Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/page_alloc.c | |
4 | * | |
5 | * Manages the free list, the system allocates free pages here. | |
6 | * Note that kmalloc() lives in slab.c | |
7 | * | |
8 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
9 | * Swap reorganised 29.12.95, Stephen Tweedie | |
10 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
11 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
12 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
13 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
14 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
15 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
16 | */ | |
17 | ||
1da177e4 LT |
18 | #include <linux/stddef.h> |
19 | #include <linux/mm.h> | |
ca79b0c2 | 20 | #include <linux/highmem.h> |
1da177e4 LT |
21 | #include <linux/swap.h> |
22 | #include <linux/interrupt.h> | |
23 | #include <linux/pagemap.h> | |
10ed273f | 24 | #include <linux/jiffies.h> |
edbe7d23 | 25 | #include <linux/memblock.h> |
1da177e4 | 26 | #include <linux/compiler.h> |
9f158333 | 27 | #include <linux/kernel.h> |
b8c73fc2 | 28 | #include <linux/kasan.h> |
1da177e4 LT |
29 | #include <linux/module.h> |
30 | #include <linux/suspend.h> | |
31 | #include <linux/pagevec.h> | |
32 | #include <linux/blkdev.h> | |
33 | #include <linux/slab.h> | |
a238ab5b | 34 | #include <linux/ratelimit.h> |
5a3135c2 | 35 | #include <linux/oom.h> |
1da177e4 LT |
36 | #include <linux/topology.h> |
37 | #include <linux/sysctl.h> | |
38 | #include <linux/cpu.h> | |
39 | #include <linux/cpuset.h> | |
bdc8cb98 | 40 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
41 | #include <linux/nodemask.h> |
42 | #include <linux/vmalloc.h> | |
a6cccdc3 | 43 | #include <linux/vmstat.h> |
4be38e35 | 44 | #include <linux/mempolicy.h> |
4b94ffdc | 45 | #include <linux/memremap.h> |
6811378e | 46 | #include <linux/stop_machine.h> |
97500a4a | 47 | #include <linux/random.h> |
c713216d MG |
48 | #include <linux/sort.h> |
49 | #include <linux/pfn.h> | |
3fcfab16 | 50 | #include <linux/backing-dev.h> |
933e312e | 51 | #include <linux/fault-inject.h> |
a5d76b54 | 52 | #include <linux/page-isolation.h> |
3ac7fe5a | 53 | #include <linux/debugobjects.h> |
dbb1f81c | 54 | #include <linux/kmemleak.h> |
56de7263 | 55 | #include <linux/compaction.h> |
0d3d062a | 56 | #include <trace/events/kmem.h> |
d379f01d | 57 | #include <trace/events/oom.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
f920e413 | 60 | #include <linux/mmu_notifier.h> |
041d3a8c | 61 | #include <linux/migrate.h> |
949f7ec5 | 62 | #include <linux/hugetlb.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
5b3cc15a | 64 | #include <linux/sched/mm.h> |
48c96a36 | 65 | #include <linux/page_owner.h> |
0e1cc95b | 66 | #include <linux/kthread.h> |
4949148a | 67 | #include <linux/memcontrol.h> |
42c269c8 | 68 | #include <linux/ftrace.h> |
d92a8cfc | 69 | #include <linux/lockdep.h> |
556b969a | 70 | #include <linux/nmi.h> |
eb414681 | 71 | #include <linux/psi.h> |
e4443149 | 72 | #include <linux/padata.h> |
4aab2be0 | 73 | #include <linux/khugepaged.h> |
ba8f3587 | 74 | #include <linux/buffer_head.h> |
7ee3d4e8 | 75 | #include <asm/sections.h> |
1da177e4 | 76 | #include <asm/tlbflush.h> |
ac924c60 | 77 | #include <asm/div64.h> |
1da177e4 | 78 | #include "internal.h" |
e900a918 | 79 | #include "shuffle.h" |
36e66c55 | 80 | #include "page_reporting.h" |
1da177e4 | 81 | |
f04a5d5d DH |
82 | /* Free Page Internal flags: for internal, non-pcp variants of free_pages(). */ |
83 | typedef int __bitwise fpi_t; | |
84 | ||
85 | /* No special request */ | |
86 | #define FPI_NONE ((__force fpi_t)0) | |
87 | ||
88 | /* | |
89 | * Skip free page reporting notification for the (possibly merged) page. | |
90 | * This does not hinder free page reporting from grabbing the page, | |
91 | * reporting it and marking it "reported" - it only skips notifying | |
92 | * the free page reporting infrastructure about a newly freed page. For | |
93 | * example, used when temporarily pulling a page from a freelist and | |
94 | * putting it back unmodified. | |
95 | */ | |
96 | #define FPI_SKIP_REPORT_NOTIFY ((__force fpi_t)BIT(0)) | |
97 | ||
47b6a24a DH |
98 | /* |
99 | * Place the (possibly merged) page to the tail of the freelist. Will ignore | |
100 | * page shuffling (relevant code - e.g., memory onlining - is expected to | |
101 | * shuffle the whole zone). | |
102 | * | |
103 | * Note: No code should rely on this flag for correctness - it's purely | |
104 | * to allow for optimizations when handing back either fresh pages | |
105 | * (memory onlining) or untouched pages (page isolation, free page | |
106 | * reporting). | |
107 | */ | |
108 | #define FPI_TO_TAIL ((__force fpi_t)BIT(1)) | |
109 | ||
2c335680 AK |
110 | /* |
111 | * Don't poison memory with KASAN (only for the tag-based modes). | |
112 | * During boot, all non-reserved memblock memory is exposed to page_alloc. | |
113 | * Poisoning all that memory lengthens boot time, especially on systems with | |
114 | * large amount of RAM. This flag is used to skip that poisoning. | |
115 | * This is only done for the tag-based KASAN modes, as those are able to | |
116 | * detect memory corruptions with the memory tags assigned by default. | |
117 | * All memory allocated normally after boot gets poisoned as usual. | |
118 | */ | |
119 | #define FPI_SKIP_KASAN_POISON ((__force fpi_t)BIT(2)) | |
120 | ||
c8e251fa CS |
121 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
122 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
74f44822 | 123 | #define MIN_PERCPU_PAGELIST_HIGH_FRACTION (8) |
c8e251fa | 124 | |
dbbee9d5 MG |
125 | struct pagesets { |
126 | local_lock_t lock; | |
127 | #if defined(CONFIG_DEBUG_INFO_BTF) && \ | |
128 | !defined(CONFIG_DEBUG_LOCK_ALLOC) && \ | |
129 | !defined(CONFIG_PAHOLE_HAS_ZEROSIZE_PERCPU_SUPPORT) | |
130 | /* | |
131 | * pahole 1.21 and earlier gets confused by zero-sized per-CPU | |
132 | * variables and produces invalid BTF. Ensure that | |
133 | * sizeof(struct pagesets) != 0 for older versions of pahole. | |
134 | */ | |
135 | char __pahole_hack; | |
136 | #warning "pahole too old to support zero-sized struct pagesets" | |
137 | #endif | |
138 | }; | |
139 | static DEFINE_PER_CPU(struct pagesets, pagesets) = { | |
140 | .lock = INIT_LOCAL_LOCK(lock), | |
141 | }; | |
142 | ||
72812019 LS |
143 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
144 | DEFINE_PER_CPU(int, numa_node); | |
145 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
146 | #endif | |
147 | ||
4518085e KW |
148 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
149 | ||
7aac7898 LS |
150 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
151 | /* | |
152 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
153 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
154 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
155 | * defined in <linux/topology.h>. | |
156 | */ | |
157 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
158 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
159 | #endif | |
160 | ||
bd233f53 | 161 | /* work_structs for global per-cpu drains */ |
d9367bd0 WY |
162 | struct pcpu_drain { |
163 | struct zone *zone; | |
164 | struct work_struct work; | |
165 | }; | |
8b885f53 JY |
166 | static DEFINE_MUTEX(pcpu_drain_mutex); |
167 | static DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain); | |
bd233f53 | 168 | |
38addce8 | 169 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 170 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
171 | EXPORT_SYMBOL(latent_entropy); |
172 | #endif | |
173 | ||
1da177e4 | 174 | /* |
13808910 | 175 | * Array of node states. |
1da177e4 | 176 | */ |
13808910 CL |
177 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
178 | [N_POSSIBLE] = NODE_MASK_ALL, | |
179 | [N_ONLINE] = { { [0] = 1UL } }, | |
180 | #ifndef CONFIG_NUMA | |
181 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
182 | #ifdef CONFIG_HIGHMEM | |
183 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 184 | #endif |
20b2f52b | 185 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
186 | [N_CPU] = { { [0] = 1UL } }, |
187 | #endif /* NUMA */ | |
188 | }; | |
189 | EXPORT_SYMBOL(node_states); | |
190 | ||
ca79b0c2 AK |
191 | atomic_long_t _totalram_pages __read_mostly; |
192 | EXPORT_SYMBOL(_totalram_pages); | |
cb45b0e9 | 193 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 194 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 195 | |
74f44822 | 196 | int percpu_pagelist_high_fraction; |
dcce284a | 197 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
51cba1eb | 198 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc); |
6471384a AP |
199 | EXPORT_SYMBOL(init_on_alloc); |
200 | ||
51cba1eb | 201 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free); |
6471384a AP |
202 | EXPORT_SYMBOL(init_on_free); |
203 | ||
04013513 VB |
204 | static bool _init_on_alloc_enabled_early __read_mostly |
205 | = IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON); | |
6471384a AP |
206 | static int __init early_init_on_alloc(char *buf) |
207 | { | |
6471384a | 208 | |
04013513 | 209 | return kstrtobool(buf, &_init_on_alloc_enabled_early); |
6471384a AP |
210 | } |
211 | early_param("init_on_alloc", early_init_on_alloc); | |
212 | ||
04013513 VB |
213 | static bool _init_on_free_enabled_early __read_mostly |
214 | = IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON); | |
6471384a AP |
215 | static int __init early_init_on_free(char *buf) |
216 | { | |
04013513 | 217 | return kstrtobool(buf, &_init_on_free_enabled_early); |
6471384a AP |
218 | } |
219 | early_param("init_on_free", early_init_on_free); | |
1da177e4 | 220 | |
bb14c2c7 VB |
221 | /* |
222 | * A cached value of the page's pageblock's migratetype, used when the page is | |
223 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
224 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
225 | * Also the migratetype set in the page does not necessarily match the pcplist | |
226 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
227 | * other index - this ensures that it will be put on the correct CMA freelist. | |
228 | */ | |
229 | static inline int get_pcppage_migratetype(struct page *page) | |
230 | { | |
231 | return page->index; | |
232 | } | |
233 | ||
234 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
235 | { | |
236 | page->index = migratetype; | |
237 | } | |
238 | ||
452aa699 RW |
239 | #ifdef CONFIG_PM_SLEEP |
240 | /* | |
241 | * The following functions are used by the suspend/hibernate code to temporarily | |
242 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
243 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
244 | * they should always be called with system_transition_mutex held |
245 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
246 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
247 | * with that modification). | |
452aa699 | 248 | */ |
c9e664f1 RW |
249 | |
250 | static gfp_t saved_gfp_mask; | |
251 | ||
252 | void pm_restore_gfp_mask(void) | |
452aa699 | 253 | { |
55f2503c | 254 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
255 | if (saved_gfp_mask) { |
256 | gfp_allowed_mask = saved_gfp_mask; | |
257 | saved_gfp_mask = 0; | |
258 | } | |
452aa699 RW |
259 | } |
260 | ||
c9e664f1 | 261 | void pm_restrict_gfp_mask(void) |
452aa699 | 262 | { |
55f2503c | 263 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
264 | WARN_ON(saved_gfp_mask); |
265 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 266 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 267 | } |
f90ac398 MG |
268 | |
269 | bool pm_suspended_storage(void) | |
270 | { | |
d0164adc | 271 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
272 | return false; |
273 | return true; | |
274 | } | |
452aa699 RW |
275 | #endif /* CONFIG_PM_SLEEP */ |
276 | ||
d9c23400 | 277 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 278 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
279 | #endif |
280 | ||
7fef431b DH |
281 | static void __free_pages_ok(struct page *page, unsigned int order, |
282 | fpi_t fpi_flags); | |
a226f6c8 | 283 | |
1da177e4 LT |
284 | /* |
285 | * results with 256, 32 in the lowmem_reserve sysctl: | |
286 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
287 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
288 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
289 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 290 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
291 | * |
292 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
293 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 294 | */ |
d3cda233 | 295 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 296 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 297 | [ZONE_DMA] = 256, |
4b51d669 | 298 | #endif |
fb0e7942 | 299 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 300 | [ZONE_DMA32] = 256, |
fb0e7942 | 301 | #endif |
d3cda233 | 302 | [ZONE_NORMAL] = 32, |
e53ef38d | 303 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 304 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 305 | #endif |
d3cda233 | 306 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 307 | }; |
1da177e4 | 308 | |
15ad7cdc | 309 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 310 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 311 | "DMA", |
4b51d669 | 312 | #endif |
fb0e7942 | 313 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 314 | "DMA32", |
fb0e7942 | 315 | #endif |
2f1b6248 | 316 | "Normal", |
e53ef38d | 317 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 318 | "HighMem", |
e53ef38d | 319 | #endif |
2a1e274a | 320 | "Movable", |
033fbae9 DW |
321 | #ifdef CONFIG_ZONE_DEVICE |
322 | "Device", | |
323 | #endif | |
2f1b6248 CL |
324 | }; |
325 | ||
c999fbd3 | 326 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
327 | "Unmovable", |
328 | "Movable", | |
329 | "Reclaimable", | |
330 | "HighAtomic", | |
331 | #ifdef CONFIG_CMA | |
332 | "CMA", | |
333 | #endif | |
334 | #ifdef CONFIG_MEMORY_ISOLATION | |
335 | "Isolate", | |
336 | #endif | |
337 | }; | |
338 | ||
ae70eddd AK |
339 | compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = { |
340 | [NULL_COMPOUND_DTOR] = NULL, | |
341 | [COMPOUND_PAGE_DTOR] = free_compound_page, | |
f1e61557 | 342 | #ifdef CONFIG_HUGETLB_PAGE |
ae70eddd | 343 | [HUGETLB_PAGE_DTOR] = free_huge_page, |
f1e61557 | 344 | #endif |
9a982250 | 345 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
ae70eddd | 346 | [TRANSHUGE_PAGE_DTOR] = free_transhuge_page, |
9a982250 | 347 | #endif |
f1e61557 KS |
348 | }; |
349 | ||
1da177e4 | 350 | int min_free_kbytes = 1024; |
42aa83cb | 351 | int user_min_free_kbytes = -1; |
1c30844d | 352 | int watermark_boost_factor __read_mostly = 15000; |
795ae7a0 | 353 | int watermark_scale_factor = 10; |
1da177e4 | 354 | |
bbe5d993 OS |
355 | static unsigned long nr_kernel_pages __initdata; |
356 | static unsigned long nr_all_pages __initdata; | |
357 | static unsigned long dma_reserve __initdata; | |
1da177e4 | 358 | |
bbe5d993 OS |
359 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata; |
360 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata; | |
7f16f91f | 361 | static unsigned long required_kernelcore __initdata; |
a5c6d650 | 362 | static unsigned long required_kernelcore_percent __initdata; |
7f16f91f | 363 | static unsigned long required_movablecore __initdata; |
a5c6d650 | 364 | static unsigned long required_movablecore_percent __initdata; |
bbe5d993 | 365 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata; |
7f16f91f | 366 | static bool mirrored_kernelcore __meminitdata; |
0ee332c1 TH |
367 | |
368 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
369 | int movable_zone; | |
370 | EXPORT_SYMBOL(movable_zone); | |
c713216d | 371 | |
418508c1 | 372 | #if MAX_NUMNODES > 1 |
b9726c26 | 373 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 374 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 375 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 376 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
377 | #endif |
378 | ||
9ef9acb0 MG |
379 | int page_group_by_mobility_disabled __read_mostly; |
380 | ||
3a80a7fa | 381 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
382 | /* |
383 | * During boot we initialize deferred pages on-demand, as needed, but once | |
384 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
385 | * and we can permanently disable that path. | |
386 | */ | |
387 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
388 | ||
389 | /* | |
390 | * Calling kasan_free_pages() only after deferred memory initialization | |
391 | * has completed. Poisoning pages during deferred memory init will greatly | |
392 | * lengthen the process and cause problem in large memory systems as the | |
393 | * deferred pages initialization is done with interrupt disabled. | |
394 | * | |
395 | * Assuming that there will be no reference to those newly initialized | |
396 | * pages before they are ever allocated, this should have no effect on | |
397 | * KASAN memory tracking as the poison will be properly inserted at page | |
398 | * allocation time. The only corner case is when pages are allocated by | |
399 | * on-demand allocation and then freed again before the deferred pages | |
400 | * initialization is done, but this is not likely to happen. | |
401 | */ | |
2c335680 | 402 | static inline void kasan_free_nondeferred_pages(struct page *page, int order, |
1bb5eab3 | 403 | bool init, fpi_t fpi_flags) |
3c0c12cc | 404 | { |
2c335680 AK |
405 | if (static_branch_unlikely(&deferred_pages)) |
406 | return; | |
407 | if (!IS_ENABLED(CONFIG_KASAN_GENERIC) && | |
408 | (fpi_flags & FPI_SKIP_KASAN_POISON)) | |
409 | return; | |
1bb5eab3 | 410 | kasan_free_pages(page, order, init); |
3c0c12cc WL |
411 | } |
412 | ||
3a80a7fa | 413 | /* Returns true if the struct page for the pfn is uninitialised */ |
0e1cc95b | 414 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 415 | { |
ef70b6f4 MG |
416 | int nid = early_pfn_to_nid(pfn); |
417 | ||
418 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
419 | return true; |
420 | ||
421 | return false; | |
422 | } | |
423 | ||
424 | /* | |
d3035be4 | 425 | * Returns true when the remaining initialisation should be deferred until |
3a80a7fa MG |
426 | * later in the boot cycle when it can be parallelised. |
427 | */ | |
d3035be4 PT |
428 | static bool __meminit |
429 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn) | |
3a80a7fa | 430 | { |
d3035be4 PT |
431 | static unsigned long prev_end_pfn, nr_initialised; |
432 | ||
433 | /* | |
434 | * prev_end_pfn static that contains the end of previous zone | |
435 | * No need to protect because called very early in boot before smp_init. | |
436 | */ | |
437 | if (prev_end_pfn != end_pfn) { | |
438 | prev_end_pfn = end_pfn; | |
439 | nr_initialised = 0; | |
440 | } | |
441 | ||
3c2c6488 | 442 | /* Always populate low zones for address-constrained allocations */ |
d3035be4 | 443 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid))) |
3a80a7fa | 444 | return false; |
23b68cfa | 445 | |
dc2da7b4 BH |
446 | if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX) |
447 | return true; | |
23b68cfa WY |
448 | /* |
449 | * We start only with one section of pages, more pages are added as | |
450 | * needed until the rest of deferred pages are initialized. | |
451 | */ | |
d3035be4 | 452 | nr_initialised++; |
23b68cfa | 453 | if ((nr_initialised > PAGES_PER_SECTION) && |
d3035be4 PT |
454 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
455 | NODE_DATA(nid)->first_deferred_pfn = pfn; | |
456 | return true; | |
3a80a7fa | 457 | } |
d3035be4 | 458 | return false; |
3a80a7fa MG |
459 | } |
460 | #else | |
2c335680 | 461 | static inline void kasan_free_nondeferred_pages(struct page *page, int order, |
1bb5eab3 | 462 | bool init, fpi_t fpi_flags) |
2c335680 AK |
463 | { |
464 | if (!IS_ENABLED(CONFIG_KASAN_GENERIC) && | |
465 | (fpi_flags & FPI_SKIP_KASAN_POISON)) | |
466 | return; | |
1bb5eab3 | 467 | kasan_free_pages(page, order, init); |
2c335680 | 468 | } |
3c0c12cc | 469 | |
3a80a7fa MG |
470 | static inline bool early_page_uninitialised(unsigned long pfn) |
471 | { | |
472 | return false; | |
473 | } | |
474 | ||
d3035be4 | 475 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn) |
3a80a7fa | 476 | { |
d3035be4 | 477 | return false; |
3a80a7fa MG |
478 | } |
479 | #endif | |
480 | ||
0b423ca2 | 481 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
ca891f41 | 482 | static inline unsigned long *get_pageblock_bitmap(const struct page *page, |
0b423ca2 MG |
483 | unsigned long pfn) |
484 | { | |
485 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 486 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
487 | #else |
488 | return page_zone(page)->pageblock_flags; | |
489 | #endif /* CONFIG_SPARSEMEM */ | |
490 | } | |
491 | ||
ca891f41 | 492 | static inline int pfn_to_bitidx(const struct page *page, unsigned long pfn) |
0b423ca2 MG |
493 | { |
494 | #ifdef CONFIG_SPARSEMEM | |
495 | pfn &= (PAGES_PER_SECTION-1); | |
0b423ca2 MG |
496 | #else |
497 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
0b423ca2 | 498 | #endif /* CONFIG_SPARSEMEM */ |
399b795b | 499 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
0b423ca2 MG |
500 | } |
501 | ||
535b81e2 | 502 | static __always_inline |
ca891f41 | 503 | unsigned long __get_pfnblock_flags_mask(const struct page *page, |
0b423ca2 | 504 | unsigned long pfn, |
0b423ca2 MG |
505 | unsigned long mask) |
506 | { | |
507 | unsigned long *bitmap; | |
508 | unsigned long bitidx, word_bitidx; | |
509 | unsigned long word; | |
510 | ||
511 | bitmap = get_pageblock_bitmap(page, pfn); | |
512 | bitidx = pfn_to_bitidx(page, pfn); | |
513 | word_bitidx = bitidx / BITS_PER_LONG; | |
514 | bitidx &= (BITS_PER_LONG-1); | |
515 | ||
516 | word = bitmap[word_bitidx]; | |
d93d5ab9 | 517 | return (word >> bitidx) & mask; |
0b423ca2 MG |
518 | } |
519 | ||
a00cda3f MCC |
520 | /** |
521 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
522 | * @page: The page within the block of interest | |
523 | * @pfn: The target page frame number | |
524 | * @mask: mask of bits that the caller is interested in | |
525 | * | |
526 | * Return: pageblock_bits flags | |
527 | */ | |
ca891f41 MWO |
528 | unsigned long get_pfnblock_flags_mask(const struct page *page, |
529 | unsigned long pfn, unsigned long mask) | |
0b423ca2 | 530 | { |
535b81e2 | 531 | return __get_pfnblock_flags_mask(page, pfn, mask); |
0b423ca2 MG |
532 | } |
533 | ||
ca891f41 MWO |
534 | static __always_inline int get_pfnblock_migratetype(const struct page *page, |
535 | unsigned long pfn) | |
0b423ca2 | 536 | { |
535b81e2 | 537 | return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK); |
0b423ca2 MG |
538 | } |
539 | ||
540 | /** | |
541 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
542 | * @page: The page within the block of interest | |
543 | * @flags: The flags to set | |
544 | * @pfn: The target page frame number | |
0b423ca2 MG |
545 | * @mask: mask of bits that the caller is interested in |
546 | */ | |
547 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
548 | unsigned long pfn, | |
0b423ca2 MG |
549 | unsigned long mask) |
550 | { | |
551 | unsigned long *bitmap; | |
552 | unsigned long bitidx, word_bitidx; | |
553 | unsigned long old_word, word; | |
554 | ||
555 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 556 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
557 | |
558 | bitmap = get_pageblock_bitmap(page, pfn); | |
559 | bitidx = pfn_to_bitidx(page, pfn); | |
560 | word_bitidx = bitidx / BITS_PER_LONG; | |
561 | bitidx &= (BITS_PER_LONG-1); | |
562 | ||
563 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
564 | ||
d93d5ab9 WY |
565 | mask <<= bitidx; |
566 | flags <<= bitidx; | |
0b423ca2 MG |
567 | |
568 | word = READ_ONCE(bitmap[word_bitidx]); | |
569 | for (;;) { | |
570 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
571 | if (word == old_word) | |
572 | break; | |
573 | word = old_word; | |
574 | } | |
575 | } | |
3a80a7fa | 576 | |
ee6f509c | 577 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 578 | { |
5d0f3f72 KM |
579 | if (unlikely(page_group_by_mobility_disabled && |
580 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
581 | migratetype = MIGRATE_UNMOVABLE; |
582 | ||
d93d5ab9 | 583 | set_pfnblock_flags_mask(page, (unsigned long)migratetype, |
535b81e2 | 584 | page_to_pfn(page), MIGRATETYPE_MASK); |
b2a0ac88 MG |
585 | } |
586 | ||
13e7444b | 587 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 588 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 589 | { |
bdc8cb98 DH |
590 | int ret = 0; |
591 | unsigned seq; | |
592 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 593 | unsigned long sp, start_pfn; |
c6a57e19 | 594 | |
bdc8cb98 DH |
595 | do { |
596 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
597 | start_pfn = zone->zone_start_pfn; |
598 | sp = zone->spanned_pages; | |
108bcc96 | 599 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
600 | ret = 1; |
601 | } while (zone_span_seqretry(zone, seq)); | |
602 | ||
b5e6a5a2 | 603 | if (ret) |
613813e8 DH |
604 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
605 | pfn, zone_to_nid(zone), zone->name, | |
606 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 607 | |
bdc8cb98 | 608 | return ret; |
c6a57e19 DH |
609 | } |
610 | ||
611 | static int page_is_consistent(struct zone *zone, struct page *page) | |
612 | { | |
14e07298 | 613 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 614 | return 0; |
1da177e4 | 615 | if (zone != page_zone(page)) |
c6a57e19 DH |
616 | return 0; |
617 | ||
618 | return 1; | |
619 | } | |
620 | /* | |
621 | * Temporary debugging check for pages not lying within a given zone. | |
622 | */ | |
d73d3c9f | 623 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
624 | { |
625 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 626 | return 1; |
c6a57e19 DH |
627 | if (!page_is_consistent(zone, page)) |
628 | return 1; | |
629 | ||
1da177e4 LT |
630 | return 0; |
631 | } | |
13e7444b | 632 | #else |
d73d3c9f | 633 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
634 | { |
635 | return 0; | |
636 | } | |
637 | #endif | |
638 | ||
82a3241a | 639 | static void bad_page(struct page *page, const char *reason) |
1da177e4 | 640 | { |
d936cf9b HD |
641 | static unsigned long resume; |
642 | static unsigned long nr_shown; | |
643 | static unsigned long nr_unshown; | |
644 | ||
645 | /* | |
646 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
647 | * or allow a steady drip of one report per second. | |
648 | */ | |
649 | if (nr_shown == 60) { | |
650 | if (time_before(jiffies, resume)) { | |
651 | nr_unshown++; | |
652 | goto out; | |
653 | } | |
654 | if (nr_unshown) { | |
ff8e8116 | 655 | pr_alert( |
1e9e6365 | 656 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
657 | nr_unshown); |
658 | nr_unshown = 0; | |
659 | } | |
660 | nr_shown = 0; | |
661 | } | |
662 | if (nr_shown++ == 0) | |
663 | resume = jiffies + 60 * HZ; | |
664 | ||
ff8e8116 | 665 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 666 | current->comm, page_to_pfn(page)); |
d2f07ec0 | 667 | dump_page(page, reason); |
3dc14741 | 668 | |
4f31888c | 669 | print_modules(); |
1da177e4 | 670 | dump_stack(); |
d936cf9b | 671 | out: |
8cc3b392 | 672 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 673 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 674 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
675 | } |
676 | ||
21d02f8f MG |
677 | static inline void free_the_page(struct page *page, unsigned int order) |
678 | { | |
679 | if (order == 0) /* Via pcp? */ | |
680 | free_unref_page(page); | |
681 | else | |
682 | __free_pages_ok(page, order, FPI_NONE); | |
683 | } | |
684 | ||
1da177e4 LT |
685 | /* |
686 | * Higher-order pages are called "compound pages". They are structured thusly: | |
687 | * | |
1d798ca3 | 688 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 689 | * |
1d798ca3 KS |
690 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
691 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 692 | * |
1d798ca3 KS |
693 | * The first tail page's ->compound_dtor holds the offset in array of compound |
694 | * page destructors. See compound_page_dtors. | |
1da177e4 | 695 | * |
1d798ca3 | 696 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 697 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 698 | */ |
d98c7a09 | 699 | |
9a982250 | 700 | void free_compound_page(struct page *page) |
d98c7a09 | 701 | { |
7ae88534 | 702 | mem_cgroup_uncharge(page); |
7fef431b | 703 | __free_pages_ok(page, compound_order(page), FPI_NONE); |
d98c7a09 HD |
704 | } |
705 | ||
d00181b9 | 706 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
707 | { |
708 | int i; | |
709 | int nr_pages = 1 << order; | |
710 | ||
18229df5 AW |
711 | __SetPageHead(page); |
712 | for (i = 1; i < nr_pages; i++) { | |
713 | struct page *p = page + i; | |
58a84aa9 | 714 | set_page_count(p, 0); |
1c290f64 | 715 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 716 | set_compound_head(p, page); |
18229df5 | 717 | } |
1378a5ee MWO |
718 | |
719 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); | |
720 | set_compound_order(page, order); | |
53f9263b | 721 | atomic_set(compound_mapcount_ptr(page), -1); |
47e29d32 JH |
722 | if (hpage_pincount_available(page)) |
723 | atomic_set(compound_pincount_ptr(page), 0); | |
18229df5 AW |
724 | } |
725 | ||
c0a32fc5 SG |
726 | #ifdef CONFIG_DEBUG_PAGEALLOC |
727 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 728 | |
8e57f8ac VB |
729 | bool _debug_pagealloc_enabled_early __read_mostly |
730 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
731 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 732 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 733 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
734 | |
735 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 736 | |
031bc574 JK |
737 | static int __init early_debug_pagealloc(char *buf) |
738 | { | |
8e57f8ac | 739 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
740 | } |
741 | early_param("debug_pagealloc", early_debug_pagealloc); | |
742 | ||
c0a32fc5 SG |
743 | static int __init debug_guardpage_minorder_setup(char *buf) |
744 | { | |
745 | unsigned long res; | |
746 | ||
747 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 748 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
749 | return 0; |
750 | } | |
751 | _debug_guardpage_minorder = res; | |
1170532b | 752 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
753 | return 0; |
754 | } | |
f1c1e9f7 | 755 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 756 | |
acbc15a4 | 757 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 758 | unsigned int order, int migratetype) |
c0a32fc5 | 759 | { |
e30825f1 | 760 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
761 | return false; |
762 | ||
763 | if (order >= debug_guardpage_minorder()) | |
764 | return false; | |
e30825f1 | 765 | |
3972f6bb | 766 | __SetPageGuard(page); |
2847cf95 JK |
767 | INIT_LIST_HEAD(&page->lru); |
768 | set_page_private(page, order); | |
769 | /* Guard pages are not available for any usage */ | |
770 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
771 | |
772 | return true; | |
c0a32fc5 SG |
773 | } |
774 | ||
2847cf95 JK |
775 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
776 | unsigned int order, int migratetype) | |
c0a32fc5 | 777 | { |
e30825f1 JK |
778 | if (!debug_guardpage_enabled()) |
779 | return; | |
780 | ||
3972f6bb | 781 | __ClearPageGuard(page); |
e30825f1 | 782 | |
2847cf95 JK |
783 | set_page_private(page, 0); |
784 | if (!is_migrate_isolate(migratetype)) | |
785 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
786 | } |
787 | #else | |
acbc15a4 JK |
788 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
789 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
790 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
791 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
792 | #endif |
793 | ||
04013513 VB |
794 | /* |
795 | * Enable static keys related to various memory debugging and hardening options. | |
796 | * Some override others, and depend on early params that are evaluated in the | |
797 | * order of appearance. So we need to first gather the full picture of what was | |
798 | * enabled, and then make decisions. | |
799 | */ | |
800 | void init_mem_debugging_and_hardening(void) | |
801 | { | |
9df65f52 ST |
802 | bool page_poisoning_requested = false; |
803 | ||
804 | #ifdef CONFIG_PAGE_POISONING | |
805 | /* | |
806 | * Page poisoning is debug page alloc for some arches. If | |
807 | * either of those options are enabled, enable poisoning. | |
808 | */ | |
809 | if (page_poisoning_enabled() || | |
810 | (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) && | |
811 | debug_pagealloc_enabled())) { | |
812 | static_branch_enable(&_page_poisoning_enabled); | |
813 | page_poisoning_requested = true; | |
814 | } | |
815 | #endif | |
816 | ||
04013513 | 817 | if (_init_on_alloc_enabled_early) { |
9df65f52 | 818 | if (page_poisoning_requested) |
04013513 VB |
819 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, " |
820 | "will take precedence over init_on_alloc\n"); | |
821 | else | |
822 | static_branch_enable(&init_on_alloc); | |
823 | } | |
824 | if (_init_on_free_enabled_early) { | |
9df65f52 | 825 | if (page_poisoning_requested) |
04013513 VB |
826 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, " |
827 | "will take precedence over init_on_free\n"); | |
828 | else | |
829 | static_branch_enable(&init_on_free); | |
830 | } | |
831 | ||
832 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
833 | if (!debug_pagealloc_enabled()) | |
834 | return; | |
835 | ||
836 | static_branch_enable(&_debug_pagealloc_enabled); | |
837 | ||
838 | if (!debug_guardpage_minorder()) | |
839 | return; | |
840 | ||
841 | static_branch_enable(&_debug_guardpage_enabled); | |
842 | #endif | |
843 | } | |
844 | ||
ab130f91 | 845 | static inline void set_buddy_order(struct page *page, unsigned int order) |
6aa3001b | 846 | { |
4c21e2f2 | 847 | set_page_private(page, order); |
676165a8 | 848 | __SetPageBuddy(page); |
1da177e4 LT |
849 | } |
850 | ||
1da177e4 LT |
851 | /* |
852 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 853 | * we can coalesce a page and its buddy if |
13ad59df | 854 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 855 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
856 | * (c) a page and its buddy have the same order && |
857 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 858 | * |
6e292b9b MW |
859 | * For recording whether a page is in the buddy system, we set PageBuddy. |
860 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 861 | * |
676165a8 | 862 | * For recording page's order, we use page_private(page). |
1da177e4 | 863 | */ |
fe925c0c | 864 | static inline bool page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 865 | unsigned int order) |
1da177e4 | 866 | { |
fe925c0c | 867 | if (!page_is_guard(buddy) && !PageBuddy(buddy)) |
868 | return false; | |
4c5018ce | 869 | |
ab130f91 | 870 | if (buddy_order(buddy) != order) |
fe925c0c | 871 | return false; |
c0a32fc5 | 872 | |
fe925c0c | 873 | /* |
874 | * zone check is done late to avoid uselessly calculating | |
875 | * zone/node ids for pages that could never merge. | |
876 | */ | |
877 | if (page_zone_id(page) != page_zone_id(buddy)) | |
878 | return false; | |
d34c5fa0 | 879 | |
fe925c0c | 880 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
4c5018ce | 881 | |
fe925c0c | 882 | return true; |
1da177e4 LT |
883 | } |
884 | ||
5e1f0f09 MG |
885 | #ifdef CONFIG_COMPACTION |
886 | static inline struct capture_control *task_capc(struct zone *zone) | |
887 | { | |
888 | struct capture_control *capc = current->capture_control; | |
889 | ||
deba0487 | 890 | return unlikely(capc) && |
5e1f0f09 MG |
891 | !(current->flags & PF_KTHREAD) && |
892 | !capc->page && | |
deba0487 | 893 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
894 | } |
895 | ||
896 | static inline bool | |
897 | compaction_capture(struct capture_control *capc, struct page *page, | |
898 | int order, int migratetype) | |
899 | { | |
900 | if (!capc || order != capc->cc->order) | |
901 | return false; | |
902 | ||
903 | /* Do not accidentally pollute CMA or isolated regions*/ | |
904 | if (is_migrate_cma(migratetype) || | |
905 | is_migrate_isolate(migratetype)) | |
906 | return false; | |
907 | ||
908 | /* | |
f0953a1b | 909 | * Do not let lower order allocations pollute a movable pageblock. |
5e1f0f09 MG |
910 | * This might let an unmovable request use a reclaimable pageblock |
911 | * and vice-versa but no more than normal fallback logic which can | |
912 | * have trouble finding a high-order free page. | |
913 | */ | |
914 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
915 | return false; | |
916 | ||
917 | capc->page = page; | |
918 | return true; | |
919 | } | |
920 | ||
921 | #else | |
922 | static inline struct capture_control *task_capc(struct zone *zone) | |
923 | { | |
924 | return NULL; | |
925 | } | |
926 | ||
927 | static inline bool | |
928 | compaction_capture(struct capture_control *capc, struct page *page, | |
929 | int order, int migratetype) | |
930 | { | |
931 | return false; | |
932 | } | |
933 | #endif /* CONFIG_COMPACTION */ | |
934 | ||
6ab01363 AD |
935 | /* Used for pages not on another list */ |
936 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
937 | unsigned int order, int migratetype) | |
938 | { | |
939 | struct free_area *area = &zone->free_area[order]; | |
940 | ||
941 | list_add(&page->lru, &area->free_list[migratetype]); | |
942 | area->nr_free++; | |
943 | } | |
944 | ||
945 | /* Used for pages not on another list */ | |
946 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
947 | unsigned int order, int migratetype) | |
948 | { | |
949 | struct free_area *area = &zone->free_area[order]; | |
950 | ||
951 | list_add_tail(&page->lru, &area->free_list[migratetype]); | |
952 | area->nr_free++; | |
953 | } | |
954 | ||
293ffa5e DH |
955 | /* |
956 | * Used for pages which are on another list. Move the pages to the tail | |
957 | * of the list - so the moved pages won't immediately be considered for | |
958 | * allocation again (e.g., optimization for memory onlining). | |
959 | */ | |
6ab01363 AD |
960 | static inline void move_to_free_list(struct page *page, struct zone *zone, |
961 | unsigned int order, int migratetype) | |
962 | { | |
963 | struct free_area *area = &zone->free_area[order]; | |
964 | ||
293ffa5e | 965 | list_move_tail(&page->lru, &area->free_list[migratetype]); |
6ab01363 AD |
966 | } |
967 | ||
968 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
969 | unsigned int order) | |
970 | { | |
36e66c55 AD |
971 | /* clear reported state and update reported page count */ |
972 | if (page_reported(page)) | |
973 | __ClearPageReported(page); | |
974 | ||
6ab01363 AD |
975 | list_del(&page->lru); |
976 | __ClearPageBuddy(page); | |
977 | set_page_private(page, 0); | |
978 | zone->free_area[order].nr_free--; | |
979 | } | |
980 | ||
a2129f24 AD |
981 | /* |
982 | * If this is not the largest possible page, check if the buddy | |
983 | * of the next-highest order is free. If it is, it's possible | |
984 | * that pages are being freed that will coalesce soon. In case, | |
985 | * that is happening, add the free page to the tail of the list | |
986 | * so it's less likely to be used soon and more likely to be merged | |
987 | * as a higher order page | |
988 | */ | |
989 | static inline bool | |
990 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
991 | struct page *page, unsigned int order) | |
992 | { | |
993 | struct page *higher_page, *higher_buddy; | |
994 | unsigned long combined_pfn; | |
995 | ||
996 | if (order >= MAX_ORDER - 2) | |
997 | return false; | |
998 | ||
999 | if (!pfn_valid_within(buddy_pfn)) | |
1000 | return false; | |
1001 | ||
1002 | combined_pfn = buddy_pfn & pfn; | |
1003 | higher_page = page + (combined_pfn - pfn); | |
1004 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
1005 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
1006 | ||
1007 | return pfn_valid_within(buddy_pfn) && | |
1008 | page_is_buddy(higher_page, higher_buddy, order + 1); | |
1009 | } | |
1010 | ||
1da177e4 LT |
1011 | /* |
1012 | * Freeing function for a buddy system allocator. | |
1013 | * | |
1014 | * The concept of a buddy system is to maintain direct-mapped table | |
1015 | * (containing bit values) for memory blocks of various "orders". | |
1016 | * The bottom level table contains the map for the smallest allocatable | |
1017 | * units of memory (here, pages), and each level above it describes | |
1018 | * pairs of units from the levels below, hence, "buddies". | |
1019 | * At a high level, all that happens here is marking the table entry | |
1020 | * at the bottom level available, and propagating the changes upward | |
1021 | * as necessary, plus some accounting needed to play nicely with other | |
1022 | * parts of the VM system. | |
1023 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
1024 | * free pages of length of (1 << order) and marked with PageBuddy. |
1025 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 1026 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
1027 | * other. That is, if we allocate a small block, and both were |
1028 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 1029 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 1030 | * triggers coalescing into a block of larger size. |
1da177e4 | 1031 | * |
6d49e352 | 1032 | * -- nyc |
1da177e4 LT |
1033 | */ |
1034 | ||
48db57f8 | 1035 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 1036 | unsigned long pfn, |
ed0ae21d | 1037 | struct zone *zone, unsigned int order, |
f04a5d5d | 1038 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1039 | { |
a2129f24 | 1040 | struct capture_control *capc = task_capc(zone); |
3f649ab7 | 1041 | unsigned long buddy_pfn; |
a2129f24 | 1042 | unsigned long combined_pfn; |
d9dddbf5 | 1043 | unsigned int max_order; |
a2129f24 AD |
1044 | struct page *buddy; |
1045 | bool to_tail; | |
d9dddbf5 | 1046 | |
7ad69832 | 1047 | max_order = min_t(unsigned int, MAX_ORDER - 1, pageblock_order); |
1da177e4 | 1048 | |
d29bb978 | 1049 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 1050 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 1051 | |
ed0ae21d | 1052 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 1053 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 1054 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 1055 | |
76741e77 | 1056 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 1057 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 1058 | |
d9dddbf5 | 1059 | continue_merging: |
7ad69832 | 1060 | while (order < max_order) { |
5e1f0f09 MG |
1061 | if (compaction_capture(capc, page, order, migratetype)) { |
1062 | __mod_zone_freepage_state(zone, -(1 << order), | |
1063 | migratetype); | |
1064 | return; | |
1065 | } | |
76741e77 VB |
1066 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1067 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
1068 | |
1069 | if (!pfn_valid_within(buddy_pfn)) | |
1070 | goto done_merging; | |
cb2b95e1 | 1071 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 1072 | goto done_merging; |
c0a32fc5 SG |
1073 | /* |
1074 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
1075 | * merge with it and move up one order. | |
1076 | */ | |
b03641af | 1077 | if (page_is_guard(buddy)) |
2847cf95 | 1078 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 1079 | else |
6ab01363 | 1080 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
1081 | combined_pfn = buddy_pfn & pfn; |
1082 | page = page + (combined_pfn - pfn); | |
1083 | pfn = combined_pfn; | |
1da177e4 LT |
1084 | order++; |
1085 | } | |
7ad69832 | 1086 | if (order < MAX_ORDER - 1) { |
d9dddbf5 VB |
1087 | /* If we are here, it means order is >= pageblock_order. |
1088 | * We want to prevent merge between freepages on isolate | |
1089 | * pageblock and normal pageblock. Without this, pageblock | |
1090 | * isolation could cause incorrect freepage or CMA accounting. | |
1091 | * | |
1092 | * We don't want to hit this code for the more frequent | |
1093 | * low-order merging. | |
1094 | */ | |
1095 | if (unlikely(has_isolate_pageblock(zone))) { | |
1096 | int buddy_mt; | |
1097 | ||
76741e77 VB |
1098 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1099 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
1100 | buddy_mt = get_pageblock_migratetype(buddy); |
1101 | ||
1102 | if (migratetype != buddy_mt | |
1103 | && (is_migrate_isolate(migratetype) || | |
1104 | is_migrate_isolate(buddy_mt))) | |
1105 | goto done_merging; | |
1106 | } | |
7ad69832 | 1107 | max_order = order + 1; |
d9dddbf5 VB |
1108 | goto continue_merging; |
1109 | } | |
1110 | ||
1111 | done_merging: | |
ab130f91 | 1112 | set_buddy_order(page, order); |
6dda9d55 | 1113 | |
47b6a24a DH |
1114 | if (fpi_flags & FPI_TO_TAIL) |
1115 | to_tail = true; | |
1116 | else if (is_shuffle_order(order)) | |
a2129f24 | 1117 | to_tail = shuffle_pick_tail(); |
97500a4a | 1118 | else |
a2129f24 | 1119 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 1120 | |
a2129f24 | 1121 | if (to_tail) |
6ab01363 | 1122 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 1123 | else |
6ab01363 | 1124 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
1125 | |
1126 | /* Notify page reporting subsystem of freed page */ | |
f04a5d5d | 1127 | if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY)) |
36e66c55 | 1128 | page_reporting_notify_free(order); |
1da177e4 LT |
1129 | } |
1130 | ||
7bfec6f4 MG |
1131 | /* |
1132 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1133 | * try and check multiple fields with one check. The caller must do a detailed | |
1134 | * check if necessary. | |
1135 | */ | |
1136 | static inline bool page_expected_state(struct page *page, | |
1137 | unsigned long check_flags) | |
1138 | { | |
1139 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1140 | return false; | |
1141 | ||
1142 | if (unlikely((unsigned long)page->mapping | | |
1143 | page_ref_count(page) | | |
1144 | #ifdef CONFIG_MEMCG | |
48060834 | 1145 | page->memcg_data | |
7bfec6f4 MG |
1146 | #endif |
1147 | (page->flags & check_flags))) | |
1148 | return false; | |
1149 | ||
1150 | return true; | |
1151 | } | |
1152 | ||
58b7f119 | 1153 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 1154 | { |
82a3241a | 1155 | const char *bad_reason = NULL; |
f0b791a3 | 1156 | |
53f9263b | 1157 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1158 | bad_reason = "nonzero mapcount"; |
1159 | if (unlikely(page->mapping != NULL)) | |
1160 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1161 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1162 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
1163 | if (unlikely(page->flags & flags)) { |
1164 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
1165 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
1166 | else | |
1167 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 1168 | } |
9edad6ea | 1169 | #ifdef CONFIG_MEMCG |
48060834 | 1170 | if (unlikely(page->memcg_data)) |
9edad6ea JW |
1171 | bad_reason = "page still charged to cgroup"; |
1172 | #endif | |
58b7f119 WY |
1173 | return bad_reason; |
1174 | } | |
1175 | ||
1176 | static void check_free_page_bad(struct page *page) | |
1177 | { | |
1178 | bad_page(page, | |
1179 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
1180 | } |
1181 | ||
534fe5e3 | 1182 | static inline int check_free_page(struct page *page) |
bb552ac6 | 1183 | { |
da838d4f | 1184 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 1185 | return 0; |
bb552ac6 MG |
1186 | |
1187 | /* Something has gone sideways, find it */ | |
0d0c48a2 | 1188 | check_free_page_bad(page); |
7bfec6f4 | 1189 | return 1; |
1da177e4 LT |
1190 | } |
1191 | ||
4db7548c MG |
1192 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1193 | { | |
1194 | int ret = 1; | |
1195 | ||
1196 | /* | |
1197 | * We rely page->lru.next never has bit 0 set, unless the page | |
1198 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1199 | */ | |
1200 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1201 | ||
1202 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1203 | ret = 0; | |
1204 | goto out; | |
1205 | } | |
1206 | switch (page - head_page) { | |
1207 | case 1: | |
4da1984e | 1208 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c | 1209 | if (unlikely(compound_mapcount(page))) { |
82a3241a | 1210 | bad_page(page, "nonzero compound_mapcount"); |
4db7548c MG |
1211 | goto out; |
1212 | } | |
1213 | break; | |
1214 | case 2: | |
1215 | /* | |
1216 | * the second tail page: ->mapping is | |
fa3015b7 | 1217 | * deferred_list.next -- ignore value. |
4db7548c MG |
1218 | */ |
1219 | break; | |
1220 | default: | |
1221 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 1222 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
1223 | goto out; |
1224 | } | |
1225 | break; | |
1226 | } | |
1227 | if (unlikely(!PageTail(page))) { | |
82a3241a | 1228 | bad_page(page, "PageTail not set"); |
4db7548c MG |
1229 | goto out; |
1230 | } | |
1231 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 1232 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
1233 | goto out; |
1234 | } | |
1235 | ret = 0; | |
1236 | out: | |
1237 | page->mapping = NULL; | |
1238 | clear_compound_head(page); | |
1239 | return ret; | |
1240 | } | |
1241 | ||
6471384a AP |
1242 | static void kernel_init_free_pages(struct page *page, int numpages) |
1243 | { | |
1244 | int i; | |
1245 | ||
9e15afa5 QC |
1246 | /* s390's use of memset() could override KASAN redzones. */ |
1247 | kasan_disable_current(); | |
aa1ef4d7 | 1248 | for (i = 0; i < numpages; i++) { |
acb35b17 | 1249 | u8 tag = page_kasan_tag(page + i); |
aa1ef4d7 | 1250 | page_kasan_tag_reset(page + i); |
6471384a | 1251 | clear_highpage(page + i); |
acb35b17 | 1252 | page_kasan_tag_set(page + i, tag); |
aa1ef4d7 | 1253 | } |
9e15afa5 | 1254 | kasan_enable_current(); |
6471384a AP |
1255 | } |
1256 | ||
e2769dbd | 1257 | static __always_inline bool free_pages_prepare(struct page *page, |
2c335680 | 1258 | unsigned int order, bool check_free, fpi_t fpi_flags) |
4db7548c | 1259 | { |
e2769dbd | 1260 | int bad = 0; |
1bb5eab3 | 1261 | bool init; |
4db7548c | 1262 | |
4db7548c MG |
1263 | VM_BUG_ON_PAGE(PageTail(page), page); |
1264 | ||
e2769dbd | 1265 | trace_mm_page_free(page, order); |
e2769dbd | 1266 | |
79f5f8fa OS |
1267 | if (unlikely(PageHWPoison(page)) && !order) { |
1268 | /* | |
1269 | * Do not let hwpoison pages hit pcplists/buddy | |
1270 | * Untie memcg state and reset page's owner | |
1271 | */ | |
18b2db3b | 1272 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
79f5f8fa OS |
1273 | __memcg_kmem_uncharge_page(page, order); |
1274 | reset_page_owner(page, order); | |
1275 | return false; | |
1276 | } | |
1277 | ||
e2769dbd MG |
1278 | /* |
1279 | * Check tail pages before head page information is cleared to | |
1280 | * avoid checking PageCompound for order-0 pages. | |
1281 | */ | |
1282 | if (unlikely(order)) { | |
1283 | bool compound = PageCompound(page); | |
1284 | int i; | |
1285 | ||
1286 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1287 | |
9a73f61b KS |
1288 | if (compound) |
1289 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1290 | for (i = 1; i < (1 << order); i++) { |
1291 | if (compound) | |
1292 | bad += free_tail_pages_check(page, page + i); | |
534fe5e3 | 1293 | if (unlikely(check_free_page(page + i))) { |
e2769dbd MG |
1294 | bad++; |
1295 | continue; | |
1296 | } | |
1297 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1298 | } | |
1299 | } | |
bda807d4 | 1300 | if (PageMappingFlags(page)) |
4db7548c | 1301 | page->mapping = NULL; |
18b2db3b | 1302 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
f4b00eab | 1303 | __memcg_kmem_uncharge_page(page, order); |
e2769dbd | 1304 | if (check_free) |
534fe5e3 | 1305 | bad += check_free_page(page); |
e2769dbd MG |
1306 | if (bad) |
1307 | return false; | |
4db7548c | 1308 | |
e2769dbd MG |
1309 | page_cpupid_reset_last(page); |
1310 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1311 | reset_page_owner(page, order); | |
4db7548c MG |
1312 | |
1313 | if (!PageHighMem(page)) { | |
1314 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1315 | PAGE_SIZE << order); |
4db7548c | 1316 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1317 | PAGE_SIZE << order); |
4db7548c | 1318 | } |
6471384a | 1319 | |
8db26a3d VB |
1320 | kernel_poison_pages(page, 1 << order); |
1321 | ||
f9d79e8d | 1322 | /* |
1bb5eab3 AK |
1323 | * As memory initialization might be integrated into KASAN, |
1324 | * kasan_free_pages and kernel_init_free_pages must be | |
1325 | * kept together to avoid discrepancies in behavior. | |
1326 | * | |
f9d79e8d AK |
1327 | * With hardware tag-based KASAN, memory tags must be set before the |
1328 | * page becomes unavailable via debug_pagealloc or arch_free_page. | |
1329 | */ | |
1bb5eab3 AK |
1330 | init = want_init_on_free(); |
1331 | if (init && !kasan_has_integrated_init()) | |
1332 | kernel_init_free_pages(page, 1 << order); | |
1333 | kasan_free_nondeferred_pages(page, order, init, fpi_flags); | |
f9d79e8d | 1334 | |
234fdce8 QC |
1335 | /* |
1336 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1337 | * does this. So nothing which can access the page's contents should | |
1338 | * happen after this. | |
1339 | */ | |
1340 | arch_free_page(page, order); | |
1341 | ||
77bc7fd6 | 1342 | debug_pagealloc_unmap_pages(page, 1 << order); |
d6332692 | 1343 | |
4db7548c MG |
1344 | return true; |
1345 | } | |
1346 | ||
e2769dbd | 1347 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1348 | /* |
1349 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1350 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1351 | * moved from pcp lists to free lists. | |
1352 | */ | |
1353 | static bool free_pcp_prepare(struct page *page) | |
e2769dbd | 1354 | { |
2c335680 | 1355 | return free_pages_prepare(page, 0, true, FPI_NONE); |
e2769dbd MG |
1356 | } |
1357 | ||
4462b32c | 1358 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1359 | { |
8e57f8ac | 1360 | if (debug_pagealloc_enabled_static()) |
534fe5e3 | 1361 | return check_free_page(page); |
4462b32c VB |
1362 | else |
1363 | return false; | |
e2769dbd MG |
1364 | } |
1365 | #else | |
4462b32c VB |
1366 | /* |
1367 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1368 | * moving from pcp lists to free list in order to reduce overhead. With | |
1369 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1370 | * to the pcp lists. | |
1371 | */ | |
e2769dbd MG |
1372 | static bool free_pcp_prepare(struct page *page) |
1373 | { | |
8e57f8ac | 1374 | if (debug_pagealloc_enabled_static()) |
2c335680 | 1375 | return free_pages_prepare(page, 0, true, FPI_NONE); |
4462b32c | 1376 | else |
2c335680 | 1377 | return free_pages_prepare(page, 0, false, FPI_NONE); |
e2769dbd MG |
1378 | } |
1379 | ||
4db7548c MG |
1380 | static bool bulkfree_pcp_prepare(struct page *page) |
1381 | { | |
534fe5e3 | 1382 | return check_free_page(page); |
4db7548c MG |
1383 | } |
1384 | #endif /* CONFIG_DEBUG_VM */ | |
1385 | ||
97334162 AL |
1386 | static inline void prefetch_buddy(struct page *page) |
1387 | { | |
1388 | unsigned long pfn = page_to_pfn(page); | |
1389 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1390 | struct page *buddy = page + (buddy_pfn - pfn); | |
1391 | ||
1392 | prefetch(buddy); | |
1393 | } | |
1394 | ||
1da177e4 | 1395 | /* |
5f8dcc21 | 1396 | * Frees a number of pages from the PCP lists |
1da177e4 | 1397 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1398 | * count is the number of pages to free. |
1da177e4 LT |
1399 | * |
1400 | * If the zone was previously in an "all pages pinned" state then look to | |
1401 | * see if this freeing clears that state. | |
1402 | * | |
1403 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1404 | * pinned" detection logic. | |
1405 | */ | |
5f8dcc21 MG |
1406 | static void free_pcppages_bulk(struct zone *zone, int count, |
1407 | struct per_cpu_pages *pcp) | |
1da177e4 | 1408 | { |
5f8dcc21 | 1409 | int migratetype = 0; |
a6f9edd6 | 1410 | int batch_free = 0; |
5c3ad2eb | 1411 | int prefetch_nr = READ_ONCE(pcp->batch); |
3777999d | 1412 | bool isolated_pageblocks; |
0a5f4e5b AL |
1413 | struct page *page, *tmp; |
1414 | LIST_HEAD(head); | |
f2260e6b | 1415 | |
88e8ac11 CTR |
1416 | /* |
1417 | * Ensure proper count is passed which otherwise would stuck in the | |
1418 | * below while (list_empty(list)) loop. | |
1419 | */ | |
1420 | count = min(pcp->count, count); | |
e5b31ac2 | 1421 | while (count) { |
5f8dcc21 MG |
1422 | struct list_head *list; |
1423 | ||
1424 | /* | |
a6f9edd6 MG |
1425 | * Remove pages from lists in a round-robin fashion. A |
1426 | * batch_free count is maintained that is incremented when an | |
1427 | * empty list is encountered. This is so more pages are freed | |
1428 | * off fuller lists instead of spinning excessively around empty | |
1429 | * lists | |
5f8dcc21 MG |
1430 | */ |
1431 | do { | |
a6f9edd6 | 1432 | batch_free++; |
5f8dcc21 MG |
1433 | if (++migratetype == MIGRATE_PCPTYPES) |
1434 | migratetype = 0; | |
1435 | list = &pcp->lists[migratetype]; | |
1436 | } while (list_empty(list)); | |
48db57f8 | 1437 | |
1d16871d NK |
1438 | /* This is the only non-empty list. Free them all. */ |
1439 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1440 | batch_free = count; |
1d16871d | 1441 | |
a6f9edd6 | 1442 | do { |
a16601c5 | 1443 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1444 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1445 | list_del(&page->lru); |
77ba9062 | 1446 | pcp->count--; |
aa016d14 | 1447 | |
4db7548c MG |
1448 | if (bulkfree_pcp_prepare(page)) |
1449 | continue; | |
1450 | ||
0a5f4e5b | 1451 | list_add_tail(&page->lru, &head); |
97334162 AL |
1452 | |
1453 | /* | |
1454 | * We are going to put the page back to the global | |
1455 | * pool, prefetch its buddy to speed up later access | |
1456 | * under zone->lock. It is believed the overhead of | |
1457 | * an additional test and calculating buddy_pfn here | |
1458 | * can be offset by reduced memory latency later. To | |
1459 | * avoid excessive prefetching due to large count, only | |
1460 | * prefetch buddy for the first pcp->batch nr of pages. | |
1461 | */ | |
5c3ad2eb | 1462 | if (prefetch_nr) { |
97334162 | 1463 | prefetch_buddy(page); |
5c3ad2eb VB |
1464 | prefetch_nr--; |
1465 | } | |
e5b31ac2 | 1466 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1467 | } |
0a5f4e5b | 1468 | |
dbbee9d5 MG |
1469 | /* |
1470 | * local_lock_irq held so equivalent to spin_lock_irqsave for | |
1471 | * both PREEMPT_RT and non-PREEMPT_RT configurations. | |
1472 | */ | |
0a5f4e5b AL |
1473 | spin_lock(&zone->lock); |
1474 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1475 | ||
1476 | /* | |
1477 | * Use safe version since after __free_one_page(), | |
1478 | * page->lru.next will not point to original list. | |
1479 | */ | |
1480 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1481 | int mt = get_pcppage_migratetype(page); | |
1482 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1483 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1484 | /* Pageblock could have been isolated meanwhile */ | |
1485 | if (unlikely(isolated_pageblocks)) | |
1486 | mt = get_pageblock_migratetype(page); | |
1487 | ||
f04a5d5d | 1488 | __free_one_page(page, page_to_pfn(page), zone, 0, mt, FPI_NONE); |
0a5f4e5b AL |
1489 | trace_mm_page_pcpu_drain(page, 0, mt); |
1490 | } | |
d34b0733 | 1491 | spin_unlock(&zone->lock); |
1da177e4 LT |
1492 | } |
1493 | ||
dc4b0caf MG |
1494 | static void free_one_page(struct zone *zone, |
1495 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1496 | unsigned int order, |
7fef431b | 1497 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1498 | { |
df1acc85 MG |
1499 | unsigned long flags; |
1500 | ||
1501 | spin_lock_irqsave(&zone->lock, flags); | |
ad53f92e JK |
1502 | if (unlikely(has_isolate_pageblock(zone) || |
1503 | is_migrate_isolate(migratetype))) { | |
1504 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1505 | } |
7fef431b | 1506 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); |
df1acc85 | 1507 | spin_unlock_irqrestore(&zone->lock, flags); |
48db57f8 NP |
1508 | } |
1509 | ||
1e8ce83c | 1510 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1511 | unsigned long zone, int nid) |
1e8ce83c | 1512 | { |
d0dc12e8 | 1513 | mm_zero_struct_page(page); |
1e8ce83c | 1514 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1515 | init_page_count(page); |
1516 | page_mapcount_reset(page); | |
1517 | page_cpupid_reset_last(page); | |
2813b9c0 | 1518 | page_kasan_tag_reset(page); |
1e8ce83c | 1519 | |
1e8ce83c RH |
1520 | INIT_LIST_HEAD(&page->lru); |
1521 | #ifdef WANT_PAGE_VIRTUAL | |
1522 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1523 | if (!is_highmem_idx(zone)) | |
1524 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1525 | #endif | |
1526 | } | |
1527 | ||
7e18adb4 | 1528 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1529 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1530 | { |
1531 | pg_data_t *pgdat; | |
1532 | int nid, zid; | |
1533 | ||
1534 | if (!early_page_uninitialised(pfn)) | |
1535 | return; | |
1536 | ||
1537 | nid = early_pfn_to_nid(pfn); | |
1538 | pgdat = NODE_DATA(nid); | |
1539 | ||
1540 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1541 | struct zone *zone = &pgdat->node_zones[zid]; | |
1542 | ||
1543 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1544 | break; | |
1545 | } | |
d0dc12e8 | 1546 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1547 | } |
1548 | #else | |
1549 | static inline void init_reserved_page(unsigned long pfn) | |
1550 | { | |
1551 | } | |
1552 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1553 | ||
92923ca3 NZ |
1554 | /* |
1555 | * Initialised pages do not have PageReserved set. This function is | |
1556 | * called for each range allocated by the bootmem allocator and | |
1557 | * marks the pages PageReserved. The remaining valid pages are later | |
1558 | * sent to the buddy page allocator. | |
1559 | */ | |
4b50bcc7 | 1560 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1561 | { |
1562 | unsigned long start_pfn = PFN_DOWN(start); | |
1563 | unsigned long end_pfn = PFN_UP(end); | |
1564 | ||
7e18adb4 MG |
1565 | for (; start_pfn < end_pfn; start_pfn++) { |
1566 | if (pfn_valid(start_pfn)) { | |
1567 | struct page *page = pfn_to_page(start_pfn); | |
1568 | ||
1569 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1570 | |
1571 | /* Avoid false-positive PageTail() */ | |
1572 | INIT_LIST_HEAD(&page->lru); | |
1573 | ||
d483da5b AD |
1574 | /* |
1575 | * no need for atomic set_bit because the struct | |
1576 | * page is not visible yet so nobody should | |
1577 | * access it yet. | |
1578 | */ | |
1579 | __SetPageReserved(page); | |
7e18adb4 MG |
1580 | } |
1581 | } | |
92923ca3 NZ |
1582 | } |
1583 | ||
7fef431b DH |
1584 | static void __free_pages_ok(struct page *page, unsigned int order, |
1585 | fpi_t fpi_flags) | |
ec95f53a | 1586 | { |
d34b0733 | 1587 | unsigned long flags; |
95e34412 | 1588 | int migratetype; |
dc4b0caf | 1589 | unsigned long pfn = page_to_pfn(page); |
56f0e661 | 1590 | struct zone *zone = page_zone(page); |
ec95f53a | 1591 | |
2c335680 | 1592 | if (!free_pages_prepare(page, order, true, fpi_flags)) |
ec95f53a KM |
1593 | return; |
1594 | ||
cfc47a28 | 1595 | migratetype = get_pfnblock_migratetype(page, pfn); |
dbbee9d5 | 1596 | |
56f0e661 | 1597 | spin_lock_irqsave(&zone->lock, flags); |
56f0e661 MG |
1598 | if (unlikely(has_isolate_pageblock(zone) || |
1599 | is_migrate_isolate(migratetype))) { | |
1600 | migratetype = get_pfnblock_migratetype(page, pfn); | |
1601 | } | |
1602 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); | |
1603 | spin_unlock_irqrestore(&zone->lock, flags); | |
90249993 MG |
1604 | |
1605 | __count_vm_events(PGFREE, 1 << order); | |
1da177e4 LT |
1606 | } |
1607 | ||
a9cd410a | 1608 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1609 | { |
c3993076 | 1610 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1611 | struct page *p = page; |
c3993076 | 1612 | unsigned int loop; |
a226f6c8 | 1613 | |
7fef431b DH |
1614 | /* |
1615 | * When initializing the memmap, __init_single_page() sets the refcount | |
1616 | * of all pages to 1 ("allocated"/"not free"). We have to set the | |
1617 | * refcount of all involved pages to 0. | |
1618 | */ | |
e2d0bd2b YL |
1619 | prefetchw(p); |
1620 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1621 | prefetchw(p + 1); | |
c3993076 JW |
1622 | __ClearPageReserved(p); |
1623 | set_page_count(p, 0); | |
a226f6c8 | 1624 | } |
e2d0bd2b YL |
1625 | __ClearPageReserved(p); |
1626 | set_page_count(p, 0); | |
c3993076 | 1627 | |
9705bea5 | 1628 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
7fef431b DH |
1629 | |
1630 | /* | |
1631 | * Bypass PCP and place fresh pages right to the tail, primarily | |
1632 | * relevant for memory onlining. | |
1633 | */ | |
2c335680 | 1634 | __free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON); |
a226f6c8 DH |
1635 | } |
1636 | ||
3f08a302 | 1637 | #ifdef CONFIG_NEED_MULTIPLE_NODES |
7ace9917 | 1638 | |
03e92a5e MR |
1639 | /* |
1640 | * During memory init memblocks map pfns to nids. The search is expensive and | |
1641 | * this caches recent lookups. The implementation of __early_pfn_to_nid | |
1642 | * treats start/end as pfns. | |
1643 | */ | |
1644 | struct mminit_pfnnid_cache { | |
1645 | unsigned long last_start; | |
1646 | unsigned long last_end; | |
1647 | int last_nid; | |
1648 | }; | |
75a592a4 | 1649 | |
03e92a5e | 1650 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
6f24fbd3 MR |
1651 | |
1652 | /* | |
1653 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
1654 | */ | |
03e92a5e | 1655 | static int __meminit __early_pfn_to_nid(unsigned long pfn, |
6f24fbd3 | 1656 | struct mminit_pfnnid_cache *state) |
75a592a4 | 1657 | { |
6f24fbd3 | 1658 | unsigned long start_pfn, end_pfn; |
75a592a4 MG |
1659 | int nid; |
1660 | ||
6f24fbd3 MR |
1661 | if (state->last_start <= pfn && pfn < state->last_end) |
1662 | return state->last_nid; | |
1663 | ||
1664 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); | |
1665 | if (nid != NUMA_NO_NODE) { | |
1666 | state->last_start = start_pfn; | |
1667 | state->last_end = end_pfn; | |
1668 | state->last_nid = nid; | |
1669 | } | |
7ace9917 MG |
1670 | |
1671 | return nid; | |
75a592a4 | 1672 | } |
75a592a4 | 1673 | |
75a592a4 | 1674 | int __meminit early_pfn_to_nid(unsigned long pfn) |
75a592a4 | 1675 | { |
7ace9917 | 1676 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1677 | int nid; |
1678 | ||
7ace9917 | 1679 | spin_lock(&early_pfn_lock); |
56ec43d8 | 1680 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1681 | if (nid < 0) |
e4568d38 | 1682 | nid = first_online_node; |
7ace9917 | 1683 | spin_unlock(&early_pfn_lock); |
75a592a4 | 1684 | |
7ace9917 | 1685 | return nid; |
75a592a4 | 1686 | } |
3f08a302 | 1687 | #endif /* CONFIG_NEED_MULTIPLE_NODES */ |
75a592a4 | 1688 | |
7c2ee349 | 1689 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1690 | unsigned int order) |
1691 | { | |
1692 | if (early_page_uninitialised(pfn)) | |
1693 | return; | |
a9cd410a | 1694 | __free_pages_core(page, order); |
3a80a7fa MG |
1695 | } |
1696 | ||
7cf91a98 JK |
1697 | /* |
1698 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1699 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1700 | * with the migration of free compaction scanner. The scanners then need to | |
1701 | * use only pfn_valid_within() check for arches that allow holes within | |
1702 | * pageblocks. | |
1703 | * | |
1704 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1705 | * | |
1706 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1707 | * i.e. it's possible that all pages within a zones range of pages do not | |
1708 | * belong to a single zone. We assume that a border between node0 and node1 | |
1709 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1710 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1711 | * the first and last page of a pageblock and avoid checking each individual | |
1712 | * page in a pageblock. | |
1713 | */ | |
1714 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1715 | unsigned long end_pfn, struct zone *zone) | |
1716 | { | |
1717 | struct page *start_page; | |
1718 | struct page *end_page; | |
1719 | ||
1720 | /* end_pfn is one past the range we are checking */ | |
1721 | end_pfn--; | |
1722 | ||
1723 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1724 | return NULL; | |
1725 | ||
2d070eab MH |
1726 | start_page = pfn_to_online_page(start_pfn); |
1727 | if (!start_page) | |
1728 | return NULL; | |
7cf91a98 JK |
1729 | |
1730 | if (page_zone(start_page) != zone) | |
1731 | return NULL; | |
1732 | ||
1733 | end_page = pfn_to_page(end_pfn); | |
1734 | ||
1735 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1736 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1737 | return NULL; | |
1738 | ||
1739 | return start_page; | |
1740 | } | |
1741 | ||
1742 | void set_zone_contiguous(struct zone *zone) | |
1743 | { | |
1744 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1745 | unsigned long block_end_pfn; | |
1746 | ||
1747 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1748 | for (; block_start_pfn < zone_end_pfn(zone); | |
1749 | block_start_pfn = block_end_pfn, | |
1750 | block_end_pfn += pageblock_nr_pages) { | |
1751 | ||
1752 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1753 | ||
1754 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1755 | block_end_pfn, zone)) | |
1756 | return; | |
e84fe99b | 1757 | cond_resched(); |
7cf91a98 JK |
1758 | } |
1759 | ||
1760 | /* We confirm that there is no hole */ | |
1761 | zone->contiguous = true; | |
1762 | } | |
1763 | ||
1764 | void clear_zone_contiguous(struct zone *zone) | |
1765 | { | |
1766 | zone->contiguous = false; | |
1767 | } | |
1768 | ||
7e18adb4 | 1769 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1770 | static void __init deferred_free_range(unsigned long pfn, |
1771 | unsigned long nr_pages) | |
a4de83dd | 1772 | { |
2f47a91f PT |
1773 | struct page *page; |
1774 | unsigned long i; | |
a4de83dd | 1775 | |
2f47a91f | 1776 | if (!nr_pages) |
a4de83dd MG |
1777 | return; |
1778 | ||
2f47a91f PT |
1779 | page = pfn_to_page(pfn); |
1780 | ||
a4de83dd | 1781 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1782 | if (nr_pages == pageblock_nr_pages && |
1783 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1784 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1785 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1786 | return; |
1787 | } | |
1788 | ||
e780149b XQ |
1789 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1790 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1791 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
a9cd410a | 1792 | __free_pages_core(page, 0); |
e780149b | 1793 | } |
a4de83dd MG |
1794 | } |
1795 | ||
d3cd131d NS |
1796 | /* Completion tracking for deferred_init_memmap() threads */ |
1797 | static atomic_t pgdat_init_n_undone __initdata; | |
1798 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1799 | ||
1800 | static inline void __init pgdat_init_report_one_done(void) | |
1801 | { | |
1802 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1803 | complete(&pgdat_init_all_done_comp); | |
1804 | } | |
0e1cc95b | 1805 | |
2f47a91f | 1806 | /* |
80b1f41c PT |
1807 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1808 | * | |
1809 | * First we check if pfn is valid on architectures where it is possible to have | |
1810 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1811 | * function is optimized out. | |
1812 | * | |
1813 | * Then, we check if a current large page is valid by only checking the validity | |
1814 | * of the head pfn. | |
2f47a91f | 1815 | */ |
56ec43d8 | 1816 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1817 | { |
80b1f41c PT |
1818 | if (!pfn_valid_within(pfn)) |
1819 | return false; | |
1820 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1821 | return false; | |
80b1f41c PT |
1822 | return true; |
1823 | } | |
2f47a91f | 1824 | |
80b1f41c PT |
1825 | /* |
1826 | * Free pages to buddy allocator. Try to free aligned pages in | |
1827 | * pageblock_nr_pages sizes. | |
1828 | */ | |
56ec43d8 | 1829 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1830 | unsigned long end_pfn) |
1831 | { | |
80b1f41c PT |
1832 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
1833 | unsigned long nr_free = 0; | |
2f47a91f | 1834 | |
80b1f41c | 1835 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1836 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1837 | deferred_free_range(pfn - nr_free, nr_free); |
1838 | nr_free = 0; | |
1839 | } else if (!(pfn & nr_pgmask)) { | |
1840 | deferred_free_range(pfn - nr_free, nr_free); | |
1841 | nr_free = 1; | |
80b1f41c PT |
1842 | } else { |
1843 | nr_free++; | |
1844 | } | |
1845 | } | |
1846 | /* Free the last block of pages to allocator */ | |
1847 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1848 | } |
1849 | ||
80b1f41c PT |
1850 | /* |
1851 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1852 | * by performing it only once every pageblock_nr_pages. | |
1853 | * Return number of pages initialized. | |
1854 | */ | |
56ec43d8 | 1855 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1856 | unsigned long pfn, |
1857 | unsigned long end_pfn) | |
2f47a91f | 1858 | { |
2f47a91f | 1859 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
56ec43d8 | 1860 | int nid = zone_to_nid(zone); |
2f47a91f | 1861 | unsigned long nr_pages = 0; |
56ec43d8 | 1862 | int zid = zone_idx(zone); |
2f47a91f | 1863 | struct page *page = NULL; |
2f47a91f | 1864 | |
80b1f41c | 1865 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1866 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1867 | page = NULL; |
2f47a91f | 1868 | continue; |
80b1f41c | 1869 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1870 | page = pfn_to_page(pfn); |
80b1f41c PT |
1871 | } else { |
1872 | page++; | |
2f47a91f | 1873 | } |
d0dc12e8 | 1874 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1875 | nr_pages++; |
2f47a91f | 1876 | } |
80b1f41c | 1877 | return (nr_pages); |
2f47a91f PT |
1878 | } |
1879 | ||
0e56acae AD |
1880 | /* |
1881 | * This function is meant to pre-load the iterator for the zone init. | |
1882 | * Specifically it walks through the ranges until we are caught up to the | |
1883 | * first_init_pfn value and exits there. If we never encounter the value we | |
1884 | * return false indicating there are no valid ranges left. | |
1885 | */ | |
1886 | static bool __init | |
1887 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
1888 | unsigned long *spfn, unsigned long *epfn, | |
1889 | unsigned long first_init_pfn) | |
1890 | { | |
1891 | u64 j; | |
1892 | ||
1893 | /* | |
1894 | * Start out by walking through the ranges in this zone that have | |
1895 | * already been initialized. We don't need to do anything with them | |
1896 | * so we just need to flush them out of the system. | |
1897 | */ | |
1898 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
1899 | if (*epfn <= first_init_pfn) | |
1900 | continue; | |
1901 | if (*spfn < first_init_pfn) | |
1902 | *spfn = first_init_pfn; | |
1903 | *i = j; | |
1904 | return true; | |
1905 | } | |
1906 | ||
1907 | return false; | |
1908 | } | |
1909 | ||
1910 | /* | |
1911 | * Initialize and free pages. We do it in two loops: first we initialize | |
1912 | * struct page, then free to buddy allocator, because while we are | |
1913 | * freeing pages we can access pages that are ahead (computing buddy | |
1914 | * page in __free_one_page()). | |
1915 | * | |
1916 | * In order to try and keep some memory in the cache we have the loop | |
1917 | * broken along max page order boundaries. This way we will not cause | |
1918 | * any issues with the buddy page computation. | |
1919 | */ | |
1920 | static unsigned long __init | |
1921 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
1922 | unsigned long *end_pfn) | |
1923 | { | |
1924 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
1925 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
1926 | unsigned long nr_pages = 0; | |
1927 | u64 j = *i; | |
1928 | ||
1929 | /* First we loop through and initialize the page values */ | |
1930 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
1931 | unsigned long t; | |
1932 | ||
1933 | if (mo_pfn <= *start_pfn) | |
1934 | break; | |
1935 | ||
1936 | t = min(mo_pfn, *end_pfn); | |
1937 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
1938 | ||
1939 | if (mo_pfn < *end_pfn) { | |
1940 | *start_pfn = mo_pfn; | |
1941 | break; | |
1942 | } | |
1943 | } | |
1944 | ||
1945 | /* Reset values and now loop through freeing pages as needed */ | |
1946 | swap(j, *i); | |
1947 | ||
1948 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
1949 | unsigned long t; | |
1950 | ||
1951 | if (mo_pfn <= spfn) | |
1952 | break; | |
1953 | ||
1954 | t = min(mo_pfn, epfn); | |
1955 | deferred_free_pages(spfn, t); | |
1956 | ||
1957 | if (mo_pfn <= epfn) | |
1958 | break; | |
1959 | } | |
1960 | ||
1961 | return nr_pages; | |
1962 | } | |
1963 | ||
e4443149 DJ |
1964 | static void __init |
1965 | deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, | |
1966 | void *arg) | |
1967 | { | |
1968 | unsigned long spfn, epfn; | |
1969 | struct zone *zone = arg; | |
1970 | u64 i; | |
1971 | ||
1972 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn); | |
1973 | ||
1974 | /* | |
1975 | * Initialize and free pages in MAX_ORDER sized increments so that we | |
1976 | * can avoid introducing any issues with the buddy allocator. | |
1977 | */ | |
1978 | while (spfn < end_pfn) { | |
1979 | deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
1980 | cond_resched(); | |
1981 | } | |
1982 | } | |
1983 | ||
ecd09650 DJ |
1984 | /* An arch may override for more concurrency. */ |
1985 | __weak int __init | |
1986 | deferred_page_init_max_threads(const struct cpumask *node_cpumask) | |
1987 | { | |
1988 | return 1; | |
1989 | } | |
1990 | ||
7e18adb4 | 1991 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1992 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1993 | { |
0e1cc95b | 1994 | pg_data_t *pgdat = data; |
0e56acae | 1995 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
89c7c402 | 1996 | unsigned long spfn = 0, epfn = 0; |
0e56acae | 1997 | unsigned long first_init_pfn, flags; |
7e18adb4 | 1998 | unsigned long start = jiffies; |
7e18adb4 | 1999 | struct zone *zone; |
e4443149 | 2000 | int zid, max_threads; |
2f47a91f | 2001 | u64 i; |
7e18adb4 | 2002 | |
3a2d7fa8 PT |
2003 | /* Bind memory initialisation thread to a local node if possible */ |
2004 | if (!cpumask_empty(cpumask)) | |
2005 | set_cpus_allowed_ptr(current, cpumask); | |
2006 | ||
2007 | pgdat_resize_lock(pgdat, &flags); | |
2008 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 2009 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 2010 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 2011 | pgdat_init_report_one_done(); |
0e1cc95b MG |
2012 | return 0; |
2013 | } | |
2014 | ||
7e18adb4 MG |
2015 | /* Sanity check boundaries */ |
2016 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
2017 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
2018 | pgdat->first_deferred_pfn = ULONG_MAX; | |
2019 | ||
3d060856 PT |
2020 | /* |
2021 | * Once we unlock here, the zone cannot be grown anymore, thus if an | |
2022 | * interrupt thread must allocate this early in boot, zone must be | |
2023 | * pre-grown prior to start of deferred page initialization. | |
2024 | */ | |
2025 | pgdat_resize_unlock(pgdat, &flags); | |
2026 | ||
7e18adb4 MG |
2027 | /* Only the highest zone is deferred so find it */ |
2028 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
2029 | zone = pgdat->node_zones + zid; | |
2030 | if (first_init_pfn < zone_end_pfn(zone)) | |
2031 | break; | |
2032 | } | |
0e56acae AD |
2033 | |
2034 | /* If the zone is empty somebody else may have cleared out the zone */ | |
2035 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2036 | first_init_pfn)) | |
2037 | goto zone_empty; | |
7e18adb4 | 2038 | |
ecd09650 | 2039 | max_threads = deferred_page_init_max_threads(cpumask); |
7e18adb4 | 2040 | |
117003c3 | 2041 | while (spfn < epfn) { |
e4443149 DJ |
2042 | unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION); |
2043 | struct padata_mt_job job = { | |
2044 | .thread_fn = deferred_init_memmap_chunk, | |
2045 | .fn_arg = zone, | |
2046 | .start = spfn, | |
2047 | .size = epfn_align - spfn, | |
2048 | .align = PAGES_PER_SECTION, | |
2049 | .min_chunk = PAGES_PER_SECTION, | |
2050 | .max_threads = max_threads, | |
2051 | }; | |
2052 | ||
2053 | padata_do_multithreaded(&job); | |
2054 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2055 | epfn_align); | |
117003c3 | 2056 | } |
0e56acae | 2057 | zone_empty: |
7e18adb4 MG |
2058 | /* Sanity check that the next zone really is unpopulated */ |
2059 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
2060 | ||
89c7c402 DJ |
2061 | pr_info("node %d deferred pages initialised in %ums\n", |
2062 | pgdat->node_id, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
2063 | |
2064 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
2065 | return 0; |
2066 | } | |
c9e97a19 | 2067 | |
c9e97a19 PT |
2068 | /* |
2069 | * If this zone has deferred pages, try to grow it by initializing enough | |
2070 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
2071 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
2072 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
2073 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
2074 | * | |
2075 | * Return true when zone was grown, otherwise return false. We return true even | |
2076 | * when we grow less than requested, to let the caller decide if there are | |
2077 | * enough pages to satisfy the allocation. | |
2078 | * | |
2079 | * Note: We use noinline because this function is needed only during boot, and | |
2080 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
2081 | * making sure that it is not inlined into permanent text section. | |
2082 | */ | |
2083 | static noinline bool __init | |
2084 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
2085 | { | |
c9e97a19 | 2086 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 2087 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 2088 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
2089 | unsigned long spfn, epfn, flags; |
2090 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
2091 | u64 i; |
2092 | ||
2093 | /* Only the last zone may have deferred pages */ | |
2094 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
2095 | return false; | |
2096 | ||
2097 | pgdat_resize_lock(pgdat, &flags); | |
2098 | ||
c9e97a19 PT |
2099 | /* |
2100 | * If someone grew this zone while we were waiting for spinlock, return | |
2101 | * true, as there might be enough pages already. | |
2102 | */ | |
2103 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
2104 | pgdat_resize_unlock(pgdat, &flags); | |
2105 | return true; | |
2106 | } | |
2107 | ||
0e56acae AD |
2108 | /* If the zone is empty somebody else may have cleared out the zone */ |
2109 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2110 | first_deferred_pfn)) { | |
2111 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 2112 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
2113 | /* Retry only once. */ |
2114 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
2115 | } |
2116 | ||
0e56acae AD |
2117 | /* |
2118 | * Initialize and free pages in MAX_ORDER sized increments so | |
2119 | * that we can avoid introducing any issues with the buddy | |
2120 | * allocator. | |
2121 | */ | |
2122 | while (spfn < epfn) { | |
2123 | /* update our first deferred PFN for this section */ | |
2124 | first_deferred_pfn = spfn; | |
2125 | ||
2126 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
117003c3 | 2127 | touch_nmi_watchdog(); |
c9e97a19 | 2128 | |
0e56acae AD |
2129 | /* We should only stop along section boundaries */ |
2130 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
2131 | continue; | |
c9e97a19 | 2132 | |
0e56acae | 2133 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
2134 | if (nr_pages >= nr_pages_needed) |
2135 | break; | |
2136 | } | |
2137 | ||
0e56acae | 2138 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
2139 | pgdat_resize_unlock(pgdat, &flags); |
2140 | ||
2141 | return nr_pages > 0; | |
2142 | } | |
2143 | ||
2144 | /* | |
2145 | * deferred_grow_zone() is __init, but it is called from | |
2146 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
2147 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
2148 | * and to ensure that the function body gets unloaded. | |
2149 | */ | |
2150 | static bool __ref | |
2151 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
2152 | { | |
2153 | return deferred_grow_zone(zone, order); | |
2154 | } | |
2155 | ||
7cf91a98 | 2156 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
2157 | |
2158 | void __init page_alloc_init_late(void) | |
2159 | { | |
7cf91a98 | 2160 | struct zone *zone; |
e900a918 | 2161 | int nid; |
7cf91a98 JK |
2162 | |
2163 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 2164 | |
d3cd131d NS |
2165 | /* There will be num_node_state(N_MEMORY) threads */ |
2166 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 2167 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
2168 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
2169 | } | |
2170 | ||
2171 | /* Block until all are initialised */ | |
d3cd131d | 2172 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 2173 | |
c9e97a19 PT |
2174 | /* |
2175 | * We initialized the rest of the deferred pages. Permanently disable | |
2176 | * on-demand struct page initialization. | |
2177 | */ | |
2178 | static_branch_disable(&deferred_pages); | |
2179 | ||
4248b0da MG |
2180 | /* Reinit limits that are based on free pages after the kernel is up */ |
2181 | files_maxfiles_init(); | |
7cf91a98 | 2182 | #endif |
350e88ba | 2183 | |
ba8f3587 LF |
2184 | buffer_init(); |
2185 | ||
3010f876 PT |
2186 | /* Discard memblock private memory */ |
2187 | memblock_discard(); | |
7cf91a98 | 2188 | |
e900a918 DW |
2189 | for_each_node_state(nid, N_MEMORY) |
2190 | shuffle_free_memory(NODE_DATA(nid)); | |
2191 | ||
7cf91a98 JK |
2192 | for_each_populated_zone(zone) |
2193 | set_zone_contiguous(zone); | |
7e18adb4 | 2194 | } |
7e18adb4 | 2195 | |
47118af0 | 2196 | #ifdef CONFIG_CMA |
9cf510a5 | 2197 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
2198 | void __init init_cma_reserved_pageblock(struct page *page) |
2199 | { | |
2200 | unsigned i = pageblock_nr_pages; | |
2201 | struct page *p = page; | |
2202 | ||
2203 | do { | |
2204 | __ClearPageReserved(p); | |
2205 | set_page_count(p, 0); | |
d883c6cf | 2206 | } while (++p, --i); |
47118af0 | 2207 | |
47118af0 | 2208 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
2209 | |
2210 | if (pageblock_order >= MAX_ORDER) { | |
2211 | i = pageblock_nr_pages; | |
2212 | p = page; | |
2213 | do { | |
2214 | set_page_refcounted(p); | |
2215 | __free_pages(p, MAX_ORDER - 1); | |
2216 | p += MAX_ORDER_NR_PAGES; | |
2217 | } while (i -= MAX_ORDER_NR_PAGES); | |
2218 | } else { | |
2219 | set_page_refcounted(page); | |
2220 | __free_pages(page, pageblock_order); | |
2221 | } | |
2222 | ||
3dcc0571 | 2223 | adjust_managed_page_count(page, pageblock_nr_pages); |
3c381db1 | 2224 | page_zone(page)->cma_pages += pageblock_nr_pages; |
47118af0 MN |
2225 | } |
2226 | #endif | |
1da177e4 LT |
2227 | |
2228 | /* | |
2229 | * The order of subdivision here is critical for the IO subsystem. | |
2230 | * Please do not alter this order without good reasons and regression | |
2231 | * testing. Specifically, as large blocks of memory are subdivided, | |
2232 | * the order in which smaller blocks are delivered depends on the order | |
2233 | * they're subdivided in this function. This is the primary factor | |
2234 | * influencing the order in which pages are delivered to the IO | |
2235 | * subsystem according to empirical testing, and this is also justified | |
2236 | * by considering the behavior of a buddy system containing a single | |
2237 | * large block of memory acted on by a series of small allocations. | |
2238 | * This behavior is a critical factor in sglist merging's success. | |
2239 | * | |
6d49e352 | 2240 | * -- nyc |
1da177e4 | 2241 | */ |
085cc7d5 | 2242 | static inline void expand(struct zone *zone, struct page *page, |
6ab01363 | 2243 | int low, int high, int migratetype) |
1da177e4 LT |
2244 | { |
2245 | unsigned long size = 1 << high; | |
2246 | ||
2247 | while (high > low) { | |
1da177e4 LT |
2248 | high--; |
2249 | size >>= 1; | |
309381fe | 2250 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2251 | |
acbc15a4 JK |
2252 | /* |
2253 | * Mark as guard pages (or page), that will allow to | |
2254 | * merge back to allocator when buddy will be freed. | |
2255 | * Corresponding page table entries will not be touched, | |
2256 | * pages will stay not present in virtual address space | |
2257 | */ | |
2258 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2259 | continue; |
acbc15a4 | 2260 | |
6ab01363 | 2261 | add_to_free_list(&page[size], zone, high, migratetype); |
ab130f91 | 2262 | set_buddy_order(&page[size], high); |
1da177e4 | 2263 | } |
1da177e4 LT |
2264 | } |
2265 | ||
4e611801 | 2266 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2267 | { |
f4c18e6f | 2268 | if (unlikely(page->flags & __PG_HWPOISON)) { |
e570f56c NH |
2269 | /* Don't complain about hwpoisoned pages */ |
2270 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2271 | return; | |
f4c18e6f | 2272 | } |
58b7f119 WY |
2273 | |
2274 | bad_page(page, | |
2275 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
4e611801 VB |
2276 | } |
2277 | ||
2278 | /* | |
2279 | * This page is about to be returned from the page allocator | |
2280 | */ | |
2281 | static inline int check_new_page(struct page *page) | |
2282 | { | |
2283 | if (likely(page_expected_state(page, | |
2284 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2285 | return 0; | |
2286 | ||
2287 | check_new_page_bad(page); | |
2288 | return 1; | |
2a7684a2 WF |
2289 | } |
2290 | ||
479f854a | 2291 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2292 | /* |
2293 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2294 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2295 | * also checked when pcp lists are refilled from the free lists. | |
2296 | */ | |
2297 | static inline bool check_pcp_refill(struct page *page) | |
479f854a | 2298 | { |
8e57f8ac | 2299 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2300 | return check_new_page(page); |
2301 | else | |
2302 | return false; | |
479f854a MG |
2303 | } |
2304 | ||
4462b32c | 2305 | static inline bool check_new_pcp(struct page *page) |
479f854a MG |
2306 | { |
2307 | return check_new_page(page); | |
2308 | } | |
2309 | #else | |
4462b32c VB |
2310 | /* |
2311 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2312 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2313 | * enabled, they are also checked when being allocated from the pcp lists. | |
2314 | */ | |
2315 | static inline bool check_pcp_refill(struct page *page) | |
479f854a MG |
2316 | { |
2317 | return check_new_page(page); | |
2318 | } | |
4462b32c | 2319 | static inline bool check_new_pcp(struct page *page) |
479f854a | 2320 | { |
8e57f8ac | 2321 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2322 | return check_new_page(page); |
2323 | else | |
2324 | return false; | |
479f854a MG |
2325 | } |
2326 | #endif /* CONFIG_DEBUG_VM */ | |
2327 | ||
2328 | static bool check_new_pages(struct page *page, unsigned int order) | |
2329 | { | |
2330 | int i; | |
2331 | for (i = 0; i < (1 << order); i++) { | |
2332 | struct page *p = page + i; | |
2333 | ||
2334 | if (unlikely(check_new_page(p))) | |
2335 | return true; | |
2336 | } | |
2337 | ||
2338 | return false; | |
2339 | } | |
2340 | ||
46f24fd8 JK |
2341 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2342 | gfp_t gfp_flags) | |
2343 | { | |
1bb5eab3 AK |
2344 | bool init; |
2345 | ||
46f24fd8 JK |
2346 | set_page_private(page, 0); |
2347 | set_page_refcounted(page); | |
2348 | ||
2349 | arch_alloc_page(page, order); | |
77bc7fd6 | 2350 | debug_pagealloc_map_pages(page, 1 << order); |
1bb5eab3 AK |
2351 | |
2352 | /* | |
2353 | * Page unpoisoning must happen before memory initialization. | |
2354 | * Otherwise, the poison pattern will be overwritten for __GFP_ZERO | |
2355 | * allocations and the page unpoisoning code will complain. | |
2356 | */ | |
8db26a3d | 2357 | kernel_unpoison_pages(page, 1 << order); |
862b6dee | 2358 | |
1bb5eab3 AK |
2359 | /* |
2360 | * As memory initialization might be integrated into KASAN, | |
2361 | * kasan_alloc_pages and kernel_init_free_pages must be | |
2362 | * kept together to avoid discrepancies in behavior. | |
2363 | */ | |
2364 | init = !want_init_on_free() && want_init_on_alloc(gfp_flags); | |
2365 | kasan_alloc_pages(page, order, init); | |
2366 | if (init && !kasan_has_integrated_init()) | |
862b6dee | 2367 | kernel_init_free_pages(page, 1 << order); |
1bb5eab3 AK |
2368 | |
2369 | set_page_owner(page, order, gfp_flags); | |
46f24fd8 JK |
2370 | } |
2371 | ||
479f854a | 2372 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2373 | unsigned int alloc_flags) |
2a7684a2 | 2374 | { |
46f24fd8 | 2375 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2376 | |
17cf4406 NP |
2377 | if (order && (gfp_flags & __GFP_COMP)) |
2378 | prep_compound_page(page, order); | |
2379 | ||
75379191 | 2380 | /* |
2f064f34 | 2381 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2382 | * allocate the page. The expectation is that the caller is taking |
2383 | * steps that will free more memory. The caller should avoid the page | |
2384 | * being used for !PFMEMALLOC purposes. | |
2385 | */ | |
2f064f34 MH |
2386 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2387 | set_page_pfmemalloc(page); | |
2388 | else | |
2389 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2390 | } |
2391 | ||
56fd56b8 MG |
2392 | /* |
2393 | * Go through the free lists for the given migratetype and remove | |
2394 | * the smallest available page from the freelists | |
2395 | */ | |
85ccc8fa | 2396 | static __always_inline |
728ec980 | 2397 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2398 | int migratetype) |
2399 | { | |
2400 | unsigned int current_order; | |
b8af2941 | 2401 | struct free_area *area; |
56fd56b8 MG |
2402 | struct page *page; |
2403 | ||
2404 | /* Find a page of the appropriate size in the preferred list */ | |
2405 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2406 | area = &(zone->free_area[current_order]); | |
b03641af | 2407 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2408 | if (!page) |
2409 | continue; | |
6ab01363 AD |
2410 | del_page_from_free_list(page, zone, current_order); |
2411 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 2412 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
2413 | return page; |
2414 | } | |
2415 | ||
2416 | return NULL; | |
2417 | } | |
2418 | ||
2419 | ||
b2a0ac88 MG |
2420 | /* |
2421 | * This array describes the order lists are fallen back to when | |
2422 | * the free lists for the desirable migrate type are depleted | |
2423 | */ | |
da415663 | 2424 | static int fallbacks[MIGRATE_TYPES][3] = { |
974a786e | 2425 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2426 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2427 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
47118af0 | 2428 | #ifdef CONFIG_CMA |
974a786e | 2429 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 2430 | #endif |
194159fb | 2431 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 2432 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 2433 | #endif |
b2a0ac88 MG |
2434 | }; |
2435 | ||
dc67647b | 2436 | #ifdef CONFIG_CMA |
85ccc8fa | 2437 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2438 | unsigned int order) |
2439 | { | |
2440 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2441 | } | |
2442 | #else | |
2443 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2444 | unsigned int order) { return NULL; } | |
2445 | #endif | |
2446 | ||
c361be55 | 2447 | /* |
293ffa5e | 2448 | * Move the free pages in a range to the freelist tail of the requested type. |
d9c23400 | 2449 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2450 | * boundary. If alignment is required, use move_freepages_block() |
2451 | */ | |
02aa0cdd | 2452 | static int move_freepages(struct zone *zone, |
39ddb991 | 2453 | unsigned long start_pfn, unsigned long end_pfn, |
02aa0cdd | 2454 | int migratetype, int *num_movable) |
c361be55 MG |
2455 | { |
2456 | struct page *page; | |
39ddb991 | 2457 | unsigned long pfn; |
d00181b9 | 2458 | unsigned int order; |
d100313f | 2459 | int pages_moved = 0; |
c361be55 | 2460 | |
39ddb991 KW |
2461 | for (pfn = start_pfn; pfn <= end_pfn;) { |
2462 | if (!pfn_valid_within(pfn)) { | |
2463 | pfn++; | |
c361be55 MG |
2464 | continue; |
2465 | } | |
2466 | ||
39ddb991 | 2467 | page = pfn_to_page(pfn); |
c361be55 | 2468 | if (!PageBuddy(page)) { |
02aa0cdd VB |
2469 | /* |
2470 | * We assume that pages that could be isolated for | |
2471 | * migration are movable. But we don't actually try | |
2472 | * isolating, as that would be expensive. | |
2473 | */ | |
2474 | if (num_movable && | |
2475 | (PageLRU(page) || __PageMovable(page))) | |
2476 | (*num_movable)++; | |
39ddb991 | 2477 | pfn++; |
c361be55 MG |
2478 | continue; |
2479 | } | |
2480 | ||
cd961038 DR |
2481 | /* Make sure we are not inadvertently changing nodes */ |
2482 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2483 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2484 | ||
ab130f91 | 2485 | order = buddy_order(page); |
6ab01363 | 2486 | move_to_free_list(page, zone, order, migratetype); |
39ddb991 | 2487 | pfn += 1 << order; |
d100313f | 2488 | pages_moved += 1 << order; |
c361be55 MG |
2489 | } |
2490 | ||
d100313f | 2491 | return pages_moved; |
c361be55 MG |
2492 | } |
2493 | ||
ee6f509c | 2494 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2495 | int migratetype, int *num_movable) |
c361be55 | 2496 | { |
39ddb991 | 2497 | unsigned long start_pfn, end_pfn, pfn; |
c361be55 | 2498 | |
4a222127 DR |
2499 | if (num_movable) |
2500 | *num_movable = 0; | |
2501 | ||
39ddb991 KW |
2502 | pfn = page_to_pfn(page); |
2503 | start_pfn = pfn & ~(pageblock_nr_pages - 1); | |
d9c23400 | 2504 | end_pfn = start_pfn + pageblock_nr_pages - 1; |
c361be55 MG |
2505 | |
2506 | /* Do not cross zone boundaries */ | |
108bcc96 | 2507 | if (!zone_spans_pfn(zone, start_pfn)) |
39ddb991 | 2508 | start_pfn = pfn; |
108bcc96 | 2509 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2510 | return 0; |
2511 | ||
39ddb991 | 2512 | return move_freepages(zone, start_pfn, end_pfn, migratetype, |
02aa0cdd | 2513 | num_movable); |
c361be55 MG |
2514 | } |
2515 | ||
2f66a68f MG |
2516 | static void change_pageblock_range(struct page *pageblock_page, |
2517 | int start_order, int migratetype) | |
2518 | { | |
2519 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2520 | ||
2521 | while (nr_pageblocks--) { | |
2522 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2523 | pageblock_page += pageblock_nr_pages; | |
2524 | } | |
2525 | } | |
2526 | ||
fef903ef | 2527 | /* |
9c0415eb VB |
2528 | * When we are falling back to another migratetype during allocation, try to |
2529 | * steal extra free pages from the same pageblocks to satisfy further | |
2530 | * allocations, instead of polluting multiple pageblocks. | |
2531 | * | |
2532 | * If we are stealing a relatively large buddy page, it is likely there will | |
2533 | * be more free pages in the pageblock, so try to steal them all. For | |
2534 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2535 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2536 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2537 | * pageblocks. | |
fef903ef | 2538 | */ |
4eb7dce6 JK |
2539 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2540 | { | |
2541 | /* | |
2542 | * Leaving this order check is intended, although there is | |
2543 | * relaxed order check in next check. The reason is that | |
2544 | * we can actually steal whole pageblock if this condition met, | |
2545 | * but, below check doesn't guarantee it and that is just heuristic | |
2546 | * so could be changed anytime. | |
2547 | */ | |
2548 | if (order >= pageblock_order) | |
2549 | return true; | |
2550 | ||
2551 | if (order >= pageblock_order / 2 || | |
2552 | start_mt == MIGRATE_RECLAIMABLE || | |
2553 | start_mt == MIGRATE_UNMOVABLE || | |
2554 | page_group_by_mobility_disabled) | |
2555 | return true; | |
2556 | ||
2557 | return false; | |
2558 | } | |
2559 | ||
597c8920 | 2560 | static inline bool boost_watermark(struct zone *zone) |
1c30844d MG |
2561 | { |
2562 | unsigned long max_boost; | |
2563 | ||
2564 | if (!watermark_boost_factor) | |
597c8920 | 2565 | return false; |
14f69140 HW |
2566 | /* |
2567 | * Don't bother in zones that are unlikely to produce results. | |
2568 | * On small machines, including kdump capture kernels running | |
2569 | * in a small area, boosting the watermark can cause an out of | |
2570 | * memory situation immediately. | |
2571 | */ | |
2572 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
597c8920 | 2573 | return false; |
1c30844d MG |
2574 | |
2575 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2576 | watermark_boost_factor, 10000); | |
94b3334c MG |
2577 | |
2578 | /* | |
2579 | * high watermark may be uninitialised if fragmentation occurs | |
2580 | * very early in boot so do not boost. We do not fall | |
2581 | * through and boost by pageblock_nr_pages as failing | |
2582 | * allocations that early means that reclaim is not going | |
2583 | * to help and it may even be impossible to reclaim the | |
2584 | * boosted watermark resulting in a hang. | |
2585 | */ | |
2586 | if (!max_boost) | |
597c8920 | 2587 | return false; |
94b3334c | 2588 | |
1c30844d MG |
2589 | max_boost = max(pageblock_nr_pages, max_boost); |
2590 | ||
2591 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2592 | max_boost); | |
597c8920 JW |
2593 | |
2594 | return true; | |
1c30844d MG |
2595 | } |
2596 | ||
4eb7dce6 JK |
2597 | /* |
2598 | * This function implements actual steal behaviour. If order is large enough, | |
2599 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2600 | * pageblock to our migratetype and determine how many already-allocated pages |
2601 | * are there in the pageblock with a compatible migratetype. If at least half | |
2602 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2603 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2604 | */ |
2605 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2606 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2607 | { |
ab130f91 | 2608 | unsigned int current_order = buddy_order(page); |
02aa0cdd VB |
2609 | int free_pages, movable_pages, alike_pages; |
2610 | int old_block_type; | |
2611 | ||
2612 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2613 | |
3bc48f96 VB |
2614 | /* |
2615 | * This can happen due to races and we want to prevent broken | |
2616 | * highatomic accounting. | |
2617 | */ | |
02aa0cdd | 2618 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2619 | goto single_page; |
2620 | ||
fef903ef SB |
2621 | /* Take ownership for orders >= pageblock_order */ |
2622 | if (current_order >= pageblock_order) { | |
2623 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2624 | goto single_page; |
fef903ef SB |
2625 | } |
2626 | ||
1c30844d MG |
2627 | /* |
2628 | * Boost watermarks to increase reclaim pressure to reduce the | |
2629 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2630 | * may be balanced overall and kswapd will not wake naturally. | |
2631 | */ | |
597c8920 | 2632 | if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD)) |
73444bc4 | 2633 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2634 | |
3bc48f96 VB |
2635 | /* We are not allowed to try stealing from the whole block */ |
2636 | if (!whole_block) | |
2637 | goto single_page; | |
2638 | ||
02aa0cdd VB |
2639 | free_pages = move_freepages_block(zone, page, start_type, |
2640 | &movable_pages); | |
2641 | /* | |
2642 | * Determine how many pages are compatible with our allocation. | |
2643 | * For movable allocation, it's the number of movable pages which | |
2644 | * we just obtained. For other types it's a bit more tricky. | |
2645 | */ | |
2646 | if (start_type == MIGRATE_MOVABLE) { | |
2647 | alike_pages = movable_pages; | |
2648 | } else { | |
2649 | /* | |
2650 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2651 | * to MOVABLE pageblock, consider all non-movable pages as | |
2652 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2653 | * vice versa, be conservative since we can't distinguish the | |
2654 | * exact migratetype of non-movable pages. | |
2655 | */ | |
2656 | if (old_block_type == MIGRATE_MOVABLE) | |
2657 | alike_pages = pageblock_nr_pages | |
2658 | - (free_pages + movable_pages); | |
2659 | else | |
2660 | alike_pages = 0; | |
2661 | } | |
2662 | ||
3bc48f96 | 2663 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2664 | if (!free_pages) |
3bc48f96 | 2665 | goto single_page; |
fef903ef | 2666 | |
02aa0cdd VB |
2667 | /* |
2668 | * If a sufficient number of pages in the block are either free or of | |
2669 | * comparable migratability as our allocation, claim the whole block. | |
2670 | */ | |
2671 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2672 | page_group_by_mobility_disabled) |
2673 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2674 | |
2675 | return; | |
2676 | ||
2677 | single_page: | |
6ab01363 | 2678 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
2679 | } |
2680 | ||
2149cdae JK |
2681 | /* |
2682 | * Check whether there is a suitable fallback freepage with requested order. | |
2683 | * If only_stealable is true, this function returns fallback_mt only if | |
2684 | * we can steal other freepages all together. This would help to reduce | |
2685 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2686 | */ | |
2687 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2688 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2689 | { |
2690 | int i; | |
2691 | int fallback_mt; | |
2692 | ||
2693 | if (area->nr_free == 0) | |
2694 | return -1; | |
2695 | ||
2696 | *can_steal = false; | |
2697 | for (i = 0;; i++) { | |
2698 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2699 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2700 | break; |
2701 | ||
b03641af | 2702 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2703 | continue; |
fef903ef | 2704 | |
4eb7dce6 JK |
2705 | if (can_steal_fallback(order, migratetype)) |
2706 | *can_steal = true; | |
2707 | ||
2149cdae JK |
2708 | if (!only_stealable) |
2709 | return fallback_mt; | |
2710 | ||
2711 | if (*can_steal) | |
2712 | return fallback_mt; | |
fef903ef | 2713 | } |
4eb7dce6 JK |
2714 | |
2715 | return -1; | |
fef903ef SB |
2716 | } |
2717 | ||
0aaa29a5 MG |
2718 | /* |
2719 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2720 | * there are no empty page blocks that contain a page with a suitable order | |
2721 | */ | |
2722 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2723 | unsigned int alloc_order) | |
2724 | { | |
2725 | int mt; | |
2726 | unsigned long max_managed, flags; | |
2727 | ||
2728 | /* | |
2729 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2730 | * Check is race-prone but harmless. | |
2731 | */ | |
9705bea5 | 2732 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2733 | if (zone->nr_reserved_highatomic >= max_managed) |
2734 | return; | |
2735 | ||
2736 | spin_lock_irqsave(&zone->lock, flags); | |
2737 | ||
2738 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2739 | if (zone->nr_reserved_highatomic >= max_managed) | |
2740 | goto out_unlock; | |
2741 | ||
2742 | /* Yoink! */ | |
2743 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2744 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2745 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2746 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2747 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2748 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2749 | } |
2750 | ||
2751 | out_unlock: | |
2752 | spin_unlock_irqrestore(&zone->lock, flags); | |
2753 | } | |
2754 | ||
2755 | /* | |
2756 | * Used when an allocation is about to fail under memory pressure. This | |
2757 | * potentially hurts the reliability of high-order allocations when under | |
2758 | * intense memory pressure but failed atomic allocations should be easier | |
2759 | * to recover from than an OOM. | |
29fac03b MK |
2760 | * |
2761 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2762 | * pageblock is exhausted. | |
0aaa29a5 | 2763 | */ |
29fac03b MK |
2764 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2765 | bool force) | |
0aaa29a5 MG |
2766 | { |
2767 | struct zonelist *zonelist = ac->zonelist; | |
2768 | unsigned long flags; | |
2769 | struct zoneref *z; | |
2770 | struct zone *zone; | |
2771 | struct page *page; | |
2772 | int order; | |
04c8716f | 2773 | bool ret; |
0aaa29a5 | 2774 | |
97a225e6 | 2775 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2776 | ac->nodemask) { |
29fac03b MK |
2777 | /* |
2778 | * Preserve at least one pageblock unless memory pressure | |
2779 | * is really high. | |
2780 | */ | |
2781 | if (!force && zone->nr_reserved_highatomic <= | |
2782 | pageblock_nr_pages) | |
0aaa29a5 MG |
2783 | continue; |
2784 | ||
2785 | spin_lock_irqsave(&zone->lock, flags); | |
2786 | for (order = 0; order < MAX_ORDER; order++) { | |
2787 | struct free_area *area = &(zone->free_area[order]); | |
2788 | ||
b03641af | 2789 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2790 | if (!page) |
0aaa29a5 MG |
2791 | continue; |
2792 | ||
0aaa29a5 | 2793 | /* |
4855e4a7 MK |
2794 | * In page freeing path, migratetype change is racy so |
2795 | * we can counter several free pages in a pageblock | |
f0953a1b | 2796 | * in this loop although we changed the pageblock type |
4855e4a7 MK |
2797 | * from highatomic to ac->migratetype. So we should |
2798 | * adjust the count once. | |
0aaa29a5 | 2799 | */ |
a6ffdc07 | 2800 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2801 | /* |
2802 | * It should never happen but changes to | |
2803 | * locking could inadvertently allow a per-cpu | |
2804 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2805 | * while unreserving so be safe and watch for | |
2806 | * underflows. | |
2807 | */ | |
2808 | zone->nr_reserved_highatomic -= min( | |
2809 | pageblock_nr_pages, | |
2810 | zone->nr_reserved_highatomic); | |
2811 | } | |
0aaa29a5 MG |
2812 | |
2813 | /* | |
2814 | * Convert to ac->migratetype and avoid the normal | |
2815 | * pageblock stealing heuristics. Minimally, the caller | |
2816 | * is doing the work and needs the pages. More | |
2817 | * importantly, if the block was always converted to | |
2818 | * MIGRATE_UNMOVABLE or another type then the number | |
2819 | * of pageblocks that cannot be completely freed | |
2820 | * may increase. | |
2821 | */ | |
2822 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2823 | ret = move_freepages_block(zone, page, ac->migratetype, |
2824 | NULL); | |
29fac03b MK |
2825 | if (ret) { |
2826 | spin_unlock_irqrestore(&zone->lock, flags); | |
2827 | return ret; | |
2828 | } | |
0aaa29a5 MG |
2829 | } |
2830 | spin_unlock_irqrestore(&zone->lock, flags); | |
2831 | } | |
04c8716f MK |
2832 | |
2833 | return false; | |
0aaa29a5 MG |
2834 | } |
2835 | ||
3bc48f96 VB |
2836 | /* |
2837 | * Try finding a free buddy page on the fallback list and put it on the free | |
2838 | * list of requested migratetype, possibly along with other pages from the same | |
2839 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2840 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2841 | * |
2842 | * The use of signed ints for order and current_order is a deliberate | |
2843 | * deviation from the rest of this file, to make the for loop | |
2844 | * condition simpler. | |
3bc48f96 | 2845 | */ |
85ccc8fa | 2846 | static __always_inline bool |
6bb15450 MG |
2847 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2848 | unsigned int alloc_flags) | |
b2a0ac88 | 2849 | { |
b8af2941 | 2850 | struct free_area *area; |
b002529d | 2851 | int current_order; |
6bb15450 | 2852 | int min_order = order; |
b2a0ac88 | 2853 | struct page *page; |
4eb7dce6 JK |
2854 | int fallback_mt; |
2855 | bool can_steal; | |
b2a0ac88 | 2856 | |
6bb15450 MG |
2857 | /* |
2858 | * Do not steal pages from freelists belonging to other pageblocks | |
2859 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2860 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2861 | */ | |
2862 | if (alloc_flags & ALLOC_NOFRAGMENT) | |
2863 | min_order = pageblock_order; | |
2864 | ||
7a8f58f3 VB |
2865 | /* |
2866 | * Find the largest available free page in the other list. This roughly | |
2867 | * approximates finding the pageblock with the most free pages, which | |
2868 | * would be too costly to do exactly. | |
2869 | */ | |
6bb15450 | 2870 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 2871 | --current_order) { |
4eb7dce6 JK |
2872 | area = &(zone->free_area[current_order]); |
2873 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2874 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2875 | if (fallback_mt == -1) |
2876 | continue; | |
b2a0ac88 | 2877 | |
7a8f58f3 VB |
2878 | /* |
2879 | * We cannot steal all free pages from the pageblock and the | |
2880 | * requested migratetype is movable. In that case it's better to | |
2881 | * steal and split the smallest available page instead of the | |
2882 | * largest available page, because even if the next movable | |
2883 | * allocation falls back into a different pageblock than this | |
2884 | * one, it won't cause permanent fragmentation. | |
2885 | */ | |
2886 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2887 | && current_order > order) | |
2888 | goto find_smallest; | |
b2a0ac88 | 2889 | |
7a8f58f3 VB |
2890 | goto do_steal; |
2891 | } | |
e0fff1bd | 2892 | |
7a8f58f3 | 2893 | return false; |
e0fff1bd | 2894 | |
7a8f58f3 VB |
2895 | find_smallest: |
2896 | for (current_order = order; current_order < MAX_ORDER; | |
2897 | current_order++) { | |
2898 | area = &(zone->free_area[current_order]); | |
2899 | fallback_mt = find_suitable_fallback(area, current_order, | |
2900 | start_migratetype, false, &can_steal); | |
2901 | if (fallback_mt != -1) | |
2902 | break; | |
b2a0ac88 MG |
2903 | } |
2904 | ||
7a8f58f3 VB |
2905 | /* |
2906 | * This should not happen - we already found a suitable fallback | |
2907 | * when looking for the largest page. | |
2908 | */ | |
2909 | VM_BUG_ON(current_order == MAX_ORDER); | |
2910 | ||
2911 | do_steal: | |
b03641af | 2912 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2913 | |
1c30844d MG |
2914 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
2915 | can_steal); | |
7a8f58f3 VB |
2916 | |
2917 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2918 | start_migratetype, fallback_mt); | |
2919 | ||
2920 | return true; | |
2921 | ||
b2a0ac88 MG |
2922 | } |
2923 | ||
56fd56b8 | 2924 | /* |
1da177e4 LT |
2925 | * Do the hard work of removing an element from the buddy allocator. |
2926 | * Call me with the zone->lock already held. | |
2927 | */ | |
85ccc8fa | 2928 | static __always_inline struct page * |
6bb15450 MG |
2929 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2930 | unsigned int alloc_flags) | |
1da177e4 | 2931 | { |
1da177e4 LT |
2932 | struct page *page; |
2933 | ||
ce8f86ee H |
2934 | if (IS_ENABLED(CONFIG_CMA)) { |
2935 | /* | |
2936 | * Balance movable allocations between regular and CMA areas by | |
2937 | * allocating from CMA when over half of the zone's free memory | |
2938 | * is in the CMA area. | |
2939 | */ | |
2940 | if (alloc_flags & ALLOC_CMA && | |
2941 | zone_page_state(zone, NR_FREE_CMA_PAGES) > | |
2942 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
2943 | page = __rmqueue_cma_fallback(zone, order); | |
2944 | if (page) | |
2945 | goto out; | |
2946 | } | |
16867664 | 2947 | } |
3bc48f96 | 2948 | retry: |
56fd56b8 | 2949 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2950 | if (unlikely(!page)) { |
8510e69c | 2951 | if (alloc_flags & ALLOC_CMA) |
dc67647b JK |
2952 | page = __rmqueue_cma_fallback(zone, order); |
2953 | ||
6bb15450 MG |
2954 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2955 | alloc_flags)) | |
3bc48f96 | 2956 | goto retry; |
728ec980 | 2957 | } |
ce8f86ee H |
2958 | out: |
2959 | if (page) | |
2960 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
b2a0ac88 | 2961 | return page; |
1da177e4 LT |
2962 | } |
2963 | ||
5f63b720 | 2964 | /* |
1da177e4 LT |
2965 | * Obtain a specified number of elements from the buddy allocator, all under |
2966 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2967 | * Returns the number of new pages which were placed at *list. | |
2968 | */ | |
5f63b720 | 2969 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2970 | unsigned long count, struct list_head *list, |
6bb15450 | 2971 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2972 | { |
cb66bede | 2973 | int i, allocated = 0; |
5f63b720 | 2974 | |
dbbee9d5 MG |
2975 | /* |
2976 | * local_lock_irq held so equivalent to spin_lock_irqsave for | |
2977 | * both PREEMPT_RT and non-PREEMPT_RT configurations. | |
2978 | */ | |
d34b0733 | 2979 | spin_lock(&zone->lock); |
1da177e4 | 2980 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2981 | struct page *page = __rmqueue(zone, order, migratetype, |
2982 | alloc_flags); | |
085cc7d5 | 2983 | if (unlikely(page == NULL)) |
1da177e4 | 2984 | break; |
81eabcbe | 2985 | |
479f854a MG |
2986 | if (unlikely(check_pcp_refill(page))) |
2987 | continue; | |
2988 | ||
81eabcbe | 2989 | /* |
0fac3ba5 VB |
2990 | * Split buddy pages returned by expand() are received here in |
2991 | * physical page order. The page is added to the tail of | |
2992 | * caller's list. From the callers perspective, the linked list | |
2993 | * is ordered by page number under some conditions. This is | |
2994 | * useful for IO devices that can forward direction from the | |
2995 | * head, thus also in the physical page order. This is useful | |
2996 | * for IO devices that can merge IO requests if the physical | |
2997 | * pages are ordered properly. | |
81eabcbe | 2998 | */ |
0fac3ba5 | 2999 | list_add_tail(&page->lru, list); |
cb66bede | 3000 | allocated++; |
bb14c2c7 | 3001 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
3002 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
3003 | -(1 << order)); | |
1da177e4 | 3004 | } |
a6de734b MG |
3005 | |
3006 | /* | |
3007 | * i pages were removed from the buddy list even if some leak due | |
3008 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
cb66bede | 3009 | * on i. Do not confuse with 'allocated' which is the number of |
a6de734b MG |
3010 | * pages added to the pcp list. |
3011 | */ | |
f2260e6b | 3012 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 3013 | spin_unlock(&zone->lock); |
cb66bede | 3014 | return allocated; |
1da177e4 LT |
3015 | } |
3016 | ||
4ae7c039 | 3017 | #ifdef CONFIG_NUMA |
8fce4d8e | 3018 | /* |
4037d452 CL |
3019 | * Called from the vmstat counter updater to drain pagesets of this |
3020 | * currently executing processor on remote nodes after they have | |
3021 | * expired. | |
3022 | * | |
879336c3 CL |
3023 | * Note that this function must be called with the thread pinned to |
3024 | * a single processor. | |
8fce4d8e | 3025 | */ |
4037d452 | 3026 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 3027 | { |
4ae7c039 | 3028 | unsigned long flags; |
7be12fc9 | 3029 | int to_drain, batch; |
4ae7c039 | 3030 | |
dbbee9d5 | 3031 | local_lock_irqsave(&pagesets.lock, flags); |
4db0c3c2 | 3032 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 3033 | to_drain = min(pcp->count, batch); |
77ba9062 | 3034 | if (to_drain > 0) |
2a13515c | 3035 | free_pcppages_bulk(zone, to_drain, pcp); |
dbbee9d5 | 3036 | local_unlock_irqrestore(&pagesets.lock, flags); |
4ae7c039 CL |
3037 | } |
3038 | #endif | |
3039 | ||
9f8f2172 | 3040 | /* |
93481ff0 | 3041 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
3042 | * |
3043 | * The processor must either be the current processor and the | |
3044 | * thread pinned to the current processor or a processor that | |
3045 | * is not online. | |
3046 | */ | |
93481ff0 | 3047 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 3048 | { |
c54ad30c | 3049 | unsigned long flags; |
93481ff0 | 3050 | struct per_cpu_pages *pcp; |
1da177e4 | 3051 | |
dbbee9d5 | 3052 | local_lock_irqsave(&pagesets.lock, flags); |
1da177e4 | 3053 | |
28f836b6 | 3054 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
77ba9062 | 3055 | if (pcp->count) |
93481ff0 | 3056 | free_pcppages_bulk(zone, pcp->count, pcp); |
28f836b6 | 3057 | |
dbbee9d5 | 3058 | local_unlock_irqrestore(&pagesets.lock, flags); |
93481ff0 | 3059 | } |
3dfa5721 | 3060 | |
93481ff0 VB |
3061 | /* |
3062 | * Drain pcplists of all zones on the indicated processor. | |
3063 | * | |
3064 | * The processor must either be the current processor and the | |
3065 | * thread pinned to the current processor or a processor that | |
3066 | * is not online. | |
3067 | */ | |
3068 | static void drain_pages(unsigned int cpu) | |
3069 | { | |
3070 | struct zone *zone; | |
3071 | ||
3072 | for_each_populated_zone(zone) { | |
3073 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
3074 | } |
3075 | } | |
1da177e4 | 3076 | |
9f8f2172 CL |
3077 | /* |
3078 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
3079 | * |
3080 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
3081 | * the single zone's pages. | |
9f8f2172 | 3082 | */ |
93481ff0 | 3083 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 3084 | { |
93481ff0 VB |
3085 | int cpu = smp_processor_id(); |
3086 | ||
3087 | if (zone) | |
3088 | drain_pages_zone(cpu, zone); | |
3089 | else | |
3090 | drain_pages(cpu); | |
9f8f2172 CL |
3091 | } |
3092 | ||
0ccce3b9 MG |
3093 | static void drain_local_pages_wq(struct work_struct *work) |
3094 | { | |
d9367bd0 WY |
3095 | struct pcpu_drain *drain; |
3096 | ||
3097 | drain = container_of(work, struct pcpu_drain, work); | |
3098 | ||
a459eeb7 MH |
3099 | /* |
3100 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
3101 | * we can race with cpu offline when the WQ can move this from | |
3102 | * a cpu pinned worker to an unbound one. We can operate on a different | |
f0953a1b | 3103 | * cpu which is alright but we also have to make sure to not move to |
a459eeb7 MH |
3104 | * a different one. |
3105 | */ | |
3106 | preempt_disable(); | |
d9367bd0 | 3107 | drain_local_pages(drain->zone); |
a459eeb7 | 3108 | preempt_enable(); |
0ccce3b9 MG |
3109 | } |
3110 | ||
9f8f2172 | 3111 | /* |
ec6e8c7e VB |
3112 | * The implementation of drain_all_pages(), exposing an extra parameter to |
3113 | * drain on all cpus. | |
93481ff0 | 3114 | * |
ec6e8c7e VB |
3115 | * drain_all_pages() is optimized to only execute on cpus where pcplists are |
3116 | * not empty. The check for non-emptiness can however race with a free to | |
3117 | * pcplist that has not yet increased the pcp->count from 0 to 1. Callers | |
3118 | * that need the guarantee that every CPU has drained can disable the | |
3119 | * optimizing racy check. | |
9f8f2172 | 3120 | */ |
3b1f3658 | 3121 | static void __drain_all_pages(struct zone *zone, bool force_all_cpus) |
9f8f2172 | 3122 | { |
74046494 | 3123 | int cpu; |
74046494 GBY |
3124 | |
3125 | /* | |
3126 | * Allocate in the BSS so we wont require allocation in | |
3127 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
3128 | */ | |
3129 | static cpumask_t cpus_with_pcps; | |
3130 | ||
ce612879 MH |
3131 | /* |
3132 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
3133 | * initialized. | |
3134 | */ | |
3135 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
3136 | return; | |
3137 | ||
bd233f53 MG |
3138 | /* |
3139 | * Do not drain if one is already in progress unless it's specific to | |
3140 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
3141 | * the drain to be complete when the call returns. | |
3142 | */ | |
3143 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
3144 | if (!zone) | |
3145 | return; | |
3146 | mutex_lock(&pcpu_drain_mutex); | |
3147 | } | |
0ccce3b9 | 3148 | |
74046494 GBY |
3149 | /* |
3150 | * We don't care about racing with CPU hotplug event | |
3151 | * as offline notification will cause the notified | |
3152 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
3153 | * disables preemption as part of its processing | |
3154 | */ | |
3155 | for_each_online_cpu(cpu) { | |
28f836b6 | 3156 | struct per_cpu_pages *pcp; |
93481ff0 | 3157 | struct zone *z; |
74046494 | 3158 | bool has_pcps = false; |
93481ff0 | 3159 | |
ec6e8c7e VB |
3160 | if (force_all_cpus) { |
3161 | /* | |
3162 | * The pcp.count check is racy, some callers need a | |
3163 | * guarantee that no cpu is missed. | |
3164 | */ | |
3165 | has_pcps = true; | |
3166 | } else if (zone) { | |
28f836b6 MG |
3167 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
3168 | if (pcp->count) | |
74046494 | 3169 | has_pcps = true; |
93481ff0 VB |
3170 | } else { |
3171 | for_each_populated_zone(z) { | |
28f836b6 MG |
3172 | pcp = per_cpu_ptr(z->per_cpu_pageset, cpu); |
3173 | if (pcp->count) { | |
93481ff0 VB |
3174 | has_pcps = true; |
3175 | break; | |
3176 | } | |
74046494 GBY |
3177 | } |
3178 | } | |
93481ff0 | 3179 | |
74046494 GBY |
3180 | if (has_pcps) |
3181 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
3182 | else | |
3183 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
3184 | } | |
0ccce3b9 | 3185 | |
bd233f53 | 3186 | for_each_cpu(cpu, &cpus_with_pcps) { |
d9367bd0 WY |
3187 | struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu); |
3188 | ||
3189 | drain->zone = zone; | |
3190 | INIT_WORK(&drain->work, drain_local_pages_wq); | |
3191 | queue_work_on(cpu, mm_percpu_wq, &drain->work); | |
0ccce3b9 | 3192 | } |
bd233f53 | 3193 | for_each_cpu(cpu, &cpus_with_pcps) |
d9367bd0 | 3194 | flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work); |
bd233f53 MG |
3195 | |
3196 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
3197 | } |
3198 | ||
ec6e8c7e VB |
3199 | /* |
3200 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. | |
3201 | * | |
3202 | * When zone parameter is non-NULL, spill just the single zone's pages. | |
3203 | * | |
3204 | * Note that this can be extremely slow as the draining happens in a workqueue. | |
3205 | */ | |
3206 | void drain_all_pages(struct zone *zone) | |
3207 | { | |
3208 | __drain_all_pages(zone, false); | |
3209 | } | |
3210 | ||
296699de | 3211 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3212 | |
556b969a CY |
3213 | /* |
3214 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
3215 | */ | |
3216 | #define WD_PAGE_COUNT (128*1024) | |
3217 | ||
1da177e4 LT |
3218 | void mark_free_pages(struct zone *zone) |
3219 | { | |
556b969a | 3220 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 3221 | unsigned long flags; |
7aeb09f9 | 3222 | unsigned int order, t; |
86760a2c | 3223 | struct page *page; |
1da177e4 | 3224 | |
8080fc03 | 3225 | if (zone_is_empty(zone)) |
1da177e4 LT |
3226 | return; |
3227 | ||
3228 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 3229 | |
108bcc96 | 3230 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
3231 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
3232 | if (pfn_valid(pfn)) { | |
86760a2c | 3233 | page = pfn_to_page(pfn); |
ba6b0979 | 3234 | |
556b969a CY |
3235 | if (!--page_count) { |
3236 | touch_nmi_watchdog(); | |
3237 | page_count = WD_PAGE_COUNT; | |
3238 | } | |
3239 | ||
ba6b0979 JK |
3240 | if (page_zone(page) != zone) |
3241 | continue; | |
3242 | ||
7be98234 RW |
3243 | if (!swsusp_page_is_forbidden(page)) |
3244 | swsusp_unset_page_free(page); | |
f623f0db | 3245 | } |
1da177e4 | 3246 | |
b2a0ac88 | 3247 | for_each_migratetype_order(order, t) { |
86760a2c GT |
3248 | list_for_each_entry(page, |
3249 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 3250 | unsigned long i; |
1da177e4 | 3251 | |
86760a2c | 3252 | pfn = page_to_pfn(page); |
556b969a CY |
3253 | for (i = 0; i < (1UL << order); i++) { |
3254 | if (!--page_count) { | |
3255 | touch_nmi_watchdog(); | |
3256 | page_count = WD_PAGE_COUNT; | |
3257 | } | |
7be98234 | 3258 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 3259 | } |
f623f0db | 3260 | } |
b2a0ac88 | 3261 | } |
1da177e4 LT |
3262 | spin_unlock_irqrestore(&zone->lock, flags); |
3263 | } | |
e2c55dc8 | 3264 | #endif /* CONFIG_PM */ |
1da177e4 | 3265 | |
2d4894b5 | 3266 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 3267 | { |
5f8dcc21 | 3268 | int migratetype; |
1da177e4 | 3269 | |
4db7548c | 3270 | if (!free_pcp_prepare(page)) |
9cca35d4 | 3271 | return false; |
689bcebf | 3272 | |
dc4b0caf | 3273 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 3274 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
3275 | return true; |
3276 | } | |
3277 | ||
3b12e7e9 MG |
3278 | static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch) |
3279 | { | |
3280 | int min_nr_free, max_nr_free; | |
3281 | ||
3282 | /* Check for PCP disabled or boot pageset */ | |
3283 | if (unlikely(high < batch)) | |
3284 | return 1; | |
3285 | ||
3286 | /* Leave at least pcp->batch pages on the list */ | |
3287 | min_nr_free = batch; | |
3288 | max_nr_free = high - batch; | |
3289 | ||
3290 | /* | |
3291 | * Double the number of pages freed each time there is subsequent | |
3292 | * freeing of pages without any allocation. | |
3293 | */ | |
3294 | batch <<= pcp->free_factor; | |
3295 | if (batch < max_nr_free) | |
3296 | pcp->free_factor++; | |
3297 | batch = clamp(batch, min_nr_free, max_nr_free); | |
3298 | ||
3299 | return batch; | |
3300 | } | |
3301 | ||
c49c2c47 MG |
3302 | static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone) |
3303 | { | |
3304 | int high = READ_ONCE(pcp->high); | |
3305 | ||
3306 | if (unlikely(!high)) | |
3307 | return 0; | |
3308 | ||
3309 | if (!test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags)) | |
3310 | return high; | |
3311 | ||
3312 | /* | |
3313 | * If reclaim is active, limit the number of pages that can be | |
3314 | * stored on pcp lists | |
3315 | */ | |
3316 | return min(READ_ONCE(pcp->batch) << 2, high); | |
3317 | } | |
3318 | ||
df1acc85 MG |
3319 | static void free_unref_page_commit(struct page *page, unsigned long pfn, |
3320 | int migratetype) | |
9cca35d4 MG |
3321 | { |
3322 | struct zone *zone = page_zone(page); | |
3323 | struct per_cpu_pages *pcp; | |
3b12e7e9 | 3324 | int high; |
9cca35d4 | 3325 | |
d34b0733 | 3326 | __count_vm_event(PGFREE); |
28f836b6 | 3327 | pcp = this_cpu_ptr(zone->per_cpu_pageset); |
2d4894b5 | 3328 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 3329 | pcp->count++; |
c49c2c47 | 3330 | high = nr_pcp_high(pcp, zone); |
3b12e7e9 MG |
3331 | if (pcp->count >= high) { |
3332 | int batch = READ_ONCE(pcp->batch); | |
3333 | ||
3334 | free_pcppages_bulk(zone, nr_pcp_free(pcp, high, batch), pcp); | |
3335 | } | |
9cca35d4 | 3336 | } |
5f8dcc21 | 3337 | |
9cca35d4 MG |
3338 | /* |
3339 | * Free a 0-order page | |
9cca35d4 | 3340 | */ |
2d4894b5 | 3341 | void free_unref_page(struct page *page) |
9cca35d4 MG |
3342 | { |
3343 | unsigned long flags; | |
3344 | unsigned long pfn = page_to_pfn(page); | |
df1acc85 | 3345 | int migratetype; |
9cca35d4 | 3346 | |
2d4894b5 | 3347 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3348 | return; |
3349 | ||
df1acc85 MG |
3350 | /* |
3351 | * We only track unmovable, reclaimable and movable on pcp lists. | |
3352 | * Place ISOLATE pages on the isolated list because they are being | |
3353 | * offlined but treat HIGHATOMIC as movable pages so we can get those | |
3354 | * areas back if necessary. Otherwise, we may have to free | |
3355 | * excessively into the page allocator | |
3356 | */ | |
3357 | migratetype = get_pcppage_migratetype(page); | |
3358 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) { | |
3359 | if (unlikely(is_migrate_isolate(migratetype))) { | |
3360 | free_one_page(page_zone(page), page, pfn, 0, migratetype, FPI_NONE); | |
3361 | return; | |
3362 | } | |
3363 | migratetype = MIGRATE_MOVABLE; | |
3364 | } | |
3365 | ||
dbbee9d5 | 3366 | local_lock_irqsave(&pagesets.lock, flags); |
df1acc85 | 3367 | free_unref_page_commit(page, pfn, migratetype); |
dbbee9d5 | 3368 | local_unlock_irqrestore(&pagesets.lock, flags); |
1da177e4 LT |
3369 | } |
3370 | ||
cc59850e KK |
3371 | /* |
3372 | * Free a list of 0-order pages | |
3373 | */ | |
2d4894b5 | 3374 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
3375 | { |
3376 | struct page *page, *next; | |
9cca35d4 | 3377 | unsigned long flags, pfn; |
c24ad77d | 3378 | int batch_count = 0; |
df1acc85 | 3379 | int migratetype; |
9cca35d4 MG |
3380 | |
3381 | /* Prepare pages for freeing */ | |
3382 | list_for_each_entry_safe(page, next, list, lru) { | |
3383 | pfn = page_to_pfn(page); | |
2d4894b5 | 3384 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 | 3385 | list_del(&page->lru); |
df1acc85 MG |
3386 | |
3387 | /* | |
3388 | * Free isolated pages directly to the allocator, see | |
3389 | * comment in free_unref_page. | |
3390 | */ | |
3391 | migratetype = get_pcppage_migratetype(page); | |
3392 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) { | |
3393 | if (unlikely(is_migrate_isolate(migratetype))) { | |
3394 | list_del(&page->lru); | |
3395 | free_one_page(page_zone(page), page, pfn, 0, | |
3396 | migratetype, FPI_NONE); | |
3397 | continue; | |
3398 | } | |
3399 | ||
3400 | /* | |
3401 | * Non-isolated types over MIGRATE_PCPTYPES get added | |
3402 | * to the MIGRATE_MOVABLE pcp list. | |
3403 | */ | |
3404 | set_pcppage_migratetype(page, MIGRATE_MOVABLE); | |
3405 | } | |
3406 | ||
9cca35d4 MG |
3407 | set_page_private(page, pfn); |
3408 | } | |
cc59850e | 3409 | |
dbbee9d5 | 3410 | local_lock_irqsave(&pagesets.lock, flags); |
cc59850e | 3411 | list_for_each_entry_safe(page, next, list, lru) { |
df1acc85 | 3412 | pfn = page_private(page); |
9cca35d4 | 3413 | set_page_private(page, 0); |
df1acc85 | 3414 | migratetype = get_pcppage_migratetype(page); |
2d4894b5 | 3415 | trace_mm_page_free_batched(page); |
df1acc85 | 3416 | free_unref_page_commit(page, pfn, migratetype); |
c24ad77d LS |
3417 | |
3418 | /* | |
3419 | * Guard against excessive IRQ disabled times when we get | |
3420 | * a large list of pages to free. | |
3421 | */ | |
3422 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
dbbee9d5 | 3423 | local_unlock_irqrestore(&pagesets.lock, flags); |
c24ad77d | 3424 | batch_count = 0; |
dbbee9d5 | 3425 | local_lock_irqsave(&pagesets.lock, flags); |
c24ad77d | 3426 | } |
cc59850e | 3427 | } |
dbbee9d5 | 3428 | local_unlock_irqrestore(&pagesets.lock, flags); |
cc59850e KK |
3429 | } |
3430 | ||
8dfcc9ba NP |
3431 | /* |
3432 | * split_page takes a non-compound higher-order page, and splits it into | |
3433 | * n (1<<order) sub-pages: page[0..n] | |
3434 | * Each sub-page must be freed individually. | |
3435 | * | |
3436 | * Note: this is probably too low level an operation for use in drivers. | |
3437 | * Please consult with lkml before using this in your driver. | |
3438 | */ | |
3439 | void split_page(struct page *page, unsigned int order) | |
3440 | { | |
3441 | int i; | |
3442 | ||
309381fe SL |
3443 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3444 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3445 | |
a9627bc5 | 3446 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3447 | set_page_refcounted(page + i); |
8fb156c9 | 3448 | split_page_owner(page, 1 << order); |
e1baddf8 | 3449 | split_page_memcg(page, 1 << order); |
8dfcc9ba | 3450 | } |
5853ff23 | 3451 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3452 | |
3c605096 | 3453 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3454 | { |
748446bb MG |
3455 | unsigned long watermark; |
3456 | struct zone *zone; | |
2139cbe6 | 3457 | int mt; |
748446bb MG |
3458 | |
3459 | BUG_ON(!PageBuddy(page)); | |
3460 | ||
3461 | zone = page_zone(page); | |
2e30abd1 | 3462 | mt = get_pageblock_migratetype(page); |
748446bb | 3463 | |
194159fb | 3464 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
3465 | /* |
3466 | * Obey watermarks as if the page was being allocated. We can | |
3467 | * emulate a high-order watermark check with a raised order-0 | |
3468 | * watermark, because we already know our high-order page | |
3469 | * exists. | |
3470 | */ | |
fd1444b2 | 3471 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3472 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3473 | return 0; |
3474 | ||
8fb74b9f | 3475 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3476 | } |
748446bb MG |
3477 | |
3478 | /* Remove page from free list */ | |
b03641af | 3479 | |
6ab01363 | 3480 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 3481 | |
400bc7fd | 3482 | /* |
3483 | * Set the pageblock if the isolated page is at least half of a | |
3484 | * pageblock | |
3485 | */ | |
748446bb MG |
3486 | if (order >= pageblock_order - 1) { |
3487 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3488 | for (; page < endpage; page += pageblock_nr_pages) { |
3489 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 3490 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 3491 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
3492 | set_pageblock_migratetype(page, |
3493 | MIGRATE_MOVABLE); | |
3494 | } | |
748446bb MG |
3495 | } |
3496 | ||
f3a14ced | 3497 | |
8fb74b9f | 3498 | return 1UL << order; |
1fb3f8ca MG |
3499 | } |
3500 | ||
624f58d8 AD |
3501 | /** |
3502 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
3503 | * @page: Page that was isolated | |
3504 | * @order: Order of the isolated page | |
e6a0a7ad | 3505 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
3506 | * |
3507 | * This function is meant to return a page pulled from the free lists via | |
3508 | * __isolate_free_page back to the free lists they were pulled from. | |
3509 | */ | |
3510 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
3511 | { | |
3512 | struct zone *zone = page_zone(page); | |
3513 | ||
3514 | /* zone lock should be held when this function is called */ | |
3515 | lockdep_assert_held(&zone->lock); | |
3516 | ||
3517 | /* Return isolated page to tail of freelist. */ | |
f04a5d5d | 3518 | __free_one_page(page, page_to_pfn(page), zone, order, mt, |
47b6a24a | 3519 | FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL); |
624f58d8 AD |
3520 | } |
3521 | ||
060e7417 MG |
3522 | /* |
3523 | * Update NUMA hit/miss statistics | |
3524 | * | |
3525 | * Must be called with interrupts disabled. | |
060e7417 | 3526 | */ |
3e23060b MG |
3527 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, |
3528 | long nr_account) | |
060e7417 MG |
3529 | { |
3530 | #ifdef CONFIG_NUMA | |
3a321d2a | 3531 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3532 | |
4518085e KW |
3533 | /* skip numa counters update if numa stats is disabled */ |
3534 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3535 | return; | |
3536 | ||
c1093b74 | 3537 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3538 | local_stat = NUMA_OTHER; |
060e7417 | 3539 | |
c1093b74 | 3540 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3e23060b | 3541 | __count_numa_events(z, NUMA_HIT, nr_account); |
2df26639 | 3542 | else { |
3e23060b MG |
3543 | __count_numa_events(z, NUMA_MISS, nr_account); |
3544 | __count_numa_events(preferred_zone, NUMA_FOREIGN, nr_account); | |
060e7417 | 3545 | } |
3e23060b | 3546 | __count_numa_events(z, local_stat, nr_account); |
060e7417 MG |
3547 | #endif |
3548 | } | |
3549 | ||
066b2393 | 3550 | /* Remove page from the per-cpu list, caller must protect the list */ |
3b822017 JDB |
3551 | static inline |
3552 | struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
6bb15450 | 3553 | unsigned int alloc_flags, |
453f85d4 | 3554 | struct per_cpu_pages *pcp, |
066b2393 MG |
3555 | struct list_head *list) |
3556 | { | |
3557 | struct page *page; | |
3558 | ||
3559 | do { | |
3560 | if (list_empty(list)) { | |
3561 | pcp->count += rmqueue_bulk(zone, 0, | |
5c3ad2eb | 3562 | READ_ONCE(pcp->batch), list, |
6bb15450 | 3563 | migratetype, alloc_flags); |
066b2393 MG |
3564 | if (unlikely(list_empty(list))) |
3565 | return NULL; | |
3566 | } | |
3567 | ||
453f85d4 | 3568 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
3569 | list_del(&page->lru); |
3570 | pcp->count--; | |
3571 | } while (check_new_pcp(page)); | |
3572 | ||
3573 | return page; | |
3574 | } | |
3575 | ||
3576 | /* Lock and remove page from the per-cpu list */ | |
3577 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
1c52e6d0 YS |
3578 | struct zone *zone, gfp_t gfp_flags, |
3579 | int migratetype, unsigned int alloc_flags) | |
066b2393 MG |
3580 | { |
3581 | struct per_cpu_pages *pcp; | |
3582 | struct list_head *list; | |
066b2393 | 3583 | struct page *page; |
d34b0733 | 3584 | unsigned long flags; |
066b2393 | 3585 | |
dbbee9d5 | 3586 | local_lock_irqsave(&pagesets.lock, flags); |
3b12e7e9 MG |
3587 | |
3588 | /* | |
3589 | * On allocation, reduce the number of pages that are batch freed. | |
3590 | * See nr_pcp_free() where free_factor is increased for subsequent | |
3591 | * frees. | |
3592 | */ | |
28f836b6 | 3593 | pcp = this_cpu_ptr(zone->per_cpu_pageset); |
3b12e7e9 | 3594 | pcp->free_factor >>= 1; |
066b2393 | 3595 | list = &pcp->lists[migratetype]; |
6bb15450 | 3596 | page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list); |
43c95bcc | 3597 | local_unlock_irqrestore(&pagesets.lock, flags); |
066b2393 | 3598 | if (page) { |
1c52e6d0 | 3599 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1); |
3e23060b | 3600 | zone_statistics(preferred_zone, zone, 1); |
066b2393 | 3601 | } |
066b2393 MG |
3602 | return page; |
3603 | } | |
3604 | ||
1da177e4 | 3605 | /* |
75379191 | 3606 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 3607 | */ |
0a15c3e9 | 3608 | static inline |
066b2393 | 3609 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3610 | struct zone *zone, unsigned int order, |
c603844b MG |
3611 | gfp_t gfp_flags, unsigned int alloc_flags, |
3612 | int migratetype) | |
1da177e4 LT |
3613 | { |
3614 | unsigned long flags; | |
689bcebf | 3615 | struct page *page; |
1da177e4 | 3616 | |
d34b0733 | 3617 | if (likely(order == 0)) { |
1d91df85 JK |
3618 | /* |
3619 | * MIGRATE_MOVABLE pcplist could have the pages on CMA area and | |
3620 | * we need to skip it when CMA area isn't allowed. | |
3621 | */ | |
3622 | if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA || | |
3623 | migratetype != MIGRATE_MOVABLE) { | |
3624 | page = rmqueue_pcplist(preferred_zone, zone, gfp_flags, | |
1c52e6d0 | 3625 | migratetype, alloc_flags); |
1d91df85 JK |
3626 | goto out; |
3627 | } | |
066b2393 | 3628 | } |
83b9355b | 3629 | |
066b2393 MG |
3630 | /* |
3631 | * We most definitely don't want callers attempting to | |
3632 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3633 | */ | |
3634 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3635 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3636 | |
066b2393 MG |
3637 | do { |
3638 | page = NULL; | |
1d91df85 JK |
3639 | /* |
3640 | * order-0 request can reach here when the pcplist is skipped | |
3641 | * due to non-CMA allocation context. HIGHATOMIC area is | |
3642 | * reserved for high-order atomic allocation, so order-0 | |
3643 | * request should skip it. | |
3644 | */ | |
3645 | if (order > 0 && alloc_flags & ALLOC_HARDER) { | |
066b2393 MG |
3646 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
3647 | if (page) | |
3648 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3649 | } | |
a74609fa | 3650 | if (!page) |
6bb15450 | 3651 | page = __rmqueue(zone, order, migratetype, alloc_flags); |
066b2393 | 3652 | } while (page && check_new_pages(page, order)); |
066b2393 MG |
3653 | if (!page) |
3654 | goto failed; | |
43c95bcc | 3655 | |
066b2393 MG |
3656 | __mod_zone_freepage_state(zone, -(1 << order), |
3657 | get_pcppage_migratetype(page)); | |
43c95bcc | 3658 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 3659 | |
16709d1d | 3660 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
3e23060b | 3661 | zone_statistics(preferred_zone, zone, 1); |
1da177e4 | 3662 | |
066b2393 | 3663 | out: |
73444bc4 MG |
3664 | /* Separate test+clear to avoid unnecessary atomics */ |
3665 | if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) { | |
3666 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); | |
3667 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3668 | } | |
3669 | ||
066b2393 | 3670 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 | 3671 | return page; |
a74609fa NP |
3672 | |
3673 | failed: | |
43c95bcc | 3674 | spin_unlock_irqrestore(&zone->lock, flags); |
a74609fa | 3675 | return NULL; |
1da177e4 LT |
3676 | } |
3677 | ||
933e312e AM |
3678 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3679 | ||
b2588c4b | 3680 | static struct { |
933e312e AM |
3681 | struct fault_attr attr; |
3682 | ||
621a5f7a | 3683 | bool ignore_gfp_highmem; |
71baba4b | 3684 | bool ignore_gfp_reclaim; |
54114994 | 3685 | u32 min_order; |
933e312e AM |
3686 | } fail_page_alloc = { |
3687 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3688 | .ignore_gfp_reclaim = true, |
621a5f7a | 3689 | .ignore_gfp_highmem = true, |
54114994 | 3690 | .min_order = 1, |
933e312e AM |
3691 | }; |
3692 | ||
3693 | static int __init setup_fail_page_alloc(char *str) | |
3694 | { | |
3695 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3696 | } | |
3697 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3698 | ||
af3b8544 | 3699 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3700 | { |
54114994 | 3701 | if (order < fail_page_alloc.min_order) |
deaf386e | 3702 | return false; |
933e312e | 3703 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3704 | return false; |
933e312e | 3705 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3706 | return false; |
71baba4b MG |
3707 | if (fail_page_alloc.ignore_gfp_reclaim && |
3708 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3709 | return false; |
933e312e AM |
3710 | |
3711 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3712 | } | |
3713 | ||
3714 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3715 | ||
3716 | static int __init fail_page_alloc_debugfs(void) | |
3717 | { | |
0825a6f9 | 3718 | umode_t mode = S_IFREG | 0600; |
933e312e | 3719 | struct dentry *dir; |
933e312e | 3720 | |
dd48c085 AM |
3721 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3722 | &fail_page_alloc.attr); | |
b2588c4b | 3723 | |
d9f7979c GKH |
3724 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3725 | &fail_page_alloc.ignore_gfp_reclaim); | |
3726 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3727 | &fail_page_alloc.ignore_gfp_highmem); | |
3728 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3729 | |
d9f7979c | 3730 | return 0; |
933e312e AM |
3731 | } |
3732 | ||
3733 | late_initcall(fail_page_alloc_debugfs); | |
3734 | ||
3735 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3736 | ||
3737 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3738 | ||
af3b8544 | 3739 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3740 | { |
deaf386e | 3741 | return false; |
933e312e AM |
3742 | } |
3743 | ||
3744 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3745 | ||
76cd6173 | 3746 | noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
af3b8544 BP |
3747 | { |
3748 | return __should_fail_alloc_page(gfp_mask, order); | |
3749 | } | |
3750 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3751 | ||
f27ce0e1 JK |
3752 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3753 | unsigned int order, unsigned int alloc_flags) | |
3754 | { | |
3755 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); | |
3756 | long unusable_free = (1 << order) - 1; | |
3757 | ||
3758 | /* | |
3759 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3760 | * the high-atomic reserves. This will over-estimate the size of the | |
3761 | * atomic reserve but it avoids a search. | |
3762 | */ | |
3763 | if (likely(!alloc_harder)) | |
3764 | unusable_free += z->nr_reserved_highatomic; | |
3765 | ||
3766 | #ifdef CONFIG_CMA | |
3767 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3768 | if (!(alloc_flags & ALLOC_CMA)) | |
3769 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3770 | #endif | |
3771 | ||
3772 | return unusable_free; | |
3773 | } | |
3774 | ||
1da177e4 | 3775 | /* |
97a16fc8 MG |
3776 | * Return true if free base pages are above 'mark'. For high-order checks it |
3777 | * will return true of the order-0 watermark is reached and there is at least | |
3778 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3779 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3780 | */ |
86a294a8 | 3781 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3782 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3783 | long free_pages) |
1da177e4 | 3784 | { |
d23ad423 | 3785 | long min = mark; |
1da177e4 | 3786 | int o; |
cd04ae1e | 3787 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3788 | |
0aaa29a5 | 3789 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 3790 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 3791 | |
7fb1d9fc | 3792 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3793 | min -= min / 2; |
0aaa29a5 | 3794 | |
f27ce0e1 | 3795 | if (unlikely(alloc_harder)) { |
cd04ae1e MH |
3796 | /* |
3797 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3798 | * users on the grounds that it's definitely going to be in | |
3799 | * the exit path shortly and free memory. Any allocation it | |
3800 | * makes during the free path will be small and short-lived. | |
3801 | */ | |
3802 | if (alloc_flags & ALLOC_OOM) | |
3803 | min -= min / 2; | |
3804 | else | |
3805 | min -= min / 4; | |
3806 | } | |
3807 | ||
97a16fc8 MG |
3808 | /* |
3809 | * Check watermarks for an order-0 allocation request. If these | |
3810 | * are not met, then a high-order request also cannot go ahead | |
3811 | * even if a suitable page happened to be free. | |
3812 | */ | |
97a225e6 | 3813 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 3814 | return false; |
1da177e4 | 3815 | |
97a16fc8 MG |
3816 | /* If this is an order-0 request then the watermark is fine */ |
3817 | if (!order) | |
3818 | return true; | |
3819 | ||
3820 | /* For a high-order request, check at least one suitable page is free */ | |
3821 | for (o = order; o < MAX_ORDER; o++) { | |
3822 | struct free_area *area = &z->free_area[o]; | |
3823 | int mt; | |
3824 | ||
3825 | if (!area->nr_free) | |
3826 | continue; | |
3827 | ||
97a16fc8 | 3828 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3829 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3830 | return true; |
3831 | } | |
3832 | ||
3833 | #ifdef CONFIG_CMA | |
d883c6cf | 3834 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3835 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3836 | return true; |
d883c6cf | 3837 | } |
97a16fc8 | 3838 | #endif |
76089d00 | 3839 | if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC)) |
b050e376 | 3840 | return true; |
1da177e4 | 3841 | } |
97a16fc8 | 3842 | return false; |
88f5acf8 MG |
3843 | } |
3844 | ||
7aeb09f9 | 3845 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3846 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 3847 | { |
97a225e6 | 3848 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
3849 | zone_page_state(z, NR_FREE_PAGES)); |
3850 | } | |
3851 | ||
48ee5f36 | 3852 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 3853 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 3854 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 3855 | { |
f27ce0e1 | 3856 | long free_pages; |
d883c6cf | 3857 | |
f27ce0e1 | 3858 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
3859 | |
3860 | /* | |
3861 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 3862 | * need to be calculated. |
48ee5f36 | 3863 | */ |
f27ce0e1 JK |
3864 | if (!order) { |
3865 | long fast_free; | |
3866 | ||
3867 | fast_free = free_pages; | |
3868 | fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags); | |
3869 | if (fast_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
3870 | return true; | |
3871 | } | |
48ee5f36 | 3872 | |
f80b08fc CTR |
3873 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
3874 | free_pages)) | |
3875 | return true; | |
3876 | /* | |
3877 | * Ignore watermark boosting for GFP_ATOMIC order-0 allocations | |
3878 | * when checking the min watermark. The min watermark is the | |
3879 | * point where boosting is ignored so that kswapd is woken up | |
3880 | * when below the low watermark. | |
3881 | */ | |
3882 | if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost | |
3883 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { | |
3884 | mark = z->_watermark[WMARK_MIN]; | |
3885 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
3886 | alloc_flags, free_pages); | |
3887 | } | |
3888 | ||
3889 | return false; | |
48ee5f36 MG |
3890 | } |
3891 | ||
7aeb09f9 | 3892 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 3893 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
3894 | { |
3895 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3896 | ||
3897 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3898 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3899 | ||
97a225e6 | 3900 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 3901 | free_pages); |
1da177e4 LT |
3902 | } |
3903 | ||
9276b1bc | 3904 | #ifdef CONFIG_NUMA |
957f822a DR |
3905 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3906 | { | |
e02dc017 | 3907 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3908 | node_reclaim_distance; |
957f822a | 3909 | } |
9276b1bc | 3910 | #else /* CONFIG_NUMA */ |
957f822a DR |
3911 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3912 | { | |
3913 | return true; | |
3914 | } | |
9276b1bc PJ |
3915 | #endif /* CONFIG_NUMA */ |
3916 | ||
6bb15450 MG |
3917 | /* |
3918 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3919 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3920 | * premature use of a lower zone may cause lowmem pressure problems that | |
3921 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3922 | * probably too small. It only makes sense to spread allocations to avoid | |
3923 | * fragmentation between the Normal and DMA32 zones. | |
3924 | */ | |
3925 | static inline unsigned int | |
0a79cdad | 3926 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3927 | { |
736838e9 | 3928 | unsigned int alloc_flags; |
0a79cdad | 3929 | |
736838e9 MN |
3930 | /* |
3931 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
3932 | * to save a branch. | |
3933 | */ | |
3934 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
3935 | |
3936 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3937 | if (!zone) |
3938 | return alloc_flags; | |
3939 | ||
6bb15450 | 3940 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3941 | return alloc_flags; |
6bb15450 MG |
3942 | |
3943 | /* | |
3944 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3945 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3946 | * on UMA that if Normal is populated then so is DMA32. | |
3947 | */ | |
3948 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3949 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3950 | return alloc_flags; |
6bb15450 | 3951 | |
8118b82e | 3952 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3953 | #endif /* CONFIG_ZONE_DMA32 */ |
3954 | return alloc_flags; | |
6bb15450 | 3955 | } |
6bb15450 | 3956 | |
8e3560d9 PT |
3957 | /* Must be called after current_gfp_context() which can change gfp_mask */ |
3958 | static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask, | |
3959 | unsigned int alloc_flags) | |
8510e69c JK |
3960 | { |
3961 | #ifdef CONFIG_CMA | |
8e3560d9 | 3962 | if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
8510e69c | 3963 | alloc_flags |= ALLOC_CMA; |
8510e69c JK |
3964 | #endif |
3965 | return alloc_flags; | |
3966 | } | |
3967 | ||
7fb1d9fc | 3968 | /* |
0798e519 | 3969 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3970 | * a page. |
3971 | */ | |
3972 | static struct page * | |
a9263751 VB |
3973 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3974 | const struct alloc_context *ac) | |
753ee728 | 3975 | { |
6bb15450 | 3976 | struct zoneref *z; |
5117f45d | 3977 | struct zone *zone; |
3b8c0be4 | 3978 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
6bb15450 | 3979 | bool no_fallback; |
3b8c0be4 | 3980 | |
6bb15450 | 3981 | retry: |
7fb1d9fc | 3982 | /* |
9276b1bc | 3983 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3984 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3985 | */ |
6bb15450 MG |
3986 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3987 | z = ac->preferred_zoneref; | |
30d8ec73 MN |
3988 | for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx, |
3989 | ac->nodemask) { | |
be06af00 | 3990 | struct page *page; |
e085dbc5 JW |
3991 | unsigned long mark; |
3992 | ||
664eedde MG |
3993 | if (cpusets_enabled() && |
3994 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3995 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3996 | continue; |
a756cf59 JW |
3997 | /* |
3998 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3999 | * want to get it from a node that is within its dirty |
4000 | * limit, such that no single node holds more than its | |
a756cf59 | 4001 | * proportional share of globally allowed dirty pages. |
281e3726 | 4002 | * The dirty limits take into account the node's |
a756cf59 JW |
4003 | * lowmem reserves and high watermark so that kswapd |
4004 | * should be able to balance it without having to | |
4005 | * write pages from its LRU list. | |
4006 | * | |
a756cf59 | 4007 | * XXX: For now, allow allocations to potentially |
281e3726 | 4008 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 4009 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 4010 | * which is important when on a NUMA setup the allowed |
281e3726 | 4011 | * nodes are together not big enough to reach the |
a756cf59 | 4012 | * global limit. The proper fix for these situations |
281e3726 | 4013 | * will require awareness of nodes in the |
a756cf59 JW |
4014 | * dirty-throttling and the flusher threads. |
4015 | */ | |
3b8c0be4 MG |
4016 | if (ac->spread_dirty_pages) { |
4017 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
4018 | continue; | |
4019 | ||
4020 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
4021 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
4022 | continue; | |
4023 | } | |
4024 | } | |
7fb1d9fc | 4025 | |
6bb15450 MG |
4026 | if (no_fallback && nr_online_nodes > 1 && |
4027 | zone != ac->preferred_zoneref->zone) { | |
4028 | int local_nid; | |
4029 | ||
4030 | /* | |
4031 | * If moving to a remote node, retry but allow | |
4032 | * fragmenting fallbacks. Locality is more important | |
4033 | * than fragmentation avoidance. | |
4034 | */ | |
4035 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
4036 | if (zone_to_nid(zone) != local_nid) { | |
4037 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
4038 | goto retry; | |
4039 | } | |
4040 | } | |
4041 | ||
a9214443 | 4042 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 4043 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
4044 | ac->highest_zoneidx, alloc_flags, |
4045 | gfp_mask)) { | |
fa5e084e MG |
4046 | int ret; |
4047 | ||
c9e97a19 PT |
4048 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
4049 | /* | |
4050 | * Watermark failed for this zone, but see if we can | |
4051 | * grow this zone if it contains deferred pages. | |
4052 | */ | |
4053 | if (static_branch_unlikely(&deferred_pages)) { | |
4054 | if (_deferred_grow_zone(zone, order)) | |
4055 | goto try_this_zone; | |
4056 | } | |
4057 | #endif | |
5dab2911 MG |
4058 | /* Checked here to keep the fast path fast */ |
4059 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
4060 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
4061 | goto try_this_zone; | |
4062 | ||
202e35db | 4063 | if (!node_reclaim_enabled() || |
c33d6c06 | 4064 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
4065 | continue; |
4066 | ||
a5f5f91d | 4067 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 4068 | switch (ret) { |
a5f5f91d | 4069 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 4070 | /* did not scan */ |
cd38b115 | 4071 | continue; |
a5f5f91d | 4072 | case NODE_RECLAIM_FULL: |
fa5e084e | 4073 | /* scanned but unreclaimable */ |
cd38b115 | 4074 | continue; |
fa5e084e MG |
4075 | default: |
4076 | /* did we reclaim enough */ | |
fed2719e | 4077 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 4078 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
4079 | goto try_this_zone; |
4080 | ||
fed2719e | 4081 | continue; |
0798e519 | 4082 | } |
7fb1d9fc RS |
4083 | } |
4084 | ||
fa5e084e | 4085 | try_this_zone: |
066b2393 | 4086 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 4087 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 4088 | if (page) { |
479f854a | 4089 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
4090 | |
4091 | /* | |
4092 | * If this is a high-order atomic allocation then check | |
4093 | * if the pageblock should be reserved for the future | |
4094 | */ | |
4095 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
4096 | reserve_highatomic_pageblock(page, zone, order); | |
4097 | ||
75379191 | 4098 | return page; |
c9e97a19 PT |
4099 | } else { |
4100 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
4101 | /* Try again if zone has deferred pages */ | |
4102 | if (static_branch_unlikely(&deferred_pages)) { | |
4103 | if (_deferred_grow_zone(zone, order)) | |
4104 | goto try_this_zone; | |
4105 | } | |
4106 | #endif | |
75379191 | 4107 | } |
54a6eb5c | 4108 | } |
9276b1bc | 4109 | |
6bb15450 MG |
4110 | /* |
4111 | * It's possible on a UMA machine to get through all zones that are | |
4112 | * fragmented. If avoiding fragmentation, reset and try again. | |
4113 | */ | |
4114 | if (no_fallback) { | |
4115 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
4116 | goto retry; | |
4117 | } | |
4118 | ||
4ffeaf35 | 4119 | return NULL; |
753ee728 MH |
4120 | } |
4121 | ||
9af744d7 | 4122 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 4123 | { |
a238ab5b | 4124 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
4125 | |
4126 | /* | |
4127 | * This documents exceptions given to allocations in certain | |
4128 | * contexts that are allowed to allocate outside current's set | |
4129 | * of allowed nodes. | |
4130 | */ | |
4131 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 4132 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
4133 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
4134 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 4135 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
4136 | filter &= ~SHOW_MEM_FILTER_NODES; |
4137 | ||
9af744d7 | 4138 | show_mem(filter, nodemask); |
aa187507 MH |
4139 | } |
4140 | ||
a8e99259 | 4141 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
4142 | { |
4143 | struct va_format vaf; | |
4144 | va_list args; | |
1be334e5 | 4145 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 4146 | |
0f7896f1 | 4147 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
4148 | return; |
4149 | ||
7877cdcc MH |
4150 | va_start(args, fmt); |
4151 | vaf.fmt = fmt; | |
4152 | vaf.va = &args; | |
ef8444ea | 4153 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
4154 | current->comm, &vaf, gfp_mask, &gfp_mask, |
4155 | nodemask_pr_args(nodemask)); | |
7877cdcc | 4156 | va_end(args); |
3ee9a4f0 | 4157 | |
a8e99259 | 4158 | cpuset_print_current_mems_allowed(); |
ef8444ea | 4159 | pr_cont("\n"); |
a238ab5b | 4160 | dump_stack(); |
685dbf6f | 4161 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
4162 | } |
4163 | ||
6c18ba7a MH |
4164 | static inline struct page * |
4165 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
4166 | unsigned int alloc_flags, | |
4167 | const struct alloc_context *ac) | |
4168 | { | |
4169 | struct page *page; | |
4170 | ||
4171 | page = get_page_from_freelist(gfp_mask, order, | |
4172 | alloc_flags|ALLOC_CPUSET, ac); | |
4173 | /* | |
4174 | * fallback to ignore cpuset restriction if our nodes | |
4175 | * are depleted | |
4176 | */ | |
4177 | if (!page) | |
4178 | page = get_page_from_freelist(gfp_mask, order, | |
4179 | alloc_flags, ac); | |
4180 | ||
4181 | return page; | |
4182 | } | |
4183 | ||
11e33f6a MG |
4184 | static inline struct page * |
4185 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4186 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 4187 | { |
6e0fc46d DR |
4188 | struct oom_control oc = { |
4189 | .zonelist = ac->zonelist, | |
4190 | .nodemask = ac->nodemask, | |
2a966b77 | 4191 | .memcg = NULL, |
6e0fc46d DR |
4192 | .gfp_mask = gfp_mask, |
4193 | .order = order, | |
6e0fc46d | 4194 | }; |
11e33f6a MG |
4195 | struct page *page; |
4196 | ||
9879de73 JW |
4197 | *did_some_progress = 0; |
4198 | ||
9879de73 | 4199 | /* |
dc56401f JW |
4200 | * Acquire the oom lock. If that fails, somebody else is |
4201 | * making progress for us. | |
9879de73 | 4202 | */ |
dc56401f | 4203 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 4204 | *did_some_progress = 1; |
11e33f6a | 4205 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
4206 | return NULL; |
4207 | } | |
6b1de916 | 4208 | |
11e33f6a MG |
4209 | /* |
4210 | * Go through the zonelist yet one more time, keep very high watermark | |
4211 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
4212 | * we're still under heavy pressure. But make sure that this reclaim |
4213 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
4214 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 4215 | */ |
e746bf73 TH |
4216 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
4217 | ~__GFP_DIRECT_RECLAIM, order, | |
4218 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 4219 | if (page) |
11e33f6a MG |
4220 | goto out; |
4221 | ||
06ad276a MH |
4222 | /* Coredumps can quickly deplete all memory reserves */ |
4223 | if (current->flags & PF_DUMPCORE) | |
4224 | goto out; | |
4225 | /* The OOM killer will not help higher order allocs */ | |
4226 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4227 | goto out; | |
dcda9b04 MH |
4228 | /* |
4229 | * We have already exhausted all our reclaim opportunities without any | |
4230 | * success so it is time to admit defeat. We will skip the OOM killer | |
4231 | * because it is very likely that the caller has a more reasonable | |
4232 | * fallback than shooting a random task. | |
cfb4a541 MN |
4233 | * |
4234 | * The OOM killer may not free memory on a specific node. | |
dcda9b04 | 4235 | */ |
cfb4a541 | 4236 | if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE)) |
dcda9b04 | 4237 | goto out; |
06ad276a | 4238 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 4239 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
4240 | goto out; |
4241 | if (pm_suspended_storage()) | |
4242 | goto out; | |
4243 | /* | |
4244 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
4245 | * other request to make a forward progress. | |
4246 | * We are in an unfortunate situation where out_of_memory cannot | |
4247 | * do much for this context but let's try it to at least get | |
4248 | * access to memory reserved if the current task is killed (see | |
4249 | * out_of_memory). Once filesystems are ready to handle allocation | |
4250 | * failures more gracefully we should just bail out here. | |
4251 | */ | |
4252 | ||
3c2c6488 | 4253 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 4254 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 4255 | *did_some_progress = 1; |
5020e285 | 4256 | |
6c18ba7a MH |
4257 | /* |
4258 | * Help non-failing allocations by giving them access to memory | |
4259 | * reserves | |
4260 | */ | |
4261 | if (gfp_mask & __GFP_NOFAIL) | |
4262 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 4263 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 4264 | } |
11e33f6a | 4265 | out: |
dc56401f | 4266 | mutex_unlock(&oom_lock); |
11e33f6a MG |
4267 | return page; |
4268 | } | |
4269 | ||
33c2d214 | 4270 | /* |
baf2f90b | 4271 | * Maximum number of compaction retries with a progress before OOM |
33c2d214 MH |
4272 | * killer is consider as the only way to move forward. |
4273 | */ | |
4274 | #define MAX_COMPACT_RETRIES 16 | |
4275 | ||
56de7263 MG |
4276 | #ifdef CONFIG_COMPACTION |
4277 | /* Try memory compaction for high-order allocations before reclaim */ | |
4278 | static struct page * | |
4279 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4280 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4281 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4282 | { |
5e1f0f09 | 4283 | struct page *page = NULL; |
eb414681 | 4284 | unsigned long pflags; |
499118e9 | 4285 | unsigned int noreclaim_flag; |
53853e2d VB |
4286 | |
4287 | if (!order) | |
66199712 | 4288 | return NULL; |
66199712 | 4289 | |
eb414681 | 4290 | psi_memstall_enter(&pflags); |
499118e9 | 4291 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 4292 | |
c5d01d0d | 4293 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 4294 | prio, &page); |
eb414681 | 4295 | |
499118e9 | 4296 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 4297 | psi_memstall_leave(&pflags); |
56de7263 | 4298 | |
06dac2f4 CTR |
4299 | if (*compact_result == COMPACT_SKIPPED) |
4300 | return NULL; | |
98dd3b48 VB |
4301 | /* |
4302 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
4303 | * count a compaction stall | |
4304 | */ | |
4305 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 4306 | |
5e1f0f09 MG |
4307 | /* Prep a captured page if available */ |
4308 | if (page) | |
4309 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
4310 | ||
4311 | /* Try get a page from the freelist if available */ | |
4312 | if (!page) | |
4313 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 4314 | |
98dd3b48 VB |
4315 | if (page) { |
4316 | struct zone *zone = page_zone(page); | |
53853e2d | 4317 | |
98dd3b48 VB |
4318 | zone->compact_blockskip_flush = false; |
4319 | compaction_defer_reset(zone, order, true); | |
4320 | count_vm_event(COMPACTSUCCESS); | |
4321 | return page; | |
4322 | } | |
56de7263 | 4323 | |
98dd3b48 VB |
4324 | /* |
4325 | * It's bad if compaction run occurs and fails. The most likely reason | |
4326 | * is that pages exist, but not enough to satisfy watermarks. | |
4327 | */ | |
4328 | count_vm_event(COMPACTFAIL); | |
66199712 | 4329 | |
98dd3b48 | 4330 | cond_resched(); |
56de7263 MG |
4331 | |
4332 | return NULL; | |
4333 | } | |
33c2d214 | 4334 | |
3250845d VB |
4335 | static inline bool |
4336 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
4337 | enum compact_result compact_result, | |
4338 | enum compact_priority *compact_priority, | |
d9436498 | 4339 | int *compaction_retries) |
3250845d VB |
4340 | { |
4341 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 4342 | int min_priority; |
65190cff MH |
4343 | bool ret = false; |
4344 | int retries = *compaction_retries; | |
4345 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
4346 | |
4347 | if (!order) | |
4348 | return false; | |
4349 | ||
691d9497 AT |
4350 | if (fatal_signal_pending(current)) |
4351 | return false; | |
4352 | ||
d9436498 VB |
4353 | if (compaction_made_progress(compact_result)) |
4354 | (*compaction_retries)++; | |
4355 | ||
3250845d VB |
4356 | /* |
4357 | * compaction considers all the zone as desperately out of memory | |
4358 | * so it doesn't really make much sense to retry except when the | |
4359 | * failure could be caused by insufficient priority | |
4360 | */ | |
d9436498 VB |
4361 | if (compaction_failed(compact_result)) |
4362 | goto check_priority; | |
3250845d | 4363 | |
49433085 VB |
4364 | /* |
4365 | * compaction was skipped because there are not enough order-0 pages | |
4366 | * to work with, so we retry only if it looks like reclaim can help. | |
4367 | */ | |
4368 | if (compaction_needs_reclaim(compact_result)) { | |
4369 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
4370 | goto out; | |
4371 | } | |
4372 | ||
3250845d VB |
4373 | /* |
4374 | * make sure the compaction wasn't deferred or didn't bail out early | |
4375 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
4376 | * But the next retry should use a higher priority if allowed, so |
4377 | * we don't just keep bailing out endlessly. | |
3250845d | 4378 | */ |
65190cff | 4379 | if (compaction_withdrawn(compact_result)) { |
49433085 | 4380 | goto check_priority; |
65190cff | 4381 | } |
3250845d VB |
4382 | |
4383 | /* | |
dcda9b04 | 4384 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
4385 | * costly ones because they are de facto nofail and invoke OOM |
4386 | * killer to move on while costly can fail and users are ready | |
4387 | * to cope with that. 1/4 retries is rather arbitrary but we | |
4388 | * would need much more detailed feedback from compaction to | |
4389 | * make a better decision. | |
4390 | */ | |
4391 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4392 | max_retries /= 4; | |
65190cff MH |
4393 | if (*compaction_retries <= max_retries) { |
4394 | ret = true; | |
4395 | goto out; | |
4396 | } | |
3250845d | 4397 | |
d9436498 VB |
4398 | /* |
4399 | * Make sure there are attempts at the highest priority if we exhausted | |
4400 | * all retries or failed at the lower priorities. | |
4401 | */ | |
4402 | check_priority: | |
c2033b00 VB |
4403 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
4404 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 4405 | |
c2033b00 | 4406 | if (*compact_priority > min_priority) { |
d9436498 VB |
4407 | (*compact_priority)--; |
4408 | *compaction_retries = 0; | |
65190cff | 4409 | ret = true; |
d9436498 | 4410 | } |
65190cff MH |
4411 | out: |
4412 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4413 | return ret; | |
3250845d | 4414 | } |
56de7263 MG |
4415 | #else |
4416 | static inline struct page * | |
4417 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4418 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4419 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4420 | { |
33c2d214 | 4421 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4422 | return NULL; |
4423 | } | |
33c2d214 MH |
4424 | |
4425 | static inline bool | |
86a294a8 MH |
4426 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4427 | enum compact_result compact_result, | |
a5508cd8 | 4428 | enum compact_priority *compact_priority, |
d9436498 | 4429 | int *compaction_retries) |
33c2d214 | 4430 | { |
31e49bfd MH |
4431 | struct zone *zone; |
4432 | struct zoneref *z; | |
4433 | ||
4434 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4435 | return false; | |
4436 | ||
4437 | /* | |
4438 | * There are setups with compaction disabled which would prefer to loop | |
4439 | * inside the allocator rather than hit the oom killer prematurely. | |
4440 | * Let's give them a good hope and keep retrying while the order-0 | |
4441 | * watermarks are OK. | |
4442 | */ | |
97a225e6 JK |
4443 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4444 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 4445 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 4446 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
4447 | return true; |
4448 | } | |
33c2d214 MH |
4449 | return false; |
4450 | } | |
3250845d | 4451 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4452 | |
d92a8cfc | 4453 | #ifdef CONFIG_LOCKDEP |
93781325 | 4454 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4455 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4456 | ||
f920e413 | 4457 | static bool __need_reclaim(gfp_t gfp_mask) |
d92a8cfc | 4458 | { |
d92a8cfc PZ |
4459 | /* no reclaim without waiting on it */ |
4460 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4461 | return false; | |
4462 | ||
4463 | /* this guy won't enter reclaim */ | |
2e517d68 | 4464 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4465 | return false; |
4466 | ||
d92a8cfc PZ |
4467 | if (gfp_mask & __GFP_NOLOCKDEP) |
4468 | return false; | |
4469 | ||
4470 | return true; | |
4471 | } | |
4472 | ||
93781325 OS |
4473 | void __fs_reclaim_acquire(void) |
4474 | { | |
4475 | lock_map_acquire(&__fs_reclaim_map); | |
4476 | } | |
4477 | ||
4478 | void __fs_reclaim_release(void) | |
4479 | { | |
4480 | lock_map_release(&__fs_reclaim_map); | |
4481 | } | |
4482 | ||
d92a8cfc PZ |
4483 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4484 | { | |
f920e413 DV |
4485 | gfp_mask = current_gfp_context(gfp_mask); |
4486 | ||
4487 | if (__need_reclaim(gfp_mask)) { | |
4488 | if (gfp_mask & __GFP_FS) | |
4489 | __fs_reclaim_acquire(); | |
4490 | ||
4491 | #ifdef CONFIG_MMU_NOTIFIER | |
4492 | lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); | |
4493 | lock_map_release(&__mmu_notifier_invalidate_range_start_map); | |
4494 | #endif | |
4495 | ||
4496 | } | |
d92a8cfc PZ |
4497 | } |
4498 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4499 | ||
4500 | void fs_reclaim_release(gfp_t gfp_mask) | |
4501 | { | |
f920e413 DV |
4502 | gfp_mask = current_gfp_context(gfp_mask); |
4503 | ||
4504 | if (__need_reclaim(gfp_mask)) { | |
4505 | if (gfp_mask & __GFP_FS) | |
4506 | __fs_reclaim_release(); | |
4507 | } | |
d92a8cfc PZ |
4508 | } |
4509 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4510 | #endif | |
4511 | ||
bba90710 | 4512 | /* Perform direct synchronous page reclaim */ |
2187e17b | 4513 | static unsigned long |
a9263751 VB |
4514 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4515 | const struct alloc_context *ac) | |
11e33f6a | 4516 | { |
499118e9 | 4517 | unsigned int noreclaim_flag; |
2187e17b | 4518 | unsigned long pflags, progress; |
11e33f6a MG |
4519 | |
4520 | cond_resched(); | |
4521 | ||
4522 | /* We now go into synchronous reclaim */ | |
4523 | cpuset_memory_pressure_bump(); | |
eb414681 | 4524 | psi_memstall_enter(&pflags); |
d92a8cfc | 4525 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4526 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4527 | |
a9263751 VB |
4528 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4529 | ac->nodemask); | |
11e33f6a | 4530 | |
499118e9 | 4531 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4532 | fs_reclaim_release(gfp_mask); |
eb414681 | 4533 | psi_memstall_leave(&pflags); |
11e33f6a MG |
4534 | |
4535 | cond_resched(); | |
4536 | ||
bba90710 MS |
4537 | return progress; |
4538 | } | |
4539 | ||
4540 | /* The really slow allocator path where we enter direct reclaim */ | |
4541 | static inline struct page * | |
4542 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4543 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4544 | unsigned long *did_some_progress) |
bba90710 MS |
4545 | { |
4546 | struct page *page = NULL; | |
4547 | bool drained = false; | |
4548 | ||
a9263751 | 4549 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
4550 | if (unlikely(!(*did_some_progress))) |
4551 | return NULL; | |
11e33f6a | 4552 | |
9ee493ce | 4553 | retry: |
31a6c190 | 4554 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4555 | |
4556 | /* | |
4557 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 | 4558 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
047b9967 | 4559 | * Shrink them and try again |
9ee493ce MG |
4560 | */ |
4561 | if (!page && !drained) { | |
29fac03b | 4562 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4563 | drain_all_pages(NULL); |
9ee493ce MG |
4564 | drained = true; |
4565 | goto retry; | |
4566 | } | |
4567 | ||
11e33f6a MG |
4568 | return page; |
4569 | } | |
4570 | ||
5ecd9d40 DR |
4571 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4572 | const struct alloc_context *ac) | |
3a025760 JW |
4573 | { |
4574 | struct zoneref *z; | |
4575 | struct zone *zone; | |
e1a55637 | 4576 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 4577 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 4578 | |
97a225e6 | 4579 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 4580 | ac->nodemask) { |
e1a55637 | 4581 | if (last_pgdat != zone->zone_pgdat) |
97a225e6 | 4582 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
e1a55637 MG |
4583 | last_pgdat = zone->zone_pgdat; |
4584 | } | |
3a025760 JW |
4585 | } |
4586 | ||
c603844b | 4587 | static inline unsigned int |
341ce06f PZ |
4588 | gfp_to_alloc_flags(gfp_t gfp_mask) |
4589 | { | |
c603844b | 4590 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4591 | |
736838e9 MN |
4592 | /* |
4593 | * __GFP_HIGH is assumed to be the same as ALLOC_HIGH | |
4594 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4595 | * to save two branches. | |
4596 | */ | |
e6223a3b | 4597 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
736838e9 | 4598 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 4599 | |
341ce06f PZ |
4600 | /* |
4601 | * The caller may dip into page reserves a bit more if the caller | |
4602 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4603 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 4604 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 4605 | */ |
736838e9 MN |
4606 | alloc_flags |= (__force int) |
4607 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 4608 | |
d0164adc | 4609 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 4610 | /* |
b104a35d DR |
4611 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4612 | * if it can't schedule. | |
5c3240d9 | 4613 | */ |
b104a35d | 4614 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 4615 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 4616 | /* |
b104a35d | 4617 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 4618 | * comment for __cpuset_node_allowed(). |
523b9458 | 4619 | */ |
341ce06f | 4620 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 4621 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
4622 | alloc_flags |= ALLOC_HARDER; |
4623 | ||
8e3560d9 | 4624 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags); |
8510e69c | 4625 | |
341ce06f PZ |
4626 | return alloc_flags; |
4627 | } | |
4628 | ||
cd04ae1e | 4629 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4630 | { |
cd04ae1e MH |
4631 | if (!tsk_is_oom_victim(tsk)) |
4632 | return false; | |
4633 | ||
4634 | /* | |
4635 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4636 | * only to the thread with TIF_MEMDIE set | |
4637 | */ | |
4638 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4639 | return false; |
4640 | ||
cd04ae1e MH |
4641 | return true; |
4642 | } | |
4643 | ||
4644 | /* | |
4645 | * Distinguish requests which really need access to full memory | |
4646 | * reserves from oom victims which can live with a portion of it | |
4647 | */ | |
4648 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4649 | { | |
4650 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4651 | return 0; | |
31a6c190 | 4652 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4653 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4654 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4655 | return ALLOC_NO_WATERMARKS; |
4656 | if (!in_interrupt()) { | |
4657 | if (current->flags & PF_MEMALLOC) | |
4658 | return ALLOC_NO_WATERMARKS; | |
4659 | else if (oom_reserves_allowed(current)) | |
4660 | return ALLOC_OOM; | |
4661 | } | |
31a6c190 | 4662 | |
cd04ae1e MH |
4663 | return 0; |
4664 | } | |
4665 | ||
4666 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4667 | { | |
4668 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4669 | } |
4670 | ||
0a0337e0 MH |
4671 | /* |
4672 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4673 | * for the given allocation request. | |
491d79ae JW |
4674 | * |
4675 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4676 | * without success, or when we couldn't even meet the watermark if we | |
4677 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4678 | * |
4679 | * Returns true if a retry is viable or false to enter the oom path. | |
4680 | */ | |
4681 | static inline bool | |
4682 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4683 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4684 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4685 | { |
4686 | struct zone *zone; | |
4687 | struct zoneref *z; | |
15f570bf | 4688 | bool ret = false; |
0a0337e0 | 4689 | |
423b452e VB |
4690 | /* |
4691 | * Costly allocations might have made a progress but this doesn't mean | |
4692 | * their order will become available due to high fragmentation so | |
4693 | * always increment the no progress counter for them | |
4694 | */ | |
4695 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4696 | *no_progress_loops = 0; | |
4697 | else | |
4698 | (*no_progress_loops)++; | |
4699 | ||
0a0337e0 MH |
4700 | /* |
4701 | * Make sure we converge to OOM if we cannot make any progress | |
4702 | * several times in the row. | |
4703 | */ | |
04c8716f MK |
4704 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4705 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4706 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4707 | } |
0a0337e0 | 4708 | |
bca67592 MG |
4709 | /* |
4710 | * Keep reclaiming pages while there is a chance this will lead | |
4711 | * somewhere. If none of the target zones can satisfy our allocation | |
4712 | * request even if all reclaimable pages are considered then we are | |
4713 | * screwed and have to go OOM. | |
0a0337e0 | 4714 | */ |
97a225e6 JK |
4715 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4716 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4717 | unsigned long available; |
ede37713 | 4718 | unsigned long reclaimable; |
d379f01d MH |
4719 | unsigned long min_wmark = min_wmark_pages(zone); |
4720 | bool wmark; | |
0a0337e0 | 4721 | |
5a1c84b4 | 4722 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4723 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4724 | |
4725 | /* | |
491d79ae JW |
4726 | * Would the allocation succeed if we reclaimed all |
4727 | * reclaimable pages? | |
0a0337e0 | 4728 | */ |
d379f01d | 4729 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 4730 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
4731 | trace_reclaim_retry_zone(z, order, reclaimable, |
4732 | available, min_wmark, *no_progress_loops, wmark); | |
4733 | if (wmark) { | |
ede37713 MH |
4734 | /* |
4735 | * If we didn't make any progress and have a lot of | |
4736 | * dirty + writeback pages then we should wait for | |
4737 | * an IO to complete to slow down the reclaim and | |
4738 | * prevent from pre mature OOM | |
4739 | */ | |
4740 | if (!did_some_progress) { | |
11fb9989 | 4741 | unsigned long write_pending; |
ede37713 | 4742 | |
5a1c84b4 MG |
4743 | write_pending = zone_page_state_snapshot(zone, |
4744 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 4745 | |
11fb9989 | 4746 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
4747 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
4748 | return true; | |
4749 | } | |
4750 | } | |
5a1c84b4 | 4751 | |
15f570bf MH |
4752 | ret = true; |
4753 | goto out; | |
0a0337e0 MH |
4754 | } |
4755 | } | |
4756 | ||
15f570bf MH |
4757 | out: |
4758 | /* | |
4759 | * Memory allocation/reclaim might be called from a WQ context and the | |
4760 | * current implementation of the WQ concurrency control doesn't | |
4761 | * recognize that a particular WQ is congested if the worker thread is | |
4762 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4763 | * here rather than calling cond_resched(). | |
4764 | */ | |
4765 | if (current->flags & PF_WQ_WORKER) | |
4766 | schedule_timeout_uninterruptible(1); | |
4767 | else | |
4768 | cond_resched(); | |
4769 | return ret; | |
0a0337e0 MH |
4770 | } |
4771 | ||
902b6281 VB |
4772 | static inline bool |
4773 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4774 | { | |
4775 | /* | |
4776 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4777 | * mempolicy don't intersect. This should be normally dealt with by | |
4778 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4779 | * such a way the check therein was true, and then it became false | |
4780 | * before we got our cpuset_mems_cookie here. | |
4781 | * This assumes that for all allocations, ac->nodemask can come only | |
4782 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4783 | * when it does not intersect with the cpuset restrictions) or the | |
4784 | * caller can deal with a violated nodemask. | |
4785 | */ | |
4786 | if (cpusets_enabled() && ac->nodemask && | |
4787 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4788 | ac->nodemask = NULL; | |
4789 | return true; | |
4790 | } | |
4791 | ||
4792 | /* | |
4793 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4794 | * possible to race with parallel threads in such a way that our | |
4795 | * allocation can fail while the mask is being updated. If we are about | |
4796 | * to fail, check if the cpuset changed during allocation and if so, | |
4797 | * retry. | |
4798 | */ | |
4799 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4800 | return true; | |
4801 | ||
4802 | return false; | |
4803 | } | |
4804 | ||
11e33f6a MG |
4805 | static inline struct page * |
4806 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4807 | struct alloc_context *ac) |
11e33f6a | 4808 | { |
d0164adc | 4809 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4810 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4811 | struct page *page = NULL; |
c603844b | 4812 | unsigned int alloc_flags; |
11e33f6a | 4813 | unsigned long did_some_progress; |
5ce9bfef | 4814 | enum compact_priority compact_priority; |
c5d01d0d | 4815 | enum compact_result compact_result; |
5ce9bfef VB |
4816 | int compaction_retries; |
4817 | int no_progress_loops; | |
5ce9bfef | 4818 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4819 | int reserve_flags; |
1da177e4 | 4820 | |
d0164adc MG |
4821 | /* |
4822 | * We also sanity check to catch abuse of atomic reserves being used by | |
4823 | * callers that are not in atomic context. | |
4824 | */ | |
4825 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4826 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4827 | gfp_mask &= ~__GFP_ATOMIC; | |
4828 | ||
5ce9bfef VB |
4829 | retry_cpuset: |
4830 | compaction_retries = 0; | |
4831 | no_progress_loops = 0; | |
4832 | compact_priority = DEF_COMPACT_PRIORITY; | |
4833 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4834 | |
4835 | /* | |
4836 | * The fast path uses conservative alloc_flags to succeed only until | |
4837 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4838 | * alloc_flags precisely. So we do that now. | |
4839 | */ | |
4840 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4841 | ||
e47483bc VB |
4842 | /* |
4843 | * We need to recalculate the starting point for the zonelist iterator | |
4844 | * because we might have used different nodemask in the fast path, or | |
4845 | * there was a cpuset modification and we are retrying - otherwise we | |
4846 | * could end up iterating over non-eligible zones endlessly. | |
4847 | */ | |
4848 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 4849 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
4850 | if (!ac->preferred_zoneref->zone) |
4851 | goto nopage; | |
4852 | ||
0a79cdad | 4853 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4854 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4855 | |
4856 | /* | |
4857 | * The adjusted alloc_flags might result in immediate success, so try | |
4858 | * that first | |
4859 | */ | |
4860 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4861 | if (page) | |
4862 | goto got_pg; | |
4863 | ||
a8161d1e VB |
4864 | /* |
4865 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4866 | * that we have enough base pages and don't need to reclaim. For non- |
4867 | * movable high-order allocations, do that as well, as compaction will | |
4868 | * try prevent permanent fragmentation by migrating from blocks of the | |
4869 | * same migratetype. | |
4870 | * Don't try this for allocations that are allowed to ignore | |
4871 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4872 | */ |
282722b0 VB |
4873 | if (can_direct_reclaim && |
4874 | (costly_order || | |
4875 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4876 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4877 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4878 | alloc_flags, ac, | |
a5508cd8 | 4879 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4880 | &compact_result); |
4881 | if (page) | |
4882 | goto got_pg; | |
4883 | ||
cc638f32 VB |
4884 | /* |
4885 | * Checks for costly allocations with __GFP_NORETRY, which | |
4886 | * includes some THP page fault allocations | |
4887 | */ | |
4888 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4889 | /* |
4890 | * If allocating entire pageblock(s) and compaction | |
4891 | * failed because all zones are below low watermarks | |
4892 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4893 | * order, fail immediately unless the allocator has |
4894 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4895 | * |
4896 | * Reclaim is | |
4897 | * - potentially very expensive because zones are far | |
4898 | * below their low watermarks or this is part of very | |
4899 | * bursty high order allocations, | |
4900 | * - not guaranteed to help because isolate_freepages() | |
4901 | * may not iterate over freed pages as part of its | |
4902 | * linear scan, and | |
4903 | * - unlikely to make entire pageblocks free on its | |
4904 | * own. | |
4905 | */ | |
4906 | if (compact_result == COMPACT_SKIPPED || | |
4907 | compact_result == COMPACT_DEFERRED) | |
4908 | goto nopage; | |
a8161d1e | 4909 | |
a8161d1e | 4910 | /* |
3eb2771b VB |
4911 | * Looks like reclaim/compaction is worth trying, but |
4912 | * sync compaction could be very expensive, so keep | |
25160354 | 4913 | * using async compaction. |
a8161d1e | 4914 | */ |
a5508cd8 | 4915 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4916 | } |
4917 | } | |
23771235 | 4918 | |
31a6c190 | 4919 | retry: |
23771235 | 4920 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4921 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4922 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4923 | |
cd04ae1e MH |
4924 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4925 | if (reserve_flags) | |
8e3560d9 | 4926 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags); |
23771235 | 4927 | |
e46e7b77 | 4928 | /* |
d6a24df0 VB |
4929 | * Reset the nodemask and zonelist iterators if memory policies can be |
4930 | * ignored. These allocations are high priority and system rather than | |
4931 | * user oriented. | |
e46e7b77 | 4932 | */ |
cd04ae1e | 4933 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4934 | ac->nodemask = NULL; |
e46e7b77 | 4935 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4936 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
4937 | } |
4938 | ||
23771235 | 4939 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4940 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4941 | if (page) |
4942 | goto got_pg; | |
1da177e4 | 4943 | |
d0164adc | 4944 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4945 | if (!can_direct_reclaim) |
1da177e4 LT |
4946 | goto nopage; |
4947 | ||
9a67f648 MH |
4948 | /* Avoid recursion of direct reclaim */ |
4949 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4950 | goto nopage; |
4951 | ||
a8161d1e VB |
4952 | /* Try direct reclaim and then allocating */ |
4953 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4954 | &did_some_progress); | |
4955 | if (page) | |
4956 | goto got_pg; | |
4957 | ||
4958 | /* Try direct compaction and then allocating */ | |
a9263751 | 4959 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4960 | compact_priority, &compact_result); |
56de7263 MG |
4961 | if (page) |
4962 | goto got_pg; | |
75f30861 | 4963 | |
9083905a JW |
4964 | /* Do not loop if specifically requested */ |
4965 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4966 | goto nopage; |
9083905a | 4967 | |
0a0337e0 MH |
4968 | /* |
4969 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4970 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4971 | */ |
dcda9b04 | 4972 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4973 | goto nopage; |
0a0337e0 | 4974 | |
0a0337e0 | 4975 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4976 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4977 | goto retry; |
4978 | ||
33c2d214 MH |
4979 | /* |
4980 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4981 | * reclaim is not able to make any progress because the current | |
4982 | * implementation of the compaction depends on the sufficient amount | |
4983 | * of free memory (see __compaction_suitable) | |
4984 | */ | |
4985 | if (did_some_progress > 0 && | |
86a294a8 | 4986 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4987 | compact_result, &compact_priority, |
d9436498 | 4988 | &compaction_retries)) |
33c2d214 MH |
4989 | goto retry; |
4990 | ||
902b6281 VB |
4991 | |
4992 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4993 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4994 | goto retry_cpuset; |
4995 | ||
9083905a JW |
4996 | /* Reclaim has failed us, start killing things */ |
4997 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4998 | if (page) | |
4999 | goto got_pg; | |
5000 | ||
9a67f648 | 5001 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e | 5002 | if (tsk_is_oom_victim(current) && |
8510e69c | 5003 | (alloc_flags & ALLOC_OOM || |
c288983d | 5004 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
5005 | goto nopage; |
5006 | ||
9083905a | 5007 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
5008 | if (did_some_progress) { |
5009 | no_progress_loops = 0; | |
9083905a | 5010 | goto retry; |
0a0337e0 | 5011 | } |
9083905a | 5012 | |
1da177e4 | 5013 | nopage: |
902b6281 VB |
5014 | /* Deal with possible cpuset update races before we fail */ |
5015 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
5016 | goto retry_cpuset; |
5017 | ||
9a67f648 MH |
5018 | /* |
5019 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
5020 | * we always retry | |
5021 | */ | |
5022 | if (gfp_mask & __GFP_NOFAIL) { | |
5023 | /* | |
5024 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
5025 | * of any new users that actually require GFP_NOWAIT | |
5026 | */ | |
5027 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
5028 | goto fail; | |
5029 | ||
5030 | /* | |
5031 | * PF_MEMALLOC request from this context is rather bizarre | |
5032 | * because we cannot reclaim anything and only can loop waiting | |
5033 | * for somebody to do a work for us | |
5034 | */ | |
5035 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
5036 | ||
5037 | /* | |
5038 | * non failing costly orders are a hard requirement which we | |
5039 | * are not prepared for much so let's warn about these users | |
5040 | * so that we can identify them and convert them to something | |
5041 | * else. | |
5042 | */ | |
5043 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
5044 | ||
6c18ba7a MH |
5045 | /* |
5046 | * Help non-failing allocations by giving them access to memory | |
5047 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
5048 | * could deplete whole memory reserves which would just make | |
5049 | * the situation worse | |
5050 | */ | |
5051 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
5052 | if (page) | |
5053 | goto got_pg; | |
5054 | ||
9a67f648 MH |
5055 | cond_resched(); |
5056 | goto retry; | |
5057 | } | |
5058 | fail: | |
a8e99259 | 5059 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 5060 | "page allocation failure: order:%u", order); |
1da177e4 | 5061 | got_pg: |
072bb0aa | 5062 | return page; |
1da177e4 | 5063 | } |
11e33f6a | 5064 | |
9cd75558 | 5065 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 5066 | int preferred_nid, nodemask_t *nodemask, |
8e6a930b | 5067 | struct alloc_context *ac, gfp_t *alloc_gfp, |
9cd75558 | 5068 | unsigned int *alloc_flags) |
11e33f6a | 5069 | { |
97a225e6 | 5070 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 5071 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 5072 | ac->nodemask = nodemask; |
01c0bfe0 | 5073 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 5074 | |
682a3385 | 5075 | if (cpusets_enabled()) { |
8e6a930b | 5076 | *alloc_gfp |= __GFP_HARDWALL; |
182f3d7a MS |
5077 | /* |
5078 | * When we are in the interrupt context, it is irrelevant | |
5079 | * to the current task context. It means that any node ok. | |
5080 | */ | |
5081 | if (!in_interrupt() && !ac->nodemask) | |
9cd75558 | 5082 | ac->nodemask = &cpuset_current_mems_allowed; |
51047820 VB |
5083 | else |
5084 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
5085 | } |
5086 | ||
d92a8cfc PZ |
5087 | fs_reclaim_acquire(gfp_mask); |
5088 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 5089 | |
d0164adc | 5090 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
5091 | |
5092 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 5093 | return false; |
11e33f6a | 5094 | |
8e3560d9 | 5095 | *alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags); |
d883c6cf | 5096 | |
c9ab0c4f | 5097 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 5098 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 5099 | |
e46e7b77 MG |
5100 | /* |
5101 | * The preferred zone is used for statistics but crucially it is | |
5102 | * also used as the starting point for the zonelist iterator. It | |
5103 | * may get reset for allocations that ignore memory policies. | |
5104 | */ | |
9cd75558 | 5105 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 5106 | ac->highest_zoneidx, ac->nodemask); |
a0622d05 MN |
5107 | |
5108 | return true; | |
9cd75558 MG |
5109 | } |
5110 | ||
387ba26f | 5111 | /* |
0f87d9d3 | 5112 | * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array |
387ba26f MG |
5113 | * @gfp: GFP flags for the allocation |
5114 | * @preferred_nid: The preferred NUMA node ID to allocate from | |
5115 | * @nodemask: Set of nodes to allocate from, may be NULL | |
0f87d9d3 MG |
5116 | * @nr_pages: The number of pages desired on the list or array |
5117 | * @page_list: Optional list to store the allocated pages | |
5118 | * @page_array: Optional array to store the pages | |
387ba26f MG |
5119 | * |
5120 | * This is a batched version of the page allocator that attempts to | |
0f87d9d3 MG |
5121 | * allocate nr_pages quickly. Pages are added to page_list if page_list |
5122 | * is not NULL, otherwise it is assumed that the page_array is valid. | |
387ba26f | 5123 | * |
0f87d9d3 MG |
5124 | * For lists, nr_pages is the number of pages that should be allocated. |
5125 | * | |
5126 | * For arrays, only NULL elements are populated with pages and nr_pages | |
5127 | * is the maximum number of pages that will be stored in the array. | |
5128 | * | |
5129 | * Returns the number of pages on the list or array. | |
387ba26f MG |
5130 | */ |
5131 | unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, | |
5132 | nodemask_t *nodemask, int nr_pages, | |
0f87d9d3 MG |
5133 | struct list_head *page_list, |
5134 | struct page **page_array) | |
387ba26f MG |
5135 | { |
5136 | struct page *page; | |
5137 | unsigned long flags; | |
5138 | struct zone *zone; | |
5139 | struct zoneref *z; | |
5140 | struct per_cpu_pages *pcp; | |
5141 | struct list_head *pcp_list; | |
5142 | struct alloc_context ac; | |
5143 | gfp_t alloc_gfp; | |
5144 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
3e23060b | 5145 | int nr_populated = 0, nr_account = 0; |
387ba26f | 5146 | |
ce76f9a1 | 5147 | if (unlikely(nr_pages <= 0)) |
387ba26f MG |
5148 | return 0; |
5149 | ||
0f87d9d3 MG |
5150 | /* |
5151 | * Skip populated array elements to determine if any pages need | |
5152 | * to be allocated before disabling IRQs. | |
5153 | */ | |
b08e50dd | 5154 | while (page_array && nr_populated < nr_pages && page_array[nr_populated]) |
0f87d9d3 MG |
5155 | nr_populated++; |
5156 | ||
b3b64ebd MG |
5157 | /* Already populated array? */ |
5158 | if (unlikely(page_array && nr_pages - nr_populated == 0)) | |
ff4b2b40 | 5159 | return nr_populated; |
b3b64ebd | 5160 | |
387ba26f | 5161 | /* Use the single page allocator for one page. */ |
0f87d9d3 | 5162 | if (nr_pages - nr_populated == 1) |
387ba26f MG |
5163 | goto failed; |
5164 | ||
5165 | /* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */ | |
5166 | gfp &= gfp_allowed_mask; | |
5167 | alloc_gfp = gfp; | |
5168 | if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags)) | |
5169 | return 0; | |
5170 | gfp = alloc_gfp; | |
5171 | ||
5172 | /* Find an allowed local zone that meets the low watermark. */ | |
5173 | for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) { | |
5174 | unsigned long mark; | |
5175 | ||
5176 | if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) && | |
5177 | !__cpuset_zone_allowed(zone, gfp)) { | |
5178 | continue; | |
5179 | } | |
5180 | ||
5181 | if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone && | |
5182 | zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) { | |
5183 | goto failed; | |
5184 | } | |
5185 | ||
5186 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages; | |
5187 | if (zone_watermark_fast(zone, 0, mark, | |
5188 | zonelist_zone_idx(ac.preferred_zoneref), | |
5189 | alloc_flags, gfp)) { | |
5190 | break; | |
5191 | } | |
5192 | } | |
5193 | ||
5194 | /* | |
5195 | * If there are no allowed local zones that meets the watermarks then | |
5196 | * try to allocate a single page and reclaim if necessary. | |
5197 | */ | |
ce76f9a1 | 5198 | if (unlikely(!zone)) |
387ba26f MG |
5199 | goto failed; |
5200 | ||
5201 | /* Attempt the batch allocation */ | |
dbbee9d5 | 5202 | local_lock_irqsave(&pagesets.lock, flags); |
28f836b6 | 5203 | pcp = this_cpu_ptr(zone->per_cpu_pageset); |
387ba26f MG |
5204 | pcp_list = &pcp->lists[ac.migratetype]; |
5205 | ||
0f87d9d3 MG |
5206 | while (nr_populated < nr_pages) { |
5207 | ||
5208 | /* Skip existing pages */ | |
5209 | if (page_array && page_array[nr_populated]) { | |
5210 | nr_populated++; | |
5211 | continue; | |
5212 | } | |
5213 | ||
387ba26f MG |
5214 | page = __rmqueue_pcplist(zone, ac.migratetype, alloc_flags, |
5215 | pcp, pcp_list); | |
ce76f9a1 | 5216 | if (unlikely(!page)) { |
387ba26f | 5217 | /* Try and get at least one page */ |
0f87d9d3 | 5218 | if (!nr_populated) |
387ba26f MG |
5219 | goto failed_irq; |
5220 | break; | |
5221 | } | |
3e23060b | 5222 | nr_account++; |
387ba26f MG |
5223 | |
5224 | prep_new_page(page, 0, gfp, 0); | |
0f87d9d3 MG |
5225 | if (page_list) |
5226 | list_add(&page->lru, page_list); | |
5227 | else | |
5228 | page_array[nr_populated] = page; | |
5229 | nr_populated++; | |
387ba26f MG |
5230 | } |
5231 | ||
43c95bcc MG |
5232 | local_unlock_irqrestore(&pagesets.lock, flags); |
5233 | ||
3e23060b MG |
5234 | __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account); |
5235 | zone_statistics(ac.preferred_zoneref->zone, zone, nr_account); | |
5236 | ||
0f87d9d3 | 5237 | return nr_populated; |
387ba26f MG |
5238 | |
5239 | failed_irq: | |
dbbee9d5 | 5240 | local_unlock_irqrestore(&pagesets.lock, flags); |
387ba26f MG |
5241 | |
5242 | failed: | |
5243 | page = __alloc_pages(gfp, 0, preferred_nid, nodemask); | |
5244 | if (page) { | |
0f87d9d3 MG |
5245 | if (page_list) |
5246 | list_add(&page->lru, page_list); | |
5247 | else | |
5248 | page_array[nr_populated] = page; | |
5249 | nr_populated++; | |
387ba26f MG |
5250 | } |
5251 | ||
0f87d9d3 | 5252 | return nr_populated; |
387ba26f MG |
5253 | } |
5254 | EXPORT_SYMBOL_GPL(__alloc_pages_bulk); | |
5255 | ||
9cd75558 MG |
5256 | /* |
5257 | * This is the 'heart' of the zoned buddy allocator. | |
5258 | */ | |
84172f4b | 5259 | struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, |
04ec6264 | 5260 | nodemask_t *nodemask) |
9cd75558 MG |
5261 | { |
5262 | struct page *page; | |
5263 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
8e6a930b | 5264 | gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
5265 | struct alloc_context ac = { }; |
5266 | ||
c63ae43b MH |
5267 | /* |
5268 | * There are several places where we assume that the order value is sane | |
5269 | * so bail out early if the request is out of bound. | |
5270 | */ | |
5271 | if (unlikely(order >= MAX_ORDER)) { | |
6e5e0f28 | 5272 | WARN_ON_ONCE(!(gfp & __GFP_NOWARN)); |
c63ae43b MH |
5273 | return NULL; |
5274 | } | |
5275 | ||
6e5e0f28 | 5276 | gfp &= gfp_allowed_mask; |
da6df1b0 PT |
5277 | /* |
5278 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS | |
5279 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
5280 | * from a particular context which has been marked by | |
8e3560d9 PT |
5281 | * memalloc_no{fs,io}_{save,restore}. And PF_MEMALLOC_PIN which ensures |
5282 | * movable zones are not used during allocation. | |
da6df1b0 PT |
5283 | */ |
5284 | gfp = current_gfp_context(gfp); | |
6e5e0f28 MWO |
5285 | alloc_gfp = gfp; |
5286 | if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac, | |
8e6a930b | 5287 | &alloc_gfp, &alloc_flags)) |
9cd75558 MG |
5288 | return NULL; |
5289 | ||
6bb15450 MG |
5290 | /* |
5291 | * Forbid the first pass from falling back to types that fragment | |
5292 | * memory until all local zones are considered. | |
5293 | */ | |
6e5e0f28 | 5294 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp); |
6bb15450 | 5295 | |
5117f45d | 5296 | /* First allocation attempt */ |
8e6a930b | 5297 | page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac); |
4fcb0971 MG |
5298 | if (likely(page)) |
5299 | goto out; | |
11e33f6a | 5300 | |
da6df1b0 | 5301 | alloc_gfp = gfp; |
4fcb0971 | 5302 | ac.spread_dirty_pages = false; |
23f086f9 | 5303 | |
4741526b MG |
5304 | /* |
5305 | * Restore the original nodemask if it was potentially replaced with | |
5306 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
5307 | */ | |
97ce86f9 | 5308 | ac.nodemask = nodemask; |
16096c25 | 5309 | |
8e6a930b | 5310 | page = __alloc_pages_slowpath(alloc_gfp, order, &ac); |
cc9a6c87 | 5311 | |
4fcb0971 | 5312 | out: |
6e5e0f28 MWO |
5313 | if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT) && page && |
5314 | unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) { | |
c4159a75 VD |
5315 | __free_pages(page, order); |
5316 | page = NULL; | |
4949148a VD |
5317 | } |
5318 | ||
8e6a930b | 5319 | trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype); |
4fcb0971 | 5320 | |
11e33f6a | 5321 | return page; |
1da177e4 | 5322 | } |
84172f4b | 5323 | EXPORT_SYMBOL(__alloc_pages); |
1da177e4 LT |
5324 | |
5325 | /* | |
9ea9a680 MH |
5326 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
5327 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
5328 | * you need to access high mem. | |
1da177e4 | 5329 | */ |
920c7a5d | 5330 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 5331 | { |
945a1113 AM |
5332 | struct page *page; |
5333 | ||
9ea9a680 | 5334 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
5335 | if (!page) |
5336 | return 0; | |
5337 | return (unsigned long) page_address(page); | |
5338 | } | |
1da177e4 LT |
5339 | EXPORT_SYMBOL(__get_free_pages); |
5340 | ||
920c7a5d | 5341 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 5342 | { |
945a1113 | 5343 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 5344 | } |
1da177e4 LT |
5345 | EXPORT_SYMBOL(get_zeroed_page); |
5346 | ||
7f194fbb MWO |
5347 | /** |
5348 | * __free_pages - Free pages allocated with alloc_pages(). | |
5349 | * @page: The page pointer returned from alloc_pages(). | |
5350 | * @order: The order of the allocation. | |
5351 | * | |
5352 | * This function can free multi-page allocations that are not compound | |
5353 | * pages. It does not check that the @order passed in matches that of | |
5354 | * the allocation, so it is easy to leak memory. Freeing more memory | |
5355 | * than was allocated will probably emit a warning. | |
5356 | * | |
5357 | * If the last reference to this page is speculative, it will be released | |
5358 | * by put_page() which only frees the first page of a non-compound | |
5359 | * allocation. To prevent the remaining pages from being leaked, we free | |
5360 | * the subsequent pages here. If you want to use the page's reference | |
5361 | * count to decide when to free the allocation, you should allocate a | |
5362 | * compound page, and use put_page() instead of __free_pages(). | |
5363 | * | |
5364 | * Context: May be called in interrupt context or while holding a normal | |
5365 | * spinlock, but not in NMI context or while holding a raw spinlock. | |
5366 | */ | |
742aa7fb AL |
5367 | void __free_pages(struct page *page, unsigned int order) |
5368 | { | |
5369 | if (put_page_testzero(page)) | |
5370 | free_the_page(page, order); | |
e320d301 MWO |
5371 | else if (!PageHead(page)) |
5372 | while (order-- > 0) | |
5373 | free_the_page(page + (1 << order), order); | |
742aa7fb | 5374 | } |
1da177e4 LT |
5375 | EXPORT_SYMBOL(__free_pages); |
5376 | ||
920c7a5d | 5377 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
5378 | { |
5379 | if (addr != 0) { | |
725d704e | 5380 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
5381 | __free_pages(virt_to_page((void *)addr), order); |
5382 | } | |
5383 | } | |
5384 | ||
5385 | EXPORT_SYMBOL(free_pages); | |
5386 | ||
b63ae8ca AD |
5387 | /* |
5388 | * Page Fragment: | |
5389 | * An arbitrary-length arbitrary-offset area of memory which resides | |
5390 | * within a 0 or higher order page. Multiple fragments within that page | |
5391 | * are individually refcounted, in the page's reference counter. | |
5392 | * | |
5393 | * The page_frag functions below provide a simple allocation framework for | |
5394 | * page fragments. This is used by the network stack and network device | |
5395 | * drivers to provide a backing region of memory for use as either an | |
5396 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
5397 | */ | |
2976db80 AD |
5398 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
5399 | gfp_t gfp_mask) | |
b63ae8ca AD |
5400 | { |
5401 | struct page *page = NULL; | |
5402 | gfp_t gfp = gfp_mask; | |
5403 | ||
5404 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5405 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
5406 | __GFP_NOMEMALLOC; | |
5407 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
5408 | PAGE_FRAG_CACHE_MAX_ORDER); | |
5409 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
5410 | #endif | |
5411 | if (unlikely(!page)) | |
5412 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
5413 | ||
5414 | nc->va = page ? page_address(page) : NULL; | |
5415 | ||
5416 | return page; | |
5417 | } | |
5418 | ||
2976db80 | 5419 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
5420 | { |
5421 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
5422 | ||
742aa7fb AL |
5423 | if (page_ref_sub_and_test(page, count)) |
5424 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 5425 | } |
2976db80 | 5426 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 5427 | |
b358e212 KH |
5428 | void *page_frag_alloc_align(struct page_frag_cache *nc, |
5429 | unsigned int fragsz, gfp_t gfp_mask, | |
5430 | unsigned int align_mask) | |
b63ae8ca AD |
5431 | { |
5432 | unsigned int size = PAGE_SIZE; | |
5433 | struct page *page; | |
5434 | int offset; | |
5435 | ||
5436 | if (unlikely(!nc->va)) { | |
5437 | refill: | |
2976db80 | 5438 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
5439 | if (!page) |
5440 | return NULL; | |
5441 | ||
5442 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5443 | /* if size can vary use size else just use PAGE_SIZE */ | |
5444 | size = nc->size; | |
5445 | #endif | |
5446 | /* Even if we own the page, we do not use atomic_set(). | |
5447 | * This would break get_page_unless_zero() users. | |
5448 | */ | |
86447726 | 5449 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
5450 | |
5451 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 5452 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 5453 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5454 | nc->offset = size; |
5455 | } | |
5456 | ||
5457 | offset = nc->offset - fragsz; | |
5458 | if (unlikely(offset < 0)) { | |
5459 | page = virt_to_page(nc->va); | |
5460 | ||
fe896d18 | 5461 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
5462 | goto refill; |
5463 | ||
d8c19014 DZ |
5464 | if (unlikely(nc->pfmemalloc)) { |
5465 | free_the_page(page, compound_order(page)); | |
5466 | goto refill; | |
5467 | } | |
5468 | ||
b63ae8ca AD |
5469 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
5470 | /* if size can vary use size else just use PAGE_SIZE */ | |
5471 | size = nc->size; | |
5472 | #endif | |
5473 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 5474 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
5475 | |
5476 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 5477 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5478 | offset = size - fragsz; |
5479 | } | |
5480 | ||
5481 | nc->pagecnt_bias--; | |
b358e212 | 5482 | offset &= align_mask; |
b63ae8ca AD |
5483 | nc->offset = offset; |
5484 | ||
5485 | return nc->va + offset; | |
5486 | } | |
b358e212 | 5487 | EXPORT_SYMBOL(page_frag_alloc_align); |
b63ae8ca AD |
5488 | |
5489 | /* | |
5490 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
5491 | */ | |
8c2dd3e4 | 5492 | void page_frag_free(void *addr) |
b63ae8ca AD |
5493 | { |
5494 | struct page *page = virt_to_head_page(addr); | |
5495 | ||
742aa7fb AL |
5496 | if (unlikely(put_page_testzero(page))) |
5497 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 5498 | } |
8c2dd3e4 | 5499 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 5500 | |
d00181b9 KS |
5501 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
5502 | size_t size) | |
ee85c2e1 AK |
5503 | { |
5504 | if (addr) { | |
5505 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
5506 | unsigned long used = addr + PAGE_ALIGN(size); | |
5507 | ||
5508 | split_page(virt_to_page((void *)addr), order); | |
5509 | while (used < alloc_end) { | |
5510 | free_page(used); | |
5511 | used += PAGE_SIZE; | |
5512 | } | |
5513 | } | |
5514 | return (void *)addr; | |
5515 | } | |
5516 | ||
2be0ffe2 TT |
5517 | /** |
5518 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
5519 | * @size: the number of bytes to allocate | |
63931eb9 | 5520 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
5521 | * |
5522 | * This function is similar to alloc_pages(), except that it allocates the | |
5523 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
5524 | * allocate memory in power-of-two pages. | |
5525 | * | |
5526 | * This function is also limited by MAX_ORDER. | |
5527 | * | |
5528 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
5529 | * |
5530 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
5531 | */ |
5532 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
5533 | { | |
5534 | unsigned int order = get_order(size); | |
5535 | unsigned long addr; | |
5536 | ||
63931eb9 VB |
5537 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) |
5538 | gfp_mask &= ~__GFP_COMP; | |
5539 | ||
2be0ffe2 | 5540 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 5541 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
5542 | } |
5543 | EXPORT_SYMBOL(alloc_pages_exact); | |
5544 | ||
ee85c2e1 AK |
5545 | /** |
5546 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
5547 | * pages on a node. | |
b5e6ab58 | 5548 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 5549 | * @size: the number of bytes to allocate |
63931eb9 | 5550 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
5551 | * |
5552 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
5553 | * back. | |
a862f68a MR |
5554 | * |
5555 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 5556 | */ |
e1931811 | 5557 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5558 | { |
d00181b9 | 5559 | unsigned int order = get_order(size); |
63931eb9 VB |
5560 | struct page *p; |
5561 | ||
5562 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) | |
5563 | gfp_mask &= ~__GFP_COMP; | |
5564 | ||
5565 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5566 | if (!p) |
5567 | return NULL; | |
5568 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5569 | } | |
ee85c2e1 | 5570 | |
2be0ffe2 TT |
5571 | /** |
5572 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5573 | * @virt: the value returned by alloc_pages_exact. | |
5574 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5575 | * | |
5576 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5577 | */ | |
5578 | void free_pages_exact(void *virt, size_t size) | |
5579 | { | |
5580 | unsigned long addr = (unsigned long)virt; | |
5581 | unsigned long end = addr + PAGE_ALIGN(size); | |
5582 | ||
5583 | while (addr < end) { | |
5584 | free_page(addr); | |
5585 | addr += PAGE_SIZE; | |
5586 | } | |
5587 | } | |
5588 | EXPORT_SYMBOL(free_pages_exact); | |
5589 | ||
e0fb5815 ZY |
5590 | /** |
5591 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5592 | * @offset: The zone index of the highest zone | |
5593 | * | |
a862f68a | 5594 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5595 | * high watermark within all zones at or below a given zone index. For each |
5596 | * zone, the number of pages is calculated as: | |
0e056eb5 | 5597 | * |
5598 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5599 | * |
5600 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5601 | */ |
ebec3862 | 5602 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5603 | { |
dd1a239f | 5604 | struct zoneref *z; |
54a6eb5c MG |
5605 | struct zone *zone; |
5606 | ||
e310fd43 | 5607 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5608 | unsigned long sum = 0; |
1da177e4 | 5609 | |
0e88460d | 5610 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5611 | |
54a6eb5c | 5612 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5613 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5614 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5615 | if (size > high) |
5616 | sum += size - high; | |
1da177e4 LT |
5617 | } |
5618 | ||
5619 | return sum; | |
5620 | } | |
5621 | ||
e0fb5815 ZY |
5622 | /** |
5623 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5624 | * | |
5625 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5626 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5627 | * |
5628 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5629 | * ZONE_NORMAL. | |
1da177e4 | 5630 | */ |
ebec3862 | 5631 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5632 | { |
af4ca457 | 5633 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5634 | } |
c2f1a551 | 5635 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5636 | |
08e0f6a9 | 5637 | static inline void show_node(struct zone *zone) |
1da177e4 | 5638 | { |
e5adfffc | 5639 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5640 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5641 | } |
1da177e4 | 5642 | |
d02bd27b IR |
5643 | long si_mem_available(void) |
5644 | { | |
5645 | long available; | |
5646 | unsigned long pagecache; | |
5647 | unsigned long wmark_low = 0; | |
5648 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5649 | unsigned long reclaimable; |
d02bd27b IR |
5650 | struct zone *zone; |
5651 | int lru; | |
5652 | ||
5653 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5654 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5655 | |
5656 | for_each_zone(zone) | |
a9214443 | 5657 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5658 | |
5659 | /* | |
5660 | * Estimate the amount of memory available for userspace allocations, | |
5661 | * without causing swapping. | |
5662 | */ | |
c41f012a | 5663 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5664 | |
5665 | /* | |
5666 | * Not all the page cache can be freed, otherwise the system will | |
5667 | * start swapping. Assume at least half of the page cache, or the | |
5668 | * low watermark worth of cache, needs to stay. | |
5669 | */ | |
5670 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
5671 | pagecache -= min(pagecache / 2, wmark_low); | |
5672 | available += pagecache; | |
5673 | ||
5674 | /* | |
b29940c1 VB |
5675 | * Part of the reclaimable slab and other kernel memory consists of |
5676 | * items that are in use, and cannot be freed. Cap this estimate at the | |
5677 | * low watermark. | |
d02bd27b | 5678 | */ |
d42f3245 RG |
5679 | reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) + |
5680 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
b29940c1 | 5681 | available += reclaimable - min(reclaimable / 2, wmark_low); |
034ebf65 | 5682 | |
d02bd27b IR |
5683 | if (available < 0) |
5684 | available = 0; | |
5685 | return available; | |
5686 | } | |
5687 | EXPORT_SYMBOL_GPL(si_mem_available); | |
5688 | ||
1da177e4 LT |
5689 | void si_meminfo(struct sysinfo *val) |
5690 | { | |
ca79b0c2 | 5691 | val->totalram = totalram_pages(); |
11fb9989 | 5692 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 5693 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 5694 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 5695 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 5696 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
5697 | val->mem_unit = PAGE_SIZE; |
5698 | } | |
5699 | ||
5700 | EXPORT_SYMBOL(si_meminfo); | |
5701 | ||
5702 | #ifdef CONFIG_NUMA | |
5703 | void si_meminfo_node(struct sysinfo *val, int nid) | |
5704 | { | |
cdd91a77 JL |
5705 | int zone_type; /* needs to be signed */ |
5706 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
5707 | unsigned long managed_highpages = 0; |
5708 | unsigned long free_highpages = 0; | |
1da177e4 LT |
5709 | pg_data_t *pgdat = NODE_DATA(nid); |
5710 | ||
cdd91a77 | 5711 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 5712 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 5713 | val->totalram = managed_pages; |
11fb9989 | 5714 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 5715 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 5716 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
5717 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
5718 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
5719 | ||
5720 | if (is_highmem(zone)) { | |
9705bea5 | 5721 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
5722 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
5723 | } | |
5724 | } | |
5725 | val->totalhigh = managed_highpages; | |
5726 | val->freehigh = free_highpages; | |
98d2b0eb | 5727 | #else |
fc2bd799 JK |
5728 | val->totalhigh = managed_highpages; |
5729 | val->freehigh = free_highpages; | |
98d2b0eb | 5730 | #endif |
1da177e4 LT |
5731 | val->mem_unit = PAGE_SIZE; |
5732 | } | |
5733 | #endif | |
5734 | ||
ddd588b5 | 5735 | /* |
7bf02ea2 DR |
5736 | * Determine whether the node should be displayed or not, depending on whether |
5737 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 5738 | */ |
9af744d7 | 5739 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 5740 | { |
ddd588b5 | 5741 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 5742 | return false; |
ddd588b5 | 5743 | |
9af744d7 MH |
5744 | /* |
5745 | * no node mask - aka implicit memory numa policy. Do not bother with | |
5746 | * the synchronization - read_mems_allowed_begin - because we do not | |
5747 | * have to be precise here. | |
5748 | */ | |
5749 | if (!nodemask) | |
5750 | nodemask = &cpuset_current_mems_allowed; | |
5751 | ||
5752 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
5753 | } |
5754 | ||
1da177e4 LT |
5755 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
5756 | ||
377e4f16 RV |
5757 | static void show_migration_types(unsigned char type) |
5758 | { | |
5759 | static const char types[MIGRATE_TYPES] = { | |
5760 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 5761 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
5762 | [MIGRATE_RECLAIMABLE] = 'E', |
5763 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
5764 | #ifdef CONFIG_CMA |
5765 | [MIGRATE_CMA] = 'C', | |
5766 | #endif | |
194159fb | 5767 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 5768 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 5769 | #endif |
377e4f16 RV |
5770 | }; |
5771 | char tmp[MIGRATE_TYPES + 1]; | |
5772 | char *p = tmp; | |
5773 | int i; | |
5774 | ||
5775 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
5776 | if (type & (1 << i)) | |
5777 | *p++ = types[i]; | |
5778 | } | |
5779 | ||
5780 | *p = '\0'; | |
1f84a18f | 5781 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
5782 | } |
5783 | ||
1da177e4 LT |
5784 | /* |
5785 | * Show free area list (used inside shift_scroll-lock stuff) | |
5786 | * We also calculate the percentage fragmentation. We do this by counting the | |
5787 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
5788 | * |
5789 | * Bits in @filter: | |
5790 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
5791 | * cpuset. | |
1da177e4 | 5792 | */ |
9af744d7 | 5793 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 5794 | { |
d1bfcdb8 | 5795 | unsigned long free_pcp = 0; |
c7241913 | 5796 | int cpu; |
1da177e4 | 5797 | struct zone *zone; |
599d0c95 | 5798 | pg_data_t *pgdat; |
1da177e4 | 5799 | |
ee99c71c | 5800 | for_each_populated_zone(zone) { |
9af744d7 | 5801 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5802 | continue; |
d1bfcdb8 | 5803 | |
761b0677 | 5804 | for_each_online_cpu(cpu) |
28f836b6 | 5805 | free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count; |
1da177e4 LT |
5806 | } |
5807 | ||
a731286d KM |
5808 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
5809 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
8d92890b | 5810 | " unevictable:%lu dirty:%lu writeback:%lu\n" |
d1bfcdb8 | 5811 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a | 5812 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 5813 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
5814 | global_node_page_state(NR_ACTIVE_ANON), |
5815 | global_node_page_state(NR_INACTIVE_ANON), | |
5816 | global_node_page_state(NR_ISOLATED_ANON), | |
5817 | global_node_page_state(NR_ACTIVE_FILE), | |
5818 | global_node_page_state(NR_INACTIVE_FILE), | |
5819 | global_node_page_state(NR_ISOLATED_FILE), | |
5820 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
5821 | global_node_page_state(NR_FILE_DIRTY), |
5822 | global_node_page_state(NR_WRITEBACK), | |
d42f3245 RG |
5823 | global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B), |
5824 | global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B), | |
50658e2e | 5825 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 5826 | global_node_page_state(NR_SHMEM), |
f0c0c115 | 5827 | global_node_page_state(NR_PAGETABLE), |
c41f012a MH |
5828 | global_zone_page_state(NR_BOUNCE), |
5829 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 5830 | free_pcp, |
c41f012a | 5831 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 5832 | |
599d0c95 | 5833 | for_each_online_pgdat(pgdat) { |
9af744d7 | 5834 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
5835 | continue; |
5836 | ||
599d0c95 MG |
5837 | printk("Node %d" |
5838 | " active_anon:%lukB" | |
5839 | " inactive_anon:%lukB" | |
5840 | " active_file:%lukB" | |
5841 | " inactive_file:%lukB" | |
5842 | " unevictable:%lukB" | |
5843 | " isolated(anon):%lukB" | |
5844 | " isolated(file):%lukB" | |
50658e2e | 5845 | " mapped:%lukB" |
11fb9989 MG |
5846 | " dirty:%lukB" |
5847 | " writeback:%lukB" | |
5848 | " shmem:%lukB" | |
5849 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5850 | " shmem_thp: %lukB" | |
5851 | " shmem_pmdmapped: %lukB" | |
5852 | " anon_thp: %lukB" | |
5853 | #endif | |
5854 | " writeback_tmp:%lukB" | |
991e7673 SB |
5855 | " kernel_stack:%lukB" |
5856 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5857 | " shadow_call_stack:%lukB" | |
5858 | #endif | |
f0c0c115 | 5859 | " pagetables:%lukB" |
599d0c95 MG |
5860 | " all_unreclaimable? %s" |
5861 | "\n", | |
5862 | pgdat->node_id, | |
5863 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
5864 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
5865 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
5866 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
5867 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
5868 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
5869 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 5870 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
5871 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
5872 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 5873 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 | 5874 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
57b2847d | 5875 | K(node_page_state(pgdat, NR_SHMEM_THPS)), |
a1528e21 | 5876 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)), |
69473e5d | 5877 | K(node_page_state(pgdat, NR_ANON_THPS)), |
11fb9989 | 5878 | #endif |
11fb9989 | 5879 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
991e7673 SB |
5880 | node_page_state(pgdat, NR_KERNEL_STACK_KB), |
5881 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5882 | node_page_state(pgdat, NR_KERNEL_SCS_KB), | |
5883 | #endif | |
f0c0c115 | 5884 | K(node_page_state(pgdat, NR_PAGETABLE)), |
c73322d0 JW |
5885 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
5886 | "yes" : "no"); | |
599d0c95 MG |
5887 | } |
5888 | ||
ee99c71c | 5889 | for_each_populated_zone(zone) { |
1da177e4 LT |
5890 | int i; |
5891 | ||
9af744d7 | 5892 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5893 | continue; |
d1bfcdb8 KK |
5894 | |
5895 | free_pcp = 0; | |
5896 | for_each_online_cpu(cpu) | |
28f836b6 | 5897 | free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count; |
d1bfcdb8 | 5898 | |
1da177e4 | 5899 | show_node(zone); |
1f84a18f JP |
5900 | printk(KERN_CONT |
5901 | "%s" | |
1da177e4 LT |
5902 | " free:%lukB" |
5903 | " min:%lukB" | |
5904 | " low:%lukB" | |
5905 | " high:%lukB" | |
e47b346a | 5906 | " reserved_highatomic:%luKB" |
71c799f4 MK |
5907 | " active_anon:%lukB" |
5908 | " inactive_anon:%lukB" | |
5909 | " active_file:%lukB" | |
5910 | " inactive_file:%lukB" | |
5911 | " unevictable:%lukB" | |
5a1c84b4 | 5912 | " writepending:%lukB" |
1da177e4 | 5913 | " present:%lukB" |
9feedc9d | 5914 | " managed:%lukB" |
4a0aa73f | 5915 | " mlocked:%lukB" |
4a0aa73f | 5916 | " bounce:%lukB" |
d1bfcdb8 KK |
5917 | " free_pcp:%lukB" |
5918 | " local_pcp:%ukB" | |
d1ce749a | 5919 | " free_cma:%lukB" |
1da177e4 LT |
5920 | "\n", |
5921 | zone->name, | |
88f5acf8 | 5922 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
5923 | K(min_wmark_pages(zone)), |
5924 | K(low_wmark_pages(zone)), | |
5925 | K(high_wmark_pages(zone)), | |
e47b346a | 5926 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
5927 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
5928 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
5929 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
5930 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
5931 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 5932 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 5933 | K(zone->present_pages), |
9705bea5 | 5934 | K(zone_managed_pages(zone)), |
4a0aa73f | 5935 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f | 5936 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 | 5937 | K(free_pcp), |
28f836b6 | 5938 | K(this_cpu_read(zone->per_cpu_pageset->count)), |
33e077bd | 5939 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
5940 | printk("lowmem_reserve[]:"); |
5941 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
5942 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5943 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5944 | } |
5945 | ||
ee99c71c | 5946 | for_each_populated_zone(zone) { |
d00181b9 KS |
5947 | unsigned int order; |
5948 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5949 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5950 | |
9af744d7 | 5951 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5952 | continue; |
1da177e4 | 5953 | show_node(zone); |
1f84a18f | 5954 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5955 | |
5956 | spin_lock_irqsave(&zone->lock, flags); | |
5957 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5958 | struct free_area *area = &zone->free_area[order]; |
5959 | int type; | |
5960 | ||
5961 | nr[order] = area->nr_free; | |
8f9de51a | 5962 | total += nr[order] << order; |
377e4f16 RV |
5963 | |
5964 | types[order] = 0; | |
5965 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 5966 | if (!free_area_empty(area, type)) |
377e4f16 RV |
5967 | types[order] |= 1 << type; |
5968 | } | |
1da177e4 LT |
5969 | } |
5970 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5971 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5972 | printk(KERN_CONT "%lu*%lukB ", |
5973 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5974 | if (nr[order]) |
5975 | show_migration_types(types[order]); | |
5976 | } | |
1f84a18f | 5977 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5978 | } |
5979 | ||
949f7ec5 DR |
5980 | hugetlb_show_meminfo(); |
5981 | ||
11fb9989 | 5982 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5983 | |
1da177e4 LT |
5984 | show_swap_cache_info(); |
5985 | } | |
5986 | ||
19770b32 MG |
5987 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5988 | { | |
5989 | zoneref->zone = zone; | |
5990 | zoneref->zone_idx = zone_idx(zone); | |
5991 | } | |
5992 | ||
1da177e4 LT |
5993 | /* |
5994 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5995 | * |
5996 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5997 | */ |
9d3be21b | 5998 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5999 | { |
1a93205b | 6000 | struct zone *zone; |
bc732f1d | 6001 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 6002 | int nr_zones = 0; |
02a68a5e CL |
6003 | |
6004 | do { | |
2f6726e5 | 6005 | zone_type--; |
070f8032 | 6006 | zone = pgdat->node_zones + zone_type; |
6aa303de | 6007 | if (managed_zone(zone)) { |
9d3be21b | 6008 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 6009 | check_highest_zone(zone_type); |
1da177e4 | 6010 | } |
2f6726e5 | 6011 | } while (zone_type); |
bc732f1d | 6012 | |
070f8032 | 6013 | return nr_zones; |
1da177e4 LT |
6014 | } |
6015 | ||
6016 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
6017 | |
6018 | static int __parse_numa_zonelist_order(char *s) | |
6019 | { | |
c9bff3ee | 6020 | /* |
f0953a1b | 6021 | * We used to support different zonelists modes but they turned |
c9bff3ee MH |
6022 | * out to be just not useful. Let's keep the warning in place |
6023 | * if somebody still use the cmd line parameter so that we do | |
6024 | * not fail it silently | |
6025 | */ | |
6026 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
6027 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
6028 | return -EINVAL; |
6029 | } | |
6030 | return 0; | |
6031 | } | |
6032 | ||
c9bff3ee MH |
6033 | char numa_zonelist_order[] = "Node"; |
6034 | ||
f0c0b2b8 KH |
6035 | /* |
6036 | * sysctl handler for numa_zonelist_order | |
6037 | */ | |
cccad5b9 | 6038 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 6039 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 6040 | { |
32927393 CH |
6041 | if (write) |
6042 | return __parse_numa_zonelist_order(buffer); | |
6043 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
6044 | } |
6045 | ||
6046 | ||
62bc62a8 | 6047 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
6048 | static int node_load[MAX_NUMNODES]; |
6049 | ||
1da177e4 | 6050 | /** |
4dc3b16b | 6051 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
6052 | * @node: node whose fallback list we're appending |
6053 | * @used_node_mask: nodemask_t of already used nodes | |
6054 | * | |
6055 | * We use a number of factors to determine which is the next node that should | |
6056 | * appear on a given node's fallback list. The node should not have appeared | |
6057 | * already in @node's fallback list, and it should be the next closest node | |
6058 | * according to the distance array (which contains arbitrary distance values | |
6059 | * from each node to each node in the system), and should also prefer nodes | |
6060 | * with no CPUs, since presumably they'll have very little allocation pressure | |
6061 | * on them otherwise. | |
a862f68a MR |
6062 | * |
6063 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 6064 | */ |
f0c0b2b8 | 6065 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 6066 | { |
4cf808eb | 6067 | int n, val; |
1da177e4 | 6068 | int min_val = INT_MAX; |
00ef2d2f | 6069 | int best_node = NUMA_NO_NODE; |
1da177e4 | 6070 | |
4cf808eb LT |
6071 | /* Use the local node if we haven't already */ |
6072 | if (!node_isset(node, *used_node_mask)) { | |
6073 | node_set(node, *used_node_mask); | |
6074 | return node; | |
6075 | } | |
1da177e4 | 6076 | |
4b0ef1fe | 6077 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
6078 | |
6079 | /* Don't want a node to appear more than once */ | |
6080 | if (node_isset(n, *used_node_mask)) | |
6081 | continue; | |
6082 | ||
1da177e4 LT |
6083 | /* Use the distance array to find the distance */ |
6084 | val = node_distance(node, n); | |
6085 | ||
4cf808eb LT |
6086 | /* Penalize nodes under us ("prefer the next node") */ |
6087 | val += (n < node); | |
6088 | ||
1da177e4 | 6089 | /* Give preference to headless and unused nodes */ |
b630749f | 6090 | if (!cpumask_empty(cpumask_of_node(n))) |
1da177e4 LT |
6091 | val += PENALTY_FOR_NODE_WITH_CPUS; |
6092 | ||
6093 | /* Slight preference for less loaded node */ | |
6094 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
6095 | val += node_load[n]; | |
6096 | ||
6097 | if (val < min_val) { | |
6098 | min_val = val; | |
6099 | best_node = n; | |
6100 | } | |
6101 | } | |
6102 | ||
6103 | if (best_node >= 0) | |
6104 | node_set(best_node, *used_node_mask); | |
6105 | ||
6106 | return best_node; | |
6107 | } | |
6108 | ||
f0c0b2b8 KH |
6109 | |
6110 | /* | |
6111 | * Build zonelists ordered by node and zones within node. | |
6112 | * This results in maximum locality--normal zone overflows into local | |
6113 | * DMA zone, if any--but risks exhausting DMA zone. | |
6114 | */ | |
9d3be21b MH |
6115 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
6116 | unsigned nr_nodes) | |
1da177e4 | 6117 | { |
9d3be21b MH |
6118 | struct zoneref *zonerefs; |
6119 | int i; | |
6120 | ||
6121 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
6122 | ||
6123 | for (i = 0; i < nr_nodes; i++) { | |
6124 | int nr_zones; | |
6125 | ||
6126 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 6127 | |
9d3be21b MH |
6128 | nr_zones = build_zonerefs_node(node, zonerefs); |
6129 | zonerefs += nr_zones; | |
6130 | } | |
6131 | zonerefs->zone = NULL; | |
6132 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
6133 | } |
6134 | ||
523b9458 CL |
6135 | /* |
6136 | * Build gfp_thisnode zonelists | |
6137 | */ | |
6138 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
6139 | { | |
9d3be21b MH |
6140 | struct zoneref *zonerefs; |
6141 | int nr_zones; | |
523b9458 | 6142 | |
9d3be21b MH |
6143 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
6144 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6145 | zonerefs += nr_zones; | |
6146 | zonerefs->zone = NULL; | |
6147 | zonerefs->zone_idx = 0; | |
523b9458 CL |
6148 | } |
6149 | ||
f0c0b2b8 KH |
6150 | /* |
6151 | * Build zonelists ordered by zone and nodes within zones. | |
6152 | * This results in conserving DMA zone[s] until all Normal memory is | |
6153 | * exhausted, but results in overflowing to remote node while memory | |
6154 | * may still exist in local DMA zone. | |
6155 | */ | |
f0c0b2b8 | 6156 | |
f0c0b2b8 KH |
6157 | static void build_zonelists(pg_data_t *pgdat) |
6158 | { | |
9d3be21b MH |
6159 | static int node_order[MAX_NUMNODES]; |
6160 | int node, load, nr_nodes = 0; | |
d0ddf49b | 6161 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 6162 | int local_node, prev_node; |
1da177e4 LT |
6163 | |
6164 | /* NUMA-aware ordering of nodes */ | |
6165 | local_node = pgdat->node_id; | |
62bc62a8 | 6166 | load = nr_online_nodes; |
1da177e4 | 6167 | prev_node = local_node; |
f0c0b2b8 | 6168 | |
f0c0b2b8 | 6169 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
6170 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
6171 | /* | |
6172 | * We don't want to pressure a particular node. | |
6173 | * So adding penalty to the first node in same | |
6174 | * distance group to make it round-robin. | |
6175 | */ | |
957f822a DR |
6176 | if (node_distance(local_node, node) != |
6177 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
6178 | node_load[node] = load; |
6179 | ||
9d3be21b | 6180 | node_order[nr_nodes++] = node; |
1da177e4 LT |
6181 | prev_node = node; |
6182 | load--; | |
1da177e4 | 6183 | } |
523b9458 | 6184 | |
9d3be21b | 6185 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 6186 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
6187 | } |
6188 | ||
7aac7898 LS |
6189 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6190 | /* | |
6191 | * Return node id of node used for "local" allocations. | |
6192 | * I.e., first node id of first zone in arg node's generic zonelist. | |
6193 | * Used for initializing percpu 'numa_mem', which is used primarily | |
6194 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
6195 | */ | |
6196 | int local_memory_node(int node) | |
6197 | { | |
c33d6c06 | 6198 | struct zoneref *z; |
7aac7898 | 6199 | |
c33d6c06 | 6200 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 6201 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 6202 | NULL); |
c1093b74 | 6203 | return zone_to_nid(z->zone); |
7aac7898 LS |
6204 | } |
6205 | #endif | |
f0c0b2b8 | 6206 | |
6423aa81 JK |
6207 | static void setup_min_unmapped_ratio(void); |
6208 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
6209 | #else /* CONFIG_NUMA */ |
6210 | ||
f0c0b2b8 | 6211 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 6212 | { |
19655d34 | 6213 | int node, local_node; |
9d3be21b MH |
6214 | struct zoneref *zonerefs; |
6215 | int nr_zones; | |
1da177e4 LT |
6216 | |
6217 | local_node = pgdat->node_id; | |
1da177e4 | 6218 | |
9d3be21b MH |
6219 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
6220 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6221 | zonerefs += nr_zones; | |
1da177e4 | 6222 | |
54a6eb5c MG |
6223 | /* |
6224 | * Now we build the zonelist so that it contains the zones | |
6225 | * of all the other nodes. | |
6226 | * We don't want to pressure a particular node, so when | |
6227 | * building the zones for node N, we make sure that the | |
6228 | * zones coming right after the local ones are those from | |
6229 | * node N+1 (modulo N) | |
6230 | */ | |
6231 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
6232 | if (!node_online(node)) | |
6233 | continue; | |
9d3be21b MH |
6234 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6235 | zonerefs += nr_zones; | |
1da177e4 | 6236 | } |
54a6eb5c MG |
6237 | for (node = 0; node < local_node; node++) { |
6238 | if (!node_online(node)) | |
6239 | continue; | |
9d3be21b MH |
6240 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6241 | zonerefs += nr_zones; | |
54a6eb5c MG |
6242 | } |
6243 | ||
9d3be21b MH |
6244 | zonerefs->zone = NULL; |
6245 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
6246 | } |
6247 | ||
6248 | #endif /* CONFIG_NUMA */ | |
6249 | ||
99dcc3e5 CL |
6250 | /* |
6251 | * Boot pageset table. One per cpu which is going to be used for all | |
6252 | * zones and all nodes. The parameters will be set in such a way | |
6253 | * that an item put on a list will immediately be handed over to | |
6254 | * the buddy list. This is safe since pageset manipulation is done | |
6255 | * with interrupts disabled. | |
6256 | * | |
6257 | * The boot_pagesets must be kept even after bootup is complete for | |
6258 | * unused processors and/or zones. They do play a role for bootstrapping | |
6259 | * hotplugged processors. | |
6260 | * | |
6261 | * zoneinfo_show() and maybe other functions do | |
6262 | * not check if the processor is online before following the pageset pointer. | |
6263 | * Other parts of the kernel may not check if the zone is available. | |
6264 | */ | |
28f836b6 | 6265 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats); |
952eaf81 VB |
6266 | /* These effectively disable the pcplists in the boot pageset completely */ |
6267 | #define BOOT_PAGESET_HIGH 0 | |
6268 | #define BOOT_PAGESET_BATCH 1 | |
28f836b6 MG |
6269 | static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset); |
6270 | static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats); | |
385386cf | 6271 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 6272 | |
11cd8638 | 6273 | static void __build_all_zonelists(void *data) |
1da177e4 | 6274 | { |
6811378e | 6275 | int nid; |
afb6ebb3 | 6276 | int __maybe_unused cpu; |
9adb62a5 | 6277 | pg_data_t *self = data; |
b93e0f32 MH |
6278 | static DEFINE_SPINLOCK(lock); |
6279 | ||
6280 | spin_lock(&lock); | |
9276b1bc | 6281 | |
7f9cfb31 BL |
6282 | #ifdef CONFIG_NUMA |
6283 | memset(node_load, 0, sizeof(node_load)); | |
6284 | #endif | |
9adb62a5 | 6285 | |
c1152583 WY |
6286 | /* |
6287 | * This node is hotadded and no memory is yet present. So just | |
6288 | * building zonelists is fine - no need to touch other nodes. | |
6289 | */ | |
9adb62a5 JL |
6290 | if (self && !node_online(self->node_id)) { |
6291 | build_zonelists(self); | |
c1152583 WY |
6292 | } else { |
6293 | for_each_online_node(nid) { | |
6294 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 6295 | |
c1152583 WY |
6296 | build_zonelists(pgdat); |
6297 | } | |
99dcc3e5 | 6298 | |
7aac7898 LS |
6299 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6300 | /* | |
6301 | * We now know the "local memory node" for each node-- | |
6302 | * i.e., the node of the first zone in the generic zonelist. | |
6303 | * Set up numa_mem percpu variable for on-line cpus. During | |
6304 | * boot, only the boot cpu should be on-line; we'll init the | |
6305 | * secondary cpus' numa_mem as they come on-line. During | |
6306 | * node/memory hotplug, we'll fixup all on-line cpus. | |
6307 | */ | |
d9c9a0b9 | 6308 | for_each_online_cpu(cpu) |
7aac7898 | 6309 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 6310 | #endif |
d9c9a0b9 | 6311 | } |
b93e0f32 MH |
6312 | |
6313 | spin_unlock(&lock); | |
6811378e YG |
6314 | } |
6315 | ||
061f67bc RV |
6316 | static noinline void __init |
6317 | build_all_zonelists_init(void) | |
6318 | { | |
afb6ebb3 MH |
6319 | int cpu; |
6320 | ||
061f67bc | 6321 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
6322 | |
6323 | /* | |
6324 | * Initialize the boot_pagesets that are going to be used | |
6325 | * for bootstrapping processors. The real pagesets for | |
6326 | * each zone will be allocated later when the per cpu | |
6327 | * allocator is available. | |
6328 | * | |
6329 | * boot_pagesets are used also for bootstrapping offline | |
6330 | * cpus if the system is already booted because the pagesets | |
6331 | * are needed to initialize allocators on a specific cpu too. | |
6332 | * F.e. the percpu allocator needs the page allocator which | |
6333 | * needs the percpu allocator in order to allocate its pagesets | |
6334 | * (a chicken-egg dilemma). | |
6335 | */ | |
6336 | for_each_possible_cpu(cpu) | |
28f836b6 | 6337 | per_cpu_pages_init(&per_cpu(boot_pageset, cpu), &per_cpu(boot_zonestats, cpu)); |
afb6ebb3 | 6338 | |
061f67bc RV |
6339 | mminit_verify_zonelist(); |
6340 | cpuset_init_current_mems_allowed(); | |
6341 | } | |
6342 | ||
4eaf3f64 | 6343 | /* |
4eaf3f64 | 6344 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 6345 | * |
72675e13 | 6346 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 6347 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 6348 | */ |
72675e13 | 6349 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 6350 | { |
0a18e607 DH |
6351 | unsigned long vm_total_pages; |
6352 | ||
6811378e | 6353 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 6354 | build_all_zonelists_init(); |
6811378e | 6355 | } else { |
11cd8638 | 6356 | __build_all_zonelists(pgdat); |
6811378e YG |
6357 | /* cpuset refresh routine should be here */ |
6358 | } | |
56b9413b DH |
6359 | /* Get the number of free pages beyond high watermark in all zones. */ |
6360 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
6361 | /* |
6362 | * Disable grouping by mobility if the number of pages in the | |
6363 | * system is too low to allow the mechanism to work. It would be | |
6364 | * more accurate, but expensive to check per-zone. This check is | |
6365 | * made on memory-hotadd so a system can start with mobility | |
6366 | * disabled and enable it later | |
6367 | */ | |
d9c23400 | 6368 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
6369 | page_group_by_mobility_disabled = 1; |
6370 | else | |
6371 | page_group_by_mobility_disabled = 0; | |
6372 | ||
ce0725f7 | 6373 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 6374 | nr_online_nodes, |
756a025f JP |
6375 | page_group_by_mobility_disabled ? "off" : "on", |
6376 | vm_total_pages); | |
f0c0b2b8 | 6377 | #ifdef CONFIG_NUMA |
f88dfff5 | 6378 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 6379 | #endif |
1da177e4 LT |
6380 | } |
6381 | ||
a9a9e77f PT |
6382 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
6383 | static bool __meminit | |
6384 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
6385 | { | |
a9a9e77f PT |
6386 | static struct memblock_region *r; |
6387 | ||
6388 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
6389 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
cc6de168 | 6390 | for_each_mem_region(r) { |
a9a9e77f PT |
6391 | if (*pfn < memblock_region_memory_end_pfn(r)) |
6392 | break; | |
6393 | } | |
6394 | } | |
6395 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
6396 | memblock_is_mirror(r)) { | |
6397 | *pfn = memblock_region_memory_end_pfn(r); | |
6398 | return true; | |
6399 | } | |
6400 | } | |
a9a9e77f PT |
6401 | return false; |
6402 | } | |
6403 | ||
1da177e4 LT |
6404 | /* |
6405 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 6406 | * up by memblock_free_all() once the early boot process is |
1da177e4 | 6407 | * done. Non-atomic initialization, single-pass. |
d882c006 DH |
6408 | * |
6409 | * All aligned pageblocks are initialized to the specified migratetype | |
6410 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
6411 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
1da177e4 | 6412 | */ |
ab28cb6e | 6413 | void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone, |
dc2da7b4 | 6414 | unsigned long start_pfn, unsigned long zone_end_pfn, |
d882c006 DH |
6415 | enum meminit_context context, |
6416 | struct vmem_altmap *altmap, int migratetype) | |
1da177e4 | 6417 | { |
a9a9e77f | 6418 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 6419 | struct page *page; |
1da177e4 | 6420 | |
22b31eec HD |
6421 | if (highest_memmap_pfn < end_pfn - 1) |
6422 | highest_memmap_pfn = end_pfn - 1; | |
6423 | ||
966cf44f | 6424 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
6425 | /* |
6426 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
6427 | * memory. We limit the total number of pages to initialize to just |
6428 | * those that might contain the memory mapping. We will defer the | |
6429 | * ZONE_DEVICE page initialization until after we have released | |
6430 | * the hotplug lock. | |
4b94ffdc | 6431 | */ |
966cf44f AD |
6432 | if (zone == ZONE_DEVICE) { |
6433 | if (!altmap) | |
6434 | return; | |
6435 | ||
6436 | if (start_pfn == altmap->base_pfn) | |
6437 | start_pfn += altmap->reserve; | |
6438 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
6439 | } | |
6440 | #endif | |
4b94ffdc | 6441 | |
948c436e | 6442 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 6443 | /* |
b72d0ffb AM |
6444 | * There can be holes in boot-time mem_map[]s handed to this |
6445 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 6446 | */ |
c1d0da83 | 6447 | if (context == MEMINIT_EARLY) { |
a9a9e77f PT |
6448 | if (overlap_memmap_init(zone, &pfn)) |
6449 | continue; | |
dc2da7b4 | 6450 | if (defer_init(nid, pfn, zone_end_pfn)) |
a9a9e77f | 6451 | break; |
a2f3aa02 | 6452 | } |
ac5d2539 | 6453 | |
d0dc12e8 PT |
6454 | page = pfn_to_page(pfn); |
6455 | __init_single_page(page, pfn, zone, nid); | |
c1d0da83 | 6456 | if (context == MEMINIT_HOTPLUG) |
d483da5b | 6457 | __SetPageReserved(page); |
d0dc12e8 | 6458 | |
ac5d2539 | 6459 | /* |
d882c006 DH |
6460 | * Usually, we want to mark the pageblock MIGRATE_MOVABLE, |
6461 | * such that unmovable allocations won't be scattered all | |
6462 | * over the place during system boot. | |
ac5d2539 | 6463 | */ |
4eb29bd9 | 6464 | if (IS_ALIGNED(pfn, pageblock_nr_pages)) { |
d882c006 | 6465 | set_pageblock_migratetype(page, migratetype); |
9b6e63cb | 6466 | cond_resched(); |
ac5d2539 | 6467 | } |
948c436e | 6468 | pfn++; |
1da177e4 LT |
6469 | } |
6470 | } | |
6471 | ||
966cf44f AD |
6472 | #ifdef CONFIG_ZONE_DEVICE |
6473 | void __ref memmap_init_zone_device(struct zone *zone, | |
6474 | unsigned long start_pfn, | |
1f8d75c1 | 6475 | unsigned long nr_pages, |
966cf44f AD |
6476 | struct dev_pagemap *pgmap) |
6477 | { | |
1f8d75c1 | 6478 | unsigned long pfn, end_pfn = start_pfn + nr_pages; |
966cf44f | 6479 | struct pglist_data *pgdat = zone->zone_pgdat; |
514caf23 | 6480 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
966cf44f AD |
6481 | unsigned long zone_idx = zone_idx(zone); |
6482 | unsigned long start = jiffies; | |
6483 | int nid = pgdat->node_id; | |
6484 | ||
46d945ae | 6485 | if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE)) |
966cf44f AD |
6486 | return; |
6487 | ||
6488 | /* | |
122e093c | 6489 | * The call to memmap_init should have already taken care |
966cf44f AD |
6490 | * of the pages reserved for the memmap, so we can just jump to |
6491 | * the end of that region and start processing the device pages. | |
6492 | */ | |
514caf23 | 6493 | if (altmap) { |
966cf44f | 6494 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
1f8d75c1 | 6495 | nr_pages = end_pfn - start_pfn; |
966cf44f AD |
6496 | } |
6497 | ||
6498 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
6499 | struct page *page = pfn_to_page(pfn); | |
6500 | ||
6501 | __init_single_page(page, pfn, zone_idx, nid); | |
6502 | ||
6503 | /* | |
6504 | * Mark page reserved as it will need to wait for onlining | |
6505 | * phase for it to be fully associated with a zone. | |
6506 | * | |
6507 | * We can use the non-atomic __set_bit operation for setting | |
6508 | * the flag as we are still initializing the pages. | |
6509 | */ | |
6510 | __SetPageReserved(page); | |
6511 | ||
6512 | /* | |
8a164fef CH |
6513 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer |
6514 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
6515 | * ever freed or placed on a driver-private list. | |
966cf44f AD |
6516 | */ |
6517 | page->pgmap = pgmap; | |
8a164fef | 6518 | page->zone_device_data = NULL; |
966cf44f AD |
6519 | |
6520 | /* | |
6521 | * Mark the block movable so that blocks are reserved for | |
6522 | * movable at startup. This will force kernel allocations | |
6523 | * to reserve their blocks rather than leaking throughout | |
6524 | * the address space during boot when many long-lived | |
6525 | * kernel allocations are made. | |
6526 | * | |
c1d0da83 | 6527 | * Please note that MEMINIT_HOTPLUG path doesn't clear memmap |
ba72b4c8 | 6528 | * because this is done early in section_activate() |
966cf44f | 6529 | */ |
4eb29bd9 | 6530 | if (IS_ALIGNED(pfn, pageblock_nr_pages)) { |
966cf44f AD |
6531 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
6532 | cond_resched(); | |
6533 | } | |
6534 | } | |
6535 | ||
fdc029b1 | 6536 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
1f8d75c1 | 6537 | nr_pages, jiffies_to_msecs(jiffies - start)); |
966cf44f AD |
6538 | } |
6539 | ||
6540 | #endif | |
1e548deb | 6541 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6542 | { |
7aeb09f9 | 6543 | unsigned int order, t; |
b2a0ac88 MG |
6544 | for_each_migratetype_order(order, t) { |
6545 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6546 | zone->free_area[order].nr_free = 0; |
6547 | } | |
6548 | } | |
6549 | ||
0740a50b MR |
6550 | #if !defined(CONFIG_FLAT_NODE_MEM_MAP) |
6551 | /* | |
6552 | * Only struct pages that correspond to ranges defined by memblock.memory | |
6553 | * are zeroed and initialized by going through __init_single_page() during | |
122e093c | 6554 | * memmap_init_zone_range(). |
0740a50b MR |
6555 | * |
6556 | * But, there could be struct pages that correspond to holes in | |
6557 | * memblock.memory. This can happen because of the following reasons: | |
6558 | * - physical memory bank size is not necessarily the exact multiple of the | |
6559 | * arbitrary section size | |
6560 | * - early reserved memory may not be listed in memblock.memory | |
6561 | * - memory layouts defined with memmap= kernel parameter may not align | |
6562 | * nicely with memmap sections | |
6563 | * | |
6564 | * Explicitly initialize those struct pages so that: | |
6565 | * - PG_Reserved is set | |
6566 | * - zone and node links point to zone and node that span the page if the | |
6567 | * hole is in the middle of a zone | |
6568 | * - zone and node links point to adjacent zone/node if the hole falls on | |
6569 | * the zone boundary; the pages in such holes will be prepended to the | |
6570 | * zone/node above the hole except for the trailing pages in the last | |
6571 | * section that will be appended to the zone/node below. | |
6572 | */ | |
122e093c MR |
6573 | static void __init init_unavailable_range(unsigned long spfn, |
6574 | unsigned long epfn, | |
6575 | int zone, int node) | |
0740a50b MR |
6576 | { |
6577 | unsigned long pfn; | |
6578 | u64 pgcnt = 0; | |
6579 | ||
6580 | for (pfn = spfn; pfn < epfn; pfn++) { | |
6581 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { | |
6582 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6583 | + pageblock_nr_pages - 1; | |
6584 | continue; | |
6585 | } | |
6586 | __init_single_page(pfn_to_page(pfn), pfn, zone, node); | |
6587 | __SetPageReserved(pfn_to_page(pfn)); | |
6588 | pgcnt++; | |
6589 | } | |
6590 | ||
122e093c MR |
6591 | if (pgcnt) |
6592 | pr_info("On node %d, zone %s: %lld pages in unavailable ranges", | |
6593 | node, zone_names[zone], pgcnt); | |
0740a50b MR |
6594 | } |
6595 | #else | |
122e093c MR |
6596 | static inline void init_unavailable_range(unsigned long spfn, |
6597 | unsigned long epfn, | |
6598 | int zone, int node) | |
0740a50b | 6599 | { |
0740a50b MR |
6600 | } |
6601 | #endif | |
6602 | ||
122e093c MR |
6603 | static void __init memmap_init_zone_range(struct zone *zone, |
6604 | unsigned long start_pfn, | |
6605 | unsigned long end_pfn, | |
6606 | unsigned long *hole_pfn) | |
dfb3ccd0 | 6607 | { |
3256ff83 BH |
6608 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
6609 | unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages; | |
122e093c MR |
6610 | int nid = zone_to_nid(zone), zone_id = zone_idx(zone); |
6611 | ||
6612 | start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn); | |
6613 | end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn); | |
6614 | ||
6615 | if (start_pfn >= end_pfn) | |
6616 | return; | |
6617 | ||
6618 | memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn, | |
6619 | zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE); | |
6620 | ||
6621 | if (*hole_pfn < start_pfn) | |
6622 | init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid); | |
6623 | ||
6624 | *hole_pfn = end_pfn; | |
6625 | } | |
6626 | ||
6627 | static void __init memmap_init(void) | |
6628 | { | |
73a6e474 | 6629 | unsigned long start_pfn, end_pfn; |
122e093c MR |
6630 | unsigned long hole_pfn = 0; |
6631 | int i, j, zone_id, nid; | |
73a6e474 | 6632 | |
122e093c MR |
6633 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
6634 | struct pglist_data *node = NODE_DATA(nid); | |
6635 | ||
6636 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
6637 | struct zone *zone = node->node_zones + j; | |
73a6e474 | 6638 | |
122e093c MR |
6639 | if (!populated_zone(zone)) |
6640 | continue; | |
0740a50b | 6641 | |
122e093c MR |
6642 | memmap_init_zone_range(zone, start_pfn, end_pfn, |
6643 | &hole_pfn); | |
6644 | zone_id = j; | |
6645 | } | |
73a6e474 | 6646 | } |
0740a50b MR |
6647 | |
6648 | #ifdef CONFIG_SPARSEMEM | |
6649 | /* | |
122e093c MR |
6650 | * Initialize the memory map for hole in the range [memory_end, |
6651 | * section_end]. | |
6652 | * Append the pages in this hole to the highest zone in the last | |
6653 | * node. | |
6654 | * The call to init_unavailable_range() is outside the ifdef to | |
6655 | * silence the compiler warining about zone_id set but not used; | |
6656 | * for FLATMEM it is a nop anyway | |
0740a50b | 6657 | */ |
122e093c | 6658 | end_pfn = round_up(end_pfn, PAGES_PER_SECTION); |
0740a50b | 6659 | if (hole_pfn < end_pfn) |
0740a50b | 6660 | #endif |
122e093c | 6661 | init_unavailable_range(hole_pfn, end_pfn, zone_id, nid); |
dfb3ccd0 | 6662 | } |
1da177e4 | 6663 | |
7cd2b0a3 | 6664 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 6665 | { |
3a6be87f | 6666 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
6667 | int batch; |
6668 | ||
6669 | /* | |
b92ca18e MG |
6670 | * The number of pages to batch allocate is either ~0.1% |
6671 | * of the zone or 1MB, whichever is smaller. The batch | |
6672 | * size is striking a balance between allocation latency | |
6673 | * and zone lock contention. | |
e7c8d5c9 | 6674 | */ |
b92ca18e | 6675 | batch = min(zone_managed_pages(zone) >> 10, (1024 * 1024) / PAGE_SIZE); |
e7c8d5c9 CL |
6676 | batch /= 4; /* We effectively *= 4 below */ |
6677 | if (batch < 1) | |
6678 | batch = 1; | |
6679 | ||
6680 | /* | |
0ceaacc9 NP |
6681 | * Clamp the batch to a 2^n - 1 value. Having a power |
6682 | * of 2 value was found to be more likely to have | |
6683 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 6684 | * |
0ceaacc9 NP |
6685 | * For example if 2 tasks are alternately allocating |
6686 | * batches of pages, one task can end up with a lot | |
6687 | * of pages of one half of the possible page colors | |
6688 | * and the other with pages of the other colors. | |
e7c8d5c9 | 6689 | */ |
9155203a | 6690 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 6691 | |
e7c8d5c9 | 6692 | return batch; |
3a6be87f DH |
6693 | |
6694 | #else | |
6695 | /* The deferral and batching of frees should be suppressed under NOMMU | |
6696 | * conditions. | |
6697 | * | |
6698 | * The problem is that NOMMU needs to be able to allocate large chunks | |
6699 | * of contiguous memory as there's no hardware page translation to | |
6700 | * assemble apparent contiguous memory from discontiguous pages. | |
6701 | * | |
6702 | * Queueing large contiguous runs of pages for batching, however, | |
6703 | * causes the pages to actually be freed in smaller chunks. As there | |
6704 | * can be a significant delay between the individual batches being | |
6705 | * recycled, this leads to the once large chunks of space being | |
6706 | * fragmented and becoming unavailable for high-order allocations. | |
6707 | */ | |
6708 | return 0; | |
6709 | #endif | |
e7c8d5c9 CL |
6710 | } |
6711 | ||
04f8cfea | 6712 | static int zone_highsize(struct zone *zone, int batch, int cpu_online) |
b92ca18e MG |
6713 | { |
6714 | #ifdef CONFIG_MMU | |
6715 | int high; | |
6716 | int nr_local_cpus; | |
74f44822 MG |
6717 | unsigned long total_pages; |
6718 | ||
6719 | if (!percpu_pagelist_high_fraction) { | |
6720 | /* | |
6721 | * By default, the high value of the pcp is based on the zone | |
6722 | * low watermark so that if they are full then background | |
6723 | * reclaim will not be started prematurely. | |
6724 | */ | |
6725 | total_pages = low_wmark_pages(zone); | |
6726 | } else { | |
6727 | /* | |
6728 | * If percpu_pagelist_high_fraction is configured, the high | |
6729 | * value is based on a fraction of the managed pages in the | |
6730 | * zone. | |
6731 | */ | |
6732 | total_pages = zone_managed_pages(zone) / percpu_pagelist_high_fraction; | |
6733 | } | |
b92ca18e MG |
6734 | |
6735 | /* | |
74f44822 MG |
6736 | * Split the high value across all online CPUs local to the zone. Note |
6737 | * that early in boot that CPUs may not be online yet and that during | |
6738 | * CPU hotplug that the cpumask is not yet updated when a CPU is being | |
6739 | * onlined. | |
b92ca18e | 6740 | */ |
04f8cfea | 6741 | nr_local_cpus = max(1U, cpumask_weight(cpumask_of_node(zone_to_nid(zone)))) + cpu_online; |
74f44822 | 6742 | high = total_pages / nr_local_cpus; |
b92ca18e MG |
6743 | |
6744 | /* | |
6745 | * Ensure high is at least batch*4. The multiple is based on the | |
6746 | * historical relationship between high and batch. | |
6747 | */ | |
6748 | high = max(high, batch << 2); | |
6749 | ||
6750 | return high; | |
6751 | #else | |
6752 | return 0; | |
6753 | #endif | |
6754 | } | |
6755 | ||
8d7a8fa9 | 6756 | /* |
5c3ad2eb VB |
6757 | * pcp->high and pcp->batch values are related and generally batch is lower |
6758 | * than high. They are also related to pcp->count such that count is lower | |
6759 | * than high, and as soon as it reaches high, the pcplist is flushed. | |
8d7a8fa9 | 6760 | * |
5c3ad2eb VB |
6761 | * However, guaranteeing these relations at all times would require e.g. write |
6762 | * barriers here but also careful usage of read barriers at the read side, and | |
6763 | * thus be prone to error and bad for performance. Thus the update only prevents | |
6764 | * store tearing. Any new users of pcp->batch and pcp->high should ensure they | |
6765 | * can cope with those fields changing asynchronously, and fully trust only the | |
6766 | * pcp->count field on the local CPU with interrupts disabled. | |
8d7a8fa9 CS |
6767 | * |
6768 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
6769 | * outside of boot time (or some other assurance that no concurrent updaters | |
6770 | * exist). | |
6771 | */ | |
6772 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
6773 | unsigned long batch) | |
6774 | { | |
5c3ad2eb VB |
6775 | WRITE_ONCE(pcp->batch, batch); |
6776 | WRITE_ONCE(pcp->high, high); | |
8d7a8fa9 CS |
6777 | } |
6778 | ||
28f836b6 | 6779 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats) |
2caaad41 | 6780 | { |
5f8dcc21 | 6781 | int migratetype; |
2caaad41 | 6782 | |
28f836b6 MG |
6783 | memset(pcp, 0, sizeof(*pcp)); |
6784 | memset(pzstats, 0, sizeof(*pzstats)); | |
1c6fe946 | 6785 | |
5f8dcc21 MG |
6786 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
6787 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 | 6788 | |
69a8396a VB |
6789 | /* |
6790 | * Set batch and high values safe for a boot pageset. A true percpu | |
6791 | * pageset's initialization will update them subsequently. Here we don't | |
6792 | * need to be as careful as pageset_update() as nobody can access the | |
6793 | * pageset yet. | |
6794 | */ | |
952eaf81 VB |
6795 | pcp->high = BOOT_PAGESET_HIGH; |
6796 | pcp->batch = BOOT_PAGESET_BATCH; | |
3b12e7e9 | 6797 | pcp->free_factor = 0; |
88c90dbc CS |
6798 | } |
6799 | ||
3b1f3658 | 6800 | static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high, |
ec6e8c7e VB |
6801 | unsigned long batch) |
6802 | { | |
28f836b6 | 6803 | struct per_cpu_pages *pcp; |
ec6e8c7e VB |
6804 | int cpu; |
6805 | ||
6806 | for_each_possible_cpu(cpu) { | |
28f836b6 MG |
6807 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
6808 | pageset_update(pcp, high, batch); | |
ec6e8c7e VB |
6809 | } |
6810 | } | |
6811 | ||
8ad4b1fb | 6812 | /* |
0a8b4f1d | 6813 | * Calculate and set new high and batch values for all per-cpu pagesets of a |
bbbecb35 | 6814 | * zone based on the zone's size. |
8ad4b1fb | 6815 | */ |
04f8cfea | 6816 | static void zone_set_pageset_high_and_batch(struct zone *zone, int cpu_online) |
56cef2b8 | 6817 | { |
b92ca18e | 6818 | int new_high, new_batch; |
7115ac6e | 6819 | |
b92ca18e | 6820 | new_batch = max(1, zone_batchsize(zone)); |
04f8cfea | 6821 | new_high = zone_highsize(zone, new_batch, cpu_online); |
169f6c19 | 6822 | |
952eaf81 VB |
6823 | if (zone->pageset_high == new_high && |
6824 | zone->pageset_batch == new_batch) | |
6825 | return; | |
6826 | ||
6827 | zone->pageset_high = new_high; | |
6828 | zone->pageset_batch = new_batch; | |
6829 | ||
ec6e8c7e | 6830 | __zone_set_pageset_high_and_batch(zone, new_high, new_batch); |
169f6c19 CS |
6831 | } |
6832 | ||
72675e13 | 6833 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
6834 | { |
6835 | int cpu; | |
0a8b4f1d | 6836 | |
28f836b6 MG |
6837 | /* Size may be 0 on !SMP && !NUMA */ |
6838 | if (sizeof(struct per_cpu_zonestat) > 0) | |
6839 | zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat); | |
6840 | ||
6841 | zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages); | |
0a8b4f1d | 6842 | for_each_possible_cpu(cpu) { |
28f836b6 MG |
6843 | struct per_cpu_pages *pcp; |
6844 | struct per_cpu_zonestat *pzstats; | |
6845 | ||
6846 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); | |
6847 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); | |
6848 | per_cpu_pages_init(pcp, pzstats); | |
0a8b4f1d VB |
6849 | } |
6850 | ||
04f8cfea | 6851 | zone_set_pageset_high_and_batch(zone, 0); |
319774e2 WF |
6852 | } |
6853 | ||
2caaad41 | 6854 | /* |
99dcc3e5 CL |
6855 | * Allocate per cpu pagesets and initialize them. |
6856 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 6857 | */ |
99dcc3e5 | 6858 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 6859 | { |
b4911ea2 | 6860 | struct pglist_data *pgdat; |
99dcc3e5 | 6861 | struct zone *zone; |
b418a0f9 | 6862 | int __maybe_unused cpu; |
e7c8d5c9 | 6863 | |
319774e2 WF |
6864 | for_each_populated_zone(zone) |
6865 | setup_zone_pageset(zone); | |
b4911ea2 | 6866 | |
b418a0f9 SD |
6867 | #ifdef CONFIG_NUMA |
6868 | /* | |
6869 | * Unpopulated zones continue using the boot pagesets. | |
6870 | * The numa stats for these pagesets need to be reset. | |
6871 | * Otherwise, they will end up skewing the stats of | |
6872 | * the nodes these zones are associated with. | |
6873 | */ | |
6874 | for_each_possible_cpu(cpu) { | |
28f836b6 | 6875 | struct per_cpu_zonestat *pzstats = &per_cpu(boot_zonestats, cpu); |
f19298b9 MG |
6876 | memset(pzstats->vm_numa_event, 0, |
6877 | sizeof(pzstats->vm_numa_event)); | |
b418a0f9 SD |
6878 | } |
6879 | #endif | |
6880 | ||
b4911ea2 MG |
6881 | for_each_online_pgdat(pgdat) |
6882 | pgdat->per_cpu_nodestats = | |
6883 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
6884 | } |
6885 | ||
c09b4240 | 6886 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 6887 | { |
99dcc3e5 CL |
6888 | /* |
6889 | * per cpu subsystem is not up at this point. The following code | |
6890 | * relies on the ability of the linker to provide the | |
6891 | * offset of a (static) per cpu variable into the per cpu area. | |
6892 | */ | |
28f836b6 MG |
6893 | zone->per_cpu_pageset = &boot_pageset; |
6894 | zone->per_cpu_zonestats = &boot_zonestats; | |
952eaf81 VB |
6895 | zone->pageset_high = BOOT_PAGESET_HIGH; |
6896 | zone->pageset_batch = BOOT_PAGESET_BATCH; | |
ed8ece2e | 6897 | |
b38a8725 | 6898 | if (populated_zone(zone)) |
9660ecaa HK |
6899 | pr_debug(" %s zone: %lu pages, LIFO batch:%u\n", zone->name, |
6900 | zone->present_pages, zone_batchsize(zone)); | |
ed8ece2e DH |
6901 | } |
6902 | ||
dc0bbf3b | 6903 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 6904 | unsigned long zone_start_pfn, |
b171e409 | 6905 | unsigned long size) |
ed8ece2e DH |
6906 | { |
6907 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 6908 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 6909 | |
8f416836 WY |
6910 | if (zone_idx > pgdat->nr_zones) |
6911 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 6912 | |
ed8ece2e DH |
6913 | zone->zone_start_pfn = zone_start_pfn; |
6914 | ||
708614e6 MG |
6915 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
6916 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
6917 | pgdat->node_id, | |
6918 | (unsigned long)zone_idx(zone), | |
6919 | zone_start_pfn, (zone_start_pfn + size)); | |
6920 | ||
1e548deb | 6921 | zone_init_free_lists(zone); |
9dcb8b68 | 6922 | zone->initialized = 1; |
ed8ece2e DH |
6923 | } |
6924 | ||
c713216d MG |
6925 | /** |
6926 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
6927 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
6928 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
6929 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
6930 | * |
6931 | * It returns the start and end page frame of a node based on information | |
7d018176 | 6932 | * provided by memblock_set_node(). If called for a node |
c713216d | 6933 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 6934 | * PFNs will be 0. |
c713216d | 6935 | */ |
bbe5d993 | 6936 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
6937 | unsigned long *start_pfn, unsigned long *end_pfn) |
6938 | { | |
c13291a5 | 6939 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 6940 | int i; |
c13291a5 | 6941 | |
c713216d MG |
6942 | *start_pfn = -1UL; |
6943 | *end_pfn = 0; | |
6944 | ||
c13291a5 TH |
6945 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
6946 | *start_pfn = min(*start_pfn, this_start_pfn); | |
6947 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
6948 | } |
6949 | ||
633c0666 | 6950 | if (*start_pfn == -1UL) |
c713216d | 6951 | *start_pfn = 0; |
c713216d MG |
6952 | } |
6953 | ||
2a1e274a MG |
6954 | /* |
6955 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
6956 | * assumption is made that zones within a node are ordered in monotonic | |
6957 | * increasing memory addresses so that the "highest" populated zone is used | |
6958 | */ | |
b69a7288 | 6959 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
6960 | { |
6961 | int zone_index; | |
6962 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
6963 | if (zone_index == ZONE_MOVABLE) | |
6964 | continue; | |
6965 | ||
6966 | if (arch_zone_highest_possible_pfn[zone_index] > | |
6967 | arch_zone_lowest_possible_pfn[zone_index]) | |
6968 | break; | |
6969 | } | |
6970 | ||
6971 | VM_BUG_ON(zone_index == -1); | |
6972 | movable_zone = zone_index; | |
6973 | } | |
6974 | ||
6975 | /* | |
6976 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 6977 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
6978 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
6979 | * in each node depending on the size of each node and how evenly kernelcore | |
6980 | * is distributed. This helper function adjusts the zone ranges | |
6981 | * provided by the architecture for a given node by using the end of the | |
6982 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
6983 | * zones within a node are in order of monotonic increases memory addresses | |
6984 | */ | |
bbe5d993 | 6985 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
6986 | unsigned long zone_type, |
6987 | unsigned long node_start_pfn, | |
6988 | unsigned long node_end_pfn, | |
6989 | unsigned long *zone_start_pfn, | |
6990 | unsigned long *zone_end_pfn) | |
6991 | { | |
6992 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
6993 | if (zone_movable_pfn[nid]) { | |
6994 | /* Size ZONE_MOVABLE */ | |
6995 | if (zone_type == ZONE_MOVABLE) { | |
6996 | *zone_start_pfn = zone_movable_pfn[nid]; | |
6997 | *zone_end_pfn = min(node_end_pfn, | |
6998 | arch_zone_highest_possible_pfn[movable_zone]); | |
6999 | ||
e506b996 XQ |
7000 | /* Adjust for ZONE_MOVABLE starting within this range */ |
7001 | } else if (!mirrored_kernelcore && | |
7002 | *zone_start_pfn < zone_movable_pfn[nid] && | |
7003 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
7004 | *zone_end_pfn = zone_movable_pfn[nid]; | |
7005 | ||
2a1e274a MG |
7006 | /* Check if this whole range is within ZONE_MOVABLE */ |
7007 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
7008 | *zone_start_pfn = *zone_end_pfn; | |
7009 | } | |
7010 | } | |
7011 | ||
c713216d MG |
7012 | /* |
7013 | * Return the number of pages a zone spans in a node, including holes | |
7014 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
7015 | */ | |
bbe5d993 | 7016 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 7017 | unsigned long zone_type, |
7960aedd ZY |
7018 | unsigned long node_start_pfn, |
7019 | unsigned long node_end_pfn, | |
d91749c1 | 7020 | unsigned long *zone_start_pfn, |
854e8848 | 7021 | unsigned long *zone_end_pfn) |
c713216d | 7022 | { |
299c83dc LF |
7023 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
7024 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 7025 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
7026 | if (!node_start_pfn && !node_end_pfn) |
7027 | return 0; | |
7028 | ||
7960aedd | 7029 | /* Get the start and end of the zone */ |
299c83dc LF |
7030 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
7031 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
7032 | adjust_zone_range_for_zone_movable(nid, zone_type, |
7033 | node_start_pfn, node_end_pfn, | |
d91749c1 | 7034 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
7035 | |
7036 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 7037 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
7038 | return 0; |
7039 | ||
7040 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
7041 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
7042 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
7043 | |
7044 | /* Return the spanned pages */ | |
d91749c1 | 7045 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
7046 | } |
7047 | ||
7048 | /* | |
7049 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 7050 | * then all holes in the requested range will be accounted for. |
c713216d | 7051 | */ |
bbe5d993 | 7052 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
7053 | unsigned long range_start_pfn, |
7054 | unsigned long range_end_pfn) | |
7055 | { | |
96e907d1 TH |
7056 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
7057 | unsigned long start_pfn, end_pfn; | |
7058 | int i; | |
c713216d | 7059 | |
96e907d1 TH |
7060 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
7061 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
7062 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
7063 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 7064 | } |
96e907d1 | 7065 | return nr_absent; |
c713216d MG |
7066 | } |
7067 | ||
7068 | /** | |
7069 | * absent_pages_in_range - Return number of page frames in holes within a range | |
7070 | * @start_pfn: The start PFN to start searching for holes | |
7071 | * @end_pfn: The end PFN to stop searching for holes | |
7072 | * | |
a862f68a | 7073 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
7074 | */ |
7075 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
7076 | unsigned long end_pfn) | |
7077 | { | |
7078 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
7079 | } | |
7080 | ||
7081 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 7082 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 7083 | unsigned long zone_type, |
7960aedd | 7084 | unsigned long node_start_pfn, |
854e8848 | 7085 | unsigned long node_end_pfn) |
c713216d | 7086 | { |
96e907d1 TH |
7087 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
7088 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 7089 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 7090 | unsigned long nr_absent; |
9c7cd687 | 7091 | |
b5685e92 | 7092 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
7093 | if (!node_start_pfn && !node_end_pfn) |
7094 | return 0; | |
7095 | ||
96e907d1 TH |
7096 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
7097 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 7098 | |
2a1e274a MG |
7099 | adjust_zone_range_for_zone_movable(nid, zone_type, |
7100 | node_start_pfn, node_end_pfn, | |
7101 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
7102 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
7103 | ||
7104 | /* | |
7105 | * ZONE_MOVABLE handling. | |
7106 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
7107 | * and vice versa. | |
7108 | */ | |
e506b996 XQ |
7109 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
7110 | unsigned long start_pfn, end_pfn; | |
7111 | struct memblock_region *r; | |
7112 | ||
cc6de168 | 7113 | for_each_mem_region(r) { |
e506b996 XQ |
7114 | start_pfn = clamp(memblock_region_memory_base_pfn(r), |
7115 | zone_start_pfn, zone_end_pfn); | |
7116 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
7117 | zone_start_pfn, zone_end_pfn); | |
7118 | ||
7119 | if (zone_type == ZONE_MOVABLE && | |
7120 | memblock_is_mirror(r)) | |
7121 | nr_absent += end_pfn - start_pfn; | |
7122 | ||
7123 | if (zone_type == ZONE_NORMAL && | |
7124 | !memblock_is_mirror(r)) | |
7125 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
7126 | } |
7127 | } | |
7128 | ||
7129 | return nr_absent; | |
c713216d | 7130 | } |
0e0b864e | 7131 | |
bbe5d993 | 7132 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd | 7133 | unsigned long node_start_pfn, |
854e8848 | 7134 | unsigned long node_end_pfn) |
c713216d | 7135 | { |
febd5949 | 7136 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
7137 | enum zone_type i; |
7138 | ||
febd5949 GZ |
7139 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7140 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 7141 | unsigned long zone_start_pfn, zone_end_pfn; |
3f08a302 | 7142 | unsigned long spanned, absent; |
febd5949 | 7143 | unsigned long size, real_size; |
c713216d | 7144 | |
854e8848 MR |
7145 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i, |
7146 | node_start_pfn, | |
7147 | node_end_pfn, | |
7148 | &zone_start_pfn, | |
7149 | &zone_end_pfn); | |
7150 | absent = zone_absent_pages_in_node(pgdat->node_id, i, | |
7151 | node_start_pfn, | |
7152 | node_end_pfn); | |
3f08a302 MR |
7153 | |
7154 | size = spanned; | |
7155 | real_size = size - absent; | |
7156 | ||
d91749c1 TI |
7157 | if (size) |
7158 | zone->zone_start_pfn = zone_start_pfn; | |
7159 | else | |
7160 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
7161 | zone->spanned_pages = size; |
7162 | zone->present_pages = real_size; | |
7163 | ||
7164 | totalpages += size; | |
7165 | realtotalpages += real_size; | |
7166 | } | |
7167 | ||
7168 | pgdat->node_spanned_pages = totalpages; | |
c713216d | 7169 | pgdat->node_present_pages = realtotalpages; |
9660ecaa | 7170 | pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); |
c713216d MG |
7171 | } |
7172 | ||
835c134e MG |
7173 | #ifndef CONFIG_SPARSEMEM |
7174 | /* | |
7175 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
7176 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
7177 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
7178 | * round what is now in bits to nearest long in bits, then return it in |
7179 | * bytes. | |
7180 | */ | |
7c45512d | 7181 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
7182 | { |
7183 | unsigned long usemapsize; | |
7184 | ||
7c45512d | 7185 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
7186 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
7187 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
7188 | usemapsize *= NR_PAGEBLOCK_BITS; |
7189 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
7190 | ||
7191 | return usemapsize / 8; | |
7192 | } | |
7193 | ||
7010a6ec | 7194 | static void __ref setup_usemap(struct zone *zone) |
835c134e | 7195 | { |
7010a6ec BH |
7196 | unsigned long usemapsize = usemap_size(zone->zone_start_pfn, |
7197 | zone->spanned_pages); | |
835c134e | 7198 | zone->pageblock_flags = NULL; |
23a7052a | 7199 | if (usemapsize) { |
6782832e | 7200 | zone->pageblock_flags = |
26fb3dae | 7201 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
7010a6ec | 7202 | zone_to_nid(zone)); |
23a7052a MR |
7203 | if (!zone->pageblock_flags) |
7204 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
7010a6ec | 7205 | usemapsize, zone->name, zone_to_nid(zone)); |
23a7052a | 7206 | } |
835c134e MG |
7207 | } |
7208 | #else | |
7010a6ec | 7209 | static inline void setup_usemap(struct zone *zone) {} |
835c134e MG |
7210 | #endif /* CONFIG_SPARSEMEM */ |
7211 | ||
d9c23400 | 7212 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 7213 | |
d9c23400 | 7214 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 7215 | void __init set_pageblock_order(void) |
d9c23400 | 7216 | { |
955c1cd7 AM |
7217 | unsigned int order; |
7218 | ||
d9c23400 MG |
7219 | /* Check that pageblock_nr_pages has not already been setup */ |
7220 | if (pageblock_order) | |
7221 | return; | |
7222 | ||
955c1cd7 AM |
7223 | if (HPAGE_SHIFT > PAGE_SHIFT) |
7224 | order = HUGETLB_PAGE_ORDER; | |
7225 | else | |
7226 | order = MAX_ORDER - 1; | |
7227 | ||
d9c23400 MG |
7228 | /* |
7229 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
7230 | * This value may be variable depending on boot parameters on IA64 and |
7231 | * powerpc. | |
d9c23400 MG |
7232 | */ |
7233 | pageblock_order = order; | |
7234 | } | |
7235 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7236 | ||
ba72cb8c MG |
7237 | /* |
7238 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
7239 | * is unused as pageblock_order is set at compile-time. See |
7240 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
7241 | * the kernel config | |
ba72cb8c | 7242 | */ |
03e85f9d | 7243 | void __init set_pageblock_order(void) |
ba72cb8c | 7244 | { |
ba72cb8c | 7245 | } |
d9c23400 MG |
7246 | |
7247 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7248 | ||
03e85f9d | 7249 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 7250 | unsigned long present_pages) |
01cefaef JL |
7251 | { |
7252 | unsigned long pages = spanned_pages; | |
7253 | ||
7254 | /* | |
7255 | * Provide a more accurate estimation if there are holes within | |
7256 | * the zone and SPARSEMEM is in use. If there are holes within the | |
7257 | * zone, each populated memory region may cost us one or two extra | |
7258 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 7259 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
7260 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
7261 | */ | |
7262 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
7263 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
7264 | pages = present_pages; | |
7265 | ||
7266 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
7267 | } | |
7268 | ||
ace1db39 OS |
7269 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
7270 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
7271 | { | |
364c1eeb YS |
7272 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
7273 | ||
7274 | spin_lock_init(&ds_queue->split_queue_lock); | |
7275 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
7276 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
7277 | } |
7278 | #else | |
7279 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
7280 | #endif | |
7281 | ||
7282 | #ifdef CONFIG_COMPACTION | |
7283 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
7284 | { | |
7285 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
7286 | } | |
7287 | #else | |
7288 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
7289 | #endif | |
7290 | ||
03e85f9d | 7291 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 7292 | { |
208d54e5 | 7293 | pgdat_resize_init(pgdat); |
ace1db39 | 7294 | |
ace1db39 OS |
7295 | pgdat_init_split_queue(pgdat); |
7296 | pgdat_init_kcompactd(pgdat); | |
7297 | ||
1da177e4 | 7298 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 7299 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 7300 | |
eefa864b | 7301 | pgdat_page_ext_init(pgdat); |
867e5e1d | 7302 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
7303 | } |
7304 | ||
7305 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
7306 | unsigned long remaining_pages) | |
7307 | { | |
9705bea5 | 7308 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
7309 | zone_set_nid(zone, nid); |
7310 | zone->name = zone_names[idx]; | |
7311 | zone->zone_pgdat = NODE_DATA(nid); | |
7312 | spin_lock_init(&zone->lock); | |
7313 | zone_seqlock_init(zone); | |
7314 | zone_pcp_init(zone); | |
7315 | } | |
7316 | ||
7317 | /* | |
7318 | * Set up the zone data structures | |
7319 | * - init pgdat internals | |
7320 | * - init all zones belonging to this node | |
7321 | * | |
7322 | * NOTE: this function is only called during memory hotplug | |
7323 | */ | |
7324 | #ifdef CONFIG_MEMORY_HOTPLUG | |
7325 | void __ref free_area_init_core_hotplug(int nid) | |
7326 | { | |
7327 | enum zone_type z; | |
7328 | pg_data_t *pgdat = NODE_DATA(nid); | |
7329 | ||
7330 | pgdat_init_internals(pgdat); | |
7331 | for (z = 0; z < MAX_NR_ZONES; z++) | |
7332 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
7333 | } | |
7334 | #endif | |
7335 | ||
7336 | /* | |
7337 | * Set up the zone data structures: | |
7338 | * - mark all pages reserved | |
7339 | * - mark all memory queues empty | |
7340 | * - clear the memory bitmaps | |
7341 | * | |
7342 | * NOTE: pgdat should get zeroed by caller. | |
7343 | * NOTE: this function is only called during early init. | |
7344 | */ | |
7345 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
7346 | { | |
7347 | enum zone_type j; | |
7348 | int nid = pgdat->node_id; | |
5f63b720 | 7349 | |
03e85f9d | 7350 | pgdat_init_internals(pgdat); |
385386cf JW |
7351 | pgdat->per_cpu_nodestats = &boot_nodestats; |
7352 | ||
1da177e4 LT |
7353 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7354 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 7355 | unsigned long size, freesize, memmap_pages; |
1da177e4 | 7356 | |
febd5949 | 7357 | size = zone->spanned_pages; |
e6943859 | 7358 | freesize = zone->present_pages; |
1da177e4 | 7359 | |
0e0b864e | 7360 | /* |
9feedc9d | 7361 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
7362 | * is used by this zone for memmap. This affects the watermark |
7363 | * and per-cpu initialisations | |
7364 | */ | |
e6943859 | 7365 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
7366 | if (!is_highmem_idx(j)) { |
7367 | if (freesize >= memmap_pages) { | |
7368 | freesize -= memmap_pages; | |
7369 | if (memmap_pages) | |
9660ecaa HK |
7370 | pr_debug(" %s zone: %lu pages used for memmap\n", |
7371 | zone_names[j], memmap_pages); | |
ba914f48 | 7372 | } else |
e47aa905 | 7373 | pr_warn(" %s zone: %lu memmap pages exceeds freesize %lu\n", |
ba914f48 ZH |
7374 | zone_names[j], memmap_pages, freesize); |
7375 | } | |
0e0b864e | 7376 | |
6267276f | 7377 | /* Account for reserved pages */ |
9feedc9d JL |
7378 | if (j == 0 && freesize > dma_reserve) { |
7379 | freesize -= dma_reserve; | |
9660ecaa | 7380 | pr_debug(" %s zone: %lu pages reserved\n", zone_names[0], dma_reserve); |
0e0b864e MG |
7381 | } |
7382 | ||
98d2b0eb | 7383 | if (!is_highmem_idx(j)) |
9feedc9d | 7384 | nr_kernel_pages += freesize; |
01cefaef JL |
7385 | /* Charge for highmem memmap if there are enough kernel pages */ |
7386 | else if (nr_kernel_pages > memmap_pages * 2) | |
7387 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 7388 | nr_all_pages += freesize; |
1da177e4 | 7389 | |
9feedc9d JL |
7390 | /* |
7391 | * Set an approximate value for lowmem here, it will be adjusted | |
7392 | * when the bootmem allocator frees pages into the buddy system. | |
7393 | * And all highmem pages will be managed by the buddy system. | |
7394 | */ | |
03e85f9d | 7395 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 7396 | |
d883c6cf | 7397 | if (!size) |
1da177e4 LT |
7398 | continue; |
7399 | ||
955c1cd7 | 7400 | set_pageblock_order(); |
7010a6ec | 7401 | setup_usemap(zone); |
9699ee7b | 7402 | init_currently_empty_zone(zone, zone->zone_start_pfn, size); |
1da177e4 LT |
7403 | } |
7404 | } | |
7405 | ||
0cd842f9 | 7406 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 7407 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 7408 | { |
b0aeba74 | 7409 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
7410 | unsigned long __maybe_unused offset = 0; |
7411 | ||
1da177e4 LT |
7412 | /* Skip empty nodes */ |
7413 | if (!pgdat->node_spanned_pages) | |
7414 | return; | |
7415 | ||
b0aeba74 TL |
7416 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
7417 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
7418 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
7419 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 7420 | unsigned long size, end; |
d41dee36 AW |
7421 | struct page *map; |
7422 | ||
e984bb43 BP |
7423 | /* |
7424 | * The zone's endpoints aren't required to be MAX_ORDER | |
7425 | * aligned but the node_mem_map endpoints must be in order | |
7426 | * for the buddy allocator to function correctly. | |
7427 | */ | |
108bcc96 | 7428 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
7429 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
7430 | size = (end - start) * sizeof(struct page); | |
26fb3dae MR |
7431 | map = memblock_alloc_node(size, SMP_CACHE_BYTES, |
7432 | pgdat->node_id); | |
23a7052a MR |
7433 | if (!map) |
7434 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
7435 | size, pgdat->node_id); | |
a1c34a3b | 7436 | pgdat->node_mem_map = map + offset; |
1da177e4 | 7437 | } |
0cd842f9 OS |
7438 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
7439 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
7440 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 7441 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
7442 | /* |
7443 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
7444 | */ | |
c713216d | 7445 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 7446 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d | 7447 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 7448 | mem_map -= offset; |
c713216d | 7449 | } |
1da177e4 LT |
7450 | #endif |
7451 | } | |
0cd842f9 OS |
7452 | #else |
7453 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
7454 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 7455 | |
0188dc98 OS |
7456 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
7457 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
7458 | { | |
0188dc98 OS |
7459 | pgdat->first_deferred_pfn = ULONG_MAX; |
7460 | } | |
7461 | #else | |
7462 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
7463 | #endif | |
7464 | ||
854e8848 | 7465 | static void __init free_area_init_node(int nid) |
1da177e4 | 7466 | { |
9109fb7b | 7467 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
7468 | unsigned long start_pfn = 0; |
7469 | unsigned long end_pfn = 0; | |
9109fb7b | 7470 | |
88fdf75d | 7471 | /* pg_data_t should be reset to zero when it's allocated */ |
97a225e6 | 7472 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx); |
88fdf75d | 7473 | |
854e8848 | 7474 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
88fdf75d | 7475 | |
1da177e4 | 7476 | pgdat->node_id = nid; |
854e8848 | 7477 | pgdat->node_start_pfn = start_pfn; |
75ef7184 | 7478 | pgdat->per_cpu_nodestats = NULL; |
854e8848 | 7479 | |
8d29e18a | 7480 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
7481 | (u64)start_pfn << PAGE_SHIFT, |
7482 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
854e8848 | 7483 | calculate_node_totalpages(pgdat, start_pfn, end_pfn); |
1da177e4 LT |
7484 | |
7485 | alloc_node_mem_map(pgdat); | |
0188dc98 | 7486 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 7487 | |
7f3eb55b | 7488 | free_area_init_core(pgdat); |
1da177e4 LT |
7489 | } |
7490 | ||
bc9331a1 | 7491 | void __init free_area_init_memoryless_node(int nid) |
3f08a302 | 7492 | { |
854e8848 | 7493 | free_area_init_node(nid); |
3f08a302 MR |
7494 | } |
7495 | ||
418508c1 MS |
7496 | #if MAX_NUMNODES > 1 |
7497 | /* | |
7498 | * Figure out the number of possible node ids. | |
7499 | */ | |
f9872caf | 7500 | void __init setup_nr_node_ids(void) |
418508c1 | 7501 | { |
904a9553 | 7502 | unsigned int highest; |
418508c1 | 7503 | |
904a9553 | 7504 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7505 | nr_node_ids = highest + 1; |
7506 | } | |
418508c1 MS |
7507 | #endif |
7508 | ||
1e01979c TH |
7509 | /** |
7510 | * node_map_pfn_alignment - determine the maximum internode alignment | |
7511 | * | |
7512 | * This function should be called after node map is populated and sorted. | |
7513 | * It calculates the maximum power of two alignment which can distinguish | |
7514 | * all the nodes. | |
7515 | * | |
7516 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
7517 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
7518 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
7519 | * shifted, 1GiB is enough and this function will indicate so. | |
7520 | * | |
7521 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
7522 | * model has fine enough granularity to avoid incorrect mapping for the | |
7523 | * populated node map. | |
7524 | * | |
a862f68a | 7525 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
7526 | * requirement (single node). |
7527 | */ | |
7528 | unsigned long __init node_map_pfn_alignment(void) | |
7529 | { | |
7530 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 7531 | unsigned long start, end, mask; |
98fa15f3 | 7532 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 7533 | int i, nid; |
1e01979c | 7534 | |
c13291a5 | 7535 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
7536 | if (!start || last_nid < 0 || last_nid == nid) { |
7537 | last_nid = nid; | |
7538 | last_end = end; | |
7539 | continue; | |
7540 | } | |
7541 | ||
7542 | /* | |
7543 | * Start with a mask granular enough to pin-point to the | |
7544 | * start pfn and tick off bits one-by-one until it becomes | |
7545 | * too coarse to separate the current node from the last. | |
7546 | */ | |
7547 | mask = ~((1 << __ffs(start)) - 1); | |
7548 | while (mask && last_end <= (start & (mask << 1))) | |
7549 | mask <<= 1; | |
7550 | ||
7551 | /* accumulate all internode masks */ | |
7552 | accl_mask |= mask; | |
7553 | } | |
7554 | ||
7555 | /* convert mask to number of pages */ | |
7556 | return ~accl_mask + 1; | |
7557 | } | |
7558 | ||
c713216d MG |
7559 | /** |
7560 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
7561 | * | |
a862f68a | 7562 | * Return: the minimum PFN based on information provided via |
7d018176 | 7563 | * memblock_set_node(). |
c713216d MG |
7564 | */ |
7565 | unsigned long __init find_min_pfn_with_active_regions(void) | |
7566 | { | |
8a1b25fe | 7567 | return PHYS_PFN(memblock_start_of_DRAM()); |
c713216d MG |
7568 | } |
7569 | ||
37b07e41 LS |
7570 | /* |
7571 | * early_calculate_totalpages() | |
7572 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 7573 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 7574 | */ |
484f51f8 | 7575 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 7576 | { |
7e63efef | 7577 | unsigned long totalpages = 0; |
c13291a5 TH |
7578 | unsigned long start_pfn, end_pfn; |
7579 | int i, nid; | |
7580 | ||
7581 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
7582 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 7583 | |
37b07e41 LS |
7584 | totalpages += pages; |
7585 | if (pages) | |
4b0ef1fe | 7586 | node_set_state(nid, N_MEMORY); |
37b07e41 | 7587 | } |
b8af2941 | 7588 | return totalpages; |
7e63efef MG |
7589 | } |
7590 | ||
2a1e274a MG |
7591 | /* |
7592 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
7593 | * is spread evenly between nodes as long as the nodes have enough | |
7594 | * memory. When they don't, some nodes will have more kernelcore than | |
7595 | * others | |
7596 | */ | |
b224ef85 | 7597 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
7598 | { |
7599 | int i, nid; | |
7600 | unsigned long usable_startpfn; | |
7601 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 7602 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 7603 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 7604 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 7605 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 7606 | struct memblock_region *r; |
b2f3eebe TC |
7607 | |
7608 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
7609 | find_usable_zone_for_movable(); | |
7610 | ||
7611 | /* | |
7612 | * If movable_node is specified, ignore kernelcore and movablecore | |
7613 | * options. | |
7614 | */ | |
7615 | if (movable_node_is_enabled()) { | |
cc6de168 | 7616 | for_each_mem_region(r) { |
136199f0 | 7617 | if (!memblock_is_hotpluggable(r)) |
b2f3eebe TC |
7618 | continue; |
7619 | ||
d622abf7 | 7620 | nid = memblock_get_region_node(r); |
b2f3eebe | 7621 | |
136199f0 | 7622 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
7623 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
7624 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7625 | usable_startpfn; | |
7626 | } | |
7627 | ||
7628 | goto out2; | |
7629 | } | |
2a1e274a | 7630 | |
342332e6 TI |
7631 | /* |
7632 | * If kernelcore=mirror is specified, ignore movablecore option | |
7633 | */ | |
7634 | if (mirrored_kernelcore) { | |
7635 | bool mem_below_4gb_not_mirrored = false; | |
7636 | ||
cc6de168 | 7637 | for_each_mem_region(r) { |
342332e6 TI |
7638 | if (memblock_is_mirror(r)) |
7639 | continue; | |
7640 | ||
d622abf7 | 7641 | nid = memblock_get_region_node(r); |
342332e6 TI |
7642 | |
7643 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
7644 | ||
7645 | if (usable_startpfn < 0x100000) { | |
7646 | mem_below_4gb_not_mirrored = true; | |
7647 | continue; | |
7648 | } | |
7649 | ||
7650 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
7651 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7652 | usable_startpfn; | |
7653 | } | |
7654 | ||
7655 | if (mem_below_4gb_not_mirrored) | |
633bf2fe | 7656 | pr_warn("This configuration results in unmirrored kernel memory.\n"); |
342332e6 TI |
7657 | |
7658 | goto out2; | |
7659 | } | |
7660 | ||
7e63efef | 7661 | /* |
a5c6d650 DR |
7662 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
7663 | * amount of necessary memory. | |
7664 | */ | |
7665 | if (required_kernelcore_percent) | |
7666 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
7667 | 10000UL; | |
7668 | if (required_movablecore_percent) | |
7669 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
7670 | 10000UL; | |
7671 | ||
7672 | /* | |
7673 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
7674 | * kernelcore that corresponds so that memory usable for |
7675 | * any allocation type is evenly spread. If both kernelcore | |
7676 | * and movablecore are specified, then the value of kernelcore | |
7677 | * will be used for required_kernelcore if it's greater than | |
7678 | * what movablecore would have allowed. | |
7679 | */ | |
7680 | if (required_movablecore) { | |
7e63efef MG |
7681 | unsigned long corepages; |
7682 | ||
7683 | /* | |
7684 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
7685 | * was requested by the user | |
7686 | */ | |
7687 | required_movablecore = | |
7688 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 7689 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
7690 | corepages = totalpages - required_movablecore; |
7691 | ||
7692 | required_kernelcore = max(required_kernelcore, corepages); | |
7693 | } | |
7694 | ||
bde304bd XQ |
7695 | /* |
7696 | * If kernelcore was not specified or kernelcore size is larger | |
7697 | * than totalpages, there is no ZONE_MOVABLE. | |
7698 | */ | |
7699 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 7700 | goto out; |
2a1e274a MG |
7701 | |
7702 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
7703 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
7704 | ||
7705 | restart: | |
7706 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
7707 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 7708 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
7709 | unsigned long start_pfn, end_pfn; |
7710 | ||
2a1e274a MG |
7711 | /* |
7712 | * Recalculate kernelcore_node if the division per node | |
7713 | * now exceeds what is necessary to satisfy the requested | |
7714 | * amount of memory for the kernel | |
7715 | */ | |
7716 | if (required_kernelcore < kernelcore_node) | |
7717 | kernelcore_node = required_kernelcore / usable_nodes; | |
7718 | ||
7719 | /* | |
7720 | * As the map is walked, we track how much memory is usable | |
7721 | * by the kernel using kernelcore_remaining. When it is | |
7722 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
7723 | */ | |
7724 | kernelcore_remaining = kernelcore_node; | |
7725 | ||
7726 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 7727 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
7728 | unsigned long size_pages; |
7729 | ||
c13291a5 | 7730 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
7731 | if (start_pfn >= end_pfn) |
7732 | continue; | |
7733 | ||
7734 | /* Account for what is only usable for kernelcore */ | |
7735 | if (start_pfn < usable_startpfn) { | |
7736 | unsigned long kernel_pages; | |
7737 | kernel_pages = min(end_pfn, usable_startpfn) | |
7738 | - start_pfn; | |
7739 | ||
7740 | kernelcore_remaining -= min(kernel_pages, | |
7741 | kernelcore_remaining); | |
7742 | required_kernelcore -= min(kernel_pages, | |
7743 | required_kernelcore); | |
7744 | ||
7745 | /* Continue if range is now fully accounted */ | |
7746 | if (end_pfn <= usable_startpfn) { | |
7747 | ||
7748 | /* | |
7749 | * Push zone_movable_pfn to the end so | |
7750 | * that if we have to rebalance | |
7751 | * kernelcore across nodes, we will | |
7752 | * not double account here | |
7753 | */ | |
7754 | zone_movable_pfn[nid] = end_pfn; | |
7755 | continue; | |
7756 | } | |
7757 | start_pfn = usable_startpfn; | |
7758 | } | |
7759 | ||
7760 | /* | |
7761 | * The usable PFN range for ZONE_MOVABLE is from | |
7762 | * start_pfn->end_pfn. Calculate size_pages as the | |
7763 | * number of pages used as kernelcore | |
7764 | */ | |
7765 | size_pages = end_pfn - start_pfn; | |
7766 | if (size_pages > kernelcore_remaining) | |
7767 | size_pages = kernelcore_remaining; | |
7768 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
7769 | ||
7770 | /* | |
7771 | * Some kernelcore has been met, update counts and | |
7772 | * break if the kernelcore for this node has been | |
b8af2941 | 7773 | * satisfied |
2a1e274a MG |
7774 | */ |
7775 | required_kernelcore -= min(required_kernelcore, | |
7776 | size_pages); | |
7777 | kernelcore_remaining -= size_pages; | |
7778 | if (!kernelcore_remaining) | |
7779 | break; | |
7780 | } | |
7781 | } | |
7782 | ||
7783 | /* | |
7784 | * If there is still required_kernelcore, we do another pass with one | |
7785 | * less node in the count. This will push zone_movable_pfn[nid] further | |
7786 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 7787 | * satisfied |
2a1e274a MG |
7788 | */ |
7789 | usable_nodes--; | |
7790 | if (usable_nodes && required_kernelcore > usable_nodes) | |
7791 | goto restart; | |
7792 | ||
b2f3eebe | 7793 | out2: |
2a1e274a MG |
7794 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
7795 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
7796 | zone_movable_pfn[nid] = | |
7797 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 7798 | |
20e6926d | 7799 | out: |
66918dcd | 7800 | /* restore the node_state */ |
4b0ef1fe | 7801 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
7802 | } |
7803 | ||
4b0ef1fe LJ |
7804 | /* Any regular or high memory on that node ? */ |
7805 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 7806 | { |
37b07e41 LS |
7807 | enum zone_type zone_type; |
7808 | ||
4b0ef1fe | 7809 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 7810 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 7811 | if (populated_zone(zone)) { |
7b0e0c0e OS |
7812 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
7813 | node_set_state(nid, N_HIGH_MEMORY); | |
7814 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 7815 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
7816 | break; |
7817 | } | |
37b07e41 | 7818 | } |
37b07e41 LS |
7819 | } |
7820 | ||
51930df5 | 7821 | /* |
f0953a1b | 7822 | * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For |
51930df5 MR |
7823 | * such cases we allow max_zone_pfn sorted in the descending order |
7824 | */ | |
7825 | bool __weak arch_has_descending_max_zone_pfns(void) | |
7826 | { | |
7827 | return false; | |
7828 | } | |
7829 | ||
c713216d | 7830 | /** |
9691a071 | 7831 | * free_area_init - Initialise all pg_data_t and zone data |
88ca3b94 | 7832 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
7833 | * |
7834 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 7835 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
7836 | * zone in each node and their holes is calculated. If the maximum PFN |
7837 | * between two adjacent zones match, it is assumed that the zone is empty. | |
7838 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
7839 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
7840 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
7841 | * at arch_max_dma_pfn. | |
7842 | */ | |
9691a071 | 7843 | void __init free_area_init(unsigned long *max_zone_pfn) |
c713216d | 7844 | { |
c13291a5 | 7845 | unsigned long start_pfn, end_pfn; |
51930df5 MR |
7846 | int i, nid, zone; |
7847 | bool descending; | |
a6af2bc3 | 7848 | |
c713216d MG |
7849 | /* Record where the zone boundaries are */ |
7850 | memset(arch_zone_lowest_possible_pfn, 0, | |
7851 | sizeof(arch_zone_lowest_possible_pfn)); | |
7852 | memset(arch_zone_highest_possible_pfn, 0, | |
7853 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
7854 | |
7855 | start_pfn = find_min_pfn_with_active_regions(); | |
51930df5 | 7856 | descending = arch_has_descending_max_zone_pfns(); |
90cae1fe OH |
7857 | |
7858 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
51930df5 MR |
7859 | if (descending) |
7860 | zone = MAX_NR_ZONES - i - 1; | |
7861 | else | |
7862 | zone = i; | |
7863 | ||
7864 | if (zone == ZONE_MOVABLE) | |
2a1e274a | 7865 | continue; |
90cae1fe | 7866 | |
51930df5 MR |
7867 | end_pfn = max(max_zone_pfn[zone], start_pfn); |
7868 | arch_zone_lowest_possible_pfn[zone] = start_pfn; | |
7869 | arch_zone_highest_possible_pfn[zone] = end_pfn; | |
90cae1fe OH |
7870 | |
7871 | start_pfn = end_pfn; | |
c713216d | 7872 | } |
2a1e274a MG |
7873 | |
7874 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
7875 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 7876 | find_zone_movable_pfns_for_nodes(); |
c713216d | 7877 | |
c713216d | 7878 | /* Print out the zone ranges */ |
f88dfff5 | 7879 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
7880 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7881 | if (i == ZONE_MOVABLE) | |
7882 | continue; | |
f88dfff5 | 7883 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
7884 | if (arch_zone_lowest_possible_pfn[i] == |
7885 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 7886 | pr_cont("empty\n"); |
72f0ba02 | 7887 | else |
8d29e18a JG |
7888 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
7889 | (u64)arch_zone_lowest_possible_pfn[i] | |
7890 | << PAGE_SHIFT, | |
7891 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 7892 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
7893 | } |
7894 | ||
7895 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 7896 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
7897 | for (i = 0; i < MAX_NUMNODES; i++) { |
7898 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
7899 | pr_info(" Node %d: %#018Lx\n", i, |
7900 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 7901 | } |
c713216d | 7902 | |
f46edbd1 DW |
7903 | /* |
7904 | * Print out the early node map, and initialize the | |
7905 | * subsection-map relative to active online memory ranges to | |
7906 | * enable future "sub-section" extensions of the memory map. | |
7907 | */ | |
f88dfff5 | 7908 | pr_info("Early memory node ranges\n"); |
f46edbd1 | 7909 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
8d29e18a JG |
7910 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
7911 | (u64)start_pfn << PAGE_SHIFT, | |
7912 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
f46edbd1 DW |
7913 | subsection_map_init(start_pfn, end_pfn - start_pfn); |
7914 | } | |
c713216d MG |
7915 | |
7916 | /* Initialise every node */ | |
708614e6 | 7917 | mminit_verify_pageflags_layout(); |
8ef82866 | 7918 | setup_nr_node_ids(); |
c713216d MG |
7919 | for_each_online_node(nid) { |
7920 | pg_data_t *pgdat = NODE_DATA(nid); | |
854e8848 | 7921 | free_area_init_node(nid); |
37b07e41 LS |
7922 | |
7923 | /* Any memory on that node */ | |
7924 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
7925 | node_set_state(nid, N_MEMORY); |
7926 | check_for_memory(pgdat, nid); | |
c713216d | 7927 | } |
122e093c MR |
7928 | |
7929 | memmap_init(); | |
c713216d | 7930 | } |
2a1e274a | 7931 | |
a5c6d650 DR |
7932 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
7933 | unsigned long *percent) | |
2a1e274a MG |
7934 | { |
7935 | unsigned long long coremem; | |
a5c6d650 DR |
7936 | char *endptr; |
7937 | ||
2a1e274a MG |
7938 | if (!p) |
7939 | return -EINVAL; | |
7940 | ||
a5c6d650 DR |
7941 | /* Value may be a percentage of total memory, otherwise bytes */ |
7942 | coremem = simple_strtoull(p, &endptr, 0); | |
7943 | if (*endptr == '%') { | |
7944 | /* Paranoid check for percent values greater than 100 */ | |
7945 | WARN_ON(coremem > 100); | |
2a1e274a | 7946 | |
a5c6d650 DR |
7947 | *percent = coremem; |
7948 | } else { | |
7949 | coremem = memparse(p, &p); | |
7950 | /* Paranoid check that UL is enough for the coremem value */ | |
7951 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7952 | |
a5c6d650 DR |
7953 | *core = coremem >> PAGE_SHIFT; |
7954 | *percent = 0UL; | |
7955 | } | |
2a1e274a MG |
7956 | return 0; |
7957 | } | |
ed7ed365 | 7958 | |
7e63efef MG |
7959 | /* |
7960 | * kernelcore=size sets the amount of memory for use for allocations that | |
7961 | * cannot be reclaimed or migrated. | |
7962 | */ | |
7963 | static int __init cmdline_parse_kernelcore(char *p) | |
7964 | { | |
342332e6 TI |
7965 | /* parse kernelcore=mirror */ |
7966 | if (parse_option_str(p, "mirror")) { | |
7967 | mirrored_kernelcore = true; | |
7968 | return 0; | |
7969 | } | |
7970 | ||
a5c6d650 DR |
7971 | return cmdline_parse_core(p, &required_kernelcore, |
7972 | &required_kernelcore_percent); | |
7e63efef MG |
7973 | } |
7974 | ||
7975 | /* | |
7976 | * movablecore=size sets the amount of memory for use for allocations that | |
7977 | * can be reclaimed or migrated. | |
7978 | */ | |
7979 | static int __init cmdline_parse_movablecore(char *p) | |
7980 | { | |
a5c6d650 DR |
7981 | return cmdline_parse_core(p, &required_movablecore, |
7982 | &required_movablecore_percent); | |
7e63efef MG |
7983 | } |
7984 | ||
ed7ed365 | 7985 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7986 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7987 | |
c3d5f5f0 JL |
7988 | void adjust_managed_page_count(struct page *page, long count) |
7989 | { | |
9705bea5 | 7990 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 7991 | totalram_pages_add(count); |
3dcc0571 JL |
7992 | #ifdef CONFIG_HIGHMEM |
7993 | if (PageHighMem(page)) | |
ca79b0c2 | 7994 | totalhigh_pages_add(count); |
3dcc0571 | 7995 | #endif |
c3d5f5f0 | 7996 | } |
3dcc0571 | 7997 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7998 | |
e5cb113f | 7999 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 8000 | { |
11199692 JL |
8001 | void *pos; |
8002 | unsigned long pages = 0; | |
69afade7 | 8003 | |
11199692 JL |
8004 | start = (void *)PAGE_ALIGN((unsigned long)start); |
8005 | end = (void *)((unsigned long)end & PAGE_MASK); | |
8006 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
8007 | struct page *page = virt_to_page(pos); |
8008 | void *direct_map_addr; | |
8009 | ||
8010 | /* | |
8011 | * 'direct_map_addr' might be different from 'pos' | |
8012 | * because some architectures' virt_to_page() | |
8013 | * work with aliases. Getting the direct map | |
8014 | * address ensures that we get a _writeable_ | |
8015 | * alias for the memset(). | |
8016 | */ | |
8017 | direct_map_addr = page_address(page); | |
c746170d VF |
8018 | /* |
8019 | * Perform a kasan-unchecked memset() since this memory | |
8020 | * has not been initialized. | |
8021 | */ | |
8022 | direct_map_addr = kasan_reset_tag(direct_map_addr); | |
dbe67df4 | 8023 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
8024 | memset(direct_map_addr, poison, PAGE_SIZE); |
8025 | ||
8026 | free_reserved_page(page); | |
69afade7 JL |
8027 | } |
8028 | ||
8029 | if (pages && s) | |
adb1fe9a JP |
8030 | pr_info("Freeing %s memory: %ldK\n", |
8031 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
8032 | |
8033 | return pages; | |
8034 | } | |
8035 | ||
1f9d03c5 | 8036 | void __init mem_init_print_info(void) |
7ee3d4e8 JL |
8037 | { |
8038 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
8039 | unsigned long init_code_size, init_data_size; | |
8040 | ||
8041 | physpages = get_num_physpages(); | |
8042 | codesize = _etext - _stext; | |
8043 | datasize = _edata - _sdata; | |
8044 | rosize = __end_rodata - __start_rodata; | |
8045 | bss_size = __bss_stop - __bss_start; | |
8046 | init_data_size = __init_end - __init_begin; | |
8047 | init_code_size = _einittext - _sinittext; | |
8048 | ||
8049 | /* | |
8050 | * Detect special cases and adjust section sizes accordingly: | |
8051 | * 1) .init.* may be embedded into .data sections | |
8052 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
8053 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
8054 | * 3) .rodata.* may be embedded into .text or .data sections. | |
8055 | */ | |
8056 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
8057 | do { \ |
8058 | if (start <= pos && pos < end && size > adj) \ | |
8059 | size -= adj; \ | |
8060 | } while (0) | |
7ee3d4e8 JL |
8061 | |
8062 | adj_init_size(__init_begin, __init_end, init_data_size, | |
8063 | _sinittext, init_code_size); | |
8064 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
8065 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
8066 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
8067 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
8068 | ||
8069 | #undef adj_init_size | |
8070 | ||
756a025f | 8071 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 8072 | #ifdef CONFIG_HIGHMEM |
756a025f | 8073 | ", %luK highmem" |
7ee3d4e8 | 8074 | #endif |
1f9d03c5 | 8075 | ")\n", |
756a025f JP |
8076 | nr_free_pages() << (PAGE_SHIFT - 10), |
8077 | physpages << (PAGE_SHIFT - 10), | |
8078 | codesize >> 10, datasize >> 10, rosize >> 10, | |
8079 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
ca79b0c2 | 8080 | (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10), |
1f9d03c5 | 8081 | totalcma_pages << (PAGE_SHIFT - 10) |
7ee3d4e8 | 8082 | #ifdef CONFIG_HIGHMEM |
1f9d03c5 | 8083 | , totalhigh_pages() << (PAGE_SHIFT - 10) |
7ee3d4e8 | 8084 | #endif |
1f9d03c5 | 8085 | ); |
7ee3d4e8 JL |
8086 | } |
8087 | ||
0e0b864e | 8088 | /** |
88ca3b94 RD |
8089 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
8090 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 8091 | * |
013110a7 | 8092 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
8093 | * In the DMA zone, a significant percentage may be consumed by kernel image |
8094 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
8095 | * function may optionally be used to account for unfreeable pages in the |
8096 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
8097 | * smaller per-cpu batchsize. | |
0e0b864e MG |
8098 | */ |
8099 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
8100 | { | |
8101 | dma_reserve = new_dma_reserve; | |
8102 | } | |
8103 | ||
005fd4bb | 8104 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 8105 | { |
04f8cfea | 8106 | struct zone *zone; |
1da177e4 | 8107 | |
005fd4bb SAS |
8108 | lru_add_drain_cpu(cpu); |
8109 | drain_pages(cpu); | |
9f8f2172 | 8110 | |
005fd4bb SAS |
8111 | /* |
8112 | * Spill the event counters of the dead processor | |
8113 | * into the current processors event counters. | |
8114 | * This artificially elevates the count of the current | |
8115 | * processor. | |
8116 | */ | |
8117 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 8118 | |
005fd4bb SAS |
8119 | /* |
8120 | * Zero the differential counters of the dead processor | |
8121 | * so that the vm statistics are consistent. | |
8122 | * | |
8123 | * This is only okay since the processor is dead and cannot | |
8124 | * race with what we are doing. | |
8125 | */ | |
8126 | cpu_vm_stats_fold(cpu); | |
04f8cfea MG |
8127 | |
8128 | for_each_populated_zone(zone) | |
8129 | zone_pcp_update(zone, 0); | |
8130 | ||
8131 | return 0; | |
8132 | } | |
8133 | ||
8134 | static int page_alloc_cpu_online(unsigned int cpu) | |
8135 | { | |
8136 | struct zone *zone; | |
8137 | ||
8138 | for_each_populated_zone(zone) | |
8139 | zone_pcp_update(zone, 1); | |
005fd4bb | 8140 | return 0; |
1da177e4 | 8141 | } |
1da177e4 | 8142 | |
e03a5125 NP |
8143 | #ifdef CONFIG_NUMA |
8144 | int hashdist = HASHDIST_DEFAULT; | |
8145 | ||
8146 | static int __init set_hashdist(char *str) | |
8147 | { | |
8148 | if (!str) | |
8149 | return 0; | |
8150 | hashdist = simple_strtoul(str, &str, 0); | |
8151 | return 1; | |
8152 | } | |
8153 | __setup("hashdist=", set_hashdist); | |
8154 | #endif | |
8155 | ||
1da177e4 LT |
8156 | void __init page_alloc_init(void) |
8157 | { | |
005fd4bb SAS |
8158 | int ret; |
8159 | ||
e03a5125 NP |
8160 | #ifdef CONFIG_NUMA |
8161 | if (num_node_state(N_MEMORY) == 1) | |
8162 | hashdist = 0; | |
8163 | #endif | |
8164 | ||
04f8cfea MG |
8165 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC, |
8166 | "mm/page_alloc:pcp", | |
8167 | page_alloc_cpu_online, | |
005fd4bb SAS |
8168 | page_alloc_cpu_dead); |
8169 | WARN_ON(ret < 0); | |
1da177e4 LT |
8170 | } |
8171 | ||
cb45b0e9 | 8172 | /* |
34b10060 | 8173 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
8174 | * or min_free_kbytes changes. |
8175 | */ | |
8176 | static void calculate_totalreserve_pages(void) | |
8177 | { | |
8178 | struct pglist_data *pgdat; | |
8179 | unsigned long reserve_pages = 0; | |
2f6726e5 | 8180 | enum zone_type i, j; |
cb45b0e9 HA |
8181 | |
8182 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
8183 | |
8184 | pgdat->totalreserve_pages = 0; | |
8185 | ||
cb45b0e9 HA |
8186 | for (i = 0; i < MAX_NR_ZONES; i++) { |
8187 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 8188 | long max = 0; |
9705bea5 | 8189 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
8190 | |
8191 | /* Find valid and maximum lowmem_reserve in the zone */ | |
8192 | for (j = i; j < MAX_NR_ZONES; j++) { | |
8193 | if (zone->lowmem_reserve[j] > max) | |
8194 | max = zone->lowmem_reserve[j]; | |
8195 | } | |
8196 | ||
41858966 MG |
8197 | /* we treat the high watermark as reserved pages. */ |
8198 | max += high_wmark_pages(zone); | |
cb45b0e9 | 8199 | |
3d6357de AK |
8200 | if (max > managed_pages) |
8201 | max = managed_pages; | |
a8d01437 | 8202 | |
281e3726 | 8203 | pgdat->totalreserve_pages += max; |
a8d01437 | 8204 | |
cb45b0e9 HA |
8205 | reserve_pages += max; |
8206 | } | |
8207 | } | |
8208 | totalreserve_pages = reserve_pages; | |
8209 | } | |
8210 | ||
1da177e4 LT |
8211 | /* |
8212 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 8213 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
8214 | * has a correct pages reserved value, so an adequate number of |
8215 | * pages are left in the zone after a successful __alloc_pages(). | |
8216 | */ | |
8217 | static void setup_per_zone_lowmem_reserve(void) | |
8218 | { | |
8219 | struct pglist_data *pgdat; | |
470c61d7 | 8220 | enum zone_type i, j; |
1da177e4 | 8221 | |
ec936fc5 | 8222 | for_each_online_pgdat(pgdat) { |
470c61d7 LS |
8223 | for (i = 0; i < MAX_NR_ZONES - 1; i++) { |
8224 | struct zone *zone = &pgdat->node_zones[i]; | |
8225 | int ratio = sysctl_lowmem_reserve_ratio[i]; | |
8226 | bool clear = !ratio || !zone_managed_pages(zone); | |
8227 | unsigned long managed_pages = 0; | |
8228 | ||
8229 | for (j = i + 1; j < MAX_NR_ZONES; j++) { | |
f7ec1044 LS |
8230 | struct zone *upper_zone = &pgdat->node_zones[j]; |
8231 | ||
8232 | managed_pages += zone_managed_pages(upper_zone); | |
470c61d7 | 8233 | |
f7ec1044 LS |
8234 | if (clear) |
8235 | zone->lowmem_reserve[j] = 0; | |
8236 | else | |
470c61d7 | 8237 | zone->lowmem_reserve[j] = managed_pages / ratio; |
1da177e4 LT |
8238 | } |
8239 | } | |
8240 | } | |
cb45b0e9 HA |
8241 | |
8242 | /* update totalreserve_pages */ | |
8243 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8244 | } |
8245 | ||
cfd3da1e | 8246 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
8247 | { |
8248 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
8249 | unsigned long lowmem_pages = 0; | |
8250 | struct zone *zone; | |
8251 | unsigned long flags; | |
8252 | ||
8253 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
8254 | for_each_zone(zone) { | |
8255 | if (!is_highmem(zone)) | |
9705bea5 | 8256 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
8257 | } |
8258 | ||
8259 | for_each_zone(zone) { | |
ac924c60 AM |
8260 | u64 tmp; |
8261 | ||
1125b4e3 | 8262 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 8263 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 8264 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
8265 | if (is_highmem(zone)) { |
8266 | /* | |
669ed175 NP |
8267 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
8268 | * need highmem pages, so cap pages_min to a small | |
8269 | * value here. | |
8270 | * | |
41858966 | 8271 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 8272 | * deltas control async page reclaim, and so should |
669ed175 | 8273 | * not be capped for highmem. |
1da177e4 | 8274 | */ |
90ae8d67 | 8275 | unsigned long min_pages; |
1da177e4 | 8276 | |
9705bea5 | 8277 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 8278 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 8279 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 8280 | } else { |
669ed175 NP |
8281 | /* |
8282 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
8283 | * proportionate to the zone's size. |
8284 | */ | |
a9214443 | 8285 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
8286 | } |
8287 | ||
795ae7a0 JW |
8288 | /* |
8289 | * Set the kswapd watermarks distance according to the | |
8290 | * scale factor in proportion to available memory, but | |
8291 | * ensure a minimum size on small systems. | |
8292 | */ | |
8293 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 8294 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
8295 | watermark_scale_factor, 10000)); |
8296 | ||
aa092591 | 8297 | zone->watermark_boost = 0; |
a9214443 MG |
8298 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
8299 | zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 8300 | |
1125b4e3 | 8301 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 8302 | } |
cb45b0e9 HA |
8303 | |
8304 | /* update totalreserve_pages */ | |
8305 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8306 | } |
8307 | ||
cfd3da1e MG |
8308 | /** |
8309 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
8310 | * or when memory is hot-{added|removed} | |
8311 | * | |
8312 | * Ensures that the watermark[min,low,high] values for each zone are set | |
8313 | * correctly with respect to min_free_kbytes. | |
8314 | */ | |
8315 | void setup_per_zone_wmarks(void) | |
8316 | { | |
b92ca18e | 8317 | struct zone *zone; |
b93e0f32 MH |
8318 | static DEFINE_SPINLOCK(lock); |
8319 | ||
8320 | spin_lock(&lock); | |
cfd3da1e | 8321 | __setup_per_zone_wmarks(); |
b93e0f32 | 8322 | spin_unlock(&lock); |
b92ca18e MG |
8323 | |
8324 | /* | |
8325 | * The watermark size have changed so update the pcpu batch | |
8326 | * and high limits or the limits may be inappropriate. | |
8327 | */ | |
8328 | for_each_zone(zone) | |
04f8cfea | 8329 | zone_pcp_update(zone, 0); |
cfd3da1e MG |
8330 | } |
8331 | ||
1da177e4 LT |
8332 | /* |
8333 | * Initialise min_free_kbytes. | |
8334 | * | |
8335 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 8336 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
8337 | * bandwidth does not increase linearly with machine size. We use |
8338 | * | |
b8af2941 | 8339 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
8340 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
8341 | * | |
8342 | * which yields | |
8343 | * | |
8344 | * 16MB: 512k | |
8345 | * 32MB: 724k | |
8346 | * 64MB: 1024k | |
8347 | * 128MB: 1448k | |
8348 | * 256MB: 2048k | |
8349 | * 512MB: 2896k | |
8350 | * 1024MB: 4096k | |
8351 | * 2048MB: 5792k | |
8352 | * 4096MB: 8192k | |
8353 | * 8192MB: 11584k | |
8354 | * 16384MB: 16384k | |
8355 | */ | |
1b79acc9 | 8356 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
8357 | { |
8358 | unsigned long lowmem_kbytes; | |
5f12733e | 8359 | int new_min_free_kbytes; |
1da177e4 LT |
8360 | |
8361 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
8362 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
8363 | ||
8364 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
8365 | min_free_kbytes = new_min_free_kbytes; | |
8366 | if (min_free_kbytes < 128) | |
8367 | min_free_kbytes = 128; | |
ee8eb9a5 JS |
8368 | if (min_free_kbytes > 262144) |
8369 | min_free_kbytes = 262144; | |
5f12733e MH |
8370 | } else { |
8371 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
8372 | new_min_free_kbytes, user_min_free_kbytes); | |
8373 | } | |
bc75d33f | 8374 | setup_per_zone_wmarks(); |
a6cccdc3 | 8375 | refresh_zone_stat_thresholds(); |
1da177e4 | 8376 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
8377 | |
8378 | #ifdef CONFIG_NUMA | |
8379 | setup_min_unmapped_ratio(); | |
8380 | setup_min_slab_ratio(); | |
8381 | #endif | |
8382 | ||
4aab2be0 VB |
8383 | khugepaged_min_free_kbytes_update(); |
8384 | ||
1da177e4 LT |
8385 | return 0; |
8386 | } | |
e08d3fdf | 8387 | postcore_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
8388 | |
8389 | /* | |
b8af2941 | 8390 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
8391 | * that we can call two helper functions whenever min_free_kbytes |
8392 | * changes. | |
8393 | */ | |
cccad5b9 | 8394 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8395 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8396 | { |
da8c757b HP |
8397 | int rc; |
8398 | ||
8399 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8400 | if (rc) | |
8401 | return rc; | |
8402 | ||
5f12733e MH |
8403 | if (write) { |
8404 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 8405 | setup_per_zone_wmarks(); |
5f12733e | 8406 | } |
1da177e4 LT |
8407 | return 0; |
8408 | } | |
8409 | ||
795ae7a0 | 8410 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8411 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
8412 | { |
8413 | int rc; | |
8414 | ||
8415 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8416 | if (rc) | |
8417 | return rc; | |
8418 | ||
8419 | if (write) | |
8420 | setup_per_zone_wmarks(); | |
8421 | ||
8422 | return 0; | |
8423 | } | |
8424 | ||
9614634f | 8425 | #ifdef CONFIG_NUMA |
6423aa81 | 8426 | static void setup_min_unmapped_ratio(void) |
9614634f | 8427 | { |
6423aa81 | 8428 | pg_data_t *pgdat; |
9614634f | 8429 | struct zone *zone; |
9614634f | 8430 | |
a5f5f91d | 8431 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 8432 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 8433 | |
9614634f | 8434 | for_each_zone(zone) |
9705bea5 AK |
8435 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
8436 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 8437 | } |
0ff38490 | 8438 | |
6423aa81 JK |
8439 | |
8440 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8441 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 8442 | { |
0ff38490 CL |
8443 | int rc; |
8444 | ||
8d65af78 | 8445 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
8446 | if (rc) |
8447 | return rc; | |
8448 | ||
6423aa81 JK |
8449 | setup_min_unmapped_ratio(); |
8450 | ||
8451 | return 0; | |
8452 | } | |
8453 | ||
8454 | static void setup_min_slab_ratio(void) | |
8455 | { | |
8456 | pg_data_t *pgdat; | |
8457 | struct zone *zone; | |
8458 | ||
a5f5f91d MG |
8459 | for_each_online_pgdat(pgdat) |
8460 | pgdat->min_slab_pages = 0; | |
8461 | ||
0ff38490 | 8462 | for_each_zone(zone) |
9705bea5 AK |
8463 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
8464 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
8465 | } |
8466 | ||
8467 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8468 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
8469 | { |
8470 | int rc; | |
8471 | ||
8472 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8473 | if (rc) | |
8474 | return rc; | |
8475 | ||
8476 | setup_min_slab_ratio(); | |
8477 | ||
0ff38490 CL |
8478 | return 0; |
8479 | } | |
9614634f CL |
8480 | #endif |
8481 | ||
1da177e4 LT |
8482 | /* |
8483 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
8484 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
8485 | * whenever sysctl_lowmem_reserve_ratio changes. | |
8486 | * | |
8487 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 8488 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
8489 | * if in function of the boot time zone sizes. |
8490 | */ | |
cccad5b9 | 8491 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8492 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8493 | { |
86aaf255 BH |
8494 | int i; |
8495 | ||
8d65af78 | 8496 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
8497 | |
8498 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
8499 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
8500 | sysctl_lowmem_reserve_ratio[i] = 0; | |
8501 | } | |
8502 | ||
1da177e4 LT |
8503 | setup_per_zone_lowmem_reserve(); |
8504 | return 0; | |
8505 | } | |
8506 | ||
74f44822 MG |
8507 | /* |
8508 | * percpu_pagelist_high_fraction - changes the pcp->high for each zone on each | |
8509 | * cpu. It is the fraction of total pages in each zone that a hot per cpu | |
8510 | * pagelist can have before it gets flushed back to buddy allocator. | |
8511 | */ | |
8512 | int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *table, | |
8513 | int write, void *buffer, size_t *length, loff_t *ppos) | |
8514 | { | |
8515 | struct zone *zone; | |
8516 | int old_percpu_pagelist_high_fraction; | |
8517 | int ret; | |
8518 | ||
8519 | mutex_lock(&pcp_batch_high_lock); | |
8520 | old_percpu_pagelist_high_fraction = percpu_pagelist_high_fraction; | |
8521 | ||
8522 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8523 | if (!write || ret < 0) | |
8524 | goto out; | |
8525 | ||
8526 | /* Sanity checking to avoid pcp imbalance */ | |
8527 | if (percpu_pagelist_high_fraction && | |
8528 | percpu_pagelist_high_fraction < MIN_PERCPU_PAGELIST_HIGH_FRACTION) { | |
8529 | percpu_pagelist_high_fraction = old_percpu_pagelist_high_fraction; | |
8530 | ret = -EINVAL; | |
8531 | goto out; | |
8532 | } | |
8533 | ||
8534 | /* No change? */ | |
8535 | if (percpu_pagelist_high_fraction == old_percpu_pagelist_high_fraction) | |
8536 | goto out; | |
8537 | ||
8538 | for_each_populated_zone(zone) | |
8539 | zone_set_pageset_high_and_batch(zone, 0); | |
8540 | out: | |
8541 | mutex_unlock(&pcp_batch_high_lock); | |
8542 | return ret; | |
8543 | } | |
8544 | ||
f6f34b43 SD |
8545 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
8546 | /* | |
8547 | * Returns the number of pages that arch has reserved but | |
8548 | * is not known to alloc_large_system_hash(). | |
8549 | */ | |
8550 | static unsigned long __init arch_reserved_kernel_pages(void) | |
8551 | { | |
8552 | return 0; | |
8553 | } | |
8554 | #endif | |
8555 | ||
9017217b PT |
8556 | /* |
8557 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
8558 | * machines. As memory size is increased the scale is also increased but at | |
8559 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
8560 | * quadruples the scale is increased by one, which means the size of hash table | |
8561 | * only doubles, instead of quadrupling as well. | |
8562 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
8563 | * makes sense, it is disabled on such platforms. | |
8564 | */ | |
8565 | #if __BITS_PER_LONG > 32 | |
8566 | #define ADAPT_SCALE_BASE (64ul << 30) | |
8567 | #define ADAPT_SCALE_SHIFT 2 | |
8568 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
8569 | #endif | |
8570 | ||
1da177e4 LT |
8571 | /* |
8572 | * allocate a large system hash table from bootmem | |
8573 | * - it is assumed that the hash table must contain an exact power-of-2 | |
8574 | * quantity of entries | |
8575 | * - limit is the number of hash buckets, not the total allocation size | |
8576 | */ | |
8577 | void *__init alloc_large_system_hash(const char *tablename, | |
8578 | unsigned long bucketsize, | |
8579 | unsigned long numentries, | |
8580 | int scale, | |
8581 | int flags, | |
8582 | unsigned int *_hash_shift, | |
8583 | unsigned int *_hash_mask, | |
31fe62b9 TB |
8584 | unsigned long low_limit, |
8585 | unsigned long high_limit) | |
1da177e4 | 8586 | { |
31fe62b9 | 8587 | unsigned long long max = high_limit; |
1da177e4 LT |
8588 | unsigned long log2qty, size; |
8589 | void *table = NULL; | |
3749a8f0 | 8590 | gfp_t gfp_flags; |
ec11408a | 8591 | bool virt; |
121e6f32 | 8592 | bool huge; |
1da177e4 LT |
8593 | |
8594 | /* allow the kernel cmdline to have a say */ | |
8595 | if (!numentries) { | |
8596 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 8597 | numentries = nr_kernel_pages; |
f6f34b43 | 8598 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
8599 | |
8600 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
8601 | if (PAGE_SHIFT < 20) | |
8602 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 8603 | |
9017217b PT |
8604 | #if __BITS_PER_LONG > 32 |
8605 | if (!high_limit) { | |
8606 | unsigned long adapt; | |
8607 | ||
8608 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
8609 | adapt <<= ADAPT_SCALE_SHIFT) | |
8610 | scale++; | |
8611 | } | |
8612 | #endif | |
8613 | ||
1da177e4 LT |
8614 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
8615 | if (scale > PAGE_SHIFT) | |
8616 | numentries >>= (scale - PAGE_SHIFT); | |
8617 | else | |
8618 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
8619 | |
8620 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
8621 | if (unlikely(flags & HASH_SMALL)) { |
8622 | /* Makes no sense without HASH_EARLY */ | |
8623 | WARN_ON(!(flags & HASH_EARLY)); | |
8624 | if (!(numentries >> *_hash_shift)) { | |
8625 | numentries = 1UL << *_hash_shift; | |
8626 | BUG_ON(!numentries); | |
8627 | } | |
8628 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 8629 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 8630 | } |
6e692ed3 | 8631 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
8632 | |
8633 | /* limit allocation size to 1/16 total memory by default */ | |
8634 | if (max == 0) { | |
8635 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
8636 | do_div(max, bucketsize); | |
8637 | } | |
074b8517 | 8638 | max = min(max, 0x80000000ULL); |
1da177e4 | 8639 | |
31fe62b9 TB |
8640 | if (numentries < low_limit) |
8641 | numentries = low_limit; | |
1da177e4 LT |
8642 | if (numentries > max) |
8643 | numentries = max; | |
8644 | ||
f0d1b0b3 | 8645 | log2qty = ilog2(numentries); |
1da177e4 | 8646 | |
3749a8f0 | 8647 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 8648 | do { |
ec11408a | 8649 | virt = false; |
1da177e4 | 8650 | size = bucketsize << log2qty; |
ea1f5f37 PT |
8651 | if (flags & HASH_EARLY) { |
8652 | if (flags & HASH_ZERO) | |
26fb3dae | 8653 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 8654 | else |
7e1c4e27 MR |
8655 | table = memblock_alloc_raw(size, |
8656 | SMP_CACHE_BYTES); | |
ec11408a | 8657 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
88dca4ca | 8658 | table = __vmalloc(size, gfp_flags); |
ec11408a | 8659 | virt = true; |
121e6f32 | 8660 | huge = is_vm_area_hugepages(table); |
ea1f5f37 | 8661 | } else { |
1037b83b ED |
8662 | /* |
8663 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
8664 | * some pages at the end of hash table which |
8665 | * alloc_pages_exact() automatically does | |
1037b83b | 8666 | */ |
ec11408a NP |
8667 | table = alloc_pages_exact(size, gfp_flags); |
8668 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
8669 | } |
8670 | } while (!table && size > PAGE_SIZE && --log2qty); | |
8671 | ||
8672 | if (!table) | |
8673 | panic("Failed to allocate %s hash table\n", tablename); | |
8674 | ||
ec11408a NP |
8675 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
8676 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
121e6f32 | 8677 | virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear"); |
1da177e4 LT |
8678 | |
8679 | if (_hash_shift) | |
8680 | *_hash_shift = log2qty; | |
8681 | if (_hash_mask) | |
8682 | *_hash_mask = (1 << log2qty) - 1; | |
8683 | ||
8684 | return table; | |
8685 | } | |
a117e66e | 8686 | |
a5d76b54 | 8687 | /* |
80934513 | 8688 | * This function checks whether pageblock includes unmovable pages or not. |
80934513 | 8689 | * |
b8af2941 | 8690 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
8691 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
8692 | * check without lock_page also may miss some movable non-lru pages at | |
8693 | * race condition. So you can't expect this function should be exact. | |
4a55c047 QC |
8694 | * |
8695 | * Returns a page without holding a reference. If the caller wants to | |
047b9967 | 8696 | * dereference that page (e.g., dumping), it has to make sure that it |
4a55c047 QC |
8697 | * cannot get removed (e.g., via memory unplug) concurrently. |
8698 | * | |
a5d76b54 | 8699 | */ |
4a55c047 QC |
8700 | struct page *has_unmovable_pages(struct zone *zone, struct page *page, |
8701 | int migratetype, int flags) | |
49ac8255 | 8702 | { |
1a9f2191 QC |
8703 | unsigned long iter = 0; |
8704 | unsigned long pfn = page_to_pfn(page); | |
6a654e36 | 8705 | unsigned long offset = pfn % pageblock_nr_pages; |
47118af0 | 8706 | |
1a9f2191 QC |
8707 | if (is_migrate_cma_page(page)) { |
8708 | /* | |
8709 | * CMA allocations (alloc_contig_range) really need to mark | |
8710 | * isolate CMA pageblocks even when they are not movable in fact | |
8711 | * so consider them movable here. | |
8712 | */ | |
8713 | if (is_migrate_cma(migratetype)) | |
4a55c047 | 8714 | return NULL; |
1a9f2191 | 8715 | |
3d680bdf | 8716 | return page; |
1a9f2191 | 8717 | } |
4da2ce25 | 8718 | |
6a654e36 | 8719 | for (; iter < pageblock_nr_pages - offset; iter++) { |
fe4c86c9 | 8720 | if (!pfn_valid_within(pfn + iter)) |
49ac8255 | 8721 | continue; |
29723fcc | 8722 | |
fe4c86c9 | 8723 | page = pfn_to_page(pfn + iter); |
c8721bbb | 8724 | |
c9c510dc DH |
8725 | /* |
8726 | * Both, bootmem allocations and memory holes are marked | |
8727 | * PG_reserved and are unmovable. We can even have unmovable | |
8728 | * allocations inside ZONE_MOVABLE, for example when | |
8729 | * specifying "movablecore". | |
8730 | */ | |
d7ab3672 | 8731 | if (PageReserved(page)) |
3d680bdf | 8732 | return page; |
d7ab3672 | 8733 | |
9d789999 MH |
8734 | /* |
8735 | * If the zone is movable and we have ruled out all reserved | |
8736 | * pages then it should be reasonably safe to assume the rest | |
8737 | * is movable. | |
8738 | */ | |
8739 | if (zone_idx(zone) == ZONE_MOVABLE) | |
8740 | continue; | |
8741 | ||
c8721bbb NH |
8742 | /* |
8743 | * Hugepages are not in LRU lists, but they're movable. | |
1da2f328 | 8744 | * THPs are on the LRU, but need to be counted as #small pages. |
8bb4e7a2 | 8745 | * We need not scan over tail pages because we don't |
c8721bbb NH |
8746 | * handle each tail page individually in migration. |
8747 | */ | |
1da2f328 | 8748 | if (PageHuge(page) || PageTransCompound(page)) { |
17e2e7d7 OS |
8749 | struct page *head = compound_head(page); |
8750 | unsigned int skip_pages; | |
464c7ffb | 8751 | |
1da2f328 RR |
8752 | if (PageHuge(page)) { |
8753 | if (!hugepage_migration_supported(page_hstate(head))) | |
8754 | return page; | |
8755 | } else if (!PageLRU(head) && !__PageMovable(head)) { | |
3d680bdf | 8756 | return page; |
1da2f328 | 8757 | } |
464c7ffb | 8758 | |
d8c6546b | 8759 | skip_pages = compound_nr(head) - (page - head); |
17e2e7d7 | 8760 | iter += skip_pages - 1; |
c8721bbb NH |
8761 | continue; |
8762 | } | |
8763 | ||
97d255c8 MK |
8764 | /* |
8765 | * We can't use page_count without pin a page | |
8766 | * because another CPU can free compound page. | |
8767 | * This check already skips compound tails of THP | |
0139aa7b | 8768 | * because their page->_refcount is zero at all time. |
97d255c8 | 8769 | */ |
fe896d18 | 8770 | if (!page_ref_count(page)) { |
49ac8255 | 8771 | if (PageBuddy(page)) |
ab130f91 | 8772 | iter += (1 << buddy_order(page)) - 1; |
49ac8255 KH |
8773 | continue; |
8774 | } | |
97d255c8 | 8775 | |
b023f468 WC |
8776 | /* |
8777 | * The HWPoisoned page may be not in buddy system, and | |
8778 | * page_count() is not 0. | |
8779 | */ | |
756d25be | 8780 | if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
b023f468 WC |
8781 | continue; |
8782 | ||
aa218795 DH |
8783 | /* |
8784 | * We treat all PageOffline() pages as movable when offlining | |
8785 | * to give drivers a chance to decrement their reference count | |
8786 | * in MEM_GOING_OFFLINE in order to indicate that these pages | |
8787 | * can be offlined as there are no direct references anymore. | |
8788 | * For actually unmovable PageOffline() where the driver does | |
8789 | * not support this, we will fail later when trying to actually | |
8790 | * move these pages that still have a reference count > 0. | |
8791 | * (false negatives in this function only) | |
8792 | */ | |
8793 | if ((flags & MEMORY_OFFLINE) && PageOffline(page)) | |
8794 | continue; | |
8795 | ||
fe4c86c9 | 8796 | if (__PageMovable(page) || PageLRU(page)) |
0efadf48 YX |
8797 | continue; |
8798 | ||
49ac8255 | 8799 | /* |
6b4f7799 JW |
8800 | * If there are RECLAIMABLE pages, we need to check |
8801 | * it. But now, memory offline itself doesn't call | |
8802 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 | 8803 | */ |
3d680bdf | 8804 | return page; |
49ac8255 | 8805 | } |
4a55c047 | 8806 | return NULL; |
49ac8255 KH |
8807 | } |
8808 | ||
8df995f6 | 8809 | #ifdef CONFIG_CONTIG_ALLOC |
041d3a8c MN |
8810 | static unsigned long pfn_max_align_down(unsigned long pfn) |
8811 | { | |
8812 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8813 | pageblock_nr_pages) - 1); | |
8814 | } | |
8815 | ||
8816 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
8817 | { | |
8818 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8819 | pageblock_nr_pages)); | |
8820 | } | |
8821 | ||
a1394bdd MK |
8822 | #if defined(CONFIG_DYNAMIC_DEBUG) || \ |
8823 | (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) | |
8824 | /* Usage: See admin-guide/dynamic-debug-howto.rst */ | |
8825 | static void alloc_contig_dump_pages(struct list_head *page_list) | |
8826 | { | |
8827 | DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure"); | |
8828 | ||
8829 | if (DYNAMIC_DEBUG_BRANCH(descriptor)) { | |
8830 | struct page *page; | |
8831 | ||
8832 | dump_stack(); | |
8833 | list_for_each_entry(page, page_list, lru) | |
8834 | dump_page(page, "migration failure"); | |
8835 | } | |
8836 | } | |
8837 | #else | |
8838 | static inline void alloc_contig_dump_pages(struct list_head *page_list) | |
8839 | { | |
8840 | } | |
8841 | #endif | |
8842 | ||
041d3a8c | 8843 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
8844 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
8845 | unsigned long start, unsigned long end) | |
041d3a8c MN |
8846 | { |
8847 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 8848 | unsigned int nr_reclaimed; |
041d3a8c MN |
8849 | unsigned long pfn = start; |
8850 | unsigned int tries = 0; | |
8851 | int ret = 0; | |
8b94e0b8 JK |
8852 | struct migration_target_control mtc = { |
8853 | .nid = zone_to_nid(cc->zone), | |
8854 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
8855 | }; | |
041d3a8c | 8856 | |
361a2a22 | 8857 | lru_cache_disable(); |
041d3a8c | 8858 | |
bb13ffeb | 8859 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
8860 | if (fatal_signal_pending(current)) { |
8861 | ret = -EINTR; | |
8862 | break; | |
8863 | } | |
8864 | ||
bb13ffeb MG |
8865 | if (list_empty(&cc->migratepages)) { |
8866 | cc->nr_migratepages = 0; | |
c2ad7a1f OS |
8867 | ret = isolate_migratepages_range(cc, pfn, end); |
8868 | if (ret && ret != -EAGAIN) | |
041d3a8c | 8869 | break; |
c2ad7a1f | 8870 | pfn = cc->migrate_pfn; |
041d3a8c MN |
8871 | tries = 0; |
8872 | } else if (++tries == 5) { | |
c8e28b47 | 8873 | ret = -EBUSY; |
041d3a8c MN |
8874 | break; |
8875 | } | |
8876 | ||
beb51eaa MK |
8877 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
8878 | &cc->migratepages); | |
8879 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 8880 | |
8b94e0b8 JK |
8881 | ret = migrate_pages(&cc->migratepages, alloc_migration_target, |
8882 | NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE); | |
c8e28b47 OS |
8883 | |
8884 | /* | |
8885 | * On -ENOMEM, migrate_pages() bails out right away. It is pointless | |
8886 | * to retry again over this error, so do the same here. | |
8887 | */ | |
8888 | if (ret == -ENOMEM) | |
8889 | break; | |
041d3a8c | 8890 | } |
d479960e | 8891 | |
361a2a22 | 8892 | lru_cache_enable(); |
2a6f5124 | 8893 | if (ret < 0) { |
151e084a MK |
8894 | if (ret == -EBUSY) |
8895 | alloc_contig_dump_pages(&cc->migratepages); | |
2a6f5124 SP |
8896 | putback_movable_pages(&cc->migratepages); |
8897 | return ret; | |
8898 | } | |
8899 | return 0; | |
041d3a8c MN |
8900 | } |
8901 | ||
8902 | /** | |
8903 | * alloc_contig_range() -- tries to allocate given range of pages | |
8904 | * @start: start PFN to allocate | |
8905 | * @end: one-past-the-last PFN to allocate | |
f0953a1b | 8906 | * @migratetype: migratetype of the underlying pageblocks (either |
0815f3d8 MN |
8907 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
8908 | * in range must have the same migratetype and it must | |
8909 | * be either of the two. | |
ca96b625 | 8910 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
8911 | * |
8912 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 8913 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 8914 | * |
2c7452a0 MK |
8915 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
8916 | * pageblocks in the range. Once isolated, the pageblocks should not | |
8917 | * be modified by others. | |
041d3a8c | 8918 | * |
a862f68a | 8919 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
8920 | * pages which PFN is in [start, end) are allocated for the caller and |
8921 | * need to be freed with free_contig_range(). | |
8922 | */ | |
0815f3d8 | 8923 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 8924 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 8925 | { |
041d3a8c | 8926 | unsigned long outer_start, outer_end; |
d00181b9 KS |
8927 | unsigned int order; |
8928 | int ret = 0; | |
041d3a8c | 8929 | |
bb13ffeb MG |
8930 | struct compact_control cc = { |
8931 | .nr_migratepages = 0, | |
8932 | .order = -1, | |
8933 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 8934 | .mode = MIGRATE_SYNC, |
bb13ffeb | 8935 | .ignore_skip_hint = true, |
2583d671 | 8936 | .no_set_skip_hint = true, |
7dea19f9 | 8937 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 8938 | .alloc_contig = true, |
bb13ffeb MG |
8939 | }; |
8940 | INIT_LIST_HEAD(&cc.migratepages); | |
8941 | ||
041d3a8c MN |
8942 | /* |
8943 | * What we do here is we mark all pageblocks in range as | |
8944 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
8945 | * have different sizes, and due to the way page allocator | |
8946 | * work, we align the range to biggest of the two pages so | |
8947 | * that page allocator won't try to merge buddies from | |
8948 | * different pageblocks and change MIGRATE_ISOLATE to some | |
8949 | * other migration type. | |
8950 | * | |
8951 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
8952 | * migrate the pages from an unaligned range (ie. pages that | |
8953 | * we are interested in). This will put all the pages in | |
8954 | * range back to page allocator as MIGRATE_ISOLATE. | |
8955 | * | |
8956 | * When this is done, we take the pages in range from page | |
8957 | * allocator removing them from the buddy system. This way | |
8958 | * page allocator will never consider using them. | |
8959 | * | |
8960 | * This lets us mark the pageblocks back as | |
8961 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
8962 | * aligned range but not in the unaligned, original range are | |
8963 | * put back to page allocator so that buddy can use them. | |
8964 | */ | |
8965 | ||
8966 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
d381c547 | 8967 | pfn_max_align_up(end), migratetype, 0); |
3fa0c7c7 | 8968 | if (ret) |
86a595f9 | 8969 | return ret; |
041d3a8c | 8970 | |
7612921f VB |
8971 | drain_all_pages(cc.zone); |
8972 | ||
8ef5849f JK |
8973 | /* |
8974 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
8975 | * So, just fall through. test_pages_isolated() has a tracepoint |
8976 | * which will report the busy page. | |
8977 | * | |
8978 | * It is possible that busy pages could become available before | |
8979 | * the call to test_pages_isolated, and the range will actually be | |
8980 | * allocated. So, if we fall through be sure to clear ret so that | |
8981 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 8982 | */ |
bb13ffeb | 8983 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 8984 | if (ret && ret != -EBUSY) |
041d3a8c | 8985 | goto done; |
68d68ff6 | 8986 | ret = 0; |
041d3a8c MN |
8987 | |
8988 | /* | |
8989 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
8990 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
8991 | * more, all pages in [start, end) are free in page allocator. | |
8992 | * What we are going to do is to allocate all pages from | |
8993 | * [start, end) (that is remove them from page allocator). | |
8994 | * | |
8995 | * The only problem is that pages at the beginning and at the | |
8996 | * end of interesting range may be not aligned with pages that | |
8997 | * page allocator holds, ie. they can be part of higher order | |
8998 | * pages. Because of this, we reserve the bigger range and | |
8999 | * once this is done free the pages we are not interested in. | |
9000 | * | |
9001 | * We don't have to hold zone->lock here because the pages are | |
9002 | * isolated thus they won't get removed from buddy. | |
9003 | */ | |
9004 | ||
041d3a8c MN |
9005 | order = 0; |
9006 | outer_start = start; | |
9007 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
9008 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
9009 | outer_start = start; |
9010 | break; | |
041d3a8c MN |
9011 | } |
9012 | outer_start &= ~0UL << order; | |
9013 | } | |
9014 | ||
8ef5849f | 9015 | if (outer_start != start) { |
ab130f91 | 9016 | order = buddy_order(pfn_to_page(outer_start)); |
8ef5849f JK |
9017 | |
9018 | /* | |
9019 | * outer_start page could be small order buddy page and | |
9020 | * it doesn't include start page. Adjust outer_start | |
9021 | * in this case to report failed page properly | |
9022 | * on tracepoint in test_pages_isolated() | |
9023 | */ | |
9024 | if (outer_start + (1UL << order) <= start) | |
9025 | outer_start = start; | |
9026 | } | |
9027 | ||
041d3a8c | 9028 | /* Make sure the range is really isolated. */ |
756d25be | 9029 | if (test_pages_isolated(outer_start, end, 0)) { |
041d3a8c MN |
9030 | ret = -EBUSY; |
9031 | goto done; | |
9032 | } | |
9033 | ||
49f223a9 | 9034 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 9035 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
9036 | if (!outer_end) { |
9037 | ret = -EBUSY; | |
9038 | goto done; | |
9039 | } | |
9040 | ||
9041 | /* Free head and tail (if any) */ | |
9042 | if (start != outer_start) | |
9043 | free_contig_range(outer_start, start - outer_start); | |
9044 | if (end != outer_end) | |
9045 | free_contig_range(end, outer_end - end); | |
9046 | ||
9047 | done: | |
9048 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 9049 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
9050 | return ret; |
9051 | } | |
255f5985 | 9052 | EXPORT_SYMBOL(alloc_contig_range); |
5e27a2df AK |
9053 | |
9054 | static int __alloc_contig_pages(unsigned long start_pfn, | |
9055 | unsigned long nr_pages, gfp_t gfp_mask) | |
9056 | { | |
9057 | unsigned long end_pfn = start_pfn + nr_pages; | |
9058 | ||
9059 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
9060 | gfp_mask); | |
9061 | } | |
9062 | ||
9063 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
9064 | unsigned long nr_pages) | |
9065 | { | |
9066 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
9067 | struct page *page; | |
9068 | ||
9069 | for (i = start_pfn; i < end_pfn; i++) { | |
9070 | page = pfn_to_online_page(i); | |
9071 | if (!page) | |
9072 | return false; | |
9073 | ||
9074 | if (page_zone(page) != z) | |
9075 | return false; | |
9076 | ||
9077 | if (PageReserved(page)) | |
9078 | return false; | |
5e27a2df AK |
9079 | } |
9080 | return true; | |
9081 | } | |
9082 | ||
9083 | static bool zone_spans_last_pfn(const struct zone *zone, | |
9084 | unsigned long start_pfn, unsigned long nr_pages) | |
9085 | { | |
9086 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
9087 | ||
9088 | return zone_spans_pfn(zone, last_pfn); | |
9089 | } | |
9090 | ||
9091 | /** | |
9092 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
9093 | * @nr_pages: Number of contiguous pages to allocate | |
9094 | * @gfp_mask: GFP mask to limit search and used during compaction | |
9095 | * @nid: Target node | |
9096 | * @nodemask: Mask for other possible nodes | |
9097 | * | |
9098 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
9099 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
9100 | * tried for allocation with alloc_contig_range(). This routine is intended | |
9101 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
9102 | * | |
9103 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
9104 | * power of two then the alignment is guaranteed to be to the given nr_pages | |
9105 | * (e.g. 1GB request would be aligned to 1GB). | |
9106 | * | |
9107 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
9108 | * __free_page() on each allocated page. | |
9109 | * | |
9110 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
9111 | */ | |
9112 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
9113 | int nid, nodemask_t *nodemask) | |
9114 | { | |
9115 | unsigned long ret, pfn, flags; | |
9116 | struct zonelist *zonelist; | |
9117 | struct zone *zone; | |
9118 | struct zoneref *z; | |
9119 | ||
9120 | zonelist = node_zonelist(nid, gfp_mask); | |
9121 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
9122 | gfp_zone(gfp_mask), nodemask) { | |
9123 | spin_lock_irqsave(&zone->lock, flags); | |
9124 | ||
9125 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
9126 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
9127 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
9128 | /* | |
9129 | * We release the zone lock here because | |
9130 | * alloc_contig_range() will also lock the zone | |
9131 | * at some point. If there's an allocation | |
9132 | * spinning on this lock, it may win the race | |
9133 | * and cause alloc_contig_range() to fail... | |
9134 | */ | |
9135 | spin_unlock_irqrestore(&zone->lock, flags); | |
9136 | ret = __alloc_contig_pages(pfn, nr_pages, | |
9137 | gfp_mask); | |
9138 | if (!ret) | |
9139 | return pfn_to_page(pfn); | |
9140 | spin_lock_irqsave(&zone->lock, flags); | |
9141 | } | |
9142 | pfn += nr_pages; | |
9143 | } | |
9144 | spin_unlock_irqrestore(&zone->lock, flags); | |
9145 | } | |
9146 | return NULL; | |
9147 | } | |
4eb0716e | 9148 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 9149 | |
78fa5150 | 9150 | void free_contig_range(unsigned long pfn, unsigned long nr_pages) |
041d3a8c | 9151 | { |
78fa5150 | 9152 | unsigned long count = 0; |
bcc2b02f MS |
9153 | |
9154 | for (; nr_pages--; pfn++) { | |
9155 | struct page *page = pfn_to_page(pfn); | |
9156 | ||
9157 | count += page_count(page) != 1; | |
9158 | __free_page(page); | |
9159 | } | |
78fa5150 | 9160 | WARN(count != 0, "%lu pages are still in use!\n", count); |
041d3a8c | 9161 | } |
255f5985 | 9162 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 9163 | |
0a647f38 CS |
9164 | /* |
9165 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
f0953a1b | 9166 | * page high values need to be recalculated. |
0a647f38 | 9167 | */ |
04f8cfea | 9168 | void zone_pcp_update(struct zone *zone, int cpu_online) |
4ed7e022 | 9169 | { |
c8e251fa | 9170 | mutex_lock(&pcp_batch_high_lock); |
04f8cfea | 9171 | zone_set_pageset_high_and_batch(zone, cpu_online); |
c8e251fa | 9172 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 | 9173 | } |
4ed7e022 | 9174 | |
ec6e8c7e VB |
9175 | /* |
9176 | * Effectively disable pcplists for the zone by setting the high limit to 0 | |
9177 | * and draining all cpus. A concurrent page freeing on another CPU that's about | |
9178 | * to put the page on pcplist will either finish before the drain and the page | |
9179 | * will be drained, or observe the new high limit and skip the pcplist. | |
9180 | * | |
9181 | * Must be paired with a call to zone_pcp_enable(). | |
9182 | */ | |
9183 | void zone_pcp_disable(struct zone *zone) | |
9184 | { | |
9185 | mutex_lock(&pcp_batch_high_lock); | |
9186 | __zone_set_pageset_high_and_batch(zone, 0, 1); | |
9187 | __drain_all_pages(zone, true); | |
9188 | } | |
9189 | ||
9190 | void zone_pcp_enable(struct zone *zone) | |
9191 | { | |
9192 | __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch); | |
9193 | mutex_unlock(&pcp_batch_high_lock); | |
9194 | } | |
9195 | ||
340175b7 JL |
9196 | void zone_pcp_reset(struct zone *zone) |
9197 | { | |
5a883813 | 9198 | int cpu; |
28f836b6 | 9199 | struct per_cpu_zonestat *pzstats; |
340175b7 | 9200 | |
28f836b6 | 9201 | if (zone->per_cpu_pageset != &boot_pageset) { |
5a883813 | 9202 | for_each_online_cpu(cpu) { |
28f836b6 MG |
9203 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); |
9204 | drain_zonestat(zone, pzstats); | |
5a883813 | 9205 | } |
28f836b6 MG |
9206 | free_percpu(zone->per_cpu_pageset); |
9207 | free_percpu(zone->per_cpu_zonestats); | |
9208 | zone->per_cpu_pageset = &boot_pageset; | |
9209 | zone->per_cpu_zonestats = &boot_zonestats; | |
340175b7 | 9210 | } |
340175b7 JL |
9211 | } |
9212 | ||
6dcd73d7 | 9213 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 9214 | /* |
257bea71 DH |
9215 | * All pages in the range must be in a single zone, must not contain holes, |
9216 | * must span full sections, and must be isolated before calling this function. | |
0c0e6195 | 9217 | */ |
257bea71 | 9218 | void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 9219 | { |
257bea71 | 9220 | unsigned long pfn = start_pfn; |
0c0e6195 KH |
9221 | struct page *page; |
9222 | struct zone *zone; | |
0ee5f4f3 | 9223 | unsigned int order; |
0c0e6195 | 9224 | unsigned long flags; |
5557c766 | 9225 | |
2d070eab | 9226 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
9227 | zone = page_zone(pfn_to_page(pfn)); |
9228 | spin_lock_irqsave(&zone->lock, flags); | |
0c0e6195 | 9229 | while (pfn < end_pfn) { |
0c0e6195 | 9230 | page = pfn_to_page(pfn); |
b023f468 WC |
9231 | /* |
9232 | * The HWPoisoned page may be not in buddy system, and | |
9233 | * page_count() is not 0. | |
9234 | */ | |
9235 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
9236 | pfn++; | |
b023f468 WC |
9237 | continue; |
9238 | } | |
aa218795 DH |
9239 | /* |
9240 | * At this point all remaining PageOffline() pages have a | |
9241 | * reference count of 0 and can simply be skipped. | |
9242 | */ | |
9243 | if (PageOffline(page)) { | |
9244 | BUG_ON(page_count(page)); | |
9245 | BUG_ON(PageBuddy(page)); | |
9246 | pfn++; | |
aa218795 DH |
9247 | continue; |
9248 | } | |
b023f468 | 9249 | |
0c0e6195 KH |
9250 | BUG_ON(page_count(page)); |
9251 | BUG_ON(!PageBuddy(page)); | |
ab130f91 | 9252 | order = buddy_order(page); |
6ab01363 | 9253 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
9254 | pfn += (1 << order); |
9255 | } | |
9256 | spin_unlock_irqrestore(&zone->lock, flags); | |
9257 | } | |
9258 | #endif | |
8d22ba1b | 9259 | |
8d22ba1b WF |
9260 | bool is_free_buddy_page(struct page *page) |
9261 | { | |
9262 | struct zone *zone = page_zone(page); | |
9263 | unsigned long pfn = page_to_pfn(page); | |
9264 | unsigned long flags; | |
7aeb09f9 | 9265 | unsigned int order; |
8d22ba1b WF |
9266 | |
9267 | spin_lock_irqsave(&zone->lock, flags); | |
9268 | for (order = 0; order < MAX_ORDER; order++) { | |
9269 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
9270 | ||
ab130f91 | 9271 | if (PageBuddy(page_head) && buddy_order(page_head) >= order) |
8d22ba1b WF |
9272 | break; |
9273 | } | |
9274 | spin_unlock_irqrestore(&zone->lock, flags); | |
9275 | ||
9276 | return order < MAX_ORDER; | |
9277 | } | |
d4ae9916 NH |
9278 | |
9279 | #ifdef CONFIG_MEMORY_FAILURE | |
9280 | /* | |
06be6ff3 OS |
9281 | * Break down a higher-order page in sub-pages, and keep our target out of |
9282 | * buddy allocator. | |
d4ae9916 | 9283 | */ |
06be6ff3 OS |
9284 | static void break_down_buddy_pages(struct zone *zone, struct page *page, |
9285 | struct page *target, int low, int high, | |
9286 | int migratetype) | |
9287 | { | |
9288 | unsigned long size = 1 << high; | |
9289 | struct page *current_buddy, *next_page; | |
9290 | ||
9291 | while (high > low) { | |
9292 | high--; | |
9293 | size >>= 1; | |
9294 | ||
9295 | if (target >= &page[size]) { | |
9296 | next_page = page + size; | |
9297 | current_buddy = page; | |
9298 | } else { | |
9299 | next_page = page; | |
9300 | current_buddy = page + size; | |
9301 | } | |
9302 | ||
9303 | if (set_page_guard(zone, current_buddy, high, migratetype)) | |
9304 | continue; | |
9305 | ||
9306 | if (current_buddy != target) { | |
9307 | add_to_free_list(current_buddy, zone, high, migratetype); | |
ab130f91 | 9308 | set_buddy_order(current_buddy, high); |
06be6ff3 OS |
9309 | page = next_page; |
9310 | } | |
9311 | } | |
9312 | } | |
9313 | ||
9314 | /* | |
9315 | * Take a page that will be marked as poisoned off the buddy allocator. | |
9316 | */ | |
9317 | bool take_page_off_buddy(struct page *page) | |
d4ae9916 NH |
9318 | { |
9319 | struct zone *zone = page_zone(page); | |
9320 | unsigned long pfn = page_to_pfn(page); | |
9321 | unsigned long flags; | |
9322 | unsigned int order; | |
06be6ff3 | 9323 | bool ret = false; |
d4ae9916 NH |
9324 | |
9325 | spin_lock_irqsave(&zone->lock, flags); | |
9326 | for (order = 0; order < MAX_ORDER; order++) { | |
9327 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
ab130f91 | 9328 | int page_order = buddy_order(page_head); |
d4ae9916 | 9329 | |
ab130f91 | 9330 | if (PageBuddy(page_head) && page_order >= order) { |
06be6ff3 OS |
9331 | unsigned long pfn_head = page_to_pfn(page_head); |
9332 | int migratetype = get_pfnblock_migratetype(page_head, | |
9333 | pfn_head); | |
9334 | ||
ab130f91 | 9335 | del_page_from_free_list(page_head, zone, page_order); |
06be6ff3 | 9336 | break_down_buddy_pages(zone, page_head, page, 0, |
ab130f91 | 9337 | page_order, migratetype); |
bac9c6fa DH |
9338 | if (!is_migrate_isolate(migratetype)) |
9339 | __mod_zone_freepage_state(zone, -1, migratetype); | |
06be6ff3 | 9340 | ret = true; |
d4ae9916 NH |
9341 | break; |
9342 | } | |
06be6ff3 OS |
9343 | if (page_count(page_head) > 0) |
9344 | break; | |
d4ae9916 NH |
9345 | } |
9346 | spin_unlock_irqrestore(&zone->lock, flags); | |
06be6ff3 | 9347 | return ret; |
d4ae9916 NH |
9348 | } |
9349 | #endif |