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 | { | |
789 | if (_init_on_alloc_enabled_early) { | |
790 | if (page_poisoning_enabled()) | |
791 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, " | |
792 | "will take precedence over init_on_alloc\n"); | |
793 | else | |
794 | static_branch_enable(&init_on_alloc); | |
795 | } | |
796 | if (_init_on_free_enabled_early) { | |
797 | if (page_poisoning_enabled()) | |
798 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, " | |
799 | "will take precedence over init_on_free\n"); | |
800 | else | |
801 | static_branch_enable(&init_on_free); | |
802 | } | |
803 | ||
8db26a3d VB |
804 | #ifdef CONFIG_PAGE_POISONING |
805 | /* | |
806 | * Page poisoning is debug page alloc for some arches. If | |
807 | * either of those options are enabled, enable poisoning. | |
808 | */ | |
809 | if (page_poisoning_enabled() || | |
810 | (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) && | |
811 | debug_pagealloc_enabled())) | |
812 | static_branch_enable(&_page_poisoning_enabled); | |
813 | #endif | |
814 | ||
04013513 VB |
815 | #ifdef CONFIG_DEBUG_PAGEALLOC |
816 | if (!debug_pagealloc_enabled()) | |
817 | return; | |
818 | ||
819 | static_branch_enable(&_debug_pagealloc_enabled); | |
820 | ||
821 | if (!debug_guardpage_minorder()) | |
822 | return; | |
823 | ||
824 | static_branch_enable(&_debug_guardpage_enabled); | |
825 | #endif | |
826 | } | |
827 | ||
ab130f91 | 828 | static inline void set_buddy_order(struct page *page, unsigned int order) |
6aa3001b | 829 | { |
4c21e2f2 | 830 | set_page_private(page, order); |
676165a8 | 831 | __SetPageBuddy(page); |
1da177e4 LT |
832 | } |
833 | ||
1da177e4 LT |
834 | /* |
835 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 836 | * we can coalesce a page and its buddy if |
13ad59df | 837 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 838 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
839 | * (c) a page and its buddy have the same order && |
840 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 841 | * |
6e292b9b MW |
842 | * For recording whether a page is in the buddy system, we set PageBuddy. |
843 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 844 | * |
676165a8 | 845 | * For recording page's order, we use page_private(page). |
1da177e4 | 846 | */ |
fe925c0c | 847 | static inline bool page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 848 | unsigned int order) |
1da177e4 | 849 | { |
fe925c0c | 850 | if (!page_is_guard(buddy) && !PageBuddy(buddy)) |
851 | return false; | |
4c5018ce | 852 | |
ab130f91 | 853 | if (buddy_order(buddy) != order) |
fe925c0c | 854 | return false; |
c0a32fc5 | 855 | |
fe925c0c | 856 | /* |
857 | * zone check is done late to avoid uselessly calculating | |
858 | * zone/node ids for pages that could never merge. | |
859 | */ | |
860 | if (page_zone_id(page) != page_zone_id(buddy)) | |
861 | return false; | |
d34c5fa0 | 862 | |
fe925c0c | 863 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
4c5018ce | 864 | |
fe925c0c | 865 | return true; |
1da177e4 LT |
866 | } |
867 | ||
5e1f0f09 MG |
868 | #ifdef CONFIG_COMPACTION |
869 | static inline struct capture_control *task_capc(struct zone *zone) | |
870 | { | |
871 | struct capture_control *capc = current->capture_control; | |
872 | ||
deba0487 | 873 | return unlikely(capc) && |
5e1f0f09 MG |
874 | !(current->flags & PF_KTHREAD) && |
875 | !capc->page && | |
deba0487 | 876 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
877 | } |
878 | ||
879 | static inline bool | |
880 | compaction_capture(struct capture_control *capc, struct page *page, | |
881 | int order, int migratetype) | |
882 | { | |
883 | if (!capc || order != capc->cc->order) | |
884 | return false; | |
885 | ||
886 | /* Do not accidentally pollute CMA or isolated regions*/ | |
887 | if (is_migrate_cma(migratetype) || | |
888 | is_migrate_isolate(migratetype)) | |
889 | return false; | |
890 | ||
891 | /* | |
892 | * Do not let lower order allocations polluate a movable pageblock. | |
893 | * This might let an unmovable request use a reclaimable pageblock | |
894 | * and vice-versa but no more than normal fallback logic which can | |
895 | * have trouble finding a high-order free page. | |
896 | */ | |
897 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
898 | return false; | |
899 | ||
900 | capc->page = page; | |
901 | return true; | |
902 | } | |
903 | ||
904 | #else | |
905 | static inline struct capture_control *task_capc(struct zone *zone) | |
906 | { | |
907 | return NULL; | |
908 | } | |
909 | ||
910 | static inline bool | |
911 | compaction_capture(struct capture_control *capc, struct page *page, | |
912 | int order, int migratetype) | |
913 | { | |
914 | return false; | |
915 | } | |
916 | #endif /* CONFIG_COMPACTION */ | |
917 | ||
6ab01363 AD |
918 | /* Used for pages not on another list */ |
919 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
920 | unsigned int order, int migratetype) | |
921 | { | |
922 | struct free_area *area = &zone->free_area[order]; | |
923 | ||
924 | list_add(&page->lru, &area->free_list[migratetype]); | |
925 | area->nr_free++; | |
926 | } | |
927 | ||
928 | /* Used for pages not on another list */ | |
929 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
930 | unsigned int order, int migratetype) | |
931 | { | |
932 | struct free_area *area = &zone->free_area[order]; | |
933 | ||
934 | list_add_tail(&page->lru, &area->free_list[migratetype]); | |
935 | area->nr_free++; | |
936 | } | |
937 | ||
293ffa5e DH |
938 | /* |
939 | * Used for pages which are on another list. Move the pages to the tail | |
940 | * of the list - so the moved pages won't immediately be considered for | |
941 | * allocation again (e.g., optimization for memory onlining). | |
942 | */ | |
6ab01363 AD |
943 | static inline void move_to_free_list(struct page *page, struct zone *zone, |
944 | unsigned int order, int migratetype) | |
945 | { | |
946 | struct free_area *area = &zone->free_area[order]; | |
947 | ||
293ffa5e | 948 | list_move_tail(&page->lru, &area->free_list[migratetype]); |
6ab01363 AD |
949 | } |
950 | ||
951 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
952 | unsigned int order) | |
953 | { | |
36e66c55 AD |
954 | /* clear reported state and update reported page count */ |
955 | if (page_reported(page)) | |
956 | __ClearPageReported(page); | |
957 | ||
6ab01363 AD |
958 | list_del(&page->lru); |
959 | __ClearPageBuddy(page); | |
960 | set_page_private(page, 0); | |
961 | zone->free_area[order].nr_free--; | |
962 | } | |
963 | ||
a2129f24 AD |
964 | /* |
965 | * If this is not the largest possible page, check if the buddy | |
966 | * of the next-highest order is free. If it is, it's possible | |
967 | * that pages are being freed that will coalesce soon. In case, | |
968 | * that is happening, add the free page to the tail of the list | |
969 | * so it's less likely to be used soon and more likely to be merged | |
970 | * as a higher order page | |
971 | */ | |
972 | static inline bool | |
973 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
974 | struct page *page, unsigned int order) | |
975 | { | |
976 | struct page *higher_page, *higher_buddy; | |
977 | unsigned long combined_pfn; | |
978 | ||
979 | if (order >= MAX_ORDER - 2) | |
980 | return false; | |
981 | ||
982 | if (!pfn_valid_within(buddy_pfn)) | |
983 | return false; | |
984 | ||
985 | combined_pfn = buddy_pfn & pfn; | |
986 | higher_page = page + (combined_pfn - pfn); | |
987 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
988 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
989 | ||
990 | return pfn_valid_within(buddy_pfn) && | |
991 | page_is_buddy(higher_page, higher_buddy, order + 1); | |
992 | } | |
993 | ||
1da177e4 LT |
994 | /* |
995 | * Freeing function for a buddy system allocator. | |
996 | * | |
997 | * The concept of a buddy system is to maintain direct-mapped table | |
998 | * (containing bit values) for memory blocks of various "orders". | |
999 | * The bottom level table contains the map for the smallest allocatable | |
1000 | * units of memory (here, pages), and each level above it describes | |
1001 | * pairs of units from the levels below, hence, "buddies". | |
1002 | * At a high level, all that happens here is marking the table entry | |
1003 | * at the bottom level available, and propagating the changes upward | |
1004 | * as necessary, plus some accounting needed to play nicely with other | |
1005 | * parts of the VM system. | |
1006 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
1007 | * free pages of length of (1 << order) and marked with PageBuddy. |
1008 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 1009 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
1010 | * other. That is, if we allocate a small block, and both were |
1011 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 1012 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 1013 | * triggers coalescing into a block of larger size. |
1da177e4 | 1014 | * |
6d49e352 | 1015 | * -- nyc |
1da177e4 LT |
1016 | */ |
1017 | ||
48db57f8 | 1018 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 1019 | unsigned long pfn, |
ed0ae21d | 1020 | struct zone *zone, unsigned int order, |
f04a5d5d | 1021 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1022 | { |
a2129f24 | 1023 | struct capture_control *capc = task_capc(zone); |
3f649ab7 | 1024 | unsigned long buddy_pfn; |
a2129f24 | 1025 | unsigned long combined_pfn; |
d9dddbf5 | 1026 | unsigned int max_order; |
a2129f24 AD |
1027 | struct page *buddy; |
1028 | bool to_tail; | |
d9dddbf5 | 1029 | |
7ad69832 | 1030 | max_order = min_t(unsigned int, MAX_ORDER - 1, pageblock_order); |
1da177e4 | 1031 | |
d29bb978 | 1032 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 1033 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 1034 | |
ed0ae21d | 1035 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 1036 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 1037 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 1038 | |
76741e77 | 1039 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 1040 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 1041 | |
d9dddbf5 | 1042 | continue_merging: |
7ad69832 | 1043 | while (order < max_order) { |
5e1f0f09 MG |
1044 | if (compaction_capture(capc, page, order, migratetype)) { |
1045 | __mod_zone_freepage_state(zone, -(1 << order), | |
1046 | migratetype); | |
1047 | return; | |
1048 | } | |
76741e77 VB |
1049 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1050 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
1051 | |
1052 | if (!pfn_valid_within(buddy_pfn)) | |
1053 | goto done_merging; | |
cb2b95e1 | 1054 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 1055 | goto done_merging; |
c0a32fc5 SG |
1056 | /* |
1057 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
1058 | * merge with it and move up one order. | |
1059 | */ | |
b03641af | 1060 | if (page_is_guard(buddy)) |
2847cf95 | 1061 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 1062 | else |
6ab01363 | 1063 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
1064 | combined_pfn = buddy_pfn & pfn; |
1065 | page = page + (combined_pfn - pfn); | |
1066 | pfn = combined_pfn; | |
1da177e4 LT |
1067 | order++; |
1068 | } | |
7ad69832 | 1069 | if (order < MAX_ORDER - 1) { |
d9dddbf5 VB |
1070 | /* If we are here, it means order is >= pageblock_order. |
1071 | * We want to prevent merge between freepages on isolate | |
1072 | * pageblock and normal pageblock. Without this, pageblock | |
1073 | * isolation could cause incorrect freepage or CMA accounting. | |
1074 | * | |
1075 | * We don't want to hit this code for the more frequent | |
1076 | * low-order merging. | |
1077 | */ | |
1078 | if (unlikely(has_isolate_pageblock(zone))) { | |
1079 | int buddy_mt; | |
1080 | ||
76741e77 VB |
1081 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1082 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
1083 | buddy_mt = get_pageblock_migratetype(buddy); |
1084 | ||
1085 | if (migratetype != buddy_mt | |
1086 | && (is_migrate_isolate(migratetype) || | |
1087 | is_migrate_isolate(buddy_mt))) | |
1088 | goto done_merging; | |
1089 | } | |
7ad69832 | 1090 | max_order = order + 1; |
d9dddbf5 VB |
1091 | goto continue_merging; |
1092 | } | |
1093 | ||
1094 | done_merging: | |
ab130f91 | 1095 | set_buddy_order(page, order); |
6dda9d55 | 1096 | |
47b6a24a DH |
1097 | if (fpi_flags & FPI_TO_TAIL) |
1098 | to_tail = true; | |
1099 | else if (is_shuffle_order(order)) | |
a2129f24 | 1100 | to_tail = shuffle_pick_tail(); |
97500a4a | 1101 | else |
a2129f24 | 1102 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 1103 | |
a2129f24 | 1104 | if (to_tail) |
6ab01363 | 1105 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 1106 | else |
6ab01363 | 1107 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
1108 | |
1109 | /* Notify page reporting subsystem of freed page */ | |
f04a5d5d | 1110 | if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY)) |
36e66c55 | 1111 | page_reporting_notify_free(order); |
1da177e4 LT |
1112 | } |
1113 | ||
7bfec6f4 MG |
1114 | /* |
1115 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1116 | * try and check multiple fields with one check. The caller must do a detailed | |
1117 | * check if necessary. | |
1118 | */ | |
1119 | static inline bool page_expected_state(struct page *page, | |
1120 | unsigned long check_flags) | |
1121 | { | |
1122 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1123 | return false; | |
1124 | ||
1125 | if (unlikely((unsigned long)page->mapping | | |
1126 | page_ref_count(page) | | |
1127 | #ifdef CONFIG_MEMCG | |
48060834 | 1128 | page->memcg_data | |
7bfec6f4 MG |
1129 | #endif |
1130 | (page->flags & check_flags))) | |
1131 | return false; | |
1132 | ||
1133 | return true; | |
1134 | } | |
1135 | ||
58b7f119 | 1136 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 1137 | { |
82a3241a | 1138 | const char *bad_reason = NULL; |
f0b791a3 | 1139 | |
53f9263b | 1140 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1141 | bad_reason = "nonzero mapcount"; |
1142 | if (unlikely(page->mapping != NULL)) | |
1143 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1144 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1145 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
1146 | if (unlikely(page->flags & flags)) { |
1147 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
1148 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
1149 | else | |
1150 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 1151 | } |
9edad6ea | 1152 | #ifdef CONFIG_MEMCG |
48060834 | 1153 | if (unlikely(page->memcg_data)) |
9edad6ea JW |
1154 | bad_reason = "page still charged to cgroup"; |
1155 | #endif | |
58b7f119 WY |
1156 | return bad_reason; |
1157 | } | |
1158 | ||
1159 | static void check_free_page_bad(struct page *page) | |
1160 | { | |
1161 | bad_page(page, | |
1162 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
1163 | } |
1164 | ||
534fe5e3 | 1165 | static inline int check_free_page(struct page *page) |
bb552ac6 | 1166 | { |
da838d4f | 1167 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 1168 | return 0; |
bb552ac6 MG |
1169 | |
1170 | /* Something has gone sideways, find it */ | |
0d0c48a2 | 1171 | check_free_page_bad(page); |
7bfec6f4 | 1172 | return 1; |
1da177e4 LT |
1173 | } |
1174 | ||
4db7548c MG |
1175 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1176 | { | |
1177 | int ret = 1; | |
1178 | ||
1179 | /* | |
1180 | * We rely page->lru.next never has bit 0 set, unless the page | |
1181 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1182 | */ | |
1183 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1184 | ||
1185 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1186 | ret = 0; | |
1187 | goto out; | |
1188 | } | |
1189 | switch (page - head_page) { | |
1190 | case 1: | |
4da1984e | 1191 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c | 1192 | if (unlikely(compound_mapcount(page))) { |
82a3241a | 1193 | bad_page(page, "nonzero compound_mapcount"); |
4db7548c MG |
1194 | goto out; |
1195 | } | |
1196 | break; | |
1197 | case 2: | |
1198 | /* | |
1199 | * the second tail page: ->mapping is | |
fa3015b7 | 1200 | * deferred_list.next -- ignore value. |
4db7548c MG |
1201 | */ |
1202 | break; | |
1203 | default: | |
1204 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 1205 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
1206 | goto out; |
1207 | } | |
1208 | break; | |
1209 | } | |
1210 | if (unlikely(!PageTail(page))) { | |
82a3241a | 1211 | bad_page(page, "PageTail not set"); |
4db7548c MG |
1212 | goto out; |
1213 | } | |
1214 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 1215 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
1216 | goto out; |
1217 | } | |
1218 | ret = 0; | |
1219 | out: | |
1220 | page->mapping = NULL; | |
1221 | clear_compound_head(page); | |
1222 | return ret; | |
1223 | } | |
1224 | ||
6471384a AP |
1225 | static void kernel_init_free_pages(struct page *page, int numpages) |
1226 | { | |
1227 | int i; | |
1228 | ||
9e15afa5 QC |
1229 | /* s390's use of memset() could override KASAN redzones. */ |
1230 | kasan_disable_current(); | |
aa1ef4d7 | 1231 | for (i = 0; i < numpages; i++) { |
acb35b17 | 1232 | u8 tag = page_kasan_tag(page + i); |
aa1ef4d7 | 1233 | page_kasan_tag_reset(page + i); |
6471384a | 1234 | clear_highpage(page + i); |
acb35b17 | 1235 | page_kasan_tag_set(page + i, tag); |
aa1ef4d7 | 1236 | } |
9e15afa5 | 1237 | kasan_enable_current(); |
6471384a AP |
1238 | } |
1239 | ||
e2769dbd | 1240 | static __always_inline bool free_pages_prepare(struct page *page, |
2c335680 | 1241 | unsigned int order, bool check_free, fpi_t fpi_flags) |
4db7548c | 1242 | { |
e2769dbd | 1243 | int bad = 0; |
1bb5eab3 | 1244 | bool init; |
4db7548c | 1245 | |
4db7548c MG |
1246 | VM_BUG_ON_PAGE(PageTail(page), page); |
1247 | ||
e2769dbd | 1248 | trace_mm_page_free(page, order); |
e2769dbd | 1249 | |
79f5f8fa OS |
1250 | if (unlikely(PageHWPoison(page)) && !order) { |
1251 | /* | |
1252 | * Do not let hwpoison pages hit pcplists/buddy | |
1253 | * Untie memcg state and reset page's owner | |
1254 | */ | |
18b2db3b | 1255 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
79f5f8fa OS |
1256 | __memcg_kmem_uncharge_page(page, order); |
1257 | reset_page_owner(page, order); | |
1258 | return false; | |
1259 | } | |
1260 | ||
e2769dbd MG |
1261 | /* |
1262 | * Check tail pages before head page information is cleared to | |
1263 | * avoid checking PageCompound for order-0 pages. | |
1264 | */ | |
1265 | if (unlikely(order)) { | |
1266 | bool compound = PageCompound(page); | |
1267 | int i; | |
1268 | ||
1269 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1270 | |
9a73f61b KS |
1271 | if (compound) |
1272 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1273 | for (i = 1; i < (1 << order); i++) { |
1274 | if (compound) | |
1275 | bad += free_tail_pages_check(page, page + i); | |
534fe5e3 | 1276 | if (unlikely(check_free_page(page + i))) { |
e2769dbd MG |
1277 | bad++; |
1278 | continue; | |
1279 | } | |
1280 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1281 | } | |
1282 | } | |
bda807d4 | 1283 | if (PageMappingFlags(page)) |
4db7548c | 1284 | page->mapping = NULL; |
18b2db3b | 1285 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
f4b00eab | 1286 | __memcg_kmem_uncharge_page(page, order); |
e2769dbd | 1287 | if (check_free) |
534fe5e3 | 1288 | bad += check_free_page(page); |
e2769dbd MG |
1289 | if (bad) |
1290 | return false; | |
4db7548c | 1291 | |
e2769dbd MG |
1292 | page_cpupid_reset_last(page); |
1293 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1294 | reset_page_owner(page, order); | |
4db7548c MG |
1295 | |
1296 | if (!PageHighMem(page)) { | |
1297 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1298 | PAGE_SIZE << order); |
4db7548c | 1299 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1300 | PAGE_SIZE << order); |
4db7548c | 1301 | } |
6471384a | 1302 | |
8db26a3d VB |
1303 | kernel_poison_pages(page, 1 << order); |
1304 | ||
f9d79e8d | 1305 | /* |
1bb5eab3 AK |
1306 | * As memory initialization might be integrated into KASAN, |
1307 | * kasan_free_pages and kernel_init_free_pages must be | |
1308 | * kept together to avoid discrepancies in behavior. | |
1309 | * | |
f9d79e8d AK |
1310 | * With hardware tag-based KASAN, memory tags must be set before the |
1311 | * page becomes unavailable via debug_pagealloc or arch_free_page. | |
1312 | */ | |
1bb5eab3 AK |
1313 | init = want_init_on_free(); |
1314 | if (init && !kasan_has_integrated_init()) | |
1315 | kernel_init_free_pages(page, 1 << order); | |
1316 | kasan_free_nondeferred_pages(page, order, init, fpi_flags); | |
f9d79e8d | 1317 | |
234fdce8 QC |
1318 | /* |
1319 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1320 | * does this. So nothing which can access the page's contents should | |
1321 | * happen after this. | |
1322 | */ | |
1323 | arch_free_page(page, order); | |
1324 | ||
77bc7fd6 | 1325 | debug_pagealloc_unmap_pages(page, 1 << order); |
d6332692 | 1326 | |
4db7548c MG |
1327 | return true; |
1328 | } | |
1329 | ||
e2769dbd | 1330 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1331 | /* |
1332 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1333 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1334 | * moved from pcp lists to free lists. | |
1335 | */ | |
1336 | static bool free_pcp_prepare(struct page *page) | |
e2769dbd | 1337 | { |
2c335680 | 1338 | return free_pages_prepare(page, 0, true, FPI_NONE); |
e2769dbd MG |
1339 | } |
1340 | ||
4462b32c | 1341 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1342 | { |
8e57f8ac | 1343 | if (debug_pagealloc_enabled_static()) |
534fe5e3 | 1344 | return check_free_page(page); |
4462b32c VB |
1345 | else |
1346 | return false; | |
e2769dbd MG |
1347 | } |
1348 | #else | |
4462b32c VB |
1349 | /* |
1350 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1351 | * moving from pcp lists to free list in order to reduce overhead. With | |
1352 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1353 | * to the pcp lists. | |
1354 | */ | |
e2769dbd MG |
1355 | static bool free_pcp_prepare(struct page *page) |
1356 | { | |
8e57f8ac | 1357 | if (debug_pagealloc_enabled_static()) |
2c335680 | 1358 | return free_pages_prepare(page, 0, true, FPI_NONE); |
4462b32c | 1359 | else |
2c335680 | 1360 | return free_pages_prepare(page, 0, false, FPI_NONE); |
e2769dbd MG |
1361 | } |
1362 | ||
4db7548c MG |
1363 | static bool bulkfree_pcp_prepare(struct page *page) |
1364 | { | |
534fe5e3 | 1365 | return check_free_page(page); |
4db7548c MG |
1366 | } |
1367 | #endif /* CONFIG_DEBUG_VM */ | |
1368 | ||
97334162 AL |
1369 | static inline void prefetch_buddy(struct page *page) |
1370 | { | |
1371 | unsigned long pfn = page_to_pfn(page); | |
1372 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1373 | struct page *buddy = page + (buddy_pfn - pfn); | |
1374 | ||
1375 | prefetch(buddy); | |
1376 | } | |
1377 | ||
1da177e4 | 1378 | /* |
5f8dcc21 | 1379 | * Frees a number of pages from the PCP lists |
1da177e4 | 1380 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1381 | * count is the number of pages to free. |
1da177e4 LT |
1382 | * |
1383 | * If the zone was previously in an "all pages pinned" state then look to | |
1384 | * see if this freeing clears that state. | |
1385 | * | |
1386 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1387 | * pinned" detection logic. | |
1388 | */ | |
5f8dcc21 MG |
1389 | static void free_pcppages_bulk(struct zone *zone, int count, |
1390 | struct per_cpu_pages *pcp) | |
1da177e4 | 1391 | { |
5f8dcc21 | 1392 | int migratetype = 0; |
a6f9edd6 | 1393 | int batch_free = 0; |
5c3ad2eb | 1394 | int prefetch_nr = READ_ONCE(pcp->batch); |
3777999d | 1395 | bool isolated_pageblocks; |
0a5f4e5b AL |
1396 | struct page *page, *tmp; |
1397 | LIST_HEAD(head); | |
f2260e6b | 1398 | |
88e8ac11 CTR |
1399 | /* |
1400 | * Ensure proper count is passed which otherwise would stuck in the | |
1401 | * below while (list_empty(list)) loop. | |
1402 | */ | |
1403 | count = min(pcp->count, count); | |
e5b31ac2 | 1404 | while (count) { |
5f8dcc21 MG |
1405 | struct list_head *list; |
1406 | ||
1407 | /* | |
a6f9edd6 MG |
1408 | * Remove pages from lists in a round-robin fashion. A |
1409 | * batch_free count is maintained that is incremented when an | |
1410 | * empty list is encountered. This is so more pages are freed | |
1411 | * off fuller lists instead of spinning excessively around empty | |
1412 | * lists | |
5f8dcc21 MG |
1413 | */ |
1414 | do { | |
a6f9edd6 | 1415 | batch_free++; |
5f8dcc21 MG |
1416 | if (++migratetype == MIGRATE_PCPTYPES) |
1417 | migratetype = 0; | |
1418 | list = &pcp->lists[migratetype]; | |
1419 | } while (list_empty(list)); | |
48db57f8 | 1420 | |
1d16871d NK |
1421 | /* This is the only non-empty list. Free them all. */ |
1422 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1423 | batch_free = count; |
1d16871d | 1424 | |
a6f9edd6 | 1425 | do { |
a16601c5 | 1426 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1427 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1428 | list_del(&page->lru); |
77ba9062 | 1429 | pcp->count--; |
aa016d14 | 1430 | |
4db7548c MG |
1431 | if (bulkfree_pcp_prepare(page)) |
1432 | continue; | |
1433 | ||
0a5f4e5b | 1434 | list_add_tail(&page->lru, &head); |
97334162 AL |
1435 | |
1436 | /* | |
1437 | * We are going to put the page back to the global | |
1438 | * pool, prefetch its buddy to speed up later access | |
1439 | * under zone->lock. It is believed the overhead of | |
1440 | * an additional test and calculating buddy_pfn here | |
1441 | * can be offset by reduced memory latency later. To | |
1442 | * avoid excessive prefetching due to large count, only | |
1443 | * prefetch buddy for the first pcp->batch nr of pages. | |
1444 | */ | |
5c3ad2eb | 1445 | if (prefetch_nr) { |
97334162 | 1446 | prefetch_buddy(page); |
5c3ad2eb VB |
1447 | prefetch_nr--; |
1448 | } | |
e5b31ac2 | 1449 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1450 | } |
0a5f4e5b AL |
1451 | |
1452 | spin_lock(&zone->lock); | |
1453 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1454 | ||
1455 | /* | |
1456 | * Use safe version since after __free_one_page(), | |
1457 | * page->lru.next will not point to original list. | |
1458 | */ | |
1459 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1460 | int mt = get_pcppage_migratetype(page); | |
1461 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1462 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1463 | /* Pageblock could have been isolated meanwhile */ | |
1464 | if (unlikely(isolated_pageblocks)) | |
1465 | mt = get_pageblock_migratetype(page); | |
1466 | ||
f04a5d5d | 1467 | __free_one_page(page, page_to_pfn(page), zone, 0, mt, FPI_NONE); |
0a5f4e5b AL |
1468 | trace_mm_page_pcpu_drain(page, 0, mt); |
1469 | } | |
d34b0733 | 1470 | spin_unlock(&zone->lock); |
1da177e4 LT |
1471 | } |
1472 | ||
dc4b0caf MG |
1473 | static void free_one_page(struct zone *zone, |
1474 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1475 | unsigned int order, |
7fef431b | 1476 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1477 | { |
d34b0733 | 1478 | spin_lock(&zone->lock); |
ad53f92e JK |
1479 | if (unlikely(has_isolate_pageblock(zone) || |
1480 | is_migrate_isolate(migratetype))) { | |
1481 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1482 | } |
7fef431b | 1483 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); |
d34b0733 | 1484 | spin_unlock(&zone->lock); |
48db57f8 NP |
1485 | } |
1486 | ||
1e8ce83c | 1487 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1488 | unsigned long zone, int nid) |
1e8ce83c | 1489 | { |
d0dc12e8 | 1490 | mm_zero_struct_page(page); |
1e8ce83c | 1491 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1492 | init_page_count(page); |
1493 | page_mapcount_reset(page); | |
1494 | page_cpupid_reset_last(page); | |
2813b9c0 | 1495 | page_kasan_tag_reset(page); |
1e8ce83c | 1496 | |
1e8ce83c RH |
1497 | INIT_LIST_HEAD(&page->lru); |
1498 | #ifdef WANT_PAGE_VIRTUAL | |
1499 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1500 | if (!is_highmem_idx(zone)) | |
1501 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1502 | #endif | |
1503 | } | |
1504 | ||
7e18adb4 | 1505 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1506 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1507 | { |
1508 | pg_data_t *pgdat; | |
1509 | int nid, zid; | |
1510 | ||
1511 | if (!early_page_uninitialised(pfn)) | |
1512 | return; | |
1513 | ||
1514 | nid = early_pfn_to_nid(pfn); | |
1515 | pgdat = NODE_DATA(nid); | |
1516 | ||
1517 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1518 | struct zone *zone = &pgdat->node_zones[zid]; | |
1519 | ||
1520 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1521 | break; | |
1522 | } | |
d0dc12e8 | 1523 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1524 | } |
1525 | #else | |
1526 | static inline void init_reserved_page(unsigned long pfn) | |
1527 | { | |
1528 | } | |
1529 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1530 | ||
92923ca3 NZ |
1531 | /* |
1532 | * Initialised pages do not have PageReserved set. This function is | |
1533 | * called for each range allocated by the bootmem allocator and | |
1534 | * marks the pages PageReserved. The remaining valid pages are later | |
1535 | * sent to the buddy page allocator. | |
1536 | */ | |
4b50bcc7 | 1537 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1538 | { |
1539 | unsigned long start_pfn = PFN_DOWN(start); | |
1540 | unsigned long end_pfn = PFN_UP(end); | |
1541 | ||
7e18adb4 MG |
1542 | for (; start_pfn < end_pfn; start_pfn++) { |
1543 | if (pfn_valid(start_pfn)) { | |
1544 | struct page *page = pfn_to_page(start_pfn); | |
1545 | ||
1546 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1547 | |
1548 | /* Avoid false-positive PageTail() */ | |
1549 | INIT_LIST_HEAD(&page->lru); | |
1550 | ||
d483da5b AD |
1551 | /* |
1552 | * no need for atomic set_bit because the struct | |
1553 | * page is not visible yet so nobody should | |
1554 | * access it yet. | |
1555 | */ | |
1556 | __SetPageReserved(page); | |
7e18adb4 MG |
1557 | } |
1558 | } | |
92923ca3 NZ |
1559 | } |
1560 | ||
7fef431b DH |
1561 | static void __free_pages_ok(struct page *page, unsigned int order, |
1562 | fpi_t fpi_flags) | |
ec95f53a | 1563 | { |
d34b0733 | 1564 | unsigned long flags; |
95e34412 | 1565 | int migratetype; |
dc4b0caf | 1566 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1567 | |
2c335680 | 1568 | if (!free_pages_prepare(page, order, true, fpi_flags)) |
ec95f53a KM |
1569 | return; |
1570 | ||
cfc47a28 | 1571 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1572 | local_irq_save(flags); |
1573 | __count_vm_events(PGFREE, 1 << order); | |
7fef431b DH |
1574 | free_one_page(page_zone(page), page, pfn, order, migratetype, |
1575 | fpi_flags); | |
d34b0733 | 1576 | local_irq_restore(flags); |
1da177e4 LT |
1577 | } |
1578 | ||
a9cd410a | 1579 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1580 | { |
c3993076 | 1581 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1582 | struct page *p = page; |
c3993076 | 1583 | unsigned int loop; |
a226f6c8 | 1584 | |
7fef431b DH |
1585 | /* |
1586 | * When initializing the memmap, __init_single_page() sets the refcount | |
1587 | * of all pages to 1 ("allocated"/"not free"). We have to set the | |
1588 | * refcount of all involved pages to 0. | |
1589 | */ | |
e2d0bd2b YL |
1590 | prefetchw(p); |
1591 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1592 | prefetchw(p + 1); | |
c3993076 JW |
1593 | __ClearPageReserved(p); |
1594 | set_page_count(p, 0); | |
a226f6c8 | 1595 | } |
e2d0bd2b YL |
1596 | __ClearPageReserved(p); |
1597 | set_page_count(p, 0); | |
c3993076 | 1598 | |
9705bea5 | 1599 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
7fef431b DH |
1600 | |
1601 | /* | |
1602 | * Bypass PCP and place fresh pages right to the tail, primarily | |
1603 | * relevant for memory onlining. | |
1604 | */ | |
2c335680 | 1605 | __free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON); |
a226f6c8 DH |
1606 | } |
1607 | ||
3f08a302 | 1608 | #ifdef CONFIG_NEED_MULTIPLE_NODES |
7ace9917 | 1609 | |
03e92a5e MR |
1610 | /* |
1611 | * During memory init memblocks map pfns to nids. The search is expensive and | |
1612 | * this caches recent lookups. The implementation of __early_pfn_to_nid | |
1613 | * treats start/end as pfns. | |
1614 | */ | |
1615 | struct mminit_pfnnid_cache { | |
1616 | unsigned long last_start; | |
1617 | unsigned long last_end; | |
1618 | int last_nid; | |
1619 | }; | |
75a592a4 | 1620 | |
03e92a5e | 1621 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
6f24fbd3 MR |
1622 | |
1623 | /* | |
1624 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
1625 | */ | |
03e92a5e | 1626 | static int __meminit __early_pfn_to_nid(unsigned long pfn, |
6f24fbd3 | 1627 | struct mminit_pfnnid_cache *state) |
75a592a4 | 1628 | { |
6f24fbd3 | 1629 | unsigned long start_pfn, end_pfn; |
75a592a4 MG |
1630 | int nid; |
1631 | ||
6f24fbd3 MR |
1632 | if (state->last_start <= pfn && pfn < state->last_end) |
1633 | return state->last_nid; | |
1634 | ||
1635 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); | |
1636 | if (nid != NUMA_NO_NODE) { | |
1637 | state->last_start = start_pfn; | |
1638 | state->last_end = end_pfn; | |
1639 | state->last_nid = nid; | |
1640 | } | |
7ace9917 MG |
1641 | |
1642 | return nid; | |
75a592a4 | 1643 | } |
75a592a4 | 1644 | |
75a592a4 | 1645 | int __meminit early_pfn_to_nid(unsigned long pfn) |
75a592a4 | 1646 | { |
7ace9917 | 1647 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1648 | int nid; |
1649 | ||
7ace9917 | 1650 | spin_lock(&early_pfn_lock); |
56ec43d8 | 1651 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1652 | if (nid < 0) |
e4568d38 | 1653 | nid = first_online_node; |
7ace9917 | 1654 | spin_unlock(&early_pfn_lock); |
75a592a4 | 1655 | |
7ace9917 | 1656 | return nid; |
75a592a4 | 1657 | } |
3f08a302 | 1658 | #endif /* CONFIG_NEED_MULTIPLE_NODES */ |
75a592a4 | 1659 | |
7c2ee349 | 1660 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1661 | unsigned int order) |
1662 | { | |
1663 | if (early_page_uninitialised(pfn)) | |
1664 | return; | |
a9cd410a | 1665 | __free_pages_core(page, order); |
3a80a7fa MG |
1666 | } |
1667 | ||
7cf91a98 JK |
1668 | /* |
1669 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1670 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1671 | * with the migration of free compaction scanner. The scanners then need to | |
1672 | * use only pfn_valid_within() check for arches that allow holes within | |
1673 | * pageblocks. | |
1674 | * | |
1675 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1676 | * | |
1677 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1678 | * i.e. it's possible that all pages within a zones range of pages do not | |
1679 | * belong to a single zone. We assume that a border between node0 and node1 | |
1680 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1681 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1682 | * the first and last page of a pageblock and avoid checking each individual | |
1683 | * page in a pageblock. | |
1684 | */ | |
1685 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1686 | unsigned long end_pfn, struct zone *zone) | |
1687 | { | |
1688 | struct page *start_page; | |
1689 | struct page *end_page; | |
1690 | ||
1691 | /* end_pfn is one past the range we are checking */ | |
1692 | end_pfn--; | |
1693 | ||
1694 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1695 | return NULL; | |
1696 | ||
2d070eab MH |
1697 | start_page = pfn_to_online_page(start_pfn); |
1698 | if (!start_page) | |
1699 | return NULL; | |
7cf91a98 JK |
1700 | |
1701 | if (page_zone(start_page) != zone) | |
1702 | return NULL; | |
1703 | ||
1704 | end_page = pfn_to_page(end_pfn); | |
1705 | ||
1706 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1707 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1708 | return NULL; | |
1709 | ||
1710 | return start_page; | |
1711 | } | |
1712 | ||
1713 | void set_zone_contiguous(struct zone *zone) | |
1714 | { | |
1715 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1716 | unsigned long block_end_pfn; | |
1717 | ||
1718 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1719 | for (; block_start_pfn < zone_end_pfn(zone); | |
1720 | block_start_pfn = block_end_pfn, | |
1721 | block_end_pfn += pageblock_nr_pages) { | |
1722 | ||
1723 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1724 | ||
1725 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1726 | block_end_pfn, zone)) | |
1727 | return; | |
e84fe99b | 1728 | cond_resched(); |
7cf91a98 JK |
1729 | } |
1730 | ||
1731 | /* We confirm that there is no hole */ | |
1732 | zone->contiguous = true; | |
1733 | } | |
1734 | ||
1735 | void clear_zone_contiguous(struct zone *zone) | |
1736 | { | |
1737 | zone->contiguous = false; | |
1738 | } | |
1739 | ||
7e18adb4 | 1740 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1741 | static void __init deferred_free_range(unsigned long pfn, |
1742 | unsigned long nr_pages) | |
a4de83dd | 1743 | { |
2f47a91f PT |
1744 | struct page *page; |
1745 | unsigned long i; | |
a4de83dd | 1746 | |
2f47a91f | 1747 | if (!nr_pages) |
a4de83dd MG |
1748 | return; |
1749 | ||
2f47a91f PT |
1750 | page = pfn_to_page(pfn); |
1751 | ||
a4de83dd | 1752 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1753 | if (nr_pages == pageblock_nr_pages && |
1754 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1755 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1756 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1757 | return; |
1758 | } | |
1759 | ||
e780149b XQ |
1760 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1761 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1762 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
a9cd410a | 1763 | __free_pages_core(page, 0); |
e780149b | 1764 | } |
a4de83dd MG |
1765 | } |
1766 | ||
d3cd131d NS |
1767 | /* Completion tracking for deferred_init_memmap() threads */ |
1768 | static atomic_t pgdat_init_n_undone __initdata; | |
1769 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1770 | ||
1771 | static inline void __init pgdat_init_report_one_done(void) | |
1772 | { | |
1773 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1774 | complete(&pgdat_init_all_done_comp); | |
1775 | } | |
0e1cc95b | 1776 | |
2f47a91f | 1777 | /* |
80b1f41c PT |
1778 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1779 | * | |
1780 | * First we check if pfn is valid on architectures where it is possible to have | |
1781 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1782 | * function is optimized out. | |
1783 | * | |
1784 | * Then, we check if a current large page is valid by only checking the validity | |
1785 | * of the head pfn. | |
2f47a91f | 1786 | */ |
56ec43d8 | 1787 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1788 | { |
80b1f41c PT |
1789 | if (!pfn_valid_within(pfn)) |
1790 | return false; | |
1791 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1792 | return false; | |
80b1f41c PT |
1793 | return true; |
1794 | } | |
2f47a91f | 1795 | |
80b1f41c PT |
1796 | /* |
1797 | * Free pages to buddy allocator. Try to free aligned pages in | |
1798 | * pageblock_nr_pages sizes. | |
1799 | */ | |
56ec43d8 | 1800 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1801 | unsigned long end_pfn) |
1802 | { | |
80b1f41c PT |
1803 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
1804 | unsigned long nr_free = 0; | |
2f47a91f | 1805 | |
80b1f41c | 1806 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1807 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1808 | deferred_free_range(pfn - nr_free, nr_free); |
1809 | nr_free = 0; | |
1810 | } else if (!(pfn & nr_pgmask)) { | |
1811 | deferred_free_range(pfn - nr_free, nr_free); | |
1812 | nr_free = 1; | |
80b1f41c PT |
1813 | } else { |
1814 | nr_free++; | |
1815 | } | |
1816 | } | |
1817 | /* Free the last block of pages to allocator */ | |
1818 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1819 | } |
1820 | ||
80b1f41c PT |
1821 | /* |
1822 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1823 | * by performing it only once every pageblock_nr_pages. | |
1824 | * Return number of pages initialized. | |
1825 | */ | |
56ec43d8 | 1826 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1827 | unsigned long pfn, |
1828 | unsigned long end_pfn) | |
2f47a91f | 1829 | { |
2f47a91f | 1830 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
56ec43d8 | 1831 | int nid = zone_to_nid(zone); |
2f47a91f | 1832 | unsigned long nr_pages = 0; |
56ec43d8 | 1833 | int zid = zone_idx(zone); |
2f47a91f | 1834 | struct page *page = NULL; |
2f47a91f | 1835 | |
80b1f41c | 1836 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1837 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1838 | page = NULL; |
2f47a91f | 1839 | continue; |
80b1f41c | 1840 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1841 | page = pfn_to_page(pfn); |
80b1f41c PT |
1842 | } else { |
1843 | page++; | |
2f47a91f | 1844 | } |
d0dc12e8 | 1845 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1846 | nr_pages++; |
2f47a91f | 1847 | } |
80b1f41c | 1848 | return (nr_pages); |
2f47a91f PT |
1849 | } |
1850 | ||
0e56acae AD |
1851 | /* |
1852 | * This function is meant to pre-load the iterator for the zone init. | |
1853 | * Specifically it walks through the ranges until we are caught up to the | |
1854 | * first_init_pfn value and exits there. If we never encounter the value we | |
1855 | * return false indicating there are no valid ranges left. | |
1856 | */ | |
1857 | static bool __init | |
1858 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
1859 | unsigned long *spfn, unsigned long *epfn, | |
1860 | unsigned long first_init_pfn) | |
1861 | { | |
1862 | u64 j; | |
1863 | ||
1864 | /* | |
1865 | * Start out by walking through the ranges in this zone that have | |
1866 | * already been initialized. We don't need to do anything with them | |
1867 | * so we just need to flush them out of the system. | |
1868 | */ | |
1869 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
1870 | if (*epfn <= first_init_pfn) | |
1871 | continue; | |
1872 | if (*spfn < first_init_pfn) | |
1873 | *spfn = first_init_pfn; | |
1874 | *i = j; | |
1875 | return true; | |
1876 | } | |
1877 | ||
1878 | return false; | |
1879 | } | |
1880 | ||
1881 | /* | |
1882 | * Initialize and free pages. We do it in two loops: first we initialize | |
1883 | * struct page, then free to buddy allocator, because while we are | |
1884 | * freeing pages we can access pages that are ahead (computing buddy | |
1885 | * page in __free_one_page()). | |
1886 | * | |
1887 | * In order to try and keep some memory in the cache we have the loop | |
1888 | * broken along max page order boundaries. This way we will not cause | |
1889 | * any issues with the buddy page computation. | |
1890 | */ | |
1891 | static unsigned long __init | |
1892 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
1893 | unsigned long *end_pfn) | |
1894 | { | |
1895 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
1896 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
1897 | unsigned long nr_pages = 0; | |
1898 | u64 j = *i; | |
1899 | ||
1900 | /* First we loop through and initialize the page values */ | |
1901 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
1902 | unsigned long t; | |
1903 | ||
1904 | if (mo_pfn <= *start_pfn) | |
1905 | break; | |
1906 | ||
1907 | t = min(mo_pfn, *end_pfn); | |
1908 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
1909 | ||
1910 | if (mo_pfn < *end_pfn) { | |
1911 | *start_pfn = mo_pfn; | |
1912 | break; | |
1913 | } | |
1914 | } | |
1915 | ||
1916 | /* Reset values and now loop through freeing pages as needed */ | |
1917 | swap(j, *i); | |
1918 | ||
1919 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
1920 | unsigned long t; | |
1921 | ||
1922 | if (mo_pfn <= spfn) | |
1923 | break; | |
1924 | ||
1925 | t = min(mo_pfn, epfn); | |
1926 | deferred_free_pages(spfn, t); | |
1927 | ||
1928 | if (mo_pfn <= epfn) | |
1929 | break; | |
1930 | } | |
1931 | ||
1932 | return nr_pages; | |
1933 | } | |
1934 | ||
e4443149 DJ |
1935 | static void __init |
1936 | deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, | |
1937 | void *arg) | |
1938 | { | |
1939 | unsigned long spfn, epfn; | |
1940 | struct zone *zone = arg; | |
1941 | u64 i; | |
1942 | ||
1943 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn); | |
1944 | ||
1945 | /* | |
1946 | * Initialize and free pages in MAX_ORDER sized increments so that we | |
1947 | * can avoid introducing any issues with the buddy allocator. | |
1948 | */ | |
1949 | while (spfn < end_pfn) { | |
1950 | deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
1951 | cond_resched(); | |
1952 | } | |
1953 | } | |
1954 | ||
ecd09650 DJ |
1955 | /* An arch may override for more concurrency. */ |
1956 | __weak int __init | |
1957 | deferred_page_init_max_threads(const struct cpumask *node_cpumask) | |
1958 | { | |
1959 | return 1; | |
1960 | } | |
1961 | ||
7e18adb4 | 1962 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1963 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1964 | { |
0e1cc95b | 1965 | pg_data_t *pgdat = data; |
0e56acae | 1966 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
89c7c402 | 1967 | unsigned long spfn = 0, epfn = 0; |
0e56acae | 1968 | unsigned long first_init_pfn, flags; |
7e18adb4 | 1969 | unsigned long start = jiffies; |
7e18adb4 | 1970 | struct zone *zone; |
e4443149 | 1971 | int zid, max_threads; |
2f47a91f | 1972 | u64 i; |
7e18adb4 | 1973 | |
3a2d7fa8 PT |
1974 | /* Bind memory initialisation thread to a local node if possible */ |
1975 | if (!cpumask_empty(cpumask)) | |
1976 | set_cpus_allowed_ptr(current, cpumask); | |
1977 | ||
1978 | pgdat_resize_lock(pgdat, &flags); | |
1979 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1980 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1981 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1982 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1983 | return 0; |
1984 | } | |
1985 | ||
7e18adb4 MG |
1986 | /* Sanity check boundaries */ |
1987 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1988 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1989 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1990 | ||
3d060856 PT |
1991 | /* |
1992 | * Once we unlock here, the zone cannot be grown anymore, thus if an | |
1993 | * interrupt thread must allocate this early in boot, zone must be | |
1994 | * pre-grown prior to start of deferred page initialization. | |
1995 | */ | |
1996 | pgdat_resize_unlock(pgdat, &flags); | |
1997 | ||
7e18adb4 MG |
1998 | /* Only the highest zone is deferred so find it */ |
1999 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
2000 | zone = pgdat->node_zones + zid; | |
2001 | if (first_init_pfn < zone_end_pfn(zone)) | |
2002 | break; | |
2003 | } | |
0e56acae AD |
2004 | |
2005 | /* If the zone is empty somebody else may have cleared out the zone */ | |
2006 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2007 | first_init_pfn)) | |
2008 | goto zone_empty; | |
7e18adb4 | 2009 | |
ecd09650 | 2010 | max_threads = deferred_page_init_max_threads(cpumask); |
7e18adb4 | 2011 | |
117003c3 | 2012 | while (spfn < epfn) { |
e4443149 DJ |
2013 | unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION); |
2014 | struct padata_mt_job job = { | |
2015 | .thread_fn = deferred_init_memmap_chunk, | |
2016 | .fn_arg = zone, | |
2017 | .start = spfn, | |
2018 | .size = epfn_align - spfn, | |
2019 | .align = PAGES_PER_SECTION, | |
2020 | .min_chunk = PAGES_PER_SECTION, | |
2021 | .max_threads = max_threads, | |
2022 | }; | |
2023 | ||
2024 | padata_do_multithreaded(&job); | |
2025 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2026 | epfn_align); | |
117003c3 | 2027 | } |
0e56acae | 2028 | zone_empty: |
7e18adb4 MG |
2029 | /* Sanity check that the next zone really is unpopulated */ |
2030 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
2031 | ||
89c7c402 DJ |
2032 | pr_info("node %d deferred pages initialised in %ums\n", |
2033 | pgdat->node_id, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
2034 | |
2035 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
2036 | return 0; |
2037 | } | |
c9e97a19 | 2038 | |
c9e97a19 PT |
2039 | /* |
2040 | * If this zone has deferred pages, try to grow it by initializing enough | |
2041 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
2042 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
2043 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
2044 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
2045 | * | |
2046 | * Return true when zone was grown, otherwise return false. We return true even | |
2047 | * when we grow less than requested, to let the caller decide if there are | |
2048 | * enough pages to satisfy the allocation. | |
2049 | * | |
2050 | * Note: We use noinline because this function is needed only during boot, and | |
2051 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
2052 | * making sure that it is not inlined into permanent text section. | |
2053 | */ | |
2054 | static noinline bool __init | |
2055 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
2056 | { | |
c9e97a19 | 2057 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 2058 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 2059 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
2060 | unsigned long spfn, epfn, flags; |
2061 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
2062 | u64 i; |
2063 | ||
2064 | /* Only the last zone may have deferred pages */ | |
2065 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
2066 | return false; | |
2067 | ||
2068 | pgdat_resize_lock(pgdat, &flags); | |
2069 | ||
c9e97a19 PT |
2070 | /* |
2071 | * If someone grew this zone while we were waiting for spinlock, return | |
2072 | * true, as there might be enough pages already. | |
2073 | */ | |
2074 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
2075 | pgdat_resize_unlock(pgdat, &flags); | |
2076 | return true; | |
2077 | } | |
2078 | ||
0e56acae AD |
2079 | /* If the zone is empty somebody else may have cleared out the zone */ |
2080 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2081 | first_deferred_pfn)) { | |
2082 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 2083 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
2084 | /* Retry only once. */ |
2085 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
2086 | } |
2087 | ||
0e56acae AD |
2088 | /* |
2089 | * Initialize and free pages in MAX_ORDER sized increments so | |
2090 | * that we can avoid introducing any issues with the buddy | |
2091 | * allocator. | |
2092 | */ | |
2093 | while (spfn < epfn) { | |
2094 | /* update our first deferred PFN for this section */ | |
2095 | first_deferred_pfn = spfn; | |
2096 | ||
2097 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
117003c3 | 2098 | touch_nmi_watchdog(); |
c9e97a19 | 2099 | |
0e56acae AD |
2100 | /* We should only stop along section boundaries */ |
2101 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
2102 | continue; | |
c9e97a19 | 2103 | |
0e56acae | 2104 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
2105 | if (nr_pages >= nr_pages_needed) |
2106 | break; | |
2107 | } | |
2108 | ||
0e56acae | 2109 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
2110 | pgdat_resize_unlock(pgdat, &flags); |
2111 | ||
2112 | return nr_pages > 0; | |
2113 | } | |
2114 | ||
2115 | /* | |
2116 | * deferred_grow_zone() is __init, but it is called from | |
2117 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
2118 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
2119 | * and to ensure that the function body gets unloaded. | |
2120 | */ | |
2121 | static bool __ref | |
2122 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
2123 | { | |
2124 | return deferred_grow_zone(zone, order); | |
2125 | } | |
2126 | ||
7cf91a98 | 2127 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
2128 | |
2129 | void __init page_alloc_init_late(void) | |
2130 | { | |
7cf91a98 | 2131 | struct zone *zone; |
e900a918 | 2132 | int nid; |
7cf91a98 JK |
2133 | |
2134 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 2135 | |
d3cd131d NS |
2136 | /* There will be num_node_state(N_MEMORY) threads */ |
2137 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 2138 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
2139 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
2140 | } | |
2141 | ||
2142 | /* Block until all are initialised */ | |
d3cd131d | 2143 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 2144 | |
3e8fc007 MG |
2145 | /* |
2146 | * The number of managed pages has changed due to the initialisation | |
2147 | * so the pcpu batch and high limits needs to be updated or the limits | |
2148 | * will be artificially small. | |
2149 | */ | |
2150 | for_each_populated_zone(zone) | |
2151 | zone_pcp_update(zone); | |
2152 | ||
c9e97a19 PT |
2153 | /* |
2154 | * We initialized the rest of the deferred pages. Permanently disable | |
2155 | * on-demand struct page initialization. | |
2156 | */ | |
2157 | static_branch_disable(&deferred_pages); | |
2158 | ||
4248b0da MG |
2159 | /* Reinit limits that are based on free pages after the kernel is up */ |
2160 | files_maxfiles_init(); | |
7cf91a98 | 2161 | #endif |
350e88ba | 2162 | |
ba8f3587 LF |
2163 | buffer_init(); |
2164 | ||
3010f876 PT |
2165 | /* Discard memblock private memory */ |
2166 | memblock_discard(); | |
7cf91a98 | 2167 | |
e900a918 DW |
2168 | for_each_node_state(nid, N_MEMORY) |
2169 | shuffle_free_memory(NODE_DATA(nid)); | |
2170 | ||
7cf91a98 JK |
2171 | for_each_populated_zone(zone) |
2172 | set_zone_contiguous(zone); | |
7e18adb4 | 2173 | } |
7e18adb4 | 2174 | |
47118af0 | 2175 | #ifdef CONFIG_CMA |
9cf510a5 | 2176 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
2177 | void __init init_cma_reserved_pageblock(struct page *page) |
2178 | { | |
2179 | unsigned i = pageblock_nr_pages; | |
2180 | struct page *p = page; | |
2181 | ||
2182 | do { | |
2183 | __ClearPageReserved(p); | |
2184 | set_page_count(p, 0); | |
d883c6cf | 2185 | } while (++p, --i); |
47118af0 | 2186 | |
47118af0 | 2187 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
2188 | |
2189 | if (pageblock_order >= MAX_ORDER) { | |
2190 | i = pageblock_nr_pages; | |
2191 | p = page; | |
2192 | do { | |
2193 | set_page_refcounted(p); | |
2194 | __free_pages(p, MAX_ORDER - 1); | |
2195 | p += MAX_ORDER_NR_PAGES; | |
2196 | } while (i -= MAX_ORDER_NR_PAGES); | |
2197 | } else { | |
2198 | set_page_refcounted(page); | |
2199 | __free_pages(page, pageblock_order); | |
2200 | } | |
2201 | ||
3dcc0571 | 2202 | adjust_managed_page_count(page, pageblock_nr_pages); |
3c381db1 | 2203 | page_zone(page)->cma_pages += pageblock_nr_pages; |
47118af0 MN |
2204 | } |
2205 | #endif | |
1da177e4 LT |
2206 | |
2207 | /* | |
2208 | * The order of subdivision here is critical for the IO subsystem. | |
2209 | * Please do not alter this order without good reasons and regression | |
2210 | * testing. Specifically, as large blocks of memory are subdivided, | |
2211 | * the order in which smaller blocks are delivered depends on the order | |
2212 | * they're subdivided in this function. This is the primary factor | |
2213 | * influencing the order in which pages are delivered to the IO | |
2214 | * subsystem according to empirical testing, and this is also justified | |
2215 | * by considering the behavior of a buddy system containing a single | |
2216 | * large block of memory acted on by a series of small allocations. | |
2217 | * This behavior is a critical factor in sglist merging's success. | |
2218 | * | |
6d49e352 | 2219 | * -- nyc |
1da177e4 | 2220 | */ |
085cc7d5 | 2221 | static inline void expand(struct zone *zone, struct page *page, |
6ab01363 | 2222 | int low, int high, int migratetype) |
1da177e4 LT |
2223 | { |
2224 | unsigned long size = 1 << high; | |
2225 | ||
2226 | while (high > low) { | |
1da177e4 LT |
2227 | high--; |
2228 | size >>= 1; | |
309381fe | 2229 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2230 | |
acbc15a4 JK |
2231 | /* |
2232 | * Mark as guard pages (or page), that will allow to | |
2233 | * merge back to allocator when buddy will be freed. | |
2234 | * Corresponding page table entries will not be touched, | |
2235 | * pages will stay not present in virtual address space | |
2236 | */ | |
2237 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2238 | continue; |
acbc15a4 | 2239 | |
6ab01363 | 2240 | add_to_free_list(&page[size], zone, high, migratetype); |
ab130f91 | 2241 | set_buddy_order(&page[size], high); |
1da177e4 | 2242 | } |
1da177e4 LT |
2243 | } |
2244 | ||
4e611801 | 2245 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2246 | { |
f4c18e6f | 2247 | if (unlikely(page->flags & __PG_HWPOISON)) { |
e570f56c NH |
2248 | /* Don't complain about hwpoisoned pages */ |
2249 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2250 | return; | |
f4c18e6f | 2251 | } |
58b7f119 WY |
2252 | |
2253 | bad_page(page, | |
2254 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
4e611801 VB |
2255 | } |
2256 | ||
2257 | /* | |
2258 | * This page is about to be returned from the page allocator | |
2259 | */ | |
2260 | static inline int check_new_page(struct page *page) | |
2261 | { | |
2262 | if (likely(page_expected_state(page, | |
2263 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2264 | return 0; | |
2265 | ||
2266 | check_new_page_bad(page); | |
2267 | return 1; | |
2a7684a2 WF |
2268 | } |
2269 | ||
479f854a | 2270 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2271 | /* |
2272 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2273 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2274 | * also checked when pcp lists are refilled from the free lists. | |
2275 | */ | |
2276 | static inline bool check_pcp_refill(struct page *page) | |
479f854a | 2277 | { |
8e57f8ac | 2278 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2279 | return check_new_page(page); |
2280 | else | |
2281 | return false; | |
479f854a MG |
2282 | } |
2283 | ||
4462b32c | 2284 | static inline bool check_new_pcp(struct page *page) |
479f854a MG |
2285 | { |
2286 | return check_new_page(page); | |
2287 | } | |
2288 | #else | |
4462b32c VB |
2289 | /* |
2290 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2291 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2292 | * enabled, they are also checked when being allocated from the pcp lists. | |
2293 | */ | |
2294 | static inline bool check_pcp_refill(struct page *page) | |
479f854a MG |
2295 | { |
2296 | return check_new_page(page); | |
2297 | } | |
4462b32c | 2298 | static inline bool check_new_pcp(struct page *page) |
479f854a | 2299 | { |
8e57f8ac | 2300 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2301 | return check_new_page(page); |
2302 | else | |
2303 | return false; | |
479f854a MG |
2304 | } |
2305 | #endif /* CONFIG_DEBUG_VM */ | |
2306 | ||
2307 | static bool check_new_pages(struct page *page, unsigned int order) | |
2308 | { | |
2309 | int i; | |
2310 | for (i = 0; i < (1 << order); i++) { | |
2311 | struct page *p = page + i; | |
2312 | ||
2313 | if (unlikely(check_new_page(p))) | |
2314 | return true; | |
2315 | } | |
2316 | ||
2317 | return false; | |
2318 | } | |
2319 | ||
46f24fd8 JK |
2320 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2321 | gfp_t gfp_flags) | |
2322 | { | |
1bb5eab3 AK |
2323 | bool init; |
2324 | ||
46f24fd8 JK |
2325 | set_page_private(page, 0); |
2326 | set_page_refcounted(page); | |
2327 | ||
2328 | arch_alloc_page(page, order); | |
77bc7fd6 | 2329 | debug_pagealloc_map_pages(page, 1 << order); |
1bb5eab3 AK |
2330 | |
2331 | /* | |
2332 | * Page unpoisoning must happen before memory initialization. | |
2333 | * Otherwise, the poison pattern will be overwritten for __GFP_ZERO | |
2334 | * allocations and the page unpoisoning code will complain. | |
2335 | */ | |
8db26a3d | 2336 | kernel_unpoison_pages(page, 1 << order); |
862b6dee | 2337 | |
1bb5eab3 AK |
2338 | /* |
2339 | * As memory initialization might be integrated into KASAN, | |
2340 | * kasan_alloc_pages and kernel_init_free_pages must be | |
2341 | * kept together to avoid discrepancies in behavior. | |
2342 | */ | |
2343 | init = !want_init_on_free() && want_init_on_alloc(gfp_flags); | |
2344 | kasan_alloc_pages(page, order, init); | |
2345 | if (init && !kasan_has_integrated_init()) | |
862b6dee | 2346 | kernel_init_free_pages(page, 1 << order); |
1bb5eab3 AK |
2347 | |
2348 | set_page_owner(page, order, gfp_flags); | |
46f24fd8 JK |
2349 | } |
2350 | ||
479f854a | 2351 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2352 | unsigned int alloc_flags) |
2a7684a2 | 2353 | { |
46f24fd8 | 2354 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2355 | |
17cf4406 NP |
2356 | if (order && (gfp_flags & __GFP_COMP)) |
2357 | prep_compound_page(page, order); | |
2358 | ||
75379191 | 2359 | /* |
2f064f34 | 2360 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2361 | * allocate the page. The expectation is that the caller is taking |
2362 | * steps that will free more memory. The caller should avoid the page | |
2363 | * being used for !PFMEMALLOC purposes. | |
2364 | */ | |
2f064f34 MH |
2365 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2366 | set_page_pfmemalloc(page); | |
2367 | else | |
2368 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2369 | } |
2370 | ||
56fd56b8 MG |
2371 | /* |
2372 | * Go through the free lists for the given migratetype and remove | |
2373 | * the smallest available page from the freelists | |
2374 | */ | |
85ccc8fa | 2375 | static __always_inline |
728ec980 | 2376 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2377 | int migratetype) |
2378 | { | |
2379 | unsigned int current_order; | |
b8af2941 | 2380 | struct free_area *area; |
56fd56b8 MG |
2381 | struct page *page; |
2382 | ||
2383 | /* Find a page of the appropriate size in the preferred list */ | |
2384 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2385 | area = &(zone->free_area[current_order]); | |
b03641af | 2386 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2387 | if (!page) |
2388 | continue; | |
6ab01363 AD |
2389 | del_page_from_free_list(page, zone, current_order); |
2390 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 2391 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
2392 | return page; |
2393 | } | |
2394 | ||
2395 | return NULL; | |
2396 | } | |
2397 | ||
2398 | ||
b2a0ac88 MG |
2399 | /* |
2400 | * This array describes the order lists are fallen back to when | |
2401 | * the free lists for the desirable migrate type are depleted | |
2402 | */ | |
da415663 | 2403 | static int fallbacks[MIGRATE_TYPES][3] = { |
974a786e | 2404 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2405 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2406 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
47118af0 | 2407 | #ifdef CONFIG_CMA |
974a786e | 2408 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 2409 | #endif |
194159fb | 2410 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 2411 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 2412 | #endif |
b2a0ac88 MG |
2413 | }; |
2414 | ||
dc67647b | 2415 | #ifdef CONFIG_CMA |
85ccc8fa | 2416 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2417 | unsigned int order) |
2418 | { | |
2419 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2420 | } | |
2421 | #else | |
2422 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2423 | unsigned int order) { return NULL; } | |
2424 | #endif | |
2425 | ||
c361be55 | 2426 | /* |
293ffa5e | 2427 | * Move the free pages in a range to the freelist tail of the requested type. |
d9c23400 | 2428 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2429 | * boundary. If alignment is required, use move_freepages_block() |
2430 | */ | |
02aa0cdd | 2431 | static int move_freepages(struct zone *zone, |
39ddb991 | 2432 | unsigned long start_pfn, unsigned long end_pfn, |
02aa0cdd | 2433 | int migratetype, int *num_movable) |
c361be55 MG |
2434 | { |
2435 | struct page *page; | |
39ddb991 | 2436 | unsigned long pfn; |
d00181b9 | 2437 | unsigned int order; |
d100313f | 2438 | int pages_moved = 0; |
c361be55 | 2439 | |
39ddb991 KW |
2440 | for (pfn = start_pfn; pfn <= end_pfn;) { |
2441 | if (!pfn_valid_within(pfn)) { | |
2442 | pfn++; | |
c361be55 MG |
2443 | continue; |
2444 | } | |
2445 | ||
39ddb991 | 2446 | page = pfn_to_page(pfn); |
c361be55 | 2447 | if (!PageBuddy(page)) { |
02aa0cdd VB |
2448 | /* |
2449 | * We assume that pages that could be isolated for | |
2450 | * migration are movable. But we don't actually try | |
2451 | * isolating, as that would be expensive. | |
2452 | */ | |
2453 | if (num_movable && | |
2454 | (PageLRU(page) || __PageMovable(page))) | |
2455 | (*num_movable)++; | |
39ddb991 | 2456 | pfn++; |
c361be55 MG |
2457 | continue; |
2458 | } | |
2459 | ||
cd961038 DR |
2460 | /* Make sure we are not inadvertently changing nodes */ |
2461 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2462 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2463 | ||
ab130f91 | 2464 | order = buddy_order(page); |
6ab01363 | 2465 | move_to_free_list(page, zone, order, migratetype); |
39ddb991 | 2466 | pfn += 1 << order; |
d100313f | 2467 | pages_moved += 1 << order; |
c361be55 MG |
2468 | } |
2469 | ||
d100313f | 2470 | return pages_moved; |
c361be55 MG |
2471 | } |
2472 | ||
ee6f509c | 2473 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2474 | int migratetype, int *num_movable) |
c361be55 | 2475 | { |
39ddb991 | 2476 | unsigned long start_pfn, end_pfn, pfn; |
c361be55 | 2477 | |
4a222127 DR |
2478 | if (num_movable) |
2479 | *num_movable = 0; | |
2480 | ||
39ddb991 KW |
2481 | pfn = page_to_pfn(page); |
2482 | start_pfn = pfn & ~(pageblock_nr_pages - 1); | |
d9c23400 | 2483 | end_pfn = start_pfn + pageblock_nr_pages - 1; |
c361be55 MG |
2484 | |
2485 | /* Do not cross zone boundaries */ | |
108bcc96 | 2486 | if (!zone_spans_pfn(zone, start_pfn)) |
39ddb991 | 2487 | start_pfn = pfn; |
108bcc96 | 2488 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2489 | return 0; |
2490 | ||
39ddb991 | 2491 | return move_freepages(zone, start_pfn, end_pfn, migratetype, |
02aa0cdd | 2492 | num_movable); |
c361be55 MG |
2493 | } |
2494 | ||
2f66a68f MG |
2495 | static void change_pageblock_range(struct page *pageblock_page, |
2496 | int start_order, int migratetype) | |
2497 | { | |
2498 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2499 | ||
2500 | while (nr_pageblocks--) { | |
2501 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2502 | pageblock_page += pageblock_nr_pages; | |
2503 | } | |
2504 | } | |
2505 | ||
fef903ef | 2506 | /* |
9c0415eb VB |
2507 | * When we are falling back to another migratetype during allocation, try to |
2508 | * steal extra free pages from the same pageblocks to satisfy further | |
2509 | * allocations, instead of polluting multiple pageblocks. | |
2510 | * | |
2511 | * If we are stealing a relatively large buddy page, it is likely there will | |
2512 | * be more free pages in the pageblock, so try to steal them all. For | |
2513 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2514 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2515 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2516 | * pageblocks. | |
fef903ef | 2517 | */ |
4eb7dce6 JK |
2518 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2519 | { | |
2520 | /* | |
2521 | * Leaving this order check is intended, although there is | |
2522 | * relaxed order check in next check. The reason is that | |
2523 | * we can actually steal whole pageblock if this condition met, | |
2524 | * but, below check doesn't guarantee it and that is just heuristic | |
2525 | * so could be changed anytime. | |
2526 | */ | |
2527 | if (order >= pageblock_order) | |
2528 | return true; | |
2529 | ||
2530 | if (order >= pageblock_order / 2 || | |
2531 | start_mt == MIGRATE_RECLAIMABLE || | |
2532 | start_mt == MIGRATE_UNMOVABLE || | |
2533 | page_group_by_mobility_disabled) | |
2534 | return true; | |
2535 | ||
2536 | return false; | |
2537 | } | |
2538 | ||
597c8920 | 2539 | static inline bool boost_watermark(struct zone *zone) |
1c30844d MG |
2540 | { |
2541 | unsigned long max_boost; | |
2542 | ||
2543 | if (!watermark_boost_factor) | |
597c8920 | 2544 | return false; |
14f69140 HW |
2545 | /* |
2546 | * Don't bother in zones that are unlikely to produce results. | |
2547 | * On small machines, including kdump capture kernels running | |
2548 | * in a small area, boosting the watermark can cause an out of | |
2549 | * memory situation immediately. | |
2550 | */ | |
2551 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
597c8920 | 2552 | return false; |
1c30844d MG |
2553 | |
2554 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2555 | watermark_boost_factor, 10000); | |
94b3334c MG |
2556 | |
2557 | /* | |
2558 | * high watermark may be uninitialised if fragmentation occurs | |
2559 | * very early in boot so do not boost. We do not fall | |
2560 | * through and boost by pageblock_nr_pages as failing | |
2561 | * allocations that early means that reclaim is not going | |
2562 | * to help and it may even be impossible to reclaim the | |
2563 | * boosted watermark resulting in a hang. | |
2564 | */ | |
2565 | if (!max_boost) | |
597c8920 | 2566 | return false; |
94b3334c | 2567 | |
1c30844d MG |
2568 | max_boost = max(pageblock_nr_pages, max_boost); |
2569 | ||
2570 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2571 | max_boost); | |
597c8920 JW |
2572 | |
2573 | return true; | |
1c30844d MG |
2574 | } |
2575 | ||
4eb7dce6 JK |
2576 | /* |
2577 | * This function implements actual steal behaviour. If order is large enough, | |
2578 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2579 | * pageblock to our migratetype and determine how many already-allocated pages |
2580 | * are there in the pageblock with a compatible migratetype. If at least half | |
2581 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2582 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2583 | */ |
2584 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2585 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2586 | { |
ab130f91 | 2587 | unsigned int current_order = buddy_order(page); |
02aa0cdd VB |
2588 | int free_pages, movable_pages, alike_pages; |
2589 | int old_block_type; | |
2590 | ||
2591 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2592 | |
3bc48f96 VB |
2593 | /* |
2594 | * This can happen due to races and we want to prevent broken | |
2595 | * highatomic accounting. | |
2596 | */ | |
02aa0cdd | 2597 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2598 | goto single_page; |
2599 | ||
fef903ef SB |
2600 | /* Take ownership for orders >= pageblock_order */ |
2601 | if (current_order >= pageblock_order) { | |
2602 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2603 | goto single_page; |
fef903ef SB |
2604 | } |
2605 | ||
1c30844d MG |
2606 | /* |
2607 | * Boost watermarks to increase reclaim pressure to reduce the | |
2608 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2609 | * may be balanced overall and kswapd will not wake naturally. | |
2610 | */ | |
597c8920 | 2611 | if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD)) |
73444bc4 | 2612 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2613 | |
3bc48f96 VB |
2614 | /* We are not allowed to try stealing from the whole block */ |
2615 | if (!whole_block) | |
2616 | goto single_page; | |
2617 | ||
02aa0cdd VB |
2618 | free_pages = move_freepages_block(zone, page, start_type, |
2619 | &movable_pages); | |
2620 | /* | |
2621 | * Determine how many pages are compatible with our allocation. | |
2622 | * For movable allocation, it's the number of movable pages which | |
2623 | * we just obtained. For other types it's a bit more tricky. | |
2624 | */ | |
2625 | if (start_type == MIGRATE_MOVABLE) { | |
2626 | alike_pages = movable_pages; | |
2627 | } else { | |
2628 | /* | |
2629 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2630 | * to MOVABLE pageblock, consider all non-movable pages as | |
2631 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2632 | * vice versa, be conservative since we can't distinguish the | |
2633 | * exact migratetype of non-movable pages. | |
2634 | */ | |
2635 | if (old_block_type == MIGRATE_MOVABLE) | |
2636 | alike_pages = pageblock_nr_pages | |
2637 | - (free_pages + movable_pages); | |
2638 | else | |
2639 | alike_pages = 0; | |
2640 | } | |
2641 | ||
3bc48f96 | 2642 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2643 | if (!free_pages) |
3bc48f96 | 2644 | goto single_page; |
fef903ef | 2645 | |
02aa0cdd VB |
2646 | /* |
2647 | * If a sufficient number of pages in the block are either free or of | |
2648 | * comparable migratability as our allocation, claim the whole block. | |
2649 | */ | |
2650 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2651 | page_group_by_mobility_disabled) |
2652 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2653 | |
2654 | return; | |
2655 | ||
2656 | single_page: | |
6ab01363 | 2657 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
2658 | } |
2659 | ||
2149cdae JK |
2660 | /* |
2661 | * Check whether there is a suitable fallback freepage with requested order. | |
2662 | * If only_stealable is true, this function returns fallback_mt only if | |
2663 | * we can steal other freepages all together. This would help to reduce | |
2664 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2665 | */ | |
2666 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2667 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2668 | { |
2669 | int i; | |
2670 | int fallback_mt; | |
2671 | ||
2672 | if (area->nr_free == 0) | |
2673 | return -1; | |
2674 | ||
2675 | *can_steal = false; | |
2676 | for (i = 0;; i++) { | |
2677 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2678 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2679 | break; |
2680 | ||
b03641af | 2681 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2682 | continue; |
fef903ef | 2683 | |
4eb7dce6 JK |
2684 | if (can_steal_fallback(order, migratetype)) |
2685 | *can_steal = true; | |
2686 | ||
2149cdae JK |
2687 | if (!only_stealable) |
2688 | return fallback_mt; | |
2689 | ||
2690 | if (*can_steal) | |
2691 | return fallback_mt; | |
fef903ef | 2692 | } |
4eb7dce6 JK |
2693 | |
2694 | return -1; | |
fef903ef SB |
2695 | } |
2696 | ||
0aaa29a5 MG |
2697 | /* |
2698 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2699 | * there are no empty page blocks that contain a page with a suitable order | |
2700 | */ | |
2701 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2702 | unsigned int alloc_order) | |
2703 | { | |
2704 | int mt; | |
2705 | unsigned long max_managed, flags; | |
2706 | ||
2707 | /* | |
2708 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2709 | * Check is race-prone but harmless. | |
2710 | */ | |
9705bea5 | 2711 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2712 | if (zone->nr_reserved_highatomic >= max_managed) |
2713 | return; | |
2714 | ||
2715 | spin_lock_irqsave(&zone->lock, flags); | |
2716 | ||
2717 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2718 | if (zone->nr_reserved_highatomic >= max_managed) | |
2719 | goto out_unlock; | |
2720 | ||
2721 | /* Yoink! */ | |
2722 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2723 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2724 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2725 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2726 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2727 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2728 | } |
2729 | ||
2730 | out_unlock: | |
2731 | spin_unlock_irqrestore(&zone->lock, flags); | |
2732 | } | |
2733 | ||
2734 | /* | |
2735 | * Used when an allocation is about to fail under memory pressure. This | |
2736 | * potentially hurts the reliability of high-order allocations when under | |
2737 | * intense memory pressure but failed atomic allocations should be easier | |
2738 | * to recover from than an OOM. | |
29fac03b MK |
2739 | * |
2740 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2741 | * pageblock is exhausted. | |
0aaa29a5 | 2742 | */ |
29fac03b MK |
2743 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2744 | bool force) | |
0aaa29a5 MG |
2745 | { |
2746 | struct zonelist *zonelist = ac->zonelist; | |
2747 | unsigned long flags; | |
2748 | struct zoneref *z; | |
2749 | struct zone *zone; | |
2750 | struct page *page; | |
2751 | int order; | |
04c8716f | 2752 | bool ret; |
0aaa29a5 | 2753 | |
97a225e6 | 2754 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2755 | ac->nodemask) { |
29fac03b MK |
2756 | /* |
2757 | * Preserve at least one pageblock unless memory pressure | |
2758 | * is really high. | |
2759 | */ | |
2760 | if (!force && zone->nr_reserved_highatomic <= | |
2761 | pageblock_nr_pages) | |
0aaa29a5 MG |
2762 | continue; |
2763 | ||
2764 | spin_lock_irqsave(&zone->lock, flags); | |
2765 | for (order = 0; order < MAX_ORDER; order++) { | |
2766 | struct free_area *area = &(zone->free_area[order]); | |
2767 | ||
b03641af | 2768 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2769 | if (!page) |
0aaa29a5 MG |
2770 | continue; |
2771 | ||
0aaa29a5 | 2772 | /* |
4855e4a7 MK |
2773 | * In page freeing path, migratetype change is racy so |
2774 | * we can counter several free pages in a pageblock | |
2775 | * in this loop althoug we changed the pageblock type | |
2776 | * from highatomic to ac->migratetype. So we should | |
2777 | * adjust the count once. | |
0aaa29a5 | 2778 | */ |
a6ffdc07 | 2779 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2780 | /* |
2781 | * It should never happen but changes to | |
2782 | * locking could inadvertently allow a per-cpu | |
2783 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2784 | * while unreserving so be safe and watch for | |
2785 | * underflows. | |
2786 | */ | |
2787 | zone->nr_reserved_highatomic -= min( | |
2788 | pageblock_nr_pages, | |
2789 | zone->nr_reserved_highatomic); | |
2790 | } | |
0aaa29a5 MG |
2791 | |
2792 | /* | |
2793 | * Convert to ac->migratetype and avoid the normal | |
2794 | * pageblock stealing heuristics. Minimally, the caller | |
2795 | * is doing the work and needs the pages. More | |
2796 | * importantly, if the block was always converted to | |
2797 | * MIGRATE_UNMOVABLE or another type then the number | |
2798 | * of pageblocks that cannot be completely freed | |
2799 | * may increase. | |
2800 | */ | |
2801 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2802 | ret = move_freepages_block(zone, page, ac->migratetype, |
2803 | NULL); | |
29fac03b MK |
2804 | if (ret) { |
2805 | spin_unlock_irqrestore(&zone->lock, flags); | |
2806 | return ret; | |
2807 | } | |
0aaa29a5 MG |
2808 | } |
2809 | spin_unlock_irqrestore(&zone->lock, flags); | |
2810 | } | |
04c8716f MK |
2811 | |
2812 | return false; | |
0aaa29a5 MG |
2813 | } |
2814 | ||
3bc48f96 VB |
2815 | /* |
2816 | * Try finding a free buddy page on the fallback list and put it on the free | |
2817 | * list of requested migratetype, possibly along with other pages from the same | |
2818 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2819 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2820 | * |
2821 | * The use of signed ints for order and current_order is a deliberate | |
2822 | * deviation from the rest of this file, to make the for loop | |
2823 | * condition simpler. | |
3bc48f96 | 2824 | */ |
85ccc8fa | 2825 | static __always_inline bool |
6bb15450 MG |
2826 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2827 | unsigned int alloc_flags) | |
b2a0ac88 | 2828 | { |
b8af2941 | 2829 | struct free_area *area; |
b002529d | 2830 | int current_order; |
6bb15450 | 2831 | int min_order = order; |
b2a0ac88 | 2832 | struct page *page; |
4eb7dce6 JK |
2833 | int fallback_mt; |
2834 | bool can_steal; | |
b2a0ac88 | 2835 | |
6bb15450 MG |
2836 | /* |
2837 | * Do not steal pages from freelists belonging to other pageblocks | |
2838 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2839 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2840 | */ | |
2841 | if (alloc_flags & ALLOC_NOFRAGMENT) | |
2842 | min_order = pageblock_order; | |
2843 | ||
7a8f58f3 VB |
2844 | /* |
2845 | * Find the largest available free page in the other list. This roughly | |
2846 | * approximates finding the pageblock with the most free pages, which | |
2847 | * would be too costly to do exactly. | |
2848 | */ | |
6bb15450 | 2849 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 2850 | --current_order) { |
4eb7dce6 JK |
2851 | area = &(zone->free_area[current_order]); |
2852 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2853 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2854 | if (fallback_mt == -1) |
2855 | continue; | |
b2a0ac88 | 2856 | |
7a8f58f3 VB |
2857 | /* |
2858 | * We cannot steal all free pages from the pageblock and the | |
2859 | * requested migratetype is movable. In that case it's better to | |
2860 | * steal and split the smallest available page instead of the | |
2861 | * largest available page, because even if the next movable | |
2862 | * allocation falls back into a different pageblock than this | |
2863 | * one, it won't cause permanent fragmentation. | |
2864 | */ | |
2865 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2866 | && current_order > order) | |
2867 | goto find_smallest; | |
b2a0ac88 | 2868 | |
7a8f58f3 VB |
2869 | goto do_steal; |
2870 | } | |
e0fff1bd | 2871 | |
7a8f58f3 | 2872 | return false; |
e0fff1bd | 2873 | |
7a8f58f3 VB |
2874 | find_smallest: |
2875 | for (current_order = order; current_order < MAX_ORDER; | |
2876 | current_order++) { | |
2877 | area = &(zone->free_area[current_order]); | |
2878 | fallback_mt = find_suitable_fallback(area, current_order, | |
2879 | start_migratetype, false, &can_steal); | |
2880 | if (fallback_mt != -1) | |
2881 | break; | |
b2a0ac88 MG |
2882 | } |
2883 | ||
7a8f58f3 VB |
2884 | /* |
2885 | * This should not happen - we already found a suitable fallback | |
2886 | * when looking for the largest page. | |
2887 | */ | |
2888 | VM_BUG_ON(current_order == MAX_ORDER); | |
2889 | ||
2890 | do_steal: | |
b03641af | 2891 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2892 | |
1c30844d MG |
2893 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
2894 | can_steal); | |
7a8f58f3 VB |
2895 | |
2896 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2897 | start_migratetype, fallback_mt); | |
2898 | ||
2899 | return true; | |
2900 | ||
b2a0ac88 MG |
2901 | } |
2902 | ||
56fd56b8 | 2903 | /* |
1da177e4 LT |
2904 | * Do the hard work of removing an element from the buddy allocator. |
2905 | * Call me with the zone->lock already held. | |
2906 | */ | |
85ccc8fa | 2907 | static __always_inline struct page * |
6bb15450 MG |
2908 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2909 | unsigned int alloc_flags) | |
1da177e4 | 2910 | { |
1da177e4 LT |
2911 | struct page *page; |
2912 | ||
ce8f86ee H |
2913 | if (IS_ENABLED(CONFIG_CMA)) { |
2914 | /* | |
2915 | * Balance movable allocations between regular and CMA areas by | |
2916 | * allocating from CMA when over half of the zone's free memory | |
2917 | * is in the CMA area. | |
2918 | */ | |
2919 | if (alloc_flags & ALLOC_CMA && | |
2920 | zone_page_state(zone, NR_FREE_CMA_PAGES) > | |
2921 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
2922 | page = __rmqueue_cma_fallback(zone, order); | |
2923 | if (page) | |
2924 | goto out; | |
2925 | } | |
16867664 | 2926 | } |
3bc48f96 | 2927 | retry: |
56fd56b8 | 2928 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2929 | if (unlikely(!page)) { |
8510e69c | 2930 | if (alloc_flags & ALLOC_CMA) |
dc67647b JK |
2931 | page = __rmqueue_cma_fallback(zone, order); |
2932 | ||
6bb15450 MG |
2933 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2934 | alloc_flags)) | |
3bc48f96 | 2935 | goto retry; |
728ec980 | 2936 | } |
ce8f86ee H |
2937 | out: |
2938 | if (page) | |
2939 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
b2a0ac88 | 2940 | return page; |
1da177e4 LT |
2941 | } |
2942 | ||
5f63b720 | 2943 | /* |
1da177e4 LT |
2944 | * Obtain a specified number of elements from the buddy allocator, all under |
2945 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2946 | * Returns the number of new pages which were placed at *list. | |
2947 | */ | |
5f63b720 | 2948 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2949 | unsigned long count, struct list_head *list, |
6bb15450 | 2950 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2951 | { |
cb66bede | 2952 | int i, allocated = 0; |
5f63b720 | 2953 | |
d34b0733 | 2954 | spin_lock(&zone->lock); |
1da177e4 | 2955 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2956 | struct page *page = __rmqueue(zone, order, migratetype, |
2957 | alloc_flags); | |
085cc7d5 | 2958 | if (unlikely(page == NULL)) |
1da177e4 | 2959 | break; |
81eabcbe | 2960 | |
479f854a MG |
2961 | if (unlikely(check_pcp_refill(page))) |
2962 | continue; | |
2963 | ||
81eabcbe | 2964 | /* |
0fac3ba5 VB |
2965 | * Split buddy pages returned by expand() are received here in |
2966 | * physical page order. The page is added to the tail of | |
2967 | * caller's list. From the callers perspective, the linked list | |
2968 | * is ordered by page number under some conditions. This is | |
2969 | * useful for IO devices that can forward direction from the | |
2970 | * head, thus also in the physical page order. This is useful | |
2971 | * for IO devices that can merge IO requests if the physical | |
2972 | * pages are ordered properly. | |
81eabcbe | 2973 | */ |
0fac3ba5 | 2974 | list_add_tail(&page->lru, list); |
cb66bede | 2975 | allocated++; |
bb14c2c7 | 2976 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2977 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2978 | -(1 << order)); | |
1da177e4 | 2979 | } |
a6de734b MG |
2980 | |
2981 | /* | |
2982 | * i pages were removed from the buddy list even if some leak due | |
2983 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
cb66bede | 2984 | * on i. Do not confuse with 'allocated' which is the number of |
a6de734b MG |
2985 | * pages added to the pcp list. |
2986 | */ | |
f2260e6b | 2987 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2988 | spin_unlock(&zone->lock); |
cb66bede | 2989 | return allocated; |
1da177e4 LT |
2990 | } |
2991 | ||
4ae7c039 | 2992 | #ifdef CONFIG_NUMA |
8fce4d8e | 2993 | /* |
4037d452 CL |
2994 | * Called from the vmstat counter updater to drain pagesets of this |
2995 | * currently executing processor on remote nodes after they have | |
2996 | * expired. | |
2997 | * | |
879336c3 CL |
2998 | * Note that this function must be called with the thread pinned to |
2999 | * a single processor. | |
8fce4d8e | 3000 | */ |
4037d452 | 3001 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 3002 | { |
4ae7c039 | 3003 | unsigned long flags; |
7be12fc9 | 3004 | int to_drain, batch; |
4ae7c039 | 3005 | |
4037d452 | 3006 | local_irq_save(flags); |
4db0c3c2 | 3007 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 3008 | to_drain = min(pcp->count, batch); |
77ba9062 | 3009 | if (to_drain > 0) |
2a13515c | 3010 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 3011 | local_irq_restore(flags); |
4ae7c039 CL |
3012 | } |
3013 | #endif | |
3014 | ||
9f8f2172 | 3015 | /* |
93481ff0 | 3016 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
3017 | * |
3018 | * The processor must either be the current processor and the | |
3019 | * thread pinned to the current processor or a processor that | |
3020 | * is not online. | |
3021 | */ | |
93481ff0 | 3022 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 3023 | { |
c54ad30c | 3024 | unsigned long flags; |
93481ff0 VB |
3025 | struct per_cpu_pageset *pset; |
3026 | struct per_cpu_pages *pcp; | |
1da177e4 | 3027 | |
93481ff0 VB |
3028 | local_irq_save(flags); |
3029 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 3030 | |
93481ff0 | 3031 | pcp = &pset->pcp; |
77ba9062 | 3032 | if (pcp->count) |
93481ff0 | 3033 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
3034 | local_irq_restore(flags); |
3035 | } | |
3dfa5721 | 3036 | |
93481ff0 VB |
3037 | /* |
3038 | * Drain pcplists of all zones on the indicated processor. | |
3039 | * | |
3040 | * The processor must either be the current processor and the | |
3041 | * thread pinned to the current processor or a processor that | |
3042 | * is not online. | |
3043 | */ | |
3044 | static void drain_pages(unsigned int cpu) | |
3045 | { | |
3046 | struct zone *zone; | |
3047 | ||
3048 | for_each_populated_zone(zone) { | |
3049 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
3050 | } |
3051 | } | |
1da177e4 | 3052 | |
9f8f2172 CL |
3053 | /* |
3054 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
3055 | * |
3056 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
3057 | * the single zone's pages. | |
9f8f2172 | 3058 | */ |
93481ff0 | 3059 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 3060 | { |
93481ff0 VB |
3061 | int cpu = smp_processor_id(); |
3062 | ||
3063 | if (zone) | |
3064 | drain_pages_zone(cpu, zone); | |
3065 | else | |
3066 | drain_pages(cpu); | |
9f8f2172 CL |
3067 | } |
3068 | ||
0ccce3b9 MG |
3069 | static void drain_local_pages_wq(struct work_struct *work) |
3070 | { | |
d9367bd0 WY |
3071 | struct pcpu_drain *drain; |
3072 | ||
3073 | drain = container_of(work, struct pcpu_drain, work); | |
3074 | ||
a459eeb7 MH |
3075 | /* |
3076 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
3077 | * we can race with cpu offline when the WQ can move this from | |
3078 | * a cpu pinned worker to an unbound one. We can operate on a different | |
3079 | * cpu which is allright but we also have to make sure to not move to | |
3080 | * a different one. | |
3081 | */ | |
3082 | preempt_disable(); | |
d9367bd0 | 3083 | drain_local_pages(drain->zone); |
a459eeb7 | 3084 | preempt_enable(); |
0ccce3b9 MG |
3085 | } |
3086 | ||
9f8f2172 | 3087 | /* |
ec6e8c7e VB |
3088 | * The implementation of drain_all_pages(), exposing an extra parameter to |
3089 | * drain on all cpus. | |
93481ff0 | 3090 | * |
ec6e8c7e VB |
3091 | * drain_all_pages() is optimized to only execute on cpus where pcplists are |
3092 | * not empty. The check for non-emptiness can however race with a free to | |
3093 | * pcplist that has not yet increased the pcp->count from 0 to 1. Callers | |
3094 | * that need the guarantee that every CPU has drained can disable the | |
3095 | * optimizing racy check. | |
9f8f2172 | 3096 | */ |
3b1f3658 | 3097 | static void __drain_all_pages(struct zone *zone, bool force_all_cpus) |
9f8f2172 | 3098 | { |
74046494 | 3099 | int cpu; |
74046494 GBY |
3100 | |
3101 | /* | |
3102 | * Allocate in the BSS so we wont require allocation in | |
3103 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
3104 | */ | |
3105 | static cpumask_t cpus_with_pcps; | |
3106 | ||
ce612879 MH |
3107 | /* |
3108 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
3109 | * initialized. | |
3110 | */ | |
3111 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
3112 | return; | |
3113 | ||
bd233f53 MG |
3114 | /* |
3115 | * Do not drain if one is already in progress unless it's specific to | |
3116 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
3117 | * the drain to be complete when the call returns. | |
3118 | */ | |
3119 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
3120 | if (!zone) | |
3121 | return; | |
3122 | mutex_lock(&pcpu_drain_mutex); | |
3123 | } | |
0ccce3b9 | 3124 | |
74046494 GBY |
3125 | /* |
3126 | * We don't care about racing with CPU hotplug event | |
3127 | * as offline notification will cause the notified | |
3128 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
3129 | * disables preemption as part of its processing | |
3130 | */ | |
3131 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
3132 | struct per_cpu_pageset *pcp; |
3133 | struct zone *z; | |
74046494 | 3134 | bool has_pcps = false; |
93481ff0 | 3135 | |
ec6e8c7e VB |
3136 | if (force_all_cpus) { |
3137 | /* | |
3138 | * The pcp.count check is racy, some callers need a | |
3139 | * guarantee that no cpu is missed. | |
3140 | */ | |
3141 | has_pcps = true; | |
3142 | } else if (zone) { | |
74046494 | 3143 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 3144 | if (pcp->pcp.count) |
74046494 | 3145 | has_pcps = true; |
93481ff0 VB |
3146 | } else { |
3147 | for_each_populated_zone(z) { | |
3148 | pcp = per_cpu_ptr(z->pageset, cpu); | |
3149 | if (pcp->pcp.count) { | |
3150 | has_pcps = true; | |
3151 | break; | |
3152 | } | |
74046494 GBY |
3153 | } |
3154 | } | |
93481ff0 | 3155 | |
74046494 GBY |
3156 | if (has_pcps) |
3157 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
3158 | else | |
3159 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
3160 | } | |
0ccce3b9 | 3161 | |
bd233f53 | 3162 | for_each_cpu(cpu, &cpus_with_pcps) { |
d9367bd0 WY |
3163 | struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu); |
3164 | ||
3165 | drain->zone = zone; | |
3166 | INIT_WORK(&drain->work, drain_local_pages_wq); | |
3167 | queue_work_on(cpu, mm_percpu_wq, &drain->work); | |
0ccce3b9 | 3168 | } |
bd233f53 | 3169 | for_each_cpu(cpu, &cpus_with_pcps) |
d9367bd0 | 3170 | flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work); |
bd233f53 MG |
3171 | |
3172 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
3173 | } |
3174 | ||
ec6e8c7e VB |
3175 | /* |
3176 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. | |
3177 | * | |
3178 | * When zone parameter is non-NULL, spill just the single zone's pages. | |
3179 | * | |
3180 | * Note that this can be extremely slow as the draining happens in a workqueue. | |
3181 | */ | |
3182 | void drain_all_pages(struct zone *zone) | |
3183 | { | |
3184 | __drain_all_pages(zone, false); | |
3185 | } | |
3186 | ||
296699de | 3187 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3188 | |
556b969a CY |
3189 | /* |
3190 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
3191 | */ | |
3192 | #define WD_PAGE_COUNT (128*1024) | |
3193 | ||
1da177e4 LT |
3194 | void mark_free_pages(struct zone *zone) |
3195 | { | |
556b969a | 3196 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 3197 | unsigned long flags; |
7aeb09f9 | 3198 | unsigned int order, t; |
86760a2c | 3199 | struct page *page; |
1da177e4 | 3200 | |
8080fc03 | 3201 | if (zone_is_empty(zone)) |
1da177e4 LT |
3202 | return; |
3203 | ||
3204 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 3205 | |
108bcc96 | 3206 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
3207 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
3208 | if (pfn_valid(pfn)) { | |
86760a2c | 3209 | page = pfn_to_page(pfn); |
ba6b0979 | 3210 | |
556b969a CY |
3211 | if (!--page_count) { |
3212 | touch_nmi_watchdog(); | |
3213 | page_count = WD_PAGE_COUNT; | |
3214 | } | |
3215 | ||
ba6b0979 JK |
3216 | if (page_zone(page) != zone) |
3217 | continue; | |
3218 | ||
7be98234 RW |
3219 | if (!swsusp_page_is_forbidden(page)) |
3220 | swsusp_unset_page_free(page); | |
f623f0db | 3221 | } |
1da177e4 | 3222 | |
b2a0ac88 | 3223 | for_each_migratetype_order(order, t) { |
86760a2c GT |
3224 | list_for_each_entry(page, |
3225 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 3226 | unsigned long i; |
1da177e4 | 3227 | |
86760a2c | 3228 | pfn = page_to_pfn(page); |
556b969a CY |
3229 | for (i = 0; i < (1UL << order); i++) { |
3230 | if (!--page_count) { | |
3231 | touch_nmi_watchdog(); | |
3232 | page_count = WD_PAGE_COUNT; | |
3233 | } | |
7be98234 | 3234 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 3235 | } |
f623f0db | 3236 | } |
b2a0ac88 | 3237 | } |
1da177e4 LT |
3238 | spin_unlock_irqrestore(&zone->lock, flags); |
3239 | } | |
e2c55dc8 | 3240 | #endif /* CONFIG_PM */ |
1da177e4 | 3241 | |
2d4894b5 | 3242 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 3243 | { |
5f8dcc21 | 3244 | int migratetype; |
1da177e4 | 3245 | |
4db7548c | 3246 | if (!free_pcp_prepare(page)) |
9cca35d4 | 3247 | return false; |
689bcebf | 3248 | |
dc4b0caf | 3249 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 3250 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
3251 | return true; |
3252 | } | |
3253 | ||
2d4894b5 | 3254 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
3255 | { |
3256 | struct zone *zone = page_zone(page); | |
3257 | struct per_cpu_pages *pcp; | |
3258 | int migratetype; | |
3259 | ||
3260 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 3261 | __count_vm_event(PGFREE); |
da456f14 | 3262 | |
5f8dcc21 MG |
3263 | /* |
3264 | * We only track unmovable, reclaimable and movable on pcp lists. | |
3265 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 3266 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
3267 | * areas back if necessary. Otherwise, we may have to free |
3268 | * excessively into the page allocator | |
3269 | */ | |
3270 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 3271 | if (unlikely(is_migrate_isolate(migratetype))) { |
7fef431b DH |
3272 | free_one_page(zone, page, pfn, 0, migratetype, |
3273 | FPI_NONE); | |
9cca35d4 | 3274 | return; |
5f8dcc21 MG |
3275 | } |
3276 | migratetype = MIGRATE_MOVABLE; | |
3277 | } | |
3278 | ||
99dcc3e5 | 3279 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 3280 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 3281 | pcp->count++; |
5c3ad2eb VB |
3282 | if (pcp->count >= READ_ONCE(pcp->high)) |
3283 | free_pcppages_bulk(zone, READ_ONCE(pcp->batch), pcp); | |
9cca35d4 | 3284 | } |
5f8dcc21 | 3285 | |
9cca35d4 MG |
3286 | /* |
3287 | * Free a 0-order page | |
9cca35d4 | 3288 | */ |
2d4894b5 | 3289 | void free_unref_page(struct page *page) |
9cca35d4 MG |
3290 | { |
3291 | unsigned long flags; | |
3292 | unsigned long pfn = page_to_pfn(page); | |
3293 | ||
2d4894b5 | 3294 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3295 | return; |
3296 | ||
3297 | local_irq_save(flags); | |
2d4894b5 | 3298 | free_unref_page_commit(page, pfn); |
d34b0733 | 3299 | local_irq_restore(flags); |
1da177e4 LT |
3300 | } |
3301 | ||
cc59850e KK |
3302 | /* |
3303 | * Free a list of 0-order pages | |
3304 | */ | |
2d4894b5 | 3305 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
3306 | { |
3307 | struct page *page, *next; | |
9cca35d4 | 3308 | unsigned long flags, pfn; |
c24ad77d | 3309 | int batch_count = 0; |
9cca35d4 MG |
3310 | |
3311 | /* Prepare pages for freeing */ | |
3312 | list_for_each_entry_safe(page, next, list, lru) { | |
3313 | pfn = page_to_pfn(page); | |
2d4894b5 | 3314 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3315 | list_del(&page->lru); |
3316 | set_page_private(page, pfn); | |
3317 | } | |
cc59850e | 3318 | |
9cca35d4 | 3319 | local_irq_save(flags); |
cc59850e | 3320 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
3321 | unsigned long pfn = page_private(page); |
3322 | ||
3323 | set_page_private(page, 0); | |
2d4894b5 MG |
3324 | trace_mm_page_free_batched(page); |
3325 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
3326 | |
3327 | /* | |
3328 | * Guard against excessive IRQ disabled times when we get | |
3329 | * a large list of pages to free. | |
3330 | */ | |
3331 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
3332 | local_irq_restore(flags); | |
3333 | batch_count = 0; | |
3334 | local_irq_save(flags); | |
3335 | } | |
cc59850e | 3336 | } |
9cca35d4 | 3337 | local_irq_restore(flags); |
cc59850e KK |
3338 | } |
3339 | ||
8dfcc9ba NP |
3340 | /* |
3341 | * split_page takes a non-compound higher-order page, and splits it into | |
3342 | * n (1<<order) sub-pages: page[0..n] | |
3343 | * Each sub-page must be freed individually. | |
3344 | * | |
3345 | * Note: this is probably too low level an operation for use in drivers. | |
3346 | * Please consult with lkml before using this in your driver. | |
3347 | */ | |
3348 | void split_page(struct page *page, unsigned int order) | |
3349 | { | |
3350 | int i; | |
3351 | ||
309381fe SL |
3352 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3353 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3354 | |
a9627bc5 | 3355 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3356 | set_page_refcounted(page + i); |
8fb156c9 | 3357 | split_page_owner(page, 1 << order); |
e1baddf8 | 3358 | split_page_memcg(page, 1 << order); |
8dfcc9ba | 3359 | } |
5853ff23 | 3360 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3361 | |
3c605096 | 3362 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3363 | { |
748446bb MG |
3364 | unsigned long watermark; |
3365 | struct zone *zone; | |
2139cbe6 | 3366 | int mt; |
748446bb MG |
3367 | |
3368 | BUG_ON(!PageBuddy(page)); | |
3369 | ||
3370 | zone = page_zone(page); | |
2e30abd1 | 3371 | mt = get_pageblock_migratetype(page); |
748446bb | 3372 | |
194159fb | 3373 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
3374 | /* |
3375 | * Obey watermarks as if the page was being allocated. We can | |
3376 | * emulate a high-order watermark check with a raised order-0 | |
3377 | * watermark, because we already know our high-order page | |
3378 | * exists. | |
3379 | */ | |
fd1444b2 | 3380 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3381 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3382 | return 0; |
3383 | ||
8fb74b9f | 3384 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3385 | } |
748446bb MG |
3386 | |
3387 | /* Remove page from free list */ | |
b03641af | 3388 | |
6ab01363 | 3389 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 3390 | |
400bc7fd | 3391 | /* |
3392 | * Set the pageblock if the isolated page is at least half of a | |
3393 | * pageblock | |
3394 | */ | |
748446bb MG |
3395 | if (order >= pageblock_order - 1) { |
3396 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3397 | for (; page < endpage; page += pageblock_nr_pages) { |
3398 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 3399 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 3400 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
3401 | set_pageblock_migratetype(page, |
3402 | MIGRATE_MOVABLE); | |
3403 | } | |
748446bb MG |
3404 | } |
3405 | ||
f3a14ced | 3406 | |
8fb74b9f | 3407 | return 1UL << order; |
1fb3f8ca MG |
3408 | } |
3409 | ||
624f58d8 AD |
3410 | /** |
3411 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
3412 | * @page: Page that was isolated | |
3413 | * @order: Order of the isolated page | |
e6a0a7ad | 3414 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
3415 | * |
3416 | * This function is meant to return a page pulled from the free lists via | |
3417 | * __isolate_free_page back to the free lists they were pulled from. | |
3418 | */ | |
3419 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
3420 | { | |
3421 | struct zone *zone = page_zone(page); | |
3422 | ||
3423 | /* zone lock should be held when this function is called */ | |
3424 | lockdep_assert_held(&zone->lock); | |
3425 | ||
3426 | /* Return isolated page to tail of freelist. */ | |
f04a5d5d | 3427 | __free_one_page(page, page_to_pfn(page), zone, order, mt, |
47b6a24a | 3428 | FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL); |
624f58d8 AD |
3429 | } |
3430 | ||
060e7417 MG |
3431 | /* |
3432 | * Update NUMA hit/miss statistics | |
3433 | * | |
3434 | * Must be called with interrupts disabled. | |
060e7417 | 3435 | */ |
41b6167e | 3436 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
3437 | { |
3438 | #ifdef CONFIG_NUMA | |
3a321d2a | 3439 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3440 | |
4518085e KW |
3441 | /* skip numa counters update if numa stats is disabled */ |
3442 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3443 | return; | |
3444 | ||
c1093b74 | 3445 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3446 | local_stat = NUMA_OTHER; |
060e7417 | 3447 | |
c1093b74 | 3448 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3a321d2a | 3449 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 3450 | else { |
3a321d2a KW |
3451 | __inc_numa_state(z, NUMA_MISS); |
3452 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 3453 | } |
3a321d2a | 3454 | __inc_numa_state(z, local_stat); |
060e7417 MG |
3455 | #endif |
3456 | } | |
3457 | ||
066b2393 | 3458 | /* Remove page from the per-cpu list, caller must protect the list */ |
3b822017 JDB |
3459 | static inline |
3460 | struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
6bb15450 | 3461 | unsigned int alloc_flags, |
453f85d4 | 3462 | struct per_cpu_pages *pcp, |
066b2393 MG |
3463 | struct list_head *list) |
3464 | { | |
3465 | struct page *page; | |
3466 | ||
3467 | do { | |
3468 | if (list_empty(list)) { | |
3469 | pcp->count += rmqueue_bulk(zone, 0, | |
5c3ad2eb | 3470 | READ_ONCE(pcp->batch), list, |
6bb15450 | 3471 | migratetype, alloc_flags); |
066b2393 MG |
3472 | if (unlikely(list_empty(list))) |
3473 | return NULL; | |
3474 | } | |
3475 | ||
453f85d4 | 3476 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
3477 | list_del(&page->lru); |
3478 | pcp->count--; | |
3479 | } while (check_new_pcp(page)); | |
3480 | ||
3481 | return page; | |
3482 | } | |
3483 | ||
3484 | /* Lock and remove page from the per-cpu list */ | |
3485 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
1c52e6d0 YS |
3486 | struct zone *zone, gfp_t gfp_flags, |
3487 | int migratetype, unsigned int alloc_flags) | |
066b2393 MG |
3488 | { |
3489 | struct per_cpu_pages *pcp; | |
3490 | struct list_head *list; | |
066b2393 | 3491 | struct page *page; |
d34b0733 | 3492 | unsigned long flags; |
066b2393 | 3493 | |
d34b0733 | 3494 | local_irq_save(flags); |
066b2393 MG |
3495 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3496 | list = &pcp->lists[migratetype]; | |
6bb15450 | 3497 | page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list); |
066b2393 | 3498 | if (page) { |
1c52e6d0 | 3499 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1); |
066b2393 MG |
3500 | zone_statistics(preferred_zone, zone); |
3501 | } | |
d34b0733 | 3502 | local_irq_restore(flags); |
066b2393 MG |
3503 | return page; |
3504 | } | |
3505 | ||
1da177e4 | 3506 | /* |
75379191 | 3507 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 3508 | */ |
0a15c3e9 | 3509 | static inline |
066b2393 | 3510 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3511 | struct zone *zone, unsigned int order, |
c603844b MG |
3512 | gfp_t gfp_flags, unsigned int alloc_flags, |
3513 | int migratetype) | |
1da177e4 LT |
3514 | { |
3515 | unsigned long flags; | |
689bcebf | 3516 | struct page *page; |
1da177e4 | 3517 | |
d34b0733 | 3518 | if (likely(order == 0)) { |
1d91df85 JK |
3519 | /* |
3520 | * MIGRATE_MOVABLE pcplist could have the pages on CMA area and | |
3521 | * we need to skip it when CMA area isn't allowed. | |
3522 | */ | |
3523 | if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA || | |
3524 | migratetype != MIGRATE_MOVABLE) { | |
3525 | page = rmqueue_pcplist(preferred_zone, zone, gfp_flags, | |
1c52e6d0 | 3526 | migratetype, alloc_flags); |
1d91df85 JK |
3527 | goto out; |
3528 | } | |
066b2393 | 3529 | } |
83b9355b | 3530 | |
066b2393 MG |
3531 | /* |
3532 | * We most definitely don't want callers attempting to | |
3533 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3534 | */ | |
3535 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3536 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3537 | |
066b2393 MG |
3538 | do { |
3539 | page = NULL; | |
1d91df85 JK |
3540 | /* |
3541 | * order-0 request can reach here when the pcplist is skipped | |
3542 | * due to non-CMA allocation context. HIGHATOMIC area is | |
3543 | * reserved for high-order atomic allocation, so order-0 | |
3544 | * request should skip it. | |
3545 | */ | |
3546 | if (order > 0 && alloc_flags & ALLOC_HARDER) { | |
066b2393 MG |
3547 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
3548 | if (page) | |
3549 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3550 | } | |
a74609fa | 3551 | if (!page) |
6bb15450 | 3552 | page = __rmqueue(zone, order, migratetype, alloc_flags); |
066b2393 MG |
3553 | } while (page && check_new_pages(page, order)); |
3554 | spin_unlock(&zone->lock); | |
3555 | if (!page) | |
3556 | goto failed; | |
3557 | __mod_zone_freepage_state(zone, -(1 << order), | |
3558 | get_pcppage_migratetype(page)); | |
1da177e4 | 3559 | |
16709d1d | 3560 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3561 | zone_statistics(preferred_zone, zone); |
a74609fa | 3562 | local_irq_restore(flags); |
1da177e4 | 3563 | |
066b2393 | 3564 | out: |
73444bc4 MG |
3565 | /* Separate test+clear to avoid unnecessary atomics */ |
3566 | if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) { | |
3567 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); | |
3568 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3569 | } | |
3570 | ||
066b2393 | 3571 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 | 3572 | return page; |
a74609fa NP |
3573 | |
3574 | failed: | |
3575 | local_irq_restore(flags); | |
a74609fa | 3576 | return NULL; |
1da177e4 LT |
3577 | } |
3578 | ||
933e312e AM |
3579 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3580 | ||
b2588c4b | 3581 | static struct { |
933e312e AM |
3582 | struct fault_attr attr; |
3583 | ||
621a5f7a | 3584 | bool ignore_gfp_highmem; |
71baba4b | 3585 | bool ignore_gfp_reclaim; |
54114994 | 3586 | u32 min_order; |
933e312e AM |
3587 | } fail_page_alloc = { |
3588 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3589 | .ignore_gfp_reclaim = true, |
621a5f7a | 3590 | .ignore_gfp_highmem = true, |
54114994 | 3591 | .min_order = 1, |
933e312e AM |
3592 | }; |
3593 | ||
3594 | static int __init setup_fail_page_alloc(char *str) | |
3595 | { | |
3596 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3597 | } | |
3598 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3599 | ||
af3b8544 | 3600 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3601 | { |
54114994 | 3602 | if (order < fail_page_alloc.min_order) |
deaf386e | 3603 | return false; |
933e312e | 3604 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3605 | return false; |
933e312e | 3606 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3607 | return false; |
71baba4b MG |
3608 | if (fail_page_alloc.ignore_gfp_reclaim && |
3609 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3610 | return false; |
933e312e AM |
3611 | |
3612 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3613 | } | |
3614 | ||
3615 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3616 | ||
3617 | static int __init fail_page_alloc_debugfs(void) | |
3618 | { | |
0825a6f9 | 3619 | umode_t mode = S_IFREG | 0600; |
933e312e | 3620 | struct dentry *dir; |
933e312e | 3621 | |
dd48c085 AM |
3622 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3623 | &fail_page_alloc.attr); | |
b2588c4b | 3624 | |
d9f7979c GKH |
3625 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3626 | &fail_page_alloc.ignore_gfp_reclaim); | |
3627 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3628 | &fail_page_alloc.ignore_gfp_highmem); | |
3629 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3630 | |
d9f7979c | 3631 | return 0; |
933e312e AM |
3632 | } |
3633 | ||
3634 | late_initcall(fail_page_alloc_debugfs); | |
3635 | ||
3636 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3637 | ||
3638 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3639 | ||
af3b8544 | 3640 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3641 | { |
deaf386e | 3642 | return false; |
933e312e AM |
3643 | } |
3644 | ||
3645 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3646 | ||
76cd6173 | 3647 | noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
af3b8544 BP |
3648 | { |
3649 | return __should_fail_alloc_page(gfp_mask, order); | |
3650 | } | |
3651 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3652 | ||
f27ce0e1 JK |
3653 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3654 | unsigned int order, unsigned int alloc_flags) | |
3655 | { | |
3656 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); | |
3657 | long unusable_free = (1 << order) - 1; | |
3658 | ||
3659 | /* | |
3660 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3661 | * the high-atomic reserves. This will over-estimate the size of the | |
3662 | * atomic reserve but it avoids a search. | |
3663 | */ | |
3664 | if (likely(!alloc_harder)) | |
3665 | unusable_free += z->nr_reserved_highatomic; | |
3666 | ||
3667 | #ifdef CONFIG_CMA | |
3668 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3669 | if (!(alloc_flags & ALLOC_CMA)) | |
3670 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3671 | #endif | |
3672 | ||
3673 | return unusable_free; | |
3674 | } | |
3675 | ||
1da177e4 | 3676 | /* |
97a16fc8 MG |
3677 | * Return true if free base pages are above 'mark'. For high-order checks it |
3678 | * will return true of the order-0 watermark is reached and there is at least | |
3679 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3680 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3681 | */ |
86a294a8 | 3682 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3683 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3684 | long free_pages) |
1da177e4 | 3685 | { |
d23ad423 | 3686 | long min = mark; |
1da177e4 | 3687 | int o; |
cd04ae1e | 3688 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3689 | |
0aaa29a5 | 3690 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 3691 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 3692 | |
7fb1d9fc | 3693 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3694 | min -= min / 2; |
0aaa29a5 | 3695 | |
f27ce0e1 | 3696 | if (unlikely(alloc_harder)) { |
cd04ae1e MH |
3697 | /* |
3698 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3699 | * users on the grounds that it's definitely going to be in | |
3700 | * the exit path shortly and free memory. Any allocation it | |
3701 | * makes during the free path will be small and short-lived. | |
3702 | */ | |
3703 | if (alloc_flags & ALLOC_OOM) | |
3704 | min -= min / 2; | |
3705 | else | |
3706 | min -= min / 4; | |
3707 | } | |
3708 | ||
97a16fc8 MG |
3709 | /* |
3710 | * Check watermarks for an order-0 allocation request. If these | |
3711 | * are not met, then a high-order request also cannot go ahead | |
3712 | * even if a suitable page happened to be free. | |
3713 | */ | |
97a225e6 | 3714 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 3715 | return false; |
1da177e4 | 3716 | |
97a16fc8 MG |
3717 | /* If this is an order-0 request then the watermark is fine */ |
3718 | if (!order) | |
3719 | return true; | |
3720 | ||
3721 | /* For a high-order request, check at least one suitable page is free */ | |
3722 | for (o = order; o < MAX_ORDER; o++) { | |
3723 | struct free_area *area = &z->free_area[o]; | |
3724 | int mt; | |
3725 | ||
3726 | if (!area->nr_free) | |
3727 | continue; | |
3728 | ||
97a16fc8 | 3729 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3730 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3731 | return true; |
3732 | } | |
3733 | ||
3734 | #ifdef CONFIG_CMA | |
d883c6cf | 3735 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3736 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3737 | return true; |
d883c6cf | 3738 | } |
97a16fc8 | 3739 | #endif |
76089d00 | 3740 | if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC)) |
b050e376 | 3741 | return true; |
1da177e4 | 3742 | } |
97a16fc8 | 3743 | return false; |
88f5acf8 MG |
3744 | } |
3745 | ||
7aeb09f9 | 3746 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3747 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 3748 | { |
97a225e6 | 3749 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
3750 | zone_page_state(z, NR_FREE_PAGES)); |
3751 | } | |
3752 | ||
48ee5f36 | 3753 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 3754 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 3755 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 3756 | { |
f27ce0e1 | 3757 | long free_pages; |
d883c6cf | 3758 | |
f27ce0e1 | 3759 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
3760 | |
3761 | /* | |
3762 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 3763 | * need to be calculated. |
48ee5f36 | 3764 | */ |
f27ce0e1 JK |
3765 | if (!order) { |
3766 | long fast_free; | |
3767 | ||
3768 | fast_free = free_pages; | |
3769 | fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags); | |
3770 | if (fast_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
3771 | return true; | |
3772 | } | |
48ee5f36 | 3773 | |
f80b08fc CTR |
3774 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
3775 | free_pages)) | |
3776 | return true; | |
3777 | /* | |
3778 | * Ignore watermark boosting for GFP_ATOMIC order-0 allocations | |
3779 | * when checking the min watermark. The min watermark is the | |
3780 | * point where boosting is ignored so that kswapd is woken up | |
3781 | * when below the low watermark. | |
3782 | */ | |
3783 | if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost | |
3784 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { | |
3785 | mark = z->_watermark[WMARK_MIN]; | |
3786 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
3787 | alloc_flags, free_pages); | |
3788 | } | |
3789 | ||
3790 | return false; | |
48ee5f36 MG |
3791 | } |
3792 | ||
7aeb09f9 | 3793 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 3794 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
3795 | { |
3796 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3797 | ||
3798 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3799 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3800 | ||
97a225e6 | 3801 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 3802 | free_pages); |
1da177e4 LT |
3803 | } |
3804 | ||
9276b1bc | 3805 | #ifdef CONFIG_NUMA |
957f822a DR |
3806 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3807 | { | |
e02dc017 | 3808 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3809 | node_reclaim_distance; |
957f822a | 3810 | } |
9276b1bc | 3811 | #else /* CONFIG_NUMA */ |
957f822a DR |
3812 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3813 | { | |
3814 | return true; | |
3815 | } | |
9276b1bc PJ |
3816 | #endif /* CONFIG_NUMA */ |
3817 | ||
6bb15450 MG |
3818 | /* |
3819 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3820 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3821 | * premature use of a lower zone may cause lowmem pressure problems that | |
3822 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3823 | * probably too small. It only makes sense to spread allocations to avoid | |
3824 | * fragmentation between the Normal and DMA32 zones. | |
3825 | */ | |
3826 | static inline unsigned int | |
0a79cdad | 3827 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3828 | { |
736838e9 | 3829 | unsigned int alloc_flags; |
0a79cdad | 3830 | |
736838e9 MN |
3831 | /* |
3832 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
3833 | * to save a branch. | |
3834 | */ | |
3835 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
3836 | |
3837 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3838 | if (!zone) |
3839 | return alloc_flags; | |
3840 | ||
6bb15450 | 3841 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3842 | return alloc_flags; |
6bb15450 MG |
3843 | |
3844 | /* | |
3845 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3846 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3847 | * on UMA that if Normal is populated then so is DMA32. | |
3848 | */ | |
3849 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3850 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3851 | return alloc_flags; |
6bb15450 | 3852 | |
8118b82e | 3853 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3854 | #endif /* CONFIG_ZONE_DMA32 */ |
3855 | return alloc_flags; | |
6bb15450 | 3856 | } |
6bb15450 | 3857 | |
8510e69c JK |
3858 | static inline unsigned int current_alloc_flags(gfp_t gfp_mask, |
3859 | unsigned int alloc_flags) | |
3860 | { | |
3861 | #ifdef CONFIG_CMA | |
3862 | unsigned int pflags = current->flags; | |
3863 | ||
3864 | if (!(pflags & PF_MEMALLOC_NOCMA) && | |
3865 | gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
3866 | alloc_flags |= ALLOC_CMA; | |
3867 | ||
3868 | #endif | |
3869 | return alloc_flags; | |
3870 | } | |
3871 | ||
7fb1d9fc | 3872 | /* |
0798e519 | 3873 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3874 | * a page. |
3875 | */ | |
3876 | static struct page * | |
a9263751 VB |
3877 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3878 | const struct alloc_context *ac) | |
753ee728 | 3879 | { |
6bb15450 | 3880 | struct zoneref *z; |
5117f45d | 3881 | struct zone *zone; |
3b8c0be4 | 3882 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
6bb15450 | 3883 | bool no_fallback; |
3b8c0be4 | 3884 | |
6bb15450 | 3885 | retry: |
7fb1d9fc | 3886 | /* |
9276b1bc | 3887 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3888 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3889 | */ |
6bb15450 MG |
3890 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3891 | z = ac->preferred_zoneref; | |
30d8ec73 MN |
3892 | for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx, |
3893 | ac->nodemask) { | |
be06af00 | 3894 | struct page *page; |
e085dbc5 JW |
3895 | unsigned long mark; |
3896 | ||
664eedde MG |
3897 | if (cpusets_enabled() && |
3898 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3899 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3900 | continue; |
a756cf59 JW |
3901 | /* |
3902 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3903 | * want to get it from a node that is within its dirty |
3904 | * limit, such that no single node holds more than its | |
a756cf59 | 3905 | * proportional share of globally allowed dirty pages. |
281e3726 | 3906 | * The dirty limits take into account the node's |
a756cf59 JW |
3907 | * lowmem reserves and high watermark so that kswapd |
3908 | * should be able to balance it without having to | |
3909 | * write pages from its LRU list. | |
3910 | * | |
a756cf59 | 3911 | * XXX: For now, allow allocations to potentially |
281e3726 | 3912 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3913 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3914 | * which is important when on a NUMA setup the allowed |
281e3726 | 3915 | * nodes are together not big enough to reach the |
a756cf59 | 3916 | * global limit. The proper fix for these situations |
281e3726 | 3917 | * will require awareness of nodes in the |
a756cf59 JW |
3918 | * dirty-throttling and the flusher threads. |
3919 | */ | |
3b8c0be4 MG |
3920 | if (ac->spread_dirty_pages) { |
3921 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3922 | continue; | |
3923 | ||
3924 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3925 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3926 | continue; | |
3927 | } | |
3928 | } | |
7fb1d9fc | 3929 | |
6bb15450 MG |
3930 | if (no_fallback && nr_online_nodes > 1 && |
3931 | zone != ac->preferred_zoneref->zone) { | |
3932 | int local_nid; | |
3933 | ||
3934 | /* | |
3935 | * If moving to a remote node, retry but allow | |
3936 | * fragmenting fallbacks. Locality is more important | |
3937 | * than fragmentation avoidance. | |
3938 | */ | |
3939 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3940 | if (zone_to_nid(zone) != local_nid) { | |
3941 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3942 | goto retry; | |
3943 | } | |
3944 | } | |
3945 | ||
a9214443 | 3946 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 3947 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
3948 | ac->highest_zoneidx, alloc_flags, |
3949 | gfp_mask)) { | |
fa5e084e MG |
3950 | int ret; |
3951 | ||
c9e97a19 PT |
3952 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3953 | /* | |
3954 | * Watermark failed for this zone, but see if we can | |
3955 | * grow this zone if it contains deferred pages. | |
3956 | */ | |
3957 | if (static_branch_unlikely(&deferred_pages)) { | |
3958 | if (_deferred_grow_zone(zone, order)) | |
3959 | goto try_this_zone; | |
3960 | } | |
3961 | #endif | |
5dab2911 MG |
3962 | /* Checked here to keep the fast path fast */ |
3963 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3964 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3965 | goto try_this_zone; | |
3966 | ||
a5f5f91d | 3967 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3968 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3969 | continue; |
3970 | ||
a5f5f91d | 3971 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3972 | switch (ret) { |
a5f5f91d | 3973 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3974 | /* did not scan */ |
cd38b115 | 3975 | continue; |
a5f5f91d | 3976 | case NODE_RECLAIM_FULL: |
fa5e084e | 3977 | /* scanned but unreclaimable */ |
cd38b115 | 3978 | continue; |
fa5e084e MG |
3979 | default: |
3980 | /* did we reclaim enough */ | |
fed2719e | 3981 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 3982 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
3983 | goto try_this_zone; |
3984 | ||
fed2719e | 3985 | continue; |
0798e519 | 3986 | } |
7fb1d9fc RS |
3987 | } |
3988 | ||
fa5e084e | 3989 | try_this_zone: |
066b2393 | 3990 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3991 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3992 | if (page) { |
479f854a | 3993 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3994 | |
3995 | /* | |
3996 | * If this is a high-order atomic allocation then check | |
3997 | * if the pageblock should be reserved for the future | |
3998 | */ | |
3999 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
4000 | reserve_highatomic_pageblock(page, zone, order); | |
4001 | ||
75379191 | 4002 | return page; |
c9e97a19 PT |
4003 | } else { |
4004 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
4005 | /* Try again if zone has deferred pages */ | |
4006 | if (static_branch_unlikely(&deferred_pages)) { | |
4007 | if (_deferred_grow_zone(zone, order)) | |
4008 | goto try_this_zone; | |
4009 | } | |
4010 | #endif | |
75379191 | 4011 | } |
54a6eb5c | 4012 | } |
9276b1bc | 4013 | |
6bb15450 MG |
4014 | /* |
4015 | * It's possible on a UMA machine to get through all zones that are | |
4016 | * fragmented. If avoiding fragmentation, reset and try again. | |
4017 | */ | |
4018 | if (no_fallback) { | |
4019 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
4020 | goto retry; | |
4021 | } | |
4022 | ||
4ffeaf35 | 4023 | return NULL; |
753ee728 MH |
4024 | } |
4025 | ||
9af744d7 | 4026 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 4027 | { |
a238ab5b | 4028 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
4029 | |
4030 | /* | |
4031 | * This documents exceptions given to allocations in certain | |
4032 | * contexts that are allowed to allocate outside current's set | |
4033 | * of allowed nodes. | |
4034 | */ | |
4035 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 4036 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
4037 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
4038 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 4039 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
4040 | filter &= ~SHOW_MEM_FILTER_NODES; |
4041 | ||
9af744d7 | 4042 | show_mem(filter, nodemask); |
aa187507 MH |
4043 | } |
4044 | ||
a8e99259 | 4045 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
4046 | { |
4047 | struct va_format vaf; | |
4048 | va_list args; | |
1be334e5 | 4049 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 4050 | |
0f7896f1 | 4051 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
4052 | return; |
4053 | ||
7877cdcc MH |
4054 | va_start(args, fmt); |
4055 | vaf.fmt = fmt; | |
4056 | vaf.va = &args; | |
ef8444ea | 4057 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
4058 | current->comm, &vaf, gfp_mask, &gfp_mask, |
4059 | nodemask_pr_args(nodemask)); | |
7877cdcc | 4060 | va_end(args); |
3ee9a4f0 | 4061 | |
a8e99259 | 4062 | cpuset_print_current_mems_allowed(); |
ef8444ea | 4063 | pr_cont("\n"); |
a238ab5b | 4064 | dump_stack(); |
685dbf6f | 4065 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
4066 | } |
4067 | ||
6c18ba7a MH |
4068 | static inline struct page * |
4069 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
4070 | unsigned int alloc_flags, | |
4071 | const struct alloc_context *ac) | |
4072 | { | |
4073 | struct page *page; | |
4074 | ||
4075 | page = get_page_from_freelist(gfp_mask, order, | |
4076 | alloc_flags|ALLOC_CPUSET, ac); | |
4077 | /* | |
4078 | * fallback to ignore cpuset restriction if our nodes | |
4079 | * are depleted | |
4080 | */ | |
4081 | if (!page) | |
4082 | page = get_page_from_freelist(gfp_mask, order, | |
4083 | alloc_flags, ac); | |
4084 | ||
4085 | return page; | |
4086 | } | |
4087 | ||
11e33f6a MG |
4088 | static inline struct page * |
4089 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4090 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 4091 | { |
6e0fc46d DR |
4092 | struct oom_control oc = { |
4093 | .zonelist = ac->zonelist, | |
4094 | .nodemask = ac->nodemask, | |
2a966b77 | 4095 | .memcg = NULL, |
6e0fc46d DR |
4096 | .gfp_mask = gfp_mask, |
4097 | .order = order, | |
6e0fc46d | 4098 | }; |
11e33f6a MG |
4099 | struct page *page; |
4100 | ||
9879de73 JW |
4101 | *did_some_progress = 0; |
4102 | ||
9879de73 | 4103 | /* |
dc56401f JW |
4104 | * Acquire the oom lock. If that fails, somebody else is |
4105 | * making progress for us. | |
9879de73 | 4106 | */ |
dc56401f | 4107 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 4108 | *did_some_progress = 1; |
11e33f6a | 4109 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
4110 | return NULL; |
4111 | } | |
6b1de916 | 4112 | |
11e33f6a MG |
4113 | /* |
4114 | * Go through the zonelist yet one more time, keep very high watermark | |
4115 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
4116 | * we're still under heavy pressure. But make sure that this reclaim |
4117 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
4118 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 4119 | */ |
e746bf73 TH |
4120 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
4121 | ~__GFP_DIRECT_RECLAIM, order, | |
4122 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 4123 | if (page) |
11e33f6a MG |
4124 | goto out; |
4125 | ||
06ad276a MH |
4126 | /* Coredumps can quickly deplete all memory reserves */ |
4127 | if (current->flags & PF_DUMPCORE) | |
4128 | goto out; | |
4129 | /* The OOM killer will not help higher order allocs */ | |
4130 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4131 | goto out; | |
dcda9b04 MH |
4132 | /* |
4133 | * We have already exhausted all our reclaim opportunities without any | |
4134 | * success so it is time to admit defeat. We will skip the OOM killer | |
4135 | * because it is very likely that the caller has a more reasonable | |
4136 | * fallback than shooting a random task. | |
cfb4a541 MN |
4137 | * |
4138 | * The OOM killer may not free memory on a specific node. | |
dcda9b04 | 4139 | */ |
cfb4a541 | 4140 | if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE)) |
dcda9b04 | 4141 | goto out; |
06ad276a | 4142 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 4143 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
4144 | goto out; |
4145 | if (pm_suspended_storage()) | |
4146 | goto out; | |
4147 | /* | |
4148 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
4149 | * other request to make a forward progress. | |
4150 | * We are in an unfortunate situation where out_of_memory cannot | |
4151 | * do much for this context but let's try it to at least get | |
4152 | * access to memory reserved if the current task is killed (see | |
4153 | * out_of_memory). Once filesystems are ready to handle allocation | |
4154 | * failures more gracefully we should just bail out here. | |
4155 | */ | |
4156 | ||
3c2c6488 | 4157 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 4158 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 4159 | *did_some_progress = 1; |
5020e285 | 4160 | |
6c18ba7a MH |
4161 | /* |
4162 | * Help non-failing allocations by giving them access to memory | |
4163 | * reserves | |
4164 | */ | |
4165 | if (gfp_mask & __GFP_NOFAIL) | |
4166 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 4167 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 4168 | } |
11e33f6a | 4169 | out: |
dc56401f | 4170 | mutex_unlock(&oom_lock); |
11e33f6a MG |
4171 | return page; |
4172 | } | |
4173 | ||
33c2d214 MH |
4174 | /* |
4175 | * Maximum number of compaction retries wit a progress before OOM | |
4176 | * killer is consider as the only way to move forward. | |
4177 | */ | |
4178 | #define MAX_COMPACT_RETRIES 16 | |
4179 | ||
56de7263 MG |
4180 | #ifdef CONFIG_COMPACTION |
4181 | /* Try memory compaction for high-order allocations before reclaim */ | |
4182 | static struct page * | |
4183 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4184 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4185 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4186 | { |
5e1f0f09 | 4187 | struct page *page = NULL; |
eb414681 | 4188 | unsigned long pflags; |
499118e9 | 4189 | unsigned int noreclaim_flag; |
53853e2d VB |
4190 | |
4191 | if (!order) | |
66199712 | 4192 | return NULL; |
66199712 | 4193 | |
eb414681 | 4194 | psi_memstall_enter(&pflags); |
499118e9 | 4195 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 4196 | |
c5d01d0d | 4197 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 4198 | prio, &page); |
eb414681 | 4199 | |
499118e9 | 4200 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 4201 | psi_memstall_leave(&pflags); |
56de7263 | 4202 | |
98dd3b48 VB |
4203 | /* |
4204 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
4205 | * count a compaction stall | |
4206 | */ | |
4207 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 4208 | |
5e1f0f09 MG |
4209 | /* Prep a captured page if available */ |
4210 | if (page) | |
4211 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
4212 | ||
4213 | /* Try get a page from the freelist if available */ | |
4214 | if (!page) | |
4215 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 4216 | |
98dd3b48 VB |
4217 | if (page) { |
4218 | struct zone *zone = page_zone(page); | |
53853e2d | 4219 | |
98dd3b48 VB |
4220 | zone->compact_blockskip_flush = false; |
4221 | compaction_defer_reset(zone, order, true); | |
4222 | count_vm_event(COMPACTSUCCESS); | |
4223 | return page; | |
4224 | } | |
56de7263 | 4225 | |
98dd3b48 VB |
4226 | /* |
4227 | * It's bad if compaction run occurs and fails. The most likely reason | |
4228 | * is that pages exist, but not enough to satisfy watermarks. | |
4229 | */ | |
4230 | count_vm_event(COMPACTFAIL); | |
66199712 | 4231 | |
98dd3b48 | 4232 | cond_resched(); |
56de7263 MG |
4233 | |
4234 | return NULL; | |
4235 | } | |
33c2d214 | 4236 | |
3250845d VB |
4237 | static inline bool |
4238 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
4239 | enum compact_result compact_result, | |
4240 | enum compact_priority *compact_priority, | |
d9436498 | 4241 | int *compaction_retries) |
3250845d VB |
4242 | { |
4243 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 4244 | int min_priority; |
65190cff MH |
4245 | bool ret = false; |
4246 | int retries = *compaction_retries; | |
4247 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
4248 | |
4249 | if (!order) | |
4250 | return false; | |
4251 | ||
d9436498 VB |
4252 | if (compaction_made_progress(compact_result)) |
4253 | (*compaction_retries)++; | |
4254 | ||
3250845d VB |
4255 | /* |
4256 | * compaction considers all the zone as desperately out of memory | |
4257 | * so it doesn't really make much sense to retry except when the | |
4258 | * failure could be caused by insufficient priority | |
4259 | */ | |
d9436498 VB |
4260 | if (compaction_failed(compact_result)) |
4261 | goto check_priority; | |
3250845d | 4262 | |
49433085 VB |
4263 | /* |
4264 | * compaction was skipped because there are not enough order-0 pages | |
4265 | * to work with, so we retry only if it looks like reclaim can help. | |
4266 | */ | |
4267 | if (compaction_needs_reclaim(compact_result)) { | |
4268 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
4269 | goto out; | |
4270 | } | |
4271 | ||
3250845d VB |
4272 | /* |
4273 | * make sure the compaction wasn't deferred or didn't bail out early | |
4274 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
4275 | * But the next retry should use a higher priority if allowed, so |
4276 | * we don't just keep bailing out endlessly. | |
3250845d | 4277 | */ |
65190cff | 4278 | if (compaction_withdrawn(compact_result)) { |
49433085 | 4279 | goto check_priority; |
65190cff | 4280 | } |
3250845d VB |
4281 | |
4282 | /* | |
dcda9b04 | 4283 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
4284 | * costly ones because they are de facto nofail and invoke OOM |
4285 | * killer to move on while costly can fail and users are ready | |
4286 | * to cope with that. 1/4 retries is rather arbitrary but we | |
4287 | * would need much more detailed feedback from compaction to | |
4288 | * make a better decision. | |
4289 | */ | |
4290 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4291 | max_retries /= 4; | |
65190cff MH |
4292 | if (*compaction_retries <= max_retries) { |
4293 | ret = true; | |
4294 | goto out; | |
4295 | } | |
3250845d | 4296 | |
d9436498 VB |
4297 | /* |
4298 | * Make sure there are attempts at the highest priority if we exhausted | |
4299 | * all retries or failed at the lower priorities. | |
4300 | */ | |
4301 | check_priority: | |
c2033b00 VB |
4302 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
4303 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 4304 | |
c2033b00 | 4305 | if (*compact_priority > min_priority) { |
d9436498 VB |
4306 | (*compact_priority)--; |
4307 | *compaction_retries = 0; | |
65190cff | 4308 | ret = true; |
d9436498 | 4309 | } |
65190cff MH |
4310 | out: |
4311 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4312 | return ret; | |
3250845d | 4313 | } |
56de7263 MG |
4314 | #else |
4315 | static inline struct page * | |
4316 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4317 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4318 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4319 | { |
33c2d214 | 4320 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4321 | return NULL; |
4322 | } | |
33c2d214 MH |
4323 | |
4324 | static inline bool | |
86a294a8 MH |
4325 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4326 | enum compact_result compact_result, | |
a5508cd8 | 4327 | enum compact_priority *compact_priority, |
d9436498 | 4328 | int *compaction_retries) |
33c2d214 | 4329 | { |
31e49bfd MH |
4330 | struct zone *zone; |
4331 | struct zoneref *z; | |
4332 | ||
4333 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4334 | return false; | |
4335 | ||
4336 | /* | |
4337 | * There are setups with compaction disabled which would prefer to loop | |
4338 | * inside the allocator rather than hit the oom killer prematurely. | |
4339 | * Let's give them a good hope and keep retrying while the order-0 | |
4340 | * watermarks are OK. | |
4341 | */ | |
97a225e6 JK |
4342 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4343 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 4344 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 4345 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
4346 | return true; |
4347 | } | |
33c2d214 MH |
4348 | return false; |
4349 | } | |
3250845d | 4350 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4351 | |
d92a8cfc | 4352 | #ifdef CONFIG_LOCKDEP |
93781325 | 4353 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4354 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4355 | ||
f920e413 | 4356 | static bool __need_reclaim(gfp_t gfp_mask) |
d92a8cfc | 4357 | { |
d92a8cfc PZ |
4358 | /* no reclaim without waiting on it */ |
4359 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4360 | return false; | |
4361 | ||
4362 | /* this guy won't enter reclaim */ | |
2e517d68 | 4363 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4364 | return false; |
4365 | ||
d92a8cfc PZ |
4366 | if (gfp_mask & __GFP_NOLOCKDEP) |
4367 | return false; | |
4368 | ||
4369 | return true; | |
4370 | } | |
4371 | ||
93781325 OS |
4372 | void __fs_reclaim_acquire(void) |
4373 | { | |
4374 | lock_map_acquire(&__fs_reclaim_map); | |
4375 | } | |
4376 | ||
4377 | void __fs_reclaim_release(void) | |
4378 | { | |
4379 | lock_map_release(&__fs_reclaim_map); | |
4380 | } | |
4381 | ||
d92a8cfc PZ |
4382 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4383 | { | |
f920e413 DV |
4384 | gfp_mask = current_gfp_context(gfp_mask); |
4385 | ||
4386 | if (__need_reclaim(gfp_mask)) { | |
4387 | if (gfp_mask & __GFP_FS) | |
4388 | __fs_reclaim_acquire(); | |
4389 | ||
4390 | #ifdef CONFIG_MMU_NOTIFIER | |
4391 | lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); | |
4392 | lock_map_release(&__mmu_notifier_invalidate_range_start_map); | |
4393 | #endif | |
4394 | ||
4395 | } | |
d92a8cfc PZ |
4396 | } |
4397 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4398 | ||
4399 | void fs_reclaim_release(gfp_t gfp_mask) | |
4400 | { | |
f920e413 DV |
4401 | gfp_mask = current_gfp_context(gfp_mask); |
4402 | ||
4403 | if (__need_reclaim(gfp_mask)) { | |
4404 | if (gfp_mask & __GFP_FS) | |
4405 | __fs_reclaim_release(); | |
4406 | } | |
d92a8cfc PZ |
4407 | } |
4408 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4409 | #endif | |
4410 | ||
bba90710 | 4411 | /* Perform direct synchronous page reclaim */ |
2187e17b | 4412 | static unsigned long |
a9263751 VB |
4413 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4414 | const struct alloc_context *ac) | |
11e33f6a | 4415 | { |
499118e9 | 4416 | unsigned int noreclaim_flag; |
2187e17b | 4417 | unsigned long pflags, progress; |
11e33f6a MG |
4418 | |
4419 | cond_resched(); | |
4420 | ||
4421 | /* We now go into synchronous reclaim */ | |
4422 | cpuset_memory_pressure_bump(); | |
eb414681 | 4423 | psi_memstall_enter(&pflags); |
d92a8cfc | 4424 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4425 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4426 | |
a9263751 VB |
4427 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4428 | ac->nodemask); | |
11e33f6a | 4429 | |
499118e9 | 4430 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4431 | fs_reclaim_release(gfp_mask); |
eb414681 | 4432 | psi_memstall_leave(&pflags); |
11e33f6a MG |
4433 | |
4434 | cond_resched(); | |
4435 | ||
bba90710 MS |
4436 | return progress; |
4437 | } | |
4438 | ||
4439 | /* The really slow allocator path where we enter direct reclaim */ | |
4440 | static inline struct page * | |
4441 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4442 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4443 | unsigned long *did_some_progress) |
bba90710 MS |
4444 | { |
4445 | struct page *page = NULL; | |
4446 | bool drained = false; | |
4447 | ||
a9263751 | 4448 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
4449 | if (unlikely(!(*did_some_progress))) |
4450 | return NULL; | |
11e33f6a | 4451 | |
9ee493ce | 4452 | retry: |
31a6c190 | 4453 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4454 | |
4455 | /* | |
4456 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 | 4457 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
047b9967 | 4458 | * Shrink them and try again |
9ee493ce MG |
4459 | */ |
4460 | if (!page && !drained) { | |
29fac03b | 4461 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4462 | drain_all_pages(NULL); |
9ee493ce MG |
4463 | drained = true; |
4464 | goto retry; | |
4465 | } | |
4466 | ||
11e33f6a MG |
4467 | return page; |
4468 | } | |
4469 | ||
5ecd9d40 DR |
4470 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4471 | const struct alloc_context *ac) | |
3a025760 JW |
4472 | { |
4473 | struct zoneref *z; | |
4474 | struct zone *zone; | |
e1a55637 | 4475 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 4476 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 4477 | |
97a225e6 | 4478 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 4479 | ac->nodemask) { |
e1a55637 | 4480 | if (last_pgdat != zone->zone_pgdat) |
97a225e6 | 4481 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
e1a55637 MG |
4482 | last_pgdat = zone->zone_pgdat; |
4483 | } | |
3a025760 JW |
4484 | } |
4485 | ||
c603844b | 4486 | static inline unsigned int |
341ce06f PZ |
4487 | gfp_to_alloc_flags(gfp_t gfp_mask) |
4488 | { | |
c603844b | 4489 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4490 | |
736838e9 MN |
4491 | /* |
4492 | * __GFP_HIGH is assumed to be the same as ALLOC_HIGH | |
4493 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4494 | * to save two branches. | |
4495 | */ | |
e6223a3b | 4496 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
736838e9 | 4497 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 4498 | |
341ce06f PZ |
4499 | /* |
4500 | * The caller may dip into page reserves a bit more if the caller | |
4501 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4502 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 4503 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 4504 | */ |
736838e9 MN |
4505 | alloc_flags |= (__force int) |
4506 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 4507 | |
d0164adc | 4508 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 4509 | /* |
b104a35d DR |
4510 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4511 | * if it can't schedule. | |
5c3240d9 | 4512 | */ |
b104a35d | 4513 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 4514 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 4515 | /* |
b104a35d | 4516 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 4517 | * comment for __cpuset_node_allowed(). |
523b9458 | 4518 | */ |
341ce06f | 4519 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 4520 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
4521 | alloc_flags |= ALLOC_HARDER; |
4522 | ||
8510e69c JK |
4523 | alloc_flags = current_alloc_flags(gfp_mask, alloc_flags); |
4524 | ||
341ce06f PZ |
4525 | return alloc_flags; |
4526 | } | |
4527 | ||
cd04ae1e | 4528 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4529 | { |
cd04ae1e MH |
4530 | if (!tsk_is_oom_victim(tsk)) |
4531 | return false; | |
4532 | ||
4533 | /* | |
4534 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4535 | * only to the thread with TIF_MEMDIE set | |
4536 | */ | |
4537 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4538 | return false; |
4539 | ||
cd04ae1e MH |
4540 | return true; |
4541 | } | |
4542 | ||
4543 | /* | |
4544 | * Distinguish requests which really need access to full memory | |
4545 | * reserves from oom victims which can live with a portion of it | |
4546 | */ | |
4547 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4548 | { | |
4549 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4550 | return 0; | |
31a6c190 | 4551 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4552 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4553 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4554 | return ALLOC_NO_WATERMARKS; |
4555 | if (!in_interrupt()) { | |
4556 | if (current->flags & PF_MEMALLOC) | |
4557 | return ALLOC_NO_WATERMARKS; | |
4558 | else if (oom_reserves_allowed(current)) | |
4559 | return ALLOC_OOM; | |
4560 | } | |
31a6c190 | 4561 | |
cd04ae1e MH |
4562 | return 0; |
4563 | } | |
4564 | ||
4565 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4566 | { | |
4567 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4568 | } |
4569 | ||
0a0337e0 MH |
4570 | /* |
4571 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4572 | * for the given allocation request. | |
491d79ae JW |
4573 | * |
4574 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4575 | * without success, or when we couldn't even meet the watermark if we | |
4576 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4577 | * |
4578 | * Returns true if a retry is viable or false to enter the oom path. | |
4579 | */ | |
4580 | static inline bool | |
4581 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4582 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4583 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4584 | { |
4585 | struct zone *zone; | |
4586 | struct zoneref *z; | |
15f570bf | 4587 | bool ret = false; |
0a0337e0 | 4588 | |
423b452e VB |
4589 | /* |
4590 | * Costly allocations might have made a progress but this doesn't mean | |
4591 | * their order will become available due to high fragmentation so | |
4592 | * always increment the no progress counter for them | |
4593 | */ | |
4594 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4595 | *no_progress_loops = 0; | |
4596 | else | |
4597 | (*no_progress_loops)++; | |
4598 | ||
0a0337e0 MH |
4599 | /* |
4600 | * Make sure we converge to OOM if we cannot make any progress | |
4601 | * several times in the row. | |
4602 | */ | |
04c8716f MK |
4603 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4604 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4605 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4606 | } |
0a0337e0 | 4607 | |
bca67592 MG |
4608 | /* |
4609 | * Keep reclaiming pages while there is a chance this will lead | |
4610 | * somewhere. If none of the target zones can satisfy our allocation | |
4611 | * request even if all reclaimable pages are considered then we are | |
4612 | * screwed and have to go OOM. | |
0a0337e0 | 4613 | */ |
97a225e6 JK |
4614 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4615 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4616 | unsigned long available; |
ede37713 | 4617 | unsigned long reclaimable; |
d379f01d MH |
4618 | unsigned long min_wmark = min_wmark_pages(zone); |
4619 | bool wmark; | |
0a0337e0 | 4620 | |
5a1c84b4 | 4621 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4622 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4623 | |
4624 | /* | |
491d79ae JW |
4625 | * Would the allocation succeed if we reclaimed all |
4626 | * reclaimable pages? | |
0a0337e0 | 4627 | */ |
d379f01d | 4628 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 4629 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
4630 | trace_reclaim_retry_zone(z, order, reclaimable, |
4631 | available, min_wmark, *no_progress_loops, wmark); | |
4632 | if (wmark) { | |
ede37713 MH |
4633 | /* |
4634 | * If we didn't make any progress and have a lot of | |
4635 | * dirty + writeback pages then we should wait for | |
4636 | * an IO to complete to slow down the reclaim and | |
4637 | * prevent from pre mature OOM | |
4638 | */ | |
4639 | if (!did_some_progress) { | |
11fb9989 | 4640 | unsigned long write_pending; |
ede37713 | 4641 | |
5a1c84b4 MG |
4642 | write_pending = zone_page_state_snapshot(zone, |
4643 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 4644 | |
11fb9989 | 4645 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
4646 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
4647 | return true; | |
4648 | } | |
4649 | } | |
5a1c84b4 | 4650 | |
15f570bf MH |
4651 | ret = true; |
4652 | goto out; | |
0a0337e0 MH |
4653 | } |
4654 | } | |
4655 | ||
15f570bf MH |
4656 | out: |
4657 | /* | |
4658 | * Memory allocation/reclaim might be called from a WQ context and the | |
4659 | * current implementation of the WQ concurrency control doesn't | |
4660 | * recognize that a particular WQ is congested if the worker thread is | |
4661 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4662 | * here rather than calling cond_resched(). | |
4663 | */ | |
4664 | if (current->flags & PF_WQ_WORKER) | |
4665 | schedule_timeout_uninterruptible(1); | |
4666 | else | |
4667 | cond_resched(); | |
4668 | return ret; | |
0a0337e0 MH |
4669 | } |
4670 | ||
902b6281 VB |
4671 | static inline bool |
4672 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4673 | { | |
4674 | /* | |
4675 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4676 | * mempolicy don't intersect. This should be normally dealt with by | |
4677 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4678 | * such a way the check therein was true, and then it became false | |
4679 | * before we got our cpuset_mems_cookie here. | |
4680 | * This assumes that for all allocations, ac->nodemask can come only | |
4681 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4682 | * when it does not intersect with the cpuset restrictions) or the | |
4683 | * caller can deal with a violated nodemask. | |
4684 | */ | |
4685 | if (cpusets_enabled() && ac->nodemask && | |
4686 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4687 | ac->nodemask = NULL; | |
4688 | return true; | |
4689 | } | |
4690 | ||
4691 | /* | |
4692 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4693 | * possible to race with parallel threads in such a way that our | |
4694 | * allocation can fail while the mask is being updated. If we are about | |
4695 | * to fail, check if the cpuset changed during allocation and if so, | |
4696 | * retry. | |
4697 | */ | |
4698 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4699 | return true; | |
4700 | ||
4701 | return false; | |
4702 | } | |
4703 | ||
11e33f6a MG |
4704 | static inline struct page * |
4705 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4706 | struct alloc_context *ac) |
11e33f6a | 4707 | { |
d0164adc | 4708 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4709 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4710 | struct page *page = NULL; |
c603844b | 4711 | unsigned int alloc_flags; |
11e33f6a | 4712 | unsigned long did_some_progress; |
5ce9bfef | 4713 | enum compact_priority compact_priority; |
c5d01d0d | 4714 | enum compact_result compact_result; |
5ce9bfef VB |
4715 | int compaction_retries; |
4716 | int no_progress_loops; | |
5ce9bfef | 4717 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4718 | int reserve_flags; |
1da177e4 | 4719 | |
d0164adc MG |
4720 | /* |
4721 | * We also sanity check to catch abuse of atomic reserves being used by | |
4722 | * callers that are not in atomic context. | |
4723 | */ | |
4724 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4725 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4726 | gfp_mask &= ~__GFP_ATOMIC; | |
4727 | ||
5ce9bfef VB |
4728 | retry_cpuset: |
4729 | compaction_retries = 0; | |
4730 | no_progress_loops = 0; | |
4731 | compact_priority = DEF_COMPACT_PRIORITY; | |
4732 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4733 | |
4734 | /* | |
4735 | * The fast path uses conservative alloc_flags to succeed only until | |
4736 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4737 | * alloc_flags precisely. So we do that now. | |
4738 | */ | |
4739 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4740 | ||
e47483bc VB |
4741 | /* |
4742 | * We need to recalculate the starting point for the zonelist iterator | |
4743 | * because we might have used different nodemask in the fast path, or | |
4744 | * there was a cpuset modification and we are retrying - otherwise we | |
4745 | * could end up iterating over non-eligible zones endlessly. | |
4746 | */ | |
4747 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 4748 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
4749 | if (!ac->preferred_zoneref->zone) |
4750 | goto nopage; | |
4751 | ||
0a79cdad | 4752 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4753 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4754 | |
4755 | /* | |
4756 | * The adjusted alloc_flags might result in immediate success, so try | |
4757 | * that first | |
4758 | */ | |
4759 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4760 | if (page) | |
4761 | goto got_pg; | |
4762 | ||
a8161d1e VB |
4763 | /* |
4764 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4765 | * that we have enough base pages and don't need to reclaim. For non- |
4766 | * movable high-order allocations, do that as well, as compaction will | |
4767 | * try prevent permanent fragmentation by migrating from blocks of the | |
4768 | * same migratetype. | |
4769 | * Don't try this for allocations that are allowed to ignore | |
4770 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4771 | */ |
282722b0 VB |
4772 | if (can_direct_reclaim && |
4773 | (costly_order || | |
4774 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4775 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4776 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4777 | alloc_flags, ac, | |
a5508cd8 | 4778 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4779 | &compact_result); |
4780 | if (page) | |
4781 | goto got_pg; | |
4782 | ||
cc638f32 VB |
4783 | /* |
4784 | * Checks for costly allocations with __GFP_NORETRY, which | |
4785 | * includes some THP page fault allocations | |
4786 | */ | |
4787 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4788 | /* |
4789 | * If allocating entire pageblock(s) and compaction | |
4790 | * failed because all zones are below low watermarks | |
4791 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4792 | * order, fail immediately unless the allocator has |
4793 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4794 | * |
4795 | * Reclaim is | |
4796 | * - potentially very expensive because zones are far | |
4797 | * below their low watermarks or this is part of very | |
4798 | * bursty high order allocations, | |
4799 | * - not guaranteed to help because isolate_freepages() | |
4800 | * may not iterate over freed pages as part of its | |
4801 | * linear scan, and | |
4802 | * - unlikely to make entire pageblocks free on its | |
4803 | * own. | |
4804 | */ | |
4805 | if (compact_result == COMPACT_SKIPPED || | |
4806 | compact_result == COMPACT_DEFERRED) | |
4807 | goto nopage; | |
a8161d1e | 4808 | |
a8161d1e | 4809 | /* |
3eb2771b VB |
4810 | * Looks like reclaim/compaction is worth trying, but |
4811 | * sync compaction could be very expensive, so keep | |
25160354 | 4812 | * using async compaction. |
a8161d1e | 4813 | */ |
a5508cd8 | 4814 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4815 | } |
4816 | } | |
23771235 | 4817 | |
31a6c190 | 4818 | retry: |
23771235 | 4819 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4820 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4821 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4822 | |
cd04ae1e MH |
4823 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4824 | if (reserve_flags) | |
8510e69c | 4825 | alloc_flags = current_alloc_flags(gfp_mask, reserve_flags); |
23771235 | 4826 | |
e46e7b77 | 4827 | /* |
d6a24df0 VB |
4828 | * Reset the nodemask and zonelist iterators if memory policies can be |
4829 | * ignored. These allocations are high priority and system rather than | |
4830 | * user oriented. | |
e46e7b77 | 4831 | */ |
cd04ae1e | 4832 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4833 | ac->nodemask = NULL; |
e46e7b77 | 4834 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4835 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
4836 | } |
4837 | ||
23771235 | 4838 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4839 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4840 | if (page) |
4841 | goto got_pg; | |
1da177e4 | 4842 | |
d0164adc | 4843 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4844 | if (!can_direct_reclaim) |
1da177e4 LT |
4845 | goto nopage; |
4846 | ||
9a67f648 MH |
4847 | /* Avoid recursion of direct reclaim */ |
4848 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4849 | goto nopage; |
4850 | ||
a8161d1e VB |
4851 | /* Try direct reclaim and then allocating */ |
4852 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4853 | &did_some_progress); | |
4854 | if (page) | |
4855 | goto got_pg; | |
4856 | ||
4857 | /* Try direct compaction and then allocating */ | |
a9263751 | 4858 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4859 | compact_priority, &compact_result); |
56de7263 MG |
4860 | if (page) |
4861 | goto got_pg; | |
75f30861 | 4862 | |
9083905a JW |
4863 | /* Do not loop if specifically requested */ |
4864 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4865 | goto nopage; |
9083905a | 4866 | |
0a0337e0 MH |
4867 | /* |
4868 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4869 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4870 | */ |
dcda9b04 | 4871 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4872 | goto nopage; |
0a0337e0 | 4873 | |
0a0337e0 | 4874 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4875 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4876 | goto retry; |
4877 | ||
33c2d214 MH |
4878 | /* |
4879 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4880 | * reclaim is not able to make any progress because the current | |
4881 | * implementation of the compaction depends on the sufficient amount | |
4882 | * of free memory (see __compaction_suitable) | |
4883 | */ | |
4884 | if (did_some_progress > 0 && | |
86a294a8 | 4885 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4886 | compact_result, &compact_priority, |
d9436498 | 4887 | &compaction_retries)) |
33c2d214 MH |
4888 | goto retry; |
4889 | ||
902b6281 VB |
4890 | |
4891 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4892 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4893 | goto retry_cpuset; |
4894 | ||
9083905a JW |
4895 | /* Reclaim has failed us, start killing things */ |
4896 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4897 | if (page) | |
4898 | goto got_pg; | |
4899 | ||
9a67f648 | 4900 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e | 4901 | if (tsk_is_oom_victim(current) && |
8510e69c | 4902 | (alloc_flags & ALLOC_OOM || |
c288983d | 4903 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4904 | goto nopage; |
4905 | ||
9083905a | 4906 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4907 | if (did_some_progress) { |
4908 | no_progress_loops = 0; | |
9083905a | 4909 | goto retry; |
0a0337e0 | 4910 | } |
9083905a | 4911 | |
1da177e4 | 4912 | nopage: |
902b6281 VB |
4913 | /* Deal with possible cpuset update races before we fail */ |
4914 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4915 | goto retry_cpuset; |
4916 | ||
9a67f648 MH |
4917 | /* |
4918 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4919 | * we always retry | |
4920 | */ | |
4921 | if (gfp_mask & __GFP_NOFAIL) { | |
4922 | /* | |
4923 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4924 | * of any new users that actually require GFP_NOWAIT | |
4925 | */ | |
4926 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4927 | goto fail; | |
4928 | ||
4929 | /* | |
4930 | * PF_MEMALLOC request from this context is rather bizarre | |
4931 | * because we cannot reclaim anything and only can loop waiting | |
4932 | * for somebody to do a work for us | |
4933 | */ | |
4934 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4935 | ||
4936 | /* | |
4937 | * non failing costly orders are a hard requirement which we | |
4938 | * are not prepared for much so let's warn about these users | |
4939 | * so that we can identify them and convert them to something | |
4940 | * else. | |
4941 | */ | |
4942 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4943 | ||
6c18ba7a MH |
4944 | /* |
4945 | * Help non-failing allocations by giving them access to memory | |
4946 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4947 | * could deplete whole memory reserves which would just make | |
4948 | * the situation worse | |
4949 | */ | |
4950 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4951 | if (page) | |
4952 | goto got_pg; | |
4953 | ||
9a67f648 MH |
4954 | cond_resched(); |
4955 | goto retry; | |
4956 | } | |
4957 | fail: | |
a8e99259 | 4958 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4959 | "page allocation failure: order:%u", order); |
1da177e4 | 4960 | got_pg: |
072bb0aa | 4961 | return page; |
1da177e4 | 4962 | } |
11e33f6a | 4963 | |
9cd75558 | 4964 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4965 | int preferred_nid, nodemask_t *nodemask, |
8e6a930b | 4966 | struct alloc_context *ac, gfp_t *alloc_gfp, |
9cd75558 | 4967 | unsigned int *alloc_flags) |
11e33f6a | 4968 | { |
97a225e6 | 4969 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4970 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 4971 | ac->nodemask = nodemask; |
01c0bfe0 | 4972 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 4973 | |
682a3385 | 4974 | if (cpusets_enabled()) { |
8e6a930b | 4975 | *alloc_gfp |= __GFP_HARDWALL; |
182f3d7a MS |
4976 | /* |
4977 | * When we are in the interrupt context, it is irrelevant | |
4978 | * to the current task context. It means that any node ok. | |
4979 | */ | |
4980 | if (!in_interrupt() && !ac->nodemask) | |
9cd75558 | 4981 | ac->nodemask = &cpuset_current_mems_allowed; |
51047820 VB |
4982 | else |
4983 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4984 | } |
4985 | ||
d92a8cfc PZ |
4986 | fs_reclaim_acquire(gfp_mask); |
4987 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4988 | |
d0164adc | 4989 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4990 | |
4991 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4992 | return false; |
11e33f6a | 4993 | |
8510e69c | 4994 | *alloc_flags = current_alloc_flags(gfp_mask, *alloc_flags); |
d883c6cf | 4995 | |
c9ab0c4f | 4996 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4997 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4998 | |
e46e7b77 MG |
4999 | /* |
5000 | * The preferred zone is used for statistics but crucially it is | |
5001 | * also used as the starting point for the zonelist iterator. It | |
5002 | * may get reset for allocations that ignore memory policies. | |
5003 | */ | |
9cd75558 | 5004 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 5005 | ac->highest_zoneidx, ac->nodemask); |
a0622d05 MN |
5006 | |
5007 | return true; | |
9cd75558 MG |
5008 | } |
5009 | ||
387ba26f | 5010 | /* |
0f87d9d3 | 5011 | * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array |
387ba26f MG |
5012 | * @gfp: GFP flags for the allocation |
5013 | * @preferred_nid: The preferred NUMA node ID to allocate from | |
5014 | * @nodemask: Set of nodes to allocate from, may be NULL | |
0f87d9d3 MG |
5015 | * @nr_pages: The number of pages desired on the list or array |
5016 | * @page_list: Optional list to store the allocated pages | |
5017 | * @page_array: Optional array to store the pages | |
387ba26f MG |
5018 | * |
5019 | * This is a batched version of the page allocator that attempts to | |
0f87d9d3 MG |
5020 | * allocate nr_pages quickly. Pages are added to page_list if page_list |
5021 | * is not NULL, otherwise it is assumed that the page_array is valid. | |
387ba26f | 5022 | * |
0f87d9d3 MG |
5023 | * For lists, nr_pages is the number of pages that should be allocated. |
5024 | * | |
5025 | * For arrays, only NULL elements are populated with pages and nr_pages | |
5026 | * is the maximum number of pages that will be stored in the array. | |
5027 | * | |
5028 | * Returns the number of pages on the list or array. | |
387ba26f MG |
5029 | */ |
5030 | unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, | |
5031 | nodemask_t *nodemask, int nr_pages, | |
0f87d9d3 MG |
5032 | struct list_head *page_list, |
5033 | struct page **page_array) | |
387ba26f MG |
5034 | { |
5035 | struct page *page; | |
5036 | unsigned long flags; | |
5037 | struct zone *zone; | |
5038 | struct zoneref *z; | |
5039 | struct per_cpu_pages *pcp; | |
5040 | struct list_head *pcp_list; | |
5041 | struct alloc_context ac; | |
5042 | gfp_t alloc_gfp; | |
5043 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
0f87d9d3 | 5044 | int nr_populated = 0; |
387ba26f | 5045 | |
ce76f9a1 | 5046 | if (unlikely(nr_pages <= 0)) |
387ba26f MG |
5047 | return 0; |
5048 | ||
0f87d9d3 MG |
5049 | /* |
5050 | * Skip populated array elements to determine if any pages need | |
5051 | * to be allocated before disabling IRQs. | |
5052 | */ | |
5053 | while (page_array && page_array[nr_populated] && nr_populated < nr_pages) | |
5054 | nr_populated++; | |
5055 | ||
387ba26f | 5056 | /* Use the single page allocator for one page. */ |
0f87d9d3 | 5057 | if (nr_pages - nr_populated == 1) |
387ba26f MG |
5058 | goto failed; |
5059 | ||
5060 | /* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */ | |
5061 | gfp &= gfp_allowed_mask; | |
5062 | alloc_gfp = gfp; | |
5063 | if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags)) | |
5064 | return 0; | |
5065 | gfp = alloc_gfp; | |
5066 | ||
5067 | /* Find an allowed local zone that meets the low watermark. */ | |
5068 | for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) { | |
5069 | unsigned long mark; | |
5070 | ||
5071 | if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) && | |
5072 | !__cpuset_zone_allowed(zone, gfp)) { | |
5073 | continue; | |
5074 | } | |
5075 | ||
5076 | if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone && | |
5077 | zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) { | |
5078 | goto failed; | |
5079 | } | |
5080 | ||
5081 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages; | |
5082 | if (zone_watermark_fast(zone, 0, mark, | |
5083 | zonelist_zone_idx(ac.preferred_zoneref), | |
5084 | alloc_flags, gfp)) { | |
5085 | break; | |
5086 | } | |
5087 | } | |
5088 | ||
5089 | /* | |
5090 | * If there are no allowed local zones that meets the watermarks then | |
5091 | * try to allocate a single page and reclaim if necessary. | |
5092 | */ | |
ce76f9a1 | 5093 | if (unlikely(!zone)) |
387ba26f MG |
5094 | goto failed; |
5095 | ||
5096 | /* Attempt the batch allocation */ | |
5097 | local_irq_save(flags); | |
5098 | pcp = &this_cpu_ptr(zone->pageset)->pcp; | |
5099 | pcp_list = &pcp->lists[ac.migratetype]; | |
5100 | ||
0f87d9d3 MG |
5101 | while (nr_populated < nr_pages) { |
5102 | ||
5103 | /* Skip existing pages */ | |
5104 | if (page_array && page_array[nr_populated]) { | |
5105 | nr_populated++; | |
5106 | continue; | |
5107 | } | |
5108 | ||
387ba26f MG |
5109 | page = __rmqueue_pcplist(zone, ac.migratetype, alloc_flags, |
5110 | pcp, pcp_list); | |
ce76f9a1 | 5111 | if (unlikely(!page)) { |
387ba26f | 5112 | /* Try and get at least one page */ |
0f87d9d3 | 5113 | if (!nr_populated) |
387ba26f MG |
5114 | goto failed_irq; |
5115 | break; | |
5116 | } | |
5117 | ||
5118 | /* | |
5119 | * Ideally this would be batched but the best way to do | |
5120 | * that cheaply is to first convert zone_statistics to | |
5121 | * be inaccurate per-cpu counter like vm_events to avoid | |
5122 | * a RMW cycle then do the accounting with IRQs enabled. | |
5123 | */ | |
5124 | __count_zid_vm_events(PGALLOC, zone_idx(zone), 1); | |
5125 | zone_statistics(ac.preferred_zoneref->zone, zone); | |
5126 | ||
5127 | prep_new_page(page, 0, gfp, 0); | |
0f87d9d3 MG |
5128 | if (page_list) |
5129 | list_add(&page->lru, page_list); | |
5130 | else | |
5131 | page_array[nr_populated] = page; | |
5132 | nr_populated++; | |
387ba26f MG |
5133 | } |
5134 | ||
5135 | local_irq_restore(flags); | |
5136 | ||
0f87d9d3 | 5137 | return nr_populated; |
387ba26f MG |
5138 | |
5139 | failed_irq: | |
5140 | local_irq_restore(flags); | |
5141 | ||
5142 | failed: | |
5143 | page = __alloc_pages(gfp, 0, preferred_nid, nodemask); | |
5144 | if (page) { | |
0f87d9d3 MG |
5145 | if (page_list) |
5146 | list_add(&page->lru, page_list); | |
5147 | else | |
5148 | page_array[nr_populated] = page; | |
5149 | nr_populated++; | |
387ba26f MG |
5150 | } |
5151 | ||
0f87d9d3 | 5152 | return nr_populated; |
387ba26f MG |
5153 | } |
5154 | EXPORT_SYMBOL_GPL(__alloc_pages_bulk); | |
5155 | ||
9cd75558 MG |
5156 | /* |
5157 | * This is the 'heart' of the zoned buddy allocator. | |
5158 | */ | |
84172f4b | 5159 | struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, |
04ec6264 | 5160 | nodemask_t *nodemask) |
9cd75558 MG |
5161 | { |
5162 | struct page *page; | |
5163 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
8e6a930b | 5164 | gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
5165 | struct alloc_context ac = { }; |
5166 | ||
c63ae43b MH |
5167 | /* |
5168 | * There are several places where we assume that the order value is sane | |
5169 | * so bail out early if the request is out of bound. | |
5170 | */ | |
5171 | if (unlikely(order >= MAX_ORDER)) { | |
6e5e0f28 | 5172 | WARN_ON_ONCE(!(gfp & __GFP_NOWARN)); |
c63ae43b MH |
5173 | return NULL; |
5174 | } | |
5175 | ||
6e5e0f28 MWO |
5176 | gfp &= gfp_allowed_mask; |
5177 | alloc_gfp = gfp; | |
5178 | if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac, | |
8e6a930b | 5179 | &alloc_gfp, &alloc_flags)) |
9cd75558 MG |
5180 | return NULL; |
5181 | ||
6bb15450 MG |
5182 | /* |
5183 | * Forbid the first pass from falling back to types that fragment | |
5184 | * memory until all local zones are considered. | |
5185 | */ | |
6e5e0f28 | 5186 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp); |
6bb15450 | 5187 | |
5117f45d | 5188 | /* First allocation attempt */ |
8e6a930b | 5189 | page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac); |
4fcb0971 MG |
5190 | if (likely(page)) |
5191 | goto out; | |
11e33f6a | 5192 | |
4fcb0971 | 5193 | /* |
7dea19f9 MH |
5194 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
5195 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
5196 | * from a particular context which has been marked by | |
5197 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 5198 | */ |
6e5e0f28 | 5199 | alloc_gfp = current_gfp_context(gfp); |
4fcb0971 | 5200 | ac.spread_dirty_pages = false; |
23f086f9 | 5201 | |
4741526b MG |
5202 | /* |
5203 | * Restore the original nodemask if it was potentially replaced with | |
5204 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
5205 | */ | |
97ce86f9 | 5206 | ac.nodemask = nodemask; |
16096c25 | 5207 | |
8e6a930b | 5208 | page = __alloc_pages_slowpath(alloc_gfp, order, &ac); |
cc9a6c87 | 5209 | |
4fcb0971 | 5210 | out: |
6e5e0f28 MWO |
5211 | if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT) && page && |
5212 | unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) { | |
c4159a75 VD |
5213 | __free_pages(page, order); |
5214 | page = NULL; | |
4949148a VD |
5215 | } |
5216 | ||
8e6a930b | 5217 | trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype); |
4fcb0971 | 5218 | |
11e33f6a | 5219 | return page; |
1da177e4 | 5220 | } |
84172f4b | 5221 | EXPORT_SYMBOL(__alloc_pages); |
1da177e4 LT |
5222 | |
5223 | /* | |
9ea9a680 MH |
5224 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
5225 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
5226 | * you need to access high mem. | |
1da177e4 | 5227 | */ |
920c7a5d | 5228 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 5229 | { |
945a1113 AM |
5230 | struct page *page; |
5231 | ||
9ea9a680 | 5232 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
5233 | if (!page) |
5234 | return 0; | |
5235 | return (unsigned long) page_address(page); | |
5236 | } | |
1da177e4 LT |
5237 | EXPORT_SYMBOL(__get_free_pages); |
5238 | ||
920c7a5d | 5239 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 5240 | { |
945a1113 | 5241 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 5242 | } |
1da177e4 LT |
5243 | EXPORT_SYMBOL(get_zeroed_page); |
5244 | ||
742aa7fb | 5245 | static inline void free_the_page(struct page *page, unsigned int order) |
1da177e4 | 5246 | { |
742aa7fb AL |
5247 | if (order == 0) /* Via pcp? */ |
5248 | free_unref_page(page); | |
5249 | else | |
7fef431b | 5250 | __free_pages_ok(page, order, FPI_NONE); |
1da177e4 LT |
5251 | } |
5252 | ||
7f194fbb MWO |
5253 | /** |
5254 | * __free_pages - Free pages allocated with alloc_pages(). | |
5255 | * @page: The page pointer returned from alloc_pages(). | |
5256 | * @order: The order of the allocation. | |
5257 | * | |
5258 | * This function can free multi-page allocations that are not compound | |
5259 | * pages. It does not check that the @order passed in matches that of | |
5260 | * the allocation, so it is easy to leak memory. Freeing more memory | |
5261 | * than was allocated will probably emit a warning. | |
5262 | * | |
5263 | * If the last reference to this page is speculative, it will be released | |
5264 | * by put_page() which only frees the first page of a non-compound | |
5265 | * allocation. To prevent the remaining pages from being leaked, we free | |
5266 | * the subsequent pages here. If you want to use the page's reference | |
5267 | * count to decide when to free the allocation, you should allocate a | |
5268 | * compound page, and use put_page() instead of __free_pages(). | |
5269 | * | |
5270 | * Context: May be called in interrupt context or while holding a normal | |
5271 | * spinlock, but not in NMI context or while holding a raw spinlock. | |
5272 | */ | |
742aa7fb AL |
5273 | void __free_pages(struct page *page, unsigned int order) |
5274 | { | |
5275 | if (put_page_testzero(page)) | |
5276 | free_the_page(page, order); | |
e320d301 MWO |
5277 | else if (!PageHead(page)) |
5278 | while (order-- > 0) | |
5279 | free_the_page(page + (1 << order), order); | |
742aa7fb | 5280 | } |
1da177e4 LT |
5281 | EXPORT_SYMBOL(__free_pages); |
5282 | ||
920c7a5d | 5283 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
5284 | { |
5285 | if (addr != 0) { | |
725d704e | 5286 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
5287 | __free_pages(virt_to_page((void *)addr), order); |
5288 | } | |
5289 | } | |
5290 | ||
5291 | EXPORT_SYMBOL(free_pages); | |
5292 | ||
b63ae8ca AD |
5293 | /* |
5294 | * Page Fragment: | |
5295 | * An arbitrary-length arbitrary-offset area of memory which resides | |
5296 | * within a 0 or higher order page. Multiple fragments within that page | |
5297 | * are individually refcounted, in the page's reference counter. | |
5298 | * | |
5299 | * The page_frag functions below provide a simple allocation framework for | |
5300 | * page fragments. This is used by the network stack and network device | |
5301 | * drivers to provide a backing region of memory for use as either an | |
5302 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
5303 | */ | |
2976db80 AD |
5304 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
5305 | gfp_t gfp_mask) | |
b63ae8ca AD |
5306 | { |
5307 | struct page *page = NULL; | |
5308 | gfp_t gfp = gfp_mask; | |
5309 | ||
5310 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5311 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
5312 | __GFP_NOMEMALLOC; | |
5313 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
5314 | PAGE_FRAG_CACHE_MAX_ORDER); | |
5315 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
5316 | #endif | |
5317 | if (unlikely(!page)) | |
5318 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
5319 | ||
5320 | nc->va = page ? page_address(page) : NULL; | |
5321 | ||
5322 | return page; | |
5323 | } | |
5324 | ||
2976db80 | 5325 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
5326 | { |
5327 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
5328 | ||
742aa7fb AL |
5329 | if (page_ref_sub_and_test(page, count)) |
5330 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 5331 | } |
2976db80 | 5332 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 5333 | |
b358e212 KH |
5334 | void *page_frag_alloc_align(struct page_frag_cache *nc, |
5335 | unsigned int fragsz, gfp_t gfp_mask, | |
5336 | unsigned int align_mask) | |
b63ae8ca AD |
5337 | { |
5338 | unsigned int size = PAGE_SIZE; | |
5339 | struct page *page; | |
5340 | int offset; | |
5341 | ||
5342 | if (unlikely(!nc->va)) { | |
5343 | refill: | |
2976db80 | 5344 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
5345 | if (!page) |
5346 | return NULL; | |
5347 | ||
5348 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5349 | /* if size can vary use size else just use PAGE_SIZE */ | |
5350 | size = nc->size; | |
5351 | #endif | |
5352 | /* Even if we own the page, we do not use atomic_set(). | |
5353 | * This would break get_page_unless_zero() users. | |
5354 | */ | |
86447726 | 5355 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
5356 | |
5357 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 5358 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 5359 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5360 | nc->offset = size; |
5361 | } | |
5362 | ||
5363 | offset = nc->offset - fragsz; | |
5364 | if (unlikely(offset < 0)) { | |
5365 | page = virt_to_page(nc->va); | |
5366 | ||
fe896d18 | 5367 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
5368 | goto refill; |
5369 | ||
d8c19014 DZ |
5370 | if (unlikely(nc->pfmemalloc)) { |
5371 | free_the_page(page, compound_order(page)); | |
5372 | goto refill; | |
5373 | } | |
5374 | ||
b63ae8ca AD |
5375 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
5376 | /* if size can vary use size else just use PAGE_SIZE */ | |
5377 | size = nc->size; | |
5378 | #endif | |
5379 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 5380 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
5381 | |
5382 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 5383 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5384 | offset = size - fragsz; |
5385 | } | |
5386 | ||
5387 | nc->pagecnt_bias--; | |
b358e212 | 5388 | offset &= align_mask; |
b63ae8ca AD |
5389 | nc->offset = offset; |
5390 | ||
5391 | return nc->va + offset; | |
5392 | } | |
b358e212 | 5393 | EXPORT_SYMBOL(page_frag_alloc_align); |
b63ae8ca AD |
5394 | |
5395 | /* | |
5396 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
5397 | */ | |
8c2dd3e4 | 5398 | void page_frag_free(void *addr) |
b63ae8ca AD |
5399 | { |
5400 | struct page *page = virt_to_head_page(addr); | |
5401 | ||
742aa7fb AL |
5402 | if (unlikely(put_page_testzero(page))) |
5403 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 5404 | } |
8c2dd3e4 | 5405 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 5406 | |
d00181b9 KS |
5407 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
5408 | size_t size) | |
ee85c2e1 AK |
5409 | { |
5410 | if (addr) { | |
5411 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
5412 | unsigned long used = addr + PAGE_ALIGN(size); | |
5413 | ||
5414 | split_page(virt_to_page((void *)addr), order); | |
5415 | while (used < alloc_end) { | |
5416 | free_page(used); | |
5417 | used += PAGE_SIZE; | |
5418 | } | |
5419 | } | |
5420 | return (void *)addr; | |
5421 | } | |
5422 | ||
2be0ffe2 TT |
5423 | /** |
5424 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
5425 | * @size: the number of bytes to allocate | |
63931eb9 | 5426 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
5427 | * |
5428 | * This function is similar to alloc_pages(), except that it allocates the | |
5429 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
5430 | * allocate memory in power-of-two pages. | |
5431 | * | |
5432 | * This function is also limited by MAX_ORDER. | |
5433 | * | |
5434 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
5435 | * |
5436 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
5437 | */ |
5438 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
5439 | { | |
5440 | unsigned int order = get_order(size); | |
5441 | unsigned long addr; | |
5442 | ||
63931eb9 VB |
5443 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) |
5444 | gfp_mask &= ~__GFP_COMP; | |
5445 | ||
2be0ffe2 | 5446 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 5447 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
5448 | } |
5449 | EXPORT_SYMBOL(alloc_pages_exact); | |
5450 | ||
ee85c2e1 AK |
5451 | /** |
5452 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
5453 | * pages on a node. | |
b5e6ab58 | 5454 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 5455 | * @size: the number of bytes to allocate |
63931eb9 | 5456 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
5457 | * |
5458 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
5459 | * back. | |
a862f68a MR |
5460 | * |
5461 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 5462 | */ |
e1931811 | 5463 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5464 | { |
d00181b9 | 5465 | unsigned int order = get_order(size); |
63931eb9 VB |
5466 | struct page *p; |
5467 | ||
5468 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) | |
5469 | gfp_mask &= ~__GFP_COMP; | |
5470 | ||
5471 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5472 | if (!p) |
5473 | return NULL; | |
5474 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5475 | } | |
ee85c2e1 | 5476 | |
2be0ffe2 TT |
5477 | /** |
5478 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5479 | * @virt: the value returned by alloc_pages_exact. | |
5480 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5481 | * | |
5482 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5483 | */ | |
5484 | void free_pages_exact(void *virt, size_t size) | |
5485 | { | |
5486 | unsigned long addr = (unsigned long)virt; | |
5487 | unsigned long end = addr + PAGE_ALIGN(size); | |
5488 | ||
5489 | while (addr < end) { | |
5490 | free_page(addr); | |
5491 | addr += PAGE_SIZE; | |
5492 | } | |
5493 | } | |
5494 | EXPORT_SYMBOL(free_pages_exact); | |
5495 | ||
e0fb5815 ZY |
5496 | /** |
5497 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5498 | * @offset: The zone index of the highest zone | |
5499 | * | |
a862f68a | 5500 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5501 | * high watermark within all zones at or below a given zone index. For each |
5502 | * zone, the number of pages is calculated as: | |
0e056eb5 | 5503 | * |
5504 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5505 | * |
5506 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5507 | */ |
ebec3862 | 5508 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5509 | { |
dd1a239f | 5510 | struct zoneref *z; |
54a6eb5c MG |
5511 | struct zone *zone; |
5512 | ||
e310fd43 | 5513 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5514 | unsigned long sum = 0; |
1da177e4 | 5515 | |
0e88460d | 5516 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5517 | |
54a6eb5c | 5518 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5519 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5520 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5521 | if (size > high) |
5522 | sum += size - high; | |
1da177e4 LT |
5523 | } |
5524 | ||
5525 | return sum; | |
5526 | } | |
5527 | ||
e0fb5815 ZY |
5528 | /** |
5529 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5530 | * | |
5531 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5532 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5533 | * |
5534 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5535 | * ZONE_NORMAL. | |
1da177e4 | 5536 | */ |
ebec3862 | 5537 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5538 | { |
af4ca457 | 5539 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5540 | } |
c2f1a551 | 5541 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5542 | |
08e0f6a9 | 5543 | static inline void show_node(struct zone *zone) |
1da177e4 | 5544 | { |
e5adfffc | 5545 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5546 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5547 | } |
1da177e4 | 5548 | |
d02bd27b IR |
5549 | long si_mem_available(void) |
5550 | { | |
5551 | long available; | |
5552 | unsigned long pagecache; | |
5553 | unsigned long wmark_low = 0; | |
5554 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5555 | unsigned long reclaimable; |
d02bd27b IR |
5556 | struct zone *zone; |
5557 | int lru; | |
5558 | ||
5559 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5560 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5561 | |
5562 | for_each_zone(zone) | |
a9214443 | 5563 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5564 | |
5565 | /* | |
5566 | * Estimate the amount of memory available for userspace allocations, | |
5567 | * without causing swapping. | |
5568 | */ | |
c41f012a | 5569 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5570 | |
5571 | /* | |
5572 | * Not all the page cache can be freed, otherwise the system will | |
5573 | * start swapping. Assume at least half of the page cache, or the | |
5574 | * low watermark worth of cache, needs to stay. | |
5575 | */ | |
5576 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
5577 | pagecache -= min(pagecache / 2, wmark_low); | |
5578 | available += pagecache; | |
5579 | ||
5580 | /* | |
b29940c1 VB |
5581 | * Part of the reclaimable slab and other kernel memory consists of |
5582 | * items that are in use, and cannot be freed. Cap this estimate at the | |
5583 | * low watermark. | |
d02bd27b | 5584 | */ |
d42f3245 RG |
5585 | reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) + |
5586 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
b29940c1 | 5587 | available += reclaimable - min(reclaimable / 2, wmark_low); |
034ebf65 | 5588 | |
d02bd27b IR |
5589 | if (available < 0) |
5590 | available = 0; | |
5591 | return available; | |
5592 | } | |
5593 | EXPORT_SYMBOL_GPL(si_mem_available); | |
5594 | ||
1da177e4 LT |
5595 | void si_meminfo(struct sysinfo *val) |
5596 | { | |
ca79b0c2 | 5597 | val->totalram = totalram_pages(); |
11fb9989 | 5598 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 5599 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 5600 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 5601 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 5602 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
5603 | val->mem_unit = PAGE_SIZE; |
5604 | } | |
5605 | ||
5606 | EXPORT_SYMBOL(si_meminfo); | |
5607 | ||
5608 | #ifdef CONFIG_NUMA | |
5609 | void si_meminfo_node(struct sysinfo *val, int nid) | |
5610 | { | |
cdd91a77 JL |
5611 | int zone_type; /* needs to be signed */ |
5612 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
5613 | unsigned long managed_highpages = 0; |
5614 | unsigned long free_highpages = 0; | |
1da177e4 LT |
5615 | pg_data_t *pgdat = NODE_DATA(nid); |
5616 | ||
cdd91a77 | 5617 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 5618 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 5619 | val->totalram = managed_pages; |
11fb9989 | 5620 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 5621 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 5622 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
5623 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
5624 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
5625 | ||
5626 | if (is_highmem(zone)) { | |
9705bea5 | 5627 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
5628 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
5629 | } | |
5630 | } | |
5631 | val->totalhigh = managed_highpages; | |
5632 | val->freehigh = free_highpages; | |
98d2b0eb | 5633 | #else |
fc2bd799 JK |
5634 | val->totalhigh = managed_highpages; |
5635 | val->freehigh = free_highpages; | |
98d2b0eb | 5636 | #endif |
1da177e4 LT |
5637 | val->mem_unit = PAGE_SIZE; |
5638 | } | |
5639 | #endif | |
5640 | ||
ddd588b5 | 5641 | /* |
7bf02ea2 DR |
5642 | * Determine whether the node should be displayed or not, depending on whether |
5643 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 5644 | */ |
9af744d7 | 5645 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 5646 | { |
ddd588b5 | 5647 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 5648 | return false; |
ddd588b5 | 5649 | |
9af744d7 MH |
5650 | /* |
5651 | * no node mask - aka implicit memory numa policy. Do not bother with | |
5652 | * the synchronization - read_mems_allowed_begin - because we do not | |
5653 | * have to be precise here. | |
5654 | */ | |
5655 | if (!nodemask) | |
5656 | nodemask = &cpuset_current_mems_allowed; | |
5657 | ||
5658 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
5659 | } |
5660 | ||
1da177e4 LT |
5661 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
5662 | ||
377e4f16 RV |
5663 | static void show_migration_types(unsigned char type) |
5664 | { | |
5665 | static const char types[MIGRATE_TYPES] = { | |
5666 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 5667 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
5668 | [MIGRATE_RECLAIMABLE] = 'E', |
5669 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
5670 | #ifdef CONFIG_CMA |
5671 | [MIGRATE_CMA] = 'C', | |
5672 | #endif | |
194159fb | 5673 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 5674 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 5675 | #endif |
377e4f16 RV |
5676 | }; |
5677 | char tmp[MIGRATE_TYPES + 1]; | |
5678 | char *p = tmp; | |
5679 | int i; | |
5680 | ||
5681 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
5682 | if (type & (1 << i)) | |
5683 | *p++ = types[i]; | |
5684 | } | |
5685 | ||
5686 | *p = '\0'; | |
1f84a18f | 5687 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
5688 | } |
5689 | ||
1da177e4 LT |
5690 | /* |
5691 | * Show free area list (used inside shift_scroll-lock stuff) | |
5692 | * We also calculate the percentage fragmentation. We do this by counting the | |
5693 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
5694 | * |
5695 | * Bits in @filter: | |
5696 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
5697 | * cpuset. | |
1da177e4 | 5698 | */ |
9af744d7 | 5699 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 5700 | { |
d1bfcdb8 | 5701 | unsigned long free_pcp = 0; |
c7241913 | 5702 | int cpu; |
1da177e4 | 5703 | struct zone *zone; |
599d0c95 | 5704 | pg_data_t *pgdat; |
1da177e4 | 5705 | |
ee99c71c | 5706 | for_each_populated_zone(zone) { |
9af744d7 | 5707 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5708 | continue; |
d1bfcdb8 | 5709 | |
761b0677 KK |
5710 | for_each_online_cpu(cpu) |
5711 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
5712 | } |
5713 | ||
a731286d KM |
5714 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
5715 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
8d92890b | 5716 | " unevictable:%lu dirty:%lu writeback:%lu\n" |
d1bfcdb8 | 5717 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a | 5718 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 5719 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
5720 | global_node_page_state(NR_ACTIVE_ANON), |
5721 | global_node_page_state(NR_INACTIVE_ANON), | |
5722 | global_node_page_state(NR_ISOLATED_ANON), | |
5723 | global_node_page_state(NR_ACTIVE_FILE), | |
5724 | global_node_page_state(NR_INACTIVE_FILE), | |
5725 | global_node_page_state(NR_ISOLATED_FILE), | |
5726 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
5727 | global_node_page_state(NR_FILE_DIRTY), |
5728 | global_node_page_state(NR_WRITEBACK), | |
d42f3245 RG |
5729 | global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B), |
5730 | global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B), | |
50658e2e | 5731 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 5732 | global_node_page_state(NR_SHMEM), |
f0c0c115 | 5733 | global_node_page_state(NR_PAGETABLE), |
c41f012a MH |
5734 | global_zone_page_state(NR_BOUNCE), |
5735 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 5736 | free_pcp, |
c41f012a | 5737 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 5738 | |
599d0c95 | 5739 | for_each_online_pgdat(pgdat) { |
9af744d7 | 5740 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
5741 | continue; |
5742 | ||
599d0c95 MG |
5743 | printk("Node %d" |
5744 | " active_anon:%lukB" | |
5745 | " inactive_anon:%lukB" | |
5746 | " active_file:%lukB" | |
5747 | " inactive_file:%lukB" | |
5748 | " unevictable:%lukB" | |
5749 | " isolated(anon):%lukB" | |
5750 | " isolated(file):%lukB" | |
50658e2e | 5751 | " mapped:%lukB" |
11fb9989 MG |
5752 | " dirty:%lukB" |
5753 | " writeback:%lukB" | |
5754 | " shmem:%lukB" | |
5755 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5756 | " shmem_thp: %lukB" | |
5757 | " shmem_pmdmapped: %lukB" | |
5758 | " anon_thp: %lukB" | |
5759 | #endif | |
5760 | " writeback_tmp:%lukB" | |
991e7673 SB |
5761 | " kernel_stack:%lukB" |
5762 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5763 | " shadow_call_stack:%lukB" | |
5764 | #endif | |
f0c0c115 | 5765 | " pagetables:%lukB" |
599d0c95 MG |
5766 | " all_unreclaimable? %s" |
5767 | "\n", | |
5768 | pgdat->node_id, | |
5769 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
5770 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
5771 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
5772 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
5773 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
5774 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
5775 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 5776 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
5777 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
5778 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 5779 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 | 5780 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
57b2847d | 5781 | K(node_page_state(pgdat, NR_SHMEM_THPS)), |
a1528e21 | 5782 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)), |
69473e5d | 5783 | K(node_page_state(pgdat, NR_ANON_THPS)), |
11fb9989 | 5784 | #endif |
11fb9989 | 5785 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
991e7673 SB |
5786 | node_page_state(pgdat, NR_KERNEL_STACK_KB), |
5787 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5788 | node_page_state(pgdat, NR_KERNEL_SCS_KB), | |
5789 | #endif | |
f0c0c115 | 5790 | K(node_page_state(pgdat, NR_PAGETABLE)), |
c73322d0 JW |
5791 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
5792 | "yes" : "no"); | |
599d0c95 MG |
5793 | } |
5794 | ||
ee99c71c | 5795 | for_each_populated_zone(zone) { |
1da177e4 LT |
5796 | int i; |
5797 | ||
9af744d7 | 5798 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5799 | continue; |
d1bfcdb8 KK |
5800 | |
5801 | free_pcp = 0; | |
5802 | for_each_online_cpu(cpu) | |
5803 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
5804 | ||
1da177e4 | 5805 | show_node(zone); |
1f84a18f JP |
5806 | printk(KERN_CONT |
5807 | "%s" | |
1da177e4 LT |
5808 | " free:%lukB" |
5809 | " min:%lukB" | |
5810 | " low:%lukB" | |
5811 | " high:%lukB" | |
e47b346a | 5812 | " reserved_highatomic:%luKB" |
71c799f4 MK |
5813 | " active_anon:%lukB" |
5814 | " inactive_anon:%lukB" | |
5815 | " active_file:%lukB" | |
5816 | " inactive_file:%lukB" | |
5817 | " unevictable:%lukB" | |
5a1c84b4 | 5818 | " writepending:%lukB" |
1da177e4 | 5819 | " present:%lukB" |
9feedc9d | 5820 | " managed:%lukB" |
4a0aa73f | 5821 | " mlocked:%lukB" |
4a0aa73f | 5822 | " bounce:%lukB" |
d1bfcdb8 KK |
5823 | " free_pcp:%lukB" |
5824 | " local_pcp:%ukB" | |
d1ce749a | 5825 | " free_cma:%lukB" |
1da177e4 LT |
5826 | "\n", |
5827 | zone->name, | |
88f5acf8 | 5828 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
5829 | K(min_wmark_pages(zone)), |
5830 | K(low_wmark_pages(zone)), | |
5831 | K(high_wmark_pages(zone)), | |
e47b346a | 5832 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
5833 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
5834 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
5835 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
5836 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
5837 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 5838 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 5839 | K(zone->present_pages), |
9705bea5 | 5840 | K(zone_managed_pages(zone)), |
4a0aa73f | 5841 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f | 5842 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
5843 | K(free_pcp), |
5844 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 5845 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
5846 | printk("lowmem_reserve[]:"); |
5847 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
5848 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5849 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5850 | } |
5851 | ||
ee99c71c | 5852 | for_each_populated_zone(zone) { |
d00181b9 KS |
5853 | unsigned int order; |
5854 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5855 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5856 | |
9af744d7 | 5857 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5858 | continue; |
1da177e4 | 5859 | show_node(zone); |
1f84a18f | 5860 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5861 | |
5862 | spin_lock_irqsave(&zone->lock, flags); | |
5863 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5864 | struct free_area *area = &zone->free_area[order]; |
5865 | int type; | |
5866 | ||
5867 | nr[order] = area->nr_free; | |
8f9de51a | 5868 | total += nr[order] << order; |
377e4f16 RV |
5869 | |
5870 | types[order] = 0; | |
5871 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 5872 | if (!free_area_empty(area, type)) |
377e4f16 RV |
5873 | types[order] |= 1 << type; |
5874 | } | |
1da177e4 LT |
5875 | } |
5876 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5877 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5878 | printk(KERN_CONT "%lu*%lukB ", |
5879 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5880 | if (nr[order]) |
5881 | show_migration_types(types[order]); | |
5882 | } | |
1f84a18f | 5883 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5884 | } |
5885 | ||
949f7ec5 DR |
5886 | hugetlb_show_meminfo(); |
5887 | ||
11fb9989 | 5888 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5889 | |
1da177e4 LT |
5890 | show_swap_cache_info(); |
5891 | } | |
5892 | ||
19770b32 MG |
5893 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5894 | { | |
5895 | zoneref->zone = zone; | |
5896 | zoneref->zone_idx = zone_idx(zone); | |
5897 | } | |
5898 | ||
1da177e4 LT |
5899 | /* |
5900 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5901 | * |
5902 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5903 | */ |
9d3be21b | 5904 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5905 | { |
1a93205b | 5906 | struct zone *zone; |
bc732f1d | 5907 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5908 | int nr_zones = 0; |
02a68a5e CL |
5909 | |
5910 | do { | |
2f6726e5 | 5911 | zone_type--; |
070f8032 | 5912 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5913 | if (managed_zone(zone)) { |
9d3be21b | 5914 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5915 | check_highest_zone(zone_type); |
1da177e4 | 5916 | } |
2f6726e5 | 5917 | } while (zone_type); |
bc732f1d | 5918 | |
070f8032 | 5919 | return nr_zones; |
1da177e4 LT |
5920 | } |
5921 | ||
5922 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5923 | |
5924 | static int __parse_numa_zonelist_order(char *s) | |
5925 | { | |
c9bff3ee MH |
5926 | /* |
5927 | * We used to support different zonlists modes but they turned | |
5928 | * out to be just not useful. Let's keep the warning in place | |
5929 | * if somebody still use the cmd line parameter so that we do | |
5930 | * not fail it silently | |
5931 | */ | |
5932 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5933 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5934 | return -EINVAL; |
5935 | } | |
5936 | return 0; | |
5937 | } | |
5938 | ||
c9bff3ee MH |
5939 | char numa_zonelist_order[] = "Node"; |
5940 | ||
f0c0b2b8 KH |
5941 | /* |
5942 | * sysctl handler for numa_zonelist_order | |
5943 | */ | |
cccad5b9 | 5944 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 5945 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 5946 | { |
32927393 CH |
5947 | if (write) |
5948 | return __parse_numa_zonelist_order(buffer); | |
5949 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
5950 | } |
5951 | ||
5952 | ||
62bc62a8 | 5953 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5954 | static int node_load[MAX_NUMNODES]; |
5955 | ||
1da177e4 | 5956 | /** |
4dc3b16b | 5957 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5958 | * @node: node whose fallback list we're appending |
5959 | * @used_node_mask: nodemask_t of already used nodes | |
5960 | * | |
5961 | * We use a number of factors to determine which is the next node that should | |
5962 | * appear on a given node's fallback list. The node should not have appeared | |
5963 | * already in @node's fallback list, and it should be the next closest node | |
5964 | * according to the distance array (which contains arbitrary distance values | |
5965 | * from each node to each node in the system), and should also prefer nodes | |
5966 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5967 | * on them otherwise. | |
a862f68a MR |
5968 | * |
5969 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 5970 | */ |
f0c0b2b8 | 5971 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5972 | { |
4cf808eb | 5973 | int n, val; |
1da177e4 | 5974 | int min_val = INT_MAX; |
00ef2d2f | 5975 | int best_node = NUMA_NO_NODE; |
1da177e4 | 5976 | |
4cf808eb LT |
5977 | /* Use the local node if we haven't already */ |
5978 | if (!node_isset(node, *used_node_mask)) { | |
5979 | node_set(node, *used_node_mask); | |
5980 | return node; | |
5981 | } | |
1da177e4 | 5982 | |
4b0ef1fe | 5983 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5984 | |
5985 | /* Don't want a node to appear more than once */ | |
5986 | if (node_isset(n, *used_node_mask)) | |
5987 | continue; | |
5988 | ||
1da177e4 LT |
5989 | /* Use the distance array to find the distance */ |
5990 | val = node_distance(node, n); | |
5991 | ||
4cf808eb LT |
5992 | /* Penalize nodes under us ("prefer the next node") */ |
5993 | val += (n < node); | |
5994 | ||
1da177e4 | 5995 | /* Give preference to headless and unused nodes */ |
b630749f | 5996 | if (!cpumask_empty(cpumask_of_node(n))) |
1da177e4 LT |
5997 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5998 | ||
5999 | /* Slight preference for less loaded node */ | |
6000 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
6001 | val += node_load[n]; | |
6002 | ||
6003 | if (val < min_val) { | |
6004 | min_val = val; | |
6005 | best_node = n; | |
6006 | } | |
6007 | } | |
6008 | ||
6009 | if (best_node >= 0) | |
6010 | node_set(best_node, *used_node_mask); | |
6011 | ||
6012 | return best_node; | |
6013 | } | |
6014 | ||
f0c0b2b8 KH |
6015 | |
6016 | /* | |
6017 | * Build zonelists ordered by node and zones within node. | |
6018 | * This results in maximum locality--normal zone overflows into local | |
6019 | * DMA zone, if any--but risks exhausting DMA zone. | |
6020 | */ | |
9d3be21b MH |
6021 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
6022 | unsigned nr_nodes) | |
1da177e4 | 6023 | { |
9d3be21b MH |
6024 | struct zoneref *zonerefs; |
6025 | int i; | |
6026 | ||
6027 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
6028 | ||
6029 | for (i = 0; i < nr_nodes; i++) { | |
6030 | int nr_zones; | |
6031 | ||
6032 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 6033 | |
9d3be21b MH |
6034 | nr_zones = build_zonerefs_node(node, zonerefs); |
6035 | zonerefs += nr_zones; | |
6036 | } | |
6037 | zonerefs->zone = NULL; | |
6038 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
6039 | } |
6040 | ||
523b9458 CL |
6041 | /* |
6042 | * Build gfp_thisnode zonelists | |
6043 | */ | |
6044 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
6045 | { | |
9d3be21b MH |
6046 | struct zoneref *zonerefs; |
6047 | int nr_zones; | |
523b9458 | 6048 | |
9d3be21b MH |
6049 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
6050 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6051 | zonerefs += nr_zones; | |
6052 | zonerefs->zone = NULL; | |
6053 | zonerefs->zone_idx = 0; | |
523b9458 CL |
6054 | } |
6055 | ||
f0c0b2b8 KH |
6056 | /* |
6057 | * Build zonelists ordered by zone and nodes within zones. | |
6058 | * This results in conserving DMA zone[s] until all Normal memory is | |
6059 | * exhausted, but results in overflowing to remote node while memory | |
6060 | * may still exist in local DMA zone. | |
6061 | */ | |
f0c0b2b8 | 6062 | |
f0c0b2b8 KH |
6063 | static void build_zonelists(pg_data_t *pgdat) |
6064 | { | |
9d3be21b MH |
6065 | static int node_order[MAX_NUMNODES]; |
6066 | int node, load, nr_nodes = 0; | |
d0ddf49b | 6067 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 6068 | int local_node, prev_node; |
1da177e4 LT |
6069 | |
6070 | /* NUMA-aware ordering of nodes */ | |
6071 | local_node = pgdat->node_id; | |
62bc62a8 | 6072 | load = nr_online_nodes; |
1da177e4 | 6073 | prev_node = local_node; |
f0c0b2b8 | 6074 | |
f0c0b2b8 | 6075 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
6076 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
6077 | /* | |
6078 | * We don't want to pressure a particular node. | |
6079 | * So adding penalty to the first node in same | |
6080 | * distance group to make it round-robin. | |
6081 | */ | |
957f822a DR |
6082 | if (node_distance(local_node, node) != |
6083 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
6084 | node_load[node] = load; |
6085 | ||
9d3be21b | 6086 | node_order[nr_nodes++] = node; |
1da177e4 LT |
6087 | prev_node = node; |
6088 | load--; | |
1da177e4 | 6089 | } |
523b9458 | 6090 | |
9d3be21b | 6091 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 6092 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
6093 | } |
6094 | ||
7aac7898 LS |
6095 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6096 | /* | |
6097 | * Return node id of node used for "local" allocations. | |
6098 | * I.e., first node id of first zone in arg node's generic zonelist. | |
6099 | * Used for initializing percpu 'numa_mem', which is used primarily | |
6100 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
6101 | */ | |
6102 | int local_memory_node(int node) | |
6103 | { | |
c33d6c06 | 6104 | struct zoneref *z; |
7aac7898 | 6105 | |
c33d6c06 | 6106 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 6107 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 6108 | NULL); |
c1093b74 | 6109 | return zone_to_nid(z->zone); |
7aac7898 LS |
6110 | } |
6111 | #endif | |
f0c0b2b8 | 6112 | |
6423aa81 JK |
6113 | static void setup_min_unmapped_ratio(void); |
6114 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
6115 | #else /* CONFIG_NUMA */ |
6116 | ||
f0c0b2b8 | 6117 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 6118 | { |
19655d34 | 6119 | int node, local_node; |
9d3be21b MH |
6120 | struct zoneref *zonerefs; |
6121 | int nr_zones; | |
1da177e4 LT |
6122 | |
6123 | local_node = pgdat->node_id; | |
1da177e4 | 6124 | |
9d3be21b MH |
6125 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
6126 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6127 | zonerefs += nr_zones; | |
1da177e4 | 6128 | |
54a6eb5c MG |
6129 | /* |
6130 | * Now we build the zonelist so that it contains the zones | |
6131 | * of all the other nodes. | |
6132 | * We don't want to pressure a particular node, so when | |
6133 | * building the zones for node N, we make sure that the | |
6134 | * zones coming right after the local ones are those from | |
6135 | * node N+1 (modulo N) | |
6136 | */ | |
6137 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
6138 | if (!node_online(node)) | |
6139 | continue; | |
9d3be21b MH |
6140 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6141 | zonerefs += nr_zones; | |
1da177e4 | 6142 | } |
54a6eb5c MG |
6143 | for (node = 0; node < local_node; node++) { |
6144 | if (!node_online(node)) | |
6145 | continue; | |
9d3be21b MH |
6146 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6147 | zonerefs += nr_zones; | |
54a6eb5c MG |
6148 | } |
6149 | ||
9d3be21b MH |
6150 | zonerefs->zone = NULL; |
6151 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
6152 | } |
6153 | ||
6154 | #endif /* CONFIG_NUMA */ | |
6155 | ||
99dcc3e5 CL |
6156 | /* |
6157 | * Boot pageset table. One per cpu which is going to be used for all | |
6158 | * zones and all nodes. The parameters will be set in such a way | |
6159 | * that an item put on a list will immediately be handed over to | |
6160 | * the buddy list. This is safe since pageset manipulation is done | |
6161 | * with interrupts disabled. | |
6162 | * | |
6163 | * The boot_pagesets must be kept even after bootup is complete for | |
6164 | * unused processors and/or zones. They do play a role for bootstrapping | |
6165 | * hotplugged processors. | |
6166 | * | |
6167 | * zoneinfo_show() and maybe other functions do | |
6168 | * not check if the processor is online before following the pageset pointer. | |
6169 | * Other parts of the kernel may not check if the zone is available. | |
6170 | */ | |
69a8396a | 6171 | static void pageset_init(struct per_cpu_pageset *p); |
952eaf81 VB |
6172 | /* These effectively disable the pcplists in the boot pageset completely */ |
6173 | #define BOOT_PAGESET_HIGH 0 | |
6174 | #define BOOT_PAGESET_BATCH 1 | |
99dcc3e5 | 6175 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); |
385386cf | 6176 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 6177 | |
11cd8638 | 6178 | static void __build_all_zonelists(void *data) |
1da177e4 | 6179 | { |
6811378e | 6180 | int nid; |
afb6ebb3 | 6181 | int __maybe_unused cpu; |
9adb62a5 | 6182 | pg_data_t *self = data; |
b93e0f32 MH |
6183 | static DEFINE_SPINLOCK(lock); |
6184 | ||
6185 | spin_lock(&lock); | |
9276b1bc | 6186 | |
7f9cfb31 BL |
6187 | #ifdef CONFIG_NUMA |
6188 | memset(node_load, 0, sizeof(node_load)); | |
6189 | #endif | |
9adb62a5 | 6190 | |
c1152583 WY |
6191 | /* |
6192 | * This node is hotadded and no memory is yet present. So just | |
6193 | * building zonelists is fine - no need to touch other nodes. | |
6194 | */ | |
9adb62a5 JL |
6195 | if (self && !node_online(self->node_id)) { |
6196 | build_zonelists(self); | |
c1152583 WY |
6197 | } else { |
6198 | for_each_online_node(nid) { | |
6199 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 6200 | |
c1152583 WY |
6201 | build_zonelists(pgdat); |
6202 | } | |
99dcc3e5 | 6203 | |
7aac7898 LS |
6204 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6205 | /* | |
6206 | * We now know the "local memory node" for each node-- | |
6207 | * i.e., the node of the first zone in the generic zonelist. | |
6208 | * Set up numa_mem percpu variable for on-line cpus. During | |
6209 | * boot, only the boot cpu should be on-line; we'll init the | |
6210 | * secondary cpus' numa_mem as they come on-line. During | |
6211 | * node/memory hotplug, we'll fixup all on-line cpus. | |
6212 | */ | |
d9c9a0b9 | 6213 | for_each_online_cpu(cpu) |
7aac7898 | 6214 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 6215 | #endif |
d9c9a0b9 | 6216 | } |
b93e0f32 MH |
6217 | |
6218 | spin_unlock(&lock); | |
6811378e YG |
6219 | } |
6220 | ||
061f67bc RV |
6221 | static noinline void __init |
6222 | build_all_zonelists_init(void) | |
6223 | { | |
afb6ebb3 MH |
6224 | int cpu; |
6225 | ||
061f67bc | 6226 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
6227 | |
6228 | /* | |
6229 | * Initialize the boot_pagesets that are going to be used | |
6230 | * for bootstrapping processors. The real pagesets for | |
6231 | * each zone will be allocated later when the per cpu | |
6232 | * allocator is available. | |
6233 | * | |
6234 | * boot_pagesets are used also for bootstrapping offline | |
6235 | * cpus if the system is already booted because the pagesets | |
6236 | * are needed to initialize allocators on a specific cpu too. | |
6237 | * F.e. the percpu allocator needs the page allocator which | |
6238 | * needs the percpu allocator in order to allocate its pagesets | |
6239 | * (a chicken-egg dilemma). | |
6240 | */ | |
6241 | for_each_possible_cpu(cpu) | |
69a8396a | 6242 | pageset_init(&per_cpu(boot_pageset, cpu)); |
afb6ebb3 | 6243 | |
061f67bc RV |
6244 | mminit_verify_zonelist(); |
6245 | cpuset_init_current_mems_allowed(); | |
6246 | } | |
6247 | ||
4eaf3f64 | 6248 | /* |
4eaf3f64 | 6249 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 6250 | * |
72675e13 | 6251 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 6252 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 6253 | */ |
72675e13 | 6254 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 6255 | { |
0a18e607 DH |
6256 | unsigned long vm_total_pages; |
6257 | ||
6811378e | 6258 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 6259 | build_all_zonelists_init(); |
6811378e | 6260 | } else { |
11cd8638 | 6261 | __build_all_zonelists(pgdat); |
6811378e YG |
6262 | /* cpuset refresh routine should be here */ |
6263 | } | |
56b9413b DH |
6264 | /* Get the number of free pages beyond high watermark in all zones. */ |
6265 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
6266 | /* |
6267 | * Disable grouping by mobility if the number of pages in the | |
6268 | * system is too low to allow the mechanism to work. It would be | |
6269 | * more accurate, but expensive to check per-zone. This check is | |
6270 | * made on memory-hotadd so a system can start with mobility | |
6271 | * disabled and enable it later | |
6272 | */ | |
d9c23400 | 6273 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
6274 | page_group_by_mobility_disabled = 1; |
6275 | else | |
6276 | page_group_by_mobility_disabled = 0; | |
6277 | ||
ce0725f7 | 6278 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 6279 | nr_online_nodes, |
756a025f JP |
6280 | page_group_by_mobility_disabled ? "off" : "on", |
6281 | vm_total_pages); | |
f0c0b2b8 | 6282 | #ifdef CONFIG_NUMA |
f88dfff5 | 6283 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 6284 | #endif |
1da177e4 LT |
6285 | } |
6286 | ||
a9a9e77f PT |
6287 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
6288 | static bool __meminit | |
6289 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
6290 | { | |
a9a9e77f PT |
6291 | static struct memblock_region *r; |
6292 | ||
6293 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
6294 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
cc6de168 | 6295 | for_each_mem_region(r) { |
a9a9e77f PT |
6296 | if (*pfn < memblock_region_memory_end_pfn(r)) |
6297 | break; | |
6298 | } | |
6299 | } | |
6300 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
6301 | memblock_is_mirror(r)) { | |
6302 | *pfn = memblock_region_memory_end_pfn(r); | |
6303 | return true; | |
6304 | } | |
6305 | } | |
a9a9e77f PT |
6306 | return false; |
6307 | } | |
6308 | ||
1da177e4 LT |
6309 | /* |
6310 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 6311 | * up by memblock_free_all() once the early boot process is |
1da177e4 | 6312 | * done. Non-atomic initialization, single-pass. |
d882c006 DH |
6313 | * |
6314 | * All aligned pageblocks are initialized to the specified migratetype | |
6315 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
6316 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
1da177e4 | 6317 | */ |
ab28cb6e | 6318 | void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone, |
dc2da7b4 | 6319 | unsigned long start_pfn, unsigned long zone_end_pfn, |
d882c006 DH |
6320 | enum meminit_context context, |
6321 | struct vmem_altmap *altmap, int migratetype) | |
1da177e4 | 6322 | { |
a9a9e77f | 6323 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 6324 | struct page *page; |
1da177e4 | 6325 | |
22b31eec HD |
6326 | if (highest_memmap_pfn < end_pfn - 1) |
6327 | highest_memmap_pfn = end_pfn - 1; | |
6328 | ||
966cf44f | 6329 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
6330 | /* |
6331 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
6332 | * memory. We limit the total number of pages to initialize to just |
6333 | * those that might contain the memory mapping. We will defer the | |
6334 | * ZONE_DEVICE page initialization until after we have released | |
6335 | * the hotplug lock. | |
4b94ffdc | 6336 | */ |
966cf44f AD |
6337 | if (zone == ZONE_DEVICE) { |
6338 | if (!altmap) | |
6339 | return; | |
6340 | ||
6341 | if (start_pfn == altmap->base_pfn) | |
6342 | start_pfn += altmap->reserve; | |
6343 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
6344 | } | |
6345 | #endif | |
4b94ffdc | 6346 | |
948c436e | 6347 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 6348 | /* |
b72d0ffb AM |
6349 | * There can be holes in boot-time mem_map[]s handed to this |
6350 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 6351 | */ |
c1d0da83 | 6352 | if (context == MEMINIT_EARLY) { |
a9a9e77f PT |
6353 | if (overlap_memmap_init(zone, &pfn)) |
6354 | continue; | |
dc2da7b4 | 6355 | if (defer_init(nid, pfn, zone_end_pfn)) |
a9a9e77f | 6356 | break; |
a2f3aa02 | 6357 | } |
ac5d2539 | 6358 | |
d0dc12e8 PT |
6359 | page = pfn_to_page(pfn); |
6360 | __init_single_page(page, pfn, zone, nid); | |
c1d0da83 | 6361 | if (context == MEMINIT_HOTPLUG) |
d483da5b | 6362 | __SetPageReserved(page); |
d0dc12e8 | 6363 | |
ac5d2539 | 6364 | /* |
d882c006 DH |
6365 | * Usually, we want to mark the pageblock MIGRATE_MOVABLE, |
6366 | * such that unmovable allocations won't be scattered all | |
6367 | * over the place during system boot. | |
ac5d2539 | 6368 | */ |
4eb29bd9 | 6369 | if (IS_ALIGNED(pfn, pageblock_nr_pages)) { |
d882c006 | 6370 | set_pageblock_migratetype(page, migratetype); |
9b6e63cb | 6371 | cond_resched(); |
ac5d2539 | 6372 | } |
948c436e | 6373 | pfn++; |
1da177e4 LT |
6374 | } |
6375 | } | |
6376 | ||
966cf44f AD |
6377 | #ifdef CONFIG_ZONE_DEVICE |
6378 | void __ref memmap_init_zone_device(struct zone *zone, | |
6379 | unsigned long start_pfn, | |
1f8d75c1 | 6380 | unsigned long nr_pages, |
966cf44f AD |
6381 | struct dev_pagemap *pgmap) |
6382 | { | |
1f8d75c1 | 6383 | unsigned long pfn, end_pfn = start_pfn + nr_pages; |
966cf44f | 6384 | struct pglist_data *pgdat = zone->zone_pgdat; |
514caf23 | 6385 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
966cf44f AD |
6386 | unsigned long zone_idx = zone_idx(zone); |
6387 | unsigned long start = jiffies; | |
6388 | int nid = pgdat->node_id; | |
6389 | ||
46d945ae | 6390 | if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE)) |
966cf44f AD |
6391 | return; |
6392 | ||
6393 | /* | |
6394 | * The call to memmap_init_zone should have already taken care | |
6395 | * of the pages reserved for the memmap, so we can just jump to | |
6396 | * the end of that region and start processing the device pages. | |
6397 | */ | |
514caf23 | 6398 | if (altmap) { |
966cf44f | 6399 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
1f8d75c1 | 6400 | nr_pages = end_pfn - start_pfn; |
966cf44f AD |
6401 | } |
6402 | ||
6403 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
6404 | struct page *page = pfn_to_page(pfn); | |
6405 | ||
6406 | __init_single_page(page, pfn, zone_idx, nid); | |
6407 | ||
6408 | /* | |
6409 | * Mark page reserved as it will need to wait for onlining | |
6410 | * phase for it to be fully associated with a zone. | |
6411 | * | |
6412 | * We can use the non-atomic __set_bit operation for setting | |
6413 | * the flag as we are still initializing the pages. | |
6414 | */ | |
6415 | __SetPageReserved(page); | |
6416 | ||
6417 | /* | |
8a164fef CH |
6418 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer |
6419 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
6420 | * ever freed or placed on a driver-private list. | |
966cf44f AD |
6421 | */ |
6422 | page->pgmap = pgmap; | |
8a164fef | 6423 | page->zone_device_data = NULL; |
966cf44f AD |
6424 | |
6425 | /* | |
6426 | * Mark the block movable so that blocks are reserved for | |
6427 | * movable at startup. This will force kernel allocations | |
6428 | * to reserve their blocks rather than leaking throughout | |
6429 | * the address space during boot when many long-lived | |
6430 | * kernel allocations are made. | |
6431 | * | |
c1d0da83 | 6432 | * Please note that MEMINIT_HOTPLUG path doesn't clear memmap |
ba72b4c8 | 6433 | * because this is done early in section_activate() |
966cf44f | 6434 | */ |
4eb29bd9 | 6435 | if (IS_ALIGNED(pfn, pageblock_nr_pages)) { |
966cf44f AD |
6436 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
6437 | cond_resched(); | |
6438 | } | |
6439 | } | |
6440 | ||
fdc029b1 | 6441 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
1f8d75c1 | 6442 | nr_pages, jiffies_to_msecs(jiffies - start)); |
966cf44f AD |
6443 | } |
6444 | ||
6445 | #endif | |
1e548deb | 6446 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6447 | { |
7aeb09f9 | 6448 | unsigned int order, t; |
b2a0ac88 MG |
6449 | for_each_migratetype_order(order, t) { |
6450 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6451 | zone->free_area[order].nr_free = 0; |
6452 | } | |
6453 | } | |
6454 | ||
0740a50b MR |
6455 | #if !defined(CONFIG_FLAT_NODE_MEM_MAP) |
6456 | /* | |
6457 | * Only struct pages that correspond to ranges defined by memblock.memory | |
6458 | * are zeroed and initialized by going through __init_single_page() during | |
6459 | * memmap_init_zone(). | |
6460 | * | |
6461 | * But, there could be struct pages that correspond to holes in | |
6462 | * memblock.memory. This can happen because of the following reasons: | |
6463 | * - physical memory bank size is not necessarily the exact multiple of the | |
6464 | * arbitrary section size | |
6465 | * - early reserved memory may not be listed in memblock.memory | |
6466 | * - memory layouts defined with memmap= kernel parameter may not align | |
6467 | * nicely with memmap sections | |
6468 | * | |
6469 | * Explicitly initialize those struct pages so that: | |
6470 | * - PG_Reserved is set | |
6471 | * - zone and node links point to zone and node that span the page if the | |
6472 | * hole is in the middle of a zone | |
6473 | * - zone and node links point to adjacent zone/node if the hole falls on | |
6474 | * the zone boundary; the pages in such holes will be prepended to the | |
6475 | * zone/node above the hole except for the trailing pages in the last | |
6476 | * section that will be appended to the zone/node below. | |
6477 | */ | |
6478 | static u64 __meminit init_unavailable_range(unsigned long spfn, | |
6479 | unsigned long epfn, | |
6480 | int zone, int node) | |
6481 | { | |
6482 | unsigned long pfn; | |
6483 | u64 pgcnt = 0; | |
6484 | ||
6485 | for (pfn = spfn; pfn < epfn; pfn++) { | |
6486 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { | |
6487 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6488 | + pageblock_nr_pages - 1; | |
6489 | continue; | |
6490 | } | |
6491 | __init_single_page(pfn_to_page(pfn), pfn, zone, node); | |
6492 | __SetPageReserved(pfn_to_page(pfn)); | |
6493 | pgcnt++; | |
6494 | } | |
6495 | ||
6496 | return pgcnt; | |
6497 | } | |
6498 | #else | |
6499 | static inline u64 init_unavailable_range(unsigned long spfn, unsigned long epfn, | |
6500 | int zone, int node) | |
6501 | { | |
6502 | return 0; | |
6503 | } | |
6504 | #endif | |
6505 | ||
3256ff83 | 6506 | void __meminit __weak memmap_init_zone(struct zone *zone) |
dfb3ccd0 | 6507 | { |
3256ff83 BH |
6508 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
6509 | unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages; | |
6510 | int i, nid = zone_to_nid(zone), zone_id = zone_idx(zone); | |
0740a50b | 6511 | static unsigned long hole_pfn; |
73a6e474 | 6512 | unsigned long start_pfn, end_pfn; |
0740a50b | 6513 | u64 pgcnt = 0; |
73a6e474 BH |
6514 | |
6515 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { | |
3256ff83 BH |
6516 | start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn); |
6517 | end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn); | |
73a6e474 | 6518 | |
3256ff83 BH |
6519 | if (end_pfn > start_pfn) |
6520 | memmap_init_range(end_pfn - start_pfn, nid, | |
6521 | zone_id, start_pfn, zone_end_pfn, | |
6522 | MEMINIT_EARLY, NULL, MIGRATE_MOVABLE); | |
0740a50b MR |
6523 | |
6524 | if (hole_pfn < start_pfn) | |
6525 | pgcnt += init_unavailable_range(hole_pfn, start_pfn, | |
6526 | zone_id, nid); | |
6527 | hole_pfn = end_pfn; | |
73a6e474 | 6528 | } |
0740a50b MR |
6529 | |
6530 | #ifdef CONFIG_SPARSEMEM | |
6531 | /* | |
6532 | * Initialize the hole in the range [zone_end_pfn, section_end]. | |
6533 | * If zone boundary falls in the middle of a section, this hole | |
6534 | * will be re-initialized during the call to this function for the | |
6535 | * higher zone. | |
6536 | */ | |
6537 | end_pfn = round_up(zone_end_pfn, PAGES_PER_SECTION); | |
6538 | if (hole_pfn < end_pfn) | |
6539 | pgcnt += init_unavailable_range(hole_pfn, end_pfn, | |
6540 | zone_id, nid); | |
6541 | #endif | |
6542 | ||
6543 | if (pgcnt) | |
6544 | pr_info(" %s zone: %llu pages in unavailable ranges\n", | |
6545 | zone->name, pgcnt); | |
dfb3ccd0 | 6546 | } |
1da177e4 | 6547 | |
7cd2b0a3 | 6548 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 6549 | { |
3a6be87f | 6550 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
6551 | int batch; |
6552 | ||
6553 | /* | |
6554 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 6555 | * size of the zone. |
e7c8d5c9 | 6556 | */ |
9705bea5 | 6557 | batch = zone_managed_pages(zone) / 1024; |
d8a759b5 AL |
6558 | /* But no more than a meg. */ |
6559 | if (batch * PAGE_SIZE > 1024 * 1024) | |
6560 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
6561 | batch /= 4; /* We effectively *= 4 below */ |
6562 | if (batch < 1) | |
6563 | batch = 1; | |
6564 | ||
6565 | /* | |
0ceaacc9 NP |
6566 | * Clamp the batch to a 2^n - 1 value. Having a power |
6567 | * of 2 value was found to be more likely to have | |
6568 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 6569 | * |
0ceaacc9 NP |
6570 | * For example if 2 tasks are alternately allocating |
6571 | * batches of pages, one task can end up with a lot | |
6572 | * of pages of one half of the possible page colors | |
6573 | * and the other with pages of the other colors. | |
e7c8d5c9 | 6574 | */ |
9155203a | 6575 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 6576 | |
e7c8d5c9 | 6577 | return batch; |
3a6be87f DH |
6578 | |
6579 | #else | |
6580 | /* The deferral and batching of frees should be suppressed under NOMMU | |
6581 | * conditions. | |
6582 | * | |
6583 | * The problem is that NOMMU needs to be able to allocate large chunks | |
6584 | * of contiguous memory as there's no hardware page translation to | |
6585 | * assemble apparent contiguous memory from discontiguous pages. | |
6586 | * | |
6587 | * Queueing large contiguous runs of pages for batching, however, | |
6588 | * causes the pages to actually be freed in smaller chunks. As there | |
6589 | * can be a significant delay between the individual batches being | |
6590 | * recycled, this leads to the once large chunks of space being | |
6591 | * fragmented and becoming unavailable for high-order allocations. | |
6592 | */ | |
6593 | return 0; | |
6594 | #endif | |
e7c8d5c9 CL |
6595 | } |
6596 | ||
8d7a8fa9 | 6597 | /* |
5c3ad2eb VB |
6598 | * pcp->high and pcp->batch values are related and generally batch is lower |
6599 | * than high. They are also related to pcp->count such that count is lower | |
6600 | * than high, and as soon as it reaches high, the pcplist is flushed. | |
8d7a8fa9 | 6601 | * |
5c3ad2eb VB |
6602 | * However, guaranteeing these relations at all times would require e.g. write |
6603 | * barriers here but also careful usage of read barriers at the read side, and | |
6604 | * thus be prone to error and bad for performance. Thus the update only prevents | |
6605 | * store tearing. Any new users of pcp->batch and pcp->high should ensure they | |
6606 | * can cope with those fields changing asynchronously, and fully trust only the | |
6607 | * pcp->count field on the local CPU with interrupts disabled. | |
8d7a8fa9 CS |
6608 | * |
6609 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
6610 | * outside of boot time (or some other assurance that no concurrent updaters | |
6611 | * exist). | |
6612 | */ | |
6613 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
6614 | unsigned long batch) | |
6615 | { | |
5c3ad2eb VB |
6616 | WRITE_ONCE(pcp->batch, batch); |
6617 | WRITE_ONCE(pcp->high, high); | |
8d7a8fa9 CS |
6618 | } |
6619 | ||
88c90dbc | 6620 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
6621 | { |
6622 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 6623 | int migratetype; |
2caaad41 | 6624 | |
1c6fe946 MD |
6625 | memset(p, 0, sizeof(*p)); |
6626 | ||
3dfa5721 | 6627 | pcp = &p->pcp; |
5f8dcc21 MG |
6628 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
6629 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 | 6630 | |
69a8396a VB |
6631 | /* |
6632 | * Set batch and high values safe for a boot pageset. A true percpu | |
6633 | * pageset's initialization will update them subsequently. Here we don't | |
6634 | * need to be as careful as pageset_update() as nobody can access the | |
6635 | * pageset yet. | |
6636 | */ | |
952eaf81 VB |
6637 | pcp->high = BOOT_PAGESET_HIGH; |
6638 | pcp->batch = BOOT_PAGESET_BATCH; | |
88c90dbc CS |
6639 | } |
6640 | ||
3b1f3658 | 6641 | static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high, |
ec6e8c7e VB |
6642 | unsigned long batch) |
6643 | { | |
6644 | struct per_cpu_pageset *p; | |
6645 | int cpu; | |
6646 | ||
6647 | for_each_possible_cpu(cpu) { | |
6648 | p = per_cpu_ptr(zone->pageset, cpu); | |
6649 | pageset_update(&p->pcp, high, batch); | |
6650 | } | |
6651 | } | |
6652 | ||
8ad4b1fb | 6653 | /* |
0a8b4f1d | 6654 | * Calculate and set new high and batch values for all per-cpu pagesets of a |
7115ac6e | 6655 | * zone, based on the zone's size and the percpu_pagelist_fraction sysctl. |
8ad4b1fb | 6656 | */ |
0a8b4f1d | 6657 | static void zone_set_pageset_high_and_batch(struct zone *zone) |
56cef2b8 | 6658 | { |
7115ac6e VB |
6659 | unsigned long new_high, new_batch; |
6660 | ||
6661 | if (percpu_pagelist_fraction) { | |
6662 | new_high = zone_managed_pages(zone) / percpu_pagelist_fraction; | |
6663 | new_batch = max(1UL, new_high / 4); | |
6664 | if ((new_high / 4) > (PAGE_SHIFT * 8)) | |
6665 | new_batch = PAGE_SHIFT * 8; | |
6666 | } else { | |
6667 | new_batch = zone_batchsize(zone); | |
6668 | new_high = 6 * new_batch; | |
6669 | new_batch = max(1UL, 1 * new_batch); | |
6670 | } | |
169f6c19 | 6671 | |
952eaf81 VB |
6672 | if (zone->pageset_high == new_high && |
6673 | zone->pageset_batch == new_batch) | |
6674 | return; | |
6675 | ||
6676 | zone->pageset_high = new_high; | |
6677 | zone->pageset_batch = new_batch; | |
6678 | ||
ec6e8c7e | 6679 | __zone_set_pageset_high_and_batch(zone, new_high, new_batch); |
169f6c19 CS |
6680 | } |
6681 | ||
72675e13 | 6682 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 | 6683 | { |
0a8b4f1d | 6684 | struct per_cpu_pageset *p; |
319774e2 | 6685 | int cpu; |
0a8b4f1d | 6686 | |
319774e2 | 6687 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
0a8b4f1d VB |
6688 | for_each_possible_cpu(cpu) { |
6689 | p = per_cpu_ptr(zone->pageset, cpu); | |
6690 | pageset_init(p); | |
6691 | } | |
6692 | ||
6693 | zone_set_pageset_high_and_batch(zone); | |
319774e2 WF |
6694 | } |
6695 | ||
2caaad41 | 6696 | /* |
99dcc3e5 CL |
6697 | * Allocate per cpu pagesets and initialize them. |
6698 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 6699 | */ |
99dcc3e5 | 6700 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 6701 | { |
b4911ea2 | 6702 | struct pglist_data *pgdat; |
99dcc3e5 | 6703 | struct zone *zone; |
b418a0f9 | 6704 | int __maybe_unused cpu; |
e7c8d5c9 | 6705 | |
319774e2 WF |
6706 | for_each_populated_zone(zone) |
6707 | setup_zone_pageset(zone); | |
b4911ea2 | 6708 | |
b418a0f9 SD |
6709 | #ifdef CONFIG_NUMA |
6710 | /* | |
6711 | * Unpopulated zones continue using the boot pagesets. | |
6712 | * The numa stats for these pagesets need to be reset. | |
6713 | * Otherwise, they will end up skewing the stats of | |
6714 | * the nodes these zones are associated with. | |
6715 | */ | |
6716 | for_each_possible_cpu(cpu) { | |
6717 | struct per_cpu_pageset *pcp = &per_cpu(boot_pageset, cpu); | |
6718 | memset(pcp->vm_numa_stat_diff, 0, | |
6719 | sizeof(pcp->vm_numa_stat_diff)); | |
6720 | } | |
6721 | #endif | |
6722 | ||
b4911ea2 MG |
6723 | for_each_online_pgdat(pgdat) |
6724 | pgdat->per_cpu_nodestats = | |
6725 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
6726 | } |
6727 | ||
c09b4240 | 6728 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 6729 | { |
99dcc3e5 CL |
6730 | /* |
6731 | * per cpu subsystem is not up at this point. The following code | |
6732 | * relies on the ability of the linker to provide the | |
6733 | * offset of a (static) per cpu variable into the per cpu area. | |
6734 | */ | |
6735 | zone->pageset = &boot_pageset; | |
952eaf81 VB |
6736 | zone->pageset_high = BOOT_PAGESET_HIGH; |
6737 | zone->pageset_batch = BOOT_PAGESET_BATCH; | |
ed8ece2e | 6738 | |
b38a8725 | 6739 | if (populated_zone(zone)) |
99dcc3e5 CL |
6740 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
6741 | zone->name, zone->present_pages, | |
6742 | zone_batchsize(zone)); | |
ed8ece2e DH |
6743 | } |
6744 | ||
dc0bbf3b | 6745 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 6746 | unsigned long zone_start_pfn, |
b171e409 | 6747 | unsigned long size) |
ed8ece2e DH |
6748 | { |
6749 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 6750 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 6751 | |
8f416836 WY |
6752 | if (zone_idx > pgdat->nr_zones) |
6753 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 6754 | |
ed8ece2e DH |
6755 | zone->zone_start_pfn = zone_start_pfn; |
6756 | ||
708614e6 MG |
6757 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
6758 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
6759 | pgdat->node_id, | |
6760 | (unsigned long)zone_idx(zone), | |
6761 | zone_start_pfn, (zone_start_pfn + size)); | |
6762 | ||
1e548deb | 6763 | zone_init_free_lists(zone); |
9dcb8b68 | 6764 | zone->initialized = 1; |
ed8ece2e DH |
6765 | } |
6766 | ||
c713216d MG |
6767 | /** |
6768 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
6769 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
6770 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
6771 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
6772 | * |
6773 | * It returns the start and end page frame of a node based on information | |
7d018176 | 6774 | * provided by memblock_set_node(). If called for a node |
c713216d | 6775 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 6776 | * PFNs will be 0. |
c713216d | 6777 | */ |
bbe5d993 | 6778 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
6779 | unsigned long *start_pfn, unsigned long *end_pfn) |
6780 | { | |
c13291a5 | 6781 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 6782 | int i; |
c13291a5 | 6783 | |
c713216d MG |
6784 | *start_pfn = -1UL; |
6785 | *end_pfn = 0; | |
6786 | ||
c13291a5 TH |
6787 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
6788 | *start_pfn = min(*start_pfn, this_start_pfn); | |
6789 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
6790 | } |
6791 | ||
633c0666 | 6792 | if (*start_pfn == -1UL) |
c713216d | 6793 | *start_pfn = 0; |
c713216d MG |
6794 | } |
6795 | ||
2a1e274a MG |
6796 | /* |
6797 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
6798 | * assumption is made that zones within a node are ordered in monotonic | |
6799 | * increasing memory addresses so that the "highest" populated zone is used | |
6800 | */ | |
b69a7288 | 6801 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
6802 | { |
6803 | int zone_index; | |
6804 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
6805 | if (zone_index == ZONE_MOVABLE) | |
6806 | continue; | |
6807 | ||
6808 | if (arch_zone_highest_possible_pfn[zone_index] > | |
6809 | arch_zone_lowest_possible_pfn[zone_index]) | |
6810 | break; | |
6811 | } | |
6812 | ||
6813 | VM_BUG_ON(zone_index == -1); | |
6814 | movable_zone = zone_index; | |
6815 | } | |
6816 | ||
6817 | /* | |
6818 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 6819 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
6820 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
6821 | * in each node depending on the size of each node and how evenly kernelcore | |
6822 | * is distributed. This helper function adjusts the zone ranges | |
6823 | * provided by the architecture for a given node by using the end of the | |
6824 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
6825 | * zones within a node are in order of monotonic increases memory addresses | |
6826 | */ | |
bbe5d993 | 6827 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
6828 | unsigned long zone_type, |
6829 | unsigned long node_start_pfn, | |
6830 | unsigned long node_end_pfn, | |
6831 | unsigned long *zone_start_pfn, | |
6832 | unsigned long *zone_end_pfn) | |
6833 | { | |
6834 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
6835 | if (zone_movable_pfn[nid]) { | |
6836 | /* Size ZONE_MOVABLE */ | |
6837 | if (zone_type == ZONE_MOVABLE) { | |
6838 | *zone_start_pfn = zone_movable_pfn[nid]; | |
6839 | *zone_end_pfn = min(node_end_pfn, | |
6840 | arch_zone_highest_possible_pfn[movable_zone]); | |
6841 | ||
e506b996 XQ |
6842 | /* Adjust for ZONE_MOVABLE starting within this range */ |
6843 | } else if (!mirrored_kernelcore && | |
6844 | *zone_start_pfn < zone_movable_pfn[nid] && | |
6845 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
6846 | *zone_end_pfn = zone_movable_pfn[nid]; | |
6847 | ||
2a1e274a MG |
6848 | /* Check if this whole range is within ZONE_MOVABLE */ |
6849 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
6850 | *zone_start_pfn = *zone_end_pfn; | |
6851 | } | |
6852 | } | |
6853 | ||
c713216d MG |
6854 | /* |
6855 | * Return the number of pages a zone spans in a node, including holes | |
6856 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
6857 | */ | |
bbe5d993 | 6858 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 6859 | unsigned long zone_type, |
7960aedd ZY |
6860 | unsigned long node_start_pfn, |
6861 | unsigned long node_end_pfn, | |
d91749c1 | 6862 | unsigned long *zone_start_pfn, |
854e8848 | 6863 | unsigned long *zone_end_pfn) |
c713216d | 6864 | { |
299c83dc LF |
6865 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6866 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 6867 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6868 | if (!node_start_pfn && !node_end_pfn) |
6869 | return 0; | |
6870 | ||
7960aedd | 6871 | /* Get the start and end of the zone */ |
299c83dc LF |
6872 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6873 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
6874 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6875 | node_start_pfn, node_end_pfn, | |
d91749c1 | 6876 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
6877 | |
6878 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 6879 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
6880 | return 0; |
6881 | ||
6882 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
6883 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
6884 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
6885 | |
6886 | /* Return the spanned pages */ | |
d91749c1 | 6887 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
6888 | } |
6889 | ||
6890 | /* | |
6891 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 6892 | * then all holes in the requested range will be accounted for. |
c713216d | 6893 | */ |
bbe5d993 | 6894 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
6895 | unsigned long range_start_pfn, |
6896 | unsigned long range_end_pfn) | |
6897 | { | |
96e907d1 TH |
6898 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
6899 | unsigned long start_pfn, end_pfn; | |
6900 | int i; | |
c713216d | 6901 | |
96e907d1 TH |
6902 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6903 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6904 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6905 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 6906 | } |
96e907d1 | 6907 | return nr_absent; |
c713216d MG |
6908 | } |
6909 | ||
6910 | /** | |
6911 | * absent_pages_in_range - Return number of page frames in holes within a range | |
6912 | * @start_pfn: The start PFN to start searching for holes | |
6913 | * @end_pfn: The end PFN to stop searching for holes | |
6914 | * | |
a862f68a | 6915 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
6916 | */ |
6917 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
6918 | unsigned long end_pfn) | |
6919 | { | |
6920 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
6921 | } | |
6922 | ||
6923 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 6924 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 6925 | unsigned long zone_type, |
7960aedd | 6926 | unsigned long node_start_pfn, |
854e8848 | 6927 | unsigned long node_end_pfn) |
c713216d | 6928 | { |
96e907d1 TH |
6929 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6930 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 6931 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 6932 | unsigned long nr_absent; |
9c7cd687 | 6933 | |
b5685e92 | 6934 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6935 | if (!node_start_pfn && !node_end_pfn) |
6936 | return 0; | |
6937 | ||
96e907d1 TH |
6938 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6939 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 6940 | |
2a1e274a MG |
6941 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6942 | node_start_pfn, node_end_pfn, | |
6943 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
6944 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
6945 | ||
6946 | /* | |
6947 | * ZONE_MOVABLE handling. | |
6948 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6949 | * and vice versa. | |
6950 | */ | |
e506b996 XQ |
6951 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6952 | unsigned long start_pfn, end_pfn; | |
6953 | struct memblock_region *r; | |
6954 | ||
cc6de168 | 6955 | for_each_mem_region(r) { |
e506b996 XQ |
6956 | start_pfn = clamp(memblock_region_memory_base_pfn(r), |
6957 | zone_start_pfn, zone_end_pfn); | |
6958 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6959 | zone_start_pfn, zone_end_pfn); | |
6960 | ||
6961 | if (zone_type == ZONE_MOVABLE && | |
6962 | memblock_is_mirror(r)) | |
6963 | nr_absent += end_pfn - start_pfn; | |
6964 | ||
6965 | if (zone_type == ZONE_NORMAL && | |
6966 | !memblock_is_mirror(r)) | |
6967 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6968 | } |
6969 | } | |
6970 | ||
6971 | return nr_absent; | |
c713216d | 6972 | } |
0e0b864e | 6973 | |
bbe5d993 | 6974 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd | 6975 | unsigned long node_start_pfn, |
854e8848 | 6976 | unsigned long node_end_pfn) |
c713216d | 6977 | { |
febd5949 | 6978 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6979 | enum zone_type i; |
6980 | ||
febd5949 GZ |
6981 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6982 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6983 | unsigned long zone_start_pfn, zone_end_pfn; |
3f08a302 | 6984 | unsigned long spanned, absent; |
febd5949 | 6985 | unsigned long size, real_size; |
c713216d | 6986 | |
854e8848 MR |
6987 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i, |
6988 | node_start_pfn, | |
6989 | node_end_pfn, | |
6990 | &zone_start_pfn, | |
6991 | &zone_end_pfn); | |
6992 | absent = zone_absent_pages_in_node(pgdat->node_id, i, | |
6993 | node_start_pfn, | |
6994 | node_end_pfn); | |
3f08a302 MR |
6995 | |
6996 | size = spanned; | |
6997 | real_size = size - absent; | |
6998 | ||
d91749c1 TI |
6999 | if (size) |
7000 | zone->zone_start_pfn = zone_start_pfn; | |
7001 | else | |
7002 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
7003 | zone->spanned_pages = size; |
7004 | zone->present_pages = real_size; | |
7005 | ||
7006 | totalpages += size; | |
7007 | realtotalpages += real_size; | |
7008 | } | |
7009 | ||
7010 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
7011 | pgdat->node_present_pages = realtotalpages; |
7012 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
7013 | realtotalpages); | |
7014 | } | |
7015 | ||
835c134e MG |
7016 | #ifndef CONFIG_SPARSEMEM |
7017 | /* | |
7018 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
7019 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
7020 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
7021 | * round what is now in bits to nearest long in bits, then return it in |
7022 | * bytes. | |
7023 | */ | |
7c45512d | 7024 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
7025 | { |
7026 | unsigned long usemapsize; | |
7027 | ||
7c45512d | 7028 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
7029 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
7030 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
7031 | usemapsize *= NR_PAGEBLOCK_BITS; |
7032 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
7033 | ||
7034 | return usemapsize / 8; | |
7035 | } | |
7036 | ||
7010a6ec | 7037 | static void __ref setup_usemap(struct zone *zone) |
835c134e | 7038 | { |
7010a6ec BH |
7039 | unsigned long usemapsize = usemap_size(zone->zone_start_pfn, |
7040 | zone->spanned_pages); | |
835c134e | 7041 | zone->pageblock_flags = NULL; |
23a7052a | 7042 | if (usemapsize) { |
6782832e | 7043 | zone->pageblock_flags = |
26fb3dae | 7044 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
7010a6ec | 7045 | zone_to_nid(zone)); |
23a7052a MR |
7046 | if (!zone->pageblock_flags) |
7047 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
7010a6ec | 7048 | usemapsize, zone->name, zone_to_nid(zone)); |
23a7052a | 7049 | } |
835c134e MG |
7050 | } |
7051 | #else | |
7010a6ec | 7052 | static inline void setup_usemap(struct zone *zone) {} |
835c134e MG |
7053 | #endif /* CONFIG_SPARSEMEM */ |
7054 | ||
d9c23400 | 7055 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 7056 | |
d9c23400 | 7057 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 7058 | void __init set_pageblock_order(void) |
d9c23400 | 7059 | { |
955c1cd7 AM |
7060 | unsigned int order; |
7061 | ||
d9c23400 MG |
7062 | /* Check that pageblock_nr_pages has not already been setup */ |
7063 | if (pageblock_order) | |
7064 | return; | |
7065 | ||
955c1cd7 AM |
7066 | if (HPAGE_SHIFT > PAGE_SHIFT) |
7067 | order = HUGETLB_PAGE_ORDER; | |
7068 | else | |
7069 | order = MAX_ORDER - 1; | |
7070 | ||
d9c23400 MG |
7071 | /* |
7072 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
7073 | * This value may be variable depending on boot parameters on IA64 and |
7074 | * powerpc. | |
d9c23400 MG |
7075 | */ |
7076 | pageblock_order = order; | |
7077 | } | |
7078 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7079 | ||
ba72cb8c MG |
7080 | /* |
7081 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
7082 | * is unused as pageblock_order is set at compile-time. See |
7083 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
7084 | * the kernel config | |
ba72cb8c | 7085 | */ |
03e85f9d | 7086 | void __init set_pageblock_order(void) |
ba72cb8c | 7087 | { |
ba72cb8c | 7088 | } |
d9c23400 MG |
7089 | |
7090 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7091 | ||
03e85f9d | 7092 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 7093 | unsigned long present_pages) |
01cefaef JL |
7094 | { |
7095 | unsigned long pages = spanned_pages; | |
7096 | ||
7097 | /* | |
7098 | * Provide a more accurate estimation if there are holes within | |
7099 | * the zone and SPARSEMEM is in use. If there are holes within the | |
7100 | * zone, each populated memory region may cost us one or two extra | |
7101 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 7102 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
7103 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
7104 | */ | |
7105 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
7106 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
7107 | pages = present_pages; | |
7108 | ||
7109 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
7110 | } | |
7111 | ||
ace1db39 OS |
7112 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
7113 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
7114 | { | |
364c1eeb YS |
7115 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
7116 | ||
7117 | spin_lock_init(&ds_queue->split_queue_lock); | |
7118 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
7119 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
7120 | } |
7121 | #else | |
7122 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
7123 | #endif | |
7124 | ||
7125 | #ifdef CONFIG_COMPACTION | |
7126 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
7127 | { | |
7128 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
7129 | } | |
7130 | #else | |
7131 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
7132 | #endif | |
7133 | ||
03e85f9d | 7134 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 7135 | { |
208d54e5 | 7136 | pgdat_resize_init(pgdat); |
ace1db39 | 7137 | |
ace1db39 OS |
7138 | pgdat_init_split_queue(pgdat); |
7139 | pgdat_init_kcompactd(pgdat); | |
7140 | ||
1da177e4 | 7141 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 7142 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 7143 | |
eefa864b | 7144 | pgdat_page_ext_init(pgdat); |
867e5e1d | 7145 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
7146 | } |
7147 | ||
7148 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
7149 | unsigned long remaining_pages) | |
7150 | { | |
9705bea5 | 7151 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
7152 | zone_set_nid(zone, nid); |
7153 | zone->name = zone_names[idx]; | |
7154 | zone->zone_pgdat = NODE_DATA(nid); | |
7155 | spin_lock_init(&zone->lock); | |
7156 | zone_seqlock_init(zone); | |
7157 | zone_pcp_init(zone); | |
7158 | } | |
7159 | ||
7160 | /* | |
7161 | * Set up the zone data structures | |
7162 | * - init pgdat internals | |
7163 | * - init all zones belonging to this node | |
7164 | * | |
7165 | * NOTE: this function is only called during memory hotplug | |
7166 | */ | |
7167 | #ifdef CONFIG_MEMORY_HOTPLUG | |
7168 | void __ref free_area_init_core_hotplug(int nid) | |
7169 | { | |
7170 | enum zone_type z; | |
7171 | pg_data_t *pgdat = NODE_DATA(nid); | |
7172 | ||
7173 | pgdat_init_internals(pgdat); | |
7174 | for (z = 0; z < MAX_NR_ZONES; z++) | |
7175 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
7176 | } | |
7177 | #endif | |
7178 | ||
7179 | /* | |
7180 | * Set up the zone data structures: | |
7181 | * - mark all pages reserved | |
7182 | * - mark all memory queues empty | |
7183 | * - clear the memory bitmaps | |
7184 | * | |
7185 | * NOTE: pgdat should get zeroed by caller. | |
7186 | * NOTE: this function is only called during early init. | |
7187 | */ | |
7188 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
7189 | { | |
7190 | enum zone_type j; | |
7191 | int nid = pgdat->node_id; | |
5f63b720 | 7192 | |
03e85f9d | 7193 | pgdat_init_internals(pgdat); |
385386cf JW |
7194 | pgdat->per_cpu_nodestats = &boot_nodestats; |
7195 | ||
1da177e4 LT |
7196 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7197 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 7198 | unsigned long size, freesize, memmap_pages; |
1da177e4 | 7199 | |
febd5949 | 7200 | size = zone->spanned_pages; |
e6943859 | 7201 | freesize = zone->present_pages; |
1da177e4 | 7202 | |
0e0b864e | 7203 | /* |
9feedc9d | 7204 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
7205 | * is used by this zone for memmap. This affects the watermark |
7206 | * and per-cpu initialisations | |
7207 | */ | |
e6943859 | 7208 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
7209 | if (!is_highmem_idx(j)) { |
7210 | if (freesize >= memmap_pages) { | |
7211 | freesize -= memmap_pages; | |
7212 | if (memmap_pages) | |
7213 | printk(KERN_DEBUG | |
7214 | " %s zone: %lu pages used for memmap\n", | |
7215 | zone_names[j], memmap_pages); | |
7216 | } else | |
1170532b | 7217 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
7218 | zone_names[j], memmap_pages, freesize); |
7219 | } | |
0e0b864e | 7220 | |
6267276f | 7221 | /* Account for reserved pages */ |
9feedc9d JL |
7222 | if (j == 0 && freesize > dma_reserve) { |
7223 | freesize -= dma_reserve; | |
d903ef9f | 7224 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 7225 | zone_names[0], dma_reserve); |
0e0b864e MG |
7226 | } |
7227 | ||
98d2b0eb | 7228 | if (!is_highmem_idx(j)) |
9feedc9d | 7229 | nr_kernel_pages += freesize; |
01cefaef JL |
7230 | /* Charge for highmem memmap if there are enough kernel pages */ |
7231 | else if (nr_kernel_pages > memmap_pages * 2) | |
7232 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 7233 | nr_all_pages += freesize; |
1da177e4 | 7234 | |
9feedc9d JL |
7235 | /* |
7236 | * Set an approximate value for lowmem here, it will be adjusted | |
7237 | * when the bootmem allocator frees pages into the buddy system. | |
7238 | * And all highmem pages will be managed by the buddy system. | |
7239 | */ | |
03e85f9d | 7240 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 7241 | |
d883c6cf | 7242 | if (!size) |
1da177e4 LT |
7243 | continue; |
7244 | ||
955c1cd7 | 7245 | set_pageblock_order(); |
7010a6ec | 7246 | setup_usemap(zone); |
9699ee7b | 7247 | init_currently_empty_zone(zone, zone->zone_start_pfn, size); |
3256ff83 | 7248 | memmap_init_zone(zone); |
1da177e4 LT |
7249 | } |
7250 | } | |
7251 | ||
0cd842f9 | 7252 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 7253 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 7254 | { |
b0aeba74 | 7255 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
7256 | unsigned long __maybe_unused offset = 0; |
7257 | ||
1da177e4 LT |
7258 | /* Skip empty nodes */ |
7259 | if (!pgdat->node_spanned_pages) | |
7260 | return; | |
7261 | ||
b0aeba74 TL |
7262 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
7263 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
7264 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
7265 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 7266 | unsigned long size, end; |
d41dee36 AW |
7267 | struct page *map; |
7268 | ||
e984bb43 BP |
7269 | /* |
7270 | * The zone's endpoints aren't required to be MAX_ORDER | |
7271 | * aligned but the node_mem_map endpoints must be in order | |
7272 | * for the buddy allocator to function correctly. | |
7273 | */ | |
108bcc96 | 7274 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
7275 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
7276 | size = (end - start) * sizeof(struct page); | |
26fb3dae MR |
7277 | map = memblock_alloc_node(size, SMP_CACHE_BYTES, |
7278 | pgdat->node_id); | |
23a7052a MR |
7279 | if (!map) |
7280 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
7281 | size, pgdat->node_id); | |
a1c34a3b | 7282 | pgdat->node_mem_map = map + offset; |
1da177e4 | 7283 | } |
0cd842f9 OS |
7284 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
7285 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
7286 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 7287 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
7288 | /* |
7289 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
7290 | */ | |
c713216d | 7291 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 7292 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d | 7293 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 7294 | mem_map -= offset; |
c713216d | 7295 | } |
1da177e4 LT |
7296 | #endif |
7297 | } | |
0cd842f9 OS |
7298 | #else |
7299 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
7300 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 7301 | |
0188dc98 OS |
7302 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
7303 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
7304 | { | |
0188dc98 OS |
7305 | pgdat->first_deferred_pfn = ULONG_MAX; |
7306 | } | |
7307 | #else | |
7308 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
7309 | #endif | |
7310 | ||
854e8848 | 7311 | static void __init free_area_init_node(int nid) |
1da177e4 | 7312 | { |
9109fb7b | 7313 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
7314 | unsigned long start_pfn = 0; |
7315 | unsigned long end_pfn = 0; | |
9109fb7b | 7316 | |
88fdf75d | 7317 | /* pg_data_t should be reset to zero when it's allocated */ |
97a225e6 | 7318 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx); |
88fdf75d | 7319 | |
854e8848 | 7320 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
88fdf75d | 7321 | |
1da177e4 | 7322 | pgdat->node_id = nid; |
854e8848 | 7323 | pgdat->node_start_pfn = start_pfn; |
75ef7184 | 7324 | pgdat->per_cpu_nodestats = NULL; |
854e8848 | 7325 | |
8d29e18a | 7326 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
7327 | (u64)start_pfn << PAGE_SHIFT, |
7328 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
854e8848 | 7329 | calculate_node_totalpages(pgdat, start_pfn, end_pfn); |
1da177e4 LT |
7330 | |
7331 | alloc_node_mem_map(pgdat); | |
0188dc98 | 7332 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 7333 | |
7f3eb55b | 7334 | free_area_init_core(pgdat); |
1da177e4 LT |
7335 | } |
7336 | ||
bc9331a1 | 7337 | void __init free_area_init_memoryless_node(int nid) |
3f08a302 | 7338 | { |
854e8848 | 7339 | free_area_init_node(nid); |
3f08a302 MR |
7340 | } |
7341 | ||
418508c1 MS |
7342 | #if MAX_NUMNODES > 1 |
7343 | /* | |
7344 | * Figure out the number of possible node ids. | |
7345 | */ | |
f9872caf | 7346 | void __init setup_nr_node_ids(void) |
418508c1 | 7347 | { |
904a9553 | 7348 | unsigned int highest; |
418508c1 | 7349 | |
904a9553 | 7350 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7351 | nr_node_ids = highest + 1; |
7352 | } | |
418508c1 MS |
7353 | #endif |
7354 | ||
1e01979c TH |
7355 | /** |
7356 | * node_map_pfn_alignment - determine the maximum internode alignment | |
7357 | * | |
7358 | * This function should be called after node map is populated and sorted. | |
7359 | * It calculates the maximum power of two alignment which can distinguish | |
7360 | * all the nodes. | |
7361 | * | |
7362 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
7363 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
7364 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
7365 | * shifted, 1GiB is enough and this function will indicate so. | |
7366 | * | |
7367 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
7368 | * model has fine enough granularity to avoid incorrect mapping for the | |
7369 | * populated node map. | |
7370 | * | |
a862f68a | 7371 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
7372 | * requirement (single node). |
7373 | */ | |
7374 | unsigned long __init node_map_pfn_alignment(void) | |
7375 | { | |
7376 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 7377 | unsigned long start, end, mask; |
98fa15f3 | 7378 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 7379 | int i, nid; |
1e01979c | 7380 | |
c13291a5 | 7381 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
7382 | if (!start || last_nid < 0 || last_nid == nid) { |
7383 | last_nid = nid; | |
7384 | last_end = end; | |
7385 | continue; | |
7386 | } | |
7387 | ||
7388 | /* | |
7389 | * Start with a mask granular enough to pin-point to the | |
7390 | * start pfn and tick off bits one-by-one until it becomes | |
7391 | * too coarse to separate the current node from the last. | |
7392 | */ | |
7393 | mask = ~((1 << __ffs(start)) - 1); | |
7394 | while (mask && last_end <= (start & (mask << 1))) | |
7395 | mask <<= 1; | |
7396 | ||
7397 | /* accumulate all internode masks */ | |
7398 | accl_mask |= mask; | |
7399 | } | |
7400 | ||
7401 | /* convert mask to number of pages */ | |
7402 | return ~accl_mask + 1; | |
7403 | } | |
7404 | ||
c713216d MG |
7405 | /** |
7406 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
7407 | * | |
a862f68a | 7408 | * Return: the minimum PFN based on information provided via |
7d018176 | 7409 | * memblock_set_node(). |
c713216d MG |
7410 | */ |
7411 | unsigned long __init find_min_pfn_with_active_regions(void) | |
7412 | { | |
8a1b25fe | 7413 | return PHYS_PFN(memblock_start_of_DRAM()); |
c713216d MG |
7414 | } |
7415 | ||
37b07e41 LS |
7416 | /* |
7417 | * early_calculate_totalpages() | |
7418 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 7419 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 7420 | */ |
484f51f8 | 7421 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 7422 | { |
7e63efef | 7423 | unsigned long totalpages = 0; |
c13291a5 TH |
7424 | unsigned long start_pfn, end_pfn; |
7425 | int i, nid; | |
7426 | ||
7427 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
7428 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 7429 | |
37b07e41 LS |
7430 | totalpages += pages; |
7431 | if (pages) | |
4b0ef1fe | 7432 | node_set_state(nid, N_MEMORY); |
37b07e41 | 7433 | } |
b8af2941 | 7434 | return totalpages; |
7e63efef MG |
7435 | } |
7436 | ||
2a1e274a MG |
7437 | /* |
7438 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
7439 | * is spread evenly between nodes as long as the nodes have enough | |
7440 | * memory. When they don't, some nodes will have more kernelcore than | |
7441 | * others | |
7442 | */ | |
b224ef85 | 7443 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
7444 | { |
7445 | int i, nid; | |
7446 | unsigned long usable_startpfn; | |
7447 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 7448 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 7449 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 7450 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 7451 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 7452 | struct memblock_region *r; |
b2f3eebe TC |
7453 | |
7454 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
7455 | find_usable_zone_for_movable(); | |
7456 | ||
7457 | /* | |
7458 | * If movable_node is specified, ignore kernelcore and movablecore | |
7459 | * options. | |
7460 | */ | |
7461 | if (movable_node_is_enabled()) { | |
cc6de168 | 7462 | for_each_mem_region(r) { |
136199f0 | 7463 | if (!memblock_is_hotpluggable(r)) |
b2f3eebe TC |
7464 | continue; |
7465 | ||
d622abf7 | 7466 | nid = memblock_get_region_node(r); |
b2f3eebe | 7467 | |
136199f0 | 7468 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
7469 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
7470 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7471 | usable_startpfn; | |
7472 | } | |
7473 | ||
7474 | goto out2; | |
7475 | } | |
2a1e274a | 7476 | |
342332e6 TI |
7477 | /* |
7478 | * If kernelcore=mirror is specified, ignore movablecore option | |
7479 | */ | |
7480 | if (mirrored_kernelcore) { | |
7481 | bool mem_below_4gb_not_mirrored = false; | |
7482 | ||
cc6de168 | 7483 | for_each_mem_region(r) { |
342332e6 TI |
7484 | if (memblock_is_mirror(r)) |
7485 | continue; | |
7486 | ||
d622abf7 | 7487 | nid = memblock_get_region_node(r); |
342332e6 TI |
7488 | |
7489 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
7490 | ||
7491 | if (usable_startpfn < 0x100000) { | |
7492 | mem_below_4gb_not_mirrored = true; | |
7493 | continue; | |
7494 | } | |
7495 | ||
7496 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
7497 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7498 | usable_startpfn; | |
7499 | } | |
7500 | ||
7501 | if (mem_below_4gb_not_mirrored) | |
633bf2fe | 7502 | pr_warn("This configuration results in unmirrored kernel memory.\n"); |
342332e6 TI |
7503 | |
7504 | goto out2; | |
7505 | } | |
7506 | ||
7e63efef | 7507 | /* |
a5c6d650 DR |
7508 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
7509 | * amount of necessary memory. | |
7510 | */ | |
7511 | if (required_kernelcore_percent) | |
7512 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
7513 | 10000UL; | |
7514 | if (required_movablecore_percent) | |
7515 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
7516 | 10000UL; | |
7517 | ||
7518 | /* | |
7519 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
7520 | * kernelcore that corresponds so that memory usable for |
7521 | * any allocation type is evenly spread. If both kernelcore | |
7522 | * and movablecore are specified, then the value of kernelcore | |
7523 | * will be used for required_kernelcore if it's greater than | |
7524 | * what movablecore would have allowed. | |
7525 | */ | |
7526 | if (required_movablecore) { | |
7e63efef MG |
7527 | unsigned long corepages; |
7528 | ||
7529 | /* | |
7530 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
7531 | * was requested by the user | |
7532 | */ | |
7533 | required_movablecore = | |
7534 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 7535 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
7536 | corepages = totalpages - required_movablecore; |
7537 | ||
7538 | required_kernelcore = max(required_kernelcore, corepages); | |
7539 | } | |
7540 | ||
bde304bd XQ |
7541 | /* |
7542 | * If kernelcore was not specified or kernelcore size is larger | |
7543 | * than totalpages, there is no ZONE_MOVABLE. | |
7544 | */ | |
7545 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 7546 | goto out; |
2a1e274a MG |
7547 | |
7548 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
7549 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
7550 | ||
7551 | restart: | |
7552 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
7553 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 7554 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
7555 | unsigned long start_pfn, end_pfn; |
7556 | ||
2a1e274a MG |
7557 | /* |
7558 | * Recalculate kernelcore_node if the division per node | |
7559 | * now exceeds what is necessary to satisfy the requested | |
7560 | * amount of memory for the kernel | |
7561 | */ | |
7562 | if (required_kernelcore < kernelcore_node) | |
7563 | kernelcore_node = required_kernelcore / usable_nodes; | |
7564 | ||
7565 | /* | |
7566 | * As the map is walked, we track how much memory is usable | |
7567 | * by the kernel using kernelcore_remaining. When it is | |
7568 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
7569 | */ | |
7570 | kernelcore_remaining = kernelcore_node; | |
7571 | ||
7572 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 7573 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
7574 | unsigned long size_pages; |
7575 | ||
c13291a5 | 7576 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
7577 | if (start_pfn >= end_pfn) |
7578 | continue; | |
7579 | ||
7580 | /* Account for what is only usable for kernelcore */ | |
7581 | if (start_pfn < usable_startpfn) { | |
7582 | unsigned long kernel_pages; | |
7583 | kernel_pages = min(end_pfn, usable_startpfn) | |
7584 | - start_pfn; | |
7585 | ||
7586 | kernelcore_remaining -= min(kernel_pages, | |
7587 | kernelcore_remaining); | |
7588 | required_kernelcore -= min(kernel_pages, | |
7589 | required_kernelcore); | |
7590 | ||
7591 | /* Continue if range is now fully accounted */ | |
7592 | if (end_pfn <= usable_startpfn) { | |
7593 | ||
7594 | /* | |
7595 | * Push zone_movable_pfn to the end so | |
7596 | * that if we have to rebalance | |
7597 | * kernelcore across nodes, we will | |
7598 | * not double account here | |
7599 | */ | |
7600 | zone_movable_pfn[nid] = end_pfn; | |
7601 | continue; | |
7602 | } | |
7603 | start_pfn = usable_startpfn; | |
7604 | } | |
7605 | ||
7606 | /* | |
7607 | * The usable PFN range for ZONE_MOVABLE is from | |
7608 | * start_pfn->end_pfn. Calculate size_pages as the | |
7609 | * number of pages used as kernelcore | |
7610 | */ | |
7611 | size_pages = end_pfn - start_pfn; | |
7612 | if (size_pages > kernelcore_remaining) | |
7613 | size_pages = kernelcore_remaining; | |
7614 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
7615 | ||
7616 | /* | |
7617 | * Some kernelcore has been met, update counts and | |
7618 | * break if the kernelcore for this node has been | |
b8af2941 | 7619 | * satisfied |
2a1e274a MG |
7620 | */ |
7621 | required_kernelcore -= min(required_kernelcore, | |
7622 | size_pages); | |
7623 | kernelcore_remaining -= size_pages; | |
7624 | if (!kernelcore_remaining) | |
7625 | break; | |
7626 | } | |
7627 | } | |
7628 | ||
7629 | /* | |
7630 | * If there is still required_kernelcore, we do another pass with one | |
7631 | * less node in the count. This will push zone_movable_pfn[nid] further | |
7632 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 7633 | * satisfied |
2a1e274a MG |
7634 | */ |
7635 | usable_nodes--; | |
7636 | if (usable_nodes && required_kernelcore > usable_nodes) | |
7637 | goto restart; | |
7638 | ||
b2f3eebe | 7639 | out2: |
2a1e274a MG |
7640 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
7641 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
7642 | zone_movable_pfn[nid] = | |
7643 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 7644 | |
20e6926d | 7645 | out: |
66918dcd | 7646 | /* restore the node_state */ |
4b0ef1fe | 7647 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
7648 | } |
7649 | ||
4b0ef1fe LJ |
7650 | /* Any regular or high memory on that node ? */ |
7651 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 7652 | { |
37b07e41 LS |
7653 | enum zone_type zone_type; |
7654 | ||
4b0ef1fe | 7655 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 7656 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 7657 | if (populated_zone(zone)) { |
7b0e0c0e OS |
7658 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
7659 | node_set_state(nid, N_HIGH_MEMORY); | |
7660 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 7661 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
7662 | break; |
7663 | } | |
37b07e41 | 7664 | } |
37b07e41 LS |
7665 | } |
7666 | ||
51930df5 MR |
7667 | /* |
7668 | * Some architecturs, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For | |
7669 | * such cases we allow max_zone_pfn sorted in the descending order | |
7670 | */ | |
7671 | bool __weak arch_has_descending_max_zone_pfns(void) | |
7672 | { | |
7673 | return false; | |
7674 | } | |
7675 | ||
c713216d | 7676 | /** |
9691a071 | 7677 | * free_area_init - Initialise all pg_data_t and zone data |
88ca3b94 | 7678 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
7679 | * |
7680 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 7681 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
7682 | * zone in each node and their holes is calculated. If the maximum PFN |
7683 | * between two adjacent zones match, it is assumed that the zone is empty. | |
7684 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
7685 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
7686 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
7687 | * at arch_max_dma_pfn. | |
7688 | */ | |
9691a071 | 7689 | void __init free_area_init(unsigned long *max_zone_pfn) |
c713216d | 7690 | { |
c13291a5 | 7691 | unsigned long start_pfn, end_pfn; |
51930df5 MR |
7692 | int i, nid, zone; |
7693 | bool descending; | |
a6af2bc3 | 7694 | |
c713216d MG |
7695 | /* Record where the zone boundaries are */ |
7696 | memset(arch_zone_lowest_possible_pfn, 0, | |
7697 | sizeof(arch_zone_lowest_possible_pfn)); | |
7698 | memset(arch_zone_highest_possible_pfn, 0, | |
7699 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
7700 | |
7701 | start_pfn = find_min_pfn_with_active_regions(); | |
51930df5 | 7702 | descending = arch_has_descending_max_zone_pfns(); |
90cae1fe OH |
7703 | |
7704 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
51930df5 MR |
7705 | if (descending) |
7706 | zone = MAX_NR_ZONES - i - 1; | |
7707 | else | |
7708 | zone = i; | |
7709 | ||
7710 | if (zone == ZONE_MOVABLE) | |
2a1e274a | 7711 | continue; |
90cae1fe | 7712 | |
51930df5 MR |
7713 | end_pfn = max(max_zone_pfn[zone], start_pfn); |
7714 | arch_zone_lowest_possible_pfn[zone] = start_pfn; | |
7715 | arch_zone_highest_possible_pfn[zone] = end_pfn; | |
90cae1fe OH |
7716 | |
7717 | start_pfn = end_pfn; | |
c713216d | 7718 | } |
2a1e274a MG |
7719 | |
7720 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
7721 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 7722 | find_zone_movable_pfns_for_nodes(); |
c713216d | 7723 | |
c713216d | 7724 | /* Print out the zone ranges */ |
f88dfff5 | 7725 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
7726 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7727 | if (i == ZONE_MOVABLE) | |
7728 | continue; | |
f88dfff5 | 7729 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
7730 | if (arch_zone_lowest_possible_pfn[i] == |
7731 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 7732 | pr_cont("empty\n"); |
72f0ba02 | 7733 | else |
8d29e18a JG |
7734 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
7735 | (u64)arch_zone_lowest_possible_pfn[i] | |
7736 | << PAGE_SHIFT, | |
7737 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 7738 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
7739 | } |
7740 | ||
7741 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 7742 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
7743 | for (i = 0; i < MAX_NUMNODES; i++) { |
7744 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
7745 | pr_info(" Node %d: %#018Lx\n", i, |
7746 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 7747 | } |
c713216d | 7748 | |
f46edbd1 DW |
7749 | /* |
7750 | * Print out the early node map, and initialize the | |
7751 | * subsection-map relative to active online memory ranges to | |
7752 | * enable future "sub-section" extensions of the memory map. | |
7753 | */ | |
f88dfff5 | 7754 | pr_info("Early memory node ranges\n"); |
f46edbd1 | 7755 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
8d29e18a JG |
7756 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
7757 | (u64)start_pfn << PAGE_SHIFT, | |
7758 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
f46edbd1 DW |
7759 | subsection_map_init(start_pfn, end_pfn - start_pfn); |
7760 | } | |
c713216d MG |
7761 | |
7762 | /* Initialise every node */ | |
708614e6 | 7763 | mminit_verify_pageflags_layout(); |
8ef82866 | 7764 | setup_nr_node_ids(); |
c713216d MG |
7765 | for_each_online_node(nid) { |
7766 | pg_data_t *pgdat = NODE_DATA(nid); | |
854e8848 | 7767 | free_area_init_node(nid); |
37b07e41 LS |
7768 | |
7769 | /* Any memory on that node */ | |
7770 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
7771 | node_set_state(nid, N_MEMORY); |
7772 | check_for_memory(pgdat, nid); | |
c713216d MG |
7773 | } |
7774 | } | |
2a1e274a | 7775 | |
a5c6d650 DR |
7776 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
7777 | unsigned long *percent) | |
2a1e274a MG |
7778 | { |
7779 | unsigned long long coremem; | |
a5c6d650 DR |
7780 | char *endptr; |
7781 | ||
2a1e274a MG |
7782 | if (!p) |
7783 | return -EINVAL; | |
7784 | ||
a5c6d650 DR |
7785 | /* Value may be a percentage of total memory, otherwise bytes */ |
7786 | coremem = simple_strtoull(p, &endptr, 0); | |
7787 | if (*endptr == '%') { | |
7788 | /* Paranoid check for percent values greater than 100 */ | |
7789 | WARN_ON(coremem > 100); | |
2a1e274a | 7790 | |
a5c6d650 DR |
7791 | *percent = coremem; |
7792 | } else { | |
7793 | coremem = memparse(p, &p); | |
7794 | /* Paranoid check that UL is enough for the coremem value */ | |
7795 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7796 | |
a5c6d650 DR |
7797 | *core = coremem >> PAGE_SHIFT; |
7798 | *percent = 0UL; | |
7799 | } | |
2a1e274a MG |
7800 | return 0; |
7801 | } | |
ed7ed365 | 7802 | |
7e63efef MG |
7803 | /* |
7804 | * kernelcore=size sets the amount of memory for use for allocations that | |
7805 | * cannot be reclaimed or migrated. | |
7806 | */ | |
7807 | static int __init cmdline_parse_kernelcore(char *p) | |
7808 | { | |
342332e6 TI |
7809 | /* parse kernelcore=mirror */ |
7810 | if (parse_option_str(p, "mirror")) { | |
7811 | mirrored_kernelcore = true; | |
7812 | return 0; | |
7813 | } | |
7814 | ||
a5c6d650 DR |
7815 | return cmdline_parse_core(p, &required_kernelcore, |
7816 | &required_kernelcore_percent); | |
7e63efef MG |
7817 | } |
7818 | ||
7819 | /* | |
7820 | * movablecore=size sets the amount of memory for use for allocations that | |
7821 | * can be reclaimed or migrated. | |
7822 | */ | |
7823 | static int __init cmdline_parse_movablecore(char *p) | |
7824 | { | |
a5c6d650 DR |
7825 | return cmdline_parse_core(p, &required_movablecore, |
7826 | &required_movablecore_percent); | |
7e63efef MG |
7827 | } |
7828 | ||
ed7ed365 | 7829 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7830 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7831 | |
c3d5f5f0 JL |
7832 | void adjust_managed_page_count(struct page *page, long count) |
7833 | { | |
9705bea5 | 7834 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 7835 | totalram_pages_add(count); |
3dcc0571 JL |
7836 | #ifdef CONFIG_HIGHMEM |
7837 | if (PageHighMem(page)) | |
ca79b0c2 | 7838 | totalhigh_pages_add(count); |
3dcc0571 | 7839 | #endif |
c3d5f5f0 | 7840 | } |
3dcc0571 | 7841 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7842 | |
e5cb113f | 7843 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 7844 | { |
11199692 JL |
7845 | void *pos; |
7846 | unsigned long pages = 0; | |
69afade7 | 7847 | |
11199692 JL |
7848 | start = (void *)PAGE_ALIGN((unsigned long)start); |
7849 | end = (void *)((unsigned long)end & PAGE_MASK); | |
7850 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
7851 | struct page *page = virt_to_page(pos); |
7852 | void *direct_map_addr; | |
7853 | ||
7854 | /* | |
7855 | * 'direct_map_addr' might be different from 'pos' | |
7856 | * because some architectures' virt_to_page() | |
7857 | * work with aliases. Getting the direct map | |
7858 | * address ensures that we get a _writeable_ | |
7859 | * alias for the memset(). | |
7860 | */ | |
7861 | direct_map_addr = page_address(page); | |
c746170d VF |
7862 | /* |
7863 | * Perform a kasan-unchecked memset() since this memory | |
7864 | * has not been initialized. | |
7865 | */ | |
7866 | direct_map_addr = kasan_reset_tag(direct_map_addr); | |
dbe67df4 | 7867 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
7868 | memset(direct_map_addr, poison, PAGE_SIZE); |
7869 | ||
7870 | free_reserved_page(page); | |
69afade7 JL |
7871 | } |
7872 | ||
7873 | if (pages && s) | |
adb1fe9a JP |
7874 | pr_info("Freeing %s memory: %ldK\n", |
7875 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
7876 | |
7877 | return pages; | |
7878 | } | |
7879 | ||
1f9d03c5 | 7880 | void __init mem_init_print_info(void) |
7ee3d4e8 JL |
7881 | { |
7882 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7883 | unsigned long init_code_size, init_data_size; | |
7884 | ||
7885 | physpages = get_num_physpages(); | |
7886 | codesize = _etext - _stext; | |
7887 | datasize = _edata - _sdata; | |
7888 | rosize = __end_rodata - __start_rodata; | |
7889 | bss_size = __bss_stop - __bss_start; | |
7890 | init_data_size = __init_end - __init_begin; | |
7891 | init_code_size = _einittext - _sinittext; | |
7892 | ||
7893 | /* | |
7894 | * Detect special cases and adjust section sizes accordingly: | |
7895 | * 1) .init.* may be embedded into .data sections | |
7896 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7897 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7898 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7899 | */ | |
7900 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7901 | do { \ |
7902 | if (start <= pos && pos < end && size > adj) \ | |
7903 | size -= adj; \ | |
7904 | } while (0) | |
7ee3d4e8 JL |
7905 | |
7906 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7907 | _sinittext, init_code_size); | |
7908 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7909 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7910 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7911 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7912 | ||
7913 | #undef adj_init_size | |
7914 | ||
756a025f | 7915 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7916 | #ifdef CONFIG_HIGHMEM |
756a025f | 7917 | ", %luK highmem" |
7ee3d4e8 | 7918 | #endif |
1f9d03c5 | 7919 | ")\n", |
756a025f JP |
7920 | nr_free_pages() << (PAGE_SHIFT - 10), |
7921 | physpages << (PAGE_SHIFT - 10), | |
7922 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7923 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
ca79b0c2 | 7924 | (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10), |
1f9d03c5 | 7925 | totalcma_pages << (PAGE_SHIFT - 10) |
7ee3d4e8 | 7926 | #ifdef CONFIG_HIGHMEM |
1f9d03c5 | 7927 | , totalhigh_pages() << (PAGE_SHIFT - 10) |
7ee3d4e8 | 7928 | #endif |
1f9d03c5 | 7929 | ); |
7ee3d4e8 JL |
7930 | } |
7931 | ||
0e0b864e | 7932 | /** |
88ca3b94 RD |
7933 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7934 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7935 | * |
013110a7 | 7936 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7937 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7938 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7939 | * function may optionally be used to account for unfreeable pages in the |
7940 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7941 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7942 | */ |
7943 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7944 | { | |
7945 | dma_reserve = new_dma_reserve; | |
7946 | } | |
7947 | ||
005fd4bb | 7948 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7949 | { |
1da177e4 | 7950 | |
005fd4bb SAS |
7951 | lru_add_drain_cpu(cpu); |
7952 | drain_pages(cpu); | |
9f8f2172 | 7953 | |
005fd4bb SAS |
7954 | /* |
7955 | * Spill the event counters of the dead processor | |
7956 | * into the current processors event counters. | |
7957 | * This artificially elevates the count of the current | |
7958 | * processor. | |
7959 | */ | |
7960 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7961 | |
005fd4bb SAS |
7962 | /* |
7963 | * Zero the differential counters of the dead processor | |
7964 | * so that the vm statistics are consistent. | |
7965 | * | |
7966 | * This is only okay since the processor is dead and cannot | |
7967 | * race with what we are doing. | |
7968 | */ | |
7969 | cpu_vm_stats_fold(cpu); | |
7970 | return 0; | |
1da177e4 | 7971 | } |
1da177e4 | 7972 | |
e03a5125 NP |
7973 | #ifdef CONFIG_NUMA |
7974 | int hashdist = HASHDIST_DEFAULT; | |
7975 | ||
7976 | static int __init set_hashdist(char *str) | |
7977 | { | |
7978 | if (!str) | |
7979 | return 0; | |
7980 | hashdist = simple_strtoul(str, &str, 0); | |
7981 | return 1; | |
7982 | } | |
7983 | __setup("hashdist=", set_hashdist); | |
7984 | #endif | |
7985 | ||
1da177e4 LT |
7986 | void __init page_alloc_init(void) |
7987 | { | |
005fd4bb SAS |
7988 | int ret; |
7989 | ||
e03a5125 NP |
7990 | #ifdef CONFIG_NUMA |
7991 | if (num_node_state(N_MEMORY) == 1) | |
7992 | hashdist = 0; | |
7993 | #endif | |
7994 | ||
005fd4bb SAS |
7995 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, |
7996 | "mm/page_alloc:dead", NULL, | |
7997 | page_alloc_cpu_dead); | |
7998 | WARN_ON(ret < 0); | |
1da177e4 LT |
7999 | } |
8000 | ||
cb45b0e9 | 8001 | /* |
34b10060 | 8002 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
8003 | * or min_free_kbytes changes. |
8004 | */ | |
8005 | static void calculate_totalreserve_pages(void) | |
8006 | { | |
8007 | struct pglist_data *pgdat; | |
8008 | unsigned long reserve_pages = 0; | |
2f6726e5 | 8009 | enum zone_type i, j; |
cb45b0e9 HA |
8010 | |
8011 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
8012 | |
8013 | pgdat->totalreserve_pages = 0; | |
8014 | ||
cb45b0e9 HA |
8015 | for (i = 0; i < MAX_NR_ZONES; i++) { |
8016 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 8017 | long max = 0; |
9705bea5 | 8018 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
8019 | |
8020 | /* Find valid and maximum lowmem_reserve in the zone */ | |
8021 | for (j = i; j < MAX_NR_ZONES; j++) { | |
8022 | if (zone->lowmem_reserve[j] > max) | |
8023 | max = zone->lowmem_reserve[j]; | |
8024 | } | |
8025 | ||
41858966 MG |
8026 | /* we treat the high watermark as reserved pages. */ |
8027 | max += high_wmark_pages(zone); | |
cb45b0e9 | 8028 | |
3d6357de AK |
8029 | if (max > managed_pages) |
8030 | max = managed_pages; | |
a8d01437 | 8031 | |
281e3726 | 8032 | pgdat->totalreserve_pages += max; |
a8d01437 | 8033 | |
cb45b0e9 HA |
8034 | reserve_pages += max; |
8035 | } | |
8036 | } | |
8037 | totalreserve_pages = reserve_pages; | |
8038 | } | |
8039 | ||
1da177e4 LT |
8040 | /* |
8041 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 8042 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
8043 | * has a correct pages reserved value, so an adequate number of |
8044 | * pages are left in the zone after a successful __alloc_pages(). | |
8045 | */ | |
8046 | static void setup_per_zone_lowmem_reserve(void) | |
8047 | { | |
8048 | struct pglist_data *pgdat; | |
470c61d7 | 8049 | enum zone_type i, j; |
1da177e4 | 8050 | |
ec936fc5 | 8051 | for_each_online_pgdat(pgdat) { |
470c61d7 LS |
8052 | for (i = 0; i < MAX_NR_ZONES - 1; i++) { |
8053 | struct zone *zone = &pgdat->node_zones[i]; | |
8054 | int ratio = sysctl_lowmem_reserve_ratio[i]; | |
8055 | bool clear = !ratio || !zone_managed_pages(zone); | |
8056 | unsigned long managed_pages = 0; | |
8057 | ||
8058 | for (j = i + 1; j < MAX_NR_ZONES; j++) { | |
8059 | if (clear) { | |
8060 | zone->lowmem_reserve[j] = 0; | |
d3cda233 | 8061 | } else { |
470c61d7 LS |
8062 | struct zone *upper_zone = &pgdat->node_zones[j]; |
8063 | ||
8064 | managed_pages += zone_managed_pages(upper_zone); | |
8065 | zone->lowmem_reserve[j] = managed_pages / ratio; | |
d3cda233 | 8066 | } |
1da177e4 LT |
8067 | } |
8068 | } | |
8069 | } | |
cb45b0e9 HA |
8070 | |
8071 | /* update totalreserve_pages */ | |
8072 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8073 | } |
8074 | ||
cfd3da1e | 8075 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
8076 | { |
8077 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
8078 | unsigned long lowmem_pages = 0; | |
8079 | struct zone *zone; | |
8080 | unsigned long flags; | |
8081 | ||
8082 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
8083 | for_each_zone(zone) { | |
8084 | if (!is_highmem(zone)) | |
9705bea5 | 8085 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
8086 | } |
8087 | ||
8088 | for_each_zone(zone) { | |
ac924c60 AM |
8089 | u64 tmp; |
8090 | ||
1125b4e3 | 8091 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 8092 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 8093 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
8094 | if (is_highmem(zone)) { |
8095 | /* | |
669ed175 NP |
8096 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
8097 | * need highmem pages, so cap pages_min to a small | |
8098 | * value here. | |
8099 | * | |
41858966 | 8100 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 8101 | * deltas control async page reclaim, and so should |
669ed175 | 8102 | * not be capped for highmem. |
1da177e4 | 8103 | */ |
90ae8d67 | 8104 | unsigned long min_pages; |
1da177e4 | 8105 | |
9705bea5 | 8106 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 8107 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 8108 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 8109 | } else { |
669ed175 NP |
8110 | /* |
8111 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
8112 | * proportionate to the zone's size. |
8113 | */ | |
a9214443 | 8114 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
8115 | } |
8116 | ||
795ae7a0 JW |
8117 | /* |
8118 | * Set the kswapd watermarks distance according to the | |
8119 | * scale factor in proportion to available memory, but | |
8120 | * ensure a minimum size on small systems. | |
8121 | */ | |
8122 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 8123 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
8124 | watermark_scale_factor, 10000)); |
8125 | ||
aa092591 | 8126 | zone->watermark_boost = 0; |
a9214443 MG |
8127 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
8128 | zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 8129 | |
1125b4e3 | 8130 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 8131 | } |
cb45b0e9 HA |
8132 | |
8133 | /* update totalreserve_pages */ | |
8134 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8135 | } |
8136 | ||
cfd3da1e MG |
8137 | /** |
8138 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
8139 | * or when memory is hot-{added|removed} | |
8140 | * | |
8141 | * Ensures that the watermark[min,low,high] values for each zone are set | |
8142 | * correctly with respect to min_free_kbytes. | |
8143 | */ | |
8144 | void setup_per_zone_wmarks(void) | |
8145 | { | |
b93e0f32 MH |
8146 | static DEFINE_SPINLOCK(lock); |
8147 | ||
8148 | spin_lock(&lock); | |
cfd3da1e | 8149 | __setup_per_zone_wmarks(); |
b93e0f32 | 8150 | spin_unlock(&lock); |
cfd3da1e MG |
8151 | } |
8152 | ||
1da177e4 LT |
8153 | /* |
8154 | * Initialise min_free_kbytes. | |
8155 | * | |
8156 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 8157 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
8158 | * bandwidth does not increase linearly with machine size. We use |
8159 | * | |
b8af2941 | 8160 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
8161 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
8162 | * | |
8163 | * which yields | |
8164 | * | |
8165 | * 16MB: 512k | |
8166 | * 32MB: 724k | |
8167 | * 64MB: 1024k | |
8168 | * 128MB: 1448k | |
8169 | * 256MB: 2048k | |
8170 | * 512MB: 2896k | |
8171 | * 1024MB: 4096k | |
8172 | * 2048MB: 5792k | |
8173 | * 4096MB: 8192k | |
8174 | * 8192MB: 11584k | |
8175 | * 16384MB: 16384k | |
8176 | */ | |
1b79acc9 | 8177 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
8178 | { |
8179 | unsigned long lowmem_kbytes; | |
5f12733e | 8180 | int new_min_free_kbytes; |
1da177e4 LT |
8181 | |
8182 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
8183 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
8184 | ||
8185 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
8186 | min_free_kbytes = new_min_free_kbytes; | |
8187 | if (min_free_kbytes < 128) | |
8188 | min_free_kbytes = 128; | |
ee8eb9a5 JS |
8189 | if (min_free_kbytes > 262144) |
8190 | min_free_kbytes = 262144; | |
5f12733e MH |
8191 | } else { |
8192 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
8193 | new_min_free_kbytes, user_min_free_kbytes); | |
8194 | } | |
bc75d33f | 8195 | setup_per_zone_wmarks(); |
a6cccdc3 | 8196 | refresh_zone_stat_thresholds(); |
1da177e4 | 8197 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
8198 | |
8199 | #ifdef CONFIG_NUMA | |
8200 | setup_min_unmapped_ratio(); | |
8201 | setup_min_slab_ratio(); | |
8202 | #endif | |
8203 | ||
4aab2be0 VB |
8204 | khugepaged_min_free_kbytes_update(); |
8205 | ||
1da177e4 LT |
8206 | return 0; |
8207 | } | |
e08d3fdf | 8208 | postcore_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
8209 | |
8210 | /* | |
b8af2941 | 8211 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
8212 | * that we can call two helper functions whenever min_free_kbytes |
8213 | * changes. | |
8214 | */ | |
cccad5b9 | 8215 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8216 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8217 | { |
da8c757b HP |
8218 | int rc; |
8219 | ||
8220 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8221 | if (rc) | |
8222 | return rc; | |
8223 | ||
5f12733e MH |
8224 | if (write) { |
8225 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 8226 | setup_per_zone_wmarks(); |
5f12733e | 8227 | } |
1da177e4 LT |
8228 | return 0; |
8229 | } | |
8230 | ||
795ae7a0 | 8231 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8232 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
8233 | { |
8234 | int rc; | |
8235 | ||
8236 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8237 | if (rc) | |
8238 | return rc; | |
8239 | ||
8240 | if (write) | |
8241 | setup_per_zone_wmarks(); | |
8242 | ||
8243 | return 0; | |
8244 | } | |
8245 | ||
9614634f | 8246 | #ifdef CONFIG_NUMA |
6423aa81 | 8247 | static void setup_min_unmapped_ratio(void) |
9614634f | 8248 | { |
6423aa81 | 8249 | pg_data_t *pgdat; |
9614634f | 8250 | struct zone *zone; |
9614634f | 8251 | |
a5f5f91d | 8252 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 8253 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 8254 | |
9614634f | 8255 | for_each_zone(zone) |
9705bea5 AK |
8256 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
8257 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 8258 | } |
0ff38490 | 8259 | |
6423aa81 JK |
8260 | |
8261 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8262 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 8263 | { |
0ff38490 CL |
8264 | int rc; |
8265 | ||
8d65af78 | 8266 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
8267 | if (rc) |
8268 | return rc; | |
8269 | ||
6423aa81 JK |
8270 | setup_min_unmapped_ratio(); |
8271 | ||
8272 | return 0; | |
8273 | } | |
8274 | ||
8275 | static void setup_min_slab_ratio(void) | |
8276 | { | |
8277 | pg_data_t *pgdat; | |
8278 | struct zone *zone; | |
8279 | ||
a5f5f91d MG |
8280 | for_each_online_pgdat(pgdat) |
8281 | pgdat->min_slab_pages = 0; | |
8282 | ||
0ff38490 | 8283 | for_each_zone(zone) |
9705bea5 AK |
8284 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
8285 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
8286 | } |
8287 | ||
8288 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8289 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
8290 | { |
8291 | int rc; | |
8292 | ||
8293 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8294 | if (rc) | |
8295 | return rc; | |
8296 | ||
8297 | setup_min_slab_ratio(); | |
8298 | ||
0ff38490 CL |
8299 | return 0; |
8300 | } | |
9614634f CL |
8301 | #endif |
8302 | ||
1da177e4 LT |
8303 | /* |
8304 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
8305 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
8306 | * whenever sysctl_lowmem_reserve_ratio changes. | |
8307 | * | |
8308 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 8309 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
8310 | * if in function of the boot time zone sizes. |
8311 | */ | |
cccad5b9 | 8312 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8313 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8314 | { |
86aaf255 BH |
8315 | int i; |
8316 | ||
8d65af78 | 8317 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
8318 | |
8319 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
8320 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
8321 | sysctl_lowmem_reserve_ratio[i] = 0; | |
8322 | } | |
8323 | ||
1da177e4 LT |
8324 | setup_per_zone_lowmem_reserve(); |
8325 | return 0; | |
8326 | } | |
8327 | ||
8ad4b1fb RS |
8328 | /* |
8329 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
8330 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
8331 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 8332 | */ |
cccad5b9 | 8333 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8334 | void *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
8335 | { |
8336 | struct zone *zone; | |
7cd2b0a3 | 8337 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
8338 | int ret; |
8339 | ||
7cd2b0a3 DR |
8340 | mutex_lock(&pcp_batch_high_lock); |
8341 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
8342 | ||
8d65af78 | 8343 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
8344 | if (!write || ret < 0) |
8345 | goto out; | |
8346 | ||
8347 | /* Sanity checking to avoid pcp imbalance */ | |
8348 | if (percpu_pagelist_fraction && | |
8349 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
8350 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
8351 | ret = -EINVAL; | |
8352 | goto out; | |
8353 | } | |
8354 | ||
8355 | /* No change? */ | |
8356 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
8357 | goto out; | |
c8e251fa | 8358 | |
cb1ef534 | 8359 | for_each_populated_zone(zone) |
0a8b4f1d | 8360 | zone_set_pageset_high_and_batch(zone); |
7cd2b0a3 | 8361 | out: |
c8e251fa | 8362 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 8363 | return ret; |
8ad4b1fb RS |
8364 | } |
8365 | ||
f6f34b43 SD |
8366 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
8367 | /* | |
8368 | * Returns the number of pages that arch has reserved but | |
8369 | * is not known to alloc_large_system_hash(). | |
8370 | */ | |
8371 | static unsigned long __init arch_reserved_kernel_pages(void) | |
8372 | { | |
8373 | return 0; | |
8374 | } | |
8375 | #endif | |
8376 | ||
9017217b PT |
8377 | /* |
8378 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
8379 | * machines. As memory size is increased the scale is also increased but at | |
8380 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
8381 | * quadruples the scale is increased by one, which means the size of hash table | |
8382 | * only doubles, instead of quadrupling as well. | |
8383 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
8384 | * makes sense, it is disabled on such platforms. | |
8385 | */ | |
8386 | #if __BITS_PER_LONG > 32 | |
8387 | #define ADAPT_SCALE_BASE (64ul << 30) | |
8388 | #define ADAPT_SCALE_SHIFT 2 | |
8389 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
8390 | #endif | |
8391 | ||
1da177e4 LT |
8392 | /* |
8393 | * allocate a large system hash table from bootmem | |
8394 | * - it is assumed that the hash table must contain an exact power-of-2 | |
8395 | * quantity of entries | |
8396 | * - limit is the number of hash buckets, not the total allocation size | |
8397 | */ | |
8398 | void *__init alloc_large_system_hash(const char *tablename, | |
8399 | unsigned long bucketsize, | |
8400 | unsigned long numentries, | |
8401 | int scale, | |
8402 | int flags, | |
8403 | unsigned int *_hash_shift, | |
8404 | unsigned int *_hash_mask, | |
31fe62b9 TB |
8405 | unsigned long low_limit, |
8406 | unsigned long high_limit) | |
1da177e4 | 8407 | { |
31fe62b9 | 8408 | unsigned long long max = high_limit; |
1da177e4 LT |
8409 | unsigned long log2qty, size; |
8410 | void *table = NULL; | |
3749a8f0 | 8411 | gfp_t gfp_flags; |
ec11408a | 8412 | bool virt; |
121e6f32 | 8413 | bool huge; |
1da177e4 LT |
8414 | |
8415 | /* allow the kernel cmdline to have a say */ | |
8416 | if (!numentries) { | |
8417 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 8418 | numentries = nr_kernel_pages; |
f6f34b43 | 8419 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
8420 | |
8421 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
8422 | if (PAGE_SHIFT < 20) | |
8423 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 8424 | |
9017217b PT |
8425 | #if __BITS_PER_LONG > 32 |
8426 | if (!high_limit) { | |
8427 | unsigned long adapt; | |
8428 | ||
8429 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
8430 | adapt <<= ADAPT_SCALE_SHIFT) | |
8431 | scale++; | |
8432 | } | |
8433 | #endif | |
8434 | ||
1da177e4 LT |
8435 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
8436 | if (scale > PAGE_SHIFT) | |
8437 | numentries >>= (scale - PAGE_SHIFT); | |
8438 | else | |
8439 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
8440 | |
8441 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
8442 | if (unlikely(flags & HASH_SMALL)) { |
8443 | /* Makes no sense without HASH_EARLY */ | |
8444 | WARN_ON(!(flags & HASH_EARLY)); | |
8445 | if (!(numentries >> *_hash_shift)) { | |
8446 | numentries = 1UL << *_hash_shift; | |
8447 | BUG_ON(!numentries); | |
8448 | } | |
8449 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 8450 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 8451 | } |
6e692ed3 | 8452 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
8453 | |
8454 | /* limit allocation size to 1/16 total memory by default */ | |
8455 | if (max == 0) { | |
8456 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
8457 | do_div(max, bucketsize); | |
8458 | } | |
074b8517 | 8459 | max = min(max, 0x80000000ULL); |
1da177e4 | 8460 | |
31fe62b9 TB |
8461 | if (numentries < low_limit) |
8462 | numentries = low_limit; | |
1da177e4 LT |
8463 | if (numentries > max) |
8464 | numentries = max; | |
8465 | ||
f0d1b0b3 | 8466 | log2qty = ilog2(numentries); |
1da177e4 | 8467 | |
3749a8f0 | 8468 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 8469 | do { |
ec11408a | 8470 | virt = false; |
1da177e4 | 8471 | size = bucketsize << log2qty; |
ea1f5f37 PT |
8472 | if (flags & HASH_EARLY) { |
8473 | if (flags & HASH_ZERO) | |
26fb3dae | 8474 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 8475 | else |
7e1c4e27 MR |
8476 | table = memblock_alloc_raw(size, |
8477 | SMP_CACHE_BYTES); | |
ec11408a | 8478 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
88dca4ca | 8479 | table = __vmalloc(size, gfp_flags); |
ec11408a | 8480 | virt = true; |
121e6f32 | 8481 | huge = is_vm_area_hugepages(table); |
ea1f5f37 | 8482 | } else { |
1037b83b ED |
8483 | /* |
8484 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
8485 | * some pages at the end of hash table which |
8486 | * alloc_pages_exact() automatically does | |
1037b83b | 8487 | */ |
ec11408a NP |
8488 | table = alloc_pages_exact(size, gfp_flags); |
8489 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
8490 | } |
8491 | } while (!table && size > PAGE_SIZE && --log2qty); | |
8492 | ||
8493 | if (!table) | |
8494 | panic("Failed to allocate %s hash table\n", tablename); | |
8495 | ||
ec11408a NP |
8496 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
8497 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
121e6f32 | 8498 | virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear"); |
1da177e4 LT |
8499 | |
8500 | if (_hash_shift) | |
8501 | *_hash_shift = log2qty; | |
8502 | if (_hash_mask) | |
8503 | *_hash_mask = (1 << log2qty) - 1; | |
8504 | ||
8505 | return table; | |
8506 | } | |
a117e66e | 8507 | |
a5d76b54 | 8508 | /* |
80934513 | 8509 | * This function checks whether pageblock includes unmovable pages or not. |
80934513 | 8510 | * |
b8af2941 | 8511 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
8512 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
8513 | * check without lock_page also may miss some movable non-lru pages at | |
8514 | * race condition. So you can't expect this function should be exact. | |
4a55c047 QC |
8515 | * |
8516 | * Returns a page without holding a reference. If the caller wants to | |
047b9967 | 8517 | * dereference that page (e.g., dumping), it has to make sure that it |
4a55c047 QC |
8518 | * cannot get removed (e.g., via memory unplug) concurrently. |
8519 | * | |
a5d76b54 | 8520 | */ |
4a55c047 QC |
8521 | struct page *has_unmovable_pages(struct zone *zone, struct page *page, |
8522 | int migratetype, int flags) | |
49ac8255 | 8523 | { |
1a9f2191 QC |
8524 | unsigned long iter = 0; |
8525 | unsigned long pfn = page_to_pfn(page); | |
6a654e36 | 8526 | unsigned long offset = pfn % pageblock_nr_pages; |
47118af0 | 8527 | |
1a9f2191 QC |
8528 | if (is_migrate_cma_page(page)) { |
8529 | /* | |
8530 | * CMA allocations (alloc_contig_range) really need to mark | |
8531 | * isolate CMA pageblocks even when they are not movable in fact | |
8532 | * so consider them movable here. | |
8533 | */ | |
8534 | if (is_migrate_cma(migratetype)) | |
4a55c047 | 8535 | return NULL; |
1a9f2191 | 8536 | |
3d680bdf | 8537 | return page; |
1a9f2191 | 8538 | } |
4da2ce25 | 8539 | |
6a654e36 | 8540 | for (; iter < pageblock_nr_pages - offset; iter++) { |
fe4c86c9 | 8541 | if (!pfn_valid_within(pfn + iter)) |
49ac8255 | 8542 | continue; |
29723fcc | 8543 | |
fe4c86c9 | 8544 | page = pfn_to_page(pfn + iter); |
c8721bbb | 8545 | |
c9c510dc DH |
8546 | /* |
8547 | * Both, bootmem allocations and memory holes are marked | |
8548 | * PG_reserved and are unmovable. We can even have unmovable | |
8549 | * allocations inside ZONE_MOVABLE, for example when | |
8550 | * specifying "movablecore". | |
8551 | */ | |
d7ab3672 | 8552 | if (PageReserved(page)) |
3d680bdf | 8553 | return page; |
d7ab3672 | 8554 | |
9d789999 MH |
8555 | /* |
8556 | * If the zone is movable and we have ruled out all reserved | |
8557 | * pages then it should be reasonably safe to assume the rest | |
8558 | * is movable. | |
8559 | */ | |
8560 | if (zone_idx(zone) == ZONE_MOVABLE) | |
8561 | continue; | |
8562 | ||
c8721bbb NH |
8563 | /* |
8564 | * Hugepages are not in LRU lists, but they're movable. | |
1da2f328 | 8565 | * THPs are on the LRU, but need to be counted as #small pages. |
8bb4e7a2 | 8566 | * We need not scan over tail pages because we don't |
c8721bbb NH |
8567 | * handle each tail page individually in migration. |
8568 | */ | |
1da2f328 | 8569 | if (PageHuge(page) || PageTransCompound(page)) { |
17e2e7d7 OS |
8570 | struct page *head = compound_head(page); |
8571 | unsigned int skip_pages; | |
464c7ffb | 8572 | |
1da2f328 RR |
8573 | if (PageHuge(page)) { |
8574 | if (!hugepage_migration_supported(page_hstate(head))) | |
8575 | return page; | |
8576 | } else if (!PageLRU(head) && !__PageMovable(head)) { | |
3d680bdf | 8577 | return page; |
1da2f328 | 8578 | } |
464c7ffb | 8579 | |
d8c6546b | 8580 | skip_pages = compound_nr(head) - (page - head); |
17e2e7d7 | 8581 | iter += skip_pages - 1; |
c8721bbb NH |
8582 | continue; |
8583 | } | |
8584 | ||
97d255c8 MK |
8585 | /* |
8586 | * We can't use page_count without pin a page | |
8587 | * because another CPU can free compound page. | |
8588 | * This check already skips compound tails of THP | |
0139aa7b | 8589 | * because their page->_refcount is zero at all time. |
97d255c8 | 8590 | */ |
fe896d18 | 8591 | if (!page_ref_count(page)) { |
49ac8255 | 8592 | if (PageBuddy(page)) |
ab130f91 | 8593 | iter += (1 << buddy_order(page)) - 1; |
49ac8255 KH |
8594 | continue; |
8595 | } | |
97d255c8 | 8596 | |
b023f468 WC |
8597 | /* |
8598 | * The HWPoisoned page may be not in buddy system, and | |
8599 | * page_count() is not 0. | |
8600 | */ | |
756d25be | 8601 | if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
b023f468 WC |
8602 | continue; |
8603 | ||
aa218795 DH |
8604 | /* |
8605 | * We treat all PageOffline() pages as movable when offlining | |
8606 | * to give drivers a chance to decrement their reference count | |
8607 | * in MEM_GOING_OFFLINE in order to indicate that these pages | |
8608 | * can be offlined as there are no direct references anymore. | |
8609 | * For actually unmovable PageOffline() where the driver does | |
8610 | * not support this, we will fail later when trying to actually | |
8611 | * move these pages that still have a reference count > 0. | |
8612 | * (false negatives in this function only) | |
8613 | */ | |
8614 | if ((flags & MEMORY_OFFLINE) && PageOffline(page)) | |
8615 | continue; | |
8616 | ||
fe4c86c9 | 8617 | if (__PageMovable(page) || PageLRU(page)) |
0efadf48 YX |
8618 | continue; |
8619 | ||
49ac8255 | 8620 | /* |
6b4f7799 JW |
8621 | * If there are RECLAIMABLE pages, we need to check |
8622 | * it. But now, memory offline itself doesn't call | |
8623 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 | 8624 | */ |
3d680bdf | 8625 | return page; |
49ac8255 | 8626 | } |
4a55c047 | 8627 | return NULL; |
49ac8255 KH |
8628 | } |
8629 | ||
8df995f6 | 8630 | #ifdef CONFIG_CONTIG_ALLOC |
041d3a8c MN |
8631 | static unsigned long pfn_max_align_down(unsigned long pfn) |
8632 | { | |
8633 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8634 | pageblock_nr_pages) - 1); | |
8635 | } | |
8636 | ||
8637 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
8638 | { | |
8639 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8640 | pageblock_nr_pages)); | |
8641 | } | |
8642 | ||
a1394bdd MK |
8643 | #if defined(CONFIG_DYNAMIC_DEBUG) || \ |
8644 | (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) | |
8645 | /* Usage: See admin-guide/dynamic-debug-howto.rst */ | |
8646 | static void alloc_contig_dump_pages(struct list_head *page_list) | |
8647 | { | |
8648 | DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure"); | |
8649 | ||
8650 | if (DYNAMIC_DEBUG_BRANCH(descriptor)) { | |
8651 | struct page *page; | |
8652 | ||
8653 | dump_stack(); | |
8654 | list_for_each_entry(page, page_list, lru) | |
8655 | dump_page(page, "migration failure"); | |
8656 | } | |
8657 | } | |
8658 | #else | |
8659 | static inline void alloc_contig_dump_pages(struct list_head *page_list) | |
8660 | { | |
8661 | } | |
8662 | #endif | |
8663 | ||
041d3a8c | 8664 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
8665 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
8666 | unsigned long start, unsigned long end) | |
041d3a8c MN |
8667 | { |
8668 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 8669 | unsigned int nr_reclaimed; |
041d3a8c MN |
8670 | unsigned long pfn = start; |
8671 | unsigned int tries = 0; | |
8672 | int ret = 0; | |
8b94e0b8 JK |
8673 | struct migration_target_control mtc = { |
8674 | .nid = zone_to_nid(cc->zone), | |
8675 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
8676 | }; | |
041d3a8c | 8677 | |
be49a6e1 | 8678 | migrate_prep(); |
041d3a8c | 8679 | |
bb13ffeb | 8680 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
8681 | if (fatal_signal_pending(current)) { |
8682 | ret = -EINTR; | |
8683 | break; | |
8684 | } | |
8685 | ||
bb13ffeb MG |
8686 | if (list_empty(&cc->migratepages)) { |
8687 | cc->nr_migratepages = 0; | |
edc2ca61 | 8688 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
8689 | if (!pfn) { |
8690 | ret = -EINTR; | |
8691 | break; | |
8692 | } | |
8693 | tries = 0; | |
8694 | } else if (++tries == 5) { | |
8695 | ret = ret < 0 ? ret : -EBUSY; | |
8696 | break; | |
8697 | } | |
8698 | ||
beb51eaa MK |
8699 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
8700 | &cc->migratepages); | |
8701 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 8702 | |
8b94e0b8 JK |
8703 | ret = migrate_pages(&cc->migratepages, alloc_migration_target, |
8704 | NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE); | |
041d3a8c | 8705 | } |
2a6f5124 | 8706 | if (ret < 0) { |
a1394bdd | 8707 | alloc_contig_dump_pages(&cc->migratepages); |
2a6f5124 SP |
8708 | putback_movable_pages(&cc->migratepages); |
8709 | return ret; | |
8710 | } | |
8711 | return 0; | |
041d3a8c MN |
8712 | } |
8713 | ||
8714 | /** | |
8715 | * alloc_contig_range() -- tries to allocate given range of pages | |
8716 | * @start: start PFN to allocate | |
8717 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
8718 | * @migratetype: migratetype of the underlaying pageblocks (either |
8719 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
8720 | * in range must have the same migratetype and it must | |
8721 | * be either of the two. | |
ca96b625 | 8722 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
8723 | * |
8724 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 8725 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 8726 | * |
2c7452a0 MK |
8727 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
8728 | * pageblocks in the range. Once isolated, the pageblocks should not | |
8729 | * be modified by others. | |
041d3a8c | 8730 | * |
a862f68a | 8731 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
8732 | * pages which PFN is in [start, end) are allocated for the caller and |
8733 | * need to be freed with free_contig_range(). | |
8734 | */ | |
0815f3d8 | 8735 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 8736 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 8737 | { |
041d3a8c | 8738 | unsigned long outer_start, outer_end; |
d00181b9 KS |
8739 | unsigned int order; |
8740 | int ret = 0; | |
041d3a8c | 8741 | |
bb13ffeb MG |
8742 | struct compact_control cc = { |
8743 | .nr_migratepages = 0, | |
8744 | .order = -1, | |
8745 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 8746 | .mode = MIGRATE_SYNC, |
bb13ffeb | 8747 | .ignore_skip_hint = true, |
2583d671 | 8748 | .no_set_skip_hint = true, |
7dea19f9 | 8749 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 8750 | .alloc_contig = true, |
bb13ffeb MG |
8751 | }; |
8752 | INIT_LIST_HEAD(&cc.migratepages); | |
8753 | ||
041d3a8c MN |
8754 | /* |
8755 | * What we do here is we mark all pageblocks in range as | |
8756 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
8757 | * have different sizes, and due to the way page allocator | |
8758 | * work, we align the range to biggest of the two pages so | |
8759 | * that page allocator won't try to merge buddies from | |
8760 | * different pageblocks and change MIGRATE_ISOLATE to some | |
8761 | * other migration type. | |
8762 | * | |
8763 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
8764 | * migrate the pages from an unaligned range (ie. pages that | |
8765 | * we are interested in). This will put all the pages in | |
8766 | * range back to page allocator as MIGRATE_ISOLATE. | |
8767 | * | |
8768 | * When this is done, we take the pages in range from page | |
8769 | * allocator removing them from the buddy system. This way | |
8770 | * page allocator will never consider using them. | |
8771 | * | |
8772 | * This lets us mark the pageblocks back as | |
8773 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
8774 | * aligned range but not in the unaligned, original range are | |
8775 | * put back to page allocator so that buddy can use them. | |
8776 | */ | |
8777 | ||
8778 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
d381c547 | 8779 | pfn_max_align_up(end), migratetype, 0); |
3fa0c7c7 | 8780 | if (ret) |
86a595f9 | 8781 | return ret; |
041d3a8c | 8782 | |
7612921f VB |
8783 | drain_all_pages(cc.zone); |
8784 | ||
8ef5849f JK |
8785 | /* |
8786 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
8787 | * So, just fall through. test_pages_isolated() has a tracepoint |
8788 | * which will report the busy page. | |
8789 | * | |
8790 | * It is possible that busy pages could become available before | |
8791 | * the call to test_pages_isolated, and the range will actually be | |
8792 | * allocated. So, if we fall through be sure to clear ret so that | |
8793 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 8794 | */ |
bb13ffeb | 8795 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 8796 | if (ret && ret != -EBUSY) |
041d3a8c | 8797 | goto done; |
63cd4489 | 8798 | ret =0; |
041d3a8c MN |
8799 | |
8800 | /* | |
8801 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
8802 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
8803 | * more, all pages in [start, end) are free in page allocator. | |
8804 | * What we are going to do is to allocate all pages from | |
8805 | * [start, end) (that is remove them from page allocator). | |
8806 | * | |
8807 | * The only problem is that pages at the beginning and at the | |
8808 | * end of interesting range may be not aligned with pages that | |
8809 | * page allocator holds, ie. they can be part of higher order | |
8810 | * pages. Because of this, we reserve the bigger range and | |
8811 | * once this is done free the pages we are not interested in. | |
8812 | * | |
8813 | * We don't have to hold zone->lock here because the pages are | |
8814 | * isolated thus they won't get removed from buddy. | |
8815 | */ | |
8816 | ||
041d3a8c MN |
8817 | order = 0; |
8818 | outer_start = start; | |
8819 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
8820 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
8821 | outer_start = start; |
8822 | break; | |
041d3a8c MN |
8823 | } |
8824 | outer_start &= ~0UL << order; | |
8825 | } | |
8826 | ||
8ef5849f | 8827 | if (outer_start != start) { |
ab130f91 | 8828 | order = buddy_order(pfn_to_page(outer_start)); |
8ef5849f JK |
8829 | |
8830 | /* | |
8831 | * outer_start page could be small order buddy page and | |
8832 | * it doesn't include start page. Adjust outer_start | |
8833 | * in this case to report failed page properly | |
8834 | * on tracepoint in test_pages_isolated() | |
8835 | */ | |
8836 | if (outer_start + (1UL << order) <= start) | |
8837 | outer_start = start; | |
8838 | } | |
8839 | ||
041d3a8c | 8840 | /* Make sure the range is really isolated. */ |
756d25be | 8841 | if (test_pages_isolated(outer_start, end, 0)) { |
041d3a8c MN |
8842 | ret = -EBUSY; |
8843 | goto done; | |
8844 | } | |
8845 | ||
49f223a9 | 8846 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 8847 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
8848 | if (!outer_end) { |
8849 | ret = -EBUSY; | |
8850 | goto done; | |
8851 | } | |
8852 | ||
8853 | /* Free head and tail (if any) */ | |
8854 | if (start != outer_start) | |
8855 | free_contig_range(outer_start, start - outer_start); | |
8856 | if (end != outer_end) | |
8857 | free_contig_range(end, outer_end - end); | |
8858 | ||
8859 | done: | |
8860 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 8861 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
8862 | return ret; |
8863 | } | |
255f5985 | 8864 | EXPORT_SYMBOL(alloc_contig_range); |
5e27a2df AK |
8865 | |
8866 | static int __alloc_contig_pages(unsigned long start_pfn, | |
8867 | unsigned long nr_pages, gfp_t gfp_mask) | |
8868 | { | |
8869 | unsigned long end_pfn = start_pfn + nr_pages; | |
8870 | ||
8871 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
8872 | gfp_mask); | |
8873 | } | |
8874 | ||
8875 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
8876 | unsigned long nr_pages) | |
8877 | { | |
8878 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
8879 | struct page *page; | |
8880 | ||
8881 | for (i = start_pfn; i < end_pfn; i++) { | |
8882 | page = pfn_to_online_page(i); | |
8883 | if (!page) | |
8884 | return false; | |
8885 | ||
8886 | if (page_zone(page) != z) | |
8887 | return false; | |
8888 | ||
8889 | if (PageReserved(page)) | |
8890 | return false; | |
8891 | ||
8892 | if (page_count(page) > 0) | |
8893 | return false; | |
8894 | ||
8895 | if (PageHuge(page)) | |
8896 | return false; | |
8897 | } | |
8898 | return true; | |
8899 | } | |
8900 | ||
8901 | static bool zone_spans_last_pfn(const struct zone *zone, | |
8902 | unsigned long start_pfn, unsigned long nr_pages) | |
8903 | { | |
8904 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
8905 | ||
8906 | return zone_spans_pfn(zone, last_pfn); | |
8907 | } | |
8908 | ||
8909 | /** | |
8910 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
8911 | * @nr_pages: Number of contiguous pages to allocate | |
8912 | * @gfp_mask: GFP mask to limit search and used during compaction | |
8913 | * @nid: Target node | |
8914 | * @nodemask: Mask for other possible nodes | |
8915 | * | |
8916 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
8917 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
8918 | * tried for allocation with alloc_contig_range(). This routine is intended | |
8919 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
8920 | * | |
8921 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
8922 | * power of two then the alignment is guaranteed to be to the given nr_pages | |
8923 | * (e.g. 1GB request would be aligned to 1GB). | |
8924 | * | |
8925 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
8926 | * __free_page() on each allocated page. | |
8927 | * | |
8928 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
8929 | */ | |
8930 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
8931 | int nid, nodemask_t *nodemask) | |
8932 | { | |
8933 | unsigned long ret, pfn, flags; | |
8934 | struct zonelist *zonelist; | |
8935 | struct zone *zone; | |
8936 | struct zoneref *z; | |
8937 | ||
8938 | zonelist = node_zonelist(nid, gfp_mask); | |
8939 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
8940 | gfp_zone(gfp_mask), nodemask) { | |
8941 | spin_lock_irqsave(&zone->lock, flags); | |
8942 | ||
8943 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
8944 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
8945 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
8946 | /* | |
8947 | * We release the zone lock here because | |
8948 | * alloc_contig_range() will also lock the zone | |
8949 | * at some point. If there's an allocation | |
8950 | * spinning on this lock, it may win the race | |
8951 | * and cause alloc_contig_range() to fail... | |
8952 | */ | |
8953 | spin_unlock_irqrestore(&zone->lock, flags); | |
8954 | ret = __alloc_contig_pages(pfn, nr_pages, | |
8955 | gfp_mask); | |
8956 | if (!ret) | |
8957 | return pfn_to_page(pfn); | |
8958 | spin_lock_irqsave(&zone->lock, flags); | |
8959 | } | |
8960 | pfn += nr_pages; | |
8961 | } | |
8962 | spin_unlock_irqrestore(&zone->lock, flags); | |
8963 | } | |
8964 | return NULL; | |
8965 | } | |
4eb0716e | 8966 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 8967 | |
4eb0716e | 8968 | void free_contig_range(unsigned long pfn, unsigned int nr_pages) |
041d3a8c | 8969 | { |
bcc2b02f MS |
8970 | unsigned int count = 0; |
8971 | ||
8972 | for (; nr_pages--; pfn++) { | |
8973 | struct page *page = pfn_to_page(pfn); | |
8974 | ||
8975 | count += page_count(page) != 1; | |
8976 | __free_page(page); | |
8977 | } | |
8978 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c | 8979 | } |
255f5985 | 8980 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 8981 | |
0a647f38 CS |
8982 | /* |
8983 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
8984 | * page high values need to be recalulated. | |
8985 | */ | |
4ed7e022 JL |
8986 | void __meminit zone_pcp_update(struct zone *zone) |
8987 | { | |
c8e251fa | 8988 | mutex_lock(&pcp_batch_high_lock); |
0a8b4f1d | 8989 | zone_set_pageset_high_and_batch(zone); |
c8e251fa | 8990 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 | 8991 | } |
4ed7e022 | 8992 | |
ec6e8c7e VB |
8993 | /* |
8994 | * Effectively disable pcplists for the zone by setting the high limit to 0 | |
8995 | * and draining all cpus. A concurrent page freeing on another CPU that's about | |
8996 | * to put the page on pcplist will either finish before the drain and the page | |
8997 | * will be drained, or observe the new high limit and skip the pcplist. | |
8998 | * | |
8999 | * Must be paired with a call to zone_pcp_enable(). | |
9000 | */ | |
9001 | void zone_pcp_disable(struct zone *zone) | |
9002 | { | |
9003 | mutex_lock(&pcp_batch_high_lock); | |
9004 | __zone_set_pageset_high_and_batch(zone, 0, 1); | |
9005 | __drain_all_pages(zone, true); | |
9006 | } | |
9007 | ||
9008 | void zone_pcp_enable(struct zone *zone) | |
9009 | { | |
9010 | __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch); | |
9011 | mutex_unlock(&pcp_batch_high_lock); | |
9012 | } | |
9013 | ||
340175b7 JL |
9014 | void zone_pcp_reset(struct zone *zone) |
9015 | { | |
9016 | unsigned long flags; | |
5a883813 MK |
9017 | int cpu; |
9018 | struct per_cpu_pageset *pset; | |
340175b7 JL |
9019 | |
9020 | /* avoid races with drain_pages() */ | |
9021 | local_irq_save(flags); | |
9022 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
9023 | for_each_online_cpu(cpu) { |
9024 | pset = per_cpu_ptr(zone->pageset, cpu); | |
9025 | drain_zonestat(zone, pset); | |
9026 | } | |
340175b7 JL |
9027 | free_percpu(zone->pageset); |
9028 | zone->pageset = &boot_pageset; | |
9029 | } | |
9030 | local_irq_restore(flags); | |
9031 | } | |
9032 | ||
6dcd73d7 | 9033 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 9034 | /* |
257bea71 DH |
9035 | * All pages in the range must be in a single zone, must not contain holes, |
9036 | * must span full sections, and must be isolated before calling this function. | |
0c0e6195 | 9037 | */ |
257bea71 | 9038 | void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 9039 | { |
257bea71 | 9040 | unsigned long pfn = start_pfn; |
0c0e6195 KH |
9041 | struct page *page; |
9042 | struct zone *zone; | |
0ee5f4f3 | 9043 | unsigned int order; |
0c0e6195 | 9044 | unsigned long flags; |
5557c766 | 9045 | |
2d070eab | 9046 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
9047 | zone = page_zone(pfn_to_page(pfn)); |
9048 | spin_lock_irqsave(&zone->lock, flags); | |
0c0e6195 | 9049 | while (pfn < end_pfn) { |
0c0e6195 | 9050 | page = pfn_to_page(pfn); |
b023f468 WC |
9051 | /* |
9052 | * The HWPoisoned page may be not in buddy system, and | |
9053 | * page_count() is not 0. | |
9054 | */ | |
9055 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
9056 | pfn++; | |
b023f468 WC |
9057 | continue; |
9058 | } | |
aa218795 DH |
9059 | /* |
9060 | * At this point all remaining PageOffline() pages have a | |
9061 | * reference count of 0 and can simply be skipped. | |
9062 | */ | |
9063 | if (PageOffline(page)) { | |
9064 | BUG_ON(page_count(page)); | |
9065 | BUG_ON(PageBuddy(page)); | |
9066 | pfn++; | |
aa218795 DH |
9067 | continue; |
9068 | } | |
b023f468 | 9069 | |
0c0e6195 KH |
9070 | BUG_ON(page_count(page)); |
9071 | BUG_ON(!PageBuddy(page)); | |
ab130f91 | 9072 | order = buddy_order(page); |
6ab01363 | 9073 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
9074 | pfn += (1 << order); |
9075 | } | |
9076 | spin_unlock_irqrestore(&zone->lock, flags); | |
9077 | } | |
9078 | #endif | |
8d22ba1b | 9079 | |
8d22ba1b WF |
9080 | bool is_free_buddy_page(struct page *page) |
9081 | { | |
9082 | struct zone *zone = page_zone(page); | |
9083 | unsigned long pfn = page_to_pfn(page); | |
9084 | unsigned long flags; | |
7aeb09f9 | 9085 | unsigned int order; |
8d22ba1b WF |
9086 | |
9087 | spin_lock_irqsave(&zone->lock, flags); | |
9088 | for (order = 0; order < MAX_ORDER; order++) { | |
9089 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
9090 | ||
ab130f91 | 9091 | if (PageBuddy(page_head) && buddy_order(page_head) >= order) |
8d22ba1b WF |
9092 | break; |
9093 | } | |
9094 | spin_unlock_irqrestore(&zone->lock, flags); | |
9095 | ||
9096 | return order < MAX_ORDER; | |
9097 | } | |
d4ae9916 NH |
9098 | |
9099 | #ifdef CONFIG_MEMORY_FAILURE | |
9100 | /* | |
06be6ff3 OS |
9101 | * Break down a higher-order page in sub-pages, and keep our target out of |
9102 | * buddy allocator. | |
d4ae9916 | 9103 | */ |
06be6ff3 OS |
9104 | static void break_down_buddy_pages(struct zone *zone, struct page *page, |
9105 | struct page *target, int low, int high, | |
9106 | int migratetype) | |
9107 | { | |
9108 | unsigned long size = 1 << high; | |
9109 | struct page *current_buddy, *next_page; | |
9110 | ||
9111 | while (high > low) { | |
9112 | high--; | |
9113 | size >>= 1; | |
9114 | ||
9115 | if (target >= &page[size]) { | |
9116 | next_page = page + size; | |
9117 | current_buddy = page; | |
9118 | } else { | |
9119 | next_page = page; | |
9120 | current_buddy = page + size; | |
9121 | } | |
9122 | ||
9123 | if (set_page_guard(zone, current_buddy, high, migratetype)) | |
9124 | continue; | |
9125 | ||
9126 | if (current_buddy != target) { | |
9127 | add_to_free_list(current_buddy, zone, high, migratetype); | |
ab130f91 | 9128 | set_buddy_order(current_buddy, high); |
06be6ff3 OS |
9129 | page = next_page; |
9130 | } | |
9131 | } | |
9132 | } | |
9133 | ||
9134 | /* | |
9135 | * Take a page that will be marked as poisoned off the buddy allocator. | |
9136 | */ | |
9137 | bool take_page_off_buddy(struct page *page) | |
d4ae9916 NH |
9138 | { |
9139 | struct zone *zone = page_zone(page); | |
9140 | unsigned long pfn = page_to_pfn(page); | |
9141 | unsigned long flags; | |
9142 | unsigned int order; | |
06be6ff3 | 9143 | bool ret = false; |
d4ae9916 NH |
9144 | |
9145 | spin_lock_irqsave(&zone->lock, flags); | |
9146 | for (order = 0; order < MAX_ORDER; order++) { | |
9147 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
ab130f91 | 9148 | int page_order = buddy_order(page_head); |
d4ae9916 | 9149 | |
ab130f91 | 9150 | if (PageBuddy(page_head) && page_order >= order) { |
06be6ff3 OS |
9151 | unsigned long pfn_head = page_to_pfn(page_head); |
9152 | int migratetype = get_pfnblock_migratetype(page_head, | |
9153 | pfn_head); | |
9154 | ||
ab130f91 | 9155 | del_page_from_free_list(page_head, zone, page_order); |
06be6ff3 | 9156 | break_down_buddy_pages(zone, page_head, page, 0, |
ab130f91 | 9157 | page_order, migratetype); |
06be6ff3 | 9158 | ret = true; |
d4ae9916 NH |
9159 | break; |
9160 | } | |
06be6ff3 OS |
9161 | if (page_count(page_head) > 0) |
9162 | break; | |
d4ae9916 NH |
9163 | } |
9164 | spin_unlock_irqrestore(&zone->lock, flags); | |
06be6ff3 | 9165 | return ret; |
d4ae9916 NH |
9166 | } |
9167 | #endif |