page) | |
3825 | goto got_pg; | |
3826 | ||
3827 | /* Try direct compaction and then allocating */ | |
a9263751 | 3828 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 3829 | compact_priority, &compact_result); |
56de7263 MG |
3830 | if (page) |
3831 | goto got_pg; | |
75f30861 | 3832 | |
9083905a JW |
3833 | /* Do not loop if specifically requested */ |
3834 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 3835 | goto nopage; |
9083905a | 3836 | |
0a0337e0 MH |
3837 | /* |
3838 | * Do not retry costly high order allocations unless they are | |
3839 | * __GFP_REPEAT | |
3840 | */ | |
282722b0 | 3841 | if (costly_order && !(gfp_mask & __GFP_REPEAT)) |
a8161d1e | 3842 | goto nopage; |
0a0337e0 | 3843 | |
0a0337e0 | 3844 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 3845 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
3846 | goto retry; |
3847 | ||
33c2d214 MH |
3848 | /* |
3849 | * It doesn't make any sense to retry for the compaction if the order-0 | |
3850 | * reclaim is not able to make any progress because the current | |
3851 | * implementation of the compaction depends on the sufficient amount | |
3852 | * of free memory (see __compaction_suitable) | |
3853 | */ | |
3854 | if (did_some_progress > 0 && | |
86a294a8 | 3855 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 3856 | compact_result, &compact_priority, |
d9436498 | 3857 | &compaction_retries)) |
33c2d214 MH |
3858 | goto retry; |
3859 | ||
e47483bc VB |
3860 | /* |
3861 | * It's possible we raced with cpuset update so the OOM would be | |
3862 | * premature (see below the nopage: label for full explanation). | |
3863 | */ | |
3864 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
3865 | goto retry_cpuset; | |
3866 | ||
9083905a JW |
3867 | /* Reclaim has failed us, start killing things */ |
3868 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
3869 | if (page) | |
3870 | goto got_pg; | |
3871 | ||
9a67f648 MH |
3872 | /* Avoid allocations with no watermarks from looping endlessly */ |
3873 | if (test_thread_flag(TIF_MEMDIE)) | |
3874 | goto nopage; | |
3875 | ||
9083905a | 3876 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
3877 | if (did_some_progress) { |
3878 | no_progress_loops = 0; | |
9083905a | 3879 | goto retry; |
0a0337e0 | 3880 | } |
9083905a | 3881 | |
1da177e4 | 3882 | nopage: |
5ce9bfef | 3883 | /* |
e47483bc VB |
3884 | * When updating a task's mems_allowed or mempolicy nodemask, it is |
3885 | * possible to race with parallel threads in such a way that our | |
3886 | * allocation can fail while the mask is being updated. If we are about | |
3887 | * to fail, check if the cpuset changed during allocation and if so, | |
3888 | * retry. | |
5ce9bfef VB |
3889 | */ |
3890 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
3891 | goto retry_cpuset; | |
3892 | ||
9a67f648 MH |
3893 | /* |
3894 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
3895 | * we always retry | |
3896 | */ | |
3897 | if (gfp_mask & __GFP_NOFAIL) { | |
3898 | /* | |
3899 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
3900 | * of any new users that actually require GFP_NOWAIT | |
3901 | */ | |
3902 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
3903 | goto fail; | |
3904 | ||
3905 | /* | |
3906 | * PF_MEMALLOC request from this context is rather bizarre | |
3907 | * because we cannot reclaim anything and only can loop waiting | |
3908 | * for somebody to do a work for us | |
3909 | */ | |
3910 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
3911 | ||
3912 | /* | |
3913 | * non failing costly orders are a hard requirement which we | |
3914 | * are not prepared for much so let's warn about these users | |
3915 | * so that we can identify them and convert them to something | |
3916 | * else. | |
3917 | */ | |
3918 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
3919 | ||
6c18ba7a MH |
3920 | /* |
3921 | * Help non-failing allocations by giving them access to memory | |
3922 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
3923 | * could deplete whole memory reserves which would just make | |
3924 | * the situation worse | |
3925 | */ | |
3926 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
3927 | if (page) | |
3928 | goto got_pg; | |
3929 | ||
9a67f648 MH |
3930 | cond_resched(); |
3931 | goto retry; | |
3932 | } | |
3933 | fail: | |
a8e99259 | 3934 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 3935 | "page allocation failure: order:%u", order); |
1da177e4 | 3936 | got_pg: |
072bb0aa | 3937 | return page; |
1da177e4 | 3938 | } |
11e33f6a | 3939 | |
9cd75558 MG |
3940 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
3941 | struct zonelist *zonelist, nodemask_t *nodemask, | |
3942 | struct alloc_context *ac, gfp_t *alloc_mask, | |
3943 | unsigned int *alloc_flags) | |
11e33f6a | 3944 | { |
9cd75558 MG |
3945 | ac->high_zoneidx = gfp_zone(gfp_mask); |
3946 | ac->zonelist = zonelist; | |
3947 | ac->nodemask = nodemask; | |
3948 | ac->migratetype = gfpflags_to_migratetype(gfp_mask); | |
11e33f6a | 3949 | |
682a3385 | 3950 | if (cpusets_enabled()) { |
9cd75558 | 3951 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
3952 | if (!ac->nodemask) |
3953 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
3954 | else |
3955 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
3956 | } |
3957 | ||
11e33f6a MG |
3958 | lockdep_trace_alloc(gfp_mask); |
3959 | ||
d0164adc | 3960 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
3961 | |
3962 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 3963 | return false; |
11e33f6a | 3964 | |
9cd75558 MG |
3965 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
3966 | *alloc_flags |= ALLOC_CMA; | |
3967 | ||
3968 | return true; | |
3969 | } | |
21bb9bd1 | 3970 | |
9cd75558 MG |
3971 | /* Determine whether to spread dirty pages and what the first usable zone */ |
3972 | static inline void finalise_ac(gfp_t gfp_mask, | |
3973 | unsigned int order, struct alloc_context *ac) | |
3974 | { | |
c9ab0c4f | 3975 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 3976 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 3977 | |
e46e7b77 MG |
3978 | /* |
3979 | * The preferred zone is used for statistics but crucially it is | |
3980 | * also used as the starting point for the zonelist iterator. It | |
3981 | * may get reset for allocations that ignore memory policies. | |
3982 | */ | |
9cd75558 MG |
3983 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
3984 | ac->high_zoneidx, ac->nodemask); | |
3985 | } | |
3986 | ||
3987 | /* | |
3988 | * This is the 'heart' of the zoned buddy allocator. | |
3989 | */ | |
3990 | struct page * | |
3991 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
3992 | struct zonelist *zonelist, nodemask_t *nodemask) | |
3993 | { | |
3994 | struct page *page; | |
3995 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
3996 | gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */ | |
3997 | struct alloc_context ac = { }; | |
3998 | ||
3999 | gfp_mask &= gfp_allowed_mask; | |
4000 | if (!prepare_alloc_pages(gfp_mask, order, zonelist, nodemask, &ac, &alloc_mask, &alloc_flags)) | |
4001 | return NULL; | |
4002 | ||
4003 | finalise_ac(gfp_mask, order, &ac); | |
5bb1b169 | 4004 | |
5117f45d | 4005 | /* First allocation attempt */ |
a9263751 | 4006 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4007 | if (likely(page)) |
4008 | goto out; | |
11e33f6a | 4009 | |
4fcb0971 | 4010 | /* |
7dea19f9 MH |
4011 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4012 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4013 | * from a particular context which has been marked by | |
4014 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4015 | */ |
7dea19f9 | 4016 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4017 | ac.spread_dirty_pages = false; |
23f086f9 | 4018 | |
4741526b MG |
4019 | /* |
4020 | * Restore the original nodemask if it was potentially replaced with | |
4021 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4022 | */ | |
e47483bc | 4023 | if (unlikely(ac.nodemask != nodemask)) |
4741526b | 4024 | ac.nodemask = nodemask; |
16096c25 | 4025 | |
4fcb0971 | 4026 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4027 | |
4fcb0971 | 4028 | out: |
c4159a75 VD |
4029 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
4030 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { | |
4031 | __free_pages(page, order); | |
4032 | page = NULL; | |
4949148a VD |
4033 | } |
4034 | ||
4fcb0971 MG |
4035 | if (kmemcheck_enabled && page) |
4036 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
4037 | ||
4038 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); | |
4039 | ||
11e33f6a | 4040 | return page; |
1da177e4 | 4041 | } |
d239171e | 4042 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4043 | |
4044 | /* | |
4045 | * Common helper functions. | |
4046 | */ | |
920c7a5d | 4047 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4048 | { |
945a1113 AM |
4049 | struct page *page; |
4050 | ||
4051 | /* | |
4052 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
4053 | * a highmem page | |
4054 | */ | |
4055 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
4056 | ||
1da177e4 LT |
4057 | page = alloc_pages(gfp_mask, order); |
4058 | if (!page) | |
4059 | return 0; | |
4060 | return (unsigned long) page_address(page); | |
4061 | } | |
1da177e4 LT |
4062 | EXPORT_SYMBOL(__get_free_pages); |
4063 | ||
920c7a5d | 4064 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4065 | { |
945a1113 | 4066 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4067 | } |
1da177e4 LT |
4068 | EXPORT_SYMBOL(get_zeroed_page); |
4069 | ||
920c7a5d | 4070 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 4071 | { |
b5810039 | 4072 | if (put_page_testzero(page)) { |
1da177e4 | 4073 | if (order == 0) |
b745bc85 | 4074 | free_hot_cold_page(page, false); |
1da177e4 LT |
4075 | else |
4076 | __free_pages_ok(page, order); | |
4077 | } | |
4078 | } | |
4079 | ||
4080 | EXPORT_SYMBOL(__free_pages); | |
4081 | ||
920c7a5d | 4082 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4083 | { |
4084 | if (addr != 0) { | |
725d704e | 4085 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4086 | __free_pages(virt_to_page((void *)addr), order); |
4087 | } | |
4088 | } | |
4089 | ||
4090 | EXPORT_SYMBOL(free_pages); | |
4091 | ||
b63ae8ca AD |
4092 | /* |
4093 | * Page Fragment: | |
4094 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4095 | * within a 0 or higher order page. Multiple fragments within that page | |
4096 | * are individually refcounted, in the page's reference counter. | |
4097 | * | |
4098 | * The page_frag functions below provide a simple allocation framework for | |
4099 | * page fragments. This is used by the network stack and network device | |
4100 | * drivers to provide a backing region of memory for use as either an | |
4101 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4102 | */ | |
2976db80 AD |
4103 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4104 | gfp_t gfp_mask) | |
b63ae8ca AD |
4105 | { |
4106 | struct page *page = NULL; | |
4107 | gfp_t gfp = gfp_mask; | |
4108 | ||
4109 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4110 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4111 | __GFP_NOMEMALLOC; | |
4112 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4113 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4114 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4115 | #endif | |
4116 | if (unlikely(!page)) | |
4117 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4118 | ||
4119 | nc->va = page ? page_address(page) : NULL; | |
4120 | ||
4121 | return page; | |
4122 | } | |
4123 | ||
2976db80 | 4124 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4125 | { |
4126 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4127 | ||
4128 | if (page_ref_sub_and_test(page, count)) { | |
2976db80 AD |
4129 | unsigned int order = compound_order(page); |
4130 | ||
44fdffd7 AD |
4131 | if (order == 0) |
4132 | free_hot_cold_page(page, false); | |
4133 | else | |
4134 | __free_pages_ok(page, order); | |
4135 | } | |
4136 | } | |
2976db80 | 4137 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4138 | |
8c2dd3e4 AD |
4139 | void *page_frag_alloc(struct page_frag_cache *nc, |
4140 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4141 | { |
4142 | unsigned int size = PAGE_SIZE; | |
4143 | struct page *page; | |
4144 | int offset; | |
4145 | ||
4146 | if (unlikely(!nc->va)) { | |
4147 | refill: | |
2976db80 | 4148 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4149 | if (!page) |
4150 | return NULL; | |
4151 | ||
4152 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4153 | /* if size can vary use size else just use PAGE_SIZE */ | |
4154 | size = nc->size; | |
4155 | #endif | |
4156 | /* Even if we own the page, we do not use atomic_set(). | |
4157 | * This would break get_page_unless_zero() users. | |
4158 | */ | |
fe896d18 | 4159 | page_ref_add(page, size - 1); |
b63ae8ca AD |
4160 | |
4161 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4162 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
4163 | nc->pagecnt_bias = size; |
4164 | nc->offset = size; | |
4165 | } | |
4166 | ||
4167 | offset = nc->offset - fragsz; | |
4168 | if (unlikely(offset < 0)) { | |
4169 | page = virt_to_page(nc->va); | |
4170 | ||
fe896d18 | 4171 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4172 | goto refill; |
4173 | ||
4174 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4175 | /* if size can vary use size else just use PAGE_SIZE */ | |
4176 | size = nc->size; | |
4177 | #endif | |
4178 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 4179 | set_page_count(page, size); |
b63ae8ca AD |
4180 | |
4181 | /* reset page count bias and offset to start of new frag */ | |
4182 | nc->pagecnt_bias = size; | |
4183 | offset = size - fragsz; | |
4184 | } | |
4185 | ||
4186 | nc->pagecnt_bias--; | |
4187 | nc->offset = offset; | |
4188 | ||
4189 | return nc->va + offset; | |
4190 | } | |
8c2dd3e4 | 4191 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
4192 | |
4193 | /* | |
4194 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4195 | */ | |
8c2dd3e4 | 4196 | void page_frag_free(void *addr) |
b63ae8ca AD |
4197 | { |
4198 | struct page *page = virt_to_head_page(addr); | |
4199 | ||
4200 | if (unlikely(put_page_testzero(page))) | |
4201 | __free_pages_ok(page, compound_order(page)); | |
4202 | } | |
8c2dd3e4 | 4203 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4204 | |
d00181b9 KS |
4205 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4206 | size_t size) | |
ee85c2e1 AK |
4207 | { |
4208 | if (addr) { | |
4209 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
4210 | unsigned long used = addr + PAGE_ALIGN(size); | |
4211 | ||
4212 | split_page(virt_to_page((void *)addr), order); | |
4213 | while (used < alloc_end) { | |
4214 | free_page(used); | |
4215 | used += PAGE_SIZE; | |
4216 | } | |
4217 | } | |
4218 | return (void *)addr; | |
4219 | } | |
4220 | ||
2be0ffe2 TT |
4221 | /** |
4222 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4223 | * @size: the number of bytes to allocate | |
4224 | * @gfp_mask: GFP flags for the allocation | |
4225 | * | |
4226 | * This function is similar to alloc_pages(), except that it allocates the | |
4227 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4228 | * allocate memory in power-of-two pages. | |
4229 | * | |
4230 | * This function is also limited by MAX_ORDER. | |
4231 | * | |
4232 | * Memory allocated by this function must be released by free_pages_exact(). | |
4233 | */ | |
4234 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4235 | { | |
4236 | unsigned int order = get_order(size); | |
4237 | unsigned long addr; | |
4238 | ||
4239 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 4240 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4241 | } |
4242 | EXPORT_SYMBOL(alloc_pages_exact); | |
4243 | ||
ee85c2e1 AK |
4244 | /** |
4245 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4246 | * pages on a node. | |
b5e6ab58 | 4247 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
4248 | * @size: the number of bytes to allocate |
4249 | * @gfp_mask: GFP flags for the allocation | |
4250 | * | |
4251 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4252 | * back. | |
ee85c2e1 | 4253 | */ |
e1931811 | 4254 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4255 | { |
d00181b9 | 4256 | unsigned int order = get_order(size); |
ee85c2e1 AK |
4257 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
4258 | if (!p) | |
4259 | return NULL; | |
4260 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4261 | } | |
ee85c2e1 | 4262 | |
2be0ffe2 TT |
4263 | /** |
4264 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4265 | * @virt: the value returned by alloc_pages_exact. | |
4266 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4267 | * | |
4268 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4269 | */ | |
4270 | void free_pages_exact(void *virt, size_t size) | |
4271 | { | |
4272 | unsigned long addr = (unsigned long)virt; | |
4273 | unsigned long end = addr + PAGE_ALIGN(size); | |
4274 | ||
4275 | while (addr < end) { | |
4276 | free_page(addr); | |
4277 | addr += PAGE_SIZE; | |
4278 | } | |
4279 | } | |
4280 | EXPORT_SYMBOL(free_pages_exact); | |
4281 | ||
e0fb5815 ZY |
4282 | /** |
4283 | * nr_free_zone_pages - count number of pages beyond high watermark | |
4284 | * @offset: The zone index of the highest zone | |
4285 | * | |
4286 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
4287 | * high watermark within all zones at or below a given zone index. For each | |
4288 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
4289 | * |
4290 | * nr_free_zone_pages = managed_pages - high_pages | |
e0fb5815 | 4291 | */ |
ebec3862 | 4292 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 4293 | { |
dd1a239f | 4294 | struct zoneref *z; |
54a6eb5c MG |
4295 | struct zone *zone; |
4296 | ||
e310fd43 | 4297 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 4298 | unsigned long sum = 0; |
1da177e4 | 4299 | |
0e88460d | 4300 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 4301 | |
54a6eb5c | 4302 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 4303 | unsigned long size = zone->managed_pages; |
41858966 | 4304 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
4305 | if (size > high) |
4306 | sum += size - high; | |
1da177e4 LT |
4307 | } |
4308 | ||
4309 | return sum; | |
4310 | } | |
4311 | ||
e0fb5815 ZY |
4312 | /** |
4313 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
4314 | * | |
4315 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
4316 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 4317 | */ |
ebec3862 | 4318 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 4319 | { |
af4ca457 | 4320 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 4321 | } |
c2f1a551 | 4322 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 4323 | |
e0fb5815 ZY |
4324 | /** |
4325 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
4326 | * | |
4327 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
4328 | * high watermark within all zones. | |
1da177e4 | 4329 | */ |
ebec3862 | 4330 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 4331 | { |
2a1e274a | 4332 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 4333 | } |
08e0f6a9 CL |
4334 | |
4335 | static inline void show_node(struct zone *zone) | |
1da177e4 | 4336 | { |
e5adfffc | 4337 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 4338 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 4339 | } |
1da177e4 | 4340 | |
d02bd27b IR |
4341 | long si_mem_available(void) |
4342 | { | |
4343 | long available; | |
4344 | unsigned long pagecache; | |
4345 | unsigned long wmark_low = 0; | |
4346 | unsigned long pages[NR_LRU_LISTS]; | |
4347 | struct zone *zone; | |
4348 | int lru; | |
4349 | ||
4350 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 4351 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
4352 | |
4353 | for_each_zone(zone) | |
4354 | wmark_low += zone->watermark[WMARK_LOW]; | |
4355 | ||
4356 | /* | |
4357 | * Estimate the amount of memory available for userspace allocations, | |
4358 | * without causing swapping. | |
4359 | */ | |
4360 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; | |
4361 | ||
4362 | /* | |
4363 | * Not all the page cache can be freed, otherwise the system will | |
4364 | * start swapping. Assume at least half of the page cache, or the | |
4365 | * low watermark worth of cache, needs to stay. | |
4366 | */ | |
4367 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
4368 | pagecache -= min(pagecache / 2, wmark_low); | |
4369 | available += pagecache; | |
4370 | ||
4371 | /* | |
4372 | * Part of the reclaimable slab consists of items that are in use, | |
4373 | * and cannot be freed. Cap this estimate at the low watermark. | |
4374 | */ | |
4375 | available += global_page_state(NR_SLAB_RECLAIMABLE) - | |
4376 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); | |
4377 | ||
4378 | if (available < 0) | |
4379 | available = 0; | |
4380 | return available; | |
4381 | } | |
4382 | EXPORT_SYMBOL_GPL(si_mem_available); | |
4383 | ||
1da177e4 LT |
4384 | void si_meminfo(struct sysinfo *val) |
4385 | { | |
4386 | val->totalram = totalram_pages; | |
11fb9989 | 4387 | val->sharedram = global_node_page_state(NR_SHMEM); |
d23ad423 | 4388 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 4389 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
4390 | val->totalhigh = totalhigh_pages; |
4391 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
4392 | val->mem_unit = PAGE_SIZE; |
4393 | } | |
4394 | ||
4395 | EXPORT_SYMBOL(si_meminfo); | |
4396 | ||
4397 | #ifdef CONFIG_NUMA | |
4398 | void si_meminfo_node(struct sysinfo *val, int nid) | |
4399 | { | |
cdd91a77 JL |
4400 | int zone_type; /* needs to be signed */ |
4401 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
4402 | unsigned long managed_highpages = 0; |
4403 | unsigned long free_highpages = 0; | |
1da177e4 LT |
4404 | pg_data_t *pgdat = NODE_DATA(nid); |
4405 | ||
cdd91a77 JL |
4406 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
4407 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
4408 | val->totalram = managed_pages; | |
11fb9989 | 4409 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 4410 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 4411 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
4412 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4413 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4414 | ||
4415 | if (is_highmem(zone)) { | |
4416 | managed_highpages += zone->managed_pages; | |
4417 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
4418 | } | |
4419 | } | |
4420 | val->totalhigh = managed_highpages; | |
4421 | val->freehigh = free_highpages; | |
98d2b0eb | 4422 | #else |
fc2bd799 JK |
4423 | val->totalhigh = managed_highpages; |
4424 | val->freehigh = free_highpages; | |
98d2b0eb | 4425 | #endif |
1da177e4 LT |
4426 | val->mem_unit = PAGE_SIZE; |
4427 | } | |
4428 | #endif | |
4429 | ||
ddd588b5 | 4430 | /* |
7bf02ea2 DR |
4431 | * Determine whether the node should be displayed or not, depending on whether |
4432 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 4433 | */ |
9af744d7 | 4434 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 4435 | { |
ddd588b5 | 4436 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 4437 | return false; |
ddd588b5 | 4438 | |
9af744d7 MH |
4439 | /* |
4440 | * no node mask - aka implicit memory numa policy. Do not bother with | |
4441 | * the synchronization - read_mems_allowed_begin - because we do not | |
4442 | * have to be precise here. | |
4443 | */ | |
4444 | if (!nodemask) | |
4445 | nodemask = &cpuset_current_mems_allowed; | |
4446 | ||
4447 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
4448 | } |
4449 | ||
1da177e4 LT |
4450 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4451 | ||
377e4f16 RV |
4452 | static void show_migration_types(unsigned char type) |
4453 | { | |
4454 | static const char types[MIGRATE_TYPES] = { | |
4455 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 4456 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
4457 | [MIGRATE_RECLAIMABLE] = 'E', |
4458 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
4459 | #ifdef CONFIG_CMA |
4460 | [MIGRATE_CMA] = 'C', | |
4461 | #endif | |
194159fb | 4462 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 4463 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 4464 | #endif |
377e4f16 RV |
4465 | }; |
4466 | char tmp[MIGRATE_TYPES + 1]; | |
4467 | char *p = tmp; | |
4468 | int i; | |
4469 | ||
4470 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
4471 | if (type & (1 << i)) | |
4472 | *p++ = types[i]; | |
4473 | } | |
4474 | ||
4475 | *p = '\0'; | |
1f84a18f | 4476 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
4477 | } |
4478 | ||
1da177e4 LT |
4479 | /* |
4480 | * Show free area list (used inside shift_scroll-lock stuff) | |
4481 | * We also calculate the percentage fragmentation. We do this by counting the | |
4482 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
4483 | * |
4484 | * Bits in @filter: | |
4485 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
4486 | * cpuset. | |
1da177e4 | 4487 | */ |
9af744d7 | 4488 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 4489 | { |
d1bfcdb8 | 4490 | unsigned long free_pcp = 0; |
c7241913 | 4491 | int cpu; |
1da177e4 | 4492 | struct zone *zone; |
599d0c95 | 4493 | pg_data_t *pgdat; |
1da177e4 | 4494 | |
ee99c71c | 4495 | for_each_populated_zone(zone) { |
9af744d7 | 4496 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4497 | continue; |
d1bfcdb8 | 4498 | |
761b0677 KK |
4499 | for_each_online_cpu(cpu) |
4500 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
4501 | } |
4502 | ||
a731286d KM |
4503 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4504 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
4505 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4506 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 4507 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 4508 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
4509 | global_node_page_state(NR_ACTIVE_ANON), |
4510 | global_node_page_state(NR_INACTIVE_ANON), | |
4511 | global_node_page_state(NR_ISOLATED_ANON), | |
4512 | global_node_page_state(NR_ACTIVE_FILE), | |
4513 | global_node_page_state(NR_INACTIVE_FILE), | |
4514 | global_node_page_state(NR_ISOLATED_FILE), | |
4515 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
4516 | global_node_page_state(NR_FILE_DIRTY), |
4517 | global_node_page_state(NR_WRITEBACK), | |
4518 | global_node_page_state(NR_UNSTABLE_NFS), | |
3701b033 KM |
4519 | global_page_state(NR_SLAB_RECLAIMABLE), |
4520 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 4521 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 4522 | global_node_page_state(NR_SHMEM), |
a25700a5 | 4523 | global_page_state(NR_PAGETABLE), |
d1ce749a | 4524 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
4525 | global_page_state(NR_FREE_PAGES), |
4526 | free_pcp, | |
d1ce749a | 4527 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4528 | |
599d0c95 | 4529 | for_each_online_pgdat(pgdat) { |
9af744d7 | 4530 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
4531 | continue; |
4532 | ||
599d0c95 MG |
4533 | printk("Node %d" |
4534 | " active_anon:%lukB" | |
4535 | " inactive_anon:%lukB" | |
4536 | " active_file:%lukB" | |
4537 | " inactive_file:%lukB" | |
4538 | " unevictable:%lukB" | |
4539 | " isolated(anon):%lukB" | |
4540 | " isolated(file):%lukB" | |
50658e2e | 4541 | " mapped:%lukB" |
11fb9989 MG |
4542 | " dirty:%lukB" |
4543 | " writeback:%lukB" | |
4544 | " shmem:%lukB" | |
4545 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4546 | " shmem_thp: %lukB" | |
4547 | " shmem_pmdmapped: %lukB" | |
4548 | " anon_thp: %lukB" | |
4549 | #endif | |
4550 | " writeback_tmp:%lukB" | |
4551 | " unstable:%lukB" | |
599d0c95 MG |
4552 | " all_unreclaimable? %s" |
4553 | "\n", | |
4554 | pgdat->node_id, | |
4555 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
4556 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
4557 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
4558 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
4559 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
4560 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
4561 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 4562 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
4563 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
4564 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 4565 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
4566 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4567 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
4568 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
4569 | * HPAGE_PMD_NR), | |
4570 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
4571 | #endif | |
11fb9989 MG |
4572 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
4573 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), | |
c73322d0 JW |
4574 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
4575 | "yes" : "no"); | |
599d0c95 MG |
4576 | } |
4577 | ||
ee99c71c | 4578 | for_each_populated_zone(zone) { |
1da177e4 LT |
4579 | int i; |
4580 | ||
9af744d7 | 4581 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4582 | continue; |
d1bfcdb8 KK |
4583 | |
4584 | free_pcp = 0; | |
4585 | for_each_online_cpu(cpu) | |
4586 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
4587 | ||
1da177e4 | 4588 | show_node(zone); |
1f84a18f JP |
4589 | printk(KERN_CONT |
4590 | "%s" | |
1da177e4 LT |
4591 | " free:%lukB" |
4592 | " min:%lukB" | |
4593 | " low:%lukB" | |
4594 | " high:%lukB" | |
71c799f4 MK |
4595 | " active_anon:%lukB" |
4596 | " inactive_anon:%lukB" | |
4597 | " active_file:%lukB" | |
4598 | " inactive_file:%lukB" | |
4599 | " unevictable:%lukB" | |
5a1c84b4 | 4600 | " writepending:%lukB" |
1da177e4 | 4601 | " present:%lukB" |
9feedc9d | 4602 | " managed:%lukB" |
4a0aa73f | 4603 | " mlocked:%lukB" |
4a0aa73f KM |
4604 | " slab_reclaimable:%lukB" |
4605 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 4606 | " kernel_stack:%lukB" |
4a0aa73f | 4607 | " pagetables:%lukB" |
4a0aa73f | 4608 | " bounce:%lukB" |
d1bfcdb8 KK |
4609 | " free_pcp:%lukB" |
4610 | " local_pcp:%ukB" | |
d1ce749a | 4611 | " free_cma:%lukB" |
1da177e4 LT |
4612 | "\n", |
4613 | zone->name, | |
88f5acf8 | 4614 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
4615 | K(min_wmark_pages(zone)), |
4616 | K(low_wmark_pages(zone)), | |
4617 | K(high_wmark_pages(zone)), | |
71c799f4 MK |
4618 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
4619 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
4620 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
4621 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
4622 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 4623 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 4624 | K(zone->present_pages), |
9feedc9d | 4625 | K(zone->managed_pages), |
4a0aa73f | 4626 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f KM |
4627 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
4628 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
d30dd8be | 4629 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4a0aa73f | 4630 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 4631 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
4632 | K(free_pcp), |
4633 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 4634 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
4635 | printk("lowmem_reserve[]:"); |
4636 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
4637 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
4638 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
4639 | } |
4640 | ||
ee99c71c | 4641 | for_each_populated_zone(zone) { |
d00181b9 KS |
4642 | unsigned int order; |
4643 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 4644 | unsigned char types[MAX_ORDER]; |
1da177e4 | 4645 | |
9af744d7 | 4646 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4647 | continue; |
1da177e4 | 4648 | show_node(zone); |
1f84a18f | 4649 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
4650 | |
4651 | spin_lock_irqsave(&zone->lock, flags); | |
4652 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
4653 | struct free_area *area = &zone->free_area[order]; |
4654 | int type; | |
4655 | ||
4656 | nr[order] = area->nr_free; | |
8f9de51a | 4657 | total += nr[order] << order; |
377e4f16 RV |
4658 | |
4659 | types[order] = 0; | |
4660 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
4661 | if (!list_empty(&area->free_list[type])) | |
4662 | types[order] |= 1 << type; | |
4663 | } | |
1da177e4 LT |
4664 | } |
4665 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 4666 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
4667 | printk(KERN_CONT "%lu*%lukB ", |
4668 | nr[order], K(1UL) << order); | |
377e4f16 RV |
4669 | if (nr[order]) |
4670 | show_migration_types(types[order]); | |
4671 | } | |
1f84a18f | 4672 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
4673 | } |
4674 | ||
949f7ec5 DR |
4675 | hugetlb_show_meminfo(); |
4676 | ||
11fb9989 | 4677 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 4678 | |
1da177e4 LT |
4679 | show_swap_cache_info(); |
4680 | } | |
4681 | ||
19770b32 MG |
4682 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4683 | { | |
4684 | zoneref->zone = zone; | |
4685 | zoneref->zone_idx = zone_idx(zone); | |
4686 | } | |
4687 | ||
1da177e4 LT |
4688 | /* |
4689 | * Builds allocation fallback zone lists. | |
1a93205b CL |
4690 | * |
4691 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 4692 | */ |
f0c0b2b8 | 4693 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 4694 | int nr_zones) |
1da177e4 | 4695 | { |
1a93205b | 4696 | struct zone *zone; |
bc732f1d | 4697 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
4698 | |
4699 | do { | |
2f6726e5 | 4700 | zone_type--; |
070f8032 | 4701 | zone = pgdat->node_zones + zone_type; |
6aa303de | 4702 | if (managed_zone(zone)) { |
dd1a239f MG |
4703 | zoneref_set_zone(zone, |
4704 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 4705 | check_highest_zone(zone_type); |
1da177e4 | 4706 | } |
2f6726e5 | 4707 | } while (zone_type); |
bc732f1d | 4708 | |
070f8032 | 4709 | return nr_zones; |
1da177e4 LT |
4710 | } |
4711 | ||
f0c0b2b8 KH |
4712 | |
4713 | /* | |
4714 | * zonelist_order: | |
4715 | * 0 = automatic detection of better ordering. | |
4716 | * 1 = order by ([node] distance, -zonetype) | |
4717 | * 2 = order by (-zonetype, [node] distance) | |
4718 | * | |
4719 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
4720 | * the same zonelist. So only NUMA can configure this param. | |
4721 | */ | |
4722 | #define ZONELIST_ORDER_DEFAULT 0 | |
4723 | #define ZONELIST_ORDER_NODE 1 | |
4724 | #define ZONELIST_ORDER_ZONE 2 | |
4725 | ||
4726 | /* zonelist order in the kernel. | |
4727 | * set_zonelist_order() will set this to NODE or ZONE. | |
4728 | */ | |
4729 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4730 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
4731 | ||
4732 | ||
1da177e4 | 4733 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
4734 | /* The value user specified ....changed by config */ |
4735 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4736 | /* string for sysctl */ | |
4737 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
4738 | char numa_zonelist_order[16] = "default"; | |
4739 | ||
4740 | /* | |
4741 | * interface for configure zonelist ordering. | |
4742 | * command line option "numa_zonelist_order" | |
4743 | * = "[dD]efault - default, automatic configuration. | |
4744 | * = "[nN]ode - order by node locality, then by zone within node | |
4745 | * = "[zZ]one - order by zone, then by locality within zone | |
4746 | */ | |
4747 | ||
4748 | static int __parse_numa_zonelist_order(char *s) | |
4749 | { | |
4750 | if (*s == 'd' || *s == 'D') { | |
4751 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4752 | } else if (*s == 'n' || *s == 'N') { | |
4753 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
4754 | } else if (*s == 'z' || *s == 'Z') { | |
4755 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
4756 | } else { | |
1170532b | 4757 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
f0c0b2b8 KH |
4758 | return -EINVAL; |
4759 | } | |
4760 | return 0; | |
4761 | } | |
4762 | ||
4763 | static __init int setup_numa_zonelist_order(char *s) | |
4764 | { | |
ecb256f8 VL |
4765 | int ret; |
4766 | ||
4767 | if (!s) | |
4768 | return 0; | |
4769 | ||
4770 | ret = __parse_numa_zonelist_order(s); | |
4771 | if (ret == 0) | |
4772 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
4773 | ||
4774 | return ret; | |
f0c0b2b8 KH |
4775 | } |
4776 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
4777 | ||
4778 | /* | |
4779 | * sysctl handler for numa_zonelist_order | |
4780 | */ | |
cccad5b9 | 4781 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 4782 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
4783 | loff_t *ppos) |
4784 | { | |
4785 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
4786 | int ret; | |
443c6f14 | 4787 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 4788 | |
443c6f14 | 4789 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
4790 | if (write) { |
4791 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
4792 | ret = -EINVAL; | |
4793 | goto out; | |
4794 | } | |
4795 | strcpy(saved_string, (char *)table->data); | |
4796 | } | |
8d65af78 | 4797 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 4798 | if (ret) |
443c6f14 | 4799 | goto out; |
f0c0b2b8 KH |
4800 | if (write) { |
4801 | int oldval = user_zonelist_order; | |
dacbde09 CG |
4802 | |
4803 | ret = __parse_numa_zonelist_order((char *)table->data); | |
4804 | if (ret) { | |
f0c0b2b8 KH |
4805 | /* |
4806 | * bogus value. restore saved string | |
4807 | */ | |
dacbde09 | 4808 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
4809 | NUMA_ZONELIST_ORDER_LEN); |
4810 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
4811 | } else if (oldval != user_zonelist_order) { |
4812 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 4813 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
4814 | mutex_unlock(&zonelists_mutex); |
4815 | } | |
f0c0b2b8 | 4816 | } |
443c6f14 AK |
4817 | out: |
4818 | mutex_unlock(&zl_order_mutex); | |
4819 | return ret; | |
f0c0b2b8 KH |
4820 | } |
4821 | ||
4822 | ||
62bc62a8 | 4823 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
4824 | static int node_load[MAX_NUMNODES]; |
4825 | ||
1da177e4 | 4826 | /** |
4dc3b16b | 4827 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
4828 | * @node: node whose fallback list we're appending |
4829 | * @used_node_mask: nodemask_t of already used nodes | |
4830 | * | |
4831 | * We use a number of factors to determine which is the next node that should | |
4832 | * appear on a given node's fallback list. The node should not have appeared | |
4833 | * already in @node's fallback list, and it should be the next closest node | |
4834 | * according to the distance array (which contains arbitrary distance values | |
4835 | * from each node to each node in the system), and should also prefer nodes | |
4836 | * with no CPUs, since presumably they'll have very little allocation pressure | |
4837 | * on them otherwise. | |
4838 | * It returns -1 if no node is found. | |
4839 | */ | |
f0c0b2b8 | 4840 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 4841 | { |
4cf808eb | 4842 | int n, val; |
1da177e4 | 4843 | int min_val = INT_MAX; |
00ef2d2f | 4844 | int best_node = NUMA_NO_NODE; |
a70f7302 | 4845 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 4846 | |
4cf808eb LT |
4847 | /* Use the local node if we haven't already */ |
4848 | if (!node_isset(node, *used_node_mask)) { | |
4849 | node_set(node, *used_node_mask); | |
4850 | return node; | |
4851 | } | |
1da177e4 | 4852 | |
4b0ef1fe | 4853 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
4854 | |
4855 | /* Don't want a node to appear more than once */ | |
4856 | if (node_isset(n, *used_node_mask)) | |
4857 | continue; | |
4858 | ||
1da177e4 LT |
4859 | /* Use the distance array to find the distance */ |
4860 | val = node_distance(node, n); | |
4861 | ||
4cf808eb LT |
4862 | /* Penalize nodes under us ("prefer the next node") */ |
4863 | val += (n < node); | |
4864 | ||
1da177e4 | 4865 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
4866 | tmp = cpumask_of_node(n); |
4867 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
4868 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4869 | ||
4870 | /* Slight preference for less loaded node */ | |
4871 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
4872 | val += node_load[n]; | |
4873 | ||
4874 | if (val < min_val) { | |
4875 | min_val = val; | |
4876 | best_node = n; | |
4877 | } | |
4878 | } | |
4879 | ||
4880 | if (best_node >= 0) | |
4881 | node_set(best_node, *used_node_mask); | |
4882 | ||
4883 | return best_node; | |
4884 | } | |
4885 | ||
f0c0b2b8 KH |
4886 | |
4887 | /* | |
4888 | * Build zonelists ordered by node and zones within node. | |
4889 | * This results in maximum locality--normal zone overflows into local | |
4890 | * DMA zone, if any--but risks exhausting DMA zone. | |
4891 | */ | |
4892 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 4893 | { |
f0c0b2b8 | 4894 | int j; |
1da177e4 | 4895 | struct zonelist *zonelist; |
f0c0b2b8 | 4896 | |
c9634cf0 | 4897 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
dd1a239f | 4898 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 4899 | ; |
bc732f1d | 4900 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
4901 | zonelist->_zonerefs[j].zone = NULL; |
4902 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
4903 | } |
4904 | ||
523b9458 CL |
4905 | /* |
4906 | * Build gfp_thisnode zonelists | |
4907 | */ | |
4908 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
4909 | { | |
523b9458 CL |
4910 | int j; |
4911 | struct zonelist *zonelist; | |
4912 | ||
c9634cf0 | 4913 | zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK]; |
bc732f1d | 4914 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
4915 | zonelist->_zonerefs[j].zone = NULL; |
4916 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
4917 | } |
4918 | ||
f0c0b2b8 KH |
4919 | /* |
4920 | * Build zonelists ordered by zone and nodes within zones. | |
4921 | * This results in conserving DMA zone[s] until all Normal memory is | |
4922 | * exhausted, but results in overflowing to remote node while memory | |
4923 | * may still exist in local DMA zone. | |
4924 | */ | |
4925 | static int node_order[MAX_NUMNODES]; | |
4926 | ||
4927 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
4928 | { | |
f0c0b2b8 KH |
4929 | int pos, j, node; |
4930 | int zone_type; /* needs to be signed */ | |
4931 | struct zone *z; | |
4932 | struct zonelist *zonelist; | |
4933 | ||
c9634cf0 | 4934 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
54a6eb5c MG |
4935 | pos = 0; |
4936 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
4937 | for (j = 0; j < nr_nodes; j++) { | |
4938 | node = node_order[j]; | |
4939 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
6aa303de | 4940 | if (managed_zone(z)) { |
dd1a239f MG |
4941 | zoneref_set_zone(z, |
4942 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 4943 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
4944 | } |
4945 | } | |
f0c0b2b8 | 4946 | } |
dd1a239f MG |
4947 | zonelist->_zonerefs[pos].zone = NULL; |
4948 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
4949 | } |
4950 | ||
3193913c MG |
4951 | #if defined(CONFIG_64BIT) |
4952 | /* | |
4953 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
4954 | * penalty to every machine in case the specialised case applies. Default | |
4955 | * to Node-ordering on 64-bit NUMA machines | |
4956 | */ | |
4957 | static int default_zonelist_order(void) | |
4958 | { | |
4959 | return ZONELIST_ORDER_NODE; | |
4960 | } | |
4961 | #else | |
4962 | /* | |
4963 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
4964 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
4965 | * then reclaim will occur more frequency increasing stalls and potentially | |
4966 | * be easier to OOM if a large percentage of the zone is under writeback or | |
4967 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
4968 | * Hence, default to zone ordering on 32-bit. | |
4969 | */ | |
f0c0b2b8 KH |
4970 | static int default_zonelist_order(void) |
4971 | { | |
f0c0b2b8 KH |
4972 | return ZONELIST_ORDER_ZONE; |
4973 | } | |
3193913c | 4974 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
4975 | |
4976 | static void set_zonelist_order(void) | |
4977 | { | |
4978 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
4979 | current_zonelist_order = default_zonelist_order(); | |
4980 | else | |
4981 | current_zonelist_order = user_zonelist_order; | |
4982 | } | |
4983 | ||
4984 | static void build_zonelists(pg_data_t *pgdat) | |
4985 | { | |
c00eb15a | 4986 | int i, node, load; |
1da177e4 | 4987 | nodemask_t used_mask; |
f0c0b2b8 KH |
4988 | int local_node, prev_node; |
4989 | struct zonelist *zonelist; | |
d00181b9 | 4990 | unsigned int order = current_zonelist_order; |
1da177e4 LT |
4991 | |
4992 | /* initialize zonelists */ | |
523b9458 | 4993 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 4994 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
4995 | zonelist->_zonerefs[0].zone = NULL; |
4996 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
4997 | } |
4998 | ||
4999 | /* NUMA-aware ordering of nodes */ | |
5000 | local_node = pgdat->node_id; | |
62bc62a8 | 5001 | load = nr_online_nodes; |
1da177e4 LT |
5002 | prev_node = local_node; |
5003 | nodes_clear(used_mask); | |
f0c0b2b8 | 5004 | |
f0c0b2b8 | 5005 | memset(node_order, 0, sizeof(node_order)); |
c00eb15a | 5006 | i = 0; |
f0c0b2b8 | 5007 | |
1da177e4 LT |
5008 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5009 | /* | |
5010 | * We don't want to pressure a particular node. | |
5011 | * So adding penalty to the first node in same | |
5012 | * distance group to make it round-robin. | |
5013 | */ | |
957f822a DR |
5014 | if (node_distance(local_node, node) != |
5015 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5016 | node_load[node] = load; |
5017 | ||
1da177e4 LT |
5018 | prev_node = node; |
5019 | load--; | |
f0c0b2b8 KH |
5020 | if (order == ZONELIST_ORDER_NODE) |
5021 | build_zonelists_in_node_order(pgdat, node); | |
5022 | else | |
c00eb15a | 5023 | node_order[i++] = node; /* remember order */ |
f0c0b2b8 | 5024 | } |
1da177e4 | 5025 | |
f0c0b2b8 KH |
5026 | if (order == ZONELIST_ORDER_ZONE) { |
5027 | /* calculate node order -- i.e., DMA last! */ | |
c00eb15a | 5028 | build_zonelists_in_zone_order(pgdat, i); |
1da177e4 | 5029 | } |
523b9458 CL |
5030 | |
5031 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
5032 | } |
5033 | ||
7aac7898 LS |
5034 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5035 | /* | |
5036 | * Return node id of node used for "local" allocations. | |
5037 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5038 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5039 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5040 | */ | |
5041 | int local_memory_node(int node) | |
5042 | { | |
c33d6c06 | 5043 | struct zoneref *z; |
7aac7898 | 5044 | |
c33d6c06 | 5045 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5046 | gfp_zone(GFP_KERNEL), |
c33d6c06 MG |
5047 | NULL); |
5048 | return z->zone->node; | |
7aac7898 LS |
5049 | } |
5050 | #endif | |
f0c0b2b8 | 5051 | |
6423aa81 JK |
5052 | static void setup_min_unmapped_ratio(void); |
5053 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5054 | #else /* CONFIG_NUMA */ |
5055 | ||
f0c0b2b8 KH |
5056 | static void set_zonelist_order(void) |
5057 | { | |
5058 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
5059 | } | |
5060 | ||
5061 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 5062 | { |
19655d34 | 5063 | int node, local_node; |
54a6eb5c MG |
5064 | enum zone_type j; |
5065 | struct zonelist *zonelist; | |
1da177e4 LT |
5066 | |
5067 | local_node = pgdat->node_id; | |
1da177e4 | 5068 | |
c9634cf0 | 5069 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
bc732f1d | 5070 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 5071 | |
54a6eb5c MG |
5072 | /* |
5073 | * Now we build the zonelist so that it contains the zones | |
5074 | * of all the other nodes. | |
5075 | * We don't want to pressure a particular node, so when | |
5076 | * building the zones for node N, we make sure that the | |
5077 | * zones coming right after the local ones are those from | |
5078 | * node N+1 (modulo N) | |
5079 | */ | |
5080 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5081 | if (!node_online(node)) | |
5082 | continue; | |
bc732f1d | 5083 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 5084 | } |
54a6eb5c MG |
5085 | for (node = 0; node < local_node; node++) { |
5086 | if (!node_online(node)) | |
5087 | continue; | |
bc732f1d | 5088 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
5089 | } |
5090 | ||
dd1a239f MG |
5091 | zonelist->_zonerefs[j].zone = NULL; |
5092 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
5093 | } |
5094 | ||
5095 | #endif /* CONFIG_NUMA */ | |
5096 | ||
99dcc3e5 CL |
5097 | /* |
5098 | * Boot pageset table. One per cpu which is going to be used for all | |
5099 | * zones and all nodes. The parameters will be set in such a way | |
5100 | * that an item put on a list will immediately be handed over to | |
5101 | * the buddy list. This is safe since pageset manipulation is done | |
5102 | * with interrupts disabled. | |
5103 | * | |
5104 | * The boot_pagesets must be kept even after bootup is complete for | |
5105 | * unused processors and/or zones. They do play a role for bootstrapping | |
5106 | * hotplugged processors. | |
5107 | * | |
5108 | * zoneinfo_show() and maybe other functions do | |
5109 | * not check if the processor is online before following the pageset pointer. | |
5110 | * Other parts of the kernel may not check if the zone is available. | |
5111 | */ | |
5112 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5113 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 5114 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 5115 | |
4eaf3f64 HL |
5116 | /* |
5117 | * Global mutex to protect against size modification of zonelists | |
5118 | * as well as to serialize pageset setup for the new populated zone. | |
5119 | */ | |
5120 | DEFINE_MUTEX(zonelists_mutex); | |
5121 | ||
9b1a4d38 | 5122 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 5123 | static int __build_all_zonelists(void *data) |
1da177e4 | 5124 | { |
6811378e | 5125 | int nid; |
99dcc3e5 | 5126 | int cpu; |
9adb62a5 | 5127 | pg_data_t *self = data; |
9276b1bc | 5128 | |
7f9cfb31 BL |
5129 | #ifdef CONFIG_NUMA |
5130 | memset(node_load, 0, sizeof(node_load)); | |
5131 | #endif | |
9adb62a5 JL |
5132 | |
5133 | if (self && !node_online(self->node_id)) { | |
5134 | build_zonelists(self); | |
9adb62a5 JL |
5135 | } |
5136 | ||
9276b1bc | 5137 | for_each_online_node(nid) { |
7ea1530a CL |
5138 | pg_data_t *pgdat = NODE_DATA(nid); |
5139 | ||
5140 | build_zonelists(pgdat); | |
9276b1bc | 5141 | } |
99dcc3e5 CL |
5142 | |
5143 | /* | |
5144 | * Initialize the boot_pagesets that are going to be used | |
5145 | * for bootstrapping processors. The real pagesets for | |
5146 | * each zone will be allocated later when the per cpu | |
5147 | * allocator is available. | |
5148 | * | |
5149 | * boot_pagesets are used also for bootstrapping offline | |
5150 | * cpus if the system is already booted because the pagesets | |
5151 | * are needed to initialize allocators on a specific cpu too. | |
5152 | * F.e. the percpu allocator needs the page allocator which | |
5153 | * needs the percpu allocator in order to allocate its pagesets | |
5154 | * (a chicken-egg dilemma). | |
5155 | */ | |
7aac7898 | 5156 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
5157 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
5158 | ||
7aac7898 LS |
5159 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5160 | /* | |
5161 | * We now know the "local memory node" for each node-- | |
5162 | * i.e., the node of the first zone in the generic zonelist. | |
5163 | * Set up numa_mem percpu variable for on-line cpus. During | |
5164 | * boot, only the boot cpu should be on-line; we'll init the | |
5165 | * secondary cpus' numa_mem as they come on-line. During | |
5166 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5167 | */ | |
5168 | if (cpu_online(cpu)) | |
5169 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
5170 | #endif | |
5171 | } | |
5172 | ||
6811378e YG |
5173 | return 0; |
5174 | } | |
5175 | ||
061f67bc RV |
5176 | static noinline void __init |
5177 | build_all_zonelists_init(void) | |
5178 | { | |
5179 | __build_all_zonelists(NULL); | |
5180 | mminit_verify_zonelist(); | |
5181 | cpuset_init_current_mems_allowed(); | |
5182 | } | |
5183 | ||
4eaf3f64 HL |
5184 | /* |
5185 | * Called with zonelists_mutex held always | |
5186 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
5187 | * |
5188 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
5189 | * [we're only called with non-NULL zone through __meminit paths] and | |
5190 | * (2) call of __init annotated helper build_all_zonelists_init | |
5191 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 5192 | */ |
9adb62a5 | 5193 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 5194 | { |
f0c0b2b8 KH |
5195 | set_zonelist_order(); |
5196 | ||
6811378e | 5197 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 5198 | build_all_zonelists_init(); |
6811378e | 5199 | } else { |
e9959f0f | 5200 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
5201 | if (zone) |
5202 | setup_zone_pageset(zone); | |
e9959f0f | 5203 | #endif |
dd1895e2 CS |
5204 | /* we have to stop all cpus to guarantee there is no user |
5205 | of zonelist */ | |
9adb62a5 | 5206 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
5207 | /* cpuset refresh routine should be here */ |
5208 | } | |
bd1e22b8 | 5209 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5210 | /* |
5211 | * Disable grouping by mobility if the number of pages in the | |
5212 | * system is too low to allow the mechanism to work. It would be | |
5213 | * more accurate, but expensive to check per-zone. This check is | |
5214 | * made on memory-hotadd so a system can start with mobility | |
5215 | * disabled and enable it later | |
5216 | */ | |
d9c23400 | 5217 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5218 | page_group_by_mobility_disabled = 1; |
5219 | else | |
5220 | page_group_by_mobility_disabled = 0; | |
5221 | ||
756a025f JP |
5222 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
5223 | nr_online_nodes, | |
5224 | zonelist_order_name[current_zonelist_order], | |
5225 | page_group_by_mobility_disabled ? "off" : "on", | |
5226 | vm_total_pages); | |
f0c0b2b8 | 5227 | #ifdef CONFIG_NUMA |
f88dfff5 | 5228 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5229 | #endif |
1da177e4 LT |
5230 | } |
5231 | ||
1da177e4 LT |
5232 | /* |
5233 | * Initially all pages are reserved - free ones are freed | |
5234 | * up by free_all_bootmem() once the early boot process is | |
5235 | * done. Non-atomic initialization, single-pass. | |
5236 | */ | |
c09b4240 | 5237 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 5238 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 5239 | { |
4b94ffdc | 5240 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
29751f69 | 5241 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 5242 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 5243 | unsigned long pfn; |
3a80a7fa | 5244 | unsigned long nr_initialised = 0; |
342332e6 TI |
5245 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5246 | struct memblock_region *r = NULL, *tmp; | |
5247 | #endif | |
1da177e4 | 5248 | |
22b31eec HD |
5249 | if (highest_memmap_pfn < end_pfn - 1) |
5250 | highest_memmap_pfn = end_pfn - 1; | |
5251 | ||
4b94ffdc DW |
5252 | /* |
5253 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
5254 | * memory | |
5255 | */ | |
5256 | if (altmap && start_pfn == altmap->base_pfn) | |
5257 | start_pfn += altmap->reserve; | |
5258 | ||
cbe8dd4a | 5259 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 5260 | /* |
b72d0ffb AM |
5261 | * There can be holes in boot-time mem_map[]s handed to this |
5262 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5263 | */ |
b72d0ffb AM |
5264 | if (context != MEMMAP_EARLY) |
5265 | goto not_early; | |
5266 | ||
b92df1de PB |
5267 | if (!early_pfn_valid(pfn)) { |
5268 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
5269 | /* | |
5270 | * Skip to the pfn preceding the next valid one (or | |
5271 | * end_pfn), such that we hit a valid pfn (or end_pfn) | |
5272 | * on our next iteration of the loop. | |
5273 | */ | |
5274 | pfn = memblock_next_valid_pfn(pfn, end_pfn) - 1; | |
5275 | #endif | |
b72d0ffb | 5276 | continue; |
b92df1de | 5277 | } |
b72d0ffb AM |
5278 | if (!early_pfn_in_nid(pfn, nid)) |
5279 | continue; | |
5280 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
5281 | break; | |
342332e6 TI |
5282 | |
5283 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
5284 | /* |
5285 | * Check given memblock attribute by firmware which can affect | |
5286 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
5287 | * mirrored, it's an overlapped memmap init. skip it. | |
5288 | */ | |
5289 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5290 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
5291 | for_each_memblock(memory, tmp) | |
5292 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
5293 | break; | |
5294 | r = tmp; | |
5295 | } | |
5296 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
5297 | memblock_is_mirror(r)) { | |
5298 | /* already initialized as NORMAL */ | |
5299 | pfn = memblock_region_memory_end_pfn(r); | |
5300 | continue; | |
342332e6 | 5301 | } |
a2f3aa02 | 5302 | } |
b72d0ffb | 5303 | #endif |
ac5d2539 | 5304 | |
b72d0ffb | 5305 | not_early: |
ac5d2539 MG |
5306 | /* |
5307 | * Mark the block movable so that blocks are reserved for | |
5308 | * movable at startup. This will force kernel allocations | |
5309 | * to reserve their blocks rather than leaking throughout | |
5310 | * the address space during boot when many long-lived | |
974a786e | 5311 | * kernel allocations are made. |
ac5d2539 MG |
5312 | * |
5313 | * bitmap is created for zone's valid pfn range. but memmap | |
5314 | * can be created for invalid pages (for alignment) | |
5315 | * check here not to call set_pageblock_migratetype() against | |
5316 | * pfn out of zone. | |
5317 | */ | |
5318 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
5319 | struct page *page = pfn_to_page(pfn); | |
5320 | ||
5321 | __init_single_page(page, pfn, zone, nid); | |
5322 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5323 | } else { | |
5324 | __init_single_pfn(pfn, zone, nid); | |
5325 | } | |
1da177e4 LT |
5326 | } |
5327 | } | |
5328 | ||
1e548deb | 5329 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 5330 | { |
7aeb09f9 | 5331 | unsigned int order, t; |
b2a0ac88 MG |
5332 | for_each_migratetype_order(order, t) { |
5333 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
5334 | zone->free_area[order].nr_free = 0; |
5335 | } | |
5336 | } | |
5337 | ||
5338 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
5339 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 5340 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
5341 | #endif |
5342 | ||
7cd2b0a3 | 5343 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 5344 | { |
3a6be87f | 5345 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
5346 | int batch; |
5347 | ||
5348 | /* | |
5349 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 5350 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
5351 | * |
5352 | * OK, so we don't know how big the cache is. So guess. | |
5353 | */ | |
b40da049 | 5354 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
5355 | if (batch * PAGE_SIZE > 512 * 1024) |
5356 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
5357 | batch /= 4; /* We effectively *= 4 below */ |
5358 | if (batch < 1) | |
5359 | batch = 1; | |
5360 | ||
5361 | /* | |
0ceaacc9 NP |
5362 | * Clamp the batch to a 2^n - 1 value. Having a power |
5363 | * of 2 value was found to be more likely to have | |
5364 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 5365 | * |
0ceaacc9 NP |
5366 | * For example if 2 tasks are alternately allocating |
5367 | * batches of pages, one task can end up with a lot | |
5368 | * of pages of one half of the possible page colors | |
5369 | * and the other with pages of the other colors. | |
e7c8d5c9 | 5370 | */ |
9155203a | 5371 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 5372 | |
e7c8d5c9 | 5373 | return batch; |
3a6be87f DH |
5374 | |
5375 | #else | |
5376 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5377 | * conditions. | |
5378 | * | |
5379 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5380 | * of contiguous memory as there's no hardware page translation to | |
5381 | * assemble apparent contiguous memory from discontiguous pages. | |
5382 | * | |
5383 | * Queueing large contiguous runs of pages for batching, however, | |
5384 | * causes the pages to actually be freed in smaller chunks. As there | |
5385 | * can be a significant delay between the individual batches being | |
5386 | * recycled, this leads to the once large chunks of space being | |
5387 | * fragmented and becoming unavailable for high-order allocations. | |
5388 | */ | |
5389 | return 0; | |
5390 | #endif | |
e7c8d5c9 CL |
5391 | } |
5392 | ||
8d7a8fa9 CS |
5393 | /* |
5394 | * pcp->high and pcp->batch values are related and dependent on one another: | |
5395 | * ->batch must never be higher then ->high. | |
5396 | * The following function updates them in a safe manner without read side | |
5397 | * locking. | |
5398 | * | |
5399 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
5400 | * those fields changing asynchronously (acording the the above rule). | |
5401 | * | |
5402 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5403 | * outside of boot time (or some other assurance that no concurrent updaters | |
5404 | * exist). | |
5405 | */ | |
5406 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
5407 | unsigned long batch) | |
5408 | { | |
5409 | /* start with a fail safe value for batch */ | |
5410 | pcp->batch = 1; | |
5411 | smp_wmb(); | |
5412 | ||
5413 | /* Update high, then batch, in order */ | |
5414 | pcp->high = high; | |
5415 | smp_wmb(); | |
5416 | ||
5417 | pcp->batch = batch; | |
5418 | } | |
5419 | ||
3664033c | 5420 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
5421 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5422 | { | |
8d7a8fa9 | 5423 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
5424 | } |
5425 | ||
88c90dbc | 5426 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
5427 | { |
5428 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 5429 | int migratetype; |
2caaad41 | 5430 | |
1c6fe946 MD |
5431 | memset(p, 0, sizeof(*p)); |
5432 | ||
3dfa5721 | 5433 | pcp = &p->pcp; |
2caaad41 | 5434 | pcp->count = 0; |
5f8dcc21 MG |
5435 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5436 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
5437 | } |
5438 | ||
88c90dbc CS |
5439 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5440 | { | |
5441 | pageset_init(p); | |
5442 | pageset_set_batch(p, batch); | |
5443 | } | |
5444 | ||
8ad4b1fb | 5445 | /* |
3664033c | 5446 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
5447 | * to the value high for the pageset p. |
5448 | */ | |
3664033c | 5449 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
5450 | unsigned long high) |
5451 | { | |
8d7a8fa9 CS |
5452 | unsigned long batch = max(1UL, high / 4); |
5453 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
5454 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 5455 | |
8d7a8fa9 | 5456 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
5457 | } |
5458 | ||
7cd2b0a3 DR |
5459 | static void pageset_set_high_and_batch(struct zone *zone, |
5460 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 5461 | { |
56cef2b8 | 5462 | if (percpu_pagelist_fraction) |
3664033c | 5463 | pageset_set_high(pcp, |
56cef2b8 CS |
5464 | (zone->managed_pages / |
5465 | percpu_pagelist_fraction)); | |
5466 | else | |
5467 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
5468 | } | |
5469 | ||
169f6c19 CS |
5470 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5471 | { | |
5472 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
5473 | ||
5474 | pageset_init(pcp); | |
5475 | pageset_set_high_and_batch(zone, pcp); | |
5476 | } | |
5477 | ||
4ed7e022 | 5478 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
5479 | { |
5480 | int cpu; | |
319774e2 | 5481 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5482 | for_each_possible_cpu(cpu) |
5483 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5484 | } |
5485 | ||
2caaad41 | 5486 | /* |
99dcc3e5 CL |
5487 | * Allocate per cpu pagesets and initialize them. |
5488 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5489 | */ |
99dcc3e5 | 5490 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5491 | { |
b4911ea2 | 5492 | struct pglist_data *pgdat; |
99dcc3e5 | 5493 | struct zone *zone; |
e7c8d5c9 | 5494 | |
319774e2 WF |
5495 | for_each_populated_zone(zone) |
5496 | setup_zone_pageset(zone); | |
b4911ea2 MG |
5497 | |
5498 | for_each_online_pgdat(pgdat) | |
5499 | pgdat->per_cpu_nodestats = | |
5500 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
5501 | } |
5502 | ||
c09b4240 | 5503 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5504 | { |
99dcc3e5 CL |
5505 | /* |
5506 | * per cpu subsystem is not up at this point. The following code | |
5507 | * relies on the ability of the linker to provide the | |
5508 | * offset of a (static) per cpu variable into the per cpu area. | |
5509 | */ | |
5510 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5511 | |
b38a8725 | 5512 | if (populated_zone(zone)) |
99dcc3e5 CL |
5513 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5514 | zone->name, zone->present_pages, | |
5515 | zone_batchsize(zone)); | |
ed8ece2e DH |
5516 | } |
5517 | ||
4ed7e022 | 5518 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5519 | unsigned long zone_start_pfn, |
b171e409 | 5520 | unsigned long size) |
ed8ece2e DH |
5521 | { |
5522 | struct pglist_data *pgdat = zone->zone_pgdat; | |
9dcb8b68 | 5523 | |
ed8ece2e DH |
5524 | pgdat->nr_zones = zone_idx(zone) + 1; |
5525 | ||
ed8ece2e DH |
5526 | zone->zone_start_pfn = zone_start_pfn; |
5527 | ||
708614e6 MG |
5528 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5529 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5530 | pgdat->node_id, | |
5531 | (unsigned long)zone_idx(zone), | |
5532 | zone_start_pfn, (zone_start_pfn + size)); | |
5533 | ||
1e548deb | 5534 | zone_init_free_lists(zone); |
9dcb8b68 | 5535 | zone->initialized = 1; |
718127cc YG |
5536 | |
5537 | return 0; | |
ed8ece2e DH |
5538 | } |
5539 | ||
0ee332c1 | 5540 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5541 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5542 | |
c713216d MG |
5543 | /* |
5544 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5545 | */ |
8a942fde MG |
5546 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5547 | struct mminit_pfnnid_cache *state) | |
c713216d | 5548 | { |
c13291a5 | 5549 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5550 | int nid; |
7c243c71 | 5551 | |
8a942fde MG |
5552 | if (state->last_start <= pfn && pfn < state->last_end) |
5553 | return state->last_nid; | |
c713216d | 5554 | |
e76b63f8 YL |
5555 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5556 | if (nid != -1) { | |
8a942fde MG |
5557 | state->last_start = start_pfn; |
5558 | state->last_end = end_pfn; | |
5559 | state->last_nid = nid; | |
e76b63f8 YL |
5560 | } |
5561 | ||
5562 | return nid; | |
c713216d MG |
5563 | } |
5564 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5565 | ||
c713216d | 5566 | /** |
6782832e | 5567 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5568 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5569 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5570 | * |
7d018176 ZZ |
5571 | * If an architecture guarantees that all ranges registered contain no holes |
5572 | * and may be freed, this this function may be used instead of calling | |
5573 | * memblock_free_early_nid() manually. | |
c713216d | 5574 | */ |
c13291a5 | 5575 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5576 | { |
c13291a5 TH |
5577 | unsigned long start_pfn, end_pfn; |
5578 | int i, this_nid; | |
edbe7d23 | 5579 | |
c13291a5 TH |
5580 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5581 | start_pfn = min(start_pfn, max_low_pfn); | |
5582 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5583 | |
c13291a5 | 5584 | if (start_pfn < end_pfn) |
6782832e SS |
5585 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5586 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5587 | this_nid); | |
edbe7d23 | 5588 | } |
edbe7d23 | 5589 | } |
edbe7d23 | 5590 | |
c713216d MG |
5591 | /** |
5592 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5593 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5594 | * |
7d018176 ZZ |
5595 | * If an architecture guarantees that all ranges registered contain no holes and may |
5596 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5597 | */ |
5598 | void __init sparse_memory_present_with_active_regions(int nid) | |
5599 | { | |
c13291a5 TH |
5600 | unsigned long start_pfn, end_pfn; |
5601 | int i, this_nid; | |
c713216d | 5602 | |
c13291a5 TH |
5603 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5604 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5605 | } |
5606 | ||
5607 | /** | |
5608 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5609 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5610 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5611 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5612 | * |
5613 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5614 | * provided by memblock_set_node(). If called for a node |
c713216d | 5615 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5616 | * PFNs will be 0. |
c713216d | 5617 | */ |
a3142c8e | 5618 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5619 | unsigned long *start_pfn, unsigned long *end_pfn) |
5620 | { | |
c13291a5 | 5621 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5622 | int i; |
c13291a5 | 5623 | |
c713216d MG |
5624 | *start_pfn = -1UL; |
5625 | *end_pfn = 0; | |
5626 | ||
c13291a5 TH |
5627 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5628 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5629 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5630 | } |
5631 | ||
633c0666 | 5632 | if (*start_pfn == -1UL) |
c713216d | 5633 | *start_pfn = 0; |
c713216d MG |
5634 | } |
5635 | ||
2a1e274a MG |
5636 | /* |
5637 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5638 | * assumption is made that zones within a node are ordered in monotonic | |
5639 | * increasing memory addresses so that the "highest" populated zone is used | |
5640 | */ | |
b69a7288 | 5641 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5642 | { |
5643 | int zone_index; | |
5644 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5645 | if (zone_index == ZONE_MOVABLE) | |
5646 | continue; | |
5647 | ||
5648 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5649 | arch_zone_lowest_possible_pfn[zone_index]) | |
5650 | break; | |
5651 | } | |
5652 | ||
5653 | VM_BUG_ON(zone_index == -1); | |
5654 | movable_zone = zone_index; | |
5655 | } | |
5656 | ||
5657 | /* | |
5658 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5659 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5660 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5661 | * in each node depending on the size of each node and how evenly kernelcore | |
5662 | * is distributed. This helper function adjusts the zone ranges | |
5663 | * provided by the architecture for a given node by using the end of the | |
5664 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5665 | * zones within a node are in order of monotonic increases memory addresses | |
5666 | */ | |
b69a7288 | 5667 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5668 | unsigned long zone_type, |
5669 | unsigned long node_start_pfn, | |
5670 | unsigned long node_end_pfn, | |
5671 | unsigned long *zone_start_pfn, | |
5672 | unsigned long *zone_end_pfn) | |
5673 | { | |
5674 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5675 | if (zone_movable_pfn[nid]) { | |
5676 | /* Size ZONE_MOVABLE */ | |
5677 | if (zone_type == ZONE_MOVABLE) { | |
5678 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5679 | *zone_end_pfn = min(node_end_pfn, | |
5680 | arch_zone_highest_possible_pfn[movable_zone]); | |
5681 | ||
e506b996 XQ |
5682 | /* Adjust for ZONE_MOVABLE starting within this range */ |
5683 | } else if (!mirrored_kernelcore && | |
5684 | *zone_start_pfn < zone_movable_pfn[nid] && | |
5685 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
5686 | *zone_end_pfn = zone_movable_pfn[nid]; | |
5687 | ||
2a1e274a MG |
5688 | /* Check if this whole range is within ZONE_MOVABLE */ |
5689 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5690 | *zone_start_pfn = *zone_end_pfn; | |
5691 | } | |
5692 | } | |
5693 | ||
c713216d MG |
5694 | /* |
5695 | * Return the number of pages a zone spans in a node, including holes | |
5696 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5697 | */ | |
6ea6e688 | 5698 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5699 | unsigned long zone_type, |
7960aedd ZY |
5700 | unsigned long node_start_pfn, |
5701 | unsigned long node_end_pfn, | |
d91749c1 TI |
5702 | unsigned long *zone_start_pfn, |
5703 | unsigned long *zone_end_pfn, | |
c713216d MG |
5704 | unsigned long *ignored) |
5705 | { | |
b5685e92 | 5706 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5707 | if (!node_start_pfn && !node_end_pfn) |
5708 | return 0; | |
5709 | ||
7960aedd | 5710 | /* Get the start and end of the zone */ |
d91749c1 TI |
5711 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5712 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5713 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5714 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5715 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5716 | |
5717 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5718 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5719 | return 0; |
5720 | ||
5721 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5722 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5723 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5724 | |
5725 | /* Return the spanned pages */ | |
d91749c1 | 5726 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5727 | } |
5728 | ||
5729 | /* | |
5730 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5731 | * then all holes in the requested range will be accounted for. |
c713216d | 5732 | */ |
32996250 | 5733 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5734 | unsigned long range_start_pfn, |
5735 | unsigned long range_end_pfn) | |
5736 | { | |
96e907d1 TH |
5737 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5738 | unsigned long start_pfn, end_pfn; | |
5739 | int i; | |
c713216d | 5740 | |
96e907d1 TH |
5741 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5742 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5743 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5744 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5745 | } |
96e907d1 | 5746 | return nr_absent; |
c713216d MG |
5747 | } |
5748 | ||
5749 | /** | |
5750 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5751 | * @start_pfn: The start PFN to start searching for holes | |
5752 | * @end_pfn: The end PFN to stop searching for holes | |
5753 | * | |
88ca3b94 | 5754 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5755 | */ |
5756 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5757 | unsigned long end_pfn) | |
5758 | { | |
5759 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5760 | } | |
5761 | ||
5762 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5763 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5764 | unsigned long zone_type, |
7960aedd ZY |
5765 | unsigned long node_start_pfn, |
5766 | unsigned long node_end_pfn, | |
c713216d MG |
5767 | unsigned long *ignored) |
5768 | { | |
96e907d1 TH |
5769 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5770 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5771 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5772 | unsigned long nr_absent; |
9c7cd687 | 5773 | |
b5685e92 | 5774 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5775 | if (!node_start_pfn && !node_end_pfn) |
5776 | return 0; | |
5777 | ||
96e907d1 TH |
5778 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5779 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5780 | |
2a1e274a MG |
5781 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5782 | node_start_pfn, node_end_pfn, | |
5783 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5784 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5785 | ||
5786 | /* | |
5787 | * ZONE_MOVABLE handling. | |
5788 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
5789 | * and vice versa. | |
5790 | */ | |
e506b996 XQ |
5791 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
5792 | unsigned long start_pfn, end_pfn; | |
5793 | struct memblock_region *r; | |
5794 | ||
5795 | for_each_memblock(memory, r) { | |
5796 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
5797 | zone_start_pfn, zone_end_pfn); | |
5798 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
5799 | zone_start_pfn, zone_end_pfn); | |
5800 | ||
5801 | if (zone_type == ZONE_MOVABLE && | |
5802 | memblock_is_mirror(r)) | |
5803 | nr_absent += end_pfn - start_pfn; | |
5804 | ||
5805 | if (zone_type == ZONE_NORMAL && | |
5806 | !memblock_is_mirror(r)) | |
5807 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
5808 | } |
5809 | } | |
5810 | ||
5811 | return nr_absent; | |
c713216d | 5812 | } |
0e0b864e | 5813 | |
0ee332c1 | 5814 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 5815 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5816 | unsigned long zone_type, |
7960aedd ZY |
5817 | unsigned long node_start_pfn, |
5818 | unsigned long node_end_pfn, | |
d91749c1 TI |
5819 | unsigned long *zone_start_pfn, |
5820 | unsigned long *zone_end_pfn, | |
c713216d MG |
5821 | unsigned long *zones_size) |
5822 | { | |
d91749c1 TI |
5823 | unsigned int zone; |
5824 | ||
5825 | *zone_start_pfn = node_start_pfn; | |
5826 | for (zone = 0; zone < zone_type; zone++) | |
5827 | *zone_start_pfn += zones_size[zone]; | |
5828 | ||
5829 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
5830 | ||
c713216d MG |
5831 | return zones_size[zone_type]; |
5832 | } | |
5833 | ||
6ea6e688 | 5834 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5835 | unsigned long zone_type, |
7960aedd ZY |
5836 | unsigned long node_start_pfn, |
5837 | unsigned long node_end_pfn, | |
c713216d MG |
5838 | unsigned long *zholes_size) |
5839 | { | |
5840 | if (!zholes_size) | |
5841 | return 0; | |
5842 | ||
5843 | return zholes_size[zone_type]; | |
5844 | } | |
20e6926d | 5845 | |
0ee332c1 | 5846 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5847 | |
a3142c8e | 5848 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
5849 | unsigned long node_start_pfn, |
5850 | unsigned long node_end_pfn, | |
5851 | unsigned long *zones_size, | |
5852 | unsigned long *zholes_size) | |
c713216d | 5853 | { |
febd5949 | 5854 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
5855 | enum zone_type i; |
5856 | ||
febd5949 GZ |
5857 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5858 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 5859 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 5860 | unsigned long size, real_size; |
c713216d | 5861 | |
febd5949 GZ |
5862 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5863 | node_start_pfn, | |
5864 | node_end_pfn, | |
d91749c1 TI |
5865 | &zone_start_pfn, |
5866 | &zone_end_pfn, | |
febd5949 GZ |
5867 | zones_size); |
5868 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
5869 | node_start_pfn, node_end_pfn, |
5870 | zholes_size); | |
d91749c1 TI |
5871 | if (size) |
5872 | zone->zone_start_pfn = zone_start_pfn; | |
5873 | else | |
5874 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
5875 | zone->spanned_pages = size; |
5876 | zone->present_pages = real_size; | |
5877 | ||
5878 | totalpages += size; | |
5879 | realtotalpages += real_size; | |
5880 | } | |
5881 | ||
5882 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
5883 | pgdat->node_present_pages = realtotalpages; |
5884 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
5885 | realtotalpages); | |
5886 | } | |
5887 | ||
835c134e MG |
5888 | #ifndef CONFIG_SPARSEMEM |
5889 | /* | |
5890 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
5891 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5892 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
5893 | * round what is now in bits to nearest long in bits, then return it in |
5894 | * bytes. | |
5895 | */ | |
7c45512d | 5896 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
5897 | { |
5898 | unsigned long usemapsize; | |
5899 | ||
7c45512d | 5900 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
5901 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
5902 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
5903 | usemapsize *= NR_PAGEBLOCK_BITS; |
5904 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
5905 | ||
5906 | return usemapsize / 8; | |
5907 | } | |
5908 | ||
5909 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
5910 | struct zone *zone, |
5911 | unsigned long zone_start_pfn, | |
5912 | unsigned long zonesize) | |
835c134e | 5913 | { |
7c45512d | 5914 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 5915 | zone->pageblock_flags = NULL; |
58a01a45 | 5916 | if (usemapsize) |
6782832e SS |
5917 | zone->pageblock_flags = |
5918 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5919 | pgdat->node_id); | |
835c134e MG |
5920 | } |
5921 | #else | |
7c45512d LT |
5922 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5923 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5924 | #endif /* CONFIG_SPARSEMEM */ |
5925 | ||
d9c23400 | 5926 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5927 | |
d9c23400 | 5928 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5929 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5930 | { |
955c1cd7 AM |
5931 | unsigned int order; |
5932 | ||
d9c23400 MG |
5933 | /* Check that pageblock_nr_pages has not already been setup */ |
5934 | if (pageblock_order) | |
5935 | return; | |
5936 | ||
955c1cd7 AM |
5937 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5938 | order = HUGETLB_PAGE_ORDER; | |
5939 | else | |
5940 | order = MAX_ORDER - 1; | |
5941 | ||
d9c23400 MG |
5942 | /* |
5943 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5944 | * This value may be variable depending on boot parameters on IA64 and |
5945 | * powerpc. | |
d9c23400 MG |
5946 | */ |
5947 | pageblock_order = order; | |
5948 | } | |
5949 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5950 | ||
ba72cb8c MG |
5951 | /* |
5952 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5953 | * is unused as pageblock_order is set at compile-time. See |
5954 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5955 | * the kernel config | |
ba72cb8c | 5956 | */ |
15ca220e | 5957 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5958 | { |
ba72cb8c | 5959 | } |
d9c23400 MG |
5960 | |
5961 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5962 | ||
01cefaef JL |
5963 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5964 | unsigned long present_pages) | |
5965 | { | |
5966 | unsigned long pages = spanned_pages; | |
5967 | ||
5968 | /* | |
5969 | * Provide a more accurate estimation if there are holes within | |
5970 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5971 | * zone, each populated memory region may cost us one or two extra | |
5972 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 5973 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
5974 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
5975 | */ | |
5976 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5977 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5978 | pages = present_pages; | |
5979 | ||
5980 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5981 | } | |
5982 | ||
1da177e4 LT |
5983 | /* |
5984 | * Set up the zone data structures: | |
5985 | * - mark all pages reserved | |
5986 | * - mark all memory queues empty | |
5987 | * - clear the memory bitmaps | |
6527af5d MK |
5988 | * |
5989 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 5990 | */ |
7f3eb55b | 5991 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 5992 | { |
2f1b6248 | 5993 | enum zone_type j; |
ed8ece2e | 5994 | int nid = pgdat->node_id; |
718127cc | 5995 | int ret; |
1da177e4 | 5996 | |
208d54e5 | 5997 | pgdat_resize_init(pgdat); |
8177a420 AA |
5998 | #ifdef CONFIG_NUMA_BALANCING |
5999 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
6000 | pgdat->numabalancing_migrate_nr_pages = 0; | |
6001 | pgdat->numabalancing_migrate_next_window = jiffies; | |
a3d0a918 KS |
6002 | #endif |
6003 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6004 | spin_lock_init(&pgdat->split_queue_lock); | |
6005 | INIT_LIST_HEAD(&pgdat->split_queue); | |
6006 | pgdat->split_queue_len = 0; | |
8177a420 | 6007 | #endif |
1da177e4 | 6008 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6009 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
698b1b30 VB |
6010 | #ifdef CONFIG_COMPACTION |
6011 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6012 | #endif | |
eefa864b | 6013 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6014 | spin_lock_init(&pgdat->lru_lock); |
a9dd0a83 | 6015 | lruvec_init(node_lruvec(pgdat)); |
5f63b720 | 6016 | |
1da177e4 LT |
6017 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6018 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 6019 | unsigned long size, realsize, freesize, memmap_pages; |
d91749c1 | 6020 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6021 | |
febd5949 GZ |
6022 | size = zone->spanned_pages; |
6023 | realsize = freesize = zone->present_pages; | |
1da177e4 | 6024 | |
0e0b864e | 6025 | /* |
9feedc9d | 6026 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6027 | * is used by this zone for memmap. This affects the watermark |
6028 | * and per-cpu initialisations | |
6029 | */ | |
01cefaef | 6030 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
6031 | if (!is_highmem_idx(j)) { |
6032 | if (freesize >= memmap_pages) { | |
6033 | freesize -= memmap_pages; | |
6034 | if (memmap_pages) | |
6035 | printk(KERN_DEBUG | |
6036 | " %s zone: %lu pages used for memmap\n", | |
6037 | zone_names[j], memmap_pages); | |
6038 | } else | |
1170532b | 6039 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6040 | zone_names[j], memmap_pages, freesize); |
6041 | } | |
0e0b864e | 6042 | |
6267276f | 6043 | /* Account for reserved pages */ |
9feedc9d JL |
6044 | if (j == 0 && freesize > dma_reserve) { |
6045 | freesize -= dma_reserve; | |
d903ef9f | 6046 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6047 | zone_names[0], dma_reserve); |
0e0b864e MG |
6048 | } |
6049 | ||
98d2b0eb | 6050 | if (!is_highmem_idx(j)) |
9feedc9d | 6051 | nr_kernel_pages += freesize; |
01cefaef JL |
6052 | /* Charge for highmem memmap if there are enough kernel pages */ |
6053 | else if (nr_kernel_pages > memmap_pages * 2) | |
6054 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6055 | nr_all_pages += freesize; |
1da177e4 | 6056 | |
9feedc9d JL |
6057 | /* |
6058 | * Set an approximate value for lowmem here, it will be adjusted | |
6059 | * when the bootmem allocator frees pages into the buddy system. | |
6060 | * And all highmem pages will be managed by the buddy system. | |
6061 | */ | |
6062 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 6063 | #ifdef CONFIG_NUMA |
d5f541ed | 6064 | zone->node = nid; |
9614634f | 6065 | #endif |
1da177e4 | 6066 | zone->name = zone_names[j]; |
a52633d8 | 6067 | zone->zone_pgdat = pgdat; |
1da177e4 | 6068 | spin_lock_init(&zone->lock); |
bdc8cb98 | 6069 | zone_seqlock_init(zone); |
ed8ece2e | 6070 | zone_pcp_init(zone); |
81c0a2bb | 6071 | |
1da177e4 LT |
6072 | if (!size) |
6073 | continue; | |
6074 | ||
955c1cd7 | 6075 | set_pageblock_order(); |
7c45512d | 6076 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
b171e409 | 6077 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
718127cc | 6078 | BUG_ON(ret); |
76cdd58e | 6079 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6080 | } |
6081 | } | |
6082 | ||
bd721ea7 | 6083 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6084 | { |
b0aeba74 | 6085 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6086 | unsigned long __maybe_unused offset = 0; |
6087 | ||
1da177e4 LT |
6088 | /* Skip empty nodes */ |
6089 | if (!pgdat->node_spanned_pages) | |
6090 | return; | |
6091 | ||
d41dee36 | 6092 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
b0aeba74 TL |
6093 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6094 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6095 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6096 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6097 | unsigned long size, end; |
d41dee36 AW |
6098 | struct page *map; |
6099 | ||
e984bb43 BP |
6100 | /* |
6101 | * The zone's endpoints aren't required to be MAX_ORDER | |
6102 | * aligned but the node_mem_map endpoints must be in order | |
6103 | * for the buddy allocator to function correctly. | |
6104 | */ | |
108bcc96 | 6105 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6106 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6107 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
6108 | map = alloc_remap(pgdat->node_id, size); |
6109 | if (!map) | |
6782832e SS |
6110 | map = memblock_virt_alloc_node_nopanic(size, |
6111 | pgdat->node_id); | |
a1c34a3b | 6112 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6113 | } |
12d810c1 | 6114 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6115 | /* |
6116 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6117 | */ | |
c713216d | 6118 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6119 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 6120 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 6121 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6122 | mem_map -= offset; |
0ee332c1 | 6123 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6124 | } |
1da177e4 | 6125 | #endif |
d41dee36 | 6126 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
6127 | } |
6128 | ||
9109fb7b JW |
6129 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
6130 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 6131 | { |
9109fb7b | 6132 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6133 | unsigned long start_pfn = 0; |
6134 | unsigned long end_pfn = 0; | |
9109fb7b | 6135 | |
88fdf75d | 6136 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 6137 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 6138 | |
3a80a7fa | 6139 | reset_deferred_meminit(pgdat); |
1da177e4 LT |
6140 | pgdat->node_id = nid; |
6141 | pgdat->node_start_pfn = node_start_pfn; | |
75ef7184 | 6142 | pgdat->per_cpu_nodestats = NULL; |
7960aedd ZY |
6143 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6144 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 6145 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6146 | (u64)start_pfn << PAGE_SHIFT, |
6147 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
6148 | #else |
6149 | start_pfn = node_start_pfn; | |
7960aedd ZY |
6150 | #endif |
6151 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
6152 | zones_size, zholes_size); | |
1da177e4 LT |
6153 | |
6154 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
6155 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
6156 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
6157 | nid, (unsigned long)pgdat, | |
6158 | (unsigned long)pgdat->node_mem_map); | |
6159 | #endif | |
1da177e4 | 6160 | |
7f3eb55b | 6161 | free_area_init_core(pgdat); |
1da177e4 LT |
6162 | } |
6163 | ||
0ee332c1 | 6164 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
6165 | |
6166 | #if MAX_NUMNODES > 1 | |
6167 | /* | |
6168 | * Figure out the number of possible node ids. | |
6169 | */ | |
f9872caf | 6170 | void __init setup_nr_node_ids(void) |
418508c1 | 6171 | { |
904a9553 | 6172 | unsigned int highest; |
418508c1 | 6173 | |
904a9553 | 6174 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
6175 | nr_node_ids = highest + 1; |
6176 | } | |
418508c1 MS |
6177 | #endif |
6178 | ||
1e01979c TH |
6179 | /** |
6180 | * node_map_pfn_alignment - determine the maximum internode alignment | |
6181 | * | |
6182 | * This function should be called after node map is populated and sorted. | |
6183 | * It calculates the maximum power of two alignment which can distinguish | |
6184 | * all the nodes. | |
6185 | * | |
6186 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
6187 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
6188 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
6189 | * shifted, 1GiB is enough and this function will indicate so. | |
6190 | * | |
6191 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
6192 | * model has fine enough granularity to avoid incorrect mapping for the | |
6193 | * populated node map. | |
6194 | * | |
6195 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
6196 | * requirement (single node). | |
6197 | */ | |
6198 | unsigned long __init node_map_pfn_alignment(void) | |
6199 | { | |
6200 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 6201 | unsigned long start, end, mask; |
1e01979c | 6202 | int last_nid = -1; |
c13291a5 | 6203 | int i, nid; |
1e01979c | 6204 | |
c13291a5 | 6205 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
6206 | if (!start || last_nid < 0 || last_nid == nid) { |
6207 | last_nid = nid; | |
6208 | last_end = end; | |
6209 | continue; | |
6210 | } | |
6211 | ||
6212 | /* | |
6213 | * Start with a mask granular enough to pin-point to the | |
6214 | * start pfn and tick off bits one-by-one until it becomes | |
6215 | * too coarse to separate the current node from the last. | |
6216 | */ | |
6217 | mask = ~((1 << __ffs(start)) - 1); | |
6218 | while (mask && last_end <= (start & (mask << 1))) | |
6219 | mask <<= 1; | |
6220 | ||
6221 | /* accumulate all internode masks */ | |
6222 | accl_mask |= mask; | |
6223 | } | |
6224 | ||
6225 | /* convert mask to number of pages */ | |
6226 | return ~accl_mask + 1; | |
6227 | } | |
6228 | ||
a6af2bc3 | 6229 | /* Find the lowest pfn for a node */ |
b69a7288 | 6230 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 6231 | { |
a6af2bc3 | 6232 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
6233 | unsigned long start_pfn; |
6234 | int i; | |
1abbfb41 | 6235 | |
c13291a5 TH |
6236 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
6237 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 6238 | |
a6af2bc3 | 6239 | if (min_pfn == ULONG_MAX) { |
1170532b | 6240 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
6241 | return 0; |
6242 | } | |
6243 | ||
6244 | return min_pfn; | |
c713216d MG |
6245 | } |
6246 | ||
6247 | /** | |
6248 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
6249 | * | |
6250 | * It returns the minimum PFN based on information provided via | |
7d018176 | 6251 | * memblock_set_node(). |
c713216d MG |
6252 | */ |
6253 | unsigned long __init find_min_pfn_with_active_regions(void) | |
6254 | { | |
6255 | return find_min_pfn_for_node(MAX_NUMNODES); | |
6256 | } | |
6257 | ||
37b07e41 LS |
6258 | /* |
6259 | * early_calculate_totalpages() | |
6260 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 6261 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 6262 | */ |
484f51f8 | 6263 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 6264 | { |
7e63efef | 6265 | unsigned long totalpages = 0; |
c13291a5 TH |
6266 | unsigned long start_pfn, end_pfn; |
6267 | int i, nid; | |
6268 | ||
6269 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
6270 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 6271 | |
37b07e41 LS |
6272 | totalpages += pages; |
6273 | if (pages) | |
4b0ef1fe | 6274 | node_set_state(nid, N_MEMORY); |
37b07e41 | 6275 | } |
b8af2941 | 6276 | return totalpages; |
7e63efef MG |
6277 | } |
6278 | ||
2a1e274a MG |
6279 | /* |
6280 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
6281 | * is spread evenly between nodes as long as the nodes have enough | |
6282 | * memory. When they don't, some nodes will have more kernelcore than | |
6283 | * others | |
6284 | */ | |
b224ef85 | 6285 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
6286 | { |
6287 | int i, nid; | |
6288 | unsigned long usable_startpfn; | |
6289 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 6290 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 6291 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 6292 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 6293 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 6294 | struct memblock_region *r; |
b2f3eebe TC |
6295 | |
6296 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
6297 | find_usable_zone_for_movable(); | |
6298 | ||
6299 | /* | |
6300 | * If movable_node is specified, ignore kernelcore and movablecore | |
6301 | * options. | |
6302 | */ | |
6303 | if (movable_node_is_enabled()) { | |
136199f0 EM |
6304 | for_each_memblock(memory, r) { |
6305 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
6306 | continue; |
6307 | ||
136199f0 | 6308 | nid = r->nid; |
b2f3eebe | 6309 | |
136199f0 | 6310 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
6311 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6312 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6313 | usable_startpfn; | |
6314 | } | |
6315 | ||
6316 | goto out2; | |
6317 | } | |
2a1e274a | 6318 | |
342332e6 TI |
6319 | /* |
6320 | * If kernelcore=mirror is specified, ignore movablecore option | |
6321 | */ | |
6322 | if (mirrored_kernelcore) { | |
6323 | bool mem_below_4gb_not_mirrored = false; | |
6324 | ||
6325 | for_each_memblock(memory, r) { | |
6326 | if (memblock_is_mirror(r)) | |
6327 | continue; | |
6328 | ||
6329 | nid = r->nid; | |
6330 | ||
6331 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
6332 | ||
6333 | if (usable_startpfn < 0x100000) { | |
6334 | mem_below_4gb_not_mirrored = true; | |
6335 | continue; | |
6336 | } | |
6337 | ||
6338 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
6339 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6340 | usable_startpfn; | |
6341 | } | |
6342 | ||
6343 | if (mem_below_4gb_not_mirrored) | |
6344 | pr_warn("This configuration results in unmirrored kernel memory."); | |
6345 | ||
6346 | goto out2; | |
6347 | } | |
6348 | ||
7e63efef | 6349 | /* |
b2f3eebe | 6350 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
6351 | * kernelcore that corresponds so that memory usable for |
6352 | * any allocation type is evenly spread. If both kernelcore | |
6353 | * and movablecore are specified, then the value of kernelcore | |
6354 | * will be used for required_kernelcore if it's greater than | |
6355 | * what movablecore would have allowed. | |
6356 | */ | |
6357 | if (required_movablecore) { | |
7e63efef MG |
6358 | unsigned long corepages; |
6359 | ||
6360 | /* | |
6361 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
6362 | * was requested by the user | |
6363 | */ | |
6364 | required_movablecore = | |
6365 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 6366 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
6367 | corepages = totalpages - required_movablecore; |
6368 | ||
6369 | required_kernelcore = max(required_kernelcore, corepages); | |
6370 | } | |
6371 | ||
bde304bd XQ |
6372 | /* |
6373 | * If kernelcore was not specified or kernelcore size is larger | |
6374 | * than totalpages, there is no ZONE_MOVABLE. | |
6375 | */ | |
6376 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 6377 | goto out; |
2a1e274a MG |
6378 | |
6379 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
6380 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6381 | ||
6382 | restart: | |
6383 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
6384 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 6385 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
6386 | unsigned long start_pfn, end_pfn; |
6387 | ||
2a1e274a MG |
6388 | /* |
6389 | * Recalculate kernelcore_node if the division per node | |
6390 | * now exceeds what is necessary to satisfy the requested | |
6391 | * amount of memory for the kernel | |
6392 | */ | |
6393 | if (required_kernelcore < kernelcore_node) | |
6394 | kernelcore_node = required_kernelcore / usable_nodes; | |
6395 | ||
6396 | /* | |
6397 | * As the map is walked, we track how much memory is usable | |
6398 | * by the kernel using kernelcore_remaining. When it is | |
6399 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
6400 | */ | |
6401 | kernelcore_remaining = kernelcore_node; | |
6402 | ||
6403 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 6404 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
6405 | unsigned long size_pages; |
6406 | ||
c13291a5 | 6407 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
6408 | if (start_pfn >= end_pfn) |
6409 | continue; | |
6410 | ||
6411 | /* Account for what is only usable for kernelcore */ | |
6412 | if (start_pfn < usable_startpfn) { | |
6413 | unsigned long kernel_pages; | |
6414 | kernel_pages = min(end_pfn, usable_startpfn) | |
6415 | - start_pfn; | |
6416 | ||
6417 | kernelcore_remaining -= min(kernel_pages, | |
6418 | kernelcore_remaining); | |
6419 | required_kernelcore -= min(kernel_pages, | |
6420 | required_kernelcore); | |
6421 | ||
6422 | /* Continue if range is now fully accounted */ | |
6423 | if (end_pfn <= usable_startpfn) { | |
6424 | ||
6425 | /* | |
6426 | * Push zone_movable_pfn to the end so | |
6427 | * that if we have to rebalance | |
6428 | * kernelcore across nodes, we will | |
6429 | * not double account here | |
6430 | */ | |
6431 | zone_movable_pfn[nid] = end_pfn; | |
6432 | continue; | |
6433 | } | |
6434 | start_pfn = usable_startpfn; | |
6435 | } | |
6436 | ||
6437 | /* | |
6438 | * The usable PFN range for ZONE_MOVABLE is from | |
6439 | * start_pfn->end_pfn. Calculate size_pages as the | |
6440 | * number of pages used as kernelcore | |
6441 | */ | |
6442 | size_pages = end_pfn - start_pfn; | |
6443 | if (size_pages > kernelcore_remaining) | |
6444 | size_pages = kernelcore_remaining; | |
6445 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6446 | ||
6447 | /* | |
6448 | * Some kernelcore has been met, update counts and | |
6449 | * break if the kernelcore for this node has been | |
b8af2941 | 6450 | * satisfied |
2a1e274a MG |
6451 | */ |
6452 | required_kernelcore -= min(required_kernelcore, | |
6453 | size_pages); | |
6454 | kernelcore_remaining -= size_pages; | |
6455 | if (!kernelcore_remaining) | |
6456 | break; | |
6457 | } | |
6458 | } | |
6459 | ||
6460 | /* | |
6461 | * If there is still required_kernelcore, we do another pass with one | |
6462 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6463 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6464 | * satisfied |
2a1e274a MG |
6465 | */ |
6466 | usable_nodes--; | |
6467 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6468 | goto restart; | |
6469 | ||
b2f3eebe | 6470 | out2: |
2a1e274a MG |
6471 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6472 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6473 | zone_movable_pfn[nid] = | |
6474 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6475 | |
20e6926d | 6476 | out: |
66918dcd | 6477 | /* restore the node_state */ |
4b0ef1fe | 6478 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6479 | } |
6480 | ||
4b0ef1fe LJ |
6481 | /* Any regular or high memory on that node ? */ |
6482 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6483 | { |
37b07e41 LS |
6484 | enum zone_type zone_type; |
6485 | ||
4b0ef1fe LJ |
6486 | if (N_MEMORY == N_NORMAL_MEMORY) |
6487 | return; | |
6488 | ||
6489 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 6490 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6491 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
6492 | node_set_state(nid, N_HIGH_MEMORY); |
6493 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
6494 | zone_type <= ZONE_NORMAL) | |
6495 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
6496 | break; |
6497 | } | |
37b07e41 | 6498 | } |
37b07e41 LS |
6499 | } |
6500 | ||
c713216d MG |
6501 | /** |
6502 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6503 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6504 | * |
6505 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6506 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6507 | * zone in each node and their holes is calculated. If the maximum PFN |
6508 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6509 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6510 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6511 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6512 | * at arch_max_dma_pfn. | |
6513 | */ | |
6514 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6515 | { | |
c13291a5 TH |
6516 | unsigned long start_pfn, end_pfn; |
6517 | int i, nid; | |
a6af2bc3 | 6518 | |
c713216d MG |
6519 | /* Record where the zone boundaries are */ |
6520 | memset(arch_zone_lowest_possible_pfn, 0, | |
6521 | sizeof(arch_zone_lowest_possible_pfn)); | |
6522 | memset(arch_zone_highest_possible_pfn, 0, | |
6523 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
6524 | |
6525 | start_pfn = find_min_pfn_with_active_regions(); | |
6526 | ||
6527 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6528 | if (i == ZONE_MOVABLE) |
6529 | continue; | |
90cae1fe OH |
6530 | |
6531 | end_pfn = max(max_zone_pfn[i], start_pfn); | |
6532 | arch_zone_lowest_possible_pfn[i] = start_pfn; | |
6533 | arch_zone_highest_possible_pfn[i] = end_pfn; | |
6534 | ||
6535 | start_pfn = end_pfn; | |
c713216d | 6536 | } |
2a1e274a MG |
6537 | |
6538 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6539 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6540 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6541 | |
c713216d | 6542 | /* Print out the zone ranges */ |
f88dfff5 | 6543 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6544 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6545 | if (i == ZONE_MOVABLE) | |
6546 | continue; | |
f88dfff5 | 6547 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6548 | if (arch_zone_lowest_possible_pfn[i] == |
6549 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6550 | pr_cont("empty\n"); |
72f0ba02 | 6551 | else |
8d29e18a JG |
6552 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6553 | (u64)arch_zone_lowest_possible_pfn[i] | |
6554 | << PAGE_SHIFT, | |
6555 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6556 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6557 | } |
6558 | ||
6559 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6560 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6561 | for (i = 0; i < MAX_NUMNODES; i++) { |
6562 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6563 | pr_info(" Node %d: %#018Lx\n", i, |
6564 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6565 | } |
c713216d | 6566 | |
f2d52fe5 | 6567 | /* Print out the early node map */ |
f88dfff5 | 6568 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6569 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6570 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6571 | (u64)start_pfn << PAGE_SHIFT, | |
6572 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6573 | |
6574 | /* Initialise every node */ | |
708614e6 | 6575 | mminit_verify_pageflags_layout(); |
8ef82866 | 6576 | setup_nr_node_ids(); |
c713216d MG |
6577 | for_each_online_node(nid) { |
6578 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6579 | free_area_init_node(nid, NULL, |
c713216d | 6580 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6581 | |
6582 | /* Any memory on that node */ | |
6583 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6584 | node_set_state(nid, N_MEMORY); |
6585 | check_for_memory(pgdat, nid); | |
c713216d MG |
6586 | } |
6587 | } | |
2a1e274a | 6588 | |
7e63efef | 6589 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
6590 | { |
6591 | unsigned long long coremem; | |
6592 | if (!p) | |
6593 | return -EINVAL; | |
6594 | ||
6595 | coremem = memparse(p, &p); | |
7e63efef | 6596 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 6597 | |
7e63efef | 6598 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
6599 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6600 | ||
6601 | return 0; | |
6602 | } | |
ed7ed365 | 6603 | |
7e63efef MG |
6604 | /* |
6605 | * kernelcore=size sets the amount of memory for use for allocations that | |
6606 | * cannot be reclaimed or migrated. | |
6607 | */ | |
6608 | static int __init cmdline_parse_kernelcore(char *p) | |
6609 | { | |
342332e6 TI |
6610 | /* parse kernelcore=mirror */ |
6611 | if (parse_option_str(p, "mirror")) { | |
6612 | mirrored_kernelcore = true; | |
6613 | return 0; | |
6614 | } | |
6615 | ||
7e63efef MG |
6616 | return cmdline_parse_core(p, &required_kernelcore); |
6617 | } | |
6618 | ||
6619 | /* | |
6620 | * movablecore=size sets the amount of memory for use for allocations that | |
6621 | * can be reclaimed or migrated. | |
6622 | */ | |
6623 | static int __init cmdline_parse_movablecore(char *p) | |
6624 | { | |
6625 | return cmdline_parse_core(p, &required_movablecore); | |
6626 | } | |
6627 | ||
ed7ed365 | 6628 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6629 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6630 | |
0ee332c1 | 6631 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6632 | |
c3d5f5f0 JL |
6633 | void adjust_managed_page_count(struct page *page, long count) |
6634 | { | |
6635 | spin_lock(&managed_page_count_lock); | |
6636 | page_zone(page)->managed_pages += count; | |
6637 | totalram_pages += count; | |
3dcc0571 JL |
6638 | #ifdef CONFIG_HIGHMEM |
6639 | if (PageHighMem(page)) | |
6640 | totalhigh_pages += count; | |
6641 | #endif | |
c3d5f5f0 JL |
6642 | spin_unlock(&managed_page_count_lock); |
6643 | } | |
3dcc0571 | 6644 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6645 | |
11199692 | 6646 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6647 | { |
11199692 JL |
6648 | void *pos; |
6649 | unsigned long pages = 0; | |
69afade7 | 6650 | |
11199692 JL |
6651 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6652 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6653 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 6654 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
6655 | memset(pos, poison, PAGE_SIZE); |
6656 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
6657 | } |
6658 | ||
6659 | if (pages && s) | |
adb1fe9a JP |
6660 | pr_info("Freeing %s memory: %ldK\n", |
6661 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
6662 | |
6663 | return pages; | |
6664 | } | |
11199692 | 6665 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6666 | |
cfa11e08 JL |
6667 | #ifdef CONFIG_HIGHMEM |
6668 | void free_highmem_page(struct page *page) | |
6669 | { | |
6670 | __free_reserved_page(page); | |
6671 | totalram_pages++; | |
7b4b2a0d | 6672 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6673 | totalhigh_pages++; |
6674 | } | |
6675 | #endif | |
6676 | ||
7ee3d4e8 JL |
6677 | |
6678 | void __init mem_init_print_info(const char *str) | |
6679 | { | |
6680 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
6681 | unsigned long init_code_size, init_data_size; | |
6682 | ||
6683 | physpages = get_num_physpages(); | |
6684 | codesize = _etext - _stext; | |
6685 | datasize = _edata - _sdata; | |
6686 | rosize = __end_rodata - __start_rodata; | |
6687 | bss_size = __bss_stop - __bss_start; | |
6688 | init_data_size = __init_end - __init_begin; | |
6689 | init_code_size = _einittext - _sinittext; | |
6690 | ||
6691 | /* | |
6692 | * Detect special cases and adjust section sizes accordingly: | |
6693 | * 1) .init.* may be embedded into .data sections | |
6694 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
6695 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
6696 | * 3) .rodata.* may be embedded into .text or .data sections. | |
6697 | */ | |
6698 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
6699 | do { \ |
6700 | if (start <= pos && pos < end && size > adj) \ | |
6701 | size -= adj; \ | |
6702 | } while (0) | |
7ee3d4e8 JL |
6703 | |
6704 | adj_init_size(__init_begin, __init_end, init_data_size, | |
6705 | _sinittext, init_code_size); | |
6706 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
6707 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
6708 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
6709 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
6710 | ||
6711 | #undef adj_init_size | |
6712 | ||
756a025f | 6713 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 6714 | #ifdef CONFIG_HIGHMEM |
756a025f | 6715 | ", %luK highmem" |
7ee3d4e8 | 6716 | #endif |
756a025f JP |
6717 | "%s%s)\n", |
6718 | nr_free_pages() << (PAGE_SHIFT - 10), | |
6719 | physpages << (PAGE_SHIFT - 10), | |
6720 | codesize >> 10, datasize >> 10, rosize >> 10, | |
6721 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
6722 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
6723 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 6724 | #ifdef CONFIG_HIGHMEM |
756a025f | 6725 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 6726 | #endif |
756a025f | 6727 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
6728 | } |
6729 | ||
0e0b864e | 6730 | /** |
88ca3b94 RD |
6731 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6732 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 6733 | * |
013110a7 | 6734 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
6735 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6736 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
6737 | * function may optionally be used to account for unfreeable pages in the |
6738 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
6739 | * smaller per-cpu batchsize. | |
0e0b864e MG |
6740 | */ |
6741 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
6742 | { | |
6743 | dma_reserve = new_dma_reserve; | |
6744 | } | |
6745 | ||
1da177e4 LT |
6746 | void __init free_area_init(unsigned long *zones_size) |
6747 | { | |
9109fb7b | 6748 | free_area_init_node(0, zones_size, |
1da177e4 LT |
6749 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6750 | } | |
1da177e4 | 6751 | |
005fd4bb | 6752 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 6753 | { |
1da177e4 | 6754 | |
005fd4bb SAS |
6755 | lru_add_drain_cpu(cpu); |
6756 | drain_pages(cpu); | |
9f8f2172 | 6757 | |
005fd4bb SAS |
6758 | /* |
6759 | * Spill the event counters of the dead processor | |
6760 | * into the current processors event counters. | |
6761 | * This artificially elevates the count of the current | |
6762 | * processor. | |
6763 | */ | |
6764 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 6765 | |
005fd4bb SAS |
6766 | /* |
6767 | * Zero the differential counters of the dead processor | |
6768 | * so that the vm statistics are consistent. | |
6769 | * | |
6770 | * This is only okay since the processor is dead and cannot | |
6771 | * race with what we are doing. | |
6772 | */ | |
6773 | cpu_vm_stats_fold(cpu); | |
6774 | return 0; | |
1da177e4 | 6775 | } |
1da177e4 LT |
6776 | |
6777 | void __init page_alloc_init(void) | |
6778 | { | |
005fd4bb SAS |
6779 | int ret; |
6780 | ||
6781 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, | |
6782 | "mm/page_alloc:dead", NULL, | |
6783 | page_alloc_cpu_dead); | |
6784 | WARN_ON(ret < 0); | |
1da177e4 LT |
6785 | } |
6786 | ||
cb45b0e9 | 6787 | /* |
34b10060 | 6788 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
6789 | * or min_free_kbytes changes. |
6790 | */ | |
6791 | static void calculate_totalreserve_pages(void) | |
6792 | { | |
6793 | struct pglist_data *pgdat; | |
6794 | unsigned long reserve_pages = 0; | |
2f6726e5 | 6795 | enum zone_type i, j; |
cb45b0e9 HA |
6796 | |
6797 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
6798 | |
6799 | pgdat->totalreserve_pages = 0; | |
6800 | ||
cb45b0e9 HA |
6801 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6802 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 6803 | long max = 0; |
cb45b0e9 HA |
6804 | |
6805 | /* Find valid and maximum lowmem_reserve in the zone */ | |
6806 | for (j = i; j < MAX_NR_ZONES; j++) { | |
6807 | if (zone->lowmem_reserve[j] > max) | |
6808 | max = zone->lowmem_reserve[j]; | |
6809 | } | |
6810 | ||
41858966 MG |
6811 | /* we treat the high watermark as reserved pages. */ |
6812 | max += high_wmark_pages(zone); | |
cb45b0e9 | 6813 | |
b40da049 JL |
6814 | if (max > zone->managed_pages) |
6815 | max = zone->managed_pages; | |
a8d01437 | 6816 | |
281e3726 | 6817 | pgdat->totalreserve_pages += max; |
a8d01437 | 6818 | |
cb45b0e9 HA |
6819 | reserve_pages += max; |
6820 | } | |
6821 | } | |
6822 | totalreserve_pages = reserve_pages; | |
6823 | } | |
6824 | ||
1da177e4 LT |
6825 | /* |
6826 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 6827 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
6828 | * has a correct pages reserved value, so an adequate number of |
6829 | * pages are left in the zone after a successful __alloc_pages(). | |
6830 | */ | |
6831 | static void setup_per_zone_lowmem_reserve(void) | |
6832 | { | |
6833 | struct pglist_data *pgdat; | |
2f6726e5 | 6834 | enum zone_type j, idx; |
1da177e4 | 6835 | |
ec936fc5 | 6836 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
6837 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6838 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 6839 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
6840 | |
6841 | zone->lowmem_reserve[j] = 0; | |
6842 | ||
2f6726e5 CL |
6843 | idx = j; |
6844 | while (idx) { | |
1da177e4 LT |
6845 | struct zone *lower_zone; |
6846 | ||
2f6726e5 CL |
6847 | idx--; |
6848 | ||
1da177e4 LT |
6849 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6850 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
6851 | ||
6852 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 6853 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 6854 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 6855 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
6856 | } |
6857 | } | |
6858 | } | |
cb45b0e9 HA |
6859 | |
6860 | /* update totalreserve_pages */ | |
6861 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6862 | } |
6863 | ||
cfd3da1e | 6864 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
6865 | { |
6866 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
6867 | unsigned long lowmem_pages = 0; | |
6868 | struct zone *zone; | |
6869 | unsigned long flags; | |
6870 | ||
6871 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
6872 | for_each_zone(zone) { | |
6873 | if (!is_highmem(zone)) | |
b40da049 | 6874 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
6875 | } |
6876 | ||
6877 | for_each_zone(zone) { | |
ac924c60 AM |
6878 | u64 tmp; |
6879 | ||
1125b4e3 | 6880 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 6881 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 6882 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
6883 | if (is_highmem(zone)) { |
6884 | /* | |
669ed175 NP |
6885 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6886 | * need highmem pages, so cap pages_min to a small | |
6887 | * value here. | |
6888 | * | |
41858966 | 6889 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 6890 | * deltas control asynch page reclaim, and so should |
669ed175 | 6891 | * not be capped for highmem. |
1da177e4 | 6892 | */ |
90ae8d67 | 6893 | unsigned long min_pages; |
1da177e4 | 6894 | |
b40da049 | 6895 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 6896 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 6897 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 6898 | } else { |
669ed175 NP |
6899 | /* |
6900 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
6901 | * proportionate to the zone's size. |
6902 | */ | |
41858966 | 6903 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
6904 | } |
6905 | ||
795ae7a0 JW |
6906 | /* |
6907 | * Set the kswapd watermarks distance according to the | |
6908 | * scale factor in proportion to available memory, but | |
6909 | * ensure a minimum size on small systems. | |
6910 | */ | |
6911 | tmp = max_t(u64, tmp >> 2, | |
6912 | mult_frac(zone->managed_pages, | |
6913 | watermark_scale_factor, 10000)); | |
6914 | ||
6915 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
6916 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 6917 | |
1125b4e3 | 6918 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 6919 | } |
cb45b0e9 HA |
6920 | |
6921 | /* update totalreserve_pages */ | |
6922 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6923 | } |
6924 | ||
cfd3da1e MG |
6925 | /** |
6926 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
6927 | * or when memory is hot-{added|removed} | |
6928 | * | |
6929 | * Ensures that the watermark[min,low,high] values for each zone are set | |
6930 | * correctly with respect to min_free_kbytes. | |
6931 | */ | |
6932 | void setup_per_zone_wmarks(void) | |
6933 | { | |
6934 | mutex_lock(&zonelists_mutex); | |
6935 | __setup_per_zone_wmarks(); | |
6936 | mutex_unlock(&zonelists_mutex); | |
6937 | } | |
6938 | ||
1da177e4 LT |
6939 | /* |
6940 | * Initialise min_free_kbytes. | |
6941 | * | |
6942 | * For small machines we want it small (128k min). For large machines | |
6943 | * we want it large (64MB max). But it is not linear, because network | |
6944 | * bandwidth does not increase linearly with machine size. We use | |
6945 | * | |
b8af2941 | 6946 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6947 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6948 | * | |
6949 | * which yields | |
6950 | * | |
6951 | * 16MB: 512k | |
6952 | * 32MB: 724k | |
6953 | * 64MB: 1024k | |
6954 | * 128MB: 1448k | |
6955 | * 256MB: 2048k | |
6956 | * 512MB: 2896k | |
6957 | * 1024MB: 4096k | |
6958 | * 2048MB: 5792k | |
6959 | * 4096MB: 8192k | |
6960 | * 8192MB: 11584k | |
6961 | * 16384MB: 16384k | |
6962 | */ | |
1b79acc9 | 6963 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6964 | { |
6965 | unsigned long lowmem_kbytes; | |
5f12733e | 6966 | int new_min_free_kbytes; |
1da177e4 LT |
6967 | |
6968 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6969 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6970 | ||
6971 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6972 | min_free_kbytes = new_min_free_kbytes; | |
6973 | if (min_free_kbytes < 128) | |
6974 | min_free_kbytes = 128; | |
6975 | if (min_free_kbytes > 65536) | |
6976 | min_free_kbytes = 65536; | |
6977 | } else { | |
6978 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6979 | new_min_free_kbytes, user_min_free_kbytes); | |
6980 | } | |
bc75d33f | 6981 | setup_per_zone_wmarks(); |
a6cccdc3 | 6982 | refresh_zone_stat_thresholds(); |
1da177e4 | 6983 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
6984 | |
6985 | #ifdef CONFIG_NUMA | |
6986 | setup_min_unmapped_ratio(); | |
6987 | setup_min_slab_ratio(); | |
6988 | #endif | |
6989 | ||
1da177e4 LT |
6990 | return 0; |
6991 | } | |
bc22af74 | 6992 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
6993 | |
6994 | /* | |
b8af2941 | 6995 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6996 | * that we can call two helper functions whenever min_free_kbytes |
6997 | * changes. | |
6998 | */ | |
cccad5b9 | 6999 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7000 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7001 | { |
da8c757b HP |
7002 | int rc; |
7003 | ||
7004 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7005 | if (rc) | |
7006 | return rc; | |
7007 | ||
5f12733e MH |
7008 | if (write) { |
7009 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7010 | setup_per_zone_wmarks(); |
5f12733e | 7011 | } |
1da177e4 LT |
7012 | return 0; |
7013 | } | |
7014 | ||
795ae7a0 JW |
7015 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
7016 | void __user *buffer, size_t *length, loff_t *ppos) | |
7017 | { | |
7018 | int rc; | |
7019 | ||
7020 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7021 | if (rc) | |
7022 | return rc; | |
7023 | ||
7024 | if (write) | |
7025 | setup_per_zone_wmarks(); | |
7026 | ||
7027 | return 0; | |
7028 | } | |
7029 | ||
9614634f | 7030 | #ifdef CONFIG_NUMA |
6423aa81 | 7031 | static void setup_min_unmapped_ratio(void) |
9614634f | 7032 | { |
6423aa81 | 7033 | pg_data_t *pgdat; |
9614634f | 7034 | struct zone *zone; |
9614634f | 7035 | |
a5f5f91d | 7036 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7037 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7038 | |
9614634f | 7039 | for_each_zone(zone) |
a5f5f91d | 7040 | zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages * |
9614634f | 7041 | sysctl_min_unmapped_ratio) / 100; |
9614634f | 7042 | } |
0ff38490 | 7043 | |
6423aa81 JK |
7044 | |
7045 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 7046 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7047 | { |
0ff38490 CL |
7048 | int rc; |
7049 | ||
8d65af78 | 7050 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7051 | if (rc) |
7052 | return rc; | |
7053 | ||
6423aa81 JK |
7054 | setup_min_unmapped_ratio(); |
7055 | ||
7056 | return 0; | |
7057 | } | |
7058 | ||
7059 | static void setup_min_slab_ratio(void) | |
7060 | { | |
7061 | pg_data_t *pgdat; | |
7062 | struct zone *zone; | |
7063 | ||
a5f5f91d MG |
7064 | for_each_online_pgdat(pgdat) |
7065 | pgdat->min_slab_pages = 0; | |
7066 | ||
0ff38490 | 7067 | for_each_zone(zone) |
a5f5f91d | 7068 | zone->zone_pgdat->min_slab_pages += (zone->managed_pages * |
0ff38490 | 7069 | sysctl_min_slab_ratio) / 100; |
6423aa81 JK |
7070 | } |
7071 | ||
7072 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
7073 | void __user *buffer, size_t *length, loff_t *ppos) | |
7074 | { | |
7075 | int rc; | |
7076 | ||
7077 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7078 | if (rc) | |
7079 | return rc; | |
7080 | ||
7081 | setup_min_slab_ratio(); | |
7082 | ||
0ff38490 CL |
7083 | return 0; |
7084 | } | |
9614634f CL |
7085 | #endif |
7086 | ||
1da177e4 LT |
7087 | /* |
7088 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7089 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7090 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7091 | * | |
7092 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7093 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7094 | * if in function of the boot time zone sizes. |
7095 | */ | |
cccad5b9 | 7096 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7097 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7098 | { |
8d65af78 | 7099 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
7100 | setup_per_zone_lowmem_reserve(); |
7101 | return 0; | |
7102 | } | |
7103 | ||
8ad4b1fb RS |
7104 | /* |
7105 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
7106 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
7107 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 7108 | */ |
cccad5b9 | 7109 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7110 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
7111 | { |
7112 | struct zone *zone; | |
7cd2b0a3 | 7113 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
7114 | int ret; |
7115 | ||
7cd2b0a3 DR |
7116 | mutex_lock(&pcp_batch_high_lock); |
7117 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
7118 | ||
8d65af78 | 7119 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
7120 | if (!write || ret < 0) |
7121 | goto out; | |
7122 | ||
7123 | /* Sanity checking to avoid pcp imbalance */ | |
7124 | if (percpu_pagelist_fraction && | |
7125 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
7126 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
7127 | ret = -EINVAL; | |
7128 | goto out; | |
7129 | } | |
7130 | ||
7131 | /* No change? */ | |
7132 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
7133 | goto out; | |
c8e251fa | 7134 | |
364df0eb | 7135 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
7136 | unsigned int cpu; |
7137 | ||
22a7f12b | 7138 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
7139 | pageset_set_high_and_batch(zone, |
7140 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 7141 | } |
7cd2b0a3 | 7142 | out: |
c8e251fa | 7143 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 7144 | return ret; |
8ad4b1fb RS |
7145 | } |
7146 | ||
a9919c79 | 7147 | #ifdef CONFIG_NUMA |
f034b5d4 | 7148 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 7149 | |
1da177e4 LT |
7150 | static int __init set_hashdist(char *str) |
7151 | { | |
7152 | if (!str) | |
7153 | return 0; | |
7154 | hashdist = simple_strtoul(str, &str, 0); | |
7155 | return 1; | |
7156 | } | |
7157 | __setup("hashdist=", set_hashdist); | |
7158 | #endif | |
7159 | ||
f6f34b43 SD |
7160 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
7161 | /* | |
7162 | * Returns the number of pages that arch has reserved but | |
7163 | * is not known to alloc_large_system_hash(). | |
7164 | */ | |
7165 | static unsigned long __init arch_reserved_kernel_pages(void) | |
7166 | { | |
7167 | return 0; | |
7168 | } | |
7169 | #endif | |
7170 | ||
1da177e4 LT |
7171 | /* |
7172 | * allocate a large system hash table from bootmem | |
7173 | * - it is assumed that the hash table must contain an exact power-of-2 | |
7174 | * quantity of entries | |
7175 | * - limit is the number of hash buckets, not the total allocation size | |
7176 | */ | |
7177 | void *__init alloc_large_system_hash(const char *tablename, | |
7178 | unsigned long bucketsize, | |
7179 | unsigned long numentries, | |
7180 | int scale, | |
7181 | int flags, | |
7182 | unsigned int *_hash_shift, | |
7183 | unsigned int *_hash_mask, | |
31fe62b9 TB |
7184 | unsigned long low_limit, |
7185 | unsigned long high_limit) | |
1da177e4 | 7186 | { |
31fe62b9 | 7187 | unsigned long long max = high_limit; |
1da177e4 LT |
7188 | unsigned long log2qty, size; |
7189 | void *table = NULL; | |
7190 | ||
7191 | /* allow the kernel cmdline to have a say */ | |
7192 | if (!numentries) { | |
7193 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 7194 | numentries = nr_kernel_pages; |
f6f34b43 | 7195 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
7196 | |
7197 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
7198 | if (PAGE_SHIFT < 20) | |
7199 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
7200 | |
7201 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
7202 | if (scale > PAGE_SHIFT) | |
7203 | numentries >>= (scale - PAGE_SHIFT); | |
7204 | else | |
7205 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
7206 | |
7207 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
7208 | if (unlikely(flags & HASH_SMALL)) { |
7209 | /* Makes no sense without HASH_EARLY */ | |
7210 | WARN_ON(!(flags & HASH_EARLY)); | |
7211 | if (!(numentries >> *_hash_shift)) { | |
7212 | numentries = 1UL << *_hash_shift; | |
7213 | BUG_ON(!numentries); | |
7214 | } | |
7215 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 7216 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 7217 | } |
6e692ed3 | 7218 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
7219 | |
7220 | /* limit allocation size to 1/16 total memory by default */ | |
7221 | if (max == 0) { | |
7222 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
7223 | do_div(max, bucketsize); | |
7224 | } | |
074b8517 | 7225 | max = min(max, 0x80000000ULL); |
1da177e4 | 7226 | |
31fe62b9 TB |
7227 | if (numentries < low_limit) |
7228 | numentries = low_limit; | |
1da177e4 LT |
7229 | if (numentries > max) |
7230 | numentries = max; | |
7231 | ||
f0d1b0b3 | 7232 | log2qty = ilog2(numentries); |
1da177e4 LT |
7233 | |
7234 | do { | |
7235 | size = bucketsize << log2qty; | |
7236 | if (flags & HASH_EARLY) | |
6782832e | 7237 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
7238 | else if (hashdist) |
7239 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
7240 | else { | |
1037b83b ED |
7241 | /* |
7242 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
7243 | * some pages at the end of hash table which |
7244 | * alloc_pages_exact() automatically does | |
1037b83b | 7245 | */ |
264ef8a9 | 7246 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 7247 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
7248 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
7249 | } | |
1da177e4 LT |
7250 | } |
7251 | } while (!table && size > PAGE_SIZE && --log2qty); | |
7252 | ||
7253 | if (!table) | |
7254 | panic("Failed to allocate %s hash table\n", tablename); | |
7255 | ||
1170532b JP |
7256 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
7257 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
7258 | |
7259 | if (_hash_shift) | |
7260 | *_hash_shift = log2qty; | |
7261 | if (_hash_mask) | |
7262 | *_hash_mask = (1 << log2qty) - 1; | |
7263 | ||
7264 | return table; | |
7265 | } | |
a117e66e | 7266 | |
a5d76b54 | 7267 | /* |
80934513 MK |
7268 | * This function checks whether pageblock includes unmovable pages or not. |
7269 | * If @count is not zero, it is okay to include less @count unmovable pages | |
7270 | * | |
b8af2941 | 7271 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
7272 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
7273 | * check without lock_page also may miss some movable non-lru pages at | |
7274 | * race condition. So you can't expect this function should be exact. | |
a5d76b54 | 7275 | */ |
b023f468 WC |
7276 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
7277 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
7278 | { |
7279 | unsigned long pfn, iter, found; | |
47118af0 MN |
7280 | int mt; |
7281 | ||
49ac8255 KH |
7282 | /* |
7283 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 7284 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
7285 | */ |
7286 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 7287 | return false; |
47118af0 MN |
7288 | mt = get_pageblock_migratetype(page); |
7289 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 7290 | return false; |
49ac8255 KH |
7291 | |
7292 | pfn = page_to_pfn(page); | |
7293 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
7294 | unsigned long check = pfn + iter; | |
7295 | ||
29723fcc | 7296 | if (!pfn_valid_within(check)) |
49ac8255 | 7297 | continue; |
29723fcc | 7298 | |
49ac8255 | 7299 | page = pfn_to_page(check); |
c8721bbb NH |
7300 | |
7301 | /* | |
7302 | * Hugepages are not in LRU lists, but they're movable. | |
7303 | * We need not scan over tail pages bacause we don't | |
7304 | * handle each tail page individually in migration. | |
7305 | */ | |
7306 | if (PageHuge(page)) { | |
7307 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
7308 | continue; | |
7309 | } | |
7310 | ||
97d255c8 MK |
7311 | /* |
7312 | * We can't use page_count without pin a page | |
7313 | * because another CPU can free compound page. | |
7314 | * This check already skips compound tails of THP | |
0139aa7b | 7315 | * because their page->_refcount is zero at all time. |
97d255c8 | 7316 | */ |
fe896d18 | 7317 | if (!page_ref_count(page)) { |
49ac8255 KH |
7318 | if (PageBuddy(page)) |
7319 | iter += (1 << page_order(page)) - 1; | |
7320 | continue; | |
7321 | } | |
97d255c8 | 7322 | |
b023f468 WC |
7323 | /* |
7324 | * The HWPoisoned page may be not in buddy system, and | |
7325 | * page_count() is not 0. | |
7326 | */ | |
7327 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
7328 | continue; | |
7329 | ||
0efadf48 YX |
7330 | if (__PageMovable(page)) |
7331 | continue; | |
7332 | ||
49ac8255 KH |
7333 | if (!PageLRU(page)) |
7334 | found++; | |
7335 | /* | |
6b4f7799 JW |
7336 | * If there are RECLAIMABLE pages, we need to check |
7337 | * it. But now, memory offline itself doesn't call | |
7338 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
7339 | */ |
7340 | /* | |
7341 | * If the page is not RAM, page_count()should be 0. | |
7342 | * we don't need more check. This is an _used_ not-movable page. | |
7343 | * | |
7344 | * The problematic thing here is PG_reserved pages. PG_reserved | |
7345 | * is set to both of a memory hole page and a _used_ kernel | |
7346 | * page at boot. | |
7347 | */ | |
7348 | if (found > count) | |
80934513 | 7349 | return true; |
49ac8255 | 7350 | } |
80934513 | 7351 | return false; |
49ac8255 KH |
7352 | } |
7353 | ||
7354 | bool is_pageblock_removable_nolock(struct page *page) | |
7355 | { | |
656a0706 MH |
7356 | struct zone *zone; |
7357 | unsigned long pfn; | |
687875fb MH |
7358 | |
7359 | /* | |
7360 | * We have to be careful here because we are iterating over memory | |
7361 | * sections which are not zone aware so we might end up outside of | |
7362 | * the zone but still within the section. | |
656a0706 MH |
7363 | * We have to take care about the node as well. If the node is offline |
7364 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 7365 | */ |
656a0706 MH |
7366 | if (!node_online(page_to_nid(page))) |
7367 | return false; | |
7368 | ||
7369 | zone = page_zone(page); | |
7370 | pfn = page_to_pfn(page); | |
108bcc96 | 7371 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
7372 | return false; |
7373 | ||
b023f468 | 7374 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 7375 | } |
0c0e6195 | 7376 | |
080fe206 | 7377 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7378 | |
7379 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7380 | { | |
7381 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7382 | pageblock_nr_pages) - 1); | |
7383 | } | |
7384 | ||
7385 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7386 | { | |
7387 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7388 | pageblock_nr_pages)); | |
7389 | } | |
7390 | ||
041d3a8c | 7391 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7392 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7393 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7394 | { |
7395 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7396 | unsigned long nr_reclaimed; |
041d3a8c MN |
7397 | unsigned long pfn = start; |
7398 | unsigned int tries = 0; | |
7399 | int ret = 0; | |
7400 | ||
be49a6e1 | 7401 | migrate_prep(); |
041d3a8c | 7402 | |
bb13ffeb | 7403 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7404 | if (fatal_signal_pending(current)) { |
7405 | ret = -EINTR; | |
7406 | break; | |
7407 | } | |
7408 | ||
bb13ffeb MG |
7409 | if (list_empty(&cc->migratepages)) { |
7410 | cc->nr_migratepages = 0; | |
edc2ca61 | 7411 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7412 | if (!pfn) { |
7413 | ret = -EINTR; | |
7414 | break; | |
7415 | } | |
7416 | tries = 0; | |
7417 | } else if (++tries == 5) { | |
7418 | ret = ret < 0 ? ret : -EBUSY; | |
7419 | break; | |
7420 | } | |
7421 | ||
beb51eaa MK |
7422 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7423 | &cc->migratepages); | |
7424 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7425 | |
9c620e2b | 7426 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 7427 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 7428 | } |
2a6f5124 SP |
7429 | if (ret < 0) { |
7430 | putback_movable_pages(&cc->migratepages); | |
7431 | return ret; | |
7432 | } | |
7433 | return 0; | |
041d3a8c MN |
7434 | } |
7435 | ||
7436 | /** | |
7437 | * alloc_contig_range() -- tries to allocate given range of pages | |
7438 | * @start: start PFN to allocate | |
7439 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7440 | * @migratetype: migratetype of the underlaying pageblocks (either |
7441 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7442 | * in range must have the same migratetype and it must | |
7443 | * be either of the two. | |
ca96b625 | 7444 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
7445 | * |
7446 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
7447 | * aligned, however it's the caller's responsibility to guarantee that | |
7448 | * we are the only thread that changes migrate type of pageblocks the | |
7449 | * pages fall in. | |
7450 | * | |
7451 | * The PFN range must belong to a single zone. | |
7452 | * | |
7453 | * Returns zero on success or negative error code. On success all | |
7454 | * pages which PFN is in [start, end) are allocated for the caller and | |
7455 | * need to be freed with free_contig_range(). | |
7456 | */ | |
0815f3d8 | 7457 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 7458 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 7459 | { |
041d3a8c | 7460 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7461 | unsigned int order; |
7462 | int ret = 0; | |
041d3a8c | 7463 | |
bb13ffeb MG |
7464 | struct compact_control cc = { |
7465 | .nr_migratepages = 0, | |
7466 | .order = -1, | |
7467 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7468 | .mode = MIGRATE_SYNC, |
bb13ffeb | 7469 | .ignore_skip_hint = true, |
7dea19f9 | 7470 | .gfp_mask = current_gfp_context(gfp_mask), |
bb13ffeb MG |
7471 | }; |
7472 | INIT_LIST_HEAD(&cc.migratepages); | |
7473 | ||
041d3a8c MN |
7474 | /* |
7475 | * What we do here is we mark all pageblocks in range as | |
7476 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7477 | * have different sizes, and due to the way page allocator | |
7478 | * work, we align the range to biggest of the two pages so | |
7479 | * that page allocator won't try to merge buddies from | |
7480 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7481 | * other migration type. | |
7482 | * | |
7483 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7484 | * migrate the pages from an unaligned range (ie. pages that | |
7485 | * we are interested in). This will put all the pages in | |
7486 | * range back to page allocator as MIGRATE_ISOLATE. | |
7487 | * | |
7488 | * When this is done, we take the pages in range from page | |
7489 | * allocator removing them from the buddy system. This way | |
7490 | * page allocator will never consider using them. | |
7491 | * | |
7492 | * This lets us mark the pageblocks back as | |
7493 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7494 | * aligned range but not in the unaligned, original range are | |
7495 | * put back to page allocator so that buddy can use them. | |
7496 | */ | |
7497 | ||
7498 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7499 | pfn_max_align_up(end), migratetype, |
7500 | false); | |
041d3a8c | 7501 | if (ret) |
86a595f9 | 7502 | return ret; |
041d3a8c | 7503 | |
8ef5849f JK |
7504 | /* |
7505 | * In case of -EBUSY, we'd like to know which page causes problem. | |
7506 | * So, just fall through. We will check it in test_pages_isolated(). | |
7507 | */ | |
bb13ffeb | 7508 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7509 | if (ret && ret != -EBUSY) |
041d3a8c MN |
7510 | goto done; |
7511 | ||
7512 | /* | |
7513 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7514 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7515 | * more, all pages in [start, end) are free in page allocator. | |
7516 | * What we are going to do is to allocate all pages from | |
7517 | * [start, end) (that is remove them from page allocator). | |
7518 | * | |
7519 | * The only problem is that pages at the beginning and at the | |
7520 | * end of interesting range may be not aligned with pages that | |
7521 | * page allocator holds, ie. they can be part of higher order | |
7522 | * pages. Because of this, we reserve the bigger range and | |
7523 | * once this is done free the pages we are not interested in. | |
7524 | * | |
7525 | * We don't have to hold zone->lock here because the pages are | |
7526 | * isolated thus they won't get removed from buddy. | |
7527 | */ | |
7528 | ||
7529 | lru_add_drain_all(); | |
510f5507 | 7530 | drain_all_pages(cc.zone); |
041d3a8c MN |
7531 | |
7532 | order = 0; | |
7533 | outer_start = start; | |
7534 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7535 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7536 | outer_start = start; |
7537 | break; | |
041d3a8c MN |
7538 | } |
7539 | outer_start &= ~0UL << order; | |
7540 | } | |
7541 | ||
8ef5849f JK |
7542 | if (outer_start != start) { |
7543 | order = page_order(pfn_to_page(outer_start)); | |
7544 | ||
7545 | /* | |
7546 | * outer_start page could be small order buddy page and | |
7547 | * it doesn't include start page. Adjust outer_start | |
7548 | * in this case to report failed page properly | |
7549 | * on tracepoint in test_pages_isolated() | |
7550 | */ | |
7551 | if (outer_start + (1UL << order) <= start) | |
7552 | outer_start = start; | |
7553 | } | |
7554 | ||
041d3a8c | 7555 | /* Make sure the range is really isolated. */ |
b023f468 | 7556 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
7557 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
7558 | __func__, outer_start, end); | |
041d3a8c MN |
7559 | ret = -EBUSY; |
7560 | goto done; | |
7561 | } | |
7562 | ||
49f223a9 | 7563 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7564 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7565 | if (!outer_end) { |
7566 | ret = -EBUSY; | |
7567 | goto done; | |
7568 | } | |
7569 | ||
7570 | /* Free head and tail (if any) */ | |
7571 | if (start != outer_start) | |
7572 | free_contig_range(outer_start, start - outer_start); | |
7573 | if (end != outer_end) | |
7574 | free_contig_range(end, outer_end - end); | |
7575 | ||
7576 | done: | |
7577 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7578 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7579 | return ret; |
7580 | } | |
7581 | ||
7582 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7583 | { | |
bcc2b02f MS |
7584 | unsigned int count = 0; |
7585 | ||
7586 | for (; nr_pages--; pfn++) { | |
7587 | struct page *page = pfn_to_page(pfn); | |
7588 | ||
7589 | count += page_count(page) != 1; | |
7590 | __free_page(page); | |
7591 | } | |
7592 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7593 | } |
7594 | #endif | |
7595 | ||
4ed7e022 | 7596 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7597 | /* |
7598 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7599 | * page high values need to be recalulated. | |
7600 | */ | |
4ed7e022 JL |
7601 | void __meminit zone_pcp_update(struct zone *zone) |
7602 | { | |
0a647f38 | 7603 | unsigned cpu; |
c8e251fa | 7604 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7605 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7606 | pageset_set_high_and_batch(zone, |
7607 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7608 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7609 | } |
7610 | #endif | |
7611 | ||
340175b7 JL |
7612 | void zone_pcp_reset(struct zone *zone) |
7613 | { | |
7614 | unsigned long flags; | |
5a883813 MK |
7615 | int cpu; |
7616 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7617 | |
7618 | /* avoid races with drain_pages() */ | |
7619 | local_irq_save(flags); | |
7620 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7621 | for_each_online_cpu(cpu) { |
7622 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7623 | drain_zonestat(zone, pset); | |
7624 | } | |
340175b7 JL |
7625 | free_percpu(zone->pageset); |
7626 | zone->pageset = &boot_pageset; | |
7627 | } | |
7628 | local_irq_restore(flags); | |
7629 | } | |
7630 | ||
6dcd73d7 | 7631 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 7632 | /* |
b9eb6319 JK |
7633 | * All pages in the range must be in a single zone and isolated |
7634 | * before calling this. | |
0c0e6195 KH |
7635 | */ |
7636 | void | |
7637 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7638 | { | |
7639 | struct page *page; | |
7640 | struct zone *zone; | |
7aeb09f9 | 7641 | unsigned int order, i; |
0c0e6195 KH |
7642 | unsigned long pfn; |
7643 | unsigned long flags; | |
7644 | /* find the first valid pfn */ | |
7645 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
7646 | if (pfn_valid(pfn)) | |
7647 | break; | |
7648 | if (pfn == end_pfn) | |
7649 | return; | |
7650 | zone = page_zone(pfn_to_page(pfn)); | |
7651 | spin_lock_irqsave(&zone->lock, flags); | |
7652 | pfn = start_pfn; | |
7653 | while (pfn < end_pfn) { | |
7654 | if (!pfn_valid(pfn)) { | |
7655 | pfn++; | |
7656 | continue; | |
7657 | } | |
7658 | page = pfn_to_page(pfn); | |
b023f468 WC |
7659 | /* |
7660 | * The HWPoisoned page may be not in buddy system, and | |
7661 | * page_count() is not 0. | |
7662 | */ | |
7663 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
7664 | pfn++; | |
7665 | SetPageReserved(page); | |
7666 | continue; | |
7667 | } | |
7668 | ||
0c0e6195 KH |
7669 | BUG_ON(page_count(page)); |
7670 | BUG_ON(!PageBuddy(page)); | |
7671 | order = page_order(page); | |
7672 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
7673 | pr_info("remove from free list %lx %d %lx\n", |
7674 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
7675 | #endif |
7676 | list_del(&page->lru); | |
7677 | rmv_page_order(page); | |
7678 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
7679 | for (i = 0; i < (1 << order); i++) |
7680 | SetPageReserved((page+i)); | |
7681 | pfn += (1 << order); | |
7682 | } | |
7683 | spin_unlock_irqrestore(&zone->lock, flags); | |
7684 | } | |
7685 | #endif | |
8d22ba1b | 7686 | |
8d22ba1b WF |
7687 | bool is_free_buddy_page(struct page *page) |
7688 | { | |
7689 | struct zone *zone = page_zone(page); | |
7690 | unsigned long pfn = page_to_pfn(page); | |
7691 | unsigned long flags; | |
7aeb09f9 | 7692 | unsigned int order; |
8d22ba1b WF |
7693 | |
7694 | spin_lock_irqsave(&zone->lock, flags); | |
7695 | for (order = 0; order < MAX_ORDER; order++) { | |
7696 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
7697 | ||
7698 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
7699 | break; | |
7700 | } | |
7701 | spin_unlock_irqrestore(&zone->lock, flags); | |
7702 | ||
7703 | return order < MAX_ORDER; | |
7704 | } |