]>
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> | |
ca79b0c2 | 19 | #include <linux/highmem.h> |
1da177e4 LT |
20 | #include <linux/swap.h> |
21 | #include <linux/interrupt.h> | |
22 | #include <linux/pagemap.h> | |
10ed273f | 23 | #include <linux/jiffies.h> |
edbe7d23 | 24 | #include <linux/memblock.h> |
1da177e4 | 25 | #include <linux/compiler.h> |
9f158333 | 26 | #include <linux/kernel.h> |
b8c73fc2 | 27 | #include <linux/kasan.h> |
1da177e4 LT |
28 | #include <linux/module.h> |
29 | #include <linux/suspend.h> | |
30 | #include <linux/pagevec.h> | |
31 | #include <linux/blkdev.h> | |
32 | #include <linux/slab.h> | |
a238ab5b | 33 | #include <linux/ratelimit.h> |
5a3135c2 | 34 | #include <linux/oom.h> |
1da177e4 LT |
35 | #include <linux/topology.h> |
36 | #include <linux/sysctl.h> | |
37 | #include <linux/cpu.h> | |
38 | #include <linux/cpuset.h> | |
bdc8cb98 | 39 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
40 | #include <linux/nodemask.h> |
41 | #include <linux/vmalloc.h> | |
a6cccdc3 | 42 | #include <linux/vmstat.h> |
4be38e35 | 43 | #include <linux/mempolicy.h> |
4b94ffdc | 44 | #include <linux/memremap.h> |
6811378e | 45 | #include <linux/stop_machine.h> |
c713216d MG |
46 | #include <linux/sort.h> |
47 | #include <linux/pfn.h> | |
3fcfab16 | 48 | #include <linux/backing-dev.h> |
933e312e | 49 | #include <linux/fault-inject.h> |
a5d76b54 | 50 | #include <linux/page-isolation.h> |
eefa864b | 51 | #include <linux/page_ext.h> |
3ac7fe5a | 52 | #include <linux/debugobjects.h> |
dbb1f81c | 53 | #include <linux/kmemleak.h> |
56de7263 | 54 | #include <linux/compaction.h> |
0d3d062a | 55 | #include <trace/events/kmem.h> |
d379f01d | 56 | #include <trace/events/oom.h> |
268bb0ce | 57 | #include <linux/prefetch.h> |
6e543d57 | 58 | #include <linux/mm_inline.h> |
041d3a8c | 59 | #include <linux/migrate.h> |
949f7ec5 | 60 | #include <linux/hugetlb.h> |
8bd75c77 | 61 | #include <linux/sched/rt.h> |
5b3cc15a | 62 | #include <linux/sched/mm.h> |
48c96a36 | 63 | #include <linux/page_owner.h> |
0e1cc95b | 64 | #include <linux/kthread.h> |
4949148a | 65 | #include <linux/memcontrol.h> |
42c269c8 | 66 | #include <linux/ftrace.h> |
d92a8cfc | 67 | #include <linux/lockdep.h> |
556b969a | 68 | #include <linux/nmi.h> |
eb414681 | 69 | #include <linux/psi.h> |
1da177e4 | 70 | |
7ee3d4e8 | 71 | #include <asm/sections.h> |
1da177e4 | 72 | #include <asm/tlbflush.h> |
ac924c60 | 73 | #include <asm/div64.h> |
1da177e4 LT |
74 | #include "internal.h" |
75 | ||
c8e251fa CS |
76 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
77 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 78 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 79 | |
72812019 LS |
80 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
81 | DEFINE_PER_CPU(int, numa_node); | |
82 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
83 | #endif | |
84 | ||
4518085e KW |
85 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
86 | ||
7aac7898 LS |
87 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
88 | /* | |
89 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
90 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
91 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
92 | * defined in <linux/topology.h>. | |
93 | */ | |
94 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
95 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 96 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
97 | #endif |
98 | ||
bd233f53 MG |
99 | /* work_structs for global per-cpu drains */ |
100 | DEFINE_MUTEX(pcpu_drain_mutex); | |
101 | DEFINE_PER_CPU(struct work_struct, pcpu_drain); | |
102 | ||
38addce8 | 103 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 104 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
105 | EXPORT_SYMBOL(latent_entropy); |
106 | #endif | |
107 | ||
1da177e4 | 108 | /* |
13808910 | 109 | * Array of node states. |
1da177e4 | 110 | */ |
13808910 CL |
111 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
112 | [N_POSSIBLE] = NODE_MASK_ALL, | |
113 | [N_ONLINE] = { { [0] = 1UL } }, | |
114 | #ifndef CONFIG_NUMA | |
115 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
116 | #ifdef CONFIG_HIGHMEM | |
117 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 118 | #endif |
20b2f52b | 119 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
120 | [N_CPU] = { { [0] = 1UL } }, |
121 | #endif /* NUMA */ | |
122 | }; | |
123 | EXPORT_SYMBOL(node_states); | |
124 | ||
ca79b0c2 AK |
125 | atomic_long_t _totalram_pages __read_mostly; |
126 | EXPORT_SYMBOL(_totalram_pages); | |
cb45b0e9 | 127 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 128 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 129 | |
1b76b02f | 130 | int percpu_pagelist_fraction; |
dcce284a | 131 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 132 | |
bb14c2c7 VB |
133 | /* |
134 | * A cached value of the page's pageblock's migratetype, used when the page is | |
135 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
136 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
137 | * Also the migratetype set in the page does not necessarily match the pcplist | |
138 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
139 | * other index - this ensures that it will be put on the correct CMA freelist. | |
140 | */ | |
141 | static inline int get_pcppage_migratetype(struct page *page) | |
142 | { | |
143 | return page->index; | |
144 | } | |
145 | ||
146 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
147 | { | |
148 | page->index = migratetype; | |
149 | } | |
150 | ||
452aa699 RW |
151 | #ifdef CONFIG_PM_SLEEP |
152 | /* | |
153 | * The following functions are used by the suspend/hibernate code to temporarily | |
154 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
155 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
156 | * they should always be called with system_transition_mutex held |
157 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
158 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
159 | * with that modification). | |
452aa699 | 160 | */ |
c9e664f1 RW |
161 | |
162 | static gfp_t saved_gfp_mask; | |
163 | ||
164 | void pm_restore_gfp_mask(void) | |
452aa699 | 165 | { |
55f2503c | 166 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
167 | if (saved_gfp_mask) { |
168 | gfp_allowed_mask = saved_gfp_mask; | |
169 | saved_gfp_mask = 0; | |
170 | } | |
452aa699 RW |
171 | } |
172 | ||
c9e664f1 | 173 | void pm_restrict_gfp_mask(void) |
452aa699 | 174 | { |
55f2503c | 175 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
176 | WARN_ON(saved_gfp_mask); |
177 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 178 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 179 | } |
f90ac398 MG |
180 | |
181 | bool pm_suspended_storage(void) | |
182 | { | |
d0164adc | 183 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
184 | return false; |
185 | return true; | |
186 | } | |
452aa699 RW |
187 | #endif /* CONFIG_PM_SLEEP */ |
188 | ||
d9c23400 | 189 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 190 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
191 | #endif |
192 | ||
d98c7a09 | 193 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 194 | |
1da177e4 LT |
195 | /* |
196 | * results with 256, 32 in the lowmem_reserve sysctl: | |
197 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
198 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
199 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
200 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 201 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
202 | * |
203 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
204 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 205 | */ |
d3cda233 | 206 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 207 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 208 | [ZONE_DMA] = 256, |
4b51d669 | 209 | #endif |
fb0e7942 | 210 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 211 | [ZONE_DMA32] = 256, |
fb0e7942 | 212 | #endif |
d3cda233 | 213 | [ZONE_NORMAL] = 32, |
e53ef38d | 214 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 215 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 216 | #endif |
d3cda233 | 217 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 218 | }; |
1da177e4 LT |
219 | |
220 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 221 | |
15ad7cdc | 222 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 223 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 224 | "DMA", |
4b51d669 | 225 | #endif |
fb0e7942 | 226 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 227 | "DMA32", |
fb0e7942 | 228 | #endif |
2f1b6248 | 229 | "Normal", |
e53ef38d | 230 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 231 | "HighMem", |
e53ef38d | 232 | #endif |
2a1e274a | 233 | "Movable", |
033fbae9 DW |
234 | #ifdef CONFIG_ZONE_DEVICE |
235 | "Device", | |
236 | #endif | |
2f1b6248 CL |
237 | }; |
238 | ||
60f30350 VB |
239 | char * const migratetype_names[MIGRATE_TYPES] = { |
240 | "Unmovable", | |
241 | "Movable", | |
242 | "Reclaimable", | |
243 | "HighAtomic", | |
244 | #ifdef CONFIG_CMA | |
245 | "CMA", | |
246 | #endif | |
247 | #ifdef CONFIG_MEMORY_ISOLATION | |
248 | "Isolate", | |
249 | #endif | |
250 | }; | |
251 | ||
f1e61557 KS |
252 | compound_page_dtor * const compound_page_dtors[] = { |
253 | NULL, | |
254 | free_compound_page, | |
255 | #ifdef CONFIG_HUGETLB_PAGE | |
256 | free_huge_page, | |
257 | #endif | |
9a982250 KS |
258 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
259 | free_transhuge_page, | |
260 | #endif | |
f1e61557 KS |
261 | }; |
262 | ||
1da177e4 | 263 | int min_free_kbytes = 1024; |
42aa83cb | 264 | int user_min_free_kbytes = -1; |
795ae7a0 | 265 | int watermark_scale_factor = 10; |
1da177e4 | 266 | |
7f16f91f DR |
267 | static unsigned long nr_kernel_pages __meminitdata; |
268 | static unsigned long nr_all_pages __meminitdata; | |
269 | static unsigned long dma_reserve __meminitdata; | |
1da177e4 | 270 | |
0ee332c1 | 271 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
7f16f91f DR |
272 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __meminitdata; |
273 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __meminitdata; | |
274 | static unsigned long required_kernelcore __initdata; | |
a5c6d650 | 275 | static unsigned long required_kernelcore_percent __initdata; |
7f16f91f | 276 | static unsigned long required_movablecore __initdata; |
a5c6d650 | 277 | static unsigned long required_movablecore_percent __initdata; |
7f16f91f DR |
278 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __meminitdata; |
279 | static bool mirrored_kernelcore __meminitdata; | |
0ee332c1 TH |
280 | |
281 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
282 | int movable_zone; | |
283 | EXPORT_SYMBOL(movable_zone); | |
284 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 285 | |
418508c1 MS |
286 | #if MAX_NUMNODES > 1 |
287 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 288 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 289 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 290 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
291 | #endif |
292 | ||
9ef9acb0 MG |
293 | int page_group_by_mobility_disabled __read_mostly; |
294 | ||
3a80a7fa | 295 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3a80a7fa | 296 | /* Returns true if the struct page for the pfn is uninitialised */ |
0e1cc95b | 297 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 298 | { |
ef70b6f4 MG |
299 | int nid = early_pfn_to_nid(pfn); |
300 | ||
301 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
302 | return true; |
303 | ||
304 | return false; | |
305 | } | |
306 | ||
307 | /* | |
d3035be4 | 308 | * Returns true when the remaining initialisation should be deferred until |
3a80a7fa MG |
309 | * later in the boot cycle when it can be parallelised. |
310 | */ | |
d3035be4 PT |
311 | static bool __meminit |
312 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn) | |
3a80a7fa | 313 | { |
d3035be4 PT |
314 | static unsigned long prev_end_pfn, nr_initialised; |
315 | ||
316 | /* | |
317 | * prev_end_pfn static that contains the end of previous zone | |
318 | * No need to protect because called very early in boot before smp_init. | |
319 | */ | |
320 | if (prev_end_pfn != end_pfn) { | |
321 | prev_end_pfn = end_pfn; | |
322 | nr_initialised = 0; | |
323 | } | |
324 | ||
3c2c6488 | 325 | /* Always populate low zones for address-constrained allocations */ |
d3035be4 | 326 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid))) |
3a80a7fa | 327 | return false; |
d3035be4 PT |
328 | nr_initialised++; |
329 | if ((nr_initialised > NODE_DATA(nid)->static_init_pgcnt) && | |
330 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { | |
331 | NODE_DATA(nid)->first_deferred_pfn = pfn; | |
332 | return true; | |
3a80a7fa | 333 | } |
d3035be4 | 334 | return false; |
3a80a7fa MG |
335 | } |
336 | #else | |
3a80a7fa MG |
337 | static inline bool early_page_uninitialised(unsigned long pfn) |
338 | { | |
339 | return false; | |
340 | } | |
341 | ||
d3035be4 | 342 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn) |
3a80a7fa | 343 | { |
d3035be4 | 344 | return false; |
3a80a7fa MG |
345 | } |
346 | #endif | |
347 | ||
0b423ca2 MG |
348 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
349 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
350 | unsigned long pfn) | |
351 | { | |
352 | #ifdef CONFIG_SPARSEMEM | |
353 | return __pfn_to_section(pfn)->pageblock_flags; | |
354 | #else | |
355 | return page_zone(page)->pageblock_flags; | |
356 | #endif /* CONFIG_SPARSEMEM */ | |
357 | } | |
358 | ||
359 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
360 | { | |
361 | #ifdef CONFIG_SPARSEMEM | |
362 | pfn &= (PAGES_PER_SECTION-1); | |
363 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
364 | #else | |
365 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
366 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
367 | #endif /* CONFIG_SPARSEMEM */ | |
368 | } | |
369 | ||
370 | /** | |
371 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
372 | * @page: The page within the block of interest | |
373 | * @pfn: The target page frame number | |
374 | * @end_bitidx: The last bit of interest to retrieve | |
375 | * @mask: mask of bits that the caller is interested in | |
376 | * | |
377 | * Return: pageblock_bits flags | |
378 | */ | |
379 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
380 | unsigned long pfn, | |
381 | unsigned long end_bitidx, | |
382 | unsigned long mask) | |
383 | { | |
384 | unsigned long *bitmap; | |
385 | unsigned long bitidx, word_bitidx; | |
386 | unsigned long word; | |
387 | ||
388 | bitmap = get_pageblock_bitmap(page, pfn); | |
389 | bitidx = pfn_to_bitidx(page, pfn); | |
390 | word_bitidx = bitidx / BITS_PER_LONG; | |
391 | bitidx &= (BITS_PER_LONG-1); | |
392 | ||
393 | word = bitmap[word_bitidx]; | |
394 | bitidx += end_bitidx; | |
395 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
396 | } | |
397 | ||
398 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
399 | unsigned long end_bitidx, | |
400 | unsigned long mask) | |
401 | { | |
402 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
403 | } | |
404 | ||
405 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
406 | { | |
407 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
408 | } | |
409 | ||
410 | /** | |
411 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
412 | * @page: The page within the block of interest | |
413 | * @flags: The flags to set | |
414 | * @pfn: The target page frame number | |
415 | * @end_bitidx: The last bit of interest | |
416 | * @mask: mask of bits that the caller is interested in | |
417 | */ | |
418 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
419 | unsigned long pfn, | |
420 | unsigned long end_bitidx, | |
421 | unsigned long mask) | |
422 | { | |
423 | unsigned long *bitmap; | |
424 | unsigned long bitidx, word_bitidx; | |
425 | unsigned long old_word, word; | |
426 | ||
427 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
428 | ||
429 | bitmap = get_pageblock_bitmap(page, pfn); | |
430 | bitidx = pfn_to_bitidx(page, pfn); | |
431 | word_bitidx = bitidx / BITS_PER_LONG; | |
432 | bitidx &= (BITS_PER_LONG-1); | |
433 | ||
434 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
435 | ||
436 | bitidx += end_bitidx; | |
437 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
438 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
439 | ||
440 | word = READ_ONCE(bitmap[word_bitidx]); | |
441 | for (;;) { | |
442 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
443 | if (word == old_word) | |
444 | break; | |
445 | word = old_word; | |
446 | } | |
447 | } | |
3a80a7fa | 448 | |
ee6f509c | 449 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 450 | { |
5d0f3f72 KM |
451 | if (unlikely(page_group_by_mobility_disabled && |
452 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
453 | migratetype = MIGRATE_UNMOVABLE; |
454 | ||
b2a0ac88 MG |
455 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
456 | PB_migrate, PB_migrate_end); | |
457 | } | |
458 | ||
13e7444b | 459 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 460 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 461 | { |
bdc8cb98 DH |
462 | int ret = 0; |
463 | unsigned seq; | |
464 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 465 | unsigned long sp, start_pfn; |
c6a57e19 | 466 | |
bdc8cb98 DH |
467 | do { |
468 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
469 | start_pfn = zone->zone_start_pfn; |
470 | sp = zone->spanned_pages; | |
108bcc96 | 471 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
472 | ret = 1; |
473 | } while (zone_span_seqretry(zone, seq)); | |
474 | ||
b5e6a5a2 | 475 | if (ret) |
613813e8 DH |
476 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
477 | pfn, zone_to_nid(zone), zone->name, | |
478 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 479 | |
bdc8cb98 | 480 | return ret; |
c6a57e19 DH |
481 | } |
482 | ||
483 | static int page_is_consistent(struct zone *zone, struct page *page) | |
484 | { | |
14e07298 | 485 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 486 | return 0; |
1da177e4 | 487 | if (zone != page_zone(page)) |
c6a57e19 DH |
488 | return 0; |
489 | ||
490 | return 1; | |
491 | } | |
492 | /* | |
493 | * Temporary debugging check for pages not lying within a given zone. | |
494 | */ | |
d73d3c9f | 495 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
496 | { |
497 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 498 | return 1; |
c6a57e19 DH |
499 | if (!page_is_consistent(zone, page)) |
500 | return 1; | |
501 | ||
1da177e4 LT |
502 | return 0; |
503 | } | |
13e7444b | 504 | #else |
d73d3c9f | 505 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
506 | { |
507 | return 0; | |
508 | } | |
509 | #endif | |
510 | ||
d230dec1 KS |
511 | static void bad_page(struct page *page, const char *reason, |
512 | unsigned long bad_flags) | |
1da177e4 | 513 | { |
d936cf9b HD |
514 | static unsigned long resume; |
515 | static unsigned long nr_shown; | |
516 | static unsigned long nr_unshown; | |
517 | ||
518 | /* | |
519 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
520 | * or allow a steady drip of one report per second. | |
521 | */ | |
522 | if (nr_shown == 60) { | |
523 | if (time_before(jiffies, resume)) { | |
524 | nr_unshown++; | |
525 | goto out; | |
526 | } | |
527 | if (nr_unshown) { | |
ff8e8116 | 528 | pr_alert( |
1e9e6365 | 529 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
530 | nr_unshown); |
531 | nr_unshown = 0; | |
532 | } | |
533 | nr_shown = 0; | |
534 | } | |
535 | if (nr_shown++ == 0) | |
536 | resume = jiffies + 60 * HZ; | |
537 | ||
ff8e8116 | 538 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 539 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
540 | __dump_page(page, reason); |
541 | bad_flags &= page->flags; | |
542 | if (bad_flags) | |
543 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
544 | bad_flags, &bad_flags); | |
4e462112 | 545 | dump_page_owner(page); |
3dc14741 | 546 | |
4f31888c | 547 | print_modules(); |
1da177e4 | 548 | dump_stack(); |
d936cf9b | 549 | out: |
8cc3b392 | 550 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 551 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 552 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
553 | } |
554 | ||
1da177e4 LT |
555 | /* |
556 | * Higher-order pages are called "compound pages". They are structured thusly: | |
557 | * | |
1d798ca3 | 558 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 559 | * |
1d798ca3 KS |
560 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
561 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 562 | * |
1d798ca3 KS |
563 | * The first tail page's ->compound_dtor holds the offset in array of compound |
564 | * page destructors. See compound_page_dtors. | |
1da177e4 | 565 | * |
1d798ca3 | 566 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 567 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 568 | */ |
d98c7a09 | 569 | |
9a982250 | 570 | void free_compound_page(struct page *page) |
d98c7a09 | 571 | { |
d85f3385 | 572 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
573 | } |
574 | ||
d00181b9 | 575 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
576 | { |
577 | int i; | |
578 | int nr_pages = 1 << order; | |
579 | ||
f1e61557 | 580 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
581 | set_compound_order(page, order); |
582 | __SetPageHead(page); | |
583 | for (i = 1; i < nr_pages; i++) { | |
584 | struct page *p = page + i; | |
58a84aa9 | 585 | set_page_count(p, 0); |
1c290f64 | 586 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 587 | set_compound_head(p, page); |
18229df5 | 588 | } |
53f9263b | 589 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
590 | } |
591 | ||
c0a32fc5 SG |
592 | #ifdef CONFIG_DEBUG_PAGEALLOC |
593 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
594 | bool _debug_pagealloc_enabled __read_mostly |
595 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 596 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
597 | bool _debug_guardpage_enabled __read_mostly; |
598 | ||
031bc574 JK |
599 | static int __init early_debug_pagealloc(char *buf) |
600 | { | |
601 | if (!buf) | |
602 | return -EINVAL; | |
2a138dc7 | 603 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
031bc574 JK |
604 | } |
605 | early_param("debug_pagealloc", early_debug_pagealloc); | |
606 | ||
e30825f1 JK |
607 | static bool need_debug_guardpage(void) |
608 | { | |
031bc574 JK |
609 | /* If we don't use debug_pagealloc, we don't need guard page */ |
610 | if (!debug_pagealloc_enabled()) | |
611 | return false; | |
612 | ||
f1c1e9f7 JK |
613 | if (!debug_guardpage_minorder()) |
614 | return false; | |
615 | ||
e30825f1 JK |
616 | return true; |
617 | } | |
618 | ||
619 | static void init_debug_guardpage(void) | |
620 | { | |
031bc574 JK |
621 | if (!debug_pagealloc_enabled()) |
622 | return; | |
623 | ||
f1c1e9f7 JK |
624 | if (!debug_guardpage_minorder()) |
625 | return; | |
626 | ||
e30825f1 JK |
627 | _debug_guardpage_enabled = true; |
628 | } | |
629 | ||
630 | struct page_ext_operations debug_guardpage_ops = { | |
631 | .need = need_debug_guardpage, | |
632 | .init = init_debug_guardpage, | |
633 | }; | |
c0a32fc5 SG |
634 | |
635 | static int __init debug_guardpage_minorder_setup(char *buf) | |
636 | { | |
637 | unsigned long res; | |
638 | ||
639 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 640 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
641 | return 0; |
642 | } | |
643 | _debug_guardpage_minorder = res; | |
1170532b | 644 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
645 | return 0; |
646 | } | |
f1c1e9f7 | 647 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 648 | |
acbc15a4 | 649 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 650 | unsigned int order, int migratetype) |
c0a32fc5 | 651 | { |
e30825f1 JK |
652 | struct page_ext *page_ext; |
653 | ||
654 | if (!debug_guardpage_enabled()) | |
acbc15a4 JK |
655 | return false; |
656 | ||
657 | if (order >= debug_guardpage_minorder()) | |
658 | return false; | |
e30825f1 JK |
659 | |
660 | page_ext = lookup_page_ext(page); | |
f86e4271 | 661 | if (unlikely(!page_ext)) |
acbc15a4 | 662 | return false; |
f86e4271 | 663 | |
e30825f1 JK |
664 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
665 | ||
2847cf95 JK |
666 | INIT_LIST_HEAD(&page->lru); |
667 | set_page_private(page, order); | |
668 | /* Guard pages are not available for any usage */ | |
669 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
670 | |
671 | return true; | |
c0a32fc5 SG |
672 | } |
673 | ||
2847cf95 JK |
674 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
675 | unsigned int order, int migratetype) | |
c0a32fc5 | 676 | { |
e30825f1 JK |
677 | struct page_ext *page_ext; |
678 | ||
679 | if (!debug_guardpage_enabled()) | |
680 | return; | |
681 | ||
682 | page_ext = lookup_page_ext(page); | |
f86e4271 YS |
683 | if (unlikely(!page_ext)) |
684 | return; | |
685 | ||
e30825f1 JK |
686 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
687 | ||
2847cf95 JK |
688 | set_page_private(page, 0); |
689 | if (!is_migrate_isolate(migratetype)) | |
690 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
691 | } |
692 | #else | |
980ac167 | 693 | struct page_ext_operations debug_guardpage_ops; |
acbc15a4 JK |
694 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
695 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
696 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
697 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
698 | #endif |
699 | ||
7aeb09f9 | 700 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 701 | { |
4c21e2f2 | 702 | set_page_private(page, order); |
676165a8 | 703 | __SetPageBuddy(page); |
1da177e4 LT |
704 | } |
705 | ||
706 | static inline void rmv_page_order(struct page *page) | |
707 | { | |
676165a8 | 708 | __ClearPageBuddy(page); |
4c21e2f2 | 709 | set_page_private(page, 0); |
1da177e4 LT |
710 | } |
711 | ||
1da177e4 LT |
712 | /* |
713 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 714 | * we can coalesce a page and its buddy if |
13ad59df | 715 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 716 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
717 | * (c) a page and its buddy have the same order && |
718 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 719 | * |
6e292b9b MW |
720 | * For recording whether a page is in the buddy system, we set PageBuddy. |
721 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 722 | * |
676165a8 | 723 | * For recording page's order, we use page_private(page). |
1da177e4 | 724 | */ |
cb2b95e1 | 725 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 726 | unsigned int order) |
1da177e4 | 727 | { |
c0a32fc5 | 728 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
729 | if (page_zone_id(page) != page_zone_id(buddy)) |
730 | return 0; | |
731 | ||
4c5018ce WY |
732 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
733 | ||
c0a32fc5 SG |
734 | return 1; |
735 | } | |
736 | ||
cb2b95e1 | 737 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
738 | /* |
739 | * zone check is done late to avoid uselessly | |
740 | * calculating zone/node ids for pages that could | |
741 | * never merge. | |
742 | */ | |
743 | if (page_zone_id(page) != page_zone_id(buddy)) | |
744 | return 0; | |
745 | ||
4c5018ce WY |
746 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
747 | ||
6aa3001b | 748 | return 1; |
676165a8 | 749 | } |
6aa3001b | 750 | return 0; |
1da177e4 LT |
751 | } |
752 | ||
753 | /* | |
754 | * Freeing function for a buddy system allocator. | |
755 | * | |
756 | * The concept of a buddy system is to maintain direct-mapped table | |
757 | * (containing bit values) for memory blocks of various "orders". | |
758 | * The bottom level table contains the map for the smallest allocatable | |
759 | * units of memory (here, pages), and each level above it describes | |
760 | * pairs of units from the levels below, hence, "buddies". | |
761 | * At a high level, all that happens here is marking the table entry | |
762 | * at the bottom level available, and propagating the changes upward | |
763 | * as necessary, plus some accounting needed to play nicely with other | |
764 | * parts of the VM system. | |
765 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
766 | * free pages of length of (1 << order) and marked with PageBuddy. |
767 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 768 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
769 | * other. That is, if we allocate a small block, and both were |
770 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 771 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 772 | * triggers coalescing into a block of larger size. |
1da177e4 | 773 | * |
6d49e352 | 774 | * -- nyc |
1da177e4 LT |
775 | */ |
776 | ||
48db57f8 | 777 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 778 | unsigned long pfn, |
ed0ae21d MG |
779 | struct zone *zone, unsigned int order, |
780 | int migratetype) | |
1da177e4 | 781 | { |
76741e77 VB |
782 | unsigned long combined_pfn; |
783 | unsigned long uninitialized_var(buddy_pfn); | |
6dda9d55 | 784 | struct page *buddy; |
d9dddbf5 VB |
785 | unsigned int max_order; |
786 | ||
787 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 788 | |
d29bb978 | 789 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 790 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 791 | |
ed0ae21d | 792 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 793 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 794 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 795 | |
76741e77 | 796 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 797 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 798 | |
d9dddbf5 | 799 | continue_merging: |
3c605096 | 800 | while (order < max_order - 1) { |
76741e77 VB |
801 | buddy_pfn = __find_buddy_pfn(pfn, order); |
802 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
803 | |
804 | if (!pfn_valid_within(buddy_pfn)) | |
805 | goto done_merging; | |
cb2b95e1 | 806 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 807 | goto done_merging; |
c0a32fc5 SG |
808 | /* |
809 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
810 | * merge with it and move up one order. | |
811 | */ | |
812 | if (page_is_guard(buddy)) { | |
2847cf95 | 813 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
814 | } else { |
815 | list_del(&buddy->lru); | |
816 | zone->free_area[order].nr_free--; | |
817 | rmv_page_order(buddy); | |
818 | } | |
76741e77 VB |
819 | combined_pfn = buddy_pfn & pfn; |
820 | page = page + (combined_pfn - pfn); | |
821 | pfn = combined_pfn; | |
1da177e4 LT |
822 | order++; |
823 | } | |
d9dddbf5 VB |
824 | if (max_order < MAX_ORDER) { |
825 | /* If we are here, it means order is >= pageblock_order. | |
826 | * We want to prevent merge between freepages on isolate | |
827 | * pageblock and normal pageblock. Without this, pageblock | |
828 | * isolation could cause incorrect freepage or CMA accounting. | |
829 | * | |
830 | * We don't want to hit this code for the more frequent | |
831 | * low-order merging. | |
832 | */ | |
833 | if (unlikely(has_isolate_pageblock(zone))) { | |
834 | int buddy_mt; | |
835 | ||
76741e77 VB |
836 | buddy_pfn = __find_buddy_pfn(pfn, order); |
837 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
838 | buddy_mt = get_pageblock_migratetype(buddy); |
839 | ||
840 | if (migratetype != buddy_mt | |
841 | && (is_migrate_isolate(migratetype) || | |
842 | is_migrate_isolate(buddy_mt))) | |
843 | goto done_merging; | |
844 | } | |
845 | max_order++; | |
846 | goto continue_merging; | |
847 | } | |
848 | ||
849 | done_merging: | |
1da177e4 | 850 | set_page_order(page, order); |
6dda9d55 CZ |
851 | |
852 | /* | |
853 | * If this is not the largest possible page, check if the buddy | |
854 | * of the next-highest order is free. If it is, it's possible | |
855 | * that pages are being freed that will coalesce soon. In case, | |
856 | * that is happening, add the free page to the tail of the list | |
857 | * so it's less likely to be used soon and more likely to be merged | |
858 | * as a higher order page | |
859 | */ | |
13ad59df | 860 | if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) { |
6dda9d55 | 861 | struct page *higher_page, *higher_buddy; |
76741e77 VB |
862 | combined_pfn = buddy_pfn & pfn; |
863 | higher_page = page + (combined_pfn - pfn); | |
864 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
865 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
b4fb8f66 TL |
866 | if (pfn_valid_within(buddy_pfn) && |
867 | page_is_buddy(higher_page, higher_buddy, order + 1)) { | |
6dda9d55 CZ |
868 | list_add_tail(&page->lru, |
869 | &zone->free_area[order].free_list[migratetype]); | |
870 | goto out; | |
871 | } | |
872 | } | |
873 | ||
874 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
875 | out: | |
1da177e4 LT |
876 | zone->free_area[order].nr_free++; |
877 | } | |
878 | ||
7bfec6f4 MG |
879 | /* |
880 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
881 | * try and check multiple fields with one check. The caller must do a detailed | |
882 | * check if necessary. | |
883 | */ | |
884 | static inline bool page_expected_state(struct page *page, | |
885 | unsigned long check_flags) | |
886 | { | |
887 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
888 | return false; | |
889 | ||
890 | if (unlikely((unsigned long)page->mapping | | |
891 | page_ref_count(page) | | |
892 | #ifdef CONFIG_MEMCG | |
893 | (unsigned long)page->mem_cgroup | | |
894 | #endif | |
895 | (page->flags & check_flags))) | |
896 | return false; | |
897 | ||
898 | return true; | |
899 | } | |
900 | ||
bb552ac6 | 901 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 902 | { |
7bfec6f4 MG |
903 | const char *bad_reason; |
904 | unsigned long bad_flags; | |
905 | ||
7bfec6f4 MG |
906 | bad_reason = NULL; |
907 | bad_flags = 0; | |
f0b791a3 | 908 | |
53f9263b | 909 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
910 | bad_reason = "nonzero mapcount"; |
911 | if (unlikely(page->mapping != NULL)) | |
912 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 913 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 914 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
915 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
916 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
917 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
918 | } | |
9edad6ea JW |
919 | #ifdef CONFIG_MEMCG |
920 | if (unlikely(page->mem_cgroup)) | |
921 | bad_reason = "page still charged to cgroup"; | |
922 | #endif | |
7bfec6f4 | 923 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
924 | } |
925 | ||
926 | static inline int free_pages_check(struct page *page) | |
927 | { | |
da838d4f | 928 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 929 | return 0; |
bb552ac6 MG |
930 | |
931 | /* Something has gone sideways, find it */ | |
932 | free_pages_check_bad(page); | |
7bfec6f4 | 933 | return 1; |
1da177e4 LT |
934 | } |
935 | ||
4db7548c MG |
936 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
937 | { | |
938 | int ret = 1; | |
939 | ||
940 | /* | |
941 | * We rely page->lru.next never has bit 0 set, unless the page | |
942 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
943 | */ | |
944 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
945 | ||
946 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
947 | ret = 0; | |
948 | goto out; | |
949 | } | |
950 | switch (page - head_page) { | |
951 | case 1: | |
4da1984e | 952 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c MG |
953 | if (unlikely(compound_mapcount(page))) { |
954 | bad_page(page, "nonzero compound_mapcount", 0); | |
955 | goto out; | |
956 | } | |
957 | break; | |
958 | case 2: | |
959 | /* | |
960 | * the second tail page: ->mapping is | |
fa3015b7 | 961 | * deferred_list.next -- ignore value. |
4db7548c MG |
962 | */ |
963 | break; | |
964 | default: | |
965 | if (page->mapping != TAIL_MAPPING) { | |
966 | bad_page(page, "corrupted mapping in tail page", 0); | |
967 | goto out; | |
968 | } | |
969 | break; | |
970 | } | |
971 | if (unlikely(!PageTail(page))) { | |
972 | bad_page(page, "PageTail not set", 0); | |
973 | goto out; | |
974 | } | |
975 | if (unlikely(compound_head(page) != head_page)) { | |
976 | bad_page(page, "compound_head not consistent", 0); | |
977 | goto out; | |
978 | } | |
979 | ret = 0; | |
980 | out: | |
981 | page->mapping = NULL; | |
982 | clear_compound_head(page); | |
983 | return ret; | |
984 | } | |
985 | ||
e2769dbd MG |
986 | static __always_inline bool free_pages_prepare(struct page *page, |
987 | unsigned int order, bool check_free) | |
4db7548c | 988 | { |
e2769dbd | 989 | int bad = 0; |
4db7548c | 990 | |
4db7548c MG |
991 | VM_BUG_ON_PAGE(PageTail(page), page); |
992 | ||
e2769dbd | 993 | trace_mm_page_free(page, order); |
e2769dbd MG |
994 | |
995 | /* | |
996 | * Check tail pages before head page information is cleared to | |
997 | * avoid checking PageCompound for order-0 pages. | |
998 | */ | |
999 | if (unlikely(order)) { | |
1000 | bool compound = PageCompound(page); | |
1001 | int i; | |
1002 | ||
1003 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1004 | |
9a73f61b KS |
1005 | if (compound) |
1006 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1007 | for (i = 1; i < (1 << order); i++) { |
1008 | if (compound) | |
1009 | bad += free_tail_pages_check(page, page + i); | |
1010 | if (unlikely(free_pages_check(page + i))) { | |
1011 | bad++; | |
1012 | continue; | |
1013 | } | |
1014 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1015 | } | |
1016 | } | |
bda807d4 | 1017 | if (PageMappingFlags(page)) |
4db7548c | 1018 | page->mapping = NULL; |
c4159a75 | 1019 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
4949148a | 1020 | memcg_kmem_uncharge(page, order); |
e2769dbd MG |
1021 | if (check_free) |
1022 | bad += free_pages_check(page); | |
1023 | if (bad) | |
1024 | return false; | |
4db7548c | 1025 | |
e2769dbd MG |
1026 | page_cpupid_reset_last(page); |
1027 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1028 | reset_page_owner(page, order); | |
4db7548c MG |
1029 | |
1030 | if (!PageHighMem(page)) { | |
1031 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1032 | PAGE_SIZE << order); |
4db7548c | 1033 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1034 | PAGE_SIZE << order); |
4db7548c | 1035 | } |
e2769dbd MG |
1036 | arch_free_page(page, order); |
1037 | kernel_poison_pages(page, 1 << order, 0); | |
1038 | kernel_map_pages(page, 1 << order, 0); | |
29b52de1 | 1039 | kasan_free_pages(page, order); |
4db7548c | 1040 | |
4db7548c MG |
1041 | return true; |
1042 | } | |
1043 | ||
e2769dbd MG |
1044 | #ifdef CONFIG_DEBUG_VM |
1045 | static inline bool free_pcp_prepare(struct page *page) | |
1046 | { | |
1047 | return free_pages_prepare(page, 0, true); | |
1048 | } | |
1049 | ||
1050 | static inline bool bulkfree_pcp_prepare(struct page *page) | |
1051 | { | |
1052 | return false; | |
1053 | } | |
1054 | #else | |
1055 | static bool free_pcp_prepare(struct page *page) | |
1056 | { | |
1057 | return free_pages_prepare(page, 0, false); | |
1058 | } | |
1059 | ||
4db7548c MG |
1060 | static bool bulkfree_pcp_prepare(struct page *page) |
1061 | { | |
1062 | return free_pages_check(page); | |
1063 | } | |
1064 | #endif /* CONFIG_DEBUG_VM */ | |
1065 | ||
97334162 AL |
1066 | static inline void prefetch_buddy(struct page *page) |
1067 | { | |
1068 | unsigned long pfn = page_to_pfn(page); | |
1069 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1070 | struct page *buddy = page + (buddy_pfn - pfn); | |
1071 | ||
1072 | prefetch(buddy); | |
1073 | } | |
1074 | ||
1da177e4 | 1075 | /* |
5f8dcc21 | 1076 | * Frees a number of pages from the PCP lists |
1da177e4 | 1077 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1078 | * count is the number of pages to free. |
1da177e4 LT |
1079 | * |
1080 | * If the zone was previously in an "all pages pinned" state then look to | |
1081 | * see if this freeing clears that state. | |
1082 | * | |
1083 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1084 | * pinned" detection logic. | |
1085 | */ | |
5f8dcc21 MG |
1086 | static void free_pcppages_bulk(struct zone *zone, int count, |
1087 | struct per_cpu_pages *pcp) | |
1da177e4 | 1088 | { |
5f8dcc21 | 1089 | int migratetype = 0; |
a6f9edd6 | 1090 | int batch_free = 0; |
97334162 | 1091 | int prefetch_nr = 0; |
3777999d | 1092 | bool isolated_pageblocks; |
0a5f4e5b AL |
1093 | struct page *page, *tmp; |
1094 | LIST_HEAD(head); | |
f2260e6b | 1095 | |
e5b31ac2 | 1096 | while (count) { |
5f8dcc21 MG |
1097 | struct list_head *list; |
1098 | ||
1099 | /* | |
a6f9edd6 MG |
1100 | * Remove pages from lists in a round-robin fashion. A |
1101 | * batch_free count is maintained that is incremented when an | |
1102 | * empty list is encountered. This is so more pages are freed | |
1103 | * off fuller lists instead of spinning excessively around empty | |
1104 | * lists | |
5f8dcc21 MG |
1105 | */ |
1106 | do { | |
a6f9edd6 | 1107 | batch_free++; |
5f8dcc21 MG |
1108 | if (++migratetype == MIGRATE_PCPTYPES) |
1109 | migratetype = 0; | |
1110 | list = &pcp->lists[migratetype]; | |
1111 | } while (list_empty(list)); | |
48db57f8 | 1112 | |
1d16871d NK |
1113 | /* This is the only non-empty list. Free them all. */ |
1114 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1115 | batch_free = count; |
1d16871d | 1116 | |
a6f9edd6 | 1117 | do { |
a16601c5 | 1118 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1119 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1120 | list_del(&page->lru); |
77ba9062 | 1121 | pcp->count--; |
aa016d14 | 1122 | |
4db7548c MG |
1123 | if (bulkfree_pcp_prepare(page)) |
1124 | continue; | |
1125 | ||
0a5f4e5b | 1126 | list_add_tail(&page->lru, &head); |
97334162 AL |
1127 | |
1128 | /* | |
1129 | * We are going to put the page back to the global | |
1130 | * pool, prefetch its buddy to speed up later access | |
1131 | * under zone->lock. It is believed the overhead of | |
1132 | * an additional test and calculating buddy_pfn here | |
1133 | * can be offset by reduced memory latency later. To | |
1134 | * avoid excessive prefetching due to large count, only | |
1135 | * prefetch buddy for the first pcp->batch nr of pages. | |
1136 | */ | |
1137 | if (prefetch_nr++ < pcp->batch) | |
1138 | prefetch_buddy(page); | |
e5b31ac2 | 1139 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1140 | } |
0a5f4e5b AL |
1141 | |
1142 | spin_lock(&zone->lock); | |
1143 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1144 | ||
1145 | /* | |
1146 | * Use safe version since after __free_one_page(), | |
1147 | * page->lru.next will not point to original list. | |
1148 | */ | |
1149 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1150 | int mt = get_pcppage_migratetype(page); | |
1151 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1152 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1153 | /* Pageblock could have been isolated meanwhile */ | |
1154 | if (unlikely(isolated_pageblocks)) | |
1155 | mt = get_pageblock_migratetype(page); | |
1156 | ||
1157 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); | |
1158 | trace_mm_page_pcpu_drain(page, 0, mt); | |
1159 | } | |
d34b0733 | 1160 | spin_unlock(&zone->lock); |
1da177e4 LT |
1161 | } |
1162 | ||
dc4b0caf MG |
1163 | static void free_one_page(struct zone *zone, |
1164 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1165 | unsigned int order, |
ed0ae21d | 1166 | int migratetype) |
1da177e4 | 1167 | { |
d34b0733 | 1168 | spin_lock(&zone->lock); |
ad53f92e JK |
1169 | if (unlikely(has_isolate_pageblock(zone) || |
1170 | is_migrate_isolate(migratetype))) { | |
1171 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1172 | } |
dc4b0caf | 1173 | __free_one_page(page, pfn, zone, order, migratetype); |
d34b0733 | 1174 | spin_unlock(&zone->lock); |
48db57f8 NP |
1175 | } |
1176 | ||
1e8ce83c | 1177 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1178 | unsigned long zone, int nid) |
1e8ce83c | 1179 | { |
d0dc12e8 | 1180 | mm_zero_struct_page(page); |
1e8ce83c | 1181 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1182 | init_page_count(page); |
1183 | page_mapcount_reset(page); | |
1184 | page_cpupid_reset_last(page); | |
2813b9c0 | 1185 | page_kasan_tag_reset(page); |
1e8ce83c | 1186 | |
1e8ce83c RH |
1187 | INIT_LIST_HEAD(&page->lru); |
1188 | #ifdef WANT_PAGE_VIRTUAL | |
1189 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1190 | if (!is_highmem_idx(zone)) | |
1191 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1192 | #endif | |
1193 | } | |
1194 | ||
7e18adb4 | 1195 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1196 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1197 | { |
1198 | pg_data_t *pgdat; | |
1199 | int nid, zid; | |
1200 | ||
1201 | if (!early_page_uninitialised(pfn)) | |
1202 | return; | |
1203 | ||
1204 | nid = early_pfn_to_nid(pfn); | |
1205 | pgdat = NODE_DATA(nid); | |
1206 | ||
1207 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1208 | struct zone *zone = &pgdat->node_zones[zid]; | |
1209 | ||
1210 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1211 | break; | |
1212 | } | |
d0dc12e8 | 1213 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1214 | } |
1215 | #else | |
1216 | static inline void init_reserved_page(unsigned long pfn) | |
1217 | { | |
1218 | } | |
1219 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1220 | ||
92923ca3 NZ |
1221 | /* |
1222 | * Initialised pages do not have PageReserved set. This function is | |
1223 | * called for each range allocated by the bootmem allocator and | |
1224 | * marks the pages PageReserved. The remaining valid pages are later | |
1225 | * sent to the buddy page allocator. | |
1226 | */ | |
4b50bcc7 | 1227 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1228 | { |
1229 | unsigned long start_pfn = PFN_DOWN(start); | |
1230 | unsigned long end_pfn = PFN_UP(end); | |
1231 | ||
7e18adb4 MG |
1232 | for (; start_pfn < end_pfn; start_pfn++) { |
1233 | if (pfn_valid(start_pfn)) { | |
1234 | struct page *page = pfn_to_page(start_pfn); | |
1235 | ||
1236 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1237 | |
1238 | /* Avoid false-positive PageTail() */ | |
1239 | INIT_LIST_HEAD(&page->lru); | |
1240 | ||
d483da5b AD |
1241 | /* |
1242 | * no need for atomic set_bit because the struct | |
1243 | * page is not visible yet so nobody should | |
1244 | * access it yet. | |
1245 | */ | |
1246 | __SetPageReserved(page); | |
7e18adb4 MG |
1247 | } |
1248 | } | |
92923ca3 NZ |
1249 | } |
1250 | ||
ec95f53a KM |
1251 | static void __free_pages_ok(struct page *page, unsigned int order) |
1252 | { | |
d34b0733 | 1253 | unsigned long flags; |
95e34412 | 1254 | int migratetype; |
dc4b0caf | 1255 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1256 | |
e2769dbd | 1257 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1258 | return; |
1259 | ||
cfc47a28 | 1260 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1261 | local_irq_save(flags); |
1262 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1263 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1264 | local_irq_restore(flags); |
1da177e4 LT |
1265 | } |
1266 | ||
949698a3 | 1267 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1268 | { |
c3993076 | 1269 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1270 | struct page *p = page; |
c3993076 | 1271 | unsigned int loop; |
a226f6c8 | 1272 | |
e2d0bd2b YL |
1273 | prefetchw(p); |
1274 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1275 | prefetchw(p + 1); | |
c3993076 JW |
1276 | __ClearPageReserved(p); |
1277 | set_page_count(p, 0); | |
a226f6c8 | 1278 | } |
e2d0bd2b YL |
1279 | __ClearPageReserved(p); |
1280 | set_page_count(p, 0); | |
c3993076 | 1281 | |
9705bea5 | 1282 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
c3993076 JW |
1283 | set_page_refcounted(page); |
1284 | __free_pages(page, order); | |
a226f6c8 DH |
1285 | } |
1286 | ||
75a592a4 MG |
1287 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1288 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1289 | |
75a592a4 MG |
1290 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1291 | ||
1292 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1293 | { | |
7ace9917 | 1294 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1295 | int nid; |
1296 | ||
7ace9917 | 1297 | spin_lock(&early_pfn_lock); |
75a592a4 | 1298 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1299 | if (nid < 0) |
e4568d38 | 1300 | nid = first_online_node; |
7ace9917 MG |
1301 | spin_unlock(&early_pfn_lock); |
1302 | ||
1303 | return nid; | |
75a592a4 MG |
1304 | } |
1305 | #endif | |
1306 | ||
1307 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
d73d3c9f MK |
1308 | static inline bool __meminit __maybe_unused |
1309 | meminit_pfn_in_nid(unsigned long pfn, int node, | |
1310 | struct mminit_pfnnid_cache *state) | |
75a592a4 MG |
1311 | { |
1312 | int nid; | |
1313 | ||
1314 | nid = __early_pfn_to_nid(pfn, state); | |
1315 | if (nid >= 0 && nid != node) | |
1316 | return false; | |
1317 | return true; | |
1318 | } | |
1319 | ||
1320 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1321 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1322 | { | |
1323 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1324 | } | |
1325 | ||
1326 | #else | |
1327 | ||
1328 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1329 | { | |
1330 | return true; | |
1331 | } | |
d73d3c9f MK |
1332 | static inline bool __meminit __maybe_unused |
1333 | meminit_pfn_in_nid(unsigned long pfn, int node, | |
1334 | struct mminit_pfnnid_cache *state) | |
75a592a4 MG |
1335 | { |
1336 | return true; | |
1337 | } | |
1338 | #endif | |
1339 | ||
1340 | ||
7c2ee349 | 1341 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1342 | unsigned int order) |
1343 | { | |
1344 | if (early_page_uninitialised(pfn)) | |
1345 | return; | |
949698a3 | 1346 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1347 | } |
1348 | ||
7cf91a98 JK |
1349 | /* |
1350 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1351 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1352 | * with the migration of free compaction scanner. The scanners then need to | |
1353 | * use only pfn_valid_within() check for arches that allow holes within | |
1354 | * pageblocks. | |
1355 | * | |
1356 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1357 | * | |
1358 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1359 | * i.e. it's possible that all pages within a zones range of pages do not | |
1360 | * belong to a single zone. We assume that a border between node0 and node1 | |
1361 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1362 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1363 | * the first and last page of a pageblock and avoid checking each individual | |
1364 | * page in a pageblock. | |
1365 | */ | |
1366 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1367 | unsigned long end_pfn, struct zone *zone) | |
1368 | { | |
1369 | struct page *start_page; | |
1370 | struct page *end_page; | |
1371 | ||
1372 | /* end_pfn is one past the range we are checking */ | |
1373 | end_pfn--; | |
1374 | ||
1375 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1376 | return NULL; | |
1377 | ||
2d070eab MH |
1378 | start_page = pfn_to_online_page(start_pfn); |
1379 | if (!start_page) | |
1380 | return NULL; | |
7cf91a98 JK |
1381 | |
1382 | if (page_zone(start_page) != zone) | |
1383 | return NULL; | |
1384 | ||
1385 | end_page = pfn_to_page(end_pfn); | |
1386 | ||
1387 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1388 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1389 | return NULL; | |
1390 | ||
1391 | return start_page; | |
1392 | } | |
1393 | ||
1394 | void set_zone_contiguous(struct zone *zone) | |
1395 | { | |
1396 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1397 | unsigned long block_end_pfn; | |
1398 | ||
1399 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1400 | for (; block_start_pfn < zone_end_pfn(zone); | |
1401 | block_start_pfn = block_end_pfn, | |
1402 | block_end_pfn += pageblock_nr_pages) { | |
1403 | ||
1404 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1405 | ||
1406 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1407 | block_end_pfn, zone)) | |
1408 | return; | |
1409 | } | |
1410 | ||
1411 | /* We confirm that there is no hole */ | |
1412 | zone->contiguous = true; | |
1413 | } | |
1414 | ||
1415 | void clear_zone_contiguous(struct zone *zone) | |
1416 | { | |
1417 | zone->contiguous = false; | |
1418 | } | |
1419 | ||
7e18adb4 | 1420 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1421 | static void __init deferred_free_range(unsigned long pfn, |
1422 | unsigned long nr_pages) | |
a4de83dd | 1423 | { |
2f47a91f PT |
1424 | struct page *page; |
1425 | unsigned long i; | |
a4de83dd | 1426 | |
2f47a91f | 1427 | if (!nr_pages) |
a4de83dd MG |
1428 | return; |
1429 | ||
2f47a91f PT |
1430 | page = pfn_to_page(pfn); |
1431 | ||
a4de83dd | 1432 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1433 | if (nr_pages == pageblock_nr_pages && |
1434 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1435 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
e780149b | 1436 | __free_pages_boot_core(page, pageblock_order); |
a4de83dd MG |
1437 | return; |
1438 | } | |
1439 | ||
e780149b XQ |
1440 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1441 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1442 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
949698a3 | 1443 | __free_pages_boot_core(page, 0); |
e780149b | 1444 | } |
a4de83dd MG |
1445 | } |
1446 | ||
d3cd131d NS |
1447 | /* Completion tracking for deferred_init_memmap() threads */ |
1448 | static atomic_t pgdat_init_n_undone __initdata; | |
1449 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1450 | ||
1451 | static inline void __init pgdat_init_report_one_done(void) | |
1452 | { | |
1453 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1454 | complete(&pgdat_init_all_done_comp); | |
1455 | } | |
0e1cc95b | 1456 | |
2f47a91f | 1457 | /* |
80b1f41c PT |
1458 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1459 | * | |
1460 | * First we check if pfn is valid on architectures where it is possible to have | |
1461 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1462 | * function is optimized out. | |
1463 | * | |
1464 | * Then, we check if a current large page is valid by only checking the validity | |
1465 | * of the head pfn. | |
1466 | * | |
1467 | * Finally, meminit_pfn_in_nid is checked on systems where pfns can interleave | |
1468 | * within a node: a pfn is between start and end of a node, but does not belong | |
1469 | * to this memory node. | |
2f47a91f | 1470 | */ |
80b1f41c PT |
1471 | static inline bool __init |
1472 | deferred_pfn_valid(int nid, unsigned long pfn, | |
1473 | struct mminit_pfnnid_cache *nid_init_state) | |
2f47a91f | 1474 | { |
80b1f41c PT |
1475 | if (!pfn_valid_within(pfn)) |
1476 | return false; | |
1477 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1478 | return false; | |
1479 | if (!meminit_pfn_in_nid(pfn, nid, nid_init_state)) | |
1480 | return false; | |
1481 | return true; | |
1482 | } | |
2f47a91f | 1483 | |
80b1f41c PT |
1484 | /* |
1485 | * Free pages to buddy allocator. Try to free aligned pages in | |
1486 | * pageblock_nr_pages sizes. | |
1487 | */ | |
1488 | static void __init deferred_free_pages(int nid, int zid, unsigned long pfn, | |
1489 | unsigned long end_pfn) | |
1490 | { | |
1491 | struct mminit_pfnnid_cache nid_init_state = { }; | |
1492 | unsigned long nr_pgmask = pageblock_nr_pages - 1; | |
1493 | unsigned long nr_free = 0; | |
2f47a91f | 1494 | |
80b1f41c PT |
1495 | for (; pfn < end_pfn; pfn++) { |
1496 | if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) { | |
1497 | deferred_free_range(pfn - nr_free, nr_free); | |
1498 | nr_free = 0; | |
1499 | } else if (!(pfn & nr_pgmask)) { | |
1500 | deferred_free_range(pfn - nr_free, nr_free); | |
1501 | nr_free = 1; | |
3a2d7fa8 | 1502 | touch_nmi_watchdog(); |
80b1f41c PT |
1503 | } else { |
1504 | nr_free++; | |
1505 | } | |
1506 | } | |
1507 | /* Free the last block of pages to allocator */ | |
1508 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1509 | } |
1510 | ||
80b1f41c PT |
1511 | /* |
1512 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1513 | * by performing it only once every pageblock_nr_pages. | |
1514 | * Return number of pages initialized. | |
1515 | */ | |
1516 | static unsigned long __init deferred_init_pages(int nid, int zid, | |
1517 | unsigned long pfn, | |
1518 | unsigned long end_pfn) | |
2f47a91f PT |
1519 | { |
1520 | struct mminit_pfnnid_cache nid_init_state = { }; | |
1521 | unsigned long nr_pgmask = pageblock_nr_pages - 1; | |
2f47a91f | 1522 | unsigned long nr_pages = 0; |
2f47a91f | 1523 | struct page *page = NULL; |
2f47a91f | 1524 | |
80b1f41c PT |
1525 | for (; pfn < end_pfn; pfn++) { |
1526 | if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) { | |
1527 | page = NULL; | |
2f47a91f | 1528 | continue; |
80b1f41c | 1529 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1530 | page = pfn_to_page(pfn); |
3a2d7fa8 | 1531 | touch_nmi_watchdog(); |
80b1f41c PT |
1532 | } else { |
1533 | page++; | |
2f47a91f | 1534 | } |
d0dc12e8 | 1535 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1536 | nr_pages++; |
2f47a91f | 1537 | } |
80b1f41c | 1538 | return (nr_pages); |
2f47a91f PT |
1539 | } |
1540 | ||
7e18adb4 | 1541 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1542 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1543 | { |
0e1cc95b MG |
1544 | pg_data_t *pgdat = data; |
1545 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1546 | unsigned long start = jiffies; |
1547 | unsigned long nr_pages = 0; | |
3a2d7fa8 | 1548 | unsigned long spfn, epfn, first_init_pfn, flags; |
2f47a91f PT |
1549 | phys_addr_t spa, epa; |
1550 | int zid; | |
7e18adb4 | 1551 | struct zone *zone; |
0e1cc95b | 1552 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
2f47a91f | 1553 | u64 i; |
7e18adb4 | 1554 | |
3a2d7fa8 PT |
1555 | /* Bind memory initialisation thread to a local node if possible */ |
1556 | if (!cpumask_empty(cpumask)) | |
1557 | set_cpus_allowed_ptr(current, cpumask); | |
1558 | ||
1559 | pgdat_resize_lock(pgdat, &flags); | |
1560 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1561 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1562 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1563 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1564 | return 0; |
1565 | } | |
1566 | ||
7e18adb4 MG |
1567 | /* Sanity check boundaries */ |
1568 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1569 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1570 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1571 | ||
1572 | /* Only the highest zone is deferred so find it */ | |
1573 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1574 | zone = pgdat->node_zones + zid; | |
1575 | if (first_init_pfn < zone_end_pfn(zone)) | |
1576 | break; | |
1577 | } | |
2f47a91f | 1578 | first_init_pfn = max(zone->zone_start_pfn, first_init_pfn); |
7e18adb4 | 1579 | |
80b1f41c PT |
1580 | /* |
1581 | * Initialize and free pages. We do it in two loops: first we initialize | |
1582 | * struct page, than free to buddy allocator, because while we are | |
1583 | * freeing pages we can access pages that are ahead (computing buddy | |
1584 | * page in __free_one_page()). | |
1585 | */ | |
1586 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { | |
1587 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1588 | epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa)); | |
1589 | nr_pages += deferred_init_pages(nid, zid, spfn, epfn); | |
1590 | } | |
2f47a91f PT |
1591 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { |
1592 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1593 | epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa)); | |
80b1f41c | 1594 | deferred_free_pages(nid, zid, spfn, epfn); |
7e18adb4 | 1595 | } |
3a2d7fa8 | 1596 | pgdat_resize_unlock(pgdat, &flags); |
7e18adb4 MG |
1597 | |
1598 | /* Sanity check that the next zone really is unpopulated */ | |
1599 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1600 | ||
0e1cc95b | 1601 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1602 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1603 | |
1604 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1605 | return 0; |
1606 | } | |
c9e97a19 PT |
1607 | |
1608 | /* | |
1609 | * During boot we initialize deferred pages on-demand, as needed, but once | |
1610 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
1611 | * and we can permanently disable that path. | |
1612 | */ | |
1613 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
1614 | ||
1615 | /* | |
1616 | * If this zone has deferred pages, try to grow it by initializing enough | |
1617 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
1618 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
1619 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
1620 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
1621 | * | |
1622 | * Return true when zone was grown, otherwise return false. We return true even | |
1623 | * when we grow less than requested, to let the caller decide if there are | |
1624 | * enough pages to satisfy the allocation. | |
1625 | * | |
1626 | * Note: We use noinline because this function is needed only during boot, and | |
1627 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
1628 | * making sure that it is not inlined into permanent text section. | |
1629 | */ | |
1630 | static noinline bool __init | |
1631 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
1632 | { | |
1633 | int zid = zone_idx(zone); | |
1634 | int nid = zone_to_nid(zone); | |
1635 | pg_data_t *pgdat = NODE_DATA(nid); | |
1636 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); | |
1637 | unsigned long nr_pages = 0; | |
1638 | unsigned long first_init_pfn, spfn, epfn, t, flags; | |
1639 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; | |
1640 | phys_addr_t spa, epa; | |
1641 | u64 i; | |
1642 | ||
1643 | /* Only the last zone may have deferred pages */ | |
1644 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
1645 | return false; | |
1646 | ||
1647 | pgdat_resize_lock(pgdat, &flags); | |
1648 | ||
1649 | /* | |
1650 | * If deferred pages have been initialized while we were waiting for | |
1651 | * the lock, return true, as the zone was grown. The caller will retry | |
1652 | * this zone. We won't return to this function since the caller also | |
1653 | * has this static branch. | |
1654 | */ | |
1655 | if (!static_branch_unlikely(&deferred_pages)) { | |
1656 | pgdat_resize_unlock(pgdat, &flags); | |
1657 | return true; | |
1658 | } | |
1659 | ||
1660 | /* | |
1661 | * If someone grew this zone while we were waiting for spinlock, return | |
1662 | * true, as there might be enough pages already. | |
1663 | */ | |
1664 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
1665 | pgdat_resize_unlock(pgdat, &flags); | |
1666 | return true; | |
1667 | } | |
1668 | ||
1669 | first_init_pfn = max(zone->zone_start_pfn, first_deferred_pfn); | |
1670 | ||
1671 | if (first_init_pfn >= pgdat_end_pfn(pgdat)) { | |
1672 | pgdat_resize_unlock(pgdat, &flags); | |
1673 | return false; | |
1674 | } | |
1675 | ||
1676 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { | |
1677 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1678 | epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa)); | |
1679 | ||
1680 | while (spfn < epfn && nr_pages < nr_pages_needed) { | |
1681 | t = ALIGN(spfn + PAGES_PER_SECTION, PAGES_PER_SECTION); | |
1682 | first_deferred_pfn = min(t, epfn); | |
1683 | nr_pages += deferred_init_pages(nid, zid, spfn, | |
1684 | first_deferred_pfn); | |
1685 | spfn = first_deferred_pfn; | |
1686 | } | |
1687 | ||
1688 | if (nr_pages >= nr_pages_needed) | |
1689 | break; | |
1690 | } | |
1691 | ||
1692 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) { | |
1693 | spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa)); | |
1694 | epfn = min_t(unsigned long, first_deferred_pfn, PFN_DOWN(epa)); | |
1695 | deferred_free_pages(nid, zid, spfn, epfn); | |
1696 | ||
1697 | if (first_deferred_pfn == epfn) | |
1698 | break; | |
1699 | } | |
1700 | pgdat->first_deferred_pfn = first_deferred_pfn; | |
1701 | pgdat_resize_unlock(pgdat, &flags); | |
1702 | ||
1703 | return nr_pages > 0; | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * deferred_grow_zone() is __init, but it is called from | |
1708 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
1709 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
1710 | * and to ensure that the function body gets unloaded. | |
1711 | */ | |
1712 | static bool __ref | |
1713 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
1714 | { | |
1715 | return deferred_grow_zone(zone, order); | |
1716 | } | |
1717 | ||
7cf91a98 | 1718 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1719 | |
1720 | void __init page_alloc_init_late(void) | |
1721 | { | |
7cf91a98 JK |
1722 | struct zone *zone; |
1723 | ||
1724 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1725 | int nid; |
1726 | ||
d3cd131d NS |
1727 | /* There will be num_node_state(N_MEMORY) threads */ |
1728 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1729 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1730 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1731 | } | |
1732 | ||
1733 | /* Block until all are initialised */ | |
d3cd131d | 1734 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 1735 | |
c9e97a19 PT |
1736 | /* |
1737 | * We initialized the rest of the deferred pages. Permanently disable | |
1738 | * on-demand struct page initialization. | |
1739 | */ | |
1740 | static_branch_disable(&deferred_pages); | |
1741 | ||
4248b0da MG |
1742 | /* Reinit limits that are based on free pages after the kernel is up */ |
1743 | files_maxfiles_init(); | |
7cf91a98 | 1744 | #endif |
3010f876 PT |
1745 | #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK |
1746 | /* Discard memblock private memory */ | |
1747 | memblock_discard(); | |
1748 | #endif | |
7cf91a98 JK |
1749 | |
1750 | for_each_populated_zone(zone) | |
1751 | set_zone_contiguous(zone); | |
7e18adb4 | 1752 | } |
7e18adb4 | 1753 | |
47118af0 | 1754 | #ifdef CONFIG_CMA |
9cf510a5 | 1755 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1756 | void __init init_cma_reserved_pageblock(struct page *page) |
1757 | { | |
1758 | unsigned i = pageblock_nr_pages; | |
1759 | struct page *p = page; | |
1760 | ||
1761 | do { | |
1762 | __ClearPageReserved(p); | |
1763 | set_page_count(p, 0); | |
d883c6cf | 1764 | } while (++p, --i); |
47118af0 | 1765 | |
47118af0 | 1766 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1767 | |
1768 | if (pageblock_order >= MAX_ORDER) { | |
1769 | i = pageblock_nr_pages; | |
1770 | p = page; | |
1771 | do { | |
1772 | set_page_refcounted(p); | |
1773 | __free_pages(p, MAX_ORDER - 1); | |
1774 | p += MAX_ORDER_NR_PAGES; | |
1775 | } while (i -= MAX_ORDER_NR_PAGES); | |
1776 | } else { | |
1777 | set_page_refcounted(page); | |
1778 | __free_pages(page, pageblock_order); | |
1779 | } | |
1780 | ||
3dcc0571 | 1781 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1782 | } |
1783 | #endif | |
1da177e4 LT |
1784 | |
1785 | /* | |
1786 | * The order of subdivision here is critical for the IO subsystem. | |
1787 | * Please do not alter this order without good reasons and regression | |
1788 | * testing. Specifically, as large blocks of memory are subdivided, | |
1789 | * the order in which smaller blocks are delivered depends on the order | |
1790 | * they're subdivided in this function. This is the primary factor | |
1791 | * influencing the order in which pages are delivered to the IO | |
1792 | * subsystem according to empirical testing, and this is also justified | |
1793 | * by considering the behavior of a buddy system containing a single | |
1794 | * large block of memory acted on by a series of small allocations. | |
1795 | * This behavior is a critical factor in sglist merging's success. | |
1796 | * | |
6d49e352 | 1797 | * -- nyc |
1da177e4 | 1798 | */ |
085cc7d5 | 1799 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1800 | int low, int high, struct free_area *area, |
1801 | int migratetype) | |
1da177e4 LT |
1802 | { |
1803 | unsigned long size = 1 << high; | |
1804 | ||
1805 | while (high > low) { | |
1806 | area--; | |
1807 | high--; | |
1808 | size >>= 1; | |
309381fe | 1809 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1810 | |
acbc15a4 JK |
1811 | /* |
1812 | * Mark as guard pages (or page), that will allow to | |
1813 | * merge back to allocator when buddy will be freed. | |
1814 | * Corresponding page table entries will not be touched, | |
1815 | * pages will stay not present in virtual address space | |
1816 | */ | |
1817 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 1818 | continue; |
acbc15a4 | 1819 | |
b2a0ac88 | 1820 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1821 | area->nr_free++; |
1822 | set_page_order(&page[size], high); | |
1823 | } | |
1da177e4 LT |
1824 | } |
1825 | ||
4e611801 | 1826 | static void check_new_page_bad(struct page *page) |
1da177e4 | 1827 | { |
4e611801 VB |
1828 | const char *bad_reason = NULL; |
1829 | unsigned long bad_flags = 0; | |
7bfec6f4 | 1830 | |
53f9263b | 1831 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1832 | bad_reason = "nonzero mapcount"; |
1833 | if (unlikely(page->mapping != NULL)) | |
1834 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1835 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1836 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1837 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1838 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1839 | bad_flags = __PG_HWPOISON; | |
e570f56c NH |
1840 | /* Don't complain about hwpoisoned pages */ |
1841 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1842 | return; | |
f4c18e6f | 1843 | } |
f0b791a3 DH |
1844 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1845 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1846 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1847 | } | |
9edad6ea JW |
1848 | #ifdef CONFIG_MEMCG |
1849 | if (unlikely(page->mem_cgroup)) | |
1850 | bad_reason = "page still charged to cgroup"; | |
1851 | #endif | |
4e611801 VB |
1852 | bad_page(page, bad_reason, bad_flags); |
1853 | } | |
1854 | ||
1855 | /* | |
1856 | * This page is about to be returned from the page allocator | |
1857 | */ | |
1858 | static inline int check_new_page(struct page *page) | |
1859 | { | |
1860 | if (likely(page_expected_state(page, | |
1861 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1862 | return 0; | |
1863 | ||
1864 | check_new_page_bad(page); | |
1865 | return 1; | |
2a7684a2 WF |
1866 | } |
1867 | ||
bd33ef36 | 1868 | static inline bool free_pages_prezeroed(void) |
1414c7f4 LA |
1869 | { |
1870 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
bd33ef36 | 1871 | page_poisoning_enabled(); |
1414c7f4 LA |
1872 | } |
1873 | ||
479f854a MG |
1874 | #ifdef CONFIG_DEBUG_VM |
1875 | static bool check_pcp_refill(struct page *page) | |
1876 | { | |
1877 | return false; | |
1878 | } | |
1879 | ||
1880 | static bool check_new_pcp(struct page *page) | |
1881 | { | |
1882 | return check_new_page(page); | |
1883 | } | |
1884 | #else | |
1885 | static bool check_pcp_refill(struct page *page) | |
1886 | { | |
1887 | return check_new_page(page); | |
1888 | } | |
1889 | static bool check_new_pcp(struct page *page) | |
1890 | { | |
1891 | return false; | |
1892 | } | |
1893 | #endif /* CONFIG_DEBUG_VM */ | |
1894 | ||
1895 | static bool check_new_pages(struct page *page, unsigned int order) | |
1896 | { | |
1897 | int i; | |
1898 | for (i = 0; i < (1 << order); i++) { | |
1899 | struct page *p = page + i; | |
1900 | ||
1901 | if (unlikely(check_new_page(p))) | |
1902 | return true; | |
1903 | } | |
1904 | ||
1905 | return false; | |
1906 | } | |
1907 | ||
46f24fd8 JK |
1908 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1909 | gfp_t gfp_flags) | |
1910 | { | |
1911 | set_page_private(page, 0); | |
1912 | set_page_refcounted(page); | |
1913 | ||
1914 | arch_alloc_page(page, order); | |
1915 | kernel_map_pages(page, 1 << order, 1); | |
1916 | kernel_poison_pages(page, 1 << order, 1); | |
1917 | kasan_alloc_pages(page, order); | |
1918 | set_page_owner(page, order, gfp_flags); | |
1919 | } | |
1920 | ||
479f854a | 1921 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1922 | unsigned int alloc_flags) |
2a7684a2 WF |
1923 | { |
1924 | int i; | |
689bcebf | 1925 | |
46f24fd8 | 1926 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1927 | |
bd33ef36 | 1928 | if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1929 | for (i = 0; i < (1 << order); i++) |
1930 | clear_highpage(page + i); | |
17cf4406 NP |
1931 | |
1932 | if (order && (gfp_flags & __GFP_COMP)) | |
1933 | prep_compound_page(page, order); | |
1934 | ||
75379191 | 1935 | /* |
2f064f34 | 1936 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1937 | * allocate the page. The expectation is that the caller is taking |
1938 | * steps that will free more memory. The caller should avoid the page | |
1939 | * being used for !PFMEMALLOC purposes. | |
1940 | */ | |
2f064f34 MH |
1941 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1942 | set_page_pfmemalloc(page); | |
1943 | else | |
1944 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1945 | } |
1946 | ||
56fd56b8 MG |
1947 | /* |
1948 | * Go through the free lists for the given migratetype and remove | |
1949 | * the smallest available page from the freelists | |
1950 | */ | |
85ccc8fa | 1951 | static __always_inline |
728ec980 | 1952 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
1953 | int migratetype) |
1954 | { | |
1955 | unsigned int current_order; | |
b8af2941 | 1956 | struct free_area *area; |
56fd56b8 MG |
1957 | struct page *page; |
1958 | ||
1959 | /* Find a page of the appropriate size in the preferred list */ | |
1960 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1961 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1962 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1963 | struct page, lru); |
a16601c5 GT |
1964 | if (!page) |
1965 | continue; | |
56fd56b8 MG |
1966 | list_del(&page->lru); |
1967 | rmv_page_order(page); | |
1968 | area->nr_free--; | |
56fd56b8 | 1969 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1970 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1971 | return page; |
1972 | } | |
1973 | ||
1974 | return NULL; | |
1975 | } | |
1976 | ||
1977 | ||
b2a0ac88 MG |
1978 | /* |
1979 | * This array describes the order lists are fallen back to when | |
1980 | * the free lists for the desirable migrate type are depleted | |
1981 | */ | |
47118af0 | 1982 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e | 1983 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 1984 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 1985 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
47118af0 | 1986 | #ifdef CONFIG_CMA |
974a786e | 1987 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1988 | #endif |
194159fb | 1989 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1990 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1991 | #endif |
b2a0ac88 MG |
1992 | }; |
1993 | ||
dc67647b | 1994 | #ifdef CONFIG_CMA |
85ccc8fa | 1995 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
1996 | unsigned int order) |
1997 | { | |
1998 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1999 | } | |
2000 | #else | |
2001 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2002 | unsigned int order) { return NULL; } | |
2003 | #endif | |
2004 | ||
c361be55 MG |
2005 | /* |
2006 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 2007 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2008 | * boundary. If alignment is required, use move_freepages_block() |
2009 | */ | |
02aa0cdd | 2010 | static int move_freepages(struct zone *zone, |
b69a7288 | 2011 | struct page *start_page, struct page *end_page, |
02aa0cdd | 2012 | int migratetype, int *num_movable) |
c361be55 MG |
2013 | { |
2014 | struct page *page; | |
d00181b9 | 2015 | unsigned int order; |
d100313f | 2016 | int pages_moved = 0; |
c361be55 MG |
2017 | |
2018 | #ifndef CONFIG_HOLES_IN_ZONE | |
2019 | /* | |
2020 | * page_zone is not safe to call in this context when | |
2021 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
2022 | * anyway as we check zone boundaries in move_freepages_block(). | |
2023 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 2024 | * grouping pages by mobility |
c361be55 | 2025 | */ |
3e04040d AB |
2026 | VM_BUG_ON(pfn_valid(page_to_pfn(start_page)) && |
2027 | pfn_valid(page_to_pfn(end_page)) && | |
2028 | page_zone(start_page) != page_zone(end_page)); | |
c361be55 | 2029 | #endif |
c361be55 MG |
2030 | for (page = start_page; page <= end_page;) { |
2031 | if (!pfn_valid_within(page_to_pfn(page))) { | |
2032 | page++; | |
2033 | continue; | |
2034 | } | |
2035 | ||
f073bdc5 AB |
2036 | /* Make sure we are not inadvertently changing nodes */ |
2037 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2038 | ||
c361be55 | 2039 | if (!PageBuddy(page)) { |
02aa0cdd VB |
2040 | /* |
2041 | * We assume that pages that could be isolated for | |
2042 | * migration are movable. But we don't actually try | |
2043 | * isolating, as that would be expensive. | |
2044 | */ | |
2045 | if (num_movable && | |
2046 | (PageLRU(page) || __PageMovable(page))) | |
2047 | (*num_movable)++; | |
2048 | ||
c361be55 MG |
2049 | page++; |
2050 | continue; | |
2051 | } | |
2052 | ||
2053 | order = page_order(page); | |
84be48d8 KS |
2054 | list_move(&page->lru, |
2055 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 2056 | page += 1 << order; |
d100313f | 2057 | pages_moved += 1 << order; |
c361be55 MG |
2058 | } |
2059 | ||
d100313f | 2060 | return pages_moved; |
c361be55 MG |
2061 | } |
2062 | ||
ee6f509c | 2063 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2064 | int migratetype, int *num_movable) |
c361be55 MG |
2065 | { |
2066 | unsigned long start_pfn, end_pfn; | |
2067 | struct page *start_page, *end_page; | |
2068 | ||
4a222127 DR |
2069 | if (num_movable) |
2070 | *num_movable = 0; | |
2071 | ||
c361be55 | 2072 | start_pfn = page_to_pfn(page); |
d9c23400 | 2073 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 2074 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
2075 | end_page = start_page + pageblock_nr_pages - 1; |
2076 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
2077 | |
2078 | /* Do not cross zone boundaries */ | |
108bcc96 | 2079 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 2080 | start_page = page; |
108bcc96 | 2081 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2082 | return 0; |
2083 | ||
02aa0cdd VB |
2084 | return move_freepages(zone, start_page, end_page, migratetype, |
2085 | num_movable); | |
c361be55 MG |
2086 | } |
2087 | ||
2f66a68f MG |
2088 | static void change_pageblock_range(struct page *pageblock_page, |
2089 | int start_order, int migratetype) | |
2090 | { | |
2091 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2092 | ||
2093 | while (nr_pageblocks--) { | |
2094 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2095 | pageblock_page += pageblock_nr_pages; | |
2096 | } | |
2097 | } | |
2098 | ||
fef903ef | 2099 | /* |
9c0415eb VB |
2100 | * When we are falling back to another migratetype during allocation, try to |
2101 | * steal extra free pages from the same pageblocks to satisfy further | |
2102 | * allocations, instead of polluting multiple pageblocks. | |
2103 | * | |
2104 | * If we are stealing a relatively large buddy page, it is likely there will | |
2105 | * be more free pages in the pageblock, so try to steal them all. For | |
2106 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2107 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2108 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2109 | * pageblocks. | |
fef903ef | 2110 | */ |
4eb7dce6 JK |
2111 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2112 | { | |
2113 | /* | |
2114 | * Leaving this order check is intended, although there is | |
2115 | * relaxed order check in next check. The reason is that | |
2116 | * we can actually steal whole pageblock if this condition met, | |
2117 | * but, below check doesn't guarantee it and that is just heuristic | |
2118 | * so could be changed anytime. | |
2119 | */ | |
2120 | if (order >= pageblock_order) | |
2121 | return true; | |
2122 | ||
2123 | if (order >= pageblock_order / 2 || | |
2124 | start_mt == MIGRATE_RECLAIMABLE || | |
2125 | start_mt == MIGRATE_UNMOVABLE || | |
2126 | page_group_by_mobility_disabled) | |
2127 | return true; | |
2128 | ||
2129 | return false; | |
2130 | } | |
2131 | ||
2132 | /* | |
2133 | * This function implements actual steal behaviour. If order is large enough, | |
2134 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2135 | * pageblock to our migratetype and determine how many already-allocated pages |
2136 | * are there in the pageblock with a compatible migratetype. If at least half | |
2137 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2138 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2139 | */ |
2140 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
3bc48f96 | 2141 | int start_type, bool whole_block) |
fef903ef | 2142 | { |
d00181b9 | 2143 | unsigned int current_order = page_order(page); |
3bc48f96 | 2144 | struct free_area *area; |
02aa0cdd VB |
2145 | int free_pages, movable_pages, alike_pages; |
2146 | int old_block_type; | |
2147 | ||
2148 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2149 | |
3bc48f96 VB |
2150 | /* |
2151 | * This can happen due to races and we want to prevent broken | |
2152 | * highatomic accounting. | |
2153 | */ | |
02aa0cdd | 2154 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2155 | goto single_page; |
2156 | ||
fef903ef SB |
2157 | /* Take ownership for orders >= pageblock_order */ |
2158 | if (current_order >= pageblock_order) { | |
2159 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2160 | goto single_page; |
fef903ef SB |
2161 | } |
2162 | ||
3bc48f96 VB |
2163 | /* We are not allowed to try stealing from the whole block */ |
2164 | if (!whole_block) | |
2165 | goto single_page; | |
2166 | ||
02aa0cdd VB |
2167 | free_pages = move_freepages_block(zone, page, start_type, |
2168 | &movable_pages); | |
2169 | /* | |
2170 | * Determine how many pages are compatible with our allocation. | |
2171 | * For movable allocation, it's the number of movable pages which | |
2172 | * we just obtained. For other types it's a bit more tricky. | |
2173 | */ | |
2174 | if (start_type == MIGRATE_MOVABLE) { | |
2175 | alike_pages = movable_pages; | |
2176 | } else { | |
2177 | /* | |
2178 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2179 | * to MOVABLE pageblock, consider all non-movable pages as | |
2180 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2181 | * vice versa, be conservative since we can't distinguish the | |
2182 | * exact migratetype of non-movable pages. | |
2183 | */ | |
2184 | if (old_block_type == MIGRATE_MOVABLE) | |
2185 | alike_pages = pageblock_nr_pages | |
2186 | - (free_pages + movable_pages); | |
2187 | else | |
2188 | alike_pages = 0; | |
2189 | } | |
2190 | ||
3bc48f96 | 2191 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2192 | if (!free_pages) |
3bc48f96 | 2193 | goto single_page; |
fef903ef | 2194 | |
02aa0cdd VB |
2195 | /* |
2196 | * If a sufficient number of pages in the block are either free or of | |
2197 | * comparable migratability as our allocation, claim the whole block. | |
2198 | */ | |
2199 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2200 | page_group_by_mobility_disabled) |
2201 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2202 | |
2203 | return; | |
2204 | ||
2205 | single_page: | |
2206 | area = &zone->free_area[current_order]; | |
2207 | list_move(&page->lru, &area->free_list[start_type]); | |
4eb7dce6 JK |
2208 | } |
2209 | ||
2149cdae JK |
2210 | /* |
2211 | * Check whether there is a suitable fallback freepage with requested order. | |
2212 | * If only_stealable is true, this function returns fallback_mt only if | |
2213 | * we can steal other freepages all together. This would help to reduce | |
2214 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2215 | */ | |
2216 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2217 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2218 | { |
2219 | int i; | |
2220 | int fallback_mt; | |
2221 | ||
2222 | if (area->nr_free == 0) | |
2223 | return -1; | |
2224 | ||
2225 | *can_steal = false; | |
2226 | for (i = 0;; i++) { | |
2227 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2228 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2229 | break; |
2230 | ||
2231 | if (list_empty(&area->free_list[fallback_mt])) | |
2232 | continue; | |
fef903ef | 2233 | |
4eb7dce6 JK |
2234 | if (can_steal_fallback(order, migratetype)) |
2235 | *can_steal = true; | |
2236 | ||
2149cdae JK |
2237 | if (!only_stealable) |
2238 | return fallback_mt; | |
2239 | ||
2240 | if (*can_steal) | |
2241 | return fallback_mt; | |
fef903ef | 2242 | } |
4eb7dce6 JK |
2243 | |
2244 | return -1; | |
fef903ef SB |
2245 | } |
2246 | ||
0aaa29a5 MG |
2247 | /* |
2248 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2249 | * there are no empty page blocks that contain a page with a suitable order | |
2250 | */ | |
2251 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2252 | unsigned int alloc_order) | |
2253 | { | |
2254 | int mt; | |
2255 | unsigned long max_managed, flags; | |
2256 | ||
2257 | /* | |
2258 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2259 | * Check is race-prone but harmless. | |
2260 | */ | |
9705bea5 | 2261 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2262 | if (zone->nr_reserved_highatomic >= max_managed) |
2263 | return; | |
2264 | ||
2265 | spin_lock_irqsave(&zone->lock, flags); | |
2266 | ||
2267 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2268 | if (zone->nr_reserved_highatomic >= max_managed) | |
2269 | goto out_unlock; | |
2270 | ||
2271 | /* Yoink! */ | |
2272 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2273 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2274 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2275 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2276 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2277 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2278 | } |
2279 | ||
2280 | out_unlock: | |
2281 | spin_unlock_irqrestore(&zone->lock, flags); | |
2282 | } | |
2283 | ||
2284 | /* | |
2285 | * Used when an allocation is about to fail under memory pressure. This | |
2286 | * potentially hurts the reliability of high-order allocations when under | |
2287 | * intense memory pressure but failed atomic allocations should be easier | |
2288 | * to recover from than an OOM. | |
29fac03b MK |
2289 | * |
2290 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2291 | * pageblock is exhausted. | |
0aaa29a5 | 2292 | */ |
29fac03b MK |
2293 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2294 | bool force) | |
0aaa29a5 MG |
2295 | { |
2296 | struct zonelist *zonelist = ac->zonelist; | |
2297 | unsigned long flags; | |
2298 | struct zoneref *z; | |
2299 | struct zone *zone; | |
2300 | struct page *page; | |
2301 | int order; | |
04c8716f | 2302 | bool ret; |
0aaa29a5 MG |
2303 | |
2304 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2305 | ac->nodemask) { | |
29fac03b MK |
2306 | /* |
2307 | * Preserve at least one pageblock unless memory pressure | |
2308 | * is really high. | |
2309 | */ | |
2310 | if (!force && zone->nr_reserved_highatomic <= | |
2311 | pageblock_nr_pages) | |
0aaa29a5 MG |
2312 | continue; |
2313 | ||
2314 | spin_lock_irqsave(&zone->lock, flags); | |
2315 | for (order = 0; order < MAX_ORDER; order++) { | |
2316 | struct free_area *area = &(zone->free_area[order]); | |
2317 | ||
a16601c5 GT |
2318 | page = list_first_entry_or_null( |
2319 | &area->free_list[MIGRATE_HIGHATOMIC], | |
2320 | struct page, lru); | |
2321 | if (!page) | |
0aaa29a5 MG |
2322 | continue; |
2323 | ||
0aaa29a5 | 2324 | /* |
4855e4a7 MK |
2325 | * In page freeing path, migratetype change is racy so |
2326 | * we can counter several free pages in a pageblock | |
2327 | * in this loop althoug we changed the pageblock type | |
2328 | * from highatomic to ac->migratetype. So we should | |
2329 | * adjust the count once. | |
0aaa29a5 | 2330 | */ |
a6ffdc07 | 2331 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2332 | /* |
2333 | * It should never happen but changes to | |
2334 | * locking could inadvertently allow a per-cpu | |
2335 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2336 | * while unreserving so be safe and watch for | |
2337 | * underflows. | |
2338 | */ | |
2339 | zone->nr_reserved_highatomic -= min( | |
2340 | pageblock_nr_pages, | |
2341 | zone->nr_reserved_highatomic); | |
2342 | } | |
0aaa29a5 MG |
2343 | |
2344 | /* | |
2345 | * Convert to ac->migratetype and avoid the normal | |
2346 | * pageblock stealing heuristics. Minimally, the caller | |
2347 | * is doing the work and needs the pages. More | |
2348 | * importantly, if the block was always converted to | |
2349 | * MIGRATE_UNMOVABLE or another type then the number | |
2350 | * of pageblocks that cannot be completely freed | |
2351 | * may increase. | |
2352 | */ | |
2353 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2354 | ret = move_freepages_block(zone, page, ac->migratetype, |
2355 | NULL); | |
29fac03b MK |
2356 | if (ret) { |
2357 | spin_unlock_irqrestore(&zone->lock, flags); | |
2358 | return ret; | |
2359 | } | |
0aaa29a5 MG |
2360 | } |
2361 | spin_unlock_irqrestore(&zone->lock, flags); | |
2362 | } | |
04c8716f MK |
2363 | |
2364 | return false; | |
0aaa29a5 MG |
2365 | } |
2366 | ||
3bc48f96 VB |
2367 | /* |
2368 | * Try finding a free buddy page on the fallback list and put it on the free | |
2369 | * list of requested migratetype, possibly along with other pages from the same | |
2370 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2371 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2372 | * |
2373 | * The use of signed ints for order and current_order is a deliberate | |
2374 | * deviation from the rest of this file, to make the for loop | |
2375 | * condition simpler. | |
3bc48f96 | 2376 | */ |
85ccc8fa | 2377 | static __always_inline bool |
6bb15450 MG |
2378 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2379 | unsigned int alloc_flags) | |
b2a0ac88 | 2380 | { |
b8af2941 | 2381 | struct free_area *area; |
b002529d | 2382 | int current_order; |
6bb15450 | 2383 | int min_order = order; |
b2a0ac88 | 2384 | struct page *page; |
4eb7dce6 JK |
2385 | int fallback_mt; |
2386 | bool can_steal; | |
b2a0ac88 | 2387 | |
6bb15450 MG |
2388 | /* |
2389 | * Do not steal pages from freelists belonging to other pageblocks | |
2390 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2391 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2392 | */ | |
2393 | if (alloc_flags & ALLOC_NOFRAGMENT) | |
2394 | min_order = pageblock_order; | |
2395 | ||
7a8f58f3 VB |
2396 | /* |
2397 | * Find the largest available free page in the other list. This roughly | |
2398 | * approximates finding the pageblock with the most free pages, which | |
2399 | * would be too costly to do exactly. | |
2400 | */ | |
6bb15450 | 2401 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 2402 | --current_order) { |
4eb7dce6 JK |
2403 | area = &(zone->free_area[current_order]); |
2404 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2405 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2406 | if (fallback_mt == -1) |
2407 | continue; | |
b2a0ac88 | 2408 | |
7a8f58f3 VB |
2409 | /* |
2410 | * We cannot steal all free pages from the pageblock and the | |
2411 | * requested migratetype is movable. In that case it's better to | |
2412 | * steal and split the smallest available page instead of the | |
2413 | * largest available page, because even if the next movable | |
2414 | * allocation falls back into a different pageblock than this | |
2415 | * one, it won't cause permanent fragmentation. | |
2416 | */ | |
2417 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2418 | && current_order > order) | |
2419 | goto find_smallest; | |
b2a0ac88 | 2420 | |
7a8f58f3 VB |
2421 | goto do_steal; |
2422 | } | |
e0fff1bd | 2423 | |
7a8f58f3 | 2424 | return false; |
e0fff1bd | 2425 | |
7a8f58f3 VB |
2426 | find_smallest: |
2427 | for (current_order = order; current_order < MAX_ORDER; | |
2428 | current_order++) { | |
2429 | area = &(zone->free_area[current_order]); | |
2430 | fallback_mt = find_suitable_fallback(area, current_order, | |
2431 | start_migratetype, false, &can_steal); | |
2432 | if (fallback_mt != -1) | |
2433 | break; | |
b2a0ac88 MG |
2434 | } |
2435 | ||
7a8f58f3 VB |
2436 | /* |
2437 | * This should not happen - we already found a suitable fallback | |
2438 | * when looking for the largest page. | |
2439 | */ | |
2440 | VM_BUG_ON(current_order == MAX_ORDER); | |
2441 | ||
2442 | do_steal: | |
2443 | page = list_first_entry(&area->free_list[fallback_mt], | |
2444 | struct page, lru); | |
2445 | ||
2446 | steal_suitable_fallback(zone, page, start_migratetype, can_steal); | |
2447 | ||
2448 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2449 | start_migratetype, fallback_mt); | |
2450 | ||
2451 | return true; | |
2452 | ||
b2a0ac88 MG |
2453 | } |
2454 | ||
56fd56b8 | 2455 | /* |
1da177e4 LT |
2456 | * Do the hard work of removing an element from the buddy allocator. |
2457 | * Call me with the zone->lock already held. | |
2458 | */ | |
85ccc8fa | 2459 | static __always_inline struct page * |
6bb15450 MG |
2460 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2461 | unsigned int alloc_flags) | |
1da177e4 | 2462 | { |
1da177e4 LT |
2463 | struct page *page; |
2464 | ||
3bc48f96 | 2465 | retry: |
56fd56b8 | 2466 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2467 | if (unlikely(!page)) { |
dc67647b JK |
2468 | if (migratetype == MIGRATE_MOVABLE) |
2469 | page = __rmqueue_cma_fallback(zone, order); | |
2470 | ||
6bb15450 MG |
2471 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2472 | alloc_flags)) | |
3bc48f96 | 2473 | goto retry; |
728ec980 MG |
2474 | } |
2475 | ||
0d3d062a | 2476 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2477 | return page; |
1da177e4 LT |
2478 | } |
2479 | ||
5f63b720 | 2480 | /* |
1da177e4 LT |
2481 | * Obtain a specified number of elements from the buddy allocator, all under |
2482 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2483 | * Returns the number of new pages which were placed at *list. | |
2484 | */ | |
5f63b720 | 2485 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2486 | unsigned long count, struct list_head *list, |
6bb15450 | 2487 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2488 | { |
a6de734b | 2489 | int i, alloced = 0; |
5f63b720 | 2490 | |
d34b0733 | 2491 | spin_lock(&zone->lock); |
1da177e4 | 2492 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2493 | struct page *page = __rmqueue(zone, order, migratetype, |
2494 | alloc_flags); | |
085cc7d5 | 2495 | if (unlikely(page == NULL)) |
1da177e4 | 2496 | break; |
81eabcbe | 2497 | |
479f854a MG |
2498 | if (unlikely(check_pcp_refill(page))) |
2499 | continue; | |
2500 | ||
81eabcbe | 2501 | /* |
0fac3ba5 VB |
2502 | * Split buddy pages returned by expand() are received here in |
2503 | * physical page order. The page is added to the tail of | |
2504 | * caller's list. From the callers perspective, the linked list | |
2505 | * is ordered by page number under some conditions. This is | |
2506 | * useful for IO devices that can forward direction from the | |
2507 | * head, thus also in the physical page order. This is useful | |
2508 | * for IO devices that can merge IO requests if the physical | |
2509 | * pages are ordered properly. | |
81eabcbe | 2510 | */ |
0fac3ba5 | 2511 | list_add_tail(&page->lru, list); |
a6de734b | 2512 | alloced++; |
bb14c2c7 | 2513 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2514 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2515 | -(1 << order)); | |
1da177e4 | 2516 | } |
a6de734b MG |
2517 | |
2518 | /* | |
2519 | * i pages were removed from the buddy list even if some leak due | |
2520 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2521 | * on i. Do not confuse with 'alloced' which is the number of | |
2522 | * pages added to the pcp list. | |
2523 | */ | |
f2260e6b | 2524 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2525 | spin_unlock(&zone->lock); |
a6de734b | 2526 | return alloced; |
1da177e4 LT |
2527 | } |
2528 | ||
4ae7c039 | 2529 | #ifdef CONFIG_NUMA |
8fce4d8e | 2530 | /* |
4037d452 CL |
2531 | * Called from the vmstat counter updater to drain pagesets of this |
2532 | * currently executing processor on remote nodes after they have | |
2533 | * expired. | |
2534 | * | |
879336c3 CL |
2535 | * Note that this function must be called with the thread pinned to |
2536 | * a single processor. | |
8fce4d8e | 2537 | */ |
4037d452 | 2538 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2539 | { |
4ae7c039 | 2540 | unsigned long flags; |
7be12fc9 | 2541 | int to_drain, batch; |
4ae7c039 | 2542 | |
4037d452 | 2543 | local_irq_save(flags); |
4db0c3c2 | 2544 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2545 | to_drain = min(pcp->count, batch); |
77ba9062 | 2546 | if (to_drain > 0) |
2a13515c | 2547 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 2548 | local_irq_restore(flags); |
4ae7c039 CL |
2549 | } |
2550 | #endif | |
2551 | ||
9f8f2172 | 2552 | /* |
93481ff0 | 2553 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2554 | * |
2555 | * The processor must either be the current processor and the | |
2556 | * thread pinned to the current processor or a processor that | |
2557 | * is not online. | |
2558 | */ | |
93481ff0 | 2559 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2560 | { |
c54ad30c | 2561 | unsigned long flags; |
93481ff0 VB |
2562 | struct per_cpu_pageset *pset; |
2563 | struct per_cpu_pages *pcp; | |
1da177e4 | 2564 | |
93481ff0 VB |
2565 | local_irq_save(flags); |
2566 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2567 | |
93481ff0 | 2568 | pcp = &pset->pcp; |
77ba9062 | 2569 | if (pcp->count) |
93481ff0 | 2570 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
2571 | local_irq_restore(flags); |
2572 | } | |
3dfa5721 | 2573 | |
93481ff0 VB |
2574 | /* |
2575 | * Drain pcplists of all zones on the indicated processor. | |
2576 | * | |
2577 | * The processor must either be the current processor and the | |
2578 | * thread pinned to the current processor or a processor that | |
2579 | * is not online. | |
2580 | */ | |
2581 | static void drain_pages(unsigned int cpu) | |
2582 | { | |
2583 | struct zone *zone; | |
2584 | ||
2585 | for_each_populated_zone(zone) { | |
2586 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2587 | } |
2588 | } | |
1da177e4 | 2589 | |
9f8f2172 CL |
2590 | /* |
2591 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2592 | * |
2593 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2594 | * the single zone's pages. | |
9f8f2172 | 2595 | */ |
93481ff0 | 2596 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2597 | { |
93481ff0 VB |
2598 | int cpu = smp_processor_id(); |
2599 | ||
2600 | if (zone) | |
2601 | drain_pages_zone(cpu, zone); | |
2602 | else | |
2603 | drain_pages(cpu); | |
9f8f2172 CL |
2604 | } |
2605 | ||
0ccce3b9 MG |
2606 | static void drain_local_pages_wq(struct work_struct *work) |
2607 | { | |
a459eeb7 MH |
2608 | /* |
2609 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2610 | * we can race with cpu offline when the WQ can move this from | |
2611 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2612 | * cpu which is allright but we also have to make sure to not move to | |
2613 | * a different one. | |
2614 | */ | |
2615 | preempt_disable(); | |
0ccce3b9 | 2616 | drain_local_pages(NULL); |
a459eeb7 | 2617 | preempt_enable(); |
0ccce3b9 MG |
2618 | } |
2619 | ||
9f8f2172 | 2620 | /* |
74046494 GBY |
2621 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2622 | * | |
93481ff0 VB |
2623 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2624 | * | |
0ccce3b9 | 2625 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2626 | */ |
93481ff0 | 2627 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2628 | { |
74046494 | 2629 | int cpu; |
74046494 GBY |
2630 | |
2631 | /* | |
2632 | * Allocate in the BSS so we wont require allocation in | |
2633 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2634 | */ | |
2635 | static cpumask_t cpus_with_pcps; | |
2636 | ||
ce612879 MH |
2637 | /* |
2638 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2639 | * initialized. | |
2640 | */ | |
2641 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2642 | return; | |
2643 | ||
bd233f53 MG |
2644 | /* |
2645 | * Do not drain if one is already in progress unless it's specific to | |
2646 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2647 | * the drain to be complete when the call returns. | |
2648 | */ | |
2649 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2650 | if (!zone) | |
2651 | return; | |
2652 | mutex_lock(&pcpu_drain_mutex); | |
2653 | } | |
0ccce3b9 | 2654 | |
74046494 GBY |
2655 | /* |
2656 | * We don't care about racing with CPU hotplug event | |
2657 | * as offline notification will cause the notified | |
2658 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2659 | * disables preemption as part of its processing | |
2660 | */ | |
2661 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2662 | struct per_cpu_pageset *pcp; |
2663 | struct zone *z; | |
74046494 | 2664 | bool has_pcps = false; |
93481ff0 VB |
2665 | |
2666 | if (zone) { | |
74046494 | 2667 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2668 | if (pcp->pcp.count) |
74046494 | 2669 | has_pcps = true; |
93481ff0 VB |
2670 | } else { |
2671 | for_each_populated_zone(z) { | |
2672 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2673 | if (pcp->pcp.count) { | |
2674 | has_pcps = true; | |
2675 | break; | |
2676 | } | |
74046494 GBY |
2677 | } |
2678 | } | |
93481ff0 | 2679 | |
74046494 GBY |
2680 | if (has_pcps) |
2681 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2682 | else | |
2683 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2684 | } | |
0ccce3b9 | 2685 | |
bd233f53 MG |
2686 | for_each_cpu(cpu, &cpus_with_pcps) { |
2687 | struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu); | |
2688 | INIT_WORK(work, drain_local_pages_wq); | |
ce612879 | 2689 | queue_work_on(cpu, mm_percpu_wq, work); |
0ccce3b9 | 2690 | } |
bd233f53 MG |
2691 | for_each_cpu(cpu, &cpus_with_pcps) |
2692 | flush_work(per_cpu_ptr(&pcpu_drain, cpu)); | |
2693 | ||
2694 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2695 | } |
2696 | ||
296699de | 2697 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 2698 | |
556b969a CY |
2699 | /* |
2700 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
2701 | */ | |
2702 | #define WD_PAGE_COUNT (128*1024) | |
2703 | ||
1da177e4 LT |
2704 | void mark_free_pages(struct zone *zone) |
2705 | { | |
556b969a | 2706 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 2707 | unsigned long flags; |
7aeb09f9 | 2708 | unsigned int order, t; |
86760a2c | 2709 | struct page *page; |
1da177e4 | 2710 | |
8080fc03 | 2711 | if (zone_is_empty(zone)) |
1da177e4 LT |
2712 | return; |
2713 | ||
2714 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2715 | |
108bcc96 | 2716 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2717 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2718 | if (pfn_valid(pfn)) { | |
86760a2c | 2719 | page = pfn_to_page(pfn); |
ba6b0979 | 2720 | |
556b969a CY |
2721 | if (!--page_count) { |
2722 | touch_nmi_watchdog(); | |
2723 | page_count = WD_PAGE_COUNT; | |
2724 | } | |
2725 | ||
ba6b0979 JK |
2726 | if (page_zone(page) != zone) |
2727 | continue; | |
2728 | ||
7be98234 RW |
2729 | if (!swsusp_page_is_forbidden(page)) |
2730 | swsusp_unset_page_free(page); | |
f623f0db | 2731 | } |
1da177e4 | 2732 | |
b2a0ac88 | 2733 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2734 | list_for_each_entry(page, |
2735 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2736 | unsigned long i; |
1da177e4 | 2737 | |
86760a2c | 2738 | pfn = page_to_pfn(page); |
556b969a CY |
2739 | for (i = 0; i < (1UL << order); i++) { |
2740 | if (!--page_count) { | |
2741 | touch_nmi_watchdog(); | |
2742 | page_count = WD_PAGE_COUNT; | |
2743 | } | |
7be98234 | 2744 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 2745 | } |
f623f0db | 2746 | } |
b2a0ac88 | 2747 | } |
1da177e4 LT |
2748 | spin_unlock_irqrestore(&zone->lock, flags); |
2749 | } | |
e2c55dc8 | 2750 | #endif /* CONFIG_PM */ |
1da177e4 | 2751 | |
2d4894b5 | 2752 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 2753 | { |
5f8dcc21 | 2754 | int migratetype; |
1da177e4 | 2755 | |
4db7548c | 2756 | if (!free_pcp_prepare(page)) |
9cca35d4 | 2757 | return false; |
689bcebf | 2758 | |
dc4b0caf | 2759 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2760 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
2761 | return true; |
2762 | } | |
2763 | ||
2d4894b5 | 2764 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
2765 | { |
2766 | struct zone *zone = page_zone(page); | |
2767 | struct per_cpu_pages *pcp; | |
2768 | int migratetype; | |
2769 | ||
2770 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 2771 | __count_vm_event(PGFREE); |
da456f14 | 2772 | |
5f8dcc21 MG |
2773 | /* |
2774 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2775 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 2776 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
2777 | * areas back if necessary. Otherwise, we may have to free |
2778 | * excessively into the page allocator | |
2779 | */ | |
2780 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2781 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2782 | free_one_page(zone, page, pfn, 0, migratetype); |
9cca35d4 | 2783 | return; |
5f8dcc21 MG |
2784 | } |
2785 | migratetype = MIGRATE_MOVABLE; | |
2786 | } | |
2787 | ||
99dcc3e5 | 2788 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 2789 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 2790 | pcp->count++; |
48db57f8 | 2791 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2792 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb | 2793 | free_pcppages_bulk(zone, batch, pcp); |
48db57f8 | 2794 | } |
9cca35d4 | 2795 | } |
5f8dcc21 | 2796 | |
9cca35d4 MG |
2797 | /* |
2798 | * Free a 0-order page | |
9cca35d4 | 2799 | */ |
2d4894b5 | 2800 | void free_unref_page(struct page *page) |
9cca35d4 MG |
2801 | { |
2802 | unsigned long flags; | |
2803 | unsigned long pfn = page_to_pfn(page); | |
2804 | ||
2d4894b5 | 2805 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
2806 | return; |
2807 | ||
2808 | local_irq_save(flags); | |
2d4894b5 | 2809 | free_unref_page_commit(page, pfn); |
d34b0733 | 2810 | local_irq_restore(flags); |
1da177e4 LT |
2811 | } |
2812 | ||
cc59850e KK |
2813 | /* |
2814 | * Free a list of 0-order pages | |
2815 | */ | |
2d4894b5 | 2816 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
2817 | { |
2818 | struct page *page, *next; | |
9cca35d4 | 2819 | unsigned long flags, pfn; |
c24ad77d | 2820 | int batch_count = 0; |
9cca35d4 MG |
2821 | |
2822 | /* Prepare pages for freeing */ | |
2823 | list_for_each_entry_safe(page, next, list, lru) { | |
2824 | pfn = page_to_pfn(page); | |
2d4894b5 | 2825 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
2826 | list_del(&page->lru); |
2827 | set_page_private(page, pfn); | |
2828 | } | |
cc59850e | 2829 | |
9cca35d4 | 2830 | local_irq_save(flags); |
cc59850e | 2831 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
2832 | unsigned long pfn = page_private(page); |
2833 | ||
2834 | set_page_private(page, 0); | |
2d4894b5 MG |
2835 | trace_mm_page_free_batched(page); |
2836 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
2837 | |
2838 | /* | |
2839 | * Guard against excessive IRQ disabled times when we get | |
2840 | * a large list of pages to free. | |
2841 | */ | |
2842 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
2843 | local_irq_restore(flags); | |
2844 | batch_count = 0; | |
2845 | local_irq_save(flags); | |
2846 | } | |
cc59850e | 2847 | } |
9cca35d4 | 2848 | local_irq_restore(flags); |
cc59850e KK |
2849 | } |
2850 | ||
8dfcc9ba NP |
2851 | /* |
2852 | * split_page takes a non-compound higher-order page, and splits it into | |
2853 | * n (1<<order) sub-pages: page[0..n] | |
2854 | * Each sub-page must be freed individually. | |
2855 | * | |
2856 | * Note: this is probably too low level an operation for use in drivers. | |
2857 | * Please consult with lkml before using this in your driver. | |
2858 | */ | |
2859 | void split_page(struct page *page, unsigned int order) | |
2860 | { | |
2861 | int i; | |
2862 | ||
309381fe SL |
2863 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2864 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 2865 | |
a9627bc5 | 2866 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2867 | set_page_refcounted(page + i); |
a9627bc5 | 2868 | split_page_owner(page, order); |
8dfcc9ba | 2869 | } |
5853ff23 | 2870 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2871 | |
3c605096 | 2872 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2873 | { |
748446bb MG |
2874 | unsigned long watermark; |
2875 | struct zone *zone; | |
2139cbe6 | 2876 | int mt; |
748446bb MG |
2877 | |
2878 | BUG_ON(!PageBuddy(page)); | |
2879 | ||
2880 | zone = page_zone(page); | |
2e30abd1 | 2881 | mt = get_pageblock_migratetype(page); |
748446bb | 2882 | |
194159fb | 2883 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
2884 | /* |
2885 | * Obey watermarks as if the page was being allocated. We can | |
2886 | * emulate a high-order watermark check with a raised order-0 | |
2887 | * watermark, because we already know our high-order page | |
2888 | * exists. | |
2889 | */ | |
2890 | watermark = min_wmark_pages(zone) + (1UL << order); | |
d883c6cf | 2891 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
2892 | return 0; |
2893 | ||
8fb74b9f | 2894 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2895 | } |
748446bb MG |
2896 | |
2897 | /* Remove page from free list */ | |
2898 | list_del(&page->lru); | |
2899 | zone->free_area[order].nr_free--; | |
2900 | rmv_page_order(page); | |
2139cbe6 | 2901 | |
400bc7fd | 2902 | /* |
2903 | * Set the pageblock if the isolated page is at least half of a | |
2904 | * pageblock | |
2905 | */ | |
748446bb MG |
2906 | if (order >= pageblock_order - 1) { |
2907 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2908 | for (; page < endpage; page += pageblock_nr_pages) { |
2909 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 2910 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 2911 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
2912 | set_pageblock_migratetype(page, |
2913 | MIGRATE_MOVABLE); | |
2914 | } | |
748446bb MG |
2915 | } |
2916 | ||
f3a14ced | 2917 | |
8fb74b9f | 2918 | return 1UL << order; |
1fb3f8ca MG |
2919 | } |
2920 | ||
060e7417 MG |
2921 | /* |
2922 | * Update NUMA hit/miss statistics | |
2923 | * | |
2924 | * Must be called with interrupts disabled. | |
060e7417 | 2925 | */ |
41b6167e | 2926 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
2927 | { |
2928 | #ifdef CONFIG_NUMA | |
3a321d2a | 2929 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 2930 | |
4518085e KW |
2931 | /* skip numa counters update if numa stats is disabled */ |
2932 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2933 | return; | |
2934 | ||
c1093b74 | 2935 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 2936 | local_stat = NUMA_OTHER; |
060e7417 | 2937 | |
c1093b74 | 2938 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3a321d2a | 2939 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 2940 | else { |
3a321d2a KW |
2941 | __inc_numa_state(z, NUMA_MISS); |
2942 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 2943 | } |
3a321d2a | 2944 | __inc_numa_state(z, local_stat); |
060e7417 MG |
2945 | #endif |
2946 | } | |
2947 | ||
066b2393 MG |
2948 | /* Remove page from the per-cpu list, caller must protect the list */ |
2949 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
6bb15450 | 2950 | unsigned int alloc_flags, |
453f85d4 | 2951 | struct per_cpu_pages *pcp, |
066b2393 MG |
2952 | struct list_head *list) |
2953 | { | |
2954 | struct page *page; | |
2955 | ||
2956 | do { | |
2957 | if (list_empty(list)) { | |
2958 | pcp->count += rmqueue_bulk(zone, 0, | |
2959 | pcp->batch, list, | |
6bb15450 | 2960 | migratetype, alloc_flags); |
066b2393 MG |
2961 | if (unlikely(list_empty(list))) |
2962 | return NULL; | |
2963 | } | |
2964 | ||
453f85d4 | 2965 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
2966 | list_del(&page->lru); |
2967 | pcp->count--; | |
2968 | } while (check_new_pcp(page)); | |
2969 | ||
2970 | return page; | |
2971 | } | |
2972 | ||
2973 | /* Lock and remove page from the per-cpu list */ | |
2974 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
2975 | struct zone *zone, unsigned int order, | |
6bb15450 MG |
2976 | gfp_t gfp_flags, int migratetype, |
2977 | unsigned int alloc_flags) | |
066b2393 MG |
2978 | { |
2979 | struct per_cpu_pages *pcp; | |
2980 | struct list_head *list; | |
066b2393 | 2981 | struct page *page; |
d34b0733 | 2982 | unsigned long flags; |
066b2393 | 2983 | |
d34b0733 | 2984 | local_irq_save(flags); |
066b2393 MG |
2985 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2986 | list = &pcp->lists[migratetype]; | |
6bb15450 | 2987 | page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list); |
066b2393 MG |
2988 | if (page) { |
2989 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
2990 | zone_statistics(preferred_zone, zone); | |
2991 | } | |
d34b0733 | 2992 | local_irq_restore(flags); |
066b2393 MG |
2993 | return page; |
2994 | } | |
2995 | ||
1da177e4 | 2996 | /* |
75379191 | 2997 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2998 | */ |
0a15c3e9 | 2999 | static inline |
066b2393 | 3000 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3001 | struct zone *zone, unsigned int order, |
c603844b MG |
3002 | gfp_t gfp_flags, unsigned int alloc_flags, |
3003 | int migratetype) | |
1da177e4 LT |
3004 | { |
3005 | unsigned long flags; | |
689bcebf | 3006 | struct page *page; |
1da177e4 | 3007 | |
d34b0733 | 3008 | if (likely(order == 0)) { |
066b2393 | 3009 | page = rmqueue_pcplist(preferred_zone, zone, order, |
6bb15450 | 3010 | gfp_flags, migratetype, alloc_flags); |
066b2393 MG |
3011 | goto out; |
3012 | } | |
83b9355b | 3013 | |
066b2393 MG |
3014 | /* |
3015 | * We most definitely don't want callers attempting to | |
3016 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3017 | */ | |
3018 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3019 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3020 | |
066b2393 MG |
3021 | do { |
3022 | page = NULL; | |
3023 | if (alloc_flags & ALLOC_HARDER) { | |
3024 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3025 | if (page) | |
3026 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3027 | } | |
a74609fa | 3028 | if (!page) |
6bb15450 | 3029 | page = __rmqueue(zone, order, migratetype, alloc_flags); |
066b2393 MG |
3030 | } while (page && check_new_pages(page, order)); |
3031 | spin_unlock(&zone->lock); | |
3032 | if (!page) | |
3033 | goto failed; | |
3034 | __mod_zone_freepage_state(zone, -(1 << order), | |
3035 | get_pcppage_migratetype(page)); | |
1da177e4 | 3036 | |
16709d1d | 3037 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3038 | zone_statistics(preferred_zone, zone); |
a74609fa | 3039 | local_irq_restore(flags); |
1da177e4 | 3040 | |
066b2393 MG |
3041 | out: |
3042 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); | |
1da177e4 | 3043 | return page; |
a74609fa NP |
3044 | |
3045 | failed: | |
3046 | local_irq_restore(flags); | |
a74609fa | 3047 | return NULL; |
1da177e4 LT |
3048 | } |
3049 | ||
933e312e AM |
3050 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3051 | ||
b2588c4b | 3052 | static struct { |
933e312e AM |
3053 | struct fault_attr attr; |
3054 | ||
621a5f7a | 3055 | bool ignore_gfp_highmem; |
71baba4b | 3056 | bool ignore_gfp_reclaim; |
54114994 | 3057 | u32 min_order; |
933e312e AM |
3058 | } fail_page_alloc = { |
3059 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3060 | .ignore_gfp_reclaim = true, |
621a5f7a | 3061 | .ignore_gfp_highmem = true, |
54114994 | 3062 | .min_order = 1, |
933e312e AM |
3063 | }; |
3064 | ||
3065 | static int __init setup_fail_page_alloc(char *str) | |
3066 | { | |
3067 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3068 | } | |
3069 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3070 | ||
deaf386e | 3071 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3072 | { |
54114994 | 3073 | if (order < fail_page_alloc.min_order) |
deaf386e | 3074 | return false; |
933e312e | 3075 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3076 | return false; |
933e312e | 3077 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3078 | return false; |
71baba4b MG |
3079 | if (fail_page_alloc.ignore_gfp_reclaim && |
3080 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3081 | return false; |
933e312e AM |
3082 | |
3083 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3084 | } | |
3085 | ||
3086 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3087 | ||
3088 | static int __init fail_page_alloc_debugfs(void) | |
3089 | { | |
0825a6f9 | 3090 | umode_t mode = S_IFREG | 0600; |
933e312e | 3091 | struct dentry *dir; |
933e312e | 3092 | |
dd48c085 AM |
3093 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3094 | &fail_page_alloc.attr); | |
3095 | if (IS_ERR(dir)) | |
3096 | return PTR_ERR(dir); | |
933e312e | 3097 | |
b2588c4b | 3098 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 3099 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
3100 | goto fail; |
3101 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3102 | &fail_page_alloc.ignore_gfp_highmem)) | |
3103 | goto fail; | |
3104 | if (!debugfs_create_u32("min-order", mode, dir, | |
3105 | &fail_page_alloc.min_order)) | |
3106 | goto fail; | |
3107 | ||
3108 | return 0; | |
3109 | fail: | |
dd48c085 | 3110 | debugfs_remove_recursive(dir); |
933e312e | 3111 | |
b2588c4b | 3112 | return -ENOMEM; |
933e312e AM |
3113 | } |
3114 | ||
3115 | late_initcall(fail_page_alloc_debugfs); | |
3116 | ||
3117 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3118 | ||
3119 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3120 | ||
deaf386e | 3121 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3122 | { |
deaf386e | 3123 | return false; |
933e312e AM |
3124 | } |
3125 | ||
3126 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3127 | ||
1da177e4 | 3128 | /* |
97a16fc8 MG |
3129 | * Return true if free base pages are above 'mark'. For high-order checks it |
3130 | * will return true of the order-0 watermark is reached and there is at least | |
3131 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3132 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3133 | */ |
86a294a8 MH |
3134 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
3135 | int classzone_idx, unsigned int alloc_flags, | |
3136 | long free_pages) | |
1da177e4 | 3137 | { |
d23ad423 | 3138 | long min = mark; |
1da177e4 | 3139 | int o; |
cd04ae1e | 3140 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3141 | |
0aaa29a5 | 3142 | /* free_pages may go negative - that's OK */ |
df0a6daa | 3143 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 3144 | |
7fb1d9fc | 3145 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3146 | min -= min / 2; |
0aaa29a5 MG |
3147 | |
3148 | /* | |
3149 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3150 | * the high-atomic reserves. This will over-estimate the size of the | |
3151 | * atomic reserve but it avoids a search. | |
3152 | */ | |
cd04ae1e | 3153 | if (likely(!alloc_harder)) { |
0aaa29a5 | 3154 | free_pages -= z->nr_reserved_highatomic; |
cd04ae1e MH |
3155 | } else { |
3156 | /* | |
3157 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3158 | * users on the grounds that it's definitely going to be in | |
3159 | * the exit path shortly and free memory. Any allocation it | |
3160 | * makes during the free path will be small and short-lived. | |
3161 | */ | |
3162 | if (alloc_flags & ALLOC_OOM) | |
3163 | min -= min / 2; | |
3164 | else | |
3165 | min -= min / 4; | |
3166 | } | |
3167 | ||
e2b19197 | 3168 | |
d883c6cf JK |
3169 | #ifdef CONFIG_CMA |
3170 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3171 | if (!(alloc_flags & ALLOC_CMA)) | |
3172 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); | |
3173 | #endif | |
3174 | ||
97a16fc8 MG |
3175 | /* |
3176 | * Check watermarks for an order-0 allocation request. If these | |
3177 | * are not met, then a high-order request also cannot go ahead | |
3178 | * even if a suitable page happened to be free. | |
3179 | */ | |
3180 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 3181 | return false; |
1da177e4 | 3182 | |
97a16fc8 MG |
3183 | /* If this is an order-0 request then the watermark is fine */ |
3184 | if (!order) | |
3185 | return true; | |
3186 | ||
3187 | /* For a high-order request, check at least one suitable page is free */ | |
3188 | for (o = order; o < MAX_ORDER; o++) { | |
3189 | struct free_area *area = &z->free_area[o]; | |
3190 | int mt; | |
3191 | ||
3192 | if (!area->nr_free) | |
3193 | continue; | |
3194 | ||
97a16fc8 MG |
3195 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
3196 | if (!list_empty(&area->free_list[mt])) | |
3197 | return true; | |
3198 | } | |
3199 | ||
3200 | #ifdef CONFIG_CMA | |
d883c6cf JK |
3201 | if ((alloc_flags & ALLOC_CMA) && |
3202 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
97a16fc8 | 3203 | return true; |
d883c6cf | 3204 | } |
97a16fc8 | 3205 | #endif |
b050e376 VB |
3206 | if (alloc_harder && |
3207 | !list_empty(&area->free_list[MIGRATE_HIGHATOMIC])) | |
3208 | return true; | |
1da177e4 | 3209 | } |
97a16fc8 | 3210 | return false; |
88f5acf8 MG |
3211 | } |
3212 | ||
7aeb09f9 | 3213 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 3214 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
3215 | { |
3216 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3217 | zone_page_state(z, NR_FREE_PAGES)); | |
3218 | } | |
3219 | ||
48ee5f36 MG |
3220 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
3221 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
3222 | { | |
3223 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
d883c6cf JK |
3224 | long cma_pages = 0; |
3225 | ||
3226 | #ifdef CONFIG_CMA | |
3227 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3228 | if (!(alloc_flags & ALLOC_CMA)) | |
3229 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
3230 | #endif | |
48ee5f36 MG |
3231 | |
3232 | /* | |
3233 | * Fast check for order-0 only. If this fails then the reserves | |
3234 | * need to be calculated. There is a corner case where the check | |
3235 | * passes but only the high-order atomic reserve are free. If | |
3236 | * the caller is !atomic then it'll uselessly search the free | |
3237 | * list. That corner case is then slower but it is harmless. | |
3238 | */ | |
d883c6cf | 3239 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) |
48ee5f36 MG |
3240 | return true; |
3241 | ||
3242 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3243 | free_pages); | |
3244 | } | |
3245 | ||
7aeb09f9 | 3246 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 3247 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
3248 | { |
3249 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3250 | ||
3251 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3252 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3253 | ||
e2b19197 | 3254 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 3255 | free_pages); |
1da177e4 LT |
3256 | } |
3257 | ||
9276b1bc | 3258 | #ifdef CONFIG_NUMA |
957f822a DR |
3259 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3260 | { | |
e02dc017 | 3261 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
5f7a75ac | 3262 | RECLAIM_DISTANCE; |
957f822a | 3263 | } |
9276b1bc | 3264 | #else /* CONFIG_NUMA */ |
957f822a DR |
3265 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3266 | { | |
3267 | return true; | |
3268 | } | |
9276b1bc PJ |
3269 | #endif /* CONFIG_NUMA */ |
3270 | ||
6bb15450 MG |
3271 | #ifdef CONFIG_ZONE_DMA32 |
3272 | /* | |
3273 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3274 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3275 | * premature use of a lower zone may cause lowmem pressure problems that | |
3276 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3277 | * probably too small. It only makes sense to spread allocations to avoid | |
3278 | * fragmentation between the Normal and DMA32 zones. | |
3279 | */ | |
3280 | static inline unsigned int | |
3281 | alloc_flags_nofragment(struct zone *zone) | |
3282 | { | |
3283 | if (zone_idx(zone) != ZONE_NORMAL) | |
3284 | return 0; | |
3285 | ||
3286 | /* | |
3287 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3288 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3289 | * on UMA that if Normal is populated then so is DMA32. | |
3290 | */ | |
3291 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3292 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
3293 | return 0; | |
3294 | ||
3295 | return ALLOC_NOFRAGMENT; | |
3296 | } | |
3297 | #else | |
3298 | static inline unsigned int | |
3299 | alloc_flags_nofragment(struct zone *zone) | |
3300 | { | |
3301 | return 0; | |
3302 | } | |
3303 | #endif | |
3304 | ||
7fb1d9fc | 3305 | /* |
0798e519 | 3306 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3307 | * a page. |
3308 | */ | |
3309 | static struct page * | |
a9263751 VB |
3310 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3311 | const struct alloc_context *ac) | |
753ee728 | 3312 | { |
6bb15450 | 3313 | struct zoneref *z; |
5117f45d | 3314 | struct zone *zone; |
3b8c0be4 | 3315 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
6bb15450 | 3316 | bool no_fallback; |
3b8c0be4 | 3317 | |
6bb15450 | 3318 | retry: |
7fb1d9fc | 3319 | /* |
9276b1bc | 3320 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3321 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3322 | */ |
6bb15450 MG |
3323 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3324 | z = ac->preferred_zoneref; | |
c33d6c06 | 3325 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 3326 | ac->nodemask) { |
be06af00 | 3327 | struct page *page; |
e085dbc5 JW |
3328 | unsigned long mark; |
3329 | ||
664eedde MG |
3330 | if (cpusets_enabled() && |
3331 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3332 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3333 | continue; |
a756cf59 JW |
3334 | /* |
3335 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3336 | * want to get it from a node that is within its dirty |
3337 | * limit, such that no single node holds more than its | |
a756cf59 | 3338 | * proportional share of globally allowed dirty pages. |
281e3726 | 3339 | * The dirty limits take into account the node's |
a756cf59 JW |
3340 | * lowmem reserves and high watermark so that kswapd |
3341 | * should be able to balance it without having to | |
3342 | * write pages from its LRU list. | |
3343 | * | |
a756cf59 | 3344 | * XXX: For now, allow allocations to potentially |
281e3726 | 3345 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3346 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3347 | * which is important when on a NUMA setup the allowed |
281e3726 | 3348 | * nodes are together not big enough to reach the |
a756cf59 | 3349 | * global limit. The proper fix for these situations |
281e3726 | 3350 | * will require awareness of nodes in the |
a756cf59 JW |
3351 | * dirty-throttling and the flusher threads. |
3352 | */ | |
3b8c0be4 MG |
3353 | if (ac->spread_dirty_pages) { |
3354 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3355 | continue; | |
3356 | ||
3357 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3358 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3359 | continue; | |
3360 | } | |
3361 | } | |
7fb1d9fc | 3362 | |
6bb15450 MG |
3363 | if (no_fallback && nr_online_nodes > 1 && |
3364 | zone != ac->preferred_zoneref->zone) { | |
3365 | int local_nid; | |
3366 | ||
3367 | /* | |
3368 | * If moving to a remote node, retry but allow | |
3369 | * fragmenting fallbacks. Locality is more important | |
3370 | * than fragmentation avoidance. | |
3371 | */ | |
3372 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3373 | if (zone_to_nid(zone) != local_nid) { | |
3374 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3375 | goto retry; | |
3376 | } | |
3377 | } | |
3378 | ||
e085dbc5 | 3379 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 3380 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 3381 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
3382 | int ret; |
3383 | ||
c9e97a19 PT |
3384 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3385 | /* | |
3386 | * Watermark failed for this zone, but see if we can | |
3387 | * grow this zone if it contains deferred pages. | |
3388 | */ | |
3389 | if (static_branch_unlikely(&deferred_pages)) { | |
3390 | if (_deferred_grow_zone(zone, order)) | |
3391 | goto try_this_zone; | |
3392 | } | |
3393 | #endif | |
5dab2911 MG |
3394 | /* Checked here to keep the fast path fast */ |
3395 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3396 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3397 | goto try_this_zone; | |
3398 | ||
a5f5f91d | 3399 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3400 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3401 | continue; |
3402 | ||
a5f5f91d | 3403 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3404 | switch (ret) { |
a5f5f91d | 3405 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3406 | /* did not scan */ |
cd38b115 | 3407 | continue; |
a5f5f91d | 3408 | case NODE_RECLAIM_FULL: |
fa5e084e | 3409 | /* scanned but unreclaimable */ |
cd38b115 | 3410 | continue; |
fa5e084e MG |
3411 | default: |
3412 | /* did we reclaim enough */ | |
fed2719e | 3413 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3414 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3415 | goto try_this_zone; |
3416 | ||
fed2719e | 3417 | continue; |
0798e519 | 3418 | } |
7fb1d9fc RS |
3419 | } |
3420 | ||
fa5e084e | 3421 | try_this_zone: |
066b2393 | 3422 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3423 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3424 | if (page) { |
479f854a | 3425 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3426 | |
3427 | /* | |
3428 | * If this is a high-order atomic allocation then check | |
3429 | * if the pageblock should be reserved for the future | |
3430 | */ | |
3431 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3432 | reserve_highatomic_pageblock(page, zone, order); | |
3433 | ||
75379191 | 3434 | return page; |
c9e97a19 PT |
3435 | } else { |
3436 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
3437 | /* Try again if zone has deferred pages */ | |
3438 | if (static_branch_unlikely(&deferred_pages)) { | |
3439 | if (_deferred_grow_zone(zone, order)) | |
3440 | goto try_this_zone; | |
3441 | } | |
3442 | #endif | |
75379191 | 3443 | } |
54a6eb5c | 3444 | } |
9276b1bc | 3445 | |
6bb15450 MG |
3446 | /* |
3447 | * It's possible on a UMA machine to get through all zones that are | |
3448 | * fragmented. If avoiding fragmentation, reset and try again. | |
3449 | */ | |
3450 | if (no_fallback) { | |
3451 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3452 | goto retry; | |
3453 | } | |
3454 | ||
4ffeaf35 | 3455 | return NULL; |
753ee728 MH |
3456 | } |
3457 | ||
9af744d7 | 3458 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3459 | { |
a238ab5b | 3460 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
aa187507 | 3461 | static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1); |
a238ab5b | 3462 | |
2c029a1e | 3463 | if (!__ratelimit(&show_mem_rs)) |
a238ab5b DH |
3464 | return; |
3465 | ||
3466 | /* | |
3467 | * This documents exceptions given to allocations in certain | |
3468 | * contexts that are allowed to allocate outside current's set | |
3469 | * of allowed nodes. | |
3470 | */ | |
3471 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3472 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3473 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3474 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3475 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3476 | filter &= ~SHOW_MEM_FILTER_NODES; |
3477 | ||
9af744d7 | 3478 | show_mem(filter, nodemask); |
aa187507 MH |
3479 | } |
3480 | ||
a8e99259 | 3481 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3482 | { |
3483 | struct va_format vaf; | |
3484 | va_list args; | |
3485 | static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL, | |
3486 | DEFAULT_RATELIMIT_BURST); | |
3487 | ||
0f7896f1 | 3488 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3489 | return; |
3490 | ||
7877cdcc MH |
3491 | va_start(args, fmt); |
3492 | vaf.fmt = fmt; | |
3493 | vaf.va = &args; | |
0205f755 MH |
3494 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl\n", |
3495 | current->comm, &vaf, gfp_mask, &gfp_mask, | |
3496 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3497 | va_end(args); |
3ee9a4f0 | 3498 | |
a8e99259 | 3499 | cpuset_print_current_mems_allowed(); |
3ee9a4f0 | 3500 | |
a238ab5b | 3501 | dump_stack(); |
685dbf6f | 3502 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3503 | } |
3504 | ||
6c18ba7a MH |
3505 | static inline struct page * |
3506 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3507 | unsigned int alloc_flags, | |
3508 | const struct alloc_context *ac) | |
3509 | { | |
3510 | struct page *page; | |
3511 | ||
3512 | page = get_page_from_freelist(gfp_mask, order, | |
3513 | alloc_flags|ALLOC_CPUSET, ac); | |
3514 | /* | |
3515 | * fallback to ignore cpuset restriction if our nodes | |
3516 | * are depleted | |
3517 | */ | |
3518 | if (!page) | |
3519 | page = get_page_from_freelist(gfp_mask, order, | |
3520 | alloc_flags, ac); | |
3521 | ||
3522 | return page; | |
3523 | } | |
3524 | ||
11e33f6a MG |
3525 | static inline struct page * |
3526 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3527 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3528 | { |
6e0fc46d DR |
3529 | struct oom_control oc = { |
3530 | .zonelist = ac->zonelist, | |
3531 | .nodemask = ac->nodemask, | |
2a966b77 | 3532 | .memcg = NULL, |
6e0fc46d DR |
3533 | .gfp_mask = gfp_mask, |
3534 | .order = order, | |
6e0fc46d | 3535 | }; |
11e33f6a MG |
3536 | struct page *page; |
3537 | ||
9879de73 JW |
3538 | *did_some_progress = 0; |
3539 | ||
9879de73 | 3540 | /* |
dc56401f JW |
3541 | * Acquire the oom lock. If that fails, somebody else is |
3542 | * making progress for us. | |
9879de73 | 3543 | */ |
dc56401f | 3544 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3545 | *did_some_progress = 1; |
11e33f6a | 3546 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3547 | return NULL; |
3548 | } | |
6b1de916 | 3549 | |
11e33f6a MG |
3550 | /* |
3551 | * Go through the zonelist yet one more time, keep very high watermark | |
3552 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3553 | * we're still under heavy pressure. But make sure that this reclaim |
3554 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3555 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3556 | */ |
e746bf73 TH |
3557 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3558 | ~__GFP_DIRECT_RECLAIM, order, | |
3559 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3560 | if (page) |
11e33f6a MG |
3561 | goto out; |
3562 | ||
06ad276a MH |
3563 | /* Coredumps can quickly deplete all memory reserves */ |
3564 | if (current->flags & PF_DUMPCORE) | |
3565 | goto out; | |
3566 | /* The OOM killer will not help higher order allocs */ | |
3567 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3568 | goto out; | |
dcda9b04 MH |
3569 | /* |
3570 | * We have already exhausted all our reclaim opportunities without any | |
3571 | * success so it is time to admit defeat. We will skip the OOM killer | |
3572 | * because it is very likely that the caller has a more reasonable | |
3573 | * fallback than shooting a random task. | |
3574 | */ | |
3575 | if (gfp_mask & __GFP_RETRY_MAYFAIL) | |
3576 | goto out; | |
06ad276a MH |
3577 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
3578 | if (ac->high_zoneidx < ZONE_NORMAL) | |
3579 | goto out; | |
3580 | if (pm_suspended_storage()) | |
3581 | goto out; | |
3582 | /* | |
3583 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3584 | * other request to make a forward progress. | |
3585 | * We are in an unfortunate situation where out_of_memory cannot | |
3586 | * do much for this context but let's try it to at least get | |
3587 | * access to memory reserved if the current task is killed (see | |
3588 | * out_of_memory). Once filesystems are ready to handle allocation | |
3589 | * failures more gracefully we should just bail out here. | |
3590 | */ | |
3591 | ||
3592 | /* The OOM killer may not free memory on a specific node */ | |
3593 | if (gfp_mask & __GFP_THISNODE) | |
3594 | goto out; | |
3da88fb3 | 3595 | |
3c2c6488 | 3596 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 3597 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3598 | *did_some_progress = 1; |
5020e285 | 3599 | |
6c18ba7a MH |
3600 | /* |
3601 | * Help non-failing allocations by giving them access to memory | |
3602 | * reserves | |
3603 | */ | |
3604 | if (gfp_mask & __GFP_NOFAIL) | |
3605 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3606 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3607 | } |
11e33f6a | 3608 | out: |
dc56401f | 3609 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3610 | return page; |
3611 | } | |
3612 | ||
33c2d214 MH |
3613 | /* |
3614 | * Maximum number of compaction retries wit a progress before OOM | |
3615 | * killer is consider as the only way to move forward. | |
3616 | */ | |
3617 | #define MAX_COMPACT_RETRIES 16 | |
3618 | ||
56de7263 MG |
3619 | #ifdef CONFIG_COMPACTION |
3620 | /* Try memory compaction for high-order allocations before reclaim */ | |
3621 | static struct page * | |
3622 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3623 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3624 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3625 | { |
98dd3b48 | 3626 | struct page *page; |
eb414681 | 3627 | unsigned long pflags; |
499118e9 | 3628 | unsigned int noreclaim_flag; |
53853e2d VB |
3629 | |
3630 | if (!order) | |
66199712 | 3631 | return NULL; |
66199712 | 3632 | |
eb414681 | 3633 | psi_memstall_enter(&pflags); |
499118e9 | 3634 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 3635 | |
c5d01d0d | 3636 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
c3486f53 | 3637 | prio); |
eb414681 | 3638 | |
499118e9 | 3639 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 3640 | psi_memstall_leave(&pflags); |
56de7263 | 3641 | |
c5d01d0d | 3642 | if (*compact_result <= COMPACT_INACTIVE) |
98dd3b48 | 3643 | return NULL; |
53853e2d | 3644 | |
98dd3b48 VB |
3645 | /* |
3646 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3647 | * count a compaction stall | |
3648 | */ | |
3649 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3650 | |
31a6c190 | 3651 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
53853e2d | 3652 | |
98dd3b48 VB |
3653 | if (page) { |
3654 | struct zone *zone = page_zone(page); | |
53853e2d | 3655 | |
98dd3b48 VB |
3656 | zone->compact_blockskip_flush = false; |
3657 | compaction_defer_reset(zone, order, true); | |
3658 | count_vm_event(COMPACTSUCCESS); | |
3659 | return page; | |
3660 | } | |
56de7263 | 3661 | |
98dd3b48 VB |
3662 | /* |
3663 | * It's bad if compaction run occurs and fails. The most likely reason | |
3664 | * is that pages exist, but not enough to satisfy watermarks. | |
3665 | */ | |
3666 | count_vm_event(COMPACTFAIL); | |
66199712 | 3667 | |
98dd3b48 | 3668 | cond_resched(); |
56de7263 MG |
3669 | |
3670 | return NULL; | |
3671 | } | |
33c2d214 | 3672 | |
3250845d VB |
3673 | static inline bool |
3674 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3675 | enum compact_result compact_result, | |
3676 | enum compact_priority *compact_priority, | |
d9436498 | 3677 | int *compaction_retries) |
3250845d VB |
3678 | { |
3679 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3680 | int min_priority; |
65190cff MH |
3681 | bool ret = false; |
3682 | int retries = *compaction_retries; | |
3683 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3684 | |
3685 | if (!order) | |
3686 | return false; | |
3687 | ||
d9436498 VB |
3688 | if (compaction_made_progress(compact_result)) |
3689 | (*compaction_retries)++; | |
3690 | ||
3250845d VB |
3691 | /* |
3692 | * compaction considers all the zone as desperately out of memory | |
3693 | * so it doesn't really make much sense to retry except when the | |
3694 | * failure could be caused by insufficient priority | |
3695 | */ | |
d9436498 VB |
3696 | if (compaction_failed(compact_result)) |
3697 | goto check_priority; | |
3250845d VB |
3698 | |
3699 | /* | |
3700 | * make sure the compaction wasn't deferred or didn't bail out early | |
3701 | * due to locks contention before we declare that we should give up. | |
3702 | * But do not retry if the given zonelist is not suitable for | |
3703 | * compaction. | |
3704 | */ | |
65190cff MH |
3705 | if (compaction_withdrawn(compact_result)) { |
3706 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
3707 | goto out; | |
3708 | } | |
3250845d VB |
3709 | |
3710 | /* | |
dcda9b04 | 3711 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
3712 | * costly ones because they are de facto nofail and invoke OOM |
3713 | * killer to move on while costly can fail and users are ready | |
3714 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3715 | * would need much more detailed feedback from compaction to | |
3716 | * make a better decision. | |
3717 | */ | |
3718 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3719 | max_retries /= 4; | |
65190cff MH |
3720 | if (*compaction_retries <= max_retries) { |
3721 | ret = true; | |
3722 | goto out; | |
3723 | } | |
3250845d | 3724 | |
d9436498 VB |
3725 | /* |
3726 | * Make sure there are attempts at the highest priority if we exhausted | |
3727 | * all retries or failed at the lower priorities. | |
3728 | */ | |
3729 | check_priority: | |
c2033b00 VB |
3730 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3731 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3732 | |
c2033b00 | 3733 | if (*compact_priority > min_priority) { |
d9436498 VB |
3734 | (*compact_priority)--; |
3735 | *compaction_retries = 0; | |
65190cff | 3736 | ret = true; |
d9436498 | 3737 | } |
65190cff MH |
3738 | out: |
3739 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
3740 | return ret; | |
3250845d | 3741 | } |
56de7263 MG |
3742 | #else |
3743 | static inline struct page * | |
3744 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3745 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3746 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3747 | { |
33c2d214 | 3748 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3749 | return NULL; |
3750 | } | |
33c2d214 MH |
3751 | |
3752 | static inline bool | |
86a294a8 MH |
3753 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3754 | enum compact_result compact_result, | |
a5508cd8 | 3755 | enum compact_priority *compact_priority, |
d9436498 | 3756 | int *compaction_retries) |
33c2d214 | 3757 | { |
31e49bfd MH |
3758 | struct zone *zone; |
3759 | struct zoneref *z; | |
3760 | ||
3761 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3762 | return false; | |
3763 | ||
3764 | /* | |
3765 | * There are setups with compaction disabled which would prefer to loop | |
3766 | * inside the allocator rather than hit the oom killer prematurely. | |
3767 | * Let's give them a good hope and keep retrying while the order-0 | |
3768 | * watermarks are OK. | |
3769 | */ | |
3770 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
3771 | ac->nodemask) { | |
3772 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
3773 | ac_classzone_idx(ac), alloc_flags)) | |
3774 | return true; | |
3775 | } | |
33c2d214 MH |
3776 | return false; |
3777 | } | |
3250845d | 3778 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3779 | |
d92a8cfc | 3780 | #ifdef CONFIG_LOCKDEP |
93781325 | 3781 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
3782 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
3783 | ||
3784 | static bool __need_fs_reclaim(gfp_t gfp_mask) | |
3785 | { | |
3786 | gfp_mask = current_gfp_context(gfp_mask); | |
3787 | ||
3788 | /* no reclaim without waiting on it */ | |
3789 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
3790 | return false; | |
3791 | ||
3792 | /* this guy won't enter reclaim */ | |
2e517d68 | 3793 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
3794 | return false; |
3795 | ||
3796 | /* We're only interested __GFP_FS allocations for now */ | |
3797 | if (!(gfp_mask & __GFP_FS)) | |
3798 | return false; | |
3799 | ||
3800 | if (gfp_mask & __GFP_NOLOCKDEP) | |
3801 | return false; | |
3802 | ||
3803 | return true; | |
3804 | } | |
3805 | ||
93781325 OS |
3806 | void __fs_reclaim_acquire(void) |
3807 | { | |
3808 | lock_map_acquire(&__fs_reclaim_map); | |
3809 | } | |
3810 | ||
3811 | void __fs_reclaim_release(void) | |
3812 | { | |
3813 | lock_map_release(&__fs_reclaim_map); | |
3814 | } | |
3815 | ||
d92a8cfc PZ |
3816 | void fs_reclaim_acquire(gfp_t gfp_mask) |
3817 | { | |
3818 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 3819 | __fs_reclaim_acquire(); |
d92a8cfc PZ |
3820 | } |
3821 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
3822 | ||
3823 | void fs_reclaim_release(gfp_t gfp_mask) | |
3824 | { | |
3825 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 3826 | __fs_reclaim_release(); |
d92a8cfc PZ |
3827 | } |
3828 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
3829 | #endif | |
3830 | ||
bba90710 MS |
3831 | /* Perform direct synchronous page reclaim */ |
3832 | static int | |
a9263751 VB |
3833 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3834 | const struct alloc_context *ac) | |
11e33f6a | 3835 | { |
11e33f6a | 3836 | struct reclaim_state reclaim_state; |
bba90710 | 3837 | int progress; |
499118e9 | 3838 | unsigned int noreclaim_flag; |
eb414681 | 3839 | unsigned long pflags; |
11e33f6a MG |
3840 | |
3841 | cond_resched(); | |
3842 | ||
3843 | /* We now go into synchronous reclaim */ | |
3844 | cpuset_memory_pressure_bump(); | |
eb414681 | 3845 | psi_memstall_enter(&pflags); |
d92a8cfc | 3846 | fs_reclaim_acquire(gfp_mask); |
93781325 | 3847 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 3848 | reclaim_state.reclaimed_slab = 0; |
c06b1fca | 3849 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3850 | |
a9263751 VB |
3851 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3852 | ac->nodemask); | |
11e33f6a | 3853 | |
c06b1fca | 3854 | current->reclaim_state = NULL; |
499118e9 | 3855 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 3856 | fs_reclaim_release(gfp_mask); |
eb414681 | 3857 | psi_memstall_leave(&pflags); |
11e33f6a MG |
3858 | |
3859 | cond_resched(); | |
3860 | ||
bba90710 MS |
3861 | return progress; |
3862 | } | |
3863 | ||
3864 | /* The really slow allocator path where we enter direct reclaim */ | |
3865 | static inline struct page * | |
3866 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3867 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3868 | unsigned long *did_some_progress) |
bba90710 MS |
3869 | { |
3870 | struct page *page = NULL; | |
3871 | bool drained = false; | |
3872 | ||
a9263751 | 3873 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3874 | if (unlikely(!(*did_some_progress))) |
3875 | return NULL; | |
11e33f6a | 3876 | |
9ee493ce | 3877 | retry: |
31a6c190 | 3878 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3879 | |
3880 | /* | |
3881 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3882 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3883 | * Shrink them them and try again | |
9ee493ce MG |
3884 | */ |
3885 | if (!page && !drained) { | |
29fac03b | 3886 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3887 | drain_all_pages(NULL); |
9ee493ce MG |
3888 | drained = true; |
3889 | goto retry; | |
3890 | } | |
3891 | ||
11e33f6a MG |
3892 | return page; |
3893 | } | |
3894 | ||
5ecd9d40 DR |
3895 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
3896 | const struct alloc_context *ac) | |
3a025760 JW |
3897 | { |
3898 | struct zoneref *z; | |
3899 | struct zone *zone; | |
e1a55637 | 3900 | pg_data_t *last_pgdat = NULL; |
5ecd9d40 | 3901 | enum zone_type high_zoneidx = ac->high_zoneidx; |
3a025760 | 3902 | |
5ecd9d40 DR |
3903 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx, |
3904 | ac->nodemask) { | |
e1a55637 | 3905 | if (last_pgdat != zone->zone_pgdat) |
5ecd9d40 | 3906 | wakeup_kswapd(zone, gfp_mask, order, high_zoneidx); |
e1a55637 MG |
3907 | last_pgdat = zone->zone_pgdat; |
3908 | } | |
3a025760 JW |
3909 | } |
3910 | ||
c603844b | 3911 | static inline unsigned int |
341ce06f PZ |
3912 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3913 | { | |
c603844b | 3914 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3915 | |
a56f57ff | 3916 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3917 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3918 | |
341ce06f PZ |
3919 | /* |
3920 | * The caller may dip into page reserves a bit more if the caller | |
3921 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3922 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3923 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3924 | */ |
e6223a3b | 3925 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3926 | |
d0164adc | 3927 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3928 | /* |
b104a35d DR |
3929 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3930 | * if it can't schedule. | |
5c3240d9 | 3931 | */ |
b104a35d | 3932 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3933 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3934 | /* |
b104a35d | 3935 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3936 | * comment for __cpuset_node_allowed(). |
523b9458 | 3937 | */ |
341ce06f | 3938 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3939 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3940 | alloc_flags |= ALLOC_HARDER; |
3941 | ||
d883c6cf JK |
3942 | #ifdef CONFIG_CMA |
3943 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
3944 | alloc_flags |= ALLOC_CMA; | |
3945 | #endif | |
341ce06f PZ |
3946 | return alloc_flags; |
3947 | } | |
3948 | ||
cd04ae1e | 3949 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 3950 | { |
cd04ae1e MH |
3951 | if (!tsk_is_oom_victim(tsk)) |
3952 | return false; | |
3953 | ||
3954 | /* | |
3955 | * !MMU doesn't have oom reaper so give access to memory reserves | |
3956 | * only to the thread with TIF_MEMDIE set | |
3957 | */ | |
3958 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
3959 | return false; |
3960 | ||
cd04ae1e MH |
3961 | return true; |
3962 | } | |
3963 | ||
3964 | /* | |
3965 | * Distinguish requests which really need access to full memory | |
3966 | * reserves from oom victims which can live with a portion of it | |
3967 | */ | |
3968 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
3969 | { | |
3970 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
3971 | return 0; | |
31a6c190 | 3972 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 3973 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 3974 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
3975 | return ALLOC_NO_WATERMARKS; |
3976 | if (!in_interrupt()) { | |
3977 | if (current->flags & PF_MEMALLOC) | |
3978 | return ALLOC_NO_WATERMARKS; | |
3979 | else if (oom_reserves_allowed(current)) | |
3980 | return ALLOC_OOM; | |
3981 | } | |
31a6c190 | 3982 | |
cd04ae1e MH |
3983 | return 0; |
3984 | } | |
3985 | ||
3986 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
3987 | { | |
3988 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
3989 | } |
3990 | ||
0a0337e0 MH |
3991 | /* |
3992 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
3993 | * for the given allocation request. | |
491d79ae JW |
3994 | * |
3995 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
3996 | * without success, or when we couldn't even meet the watermark if we | |
3997 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
3998 | * |
3999 | * Returns true if a retry is viable or false to enter the oom path. | |
4000 | */ | |
4001 | static inline bool | |
4002 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4003 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4004 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4005 | { |
4006 | struct zone *zone; | |
4007 | struct zoneref *z; | |
15f570bf | 4008 | bool ret = false; |
0a0337e0 | 4009 | |
423b452e VB |
4010 | /* |
4011 | * Costly allocations might have made a progress but this doesn't mean | |
4012 | * their order will become available due to high fragmentation so | |
4013 | * always increment the no progress counter for them | |
4014 | */ | |
4015 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4016 | *no_progress_loops = 0; | |
4017 | else | |
4018 | (*no_progress_loops)++; | |
4019 | ||
0a0337e0 MH |
4020 | /* |
4021 | * Make sure we converge to OOM if we cannot make any progress | |
4022 | * several times in the row. | |
4023 | */ | |
04c8716f MK |
4024 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4025 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4026 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4027 | } |
0a0337e0 | 4028 | |
bca67592 MG |
4029 | /* |
4030 | * Keep reclaiming pages while there is a chance this will lead | |
4031 | * somewhere. If none of the target zones can satisfy our allocation | |
4032 | * request even if all reclaimable pages are considered then we are | |
4033 | * screwed and have to go OOM. | |
0a0337e0 MH |
4034 | */ |
4035 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
4036 | ac->nodemask) { | |
4037 | unsigned long available; | |
ede37713 | 4038 | unsigned long reclaimable; |
d379f01d MH |
4039 | unsigned long min_wmark = min_wmark_pages(zone); |
4040 | bool wmark; | |
0a0337e0 | 4041 | |
5a1c84b4 | 4042 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4043 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4044 | |
4045 | /* | |
491d79ae JW |
4046 | * Would the allocation succeed if we reclaimed all |
4047 | * reclaimable pages? | |
0a0337e0 | 4048 | */ |
d379f01d MH |
4049 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
4050 | ac_classzone_idx(ac), alloc_flags, available); | |
4051 | trace_reclaim_retry_zone(z, order, reclaimable, | |
4052 | available, min_wmark, *no_progress_loops, wmark); | |
4053 | if (wmark) { | |
ede37713 MH |
4054 | /* |
4055 | * If we didn't make any progress and have a lot of | |
4056 | * dirty + writeback pages then we should wait for | |
4057 | * an IO to complete to slow down the reclaim and | |
4058 | * prevent from pre mature OOM | |
4059 | */ | |
4060 | if (!did_some_progress) { | |
11fb9989 | 4061 | unsigned long write_pending; |
ede37713 | 4062 | |
5a1c84b4 MG |
4063 | write_pending = zone_page_state_snapshot(zone, |
4064 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 4065 | |
11fb9989 | 4066 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
4067 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
4068 | return true; | |
4069 | } | |
4070 | } | |
5a1c84b4 | 4071 | |
15f570bf MH |
4072 | ret = true; |
4073 | goto out; | |
0a0337e0 MH |
4074 | } |
4075 | } | |
4076 | ||
15f570bf MH |
4077 | out: |
4078 | /* | |
4079 | * Memory allocation/reclaim might be called from a WQ context and the | |
4080 | * current implementation of the WQ concurrency control doesn't | |
4081 | * recognize that a particular WQ is congested if the worker thread is | |
4082 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4083 | * here rather than calling cond_resched(). | |
4084 | */ | |
4085 | if (current->flags & PF_WQ_WORKER) | |
4086 | schedule_timeout_uninterruptible(1); | |
4087 | else | |
4088 | cond_resched(); | |
4089 | return ret; | |
0a0337e0 MH |
4090 | } |
4091 | ||
902b6281 VB |
4092 | static inline bool |
4093 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4094 | { | |
4095 | /* | |
4096 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4097 | * mempolicy don't intersect. This should be normally dealt with by | |
4098 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4099 | * such a way the check therein was true, and then it became false | |
4100 | * before we got our cpuset_mems_cookie here. | |
4101 | * This assumes that for all allocations, ac->nodemask can come only | |
4102 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4103 | * when it does not intersect with the cpuset restrictions) or the | |
4104 | * caller can deal with a violated nodemask. | |
4105 | */ | |
4106 | if (cpusets_enabled() && ac->nodemask && | |
4107 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4108 | ac->nodemask = NULL; | |
4109 | return true; | |
4110 | } | |
4111 | ||
4112 | /* | |
4113 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4114 | * possible to race with parallel threads in such a way that our | |
4115 | * allocation can fail while the mask is being updated. If we are about | |
4116 | * to fail, check if the cpuset changed during allocation and if so, | |
4117 | * retry. | |
4118 | */ | |
4119 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4120 | return true; | |
4121 | ||
4122 | return false; | |
4123 | } | |
4124 | ||
11e33f6a MG |
4125 | static inline struct page * |
4126 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4127 | struct alloc_context *ac) |
11e33f6a | 4128 | { |
d0164adc | 4129 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4130 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4131 | struct page *page = NULL; |
c603844b | 4132 | unsigned int alloc_flags; |
11e33f6a | 4133 | unsigned long did_some_progress; |
5ce9bfef | 4134 | enum compact_priority compact_priority; |
c5d01d0d | 4135 | enum compact_result compact_result; |
5ce9bfef VB |
4136 | int compaction_retries; |
4137 | int no_progress_loops; | |
5ce9bfef | 4138 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4139 | int reserve_flags; |
1da177e4 | 4140 | |
d0164adc MG |
4141 | /* |
4142 | * We also sanity check to catch abuse of atomic reserves being used by | |
4143 | * callers that are not in atomic context. | |
4144 | */ | |
4145 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4146 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4147 | gfp_mask &= ~__GFP_ATOMIC; | |
4148 | ||
5ce9bfef VB |
4149 | retry_cpuset: |
4150 | compaction_retries = 0; | |
4151 | no_progress_loops = 0; | |
4152 | compact_priority = DEF_COMPACT_PRIORITY; | |
4153 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4154 | |
4155 | /* | |
4156 | * The fast path uses conservative alloc_flags to succeed only until | |
4157 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4158 | * alloc_flags precisely. So we do that now. | |
4159 | */ | |
4160 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4161 | ||
e47483bc VB |
4162 | /* |
4163 | * We need to recalculate the starting point for the zonelist iterator | |
4164 | * because we might have used different nodemask in the fast path, or | |
4165 | * there was a cpuset modification and we are retrying - otherwise we | |
4166 | * could end up iterating over non-eligible zones endlessly. | |
4167 | */ | |
4168 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
4169 | ac->high_zoneidx, ac->nodemask); | |
4170 | if (!ac->preferred_zoneref->zone) | |
4171 | goto nopage; | |
4172 | ||
23771235 | 4173 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
5ecd9d40 | 4174 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4175 | |
4176 | /* | |
4177 | * The adjusted alloc_flags might result in immediate success, so try | |
4178 | * that first | |
4179 | */ | |
4180 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4181 | if (page) | |
4182 | goto got_pg; | |
4183 | ||
a8161d1e VB |
4184 | /* |
4185 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4186 | * that we have enough base pages and don't need to reclaim. For non- |
4187 | * movable high-order allocations, do that as well, as compaction will | |
4188 | * try prevent permanent fragmentation by migrating from blocks of the | |
4189 | * same migratetype. | |
4190 | * Don't try this for allocations that are allowed to ignore | |
4191 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4192 | */ |
282722b0 VB |
4193 | if (can_direct_reclaim && |
4194 | (costly_order || | |
4195 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4196 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4197 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4198 | alloc_flags, ac, | |
a5508cd8 | 4199 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4200 | &compact_result); |
4201 | if (page) | |
4202 | goto got_pg; | |
4203 | ||
3eb2771b VB |
4204 | /* |
4205 | * Checks for costly allocations with __GFP_NORETRY, which | |
4206 | * includes THP page fault allocations | |
4207 | */ | |
282722b0 | 4208 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { |
a8161d1e VB |
4209 | /* |
4210 | * If compaction is deferred for high-order allocations, | |
4211 | * it is because sync compaction recently failed. If | |
4212 | * this is the case and the caller requested a THP | |
4213 | * allocation, we do not want to heavily disrupt the | |
4214 | * system, so we fail the allocation instead of entering | |
4215 | * direct reclaim. | |
4216 | */ | |
4217 | if (compact_result == COMPACT_DEFERRED) | |
4218 | goto nopage; | |
4219 | ||
a8161d1e | 4220 | /* |
3eb2771b VB |
4221 | * Looks like reclaim/compaction is worth trying, but |
4222 | * sync compaction could be very expensive, so keep | |
25160354 | 4223 | * using async compaction. |
a8161d1e | 4224 | */ |
a5508cd8 | 4225 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4226 | } |
4227 | } | |
23771235 | 4228 | |
31a6c190 | 4229 | retry: |
23771235 | 4230 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
31a6c190 | 4231 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
5ecd9d40 | 4232 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4233 | |
cd04ae1e MH |
4234 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4235 | if (reserve_flags) | |
4236 | alloc_flags = reserve_flags; | |
23771235 | 4237 | |
e46e7b77 | 4238 | /* |
d6a24df0 VB |
4239 | * Reset the nodemask and zonelist iterators if memory policies can be |
4240 | * ignored. These allocations are high priority and system rather than | |
4241 | * user oriented. | |
e46e7b77 | 4242 | */ |
cd04ae1e | 4243 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4244 | ac->nodemask = NULL; |
e46e7b77 MG |
4245 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4246 | ac->high_zoneidx, ac->nodemask); | |
4247 | } | |
4248 | ||
23771235 | 4249 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4250 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4251 | if (page) |
4252 | goto got_pg; | |
1da177e4 | 4253 | |
d0164adc | 4254 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4255 | if (!can_direct_reclaim) |
1da177e4 LT |
4256 | goto nopage; |
4257 | ||
9a67f648 MH |
4258 | /* Avoid recursion of direct reclaim */ |
4259 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4260 | goto nopage; |
4261 | ||
a8161d1e VB |
4262 | /* Try direct reclaim and then allocating */ |
4263 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4264 | &did_some_progress); | |
4265 | if (page) | |
4266 | goto got_pg; | |
4267 | ||
4268 | /* Try direct compaction and then allocating */ | |
a9263751 | 4269 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4270 | compact_priority, &compact_result); |
56de7263 MG |
4271 | if (page) |
4272 | goto got_pg; | |
75f30861 | 4273 | |
9083905a JW |
4274 | /* Do not loop if specifically requested */ |
4275 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4276 | goto nopage; |
9083905a | 4277 | |
0a0337e0 MH |
4278 | /* |
4279 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4280 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4281 | */ |
dcda9b04 | 4282 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4283 | goto nopage; |
0a0337e0 | 4284 | |
0a0337e0 | 4285 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4286 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4287 | goto retry; |
4288 | ||
33c2d214 MH |
4289 | /* |
4290 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4291 | * reclaim is not able to make any progress because the current | |
4292 | * implementation of the compaction depends on the sufficient amount | |
4293 | * of free memory (see __compaction_suitable) | |
4294 | */ | |
4295 | if (did_some_progress > 0 && | |
86a294a8 | 4296 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4297 | compact_result, &compact_priority, |
d9436498 | 4298 | &compaction_retries)) |
33c2d214 MH |
4299 | goto retry; |
4300 | ||
902b6281 VB |
4301 | |
4302 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4303 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4304 | goto retry_cpuset; |
4305 | ||
9083905a JW |
4306 | /* Reclaim has failed us, start killing things */ |
4307 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4308 | if (page) | |
4309 | goto got_pg; | |
4310 | ||
9a67f648 | 4311 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e MH |
4312 | if (tsk_is_oom_victim(current) && |
4313 | (alloc_flags == ALLOC_OOM || | |
c288983d | 4314 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4315 | goto nopage; |
4316 | ||
9083905a | 4317 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4318 | if (did_some_progress) { |
4319 | no_progress_loops = 0; | |
9083905a | 4320 | goto retry; |
0a0337e0 | 4321 | } |
9083905a | 4322 | |
1da177e4 | 4323 | nopage: |
902b6281 VB |
4324 | /* Deal with possible cpuset update races before we fail */ |
4325 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4326 | goto retry_cpuset; |
4327 | ||
9a67f648 MH |
4328 | /* |
4329 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4330 | * we always retry | |
4331 | */ | |
4332 | if (gfp_mask & __GFP_NOFAIL) { | |
4333 | /* | |
4334 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4335 | * of any new users that actually require GFP_NOWAIT | |
4336 | */ | |
4337 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4338 | goto fail; | |
4339 | ||
4340 | /* | |
4341 | * PF_MEMALLOC request from this context is rather bizarre | |
4342 | * because we cannot reclaim anything and only can loop waiting | |
4343 | * for somebody to do a work for us | |
4344 | */ | |
4345 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4346 | ||
4347 | /* | |
4348 | * non failing costly orders are a hard requirement which we | |
4349 | * are not prepared for much so let's warn about these users | |
4350 | * so that we can identify them and convert them to something | |
4351 | * else. | |
4352 | */ | |
4353 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4354 | ||
6c18ba7a MH |
4355 | /* |
4356 | * Help non-failing allocations by giving them access to memory | |
4357 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4358 | * could deplete whole memory reserves which would just make | |
4359 | * the situation worse | |
4360 | */ | |
4361 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4362 | if (page) | |
4363 | goto got_pg; | |
4364 | ||
9a67f648 MH |
4365 | cond_resched(); |
4366 | goto retry; | |
4367 | } | |
4368 | fail: | |
a8e99259 | 4369 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4370 | "page allocation failure: order:%u", order); |
1da177e4 | 4371 | got_pg: |
072bb0aa | 4372 | return page; |
1da177e4 | 4373 | } |
11e33f6a | 4374 | |
9cd75558 | 4375 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4376 | int preferred_nid, nodemask_t *nodemask, |
9cd75558 MG |
4377 | struct alloc_context *ac, gfp_t *alloc_mask, |
4378 | unsigned int *alloc_flags) | |
11e33f6a | 4379 | { |
9cd75558 | 4380 | ac->high_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4381 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 MG |
4382 | ac->nodemask = nodemask; |
4383 | ac->migratetype = gfpflags_to_migratetype(gfp_mask); | |
11e33f6a | 4384 | |
682a3385 | 4385 | if (cpusets_enabled()) { |
9cd75558 | 4386 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
4387 | if (!ac->nodemask) |
4388 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
4389 | else |
4390 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4391 | } |
4392 | ||
d92a8cfc PZ |
4393 | fs_reclaim_acquire(gfp_mask); |
4394 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4395 | |
d0164adc | 4396 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4397 | |
4398 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4399 | return false; |
11e33f6a | 4400 | |
d883c6cf JK |
4401 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
4402 | *alloc_flags |= ALLOC_CMA; | |
4403 | ||
9cd75558 MG |
4404 | return true; |
4405 | } | |
21bb9bd1 | 4406 | |
9cd75558 | 4407 | /* Determine whether to spread dirty pages and what the first usable zone */ |
a380b40a | 4408 | static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac) |
9cd75558 | 4409 | { |
c9ab0c4f | 4410 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4411 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4412 | |
e46e7b77 MG |
4413 | /* |
4414 | * The preferred zone is used for statistics but crucially it is | |
4415 | * also used as the starting point for the zonelist iterator. It | |
4416 | * may get reset for allocations that ignore memory policies. | |
4417 | */ | |
9cd75558 MG |
4418 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4419 | ac->high_zoneidx, ac->nodemask); | |
4420 | } | |
4421 | ||
4422 | /* | |
4423 | * This is the 'heart' of the zoned buddy allocator. | |
4424 | */ | |
4425 | struct page * | |
04ec6264 VB |
4426 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
4427 | nodemask_t *nodemask) | |
9cd75558 MG |
4428 | { |
4429 | struct page *page; | |
4430 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
f19360f0 | 4431 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4432 | struct alloc_context ac = { }; |
4433 | ||
c63ae43b MH |
4434 | /* |
4435 | * There are several places where we assume that the order value is sane | |
4436 | * so bail out early if the request is out of bound. | |
4437 | */ | |
4438 | if (unlikely(order >= MAX_ORDER)) { | |
4439 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
4440 | return NULL; | |
4441 | } | |
4442 | ||
9cd75558 | 4443 | gfp_mask &= gfp_allowed_mask; |
f19360f0 | 4444 | alloc_mask = gfp_mask; |
04ec6264 | 4445 | if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags)) |
9cd75558 MG |
4446 | return NULL; |
4447 | ||
a380b40a | 4448 | finalise_ac(gfp_mask, &ac); |
5bb1b169 | 4449 | |
6bb15450 MG |
4450 | /* |
4451 | * Forbid the first pass from falling back to types that fragment | |
4452 | * memory until all local zones are considered. | |
4453 | */ | |
4454 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone); | |
4455 | ||
5117f45d | 4456 | /* First allocation attempt */ |
a9263751 | 4457 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4458 | if (likely(page)) |
4459 | goto out; | |
11e33f6a | 4460 | |
4fcb0971 | 4461 | /* |
7dea19f9 MH |
4462 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4463 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4464 | * from a particular context which has been marked by | |
4465 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4466 | */ |
7dea19f9 | 4467 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4468 | ac.spread_dirty_pages = false; |
23f086f9 | 4469 | |
4741526b MG |
4470 | /* |
4471 | * Restore the original nodemask if it was potentially replaced with | |
4472 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4473 | */ | |
e47483bc | 4474 | if (unlikely(ac.nodemask != nodemask)) |
4741526b | 4475 | ac.nodemask = nodemask; |
16096c25 | 4476 | |
4fcb0971 | 4477 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4478 | |
4fcb0971 | 4479 | out: |
c4159a75 VD |
4480 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
4481 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { | |
4482 | __free_pages(page, order); | |
4483 | page = NULL; | |
4949148a VD |
4484 | } |
4485 | ||
4fcb0971 MG |
4486 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
4487 | ||
11e33f6a | 4488 | return page; |
1da177e4 | 4489 | } |
d239171e | 4490 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4491 | |
4492 | /* | |
9ea9a680 MH |
4493 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4494 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4495 | * you need to access high mem. | |
1da177e4 | 4496 | */ |
920c7a5d | 4497 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4498 | { |
945a1113 AM |
4499 | struct page *page; |
4500 | ||
9ea9a680 | 4501 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4502 | if (!page) |
4503 | return 0; | |
4504 | return (unsigned long) page_address(page); | |
4505 | } | |
1da177e4 LT |
4506 | EXPORT_SYMBOL(__get_free_pages); |
4507 | ||
920c7a5d | 4508 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4509 | { |
945a1113 | 4510 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4511 | } |
1da177e4 LT |
4512 | EXPORT_SYMBOL(get_zeroed_page); |
4513 | ||
742aa7fb | 4514 | static inline void free_the_page(struct page *page, unsigned int order) |
1da177e4 | 4515 | { |
742aa7fb AL |
4516 | if (order == 0) /* Via pcp? */ |
4517 | free_unref_page(page); | |
4518 | else | |
4519 | __free_pages_ok(page, order); | |
1da177e4 LT |
4520 | } |
4521 | ||
742aa7fb AL |
4522 | void __free_pages(struct page *page, unsigned int order) |
4523 | { | |
4524 | if (put_page_testzero(page)) | |
4525 | free_the_page(page, order); | |
4526 | } | |
1da177e4 LT |
4527 | EXPORT_SYMBOL(__free_pages); |
4528 | ||
920c7a5d | 4529 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4530 | { |
4531 | if (addr != 0) { | |
725d704e | 4532 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4533 | __free_pages(virt_to_page((void *)addr), order); |
4534 | } | |
4535 | } | |
4536 | ||
4537 | EXPORT_SYMBOL(free_pages); | |
4538 | ||
b63ae8ca AD |
4539 | /* |
4540 | * Page Fragment: | |
4541 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4542 | * within a 0 or higher order page. Multiple fragments within that page | |
4543 | * are individually refcounted, in the page's reference counter. | |
4544 | * | |
4545 | * The page_frag functions below provide a simple allocation framework for | |
4546 | * page fragments. This is used by the network stack and network device | |
4547 | * drivers to provide a backing region of memory for use as either an | |
4548 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4549 | */ | |
2976db80 AD |
4550 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4551 | gfp_t gfp_mask) | |
b63ae8ca AD |
4552 | { |
4553 | struct page *page = NULL; | |
4554 | gfp_t gfp = gfp_mask; | |
4555 | ||
4556 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4557 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4558 | __GFP_NOMEMALLOC; | |
4559 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4560 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4561 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4562 | #endif | |
4563 | if (unlikely(!page)) | |
4564 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4565 | ||
4566 | nc->va = page ? page_address(page) : NULL; | |
4567 | ||
4568 | return page; | |
4569 | } | |
4570 | ||
2976db80 | 4571 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4572 | { |
4573 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4574 | ||
742aa7fb AL |
4575 | if (page_ref_sub_and_test(page, count)) |
4576 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 4577 | } |
2976db80 | 4578 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4579 | |
8c2dd3e4 AD |
4580 | void *page_frag_alloc(struct page_frag_cache *nc, |
4581 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4582 | { |
4583 | unsigned int size = PAGE_SIZE; | |
4584 | struct page *page; | |
4585 | int offset; | |
4586 | ||
4587 | if (unlikely(!nc->va)) { | |
4588 | refill: | |
2976db80 | 4589 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4590 | if (!page) |
4591 | return NULL; | |
4592 | ||
4593 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4594 | /* if size can vary use size else just use PAGE_SIZE */ | |
4595 | size = nc->size; | |
4596 | #endif | |
4597 | /* Even if we own the page, we do not use atomic_set(). | |
4598 | * This would break get_page_unless_zero() users. | |
4599 | */ | |
fe896d18 | 4600 | page_ref_add(page, size - 1); |
b63ae8ca AD |
4601 | |
4602 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4603 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
4604 | nc->pagecnt_bias = size; |
4605 | nc->offset = size; | |
4606 | } | |
4607 | ||
4608 | offset = nc->offset - fragsz; | |
4609 | if (unlikely(offset < 0)) { | |
4610 | page = virt_to_page(nc->va); | |
4611 | ||
fe896d18 | 4612 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4613 | goto refill; |
4614 | ||
4615 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4616 | /* if size can vary use size else just use PAGE_SIZE */ | |
4617 | size = nc->size; | |
4618 | #endif | |
4619 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 4620 | set_page_count(page, size); |
b63ae8ca AD |
4621 | |
4622 | /* reset page count bias and offset to start of new frag */ | |
4623 | nc->pagecnt_bias = size; | |
4624 | offset = size - fragsz; | |
4625 | } | |
4626 | ||
4627 | nc->pagecnt_bias--; | |
4628 | nc->offset = offset; | |
4629 | ||
4630 | return nc->va + offset; | |
4631 | } | |
8c2dd3e4 | 4632 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
4633 | |
4634 | /* | |
4635 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4636 | */ | |
8c2dd3e4 | 4637 | void page_frag_free(void *addr) |
b63ae8ca AD |
4638 | { |
4639 | struct page *page = virt_to_head_page(addr); | |
4640 | ||
742aa7fb AL |
4641 | if (unlikely(put_page_testzero(page))) |
4642 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 4643 | } |
8c2dd3e4 | 4644 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4645 | |
d00181b9 KS |
4646 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4647 | size_t size) | |
ee85c2e1 AK |
4648 | { |
4649 | if (addr) { | |
4650 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
4651 | unsigned long used = addr + PAGE_ALIGN(size); | |
4652 | ||
4653 | split_page(virt_to_page((void *)addr), order); | |
4654 | while (used < alloc_end) { | |
4655 | free_page(used); | |
4656 | used += PAGE_SIZE; | |
4657 | } | |
4658 | } | |
4659 | return (void *)addr; | |
4660 | } | |
4661 | ||
2be0ffe2 TT |
4662 | /** |
4663 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4664 | * @size: the number of bytes to allocate | |
4665 | * @gfp_mask: GFP flags for the allocation | |
4666 | * | |
4667 | * This function is similar to alloc_pages(), except that it allocates the | |
4668 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4669 | * allocate memory in power-of-two pages. | |
4670 | * | |
4671 | * This function is also limited by MAX_ORDER. | |
4672 | * | |
4673 | * Memory allocated by this function must be released by free_pages_exact(). | |
4674 | */ | |
4675 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4676 | { | |
4677 | unsigned int order = get_order(size); | |
4678 | unsigned long addr; | |
4679 | ||
4680 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 4681 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4682 | } |
4683 | EXPORT_SYMBOL(alloc_pages_exact); | |
4684 | ||
ee85c2e1 AK |
4685 | /** |
4686 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4687 | * pages on a node. | |
b5e6ab58 | 4688 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
4689 | * @size: the number of bytes to allocate |
4690 | * @gfp_mask: GFP flags for the allocation | |
4691 | * | |
4692 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4693 | * back. | |
ee85c2e1 | 4694 | */ |
e1931811 | 4695 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4696 | { |
d00181b9 | 4697 | unsigned int order = get_order(size); |
ee85c2e1 AK |
4698 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
4699 | if (!p) | |
4700 | return NULL; | |
4701 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4702 | } | |
ee85c2e1 | 4703 | |
2be0ffe2 TT |
4704 | /** |
4705 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4706 | * @virt: the value returned by alloc_pages_exact. | |
4707 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4708 | * | |
4709 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4710 | */ | |
4711 | void free_pages_exact(void *virt, size_t size) | |
4712 | { | |
4713 | unsigned long addr = (unsigned long)virt; | |
4714 | unsigned long end = addr + PAGE_ALIGN(size); | |
4715 | ||
4716 | while (addr < end) { | |
4717 | free_page(addr); | |
4718 | addr += PAGE_SIZE; | |
4719 | } | |
4720 | } | |
4721 | EXPORT_SYMBOL(free_pages_exact); | |
4722 | ||
e0fb5815 ZY |
4723 | /** |
4724 | * nr_free_zone_pages - count number of pages beyond high watermark | |
4725 | * @offset: The zone index of the highest zone | |
4726 | * | |
4727 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
4728 | * high watermark within all zones at or below a given zone index. For each | |
4729 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
4730 | * |
4731 | * nr_free_zone_pages = managed_pages - high_pages | |
e0fb5815 | 4732 | */ |
ebec3862 | 4733 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 4734 | { |
dd1a239f | 4735 | struct zoneref *z; |
54a6eb5c MG |
4736 | struct zone *zone; |
4737 | ||
e310fd43 | 4738 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 4739 | unsigned long sum = 0; |
1da177e4 | 4740 | |
0e88460d | 4741 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 4742 | |
54a6eb5c | 4743 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 4744 | unsigned long size = zone_managed_pages(zone); |
41858966 | 4745 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
4746 | if (size > high) |
4747 | sum += size - high; | |
1da177e4 LT |
4748 | } |
4749 | ||
4750 | return sum; | |
4751 | } | |
4752 | ||
e0fb5815 ZY |
4753 | /** |
4754 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
4755 | * | |
4756 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
4757 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 4758 | */ |
ebec3862 | 4759 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 4760 | { |
af4ca457 | 4761 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 4762 | } |
c2f1a551 | 4763 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 4764 | |
e0fb5815 ZY |
4765 | /** |
4766 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
4767 | * | |
4768 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
4769 | * high watermark within all zones. | |
1da177e4 | 4770 | */ |
ebec3862 | 4771 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 4772 | { |
2a1e274a | 4773 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 4774 | } |
08e0f6a9 CL |
4775 | |
4776 | static inline void show_node(struct zone *zone) | |
1da177e4 | 4777 | { |
e5adfffc | 4778 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 4779 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 4780 | } |
1da177e4 | 4781 | |
d02bd27b IR |
4782 | long si_mem_available(void) |
4783 | { | |
4784 | long available; | |
4785 | unsigned long pagecache; | |
4786 | unsigned long wmark_low = 0; | |
4787 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 4788 | unsigned long reclaimable; |
d02bd27b IR |
4789 | struct zone *zone; |
4790 | int lru; | |
4791 | ||
4792 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 4793 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
4794 | |
4795 | for_each_zone(zone) | |
4796 | wmark_low += zone->watermark[WMARK_LOW]; | |
4797 | ||
4798 | /* | |
4799 | * Estimate the amount of memory available for userspace allocations, | |
4800 | * without causing swapping. | |
4801 | */ | |
c41f012a | 4802 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
4803 | |
4804 | /* | |
4805 | * Not all the page cache can be freed, otherwise the system will | |
4806 | * start swapping. Assume at least half of the page cache, or the | |
4807 | * low watermark worth of cache, needs to stay. | |
4808 | */ | |
4809 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
4810 | pagecache -= min(pagecache / 2, wmark_low); | |
4811 | available += pagecache; | |
4812 | ||
4813 | /* | |
b29940c1 VB |
4814 | * Part of the reclaimable slab and other kernel memory consists of |
4815 | * items that are in use, and cannot be freed. Cap this estimate at the | |
4816 | * low watermark. | |
d02bd27b | 4817 | */ |
b29940c1 VB |
4818 | reclaimable = global_node_page_state(NR_SLAB_RECLAIMABLE) + |
4819 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
4820 | available += reclaimable - min(reclaimable / 2, wmark_low); | |
034ebf65 | 4821 | |
d02bd27b IR |
4822 | if (available < 0) |
4823 | available = 0; | |
4824 | return available; | |
4825 | } | |
4826 | EXPORT_SYMBOL_GPL(si_mem_available); | |
4827 | ||
1da177e4 LT |
4828 | void si_meminfo(struct sysinfo *val) |
4829 | { | |
ca79b0c2 | 4830 | val->totalram = totalram_pages(); |
11fb9989 | 4831 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 4832 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 4833 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 4834 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 4835 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
4836 | val->mem_unit = PAGE_SIZE; |
4837 | } | |
4838 | ||
4839 | EXPORT_SYMBOL(si_meminfo); | |
4840 | ||
4841 | #ifdef CONFIG_NUMA | |
4842 | void si_meminfo_node(struct sysinfo *val, int nid) | |
4843 | { | |
cdd91a77 JL |
4844 | int zone_type; /* needs to be signed */ |
4845 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
4846 | unsigned long managed_highpages = 0; |
4847 | unsigned long free_highpages = 0; | |
1da177e4 LT |
4848 | pg_data_t *pgdat = NODE_DATA(nid); |
4849 | ||
cdd91a77 | 4850 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 4851 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 4852 | val->totalram = managed_pages; |
11fb9989 | 4853 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 4854 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 4855 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
4856 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4857 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4858 | ||
4859 | if (is_highmem(zone)) { | |
9705bea5 | 4860 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
4861 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
4862 | } | |
4863 | } | |
4864 | val->totalhigh = managed_highpages; | |
4865 | val->freehigh = free_highpages; | |
98d2b0eb | 4866 | #else |
fc2bd799 JK |
4867 | val->totalhigh = managed_highpages; |
4868 | val->freehigh = free_highpages; | |
98d2b0eb | 4869 | #endif |
1da177e4 LT |
4870 | val->mem_unit = PAGE_SIZE; |
4871 | } | |
4872 | #endif | |
4873 | ||
ddd588b5 | 4874 | /* |
7bf02ea2 DR |
4875 | * Determine whether the node should be displayed or not, depending on whether |
4876 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 4877 | */ |
9af744d7 | 4878 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 4879 | { |
ddd588b5 | 4880 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 4881 | return false; |
ddd588b5 | 4882 | |
9af744d7 MH |
4883 | /* |
4884 | * no node mask - aka implicit memory numa policy. Do not bother with | |
4885 | * the synchronization - read_mems_allowed_begin - because we do not | |
4886 | * have to be precise here. | |
4887 | */ | |
4888 | if (!nodemask) | |
4889 | nodemask = &cpuset_current_mems_allowed; | |
4890 | ||
4891 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
4892 | } |
4893 | ||
1da177e4 LT |
4894 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4895 | ||
377e4f16 RV |
4896 | static void show_migration_types(unsigned char type) |
4897 | { | |
4898 | static const char types[MIGRATE_TYPES] = { | |
4899 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 4900 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
4901 | [MIGRATE_RECLAIMABLE] = 'E', |
4902 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
4903 | #ifdef CONFIG_CMA |
4904 | [MIGRATE_CMA] = 'C', | |
4905 | #endif | |
194159fb | 4906 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 4907 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 4908 | #endif |
377e4f16 RV |
4909 | }; |
4910 | char tmp[MIGRATE_TYPES + 1]; | |
4911 | char *p = tmp; | |
4912 | int i; | |
4913 | ||
4914 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
4915 | if (type & (1 << i)) | |
4916 | *p++ = types[i]; | |
4917 | } | |
4918 | ||
4919 | *p = '\0'; | |
1f84a18f | 4920 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
4921 | } |
4922 | ||
1da177e4 LT |
4923 | /* |
4924 | * Show free area list (used inside shift_scroll-lock stuff) | |
4925 | * We also calculate the percentage fragmentation. We do this by counting the | |
4926 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
4927 | * |
4928 | * Bits in @filter: | |
4929 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
4930 | * cpuset. | |
1da177e4 | 4931 | */ |
9af744d7 | 4932 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 4933 | { |
d1bfcdb8 | 4934 | unsigned long free_pcp = 0; |
c7241913 | 4935 | int cpu; |
1da177e4 | 4936 | struct zone *zone; |
599d0c95 | 4937 | pg_data_t *pgdat; |
1da177e4 | 4938 | |
ee99c71c | 4939 | for_each_populated_zone(zone) { |
9af744d7 | 4940 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 4941 | continue; |
d1bfcdb8 | 4942 | |
761b0677 KK |
4943 | for_each_online_cpu(cpu) |
4944 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
4945 | } |
4946 | ||
a731286d KM |
4947 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4948 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
4949 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4950 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 4951 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 4952 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
4953 | global_node_page_state(NR_ACTIVE_ANON), |
4954 | global_node_page_state(NR_INACTIVE_ANON), | |
4955 | global_node_page_state(NR_ISOLATED_ANON), | |
4956 | global_node_page_state(NR_ACTIVE_FILE), | |
4957 | global_node_page_state(NR_INACTIVE_FILE), | |
4958 | global_node_page_state(NR_ISOLATED_FILE), | |
4959 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
4960 | global_node_page_state(NR_FILE_DIRTY), |
4961 | global_node_page_state(NR_WRITEBACK), | |
4962 | global_node_page_state(NR_UNSTABLE_NFS), | |
d507e2eb JW |
4963 | global_node_page_state(NR_SLAB_RECLAIMABLE), |
4964 | global_node_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 4965 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 4966 | global_node_page_state(NR_SHMEM), |
c41f012a MH |
4967 | global_zone_page_state(NR_PAGETABLE), |
4968 | global_zone_page_state(NR_BOUNCE), | |
4969 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 4970 | free_pcp, |
c41f012a | 4971 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4972 | |
599d0c95 | 4973 | for_each_online_pgdat(pgdat) { |
9af744d7 | 4974 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
4975 | continue; |
4976 | ||
599d0c95 MG |
4977 | printk("Node %d" |
4978 | " active_anon:%lukB" | |
4979 | " inactive_anon:%lukB" | |
4980 | " active_file:%lukB" | |
4981 | " inactive_file:%lukB" | |
4982 | " unevictable:%lukB" | |
4983 | " isolated(anon):%lukB" | |
4984 | " isolated(file):%lukB" | |
50658e2e | 4985 | " mapped:%lukB" |
11fb9989 MG |
4986 | " dirty:%lukB" |
4987 | " writeback:%lukB" | |
4988 | " shmem:%lukB" | |
4989 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4990 | " shmem_thp: %lukB" | |
4991 | " shmem_pmdmapped: %lukB" | |
4992 | " anon_thp: %lukB" | |
4993 | #endif | |
4994 | " writeback_tmp:%lukB" | |
4995 | " unstable:%lukB" | |
599d0c95 MG |
4996 | " all_unreclaimable? %s" |
4997 | "\n", | |
4998 | pgdat->node_id, | |
4999 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
5000 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
5001 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
5002 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
5003 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
5004 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
5005 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 5006 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
5007 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
5008 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 5009 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
5010 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5011 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
5012 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
5013 | * HPAGE_PMD_NR), | |
5014 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
5015 | #endif | |
11fb9989 MG |
5016 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
5017 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), | |
c73322d0 JW |
5018 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
5019 | "yes" : "no"); | |
599d0c95 MG |
5020 | } |
5021 | ||
ee99c71c | 5022 | for_each_populated_zone(zone) { |
1da177e4 LT |
5023 | int i; |
5024 | ||
9af744d7 | 5025 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5026 | continue; |
d1bfcdb8 KK |
5027 | |
5028 | free_pcp = 0; | |
5029 | for_each_online_cpu(cpu) | |
5030 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
5031 | ||
1da177e4 | 5032 | show_node(zone); |
1f84a18f JP |
5033 | printk(KERN_CONT |
5034 | "%s" | |
1da177e4 LT |
5035 | " free:%lukB" |
5036 | " min:%lukB" | |
5037 | " low:%lukB" | |
5038 | " high:%lukB" | |
71c799f4 MK |
5039 | " active_anon:%lukB" |
5040 | " inactive_anon:%lukB" | |
5041 | " active_file:%lukB" | |
5042 | " inactive_file:%lukB" | |
5043 | " unevictable:%lukB" | |
5a1c84b4 | 5044 | " writepending:%lukB" |
1da177e4 | 5045 | " present:%lukB" |
9feedc9d | 5046 | " managed:%lukB" |
4a0aa73f | 5047 | " mlocked:%lukB" |
c6a7f572 | 5048 | " kernel_stack:%lukB" |
4a0aa73f | 5049 | " pagetables:%lukB" |
4a0aa73f | 5050 | " bounce:%lukB" |
d1bfcdb8 KK |
5051 | " free_pcp:%lukB" |
5052 | " local_pcp:%ukB" | |
d1ce749a | 5053 | " free_cma:%lukB" |
1da177e4 LT |
5054 | "\n", |
5055 | zone->name, | |
88f5acf8 | 5056 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
5057 | K(min_wmark_pages(zone)), |
5058 | K(low_wmark_pages(zone)), | |
5059 | K(high_wmark_pages(zone)), | |
71c799f4 MK |
5060 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
5061 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
5062 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
5063 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
5064 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 5065 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 5066 | K(zone->present_pages), |
9705bea5 | 5067 | K(zone_managed_pages(zone)), |
4a0aa73f | 5068 | K(zone_page_state(zone, NR_MLOCK)), |
d30dd8be | 5069 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4a0aa73f | 5070 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 5071 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
5072 | K(free_pcp), |
5073 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 5074 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
5075 | printk("lowmem_reserve[]:"); |
5076 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
5077 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5078 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5079 | } |
5080 | ||
ee99c71c | 5081 | for_each_populated_zone(zone) { |
d00181b9 KS |
5082 | unsigned int order; |
5083 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5084 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5085 | |
9af744d7 | 5086 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5087 | continue; |
1da177e4 | 5088 | show_node(zone); |
1f84a18f | 5089 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5090 | |
5091 | spin_lock_irqsave(&zone->lock, flags); | |
5092 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5093 | struct free_area *area = &zone->free_area[order]; |
5094 | int type; | |
5095 | ||
5096 | nr[order] = area->nr_free; | |
8f9de51a | 5097 | total += nr[order] << order; |
377e4f16 RV |
5098 | |
5099 | types[order] = 0; | |
5100 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
5101 | if (!list_empty(&area->free_list[type])) | |
5102 | types[order] |= 1 << type; | |
5103 | } | |
1da177e4 LT |
5104 | } |
5105 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5106 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5107 | printk(KERN_CONT "%lu*%lukB ", |
5108 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5109 | if (nr[order]) |
5110 | show_migration_types(types[order]); | |
5111 | } | |
1f84a18f | 5112 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5113 | } |
5114 | ||
949f7ec5 DR |
5115 | hugetlb_show_meminfo(); |
5116 | ||
11fb9989 | 5117 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5118 | |
1da177e4 LT |
5119 | show_swap_cache_info(); |
5120 | } | |
5121 | ||
19770b32 MG |
5122 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5123 | { | |
5124 | zoneref->zone = zone; | |
5125 | zoneref->zone_idx = zone_idx(zone); | |
5126 | } | |
5127 | ||
1da177e4 LT |
5128 | /* |
5129 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5130 | * |
5131 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5132 | */ |
9d3be21b | 5133 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5134 | { |
1a93205b | 5135 | struct zone *zone; |
bc732f1d | 5136 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5137 | int nr_zones = 0; |
02a68a5e CL |
5138 | |
5139 | do { | |
2f6726e5 | 5140 | zone_type--; |
070f8032 | 5141 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5142 | if (managed_zone(zone)) { |
9d3be21b | 5143 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5144 | check_highest_zone(zone_type); |
1da177e4 | 5145 | } |
2f6726e5 | 5146 | } while (zone_type); |
bc732f1d | 5147 | |
070f8032 | 5148 | return nr_zones; |
1da177e4 LT |
5149 | } |
5150 | ||
5151 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5152 | |
5153 | static int __parse_numa_zonelist_order(char *s) | |
5154 | { | |
c9bff3ee MH |
5155 | /* |
5156 | * We used to support different zonlists modes but they turned | |
5157 | * out to be just not useful. Let's keep the warning in place | |
5158 | * if somebody still use the cmd line parameter so that we do | |
5159 | * not fail it silently | |
5160 | */ | |
5161 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5162 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5163 | return -EINVAL; |
5164 | } | |
5165 | return 0; | |
5166 | } | |
5167 | ||
5168 | static __init int setup_numa_zonelist_order(char *s) | |
5169 | { | |
ecb256f8 VL |
5170 | if (!s) |
5171 | return 0; | |
5172 | ||
c9bff3ee | 5173 | return __parse_numa_zonelist_order(s); |
f0c0b2b8 KH |
5174 | } |
5175 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
5176 | ||
c9bff3ee MH |
5177 | char numa_zonelist_order[] = "Node"; |
5178 | ||
f0c0b2b8 KH |
5179 | /* |
5180 | * sysctl handler for numa_zonelist_order | |
5181 | */ | |
cccad5b9 | 5182 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 5183 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
5184 | loff_t *ppos) |
5185 | { | |
c9bff3ee | 5186 | char *str; |
f0c0b2b8 KH |
5187 | int ret; |
5188 | ||
c9bff3ee MH |
5189 | if (!write) |
5190 | return proc_dostring(table, write, buffer, length, ppos); | |
5191 | str = memdup_user_nul(buffer, 16); | |
5192 | if (IS_ERR(str)) | |
5193 | return PTR_ERR(str); | |
dacbde09 | 5194 | |
c9bff3ee MH |
5195 | ret = __parse_numa_zonelist_order(str); |
5196 | kfree(str); | |
443c6f14 | 5197 | return ret; |
f0c0b2b8 KH |
5198 | } |
5199 | ||
5200 | ||
62bc62a8 | 5201 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5202 | static int node_load[MAX_NUMNODES]; |
5203 | ||
1da177e4 | 5204 | /** |
4dc3b16b | 5205 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5206 | * @node: node whose fallback list we're appending |
5207 | * @used_node_mask: nodemask_t of already used nodes | |
5208 | * | |
5209 | * We use a number of factors to determine which is the next node that should | |
5210 | * appear on a given node's fallback list. The node should not have appeared | |
5211 | * already in @node's fallback list, and it should be the next closest node | |
5212 | * according to the distance array (which contains arbitrary distance values | |
5213 | * from each node to each node in the system), and should also prefer nodes | |
5214 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5215 | * on them otherwise. | |
5216 | * It returns -1 if no node is found. | |
5217 | */ | |
f0c0b2b8 | 5218 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5219 | { |
4cf808eb | 5220 | int n, val; |
1da177e4 | 5221 | int min_val = INT_MAX; |
00ef2d2f | 5222 | int best_node = NUMA_NO_NODE; |
a70f7302 | 5223 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 5224 | |
4cf808eb LT |
5225 | /* Use the local node if we haven't already */ |
5226 | if (!node_isset(node, *used_node_mask)) { | |
5227 | node_set(node, *used_node_mask); | |
5228 | return node; | |
5229 | } | |
1da177e4 | 5230 | |
4b0ef1fe | 5231 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5232 | |
5233 | /* Don't want a node to appear more than once */ | |
5234 | if (node_isset(n, *used_node_mask)) | |
5235 | continue; | |
5236 | ||
1da177e4 LT |
5237 | /* Use the distance array to find the distance */ |
5238 | val = node_distance(node, n); | |
5239 | ||
4cf808eb LT |
5240 | /* Penalize nodes under us ("prefer the next node") */ |
5241 | val += (n < node); | |
5242 | ||
1da177e4 | 5243 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
5244 | tmp = cpumask_of_node(n); |
5245 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
5246 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5247 | ||
5248 | /* Slight preference for less loaded node */ | |
5249 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
5250 | val += node_load[n]; | |
5251 | ||
5252 | if (val < min_val) { | |
5253 | min_val = val; | |
5254 | best_node = n; | |
5255 | } | |
5256 | } | |
5257 | ||
5258 | if (best_node >= 0) | |
5259 | node_set(best_node, *used_node_mask); | |
5260 | ||
5261 | return best_node; | |
5262 | } | |
5263 | ||
f0c0b2b8 KH |
5264 | |
5265 | /* | |
5266 | * Build zonelists ordered by node and zones within node. | |
5267 | * This results in maximum locality--normal zone overflows into local | |
5268 | * DMA zone, if any--but risks exhausting DMA zone. | |
5269 | */ | |
9d3be21b MH |
5270 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5271 | unsigned nr_nodes) | |
1da177e4 | 5272 | { |
9d3be21b MH |
5273 | struct zoneref *zonerefs; |
5274 | int i; | |
5275 | ||
5276 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5277 | ||
5278 | for (i = 0; i < nr_nodes; i++) { | |
5279 | int nr_zones; | |
5280 | ||
5281 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5282 | |
9d3be21b MH |
5283 | nr_zones = build_zonerefs_node(node, zonerefs); |
5284 | zonerefs += nr_zones; | |
5285 | } | |
5286 | zonerefs->zone = NULL; | |
5287 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5288 | } |
5289 | ||
523b9458 CL |
5290 | /* |
5291 | * Build gfp_thisnode zonelists | |
5292 | */ | |
5293 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5294 | { | |
9d3be21b MH |
5295 | struct zoneref *zonerefs; |
5296 | int nr_zones; | |
523b9458 | 5297 | |
9d3be21b MH |
5298 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5299 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5300 | zonerefs += nr_zones; | |
5301 | zonerefs->zone = NULL; | |
5302 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5303 | } |
5304 | ||
f0c0b2b8 KH |
5305 | /* |
5306 | * Build zonelists ordered by zone and nodes within zones. | |
5307 | * This results in conserving DMA zone[s] until all Normal memory is | |
5308 | * exhausted, but results in overflowing to remote node while memory | |
5309 | * may still exist in local DMA zone. | |
5310 | */ | |
f0c0b2b8 | 5311 | |
f0c0b2b8 KH |
5312 | static void build_zonelists(pg_data_t *pgdat) |
5313 | { | |
9d3be21b MH |
5314 | static int node_order[MAX_NUMNODES]; |
5315 | int node, load, nr_nodes = 0; | |
1da177e4 | 5316 | nodemask_t used_mask; |
f0c0b2b8 | 5317 | int local_node, prev_node; |
1da177e4 LT |
5318 | |
5319 | /* NUMA-aware ordering of nodes */ | |
5320 | local_node = pgdat->node_id; | |
62bc62a8 | 5321 | load = nr_online_nodes; |
1da177e4 LT |
5322 | prev_node = local_node; |
5323 | nodes_clear(used_mask); | |
f0c0b2b8 | 5324 | |
f0c0b2b8 | 5325 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5326 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5327 | /* | |
5328 | * We don't want to pressure a particular node. | |
5329 | * So adding penalty to the first node in same | |
5330 | * distance group to make it round-robin. | |
5331 | */ | |
957f822a DR |
5332 | if (node_distance(local_node, node) != |
5333 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5334 | node_load[node] = load; |
5335 | ||
9d3be21b | 5336 | node_order[nr_nodes++] = node; |
1da177e4 LT |
5337 | prev_node = node; |
5338 | load--; | |
1da177e4 | 5339 | } |
523b9458 | 5340 | |
9d3be21b | 5341 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5342 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
5343 | } |
5344 | ||
7aac7898 LS |
5345 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5346 | /* | |
5347 | * Return node id of node used for "local" allocations. | |
5348 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5349 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5350 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5351 | */ | |
5352 | int local_memory_node(int node) | |
5353 | { | |
c33d6c06 | 5354 | struct zoneref *z; |
7aac7898 | 5355 | |
c33d6c06 | 5356 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5357 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 5358 | NULL); |
c1093b74 | 5359 | return zone_to_nid(z->zone); |
7aac7898 LS |
5360 | } |
5361 | #endif | |
f0c0b2b8 | 5362 | |
6423aa81 JK |
5363 | static void setup_min_unmapped_ratio(void); |
5364 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5365 | #else /* CONFIG_NUMA */ |
5366 | ||
f0c0b2b8 | 5367 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5368 | { |
19655d34 | 5369 | int node, local_node; |
9d3be21b MH |
5370 | struct zoneref *zonerefs; |
5371 | int nr_zones; | |
1da177e4 LT |
5372 | |
5373 | local_node = pgdat->node_id; | |
1da177e4 | 5374 | |
9d3be21b MH |
5375 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5376 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5377 | zonerefs += nr_zones; | |
1da177e4 | 5378 | |
54a6eb5c MG |
5379 | /* |
5380 | * Now we build the zonelist so that it contains the zones | |
5381 | * of all the other nodes. | |
5382 | * We don't want to pressure a particular node, so when | |
5383 | * building the zones for node N, we make sure that the | |
5384 | * zones coming right after the local ones are those from | |
5385 | * node N+1 (modulo N) | |
5386 | */ | |
5387 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5388 | if (!node_online(node)) | |
5389 | continue; | |
9d3be21b MH |
5390 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5391 | zonerefs += nr_zones; | |
1da177e4 | 5392 | } |
54a6eb5c MG |
5393 | for (node = 0; node < local_node; node++) { |
5394 | if (!node_online(node)) | |
5395 | continue; | |
9d3be21b MH |
5396 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5397 | zonerefs += nr_zones; | |
54a6eb5c MG |
5398 | } |
5399 | ||
9d3be21b MH |
5400 | zonerefs->zone = NULL; |
5401 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5402 | } |
5403 | ||
5404 | #endif /* CONFIG_NUMA */ | |
5405 | ||
99dcc3e5 CL |
5406 | /* |
5407 | * Boot pageset table. One per cpu which is going to be used for all | |
5408 | * zones and all nodes. The parameters will be set in such a way | |
5409 | * that an item put on a list will immediately be handed over to | |
5410 | * the buddy list. This is safe since pageset manipulation is done | |
5411 | * with interrupts disabled. | |
5412 | * | |
5413 | * The boot_pagesets must be kept even after bootup is complete for | |
5414 | * unused processors and/or zones. They do play a role for bootstrapping | |
5415 | * hotplugged processors. | |
5416 | * | |
5417 | * zoneinfo_show() and maybe other functions do | |
5418 | * not check if the processor is online before following the pageset pointer. | |
5419 | * Other parts of the kernel may not check if the zone is available. | |
5420 | */ | |
5421 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5422 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
385386cf | 5423 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 5424 | |
11cd8638 | 5425 | static void __build_all_zonelists(void *data) |
1da177e4 | 5426 | { |
6811378e | 5427 | int nid; |
afb6ebb3 | 5428 | int __maybe_unused cpu; |
9adb62a5 | 5429 | pg_data_t *self = data; |
b93e0f32 MH |
5430 | static DEFINE_SPINLOCK(lock); |
5431 | ||
5432 | spin_lock(&lock); | |
9276b1bc | 5433 | |
7f9cfb31 BL |
5434 | #ifdef CONFIG_NUMA |
5435 | memset(node_load, 0, sizeof(node_load)); | |
5436 | #endif | |
9adb62a5 | 5437 | |
c1152583 WY |
5438 | /* |
5439 | * This node is hotadded and no memory is yet present. So just | |
5440 | * building zonelists is fine - no need to touch other nodes. | |
5441 | */ | |
9adb62a5 JL |
5442 | if (self && !node_online(self->node_id)) { |
5443 | build_zonelists(self); | |
c1152583 WY |
5444 | } else { |
5445 | for_each_online_node(nid) { | |
5446 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 5447 | |
c1152583 WY |
5448 | build_zonelists(pgdat); |
5449 | } | |
99dcc3e5 | 5450 | |
7aac7898 LS |
5451 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5452 | /* | |
5453 | * We now know the "local memory node" for each node-- | |
5454 | * i.e., the node of the first zone in the generic zonelist. | |
5455 | * Set up numa_mem percpu variable for on-line cpus. During | |
5456 | * boot, only the boot cpu should be on-line; we'll init the | |
5457 | * secondary cpus' numa_mem as they come on-line. During | |
5458 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5459 | */ | |
d9c9a0b9 | 5460 | for_each_online_cpu(cpu) |
7aac7898 | 5461 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5462 | #endif |
d9c9a0b9 | 5463 | } |
b93e0f32 MH |
5464 | |
5465 | spin_unlock(&lock); | |
6811378e YG |
5466 | } |
5467 | ||
061f67bc RV |
5468 | static noinline void __init |
5469 | build_all_zonelists_init(void) | |
5470 | { | |
afb6ebb3 MH |
5471 | int cpu; |
5472 | ||
061f67bc | 5473 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5474 | |
5475 | /* | |
5476 | * Initialize the boot_pagesets that are going to be used | |
5477 | * for bootstrapping processors. The real pagesets for | |
5478 | * each zone will be allocated later when the per cpu | |
5479 | * allocator is available. | |
5480 | * | |
5481 | * boot_pagesets are used also for bootstrapping offline | |
5482 | * cpus if the system is already booted because the pagesets | |
5483 | * are needed to initialize allocators on a specific cpu too. | |
5484 | * F.e. the percpu allocator needs the page allocator which | |
5485 | * needs the percpu allocator in order to allocate its pagesets | |
5486 | * (a chicken-egg dilemma). | |
5487 | */ | |
5488 | for_each_possible_cpu(cpu) | |
5489 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); | |
5490 | ||
061f67bc RV |
5491 | mminit_verify_zonelist(); |
5492 | cpuset_init_current_mems_allowed(); | |
5493 | } | |
5494 | ||
4eaf3f64 | 5495 | /* |
4eaf3f64 | 5496 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5497 | * |
72675e13 | 5498 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5499 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5500 | */ |
72675e13 | 5501 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e YG |
5502 | { |
5503 | if (system_state == SYSTEM_BOOTING) { | |
061f67bc | 5504 | build_all_zonelists_init(); |
6811378e | 5505 | } else { |
11cd8638 | 5506 | __build_all_zonelists(pgdat); |
6811378e YG |
5507 | /* cpuset refresh routine should be here */ |
5508 | } | |
bd1e22b8 | 5509 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5510 | /* |
5511 | * Disable grouping by mobility if the number of pages in the | |
5512 | * system is too low to allow the mechanism to work. It would be | |
5513 | * more accurate, but expensive to check per-zone. This check is | |
5514 | * made on memory-hotadd so a system can start with mobility | |
5515 | * disabled and enable it later | |
5516 | */ | |
d9c23400 | 5517 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5518 | page_group_by_mobility_disabled = 1; |
5519 | else | |
5520 | page_group_by_mobility_disabled = 0; | |
5521 | ||
c9bff3ee | 5522 | pr_info("Built %i zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5523 | nr_online_nodes, |
756a025f JP |
5524 | page_group_by_mobility_disabled ? "off" : "on", |
5525 | vm_total_pages); | |
f0c0b2b8 | 5526 | #ifdef CONFIG_NUMA |
f88dfff5 | 5527 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5528 | #endif |
1da177e4 LT |
5529 | } |
5530 | ||
a9a9e77f PT |
5531 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
5532 | static bool __meminit | |
5533 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
5534 | { | |
5535 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
5536 | static struct memblock_region *r; | |
5537 | ||
5538 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5539 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
5540 | for_each_memblock(memory, r) { | |
5541 | if (*pfn < memblock_region_memory_end_pfn(r)) | |
5542 | break; | |
5543 | } | |
5544 | } | |
5545 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
5546 | memblock_is_mirror(r)) { | |
5547 | *pfn = memblock_region_memory_end_pfn(r); | |
5548 | return true; | |
5549 | } | |
5550 | } | |
5551 | #endif | |
5552 | return false; | |
5553 | } | |
5554 | ||
1da177e4 LT |
5555 | /* |
5556 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 5557 | * up by memblock_free_all() once the early boot process is |
1da177e4 LT |
5558 | * done. Non-atomic initialization, single-pass. |
5559 | */ | |
c09b4240 | 5560 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a99583e7 CH |
5561 | unsigned long start_pfn, enum memmap_context context, |
5562 | struct vmem_altmap *altmap) | |
1da177e4 | 5563 | { |
a9a9e77f | 5564 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 5565 | struct page *page; |
1da177e4 | 5566 | |
22b31eec HD |
5567 | if (highest_memmap_pfn < end_pfn - 1) |
5568 | highest_memmap_pfn = end_pfn - 1; | |
5569 | ||
966cf44f | 5570 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
5571 | /* |
5572 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
5573 | * memory. We limit the total number of pages to initialize to just |
5574 | * those that might contain the memory mapping. We will defer the | |
5575 | * ZONE_DEVICE page initialization until after we have released | |
5576 | * the hotplug lock. | |
4b94ffdc | 5577 | */ |
966cf44f AD |
5578 | if (zone == ZONE_DEVICE) { |
5579 | if (!altmap) | |
5580 | return; | |
5581 | ||
5582 | if (start_pfn == altmap->base_pfn) | |
5583 | start_pfn += altmap->reserve; | |
5584 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
5585 | } | |
5586 | #endif | |
4b94ffdc | 5587 | |
cbe8dd4a | 5588 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 5589 | /* |
b72d0ffb AM |
5590 | * There can be holes in boot-time mem_map[]s handed to this |
5591 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5592 | */ |
a9a9e77f PT |
5593 | if (context == MEMMAP_EARLY) { |
5594 | if (!early_pfn_valid(pfn)) | |
b72d0ffb | 5595 | continue; |
a9a9e77f PT |
5596 | if (!early_pfn_in_nid(pfn, nid)) |
5597 | continue; | |
5598 | if (overlap_memmap_init(zone, &pfn)) | |
5599 | continue; | |
5600 | if (defer_init(nid, pfn, end_pfn)) | |
5601 | break; | |
a2f3aa02 | 5602 | } |
ac5d2539 | 5603 | |
d0dc12e8 PT |
5604 | page = pfn_to_page(pfn); |
5605 | __init_single_page(page, pfn, zone, nid); | |
5606 | if (context == MEMMAP_HOTPLUG) | |
d483da5b | 5607 | __SetPageReserved(page); |
d0dc12e8 | 5608 | |
ac5d2539 MG |
5609 | /* |
5610 | * Mark the block movable so that blocks are reserved for | |
5611 | * movable at startup. This will force kernel allocations | |
5612 | * to reserve their blocks rather than leaking throughout | |
5613 | * the address space during boot when many long-lived | |
974a786e | 5614 | * kernel allocations are made. |
ac5d2539 MG |
5615 | * |
5616 | * bitmap is created for zone's valid pfn range. but memmap | |
5617 | * can be created for invalid pages (for alignment) | |
5618 | * check here not to call set_pageblock_migratetype() against | |
5619 | * pfn out of zone. | |
5620 | */ | |
5621 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
ac5d2539 | 5622 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
9b6e63cb | 5623 | cond_resched(); |
ac5d2539 | 5624 | } |
1da177e4 | 5625 | } |
2830bf6f MZ |
5626 | #ifdef CONFIG_SPARSEMEM |
5627 | /* | |
5628 | * If the zone does not span the rest of the section then | |
5629 | * we should at least initialize those pages. Otherwise we | |
5630 | * could blow up on a poisoned page in some paths which depend | |
5631 | * on full sections being initialized (e.g. memory hotplug). | |
5632 | */ | |
5633 | while (end_pfn % PAGES_PER_SECTION) { | |
5634 | __init_single_page(pfn_to_page(end_pfn), end_pfn, zone, nid); | |
5635 | end_pfn++; | |
5636 | } | |
5637 | #endif | |
1da177e4 LT |
5638 | } |
5639 | ||
966cf44f AD |
5640 | #ifdef CONFIG_ZONE_DEVICE |
5641 | void __ref memmap_init_zone_device(struct zone *zone, | |
5642 | unsigned long start_pfn, | |
5643 | unsigned long size, | |
5644 | struct dev_pagemap *pgmap) | |
5645 | { | |
5646 | unsigned long pfn, end_pfn = start_pfn + size; | |
5647 | struct pglist_data *pgdat = zone->zone_pgdat; | |
5648 | unsigned long zone_idx = zone_idx(zone); | |
5649 | unsigned long start = jiffies; | |
5650 | int nid = pgdat->node_id; | |
5651 | ||
5652 | if (WARN_ON_ONCE(!pgmap || !is_dev_zone(zone))) | |
5653 | return; | |
5654 | ||
5655 | /* | |
5656 | * The call to memmap_init_zone should have already taken care | |
5657 | * of the pages reserved for the memmap, so we can just jump to | |
5658 | * the end of that region and start processing the device pages. | |
5659 | */ | |
5660 | if (pgmap->altmap_valid) { | |
5661 | struct vmem_altmap *altmap = &pgmap->altmap; | |
5662 | ||
5663 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
5664 | size = end_pfn - start_pfn; | |
5665 | } | |
5666 | ||
5667 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
5668 | struct page *page = pfn_to_page(pfn); | |
5669 | ||
5670 | __init_single_page(page, pfn, zone_idx, nid); | |
5671 | ||
5672 | /* | |
5673 | * Mark page reserved as it will need to wait for onlining | |
5674 | * phase for it to be fully associated with a zone. | |
5675 | * | |
5676 | * We can use the non-atomic __set_bit operation for setting | |
5677 | * the flag as we are still initializing the pages. | |
5678 | */ | |
5679 | __SetPageReserved(page); | |
5680 | ||
5681 | /* | |
5682 | * ZONE_DEVICE pages union ->lru with a ->pgmap back | |
5683 | * pointer and hmm_data. It is a bug if a ZONE_DEVICE | |
5684 | * page is ever freed or placed on a driver-private list. | |
5685 | */ | |
5686 | page->pgmap = pgmap; | |
5687 | page->hmm_data = 0; | |
5688 | ||
5689 | /* | |
5690 | * Mark the block movable so that blocks are reserved for | |
5691 | * movable at startup. This will force kernel allocations | |
5692 | * to reserve their blocks rather than leaking throughout | |
5693 | * the address space during boot when many long-lived | |
5694 | * kernel allocations are made. | |
5695 | * | |
5696 | * bitmap is created for zone's valid pfn range. but memmap | |
5697 | * can be created for invalid pages (for alignment) | |
5698 | * check here not to call set_pageblock_migratetype() against | |
5699 | * pfn out of zone. | |
5700 | * | |
5701 | * Please note that MEMMAP_HOTPLUG path doesn't clear memmap | |
5702 | * because this is done early in sparse_add_one_section | |
5703 | */ | |
5704 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
5705 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5706 | cond_resched(); | |
5707 | } | |
5708 | } | |
5709 | ||
5710 | pr_info("%s initialised, %lu pages in %ums\n", dev_name(pgmap->dev), | |
5711 | size, jiffies_to_msecs(jiffies - start)); | |
5712 | } | |
5713 | ||
5714 | #endif | |
1e548deb | 5715 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 5716 | { |
7aeb09f9 | 5717 | unsigned int order, t; |
b2a0ac88 MG |
5718 | for_each_migratetype_order(order, t) { |
5719 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
5720 | zone->free_area[order].nr_free = 0; |
5721 | } | |
5722 | } | |
5723 | ||
dfb3ccd0 PT |
5724 | void __meminit __weak memmap_init(unsigned long size, int nid, |
5725 | unsigned long zone, unsigned long start_pfn) | |
5726 | { | |
5727 | memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY, NULL); | |
5728 | } | |
1da177e4 | 5729 | |
7cd2b0a3 | 5730 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 5731 | { |
3a6be87f | 5732 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
5733 | int batch; |
5734 | ||
5735 | /* | |
5736 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 5737 | * size of the zone. |
e7c8d5c9 | 5738 | */ |
9705bea5 | 5739 | batch = zone_managed_pages(zone) / 1024; |
d8a759b5 AL |
5740 | /* But no more than a meg. */ |
5741 | if (batch * PAGE_SIZE > 1024 * 1024) | |
5742 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
5743 | batch /= 4; /* We effectively *= 4 below */ |
5744 | if (batch < 1) | |
5745 | batch = 1; | |
5746 | ||
5747 | /* | |
0ceaacc9 NP |
5748 | * Clamp the batch to a 2^n - 1 value. Having a power |
5749 | * of 2 value was found to be more likely to have | |
5750 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 5751 | * |
0ceaacc9 NP |
5752 | * For example if 2 tasks are alternately allocating |
5753 | * batches of pages, one task can end up with a lot | |
5754 | * of pages of one half of the possible page colors | |
5755 | * and the other with pages of the other colors. | |
e7c8d5c9 | 5756 | */ |
9155203a | 5757 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 5758 | |
e7c8d5c9 | 5759 | return batch; |
3a6be87f DH |
5760 | |
5761 | #else | |
5762 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5763 | * conditions. | |
5764 | * | |
5765 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5766 | * of contiguous memory as there's no hardware page translation to | |
5767 | * assemble apparent contiguous memory from discontiguous pages. | |
5768 | * | |
5769 | * Queueing large contiguous runs of pages for batching, however, | |
5770 | * causes the pages to actually be freed in smaller chunks. As there | |
5771 | * can be a significant delay between the individual batches being | |
5772 | * recycled, this leads to the once large chunks of space being | |
5773 | * fragmented and becoming unavailable for high-order allocations. | |
5774 | */ | |
5775 | return 0; | |
5776 | #endif | |
e7c8d5c9 CL |
5777 | } |
5778 | ||
8d7a8fa9 CS |
5779 | /* |
5780 | * pcp->high and pcp->batch values are related and dependent on one another: | |
5781 | * ->batch must never be higher then ->high. | |
5782 | * The following function updates them in a safe manner without read side | |
5783 | * locking. | |
5784 | * | |
5785 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
5786 | * those fields changing asynchronously (acording the the above rule). | |
5787 | * | |
5788 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5789 | * outside of boot time (or some other assurance that no concurrent updaters | |
5790 | * exist). | |
5791 | */ | |
5792 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
5793 | unsigned long batch) | |
5794 | { | |
5795 | /* start with a fail safe value for batch */ | |
5796 | pcp->batch = 1; | |
5797 | smp_wmb(); | |
5798 | ||
5799 | /* Update high, then batch, in order */ | |
5800 | pcp->high = high; | |
5801 | smp_wmb(); | |
5802 | ||
5803 | pcp->batch = batch; | |
5804 | } | |
5805 | ||
3664033c | 5806 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
5807 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5808 | { | |
8d7a8fa9 | 5809 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
5810 | } |
5811 | ||
88c90dbc | 5812 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
5813 | { |
5814 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 5815 | int migratetype; |
2caaad41 | 5816 | |
1c6fe946 MD |
5817 | memset(p, 0, sizeof(*p)); |
5818 | ||
3dfa5721 | 5819 | pcp = &p->pcp; |
5f8dcc21 MG |
5820 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5821 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
5822 | } |
5823 | ||
88c90dbc CS |
5824 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5825 | { | |
5826 | pageset_init(p); | |
5827 | pageset_set_batch(p, batch); | |
5828 | } | |
5829 | ||
8ad4b1fb | 5830 | /* |
3664033c | 5831 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
5832 | * to the value high for the pageset p. |
5833 | */ | |
3664033c | 5834 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
5835 | unsigned long high) |
5836 | { | |
8d7a8fa9 CS |
5837 | unsigned long batch = max(1UL, high / 4); |
5838 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
5839 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 5840 | |
8d7a8fa9 | 5841 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
5842 | } |
5843 | ||
7cd2b0a3 DR |
5844 | static void pageset_set_high_and_batch(struct zone *zone, |
5845 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 5846 | { |
56cef2b8 | 5847 | if (percpu_pagelist_fraction) |
3664033c | 5848 | pageset_set_high(pcp, |
9705bea5 | 5849 | (zone_managed_pages(zone) / |
56cef2b8 CS |
5850 | percpu_pagelist_fraction)); |
5851 | else | |
5852 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
5853 | } | |
5854 | ||
169f6c19 CS |
5855 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5856 | { | |
5857 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
5858 | ||
5859 | pageset_init(pcp); | |
5860 | pageset_set_high_and_batch(zone, pcp); | |
5861 | } | |
5862 | ||
72675e13 | 5863 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
5864 | { |
5865 | int cpu; | |
319774e2 | 5866 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5867 | for_each_possible_cpu(cpu) |
5868 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5869 | } |
5870 | ||
2caaad41 | 5871 | /* |
99dcc3e5 CL |
5872 | * Allocate per cpu pagesets and initialize them. |
5873 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5874 | */ |
99dcc3e5 | 5875 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5876 | { |
b4911ea2 | 5877 | struct pglist_data *pgdat; |
99dcc3e5 | 5878 | struct zone *zone; |
e7c8d5c9 | 5879 | |
319774e2 WF |
5880 | for_each_populated_zone(zone) |
5881 | setup_zone_pageset(zone); | |
b4911ea2 MG |
5882 | |
5883 | for_each_online_pgdat(pgdat) | |
5884 | pgdat->per_cpu_nodestats = | |
5885 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
5886 | } |
5887 | ||
c09b4240 | 5888 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5889 | { |
99dcc3e5 CL |
5890 | /* |
5891 | * per cpu subsystem is not up at this point. The following code | |
5892 | * relies on the ability of the linker to provide the | |
5893 | * offset of a (static) per cpu variable into the per cpu area. | |
5894 | */ | |
5895 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5896 | |
b38a8725 | 5897 | if (populated_zone(zone)) |
99dcc3e5 CL |
5898 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5899 | zone->name, zone->present_pages, | |
5900 | zone_batchsize(zone)); | |
ed8ece2e DH |
5901 | } |
5902 | ||
dc0bbf3b | 5903 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5904 | unsigned long zone_start_pfn, |
b171e409 | 5905 | unsigned long size) |
ed8ece2e DH |
5906 | { |
5907 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 5908 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 5909 | |
8f416836 WY |
5910 | if (zone_idx > pgdat->nr_zones) |
5911 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 5912 | |
ed8ece2e DH |
5913 | zone->zone_start_pfn = zone_start_pfn; |
5914 | ||
708614e6 MG |
5915 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5916 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5917 | pgdat->node_id, | |
5918 | (unsigned long)zone_idx(zone), | |
5919 | zone_start_pfn, (zone_start_pfn + size)); | |
5920 | ||
1e548deb | 5921 | zone_init_free_lists(zone); |
9dcb8b68 | 5922 | zone->initialized = 1; |
ed8ece2e DH |
5923 | } |
5924 | ||
0ee332c1 | 5925 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5926 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5927 | |
c713216d MG |
5928 | /* |
5929 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5930 | */ |
8a942fde MG |
5931 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5932 | struct mminit_pfnnid_cache *state) | |
c713216d | 5933 | { |
c13291a5 | 5934 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5935 | int nid; |
7c243c71 | 5936 | |
8a942fde MG |
5937 | if (state->last_start <= pfn && pfn < state->last_end) |
5938 | return state->last_nid; | |
c713216d | 5939 | |
e76b63f8 YL |
5940 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5941 | if (nid != -1) { | |
8a942fde MG |
5942 | state->last_start = start_pfn; |
5943 | state->last_end = end_pfn; | |
5944 | state->last_nid = nid; | |
e76b63f8 YL |
5945 | } |
5946 | ||
5947 | return nid; | |
c713216d MG |
5948 | } |
5949 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5950 | ||
c713216d | 5951 | /** |
6782832e | 5952 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5953 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5954 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5955 | * |
7d018176 ZZ |
5956 | * If an architecture guarantees that all ranges registered contain no holes |
5957 | * and may be freed, this this function may be used instead of calling | |
5958 | * memblock_free_early_nid() manually. | |
c713216d | 5959 | */ |
c13291a5 | 5960 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5961 | { |
c13291a5 TH |
5962 | unsigned long start_pfn, end_pfn; |
5963 | int i, this_nid; | |
edbe7d23 | 5964 | |
c13291a5 TH |
5965 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5966 | start_pfn = min(start_pfn, max_low_pfn); | |
5967 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5968 | |
c13291a5 | 5969 | if (start_pfn < end_pfn) |
6782832e SS |
5970 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5971 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5972 | this_nid); | |
edbe7d23 | 5973 | } |
edbe7d23 | 5974 | } |
edbe7d23 | 5975 | |
c713216d MG |
5976 | /** |
5977 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5978 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5979 | * |
7d018176 ZZ |
5980 | * If an architecture guarantees that all ranges registered contain no holes and may |
5981 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5982 | */ |
5983 | void __init sparse_memory_present_with_active_regions(int nid) | |
5984 | { | |
c13291a5 TH |
5985 | unsigned long start_pfn, end_pfn; |
5986 | int i, this_nid; | |
c713216d | 5987 | |
c13291a5 TH |
5988 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5989 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5990 | } |
5991 | ||
5992 | /** | |
5993 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5994 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5995 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5996 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5997 | * |
5998 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5999 | * provided by memblock_set_node(). If called for a node |
c713216d | 6000 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 6001 | * PFNs will be 0. |
c713216d | 6002 | */ |
a3142c8e | 6003 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
6004 | unsigned long *start_pfn, unsigned long *end_pfn) |
6005 | { | |
c13291a5 | 6006 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 6007 | int i; |
c13291a5 | 6008 | |
c713216d MG |
6009 | *start_pfn = -1UL; |
6010 | *end_pfn = 0; | |
6011 | ||
c13291a5 TH |
6012 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
6013 | *start_pfn = min(*start_pfn, this_start_pfn); | |
6014 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
6015 | } |
6016 | ||
633c0666 | 6017 | if (*start_pfn == -1UL) |
c713216d | 6018 | *start_pfn = 0; |
c713216d MG |
6019 | } |
6020 | ||
2a1e274a MG |
6021 | /* |
6022 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
6023 | * assumption is made that zones within a node are ordered in monotonic | |
6024 | * increasing memory addresses so that the "highest" populated zone is used | |
6025 | */ | |
b69a7288 | 6026 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
6027 | { |
6028 | int zone_index; | |
6029 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
6030 | if (zone_index == ZONE_MOVABLE) | |
6031 | continue; | |
6032 | ||
6033 | if (arch_zone_highest_possible_pfn[zone_index] > | |
6034 | arch_zone_lowest_possible_pfn[zone_index]) | |
6035 | break; | |
6036 | } | |
6037 | ||
6038 | VM_BUG_ON(zone_index == -1); | |
6039 | movable_zone = zone_index; | |
6040 | } | |
6041 | ||
6042 | /* | |
6043 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 6044 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
6045 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
6046 | * in each node depending on the size of each node and how evenly kernelcore | |
6047 | * is distributed. This helper function adjusts the zone ranges | |
6048 | * provided by the architecture for a given node by using the end of the | |
6049 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
6050 | * zones within a node are in order of monotonic increases memory addresses | |
6051 | */ | |
b69a7288 | 6052 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
6053 | unsigned long zone_type, |
6054 | unsigned long node_start_pfn, | |
6055 | unsigned long node_end_pfn, | |
6056 | unsigned long *zone_start_pfn, | |
6057 | unsigned long *zone_end_pfn) | |
6058 | { | |
6059 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
6060 | if (zone_movable_pfn[nid]) { | |
6061 | /* Size ZONE_MOVABLE */ | |
6062 | if (zone_type == ZONE_MOVABLE) { | |
6063 | *zone_start_pfn = zone_movable_pfn[nid]; | |
6064 | *zone_end_pfn = min(node_end_pfn, | |
6065 | arch_zone_highest_possible_pfn[movable_zone]); | |
6066 | ||
e506b996 XQ |
6067 | /* Adjust for ZONE_MOVABLE starting within this range */ |
6068 | } else if (!mirrored_kernelcore && | |
6069 | *zone_start_pfn < zone_movable_pfn[nid] && | |
6070 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
6071 | *zone_end_pfn = zone_movable_pfn[nid]; | |
6072 | ||
2a1e274a MG |
6073 | /* Check if this whole range is within ZONE_MOVABLE */ |
6074 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
6075 | *zone_start_pfn = *zone_end_pfn; | |
6076 | } | |
6077 | } | |
6078 | ||
c713216d MG |
6079 | /* |
6080 | * Return the number of pages a zone spans in a node, including holes | |
6081 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
6082 | */ | |
6ea6e688 | 6083 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 6084 | unsigned long zone_type, |
7960aedd ZY |
6085 | unsigned long node_start_pfn, |
6086 | unsigned long node_end_pfn, | |
d91749c1 TI |
6087 | unsigned long *zone_start_pfn, |
6088 | unsigned long *zone_end_pfn, | |
c713216d MG |
6089 | unsigned long *ignored) |
6090 | { | |
b5685e92 | 6091 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6092 | if (!node_start_pfn && !node_end_pfn) |
6093 | return 0; | |
6094 | ||
7960aedd | 6095 | /* Get the start and end of the zone */ |
d91749c1 TI |
6096 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
6097 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
6098 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6099 | node_start_pfn, node_end_pfn, | |
d91749c1 | 6100 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
6101 | |
6102 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 6103 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
6104 | return 0; |
6105 | ||
6106 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
6107 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
6108 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
6109 | |
6110 | /* Return the spanned pages */ | |
d91749c1 | 6111 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
6112 | } |
6113 | ||
6114 | /* | |
6115 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 6116 | * then all holes in the requested range will be accounted for. |
c713216d | 6117 | */ |
32996250 | 6118 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
6119 | unsigned long range_start_pfn, |
6120 | unsigned long range_end_pfn) | |
6121 | { | |
96e907d1 TH |
6122 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
6123 | unsigned long start_pfn, end_pfn; | |
6124 | int i; | |
c713216d | 6125 | |
96e907d1 TH |
6126 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6127 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6128 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6129 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 6130 | } |
96e907d1 | 6131 | return nr_absent; |
c713216d MG |
6132 | } |
6133 | ||
6134 | /** | |
6135 | * absent_pages_in_range - Return number of page frames in holes within a range | |
6136 | * @start_pfn: The start PFN to start searching for holes | |
6137 | * @end_pfn: The end PFN to stop searching for holes | |
6138 | * | |
88ca3b94 | 6139 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
6140 | */ |
6141 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
6142 | unsigned long end_pfn) | |
6143 | { | |
6144 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
6145 | } | |
6146 | ||
6147 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 6148 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 6149 | unsigned long zone_type, |
7960aedd ZY |
6150 | unsigned long node_start_pfn, |
6151 | unsigned long node_end_pfn, | |
c713216d MG |
6152 | unsigned long *ignored) |
6153 | { | |
96e907d1 TH |
6154 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6155 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 6156 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 6157 | unsigned long nr_absent; |
9c7cd687 | 6158 | |
b5685e92 | 6159 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6160 | if (!node_start_pfn && !node_end_pfn) |
6161 | return 0; | |
6162 | ||
96e907d1 TH |
6163 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6164 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 6165 | |
2a1e274a MG |
6166 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6167 | node_start_pfn, node_end_pfn, | |
6168 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
6169 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
6170 | ||
6171 | /* | |
6172 | * ZONE_MOVABLE handling. | |
6173 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6174 | * and vice versa. | |
6175 | */ | |
e506b996 XQ |
6176 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6177 | unsigned long start_pfn, end_pfn; | |
6178 | struct memblock_region *r; | |
6179 | ||
6180 | for_each_memblock(memory, r) { | |
6181 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
6182 | zone_start_pfn, zone_end_pfn); | |
6183 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6184 | zone_start_pfn, zone_end_pfn); | |
6185 | ||
6186 | if (zone_type == ZONE_MOVABLE && | |
6187 | memblock_is_mirror(r)) | |
6188 | nr_absent += end_pfn - start_pfn; | |
6189 | ||
6190 | if (zone_type == ZONE_NORMAL && | |
6191 | !memblock_is_mirror(r)) | |
6192 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6193 | } |
6194 | } | |
6195 | ||
6196 | return nr_absent; | |
c713216d | 6197 | } |
0e0b864e | 6198 | |
0ee332c1 | 6199 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 6200 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 6201 | unsigned long zone_type, |
7960aedd ZY |
6202 | unsigned long node_start_pfn, |
6203 | unsigned long node_end_pfn, | |
d91749c1 TI |
6204 | unsigned long *zone_start_pfn, |
6205 | unsigned long *zone_end_pfn, | |
c713216d MG |
6206 | unsigned long *zones_size) |
6207 | { | |
d91749c1 TI |
6208 | unsigned int zone; |
6209 | ||
6210 | *zone_start_pfn = node_start_pfn; | |
6211 | for (zone = 0; zone < zone_type; zone++) | |
6212 | *zone_start_pfn += zones_size[zone]; | |
6213 | ||
6214 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
6215 | ||
c713216d MG |
6216 | return zones_size[zone_type]; |
6217 | } | |
6218 | ||
6ea6e688 | 6219 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 6220 | unsigned long zone_type, |
7960aedd ZY |
6221 | unsigned long node_start_pfn, |
6222 | unsigned long node_end_pfn, | |
c713216d MG |
6223 | unsigned long *zholes_size) |
6224 | { | |
6225 | if (!zholes_size) | |
6226 | return 0; | |
6227 | ||
6228 | return zholes_size[zone_type]; | |
6229 | } | |
20e6926d | 6230 | |
0ee332c1 | 6231 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6232 | |
a3142c8e | 6233 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
6234 | unsigned long node_start_pfn, |
6235 | unsigned long node_end_pfn, | |
6236 | unsigned long *zones_size, | |
6237 | unsigned long *zholes_size) | |
c713216d | 6238 | { |
febd5949 | 6239 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6240 | enum zone_type i; |
6241 | ||
febd5949 GZ |
6242 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6243 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6244 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 6245 | unsigned long size, real_size; |
c713216d | 6246 | |
febd5949 GZ |
6247 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
6248 | node_start_pfn, | |
6249 | node_end_pfn, | |
d91749c1 TI |
6250 | &zone_start_pfn, |
6251 | &zone_end_pfn, | |
febd5949 GZ |
6252 | zones_size); |
6253 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
6254 | node_start_pfn, node_end_pfn, |
6255 | zholes_size); | |
d91749c1 TI |
6256 | if (size) |
6257 | zone->zone_start_pfn = zone_start_pfn; | |
6258 | else | |
6259 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
6260 | zone->spanned_pages = size; |
6261 | zone->present_pages = real_size; | |
6262 | ||
6263 | totalpages += size; | |
6264 | realtotalpages += real_size; | |
6265 | } | |
6266 | ||
6267 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
6268 | pgdat->node_present_pages = realtotalpages; |
6269 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
6270 | realtotalpages); | |
6271 | } | |
6272 | ||
835c134e MG |
6273 | #ifndef CONFIG_SPARSEMEM |
6274 | /* | |
6275 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
6276 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
6277 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
6278 | * round what is now in bits to nearest long in bits, then return it in |
6279 | * bytes. | |
6280 | */ | |
7c45512d | 6281 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
6282 | { |
6283 | unsigned long usemapsize; | |
6284 | ||
7c45512d | 6285 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
6286 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
6287 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
6288 | usemapsize *= NR_PAGEBLOCK_BITS; |
6289 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
6290 | ||
6291 | return usemapsize / 8; | |
6292 | } | |
6293 | ||
7cc2a959 | 6294 | static void __ref setup_usemap(struct pglist_data *pgdat, |
7c45512d LT |
6295 | struct zone *zone, |
6296 | unsigned long zone_start_pfn, | |
6297 | unsigned long zonesize) | |
835c134e | 6298 | { |
7c45512d | 6299 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 6300 | zone->pageblock_flags = NULL; |
58a01a45 | 6301 | if (usemapsize) |
6782832e | 6302 | zone->pageblock_flags = |
eb31d559 | 6303 | memblock_alloc_node_nopanic(usemapsize, |
6782832e | 6304 | pgdat->node_id); |
835c134e MG |
6305 | } |
6306 | #else | |
7c45512d LT |
6307 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
6308 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
6309 | #endif /* CONFIG_SPARSEMEM */ |
6310 | ||
d9c23400 | 6311 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 6312 | |
d9c23400 | 6313 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 6314 | void __init set_pageblock_order(void) |
d9c23400 | 6315 | { |
955c1cd7 AM |
6316 | unsigned int order; |
6317 | ||
d9c23400 MG |
6318 | /* Check that pageblock_nr_pages has not already been setup */ |
6319 | if (pageblock_order) | |
6320 | return; | |
6321 | ||
955c1cd7 AM |
6322 | if (HPAGE_SHIFT > PAGE_SHIFT) |
6323 | order = HUGETLB_PAGE_ORDER; | |
6324 | else | |
6325 | order = MAX_ORDER - 1; | |
6326 | ||
d9c23400 MG |
6327 | /* |
6328 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
6329 | * This value may be variable depending on boot parameters on IA64 and |
6330 | * powerpc. | |
d9c23400 MG |
6331 | */ |
6332 | pageblock_order = order; | |
6333 | } | |
6334 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6335 | ||
ba72cb8c MG |
6336 | /* |
6337 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
6338 | * is unused as pageblock_order is set at compile-time. See |
6339 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
6340 | * the kernel config | |
ba72cb8c | 6341 | */ |
03e85f9d | 6342 | void __init set_pageblock_order(void) |
ba72cb8c | 6343 | { |
ba72cb8c | 6344 | } |
d9c23400 MG |
6345 | |
6346 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6347 | ||
03e85f9d | 6348 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 6349 | unsigned long present_pages) |
01cefaef JL |
6350 | { |
6351 | unsigned long pages = spanned_pages; | |
6352 | ||
6353 | /* | |
6354 | * Provide a more accurate estimation if there are holes within | |
6355 | * the zone and SPARSEMEM is in use. If there are holes within the | |
6356 | * zone, each populated memory region may cost us one or two extra | |
6357 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 6358 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
6359 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
6360 | */ | |
6361 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
6362 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
6363 | pages = present_pages; | |
6364 | ||
6365 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
6366 | } | |
6367 | ||
ace1db39 OS |
6368 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6369 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
6370 | { | |
6371 | spin_lock_init(&pgdat->split_queue_lock); | |
6372 | INIT_LIST_HEAD(&pgdat->split_queue); | |
6373 | pgdat->split_queue_len = 0; | |
6374 | } | |
6375 | #else | |
6376 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
6377 | #endif | |
6378 | ||
6379 | #ifdef CONFIG_COMPACTION | |
6380 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
6381 | { | |
6382 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6383 | } | |
6384 | #else | |
6385 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
6386 | #endif | |
6387 | ||
03e85f9d | 6388 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 6389 | { |
208d54e5 | 6390 | pgdat_resize_init(pgdat); |
ace1db39 | 6391 | |
ace1db39 OS |
6392 | pgdat_init_split_queue(pgdat); |
6393 | pgdat_init_kcompactd(pgdat); | |
6394 | ||
1da177e4 | 6395 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6396 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 6397 | |
eefa864b | 6398 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6399 | spin_lock_init(&pgdat->lru_lock); |
a9dd0a83 | 6400 | lruvec_init(node_lruvec(pgdat)); |
03e85f9d OS |
6401 | } |
6402 | ||
6403 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
6404 | unsigned long remaining_pages) | |
6405 | { | |
9705bea5 | 6406 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
6407 | zone_set_nid(zone, nid); |
6408 | zone->name = zone_names[idx]; | |
6409 | zone->zone_pgdat = NODE_DATA(nid); | |
6410 | spin_lock_init(&zone->lock); | |
6411 | zone_seqlock_init(zone); | |
6412 | zone_pcp_init(zone); | |
6413 | } | |
6414 | ||
6415 | /* | |
6416 | * Set up the zone data structures | |
6417 | * - init pgdat internals | |
6418 | * - init all zones belonging to this node | |
6419 | * | |
6420 | * NOTE: this function is only called during memory hotplug | |
6421 | */ | |
6422 | #ifdef CONFIG_MEMORY_HOTPLUG | |
6423 | void __ref free_area_init_core_hotplug(int nid) | |
6424 | { | |
6425 | enum zone_type z; | |
6426 | pg_data_t *pgdat = NODE_DATA(nid); | |
6427 | ||
6428 | pgdat_init_internals(pgdat); | |
6429 | for (z = 0; z < MAX_NR_ZONES; z++) | |
6430 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
6431 | } | |
6432 | #endif | |
6433 | ||
6434 | /* | |
6435 | * Set up the zone data structures: | |
6436 | * - mark all pages reserved | |
6437 | * - mark all memory queues empty | |
6438 | * - clear the memory bitmaps | |
6439 | * | |
6440 | * NOTE: pgdat should get zeroed by caller. | |
6441 | * NOTE: this function is only called during early init. | |
6442 | */ | |
6443 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
6444 | { | |
6445 | enum zone_type j; | |
6446 | int nid = pgdat->node_id; | |
5f63b720 | 6447 | |
03e85f9d | 6448 | pgdat_init_internals(pgdat); |
385386cf JW |
6449 | pgdat->per_cpu_nodestats = &boot_nodestats; |
6450 | ||
1da177e4 LT |
6451 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6452 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 6453 | unsigned long size, freesize, memmap_pages; |
d91749c1 | 6454 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6455 | |
febd5949 | 6456 | size = zone->spanned_pages; |
e6943859 | 6457 | freesize = zone->present_pages; |
1da177e4 | 6458 | |
0e0b864e | 6459 | /* |
9feedc9d | 6460 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6461 | * is used by this zone for memmap. This affects the watermark |
6462 | * and per-cpu initialisations | |
6463 | */ | |
e6943859 | 6464 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
6465 | if (!is_highmem_idx(j)) { |
6466 | if (freesize >= memmap_pages) { | |
6467 | freesize -= memmap_pages; | |
6468 | if (memmap_pages) | |
6469 | printk(KERN_DEBUG | |
6470 | " %s zone: %lu pages used for memmap\n", | |
6471 | zone_names[j], memmap_pages); | |
6472 | } else | |
1170532b | 6473 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6474 | zone_names[j], memmap_pages, freesize); |
6475 | } | |
0e0b864e | 6476 | |
6267276f | 6477 | /* Account for reserved pages */ |
9feedc9d JL |
6478 | if (j == 0 && freesize > dma_reserve) { |
6479 | freesize -= dma_reserve; | |
d903ef9f | 6480 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6481 | zone_names[0], dma_reserve); |
0e0b864e MG |
6482 | } |
6483 | ||
98d2b0eb | 6484 | if (!is_highmem_idx(j)) |
9feedc9d | 6485 | nr_kernel_pages += freesize; |
01cefaef JL |
6486 | /* Charge for highmem memmap if there are enough kernel pages */ |
6487 | else if (nr_kernel_pages > memmap_pages * 2) | |
6488 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6489 | nr_all_pages += freesize; |
1da177e4 | 6490 | |
9feedc9d JL |
6491 | /* |
6492 | * Set an approximate value for lowmem here, it will be adjusted | |
6493 | * when the bootmem allocator frees pages into the buddy system. | |
6494 | * And all highmem pages will be managed by the buddy system. | |
6495 | */ | |
03e85f9d | 6496 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 6497 | |
d883c6cf | 6498 | if (!size) |
1da177e4 LT |
6499 | continue; |
6500 | ||
955c1cd7 | 6501 | set_pageblock_order(); |
d883c6cf JK |
6502 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
6503 | init_currently_empty_zone(zone, zone_start_pfn, size); | |
76cdd58e | 6504 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6505 | } |
6506 | } | |
6507 | ||
0cd842f9 | 6508 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 6509 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6510 | { |
b0aeba74 | 6511 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6512 | unsigned long __maybe_unused offset = 0; |
6513 | ||
1da177e4 LT |
6514 | /* Skip empty nodes */ |
6515 | if (!pgdat->node_spanned_pages) | |
6516 | return; | |
6517 | ||
b0aeba74 TL |
6518 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6519 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6520 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6521 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6522 | unsigned long size, end; |
d41dee36 AW |
6523 | struct page *map; |
6524 | ||
e984bb43 BP |
6525 | /* |
6526 | * The zone's endpoints aren't required to be MAX_ORDER | |
6527 | * aligned but the node_mem_map endpoints must be in order | |
6528 | * for the buddy allocator to function correctly. | |
6529 | */ | |
108bcc96 | 6530 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6531 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6532 | size = (end - start) * sizeof(struct page); | |
eb31d559 | 6533 | map = memblock_alloc_node_nopanic(size, pgdat->node_id); |
a1c34a3b | 6534 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6535 | } |
0cd842f9 OS |
6536 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
6537 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
6538 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 6539 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6540 | /* |
6541 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6542 | */ | |
c713216d | 6543 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6544 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 6545 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 6546 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6547 | mem_map -= offset; |
0ee332c1 | 6548 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6549 | } |
1da177e4 LT |
6550 | #endif |
6551 | } | |
0cd842f9 OS |
6552 | #else |
6553 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
6554 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 6555 | |
0188dc98 OS |
6556 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
6557 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
6558 | { | |
6559 | /* | |
6560 | * We start only with one section of pages, more pages are added as | |
6561 | * needed until the rest of deferred pages are initialized. | |
6562 | */ | |
6563 | pgdat->static_init_pgcnt = min_t(unsigned long, PAGES_PER_SECTION, | |
6564 | pgdat->node_spanned_pages); | |
6565 | pgdat->first_deferred_pfn = ULONG_MAX; | |
6566 | } | |
6567 | #else | |
6568 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
6569 | #endif | |
6570 | ||
03e85f9d | 6571 | void __init free_area_init_node(int nid, unsigned long *zones_size, |
7cc2a959 PT |
6572 | unsigned long node_start_pfn, |
6573 | unsigned long *zholes_size) | |
1da177e4 | 6574 | { |
9109fb7b | 6575 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6576 | unsigned long start_pfn = 0; |
6577 | unsigned long end_pfn = 0; | |
9109fb7b | 6578 | |
88fdf75d | 6579 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 6580 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 6581 | |
1da177e4 LT |
6582 | pgdat->node_id = nid; |
6583 | pgdat->node_start_pfn = node_start_pfn; | |
75ef7184 | 6584 | pgdat->per_cpu_nodestats = NULL; |
7960aedd ZY |
6585 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6586 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 6587 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6588 | (u64)start_pfn << PAGE_SHIFT, |
6589 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
6590 | #else |
6591 | start_pfn = node_start_pfn; | |
7960aedd ZY |
6592 | #endif |
6593 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
6594 | zones_size, zholes_size); | |
1da177e4 LT |
6595 | |
6596 | alloc_node_mem_map(pgdat); | |
0188dc98 | 6597 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 6598 | |
7f3eb55b | 6599 | free_area_init_core(pgdat); |
1da177e4 LT |
6600 | } |
6601 | ||
aca52c39 | 6602 | #if !defined(CONFIG_FLAT_NODE_MEM_MAP) |
ec393a0f PT |
6603 | /* |
6604 | * Zero all valid struct pages in range [spfn, epfn), return number of struct | |
6605 | * pages zeroed | |
6606 | */ | |
6607 | static u64 zero_pfn_range(unsigned long spfn, unsigned long epfn) | |
6608 | { | |
6609 | unsigned long pfn; | |
6610 | u64 pgcnt = 0; | |
6611 | ||
6612 | for (pfn = spfn; pfn < epfn; pfn++) { | |
6613 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { | |
6614 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6615 | + pageblock_nr_pages - 1; | |
6616 | continue; | |
6617 | } | |
6618 | mm_zero_struct_page(pfn_to_page(pfn)); | |
6619 | pgcnt++; | |
6620 | } | |
6621 | ||
6622 | return pgcnt; | |
6623 | } | |
6624 | ||
a4a3ede2 PT |
6625 | /* |
6626 | * Only struct pages that are backed by physical memory are zeroed and | |
6627 | * initialized by going through __init_single_page(). But, there are some | |
6628 | * struct pages which are reserved in memblock allocator and their fields | |
6629 | * may be accessed (for example page_to_pfn() on some configuration accesses | |
6630 | * flags). We must explicitly zero those struct pages. | |
907ec5fc NH |
6631 | * |
6632 | * This function also addresses a similar issue where struct pages are left | |
6633 | * uninitialized because the physical address range is not covered by | |
6634 | * memblock.memory or memblock.reserved. That could happen when memblock | |
6635 | * layout is manually configured via memmap=. | |
a4a3ede2 | 6636 | */ |
03e85f9d | 6637 | void __init zero_resv_unavail(void) |
a4a3ede2 PT |
6638 | { |
6639 | phys_addr_t start, end; | |
a4a3ede2 | 6640 | u64 i, pgcnt; |
907ec5fc | 6641 | phys_addr_t next = 0; |
a4a3ede2 PT |
6642 | |
6643 | /* | |
907ec5fc | 6644 | * Loop through unavailable ranges not covered by memblock.memory. |
a4a3ede2 PT |
6645 | */ |
6646 | pgcnt = 0; | |
907ec5fc NH |
6647 | for_each_mem_range(i, &memblock.memory, NULL, |
6648 | NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) { | |
ec393a0f PT |
6649 | if (next < start) |
6650 | pgcnt += zero_pfn_range(PFN_DOWN(next), PFN_UP(start)); | |
907ec5fc NH |
6651 | next = end; |
6652 | } | |
ec393a0f | 6653 | pgcnt += zero_pfn_range(PFN_DOWN(next), max_pfn); |
907ec5fc | 6654 | |
a4a3ede2 PT |
6655 | /* |
6656 | * Struct pages that do not have backing memory. This could be because | |
6657 | * firmware is using some of this memory, or for some other reasons. | |
a4a3ede2 PT |
6658 | */ |
6659 | if (pgcnt) | |
907ec5fc | 6660 | pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt); |
a4a3ede2 | 6661 | } |
aca52c39 | 6662 | #endif /* !CONFIG_FLAT_NODE_MEM_MAP */ |
a4a3ede2 | 6663 | |
0ee332c1 | 6664 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
6665 | |
6666 | #if MAX_NUMNODES > 1 | |
6667 | /* | |
6668 | * Figure out the number of possible node ids. | |
6669 | */ | |
f9872caf | 6670 | void __init setup_nr_node_ids(void) |
418508c1 | 6671 | { |
904a9553 | 6672 | unsigned int highest; |
418508c1 | 6673 | |
904a9553 | 6674 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
6675 | nr_node_ids = highest + 1; |
6676 | } | |
418508c1 MS |
6677 | #endif |
6678 | ||
1e01979c TH |
6679 | /** |
6680 | * node_map_pfn_alignment - determine the maximum internode alignment | |
6681 | * | |
6682 | * This function should be called after node map is populated and sorted. | |
6683 | * It calculates the maximum power of two alignment which can distinguish | |
6684 | * all the nodes. | |
6685 | * | |
6686 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
6687 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
6688 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
6689 | * shifted, 1GiB is enough and this function will indicate so. | |
6690 | * | |
6691 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
6692 | * model has fine enough granularity to avoid incorrect mapping for the | |
6693 | * populated node map. | |
6694 | * | |
6695 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
6696 | * requirement (single node). | |
6697 | */ | |
6698 | unsigned long __init node_map_pfn_alignment(void) | |
6699 | { | |
6700 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 6701 | unsigned long start, end, mask; |
1e01979c | 6702 | int last_nid = -1; |
c13291a5 | 6703 | int i, nid; |
1e01979c | 6704 | |
c13291a5 | 6705 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
6706 | if (!start || last_nid < 0 || last_nid == nid) { |
6707 | last_nid = nid; | |
6708 | last_end = end; | |
6709 | continue; | |
6710 | } | |
6711 | ||
6712 | /* | |
6713 | * Start with a mask granular enough to pin-point to the | |
6714 | * start pfn and tick off bits one-by-one until it becomes | |
6715 | * too coarse to separate the current node from the last. | |
6716 | */ | |
6717 | mask = ~((1 << __ffs(start)) - 1); | |
6718 | while (mask && last_end <= (start & (mask << 1))) | |
6719 | mask <<= 1; | |
6720 | ||
6721 | /* accumulate all internode masks */ | |
6722 | accl_mask |= mask; | |
6723 | } | |
6724 | ||
6725 | /* convert mask to number of pages */ | |
6726 | return ~accl_mask + 1; | |
6727 | } | |
6728 | ||
a6af2bc3 | 6729 | /* Find the lowest pfn for a node */ |
b69a7288 | 6730 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 6731 | { |
a6af2bc3 | 6732 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
6733 | unsigned long start_pfn; |
6734 | int i; | |
1abbfb41 | 6735 | |
c13291a5 TH |
6736 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
6737 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 6738 | |
a6af2bc3 | 6739 | if (min_pfn == ULONG_MAX) { |
1170532b | 6740 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
6741 | return 0; |
6742 | } | |
6743 | ||
6744 | return min_pfn; | |
c713216d MG |
6745 | } |
6746 | ||
6747 | /** | |
6748 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
6749 | * | |
6750 | * It returns the minimum PFN based on information provided via | |
7d018176 | 6751 | * memblock_set_node(). |
c713216d MG |
6752 | */ |
6753 | unsigned long __init find_min_pfn_with_active_regions(void) | |
6754 | { | |
6755 | return find_min_pfn_for_node(MAX_NUMNODES); | |
6756 | } | |
6757 | ||
37b07e41 LS |
6758 | /* |
6759 | * early_calculate_totalpages() | |
6760 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 6761 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 6762 | */ |
484f51f8 | 6763 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 6764 | { |
7e63efef | 6765 | unsigned long totalpages = 0; |
c13291a5 TH |
6766 | unsigned long start_pfn, end_pfn; |
6767 | int i, nid; | |
6768 | ||
6769 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
6770 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 6771 | |
37b07e41 LS |
6772 | totalpages += pages; |
6773 | if (pages) | |
4b0ef1fe | 6774 | node_set_state(nid, N_MEMORY); |
37b07e41 | 6775 | } |
b8af2941 | 6776 | return totalpages; |
7e63efef MG |
6777 | } |
6778 | ||
2a1e274a MG |
6779 | /* |
6780 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
6781 | * is spread evenly between nodes as long as the nodes have enough | |
6782 | * memory. When they don't, some nodes will have more kernelcore than | |
6783 | * others | |
6784 | */ | |
b224ef85 | 6785 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
6786 | { |
6787 | int i, nid; | |
6788 | unsigned long usable_startpfn; | |
6789 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 6790 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 6791 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 6792 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 6793 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 6794 | struct memblock_region *r; |
b2f3eebe TC |
6795 | |
6796 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
6797 | find_usable_zone_for_movable(); | |
6798 | ||
6799 | /* | |
6800 | * If movable_node is specified, ignore kernelcore and movablecore | |
6801 | * options. | |
6802 | */ | |
6803 | if (movable_node_is_enabled()) { | |
136199f0 EM |
6804 | for_each_memblock(memory, r) { |
6805 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
6806 | continue; |
6807 | ||
136199f0 | 6808 | nid = r->nid; |
b2f3eebe | 6809 | |
136199f0 | 6810 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
6811 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6812 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6813 | usable_startpfn; | |
6814 | } | |
6815 | ||
6816 | goto out2; | |
6817 | } | |
2a1e274a | 6818 | |
342332e6 TI |
6819 | /* |
6820 | * If kernelcore=mirror is specified, ignore movablecore option | |
6821 | */ | |
6822 | if (mirrored_kernelcore) { | |
6823 | bool mem_below_4gb_not_mirrored = false; | |
6824 | ||
6825 | for_each_memblock(memory, r) { | |
6826 | if (memblock_is_mirror(r)) | |
6827 | continue; | |
6828 | ||
6829 | nid = r->nid; | |
6830 | ||
6831 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
6832 | ||
6833 | if (usable_startpfn < 0x100000) { | |
6834 | mem_below_4gb_not_mirrored = true; | |
6835 | continue; | |
6836 | } | |
6837 | ||
6838 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
6839 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
6840 | usable_startpfn; | |
6841 | } | |
6842 | ||
6843 | if (mem_below_4gb_not_mirrored) | |
6844 | pr_warn("This configuration results in unmirrored kernel memory."); | |
6845 | ||
6846 | goto out2; | |
6847 | } | |
6848 | ||
7e63efef | 6849 | /* |
a5c6d650 DR |
6850 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
6851 | * amount of necessary memory. | |
6852 | */ | |
6853 | if (required_kernelcore_percent) | |
6854 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
6855 | 10000UL; | |
6856 | if (required_movablecore_percent) | |
6857 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
6858 | 10000UL; | |
6859 | ||
6860 | /* | |
6861 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
6862 | * kernelcore that corresponds so that memory usable for |
6863 | * any allocation type is evenly spread. If both kernelcore | |
6864 | * and movablecore are specified, then the value of kernelcore | |
6865 | * will be used for required_kernelcore if it's greater than | |
6866 | * what movablecore would have allowed. | |
6867 | */ | |
6868 | if (required_movablecore) { | |
7e63efef MG |
6869 | unsigned long corepages; |
6870 | ||
6871 | /* | |
6872 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
6873 | * was requested by the user | |
6874 | */ | |
6875 | required_movablecore = | |
6876 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 6877 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
6878 | corepages = totalpages - required_movablecore; |
6879 | ||
6880 | required_kernelcore = max(required_kernelcore, corepages); | |
6881 | } | |
6882 | ||
bde304bd XQ |
6883 | /* |
6884 | * If kernelcore was not specified or kernelcore size is larger | |
6885 | * than totalpages, there is no ZONE_MOVABLE. | |
6886 | */ | |
6887 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 6888 | goto out; |
2a1e274a MG |
6889 | |
6890 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
6891 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6892 | ||
6893 | restart: | |
6894 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
6895 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 6896 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
6897 | unsigned long start_pfn, end_pfn; |
6898 | ||
2a1e274a MG |
6899 | /* |
6900 | * Recalculate kernelcore_node if the division per node | |
6901 | * now exceeds what is necessary to satisfy the requested | |
6902 | * amount of memory for the kernel | |
6903 | */ | |
6904 | if (required_kernelcore < kernelcore_node) | |
6905 | kernelcore_node = required_kernelcore / usable_nodes; | |
6906 | ||
6907 | /* | |
6908 | * As the map is walked, we track how much memory is usable | |
6909 | * by the kernel using kernelcore_remaining. When it is | |
6910 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
6911 | */ | |
6912 | kernelcore_remaining = kernelcore_node; | |
6913 | ||
6914 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 6915 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
6916 | unsigned long size_pages; |
6917 | ||
c13291a5 | 6918 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
6919 | if (start_pfn >= end_pfn) |
6920 | continue; | |
6921 | ||
6922 | /* Account for what is only usable for kernelcore */ | |
6923 | if (start_pfn < usable_startpfn) { | |
6924 | unsigned long kernel_pages; | |
6925 | kernel_pages = min(end_pfn, usable_startpfn) | |
6926 | - start_pfn; | |
6927 | ||
6928 | kernelcore_remaining -= min(kernel_pages, | |
6929 | kernelcore_remaining); | |
6930 | required_kernelcore -= min(kernel_pages, | |
6931 | required_kernelcore); | |
6932 | ||
6933 | /* Continue if range is now fully accounted */ | |
6934 | if (end_pfn <= usable_startpfn) { | |
6935 | ||
6936 | /* | |
6937 | * Push zone_movable_pfn to the end so | |
6938 | * that if we have to rebalance | |
6939 | * kernelcore across nodes, we will | |
6940 | * not double account here | |
6941 | */ | |
6942 | zone_movable_pfn[nid] = end_pfn; | |
6943 | continue; | |
6944 | } | |
6945 | start_pfn = usable_startpfn; | |
6946 | } | |
6947 | ||
6948 | /* | |
6949 | * The usable PFN range for ZONE_MOVABLE is from | |
6950 | * start_pfn->end_pfn. Calculate size_pages as the | |
6951 | * number of pages used as kernelcore | |
6952 | */ | |
6953 | size_pages = end_pfn - start_pfn; | |
6954 | if (size_pages > kernelcore_remaining) | |
6955 | size_pages = kernelcore_remaining; | |
6956 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6957 | ||
6958 | /* | |
6959 | * Some kernelcore has been met, update counts and | |
6960 | * break if the kernelcore for this node has been | |
b8af2941 | 6961 | * satisfied |
2a1e274a MG |
6962 | */ |
6963 | required_kernelcore -= min(required_kernelcore, | |
6964 | size_pages); | |
6965 | kernelcore_remaining -= size_pages; | |
6966 | if (!kernelcore_remaining) | |
6967 | break; | |
6968 | } | |
6969 | } | |
6970 | ||
6971 | /* | |
6972 | * If there is still required_kernelcore, we do another pass with one | |
6973 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6974 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6975 | * satisfied |
2a1e274a MG |
6976 | */ |
6977 | usable_nodes--; | |
6978 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6979 | goto restart; | |
6980 | ||
b2f3eebe | 6981 | out2: |
2a1e274a MG |
6982 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6983 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6984 | zone_movable_pfn[nid] = | |
6985 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6986 | |
20e6926d | 6987 | out: |
66918dcd | 6988 | /* restore the node_state */ |
4b0ef1fe | 6989 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6990 | } |
6991 | ||
4b0ef1fe LJ |
6992 | /* Any regular or high memory on that node ? */ |
6993 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6994 | { |
37b07e41 LS |
6995 | enum zone_type zone_type; |
6996 | ||
4b0ef1fe | 6997 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 6998 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6999 | if (populated_zone(zone)) { |
7b0e0c0e OS |
7000 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
7001 | node_set_state(nid, N_HIGH_MEMORY); | |
7002 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 7003 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
7004 | break; |
7005 | } | |
37b07e41 | 7006 | } |
37b07e41 LS |
7007 | } |
7008 | ||
c713216d MG |
7009 | /** |
7010 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 7011 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
7012 | * |
7013 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 7014 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
7015 | * zone in each node and their holes is calculated. If the maximum PFN |
7016 | * between two adjacent zones match, it is assumed that the zone is empty. | |
7017 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
7018 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
7019 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
7020 | * at arch_max_dma_pfn. | |
7021 | */ | |
7022 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
7023 | { | |
c13291a5 TH |
7024 | unsigned long start_pfn, end_pfn; |
7025 | int i, nid; | |
a6af2bc3 | 7026 | |
c713216d MG |
7027 | /* Record where the zone boundaries are */ |
7028 | memset(arch_zone_lowest_possible_pfn, 0, | |
7029 | sizeof(arch_zone_lowest_possible_pfn)); | |
7030 | memset(arch_zone_highest_possible_pfn, 0, | |
7031 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
7032 | |
7033 | start_pfn = find_min_pfn_with_active_regions(); | |
7034 | ||
7035 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
7036 | if (i == ZONE_MOVABLE) |
7037 | continue; | |
90cae1fe OH |
7038 | |
7039 | end_pfn = max(max_zone_pfn[i], start_pfn); | |
7040 | arch_zone_lowest_possible_pfn[i] = start_pfn; | |
7041 | arch_zone_highest_possible_pfn[i] = end_pfn; | |
7042 | ||
7043 | start_pfn = end_pfn; | |
c713216d | 7044 | } |
2a1e274a MG |
7045 | |
7046 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
7047 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 7048 | find_zone_movable_pfns_for_nodes(); |
c713216d | 7049 | |
c713216d | 7050 | /* Print out the zone ranges */ |
f88dfff5 | 7051 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
7052 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7053 | if (i == ZONE_MOVABLE) | |
7054 | continue; | |
f88dfff5 | 7055 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
7056 | if (arch_zone_lowest_possible_pfn[i] == |
7057 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 7058 | pr_cont("empty\n"); |
72f0ba02 | 7059 | else |
8d29e18a JG |
7060 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
7061 | (u64)arch_zone_lowest_possible_pfn[i] | |
7062 | << PAGE_SHIFT, | |
7063 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 7064 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
7065 | } |
7066 | ||
7067 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 7068 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
7069 | for (i = 0; i < MAX_NUMNODES; i++) { |
7070 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
7071 | pr_info(" Node %d: %#018Lx\n", i, |
7072 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 7073 | } |
c713216d | 7074 | |
f2d52fe5 | 7075 | /* Print out the early node map */ |
f88dfff5 | 7076 | pr_info("Early memory node ranges\n"); |
c13291a5 | 7077 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
7078 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
7079 | (u64)start_pfn << PAGE_SHIFT, | |
7080 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
7081 | |
7082 | /* Initialise every node */ | |
708614e6 | 7083 | mminit_verify_pageflags_layout(); |
8ef82866 | 7084 | setup_nr_node_ids(); |
e181ae0c | 7085 | zero_resv_unavail(); |
c713216d MG |
7086 | for_each_online_node(nid) { |
7087 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 7088 | free_area_init_node(nid, NULL, |
c713216d | 7089 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
7090 | |
7091 | /* Any memory on that node */ | |
7092 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
7093 | node_set_state(nid, N_MEMORY); |
7094 | check_for_memory(pgdat, nid); | |
c713216d MG |
7095 | } |
7096 | } | |
2a1e274a | 7097 | |
a5c6d650 DR |
7098 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
7099 | unsigned long *percent) | |
2a1e274a MG |
7100 | { |
7101 | unsigned long long coremem; | |
a5c6d650 DR |
7102 | char *endptr; |
7103 | ||
2a1e274a MG |
7104 | if (!p) |
7105 | return -EINVAL; | |
7106 | ||
a5c6d650 DR |
7107 | /* Value may be a percentage of total memory, otherwise bytes */ |
7108 | coremem = simple_strtoull(p, &endptr, 0); | |
7109 | if (*endptr == '%') { | |
7110 | /* Paranoid check for percent values greater than 100 */ | |
7111 | WARN_ON(coremem > 100); | |
2a1e274a | 7112 | |
a5c6d650 DR |
7113 | *percent = coremem; |
7114 | } else { | |
7115 | coremem = memparse(p, &p); | |
7116 | /* Paranoid check that UL is enough for the coremem value */ | |
7117 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7118 | |
a5c6d650 DR |
7119 | *core = coremem >> PAGE_SHIFT; |
7120 | *percent = 0UL; | |
7121 | } | |
2a1e274a MG |
7122 | return 0; |
7123 | } | |
ed7ed365 | 7124 | |
7e63efef MG |
7125 | /* |
7126 | * kernelcore=size sets the amount of memory for use for allocations that | |
7127 | * cannot be reclaimed or migrated. | |
7128 | */ | |
7129 | static int __init cmdline_parse_kernelcore(char *p) | |
7130 | { | |
342332e6 TI |
7131 | /* parse kernelcore=mirror */ |
7132 | if (parse_option_str(p, "mirror")) { | |
7133 | mirrored_kernelcore = true; | |
7134 | return 0; | |
7135 | } | |
7136 | ||
a5c6d650 DR |
7137 | return cmdline_parse_core(p, &required_kernelcore, |
7138 | &required_kernelcore_percent); | |
7e63efef MG |
7139 | } |
7140 | ||
7141 | /* | |
7142 | * movablecore=size sets the amount of memory for use for allocations that | |
7143 | * can be reclaimed or migrated. | |
7144 | */ | |
7145 | static int __init cmdline_parse_movablecore(char *p) | |
7146 | { | |
a5c6d650 DR |
7147 | return cmdline_parse_core(p, &required_movablecore, |
7148 | &required_movablecore_percent); | |
7e63efef MG |
7149 | } |
7150 | ||
ed7ed365 | 7151 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7152 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7153 | |
0ee332c1 | 7154 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 7155 | |
c3d5f5f0 JL |
7156 | void adjust_managed_page_count(struct page *page, long count) |
7157 | { | |
9705bea5 | 7158 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 7159 | totalram_pages_add(count); |
3dcc0571 JL |
7160 | #ifdef CONFIG_HIGHMEM |
7161 | if (PageHighMem(page)) | |
ca79b0c2 | 7162 | totalhigh_pages_add(count); |
3dcc0571 | 7163 | #endif |
c3d5f5f0 | 7164 | } |
3dcc0571 | 7165 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7166 | |
11199692 | 7167 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 7168 | { |
11199692 JL |
7169 | void *pos; |
7170 | unsigned long pages = 0; | |
69afade7 | 7171 | |
11199692 JL |
7172 | start = (void *)PAGE_ALIGN((unsigned long)start); |
7173 | end = (void *)((unsigned long)end & PAGE_MASK); | |
7174 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
7175 | struct page *page = virt_to_page(pos); |
7176 | void *direct_map_addr; | |
7177 | ||
7178 | /* | |
7179 | * 'direct_map_addr' might be different from 'pos' | |
7180 | * because some architectures' virt_to_page() | |
7181 | * work with aliases. Getting the direct map | |
7182 | * address ensures that we get a _writeable_ | |
7183 | * alias for the memset(). | |
7184 | */ | |
7185 | direct_map_addr = page_address(page); | |
dbe67df4 | 7186 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
7187 | memset(direct_map_addr, poison, PAGE_SIZE); |
7188 | ||
7189 | free_reserved_page(page); | |
69afade7 JL |
7190 | } |
7191 | ||
7192 | if (pages && s) | |
adb1fe9a JP |
7193 | pr_info("Freeing %s memory: %ldK\n", |
7194 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
7195 | |
7196 | return pages; | |
7197 | } | |
11199692 | 7198 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 7199 | |
cfa11e08 JL |
7200 | #ifdef CONFIG_HIGHMEM |
7201 | void free_highmem_page(struct page *page) | |
7202 | { | |
7203 | __free_reserved_page(page); | |
ca79b0c2 | 7204 | totalram_pages_inc(); |
9705bea5 | 7205 | atomic_long_inc(&page_zone(page)->managed_pages); |
ca79b0c2 | 7206 | totalhigh_pages_inc(); |
cfa11e08 JL |
7207 | } |
7208 | #endif | |
7209 | ||
7ee3d4e8 JL |
7210 | |
7211 | void __init mem_init_print_info(const char *str) | |
7212 | { | |
7213 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7214 | unsigned long init_code_size, init_data_size; | |
7215 | ||
7216 | physpages = get_num_physpages(); | |
7217 | codesize = _etext - _stext; | |
7218 | datasize = _edata - _sdata; | |
7219 | rosize = __end_rodata - __start_rodata; | |
7220 | bss_size = __bss_stop - __bss_start; | |
7221 | init_data_size = __init_end - __init_begin; | |
7222 | init_code_size = _einittext - _sinittext; | |
7223 | ||
7224 | /* | |
7225 | * Detect special cases and adjust section sizes accordingly: | |
7226 | * 1) .init.* may be embedded into .data sections | |
7227 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7228 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7229 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7230 | */ | |
7231 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7232 | do { \ |
7233 | if (start <= pos && pos < end && size > adj) \ | |
7234 | size -= adj; \ | |
7235 | } while (0) | |
7ee3d4e8 JL |
7236 | |
7237 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7238 | _sinittext, init_code_size); | |
7239 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7240 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7241 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7242 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7243 | ||
7244 | #undef adj_init_size | |
7245 | ||
756a025f | 7246 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7247 | #ifdef CONFIG_HIGHMEM |
756a025f | 7248 | ", %luK highmem" |
7ee3d4e8 | 7249 | #endif |
756a025f JP |
7250 | "%s%s)\n", |
7251 | nr_free_pages() << (PAGE_SHIFT - 10), | |
7252 | physpages << (PAGE_SHIFT - 10), | |
7253 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7254 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
ca79b0c2 | 7255 | (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10), |
756a025f | 7256 | totalcma_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7257 | #ifdef CONFIG_HIGHMEM |
ca79b0c2 | 7258 | totalhigh_pages() << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7259 | #endif |
756a025f | 7260 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
7261 | } |
7262 | ||
0e0b864e | 7263 | /** |
88ca3b94 RD |
7264 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7265 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7266 | * |
013110a7 | 7267 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7268 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7269 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7270 | * function may optionally be used to account for unfreeable pages in the |
7271 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7272 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7273 | */ |
7274 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7275 | { | |
7276 | dma_reserve = new_dma_reserve; | |
7277 | } | |
7278 | ||
1da177e4 LT |
7279 | void __init free_area_init(unsigned long *zones_size) |
7280 | { | |
e181ae0c | 7281 | zero_resv_unavail(); |
9109fb7b | 7282 | free_area_init_node(0, zones_size, |
1da177e4 LT |
7283 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
7284 | } | |
1da177e4 | 7285 | |
005fd4bb | 7286 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7287 | { |
1da177e4 | 7288 | |
005fd4bb SAS |
7289 | lru_add_drain_cpu(cpu); |
7290 | drain_pages(cpu); | |
9f8f2172 | 7291 | |
005fd4bb SAS |
7292 | /* |
7293 | * Spill the event counters of the dead processor | |
7294 | * into the current processors event counters. | |
7295 | * This artificially elevates the count of the current | |
7296 | * processor. | |
7297 | */ | |
7298 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7299 | |
005fd4bb SAS |
7300 | /* |
7301 | * Zero the differential counters of the dead processor | |
7302 | * so that the vm statistics are consistent. | |
7303 | * | |
7304 | * This is only okay since the processor is dead and cannot | |
7305 | * race with what we are doing. | |
7306 | */ | |
7307 | cpu_vm_stats_fold(cpu); | |
7308 | return 0; | |
1da177e4 | 7309 | } |
1da177e4 LT |
7310 | |
7311 | void __init page_alloc_init(void) | |
7312 | { | |
005fd4bb SAS |
7313 | int ret; |
7314 | ||
7315 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, | |
7316 | "mm/page_alloc:dead", NULL, | |
7317 | page_alloc_cpu_dead); | |
7318 | WARN_ON(ret < 0); | |
1da177e4 LT |
7319 | } |
7320 | ||
cb45b0e9 | 7321 | /* |
34b10060 | 7322 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
7323 | * or min_free_kbytes changes. |
7324 | */ | |
7325 | static void calculate_totalreserve_pages(void) | |
7326 | { | |
7327 | struct pglist_data *pgdat; | |
7328 | unsigned long reserve_pages = 0; | |
2f6726e5 | 7329 | enum zone_type i, j; |
cb45b0e9 HA |
7330 | |
7331 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
7332 | |
7333 | pgdat->totalreserve_pages = 0; | |
7334 | ||
cb45b0e9 HA |
7335 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7336 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 7337 | long max = 0; |
9705bea5 | 7338 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
7339 | |
7340 | /* Find valid and maximum lowmem_reserve in the zone */ | |
7341 | for (j = i; j < MAX_NR_ZONES; j++) { | |
7342 | if (zone->lowmem_reserve[j] > max) | |
7343 | max = zone->lowmem_reserve[j]; | |
7344 | } | |
7345 | ||
41858966 MG |
7346 | /* we treat the high watermark as reserved pages. */ |
7347 | max += high_wmark_pages(zone); | |
cb45b0e9 | 7348 | |
3d6357de AK |
7349 | if (max > managed_pages) |
7350 | max = managed_pages; | |
a8d01437 | 7351 | |
281e3726 | 7352 | pgdat->totalreserve_pages += max; |
a8d01437 | 7353 | |
cb45b0e9 HA |
7354 | reserve_pages += max; |
7355 | } | |
7356 | } | |
7357 | totalreserve_pages = reserve_pages; | |
7358 | } | |
7359 | ||
1da177e4 LT |
7360 | /* |
7361 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 7362 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
7363 | * has a correct pages reserved value, so an adequate number of |
7364 | * pages are left in the zone after a successful __alloc_pages(). | |
7365 | */ | |
7366 | static void setup_per_zone_lowmem_reserve(void) | |
7367 | { | |
7368 | struct pglist_data *pgdat; | |
2f6726e5 | 7369 | enum zone_type j, idx; |
1da177e4 | 7370 | |
ec936fc5 | 7371 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
7372 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7373 | struct zone *zone = pgdat->node_zones + j; | |
9705bea5 | 7374 | unsigned long managed_pages = zone_managed_pages(zone); |
1da177e4 LT |
7375 | |
7376 | zone->lowmem_reserve[j] = 0; | |
7377 | ||
2f6726e5 CL |
7378 | idx = j; |
7379 | while (idx) { | |
1da177e4 LT |
7380 | struct zone *lower_zone; |
7381 | ||
2f6726e5 | 7382 | idx--; |
1da177e4 | 7383 | lower_zone = pgdat->node_zones + idx; |
d3cda233 JK |
7384 | |
7385 | if (sysctl_lowmem_reserve_ratio[idx] < 1) { | |
7386 | sysctl_lowmem_reserve_ratio[idx] = 0; | |
7387 | lower_zone->lowmem_reserve[j] = 0; | |
7388 | } else { | |
7389 | lower_zone->lowmem_reserve[j] = | |
7390 | managed_pages / sysctl_lowmem_reserve_ratio[idx]; | |
7391 | } | |
9705bea5 | 7392 | managed_pages += zone_managed_pages(lower_zone); |
1da177e4 LT |
7393 | } |
7394 | } | |
7395 | } | |
cb45b0e9 HA |
7396 | |
7397 | /* update totalreserve_pages */ | |
7398 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7399 | } |
7400 | ||
cfd3da1e | 7401 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
7402 | { |
7403 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
7404 | unsigned long lowmem_pages = 0; | |
7405 | struct zone *zone; | |
7406 | unsigned long flags; | |
7407 | ||
7408 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
7409 | for_each_zone(zone) { | |
7410 | if (!is_highmem(zone)) | |
9705bea5 | 7411 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
7412 | } |
7413 | ||
7414 | for_each_zone(zone) { | |
ac924c60 AM |
7415 | u64 tmp; |
7416 | ||
1125b4e3 | 7417 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 7418 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 7419 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
7420 | if (is_highmem(zone)) { |
7421 | /* | |
669ed175 NP |
7422 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
7423 | * need highmem pages, so cap pages_min to a small | |
7424 | * value here. | |
7425 | * | |
41858966 | 7426 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 7427 | * deltas control asynch page reclaim, and so should |
669ed175 | 7428 | * not be capped for highmem. |
1da177e4 | 7429 | */ |
90ae8d67 | 7430 | unsigned long min_pages; |
1da177e4 | 7431 | |
9705bea5 | 7432 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 7433 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 7434 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 7435 | } else { |
669ed175 NP |
7436 | /* |
7437 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
7438 | * proportionate to the zone's size. |
7439 | */ | |
41858966 | 7440 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
7441 | } |
7442 | ||
795ae7a0 JW |
7443 | /* |
7444 | * Set the kswapd watermarks distance according to the | |
7445 | * scale factor in proportion to available memory, but | |
7446 | * ensure a minimum size on small systems. | |
7447 | */ | |
7448 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 7449 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
7450 | watermark_scale_factor, 10000)); |
7451 | ||
7452 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
7453 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 7454 | |
1125b4e3 | 7455 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 7456 | } |
cb45b0e9 HA |
7457 | |
7458 | /* update totalreserve_pages */ | |
7459 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7460 | } |
7461 | ||
cfd3da1e MG |
7462 | /** |
7463 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
7464 | * or when memory is hot-{added|removed} | |
7465 | * | |
7466 | * Ensures that the watermark[min,low,high] values for each zone are set | |
7467 | * correctly with respect to min_free_kbytes. | |
7468 | */ | |
7469 | void setup_per_zone_wmarks(void) | |
7470 | { | |
b93e0f32 MH |
7471 | static DEFINE_SPINLOCK(lock); |
7472 | ||
7473 | spin_lock(&lock); | |
cfd3da1e | 7474 | __setup_per_zone_wmarks(); |
b93e0f32 | 7475 | spin_unlock(&lock); |
cfd3da1e MG |
7476 | } |
7477 | ||
1da177e4 LT |
7478 | /* |
7479 | * Initialise min_free_kbytes. | |
7480 | * | |
7481 | * For small machines we want it small (128k min). For large machines | |
7482 | * we want it large (64MB max). But it is not linear, because network | |
7483 | * bandwidth does not increase linearly with machine size. We use | |
7484 | * | |
b8af2941 | 7485 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
7486 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
7487 | * | |
7488 | * which yields | |
7489 | * | |
7490 | * 16MB: 512k | |
7491 | * 32MB: 724k | |
7492 | * 64MB: 1024k | |
7493 | * 128MB: 1448k | |
7494 | * 256MB: 2048k | |
7495 | * 512MB: 2896k | |
7496 | * 1024MB: 4096k | |
7497 | * 2048MB: 5792k | |
7498 | * 4096MB: 8192k | |
7499 | * 8192MB: 11584k | |
7500 | * 16384MB: 16384k | |
7501 | */ | |
1b79acc9 | 7502 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
7503 | { |
7504 | unsigned long lowmem_kbytes; | |
5f12733e | 7505 | int new_min_free_kbytes; |
1da177e4 LT |
7506 | |
7507 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
7508 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
7509 | ||
7510 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
7511 | min_free_kbytes = new_min_free_kbytes; | |
7512 | if (min_free_kbytes < 128) | |
7513 | min_free_kbytes = 128; | |
7514 | if (min_free_kbytes > 65536) | |
7515 | min_free_kbytes = 65536; | |
7516 | } else { | |
7517 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
7518 | new_min_free_kbytes, user_min_free_kbytes); | |
7519 | } | |
bc75d33f | 7520 | setup_per_zone_wmarks(); |
a6cccdc3 | 7521 | refresh_zone_stat_thresholds(); |
1da177e4 | 7522 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
7523 | |
7524 | #ifdef CONFIG_NUMA | |
7525 | setup_min_unmapped_ratio(); | |
7526 | setup_min_slab_ratio(); | |
7527 | #endif | |
7528 | ||
1da177e4 LT |
7529 | return 0; |
7530 | } | |
bc22af74 | 7531 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
7532 | |
7533 | /* | |
b8af2941 | 7534 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
7535 | * that we can call two helper functions whenever min_free_kbytes |
7536 | * changes. | |
7537 | */ | |
cccad5b9 | 7538 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7539 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7540 | { |
da8c757b HP |
7541 | int rc; |
7542 | ||
7543 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7544 | if (rc) | |
7545 | return rc; | |
7546 | ||
5f12733e MH |
7547 | if (write) { |
7548 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7549 | setup_per_zone_wmarks(); |
5f12733e | 7550 | } |
1da177e4 LT |
7551 | return 0; |
7552 | } | |
7553 | ||
795ae7a0 JW |
7554 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
7555 | void __user *buffer, size_t *length, loff_t *ppos) | |
7556 | { | |
7557 | int rc; | |
7558 | ||
7559 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7560 | if (rc) | |
7561 | return rc; | |
7562 | ||
7563 | if (write) | |
7564 | setup_per_zone_wmarks(); | |
7565 | ||
7566 | return 0; | |
7567 | } | |
7568 | ||
9614634f | 7569 | #ifdef CONFIG_NUMA |
6423aa81 | 7570 | static void setup_min_unmapped_ratio(void) |
9614634f | 7571 | { |
6423aa81 | 7572 | pg_data_t *pgdat; |
9614634f | 7573 | struct zone *zone; |
9614634f | 7574 | |
a5f5f91d | 7575 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7576 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7577 | |
9614634f | 7578 | for_each_zone(zone) |
9705bea5 AK |
7579 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
7580 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 7581 | } |
0ff38490 | 7582 | |
6423aa81 JK |
7583 | |
7584 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 7585 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7586 | { |
0ff38490 CL |
7587 | int rc; |
7588 | ||
8d65af78 | 7589 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7590 | if (rc) |
7591 | return rc; | |
7592 | ||
6423aa81 JK |
7593 | setup_min_unmapped_ratio(); |
7594 | ||
7595 | return 0; | |
7596 | } | |
7597 | ||
7598 | static void setup_min_slab_ratio(void) | |
7599 | { | |
7600 | pg_data_t *pgdat; | |
7601 | struct zone *zone; | |
7602 | ||
a5f5f91d MG |
7603 | for_each_online_pgdat(pgdat) |
7604 | pgdat->min_slab_pages = 0; | |
7605 | ||
0ff38490 | 7606 | for_each_zone(zone) |
9705bea5 AK |
7607 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
7608 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
7609 | } |
7610 | ||
7611 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
7612 | void __user *buffer, size_t *length, loff_t *ppos) | |
7613 | { | |
7614 | int rc; | |
7615 | ||
7616 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7617 | if (rc) | |
7618 | return rc; | |
7619 | ||
7620 | setup_min_slab_ratio(); | |
7621 | ||
0ff38490 CL |
7622 | return 0; |
7623 | } | |
9614634f CL |
7624 | #endif |
7625 | ||
1da177e4 LT |
7626 | /* |
7627 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7628 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7629 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7630 | * | |
7631 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7632 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7633 | * if in function of the boot time zone sizes. |
7634 | */ | |
cccad5b9 | 7635 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7636 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7637 | { |
8d65af78 | 7638 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
7639 | setup_per_zone_lowmem_reserve(); |
7640 | return 0; | |
7641 | } | |
7642 | ||
8ad4b1fb RS |
7643 | /* |
7644 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
7645 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
7646 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 7647 | */ |
cccad5b9 | 7648 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7649 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
7650 | { |
7651 | struct zone *zone; | |
7cd2b0a3 | 7652 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
7653 | int ret; |
7654 | ||
7cd2b0a3 DR |
7655 | mutex_lock(&pcp_batch_high_lock); |
7656 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
7657 | ||
8d65af78 | 7658 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
7659 | if (!write || ret < 0) |
7660 | goto out; | |
7661 | ||
7662 | /* Sanity checking to avoid pcp imbalance */ | |
7663 | if (percpu_pagelist_fraction && | |
7664 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
7665 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
7666 | ret = -EINVAL; | |
7667 | goto out; | |
7668 | } | |
7669 | ||
7670 | /* No change? */ | |
7671 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
7672 | goto out; | |
c8e251fa | 7673 | |
364df0eb | 7674 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
7675 | unsigned int cpu; |
7676 | ||
22a7f12b | 7677 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
7678 | pageset_set_high_and_batch(zone, |
7679 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 7680 | } |
7cd2b0a3 | 7681 | out: |
c8e251fa | 7682 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 7683 | return ret; |
8ad4b1fb RS |
7684 | } |
7685 | ||
a9919c79 | 7686 | #ifdef CONFIG_NUMA |
f034b5d4 | 7687 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 7688 | |
1da177e4 LT |
7689 | static int __init set_hashdist(char *str) |
7690 | { | |
7691 | if (!str) | |
7692 | return 0; | |
7693 | hashdist = simple_strtoul(str, &str, 0); | |
7694 | return 1; | |
7695 | } | |
7696 | __setup("hashdist=", set_hashdist); | |
7697 | #endif | |
7698 | ||
f6f34b43 SD |
7699 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
7700 | /* | |
7701 | * Returns the number of pages that arch has reserved but | |
7702 | * is not known to alloc_large_system_hash(). | |
7703 | */ | |
7704 | static unsigned long __init arch_reserved_kernel_pages(void) | |
7705 | { | |
7706 | return 0; | |
7707 | } | |
7708 | #endif | |
7709 | ||
9017217b PT |
7710 | /* |
7711 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
7712 | * machines. As memory size is increased the scale is also increased but at | |
7713 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
7714 | * quadruples the scale is increased by one, which means the size of hash table | |
7715 | * only doubles, instead of quadrupling as well. | |
7716 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
7717 | * makes sense, it is disabled on such platforms. | |
7718 | */ | |
7719 | #if __BITS_PER_LONG > 32 | |
7720 | #define ADAPT_SCALE_BASE (64ul << 30) | |
7721 | #define ADAPT_SCALE_SHIFT 2 | |
7722 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
7723 | #endif | |
7724 | ||
1da177e4 LT |
7725 | /* |
7726 | * allocate a large system hash table from bootmem | |
7727 | * - it is assumed that the hash table must contain an exact power-of-2 | |
7728 | * quantity of entries | |
7729 | * - limit is the number of hash buckets, not the total allocation size | |
7730 | */ | |
7731 | void *__init alloc_large_system_hash(const char *tablename, | |
7732 | unsigned long bucketsize, | |
7733 | unsigned long numentries, | |
7734 | int scale, | |
7735 | int flags, | |
7736 | unsigned int *_hash_shift, | |
7737 | unsigned int *_hash_mask, | |
31fe62b9 TB |
7738 | unsigned long low_limit, |
7739 | unsigned long high_limit) | |
1da177e4 | 7740 | { |
31fe62b9 | 7741 | unsigned long long max = high_limit; |
1da177e4 LT |
7742 | unsigned long log2qty, size; |
7743 | void *table = NULL; | |
3749a8f0 | 7744 | gfp_t gfp_flags; |
1da177e4 LT |
7745 | |
7746 | /* allow the kernel cmdline to have a say */ | |
7747 | if (!numentries) { | |
7748 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 7749 | numentries = nr_kernel_pages; |
f6f34b43 | 7750 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
7751 | |
7752 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
7753 | if (PAGE_SHIFT < 20) | |
7754 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 7755 | |
9017217b PT |
7756 | #if __BITS_PER_LONG > 32 |
7757 | if (!high_limit) { | |
7758 | unsigned long adapt; | |
7759 | ||
7760 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
7761 | adapt <<= ADAPT_SCALE_SHIFT) | |
7762 | scale++; | |
7763 | } | |
7764 | #endif | |
7765 | ||
1da177e4 LT |
7766 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
7767 | if (scale > PAGE_SHIFT) | |
7768 | numentries >>= (scale - PAGE_SHIFT); | |
7769 | else | |
7770 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
7771 | |
7772 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
7773 | if (unlikely(flags & HASH_SMALL)) { |
7774 | /* Makes no sense without HASH_EARLY */ | |
7775 | WARN_ON(!(flags & HASH_EARLY)); | |
7776 | if (!(numentries >> *_hash_shift)) { | |
7777 | numentries = 1UL << *_hash_shift; | |
7778 | BUG_ON(!numentries); | |
7779 | } | |
7780 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 7781 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 7782 | } |
6e692ed3 | 7783 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
7784 | |
7785 | /* limit allocation size to 1/16 total memory by default */ | |
7786 | if (max == 0) { | |
7787 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
7788 | do_div(max, bucketsize); | |
7789 | } | |
074b8517 | 7790 | max = min(max, 0x80000000ULL); |
1da177e4 | 7791 | |
31fe62b9 TB |
7792 | if (numentries < low_limit) |
7793 | numentries = low_limit; | |
1da177e4 LT |
7794 | if (numentries > max) |
7795 | numentries = max; | |
7796 | ||
f0d1b0b3 | 7797 | log2qty = ilog2(numentries); |
1da177e4 | 7798 | |
3749a8f0 | 7799 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 LT |
7800 | do { |
7801 | size = bucketsize << log2qty; | |
ea1f5f37 PT |
7802 | if (flags & HASH_EARLY) { |
7803 | if (flags & HASH_ZERO) | |
7e1c4e27 MR |
7804 | table = memblock_alloc_nopanic(size, |
7805 | SMP_CACHE_BYTES); | |
ea1f5f37 | 7806 | else |
7e1c4e27 MR |
7807 | table = memblock_alloc_raw(size, |
7808 | SMP_CACHE_BYTES); | |
ea1f5f37 | 7809 | } else if (hashdist) { |
3749a8f0 | 7810 | table = __vmalloc(size, gfp_flags, PAGE_KERNEL); |
ea1f5f37 | 7811 | } else { |
1037b83b ED |
7812 | /* |
7813 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
7814 | * some pages at the end of hash table which |
7815 | * alloc_pages_exact() automatically does | |
1037b83b | 7816 | */ |
264ef8a9 | 7817 | if (get_order(size) < MAX_ORDER) { |
3749a8f0 PT |
7818 | table = alloc_pages_exact(size, gfp_flags); |
7819 | kmemleak_alloc(table, size, 1, gfp_flags); | |
264ef8a9 | 7820 | } |
1da177e4 LT |
7821 | } |
7822 | } while (!table && size > PAGE_SIZE && --log2qty); | |
7823 | ||
7824 | if (!table) | |
7825 | panic("Failed to allocate %s hash table\n", tablename); | |
7826 | ||
1170532b JP |
7827 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
7828 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
7829 | |
7830 | if (_hash_shift) | |
7831 | *_hash_shift = log2qty; | |
7832 | if (_hash_mask) | |
7833 | *_hash_mask = (1 << log2qty) - 1; | |
7834 | ||
7835 | return table; | |
7836 | } | |
a117e66e | 7837 | |
a5d76b54 | 7838 | /* |
80934513 MK |
7839 | * This function checks whether pageblock includes unmovable pages or not. |
7840 | * If @count is not zero, it is okay to include less @count unmovable pages | |
7841 | * | |
b8af2941 | 7842 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
7843 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
7844 | * check without lock_page also may miss some movable non-lru pages at | |
7845 | * race condition. So you can't expect this function should be exact. | |
a5d76b54 | 7846 | */ |
b023f468 | 7847 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
d381c547 | 7848 | int migratetype, int flags) |
49ac8255 KH |
7849 | { |
7850 | unsigned long pfn, iter, found; | |
47118af0 | 7851 | |
49ac8255 | 7852 | /* |
15c30bc0 MH |
7853 | * TODO we could make this much more efficient by not checking every |
7854 | * page in the range if we know all of them are in MOVABLE_ZONE and | |
7855 | * that the movable zone guarantees that pages are migratable but | |
7856 | * the later is not the case right now unfortunatelly. E.g. movablecore | |
7857 | * can still lead to having bootmem allocations in zone_movable. | |
49ac8255 | 7858 | */ |
49ac8255 | 7859 | |
4da2ce25 MH |
7860 | /* |
7861 | * CMA allocations (alloc_contig_range) really need to mark isolate | |
7862 | * CMA pageblocks even when they are not movable in fact so consider | |
7863 | * them movable here. | |
7864 | */ | |
7865 | if (is_migrate_cma(migratetype) && | |
7866 | is_migrate_cma(get_pageblock_migratetype(page))) | |
7867 | return false; | |
7868 | ||
49ac8255 KH |
7869 | pfn = page_to_pfn(page); |
7870 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
7871 | unsigned long check = pfn + iter; | |
7872 | ||
29723fcc | 7873 | if (!pfn_valid_within(check)) |
49ac8255 | 7874 | continue; |
29723fcc | 7875 | |
49ac8255 | 7876 | page = pfn_to_page(check); |
c8721bbb | 7877 | |
d7ab3672 | 7878 | if (PageReserved(page)) |
15c30bc0 | 7879 | goto unmovable; |
d7ab3672 | 7880 | |
9d789999 MH |
7881 | /* |
7882 | * If the zone is movable and we have ruled out all reserved | |
7883 | * pages then it should be reasonably safe to assume the rest | |
7884 | * is movable. | |
7885 | */ | |
7886 | if (zone_idx(zone) == ZONE_MOVABLE) | |
7887 | continue; | |
7888 | ||
c8721bbb NH |
7889 | /* |
7890 | * Hugepages are not in LRU lists, but they're movable. | |
7891 | * We need not scan over tail pages bacause we don't | |
7892 | * handle each tail page individually in migration. | |
7893 | */ | |
7894 | if (PageHuge(page)) { | |
17e2e7d7 OS |
7895 | struct page *head = compound_head(page); |
7896 | unsigned int skip_pages; | |
464c7ffb | 7897 | |
17e2e7d7 | 7898 | if (!hugepage_migration_supported(page_hstate(head))) |
464c7ffb AK |
7899 | goto unmovable; |
7900 | ||
17e2e7d7 OS |
7901 | skip_pages = (1 << compound_order(head)) - (page - head); |
7902 | iter += skip_pages - 1; | |
c8721bbb NH |
7903 | continue; |
7904 | } | |
7905 | ||
97d255c8 MK |
7906 | /* |
7907 | * We can't use page_count without pin a page | |
7908 | * because another CPU can free compound page. | |
7909 | * This check already skips compound tails of THP | |
0139aa7b | 7910 | * because their page->_refcount is zero at all time. |
97d255c8 | 7911 | */ |
fe896d18 | 7912 | if (!page_ref_count(page)) { |
49ac8255 KH |
7913 | if (PageBuddy(page)) |
7914 | iter += (1 << page_order(page)) - 1; | |
7915 | continue; | |
7916 | } | |
97d255c8 | 7917 | |
b023f468 WC |
7918 | /* |
7919 | * The HWPoisoned page may be not in buddy system, and | |
7920 | * page_count() is not 0. | |
7921 | */ | |
d381c547 | 7922 | if ((flags & SKIP_HWPOISON) && PageHWPoison(page)) |
b023f468 WC |
7923 | continue; |
7924 | ||
0efadf48 YX |
7925 | if (__PageMovable(page)) |
7926 | continue; | |
7927 | ||
49ac8255 KH |
7928 | if (!PageLRU(page)) |
7929 | found++; | |
7930 | /* | |
6b4f7799 JW |
7931 | * If there are RECLAIMABLE pages, we need to check |
7932 | * it. But now, memory offline itself doesn't call | |
7933 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
7934 | */ |
7935 | /* | |
7936 | * If the page is not RAM, page_count()should be 0. | |
7937 | * we don't need more check. This is an _used_ not-movable page. | |
7938 | * | |
7939 | * The problematic thing here is PG_reserved pages. PG_reserved | |
7940 | * is set to both of a memory hole page and a _used_ kernel | |
7941 | * page at boot. | |
7942 | */ | |
7943 | if (found > count) | |
15c30bc0 | 7944 | goto unmovable; |
49ac8255 | 7945 | } |
80934513 | 7946 | return false; |
15c30bc0 MH |
7947 | unmovable: |
7948 | WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE); | |
d381c547 MH |
7949 | if (flags & REPORT_FAILURE) |
7950 | dump_page(pfn_to_page(pfn+iter), "unmovable page"); | |
15c30bc0 | 7951 | return true; |
49ac8255 KH |
7952 | } |
7953 | ||
080fe206 | 7954 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7955 | |
7956 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7957 | { | |
7958 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7959 | pageblock_nr_pages) - 1); | |
7960 | } | |
7961 | ||
7962 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7963 | { | |
7964 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7965 | pageblock_nr_pages)); | |
7966 | } | |
7967 | ||
041d3a8c | 7968 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7969 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7970 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7971 | { |
7972 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7973 | unsigned long nr_reclaimed; |
041d3a8c MN |
7974 | unsigned long pfn = start; |
7975 | unsigned int tries = 0; | |
7976 | int ret = 0; | |
7977 | ||
be49a6e1 | 7978 | migrate_prep(); |
041d3a8c | 7979 | |
bb13ffeb | 7980 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7981 | if (fatal_signal_pending(current)) { |
7982 | ret = -EINTR; | |
7983 | break; | |
7984 | } | |
7985 | ||
bb13ffeb MG |
7986 | if (list_empty(&cc->migratepages)) { |
7987 | cc->nr_migratepages = 0; | |
edc2ca61 | 7988 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7989 | if (!pfn) { |
7990 | ret = -EINTR; | |
7991 | break; | |
7992 | } | |
7993 | tries = 0; | |
7994 | } else if (++tries == 5) { | |
7995 | ret = ret < 0 ? ret : -EBUSY; | |
7996 | break; | |
7997 | } | |
7998 | ||
beb51eaa MK |
7999 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
8000 | &cc->migratepages); | |
8001 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 8002 | |
9c620e2b | 8003 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
31025351 | 8004 | NULL, 0, cc->mode, MR_CONTIG_RANGE); |
041d3a8c | 8005 | } |
2a6f5124 SP |
8006 | if (ret < 0) { |
8007 | putback_movable_pages(&cc->migratepages); | |
8008 | return ret; | |
8009 | } | |
8010 | return 0; | |
041d3a8c MN |
8011 | } |
8012 | ||
8013 | /** | |
8014 | * alloc_contig_range() -- tries to allocate given range of pages | |
8015 | * @start: start PFN to allocate | |
8016 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
8017 | * @migratetype: migratetype of the underlaying pageblocks (either |
8018 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
8019 | * in range must have the same migratetype and it must | |
8020 | * be either of the two. | |
ca96b625 | 8021 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
8022 | * |
8023 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 8024 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 8025 | * |
2c7452a0 MK |
8026 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
8027 | * pageblocks in the range. Once isolated, the pageblocks should not | |
8028 | * be modified by others. | |
041d3a8c MN |
8029 | * |
8030 | * Returns zero on success or negative error code. On success all | |
8031 | * pages which PFN is in [start, end) are allocated for the caller and | |
8032 | * need to be freed with free_contig_range(). | |
8033 | */ | |
0815f3d8 | 8034 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 8035 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 8036 | { |
041d3a8c | 8037 | unsigned long outer_start, outer_end; |
d00181b9 KS |
8038 | unsigned int order; |
8039 | int ret = 0; | |
041d3a8c | 8040 | |
bb13ffeb MG |
8041 | struct compact_control cc = { |
8042 | .nr_migratepages = 0, | |
8043 | .order = -1, | |
8044 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 8045 | .mode = MIGRATE_SYNC, |
bb13ffeb | 8046 | .ignore_skip_hint = true, |
2583d671 | 8047 | .no_set_skip_hint = true, |
7dea19f9 | 8048 | .gfp_mask = current_gfp_context(gfp_mask), |
bb13ffeb MG |
8049 | }; |
8050 | INIT_LIST_HEAD(&cc.migratepages); | |
8051 | ||
041d3a8c MN |
8052 | /* |
8053 | * What we do here is we mark all pageblocks in range as | |
8054 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
8055 | * have different sizes, and due to the way page allocator | |
8056 | * work, we align the range to biggest of the two pages so | |
8057 | * that page allocator won't try to merge buddies from | |
8058 | * different pageblocks and change MIGRATE_ISOLATE to some | |
8059 | * other migration type. | |
8060 | * | |
8061 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
8062 | * migrate the pages from an unaligned range (ie. pages that | |
8063 | * we are interested in). This will put all the pages in | |
8064 | * range back to page allocator as MIGRATE_ISOLATE. | |
8065 | * | |
8066 | * When this is done, we take the pages in range from page | |
8067 | * allocator removing them from the buddy system. This way | |
8068 | * page allocator will never consider using them. | |
8069 | * | |
8070 | * This lets us mark the pageblocks back as | |
8071 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
8072 | * aligned range but not in the unaligned, original range are | |
8073 | * put back to page allocator so that buddy can use them. | |
8074 | */ | |
8075 | ||
8076 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
d381c547 | 8077 | pfn_max_align_up(end), migratetype, 0); |
041d3a8c | 8078 | if (ret) |
86a595f9 | 8079 | return ret; |
041d3a8c | 8080 | |
8ef5849f JK |
8081 | /* |
8082 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
8083 | * So, just fall through. test_pages_isolated() has a tracepoint |
8084 | * which will report the busy page. | |
8085 | * | |
8086 | * It is possible that busy pages could become available before | |
8087 | * the call to test_pages_isolated, and the range will actually be | |
8088 | * allocated. So, if we fall through be sure to clear ret so that | |
8089 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 8090 | */ |
bb13ffeb | 8091 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 8092 | if (ret && ret != -EBUSY) |
041d3a8c | 8093 | goto done; |
63cd4489 | 8094 | ret =0; |
041d3a8c MN |
8095 | |
8096 | /* | |
8097 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
8098 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
8099 | * more, all pages in [start, end) are free in page allocator. | |
8100 | * What we are going to do is to allocate all pages from | |
8101 | * [start, end) (that is remove them from page allocator). | |
8102 | * | |
8103 | * The only problem is that pages at the beginning and at the | |
8104 | * end of interesting range may be not aligned with pages that | |
8105 | * page allocator holds, ie. they can be part of higher order | |
8106 | * pages. Because of this, we reserve the bigger range and | |
8107 | * once this is done free the pages we are not interested in. | |
8108 | * | |
8109 | * We don't have to hold zone->lock here because the pages are | |
8110 | * isolated thus they won't get removed from buddy. | |
8111 | */ | |
8112 | ||
8113 | lru_add_drain_all(); | |
510f5507 | 8114 | drain_all_pages(cc.zone); |
041d3a8c MN |
8115 | |
8116 | order = 0; | |
8117 | outer_start = start; | |
8118 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
8119 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
8120 | outer_start = start; |
8121 | break; | |
041d3a8c MN |
8122 | } |
8123 | outer_start &= ~0UL << order; | |
8124 | } | |
8125 | ||
8ef5849f JK |
8126 | if (outer_start != start) { |
8127 | order = page_order(pfn_to_page(outer_start)); | |
8128 | ||
8129 | /* | |
8130 | * outer_start page could be small order buddy page and | |
8131 | * it doesn't include start page. Adjust outer_start | |
8132 | * in this case to report failed page properly | |
8133 | * on tracepoint in test_pages_isolated() | |
8134 | */ | |
8135 | if (outer_start + (1UL << order) <= start) | |
8136 | outer_start = start; | |
8137 | } | |
8138 | ||
041d3a8c | 8139 | /* Make sure the range is really isolated. */ |
b023f468 | 8140 | if (test_pages_isolated(outer_start, end, false)) { |
75dddef3 | 8141 | pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n", |
dae803e1 | 8142 | __func__, outer_start, end); |
041d3a8c MN |
8143 | ret = -EBUSY; |
8144 | goto done; | |
8145 | } | |
8146 | ||
49f223a9 | 8147 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 8148 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
8149 | if (!outer_end) { |
8150 | ret = -EBUSY; | |
8151 | goto done; | |
8152 | } | |
8153 | ||
8154 | /* Free head and tail (if any) */ | |
8155 | if (start != outer_start) | |
8156 | free_contig_range(outer_start, start - outer_start); | |
8157 | if (end != outer_end) | |
8158 | free_contig_range(end, outer_end - end); | |
8159 | ||
8160 | done: | |
8161 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 8162 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
8163 | return ret; |
8164 | } | |
8165 | ||
8166 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
8167 | { | |
bcc2b02f MS |
8168 | unsigned int count = 0; |
8169 | ||
8170 | for (; nr_pages--; pfn++) { | |
8171 | struct page *page = pfn_to_page(pfn); | |
8172 | ||
8173 | count += page_count(page) != 1; | |
8174 | __free_page(page); | |
8175 | } | |
8176 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
8177 | } |
8178 | #endif | |
8179 | ||
d883c6cf | 8180 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
8181 | /* |
8182 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
8183 | * page high values need to be recalulated. | |
8184 | */ | |
4ed7e022 JL |
8185 | void __meminit zone_pcp_update(struct zone *zone) |
8186 | { | |
0a647f38 | 8187 | unsigned cpu; |
c8e251fa | 8188 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 8189 | for_each_possible_cpu(cpu) |
169f6c19 CS |
8190 | pageset_set_high_and_batch(zone, |
8191 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 8192 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
8193 | } |
8194 | #endif | |
8195 | ||
340175b7 JL |
8196 | void zone_pcp_reset(struct zone *zone) |
8197 | { | |
8198 | unsigned long flags; | |
5a883813 MK |
8199 | int cpu; |
8200 | struct per_cpu_pageset *pset; | |
340175b7 JL |
8201 | |
8202 | /* avoid races with drain_pages() */ | |
8203 | local_irq_save(flags); | |
8204 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
8205 | for_each_online_cpu(cpu) { |
8206 | pset = per_cpu_ptr(zone->pageset, cpu); | |
8207 | drain_zonestat(zone, pset); | |
8208 | } | |
340175b7 JL |
8209 | free_percpu(zone->pageset); |
8210 | zone->pageset = &boot_pageset; | |
8211 | } | |
8212 | local_irq_restore(flags); | |
8213 | } | |
8214 | ||
6dcd73d7 | 8215 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 8216 | /* |
b9eb6319 JK |
8217 | * All pages in the range must be in a single zone and isolated |
8218 | * before calling this. | |
0c0e6195 KH |
8219 | */ |
8220 | void | |
8221 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
8222 | { | |
8223 | struct page *page; | |
8224 | struct zone *zone; | |
7aeb09f9 | 8225 | unsigned int order, i; |
0c0e6195 KH |
8226 | unsigned long pfn; |
8227 | unsigned long flags; | |
8228 | /* find the first valid pfn */ | |
8229 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
8230 | if (pfn_valid(pfn)) | |
8231 | break; | |
8232 | if (pfn == end_pfn) | |
8233 | return; | |
2d070eab | 8234 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
8235 | zone = page_zone(pfn_to_page(pfn)); |
8236 | spin_lock_irqsave(&zone->lock, flags); | |
8237 | pfn = start_pfn; | |
8238 | while (pfn < end_pfn) { | |
8239 | if (!pfn_valid(pfn)) { | |
8240 | pfn++; | |
8241 | continue; | |
8242 | } | |
8243 | page = pfn_to_page(pfn); | |
b023f468 WC |
8244 | /* |
8245 | * The HWPoisoned page may be not in buddy system, and | |
8246 | * page_count() is not 0. | |
8247 | */ | |
8248 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
8249 | pfn++; | |
8250 | SetPageReserved(page); | |
8251 | continue; | |
8252 | } | |
8253 | ||
0c0e6195 KH |
8254 | BUG_ON(page_count(page)); |
8255 | BUG_ON(!PageBuddy(page)); | |
8256 | order = page_order(page); | |
8257 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
8258 | pr_info("remove from free list %lx %d %lx\n", |
8259 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
8260 | #endif |
8261 | list_del(&page->lru); | |
8262 | rmv_page_order(page); | |
8263 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
8264 | for (i = 0; i < (1 << order); i++) |
8265 | SetPageReserved((page+i)); | |
8266 | pfn += (1 << order); | |
8267 | } | |
8268 | spin_unlock_irqrestore(&zone->lock, flags); | |
8269 | } | |
8270 | #endif | |
8d22ba1b | 8271 | |
8d22ba1b WF |
8272 | bool is_free_buddy_page(struct page *page) |
8273 | { | |
8274 | struct zone *zone = page_zone(page); | |
8275 | unsigned long pfn = page_to_pfn(page); | |
8276 | unsigned long flags; | |
7aeb09f9 | 8277 | unsigned int order; |
8d22ba1b WF |
8278 | |
8279 | spin_lock_irqsave(&zone->lock, flags); | |
8280 | for (order = 0; order < MAX_ORDER; order++) { | |
8281 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8282 | ||
8283 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
8284 | break; | |
8285 | } | |
8286 | spin_unlock_irqrestore(&zone->lock, flags); | |
8287 | ||
8288 | return order < MAX_ORDER; | |
8289 | } | |
d4ae9916 NH |
8290 | |
8291 | #ifdef CONFIG_MEMORY_FAILURE | |
8292 | /* | |
8293 | * Set PG_hwpoison flag if a given page is confirmed to be a free page. This | |
8294 | * test is performed under the zone lock to prevent a race against page | |
8295 | * allocation. | |
8296 | */ | |
8297 | bool set_hwpoison_free_buddy_page(struct page *page) | |
8298 | { | |
8299 | struct zone *zone = page_zone(page); | |
8300 | unsigned long pfn = page_to_pfn(page); | |
8301 | unsigned long flags; | |
8302 | unsigned int order; | |
8303 | bool hwpoisoned = false; | |
8304 | ||
8305 | spin_lock_irqsave(&zone->lock, flags); | |
8306 | for (order = 0; order < MAX_ORDER; order++) { | |
8307 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8308 | ||
8309 | if (PageBuddy(page_head) && page_order(page_head) >= order) { | |
8310 | if (!TestSetPageHWPoison(page)) | |
8311 | hwpoisoned = true; | |
8312 | break; | |
8313 | } | |
8314 | } | |
8315 | spin_unlock_irqrestore(&zone->lock, flags); | |
8316 | ||
8317 | return hwpoisoned; | |
8318 | } | |
8319 | #endif |