]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/stddef.h> |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
10ed273f | 22 | #include <linux/jiffies.h> |
1da177e4 | 23 | #include <linux/bootmem.h> |
edbe7d23 | 24 | #include <linux/memblock.h> |
1da177e4 | 25 | #include <linux/compiler.h> |
9f158333 | 26 | #include <linux/kernel.h> |
b1eeab67 | 27 | #include <linux/kmemcheck.h> |
b8c73fc2 | 28 | #include <linux/kasan.h> |
1da177e4 LT |
29 | #include <linux/module.h> |
30 | #include <linux/suspend.h> | |
31 | #include <linux/pagevec.h> | |
32 | #include <linux/blkdev.h> | |
33 | #include <linux/slab.h> | |
a238ab5b | 34 | #include <linux/ratelimit.h> |
5a3135c2 | 35 | #include <linux/oom.h> |
1da177e4 LT |
36 | #include <linux/notifier.h> |
37 | #include <linux/topology.h> | |
38 | #include <linux/sysctl.h> | |
39 | #include <linux/cpu.h> | |
40 | #include <linux/cpuset.h> | |
bdc8cb98 | 41 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
42 | #include <linux/nodemask.h> |
43 | #include <linux/vmalloc.h> | |
a6cccdc3 | 44 | #include <linux/vmstat.h> |
4be38e35 | 45 | #include <linux/mempolicy.h> |
4b94ffdc | 46 | #include <linux/memremap.h> |
6811378e | 47 | #include <linux/stop_machine.h> |
c713216d MG |
48 | #include <linux/sort.h> |
49 | #include <linux/pfn.h> | |
3fcfab16 | 50 | #include <linux/backing-dev.h> |
933e312e | 51 | #include <linux/fault-inject.h> |
a5d76b54 | 52 | #include <linux/page-isolation.h> |
eefa864b | 53 | #include <linux/page_ext.h> |
3ac7fe5a | 54 | #include <linux/debugobjects.h> |
dbb1f81c | 55 | #include <linux/kmemleak.h> |
56de7263 | 56 | #include <linux/compaction.h> |
0d3d062a | 57 | #include <trace/events/kmem.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
041d3a8c | 60 | #include <linux/migrate.h> |
e30825f1 | 61 | #include <linux/page_ext.h> |
949f7ec5 | 62 | #include <linux/hugetlb.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
48c96a36 | 64 | #include <linux/page_owner.h> |
0e1cc95b | 65 | #include <linux/kthread.h> |
1da177e4 | 66 | |
7ee3d4e8 | 67 | #include <asm/sections.h> |
1da177e4 | 68 | #include <asm/tlbflush.h> |
ac924c60 | 69 | #include <asm/div64.h> |
1da177e4 LT |
70 | #include "internal.h" |
71 | ||
c8e251fa CS |
72 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
73 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 74 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 75 | |
72812019 LS |
76 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
77 | DEFINE_PER_CPU(int, numa_node); | |
78 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
79 | #endif | |
80 | ||
7aac7898 LS |
81 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
82 | /* | |
83 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
84 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
85 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
86 | * defined in <linux/topology.h>. | |
87 | */ | |
88 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
89 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 90 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
91 | #endif |
92 | ||
1da177e4 | 93 | /* |
13808910 | 94 | * Array of node states. |
1da177e4 | 95 | */ |
13808910 CL |
96 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
97 | [N_POSSIBLE] = NODE_MASK_ALL, | |
98 | [N_ONLINE] = { { [0] = 1UL } }, | |
99 | #ifndef CONFIG_NUMA | |
100 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
101 | #ifdef CONFIG_HIGHMEM | |
102 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
103 | #endif |
104 | #ifdef CONFIG_MOVABLE_NODE | |
105 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
106 | #endif |
107 | [N_CPU] = { { [0] = 1UL } }, | |
108 | #endif /* NUMA */ | |
109 | }; | |
110 | EXPORT_SYMBOL(node_states); | |
111 | ||
c3d5f5f0 JL |
112 | /* Protect totalram_pages and zone->managed_pages */ |
113 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
114 | ||
6c231b7b | 115 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 116 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 117 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 118 | |
1b76b02f | 119 | int percpu_pagelist_fraction; |
dcce284a | 120 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 121 | |
bb14c2c7 VB |
122 | /* |
123 | * A cached value of the page's pageblock's migratetype, used when the page is | |
124 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
125 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
126 | * Also the migratetype set in the page does not necessarily match the pcplist | |
127 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
128 | * other index - this ensures that it will be put on the correct CMA freelist. | |
129 | */ | |
130 | static inline int get_pcppage_migratetype(struct page *page) | |
131 | { | |
132 | return page->index; | |
133 | } | |
134 | ||
135 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
136 | { | |
137 | page->index = migratetype; | |
138 | } | |
139 | ||
452aa699 RW |
140 | #ifdef CONFIG_PM_SLEEP |
141 | /* | |
142 | * The following functions are used by the suspend/hibernate code to temporarily | |
143 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
144 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
145 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
146 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
147 | * guaranteed not to run in parallel with that modification). | |
148 | */ | |
c9e664f1 RW |
149 | |
150 | static gfp_t saved_gfp_mask; | |
151 | ||
152 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
153 | { |
154 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
155 | if (saved_gfp_mask) { |
156 | gfp_allowed_mask = saved_gfp_mask; | |
157 | saved_gfp_mask = 0; | |
158 | } | |
452aa699 RW |
159 | } |
160 | ||
c9e664f1 | 161 | void pm_restrict_gfp_mask(void) |
452aa699 | 162 | { |
452aa699 | 163 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
164 | WARN_ON(saved_gfp_mask); |
165 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 166 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 167 | } |
f90ac398 MG |
168 | |
169 | bool pm_suspended_storage(void) | |
170 | { | |
d0164adc | 171 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
172 | return false; |
173 | return true; | |
174 | } | |
452aa699 RW |
175 | #endif /* CONFIG_PM_SLEEP */ |
176 | ||
d9c23400 | 177 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 178 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
179 | #endif |
180 | ||
d98c7a09 | 181 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 182 | |
1da177e4 LT |
183 | /* |
184 | * results with 256, 32 in the lowmem_reserve sysctl: | |
185 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
186 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
187 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
188 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 189 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
190 | * |
191 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
192 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 193 | */ |
2f1b6248 | 194 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 195 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 196 | 256, |
4b51d669 | 197 | #endif |
fb0e7942 | 198 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 199 | 256, |
fb0e7942 | 200 | #endif |
e53ef38d | 201 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 202 | 32, |
e53ef38d | 203 | #endif |
2a1e274a | 204 | 32, |
2f1b6248 | 205 | }; |
1da177e4 LT |
206 | |
207 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 208 | |
15ad7cdc | 209 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 210 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 211 | "DMA", |
4b51d669 | 212 | #endif |
fb0e7942 | 213 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 214 | "DMA32", |
fb0e7942 | 215 | #endif |
2f1b6248 | 216 | "Normal", |
e53ef38d | 217 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 218 | "HighMem", |
e53ef38d | 219 | #endif |
2a1e274a | 220 | "Movable", |
033fbae9 DW |
221 | #ifdef CONFIG_ZONE_DEVICE |
222 | "Device", | |
223 | #endif | |
2f1b6248 CL |
224 | }; |
225 | ||
60f30350 VB |
226 | char * const migratetype_names[MIGRATE_TYPES] = { |
227 | "Unmovable", | |
228 | "Movable", | |
229 | "Reclaimable", | |
230 | "HighAtomic", | |
231 | #ifdef CONFIG_CMA | |
232 | "CMA", | |
233 | #endif | |
234 | #ifdef CONFIG_MEMORY_ISOLATION | |
235 | "Isolate", | |
236 | #endif | |
237 | }; | |
238 | ||
f1e61557 KS |
239 | compound_page_dtor * const compound_page_dtors[] = { |
240 | NULL, | |
241 | free_compound_page, | |
242 | #ifdef CONFIG_HUGETLB_PAGE | |
243 | free_huge_page, | |
244 | #endif | |
9a982250 KS |
245 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
246 | free_transhuge_page, | |
247 | #endif | |
f1e61557 KS |
248 | }; |
249 | ||
1da177e4 | 250 | int min_free_kbytes = 1024; |
42aa83cb | 251 | int user_min_free_kbytes = -1; |
795ae7a0 | 252 | int watermark_scale_factor = 10; |
1da177e4 | 253 | |
2c85f51d JB |
254 | static unsigned long __meminitdata nr_kernel_pages; |
255 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 256 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 257 | |
0ee332c1 TH |
258 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
259 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
260 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
261 | static unsigned long __initdata required_kernelcore; | |
262 | static unsigned long __initdata required_movablecore; | |
263 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
342332e6 | 264 | static bool mirrored_kernelcore; |
0ee332c1 TH |
265 | |
266 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
267 | int movable_zone; | |
268 | EXPORT_SYMBOL(movable_zone); | |
269 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 270 | |
418508c1 MS |
271 | #if MAX_NUMNODES > 1 |
272 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 273 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 274 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 275 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
276 | #endif |
277 | ||
9ef9acb0 MG |
278 | int page_group_by_mobility_disabled __read_mostly; |
279 | ||
3a80a7fa MG |
280 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
281 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
282 | { | |
283 | pgdat->first_deferred_pfn = ULONG_MAX; | |
284 | } | |
285 | ||
286 | /* Returns true if the struct page for the pfn is uninitialised */ | |
0e1cc95b | 287 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 288 | { |
ae026b2a | 289 | if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn) |
3a80a7fa MG |
290 | return true; |
291 | ||
292 | return false; | |
293 | } | |
294 | ||
7e18adb4 MG |
295 | static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid) |
296 | { | |
297 | if (pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
298 | return true; | |
299 | ||
300 | return false; | |
301 | } | |
302 | ||
3a80a7fa MG |
303 | /* |
304 | * Returns false when the remaining initialisation should be deferred until | |
305 | * later in the boot cycle when it can be parallelised. | |
306 | */ | |
307 | static inline bool update_defer_init(pg_data_t *pgdat, | |
308 | unsigned long pfn, unsigned long zone_end, | |
309 | unsigned long *nr_initialised) | |
310 | { | |
987b3095 LZ |
311 | unsigned long max_initialise; |
312 | ||
3a80a7fa MG |
313 | /* Always populate low zones for address-contrained allocations */ |
314 | if (zone_end < pgdat_end_pfn(pgdat)) | |
315 | return true; | |
987b3095 LZ |
316 | /* |
317 | * Initialise at least 2G of a node but also take into account that | |
318 | * two large system hashes that can take up 1GB for 0.25TB/node. | |
319 | */ | |
320 | max_initialise = max(2UL << (30 - PAGE_SHIFT), | |
321 | (pgdat->node_spanned_pages >> 8)); | |
3a80a7fa | 322 | |
3a80a7fa | 323 | (*nr_initialised)++; |
987b3095 | 324 | if ((*nr_initialised > max_initialise) && |
3a80a7fa MG |
325 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
326 | pgdat->first_deferred_pfn = pfn; | |
327 | return false; | |
328 | } | |
329 | ||
330 | return true; | |
331 | } | |
332 | #else | |
333 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
334 | { | |
335 | } | |
336 | ||
337 | static inline bool early_page_uninitialised(unsigned long pfn) | |
338 | { | |
339 | return false; | |
340 | } | |
341 | ||
7e18adb4 MG |
342 | static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid) |
343 | { | |
344 | return false; | |
345 | } | |
346 | ||
3a80a7fa MG |
347 | static inline bool update_defer_init(pg_data_t *pgdat, |
348 | unsigned long pfn, unsigned long zone_end, | |
349 | unsigned long *nr_initialised) | |
350 | { | |
351 | return true; | |
352 | } | |
353 | #endif | |
354 | ||
0b423ca2 MG |
355 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
356 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
357 | unsigned long pfn) | |
358 | { | |
359 | #ifdef CONFIG_SPARSEMEM | |
360 | return __pfn_to_section(pfn)->pageblock_flags; | |
361 | #else | |
362 | return page_zone(page)->pageblock_flags; | |
363 | #endif /* CONFIG_SPARSEMEM */ | |
364 | } | |
365 | ||
366 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
367 | { | |
368 | #ifdef CONFIG_SPARSEMEM | |
369 | pfn &= (PAGES_PER_SECTION-1); | |
370 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
371 | #else | |
372 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
373 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
374 | #endif /* CONFIG_SPARSEMEM */ | |
375 | } | |
376 | ||
377 | /** | |
378 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
379 | * @page: The page within the block of interest | |
380 | * @pfn: The target page frame number | |
381 | * @end_bitidx: The last bit of interest to retrieve | |
382 | * @mask: mask of bits that the caller is interested in | |
383 | * | |
384 | * Return: pageblock_bits flags | |
385 | */ | |
386 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
387 | unsigned long pfn, | |
388 | unsigned long end_bitidx, | |
389 | unsigned long mask) | |
390 | { | |
391 | unsigned long *bitmap; | |
392 | unsigned long bitidx, word_bitidx; | |
393 | unsigned long word; | |
394 | ||
395 | bitmap = get_pageblock_bitmap(page, pfn); | |
396 | bitidx = pfn_to_bitidx(page, pfn); | |
397 | word_bitidx = bitidx / BITS_PER_LONG; | |
398 | bitidx &= (BITS_PER_LONG-1); | |
399 | ||
400 | word = bitmap[word_bitidx]; | |
401 | bitidx += end_bitidx; | |
402 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
403 | } | |
404 | ||
405 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
406 | unsigned long end_bitidx, | |
407 | unsigned long mask) | |
408 | { | |
409 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
410 | } | |
411 | ||
412 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
413 | { | |
414 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
415 | } | |
416 | ||
417 | /** | |
418 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
419 | * @page: The page within the block of interest | |
420 | * @flags: The flags to set | |
421 | * @pfn: The target page frame number | |
422 | * @end_bitidx: The last bit of interest | |
423 | * @mask: mask of bits that the caller is interested in | |
424 | */ | |
425 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
426 | unsigned long pfn, | |
427 | unsigned long end_bitidx, | |
428 | unsigned long mask) | |
429 | { | |
430 | unsigned long *bitmap; | |
431 | unsigned long bitidx, word_bitidx; | |
432 | unsigned long old_word, word; | |
433 | ||
434 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
435 | ||
436 | bitmap = get_pageblock_bitmap(page, pfn); | |
437 | bitidx = pfn_to_bitidx(page, pfn); | |
438 | word_bitidx = bitidx / BITS_PER_LONG; | |
439 | bitidx &= (BITS_PER_LONG-1); | |
440 | ||
441 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
442 | ||
443 | bitidx += end_bitidx; | |
444 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
445 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
446 | ||
447 | word = READ_ONCE(bitmap[word_bitidx]); | |
448 | for (;;) { | |
449 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
450 | if (word == old_word) | |
451 | break; | |
452 | word = old_word; | |
453 | } | |
454 | } | |
3a80a7fa | 455 | |
ee6f509c | 456 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 457 | { |
5d0f3f72 KM |
458 | if (unlikely(page_group_by_mobility_disabled && |
459 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
460 | migratetype = MIGRATE_UNMOVABLE; |
461 | ||
b2a0ac88 MG |
462 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
463 | PB_migrate, PB_migrate_end); | |
464 | } | |
465 | ||
13e7444b | 466 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 467 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 468 | { |
bdc8cb98 DH |
469 | int ret = 0; |
470 | unsigned seq; | |
471 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 472 | unsigned long sp, start_pfn; |
c6a57e19 | 473 | |
bdc8cb98 DH |
474 | do { |
475 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
476 | start_pfn = zone->zone_start_pfn; |
477 | sp = zone->spanned_pages; | |
108bcc96 | 478 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
479 | ret = 1; |
480 | } while (zone_span_seqretry(zone, seq)); | |
481 | ||
b5e6a5a2 | 482 | if (ret) |
613813e8 DH |
483 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
484 | pfn, zone_to_nid(zone), zone->name, | |
485 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 486 | |
bdc8cb98 | 487 | return ret; |
c6a57e19 DH |
488 | } |
489 | ||
490 | static int page_is_consistent(struct zone *zone, struct page *page) | |
491 | { | |
14e07298 | 492 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 493 | return 0; |
1da177e4 | 494 | if (zone != page_zone(page)) |
c6a57e19 DH |
495 | return 0; |
496 | ||
497 | return 1; | |
498 | } | |
499 | /* | |
500 | * Temporary debugging check for pages not lying within a given zone. | |
501 | */ | |
502 | static int bad_range(struct zone *zone, struct page *page) | |
503 | { | |
504 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 505 | return 1; |
c6a57e19 DH |
506 | if (!page_is_consistent(zone, page)) |
507 | return 1; | |
508 | ||
1da177e4 LT |
509 | return 0; |
510 | } | |
13e7444b NP |
511 | #else |
512 | static inline int bad_range(struct zone *zone, struct page *page) | |
513 | { | |
514 | return 0; | |
515 | } | |
516 | #endif | |
517 | ||
d230dec1 KS |
518 | static void bad_page(struct page *page, const char *reason, |
519 | unsigned long bad_flags) | |
1da177e4 | 520 | { |
d936cf9b HD |
521 | static unsigned long resume; |
522 | static unsigned long nr_shown; | |
523 | static unsigned long nr_unshown; | |
524 | ||
2a7684a2 WF |
525 | /* Don't complain about poisoned pages */ |
526 | if (PageHWPoison(page)) { | |
22b751c3 | 527 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
528 | return; |
529 | } | |
530 | ||
d936cf9b HD |
531 | /* |
532 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
533 | * or allow a steady drip of one report per second. | |
534 | */ | |
535 | if (nr_shown == 60) { | |
536 | if (time_before(jiffies, resume)) { | |
537 | nr_unshown++; | |
538 | goto out; | |
539 | } | |
540 | if (nr_unshown) { | |
ff8e8116 | 541 | pr_alert( |
1e9e6365 | 542 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
543 | nr_unshown); |
544 | nr_unshown = 0; | |
545 | } | |
546 | nr_shown = 0; | |
547 | } | |
548 | if (nr_shown++ == 0) | |
549 | resume = jiffies + 60 * HZ; | |
550 | ||
ff8e8116 | 551 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 552 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
553 | __dump_page(page, reason); |
554 | bad_flags &= page->flags; | |
555 | if (bad_flags) | |
556 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
557 | bad_flags, &bad_flags); | |
4e462112 | 558 | dump_page_owner(page); |
3dc14741 | 559 | |
4f31888c | 560 | print_modules(); |
1da177e4 | 561 | dump_stack(); |
d936cf9b | 562 | out: |
8cc3b392 | 563 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 564 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 565 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
566 | } |
567 | ||
1da177e4 LT |
568 | /* |
569 | * Higher-order pages are called "compound pages". They are structured thusly: | |
570 | * | |
1d798ca3 | 571 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 572 | * |
1d798ca3 KS |
573 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
574 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 575 | * |
1d798ca3 KS |
576 | * The first tail page's ->compound_dtor holds the offset in array of compound |
577 | * page destructors. See compound_page_dtors. | |
1da177e4 | 578 | * |
1d798ca3 | 579 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 580 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 581 | */ |
d98c7a09 | 582 | |
9a982250 | 583 | void free_compound_page(struct page *page) |
d98c7a09 | 584 | { |
d85f3385 | 585 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
586 | } |
587 | ||
d00181b9 | 588 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
589 | { |
590 | int i; | |
591 | int nr_pages = 1 << order; | |
592 | ||
f1e61557 | 593 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
594 | set_compound_order(page, order); |
595 | __SetPageHead(page); | |
596 | for (i = 1; i < nr_pages; i++) { | |
597 | struct page *p = page + i; | |
58a84aa9 | 598 | set_page_count(p, 0); |
1c290f64 | 599 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 600 | set_compound_head(p, page); |
18229df5 | 601 | } |
53f9263b | 602 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
603 | } |
604 | ||
c0a32fc5 SG |
605 | #ifdef CONFIG_DEBUG_PAGEALLOC |
606 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
607 | bool _debug_pagealloc_enabled __read_mostly |
608 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 609 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
610 | bool _debug_guardpage_enabled __read_mostly; |
611 | ||
031bc574 JK |
612 | static int __init early_debug_pagealloc(char *buf) |
613 | { | |
614 | if (!buf) | |
615 | return -EINVAL; | |
616 | ||
617 | if (strcmp(buf, "on") == 0) | |
618 | _debug_pagealloc_enabled = true; | |
619 | ||
ea6eabb0 CB |
620 | if (strcmp(buf, "off") == 0) |
621 | _debug_pagealloc_enabled = false; | |
622 | ||
031bc574 JK |
623 | return 0; |
624 | } | |
625 | early_param("debug_pagealloc", early_debug_pagealloc); | |
626 | ||
e30825f1 JK |
627 | static bool need_debug_guardpage(void) |
628 | { | |
031bc574 JK |
629 | /* If we don't use debug_pagealloc, we don't need guard page */ |
630 | if (!debug_pagealloc_enabled()) | |
631 | return false; | |
632 | ||
e30825f1 JK |
633 | return true; |
634 | } | |
635 | ||
636 | static void init_debug_guardpage(void) | |
637 | { | |
031bc574 JK |
638 | if (!debug_pagealloc_enabled()) |
639 | return; | |
640 | ||
e30825f1 JK |
641 | _debug_guardpage_enabled = true; |
642 | } | |
643 | ||
644 | struct page_ext_operations debug_guardpage_ops = { | |
645 | .need = need_debug_guardpage, | |
646 | .init = init_debug_guardpage, | |
647 | }; | |
c0a32fc5 SG |
648 | |
649 | static int __init debug_guardpage_minorder_setup(char *buf) | |
650 | { | |
651 | unsigned long res; | |
652 | ||
653 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 654 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
655 | return 0; |
656 | } | |
657 | _debug_guardpage_minorder = res; | |
1170532b | 658 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
659 | return 0; |
660 | } | |
661 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
662 | ||
2847cf95 JK |
663 | static inline void set_page_guard(struct zone *zone, struct page *page, |
664 | unsigned int order, int migratetype) | |
c0a32fc5 | 665 | { |
e30825f1 JK |
666 | struct page_ext *page_ext; |
667 | ||
668 | if (!debug_guardpage_enabled()) | |
669 | return; | |
670 | ||
671 | page_ext = lookup_page_ext(page); | |
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); | |
c0a32fc5 SG |
678 | } |
679 | ||
2847cf95 JK |
680 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
681 | unsigned int order, int migratetype) | |
c0a32fc5 | 682 | { |
e30825f1 JK |
683 | struct page_ext *page_ext; |
684 | ||
685 | if (!debug_guardpage_enabled()) | |
686 | return; | |
687 | ||
688 | page_ext = lookup_page_ext(page); | |
689 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
690 | ||
2847cf95 JK |
691 | set_page_private(page, 0); |
692 | if (!is_migrate_isolate(migratetype)) | |
693 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
694 | } |
695 | #else | |
e30825f1 | 696 | struct page_ext_operations debug_guardpage_ops = { NULL, }; |
2847cf95 JK |
697 | static inline void set_page_guard(struct zone *zone, struct page *page, |
698 | unsigned int order, int migratetype) {} | |
699 | static inline void clear_page_guard(struct zone *zone, struct page *page, | |
700 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
701 | #endif |
702 | ||
7aeb09f9 | 703 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 704 | { |
4c21e2f2 | 705 | set_page_private(page, order); |
676165a8 | 706 | __SetPageBuddy(page); |
1da177e4 LT |
707 | } |
708 | ||
709 | static inline void rmv_page_order(struct page *page) | |
710 | { | |
676165a8 | 711 | __ClearPageBuddy(page); |
4c21e2f2 | 712 | set_page_private(page, 0); |
1da177e4 LT |
713 | } |
714 | ||
1da177e4 LT |
715 | /* |
716 | * This function checks whether a page is free && is the buddy | |
717 | * we can do coalesce a page and its buddy if | |
13e7444b | 718 | * (a) the buddy is not in a hole && |
676165a8 | 719 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
720 | * (c) a page and its buddy have the same order && |
721 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 722 | * |
cf6fe945 WSH |
723 | * For recording whether a page is in the buddy system, we set ->_mapcount |
724 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
725 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
726 | * serialized by zone->lock. | |
1da177e4 | 727 | * |
676165a8 | 728 | * For recording page's order, we use page_private(page). |
1da177e4 | 729 | */ |
cb2b95e1 | 730 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 731 | unsigned int order) |
1da177e4 | 732 | { |
14e07298 | 733 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 734 | return 0; |
13e7444b | 735 | |
c0a32fc5 | 736 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
737 | if (page_zone_id(page) != page_zone_id(buddy)) |
738 | return 0; | |
739 | ||
4c5018ce WY |
740 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
741 | ||
c0a32fc5 SG |
742 | return 1; |
743 | } | |
744 | ||
cb2b95e1 | 745 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
746 | /* |
747 | * zone check is done late to avoid uselessly | |
748 | * calculating zone/node ids for pages that could | |
749 | * never merge. | |
750 | */ | |
751 | if (page_zone_id(page) != page_zone_id(buddy)) | |
752 | return 0; | |
753 | ||
4c5018ce WY |
754 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
755 | ||
6aa3001b | 756 | return 1; |
676165a8 | 757 | } |
6aa3001b | 758 | return 0; |
1da177e4 LT |
759 | } |
760 | ||
761 | /* | |
762 | * Freeing function for a buddy system allocator. | |
763 | * | |
764 | * The concept of a buddy system is to maintain direct-mapped table | |
765 | * (containing bit values) for memory blocks of various "orders". | |
766 | * The bottom level table contains the map for the smallest allocatable | |
767 | * units of memory (here, pages), and each level above it describes | |
768 | * pairs of units from the levels below, hence, "buddies". | |
769 | * At a high level, all that happens here is marking the table entry | |
770 | * at the bottom level available, and propagating the changes upward | |
771 | * as necessary, plus some accounting needed to play nicely with other | |
772 | * parts of the VM system. | |
773 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
774 | * free pages of length of (1 << order) and marked with _mapcount |
775 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
776 | * field. | |
1da177e4 | 777 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
778 | * other. That is, if we allocate a small block, and both were |
779 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 780 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 781 | * triggers coalescing into a block of larger size. |
1da177e4 | 782 | * |
6d49e352 | 783 | * -- nyc |
1da177e4 LT |
784 | */ |
785 | ||
48db57f8 | 786 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 787 | unsigned long pfn, |
ed0ae21d MG |
788 | struct zone *zone, unsigned int order, |
789 | int migratetype) | |
1da177e4 LT |
790 | { |
791 | unsigned long page_idx; | |
6dda9d55 | 792 | unsigned long combined_idx; |
43506fad | 793 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 794 | struct page *buddy; |
d9dddbf5 VB |
795 | unsigned int max_order; |
796 | ||
797 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 798 | |
d29bb978 | 799 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 800 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 801 | |
ed0ae21d | 802 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 803 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 804 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 805 | |
d9dddbf5 | 806 | page_idx = pfn & ((1 << MAX_ORDER) - 1); |
1da177e4 | 807 | |
309381fe SL |
808 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
809 | VM_BUG_ON_PAGE(bad_range(zone, page), page); | |
1da177e4 | 810 | |
d9dddbf5 | 811 | continue_merging: |
3c605096 | 812 | while (order < max_order - 1) { |
43506fad KC |
813 | buddy_idx = __find_buddy_index(page_idx, order); |
814 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 815 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 816 | goto done_merging; |
c0a32fc5 SG |
817 | /* |
818 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
819 | * merge with it and move up one order. | |
820 | */ | |
821 | if (page_is_guard(buddy)) { | |
2847cf95 | 822 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
823 | } else { |
824 | list_del(&buddy->lru); | |
825 | zone->free_area[order].nr_free--; | |
826 | rmv_page_order(buddy); | |
827 | } | |
43506fad | 828 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
829 | page = page + (combined_idx - page_idx); |
830 | page_idx = combined_idx; | |
831 | order++; | |
832 | } | |
d9dddbf5 VB |
833 | if (max_order < MAX_ORDER) { |
834 | /* If we are here, it means order is >= pageblock_order. | |
835 | * We want to prevent merge between freepages on isolate | |
836 | * pageblock and normal pageblock. Without this, pageblock | |
837 | * isolation could cause incorrect freepage or CMA accounting. | |
838 | * | |
839 | * We don't want to hit this code for the more frequent | |
840 | * low-order merging. | |
841 | */ | |
842 | if (unlikely(has_isolate_pageblock(zone))) { | |
843 | int buddy_mt; | |
844 | ||
845 | buddy_idx = __find_buddy_index(page_idx, order); | |
846 | buddy = page + (buddy_idx - page_idx); | |
847 | buddy_mt = get_pageblock_migratetype(buddy); | |
848 | ||
849 | if (migratetype != buddy_mt | |
850 | && (is_migrate_isolate(migratetype) || | |
851 | is_migrate_isolate(buddy_mt))) | |
852 | goto done_merging; | |
853 | } | |
854 | max_order++; | |
855 | goto continue_merging; | |
856 | } | |
857 | ||
858 | done_merging: | |
1da177e4 | 859 | set_page_order(page, order); |
6dda9d55 CZ |
860 | |
861 | /* | |
862 | * If this is not the largest possible page, check if the buddy | |
863 | * of the next-highest order is free. If it is, it's possible | |
864 | * that pages are being freed that will coalesce soon. In case, | |
865 | * that is happening, add the free page to the tail of the list | |
866 | * so it's less likely to be used soon and more likely to be merged | |
867 | * as a higher order page | |
868 | */ | |
b7f50cfa | 869 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 870 | struct page *higher_page, *higher_buddy; |
43506fad KC |
871 | combined_idx = buddy_idx & page_idx; |
872 | higher_page = page + (combined_idx - page_idx); | |
873 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 874 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
875 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
876 | list_add_tail(&page->lru, | |
877 | &zone->free_area[order].free_list[migratetype]); | |
878 | goto out; | |
879 | } | |
880 | } | |
881 | ||
882 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
883 | out: | |
1da177e4 LT |
884 | zone->free_area[order].nr_free++; |
885 | } | |
886 | ||
7bfec6f4 MG |
887 | /* |
888 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
889 | * try and check multiple fields with one check. The caller must do a detailed | |
890 | * check if necessary. | |
891 | */ | |
892 | static inline bool page_expected_state(struct page *page, | |
893 | unsigned long check_flags) | |
894 | { | |
895 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
896 | return false; | |
897 | ||
898 | if (unlikely((unsigned long)page->mapping | | |
899 | page_ref_count(page) | | |
900 | #ifdef CONFIG_MEMCG | |
901 | (unsigned long)page->mem_cgroup | | |
902 | #endif | |
903 | (page->flags & check_flags))) | |
904 | return false; | |
905 | ||
906 | return true; | |
907 | } | |
908 | ||
bb552ac6 | 909 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 910 | { |
7bfec6f4 MG |
911 | const char *bad_reason; |
912 | unsigned long bad_flags; | |
913 | ||
7bfec6f4 MG |
914 | bad_reason = NULL; |
915 | bad_flags = 0; | |
f0b791a3 | 916 | |
53f9263b | 917 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
918 | bad_reason = "nonzero mapcount"; |
919 | if (unlikely(page->mapping != NULL)) | |
920 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 921 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 922 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
923 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
924 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
925 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
926 | } | |
9edad6ea JW |
927 | #ifdef CONFIG_MEMCG |
928 | if (unlikely(page->mem_cgroup)) | |
929 | bad_reason = "page still charged to cgroup"; | |
930 | #endif | |
7bfec6f4 | 931 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
932 | } |
933 | ||
934 | static inline int free_pages_check(struct page *page) | |
935 | { | |
da838d4f | 936 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 937 | return 0; |
bb552ac6 MG |
938 | |
939 | /* Something has gone sideways, find it */ | |
940 | free_pages_check_bad(page); | |
7bfec6f4 | 941 | return 1; |
1da177e4 LT |
942 | } |
943 | ||
4db7548c MG |
944 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
945 | { | |
946 | int ret = 1; | |
947 | ||
948 | /* | |
949 | * We rely page->lru.next never has bit 0 set, unless the page | |
950 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
951 | */ | |
952 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
953 | ||
954 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
955 | ret = 0; | |
956 | goto out; | |
957 | } | |
958 | switch (page - head_page) { | |
959 | case 1: | |
960 | /* the first tail page: ->mapping is compound_mapcount() */ | |
961 | if (unlikely(compound_mapcount(page))) { | |
962 | bad_page(page, "nonzero compound_mapcount", 0); | |
963 | goto out; | |
964 | } | |
965 | break; | |
966 | case 2: | |
967 | /* | |
968 | * the second tail page: ->mapping is | |
969 | * page_deferred_list().next -- ignore value. | |
970 | */ | |
971 | break; | |
972 | default: | |
973 | if (page->mapping != TAIL_MAPPING) { | |
974 | bad_page(page, "corrupted mapping in tail page", 0); | |
975 | goto out; | |
976 | } | |
977 | break; | |
978 | } | |
979 | if (unlikely(!PageTail(page))) { | |
980 | bad_page(page, "PageTail not set", 0); | |
981 | goto out; | |
982 | } | |
983 | if (unlikely(compound_head(page) != head_page)) { | |
984 | bad_page(page, "compound_head not consistent", 0); | |
985 | goto out; | |
986 | } | |
987 | ret = 0; | |
988 | out: | |
989 | page->mapping = NULL; | |
990 | clear_compound_head(page); | |
991 | return ret; | |
992 | } | |
993 | ||
e2769dbd MG |
994 | static __always_inline bool free_pages_prepare(struct page *page, |
995 | unsigned int order, bool check_free) | |
4db7548c | 996 | { |
e2769dbd | 997 | int bad = 0; |
4db7548c | 998 | |
4db7548c MG |
999 | VM_BUG_ON_PAGE(PageTail(page), page); |
1000 | ||
e2769dbd MG |
1001 | trace_mm_page_free(page, order); |
1002 | kmemcheck_free_shadow(page, order); | |
1003 | kasan_free_pages(page, order); | |
1004 | ||
1005 | /* | |
1006 | * Check tail pages before head page information is cleared to | |
1007 | * avoid checking PageCompound for order-0 pages. | |
1008 | */ | |
1009 | if (unlikely(order)) { | |
1010 | bool compound = PageCompound(page); | |
1011 | int i; | |
1012 | ||
1013 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1014 | |
e2769dbd MG |
1015 | for (i = 1; i < (1 << order); i++) { |
1016 | if (compound) | |
1017 | bad += free_tail_pages_check(page, page + i); | |
1018 | if (unlikely(free_pages_check(page + i))) { | |
1019 | bad++; | |
1020 | continue; | |
1021 | } | |
1022 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1023 | } | |
1024 | } | |
4db7548c MG |
1025 | if (PageAnonHead(page)) |
1026 | page->mapping = NULL; | |
e2769dbd MG |
1027 | if (check_free) |
1028 | bad += free_pages_check(page); | |
1029 | if (bad) | |
1030 | return false; | |
4db7548c | 1031 | |
e2769dbd MG |
1032 | page_cpupid_reset_last(page); |
1033 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1034 | reset_page_owner(page, order); | |
4db7548c MG |
1035 | |
1036 | if (!PageHighMem(page)) { | |
1037 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1038 | PAGE_SIZE << order); |
4db7548c | 1039 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1040 | PAGE_SIZE << order); |
4db7548c | 1041 | } |
e2769dbd MG |
1042 | arch_free_page(page, order); |
1043 | kernel_poison_pages(page, 1 << order, 0); | |
1044 | kernel_map_pages(page, 1 << order, 0); | |
4db7548c | 1045 | |
4db7548c MG |
1046 | return true; |
1047 | } | |
1048 | ||
e2769dbd MG |
1049 | #ifdef CONFIG_DEBUG_VM |
1050 | static inline bool free_pcp_prepare(struct page *page) | |
1051 | { | |
1052 | return free_pages_prepare(page, 0, true); | |
1053 | } | |
1054 | ||
1055 | static inline bool bulkfree_pcp_prepare(struct page *page) | |
1056 | { | |
1057 | return false; | |
1058 | } | |
1059 | #else | |
1060 | static bool free_pcp_prepare(struct page *page) | |
1061 | { | |
1062 | return free_pages_prepare(page, 0, false); | |
1063 | } | |
1064 | ||
4db7548c MG |
1065 | static bool bulkfree_pcp_prepare(struct page *page) |
1066 | { | |
1067 | return free_pages_check(page); | |
1068 | } | |
1069 | #endif /* CONFIG_DEBUG_VM */ | |
1070 | ||
1da177e4 | 1071 | /* |
5f8dcc21 | 1072 | * Frees a number of pages from the PCP lists |
1da177e4 | 1073 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1074 | * count is the number of pages to free. |
1da177e4 LT |
1075 | * |
1076 | * If the zone was previously in an "all pages pinned" state then look to | |
1077 | * see if this freeing clears that state. | |
1078 | * | |
1079 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1080 | * pinned" detection logic. | |
1081 | */ | |
5f8dcc21 MG |
1082 | static void free_pcppages_bulk(struct zone *zone, int count, |
1083 | struct per_cpu_pages *pcp) | |
1da177e4 | 1084 | { |
5f8dcc21 | 1085 | int migratetype = 0; |
a6f9edd6 | 1086 | int batch_free = 0; |
0d5d823a | 1087 | unsigned long nr_scanned; |
3777999d | 1088 | bool isolated_pageblocks; |
5f8dcc21 | 1089 | |
c54ad30c | 1090 | spin_lock(&zone->lock); |
3777999d | 1091 | isolated_pageblocks = has_isolate_pageblock(zone); |
0d5d823a MG |
1092 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
1093 | if (nr_scanned) | |
1094 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 1095 | |
e5b31ac2 | 1096 | while (count) { |
48db57f8 | 1097 | struct page *page; |
5f8dcc21 MG |
1098 | struct list_head *list; |
1099 | ||
1100 | /* | |
a6f9edd6 MG |
1101 | * Remove pages from lists in a round-robin fashion. A |
1102 | * batch_free count is maintained that is incremented when an | |
1103 | * empty list is encountered. This is so more pages are freed | |
1104 | * off fuller lists instead of spinning excessively around empty | |
1105 | * lists | |
5f8dcc21 MG |
1106 | */ |
1107 | do { | |
a6f9edd6 | 1108 | batch_free++; |
5f8dcc21 MG |
1109 | if (++migratetype == MIGRATE_PCPTYPES) |
1110 | migratetype = 0; | |
1111 | list = &pcp->lists[migratetype]; | |
1112 | } while (list_empty(list)); | |
48db57f8 | 1113 | |
1d16871d NK |
1114 | /* This is the only non-empty list. Free them all. */ |
1115 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1116 | batch_free = count; |
1d16871d | 1117 | |
a6f9edd6 | 1118 | do { |
770c8aaa BZ |
1119 | int mt; /* migratetype of the to-be-freed page */ |
1120 | ||
a16601c5 | 1121 | page = list_last_entry(list, struct page, lru); |
a6f9edd6 MG |
1122 | /* must delete as __free_one_page list manipulates */ |
1123 | list_del(&page->lru); | |
aa016d14 | 1124 | |
bb14c2c7 | 1125 | mt = get_pcppage_migratetype(page); |
aa016d14 VB |
1126 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1127 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1128 | /* Pageblock could have been isolated meanwhile */ | |
3777999d | 1129 | if (unlikely(isolated_pageblocks)) |
51bb1a40 | 1130 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 1131 | |
4db7548c MG |
1132 | if (bulkfree_pcp_prepare(page)) |
1133 | continue; | |
1134 | ||
dc4b0caf | 1135 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 1136 | trace_mm_page_pcpu_drain(page, 0, mt); |
e5b31ac2 | 1137 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1138 | } |
c54ad30c | 1139 | spin_unlock(&zone->lock); |
1da177e4 LT |
1140 | } |
1141 | ||
dc4b0caf MG |
1142 | static void free_one_page(struct zone *zone, |
1143 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1144 | unsigned int order, |
ed0ae21d | 1145 | int migratetype) |
1da177e4 | 1146 | { |
0d5d823a | 1147 | unsigned long nr_scanned; |
006d22d9 | 1148 | spin_lock(&zone->lock); |
0d5d823a MG |
1149 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
1150 | if (nr_scanned) | |
1151 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 1152 | |
ad53f92e JK |
1153 | if (unlikely(has_isolate_pageblock(zone) || |
1154 | is_migrate_isolate(migratetype))) { | |
1155 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1156 | } |
dc4b0caf | 1157 | __free_one_page(page, pfn, zone, order, migratetype); |
006d22d9 | 1158 | spin_unlock(&zone->lock); |
48db57f8 NP |
1159 | } |
1160 | ||
1e8ce83c RH |
1161 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
1162 | unsigned long zone, int nid) | |
1163 | { | |
1e8ce83c | 1164 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1165 | init_page_count(page); |
1166 | page_mapcount_reset(page); | |
1167 | page_cpupid_reset_last(page); | |
1e8ce83c | 1168 | |
1e8ce83c RH |
1169 | INIT_LIST_HEAD(&page->lru); |
1170 | #ifdef WANT_PAGE_VIRTUAL | |
1171 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1172 | if (!is_highmem_idx(zone)) | |
1173 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1174 | #endif | |
1175 | } | |
1176 | ||
1177 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
1178 | int nid) | |
1179 | { | |
1180 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
1181 | } | |
1182 | ||
7e18adb4 MG |
1183 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1184 | static void init_reserved_page(unsigned long pfn) | |
1185 | { | |
1186 | pg_data_t *pgdat; | |
1187 | int nid, zid; | |
1188 | ||
1189 | if (!early_page_uninitialised(pfn)) | |
1190 | return; | |
1191 | ||
1192 | nid = early_pfn_to_nid(pfn); | |
1193 | pgdat = NODE_DATA(nid); | |
1194 | ||
1195 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1196 | struct zone *zone = &pgdat->node_zones[zid]; | |
1197 | ||
1198 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1199 | break; | |
1200 | } | |
1201 | __init_single_pfn(pfn, zid, nid); | |
1202 | } | |
1203 | #else | |
1204 | static inline void init_reserved_page(unsigned long pfn) | |
1205 | { | |
1206 | } | |
1207 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1208 | ||
92923ca3 NZ |
1209 | /* |
1210 | * Initialised pages do not have PageReserved set. This function is | |
1211 | * called for each range allocated by the bootmem allocator and | |
1212 | * marks the pages PageReserved. The remaining valid pages are later | |
1213 | * sent to the buddy page allocator. | |
1214 | */ | |
7e18adb4 | 1215 | void __meminit reserve_bootmem_region(unsigned long start, unsigned long end) |
92923ca3 NZ |
1216 | { |
1217 | unsigned long start_pfn = PFN_DOWN(start); | |
1218 | unsigned long end_pfn = PFN_UP(end); | |
1219 | ||
7e18adb4 MG |
1220 | for (; start_pfn < end_pfn; start_pfn++) { |
1221 | if (pfn_valid(start_pfn)) { | |
1222 | struct page *page = pfn_to_page(start_pfn); | |
1223 | ||
1224 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1225 | |
1226 | /* Avoid false-positive PageTail() */ | |
1227 | INIT_LIST_HEAD(&page->lru); | |
1228 | ||
7e18adb4 MG |
1229 | SetPageReserved(page); |
1230 | } | |
1231 | } | |
92923ca3 NZ |
1232 | } |
1233 | ||
ec95f53a KM |
1234 | static void __free_pages_ok(struct page *page, unsigned int order) |
1235 | { | |
1236 | unsigned long flags; | |
95e34412 | 1237 | int migratetype; |
dc4b0caf | 1238 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1239 | |
e2769dbd | 1240 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1241 | return; |
1242 | ||
cfc47a28 | 1243 | migratetype = get_pfnblock_migratetype(page, pfn); |
c54ad30c | 1244 | local_irq_save(flags); |
f8891e5e | 1245 | __count_vm_events(PGFREE, 1 << order); |
dc4b0caf | 1246 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
c54ad30c | 1247 | local_irq_restore(flags); |
1da177e4 LT |
1248 | } |
1249 | ||
949698a3 | 1250 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1251 | { |
c3993076 | 1252 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1253 | struct page *p = page; |
c3993076 | 1254 | unsigned int loop; |
a226f6c8 | 1255 | |
e2d0bd2b YL |
1256 | prefetchw(p); |
1257 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1258 | prefetchw(p + 1); | |
c3993076 JW |
1259 | __ClearPageReserved(p); |
1260 | set_page_count(p, 0); | |
a226f6c8 | 1261 | } |
e2d0bd2b YL |
1262 | __ClearPageReserved(p); |
1263 | set_page_count(p, 0); | |
c3993076 | 1264 | |
e2d0bd2b | 1265 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1266 | set_page_refcounted(page); |
1267 | __free_pages(page, order); | |
a226f6c8 DH |
1268 | } |
1269 | ||
75a592a4 MG |
1270 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1271 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1272 | |
75a592a4 MG |
1273 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1274 | ||
1275 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1276 | { | |
7ace9917 | 1277 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1278 | int nid; |
1279 | ||
7ace9917 | 1280 | spin_lock(&early_pfn_lock); |
75a592a4 | 1281 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 MG |
1282 | if (nid < 0) |
1283 | nid = 0; | |
1284 | spin_unlock(&early_pfn_lock); | |
1285 | ||
1286 | return nid; | |
75a592a4 MG |
1287 | } |
1288 | #endif | |
1289 | ||
1290 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
1291 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1292 | struct mminit_pfnnid_cache *state) | |
1293 | { | |
1294 | int nid; | |
1295 | ||
1296 | nid = __early_pfn_to_nid(pfn, state); | |
1297 | if (nid >= 0 && nid != node) | |
1298 | return false; | |
1299 | return true; | |
1300 | } | |
1301 | ||
1302 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1303 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1304 | { | |
1305 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1306 | } | |
1307 | ||
1308 | #else | |
1309 | ||
1310 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1311 | { | |
1312 | return true; | |
1313 | } | |
1314 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1315 | struct mminit_pfnnid_cache *state) | |
1316 | { | |
1317 | return true; | |
1318 | } | |
1319 | #endif | |
1320 | ||
1321 | ||
0e1cc95b | 1322 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1323 | unsigned int order) |
1324 | { | |
1325 | if (early_page_uninitialised(pfn)) | |
1326 | return; | |
949698a3 | 1327 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1328 | } |
1329 | ||
7cf91a98 JK |
1330 | /* |
1331 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1332 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1333 | * with the migration of free compaction scanner. The scanners then need to | |
1334 | * use only pfn_valid_within() check for arches that allow holes within | |
1335 | * pageblocks. | |
1336 | * | |
1337 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1338 | * | |
1339 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1340 | * i.e. it's possible that all pages within a zones range of pages do not | |
1341 | * belong to a single zone. We assume that a border between node0 and node1 | |
1342 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1343 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1344 | * the first and last page of a pageblock and avoid checking each individual | |
1345 | * page in a pageblock. | |
1346 | */ | |
1347 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1348 | unsigned long end_pfn, struct zone *zone) | |
1349 | { | |
1350 | struct page *start_page; | |
1351 | struct page *end_page; | |
1352 | ||
1353 | /* end_pfn is one past the range we are checking */ | |
1354 | end_pfn--; | |
1355 | ||
1356 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1357 | return NULL; | |
1358 | ||
1359 | start_page = pfn_to_page(start_pfn); | |
1360 | ||
1361 | if (page_zone(start_page) != zone) | |
1362 | return NULL; | |
1363 | ||
1364 | end_page = pfn_to_page(end_pfn); | |
1365 | ||
1366 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1367 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1368 | return NULL; | |
1369 | ||
1370 | return start_page; | |
1371 | } | |
1372 | ||
1373 | void set_zone_contiguous(struct zone *zone) | |
1374 | { | |
1375 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1376 | unsigned long block_end_pfn; | |
1377 | ||
1378 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1379 | for (; block_start_pfn < zone_end_pfn(zone); | |
1380 | block_start_pfn = block_end_pfn, | |
1381 | block_end_pfn += pageblock_nr_pages) { | |
1382 | ||
1383 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1384 | ||
1385 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1386 | block_end_pfn, zone)) | |
1387 | return; | |
1388 | } | |
1389 | ||
1390 | /* We confirm that there is no hole */ | |
1391 | zone->contiguous = true; | |
1392 | } | |
1393 | ||
1394 | void clear_zone_contiguous(struct zone *zone) | |
1395 | { | |
1396 | zone->contiguous = false; | |
1397 | } | |
1398 | ||
7e18adb4 | 1399 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
0e1cc95b | 1400 | static void __init deferred_free_range(struct page *page, |
a4de83dd MG |
1401 | unsigned long pfn, int nr_pages) |
1402 | { | |
1403 | int i; | |
1404 | ||
1405 | if (!page) | |
1406 | return; | |
1407 | ||
1408 | /* Free a large naturally-aligned chunk if possible */ | |
1409 | if (nr_pages == MAX_ORDER_NR_PAGES && | |
1410 | (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) { | |
ac5d2539 | 1411 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
949698a3 | 1412 | __free_pages_boot_core(page, MAX_ORDER-1); |
a4de83dd MG |
1413 | return; |
1414 | } | |
1415 | ||
949698a3 LZ |
1416 | for (i = 0; i < nr_pages; i++, page++) |
1417 | __free_pages_boot_core(page, 0); | |
a4de83dd MG |
1418 | } |
1419 | ||
d3cd131d NS |
1420 | /* Completion tracking for deferred_init_memmap() threads */ |
1421 | static atomic_t pgdat_init_n_undone __initdata; | |
1422 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1423 | ||
1424 | static inline void __init pgdat_init_report_one_done(void) | |
1425 | { | |
1426 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1427 | complete(&pgdat_init_all_done_comp); | |
1428 | } | |
0e1cc95b | 1429 | |
7e18adb4 | 1430 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1431 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1432 | { |
0e1cc95b MG |
1433 | pg_data_t *pgdat = data; |
1434 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1435 | struct mminit_pfnnid_cache nid_init_state = { }; |
1436 | unsigned long start = jiffies; | |
1437 | unsigned long nr_pages = 0; | |
1438 | unsigned long walk_start, walk_end; | |
1439 | int i, zid; | |
1440 | struct zone *zone; | |
7e18adb4 | 1441 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
0e1cc95b | 1442 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
7e18adb4 | 1443 | |
0e1cc95b | 1444 | if (first_init_pfn == ULONG_MAX) { |
d3cd131d | 1445 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1446 | return 0; |
1447 | } | |
1448 | ||
1449 | /* Bind memory initialisation thread to a local node if possible */ | |
1450 | if (!cpumask_empty(cpumask)) | |
1451 | set_cpus_allowed_ptr(current, cpumask); | |
7e18adb4 MG |
1452 | |
1453 | /* Sanity check boundaries */ | |
1454 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1455 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1456 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1457 | ||
1458 | /* Only the highest zone is deferred so find it */ | |
1459 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1460 | zone = pgdat->node_zones + zid; | |
1461 | if (first_init_pfn < zone_end_pfn(zone)) | |
1462 | break; | |
1463 | } | |
1464 | ||
1465 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { | |
1466 | unsigned long pfn, end_pfn; | |
54608c3f | 1467 | struct page *page = NULL; |
a4de83dd MG |
1468 | struct page *free_base_page = NULL; |
1469 | unsigned long free_base_pfn = 0; | |
1470 | int nr_to_free = 0; | |
7e18adb4 MG |
1471 | |
1472 | end_pfn = min(walk_end, zone_end_pfn(zone)); | |
1473 | pfn = first_init_pfn; | |
1474 | if (pfn < walk_start) | |
1475 | pfn = walk_start; | |
1476 | if (pfn < zone->zone_start_pfn) | |
1477 | pfn = zone->zone_start_pfn; | |
1478 | ||
1479 | for (; pfn < end_pfn; pfn++) { | |
54608c3f | 1480 | if (!pfn_valid_within(pfn)) |
a4de83dd | 1481 | goto free_range; |
7e18adb4 | 1482 | |
54608c3f MG |
1483 | /* |
1484 | * Ensure pfn_valid is checked every | |
1485 | * MAX_ORDER_NR_PAGES for memory holes | |
1486 | */ | |
1487 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
1488 | if (!pfn_valid(pfn)) { | |
1489 | page = NULL; | |
a4de83dd | 1490 | goto free_range; |
54608c3f MG |
1491 | } |
1492 | } | |
1493 | ||
1494 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { | |
1495 | page = NULL; | |
a4de83dd | 1496 | goto free_range; |
54608c3f MG |
1497 | } |
1498 | ||
1499 | /* Minimise pfn page lookups and scheduler checks */ | |
1500 | if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) { | |
1501 | page++; | |
1502 | } else { | |
a4de83dd MG |
1503 | nr_pages += nr_to_free; |
1504 | deferred_free_range(free_base_page, | |
1505 | free_base_pfn, nr_to_free); | |
1506 | free_base_page = NULL; | |
1507 | free_base_pfn = nr_to_free = 0; | |
1508 | ||
54608c3f MG |
1509 | page = pfn_to_page(pfn); |
1510 | cond_resched(); | |
1511 | } | |
7e18adb4 MG |
1512 | |
1513 | if (page->flags) { | |
1514 | VM_BUG_ON(page_zone(page) != zone); | |
a4de83dd | 1515 | goto free_range; |
7e18adb4 MG |
1516 | } |
1517 | ||
1518 | __init_single_page(page, pfn, zid, nid); | |
a4de83dd MG |
1519 | if (!free_base_page) { |
1520 | free_base_page = page; | |
1521 | free_base_pfn = pfn; | |
1522 | nr_to_free = 0; | |
1523 | } | |
1524 | nr_to_free++; | |
1525 | ||
1526 | /* Where possible, batch up pages for a single free */ | |
1527 | continue; | |
1528 | free_range: | |
1529 | /* Free the current block of pages to allocator */ | |
1530 | nr_pages += nr_to_free; | |
1531 | deferred_free_range(free_base_page, free_base_pfn, | |
1532 | nr_to_free); | |
1533 | free_base_page = NULL; | |
1534 | free_base_pfn = nr_to_free = 0; | |
7e18adb4 | 1535 | } |
a4de83dd | 1536 | |
7e18adb4 MG |
1537 | first_init_pfn = max(end_pfn, first_init_pfn); |
1538 | } | |
1539 | ||
1540 | /* Sanity check that the next zone really is unpopulated */ | |
1541 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1542 | ||
0e1cc95b | 1543 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1544 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1545 | |
1546 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1547 | return 0; |
1548 | } | |
7cf91a98 | 1549 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1550 | |
1551 | void __init page_alloc_init_late(void) | |
1552 | { | |
7cf91a98 JK |
1553 | struct zone *zone; |
1554 | ||
1555 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1556 | int nid; |
1557 | ||
d3cd131d NS |
1558 | /* There will be num_node_state(N_MEMORY) threads */ |
1559 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1560 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1561 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1562 | } | |
1563 | ||
1564 | /* Block until all are initialised */ | |
d3cd131d | 1565 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da MG |
1566 | |
1567 | /* Reinit limits that are based on free pages after the kernel is up */ | |
1568 | files_maxfiles_init(); | |
7cf91a98 JK |
1569 | #endif |
1570 | ||
1571 | for_each_populated_zone(zone) | |
1572 | set_zone_contiguous(zone); | |
7e18adb4 | 1573 | } |
7e18adb4 | 1574 | |
47118af0 | 1575 | #ifdef CONFIG_CMA |
9cf510a5 | 1576 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1577 | void __init init_cma_reserved_pageblock(struct page *page) |
1578 | { | |
1579 | unsigned i = pageblock_nr_pages; | |
1580 | struct page *p = page; | |
1581 | ||
1582 | do { | |
1583 | __ClearPageReserved(p); | |
1584 | set_page_count(p, 0); | |
1585 | } while (++p, --i); | |
1586 | ||
47118af0 | 1587 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1588 | |
1589 | if (pageblock_order >= MAX_ORDER) { | |
1590 | i = pageblock_nr_pages; | |
1591 | p = page; | |
1592 | do { | |
1593 | set_page_refcounted(p); | |
1594 | __free_pages(p, MAX_ORDER - 1); | |
1595 | p += MAX_ORDER_NR_PAGES; | |
1596 | } while (i -= MAX_ORDER_NR_PAGES); | |
1597 | } else { | |
1598 | set_page_refcounted(page); | |
1599 | __free_pages(page, pageblock_order); | |
1600 | } | |
1601 | ||
3dcc0571 | 1602 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1603 | } |
1604 | #endif | |
1da177e4 LT |
1605 | |
1606 | /* | |
1607 | * The order of subdivision here is critical for the IO subsystem. | |
1608 | * Please do not alter this order without good reasons and regression | |
1609 | * testing. Specifically, as large blocks of memory are subdivided, | |
1610 | * the order in which smaller blocks are delivered depends on the order | |
1611 | * they're subdivided in this function. This is the primary factor | |
1612 | * influencing the order in which pages are delivered to the IO | |
1613 | * subsystem according to empirical testing, and this is also justified | |
1614 | * by considering the behavior of a buddy system containing a single | |
1615 | * large block of memory acted on by a series of small allocations. | |
1616 | * This behavior is a critical factor in sglist merging's success. | |
1617 | * | |
6d49e352 | 1618 | * -- nyc |
1da177e4 | 1619 | */ |
085cc7d5 | 1620 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1621 | int low, int high, struct free_area *area, |
1622 | int migratetype) | |
1da177e4 LT |
1623 | { |
1624 | unsigned long size = 1 << high; | |
1625 | ||
1626 | while (high > low) { | |
1627 | area--; | |
1628 | high--; | |
1629 | size >>= 1; | |
309381fe | 1630 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1631 | |
2847cf95 | 1632 | if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && |
e30825f1 | 1633 | debug_guardpage_enabled() && |
2847cf95 | 1634 | high < debug_guardpage_minorder()) { |
c0a32fc5 SG |
1635 | /* |
1636 | * Mark as guard pages (or page), that will allow to | |
1637 | * merge back to allocator when buddy will be freed. | |
1638 | * Corresponding page table entries will not be touched, | |
1639 | * pages will stay not present in virtual address space | |
1640 | */ | |
2847cf95 | 1641 | set_page_guard(zone, &page[size], high, migratetype); |
c0a32fc5 SG |
1642 | continue; |
1643 | } | |
b2a0ac88 | 1644 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1645 | area->nr_free++; |
1646 | set_page_order(&page[size], high); | |
1647 | } | |
1da177e4 LT |
1648 | } |
1649 | ||
4e611801 | 1650 | static void check_new_page_bad(struct page *page) |
1da177e4 | 1651 | { |
4e611801 VB |
1652 | const char *bad_reason = NULL; |
1653 | unsigned long bad_flags = 0; | |
7bfec6f4 | 1654 | |
53f9263b | 1655 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1656 | bad_reason = "nonzero mapcount"; |
1657 | if (unlikely(page->mapping != NULL)) | |
1658 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1659 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1660 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1661 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1662 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1663 | bad_flags = __PG_HWPOISON; | |
1664 | } | |
f0b791a3 DH |
1665 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1666 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1667 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1668 | } | |
9edad6ea JW |
1669 | #ifdef CONFIG_MEMCG |
1670 | if (unlikely(page->mem_cgroup)) | |
1671 | bad_reason = "page still charged to cgroup"; | |
1672 | #endif | |
4e611801 VB |
1673 | bad_page(page, bad_reason, bad_flags); |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * This page is about to be returned from the page allocator | |
1678 | */ | |
1679 | static inline int check_new_page(struct page *page) | |
1680 | { | |
1681 | if (likely(page_expected_state(page, | |
1682 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1683 | return 0; | |
1684 | ||
1685 | check_new_page_bad(page); | |
1686 | return 1; | |
2a7684a2 WF |
1687 | } |
1688 | ||
1414c7f4 LA |
1689 | static inline bool free_pages_prezeroed(bool poisoned) |
1690 | { | |
1691 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
1692 | page_poisoning_enabled() && poisoned; | |
1693 | } | |
1694 | ||
479f854a MG |
1695 | #ifdef CONFIG_DEBUG_VM |
1696 | static bool check_pcp_refill(struct page *page) | |
1697 | { | |
1698 | return false; | |
1699 | } | |
1700 | ||
1701 | static bool check_new_pcp(struct page *page) | |
1702 | { | |
1703 | return check_new_page(page); | |
1704 | } | |
1705 | #else | |
1706 | static bool check_pcp_refill(struct page *page) | |
1707 | { | |
1708 | return check_new_page(page); | |
1709 | } | |
1710 | static bool check_new_pcp(struct page *page) | |
1711 | { | |
1712 | return false; | |
1713 | } | |
1714 | #endif /* CONFIG_DEBUG_VM */ | |
1715 | ||
1716 | static bool check_new_pages(struct page *page, unsigned int order) | |
1717 | { | |
1718 | int i; | |
1719 | for (i = 0; i < (1 << order); i++) { | |
1720 | struct page *p = page + i; | |
1721 | ||
1722 | if (unlikely(check_new_page(p))) | |
1723 | return true; | |
1724 | } | |
1725 | ||
1726 | return false; | |
1727 | } | |
1728 | ||
1729 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, | |
c603844b | 1730 | unsigned int alloc_flags) |
2a7684a2 WF |
1731 | { |
1732 | int i; | |
1414c7f4 | 1733 | bool poisoned = true; |
2a7684a2 WF |
1734 | |
1735 | for (i = 0; i < (1 << order); i++) { | |
1736 | struct page *p = page + i; | |
1414c7f4 LA |
1737 | if (poisoned) |
1738 | poisoned &= page_is_poisoned(p); | |
2a7684a2 | 1739 | } |
689bcebf | 1740 | |
4c21e2f2 | 1741 | set_page_private(page, 0); |
7835e98b | 1742 | set_page_refcounted(page); |
cc102509 NP |
1743 | |
1744 | arch_alloc_page(page, order); | |
1da177e4 | 1745 | kernel_map_pages(page, 1 << order, 1); |
8823b1db | 1746 | kernel_poison_pages(page, 1 << order, 1); |
b8c73fc2 | 1747 | kasan_alloc_pages(page, order); |
17cf4406 | 1748 | |
1414c7f4 | 1749 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1750 | for (i = 0; i < (1 << order); i++) |
1751 | clear_highpage(page + i); | |
17cf4406 NP |
1752 | |
1753 | if (order && (gfp_flags & __GFP_COMP)) | |
1754 | prep_compound_page(page, order); | |
1755 | ||
48c96a36 JK |
1756 | set_page_owner(page, order, gfp_flags); |
1757 | ||
75379191 | 1758 | /* |
2f064f34 | 1759 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1760 | * allocate the page. The expectation is that the caller is taking |
1761 | * steps that will free more memory. The caller should avoid the page | |
1762 | * being used for !PFMEMALLOC purposes. | |
1763 | */ | |
2f064f34 MH |
1764 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1765 | set_page_pfmemalloc(page); | |
1766 | else | |
1767 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1768 | } |
1769 | ||
56fd56b8 MG |
1770 | /* |
1771 | * Go through the free lists for the given migratetype and remove | |
1772 | * the smallest available page from the freelists | |
1773 | */ | |
728ec980 MG |
1774 | static inline |
1775 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1776 | int migratetype) |
1777 | { | |
1778 | unsigned int current_order; | |
b8af2941 | 1779 | struct free_area *area; |
56fd56b8 MG |
1780 | struct page *page; |
1781 | ||
1782 | /* Find a page of the appropriate size in the preferred list */ | |
1783 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1784 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1785 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1786 | struct page, lru); |
a16601c5 GT |
1787 | if (!page) |
1788 | continue; | |
56fd56b8 MG |
1789 | list_del(&page->lru); |
1790 | rmv_page_order(page); | |
1791 | area->nr_free--; | |
56fd56b8 | 1792 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1793 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1794 | return page; |
1795 | } | |
1796 | ||
1797 | return NULL; | |
1798 | } | |
1799 | ||
1800 | ||
b2a0ac88 MG |
1801 | /* |
1802 | * This array describes the order lists are fallen back to when | |
1803 | * the free lists for the desirable migrate type are depleted | |
1804 | */ | |
47118af0 | 1805 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1806 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1807 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1808 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1809 | #ifdef CONFIG_CMA |
974a786e | 1810 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1811 | #endif |
194159fb | 1812 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1813 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1814 | #endif |
b2a0ac88 MG |
1815 | }; |
1816 | ||
dc67647b JK |
1817 | #ifdef CONFIG_CMA |
1818 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1819 | unsigned int order) | |
1820 | { | |
1821 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1822 | } | |
1823 | #else | |
1824 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1825 | unsigned int order) { return NULL; } | |
1826 | #endif | |
1827 | ||
c361be55 MG |
1828 | /* |
1829 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1830 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1831 | * boundary. If alignment is required, use move_freepages_block() |
1832 | */ | |
435b405c | 1833 | int move_freepages(struct zone *zone, |
b69a7288 AB |
1834 | struct page *start_page, struct page *end_page, |
1835 | int migratetype) | |
c361be55 MG |
1836 | { |
1837 | struct page *page; | |
d00181b9 | 1838 | unsigned int order; |
d100313f | 1839 | int pages_moved = 0; |
c361be55 MG |
1840 | |
1841 | #ifndef CONFIG_HOLES_IN_ZONE | |
1842 | /* | |
1843 | * page_zone is not safe to call in this context when | |
1844 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1845 | * anyway as we check zone boundaries in move_freepages_block(). | |
1846 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1847 | * grouping pages by mobility |
c361be55 | 1848 | */ |
97ee4ba7 | 1849 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1850 | #endif |
1851 | ||
1852 | for (page = start_page; page <= end_page;) { | |
344c790e | 1853 | /* Make sure we are not inadvertently changing nodes */ |
309381fe | 1854 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
344c790e | 1855 | |
c361be55 MG |
1856 | if (!pfn_valid_within(page_to_pfn(page))) { |
1857 | page++; | |
1858 | continue; | |
1859 | } | |
1860 | ||
1861 | if (!PageBuddy(page)) { | |
1862 | page++; | |
1863 | continue; | |
1864 | } | |
1865 | ||
1866 | order = page_order(page); | |
84be48d8 KS |
1867 | list_move(&page->lru, |
1868 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 1869 | page += 1 << order; |
d100313f | 1870 | pages_moved += 1 << order; |
c361be55 MG |
1871 | } |
1872 | ||
d100313f | 1873 | return pages_moved; |
c361be55 MG |
1874 | } |
1875 | ||
ee6f509c | 1876 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1877 | int migratetype) |
c361be55 MG |
1878 | { |
1879 | unsigned long start_pfn, end_pfn; | |
1880 | struct page *start_page, *end_page; | |
1881 | ||
1882 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1883 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1884 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1885 | end_page = start_page + pageblock_nr_pages - 1; |
1886 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1887 | |
1888 | /* Do not cross zone boundaries */ | |
108bcc96 | 1889 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1890 | start_page = page; |
108bcc96 | 1891 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1892 | return 0; |
1893 | ||
1894 | return move_freepages(zone, start_page, end_page, migratetype); | |
1895 | } | |
1896 | ||
2f66a68f MG |
1897 | static void change_pageblock_range(struct page *pageblock_page, |
1898 | int start_order, int migratetype) | |
1899 | { | |
1900 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1901 | ||
1902 | while (nr_pageblocks--) { | |
1903 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1904 | pageblock_page += pageblock_nr_pages; | |
1905 | } | |
1906 | } | |
1907 | ||
fef903ef | 1908 | /* |
9c0415eb VB |
1909 | * When we are falling back to another migratetype during allocation, try to |
1910 | * steal extra free pages from the same pageblocks to satisfy further | |
1911 | * allocations, instead of polluting multiple pageblocks. | |
1912 | * | |
1913 | * If we are stealing a relatively large buddy page, it is likely there will | |
1914 | * be more free pages in the pageblock, so try to steal them all. For | |
1915 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1916 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1917 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1918 | * pageblocks. | |
fef903ef | 1919 | */ |
4eb7dce6 JK |
1920 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1921 | { | |
1922 | /* | |
1923 | * Leaving this order check is intended, although there is | |
1924 | * relaxed order check in next check. The reason is that | |
1925 | * we can actually steal whole pageblock if this condition met, | |
1926 | * but, below check doesn't guarantee it and that is just heuristic | |
1927 | * so could be changed anytime. | |
1928 | */ | |
1929 | if (order >= pageblock_order) | |
1930 | return true; | |
1931 | ||
1932 | if (order >= pageblock_order / 2 || | |
1933 | start_mt == MIGRATE_RECLAIMABLE || | |
1934 | start_mt == MIGRATE_UNMOVABLE || | |
1935 | page_group_by_mobility_disabled) | |
1936 | return true; | |
1937 | ||
1938 | return false; | |
1939 | } | |
1940 | ||
1941 | /* | |
1942 | * This function implements actual steal behaviour. If order is large enough, | |
1943 | * we can steal whole pageblock. If not, we first move freepages in this | |
1944 | * pageblock and check whether half of pages are moved or not. If half of | |
1945 | * pages are moved, we can change migratetype of pageblock and permanently | |
1946 | * use it's pages as requested migratetype in the future. | |
1947 | */ | |
1948 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1949 | int start_type) | |
fef903ef | 1950 | { |
d00181b9 | 1951 | unsigned int current_order = page_order(page); |
4eb7dce6 | 1952 | int pages; |
fef903ef | 1953 | |
fef903ef SB |
1954 | /* Take ownership for orders >= pageblock_order */ |
1955 | if (current_order >= pageblock_order) { | |
1956 | change_pageblock_range(page, current_order, start_type); | |
3a1086fb | 1957 | return; |
fef903ef SB |
1958 | } |
1959 | ||
4eb7dce6 | 1960 | pages = move_freepages_block(zone, page, start_type); |
fef903ef | 1961 | |
4eb7dce6 JK |
1962 | /* Claim the whole block if over half of it is free */ |
1963 | if (pages >= (1 << (pageblock_order-1)) || | |
1964 | page_group_by_mobility_disabled) | |
1965 | set_pageblock_migratetype(page, start_type); | |
1966 | } | |
1967 | ||
2149cdae JK |
1968 | /* |
1969 | * Check whether there is a suitable fallback freepage with requested order. | |
1970 | * If only_stealable is true, this function returns fallback_mt only if | |
1971 | * we can steal other freepages all together. This would help to reduce | |
1972 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1973 | */ | |
1974 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1975 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1976 | { |
1977 | int i; | |
1978 | int fallback_mt; | |
1979 | ||
1980 | if (area->nr_free == 0) | |
1981 | return -1; | |
1982 | ||
1983 | *can_steal = false; | |
1984 | for (i = 0;; i++) { | |
1985 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 1986 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
1987 | break; |
1988 | ||
1989 | if (list_empty(&area->free_list[fallback_mt])) | |
1990 | continue; | |
fef903ef | 1991 | |
4eb7dce6 JK |
1992 | if (can_steal_fallback(order, migratetype)) |
1993 | *can_steal = true; | |
1994 | ||
2149cdae JK |
1995 | if (!only_stealable) |
1996 | return fallback_mt; | |
1997 | ||
1998 | if (*can_steal) | |
1999 | return fallback_mt; | |
fef903ef | 2000 | } |
4eb7dce6 JK |
2001 | |
2002 | return -1; | |
fef903ef SB |
2003 | } |
2004 | ||
0aaa29a5 MG |
2005 | /* |
2006 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2007 | * there are no empty page blocks that contain a page with a suitable order | |
2008 | */ | |
2009 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2010 | unsigned int alloc_order) | |
2011 | { | |
2012 | int mt; | |
2013 | unsigned long max_managed, flags; | |
2014 | ||
2015 | /* | |
2016 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2017 | * Check is race-prone but harmless. | |
2018 | */ | |
2019 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
2020 | if (zone->nr_reserved_highatomic >= max_managed) | |
2021 | return; | |
2022 | ||
2023 | spin_lock_irqsave(&zone->lock, flags); | |
2024 | ||
2025 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2026 | if (zone->nr_reserved_highatomic >= max_managed) | |
2027 | goto out_unlock; | |
2028 | ||
2029 | /* Yoink! */ | |
2030 | mt = get_pageblock_migratetype(page); | |
2031 | if (mt != MIGRATE_HIGHATOMIC && | |
2032 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { | |
2033 | zone->nr_reserved_highatomic += pageblock_nr_pages; | |
2034 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
2035 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); | |
2036 | } | |
2037 | ||
2038 | out_unlock: | |
2039 | spin_unlock_irqrestore(&zone->lock, flags); | |
2040 | } | |
2041 | ||
2042 | /* | |
2043 | * Used when an allocation is about to fail under memory pressure. This | |
2044 | * potentially hurts the reliability of high-order allocations when under | |
2045 | * intense memory pressure but failed atomic allocations should be easier | |
2046 | * to recover from than an OOM. | |
2047 | */ | |
2048 | static void unreserve_highatomic_pageblock(const struct alloc_context *ac) | |
2049 | { | |
2050 | struct zonelist *zonelist = ac->zonelist; | |
2051 | unsigned long flags; | |
2052 | struct zoneref *z; | |
2053 | struct zone *zone; | |
2054 | struct page *page; | |
2055 | int order; | |
2056 | ||
2057 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2058 | ac->nodemask) { | |
2059 | /* Preserve at least one pageblock */ | |
2060 | if (zone->nr_reserved_highatomic <= pageblock_nr_pages) | |
2061 | continue; | |
2062 | ||
2063 | spin_lock_irqsave(&zone->lock, flags); | |
2064 | for (order = 0; order < MAX_ORDER; order++) { | |
2065 | struct free_area *area = &(zone->free_area[order]); | |
2066 | ||
a16601c5 GT |
2067 | page = list_first_entry_or_null( |
2068 | &area->free_list[MIGRATE_HIGHATOMIC], | |
2069 | struct page, lru); | |
2070 | if (!page) | |
0aaa29a5 MG |
2071 | continue; |
2072 | ||
0aaa29a5 MG |
2073 | /* |
2074 | * It should never happen but changes to locking could | |
2075 | * inadvertently allow a per-cpu drain to add pages | |
2076 | * to MIGRATE_HIGHATOMIC while unreserving so be safe | |
2077 | * and watch for underflows. | |
2078 | */ | |
2079 | zone->nr_reserved_highatomic -= min(pageblock_nr_pages, | |
2080 | zone->nr_reserved_highatomic); | |
2081 | ||
2082 | /* | |
2083 | * Convert to ac->migratetype and avoid the normal | |
2084 | * pageblock stealing heuristics. Minimally, the caller | |
2085 | * is doing the work and needs the pages. More | |
2086 | * importantly, if the block was always converted to | |
2087 | * MIGRATE_UNMOVABLE or another type then the number | |
2088 | * of pageblocks that cannot be completely freed | |
2089 | * may increase. | |
2090 | */ | |
2091 | set_pageblock_migratetype(page, ac->migratetype); | |
2092 | move_freepages_block(zone, page, ac->migratetype); | |
2093 | spin_unlock_irqrestore(&zone->lock, flags); | |
2094 | return; | |
2095 | } | |
2096 | spin_unlock_irqrestore(&zone->lock, flags); | |
2097 | } | |
2098 | } | |
2099 | ||
b2a0ac88 | 2100 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 2101 | static inline struct page * |
7aeb09f9 | 2102 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 2103 | { |
b8af2941 | 2104 | struct free_area *area; |
7aeb09f9 | 2105 | unsigned int current_order; |
b2a0ac88 | 2106 | struct page *page; |
4eb7dce6 JK |
2107 | int fallback_mt; |
2108 | bool can_steal; | |
b2a0ac88 MG |
2109 | |
2110 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
2111 | for (current_order = MAX_ORDER-1; |
2112 | current_order >= order && current_order <= MAX_ORDER-1; | |
2113 | --current_order) { | |
4eb7dce6 JK |
2114 | area = &(zone->free_area[current_order]); |
2115 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2116 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2117 | if (fallback_mt == -1) |
2118 | continue; | |
b2a0ac88 | 2119 | |
a16601c5 | 2120 | page = list_first_entry(&area->free_list[fallback_mt], |
4eb7dce6 JK |
2121 | struct page, lru); |
2122 | if (can_steal) | |
2123 | steal_suitable_fallback(zone, page, start_migratetype); | |
b2a0ac88 | 2124 | |
4eb7dce6 JK |
2125 | /* Remove the page from the freelists */ |
2126 | area->nr_free--; | |
2127 | list_del(&page->lru); | |
2128 | rmv_page_order(page); | |
3a1086fb | 2129 | |
4eb7dce6 JK |
2130 | expand(zone, page, order, current_order, area, |
2131 | start_migratetype); | |
2132 | /* | |
bb14c2c7 | 2133 | * The pcppage_migratetype may differ from pageblock's |
4eb7dce6 | 2134 | * migratetype depending on the decisions in |
bb14c2c7 VB |
2135 | * find_suitable_fallback(). This is OK as long as it does not |
2136 | * differ for MIGRATE_CMA pageblocks. Those can be used as | |
2137 | * fallback only via special __rmqueue_cma_fallback() function | |
4eb7dce6 | 2138 | */ |
bb14c2c7 | 2139 | set_pcppage_migratetype(page, start_migratetype); |
e0fff1bd | 2140 | |
4eb7dce6 JK |
2141 | trace_mm_page_alloc_extfrag(page, order, current_order, |
2142 | start_migratetype, fallback_mt); | |
e0fff1bd | 2143 | |
4eb7dce6 | 2144 | return page; |
b2a0ac88 MG |
2145 | } |
2146 | ||
728ec980 | 2147 | return NULL; |
b2a0ac88 MG |
2148 | } |
2149 | ||
56fd56b8 | 2150 | /* |
1da177e4 LT |
2151 | * Do the hard work of removing an element from the buddy allocator. |
2152 | * Call me with the zone->lock already held. | |
2153 | */ | |
b2a0ac88 | 2154 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
6ac0206b | 2155 | int migratetype) |
1da177e4 | 2156 | { |
1da177e4 LT |
2157 | struct page *page; |
2158 | ||
56fd56b8 | 2159 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2160 | if (unlikely(!page)) { |
dc67647b JK |
2161 | if (migratetype == MIGRATE_MOVABLE) |
2162 | page = __rmqueue_cma_fallback(zone, order); | |
2163 | ||
2164 | if (!page) | |
2165 | page = __rmqueue_fallback(zone, order, migratetype); | |
728ec980 MG |
2166 | } |
2167 | ||
0d3d062a | 2168 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2169 | return page; |
1da177e4 LT |
2170 | } |
2171 | ||
5f63b720 | 2172 | /* |
1da177e4 LT |
2173 | * Obtain a specified number of elements from the buddy allocator, all under |
2174 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2175 | * Returns the number of new pages which were placed at *list. | |
2176 | */ | |
5f63b720 | 2177 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2178 | unsigned long count, struct list_head *list, |
b745bc85 | 2179 | int migratetype, bool cold) |
1da177e4 | 2180 | { |
5bcc9f86 | 2181 | int i; |
5f63b720 | 2182 | |
c54ad30c | 2183 | spin_lock(&zone->lock); |
1da177e4 | 2184 | for (i = 0; i < count; ++i) { |
6ac0206b | 2185 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 2186 | if (unlikely(page == NULL)) |
1da177e4 | 2187 | break; |
81eabcbe | 2188 | |
479f854a MG |
2189 | if (unlikely(check_pcp_refill(page))) |
2190 | continue; | |
2191 | ||
81eabcbe MG |
2192 | /* |
2193 | * Split buddy pages returned by expand() are received here | |
2194 | * in physical page order. The page is added to the callers and | |
2195 | * list and the list head then moves forward. From the callers | |
2196 | * perspective, the linked list is ordered by page number in | |
2197 | * some conditions. This is useful for IO devices that can | |
2198 | * merge IO requests if the physical pages are ordered | |
2199 | * properly. | |
2200 | */ | |
b745bc85 | 2201 | if (likely(!cold)) |
e084b2d9 MG |
2202 | list_add(&page->lru, list); |
2203 | else | |
2204 | list_add_tail(&page->lru, list); | |
81eabcbe | 2205 | list = &page->lru; |
bb14c2c7 | 2206 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2207 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2208 | -(1 << order)); | |
1da177e4 | 2209 | } |
f2260e6b | 2210 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 2211 | spin_unlock(&zone->lock); |
085cc7d5 | 2212 | return i; |
1da177e4 LT |
2213 | } |
2214 | ||
4ae7c039 | 2215 | #ifdef CONFIG_NUMA |
8fce4d8e | 2216 | /* |
4037d452 CL |
2217 | * Called from the vmstat counter updater to drain pagesets of this |
2218 | * currently executing processor on remote nodes after they have | |
2219 | * expired. | |
2220 | * | |
879336c3 CL |
2221 | * Note that this function must be called with the thread pinned to |
2222 | * a single processor. | |
8fce4d8e | 2223 | */ |
4037d452 | 2224 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2225 | { |
4ae7c039 | 2226 | unsigned long flags; |
7be12fc9 | 2227 | int to_drain, batch; |
4ae7c039 | 2228 | |
4037d452 | 2229 | local_irq_save(flags); |
4db0c3c2 | 2230 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2231 | to_drain = min(pcp->count, batch); |
2a13515c KM |
2232 | if (to_drain > 0) { |
2233 | free_pcppages_bulk(zone, to_drain, pcp); | |
2234 | pcp->count -= to_drain; | |
2235 | } | |
4037d452 | 2236 | local_irq_restore(flags); |
4ae7c039 CL |
2237 | } |
2238 | #endif | |
2239 | ||
9f8f2172 | 2240 | /* |
93481ff0 | 2241 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2242 | * |
2243 | * The processor must either be the current processor and the | |
2244 | * thread pinned to the current processor or a processor that | |
2245 | * is not online. | |
2246 | */ | |
93481ff0 | 2247 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2248 | { |
c54ad30c | 2249 | unsigned long flags; |
93481ff0 VB |
2250 | struct per_cpu_pageset *pset; |
2251 | struct per_cpu_pages *pcp; | |
1da177e4 | 2252 | |
93481ff0 VB |
2253 | local_irq_save(flags); |
2254 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2255 | |
93481ff0 VB |
2256 | pcp = &pset->pcp; |
2257 | if (pcp->count) { | |
2258 | free_pcppages_bulk(zone, pcp->count, pcp); | |
2259 | pcp->count = 0; | |
2260 | } | |
2261 | local_irq_restore(flags); | |
2262 | } | |
3dfa5721 | 2263 | |
93481ff0 VB |
2264 | /* |
2265 | * Drain pcplists of all zones on the indicated processor. | |
2266 | * | |
2267 | * The processor must either be the current processor and the | |
2268 | * thread pinned to the current processor or a processor that | |
2269 | * is not online. | |
2270 | */ | |
2271 | static void drain_pages(unsigned int cpu) | |
2272 | { | |
2273 | struct zone *zone; | |
2274 | ||
2275 | for_each_populated_zone(zone) { | |
2276 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2277 | } |
2278 | } | |
1da177e4 | 2279 | |
9f8f2172 CL |
2280 | /* |
2281 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2282 | * |
2283 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2284 | * the single zone's pages. | |
9f8f2172 | 2285 | */ |
93481ff0 | 2286 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2287 | { |
93481ff0 VB |
2288 | int cpu = smp_processor_id(); |
2289 | ||
2290 | if (zone) | |
2291 | drain_pages_zone(cpu, zone); | |
2292 | else | |
2293 | drain_pages(cpu); | |
9f8f2172 CL |
2294 | } |
2295 | ||
2296 | /* | |
74046494 GBY |
2297 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2298 | * | |
93481ff0 VB |
2299 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2300 | * | |
74046494 GBY |
2301 | * Note that this code is protected against sending an IPI to an offline |
2302 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
2303 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
2304 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
2305 | * before the call to on_each_cpu_mask(). | |
9f8f2172 | 2306 | */ |
93481ff0 | 2307 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2308 | { |
74046494 | 2309 | int cpu; |
74046494 GBY |
2310 | |
2311 | /* | |
2312 | * Allocate in the BSS so we wont require allocation in | |
2313 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2314 | */ | |
2315 | static cpumask_t cpus_with_pcps; | |
2316 | ||
2317 | /* | |
2318 | * We don't care about racing with CPU hotplug event | |
2319 | * as offline notification will cause the notified | |
2320 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2321 | * disables preemption as part of its processing | |
2322 | */ | |
2323 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2324 | struct per_cpu_pageset *pcp; |
2325 | struct zone *z; | |
74046494 | 2326 | bool has_pcps = false; |
93481ff0 VB |
2327 | |
2328 | if (zone) { | |
74046494 | 2329 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2330 | if (pcp->pcp.count) |
74046494 | 2331 | has_pcps = true; |
93481ff0 VB |
2332 | } else { |
2333 | for_each_populated_zone(z) { | |
2334 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2335 | if (pcp->pcp.count) { | |
2336 | has_pcps = true; | |
2337 | break; | |
2338 | } | |
74046494 GBY |
2339 | } |
2340 | } | |
93481ff0 | 2341 | |
74046494 GBY |
2342 | if (has_pcps) |
2343 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2344 | else | |
2345 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2346 | } | |
93481ff0 VB |
2347 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
2348 | zone, 1); | |
9f8f2172 CL |
2349 | } |
2350 | ||
296699de | 2351 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
2352 | |
2353 | void mark_free_pages(struct zone *zone) | |
2354 | { | |
f623f0db RW |
2355 | unsigned long pfn, max_zone_pfn; |
2356 | unsigned long flags; | |
7aeb09f9 | 2357 | unsigned int order, t; |
86760a2c | 2358 | struct page *page; |
1da177e4 | 2359 | |
8080fc03 | 2360 | if (zone_is_empty(zone)) |
1da177e4 LT |
2361 | return; |
2362 | ||
2363 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2364 | |
108bcc96 | 2365 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2366 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2367 | if (pfn_valid(pfn)) { | |
86760a2c | 2368 | page = pfn_to_page(pfn); |
ba6b0979 JK |
2369 | |
2370 | if (page_zone(page) != zone) | |
2371 | continue; | |
2372 | ||
7be98234 RW |
2373 | if (!swsusp_page_is_forbidden(page)) |
2374 | swsusp_unset_page_free(page); | |
f623f0db | 2375 | } |
1da177e4 | 2376 | |
b2a0ac88 | 2377 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2378 | list_for_each_entry(page, |
2379 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2380 | unsigned long i; |
1da177e4 | 2381 | |
86760a2c | 2382 | pfn = page_to_pfn(page); |
f623f0db | 2383 | for (i = 0; i < (1UL << order); i++) |
7be98234 | 2384 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 2385 | } |
b2a0ac88 | 2386 | } |
1da177e4 LT |
2387 | spin_unlock_irqrestore(&zone->lock, flags); |
2388 | } | |
e2c55dc8 | 2389 | #endif /* CONFIG_PM */ |
1da177e4 | 2390 | |
1da177e4 LT |
2391 | /* |
2392 | * Free a 0-order page | |
b745bc85 | 2393 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 2394 | */ |
b745bc85 | 2395 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
2396 | { |
2397 | struct zone *zone = page_zone(page); | |
2398 | struct per_cpu_pages *pcp; | |
2399 | unsigned long flags; | |
dc4b0caf | 2400 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 2401 | int migratetype; |
1da177e4 | 2402 | |
4db7548c | 2403 | if (!free_pcp_prepare(page)) |
689bcebf HD |
2404 | return; |
2405 | ||
dc4b0caf | 2406 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2407 | set_pcppage_migratetype(page, migratetype); |
1da177e4 | 2408 | local_irq_save(flags); |
f8891e5e | 2409 | __count_vm_event(PGFREE); |
da456f14 | 2410 | |
5f8dcc21 MG |
2411 | /* |
2412 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2413 | * Free ISOLATE pages back to the allocator because they are being | |
2414 | * offlined but treat RESERVE as movable pages so we can get those | |
2415 | * areas back if necessary. Otherwise, we may have to free | |
2416 | * excessively into the page allocator | |
2417 | */ | |
2418 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2419 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2420 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
2421 | goto out; |
2422 | } | |
2423 | migratetype = MIGRATE_MOVABLE; | |
2424 | } | |
2425 | ||
99dcc3e5 | 2426 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 2427 | if (!cold) |
5f8dcc21 | 2428 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
2429 | else |
2430 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 2431 | pcp->count++; |
48db57f8 | 2432 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2433 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
2434 | free_pcppages_bulk(zone, batch, pcp); |
2435 | pcp->count -= batch; | |
48db57f8 | 2436 | } |
5f8dcc21 MG |
2437 | |
2438 | out: | |
1da177e4 | 2439 | local_irq_restore(flags); |
1da177e4 LT |
2440 | } |
2441 | ||
cc59850e KK |
2442 | /* |
2443 | * Free a list of 0-order pages | |
2444 | */ | |
b745bc85 | 2445 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
2446 | { |
2447 | struct page *page, *next; | |
2448 | ||
2449 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 2450 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
2451 | free_hot_cold_page(page, cold); |
2452 | } | |
2453 | } | |
2454 | ||
8dfcc9ba NP |
2455 | /* |
2456 | * split_page takes a non-compound higher-order page, and splits it into | |
2457 | * n (1<<order) sub-pages: page[0..n] | |
2458 | * Each sub-page must be freed individually. | |
2459 | * | |
2460 | * Note: this is probably too low level an operation for use in drivers. | |
2461 | * Please consult with lkml before using this in your driver. | |
2462 | */ | |
2463 | void split_page(struct page *page, unsigned int order) | |
2464 | { | |
2465 | int i; | |
e2cfc911 | 2466 | gfp_t gfp_mask; |
8dfcc9ba | 2467 | |
309381fe SL |
2468 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2469 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
2470 | |
2471 | #ifdef CONFIG_KMEMCHECK | |
2472 | /* | |
2473 | * Split shadow pages too, because free(page[0]) would | |
2474 | * otherwise free the whole shadow. | |
2475 | */ | |
2476 | if (kmemcheck_page_is_tracked(page)) | |
2477 | split_page(virt_to_page(page[0].shadow), order); | |
2478 | #endif | |
2479 | ||
e2cfc911 JK |
2480 | gfp_mask = get_page_owner_gfp(page); |
2481 | set_page_owner(page, 0, gfp_mask); | |
48c96a36 | 2482 | for (i = 1; i < (1 << order); i++) { |
7835e98b | 2483 | set_page_refcounted(page + i); |
e2cfc911 | 2484 | set_page_owner(page + i, 0, gfp_mask); |
48c96a36 | 2485 | } |
8dfcc9ba | 2486 | } |
5853ff23 | 2487 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2488 | |
3c605096 | 2489 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2490 | { |
748446bb MG |
2491 | unsigned long watermark; |
2492 | struct zone *zone; | |
2139cbe6 | 2493 | int mt; |
748446bb MG |
2494 | |
2495 | BUG_ON(!PageBuddy(page)); | |
2496 | ||
2497 | zone = page_zone(page); | |
2e30abd1 | 2498 | mt = get_pageblock_migratetype(page); |
748446bb | 2499 | |
194159fb | 2500 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
2501 | /* Obey watermarks as if the page was being allocated */ |
2502 | watermark = low_wmark_pages(zone) + (1 << order); | |
2503 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
2504 | return 0; | |
2505 | ||
8fb74b9f | 2506 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2507 | } |
748446bb MG |
2508 | |
2509 | /* Remove page from free list */ | |
2510 | list_del(&page->lru); | |
2511 | zone->free_area[order].nr_free--; | |
2512 | rmv_page_order(page); | |
2139cbe6 | 2513 | |
e2cfc911 | 2514 | set_page_owner(page, order, __GFP_MOVABLE); |
f3a14ced | 2515 | |
8fb74b9f | 2516 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
2517 | if (order >= pageblock_order - 1) { |
2518 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2519 | for (; page < endpage; page += pageblock_nr_pages) { |
2520 | int mt = get_pageblock_migratetype(page); | |
194159fb | 2521 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
2522 | set_pageblock_migratetype(page, |
2523 | MIGRATE_MOVABLE); | |
2524 | } | |
748446bb MG |
2525 | } |
2526 | ||
f3a14ced | 2527 | |
8fb74b9f | 2528 | return 1UL << order; |
1fb3f8ca MG |
2529 | } |
2530 | ||
2531 | /* | |
2532 | * Similar to split_page except the page is already free. As this is only | |
2533 | * being used for migration, the migratetype of the block also changes. | |
2534 | * As this is called with interrupts disabled, the caller is responsible | |
2535 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
2536 | * are enabled. | |
2537 | * | |
2538 | * Note: this is probably too low level an operation for use in drivers. | |
2539 | * Please consult with lkml before using this in your driver. | |
2540 | */ | |
2541 | int split_free_page(struct page *page) | |
2542 | { | |
2543 | unsigned int order; | |
2544 | int nr_pages; | |
2545 | ||
1fb3f8ca MG |
2546 | order = page_order(page); |
2547 | ||
8fb74b9f | 2548 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
2549 | if (!nr_pages) |
2550 | return 0; | |
2551 | ||
2552 | /* Split into individual pages */ | |
2553 | set_page_refcounted(page); | |
2554 | split_page(page, order); | |
2555 | return nr_pages; | |
748446bb MG |
2556 | } |
2557 | ||
060e7417 MG |
2558 | /* |
2559 | * Update NUMA hit/miss statistics | |
2560 | * | |
2561 | * Must be called with interrupts disabled. | |
2562 | * | |
2563 | * When __GFP_OTHER_NODE is set assume the node of the preferred | |
2564 | * zone is the local node. This is useful for daemons who allocate | |
2565 | * memory on behalf of other processes. | |
2566 | */ | |
2567 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, | |
2568 | gfp_t flags) | |
2569 | { | |
2570 | #ifdef CONFIG_NUMA | |
2571 | int local_nid = numa_node_id(); | |
2572 | enum zone_stat_item local_stat = NUMA_LOCAL; | |
2573 | ||
2574 | if (unlikely(flags & __GFP_OTHER_NODE)) { | |
2575 | local_stat = NUMA_OTHER; | |
2576 | local_nid = preferred_zone->node; | |
2577 | } | |
2578 | ||
2579 | if (z->node == local_nid) { | |
2580 | __inc_zone_state(z, NUMA_HIT); | |
2581 | __inc_zone_state(z, local_stat); | |
2582 | } else { | |
2583 | __inc_zone_state(z, NUMA_MISS); | |
2584 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); | |
2585 | } | |
2586 | #endif | |
2587 | } | |
2588 | ||
1da177e4 | 2589 | /* |
75379191 | 2590 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2591 | */ |
0a15c3e9 MG |
2592 | static inline |
2593 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
7aeb09f9 | 2594 | struct zone *zone, unsigned int order, |
c603844b MG |
2595 | gfp_t gfp_flags, unsigned int alloc_flags, |
2596 | int migratetype) | |
1da177e4 LT |
2597 | { |
2598 | unsigned long flags; | |
689bcebf | 2599 | struct page *page; |
b745bc85 | 2600 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
1da177e4 | 2601 | |
48db57f8 | 2602 | if (likely(order == 0)) { |
1da177e4 | 2603 | struct per_cpu_pages *pcp; |
5f8dcc21 | 2604 | struct list_head *list; |
1da177e4 | 2605 | |
1da177e4 | 2606 | local_irq_save(flags); |
479f854a MG |
2607 | do { |
2608 | pcp = &this_cpu_ptr(zone->pageset)->pcp; | |
2609 | list = &pcp->lists[migratetype]; | |
2610 | if (list_empty(list)) { | |
2611 | pcp->count += rmqueue_bulk(zone, 0, | |
2612 | pcp->batch, list, | |
2613 | migratetype, cold); | |
2614 | if (unlikely(list_empty(list))) | |
2615 | goto failed; | |
2616 | } | |
b92a6edd | 2617 | |
479f854a MG |
2618 | if (cold) |
2619 | page = list_last_entry(list, struct page, lru); | |
2620 | else | |
2621 | page = list_first_entry(list, struct page, lru); | |
2622 | } while (page && check_new_pcp(page)); | |
5f8dcc21 | 2623 | |
754078eb | 2624 | __dec_zone_state(zone, NR_ALLOC_BATCH); |
b92a6edd MG |
2625 | list_del(&page->lru); |
2626 | pcp->count--; | |
7fb1d9fc | 2627 | } else { |
0f352e53 MH |
2628 | /* |
2629 | * We most definitely don't want callers attempting to | |
2630 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2631 | */ | |
2632 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
1da177e4 | 2633 | spin_lock_irqsave(&zone->lock, flags); |
0aaa29a5 | 2634 | |
479f854a MG |
2635 | do { |
2636 | page = NULL; | |
2637 | if (alloc_flags & ALLOC_HARDER) { | |
2638 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2639 | if (page) | |
2640 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
2641 | } | |
2642 | if (!page) | |
2643 | page = __rmqueue(zone, order, migratetype); | |
2644 | } while (page && check_new_pages(page, order)); | |
a74609fa NP |
2645 | spin_unlock(&zone->lock); |
2646 | if (!page) | |
2647 | goto failed; | |
754078eb | 2648 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
d1ce749a | 2649 | __mod_zone_freepage_state(zone, -(1 << order), |
bb14c2c7 | 2650 | get_pcppage_migratetype(page)); |
1da177e4 LT |
2651 | } |
2652 | ||
abe5f972 | 2653 | if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 && |
57054651 JW |
2654 | !test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) |
2655 | set_bit(ZONE_FAIR_DEPLETED, &zone->flags); | |
27329369 | 2656 | |
f8891e5e | 2657 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 2658 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 2659 | local_irq_restore(flags); |
1da177e4 | 2660 | |
309381fe | 2661 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 2662 | return page; |
a74609fa NP |
2663 | |
2664 | failed: | |
2665 | local_irq_restore(flags); | |
a74609fa | 2666 | return NULL; |
1da177e4 LT |
2667 | } |
2668 | ||
933e312e AM |
2669 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2670 | ||
b2588c4b | 2671 | static struct { |
933e312e AM |
2672 | struct fault_attr attr; |
2673 | ||
621a5f7a | 2674 | bool ignore_gfp_highmem; |
71baba4b | 2675 | bool ignore_gfp_reclaim; |
54114994 | 2676 | u32 min_order; |
933e312e AM |
2677 | } fail_page_alloc = { |
2678 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 2679 | .ignore_gfp_reclaim = true, |
621a5f7a | 2680 | .ignore_gfp_highmem = true, |
54114994 | 2681 | .min_order = 1, |
933e312e AM |
2682 | }; |
2683 | ||
2684 | static int __init setup_fail_page_alloc(char *str) | |
2685 | { | |
2686 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
2687 | } | |
2688 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
2689 | ||
deaf386e | 2690 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2691 | { |
54114994 | 2692 | if (order < fail_page_alloc.min_order) |
deaf386e | 2693 | return false; |
933e312e | 2694 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 2695 | return false; |
933e312e | 2696 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 2697 | return false; |
71baba4b MG |
2698 | if (fail_page_alloc.ignore_gfp_reclaim && |
2699 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 2700 | return false; |
933e312e AM |
2701 | |
2702 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
2703 | } | |
2704 | ||
2705 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
2706 | ||
2707 | static int __init fail_page_alloc_debugfs(void) | |
2708 | { | |
f4ae40a6 | 2709 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 2710 | struct dentry *dir; |
933e312e | 2711 | |
dd48c085 AM |
2712 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2713 | &fail_page_alloc.attr); | |
2714 | if (IS_ERR(dir)) | |
2715 | return PTR_ERR(dir); | |
933e312e | 2716 | |
b2588c4b | 2717 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 2718 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
2719 | goto fail; |
2720 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
2721 | &fail_page_alloc.ignore_gfp_highmem)) | |
2722 | goto fail; | |
2723 | if (!debugfs_create_u32("min-order", mode, dir, | |
2724 | &fail_page_alloc.min_order)) | |
2725 | goto fail; | |
2726 | ||
2727 | return 0; | |
2728 | fail: | |
dd48c085 | 2729 | debugfs_remove_recursive(dir); |
933e312e | 2730 | |
b2588c4b | 2731 | return -ENOMEM; |
933e312e AM |
2732 | } |
2733 | ||
2734 | late_initcall(fail_page_alloc_debugfs); | |
2735 | ||
2736 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2737 | ||
2738 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2739 | ||
deaf386e | 2740 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2741 | { |
deaf386e | 2742 | return false; |
933e312e AM |
2743 | } |
2744 | ||
2745 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2746 | ||
1da177e4 | 2747 | /* |
97a16fc8 MG |
2748 | * Return true if free base pages are above 'mark'. For high-order checks it |
2749 | * will return true of the order-0 watermark is reached and there is at least | |
2750 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
2751 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 2752 | */ |
7aeb09f9 | 2753 | static bool __zone_watermark_ok(struct zone *z, unsigned int order, |
c603844b MG |
2754 | unsigned long mark, int classzone_idx, |
2755 | unsigned int alloc_flags, | |
7aeb09f9 | 2756 | long free_pages) |
1da177e4 | 2757 | { |
d23ad423 | 2758 | long min = mark; |
1da177e4 | 2759 | int o; |
c603844b | 2760 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
1da177e4 | 2761 | |
0aaa29a5 | 2762 | /* free_pages may go negative - that's OK */ |
df0a6daa | 2763 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 2764 | |
7fb1d9fc | 2765 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2766 | min -= min / 2; |
0aaa29a5 MG |
2767 | |
2768 | /* | |
2769 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
2770 | * the high-atomic reserves. This will over-estimate the size of the | |
2771 | * atomic reserve but it avoids a search. | |
2772 | */ | |
97a16fc8 | 2773 | if (likely(!alloc_harder)) |
0aaa29a5 MG |
2774 | free_pages -= z->nr_reserved_highatomic; |
2775 | else | |
1da177e4 | 2776 | min -= min / 4; |
e2b19197 | 2777 | |
d95ea5d1 BZ |
2778 | #ifdef CONFIG_CMA |
2779 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2780 | if (!(alloc_flags & ALLOC_CMA)) | |
97a16fc8 | 2781 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2782 | #endif |
026b0814 | 2783 | |
97a16fc8 MG |
2784 | /* |
2785 | * Check watermarks for an order-0 allocation request. If these | |
2786 | * are not met, then a high-order request also cannot go ahead | |
2787 | * even if a suitable page happened to be free. | |
2788 | */ | |
2789 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 2790 | return false; |
1da177e4 | 2791 | |
97a16fc8 MG |
2792 | /* If this is an order-0 request then the watermark is fine */ |
2793 | if (!order) | |
2794 | return true; | |
2795 | ||
2796 | /* For a high-order request, check at least one suitable page is free */ | |
2797 | for (o = order; o < MAX_ORDER; o++) { | |
2798 | struct free_area *area = &z->free_area[o]; | |
2799 | int mt; | |
2800 | ||
2801 | if (!area->nr_free) | |
2802 | continue; | |
2803 | ||
2804 | if (alloc_harder) | |
2805 | return true; | |
1da177e4 | 2806 | |
97a16fc8 MG |
2807 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
2808 | if (!list_empty(&area->free_list[mt])) | |
2809 | return true; | |
2810 | } | |
2811 | ||
2812 | #ifdef CONFIG_CMA | |
2813 | if ((alloc_flags & ALLOC_CMA) && | |
2814 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
2815 | return true; | |
2816 | } | |
2817 | #endif | |
1da177e4 | 2818 | } |
97a16fc8 | 2819 | return false; |
88f5acf8 MG |
2820 | } |
2821 | ||
7aeb09f9 | 2822 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 2823 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
2824 | { |
2825 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2826 | zone_page_state(z, NR_FREE_PAGES)); | |
2827 | } | |
2828 | ||
48ee5f36 MG |
2829 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
2830 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
2831 | { | |
2832 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2833 | long cma_pages = 0; | |
2834 | ||
2835 | #ifdef CONFIG_CMA | |
2836 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2837 | if (!(alloc_flags & ALLOC_CMA)) | |
2838 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
2839 | #endif | |
2840 | ||
2841 | /* | |
2842 | * Fast check for order-0 only. If this fails then the reserves | |
2843 | * need to be calculated. There is a corner case where the check | |
2844 | * passes but only the high-order atomic reserve are free. If | |
2845 | * the caller is !atomic then it'll uselessly search the free | |
2846 | * list. That corner case is then slower but it is harmless. | |
2847 | */ | |
2848 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) | |
2849 | return true; | |
2850 | ||
2851 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2852 | free_pages); | |
2853 | } | |
2854 | ||
7aeb09f9 | 2855 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 2856 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
2857 | { |
2858 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2859 | ||
2860 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2861 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2862 | ||
e2b19197 | 2863 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 2864 | free_pages); |
1da177e4 LT |
2865 | } |
2866 | ||
9276b1bc | 2867 | #ifdef CONFIG_NUMA |
81c0a2bb JW |
2868 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2869 | { | |
fff4068c | 2870 | return local_zone->node == zone->node; |
81c0a2bb JW |
2871 | } |
2872 | ||
957f822a DR |
2873 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2874 | { | |
5f7a75ac MG |
2875 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
2876 | RECLAIM_DISTANCE; | |
957f822a | 2877 | } |
9276b1bc | 2878 | #else /* CONFIG_NUMA */ |
81c0a2bb JW |
2879 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2880 | { | |
2881 | return true; | |
2882 | } | |
2883 | ||
957f822a DR |
2884 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2885 | { | |
2886 | return true; | |
2887 | } | |
9276b1bc PJ |
2888 | #endif /* CONFIG_NUMA */ |
2889 | ||
4ffeaf35 MG |
2890 | static void reset_alloc_batches(struct zone *preferred_zone) |
2891 | { | |
2892 | struct zone *zone = preferred_zone->zone_pgdat->node_zones; | |
2893 | ||
2894 | do { | |
2895 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
2896 | high_wmark_pages(zone) - low_wmark_pages(zone) - | |
2897 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
57054651 | 2898 | clear_bit(ZONE_FAIR_DEPLETED, &zone->flags); |
4ffeaf35 MG |
2899 | } while (zone++ != preferred_zone); |
2900 | } | |
2901 | ||
7fb1d9fc | 2902 | /* |
0798e519 | 2903 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
2904 | * a page. |
2905 | */ | |
2906 | static struct page * | |
a9263751 VB |
2907 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
2908 | const struct alloc_context *ac) | |
753ee728 | 2909 | { |
c33d6c06 | 2910 | struct zoneref *z = ac->preferred_zoneref; |
5117f45d | 2911 | struct zone *zone; |
30534755 MG |
2912 | bool fair_skipped = false; |
2913 | bool apply_fair = (alloc_flags & ALLOC_FAIR); | |
54a6eb5c | 2914 | |
9276b1bc | 2915 | zonelist_scan: |
7fb1d9fc | 2916 | /* |
9276b1bc | 2917 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 2918 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 2919 | */ |
c33d6c06 | 2920 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 2921 | ac->nodemask) { |
be06af00 | 2922 | struct page *page; |
e085dbc5 JW |
2923 | unsigned long mark; |
2924 | ||
664eedde MG |
2925 | if (cpusets_enabled() && |
2926 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 2927 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 2928 | continue; |
81c0a2bb JW |
2929 | /* |
2930 | * Distribute pages in proportion to the individual | |
2931 | * zone size to ensure fair page aging. The zone a | |
2932 | * page was allocated in should have no effect on the | |
2933 | * time the page has in memory before being reclaimed. | |
81c0a2bb | 2934 | */ |
30534755 | 2935 | if (apply_fair) { |
57054651 | 2936 | if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) { |
fa379b95 | 2937 | fair_skipped = true; |
3a025760 | 2938 | continue; |
4ffeaf35 | 2939 | } |
c33d6c06 | 2940 | if (!zone_local(ac->preferred_zoneref->zone, zone)) { |
30534755 MG |
2941 | if (fair_skipped) |
2942 | goto reset_fair; | |
2943 | apply_fair = false; | |
2944 | } | |
81c0a2bb | 2945 | } |
a756cf59 JW |
2946 | /* |
2947 | * When allocating a page cache page for writing, we | |
2948 | * want to get it from a zone that is within its dirty | |
2949 | * limit, such that no single zone holds more than its | |
2950 | * proportional share of globally allowed dirty pages. | |
2951 | * The dirty limits take into account the zone's | |
2952 | * lowmem reserves and high watermark so that kswapd | |
2953 | * should be able to balance it without having to | |
2954 | * write pages from its LRU list. | |
2955 | * | |
2956 | * This may look like it could increase pressure on | |
2957 | * lower zones by failing allocations in higher zones | |
2958 | * before they are full. But the pages that do spill | |
2959 | * over are limited as the lower zones are protected | |
2960 | * by this very same mechanism. It should not become | |
2961 | * a practical burden to them. | |
2962 | * | |
2963 | * XXX: For now, allow allocations to potentially | |
2964 | * exceed the per-zone dirty limit in the slowpath | |
c9ab0c4f | 2965 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 JW |
2966 | * which is important when on a NUMA setup the allowed |
2967 | * zones are together not big enough to reach the | |
2968 | * global limit. The proper fix for these situations | |
2969 | * will require awareness of zones in the | |
2970 | * dirty-throttling and the flusher threads. | |
2971 | */ | |
c9ab0c4f | 2972 | if (ac->spread_dirty_pages && !zone_dirty_ok(zone)) |
800a1e75 | 2973 | continue; |
7fb1d9fc | 2974 | |
e085dbc5 | 2975 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 2976 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 2977 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
2978 | int ret; |
2979 | ||
5dab2911 MG |
2980 | /* Checked here to keep the fast path fast */ |
2981 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
2982 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
2983 | goto try_this_zone; | |
2984 | ||
957f822a | 2985 | if (zone_reclaim_mode == 0 || |
c33d6c06 | 2986 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
2987 | continue; |
2988 | ||
fa5e084e MG |
2989 | ret = zone_reclaim(zone, gfp_mask, order); |
2990 | switch (ret) { | |
2991 | case ZONE_RECLAIM_NOSCAN: | |
2992 | /* did not scan */ | |
cd38b115 | 2993 | continue; |
fa5e084e MG |
2994 | case ZONE_RECLAIM_FULL: |
2995 | /* scanned but unreclaimable */ | |
cd38b115 | 2996 | continue; |
fa5e084e MG |
2997 | default: |
2998 | /* did we reclaim enough */ | |
fed2719e | 2999 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3000 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3001 | goto try_this_zone; |
3002 | ||
fed2719e | 3003 | continue; |
0798e519 | 3004 | } |
7fb1d9fc RS |
3005 | } |
3006 | ||
fa5e084e | 3007 | try_this_zone: |
c33d6c06 | 3008 | page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3009 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3010 | if (page) { |
479f854a | 3011 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3012 | |
3013 | /* | |
3014 | * If this is a high-order atomic allocation then check | |
3015 | * if the pageblock should be reserved for the future | |
3016 | */ | |
3017 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3018 | reserve_highatomic_pageblock(page, zone, order); | |
3019 | ||
75379191 VB |
3020 | return page; |
3021 | } | |
54a6eb5c | 3022 | } |
9276b1bc | 3023 | |
4ffeaf35 MG |
3024 | /* |
3025 | * The first pass makes sure allocations are spread fairly within the | |
3026 | * local node. However, the local node might have free pages left | |
3027 | * after the fairness batches are exhausted, and remote zones haven't | |
3028 | * even been considered yet. Try once more without fairness, and | |
3029 | * include remote zones now, before entering the slowpath and waking | |
3030 | * kswapd: prefer spilling to a remote zone over swapping locally. | |
3031 | */ | |
30534755 MG |
3032 | if (fair_skipped) { |
3033 | reset_fair: | |
3034 | apply_fair = false; | |
3035 | fair_skipped = false; | |
c33d6c06 | 3036 | reset_alloc_batches(ac->preferred_zoneref->zone); |
4ffeaf35 | 3037 | goto zonelist_scan; |
30534755 | 3038 | } |
4ffeaf35 MG |
3039 | |
3040 | return NULL; | |
753ee728 MH |
3041 | } |
3042 | ||
29423e77 DR |
3043 | /* |
3044 | * Large machines with many possible nodes should not always dump per-node | |
3045 | * meminfo in irq context. | |
3046 | */ | |
3047 | static inline bool should_suppress_show_mem(void) | |
3048 | { | |
3049 | bool ret = false; | |
3050 | ||
3051 | #if NODES_SHIFT > 8 | |
3052 | ret = in_interrupt(); | |
3053 | #endif | |
3054 | return ret; | |
3055 | } | |
3056 | ||
a238ab5b DH |
3057 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
3058 | DEFAULT_RATELIMIT_INTERVAL, | |
3059 | DEFAULT_RATELIMIT_BURST); | |
3060 | ||
d00181b9 | 3061 | void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...) |
a238ab5b | 3062 | { |
a238ab5b DH |
3063 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
3064 | ||
c0a32fc5 SG |
3065 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
3066 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
3067 | return; |
3068 | ||
3069 | /* | |
3070 | * This documents exceptions given to allocations in certain | |
3071 | * contexts that are allowed to allocate outside current's set | |
3072 | * of allowed nodes. | |
3073 | */ | |
3074 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
3075 | if (test_thread_flag(TIF_MEMDIE) || | |
3076 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
3077 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3078 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3079 | filter &= ~SHOW_MEM_FILTER_NODES; |
3080 | ||
3081 | if (fmt) { | |
3ee9a4f0 JP |
3082 | struct va_format vaf; |
3083 | va_list args; | |
3084 | ||
a238ab5b | 3085 | va_start(args, fmt); |
3ee9a4f0 JP |
3086 | |
3087 | vaf.fmt = fmt; | |
3088 | vaf.va = &args; | |
3089 | ||
3090 | pr_warn("%pV", &vaf); | |
3091 | ||
a238ab5b DH |
3092 | va_end(args); |
3093 | } | |
3094 | ||
c5c990e8 VB |
3095 | pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n", |
3096 | current->comm, order, gfp_mask, &gfp_mask); | |
a238ab5b DH |
3097 | dump_stack(); |
3098 | if (!should_suppress_show_mem()) | |
3099 | show_mem(filter); | |
3100 | } | |
3101 | ||
11e33f6a MG |
3102 | static inline struct page * |
3103 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3104 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3105 | { |
6e0fc46d DR |
3106 | struct oom_control oc = { |
3107 | .zonelist = ac->zonelist, | |
3108 | .nodemask = ac->nodemask, | |
3109 | .gfp_mask = gfp_mask, | |
3110 | .order = order, | |
6e0fc46d | 3111 | }; |
11e33f6a MG |
3112 | struct page *page; |
3113 | ||
9879de73 JW |
3114 | *did_some_progress = 0; |
3115 | ||
9879de73 | 3116 | /* |
dc56401f JW |
3117 | * Acquire the oom lock. If that fails, somebody else is |
3118 | * making progress for us. | |
9879de73 | 3119 | */ |
dc56401f | 3120 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3121 | *did_some_progress = 1; |
11e33f6a | 3122 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3123 | return NULL; |
3124 | } | |
6b1de916 | 3125 | |
11e33f6a MG |
3126 | /* |
3127 | * Go through the zonelist yet one more time, keep very high watermark | |
3128 | * here, this is only to catch a parallel oom killing, we must fail if | |
3129 | * we're still under heavy pressure. | |
3130 | */ | |
a9263751 VB |
3131 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
3132 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3133 | if (page) |
11e33f6a MG |
3134 | goto out; |
3135 | ||
4365a567 | 3136 | if (!(gfp_mask & __GFP_NOFAIL)) { |
9879de73 JW |
3137 | /* Coredumps can quickly deplete all memory reserves */ |
3138 | if (current->flags & PF_DUMPCORE) | |
3139 | goto out; | |
4365a567 KH |
3140 | /* The OOM killer will not help higher order allocs */ |
3141 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3142 | goto out; | |
03668b3c | 3143 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
a9263751 | 3144 | if (ac->high_zoneidx < ZONE_NORMAL) |
03668b3c | 3145 | goto out; |
9083905a JW |
3146 | if (pm_suspended_storage()) |
3147 | goto out; | |
3da88fb3 MH |
3148 | /* |
3149 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3150 | * other request to make a forward progress. | |
3151 | * We are in an unfortunate situation where out_of_memory cannot | |
3152 | * do much for this context but let's try it to at least get | |
3153 | * access to memory reserved if the current task is killed (see | |
3154 | * out_of_memory). Once filesystems are ready to handle allocation | |
3155 | * failures more gracefully we should just bail out here. | |
3156 | */ | |
3157 | ||
4167e9b2 | 3158 | /* The OOM killer may not free memory on a specific node */ |
4365a567 KH |
3159 | if (gfp_mask & __GFP_THISNODE) |
3160 | goto out; | |
3161 | } | |
11e33f6a | 3162 | /* Exhausted what can be done so it's blamo time */ |
5020e285 | 3163 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3164 | *did_some_progress = 1; |
5020e285 MH |
3165 | |
3166 | if (gfp_mask & __GFP_NOFAIL) { | |
3167 | page = get_page_from_freelist(gfp_mask, order, | |
3168 | ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac); | |
3169 | /* | |
3170 | * fallback to ignore cpuset restriction if our nodes | |
3171 | * are depleted | |
3172 | */ | |
3173 | if (!page) | |
3174 | page = get_page_from_freelist(gfp_mask, order, | |
3175 | ALLOC_NO_WATERMARKS, ac); | |
3176 | } | |
3177 | } | |
11e33f6a | 3178 | out: |
dc56401f | 3179 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3180 | return page; |
3181 | } | |
3182 | ||
56de7263 MG |
3183 | #ifdef CONFIG_COMPACTION |
3184 | /* Try memory compaction for high-order allocations before reclaim */ | |
3185 | static struct page * | |
3186 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3187 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 VB |
3188 | enum migrate_mode mode, int *contended_compaction, |
3189 | bool *deferred_compaction) | |
56de7263 | 3190 | { |
53853e2d | 3191 | unsigned long compact_result; |
98dd3b48 | 3192 | struct page *page; |
53853e2d VB |
3193 | |
3194 | if (!order) | |
66199712 | 3195 | return NULL; |
66199712 | 3196 | |
c06b1fca | 3197 | current->flags |= PF_MEMALLOC; |
1a6d53a1 VB |
3198 | compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
3199 | mode, contended_compaction); | |
c06b1fca | 3200 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 3201 | |
98dd3b48 VB |
3202 | switch (compact_result) { |
3203 | case COMPACT_DEFERRED: | |
53853e2d | 3204 | *deferred_compaction = true; |
98dd3b48 VB |
3205 | /* fall-through */ |
3206 | case COMPACT_SKIPPED: | |
3207 | return NULL; | |
3208 | default: | |
3209 | break; | |
3210 | } | |
53853e2d | 3211 | |
98dd3b48 VB |
3212 | /* |
3213 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3214 | * count a compaction stall | |
3215 | */ | |
3216 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3217 | |
a9263751 VB |
3218 | page = get_page_from_freelist(gfp_mask, order, |
3219 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
53853e2d | 3220 | |
98dd3b48 VB |
3221 | if (page) { |
3222 | struct zone *zone = page_zone(page); | |
53853e2d | 3223 | |
98dd3b48 VB |
3224 | zone->compact_blockskip_flush = false; |
3225 | compaction_defer_reset(zone, order, true); | |
3226 | count_vm_event(COMPACTSUCCESS); | |
3227 | return page; | |
3228 | } | |
56de7263 | 3229 | |
98dd3b48 VB |
3230 | /* |
3231 | * It's bad if compaction run occurs and fails. The most likely reason | |
3232 | * is that pages exist, but not enough to satisfy watermarks. | |
3233 | */ | |
3234 | count_vm_event(COMPACTFAIL); | |
66199712 | 3235 | |
98dd3b48 | 3236 | cond_resched(); |
56de7263 MG |
3237 | |
3238 | return NULL; | |
3239 | } | |
3240 | #else | |
3241 | static inline struct page * | |
3242 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3243 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 VB |
3244 | enum migrate_mode mode, int *contended_compaction, |
3245 | bool *deferred_compaction) | |
56de7263 MG |
3246 | { |
3247 | return NULL; | |
3248 | } | |
3249 | #endif /* CONFIG_COMPACTION */ | |
3250 | ||
bba90710 MS |
3251 | /* Perform direct synchronous page reclaim */ |
3252 | static int | |
a9263751 VB |
3253 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3254 | const struct alloc_context *ac) | |
11e33f6a | 3255 | { |
11e33f6a | 3256 | struct reclaim_state reclaim_state; |
bba90710 | 3257 | int progress; |
11e33f6a MG |
3258 | |
3259 | cond_resched(); | |
3260 | ||
3261 | /* We now go into synchronous reclaim */ | |
3262 | cpuset_memory_pressure_bump(); | |
c06b1fca | 3263 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
3264 | lockdep_set_current_reclaim_state(gfp_mask); |
3265 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 3266 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3267 | |
a9263751 VB |
3268 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3269 | ac->nodemask); | |
11e33f6a | 3270 | |
c06b1fca | 3271 | current->reclaim_state = NULL; |
11e33f6a | 3272 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 3273 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
3274 | |
3275 | cond_resched(); | |
3276 | ||
bba90710 MS |
3277 | return progress; |
3278 | } | |
3279 | ||
3280 | /* The really slow allocator path where we enter direct reclaim */ | |
3281 | static inline struct page * | |
3282 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3283 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3284 | unsigned long *did_some_progress) |
bba90710 MS |
3285 | { |
3286 | struct page *page = NULL; | |
3287 | bool drained = false; | |
3288 | ||
a9263751 | 3289 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3290 | if (unlikely(!(*did_some_progress))) |
3291 | return NULL; | |
11e33f6a | 3292 | |
9ee493ce | 3293 | retry: |
a9263751 VB |
3294 | page = get_page_from_freelist(gfp_mask, order, |
3295 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
9ee493ce MG |
3296 | |
3297 | /* | |
3298 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3299 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3300 | * Shrink them them and try again | |
9ee493ce MG |
3301 | */ |
3302 | if (!page && !drained) { | |
0aaa29a5 | 3303 | unreserve_highatomic_pageblock(ac); |
93481ff0 | 3304 | drain_all_pages(NULL); |
9ee493ce MG |
3305 | drained = true; |
3306 | goto retry; | |
3307 | } | |
3308 | ||
11e33f6a MG |
3309 | return page; |
3310 | } | |
3311 | ||
a9263751 | 3312 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
3313 | { |
3314 | struct zoneref *z; | |
3315 | struct zone *zone; | |
3316 | ||
a9263751 VB |
3317 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3318 | ac->high_zoneidx, ac->nodemask) | |
93ea9964 | 3319 | wakeup_kswapd(zone, order, ac_classzone_idx(ac)); |
3a025760 JW |
3320 | } |
3321 | ||
c603844b | 3322 | static inline unsigned int |
341ce06f PZ |
3323 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3324 | { | |
c603844b | 3325 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3326 | |
a56f57ff | 3327 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3328 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3329 | |
341ce06f PZ |
3330 | /* |
3331 | * The caller may dip into page reserves a bit more if the caller | |
3332 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3333 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3334 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3335 | */ |
e6223a3b | 3336 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3337 | |
d0164adc | 3338 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3339 | /* |
b104a35d DR |
3340 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3341 | * if it can't schedule. | |
5c3240d9 | 3342 | */ |
b104a35d | 3343 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3344 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3345 | /* |
b104a35d | 3346 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3347 | * comment for __cpuset_node_allowed(). |
523b9458 | 3348 | */ |
341ce06f | 3349 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3350 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3351 | alloc_flags |= ALLOC_HARDER; |
3352 | ||
b37f1dd0 MG |
3353 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
3354 | if (gfp_mask & __GFP_MEMALLOC) | |
3355 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
3356 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
3357 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
3358 | else if (!in_interrupt() && | |
3359 | ((current->flags & PF_MEMALLOC) || | |
3360 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 3361 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 3362 | } |
d95ea5d1 | 3363 | #ifdef CONFIG_CMA |
43e7a34d | 3364 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
3365 | alloc_flags |= ALLOC_CMA; |
3366 | #endif | |
341ce06f PZ |
3367 | return alloc_flags; |
3368 | } | |
3369 | ||
072bb0aa MG |
3370 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3371 | { | |
b37f1dd0 | 3372 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
3373 | } |
3374 | ||
d0164adc MG |
3375 | static inline bool is_thp_gfp_mask(gfp_t gfp_mask) |
3376 | { | |
3377 | return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE; | |
3378 | } | |
3379 | ||
11e33f6a MG |
3380 | static inline struct page * |
3381 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3382 | struct alloc_context *ac) |
11e33f6a | 3383 | { |
d0164adc | 3384 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
11e33f6a | 3385 | struct page *page = NULL; |
c603844b | 3386 | unsigned int alloc_flags; |
11e33f6a MG |
3387 | unsigned long pages_reclaimed = 0; |
3388 | unsigned long did_some_progress; | |
e0b9daeb | 3389 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
66199712 | 3390 | bool deferred_compaction = false; |
1f9efdef | 3391 | int contended_compaction = COMPACT_CONTENDED_NONE; |
1da177e4 | 3392 | |
72807a74 MG |
3393 | /* |
3394 | * In the slowpath, we sanity check order to avoid ever trying to | |
3395 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
3396 | * be using allocators in order of preference for an area that is | |
3397 | * too large. | |
3398 | */ | |
1fc28b70 MG |
3399 | if (order >= MAX_ORDER) { |
3400 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 3401 | return NULL; |
1fc28b70 | 3402 | } |
1da177e4 | 3403 | |
d0164adc MG |
3404 | /* |
3405 | * We also sanity check to catch abuse of atomic reserves being used by | |
3406 | * callers that are not in atomic context. | |
3407 | */ | |
3408 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
3409 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
3410 | gfp_mask &= ~__GFP_ATOMIC; | |
3411 | ||
9879de73 | 3412 | retry: |
d0164adc | 3413 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
a9263751 | 3414 | wake_all_kswapds(order, ac); |
1da177e4 | 3415 | |
9bf2229f | 3416 | /* |
7fb1d9fc RS |
3417 | * OK, we're below the kswapd watermark and have kicked background |
3418 | * reclaim. Now things get more complex, so set up alloc_flags according | |
3419 | * to how we want to proceed. | |
9bf2229f | 3420 | */ |
341ce06f | 3421 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 3422 | |
341ce06f | 3423 | /* This is the last chance, in general, before the goto nopage. */ |
a9263751 VB |
3424 | page = get_page_from_freelist(gfp_mask, order, |
3425 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
7fb1d9fc RS |
3426 | if (page) |
3427 | goto got_pg; | |
1da177e4 | 3428 | |
11e33f6a | 3429 | /* Allocate without watermarks if the context allows */ |
341ce06f | 3430 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
3431 | /* |
3432 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
3433 | * the allocation is high priority and these type of | |
3434 | * allocations are system rather than user orientated | |
3435 | */ | |
a9263751 | 3436 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
33d53103 MH |
3437 | page = get_page_from_freelist(gfp_mask, order, |
3438 | ALLOC_NO_WATERMARKS, ac); | |
3439 | if (page) | |
3440 | goto got_pg; | |
1da177e4 LT |
3441 | } |
3442 | ||
d0164adc MG |
3443 | /* Caller is not willing to reclaim, we can't balance anything */ |
3444 | if (!can_direct_reclaim) { | |
aed0a0e3 | 3445 | /* |
33d53103 MH |
3446 | * All existing users of the __GFP_NOFAIL are blockable, so warn |
3447 | * of any new users that actually allow this type of allocation | |
3448 | * to fail. | |
aed0a0e3 DR |
3449 | */ |
3450 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 3451 | goto nopage; |
aed0a0e3 | 3452 | } |
1da177e4 | 3453 | |
341ce06f | 3454 | /* Avoid recursion of direct reclaim */ |
33d53103 MH |
3455 | if (current->flags & PF_MEMALLOC) { |
3456 | /* | |
3457 | * __GFP_NOFAIL request from this context is rather bizarre | |
3458 | * because we cannot reclaim anything and only can loop waiting | |
3459 | * for somebody to do a work for us. | |
3460 | */ | |
3461 | if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { | |
3462 | cond_resched(); | |
3463 | goto retry; | |
3464 | } | |
341ce06f | 3465 | goto nopage; |
33d53103 | 3466 | } |
341ce06f | 3467 | |
6583bb64 DR |
3468 | /* Avoid allocations with no watermarks from looping endlessly */ |
3469 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
3470 | goto nopage; | |
3471 | ||
77f1fe6b MG |
3472 | /* |
3473 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
3474 | * attempts after direct reclaim are synchronous | |
3475 | */ | |
a9263751 VB |
3476 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
3477 | migration_mode, | |
3478 | &contended_compaction, | |
53853e2d | 3479 | &deferred_compaction); |
56de7263 MG |
3480 | if (page) |
3481 | goto got_pg; | |
75f30861 | 3482 | |
1f9efdef | 3483 | /* Checks for THP-specific high-order allocations */ |
d0164adc | 3484 | if (is_thp_gfp_mask(gfp_mask)) { |
1f9efdef VB |
3485 | /* |
3486 | * If compaction is deferred for high-order allocations, it is | |
3487 | * because sync compaction recently failed. If this is the case | |
3488 | * and the caller requested a THP allocation, we do not want | |
3489 | * to heavily disrupt the system, so we fail the allocation | |
3490 | * instead of entering direct reclaim. | |
3491 | */ | |
3492 | if (deferred_compaction) | |
3493 | goto nopage; | |
3494 | ||
3495 | /* | |
3496 | * In all zones where compaction was attempted (and not | |
3497 | * deferred or skipped), lock contention has been detected. | |
3498 | * For THP allocation we do not want to disrupt the others | |
3499 | * so we fallback to base pages instead. | |
3500 | */ | |
3501 | if (contended_compaction == COMPACT_CONTENDED_LOCK) | |
3502 | goto nopage; | |
3503 | ||
3504 | /* | |
3505 | * If compaction was aborted due to need_resched(), we do not | |
3506 | * want to further increase allocation latency, unless it is | |
3507 | * khugepaged trying to collapse. | |
3508 | */ | |
3509 | if (contended_compaction == COMPACT_CONTENDED_SCHED | |
3510 | && !(current->flags & PF_KTHREAD)) | |
3511 | goto nopage; | |
3512 | } | |
66199712 | 3513 | |
8fe78048 DR |
3514 | /* |
3515 | * It can become very expensive to allocate transparent hugepages at | |
3516 | * fault, so use asynchronous memory compaction for THP unless it is | |
3517 | * khugepaged trying to collapse. | |
3518 | */ | |
d0164adc | 3519 | if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD)) |
8fe78048 DR |
3520 | migration_mode = MIGRATE_SYNC_LIGHT; |
3521 | ||
11e33f6a | 3522 | /* Try direct reclaim and then allocating */ |
a9263751 VB |
3523 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, |
3524 | &did_some_progress); | |
11e33f6a MG |
3525 | if (page) |
3526 | goto got_pg; | |
1da177e4 | 3527 | |
9083905a JW |
3528 | /* Do not loop if specifically requested */ |
3529 | if (gfp_mask & __GFP_NORETRY) | |
3530 | goto noretry; | |
3531 | ||
3532 | /* Keep reclaiming pages as long as there is reasonable progress */ | |
a41f24ea | 3533 | pages_reclaimed += did_some_progress; |
9083905a JW |
3534 | if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) || |
3535 | ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) { | |
11e33f6a | 3536 | /* Wait for some write requests to complete then retry */ |
c33d6c06 | 3537 | wait_iff_congested(ac->preferred_zoneref->zone, BLK_RW_ASYNC, HZ/50); |
9879de73 | 3538 | goto retry; |
1da177e4 LT |
3539 | } |
3540 | ||
9083905a JW |
3541 | /* Reclaim has failed us, start killing things */ |
3542 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
3543 | if (page) | |
3544 | goto got_pg; | |
3545 | ||
3546 | /* Retry as long as the OOM killer is making progress */ | |
3547 | if (did_some_progress) | |
3548 | goto retry; | |
3549 | ||
3550 | noretry: | |
3551 | /* | |
3552 | * High-order allocations do not necessarily loop after | |
3553 | * direct reclaim and reclaim/compaction depends on compaction | |
3554 | * being called after reclaim so call directly if necessary | |
3555 | */ | |
3556 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, | |
3557 | ac, migration_mode, | |
3558 | &contended_compaction, | |
3559 | &deferred_compaction); | |
3560 | if (page) | |
3561 | goto got_pg; | |
1da177e4 | 3562 | nopage: |
a238ab5b | 3563 | warn_alloc_failed(gfp_mask, order, NULL); |
1da177e4 | 3564 | got_pg: |
072bb0aa | 3565 | return page; |
1da177e4 | 3566 | } |
11e33f6a MG |
3567 | |
3568 | /* | |
3569 | * This is the 'heart' of the zoned buddy allocator. | |
3570 | */ | |
3571 | struct page * | |
3572 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
3573 | struct zonelist *zonelist, nodemask_t *nodemask) | |
3574 | { | |
5bb1b169 | 3575 | struct page *page; |
cc9a6c87 | 3576 | unsigned int cpuset_mems_cookie; |
c603844b | 3577 | unsigned int alloc_flags = ALLOC_WMARK_LOW|ALLOC_FAIR; |
83d4ca81 | 3578 | gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */ |
a9263751 VB |
3579 | struct alloc_context ac = { |
3580 | .high_zoneidx = gfp_zone(gfp_mask), | |
682a3385 | 3581 | .zonelist = zonelist, |
a9263751 VB |
3582 | .nodemask = nodemask, |
3583 | .migratetype = gfpflags_to_migratetype(gfp_mask), | |
3584 | }; | |
11e33f6a | 3585 | |
682a3385 | 3586 | if (cpusets_enabled()) { |
83d4ca81 | 3587 | alloc_mask |= __GFP_HARDWALL; |
682a3385 MG |
3588 | alloc_flags |= ALLOC_CPUSET; |
3589 | if (!ac.nodemask) | |
3590 | ac.nodemask = &cpuset_current_mems_allowed; | |
3591 | } | |
3592 | ||
dcce284a BH |
3593 | gfp_mask &= gfp_allowed_mask; |
3594 | ||
11e33f6a MG |
3595 | lockdep_trace_alloc(gfp_mask); |
3596 | ||
d0164adc | 3597 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
3598 | |
3599 | if (should_fail_alloc_page(gfp_mask, order)) | |
3600 | return NULL; | |
3601 | ||
3602 | /* | |
3603 | * Check the zones suitable for the gfp_mask contain at least one | |
3604 | * valid zone. It's possible to have an empty zonelist as a result | |
4167e9b2 | 3605 | * of __GFP_THISNODE and a memoryless node |
11e33f6a MG |
3606 | */ |
3607 | if (unlikely(!zonelist->_zonerefs->zone)) | |
3608 | return NULL; | |
3609 | ||
a9263751 | 3610 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
21bb9bd1 VB |
3611 | alloc_flags |= ALLOC_CMA; |
3612 | ||
cc9a6c87 | 3613 | retry_cpuset: |
d26914d1 | 3614 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3615 | |
c9ab0c4f MG |
3616 | /* Dirty zone balancing only done in the fast path */ |
3617 | ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE); | |
3618 | ||
5117f45d | 3619 | /* The preferred zone is used for statistics later */ |
c33d6c06 MG |
3620 | ac.preferred_zoneref = first_zones_zonelist(ac.zonelist, |
3621 | ac.high_zoneidx, ac.nodemask); | |
3622 | if (!ac.preferred_zoneref) { | |
5bb1b169 | 3623 | page = NULL; |
4fcb0971 | 3624 | goto no_zone; |
5bb1b169 MG |
3625 | } |
3626 | ||
5117f45d | 3627 | /* First allocation attempt */ |
a9263751 | 3628 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
3629 | if (likely(page)) |
3630 | goto out; | |
11e33f6a | 3631 | |
4fcb0971 MG |
3632 | /* |
3633 | * Runtime PM, block IO and its error handling path can deadlock | |
3634 | * because I/O on the device might not complete. | |
3635 | */ | |
3636 | alloc_mask = memalloc_noio_flags(gfp_mask); | |
3637 | ac.spread_dirty_pages = false; | |
23f086f9 | 3638 | |
4741526b MG |
3639 | /* |
3640 | * Restore the original nodemask if it was potentially replaced with | |
3641 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
3642 | */ | |
3643 | if (cpusets_enabled()) | |
3644 | ac.nodemask = nodemask; | |
4fcb0971 | 3645 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 3646 | |
4fcb0971 | 3647 | no_zone: |
cc9a6c87 MG |
3648 | /* |
3649 | * When updating a task's mems_allowed, it is possible to race with | |
3650 | * parallel threads in such a way that an allocation can fail while | |
3651 | * the mask is being updated. If a page allocation is about to fail, | |
3652 | * check if the cpuset changed during allocation and if so, retry. | |
3653 | */ | |
83d4ca81 MG |
3654 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) { |
3655 | alloc_mask = gfp_mask; | |
cc9a6c87 | 3656 | goto retry_cpuset; |
83d4ca81 | 3657 | } |
cc9a6c87 | 3658 | |
4fcb0971 MG |
3659 | out: |
3660 | if (kmemcheck_enabled && page) | |
3661 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
3662 | ||
3663 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); | |
3664 | ||
11e33f6a | 3665 | return page; |
1da177e4 | 3666 | } |
d239171e | 3667 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
3668 | |
3669 | /* | |
3670 | * Common helper functions. | |
3671 | */ | |
920c7a5d | 3672 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 3673 | { |
945a1113 AM |
3674 | struct page *page; |
3675 | ||
3676 | /* | |
3677 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
3678 | * a highmem page | |
3679 | */ | |
3680 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
3681 | ||
1da177e4 LT |
3682 | page = alloc_pages(gfp_mask, order); |
3683 | if (!page) | |
3684 | return 0; | |
3685 | return (unsigned long) page_address(page); | |
3686 | } | |
1da177e4 LT |
3687 | EXPORT_SYMBOL(__get_free_pages); |
3688 | ||
920c7a5d | 3689 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 3690 | { |
945a1113 | 3691 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 3692 | } |
1da177e4 LT |
3693 | EXPORT_SYMBOL(get_zeroed_page); |
3694 | ||
920c7a5d | 3695 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 3696 | { |
b5810039 | 3697 | if (put_page_testzero(page)) { |
1da177e4 | 3698 | if (order == 0) |
b745bc85 | 3699 | free_hot_cold_page(page, false); |
1da177e4 LT |
3700 | else |
3701 | __free_pages_ok(page, order); | |
3702 | } | |
3703 | } | |
3704 | ||
3705 | EXPORT_SYMBOL(__free_pages); | |
3706 | ||
920c7a5d | 3707 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
3708 | { |
3709 | if (addr != 0) { | |
725d704e | 3710 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
3711 | __free_pages(virt_to_page((void *)addr), order); |
3712 | } | |
3713 | } | |
3714 | ||
3715 | EXPORT_SYMBOL(free_pages); | |
3716 | ||
b63ae8ca AD |
3717 | /* |
3718 | * Page Fragment: | |
3719 | * An arbitrary-length arbitrary-offset area of memory which resides | |
3720 | * within a 0 or higher order page. Multiple fragments within that page | |
3721 | * are individually refcounted, in the page's reference counter. | |
3722 | * | |
3723 | * The page_frag functions below provide a simple allocation framework for | |
3724 | * page fragments. This is used by the network stack and network device | |
3725 | * drivers to provide a backing region of memory for use as either an | |
3726 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
3727 | */ | |
3728 | static struct page *__page_frag_refill(struct page_frag_cache *nc, | |
3729 | gfp_t gfp_mask) | |
3730 | { | |
3731 | struct page *page = NULL; | |
3732 | gfp_t gfp = gfp_mask; | |
3733 | ||
3734 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3735 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
3736 | __GFP_NOMEMALLOC; | |
3737 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
3738 | PAGE_FRAG_CACHE_MAX_ORDER); | |
3739 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
3740 | #endif | |
3741 | if (unlikely(!page)) | |
3742 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
3743 | ||
3744 | nc->va = page ? page_address(page) : NULL; | |
3745 | ||
3746 | return page; | |
3747 | } | |
3748 | ||
3749 | void *__alloc_page_frag(struct page_frag_cache *nc, | |
3750 | unsigned int fragsz, gfp_t gfp_mask) | |
3751 | { | |
3752 | unsigned int size = PAGE_SIZE; | |
3753 | struct page *page; | |
3754 | int offset; | |
3755 | ||
3756 | if (unlikely(!nc->va)) { | |
3757 | refill: | |
3758 | page = __page_frag_refill(nc, gfp_mask); | |
3759 | if (!page) | |
3760 | return NULL; | |
3761 | ||
3762 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3763 | /* if size can vary use size else just use PAGE_SIZE */ | |
3764 | size = nc->size; | |
3765 | #endif | |
3766 | /* Even if we own the page, we do not use atomic_set(). | |
3767 | * This would break get_page_unless_zero() users. | |
3768 | */ | |
fe896d18 | 3769 | page_ref_add(page, size - 1); |
b63ae8ca AD |
3770 | |
3771 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 3772 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
3773 | nc->pagecnt_bias = size; |
3774 | nc->offset = size; | |
3775 | } | |
3776 | ||
3777 | offset = nc->offset - fragsz; | |
3778 | if (unlikely(offset < 0)) { | |
3779 | page = virt_to_page(nc->va); | |
3780 | ||
fe896d18 | 3781 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
3782 | goto refill; |
3783 | ||
3784 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3785 | /* if size can vary use size else just use PAGE_SIZE */ | |
3786 | size = nc->size; | |
3787 | #endif | |
3788 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 3789 | set_page_count(page, size); |
b63ae8ca AD |
3790 | |
3791 | /* reset page count bias and offset to start of new frag */ | |
3792 | nc->pagecnt_bias = size; | |
3793 | offset = size - fragsz; | |
3794 | } | |
3795 | ||
3796 | nc->pagecnt_bias--; | |
3797 | nc->offset = offset; | |
3798 | ||
3799 | return nc->va + offset; | |
3800 | } | |
3801 | EXPORT_SYMBOL(__alloc_page_frag); | |
3802 | ||
3803 | /* | |
3804 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
3805 | */ | |
3806 | void __free_page_frag(void *addr) | |
3807 | { | |
3808 | struct page *page = virt_to_head_page(addr); | |
3809 | ||
3810 | if (unlikely(put_page_testzero(page))) | |
3811 | __free_pages_ok(page, compound_order(page)); | |
3812 | } | |
3813 | EXPORT_SYMBOL(__free_page_frag); | |
3814 | ||
6a1a0d3b | 3815 | /* |
52383431 | 3816 | * alloc_kmem_pages charges newly allocated pages to the kmem resource counter |
a9bb7e62 VD |
3817 | * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is |
3818 | * equivalent to alloc_pages. | |
6a1a0d3b | 3819 | * |
52383431 VD |
3820 | * It should be used when the caller would like to use kmalloc, but since the |
3821 | * allocation is large, it has to fall back to the page allocator. | |
3822 | */ | |
3823 | struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order) | |
3824 | { | |
3825 | struct page *page; | |
52383431 | 3826 | |
52383431 | 3827 | page = alloc_pages(gfp_mask, order); |
d05e83a6 VD |
3828 | if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) { |
3829 | __free_pages(page, order); | |
3830 | page = NULL; | |
3831 | } | |
52383431 VD |
3832 | return page; |
3833 | } | |
3834 | ||
3835 | struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
3836 | { | |
3837 | struct page *page; | |
52383431 | 3838 | |
52383431 | 3839 | page = alloc_pages_node(nid, gfp_mask, order); |
d05e83a6 VD |
3840 | if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) { |
3841 | __free_pages(page, order); | |
3842 | page = NULL; | |
3843 | } | |
52383431 VD |
3844 | return page; |
3845 | } | |
3846 | ||
3847 | /* | |
3848 | * __free_kmem_pages and free_kmem_pages will free pages allocated with | |
3849 | * alloc_kmem_pages. | |
6a1a0d3b | 3850 | */ |
52383431 | 3851 | void __free_kmem_pages(struct page *page, unsigned int order) |
6a1a0d3b | 3852 | { |
d05e83a6 | 3853 | memcg_kmem_uncharge(page, order); |
6a1a0d3b GC |
3854 | __free_pages(page, order); |
3855 | } | |
3856 | ||
52383431 | 3857 | void free_kmem_pages(unsigned long addr, unsigned int order) |
6a1a0d3b GC |
3858 | { |
3859 | if (addr != 0) { | |
3860 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
52383431 | 3861 | __free_kmem_pages(virt_to_page((void *)addr), order); |
6a1a0d3b GC |
3862 | } |
3863 | } | |
3864 | ||
d00181b9 KS |
3865 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
3866 | size_t size) | |
ee85c2e1 AK |
3867 | { |
3868 | if (addr) { | |
3869 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
3870 | unsigned long used = addr + PAGE_ALIGN(size); | |
3871 | ||
3872 | split_page(virt_to_page((void *)addr), order); | |
3873 | while (used < alloc_end) { | |
3874 | free_page(used); | |
3875 | used += PAGE_SIZE; | |
3876 | } | |
3877 | } | |
3878 | return (void *)addr; | |
3879 | } | |
3880 | ||
2be0ffe2 TT |
3881 | /** |
3882 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
3883 | * @size: the number of bytes to allocate | |
3884 | * @gfp_mask: GFP flags for the allocation | |
3885 | * | |
3886 | * This function is similar to alloc_pages(), except that it allocates the | |
3887 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
3888 | * allocate memory in power-of-two pages. | |
3889 | * | |
3890 | * This function is also limited by MAX_ORDER. | |
3891 | * | |
3892 | * Memory allocated by this function must be released by free_pages_exact(). | |
3893 | */ | |
3894 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
3895 | { | |
3896 | unsigned int order = get_order(size); | |
3897 | unsigned long addr; | |
3898 | ||
3899 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 3900 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
3901 | } |
3902 | EXPORT_SYMBOL(alloc_pages_exact); | |
3903 | ||
ee85c2e1 AK |
3904 | /** |
3905 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
3906 | * pages on a node. | |
b5e6ab58 | 3907 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
3908 | * @size: the number of bytes to allocate |
3909 | * @gfp_mask: GFP flags for the allocation | |
3910 | * | |
3911 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
3912 | * back. | |
ee85c2e1 | 3913 | */ |
e1931811 | 3914 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 3915 | { |
d00181b9 | 3916 | unsigned int order = get_order(size); |
ee85c2e1 AK |
3917 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
3918 | if (!p) | |
3919 | return NULL; | |
3920 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
3921 | } | |
ee85c2e1 | 3922 | |
2be0ffe2 TT |
3923 | /** |
3924 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
3925 | * @virt: the value returned by alloc_pages_exact. | |
3926 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
3927 | * | |
3928 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
3929 | */ | |
3930 | void free_pages_exact(void *virt, size_t size) | |
3931 | { | |
3932 | unsigned long addr = (unsigned long)virt; | |
3933 | unsigned long end = addr + PAGE_ALIGN(size); | |
3934 | ||
3935 | while (addr < end) { | |
3936 | free_page(addr); | |
3937 | addr += PAGE_SIZE; | |
3938 | } | |
3939 | } | |
3940 | EXPORT_SYMBOL(free_pages_exact); | |
3941 | ||
e0fb5815 ZY |
3942 | /** |
3943 | * nr_free_zone_pages - count number of pages beyond high watermark | |
3944 | * @offset: The zone index of the highest zone | |
3945 | * | |
3946 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
3947 | * high watermark within all zones at or below a given zone index. For each | |
3948 | * zone, the number of pages is calculated as: | |
834405c3 | 3949 | * managed_pages - high_pages |
e0fb5815 | 3950 | */ |
ebec3862 | 3951 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 3952 | { |
dd1a239f | 3953 | struct zoneref *z; |
54a6eb5c MG |
3954 | struct zone *zone; |
3955 | ||
e310fd43 | 3956 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 3957 | unsigned long sum = 0; |
1da177e4 | 3958 | |
0e88460d | 3959 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 3960 | |
54a6eb5c | 3961 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 3962 | unsigned long size = zone->managed_pages; |
41858966 | 3963 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
3964 | if (size > high) |
3965 | sum += size - high; | |
1da177e4 LT |
3966 | } |
3967 | ||
3968 | return sum; | |
3969 | } | |
3970 | ||
e0fb5815 ZY |
3971 | /** |
3972 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
3973 | * | |
3974 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
3975 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 3976 | */ |
ebec3862 | 3977 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 3978 | { |
af4ca457 | 3979 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 3980 | } |
c2f1a551 | 3981 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 3982 | |
e0fb5815 ZY |
3983 | /** |
3984 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
3985 | * | |
3986 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
3987 | * high watermark within all zones. | |
1da177e4 | 3988 | */ |
ebec3862 | 3989 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 3990 | { |
2a1e274a | 3991 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 3992 | } |
08e0f6a9 CL |
3993 | |
3994 | static inline void show_node(struct zone *zone) | |
1da177e4 | 3995 | { |
e5adfffc | 3996 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 3997 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 3998 | } |
1da177e4 | 3999 | |
d02bd27b IR |
4000 | long si_mem_available(void) |
4001 | { | |
4002 | long available; | |
4003 | unsigned long pagecache; | |
4004 | unsigned long wmark_low = 0; | |
4005 | unsigned long pages[NR_LRU_LISTS]; | |
4006 | struct zone *zone; | |
4007 | int lru; | |
4008 | ||
4009 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
4010 | pages[lru] = global_page_state(NR_LRU_BASE + lru); | |
4011 | ||
4012 | for_each_zone(zone) | |
4013 | wmark_low += zone->watermark[WMARK_LOW]; | |
4014 | ||
4015 | /* | |
4016 | * Estimate the amount of memory available for userspace allocations, | |
4017 | * without causing swapping. | |
4018 | */ | |
4019 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; | |
4020 | ||
4021 | /* | |
4022 | * Not all the page cache can be freed, otherwise the system will | |
4023 | * start swapping. Assume at least half of the page cache, or the | |
4024 | * low watermark worth of cache, needs to stay. | |
4025 | */ | |
4026 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
4027 | pagecache -= min(pagecache / 2, wmark_low); | |
4028 | available += pagecache; | |
4029 | ||
4030 | /* | |
4031 | * Part of the reclaimable slab consists of items that are in use, | |
4032 | * and cannot be freed. Cap this estimate at the low watermark. | |
4033 | */ | |
4034 | available += global_page_state(NR_SLAB_RECLAIMABLE) - | |
4035 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); | |
4036 | ||
4037 | if (available < 0) | |
4038 | available = 0; | |
4039 | return available; | |
4040 | } | |
4041 | EXPORT_SYMBOL_GPL(si_mem_available); | |
4042 | ||
1da177e4 LT |
4043 | void si_meminfo(struct sysinfo *val) |
4044 | { | |
4045 | val->totalram = totalram_pages; | |
cc7452b6 | 4046 | val->sharedram = global_page_state(NR_SHMEM); |
d23ad423 | 4047 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 4048 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
4049 | val->totalhigh = totalhigh_pages; |
4050 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
4051 | val->mem_unit = PAGE_SIZE; |
4052 | } | |
4053 | ||
4054 | EXPORT_SYMBOL(si_meminfo); | |
4055 | ||
4056 | #ifdef CONFIG_NUMA | |
4057 | void si_meminfo_node(struct sysinfo *val, int nid) | |
4058 | { | |
cdd91a77 JL |
4059 | int zone_type; /* needs to be signed */ |
4060 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
4061 | unsigned long managed_highpages = 0; |
4062 | unsigned long free_highpages = 0; | |
1da177e4 LT |
4063 | pg_data_t *pgdat = NODE_DATA(nid); |
4064 | ||
cdd91a77 JL |
4065 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
4066 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
4067 | val->totalram = managed_pages; | |
cc7452b6 | 4068 | val->sharedram = node_page_state(nid, NR_SHMEM); |
d23ad423 | 4069 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 4070 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
4071 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4072 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4073 | ||
4074 | if (is_highmem(zone)) { | |
4075 | managed_highpages += zone->managed_pages; | |
4076 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
4077 | } | |
4078 | } | |
4079 | val->totalhigh = managed_highpages; | |
4080 | val->freehigh = free_highpages; | |
98d2b0eb | 4081 | #else |
fc2bd799 JK |
4082 | val->totalhigh = managed_highpages; |
4083 | val->freehigh = free_highpages; | |
98d2b0eb | 4084 | #endif |
1da177e4 LT |
4085 | val->mem_unit = PAGE_SIZE; |
4086 | } | |
4087 | #endif | |
4088 | ||
ddd588b5 | 4089 | /* |
7bf02ea2 DR |
4090 | * Determine whether the node should be displayed or not, depending on whether |
4091 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 4092 | */ |
7bf02ea2 | 4093 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
4094 | { |
4095 | bool ret = false; | |
cc9a6c87 | 4096 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
4097 | |
4098 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
4099 | goto out; | |
4100 | ||
cc9a6c87 | 4101 | do { |
d26914d1 | 4102 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 4103 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 4104 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
4105 | out: |
4106 | return ret; | |
4107 | } | |
4108 | ||
1da177e4 LT |
4109 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4110 | ||
377e4f16 RV |
4111 | static void show_migration_types(unsigned char type) |
4112 | { | |
4113 | static const char types[MIGRATE_TYPES] = { | |
4114 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 4115 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
4116 | [MIGRATE_RECLAIMABLE] = 'E', |
4117 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
4118 | #ifdef CONFIG_CMA |
4119 | [MIGRATE_CMA] = 'C', | |
4120 | #endif | |
194159fb | 4121 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 4122 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 4123 | #endif |
377e4f16 RV |
4124 | }; |
4125 | char tmp[MIGRATE_TYPES + 1]; | |
4126 | char *p = tmp; | |
4127 | int i; | |
4128 | ||
4129 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
4130 | if (type & (1 << i)) | |
4131 | *p++ = types[i]; | |
4132 | } | |
4133 | ||
4134 | *p = '\0'; | |
4135 | printk("(%s) ", tmp); | |
4136 | } | |
4137 | ||
1da177e4 LT |
4138 | /* |
4139 | * Show free area list (used inside shift_scroll-lock stuff) | |
4140 | * We also calculate the percentage fragmentation. We do this by counting the | |
4141 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
4142 | * |
4143 | * Bits in @filter: | |
4144 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
4145 | * cpuset. | |
1da177e4 | 4146 | */ |
7bf02ea2 | 4147 | void show_free_areas(unsigned int filter) |
1da177e4 | 4148 | { |
d1bfcdb8 | 4149 | unsigned long free_pcp = 0; |
c7241913 | 4150 | int cpu; |
1da177e4 LT |
4151 | struct zone *zone; |
4152 | ||
ee99c71c | 4153 | for_each_populated_zone(zone) { |
7bf02ea2 | 4154 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4155 | continue; |
d1bfcdb8 | 4156 | |
761b0677 KK |
4157 | for_each_online_cpu(cpu) |
4158 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
4159 | } |
4160 | ||
a731286d KM |
4161 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4162 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
4163 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4164 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 4165 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 4166 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
4f98a2fe | 4167 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 4168 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
4169 | global_page_state(NR_ISOLATED_ANON), |
4170 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 4171 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 4172 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 4173 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 4174 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 4175 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 4176 | global_page_state(NR_UNSTABLE_NFS), |
3701b033 KM |
4177 | global_page_state(NR_SLAB_RECLAIMABLE), |
4178 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 4179 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 4180 | global_page_state(NR_SHMEM), |
a25700a5 | 4181 | global_page_state(NR_PAGETABLE), |
d1ce749a | 4182 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
4183 | global_page_state(NR_FREE_PAGES), |
4184 | free_pcp, | |
d1ce749a | 4185 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4186 | |
ee99c71c | 4187 | for_each_populated_zone(zone) { |
1da177e4 LT |
4188 | int i; |
4189 | ||
7bf02ea2 | 4190 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4191 | continue; |
d1bfcdb8 KK |
4192 | |
4193 | free_pcp = 0; | |
4194 | for_each_online_cpu(cpu) | |
4195 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
4196 | ||
1da177e4 LT |
4197 | show_node(zone); |
4198 | printk("%s" | |
4199 | " free:%lukB" | |
4200 | " min:%lukB" | |
4201 | " low:%lukB" | |
4202 | " high:%lukB" | |
4f98a2fe RR |
4203 | " active_anon:%lukB" |
4204 | " inactive_anon:%lukB" | |
4205 | " active_file:%lukB" | |
4206 | " inactive_file:%lukB" | |
7b854121 | 4207 | " unevictable:%lukB" |
a731286d KM |
4208 | " isolated(anon):%lukB" |
4209 | " isolated(file):%lukB" | |
1da177e4 | 4210 | " present:%lukB" |
9feedc9d | 4211 | " managed:%lukB" |
4a0aa73f KM |
4212 | " mlocked:%lukB" |
4213 | " dirty:%lukB" | |
4214 | " writeback:%lukB" | |
4215 | " mapped:%lukB" | |
4b02108a | 4216 | " shmem:%lukB" |
4a0aa73f KM |
4217 | " slab_reclaimable:%lukB" |
4218 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 4219 | " kernel_stack:%lukB" |
4a0aa73f KM |
4220 | " pagetables:%lukB" |
4221 | " unstable:%lukB" | |
4222 | " bounce:%lukB" | |
d1bfcdb8 KK |
4223 | " free_pcp:%lukB" |
4224 | " local_pcp:%ukB" | |
d1ce749a | 4225 | " free_cma:%lukB" |
4a0aa73f | 4226 | " writeback_tmp:%lukB" |
1da177e4 LT |
4227 | " pages_scanned:%lu" |
4228 | " all_unreclaimable? %s" | |
4229 | "\n", | |
4230 | zone->name, | |
88f5acf8 | 4231 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
4232 | K(min_wmark_pages(zone)), |
4233 | K(low_wmark_pages(zone)), | |
4234 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
4235 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
4236 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
4237 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
4238 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 4239 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
4240 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
4241 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 4242 | K(zone->present_pages), |
9feedc9d | 4243 | K(zone->managed_pages), |
4a0aa73f KM |
4244 | K(zone_page_state(zone, NR_MLOCK)), |
4245 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
4246 | K(zone_page_state(zone, NR_WRITEBACK)), | |
4247 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 4248 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
4249 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
4250 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
4251 | zone_page_state(zone, NR_KERNEL_STACK) * |
4252 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
4253 | K(zone_page_state(zone, NR_PAGETABLE)), |
4254 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
4255 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1bfcdb8 KK |
4256 | K(free_pcp), |
4257 | K(this_cpu_read(zone->pageset->pcp.count)), | |
d1ce749a | 4258 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 4259 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
0d5d823a | 4260 | K(zone_page_state(zone, NR_PAGES_SCANNED)), |
6e543d57 | 4261 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
4262 | ); |
4263 | printk("lowmem_reserve[]:"); | |
4264 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3484b2de | 4265 | printk(" %ld", zone->lowmem_reserve[i]); |
1da177e4 LT |
4266 | printk("\n"); |
4267 | } | |
4268 | ||
ee99c71c | 4269 | for_each_populated_zone(zone) { |
d00181b9 KS |
4270 | unsigned int order; |
4271 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 4272 | unsigned char types[MAX_ORDER]; |
1da177e4 | 4273 | |
7bf02ea2 | 4274 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4275 | continue; |
1da177e4 LT |
4276 | show_node(zone); |
4277 | printk("%s: ", zone->name); | |
1da177e4 LT |
4278 | |
4279 | spin_lock_irqsave(&zone->lock, flags); | |
4280 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
4281 | struct free_area *area = &zone->free_area[order]; |
4282 | int type; | |
4283 | ||
4284 | nr[order] = area->nr_free; | |
8f9de51a | 4285 | total += nr[order] << order; |
377e4f16 RV |
4286 | |
4287 | types[order] = 0; | |
4288 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
4289 | if (!list_empty(&area->free_list[type])) | |
4290 | types[order] |= 1 << type; | |
4291 | } | |
1da177e4 LT |
4292 | } |
4293 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 4294 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 4295 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
4296 | if (nr[order]) |
4297 | show_migration_types(types[order]); | |
4298 | } | |
1da177e4 LT |
4299 | printk("= %lukB\n", K(total)); |
4300 | } | |
4301 | ||
949f7ec5 DR |
4302 | hugetlb_show_meminfo(); |
4303 | ||
e6f3602d LW |
4304 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
4305 | ||
1da177e4 LT |
4306 | show_swap_cache_info(); |
4307 | } | |
4308 | ||
19770b32 MG |
4309 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4310 | { | |
4311 | zoneref->zone = zone; | |
4312 | zoneref->zone_idx = zone_idx(zone); | |
4313 | } | |
4314 | ||
1da177e4 LT |
4315 | /* |
4316 | * Builds allocation fallback zone lists. | |
1a93205b CL |
4317 | * |
4318 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 4319 | */ |
f0c0b2b8 | 4320 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 4321 | int nr_zones) |
1da177e4 | 4322 | { |
1a93205b | 4323 | struct zone *zone; |
bc732f1d | 4324 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
4325 | |
4326 | do { | |
2f6726e5 | 4327 | zone_type--; |
070f8032 | 4328 | zone = pgdat->node_zones + zone_type; |
1a93205b | 4329 | if (populated_zone(zone)) { |
dd1a239f MG |
4330 | zoneref_set_zone(zone, |
4331 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 4332 | check_highest_zone(zone_type); |
1da177e4 | 4333 | } |
2f6726e5 | 4334 | } while (zone_type); |
bc732f1d | 4335 | |
070f8032 | 4336 | return nr_zones; |
1da177e4 LT |
4337 | } |
4338 | ||
f0c0b2b8 KH |
4339 | |
4340 | /* | |
4341 | * zonelist_order: | |
4342 | * 0 = automatic detection of better ordering. | |
4343 | * 1 = order by ([node] distance, -zonetype) | |
4344 | * 2 = order by (-zonetype, [node] distance) | |
4345 | * | |
4346 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
4347 | * the same zonelist. So only NUMA can configure this param. | |
4348 | */ | |
4349 | #define ZONELIST_ORDER_DEFAULT 0 | |
4350 | #define ZONELIST_ORDER_NODE 1 | |
4351 | #define ZONELIST_ORDER_ZONE 2 | |
4352 | ||
4353 | /* zonelist order in the kernel. | |
4354 | * set_zonelist_order() will set this to NODE or ZONE. | |
4355 | */ | |
4356 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4357 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
4358 | ||
4359 | ||
1da177e4 | 4360 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
4361 | /* The value user specified ....changed by config */ |
4362 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4363 | /* string for sysctl */ | |
4364 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
4365 | char numa_zonelist_order[16] = "default"; | |
4366 | ||
4367 | /* | |
4368 | * interface for configure zonelist ordering. | |
4369 | * command line option "numa_zonelist_order" | |
4370 | * = "[dD]efault - default, automatic configuration. | |
4371 | * = "[nN]ode - order by node locality, then by zone within node | |
4372 | * = "[zZ]one - order by zone, then by locality within zone | |
4373 | */ | |
4374 | ||
4375 | static int __parse_numa_zonelist_order(char *s) | |
4376 | { | |
4377 | if (*s == 'd' || *s == 'D') { | |
4378 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4379 | } else if (*s == 'n' || *s == 'N') { | |
4380 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
4381 | } else if (*s == 'z' || *s == 'Z') { | |
4382 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
4383 | } else { | |
1170532b | 4384 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
f0c0b2b8 KH |
4385 | return -EINVAL; |
4386 | } | |
4387 | return 0; | |
4388 | } | |
4389 | ||
4390 | static __init int setup_numa_zonelist_order(char *s) | |
4391 | { | |
ecb256f8 VL |
4392 | int ret; |
4393 | ||
4394 | if (!s) | |
4395 | return 0; | |
4396 | ||
4397 | ret = __parse_numa_zonelist_order(s); | |
4398 | if (ret == 0) | |
4399 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
4400 | ||
4401 | return ret; | |
f0c0b2b8 KH |
4402 | } |
4403 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
4404 | ||
4405 | /* | |
4406 | * sysctl handler for numa_zonelist_order | |
4407 | */ | |
cccad5b9 | 4408 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 4409 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
4410 | loff_t *ppos) |
4411 | { | |
4412 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
4413 | int ret; | |
443c6f14 | 4414 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 4415 | |
443c6f14 | 4416 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
4417 | if (write) { |
4418 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
4419 | ret = -EINVAL; | |
4420 | goto out; | |
4421 | } | |
4422 | strcpy(saved_string, (char *)table->data); | |
4423 | } | |
8d65af78 | 4424 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 4425 | if (ret) |
443c6f14 | 4426 | goto out; |
f0c0b2b8 KH |
4427 | if (write) { |
4428 | int oldval = user_zonelist_order; | |
dacbde09 CG |
4429 | |
4430 | ret = __parse_numa_zonelist_order((char *)table->data); | |
4431 | if (ret) { | |
f0c0b2b8 KH |
4432 | /* |
4433 | * bogus value. restore saved string | |
4434 | */ | |
dacbde09 | 4435 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
4436 | NUMA_ZONELIST_ORDER_LEN); |
4437 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
4438 | } else if (oldval != user_zonelist_order) { |
4439 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 4440 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
4441 | mutex_unlock(&zonelists_mutex); |
4442 | } | |
f0c0b2b8 | 4443 | } |
443c6f14 AK |
4444 | out: |
4445 | mutex_unlock(&zl_order_mutex); | |
4446 | return ret; | |
f0c0b2b8 KH |
4447 | } |
4448 | ||
4449 | ||
62bc62a8 | 4450 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
4451 | static int node_load[MAX_NUMNODES]; |
4452 | ||
1da177e4 | 4453 | /** |
4dc3b16b | 4454 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
4455 | * @node: node whose fallback list we're appending |
4456 | * @used_node_mask: nodemask_t of already used nodes | |
4457 | * | |
4458 | * We use a number of factors to determine which is the next node that should | |
4459 | * appear on a given node's fallback list. The node should not have appeared | |
4460 | * already in @node's fallback list, and it should be the next closest node | |
4461 | * according to the distance array (which contains arbitrary distance values | |
4462 | * from each node to each node in the system), and should also prefer nodes | |
4463 | * with no CPUs, since presumably they'll have very little allocation pressure | |
4464 | * on them otherwise. | |
4465 | * It returns -1 if no node is found. | |
4466 | */ | |
f0c0b2b8 | 4467 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 4468 | { |
4cf808eb | 4469 | int n, val; |
1da177e4 | 4470 | int min_val = INT_MAX; |
00ef2d2f | 4471 | int best_node = NUMA_NO_NODE; |
a70f7302 | 4472 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 4473 | |
4cf808eb LT |
4474 | /* Use the local node if we haven't already */ |
4475 | if (!node_isset(node, *used_node_mask)) { | |
4476 | node_set(node, *used_node_mask); | |
4477 | return node; | |
4478 | } | |
1da177e4 | 4479 | |
4b0ef1fe | 4480 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
4481 | |
4482 | /* Don't want a node to appear more than once */ | |
4483 | if (node_isset(n, *used_node_mask)) | |
4484 | continue; | |
4485 | ||
1da177e4 LT |
4486 | /* Use the distance array to find the distance */ |
4487 | val = node_distance(node, n); | |
4488 | ||
4cf808eb LT |
4489 | /* Penalize nodes under us ("prefer the next node") */ |
4490 | val += (n < node); | |
4491 | ||
1da177e4 | 4492 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
4493 | tmp = cpumask_of_node(n); |
4494 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
4495 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4496 | ||
4497 | /* Slight preference for less loaded node */ | |
4498 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
4499 | val += node_load[n]; | |
4500 | ||
4501 | if (val < min_val) { | |
4502 | min_val = val; | |
4503 | best_node = n; | |
4504 | } | |
4505 | } | |
4506 | ||
4507 | if (best_node >= 0) | |
4508 | node_set(best_node, *used_node_mask); | |
4509 | ||
4510 | return best_node; | |
4511 | } | |
4512 | ||
f0c0b2b8 KH |
4513 | |
4514 | /* | |
4515 | * Build zonelists ordered by node and zones within node. | |
4516 | * This results in maximum locality--normal zone overflows into local | |
4517 | * DMA zone, if any--but risks exhausting DMA zone. | |
4518 | */ | |
4519 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 4520 | { |
f0c0b2b8 | 4521 | int j; |
1da177e4 | 4522 | struct zonelist *zonelist; |
f0c0b2b8 | 4523 | |
54a6eb5c | 4524 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 4525 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 4526 | ; |
bc732f1d | 4527 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
4528 | zonelist->_zonerefs[j].zone = NULL; |
4529 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
4530 | } |
4531 | ||
523b9458 CL |
4532 | /* |
4533 | * Build gfp_thisnode zonelists | |
4534 | */ | |
4535 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
4536 | { | |
523b9458 CL |
4537 | int j; |
4538 | struct zonelist *zonelist; | |
4539 | ||
54a6eb5c | 4540 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 4541 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
4542 | zonelist->_zonerefs[j].zone = NULL; |
4543 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
4544 | } |
4545 | ||
f0c0b2b8 KH |
4546 | /* |
4547 | * Build zonelists ordered by zone and nodes within zones. | |
4548 | * This results in conserving DMA zone[s] until all Normal memory is | |
4549 | * exhausted, but results in overflowing to remote node while memory | |
4550 | * may still exist in local DMA zone. | |
4551 | */ | |
4552 | static int node_order[MAX_NUMNODES]; | |
4553 | ||
4554 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
4555 | { | |
f0c0b2b8 KH |
4556 | int pos, j, node; |
4557 | int zone_type; /* needs to be signed */ | |
4558 | struct zone *z; | |
4559 | struct zonelist *zonelist; | |
4560 | ||
54a6eb5c MG |
4561 | zonelist = &pgdat->node_zonelists[0]; |
4562 | pos = 0; | |
4563 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
4564 | for (j = 0; j < nr_nodes; j++) { | |
4565 | node = node_order[j]; | |
4566 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
4567 | if (populated_zone(z)) { | |
dd1a239f MG |
4568 | zoneref_set_zone(z, |
4569 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 4570 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
4571 | } |
4572 | } | |
f0c0b2b8 | 4573 | } |
dd1a239f MG |
4574 | zonelist->_zonerefs[pos].zone = NULL; |
4575 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
4576 | } |
4577 | ||
3193913c MG |
4578 | #if defined(CONFIG_64BIT) |
4579 | /* | |
4580 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
4581 | * penalty to every machine in case the specialised case applies. Default | |
4582 | * to Node-ordering on 64-bit NUMA machines | |
4583 | */ | |
4584 | static int default_zonelist_order(void) | |
4585 | { | |
4586 | return ZONELIST_ORDER_NODE; | |
4587 | } | |
4588 | #else | |
4589 | /* | |
4590 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
4591 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
4592 | * then reclaim will occur more frequency increasing stalls and potentially | |
4593 | * be easier to OOM if a large percentage of the zone is under writeback or | |
4594 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
4595 | * Hence, default to zone ordering on 32-bit. | |
4596 | */ | |
f0c0b2b8 KH |
4597 | static int default_zonelist_order(void) |
4598 | { | |
f0c0b2b8 KH |
4599 | return ZONELIST_ORDER_ZONE; |
4600 | } | |
3193913c | 4601 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
4602 | |
4603 | static void set_zonelist_order(void) | |
4604 | { | |
4605 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
4606 | current_zonelist_order = default_zonelist_order(); | |
4607 | else | |
4608 | current_zonelist_order = user_zonelist_order; | |
4609 | } | |
4610 | ||
4611 | static void build_zonelists(pg_data_t *pgdat) | |
4612 | { | |
c00eb15a | 4613 | int i, node, load; |
1da177e4 | 4614 | nodemask_t used_mask; |
f0c0b2b8 KH |
4615 | int local_node, prev_node; |
4616 | struct zonelist *zonelist; | |
d00181b9 | 4617 | unsigned int order = current_zonelist_order; |
1da177e4 LT |
4618 | |
4619 | /* initialize zonelists */ | |
523b9458 | 4620 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 4621 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
4622 | zonelist->_zonerefs[0].zone = NULL; |
4623 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
4624 | } |
4625 | ||
4626 | /* NUMA-aware ordering of nodes */ | |
4627 | local_node = pgdat->node_id; | |
62bc62a8 | 4628 | load = nr_online_nodes; |
1da177e4 LT |
4629 | prev_node = local_node; |
4630 | nodes_clear(used_mask); | |
f0c0b2b8 | 4631 | |
f0c0b2b8 | 4632 | memset(node_order, 0, sizeof(node_order)); |
c00eb15a | 4633 | i = 0; |
f0c0b2b8 | 4634 | |
1da177e4 LT |
4635 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
4636 | /* | |
4637 | * We don't want to pressure a particular node. | |
4638 | * So adding penalty to the first node in same | |
4639 | * distance group to make it round-robin. | |
4640 | */ | |
957f822a DR |
4641 | if (node_distance(local_node, node) != |
4642 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
4643 | node_load[node] = load; |
4644 | ||
1da177e4 LT |
4645 | prev_node = node; |
4646 | load--; | |
f0c0b2b8 KH |
4647 | if (order == ZONELIST_ORDER_NODE) |
4648 | build_zonelists_in_node_order(pgdat, node); | |
4649 | else | |
c00eb15a | 4650 | node_order[i++] = node; /* remember order */ |
f0c0b2b8 | 4651 | } |
1da177e4 | 4652 | |
f0c0b2b8 KH |
4653 | if (order == ZONELIST_ORDER_ZONE) { |
4654 | /* calculate node order -- i.e., DMA last! */ | |
c00eb15a | 4655 | build_zonelists_in_zone_order(pgdat, i); |
1da177e4 | 4656 | } |
523b9458 CL |
4657 | |
4658 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
4659 | } |
4660 | ||
7aac7898 LS |
4661 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4662 | /* | |
4663 | * Return node id of node used for "local" allocations. | |
4664 | * I.e., first node id of first zone in arg node's generic zonelist. | |
4665 | * Used for initializing percpu 'numa_mem', which is used primarily | |
4666 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
4667 | */ | |
4668 | int local_memory_node(int node) | |
4669 | { | |
c33d6c06 | 4670 | struct zoneref *z; |
7aac7898 | 4671 | |
c33d6c06 | 4672 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 4673 | gfp_zone(GFP_KERNEL), |
c33d6c06 MG |
4674 | NULL); |
4675 | return z->zone->node; | |
7aac7898 LS |
4676 | } |
4677 | #endif | |
f0c0b2b8 | 4678 | |
1da177e4 LT |
4679 | #else /* CONFIG_NUMA */ |
4680 | ||
f0c0b2b8 KH |
4681 | static void set_zonelist_order(void) |
4682 | { | |
4683 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
4684 | } | |
4685 | ||
4686 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 4687 | { |
19655d34 | 4688 | int node, local_node; |
54a6eb5c MG |
4689 | enum zone_type j; |
4690 | struct zonelist *zonelist; | |
1da177e4 LT |
4691 | |
4692 | local_node = pgdat->node_id; | |
1da177e4 | 4693 | |
54a6eb5c | 4694 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 4695 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 4696 | |
54a6eb5c MG |
4697 | /* |
4698 | * Now we build the zonelist so that it contains the zones | |
4699 | * of all the other nodes. | |
4700 | * We don't want to pressure a particular node, so when | |
4701 | * building the zones for node N, we make sure that the | |
4702 | * zones coming right after the local ones are those from | |
4703 | * node N+1 (modulo N) | |
4704 | */ | |
4705 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
4706 | if (!node_online(node)) | |
4707 | continue; | |
bc732f1d | 4708 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 4709 | } |
54a6eb5c MG |
4710 | for (node = 0; node < local_node; node++) { |
4711 | if (!node_online(node)) | |
4712 | continue; | |
bc732f1d | 4713 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
4714 | } |
4715 | ||
dd1a239f MG |
4716 | zonelist->_zonerefs[j].zone = NULL; |
4717 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
4718 | } |
4719 | ||
4720 | #endif /* CONFIG_NUMA */ | |
4721 | ||
99dcc3e5 CL |
4722 | /* |
4723 | * Boot pageset table. One per cpu which is going to be used for all | |
4724 | * zones and all nodes. The parameters will be set in such a way | |
4725 | * that an item put on a list will immediately be handed over to | |
4726 | * the buddy list. This is safe since pageset manipulation is done | |
4727 | * with interrupts disabled. | |
4728 | * | |
4729 | * The boot_pagesets must be kept even after bootup is complete for | |
4730 | * unused processors and/or zones. They do play a role for bootstrapping | |
4731 | * hotplugged processors. | |
4732 | * | |
4733 | * zoneinfo_show() and maybe other functions do | |
4734 | * not check if the processor is online before following the pageset pointer. | |
4735 | * Other parts of the kernel may not check if the zone is available. | |
4736 | */ | |
4737 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
4738 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 4739 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 4740 | |
4eaf3f64 HL |
4741 | /* |
4742 | * Global mutex to protect against size modification of zonelists | |
4743 | * as well as to serialize pageset setup for the new populated zone. | |
4744 | */ | |
4745 | DEFINE_MUTEX(zonelists_mutex); | |
4746 | ||
9b1a4d38 | 4747 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 4748 | static int __build_all_zonelists(void *data) |
1da177e4 | 4749 | { |
6811378e | 4750 | int nid; |
99dcc3e5 | 4751 | int cpu; |
9adb62a5 | 4752 | pg_data_t *self = data; |
9276b1bc | 4753 | |
7f9cfb31 BL |
4754 | #ifdef CONFIG_NUMA |
4755 | memset(node_load, 0, sizeof(node_load)); | |
4756 | #endif | |
9adb62a5 JL |
4757 | |
4758 | if (self && !node_online(self->node_id)) { | |
4759 | build_zonelists(self); | |
9adb62a5 JL |
4760 | } |
4761 | ||
9276b1bc | 4762 | for_each_online_node(nid) { |
7ea1530a CL |
4763 | pg_data_t *pgdat = NODE_DATA(nid); |
4764 | ||
4765 | build_zonelists(pgdat); | |
9276b1bc | 4766 | } |
99dcc3e5 CL |
4767 | |
4768 | /* | |
4769 | * Initialize the boot_pagesets that are going to be used | |
4770 | * for bootstrapping processors. The real pagesets for | |
4771 | * each zone will be allocated later when the per cpu | |
4772 | * allocator is available. | |
4773 | * | |
4774 | * boot_pagesets are used also for bootstrapping offline | |
4775 | * cpus if the system is already booted because the pagesets | |
4776 | * are needed to initialize allocators on a specific cpu too. | |
4777 | * F.e. the percpu allocator needs the page allocator which | |
4778 | * needs the percpu allocator in order to allocate its pagesets | |
4779 | * (a chicken-egg dilemma). | |
4780 | */ | |
7aac7898 | 4781 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
4782 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
4783 | ||
7aac7898 LS |
4784 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4785 | /* | |
4786 | * We now know the "local memory node" for each node-- | |
4787 | * i.e., the node of the first zone in the generic zonelist. | |
4788 | * Set up numa_mem percpu variable for on-line cpus. During | |
4789 | * boot, only the boot cpu should be on-line; we'll init the | |
4790 | * secondary cpus' numa_mem as they come on-line. During | |
4791 | * node/memory hotplug, we'll fixup all on-line cpus. | |
4792 | */ | |
4793 | if (cpu_online(cpu)) | |
4794 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
4795 | #endif | |
4796 | } | |
4797 | ||
6811378e YG |
4798 | return 0; |
4799 | } | |
4800 | ||
061f67bc RV |
4801 | static noinline void __init |
4802 | build_all_zonelists_init(void) | |
4803 | { | |
4804 | __build_all_zonelists(NULL); | |
4805 | mminit_verify_zonelist(); | |
4806 | cpuset_init_current_mems_allowed(); | |
4807 | } | |
4808 | ||
4eaf3f64 HL |
4809 | /* |
4810 | * Called with zonelists_mutex held always | |
4811 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
4812 | * |
4813 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
4814 | * [we're only called with non-NULL zone through __meminit paths] and | |
4815 | * (2) call of __init annotated helper build_all_zonelists_init | |
4816 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 4817 | */ |
9adb62a5 | 4818 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 4819 | { |
f0c0b2b8 KH |
4820 | set_zonelist_order(); |
4821 | ||
6811378e | 4822 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 4823 | build_all_zonelists_init(); |
6811378e | 4824 | } else { |
e9959f0f | 4825 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
4826 | if (zone) |
4827 | setup_zone_pageset(zone); | |
e9959f0f | 4828 | #endif |
dd1895e2 CS |
4829 | /* we have to stop all cpus to guarantee there is no user |
4830 | of zonelist */ | |
9adb62a5 | 4831 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
4832 | /* cpuset refresh routine should be here */ |
4833 | } | |
bd1e22b8 | 4834 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
4835 | /* |
4836 | * Disable grouping by mobility if the number of pages in the | |
4837 | * system is too low to allow the mechanism to work. It would be | |
4838 | * more accurate, but expensive to check per-zone. This check is | |
4839 | * made on memory-hotadd so a system can start with mobility | |
4840 | * disabled and enable it later | |
4841 | */ | |
d9c23400 | 4842 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
4843 | page_group_by_mobility_disabled = 1; |
4844 | else | |
4845 | page_group_by_mobility_disabled = 0; | |
4846 | ||
756a025f JP |
4847 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
4848 | nr_online_nodes, | |
4849 | zonelist_order_name[current_zonelist_order], | |
4850 | page_group_by_mobility_disabled ? "off" : "on", | |
4851 | vm_total_pages); | |
f0c0b2b8 | 4852 | #ifdef CONFIG_NUMA |
f88dfff5 | 4853 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 4854 | #endif |
1da177e4 LT |
4855 | } |
4856 | ||
4857 | /* | |
4858 | * Helper functions to size the waitqueue hash table. | |
4859 | * Essentially these want to choose hash table sizes sufficiently | |
4860 | * large so that collisions trying to wait on pages are rare. | |
4861 | * But in fact, the number of active page waitqueues on typical | |
4862 | * systems is ridiculously low, less than 200. So this is even | |
4863 | * conservative, even though it seems large. | |
4864 | * | |
4865 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
4866 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
4867 | */ | |
4868 | #define PAGES_PER_WAITQUEUE 256 | |
4869 | ||
cca448fe | 4870 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 4871 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
4872 | { |
4873 | unsigned long size = 1; | |
4874 | ||
4875 | pages /= PAGES_PER_WAITQUEUE; | |
4876 | ||
4877 | while (size < pages) | |
4878 | size <<= 1; | |
4879 | ||
4880 | /* | |
4881 | * Once we have dozens or even hundreds of threads sleeping | |
4882 | * on IO we've got bigger problems than wait queue collision. | |
4883 | * Limit the size of the wait table to a reasonable size. | |
4884 | */ | |
4885 | size = min(size, 4096UL); | |
4886 | ||
4887 | return max(size, 4UL); | |
4888 | } | |
cca448fe YG |
4889 | #else |
4890 | /* | |
4891 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
4892 | * a suitable size for its wait_table. So we use the maximum size now. | |
4893 | * | |
4894 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
4895 | * | |
4896 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
4897 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
4898 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
4899 | * | |
4900 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
4901 | * or more by the traditional way. (See above). It equals: | |
4902 | * | |
4903 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
4904 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
4905 | * powerpc (64K page size) : = (32G +16M)byte. | |
4906 | */ | |
4907 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
4908 | { | |
4909 | return 4096UL; | |
4910 | } | |
4911 | #endif | |
1da177e4 LT |
4912 | |
4913 | /* | |
4914 | * This is an integer logarithm so that shifts can be used later | |
4915 | * to extract the more random high bits from the multiplicative | |
4916 | * hash function before the remainder is taken. | |
4917 | */ | |
4918 | static inline unsigned long wait_table_bits(unsigned long size) | |
4919 | { | |
4920 | return ffz(~size); | |
4921 | } | |
4922 | ||
1da177e4 LT |
4923 | /* |
4924 | * Initially all pages are reserved - free ones are freed | |
4925 | * up by free_all_bootmem() once the early boot process is | |
4926 | * done. Non-atomic initialization, single-pass. | |
4927 | */ | |
c09b4240 | 4928 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 4929 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 4930 | { |
4b94ffdc | 4931 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
29751f69 | 4932 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 4933 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 4934 | unsigned long pfn; |
3a80a7fa | 4935 | unsigned long nr_initialised = 0; |
342332e6 TI |
4936 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
4937 | struct memblock_region *r = NULL, *tmp; | |
4938 | #endif | |
1da177e4 | 4939 | |
22b31eec HD |
4940 | if (highest_memmap_pfn < end_pfn - 1) |
4941 | highest_memmap_pfn = end_pfn - 1; | |
4942 | ||
4b94ffdc DW |
4943 | /* |
4944 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
4945 | * memory | |
4946 | */ | |
4947 | if (altmap && start_pfn == altmap->base_pfn) | |
4948 | start_pfn += altmap->reserve; | |
4949 | ||
cbe8dd4a | 4950 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 4951 | /* |
b72d0ffb AM |
4952 | * There can be holes in boot-time mem_map[]s handed to this |
4953 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 4954 | */ |
b72d0ffb AM |
4955 | if (context != MEMMAP_EARLY) |
4956 | goto not_early; | |
4957 | ||
4958 | if (!early_pfn_valid(pfn)) | |
4959 | continue; | |
4960 | if (!early_pfn_in_nid(pfn, nid)) | |
4961 | continue; | |
4962 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
4963 | break; | |
342332e6 TI |
4964 | |
4965 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
4966 | /* |
4967 | * If not mirrored_kernelcore and ZONE_MOVABLE exists, range | |
4968 | * from zone_movable_pfn[nid] to end of each node should be | |
4969 | * ZONE_MOVABLE not ZONE_NORMAL. skip it. | |
4970 | */ | |
4971 | if (!mirrored_kernelcore && zone_movable_pfn[nid]) | |
4972 | if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid]) | |
4973 | continue; | |
342332e6 | 4974 | |
b72d0ffb AM |
4975 | /* |
4976 | * Check given memblock attribute by firmware which can affect | |
4977 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
4978 | * mirrored, it's an overlapped memmap init. skip it. | |
4979 | */ | |
4980 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
4981 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
4982 | for_each_memblock(memory, tmp) | |
4983 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
4984 | break; | |
4985 | r = tmp; | |
4986 | } | |
4987 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
4988 | memblock_is_mirror(r)) { | |
4989 | /* already initialized as NORMAL */ | |
4990 | pfn = memblock_region_memory_end_pfn(r); | |
4991 | continue; | |
342332e6 | 4992 | } |
a2f3aa02 | 4993 | } |
b72d0ffb | 4994 | #endif |
ac5d2539 | 4995 | |
b72d0ffb | 4996 | not_early: |
ac5d2539 MG |
4997 | /* |
4998 | * Mark the block movable so that blocks are reserved for | |
4999 | * movable at startup. This will force kernel allocations | |
5000 | * to reserve their blocks rather than leaking throughout | |
5001 | * the address space during boot when many long-lived | |
974a786e | 5002 | * kernel allocations are made. |
ac5d2539 MG |
5003 | * |
5004 | * bitmap is created for zone's valid pfn range. but memmap | |
5005 | * can be created for invalid pages (for alignment) | |
5006 | * check here not to call set_pageblock_migratetype() against | |
5007 | * pfn out of zone. | |
5008 | */ | |
5009 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
5010 | struct page *page = pfn_to_page(pfn); | |
5011 | ||
5012 | __init_single_page(page, pfn, zone, nid); | |
5013 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5014 | } else { | |
5015 | __init_single_pfn(pfn, zone, nid); | |
5016 | } | |
1da177e4 LT |
5017 | } |
5018 | } | |
5019 | ||
1e548deb | 5020 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 5021 | { |
7aeb09f9 | 5022 | unsigned int order, t; |
b2a0ac88 MG |
5023 | for_each_migratetype_order(order, t) { |
5024 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
5025 | zone->free_area[order].nr_free = 0; |
5026 | } | |
5027 | } | |
5028 | ||
5029 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
5030 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 5031 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
5032 | #endif |
5033 | ||
7cd2b0a3 | 5034 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 5035 | { |
3a6be87f | 5036 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
5037 | int batch; |
5038 | ||
5039 | /* | |
5040 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 5041 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
5042 | * |
5043 | * OK, so we don't know how big the cache is. So guess. | |
5044 | */ | |
b40da049 | 5045 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
5046 | if (batch * PAGE_SIZE > 512 * 1024) |
5047 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
5048 | batch /= 4; /* We effectively *= 4 below */ |
5049 | if (batch < 1) | |
5050 | batch = 1; | |
5051 | ||
5052 | /* | |
0ceaacc9 NP |
5053 | * Clamp the batch to a 2^n - 1 value. Having a power |
5054 | * of 2 value was found to be more likely to have | |
5055 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 5056 | * |
0ceaacc9 NP |
5057 | * For example if 2 tasks are alternately allocating |
5058 | * batches of pages, one task can end up with a lot | |
5059 | * of pages of one half of the possible page colors | |
5060 | * and the other with pages of the other colors. | |
e7c8d5c9 | 5061 | */ |
9155203a | 5062 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 5063 | |
e7c8d5c9 | 5064 | return batch; |
3a6be87f DH |
5065 | |
5066 | #else | |
5067 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5068 | * conditions. | |
5069 | * | |
5070 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5071 | * of contiguous memory as there's no hardware page translation to | |
5072 | * assemble apparent contiguous memory from discontiguous pages. | |
5073 | * | |
5074 | * Queueing large contiguous runs of pages for batching, however, | |
5075 | * causes the pages to actually be freed in smaller chunks. As there | |
5076 | * can be a significant delay between the individual batches being | |
5077 | * recycled, this leads to the once large chunks of space being | |
5078 | * fragmented and becoming unavailable for high-order allocations. | |
5079 | */ | |
5080 | return 0; | |
5081 | #endif | |
e7c8d5c9 CL |
5082 | } |
5083 | ||
8d7a8fa9 CS |
5084 | /* |
5085 | * pcp->high and pcp->batch values are related and dependent on one another: | |
5086 | * ->batch must never be higher then ->high. | |
5087 | * The following function updates them in a safe manner without read side | |
5088 | * locking. | |
5089 | * | |
5090 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
5091 | * those fields changing asynchronously (acording the the above rule). | |
5092 | * | |
5093 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5094 | * outside of boot time (or some other assurance that no concurrent updaters | |
5095 | * exist). | |
5096 | */ | |
5097 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
5098 | unsigned long batch) | |
5099 | { | |
5100 | /* start with a fail safe value for batch */ | |
5101 | pcp->batch = 1; | |
5102 | smp_wmb(); | |
5103 | ||
5104 | /* Update high, then batch, in order */ | |
5105 | pcp->high = high; | |
5106 | smp_wmb(); | |
5107 | ||
5108 | pcp->batch = batch; | |
5109 | } | |
5110 | ||
3664033c | 5111 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
5112 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5113 | { | |
8d7a8fa9 | 5114 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
5115 | } |
5116 | ||
88c90dbc | 5117 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
5118 | { |
5119 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 5120 | int migratetype; |
2caaad41 | 5121 | |
1c6fe946 MD |
5122 | memset(p, 0, sizeof(*p)); |
5123 | ||
3dfa5721 | 5124 | pcp = &p->pcp; |
2caaad41 | 5125 | pcp->count = 0; |
5f8dcc21 MG |
5126 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5127 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
5128 | } |
5129 | ||
88c90dbc CS |
5130 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5131 | { | |
5132 | pageset_init(p); | |
5133 | pageset_set_batch(p, batch); | |
5134 | } | |
5135 | ||
8ad4b1fb | 5136 | /* |
3664033c | 5137 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
5138 | * to the value high for the pageset p. |
5139 | */ | |
3664033c | 5140 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
5141 | unsigned long high) |
5142 | { | |
8d7a8fa9 CS |
5143 | unsigned long batch = max(1UL, high / 4); |
5144 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
5145 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 5146 | |
8d7a8fa9 | 5147 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
5148 | } |
5149 | ||
7cd2b0a3 DR |
5150 | static void pageset_set_high_and_batch(struct zone *zone, |
5151 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 5152 | { |
56cef2b8 | 5153 | if (percpu_pagelist_fraction) |
3664033c | 5154 | pageset_set_high(pcp, |
56cef2b8 CS |
5155 | (zone->managed_pages / |
5156 | percpu_pagelist_fraction)); | |
5157 | else | |
5158 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
5159 | } | |
5160 | ||
169f6c19 CS |
5161 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5162 | { | |
5163 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
5164 | ||
5165 | pageset_init(pcp); | |
5166 | pageset_set_high_and_batch(zone, pcp); | |
5167 | } | |
5168 | ||
4ed7e022 | 5169 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
5170 | { |
5171 | int cpu; | |
319774e2 | 5172 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5173 | for_each_possible_cpu(cpu) |
5174 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5175 | } |
5176 | ||
2caaad41 | 5177 | /* |
99dcc3e5 CL |
5178 | * Allocate per cpu pagesets and initialize them. |
5179 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5180 | */ |
99dcc3e5 | 5181 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5182 | { |
99dcc3e5 | 5183 | struct zone *zone; |
e7c8d5c9 | 5184 | |
319774e2 WF |
5185 | for_each_populated_zone(zone) |
5186 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
5187 | } |
5188 | ||
577a32f6 | 5189 | static noinline __init_refok |
cca448fe | 5190 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
5191 | { |
5192 | int i; | |
cca448fe | 5193 | size_t alloc_size; |
ed8ece2e DH |
5194 | |
5195 | /* | |
5196 | * The per-page waitqueue mechanism uses hashed waitqueues | |
5197 | * per zone. | |
5198 | */ | |
02b694de YG |
5199 | zone->wait_table_hash_nr_entries = |
5200 | wait_table_hash_nr_entries(zone_size_pages); | |
5201 | zone->wait_table_bits = | |
5202 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
5203 | alloc_size = zone->wait_table_hash_nr_entries |
5204 | * sizeof(wait_queue_head_t); | |
5205 | ||
cd94b9db | 5206 | if (!slab_is_available()) { |
cca448fe | 5207 | zone->wait_table = (wait_queue_head_t *) |
6782832e SS |
5208 | memblock_virt_alloc_node_nopanic( |
5209 | alloc_size, zone->zone_pgdat->node_id); | |
cca448fe YG |
5210 | } else { |
5211 | /* | |
5212 | * This case means that a zone whose size was 0 gets new memory | |
5213 | * via memory hot-add. | |
5214 | * But it may be the case that a new node was hot-added. In | |
5215 | * this case vmalloc() will not be able to use this new node's | |
5216 | * memory - this wait_table must be initialized to use this new | |
5217 | * node itself as well. | |
5218 | * To use this new node's memory, further consideration will be | |
5219 | * necessary. | |
5220 | */ | |
8691f3a7 | 5221 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
5222 | } |
5223 | if (!zone->wait_table) | |
5224 | return -ENOMEM; | |
ed8ece2e | 5225 | |
b8af2941 | 5226 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 5227 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
5228 | |
5229 | return 0; | |
ed8ece2e DH |
5230 | } |
5231 | ||
c09b4240 | 5232 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5233 | { |
99dcc3e5 CL |
5234 | /* |
5235 | * per cpu subsystem is not up at this point. The following code | |
5236 | * relies on the ability of the linker to provide the | |
5237 | * offset of a (static) per cpu variable into the per cpu area. | |
5238 | */ | |
5239 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5240 | |
b38a8725 | 5241 | if (populated_zone(zone)) |
99dcc3e5 CL |
5242 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5243 | zone->name, zone->present_pages, | |
5244 | zone_batchsize(zone)); | |
ed8ece2e DH |
5245 | } |
5246 | ||
4ed7e022 | 5247 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5248 | unsigned long zone_start_pfn, |
b171e409 | 5249 | unsigned long size) |
ed8ece2e DH |
5250 | { |
5251 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
5252 | int ret; |
5253 | ret = zone_wait_table_init(zone, size); | |
5254 | if (ret) | |
5255 | return ret; | |
ed8ece2e DH |
5256 | pgdat->nr_zones = zone_idx(zone) + 1; |
5257 | ||
ed8ece2e DH |
5258 | zone->zone_start_pfn = zone_start_pfn; |
5259 | ||
708614e6 MG |
5260 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5261 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5262 | pgdat->node_id, | |
5263 | (unsigned long)zone_idx(zone), | |
5264 | zone_start_pfn, (zone_start_pfn + size)); | |
5265 | ||
1e548deb | 5266 | zone_init_free_lists(zone); |
718127cc YG |
5267 | |
5268 | return 0; | |
ed8ece2e DH |
5269 | } |
5270 | ||
0ee332c1 | 5271 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5272 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5273 | |
c713216d MG |
5274 | /* |
5275 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5276 | */ |
8a942fde MG |
5277 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5278 | struct mminit_pfnnid_cache *state) | |
c713216d | 5279 | { |
c13291a5 | 5280 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5281 | int nid; |
7c243c71 | 5282 | |
8a942fde MG |
5283 | if (state->last_start <= pfn && pfn < state->last_end) |
5284 | return state->last_nid; | |
c713216d | 5285 | |
e76b63f8 YL |
5286 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5287 | if (nid != -1) { | |
8a942fde MG |
5288 | state->last_start = start_pfn; |
5289 | state->last_end = end_pfn; | |
5290 | state->last_nid = nid; | |
e76b63f8 YL |
5291 | } |
5292 | ||
5293 | return nid; | |
c713216d MG |
5294 | } |
5295 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5296 | ||
c713216d | 5297 | /** |
6782832e | 5298 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5299 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5300 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5301 | * |
7d018176 ZZ |
5302 | * If an architecture guarantees that all ranges registered contain no holes |
5303 | * and may be freed, this this function may be used instead of calling | |
5304 | * memblock_free_early_nid() manually. | |
c713216d | 5305 | */ |
c13291a5 | 5306 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5307 | { |
c13291a5 TH |
5308 | unsigned long start_pfn, end_pfn; |
5309 | int i, this_nid; | |
edbe7d23 | 5310 | |
c13291a5 TH |
5311 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5312 | start_pfn = min(start_pfn, max_low_pfn); | |
5313 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5314 | |
c13291a5 | 5315 | if (start_pfn < end_pfn) |
6782832e SS |
5316 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5317 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5318 | this_nid); | |
edbe7d23 | 5319 | } |
edbe7d23 | 5320 | } |
edbe7d23 | 5321 | |
c713216d MG |
5322 | /** |
5323 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5324 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5325 | * |
7d018176 ZZ |
5326 | * If an architecture guarantees that all ranges registered contain no holes and may |
5327 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5328 | */ |
5329 | void __init sparse_memory_present_with_active_regions(int nid) | |
5330 | { | |
c13291a5 TH |
5331 | unsigned long start_pfn, end_pfn; |
5332 | int i, this_nid; | |
c713216d | 5333 | |
c13291a5 TH |
5334 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5335 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5336 | } |
5337 | ||
5338 | /** | |
5339 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5340 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5341 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5342 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5343 | * |
5344 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5345 | * provided by memblock_set_node(). If called for a node |
c713216d | 5346 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5347 | * PFNs will be 0. |
c713216d | 5348 | */ |
a3142c8e | 5349 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5350 | unsigned long *start_pfn, unsigned long *end_pfn) |
5351 | { | |
c13291a5 | 5352 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5353 | int i; |
c13291a5 | 5354 | |
c713216d MG |
5355 | *start_pfn = -1UL; |
5356 | *end_pfn = 0; | |
5357 | ||
c13291a5 TH |
5358 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5359 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5360 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5361 | } |
5362 | ||
633c0666 | 5363 | if (*start_pfn == -1UL) |
c713216d | 5364 | *start_pfn = 0; |
c713216d MG |
5365 | } |
5366 | ||
2a1e274a MG |
5367 | /* |
5368 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5369 | * assumption is made that zones within a node are ordered in monotonic | |
5370 | * increasing memory addresses so that the "highest" populated zone is used | |
5371 | */ | |
b69a7288 | 5372 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5373 | { |
5374 | int zone_index; | |
5375 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5376 | if (zone_index == ZONE_MOVABLE) | |
5377 | continue; | |
5378 | ||
5379 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5380 | arch_zone_lowest_possible_pfn[zone_index]) | |
5381 | break; | |
5382 | } | |
5383 | ||
5384 | VM_BUG_ON(zone_index == -1); | |
5385 | movable_zone = zone_index; | |
5386 | } | |
5387 | ||
5388 | /* | |
5389 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5390 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5391 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5392 | * in each node depending on the size of each node and how evenly kernelcore | |
5393 | * is distributed. This helper function adjusts the zone ranges | |
5394 | * provided by the architecture for a given node by using the end of the | |
5395 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5396 | * zones within a node are in order of monotonic increases memory addresses | |
5397 | */ | |
b69a7288 | 5398 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5399 | unsigned long zone_type, |
5400 | unsigned long node_start_pfn, | |
5401 | unsigned long node_end_pfn, | |
5402 | unsigned long *zone_start_pfn, | |
5403 | unsigned long *zone_end_pfn) | |
5404 | { | |
5405 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5406 | if (zone_movable_pfn[nid]) { | |
5407 | /* Size ZONE_MOVABLE */ | |
5408 | if (zone_type == ZONE_MOVABLE) { | |
5409 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5410 | *zone_end_pfn = min(node_end_pfn, | |
5411 | arch_zone_highest_possible_pfn[movable_zone]); | |
5412 | ||
2a1e274a MG |
5413 | /* Check if this whole range is within ZONE_MOVABLE */ |
5414 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5415 | *zone_start_pfn = *zone_end_pfn; | |
5416 | } | |
5417 | } | |
5418 | ||
c713216d MG |
5419 | /* |
5420 | * Return the number of pages a zone spans in a node, including holes | |
5421 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5422 | */ | |
6ea6e688 | 5423 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5424 | unsigned long zone_type, |
7960aedd ZY |
5425 | unsigned long node_start_pfn, |
5426 | unsigned long node_end_pfn, | |
d91749c1 TI |
5427 | unsigned long *zone_start_pfn, |
5428 | unsigned long *zone_end_pfn, | |
c713216d MG |
5429 | unsigned long *ignored) |
5430 | { | |
b5685e92 | 5431 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5432 | if (!node_start_pfn && !node_end_pfn) |
5433 | return 0; | |
5434 | ||
7960aedd | 5435 | /* Get the start and end of the zone */ |
d91749c1 TI |
5436 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5437 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5438 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5439 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5440 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5441 | |
5442 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5443 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5444 | return 0; |
5445 | ||
5446 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5447 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5448 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5449 | |
5450 | /* Return the spanned pages */ | |
d91749c1 | 5451 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5452 | } |
5453 | ||
5454 | /* | |
5455 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5456 | * then all holes in the requested range will be accounted for. |
c713216d | 5457 | */ |
32996250 | 5458 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5459 | unsigned long range_start_pfn, |
5460 | unsigned long range_end_pfn) | |
5461 | { | |
96e907d1 TH |
5462 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5463 | unsigned long start_pfn, end_pfn; | |
5464 | int i; | |
c713216d | 5465 | |
96e907d1 TH |
5466 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5467 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5468 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5469 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5470 | } |
96e907d1 | 5471 | return nr_absent; |
c713216d MG |
5472 | } |
5473 | ||
5474 | /** | |
5475 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5476 | * @start_pfn: The start PFN to start searching for holes | |
5477 | * @end_pfn: The end PFN to stop searching for holes | |
5478 | * | |
88ca3b94 | 5479 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5480 | */ |
5481 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5482 | unsigned long end_pfn) | |
5483 | { | |
5484 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5485 | } | |
5486 | ||
5487 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5488 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5489 | unsigned long zone_type, |
7960aedd ZY |
5490 | unsigned long node_start_pfn, |
5491 | unsigned long node_end_pfn, | |
c713216d MG |
5492 | unsigned long *ignored) |
5493 | { | |
96e907d1 TH |
5494 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5495 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5496 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5497 | unsigned long nr_absent; |
9c7cd687 | 5498 | |
b5685e92 | 5499 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5500 | if (!node_start_pfn && !node_end_pfn) |
5501 | return 0; | |
5502 | ||
96e907d1 TH |
5503 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5504 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5505 | |
2a1e274a MG |
5506 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5507 | node_start_pfn, node_end_pfn, | |
5508 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5509 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5510 | ||
5511 | /* | |
5512 | * ZONE_MOVABLE handling. | |
5513 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
5514 | * and vice versa. | |
5515 | */ | |
5516 | if (zone_movable_pfn[nid]) { | |
5517 | if (mirrored_kernelcore) { | |
5518 | unsigned long start_pfn, end_pfn; | |
5519 | struct memblock_region *r; | |
5520 | ||
5521 | for_each_memblock(memory, r) { | |
5522 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
5523 | zone_start_pfn, zone_end_pfn); | |
5524 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
5525 | zone_start_pfn, zone_end_pfn); | |
5526 | ||
5527 | if (zone_type == ZONE_MOVABLE && | |
5528 | memblock_is_mirror(r)) | |
5529 | nr_absent += end_pfn - start_pfn; | |
5530 | ||
5531 | if (zone_type == ZONE_NORMAL && | |
5532 | !memblock_is_mirror(r)) | |
5533 | nr_absent += end_pfn - start_pfn; | |
5534 | } | |
5535 | } else { | |
5536 | if (zone_type == ZONE_NORMAL) | |
5537 | nr_absent += node_end_pfn - zone_movable_pfn[nid]; | |
5538 | } | |
5539 | } | |
5540 | ||
5541 | return nr_absent; | |
c713216d | 5542 | } |
0e0b864e | 5543 | |
0ee332c1 | 5544 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 5545 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5546 | unsigned long zone_type, |
7960aedd ZY |
5547 | unsigned long node_start_pfn, |
5548 | unsigned long node_end_pfn, | |
d91749c1 TI |
5549 | unsigned long *zone_start_pfn, |
5550 | unsigned long *zone_end_pfn, | |
c713216d MG |
5551 | unsigned long *zones_size) |
5552 | { | |
d91749c1 TI |
5553 | unsigned int zone; |
5554 | ||
5555 | *zone_start_pfn = node_start_pfn; | |
5556 | for (zone = 0; zone < zone_type; zone++) | |
5557 | *zone_start_pfn += zones_size[zone]; | |
5558 | ||
5559 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
5560 | ||
c713216d MG |
5561 | return zones_size[zone_type]; |
5562 | } | |
5563 | ||
6ea6e688 | 5564 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5565 | unsigned long zone_type, |
7960aedd ZY |
5566 | unsigned long node_start_pfn, |
5567 | unsigned long node_end_pfn, | |
c713216d MG |
5568 | unsigned long *zholes_size) |
5569 | { | |
5570 | if (!zholes_size) | |
5571 | return 0; | |
5572 | ||
5573 | return zholes_size[zone_type]; | |
5574 | } | |
20e6926d | 5575 | |
0ee332c1 | 5576 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5577 | |
a3142c8e | 5578 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
5579 | unsigned long node_start_pfn, |
5580 | unsigned long node_end_pfn, | |
5581 | unsigned long *zones_size, | |
5582 | unsigned long *zholes_size) | |
c713216d | 5583 | { |
febd5949 | 5584 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
5585 | enum zone_type i; |
5586 | ||
febd5949 GZ |
5587 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5588 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 5589 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 5590 | unsigned long size, real_size; |
c713216d | 5591 | |
febd5949 GZ |
5592 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5593 | node_start_pfn, | |
5594 | node_end_pfn, | |
d91749c1 TI |
5595 | &zone_start_pfn, |
5596 | &zone_end_pfn, | |
febd5949 GZ |
5597 | zones_size); |
5598 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
5599 | node_start_pfn, node_end_pfn, |
5600 | zholes_size); | |
d91749c1 TI |
5601 | if (size) |
5602 | zone->zone_start_pfn = zone_start_pfn; | |
5603 | else | |
5604 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
5605 | zone->spanned_pages = size; |
5606 | zone->present_pages = real_size; | |
5607 | ||
5608 | totalpages += size; | |
5609 | realtotalpages += real_size; | |
5610 | } | |
5611 | ||
5612 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
5613 | pgdat->node_present_pages = realtotalpages; |
5614 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
5615 | realtotalpages); | |
5616 | } | |
5617 | ||
835c134e MG |
5618 | #ifndef CONFIG_SPARSEMEM |
5619 | /* | |
5620 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
5621 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5622 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
5623 | * round what is now in bits to nearest long in bits, then return it in |
5624 | * bytes. | |
5625 | */ | |
7c45512d | 5626 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
5627 | { |
5628 | unsigned long usemapsize; | |
5629 | ||
7c45512d | 5630 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
5631 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
5632 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
5633 | usemapsize *= NR_PAGEBLOCK_BITS; |
5634 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
5635 | ||
5636 | return usemapsize / 8; | |
5637 | } | |
5638 | ||
5639 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
5640 | struct zone *zone, |
5641 | unsigned long zone_start_pfn, | |
5642 | unsigned long zonesize) | |
835c134e | 5643 | { |
7c45512d | 5644 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 5645 | zone->pageblock_flags = NULL; |
58a01a45 | 5646 | if (usemapsize) |
6782832e SS |
5647 | zone->pageblock_flags = |
5648 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5649 | pgdat->node_id); | |
835c134e MG |
5650 | } |
5651 | #else | |
7c45512d LT |
5652 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5653 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5654 | #endif /* CONFIG_SPARSEMEM */ |
5655 | ||
d9c23400 | 5656 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5657 | |
d9c23400 | 5658 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5659 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5660 | { |
955c1cd7 AM |
5661 | unsigned int order; |
5662 | ||
d9c23400 MG |
5663 | /* Check that pageblock_nr_pages has not already been setup */ |
5664 | if (pageblock_order) | |
5665 | return; | |
5666 | ||
955c1cd7 AM |
5667 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5668 | order = HUGETLB_PAGE_ORDER; | |
5669 | else | |
5670 | order = MAX_ORDER - 1; | |
5671 | ||
d9c23400 MG |
5672 | /* |
5673 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5674 | * This value may be variable depending on boot parameters on IA64 and |
5675 | * powerpc. | |
d9c23400 MG |
5676 | */ |
5677 | pageblock_order = order; | |
5678 | } | |
5679 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5680 | ||
ba72cb8c MG |
5681 | /* |
5682 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5683 | * is unused as pageblock_order is set at compile-time. See |
5684 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5685 | * the kernel config | |
ba72cb8c | 5686 | */ |
15ca220e | 5687 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5688 | { |
ba72cb8c | 5689 | } |
d9c23400 MG |
5690 | |
5691 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5692 | ||
01cefaef JL |
5693 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5694 | unsigned long present_pages) | |
5695 | { | |
5696 | unsigned long pages = spanned_pages; | |
5697 | ||
5698 | /* | |
5699 | * Provide a more accurate estimation if there are holes within | |
5700 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5701 | * zone, each populated memory region may cost us one or two extra | |
5702 | * memmap pages due to alignment because memmap pages for each | |
5703 | * populated regions may not naturally algined on page boundary. | |
5704 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
5705 | */ | |
5706 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5707 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5708 | pages = present_pages; | |
5709 | ||
5710 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5711 | } | |
5712 | ||
1da177e4 LT |
5713 | /* |
5714 | * Set up the zone data structures: | |
5715 | * - mark all pages reserved | |
5716 | * - mark all memory queues empty | |
5717 | * - clear the memory bitmaps | |
6527af5d MK |
5718 | * |
5719 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 5720 | */ |
7f3eb55b | 5721 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 5722 | { |
2f1b6248 | 5723 | enum zone_type j; |
ed8ece2e | 5724 | int nid = pgdat->node_id; |
718127cc | 5725 | int ret; |
1da177e4 | 5726 | |
208d54e5 | 5727 | pgdat_resize_init(pgdat); |
8177a420 AA |
5728 | #ifdef CONFIG_NUMA_BALANCING |
5729 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
5730 | pgdat->numabalancing_migrate_nr_pages = 0; | |
5731 | pgdat->numabalancing_migrate_next_window = jiffies; | |
a3d0a918 KS |
5732 | #endif |
5733 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5734 | spin_lock_init(&pgdat->split_queue_lock); | |
5735 | INIT_LIST_HEAD(&pgdat->split_queue); | |
5736 | pgdat->split_queue_len = 0; | |
8177a420 | 5737 | #endif |
1da177e4 | 5738 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 5739 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
698b1b30 VB |
5740 | #ifdef CONFIG_COMPACTION |
5741 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
5742 | #endif | |
eefa864b | 5743 | pgdat_page_ext_init(pgdat); |
5f63b720 | 5744 | |
1da177e4 LT |
5745 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5746 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 5747 | unsigned long size, realsize, freesize, memmap_pages; |
d91749c1 | 5748 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 5749 | |
febd5949 GZ |
5750 | size = zone->spanned_pages; |
5751 | realsize = freesize = zone->present_pages; | |
1da177e4 | 5752 | |
0e0b864e | 5753 | /* |
9feedc9d | 5754 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
5755 | * is used by this zone for memmap. This affects the watermark |
5756 | * and per-cpu initialisations | |
5757 | */ | |
01cefaef | 5758 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
5759 | if (!is_highmem_idx(j)) { |
5760 | if (freesize >= memmap_pages) { | |
5761 | freesize -= memmap_pages; | |
5762 | if (memmap_pages) | |
5763 | printk(KERN_DEBUG | |
5764 | " %s zone: %lu pages used for memmap\n", | |
5765 | zone_names[j], memmap_pages); | |
5766 | } else | |
1170532b | 5767 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
5768 | zone_names[j], memmap_pages, freesize); |
5769 | } | |
0e0b864e | 5770 | |
6267276f | 5771 | /* Account for reserved pages */ |
9feedc9d JL |
5772 | if (j == 0 && freesize > dma_reserve) { |
5773 | freesize -= dma_reserve; | |
d903ef9f | 5774 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 5775 | zone_names[0], dma_reserve); |
0e0b864e MG |
5776 | } |
5777 | ||
98d2b0eb | 5778 | if (!is_highmem_idx(j)) |
9feedc9d | 5779 | nr_kernel_pages += freesize; |
01cefaef JL |
5780 | /* Charge for highmem memmap if there are enough kernel pages */ |
5781 | else if (nr_kernel_pages > memmap_pages * 2) | |
5782 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 5783 | nr_all_pages += freesize; |
1da177e4 | 5784 | |
9feedc9d JL |
5785 | /* |
5786 | * Set an approximate value for lowmem here, it will be adjusted | |
5787 | * when the bootmem allocator frees pages into the buddy system. | |
5788 | * And all highmem pages will be managed by the buddy system. | |
5789 | */ | |
5790 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 5791 | #ifdef CONFIG_NUMA |
d5f541ed | 5792 | zone->node = nid; |
9feedc9d | 5793 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 5794 | / 100; |
9feedc9d | 5795 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 5796 | #endif |
1da177e4 LT |
5797 | zone->name = zone_names[j]; |
5798 | spin_lock_init(&zone->lock); | |
5799 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 5800 | zone_seqlock_init(zone); |
1da177e4 | 5801 | zone->zone_pgdat = pgdat; |
ed8ece2e | 5802 | zone_pcp_init(zone); |
81c0a2bb JW |
5803 | |
5804 | /* For bootup, initialized properly in watermark setup */ | |
5805 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
5806 | ||
bea8c150 | 5807 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
5808 | if (!size) |
5809 | continue; | |
5810 | ||
955c1cd7 | 5811 | set_pageblock_order(); |
7c45512d | 5812 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
b171e409 | 5813 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
718127cc | 5814 | BUG_ON(ret); |
76cdd58e | 5815 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
5816 | } |
5817 | } | |
5818 | ||
577a32f6 | 5819 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 5820 | { |
b0aeba74 | 5821 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
5822 | unsigned long __maybe_unused offset = 0; |
5823 | ||
1da177e4 LT |
5824 | /* Skip empty nodes */ |
5825 | if (!pgdat->node_spanned_pages) | |
5826 | return; | |
5827 | ||
d41dee36 | 5828 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
b0aeba74 TL |
5829 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
5830 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
5831 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
5832 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 5833 | unsigned long size, end; |
d41dee36 AW |
5834 | struct page *map; |
5835 | ||
e984bb43 BP |
5836 | /* |
5837 | * The zone's endpoints aren't required to be MAX_ORDER | |
5838 | * aligned but the node_mem_map endpoints must be in order | |
5839 | * for the buddy allocator to function correctly. | |
5840 | */ | |
108bcc96 | 5841 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
5842 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
5843 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
5844 | map = alloc_remap(pgdat->node_id, size); |
5845 | if (!map) | |
6782832e SS |
5846 | map = memblock_virt_alloc_node_nopanic(size, |
5847 | pgdat->node_id); | |
a1c34a3b | 5848 | pgdat->node_mem_map = map + offset; |
1da177e4 | 5849 | } |
12d810c1 | 5850 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
5851 | /* |
5852 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
5853 | */ | |
c713216d | 5854 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 5855 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 5856 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 5857 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 5858 | mem_map -= offset; |
0ee332c1 | 5859 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5860 | } |
1da177e4 | 5861 | #endif |
d41dee36 | 5862 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
5863 | } |
5864 | ||
9109fb7b JW |
5865 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
5866 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 5867 | { |
9109fb7b | 5868 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
5869 | unsigned long start_pfn = 0; |
5870 | unsigned long end_pfn = 0; | |
9109fb7b | 5871 | |
88fdf75d | 5872 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 5873 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 5874 | |
3a80a7fa | 5875 | reset_deferred_meminit(pgdat); |
1da177e4 LT |
5876 | pgdat->node_id = nid; |
5877 | pgdat->node_start_pfn = node_start_pfn; | |
7960aedd ZY |
5878 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5879 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 5880 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
5881 | (u64)start_pfn << PAGE_SHIFT, |
5882 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
5883 | #else |
5884 | start_pfn = node_start_pfn; | |
7960aedd ZY |
5885 | #endif |
5886 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
5887 | zones_size, zholes_size); | |
1da177e4 LT |
5888 | |
5889 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
5890 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
5891 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
5892 | nid, (unsigned long)pgdat, | |
5893 | (unsigned long)pgdat->node_mem_map); | |
5894 | #endif | |
1da177e4 | 5895 | |
7f3eb55b | 5896 | free_area_init_core(pgdat); |
1da177e4 LT |
5897 | } |
5898 | ||
0ee332c1 | 5899 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
5900 | |
5901 | #if MAX_NUMNODES > 1 | |
5902 | /* | |
5903 | * Figure out the number of possible node ids. | |
5904 | */ | |
f9872caf | 5905 | void __init setup_nr_node_ids(void) |
418508c1 | 5906 | { |
904a9553 | 5907 | unsigned int highest; |
418508c1 | 5908 | |
904a9553 | 5909 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
5910 | nr_node_ids = highest + 1; |
5911 | } | |
418508c1 MS |
5912 | #endif |
5913 | ||
1e01979c TH |
5914 | /** |
5915 | * node_map_pfn_alignment - determine the maximum internode alignment | |
5916 | * | |
5917 | * This function should be called after node map is populated and sorted. | |
5918 | * It calculates the maximum power of two alignment which can distinguish | |
5919 | * all the nodes. | |
5920 | * | |
5921 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
5922 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
5923 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
5924 | * shifted, 1GiB is enough and this function will indicate so. | |
5925 | * | |
5926 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
5927 | * model has fine enough granularity to avoid incorrect mapping for the | |
5928 | * populated node map. | |
5929 | * | |
5930 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
5931 | * requirement (single node). | |
5932 | */ | |
5933 | unsigned long __init node_map_pfn_alignment(void) | |
5934 | { | |
5935 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 5936 | unsigned long start, end, mask; |
1e01979c | 5937 | int last_nid = -1; |
c13291a5 | 5938 | int i, nid; |
1e01979c | 5939 | |
c13291a5 | 5940 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5941 | if (!start || last_nid < 0 || last_nid == nid) { |
5942 | last_nid = nid; | |
5943 | last_end = end; | |
5944 | continue; | |
5945 | } | |
5946 | ||
5947 | /* | |
5948 | * Start with a mask granular enough to pin-point to the | |
5949 | * start pfn and tick off bits one-by-one until it becomes | |
5950 | * too coarse to separate the current node from the last. | |
5951 | */ | |
5952 | mask = ~((1 << __ffs(start)) - 1); | |
5953 | while (mask && last_end <= (start & (mask << 1))) | |
5954 | mask <<= 1; | |
5955 | ||
5956 | /* accumulate all internode masks */ | |
5957 | accl_mask |= mask; | |
5958 | } | |
5959 | ||
5960 | /* convert mask to number of pages */ | |
5961 | return ~accl_mask + 1; | |
5962 | } | |
5963 | ||
a6af2bc3 | 5964 | /* Find the lowest pfn for a node */ |
b69a7288 | 5965 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 5966 | { |
a6af2bc3 | 5967 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
5968 | unsigned long start_pfn; |
5969 | int i; | |
1abbfb41 | 5970 | |
c13291a5 TH |
5971 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
5972 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 5973 | |
a6af2bc3 | 5974 | if (min_pfn == ULONG_MAX) { |
1170532b | 5975 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
5976 | return 0; |
5977 | } | |
5978 | ||
5979 | return min_pfn; | |
c713216d MG |
5980 | } |
5981 | ||
5982 | /** | |
5983 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
5984 | * | |
5985 | * It returns the minimum PFN based on information provided via | |
7d018176 | 5986 | * memblock_set_node(). |
c713216d MG |
5987 | */ |
5988 | unsigned long __init find_min_pfn_with_active_regions(void) | |
5989 | { | |
5990 | return find_min_pfn_for_node(MAX_NUMNODES); | |
5991 | } | |
5992 | ||
37b07e41 LS |
5993 | /* |
5994 | * early_calculate_totalpages() | |
5995 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 5996 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 5997 | */ |
484f51f8 | 5998 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 5999 | { |
7e63efef | 6000 | unsigned long totalpages = 0; |
c13291a5 TH |
6001 | unsigned long start_pfn, end_pfn; |
6002 | int i, nid; | |
6003 | ||
6004 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
6005 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 6006 | |
37b07e41 LS |
6007 | totalpages += pages; |
6008 | if (pages) | |
4b0ef1fe | 6009 | node_set_state(nid, N_MEMORY); |
37b07e41 | 6010 | } |
b8af2941 | 6011 | return totalpages; |
7e63efef MG |
6012 | } |
6013 | ||
2a1e274a MG |
6014 | /* |
6015 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
6016 | * is spread evenly between nodes as long as the nodes have enough | |
6017 | * memory. When they don't, some nodes will have more kernelcore than | |
6018 | * others | |
6019 | */ | |
b224ef85 | 6020 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
6021 | { |
6022 | int i, nid; | |
6023 | unsigned long usable_startpfn; | |
6024 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 6025 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 6026 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 6027 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 6028 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 6029 | struct memblock_region *r; |
b2f3eebe TC |
6030 | |
6031 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
6032 | find_usable_zone_for_movable(); | |
6033 | ||
6034 | /* | |
6035 | * If movable_node is specified, ignore kernelcore and movablecore | |
6036 | * options. | |
6037 | */ | |
6038 | if (movable_node_is_enabled()) { | |
136199f0 EM |
6039 | for_each_memblock(memory, r) { |
6040 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
6041 | continue; |
6042 | ||
136199f0 | 6043 | nid = r->nid; |
b2f3eebe | 6044 | |
136199f0 | 6045 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
6046 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6047 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6048 | usable_startpfn; | |
6049 | } | |
6050 | ||
6051 | goto out2; | |
6052 | } | |
2a1e274a | 6053 | |
342332e6 TI |
6054 | /* |
6055 | * If kernelcore=mirror is specified, ignore movablecore option | |
6056 | */ | |
6057 | if (mirrored_kernelcore) { | |
6058 | bool mem_below_4gb_not_mirrored = false; | |
6059 | ||
6060 | for_each_memblock(memory, r) { | |
6061 | if (memblock_is_mirror(r)) | |
6062 | continue; | |
6063 | ||
6064 | nid = r->nid; | |
6065 | ||
6066 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
6067 | ||
6068 | if (usable_startpfn < 0x100000) { | |
6069 | mem_below_4gb_not_mirrored = true; | |
6070 | continue; | |
6071 | } | |
6072 | ||
6073 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
6074 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6075 | usable_startpfn; | |
6076 | } | |
6077 | ||
6078 | if (mem_below_4gb_not_mirrored) | |
6079 | pr_warn("This configuration results in unmirrored kernel memory."); | |
6080 | ||
6081 | goto out2; | |
6082 | } | |
6083 | ||
7e63efef | 6084 | /* |
b2f3eebe | 6085 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
6086 | * kernelcore that corresponds so that memory usable for |
6087 | * any allocation type is evenly spread. If both kernelcore | |
6088 | * and movablecore are specified, then the value of kernelcore | |
6089 | * will be used for required_kernelcore if it's greater than | |
6090 | * what movablecore would have allowed. | |
6091 | */ | |
6092 | if (required_movablecore) { | |
7e63efef MG |
6093 | unsigned long corepages; |
6094 | ||
6095 | /* | |
6096 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
6097 | * was requested by the user | |
6098 | */ | |
6099 | required_movablecore = | |
6100 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 6101 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
6102 | corepages = totalpages - required_movablecore; |
6103 | ||
6104 | required_kernelcore = max(required_kernelcore, corepages); | |
6105 | } | |
6106 | ||
bde304bd XQ |
6107 | /* |
6108 | * If kernelcore was not specified or kernelcore size is larger | |
6109 | * than totalpages, there is no ZONE_MOVABLE. | |
6110 | */ | |
6111 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 6112 | goto out; |
2a1e274a MG |
6113 | |
6114 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
6115 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6116 | ||
6117 | restart: | |
6118 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
6119 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 6120 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
6121 | unsigned long start_pfn, end_pfn; |
6122 | ||
2a1e274a MG |
6123 | /* |
6124 | * Recalculate kernelcore_node if the division per node | |
6125 | * now exceeds what is necessary to satisfy the requested | |
6126 | * amount of memory for the kernel | |
6127 | */ | |
6128 | if (required_kernelcore < kernelcore_node) | |
6129 | kernelcore_node = required_kernelcore / usable_nodes; | |
6130 | ||
6131 | /* | |
6132 | * As the map is walked, we track how much memory is usable | |
6133 | * by the kernel using kernelcore_remaining. When it is | |
6134 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
6135 | */ | |
6136 | kernelcore_remaining = kernelcore_node; | |
6137 | ||
6138 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 6139 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
6140 | unsigned long size_pages; |
6141 | ||
c13291a5 | 6142 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
6143 | if (start_pfn >= end_pfn) |
6144 | continue; | |
6145 | ||
6146 | /* Account for what is only usable for kernelcore */ | |
6147 | if (start_pfn < usable_startpfn) { | |
6148 | unsigned long kernel_pages; | |
6149 | kernel_pages = min(end_pfn, usable_startpfn) | |
6150 | - start_pfn; | |
6151 | ||
6152 | kernelcore_remaining -= min(kernel_pages, | |
6153 | kernelcore_remaining); | |
6154 | required_kernelcore -= min(kernel_pages, | |
6155 | required_kernelcore); | |
6156 | ||
6157 | /* Continue if range is now fully accounted */ | |
6158 | if (end_pfn <= usable_startpfn) { | |
6159 | ||
6160 | /* | |
6161 | * Push zone_movable_pfn to the end so | |
6162 | * that if we have to rebalance | |
6163 | * kernelcore across nodes, we will | |
6164 | * not double account here | |
6165 | */ | |
6166 | zone_movable_pfn[nid] = end_pfn; | |
6167 | continue; | |
6168 | } | |
6169 | start_pfn = usable_startpfn; | |
6170 | } | |
6171 | ||
6172 | /* | |
6173 | * The usable PFN range for ZONE_MOVABLE is from | |
6174 | * start_pfn->end_pfn. Calculate size_pages as the | |
6175 | * number of pages used as kernelcore | |
6176 | */ | |
6177 | size_pages = end_pfn - start_pfn; | |
6178 | if (size_pages > kernelcore_remaining) | |
6179 | size_pages = kernelcore_remaining; | |
6180 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6181 | ||
6182 | /* | |
6183 | * Some kernelcore has been met, update counts and | |
6184 | * break if the kernelcore for this node has been | |
b8af2941 | 6185 | * satisfied |
2a1e274a MG |
6186 | */ |
6187 | required_kernelcore -= min(required_kernelcore, | |
6188 | size_pages); | |
6189 | kernelcore_remaining -= size_pages; | |
6190 | if (!kernelcore_remaining) | |
6191 | break; | |
6192 | } | |
6193 | } | |
6194 | ||
6195 | /* | |
6196 | * If there is still required_kernelcore, we do another pass with one | |
6197 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6198 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6199 | * satisfied |
2a1e274a MG |
6200 | */ |
6201 | usable_nodes--; | |
6202 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6203 | goto restart; | |
6204 | ||
b2f3eebe | 6205 | out2: |
2a1e274a MG |
6206 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6207 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6208 | zone_movable_pfn[nid] = | |
6209 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6210 | |
20e6926d | 6211 | out: |
66918dcd | 6212 | /* restore the node_state */ |
4b0ef1fe | 6213 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6214 | } |
6215 | ||
4b0ef1fe LJ |
6216 | /* Any regular or high memory on that node ? */ |
6217 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6218 | { |
37b07e41 LS |
6219 | enum zone_type zone_type; |
6220 | ||
4b0ef1fe LJ |
6221 | if (N_MEMORY == N_NORMAL_MEMORY) |
6222 | return; | |
6223 | ||
6224 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 6225 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6226 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
6227 | node_set_state(nid, N_HIGH_MEMORY); |
6228 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
6229 | zone_type <= ZONE_NORMAL) | |
6230 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
6231 | break; |
6232 | } | |
37b07e41 | 6233 | } |
37b07e41 LS |
6234 | } |
6235 | ||
c713216d MG |
6236 | /** |
6237 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6238 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6239 | * |
6240 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6241 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6242 | * zone in each node and their holes is calculated. If the maximum PFN |
6243 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6244 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6245 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6246 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6247 | * at arch_max_dma_pfn. | |
6248 | */ | |
6249 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6250 | { | |
c13291a5 TH |
6251 | unsigned long start_pfn, end_pfn; |
6252 | int i, nid; | |
a6af2bc3 | 6253 | |
c713216d MG |
6254 | /* Record where the zone boundaries are */ |
6255 | memset(arch_zone_lowest_possible_pfn, 0, | |
6256 | sizeof(arch_zone_lowest_possible_pfn)); | |
6257 | memset(arch_zone_highest_possible_pfn, 0, | |
6258 | sizeof(arch_zone_highest_possible_pfn)); | |
6259 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
6260 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
6261 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6262 | if (i == ZONE_MOVABLE) |
6263 | continue; | |
c713216d MG |
6264 | arch_zone_lowest_possible_pfn[i] = |
6265 | arch_zone_highest_possible_pfn[i-1]; | |
6266 | arch_zone_highest_possible_pfn[i] = | |
6267 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
6268 | } | |
2a1e274a MG |
6269 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
6270 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
6271 | ||
6272 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6273 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6274 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6275 | |
c713216d | 6276 | /* Print out the zone ranges */ |
f88dfff5 | 6277 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6278 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6279 | if (i == ZONE_MOVABLE) | |
6280 | continue; | |
f88dfff5 | 6281 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6282 | if (arch_zone_lowest_possible_pfn[i] == |
6283 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6284 | pr_cont("empty\n"); |
72f0ba02 | 6285 | else |
8d29e18a JG |
6286 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6287 | (u64)arch_zone_lowest_possible_pfn[i] | |
6288 | << PAGE_SHIFT, | |
6289 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6290 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6291 | } |
6292 | ||
6293 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6294 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6295 | for (i = 0; i < MAX_NUMNODES; i++) { |
6296 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6297 | pr_info(" Node %d: %#018Lx\n", i, |
6298 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6299 | } |
c713216d | 6300 | |
f2d52fe5 | 6301 | /* Print out the early node map */ |
f88dfff5 | 6302 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6303 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6304 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6305 | (u64)start_pfn << PAGE_SHIFT, | |
6306 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6307 | |
6308 | /* Initialise every node */ | |
708614e6 | 6309 | mminit_verify_pageflags_layout(); |
8ef82866 | 6310 | setup_nr_node_ids(); |
c713216d MG |
6311 | for_each_online_node(nid) { |
6312 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6313 | free_area_init_node(nid, NULL, |
c713216d | 6314 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6315 | |
6316 | /* Any memory on that node */ | |
6317 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6318 | node_set_state(nid, N_MEMORY); |
6319 | check_for_memory(pgdat, nid); | |
c713216d MG |
6320 | } |
6321 | } | |
2a1e274a | 6322 | |
7e63efef | 6323 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
6324 | { |
6325 | unsigned long long coremem; | |
6326 | if (!p) | |
6327 | return -EINVAL; | |
6328 | ||
6329 | coremem = memparse(p, &p); | |
7e63efef | 6330 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 6331 | |
7e63efef | 6332 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
6333 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6334 | ||
6335 | return 0; | |
6336 | } | |
ed7ed365 | 6337 | |
7e63efef MG |
6338 | /* |
6339 | * kernelcore=size sets the amount of memory for use for allocations that | |
6340 | * cannot be reclaimed or migrated. | |
6341 | */ | |
6342 | static int __init cmdline_parse_kernelcore(char *p) | |
6343 | { | |
342332e6 TI |
6344 | /* parse kernelcore=mirror */ |
6345 | if (parse_option_str(p, "mirror")) { | |
6346 | mirrored_kernelcore = true; | |
6347 | return 0; | |
6348 | } | |
6349 | ||
7e63efef MG |
6350 | return cmdline_parse_core(p, &required_kernelcore); |
6351 | } | |
6352 | ||
6353 | /* | |
6354 | * movablecore=size sets the amount of memory for use for allocations that | |
6355 | * can be reclaimed or migrated. | |
6356 | */ | |
6357 | static int __init cmdline_parse_movablecore(char *p) | |
6358 | { | |
6359 | return cmdline_parse_core(p, &required_movablecore); | |
6360 | } | |
6361 | ||
ed7ed365 | 6362 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6363 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6364 | |
0ee332c1 | 6365 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6366 | |
c3d5f5f0 JL |
6367 | void adjust_managed_page_count(struct page *page, long count) |
6368 | { | |
6369 | spin_lock(&managed_page_count_lock); | |
6370 | page_zone(page)->managed_pages += count; | |
6371 | totalram_pages += count; | |
3dcc0571 JL |
6372 | #ifdef CONFIG_HIGHMEM |
6373 | if (PageHighMem(page)) | |
6374 | totalhigh_pages += count; | |
6375 | #endif | |
c3d5f5f0 JL |
6376 | spin_unlock(&managed_page_count_lock); |
6377 | } | |
3dcc0571 | 6378 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6379 | |
11199692 | 6380 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6381 | { |
11199692 JL |
6382 | void *pos; |
6383 | unsigned long pages = 0; | |
69afade7 | 6384 | |
11199692 JL |
6385 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6386 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6387 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 6388 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
6389 | memset(pos, poison, PAGE_SIZE); |
6390 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
6391 | } |
6392 | ||
6393 | if (pages && s) | |
11199692 | 6394 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
6395 | s, pages << (PAGE_SHIFT - 10), start, end); |
6396 | ||
6397 | return pages; | |
6398 | } | |
11199692 | 6399 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6400 | |
cfa11e08 JL |
6401 | #ifdef CONFIG_HIGHMEM |
6402 | void free_highmem_page(struct page *page) | |
6403 | { | |
6404 | __free_reserved_page(page); | |
6405 | totalram_pages++; | |
7b4b2a0d | 6406 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6407 | totalhigh_pages++; |
6408 | } | |
6409 | #endif | |
6410 | ||
7ee3d4e8 JL |
6411 | |
6412 | void __init mem_init_print_info(const char *str) | |
6413 | { | |
6414 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
6415 | unsigned long init_code_size, init_data_size; | |
6416 | ||
6417 | physpages = get_num_physpages(); | |
6418 | codesize = _etext - _stext; | |
6419 | datasize = _edata - _sdata; | |
6420 | rosize = __end_rodata - __start_rodata; | |
6421 | bss_size = __bss_stop - __bss_start; | |
6422 | init_data_size = __init_end - __init_begin; | |
6423 | init_code_size = _einittext - _sinittext; | |
6424 | ||
6425 | /* | |
6426 | * Detect special cases and adjust section sizes accordingly: | |
6427 | * 1) .init.* may be embedded into .data sections | |
6428 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
6429 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
6430 | * 3) .rodata.* may be embedded into .text or .data sections. | |
6431 | */ | |
6432 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
6433 | do { \ |
6434 | if (start <= pos && pos < end && size > adj) \ | |
6435 | size -= adj; \ | |
6436 | } while (0) | |
7ee3d4e8 JL |
6437 | |
6438 | adj_init_size(__init_begin, __init_end, init_data_size, | |
6439 | _sinittext, init_code_size); | |
6440 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
6441 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
6442 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
6443 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
6444 | ||
6445 | #undef adj_init_size | |
6446 | ||
756a025f | 6447 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 6448 | #ifdef CONFIG_HIGHMEM |
756a025f | 6449 | ", %luK highmem" |
7ee3d4e8 | 6450 | #endif |
756a025f JP |
6451 | "%s%s)\n", |
6452 | nr_free_pages() << (PAGE_SHIFT - 10), | |
6453 | physpages << (PAGE_SHIFT - 10), | |
6454 | codesize >> 10, datasize >> 10, rosize >> 10, | |
6455 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
6456 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
6457 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 6458 | #ifdef CONFIG_HIGHMEM |
756a025f | 6459 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 6460 | #endif |
756a025f | 6461 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
6462 | } |
6463 | ||
0e0b864e | 6464 | /** |
88ca3b94 RD |
6465 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6466 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 6467 | * |
013110a7 | 6468 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
6469 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6470 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
6471 | * function may optionally be used to account for unfreeable pages in the |
6472 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
6473 | * smaller per-cpu batchsize. | |
0e0b864e MG |
6474 | */ |
6475 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
6476 | { | |
6477 | dma_reserve = new_dma_reserve; | |
6478 | } | |
6479 | ||
1da177e4 LT |
6480 | void __init free_area_init(unsigned long *zones_size) |
6481 | { | |
9109fb7b | 6482 | free_area_init_node(0, zones_size, |
1da177e4 LT |
6483 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6484 | } | |
1da177e4 | 6485 | |
1da177e4 LT |
6486 | static int page_alloc_cpu_notify(struct notifier_block *self, |
6487 | unsigned long action, void *hcpu) | |
6488 | { | |
6489 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 6490 | |
8bb78442 | 6491 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 6492 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
6493 | drain_pages(cpu); |
6494 | ||
6495 | /* | |
6496 | * Spill the event counters of the dead processor | |
6497 | * into the current processors event counters. | |
6498 | * This artificially elevates the count of the current | |
6499 | * processor. | |
6500 | */ | |
f8891e5e | 6501 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
6502 | |
6503 | /* | |
6504 | * Zero the differential counters of the dead processor | |
6505 | * so that the vm statistics are consistent. | |
6506 | * | |
6507 | * This is only okay since the processor is dead and cannot | |
6508 | * race with what we are doing. | |
6509 | */ | |
2bb921e5 | 6510 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
6511 | } |
6512 | return NOTIFY_OK; | |
6513 | } | |
1da177e4 LT |
6514 | |
6515 | void __init page_alloc_init(void) | |
6516 | { | |
6517 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
6518 | } | |
6519 | ||
cb45b0e9 | 6520 | /* |
34b10060 | 6521 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
6522 | * or min_free_kbytes changes. |
6523 | */ | |
6524 | static void calculate_totalreserve_pages(void) | |
6525 | { | |
6526 | struct pglist_data *pgdat; | |
6527 | unsigned long reserve_pages = 0; | |
2f6726e5 | 6528 | enum zone_type i, j; |
cb45b0e9 HA |
6529 | |
6530 | for_each_online_pgdat(pgdat) { | |
6531 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
6532 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 6533 | long max = 0; |
cb45b0e9 HA |
6534 | |
6535 | /* Find valid and maximum lowmem_reserve in the zone */ | |
6536 | for (j = i; j < MAX_NR_ZONES; j++) { | |
6537 | if (zone->lowmem_reserve[j] > max) | |
6538 | max = zone->lowmem_reserve[j]; | |
6539 | } | |
6540 | ||
41858966 MG |
6541 | /* we treat the high watermark as reserved pages. */ |
6542 | max += high_wmark_pages(zone); | |
cb45b0e9 | 6543 | |
b40da049 JL |
6544 | if (max > zone->managed_pages) |
6545 | max = zone->managed_pages; | |
a8d01437 JW |
6546 | |
6547 | zone->totalreserve_pages = max; | |
6548 | ||
cb45b0e9 HA |
6549 | reserve_pages += max; |
6550 | } | |
6551 | } | |
6552 | totalreserve_pages = reserve_pages; | |
6553 | } | |
6554 | ||
1da177e4 LT |
6555 | /* |
6556 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 6557 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
6558 | * has a correct pages reserved value, so an adequate number of |
6559 | * pages are left in the zone after a successful __alloc_pages(). | |
6560 | */ | |
6561 | static void setup_per_zone_lowmem_reserve(void) | |
6562 | { | |
6563 | struct pglist_data *pgdat; | |
2f6726e5 | 6564 | enum zone_type j, idx; |
1da177e4 | 6565 | |
ec936fc5 | 6566 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
6567 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6568 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 6569 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
6570 | |
6571 | zone->lowmem_reserve[j] = 0; | |
6572 | ||
2f6726e5 CL |
6573 | idx = j; |
6574 | while (idx) { | |
1da177e4 LT |
6575 | struct zone *lower_zone; |
6576 | ||
2f6726e5 CL |
6577 | idx--; |
6578 | ||
1da177e4 LT |
6579 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6580 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
6581 | ||
6582 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 6583 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 6584 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 6585 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
6586 | } |
6587 | } | |
6588 | } | |
cb45b0e9 HA |
6589 | |
6590 | /* update totalreserve_pages */ | |
6591 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6592 | } |
6593 | ||
cfd3da1e | 6594 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
6595 | { |
6596 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
6597 | unsigned long lowmem_pages = 0; | |
6598 | struct zone *zone; | |
6599 | unsigned long flags; | |
6600 | ||
6601 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
6602 | for_each_zone(zone) { | |
6603 | if (!is_highmem(zone)) | |
b40da049 | 6604 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
6605 | } |
6606 | ||
6607 | for_each_zone(zone) { | |
ac924c60 AM |
6608 | u64 tmp; |
6609 | ||
1125b4e3 | 6610 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 6611 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 6612 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
6613 | if (is_highmem(zone)) { |
6614 | /* | |
669ed175 NP |
6615 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6616 | * need highmem pages, so cap pages_min to a small | |
6617 | * value here. | |
6618 | * | |
41858966 | 6619 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 6620 | * deltas control asynch page reclaim, and so should |
669ed175 | 6621 | * not be capped for highmem. |
1da177e4 | 6622 | */ |
90ae8d67 | 6623 | unsigned long min_pages; |
1da177e4 | 6624 | |
b40da049 | 6625 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 6626 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 6627 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 6628 | } else { |
669ed175 NP |
6629 | /* |
6630 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
6631 | * proportionate to the zone's size. |
6632 | */ | |
41858966 | 6633 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
6634 | } |
6635 | ||
795ae7a0 JW |
6636 | /* |
6637 | * Set the kswapd watermarks distance according to the | |
6638 | * scale factor in proportion to available memory, but | |
6639 | * ensure a minimum size on small systems. | |
6640 | */ | |
6641 | tmp = max_t(u64, tmp >> 2, | |
6642 | mult_frac(zone->managed_pages, | |
6643 | watermark_scale_factor, 10000)); | |
6644 | ||
6645 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
6646 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 6647 | |
81c0a2bb | 6648 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
abe5f972 JW |
6649 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
6650 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
81c0a2bb | 6651 | |
1125b4e3 | 6652 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 6653 | } |
cb45b0e9 HA |
6654 | |
6655 | /* update totalreserve_pages */ | |
6656 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6657 | } |
6658 | ||
cfd3da1e MG |
6659 | /** |
6660 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
6661 | * or when memory is hot-{added|removed} | |
6662 | * | |
6663 | * Ensures that the watermark[min,low,high] values for each zone are set | |
6664 | * correctly with respect to min_free_kbytes. | |
6665 | */ | |
6666 | void setup_per_zone_wmarks(void) | |
6667 | { | |
6668 | mutex_lock(&zonelists_mutex); | |
6669 | __setup_per_zone_wmarks(); | |
6670 | mutex_unlock(&zonelists_mutex); | |
6671 | } | |
6672 | ||
55a4462a | 6673 | /* |
556adecb RR |
6674 | * The inactive anon list should be small enough that the VM never has to |
6675 | * do too much work, but large enough that each inactive page has a chance | |
6676 | * to be referenced again before it is swapped out. | |
6677 | * | |
6678 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
6679 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
6680 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
6681 | * the anonymous pages are kept on the inactive list. | |
6682 | * | |
6683 | * total target max | |
6684 | * memory ratio inactive anon | |
6685 | * ------------------------------------- | |
6686 | * 10MB 1 5MB | |
6687 | * 100MB 1 50MB | |
6688 | * 1GB 3 250MB | |
6689 | * 10GB 10 0.9GB | |
6690 | * 100GB 31 3GB | |
6691 | * 1TB 101 10GB | |
6692 | * 10TB 320 32GB | |
6693 | */ | |
1b79acc9 | 6694 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 6695 | { |
96cb4df5 | 6696 | unsigned int gb, ratio; |
556adecb | 6697 | |
96cb4df5 | 6698 | /* Zone size in gigabytes */ |
b40da049 | 6699 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 6700 | if (gb) |
556adecb | 6701 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
6702 | else |
6703 | ratio = 1; | |
556adecb | 6704 | |
96cb4df5 MK |
6705 | zone->inactive_ratio = ratio; |
6706 | } | |
556adecb | 6707 | |
839a4fcc | 6708 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
6709 | { |
6710 | struct zone *zone; | |
6711 | ||
6712 | for_each_zone(zone) | |
6713 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
6714 | } |
6715 | ||
1da177e4 LT |
6716 | /* |
6717 | * Initialise min_free_kbytes. | |
6718 | * | |
6719 | * For small machines we want it small (128k min). For large machines | |
6720 | * we want it large (64MB max). But it is not linear, because network | |
6721 | * bandwidth does not increase linearly with machine size. We use | |
6722 | * | |
b8af2941 | 6723 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6724 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6725 | * | |
6726 | * which yields | |
6727 | * | |
6728 | * 16MB: 512k | |
6729 | * 32MB: 724k | |
6730 | * 64MB: 1024k | |
6731 | * 128MB: 1448k | |
6732 | * 256MB: 2048k | |
6733 | * 512MB: 2896k | |
6734 | * 1024MB: 4096k | |
6735 | * 2048MB: 5792k | |
6736 | * 4096MB: 8192k | |
6737 | * 8192MB: 11584k | |
6738 | * 16384MB: 16384k | |
6739 | */ | |
1b79acc9 | 6740 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6741 | { |
6742 | unsigned long lowmem_kbytes; | |
5f12733e | 6743 | int new_min_free_kbytes; |
1da177e4 LT |
6744 | |
6745 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6746 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6747 | ||
6748 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6749 | min_free_kbytes = new_min_free_kbytes; | |
6750 | if (min_free_kbytes < 128) | |
6751 | min_free_kbytes = 128; | |
6752 | if (min_free_kbytes > 65536) | |
6753 | min_free_kbytes = 65536; | |
6754 | } else { | |
6755 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6756 | new_min_free_kbytes, user_min_free_kbytes); | |
6757 | } | |
bc75d33f | 6758 | setup_per_zone_wmarks(); |
a6cccdc3 | 6759 | refresh_zone_stat_thresholds(); |
1da177e4 | 6760 | setup_per_zone_lowmem_reserve(); |
556adecb | 6761 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
6762 | return 0; |
6763 | } | |
bc22af74 | 6764 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
6765 | |
6766 | /* | |
b8af2941 | 6767 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6768 | * that we can call two helper functions whenever min_free_kbytes |
6769 | * changes. | |
6770 | */ | |
cccad5b9 | 6771 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6772 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6773 | { |
da8c757b HP |
6774 | int rc; |
6775 | ||
6776 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6777 | if (rc) | |
6778 | return rc; | |
6779 | ||
5f12733e MH |
6780 | if (write) { |
6781 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6782 | setup_per_zone_wmarks(); |
5f12733e | 6783 | } |
1da177e4 LT |
6784 | return 0; |
6785 | } | |
6786 | ||
795ae7a0 JW |
6787 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
6788 | void __user *buffer, size_t *length, loff_t *ppos) | |
6789 | { | |
6790 | int rc; | |
6791 | ||
6792 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6793 | if (rc) | |
6794 | return rc; | |
6795 | ||
6796 | if (write) | |
6797 | setup_per_zone_wmarks(); | |
6798 | ||
6799 | return 0; | |
6800 | } | |
6801 | ||
9614634f | 6802 | #ifdef CONFIG_NUMA |
cccad5b9 | 6803 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6804 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
6805 | { |
6806 | struct zone *zone; | |
6807 | int rc; | |
6808 | ||
8d65af78 | 6809 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
6810 | if (rc) |
6811 | return rc; | |
6812 | ||
6813 | for_each_zone(zone) | |
b40da049 | 6814 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
6815 | sysctl_min_unmapped_ratio) / 100; |
6816 | return 0; | |
6817 | } | |
0ff38490 | 6818 | |
cccad5b9 | 6819 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6820 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
6821 | { |
6822 | struct zone *zone; | |
6823 | int rc; | |
6824 | ||
8d65af78 | 6825 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
6826 | if (rc) |
6827 | return rc; | |
6828 | ||
6829 | for_each_zone(zone) | |
b40da049 | 6830 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
6831 | sysctl_min_slab_ratio) / 100; |
6832 | return 0; | |
6833 | } | |
9614634f CL |
6834 | #endif |
6835 | ||
1da177e4 LT |
6836 | /* |
6837 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
6838 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
6839 | * whenever sysctl_lowmem_reserve_ratio changes. | |
6840 | * | |
6841 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 6842 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
6843 | * if in function of the boot time zone sizes. |
6844 | */ | |
cccad5b9 | 6845 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6846 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6847 | { |
8d65af78 | 6848 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
6849 | setup_per_zone_lowmem_reserve(); |
6850 | return 0; | |
6851 | } | |
6852 | ||
8ad4b1fb RS |
6853 | /* |
6854 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
6855 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
6856 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 6857 | */ |
cccad5b9 | 6858 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6859 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
6860 | { |
6861 | struct zone *zone; | |
7cd2b0a3 | 6862 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
6863 | int ret; |
6864 | ||
7cd2b0a3 DR |
6865 | mutex_lock(&pcp_batch_high_lock); |
6866 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
6867 | ||
8d65af78 | 6868 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
6869 | if (!write || ret < 0) |
6870 | goto out; | |
6871 | ||
6872 | /* Sanity checking to avoid pcp imbalance */ | |
6873 | if (percpu_pagelist_fraction && | |
6874 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
6875 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
6876 | ret = -EINVAL; | |
6877 | goto out; | |
6878 | } | |
6879 | ||
6880 | /* No change? */ | |
6881 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
6882 | goto out; | |
c8e251fa | 6883 | |
364df0eb | 6884 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
6885 | unsigned int cpu; |
6886 | ||
22a7f12b | 6887 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
6888 | pageset_set_high_and_batch(zone, |
6889 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 6890 | } |
7cd2b0a3 | 6891 | out: |
c8e251fa | 6892 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6893 | return ret; |
8ad4b1fb RS |
6894 | } |
6895 | ||
a9919c79 | 6896 | #ifdef CONFIG_NUMA |
f034b5d4 | 6897 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 6898 | |
1da177e4 LT |
6899 | static int __init set_hashdist(char *str) |
6900 | { | |
6901 | if (!str) | |
6902 | return 0; | |
6903 | hashdist = simple_strtoul(str, &str, 0); | |
6904 | return 1; | |
6905 | } | |
6906 | __setup("hashdist=", set_hashdist); | |
6907 | #endif | |
6908 | ||
6909 | /* | |
6910 | * allocate a large system hash table from bootmem | |
6911 | * - it is assumed that the hash table must contain an exact power-of-2 | |
6912 | * quantity of entries | |
6913 | * - limit is the number of hash buckets, not the total allocation size | |
6914 | */ | |
6915 | void *__init alloc_large_system_hash(const char *tablename, | |
6916 | unsigned long bucketsize, | |
6917 | unsigned long numentries, | |
6918 | int scale, | |
6919 | int flags, | |
6920 | unsigned int *_hash_shift, | |
6921 | unsigned int *_hash_mask, | |
31fe62b9 TB |
6922 | unsigned long low_limit, |
6923 | unsigned long high_limit) | |
1da177e4 | 6924 | { |
31fe62b9 | 6925 | unsigned long long max = high_limit; |
1da177e4 LT |
6926 | unsigned long log2qty, size; |
6927 | void *table = NULL; | |
6928 | ||
6929 | /* allow the kernel cmdline to have a say */ | |
6930 | if (!numentries) { | |
6931 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 6932 | numentries = nr_kernel_pages; |
a7e83318 JZ |
6933 | |
6934 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
6935 | if (PAGE_SHIFT < 20) | |
6936 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
6937 | |
6938 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
6939 | if (scale > PAGE_SHIFT) | |
6940 | numentries >>= (scale - PAGE_SHIFT); | |
6941 | else | |
6942 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
6943 | |
6944 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
6945 | if (unlikely(flags & HASH_SMALL)) { |
6946 | /* Makes no sense without HASH_EARLY */ | |
6947 | WARN_ON(!(flags & HASH_EARLY)); | |
6948 | if (!(numentries >> *_hash_shift)) { | |
6949 | numentries = 1UL << *_hash_shift; | |
6950 | BUG_ON(!numentries); | |
6951 | } | |
6952 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 6953 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 6954 | } |
6e692ed3 | 6955 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
6956 | |
6957 | /* limit allocation size to 1/16 total memory by default */ | |
6958 | if (max == 0) { | |
6959 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
6960 | do_div(max, bucketsize); | |
6961 | } | |
074b8517 | 6962 | max = min(max, 0x80000000ULL); |
1da177e4 | 6963 | |
31fe62b9 TB |
6964 | if (numentries < low_limit) |
6965 | numentries = low_limit; | |
1da177e4 LT |
6966 | if (numentries > max) |
6967 | numentries = max; | |
6968 | ||
f0d1b0b3 | 6969 | log2qty = ilog2(numentries); |
1da177e4 LT |
6970 | |
6971 | do { | |
6972 | size = bucketsize << log2qty; | |
6973 | if (flags & HASH_EARLY) | |
6782832e | 6974 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
6975 | else if (hashdist) |
6976 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
6977 | else { | |
1037b83b ED |
6978 | /* |
6979 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
6980 | * some pages at the end of hash table which |
6981 | * alloc_pages_exact() automatically does | |
1037b83b | 6982 | */ |
264ef8a9 | 6983 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 6984 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
6985 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
6986 | } | |
1da177e4 LT |
6987 | } |
6988 | } while (!table && size > PAGE_SIZE && --log2qty); | |
6989 | ||
6990 | if (!table) | |
6991 | panic("Failed to allocate %s hash table\n", tablename); | |
6992 | ||
1170532b JP |
6993 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
6994 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
6995 | |
6996 | if (_hash_shift) | |
6997 | *_hash_shift = log2qty; | |
6998 | if (_hash_mask) | |
6999 | *_hash_mask = (1 << log2qty) - 1; | |
7000 | ||
7001 | return table; | |
7002 | } | |
a117e66e | 7003 | |
a5d76b54 | 7004 | /* |
80934513 MK |
7005 | * This function checks whether pageblock includes unmovable pages or not. |
7006 | * If @count is not zero, it is okay to include less @count unmovable pages | |
7007 | * | |
b8af2941 | 7008 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
7009 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
7010 | * expect this function should be exact. | |
a5d76b54 | 7011 | */ |
b023f468 WC |
7012 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
7013 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
7014 | { |
7015 | unsigned long pfn, iter, found; | |
47118af0 MN |
7016 | int mt; |
7017 | ||
49ac8255 KH |
7018 | /* |
7019 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 7020 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
7021 | */ |
7022 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 7023 | return false; |
47118af0 MN |
7024 | mt = get_pageblock_migratetype(page); |
7025 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 7026 | return false; |
49ac8255 KH |
7027 | |
7028 | pfn = page_to_pfn(page); | |
7029 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
7030 | unsigned long check = pfn + iter; | |
7031 | ||
29723fcc | 7032 | if (!pfn_valid_within(check)) |
49ac8255 | 7033 | continue; |
29723fcc | 7034 | |
49ac8255 | 7035 | page = pfn_to_page(check); |
c8721bbb NH |
7036 | |
7037 | /* | |
7038 | * Hugepages are not in LRU lists, but they're movable. | |
7039 | * We need not scan over tail pages bacause we don't | |
7040 | * handle each tail page individually in migration. | |
7041 | */ | |
7042 | if (PageHuge(page)) { | |
7043 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
7044 | continue; | |
7045 | } | |
7046 | ||
97d255c8 MK |
7047 | /* |
7048 | * We can't use page_count without pin a page | |
7049 | * because another CPU can free compound page. | |
7050 | * This check already skips compound tails of THP | |
0139aa7b | 7051 | * because their page->_refcount is zero at all time. |
97d255c8 | 7052 | */ |
fe896d18 | 7053 | if (!page_ref_count(page)) { |
49ac8255 KH |
7054 | if (PageBuddy(page)) |
7055 | iter += (1 << page_order(page)) - 1; | |
7056 | continue; | |
7057 | } | |
97d255c8 | 7058 | |
b023f468 WC |
7059 | /* |
7060 | * The HWPoisoned page may be not in buddy system, and | |
7061 | * page_count() is not 0. | |
7062 | */ | |
7063 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
7064 | continue; | |
7065 | ||
49ac8255 KH |
7066 | if (!PageLRU(page)) |
7067 | found++; | |
7068 | /* | |
6b4f7799 JW |
7069 | * If there are RECLAIMABLE pages, we need to check |
7070 | * it. But now, memory offline itself doesn't call | |
7071 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
7072 | */ |
7073 | /* | |
7074 | * If the page is not RAM, page_count()should be 0. | |
7075 | * we don't need more check. This is an _used_ not-movable page. | |
7076 | * | |
7077 | * The problematic thing here is PG_reserved pages. PG_reserved | |
7078 | * is set to both of a memory hole page and a _used_ kernel | |
7079 | * page at boot. | |
7080 | */ | |
7081 | if (found > count) | |
80934513 | 7082 | return true; |
49ac8255 | 7083 | } |
80934513 | 7084 | return false; |
49ac8255 KH |
7085 | } |
7086 | ||
7087 | bool is_pageblock_removable_nolock(struct page *page) | |
7088 | { | |
656a0706 MH |
7089 | struct zone *zone; |
7090 | unsigned long pfn; | |
687875fb MH |
7091 | |
7092 | /* | |
7093 | * We have to be careful here because we are iterating over memory | |
7094 | * sections which are not zone aware so we might end up outside of | |
7095 | * the zone but still within the section. | |
656a0706 MH |
7096 | * We have to take care about the node as well. If the node is offline |
7097 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 7098 | */ |
656a0706 MH |
7099 | if (!node_online(page_to_nid(page))) |
7100 | return false; | |
7101 | ||
7102 | zone = page_zone(page); | |
7103 | pfn = page_to_pfn(page); | |
108bcc96 | 7104 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
7105 | return false; |
7106 | ||
b023f468 | 7107 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 7108 | } |
0c0e6195 | 7109 | |
080fe206 | 7110 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7111 | |
7112 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7113 | { | |
7114 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7115 | pageblock_nr_pages) - 1); | |
7116 | } | |
7117 | ||
7118 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7119 | { | |
7120 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7121 | pageblock_nr_pages)); | |
7122 | } | |
7123 | ||
041d3a8c | 7124 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7125 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7126 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7127 | { |
7128 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7129 | unsigned long nr_reclaimed; |
041d3a8c MN |
7130 | unsigned long pfn = start; |
7131 | unsigned int tries = 0; | |
7132 | int ret = 0; | |
7133 | ||
be49a6e1 | 7134 | migrate_prep(); |
041d3a8c | 7135 | |
bb13ffeb | 7136 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7137 | if (fatal_signal_pending(current)) { |
7138 | ret = -EINTR; | |
7139 | break; | |
7140 | } | |
7141 | ||
bb13ffeb MG |
7142 | if (list_empty(&cc->migratepages)) { |
7143 | cc->nr_migratepages = 0; | |
edc2ca61 | 7144 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7145 | if (!pfn) { |
7146 | ret = -EINTR; | |
7147 | break; | |
7148 | } | |
7149 | tries = 0; | |
7150 | } else if (++tries == 5) { | |
7151 | ret = ret < 0 ? ret : -EBUSY; | |
7152 | break; | |
7153 | } | |
7154 | ||
beb51eaa MK |
7155 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7156 | &cc->migratepages); | |
7157 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7158 | |
9c620e2b | 7159 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 7160 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 7161 | } |
2a6f5124 SP |
7162 | if (ret < 0) { |
7163 | putback_movable_pages(&cc->migratepages); | |
7164 | return ret; | |
7165 | } | |
7166 | return 0; | |
041d3a8c MN |
7167 | } |
7168 | ||
7169 | /** | |
7170 | * alloc_contig_range() -- tries to allocate given range of pages | |
7171 | * @start: start PFN to allocate | |
7172 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7173 | * @migratetype: migratetype of the underlaying pageblocks (either |
7174 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7175 | * in range must have the same migratetype and it must | |
7176 | * be either of the two. | |
041d3a8c MN |
7177 | * |
7178 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
7179 | * aligned, however it's the caller's responsibility to guarantee that | |
7180 | * we are the only thread that changes migrate type of pageblocks the | |
7181 | * pages fall in. | |
7182 | * | |
7183 | * The PFN range must belong to a single zone. | |
7184 | * | |
7185 | * Returns zero on success or negative error code. On success all | |
7186 | * pages which PFN is in [start, end) are allocated for the caller and | |
7187 | * need to be freed with free_contig_range(). | |
7188 | */ | |
0815f3d8 MN |
7189 | int alloc_contig_range(unsigned long start, unsigned long end, |
7190 | unsigned migratetype) | |
041d3a8c | 7191 | { |
041d3a8c | 7192 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7193 | unsigned int order; |
7194 | int ret = 0; | |
041d3a8c | 7195 | |
bb13ffeb MG |
7196 | struct compact_control cc = { |
7197 | .nr_migratepages = 0, | |
7198 | .order = -1, | |
7199 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7200 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
7201 | .ignore_skip_hint = true, |
7202 | }; | |
7203 | INIT_LIST_HEAD(&cc.migratepages); | |
7204 | ||
041d3a8c MN |
7205 | /* |
7206 | * What we do here is we mark all pageblocks in range as | |
7207 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7208 | * have different sizes, and due to the way page allocator | |
7209 | * work, we align the range to biggest of the two pages so | |
7210 | * that page allocator won't try to merge buddies from | |
7211 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7212 | * other migration type. | |
7213 | * | |
7214 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7215 | * migrate the pages from an unaligned range (ie. pages that | |
7216 | * we are interested in). This will put all the pages in | |
7217 | * range back to page allocator as MIGRATE_ISOLATE. | |
7218 | * | |
7219 | * When this is done, we take the pages in range from page | |
7220 | * allocator removing them from the buddy system. This way | |
7221 | * page allocator will never consider using them. | |
7222 | * | |
7223 | * This lets us mark the pageblocks back as | |
7224 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7225 | * aligned range but not in the unaligned, original range are | |
7226 | * put back to page allocator so that buddy can use them. | |
7227 | */ | |
7228 | ||
7229 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7230 | pfn_max_align_up(end), migratetype, |
7231 | false); | |
041d3a8c | 7232 | if (ret) |
86a595f9 | 7233 | return ret; |
041d3a8c | 7234 | |
8ef5849f JK |
7235 | /* |
7236 | * In case of -EBUSY, we'd like to know which page causes problem. | |
7237 | * So, just fall through. We will check it in test_pages_isolated(). | |
7238 | */ | |
bb13ffeb | 7239 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7240 | if (ret && ret != -EBUSY) |
041d3a8c MN |
7241 | goto done; |
7242 | ||
7243 | /* | |
7244 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7245 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7246 | * more, all pages in [start, end) are free in page allocator. | |
7247 | * What we are going to do is to allocate all pages from | |
7248 | * [start, end) (that is remove them from page allocator). | |
7249 | * | |
7250 | * The only problem is that pages at the beginning and at the | |
7251 | * end of interesting range may be not aligned with pages that | |
7252 | * page allocator holds, ie. they can be part of higher order | |
7253 | * pages. Because of this, we reserve the bigger range and | |
7254 | * once this is done free the pages we are not interested in. | |
7255 | * | |
7256 | * We don't have to hold zone->lock here because the pages are | |
7257 | * isolated thus they won't get removed from buddy. | |
7258 | */ | |
7259 | ||
7260 | lru_add_drain_all(); | |
510f5507 | 7261 | drain_all_pages(cc.zone); |
041d3a8c MN |
7262 | |
7263 | order = 0; | |
7264 | outer_start = start; | |
7265 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7266 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7267 | outer_start = start; |
7268 | break; | |
041d3a8c MN |
7269 | } |
7270 | outer_start &= ~0UL << order; | |
7271 | } | |
7272 | ||
8ef5849f JK |
7273 | if (outer_start != start) { |
7274 | order = page_order(pfn_to_page(outer_start)); | |
7275 | ||
7276 | /* | |
7277 | * outer_start page could be small order buddy page and | |
7278 | * it doesn't include start page. Adjust outer_start | |
7279 | * in this case to report failed page properly | |
7280 | * on tracepoint in test_pages_isolated() | |
7281 | */ | |
7282 | if (outer_start + (1UL << order) <= start) | |
7283 | outer_start = start; | |
7284 | } | |
7285 | ||
041d3a8c | 7286 | /* Make sure the range is really isolated. */ |
b023f468 | 7287 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
7288 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
7289 | __func__, outer_start, end); | |
041d3a8c MN |
7290 | ret = -EBUSY; |
7291 | goto done; | |
7292 | } | |
7293 | ||
49f223a9 | 7294 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7295 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7296 | if (!outer_end) { |
7297 | ret = -EBUSY; | |
7298 | goto done; | |
7299 | } | |
7300 | ||
7301 | /* Free head and tail (if any) */ | |
7302 | if (start != outer_start) | |
7303 | free_contig_range(outer_start, start - outer_start); | |
7304 | if (end != outer_end) | |
7305 | free_contig_range(end, outer_end - end); | |
7306 | ||
7307 | done: | |
7308 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7309 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7310 | return ret; |
7311 | } | |
7312 | ||
7313 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7314 | { | |
bcc2b02f MS |
7315 | unsigned int count = 0; |
7316 | ||
7317 | for (; nr_pages--; pfn++) { | |
7318 | struct page *page = pfn_to_page(pfn); | |
7319 | ||
7320 | count += page_count(page) != 1; | |
7321 | __free_page(page); | |
7322 | } | |
7323 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7324 | } |
7325 | #endif | |
7326 | ||
4ed7e022 | 7327 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7328 | /* |
7329 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7330 | * page high values need to be recalulated. | |
7331 | */ | |
4ed7e022 JL |
7332 | void __meminit zone_pcp_update(struct zone *zone) |
7333 | { | |
0a647f38 | 7334 | unsigned cpu; |
c8e251fa | 7335 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7336 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7337 | pageset_set_high_and_batch(zone, |
7338 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7339 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7340 | } |
7341 | #endif | |
7342 | ||
340175b7 JL |
7343 | void zone_pcp_reset(struct zone *zone) |
7344 | { | |
7345 | unsigned long flags; | |
5a883813 MK |
7346 | int cpu; |
7347 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7348 | |
7349 | /* avoid races with drain_pages() */ | |
7350 | local_irq_save(flags); | |
7351 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7352 | for_each_online_cpu(cpu) { |
7353 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7354 | drain_zonestat(zone, pset); | |
7355 | } | |
340175b7 JL |
7356 | free_percpu(zone->pageset); |
7357 | zone->pageset = &boot_pageset; | |
7358 | } | |
7359 | local_irq_restore(flags); | |
7360 | } | |
7361 | ||
6dcd73d7 | 7362 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 7363 | /* |
b9eb6319 JK |
7364 | * All pages in the range must be in a single zone and isolated |
7365 | * before calling this. | |
0c0e6195 KH |
7366 | */ |
7367 | void | |
7368 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7369 | { | |
7370 | struct page *page; | |
7371 | struct zone *zone; | |
7aeb09f9 | 7372 | unsigned int order, i; |
0c0e6195 KH |
7373 | unsigned long pfn; |
7374 | unsigned long flags; | |
7375 | /* find the first valid pfn */ | |
7376 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
7377 | if (pfn_valid(pfn)) | |
7378 | break; | |
7379 | if (pfn == end_pfn) | |
7380 | return; | |
7381 | zone = page_zone(pfn_to_page(pfn)); | |
7382 | spin_lock_irqsave(&zone->lock, flags); | |
7383 | pfn = start_pfn; | |
7384 | while (pfn < end_pfn) { | |
7385 | if (!pfn_valid(pfn)) { | |
7386 | pfn++; | |
7387 | continue; | |
7388 | } | |
7389 | page = pfn_to_page(pfn); | |
b023f468 WC |
7390 | /* |
7391 | * The HWPoisoned page may be not in buddy system, and | |
7392 | * page_count() is not 0. | |
7393 | */ | |
7394 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
7395 | pfn++; | |
7396 | SetPageReserved(page); | |
7397 | continue; | |
7398 | } | |
7399 | ||
0c0e6195 KH |
7400 | BUG_ON(page_count(page)); |
7401 | BUG_ON(!PageBuddy(page)); | |
7402 | order = page_order(page); | |
7403 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
7404 | pr_info("remove from free list %lx %d %lx\n", |
7405 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
7406 | #endif |
7407 | list_del(&page->lru); | |
7408 | rmv_page_order(page); | |
7409 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
7410 | for (i = 0; i < (1 << order); i++) |
7411 | SetPageReserved((page+i)); | |
7412 | pfn += (1 << order); | |
7413 | } | |
7414 | spin_unlock_irqrestore(&zone->lock, flags); | |
7415 | } | |
7416 | #endif | |
8d22ba1b | 7417 | |
8d22ba1b WF |
7418 | bool is_free_buddy_page(struct page *page) |
7419 | { | |
7420 | struct zone *zone = page_zone(page); | |
7421 | unsigned long pfn = page_to_pfn(page); | |
7422 | unsigned long flags; | |
7aeb09f9 | 7423 | unsigned int order; |
8d22ba1b WF |
7424 | |
7425 | spin_lock_irqsave(&zone->lock, flags); | |
7426 | for (order = 0; order < MAX_ORDER; order++) { | |
7427 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
7428 | ||
7429 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
7430 | break; | |
7431 | } | |
7432 | spin_unlock_irqrestore(&zone->lock, flags); | |
7433 | ||
7434 | return order < MAX_ORDER; | |
7435 | } |