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); | |
7cd2b0a3 | 123 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 124 | |
72812019 LS |
125 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
126 | DEFINE_PER_CPU(int, numa_node); | |
127 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
128 | #endif | |
129 | ||
4518085e KW |
130 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
131 | ||
7aac7898 LS |
132 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
133 | /* | |
134 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
135 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
136 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
137 | * defined in <linux/topology.h>. | |
138 | */ | |
139 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
140 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
141 | #endif | |
142 | ||
bd233f53 | 143 | /* work_structs for global per-cpu drains */ |
d9367bd0 WY |
144 | struct pcpu_drain { |
145 | struct zone *zone; | |
146 | struct work_struct work; | |
147 | }; | |
8b885f53 JY |
148 | static DEFINE_MUTEX(pcpu_drain_mutex); |
149 | static DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain); | |
bd233f53 | 150 | |
38addce8 | 151 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 152 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
153 | EXPORT_SYMBOL(latent_entropy); |
154 | #endif | |
155 | ||
1da177e4 | 156 | /* |
13808910 | 157 | * Array of node states. |
1da177e4 | 158 | */ |
13808910 CL |
159 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
160 | [N_POSSIBLE] = NODE_MASK_ALL, | |
161 | [N_ONLINE] = { { [0] = 1UL } }, | |
162 | #ifndef CONFIG_NUMA | |
163 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
164 | #ifdef CONFIG_HIGHMEM | |
165 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 166 | #endif |
20b2f52b | 167 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
168 | [N_CPU] = { { [0] = 1UL } }, |
169 | #endif /* NUMA */ | |
170 | }; | |
171 | EXPORT_SYMBOL(node_states); | |
172 | ||
ca79b0c2 AK |
173 | atomic_long_t _totalram_pages __read_mostly; |
174 | EXPORT_SYMBOL(_totalram_pages); | |
cb45b0e9 | 175 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 176 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 177 | |
1b76b02f | 178 | int percpu_pagelist_fraction; |
dcce284a | 179 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
51cba1eb | 180 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc); |
6471384a AP |
181 | EXPORT_SYMBOL(init_on_alloc); |
182 | ||
51cba1eb | 183 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free); |
6471384a AP |
184 | EXPORT_SYMBOL(init_on_free); |
185 | ||
04013513 VB |
186 | static bool _init_on_alloc_enabled_early __read_mostly |
187 | = IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON); | |
6471384a AP |
188 | static int __init early_init_on_alloc(char *buf) |
189 | { | |
6471384a | 190 | |
04013513 | 191 | return kstrtobool(buf, &_init_on_alloc_enabled_early); |
6471384a AP |
192 | } |
193 | early_param("init_on_alloc", early_init_on_alloc); | |
194 | ||
04013513 VB |
195 | static bool _init_on_free_enabled_early __read_mostly |
196 | = IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON); | |
6471384a AP |
197 | static int __init early_init_on_free(char *buf) |
198 | { | |
04013513 | 199 | return kstrtobool(buf, &_init_on_free_enabled_early); |
6471384a AP |
200 | } |
201 | early_param("init_on_free", early_init_on_free); | |
1da177e4 | 202 | |
bb14c2c7 VB |
203 | /* |
204 | * A cached value of the page's pageblock's migratetype, used when the page is | |
205 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
206 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
207 | * Also the migratetype set in the page does not necessarily match the pcplist | |
208 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
209 | * other index - this ensures that it will be put on the correct CMA freelist. | |
210 | */ | |
211 | static inline int get_pcppage_migratetype(struct page *page) | |
212 | { | |
213 | return page->index; | |
214 | } | |
215 | ||
216 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
217 | { | |
218 | page->index = migratetype; | |
219 | } | |
220 | ||
452aa699 RW |
221 | #ifdef CONFIG_PM_SLEEP |
222 | /* | |
223 | * The following functions are used by the suspend/hibernate code to temporarily | |
224 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
225 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
226 | * they should always be called with system_transition_mutex held |
227 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
228 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
229 | * with that modification). | |
452aa699 | 230 | */ |
c9e664f1 RW |
231 | |
232 | static gfp_t saved_gfp_mask; | |
233 | ||
234 | void pm_restore_gfp_mask(void) | |
452aa699 | 235 | { |
55f2503c | 236 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
237 | if (saved_gfp_mask) { |
238 | gfp_allowed_mask = saved_gfp_mask; | |
239 | saved_gfp_mask = 0; | |
240 | } | |
452aa699 RW |
241 | } |
242 | ||
c9e664f1 | 243 | void pm_restrict_gfp_mask(void) |
452aa699 | 244 | { |
55f2503c | 245 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
246 | WARN_ON(saved_gfp_mask); |
247 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 248 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 249 | } |
f90ac398 MG |
250 | |
251 | bool pm_suspended_storage(void) | |
252 | { | |
d0164adc | 253 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
254 | return false; |
255 | return true; | |
256 | } | |
452aa699 RW |
257 | #endif /* CONFIG_PM_SLEEP */ |
258 | ||
d9c23400 | 259 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 260 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
261 | #endif |
262 | ||
7fef431b DH |
263 | static void __free_pages_ok(struct page *page, unsigned int order, |
264 | fpi_t fpi_flags); | |
a226f6c8 | 265 | |
1da177e4 LT |
266 | /* |
267 | * results with 256, 32 in the lowmem_reserve sysctl: | |
268 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
269 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
270 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
271 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 272 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
273 | * |
274 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
275 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 276 | */ |
d3cda233 | 277 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 278 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 279 | [ZONE_DMA] = 256, |
4b51d669 | 280 | #endif |
fb0e7942 | 281 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 282 | [ZONE_DMA32] = 256, |
fb0e7942 | 283 | #endif |
d3cda233 | 284 | [ZONE_NORMAL] = 32, |
e53ef38d | 285 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 286 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 287 | #endif |
d3cda233 | 288 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 289 | }; |
1da177e4 | 290 | |
15ad7cdc | 291 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 292 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 293 | "DMA", |
4b51d669 | 294 | #endif |
fb0e7942 | 295 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 296 | "DMA32", |
fb0e7942 | 297 | #endif |
2f1b6248 | 298 | "Normal", |
e53ef38d | 299 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 300 | "HighMem", |
e53ef38d | 301 | #endif |
2a1e274a | 302 | "Movable", |
033fbae9 DW |
303 | #ifdef CONFIG_ZONE_DEVICE |
304 | "Device", | |
305 | #endif | |
2f1b6248 CL |
306 | }; |
307 | ||
c999fbd3 | 308 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
309 | "Unmovable", |
310 | "Movable", | |
311 | "Reclaimable", | |
312 | "HighAtomic", | |
313 | #ifdef CONFIG_CMA | |
314 | "CMA", | |
315 | #endif | |
316 | #ifdef CONFIG_MEMORY_ISOLATION | |
317 | "Isolate", | |
318 | #endif | |
319 | }; | |
320 | ||
ae70eddd AK |
321 | compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = { |
322 | [NULL_COMPOUND_DTOR] = NULL, | |
323 | [COMPOUND_PAGE_DTOR] = free_compound_page, | |
f1e61557 | 324 | #ifdef CONFIG_HUGETLB_PAGE |
ae70eddd | 325 | [HUGETLB_PAGE_DTOR] = free_huge_page, |
f1e61557 | 326 | #endif |
9a982250 | 327 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
ae70eddd | 328 | [TRANSHUGE_PAGE_DTOR] = free_transhuge_page, |
9a982250 | 329 | #endif |
f1e61557 KS |
330 | }; |
331 | ||
1da177e4 | 332 | int min_free_kbytes = 1024; |
42aa83cb | 333 | int user_min_free_kbytes = -1; |
24512228 MG |
334 | #ifdef CONFIG_DISCONTIGMEM |
335 | /* | |
336 | * DiscontigMem defines memory ranges as separate pg_data_t even if the ranges | |
337 | * are not on separate NUMA nodes. Functionally this works but with | |
338 | * watermark_boost_factor, it can reclaim prematurely as the ranges can be | |
339 | * quite small. By default, do not boost watermarks on discontigmem as in | |
340 | * many cases very high-order allocations like THP are likely to be | |
341 | * unsupported and the premature reclaim offsets the advantage of long-term | |
342 | * fragmentation avoidance. | |
343 | */ | |
344 | int watermark_boost_factor __read_mostly; | |
345 | #else | |
1c30844d | 346 | int watermark_boost_factor __read_mostly = 15000; |
24512228 | 347 | #endif |
795ae7a0 | 348 | int watermark_scale_factor = 10; |
1da177e4 | 349 | |
bbe5d993 OS |
350 | static unsigned long nr_kernel_pages __initdata; |
351 | static unsigned long nr_all_pages __initdata; | |
352 | static unsigned long dma_reserve __initdata; | |
1da177e4 | 353 | |
bbe5d993 OS |
354 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata; |
355 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata; | |
7f16f91f | 356 | static unsigned long required_kernelcore __initdata; |
a5c6d650 | 357 | static unsigned long required_kernelcore_percent __initdata; |
7f16f91f | 358 | static unsigned long required_movablecore __initdata; |
a5c6d650 | 359 | static unsigned long required_movablecore_percent __initdata; |
bbe5d993 | 360 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata; |
7f16f91f | 361 | static bool mirrored_kernelcore __meminitdata; |
0ee332c1 TH |
362 | |
363 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
364 | int movable_zone; | |
365 | EXPORT_SYMBOL(movable_zone); | |
c713216d | 366 | |
418508c1 | 367 | #if MAX_NUMNODES > 1 |
b9726c26 | 368 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 369 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 370 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 371 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
372 | #endif |
373 | ||
9ef9acb0 MG |
374 | int page_group_by_mobility_disabled __read_mostly; |
375 | ||
3a80a7fa | 376 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
377 | /* |
378 | * During boot we initialize deferred pages on-demand, as needed, but once | |
379 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
380 | * and we can permanently disable that path. | |
381 | */ | |
382 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
383 | ||
384 | /* | |
385 | * Calling kasan_free_pages() only after deferred memory initialization | |
386 | * has completed. Poisoning pages during deferred memory init will greatly | |
387 | * lengthen the process and cause problem in large memory systems as the | |
388 | * deferred pages initialization is done with interrupt disabled. | |
389 | * | |
390 | * Assuming that there will be no reference to those newly initialized | |
391 | * pages before they are ever allocated, this should have no effect on | |
392 | * KASAN memory tracking as the poison will be properly inserted at page | |
393 | * allocation time. The only corner case is when pages are allocated by | |
394 | * on-demand allocation and then freed again before the deferred pages | |
395 | * initialization is done, but this is not likely to happen. | |
396 | */ | |
2c335680 | 397 | static inline void kasan_free_nondeferred_pages(struct page *page, int order, |
1bb5eab3 | 398 | bool init, fpi_t fpi_flags) |
3c0c12cc | 399 | { |
2c335680 AK |
400 | if (static_branch_unlikely(&deferred_pages)) |
401 | return; | |
402 | if (!IS_ENABLED(CONFIG_KASAN_GENERIC) && | |
403 | (fpi_flags & FPI_SKIP_KASAN_POISON)) | |
404 | return; | |
1bb5eab3 | 405 | kasan_free_pages(page, order, init); |
3c0c12cc WL |
406 | } |
407 | ||
3a80a7fa | 408 | /* Returns true if the struct page for the pfn is uninitialised */ |
0e1cc95b | 409 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 410 | { |
ef70b6f4 MG |
411 | int nid = early_pfn_to_nid(pfn); |
412 | ||
413 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
414 | return true; |
415 | ||
416 | return false; | |
417 | } | |
418 | ||
419 | /* | |
d3035be4 | 420 | * Returns true when the remaining initialisation should be deferred until |
3a80a7fa MG |
421 | * later in the boot cycle when it can be parallelised. |
422 | */ | |
d3035be4 PT |
423 | static bool __meminit |
424 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn) | |
3a80a7fa | 425 | { |
d3035be4 PT |
426 | static unsigned long prev_end_pfn, nr_initialised; |
427 | ||
428 | /* | |
429 | * prev_end_pfn static that contains the end of previous zone | |
430 | * No need to protect because called very early in boot before smp_init. | |
431 | */ | |
432 | if (prev_end_pfn != end_pfn) { | |
433 | prev_end_pfn = end_pfn; | |
434 | nr_initialised = 0; | |
435 | } | |
436 | ||
3c2c6488 | 437 | /* Always populate low zones for address-constrained allocations */ |
d3035be4 | 438 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid))) |
3a80a7fa | 439 | return false; |
23b68cfa | 440 | |
dc2da7b4 BH |
441 | if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX) |
442 | return true; | |
23b68cfa WY |
443 | /* |
444 | * We start only with one section of pages, more pages are added as | |
445 | * needed until the rest of deferred pages are initialized. | |
446 | */ | |
d3035be4 | 447 | nr_initialised++; |
23b68cfa | 448 | if ((nr_initialised > PAGES_PER_SECTION) && |
d3035be4 PT |
449 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
450 | NODE_DATA(nid)->first_deferred_pfn = pfn; | |
451 | return true; | |
3a80a7fa | 452 | } |
d3035be4 | 453 | return false; |
3a80a7fa MG |
454 | } |
455 | #else | |
2c335680 | 456 | static inline void kasan_free_nondeferred_pages(struct page *page, int order, |
1bb5eab3 | 457 | bool init, fpi_t fpi_flags) |
2c335680 AK |
458 | { |
459 | if (!IS_ENABLED(CONFIG_KASAN_GENERIC) && | |
460 | (fpi_flags & FPI_SKIP_KASAN_POISON)) | |
461 | return; | |
1bb5eab3 | 462 | kasan_free_pages(page, order, init); |
2c335680 | 463 | } |
3c0c12cc | 464 | |
3a80a7fa MG |
465 | static inline bool early_page_uninitialised(unsigned long pfn) |
466 | { | |
467 | return false; | |
468 | } | |
469 | ||
d3035be4 | 470 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn) |
3a80a7fa | 471 | { |
d3035be4 | 472 | return false; |
3a80a7fa MG |
473 | } |
474 | #endif | |
475 | ||
0b423ca2 MG |
476 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
477 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
478 | unsigned long pfn) | |
479 | { | |
480 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 481 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
482 | #else |
483 | return page_zone(page)->pageblock_flags; | |
484 | #endif /* CONFIG_SPARSEMEM */ | |
485 | } | |
486 | ||
487 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
488 | { | |
489 | #ifdef CONFIG_SPARSEMEM | |
490 | pfn &= (PAGES_PER_SECTION-1); | |
0b423ca2 MG |
491 | #else |
492 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
0b423ca2 | 493 | #endif /* CONFIG_SPARSEMEM */ |
399b795b | 494 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
0b423ca2 MG |
495 | } |
496 | ||
535b81e2 WY |
497 | static __always_inline |
498 | unsigned long __get_pfnblock_flags_mask(struct page *page, | |
0b423ca2 | 499 | unsigned long pfn, |
0b423ca2 MG |
500 | unsigned long mask) |
501 | { | |
502 | unsigned long *bitmap; | |
503 | unsigned long bitidx, word_bitidx; | |
504 | unsigned long word; | |
505 | ||
506 | bitmap = get_pageblock_bitmap(page, pfn); | |
507 | bitidx = pfn_to_bitidx(page, pfn); | |
508 | word_bitidx = bitidx / BITS_PER_LONG; | |
509 | bitidx &= (BITS_PER_LONG-1); | |
510 | ||
511 | word = bitmap[word_bitidx]; | |
d93d5ab9 | 512 | return (word >> bitidx) & mask; |
0b423ca2 MG |
513 | } |
514 | ||
a00cda3f MCC |
515 | /** |
516 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
517 | * @page: The page within the block of interest | |
518 | * @pfn: The target page frame number | |
519 | * @mask: mask of bits that the caller is interested in | |
520 | * | |
521 | * Return: pageblock_bits flags | |
522 | */ | |
0b423ca2 | 523 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
0b423ca2 MG |
524 | unsigned long mask) |
525 | { | |
535b81e2 | 526 | return __get_pfnblock_flags_mask(page, pfn, mask); |
0b423ca2 MG |
527 | } |
528 | ||
529 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
530 | { | |
535b81e2 | 531 | return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK); |
0b423ca2 MG |
532 | } |
533 | ||
534 | /** | |
535 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
536 | * @page: The page within the block of interest | |
537 | * @flags: The flags to set | |
538 | * @pfn: The target page frame number | |
0b423ca2 MG |
539 | * @mask: mask of bits that the caller is interested in |
540 | */ | |
541 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
542 | unsigned long pfn, | |
0b423ca2 MG |
543 | unsigned long mask) |
544 | { | |
545 | unsigned long *bitmap; | |
546 | unsigned long bitidx, word_bitidx; | |
547 | unsigned long old_word, word; | |
548 | ||
549 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 550 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
551 | |
552 | bitmap = get_pageblock_bitmap(page, pfn); | |
553 | bitidx = pfn_to_bitidx(page, pfn); | |
554 | word_bitidx = bitidx / BITS_PER_LONG; | |
555 | bitidx &= (BITS_PER_LONG-1); | |
556 | ||
557 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
558 | ||
d93d5ab9 WY |
559 | mask <<= bitidx; |
560 | flags <<= bitidx; | |
0b423ca2 MG |
561 | |
562 | word = READ_ONCE(bitmap[word_bitidx]); | |
563 | for (;;) { | |
564 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
565 | if (word == old_word) | |
566 | break; | |
567 | word = old_word; | |
568 | } | |
569 | } | |
3a80a7fa | 570 | |
ee6f509c | 571 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 572 | { |
5d0f3f72 KM |
573 | if (unlikely(page_group_by_mobility_disabled && |
574 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
575 | migratetype = MIGRATE_UNMOVABLE; |
576 | ||
d93d5ab9 | 577 | set_pfnblock_flags_mask(page, (unsigned long)migratetype, |
535b81e2 | 578 | page_to_pfn(page), MIGRATETYPE_MASK); |
b2a0ac88 MG |
579 | } |
580 | ||
13e7444b | 581 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 582 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 583 | { |
bdc8cb98 DH |
584 | int ret = 0; |
585 | unsigned seq; | |
586 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 587 | unsigned long sp, start_pfn; |
c6a57e19 | 588 | |
bdc8cb98 DH |
589 | do { |
590 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
591 | start_pfn = zone->zone_start_pfn; |
592 | sp = zone->spanned_pages; | |
108bcc96 | 593 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
594 | ret = 1; |
595 | } while (zone_span_seqretry(zone, seq)); | |
596 | ||
b5e6a5a2 | 597 | if (ret) |
613813e8 DH |
598 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
599 | pfn, zone_to_nid(zone), zone->name, | |
600 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 601 | |
bdc8cb98 | 602 | return ret; |
c6a57e19 DH |
603 | } |
604 | ||
605 | static int page_is_consistent(struct zone *zone, struct page *page) | |
606 | { | |
14e07298 | 607 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 608 | return 0; |
1da177e4 | 609 | if (zone != page_zone(page)) |
c6a57e19 DH |
610 | return 0; |
611 | ||
612 | return 1; | |
613 | } | |
614 | /* | |
615 | * Temporary debugging check for pages not lying within a given zone. | |
616 | */ | |
d73d3c9f | 617 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
618 | { |
619 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 620 | return 1; |
c6a57e19 DH |
621 | if (!page_is_consistent(zone, page)) |
622 | return 1; | |
623 | ||
1da177e4 LT |
624 | return 0; |
625 | } | |
13e7444b | 626 | #else |
d73d3c9f | 627 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
628 | { |
629 | return 0; | |
630 | } | |
631 | #endif | |
632 | ||
82a3241a | 633 | static void bad_page(struct page *page, const char *reason) |
1da177e4 | 634 | { |
d936cf9b HD |
635 | static unsigned long resume; |
636 | static unsigned long nr_shown; | |
637 | static unsigned long nr_unshown; | |
638 | ||
639 | /* | |
640 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
641 | * or allow a steady drip of one report per second. | |
642 | */ | |
643 | if (nr_shown == 60) { | |
644 | if (time_before(jiffies, resume)) { | |
645 | nr_unshown++; | |
646 | goto out; | |
647 | } | |
648 | if (nr_unshown) { | |
ff8e8116 | 649 | pr_alert( |
1e9e6365 | 650 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
651 | nr_unshown); |
652 | nr_unshown = 0; | |
653 | } | |
654 | nr_shown = 0; | |
655 | } | |
656 | if (nr_shown++ == 0) | |
657 | resume = jiffies + 60 * HZ; | |
658 | ||
ff8e8116 | 659 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 660 | current->comm, page_to_pfn(page)); |
ff8e8116 | 661 | __dump_page(page, reason); |
4e462112 | 662 | dump_page_owner(page); |
3dc14741 | 663 | |
4f31888c | 664 | print_modules(); |
1da177e4 | 665 | dump_stack(); |
d936cf9b | 666 | out: |
8cc3b392 | 667 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 668 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 669 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
670 | } |
671 | ||
1da177e4 LT |
672 | /* |
673 | * Higher-order pages are called "compound pages". They are structured thusly: | |
674 | * | |
1d798ca3 | 675 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 676 | * |
1d798ca3 KS |
677 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
678 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 679 | * |
1d798ca3 KS |
680 | * The first tail page's ->compound_dtor holds the offset in array of compound |
681 | * page destructors. See compound_page_dtors. | |
1da177e4 | 682 | * |
1d798ca3 | 683 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 684 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 685 | */ |
d98c7a09 | 686 | |
9a982250 | 687 | void free_compound_page(struct page *page) |
d98c7a09 | 688 | { |
7ae88534 | 689 | mem_cgroup_uncharge(page); |
7fef431b | 690 | __free_pages_ok(page, compound_order(page), FPI_NONE); |
d98c7a09 HD |
691 | } |
692 | ||
d00181b9 | 693 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
694 | { |
695 | int i; | |
696 | int nr_pages = 1 << order; | |
697 | ||
18229df5 AW |
698 | __SetPageHead(page); |
699 | for (i = 1; i < nr_pages; i++) { | |
700 | struct page *p = page + i; | |
58a84aa9 | 701 | set_page_count(p, 0); |
1c290f64 | 702 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 703 | set_compound_head(p, page); |
18229df5 | 704 | } |
1378a5ee MWO |
705 | |
706 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); | |
707 | set_compound_order(page, order); | |
53f9263b | 708 | atomic_set(compound_mapcount_ptr(page), -1); |
47e29d32 JH |
709 | if (hpage_pincount_available(page)) |
710 | atomic_set(compound_pincount_ptr(page), 0); | |
18229df5 AW |
711 | } |
712 | ||
c0a32fc5 SG |
713 | #ifdef CONFIG_DEBUG_PAGEALLOC |
714 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 715 | |
8e57f8ac VB |
716 | bool _debug_pagealloc_enabled_early __read_mostly |
717 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
718 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 719 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 720 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
721 | |
722 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 723 | |
031bc574 JK |
724 | static int __init early_debug_pagealloc(char *buf) |
725 | { | |
8e57f8ac | 726 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
727 | } |
728 | early_param("debug_pagealloc", early_debug_pagealloc); | |
729 | ||
c0a32fc5 SG |
730 | static int __init debug_guardpage_minorder_setup(char *buf) |
731 | { | |
732 | unsigned long res; | |
733 | ||
734 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 735 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
736 | return 0; |
737 | } | |
738 | _debug_guardpage_minorder = res; | |
1170532b | 739 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
740 | return 0; |
741 | } | |
f1c1e9f7 | 742 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 743 | |
acbc15a4 | 744 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 745 | unsigned int order, int migratetype) |
c0a32fc5 | 746 | { |
e30825f1 | 747 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
748 | return false; |
749 | ||
750 | if (order >= debug_guardpage_minorder()) | |
751 | return false; | |
e30825f1 | 752 | |
3972f6bb | 753 | __SetPageGuard(page); |
2847cf95 JK |
754 | INIT_LIST_HEAD(&page->lru); |
755 | set_page_private(page, order); | |
756 | /* Guard pages are not available for any usage */ | |
757 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
758 | |
759 | return true; | |
c0a32fc5 SG |
760 | } |
761 | ||
2847cf95 JK |
762 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
763 | unsigned int order, int migratetype) | |
c0a32fc5 | 764 | { |
e30825f1 JK |
765 | if (!debug_guardpage_enabled()) |
766 | return; | |
767 | ||
3972f6bb | 768 | __ClearPageGuard(page); |
e30825f1 | 769 | |
2847cf95 JK |
770 | set_page_private(page, 0); |
771 | if (!is_migrate_isolate(migratetype)) | |
772 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
773 | } |
774 | #else | |
acbc15a4 JK |
775 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
776 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
777 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
778 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
779 | #endif |
780 | ||
04013513 VB |
781 | /* |
782 | * Enable static keys related to various memory debugging and hardening options. | |
783 | * Some override others, and depend on early params that are evaluated in the | |
784 | * order of appearance. So we need to first gather the full picture of what was | |
785 | * enabled, and then make decisions. | |
786 | */ | |
787 | void init_mem_debugging_and_hardening(void) | |
788 | { | |
9df65f52 ST |
789 | bool page_poisoning_requested = false; |
790 | ||
791 | #ifdef CONFIG_PAGE_POISONING | |
792 | /* | |
793 | * Page poisoning is debug page alloc for some arches. If | |
794 | * either of those options are enabled, enable poisoning. | |
795 | */ | |
796 | if (page_poisoning_enabled() || | |
797 | (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) && | |
798 | debug_pagealloc_enabled())) { | |
799 | static_branch_enable(&_page_poisoning_enabled); | |
800 | page_poisoning_requested = true; | |
801 | } | |
802 | #endif | |
803 | ||
04013513 | 804 | if (_init_on_alloc_enabled_early) { |
9df65f52 | 805 | if (page_poisoning_requested) |
04013513 VB |
806 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, " |
807 | "will take precedence over init_on_alloc\n"); | |
808 | else | |
809 | static_branch_enable(&init_on_alloc); | |
810 | } | |
811 | if (_init_on_free_enabled_early) { | |
9df65f52 | 812 | if (page_poisoning_requested) |
04013513 VB |
813 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, " |
814 | "will take precedence over init_on_free\n"); | |
815 | else | |
816 | static_branch_enable(&init_on_free); | |
817 | } | |
818 | ||
819 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
820 | if (!debug_pagealloc_enabled()) | |
821 | return; | |
822 | ||
823 | static_branch_enable(&_debug_pagealloc_enabled); | |
824 | ||
825 | if (!debug_guardpage_minorder()) | |
826 | return; | |
827 | ||
828 | static_branch_enable(&_debug_guardpage_enabled); | |
829 | #endif | |
830 | } | |
831 | ||
ab130f91 | 832 | static inline void set_buddy_order(struct page *page, unsigned int order) |
6aa3001b | 833 | { |
4c21e2f2 | 834 | set_page_private(page, order); |
676165a8 | 835 | __SetPageBuddy(page); |
1da177e4 LT |
836 | } |
837 | ||
1da177e4 LT |
838 | /* |
839 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 840 | * we can coalesce a page and its buddy if |
13ad59df | 841 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 842 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
843 | * (c) a page and its buddy have the same order && |
844 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 845 | * |
6e292b9b MW |
846 | * For recording whether a page is in the buddy system, we set PageBuddy. |
847 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 848 | * |
676165a8 | 849 | * For recording page's order, we use page_private(page). |
1da177e4 | 850 | */ |
fe925c0c | 851 | static inline bool page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 852 | unsigned int order) |
1da177e4 | 853 | { |
fe925c0c | 854 | if (!page_is_guard(buddy) && !PageBuddy(buddy)) |
855 | return false; | |
4c5018ce | 856 | |
ab130f91 | 857 | if (buddy_order(buddy) != order) |
fe925c0c | 858 | return false; |
c0a32fc5 | 859 | |
fe925c0c | 860 | /* |
861 | * zone check is done late to avoid uselessly calculating | |
862 | * zone/node ids for pages that could never merge. | |
863 | */ | |
864 | if (page_zone_id(page) != page_zone_id(buddy)) | |
865 | return false; | |
d34c5fa0 | 866 | |
fe925c0c | 867 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
4c5018ce | 868 | |
fe925c0c | 869 | return true; |
1da177e4 LT |
870 | } |
871 | ||
5e1f0f09 MG |
872 | #ifdef CONFIG_COMPACTION |
873 | static inline struct capture_control *task_capc(struct zone *zone) | |
874 | { | |
875 | struct capture_control *capc = current->capture_control; | |
876 | ||
deba0487 | 877 | return unlikely(capc) && |
5e1f0f09 MG |
878 | !(current->flags & PF_KTHREAD) && |
879 | !capc->page && | |
deba0487 | 880 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
881 | } |
882 | ||
883 | static inline bool | |
884 | compaction_capture(struct capture_control *capc, struct page *page, | |
885 | int order, int migratetype) | |
886 | { | |
887 | if (!capc || order != capc->cc->order) | |
888 | return false; | |
889 | ||
890 | /* Do not accidentally pollute CMA or isolated regions*/ | |
891 | if (is_migrate_cma(migratetype) || | |
892 | is_migrate_isolate(migratetype)) | |
893 | return false; | |
894 | ||
895 | /* | |
f0953a1b | 896 | * Do not let lower order allocations pollute a movable pageblock. |
5e1f0f09 MG |
897 | * This might let an unmovable request use a reclaimable pageblock |
898 | * and vice-versa but no more than normal fallback logic which can | |
899 | * have trouble finding a high-order free page. | |
900 | */ | |
901 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
902 | return false; | |
903 | ||
904 | capc->page = page; | |
905 | return true; | |
906 | } | |
907 | ||
908 | #else | |
909 | static inline struct capture_control *task_capc(struct zone *zone) | |
910 | { | |
911 | return NULL; | |
912 | } | |
913 | ||
914 | static inline bool | |
915 | compaction_capture(struct capture_control *capc, struct page *page, | |
916 | int order, int migratetype) | |
917 | { | |
918 | return false; | |
919 | } | |
920 | #endif /* CONFIG_COMPACTION */ | |
921 | ||
6ab01363 AD |
922 | /* Used for pages not on another list */ |
923 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
924 | unsigned int order, int migratetype) | |
925 | { | |
926 | struct free_area *area = &zone->free_area[order]; | |
927 | ||
928 | list_add(&page->lru, &area->free_list[migratetype]); | |
929 | area->nr_free++; | |
930 | } | |
931 | ||
932 | /* Used for pages not on another list */ | |
933 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
934 | unsigned int order, int migratetype) | |
935 | { | |
936 | struct free_area *area = &zone->free_area[order]; | |
937 | ||
938 | list_add_tail(&page->lru, &area->free_list[migratetype]); | |
939 | area->nr_free++; | |
940 | } | |
941 | ||
293ffa5e DH |
942 | /* |
943 | * Used for pages which are on another list. Move the pages to the tail | |
944 | * of the list - so the moved pages won't immediately be considered for | |
945 | * allocation again (e.g., optimization for memory onlining). | |
946 | */ | |
6ab01363 AD |
947 | static inline void move_to_free_list(struct page *page, struct zone *zone, |
948 | unsigned int order, int migratetype) | |
949 | { | |
950 | struct free_area *area = &zone->free_area[order]; | |
951 | ||
293ffa5e | 952 | list_move_tail(&page->lru, &area->free_list[migratetype]); |
6ab01363 AD |
953 | } |
954 | ||
955 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
956 | unsigned int order) | |
957 | { | |
36e66c55 AD |
958 | /* clear reported state and update reported page count */ |
959 | if (page_reported(page)) | |
960 | __ClearPageReported(page); | |
961 | ||
6ab01363 AD |
962 | list_del(&page->lru); |
963 | __ClearPageBuddy(page); | |
964 | set_page_private(page, 0); | |
965 | zone->free_area[order].nr_free--; | |
966 | } | |
967 | ||
a2129f24 AD |
968 | /* |
969 | * If this is not the largest possible page, check if the buddy | |
970 | * of the next-highest order is free. If it is, it's possible | |
971 | * that pages are being freed that will coalesce soon. In case, | |
972 | * that is happening, add the free page to the tail of the list | |
973 | * so it's less likely to be used soon and more likely to be merged | |
974 | * as a higher order page | |
975 | */ | |
976 | static inline bool | |
977 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
978 | struct page *page, unsigned int order) | |
979 | { | |
980 | struct page *higher_page, *higher_buddy; | |
981 | unsigned long combined_pfn; | |
982 | ||
983 | if (order >= MAX_ORDER - 2) | |
984 | return false; | |
985 | ||
986 | if (!pfn_valid_within(buddy_pfn)) | |
987 | return false; | |
988 | ||
989 | combined_pfn = buddy_pfn & pfn; | |
990 | higher_page = page + (combined_pfn - pfn); | |
991 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
992 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
993 | ||
994 | return pfn_valid_within(buddy_pfn) && | |
995 | page_is_buddy(higher_page, higher_buddy, order + 1); | |
996 | } | |
997 | ||
1da177e4 LT |
998 | /* |
999 | * Freeing function for a buddy system allocator. | |
1000 | * | |
1001 | * The concept of a buddy system is to maintain direct-mapped table | |
1002 | * (containing bit values) for memory blocks of various "orders". | |
1003 | * The bottom level table contains the map for the smallest allocatable | |
1004 | * units of memory (here, pages), and each level above it describes | |
1005 | * pairs of units from the levels below, hence, "buddies". | |
1006 | * At a high level, all that happens here is marking the table entry | |
1007 | * at the bottom level available, and propagating the changes upward | |
1008 | * as necessary, plus some accounting needed to play nicely with other | |
1009 | * parts of the VM system. | |
1010 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
1011 | * free pages of length of (1 << order) and marked with PageBuddy. |
1012 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 1013 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
1014 | * other. That is, if we allocate a small block, and both were |
1015 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 1016 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 1017 | * triggers coalescing into a block of larger size. |
1da177e4 | 1018 | * |
6d49e352 | 1019 | * -- nyc |
1da177e4 LT |
1020 | */ |
1021 | ||
48db57f8 | 1022 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 1023 | unsigned long pfn, |
ed0ae21d | 1024 | struct zone *zone, unsigned int order, |
f04a5d5d | 1025 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1026 | { |
a2129f24 | 1027 | struct capture_control *capc = task_capc(zone); |
3f649ab7 | 1028 | unsigned long buddy_pfn; |
a2129f24 | 1029 | unsigned long combined_pfn; |
d9dddbf5 | 1030 | unsigned int max_order; |
a2129f24 AD |
1031 | struct page *buddy; |
1032 | bool to_tail; | |
d9dddbf5 | 1033 | |
7ad69832 | 1034 | max_order = min_t(unsigned int, MAX_ORDER - 1, pageblock_order); |
1da177e4 | 1035 | |
d29bb978 | 1036 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 1037 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 1038 | |
ed0ae21d | 1039 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 1040 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 1041 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 1042 | |
76741e77 | 1043 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 1044 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 1045 | |
d9dddbf5 | 1046 | continue_merging: |
7ad69832 | 1047 | while (order < max_order) { |
5e1f0f09 MG |
1048 | if (compaction_capture(capc, page, order, migratetype)) { |
1049 | __mod_zone_freepage_state(zone, -(1 << order), | |
1050 | migratetype); | |
1051 | return; | |
1052 | } | |
76741e77 VB |
1053 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1054 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
1055 | |
1056 | if (!pfn_valid_within(buddy_pfn)) | |
1057 | goto done_merging; | |
cb2b95e1 | 1058 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 1059 | goto done_merging; |
c0a32fc5 SG |
1060 | /* |
1061 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
1062 | * merge with it and move up one order. | |
1063 | */ | |
b03641af | 1064 | if (page_is_guard(buddy)) |
2847cf95 | 1065 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 1066 | else |
6ab01363 | 1067 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
1068 | combined_pfn = buddy_pfn & pfn; |
1069 | page = page + (combined_pfn - pfn); | |
1070 | pfn = combined_pfn; | |
1da177e4 LT |
1071 | order++; |
1072 | } | |
7ad69832 | 1073 | if (order < MAX_ORDER - 1) { |
d9dddbf5 VB |
1074 | /* If we are here, it means order is >= pageblock_order. |
1075 | * We want to prevent merge between freepages on isolate | |
1076 | * pageblock and normal pageblock. Without this, pageblock | |
1077 | * isolation could cause incorrect freepage or CMA accounting. | |
1078 | * | |
1079 | * We don't want to hit this code for the more frequent | |
1080 | * low-order merging. | |
1081 | */ | |
1082 | if (unlikely(has_isolate_pageblock(zone))) { | |
1083 | int buddy_mt; | |
1084 | ||
76741e77 VB |
1085 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1086 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
1087 | buddy_mt = get_pageblock_migratetype(buddy); |
1088 | ||
1089 | if (migratetype != buddy_mt | |
1090 | && (is_migrate_isolate(migratetype) || | |
1091 | is_migrate_isolate(buddy_mt))) | |
1092 | goto done_merging; | |
1093 | } | |
7ad69832 | 1094 | max_order = order + 1; |
d9dddbf5 VB |
1095 | goto continue_merging; |
1096 | } | |
1097 | ||
1098 | done_merging: | |
ab130f91 | 1099 | set_buddy_order(page, order); |
6dda9d55 | 1100 | |
47b6a24a DH |
1101 | if (fpi_flags & FPI_TO_TAIL) |
1102 | to_tail = true; | |
1103 | else if (is_shuffle_order(order)) | |
a2129f24 | 1104 | to_tail = shuffle_pick_tail(); |
97500a4a | 1105 | else |
a2129f24 | 1106 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 1107 | |
a2129f24 | 1108 | if (to_tail) |
6ab01363 | 1109 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 1110 | else |
6ab01363 | 1111 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
1112 | |
1113 | /* Notify page reporting subsystem of freed page */ | |
f04a5d5d | 1114 | if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY)) |
36e66c55 | 1115 | page_reporting_notify_free(order); |
1da177e4 LT |
1116 | } |
1117 | ||
7bfec6f4 MG |
1118 | /* |
1119 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1120 | * try and check multiple fields with one check. The caller must do a detailed | |
1121 | * check if necessary. | |
1122 | */ | |
1123 | static inline bool page_expected_state(struct page *page, | |
1124 | unsigned long check_flags) | |
1125 | { | |
1126 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1127 | return false; | |
1128 | ||
1129 | if (unlikely((unsigned long)page->mapping | | |
1130 | page_ref_count(page) | | |
1131 | #ifdef CONFIG_MEMCG | |
48060834 | 1132 | page->memcg_data | |
7bfec6f4 MG |
1133 | #endif |
1134 | (page->flags & check_flags))) | |
1135 | return false; | |
1136 | ||
1137 | return true; | |
1138 | } | |
1139 | ||
58b7f119 | 1140 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 1141 | { |
82a3241a | 1142 | const char *bad_reason = NULL; |
f0b791a3 | 1143 | |
53f9263b | 1144 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1145 | bad_reason = "nonzero mapcount"; |
1146 | if (unlikely(page->mapping != NULL)) | |
1147 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1148 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1149 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
1150 | if (unlikely(page->flags & flags)) { |
1151 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
1152 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
1153 | else | |
1154 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 1155 | } |
9edad6ea | 1156 | #ifdef CONFIG_MEMCG |
48060834 | 1157 | if (unlikely(page->memcg_data)) |
9edad6ea JW |
1158 | bad_reason = "page still charged to cgroup"; |
1159 | #endif | |
58b7f119 WY |
1160 | return bad_reason; |
1161 | } | |
1162 | ||
1163 | static void check_free_page_bad(struct page *page) | |
1164 | { | |
1165 | bad_page(page, | |
1166 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
1167 | } |
1168 | ||
534fe5e3 | 1169 | static inline int check_free_page(struct page *page) |
bb552ac6 | 1170 | { |
da838d4f | 1171 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 1172 | return 0; |
bb552ac6 MG |
1173 | |
1174 | /* Something has gone sideways, find it */ | |
0d0c48a2 | 1175 | check_free_page_bad(page); |
7bfec6f4 | 1176 | return 1; |
1da177e4 LT |
1177 | } |
1178 | ||
4db7548c MG |
1179 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1180 | { | |
1181 | int ret = 1; | |
1182 | ||
1183 | /* | |
1184 | * We rely page->lru.next never has bit 0 set, unless the page | |
1185 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1186 | */ | |
1187 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1188 | ||
1189 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1190 | ret = 0; | |
1191 | goto out; | |
1192 | } | |
1193 | switch (page - head_page) { | |
1194 | case 1: | |
4da1984e | 1195 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c | 1196 | if (unlikely(compound_mapcount(page))) { |
82a3241a | 1197 | bad_page(page, "nonzero compound_mapcount"); |
4db7548c MG |
1198 | goto out; |
1199 | } | |
1200 | break; | |
1201 | case 2: | |
1202 | /* | |
1203 | * the second tail page: ->mapping is | |
fa3015b7 | 1204 | * deferred_list.next -- ignore value. |
4db7548c MG |
1205 | */ |
1206 | break; | |
1207 | default: | |
1208 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 1209 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
1210 | goto out; |
1211 | } | |
1212 | break; | |
1213 | } | |
1214 | if (unlikely(!PageTail(page))) { | |
82a3241a | 1215 | bad_page(page, "PageTail not set"); |
4db7548c MG |
1216 | goto out; |
1217 | } | |
1218 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 1219 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
1220 | goto out; |
1221 | } | |
1222 | ret = 0; | |
1223 | out: | |
1224 | page->mapping = NULL; | |
1225 | clear_compound_head(page); | |
1226 | return ret; | |
1227 | } | |
1228 | ||
6471384a AP |
1229 | static void kernel_init_free_pages(struct page *page, int numpages) |
1230 | { | |
1231 | int i; | |
1232 | ||
9e15afa5 QC |
1233 | /* s390's use of memset() could override KASAN redzones. */ |
1234 | kasan_disable_current(); | |
aa1ef4d7 | 1235 | for (i = 0; i < numpages; i++) { |
acb35b17 | 1236 | u8 tag = page_kasan_tag(page + i); |
aa1ef4d7 | 1237 | page_kasan_tag_reset(page + i); |
6471384a | 1238 | clear_highpage(page + i); |
acb35b17 | 1239 | page_kasan_tag_set(page + i, tag); |
aa1ef4d7 | 1240 | } |
9e15afa5 | 1241 | kasan_enable_current(); |
6471384a AP |
1242 | } |
1243 | ||
e2769dbd | 1244 | static __always_inline bool free_pages_prepare(struct page *page, |
2c335680 | 1245 | unsigned int order, bool check_free, fpi_t fpi_flags) |
4db7548c | 1246 | { |
e2769dbd | 1247 | int bad = 0; |
1bb5eab3 | 1248 | bool init; |
4db7548c | 1249 | |
4db7548c MG |
1250 | VM_BUG_ON_PAGE(PageTail(page), page); |
1251 | ||
e2769dbd | 1252 | trace_mm_page_free(page, order); |
e2769dbd | 1253 | |
79f5f8fa OS |
1254 | if (unlikely(PageHWPoison(page)) && !order) { |
1255 | /* | |
1256 | * Do not let hwpoison pages hit pcplists/buddy | |
1257 | * Untie memcg state and reset page's owner | |
1258 | */ | |
18b2db3b | 1259 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
79f5f8fa OS |
1260 | __memcg_kmem_uncharge_page(page, order); |
1261 | reset_page_owner(page, order); | |
1262 | return false; | |
1263 | } | |
1264 | ||
e2769dbd MG |
1265 | /* |
1266 | * Check tail pages before head page information is cleared to | |
1267 | * avoid checking PageCompound for order-0 pages. | |
1268 | */ | |
1269 | if (unlikely(order)) { | |
1270 | bool compound = PageCompound(page); | |
1271 | int i; | |
1272 | ||
1273 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1274 | |
9a73f61b KS |
1275 | if (compound) |
1276 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1277 | for (i = 1; i < (1 << order); i++) { |
1278 | if (compound) | |
1279 | bad += free_tail_pages_check(page, page + i); | |
534fe5e3 | 1280 | if (unlikely(check_free_page(page + i))) { |
e2769dbd MG |
1281 | bad++; |
1282 | continue; | |
1283 | } | |
1284 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1285 | } | |
1286 | } | |
bda807d4 | 1287 | if (PageMappingFlags(page)) |
4db7548c | 1288 | page->mapping = NULL; |
18b2db3b | 1289 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
f4b00eab | 1290 | __memcg_kmem_uncharge_page(page, order); |
e2769dbd | 1291 | if (check_free) |
534fe5e3 | 1292 | bad += check_free_page(page); |
e2769dbd MG |
1293 | if (bad) |
1294 | return false; | |
4db7548c | 1295 | |
e2769dbd MG |
1296 | page_cpupid_reset_last(page); |
1297 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1298 | reset_page_owner(page, order); | |
4db7548c MG |
1299 | |
1300 | if (!PageHighMem(page)) { | |
1301 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1302 | PAGE_SIZE << order); |
4db7548c | 1303 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1304 | PAGE_SIZE << order); |
4db7548c | 1305 | } |
6471384a | 1306 | |
8db26a3d VB |
1307 | kernel_poison_pages(page, 1 << order); |
1308 | ||
f9d79e8d | 1309 | /* |
1bb5eab3 AK |
1310 | * As memory initialization might be integrated into KASAN, |
1311 | * kasan_free_pages and kernel_init_free_pages must be | |
1312 | * kept together to avoid discrepancies in behavior. | |
1313 | * | |
f9d79e8d AK |
1314 | * With hardware tag-based KASAN, memory tags must be set before the |
1315 | * page becomes unavailable via debug_pagealloc or arch_free_page. | |
1316 | */ | |
1bb5eab3 AK |
1317 | init = want_init_on_free(); |
1318 | if (init && !kasan_has_integrated_init()) | |
1319 | kernel_init_free_pages(page, 1 << order); | |
1320 | kasan_free_nondeferred_pages(page, order, init, fpi_flags); | |
f9d79e8d | 1321 | |
234fdce8 QC |
1322 | /* |
1323 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1324 | * does this. So nothing which can access the page's contents should | |
1325 | * happen after this. | |
1326 | */ | |
1327 | arch_free_page(page, order); | |
1328 | ||
77bc7fd6 | 1329 | debug_pagealloc_unmap_pages(page, 1 << order); |
d6332692 | 1330 | |
4db7548c MG |
1331 | return true; |
1332 | } | |
1333 | ||
e2769dbd | 1334 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1335 | /* |
1336 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1337 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1338 | * moved from pcp lists to free lists. | |
1339 | */ | |
1340 | static bool free_pcp_prepare(struct page *page) | |
e2769dbd | 1341 | { |
2c335680 | 1342 | return free_pages_prepare(page, 0, true, FPI_NONE); |
e2769dbd MG |
1343 | } |
1344 | ||
4462b32c | 1345 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1346 | { |
8e57f8ac | 1347 | if (debug_pagealloc_enabled_static()) |
534fe5e3 | 1348 | return check_free_page(page); |
4462b32c VB |
1349 | else |
1350 | return false; | |
e2769dbd MG |
1351 | } |
1352 | #else | |
4462b32c VB |
1353 | /* |
1354 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1355 | * moving from pcp lists to free list in order to reduce overhead. With | |
1356 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1357 | * to the pcp lists. | |
1358 | */ | |
e2769dbd MG |
1359 | static bool free_pcp_prepare(struct page *page) |
1360 | { | |
8e57f8ac | 1361 | if (debug_pagealloc_enabled_static()) |
2c335680 | 1362 | return free_pages_prepare(page, 0, true, FPI_NONE); |
4462b32c | 1363 | else |
2c335680 | 1364 | return free_pages_prepare(page, 0, false, FPI_NONE); |
e2769dbd MG |
1365 | } |
1366 | ||
4db7548c MG |
1367 | static bool bulkfree_pcp_prepare(struct page *page) |
1368 | { | |
534fe5e3 | 1369 | return check_free_page(page); |
4db7548c MG |
1370 | } |
1371 | #endif /* CONFIG_DEBUG_VM */ | |
1372 | ||
97334162 AL |
1373 | static inline void prefetch_buddy(struct page *page) |
1374 | { | |
1375 | unsigned long pfn = page_to_pfn(page); | |
1376 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1377 | struct page *buddy = page + (buddy_pfn - pfn); | |
1378 | ||
1379 | prefetch(buddy); | |
1380 | } | |
1381 | ||
1da177e4 | 1382 | /* |
5f8dcc21 | 1383 | * Frees a number of pages from the PCP lists |
1da177e4 | 1384 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1385 | * count is the number of pages to free. |
1da177e4 LT |
1386 | * |
1387 | * If the zone was previously in an "all pages pinned" state then look to | |
1388 | * see if this freeing clears that state. | |
1389 | * | |
1390 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1391 | * pinned" detection logic. | |
1392 | */ | |
5f8dcc21 MG |
1393 | static void free_pcppages_bulk(struct zone *zone, int count, |
1394 | struct per_cpu_pages *pcp) | |
1da177e4 | 1395 | { |
5f8dcc21 | 1396 | int migratetype = 0; |
a6f9edd6 | 1397 | int batch_free = 0; |
5c3ad2eb | 1398 | int prefetch_nr = READ_ONCE(pcp->batch); |
3777999d | 1399 | bool isolated_pageblocks; |
0a5f4e5b AL |
1400 | struct page *page, *tmp; |
1401 | LIST_HEAD(head); | |
f2260e6b | 1402 | |
88e8ac11 CTR |
1403 | /* |
1404 | * Ensure proper count is passed which otherwise would stuck in the | |
1405 | * below while (list_empty(list)) loop. | |
1406 | */ | |
1407 | count = min(pcp->count, count); | |
e5b31ac2 | 1408 | while (count) { |
5f8dcc21 MG |
1409 | struct list_head *list; |
1410 | ||
1411 | /* | |
a6f9edd6 MG |
1412 | * Remove pages from lists in a round-robin fashion. A |
1413 | * batch_free count is maintained that is incremented when an | |
1414 | * empty list is encountered. This is so more pages are freed | |
1415 | * off fuller lists instead of spinning excessively around empty | |
1416 | * lists | |
5f8dcc21 MG |
1417 | */ |
1418 | do { | |
a6f9edd6 | 1419 | batch_free++; |
5f8dcc21 MG |
1420 | if (++migratetype == MIGRATE_PCPTYPES) |
1421 | migratetype = 0; | |
1422 | list = &pcp->lists[migratetype]; | |
1423 | } while (list_empty(list)); | |
48db57f8 | 1424 | |
1d16871d NK |
1425 | /* This is the only non-empty list. Free them all. */ |
1426 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1427 | batch_free = count; |
1d16871d | 1428 | |
a6f9edd6 | 1429 | do { |
a16601c5 | 1430 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1431 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1432 | list_del(&page->lru); |
77ba9062 | 1433 | pcp->count--; |
aa016d14 | 1434 | |
4db7548c MG |
1435 | if (bulkfree_pcp_prepare(page)) |
1436 | continue; | |
1437 | ||
0a5f4e5b | 1438 | list_add_tail(&page->lru, &head); |
97334162 AL |
1439 | |
1440 | /* | |
1441 | * We are going to put the page back to the global | |
1442 | * pool, prefetch its buddy to speed up later access | |
1443 | * under zone->lock. It is believed the overhead of | |
1444 | * an additional test and calculating buddy_pfn here | |
1445 | * can be offset by reduced memory latency later. To | |
1446 | * avoid excessive prefetching due to large count, only | |
1447 | * prefetch buddy for the first pcp->batch nr of pages. | |
1448 | */ | |
5c3ad2eb | 1449 | if (prefetch_nr) { |
97334162 | 1450 | prefetch_buddy(page); |
5c3ad2eb VB |
1451 | prefetch_nr--; |
1452 | } | |
e5b31ac2 | 1453 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1454 | } |
0a5f4e5b AL |
1455 | |
1456 | spin_lock(&zone->lock); | |
1457 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1458 | ||
1459 | /* | |
1460 | * Use safe version since after __free_one_page(), | |
1461 | * page->lru.next will not point to original list. | |
1462 | */ | |
1463 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1464 | int mt = get_pcppage_migratetype(page); | |
1465 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1466 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1467 | /* Pageblock could have been isolated meanwhile */ | |
1468 | if (unlikely(isolated_pageblocks)) | |
1469 | mt = get_pageblock_migratetype(page); | |
1470 | ||
f04a5d5d | 1471 | __free_one_page(page, page_to_pfn(page), zone, 0, mt, FPI_NONE); |
0a5f4e5b AL |
1472 | trace_mm_page_pcpu_drain(page, 0, mt); |
1473 | } | |
d34b0733 | 1474 | spin_unlock(&zone->lock); |
1da177e4 LT |
1475 | } |
1476 | ||
dc4b0caf MG |
1477 | static void free_one_page(struct zone *zone, |
1478 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1479 | unsigned int order, |
7fef431b | 1480 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1481 | { |
d34b0733 | 1482 | spin_lock(&zone->lock); |
ad53f92e JK |
1483 | if (unlikely(has_isolate_pageblock(zone) || |
1484 | is_migrate_isolate(migratetype))) { | |
1485 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1486 | } |
7fef431b | 1487 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); |
d34b0733 | 1488 | spin_unlock(&zone->lock); |
48db57f8 NP |
1489 | } |
1490 | ||
1e8ce83c | 1491 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1492 | unsigned long zone, int nid) |
1e8ce83c | 1493 | { |
d0dc12e8 | 1494 | mm_zero_struct_page(page); |
1e8ce83c | 1495 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1496 | init_page_count(page); |
1497 | page_mapcount_reset(page); | |
1498 | page_cpupid_reset_last(page); | |
2813b9c0 | 1499 | page_kasan_tag_reset(page); |
1e8ce83c | 1500 | |
1e8ce83c RH |
1501 | INIT_LIST_HEAD(&page->lru); |
1502 | #ifdef WANT_PAGE_VIRTUAL | |
1503 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1504 | if (!is_highmem_idx(zone)) | |
1505 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1506 | #endif | |
1507 | } | |
1508 | ||
7e18adb4 | 1509 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1510 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1511 | { |
1512 | pg_data_t *pgdat; | |
1513 | int nid, zid; | |
1514 | ||
1515 | if (!early_page_uninitialised(pfn)) | |
1516 | return; | |
1517 | ||
1518 | nid = early_pfn_to_nid(pfn); | |
1519 | pgdat = NODE_DATA(nid); | |
1520 | ||
1521 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1522 | struct zone *zone = &pgdat->node_zones[zid]; | |
1523 | ||
1524 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1525 | break; | |
1526 | } | |
d0dc12e8 | 1527 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1528 | } |
1529 | #else | |
1530 | static inline void init_reserved_page(unsigned long pfn) | |
1531 | { | |
1532 | } | |
1533 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1534 | ||
92923ca3 NZ |
1535 | /* |
1536 | * Initialised pages do not have PageReserved set. This function is | |
1537 | * called for each range allocated by the bootmem allocator and | |
1538 | * marks the pages PageReserved. The remaining valid pages are later | |
1539 | * sent to the buddy page allocator. | |
1540 | */ | |
4b50bcc7 | 1541 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1542 | { |
1543 | unsigned long start_pfn = PFN_DOWN(start); | |
1544 | unsigned long end_pfn = PFN_UP(end); | |
1545 | ||
7e18adb4 MG |
1546 | for (; start_pfn < end_pfn; start_pfn++) { |
1547 | if (pfn_valid(start_pfn)) { | |
1548 | struct page *page = pfn_to_page(start_pfn); | |
1549 | ||
1550 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1551 | |
1552 | /* Avoid false-positive PageTail() */ | |
1553 | INIT_LIST_HEAD(&page->lru); | |
1554 | ||
d483da5b AD |
1555 | /* |
1556 | * no need for atomic set_bit because the struct | |
1557 | * page is not visible yet so nobody should | |
1558 | * access it yet. | |
1559 | */ | |
1560 | __SetPageReserved(page); | |
7e18adb4 MG |
1561 | } |
1562 | } | |
92923ca3 NZ |
1563 | } |
1564 | ||
7fef431b DH |
1565 | static void __free_pages_ok(struct page *page, unsigned int order, |
1566 | fpi_t fpi_flags) | |
ec95f53a | 1567 | { |
d34b0733 | 1568 | unsigned long flags; |
95e34412 | 1569 | int migratetype; |
dc4b0caf | 1570 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1571 | |
2c335680 | 1572 | if (!free_pages_prepare(page, order, true, fpi_flags)) |
ec95f53a KM |
1573 | return; |
1574 | ||
cfc47a28 | 1575 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1576 | local_irq_save(flags); |
1577 | __count_vm_events(PGFREE, 1 << order); | |
7fef431b DH |
1578 | free_one_page(page_zone(page), page, pfn, order, migratetype, |
1579 | fpi_flags); | |
d34b0733 | 1580 | local_irq_restore(flags); |
1da177e4 LT |
1581 | } |
1582 | ||
a9cd410a | 1583 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1584 | { |
c3993076 | 1585 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1586 | struct page *p = page; |
c3993076 | 1587 | unsigned int loop; |
a226f6c8 | 1588 | |
7fef431b DH |
1589 | /* |
1590 | * When initializing the memmap, __init_single_page() sets the refcount | |
1591 | * of all pages to 1 ("allocated"/"not free"). We have to set the | |
1592 | * refcount of all involved pages to 0. | |
1593 | */ | |
e2d0bd2b YL |
1594 | prefetchw(p); |
1595 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1596 | prefetchw(p + 1); | |
c3993076 JW |
1597 | __ClearPageReserved(p); |
1598 | set_page_count(p, 0); | |
a226f6c8 | 1599 | } |
e2d0bd2b YL |
1600 | __ClearPageReserved(p); |
1601 | set_page_count(p, 0); | |
c3993076 | 1602 | |
9705bea5 | 1603 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
7fef431b DH |
1604 | |
1605 | /* | |
1606 | * Bypass PCP and place fresh pages right to the tail, primarily | |
1607 | * relevant for memory onlining. | |
1608 | */ | |
2c335680 | 1609 | __free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON); |
a226f6c8 DH |
1610 | } |
1611 | ||
3f08a302 | 1612 | #ifdef CONFIG_NEED_MULTIPLE_NODES |
7ace9917 | 1613 | |
03e92a5e MR |
1614 | /* |
1615 | * During memory init memblocks map pfns to nids. The search is expensive and | |
1616 | * this caches recent lookups. The implementation of __early_pfn_to_nid | |
1617 | * treats start/end as pfns. | |
1618 | */ | |
1619 | struct mminit_pfnnid_cache { | |
1620 | unsigned long last_start; | |
1621 | unsigned long last_end; | |
1622 | int last_nid; | |
1623 | }; | |
75a592a4 | 1624 | |
03e92a5e | 1625 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
6f24fbd3 MR |
1626 | |
1627 | /* | |
1628 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
1629 | */ | |
03e92a5e | 1630 | static int __meminit __early_pfn_to_nid(unsigned long pfn, |
6f24fbd3 | 1631 | struct mminit_pfnnid_cache *state) |
75a592a4 | 1632 | { |
6f24fbd3 | 1633 | unsigned long start_pfn, end_pfn; |
75a592a4 MG |
1634 | int nid; |
1635 | ||
6f24fbd3 MR |
1636 | if (state->last_start <= pfn && pfn < state->last_end) |
1637 | return state->last_nid; | |
1638 | ||
1639 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); | |
1640 | if (nid != NUMA_NO_NODE) { | |
1641 | state->last_start = start_pfn; | |
1642 | state->last_end = end_pfn; | |
1643 | state->last_nid = nid; | |
1644 | } | |
7ace9917 MG |
1645 | |
1646 | return nid; | |
75a592a4 | 1647 | } |
75a592a4 | 1648 | |
75a592a4 | 1649 | int __meminit early_pfn_to_nid(unsigned long pfn) |
75a592a4 | 1650 | { |
7ace9917 | 1651 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1652 | int nid; |
1653 | ||
7ace9917 | 1654 | spin_lock(&early_pfn_lock); |
56ec43d8 | 1655 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1656 | if (nid < 0) |
e4568d38 | 1657 | nid = first_online_node; |
7ace9917 | 1658 | spin_unlock(&early_pfn_lock); |
75a592a4 | 1659 | |
7ace9917 | 1660 | return nid; |
75a592a4 | 1661 | } |
3f08a302 | 1662 | #endif /* CONFIG_NEED_MULTIPLE_NODES */ |
75a592a4 | 1663 | |
7c2ee349 | 1664 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1665 | unsigned int order) |
1666 | { | |
1667 | if (early_page_uninitialised(pfn)) | |
1668 | return; | |
a9cd410a | 1669 | __free_pages_core(page, order); |
3a80a7fa MG |
1670 | } |
1671 | ||
7cf91a98 JK |
1672 | /* |
1673 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1674 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1675 | * with the migration of free compaction scanner. The scanners then need to | |
1676 | * use only pfn_valid_within() check for arches that allow holes within | |
1677 | * pageblocks. | |
1678 | * | |
1679 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1680 | * | |
1681 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1682 | * i.e. it's possible that all pages within a zones range of pages do not | |
1683 | * belong to a single zone. We assume that a border between node0 and node1 | |
1684 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1685 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1686 | * the first and last page of a pageblock and avoid checking each individual | |
1687 | * page in a pageblock. | |
1688 | */ | |
1689 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1690 | unsigned long end_pfn, struct zone *zone) | |
1691 | { | |
1692 | struct page *start_page; | |
1693 | struct page *end_page; | |
1694 | ||
1695 | /* end_pfn is one past the range we are checking */ | |
1696 | end_pfn--; | |
1697 | ||
1698 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1699 | return NULL; | |
1700 | ||
2d070eab MH |
1701 | start_page = pfn_to_online_page(start_pfn); |
1702 | if (!start_page) | |
1703 | return NULL; | |
7cf91a98 JK |
1704 | |
1705 | if (page_zone(start_page) != zone) | |
1706 | return NULL; | |
1707 | ||
1708 | end_page = pfn_to_page(end_pfn); | |
1709 | ||
1710 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1711 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1712 | return NULL; | |
1713 | ||
1714 | return start_page; | |
1715 | } | |
1716 | ||
1717 | void set_zone_contiguous(struct zone *zone) | |
1718 | { | |
1719 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1720 | unsigned long block_end_pfn; | |
1721 | ||
1722 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1723 | for (; block_start_pfn < zone_end_pfn(zone); | |
1724 | block_start_pfn = block_end_pfn, | |
1725 | block_end_pfn += pageblock_nr_pages) { | |
1726 | ||
1727 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1728 | ||
1729 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1730 | block_end_pfn, zone)) | |
1731 | return; | |
e84fe99b | 1732 | cond_resched(); |
7cf91a98 JK |
1733 | } |
1734 | ||
1735 | /* We confirm that there is no hole */ | |
1736 | zone->contiguous = true; | |
1737 | } | |
1738 | ||
1739 | void clear_zone_contiguous(struct zone *zone) | |
1740 | { | |
1741 | zone->contiguous = false; | |
1742 | } | |
1743 | ||
7e18adb4 | 1744 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1745 | static void __init deferred_free_range(unsigned long pfn, |
1746 | unsigned long nr_pages) | |
a4de83dd | 1747 | { |
2f47a91f PT |
1748 | struct page *page; |
1749 | unsigned long i; | |
a4de83dd | 1750 | |
2f47a91f | 1751 | if (!nr_pages) |
a4de83dd MG |
1752 | return; |
1753 | ||
2f47a91f PT |
1754 | page = pfn_to_page(pfn); |
1755 | ||
a4de83dd | 1756 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1757 | if (nr_pages == pageblock_nr_pages && |
1758 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1759 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1760 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1761 | return; |
1762 | } | |
1763 | ||
e780149b XQ |
1764 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1765 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1766 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
a9cd410a | 1767 | __free_pages_core(page, 0); |
e780149b | 1768 | } |
a4de83dd MG |
1769 | } |
1770 | ||
d3cd131d NS |
1771 | /* Completion tracking for deferred_init_memmap() threads */ |
1772 | static atomic_t pgdat_init_n_undone __initdata; | |
1773 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1774 | ||
1775 | static inline void __init pgdat_init_report_one_done(void) | |
1776 | { | |
1777 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1778 | complete(&pgdat_init_all_done_comp); | |
1779 | } | |
0e1cc95b | 1780 | |
2f47a91f | 1781 | /* |
80b1f41c PT |
1782 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1783 | * | |
1784 | * First we check if pfn is valid on architectures where it is possible to have | |
1785 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1786 | * function is optimized out. | |
1787 | * | |
1788 | * Then, we check if a current large page is valid by only checking the validity | |
1789 | * of the head pfn. | |
2f47a91f | 1790 | */ |
56ec43d8 | 1791 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1792 | { |
80b1f41c PT |
1793 | if (!pfn_valid_within(pfn)) |
1794 | return false; | |
1795 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1796 | return false; | |
80b1f41c PT |
1797 | return true; |
1798 | } | |
2f47a91f | 1799 | |
80b1f41c PT |
1800 | /* |
1801 | * Free pages to buddy allocator. Try to free aligned pages in | |
1802 | * pageblock_nr_pages sizes. | |
1803 | */ | |
56ec43d8 | 1804 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1805 | unsigned long end_pfn) |
1806 | { | |
80b1f41c PT |
1807 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
1808 | unsigned long nr_free = 0; | |
2f47a91f | 1809 | |
80b1f41c | 1810 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1811 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1812 | deferred_free_range(pfn - nr_free, nr_free); |
1813 | nr_free = 0; | |
1814 | } else if (!(pfn & nr_pgmask)) { | |
1815 | deferred_free_range(pfn - nr_free, nr_free); | |
1816 | nr_free = 1; | |
80b1f41c PT |
1817 | } else { |
1818 | nr_free++; | |
1819 | } | |
1820 | } | |
1821 | /* Free the last block of pages to allocator */ | |
1822 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1823 | } |
1824 | ||
80b1f41c PT |
1825 | /* |
1826 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1827 | * by performing it only once every pageblock_nr_pages. | |
1828 | * Return number of pages initialized. | |
1829 | */ | |
56ec43d8 | 1830 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1831 | unsigned long pfn, |
1832 | unsigned long end_pfn) | |
2f47a91f | 1833 | { |
2f47a91f | 1834 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
56ec43d8 | 1835 | int nid = zone_to_nid(zone); |
2f47a91f | 1836 | unsigned long nr_pages = 0; |
56ec43d8 | 1837 | int zid = zone_idx(zone); |
2f47a91f | 1838 | struct page *page = NULL; |
2f47a91f | 1839 | |
80b1f41c | 1840 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1841 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1842 | page = NULL; |
2f47a91f | 1843 | continue; |
80b1f41c | 1844 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1845 | page = pfn_to_page(pfn); |
80b1f41c PT |
1846 | } else { |
1847 | page++; | |
2f47a91f | 1848 | } |
d0dc12e8 | 1849 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1850 | nr_pages++; |
2f47a91f | 1851 | } |
80b1f41c | 1852 | return (nr_pages); |
2f47a91f PT |
1853 | } |
1854 | ||
0e56acae AD |
1855 | /* |
1856 | * This function is meant to pre-load the iterator for the zone init. | |
1857 | * Specifically it walks through the ranges until we are caught up to the | |
1858 | * first_init_pfn value and exits there. If we never encounter the value we | |
1859 | * return false indicating there are no valid ranges left. | |
1860 | */ | |
1861 | static bool __init | |
1862 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
1863 | unsigned long *spfn, unsigned long *epfn, | |
1864 | unsigned long first_init_pfn) | |
1865 | { | |
1866 | u64 j; | |
1867 | ||
1868 | /* | |
1869 | * Start out by walking through the ranges in this zone that have | |
1870 | * already been initialized. We don't need to do anything with them | |
1871 | * so we just need to flush them out of the system. | |
1872 | */ | |
1873 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
1874 | if (*epfn <= first_init_pfn) | |
1875 | continue; | |
1876 | if (*spfn < first_init_pfn) | |
1877 | *spfn = first_init_pfn; | |
1878 | *i = j; | |
1879 | return true; | |
1880 | } | |
1881 | ||
1882 | return false; | |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * Initialize and free pages. We do it in two loops: first we initialize | |
1887 | * struct page, then free to buddy allocator, because while we are | |
1888 | * freeing pages we can access pages that are ahead (computing buddy | |
1889 | * page in __free_one_page()). | |
1890 | * | |
1891 | * In order to try and keep some memory in the cache we have the loop | |
1892 | * broken along max page order boundaries. This way we will not cause | |
1893 | * any issues with the buddy page computation. | |
1894 | */ | |
1895 | static unsigned long __init | |
1896 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
1897 | unsigned long *end_pfn) | |
1898 | { | |
1899 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
1900 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
1901 | unsigned long nr_pages = 0; | |
1902 | u64 j = *i; | |
1903 | ||
1904 | /* First we loop through and initialize the page values */ | |
1905 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
1906 | unsigned long t; | |
1907 | ||
1908 | if (mo_pfn <= *start_pfn) | |
1909 | break; | |
1910 | ||
1911 | t = min(mo_pfn, *end_pfn); | |
1912 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
1913 | ||
1914 | if (mo_pfn < *end_pfn) { | |
1915 | *start_pfn = mo_pfn; | |
1916 | break; | |
1917 | } | |
1918 | } | |
1919 | ||
1920 | /* Reset values and now loop through freeing pages as needed */ | |
1921 | swap(j, *i); | |
1922 | ||
1923 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
1924 | unsigned long t; | |
1925 | ||
1926 | if (mo_pfn <= spfn) | |
1927 | break; | |
1928 | ||
1929 | t = min(mo_pfn, epfn); | |
1930 | deferred_free_pages(spfn, t); | |
1931 | ||
1932 | if (mo_pfn <= epfn) | |
1933 | break; | |
1934 | } | |
1935 | ||
1936 | return nr_pages; | |
1937 | } | |
1938 | ||
e4443149 DJ |
1939 | static void __init |
1940 | deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, | |
1941 | void *arg) | |
1942 | { | |
1943 | unsigned long spfn, epfn; | |
1944 | struct zone *zone = arg; | |
1945 | u64 i; | |
1946 | ||
1947 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn); | |
1948 | ||
1949 | /* | |
1950 | * Initialize and free pages in MAX_ORDER sized increments so that we | |
1951 | * can avoid introducing any issues with the buddy allocator. | |
1952 | */ | |
1953 | while (spfn < end_pfn) { | |
1954 | deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
1955 | cond_resched(); | |
1956 | } | |
1957 | } | |
1958 | ||
ecd09650 DJ |
1959 | /* An arch may override for more concurrency. */ |
1960 | __weak int __init | |
1961 | deferred_page_init_max_threads(const struct cpumask *node_cpumask) | |
1962 | { | |
1963 | return 1; | |
1964 | } | |
1965 | ||
7e18adb4 | 1966 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1967 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1968 | { |
0e1cc95b | 1969 | pg_data_t *pgdat = data; |
0e56acae | 1970 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
89c7c402 | 1971 | unsigned long spfn = 0, epfn = 0; |
0e56acae | 1972 | unsigned long first_init_pfn, flags; |
7e18adb4 | 1973 | unsigned long start = jiffies; |
7e18adb4 | 1974 | struct zone *zone; |
e4443149 | 1975 | int zid, max_threads; |
2f47a91f | 1976 | u64 i; |
7e18adb4 | 1977 | |
3a2d7fa8 PT |
1978 | /* Bind memory initialisation thread to a local node if possible */ |
1979 | if (!cpumask_empty(cpumask)) | |
1980 | set_cpus_allowed_ptr(current, cpumask); | |
1981 | ||
1982 | pgdat_resize_lock(pgdat, &flags); | |
1983 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1984 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1985 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1986 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1987 | return 0; |
1988 | } | |
1989 | ||
7e18adb4 MG |
1990 | /* Sanity check boundaries */ |
1991 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1992 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1993 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1994 | ||
3d060856 PT |
1995 | /* |
1996 | * Once we unlock here, the zone cannot be grown anymore, thus if an | |
1997 | * interrupt thread must allocate this early in boot, zone must be | |
1998 | * pre-grown prior to start of deferred page initialization. | |
1999 | */ | |
2000 | pgdat_resize_unlock(pgdat, &flags); | |
2001 | ||
7e18adb4 MG |
2002 | /* Only the highest zone is deferred so find it */ |
2003 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
2004 | zone = pgdat->node_zones + zid; | |
2005 | if (first_init_pfn < zone_end_pfn(zone)) | |
2006 | break; | |
2007 | } | |
0e56acae AD |
2008 | |
2009 | /* If the zone is empty somebody else may have cleared out the zone */ | |
2010 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2011 | first_init_pfn)) | |
2012 | goto zone_empty; | |
7e18adb4 | 2013 | |
ecd09650 | 2014 | max_threads = deferred_page_init_max_threads(cpumask); |
7e18adb4 | 2015 | |
117003c3 | 2016 | while (spfn < epfn) { |
e4443149 DJ |
2017 | unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION); |
2018 | struct padata_mt_job job = { | |
2019 | .thread_fn = deferred_init_memmap_chunk, | |
2020 | .fn_arg = zone, | |
2021 | .start = spfn, | |
2022 | .size = epfn_align - spfn, | |
2023 | .align = PAGES_PER_SECTION, | |
2024 | .min_chunk = PAGES_PER_SECTION, | |
2025 | .max_threads = max_threads, | |
2026 | }; | |
2027 | ||
2028 | padata_do_multithreaded(&job); | |
2029 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2030 | epfn_align); | |
117003c3 | 2031 | } |
0e56acae | 2032 | zone_empty: |
7e18adb4 MG |
2033 | /* Sanity check that the next zone really is unpopulated */ |
2034 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
2035 | ||
89c7c402 DJ |
2036 | pr_info("node %d deferred pages initialised in %ums\n", |
2037 | pgdat->node_id, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
2038 | |
2039 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
2040 | return 0; |
2041 | } | |
c9e97a19 | 2042 | |
c9e97a19 PT |
2043 | /* |
2044 | * If this zone has deferred pages, try to grow it by initializing enough | |
2045 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
2046 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
2047 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
2048 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
2049 | * | |
2050 | * Return true when zone was grown, otherwise return false. We return true even | |
2051 | * when we grow less than requested, to let the caller decide if there are | |
2052 | * enough pages to satisfy the allocation. | |
2053 | * | |
2054 | * Note: We use noinline because this function is needed only during boot, and | |
2055 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
2056 | * making sure that it is not inlined into permanent text section. | |
2057 | */ | |
2058 | static noinline bool __init | |
2059 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
2060 | { | |
c9e97a19 | 2061 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 2062 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 2063 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
2064 | unsigned long spfn, epfn, flags; |
2065 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
2066 | u64 i; |
2067 | ||
2068 | /* Only the last zone may have deferred pages */ | |
2069 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
2070 | return false; | |
2071 | ||
2072 | pgdat_resize_lock(pgdat, &flags); | |
2073 | ||
c9e97a19 PT |
2074 | /* |
2075 | * If someone grew this zone while we were waiting for spinlock, return | |
2076 | * true, as there might be enough pages already. | |
2077 | */ | |
2078 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
2079 | pgdat_resize_unlock(pgdat, &flags); | |
2080 | return true; | |
2081 | } | |
2082 | ||
0e56acae AD |
2083 | /* If the zone is empty somebody else may have cleared out the zone */ |
2084 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2085 | first_deferred_pfn)) { | |
2086 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 2087 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
2088 | /* Retry only once. */ |
2089 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
2090 | } |
2091 | ||
0e56acae AD |
2092 | /* |
2093 | * Initialize and free pages in MAX_ORDER sized increments so | |
2094 | * that we can avoid introducing any issues with the buddy | |
2095 | * allocator. | |
2096 | */ | |
2097 | while (spfn < epfn) { | |
2098 | /* update our first deferred PFN for this section */ | |
2099 | first_deferred_pfn = spfn; | |
2100 | ||
2101 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
117003c3 | 2102 | touch_nmi_watchdog(); |
c9e97a19 | 2103 | |
0e56acae AD |
2104 | /* We should only stop along section boundaries */ |
2105 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
2106 | continue; | |
c9e97a19 | 2107 | |
0e56acae | 2108 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
2109 | if (nr_pages >= nr_pages_needed) |
2110 | break; | |
2111 | } | |
2112 | ||
0e56acae | 2113 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
2114 | pgdat_resize_unlock(pgdat, &flags); |
2115 | ||
2116 | return nr_pages > 0; | |
2117 | } | |
2118 | ||
2119 | /* | |
2120 | * deferred_grow_zone() is __init, but it is called from | |
2121 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
2122 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
2123 | * and to ensure that the function body gets unloaded. | |
2124 | */ | |
2125 | static bool __ref | |
2126 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
2127 | { | |
2128 | return deferred_grow_zone(zone, order); | |
2129 | } | |
2130 | ||
7cf91a98 | 2131 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
2132 | |
2133 | void __init page_alloc_init_late(void) | |
2134 | { | |
7cf91a98 | 2135 | struct zone *zone; |
e900a918 | 2136 | int nid; |
7cf91a98 JK |
2137 | |
2138 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 2139 | |
d3cd131d NS |
2140 | /* There will be num_node_state(N_MEMORY) threads */ |
2141 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 2142 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
2143 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
2144 | } | |
2145 | ||
2146 | /* Block until all are initialised */ | |
d3cd131d | 2147 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 2148 | |
3e8fc007 MG |
2149 | /* |
2150 | * The number of managed pages has changed due to the initialisation | |
2151 | * so the pcpu batch and high limits needs to be updated or the limits | |
2152 | * will be artificially small. | |
2153 | */ | |
2154 | for_each_populated_zone(zone) | |
2155 | zone_pcp_update(zone); | |
2156 | ||
c9e97a19 PT |
2157 | /* |
2158 | * We initialized the rest of the deferred pages. Permanently disable | |
2159 | * on-demand struct page initialization. | |
2160 | */ | |
2161 | static_branch_disable(&deferred_pages); | |
2162 | ||
4248b0da MG |
2163 | /* Reinit limits that are based on free pages after the kernel is up */ |
2164 | files_maxfiles_init(); | |
7cf91a98 | 2165 | #endif |
350e88ba | 2166 | |
ba8f3587 LF |
2167 | buffer_init(); |
2168 | ||
3010f876 PT |
2169 | /* Discard memblock private memory */ |
2170 | memblock_discard(); | |
7cf91a98 | 2171 | |
e900a918 DW |
2172 | for_each_node_state(nid, N_MEMORY) |
2173 | shuffle_free_memory(NODE_DATA(nid)); | |
2174 | ||
7cf91a98 JK |
2175 | for_each_populated_zone(zone) |
2176 | set_zone_contiguous(zone); | |
7e18adb4 | 2177 | } |
7e18adb4 | 2178 | |
47118af0 | 2179 | #ifdef CONFIG_CMA |
9cf510a5 | 2180 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
2181 | void __init init_cma_reserved_pageblock(struct page *page) |
2182 | { | |
2183 | unsigned i = pageblock_nr_pages; | |
2184 | struct page *p = page; | |
2185 | ||
2186 | do { | |
2187 | __ClearPageReserved(p); | |
2188 | set_page_count(p, 0); | |
d883c6cf | 2189 | } while (++p, --i); |
47118af0 | 2190 | |
47118af0 | 2191 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
2192 | |
2193 | if (pageblock_order >= MAX_ORDER) { | |
2194 | i = pageblock_nr_pages; | |
2195 | p = page; | |
2196 | do { | |
2197 | set_page_refcounted(p); | |
2198 | __free_pages(p, MAX_ORDER - 1); | |
2199 | p += MAX_ORDER_NR_PAGES; | |
2200 | } while (i -= MAX_ORDER_NR_PAGES); | |
2201 | } else { | |
2202 | set_page_refcounted(page); | |
2203 | __free_pages(page, pageblock_order); | |
2204 | } | |
2205 | ||
3dcc0571 | 2206 | adjust_managed_page_count(page, pageblock_nr_pages); |
3c381db1 | 2207 | page_zone(page)->cma_pages += pageblock_nr_pages; |
47118af0 MN |
2208 | } |
2209 | #endif | |
1da177e4 LT |
2210 | |
2211 | /* | |
2212 | * The order of subdivision here is critical for the IO subsystem. | |
2213 | * Please do not alter this order without good reasons and regression | |
2214 | * testing. Specifically, as large blocks of memory are subdivided, | |
2215 | * the order in which smaller blocks are delivered depends on the order | |
2216 | * they're subdivided in this function. This is the primary factor | |
2217 | * influencing the order in which pages are delivered to the IO | |
2218 | * subsystem according to empirical testing, and this is also justified | |
2219 | * by considering the behavior of a buddy system containing a single | |
2220 | * large block of memory acted on by a series of small allocations. | |
2221 | * This behavior is a critical factor in sglist merging's success. | |
2222 | * | |
6d49e352 | 2223 | * -- nyc |
1da177e4 | 2224 | */ |
085cc7d5 | 2225 | static inline void expand(struct zone *zone, struct page *page, |
6ab01363 | 2226 | int low, int high, int migratetype) |
1da177e4 LT |
2227 | { |
2228 | unsigned long size = 1 << high; | |
2229 | ||
2230 | while (high > low) { | |
1da177e4 LT |
2231 | high--; |
2232 | size >>= 1; | |
309381fe | 2233 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2234 | |
acbc15a4 JK |
2235 | /* |
2236 | * Mark as guard pages (or page), that will allow to | |
2237 | * merge back to allocator when buddy will be freed. | |
2238 | * Corresponding page table entries will not be touched, | |
2239 | * pages will stay not present in virtual address space | |
2240 | */ | |
2241 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2242 | continue; |
acbc15a4 | 2243 | |
6ab01363 | 2244 | add_to_free_list(&page[size], zone, high, migratetype); |
ab130f91 | 2245 | set_buddy_order(&page[size], high); |
1da177e4 | 2246 | } |
1da177e4 LT |
2247 | } |
2248 | ||
4e611801 | 2249 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2250 | { |
f4c18e6f | 2251 | if (unlikely(page->flags & __PG_HWPOISON)) { |
e570f56c NH |
2252 | /* Don't complain about hwpoisoned pages */ |
2253 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2254 | return; | |
f4c18e6f | 2255 | } |
58b7f119 WY |
2256 | |
2257 | bad_page(page, | |
2258 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
4e611801 VB |
2259 | } |
2260 | ||
2261 | /* | |
2262 | * This page is about to be returned from the page allocator | |
2263 | */ | |
2264 | static inline int check_new_page(struct page *page) | |
2265 | { | |
2266 | if (likely(page_expected_state(page, | |
2267 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2268 | return 0; | |
2269 | ||
2270 | check_new_page_bad(page); | |
2271 | return 1; | |
2a7684a2 WF |
2272 | } |
2273 | ||
479f854a | 2274 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2275 | /* |
2276 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2277 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2278 | * also checked when pcp lists are refilled from the free lists. | |
2279 | */ | |
2280 | static inline bool check_pcp_refill(struct page *page) | |
479f854a | 2281 | { |
8e57f8ac | 2282 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2283 | return check_new_page(page); |
2284 | else | |
2285 | return false; | |
479f854a MG |
2286 | } |
2287 | ||
4462b32c | 2288 | static inline bool check_new_pcp(struct page *page) |
479f854a MG |
2289 | { |
2290 | return check_new_page(page); | |
2291 | } | |
2292 | #else | |
4462b32c VB |
2293 | /* |
2294 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2295 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2296 | * enabled, they are also checked when being allocated from the pcp lists. | |
2297 | */ | |
2298 | static inline bool check_pcp_refill(struct page *page) | |
479f854a MG |
2299 | { |
2300 | return check_new_page(page); | |
2301 | } | |
4462b32c | 2302 | static inline bool check_new_pcp(struct page *page) |
479f854a | 2303 | { |
8e57f8ac | 2304 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2305 | return check_new_page(page); |
2306 | else | |
2307 | return false; | |
479f854a MG |
2308 | } |
2309 | #endif /* CONFIG_DEBUG_VM */ | |
2310 | ||
2311 | static bool check_new_pages(struct page *page, unsigned int order) | |
2312 | { | |
2313 | int i; | |
2314 | for (i = 0; i < (1 << order); i++) { | |
2315 | struct page *p = page + i; | |
2316 | ||
2317 | if (unlikely(check_new_page(p))) | |
2318 | return true; | |
2319 | } | |
2320 | ||
2321 | return false; | |
2322 | } | |
2323 | ||
46f24fd8 JK |
2324 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2325 | gfp_t gfp_flags) | |
2326 | { | |
1bb5eab3 AK |
2327 | bool init; |
2328 | ||
46f24fd8 JK |
2329 | set_page_private(page, 0); |
2330 | set_page_refcounted(page); | |
2331 | ||
2332 | arch_alloc_page(page, order); | |
77bc7fd6 | 2333 | debug_pagealloc_map_pages(page, 1 << order); |
1bb5eab3 AK |
2334 | |
2335 | /* | |
2336 | * Page unpoisoning must happen before memory initialization. | |
2337 | * Otherwise, the poison pattern will be overwritten for __GFP_ZERO | |
2338 | * allocations and the page unpoisoning code will complain. | |
2339 | */ | |
8db26a3d | 2340 | kernel_unpoison_pages(page, 1 << order); |
862b6dee | 2341 | |
1bb5eab3 AK |
2342 | /* |
2343 | * As memory initialization might be integrated into KASAN, | |
2344 | * kasan_alloc_pages and kernel_init_free_pages must be | |
2345 | * kept together to avoid discrepancies in behavior. | |
2346 | */ | |
2347 | init = !want_init_on_free() && want_init_on_alloc(gfp_flags); | |
2348 | kasan_alloc_pages(page, order, init); | |
2349 | if (init && !kasan_has_integrated_init()) | |
862b6dee | 2350 | kernel_init_free_pages(page, 1 << order); |
1bb5eab3 AK |
2351 | |
2352 | set_page_owner(page, order, gfp_flags); | |
46f24fd8 JK |
2353 | } |
2354 | ||
479f854a | 2355 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2356 | unsigned int alloc_flags) |
2a7684a2 | 2357 | { |
46f24fd8 | 2358 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2359 | |
17cf4406 NP |
2360 | if (order && (gfp_flags & __GFP_COMP)) |
2361 | prep_compound_page(page, order); | |
2362 | ||
75379191 | 2363 | /* |
2f064f34 | 2364 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2365 | * allocate the page. The expectation is that the caller is taking |
2366 | * steps that will free more memory. The caller should avoid the page | |
2367 | * being used for !PFMEMALLOC purposes. | |
2368 | */ | |
2f064f34 MH |
2369 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2370 | set_page_pfmemalloc(page); | |
2371 | else | |
2372 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2373 | } |
2374 | ||
56fd56b8 MG |
2375 | /* |
2376 | * Go through the free lists for the given migratetype and remove | |
2377 | * the smallest available page from the freelists | |
2378 | */ | |
85ccc8fa | 2379 | static __always_inline |
728ec980 | 2380 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2381 | int migratetype) |
2382 | { | |
2383 | unsigned int current_order; | |
b8af2941 | 2384 | struct free_area *area; |
56fd56b8 MG |
2385 | struct page *page; |
2386 | ||
2387 | /* Find a page of the appropriate size in the preferred list */ | |
2388 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2389 | area = &(zone->free_area[current_order]); | |
b03641af | 2390 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2391 | if (!page) |
2392 | continue; | |
6ab01363 AD |
2393 | del_page_from_free_list(page, zone, current_order); |
2394 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 2395 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
2396 | return page; |
2397 | } | |
2398 | ||
2399 | return NULL; | |
2400 | } | |
2401 | ||
2402 | ||
b2a0ac88 MG |
2403 | /* |
2404 | * This array describes the order lists are fallen back to when | |
2405 | * the free lists for the desirable migrate type are depleted | |
2406 | */ | |
da415663 | 2407 | static int fallbacks[MIGRATE_TYPES][3] = { |
974a786e | 2408 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2409 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2410 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
47118af0 | 2411 | #ifdef CONFIG_CMA |
974a786e | 2412 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 2413 | #endif |
194159fb | 2414 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 2415 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 2416 | #endif |
b2a0ac88 MG |
2417 | }; |
2418 | ||
dc67647b | 2419 | #ifdef CONFIG_CMA |
85ccc8fa | 2420 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2421 | unsigned int order) |
2422 | { | |
2423 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2424 | } | |
2425 | #else | |
2426 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2427 | unsigned int order) { return NULL; } | |
2428 | #endif | |
2429 | ||
c361be55 | 2430 | /* |
293ffa5e | 2431 | * Move the free pages in a range to the freelist tail of the requested type. |
d9c23400 | 2432 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2433 | * boundary. If alignment is required, use move_freepages_block() |
2434 | */ | |
02aa0cdd | 2435 | static int move_freepages(struct zone *zone, |
39ddb991 | 2436 | unsigned long start_pfn, unsigned long end_pfn, |
02aa0cdd | 2437 | int migratetype, int *num_movable) |
c361be55 MG |
2438 | { |
2439 | struct page *page; | |
39ddb991 | 2440 | unsigned long pfn; |
d00181b9 | 2441 | unsigned int order; |
d100313f | 2442 | int pages_moved = 0; |
c361be55 | 2443 | |
39ddb991 KW |
2444 | for (pfn = start_pfn; pfn <= end_pfn;) { |
2445 | if (!pfn_valid_within(pfn)) { | |
2446 | pfn++; | |
c361be55 MG |
2447 | continue; |
2448 | } | |
2449 | ||
39ddb991 | 2450 | page = pfn_to_page(pfn); |
c361be55 | 2451 | if (!PageBuddy(page)) { |
02aa0cdd VB |
2452 | /* |
2453 | * We assume that pages that could be isolated for | |
2454 | * migration are movable. But we don't actually try | |
2455 | * isolating, as that would be expensive. | |
2456 | */ | |
2457 | if (num_movable && | |
2458 | (PageLRU(page) || __PageMovable(page))) | |
2459 | (*num_movable)++; | |
39ddb991 | 2460 | pfn++; |
c361be55 MG |
2461 | continue; |
2462 | } | |
2463 | ||
cd961038 DR |
2464 | /* Make sure we are not inadvertently changing nodes */ |
2465 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2466 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2467 | ||
ab130f91 | 2468 | order = buddy_order(page); |
6ab01363 | 2469 | move_to_free_list(page, zone, order, migratetype); |
39ddb991 | 2470 | pfn += 1 << order; |
d100313f | 2471 | pages_moved += 1 << order; |
c361be55 MG |
2472 | } |
2473 | ||
d100313f | 2474 | return pages_moved; |
c361be55 MG |
2475 | } |
2476 | ||
ee6f509c | 2477 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2478 | int migratetype, int *num_movable) |
c361be55 | 2479 | { |
39ddb991 | 2480 | unsigned long start_pfn, end_pfn, pfn; |
c361be55 | 2481 | |
4a222127 DR |
2482 | if (num_movable) |
2483 | *num_movable = 0; | |
2484 | ||
39ddb991 KW |
2485 | pfn = page_to_pfn(page); |
2486 | start_pfn = pfn & ~(pageblock_nr_pages - 1); | |
d9c23400 | 2487 | end_pfn = start_pfn + pageblock_nr_pages - 1; |
c361be55 MG |
2488 | |
2489 | /* Do not cross zone boundaries */ | |
108bcc96 | 2490 | if (!zone_spans_pfn(zone, start_pfn)) |
39ddb991 | 2491 | start_pfn = pfn; |
108bcc96 | 2492 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2493 | return 0; |
2494 | ||
39ddb991 | 2495 | return move_freepages(zone, start_pfn, end_pfn, migratetype, |
02aa0cdd | 2496 | num_movable); |
c361be55 MG |
2497 | } |
2498 | ||
2f66a68f MG |
2499 | static void change_pageblock_range(struct page *pageblock_page, |
2500 | int start_order, int migratetype) | |
2501 | { | |
2502 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2503 | ||
2504 | while (nr_pageblocks--) { | |
2505 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2506 | pageblock_page += pageblock_nr_pages; | |
2507 | } | |
2508 | } | |
2509 | ||
fef903ef | 2510 | /* |
9c0415eb VB |
2511 | * When we are falling back to another migratetype during allocation, try to |
2512 | * steal extra free pages from the same pageblocks to satisfy further | |
2513 | * allocations, instead of polluting multiple pageblocks. | |
2514 | * | |
2515 | * If we are stealing a relatively large buddy page, it is likely there will | |
2516 | * be more free pages in the pageblock, so try to steal them all. For | |
2517 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2518 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2519 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2520 | * pageblocks. | |
fef903ef | 2521 | */ |
4eb7dce6 JK |
2522 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2523 | { | |
2524 | /* | |
2525 | * Leaving this order check is intended, although there is | |
2526 | * relaxed order check in next check. The reason is that | |
2527 | * we can actually steal whole pageblock if this condition met, | |
2528 | * but, below check doesn't guarantee it and that is just heuristic | |
2529 | * so could be changed anytime. | |
2530 | */ | |
2531 | if (order >= pageblock_order) | |
2532 | return true; | |
2533 | ||
2534 | if (order >= pageblock_order / 2 || | |
2535 | start_mt == MIGRATE_RECLAIMABLE || | |
2536 | start_mt == MIGRATE_UNMOVABLE || | |
2537 | page_group_by_mobility_disabled) | |
2538 | return true; | |
2539 | ||
2540 | return false; | |
2541 | } | |
2542 | ||
597c8920 | 2543 | static inline bool boost_watermark(struct zone *zone) |
1c30844d MG |
2544 | { |
2545 | unsigned long max_boost; | |
2546 | ||
2547 | if (!watermark_boost_factor) | |
597c8920 | 2548 | return false; |
14f69140 HW |
2549 | /* |
2550 | * Don't bother in zones that are unlikely to produce results. | |
2551 | * On small machines, including kdump capture kernels running | |
2552 | * in a small area, boosting the watermark can cause an out of | |
2553 | * memory situation immediately. | |
2554 | */ | |
2555 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
597c8920 | 2556 | return false; |
1c30844d MG |
2557 | |
2558 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2559 | watermark_boost_factor, 10000); | |
94b3334c MG |
2560 | |
2561 | /* | |
2562 | * high watermark may be uninitialised if fragmentation occurs | |
2563 | * very early in boot so do not boost. We do not fall | |
2564 | * through and boost by pageblock_nr_pages as failing | |
2565 | * allocations that early means that reclaim is not going | |
2566 | * to help and it may even be impossible to reclaim the | |
2567 | * boosted watermark resulting in a hang. | |
2568 | */ | |
2569 | if (!max_boost) | |
597c8920 | 2570 | return false; |
94b3334c | 2571 | |
1c30844d MG |
2572 | max_boost = max(pageblock_nr_pages, max_boost); |
2573 | ||
2574 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2575 | max_boost); | |
597c8920 JW |
2576 | |
2577 | return true; | |
1c30844d MG |
2578 | } |
2579 | ||
4eb7dce6 JK |
2580 | /* |
2581 | * This function implements actual steal behaviour. If order is large enough, | |
2582 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2583 | * pageblock to our migratetype and determine how many already-allocated pages |
2584 | * are there in the pageblock with a compatible migratetype. If at least half | |
2585 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2586 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2587 | */ |
2588 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2589 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2590 | { |
ab130f91 | 2591 | unsigned int current_order = buddy_order(page); |
02aa0cdd VB |
2592 | int free_pages, movable_pages, alike_pages; |
2593 | int old_block_type; | |
2594 | ||
2595 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2596 | |
3bc48f96 VB |
2597 | /* |
2598 | * This can happen due to races and we want to prevent broken | |
2599 | * highatomic accounting. | |
2600 | */ | |
02aa0cdd | 2601 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2602 | goto single_page; |
2603 | ||
fef903ef SB |
2604 | /* Take ownership for orders >= pageblock_order */ |
2605 | if (current_order >= pageblock_order) { | |
2606 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2607 | goto single_page; |
fef903ef SB |
2608 | } |
2609 | ||
1c30844d MG |
2610 | /* |
2611 | * Boost watermarks to increase reclaim pressure to reduce the | |
2612 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2613 | * may be balanced overall and kswapd will not wake naturally. | |
2614 | */ | |
597c8920 | 2615 | if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD)) |
73444bc4 | 2616 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2617 | |
3bc48f96 VB |
2618 | /* We are not allowed to try stealing from the whole block */ |
2619 | if (!whole_block) | |
2620 | goto single_page; | |
2621 | ||
02aa0cdd VB |
2622 | free_pages = move_freepages_block(zone, page, start_type, |
2623 | &movable_pages); | |
2624 | /* | |
2625 | * Determine how many pages are compatible with our allocation. | |
2626 | * For movable allocation, it's the number of movable pages which | |
2627 | * we just obtained. For other types it's a bit more tricky. | |
2628 | */ | |
2629 | if (start_type == MIGRATE_MOVABLE) { | |
2630 | alike_pages = movable_pages; | |
2631 | } else { | |
2632 | /* | |
2633 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2634 | * to MOVABLE pageblock, consider all non-movable pages as | |
2635 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2636 | * vice versa, be conservative since we can't distinguish the | |
2637 | * exact migratetype of non-movable pages. | |
2638 | */ | |
2639 | if (old_block_type == MIGRATE_MOVABLE) | |
2640 | alike_pages = pageblock_nr_pages | |
2641 | - (free_pages + movable_pages); | |
2642 | else | |
2643 | alike_pages = 0; | |
2644 | } | |
2645 | ||
3bc48f96 | 2646 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2647 | if (!free_pages) |
3bc48f96 | 2648 | goto single_page; |
fef903ef | 2649 | |
02aa0cdd VB |
2650 | /* |
2651 | * If a sufficient number of pages in the block are either free or of | |
2652 | * comparable migratability as our allocation, claim the whole block. | |
2653 | */ | |
2654 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2655 | page_group_by_mobility_disabled) |
2656 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2657 | |
2658 | return; | |
2659 | ||
2660 | single_page: | |
6ab01363 | 2661 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
2662 | } |
2663 | ||
2149cdae JK |
2664 | /* |
2665 | * Check whether there is a suitable fallback freepage with requested order. | |
2666 | * If only_stealable is true, this function returns fallback_mt only if | |
2667 | * we can steal other freepages all together. This would help to reduce | |
2668 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2669 | */ | |
2670 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2671 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2672 | { |
2673 | int i; | |
2674 | int fallback_mt; | |
2675 | ||
2676 | if (area->nr_free == 0) | |
2677 | return -1; | |
2678 | ||
2679 | *can_steal = false; | |
2680 | for (i = 0;; i++) { | |
2681 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2682 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2683 | break; |
2684 | ||
b03641af | 2685 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2686 | continue; |
fef903ef | 2687 | |
4eb7dce6 JK |
2688 | if (can_steal_fallback(order, migratetype)) |
2689 | *can_steal = true; | |
2690 | ||
2149cdae JK |
2691 | if (!only_stealable) |
2692 | return fallback_mt; | |
2693 | ||
2694 | if (*can_steal) | |
2695 | return fallback_mt; | |
fef903ef | 2696 | } |
4eb7dce6 JK |
2697 | |
2698 | return -1; | |
fef903ef SB |
2699 | } |
2700 | ||
0aaa29a5 MG |
2701 | /* |
2702 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2703 | * there are no empty page blocks that contain a page with a suitable order | |
2704 | */ | |
2705 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2706 | unsigned int alloc_order) | |
2707 | { | |
2708 | int mt; | |
2709 | unsigned long max_managed, flags; | |
2710 | ||
2711 | /* | |
2712 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2713 | * Check is race-prone but harmless. | |
2714 | */ | |
9705bea5 | 2715 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2716 | if (zone->nr_reserved_highatomic >= max_managed) |
2717 | return; | |
2718 | ||
2719 | spin_lock_irqsave(&zone->lock, flags); | |
2720 | ||
2721 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2722 | if (zone->nr_reserved_highatomic >= max_managed) | |
2723 | goto out_unlock; | |
2724 | ||
2725 | /* Yoink! */ | |
2726 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2727 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2728 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2729 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2730 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2731 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2732 | } |
2733 | ||
2734 | out_unlock: | |
2735 | spin_unlock_irqrestore(&zone->lock, flags); | |
2736 | } | |
2737 | ||
2738 | /* | |
2739 | * Used when an allocation is about to fail under memory pressure. This | |
2740 | * potentially hurts the reliability of high-order allocations when under | |
2741 | * intense memory pressure but failed atomic allocations should be easier | |
2742 | * to recover from than an OOM. | |
29fac03b MK |
2743 | * |
2744 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2745 | * pageblock is exhausted. | |
0aaa29a5 | 2746 | */ |
29fac03b MK |
2747 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2748 | bool force) | |
0aaa29a5 MG |
2749 | { |
2750 | struct zonelist *zonelist = ac->zonelist; | |
2751 | unsigned long flags; | |
2752 | struct zoneref *z; | |
2753 | struct zone *zone; | |
2754 | struct page *page; | |
2755 | int order; | |
04c8716f | 2756 | bool ret; |
0aaa29a5 | 2757 | |
97a225e6 | 2758 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2759 | ac->nodemask) { |
29fac03b MK |
2760 | /* |
2761 | * Preserve at least one pageblock unless memory pressure | |
2762 | * is really high. | |
2763 | */ | |
2764 | if (!force && zone->nr_reserved_highatomic <= | |
2765 | pageblock_nr_pages) | |
0aaa29a5 MG |
2766 | continue; |
2767 | ||
2768 | spin_lock_irqsave(&zone->lock, flags); | |
2769 | for (order = 0; order < MAX_ORDER; order++) { | |
2770 | struct free_area *area = &(zone->free_area[order]); | |
2771 | ||
b03641af | 2772 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2773 | if (!page) |
0aaa29a5 MG |
2774 | continue; |
2775 | ||
0aaa29a5 | 2776 | /* |
4855e4a7 MK |
2777 | * In page freeing path, migratetype change is racy so |
2778 | * we can counter several free pages in a pageblock | |
f0953a1b | 2779 | * in this loop although we changed the pageblock type |
4855e4a7 MK |
2780 | * from highatomic to ac->migratetype. So we should |
2781 | * adjust the count once. | |
0aaa29a5 | 2782 | */ |
a6ffdc07 | 2783 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2784 | /* |
2785 | * It should never happen but changes to | |
2786 | * locking could inadvertently allow a per-cpu | |
2787 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2788 | * while unreserving so be safe and watch for | |
2789 | * underflows. | |
2790 | */ | |
2791 | zone->nr_reserved_highatomic -= min( | |
2792 | pageblock_nr_pages, | |
2793 | zone->nr_reserved_highatomic); | |
2794 | } | |
0aaa29a5 MG |
2795 | |
2796 | /* | |
2797 | * Convert to ac->migratetype and avoid the normal | |
2798 | * pageblock stealing heuristics. Minimally, the caller | |
2799 | * is doing the work and needs the pages. More | |
2800 | * importantly, if the block was always converted to | |
2801 | * MIGRATE_UNMOVABLE or another type then the number | |
2802 | * of pageblocks that cannot be completely freed | |
2803 | * may increase. | |
2804 | */ | |
2805 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2806 | ret = move_freepages_block(zone, page, ac->migratetype, |
2807 | NULL); | |
29fac03b MK |
2808 | if (ret) { |
2809 | spin_unlock_irqrestore(&zone->lock, flags); | |
2810 | return ret; | |
2811 | } | |
0aaa29a5 MG |
2812 | } |
2813 | spin_unlock_irqrestore(&zone->lock, flags); | |
2814 | } | |
04c8716f MK |
2815 | |
2816 | return false; | |
0aaa29a5 MG |
2817 | } |
2818 | ||
3bc48f96 VB |
2819 | /* |
2820 | * Try finding a free buddy page on the fallback list and put it on the free | |
2821 | * list of requested migratetype, possibly along with other pages from the same | |
2822 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2823 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2824 | * |
2825 | * The use of signed ints for order and current_order is a deliberate | |
2826 | * deviation from the rest of this file, to make the for loop | |
2827 | * condition simpler. | |
3bc48f96 | 2828 | */ |
85ccc8fa | 2829 | static __always_inline bool |
6bb15450 MG |
2830 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2831 | unsigned int alloc_flags) | |
b2a0ac88 | 2832 | { |
b8af2941 | 2833 | struct free_area *area; |
b002529d | 2834 | int current_order; |
6bb15450 | 2835 | int min_order = order; |
b2a0ac88 | 2836 | struct page *page; |
4eb7dce6 JK |
2837 | int fallback_mt; |
2838 | bool can_steal; | |
b2a0ac88 | 2839 | |
6bb15450 MG |
2840 | /* |
2841 | * Do not steal pages from freelists belonging to other pageblocks | |
2842 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2843 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2844 | */ | |
2845 | if (alloc_flags & ALLOC_NOFRAGMENT) | |
2846 | min_order = pageblock_order; | |
2847 | ||
7a8f58f3 VB |
2848 | /* |
2849 | * Find the largest available free page in the other list. This roughly | |
2850 | * approximates finding the pageblock with the most free pages, which | |
2851 | * would be too costly to do exactly. | |
2852 | */ | |
6bb15450 | 2853 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 2854 | --current_order) { |
4eb7dce6 JK |
2855 | area = &(zone->free_area[current_order]); |
2856 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2857 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2858 | if (fallback_mt == -1) |
2859 | continue; | |
b2a0ac88 | 2860 | |
7a8f58f3 VB |
2861 | /* |
2862 | * We cannot steal all free pages from the pageblock and the | |
2863 | * requested migratetype is movable. In that case it's better to | |
2864 | * steal and split the smallest available page instead of the | |
2865 | * largest available page, because even if the next movable | |
2866 | * allocation falls back into a different pageblock than this | |
2867 | * one, it won't cause permanent fragmentation. | |
2868 | */ | |
2869 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2870 | && current_order > order) | |
2871 | goto find_smallest; | |
b2a0ac88 | 2872 | |
7a8f58f3 VB |
2873 | goto do_steal; |
2874 | } | |
e0fff1bd | 2875 | |
7a8f58f3 | 2876 | return false; |
e0fff1bd | 2877 | |
7a8f58f3 VB |
2878 | find_smallest: |
2879 | for (current_order = order; current_order < MAX_ORDER; | |
2880 | current_order++) { | |
2881 | area = &(zone->free_area[current_order]); | |
2882 | fallback_mt = find_suitable_fallback(area, current_order, | |
2883 | start_migratetype, false, &can_steal); | |
2884 | if (fallback_mt != -1) | |
2885 | break; | |
b2a0ac88 MG |
2886 | } |
2887 | ||
7a8f58f3 VB |
2888 | /* |
2889 | * This should not happen - we already found a suitable fallback | |
2890 | * when looking for the largest page. | |
2891 | */ | |
2892 | VM_BUG_ON(current_order == MAX_ORDER); | |
2893 | ||
2894 | do_steal: | |
b03641af | 2895 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2896 | |
1c30844d MG |
2897 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
2898 | can_steal); | |
7a8f58f3 VB |
2899 | |
2900 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2901 | start_migratetype, fallback_mt); | |
2902 | ||
2903 | return true; | |
2904 | ||
b2a0ac88 MG |
2905 | } |
2906 | ||
56fd56b8 | 2907 | /* |
1da177e4 LT |
2908 | * Do the hard work of removing an element from the buddy allocator. |
2909 | * Call me with the zone->lock already held. | |
2910 | */ | |
85ccc8fa | 2911 | static __always_inline struct page * |
6bb15450 MG |
2912 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2913 | unsigned int alloc_flags) | |
1da177e4 | 2914 | { |
1da177e4 LT |
2915 | struct page *page; |
2916 | ||
ce8f86ee H |
2917 | if (IS_ENABLED(CONFIG_CMA)) { |
2918 | /* | |
2919 | * Balance movable allocations between regular and CMA areas by | |
2920 | * allocating from CMA when over half of the zone's free memory | |
2921 | * is in the CMA area. | |
2922 | */ | |
2923 | if (alloc_flags & ALLOC_CMA && | |
2924 | zone_page_state(zone, NR_FREE_CMA_PAGES) > | |
2925 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
2926 | page = __rmqueue_cma_fallback(zone, order); | |
2927 | if (page) | |
2928 | goto out; | |
2929 | } | |
16867664 | 2930 | } |
3bc48f96 | 2931 | retry: |
56fd56b8 | 2932 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2933 | if (unlikely(!page)) { |
8510e69c | 2934 | if (alloc_flags & ALLOC_CMA) |
dc67647b JK |
2935 | page = __rmqueue_cma_fallback(zone, order); |
2936 | ||
6bb15450 MG |
2937 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2938 | alloc_flags)) | |
3bc48f96 | 2939 | goto retry; |
728ec980 | 2940 | } |
ce8f86ee H |
2941 | out: |
2942 | if (page) | |
2943 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
b2a0ac88 | 2944 | return page; |
1da177e4 LT |
2945 | } |
2946 | ||
5f63b720 | 2947 | /* |
1da177e4 LT |
2948 | * Obtain a specified number of elements from the buddy allocator, all under |
2949 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2950 | * Returns the number of new pages which were placed at *list. | |
2951 | */ | |
5f63b720 | 2952 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2953 | unsigned long count, struct list_head *list, |
6bb15450 | 2954 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2955 | { |
cb66bede | 2956 | int i, allocated = 0; |
5f63b720 | 2957 | |
d34b0733 | 2958 | spin_lock(&zone->lock); |
1da177e4 | 2959 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2960 | struct page *page = __rmqueue(zone, order, migratetype, |
2961 | alloc_flags); | |
085cc7d5 | 2962 | if (unlikely(page == NULL)) |
1da177e4 | 2963 | break; |
81eabcbe | 2964 | |
479f854a MG |
2965 | if (unlikely(check_pcp_refill(page))) |
2966 | continue; | |
2967 | ||
81eabcbe | 2968 | /* |
0fac3ba5 VB |
2969 | * Split buddy pages returned by expand() are received here in |
2970 | * physical page order. The page is added to the tail of | |
2971 | * caller's list. From the callers perspective, the linked list | |
2972 | * is ordered by page number under some conditions. This is | |
2973 | * useful for IO devices that can forward direction from the | |
2974 | * head, thus also in the physical page order. This is useful | |
2975 | * for IO devices that can merge IO requests if the physical | |
2976 | * pages are ordered properly. | |
81eabcbe | 2977 | */ |
0fac3ba5 | 2978 | list_add_tail(&page->lru, list); |
cb66bede | 2979 | allocated++; |
bb14c2c7 | 2980 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2981 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2982 | -(1 << order)); | |
1da177e4 | 2983 | } |
a6de734b MG |
2984 | |
2985 | /* | |
2986 | * i pages were removed from the buddy list even if some leak due | |
2987 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
cb66bede | 2988 | * on i. Do not confuse with 'allocated' which is the number of |
a6de734b MG |
2989 | * pages added to the pcp list. |
2990 | */ | |
f2260e6b | 2991 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2992 | spin_unlock(&zone->lock); |
cb66bede | 2993 | return allocated; |
1da177e4 LT |
2994 | } |
2995 | ||
4ae7c039 | 2996 | #ifdef CONFIG_NUMA |
8fce4d8e | 2997 | /* |
4037d452 CL |
2998 | * Called from the vmstat counter updater to drain pagesets of this |
2999 | * currently executing processor on remote nodes after they have | |
3000 | * expired. | |
3001 | * | |
879336c3 CL |
3002 | * Note that this function must be called with the thread pinned to |
3003 | * a single processor. | |
8fce4d8e | 3004 | */ |
4037d452 | 3005 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 3006 | { |
4ae7c039 | 3007 | unsigned long flags; |
7be12fc9 | 3008 | int to_drain, batch; |
4ae7c039 | 3009 | |
4037d452 | 3010 | local_irq_save(flags); |
4db0c3c2 | 3011 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 3012 | to_drain = min(pcp->count, batch); |
77ba9062 | 3013 | if (to_drain > 0) |
2a13515c | 3014 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 3015 | local_irq_restore(flags); |
4ae7c039 CL |
3016 | } |
3017 | #endif | |
3018 | ||
9f8f2172 | 3019 | /* |
93481ff0 | 3020 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
3021 | * |
3022 | * The processor must either be the current processor and the | |
3023 | * thread pinned to the current processor or a processor that | |
3024 | * is not online. | |
3025 | */ | |
93481ff0 | 3026 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 3027 | { |
c54ad30c | 3028 | unsigned long flags; |
93481ff0 VB |
3029 | struct per_cpu_pageset *pset; |
3030 | struct per_cpu_pages *pcp; | |
1da177e4 | 3031 | |
93481ff0 VB |
3032 | local_irq_save(flags); |
3033 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 3034 | |
93481ff0 | 3035 | pcp = &pset->pcp; |
77ba9062 | 3036 | if (pcp->count) |
93481ff0 | 3037 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
3038 | local_irq_restore(flags); |
3039 | } | |
3dfa5721 | 3040 | |
93481ff0 VB |
3041 | /* |
3042 | * Drain pcplists of all zones on the indicated processor. | |
3043 | * | |
3044 | * The processor must either be the current processor and the | |
3045 | * thread pinned to the current processor or a processor that | |
3046 | * is not online. | |
3047 | */ | |
3048 | static void drain_pages(unsigned int cpu) | |
3049 | { | |
3050 | struct zone *zone; | |
3051 | ||
3052 | for_each_populated_zone(zone) { | |
3053 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
3054 | } |
3055 | } | |
1da177e4 | 3056 | |
9f8f2172 CL |
3057 | /* |
3058 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
3059 | * |
3060 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
3061 | * the single zone's pages. | |
9f8f2172 | 3062 | */ |
93481ff0 | 3063 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 3064 | { |
93481ff0 VB |
3065 | int cpu = smp_processor_id(); |
3066 | ||
3067 | if (zone) | |
3068 | drain_pages_zone(cpu, zone); | |
3069 | else | |
3070 | drain_pages(cpu); | |
9f8f2172 CL |
3071 | } |
3072 | ||
0ccce3b9 MG |
3073 | static void drain_local_pages_wq(struct work_struct *work) |
3074 | { | |
d9367bd0 WY |
3075 | struct pcpu_drain *drain; |
3076 | ||
3077 | drain = container_of(work, struct pcpu_drain, work); | |
3078 | ||
a459eeb7 MH |
3079 | /* |
3080 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
3081 | * we can race with cpu offline when the WQ can move this from | |
3082 | * a cpu pinned worker to an unbound one. We can operate on a different | |
f0953a1b | 3083 | * cpu which is alright but we also have to make sure to not move to |
a459eeb7 MH |
3084 | * a different one. |
3085 | */ | |
3086 | preempt_disable(); | |
d9367bd0 | 3087 | drain_local_pages(drain->zone); |
a459eeb7 | 3088 | preempt_enable(); |
0ccce3b9 MG |
3089 | } |
3090 | ||
9f8f2172 | 3091 | /* |
ec6e8c7e VB |
3092 | * The implementation of drain_all_pages(), exposing an extra parameter to |
3093 | * drain on all cpus. | |
93481ff0 | 3094 | * |
ec6e8c7e VB |
3095 | * drain_all_pages() is optimized to only execute on cpus where pcplists are |
3096 | * not empty. The check for non-emptiness can however race with a free to | |
3097 | * pcplist that has not yet increased the pcp->count from 0 to 1. Callers | |
3098 | * that need the guarantee that every CPU has drained can disable the | |
3099 | * optimizing racy check. | |
9f8f2172 | 3100 | */ |
3b1f3658 | 3101 | static void __drain_all_pages(struct zone *zone, bool force_all_cpus) |
9f8f2172 | 3102 | { |
74046494 | 3103 | int cpu; |
74046494 GBY |
3104 | |
3105 | /* | |
3106 | * Allocate in the BSS so we wont require allocation in | |
3107 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
3108 | */ | |
3109 | static cpumask_t cpus_with_pcps; | |
3110 | ||
ce612879 MH |
3111 | /* |
3112 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
3113 | * initialized. | |
3114 | */ | |
3115 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
3116 | return; | |
3117 | ||
bd233f53 MG |
3118 | /* |
3119 | * Do not drain if one is already in progress unless it's specific to | |
3120 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
3121 | * the drain to be complete when the call returns. | |
3122 | */ | |
3123 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
3124 | if (!zone) | |
3125 | return; | |
3126 | mutex_lock(&pcpu_drain_mutex); | |
3127 | } | |
0ccce3b9 | 3128 | |
74046494 GBY |
3129 | /* |
3130 | * We don't care about racing with CPU hotplug event | |
3131 | * as offline notification will cause the notified | |
3132 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
3133 | * disables preemption as part of its processing | |
3134 | */ | |
3135 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
3136 | struct per_cpu_pageset *pcp; |
3137 | struct zone *z; | |
74046494 | 3138 | bool has_pcps = false; |
93481ff0 | 3139 | |
ec6e8c7e VB |
3140 | if (force_all_cpus) { |
3141 | /* | |
3142 | * The pcp.count check is racy, some callers need a | |
3143 | * guarantee that no cpu is missed. | |
3144 | */ | |
3145 | has_pcps = true; | |
3146 | } else if (zone) { | |
74046494 | 3147 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 3148 | if (pcp->pcp.count) |
74046494 | 3149 | has_pcps = true; |
93481ff0 VB |
3150 | } else { |
3151 | for_each_populated_zone(z) { | |
3152 | pcp = per_cpu_ptr(z->pageset, cpu); | |
3153 | if (pcp->pcp.count) { | |
3154 | has_pcps = true; | |
3155 | break; | |
3156 | } | |
74046494 GBY |
3157 | } |
3158 | } | |
93481ff0 | 3159 | |
74046494 GBY |
3160 | if (has_pcps) |
3161 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
3162 | else | |
3163 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
3164 | } | |
0ccce3b9 | 3165 | |
bd233f53 | 3166 | for_each_cpu(cpu, &cpus_with_pcps) { |
d9367bd0 WY |
3167 | struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu); |
3168 | ||
3169 | drain->zone = zone; | |
3170 | INIT_WORK(&drain->work, drain_local_pages_wq); | |
3171 | queue_work_on(cpu, mm_percpu_wq, &drain->work); | |
0ccce3b9 | 3172 | } |
bd233f53 | 3173 | for_each_cpu(cpu, &cpus_with_pcps) |
d9367bd0 | 3174 | flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work); |
bd233f53 MG |
3175 | |
3176 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
3177 | } |
3178 | ||
ec6e8c7e VB |
3179 | /* |
3180 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. | |
3181 | * | |
3182 | * When zone parameter is non-NULL, spill just the single zone's pages. | |
3183 | * | |
3184 | * Note that this can be extremely slow as the draining happens in a workqueue. | |
3185 | */ | |
3186 | void drain_all_pages(struct zone *zone) | |
3187 | { | |
3188 | __drain_all_pages(zone, false); | |
3189 | } | |
3190 | ||
296699de | 3191 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3192 | |
556b969a CY |
3193 | /* |
3194 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
3195 | */ | |
3196 | #define WD_PAGE_COUNT (128*1024) | |
3197 | ||
1da177e4 LT |
3198 | void mark_free_pages(struct zone *zone) |
3199 | { | |
556b969a | 3200 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 3201 | unsigned long flags; |
7aeb09f9 | 3202 | unsigned int order, t; |
86760a2c | 3203 | struct page *page; |
1da177e4 | 3204 | |
8080fc03 | 3205 | if (zone_is_empty(zone)) |
1da177e4 LT |
3206 | return; |
3207 | ||
3208 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 3209 | |
108bcc96 | 3210 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
3211 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
3212 | if (pfn_valid(pfn)) { | |
86760a2c | 3213 | page = pfn_to_page(pfn); |
ba6b0979 | 3214 | |
556b969a CY |
3215 | if (!--page_count) { |
3216 | touch_nmi_watchdog(); | |
3217 | page_count = WD_PAGE_COUNT; | |
3218 | } | |
3219 | ||
ba6b0979 JK |
3220 | if (page_zone(page) != zone) |
3221 | continue; | |
3222 | ||
7be98234 RW |
3223 | if (!swsusp_page_is_forbidden(page)) |
3224 | swsusp_unset_page_free(page); | |
f623f0db | 3225 | } |
1da177e4 | 3226 | |
b2a0ac88 | 3227 | for_each_migratetype_order(order, t) { |
86760a2c GT |
3228 | list_for_each_entry(page, |
3229 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 3230 | unsigned long i; |
1da177e4 | 3231 | |
86760a2c | 3232 | pfn = page_to_pfn(page); |
556b969a CY |
3233 | for (i = 0; i < (1UL << order); i++) { |
3234 | if (!--page_count) { | |
3235 | touch_nmi_watchdog(); | |
3236 | page_count = WD_PAGE_COUNT; | |
3237 | } | |
7be98234 | 3238 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 3239 | } |
f623f0db | 3240 | } |
b2a0ac88 | 3241 | } |
1da177e4 LT |
3242 | spin_unlock_irqrestore(&zone->lock, flags); |
3243 | } | |
e2c55dc8 | 3244 | #endif /* CONFIG_PM */ |
1da177e4 | 3245 | |
2d4894b5 | 3246 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 3247 | { |
5f8dcc21 | 3248 | int migratetype; |
1da177e4 | 3249 | |
4db7548c | 3250 | if (!free_pcp_prepare(page)) |
9cca35d4 | 3251 | return false; |
689bcebf | 3252 | |
dc4b0caf | 3253 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 3254 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
3255 | return true; |
3256 | } | |
3257 | ||
2d4894b5 | 3258 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
3259 | { |
3260 | struct zone *zone = page_zone(page); | |
3261 | struct per_cpu_pages *pcp; | |
3262 | int migratetype; | |
3263 | ||
3264 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 3265 | __count_vm_event(PGFREE); |
da456f14 | 3266 | |
5f8dcc21 MG |
3267 | /* |
3268 | * We only track unmovable, reclaimable and movable on pcp lists. | |
3269 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 3270 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
3271 | * areas back if necessary. Otherwise, we may have to free |
3272 | * excessively into the page allocator | |
3273 | */ | |
3274 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 3275 | if (unlikely(is_migrate_isolate(migratetype))) { |
7fef431b DH |
3276 | free_one_page(zone, page, pfn, 0, migratetype, |
3277 | FPI_NONE); | |
9cca35d4 | 3278 | return; |
5f8dcc21 MG |
3279 | } |
3280 | migratetype = MIGRATE_MOVABLE; | |
3281 | } | |
3282 | ||
99dcc3e5 | 3283 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 3284 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 3285 | pcp->count++; |
5c3ad2eb VB |
3286 | if (pcp->count >= READ_ONCE(pcp->high)) |
3287 | free_pcppages_bulk(zone, READ_ONCE(pcp->batch), pcp); | |
9cca35d4 | 3288 | } |
5f8dcc21 | 3289 | |
9cca35d4 MG |
3290 | /* |
3291 | * Free a 0-order page | |
9cca35d4 | 3292 | */ |
2d4894b5 | 3293 | void free_unref_page(struct page *page) |
9cca35d4 MG |
3294 | { |
3295 | unsigned long flags; | |
3296 | unsigned long pfn = page_to_pfn(page); | |
3297 | ||
2d4894b5 | 3298 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3299 | return; |
3300 | ||
3301 | local_irq_save(flags); | |
2d4894b5 | 3302 | free_unref_page_commit(page, pfn); |
d34b0733 | 3303 | local_irq_restore(flags); |
1da177e4 LT |
3304 | } |
3305 | ||
cc59850e KK |
3306 | /* |
3307 | * Free a list of 0-order pages | |
3308 | */ | |
2d4894b5 | 3309 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
3310 | { |
3311 | struct page *page, *next; | |
9cca35d4 | 3312 | unsigned long flags, pfn; |
c24ad77d | 3313 | int batch_count = 0; |
9cca35d4 MG |
3314 | |
3315 | /* Prepare pages for freeing */ | |
3316 | list_for_each_entry_safe(page, next, list, lru) { | |
3317 | pfn = page_to_pfn(page); | |
2d4894b5 | 3318 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3319 | list_del(&page->lru); |
3320 | set_page_private(page, pfn); | |
3321 | } | |
cc59850e | 3322 | |
9cca35d4 | 3323 | local_irq_save(flags); |
cc59850e | 3324 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
3325 | unsigned long pfn = page_private(page); |
3326 | ||
3327 | set_page_private(page, 0); | |
2d4894b5 MG |
3328 | trace_mm_page_free_batched(page); |
3329 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
3330 | |
3331 | /* | |
3332 | * Guard against excessive IRQ disabled times when we get | |
3333 | * a large list of pages to free. | |
3334 | */ | |
3335 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
3336 | local_irq_restore(flags); | |
3337 | batch_count = 0; | |
3338 | local_irq_save(flags); | |
3339 | } | |
cc59850e | 3340 | } |
9cca35d4 | 3341 | local_irq_restore(flags); |
cc59850e KK |
3342 | } |
3343 | ||
8dfcc9ba NP |
3344 | /* |
3345 | * split_page takes a non-compound higher-order page, and splits it into | |
3346 | * n (1<<order) sub-pages: page[0..n] | |
3347 | * Each sub-page must be freed individually. | |
3348 | * | |
3349 | * Note: this is probably too low level an operation for use in drivers. | |
3350 | * Please consult with lkml before using this in your driver. | |
3351 | */ | |
3352 | void split_page(struct page *page, unsigned int order) | |
3353 | { | |
3354 | int i; | |
3355 | ||
309381fe SL |
3356 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3357 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3358 | |
a9627bc5 | 3359 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3360 | set_page_refcounted(page + i); |
8fb156c9 | 3361 | split_page_owner(page, 1 << order); |
e1baddf8 | 3362 | split_page_memcg(page, 1 << order); |
8dfcc9ba | 3363 | } |
5853ff23 | 3364 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3365 | |
3c605096 | 3366 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3367 | { |
748446bb MG |
3368 | unsigned long watermark; |
3369 | struct zone *zone; | |
2139cbe6 | 3370 | int mt; |
748446bb MG |
3371 | |
3372 | BUG_ON(!PageBuddy(page)); | |
3373 | ||
3374 | zone = page_zone(page); | |
2e30abd1 | 3375 | mt = get_pageblock_migratetype(page); |
748446bb | 3376 | |
194159fb | 3377 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
3378 | /* |
3379 | * Obey watermarks as if the page was being allocated. We can | |
3380 | * emulate a high-order watermark check with a raised order-0 | |
3381 | * watermark, because we already know our high-order page | |
3382 | * exists. | |
3383 | */ | |
fd1444b2 | 3384 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3385 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3386 | return 0; |
3387 | ||
8fb74b9f | 3388 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3389 | } |
748446bb MG |
3390 | |
3391 | /* Remove page from free list */ | |
b03641af | 3392 | |
6ab01363 | 3393 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 3394 | |
400bc7fd | 3395 | /* |
3396 | * Set the pageblock if the isolated page is at least half of a | |
3397 | * pageblock | |
3398 | */ | |
748446bb MG |
3399 | if (order >= pageblock_order - 1) { |
3400 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3401 | for (; page < endpage; page += pageblock_nr_pages) { |
3402 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 3403 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 3404 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
3405 | set_pageblock_migratetype(page, |
3406 | MIGRATE_MOVABLE); | |
3407 | } | |
748446bb MG |
3408 | } |
3409 | ||
f3a14ced | 3410 | |
8fb74b9f | 3411 | return 1UL << order; |
1fb3f8ca MG |
3412 | } |
3413 | ||
624f58d8 AD |
3414 | /** |
3415 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
3416 | * @page: Page that was isolated | |
3417 | * @order: Order of the isolated page | |
e6a0a7ad | 3418 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
3419 | * |
3420 | * This function is meant to return a page pulled from the free lists via | |
3421 | * __isolate_free_page back to the free lists they were pulled from. | |
3422 | */ | |
3423 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
3424 | { | |
3425 | struct zone *zone = page_zone(page); | |
3426 | ||
3427 | /* zone lock should be held when this function is called */ | |
3428 | lockdep_assert_held(&zone->lock); | |
3429 | ||
3430 | /* Return isolated page to tail of freelist. */ | |
f04a5d5d | 3431 | __free_one_page(page, page_to_pfn(page), zone, order, mt, |
47b6a24a | 3432 | FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL); |
624f58d8 AD |
3433 | } |
3434 | ||
060e7417 MG |
3435 | /* |
3436 | * Update NUMA hit/miss statistics | |
3437 | * | |
3438 | * Must be called with interrupts disabled. | |
060e7417 | 3439 | */ |
41b6167e | 3440 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
3441 | { |
3442 | #ifdef CONFIG_NUMA | |
3a321d2a | 3443 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3444 | |
4518085e KW |
3445 | /* skip numa counters update if numa stats is disabled */ |
3446 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3447 | return; | |
3448 | ||
c1093b74 | 3449 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3450 | local_stat = NUMA_OTHER; |
060e7417 | 3451 | |
c1093b74 | 3452 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3a321d2a | 3453 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 3454 | else { |
3a321d2a KW |
3455 | __inc_numa_state(z, NUMA_MISS); |
3456 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 3457 | } |
3a321d2a | 3458 | __inc_numa_state(z, local_stat); |
060e7417 MG |
3459 | #endif |
3460 | } | |
3461 | ||
066b2393 | 3462 | /* Remove page from the per-cpu list, caller must protect the list */ |
3b822017 JDB |
3463 | static inline |
3464 | struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
6bb15450 | 3465 | unsigned int alloc_flags, |
453f85d4 | 3466 | struct per_cpu_pages *pcp, |
066b2393 MG |
3467 | struct list_head *list) |
3468 | { | |
3469 | struct page *page; | |
3470 | ||
3471 | do { | |
3472 | if (list_empty(list)) { | |
3473 | pcp->count += rmqueue_bulk(zone, 0, | |
5c3ad2eb | 3474 | READ_ONCE(pcp->batch), list, |
6bb15450 | 3475 | migratetype, alloc_flags); |
066b2393 MG |
3476 | if (unlikely(list_empty(list))) |
3477 | return NULL; | |
3478 | } | |
3479 | ||
453f85d4 | 3480 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
3481 | list_del(&page->lru); |
3482 | pcp->count--; | |
3483 | } while (check_new_pcp(page)); | |
3484 | ||
3485 | return page; | |
3486 | } | |
3487 | ||
3488 | /* Lock and remove page from the per-cpu list */ | |
3489 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
1c52e6d0 YS |
3490 | struct zone *zone, gfp_t gfp_flags, |
3491 | int migratetype, unsigned int alloc_flags) | |
066b2393 MG |
3492 | { |
3493 | struct per_cpu_pages *pcp; | |
3494 | struct list_head *list; | |
066b2393 | 3495 | struct page *page; |
d34b0733 | 3496 | unsigned long flags; |
066b2393 | 3497 | |
d34b0733 | 3498 | local_irq_save(flags); |
066b2393 MG |
3499 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3500 | list = &pcp->lists[migratetype]; | |
6bb15450 | 3501 | page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list); |
066b2393 | 3502 | if (page) { |
1c52e6d0 | 3503 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1); |
066b2393 MG |
3504 | zone_statistics(preferred_zone, zone); |
3505 | } | |
d34b0733 | 3506 | local_irq_restore(flags); |
066b2393 MG |
3507 | return page; |
3508 | } | |
3509 | ||
1da177e4 | 3510 | /* |
75379191 | 3511 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 3512 | */ |
0a15c3e9 | 3513 | static inline |
066b2393 | 3514 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3515 | struct zone *zone, unsigned int order, |
c603844b MG |
3516 | gfp_t gfp_flags, unsigned int alloc_flags, |
3517 | int migratetype) | |
1da177e4 LT |
3518 | { |
3519 | unsigned long flags; | |
689bcebf | 3520 | struct page *page; |
1da177e4 | 3521 | |
d34b0733 | 3522 | if (likely(order == 0)) { |
1d91df85 JK |
3523 | /* |
3524 | * MIGRATE_MOVABLE pcplist could have the pages on CMA area and | |
3525 | * we need to skip it when CMA area isn't allowed. | |
3526 | */ | |
3527 | if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA || | |
3528 | migratetype != MIGRATE_MOVABLE) { | |
3529 | page = rmqueue_pcplist(preferred_zone, zone, gfp_flags, | |
1c52e6d0 | 3530 | migratetype, alloc_flags); |
1d91df85 JK |
3531 | goto out; |
3532 | } | |
066b2393 | 3533 | } |
83b9355b | 3534 | |
066b2393 MG |
3535 | /* |
3536 | * We most definitely don't want callers attempting to | |
3537 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3538 | */ | |
3539 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3540 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3541 | |
066b2393 MG |
3542 | do { |
3543 | page = NULL; | |
1d91df85 JK |
3544 | /* |
3545 | * order-0 request can reach here when the pcplist is skipped | |
3546 | * due to non-CMA allocation context. HIGHATOMIC area is | |
3547 | * reserved for high-order atomic allocation, so order-0 | |
3548 | * request should skip it. | |
3549 | */ | |
3550 | if (order > 0 && alloc_flags & ALLOC_HARDER) { | |
066b2393 MG |
3551 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
3552 | if (page) | |
3553 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3554 | } | |
a74609fa | 3555 | if (!page) |
6bb15450 | 3556 | page = __rmqueue(zone, order, migratetype, alloc_flags); |
066b2393 MG |
3557 | } while (page && check_new_pages(page, order)); |
3558 | spin_unlock(&zone->lock); | |
3559 | if (!page) | |
3560 | goto failed; | |
3561 | __mod_zone_freepage_state(zone, -(1 << order), | |
3562 | get_pcppage_migratetype(page)); | |
1da177e4 | 3563 | |
16709d1d | 3564 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3565 | zone_statistics(preferred_zone, zone); |
a74609fa | 3566 | local_irq_restore(flags); |
1da177e4 | 3567 | |
066b2393 | 3568 | out: |
73444bc4 MG |
3569 | /* Separate test+clear to avoid unnecessary atomics */ |
3570 | if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) { | |
3571 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); | |
3572 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3573 | } | |
3574 | ||
066b2393 | 3575 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 | 3576 | return page; |
a74609fa NP |
3577 | |
3578 | failed: | |
3579 | local_irq_restore(flags); | |
a74609fa | 3580 | return NULL; |
1da177e4 LT |
3581 | } |
3582 | ||
933e312e AM |
3583 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3584 | ||
b2588c4b | 3585 | static struct { |
933e312e AM |
3586 | struct fault_attr attr; |
3587 | ||
621a5f7a | 3588 | bool ignore_gfp_highmem; |
71baba4b | 3589 | bool ignore_gfp_reclaim; |
54114994 | 3590 | u32 min_order; |
933e312e AM |
3591 | } fail_page_alloc = { |
3592 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3593 | .ignore_gfp_reclaim = true, |
621a5f7a | 3594 | .ignore_gfp_highmem = true, |
54114994 | 3595 | .min_order = 1, |
933e312e AM |
3596 | }; |
3597 | ||
3598 | static int __init setup_fail_page_alloc(char *str) | |
3599 | { | |
3600 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3601 | } | |
3602 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3603 | ||
af3b8544 | 3604 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3605 | { |
54114994 | 3606 | if (order < fail_page_alloc.min_order) |
deaf386e | 3607 | return false; |
933e312e | 3608 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3609 | return false; |
933e312e | 3610 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3611 | return false; |
71baba4b MG |
3612 | if (fail_page_alloc.ignore_gfp_reclaim && |
3613 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3614 | return false; |
933e312e AM |
3615 | |
3616 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3617 | } | |
3618 | ||
3619 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3620 | ||
3621 | static int __init fail_page_alloc_debugfs(void) | |
3622 | { | |
0825a6f9 | 3623 | umode_t mode = S_IFREG | 0600; |
933e312e | 3624 | struct dentry *dir; |
933e312e | 3625 | |
dd48c085 AM |
3626 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3627 | &fail_page_alloc.attr); | |
b2588c4b | 3628 | |
d9f7979c GKH |
3629 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3630 | &fail_page_alloc.ignore_gfp_reclaim); | |
3631 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3632 | &fail_page_alloc.ignore_gfp_highmem); | |
3633 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3634 | |
d9f7979c | 3635 | return 0; |
933e312e AM |
3636 | } |
3637 | ||
3638 | late_initcall(fail_page_alloc_debugfs); | |
3639 | ||
3640 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3641 | ||
3642 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3643 | ||
af3b8544 | 3644 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3645 | { |
deaf386e | 3646 | return false; |
933e312e AM |
3647 | } |
3648 | ||
3649 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3650 | ||
76cd6173 | 3651 | noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
af3b8544 BP |
3652 | { |
3653 | return __should_fail_alloc_page(gfp_mask, order); | |
3654 | } | |
3655 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3656 | ||
f27ce0e1 JK |
3657 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3658 | unsigned int order, unsigned int alloc_flags) | |
3659 | { | |
3660 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); | |
3661 | long unusable_free = (1 << order) - 1; | |
3662 | ||
3663 | /* | |
3664 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3665 | * the high-atomic reserves. This will over-estimate the size of the | |
3666 | * atomic reserve but it avoids a search. | |
3667 | */ | |
3668 | if (likely(!alloc_harder)) | |
3669 | unusable_free += z->nr_reserved_highatomic; | |
3670 | ||
3671 | #ifdef CONFIG_CMA | |
3672 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3673 | if (!(alloc_flags & ALLOC_CMA)) | |
3674 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3675 | #endif | |
3676 | ||
3677 | return unusable_free; | |
3678 | } | |
3679 | ||
1da177e4 | 3680 | /* |
97a16fc8 MG |
3681 | * Return true if free base pages are above 'mark'. For high-order checks it |
3682 | * will return true of the order-0 watermark is reached and there is at least | |
3683 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3684 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3685 | */ |
86a294a8 | 3686 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3687 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3688 | long free_pages) |
1da177e4 | 3689 | { |
d23ad423 | 3690 | long min = mark; |
1da177e4 | 3691 | int o; |
cd04ae1e | 3692 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3693 | |
0aaa29a5 | 3694 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 3695 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 3696 | |
7fb1d9fc | 3697 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3698 | min -= min / 2; |
0aaa29a5 | 3699 | |
f27ce0e1 | 3700 | if (unlikely(alloc_harder)) { |
cd04ae1e MH |
3701 | /* |
3702 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3703 | * users on the grounds that it's definitely going to be in | |
3704 | * the exit path shortly and free memory. Any allocation it | |
3705 | * makes during the free path will be small and short-lived. | |
3706 | */ | |
3707 | if (alloc_flags & ALLOC_OOM) | |
3708 | min -= min / 2; | |
3709 | else | |
3710 | min -= min / 4; | |
3711 | } | |
3712 | ||
97a16fc8 MG |
3713 | /* |
3714 | * Check watermarks for an order-0 allocation request. If these | |
3715 | * are not met, then a high-order request also cannot go ahead | |
3716 | * even if a suitable page happened to be free. | |
3717 | */ | |
97a225e6 | 3718 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 3719 | return false; |
1da177e4 | 3720 | |
97a16fc8 MG |
3721 | /* If this is an order-0 request then the watermark is fine */ |
3722 | if (!order) | |
3723 | return true; | |
3724 | ||
3725 | /* For a high-order request, check at least one suitable page is free */ | |
3726 | for (o = order; o < MAX_ORDER; o++) { | |
3727 | struct free_area *area = &z->free_area[o]; | |
3728 | int mt; | |
3729 | ||
3730 | if (!area->nr_free) | |
3731 | continue; | |
3732 | ||
97a16fc8 | 3733 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3734 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3735 | return true; |
3736 | } | |
3737 | ||
3738 | #ifdef CONFIG_CMA | |
d883c6cf | 3739 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3740 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3741 | return true; |
d883c6cf | 3742 | } |
97a16fc8 | 3743 | #endif |
76089d00 | 3744 | if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC)) |
b050e376 | 3745 | return true; |
1da177e4 | 3746 | } |
97a16fc8 | 3747 | return false; |
88f5acf8 MG |
3748 | } |
3749 | ||
7aeb09f9 | 3750 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3751 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 3752 | { |
97a225e6 | 3753 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
3754 | zone_page_state(z, NR_FREE_PAGES)); |
3755 | } | |
3756 | ||
48ee5f36 | 3757 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 3758 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 3759 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 3760 | { |
f27ce0e1 | 3761 | long free_pages; |
d883c6cf | 3762 | |
f27ce0e1 | 3763 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
3764 | |
3765 | /* | |
3766 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 3767 | * need to be calculated. |
48ee5f36 | 3768 | */ |
f27ce0e1 JK |
3769 | if (!order) { |
3770 | long fast_free; | |
3771 | ||
3772 | fast_free = free_pages; | |
3773 | fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags); | |
3774 | if (fast_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
3775 | return true; | |
3776 | } | |
48ee5f36 | 3777 | |
f80b08fc CTR |
3778 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
3779 | free_pages)) | |
3780 | return true; | |
3781 | /* | |
3782 | * Ignore watermark boosting for GFP_ATOMIC order-0 allocations | |
3783 | * when checking the min watermark. The min watermark is the | |
3784 | * point where boosting is ignored so that kswapd is woken up | |
3785 | * when below the low watermark. | |
3786 | */ | |
3787 | if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost | |
3788 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { | |
3789 | mark = z->_watermark[WMARK_MIN]; | |
3790 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
3791 | alloc_flags, free_pages); | |
3792 | } | |
3793 | ||
3794 | return false; | |
48ee5f36 MG |
3795 | } |
3796 | ||
7aeb09f9 | 3797 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 3798 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
3799 | { |
3800 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3801 | ||
3802 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3803 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3804 | ||
97a225e6 | 3805 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 3806 | free_pages); |
1da177e4 LT |
3807 | } |
3808 | ||
9276b1bc | 3809 | #ifdef CONFIG_NUMA |
957f822a DR |
3810 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3811 | { | |
e02dc017 | 3812 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3813 | node_reclaim_distance; |
957f822a | 3814 | } |
9276b1bc | 3815 | #else /* CONFIG_NUMA */ |
957f822a DR |
3816 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3817 | { | |
3818 | return true; | |
3819 | } | |
9276b1bc PJ |
3820 | #endif /* CONFIG_NUMA */ |
3821 | ||
6bb15450 MG |
3822 | /* |
3823 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3824 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3825 | * premature use of a lower zone may cause lowmem pressure problems that | |
3826 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3827 | * probably too small. It only makes sense to spread allocations to avoid | |
3828 | * fragmentation between the Normal and DMA32 zones. | |
3829 | */ | |
3830 | static inline unsigned int | |
0a79cdad | 3831 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3832 | { |
736838e9 | 3833 | unsigned int alloc_flags; |
0a79cdad | 3834 | |
736838e9 MN |
3835 | /* |
3836 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
3837 | * to save a branch. | |
3838 | */ | |
3839 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
3840 | |
3841 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3842 | if (!zone) |
3843 | return alloc_flags; | |
3844 | ||
6bb15450 | 3845 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3846 | return alloc_flags; |
6bb15450 MG |
3847 | |
3848 | /* | |
3849 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3850 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3851 | * on UMA that if Normal is populated then so is DMA32. | |
3852 | */ | |
3853 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3854 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3855 | return alloc_flags; |
6bb15450 | 3856 | |
8118b82e | 3857 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3858 | #endif /* CONFIG_ZONE_DMA32 */ |
3859 | return alloc_flags; | |
6bb15450 | 3860 | } |
6bb15450 | 3861 | |
8e3560d9 PT |
3862 | /* Must be called after current_gfp_context() which can change gfp_mask */ |
3863 | static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask, | |
3864 | unsigned int alloc_flags) | |
8510e69c JK |
3865 | { |
3866 | #ifdef CONFIG_CMA | |
8e3560d9 | 3867 | if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
8510e69c | 3868 | alloc_flags |= ALLOC_CMA; |
8510e69c JK |
3869 | #endif |
3870 | return alloc_flags; | |
3871 | } | |
3872 | ||
7fb1d9fc | 3873 | /* |
0798e519 | 3874 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3875 | * a page. |
3876 | */ | |
3877 | static struct page * | |
a9263751 VB |
3878 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3879 | const struct alloc_context *ac) | |
753ee728 | 3880 | { |
6bb15450 | 3881 | struct zoneref *z; |
5117f45d | 3882 | struct zone *zone; |
3b8c0be4 | 3883 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
6bb15450 | 3884 | bool no_fallback; |
3b8c0be4 | 3885 | |
6bb15450 | 3886 | retry: |
7fb1d9fc | 3887 | /* |
9276b1bc | 3888 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3889 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3890 | */ |
6bb15450 MG |
3891 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3892 | z = ac->preferred_zoneref; | |
30d8ec73 MN |
3893 | for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx, |
3894 | ac->nodemask) { | |
be06af00 | 3895 | struct page *page; |
e085dbc5 JW |
3896 | unsigned long mark; |
3897 | ||
664eedde MG |
3898 | if (cpusets_enabled() && |
3899 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3900 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3901 | continue; |
a756cf59 JW |
3902 | /* |
3903 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3904 | * want to get it from a node that is within its dirty |
3905 | * limit, such that no single node holds more than its | |
a756cf59 | 3906 | * proportional share of globally allowed dirty pages. |
281e3726 | 3907 | * The dirty limits take into account the node's |
a756cf59 JW |
3908 | * lowmem reserves and high watermark so that kswapd |
3909 | * should be able to balance it without having to | |
3910 | * write pages from its LRU list. | |
3911 | * | |
a756cf59 | 3912 | * XXX: For now, allow allocations to potentially |
281e3726 | 3913 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3914 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3915 | * which is important when on a NUMA setup the allowed |
281e3726 | 3916 | * nodes are together not big enough to reach the |
a756cf59 | 3917 | * global limit. The proper fix for these situations |
281e3726 | 3918 | * will require awareness of nodes in the |
a756cf59 JW |
3919 | * dirty-throttling and the flusher threads. |
3920 | */ | |
3b8c0be4 MG |
3921 | if (ac->spread_dirty_pages) { |
3922 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3923 | continue; | |
3924 | ||
3925 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3926 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3927 | continue; | |
3928 | } | |
3929 | } | |
7fb1d9fc | 3930 | |
6bb15450 MG |
3931 | if (no_fallback && nr_online_nodes > 1 && |
3932 | zone != ac->preferred_zoneref->zone) { | |
3933 | int local_nid; | |
3934 | ||
3935 | /* | |
3936 | * If moving to a remote node, retry but allow | |
3937 | * fragmenting fallbacks. Locality is more important | |
3938 | * than fragmentation avoidance. | |
3939 | */ | |
3940 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3941 | if (zone_to_nid(zone) != local_nid) { | |
3942 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3943 | goto retry; | |
3944 | } | |
3945 | } | |
3946 | ||
a9214443 | 3947 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 3948 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
3949 | ac->highest_zoneidx, alloc_flags, |
3950 | gfp_mask)) { | |
fa5e084e MG |
3951 | int ret; |
3952 | ||
c9e97a19 PT |
3953 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3954 | /* | |
3955 | * Watermark failed for this zone, but see if we can | |
3956 | * grow this zone if it contains deferred pages. | |
3957 | */ | |
3958 | if (static_branch_unlikely(&deferred_pages)) { | |
3959 | if (_deferred_grow_zone(zone, order)) | |
3960 | goto try_this_zone; | |
3961 | } | |
3962 | #endif | |
5dab2911 MG |
3963 | /* Checked here to keep the fast path fast */ |
3964 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3965 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3966 | goto try_this_zone; | |
3967 | ||
202e35db | 3968 | if (!node_reclaim_enabled() || |
c33d6c06 | 3969 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3970 | continue; |
3971 | ||
a5f5f91d | 3972 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3973 | switch (ret) { |
a5f5f91d | 3974 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3975 | /* did not scan */ |
cd38b115 | 3976 | continue; |
a5f5f91d | 3977 | case NODE_RECLAIM_FULL: |
fa5e084e | 3978 | /* scanned but unreclaimable */ |
cd38b115 | 3979 | continue; |
fa5e084e MG |
3980 | default: |
3981 | /* did we reclaim enough */ | |
fed2719e | 3982 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 3983 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
3984 | goto try_this_zone; |
3985 | ||
fed2719e | 3986 | continue; |
0798e519 | 3987 | } |
7fb1d9fc RS |
3988 | } |
3989 | ||
fa5e084e | 3990 | try_this_zone: |
066b2393 | 3991 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3992 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3993 | if (page) { |
479f854a | 3994 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3995 | |
3996 | /* | |
3997 | * If this is a high-order atomic allocation then check | |
3998 | * if the pageblock should be reserved for the future | |
3999 | */ | |
4000 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
4001 | reserve_highatomic_pageblock(page, zone, order); | |
4002 | ||
75379191 | 4003 | return page; |
c9e97a19 PT |
4004 | } else { |
4005 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
4006 | /* Try again if zone has deferred pages */ | |
4007 | if (static_branch_unlikely(&deferred_pages)) { | |
4008 | if (_deferred_grow_zone(zone, order)) | |
4009 | goto try_this_zone; | |
4010 | } | |
4011 | #endif | |
75379191 | 4012 | } |
54a6eb5c | 4013 | } |
9276b1bc | 4014 | |
6bb15450 MG |
4015 | /* |
4016 | * It's possible on a UMA machine to get through all zones that are | |
4017 | * fragmented. If avoiding fragmentation, reset and try again. | |
4018 | */ | |
4019 | if (no_fallback) { | |
4020 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
4021 | goto retry; | |
4022 | } | |
4023 | ||
4ffeaf35 | 4024 | return NULL; |
753ee728 MH |
4025 | } |
4026 | ||
9af744d7 | 4027 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 4028 | { |
a238ab5b | 4029 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
4030 | |
4031 | /* | |
4032 | * This documents exceptions given to allocations in certain | |
4033 | * contexts that are allowed to allocate outside current's set | |
4034 | * of allowed nodes. | |
4035 | */ | |
4036 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 4037 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
4038 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
4039 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 4040 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
4041 | filter &= ~SHOW_MEM_FILTER_NODES; |
4042 | ||
9af744d7 | 4043 | show_mem(filter, nodemask); |
aa187507 MH |
4044 | } |
4045 | ||
a8e99259 | 4046 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
4047 | { |
4048 | struct va_format vaf; | |
4049 | va_list args; | |
1be334e5 | 4050 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 4051 | |
0f7896f1 | 4052 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
4053 | return; |
4054 | ||
7877cdcc MH |
4055 | va_start(args, fmt); |
4056 | vaf.fmt = fmt; | |
4057 | vaf.va = &args; | |
ef8444ea | 4058 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
4059 | current->comm, &vaf, gfp_mask, &gfp_mask, |
4060 | nodemask_pr_args(nodemask)); | |
7877cdcc | 4061 | va_end(args); |
3ee9a4f0 | 4062 | |
a8e99259 | 4063 | cpuset_print_current_mems_allowed(); |
ef8444ea | 4064 | pr_cont("\n"); |
a238ab5b | 4065 | dump_stack(); |
685dbf6f | 4066 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
4067 | } |
4068 | ||
6c18ba7a MH |
4069 | static inline struct page * |
4070 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
4071 | unsigned int alloc_flags, | |
4072 | const struct alloc_context *ac) | |
4073 | { | |
4074 | struct page *page; | |
4075 | ||
4076 | page = get_page_from_freelist(gfp_mask, order, | |
4077 | alloc_flags|ALLOC_CPUSET, ac); | |
4078 | /* | |
4079 | * fallback to ignore cpuset restriction if our nodes | |
4080 | * are depleted | |
4081 | */ | |
4082 | if (!page) | |
4083 | page = get_page_from_freelist(gfp_mask, order, | |
4084 | alloc_flags, ac); | |
4085 | ||
4086 | return page; | |
4087 | } | |
4088 | ||
11e33f6a MG |
4089 | static inline struct page * |
4090 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4091 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 4092 | { |
6e0fc46d DR |
4093 | struct oom_control oc = { |
4094 | .zonelist = ac->zonelist, | |
4095 | .nodemask = ac->nodemask, | |
2a966b77 | 4096 | .memcg = NULL, |
6e0fc46d DR |
4097 | .gfp_mask = gfp_mask, |
4098 | .order = order, | |
6e0fc46d | 4099 | }; |
11e33f6a MG |
4100 | struct page *page; |
4101 | ||
9879de73 JW |
4102 | *did_some_progress = 0; |
4103 | ||
9879de73 | 4104 | /* |
dc56401f JW |
4105 | * Acquire the oom lock. If that fails, somebody else is |
4106 | * making progress for us. | |
9879de73 | 4107 | */ |
dc56401f | 4108 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 4109 | *did_some_progress = 1; |
11e33f6a | 4110 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
4111 | return NULL; |
4112 | } | |
6b1de916 | 4113 | |
11e33f6a MG |
4114 | /* |
4115 | * Go through the zonelist yet one more time, keep very high watermark | |
4116 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
4117 | * we're still under heavy pressure. But make sure that this reclaim |
4118 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
4119 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 4120 | */ |
e746bf73 TH |
4121 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
4122 | ~__GFP_DIRECT_RECLAIM, order, | |
4123 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 4124 | if (page) |
11e33f6a MG |
4125 | goto out; |
4126 | ||
06ad276a MH |
4127 | /* Coredumps can quickly deplete all memory reserves */ |
4128 | if (current->flags & PF_DUMPCORE) | |
4129 | goto out; | |
4130 | /* The OOM killer will not help higher order allocs */ | |
4131 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4132 | goto out; | |
dcda9b04 MH |
4133 | /* |
4134 | * We have already exhausted all our reclaim opportunities without any | |
4135 | * success so it is time to admit defeat. We will skip the OOM killer | |
4136 | * because it is very likely that the caller has a more reasonable | |
4137 | * fallback than shooting a random task. | |
cfb4a541 MN |
4138 | * |
4139 | * The OOM killer may not free memory on a specific node. | |
dcda9b04 | 4140 | */ |
cfb4a541 | 4141 | if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE)) |
dcda9b04 | 4142 | goto out; |
06ad276a | 4143 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 4144 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
4145 | goto out; |
4146 | if (pm_suspended_storage()) | |
4147 | goto out; | |
4148 | /* | |
4149 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
4150 | * other request to make a forward progress. | |
4151 | * We are in an unfortunate situation where out_of_memory cannot | |
4152 | * do much for this context but let's try it to at least get | |
4153 | * access to memory reserved if the current task is killed (see | |
4154 | * out_of_memory). Once filesystems are ready to handle allocation | |
4155 | * failures more gracefully we should just bail out here. | |
4156 | */ | |
4157 | ||
3c2c6488 | 4158 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 4159 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 4160 | *did_some_progress = 1; |
5020e285 | 4161 | |
6c18ba7a MH |
4162 | /* |
4163 | * Help non-failing allocations by giving them access to memory | |
4164 | * reserves | |
4165 | */ | |
4166 | if (gfp_mask & __GFP_NOFAIL) | |
4167 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 4168 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 4169 | } |
11e33f6a | 4170 | out: |
dc56401f | 4171 | mutex_unlock(&oom_lock); |
11e33f6a MG |
4172 | return page; |
4173 | } | |
4174 | ||
33c2d214 | 4175 | /* |
baf2f90b | 4176 | * Maximum number of compaction retries with a progress before OOM |
33c2d214 MH |
4177 | * killer is consider as the only way to move forward. |
4178 | */ | |
4179 | #define MAX_COMPACT_RETRIES 16 | |
4180 | ||
56de7263 MG |
4181 | #ifdef CONFIG_COMPACTION |
4182 | /* Try memory compaction for high-order allocations before reclaim */ | |
4183 | static struct page * | |
4184 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4185 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4186 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4187 | { |
5e1f0f09 | 4188 | struct page *page = NULL; |
eb414681 | 4189 | unsigned long pflags; |
499118e9 | 4190 | unsigned int noreclaim_flag; |
53853e2d VB |
4191 | |
4192 | if (!order) | |
66199712 | 4193 | return NULL; |
66199712 | 4194 | |
eb414681 | 4195 | psi_memstall_enter(&pflags); |
499118e9 | 4196 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 4197 | |
c5d01d0d | 4198 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 4199 | prio, &page); |
eb414681 | 4200 | |
499118e9 | 4201 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 4202 | psi_memstall_leave(&pflags); |
56de7263 | 4203 | |
06dac2f4 CTR |
4204 | if (*compact_result == COMPACT_SKIPPED) |
4205 | return NULL; | |
98dd3b48 VB |
4206 | /* |
4207 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
4208 | * count a compaction stall | |
4209 | */ | |
4210 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 4211 | |
5e1f0f09 MG |
4212 | /* Prep a captured page if available */ |
4213 | if (page) | |
4214 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
4215 | ||
4216 | /* Try get a page from the freelist if available */ | |
4217 | if (!page) | |
4218 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 4219 | |
98dd3b48 VB |
4220 | if (page) { |
4221 | struct zone *zone = page_zone(page); | |
53853e2d | 4222 | |
98dd3b48 VB |
4223 | zone->compact_blockskip_flush = false; |
4224 | compaction_defer_reset(zone, order, true); | |
4225 | count_vm_event(COMPACTSUCCESS); | |
4226 | return page; | |
4227 | } | |
56de7263 | 4228 | |
98dd3b48 VB |
4229 | /* |
4230 | * It's bad if compaction run occurs and fails. The most likely reason | |
4231 | * is that pages exist, but not enough to satisfy watermarks. | |
4232 | */ | |
4233 | count_vm_event(COMPACTFAIL); | |
66199712 | 4234 | |
98dd3b48 | 4235 | cond_resched(); |
56de7263 MG |
4236 | |
4237 | return NULL; | |
4238 | } | |
33c2d214 | 4239 | |
3250845d VB |
4240 | static inline bool |
4241 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
4242 | enum compact_result compact_result, | |
4243 | enum compact_priority *compact_priority, | |
d9436498 | 4244 | int *compaction_retries) |
3250845d VB |
4245 | { |
4246 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 4247 | int min_priority; |
65190cff MH |
4248 | bool ret = false; |
4249 | int retries = *compaction_retries; | |
4250 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
4251 | |
4252 | if (!order) | |
4253 | return false; | |
4254 | ||
d9436498 VB |
4255 | if (compaction_made_progress(compact_result)) |
4256 | (*compaction_retries)++; | |
4257 | ||
3250845d VB |
4258 | /* |
4259 | * compaction considers all the zone as desperately out of memory | |
4260 | * so it doesn't really make much sense to retry except when the | |
4261 | * failure could be caused by insufficient priority | |
4262 | */ | |
d9436498 VB |
4263 | if (compaction_failed(compact_result)) |
4264 | goto check_priority; | |
3250845d | 4265 | |
49433085 VB |
4266 | /* |
4267 | * compaction was skipped because there are not enough order-0 pages | |
4268 | * to work with, so we retry only if it looks like reclaim can help. | |
4269 | */ | |
4270 | if (compaction_needs_reclaim(compact_result)) { | |
4271 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
4272 | goto out; | |
4273 | } | |
4274 | ||
3250845d VB |
4275 | /* |
4276 | * make sure the compaction wasn't deferred or didn't bail out early | |
4277 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
4278 | * But the next retry should use a higher priority if allowed, so |
4279 | * we don't just keep bailing out endlessly. | |
3250845d | 4280 | */ |
65190cff | 4281 | if (compaction_withdrawn(compact_result)) { |
49433085 | 4282 | goto check_priority; |
65190cff | 4283 | } |
3250845d VB |
4284 | |
4285 | /* | |
dcda9b04 | 4286 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
4287 | * costly ones because they are de facto nofail and invoke OOM |
4288 | * killer to move on while costly can fail and users are ready | |
4289 | * to cope with that. 1/4 retries is rather arbitrary but we | |
4290 | * would need much more detailed feedback from compaction to | |
4291 | * make a better decision. | |
4292 | */ | |
4293 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4294 | max_retries /= 4; | |
65190cff MH |
4295 | if (*compaction_retries <= max_retries) { |
4296 | ret = true; | |
4297 | goto out; | |
4298 | } | |
3250845d | 4299 | |
d9436498 VB |
4300 | /* |
4301 | * Make sure there are attempts at the highest priority if we exhausted | |
4302 | * all retries or failed at the lower priorities. | |
4303 | */ | |
4304 | check_priority: | |
c2033b00 VB |
4305 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
4306 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 4307 | |
c2033b00 | 4308 | if (*compact_priority > min_priority) { |
d9436498 VB |
4309 | (*compact_priority)--; |
4310 | *compaction_retries = 0; | |
65190cff | 4311 | ret = true; |
d9436498 | 4312 | } |
65190cff MH |
4313 | out: |
4314 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4315 | return ret; | |
3250845d | 4316 | } |
56de7263 MG |
4317 | #else |
4318 | static inline struct page * | |
4319 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4320 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4321 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4322 | { |
33c2d214 | 4323 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4324 | return NULL; |
4325 | } | |
33c2d214 MH |
4326 | |
4327 | static inline bool | |
86a294a8 MH |
4328 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4329 | enum compact_result compact_result, | |
a5508cd8 | 4330 | enum compact_priority *compact_priority, |
d9436498 | 4331 | int *compaction_retries) |
33c2d214 | 4332 | { |
31e49bfd MH |
4333 | struct zone *zone; |
4334 | struct zoneref *z; | |
4335 | ||
4336 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4337 | return false; | |
4338 | ||
4339 | /* | |
4340 | * There are setups with compaction disabled which would prefer to loop | |
4341 | * inside the allocator rather than hit the oom killer prematurely. | |
4342 | * Let's give them a good hope and keep retrying while the order-0 | |
4343 | * watermarks are OK. | |
4344 | */ | |
97a225e6 JK |
4345 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4346 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 4347 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 4348 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
4349 | return true; |
4350 | } | |
33c2d214 MH |
4351 | return false; |
4352 | } | |
3250845d | 4353 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4354 | |
d92a8cfc | 4355 | #ifdef CONFIG_LOCKDEP |
93781325 | 4356 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4357 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4358 | ||
f920e413 | 4359 | static bool __need_reclaim(gfp_t gfp_mask) |
d92a8cfc | 4360 | { |
d92a8cfc PZ |
4361 | /* no reclaim without waiting on it */ |
4362 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4363 | return false; | |
4364 | ||
4365 | /* this guy won't enter reclaim */ | |
2e517d68 | 4366 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4367 | return false; |
4368 | ||
d92a8cfc PZ |
4369 | if (gfp_mask & __GFP_NOLOCKDEP) |
4370 | return false; | |
4371 | ||
4372 | return true; | |
4373 | } | |
4374 | ||
93781325 OS |
4375 | void __fs_reclaim_acquire(void) |
4376 | { | |
4377 | lock_map_acquire(&__fs_reclaim_map); | |
4378 | } | |
4379 | ||
4380 | void __fs_reclaim_release(void) | |
4381 | { | |
4382 | lock_map_release(&__fs_reclaim_map); | |
4383 | } | |
4384 | ||
d92a8cfc PZ |
4385 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4386 | { | |
f920e413 DV |
4387 | gfp_mask = current_gfp_context(gfp_mask); |
4388 | ||
4389 | if (__need_reclaim(gfp_mask)) { | |
4390 | if (gfp_mask & __GFP_FS) | |
4391 | __fs_reclaim_acquire(); | |
4392 | ||
4393 | #ifdef CONFIG_MMU_NOTIFIER | |
4394 | lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); | |
4395 | lock_map_release(&__mmu_notifier_invalidate_range_start_map); | |
4396 | #endif | |
4397 | ||
4398 | } | |
d92a8cfc PZ |
4399 | } |
4400 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4401 | ||
4402 | void fs_reclaim_release(gfp_t gfp_mask) | |
4403 | { | |
f920e413 DV |
4404 | gfp_mask = current_gfp_context(gfp_mask); |
4405 | ||
4406 | if (__need_reclaim(gfp_mask)) { | |
4407 | if (gfp_mask & __GFP_FS) | |
4408 | __fs_reclaim_release(); | |
4409 | } | |
d92a8cfc PZ |
4410 | } |
4411 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4412 | #endif | |
4413 | ||
bba90710 | 4414 | /* Perform direct synchronous page reclaim */ |
2187e17b | 4415 | static unsigned long |
a9263751 VB |
4416 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4417 | const struct alloc_context *ac) | |
11e33f6a | 4418 | { |
499118e9 | 4419 | unsigned int noreclaim_flag; |
2187e17b | 4420 | unsigned long pflags, progress; |
11e33f6a MG |
4421 | |
4422 | cond_resched(); | |
4423 | ||
4424 | /* We now go into synchronous reclaim */ | |
4425 | cpuset_memory_pressure_bump(); | |
eb414681 | 4426 | psi_memstall_enter(&pflags); |
d92a8cfc | 4427 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4428 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4429 | |
a9263751 VB |
4430 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4431 | ac->nodemask); | |
11e33f6a | 4432 | |
499118e9 | 4433 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4434 | fs_reclaim_release(gfp_mask); |
eb414681 | 4435 | psi_memstall_leave(&pflags); |
11e33f6a MG |
4436 | |
4437 | cond_resched(); | |
4438 | ||
bba90710 MS |
4439 | return progress; |
4440 | } | |
4441 | ||
4442 | /* The really slow allocator path where we enter direct reclaim */ | |
4443 | static inline struct page * | |
4444 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4445 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4446 | unsigned long *did_some_progress) |
bba90710 MS |
4447 | { |
4448 | struct page *page = NULL; | |
4449 | bool drained = false; | |
4450 | ||
a9263751 | 4451 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
4452 | if (unlikely(!(*did_some_progress))) |
4453 | return NULL; | |
11e33f6a | 4454 | |
9ee493ce | 4455 | retry: |
31a6c190 | 4456 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4457 | |
4458 | /* | |
4459 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 | 4460 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
047b9967 | 4461 | * Shrink them and try again |
9ee493ce MG |
4462 | */ |
4463 | if (!page && !drained) { | |
29fac03b | 4464 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4465 | drain_all_pages(NULL); |
9ee493ce MG |
4466 | drained = true; |
4467 | goto retry; | |
4468 | } | |
4469 | ||
11e33f6a MG |
4470 | return page; |
4471 | } | |
4472 | ||
5ecd9d40 DR |
4473 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4474 | const struct alloc_context *ac) | |
3a025760 JW |
4475 | { |
4476 | struct zoneref *z; | |
4477 | struct zone *zone; | |
e1a55637 | 4478 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 4479 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 4480 | |
97a225e6 | 4481 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 4482 | ac->nodemask) { |
e1a55637 | 4483 | if (last_pgdat != zone->zone_pgdat) |
97a225e6 | 4484 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
e1a55637 MG |
4485 | last_pgdat = zone->zone_pgdat; |
4486 | } | |
3a025760 JW |
4487 | } |
4488 | ||
c603844b | 4489 | static inline unsigned int |
341ce06f PZ |
4490 | gfp_to_alloc_flags(gfp_t gfp_mask) |
4491 | { | |
c603844b | 4492 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4493 | |
736838e9 MN |
4494 | /* |
4495 | * __GFP_HIGH is assumed to be the same as ALLOC_HIGH | |
4496 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4497 | * to save two branches. | |
4498 | */ | |
e6223a3b | 4499 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
736838e9 | 4500 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 4501 | |
341ce06f PZ |
4502 | /* |
4503 | * The caller may dip into page reserves a bit more if the caller | |
4504 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4505 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 4506 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 4507 | */ |
736838e9 MN |
4508 | alloc_flags |= (__force int) |
4509 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 4510 | |
d0164adc | 4511 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 4512 | /* |
b104a35d DR |
4513 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4514 | * if it can't schedule. | |
5c3240d9 | 4515 | */ |
b104a35d | 4516 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 4517 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 4518 | /* |
b104a35d | 4519 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 4520 | * comment for __cpuset_node_allowed(). |
523b9458 | 4521 | */ |
341ce06f | 4522 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 4523 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
4524 | alloc_flags |= ALLOC_HARDER; |
4525 | ||
8e3560d9 | 4526 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags); |
8510e69c | 4527 | |
341ce06f PZ |
4528 | return alloc_flags; |
4529 | } | |
4530 | ||
cd04ae1e | 4531 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4532 | { |
cd04ae1e MH |
4533 | if (!tsk_is_oom_victim(tsk)) |
4534 | return false; | |
4535 | ||
4536 | /* | |
4537 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4538 | * only to the thread with TIF_MEMDIE set | |
4539 | */ | |
4540 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4541 | return false; |
4542 | ||
cd04ae1e MH |
4543 | return true; |
4544 | } | |
4545 | ||
4546 | /* | |
4547 | * Distinguish requests which really need access to full memory | |
4548 | * reserves from oom victims which can live with a portion of it | |
4549 | */ | |
4550 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4551 | { | |
4552 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4553 | return 0; | |
31a6c190 | 4554 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4555 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4556 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4557 | return ALLOC_NO_WATERMARKS; |
4558 | if (!in_interrupt()) { | |
4559 | if (current->flags & PF_MEMALLOC) | |
4560 | return ALLOC_NO_WATERMARKS; | |
4561 | else if (oom_reserves_allowed(current)) | |
4562 | return ALLOC_OOM; | |
4563 | } | |
31a6c190 | 4564 | |
cd04ae1e MH |
4565 | return 0; |
4566 | } | |
4567 | ||
4568 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4569 | { | |
4570 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4571 | } |
4572 | ||
0a0337e0 MH |
4573 | /* |
4574 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4575 | * for the given allocation request. | |
491d79ae JW |
4576 | * |
4577 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4578 | * without success, or when we couldn't even meet the watermark if we | |
4579 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4580 | * |
4581 | * Returns true if a retry is viable or false to enter the oom path. | |
4582 | */ | |
4583 | static inline bool | |
4584 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4585 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4586 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4587 | { |
4588 | struct zone *zone; | |
4589 | struct zoneref *z; | |
15f570bf | 4590 | bool ret = false; |
0a0337e0 | 4591 | |
423b452e VB |
4592 | /* |
4593 | * Costly allocations might have made a progress but this doesn't mean | |
4594 | * their order will become available due to high fragmentation so | |
4595 | * always increment the no progress counter for them | |
4596 | */ | |
4597 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4598 | *no_progress_loops = 0; | |
4599 | else | |
4600 | (*no_progress_loops)++; | |
4601 | ||
0a0337e0 MH |
4602 | /* |
4603 | * Make sure we converge to OOM if we cannot make any progress | |
4604 | * several times in the row. | |
4605 | */ | |
04c8716f MK |
4606 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4607 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4608 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4609 | } |
0a0337e0 | 4610 | |
bca67592 MG |
4611 | /* |
4612 | * Keep reclaiming pages while there is a chance this will lead | |
4613 | * somewhere. If none of the target zones can satisfy our allocation | |
4614 | * request even if all reclaimable pages are considered then we are | |
4615 | * screwed and have to go OOM. | |
0a0337e0 | 4616 | */ |
97a225e6 JK |
4617 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4618 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4619 | unsigned long available; |
ede37713 | 4620 | unsigned long reclaimable; |
d379f01d MH |
4621 | unsigned long min_wmark = min_wmark_pages(zone); |
4622 | bool wmark; | |
0a0337e0 | 4623 | |
5a1c84b4 | 4624 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4625 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4626 | |
4627 | /* | |
491d79ae JW |
4628 | * Would the allocation succeed if we reclaimed all |
4629 | * reclaimable pages? | |
0a0337e0 | 4630 | */ |
d379f01d | 4631 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 4632 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
4633 | trace_reclaim_retry_zone(z, order, reclaimable, |
4634 | available, min_wmark, *no_progress_loops, wmark); | |
4635 | if (wmark) { | |
ede37713 MH |
4636 | /* |
4637 | * If we didn't make any progress and have a lot of | |
4638 | * dirty + writeback pages then we should wait for | |
4639 | * an IO to complete to slow down the reclaim and | |
4640 | * prevent from pre mature OOM | |
4641 | */ | |
4642 | if (!did_some_progress) { | |
11fb9989 | 4643 | unsigned long write_pending; |
ede37713 | 4644 | |
5a1c84b4 MG |
4645 | write_pending = zone_page_state_snapshot(zone, |
4646 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 4647 | |
11fb9989 | 4648 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
4649 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
4650 | return true; | |
4651 | } | |
4652 | } | |
5a1c84b4 | 4653 | |
15f570bf MH |
4654 | ret = true; |
4655 | goto out; | |
0a0337e0 MH |
4656 | } |
4657 | } | |
4658 | ||
15f570bf MH |
4659 | out: |
4660 | /* | |
4661 | * Memory allocation/reclaim might be called from a WQ context and the | |
4662 | * current implementation of the WQ concurrency control doesn't | |
4663 | * recognize that a particular WQ is congested if the worker thread is | |
4664 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4665 | * here rather than calling cond_resched(). | |
4666 | */ | |
4667 | if (current->flags & PF_WQ_WORKER) | |
4668 | schedule_timeout_uninterruptible(1); | |
4669 | else | |
4670 | cond_resched(); | |
4671 | return ret; | |
0a0337e0 MH |
4672 | } |
4673 | ||
902b6281 VB |
4674 | static inline bool |
4675 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4676 | { | |
4677 | /* | |
4678 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4679 | * mempolicy don't intersect. This should be normally dealt with by | |
4680 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4681 | * such a way the check therein was true, and then it became false | |
4682 | * before we got our cpuset_mems_cookie here. | |
4683 | * This assumes that for all allocations, ac->nodemask can come only | |
4684 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4685 | * when it does not intersect with the cpuset restrictions) or the | |
4686 | * caller can deal with a violated nodemask. | |
4687 | */ | |
4688 | if (cpusets_enabled() && ac->nodemask && | |
4689 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4690 | ac->nodemask = NULL; | |
4691 | return true; | |
4692 | } | |
4693 | ||
4694 | /* | |
4695 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4696 | * possible to race with parallel threads in such a way that our | |
4697 | * allocation can fail while the mask is being updated. If we are about | |
4698 | * to fail, check if the cpuset changed during allocation and if so, | |
4699 | * retry. | |
4700 | */ | |
4701 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4702 | return true; | |
4703 | ||
4704 | return false; | |
4705 | } | |
4706 | ||
11e33f6a MG |
4707 | static inline struct page * |
4708 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4709 | struct alloc_context *ac) |
11e33f6a | 4710 | { |
d0164adc | 4711 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4712 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4713 | struct page *page = NULL; |
c603844b | 4714 | unsigned int alloc_flags; |
11e33f6a | 4715 | unsigned long did_some_progress; |
5ce9bfef | 4716 | enum compact_priority compact_priority; |
c5d01d0d | 4717 | enum compact_result compact_result; |
5ce9bfef VB |
4718 | int compaction_retries; |
4719 | int no_progress_loops; | |
5ce9bfef | 4720 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4721 | int reserve_flags; |
1da177e4 | 4722 | |
d0164adc MG |
4723 | /* |
4724 | * We also sanity check to catch abuse of atomic reserves being used by | |
4725 | * callers that are not in atomic context. | |
4726 | */ | |
4727 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4728 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4729 | gfp_mask &= ~__GFP_ATOMIC; | |
4730 | ||
5ce9bfef VB |
4731 | retry_cpuset: |
4732 | compaction_retries = 0; | |
4733 | no_progress_loops = 0; | |
4734 | compact_priority = DEF_COMPACT_PRIORITY; | |
4735 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4736 | |
4737 | /* | |
4738 | * The fast path uses conservative alloc_flags to succeed only until | |
4739 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4740 | * alloc_flags precisely. So we do that now. | |
4741 | */ | |
4742 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4743 | ||
e47483bc VB |
4744 | /* |
4745 | * We need to recalculate the starting point for the zonelist iterator | |
4746 | * because we might have used different nodemask in the fast path, or | |
4747 | * there was a cpuset modification and we are retrying - otherwise we | |
4748 | * could end up iterating over non-eligible zones endlessly. | |
4749 | */ | |
4750 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 4751 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
4752 | if (!ac->preferred_zoneref->zone) |
4753 | goto nopage; | |
4754 | ||
0a79cdad | 4755 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4756 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4757 | |
4758 | /* | |
4759 | * The adjusted alloc_flags might result in immediate success, so try | |
4760 | * that first | |
4761 | */ | |
4762 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4763 | if (page) | |
4764 | goto got_pg; | |
4765 | ||
a8161d1e VB |
4766 | /* |
4767 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4768 | * that we have enough base pages and don't need to reclaim. For non- |
4769 | * movable high-order allocations, do that as well, as compaction will | |
4770 | * try prevent permanent fragmentation by migrating from blocks of the | |
4771 | * same migratetype. | |
4772 | * Don't try this for allocations that are allowed to ignore | |
4773 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4774 | */ |
282722b0 VB |
4775 | if (can_direct_reclaim && |
4776 | (costly_order || | |
4777 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4778 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4779 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4780 | alloc_flags, ac, | |
a5508cd8 | 4781 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4782 | &compact_result); |
4783 | if (page) | |
4784 | goto got_pg; | |
4785 | ||
cc638f32 VB |
4786 | /* |
4787 | * Checks for costly allocations with __GFP_NORETRY, which | |
4788 | * includes some THP page fault allocations | |
4789 | */ | |
4790 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4791 | /* |
4792 | * If allocating entire pageblock(s) and compaction | |
4793 | * failed because all zones are below low watermarks | |
4794 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4795 | * order, fail immediately unless the allocator has |
4796 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4797 | * |
4798 | * Reclaim is | |
4799 | * - potentially very expensive because zones are far | |
4800 | * below their low watermarks or this is part of very | |
4801 | * bursty high order allocations, | |
4802 | * - not guaranteed to help because isolate_freepages() | |
4803 | * may not iterate over freed pages as part of its | |
4804 | * linear scan, and | |
4805 | * - unlikely to make entire pageblocks free on its | |
4806 | * own. | |
4807 | */ | |
4808 | if (compact_result == COMPACT_SKIPPED || | |
4809 | compact_result == COMPACT_DEFERRED) | |
4810 | goto nopage; | |
a8161d1e | 4811 | |
a8161d1e | 4812 | /* |
3eb2771b VB |
4813 | * Looks like reclaim/compaction is worth trying, but |
4814 | * sync compaction could be very expensive, so keep | |
25160354 | 4815 | * using async compaction. |
a8161d1e | 4816 | */ |
a5508cd8 | 4817 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4818 | } |
4819 | } | |
23771235 | 4820 | |
31a6c190 | 4821 | retry: |
23771235 | 4822 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4823 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4824 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4825 | |
cd04ae1e MH |
4826 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4827 | if (reserve_flags) | |
8e3560d9 | 4828 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags); |
23771235 | 4829 | |
e46e7b77 | 4830 | /* |
d6a24df0 VB |
4831 | * Reset the nodemask and zonelist iterators if memory policies can be |
4832 | * ignored. These allocations are high priority and system rather than | |
4833 | * user oriented. | |
e46e7b77 | 4834 | */ |
cd04ae1e | 4835 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4836 | ac->nodemask = NULL; |
e46e7b77 | 4837 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4838 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
4839 | } |
4840 | ||
23771235 | 4841 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4842 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4843 | if (page) |
4844 | goto got_pg; | |
1da177e4 | 4845 | |
d0164adc | 4846 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4847 | if (!can_direct_reclaim) |
1da177e4 LT |
4848 | goto nopage; |
4849 | ||
9a67f648 MH |
4850 | /* Avoid recursion of direct reclaim */ |
4851 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4852 | goto nopage; |
4853 | ||
a8161d1e VB |
4854 | /* Try direct reclaim and then allocating */ |
4855 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4856 | &did_some_progress); | |
4857 | if (page) | |
4858 | goto got_pg; | |
4859 | ||
4860 | /* Try direct compaction and then allocating */ | |
a9263751 | 4861 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4862 | compact_priority, &compact_result); |
56de7263 MG |
4863 | if (page) |
4864 | goto got_pg; | |
75f30861 | 4865 | |
9083905a JW |
4866 | /* Do not loop if specifically requested */ |
4867 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4868 | goto nopage; |
9083905a | 4869 | |
0a0337e0 MH |
4870 | /* |
4871 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4872 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4873 | */ |
dcda9b04 | 4874 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4875 | goto nopage; |
0a0337e0 | 4876 | |
0a0337e0 | 4877 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4878 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4879 | goto retry; |
4880 | ||
33c2d214 MH |
4881 | /* |
4882 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4883 | * reclaim is not able to make any progress because the current | |
4884 | * implementation of the compaction depends on the sufficient amount | |
4885 | * of free memory (see __compaction_suitable) | |
4886 | */ | |
4887 | if (did_some_progress > 0 && | |
86a294a8 | 4888 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4889 | compact_result, &compact_priority, |
d9436498 | 4890 | &compaction_retries)) |
33c2d214 MH |
4891 | goto retry; |
4892 | ||
902b6281 VB |
4893 | |
4894 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4895 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4896 | goto retry_cpuset; |
4897 | ||
9083905a JW |
4898 | /* Reclaim has failed us, start killing things */ |
4899 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4900 | if (page) | |
4901 | goto got_pg; | |
4902 | ||
9a67f648 | 4903 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e | 4904 | if (tsk_is_oom_victim(current) && |
8510e69c | 4905 | (alloc_flags & ALLOC_OOM || |
c288983d | 4906 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4907 | goto nopage; |
4908 | ||
9083905a | 4909 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4910 | if (did_some_progress) { |
4911 | no_progress_loops = 0; | |
9083905a | 4912 | goto retry; |
0a0337e0 | 4913 | } |
9083905a | 4914 | |
1da177e4 | 4915 | nopage: |
902b6281 VB |
4916 | /* Deal with possible cpuset update races before we fail */ |
4917 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4918 | goto retry_cpuset; |
4919 | ||
9a67f648 MH |
4920 | /* |
4921 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4922 | * we always retry | |
4923 | */ | |
4924 | if (gfp_mask & __GFP_NOFAIL) { | |
4925 | /* | |
4926 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4927 | * of any new users that actually require GFP_NOWAIT | |
4928 | */ | |
4929 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4930 | goto fail; | |
4931 | ||
4932 | /* | |
4933 | * PF_MEMALLOC request from this context is rather bizarre | |
4934 | * because we cannot reclaim anything and only can loop waiting | |
4935 | * for somebody to do a work for us | |
4936 | */ | |
4937 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4938 | ||
4939 | /* | |
4940 | * non failing costly orders are a hard requirement which we | |
4941 | * are not prepared for much so let's warn about these users | |
4942 | * so that we can identify them and convert them to something | |
4943 | * else. | |
4944 | */ | |
4945 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4946 | ||
6c18ba7a MH |
4947 | /* |
4948 | * Help non-failing allocations by giving them access to memory | |
4949 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4950 | * could deplete whole memory reserves which would just make | |
4951 | * the situation worse | |
4952 | */ | |
4953 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4954 | if (page) | |
4955 | goto got_pg; | |
4956 | ||
9a67f648 MH |
4957 | cond_resched(); |
4958 | goto retry; | |
4959 | } | |
4960 | fail: | |
a8e99259 | 4961 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4962 | "page allocation failure: order:%u", order); |
1da177e4 | 4963 | got_pg: |
072bb0aa | 4964 | return page; |
1da177e4 | 4965 | } |
11e33f6a | 4966 | |
9cd75558 | 4967 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4968 | int preferred_nid, nodemask_t *nodemask, |
8e6a930b | 4969 | struct alloc_context *ac, gfp_t *alloc_gfp, |
9cd75558 | 4970 | unsigned int *alloc_flags) |
11e33f6a | 4971 | { |
97a225e6 | 4972 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4973 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 4974 | ac->nodemask = nodemask; |
01c0bfe0 | 4975 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 4976 | |
682a3385 | 4977 | if (cpusets_enabled()) { |
8e6a930b | 4978 | *alloc_gfp |= __GFP_HARDWALL; |
182f3d7a MS |
4979 | /* |
4980 | * When we are in the interrupt context, it is irrelevant | |
4981 | * to the current task context. It means that any node ok. | |
4982 | */ | |
4983 | if (!in_interrupt() && !ac->nodemask) | |
9cd75558 | 4984 | ac->nodemask = &cpuset_current_mems_allowed; |
51047820 VB |
4985 | else |
4986 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4987 | } |
4988 | ||
d92a8cfc PZ |
4989 | fs_reclaim_acquire(gfp_mask); |
4990 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4991 | |
d0164adc | 4992 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4993 | |
4994 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4995 | return false; |
11e33f6a | 4996 | |
8e3560d9 | 4997 | *alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags); |
d883c6cf | 4998 | |
c9ab0c4f | 4999 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 5000 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 5001 | |
e46e7b77 MG |
5002 | /* |
5003 | * The preferred zone is used for statistics but crucially it is | |
5004 | * also used as the starting point for the zonelist iterator. It | |
5005 | * may get reset for allocations that ignore memory policies. | |
5006 | */ | |
9cd75558 | 5007 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 5008 | ac->highest_zoneidx, ac->nodemask); |
a0622d05 MN |
5009 | |
5010 | return true; | |
9cd75558 MG |
5011 | } |
5012 | ||
387ba26f | 5013 | /* |
0f87d9d3 | 5014 | * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array |
387ba26f MG |
5015 | * @gfp: GFP flags for the allocation |
5016 | * @preferred_nid: The preferred NUMA node ID to allocate from | |
5017 | * @nodemask: Set of nodes to allocate from, may be NULL | |
0f87d9d3 MG |
5018 | * @nr_pages: The number of pages desired on the list or array |
5019 | * @page_list: Optional list to store the allocated pages | |
5020 | * @page_array: Optional array to store the pages | |
387ba26f MG |
5021 | * |
5022 | * This is a batched version of the page allocator that attempts to | |
0f87d9d3 MG |
5023 | * allocate nr_pages quickly. Pages are added to page_list if page_list |
5024 | * is not NULL, otherwise it is assumed that the page_array is valid. | |
387ba26f | 5025 | * |
0f87d9d3 MG |
5026 | * For lists, nr_pages is the number of pages that should be allocated. |
5027 | * | |
5028 | * For arrays, only NULL elements are populated with pages and nr_pages | |
5029 | * is the maximum number of pages that will be stored in the array. | |
5030 | * | |
5031 | * Returns the number of pages on the list or array. | |
387ba26f MG |
5032 | */ |
5033 | unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, | |
5034 | nodemask_t *nodemask, int nr_pages, | |
0f87d9d3 MG |
5035 | struct list_head *page_list, |
5036 | struct page **page_array) | |
387ba26f MG |
5037 | { |
5038 | struct page *page; | |
5039 | unsigned long flags; | |
5040 | struct zone *zone; | |
5041 | struct zoneref *z; | |
5042 | struct per_cpu_pages *pcp; | |
5043 | struct list_head *pcp_list; | |
5044 | struct alloc_context ac; | |
5045 | gfp_t alloc_gfp; | |
5046 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
0f87d9d3 | 5047 | int nr_populated = 0; |
387ba26f | 5048 | |
ce76f9a1 | 5049 | if (unlikely(nr_pages <= 0)) |
387ba26f MG |
5050 | return 0; |
5051 | ||
0f87d9d3 MG |
5052 | /* |
5053 | * Skip populated array elements to determine if any pages need | |
5054 | * to be allocated before disabling IRQs. | |
5055 | */ | |
b08e50dd | 5056 | while (page_array && nr_populated < nr_pages && page_array[nr_populated]) |
0f87d9d3 MG |
5057 | nr_populated++; |
5058 | ||
387ba26f | 5059 | /* Use the single page allocator for one page. */ |
0f87d9d3 | 5060 | if (nr_pages - nr_populated == 1) |
387ba26f MG |
5061 | goto failed; |
5062 | ||
5063 | /* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */ | |
5064 | gfp &= gfp_allowed_mask; | |
5065 | alloc_gfp = gfp; | |
5066 | if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags)) | |
5067 | return 0; | |
5068 | gfp = alloc_gfp; | |
5069 | ||
5070 | /* Find an allowed local zone that meets the low watermark. */ | |
5071 | for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) { | |
5072 | unsigned long mark; | |
5073 | ||
5074 | if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) && | |
5075 | !__cpuset_zone_allowed(zone, gfp)) { | |
5076 | continue; | |
5077 | } | |
5078 | ||
5079 | if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone && | |
5080 | zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) { | |
5081 | goto failed; | |
5082 | } | |
5083 | ||
5084 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages; | |
5085 | if (zone_watermark_fast(zone, 0, mark, | |
5086 | zonelist_zone_idx(ac.preferred_zoneref), | |
5087 | alloc_flags, gfp)) { | |
5088 | break; | |
5089 | } | |
5090 | } | |
5091 | ||
5092 | /* | |
5093 | * If there are no allowed local zones that meets the watermarks then | |
5094 | * try to allocate a single page and reclaim if necessary. | |
5095 | */ | |
ce76f9a1 | 5096 | if (unlikely(!zone)) |
387ba26f MG |
5097 | goto failed; |
5098 | ||
5099 | /* Attempt the batch allocation */ | |
5100 | local_irq_save(flags); | |
5101 | pcp = &this_cpu_ptr(zone->pageset)->pcp; | |
5102 | pcp_list = &pcp->lists[ac.migratetype]; | |
5103 | ||
0f87d9d3 MG |
5104 | while (nr_populated < nr_pages) { |
5105 | ||
5106 | /* Skip existing pages */ | |
5107 | if (page_array && page_array[nr_populated]) { | |
5108 | nr_populated++; | |
5109 | continue; | |
5110 | } | |
5111 | ||
387ba26f MG |
5112 | page = __rmqueue_pcplist(zone, ac.migratetype, alloc_flags, |
5113 | pcp, pcp_list); | |
ce76f9a1 | 5114 | if (unlikely(!page)) { |
387ba26f | 5115 | /* Try and get at least one page */ |
0f87d9d3 | 5116 | if (!nr_populated) |
387ba26f MG |
5117 | goto failed_irq; |
5118 | break; | |
5119 | } | |
5120 | ||
5121 | /* | |
5122 | * Ideally this would be batched but the best way to do | |
5123 | * that cheaply is to first convert zone_statistics to | |
5124 | * be inaccurate per-cpu counter like vm_events to avoid | |
5125 | * a RMW cycle then do the accounting with IRQs enabled. | |
5126 | */ | |
5127 | __count_zid_vm_events(PGALLOC, zone_idx(zone), 1); | |
5128 | zone_statistics(ac.preferred_zoneref->zone, zone); | |
5129 | ||
5130 | prep_new_page(page, 0, gfp, 0); | |
0f87d9d3 MG |
5131 | if (page_list) |
5132 | list_add(&page->lru, page_list); | |
5133 | else | |
5134 | page_array[nr_populated] = page; | |
5135 | nr_populated++; | |
387ba26f MG |
5136 | } |
5137 | ||
5138 | local_irq_restore(flags); | |
5139 | ||
0f87d9d3 | 5140 | return nr_populated; |
387ba26f MG |
5141 | |
5142 | failed_irq: | |
5143 | local_irq_restore(flags); | |
5144 | ||
5145 | failed: | |
5146 | page = __alloc_pages(gfp, 0, preferred_nid, nodemask); | |
5147 | if (page) { | |
0f87d9d3 MG |
5148 | if (page_list) |
5149 | list_add(&page->lru, page_list); | |
5150 | else | |
5151 | page_array[nr_populated] = page; | |
5152 | nr_populated++; | |
387ba26f MG |
5153 | } |
5154 | ||
0f87d9d3 | 5155 | return nr_populated; |
387ba26f MG |
5156 | } |
5157 | EXPORT_SYMBOL_GPL(__alloc_pages_bulk); | |
5158 | ||
9cd75558 MG |
5159 | /* |
5160 | * This is the 'heart' of the zoned buddy allocator. | |
5161 | */ | |
84172f4b | 5162 | struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, |
04ec6264 | 5163 | nodemask_t *nodemask) |
9cd75558 MG |
5164 | { |
5165 | struct page *page; | |
5166 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
8e6a930b | 5167 | gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
5168 | struct alloc_context ac = { }; |
5169 | ||
c63ae43b MH |
5170 | /* |
5171 | * There are several places where we assume that the order value is sane | |
5172 | * so bail out early if the request is out of bound. | |
5173 | */ | |
5174 | if (unlikely(order >= MAX_ORDER)) { | |
6e5e0f28 | 5175 | WARN_ON_ONCE(!(gfp & __GFP_NOWARN)); |
c63ae43b MH |
5176 | return NULL; |
5177 | } | |
5178 | ||
6e5e0f28 | 5179 | gfp &= gfp_allowed_mask; |
da6df1b0 PT |
5180 | /* |
5181 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS | |
5182 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
5183 | * from a particular context which has been marked by | |
8e3560d9 PT |
5184 | * memalloc_no{fs,io}_{save,restore}. And PF_MEMALLOC_PIN which ensures |
5185 | * movable zones are not used during allocation. | |
da6df1b0 PT |
5186 | */ |
5187 | gfp = current_gfp_context(gfp); | |
6e5e0f28 MWO |
5188 | alloc_gfp = gfp; |
5189 | if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac, | |
8e6a930b | 5190 | &alloc_gfp, &alloc_flags)) |
9cd75558 MG |
5191 | return NULL; |
5192 | ||
6bb15450 MG |
5193 | /* |
5194 | * Forbid the first pass from falling back to types that fragment | |
5195 | * memory until all local zones are considered. | |
5196 | */ | |
6e5e0f28 | 5197 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp); |
6bb15450 | 5198 | |
5117f45d | 5199 | /* First allocation attempt */ |
8e6a930b | 5200 | page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac); |
4fcb0971 MG |
5201 | if (likely(page)) |
5202 | goto out; | |
11e33f6a | 5203 | |
da6df1b0 | 5204 | alloc_gfp = gfp; |
4fcb0971 | 5205 | ac.spread_dirty_pages = false; |
23f086f9 | 5206 | |
4741526b MG |
5207 | /* |
5208 | * Restore the original nodemask if it was potentially replaced with | |
5209 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
5210 | */ | |
97ce86f9 | 5211 | ac.nodemask = nodemask; |
16096c25 | 5212 | |
8e6a930b | 5213 | page = __alloc_pages_slowpath(alloc_gfp, order, &ac); |
cc9a6c87 | 5214 | |
4fcb0971 | 5215 | out: |
6e5e0f28 MWO |
5216 | if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT) && page && |
5217 | unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) { | |
c4159a75 VD |
5218 | __free_pages(page, order); |
5219 | page = NULL; | |
4949148a VD |
5220 | } |
5221 | ||
8e6a930b | 5222 | trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype); |
4fcb0971 | 5223 | |
11e33f6a | 5224 | return page; |
1da177e4 | 5225 | } |
84172f4b | 5226 | EXPORT_SYMBOL(__alloc_pages); |
1da177e4 LT |
5227 | |
5228 | /* | |
9ea9a680 MH |
5229 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
5230 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
5231 | * you need to access high mem. | |
1da177e4 | 5232 | */ |
920c7a5d | 5233 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 5234 | { |
945a1113 AM |
5235 | struct page *page; |
5236 | ||
9ea9a680 | 5237 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
5238 | if (!page) |
5239 | return 0; | |
5240 | return (unsigned long) page_address(page); | |
5241 | } | |
1da177e4 LT |
5242 | EXPORT_SYMBOL(__get_free_pages); |
5243 | ||
920c7a5d | 5244 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 5245 | { |
945a1113 | 5246 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 5247 | } |
1da177e4 LT |
5248 | EXPORT_SYMBOL(get_zeroed_page); |
5249 | ||
742aa7fb | 5250 | static inline void free_the_page(struct page *page, unsigned int order) |
1da177e4 | 5251 | { |
742aa7fb AL |
5252 | if (order == 0) /* Via pcp? */ |
5253 | free_unref_page(page); | |
5254 | else | |
7fef431b | 5255 | __free_pages_ok(page, order, FPI_NONE); |
1da177e4 LT |
5256 | } |
5257 | ||
7f194fbb MWO |
5258 | /** |
5259 | * __free_pages - Free pages allocated with alloc_pages(). | |
5260 | * @page: The page pointer returned from alloc_pages(). | |
5261 | * @order: The order of the allocation. | |
5262 | * | |
5263 | * This function can free multi-page allocations that are not compound | |
5264 | * pages. It does not check that the @order passed in matches that of | |
5265 | * the allocation, so it is easy to leak memory. Freeing more memory | |
5266 | * than was allocated will probably emit a warning. | |
5267 | * | |
5268 | * If the last reference to this page is speculative, it will be released | |
5269 | * by put_page() which only frees the first page of a non-compound | |
5270 | * allocation. To prevent the remaining pages from being leaked, we free | |
5271 | * the subsequent pages here. If you want to use the page's reference | |
5272 | * count to decide when to free the allocation, you should allocate a | |
5273 | * compound page, and use put_page() instead of __free_pages(). | |
5274 | * | |
5275 | * Context: May be called in interrupt context or while holding a normal | |
5276 | * spinlock, but not in NMI context or while holding a raw spinlock. | |
5277 | */ | |
742aa7fb AL |
5278 | void __free_pages(struct page *page, unsigned int order) |
5279 | { | |
5280 | if (put_page_testzero(page)) | |
5281 | free_the_page(page, order); | |
e320d301 MWO |
5282 | else if (!PageHead(page)) |
5283 | while (order-- > 0) | |
5284 | free_the_page(page + (1 << order), order); | |
742aa7fb | 5285 | } |
1da177e4 LT |
5286 | EXPORT_SYMBOL(__free_pages); |
5287 | ||
920c7a5d | 5288 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
5289 | { |
5290 | if (addr != 0) { | |
725d704e | 5291 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
5292 | __free_pages(virt_to_page((void *)addr), order); |
5293 | } | |
5294 | } | |
5295 | ||
5296 | EXPORT_SYMBOL(free_pages); | |
5297 | ||
b63ae8ca AD |
5298 | /* |
5299 | * Page Fragment: | |
5300 | * An arbitrary-length arbitrary-offset area of memory which resides | |
5301 | * within a 0 or higher order page. Multiple fragments within that page | |
5302 | * are individually refcounted, in the page's reference counter. | |
5303 | * | |
5304 | * The page_frag functions below provide a simple allocation framework for | |
5305 | * page fragments. This is used by the network stack and network device | |
5306 | * drivers to provide a backing region of memory for use as either an | |
5307 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
5308 | */ | |
2976db80 AD |
5309 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
5310 | gfp_t gfp_mask) | |
b63ae8ca AD |
5311 | { |
5312 | struct page *page = NULL; | |
5313 | gfp_t gfp = gfp_mask; | |
5314 | ||
5315 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5316 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
5317 | __GFP_NOMEMALLOC; | |
5318 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
5319 | PAGE_FRAG_CACHE_MAX_ORDER); | |
5320 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
5321 | #endif | |
5322 | if (unlikely(!page)) | |
5323 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
5324 | ||
5325 | nc->va = page ? page_address(page) : NULL; | |
5326 | ||
5327 | return page; | |
5328 | } | |
5329 | ||
2976db80 | 5330 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
5331 | { |
5332 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
5333 | ||
742aa7fb AL |
5334 | if (page_ref_sub_and_test(page, count)) |
5335 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 5336 | } |
2976db80 | 5337 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 5338 | |
b358e212 KH |
5339 | void *page_frag_alloc_align(struct page_frag_cache *nc, |
5340 | unsigned int fragsz, gfp_t gfp_mask, | |
5341 | unsigned int align_mask) | |
b63ae8ca AD |
5342 | { |
5343 | unsigned int size = PAGE_SIZE; | |
5344 | struct page *page; | |
5345 | int offset; | |
5346 | ||
5347 | if (unlikely(!nc->va)) { | |
5348 | refill: | |
2976db80 | 5349 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
5350 | if (!page) |
5351 | return NULL; | |
5352 | ||
5353 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5354 | /* if size can vary use size else just use PAGE_SIZE */ | |
5355 | size = nc->size; | |
5356 | #endif | |
5357 | /* Even if we own the page, we do not use atomic_set(). | |
5358 | * This would break get_page_unless_zero() users. | |
5359 | */ | |
86447726 | 5360 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
5361 | |
5362 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 5363 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 5364 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5365 | nc->offset = size; |
5366 | } | |
5367 | ||
5368 | offset = nc->offset - fragsz; | |
5369 | if (unlikely(offset < 0)) { | |
5370 | page = virt_to_page(nc->va); | |
5371 | ||
fe896d18 | 5372 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
5373 | goto refill; |
5374 | ||
d8c19014 DZ |
5375 | if (unlikely(nc->pfmemalloc)) { |
5376 | free_the_page(page, compound_order(page)); | |
5377 | goto refill; | |
5378 | } | |
5379 | ||
b63ae8ca AD |
5380 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
5381 | /* if size can vary use size else just use PAGE_SIZE */ | |
5382 | size = nc->size; | |
5383 | #endif | |
5384 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 5385 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
5386 | |
5387 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 5388 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5389 | offset = size - fragsz; |
5390 | } | |
5391 | ||
5392 | nc->pagecnt_bias--; | |
b358e212 | 5393 | offset &= align_mask; |
b63ae8ca AD |
5394 | nc->offset = offset; |
5395 | ||
5396 | return nc->va + offset; | |
5397 | } | |
b358e212 | 5398 | EXPORT_SYMBOL(page_frag_alloc_align); |
b63ae8ca AD |
5399 | |
5400 | /* | |
5401 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
5402 | */ | |
8c2dd3e4 | 5403 | void page_frag_free(void *addr) |
b63ae8ca AD |
5404 | { |
5405 | struct page *page = virt_to_head_page(addr); | |
5406 | ||
742aa7fb AL |
5407 | if (unlikely(put_page_testzero(page))) |
5408 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 5409 | } |
8c2dd3e4 | 5410 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 5411 | |
d00181b9 KS |
5412 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
5413 | size_t size) | |
ee85c2e1 AK |
5414 | { |
5415 | if (addr) { | |
5416 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
5417 | unsigned long used = addr + PAGE_ALIGN(size); | |
5418 | ||
5419 | split_page(virt_to_page((void *)addr), order); | |
5420 | while (used < alloc_end) { | |
5421 | free_page(used); | |
5422 | used += PAGE_SIZE; | |
5423 | } | |
5424 | } | |
5425 | return (void *)addr; | |
5426 | } | |
5427 | ||
2be0ffe2 TT |
5428 | /** |
5429 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
5430 | * @size: the number of bytes to allocate | |
63931eb9 | 5431 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
5432 | * |
5433 | * This function is similar to alloc_pages(), except that it allocates the | |
5434 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
5435 | * allocate memory in power-of-two pages. | |
5436 | * | |
5437 | * This function is also limited by MAX_ORDER. | |
5438 | * | |
5439 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
5440 | * |
5441 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
5442 | */ |
5443 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
5444 | { | |
5445 | unsigned int order = get_order(size); | |
5446 | unsigned long addr; | |
5447 | ||
63931eb9 VB |
5448 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) |
5449 | gfp_mask &= ~__GFP_COMP; | |
5450 | ||
2be0ffe2 | 5451 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 5452 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
5453 | } |
5454 | EXPORT_SYMBOL(alloc_pages_exact); | |
5455 | ||
ee85c2e1 AK |
5456 | /** |
5457 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
5458 | * pages on a node. | |
b5e6ab58 | 5459 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 5460 | * @size: the number of bytes to allocate |
63931eb9 | 5461 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
5462 | * |
5463 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
5464 | * back. | |
a862f68a MR |
5465 | * |
5466 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 5467 | */ |
e1931811 | 5468 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5469 | { |
d00181b9 | 5470 | unsigned int order = get_order(size); |
63931eb9 VB |
5471 | struct page *p; |
5472 | ||
5473 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) | |
5474 | gfp_mask &= ~__GFP_COMP; | |
5475 | ||
5476 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5477 | if (!p) |
5478 | return NULL; | |
5479 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5480 | } | |
ee85c2e1 | 5481 | |
2be0ffe2 TT |
5482 | /** |
5483 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5484 | * @virt: the value returned by alloc_pages_exact. | |
5485 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5486 | * | |
5487 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5488 | */ | |
5489 | void free_pages_exact(void *virt, size_t size) | |
5490 | { | |
5491 | unsigned long addr = (unsigned long)virt; | |
5492 | unsigned long end = addr + PAGE_ALIGN(size); | |
5493 | ||
5494 | while (addr < end) { | |
5495 | free_page(addr); | |
5496 | addr += PAGE_SIZE; | |
5497 | } | |
5498 | } | |
5499 | EXPORT_SYMBOL(free_pages_exact); | |
5500 | ||
e0fb5815 ZY |
5501 | /** |
5502 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5503 | * @offset: The zone index of the highest zone | |
5504 | * | |
a862f68a | 5505 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5506 | * high watermark within all zones at or below a given zone index. For each |
5507 | * zone, the number of pages is calculated as: | |
0e056eb5 | 5508 | * |
5509 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5510 | * |
5511 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5512 | */ |
ebec3862 | 5513 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5514 | { |
dd1a239f | 5515 | struct zoneref *z; |
54a6eb5c MG |
5516 | struct zone *zone; |
5517 | ||
e310fd43 | 5518 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5519 | unsigned long sum = 0; |
1da177e4 | 5520 | |
0e88460d | 5521 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5522 | |
54a6eb5c | 5523 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5524 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5525 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5526 | if (size > high) |
5527 | sum += size - high; | |
1da177e4 LT |
5528 | } |
5529 | ||
5530 | return sum; | |
5531 | } | |
5532 | ||
e0fb5815 ZY |
5533 | /** |
5534 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5535 | * | |
5536 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5537 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5538 | * |
5539 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5540 | * ZONE_NORMAL. | |
1da177e4 | 5541 | */ |
ebec3862 | 5542 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5543 | { |
af4ca457 | 5544 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5545 | } |
c2f1a551 | 5546 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5547 | |
08e0f6a9 | 5548 | static inline void show_node(struct zone *zone) |
1da177e4 | 5549 | { |
e5adfffc | 5550 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5551 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5552 | } |
1da177e4 | 5553 | |
d02bd27b IR |
5554 | long si_mem_available(void) |
5555 | { | |
5556 | long available; | |
5557 | unsigned long pagecache; | |
5558 | unsigned long wmark_low = 0; | |
5559 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5560 | unsigned long reclaimable; |
d02bd27b IR |
5561 | struct zone *zone; |
5562 | int lru; | |
5563 | ||
5564 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5565 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5566 | |
5567 | for_each_zone(zone) | |
a9214443 | 5568 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5569 | |
5570 | /* | |
5571 | * Estimate the amount of memory available for userspace allocations, | |
5572 | * without causing swapping. | |
5573 | */ | |
c41f012a | 5574 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5575 | |
5576 | /* | |
5577 | * Not all the page cache can be freed, otherwise the system will | |
5578 | * start swapping. Assume at least half of the page cache, or the | |
5579 | * low watermark worth of cache, needs to stay. | |
5580 | */ | |
5581 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
5582 | pagecache -= min(pagecache / 2, wmark_low); | |
5583 | available += pagecache; | |
5584 | ||
5585 | /* | |
b29940c1 VB |
5586 | * Part of the reclaimable slab and other kernel memory consists of |
5587 | * items that are in use, and cannot be freed. Cap this estimate at the | |
5588 | * low watermark. | |
d02bd27b | 5589 | */ |
d42f3245 RG |
5590 | reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) + |
5591 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
b29940c1 | 5592 | available += reclaimable - min(reclaimable / 2, wmark_low); |
034ebf65 | 5593 | |
d02bd27b IR |
5594 | if (available < 0) |
5595 | available = 0; | |
5596 | return available; | |
5597 | } | |
5598 | EXPORT_SYMBOL_GPL(si_mem_available); | |
5599 | ||
1da177e4 LT |
5600 | void si_meminfo(struct sysinfo *val) |
5601 | { | |
ca79b0c2 | 5602 | val->totalram = totalram_pages(); |
11fb9989 | 5603 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 5604 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 5605 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 5606 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 5607 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
5608 | val->mem_unit = PAGE_SIZE; |
5609 | } | |
5610 | ||
5611 | EXPORT_SYMBOL(si_meminfo); | |
5612 | ||
5613 | #ifdef CONFIG_NUMA | |
5614 | void si_meminfo_node(struct sysinfo *val, int nid) | |
5615 | { | |
cdd91a77 JL |
5616 | int zone_type; /* needs to be signed */ |
5617 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
5618 | unsigned long managed_highpages = 0; |
5619 | unsigned long free_highpages = 0; | |
1da177e4 LT |
5620 | pg_data_t *pgdat = NODE_DATA(nid); |
5621 | ||
cdd91a77 | 5622 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 5623 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 5624 | val->totalram = managed_pages; |
11fb9989 | 5625 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 5626 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 5627 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
5628 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
5629 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
5630 | ||
5631 | if (is_highmem(zone)) { | |
9705bea5 | 5632 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
5633 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
5634 | } | |
5635 | } | |
5636 | val->totalhigh = managed_highpages; | |
5637 | val->freehigh = free_highpages; | |
98d2b0eb | 5638 | #else |
fc2bd799 JK |
5639 | val->totalhigh = managed_highpages; |
5640 | val->freehigh = free_highpages; | |
98d2b0eb | 5641 | #endif |
1da177e4 LT |
5642 | val->mem_unit = PAGE_SIZE; |
5643 | } | |
5644 | #endif | |
5645 | ||
ddd588b5 | 5646 | /* |
7bf02ea2 DR |
5647 | * Determine whether the node should be displayed or not, depending on whether |
5648 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 5649 | */ |
9af744d7 | 5650 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 5651 | { |
ddd588b5 | 5652 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 5653 | return false; |
ddd588b5 | 5654 | |
9af744d7 MH |
5655 | /* |
5656 | * no node mask - aka implicit memory numa policy. Do not bother with | |
5657 | * the synchronization - read_mems_allowed_begin - because we do not | |
5658 | * have to be precise here. | |
5659 | */ | |
5660 | if (!nodemask) | |
5661 | nodemask = &cpuset_current_mems_allowed; | |
5662 | ||
5663 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
5664 | } |
5665 | ||
1da177e4 LT |
5666 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
5667 | ||
377e4f16 RV |
5668 | static void show_migration_types(unsigned char type) |
5669 | { | |
5670 | static const char types[MIGRATE_TYPES] = { | |
5671 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 5672 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
5673 | [MIGRATE_RECLAIMABLE] = 'E', |
5674 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
5675 | #ifdef CONFIG_CMA |
5676 | [MIGRATE_CMA] = 'C', | |
5677 | #endif | |
194159fb | 5678 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 5679 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 5680 | #endif |
377e4f16 RV |
5681 | }; |
5682 | char tmp[MIGRATE_TYPES + 1]; | |
5683 | char *p = tmp; | |
5684 | int i; | |
5685 | ||
5686 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
5687 | if (type & (1 << i)) | |
5688 | *p++ = types[i]; | |
5689 | } | |
5690 | ||
5691 | *p = '\0'; | |
1f84a18f | 5692 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
5693 | } |
5694 | ||
1da177e4 LT |
5695 | /* |
5696 | * Show free area list (used inside shift_scroll-lock stuff) | |
5697 | * We also calculate the percentage fragmentation. We do this by counting the | |
5698 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
5699 | * |
5700 | * Bits in @filter: | |
5701 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
5702 | * cpuset. | |
1da177e4 | 5703 | */ |
9af744d7 | 5704 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 5705 | { |
d1bfcdb8 | 5706 | unsigned long free_pcp = 0; |
c7241913 | 5707 | int cpu; |
1da177e4 | 5708 | struct zone *zone; |
599d0c95 | 5709 | pg_data_t *pgdat; |
1da177e4 | 5710 | |
ee99c71c | 5711 | for_each_populated_zone(zone) { |
9af744d7 | 5712 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5713 | continue; |
d1bfcdb8 | 5714 | |
761b0677 KK |
5715 | for_each_online_cpu(cpu) |
5716 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
5717 | } |
5718 | ||
a731286d KM |
5719 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
5720 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
8d92890b | 5721 | " unevictable:%lu dirty:%lu writeback:%lu\n" |
d1bfcdb8 | 5722 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a | 5723 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 5724 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
5725 | global_node_page_state(NR_ACTIVE_ANON), |
5726 | global_node_page_state(NR_INACTIVE_ANON), | |
5727 | global_node_page_state(NR_ISOLATED_ANON), | |
5728 | global_node_page_state(NR_ACTIVE_FILE), | |
5729 | global_node_page_state(NR_INACTIVE_FILE), | |
5730 | global_node_page_state(NR_ISOLATED_FILE), | |
5731 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
5732 | global_node_page_state(NR_FILE_DIRTY), |
5733 | global_node_page_state(NR_WRITEBACK), | |
d42f3245 RG |
5734 | global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B), |
5735 | global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B), | |
50658e2e | 5736 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 5737 | global_node_page_state(NR_SHMEM), |
f0c0c115 | 5738 | global_node_page_state(NR_PAGETABLE), |
c41f012a MH |
5739 | global_zone_page_state(NR_BOUNCE), |
5740 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 5741 | free_pcp, |
c41f012a | 5742 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 5743 | |
599d0c95 | 5744 | for_each_online_pgdat(pgdat) { |
9af744d7 | 5745 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
5746 | continue; |
5747 | ||
599d0c95 MG |
5748 | printk("Node %d" |
5749 | " active_anon:%lukB" | |
5750 | " inactive_anon:%lukB" | |
5751 | " active_file:%lukB" | |
5752 | " inactive_file:%lukB" | |
5753 | " unevictable:%lukB" | |
5754 | " isolated(anon):%lukB" | |
5755 | " isolated(file):%lukB" | |
50658e2e | 5756 | " mapped:%lukB" |
11fb9989 MG |
5757 | " dirty:%lukB" |
5758 | " writeback:%lukB" | |
5759 | " shmem:%lukB" | |
5760 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5761 | " shmem_thp: %lukB" | |
5762 | " shmem_pmdmapped: %lukB" | |
5763 | " anon_thp: %lukB" | |
5764 | #endif | |
5765 | " writeback_tmp:%lukB" | |
991e7673 SB |
5766 | " kernel_stack:%lukB" |
5767 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5768 | " shadow_call_stack:%lukB" | |
5769 | #endif | |
f0c0c115 | 5770 | " pagetables:%lukB" |
599d0c95 MG |
5771 | " all_unreclaimable? %s" |
5772 | "\n", | |
5773 | pgdat->node_id, | |
5774 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
5775 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
5776 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
5777 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
5778 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
5779 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
5780 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 5781 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
5782 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
5783 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 5784 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 | 5785 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
57b2847d | 5786 | K(node_page_state(pgdat, NR_SHMEM_THPS)), |
a1528e21 | 5787 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)), |
69473e5d | 5788 | K(node_page_state(pgdat, NR_ANON_THPS)), |
11fb9989 | 5789 | #endif |
11fb9989 | 5790 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
991e7673 SB |
5791 | node_page_state(pgdat, NR_KERNEL_STACK_KB), |
5792 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5793 | node_page_state(pgdat, NR_KERNEL_SCS_KB), | |
5794 | #endif | |
f0c0c115 | 5795 | K(node_page_state(pgdat, NR_PAGETABLE)), |
c73322d0 JW |
5796 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
5797 | "yes" : "no"); | |
599d0c95 MG |
5798 | } |
5799 | ||
ee99c71c | 5800 | for_each_populated_zone(zone) { |
1da177e4 LT |
5801 | int i; |
5802 | ||
9af744d7 | 5803 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5804 | continue; |
d1bfcdb8 KK |
5805 | |
5806 | free_pcp = 0; | |
5807 | for_each_online_cpu(cpu) | |
5808 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
5809 | ||
1da177e4 | 5810 | show_node(zone); |
1f84a18f JP |
5811 | printk(KERN_CONT |
5812 | "%s" | |
1da177e4 LT |
5813 | " free:%lukB" |
5814 | " min:%lukB" | |
5815 | " low:%lukB" | |
5816 | " high:%lukB" | |
e47b346a | 5817 | " reserved_highatomic:%luKB" |
71c799f4 MK |
5818 | " active_anon:%lukB" |
5819 | " inactive_anon:%lukB" | |
5820 | " active_file:%lukB" | |
5821 | " inactive_file:%lukB" | |
5822 | " unevictable:%lukB" | |
5a1c84b4 | 5823 | " writepending:%lukB" |
1da177e4 | 5824 | " present:%lukB" |
9feedc9d | 5825 | " managed:%lukB" |
4a0aa73f | 5826 | " mlocked:%lukB" |
4a0aa73f | 5827 | " bounce:%lukB" |
d1bfcdb8 KK |
5828 | " free_pcp:%lukB" |
5829 | " local_pcp:%ukB" | |
d1ce749a | 5830 | " free_cma:%lukB" |
1da177e4 LT |
5831 | "\n", |
5832 | zone->name, | |
88f5acf8 | 5833 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
5834 | K(min_wmark_pages(zone)), |
5835 | K(low_wmark_pages(zone)), | |
5836 | K(high_wmark_pages(zone)), | |
e47b346a | 5837 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
5838 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
5839 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
5840 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
5841 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
5842 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 5843 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 5844 | K(zone->present_pages), |
9705bea5 | 5845 | K(zone_managed_pages(zone)), |
4a0aa73f | 5846 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f | 5847 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
5848 | K(free_pcp), |
5849 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 5850 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
5851 | printk("lowmem_reserve[]:"); |
5852 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
5853 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5854 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5855 | } |
5856 | ||
ee99c71c | 5857 | for_each_populated_zone(zone) { |
d00181b9 KS |
5858 | unsigned int order; |
5859 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5860 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5861 | |
9af744d7 | 5862 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5863 | continue; |
1da177e4 | 5864 | show_node(zone); |
1f84a18f | 5865 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5866 | |
5867 | spin_lock_irqsave(&zone->lock, flags); | |
5868 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5869 | struct free_area *area = &zone->free_area[order]; |
5870 | int type; | |
5871 | ||
5872 | nr[order] = area->nr_free; | |
8f9de51a | 5873 | total += nr[order] << order; |
377e4f16 RV |
5874 | |
5875 | types[order] = 0; | |
5876 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 5877 | if (!free_area_empty(area, type)) |
377e4f16 RV |
5878 | types[order] |= 1 << type; |
5879 | } | |
1da177e4 LT |
5880 | } |
5881 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5882 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5883 | printk(KERN_CONT "%lu*%lukB ", |
5884 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5885 | if (nr[order]) |
5886 | show_migration_types(types[order]); | |
5887 | } | |
1f84a18f | 5888 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5889 | } |
5890 | ||
949f7ec5 DR |
5891 | hugetlb_show_meminfo(); |
5892 | ||
11fb9989 | 5893 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5894 | |
1da177e4 LT |
5895 | show_swap_cache_info(); |
5896 | } | |
5897 | ||
19770b32 MG |
5898 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5899 | { | |
5900 | zoneref->zone = zone; | |
5901 | zoneref->zone_idx = zone_idx(zone); | |
5902 | } | |
5903 | ||
1da177e4 LT |
5904 | /* |
5905 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5906 | * |
5907 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5908 | */ |
9d3be21b | 5909 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5910 | { |
1a93205b | 5911 | struct zone *zone; |
bc732f1d | 5912 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5913 | int nr_zones = 0; |
02a68a5e CL |
5914 | |
5915 | do { | |
2f6726e5 | 5916 | zone_type--; |
070f8032 | 5917 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5918 | if (managed_zone(zone)) { |
9d3be21b | 5919 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5920 | check_highest_zone(zone_type); |
1da177e4 | 5921 | } |
2f6726e5 | 5922 | } while (zone_type); |
bc732f1d | 5923 | |
070f8032 | 5924 | return nr_zones; |
1da177e4 LT |
5925 | } |
5926 | ||
5927 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5928 | |
5929 | static int __parse_numa_zonelist_order(char *s) | |
5930 | { | |
c9bff3ee | 5931 | /* |
f0953a1b | 5932 | * We used to support different zonelists modes but they turned |
c9bff3ee MH |
5933 | * out to be just not useful. Let's keep the warning in place |
5934 | * if somebody still use the cmd line parameter so that we do | |
5935 | * not fail it silently | |
5936 | */ | |
5937 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5938 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5939 | return -EINVAL; |
5940 | } | |
5941 | return 0; | |
5942 | } | |
5943 | ||
c9bff3ee MH |
5944 | char numa_zonelist_order[] = "Node"; |
5945 | ||
f0c0b2b8 KH |
5946 | /* |
5947 | * sysctl handler for numa_zonelist_order | |
5948 | */ | |
cccad5b9 | 5949 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 5950 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 5951 | { |
32927393 CH |
5952 | if (write) |
5953 | return __parse_numa_zonelist_order(buffer); | |
5954 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
5955 | } |
5956 | ||
5957 | ||
62bc62a8 | 5958 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5959 | static int node_load[MAX_NUMNODES]; |
5960 | ||
1da177e4 | 5961 | /** |
4dc3b16b | 5962 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5963 | * @node: node whose fallback list we're appending |
5964 | * @used_node_mask: nodemask_t of already used nodes | |
5965 | * | |
5966 | * We use a number of factors to determine which is the next node that should | |
5967 | * appear on a given node's fallback list. The node should not have appeared | |
5968 | * already in @node's fallback list, and it should be the next closest node | |
5969 | * according to the distance array (which contains arbitrary distance values | |
5970 | * from each node to each node in the system), and should also prefer nodes | |
5971 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5972 | * on them otherwise. | |
a862f68a MR |
5973 | * |
5974 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 5975 | */ |
f0c0b2b8 | 5976 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5977 | { |
4cf808eb | 5978 | int n, val; |
1da177e4 | 5979 | int min_val = INT_MAX; |
00ef2d2f | 5980 | int best_node = NUMA_NO_NODE; |
1da177e4 | 5981 | |
4cf808eb LT |
5982 | /* Use the local node if we haven't already */ |
5983 | if (!node_isset(node, *used_node_mask)) { | |
5984 | node_set(node, *used_node_mask); | |
5985 | return node; | |
5986 | } | |
1da177e4 | 5987 | |
4b0ef1fe | 5988 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5989 | |
5990 | /* Don't want a node to appear more than once */ | |
5991 | if (node_isset(n, *used_node_mask)) | |
5992 | continue; | |
5993 | ||
1da177e4 LT |
5994 | /* Use the distance array to find the distance */ |
5995 | val = node_distance(node, n); | |
5996 | ||
4cf808eb LT |
5997 | /* Penalize nodes under us ("prefer the next node") */ |
5998 | val += (n < node); | |
5999 | ||
1da177e4 | 6000 | /* Give preference to headless and unused nodes */ |
b630749f | 6001 | if (!cpumask_empty(cpumask_of_node(n))) |
1da177e4 LT |
6002 | val += PENALTY_FOR_NODE_WITH_CPUS; |
6003 | ||
6004 | /* Slight preference for less loaded node */ | |
6005 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
6006 | val += node_load[n]; | |
6007 | ||
6008 | if (val < min_val) { | |
6009 | min_val = val; | |
6010 | best_node = n; | |
6011 | } | |
6012 | } | |
6013 | ||
6014 | if (best_node >= 0) | |
6015 | node_set(best_node, *used_node_mask); | |
6016 | ||
6017 | return best_node; | |
6018 | } | |
6019 | ||
f0c0b2b8 KH |
6020 | |
6021 | /* | |
6022 | * Build zonelists ordered by node and zones within node. | |
6023 | * This results in maximum locality--normal zone overflows into local | |
6024 | * DMA zone, if any--but risks exhausting DMA zone. | |
6025 | */ | |
9d3be21b MH |
6026 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
6027 | unsigned nr_nodes) | |
1da177e4 | 6028 | { |
9d3be21b MH |
6029 | struct zoneref *zonerefs; |
6030 | int i; | |
6031 | ||
6032 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
6033 | ||
6034 | for (i = 0; i < nr_nodes; i++) { | |
6035 | int nr_zones; | |
6036 | ||
6037 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 6038 | |
9d3be21b MH |
6039 | nr_zones = build_zonerefs_node(node, zonerefs); |
6040 | zonerefs += nr_zones; | |
6041 | } | |
6042 | zonerefs->zone = NULL; | |
6043 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
6044 | } |
6045 | ||
523b9458 CL |
6046 | /* |
6047 | * Build gfp_thisnode zonelists | |
6048 | */ | |
6049 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
6050 | { | |
9d3be21b MH |
6051 | struct zoneref *zonerefs; |
6052 | int nr_zones; | |
523b9458 | 6053 | |
9d3be21b MH |
6054 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
6055 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6056 | zonerefs += nr_zones; | |
6057 | zonerefs->zone = NULL; | |
6058 | zonerefs->zone_idx = 0; | |
523b9458 CL |
6059 | } |
6060 | ||
f0c0b2b8 KH |
6061 | /* |
6062 | * Build zonelists ordered by zone and nodes within zones. | |
6063 | * This results in conserving DMA zone[s] until all Normal memory is | |
6064 | * exhausted, but results in overflowing to remote node while memory | |
6065 | * may still exist in local DMA zone. | |
6066 | */ | |
f0c0b2b8 | 6067 | |
f0c0b2b8 KH |
6068 | static void build_zonelists(pg_data_t *pgdat) |
6069 | { | |
9d3be21b MH |
6070 | static int node_order[MAX_NUMNODES]; |
6071 | int node, load, nr_nodes = 0; | |
d0ddf49b | 6072 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 6073 | int local_node, prev_node; |
1da177e4 LT |
6074 | |
6075 | /* NUMA-aware ordering of nodes */ | |
6076 | local_node = pgdat->node_id; | |
62bc62a8 | 6077 | load = nr_online_nodes; |
1da177e4 | 6078 | prev_node = local_node; |
f0c0b2b8 | 6079 | |
f0c0b2b8 | 6080 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
6081 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
6082 | /* | |
6083 | * We don't want to pressure a particular node. | |
6084 | * So adding penalty to the first node in same | |
6085 | * distance group to make it round-robin. | |
6086 | */ | |
957f822a DR |
6087 | if (node_distance(local_node, node) != |
6088 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
6089 | node_load[node] = load; |
6090 | ||
9d3be21b | 6091 | node_order[nr_nodes++] = node; |
1da177e4 LT |
6092 | prev_node = node; |
6093 | load--; | |
1da177e4 | 6094 | } |
523b9458 | 6095 | |
9d3be21b | 6096 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 6097 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
6098 | } |
6099 | ||
7aac7898 LS |
6100 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6101 | /* | |
6102 | * Return node id of node used for "local" allocations. | |
6103 | * I.e., first node id of first zone in arg node's generic zonelist. | |
6104 | * Used for initializing percpu 'numa_mem', which is used primarily | |
6105 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
6106 | */ | |
6107 | int local_memory_node(int node) | |
6108 | { | |
c33d6c06 | 6109 | struct zoneref *z; |
7aac7898 | 6110 | |
c33d6c06 | 6111 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 6112 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 6113 | NULL); |
c1093b74 | 6114 | return zone_to_nid(z->zone); |
7aac7898 LS |
6115 | } |
6116 | #endif | |
f0c0b2b8 | 6117 | |
6423aa81 JK |
6118 | static void setup_min_unmapped_ratio(void); |
6119 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
6120 | #else /* CONFIG_NUMA */ |
6121 | ||
f0c0b2b8 | 6122 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 6123 | { |
19655d34 | 6124 | int node, local_node; |
9d3be21b MH |
6125 | struct zoneref *zonerefs; |
6126 | int nr_zones; | |
1da177e4 LT |
6127 | |
6128 | local_node = pgdat->node_id; | |
1da177e4 | 6129 | |
9d3be21b MH |
6130 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
6131 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6132 | zonerefs += nr_zones; | |
1da177e4 | 6133 | |
54a6eb5c MG |
6134 | /* |
6135 | * Now we build the zonelist so that it contains the zones | |
6136 | * of all the other nodes. | |
6137 | * We don't want to pressure a particular node, so when | |
6138 | * building the zones for node N, we make sure that the | |
6139 | * zones coming right after the local ones are those from | |
6140 | * node N+1 (modulo N) | |
6141 | */ | |
6142 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
6143 | if (!node_online(node)) | |
6144 | continue; | |
9d3be21b MH |
6145 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6146 | zonerefs += nr_zones; | |
1da177e4 | 6147 | } |
54a6eb5c MG |
6148 | for (node = 0; node < local_node; node++) { |
6149 | if (!node_online(node)) | |
6150 | continue; | |
9d3be21b MH |
6151 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6152 | zonerefs += nr_zones; | |
54a6eb5c MG |
6153 | } |
6154 | ||
9d3be21b MH |
6155 | zonerefs->zone = NULL; |
6156 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
6157 | } |
6158 | ||
6159 | #endif /* CONFIG_NUMA */ | |
6160 | ||
99dcc3e5 CL |
6161 | /* |
6162 | * Boot pageset table. One per cpu which is going to be used for all | |
6163 | * zones and all nodes. The parameters will be set in such a way | |
6164 | * that an item put on a list will immediately be handed over to | |
6165 | * the buddy list. This is safe since pageset manipulation is done | |
6166 | * with interrupts disabled. | |
6167 | * | |
6168 | * The boot_pagesets must be kept even after bootup is complete for | |
6169 | * unused processors and/or zones. They do play a role for bootstrapping | |
6170 | * hotplugged processors. | |
6171 | * | |
6172 | * zoneinfo_show() and maybe other functions do | |
6173 | * not check if the processor is online before following the pageset pointer. | |
6174 | * Other parts of the kernel may not check if the zone is available. | |
6175 | */ | |
69a8396a | 6176 | static void pageset_init(struct per_cpu_pageset *p); |
952eaf81 VB |
6177 | /* These effectively disable the pcplists in the boot pageset completely */ |
6178 | #define BOOT_PAGESET_HIGH 0 | |
6179 | #define BOOT_PAGESET_BATCH 1 | |
99dcc3e5 | 6180 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); |
385386cf | 6181 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 6182 | |
11cd8638 | 6183 | static void __build_all_zonelists(void *data) |
1da177e4 | 6184 | { |
6811378e | 6185 | int nid; |
afb6ebb3 | 6186 | int __maybe_unused cpu; |
9adb62a5 | 6187 | pg_data_t *self = data; |
b93e0f32 MH |
6188 | static DEFINE_SPINLOCK(lock); |
6189 | ||
6190 | spin_lock(&lock); | |
9276b1bc | 6191 | |
7f9cfb31 BL |
6192 | #ifdef CONFIG_NUMA |
6193 | memset(node_load, 0, sizeof(node_load)); | |
6194 | #endif | |
9adb62a5 | 6195 | |
c1152583 WY |
6196 | /* |
6197 | * This node is hotadded and no memory is yet present. So just | |
6198 | * building zonelists is fine - no need to touch other nodes. | |
6199 | */ | |
9adb62a5 JL |
6200 | if (self && !node_online(self->node_id)) { |
6201 | build_zonelists(self); | |
c1152583 WY |
6202 | } else { |
6203 | for_each_online_node(nid) { | |
6204 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 6205 | |
c1152583 WY |
6206 | build_zonelists(pgdat); |
6207 | } | |
99dcc3e5 | 6208 | |
7aac7898 LS |
6209 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6210 | /* | |
6211 | * We now know the "local memory node" for each node-- | |
6212 | * i.e., the node of the first zone in the generic zonelist. | |
6213 | * Set up numa_mem percpu variable for on-line cpus. During | |
6214 | * boot, only the boot cpu should be on-line; we'll init the | |
6215 | * secondary cpus' numa_mem as they come on-line. During | |
6216 | * node/memory hotplug, we'll fixup all on-line cpus. | |
6217 | */ | |
d9c9a0b9 | 6218 | for_each_online_cpu(cpu) |
7aac7898 | 6219 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 6220 | #endif |
d9c9a0b9 | 6221 | } |
b93e0f32 MH |
6222 | |
6223 | spin_unlock(&lock); | |
6811378e YG |
6224 | } |
6225 | ||
061f67bc RV |
6226 | static noinline void __init |
6227 | build_all_zonelists_init(void) | |
6228 | { | |
afb6ebb3 MH |
6229 | int cpu; |
6230 | ||
061f67bc | 6231 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
6232 | |
6233 | /* | |
6234 | * Initialize the boot_pagesets that are going to be used | |
6235 | * for bootstrapping processors. The real pagesets for | |
6236 | * each zone will be allocated later when the per cpu | |
6237 | * allocator is available. | |
6238 | * | |
6239 | * boot_pagesets are used also for bootstrapping offline | |
6240 | * cpus if the system is already booted because the pagesets | |
6241 | * are needed to initialize allocators on a specific cpu too. | |
6242 | * F.e. the percpu allocator needs the page allocator which | |
6243 | * needs the percpu allocator in order to allocate its pagesets | |
6244 | * (a chicken-egg dilemma). | |
6245 | */ | |
6246 | for_each_possible_cpu(cpu) | |
69a8396a | 6247 | pageset_init(&per_cpu(boot_pageset, cpu)); |
afb6ebb3 | 6248 | |
061f67bc RV |
6249 | mminit_verify_zonelist(); |
6250 | cpuset_init_current_mems_allowed(); | |
6251 | } | |
6252 | ||
4eaf3f64 | 6253 | /* |
4eaf3f64 | 6254 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 6255 | * |
72675e13 | 6256 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 6257 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 6258 | */ |
72675e13 | 6259 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 6260 | { |
0a18e607 DH |
6261 | unsigned long vm_total_pages; |
6262 | ||
6811378e | 6263 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 6264 | build_all_zonelists_init(); |
6811378e | 6265 | } else { |
11cd8638 | 6266 | __build_all_zonelists(pgdat); |
6811378e YG |
6267 | /* cpuset refresh routine should be here */ |
6268 | } | |
56b9413b DH |
6269 | /* Get the number of free pages beyond high watermark in all zones. */ |
6270 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
6271 | /* |
6272 | * Disable grouping by mobility if the number of pages in the | |
6273 | * system is too low to allow the mechanism to work. It would be | |
6274 | * more accurate, but expensive to check per-zone. This check is | |
6275 | * made on memory-hotadd so a system can start with mobility | |
6276 | * disabled and enable it later | |
6277 | */ | |
d9c23400 | 6278 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
6279 | page_group_by_mobility_disabled = 1; |
6280 | else | |
6281 | page_group_by_mobility_disabled = 0; | |
6282 | ||
ce0725f7 | 6283 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 6284 | nr_online_nodes, |
756a025f JP |
6285 | page_group_by_mobility_disabled ? "off" : "on", |
6286 | vm_total_pages); | |
f0c0b2b8 | 6287 | #ifdef CONFIG_NUMA |
f88dfff5 | 6288 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 6289 | #endif |
1da177e4 LT |
6290 | } |
6291 | ||
a9a9e77f PT |
6292 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
6293 | static bool __meminit | |
6294 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
6295 | { | |
a9a9e77f PT |
6296 | static struct memblock_region *r; |
6297 | ||
6298 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
6299 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
cc6de168 | 6300 | for_each_mem_region(r) { |
a9a9e77f PT |
6301 | if (*pfn < memblock_region_memory_end_pfn(r)) |
6302 | break; | |
6303 | } | |
6304 | } | |
6305 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
6306 | memblock_is_mirror(r)) { | |
6307 | *pfn = memblock_region_memory_end_pfn(r); | |
6308 | return true; | |
6309 | } | |
6310 | } | |
a9a9e77f PT |
6311 | return false; |
6312 | } | |
6313 | ||
1da177e4 LT |
6314 | /* |
6315 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 6316 | * up by memblock_free_all() once the early boot process is |
1da177e4 | 6317 | * done. Non-atomic initialization, single-pass. |
d882c006 DH |
6318 | * |
6319 | * All aligned pageblocks are initialized to the specified migratetype | |
6320 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
6321 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
1da177e4 | 6322 | */ |
ab28cb6e | 6323 | void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone, |
dc2da7b4 | 6324 | unsigned long start_pfn, unsigned long zone_end_pfn, |
d882c006 DH |
6325 | enum meminit_context context, |
6326 | struct vmem_altmap *altmap, int migratetype) | |
1da177e4 | 6327 | { |
a9a9e77f | 6328 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 6329 | struct page *page; |
1da177e4 | 6330 | |
22b31eec HD |
6331 | if (highest_memmap_pfn < end_pfn - 1) |
6332 | highest_memmap_pfn = end_pfn - 1; | |
6333 | ||
966cf44f | 6334 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
6335 | /* |
6336 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
6337 | * memory. We limit the total number of pages to initialize to just |
6338 | * those that might contain the memory mapping. We will defer the | |
6339 | * ZONE_DEVICE page initialization until after we have released | |
6340 | * the hotplug lock. | |
4b94ffdc | 6341 | */ |
966cf44f AD |
6342 | if (zone == ZONE_DEVICE) { |
6343 | if (!altmap) | |
6344 | return; | |
6345 | ||
6346 | if (start_pfn == altmap->base_pfn) | |
6347 | start_pfn += altmap->reserve; | |
6348 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
6349 | } | |
6350 | #endif | |
4b94ffdc | 6351 | |
948c436e | 6352 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 6353 | /* |
b72d0ffb AM |
6354 | * There can be holes in boot-time mem_map[]s handed to this |
6355 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 6356 | */ |
c1d0da83 | 6357 | if (context == MEMINIT_EARLY) { |
a9a9e77f PT |
6358 | if (overlap_memmap_init(zone, &pfn)) |
6359 | continue; | |
dc2da7b4 | 6360 | if (defer_init(nid, pfn, zone_end_pfn)) |
a9a9e77f | 6361 | break; |
a2f3aa02 | 6362 | } |
ac5d2539 | 6363 | |
d0dc12e8 PT |
6364 | page = pfn_to_page(pfn); |
6365 | __init_single_page(page, pfn, zone, nid); | |
c1d0da83 | 6366 | if (context == MEMINIT_HOTPLUG) |
d483da5b | 6367 | __SetPageReserved(page); |
d0dc12e8 | 6368 | |
ac5d2539 | 6369 | /* |
d882c006 DH |
6370 | * Usually, we want to mark the pageblock MIGRATE_MOVABLE, |
6371 | * such that unmovable allocations won't be scattered all | |
6372 | * over the place during system boot. | |
ac5d2539 | 6373 | */ |
4eb29bd9 | 6374 | if (IS_ALIGNED(pfn, pageblock_nr_pages)) { |
d882c006 | 6375 | set_pageblock_migratetype(page, migratetype); |
9b6e63cb | 6376 | cond_resched(); |
ac5d2539 | 6377 | } |
948c436e | 6378 | pfn++; |
1da177e4 LT |
6379 | } |
6380 | } | |
6381 | ||
966cf44f AD |
6382 | #ifdef CONFIG_ZONE_DEVICE |
6383 | void __ref memmap_init_zone_device(struct zone *zone, | |
6384 | unsigned long start_pfn, | |
1f8d75c1 | 6385 | unsigned long nr_pages, |
966cf44f AD |
6386 | struct dev_pagemap *pgmap) |
6387 | { | |
1f8d75c1 | 6388 | unsigned long pfn, end_pfn = start_pfn + nr_pages; |
966cf44f | 6389 | struct pglist_data *pgdat = zone->zone_pgdat; |
514caf23 | 6390 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
966cf44f AD |
6391 | unsigned long zone_idx = zone_idx(zone); |
6392 | unsigned long start = jiffies; | |
6393 | int nid = pgdat->node_id; | |
6394 | ||
46d945ae | 6395 | if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE)) |
966cf44f AD |
6396 | return; |
6397 | ||
6398 | /* | |
6399 | * The call to memmap_init_zone should have already taken care | |
6400 | * of the pages reserved for the memmap, so we can just jump to | |
6401 | * the end of that region and start processing the device pages. | |
6402 | */ | |
514caf23 | 6403 | if (altmap) { |
966cf44f | 6404 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
1f8d75c1 | 6405 | nr_pages = end_pfn - start_pfn; |
966cf44f AD |
6406 | } |
6407 | ||
6408 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
6409 | struct page *page = pfn_to_page(pfn); | |
6410 | ||
6411 | __init_single_page(page, pfn, zone_idx, nid); | |
6412 | ||
6413 | /* | |
6414 | * Mark page reserved as it will need to wait for onlining | |
6415 | * phase for it to be fully associated with a zone. | |
6416 | * | |
6417 | * We can use the non-atomic __set_bit operation for setting | |
6418 | * the flag as we are still initializing the pages. | |
6419 | */ | |
6420 | __SetPageReserved(page); | |
6421 | ||
6422 | /* | |
8a164fef CH |
6423 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer |
6424 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
6425 | * ever freed or placed on a driver-private list. | |
966cf44f AD |
6426 | */ |
6427 | page->pgmap = pgmap; | |
8a164fef | 6428 | page->zone_device_data = NULL; |
966cf44f AD |
6429 | |
6430 | /* | |
6431 | * Mark the block movable so that blocks are reserved for | |
6432 | * movable at startup. This will force kernel allocations | |
6433 | * to reserve their blocks rather than leaking throughout | |
6434 | * the address space during boot when many long-lived | |
6435 | * kernel allocations are made. | |
6436 | * | |
c1d0da83 | 6437 | * Please note that MEMINIT_HOTPLUG path doesn't clear memmap |
ba72b4c8 | 6438 | * because this is done early in section_activate() |
966cf44f | 6439 | */ |
4eb29bd9 | 6440 | if (IS_ALIGNED(pfn, pageblock_nr_pages)) { |
966cf44f AD |
6441 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
6442 | cond_resched(); | |
6443 | } | |
6444 | } | |
6445 | ||
fdc029b1 | 6446 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
1f8d75c1 | 6447 | nr_pages, jiffies_to_msecs(jiffies - start)); |
966cf44f AD |
6448 | } |
6449 | ||
6450 | #endif | |
1e548deb | 6451 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6452 | { |
7aeb09f9 | 6453 | unsigned int order, t; |
b2a0ac88 MG |
6454 | for_each_migratetype_order(order, t) { |
6455 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6456 | zone->free_area[order].nr_free = 0; |
6457 | } | |
6458 | } | |
6459 | ||
0740a50b MR |
6460 | #if !defined(CONFIG_FLAT_NODE_MEM_MAP) |
6461 | /* | |
6462 | * Only struct pages that correspond to ranges defined by memblock.memory | |
6463 | * are zeroed and initialized by going through __init_single_page() during | |
6464 | * memmap_init_zone(). | |
6465 | * | |
6466 | * But, there could be struct pages that correspond to holes in | |
6467 | * memblock.memory. This can happen because of the following reasons: | |
6468 | * - physical memory bank size is not necessarily the exact multiple of the | |
6469 | * arbitrary section size | |
6470 | * - early reserved memory may not be listed in memblock.memory | |
6471 | * - memory layouts defined with memmap= kernel parameter may not align | |
6472 | * nicely with memmap sections | |
6473 | * | |
6474 | * Explicitly initialize those struct pages so that: | |
6475 | * - PG_Reserved is set | |
6476 | * - zone and node links point to zone and node that span the page if the | |
6477 | * hole is in the middle of a zone | |
6478 | * - zone and node links point to adjacent zone/node if the hole falls on | |
6479 | * the zone boundary; the pages in such holes will be prepended to the | |
6480 | * zone/node above the hole except for the trailing pages in the last | |
6481 | * section that will be appended to the zone/node below. | |
6482 | */ | |
6483 | static u64 __meminit init_unavailable_range(unsigned long spfn, | |
6484 | unsigned long epfn, | |
6485 | int zone, int node) | |
6486 | { | |
6487 | unsigned long pfn; | |
6488 | u64 pgcnt = 0; | |
6489 | ||
6490 | for (pfn = spfn; pfn < epfn; pfn++) { | |
6491 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { | |
6492 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6493 | + pageblock_nr_pages - 1; | |
6494 | continue; | |
6495 | } | |
6496 | __init_single_page(pfn_to_page(pfn), pfn, zone, node); | |
6497 | __SetPageReserved(pfn_to_page(pfn)); | |
6498 | pgcnt++; | |
6499 | } | |
6500 | ||
6501 | return pgcnt; | |
6502 | } | |
6503 | #else | |
6504 | static inline u64 init_unavailable_range(unsigned long spfn, unsigned long epfn, | |
6505 | int zone, int node) | |
6506 | { | |
6507 | return 0; | |
6508 | } | |
6509 | #endif | |
6510 | ||
3256ff83 | 6511 | void __meminit __weak memmap_init_zone(struct zone *zone) |
dfb3ccd0 | 6512 | { |
3256ff83 BH |
6513 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
6514 | unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages; | |
6515 | int i, nid = zone_to_nid(zone), zone_id = zone_idx(zone); | |
0740a50b | 6516 | static unsigned long hole_pfn; |
73a6e474 | 6517 | unsigned long start_pfn, end_pfn; |
0740a50b | 6518 | u64 pgcnt = 0; |
73a6e474 BH |
6519 | |
6520 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { | |
3256ff83 BH |
6521 | start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn); |
6522 | end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn); | |
73a6e474 | 6523 | |
3256ff83 BH |
6524 | if (end_pfn > start_pfn) |
6525 | memmap_init_range(end_pfn - start_pfn, nid, | |
6526 | zone_id, start_pfn, zone_end_pfn, | |
6527 | MEMINIT_EARLY, NULL, MIGRATE_MOVABLE); | |
0740a50b MR |
6528 | |
6529 | if (hole_pfn < start_pfn) | |
6530 | pgcnt += init_unavailable_range(hole_pfn, start_pfn, | |
6531 | zone_id, nid); | |
6532 | hole_pfn = end_pfn; | |
73a6e474 | 6533 | } |
0740a50b MR |
6534 | |
6535 | #ifdef CONFIG_SPARSEMEM | |
6536 | /* | |
6537 | * Initialize the hole in the range [zone_end_pfn, section_end]. | |
6538 | * If zone boundary falls in the middle of a section, this hole | |
6539 | * will be re-initialized during the call to this function for the | |
6540 | * higher zone. | |
6541 | */ | |
6542 | end_pfn = round_up(zone_end_pfn, PAGES_PER_SECTION); | |
6543 | if (hole_pfn < end_pfn) | |
6544 | pgcnt += init_unavailable_range(hole_pfn, end_pfn, | |
6545 | zone_id, nid); | |
6546 | #endif | |
6547 | ||
6548 | if (pgcnt) | |
6549 | pr_info(" %s zone: %llu pages in unavailable ranges\n", | |
6550 | zone->name, pgcnt); | |
dfb3ccd0 | 6551 | } |
1da177e4 | 6552 | |
7cd2b0a3 | 6553 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 6554 | { |
3a6be87f | 6555 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
6556 | int batch; |
6557 | ||
6558 | /* | |
6559 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 6560 | * size of the zone. |
e7c8d5c9 | 6561 | */ |
9705bea5 | 6562 | batch = zone_managed_pages(zone) / 1024; |
d8a759b5 AL |
6563 | /* But no more than a meg. */ |
6564 | if (batch * PAGE_SIZE > 1024 * 1024) | |
6565 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
6566 | batch /= 4; /* We effectively *= 4 below */ |
6567 | if (batch < 1) | |
6568 | batch = 1; | |
6569 | ||
6570 | /* | |
0ceaacc9 NP |
6571 | * Clamp the batch to a 2^n - 1 value. Having a power |
6572 | * of 2 value was found to be more likely to have | |
6573 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 6574 | * |
0ceaacc9 NP |
6575 | * For example if 2 tasks are alternately allocating |
6576 | * batches of pages, one task can end up with a lot | |
6577 | * of pages of one half of the possible page colors | |
6578 | * and the other with pages of the other colors. | |
e7c8d5c9 | 6579 | */ |
9155203a | 6580 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 6581 | |
e7c8d5c9 | 6582 | return batch; |
3a6be87f DH |
6583 | |
6584 | #else | |
6585 | /* The deferral and batching of frees should be suppressed under NOMMU | |
6586 | * conditions. | |
6587 | * | |
6588 | * The problem is that NOMMU needs to be able to allocate large chunks | |
6589 | * of contiguous memory as there's no hardware page translation to | |
6590 | * assemble apparent contiguous memory from discontiguous pages. | |
6591 | * | |
6592 | * Queueing large contiguous runs of pages for batching, however, | |
6593 | * causes the pages to actually be freed in smaller chunks. As there | |
6594 | * can be a significant delay between the individual batches being | |
6595 | * recycled, this leads to the once large chunks of space being | |
6596 | * fragmented and becoming unavailable for high-order allocations. | |
6597 | */ | |
6598 | return 0; | |
6599 | #endif | |
e7c8d5c9 CL |
6600 | } |
6601 | ||
8d7a8fa9 | 6602 | /* |
5c3ad2eb VB |
6603 | * pcp->high and pcp->batch values are related and generally batch is lower |
6604 | * than high. They are also related to pcp->count such that count is lower | |
6605 | * than high, and as soon as it reaches high, the pcplist is flushed. | |
8d7a8fa9 | 6606 | * |
5c3ad2eb VB |
6607 | * However, guaranteeing these relations at all times would require e.g. write |
6608 | * barriers here but also careful usage of read barriers at the read side, and | |
6609 | * thus be prone to error and bad for performance. Thus the update only prevents | |
6610 | * store tearing. Any new users of pcp->batch and pcp->high should ensure they | |
6611 | * can cope with those fields changing asynchronously, and fully trust only the | |
6612 | * pcp->count field on the local CPU with interrupts disabled. | |
8d7a8fa9 CS |
6613 | * |
6614 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
6615 | * outside of boot time (or some other assurance that no concurrent updaters | |
6616 | * exist). | |
6617 | */ | |
6618 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
6619 | unsigned long batch) | |
6620 | { | |
5c3ad2eb VB |
6621 | WRITE_ONCE(pcp->batch, batch); |
6622 | WRITE_ONCE(pcp->high, high); | |
8d7a8fa9 CS |
6623 | } |
6624 | ||
88c90dbc | 6625 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
6626 | { |
6627 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 6628 | int migratetype; |
2caaad41 | 6629 | |
1c6fe946 MD |
6630 | memset(p, 0, sizeof(*p)); |
6631 | ||
3dfa5721 | 6632 | pcp = &p->pcp; |
5f8dcc21 MG |
6633 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
6634 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 | 6635 | |
69a8396a VB |
6636 | /* |
6637 | * Set batch and high values safe for a boot pageset. A true percpu | |
6638 | * pageset's initialization will update them subsequently. Here we don't | |
6639 | * need to be as careful as pageset_update() as nobody can access the | |
6640 | * pageset yet. | |
6641 | */ | |
952eaf81 VB |
6642 | pcp->high = BOOT_PAGESET_HIGH; |
6643 | pcp->batch = BOOT_PAGESET_BATCH; | |
88c90dbc CS |
6644 | } |
6645 | ||
3b1f3658 | 6646 | static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high, |
ec6e8c7e VB |
6647 | unsigned long batch) |
6648 | { | |
6649 | struct per_cpu_pageset *p; | |
6650 | int cpu; | |
6651 | ||
6652 | for_each_possible_cpu(cpu) { | |
6653 | p = per_cpu_ptr(zone->pageset, cpu); | |
6654 | pageset_update(&p->pcp, high, batch); | |
6655 | } | |
6656 | } | |
6657 | ||
8ad4b1fb | 6658 | /* |
0a8b4f1d | 6659 | * Calculate and set new high and batch values for all per-cpu pagesets of a |
7115ac6e | 6660 | * zone, based on the zone's size and the percpu_pagelist_fraction sysctl. |
8ad4b1fb | 6661 | */ |
0a8b4f1d | 6662 | static void zone_set_pageset_high_and_batch(struct zone *zone) |
56cef2b8 | 6663 | { |
7115ac6e VB |
6664 | unsigned long new_high, new_batch; |
6665 | ||
6666 | if (percpu_pagelist_fraction) { | |
6667 | new_high = zone_managed_pages(zone) / percpu_pagelist_fraction; | |
6668 | new_batch = max(1UL, new_high / 4); | |
6669 | if ((new_high / 4) > (PAGE_SHIFT * 8)) | |
6670 | new_batch = PAGE_SHIFT * 8; | |
6671 | } else { | |
6672 | new_batch = zone_batchsize(zone); | |
6673 | new_high = 6 * new_batch; | |
6674 | new_batch = max(1UL, 1 * new_batch); | |
6675 | } | |
169f6c19 | 6676 | |
952eaf81 VB |
6677 | if (zone->pageset_high == new_high && |
6678 | zone->pageset_batch == new_batch) | |
6679 | return; | |
6680 | ||
6681 | zone->pageset_high = new_high; | |
6682 | zone->pageset_batch = new_batch; | |
6683 | ||
ec6e8c7e | 6684 | __zone_set_pageset_high_and_batch(zone, new_high, new_batch); |
169f6c19 CS |
6685 | } |
6686 | ||
72675e13 | 6687 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 | 6688 | { |
0a8b4f1d | 6689 | struct per_cpu_pageset *p; |
319774e2 | 6690 | int cpu; |
0a8b4f1d | 6691 | |
319774e2 | 6692 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
0a8b4f1d VB |
6693 | for_each_possible_cpu(cpu) { |
6694 | p = per_cpu_ptr(zone->pageset, cpu); | |
6695 | pageset_init(p); | |
6696 | } | |
6697 | ||
6698 | zone_set_pageset_high_and_batch(zone); | |
319774e2 WF |
6699 | } |
6700 | ||
2caaad41 | 6701 | /* |
99dcc3e5 CL |
6702 | * Allocate per cpu pagesets and initialize them. |
6703 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 6704 | */ |
99dcc3e5 | 6705 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 6706 | { |
b4911ea2 | 6707 | struct pglist_data *pgdat; |
99dcc3e5 | 6708 | struct zone *zone; |
b418a0f9 | 6709 | int __maybe_unused cpu; |
e7c8d5c9 | 6710 | |
319774e2 WF |
6711 | for_each_populated_zone(zone) |
6712 | setup_zone_pageset(zone); | |
b4911ea2 | 6713 | |
b418a0f9 SD |
6714 | #ifdef CONFIG_NUMA |
6715 | /* | |
6716 | * Unpopulated zones continue using the boot pagesets. | |
6717 | * The numa stats for these pagesets need to be reset. | |
6718 | * Otherwise, they will end up skewing the stats of | |
6719 | * the nodes these zones are associated with. | |
6720 | */ | |
6721 | for_each_possible_cpu(cpu) { | |
6722 | struct per_cpu_pageset *pcp = &per_cpu(boot_pageset, cpu); | |
6723 | memset(pcp->vm_numa_stat_diff, 0, | |
6724 | sizeof(pcp->vm_numa_stat_diff)); | |
6725 | } | |
6726 | #endif | |
6727 | ||
b4911ea2 MG |
6728 | for_each_online_pgdat(pgdat) |
6729 | pgdat->per_cpu_nodestats = | |
6730 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
6731 | } |
6732 | ||
c09b4240 | 6733 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 6734 | { |
99dcc3e5 CL |
6735 | /* |
6736 | * per cpu subsystem is not up at this point. The following code | |
6737 | * relies on the ability of the linker to provide the | |
6738 | * offset of a (static) per cpu variable into the per cpu area. | |
6739 | */ | |
6740 | zone->pageset = &boot_pageset; | |
952eaf81 VB |
6741 | zone->pageset_high = BOOT_PAGESET_HIGH; |
6742 | zone->pageset_batch = BOOT_PAGESET_BATCH; | |
ed8ece2e | 6743 | |
b38a8725 | 6744 | if (populated_zone(zone)) |
99dcc3e5 CL |
6745 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
6746 | zone->name, zone->present_pages, | |
6747 | zone_batchsize(zone)); | |
ed8ece2e DH |
6748 | } |
6749 | ||
dc0bbf3b | 6750 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 6751 | unsigned long zone_start_pfn, |
b171e409 | 6752 | unsigned long size) |
ed8ece2e DH |
6753 | { |
6754 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 6755 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 6756 | |
8f416836 WY |
6757 | if (zone_idx > pgdat->nr_zones) |
6758 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 6759 | |
ed8ece2e DH |
6760 | zone->zone_start_pfn = zone_start_pfn; |
6761 | ||
708614e6 MG |
6762 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
6763 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
6764 | pgdat->node_id, | |
6765 | (unsigned long)zone_idx(zone), | |
6766 | zone_start_pfn, (zone_start_pfn + size)); | |
6767 | ||
1e548deb | 6768 | zone_init_free_lists(zone); |
9dcb8b68 | 6769 | zone->initialized = 1; |
ed8ece2e DH |
6770 | } |
6771 | ||
c713216d MG |
6772 | /** |
6773 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
6774 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
6775 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
6776 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
6777 | * |
6778 | * It returns the start and end page frame of a node based on information | |
7d018176 | 6779 | * provided by memblock_set_node(). If called for a node |
c713216d | 6780 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 6781 | * PFNs will be 0. |
c713216d | 6782 | */ |
bbe5d993 | 6783 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
6784 | unsigned long *start_pfn, unsigned long *end_pfn) |
6785 | { | |
c13291a5 | 6786 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 6787 | int i; |
c13291a5 | 6788 | |
c713216d MG |
6789 | *start_pfn = -1UL; |
6790 | *end_pfn = 0; | |
6791 | ||
c13291a5 TH |
6792 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
6793 | *start_pfn = min(*start_pfn, this_start_pfn); | |
6794 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
6795 | } |
6796 | ||
633c0666 | 6797 | if (*start_pfn == -1UL) |
c713216d | 6798 | *start_pfn = 0; |
c713216d MG |
6799 | } |
6800 | ||
2a1e274a MG |
6801 | /* |
6802 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
6803 | * assumption is made that zones within a node are ordered in monotonic | |
6804 | * increasing memory addresses so that the "highest" populated zone is used | |
6805 | */ | |
b69a7288 | 6806 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
6807 | { |
6808 | int zone_index; | |
6809 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
6810 | if (zone_index == ZONE_MOVABLE) | |
6811 | continue; | |
6812 | ||
6813 | if (arch_zone_highest_possible_pfn[zone_index] > | |
6814 | arch_zone_lowest_possible_pfn[zone_index]) | |
6815 | break; | |
6816 | } | |
6817 | ||
6818 | VM_BUG_ON(zone_index == -1); | |
6819 | movable_zone = zone_index; | |
6820 | } | |
6821 | ||
6822 | /* | |
6823 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 6824 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
6825 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
6826 | * in each node depending on the size of each node and how evenly kernelcore | |
6827 | * is distributed. This helper function adjusts the zone ranges | |
6828 | * provided by the architecture for a given node by using the end of the | |
6829 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
6830 | * zones within a node are in order of monotonic increases memory addresses | |
6831 | */ | |
bbe5d993 | 6832 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
6833 | unsigned long zone_type, |
6834 | unsigned long node_start_pfn, | |
6835 | unsigned long node_end_pfn, | |
6836 | unsigned long *zone_start_pfn, | |
6837 | unsigned long *zone_end_pfn) | |
6838 | { | |
6839 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
6840 | if (zone_movable_pfn[nid]) { | |
6841 | /* Size ZONE_MOVABLE */ | |
6842 | if (zone_type == ZONE_MOVABLE) { | |
6843 | *zone_start_pfn = zone_movable_pfn[nid]; | |
6844 | *zone_end_pfn = min(node_end_pfn, | |
6845 | arch_zone_highest_possible_pfn[movable_zone]); | |
6846 | ||
e506b996 XQ |
6847 | /* Adjust for ZONE_MOVABLE starting within this range */ |
6848 | } else if (!mirrored_kernelcore && | |
6849 | *zone_start_pfn < zone_movable_pfn[nid] && | |
6850 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
6851 | *zone_end_pfn = zone_movable_pfn[nid]; | |
6852 | ||
2a1e274a MG |
6853 | /* Check if this whole range is within ZONE_MOVABLE */ |
6854 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
6855 | *zone_start_pfn = *zone_end_pfn; | |
6856 | } | |
6857 | } | |
6858 | ||
c713216d MG |
6859 | /* |
6860 | * Return the number of pages a zone spans in a node, including holes | |
6861 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
6862 | */ | |
bbe5d993 | 6863 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 6864 | unsigned long zone_type, |
7960aedd ZY |
6865 | unsigned long node_start_pfn, |
6866 | unsigned long node_end_pfn, | |
d91749c1 | 6867 | unsigned long *zone_start_pfn, |
854e8848 | 6868 | unsigned long *zone_end_pfn) |
c713216d | 6869 | { |
299c83dc LF |
6870 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6871 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 6872 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6873 | if (!node_start_pfn && !node_end_pfn) |
6874 | return 0; | |
6875 | ||
7960aedd | 6876 | /* Get the start and end of the zone */ |
299c83dc LF |
6877 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6878 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
6879 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6880 | node_start_pfn, node_end_pfn, | |
d91749c1 | 6881 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
6882 | |
6883 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 6884 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
6885 | return 0; |
6886 | ||
6887 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
6888 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
6889 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
6890 | |
6891 | /* Return the spanned pages */ | |
d91749c1 | 6892 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
6893 | } |
6894 | ||
6895 | /* | |
6896 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 6897 | * then all holes in the requested range will be accounted for. |
c713216d | 6898 | */ |
bbe5d993 | 6899 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
6900 | unsigned long range_start_pfn, |
6901 | unsigned long range_end_pfn) | |
6902 | { | |
96e907d1 TH |
6903 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
6904 | unsigned long start_pfn, end_pfn; | |
6905 | int i; | |
c713216d | 6906 | |
96e907d1 TH |
6907 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6908 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6909 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6910 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 6911 | } |
96e907d1 | 6912 | return nr_absent; |
c713216d MG |
6913 | } |
6914 | ||
6915 | /** | |
6916 | * absent_pages_in_range - Return number of page frames in holes within a range | |
6917 | * @start_pfn: The start PFN to start searching for holes | |
6918 | * @end_pfn: The end PFN to stop searching for holes | |
6919 | * | |
a862f68a | 6920 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
6921 | */ |
6922 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
6923 | unsigned long end_pfn) | |
6924 | { | |
6925 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
6926 | } | |
6927 | ||
6928 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 6929 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 6930 | unsigned long zone_type, |
7960aedd | 6931 | unsigned long node_start_pfn, |
854e8848 | 6932 | unsigned long node_end_pfn) |
c713216d | 6933 | { |
96e907d1 TH |
6934 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6935 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 6936 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 6937 | unsigned long nr_absent; |
9c7cd687 | 6938 | |
b5685e92 | 6939 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6940 | if (!node_start_pfn && !node_end_pfn) |
6941 | return 0; | |
6942 | ||
96e907d1 TH |
6943 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6944 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 6945 | |
2a1e274a MG |
6946 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6947 | node_start_pfn, node_end_pfn, | |
6948 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
6949 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
6950 | ||
6951 | /* | |
6952 | * ZONE_MOVABLE handling. | |
6953 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6954 | * and vice versa. | |
6955 | */ | |
e506b996 XQ |
6956 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6957 | unsigned long start_pfn, end_pfn; | |
6958 | struct memblock_region *r; | |
6959 | ||
cc6de168 | 6960 | for_each_mem_region(r) { |
e506b996 XQ |
6961 | start_pfn = clamp(memblock_region_memory_base_pfn(r), |
6962 | zone_start_pfn, zone_end_pfn); | |
6963 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6964 | zone_start_pfn, zone_end_pfn); | |
6965 | ||
6966 | if (zone_type == ZONE_MOVABLE && | |
6967 | memblock_is_mirror(r)) | |
6968 | nr_absent += end_pfn - start_pfn; | |
6969 | ||
6970 | if (zone_type == ZONE_NORMAL && | |
6971 | !memblock_is_mirror(r)) | |
6972 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6973 | } |
6974 | } | |
6975 | ||
6976 | return nr_absent; | |
c713216d | 6977 | } |
0e0b864e | 6978 | |
bbe5d993 | 6979 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd | 6980 | unsigned long node_start_pfn, |
854e8848 | 6981 | unsigned long node_end_pfn) |
c713216d | 6982 | { |
febd5949 | 6983 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6984 | enum zone_type i; |
6985 | ||
febd5949 GZ |
6986 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6987 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6988 | unsigned long zone_start_pfn, zone_end_pfn; |
3f08a302 | 6989 | unsigned long spanned, absent; |
febd5949 | 6990 | unsigned long size, real_size; |
c713216d | 6991 | |
854e8848 MR |
6992 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i, |
6993 | node_start_pfn, | |
6994 | node_end_pfn, | |
6995 | &zone_start_pfn, | |
6996 | &zone_end_pfn); | |
6997 | absent = zone_absent_pages_in_node(pgdat->node_id, i, | |
6998 | node_start_pfn, | |
6999 | node_end_pfn); | |
3f08a302 MR |
7000 | |
7001 | size = spanned; | |
7002 | real_size = size - absent; | |
7003 | ||
d91749c1 TI |
7004 | if (size) |
7005 | zone->zone_start_pfn = zone_start_pfn; | |
7006 | else | |
7007 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
7008 | zone->spanned_pages = size; |
7009 | zone->present_pages = real_size; | |
7010 | ||
7011 | totalpages += size; | |
7012 | realtotalpages += real_size; | |
7013 | } | |
7014 | ||
7015 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
7016 | pgdat->node_present_pages = realtotalpages; |
7017 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
7018 | realtotalpages); | |
7019 | } | |
7020 | ||
835c134e MG |
7021 | #ifndef CONFIG_SPARSEMEM |
7022 | /* | |
7023 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
7024 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
7025 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
7026 | * round what is now in bits to nearest long in bits, then return it in |
7027 | * bytes. | |
7028 | */ | |
7c45512d | 7029 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
7030 | { |
7031 | unsigned long usemapsize; | |
7032 | ||
7c45512d | 7033 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
7034 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
7035 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
7036 | usemapsize *= NR_PAGEBLOCK_BITS; |
7037 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
7038 | ||
7039 | return usemapsize / 8; | |
7040 | } | |
7041 | ||
7010a6ec | 7042 | static void __ref setup_usemap(struct zone *zone) |
835c134e | 7043 | { |
7010a6ec BH |
7044 | unsigned long usemapsize = usemap_size(zone->zone_start_pfn, |
7045 | zone->spanned_pages); | |
835c134e | 7046 | zone->pageblock_flags = NULL; |
23a7052a | 7047 | if (usemapsize) { |
6782832e | 7048 | zone->pageblock_flags = |
26fb3dae | 7049 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
7010a6ec | 7050 | zone_to_nid(zone)); |
23a7052a MR |
7051 | if (!zone->pageblock_flags) |
7052 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
7010a6ec | 7053 | usemapsize, zone->name, zone_to_nid(zone)); |
23a7052a | 7054 | } |
835c134e MG |
7055 | } |
7056 | #else | |
7010a6ec | 7057 | static inline void setup_usemap(struct zone *zone) {} |
835c134e MG |
7058 | #endif /* CONFIG_SPARSEMEM */ |
7059 | ||
d9c23400 | 7060 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 7061 | |
d9c23400 | 7062 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 7063 | void __init set_pageblock_order(void) |
d9c23400 | 7064 | { |
955c1cd7 AM |
7065 | unsigned int order; |
7066 | ||
d9c23400 MG |
7067 | /* Check that pageblock_nr_pages has not already been setup */ |
7068 | if (pageblock_order) | |
7069 | return; | |
7070 | ||
955c1cd7 AM |
7071 | if (HPAGE_SHIFT > PAGE_SHIFT) |
7072 | order = HUGETLB_PAGE_ORDER; | |
7073 | else | |
7074 | order = MAX_ORDER - 1; | |
7075 | ||
d9c23400 MG |
7076 | /* |
7077 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
7078 | * This value may be variable depending on boot parameters on IA64 and |
7079 | * powerpc. | |
d9c23400 MG |
7080 | */ |
7081 | pageblock_order = order; | |
7082 | } | |
7083 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7084 | ||
ba72cb8c MG |
7085 | /* |
7086 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
7087 | * is unused as pageblock_order is set at compile-time. See |
7088 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
7089 | * the kernel config | |
ba72cb8c | 7090 | */ |
03e85f9d | 7091 | void __init set_pageblock_order(void) |
ba72cb8c | 7092 | { |
ba72cb8c | 7093 | } |
d9c23400 MG |
7094 | |
7095 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7096 | ||
03e85f9d | 7097 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 7098 | unsigned long present_pages) |
01cefaef JL |
7099 | { |
7100 | unsigned long pages = spanned_pages; | |
7101 | ||
7102 | /* | |
7103 | * Provide a more accurate estimation if there are holes within | |
7104 | * the zone and SPARSEMEM is in use. If there are holes within the | |
7105 | * zone, each populated memory region may cost us one or two extra | |
7106 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 7107 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
7108 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
7109 | */ | |
7110 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
7111 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
7112 | pages = present_pages; | |
7113 | ||
7114 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
7115 | } | |
7116 | ||
ace1db39 OS |
7117 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
7118 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
7119 | { | |
364c1eeb YS |
7120 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
7121 | ||
7122 | spin_lock_init(&ds_queue->split_queue_lock); | |
7123 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
7124 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
7125 | } |
7126 | #else | |
7127 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
7128 | #endif | |
7129 | ||
7130 | #ifdef CONFIG_COMPACTION | |
7131 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
7132 | { | |
7133 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
7134 | } | |
7135 | #else | |
7136 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
7137 | #endif | |
7138 | ||
03e85f9d | 7139 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 7140 | { |
208d54e5 | 7141 | pgdat_resize_init(pgdat); |
ace1db39 | 7142 | |
ace1db39 OS |
7143 | pgdat_init_split_queue(pgdat); |
7144 | pgdat_init_kcompactd(pgdat); | |
7145 | ||
1da177e4 | 7146 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 7147 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 7148 | |
eefa864b | 7149 | pgdat_page_ext_init(pgdat); |
867e5e1d | 7150 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
7151 | } |
7152 | ||
7153 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
7154 | unsigned long remaining_pages) | |
7155 | { | |
9705bea5 | 7156 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
7157 | zone_set_nid(zone, nid); |
7158 | zone->name = zone_names[idx]; | |
7159 | zone->zone_pgdat = NODE_DATA(nid); | |
7160 | spin_lock_init(&zone->lock); | |
7161 | zone_seqlock_init(zone); | |
7162 | zone_pcp_init(zone); | |
7163 | } | |
7164 | ||
7165 | /* | |
7166 | * Set up the zone data structures | |
7167 | * - init pgdat internals | |
7168 | * - init all zones belonging to this node | |
7169 | * | |
7170 | * NOTE: this function is only called during memory hotplug | |
7171 | */ | |
7172 | #ifdef CONFIG_MEMORY_HOTPLUG | |
7173 | void __ref free_area_init_core_hotplug(int nid) | |
7174 | { | |
7175 | enum zone_type z; | |
7176 | pg_data_t *pgdat = NODE_DATA(nid); | |
7177 | ||
7178 | pgdat_init_internals(pgdat); | |
7179 | for (z = 0; z < MAX_NR_ZONES; z++) | |
7180 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
7181 | } | |
7182 | #endif | |
7183 | ||
7184 | /* | |
7185 | * Set up the zone data structures: | |
7186 | * - mark all pages reserved | |
7187 | * - mark all memory queues empty | |
7188 | * - clear the memory bitmaps | |
7189 | * | |
7190 | * NOTE: pgdat should get zeroed by caller. | |
7191 | * NOTE: this function is only called during early init. | |
7192 | */ | |
7193 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
7194 | { | |
7195 | enum zone_type j; | |
7196 | int nid = pgdat->node_id; | |
5f63b720 | 7197 | |
03e85f9d | 7198 | pgdat_init_internals(pgdat); |
385386cf JW |
7199 | pgdat->per_cpu_nodestats = &boot_nodestats; |
7200 | ||
1da177e4 LT |
7201 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7202 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 7203 | unsigned long size, freesize, memmap_pages; |
1da177e4 | 7204 | |
febd5949 | 7205 | size = zone->spanned_pages; |
e6943859 | 7206 | freesize = zone->present_pages; |
1da177e4 | 7207 | |
0e0b864e | 7208 | /* |
9feedc9d | 7209 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
7210 | * is used by this zone for memmap. This affects the watermark |
7211 | * and per-cpu initialisations | |
7212 | */ | |
e6943859 | 7213 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
7214 | if (!is_highmem_idx(j)) { |
7215 | if (freesize >= memmap_pages) { | |
7216 | freesize -= memmap_pages; | |
7217 | if (memmap_pages) | |
7218 | printk(KERN_DEBUG | |
7219 | " %s zone: %lu pages used for memmap\n", | |
7220 | zone_names[j], memmap_pages); | |
7221 | } else | |
1170532b | 7222 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
7223 | zone_names[j], memmap_pages, freesize); |
7224 | } | |
0e0b864e | 7225 | |
6267276f | 7226 | /* Account for reserved pages */ |
9feedc9d JL |
7227 | if (j == 0 && freesize > dma_reserve) { |
7228 | freesize -= dma_reserve; | |
d903ef9f | 7229 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 7230 | zone_names[0], dma_reserve); |
0e0b864e MG |
7231 | } |
7232 | ||
98d2b0eb | 7233 | if (!is_highmem_idx(j)) |
9feedc9d | 7234 | nr_kernel_pages += freesize; |
01cefaef JL |
7235 | /* Charge for highmem memmap if there are enough kernel pages */ |
7236 | else if (nr_kernel_pages > memmap_pages * 2) | |
7237 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 7238 | nr_all_pages += freesize; |
1da177e4 | 7239 | |
9feedc9d JL |
7240 | /* |
7241 | * Set an approximate value for lowmem here, it will be adjusted | |
7242 | * when the bootmem allocator frees pages into the buddy system. | |
7243 | * And all highmem pages will be managed by the buddy system. | |
7244 | */ | |
03e85f9d | 7245 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 7246 | |
d883c6cf | 7247 | if (!size) |
1da177e4 LT |
7248 | continue; |
7249 | ||
955c1cd7 | 7250 | set_pageblock_order(); |
7010a6ec | 7251 | setup_usemap(zone); |
9699ee7b | 7252 | init_currently_empty_zone(zone, zone->zone_start_pfn, size); |
3256ff83 | 7253 | memmap_init_zone(zone); |
1da177e4 LT |
7254 | } |
7255 | } | |
7256 | ||
0cd842f9 | 7257 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 7258 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 7259 | { |
b0aeba74 | 7260 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
7261 | unsigned long __maybe_unused offset = 0; |
7262 | ||
1da177e4 LT |
7263 | /* Skip empty nodes */ |
7264 | if (!pgdat->node_spanned_pages) | |
7265 | return; | |
7266 | ||
b0aeba74 TL |
7267 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
7268 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
7269 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
7270 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 7271 | unsigned long size, end; |
d41dee36 AW |
7272 | struct page *map; |
7273 | ||
e984bb43 BP |
7274 | /* |
7275 | * The zone's endpoints aren't required to be MAX_ORDER | |
7276 | * aligned but the node_mem_map endpoints must be in order | |
7277 | * for the buddy allocator to function correctly. | |
7278 | */ | |
108bcc96 | 7279 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
7280 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
7281 | size = (end - start) * sizeof(struct page); | |
26fb3dae MR |
7282 | map = memblock_alloc_node(size, SMP_CACHE_BYTES, |
7283 | pgdat->node_id); | |
23a7052a MR |
7284 | if (!map) |
7285 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
7286 | size, pgdat->node_id); | |
a1c34a3b | 7287 | pgdat->node_mem_map = map + offset; |
1da177e4 | 7288 | } |
0cd842f9 OS |
7289 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
7290 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
7291 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 7292 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
7293 | /* |
7294 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
7295 | */ | |
c713216d | 7296 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 7297 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d | 7298 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 7299 | mem_map -= offset; |
c713216d | 7300 | } |
1da177e4 LT |
7301 | #endif |
7302 | } | |
0cd842f9 OS |
7303 | #else |
7304 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
7305 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 7306 | |
0188dc98 OS |
7307 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
7308 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
7309 | { | |
0188dc98 OS |
7310 | pgdat->first_deferred_pfn = ULONG_MAX; |
7311 | } | |
7312 | #else | |
7313 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
7314 | #endif | |
7315 | ||
854e8848 | 7316 | static void __init free_area_init_node(int nid) |
1da177e4 | 7317 | { |
9109fb7b | 7318 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
7319 | unsigned long start_pfn = 0; |
7320 | unsigned long end_pfn = 0; | |
9109fb7b | 7321 | |
88fdf75d | 7322 | /* pg_data_t should be reset to zero when it's allocated */ |
97a225e6 | 7323 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx); |
88fdf75d | 7324 | |
854e8848 | 7325 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
88fdf75d | 7326 | |
1da177e4 | 7327 | pgdat->node_id = nid; |
854e8848 | 7328 | pgdat->node_start_pfn = start_pfn; |
75ef7184 | 7329 | pgdat->per_cpu_nodestats = NULL; |
854e8848 | 7330 | |
8d29e18a | 7331 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
7332 | (u64)start_pfn << PAGE_SHIFT, |
7333 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
854e8848 | 7334 | calculate_node_totalpages(pgdat, start_pfn, end_pfn); |
1da177e4 LT |
7335 | |
7336 | alloc_node_mem_map(pgdat); | |
0188dc98 | 7337 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 7338 | |
7f3eb55b | 7339 | free_area_init_core(pgdat); |
1da177e4 LT |
7340 | } |
7341 | ||
bc9331a1 | 7342 | void __init free_area_init_memoryless_node(int nid) |
3f08a302 | 7343 | { |
854e8848 | 7344 | free_area_init_node(nid); |
3f08a302 MR |
7345 | } |
7346 | ||
418508c1 MS |
7347 | #if MAX_NUMNODES > 1 |
7348 | /* | |
7349 | * Figure out the number of possible node ids. | |
7350 | */ | |
f9872caf | 7351 | void __init setup_nr_node_ids(void) |
418508c1 | 7352 | { |
904a9553 | 7353 | unsigned int highest; |
418508c1 | 7354 | |
904a9553 | 7355 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7356 | nr_node_ids = highest + 1; |
7357 | } | |
418508c1 MS |
7358 | #endif |
7359 | ||
1e01979c TH |
7360 | /** |
7361 | * node_map_pfn_alignment - determine the maximum internode alignment | |
7362 | * | |
7363 | * This function should be called after node map is populated and sorted. | |
7364 | * It calculates the maximum power of two alignment which can distinguish | |
7365 | * all the nodes. | |
7366 | * | |
7367 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
7368 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
7369 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
7370 | * shifted, 1GiB is enough and this function will indicate so. | |
7371 | * | |
7372 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
7373 | * model has fine enough granularity to avoid incorrect mapping for the | |
7374 | * populated node map. | |
7375 | * | |
a862f68a | 7376 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
7377 | * requirement (single node). |
7378 | */ | |
7379 | unsigned long __init node_map_pfn_alignment(void) | |
7380 | { | |
7381 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 7382 | unsigned long start, end, mask; |
98fa15f3 | 7383 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 7384 | int i, nid; |
1e01979c | 7385 | |
c13291a5 | 7386 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
7387 | if (!start || last_nid < 0 || last_nid == nid) { |
7388 | last_nid = nid; | |
7389 | last_end = end; | |
7390 | continue; | |
7391 | } | |
7392 | ||
7393 | /* | |
7394 | * Start with a mask granular enough to pin-point to the | |
7395 | * start pfn and tick off bits one-by-one until it becomes | |
7396 | * too coarse to separate the current node from the last. | |
7397 | */ | |
7398 | mask = ~((1 << __ffs(start)) - 1); | |
7399 | while (mask && last_end <= (start & (mask << 1))) | |
7400 | mask <<= 1; | |
7401 | ||
7402 | /* accumulate all internode masks */ | |
7403 | accl_mask |= mask; | |
7404 | } | |
7405 | ||
7406 | /* convert mask to number of pages */ | |
7407 | return ~accl_mask + 1; | |
7408 | } | |
7409 | ||
c713216d MG |
7410 | /** |
7411 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
7412 | * | |
a862f68a | 7413 | * Return: the minimum PFN based on information provided via |
7d018176 | 7414 | * memblock_set_node(). |
c713216d MG |
7415 | */ |
7416 | unsigned long __init find_min_pfn_with_active_regions(void) | |
7417 | { | |
8a1b25fe | 7418 | return PHYS_PFN(memblock_start_of_DRAM()); |
c713216d MG |
7419 | } |
7420 | ||
37b07e41 LS |
7421 | /* |
7422 | * early_calculate_totalpages() | |
7423 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 7424 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 7425 | */ |
484f51f8 | 7426 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 7427 | { |
7e63efef | 7428 | unsigned long totalpages = 0; |
c13291a5 TH |
7429 | unsigned long start_pfn, end_pfn; |
7430 | int i, nid; | |
7431 | ||
7432 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
7433 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 7434 | |
37b07e41 LS |
7435 | totalpages += pages; |
7436 | if (pages) | |
4b0ef1fe | 7437 | node_set_state(nid, N_MEMORY); |
37b07e41 | 7438 | } |
b8af2941 | 7439 | return totalpages; |
7e63efef MG |
7440 | } |
7441 | ||
2a1e274a MG |
7442 | /* |
7443 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
7444 | * is spread evenly between nodes as long as the nodes have enough | |
7445 | * memory. When they don't, some nodes will have more kernelcore than | |
7446 | * others | |
7447 | */ | |
b224ef85 | 7448 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
7449 | { |
7450 | int i, nid; | |
7451 | unsigned long usable_startpfn; | |
7452 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 7453 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 7454 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 7455 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 7456 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 7457 | struct memblock_region *r; |
b2f3eebe TC |
7458 | |
7459 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
7460 | find_usable_zone_for_movable(); | |
7461 | ||
7462 | /* | |
7463 | * If movable_node is specified, ignore kernelcore and movablecore | |
7464 | * options. | |
7465 | */ | |
7466 | if (movable_node_is_enabled()) { | |
cc6de168 | 7467 | for_each_mem_region(r) { |
136199f0 | 7468 | if (!memblock_is_hotpluggable(r)) |
b2f3eebe TC |
7469 | continue; |
7470 | ||
d622abf7 | 7471 | nid = memblock_get_region_node(r); |
b2f3eebe | 7472 | |
136199f0 | 7473 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
7474 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
7475 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7476 | usable_startpfn; | |
7477 | } | |
7478 | ||
7479 | goto out2; | |
7480 | } | |
2a1e274a | 7481 | |
342332e6 TI |
7482 | /* |
7483 | * If kernelcore=mirror is specified, ignore movablecore option | |
7484 | */ | |
7485 | if (mirrored_kernelcore) { | |
7486 | bool mem_below_4gb_not_mirrored = false; | |
7487 | ||
cc6de168 | 7488 | for_each_mem_region(r) { |
342332e6 TI |
7489 | if (memblock_is_mirror(r)) |
7490 | continue; | |
7491 | ||
d622abf7 | 7492 | nid = memblock_get_region_node(r); |
342332e6 TI |
7493 | |
7494 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
7495 | ||
7496 | if (usable_startpfn < 0x100000) { | |
7497 | mem_below_4gb_not_mirrored = true; | |
7498 | continue; | |
7499 | } | |
7500 | ||
7501 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
7502 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7503 | usable_startpfn; | |
7504 | } | |
7505 | ||
7506 | if (mem_below_4gb_not_mirrored) | |
633bf2fe | 7507 | pr_warn("This configuration results in unmirrored kernel memory.\n"); |
342332e6 TI |
7508 | |
7509 | goto out2; | |
7510 | } | |
7511 | ||
7e63efef | 7512 | /* |
a5c6d650 DR |
7513 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
7514 | * amount of necessary memory. | |
7515 | */ | |
7516 | if (required_kernelcore_percent) | |
7517 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
7518 | 10000UL; | |
7519 | if (required_movablecore_percent) | |
7520 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
7521 | 10000UL; | |
7522 | ||
7523 | /* | |
7524 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
7525 | * kernelcore that corresponds so that memory usable for |
7526 | * any allocation type is evenly spread. If both kernelcore | |
7527 | * and movablecore are specified, then the value of kernelcore | |
7528 | * will be used for required_kernelcore if it's greater than | |
7529 | * what movablecore would have allowed. | |
7530 | */ | |
7531 | if (required_movablecore) { | |
7e63efef MG |
7532 | unsigned long corepages; |
7533 | ||
7534 | /* | |
7535 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
7536 | * was requested by the user | |
7537 | */ | |
7538 | required_movablecore = | |
7539 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 7540 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
7541 | corepages = totalpages - required_movablecore; |
7542 | ||
7543 | required_kernelcore = max(required_kernelcore, corepages); | |
7544 | } | |
7545 | ||
bde304bd XQ |
7546 | /* |
7547 | * If kernelcore was not specified or kernelcore size is larger | |
7548 | * than totalpages, there is no ZONE_MOVABLE. | |
7549 | */ | |
7550 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 7551 | goto out; |
2a1e274a MG |
7552 | |
7553 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
7554 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
7555 | ||
7556 | restart: | |
7557 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
7558 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 7559 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
7560 | unsigned long start_pfn, end_pfn; |
7561 | ||
2a1e274a MG |
7562 | /* |
7563 | * Recalculate kernelcore_node if the division per node | |
7564 | * now exceeds what is necessary to satisfy the requested | |
7565 | * amount of memory for the kernel | |
7566 | */ | |
7567 | if (required_kernelcore < kernelcore_node) | |
7568 | kernelcore_node = required_kernelcore / usable_nodes; | |
7569 | ||
7570 | /* | |
7571 | * As the map is walked, we track how much memory is usable | |
7572 | * by the kernel using kernelcore_remaining. When it is | |
7573 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
7574 | */ | |
7575 | kernelcore_remaining = kernelcore_node; | |
7576 | ||
7577 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 7578 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
7579 | unsigned long size_pages; |
7580 | ||
c13291a5 | 7581 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
7582 | if (start_pfn >= end_pfn) |
7583 | continue; | |
7584 | ||
7585 | /* Account for what is only usable for kernelcore */ | |
7586 | if (start_pfn < usable_startpfn) { | |
7587 | unsigned long kernel_pages; | |
7588 | kernel_pages = min(end_pfn, usable_startpfn) | |
7589 | - start_pfn; | |
7590 | ||
7591 | kernelcore_remaining -= min(kernel_pages, | |
7592 | kernelcore_remaining); | |
7593 | required_kernelcore -= min(kernel_pages, | |
7594 | required_kernelcore); | |
7595 | ||
7596 | /* Continue if range is now fully accounted */ | |
7597 | if (end_pfn <= usable_startpfn) { | |
7598 | ||
7599 | /* | |
7600 | * Push zone_movable_pfn to the end so | |
7601 | * that if we have to rebalance | |
7602 | * kernelcore across nodes, we will | |
7603 | * not double account here | |
7604 | */ | |
7605 | zone_movable_pfn[nid] = end_pfn; | |
7606 | continue; | |
7607 | } | |
7608 | start_pfn = usable_startpfn; | |
7609 | } | |
7610 | ||
7611 | /* | |
7612 | * The usable PFN range for ZONE_MOVABLE is from | |
7613 | * start_pfn->end_pfn. Calculate size_pages as the | |
7614 | * number of pages used as kernelcore | |
7615 | */ | |
7616 | size_pages = end_pfn - start_pfn; | |
7617 | if (size_pages > kernelcore_remaining) | |
7618 | size_pages = kernelcore_remaining; | |
7619 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
7620 | ||
7621 | /* | |
7622 | * Some kernelcore has been met, update counts and | |
7623 | * break if the kernelcore for this node has been | |
b8af2941 | 7624 | * satisfied |
2a1e274a MG |
7625 | */ |
7626 | required_kernelcore -= min(required_kernelcore, | |
7627 | size_pages); | |
7628 | kernelcore_remaining -= size_pages; | |
7629 | if (!kernelcore_remaining) | |
7630 | break; | |
7631 | } | |
7632 | } | |
7633 | ||
7634 | /* | |
7635 | * If there is still required_kernelcore, we do another pass with one | |
7636 | * less node in the count. This will push zone_movable_pfn[nid] further | |
7637 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 7638 | * satisfied |
2a1e274a MG |
7639 | */ |
7640 | usable_nodes--; | |
7641 | if (usable_nodes && required_kernelcore > usable_nodes) | |
7642 | goto restart; | |
7643 | ||
b2f3eebe | 7644 | out2: |
2a1e274a MG |
7645 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
7646 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
7647 | zone_movable_pfn[nid] = | |
7648 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 7649 | |
20e6926d | 7650 | out: |
66918dcd | 7651 | /* restore the node_state */ |
4b0ef1fe | 7652 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
7653 | } |
7654 | ||
4b0ef1fe LJ |
7655 | /* Any regular or high memory on that node ? */ |
7656 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 7657 | { |
37b07e41 LS |
7658 | enum zone_type zone_type; |
7659 | ||
4b0ef1fe | 7660 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 7661 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 7662 | if (populated_zone(zone)) { |
7b0e0c0e OS |
7663 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
7664 | node_set_state(nid, N_HIGH_MEMORY); | |
7665 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 7666 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
7667 | break; |
7668 | } | |
37b07e41 | 7669 | } |
37b07e41 LS |
7670 | } |
7671 | ||
51930df5 | 7672 | /* |
f0953a1b | 7673 | * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For |
51930df5 MR |
7674 | * such cases we allow max_zone_pfn sorted in the descending order |
7675 | */ | |
7676 | bool __weak arch_has_descending_max_zone_pfns(void) | |
7677 | { | |
7678 | return false; | |
7679 | } | |
7680 | ||
c713216d | 7681 | /** |
9691a071 | 7682 | * free_area_init - Initialise all pg_data_t and zone data |
88ca3b94 | 7683 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
7684 | * |
7685 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 7686 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
7687 | * zone in each node and their holes is calculated. If the maximum PFN |
7688 | * between two adjacent zones match, it is assumed that the zone is empty. | |
7689 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
7690 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
7691 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
7692 | * at arch_max_dma_pfn. | |
7693 | */ | |
9691a071 | 7694 | void __init free_area_init(unsigned long *max_zone_pfn) |
c713216d | 7695 | { |
c13291a5 | 7696 | unsigned long start_pfn, end_pfn; |
51930df5 MR |
7697 | int i, nid, zone; |
7698 | bool descending; | |
a6af2bc3 | 7699 | |
c713216d MG |
7700 | /* Record where the zone boundaries are */ |
7701 | memset(arch_zone_lowest_possible_pfn, 0, | |
7702 | sizeof(arch_zone_lowest_possible_pfn)); | |
7703 | memset(arch_zone_highest_possible_pfn, 0, | |
7704 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
7705 | |
7706 | start_pfn = find_min_pfn_with_active_regions(); | |
51930df5 | 7707 | descending = arch_has_descending_max_zone_pfns(); |
90cae1fe OH |
7708 | |
7709 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
51930df5 MR |
7710 | if (descending) |
7711 | zone = MAX_NR_ZONES - i - 1; | |
7712 | else | |
7713 | zone = i; | |
7714 | ||
7715 | if (zone == ZONE_MOVABLE) | |
2a1e274a | 7716 | continue; |
90cae1fe | 7717 | |
51930df5 MR |
7718 | end_pfn = max(max_zone_pfn[zone], start_pfn); |
7719 | arch_zone_lowest_possible_pfn[zone] = start_pfn; | |
7720 | arch_zone_highest_possible_pfn[zone] = end_pfn; | |
90cae1fe OH |
7721 | |
7722 | start_pfn = end_pfn; | |
c713216d | 7723 | } |
2a1e274a MG |
7724 | |
7725 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
7726 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 7727 | find_zone_movable_pfns_for_nodes(); |
c713216d | 7728 | |
c713216d | 7729 | /* Print out the zone ranges */ |
f88dfff5 | 7730 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
7731 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7732 | if (i == ZONE_MOVABLE) | |
7733 | continue; | |
f88dfff5 | 7734 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
7735 | if (arch_zone_lowest_possible_pfn[i] == |
7736 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 7737 | pr_cont("empty\n"); |
72f0ba02 | 7738 | else |
8d29e18a JG |
7739 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
7740 | (u64)arch_zone_lowest_possible_pfn[i] | |
7741 | << PAGE_SHIFT, | |
7742 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 7743 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
7744 | } |
7745 | ||
7746 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 7747 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
7748 | for (i = 0; i < MAX_NUMNODES; i++) { |
7749 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
7750 | pr_info(" Node %d: %#018Lx\n", i, |
7751 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 7752 | } |
c713216d | 7753 | |
f46edbd1 DW |
7754 | /* |
7755 | * Print out the early node map, and initialize the | |
7756 | * subsection-map relative to active online memory ranges to | |
7757 | * enable future "sub-section" extensions of the memory map. | |
7758 | */ | |
f88dfff5 | 7759 | pr_info("Early memory node ranges\n"); |
f46edbd1 | 7760 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
8d29e18a JG |
7761 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
7762 | (u64)start_pfn << PAGE_SHIFT, | |
7763 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
f46edbd1 DW |
7764 | subsection_map_init(start_pfn, end_pfn - start_pfn); |
7765 | } | |
c713216d MG |
7766 | |
7767 | /* Initialise every node */ | |
708614e6 | 7768 | mminit_verify_pageflags_layout(); |
8ef82866 | 7769 | setup_nr_node_ids(); |
c713216d MG |
7770 | for_each_online_node(nid) { |
7771 | pg_data_t *pgdat = NODE_DATA(nid); | |
854e8848 | 7772 | free_area_init_node(nid); |
37b07e41 LS |
7773 | |
7774 | /* Any memory on that node */ | |
7775 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
7776 | node_set_state(nid, N_MEMORY); |
7777 | check_for_memory(pgdat, nid); | |
c713216d MG |
7778 | } |
7779 | } | |
2a1e274a | 7780 | |
a5c6d650 DR |
7781 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
7782 | unsigned long *percent) | |
2a1e274a MG |
7783 | { |
7784 | unsigned long long coremem; | |
a5c6d650 DR |
7785 | char *endptr; |
7786 | ||
2a1e274a MG |
7787 | if (!p) |
7788 | return -EINVAL; | |
7789 | ||
a5c6d650 DR |
7790 | /* Value may be a percentage of total memory, otherwise bytes */ |
7791 | coremem = simple_strtoull(p, &endptr, 0); | |
7792 | if (*endptr == '%') { | |
7793 | /* Paranoid check for percent values greater than 100 */ | |
7794 | WARN_ON(coremem > 100); | |
2a1e274a | 7795 | |
a5c6d650 DR |
7796 | *percent = coremem; |
7797 | } else { | |
7798 | coremem = memparse(p, &p); | |
7799 | /* Paranoid check that UL is enough for the coremem value */ | |
7800 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7801 | |
a5c6d650 DR |
7802 | *core = coremem >> PAGE_SHIFT; |
7803 | *percent = 0UL; | |
7804 | } | |
2a1e274a MG |
7805 | return 0; |
7806 | } | |
ed7ed365 | 7807 | |
7e63efef MG |
7808 | /* |
7809 | * kernelcore=size sets the amount of memory for use for allocations that | |
7810 | * cannot be reclaimed or migrated. | |
7811 | */ | |
7812 | static int __init cmdline_parse_kernelcore(char *p) | |
7813 | { | |
342332e6 TI |
7814 | /* parse kernelcore=mirror */ |
7815 | if (parse_option_str(p, "mirror")) { | |
7816 | mirrored_kernelcore = true; | |
7817 | return 0; | |
7818 | } | |
7819 | ||
a5c6d650 DR |
7820 | return cmdline_parse_core(p, &required_kernelcore, |
7821 | &required_kernelcore_percent); | |
7e63efef MG |
7822 | } |
7823 | ||
7824 | /* | |
7825 | * movablecore=size sets the amount of memory for use for allocations that | |
7826 | * can be reclaimed or migrated. | |
7827 | */ | |
7828 | static int __init cmdline_parse_movablecore(char *p) | |
7829 | { | |
a5c6d650 DR |
7830 | return cmdline_parse_core(p, &required_movablecore, |
7831 | &required_movablecore_percent); | |
7e63efef MG |
7832 | } |
7833 | ||
ed7ed365 | 7834 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7835 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7836 | |
c3d5f5f0 JL |
7837 | void adjust_managed_page_count(struct page *page, long count) |
7838 | { | |
9705bea5 | 7839 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 7840 | totalram_pages_add(count); |
3dcc0571 JL |
7841 | #ifdef CONFIG_HIGHMEM |
7842 | if (PageHighMem(page)) | |
ca79b0c2 | 7843 | totalhigh_pages_add(count); |
3dcc0571 | 7844 | #endif |
c3d5f5f0 | 7845 | } |
3dcc0571 | 7846 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7847 | |
e5cb113f | 7848 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 7849 | { |
11199692 JL |
7850 | void *pos; |
7851 | unsigned long pages = 0; | |
69afade7 | 7852 | |
11199692 JL |
7853 | start = (void *)PAGE_ALIGN((unsigned long)start); |
7854 | end = (void *)((unsigned long)end & PAGE_MASK); | |
7855 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
7856 | struct page *page = virt_to_page(pos); |
7857 | void *direct_map_addr; | |
7858 | ||
7859 | /* | |
7860 | * 'direct_map_addr' might be different from 'pos' | |
7861 | * because some architectures' virt_to_page() | |
7862 | * work with aliases. Getting the direct map | |
7863 | * address ensures that we get a _writeable_ | |
7864 | * alias for the memset(). | |
7865 | */ | |
7866 | direct_map_addr = page_address(page); | |
c746170d VF |
7867 | /* |
7868 | * Perform a kasan-unchecked memset() since this memory | |
7869 | * has not been initialized. | |
7870 | */ | |
7871 | direct_map_addr = kasan_reset_tag(direct_map_addr); | |
dbe67df4 | 7872 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
7873 | memset(direct_map_addr, poison, PAGE_SIZE); |
7874 | ||
7875 | free_reserved_page(page); | |
69afade7 JL |
7876 | } |
7877 | ||
7878 | if (pages && s) | |
adb1fe9a JP |
7879 | pr_info("Freeing %s memory: %ldK\n", |
7880 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
7881 | |
7882 | return pages; | |
7883 | } | |
7884 | ||
1f9d03c5 | 7885 | void __init mem_init_print_info(void) |
7ee3d4e8 JL |
7886 | { |
7887 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7888 | unsigned long init_code_size, init_data_size; | |
7889 | ||
7890 | physpages = get_num_physpages(); | |
7891 | codesize = _etext - _stext; | |
7892 | datasize = _edata - _sdata; | |
7893 | rosize = __end_rodata - __start_rodata; | |
7894 | bss_size = __bss_stop - __bss_start; | |
7895 | init_data_size = __init_end - __init_begin; | |
7896 | init_code_size = _einittext - _sinittext; | |
7897 | ||
7898 | /* | |
7899 | * Detect special cases and adjust section sizes accordingly: | |
7900 | * 1) .init.* may be embedded into .data sections | |
7901 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7902 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7903 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7904 | */ | |
7905 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7906 | do { \ |
7907 | if (start <= pos && pos < end && size > adj) \ | |
7908 | size -= adj; \ | |
7909 | } while (0) | |
7ee3d4e8 JL |
7910 | |
7911 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7912 | _sinittext, init_code_size); | |
7913 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7914 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7915 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7916 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7917 | ||
7918 | #undef adj_init_size | |
7919 | ||
756a025f | 7920 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7921 | #ifdef CONFIG_HIGHMEM |
756a025f | 7922 | ", %luK highmem" |
7ee3d4e8 | 7923 | #endif |
1f9d03c5 | 7924 | ")\n", |
756a025f JP |
7925 | nr_free_pages() << (PAGE_SHIFT - 10), |
7926 | physpages << (PAGE_SHIFT - 10), | |
7927 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7928 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
ca79b0c2 | 7929 | (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10), |
1f9d03c5 | 7930 | totalcma_pages << (PAGE_SHIFT - 10) |
7ee3d4e8 | 7931 | #ifdef CONFIG_HIGHMEM |
1f9d03c5 | 7932 | , totalhigh_pages() << (PAGE_SHIFT - 10) |
7ee3d4e8 | 7933 | #endif |
1f9d03c5 | 7934 | ); |
7ee3d4e8 JL |
7935 | } |
7936 | ||
0e0b864e | 7937 | /** |
88ca3b94 RD |
7938 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7939 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7940 | * |
013110a7 | 7941 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7942 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7943 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7944 | * function may optionally be used to account for unfreeable pages in the |
7945 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7946 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7947 | */ |
7948 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7949 | { | |
7950 | dma_reserve = new_dma_reserve; | |
7951 | } | |
7952 | ||
005fd4bb | 7953 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7954 | { |
1da177e4 | 7955 | |
005fd4bb SAS |
7956 | lru_add_drain_cpu(cpu); |
7957 | drain_pages(cpu); | |
9f8f2172 | 7958 | |
005fd4bb SAS |
7959 | /* |
7960 | * Spill the event counters of the dead processor | |
7961 | * into the current processors event counters. | |
7962 | * This artificially elevates the count of the current | |
7963 | * processor. | |
7964 | */ | |
7965 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7966 | |
005fd4bb SAS |
7967 | /* |
7968 | * Zero the differential counters of the dead processor | |
7969 | * so that the vm statistics are consistent. | |
7970 | * | |
7971 | * This is only okay since the processor is dead and cannot | |
7972 | * race with what we are doing. | |
7973 | */ | |
7974 | cpu_vm_stats_fold(cpu); | |
7975 | return 0; | |
1da177e4 | 7976 | } |
1da177e4 | 7977 | |
e03a5125 NP |
7978 | #ifdef CONFIG_NUMA |
7979 | int hashdist = HASHDIST_DEFAULT; | |
7980 | ||
7981 | static int __init set_hashdist(char *str) | |
7982 | { | |
7983 | if (!str) | |
7984 | return 0; | |
7985 | hashdist = simple_strtoul(str, &str, 0); | |
7986 | return 1; | |
7987 | } | |
7988 | __setup("hashdist=", set_hashdist); | |
7989 | #endif | |
7990 | ||
1da177e4 LT |
7991 | void __init page_alloc_init(void) |
7992 | { | |
005fd4bb SAS |
7993 | int ret; |
7994 | ||
e03a5125 NP |
7995 | #ifdef CONFIG_NUMA |
7996 | if (num_node_state(N_MEMORY) == 1) | |
7997 | hashdist = 0; | |
7998 | #endif | |
7999 | ||
005fd4bb SAS |
8000 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, |
8001 | "mm/page_alloc:dead", NULL, | |
8002 | page_alloc_cpu_dead); | |
8003 | WARN_ON(ret < 0); | |
1da177e4 LT |
8004 | } |
8005 | ||
cb45b0e9 | 8006 | /* |
34b10060 | 8007 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
8008 | * or min_free_kbytes changes. |
8009 | */ | |
8010 | static void calculate_totalreserve_pages(void) | |
8011 | { | |
8012 | struct pglist_data *pgdat; | |
8013 | unsigned long reserve_pages = 0; | |
2f6726e5 | 8014 | enum zone_type i, j; |
cb45b0e9 HA |
8015 | |
8016 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
8017 | |
8018 | pgdat->totalreserve_pages = 0; | |
8019 | ||
cb45b0e9 HA |
8020 | for (i = 0; i < MAX_NR_ZONES; i++) { |
8021 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 8022 | long max = 0; |
9705bea5 | 8023 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
8024 | |
8025 | /* Find valid and maximum lowmem_reserve in the zone */ | |
8026 | for (j = i; j < MAX_NR_ZONES; j++) { | |
8027 | if (zone->lowmem_reserve[j] > max) | |
8028 | max = zone->lowmem_reserve[j]; | |
8029 | } | |
8030 | ||
41858966 MG |
8031 | /* we treat the high watermark as reserved pages. */ |
8032 | max += high_wmark_pages(zone); | |
cb45b0e9 | 8033 | |
3d6357de AK |
8034 | if (max > managed_pages) |
8035 | max = managed_pages; | |
a8d01437 | 8036 | |
281e3726 | 8037 | pgdat->totalreserve_pages += max; |
a8d01437 | 8038 | |
cb45b0e9 HA |
8039 | reserve_pages += max; |
8040 | } | |
8041 | } | |
8042 | totalreserve_pages = reserve_pages; | |
8043 | } | |
8044 | ||
1da177e4 LT |
8045 | /* |
8046 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 8047 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
8048 | * has a correct pages reserved value, so an adequate number of |
8049 | * pages are left in the zone after a successful __alloc_pages(). | |
8050 | */ | |
8051 | static void setup_per_zone_lowmem_reserve(void) | |
8052 | { | |
8053 | struct pglist_data *pgdat; | |
470c61d7 | 8054 | enum zone_type i, j; |
1da177e4 | 8055 | |
ec936fc5 | 8056 | for_each_online_pgdat(pgdat) { |
470c61d7 LS |
8057 | for (i = 0; i < MAX_NR_ZONES - 1; i++) { |
8058 | struct zone *zone = &pgdat->node_zones[i]; | |
8059 | int ratio = sysctl_lowmem_reserve_ratio[i]; | |
8060 | bool clear = !ratio || !zone_managed_pages(zone); | |
8061 | unsigned long managed_pages = 0; | |
8062 | ||
8063 | for (j = i + 1; j < MAX_NR_ZONES; j++) { | |
8064 | if (clear) { | |
8065 | zone->lowmem_reserve[j] = 0; | |
d3cda233 | 8066 | } else { |
470c61d7 LS |
8067 | struct zone *upper_zone = &pgdat->node_zones[j]; |
8068 | ||
8069 | managed_pages += zone_managed_pages(upper_zone); | |
8070 | zone->lowmem_reserve[j] = managed_pages / ratio; | |
d3cda233 | 8071 | } |
1da177e4 LT |
8072 | } |
8073 | } | |
8074 | } | |
cb45b0e9 HA |
8075 | |
8076 | /* update totalreserve_pages */ | |
8077 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8078 | } |
8079 | ||
cfd3da1e | 8080 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
8081 | { |
8082 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
8083 | unsigned long lowmem_pages = 0; | |
8084 | struct zone *zone; | |
8085 | unsigned long flags; | |
8086 | ||
8087 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
8088 | for_each_zone(zone) { | |
8089 | if (!is_highmem(zone)) | |
9705bea5 | 8090 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
8091 | } |
8092 | ||
8093 | for_each_zone(zone) { | |
ac924c60 AM |
8094 | u64 tmp; |
8095 | ||
1125b4e3 | 8096 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 8097 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 8098 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
8099 | if (is_highmem(zone)) { |
8100 | /* | |
669ed175 NP |
8101 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
8102 | * need highmem pages, so cap pages_min to a small | |
8103 | * value here. | |
8104 | * | |
41858966 | 8105 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 8106 | * deltas control async page reclaim, and so should |
669ed175 | 8107 | * not be capped for highmem. |
1da177e4 | 8108 | */ |
90ae8d67 | 8109 | unsigned long min_pages; |
1da177e4 | 8110 | |
9705bea5 | 8111 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 8112 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 8113 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 8114 | } else { |
669ed175 NP |
8115 | /* |
8116 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
8117 | * proportionate to the zone's size. |
8118 | */ | |
a9214443 | 8119 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
8120 | } |
8121 | ||
795ae7a0 JW |
8122 | /* |
8123 | * Set the kswapd watermarks distance according to the | |
8124 | * scale factor in proportion to available memory, but | |
8125 | * ensure a minimum size on small systems. | |
8126 | */ | |
8127 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 8128 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
8129 | watermark_scale_factor, 10000)); |
8130 | ||
aa092591 | 8131 | zone->watermark_boost = 0; |
a9214443 MG |
8132 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
8133 | zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 8134 | |
1125b4e3 | 8135 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 8136 | } |
cb45b0e9 HA |
8137 | |
8138 | /* update totalreserve_pages */ | |
8139 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8140 | } |
8141 | ||
cfd3da1e MG |
8142 | /** |
8143 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
8144 | * or when memory is hot-{added|removed} | |
8145 | * | |
8146 | * Ensures that the watermark[min,low,high] values for each zone are set | |
8147 | * correctly with respect to min_free_kbytes. | |
8148 | */ | |
8149 | void setup_per_zone_wmarks(void) | |
8150 | { | |
b93e0f32 MH |
8151 | static DEFINE_SPINLOCK(lock); |
8152 | ||
8153 | spin_lock(&lock); | |
cfd3da1e | 8154 | __setup_per_zone_wmarks(); |
b93e0f32 | 8155 | spin_unlock(&lock); |
cfd3da1e MG |
8156 | } |
8157 | ||
1da177e4 LT |
8158 | /* |
8159 | * Initialise min_free_kbytes. | |
8160 | * | |
8161 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 8162 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
8163 | * bandwidth does not increase linearly with machine size. We use |
8164 | * | |
b8af2941 | 8165 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
8166 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
8167 | * | |
8168 | * which yields | |
8169 | * | |
8170 | * 16MB: 512k | |
8171 | * 32MB: 724k | |
8172 | * 64MB: 1024k | |
8173 | * 128MB: 1448k | |
8174 | * 256MB: 2048k | |
8175 | * 512MB: 2896k | |
8176 | * 1024MB: 4096k | |
8177 | * 2048MB: 5792k | |
8178 | * 4096MB: 8192k | |
8179 | * 8192MB: 11584k | |
8180 | * 16384MB: 16384k | |
8181 | */ | |
1b79acc9 | 8182 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
8183 | { |
8184 | unsigned long lowmem_kbytes; | |
5f12733e | 8185 | int new_min_free_kbytes; |
1da177e4 LT |
8186 | |
8187 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
8188 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
8189 | ||
8190 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
8191 | min_free_kbytes = new_min_free_kbytes; | |
8192 | if (min_free_kbytes < 128) | |
8193 | min_free_kbytes = 128; | |
ee8eb9a5 JS |
8194 | if (min_free_kbytes > 262144) |
8195 | min_free_kbytes = 262144; | |
5f12733e MH |
8196 | } else { |
8197 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
8198 | new_min_free_kbytes, user_min_free_kbytes); | |
8199 | } | |
bc75d33f | 8200 | setup_per_zone_wmarks(); |
a6cccdc3 | 8201 | refresh_zone_stat_thresholds(); |
1da177e4 | 8202 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
8203 | |
8204 | #ifdef CONFIG_NUMA | |
8205 | setup_min_unmapped_ratio(); | |
8206 | setup_min_slab_ratio(); | |
8207 | #endif | |
8208 | ||
4aab2be0 VB |
8209 | khugepaged_min_free_kbytes_update(); |
8210 | ||
1da177e4 LT |
8211 | return 0; |
8212 | } | |
e08d3fdf | 8213 | postcore_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
8214 | |
8215 | /* | |
b8af2941 | 8216 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
8217 | * that we can call two helper functions whenever min_free_kbytes |
8218 | * changes. | |
8219 | */ | |
cccad5b9 | 8220 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8221 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8222 | { |
da8c757b HP |
8223 | int rc; |
8224 | ||
8225 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8226 | if (rc) | |
8227 | return rc; | |
8228 | ||
5f12733e MH |
8229 | if (write) { |
8230 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 8231 | setup_per_zone_wmarks(); |
5f12733e | 8232 | } |
1da177e4 LT |
8233 | return 0; |
8234 | } | |
8235 | ||
795ae7a0 | 8236 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8237 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
8238 | { |
8239 | int rc; | |
8240 | ||
8241 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8242 | if (rc) | |
8243 | return rc; | |
8244 | ||
8245 | if (write) | |
8246 | setup_per_zone_wmarks(); | |
8247 | ||
8248 | return 0; | |
8249 | } | |
8250 | ||
9614634f | 8251 | #ifdef CONFIG_NUMA |
6423aa81 | 8252 | static void setup_min_unmapped_ratio(void) |
9614634f | 8253 | { |
6423aa81 | 8254 | pg_data_t *pgdat; |
9614634f | 8255 | struct zone *zone; |
9614634f | 8256 | |
a5f5f91d | 8257 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 8258 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 8259 | |
9614634f | 8260 | for_each_zone(zone) |
9705bea5 AK |
8261 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
8262 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 8263 | } |
0ff38490 | 8264 | |
6423aa81 JK |
8265 | |
8266 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8267 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 8268 | { |
0ff38490 CL |
8269 | int rc; |
8270 | ||
8d65af78 | 8271 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
8272 | if (rc) |
8273 | return rc; | |
8274 | ||
6423aa81 JK |
8275 | setup_min_unmapped_ratio(); |
8276 | ||
8277 | return 0; | |
8278 | } | |
8279 | ||
8280 | static void setup_min_slab_ratio(void) | |
8281 | { | |
8282 | pg_data_t *pgdat; | |
8283 | struct zone *zone; | |
8284 | ||
a5f5f91d MG |
8285 | for_each_online_pgdat(pgdat) |
8286 | pgdat->min_slab_pages = 0; | |
8287 | ||
0ff38490 | 8288 | for_each_zone(zone) |
9705bea5 AK |
8289 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
8290 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
8291 | } |
8292 | ||
8293 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8294 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
8295 | { |
8296 | int rc; | |
8297 | ||
8298 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8299 | if (rc) | |
8300 | return rc; | |
8301 | ||
8302 | setup_min_slab_ratio(); | |
8303 | ||
0ff38490 CL |
8304 | return 0; |
8305 | } | |
9614634f CL |
8306 | #endif |
8307 | ||
1da177e4 LT |
8308 | /* |
8309 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
8310 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
8311 | * whenever sysctl_lowmem_reserve_ratio changes. | |
8312 | * | |
8313 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 8314 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
8315 | * if in function of the boot time zone sizes. |
8316 | */ | |
cccad5b9 | 8317 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8318 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8319 | { |
86aaf255 BH |
8320 | int i; |
8321 | ||
8d65af78 | 8322 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
8323 | |
8324 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
8325 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
8326 | sysctl_lowmem_reserve_ratio[i] = 0; | |
8327 | } | |
8328 | ||
1da177e4 LT |
8329 | setup_per_zone_lowmem_reserve(); |
8330 | return 0; | |
8331 | } | |
8332 | ||
8ad4b1fb RS |
8333 | /* |
8334 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
8335 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
8336 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 8337 | */ |
cccad5b9 | 8338 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8339 | void *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
8340 | { |
8341 | struct zone *zone; | |
7cd2b0a3 | 8342 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
8343 | int ret; |
8344 | ||
7cd2b0a3 DR |
8345 | mutex_lock(&pcp_batch_high_lock); |
8346 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
8347 | ||
8d65af78 | 8348 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
8349 | if (!write || ret < 0) |
8350 | goto out; | |
8351 | ||
8352 | /* Sanity checking to avoid pcp imbalance */ | |
8353 | if (percpu_pagelist_fraction && | |
8354 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
8355 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
8356 | ret = -EINVAL; | |
8357 | goto out; | |
8358 | } | |
8359 | ||
8360 | /* No change? */ | |
8361 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
8362 | goto out; | |
c8e251fa | 8363 | |
cb1ef534 | 8364 | for_each_populated_zone(zone) |
0a8b4f1d | 8365 | zone_set_pageset_high_and_batch(zone); |
7cd2b0a3 | 8366 | out: |
c8e251fa | 8367 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 8368 | return ret; |
8ad4b1fb RS |
8369 | } |
8370 | ||
f6f34b43 SD |
8371 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
8372 | /* | |
8373 | * Returns the number of pages that arch has reserved but | |
8374 | * is not known to alloc_large_system_hash(). | |
8375 | */ | |
8376 | static unsigned long __init arch_reserved_kernel_pages(void) | |
8377 | { | |
8378 | return 0; | |
8379 | } | |
8380 | #endif | |
8381 | ||
9017217b PT |
8382 | /* |
8383 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
8384 | * machines. As memory size is increased the scale is also increased but at | |
8385 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
8386 | * quadruples the scale is increased by one, which means the size of hash table | |
8387 | * only doubles, instead of quadrupling as well. | |
8388 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
8389 | * makes sense, it is disabled on such platforms. | |
8390 | */ | |
8391 | #if __BITS_PER_LONG > 32 | |
8392 | #define ADAPT_SCALE_BASE (64ul << 30) | |
8393 | #define ADAPT_SCALE_SHIFT 2 | |
8394 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
8395 | #endif | |
8396 | ||
1da177e4 LT |
8397 | /* |
8398 | * allocate a large system hash table from bootmem | |
8399 | * - it is assumed that the hash table must contain an exact power-of-2 | |
8400 | * quantity of entries | |
8401 | * - limit is the number of hash buckets, not the total allocation size | |
8402 | */ | |
8403 | void *__init alloc_large_system_hash(const char *tablename, | |
8404 | unsigned long bucketsize, | |
8405 | unsigned long numentries, | |
8406 | int scale, | |
8407 | int flags, | |
8408 | unsigned int *_hash_shift, | |
8409 | unsigned int *_hash_mask, | |
31fe62b9 TB |
8410 | unsigned long low_limit, |
8411 | unsigned long high_limit) | |
1da177e4 | 8412 | { |
31fe62b9 | 8413 | unsigned long long max = high_limit; |
1da177e4 LT |
8414 | unsigned long log2qty, size; |
8415 | void *table = NULL; | |
3749a8f0 | 8416 | gfp_t gfp_flags; |
ec11408a | 8417 | bool virt; |
121e6f32 | 8418 | bool huge; |
1da177e4 LT |
8419 | |
8420 | /* allow the kernel cmdline to have a say */ | |
8421 | if (!numentries) { | |
8422 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 8423 | numentries = nr_kernel_pages; |
f6f34b43 | 8424 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
8425 | |
8426 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
8427 | if (PAGE_SHIFT < 20) | |
8428 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 8429 | |
9017217b PT |
8430 | #if __BITS_PER_LONG > 32 |
8431 | if (!high_limit) { | |
8432 | unsigned long adapt; | |
8433 | ||
8434 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
8435 | adapt <<= ADAPT_SCALE_SHIFT) | |
8436 | scale++; | |
8437 | } | |
8438 | #endif | |
8439 | ||
1da177e4 LT |
8440 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
8441 | if (scale > PAGE_SHIFT) | |
8442 | numentries >>= (scale - PAGE_SHIFT); | |
8443 | else | |
8444 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
8445 | |
8446 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
8447 | if (unlikely(flags & HASH_SMALL)) { |
8448 | /* Makes no sense without HASH_EARLY */ | |
8449 | WARN_ON(!(flags & HASH_EARLY)); | |
8450 | if (!(numentries >> *_hash_shift)) { | |
8451 | numentries = 1UL << *_hash_shift; | |
8452 | BUG_ON(!numentries); | |
8453 | } | |
8454 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 8455 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 8456 | } |
6e692ed3 | 8457 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
8458 | |
8459 | /* limit allocation size to 1/16 total memory by default */ | |
8460 | if (max == 0) { | |
8461 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
8462 | do_div(max, bucketsize); | |
8463 | } | |
074b8517 | 8464 | max = min(max, 0x80000000ULL); |
1da177e4 | 8465 | |
31fe62b9 TB |
8466 | if (numentries < low_limit) |
8467 | numentries = low_limit; | |
1da177e4 LT |
8468 | if (numentries > max) |
8469 | numentries = max; | |
8470 | ||
f0d1b0b3 | 8471 | log2qty = ilog2(numentries); |
1da177e4 | 8472 | |
3749a8f0 | 8473 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 8474 | do { |
ec11408a | 8475 | virt = false; |
1da177e4 | 8476 | size = bucketsize << log2qty; |
ea1f5f37 PT |
8477 | if (flags & HASH_EARLY) { |
8478 | if (flags & HASH_ZERO) | |
26fb3dae | 8479 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 8480 | else |
7e1c4e27 MR |
8481 | table = memblock_alloc_raw(size, |
8482 | SMP_CACHE_BYTES); | |
ec11408a | 8483 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
88dca4ca | 8484 | table = __vmalloc(size, gfp_flags); |
ec11408a | 8485 | virt = true; |
121e6f32 | 8486 | huge = is_vm_area_hugepages(table); |
ea1f5f37 | 8487 | } else { |
1037b83b ED |
8488 | /* |
8489 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
8490 | * some pages at the end of hash table which |
8491 | * alloc_pages_exact() automatically does | |
1037b83b | 8492 | */ |
ec11408a NP |
8493 | table = alloc_pages_exact(size, gfp_flags); |
8494 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
8495 | } |
8496 | } while (!table && size > PAGE_SIZE && --log2qty); | |
8497 | ||
8498 | if (!table) | |
8499 | panic("Failed to allocate %s hash table\n", tablename); | |
8500 | ||
ec11408a NP |
8501 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
8502 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
121e6f32 | 8503 | virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear"); |
1da177e4 LT |
8504 | |
8505 | if (_hash_shift) | |
8506 | *_hash_shift = log2qty; | |
8507 | if (_hash_mask) | |
8508 | *_hash_mask = (1 << log2qty) - 1; | |
8509 | ||
8510 | return table; | |
8511 | } | |
a117e66e | 8512 | |
a5d76b54 | 8513 | /* |
80934513 | 8514 | * This function checks whether pageblock includes unmovable pages or not. |
80934513 | 8515 | * |
b8af2941 | 8516 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
8517 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
8518 | * check without lock_page also may miss some movable non-lru pages at | |
8519 | * race condition. So you can't expect this function should be exact. | |
4a55c047 QC |
8520 | * |
8521 | * Returns a page without holding a reference. If the caller wants to | |
047b9967 | 8522 | * dereference that page (e.g., dumping), it has to make sure that it |
4a55c047 QC |
8523 | * cannot get removed (e.g., via memory unplug) concurrently. |
8524 | * | |
a5d76b54 | 8525 | */ |
4a55c047 QC |
8526 | struct page *has_unmovable_pages(struct zone *zone, struct page *page, |
8527 | int migratetype, int flags) | |
49ac8255 | 8528 | { |
1a9f2191 QC |
8529 | unsigned long iter = 0; |
8530 | unsigned long pfn = page_to_pfn(page); | |
6a654e36 | 8531 | unsigned long offset = pfn % pageblock_nr_pages; |
47118af0 | 8532 | |
1a9f2191 QC |
8533 | if (is_migrate_cma_page(page)) { |
8534 | /* | |
8535 | * CMA allocations (alloc_contig_range) really need to mark | |
8536 | * isolate CMA pageblocks even when they are not movable in fact | |
8537 | * so consider them movable here. | |
8538 | */ | |
8539 | if (is_migrate_cma(migratetype)) | |
4a55c047 | 8540 | return NULL; |
1a9f2191 | 8541 | |
3d680bdf | 8542 | return page; |
1a9f2191 | 8543 | } |
4da2ce25 | 8544 | |
6a654e36 | 8545 | for (; iter < pageblock_nr_pages - offset; iter++) { |
fe4c86c9 | 8546 | if (!pfn_valid_within(pfn + iter)) |
49ac8255 | 8547 | continue; |
29723fcc | 8548 | |
fe4c86c9 | 8549 | page = pfn_to_page(pfn + iter); |
c8721bbb | 8550 | |
c9c510dc DH |
8551 | /* |
8552 | * Both, bootmem allocations and memory holes are marked | |
8553 | * PG_reserved and are unmovable. We can even have unmovable | |
8554 | * allocations inside ZONE_MOVABLE, for example when | |
8555 | * specifying "movablecore". | |
8556 | */ | |
d7ab3672 | 8557 | if (PageReserved(page)) |
3d680bdf | 8558 | return page; |
d7ab3672 | 8559 | |
9d789999 MH |
8560 | /* |
8561 | * If the zone is movable and we have ruled out all reserved | |
8562 | * pages then it should be reasonably safe to assume the rest | |
8563 | * is movable. | |
8564 | */ | |
8565 | if (zone_idx(zone) == ZONE_MOVABLE) | |
8566 | continue; | |
8567 | ||
c8721bbb NH |
8568 | /* |
8569 | * Hugepages are not in LRU lists, but they're movable. | |
1da2f328 | 8570 | * THPs are on the LRU, but need to be counted as #small pages. |
8bb4e7a2 | 8571 | * We need not scan over tail pages because we don't |
c8721bbb NH |
8572 | * handle each tail page individually in migration. |
8573 | */ | |
1da2f328 | 8574 | if (PageHuge(page) || PageTransCompound(page)) { |
17e2e7d7 OS |
8575 | struct page *head = compound_head(page); |
8576 | unsigned int skip_pages; | |
464c7ffb | 8577 | |
1da2f328 RR |
8578 | if (PageHuge(page)) { |
8579 | if (!hugepage_migration_supported(page_hstate(head))) | |
8580 | return page; | |
8581 | } else if (!PageLRU(head) && !__PageMovable(head)) { | |
3d680bdf | 8582 | return page; |
1da2f328 | 8583 | } |
464c7ffb | 8584 | |
d8c6546b | 8585 | skip_pages = compound_nr(head) - (page - head); |
17e2e7d7 | 8586 | iter += skip_pages - 1; |
c8721bbb NH |
8587 | continue; |
8588 | } | |
8589 | ||
97d255c8 MK |
8590 | /* |
8591 | * We can't use page_count without pin a page | |
8592 | * because another CPU can free compound page. | |
8593 | * This check already skips compound tails of THP | |
0139aa7b | 8594 | * because their page->_refcount is zero at all time. |
97d255c8 | 8595 | */ |
fe896d18 | 8596 | if (!page_ref_count(page)) { |
49ac8255 | 8597 | if (PageBuddy(page)) |
ab130f91 | 8598 | iter += (1 << buddy_order(page)) - 1; |
49ac8255 KH |
8599 | continue; |
8600 | } | |
97d255c8 | 8601 | |
b023f468 WC |
8602 | /* |
8603 | * The HWPoisoned page may be not in buddy system, and | |
8604 | * page_count() is not 0. | |
8605 | */ | |
756d25be | 8606 | if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
b023f468 WC |
8607 | continue; |
8608 | ||
aa218795 DH |
8609 | /* |
8610 | * We treat all PageOffline() pages as movable when offlining | |
8611 | * to give drivers a chance to decrement their reference count | |
8612 | * in MEM_GOING_OFFLINE in order to indicate that these pages | |
8613 | * can be offlined as there are no direct references anymore. | |
8614 | * For actually unmovable PageOffline() where the driver does | |
8615 | * not support this, we will fail later when trying to actually | |
8616 | * move these pages that still have a reference count > 0. | |
8617 | * (false negatives in this function only) | |
8618 | */ | |
8619 | if ((flags & MEMORY_OFFLINE) && PageOffline(page)) | |
8620 | continue; | |
8621 | ||
fe4c86c9 | 8622 | if (__PageMovable(page) || PageLRU(page)) |
0efadf48 YX |
8623 | continue; |
8624 | ||
49ac8255 | 8625 | /* |
6b4f7799 JW |
8626 | * If there are RECLAIMABLE pages, we need to check |
8627 | * it. But now, memory offline itself doesn't call | |
8628 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 | 8629 | */ |
3d680bdf | 8630 | return page; |
49ac8255 | 8631 | } |
4a55c047 | 8632 | return NULL; |
49ac8255 KH |
8633 | } |
8634 | ||
8df995f6 | 8635 | #ifdef CONFIG_CONTIG_ALLOC |
041d3a8c MN |
8636 | static unsigned long pfn_max_align_down(unsigned long pfn) |
8637 | { | |
8638 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8639 | pageblock_nr_pages) - 1); | |
8640 | } | |
8641 | ||
8642 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
8643 | { | |
8644 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8645 | pageblock_nr_pages)); | |
8646 | } | |
8647 | ||
a1394bdd MK |
8648 | #if defined(CONFIG_DYNAMIC_DEBUG) || \ |
8649 | (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) | |
8650 | /* Usage: See admin-guide/dynamic-debug-howto.rst */ | |
8651 | static void alloc_contig_dump_pages(struct list_head *page_list) | |
8652 | { | |
8653 | DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure"); | |
8654 | ||
8655 | if (DYNAMIC_DEBUG_BRANCH(descriptor)) { | |
8656 | struct page *page; | |
8657 | ||
8658 | dump_stack(); | |
8659 | list_for_each_entry(page, page_list, lru) | |
8660 | dump_page(page, "migration failure"); | |
8661 | } | |
8662 | } | |
8663 | #else | |
8664 | static inline void alloc_contig_dump_pages(struct list_head *page_list) | |
8665 | { | |
8666 | } | |
8667 | #endif | |
8668 | ||
041d3a8c | 8669 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
8670 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
8671 | unsigned long start, unsigned long end) | |
041d3a8c MN |
8672 | { |
8673 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 8674 | unsigned int nr_reclaimed; |
041d3a8c MN |
8675 | unsigned long pfn = start; |
8676 | unsigned int tries = 0; | |
8677 | int ret = 0; | |
8b94e0b8 JK |
8678 | struct migration_target_control mtc = { |
8679 | .nid = zone_to_nid(cc->zone), | |
8680 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
8681 | }; | |
041d3a8c | 8682 | |
361a2a22 | 8683 | lru_cache_disable(); |
041d3a8c | 8684 | |
bb13ffeb | 8685 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
8686 | if (fatal_signal_pending(current)) { |
8687 | ret = -EINTR; | |
8688 | break; | |
8689 | } | |
8690 | ||
bb13ffeb MG |
8691 | if (list_empty(&cc->migratepages)) { |
8692 | cc->nr_migratepages = 0; | |
c2ad7a1f OS |
8693 | ret = isolate_migratepages_range(cc, pfn, end); |
8694 | if (ret && ret != -EAGAIN) | |
041d3a8c | 8695 | break; |
c2ad7a1f | 8696 | pfn = cc->migrate_pfn; |
041d3a8c MN |
8697 | tries = 0; |
8698 | } else if (++tries == 5) { | |
c8e28b47 | 8699 | ret = -EBUSY; |
041d3a8c MN |
8700 | break; |
8701 | } | |
8702 | ||
beb51eaa MK |
8703 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
8704 | &cc->migratepages); | |
8705 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 8706 | |
8b94e0b8 JK |
8707 | ret = migrate_pages(&cc->migratepages, alloc_migration_target, |
8708 | NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE); | |
c8e28b47 OS |
8709 | |
8710 | /* | |
8711 | * On -ENOMEM, migrate_pages() bails out right away. It is pointless | |
8712 | * to retry again over this error, so do the same here. | |
8713 | */ | |
8714 | if (ret == -ENOMEM) | |
8715 | break; | |
041d3a8c | 8716 | } |
d479960e | 8717 | |
361a2a22 | 8718 | lru_cache_enable(); |
2a6f5124 | 8719 | if (ret < 0) { |
a1394bdd | 8720 | alloc_contig_dump_pages(&cc->migratepages); |
2a6f5124 SP |
8721 | putback_movable_pages(&cc->migratepages); |
8722 | return ret; | |
8723 | } | |
8724 | return 0; | |
041d3a8c MN |
8725 | } |
8726 | ||
8727 | /** | |
8728 | * alloc_contig_range() -- tries to allocate given range of pages | |
8729 | * @start: start PFN to allocate | |
8730 | * @end: one-past-the-last PFN to allocate | |
f0953a1b | 8731 | * @migratetype: migratetype of the underlying pageblocks (either |
0815f3d8 MN |
8732 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
8733 | * in range must have the same migratetype and it must | |
8734 | * be either of the two. | |
ca96b625 | 8735 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
8736 | * |
8737 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 8738 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 8739 | * |
2c7452a0 MK |
8740 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
8741 | * pageblocks in the range. Once isolated, the pageblocks should not | |
8742 | * be modified by others. | |
041d3a8c | 8743 | * |
a862f68a | 8744 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
8745 | * pages which PFN is in [start, end) are allocated for the caller and |
8746 | * need to be freed with free_contig_range(). | |
8747 | */ | |
0815f3d8 | 8748 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 8749 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 8750 | { |
041d3a8c | 8751 | unsigned long outer_start, outer_end; |
d00181b9 KS |
8752 | unsigned int order; |
8753 | int ret = 0; | |
041d3a8c | 8754 | |
bb13ffeb MG |
8755 | struct compact_control cc = { |
8756 | .nr_migratepages = 0, | |
8757 | .order = -1, | |
8758 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 8759 | .mode = MIGRATE_SYNC, |
bb13ffeb | 8760 | .ignore_skip_hint = true, |
2583d671 | 8761 | .no_set_skip_hint = true, |
7dea19f9 | 8762 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 8763 | .alloc_contig = true, |
bb13ffeb MG |
8764 | }; |
8765 | INIT_LIST_HEAD(&cc.migratepages); | |
8766 | ||
041d3a8c MN |
8767 | /* |
8768 | * What we do here is we mark all pageblocks in range as | |
8769 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
8770 | * have different sizes, and due to the way page allocator | |
8771 | * work, we align the range to biggest of the two pages so | |
8772 | * that page allocator won't try to merge buddies from | |
8773 | * different pageblocks and change MIGRATE_ISOLATE to some | |
8774 | * other migration type. | |
8775 | * | |
8776 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
8777 | * migrate the pages from an unaligned range (ie. pages that | |
8778 | * we are interested in). This will put all the pages in | |
8779 | * range back to page allocator as MIGRATE_ISOLATE. | |
8780 | * | |
8781 | * When this is done, we take the pages in range from page | |
8782 | * allocator removing them from the buddy system. This way | |
8783 | * page allocator will never consider using them. | |
8784 | * | |
8785 | * This lets us mark the pageblocks back as | |
8786 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
8787 | * aligned range but not in the unaligned, original range are | |
8788 | * put back to page allocator so that buddy can use them. | |
8789 | */ | |
8790 | ||
8791 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
d381c547 | 8792 | pfn_max_align_up(end), migratetype, 0); |
3fa0c7c7 | 8793 | if (ret) |
86a595f9 | 8794 | return ret; |
041d3a8c | 8795 | |
7612921f VB |
8796 | drain_all_pages(cc.zone); |
8797 | ||
8ef5849f JK |
8798 | /* |
8799 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
8800 | * So, just fall through. test_pages_isolated() has a tracepoint |
8801 | * which will report the busy page. | |
8802 | * | |
8803 | * It is possible that busy pages could become available before | |
8804 | * the call to test_pages_isolated, and the range will actually be | |
8805 | * allocated. So, if we fall through be sure to clear ret so that | |
8806 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 8807 | */ |
bb13ffeb | 8808 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 8809 | if (ret && ret != -EBUSY) |
041d3a8c | 8810 | goto done; |
68d68ff6 | 8811 | ret = 0; |
041d3a8c MN |
8812 | |
8813 | /* | |
8814 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
8815 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
8816 | * more, all pages in [start, end) are free in page allocator. | |
8817 | * What we are going to do is to allocate all pages from | |
8818 | * [start, end) (that is remove them from page allocator). | |
8819 | * | |
8820 | * The only problem is that pages at the beginning and at the | |
8821 | * end of interesting range may be not aligned with pages that | |
8822 | * page allocator holds, ie. they can be part of higher order | |
8823 | * pages. Because of this, we reserve the bigger range and | |
8824 | * once this is done free the pages we are not interested in. | |
8825 | * | |
8826 | * We don't have to hold zone->lock here because the pages are | |
8827 | * isolated thus they won't get removed from buddy. | |
8828 | */ | |
8829 | ||
041d3a8c MN |
8830 | order = 0; |
8831 | outer_start = start; | |
8832 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
8833 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
8834 | outer_start = start; |
8835 | break; | |
041d3a8c MN |
8836 | } |
8837 | outer_start &= ~0UL << order; | |
8838 | } | |
8839 | ||
8ef5849f | 8840 | if (outer_start != start) { |
ab130f91 | 8841 | order = buddy_order(pfn_to_page(outer_start)); |
8ef5849f JK |
8842 | |
8843 | /* | |
8844 | * outer_start page could be small order buddy page and | |
8845 | * it doesn't include start page. Adjust outer_start | |
8846 | * in this case to report failed page properly | |
8847 | * on tracepoint in test_pages_isolated() | |
8848 | */ | |
8849 | if (outer_start + (1UL << order) <= start) | |
8850 | outer_start = start; | |
8851 | } | |
8852 | ||
041d3a8c | 8853 | /* Make sure the range is really isolated. */ |
756d25be | 8854 | if (test_pages_isolated(outer_start, end, 0)) { |
041d3a8c MN |
8855 | ret = -EBUSY; |
8856 | goto done; | |
8857 | } | |
8858 | ||
49f223a9 | 8859 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 8860 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
8861 | if (!outer_end) { |
8862 | ret = -EBUSY; | |
8863 | goto done; | |
8864 | } | |
8865 | ||
8866 | /* Free head and tail (if any) */ | |
8867 | if (start != outer_start) | |
8868 | free_contig_range(outer_start, start - outer_start); | |
8869 | if (end != outer_end) | |
8870 | free_contig_range(end, outer_end - end); | |
8871 | ||
8872 | done: | |
8873 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 8874 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
8875 | return ret; |
8876 | } | |
255f5985 | 8877 | EXPORT_SYMBOL(alloc_contig_range); |
5e27a2df AK |
8878 | |
8879 | static int __alloc_contig_pages(unsigned long start_pfn, | |
8880 | unsigned long nr_pages, gfp_t gfp_mask) | |
8881 | { | |
8882 | unsigned long end_pfn = start_pfn + nr_pages; | |
8883 | ||
8884 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
8885 | gfp_mask); | |
8886 | } | |
8887 | ||
8888 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
8889 | unsigned long nr_pages) | |
8890 | { | |
8891 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
8892 | struct page *page; | |
8893 | ||
8894 | for (i = start_pfn; i < end_pfn; i++) { | |
8895 | page = pfn_to_online_page(i); | |
8896 | if (!page) | |
8897 | return false; | |
8898 | ||
8899 | if (page_zone(page) != z) | |
8900 | return false; | |
8901 | ||
8902 | if (PageReserved(page)) | |
8903 | return false; | |
5e27a2df AK |
8904 | } |
8905 | return true; | |
8906 | } | |
8907 | ||
8908 | static bool zone_spans_last_pfn(const struct zone *zone, | |
8909 | unsigned long start_pfn, unsigned long nr_pages) | |
8910 | { | |
8911 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
8912 | ||
8913 | return zone_spans_pfn(zone, last_pfn); | |
8914 | } | |
8915 | ||
8916 | /** | |
8917 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
8918 | * @nr_pages: Number of contiguous pages to allocate | |
8919 | * @gfp_mask: GFP mask to limit search and used during compaction | |
8920 | * @nid: Target node | |
8921 | * @nodemask: Mask for other possible nodes | |
8922 | * | |
8923 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
8924 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
8925 | * tried for allocation with alloc_contig_range(). This routine is intended | |
8926 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
8927 | * | |
8928 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
8929 | * power of two then the alignment is guaranteed to be to the given nr_pages | |
8930 | * (e.g. 1GB request would be aligned to 1GB). | |
8931 | * | |
8932 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
8933 | * __free_page() on each allocated page. | |
8934 | * | |
8935 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
8936 | */ | |
8937 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
8938 | int nid, nodemask_t *nodemask) | |
8939 | { | |
8940 | unsigned long ret, pfn, flags; | |
8941 | struct zonelist *zonelist; | |
8942 | struct zone *zone; | |
8943 | struct zoneref *z; | |
8944 | ||
8945 | zonelist = node_zonelist(nid, gfp_mask); | |
8946 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
8947 | gfp_zone(gfp_mask), nodemask) { | |
8948 | spin_lock_irqsave(&zone->lock, flags); | |
8949 | ||
8950 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
8951 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
8952 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
8953 | /* | |
8954 | * We release the zone lock here because | |
8955 | * alloc_contig_range() will also lock the zone | |
8956 | * at some point. If there's an allocation | |
8957 | * spinning on this lock, it may win the race | |
8958 | * and cause alloc_contig_range() to fail... | |
8959 | */ | |
8960 | spin_unlock_irqrestore(&zone->lock, flags); | |
8961 | ret = __alloc_contig_pages(pfn, nr_pages, | |
8962 | gfp_mask); | |
8963 | if (!ret) | |
8964 | return pfn_to_page(pfn); | |
8965 | spin_lock_irqsave(&zone->lock, flags); | |
8966 | } | |
8967 | pfn += nr_pages; | |
8968 | } | |
8969 | spin_unlock_irqrestore(&zone->lock, flags); | |
8970 | } | |
8971 | return NULL; | |
8972 | } | |
4eb0716e | 8973 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 8974 | |
78fa5150 | 8975 | void free_contig_range(unsigned long pfn, unsigned long nr_pages) |
041d3a8c | 8976 | { |
78fa5150 | 8977 | unsigned long count = 0; |
bcc2b02f MS |
8978 | |
8979 | for (; nr_pages--; pfn++) { | |
8980 | struct page *page = pfn_to_page(pfn); | |
8981 | ||
8982 | count += page_count(page) != 1; | |
8983 | __free_page(page); | |
8984 | } | |
78fa5150 | 8985 | WARN(count != 0, "%lu pages are still in use!\n", count); |
041d3a8c | 8986 | } |
255f5985 | 8987 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 8988 | |
0a647f38 CS |
8989 | /* |
8990 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
f0953a1b | 8991 | * page high values need to be recalculated. |
0a647f38 | 8992 | */ |
4ed7e022 JL |
8993 | void __meminit zone_pcp_update(struct zone *zone) |
8994 | { | |
c8e251fa | 8995 | mutex_lock(&pcp_batch_high_lock); |
0a8b4f1d | 8996 | zone_set_pageset_high_and_batch(zone); |
c8e251fa | 8997 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 | 8998 | } |
4ed7e022 | 8999 | |
ec6e8c7e VB |
9000 | /* |
9001 | * Effectively disable pcplists for the zone by setting the high limit to 0 | |
9002 | * and draining all cpus. A concurrent page freeing on another CPU that's about | |
9003 | * to put the page on pcplist will either finish before the drain and the page | |
9004 | * will be drained, or observe the new high limit and skip the pcplist. | |
9005 | * | |
9006 | * Must be paired with a call to zone_pcp_enable(). | |
9007 | */ | |
9008 | void zone_pcp_disable(struct zone *zone) | |
9009 | { | |
9010 | mutex_lock(&pcp_batch_high_lock); | |
9011 | __zone_set_pageset_high_and_batch(zone, 0, 1); | |
9012 | __drain_all_pages(zone, true); | |
9013 | } | |
9014 | ||
9015 | void zone_pcp_enable(struct zone *zone) | |
9016 | { | |
9017 | __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch); | |
9018 | mutex_unlock(&pcp_batch_high_lock); | |
9019 | } | |
9020 | ||
340175b7 JL |
9021 | void zone_pcp_reset(struct zone *zone) |
9022 | { | |
5a883813 MK |
9023 | int cpu; |
9024 | struct per_cpu_pageset *pset; | |
340175b7 | 9025 | |
340175b7 | 9026 | if (zone->pageset != &boot_pageset) { |
5a883813 MK |
9027 | for_each_online_cpu(cpu) { |
9028 | pset = per_cpu_ptr(zone->pageset, cpu); | |
9029 | drain_zonestat(zone, pset); | |
9030 | } | |
340175b7 JL |
9031 | free_percpu(zone->pageset); |
9032 | zone->pageset = &boot_pageset; | |
9033 | } | |
340175b7 JL |
9034 | } |
9035 | ||
6dcd73d7 | 9036 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 9037 | /* |
257bea71 DH |
9038 | * All pages in the range must be in a single zone, must not contain holes, |
9039 | * must span full sections, and must be isolated before calling this function. | |
0c0e6195 | 9040 | */ |
257bea71 | 9041 | void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 9042 | { |
257bea71 | 9043 | unsigned long pfn = start_pfn; |
0c0e6195 KH |
9044 | struct page *page; |
9045 | struct zone *zone; | |
0ee5f4f3 | 9046 | unsigned int order; |
0c0e6195 | 9047 | unsigned long flags; |
5557c766 | 9048 | |
2d070eab | 9049 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
9050 | zone = page_zone(pfn_to_page(pfn)); |
9051 | spin_lock_irqsave(&zone->lock, flags); | |
0c0e6195 | 9052 | while (pfn < end_pfn) { |
0c0e6195 | 9053 | page = pfn_to_page(pfn); |
b023f468 WC |
9054 | /* |
9055 | * The HWPoisoned page may be not in buddy system, and | |
9056 | * page_count() is not 0. | |
9057 | */ | |
9058 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
9059 | pfn++; | |
b023f468 WC |
9060 | continue; |
9061 | } | |
aa218795 DH |
9062 | /* |
9063 | * At this point all remaining PageOffline() pages have a | |
9064 | * reference count of 0 and can simply be skipped. | |
9065 | */ | |
9066 | if (PageOffline(page)) { | |
9067 | BUG_ON(page_count(page)); | |
9068 | BUG_ON(PageBuddy(page)); | |
9069 | pfn++; | |
aa218795 DH |
9070 | continue; |
9071 | } | |
b023f468 | 9072 | |
0c0e6195 KH |
9073 | BUG_ON(page_count(page)); |
9074 | BUG_ON(!PageBuddy(page)); | |
ab130f91 | 9075 | order = buddy_order(page); |
6ab01363 | 9076 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
9077 | pfn += (1 << order); |
9078 | } | |
9079 | spin_unlock_irqrestore(&zone->lock, flags); | |
9080 | } | |
9081 | #endif | |
8d22ba1b | 9082 | |
8d22ba1b WF |
9083 | bool is_free_buddy_page(struct page *page) |
9084 | { | |
9085 | struct zone *zone = page_zone(page); | |
9086 | unsigned long pfn = page_to_pfn(page); | |
9087 | unsigned long flags; | |
7aeb09f9 | 9088 | unsigned int order; |
8d22ba1b WF |
9089 | |
9090 | spin_lock_irqsave(&zone->lock, flags); | |
9091 | for (order = 0; order < MAX_ORDER; order++) { | |
9092 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
9093 | ||
ab130f91 | 9094 | if (PageBuddy(page_head) && buddy_order(page_head) >= order) |
8d22ba1b WF |
9095 | break; |
9096 | } | |
9097 | spin_unlock_irqrestore(&zone->lock, flags); | |
9098 | ||
9099 | return order < MAX_ORDER; | |
9100 | } | |
d4ae9916 NH |
9101 | |
9102 | #ifdef CONFIG_MEMORY_FAILURE | |
9103 | /* | |
06be6ff3 OS |
9104 | * Break down a higher-order page in sub-pages, and keep our target out of |
9105 | * buddy allocator. | |
d4ae9916 | 9106 | */ |
06be6ff3 OS |
9107 | static void break_down_buddy_pages(struct zone *zone, struct page *page, |
9108 | struct page *target, int low, int high, | |
9109 | int migratetype) | |
9110 | { | |
9111 | unsigned long size = 1 << high; | |
9112 | struct page *current_buddy, *next_page; | |
9113 | ||
9114 | while (high > low) { | |
9115 | high--; | |
9116 | size >>= 1; | |
9117 | ||
9118 | if (target >= &page[size]) { | |
9119 | next_page = page + size; | |
9120 | current_buddy = page; | |
9121 | } else { | |
9122 | next_page = page; | |
9123 | current_buddy = page + size; | |
9124 | } | |
9125 | ||
9126 | if (set_page_guard(zone, current_buddy, high, migratetype)) | |
9127 | continue; | |
9128 | ||
9129 | if (current_buddy != target) { | |
9130 | add_to_free_list(current_buddy, zone, high, migratetype); | |
ab130f91 | 9131 | set_buddy_order(current_buddy, high); |
06be6ff3 OS |
9132 | page = next_page; |
9133 | } | |
9134 | } | |
9135 | } | |
9136 | ||
9137 | /* | |
9138 | * Take a page that will be marked as poisoned off the buddy allocator. | |
9139 | */ | |
9140 | bool take_page_off_buddy(struct page *page) | |
d4ae9916 NH |
9141 | { |
9142 | struct zone *zone = page_zone(page); | |
9143 | unsigned long pfn = page_to_pfn(page); | |
9144 | unsigned long flags; | |
9145 | unsigned int order; | |
06be6ff3 | 9146 | bool ret = false; |
d4ae9916 NH |
9147 | |
9148 | spin_lock_irqsave(&zone->lock, flags); | |
9149 | for (order = 0; order < MAX_ORDER; order++) { | |
9150 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
ab130f91 | 9151 | int page_order = buddy_order(page_head); |
d4ae9916 | 9152 | |
ab130f91 | 9153 | if (PageBuddy(page_head) && page_order >= order) { |
06be6ff3 OS |
9154 | unsigned long pfn_head = page_to_pfn(page_head); |
9155 | int migratetype = get_pfnblock_migratetype(page_head, | |
9156 | pfn_head); | |
9157 | ||
ab130f91 | 9158 | del_page_from_free_list(page_head, zone, page_order); |
06be6ff3 | 9159 | break_down_buddy_pages(zone, page_head, page, 0, |
ab130f91 | 9160 | page_order, migratetype); |
bac9c6fa DH |
9161 | if (!is_migrate_isolate(migratetype)) |
9162 | __mod_zone_freepage_state(zone, -1, migratetype); | |
06be6ff3 | 9163 | ret = true; |
d4ae9916 NH |
9164 | break; |
9165 | } | |
06be6ff3 OS |
9166 | if (page_count(page_head) > 0) |
9167 | break; | |
d4ae9916 NH |
9168 | } |
9169 | spin_unlock_irqrestore(&zone->lock, flags); | |
06be6ff3 | 9170 | return ret; |
d4ae9916 NH |
9171 | } |
9172 | #endif |