]>
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 | 21 | #include <linux/swap.h> |
bf181c58 | 22 | #include <linux/swapops.h> |
1da177e4 LT |
23 | #include <linux/interrupt.h> |
24 | #include <linux/pagemap.h> | |
10ed273f | 25 | #include <linux/jiffies.h> |
edbe7d23 | 26 | #include <linux/memblock.h> |
1da177e4 | 27 | #include <linux/compiler.h> |
9f158333 | 28 | #include <linux/kernel.h> |
b8c73fc2 | 29 | #include <linux/kasan.h> |
b073d7f8 | 30 | #include <linux/kmsan.h> |
1da177e4 LT |
31 | #include <linux/module.h> |
32 | #include <linux/suspend.h> | |
33 | #include <linux/pagevec.h> | |
34 | #include <linux/blkdev.h> | |
35 | #include <linux/slab.h> | |
a238ab5b | 36 | #include <linux/ratelimit.h> |
5a3135c2 | 37 | #include <linux/oom.h> |
1da177e4 LT |
38 | #include <linux/topology.h> |
39 | #include <linux/sysctl.h> | |
40 | #include <linux/cpu.h> | |
41 | #include <linux/cpuset.h> | |
bdc8cb98 | 42 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
43 | #include <linux/nodemask.h> |
44 | #include <linux/vmalloc.h> | |
a6cccdc3 | 45 | #include <linux/vmstat.h> |
4be38e35 | 46 | #include <linux/mempolicy.h> |
4b94ffdc | 47 | #include <linux/memremap.h> |
6811378e | 48 | #include <linux/stop_machine.h> |
97500a4a | 49 | #include <linux/random.h> |
c713216d MG |
50 | #include <linux/sort.h> |
51 | #include <linux/pfn.h> | |
3fcfab16 | 52 | #include <linux/backing-dev.h> |
933e312e | 53 | #include <linux/fault-inject.h> |
a5d76b54 | 54 | #include <linux/page-isolation.h> |
3ac7fe5a | 55 | #include <linux/debugobjects.h> |
dbb1f81c | 56 | #include <linux/kmemleak.h> |
56de7263 | 57 | #include <linux/compaction.h> |
0d3d062a | 58 | #include <trace/events/kmem.h> |
d379f01d | 59 | #include <trace/events/oom.h> |
268bb0ce | 60 | #include <linux/prefetch.h> |
6e543d57 | 61 | #include <linux/mm_inline.h> |
f920e413 | 62 | #include <linux/mmu_notifier.h> |
041d3a8c | 63 | #include <linux/migrate.h> |
949f7ec5 | 64 | #include <linux/hugetlb.h> |
8bd75c77 | 65 | #include <linux/sched/rt.h> |
5b3cc15a | 66 | #include <linux/sched/mm.h> |
48c96a36 | 67 | #include <linux/page_owner.h> |
df4e817b | 68 | #include <linux/page_table_check.h> |
0e1cc95b | 69 | #include <linux/kthread.h> |
4949148a | 70 | #include <linux/memcontrol.h> |
42c269c8 | 71 | #include <linux/ftrace.h> |
d92a8cfc | 72 | #include <linux/lockdep.h> |
556b969a | 73 | #include <linux/nmi.h> |
eb414681 | 74 | #include <linux/psi.h> |
e4443149 | 75 | #include <linux/padata.h> |
4aab2be0 | 76 | #include <linux/khugepaged.h> |
ba8f3587 | 77 | #include <linux/buffer_head.h> |
5bf18281 | 78 | #include <linux/delayacct.h> |
7ee3d4e8 | 79 | #include <asm/sections.h> |
1da177e4 | 80 | #include <asm/tlbflush.h> |
ac924c60 | 81 | #include <asm/div64.h> |
1da177e4 | 82 | #include "internal.h" |
e900a918 | 83 | #include "shuffle.h" |
36e66c55 | 84 | #include "page_reporting.h" |
014bb1de | 85 | #include "swap.h" |
1da177e4 | 86 | |
f04a5d5d DH |
87 | /* Free Page Internal flags: for internal, non-pcp variants of free_pages(). */ |
88 | typedef int __bitwise fpi_t; | |
89 | ||
90 | /* No special request */ | |
91 | #define FPI_NONE ((__force fpi_t)0) | |
92 | ||
93 | /* | |
94 | * Skip free page reporting notification for the (possibly merged) page. | |
95 | * This does not hinder free page reporting from grabbing the page, | |
96 | * reporting it and marking it "reported" - it only skips notifying | |
97 | * the free page reporting infrastructure about a newly freed page. For | |
98 | * example, used when temporarily pulling a page from a freelist and | |
99 | * putting it back unmodified. | |
100 | */ | |
101 | #define FPI_SKIP_REPORT_NOTIFY ((__force fpi_t)BIT(0)) | |
102 | ||
47b6a24a DH |
103 | /* |
104 | * Place the (possibly merged) page to the tail of the freelist. Will ignore | |
105 | * page shuffling (relevant code - e.g., memory onlining - is expected to | |
106 | * shuffle the whole zone). | |
107 | * | |
108 | * Note: No code should rely on this flag for correctness - it's purely | |
109 | * to allow for optimizations when handing back either fresh pages | |
110 | * (memory onlining) or untouched pages (page isolation, free page | |
111 | * reporting). | |
112 | */ | |
113 | #define FPI_TO_TAIL ((__force fpi_t)BIT(1)) | |
114 | ||
2c335680 AK |
115 | /* |
116 | * Don't poison memory with KASAN (only for the tag-based modes). | |
117 | * During boot, all non-reserved memblock memory is exposed to page_alloc. | |
118 | * Poisoning all that memory lengthens boot time, especially on systems with | |
119 | * large amount of RAM. This flag is used to skip that poisoning. | |
120 | * This is only done for the tag-based KASAN modes, as those are able to | |
121 | * detect memory corruptions with the memory tags assigned by default. | |
122 | * All memory allocated normally after boot gets poisoned as usual. | |
123 | */ | |
124 | #define FPI_SKIP_KASAN_POISON ((__force fpi_t)BIT(2)) | |
125 | ||
c8e251fa CS |
126 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
127 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
74f44822 | 128 | #define MIN_PERCPU_PAGELIST_HIGH_FRACTION (8) |
c8e251fa | 129 | |
4b23a68f MG |
130 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT) |
131 | /* | |
132 | * On SMP, spin_trylock is sufficient protection. | |
133 | * On PREEMPT_RT, spin_trylock is equivalent on both SMP and UP. | |
134 | */ | |
135 | #define pcp_trylock_prepare(flags) do { } while (0) | |
136 | #define pcp_trylock_finish(flag) do { } while (0) | |
137 | #else | |
138 | ||
139 | /* UP spin_trylock always succeeds so disable IRQs to prevent re-entrancy. */ | |
140 | #define pcp_trylock_prepare(flags) local_irq_save(flags) | |
141 | #define pcp_trylock_finish(flags) local_irq_restore(flags) | |
142 | #endif | |
143 | ||
01b44456 MG |
144 | /* |
145 | * Locking a pcp requires a PCP lookup followed by a spinlock. To avoid | |
146 | * a migration causing the wrong PCP to be locked and remote memory being | |
147 | * potentially allocated, pin the task to the CPU for the lookup+lock. | |
148 | * preempt_disable is used on !RT because it is faster than migrate_disable. | |
149 | * migrate_disable is used on RT because otherwise RT spinlock usage is | |
150 | * interfered with and a high priority task cannot preempt the allocator. | |
151 | */ | |
152 | #ifndef CONFIG_PREEMPT_RT | |
153 | #define pcpu_task_pin() preempt_disable() | |
154 | #define pcpu_task_unpin() preempt_enable() | |
155 | #else | |
156 | #define pcpu_task_pin() migrate_disable() | |
157 | #define pcpu_task_unpin() migrate_enable() | |
158 | #endif | |
c8e251fa | 159 | |
01b44456 MG |
160 | /* |
161 | * Generic helper to lookup and a per-cpu variable with an embedded spinlock. | |
162 | * Return value should be used with equivalent unlock helper. | |
163 | */ | |
164 | #define pcpu_spin_lock(type, member, ptr) \ | |
165 | ({ \ | |
166 | type *_ret; \ | |
167 | pcpu_task_pin(); \ | |
168 | _ret = this_cpu_ptr(ptr); \ | |
169 | spin_lock(&_ret->member); \ | |
170 | _ret; \ | |
171 | }) | |
172 | ||
57490774 | 173 | #define pcpu_spin_trylock(type, member, ptr) \ |
01b44456 MG |
174 | ({ \ |
175 | type *_ret; \ | |
176 | pcpu_task_pin(); \ | |
177 | _ret = this_cpu_ptr(ptr); \ | |
57490774 | 178 | if (!spin_trylock(&_ret->member)) { \ |
01b44456 MG |
179 | pcpu_task_unpin(); \ |
180 | _ret = NULL; \ | |
181 | } \ | |
182 | _ret; \ | |
183 | }) | |
184 | ||
185 | #define pcpu_spin_unlock(member, ptr) \ | |
186 | ({ \ | |
187 | spin_unlock(&ptr->member); \ | |
188 | pcpu_task_unpin(); \ | |
189 | }) | |
190 | ||
01b44456 MG |
191 | /* struct per_cpu_pages specific helpers. */ |
192 | #define pcp_spin_lock(ptr) \ | |
193 | pcpu_spin_lock(struct per_cpu_pages, lock, ptr) | |
194 | ||
57490774 MG |
195 | #define pcp_spin_trylock(ptr) \ |
196 | pcpu_spin_trylock(struct per_cpu_pages, lock, ptr) | |
01b44456 MG |
197 | |
198 | #define pcp_spin_unlock(ptr) \ | |
199 | pcpu_spin_unlock(lock, ptr) | |
200 | ||
72812019 LS |
201 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
202 | DEFINE_PER_CPU(int, numa_node); | |
203 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
204 | #endif | |
205 | ||
4518085e KW |
206 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
207 | ||
7aac7898 LS |
208 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
209 | /* | |
210 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
211 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
212 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
213 | * defined in <linux/topology.h>. | |
214 | */ | |
215 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
216 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
217 | #endif | |
218 | ||
8b885f53 | 219 | static DEFINE_MUTEX(pcpu_drain_mutex); |
bd233f53 | 220 | |
38addce8 | 221 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 222 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
223 | EXPORT_SYMBOL(latent_entropy); |
224 | #endif | |
225 | ||
1da177e4 | 226 | /* |
13808910 | 227 | * Array of node states. |
1da177e4 | 228 | */ |
13808910 CL |
229 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
230 | [N_POSSIBLE] = NODE_MASK_ALL, | |
231 | [N_ONLINE] = { { [0] = 1UL } }, | |
232 | #ifndef CONFIG_NUMA | |
233 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
234 | #ifdef CONFIG_HIGHMEM | |
235 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 236 | #endif |
20b2f52b | 237 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
238 | [N_CPU] = { { [0] = 1UL } }, |
239 | #endif /* NUMA */ | |
240 | }; | |
241 | EXPORT_SYMBOL(node_states); | |
242 | ||
ca79b0c2 AK |
243 | atomic_long_t _totalram_pages __read_mostly; |
244 | EXPORT_SYMBOL(_totalram_pages); | |
cb45b0e9 | 245 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 246 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 247 | |
74f44822 | 248 | int percpu_pagelist_high_fraction; |
dcce284a | 249 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
51cba1eb | 250 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc); |
6471384a AP |
251 | EXPORT_SYMBOL(init_on_alloc); |
252 | ||
51cba1eb | 253 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free); |
6471384a AP |
254 | EXPORT_SYMBOL(init_on_free); |
255 | ||
04013513 VB |
256 | static bool _init_on_alloc_enabled_early __read_mostly |
257 | = IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON); | |
6471384a AP |
258 | static int __init early_init_on_alloc(char *buf) |
259 | { | |
6471384a | 260 | |
04013513 | 261 | return kstrtobool(buf, &_init_on_alloc_enabled_early); |
6471384a AP |
262 | } |
263 | early_param("init_on_alloc", early_init_on_alloc); | |
264 | ||
04013513 VB |
265 | static bool _init_on_free_enabled_early __read_mostly |
266 | = IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON); | |
6471384a AP |
267 | static int __init early_init_on_free(char *buf) |
268 | { | |
04013513 | 269 | return kstrtobool(buf, &_init_on_free_enabled_early); |
6471384a AP |
270 | } |
271 | early_param("init_on_free", early_init_on_free); | |
1da177e4 | 272 | |
bb14c2c7 VB |
273 | /* |
274 | * A cached value of the page's pageblock's migratetype, used when the page is | |
275 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
276 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
277 | * Also the migratetype set in the page does not necessarily match the pcplist | |
278 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
279 | * other index - this ensures that it will be put on the correct CMA freelist. | |
280 | */ | |
281 | static inline int get_pcppage_migratetype(struct page *page) | |
282 | { | |
283 | return page->index; | |
284 | } | |
285 | ||
286 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
287 | { | |
288 | page->index = migratetype; | |
289 | } | |
290 | ||
452aa699 RW |
291 | #ifdef CONFIG_PM_SLEEP |
292 | /* | |
293 | * The following functions are used by the suspend/hibernate code to temporarily | |
294 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
295 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
296 | * they should always be called with system_transition_mutex held |
297 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
298 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
299 | * with that modification). | |
452aa699 | 300 | */ |
c9e664f1 RW |
301 | |
302 | static gfp_t saved_gfp_mask; | |
303 | ||
304 | void pm_restore_gfp_mask(void) | |
452aa699 | 305 | { |
55f2503c | 306 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
307 | if (saved_gfp_mask) { |
308 | gfp_allowed_mask = saved_gfp_mask; | |
309 | saved_gfp_mask = 0; | |
310 | } | |
452aa699 RW |
311 | } |
312 | ||
c9e664f1 | 313 | void pm_restrict_gfp_mask(void) |
452aa699 | 314 | { |
55f2503c | 315 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
316 | WARN_ON(saved_gfp_mask); |
317 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 318 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 319 | } |
f90ac398 MG |
320 | |
321 | bool pm_suspended_storage(void) | |
322 | { | |
d0164adc | 323 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
324 | return false; |
325 | return true; | |
326 | } | |
452aa699 RW |
327 | #endif /* CONFIG_PM_SLEEP */ |
328 | ||
d9c23400 | 329 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 330 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
331 | #endif |
332 | ||
7fef431b DH |
333 | static void __free_pages_ok(struct page *page, unsigned int order, |
334 | fpi_t fpi_flags); | |
a226f6c8 | 335 | |
1da177e4 LT |
336 | /* |
337 | * results with 256, 32 in the lowmem_reserve sysctl: | |
338 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
339 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
340 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
341 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 342 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
343 | * |
344 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
345 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 346 | */ |
d3cda233 | 347 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 348 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 349 | [ZONE_DMA] = 256, |
4b51d669 | 350 | #endif |
fb0e7942 | 351 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 352 | [ZONE_DMA32] = 256, |
fb0e7942 | 353 | #endif |
d3cda233 | 354 | [ZONE_NORMAL] = 32, |
e53ef38d | 355 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 356 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 357 | #endif |
d3cda233 | 358 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 359 | }; |
1da177e4 | 360 | |
15ad7cdc | 361 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 362 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 363 | "DMA", |
4b51d669 | 364 | #endif |
fb0e7942 | 365 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 366 | "DMA32", |
fb0e7942 | 367 | #endif |
2f1b6248 | 368 | "Normal", |
e53ef38d | 369 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 370 | "HighMem", |
e53ef38d | 371 | #endif |
2a1e274a | 372 | "Movable", |
033fbae9 DW |
373 | #ifdef CONFIG_ZONE_DEVICE |
374 | "Device", | |
375 | #endif | |
2f1b6248 CL |
376 | }; |
377 | ||
c999fbd3 | 378 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
379 | "Unmovable", |
380 | "Movable", | |
381 | "Reclaimable", | |
382 | "HighAtomic", | |
383 | #ifdef CONFIG_CMA | |
384 | "CMA", | |
385 | #endif | |
386 | #ifdef CONFIG_MEMORY_ISOLATION | |
387 | "Isolate", | |
388 | #endif | |
389 | }; | |
390 | ||
ae70eddd AK |
391 | compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = { |
392 | [NULL_COMPOUND_DTOR] = NULL, | |
393 | [COMPOUND_PAGE_DTOR] = free_compound_page, | |
f1e61557 | 394 | #ifdef CONFIG_HUGETLB_PAGE |
ae70eddd | 395 | [HUGETLB_PAGE_DTOR] = free_huge_page, |
f1e61557 | 396 | #endif |
9a982250 | 397 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
ae70eddd | 398 | [TRANSHUGE_PAGE_DTOR] = free_transhuge_page, |
9a982250 | 399 | #endif |
f1e61557 KS |
400 | }; |
401 | ||
1da177e4 | 402 | int min_free_kbytes = 1024; |
42aa83cb | 403 | int user_min_free_kbytes = -1; |
1c30844d | 404 | int watermark_boost_factor __read_mostly = 15000; |
795ae7a0 | 405 | int watermark_scale_factor = 10; |
1da177e4 | 406 | |
bbe5d993 OS |
407 | static unsigned long nr_kernel_pages __initdata; |
408 | static unsigned long nr_all_pages __initdata; | |
409 | static unsigned long dma_reserve __initdata; | |
1da177e4 | 410 | |
bbe5d993 OS |
411 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata; |
412 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata; | |
7f16f91f | 413 | static unsigned long required_kernelcore __initdata; |
a5c6d650 | 414 | static unsigned long required_kernelcore_percent __initdata; |
7f16f91f | 415 | static unsigned long required_movablecore __initdata; |
a5c6d650 | 416 | static unsigned long required_movablecore_percent __initdata; |
bbe5d993 | 417 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata; |
902c2d91 | 418 | bool mirrored_kernelcore __initdata_memblock; |
0ee332c1 TH |
419 | |
420 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
421 | int movable_zone; | |
422 | EXPORT_SYMBOL(movable_zone); | |
c713216d | 423 | |
418508c1 | 424 | #if MAX_NUMNODES > 1 |
b9726c26 | 425 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 426 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 427 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 428 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
429 | #endif |
430 | ||
9ef9acb0 MG |
431 | int page_group_by_mobility_disabled __read_mostly; |
432 | ||
3a80a7fa | 433 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
434 | /* |
435 | * During boot we initialize deferred pages on-demand, as needed, but once | |
436 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
437 | * and we can permanently disable that path. | |
438 | */ | |
439 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
440 | ||
94ae8b83 | 441 | static inline bool deferred_pages_enabled(void) |
3c0c12cc | 442 | { |
94ae8b83 | 443 | return static_branch_unlikely(&deferred_pages); |
3c0c12cc WL |
444 | } |
445 | ||
fc574488 MRI |
446 | /* Returns true if the struct page for the pfn is initialised */ |
447 | static inline bool __meminit early_page_initialised(unsigned long pfn) | |
3a80a7fa | 448 | { |
ef70b6f4 MG |
449 | int nid = early_pfn_to_nid(pfn); |
450 | ||
451 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
fc574488 | 452 | return false; |
3a80a7fa | 453 | |
fc574488 | 454 | return true; |
3a80a7fa MG |
455 | } |
456 | ||
457 | /* | |
d3035be4 | 458 | * Returns true when the remaining initialisation should be deferred until |
3a80a7fa MG |
459 | * later in the boot cycle when it can be parallelised. |
460 | */ | |
d3035be4 PT |
461 | static bool __meminit |
462 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn) | |
3a80a7fa | 463 | { |
d3035be4 PT |
464 | static unsigned long prev_end_pfn, nr_initialised; |
465 | ||
c4f20f14 LZ |
466 | if (early_page_ext_enabled()) |
467 | return false; | |
d3035be4 PT |
468 | /* |
469 | * prev_end_pfn static that contains the end of previous zone | |
470 | * No need to protect because called very early in boot before smp_init. | |
471 | */ | |
472 | if (prev_end_pfn != end_pfn) { | |
473 | prev_end_pfn = end_pfn; | |
474 | nr_initialised = 0; | |
475 | } | |
476 | ||
3c2c6488 | 477 | /* Always populate low zones for address-constrained allocations */ |
d3035be4 | 478 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid))) |
3a80a7fa | 479 | return false; |
23b68cfa | 480 | |
dc2da7b4 BH |
481 | if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX) |
482 | return true; | |
23b68cfa WY |
483 | /* |
484 | * We start only with one section of pages, more pages are added as | |
485 | * needed until the rest of deferred pages are initialized. | |
486 | */ | |
d3035be4 | 487 | nr_initialised++; |
23b68cfa | 488 | if ((nr_initialised > PAGES_PER_SECTION) && |
d3035be4 PT |
489 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
490 | NODE_DATA(nid)->first_deferred_pfn = pfn; | |
491 | return true; | |
3a80a7fa | 492 | } |
d3035be4 | 493 | return false; |
3a80a7fa MG |
494 | } |
495 | #else | |
94ae8b83 | 496 | static inline bool deferred_pages_enabled(void) |
2c335680 | 497 | { |
94ae8b83 | 498 | return false; |
2c335680 | 499 | } |
3c0c12cc | 500 | |
fc574488 | 501 | static inline bool early_page_initialised(unsigned long pfn) |
3a80a7fa | 502 | { |
fc574488 | 503 | return true; |
3a80a7fa MG |
504 | } |
505 | ||
d3035be4 | 506 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn) |
3a80a7fa | 507 | { |
d3035be4 | 508 | return false; |
3a80a7fa MG |
509 | } |
510 | #endif | |
511 | ||
0b423ca2 | 512 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
ca891f41 | 513 | static inline unsigned long *get_pageblock_bitmap(const struct page *page, |
0b423ca2 MG |
514 | unsigned long pfn) |
515 | { | |
516 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 517 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
518 | #else |
519 | return page_zone(page)->pageblock_flags; | |
520 | #endif /* CONFIG_SPARSEMEM */ | |
521 | } | |
522 | ||
ca891f41 | 523 | static inline int pfn_to_bitidx(const struct page *page, unsigned long pfn) |
0b423ca2 MG |
524 | { |
525 | #ifdef CONFIG_SPARSEMEM | |
526 | pfn &= (PAGES_PER_SECTION-1); | |
0b423ca2 | 527 | #else |
4f9bc69a | 528 | pfn = pfn - pageblock_start_pfn(page_zone(page)->zone_start_pfn); |
0b423ca2 | 529 | #endif /* CONFIG_SPARSEMEM */ |
399b795b | 530 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
0b423ca2 MG |
531 | } |
532 | ||
535b81e2 | 533 | static __always_inline |
ca891f41 | 534 | unsigned long __get_pfnblock_flags_mask(const struct page *page, |
0b423ca2 | 535 | unsigned long pfn, |
0b423ca2 MG |
536 | unsigned long mask) |
537 | { | |
538 | unsigned long *bitmap; | |
539 | unsigned long bitidx, word_bitidx; | |
540 | unsigned long word; | |
541 | ||
542 | bitmap = get_pageblock_bitmap(page, pfn); | |
543 | bitidx = pfn_to_bitidx(page, pfn); | |
544 | word_bitidx = bitidx / BITS_PER_LONG; | |
545 | bitidx &= (BITS_PER_LONG-1); | |
1c563432 MK |
546 | /* |
547 | * This races, without locks, with set_pfnblock_flags_mask(). Ensure | |
548 | * a consistent read of the memory array, so that results, even though | |
549 | * racy, are not corrupted. | |
550 | */ | |
551 | word = READ_ONCE(bitmap[word_bitidx]); | |
d93d5ab9 | 552 | return (word >> bitidx) & mask; |
0b423ca2 MG |
553 | } |
554 | ||
a00cda3f MCC |
555 | /** |
556 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
557 | * @page: The page within the block of interest | |
558 | * @pfn: The target page frame number | |
559 | * @mask: mask of bits that the caller is interested in | |
560 | * | |
561 | * Return: pageblock_bits flags | |
562 | */ | |
ca891f41 MWO |
563 | unsigned long get_pfnblock_flags_mask(const struct page *page, |
564 | unsigned long pfn, unsigned long mask) | |
0b423ca2 | 565 | { |
535b81e2 | 566 | return __get_pfnblock_flags_mask(page, pfn, mask); |
0b423ca2 MG |
567 | } |
568 | ||
ca891f41 MWO |
569 | static __always_inline int get_pfnblock_migratetype(const struct page *page, |
570 | unsigned long pfn) | |
0b423ca2 | 571 | { |
535b81e2 | 572 | return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK); |
0b423ca2 MG |
573 | } |
574 | ||
575 | /** | |
576 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
577 | * @page: The page within the block of interest | |
578 | * @flags: The flags to set | |
579 | * @pfn: The target page frame number | |
0b423ca2 MG |
580 | * @mask: mask of bits that the caller is interested in |
581 | */ | |
582 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
583 | unsigned long pfn, | |
0b423ca2 MG |
584 | unsigned long mask) |
585 | { | |
586 | unsigned long *bitmap; | |
587 | unsigned long bitidx, word_bitidx; | |
04ec0061 | 588 | unsigned long word; |
0b423ca2 MG |
589 | |
590 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 591 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
592 | |
593 | bitmap = get_pageblock_bitmap(page, pfn); | |
594 | bitidx = pfn_to_bitidx(page, pfn); | |
595 | word_bitidx = bitidx / BITS_PER_LONG; | |
596 | bitidx &= (BITS_PER_LONG-1); | |
597 | ||
598 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
599 | ||
d93d5ab9 WY |
600 | mask <<= bitidx; |
601 | flags <<= bitidx; | |
0b423ca2 MG |
602 | |
603 | word = READ_ONCE(bitmap[word_bitidx]); | |
04ec0061 UB |
604 | do { |
605 | } while (!try_cmpxchg(&bitmap[word_bitidx], &word, (word & ~mask) | flags)); | |
0b423ca2 | 606 | } |
3a80a7fa | 607 | |
ee6f509c | 608 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 609 | { |
5d0f3f72 KM |
610 | if (unlikely(page_group_by_mobility_disabled && |
611 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
612 | migratetype = MIGRATE_UNMOVABLE; |
613 | ||
d93d5ab9 | 614 | set_pfnblock_flags_mask(page, (unsigned long)migratetype, |
535b81e2 | 615 | page_to_pfn(page), MIGRATETYPE_MASK); |
b2a0ac88 MG |
616 | } |
617 | ||
13e7444b | 618 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 619 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 620 | { |
bdc8cb98 DH |
621 | int ret = 0; |
622 | unsigned seq; | |
623 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 624 | unsigned long sp, start_pfn; |
c6a57e19 | 625 | |
bdc8cb98 DH |
626 | do { |
627 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
628 | start_pfn = zone->zone_start_pfn; |
629 | sp = zone->spanned_pages; | |
108bcc96 | 630 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
631 | ret = 1; |
632 | } while (zone_span_seqretry(zone, seq)); | |
633 | ||
b5e6a5a2 | 634 | if (ret) |
613813e8 DH |
635 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
636 | pfn, zone_to_nid(zone), zone->name, | |
637 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 638 | |
bdc8cb98 | 639 | return ret; |
c6a57e19 DH |
640 | } |
641 | ||
642 | static int page_is_consistent(struct zone *zone, struct page *page) | |
643 | { | |
1da177e4 | 644 | if (zone != page_zone(page)) |
c6a57e19 DH |
645 | return 0; |
646 | ||
647 | return 1; | |
648 | } | |
649 | /* | |
650 | * Temporary debugging check for pages not lying within a given zone. | |
651 | */ | |
d73d3c9f | 652 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
653 | { |
654 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 655 | return 1; |
c6a57e19 DH |
656 | if (!page_is_consistent(zone, page)) |
657 | return 1; | |
658 | ||
1da177e4 LT |
659 | return 0; |
660 | } | |
13e7444b | 661 | #else |
d73d3c9f | 662 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
663 | { |
664 | return 0; | |
665 | } | |
666 | #endif | |
667 | ||
82a3241a | 668 | static void bad_page(struct page *page, const char *reason) |
1da177e4 | 669 | { |
d936cf9b HD |
670 | static unsigned long resume; |
671 | static unsigned long nr_shown; | |
672 | static unsigned long nr_unshown; | |
673 | ||
674 | /* | |
675 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
676 | * or allow a steady drip of one report per second. | |
677 | */ | |
678 | if (nr_shown == 60) { | |
679 | if (time_before(jiffies, resume)) { | |
680 | nr_unshown++; | |
681 | goto out; | |
682 | } | |
683 | if (nr_unshown) { | |
ff8e8116 | 684 | pr_alert( |
1e9e6365 | 685 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
686 | nr_unshown); |
687 | nr_unshown = 0; | |
688 | } | |
689 | nr_shown = 0; | |
690 | } | |
691 | if (nr_shown++ == 0) | |
692 | resume = jiffies + 60 * HZ; | |
693 | ||
ff8e8116 | 694 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 695 | current->comm, page_to_pfn(page)); |
d2f07ec0 | 696 | dump_page(page, reason); |
3dc14741 | 697 | |
4f31888c | 698 | print_modules(); |
1da177e4 | 699 | dump_stack(); |
d936cf9b | 700 | out: |
8cc3b392 | 701 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 702 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 703 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
704 | } |
705 | ||
44042b44 MG |
706 | static inline unsigned int order_to_pindex(int migratetype, int order) |
707 | { | |
708 | int base = order; | |
709 | ||
710 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
711 | if (order > PAGE_ALLOC_COSTLY_ORDER) { | |
712 | VM_BUG_ON(order != pageblock_order); | |
5d0a661d | 713 | return NR_LOWORDER_PCP_LISTS; |
44042b44 MG |
714 | } |
715 | #else | |
716 | VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER); | |
717 | #endif | |
718 | ||
719 | return (MIGRATE_PCPTYPES * base) + migratetype; | |
720 | } | |
721 | ||
722 | static inline int pindex_to_order(unsigned int pindex) | |
723 | { | |
724 | int order = pindex / MIGRATE_PCPTYPES; | |
725 | ||
726 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5d0a661d | 727 | if (pindex == NR_LOWORDER_PCP_LISTS) |
44042b44 | 728 | order = pageblock_order; |
44042b44 MG |
729 | #else |
730 | VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER); | |
731 | #endif | |
732 | ||
733 | return order; | |
734 | } | |
735 | ||
736 | static inline bool pcp_allowed_order(unsigned int order) | |
737 | { | |
738 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
739 | return true; | |
740 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
741 | if (order == pageblock_order) | |
742 | return true; | |
743 | #endif | |
744 | return false; | |
745 | } | |
746 | ||
21d02f8f MG |
747 | static inline void free_the_page(struct page *page, unsigned int order) |
748 | { | |
44042b44 MG |
749 | if (pcp_allowed_order(order)) /* Via pcp? */ |
750 | free_unref_page(page, order); | |
21d02f8f MG |
751 | else |
752 | __free_pages_ok(page, order, FPI_NONE); | |
753 | } | |
754 | ||
1da177e4 LT |
755 | /* |
756 | * Higher-order pages are called "compound pages". They are structured thusly: | |
757 | * | |
1d798ca3 | 758 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 759 | * |
1d798ca3 KS |
760 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
761 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 762 | * |
1d798ca3 KS |
763 | * The first tail page's ->compound_dtor holds the offset in array of compound |
764 | * page destructors. See compound_page_dtors. | |
1da177e4 | 765 | * |
1d798ca3 | 766 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 767 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 768 | */ |
d98c7a09 | 769 | |
9a982250 | 770 | void free_compound_page(struct page *page) |
d98c7a09 | 771 | { |
bbc6b703 | 772 | mem_cgroup_uncharge(page_folio(page)); |
44042b44 | 773 | free_the_page(page, compound_order(page)); |
d98c7a09 HD |
774 | } |
775 | ||
5b24eeef JM |
776 | static void prep_compound_head(struct page *page, unsigned int order) |
777 | { | |
94688e8e MWO |
778 | struct folio *folio = (struct folio *)page; |
779 | ||
5b24eeef JM |
780 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
781 | set_compound_order(page, order); | |
65a689f3 MWO |
782 | atomic_set(&folio->_entire_mapcount, -1); |
783 | atomic_set(&folio->_nr_pages_mapped, 0); | |
94688e8e | 784 | atomic_set(&folio->_pincount, 0); |
5b24eeef JM |
785 | } |
786 | ||
787 | static void prep_compound_tail(struct page *head, int tail_idx) | |
788 | { | |
789 | struct page *p = head + tail_idx; | |
790 | ||
791 | p->mapping = TAIL_MAPPING; | |
792 | set_compound_head(p, head); | |
5aae9265 | 793 | set_page_private(p, 0); |
5b24eeef JM |
794 | } |
795 | ||
d00181b9 | 796 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
797 | { |
798 | int i; | |
799 | int nr_pages = 1 << order; | |
800 | ||
18229df5 | 801 | __SetPageHead(page); |
5b24eeef JM |
802 | for (i = 1; i < nr_pages; i++) |
803 | prep_compound_tail(page, i); | |
1378a5ee | 804 | |
5b24eeef | 805 | prep_compound_head(page, order); |
18229df5 AW |
806 | } |
807 | ||
5375336c MWO |
808 | void destroy_large_folio(struct folio *folio) |
809 | { | |
a60d5942 | 810 | enum compound_dtor_id dtor = folio->_folio_dtor; |
5375336c MWO |
811 | |
812 | VM_BUG_ON_FOLIO(dtor >= NR_COMPOUND_DTORS, folio); | |
813 | compound_page_dtors[dtor](&folio->page); | |
814 | } | |
815 | ||
c0a32fc5 SG |
816 | #ifdef CONFIG_DEBUG_PAGEALLOC |
817 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 818 | |
8e57f8ac VB |
819 | bool _debug_pagealloc_enabled_early __read_mostly |
820 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
821 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 822 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 823 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
824 | |
825 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 826 | |
031bc574 JK |
827 | static int __init early_debug_pagealloc(char *buf) |
828 | { | |
8e57f8ac | 829 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
830 | } |
831 | early_param("debug_pagealloc", early_debug_pagealloc); | |
832 | ||
c0a32fc5 SG |
833 | static int __init debug_guardpage_minorder_setup(char *buf) |
834 | { | |
835 | unsigned long res; | |
836 | ||
837 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 838 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
839 | return 0; |
840 | } | |
841 | _debug_guardpage_minorder = res; | |
1170532b | 842 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
843 | return 0; |
844 | } | |
f1c1e9f7 | 845 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 846 | |
acbc15a4 | 847 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 848 | unsigned int order, int migratetype) |
c0a32fc5 | 849 | { |
e30825f1 | 850 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
851 | return false; |
852 | ||
853 | if (order >= debug_guardpage_minorder()) | |
854 | return false; | |
e30825f1 | 855 | |
3972f6bb | 856 | __SetPageGuard(page); |
bf75f200 | 857 | INIT_LIST_HEAD(&page->buddy_list); |
2847cf95 JK |
858 | set_page_private(page, order); |
859 | /* Guard pages are not available for any usage */ | |
b3618455 ML |
860 | if (!is_migrate_isolate(migratetype)) |
861 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
862 | |
863 | return true; | |
c0a32fc5 SG |
864 | } |
865 | ||
2847cf95 JK |
866 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
867 | unsigned int order, int migratetype) | |
c0a32fc5 | 868 | { |
e30825f1 JK |
869 | if (!debug_guardpage_enabled()) |
870 | return; | |
871 | ||
3972f6bb | 872 | __ClearPageGuard(page); |
e30825f1 | 873 | |
2847cf95 JK |
874 | set_page_private(page, 0); |
875 | if (!is_migrate_isolate(migratetype)) | |
876 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
877 | } |
878 | #else | |
acbc15a4 JK |
879 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
880 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
881 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
882 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
883 | #endif |
884 | ||
04013513 VB |
885 | /* |
886 | * Enable static keys related to various memory debugging and hardening options. | |
887 | * Some override others, and depend on early params that are evaluated in the | |
888 | * order of appearance. So we need to first gather the full picture of what was | |
889 | * enabled, and then make decisions. | |
890 | */ | |
5749fcc5 | 891 | void __init init_mem_debugging_and_hardening(void) |
04013513 | 892 | { |
9df65f52 ST |
893 | bool page_poisoning_requested = false; |
894 | ||
895 | #ifdef CONFIG_PAGE_POISONING | |
896 | /* | |
897 | * Page poisoning is debug page alloc for some arches. If | |
898 | * either of those options are enabled, enable poisoning. | |
899 | */ | |
900 | if (page_poisoning_enabled() || | |
901 | (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) && | |
902 | debug_pagealloc_enabled())) { | |
903 | static_branch_enable(&_page_poisoning_enabled); | |
904 | page_poisoning_requested = true; | |
905 | } | |
906 | #endif | |
907 | ||
69e5d322 ST |
908 | if ((_init_on_alloc_enabled_early || _init_on_free_enabled_early) && |
909 | page_poisoning_requested) { | |
910 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, " | |
911 | "will take precedence over init_on_alloc and init_on_free\n"); | |
912 | _init_on_alloc_enabled_early = false; | |
913 | _init_on_free_enabled_early = false; | |
04013513 VB |
914 | } |
915 | ||
69e5d322 ST |
916 | if (_init_on_alloc_enabled_early) |
917 | static_branch_enable(&init_on_alloc); | |
918 | else | |
919 | static_branch_disable(&init_on_alloc); | |
920 | ||
921 | if (_init_on_free_enabled_early) | |
922 | static_branch_enable(&init_on_free); | |
923 | else | |
924 | static_branch_disable(&init_on_free); | |
925 | ||
42eaa27d AP |
926 | if (IS_ENABLED(CONFIG_KMSAN) && |
927 | (_init_on_alloc_enabled_early || _init_on_free_enabled_early)) | |
928 | pr_info("mem auto-init: please make sure init_on_alloc and init_on_free are disabled when running KMSAN\n"); | |
929 | ||
04013513 VB |
930 | #ifdef CONFIG_DEBUG_PAGEALLOC |
931 | if (!debug_pagealloc_enabled()) | |
932 | return; | |
933 | ||
934 | static_branch_enable(&_debug_pagealloc_enabled); | |
935 | ||
936 | if (!debug_guardpage_minorder()) | |
937 | return; | |
938 | ||
939 | static_branch_enable(&_debug_guardpage_enabled); | |
940 | #endif | |
941 | } | |
942 | ||
ab130f91 | 943 | static inline void set_buddy_order(struct page *page, unsigned int order) |
6aa3001b | 944 | { |
4c21e2f2 | 945 | set_page_private(page, order); |
676165a8 | 946 | __SetPageBuddy(page); |
1da177e4 LT |
947 | } |
948 | ||
5e1f0f09 MG |
949 | #ifdef CONFIG_COMPACTION |
950 | static inline struct capture_control *task_capc(struct zone *zone) | |
951 | { | |
952 | struct capture_control *capc = current->capture_control; | |
953 | ||
deba0487 | 954 | return unlikely(capc) && |
5e1f0f09 MG |
955 | !(current->flags & PF_KTHREAD) && |
956 | !capc->page && | |
deba0487 | 957 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
958 | } |
959 | ||
960 | static inline bool | |
961 | compaction_capture(struct capture_control *capc, struct page *page, | |
962 | int order, int migratetype) | |
963 | { | |
964 | if (!capc || order != capc->cc->order) | |
965 | return false; | |
966 | ||
967 | /* Do not accidentally pollute CMA or isolated regions*/ | |
968 | if (is_migrate_cma(migratetype) || | |
969 | is_migrate_isolate(migratetype)) | |
970 | return false; | |
971 | ||
972 | /* | |
f0953a1b | 973 | * Do not let lower order allocations pollute a movable pageblock. |
5e1f0f09 MG |
974 | * This might let an unmovable request use a reclaimable pageblock |
975 | * and vice-versa but no more than normal fallback logic which can | |
976 | * have trouble finding a high-order free page. | |
977 | */ | |
978 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
979 | return false; | |
980 | ||
981 | capc->page = page; | |
982 | return true; | |
983 | } | |
984 | ||
985 | #else | |
986 | static inline struct capture_control *task_capc(struct zone *zone) | |
987 | { | |
988 | return NULL; | |
989 | } | |
990 | ||
991 | static inline bool | |
992 | compaction_capture(struct capture_control *capc, struct page *page, | |
993 | int order, int migratetype) | |
994 | { | |
995 | return false; | |
996 | } | |
997 | #endif /* CONFIG_COMPACTION */ | |
998 | ||
6ab01363 AD |
999 | /* Used for pages not on another list */ |
1000 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
1001 | unsigned int order, int migratetype) | |
1002 | { | |
1003 | struct free_area *area = &zone->free_area[order]; | |
1004 | ||
bf75f200 | 1005 | list_add(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
1006 | area->nr_free++; |
1007 | } | |
1008 | ||
1009 | /* Used for pages not on another list */ | |
1010 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
1011 | unsigned int order, int migratetype) | |
1012 | { | |
1013 | struct free_area *area = &zone->free_area[order]; | |
1014 | ||
bf75f200 | 1015 | list_add_tail(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
1016 | area->nr_free++; |
1017 | } | |
1018 | ||
293ffa5e DH |
1019 | /* |
1020 | * Used for pages which are on another list. Move the pages to the tail | |
1021 | * of the list - so the moved pages won't immediately be considered for | |
1022 | * allocation again (e.g., optimization for memory onlining). | |
1023 | */ | |
6ab01363 AD |
1024 | static inline void move_to_free_list(struct page *page, struct zone *zone, |
1025 | unsigned int order, int migratetype) | |
1026 | { | |
1027 | struct free_area *area = &zone->free_area[order]; | |
1028 | ||
bf75f200 | 1029 | list_move_tail(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
1030 | } |
1031 | ||
1032 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
1033 | unsigned int order) | |
1034 | { | |
36e66c55 AD |
1035 | /* clear reported state and update reported page count */ |
1036 | if (page_reported(page)) | |
1037 | __ClearPageReported(page); | |
1038 | ||
bf75f200 | 1039 | list_del(&page->buddy_list); |
6ab01363 AD |
1040 | __ClearPageBuddy(page); |
1041 | set_page_private(page, 0); | |
1042 | zone->free_area[order].nr_free--; | |
1043 | } | |
1044 | ||
a2129f24 AD |
1045 | /* |
1046 | * If this is not the largest possible page, check if the buddy | |
1047 | * of the next-highest order is free. If it is, it's possible | |
1048 | * that pages are being freed that will coalesce soon. In case, | |
1049 | * that is happening, add the free page to the tail of the list | |
1050 | * so it's less likely to be used soon and more likely to be merged | |
1051 | * as a higher order page | |
1052 | */ | |
1053 | static inline bool | |
1054 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
1055 | struct page *page, unsigned int order) | |
1056 | { | |
8170ac47 ZY |
1057 | unsigned long higher_page_pfn; |
1058 | struct page *higher_page; | |
a2129f24 AD |
1059 | |
1060 | if (order >= MAX_ORDER - 2) | |
1061 | return false; | |
1062 | ||
8170ac47 ZY |
1063 | higher_page_pfn = buddy_pfn & pfn; |
1064 | higher_page = page + (higher_page_pfn - pfn); | |
a2129f24 | 1065 | |
8170ac47 ZY |
1066 | return find_buddy_page_pfn(higher_page, higher_page_pfn, order + 1, |
1067 | NULL) != NULL; | |
a2129f24 AD |
1068 | } |
1069 | ||
1da177e4 LT |
1070 | /* |
1071 | * Freeing function for a buddy system allocator. | |
1072 | * | |
1073 | * The concept of a buddy system is to maintain direct-mapped table | |
1074 | * (containing bit values) for memory blocks of various "orders". | |
1075 | * The bottom level table contains the map for the smallest allocatable | |
1076 | * units of memory (here, pages), and each level above it describes | |
1077 | * pairs of units from the levels below, hence, "buddies". | |
1078 | * At a high level, all that happens here is marking the table entry | |
1079 | * at the bottom level available, and propagating the changes upward | |
1080 | * as necessary, plus some accounting needed to play nicely with other | |
1081 | * parts of the VM system. | |
1082 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
1083 | * free pages of length of (1 << order) and marked with PageBuddy. |
1084 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 1085 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
1086 | * other. That is, if we allocate a small block, and both were |
1087 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 1088 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 1089 | * triggers coalescing into a block of larger size. |
1da177e4 | 1090 | * |
6d49e352 | 1091 | * -- nyc |
1da177e4 LT |
1092 | */ |
1093 | ||
48db57f8 | 1094 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 1095 | unsigned long pfn, |
ed0ae21d | 1096 | struct zone *zone, unsigned int order, |
f04a5d5d | 1097 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1098 | { |
a2129f24 | 1099 | struct capture_control *capc = task_capc(zone); |
dae37a5d | 1100 | unsigned long buddy_pfn = 0; |
a2129f24 | 1101 | unsigned long combined_pfn; |
a2129f24 AD |
1102 | struct page *buddy; |
1103 | bool to_tail; | |
d9dddbf5 | 1104 | |
d29bb978 | 1105 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 1106 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 1107 | |
ed0ae21d | 1108 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 1109 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 1110 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 1111 | |
76741e77 | 1112 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 1113 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 1114 | |
bb0e28eb | 1115 | while (order < MAX_ORDER - 1) { |
5e1f0f09 MG |
1116 | if (compaction_capture(capc, page, order, migratetype)) { |
1117 | __mod_zone_freepage_state(zone, -(1 << order), | |
1118 | migratetype); | |
1119 | return; | |
1120 | } | |
13ad59df | 1121 | |
8170ac47 ZY |
1122 | buddy = find_buddy_page_pfn(page, pfn, order, &buddy_pfn); |
1123 | if (!buddy) | |
d9dddbf5 | 1124 | goto done_merging; |
bb0e28eb ZY |
1125 | |
1126 | if (unlikely(order >= pageblock_order)) { | |
1127 | /* | |
1128 | * We want to prevent merge between freepages on pageblock | |
1129 | * without fallbacks and normal pageblock. Without this, | |
1130 | * pageblock isolation could cause incorrect freepage or CMA | |
1131 | * accounting or HIGHATOMIC accounting. | |
1132 | */ | |
1133 | int buddy_mt = get_pageblock_migratetype(buddy); | |
1134 | ||
1135 | if (migratetype != buddy_mt | |
1136 | && (!migratetype_is_mergeable(migratetype) || | |
1137 | !migratetype_is_mergeable(buddy_mt))) | |
1138 | goto done_merging; | |
1139 | } | |
1140 | ||
c0a32fc5 SG |
1141 | /* |
1142 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
1143 | * merge with it and move up one order. | |
1144 | */ | |
b03641af | 1145 | if (page_is_guard(buddy)) |
2847cf95 | 1146 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 1147 | else |
6ab01363 | 1148 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
1149 | combined_pfn = buddy_pfn & pfn; |
1150 | page = page + (combined_pfn - pfn); | |
1151 | pfn = combined_pfn; | |
1da177e4 LT |
1152 | order++; |
1153 | } | |
d9dddbf5 VB |
1154 | |
1155 | done_merging: | |
ab130f91 | 1156 | set_buddy_order(page, order); |
6dda9d55 | 1157 | |
47b6a24a DH |
1158 | if (fpi_flags & FPI_TO_TAIL) |
1159 | to_tail = true; | |
1160 | else if (is_shuffle_order(order)) | |
a2129f24 | 1161 | to_tail = shuffle_pick_tail(); |
97500a4a | 1162 | else |
a2129f24 | 1163 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 1164 | |
a2129f24 | 1165 | if (to_tail) |
6ab01363 | 1166 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 1167 | else |
6ab01363 | 1168 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
1169 | |
1170 | /* Notify page reporting subsystem of freed page */ | |
f04a5d5d | 1171 | if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY)) |
36e66c55 | 1172 | page_reporting_notify_free(order); |
1da177e4 LT |
1173 | } |
1174 | ||
b2c9e2fb ZY |
1175 | /** |
1176 | * split_free_page() -- split a free page at split_pfn_offset | |
1177 | * @free_page: the original free page | |
1178 | * @order: the order of the page | |
1179 | * @split_pfn_offset: split offset within the page | |
1180 | * | |
86d28b07 ZY |
1181 | * Return -ENOENT if the free page is changed, otherwise 0 |
1182 | * | |
b2c9e2fb ZY |
1183 | * It is used when the free page crosses two pageblocks with different migratetypes |
1184 | * at split_pfn_offset within the page. The split free page will be put into | |
1185 | * separate migratetype lists afterwards. Otherwise, the function achieves | |
1186 | * nothing. | |
1187 | */ | |
86d28b07 ZY |
1188 | int split_free_page(struct page *free_page, |
1189 | unsigned int order, unsigned long split_pfn_offset) | |
b2c9e2fb ZY |
1190 | { |
1191 | struct zone *zone = page_zone(free_page); | |
1192 | unsigned long free_page_pfn = page_to_pfn(free_page); | |
1193 | unsigned long pfn; | |
1194 | unsigned long flags; | |
1195 | int free_page_order; | |
86d28b07 ZY |
1196 | int mt; |
1197 | int ret = 0; | |
b2c9e2fb | 1198 | |
88ee1343 | 1199 | if (split_pfn_offset == 0) |
86d28b07 | 1200 | return ret; |
88ee1343 | 1201 | |
b2c9e2fb | 1202 | spin_lock_irqsave(&zone->lock, flags); |
86d28b07 ZY |
1203 | |
1204 | if (!PageBuddy(free_page) || buddy_order(free_page) != order) { | |
1205 | ret = -ENOENT; | |
1206 | goto out; | |
1207 | } | |
1208 | ||
1209 | mt = get_pageblock_migratetype(free_page); | |
1210 | if (likely(!is_migrate_isolate(mt))) | |
1211 | __mod_zone_freepage_state(zone, -(1UL << order), mt); | |
1212 | ||
b2c9e2fb ZY |
1213 | del_page_from_free_list(free_page, zone, order); |
1214 | for (pfn = free_page_pfn; | |
1215 | pfn < free_page_pfn + (1UL << order);) { | |
1216 | int mt = get_pfnblock_migratetype(pfn_to_page(pfn), pfn); | |
1217 | ||
86d28b07 | 1218 | free_page_order = min_t(unsigned int, |
88ee1343 ZY |
1219 | pfn ? __ffs(pfn) : order, |
1220 | __fls(split_pfn_offset)); | |
b2c9e2fb ZY |
1221 | __free_one_page(pfn_to_page(pfn), pfn, zone, free_page_order, |
1222 | mt, FPI_NONE); | |
1223 | pfn += 1UL << free_page_order; | |
1224 | split_pfn_offset -= (1UL << free_page_order); | |
1225 | /* we have done the first part, now switch to second part */ | |
1226 | if (split_pfn_offset == 0) | |
1227 | split_pfn_offset = (1UL << order) - (pfn - free_page_pfn); | |
1228 | } | |
86d28b07 | 1229 | out: |
b2c9e2fb | 1230 | spin_unlock_irqrestore(&zone->lock, flags); |
86d28b07 | 1231 | return ret; |
b2c9e2fb | 1232 | } |
7bfec6f4 MG |
1233 | /* |
1234 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1235 | * try and check multiple fields with one check. The caller must do a detailed | |
1236 | * check if necessary. | |
1237 | */ | |
1238 | static inline bool page_expected_state(struct page *page, | |
1239 | unsigned long check_flags) | |
1240 | { | |
1241 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1242 | return false; | |
1243 | ||
1244 | if (unlikely((unsigned long)page->mapping | | |
1245 | page_ref_count(page) | | |
1246 | #ifdef CONFIG_MEMCG | |
48060834 | 1247 | page->memcg_data | |
7bfec6f4 MG |
1248 | #endif |
1249 | (page->flags & check_flags))) | |
1250 | return false; | |
1251 | ||
1252 | return true; | |
1253 | } | |
1254 | ||
58b7f119 | 1255 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 1256 | { |
82a3241a | 1257 | const char *bad_reason = NULL; |
f0b791a3 | 1258 | |
53f9263b | 1259 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1260 | bad_reason = "nonzero mapcount"; |
1261 | if (unlikely(page->mapping != NULL)) | |
1262 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1263 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1264 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
1265 | if (unlikely(page->flags & flags)) { |
1266 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
1267 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
1268 | else | |
1269 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 1270 | } |
9edad6ea | 1271 | #ifdef CONFIG_MEMCG |
48060834 | 1272 | if (unlikely(page->memcg_data)) |
9edad6ea JW |
1273 | bad_reason = "page still charged to cgroup"; |
1274 | #endif | |
58b7f119 WY |
1275 | return bad_reason; |
1276 | } | |
1277 | ||
a8368cd8 | 1278 | static void free_page_is_bad_report(struct page *page) |
58b7f119 WY |
1279 | { |
1280 | bad_page(page, | |
1281 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
1282 | } |
1283 | ||
a8368cd8 | 1284 | static inline bool free_page_is_bad(struct page *page) |
bb552ac6 | 1285 | { |
da838d4f | 1286 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
a8368cd8 | 1287 | return false; |
bb552ac6 MG |
1288 | |
1289 | /* Something has gone sideways, find it */ | |
a8368cd8 AM |
1290 | free_page_is_bad_report(page); |
1291 | return true; | |
1da177e4 LT |
1292 | } |
1293 | ||
4db7548c MG |
1294 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1295 | { | |
94688e8e | 1296 | struct folio *folio = (struct folio *)head_page; |
4db7548c MG |
1297 | int ret = 1; |
1298 | ||
1299 | /* | |
1300 | * We rely page->lru.next never has bit 0 set, unless the page | |
1301 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1302 | */ | |
1303 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1304 | ||
1305 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1306 | ret = 0; | |
1307 | goto out; | |
1308 | } | |
1309 | switch (page - head_page) { | |
1310 | case 1: | |
cb67f428 | 1311 | /* the first tail page: these may be in place of ->mapping */ |
65a689f3 MWO |
1312 | if (unlikely(folio_entire_mapcount(folio))) { |
1313 | bad_page(page, "nonzero entire_mapcount"); | |
4db7548c MG |
1314 | goto out; |
1315 | } | |
65a689f3 MWO |
1316 | if (unlikely(atomic_read(&folio->_nr_pages_mapped))) { |
1317 | bad_page(page, "nonzero nr_pages_mapped"); | |
cb67f428 HD |
1318 | goto out; |
1319 | } | |
94688e8e MWO |
1320 | if (unlikely(atomic_read(&folio->_pincount))) { |
1321 | bad_page(page, "nonzero pincount"); | |
cb67f428 HD |
1322 | goto out; |
1323 | } | |
4db7548c MG |
1324 | break; |
1325 | case 2: | |
1326 | /* | |
1327 | * the second tail page: ->mapping is | |
fa3015b7 | 1328 | * deferred_list.next -- ignore value. |
4db7548c MG |
1329 | */ |
1330 | break; | |
1331 | default: | |
1332 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 1333 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
1334 | goto out; |
1335 | } | |
1336 | break; | |
1337 | } | |
1338 | if (unlikely(!PageTail(page))) { | |
82a3241a | 1339 | bad_page(page, "PageTail not set"); |
4db7548c MG |
1340 | goto out; |
1341 | } | |
1342 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 1343 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
1344 | goto out; |
1345 | } | |
1346 | ret = 0; | |
1347 | out: | |
1348 | page->mapping = NULL; | |
1349 | clear_compound_head(page); | |
1350 | return ret; | |
1351 | } | |
1352 | ||
94ae8b83 AK |
1353 | /* |
1354 | * Skip KASAN memory poisoning when either: | |
1355 | * | |
1356 | * 1. Deferred memory initialization has not yet completed, | |
1357 | * see the explanation below. | |
1358 | * 2. Skipping poisoning is requested via FPI_SKIP_KASAN_POISON, | |
1359 | * see the comment next to it. | |
1360 | * 3. Skipping poisoning is requested via __GFP_SKIP_KASAN_POISON, | |
1361 | * see the comment next to it. | |
44383cef AK |
1362 | * 4. The allocation is excluded from being checked due to sampling, |
1363 | * see the call to kasan_unpoison_pages. | |
94ae8b83 AK |
1364 | * |
1365 | * Poisoning pages during deferred memory init will greatly lengthen the | |
1366 | * process and cause problem in large memory systems as the deferred pages | |
1367 | * initialization is done with interrupt disabled. | |
1368 | * | |
1369 | * Assuming that there will be no reference to those newly initialized | |
1370 | * pages before they are ever allocated, this should have no effect on | |
1371 | * KASAN memory tracking as the poison will be properly inserted at page | |
1372 | * allocation time. The only corner case is when pages are allocated by | |
1373 | * on-demand allocation and then freed again before the deferred pages | |
1374 | * initialization is done, but this is not likely to happen. | |
1375 | */ | |
1376 | static inline bool should_skip_kasan_poison(struct page *page, fpi_t fpi_flags) | |
1377 | { | |
1378 | return deferred_pages_enabled() || | |
1379 | (!IS_ENABLED(CONFIG_KASAN_GENERIC) && | |
1380 | (fpi_flags & FPI_SKIP_KASAN_POISON)) || | |
1381 | PageSkipKASanPoison(page); | |
1382 | } | |
1383 | ||
aeaec8e2 | 1384 | static void kernel_init_pages(struct page *page, int numpages) |
6471384a AP |
1385 | { |
1386 | int i; | |
1387 | ||
9e15afa5 QC |
1388 | /* s390's use of memset() could override KASAN redzones. */ |
1389 | kasan_disable_current(); | |
d9da8f6c AK |
1390 | for (i = 0; i < numpages; i++) |
1391 | clear_highpage_kasan_tagged(page + i); | |
9e15afa5 | 1392 | kasan_enable_current(); |
6471384a AP |
1393 | } |
1394 | ||
e2769dbd | 1395 | static __always_inline bool free_pages_prepare(struct page *page, |
2c335680 | 1396 | unsigned int order, bool check_free, fpi_t fpi_flags) |
4db7548c | 1397 | { |
e2769dbd | 1398 | int bad = 0; |
c3525330 | 1399 | bool init = want_init_on_free(); |
4db7548c | 1400 | |
4db7548c MG |
1401 | VM_BUG_ON_PAGE(PageTail(page), page); |
1402 | ||
e2769dbd | 1403 | trace_mm_page_free(page, order); |
b073d7f8 | 1404 | kmsan_free_page(page, order); |
e2769dbd | 1405 | |
79f5f8fa OS |
1406 | if (unlikely(PageHWPoison(page)) && !order) { |
1407 | /* | |
1408 | * Do not let hwpoison pages hit pcplists/buddy | |
1409 | * Untie memcg state and reset page's owner | |
1410 | */ | |
18b2db3b | 1411 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
79f5f8fa OS |
1412 | __memcg_kmem_uncharge_page(page, order); |
1413 | reset_page_owner(page, order); | |
df4e817b | 1414 | page_table_check_free(page, order); |
79f5f8fa OS |
1415 | return false; |
1416 | } | |
1417 | ||
e2769dbd MG |
1418 | /* |
1419 | * Check tail pages before head page information is cleared to | |
1420 | * avoid checking PageCompound for order-0 pages. | |
1421 | */ | |
1422 | if (unlikely(order)) { | |
1423 | bool compound = PageCompound(page); | |
1424 | int i; | |
1425 | ||
1426 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1427 | |
cb67f428 | 1428 | if (compound) |
eac96c3e | 1429 | ClearPageHasHWPoisoned(page); |
e2769dbd MG |
1430 | for (i = 1; i < (1 << order); i++) { |
1431 | if (compound) | |
1432 | bad += free_tail_pages_check(page, page + i); | |
a8368cd8 | 1433 | if (unlikely(free_page_is_bad(page + i))) { |
e2769dbd MG |
1434 | bad++; |
1435 | continue; | |
1436 | } | |
1437 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1438 | } | |
1439 | } | |
bda807d4 | 1440 | if (PageMappingFlags(page)) |
4db7548c | 1441 | page->mapping = NULL; |
18b2db3b | 1442 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
f4b00eab | 1443 | __memcg_kmem_uncharge_page(page, order); |
a8368cd8 AM |
1444 | if (check_free && free_page_is_bad(page)) |
1445 | bad++; | |
e2769dbd MG |
1446 | if (bad) |
1447 | return false; | |
4db7548c | 1448 | |
e2769dbd MG |
1449 | page_cpupid_reset_last(page); |
1450 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1451 | reset_page_owner(page, order); | |
df4e817b | 1452 | page_table_check_free(page, order); |
4db7548c MG |
1453 | |
1454 | if (!PageHighMem(page)) { | |
1455 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1456 | PAGE_SIZE << order); |
4db7548c | 1457 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1458 | PAGE_SIZE << order); |
4db7548c | 1459 | } |
6471384a | 1460 | |
8db26a3d VB |
1461 | kernel_poison_pages(page, 1 << order); |
1462 | ||
f9d79e8d | 1463 | /* |
1bb5eab3 | 1464 | * As memory initialization might be integrated into KASAN, |
7c13c163 | 1465 | * KASAN poisoning and memory initialization code must be |
1bb5eab3 AK |
1466 | * kept together to avoid discrepancies in behavior. |
1467 | * | |
f9d79e8d AK |
1468 | * With hardware tag-based KASAN, memory tags must be set before the |
1469 | * page becomes unavailable via debug_pagealloc or arch_free_page. | |
1470 | */ | |
487a32ec | 1471 | if (!should_skip_kasan_poison(page, fpi_flags)) { |
c3525330 | 1472 | kasan_poison_pages(page, order, init); |
f9d79e8d | 1473 | |
db8a0477 AK |
1474 | /* Memory is already initialized if KASAN did it internally. */ |
1475 | if (kasan_has_integrated_init()) | |
1476 | init = false; | |
1477 | } | |
1478 | if (init) | |
aeaec8e2 | 1479 | kernel_init_pages(page, 1 << order); |
db8a0477 | 1480 | |
234fdce8 QC |
1481 | /* |
1482 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1483 | * does this. So nothing which can access the page's contents should | |
1484 | * happen after this. | |
1485 | */ | |
1486 | arch_free_page(page, order); | |
1487 | ||
77bc7fd6 | 1488 | debug_pagealloc_unmap_pages(page, 1 << order); |
d6332692 | 1489 | |
4db7548c MG |
1490 | return true; |
1491 | } | |
1492 | ||
e2769dbd | 1493 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1494 | /* |
1495 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1496 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1497 | * moved from pcp lists to free lists. | |
1498 | */ | |
44042b44 | 1499 | static bool free_pcp_prepare(struct page *page, unsigned int order) |
e2769dbd | 1500 | { |
44042b44 | 1501 | return free_pages_prepare(page, order, true, FPI_NONE); |
e2769dbd MG |
1502 | } |
1503 | ||
d452289f | 1504 | /* return true if this page has an inappropriate state */ |
4462b32c | 1505 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1506 | { |
8e57f8ac | 1507 | if (debug_pagealloc_enabled_static()) |
a8368cd8 | 1508 | return free_page_is_bad(page); |
4462b32c VB |
1509 | else |
1510 | return false; | |
e2769dbd MG |
1511 | } |
1512 | #else | |
4462b32c VB |
1513 | /* |
1514 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1515 | * moving from pcp lists to free list in order to reduce overhead. With | |
1516 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1517 | * to the pcp lists. | |
1518 | */ | |
44042b44 | 1519 | static bool free_pcp_prepare(struct page *page, unsigned int order) |
e2769dbd | 1520 | { |
8e57f8ac | 1521 | if (debug_pagealloc_enabled_static()) |
44042b44 | 1522 | return free_pages_prepare(page, order, true, FPI_NONE); |
4462b32c | 1523 | else |
44042b44 | 1524 | return free_pages_prepare(page, order, false, FPI_NONE); |
e2769dbd MG |
1525 | } |
1526 | ||
4db7548c MG |
1527 | static bool bulkfree_pcp_prepare(struct page *page) |
1528 | { | |
a8368cd8 | 1529 | return free_page_is_bad(page); |
4db7548c MG |
1530 | } |
1531 | #endif /* CONFIG_DEBUG_VM */ | |
1532 | ||
1da177e4 | 1533 | /* |
5f8dcc21 | 1534 | * Frees a number of pages from the PCP lists |
7cba630b | 1535 | * Assumes all pages on list are in same zone. |
207f36ee | 1536 | * count is the number of pages to free. |
1da177e4 | 1537 | */ |
5f8dcc21 | 1538 | static void free_pcppages_bulk(struct zone *zone, int count, |
fd56eef2 MG |
1539 | struct per_cpu_pages *pcp, |
1540 | int pindex) | |
1da177e4 | 1541 | { |
57490774 | 1542 | unsigned long flags; |
35b6d770 MG |
1543 | int min_pindex = 0; |
1544 | int max_pindex = NR_PCP_LISTS - 1; | |
44042b44 | 1545 | unsigned int order; |
3777999d | 1546 | bool isolated_pageblocks; |
8b10b465 | 1547 | struct page *page; |
f2260e6b | 1548 | |
88e8ac11 CTR |
1549 | /* |
1550 | * Ensure proper count is passed which otherwise would stuck in the | |
1551 | * below while (list_empty(list)) loop. | |
1552 | */ | |
1553 | count = min(pcp->count, count); | |
d61372bc MG |
1554 | |
1555 | /* Ensure requested pindex is drained first. */ | |
1556 | pindex = pindex - 1; | |
1557 | ||
57490774 | 1558 | spin_lock_irqsave(&zone->lock, flags); |
8b10b465 MG |
1559 | isolated_pageblocks = has_isolate_pageblock(zone); |
1560 | ||
44042b44 | 1561 | while (count > 0) { |
5f8dcc21 | 1562 | struct list_head *list; |
fd56eef2 | 1563 | int nr_pages; |
5f8dcc21 | 1564 | |
fd56eef2 | 1565 | /* Remove pages from lists in a round-robin fashion. */ |
5f8dcc21 | 1566 | do { |
35b6d770 MG |
1567 | if (++pindex > max_pindex) |
1568 | pindex = min_pindex; | |
44042b44 | 1569 | list = &pcp->lists[pindex]; |
35b6d770 MG |
1570 | if (!list_empty(list)) |
1571 | break; | |
1572 | ||
1573 | if (pindex == max_pindex) | |
1574 | max_pindex--; | |
1575 | if (pindex == min_pindex) | |
1576 | min_pindex++; | |
1577 | } while (1); | |
48db57f8 | 1578 | |
44042b44 | 1579 | order = pindex_to_order(pindex); |
fd56eef2 | 1580 | nr_pages = 1 << order; |
a6f9edd6 | 1581 | do { |
8b10b465 MG |
1582 | int mt; |
1583 | ||
bf75f200 | 1584 | page = list_last_entry(list, struct page, pcp_list); |
8b10b465 MG |
1585 | mt = get_pcppage_migratetype(page); |
1586 | ||
0a5f4e5b | 1587 | /* must delete to avoid corrupting pcp list */ |
bf75f200 | 1588 | list_del(&page->pcp_list); |
fd56eef2 MG |
1589 | count -= nr_pages; |
1590 | pcp->count -= nr_pages; | |
aa016d14 | 1591 | |
4db7548c MG |
1592 | if (bulkfree_pcp_prepare(page)) |
1593 | continue; | |
1594 | ||
8b10b465 MG |
1595 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1596 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1597 | /* Pageblock could have been isolated meanwhile */ | |
1598 | if (unlikely(isolated_pageblocks)) | |
1599 | mt = get_pageblock_migratetype(page); | |
0a5f4e5b | 1600 | |
8b10b465 MG |
1601 | __free_one_page(page, page_to_pfn(page), zone, order, mt, FPI_NONE); |
1602 | trace_mm_page_pcpu_drain(page, order, mt); | |
1603 | } while (count > 0 && !list_empty(list)); | |
0a5f4e5b | 1604 | } |
8b10b465 | 1605 | |
57490774 | 1606 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 LT |
1607 | } |
1608 | ||
dc4b0caf MG |
1609 | static void free_one_page(struct zone *zone, |
1610 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1611 | unsigned int order, |
7fef431b | 1612 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1613 | { |
df1acc85 MG |
1614 | unsigned long flags; |
1615 | ||
1616 | spin_lock_irqsave(&zone->lock, flags); | |
ad53f92e JK |
1617 | if (unlikely(has_isolate_pageblock(zone) || |
1618 | is_migrate_isolate(migratetype))) { | |
1619 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1620 | } |
7fef431b | 1621 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); |
df1acc85 | 1622 | spin_unlock_irqrestore(&zone->lock, flags); |
48db57f8 NP |
1623 | } |
1624 | ||
1e8ce83c | 1625 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1626 | unsigned long zone, int nid) |
1e8ce83c | 1627 | { |
d0dc12e8 | 1628 | mm_zero_struct_page(page); |
1e8ce83c | 1629 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1630 | init_page_count(page); |
1631 | page_mapcount_reset(page); | |
1632 | page_cpupid_reset_last(page); | |
2813b9c0 | 1633 | page_kasan_tag_reset(page); |
1e8ce83c | 1634 | |
1e8ce83c RH |
1635 | INIT_LIST_HEAD(&page->lru); |
1636 | #ifdef WANT_PAGE_VIRTUAL | |
1637 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1638 | if (!is_highmem_idx(zone)) | |
1639 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1640 | #endif | |
1641 | } | |
1642 | ||
7e18adb4 | 1643 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1644 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1645 | { |
1646 | pg_data_t *pgdat; | |
1647 | int nid, zid; | |
1648 | ||
fc574488 | 1649 | if (early_page_initialised(pfn)) |
7e18adb4 MG |
1650 | return; |
1651 | ||
1652 | nid = early_pfn_to_nid(pfn); | |
1653 | pgdat = NODE_DATA(nid); | |
1654 | ||
1655 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1656 | struct zone *zone = &pgdat->node_zones[zid]; | |
1657 | ||
86fb05b9 | 1658 | if (zone_spans_pfn(zone, pfn)) |
7e18adb4 MG |
1659 | break; |
1660 | } | |
d0dc12e8 | 1661 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1662 | } |
1663 | #else | |
1664 | static inline void init_reserved_page(unsigned long pfn) | |
1665 | { | |
1666 | } | |
1667 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1668 | ||
92923ca3 NZ |
1669 | /* |
1670 | * Initialised pages do not have PageReserved set. This function is | |
1671 | * called for each range allocated by the bootmem allocator and | |
1672 | * marks the pages PageReserved. The remaining valid pages are later | |
1673 | * sent to the buddy page allocator. | |
1674 | */ | |
4b50bcc7 | 1675 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1676 | { |
1677 | unsigned long start_pfn = PFN_DOWN(start); | |
1678 | unsigned long end_pfn = PFN_UP(end); | |
1679 | ||
7e18adb4 MG |
1680 | for (; start_pfn < end_pfn; start_pfn++) { |
1681 | if (pfn_valid(start_pfn)) { | |
1682 | struct page *page = pfn_to_page(start_pfn); | |
1683 | ||
1684 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1685 | |
1686 | /* Avoid false-positive PageTail() */ | |
1687 | INIT_LIST_HEAD(&page->lru); | |
1688 | ||
d483da5b AD |
1689 | /* |
1690 | * no need for atomic set_bit because the struct | |
1691 | * page is not visible yet so nobody should | |
1692 | * access it yet. | |
1693 | */ | |
1694 | __SetPageReserved(page); | |
7e18adb4 MG |
1695 | } |
1696 | } | |
92923ca3 NZ |
1697 | } |
1698 | ||
7fef431b DH |
1699 | static void __free_pages_ok(struct page *page, unsigned int order, |
1700 | fpi_t fpi_flags) | |
ec95f53a | 1701 | { |
d34b0733 | 1702 | unsigned long flags; |
95e34412 | 1703 | int migratetype; |
dc4b0caf | 1704 | unsigned long pfn = page_to_pfn(page); |
56f0e661 | 1705 | struct zone *zone = page_zone(page); |
ec95f53a | 1706 | |
2c335680 | 1707 | if (!free_pages_prepare(page, order, true, fpi_flags)) |
ec95f53a KM |
1708 | return; |
1709 | ||
ac4b2901 DW |
1710 | /* |
1711 | * Calling get_pfnblock_migratetype() without spin_lock_irqsave() here | |
1712 | * is used to avoid calling get_pfnblock_migratetype() under the lock. | |
1713 | * This will reduce the lock holding time. | |
1714 | */ | |
cfc47a28 | 1715 | migratetype = get_pfnblock_migratetype(page, pfn); |
dbbee9d5 | 1716 | |
56f0e661 | 1717 | spin_lock_irqsave(&zone->lock, flags); |
56f0e661 MG |
1718 | if (unlikely(has_isolate_pageblock(zone) || |
1719 | is_migrate_isolate(migratetype))) { | |
1720 | migratetype = get_pfnblock_migratetype(page, pfn); | |
1721 | } | |
1722 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); | |
1723 | spin_unlock_irqrestore(&zone->lock, flags); | |
90249993 | 1724 | |
d34b0733 | 1725 | __count_vm_events(PGFREE, 1 << order); |
1da177e4 LT |
1726 | } |
1727 | ||
a9cd410a | 1728 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1729 | { |
c3993076 | 1730 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1731 | struct page *p = page; |
c3993076 | 1732 | unsigned int loop; |
a226f6c8 | 1733 | |
7fef431b DH |
1734 | /* |
1735 | * When initializing the memmap, __init_single_page() sets the refcount | |
1736 | * of all pages to 1 ("allocated"/"not free"). We have to set the | |
1737 | * refcount of all involved pages to 0. | |
1738 | */ | |
e2d0bd2b YL |
1739 | prefetchw(p); |
1740 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1741 | prefetchw(p + 1); | |
c3993076 JW |
1742 | __ClearPageReserved(p); |
1743 | set_page_count(p, 0); | |
a226f6c8 | 1744 | } |
e2d0bd2b YL |
1745 | __ClearPageReserved(p); |
1746 | set_page_count(p, 0); | |
c3993076 | 1747 | |
9705bea5 | 1748 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
7fef431b DH |
1749 | |
1750 | /* | |
1751 | * Bypass PCP and place fresh pages right to the tail, primarily | |
1752 | * relevant for memory onlining. | |
1753 | */ | |
2c335680 | 1754 | __free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON); |
a226f6c8 DH |
1755 | } |
1756 | ||
a9ee6cf5 | 1757 | #ifdef CONFIG_NUMA |
7ace9917 | 1758 | |
03e92a5e MR |
1759 | /* |
1760 | * During memory init memblocks map pfns to nids. The search is expensive and | |
1761 | * this caches recent lookups. The implementation of __early_pfn_to_nid | |
1762 | * treats start/end as pfns. | |
1763 | */ | |
1764 | struct mminit_pfnnid_cache { | |
1765 | unsigned long last_start; | |
1766 | unsigned long last_end; | |
1767 | int last_nid; | |
1768 | }; | |
75a592a4 | 1769 | |
03e92a5e | 1770 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
6f24fbd3 MR |
1771 | |
1772 | /* | |
1773 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
1774 | */ | |
03e92a5e | 1775 | static int __meminit __early_pfn_to_nid(unsigned long pfn, |
6f24fbd3 | 1776 | struct mminit_pfnnid_cache *state) |
75a592a4 | 1777 | { |
6f24fbd3 | 1778 | unsigned long start_pfn, end_pfn; |
75a592a4 MG |
1779 | int nid; |
1780 | ||
6f24fbd3 MR |
1781 | if (state->last_start <= pfn && pfn < state->last_end) |
1782 | return state->last_nid; | |
1783 | ||
1784 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); | |
1785 | if (nid != NUMA_NO_NODE) { | |
1786 | state->last_start = start_pfn; | |
1787 | state->last_end = end_pfn; | |
1788 | state->last_nid = nid; | |
1789 | } | |
7ace9917 MG |
1790 | |
1791 | return nid; | |
75a592a4 | 1792 | } |
75a592a4 | 1793 | |
75a592a4 | 1794 | int __meminit early_pfn_to_nid(unsigned long pfn) |
75a592a4 | 1795 | { |
7ace9917 | 1796 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1797 | int nid; |
1798 | ||
7ace9917 | 1799 | spin_lock(&early_pfn_lock); |
56ec43d8 | 1800 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1801 | if (nid < 0) |
e4568d38 | 1802 | nid = first_online_node; |
7ace9917 | 1803 | spin_unlock(&early_pfn_lock); |
75a592a4 | 1804 | |
7ace9917 | 1805 | return nid; |
75a592a4 | 1806 | } |
a9ee6cf5 | 1807 | #endif /* CONFIG_NUMA */ |
75a592a4 | 1808 | |
7c2ee349 | 1809 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1810 | unsigned int order) |
1811 | { | |
fc574488 | 1812 | if (!early_page_initialised(pfn)) |
3a80a7fa | 1813 | return; |
3c206509 AP |
1814 | if (!kmsan_memblock_free_pages(page, order)) { |
1815 | /* KMSAN will take care of these pages. */ | |
1816 | return; | |
1817 | } | |
a9cd410a | 1818 | __free_pages_core(page, order); |
3a80a7fa MG |
1819 | } |
1820 | ||
7cf91a98 JK |
1821 | /* |
1822 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1823 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
859a85dd | 1824 | * with the migration of free compaction scanner. |
7cf91a98 JK |
1825 | * |
1826 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1827 | * | |
1828 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1829 | * i.e. it's possible that all pages within a zones range of pages do not | |
1830 | * belong to a single zone. We assume that a border between node0 and node1 | |
1831 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1832 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1833 | * the first and last page of a pageblock and avoid checking each individual | |
1834 | * page in a pageblock. | |
1835 | */ | |
1836 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1837 | unsigned long end_pfn, struct zone *zone) | |
1838 | { | |
1839 | struct page *start_page; | |
1840 | struct page *end_page; | |
1841 | ||
1842 | /* end_pfn is one past the range we are checking */ | |
1843 | end_pfn--; | |
1844 | ||
1845 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1846 | return NULL; | |
1847 | ||
2d070eab MH |
1848 | start_page = pfn_to_online_page(start_pfn); |
1849 | if (!start_page) | |
1850 | return NULL; | |
7cf91a98 JK |
1851 | |
1852 | if (page_zone(start_page) != zone) | |
1853 | return NULL; | |
1854 | ||
1855 | end_page = pfn_to_page(end_pfn); | |
1856 | ||
1857 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1858 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1859 | return NULL; | |
1860 | ||
1861 | return start_page; | |
1862 | } | |
1863 | ||
1864 | void set_zone_contiguous(struct zone *zone) | |
1865 | { | |
1866 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1867 | unsigned long block_end_pfn; | |
1868 | ||
4f9bc69a | 1869 | block_end_pfn = pageblock_end_pfn(block_start_pfn); |
7cf91a98 JK |
1870 | for (; block_start_pfn < zone_end_pfn(zone); |
1871 | block_start_pfn = block_end_pfn, | |
1872 | block_end_pfn += pageblock_nr_pages) { | |
1873 | ||
1874 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1875 | ||
1876 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1877 | block_end_pfn, zone)) | |
1878 | return; | |
e84fe99b | 1879 | cond_resched(); |
7cf91a98 JK |
1880 | } |
1881 | ||
1882 | /* We confirm that there is no hole */ | |
1883 | zone->contiguous = true; | |
1884 | } | |
1885 | ||
1886 | void clear_zone_contiguous(struct zone *zone) | |
1887 | { | |
1888 | zone->contiguous = false; | |
1889 | } | |
1890 | ||
7e18adb4 | 1891 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1892 | static void __init deferred_free_range(unsigned long pfn, |
1893 | unsigned long nr_pages) | |
a4de83dd | 1894 | { |
2f47a91f PT |
1895 | struct page *page; |
1896 | unsigned long i; | |
a4de83dd | 1897 | |
2f47a91f | 1898 | if (!nr_pages) |
a4de83dd MG |
1899 | return; |
1900 | ||
2f47a91f PT |
1901 | page = pfn_to_page(pfn); |
1902 | ||
a4de83dd | 1903 | /* Free a large naturally-aligned chunk if possible */ |
ee0913c4 | 1904 | if (nr_pages == pageblock_nr_pages && pageblock_aligned(pfn)) { |
ac5d2539 | 1905 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1906 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1907 | return; |
1908 | } | |
1909 | ||
e780149b | 1910 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
ee0913c4 | 1911 | if (pageblock_aligned(pfn)) |
e780149b | 1912 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1913 | __free_pages_core(page, 0); |
e780149b | 1914 | } |
a4de83dd MG |
1915 | } |
1916 | ||
d3cd131d NS |
1917 | /* Completion tracking for deferred_init_memmap() threads */ |
1918 | static atomic_t pgdat_init_n_undone __initdata; | |
1919 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1920 | ||
1921 | static inline void __init pgdat_init_report_one_done(void) | |
1922 | { | |
1923 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1924 | complete(&pgdat_init_all_done_comp); | |
1925 | } | |
0e1cc95b | 1926 | |
2f47a91f | 1927 | /* |
80b1f41c PT |
1928 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1929 | * | |
c9b3637f | 1930 | * We check if a current large page is valid by only checking the validity |
80b1f41c | 1931 | * of the head pfn. |
2f47a91f | 1932 | */ |
56ec43d8 | 1933 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1934 | { |
ee0913c4 | 1935 | if (pageblock_aligned(pfn) && !pfn_valid(pfn)) |
80b1f41c | 1936 | return false; |
80b1f41c PT |
1937 | return true; |
1938 | } | |
2f47a91f | 1939 | |
80b1f41c PT |
1940 | /* |
1941 | * Free pages to buddy allocator. Try to free aligned pages in | |
1942 | * pageblock_nr_pages sizes. | |
1943 | */ | |
56ec43d8 | 1944 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1945 | unsigned long end_pfn) |
1946 | { | |
80b1f41c | 1947 | unsigned long nr_free = 0; |
2f47a91f | 1948 | |
80b1f41c | 1949 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1950 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1951 | deferred_free_range(pfn - nr_free, nr_free); |
1952 | nr_free = 0; | |
ee0913c4 | 1953 | } else if (pageblock_aligned(pfn)) { |
80b1f41c PT |
1954 | deferred_free_range(pfn - nr_free, nr_free); |
1955 | nr_free = 1; | |
80b1f41c PT |
1956 | } else { |
1957 | nr_free++; | |
1958 | } | |
1959 | } | |
1960 | /* Free the last block of pages to allocator */ | |
1961 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1962 | } |
1963 | ||
80b1f41c PT |
1964 | /* |
1965 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1966 | * by performing it only once every pageblock_nr_pages. | |
1967 | * Return number of pages initialized. | |
1968 | */ | |
56ec43d8 | 1969 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1970 | unsigned long pfn, |
1971 | unsigned long end_pfn) | |
2f47a91f | 1972 | { |
56ec43d8 | 1973 | int nid = zone_to_nid(zone); |
2f47a91f | 1974 | unsigned long nr_pages = 0; |
56ec43d8 | 1975 | int zid = zone_idx(zone); |
2f47a91f | 1976 | struct page *page = NULL; |
2f47a91f | 1977 | |
80b1f41c | 1978 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1979 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1980 | page = NULL; |
2f47a91f | 1981 | continue; |
ee0913c4 | 1982 | } else if (!page || pageblock_aligned(pfn)) { |
2f47a91f | 1983 | page = pfn_to_page(pfn); |
80b1f41c PT |
1984 | } else { |
1985 | page++; | |
2f47a91f | 1986 | } |
d0dc12e8 | 1987 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1988 | nr_pages++; |
2f47a91f | 1989 | } |
80b1f41c | 1990 | return (nr_pages); |
2f47a91f PT |
1991 | } |
1992 | ||
0e56acae AD |
1993 | /* |
1994 | * This function is meant to pre-load the iterator for the zone init. | |
1995 | * Specifically it walks through the ranges until we are caught up to the | |
1996 | * first_init_pfn value and exits there. If we never encounter the value we | |
1997 | * return false indicating there are no valid ranges left. | |
1998 | */ | |
1999 | static bool __init | |
2000 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
2001 | unsigned long *spfn, unsigned long *epfn, | |
2002 | unsigned long first_init_pfn) | |
2003 | { | |
2004 | u64 j; | |
2005 | ||
2006 | /* | |
2007 | * Start out by walking through the ranges in this zone that have | |
2008 | * already been initialized. We don't need to do anything with them | |
2009 | * so we just need to flush them out of the system. | |
2010 | */ | |
2011 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
2012 | if (*epfn <= first_init_pfn) | |
2013 | continue; | |
2014 | if (*spfn < first_init_pfn) | |
2015 | *spfn = first_init_pfn; | |
2016 | *i = j; | |
2017 | return true; | |
2018 | } | |
2019 | ||
2020 | return false; | |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * Initialize and free pages. We do it in two loops: first we initialize | |
2025 | * struct page, then free to buddy allocator, because while we are | |
2026 | * freeing pages we can access pages that are ahead (computing buddy | |
2027 | * page in __free_one_page()). | |
2028 | * | |
2029 | * In order to try and keep some memory in the cache we have the loop | |
2030 | * broken along max page order boundaries. This way we will not cause | |
2031 | * any issues with the buddy page computation. | |
2032 | */ | |
2033 | static unsigned long __init | |
2034 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
2035 | unsigned long *end_pfn) | |
2036 | { | |
2037 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
2038 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
2039 | unsigned long nr_pages = 0; | |
2040 | u64 j = *i; | |
2041 | ||
2042 | /* First we loop through and initialize the page values */ | |
2043 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
2044 | unsigned long t; | |
2045 | ||
2046 | if (mo_pfn <= *start_pfn) | |
2047 | break; | |
2048 | ||
2049 | t = min(mo_pfn, *end_pfn); | |
2050 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
2051 | ||
2052 | if (mo_pfn < *end_pfn) { | |
2053 | *start_pfn = mo_pfn; | |
2054 | break; | |
2055 | } | |
2056 | } | |
2057 | ||
2058 | /* Reset values and now loop through freeing pages as needed */ | |
2059 | swap(j, *i); | |
2060 | ||
2061 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
2062 | unsigned long t; | |
2063 | ||
2064 | if (mo_pfn <= spfn) | |
2065 | break; | |
2066 | ||
2067 | t = min(mo_pfn, epfn); | |
2068 | deferred_free_pages(spfn, t); | |
2069 | ||
2070 | if (mo_pfn <= epfn) | |
2071 | break; | |
2072 | } | |
2073 | ||
2074 | return nr_pages; | |
2075 | } | |
2076 | ||
e4443149 DJ |
2077 | static void __init |
2078 | deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, | |
2079 | void *arg) | |
2080 | { | |
2081 | unsigned long spfn, epfn; | |
2082 | struct zone *zone = arg; | |
2083 | u64 i; | |
2084 | ||
2085 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn); | |
2086 | ||
2087 | /* | |
2088 | * Initialize and free pages in MAX_ORDER sized increments so that we | |
2089 | * can avoid introducing any issues with the buddy allocator. | |
2090 | */ | |
2091 | while (spfn < end_pfn) { | |
2092 | deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
2093 | cond_resched(); | |
2094 | } | |
2095 | } | |
2096 | ||
ecd09650 DJ |
2097 | /* An arch may override for more concurrency. */ |
2098 | __weak int __init | |
2099 | deferred_page_init_max_threads(const struct cpumask *node_cpumask) | |
2100 | { | |
2101 | return 1; | |
2102 | } | |
2103 | ||
7e18adb4 | 2104 | /* Initialise remaining memory on a node */ |
0e1cc95b | 2105 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 2106 | { |
0e1cc95b | 2107 | pg_data_t *pgdat = data; |
0e56acae | 2108 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
89c7c402 | 2109 | unsigned long spfn = 0, epfn = 0; |
0e56acae | 2110 | unsigned long first_init_pfn, flags; |
7e18adb4 | 2111 | unsigned long start = jiffies; |
7e18adb4 | 2112 | struct zone *zone; |
e4443149 | 2113 | int zid, max_threads; |
2f47a91f | 2114 | u64 i; |
7e18adb4 | 2115 | |
3a2d7fa8 PT |
2116 | /* Bind memory initialisation thread to a local node if possible */ |
2117 | if (!cpumask_empty(cpumask)) | |
2118 | set_cpus_allowed_ptr(current, cpumask); | |
2119 | ||
2120 | pgdat_resize_lock(pgdat, &flags); | |
2121 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 2122 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 2123 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 2124 | pgdat_init_report_one_done(); |
0e1cc95b MG |
2125 | return 0; |
2126 | } | |
2127 | ||
7e18adb4 MG |
2128 | /* Sanity check boundaries */ |
2129 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
2130 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
2131 | pgdat->first_deferred_pfn = ULONG_MAX; | |
2132 | ||
3d060856 PT |
2133 | /* |
2134 | * Once we unlock here, the zone cannot be grown anymore, thus if an | |
2135 | * interrupt thread must allocate this early in boot, zone must be | |
2136 | * pre-grown prior to start of deferred page initialization. | |
2137 | */ | |
2138 | pgdat_resize_unlock(pgdat, &flags); | |
2139 | ||
7e18adb4 MG |
2140 | /* Only the highest zone is deferred so find it */ |
2141 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
2142 | zone = pgdat->node_zones + zid; | |
2143 | if (first_init_pfn < zone_end_pfn(zone)) | |
2144 | break; | |
2145 | } | |
0e56acae AD |
2146 | |
2147 | /* If the zone is empty somebody else may have cleared out the zone */ | |
2148 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2149 | first_init_pfn)) | |
2150 | goto zone_empty; | |
7e18adb4 | 2151 | |
ecd09650 | 2152 | max_threads = deferred_page_init_max_threads(cpumask); |
7e18adb4 | 2153 | |
117003c3 | 2154 | while (spfn < epfn) { |
e4443149 DJ |
2155 | unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION); |
2156 | struct padata_mt_job job = { | |
2157 | .thread_fn = deferred_init_memmap_chunk, | |
2158 | .fn_arg = zone, | |
2159 | .start = spfn, | |
2160 | .size = epfn_align - spfn, | |
2161 | .align = PAGES_PER_SECTION, | |
2162 | .min_chunk = PAGES_PER_SECTION, | |
2163 | .max_threads = max_threads, | |
2164 | }; | |
2165 | ||
2166 | padata_do_multithreaded(&job); | |
2167 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2168 | epfn_align); | |
117003c3 | 2169 | } |
0e56acae | 2170 | zone_empty: |
7e18adb4 MG |
2171 | /* Sanity check that the next zone really is unpopulated */ |
2172 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
2173 | ||
89c7c402 DJ |
2174 | pr_info("node %d deferred pages initialised in %ums\n", |
2175 | pgdat->node_id, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
2176 | |
2177 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
2178 | return 0; |
2179 | } | |
c9e97a19 | 2180 | |
c9e97a19 PT |
2181 | /* |
2182 | * If this zone has deferred pages, try to grow it by initializing enough | |
2183 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
2184 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
2185 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
2186 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
2187 | * | |
2188 | * Return true when zone was grown, otherwise return false. We return true even | |
2189 | * when we grow less than requested, to let the caller decide if there are | |
2190 | * enough pages to satisfy the allocation. | |
2191 | * | |
2192 | * Note: We use noinline because this function is needed only during boot, and | |
2193 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
2194 | * making sure that it is not inlined into permanent text section. | |
2195 | */ | |
2196 | static noinline bool __init | |
2197 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
2198 | { | |
c9e97a19 | 2199 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 2200 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 2201 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
2202 | unsigned long spfn, epfn, flags; |
2203 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
2204 | u64 i; |
2205 | ||
2206 | /* Only the last zone may have deferred pages */ | |
2207 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
2208 | return false; | |
2209 | ||
2210 | pgdat_resize_lock(pgdat, &flags); | |
2211 | ||
c9e97a19 PT |
2212 | /* |
2213 | * If someone grew this zone while we were waiting for spinlock, return | |
2214 | * true, as there might be enough pages already. | |
2215 | */ | |
2216 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
2217 | pgdat_resize_unlock(pgdat, &flags); | |
2218 | return true; | |
2219 | } | |
2220 | ||
0e56acae AD |
2221 | /* If the zone is empty somebody else may have cleared out the zone */ |
2222 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2223 | first_deferred_pfn)) { | |
2224 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 2225 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
2226 | /* Retry only once. */ |
2227 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
2228 | } |
2229 | ||
0e56acae AD |
2230 | /* |
2231 | * Initialize and free pages in MAX_ORDER sized increments so | |
2232 | * that we can avoid introducing any issues with the buddy | |
2233 | * allocator. | |
2234 | */ | |
2235 | while (spfn < epfn) { | |
2236 | /* update our first deferred PFN for this section */ | |
2237 | first_deferred_pfn = spfn; | |
2238 | ||
2239 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
117003c3 | 2240 | touch_nmi_watchdog(); |
c9e97a19 | 2241 | |
0e56acae AD |
2242 | /* We should only stop along section boundaries */ |
2243 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
2244 | continue; | |
c9e97a19 | 2245 | |
0e56acae | 2246 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
2247 | if (nr_pages >= nr_pages_needed) |
2248 | break; | |
2249 | } | |
2250 | ||
0e56acae | 2251 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
2252 | pgdat_resize_unlock(pgdat, &flags); |
2253 | ||
2254 | return nr_pages > 0; | |
2255 | } | |
2256 | ||
2257 | /* | |
2258 | * deferred_grow_zone() is __init, but it is called from | |
2259 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
2260 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
2261 | * and to ensure that the function body gets unloaded. | |
2262 | */ | |
2263 | static bool __ref | |
2264 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
2265 | { | |
2266 | return deferred_grow_zone(zone, order); | |
2267 | } | |
2268 | ||
7cf91a98 | 2269 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
2270 | |
2271 | void __init page_alloc_init_late(void) | |
2272 | { | |
7cf91a98 | 2273 | struct zone *zone; |
e900a918 | 2274 | int nid; |
7cf91a98 JK |
2275 | |
2276 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 2277 | |
d3cd131d NS |
2278 | /* There will be num_node_state(N_MEMORY) threads */ |
2279 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 2280 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
2281 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
2282 | } | |
2283 | ||
2284 | /* Block until all are initialised */ | |
d3cd131d | 2285 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 2286 | |
c9e97a19 PT |
2287 | /* |
2288 | * We initialized the rest of the deferred pages. Permanently disable | |
2289 | * on-demand struct page initialization. | |
2290 | */ | |
2291 | static_branch_disable(&deferred_pages); | |
2292 | ||
4248b0da MG |
2293 | /* Reinit limits that are based on free pages after the kernel is up */ |
2294 | files_maxfiles_init(); | |
7cf91a98 | 2295 | #endif |
350e88ba | 2296 | |
ba8f3587 LF |
2297 | buffer_init(); |
2298 | ||
3010f876 PT |
2299 | /* Discard memblock private memory */ |
2300 | memblock_discard(); | |
7cf91a98 | 2301 | |
e900a918 DW |
2302 | for_each_node_state(nid, N_MEMORY) |
2303 | shuffle_free_memory(NODE_DATA(nid)); | |
2304 | ||
7cf91a98 JK |
2305 | for_each_populated_zone(zone) |
2306 | set_zone_contiguous(zone); | |
7e18adb4 | 2307 | } |
7e18adb4 | 2308 | |
47118af0 | 2309 | #ifdef CONFIG_CMA |
9cf510a5 | 2310 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
2311 | void __init init_cma_reserved_pageblock(struct page *page) |
2312 | { | |
2313 | unsigned i = pageblock_nr_pages; | |
2314 | struct page *p = page; | |
2315 | ||
2316 | do { | |
2317 | __ClearPageReserved(p); | |
2318 | set_page_count(p, 0); | |
d883c6cf | 2319 | } while (++p, --i); |
47118af0 | 2320 | |
47118af0 | 2321 | set_pageblock_migratetype(page, MIGRATE_CMA); |
b3d40a2b DH |
2322 | set_page_refcounted(page); |
2323 | __free_pages(page, pageblock_order); | |
dc78327c | 2324 | |
3dcc0571 | 2325 | adjust_managed_page_count(page, pageblock_nr_pages); |
3c381db1 | 2326 | page_zone(page)->cma_pages += pageblock_nr_pages; |
47118af0 MN |
2327 | } |
2328 | #endif | |
1da177e4 LT |
2329 | |
2330 | /* | |
2331 | * The order of subdivision here is critical for the IO subsystem. | |
2332 | * Please do not alter this order without good reasons and regression | |
2333 | * testing. Specifically, as large blocks of memory are subdivided, | |
2334 | * the order in which smaller blocks are delivered depends on the order | |
2335 | * they're subdivided in this function. This is the primary factor | |
2336 | * influencing the order in which pages are delivered to the IO | |
2337 | * subsystem according to empirical testing, and this is also justified | |
2338 | * by considering the behavior of a buddy system containing a single | |
2339 | * large block of memory acted on by a series of small allocations. | |
2340 | * This behavior is a critical factor in sglist merging's success. | |
2341 | * | |
6d49e352 | 2342 | * -- nyc |
1da177e4 | 2343 | */ |
085cc7d5 | 2344 | static inline void expand(struct zone *zone, struct page *page, |
6ab01363 | 2345 | int low, int high, int migratetype) |
1da177e4 LT |
2346 | { |
2347 | unsigned long size = 1 << high; | |
2348 | ||
2349 | while (high > low) { | |
1da177e4 LT |
2350 | high--; |
2351 | size >>= 1; | |
309381fe | 2352 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2353 | |
acbc15a4 JK |
2354 | /* |
2355 | * Mark as guard pages (or page), that will allow to | |
2356 | * merge back to allocator when buddy will be freed. | |
2357 | * Corresponding page table entries will not be touched, | |
2358 | * pages will stay not present in virtual address space | |
2359 | */ | |
2360 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2361 | continue; |
acbc15a4 | 2362 | |
6ab01363 | 2363 | add_to_free_list(&page[size], zone, high, migratetype); |
ab130f91 | 2364 | set_buddy_order(&page[size], high); |
1da177e4 | 2365 | } |
1da177e4 LT |
2366 | } |
2367 | ||
4e611801 | 2368 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2369 | { |
f4c18e6f | 2370 | if (unlikely(page->flags & __PG_HWPOISON)) { |
e570f56c NH |
2371 | /* Don't complain about hwpoisoned pages */ |
2372 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2373 | return; | |
f4c18e6f | 2374 | } |
58b7f119 WY |
2375 | |
2376 | bad_page(page, | |
2377 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
4e611801 VB |
2378 | } |
2379 | ||
2380 | /* | |
2381 | * This page is about to be returned from the page allocator | |
2382 | */ | |
2383 | static inline int check_new_page(struct page *page) | |
2384 | { | |
2385 | if (likely(page_expected_state(page, | |
2386 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2387 | return 0; | |
2388 | ||
2389 | check_new_page_bad(page); | |
2390 | return 1; | |
2a7684a2 WF |
2391 | } |
2392 | ||
77fe7f13 MG |
2393 | static bool check_new_pages(struct page *page, unsigned int order) |
2394 | { | |
2395 | int i; | |
2396 | for (i = 0; i < (1 << order); i++) { | |
2397 | struct page *p = page + i; | |
2398 | ||
2399 | if (unlikely(check_new_page(p))) | |
2400 | return true; | |
2401 | } | |
2402 | ||
2403 | return false; | |
2404 | } | |
2405 | ||
479f854a | 2406 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2407 | /* |
2408 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2409 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2410 | * also checked when pcp lists are refilled from the free lists. | |
2411 | */ | |
77fe7f13 | 2412 | static inline bool check_pcp_refill(struct page *page, unsigned int order) |
479f854a | 2413 | { |
8e57f8ac | 2414 | if (debug_pagealloc_enabled_static()) |
77fe7f13 | 2415 | return check_new_pages(page, order); |
4462b32c VB |
2416 | else |
2417 | return false; | |
479f854a MG |
2418 | } |
2419 | ||
77fe7f13 | 2420 | static inline bool check_new_pcp(struct page *page, unsigned int order) |
479f854a | 2421 | { |
77fe7f13 | 2422 | return check_new_pages(page, order); |
479f854a MG |
2423 | } |
2424 | #else | |
4462b32c VB |
2425 | /* |
2426 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2427 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2428 | * enabled, they are also checked when being allocated from the pcp lists. | |
2429 | */ | |
77fe7f13 | 2430 | static inline bool check_pcp_refill(struct page *page, unsigned int order) |
479f854a | 2431 | { |
77fe7f13 | 2432 | return check_new_pages(page, order); |
479f854a | 2433 | } |
77fe7f13 | 2434 | static inline bool check_new_pcp(struct page *page, unsigned int order) |
479f854a | 2435 | { |
8e57f8ac | 2436 | if (debug_pagealloc_enabled_static()) |
77fe7f13 | 2437 | return check_new_pages(page, order); |
4462b32c VB |
2438 | else |
2439 | return false; | |
479f854a MG |
2440 | } |
2441 | #endif /* CONFIG_DEBUG_VM */ | |
2442 | ||
6d05141a | 2443 | static inline bool should_skip_kasan_unpoison(gfp_t flags) |
53ae233c AK |
2444 | { |
2445 | /* Don't skip if a software KASAN mode is enabled. */ | |
2446 | if (IS_ENABLED(CONFIG_KASAN_GENERIC) || | |
2447 | IS_ENABLED(CONFIG_KASAN_SW_TAGS)) | |
2448 | return false; | |
2449 | ||
2450 | /* Skip, if hardware tag-based KASAN is not enabled. */ | |
2451 | if (!kasan_hw_tags_enabled()) | |
2452 | return true; | |
2453 | ||
2454 | /* | |
6d05141a CM |
2455 | * With hardware tag-based KASAN enabled, skip if this has been |
2456 | * requested via __GFP_SKIP_KASAN_UNPOISON. | |
53ae233c | 2457 | */ |
6d05141a | 2458 | return flags & __GFP_SKIP_KASAN_UNPOISON; |
53ae233c AK |
2459 | } |
2460 | ||
9353ffa6 AK |
2461 | static inline bool should_skip_init(gfp_t flags) |
2462 | { | |
2463 | /* Don't skip, if hardware tag-based KASAN is not enabled. */ | |
2464 | if (!kasan_hw_tags_enabled()) | |
2465 | return false; | |
2466 | ||
2467 | /* For hardware tag-based KASAN, skip if requested. */ | |
2468 | return (flags & __GFP_SKIP_ZERO); | |
2469 | } | |
2470 | ||
46f24fd8 JK |
2471 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2472 | gfp_t gfp_flags) | |
2473 | { | |
9353ffa6 AK |
2474 | bool init = !want_init_on_free() && want_init_on_alloc(gfp_flags) && |
2475 | !should_skip_init(gfp_flags); | |
44383cef AK |
2476 | bool zero_tags = init && (gfp_flags & __GFP_ZEROTAGS); |
2477 | bool reset_tags = !zero_tags; | |
70c248ac | 2478 | int i; |
b42090ae | 2479 | |
46f24fd8 JK |
2480 | set_page_private(page, 0); |
2481 | set_page_refcounted(page); | |
2482 | ||
2483 | arch_alloc_page(page, order); | |
77bc7fd6 | 2484 | debug_pagealloc_map_pages(page, 1 << order); |
1bb5eab3 AK |
2485 | |
2486 | /* | |
2487 | * Page unpoisoning must happen before memory initialization. | |
2488 | * Otherwise, the poison pattern will be overwritten for __GFP_ZERO | |
2489 | * allocations and the page unpoisoning code will complain. | |
2490 | */ | |
8db26a3d | 2491 | kernel_unpoison_pages(page, 1 << order); |
862b6dee | 2492 | |
1bb5eab3 AK |
2493 | /* |
2494 | * As memory initialization might be integrated into KASAN, | |
b42090ae | 2495 | * KASAN unpoisoning and memory initializion code must be |
1bb5eab3 AK |
2496 | * kept together to avoid discrepancies in behavior. |
2497 | */ | |
9294b128 AK |
2498 | |
2499 | /* | |
44383cef AK |
2500 | * If memory tags should be zeroed |
2501 | * (which happens only when memory should be initialized as well). | |
9294b128 | 2502 | */ |
44383cef | 2503 | if (zero_tags) { |
9294b128 AK |
2504 | /* Initialize both memory and tags. */ |
2505 | for (i = 0; i != 1 << order; ++i) | |
2506 | tag_clear_highpage(page + i); | |
2507 | ||
44383cef | 2508 | /* Take note that memory was initialized by the loop above. */ |
9294b128 AK |
2509 | init = false; |
2510 | } | |
6d05141a | 2511 | if (!should_skip_kasan_unpoison(gfp_flags)) { |
44383cef AK |
2512 | /* Try unpoisoning (or setting tags) and initializing memory. */ |
2513 | if (kasan_unpoison_pages(page, order, init)) { | |
2514 | /* Take note that memory was initialized by KASAN. */ | |
2515 | if (kasan_has_integrated_init()) | |
2516 | init = false; | |
2517 | /* Take note that memory tags were set by KASAN. */ | |
2518 | reset_tags = false; | |
2519 | } else { | |
2520 | /* | |
2521 | * KASAN decided to exclude this allocation from being | |
2522 | * poisoned due to sampling. Skip poisoning as well. | |
2523 | */ | |
2524 | SetPageSkipKASanPoison(page); | |
2525 | } | |
2526 | } | |
2527 | /* | |
2528 | * If memory tags have not been set, reset the page tags to ensure | |
2529 | * page_address() dereferencing does not fault. | |
2530 | */ | |
2531 | if (reset_tags) { | |
70c248ac CM |
2532 | for (i = 0; i != 1 << order; ++i) |
2533 | page_kasan_tag_reset(page + i); | |
7a3b8353 | 2534 | } |
44383cef | 2535 | /* If memory is still not initialized, initialize it now. */ |
7e3cbba6 | 2536 | if (init) |
aeaec8e2 | 2537 | kernel_init_pages(page, 1 << order); |
89b27116 AK |
2538 | /* Propagate __GFP_SKIP_KASAN_POISON to page flags. */ |
2539 | if (kasan_hw_tags_enabled() && (gfp_flags & __GFP_SKIP_KASAN_POISON)) | |
2540 | SetPageSkipKASanPoison(page); | |
1bb5eab3 AK |
2541 | |
2542 | set_page_owner(page, order, gfp_flags); | |
df4e817b | 2543 | page_table_check_alloc(page, order); |
46f24fd8 JK |
2544 | } |
2545 | ||
479f854a | 2546 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2547 | unsigned int alloc_flags) |
2a7684a2 | 2548 | { |
46f24fd8 | 2549 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2550 | |
17cf4406 NP |
2551 | if (order && (gfp_flags & __GFP_COMP)) |
2552 | prep_compound_page(page, order); | |
2553 | ||
75379191 | 2554 | /* |
2f064f34 | 2555 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2556 | * allocate the page. The expectation is that the caller is taking |
2557 | * steps that will free more memory. The caller should avoid the page | |
2558 | * being used for !PFMEMALLOC purposes. | |
2559 | */ | |
2f064f34 MH |
2560 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2561 | set_page_pfmemalloc(page); | |
2562 | else | |
2563 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2564 | } |
2565 | ||
56fd56b8 MG |
2566 | /* |
2567 | * Go through the free lists for the given migratetype and remove | |
2568 | * the smallest available page from the freelists | |
2569 | */ | |
85ccc8fa | 2570 | static __always_inline |
728ec980 | 2571 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2572 | int migratetype) |
2573 | { | |
2574 | unsigned int current_order; | |
b8af2941 | 2575 | struct free_area *area; |
56fd56b8 MG |
2576 | struct page *page; |
2577 | ||
2578 | /* Find a page of the appropriate size in the preferred list */ | |
2579 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2580 | area = &(zone->free_area[current_order]); | |
b03641af | 2581 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2582 | if (!page) |
2583 | continue; | |
6ab01363 AD |
2584 | del_page_from_free_list(page, zone, current_order); |
2585 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 2586 | set_pcppage_migratetype(page, migratetype); |
10e0f753 WY |
2587 | trace_mm_page_alloc_zone_locked(page, order, migratetype, |
2588 | pcp_allowed_order(order) && | |
2589 | migratetype < MIGRATE_PCPTYPES); | |
56fd56b8 MG |
2590 | return page; |
2591 | } | |
2592 | ||
2593 | return NULL; | |
2594 | } | |
2595 | ||
2596 | ||
b2a0ac88 MG |
2597 | /* |
2598 | * This array describes the order lists are fallen back to when | |
2599 | * the free lists for the desirable migrate type are depleted | |
1dd214b8 ZY |
2600 | * |
2601 | * The other migratetypes do not have fallbacks. | |
b2a0ac88 | 2602 | */ |
da415663 | 2603 | static int fallbacks[MIGRATE_TYPES][3] = { |
974a786e | 2604 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2605 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2606 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
b2a0ac88 MG |
2607 | }; |
2608 | ||
dc67647b | 2609 | #ifdef CONFIG_CMA |
85ccc8fa | 2610 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2611 | unsigned int order) |
2612 | { | |
2613 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2614 | } | |
2615 | #else | |
2616 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2617 | unsigned int order) { return NULL; } | |
2618 | #endif | |
2619 | ||
c361be55 | 2620 | /* |
293ffa5e | 2621 | * Move the free pages in a range to the freelist tail of the requested type. |
d9c23400 | 2622 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2623 | * boundary. If alignment is required, use move_freepages_block() |
2624 | */ | |
02aa0cdd | 2625 | static int move_freepages(struct zone *zone, |
39ddb991 | 2626 | unsigned long start_pfn, unsigned long end_pfn, |
02aa0cdd | 2627 | int migratetype, int *num_movable) |
c361be55 MG |
2628 | { |
2629 | struct page *page; | |
39ddb991 | 2630 | unsigned long pfn; |
d00181b9 | 2631 | unsigned int order; |
d100313f | 2632 | int pages_moved = 0; |
c361be55 | 2633 | |
39ddb991 | 2634 | for (pfn = start_pfn; pfn <= end_pfn;) { |
39ddb991 | 2635 | page = pfn_to_page(pfn); |
c361be55 | 2636 | if (!PageBuddy(page)) { |
02aa0cdd VB |
2637 | /* |
2638 | * We assume that pages that could be isolated for | |
2639 | * migration are movable. But we don't actually try | |
2640 | * isolating, as that would be expensive. | |
2641 | */ | |
2642 | if (num_movable && | |
2643 | (PageLRU(page) || __PageMovable(page))) | |
2644 | (*num_movable)++; | |
39ddb991 | 2645 | pfn++; |
c361be55 MG |
2646 | continue; |
2647 | } | |
2648 | ||
cd961038 DR |
2649 | /* Make sure we are not inadvertently changing nodes */ |
2650 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2651 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2652 | ||
ab130f91 | 2653 | order = buddy_order(page); |
6ab01363 | 2654 | move_to_free_list(page, zone, order, migratetype); |
39ddb991 | 2655 | pfn += 1 << order; |
d100313f | 2656 | pages_moved += 1 << order; |
c361be55 MG |
2657 | } |
2658 | ||
d100313f | 2659 | return pages_moved; |
c361be55 MG |
2660 | } |
2661 | ||
ee6f509c | 2662 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2663 | int migratetype, int *num_movable) |
c361be55 | 2664 | { |
39ddb991 | 2665 | unsigned long start_pfn, end_pfn, pfn; |
c361be55 | 2666 | |
4a222127 DR |
2667 | if (num_movable) |
2668 | *num_movable = 0; | |
2669 | ||
39ddb991 | 2670 | pfn = page_to_pfn(page); |
4f9bc69a KW |
2671 | start_pfn = pageblock_start_pfn(pfn); |
2672 | end_pfn = pageblock_end_pfn(pfn) - 1; | |
c361be55 MG |
2673 | |
2674 | /* Do not cross zone boundaries */ | |
108bcc96 | 2675 | if (!zone_spans_pfn(zone, start_pfn)) |
39ddb991 | 2676 | start_pfn = pfn; |
108bcc96 | 2677 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2678 | return 0; |
2679 | ||
39ddb991 | 2680 | return move_freepages(zone, start_pfn, end_pfn, migratetype, |
02aa0cdd | 2681 | num_movable); |
c361be55 MG |
2682 | } |
2683 | ||
2f66a68f MG |
2684 | static void change_pageblock_range(struct page *pageblock_page, |
2685 | int start_order, int migratetype) | |
2686 | { | |
2687 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2688 | ||
2689 | while (nr_pageblocks--) { | |
2690 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2691 | pageblock_page += pageblock_nr_pages; | |
2692 | } | |
2693 | } | |
2694 | ||
fef903ef | 2695 | /* |
9c0415eb VB |
2696 | * When we are falling back to another migratetype during allocation, try to |
2697 | * steal extra free pages from the same pageblocks to satisfy further | |
2698 | * allocations, instead of polluting multiple pageblocks. | |
2699 | * | |
2700 | * If we are stealing a relatively large buddy page, it is likely there will | |
2701 | * be more free pages in the pageblock, so try to steal them all. For | |
2702 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2703 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2704 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2705 | * pageblocks. | |
fef903ef | 2706 | */ |
4eb7dce6 JK |
2707 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2708 | { | |
2709 | /* | |
2710 | * Leaving this order check is intended, although there is | |
2711 | * relaxed order check in next check. The reason is that | |
2712 | * we can actually steal whole pageblock if this condition met, | |
2713 | * but, below check doesn't guarantee it and that is just heuristic | |
2714 | * so could be changed anytime. | |
2715 | */ | |
2716 | if (order >= pageblock_order) | |
2717 | return true; | |
2718 | ||
2719 | if (order >= pageblock_order / 2 || | |
2720 | start_mt == MIGRATE_RECLAIMABLE || | |
2721 | start_mt == MIGRATE_UNMOVABLE || | |
2722 | page_group_by_mobility_disabled) | |
2723 | return true; | |
2724 | ||
2725 | return false; | |
2726 | } | |
2727 | ||
597c8920 | 2728 | static inline bool boost_watermark(struct zone *zone) |
1c30844d MG |
2729 | { |
2730 | unsigned long max_boost; | |
2731 | ||
2732 | if (!watermark_boost_factor) | |
597c8920 | 2733 | return false; |
14f69140 HW |
2734 | /* |
2735 | * Don't bother in zones that are unlikely to produce results. | |
2736 | * On small machines, including kdump capture kernels running | |
2737 | * in a small area, boosting the watermark can cause an out of | |
2738 | * memory situation immediately. | |
2739 | */ | |
2740 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
597c8920 | 2741 | return false; |
1c30844d MG |
2742 | |
2743 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2744 | watermark_boost_factor, 10000); | |
94b3334c MG |
2745 | |
2746 | /* | |
2747 | * high watermark may be uninitialised if fragmentation occurs | |
2748 | * very early in boot so do not boost. We do not fall | |
2749 | * through and boost by pageblock_nr_pages as failing | |
2750 | * allocations that early means that reclaim is not going | |
2751 | * to help and it may even be impossible to reclaim the | |
2752 | * boosted watermark resulting in a hang. | |
2753 | */ | |
2754 | if (!max_boost) | |
597c8920 | 2755 | return false; |
94b3334c | 2756 | |
1c30844d MG |
2757 | max_boost = max(pageblock_nr_pages, max_boost); |
2758 | ||
2759 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2760 | max_boost); | |
597c8920 JW |
2761 | |
2762 | return true; | |
1c30844d MG |
2763 | } |
2764 | ||
4eb7dce6 JK |
2765 | /* |
2766 | * This function implements actual steal behaviour. If order is large enough, | |
2767 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2768 | * pageblock to our migratetype and determine how many already-allocated pages |
2769 | * are there in the pageblock with a compatible migratetype. If at least half | |
2770 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2771 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2772 | */ |
2773 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2774 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2775 | { |
ab130f91 | 2776 | unsigned int current_order = buddy_order(page); |
02aa0cdd VB |
2777 | int free_pages, movable_pages, alike_pages; |
2778 | int old_block_type; | |
2779 | ||
2780 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2781 | |
3bc48f96 VB |
2782 | /* |
2783 | * This can happen due to races and we want to prevent broken | |
2784 | * highatomic accounting. | |
2785 | */ | |
02aa0cdd | 2786 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2787 | goto single_page; |
2788 | ||
fef903ef SB |
2789 | /* Take ownership for orders >= pageblock_order */ |
2790 | if (current_order >= pageblock_order) { | |
2791 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2792 | goto single_page; |
fef903ef SB |
2793 | } |
2794 | ||
1c30844d MG |
2795 | /* |
2796 | * Boost watermarks to increase reclaim pressure to reduce the | |
2797 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2798 | * may be balanced overall and kswapd will not wake naturally. | |
2799 | */ | |
597c8920 | 2800 | if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD)) |
73444bc4 | 2801 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2802 | |
3bc48f96 VB |
2803 | /* We are not allowed to try stealing from the whole block */ |
2804 | if (!whole_block) | |
2805 | goto single_page; | |
2806 | ||
02aa0cdd VB |
2807 | free_pages = move_freepages_block(zone, page, start_type, |
2808 | &movable_pages); | |
2809 | /* | |
2810 | * Determine how many pages are compatible with our allocation. | |
2811 | * For movable allocation, it's the number of movable pages which | |
2812 | * we just obtained. For other types it's a bit more tricky. | |
2813 | */ | |
2814 | if (start_type == MIGRATE_MOVABLE) { | |
2815 | alike_pages = movable_pages; | |
2816 | } else { | |
2817 | /* | |
2818 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2819 | * to MOVABLE pageblock, consider all non-movable pages as | |
2820 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2821 | * vice versa, be conservative since we can't distinguish the | |
2822 | * exact migratetype of non-movable pages. | |
2823 | */ | |
2824 | if (old_block_type == MIGRATE_MOVABLE) | |
2825 | alike_pages = pageblock_nr_pages | |
2826 | - (free_pages + movable_pages); | |
2827 | else | |
2828 | alike_pages = 0; | |
2829 | } | |
2830 | ||
3bc48f96 | 2831 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2832 | if (!free_pages) |
3bc48f96 | 2833 | goto single_page; |
fef903ef | 2834 | |
02aa0cdd VB |
2835 | /* |
2836 | * If a sufficient number of pages in the block are either free or of | |
2837 | * comparable migratability as our allocation, claim the whole block. | |
2838 | */ | |
2839 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2840 | page_group_by_mobility_disabled) |
2841 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2842 | |
2843 | return; | |
2844 | ||
2845 | single_page: | |
6ab01363 | 2846 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
2847 | } |
2848 | ||
2149cdae JK |
2849 | /* |
2850 | * Check whether there is a suitable fallback freepage with requested order. | |
2851 | * If only_stealable is true, this function returns fallback_mt only if | |
2852 | * we can steal other freepages all together. This would help to reduce | |
2853 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2854 | */ | |
2855 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2856 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2857 | { |
2858 | int i; | |
2859 | int fallback_mt; | |
2860 | ||
2861 | if (area->nr_free == 0) | |
2862 | return -1; | |
2863 | ||
2864 | *can_steal = false; | |
2865 | for (i = 0;; i++) { | |
2866 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2867 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2868 | break; |
2869 | ||
b03641af | 2870 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2871 | continue; |
fef903ef | 2872 | |
4eb7dce6 JK |
2873 | if (can_steal_fallback(order, migratetype)) |
2874 | *can_steal = true; | |
2875 | ||
2149cdae JK |
2876 | if (!only_stealable) |
2877 | return fallback_mt; | |
2878 | ||
2879 | if (*can_steal) | |
2880 | return fallback_mt; | |
fef903ef | 2881 | } |
4eb7dce6 JK |
2882 | |
2883 | return -1; | |
fef903ef SB |
2884 | } |
2885 | ||
0aaa29a5 MG |
2886 | /* |
2887 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2888 | * there are no empty page blocks that contain a page with a suitable order | |
2889 | */ | |
2890 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2891 | unsigned int alloc_order) | |
2892 | { | |
2893 | int mt; | |
2894 | unsigned long max_managed, flags; | |
2895 | ||
2896 | /* | |
2897 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2898 | * Check is race-prone but harmless. | |
2899 | */ | |
9705bea5 | 2900 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2901 | if (zone->nr_reserved_highatomic >= max_managed) |
2902 | return; | |
2903 | ||
2904 | spin_lock_irqsave(&zone->lock, flags); | |
2905 | ||
2906 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2907 | if (zone->nr_reserved_highatomic >= max_managed) | |
2908 | goto out_unlock; | |
2909 | ||
2910 | /* Yoink! */ | |
2911 | mt = get_pageblock_migratetype(page); | |
1dd214b8 ZY |
2912 | /* Only reserve normal pageblocks (i.e., they can merge with others) */ |
2913 | if (migratetype_is_mergeable(mt)) { | |
0aaa29a5 MG |
2914 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2915 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2916 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2917 | } |
2918 | ||
2919 | out_unlock: | |
2920 | spin_unlock_irqrestore(&zone->lock, flags); | |
2921 | } | |
2922 | ||
2923 | /* | |
2924 | * Used when an allocation is about to fail under memory pressure. This | |
2925 | * potentially hurts the reliability of high-order allocations when under | |
2926 | * intense memory pressure but failed atomic allocations should be easier | |
2927 | * to recover from than an OOM. | |
29fac03b MK |
2928 | * |
2929 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2930 | * pageblock is exhausted. | |
0aaa29a5 | 2931 | */ |
29fac03b MK |
2932 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2933 | bool force) | |
0aaa29a5 MG |
2934 | { |
2935 | struct zonelist *zonelist = ac->zonelist; | |
2936 | unsigned long flags; | |
2937 | struct zoneref *z; | |
2938 | struct zone *zone; | |
2939 | struct page *page; | |
2940 | int order; | |
04c8716f | 2941 | bool ret; |
0aaa29a5 | 2942 | |
97a225e6 | 2943 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2944 | ac->nodemask) { |
29fac03b MK |
2945 | /* |
2946 | * Preserve at least one pageblock unless memory pressure | |
2947 | * is really high. | |
2948 | */ | |
2949 | if (!force && zone->nr_reserved_highatomic <= | |
2950 | pageblock_nr_pages) | |
0aaa29a5 MG |
2951 | continue; |
2952 | ||
2953 | spin_lock_irqsave(&zone->lock, flags); | |
2954 | for (order = 0; order < MAX_ORDER; order++) { | |
2955 | struct free_area *area = &(zone->free_area[order]); | |
2956 | ||
b03641af | 2957 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2958 | if (!page) |
0aaa29a5 MG |
2959 | continue; |
2960 | ||
0aaa29a5 | 2961 | /* |
4855e4a7 MK |
2962 | * In page freeing path, migratetype change is racy so |
2963 | * we can counter several free pages in a pageblock | |
f0953a1b | 2964 | * in this loop although we changed the pageblock type |
4855e4a7 MK |
2965 | * from highatomic to ac->migratetype. So we should |
2966 | * adjust the count once. | |
0aaa29a5 | 2967 | */ |
a6ffdc07 | 2968 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2969 | /* |
2970 | * It should never happen but changes to | |
2971 | * locking could inadvertently allow a per-cpu | |
2972 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2973 | * while unreserving so be safe and watch for | |
2974 | * underflows. | |
2975 | */ | |
2976 | zone->nr_reserved_highatomic -= min( | |
2977 | pageblock_nr_pages, | |
2978 | zone->nr_reserved_highatomic); | |
2979 | } | |
0aaa29a5 MG |
2980 | |
2981 | /* | |
2982 | * Convert to ac->migratetype and avoid the normal | |
2983 | * pageblock stealing heuristics. Minimally, the caller | |
2984 | * is doing the work and needs the pages. More | |
2985 | * importantly, if the block was always converted to | |
2986 | * MIGRATE_UNMOVABLE or another type then the number | |
2987 | * of pageblocks that cannot be completely freed | |
2988 | * may increase. | |
2989 | */ | |
2990 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2991 | ret = move_freepages_block(zone, page, ac->migratetype, |
2992 | NULL); | |
29fac03b MK |
2993 | if (ret) { |
2994 | spin_unlock_irqrestore(&zone->lock, flags); | |
2995 | return ret; | |
2996 | } | |
0aaa29a5 MG |
2997 | } |
2998 | spin_unlock_irqrestore(&zone->lock, flags); | |
2999 | } | |
04c8716f MK |
3000 | |
3001 | return false; | |
0aaa29a5 MG |
3002 | } |
3003 | ||
3bc48f96 VB |
3004 | /* |
3005 | * Try finding a free buddy page on the fallback list and put it on the free | |
3006 | * list of requested migratetype, possibly along with other pages from the same | |
3007 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
3008 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
3009 | * |
3010 | * The use of signed ints for order and current_order is a deliberate | |
3011 | * deviation from the rest of this file, to make the for loop | |
3012 | * condition simpler. | |
3bc48f96 | 3013 | */ |
85ccc8fa | 3014 | static __always_inline bool |
6bb15450 MG |
3015 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
3016 | unsigned int alloc_flags) | |
b2a0ac88 | 3017 | { |
b8af2941 | 3018 | struct free_area *area; |
b002529d | 3019 | int current_order; |
6bb15450 | 3020 | int min_order = order; |
b2a0ac88 | 3021 | struct page *page; |
4eb7dce6 JK |
3022 | int fallback_mt; |
3023 | bool can_steal; | |
b2a0ac88 | 3024 | |
6bb15450 MG |
3025 | /* |
3026 | * Do not steal pages from freelists belonging to other pageblocks | |
3027 | * i.e. orders < pageblock_order. If there are no local zones free, | |
3028 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
3029 | */ | |
e933dc4a | 3030 | if (order < pageblock_order && alloc_flags & ALLOC_NOFRAGMENT) |
6bb15450 MG |
3031 | min_order = pageblock_order; |
3032 | ||
7a8f58f3 VB |
3033 | /* |
3034 | * Find the largest available free page in the other list. This roughly | |
3035 | * approximates finding the pageblock with the most free pages, which | |
3036 | * would be too costly to do exactly. | |
3037 | */ | |
6bb15450 | 3038 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 3039 | --current_order) { |
4eb7dce6 JK |
3040 | area = &(zone->free_area[current_order]); |
3041 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 3042 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
3043 | if (fallback_mt == -1) |
3044 | continue; | |
b2a0ac88 | 3045 | |
7a8f58f3 VB |
3046 | /* |
3047 | * We cannot steal all free pages from the pageblock and the | |
3048 | * requested migratetype is movable. In that case it's better to | |
3049 | * steal and split the smallest available page instead of the | |
3050 | * largest available page, because even if the next movable | |
3051 | * allocation falls back into a different pageblock than this | |
3052 | * one, it won't cause permanent fragmentation. | |
3053 | */ | |
3054 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
3055 | && current_order > order) | |
3056 | goto find_smallest; | |
b2a0ac88 | 3057 | |
7a8f58f3 VB |
3058 | goto do_steal; |
3059 | } | |
e0fff1bd | 3060 | |
7a8f58f3 | 3061 | return false; |
e0fff1bd | 3062 | |
7a8f58f3 VB |
3063 | find_smallest: |
3064 | for (current_order = order; current_order < MAX_ORDER; | |
3065 | current_order++) { | |
3066 | area = &(zone->free_area[current_order]); | |
3067 | fallback_mt = find_suitable_fallback(area, current_order, | |
3068 | start_migratetype, false, &can_steal); | |
3069 | if (fallback_mt != -1) | |
3070 | break; | |
b2a0ac88 MG |
3071 | } |
3072 | ||
7a8f58f3 VB |
3073 | /* |
3074 | * This should not happen - we already found a suitable fallback | |
3075 | * when looking for the largest page. | |
3076 | */ | |
3077 | VM_BUG_ON(current_order == MAX_ORDER); | |
3078 | ||
3079 | do_steal: | |
b03641af | 3080 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 3081 | |
1c30844d MG |
3082 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
3083 | can_steal); | |
7a8f58f3 VB |
3084 | |
3085 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
3086 | start_migratetype, fallback_mt); | |
3087 | ||
3088 | return true; | |
3089 | ||
b2a0ac88 MG |
3090 | } |
3091 | ||
56fd56b8 | 3092 | /* |
1da177e4 LT |
3093 | * Do the hard work of removing an element from the buddy allocator. |
3094 | * Call me with the zone->lock already held. | |
3095 | */ | |
85ccc8fa | 3096 | static __always_inline struct page * |
6bb15450 MG |
3097 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
3098 | unsigned int alloc_flags) | |
1da177e4 | 3099 | { |
1da177e4 LT |
3100 | struct page *page; |
3101 | ||
ce8f86ee H |
3102 | if (IS_ENABLED(CONFIG_CMA)) { |
3103 | /* | |
3104 | * Balance movable allocations between regular and CMA areas by | |
3105 | * allocating from CMA when over half of the zone's free memory | |
3106 | * is in the CMA area. | |
3107 | */ | |
3108 | if (alloc_flags & ALLOC_CMA && | |
3109 | zone_page_state(zone, NR_FREE_CMA_PAGES) > | |
3110 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
3111 | page = __rmqueue_cma_fallback(zone, order); | |
3112 | if (page) | |
10e0f753 | 3113 | return page; |
ce8f86ee | 3114 | } |
16867664 | 3115 | } |
3bc48f96 | 3116 | retry: |
56fd56b8 | 3117 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 3118 | if (unlikely(!page)) { |
8510e69c | 3119 | if (alloc_flags & ALLOC_CMA) |
dc67647b JK |
3120 | page = __rmqueue_cma_fallback(zone, order); |
3121 | ||
6bb15450 MG |
3122 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
3123 | alloc_flags)) | |
3bc48f96 | 3124 | goto retry; |
728ec980 | 3125 | } |
b2a0ac88 | 3126 | return page; |
1da177e4 LT |
3127 | } |
3128 | ||
5f63b720 | 3129 | /* |
1da177e4 LT |
3130 | * Obtain a specified number of elements from the buddy allocator, all under |
3131 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
3132 | * Returns the number of new pages which were placed at *list. | |
3133 | */ | |
5f63b720 | 3134 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 3135 | unsigned long count, struct list_head *list, |
6bb15450 | 3136 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 3137 | { |
57490774 | 3138 | unsigned long flags; |
cb66bede | 3139 | int i, allocated = 0; |
5f63b720 | 3140 | |
57490774 | 3141 | spin_lock_irqsave(&zone->lock, flags); |
1da177e4 | 3142 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
3143 | struct page *page = __rmqueue(zone, order, migratetype, |
3144 | alloc_flags); | |
085cc7d5 | 3145 | if (unlikely(page == NULL)) |
1da177e4 | 3146 | break; |
81eabcbe | 3147 | |
77fe7f13 | 3148 | if (unlikely(check_pcp_refill(page, order))) |
479f854a MG |
3149 | continue; |
3150 | ||
81eabcbe | 3151 | /* |
0fac3ba5 VB |
3152 | * Split buddy pages returned by expand() are received here in |
3153 | * physical page order. The page is added to the tail of | |
3154 | * caller's list. From the callers perspective, the linked list | |
3155 | * is ordered by page number under some conditions. This is | |
3156 | * useful for IO devices that can forward direction from the | |
3157 | * head, thus also in the physical page order. This is useful | |
3158 | * for IO devices that can merge IO requests if the physical | |
3159 | * pages are ordered properly. | |
81eabcbe | 3160 | */ |
bf75f200 | 3161 | list_add_tail(&page->pcp_list, list); |
cb66bede | 3162 | allocated++; |
bb14c2c7 | 3163 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
3164 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
3165 | -(1 << order)); | |
1da177e4 | 3166 | } |
a6de734b MG |
3167 | |
3168 | /* | |
3169 | * i pages were removed from the buddy list even if some leak due | |
3170 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
cb66bede | 3171 | * on i. Do not confuse with 'allocated' which is the number of |
a6de734b MG |
3172 | * pages added to the pcp list. |
3173 | */ | |
f2260e6b | 3174 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
57490774 | 3175 | spin_unlock_irqrestore(&zone->lock, flags); |
cb66bede | 3176 | return allocated; |
1da177e4 LT |
3177 | } |
3178 | ||
4ae7c039 | 3179 | #ifdef CONFIG_NUMA |
8fce4d8e | 3180 | /* |
4037d452 CL |
3181 | * Called from the vmstat counter updater to drain pagesets of this |
3182 | * currently executing processor on remote nodes after they have | |
3183 | * expired. | |
8fce4d8e | 3184 | */ |
4037d452 | 3185 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 3186 | { |
7be12fc9 | 3187 | int to_drain, batch; |
4ae7c039 | 3188 | |
4db0c3c2 | 3189 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 3190 | to_drain = min(pcp->count, batch); |
4b23a68f | 3191 | if (to_drain > 0) { |
57490774 | 3192 | spin_lock(&pcp->lock); |
fd56eef2 | 3193 | free_pcppages_bulk(zone, to_drain, pcp, 0); |
57490774 | 3194 | spin_unlock(&pcp->lock); |
4b23a68f | 3195 | } |
4ae7c039 CL |
3196 | } |
3197 | #endif | |
3198 | ||
9f8f2172 | 3199 | /* |
93481ff0 | 3200 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 | 3201 | */ |
93481ff0 | 3202 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 3203 | { |
93481ff0 | 3204 | struct per_cpu_pages *pcp; |
1da177e4 | 3205 | |
28f836b6 | 3206 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
4b23a68f | 3207 | if (pcp->count) { |
57490774 | 3208 | spin_lock(&pcp->lock); |
4b23a68f | 3209 | free_pcppages_bulk(zone, pcp->count, pcp, 0); |
57490774 | 3210 | spin_unlock(&pcp->lock); |
4b23a68f | 3211 | } |
93481ff0 | 3212 | } |
3dfa5721 | 3213 | |
93481ff0 VB |
3214 | /* |
3215 | * Drain pcplists of all zones on the indicated processor. | |
93481ff0 VB |
3216 | */ |
3217 | static void drain_pages(unsigned int cpu) | |
3218 | { | |
3219 | struct zone *zone; | |
3220 | ||
3221 | for_each_populated_zone(zone) { | |
3222 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
3223 | } |
3224 | } | |
1da177e4 | 3225 | |
9f8f2172 CL |
3226 | /* |
3227 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
3228 | */ | |
93481ff0 | 3229 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 3230 | { |
93481ff0 VB |
3231 | int cpu = smp_processor_id(); |
3232 | ||
3233 | if (zone) | |
3234 | drain_pages_zone(cpu, zone); | |
3235 | else | |
3236 | drain_pages(cpu); | |
9f8f2172 CL |
3237 | } |
3238 | ||
3239 | /* | |
ec6e8c7e VB |
3240 | * The implementation of drain_all_pages(), exposing an extra parameter to |
3241 | * drain on all cpus. | |
93481ff0 | 3242 | * |
ec6e8c7e VB |
3243 | * drain_all_pages() is optimized to only execute on cpus where pcplists are |
3244 | * not empty. The check for non-emptiness can however race with a free to | |
3245 | * pcplist that has not yet increased the pcp->count from 0 to 1. Callers | |
3246 | * that need the guarantee that every CPU has drained can disable the | |
3247 | * optimizing racy check. | |
9f8f2172 | 3248 | */ |
3b1f3658 | 3249 | static void __drain_all_pages(struct zone *zone, bool force_all_cpus) |
9f8f2172 | 3250 | { |
74046494 | 3251 | int cpu; |
74046494 GBY |
3252 | |
3253 | /* | |
041711ce | 3254 | * Allocate in the BSS so we won't require allocation in |
74046494 GBY |
3255 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y |
3256 | */ | |
3257 | static cpumask_t cpus_with_pcps; | |
3258 | ||
bd233f53 MG |
3259 | /* |
3260 | * Do not drain if one is already in progress unless it's specific to | |
3261 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
3262 | * the drain to be complete when the call returns. | |
3263 | */ | |
3264 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
3265 | if (!zone) | |
3266 | return; | |
3267 | mutex_lock(&pcpu_drain_mutex); | |
3268 | } | |
0ccce3b9 | 3269 | |
74046494 GBY |
3270 | /* |
3271 | * We don't care about racing with CPU hotplug event | |
3272 | * as offline notification will cause the notified | |
3273 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
3274 | * disables preemption as part of its processing | |
3275 | */ | |
3276 | for_each_online_cpu(cpu) { | |
28f836b6 | 3277 | struct per_cpu_pages *pcp; |
93481ff0 | 3278 | struct zone *z; |
74046494 | 3279 | bool has_pcps = false; |
93481ff0 | 3280 | |
ec6e8c7e VB |
3281 | if (force_all_cpus) { |
3282 | /* | |
3283 | * The pcp.count check is racy, some callers need a | |
3284 | * guarantee that no cpu is missed. | |
3285 | */ | |
3286 | has_pcps = true; | |
3287 | } else if (zone) { | |
28f836b6 MG |
3288 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
3289 | if (pcp->count) | |
74046494 | 3290 | has_pcps = true; |
93481ff0 VB |
3291 | } else { |
3292 | for_each_populated_zone(z) { | |
28f836b6 MG |
3293 | pcp = per_cpu_ptr(z->per_cpu_pageset, cpu); |
3294 | if (pcp->count) { | |
93481ff0 VB |
3295 | has_pcps = true; |
3296 | break; | |
3297 | } | |
74046494 GBY |
3298 | } |
3299 | } | |
93481ff0 | 3300 | |
74046494 GBY |
3301 | if (has_pcps) |
3302 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
3303 | else | |
3304 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
3305 | } | |
0ccce3b9 | 3306 | |
bd233f53 | 3307 | for_each_cpu(cpu, &cpus_with_pcps) { |
443c2acc NSJ |
3308 | if (zone) |
3309 | drain_pages_zone(cpu, zone); | |
3310 | else | |
3311 | drain_pages(cpu); | |
0ccce3b9 | 3312 | } |
bd233f53 MG |
3313 | |
3314 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
3315 | } |
3316 | ||
ec6e8c7e VB |
3317 | /* |
3318 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. | |
3319 | * | |
3320 | * When zone parameter is non-NULL, spill just the single zone's pages. | |
ec6e8c7e VB |
3321 | */ |
3322 | void drain_all_pages(struct zone *zone) | |
3323 | { | |
3324 | __drain_all_pages(zone, false); | |
3325 | } | |
3326 | ||
296699de | 3327 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3328 | |
556b969a CY |
3329 | /* |
3330 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
3331 | */ | |
3332 | #define WD_PAGE_COUNT (128*1024) | |
3333 | ||
1da177e4 LT |
3334 | void mark_free_pages(struct zone *zone) |
3335 | { | |
556b969a | 3336 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 3337 | unsigned long flags; |
7aeb09f9 | 3338 | unsigned int order, t; |
86760a2c | 3339 | struct page *page; |
1da177e4 | 3340 | |
8080fc03 | 3341 | if (zone_is_empty(zone)) |
1da177e4 LT |
3342 | return; |
3343 | ||
3344 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 3345 | |
108bcc96 | 3346 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
3347 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
3348 | if (pfn_valid(pfn)) { | |
86760a2c | 3349 | page = pfn_to_page(pfn); |
ba6b0979 | 3350 | |
556b969a CY |
3351 | if (!--page_count) { |
3352 | touch_nmi_watchdog(); | |
3353 | page_count = WD_PAGE_COUNT; | |
3354 | } | |
3355 | ||
ba6b0979 JK |
3356 | if (page_zone(page) != zone) |
3357 | continue; | |
3358 | ||
7be98234 RW |
3359 | if (!swsusp_page_is_forbidden(page)) |
3360 | swsusp_unset_page_free(page); | |
f623f0db | 3361 | } |
1da177e4 | 3362 | |
b2a0ac88 | 3363 | for_each_migratetype_order(order, t) { |
86760a2c | 3364 | list_for_each_entry(page, |
bf75f200 | 3365 | &zone->free_area[order].free_list[t], buddy_list) { |
f623f0db | 3366 | unsigned long i; |
1da177e4 | 3367 | |
86760a2c | 3368 | pfn = page_to_pfn(page); |
556b969a CY |
3369 | for (i = 0; i < (1UL << order); i++) { |
3370 | if (!--page_count) { | |
3371 | touch_nmi_watchdog(); | |
3372 | page_count = WD_PAGE_COUNT; | |
3373 | } | |
7be98234 | 3374 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 3375 | } |
f623f0db | 3376 | } |
b2a0ac88 | 3377 | } |
1da177e4 LT |
3378 | spin_unlock_irqrestore(&zone->lock, flags); |
3379 | } | |
e2c55dc8 | 3380 | #endif /* CONFIG_PM */ |
1da177e4 | 3381 | |
44042b44 MG |
3382 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn, |
3383 | unsigned int order) | |
1da177e4 | 3384 | { |
5f8dcc21 | 3385 | int migratetype; |
1da177e4 | 3386 | |
44042b44 | 3387 | if (!free_pcp_prepare(page, order)) |
9cca35d4 | 3388 | return false; |
689bcebf | 3389 | |
dc4b0caf | 3390 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 3391 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
3392 | return true; |
3393 | } | |
3394 | ||
f26b3fa0 MG |
3395 | static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch, |
3396 | bool free_high) | |
3b12e7e9 MG |
3397 | { |
3398 | int min_nr_free, max_nr_free; | |
3399 | ||
f26b3fa0 MG |
3400 | /* Free everything if batch freeing high-order pages. */ |
3401 | if (unlikely(free_high)) | |
3402 | return pcp->count; | |
3403 | ||
3b12e7e9 MG |
3404 | /* Check for PCP disabled or boot pageset */ |
3405 | if (unlikely(high < batch)) | |
3406 | return 1; | |
3407 | ||
3408 | /* Leave at least pcp->batch pages on the list */ | |
3409 | min_nr_free = batch; | |
3410 | max_nr_free = high - batch; | |
3411 | ||
3412 | /* | |
3413 | * Double the number of pages freed each time there is subsequent | |
3414 | * freeing of pages without any allocation. | |
3415 | */ | |
3416 | batch <<= pcp->free_factor; | |
3417 | if (batch < max_nr_free) | |
3418 | pcp->free_factor++; | |
3419 | batch = clamp(batch, min_nr_free, max_nr_free); | |
3420 | ||
3421 | return batch; | |
3422 | } | |
3423 | ||
f26b3fa0 MG |
3424 | static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone, |
3425 | bool free_high) | |
c49c2c47 MG |
3426 | { |
3427 | int high = READ_ONCE(pcp->high); | |
3428 | ||
f26b3fa0 | 3429 | if (unlikely(!high || free_high)) |
c49c2c47 MG |
3430 | return 0; |
3431 | ||
3432 | if (!test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags)) | |
3433 | return high; | |
3434 | ||
3435 | /* | |
3436 | * If reclaim is active, limit the number of pages that can be | |
3437 | * stored on pcp lists | |
3438 | */ | |
3439 | return min(READ_ONCE(pcp->batch) << 2, high); | |
3440 | } | |
3441 | ||
4b23a68f MG |
3442 | static void free_unref_page_commit(struct zone *zone, struct per_cpu_pages *pcp, |
3443 | struct page *page, int migratetype, | |
56651377 | 3444 | unsigned int order) |
9cca35d4 | 3445 | { |
3b12e7e9 | 3446 | int high; |
44042b44 | 3447 | int pindex; |
f26b3fa0 | 3448 | bool free_high; |
9cca35d4 | 3449 | |
15cd9004 | 3450 | __count_vm_events(PGFREE, 1 << order); |
44042b44 | 3451 | pindex = order_to_pindex(migratetype, order); |
bf75f200 | 3452 | list_add(&page->pcp_list, &pcp->lists[pindex]); |
44042b44 | 3453 | pcp->count += 1 << order; |
f26b3fa0 MG |
3454 | |
3455 | /* | |
3456 | * As high-order pages other than THP's stored on PCP can contribute | |
3457 | * to fragmentation, limit the number stored when PCP is heavily | |
3458 | * freeing without allocation. The remainder after bulk freeing | |
3459 | * stops will be drained from vmstat refresh context. | |
3460 | */ | |
3461 | free_high = (pcp->free_factor && order && order <= PAGE_ALLOC_COSTLY_ORDER); | |
3462 | ||
3463 | high = nr_pcp_high(pcp, zone, free_high); | |
3b12e7e9 MG |
3464 | if (pcp->count >= high) { |
3465 | int batch = READ_ONCE(pcp->batch); | |
3466 | ||
f26b3fa0 | 3467 | free_pcppages_bulk(zone, nr_pcp_free(pcp, high, batch, free_high), pcp, pindex); |
3b12e7e9 | 3468 | } |
9cca35d4 | 3469 | } |
5f8dcc21 | 3470 | |
9cca35d4 | 3471 | /* |
44042b44 | 3472 | * Free a pcp page |
9cca35d4 | 3473 | */ |
44042b44 | 3474 | void free_unref_page(struct page *page, unsigned int order) |
9cca35d4 | 3475 | { |
4b23a68f MG |
3476 | unsigned long __maybe_unused UP_flags; |
3477 | struct per_cpu_pages *pcp; | |
3478 | struct zone *zone; | |
9cca35d4 | 3479 | unsigned long pfn = page_to_pfn(page); |
df1acc85 | 3480 | int migratetype; |
9cca35d4 | 3481 | |
44042b44 | 3482 | if (!free_unref_page_prepare(page, pfn, order)) |
9cca35d4 | 3483 | return; |
da456f14 | 3484 | |
5f8dcc21 MG |
3485 | /* |
3486 | * We only track unmovable, reclaimable and movable on pcp lists. | |
df1acc85 | 3487 | * Place ISOLATE pages on the isolated list because they are being |
a6ffdc07 | 3488 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
3489 | * areas back if necessary. Otherwise, we may have to free |
3490 | * excessively into the page allocator | |
3491 | */ | |
df1acc85 MG |
3492 | migratetype = get_pcppage_migratetype(page); |
3493 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) { | |
194159fb | 3494 | if (unlikely(is_migrate_isolate(migratetype))) { |
44042b44 | 3495 | free_one_page(page_zone(page), page, pfn, order, migratetype, FPI_NONE); |
9cca35d4 | 3496 | return; |
5f8dcc21 MG |
3497 | } |
3498 | migratetype = MIGRATE_MOVABLE; | |
3499 | } | |
3500 | ||
4b23a68f MG |
3501 | zone = page_zone(page); |
3502 | pcp_trylock_prepare(UP_flags); | |
57490774 | 3503 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 3504 | if (pcp) { |
4b23a68f | 3505 | free_unref_page_commit(zone, pcp, page, migratetype, order); |
57490774 | 3506 | pcp_spin_unlock(pcp); |
4b23a68f MG |
3507 | } else { |
3508 | free_one_page(zone, page, pfn, order, migratetype, FPI_NONE); | |
3509 | } | |
3510 | pcp_trylock_finish(UP_flags); | |
1da177e4 LT |
3511 | } |
3512 | ||
cc59850e KK |
3513 | /* |
3514 | * Free a list of 0-order pages | |
3515 | */ | |
2d4894b5 | 3516 | void free_unref_page_list(struct list_head *list) |
cc59850e | 3517 | { |
57490774 | 3518 | unsigned long __maybe_unused UP_flags; |
cc59850e | 3519 | struct page *page, *next; |
4b23a68f MG |
3520 | struct per_cpu_pages *pcp = NULL; |
3521 | struct zone *locked_zone = NULL; | |
c24ad77d | 3522 | int batch_count = 0; |
df1acc85 | 3523 | int migratetype; |
9cca35d4 MG |
3524 | |
3525 | /* Prepare pages for freeing */ | |
3526 | list_for_each_entry_safe(page, next, list, lru) { | |
56651377 | 3527 | unsigned long pfn = page_to_pfn(page); |
053cfda1 | 3528 | if (!free_unref_page_prepare(page, pfn, 0)) { |
9cca35d4 | 3529 | list_del(&page->lru); |
053cfda1 ML |
3530 | continue; |
3531 | } | |
df1acc85 MG |
3532 | |
3533 | /* | |
3534 | * Free isolated pages directly to the allocator, see | |
3535 | * comment in free_unref_page. | |
3536 | */ | |
3537 | migratetype = get_pcppage_migratetype(page); | |
47aef601 DB |
3538 | if (unlikely(is_migrate_isolate(migratetype))) { |
3539 | list_del(&page->lru); | |
3540 | free_one_page(page_zone(page), page, pfn, 0, migratetype, FPI_NONE); | |
3541 | continue; | |
df1acc85 | 3542 | } |
9cca35d4 | 3543 | } |
cc59850e KK |
3544 | |
3545 | list_for_each_entry_safe(page, next, list, lru) { | |
4b23a68f MG |
3546 | struct zone *zone = page_zone(page); |
3547 | ||
c3e58a70 | 3548 | list_del(&page->lru); |
57490774 | 3549 | migratetype = get_pcppage_migratetype(page); |
c3e58a70 | 3550 | |
a4bafffb MG |
3551 | /* |
3552 | * Either different zone requiring a different pcp lock or | |
3553 | * excessive lock hold times when freeing a large list of | |
3554 | * pages. | |
3555 | */ | |
3556 | if (zone != locked_zone || batch_count == SWAP_CLUSTER_MAX) { | |
57490774 MG |
3557 | if (pcp) { |
3558 | pcp_spin_unlock(pcp); | |
3559 | pcp_trylock_finish(UP_flags); | |
3560 | } | |
01b44456 | 3561 | |
a4bafffb MG |
3562 | batch_count = 0; |
3563 | ||
57490774 MG |
3564 | /* |
3565 | * trylock is necessary as pages may be getting freed | |
3566 | * from IRQ or SoftIRQ context after an IO completion. | |
3567 | */ | |
3568 | pcp_trylock_prepare(UP_flags); | |
3569 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); | |
3570 | if (unlikely(!pcp)) { | |
3571 | pcp_trylock_finish(UP_flags); | |
3572 | free_one_page(zone, page, page_to_pfn(page), | |
3573 | 0, migratetype, FPI_NONE); | |
3574 | locked_zone = NULL; | |
3575 | continue; | |
3576 | } | |
4b23a68f | 3577 | locked_zone = zone; |
4b23a68f MG |
3578 | } |
3579 | ||
47aef601 DB |
3580 | /* |
3581 | * Non-isolated types over MIGRATE_PCPTYPES get added | |
3582 | * to the MIGRATE_MOVABLE pcp list. | |
3583 | */ | |
47aef601 DB |
3584 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) |
3585 | migratetype = MIGRATE_MOVABLE; | |
3586 | ||
2d4894b5 | 3587 | trace_mm_page_free_batched(page); |
4b23a68f | 3588 | free_unref_page_commit(zone, pcp, page, migratetype, 0); |
a4bafffb | 3589 | batch_count++; |
cc59850e | 3590 | } |
4b23a68f | 3591 | |
57490774 MG |
3592 | if (pcp) { |
3593 | pcp_spin_unlock(pcp); | |
3594 | pcp_trylock_finish(UP_flags); | |
3595 | } | |
cc59850e KK |
3596 | } |
3597 | ||
8dfcc9ba NP |
3598 | /* |
3599 | * split_page takes a non-compound higher-order page, and splits it into | |
3600 | * n (1<<order) sub-pages: page[0..n] | |
3601 | * Each sub-page must be freed individually. | |
3602 | * | |
3603 | * Note: this is probably too low level an operation for use in drivers. | |
3604 | * Please consult with lkml before using this in your driver. | |
3605 | */ | |
3606 | void split_page(struct page *page, unsigned int order) | |
3607 | { | |
3608 | int i; | |
3609 | ||
309381fe SL |
3610 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3611 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3612 | |
a9627bc5 | 3613 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3614 | set_page_refcounted(page + i); |
8fb156c9 | 3615 | split_page_owner(page, 1 << order); |
e1baddf8 | 3616 | split_page_memcg(page, 1 << order); |
8dfcc9ba | 3617 | } |
5853ff23 | 3618 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3619 | |
3c605096 | 3620 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3621 | { |
9a157dd8 KW |
3622 | struct zone *zone = page_zone(page); |
3623 | int mt = get_pageblock_migratetype(page); | |
748446bb | 3624 | |
194159fb | 3625 | if (!is_migrate_isolate(mt)) { |
9a157dd8 | 3626 | unsigned long watermark; |
8348faf9 VB |
3627 | /* |
3628 | * Obey watermarks as if the page was being allocated. We can | |
3629 | * emulate a high-order watermark check with a raised order-0 | |
3630 | * watermark, because we already know our high-order page | |
3631 | * exists. | |
3632 | */ | |
fd1444b2 | 3633 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3634 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3635 | return 0; |
3636 | ||
8fb74b9f | 3637 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3638 | } |
748446bb | 3639 | |
6ab01363 | 3640 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 3641 | |
400bc7fd | 3642 | /* |
3643 | * Set the pageblock if the isolated page is at least half of a | |
3644 | * pageblock | |
3645 | */ | |
748446bb MG |
3646 | if (order >= pageblock_order - 1) { |
3647 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3648 | for (; page < endpage; page += pageblock_nr_pages) { |
3649 | int mt = get_pageblock_migratetype(page); | |
1dd214b8 ZY |
3650 | /* |
3651 | * Only change normal pageblocks (i.e., they can merge | |
3652 | * with others) | |
3653 | */ | |
3654 | if (migratetype_is_mergeable(mt)) | |
47118af0 MN |
3655 | set_pageblock_migratetype(page, |
3656 | MIGRATE_MOVABLE); | |
3657 | } | |
748446bb MG |
3658 | } |
3659 | ||
8fb74b9f | 3660 | return 1UL << order; |
1fb3f8ca MG |
3661 | } |
3662 | ||
624f58d8 AD |
3663 | /** |
3664 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
3665 | * @page: Page that was isolated | |
3666 | * @order: Order of the isolated page | |
e6a0a7ad | 3667 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
3668 | * |
3669 | * This function is meant to return a page pulled from the free lists via | |
3670 | * __isolate_free_page back to the free lists they were pulled from. | |
3671 | */ | |
3672 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
3673 | { | |
3674 | struct zone *zone = page_zone(page); | |
3675 | ||
3676 | /* zone lock should be held when this function is called */ | |
3677 | lockdep_assert_held(&zone->lock); | |
3678 | ||
3679 | /* Return isolated page to tail of freelist. */ | |
f04a5d5d | 3680 | __free_one_page(page, page_to_pfn(page), zone, order, mt, |
47b6a24a | 3681 | FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL); |
624f58d8 AD |
3682 | } |
3683 | ||
060e7417 MG |
3684 | /* |
3685 | * Update NUMA hit/miss statistics | |
060e7417 | 3686 | */ |
3e23060b MG |
3687 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, |
3688 | long nr_account) | |
060e7417 MG |
3689 | { |
3690 | #ifdef CONFIG_NUMA | |
3a321d2a | 3691 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3692 | |
4518085e KW |
3693 | /* skip numa counters update if numa stats is disabled */ |
3694 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3695 | return; | |
3696 | ||
c1093b74 | 3697 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3698 | local_stat = NUMA_OTHER; |
060e7417 | 3699 | |
c1093b74 | 3700 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3e23060b | 3701 | __count_numa_events(z, NUMA_HIT, nr_account); |
2df26639 | 3702 | else { |
3e23060b MG |
3703 | __count_numa_events(z, NUMA_MISS, nr_account); |
3704 | __count_numa_events(preferred_zone, NUMA_FOREIGN, nr_account); | |
060e7417 | 3705 | } |
3e23060b | 3706 | __count_numa_events(z, local_stat, nr_account); |
060e7417 MG |
3707 | #endif |
3708 | } | |
3709 | ||
589d9973 MG |
3710 | static __always_inline |
3711 | struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, | |
3712 | unsigned int order, unsigned int alloc_flags, | |
3713 | int migratetype) | |
3714 | { | |
3715 | struct page *page; | |
3716 | unsigned long flags; | |
3717 | ||
3718 | do { | |
3719 | page = NULL; | |
3720 | spin_lock_irqsave(&zone->lock, flags); | |
3721 | /* | |
3722 | * order-0 request can reach here when the pcplist is skipped | |
3723 | * due to non-CMA allocation context. HIGHATOMIC area is | |
3724 | * reserved for high-order atomic allocation, so order-0 | |
3725 | * request should skip it. | |
3726 | */ | |
eb2e2b42 | 3727 | if (alloc_flags & ALLOC_HIGHATOMIC) |
589d9973 MG |
3728 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
3729 | if (!page) { | |
3730 | page = __rmqueue(zone, order, migratetype, alloc_flags); | |
eb2e2b42 MG |
3731 | |
3732 | /* | |
3733 | * If the allocation fails, allow OOM handling access | |
3734 | * to HIGHATOMIC reserves as failing now is worse than | |
3735 | * failing a high-order atomic allocation in the | |
3736 | * future. | |
3737 | */ | |
3738 | if (!page && (alloc_flags & ALLOC_OOM)) | |
3739 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3740 | ||
589d9973 MG |
3741 | if (!page) { |
3742 | spin_unlock_irqrestore(&zone->lock, flags); | |
3743 | return NULL; | |
3744 | } | |
3745 | } | |
3746 | __mod_zone_freepage_state(zone, -(1 << order), | |
3747 | get_pcppage_migratetype(page)); | |
3748 | spin_unlock_irqrestore(&zone->lock, flags); | |
3749 | } while (check_new_pages(page, order)); | |
3750 | ||
3751 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
3752 | zone_statistics(preferred_zone, zone, 1); | |
3753 | ||
3754 | return page; | |
3755 | } | |
3756 | ||
066b2393 | 3757 | /* Remove page from the per-cpu list, caller must protect the list */ |
3b822017 | 3758 | static inline |
44042b44 MG |
3759 | struct page *__rmqueue_pcplist(struct zone *zone, unsigned int order, |
3760 | int migratetype, | |
6bb15450 | 3761 | unsigned int alloc_flags, |
453f85d4 | 3762 | struct per_cpu_pages *pcp, |
066b2393 MG |
3763 | struct list_head *list) |
3764 | { | |
3765 | struct page *page; | |
3766 | ||
3767 | do { | |
3768 | if (list_empty(list)) { | |
44042b44 MG |
3769 | int batch = READ_ONCE(pcp->batch); |
3770 | int alloced; | |
3771 | ||
3772 | /* | |
3773 | * Scale batch relative to order if batch implies | |
3774 | * free pages can be stored on the PCP. Batch can | |
3775 | * be 1 for small zones or for boot pagesets which | |
3776 | * should never store free pages as the pages may | |
3777 | * belong to arbitrary zones. | |
3778 | */ | |
3779 | if (batch > 1) | |
3780 | batch = max(batch >> order, 2); | |
3781 | alloced = rmqueue_bulk(zone, order, | |
3782 | batch, list, | |
6bb15450 | 3783 | migratetype, alloc_flags); |
44042b44 MG |
3784 | |
3785 | pcp->count += alloced << order; | |
066b2393 MG |
3786 | if (unlikely(list_empty(list))) |
3787 | return NULL; | |
3788 | } | |
3789 | ||
bf75f200 MG |
3790 | page = list_first_entry(list, struct page, pcp_list); |
3791 | list_del(&page->pcp_list); | |
44042b44 | 3792 | pcp->count -= 1 << order; |
77fe7f13 | 3793 | } while (check_new_pcp(page, order)); |
066b2393 MG |
3794 | |
3795 | return page; | |
3796 | } | |
3797 | ||
3798 | /* Lock and remove page from the per-cpu list */ | |
3799 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
44042b44 | 3800 | struct zone *zone, unsigned int order, |
663d0cfd | 3801 | int migratetype, unsigned int alloc_flags) |
066b2393 MG |
3802 | { |
3803 | struct per_cpu_pages *pcp; | |
3804 | struct list_head *list; | |
066b2393 | 3805 | struct page *page; |
4b23a68f | 3806 | unsigned long __maybe_unused UP_flags; |
066b2393 | 3807 | |
57490774 | 3808 | /* spin_trylock may fail due to a parallel drain or IRQ reentrancy. */ |
4b23a68f | 3809 | pcp_trylock_prepare(UP_flags); |
57490774 | 3810 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 3811 | if (!pcp) { |
4b23a68f | 3812 | pcp_trylock_finish(UP_flags); |
4b23a68f MG |
3813 | return NULL; |
3814 | } | |
3b12e7e9 MG |
3815 | |
3816 | /* | |
3817 | * On allocation, reduce the number of pages that are batch freed. | |
3818 | * See nr_pcp_free() where free_factor is increased for subsequent | |
3819 | * frees. | |
3820 | */ | |
3b12e7e9 | 3821 | pcp->free_factor >>= 1; |
44042b44 MG |
3822 | list = &pcp->lists[order_to_pindex(migratetype, order)]; |
3823 | page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list); | |
57490774 | 3824 | pcp_spin_unlock(pcp); |
4b23a68f | 3825 | pcp_trylock_finish(UP_flags); |
066b2393 | 3826 | if (page) { |
15cd9004 | 3827 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
3e23060b | 3828 | zone_statistics(preferred_zone, zone, 1); |
066b2393 | 3829 | } |
066b2393 MG |
3830 | return page; |
3831 | } | |
3832 | ||
1da177e4 | 3833 | /* |
a57ae9ef RX |
3834 | * Allocate a page from the given zone. |
3835 | * Use pcplists for THP or "cheap" high-order allocations. | |
1da177e4 | 3836 | */ |
b073d7f8 AP |
3837 | |
3838 | /* | |
3839 | * Do not instrument rmqueue() with KMSAN. This function may call | |
3840 | * __msan_poison_alloca() through a call to set_pfnblock_flags_mask(). | |
3841 | * If __msan_poison_alloca() attempts to allocate pages for the stack depot, it | |
3842 | * may call rmqueue() again, which will result in a deadlock. | |
1da177e4 | 3843 | */ |
b073d7f8 | 3844 | __no_sanitize_memory |
0a15c3e9 | 3845 | static inline |
066b2393 | 3846 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3847 | struct zone *zone, unsigned int order, |
c603844b MG |
3848 | gfp_t gfp_flags, unsigned int alloc_flags, |
3849 | int migratetype) | |
1da177e4 | 3850 | { |
689bcebf | 3851 | struct page *page; |
1da177e4 | 3852 | |
589d9973 MG |
3853 | /* |
3854 | * We most definitely don't want callers attempting to | |
3855 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3856 | */ | |
3857 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3858 | ||
44042b44 | 3859 | if (likely(pcp_allowed_order(order))) { |
1d91df85 JK |
3860 | /* |
3861 | * MIGRATE_MOVABLE pcplist could have the pages on CMA area and | |
3862 | * we need to skip it when CMA area isn't allowed. | |
3863 | */ | |
3864 | if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA || | |
3865 | migratetype != MIGRATE_MOVABLE) { | |
44042b44 | 3866 | page = rmqueue_pcplist(preferred_zone, zone, order, |
663d0cfd | 3867 | migratetype, alloc_flags); |
4b23a68f MG |
3868 | if (likely(page)) |
3869 | goto out; | |
1d91df85 | 3870 | } |
066b2393 | 3871 | } |
83b9355b | 3872 | |
589d9973 MG |
3873 | page = rmqueue_buddy(preferred_zone, zone, order, alloc_flags, |
3874 | migratetype); | |
1da177e4 | 3875 | |
066b2393 | 3876 | out: |
73444bc4 | 3877 | /* Separate test+clear to avoid unnecessary atomics */ |
e2a66c21 | 3878 | if (unlikely(test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags))) { |
73444bc4 MG |
3879 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
3880 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3881 | } | |
3882 | ||
066b2393 | 3883 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 LT |
3884 | return page; |
3885 | } | |
3886 | ||
933e312e AM |
3887 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3888 | ||
b2588c4b | 3889 | static struct { |
933e312e AM |
3890 | struct fault_attr attr; |
3891 | ||
621a5f7a | 3892 | bool ignore_gfp_highmem; |
71baba4b | 3893 | bool ignore_gfp_reclaim; |
54114994 | 3894 | u32 min_order; |
933e312e AM |
3895 | } fail_page_alloc = { |
3896 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3897 | .ignore_gfp_reclaim = true, |
621a5f7a | 3898 | .ignore_gfp_highmem = true, |
54114994 | 3899 | .min_order = 1, |
933e312e AM |
3900 | }; |
3901 | ||
3902 | static int __init setup_fail_page_alloc(char *str) | |
3903 | { | |
3904 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3905 | } | |
3906 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3907 | ||
af3b8544 | 3908 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3909 | { |
ea4452de QZ |
3910 | int flags = 0; |
3911 | ||
54114994 | 3912 | if (order < fail_page_alloc.min_order) |
deaf386e | 3913 | return false; |
933e312e | 3914 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3915 | return false; |
933e312e | 3916 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3917 | return false; |
71baba4b MG |
3918 | if (fail_page_alloc.ignore_gfp_reclaim && |
3919 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3920 | return false; |
933e312e | 3921 | |
ea4452de | 3922 | /* See comment in __should_failslab() */ |
3f913fc5 | 3923 | if (gfp_mask & __GFP_NOWARN) |
ea4452de | 3924 | flags |= FAULT_NOWARN; |
3f913fc5 | 3925 | |
ea4452de | 3926 | return should_fail_ex(&fail_page_alloc.attr, 1 << order, flags); |
933e312e AM |
3927 | } |
3928 | ||
3929 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3930 | ||
3931 | static int __init fail_page_alloc_debugfs(void) | |
3932 | { | |
0825a6f9 | 3933 | umode_t mode = S_IFREG | 0600; |
933e312e | 3934 | struct dentry *dir; |
933e312e | 3935 | |
dd48c085 AM |
3936 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3937 | &fail_page_alloc.attr); | |
b2588c4b | 3938 | |
d9f7979c GKH |
3939 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3940 | &fail_page_alloc.ignore_gfp_reclaim); | |
3941 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3942 | &fail_page_alloc.ignore_gfp_highmem); | |
3943 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3944 | |
d9f7979c | 3945 | return 0; |
933e312e AM |
3946 | } |
3947 | ||
3948 | late_initcall(fail_page_alloc_debugfs); | |
3949 | ||
3950 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3951 | ||
3952 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3953 | ||
af3b8544 | 3954 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3955 | { |
deaf386e | 3956 | return false; |
933e312e AM |
3957 | } |
3958 | ||
3959 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3960 | ||
54aa3866 | 3961 | noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
af3b8544 BP |
3962 | { |
3963 | return __should_fail_alloc_page(gfp_mask, order); | |
3964 | } | |
3965 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3966 | ||
f27ce0e1 JK |
3967 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3968 | unsigned int order, unsigned int alloc_flags) | |
3969 | { | |
f27ce0e1 JK |
3970 | long unusable_free = (1 << order) - 1; |
3971 | ||
3972 | /* | |
ab350885 MG |
3973 | * If the caller does not have rights to reserves below the min |
3974 | * watermark then subtract the high-atomic reserves. This will | |
3975 | * over-estimate the size of the atomic reserve but it avoids a search. | |
f27ce0e1 | 3976 | */ |
ab350885 | 3977 | if (likely(!(alloc_flags & ALLOC_RESERVES))) |
f27ce0e1 JK |
3978 | unusable_free += z->nr_reserved_highatomic; |
3979 | ||
3980 | #ifdef CONFIG_CMA | |
3981 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3982 | if (!(alloc_flags & ALLOC_CMA)) | |
3983 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3984 | #endif | |
3985 | ||
3986 | return unusable_free; | |
3987 | } | |
3988 | ||
1da177e4 | 3989 | /* |
97a16fc8 MG |
3990 | * Return true if free base pages are above 'mark'. For high-order checks it |
3991 | * will return true of the order-0 watermark is reached and there is at least | |
3992 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3993 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3994 | */ |
86a294a8 | 3995 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3996 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3997 | long free_pages) |
1da177e4 | 3998 | { |
d23ad423 | 3999 | long min = mark; |
1da177e4 LT |
4000 | int o; |
4001 | ||
0aaa29a5 | 4002 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 4003 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 4004 | |
ab350885 MG |
4005 | if (unlikely(alloc_flags & ALLOC_RESERVES)) { |
4006 | /* | |
4007 | * __GFP_HIGH allows access to 50% of the min reserve as well | |
4008 | * as OOM. | |
4009 | */ | |
1ebbb218 | 4010 | if (alloc_flags & ALLOC_MIN_RESERVE) { |
ab350885 | 4011 | min -= min / 2; |
0aaa29a5 | 4012 | |
1ebbb218 MG |
4013 | /* |
4014 | * Non-blocking allocations (e.g. GFP_ATOMIC) can | |
4015 | * access more reserves than just __GFP_HIGH. Other | |
4016 | * non-blocking allocations requests such as GFP_NOWAIT | |
4017 | * or (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) do not get | |
4018 | * access to the min reserve. | |
4019 | */ | |
4020 | if (alloc_flags & ALLOC_NON_BLOCK) | |
4021 | min -= min / 4; | |
4022 | } | |
ab350885 MG |
4023 | |
4024 | /* | |
4025 | * OOM victims can try even harder than the normal reserve | |
cd04ae1e MH |
4026 | * users on the grounds that it's definitely going to be in |
4027 | * the exit path shortly and free memory. Any allocation it | |
4028 | * makes during the free path will be small and short-lived. | |
4029 | */ | |
4030 | if (alloc_flags & ALLOC_OOM) | |
4031 | min -= min / 2; | |
cd04ae1e MH |
4032 | } |
4033 | ||
97a16fc8 MG |
4034 | /* |
4035 | * Check watermarks for an order-0 allocation request. If these | |
4036 | * are not met, then a high-order request also cannot go ahead | |
4037 | * even if a suitable page happened to be free. | |
4038 | */ | |
97a225e6 | 4039 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 4040 | return false; |
1da177e4 | 4041 | |
97a16fc8 MG |
4042 | /* If this is an order-0 request then the watermark is fine */ |
4043 | if (!order) | |
4044 | return true; | |
4045 | ||
4046 | /* For a high-order request, check at least one suitable page is free */ | |
4047 | for (o = order; o < MAX_ORDER; o++) { | |
4048 | struct free_area *area = &z->free_area[o]; | |
4049 | int mt; | |
4050 | ||
4051 | if (!area->nr_free) | |
4052 | continue; | |
4053 | ||
97a16fc8 | 4054 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 4055 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
4056 | return true; |
4057 | } | |
4058 | ||
4059 | #ifdef CONFIG_CMA | |
d883c6cf | 4060 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 4061 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 4062 | return true; |
d883c6cf | 4063 | } |
97a16fc8 | 4064 | #endif |
eb2e2b42 MG |
4065 | if ((alloc_flags & (ALLOC_HIGHATOMIC|ALLOC_OOM)) && |
4066 | !free_area_empty(area, MIGRATE_HIGHATOMIC)) { | |
b050e376 | 4067 | return true; |
eb2e2b42 | 4068 | } |
1da177e4 | 4069 | } |
97a16fc8 | 4070 | return false; |
88f5acf8 MG |
4071 | } |
4072 | ||
7aeb09f9 | 4073 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 4074 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 4075 | { |
97a225e6 | 4076 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
4077 | zone_page_state(z, NR_FREE_PAGES)); |
4078 | } | |
4079 | ||
48ee5f36 | 4080 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 4081 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 4082 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 4083 | { |
f27ce0e1 | 4084 | long free_pages; |
d883c6cf | 4085 | |
f27ce0e1 | 4086 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
4087 | |
4088 | /* | |
4089 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 4090 | * need to be calculated. |
48ee5f36 | 4091 | */ |
f27ce0e1 | 4092 | if (!order) { |
9282012f JK |
4093 | long usable_free; |
4094 | long reserved; | |
f27ce0e1 | 4095 | |
9282012f JK |
4096 | usable_free = free_pages; |
4097 | reserved = __zone_watermark_unusable_free(z, 0, alloc_flags); | |
4098 | ||
4099 | /* reserved may over estimate high-atomic reserves. */ | |
4100 | usable_free -= min(usable_free, reserved); | |
4101 | if (usable_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
f27ce0e1 JK |
4102 | return true; |
4103 | } | |
48ee5f36 | 4104 | |
f80b08fc CTR |
4105 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
4106 | free_pages)) | |
4107 | return true; | |
4108 | /* | |
4109 | * Ignore watermark boosting for GFP_ATOMIC order-0 allocations | |
4110 | * when checking the min watermark. The min watermark is the | |
4111 | * point where boosting is ignored so that kswapd is woken up | |
4112 | * when below the low watermark. | |
4113 | */ | |
4114 | if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost | |
4115 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { | |
4116 | mark = z->_watermark[WMARK_MIN]; | |
4117 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
4118 | alloc_flags, free_pages); | |
4119 | } | |
4120 | ||
4121 | return false; | |
48ee5f36 MG |
4122 | } |
4123 | ||
7aeb09f9 | 4124 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 4125 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
4126 | { |
4127 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
4128 | ||
4129 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
4130 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
4131 | ||
97a225e6 | 4132 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 4133 | free_pages); |
1da177e4 LT |
4134 | } |
4135 | ||
9276b1bc | 4136 | #ifdef CONFIG_NUMA |
61bb6cd2 GU |
4137 | int __read_mostly node_reclaim_distance = RECLAIM_DISTANCE; |
4138 | ||
957f822a DR |
4139 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
4140 | { | |
e02dc017 | 4141 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 4142 | node_reclaim_distance; |
957f822a | 4143 | } |
9276b1bc | 4144 | #else /* CONFIG_NUMA */ |
957f822a DR |
4145 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
4146 | { | |
4147 | return true; | |
4148 | } | |
9276b1bc PJ |
4149 | #endif /* CONFIG_NUMA */ |
4150 | ||
6bb15450 MG |
4151 | /* |
4152 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
4153 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
4154 | * premature use of a lower zone may cause lowmem pressure problems that | |
4155 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
4156 | * probably too small. It only makes sense to spread allocations to avoid | |
4157 | * fragmentation between the Normal and DMA32 zones. | |
4158 | */ | |
4159 | static inline unsigned int | |
0a79cdad | 4160 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 4161 | { |
736838e9 | 4162 | unsigned int alloc_flags; |
0a79cdad | 4163 | |
736838e9 MN |
4164 | /* |
4165 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4166 | * to save a branch. | |
4167 | */ | |
4168 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
4169 | |
4170 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
4171 | if (!zone) |
4172 | return alloc_flags; | |
4173 | ||
6bb15450 | 4174 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 4175 | return alloc_flags; |
6bb15450 MG |
4176 | |
4177 | /* | |
4178 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
4179 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
4180 | * on UMA that if Normal is populated then so is DMA32. | |
4181 | */ | |
4182 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
4183 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 4184 | return alloc_flags; |
6bb15450 | 4185 | |
8118b82e | 4186 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
4187 | #endif /* CONFIG_ZONE_DMA32 */ |
4188 | return alloc_flags; | |
6bb15450 | 4189 | } |
6bb15450 | 4190 | |
8e3560d9 PT |
4191 | /* Must be called after current_gfp_context() which can change gfp_mask */ |
4192 | static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask, | |
4193 | unsigned int alloc_flags) | |
8510e69c JK |
4194 | { |
4195 | #ifdef CONFIG_CMA | |
8e3560d9 | 4196 | if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
8510e69c | 4197 | alloc_flags |= ALLOC_CMA; |
8510e69c JK |
4198 | #endif |
4199 | return alloc_flags; | |
4200 | } | |
4201 | ||
7fb1d9fc | 4202 | /* |
0798e519 | 4203 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
4204 | * a page. |
4205 | */ | |
4206 | static struct page * | |
a9263751 VB |
4207 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
4208 | const struct alloc_context *ac) | |
753ee728 | 4209 | { |
6bb15450 | 4210 | struct zoneref *z; |
5117f45d | 4211 | struct zone *zone; |
8a87d695 WY |
4212 | struct pglist_data *last_pgdat = NULL; |
4213 | bool last_pgdat_dirty_ok = false; | |
6bb15450 | 4214 | bool no_fallback; |
3b8c0be4 | 4215 | |
6bb15450 | 4216 | retry: |
7fb1d9fc | 4217 | /* |
9276b1bc | 4218 | * Scan zonelist, looking for a zone with enough free. |
189cdcfe | 4219 | * See also __cpuset_node_allowed() comment in kernel/cgroup/cpuset.c. |
7fb1d9fc | 4220 | */ |
6bb15450 MG |
4221 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
4222 | z = ac->preferred_zoneref; | |
30d8ec73 MN |
4223 | for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx, |
4224 | ac->nodemask) { | |
be06af00 | 4225 | struct page *page; |
e085dbc5 JW |
4226 | unsigned long mark; |
4227 | ||
664eedde MG |
4228 | if (cpusets_enabled() && |
4229 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 4230 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 4231 | continue; |
a756cf59 JW |
4232 | /* |
4233 | * When allocating a page cache page for writing, we | |
281e3726 MG |
4234 | * want to get it from a node that is within its dirty |
4235 | * limit, such that no single node holds more than its | |
a756cf59 | 4236 | * proportional share of globally allowed dirty pages. |
281e3726 | 4237 | * The dirty limits take into account the node's |
a756cf59 JW |
4238 | * lowmem reserves and high watermark so that kswapd |
4239 | * should be able to balance it without having to | |
4240 | * write pages from its LRU list. | |
4241 | * | |
a756cf59 | 4242 | * XXX: For now, allow allocations to potentially |
281e3726 | 4243 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 4244 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 4245 | * which is important when on a NUMA setup the allowed |
281e3726 | 4246 | * nodes are together not big enough to reach the |
a756cf59 | 4247 | * global limit. The proper fix for these situations |
281e3726 | 4248 | * will require awareness of nodes in the |
a756cf59 JW |
4249 | * dirty-throttling and the flusher threads. |
4250 | */ | |
3b8c0be4 | 4251 | if (ac->spread_dirty_pages) { |
8a87d695 WY |
4252 | if (last_pgdat != zone->zone_pgdat) { |
4253 | last_pgdat = zone->zone_pgdat; | |
4254 | last_pgdat_dirty_ok = node_dirty_ok(zone->zone_pgdat); | |
4255 | } | |
3b8c0be4 | 4256 | |
8a87d695 | 4257 | if (!last_pgdat_dirty_ok) |
3b8c0be4 | 4258 | continue; |
3b8c0be4 | 4259 | } |
7fb1d9fc | 4260 | |
6bb15450 MG |
4261 | if (no_fallback && nr_online_nodes > 1 && |
4262 | zone != ac->preferred_zoneref->zone) { | |
4263 | int local_nid; | |
4264 | ||
4265 | /* | |
4266 | * If moving to a remote node, retry but allow | |
4267 | * fragmenting fallbacks. Locality is more important | |
4268 | * than fragmentation avoidance. | |
4269 | */ | |
4270 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
4271 | if (zone_to_nid(zone) != local_nid) { | |
4272 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
4273 | goto retry; | |
4274 | } | |
4275 | } | |
4276 | ||
a9214443 | 4277 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 4278 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
4279 | ac->highest_zoneidx, alloc_flags, |
4280 | gfp_mask)) { | |
fa5e084e MG |
4281 | int ret; |
4282 | ||
c9e97a19 PT |
4283 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
4284 | /* | |
4285 | * Watermark failed for this zone, but see if we can | |
4286 | * grow this zone if it contains deferred pages. | |
4287 | */ | |
4288 | if (static_branch_unlikely(&deferred_pages)) { | |
4289 | if (_deferred_grow_zone(zone, order)) | |
4290 | goto try_this_zone; | |
4291 | } | |
4292 | #endif | |
5dab2911 MG |
4293 | /* Checked here to keep the fast path fast */ |
4294 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
4295 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
4296 | goto try_this_zone; | |
4297 | ||
202e35db | 4298 | if (!node_reclaim_enabled() || |
c33d6c06 | 4299 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
4300 | continue; |
4301 | ||
a5f5f91d | 4302 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 4303 | switch (ret) { |
a5f5f91d | 4304 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 4305 | /* did not scan */ |
cd38b115 | 4306 | continue; |
a5f5f91d | 4307 | case NODE_RECLAIM_FULL: |
fa5e084e | 4308 | /* scanned but unreclaimable */ |
cd38b115 | 4309 | continue; |
fa5e084e MG |
4310 | default: |
4311 | /* did we reclaim enough */ | |
fed2719e | 4312 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 4313 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
4314 | goto try_this_zone; |
4315 | ||
fed2719e | 4316 | continue; |
0798e519 | 4317 | } |
7fb1d9fc RS |
4318 | } |
4319 | ||
fa5e084e | 4320 | try_this_zone: |
066b2393 | 4321 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 4322 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 4323 | if (page) { |
479f854a | 4324 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
4325 | |
4326 | /* | |
4327 | * If this is a high-order atomic allocation then check | |
4328 | * if the pageblock should be reserved for the future | |
4329 | */ | |
eb2e2b42 | 4330 | if (unlikely(alloc_flags & ALLOC_HIGHATOMIC)) |
0aaa29a5 MG |
4331 | reserve_highatomic_pageblock(page, zone, order); |
4332 | ||
75379191 | 4333 | return page; |
c9e97a19 PT |
4334 | } else { |
4335 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
4336 | /* Try again if zone has deferred pages */ | |
4337 | if (static_branch_unlikely(&deferred_pages)) { | |
4338 | if (_deferred_grow_zone(zone, order)) | |
4339 | goto try_this_zone; | |
4340 | } | |
4341 | #endif | |
75379191 | 4342 | } |
54a6eb5c | 4343 | } |
9276b1bc | 4344 | |
6bb15450 MG |
4345 | /* |
4346 | * It's possible on a UMA machine to get through all zones that are | |
4347 | * fragmented. If avoiding fragmentation, reset and try again. | |
4348 | */ | |
4349 | if (no_fallback) { | |
4350 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
4351 | goto retry; | |
4352 | } | |
4353 | ||
4ffeaf35 | 4354 | return NULL; |
753ee728 MH |
4355 | } |
4356 | ||
9af744d7 | 4357 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 4358 | { |
a238ab5b | 4359 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
4360 | |
4361 | /* | |
4362 | * This documents exceptions given to allocations in certain | |
4363 | * contexts that are allowed to allocate outside current's set | |
4364 | * of allowed nodes. | |
4365 | */ | |
4366 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 4367 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
4368 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
4369 | filter &= ~SHOW_MEM_FILTER_NODES; | |
88dc6f20 | 4370 | if (!in_task() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
4371 | filter &= ~SHOW_MEM_FILTER_NODES; |
4372 | ||
974f4367 | 4373 | __show_mem(filter, nodemask, gfp_zone(gfp_mask)); |
aa187507 MH |
4374 | } |
4375 | ||
a8e99259 | 4376 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
4377 | { |
4378 | struct va_format vaf; | |
4379 | va_list args; | |
1be334e5 | 4380 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 4381 | |
c4dc63f0 BH |
4382 | if ((gfp_mask & __GFP_NOWARN) || |
4383 | !__ratelimit(&nopage_rs) || | |
4384 | ((gfp_mask & __GFP_DMA) && !has_managed_dma())) | |
aa187507 MH |
4385 | return; |
4386 | ||
7877cdcc MH |
4387 | va_start(args, fmt); |
4388 | vaf.fmt = fmt; | |
4389 | vaf.va = &args; | |
ef8444ea | 4390 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
4391 | current->comm, &vaf, gfp_mask, &gfp_mask, |
4392 | nodemask_pr_args(nodemask)); | |
7877cdcc | 4393 | va_end(args); |
3ee9a4f0 | 4394 | |
a8e99259 | 4395 | cpuset_print_current_mems_allowed(); |
ef8444ea | 4396 | pr_cont("\n"); |
a238ab5b | 4397 | dump_stack(); |
685dbf6f | 4398 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
4399 | } |
4400 | ||
6c18ba7a MH |
4401 | static inline struct page * |
4402 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
4403 | unsigned int alloc_flags, | |
4404 | const struct alloc_context *ac) | |
4405 | { | |
4406 | struct page *page; | |
4407 | ||
4408 | page = get_page_from_freelist(gfp_mask, order, | |
4409 | alloc_flags|ALLOC_CPUSET, ac); | |
4410 | /* | |
4411 | * fallback to ignore cpuset restriction if our nodes | |
4412 | * are depleted | |
4413 | */ | |
4414 | if (!page) | |
4415 | page = get_page_from_freelist(gfp_mask, order, | |
4416 | alloc_flags, ac); | |
4417 | ||
4418 | return page; | |
4419 | } | |
4420 | ||
11e33f6a MG |
4421 | static inline struct page * |
4422 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4423 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 4424 | { |
6e0fc46d DR |
4425 | struct oom_control oc = { |
4426 | .zonelist = ac->zonelist, | |
4427 | .nodemask = ac->nodemask, | |
2a966b77 | 4428 | .memcg = NULL, |
6e0fc46d DR |
4429 | .gfp_mask = gfp_mask, |
4430 | .order = order, | |
6e0fc46d | 4431 | }; |
11e33f6a MG |
4432 | struct page *page; |
4433 | ||
9879de73 JW |
4434 | *did_some_progress = 0; |
4435 | ||
9879de73 | 4436 | /* |
dc56401f JW |
4437 | * Acquire the oom lock. If that fails, somebody else is |
4438 | * making progress for us. | |
9879de73 | 4439 | */ |
dc56401f | 4440 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 4441 | *did_some_progress = 1; |
11e33f6a | 4442 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
4443 | return NULL; |
4444 | } | |
6b1de916 | 4445 | |
11e33f6a MG |
4446 | /* |
4447 | * Go through the zonelist yet one more time, keep very high watermark | |
4448 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
4449 | * we're still under heavy pressure. But make sure that this reclaim |
4450 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
4451 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 4452 | */ |
e746bf73 TH |
4453 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
4454 | ~__GFP_DIRECT_RECLAIM, order, | |
4455 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 4456 | if (page) |
11e33f6a MG |
4457 | goto out; |
4458 | ||
06ad276a MH |
4459 | /* Coredumps can quickly deplete all memory reserves */ |
4460 | if (current->flags & PF_DUMPCORE) | |
4461 | goto out; | |
4462 | /* The OOM killer will not help higher order allocs */ | |
4463 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4464 | goto out; | |
dcda9b04 MH |
4465 | /* |
4466 | * We have already exhausted all our reclaim opportunities without any | |
4467 | * success so it is time to admit defeat. We will skip the OOM killer | |
4468 | * because it is very likely that the caller has a more reasonable | |
4469 | * fallback than shooting a random task. | |
cfb4a541 MN |
4470 | * |
4471 | * The OOM killer may not free memory on a specific node. | |
dcda9b04 | 4472 | */ |
cfb4a541 | 4473 | if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE)) |
dcda9b04 | 4474 | goto out; |
06ad276a | 4475 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 4476 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
4477 | goto out; |
4478 | if (pm_suspended_storage()) | |
4479 | goto out; | |
4480 | /* | |
4481 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
4482 | * other request to make a forward progress. | |
4483 | * We are in an unfortunate situation where out_of_memory cannot | |
4484 | * do much for this context but let's try it to at least get | |
4485 | * access to memory reserved if the current task is killed (see | |
4486 | * out_of_memory). Once filesystems are ready to handle allocation | |
4487 | * failures more gracefully we should just bail out here. | |
4488 | */ | |
4489 | ||
3c2c6488 | 4490 | /* Exhausted what can be done so it's blame time */ |
3f913fc5 QZ |
4491 | if (out_of_memory(&oc) || |
4492 | WARN_ON_ONCE_GFP(gfp_mask & __GFP_NOFAIL, gfp_mask)) { | |
c32b3cbe | 4493 | *did_some_progress = 1; |
5020e285 | 4494 | |
6c18ba7a MH |
4495 | /* |
4496 | * Help non-failing allocations by giving them access to memory | |
4497 | * reserves | |
4498 | */ | |
4499 | if (gfp_mask & __GFP_NOFAIL) | |
4500 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 4501 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 4502 | } |
11e33f6a | 4503 | out: |
dc56401f | 4504 | mutex_unlock(&oom_lock); |
11e33f6a MG |
4505 | return page; |
4506 | } | |
4507 | ||
33c2d214 | 4508 | /* |
baf2f90b | 4509 | * Maximum number of compaction retries with a progress before OOM |
33c2d214 MH |
4510 | * killer is consider as the only way to move forward. |
4511 | */ | |
4512 | #define MAX_COMPACT_RETRIES 16 | |
4513 | ||
56de7263 MG |
4514 | #ifdef CONFIG_COMPACTION |
4515 | /* Try memory compaction for high-order allocations before reclaim */ | |
4516 | static struct page * | |
4517 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4518 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4519 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4520 | { |
5e1f0f09 | 4521 | struct page *page = NULL; |
eb414681 | 4522 | unsigned long pflags; |
499118e9 | 4523 | unsigned int noreclaim_flag; |
53853e2d VB |
4524 | |
4525 | if (!order) | |
66199712 | 4526 | return NULL; |
66199712 | 4527 | |
eb414681 | 4528 | psi_memstall_enter(&pflags); |
5bf18281 | 4529 | delayacct_compact_start(); |
499118e9 | 4530 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 4531 | |
c5d01d0d | 4532 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 4533 | prio, &page); |
eb414681 | 4534 | |
499118e9 | 4535 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 4536 | psi_memstall_leave(&pflags); |
5bf18281 | 4537 | delayacct_compact_end(); |
56de7263 | 4538 | |
06dac2f4 CTR |
4539 | if (*compact_result == COMPACT_SKIPPED) |
4540 | return NULL; | |
98dd3b48 VB |
4541 | /* |
4542 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
4543 | * count a compaction stall | |
4544 | */ | |
4545 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 4546 | |
5e1f0f09 MG |
4547 | /* Prep a captured page if available */ |
4548 | if (page) | |
4549 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
4550 | ||
4551 | /* Try get a page from the freelist if available */ | |
4552 | if (!page) | |
4553 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 4554 | |
98dd3b48 VB |
4555 | if (page) { |
4556 | struct zone *zone = page_zone(page); | |
53853e2d | 4557 | |
98dd3b48 VB |
4558 | zone->compact_blockskip_flush = false; |
4559 | compaction_defer_reset(zone, order, true); | |
4560 | count_vm_event(COMPACTSUCCESS); | |
4561 | return page; | |
4562 | } | |
56de7263 | 4563 | |
98dd3b48 VB |
4564 | /* |
4565 | * It's bad if compaction run occurs and fails. The most likely reason | |
4566 | * is that pages exist, but not enough to satisfy watermarks. | |
4567 | */ | |
4568 | count_vm_event(COMPACTFAIL); | |
66199712 | 4569 | |
98dd3b48 | 4570 | cond_resched(); |
56de7263 MG |
4571 | |
4572 | return NULL; | |
4573 | } | |
33c2d214 | 4574 | |
3250845d VB |
4575 | static inline bool |
4576 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
4577 | enum compact_result compact_result, | |
4578 | enum compact_priority *compact_priority, | |
d9436498 | 4579 | int *compaction_retries) |
3250845d VB |
4580 | { |
4581 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 4582 | int min_priority; |
65190cff MH |
4583 | bool ret = false; |
4584 | int retries = *compaction_retries; | |
4585 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
4586 | |
4587 | if (!order) | |
4588 | return false; | |
4589 | ||
691d9497 AT |
4590 | if (fatal_signal_pending(current)) |
4591 | return false; | |
4592 | ||
d9436498 VB |
4593 | if (compaction_made_progress(compact_result)) |
4594 | (*compaction_retries)++; | |
4595 | ||
3250845d VB |
4596 | /* |
4597 | * compaction considers all the zone as desperately out of memory | |
4598 | * so it doesn't really make much sense to retry except when the | |
4599 | * failure could be caused by insufficient priority | |
4600 | */ | |
d9436498 VB |
4601 | if (compaction_failed(compact_result)) |
4602 | goto check_priority; | |
3250845d | 4603 | |
49433085 VB |
4604 | /* |
4605 | * compaction was skipped because there are not enough order-0 pages | |
4606 | * to work with, so we retry only if it looks like reclaim can help. | |
4607 | */ | |
4608 | if (compaction_needs_reclaim(compact_result)) { | |
4609 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
4610 | goto out; | |
4611 | } | |
4612 | ||
3250845d VB |
4613 | /* |
4614 | * make sure the compaction wasn't deferred or didn't bail out early | |
4615 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
4616 | * But the next retry should use a higher priority if allowed, so |
4617 | * we don't just keep bailing out endlessly. | |
3250845d | 4618 | */ |
65190cff | 4619 | if (compaction_withdrawn(compact_result)) { |
49433085 | 4620 | goto check_priority; |
65190cff | 4621 | } |
3250845d VB |
4622 | |
4623 | /* | |
dcda9b04 | 4624 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
4625 | * costly ones because they are de facto nofail and invoke OOM |
4626 | * killer to move on while costly can fail and users are ready | |
4627 | * to cope with that. 1/4 retries is rather arbitrary but we | |
4628 | * would need much more detailed feedback from compaction to | |
4629 | * make a better decision. | |
4630 | */ | |
4631 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4632 | max_retries /= 4; | |
65190cff MH |
4633 | if (*compaction_retries <= max_retries) { |
4634 | ret = true; | |
4635 | goto out; | |
4636 | } | |
3250845d | 4637 | |
d9436498 VB |
4638 | /* |
4639 | * Make sure there are attempts at the highest priority if we exhausted | |
4640 | * all retries or failed at the lower priorities. | |
4641 | */ | |
4642 | check_priority: | |
c2033b00 VB |
4643 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
4644 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 4645 | |
c2033b00 | 4646 | if (*compact_priority > min_priority) { |
d9436498 VB |
4647 | (*compact_priority)--; |
4648 | *compaction_retries = 0; | |
65190cff | 4649 | ret = true; |
d9436498 | 4650 | } |
65190cff MH |
4651 | out: |
4652 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4653 | return ret; | |
3250845d | 4654 | } |
56de7263 MG |
4655 | #else |
4656 | static inline struct page * | |
4657 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4658 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4659 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4660 | { |
33c2d214 | 4661 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4662 | return NULL; |
4663 | } | |
33c2d214 MH |
4664 | |
4665 | static inline bool | |
86a294a8 MH |
4666 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4667 | enum compact_result compact_result, | |
a5508cd8 | 4668 | enum compact_priority *compact_priority, |
d9436498 | 4669 | int *compaction_retries) |
33c2d214 | 4670 | { |
31e49bfd MH |
4671 | struct zone *zone; |
4672 | struct zoneref *z; | |
4673 | ||
4674 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4675 | return false; | |
4676 | ||
4677 | /* | |
4678 | * There are setups with compaction disabled which would prefer to loop | |
4679 | * inside the allocator rather than hit the oom killer prematurely. | |
4680 | * Let's give them a good hope and keep retrying while the order-0 | |
4681 | * watermarks are OK. | |
4682 | */ | |
97a225e6 JK |
4683 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4684 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 4685 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 4686 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
4687 | return true; |
4688 | } | |
33c2d214 MH |
4689 | return false; |
4690 | } | |
3250845d | 4691 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4692 | |
d92a8cfc | 4693 | #ifdef CONFIG_LOCKDEP |
93781325 | 4694 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4695 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4696 | ||
f920e413 | 4697 | static bool __need_reclaim(gfp_t gfp_mask) |
d92a8cfc | 4698 | { |
d92a8cfc PZ |
4699 | /* no reclaim without waiting on it */ |
4700 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4701 | return false; | |
4702 | ||
4703 | /* this guy won't enter reclaim */ | |
2e517d68 | 4704 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4705 | return false; |
4706 | ||
d92a8cfc PZ |
4707 | if (gfp_mask & __GFP_NOLOCKDEP) |
4708 | return false; | |
4709 | ||
4710 | return true; | |
4711 | } | |
4712 | ||
4f3eaf45 | 4713 | void __fs_reclaim_acquire(unsigned long ip) |
93781325 | 4714 | { |
4f3eaf45 | 4715 | lock_acquire_exclusive(&__fs_reclaim_map, 0, 0, NULL, ip); |
93781325 OS |
4716 | } |
4717 | ||
4f3eaf45 | 4718 | void __fs_reclaim_release(unsigned long ip) |
93781325 | 4719 | { |
4f3eaf45 | 4720 | lock_release(&__fs_reclaim_map, ip); |
93781325 OS |
4721 | } |
4722 | ||
d92a8cfc PZ |
4723 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4724 | { | |
f920e413 DV |
4725 | gfp_mask = current_gfp_context(gfp_mask); |
4726 | ||
4727 | if (__need_reclaim(gfp_mask)) { | |
4728 | if (gfp_mask & __GFP_FS) | |
4f3eaf45 | 4729 | __fs_reclaim_acquire(_RET_IP_); |
f920e413 DV |
4730 | |
4731 | #ifdef CONFIG_MMU_NOTIFIER | |
4732 | lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); | |
4733 | lock_map_release(&__mmu_notifier_invalidate_range_start_map); | |
4734 | #endif | |
4735 | ||
4736 | } | |
d92a8cfc PZ |
4737 | } |
4738 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4739 | ||
4740 | void fs_reclaim_release(gfp_t gfp_mask) | |
4741 | { | |
f920e413 DV |
4742 | gfp_mask = current_gfp_context(gfp_mask); |
4743 | ||
4744 | if (__need_reclaim(gfp_mask)) { | |
4745 | if (gfp_mask & __GFP_FS) | |
4f3eaf45 | 4746 | __fs_reclaim_release(_RET_IP_); |
f920e413 | 4747 | } |
d92a8cfc PZ |
4748 | } |
4749 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4750 | #endif | |
4751 | ||
3d36424b MG |
4752 | /* |
4753 | * Zonelists may change due to hotplug during allocation. Detect when zonelists | |
4754 | * have been rebuilt so allocation retries. Reader side does not lock and | |
4755 | * retries the allocation if zonelist changes. Writer side is protected by the | |
4756 | * embedded spin_lock. | |
4757 | */ | |
4758 | static DEFINE_SEQLOCK(zonelist_update_seq); | |
4759 | ||
4760 | static unsigned int zonelist_iter_begin(void) | |
4761 | { | |
4762 | if (IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) | |
4763 | return read_seqbegin(&zonelist_update_seq); | |
4764 | ||
4765 | return 0; | |
4766 | } | |
4767 | ||
4768 | static unsigned int check_retry_zonelist(unsigned int seq) | |
4769 | { | |
4770 | if (IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) | |
4771 | return read_seqretry(&zonelist_update_seq, seq); | |
4772 | ||
4773 | return seq; | |
4774 | } | |
4775 | ||
bba90710 | 4776 | /* Perform direct synchronous page reclaim */ |
2187e17b | 4777 | static unsigned long |
a9263751 VB |
4778 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4779 | const struct alloc_context *ac) | |
11e33f6a | 4780 | { |
499118e9 | 4781 | unsigned int noreclaim_flag; |
fa7fc75f | 4782 | unsigned long progress; |
11e33f6a MG |
4783 | |
4784 | cond_resched(); | |
4785 | ||
4786 | /* We now go into synchronous reclaim */ | |
4787 | cpuset_memory_pressure_bump(); | |
d92a8cfc | 4788 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4789 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4790 | |
a9263751 VB |
4791 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4792 | ac->nodemask); | |
11e33f6a | 4793 | |
499118e9 | 4794 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4795 | fs_reclaim_release(gfp_mask); |
11e33f6a MG |
4796 | |
4797 | cond_resched(); | |
4798 | ||
bba90710 MS |
4799 | return progress; |
4800 | } | |
4801 | ||
4802 | /* The really slow allocator path where we enter direct reclaim */ | |
4803 | static inline struct page * | |
4804 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4805 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4806 | unsigned long *did_some_progress) |
bba90710 MS |
4807 | { |
4808 | struct page *page = NULL; | |
fa7fc75f | 4809 | unsigned long pflags; |
bba90710 MS |
4810 | bool drained = false; |
4811 | ||
fa7fc75f | 4812 | psi_memstall_enter(&pflags); |
a9263751 | 4813 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce | 4814 | if (unlikely(!(*did_some_progress))) |
fa7fc75f | 4815 | goto out; |
11e33f6a | 4816 | |
9ee493ce | 4817 | retry: |
31a6c190 | 4818 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4819 | |
4820 | /* | |
4821 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 | 4822 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
047b9967 | 4823 | * Shrink them and try again |
9ee493ce MG |
4824 | */ |
4825 | if (!page && !drained) { | |
29fac03b | 4826 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4827 | drain_all_pages(NULL); |
9ee493ce MG |
4828 | drained = true; |
4829 | goto retry; | |
4830 | } | |
fa7fc75f SB |
4831 | out: |
4832 | psi_memstall_leave(&pflags); | |
9ee493ce | 4833 | |
11e33f6a MG |
4834 | return page; |
4835 | } | |
4836 | ||
5ecd9d40 DR |
4837 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4838 | const struct alloc_context *ac) | |
3a025760 JW |
4839 | { |
4840 | struct zoneref *z; | |
4841 | struct zone *zone; | |
e1a55637 | 4842 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 4843 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 4844 | |
97a225e6 | 4845 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 4846 | ac->nodemask) { |
bc53008e WY |
4847 | if (!managed_zone(zone)) |
4848 | continue; | |
d137a7cb | 4849 | if (last_pgdat != zone->zone_pgdat) { |
97a225e6 | 4850 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
d137a7cb CW |
4851 | last_pgdat = zone->zone_pgdat; |
4852 | } | |
e1a55637 | 4853 | } |
3a025760 JW |
4854 | } |
4855 | ||
c603844b | 4856 | static inline unsigned int |
eb2e2b42 | 4857 | gfp_to_alloc_flags(gfp_t gfp_mask, unsigned int order) |
341ce06f | 4858 | { |
c603844b | 4859 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4860 | |
736838e9 | 4861 | /* |
524c4807 | 4862 | * __GFP_HIGH is assumed to be the same as ALLOC_MIN_RESERVE |
736838e9 MN |
4863 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD |
4864 | * to save two branches. | |
4865 | */ | |
524c4807 | 4866 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_MIN_RESERVE); |
736838e9 | 4867 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 4868 | |
341ce06f PZ |
4869 | /* |
4870 | * The caller may dip into page reserves a bit more if the caller | |
4871 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4872 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1ebbb218 | 4873 | * set both ALLOC_NON_BLOCK and ALLOC_MIN_RESERVE(__GFP_HIGH). |
341ce06f | 4874 | */ |
736838e9 MN |
4875 | alloc_flags |= (__force int) |
4876 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 4877 | |
1ebbb218 | 4878 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) { |
5c3240d9 | 4879 | /* |
b104a35d DR |
4880 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4881 | * if it can't schedule. | |
5c3240d9 | 4882 | */ |
eb2e2b42 | 4883 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { |
1ebbb218 | 4884 | alloc_flags |= ALLOC_NON_BLOCK; |
eb2e2b42 MG |
4885 | |
4886 | if (order > 0) | |
4887 | alloc_flags |= ALLOC_HIGHATOMIC; | |
4888 | } | |
4889 | ||
523b9458 | 4890 | /* |
1ebbb218 MG |
4891 | * Ignore cpuset mems for non-blocking __GFP_HIGH (probably |
4892 | * GFP_ATOMIC) rather than fail, see the comment for | |
4893 | * __cpuset_node_allowed(). | |
523b9458 | 4894 | */ |
1ebbb218 MG |
4895 | if (alloc_flags & ALLOC_MIN_RESERVE) |
4896 | alloc_flags &= ~ALLOC_CPUSET; | |
88dc6f20 | 4897 | } else if (unlikely(rt_task(current)) && in_task()) |
c988dcbe | 4898 | alloc_flags |= ALLOC_MIN_RESERVE; |
341ce06f | 4899 | |
8e3560d9 | 4900 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags); |
8510e69c | 4901 | |
341ce06f PZ |
4902 | return alloc_flags; |
4903 | } | |
4904 | ||
cd04ae1e | 4905 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4906 | { |
cd04ae1e MH |
4907 | if (!tsk_is_oom_victim(tsk)) |
4908 | return false; | |
4909 | ||
4910 | /* | |
4911 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4912 | * only to the thread with TIF_MEMDIE set | |
4913 | */ | |
4914 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4915 | return false; |
4916 | ||
cd04ae1e MH |
4917 | return true; |
4918 | } | |
4919 | ||
4920 | /* | |
4921 | * Distinguish requests which really need access to full memory | |
4922 | * reserves from oom victims which can live with a portion of it | |
4923 | */ | |
4924 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4925 | { | |
4926 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4927 | return 0; | |
31a6c190 | 4928 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4929 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4930 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4931 | return ALLOC_NO_WATERMARKS; |
4932 | if (!in_interrupt()) { | |
4933 | if (current->flags & PF_MEMALLOC) | |
4934 | return ALLOC_NO_WATERMARKS; | |
4935 | else if (oom_reserves_allowed(current)) | |
4936 | return ALLOC_OOM; | |
4937 | } | |
31a6c190 | 4938 | |
cd04ae1e MH |
4939 | return 0; |
4940 | } | |
4941 | ||
4942 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4943 | { | |
4944 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4945 | } |
4946 | ||
0a0337e0 MH |
4947 | /* |
4948 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4949 | * for the given allocation request. | |
491d79ae JW |
4950 | * |
4951 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4952 | * without success, or when we couldn't even meet the watermark if we | |
4953 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4954 | * |
4955 | * Returns true if a retry is viable or false to enter the oom path. | |
4956 | */ | |
4957 | static inline bool | |
4958 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4959 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4960 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4961 | { |
4962 | struct zone *zone; | |
4963 | struct zoneref *z; | |
15f570bf | 4964 | bool ret = false; |
0a0337e0 | 4965 | |
423b452e VB |
4966 | /* |
4967 | * Costly allocations might have made a progress but this doesn't mean | |
4968 | * their order will become available due to high fragmentation so | |
4969 | * always increment the no progress counter for them | |
4970 | */ | |
4971 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4972 | *no_progress_loops = 0; | |
4973 | else | |
4974 | (*no_progress_loops)++; | |
4975 | ||
0a0337e0 MH |
4976 | /* |
4977 | * Make sure we converge to OOM if we cannot make any progress | |
4978 | * several times in the row. | |
4979 | */ | |
04c8716f MK |
4980 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4981 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4982 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4983 | } |
0a0337e0 | 4984 | |
bca67592 MG |
4985 | /* |
4986 | * Keep reclaiming pages while there is a chance this will lead | |
4987 | * somewhere. If none of the target zones can satisfy our allocation | |
4988 | * request even if all reclaimable pages are considered then we are | |
4989 | * screwed and have to go OOM. | |
0a0337e0 | 4990 | */ |
97a225e6 JK |
4991 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4992 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4993 | unsigned long available; |
ede37713 | 4994 | unsigned long reclaimable; |
d379f01d MH |
4995 | unsigned long min_wmark = min_wmark_pages(zone); |
4996 | bool wmark; | |
0a0337e0 | 4997 | |
5a1c84b4 | 4998 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4999 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
5000 | |
5001 | /* | |
491d79ae JW |
5002 | * Would the allocation succeed if we reclaimed all |
5003 | * reclaimable pages? | |
0a0337e0 | 5004 | */ |
d379f01d | 5005 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 5006 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
5007 | trace_reclaim_retry_zone(z, order, reclaimable, |
5008 | available, min_wmark, *no_progress_loops, wmark); | |
5009 | if (wmark) { | |
15f570bf | 5010 | ret = true; |
132b0d21 | 5011 | break; |
0a0337e0 MH |
5012 | } |
5013 | } | |
5014 | ||
15f570bf MH |
5015 | /* |
5016 | * Memory allocation/reclaim might be called from a WQ context and the | |
5017 | * current implementation of the WQ concurrency control doesn't | |
5018 | * recognize that a particular WQ is congested if the worker thread is | |
5019 | * looping without ever sleeping. Therefore we have to do a short sleep | |
5020 | * here rather than calling cond_resched(). | |
5021 | */ | |
5022 | if (current->flags & PF_WQ_WORKER) | |
5023 | schedule_timeout_uninterruptible(1); | |
5024 | else | |
5025 | cond_resched(); | |
5026 | return ret; | |
0a0337e0 MH |
5027 | } |
5028 | ||
902b6281 VB |
5029 | static inline bool |
5030 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
5031 | { | |
5032 | /* | |
5033 | * It's possible that cpuset's mems_allowed and the nodemask from | |
5034 | * mempolicy don't intersect. This should be normally dealt with by | |
5035 | * policy_nodemask(), but it's possible to race with cpuset update in | |
5036 | * such a way the check therein was true, and then it became false | |
5037 | * before we got our cpuset_mems_cookie here. | |
5038 | * This assumes that for all allocations, ac->nodemask can come only | |
5039 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
5040 | * when it does not intersect with the cpuset restrictions) or the | |
5041 | * caller can deal with a violated nodemask. | |
5042 | */ | |
5043 | if (cpusets_enabled() && ac->nodemask && | |
5044 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
5045 | ac->nodemask = NULL; | |
5046 | return true; | |
5047 | } | |
5048 | ||
5049 | /* | |
5050 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
5051 | * possible to race with parallel threads in such a way that our | |
5052 | * allocation can fail while the mask is being updated. If we are about | |
5053 | * to fail, check if the cpuset changed during allocation and if so, | |
5054 | * retry. | |
5055 | */ | |
5056 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
5057 | return true; | |
5058 | ||
5059 | return false; | |
5060 | } | |
5061 | ||
11e33f6a MG |
5062 | static inline struct page * |
5063 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 5064 | struct alloc_context *ac) |
11e33f6a | 5065 | { |
d0164adc | 5066 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 5067 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 5068 | struct page *page = NULL; |
c603844b | 5069 | unsigned int alloc_flags; |
11e33f6a | 5070 | unsigned long did_some_progress; |
5ce9bfef | 5071 | enum compact_priority compact_priority; |
c5d01d0d | 5072 | enum compact_result compact_result; |
5ce9bfef VB |
5073 | int compaction_retries; |
5074 | int no_progress_loops; | |
5ce9bfef | 5075 | unsigned int cpuset_mems_cookie; |
3d36424b | 5076 | unsigned int zonelist_iter_cookie; |
cd04ae1e | 5077 | int reserve_flags; |
1da177e4 | 5078 | |
d0164adc MG |
5079 | /* |
5080 | * We also sanity check to catch abuse of atomic reserves being used by | |
5081 | * callers that are not in atomic context. | |
5082 | */ | |
5083 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
5084 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
5085 | gfp_mask &= ~__GFP_ATOMIC; | |
5086 | ||
3d36424b | 5087 | restart: |
5ce9bfef VB |
5088 | compaction_retries = 0; |
5089 | no_progress_loops = 0; | |
5090 | compact_priority = DEF_COMPACT_PRIORITY; | |
5091 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
3d36424b | 5092 | zonelist_iter_cookie = zonelist_iter_begin(); |
9a67f648 MH |
5093 | |
5094 | /* | |
5095 | * The fast path uses conservative alloc_flags to succeed only until | |
5096 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
5097 | * alloc_flags precisely. So we do that now. | |
5098 | */ | |
eb2e2b42 | 5099 | alloc_flags = gfp_to_alloc_flags(gfp_mask, order); |
9a67f648 | 5100 | |
e47483bc VB |
5101 | /* |
5102 | * We need to recalculate the starting point for the zonelist iterator | |
5103 | * because we might have used different nodemask in the fast path, or | |
5104 | * there was a cpuset modification and we are retrying - otherwise we | |
5105 | * could end up iterating over non-eligible zones endlessly. | |
5106 | */ | |
5107 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 5108 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
5109 | if (!ac->preferred_zoneref->zone) |
5110 | goto nopage; | |
5111 | ||
8ca1b5a4 FT |
5112 | /* |
5113 | * Check for insane configurations where the cpuset doesn't contain | |
5114 | * any suitable zone to satisfy the request - e.g. non-movable | |
5115 | * GFP_HIGHUSER allocations from MOVABLE nodes only. | |
5116 | */ | |
5117 | if (cpusets_insane_config() && (gfp_mask & __GFP_HARDWALL)) { | |
5118 | struct zoneref *z = first_zones_zonelist(ac->zonelist, | |
5119 | ac->highest_zoneidx, | |
5120 | &cpuset_current_mems_allowed); | |
5121 | if (!z->zone) | |
5122 | goto nopage; | |
5123 | } | |
5124 | ||
0a79cdad | 5125 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 5126 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
5127 | |
5128 | /* | |
5129 | * The adjusted alloc_flags might result in immediate success, so try | |
5130 | * that first | |
5131 | */ | |
5132 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
5133 | if (page) | |
5134 | goto got_pg; | |
5135 | ||
a8161d1e VB |
5136 | /* |
5137 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
5138 | * that we have enough base pages and don't need to reclaim. For non- |
5139 | * movable high-order allocations, do that as well, as compaction will | |
5140 | * try prevent permanent fragmentation by migrating from blocks of the | |
5141 | * same migratetype. | |
5142 | * Don't try this for allocations that are allowed to ignore | |
5143 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 5144 | */ |
282722b0 VB |
5145 | if (can_direct_reclaim && |
5146 | (costly_order || | |
5147 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
5148 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
5149 | page = __alloc_pages_direct_compact(gfp_mask, order, |
5150 | alloc_flags, ac, | |
a5508cd8 | 5151 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
5152 | &compact_result); |
5153 | if (page) | |
5154 | goto got_pg; | |
5155 | ||
cc638f32 VB |
5156 | /* |
5157 | * Checks for costly allocations with __GFP_NORETRY, which | |
5158 | * includes some THP page fault allocations | |
5159 | */ | |
5160 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
5161 | /* |
5162 | * If allocating entire pageblock(s) and compaction | |
5163 | * failed because all zones are below low watermarks | |
5164 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
5165 | * order, fail immediately unless the allocator has |
5166 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
5167 | * |
5168 | * Reclaim is | |
5169 | * - potentially very expensive because zones are far | |
5170 | * below their low watermarks or this is part of very | |
5171 | * bursty high order allocations, | |
5172 | * - not guaranteed to help because isolate_freepages() | |
5173 | * may not iterate over freed pages as part of its | |
5174 | * linear scan, and | |
5175 | * - unlikely to make entire pageblocks free on its | |
5176 | * own. | |
5177 | */ | |
5178 | if (compact_result == COMPACT_SKIPPED || | |
5179 | compact_result == COMPACT_DEFERRED) | |
5180 | goto nopage; | |
a8161d1e | 5181 | |
a8161d1e | 5182 | /* |
3eb2771b VB |
5183 | * Looks like reclaim/compaction is worth trying, but |
5184 | * sync compaction could be very expensive, so keep | |
25160354 | 5185 | * using async compaction. |
a8161d1e | 5186 | */ |
a5508cd8 | 5187 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
5188 | } |
5189 | } | |
23771235 | 5190 | |
31a6c190 | 5191 | retry: |
23771235 | 5192 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 5193 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 5194 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 5195 | |
cd04ae1e MH |
5196 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
5197 | if (reserve_flags) | |
ce96fa62 ML |
5198 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags) | |
5199 | (alloc_flags & ALLOC_KSWAPD); | |
23771235 | 5200 | |
e46e7b77 | 5201 | /* |
d6a24df0 VB |
5202 | * Reset the nodemask and zonelist iterators if memory policies can be |
5203 | * ignored. These allocations are high priority and system rather than | |
5204 | * user oriented. | |
e46e7b77 | 5205 | */ |
cd04ae1e | 5206 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 5207 | ac->nodemask = NULL; |
e46e7b77 | 5208 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 5209 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
5210 | } |
5211 | ||
23771235 | 5212 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 5213 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
5214 | if (page) |
5215 | goto got_pg; | |
1da177e4 | 5216 | |
d0164adc | 5217 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 5218 | if (!can_direct_reclaim) |
1da177e4 LT |
5219 | goto nopage; |
5220 | ||
9a67f648 MH |
5221 | /* Avoid recursion of direct reclaim */ |
5222 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
5223 | goto nopage; |
5224 | ||
a8161d1e VB |
5225 | /* Try direct reclaim and then allocating */ |
5226 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
5227 | &did_some_progress); | |
5228 | if (page) | |
5229 | goto got_pg; | |
5230 | ||
5231 | /* Try direct compaction and then allocating */ | |
a9263751 | 5232 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 5233 | compact_priority, &compact_result); |
56de7263 MG |
5234 | if (page) |
5235 | goto got_pg; | |
75f30861 | 5236 | |
9083905a JW |
5237 | /* Do not loop if specifically requested */ |
5238 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 5239 | goto nopage; |
9083905a | 5240 | |
0a0337e0 MH |
5241 | /* |
5242 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 5243 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 5244 | */ |
dcda9b04 | 5245 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 5246 | goto nopage; |
0a0337e0 | 5247 | |
0a0337e0 | 5248 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 5249 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
5250 | goto retry; |
5251 | ||
33c2d214 MH |
5252 | /* |
5253 | * It doesn't make any sense to retry for the compaction if the order-0 | |
5254 | * reclaim is not able to make any progress because the current | |
5255 | * implementation of the compaction depends on the sufficient amount | |
5256 | * of free memory (see __compaction_suitable) | |
5257 | */ | |
5258 | if (did_some_progress > 0 && | |
86a294a8 | 5259 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 5260 | compact_result, &compact_priority, |
d9436498 | 5261 | &compaction_retries)) |
33c2d214 MH |
5262 | goto retry; |
5263 | ||
902b6281 | 5264 | |
3d36424b MG |
5265 | /* |
5266 | * Deal with possible cpuset update races or zonelist updates to avoid | |
5267 | * a unnecessary OOM kill. | |
5268 | */ | |
5269 | if (check_retry_cpuset(cpuset_mems_cookie, ac) || | |
5270 | check_retry_zonelist(zonelist_iter_cookie)) | |
5271 | goto restart; | |
e47483bc | 5272 | |
9083905a JW |
5273 | /* Reclaim has failed us, start killing things */ |
5274 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
5275 | if (page) | |
5276 | goto got_pg; | |
5277 | ||
9a67f648 | 5278 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e | 5279 | if (tsk_is_oom_victim(current) && |
8510e69c | 5280 | (alloc_flags & ALLOC_OOM || |
c288983d | 5281 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
5282 | goto nopage; |
5283 | ||
9083905a | 5284 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
5285 | if (did_some_progress) { |
5286 | no_progress_loops = 0; | |
9083905a | 5287 | goto retry; |
0a0337e0 | 5288 | } |
9083905a | 5289 | |
1da177e4 | 5290 | nopage: |
3d36424b MG |
5291 | /* |
5292 | * Deal with possible cpuset update races or zonelist updates to avoid | |
5293 | * a unnecessary OOM kill. | |
5294 | */ | |
5295 | if (check_retry_cpuset(cpuset_mems_cookie, ac) || | |
5296 | check_retry_zonelist(zonelist_iter_cookie)) | |
5297 | goto restart; | |
5ce9bfef | 5298 | |
9a67f648 MH |
5299 | /* |
5300 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
5301 | * we always retry | |
5302 | */ | |
5303 | if (gfp_mask & __GFP_NOFAIL) { | |
5304 | /* | |
5305 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
5306 | * of any new users that actually require GFP_NOWAIT | |
5307 | */ | |
3f913fc5 | 5308 | if (WARN_ON_ONCE_GFP(!can_direct_reclaim, gfp_mask)) |
9a67f648 MH |
5309 | goto fail; |
5310 | ||
5311 | /* | |
5312 | * PF_MEMALLOC request from this context is rather bizarre | |
5313 | * because we cannot reclaim anything and only can loop waiting | |
5314 | * for somebody to do a work for us | |
5315 | */ | |
3f913fc5 | 5316 | WARN_ON_ONCE_GFP(current->flags & PF_MEMALLOC, gfp_mask); |
9a67f648 MH |
5317 | |
5318 | /* | |
5319 | * non failing costly orders are a hard requirement which we | |
5320 | * are not prepared for much so let's warn about these users | |
5321 | * so that we can identify them and convert them to something | |
5322 | * else. | |
5323 | */ | |
896c4d52 | 5324 | WARN_ON_ONCE_GFP(costly_order, gfp_mask); |
9a67f648 | 5325 | |
6c18ba7a | 5326 | /* |
1ebbb218 MG |
5327 | * Help non-failing allocations by giving some access to memory |
5328 | * reserves normally used for high priority non-blocking | |
5329 | * allocations but do not use ALLOC_NO_WATERMARKS because this | |
6c18ba7a | 5330 | * could deplete whole memory reserves which would just make |
1ebbb218 | 5331 | * the situation worse. |
6c18ba7a | 5332 | */ |
1ebbb218 | 5333 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_MIN_RESERVE, ac); |
6c18ba7a MH |
5334 | if (page) |
5335 | goto got_pg; | |
5336 | ||
9a67f648 MH |
5337 | cond_resched(); |
5338 | goto retry; | |
5339 | } | |
5340 | fail: | |
a8e99259 | 5341 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 5342 | "page allocation failure: order:%u", order); |
1da177e4 | 5343 | got_pg: |
072bb0aa | 5344 | return page; |
1da177e4 | 5345 | } |
11e33f6a | 5346 | |
9cd75558 | 5347 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 5348 | int preferred_nid, nodemask_t *nodemask, |
8e6a930b | 5349 | struct alloc_context *ac, gfp_t *alloc_gfp, |
9cd75558 | 5350 | unsigned int *alloc_flags) |
11e33f6a | 5351 | { |
97a225e6 | 5352 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 5353 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 5354 | ac->nodemask = nodemask; |
01c0bfe0 | 5355 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 5356 | |
682a3385 | 5357 | if (cpusets_enabled()) { |
8e6a930b | 5358 | *alloc_gfp |= __GFP_HARDWALL; |
182f3d7a MS |
5359 | /* |
5360 | * When we are in the interrupt context, it is irrelevant | |
5361 | * to the current task context. It means that any node ok. | |
5362 | */ | |
88dc6f20 | 5363 | if (in_task() && !ac->nodemask) |
9cd75558 | 5364 | ac->nodemask = &cpuset_current_mems_allowed; |
51047820 VB |
5365 | else |
5366 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
5367 | } |
5368 | ||
446ec838 | 5369 | might_alloc(gfp_mask); |
11e33f6a MG |
5370 | |
5371 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 5372 | return false; |
11e33f6a | 5373 | |
8e3560d9 | 5374 | *alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags); |
d883c6cf | 5375 | |
c9ab0c4f | 5376 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 5377 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 5378 | |
e46e7b77 MG |
5379 | /* |
5380 | * The preferred zone is used for statistics but crucially it is | |
5381 | * also used as the starting point for the zonelist iterator. It | |
5382 | * may get reset for allocations that ignore memory policies. | |
5383 | */ | |
9cd75558 | 5384 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 5385 | ac->highest_zoneidx, ac->nodemask); |
a0622d05 MN |
5386 | |
5387 | return true; | |
9cd75558 MG |
5388 | } |
5389 | ||
387ba26f | 5390 | /* |
0f87d9d3 | 5391 | * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array |
387ba26f MG |
5392 | * @gfp: GFP flags for the allocation |
5393 | * @preferred_nid: The preferred NUMA node ID to allocate from | |
5394 | * @nodemask: Set of nodes to allocate from, may be NULL | |
0f87d9d3 MG |
5395 | * @nr_pages: The number of pages desired on the list or array |
5396 | * @page_list: Optional list to store the allocated pages | |
5397 | * @page_array: Optional array to store the pages | |
387ba26f MG |
5398 | * |
5399 | * This is a batched version of the page allocator that attempts to | |
0f87d9d3 MG |
5400 | * allocate nr_pages quickly. Pages are added to page_list if page_list |
5401 | * is not NULL, otherwise it is assumed that the page_array is valid. | |
387ba26f | 5402 | * |
0f87d9d3 MG |
5403 | * For lists, nr_pages is the number of pages that should be allocated. |
5404 | * | |
5405 | * For arrays, only NULL elements are populated with pages and nr_pages | |
5406 | * is the maximum number of pages that will be stored in the array. | |
5407 | * | |
5408 | * Returns the number of pages on the list or array. | |
387ba26f MG |
5409 | */ |
5410 | unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, | |
5411 | nodemask_t *nodemask, int nr_pages, | |
0f87d9d3 MG |
5412 | struct list_head *page_list, |
5413 | struct page **page_array) | |
387ba26f MG |
5414 | { |
5415 | struct page *page; | |
4b23a68f | 5416 | unsigned long __maybe_unused UP_flags; |
387ba26f MG |
5417 | struct zone *zone; |
5418 | struct zoneref *z; | |
5419 | struct per_cpu_pages *pcp; | |
5420 | struct list_head *pcp_list; | |
5421 | struct alloc_context ac; | |
5422 | gfp_t alloc_gfp; | |
5423 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
3e23060b | 5424 | int nr_populated = 0, nr_account = 0; |
387ba26f | 5425 | |
0f87d9d3 MG |
5426 | /* |
5427 | * Skip populated array elements to determine if any pages need | |
5428 | * to be allocated before disabling IRQs. | |
5429 | */ | |
b08e50dd | 5430 | while (page_array && nr_populated < nr_pages && page_array[nr_populated]) |
0f87d9d3 MG |
5431 | nr_populated++; |
5432 | ||
06147843 CL |
5433 | /* No pages requested? */ |
5434 | if (unlikely(nr_pages <= 0)) | |
5435 | goto out; | |
5436 | ||
b3b64ebd MG |
5437 | /* Already populated array? */ |
5438 | if (unlikely(page_array && nr_pages - nr_populated == 0)) | |
06147843 | 5439 | goto out; |
b3b64ebd | 5440 | |
8dcb3060 SB |
5441 | /* Bulk allocator does not support memcg accounting. */ |
5442 | if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT)) | |
5443 | goto failed; | |
5444 | ||
387ba26f | 5445 | /* Use the single page allocator for one page. */ |
0f87d9d3 | 5446 | if (nr_pages - nr_populated == 1) |
387ba26f MG |
5447 | goto failed; |
5448 | ||
187ad460 MG |
5449 | #ifdef CONFIG_PAGE_OWNER |
5450 | /* | |
5451 | * PAGE_OWNER may recurse into the allocator to allocate space to | |
5452 | * save the stack with pagesets.lock held. Releasing/reacquiring | |
5453 | * removes much of the performance benefit of bulk allocation so | |
5454 | * force the caller to allocate one page at a time as it'll have | |
5455 | * similar performance to added complexity to the bulk allocator. | |
5456 | */ | |
5457 | if (static_branch_unlikely(&page_owner_inited)) | |
5458 | goto failed; | |
5459 | #endif | |
5460 | ||
387ba26f MG |
5461 | /* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */ |
5462 | gfp &= gfp_allowed_mask; | |
5463 | alloc_gfp = gfp; | |
5464 | if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags)) | |
06147843 | 5465 | goto out; |
387ba26f MG |
5466 | gfp = alloc_gfp; |
5467 | ||
5468 | /* Find an allowed local zone that meets the low watermark. */ | |
5469 | for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) { | |
5470 | unsigned long mark; | |
5471 | ||
5472 | if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) && | |
5473 | !__cpuset_zone_allowed(zone, gfp)) { | |
5474 | continue; | |
5475 | } | |
5476 | ||
5477 | if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone && | |
5478 | zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) { | |
5479 | goto failed; | |
5480 | } | |
5481 | ||
5482 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages; | |
5483 | if (zone_watermark_fast(zone, 0, mark, | |
5484 | zonelist_zone_idx(ac.preferred_zoneref), | |
5485 | alloc_flags, gfp)) { | |
5486 | break; | |
5487 | } | |
5488 | } | |
5489 | ||
5490 | /* | |
5491 | * If there are no allowed local zones that meets the watermarks then | |
5492 | * try to allocate a single page and reclaim if necessary. | |
5493 | */ | |
ce76f9a1 | 5494 | if (unlikely(!zone)) |
387ba26f MG |
5495 | goto failed; |
5496 | ||
57490774 | 5497 | /* spin_trylock may fail due to a parallel drain or IRQ reentrancy. */ |
4b23a68f | 5498 | pcp_trylock_prepare(UP_flags); |
57490774 | 5499 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 5500 | if (!pcp) |
4b23a68f | 5501 | goto failed_irq; |
387ba26f | 5502 | |
387ba26f | 5503 | /* Attempt the batch allocation */ |
44042b44 | 5504 | pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)]; |
0f87d9d3 MG |
5505 | while (nr_populated < nr_pages) { |
5506 | ||
5507 | /* Skip existing pages */ | |
5508 | if (page_array && page_array[nr_populated]) { | |
5509 | nr_populated++; | |
5510 | continue; | |
5511 | } | |
5512 | ||
44042b44 | 5513 | page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags, |
387ba26f | 5514 | pcp, pcp_list); |
ce76f9a1 | 5515 | if (unlikely(!page)) { |
c572e488 | 5516 | /* Try and allocate at least one page */ |
4b23a68f | 5517 | if (!nr_account) { |
57490774 | 5518 | pcp_spin_unlock(pcp); |
387ba26f | 5519 | goto failed_irq; |
4b23a68f | 5520 | } |
387ba26f MG |
5521 | break; |
5522 | } | |
3e23060b | 5523 | nr_account++; |
387ba26f MG |
5524 | |
5525 | prep_new_page(page, 0, gfp, 0); | |
0f87d9d3 MG |
5526 | if (page_list) |
5527 | list_add(&page->lru, page_list); | |
5528 | else | |
5529 | page_array[nr_populated] = page; | |
5530 | nr_populated++; | |
387ba26f MG |
5531 | } |
5532 | ||
57490774 | 5533 | pcp_spin_unlock(pcp); |
4b23a68f | 5534 | pcp_trylock_finish(UP_flags); |
43c95bcc | 5535 | |
3e23060b MG |
5536 | __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account); |
5537 | zone_statistics(ac.preferred_zoneref->zone, zone, nr_account); | |
387ba26f | 5538 | |
06147843 | 5539 | out: |
0f87d9d3 | 5540 | return nr_populated; |
387ba26f MG |
5541 | |
5542 | failed_irq: | |
4b23a68f | 5543 | pcp_trylock_finish(UP_flags); |
387ba26f MG |
5544 | |
5545 | failed: | |
5546 | page = __alloc_pages(gfp, 0, preferred_nid, nodemask); | |
5547 | if (page) { | |
0f87d9d3 MG |
5548 | if (page_list) |
5549 | list_add(&page->lru, page_list); | |
5550 | else | |
5551 | page_array[nr_populated] = page; | |
5552 | nr_populated++; | |
387ba26f MG |
5553 | } |
5554 | ||
06147843 | 5555 | goto out; |
387ba26f MG |
5556 | } |
5557 | EXPORT_SYMBOL_GPL(__alloc_pages_bulk); | |
5558 | ||
9cd75558 MG |
5559 | /* |
5560 | * This is the 'heart' of the zoned buddy allocator. | |
5561 | */ | |
84172f4b | 5562 | struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, |
04ec6264 | 5563 | nodemask_t *nodemask) |
9cd75558 MG |
5564 | { |
5565 | struct page *page; | |
5566 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
8e6a930b | 5567 | gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
5568 | struct alloc_context ac = { }; |
5569 | ||
c63ae43b MH |
5570 | /* |
5571 | * There are several places where we assume that the order value is sane | |
5572 | * so bail out early if the request is out of bound. | |
5573 | */ | |
3f913fc5 | 5574 | if (WARN_ON_ONCE_GFP(order >= MAX_ORDER, gfp)) |
c63ae43b | 5575 | return NULL; |
c63ae43b | 5576 | |
6e5e0f28 | 5577 | gfp &= gfp_allowed_mask; |
da6df1b0 PT |
5578 | /* |
5579 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS | |
5580 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
5581 | * from a particular context which has been marked by | |
8e3560d9 PT |
5582 | * memalloc_no{fs,io}_{save,restore}. And PF_MEMALLOC_PIN which ensures |
5583 | * movable zones are not used during allocation. | |
da6df1b0 PT |
5584 | */ |
5585 | gfp = current_gfp_context(gfp); | |
6e5e0f28 MWO |
5586 | alloc_gfp = gfp; |
5587 | if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac, | |
8e6a930b | 5588 | &alloc_gfp, &alloc_flags)) |
9cd75558 MG |
5589 | return NULL; |
5590 | ||
6bb15450 MG |
5591 | /* |
5592 | * Forbid the first pass from falling back to types that fragment | |
5593 | * memory until all local zones are considered. | |
5594 | */ | |
6e5e0f28 | 5595 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp); |
6bb15450 | 5596 | |
5117f45d | 5597 | /* First allocation attempt */ |
8e6a930b | 5598 | page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac); |
4fcb0971 MG |
5599 | if (likely(page)) |
5600 | goto out; | |
11e33f6a | 5601 | |
da6df1b0 | 5602 | alloc_gfp = gfp; |
4fcb0971 | 5603 | ac.spread_dirty_pages = false; |
23f086f9 | 5604 | |
4741526b MG |
5605 | /* |
5606 | * Restore the original nodemask if it was potentially replaced with | |
5607 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
5608 | */ | |
97ce86f9 | 5609 | ac.nodemask = nodemask; |
16096c25 | 5610 | |
8e6a930b | 5611 | page = __alloc_pages_slowpath(alloc_gfp, order, &ac); |
cc9a6c87 | 5612 | |
4fcb0971 | 5613 | out: |
6e5e0f28 MWO |
5614 | if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT) && page && |
5615 | unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) { | |
c4159a75 VD |
5616 | __free_pages(page, order); |
5617 | page = NULL; | |
4949148a VD |
5618 | } |
5619 | ||
8e6a930b | 5620 | trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype); |
b073d7f8 | 5621 | kmsan_alloc_page(page, order, alloc_gfp); |
4fcb0971 | 5622 | |
11e33f6a | 5623 | return page; |
1da177e4 | 5624 | } |
84172f4b | 5625 | EXPORT_SYMBOL(__alloc_pages); |
1da177e4 | 5626 | |
cc09cb13 MWO |
5627 | struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid, |
5628 | nodemask_t *nodemask) | |
5629 | { | |
5630 | struct page *page = __alloc_pages(gfp | __GFP_COMP, order, | |
5631 | preferred_nid, nodemask); | |
5632 | ||
5633 | if (page && order > 1) | |
5634 | prep_transhuge_page(page); | |
5635 | return (struct folio *)page; | |
5636 | } | |
5637 | EXPORT_SYMBOL(__folio_alloc); | |
5638 | ||
1da177e4 | 5639 | /* |
9ea9a680 MH |
5640 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
5641 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
5642 | * you need to access high mem. | |
1da177e4 | 5643 | */ |
920c7a5d | 5644 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 5645 | { |
945a1113 AM |
5646 | struct page *page; |
5647 | ||
9ea9a680 | 5648 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
5649 | if (!page) |
5650 | return 0; | |
5651 | return (unsigned long) page_address(page); | |
5652 | } | |
1da177e4 LT |
5653 | EXPORT_SYMBOL(__get_free_pages); |
5654 | ||
920c7a5d | 5655 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 5656 | { |
945a1113 | 5657 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 5658 | } |
1da177e4 LT |
5659 | EXPORT_SYMBOL(get_zeroed_page); |
5660 | ||
7f194fbb MWO |
5661 | /** |
5662 | * __free_pages - Free pages allocated with alloc_pages(). | |
5663 | * @page: The page pointer returned from alloc_pages(). | |
5664 | * @order: The order of the allocation. | |
5665 | * | |
5666 | * This function can free multi-page allocations that are not compound | |
5667 | * pages. It does not check that the @order passed in matches that of | |
5668 | * the allocation, so it is easy to leak memory. Freeing more memory | |
5669 | * than was allocated will probably emit a warning. | |
5670 | * | |
5671 | * If the last reference to this page is speculative, it will be released | |
5672 | * by put_page() which only frees the first page of a non-compound | |
5673 | * allocation. To prevent the remaining pages from being leaked, we free | |
5674 | * the subsequent pages here. If you want to use the page's reference | |
5675 | * count to decide when to free the allocation, you should allocate a | |
5676 | * compound page, and use put_page() instead of __free_pages(). | |
5677 | * | |
5678 | * Context: May be called in interrupt context or while holding a normal | |
5679 | * spinlock, but not in NMI context or while holding a raw spinlock. | |
5680 | */ | |
742aa7fb AL |
5681 | void __free_pages(struct page *page, unsigned int order) |
5682 | { | |
5683 | if (put_page_testzero(page)) | |
5684 | free_the_page(page, order); | |
e320d301 MWO |
5685 | else if (!PageHead(page)) |
5686 | while (order-- > 0) | |
5687 | free_the_page(page + (1 << order), order); | |
742aa7fb | 5688 | } |
1da177e4 LT |
5689 | EXPORT_SYMBOL(__free_pages); |
5690 | ||
920c7a5d | 5691 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
5692 | { |
5693 | if (addr != 0) { | |
725d704e | 5694 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
5695 | __free_pages(virt_to_page((void *)addr), order); |
5696 | } | |
5697 | } | |
5698 | ||
5699 | EXPORT_SYMBOL(free_pages); | |
5700 | ||
b63ae8ca AD |
5701 | /* |
5702 | * Page Fragment: | |
5703 | * An arbitrary-length arbitrary-offset area of memory which resides | |
5704 | * within a 0 or higher order page. Multiple fragments within that page | |
5705 | * are individually refcounted, in the page's reference counter. | |
5706 | * | |
5707 | * The page_frag functions below provide a simple allocation framework for | |
5708 | * page fragments. This is used by the network stack and network device | |
5709 | * drivers to provide a backing region of memory for use as either an | |
5710 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
5711 | */ | |
2976db80 AD |
5712 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
5713 | gfp_t gfp_mask) | |
b63ae8ca AD |
5714 | { |
5715 | struct page *page = NULL; | |
5716 | gfp_t gfp = gfp_mask; | |
5717 | ||
5718 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5719 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
5720 | __GFP_NOMEMALLOC; | |
5721 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
5722 | PAGE_FRAG_CACHE_MAX_ORDER); | |
5723 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
5724 | #endif | |
5725 | if (unlikely(!page)) | |
5726 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
5727 | ||
5728 | nc->va = page ? page_address(page) : NULL; | |
5729 | ||
5730 | return page; | |
5731 | } | |
5732 | ||
2976db80 | 5733 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
5734 | { |
5735 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
5736 | ||
742aa7fb AL |
5737 | if (page_ref_sub_and_test(page, count)) |
5738 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 5739 | } |
2976db80 | 5740 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 5741 | |
b358e212 KH |
5742 | void *page_frag_alloc_align(struct page_frag_cache *nc, |
5743 | unsigned int fragsz, gfp_t gfp_mask, | |
5744 | unsigned int align_mask) | |
b63ae8ca AD |
5745 | { |
5746 | unsigned int size = PAGE_SIZE; | |
5747 | struct page *page; | |
5748 | int offset; | |
5749 | ||
5750 | if (unlikely(!nc->va)) { | |
5751 | refill: | |
2976db80 | 5752 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
5753 | if (!page) |
5754 | return NULL; | |
5755 | ||
5756 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5757 | /* if size can vary use size else just use PAGE_SIZE */ | |
5758 | size = nc->size; | |
5759 | #endif | |
5760 | /* Even if we own the page, we do not use atomic_set(). | |
5761 | * This would break get_page_unless_zero() users. | |
5762 | */ | |
86447726 | 5763 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
5764 | |
5765 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 5766 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 5767 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5768 | nc->offset = size; |
5769 | } | |
5770 | ||
5771 | offset = nc->offset - fragsz; | |
5772 | if (unlikely(offset < 0)) { | |
5773 | page = virt_to_page(nc->va); | |
5774 | ||
fe896d18 | 5775 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
5776 | goto refill; |
5777 | ||
d8c19014 DZ |
5778 | if (unlikely(nc->pfmemalloc)) { |
5779 | free_the_page(page, compound_order(page)); | |
5780 | goto refill; | |
5781 | } | |
5782 | ||
b63ae8ca AD |
5783 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
5784 | /* if size can vary use size else just use PAGE_SIZE */ | |
5785 | size = nc->size; | |
5786 | #endif | |
5787 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 5788 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
5789 | |
5790 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 5791 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca | 5792 | offset = size - fragsz; |
dac22531 ML |
5793 | if (unlikely(offset < 0)) { |
5794 | /* | |
5795 | * The caller is trying to allocate a fragment | |
5796 | * with fragsz > PAGE_SIZE but the cache isn't big | |
5797 | * enough to satisfy the request, this may | |
5798 | * happen in low memory conditions. | |
5799 | * We don't release the cache page because | |
5800 | * it could make memory pressure worse | |
5801 | * so we simply return NULL here. | |
5802 | */ | |
5803 | return NULL; | |
5804 | } | |
b63ae8ca AD |
5805 | } |
5806 | ||
5807 | nc->pagecnt_bias--; | |
b358e212 | 5808 | offset &= align_mask; |
b63ae8ca AD |
5809 | nc->offset = offset; |
5810 | ||
5811 | return nc->va + offset; | |
5812 | } | |
b358e212 | 5813 | EXPORT_SYMBOL(page_frag_alloc_align); |
b63ae8ca AD |
5814 | |
5815 | /* | |
5816 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
5817 | */ | |
8c2dd3e4 | 5818 | void page_frag_free(void *addr) |
b63ae8ca AD |
5819 | { |
5820 | struct page *page = virt_to_head_page(addr); | |
5821 | ||
742aa7fb AL |
5822 | if (unlikely(put_page_testzero(page))) |
5823 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 5824 | } |
8c2dd3e4 | 5825 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 5826 | |
d00181b9 KS |
5827 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
5828 | size_t size) | |
ee85c2e1 AK |
5829 | { |
5830 | if (addr) { | |
df48a5f7 LH |
5831 | unsigned long nr = DIV_ROUND_UP(size, PAGE_SIZE); |
5832 | struct page *page = virt_to_page((void *)addr); | |
5833 | struct page *last = page + nr; | |
5834 | ||
5835 | split_page_owner(page, 1 << order); | |
5836 | split_page_memcg(page, 1 << order); | |
5837 | while (page < --last) | |
5838 | set_page_refcounted(last); | |
5839 | ||
5840 | last = page + (1UL << order); | |
5841 | for (page += nr; page < last; page++) | |
5842 | __free_pages_ok(page, 0, FPI_TO_TAIL); | |
ee85c2e1 AK |
5843 | } |
5844 | return (void *)addr; | |
5845 | } | |
5846 | ||
2be0ffe2 TT |
5847 | /** |
5848 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
5849 | * @size: the number of bytes to allocate | |
63931eb9 | 5850 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
5851 | * |
5852 | * This function is similar to alloc_pages(), except that it allocates the | |
5853 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
5854 | * allocate memory in power-of-two pages. | |
5855 | * | |
5856 | * This function is also limited by MAX_ORDER. | |
5857 | * | |
5858 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
5859 | * |
5860 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
5861 | */ |
5862 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
5863 | { | |
5864 | unsigned int order = get_order(size); | |
5865 | unsigned long addr; | |
5866 | ||
ba7f1b9e ML |
5867 | if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM))) |
5868 | gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM); | |
63931eb9 | 5869 | |
2be0ffe2 | 5870 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 5871 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
5872 | } |
5873 | EXPORT_SYMBOL(alloc_pages_exact); | |
5874 | ||
ee85c2e1 AK |
5875 | /** |
5876 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
5877 | * pages on a node. | |
b5e6ab58 | 5878 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 5879 | * @size: the number of bytes to allocate |
63931eb9 | 5880 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
5881 | * |
5882 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
5883 | * back. | |
a862f68a MR |
5884 | * |
5885 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 5886 | */ |
e1931811 | 5887 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5888 | { |
d00181b9 | 5889 | unsigned int order = get_order(size); |
63931eb9 VB |
5890 | struct page *p; |
5891 | ||
ba7f1b9e ML |
5892 | if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM))) |
5893 | gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM); | |
63931eb9 VB |
5894 | |
5895 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5896 | if (!p) |
5897 | return NULL; | |
5898 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5899 | } | |
ee85c2e1 | 5900 | |
2be0ffe2 TT |
5901 | /** |
5902 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5903 | * @virt: the value returned by alloc_pages_exact. | |
5904 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5905 | * | |
5906 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5907 | */ | |
5908 | void free_pages_exact(void *virt, size_t size) | |
5909 | { | |
5910 | unsigned long addr = (unsigned long)virt; | |
5911 | unsigned long end = addr + PAGE_ALIGN(size); | |
5912 | ||
5913 | while (addr < end) { | |
5914 | free_page(addr); | |
5915 | addr += PAGE_SIZE; | |
5916 | } | |
5917 | } | |
5918 | EXPORT_SYMBOL(free_pages_exact); | |
5919 | ||
e0fb5815 ZY |
5920 | /** |
5921 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5922 | * @offset: The zone index of the highest zone | |
5923 | * | |
a862f68a | 5924 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5925 | * high watermark within all zones at or below a given zone index. For each |
5926 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
5927 | * |
5928 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5929 | * |
5930 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5931 | */ |
ebec3862 | 5932 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5933 | { |
dd1a239f | 5934 | struct zoneref *z; |
54a6eb5c MG |
5935 | struct zone *zone; |
5936 | ||
e310fd43 | 5937 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5938 | unsigned long sum = 0; |
1da177e4 | 5939 | |
0e88460d | 5940 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5941 | |
54a6eb5c | 5942 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5943 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5944 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5945 | if (size > high) |
5946 | sum += size - high; | |
1da177e4 LT |
5947 | } |
5948 | ||
5949 | return sum; | |
5950 | } | |
5951 | ||
e0fb5815 ZY |
5952 | /** |
5953 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5954 | * | |
5955 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5956 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5957 | * |
5958 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5959 | * ZONE_NORMAL. | |
1da177e4 | 5960 | */ |
ebec3862 | 5961 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5962 | { |
af4ca457 | 5963 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5964 | } |
c2f1a551 | 5965 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5966 | |
08e0f6a9 | 5967 | static inline void show_node(struct zone *zone) |
1da177e4 | 5968 | { |
e5adfffc | 5969 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5970 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5971 | } |
1da177e4 | 5972 | |
d02bd27b IR |
5973 | long si_mem_available(void) |
5974 | { | |
5975 | long available; | |
5976 | unsigned long pagecache; | |
5977 | unsigned long wmark_low = 0; | |
5978 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5979 | unsigned long reclaimable; |
d02bd27b IR |
5980 | struct zone *zone; |
5981 | int lru; | |
5982 | ||
5983 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5984 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5985 | |
5986 | for_each_zone(zone) | |
a9214443 | 5987 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5988 | |
5989 | /* | |
5990 | * Estimate the amount of memory available for userspace allocations, | |
ade63b41 | 5991 | * without causing swapping or OOM. |
d02bd27b | 5992 | */ |
c41f012a | 5993 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5994 | |
5995 | /* | |
5996 | * Not all the page cache can be freed, otherwise the system will | |
ade63b41 YY |
5997 | * start swapping or thrashing. Assume at least half of the page |
5998 | * cache, or the low watermark worth of cache, needs to stay. | |
d02bd27b IR |
5999 | */ |
6000 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
6001 | pagecache -= min(pagecache / 2, wmark_low); | |
6002 | available += pagecache; | |
6003 | ||
6004 | /* | |
b29940c1 VB |
6005 | * Part of the reclaimable slab and other kernel memory consists of |
6006 | * items that are in use, and cannot be freed. Cap this estimate at the | |
6007 | * low watermark. | |
d02bd27b | 6008 | */ |
d42f3245 RG |
6009 | reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) + |
6010 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
b29940c1 | 6011 | available += reclaimable - min(reclaimable / 2, wmark_low); |
034ebf65 | 6012 | |
d02bd27b IR |
6013 | if (available < 0) |
6014 | available = 0; | |
6015 | return available; | |
6016 | } | |
6017 | EXPORT_SYMBOL_GPL(si_mem_available); | |
6018 | ||
1da177e4 LT |
6019 | void si_meminfo(struct sysinfo *val) |
6020 | { | |
ca79b0c2 | 6021 | val->totalram = totalram_pages(); |
11fb9989 | 6022 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 6023 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 6024 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 6025 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 6026 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
6027 | val->mem_unit = PAGE_SIZE; |
6028 | } | |
6029 | ||
6030 | EXPORT_SYMBOL(si_meminfo); | |
6031 | ||
6032 | #ifdef CONFIG_NUMA | |
6033 | void si_meminfo_node(struct sysinfo *val, int nid) | |
6034 | { | |
cdd91a77 JL |
6035 | int zone_type; /* needs to be signed */ |
6036 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
6037 | unsigned long managed_highpages = 0; |
6038 | unsigned long free_highpages = 0; | |
1da177e4 LT |
6039 | pg_data_t *pgdat = NODE_DATA(nid); |
6040 | ||
cdd91a77 | 6041 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 6042 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 6043 | val->totalram = managed_pages; |
11fb9989 | 6044 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 6045 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 6046 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
6047 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
6048 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
6049 | ||
6050 | if (is_highmem(zone)) { | |
9705bea5 | 6051 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
6052 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
6053 | } | |
6054 | } | |
6055 | val->totalhigh = managed_highpages; | |
6056 | val->freehigh = free_highpages; | |
98d2b0eb | 6057 | #else |
fc2bd799 JK |
6058 | val->totalhigh = managed_highpages; |
6059 | val->freehigh = free_highpages; | |
98d2b0eb | 6060 | #endif |
1da177e4 LT |
6061 | val->mem_unit = PAGE_SIZE; |
6062 | } | |
6063 | #endif | |
6064 | ||
ddd588b5 | 6065 | /* |
7bf02ea2 DR |
6066 | * Determine whether the node should be displayed or not, depending on whether |
6067 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 6068 | */ |
9af744d7 | 6069 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 6070 | { |
ddd588b5 | 6071 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 6072 | return false; |
ddd588b5 | 6073 | |
9af744d7 MH |
6074 | /* |
6075 | * no node mask - aka implicit memory numa policy. Do not bother with | |
6076 | * the synchronization - read_mems_allowed_begin - because we do not | |
6077 | * have to be precise here. | |
6078 | */ | |
6079 | if (!nodemask) | |
6080 | nodemask = &cpuset_current_mems_allowed; | |
6081 | ||
6082 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
6083 | } |
6084 | ||
1da177e4 LT |
6085 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
6086 | ||
377e4f16 RV |
6087 | static void show_migration_types(unsigned char type) |
6088 | { | |
6089 | static const char types[MIGRATE_TYPES] = { | |
6090 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 6091 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
6092 | [MIGRATE_RECLAIMABLE] = 'E', |
6093 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
6094 | #ifdef CONFIG_CMA |
6095 | [MIGRATE_CMA] = 'C', | |
6096 | #endif | |
194159fb | 6097 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 6098 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 6099 | #endif |
377e4f16 RV |
6100 | }; |
6101 | char tmp[MIGRATE_TYPES + 1]; | |
6102 | char *p = tmp; | |
6103 | int i; | |
6104 | ||
6105 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
6106 | if (type & (1 << i)) | |
6107 | *p++ = types[i]; | |
6108 | } | |
6109 | ||
6110 | *p = '\0'; | |
1f84a18f | 6111 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
6112 | } |
6113 | ||
974f4367 MH |
6114 | static bool node_has_managed_zones(pg_data_t *pgdat, int max_zone_idx) |
6115 | { | |
6116 | int zone_idx; | |
6117 | for (zone_idx = 0; zone_idx <= max_zone_idx; zone_idx++) | |
6118 | if (zone_managed_pages(pgdat->node_zones + zone_idx)) | |
6119 | return true; | |
6120 | return false; | |
6121 | } | |
6122 | ||
1da177e4 LT |
6123 | /* |
6124 | * Show free area list (used inside shift_scroll-lock stuff) | |
6125 | * We also calculate the percentage fragmentation. We do this by counting the | |
6126 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
6127 | * |
6128 | * Bits in @filter: | |
6129 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
6130 | * cpuset. | |
1da177e4 | 6131 | */ |
974f4367 | 6132 | void __show_free_areas(unsigned int filter, nodemask_t *nodemask, int max_zone_idx) |
1da177e4 | 6133 | { |
d1bfcdb8 | 6134 | unsigned long free_pcp = 0; |
dcadcf1c | 6135 | int cpu, nid; |
1da177e4 | 6136 | struct zone *zone; |
599d0c95 | 6137 | pg_data_t *pgdat; |
1da177e4 | 6138 | |
ee99c71c | 6139 | for_each_populated_zone(zone) { |
974f4367 MH |
6140 | if (zone_idx(zone) > max_zone_idx) |
6141 | continue; | |
9af744d7 | 6142 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 6143 | continue; |
d1bfcdb8 | 6144 | |
761b0677 | 6145 | for_each_online_cpu(cpu) |
28f836b6 | 6146 | free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count; |
1da177e4 LT |
6147 | } |
6148 | ||
a731286d KM |
6149 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
6150 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
8d92890b | 6151 | " unevictable:%lu dirty:%lu writeback:%lu\n" |
d1bfcdb8 | 6152 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
ebc97a52 YA |
6153 | " mapped:%lu shmem:%lu pagetables:%lu\n" |
6154 | " sec_pagetables:%lu bounce:%lu\n" | |
eb2169ce | 6155 | " kernel_misc_reclaimable:%lu\n" |
d1bfcdb8 | 6156 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
6157 | global_node_page_state(NR_ACTIVE_ANON), |
6158 | global_node_page_state(NR_INACTIVE_ANON), | |
6159 | global_node_page_state(NR_ISOLATED_ANON), | |
6160 | global_node_page_state(NR_ACTIVE_FILE), | |
6161 | global_node_page_state(NR_INACTIVE_FILE), | |
6162 | global_node_page_state(NR_ISOLATED_FILE), | |
6163 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
6164 | global_node_page_state(NR_FILE_DIRTY), |
6165 | global_node_page_state(NR_WRITEBACK), | |
d42f3245 RG |
6166 | global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B), |
6167 | global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B), | |
50658e2e | 6168 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 6169 | global_node_page_state(NR_SHMEM), |
f0c0c115 | 6170 | global_node_page_state(NR_PAGETABLE), |
ebc97a52 | 6171 | global_node_page_state(NR_SECONDARY_PAGETABLE), |
c41f012a | 6172 | global_zone_page_state(NR_BOUNCE), |
eb2169ce | 6173 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE), |
c41f012a | 6174 | global_zone_page_state(NR_FREE_PAGES), |
d1bfcdb8 | 6175 | free_pcp, |
c41f012a | 6176 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 6177 | |
599d0c95 | 6178 | for_each_online_pgdat(pgdat) { |
9af744d7 | 6179 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb | 6180 | continue; |
974f4367 MH |
6181 | if (!node_has_managed_zones(pgdat, max_zone_idx)) |
6182 | continue; | |
c02e50bb | 6183 | |
599d0c95 MG |
6184 | printk("Node %d" |
6185 | " active_anon:%lukB" | |
6186 | " inactive_anon:%lukB" | |
6187 | " active_file:%lukB" | |
6188 | " inactive_file:%lukB" | |
6189 | " unevictable:%lukB" | |
6190 | " isolated(anon):%lukB" | |
6191 | " isolated(file):%lukB" | |
50658e2e | 6192 | " mapped:%lukB" |
11fb9989 MG |
6193 | " dirty:%lukB" |
6194 | " writeback:%lukB" | |
6195 | " shmem:%lukB" | |
6196 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6197 | " shmem_thp: %lukB" | |
6198 | " shmem_pmdmapped: %lukB" | |
6199 | " anon_thp: %lukB" | |
6200 | #endif | |
6201 | " writeback_tmp:%lukB" | |
991e7673 SB |
6202 | " kernel_stack:%lukB" |
6203 | #ifdef CONFIG_SHADOW_CALL_STACK | |
6204 | " shadow_call_stack:%lukB" | |
6205 | #endif | |
f0c0c115 | 6206 | " pagetables:%lukB" |
ebc97a52 | 6207 | " sec_pagetables:%lukB" |
599d0c95 MG |
6208 | " all_unreclaimable? %s" |
6209 | "\n", | |
6210 | pgdat->node_id, | |
6211 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
6212 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
6213 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
6214 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
6215 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
6216 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
6217 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 6218 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
6219 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
6220 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 6221 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 | 6222 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
57b2847d | 6223 | K(node_page_state(pgdat, NR_SHMEM_THPS)), |
a1528e21 | 6224 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)), |
69473e5d | 6225 | K(node_page_state(pgdat, NR_ANON_THPS)), |
11fb9989 | 6226 | #endif |
11fb9989 | 6227 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
991e7673 SB |
6228 | node_page_state(pgdat, NR_KERNEL_STACK_KB), |
6229 | #ifdef CONFIG_SHADOW_CALL_STACK | |
6230 | node_page_state(pgdat, NR_KERNEL_SCS_KB), | |
6231 | #endif | |
f0c0c115 | 6232 | K(node_page_state(pgdat, NR_PAGETABLE)), |
ebc97a52 | 6233 | K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)), |
c73322d0 JW |
6234 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
6235 | "yes" : "no"); | |
599d0c95 MG |
6236 | } |
6237 | ||
ee99c71c | 6238 | for_each_populated_zone(zone) { |
1da177e4 LT |
6239 | int i; |
6240 | ||
974f4367 MH |
6241 | if (zone_idx(zone) > max_zone_idx) |
6242 | continue; | |
9af744d7 | 6243 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 6244 | continue; |
d1bfcdb8 KK |
6245 | |
6246 | free_pcp = 0; | |
6247 | for_each_online_cpu(cpu) | |
28f836b6 | 6248 | free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count; |
d1bfcdb8 | 6249 | |
1da177e4 | 6250 | show_node(zone); |
1f84a18f JP |
6251 | printk(KERN_CONT |
6252 | "%s" | |
1da177e4 | 6253 | " free:%lukB" |
a6ea8b5b | 6254 | " boost:%lukB" |
1da177e4 LT |
6255 | " min:%lukB" |
6256 | " low:%lukB" | |
6257 | " high:%lukB" | |
e47b346a | 6258 | " reserved_highatomic:%luKB" |
71c799f4 MK |
6259 | " active_anon:%lukB" |
6260 | " inactive_anon:%lukB" | |
6261 | " active_file:%lukB" | |
6262 | " inactive_file:%lukB" | |
6263 | " unevictable:%lukB" | |
5a1c84b4 | 6264 | " writepending:%lukB" |
1da177e4 | 6265 | " present:%lukB" |
9feedc9d | 6266 | " managed:%lukB" |
4a0aa73f | 6267 | " mlocked:%lukB" |
4a0aa73f | 6268 | " bounce:%lukB" |
d1bfcdb8 KK |
6269 | " free_pcp:%lukB" |
6270 | " local_pcp:%ukB" | |
d1ce749a | 6271 | " free_cma:%lukB" |
1da177e4 LT |
6272 | "\n", |
6273 | zone->name, | |
88f5acf8 | 6274 | K(zone_page_state(zone, NR_FREE_PAGES)), |
a6ea8b5b | 6275 | K(zone->watermark_boost), |
41858966 MG |
6276 | K(min_wmark_pages(zone)), |
6277 | K(low_wmark_pages(zone)), | |
6278 | K(high_wmark_pages(zone)), | |
e47b346a | 6279 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
6280 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
6281 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
6282 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
6283 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
6284 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 6285 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 6286 | K(zone->present_pages), |
9705bea5 | 6287 | K(zone_managed_pages(zone)), |
4a0aa73f | 6288 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f | 6289 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 | 6290 | K(free_pcp), |
28f836b6 | 6291 | K(this_cpu_read(zone->per_cpu_pageset->count)), |
33e077bd | 6292 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
6293 | printk("lowmem_reserve[]:"); |
6294 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
6295 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
6296 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
6297 | } |
6298 | ||
ee99c71c | 6299 | for_each_populated_zone(zone) { |
d00181b9 KS |
6300 | unsigned int order; |
6301 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 6302 | unsigned char types[MAX_ORDER]; |
1da177e4 | 6303 | |
974f4367 MH |
6304 | if (zone_idx(zone) > max_zone_idx) |
6305 | continue; | |
9af744d7 | 6306 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 6307 | continue; |
1da177e4 | 6308 | show_node(zone); |
1f84a18f | 6309 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
6310 | |
6311 | spin_lock_irqsave(&zone->lock, flags); | |
6312 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
6313 | struct free_area *area = &zone->free_area[order]; |
6314 | int type; | |
6315 | ||
6316 | nr[order] = area->nr_free; | |
8f9de51a | 6317 | total += nr[order] << order; |
377e4f16 RV |
6318 | |
6319 | types[order] = 0; | |
6320 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 6321 | if (!free_area_empty(area, type)) |
377e4f16 RV |
6322 | types[order] |= 1 << type; |
6323 | } | |
1da177e4 LT |
6324 | } |
6325 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 6326 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
6327 | printk(KERN_CONT "%lu*%lukB ", |
6328 | nr[order], K(1UL) << order); | |
377e4f16 RV |
6329 | if (nr[order]) |
6330 | show_migration_types(types[order]); | |
6331 | } | |
1f84a18f | 6332 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
6333 | } |
6334 | ||
dcadcf1c GL |
6335 | for_each_online_node(nid) { |
6336 | if (show_mem_node_skip(filter, nid, nodemask)) | |
6337 | continue; | |
6338 | hugetlb_show_meminfo_node(nid); | |
6339 | } | |
949f7ec5 | 6340 | |
11fb9989 | 6341 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 6342 | |
1da177e4 LT |
6343 | show_swap_cache_info(); |
6344 | } | |
6345 | ||
19770b32 MG |
6346 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
6347 | { | |
6348 | zoneref->zone = zone; | |
6349 | zoneref->zone_idx = zone_idx(zone); | |
6350 | } | |
6351 | ||
1da177e4 LT |
6352 | /* |
6353 | * Builds allocation fallback zone lists. | |
1a93205b CL |
6354 | * |
6355 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 6356 | */ |
9d3be21b | 6357 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 6358 | { |
1a93205b | 6359 | struct zone *zone; |
bc732f1d | 6360 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 6361 | int nr_zones = 0; |
02a68a5e CL |
6362 | |
6363 | do { | |
2f6726e5 | 6364 | zone_type--; |
070f8032 | 6365 | zone = pgdat->node_zones + zone_type; |
e553f62f | 6366 | if (populated_zone(zone)) { |
9d3be21b | 6367 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 6368 | check_highest_zone(zone_type); |
1da177e4 | 6369 | } |
2f6726e5 | 6370 | } while (zone_type); |
bc732f1d | 6371 | |
070f8032 | 6372 | return nr_zones; |
1da177e4 LT |
6373 | } |
6374 | ||
6375 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
6376 | |
6377 | static int __parse_numa_zonelist_order(char *s) | |
6378 | { | |
c9bff3ee | 6379 | /* |
f0953a1b | 6380 | * We used to support different zonelists modes but they turned |
c9bff3ee MH |
6381 | * out to be just not useful. Let's keep the warning in place |
6382 | * if somebody still use the cmd line parameter so that we do | |
6383 | * not fail it silently | |
6384 | */ | |
6385 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
6386 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
6387 | return -EINVAL; |
6388 | } | |
6389 | return 0; | |
6390 | } | |
6391 | ||
c9bff3ee MH |
6392 | char numa_zonelist_order[] = "Node"; |
6393 | ||
f0c0b2b8 KH |
6394 | /* |
6395 | * sysctl handler for numa_zonelist_order | |
6396 | */ | |
cccad5b9 | 6397 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 6398 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 6399 | { |
32927393 CH |
6400 | if (write) |
6401 | return __parse_numa_zonelist_order(buffer); | |
6402 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
6403 | } |
6404 | ||
6405 | ||
f0c0b2b8 KH |
6406 | static int node_load[MAX_NUMNODES]; |
6407 | ||
1da177e4 | 6408 | /** |
4dc3b16b | 6409 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
6410 | * @node: node whose fallback list we're appending |
6411 | * @used_node_mask: nodemask_t of already used nodes | |
6412 | * | |
6413 | * We use a number of factors to determine which is the next node that should | |
6414 | * appear on a given node's fallback list. The node should not have appeared | |
6415 | * already in @node's fallback list, and it should be the next closest node | |
6416 | * according to the distance array (which contains arbitrary distance values | |
6417 | * from each node to each node in the system), and should also prefer nodes | |
6418 | * with no CPUs, since presumably they'll have very little allocation pressure | |
6419 | * on them otherwise. | |
a862f68a MR |
6420 | * |
6421 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 6422 | */ |
79c28a41 | 6423 | int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 6424 | { |
4cf808eb | 6425 | int n, val; |
1da177e4 | 6426 | int min_val = INT_MAX; |
00ef2d2f | 6427 | int best_node = NUMA_NO_NODE; |
1da177e4 | 6428 | |
4cf808eb LT |
6429 | /* Use the local node if we haven't already */ |
6430 | if (!node_isset(node, *used_node_mask)) { | |
6431 | node_set(node, *used_node_mask); | |
6432 | return node; | |
6433 | } | |
1da177e4 | 6434 | |
4b0ef1fe | 6435 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
6436 | |
6437 | /* Don't want a node to appear more than once */ | |
6438 | if (node_isset(n, *used_node_mask)) | |
6439 | continue; | |
6440 | ||
1da177e4 LT |
6441 | /* Use the distance array to find the distance */ |
6442 | val = node_distance(node, n); | |
6443 | ||
4cf808eb LT |
6444 | /* Penalize nodes under us ("prefer the next node") */ |
6445 | val += (n < node); | |
6446 | ||
1da177e4 | 6447 | /* Give preference to headless and unused nodes */ |
b630749f | 6448 | if (!cpumask_empty(cpumask_of_node(n))) |
1da177e4 LT |
6449 | val += PENALTY_FOR_NODE_WITH_CPUS; |
6450 | ||
6451 | /* Slight preference for less loaded node */ | |
37931324 | 6452 | val *= MAX_NUMNODES; |
1da177e4 LT |
6453 | val += node_load[n]; |
6454 | ||
6455 | if (val < min_val) { | |
6456 | min_val = val; | |
6457 | best_node = n; | |
6458 | } | |
6459 | } | |
6460 | ||
6461 | if (best_node >= 0) | |
6462 | node_set(best_node, *used_node_mask); | |
6463 | ||
6464 | return best_node; | |
6465 | } | |
6466 | ||
f0c0b2b8 KH |
6467 | |
6468 | /* | |
6469 | * Build zonelists ordered by node and zones within node. | |
6470 | * This results in maximum locality--normal zone overflows into local | |
6471 | * DMA zone, if any--but risks exhausting DMA zone. | |
6472 | */ | |
9d3be21b MH |
6473 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
6474 | unsigned nr_nodes) | |
1da177e4 | 6475 | { |
9d3be21b MH |
6476 | struct zoneref *zonerefs; |
6477 | int i; | |
6478 | ||
6479 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
6480 | ||
6481 | for (i = 0; i < nr_nodes; i++) { | |
6482 | int nr_zones; | |
6483 | ||
6484 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 6485 | |
9d3be21b MH |
6486 | nr_zones = build_zonerefs_node(node, zonerefs); |
6487 | zonerefs += nr_zones; | |
6488 | } | |
6489 | zonerefs->zone = NULL; | |
6490 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
6491 | } |
6492 | ||
523b9458 CL |
6493 | /* |
6494 | * Build gfp_thisnode zonelists | |
6495 | */ | |
6496 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
6497 | { | |
9d3be21b MH |
6498 | struct zoneref *zonerefs; |
6499 | int nr_zones; | |
523b9458 | 6500 | |
9d3be21b MH |
6501 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
6502 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6503 | zonerefs += nr_zones; | |
6504 | zonerefs->zone = NULL; | |
6505 | zonerefs->zone_idx = 0; | |
523b9458 CL |
6506 | } |
6507 | ||
f0c0b2b8 KH |
6508 | /* |
6509 | * Build zonelists ordered by zone and nodes within zones. | |
6510 | * This results in conserving DMA zone[s] until all Normal memory is | |
6511 | * exhausted, but results in overflowing to remote node while memory | |
6512 | * may still exist in local DMA zone. | |
6513 | */ | |
f0c0b2b8 | 6514 | |
f0c0b2b8 KH |
6515 | static void build_zonelists(pg_data_t *pgdat) |
6516 | { | |
9d3be21b | 6517 | static int node_order[MAX_NUMNODES]; |
37931324 | 6518 | int node, nr_nodes = 0; |
d0ddf49b | 6519 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 6520 | int local_node, prev_node; |
1da177e4 LT |
6521 | |
6522 | /* NUMA-aware ordering of nodes */ | |
6523 | local_node = pgdat->node_id; | |
1da177e4 | 6524 | prev_node = local_node; |
f0c0b2b8 | 6525 | |
f0c0b2b8 | 6526 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
6527 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
6528 | /* | |
6529 | * We don't want to pressure a particular node. | |
6530 | * So adding penalty to the first node in same | |
6531 | * distance group to make it round-robin. | |
6532 | */ | |
957f822a DR |
6533 | if (node_distance(local_node, node) != |
6534 | node_distance(local_node, prev_node)) | |
37931324 | 6535 | node_load[node] += 1; |
f0c0b2b8 | 6536 | |
9d3be21b | 6537 | node_order[nr_nodes++] = node; |
1da177e4 | 6538 | prev_node = node; |
1da177e4 | 6539 | } |
523b9458 | 6540 | |
9d3be21b | 6541 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 6542 | build_thisnode_zonelists(pgdat); |
6cf25392 BR |
6543 | pr_info("Fallback order for Node %d: ", local_node); |
6544 | for (node = 0; node < nr_nodes; node++) | |
6545 | pr_cont("%d ", node_order[node]); | |
6546 | pr_cont("\n"); | |
1da177e4 LT |
6547 | } |
6548 | ||
7aac7898 LS |
6549 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6550 | /* | |
6551 | * Return node id of node used for "local" allocations. | |
6552 | * I.e., first node id of first zone in arg node's generic zonelist. | |
6553 | * Used for initializing percpu 'numa_mem', which is used primarily | |
6554 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
6555 | */ | |
6556 | int local_memory_node(int node) | |
6557 | { | |
c33d6c06 | 6558 | struct zoneref *z; |
7aac7898 | 6559 | |
c33d6c06 | 6560 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 6561 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 6562 | NULL); |
c1093b74 | 6563 | return zone_to_nid(z->zone); |
7aac7898 LS |
6564 | } |
6565 | #endif | |
f0c0b2b8 | 6566 | |
6423aa81 JK |
6567 | static void setup_min_unmapped_ratio(void); |
6568 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
6569 | #else /* CONFIG_NUMA */ |
6570 | ||
f0c0b2b8 | 6571 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 6572 | { |
19655d34 | 6573 | int node, local_node; |
9d3be21b MH |
6574 | struct zoneref *zonerefs; |
6575 | int nr_zones; | |
1da177e4 LT |
6576 | |
6577 | local_node = pgdat->node_id; | |
1da177e4 | 6578 | |
9d3be21b MH |
6579 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
6580 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6581 | zonerefs += nr_zones; | |
1da177e4 | 6582 | |
54a6eb5c MG |
6583 | /* |
6584 | * Now we build the zonelist so that it contains the zones | |
6585 | * of all the other nodes. | |
6586 | * We don't want to pressure a particular node, so when | |
6587 | * building the zones for node N, we make sure that the | |
6588 | * zones coming right after the local ones are those from | |
6589 | * node N+1 (modulo N) | |
6590 | */ | |
6591 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
6592 | if (!node_online(node)) | |
6593 | continue; | |
9d3be21b MH |
6594 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6595 | zonerefs += nr_zones; | |
1da177e4 | 6596 | } |
54a6eb5c MG |
6597 | for (node = 0; node < local_node; node++) { |
6598 | if (!node_online(node)) | |
6599 | continue; | |
9d3be21b MH |
6600 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6601 | zonerefs += nr_zones; | |
54a6eb5c MG |
6602 | } |
6603 | ||
9d3be21b MH |
6604 | zonerefs->zone = NULL; |
6605 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
6606 | } |
6607 | ||
6608 | #endif /* CONFIG_NUMA */ | |
6609 | ||
99dcc3e5 CL |
6610 | /* |
6611 | * Boot pageset table. One per cpu which is going to be used for all | |
6612 | * zones and all nodes. The parameters will be set in such a way | |
6613 | * that an item put on a list will immediately be handed over to | |
6614 | * the buddy list. This is safe since pageset manipulation is done | |
6615 | * with interrupts disabled. | |
6616 | * | |
6617 | * The boot_pagesets must be kept even after bootup is complete for | |
6618 | * unused processors and/or zones. They do play a role for bootstrapping | |
6619 | * hotplugged processors. | |
6620 | * | |
6621 | * zoneinfo_show() and maybe other functions do | |
6622 | * not check if the processor is online before following the pageset pointer. | |
6623 | * Other parts of the kernel may not check if the zone is available. | |
6624 | */ | |
28f836b6 | 6625 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats); |
952eaf81 VB |
6626 | /* These effectively disable the pcplists in the boot pageset completely */ |
6627 | #define BOOT_PAGESET_HIGH 0 | |
6628 | #define BOOT_PAGESET_BATCH 1 | |
28f836b6 MG |
6629 | static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset); |
6630 | static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats); | |
6dc2c87a | 6631 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 6632 | |
11cd8638 | 6633 | static void __build_all_zonelists(void *data) |
1da177e4 | 6634 | { |
6811378e | 6635 | int nid; |
afb6ebb3 | 6636 | int __maybe_unused cpu; |
9adb62a5 | 6637 | pg_data_t *self = data; |
b93e0f32 | 6638 | |
3d36424b | 6639 | write_seqlock(&zonelist_update_seq); |
9276b1bc | 6640 | |
7f9cfb31 BL |
6641 | #ifdef CONFIG_NUMA |
6642 | memset(node_load, 0, sizeof(node_load)); | |
6643 | #endif | |
9adb62a5 | 6644 | |
c1152583 WY |
6645 | /* |
6646 | * This node is hotadded and no memory is yet present. So just | |
6647 | * building zonelists is fine - no need to touch other nodes. | |
6648 | */ | |
9adb62a5 JL |
6649 | if (self && !node_online(self->node_id)) { |
6650 | build_zonelists(self); | |
c1152583 | 6651 | } else { |
09f49dca MH |
6652 | /* |
6653 | * All possible nodes have pgdat preallocated | |
6654 | * in free_area_init | |
6655 | */ | |
6656 | for_each_node(nid) { | |
c1152583 | 6657 | pg_data_t *pgdat = NODE_DATA(nid); |
7ea1530a | 6658 | |
c1152583 WY |
6659 | build_zonelists(pgdat); |
6660 | } | |
99dcc3e5 | 6661 | |
7aac7898 LS |
6662 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6663 | /* | |
6664 | * We now know the "local memory node" for each node-- | |
6665 | * i.e., the node of the first zone in the generic zonelist. | |
6666 | * Set up numa_mem percpu variable for on-line cpus. During | |
6667 | * boot, only the boot cpu should be on-line; we'll init the | |
6668 | * secondary cpus' numa_mem as they come on-line. During | |
6669 | * node/memory hotplug, we'll fixup all on-line cpus. | |
6670 | */ | |
d9c9a0b9 | 6671 | for_each_online_cpu(cpu) |
7aac7898 | 6672 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 6673 | #endif |
d9c9a0b9 | 6674 | } |
b93e0f32 | 6675 | |
3d36424b | 6676 | write_sequnlock(&zonelist_update_seq); |
6811378e YG |
6677 | } |
6678 | ||
061f67bc RV |
6679 | static noinline void __init |
6680 | build_all_zonelists_init(void) | |
6681 | { | |
afb6ebb3 MH |
6682 | int cpu; |
6683 | ||
061f67bc | 6684 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
6685 | |
6686 | /* | |
6687 | * Initialize the boot_pagesets that are going to be used | |
6688 | * for bootstrapping processors. The real pagesets for | |
6689 | * each zone will be allocated later when the per cpu | |
6690 | * allocator is available. | |
6691 | * | |
6692 | * boot_pagesets are used also for bootstrapping offline | |
6693 | * cpus if the system is already booted because the pagesets | |
6694 | * are needed to initialize allocators on a specific cpu too. | |
6695 | * F.e. the percpu allocator needs the page allocator which | |
6696 | * needs the percpu allocator in order to allocate its pagesets | |
6697 | * (a chicken-egg dilemma). | |
6698 | */ | |
6699 | for_each_possible_cpu(cpu) | |
28f836b6 | 6700 | per_cpu_pages_init(&per_cpu(boot_pageset, cpu), &per_cpu(boot_zonestats, cpu)); |
afb6ebb3 | 6701 | |
061f67bc RV |
6702 | mminit_verify_zonelist(); |
6703 | cpuset_init_current_mems_allowed(); | |
6704 | } | |
6705 | ||
4eaf3f64 | 6706 | /* |
4eaf3f64 | 6707 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 6708 | * |
72675e13 | 6709 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 6710 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 6711 | */ |
72675e13 | 6712 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 6713 | { |
0a18e607 DH |
6714 | unsigned long vm_total_pages; |
6715 | ||
6811378e | 6716 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 6717 | build_all_zonelists_init(); |
6811378e | 6718 | } else { |
11cd8638 | 6719 | __build_all_zonelists(pgdat); |
6811378e YG |
6720 | /* cpuset refresh routine should be here */ |
6721 | } | |
56b9413b DH |
6722 | /* Get the number of free pages beyond high watermark in all zones. */ |
6723 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
6724 | /* |
6725 | * Disable grouping by mobility if the number of pages in the | |
6726 | * system is too low to allow the mechanism to work. It would be | |
6727 | * more accurate, but expensive to check per-zone. This check is | |
6728 | * made on memory-hotadd so a system can start with mobility | |
6729 | * disabled and enable it later | |
6730 | */ | |
d9c23400 | 6731 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
6732 | page_group_by_mobility_disabled = 1; |
6733 | else | |
6734 | page_group_by_mobility_disabled = 0; | |
6735 | ||
ce0725f7 | 6736 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 6737 | nr_online_nodes, |
756a025f JP |
6738 | page_group_by_mobility_disabled ? "off" : "on", |
6739 | vm_total_pages); | |
f0c0b2b8 | 6740 | #ifdef CONFIG_NUMA |
f88dfff5 | 6741 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 6742 | #endif |
1da177e4 LT |
6743 | } |
6744 | ||
a9a9e77f PT |
6745 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
6746 | static bool __meminit | |
6747 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
6748 | { | |
a9a9e77f PT |
6749 | static struct memblock_region *r; |
6750 | ||
6751 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
6752 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
cc6de168 | 6753 | for_each_mem_region(r) { |
a9a9e77f PT |
6754 | if (*pfn < memblock_region_memory_end_pfn(r)) |
6755 | break; | |
6756 | } | |
6757 | } | |
6758 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
6759 | memblock_is_mirror(r)) { | |
6760 | *pfn = memblock_region_memory_end_pfn(r); | |
6761 | return true; | |
6762 | } | |
6763 | } | |
a9a9e77f PT |
6764 | return false; |
6765 | } | |
6766 | ||
1da177e4 LT |
6767 | /* |
6768 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 6769 | * up by memblock_free_all() once the early boot process is |
1da177e4 | 6770 | * done. Non-atomic initialization, single-pass. |
d882c006 DH |
6771 | * |
6772 | * All aligned pageblocks are initialized to the specified migratetype | |
6773 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
6774 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
1da177e4 | 6775 | */ |
ab28cb6e | 6776 | void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone, |
dc2da7b4 | 6777 | unsigned long start_pfn, unsigned long zone_end_pfn, |
d882c006 DH |
6778 | enum meminit_context context, |
6779 | struct vmem_altmap *altmap, int migratetype) | |
1da177e4 | 6780 | { |
a9a9e77f | 6781 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 6782 | struct page *page; |
1da177e4 | 6783 | |
22b31eec HD |
6784 | if (highest_memmap_pfn < end_pfn - 1) |
6785 | highest_memmap_pfn = end_pfn - 1; | |
6786 | ||
966cf44f | 6787 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
6788 | /* |
6789 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
6790 | * memory. We limit the total number of pages to initialize to just |
6791 | * those that might contain the memory mapping. We will defer the | |
6792 | * ZONE_DEVICE page initialization until after we have released | |
6793 | * the hotplug lock. | |
4b94ffdc | 6794 | */ |
966cf44f AD |
6795 | if (zone == ZONE_DEVICE) { |
6796 | if (!altmap) | |
6797 | return; | |
6798 | ||
6799 | if (start_pfn == altmap->base_pfn) | |
6800 | start_pfn += altmap->reserve; | |
6801 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
6802 | } | |
6803 | #endif | |
4b94ffdc | 6804 | |
948c436e | 6805 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 6806 | /* |
b72d0ffb AM |
6807 | * There can be holes in boot-time mem_map[]s handed to this |
6808 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 6809 | */ |
c1d0da83 | 6810 | if (context == MEMINIT_EARLY) { |
a9a9e77f PT |
6811 | if (overlap_memmap_init(zone, &pfn)) |
6812 | continue; | |
dc2da7b4 | 6813 | if (defer_init(nid, pfn, zone_end_pfn)) |
a9a9e77f | 6814 | break; |
a2f3aa02 | 6815 | } |
ac5d2539 | 6816 | |
d0dc12e8 PT |
6817 | page = pfn_to_page(pfn); |
6818 | __init_single_page(page, pfn, zone, nid); | |
c1d0da83 | 6819 | if (context == MEMINIT_HOTPLUG) |
d483da5b | 6820 | __SetPageReserved(page); |
d0dc12e8 | 6821 | |
ac5d2539 | 6822 | /* |
d882c006 DH |
6823 | * Usually, we want to mark the pageblock MIGRATE_MOVABLE, |
6824 | * such that unmovable allocations won't be scattered all | |
6825 | * over the place during system boot. | |
ac5d2539 | 6826 | */ |
ee0913c4 | 6827 | if (pageblock_aligned(pfn)) { |
d882c006 | 6828 | set_pageblock_migratetype(page, migratetype); |
9b6e63cb | 6829 | cond_resched(); |
ac5d2539 | 6830 | } |
948c436e | 6831 | pfn++; |
1da177e4 LT |
6832 | } |
6833 | } | |
6834 | ||
966cf44f | 6835 | #ifdef CONFIG_ZONE_DEVICE |
46487e00 JM |
6836 | static void __ref __init_zone_device_page(struct page *page, unsigned long pfn, |
6837 | unsigned long zone_idx, int nid, | |
6838 | struct dev_pagemap *pgmap) | |
6839 | { | |
6840 | ||
6841 | __init_single_page(page, pfn, zone_idx, nid); | |
6842 | ||
6843 | /* | |
6844 | * Mark page reserved as it will need to wait for onlining | |
6845 | * phase for it to be fully associated with a zone. | |
6846 | * | |
6847 | * We can use the non-atomic __set_bit operation for setting | |
6848 | * the flag as we are still initializing the pages. | |
6849 | */ | |
6850 | __SetPageReserved(page); | |
6851 | ||
6852 | /* | |
6853 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer | |
6854 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
6855 | * ever freed or placed on a driver-private list. | |
6856 | */ | |
6857 | page->pgmap = pgmap; | |
6858 | page->zone_device_data = NULL; | |
6859 | ||
6860 | /* | |
6861 | * Mark the block movable so that blocks are reserved for | |
6862 | * movable at startup. This will force kernel allocations | |
6863 | * to reserve their blocks rather than leaking throughout | |
6864 | * the address space during boot when many long-lived | |
6865 | * kernel allocations are made. | |
6866 | * | |
6867 | * Please note that MEMINIT_HOTPLUG path doesn't clear memmap | |
6868 | * because this is done early in section_activate() | |
6869 | */ | |
ee0913c4 | 6870 | if (pageblock_aligned(pfn)) { |
46487e00 JM |
6871 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
6872 | cond_resched(); | |
6873 | } | |
ef233450 AP |
6874 | |
6875 | /* | |
6876 | * ZONE_DEVICE pages are released directly to the driver page allocator | |
6877 | * which will set the page count to 1 when allocating the page. | |
6878 | */ | |
6879 | if (pgmap->type == MEMORY_DEVICE_PRIVATE || | |
6880 | pgmap->type == MEMORY_DEVICE_COHERENT) | |
6881 | set_page_count(page, 0); | |
46487e00 JM |
6882 | } |
6883 | ||
6fd3620b JM |
6884 | /* |
6885 | * With compound page geometry and when struct pages are stored in ram most | |
6886 | * tail pages are reused. Consequently, the amount of unique struct pages to | |
6887 | * initialize is a lot smaller that the total amount of struct pages being | |
6888 | * mapped. This is a paired / mild layering violation with explicit knowledge | |
6889 | * of how the sparse_vmemmap internals handle compound pages in the lack | |
6890 | * of an altmap. See vmemmap_populate_compound_pages(). | |
6891 | */ | |
6892 | static inline unsigned long compound_nr_pages(struct vmem_altmap *altmap, | |
6893 | unsigned long nr_pages) | |
6894 | { | |
6895 | return is_power_of_2(sizeof(struct page)) && | |
6896 | !altmap ? 2 * (PAGE_SIZE / sizeof(struct page)) : nr_pages; | |
6897 | } | |
6898 | ||
c4386bd8 JM |
6899 | static void __ref memmap_init_compound(struct page *head, |
6900 | unsigned long head_pfn, | |
6901 | unsigned long zone_idx, int nid, | |
6902 | struct dev_pagemap *pgmap, | |
6903 | unsigned long nr_pages) | |
6904 | { | |
6905 | unsigned long pfn, end_pfn = head_pfn + nr_pages; | |
6906 | unsigned int order = pgmap->vmemmap_shift; | |
6907 | ||
6908 | __SetPageHead(head); | |
6909 | for (pfn = head_pfn + 1; pfn < end_pfn; pfn++) { | |
6910 | struct page *page = pfn_to_page(pfn); | |
6911 | ||
6912 | __init_zone_device_page(page, pfn, zone_idx, nid, pgmap); | |
6913 | prep_compound_tail(head, pfn - head_pfn); | |
6914 | set_page_count(page, 0); | |
6915 | ||
6916 | /* | |
cb67f428 HD |
6917 | * The first tail page stores important compound page info. |
6918 | * Call prep_compound_head() after the first tail page has | |
6919 | * been initialized, to not have the data overwritten. | |
c4386bd8 | 6920 | */ |
cb67f428 | 6921 | if (pfn == head_pfn + 1) |
c4386bd8 JM |
6922 | prep_compound_head(head, order); |
6923 | } | |
6924 | } | |
6925 | ||
966cf44f AD |
6926 | void __ref memmap_init_zone_device(struct zone *zone, |
6927 | unsigned long start_pfn, | |
1f8d75c1 | 6928 | unsigned long nr_pages, |
966cf44f AD |
6929 | struct dev_pagemap *pgmap) |
6930 | { | |
1f8d75c1 | 6931 | unsigned long pfn, end_pfn = start_pfn + nr_pages; |
966cf44f | 6932 | struct pglist_data *pgdat = zone->zone_pgdat; |
514caf23 | 6933 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
c4386bd8 | 6934 | unsigned int pfns_per_compound = pgmap_vmemmap_nr(pgmap); |
966cf44f AD |
6935 | unsigned long zone_idx = zone_idx(zone); |
6936 | unsigned long start = jiffies; | |
6937 | int nid = pgdat->node_id; | |
6938 | ||
c0352904 | 6939 | if (WARN_ON_ONCE(!pgmap || zone_idx != ZONE_DEVICE)) |
966cf44f AD |
6940 | return; |
6941 | ||
6942 | /* | |
122e093c | 6943 | * The call to memmap_init should have already taken care |
966cf44f AD |
6944 | * of the pages reserved for the memmap, so we can just jump to |
6945 | * the end of that region and start processing the device pages. | |
6946 | */ | |
514caf23 | 6947 | if (altmap) { |
966cf44f | 6948 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
1f8d75c1 | 6949 | nr_pages = end_pfn - start_pfn; |
966cf44f AD |
6950 | } |
6951 | ||
c4386bd8 | 6952 | for (pfn = start_pfn; pfn < end_pfn; pfn += pfns_per_compound) { |
966cf44f AD |
6953 | struct page *page = pfn_to_page(pfn); |
6954 | ||
46487e00 | 6955 | __init_zone_device_page(page, pfn, zone_idx, nid, pgmap); |
c4386bd8 JM |
6956 | |
6957 | if (pfns_per_compound == 1) | |
6958 | continue; | |
6959 | ||
6960 | memmap_init_compound(page, pfn, zone_idx, nid, pgmap, | |
6fd3620b | 6961 | compound_nr_pages(altmap, pfns_per_compound)); |
966cf44f AD |
6962 | } |
6963 | ||
fdc029b1 | 6964 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
1f8d75c1 | 6965 | nr_pages, jiffies_to_msecs(jiffies - start)); |
966cf44f AD |
6966 | } |
6967 | ||
6968 | #endif | |
1e548deb | 6969 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6970 | { |
7aeb09f9 | 6971 | unsigned int order, t; |
b2a0ac88 MG |
6972 | for_each_migratetype_order(order, t) { |
6973 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6974 | zone->free_area[order].nr_free = 0; |
6975 | } | |
6976 | } | |
6977 | ||
0740a50b MR |
6978 | /* |
6979 | * Only struct pages that correspond to ranges defined by memblock.memory | |
6980 | * are zeroed and initialized by going through __init_single_page() during | |
122e093c | 6981 | * memmap_init_zone_range(). |
0740a50b MR |
6982 | * |
6983 | * But, there could be struct pages that correspond to holes in | |
6984 | * memblock.memory. This can happen because of the following reasons: | |
6985 | * - physical memory bank size is not necessarily the exact multiple of the | |
6986 | * arbitrary section size | |
6987 | * - early reserved memory may not be listed in memblock.memory | |
6988 | * - memory layouts defined with memmap= kernel parameter may not align | |
6989 | * nicely with memmap sections | |
6990 | * | |
6991 | * Explicitly initialize those struct pages so that: | |
6992 | * - PG_Reserved is set | |
6993 | * - zone and node links point to zone and node that span the page if the | |
6994 | * hole is in the middle of a zone | |
6995 | * - zone and node links point to adjacent zone/node if the hole falls on | |
6996 | * the zone boundary; the pages in such holes will be prepended to the | |
6997 | * zone/node above the hole except for the trailing pages in the last | |
6998 | * section that will be appended to the zone/node below. | |
6999 | */ | |
122e093c MR |
7000 | static void __init init_unavailable_range(unsigned long spfn, |
7001 | unsigned long epfn, | |
7002 | int zone, int node) | |
0740a50b MR |
7003 | { |
7004 | unsigned long pfn; | |
7005 | u64 pgcnt = 0; | |
7006 | ||
7007 | for (pfn = spfn; pfn < epfn; pfn++) { | |
4f9bc69a KW |
7008 | if (!pfn_valid(pageblock_start_pfn(pfn))) { |
7009 | pfn = pageblock_end_pfn(pfn) - 1; | |
0740a50b MR |
7010 | continue; |
7011 | } | |
7012 | __init_single_page(pfn_to_page(pfn), pfn, zone, node); | |
7013 | __SetPageReserved(pfn_to_page(pfn)); | |
7014 | pgcnt++; | |
7015 | } | |
7016 | ||
122e093c MR |
7017 | if (pgcnt) |
7018 | pr_info("On node %d, zone %s: %lld pages in unavailable ranges", | |
7019 | node, zone_names[zone], pgcnt); | |
0740a50b | 7020 | } |
0740a50b | 7021 | |
122e093c MR |
7022 | static void __init memmap_init_zone_range(struct zone *zone, |
7023 | unsigned long start_pfn, | |
7024 | unsigned long end_pfn, | |
7025 | unsigned long *hole_pfn) | |
dfb3ccd0 | 7026 | { |
3256ff83 BH |
7027 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
7028 | unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages; | |
122e093c MR |
7029 | int nid = zone_to_nid(zone), zone_id = zone_idx(zone); |
7030 | ||
7031 | start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn); | |
7032 | end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn); | |
7033 | ||
7034 | if (start_pfn >= end_pfn) | |
7035 | return; | |
7036 | ||
7037 | memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn, | |
7038 | zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE); | |
7039 | ||
7040 | if (*hole_pfn < start_pfn) | |
7041 | init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid); | |
7042 | ||
7043 | *hole_pfn = end_pfn; | |
7044 | } | |
7045 | ||
7046 | static void __init memmap_init(void) | |
7047 | { | |
73a6e474 | 7048 | unsigned long start_pfn, end_pfn; |
122e093c | 7049 | unsigned long hole_pfn = 0; |
b346075f | 7050 | int i, j, zone_id = 0, nid; |
73a6e474 | 7051 | |
122e093c MR |
7052 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
7053 | struct pglist_data *node = NODE_DATA(nid); | |
73a6e474 | 7054 | |
122e093c MR |
7055 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7056 | struct zone *zone = node->node_zones + j; | |
0740a50b | 7057 | |
122e093c MR |
7058 | if (!populated_zone(zone)) |
7059 | continue; | |
0740a50b | 7060 | |
122e093c MR |
7061 | memmap_init_zone_range(zone, start_pfn, end_pfn, |
7062 | &hole_pfn); | |
7063 | zone_id = j; | |
7064 | } | |
73a6e474 | 7065 | } |
0740a50b MR |
7066 | |
7067 | #ifdef CONFIG_SPARSEMEM | |
7068 | /* | |
122e093c MR |
7069 | * Initialize the memory map for hole in the range [memory_end, |
7070 | * section_end]. | |
7071 | * Append the pages in this hole to the highest zone in the last | |
7072 | * node. | |
7073 | * The call to init_unavailable_range() is outside the ifdef to | |
7074 | * silence the compiler warining about zone_id set but not used; | |
7075 | * for FLATMEM it is a nop anyway | |
0740a50b | 7076 | */ |
122e093c | 7077 | end_pfn = round_up(end_pfn, PAGES_PER_SECTION); |
0740a50b | 7078 | if (hole_pfn < end_pfn) |
0740a50b | 7079 | #endif |
122e093c | 7080 | init_unavailable_range(hole_pfn, end_pfn, zone_id, nid); |
dfb3ccd0 | 7081 | } |
1da177e4 | 7082 | |
c803b3c8 MR |
7083 | void __init *memmap_alloc(phys_addr_t size, phys_addr_t align, |
7084 | phys_addr_t min_addr, int nid, bool exact_nid) | |
7085 | { | |
7086 | void *ptr; | |
7087 | ||
7088 | if (exact_nid) | |
7089 | ptr = memblock_alloc_exact_nid_raw(size, align, min_addr, | |
7090 | MEMBLOCK_ALLOC_ACCESSIBLE, | |
7091 | nid); | |
7092 | else | |
7093 | ptr = memblock_alloc_try_nid_raw(size, align, min_addr, | |
7094 | MEMBLOCK_ALLOC_ACCESSIBLE, | |
7095 | nid); | |
7096 | ||
7097 | if (ptr && size > 0) | |
7098 | page_init_poison(ptr, size); | |
7099 | ||
7100 | return ptr; | |
7101 | } | |
7102 | ||
7cd2b0a3 | 7103 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 7104 | { |
3a6be87f | 7105 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
7106 | int batch; |
7107 | ||
7108 | /* | |
b92ca18e MG |
7109 | * The number of pages to batch allocate is either ~0.1% |
7110 | * of the zone or 1MB, whichever is smaller. The batch | |
7111 | * size is striking a balance between allocation latency | |
7112 | * and zone lock contention. | |
e7c8d5c9 | 7113 | */ |
c940e020 | 7114 | batch = min(zone_managed_pages(zone) >> 10, SZ_1M / PAGE_SIZE); |
e7c8d5c9 CL |
7115 | batch /= 4; /* We effectively *= 4 below */ |
7116 | if (batch < 1) | |
7117 | batch = 1; | |
7118 | ||
7119 | /* | |
0ceaacc9 NP |
7120 | * Clamp the batch to a 2^n - 1 value. Having a power |
7121 | * of 2 value was found to be more likely to have | |
7122 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 7123 | * |
0ceaacc9 NP |
7124 | * For example if 2 tasks are alternately allocating |
7125 | * batches of pages, one task can end up with a lot | |
7126 | * of pages of one half of the possible page colors | |
7127 | * and the other with pages of the other colors. | |
e7c8d5c9 | 7128 | */ |
9155203a | 7129 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 7130 | |
e7c8d5c9 | 7131 | return batch; |
3a6be87f DH |
7132 | |
7133 | #else | |
7134 | /* The deferral and batching of frees should be suppressed under NOMMU | |
7135 | * conditions. | |
7136 | * | |
7137 | * The problem is that NOMMU needs to be able to allocate large chunks | |
7138 | * of contiguous memory as there's no hardware page translation to | |
7139 | * assemble apparent contiguous memory from discontiguous pages. | |
7140 | * | |
7141 | * Queueing large contiguous runs of pages for batching, however, | |
7142 | * causes the pages to actually be freed in smaller chunks. As there | |
7143 | * can be a significant delay between the individual batches being | |
7144 | * recycled, this leads to the once large chunks of space being | |
7145 | * fragmented and becoming unavailable for high-order allocations. | |
7146 | */ | |
7147 | return 0; | |
7148 | #endif | |
e7c8d5c9 CL |
7149 | } |
7150 | ||
04f8cfea | 7151 | static int zone_highsize(struct zone *zone, int batch, int cpu_online) |
b92ca18e MG |
7152 | { |
7153 | #ifdef CONFIG_MMU | |
7154 | int high; | |
203c06ee | 7155 | int nr_split_cpus; |
74f44822 MG |
7156 | unsigned long total_pages; |
7157 | ||
7158 | if (!percpu_pagelist_high_fraction) { | |
7159 | /* | |
7160 | * By default, the high value of the pcp is based on the zone | |
7161 | * low watermark so that if they are full then background | |
7162 | * reclaim will not be started prematurely. | |
7163 | */ | |
7164 | total_pages = low_wmark_pages(zone); | |
7165 | } else { | |
7166 | /* | |
7167 | * If percpu_pagelist_high_fraction is configured, the high | |
7168 | * value is based on a fraction of the managed pages in the | |
7169 | * zone. | |
7170 | */ | |
7171 | total_pages = zone_managed_pages(zone) / percpu_pagelist_high_fraction; | |
7172 | } | |
b92ca18e MG |
7173 | |
7174 | /* | |
74f44822 MG |
7175 | * Split the high value across all online CPUs local to the zone. Note |
7176 | * that early in boot that CPUs may not be online yet and that during | |
7177 | * CPU hotplug that the cpumask is not yet updated when a CPU is being | |
203c06ee MG |
7178 | * onlined. For memory nodes that have no CPUs, split pcp->high across |
7179 | * all online CPUs to mitigate the risk that reclaim is triggered | |
7180 | * prematurely due to pages stored on pcp lists. | |
b92ca18e | 7181 | */ |
203c06ee MG |
7182 | nr_split_cpus = cpumask_weight(cpumask_of_node(zone_to_nid(zone))) + cpu_online; |
7183 | if (!nr_split_cpus) | |
7184 | nr_split_cpus = num_online_cpus(); | |
7185 | high = total_pages / nr_split_cpus; | |
b92ca18e MG |
7186 | |
7187 | /* | |
7188 | * Ensure high is at least batch*4. The multiple is based on the | |
7189 | * historical relationship between high and batch. | |
7190 | */ | |
7191 | high = max(high, batch << 2); | |
7192 | ||
7193 | return high; | |
7194 | #else | |
7195 | return 0; | |
7196 | #endif | |
7197 | } | |
7198 | ||
8d7a8fa9 | 7199 | /* |
5c3ad2eb VB |
7200 | * pcp->high and pcp->batch values are related and generally batch is lower |
7201 | * than high. They are also related to pcp->count such that count is lower | |
7202 | * than high, and as soon as it reaches high, the pcplist is flushed. | |
8d7a8fa9 | 7203 | * |
5c3ad2eb VB |
7204 | * However, guaranteeing these relations at all times would require e.g. write |
7205 | * barriers here but also careful usage of read barriers at the read side, and | |
7206 | * thus be prone to error and bad for performance. Thus the update only prevents | |
7207 | * store tearing. Any new users of pcp->batch and pcp->high should ensure they | |
7208 | * can cope with those fields changing asynchronously, and fully trust only the | |
7209 | * pcp->count field on the local CPU with interrupts disabled. | |
8d7a8fa9 CS |
7210 | * |
7211 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
7212 | * outside of boot time (or some other assurance that no concurrent updaters | |
7213 | * exist). | |
7214 | */ | |
7215 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
7216 | unsigned long batch) | |
7217 | { | |
5c3ad2eb VB |
7218 | WRITE_ONCE(pcp->batch, batch); |
7219 | WRITE_ONCE(pcp->high, high); | |
8d7a8fa9 CS |
7220 | } |
7221 | ||
28f836b6 | 7222 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats) |
2caaad41 | 7223 | { |
44042b44 | 7224 | int pindex; |
2caaad41 | 7225 | |
28f836b6 MG |
7226 | memset(pcp, 0, sizeof(*pcp)); |
7227 | memset(pzstats, 0, sizeof(*pzstats)); | |
1c6fe946 | 7228 | |
4b23a68f | 7229 | spin_lock_init(&pcp->lock); |
44042b44 MG |
7230 | for (pindex = 0; pindex < NR_PCP_LISTS; pindex++) |
7231 | INIT_LIST_HEAD(&pcp->lists[pindex]); | |
2caaad41 | 7232 | |
69a8396a VB |
7233 | /* |
7234 | * Set batch and high values safe for a boot pageset. A true percpu | |
7235 | * pageset's initialization will update them subsequently. Here we don't | |
7236 | * need to be as careful as pageset_update() as nobody can access the | |
7237 | * pageset yet. | |
7238 | */ | |
952eaf81 VB |
7239 | pcp->high = BOOT_PAGESET_HIGH; |
7240 | pcp->batch = BOOT_PAGESET_BATCH; | |
3b12e7e9 | 7241 | pcp->free_factor = 0; |
88c90dbc CS |
7242 | } |
7243 | ||
3b1f3658 | 7244 | static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high, |
ec6e8c7e VB |
7245 | unsigned long batch) |
7246 | { | |
28f836b6 | 7247 | struct per_cpu_pages *pcp; |
ec6e8c7e VB |
7248 | int cpu; |
7249 | ||
7250 | for_each_possible_cpu(cpu) { | |
28f836b6 MG |
7251 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
7252 | pageset_update(pcp, high, batch); | |
ec6e8c7e VB |
7253 | } |
7254 | } | |
7255 | ||
8ad4b1fb | 7256 | /* |
0a8b4f1d | 7257 | * Calculate and set new high and batch values for all per-cpu pagesets of a |
bbbecb35 | 7258 | * zone based on the zone's size. |
8ad4b1fb | 7259 | */ |
04f8cfea | 7260 | static void zone_set_pageset_high_and_batch(struct zone *zone, int cpu_online) |
56cef2b8 | 7261 | { |
b92ca18e | 7262 | int new_high, new_batch; |
7115ac6e | 7263 | |
b92ca18e | 7264 | new_batch = max(1, zone_batchsize(zone)); |
04f8cfea | 7265 | new_high = zone_highsize(zone, new_batch, cpu_online); |
169f6c19 | 7266 | |
952eaf81 VB |
7267 | if (zone->pageset_high == new_high && |
7268 | zone->pageset_batch == new_batch) | |
7269 | return; | |
7270 | ||
7271 | zone->pageset_high = new_high; | |
7272 | zone->pageset_batch = new_batch; | |
7273 | ||
ec6e8c7e | 7274 | __zone_set_pageset_high_and_batch(zone, new_high, new_batch); |
169f6c19 CS |
7275 | } |
7276 | ||
72675e13 | 7277 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
7278 | { |
7279 | int cpu; | |
0a8b4f1d | 7280 | |
28f836b6 MG |
7281 | /* Size may be 0 on !SMP && !NUMA */ |
7282 | if (sizeof(struct per_cpu_zonestat) > 0) | |
7283 | zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat); | |
7284 | ||
7285 | zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages); | |
0a8b4f1d | 7286 | for_each_possible_cpu(cpu) { |
28f836b6 MG |
7287 | struct per_cpu_pages *pcp; |
7288 | struct per_cpu_zonestat *pzstats; | |
7289 | ||
7290 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); | |
7291 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); | |
7292 | per_cpu_pages_init(pcp, pzstats); | |
0a8b4f1d VB |
7293 | } |
7294 | ||
04f8cfea | 7295 | zone_set_pageset_high_and_batch(zone, 0); |
319774e2 WF |
7296 | } |
7297 | ||
b89f1735 ML |
7298 | /* |
7299 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7300 | * page high values need to be recalculated. | |
7301 | */ | |
7302 | static void zone_pcp_update(struct zone *zone, int cpu_online) | |
7303 | { | |
7304 | mutex_lock(&pcp_batch_high_lock); | |
7305 | zone_set_pageset_high_and_batch(zone, cpu_online); | |
7306 | mutex_unlock(&pcp_batch_high_lock); | |
7307 | } | |
7308 | ||
2caaad41 | 7309 | /* |
99dcc3e5 CL |
7310 | * Allocate per cpu pagesets and initialize them. |
7311 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 7312 | */ |
99dcc3e5 | 7313 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 7314 | { |
b4911ea2 | 7315 | struct pglist_data *pgdat; |
99dcc3e5 | 7316 | struct zone *zone; |
b418a0f9 | 7317 | int __maybe_unused cpu; |
e7c8d5c9 | 7318 | |
319774e2 WF |
7319 | for_each_populated_zone(zone) |
7320 | setup_zone_pageset(zone); | |
b4911ea2 | 7321 | |
b418a0f9 SD |
7322 | #ifdef CONFIG_NUMA |
7323 | /* | |
7324 | * Unpopulated zones continue using the boot pagesets. | |
7325 | * The numa stats for these pagesets need to be reset. | |
7326 | * Otherwise, they will end up skewing the stats of | |
7327 | * the nodes these zones are associated with. | |
7328 | */ | |
7329 | for_each_possible_cpu(cpu) { | |
28f836b6 | 7330 | struct per_cpu_zonestat *pzstats = &per_cpu(boot_zonestats, cpu); |
f19298b9 MG |
7331 | memset(pzstats->vm_numa_event, 0, |
7332 | sizeof(pzstats->vm_numa_event)); | |
b418a0f9 SD |
7333 | } |
7334 | #endif | |
7335 | ||
b4911ea2 MG |
7336 | for_each_online_pgdat(pgdat) |
7337 | pgdat->per_cpu_nodestats = | |
7338 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
7339 | } |
7340 | ||
c09b4240 | 7341 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 7342 | { |
99dcc3e5 CL |
7343 | /* |
7344 | * per cpu subsystem is not up at this point. The following code | |
7345 | * relies on the ability of the linker to provide the | |
7346 | * offset of a (static) per cpu variable into the per cpu area. | |
7347 | */ | |
28f836b6 MG |
7348 | zone->per_cpu_pageset = &boot_pageset; |
7349 | zone->per_cpu_zonestats = &boot_zonestats; | |
952eaf81 VB |
7350 | zone->pageset_high = BOOT_PAGESET_HIGH; |
7351 | zone->pageset_batch = BOOT_PAGESET_BATCH; | |
ed8ece2e | 7352 | |
b38a8725 | 7353 | if (populated_zone(zone)) |
9660ecaa HK |
7354 | pr_debug(" %s zone: %lu pages, LIFO batch:%u\n", zone->name, |
7355 | zone->present_pages, zone_batchsize(zone)); | |
ed8ece2e DH |
7356 | } |
7357 | ||
dc0bbf3b | 7358 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 7359 | unsigned long zone_start_pfn, |
b171e409 | 7360 | unsigned long size) |
ed8ece2e DH |
7361 | { |
7362 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 7363 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 7364 | |
8f416836 WY |
7365 | if (zone_idx > pgdat->nr_zones) |
7366 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 7367 | |
ed8ece2e DH |
7368 | zone->zone_start_pfn = zone_start_pfn; |
7369 | ||
708614e6 MG |
7370 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
7371 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
7372 | pgdat->node_id, | |
7373 | (unsigned long)zone_idx(zone), | |
7374 | zone_start_pfn, (zone_start_pfn + size)); | |
7375 | ||
1e548deb | 7376 | zone_init_free_lists(zone); |
9dcb8b68 | 7377 | zone->initialized = 1; |
ed8ece2e DH |
7378 | } |
7379 | ||
c713216d MG |
7380 | /** |
7381 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
7382 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
7383 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
7384 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
7385 | * |
7386 | * It returns the start and end page frame of a node based on information | |
7d018176 | 7387 | * provided by memblock_set_node(). If called for a node |
c713216d | 7388 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 7389 | * PFNs will be 0. |
c713216d | 7390 | */ |
bbe5d993 | 7391 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
7392 | unsigned long *start_pfn, unsigned long *end_pfn) |
7393 | { | |
c13291a5 | 7394 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 7395 | int i; |
c13291a5 | 7396 | |
c713216d MG |
7397 | *start_pfn = -1UL; |
7398 | *end_pfn = 0; | |
7399 | ||
c13291a5 TH |
7400 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
7401 | *start_pfn = min(*start_pfn, this_start_pfn); | |
7402 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
7403 | } |
7404 | ||
633c0666 | 7405 | if (*start_pfn == -1UL) |
c713216d | 7406 | *start_pfn = 0; |
c713216d MG |
7407 | } |
7408 | ||
2a1e274a MG |
7409 | /* |
7410 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
7411 | * assumption is made that zones within a node are ordered in monotonic | |
7412 | * increasing memory addresses so that the "highest" populated zone is used | |
7413 | */ | |
b69a7288 | 7414 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
7415 | { |
7416 | int zone_index; | |
7417 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
7418 | if (zone_index == ZONE_MOVABLE) | |
7419 | continue; | |
7420 | ||
7421 | if (arch_zone_highest_possible_pfn[zone_index] > | |
7422 | arch_zone_lowest_possible_pfn[zone_index]) | |
7423 | break; | |
7424 | } | |
7425 | ||
7426 | VM_BUG_ON(zone_index == -1); | |
7427 | movable_zone = zone_index; | |
7428 | } | |
7429 | ||
7430 | /* | |
7431 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 7432 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
7433 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
7434 | * in each node depending on the size of each node and how evenly kernelcore | |
7435 | * is distributed. This helper function adjusts the zone ranges | |
7436 | * provided by the architecture for a given node by using the end of the | |
7437 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
7438 | * zones within a node are in order of monotonic increases memory addresses | |
7439 | */ | |
bbe5d993 | 7440 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
7441 | unsigned long zone_type, |
7442 | unsigned long node_start_pfn, | |
7443 | unsigned long node_end_pfn, | |
7444 | unsigned long *zone_start_pfn, | |
7445 | unsigned long *zone_end_pfn) | |
7446 | { | |
7447 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
7448 | if (zone_movable_pfn[nid]) { | |
7449 | /* Size ZONE_MOVABLE */ | |
7450 | if (zone_type == ZONE_MOVABLE) { | |
7451 | *zone_start_pfn = zone_movable_pfn[nid]; | |
7452 | *zone_end_pfn = min(node_end_pfn, | |
7453 | arch_zone_highest_possible_pfn[movable_zone]); | |
7454 | ||
e506b996 XQ |
7455 | /* Adjust for ZONE_MOVABLE starting within this range */ |
7456 | } else if (!mirrored_kernelcore && | |
7457 | *zone_start_pfn < zone_movable_pfn[nid] && | |
7458 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
7459 | *zone_end_pfn = zone_movable_pfn[nid]; | |
7460 | ||
2a1e274a MG |
7461 | /* Check if this whole range is within ZONE_MOVABLE */ |
7462 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
7463 | *zone_start_pfn = *zone_end_pfn; | |
7464 | } | |
7465 | } | |
7466 | ||
c713216d MG |
7467 | /* |
7468 | * Return the number of pages a zone spans in a node, including holes | |
7469 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
7470 | */ | |
bbe5d993 | 7471 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 7472 | unsigned long zone_type, |
7960aedd ZY |
7473 | unsigned long node_start_pfn, |
7474 | unsigned long node_end_pfn, | |
d91749c1 | 7475 | unsigned long *zone_start_pfn, |
854e8848 | 7476 | unsigned long *zone_end_pfn) |
c713216d | 7477 | { |
299c83dc LF |
7478 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
7479 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 7480 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
7481 | if (!node_start_pfn && !node_end_pfn) |
7482 | return 0; | |
7483 | ||
7960aedd | 7484 | /* Get the start and end of the zone */ |
299c83dc LF |
7485 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
7486 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
7487 | adjust_zone_range_for_zone_movable(nid, zone_type, |
7488 | node_start_pfn, node_end_pfn, | |
d91749c1 | 7489 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
7490 | |
7491 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 7492 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
7493 | return 0; |
7494 | ||
7495 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
7496 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
7497 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
7498 | |
7499 | /* Return the spanned pages */ | |
d91749c1 | 7500 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
7501 | } |
7502 | ||
7503 | /* | |
7504 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 7505 | * then all holes in the requested range will be accounted for. |
c713216d | 7506 | */ |
bbe5d993 | 7507 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
7508 | unsigned long range_start_pfn, |
7509 | unsigned long range_end_pfn) | |
7510 | { | |
96e907d1 TH |
7511 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
7512 | unsigned long start_pfn, end_pfn; | |
7513 | int i; | |
c713216d | 7514 | |
96e907d1 TH |
7515 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
7516 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
7517 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
7518 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 7519 | } |
96e907d1 | 7520 | return nr_absent; |
c713216d MG |
7521 | } |
7522 | ||
7523 | /** | |
7524 | * absent_pages_in_range - Return number of page frames in holes within a range | |
7525 | * @start_pfn: The start PFN to start searching for holes | |
7526 | * @end_pfn: The end PFN to stop searching for holes | |
7527 | * | |
a862f68a | 7528 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
7529 | */ |
7530 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
7531 | unsigned long end_pfn) | |
7532 | { | |
7533 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
7534 | } | |
7535 | ||
7536 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 7537 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 7538 | unsigned long zone_type, |
7960aedd | 7539 | unsigned long node_start_pfn, |
854e8848 | 7540 | unsigned long node_end_pfn) |
c713216d | 7541 | { |
96e907d1 TH |
7542 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
7543 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 7544 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 7545 | unsigned long nr_absent; |
9c7cd687 | 7546 | |
b5685e92 | 7547 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
7548 | if (!node_start_pfn && !node_end_pfn) |
7549 | return 0; | |
7550 | ||
96e907d1 TH |
7551 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
7552 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 7553 | |
2a1e274a MG |
7554 | adjust_zone_range_for_zone_movable(nid, zone_type, |
7555 | node_start_pfn, node_end_pfn, | |
7556 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
7557 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
7558 | ||
7559 | /* | |
7560 | * ZONE_MOVABLE handling. | |
7561 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
7562 | * and vice versa. | |
7563 | */ | |
e506b996 XQ |
7564 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
7565 | unsigned long start_pfn, end_pfn; | |
7566 | struct memblock_region *r; | |
7567 | ||
cc6de168 | 7568 | for_each_mem_region(r) { |
e506b996 XQ |
7569 | start_pfn = clamp(memblock_region_memory_base_pfn(r), |
7570 | zone_start_pfn, zone_end_pfn); | |
7571 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
7572 | zone_start_pfn, zone_end_pfn); | |
7573 | ||
7574 | if (zone_type == ZONE_MOVABLE && | |
7575 | memblock_is_mirror(r)) | |
7576 | nr_absent += end_pfn - start_pfn; | |
7577 | ||
7578 | if (zone_type == ZONE_NORMAL && | |
7579 | !memblock_is_mirror(r)) | |
7580 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
7581 | } |
7582 | } | |
7583 | ||
7584 | return nr_absent; | |
c713216d | 7585 | } |
0e0b864e | 7586 | |
bbe5d993 | 7587 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd | 7588 | unsigned long node_start_pfn, |
854e8848 | 7589 | unsigned long node_end_pfn) |
c713216d | 7590 | { |
febd5949 | 7591 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
7592 | enum zone_type i; |
7593 | ||
febd5949 GZ |
7594 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7595 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 7596 | unsigned long zone_start_pfn, zone_end_pfn; |
3f08a302 | 7597 | unsigned long spanned, absent; |
febd5949 | 7598 | unsigned long size, real_size; |
c713216d | 7599 | |
854e8848 MR |
7600 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i, |
7601 | node_start_pfn, | |
7602 | node_end_pfn, | |
7603 | &zone_start_pfn, | |
7604 | &zone_end_pfn); | |
7605 | absent = zone_absent_pages_in_node(pgdat->node_id, i, | |
7606 | node_start_pfn, | |
7607 | node_end_pfn); | |
3f08a302 MR |
7608 | |
7609 | size = spanned; | |
7610 | real_size = size - absent; | |
7611 | ||
d91749c1 TI |
7612 | if (size) |
7613 | zone->zone_start_pfn = zone_start_pfn; | |
7614 | else | |
7615 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
7616 | zone->spanned_pages = size; |
7617 | zone->present_pages = real_size; | |
4b097002 DH |
7618 | #if defined(CONFIG_MEMORY_HOTPLUG) |
7619 | zone->present_early_pages = real_size; | |
7620 | #endif | |
febd5949 GZ |
7621 | |
7622 | totalpages += size; | |
7623 | realtotalpages += real_size; | |
7624 | } | |
7625 | ||
7626 | pgdat->node_spanned_pages = totalpages; | |
c713216d | 7627 | pgdat->node_present_pages = realtotalpages; |
9660ecaa | 7628 | pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); |
c713216d MG |
7629 | } |
7630 | ||
835c134e MG |
7631 | #ifndef CONFIG_SPARSEMEM |
7632 | /* | |
7633 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
7634 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
7635 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
7636 | * round what is now in bits to nearest long in bits, then return it in |
7637 | * bytes. | |
7638 | */ | |
7c45512d | 7639 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
7640 | { |
7641 | unsigned long usemapsize; | |
7642 | ||
7c45512d | 7643 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
7644 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
7645 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
7646 | usemapsize *= NR_PAGEBLOCK_BITS; |
7647 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
7648 | ||
7649 | return usemapsize / 8; | |
7650 | } | |
7651 | ||
7010a6ec | 7652 | static void __ref setup_usemap(struct zone *zone) |
835c134e | 7653 | { |
7010a6ec BH |
7654 | unsigned long usemapsize = usemap_size(zone->zone_start_pfn, |
7655 | zone->spanned_pages); | |
835c134e | 7656 | zone->pageblock_flags = NULL; |
23a7052a | 7657 | if (usemapsize) { |
6782832e | 7658 | zone->pageblock_flags = |
26fb3dae | 7659 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
7010a6ec | 7660 | zone_to_nid(zone)); |
23a7052a MR |
7661 | if (!zone->pageblock_flags) |
7662 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
7010a6ec | 7663 | usemapsize, zone->name, zone_to_nid(zone)); |
23a7052a | 7664 | } |
835c134e MG |
7665 | } |
7666 | #else | |
7010a6ec | 7667 | static inline void setup_usemap(struct zone *zone) {} |
835c134e MG |
7668 | #endif /* CONFIG_SPARSEMEM */ |
7669 | ||
d9c23400 | 7670 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 7671 | |
d9c23400 | 7672 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 7673 | void __init set_pageblock_order(void) |
d9c23400 | 7674 | { |
b3d40a2b | 7675 | unsigned int order = MAX_ORDER - 1; |
955c1cd7 | 7676 | |
d9c23400 MG |
7677 | /* Check that pageblock_nr_pages has not already been setup */ |
7678 | if (pageblock_order) | |
7679 | return; | |
7680 | ||
b3d40a2b DH |
7681 | /* Don't let pageblocks exceed the maximum allocation granularity. */ |
7682 | if (HPAGE_SHIFT > PAGE_SHIFT && HUGETLB_PAGE_ORDER < order) | |
955c1cd7 | 7683 | order = HUGETLB_PAGE_ORDER; |
955c1cd7 | 7684 | |
d9c23400 MG |
7685 | /* |
7686 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
7687 | * This value may be variable depending on boot parameters on IA64 and |
7688 | * powerpc. | |
d9c23400 MG |
7689 | */ |
7690 | pageblock_order = order; | |
7691 | } | |
7692 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7693 | ||
ba72cb8c MG |
7694 | /* |
7695 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
7696 | * is unused as pageblock_order is set at compile-time. See |
7697 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
7698 | * the kernel config | |
ba72cb8c | 7699 | */ |
03e85f9d | 7700 | void __init set_pageblock_order(void) |
ba72cb8c | 7701 | { |
ba72cb8c | 7702 | } |
d9c23400 MG |
7703 | |
7704 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7705 | ||
03e85f9d | 7706 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 7707 | unsigned long present_pages) |
01cefaef JL |
7708 | { |
7709 | unsigned long pages = spanned_pages; | |
7710 | ||
7711 | /* | |
7712 | * Provide a more accurate estimation if there are holes within | |
7713 | * the zone and SPARSEMEM is in use. If there are holes within the | |
7714 | * zone, each populated memory region may cost us one or two extra | |
7715 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 7716 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
7717 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
7718 | */ | |
7719 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
7720 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
7721 | pages = present_pages; | |
7722 | ||
7723 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
7724 | } | |
7725 | ||
ace1db39 OS |
7726 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
7727 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
7728 | { | |
364c1eeb YS |
7729 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
7730 | ||
7731 | spin_lock_init(&ds_queue->split_queue_lock); | |
7732 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
7733 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
7734 | } |
7735 | #else | |
7736 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
7737 | #endif | |
7738 | ||
7739 | #ifdef CONFIG_COMPACTION | |
7740 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
7741 | { | |
7742 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
7743 | } | |
7744 | #else | |
7745 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
7746 | #endif | |
7747 | ||
03e85f9d | 7748 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 7749 | { |
8cd7c588 MG |
7750 | int i; |
7751 | ||
208d54e5 | 7752 | pgdat_resize_init(pgdat); |
b4a0215e | 7753 | pgdat_kswapd_lock_init(pgdat); |
ace1db39 | 7754 | |
ace1db39 OS |
7755 | pgdat_init_split_queue(pgdat); |
7756 | pgdat_init_kcompactd(pgdat); | |
7757 | ||
1da177e4 | 7758 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 7759 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 7760 | |
8cd7c588 MG |
7761 | for (i = 0; i < NR_VMSCAN_THROTTLE; i++) |
7762 | init_waitqueue_head(&pgdat->reclaim_wait[i]); | |
7763 | ||
eefa864b | 7764 | pgdat_page_ext_init(pgdat); |
867e5e1d | 7765 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
7766 | } |
7767 | ||
7768 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
7769 | unsigned long remaining_pages) | |
7770 | { | |
9705bea5 | 7771 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
7772 | zone_set_nid(zone, nid); |
7773 | zone->name = zone_names[idx]; | |
7774 | zone->zone_pgdat = NODE_DATA(nid); | |
7775 | spin_lock_init(&zone->lock); | |
7776 | zone_seqlock_init(zone); | |
7777 | zone_pcp_init(zone); | |
7778 | } | |
7779 | ||
7780 | /* | |
7781 | * Set up the zone data structures | |
7782 | * - init pgdat internals | |
7783 | * - init all zones belonging to this node | |
7784 | * | |
7785 | * NOTE: this function is only called during memory hotplug | |
7786 | */ | |
7787 | #ifdef CONFIG_MEMORY_HOTPLUG | |
70b5b46a | 7788 | void __ref free_area_init_core_hotplug(struct pglist_data *pgdat) |
03e85f9d | 7789 | { |
70b5b46a | 7790 | int nid = pgdat->node_id; |
03e85f9d | 7791 | enum zone_type z; |
70b5b46a | 7792 | int cpu; |
03e85f9d OS |
7793 | |
7794 | pgdat_init_internals(pgdat); | |
70b5b46a MH |
7795 | |
7796 | if (pgdat->per_cpu_nodestats == &boot_nodestats) | |
7797 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); | |
7798 | ||
7799 | /* | |
7800 | * Reset the nr_zones, order and highest_zoneidx before reuse. | |
7801 | * Note that kswapd will init kswapd_highest_zoneidx properly | |
7802 | * when it starts in the near future. | |
7803 | */ | |
7804 | pgdat->nr_zones = 0; | |
7805 | pgdat->kswapd_order = 0; | |
7806 | pgdat->kswapd_highest_zoneidx = 0; | |
7807 | pgdat->node_start_pfn = 0; | |
7808 | for_each_online_cpu(cpu) { | |
7809 | struct per_cpu_nodestat *p; | |
7810 | ||
7811 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
7812 | memset(p, 0, sizeof(*p)); | |
7813 | } | |
7814 | ||
03e85f9d OS |
7815 | for (z = 0; z < MAX_NR_ZONES; z++) |
7816 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
7817 | } | |
7818 | #endif | |
7819 | ||
7820 | /* | |
7821 | * Set up the zone data structures: | |
7822 | * - mark all pages reserved | |
7823 | * - mark all memory queues empty | |
7824 | * - clear the memory bitmaps | |
7825 | * | |
7826 | * NOTE: pgdat should get zeroed by caller. | |
7827 | * NOTE: this function is only called during early init. | |
7828 | */ | |
7829 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
7830 | { | |
7831 | enum zone_type j; | |
7832 | int nid = pgdat->node_id; | |
5f63b720 | 7833 | |
03e85f9d | 7834 | pgdat_init_internals(pgdat); |
385386cf JW |
7835 | pgdat->per_cpu_nodestats = &boot_nodestats; |
7836 | ||
1da177e4 LT |
7837 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7838 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 7839 | unsigned long size, freesize, memmap_pages; |
1da177e4 | 7840 | |
febd5949 | 7841 | size = zone->spanned_pages; |
e6943859 | 7842 | freesize = zone->present_pages; |
1da177e4 | 7843 | |
0e0b864e | 7844 | /* |
9feedc9d | 7845 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
7846 | * is used by this zone for memmap. This affects the watermark |
7847 | * and per-cpu initialisations | |
7848 | */ | |
e6943859 | 7849 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
7850 | if (!is_highmem_idx(j)) { |
7851 | if (freesize >= memmap_pages) { | |
7852 | freesize -= memmap_pages; | |
7853 | if (memmap_pages) | |
9660ecaa HK |
7854 | pr_debug(" %s zone: %lu pages used for memmap\n", |
7855 | zone_names[j], memmap_pages); | |
ba914f48 | 7856 | } else |
e47aa905 | 7857 | pr_warn(" %s zone: %lu memmap pages exceeds freesize %lu\n", |
ba914f48 ZH |
7858 | zone_names[j], memmap_pages, freesize); |
7859 | } | |
0e0b864e | 7860 | |
6267276f | 7861 | /* Account for reserved pages */ |
9feedc9d JL |
7862 | if (j == 0 && freesize > dma_reserve) { |
7863 | freesize -= dma_reserve; | |
9660ecaa | 7864 | pr_debug(" %s zone: %lu pages reserved\n", zone_names[0], dma_reserve); |
0e0b864e MG |
7865 | } |
7866 | ||
98d2b0eb | 7867 | if (!is_highmem_idx(j)) |
9feedc9d | 7868 | nr_kernel_pages += freesize; |
01cefaef JL |
7869 | /* Charge for highmem memmap if there are enough kernel pages */ |
7870 | else if (nr_kernel_pages > memmap_pages * 2) | |
7871 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 7872 | nr_all_pages += freesize; |
1da177e4 | 7873 | |
9feedc9d JL |
7874 | /* |
7875 | * Set an approximate value for lowmem here, it will be adjusted | |
7876 | * when the bootmem allocator frees pages into the buddy system. | |
7877 | * And all highmem pages will be managed by the buddy system. | |
7878 | */ | |
03e85f9d | 7879 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 7880 | |
d883c6cf | 7881 | if (!size) |
1da177e4 LT |
7882 | continue; |
7883 | ||
955c1cd7 | 7884 | set_pageblock_order(); |
7010a6ec | 7885 | setup_usemap(zone); |
9699ee7b | 7886 | init_currently_empty_zone(zone, zone->zone_start_pfn, size); |
1da177e4 LT |
7887 | } |
7888 | } | |
7889 | ||
43b02ba9 | 7890 | #ifdef CONFIG_FLATMEM |
3b446da6 | 7891 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 7892 | { |
b0aeba74 | 7893 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
7894 | unsigned long __maybe_unused offset = 0; |
7895 | ||
1da177e4 LT |
7896 | /* Skip empty nodes */ |
7897 | if (!pgdat->node_spanned_pages) | |
7898 | return; | |
7899 | ||
b0aeba74 TL |
7900 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
7901 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
7902 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
7903 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 7904 | unsigned long size, end; |
d41dee36 AW |
7905 | struct page *map; |
7906 | ||
e984bb43 BP |
7907 | /* |
7908 | * The zone's endpoints aren't required to be MAX_ORDER | |
7909 | * aligned but the node_mem_map endpoints must be in order | |
7910 | * for the buddy allocator to function correctly. | |
7911 | */ | |
108bcc96 | 7912 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
7913 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
7914 | size = (end - start) * sizeof(struct page); | |
c803b3c8 MR |
7915 | map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT, |
7916 | pgdat->node_id, false); | |
23a7052a MR |
7917 | if (!map) |
7918 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
7919 | size, pgdat->node_id); | |
a1c34a3b | 7920 | pgdat->node_mem_map = map + offset; |
1da177e4 | 7921 | } |
0cd842f9 OS |
7922 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
7923 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
7924 | (unsigned long)pgdat->node_mem_map); | |
a9ee6cf5 | 7925 | #ifndef CONFIG_NUMA |
1da177e4 LT |
7926 | /* |
7927 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
7928 | */ | |
c713216d | 7929 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 7930 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d | 7931 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 7932 | mem_map -= offset; |
c713216d | 7933 | } |
1da177e4 LT |
7934 | #endif |
7935 | } | |
0cd842f9 | 7936 | #else |
3b446da6 | 7937 | static inline void alloc_node_mem_map(struct pglist_data *pgdat) { } |
43b02ba9 | 7938 | #endif /* CONFIG_FLATMEM */ |
1da177e4 | 7939 | |
0188dc98 OS |
7940 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
7941 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
7942 | { | |
0188dc98 OS |
7943 | pgdat->first_deferred_pfn = ULONG_MAX; |
7944 | } | |
7945 | #else | |
7946 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
7947 | #endif | |
7948 | ||
854e8848 | 7949 | static void __init free_area_init_node(int nid) |
1da177e4 | 7950 | { |
9109fb7b | 7951 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
7952 | unsigned long start_pfn = 0; |
7953 | unsigned long end_pfn = 0; | |
9109fb7b | 7954 | |
88fdf75d | 7955 | /* pg_data_t should be reset to zero when it's allocated */ |
97a225e6 | 7956 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx); |
88fdf75d | 7957 | |
854e8848 | 7958 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
88fdf75d | 7959 | |
1da177e4 | 7960 | pgdat->node_id = nid; |
854e8848 | 7961 | pgdat->node_start_pfn = start_pfn; |
75ef7184 | 7962 | pgdat->per_cpu_nodestats = NULL; |
854e8848 | 7963 | |
7c30daac MH |
7964 | if (start_pfn != end_pfn) { |
7965 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, | |
7966 | (u64)start_pfn << PAGE_SHIFT, | |
7967 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
7968 | } else { | |
7969 | pr_info("Initmem setup node %d as memoryless\n", nid); | |
7970 | } | |
7971 | ||
854e8848 | 7972 | calculate_node_totalpages(pgdat, start_pfn, end_pfn); |
1da177e4 LT |
7973 | |
7974 | alloc_node_mem_map(pgdat); | |
0188dc98 | 7975 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 7976 | |
7f3eb55b | 7977 | free_area_init_core(pgdat); |
e4dde56c | 7978 | lru_gen_init_pgdat(pgdat); |
1da177e4 LT |
7979 | } |
7980 | ||
1ca75fa7 | 7981 | static void __init free_area_init_memoryless_node(int nid) |
3f08a302 | 7982 | { |
854e8848 | 7983 | free_area_init_node(nid); |
3f08a302 MR |
7984 | } |
7985 | ||
418508c1 MS |
7986 | #if MAX_NUMNODES > 1 |
7987 | /* | |
7988 | * Figure out the number of possible node ids. | |
7989 | */ | |
f9872caf | 7990 | void __init setup_nr_node_ids(void) |
418508c1 | 7991 | { |
904a9553 | 7992 | unsigned int highest; |
418508c1 | 7993 | |
904a9553 | 7994 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7995 | nr_node_ids = highest + 1; |
7996 | } | |
418508c1 MS |
7997 | #endif |
7998 | ||
1e01979c TH |
7999 | /** |
8000 | * node_map_pfn_alignment - determine the maximum internode alignment | |
8001 | * | |
8002 | * This function should be called after node map is populated and sorted. | |
8003 | * It calculates the maximum power of two alignment which can distinguish | |
8004 | * all the nodes. | |
8005 | * | |
8006 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
8007 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
8008 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
8009 | * shifted, 1GiB is enough and this function will indicate so. | |
8010 | * | |
8011 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
8012 | * model has fine enough granularity to avoid incorrect mapping for the | |
8013 | * populated node map. | |
8014 | * | |
a862f68a | 8015 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
8016 | * requirement (single node). |
8017 | */ | |
8018 | unsigned long __init node_map_pfn_alignment(void) | |
8019 | { | |
8020 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 8021 | unsigned long start, end, mask; |
98fa15f3 | 8022 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 8023 | int i, nid; |
1e01979c | 8024 | |
c13291a5 | 8025 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
8026 | if (!start || last_nid < 0 || last_nid == nid) { |
8027 | last_nid = nid; | |
8028 | last_end = end; | |
8029 | continue; | |
8030 | } | |
8031 | ||
8032 | /* | |
8033 | * Start with a mask granular enough to pin-point to the | |
8034 | * start pfn and tick off bits one-by-one until it becomes | |
8035 | * too coarse to separate the current node from the last. | |
8036 | */ | |
8037 | mask = ~((1 << __ffs(start)) - 1); | |
8038 | while (mask && last_end <= (start & (mask << 1))) | |
8039 | mask <<= 1; | |
8040 | ||
8041 | /* accumulate all internode masks */ | |
8042 | accl_mask |= mask; | |
8043 | } | |
8044 | ||
8045 | /* convert mask to number of pages */ | |
8046 | return ~accl_mask + 1; | |
8047 | } | |
8048 | ||
37b07e41 LS |
8049 | /* |
8050 | * early_calculate_totalpages() | |
8051 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 8052 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 8053 | */ |
484f51f8 | 8054 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 8055 | { |
7e63efef | 8056 | unsigned long totalpages = 0; |
c13291a5 TH |
8057 | unsigned long start_pfn, end_pfn; |
8058 | int i, nid; | |
8059 | ||
8060 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
8061 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 8062 | |
37b07e41 LS |
8063 | totalpages += pages; |
8064 | if (pages) | |
4b0ef1fe | 8065 | node_set_state(nid, N_MEMORY); |
37b07e41 | 8066 | } |
b8af2941 | 8067 | return totalpages; |
7e63efef MG |
8068 | } |
8069 | ||
2a1e274a MG |
8070 | /* |
8071 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
8072 | * is spread evenly between nodes as long as the nodes have enough | |
8073 | * memory. When they don't, some nodes will have more kernelcore than | |
8074 | * others | |
8075 | */ | |
b224ef85 | 8076 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
8077 | { |
8078 | int i, nid; | |
8079 | unsigned long usable_startpfn; | |
8080 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 8081 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 8082 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 8083 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 8084 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 8085 | struct memblock_region *r; |
b2f3eebe TC |
8086 | |
8087 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
8088 | find_usable_zone_for_movable(); | |
8089 | ||
8090 | /* | |
8091 | * If movable_node is specified, ignore kernelcore and movablecore | |
8092 | * options. | |
8093 | */ | |
8094 | if (movable_node_is_enabled()) { | |
cc6de168 | 8095 | for_each_mem_region(r) { |
136199f0 | 8096 | if (!memblock_is_hotpluggable(r)) |
b2f3eebe TC |
8097 | continue; |
8098 | ||
d622abf7 | 8099 | nid = memblock_get_region_node(r); |
b2f3eebe | 8100 | |
136199f0 | 8101 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
8102 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
8103 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
8104 | usable_startpfn; | |
8105 | } | |
8106 | ||
8107 | goto out2; | |
8108 | } | |
2a1e274a | 8109 | |
342332e6 TI |
8110 | /* |
8111 | * If kernelcore=mirror is specified, ignore movablecore option | |
8112 | */ | |
8113 | if (mirrored_kernelcore) { | |
8114 | bool mem_below_4gb_not_mirrored = false; | |
8115 | ||
cc6de168 | 8116 | for_each_mem_region(r) { |
342332e6 TI |
8117 | if (memblock_is_mirror(r)) |
8118 | continue; | |
8119 | ||
d622abf7 | 8120 | nid = memblock_get_region_node(r); |
342332e6 TI |
8121 | |
8122 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
8123 | ||
aa282a15 | 8124 | if (usable_startpfn < PHYS_PFN(SZ_4G)) { |
342332e6 TI |
8125 | mem_below_4gb_not_mirrored = true; |
8126 | continue; | |
8127 | } | |
8128 | ||
8129 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
8130 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
8131 | usable_startpfn; | |
8132 | } | |
8133 | ||
8134 | if (mem_below_4gb_not_mirrored) | |
633bf2fe | 8135 | pr_warn("This configuration results in unmirrored kernel memory.\n"); |
342332e6 TI |
8136 | |
8137 | goto out2; | |
8138 | } | |
8139 | ||
7e63efef | 8140 | /* |
a5c6d650 DR |
8141 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
8142 | * amount of necessary memory. | |
8143 | */ | |
8144 | if (required_kernelcore_percent) | |
8145 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
8146 | 10000UL; | |
8147 | if (required_movablecore_percent) | |
8148 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
8149 | 10000UL; | |
8150 | ||
8151 | /* | |
8152 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
8153 | * kernelcore that corresponds so that memory usable for |
8154 | * any allocation type is evenly spread. If both kernelcore | |
8155 | * and movablecore are specified, then the value of kernelcore | |
8156 | * will be used for required_kernelcore if it's greater than | |
8157 | * what movablecore would have allowed. | |
8158 | */ | |
8159 | if (required_movablecore) { | |
7e63efef MG |
8160 | unsigned long corepages; |
8161 | ||
8162 | /* | |
8163 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
8164 | * was requested by the user | |
8165 | */ | |
8166 | required_movablecore = | |
8167 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 8168 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
8169 | corepages = totalpages - required_movablecore; |
8170 | ||
8171 | required_kernelcore = max(required_kernelcore, corepages); | |
8172 | } | |
8173 | ||
bde304bd XQ |
8174 | /* |
8175 | * If kernelcore was not specified or kernelcore size is larger | |
8176 | * than totalpages, there is no ZONE_MOVABLE. | |
8177 | */ | |
8178 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 8179 | goto out; |
2a1e274a MG |
8180 | |
8181 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
8182 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
8183 | ||
8184 | restart: | |
8185 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
8186 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 8187 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
8188 | unsigned long start_pfn, end_pfn; |
8189 | ||
2a1e274a MG |
8190 | /* |
8191 | * Recalculate kernelcore_node if the division per node | |
8192 | * now exceeds what is necessary to satisfy the requested | |
8193 | * amount of memory for the kernel | |
8194 | */ | |
8195 | if (required_kernelcore < kernelcore_node) | |
8196 | kernelcore_node = required_kernelcore / usable_nodes; | |
8197 | ||
8198 | /* | |
8199 | * As the map is walked, we track how much memory is usable | |
8200 | * by the kernel using kernelcore_remaining. When it is | |
8201 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
8202 | */ | |
8203 | kernelcore_remaining = kernelcore_node; | |
8204 | ||
8205 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 8206 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
8207 | unsigned long size_pages; |
8208 | ||
c13291a5 | 8209 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
8210 | if (start_pfn >= end_pfn) |
8211 | continue; | |
8212 | ||
8213 | /* Account for what is only usable for kernelcore */ | |
8214 | if (start_pfn < usable_startpfn) { | |
8215 | unsigned long kernel_pages; | |
8216 | kernel_pages = min(end_pfn, usable_startpfn) | |
8217 | - start_pfn; | |
8218 | ||
8219 | kernelcore_remaining -= min(kernel_pages, | |
8220 | kernelcore_remaining); | |
8221 | required_kernelcore -= min(kernel_pages, | |
8222 | required_kernelcore); | |
8223 | ||
8224 | /* Continue if range is now fully accounted */ | |
8225 | if (end_pfn <= usable_startpfn) { | |
8226 | ||
8227 | /* | |
8228 | * Push zone_movable_pfn to the end so | |
8229 | * that if we have to rebalance | |
8230 | * kernelcore across nodes, we will | |
8231 | * not double account here | |
8232 | */ | |
8233 | zone_movable_pfn[nid] = end_pfn; | |
8234 | continue; | |
8235 | } | |
8236 | start_pfn = usable_startpfn; | |
8237 | } | |
8238 | ||
8239 | /* | |
8240 | * The usable PFN range for ZONE_MOVABLE is from | |
8241 | * start_pfn->end_pfn. Calculate size_pages as the | |
8242 | * number of pages used as kernelcore | |
8243 | */ | |
8244 | size_pages = end_pfn - start_pfn; | |
8245 | if (size_pages > kernelcore_remaining) | |
8246 | size_pages = kernelcore_remaining; | |
8247 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
8248 | ||
8249 | /* | |
8250 | * Some kernelcore has been met, update counts and | |
8251 | * break if the kernelcore for this node has been | |
b8af2941 | 8252 | * satisfied |
2a1e274a MG |
8253 | */ |
8254 | required_kernelcore -= min(required_kernelcore, | |
8255 | size_pages); | |
8256 | kernelcore_remaining -= size_pages; | |
8257 | if (!kernelcore_remaining) | |
8258 | break; | |
8259 | } | |
8260 | } | |
8261 | ||
8262 | /* | |
8263 | * If there is still required_kernelcore, we do another pass with one | |
8264 | * less node in the count. This will push zone_movable_pfn[nid] further | |
8265 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 8266 | * satisfied |
2a1e274a MG |
8267 | */ |
8268 | usable_nodes--; | |
8269 | if (usable_nodes && required_kernelcore > usable_nodes) | |
8270 | goto restart; | |
8271 | ||
b2f3eebe | 8272 | out2: |
2a1e274a | 8273 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
ddbc84f3 AP |
8274 | for (nid = 0; nid < MAX_NUMNODES; nid++) { |
8275 | unsigned long start_pfn, end_pfn; | |
8276 | ||
2a1e274a MG |
8277 | zone_movable_pfn[nid] = |
8278 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 8279 | |
ddbc84f3 AP |
8280 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
8281 | if (zone_movable_pfn[nid] >= end_pfn) | |
8282 | zone_movable_pfn[nid] = 0; | |
8283 | } | |
8284 | ||
20e6926d | 8285 | out: |
66918dcd | 8286 | /* restore the node_state */ |
4b0ef1fe | 8287 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
8288 | } |
8289 | ||
4b0ef1fe LJ |
8290 | /* Any regular or high memory on that node ? */ |
8291 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 8292 | { |
37b07e41 LS |
8293 | enum zone_type zone_type; |
8294 | ||
4b0ef1fe | 8295 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 8296 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 8297 | if (populated_zone(zone)) { |
7b0e0c0e OS |
8298 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
8299 | node_set_state(nid, N_HIGH_MEMORY); | |
8300 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 8301 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
8302 | break; |
8303 | } | |
37b07e41 | 8304 | } |
37b07e41 LS |
8305 | } |
8306 | ||
51930df5 | 8307 | /* |
f0953a1b | 8308 | * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For |
51930df5 MR |
8309 | * such cases we allow max_zone_pfn sorted in the descending order |
8310 | */ | |
8311 | bool __weak arch_has_descending_max_zone_pfns(void) | |
8312 | { | |
8313 | return false; | |
8314 | } | |
8315 | ||
c713216d | 8316 | /** |
9691a071 | 8317 | * free_area_init - Initialise all pg_data_t and zone data |
88ca3b94 | 8318 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
8319 | * |
8320 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 8321 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
8322 | * zone in each node and their holes is calculated. If the maximum PFN |
8323 | * between two adjacent zones match, it is assumed that the zone is empty. | |
8324 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
8325 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
8326 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
8327 | * at arch_max_dma_pfn. | |
8328 | */ | |
9691a071 | 8329 | void __init free_area_init(unsigned long *max_zone_pfn) |
c713216d | 8330 | { |
c13291a5 | 8331 | unsigned long start_pfn, end_pfn; |
51930df5 MR |
8332 | int i, nid, zone; |
8333 | bool descending; | |
a6af2bc3 | 8334 | |
c713216d MG |
8335 | /* Record where the zone boundaries are */ |
8336 | memset(arch_zone_lowest_possible_pfn, 0, | |
8337 | sizeof(arch_zone_lowest_possible_pfn)); | |
8338 | memset(arch_zone_highest_possible_pfn, 0, | |
8339 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe | 8340 | |
fb70c487 | 8341 | start_pfn = PHYS_PFN(memblock_start_of_DRAM()); |
51930df5 | 8342 | descending = arch_has_descending_max_zone_pfns(); |
90cae1fe OH |
8343 | |
8344 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
51930df5 MR |
8345 | if (descending) |
8346 | zone = MAX_NR_ZONES - i - 1; | |
8347 | else | |
8348 | zone = i; | |
8349 | ||
8350 | if (zone == ZONE_MOVABLE) | |
2a1e274a | 8351 | continue; |
90cae1fe | 8352 | |
51930df5 MR |
8353 | end_pfn = max(max_zone_pfn[zone], start_pfn); |
8354 | arch_zone_lowest_possible_pfn[zone] = start_pfn; | |
8355 | arch_zone_highest_possible_pfn[zone] = end_pfn; | |
90cae1fe OH |
8356 | |
8357 | start_pfn = end_pfn; | |
c713216d | 8358 | } |
2a1e274a MG |
8359 | |
8360 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
8361 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 8362 | find_zone_movable_pfns_for_nodes(); |
c713216d | 8363 | |
c713216d | 8364 | /* Print out the zone ranges */ |
f88dfff5 | 8365 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
8366 | for (i = 0; i < MAX_NR_ZONES; i++) { |
8367 | if (i == ZONE_MOVABLE) | |
8368 | continue; | |
f88dfff5 | 8369 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
8370 | if (arch_zone_lowest_possible_pfn[i] == |
8371 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 8372 | pr_cont("empty\n"); |
72f0ba02 | 8373 | else |
8d29e18a JG |
8374 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
8375 | (u64)arch_zone_lowest_possible_pfn[i] | |
8376 | << PAGE_SHIFT, | |
8377 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 8378 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
8379 | } |
8380 | ||
8381 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 8382 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
8383 | for (i = 0; i < MAX_NUMNODES; i++) { |
8384 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
8385 | pr_info(" Node %d: %#018Lx\n", i, |
8386 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 8387 | } |
c713216d | 8388 | |
f46edbd1 DW |
8389 | /* |
8390 | * Print out the early node map, and initialize the | |
8391 | * subsection-map relative to active online memory ranges to | |
8392 | * enable future "sub-section" extensions of the memory map. | |
8393 | */ | |
f88dfff5 | 8394 | pr_info("Early memory node ranges\n"); |
f46edbd1 | 8395 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
8d29e18a JG |
8396 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
8397 | (u64)start_pfn << PAGE_SHIFT, | |
8398 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
f46edbd1 DW |
8399 | subsection_map_init(start_pfn, end_pfn - start_pfn); |
8400 | } | |
c713216d MG |
8401 | |
8402 | /* Initialise every node */ | |
708614e6 | 8403 | mminit_verify_pageflags_layout(); |
8ef82866 | 8404 | setup_nr_node_ids(); |
09f49dca MH |
8405 | for_each_node(nid) { |
8406 | pg_data_t *pgdat; | |
8407 | ||
8408 | if (!node_online(nid)) { | |
8409 | pr_info("Initializing node %d as memoryless\n", nid); | |
8410 | ||
8411 | /* Allocator not initialized yet */ | |
8412 | pgdat = arch_alloc_nodedata(nid); | |
8413 | if (!pgdat) { | |
8414 | pr_err("Cannot allocate %zuB for node %d.\n", | |
8415 | sizeof(*pgdat), nid); | |
8416 | continue; | |
8417 | } | |
8418 | arch_refresh_nodedata(nid, pgdat); | |
8419 | free_area_init_memoryless_node(nid); | |
8420 | ||
8421 | /* | |
8422 | * We do not want to confuse userspace by sysfs | |
8423 | * files/directories for node without any memory | |
8424 | * attached to it, so this node is not marked as | |
8425 | * N_MEMORY and not marked online so that no sysfs | |
8426 | * hierarchy will be created via register_one_node for | |
8427 | * it. The pgdat will get fully initialized by | |
8428 | * hotadd_init_pgdat() when memory is hotplugged into | |
8429 | * this node. | |
8430 | */ | |
8431 | continue; | |
8432 | } | |
8433 | ||
8434 | pgdat = NODE_DATA(nid); | |
854e8848 | 8435 | free_area_init_node(nid); |
37b07e41 LS |
8436 | |
8437 | /* Any memory on that node */ | |
8438 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
8439 | node_set_state(nid, N_MEMORY); |
8440 | check_for_memory(pgdat, nid); | |
c713216d | 8441 | } |
122e093c MR |
8442 | |
8443 | memmap_init(); | |
c713216d | 8444 | } |
2a1e274a | 8445 | |
a5c6d650 DR |
8446 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
8447 | unsigned long *percent) | |
2a1e274a MG |
8448 | { |
8449 | unsigned long long coremem; | |
a5c6d650 DR |
8450 | char *endptr; |
8451 | ||
2a1e274a MG |
8452 | if (!p) |
8453 | return -EINVAL; | |
8454 | ||
a5c6d650 DR |
8455 | /* Value may be a percentage of total memory, otherwise bytes */ |
8456 | coremem = simple_strtoull(p, &endptr, 0); | |
8457 | if (*endptr == '%') { | |
8458 | /* Paranoid check for percent values greater than 100 */ | |
8459 | WARN_ON(coremem > 100); | |
2a1e274a | 8460 | |
a5c6d650 DR |
8461 | *percent = coremem; |
8462 | } else { | |
8463 | coremem = memparse(p, &p); | |
8464 | /* Paranoid check that UL is enough for the coremem value */ | |
8465 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 8466 | |
a5c6d650 DR |
8467 | *core = coremem >> PAGE_SHIFT; |
8468 | *percent = 0UL; | |
8469 | } | |
2a1e274a MG |
8470 | return 0; |
8471 | } | |
ed7ed365 | 8472 | |
7e63efef MG |
8473 | /* |
8474 | * kernelcore=size sets the amount of memory for use for allocations that | |
8475 | * cannot be reclaimed or migrated. | |
8476 | */ | |
8477 | static int __init cmdline_parse_kernelcore(char *p) | |
8478 | { | |
342332e6 TI |
8479 | /* parse kernelcore=mirror */ |
8480 | if (parse_option_str(p, "mirror")) { | |
8481 | mirrored_kernelcore = true; | |
8482 | return 0; | |
8483 | } | |
8484 | ||
a5c6d650 DR |
8485 | return cmdline_parse_core(p, &required_kernelcore, |
8486 | &required_kernelcore_percent); | |
7e63efef MG |
8487 | } |
8488 | ||
8489 | /* | |
8490 | * movablecore=size sets the amount of memory for use for allocations that | |
8491 | * can be reclaimed or migrated. | |
8492 | */ | |
8493 | static int __init cmdline_parse_movablecore(char *p) | |
8494 | { | |
a5c6d650 DR |
8495 | return cmdline_parse_core(p, &required_movablecore, |
8496 | &required_movablecore_percent); | |
7e63efef MG |
8497 | } |
8498 | ||
ed7ed365 | 8499 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 8500 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 8501 | |
c3d5f5f0 JL |
8502 | void adjust_managed_page_count(struct page *page, long count) |
8503 | { | |
9705bea5 | 8504 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 8505 | totalram_pages_add(count); |
3dcc0571 JL |
8506 | #ifdef CONFIG_HIGHMEM |
8507 | if (PageHighMem(page)) | |
ca79b0c2 | 8508 | totalhigh_pages_add(count); |
3dcc0571 | 8509 | #endif |
c3d5f5f0 | 8510 | } |
3dcc0571 | 8511 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 8512 | |
e5cb113f | 8513 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 8514 | { |
11199692 JL |
8515 | void *pos; |
8516 | unsigned long pages = 0; | |
69afade7 | 8517 | |
11199692 JL |
8518 | start = (void *)PAGE_ALIGN((unsigned long)start); |
8519 | end = (void *)((unsigned long)end & PAGE_MASK); | |
8520 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
8521 | struct page *page = virt_to_page(pos); |
8522 | void *direct_map_addr; | |
8523 | ||
8524 | /* | |
8525 | * 'direct_map_addr' might be different from 'pos' | |
8526 | * because some architectures' virt_to_page() | |
8527 | * work with aliases. Getting the direct map | |
8528 | * address ensures that we get a _writeable_ | |
8529 | * alias for the memset(). | |
8530 | */ | |
8531 | direct_map_addr = page_address(page); | |
c746170d VF |
8532 | /* |
8533 | * Perform a kasan-unchecked memset() since this memory | |
8534 | * has not been initialized. | |
8535 | */ | |
8536 | direct_map_addr = kasan_reset_tag(direct_map_addr); | |
dbe67df4 | 8537 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
8538 | memset(direct_map_addr, poison, PAGE_SIZE); |
8539 | ||
8540 | free_reserved_page(page); | |
69afade7 JL |
8541 | } |
8542 | ||
8543 | if (pages && s) | |
ff7ed9e4 | 8544 | pr_info("Freeing %s memory: %ldK\n", s, K(pages)); |
69afade7 JL |
8545 | |
8546 | return pages; | |
8547 | } | |
8548 | ||
1f9d03c5 | 8549 | void __init mem_init_print_info(void) |
7ee3d4e8 JL |
8550 | { |
8551 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
8552 | unsigned long init_code_size, init_data_size; | |
8553 | ||
8554 | physpages = get_num_physpages(); | |
8555 | codesize = _etext - _stext; | |
8556 | datasize = _edata - _sdata; | |
8557 | rosize = __end_rodata - __start_rodata; | |
8558 | bss_size = __bss_stop - __bss_start; | |
8559 | init_data_size = __init_end - __init_begin; | |
8560 | init_code_size = _einittext - _sinittext; | |
8561 | ||
8562 | /* | |
8563 | * Detect special cases and adjust section sizes accordingly: | |
8564 | * 1) .init.* may be embedded into .data sections | |
8565 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
8566 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
8567 | * 3) .rodata.* may be embedded into .text or .data sections. | |
8568 | */ | |
8569 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 | 8570 | do { \ |
ca831f29 | 8571 | if (&start[0] <= &pos[0] && &pos[0] < &end[0] && size > adj) \ |
b8af2941 PK |
8572 | size -= adj; \ |
8573 | } while (0) | |
7ee3d4e8 JL |
8574 | |
8575 | adj_init_size(__init_begin, __init_end, init_data_size, | |
8576 | _sinittext, init_code_size); | |
8577 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
8578 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
8579 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
8580 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
8581 | ||
8582 | #undef adj_init_size | |
8583 | ||
756a025f | 8584 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 8585 | #ifdef CONFIG_HIGHMEM |
756a025f | 8586 | ", %luK highmem" |
7ee3d4e8 | 8587 | #endif |
1f9d03c5 | 8588 | ")\n", |
ff7ed9e4 | 8589 | K(nr_free_pages()), K(physpages), |
c940e020 ML |
8590 | codesize / SZ_1K, datasize / SZ_1K, rosize / SZ_1K, |
8591 | (init_data_size + init_code_size) / SZ_1K, bss_size / SZ_1K, | |
ff7ed9e4 ML |
8592 | K(physpages - totalram_pages() - totalcma_pages), |
8593 | K(totalcma_pages) | |
7ee3d4e8 | 8594 | #ifdef CONFIG_HIGHMEM |
ff7ed9e4 | 8595 | , K(totalhigh_pages()) |
7ee3d4e8 | 8596 | #endif |
1f9d03c5 | 8597 | ); |
7ee3d4e8 JL |
8598 | } |
8599 | ||
0e0b864e | 8600 | /** |
88ca3b94 RD |
8601 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
8602 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 8603 | * |
013110a7 | 8604 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
8605 | * In the DMA zone, a significant percentage may be consumed by kernel image |
8606 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
8607 | * function may optionally be used to account for unfreeable pages in the |
8608 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
8609 | * smaller per-cpu batchsize. | |
0e0b864e MG |
8610 | */ |
8611 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
8612 | { | |
8613 | dma_reserve = new_dma_reserve; | |
8614 | } | |
8615 | ||
005fd4bb | 8616 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 8617 | { |
04f8cfea | 8618 | struct zone *zone; |
1da177e4 | 8619 | |
005fd4bb | 8620 | lru_add_drain_cpu(cpu); |
96f97c43 | 8621 | mlock_drain_remote(cpu); |
005fd4bb | 8622 | drain_pages(cpu); |
9f8f2172 | 8623 | |
005fd4bb SAS |
8624 | /* |
8625 | * Spill the event counters of the dead processor | |
8626 | * into the current processors event counters. | |
8627 | * This artificially elevates the count of the current | |
8628 | * processor. | |
8629 | */ | |
8630 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 8631 | |
005fd4bb SAS |
8632 | /* |
8633 | * Zero the differential counters of the dead processor | |
8634 | * so that the vm statistics are consistent. | |
8635 | * | |
8636 | * This is only okay since the processor is dead and cannot | |
8637 | * race with what we are doing. | |
8638 | */ | |
8639 | cpu_vm_stats_fold(cpu); | |
04f8cfea MG |
8640 | |
8641 | for_each_populated_zone(zone) | |
8642 | zone_pcp_update(zone, 0); | |
8643 | ||
8644 | return 0; | |
8645 | } | |
8646 | ||
8647 | static int page_alloc_cpu_online(unsigned int cpu) | |
8648 | { | |
8649 | struct zone *zone; | |
8650 | ||
8651 | for_each_populated_zone(zone) | |
8652 | zone_pcp_update(zone, 1); | |
005fd4bb | 8653 | return 0; |
1da177e4 | 8654 | } |
1da177e4 | 8655 | |
e03a5125 NP |
8656 | #ifdef CONFIG_NUMA |
8657 | int hashdist = HASHDIST_DEFAULT; | |
8658 | ||
8659 | static int __init set_hashdist(char *str) | |
8660 | { | |
8661 | if (!str) | |
8662 | return 0; | |
8663 | hashdist = simple_strtoul(str, &str, 0); | |
8664 | return 1; | |
8665 | } | |
8666 | __setup("hashdist=", set_hashdist); | |
8667 | #endif | |
8668 | ||
1da177e4 LT |
8669 | void __init page_alloc_init(void) |
8670 | { | |
005fd4bb SAS |
8671 | int ret; |
8672 | ||
e03a5125 NP |
8673 | #ifdef CONFIG_NUMA |
8674 | if (num_node_state(N_MEMORY) == 1) | |
8675 | hashdist = 0; | |
8676 | #endif | |
8677 | ||
04f8cfea MG |
8678 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC, |
8679 | "mm/page_alloc:pcp", | |
8680 | page_alloc_cpu_online, | |
005fd4bb SAS |
8681 | page_alloc_cpu_dead); |
8682 | WARN_ON(ret < 0); | |
1da177e4 LT |
8683 | } |
8684 | ||
cb45b0e9 | 8685 | /* |
34b10060 | 8686 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
8687 | * or min_free_kbytes changes. |
8688 | */ | |
8689 | static void calculate_totalreserve_pages(void) | |
8690 | { | |
8691 | struct pglist_data *pgdat; | |
8692 | unsigned long reserve_pages = 0; | |
2f6726e5 | 8693 | enum zone_type i, j; |
cb45b0e9 HA |
8694 | |
8695 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
8696 | |
8697 | pgdat->totalreserve_pages = 0; | |
8698 | ||
cb45b0e9 HA |
8699 | for (i = 0; i < MAX_NR_ZONES; i++) { |
8700 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 8701 | long max = 0; |
9705bea5 | 8702 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
8703 | |
8704 | /* Find valid and maximum lowmem_reserve in the zone */ | |
8705 | for (j = i; j < MAX_NR_ZONES; j++) { | |
8706 | if (zone->lowmem_reserve[j] > max) | |
8707 | max = zone->lowmem_reserve[j]; | |
8708 | } | |
8709 | ||
41858966 MG |
8710 | /* we treat the high watermark as reserved pages. */ |
8711 | max += high_wmark_pages(zone); | |
cb45b0e9 | 8712 | |
3d6357de AK |
8713 | if (max > managed_pages) |
8714 | max = managed_pages; | |
a8d01437 | 8715 | |
281e3726 | 8716 | pgdat->totalreserve_pages += max; |
a8d01437 | 8717 | |
cb45b0e9 HA |
8718 | reserve_pages += max; |
8719 | } | |
8720 | } | |
8721 | totalreserve_pages = reserve_pages; | |
8722 | } | |
8723 | ||
1da177e4 LT |
8724 | /* |
8725 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 8726 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
8727 | * has a correct pages reserved value, so an adequate number of |
8728 | * pages are left in the zone after a successful __alloc_pages(). | |
8729 | */ | |
8730 | static void setup_per_zone_lowmem_reserve(void) | |
8731 | { | |
8732 | struct pglist_data *pgdat; | |
470c61d7 | 8733 | enum zone_type i, j; |
1da177e4 | 8734 | |
ec936fc5 | 8735 | for_each_online_pgdat(pgdat) { |
470c61d7 LS |
8736 | for (i = 0; i < MAX_NR_ZONES - 1; i++) { |
8737 | struct zone *zone = &pgdat->node_zones[i]; | |
8738 | int ratio = sysctl_lowmem_reserve_ratio[i]; | |
8739 | bool clear = !ratio || !zone_managed_pages(zone); | |
8740 | unsigned long managed_pages = 0; | |
8741 | ||
8742 | for (j = i + 1; j < MAX_NR_ZONES; j++) { | |
f7ec1044 LS |
8743 | struct zone *upper_zone = &pgdat->node_zones[j]; |
8744 | ||
8745 | managed_pages += zone_managed_pages(upper_zone); | |
470c61d7 | 8746 | |
f7ec1044 LS |
8747 | if (clear) |
8748 | zone->lowmem_reserve[j] = 0; | |
8749 | else | |
470c61d7 | 8750 | zone->lowmem_reserve[j] = managed_pages / ratio; |
1da177e4 LT |
8751 | } |
8752 | } | |
8753 | } | |
cb45b0e9 HA |
8754 | |
8755 | /* update totalreserve_pages */ | |
8756 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8757 | } |
8758 | ||
cfd3da1e | 8759 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
8760 | { |
8761 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
8762 | unsigned long lowmem_pages = 0; | |
8763 | struct zone *zone; | |
8764 | unsigned long flags; | |
8765 | ||
8766 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
8767 | for_each_zone(zone) { | |
8768 | if (!is_highmem(zone)) | |
9705bea5 | 8769 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
8770 | } |
8771 | ||
8772 | for_each_zone(zone) { | |
ac924c60 AM |
8773 | u64 tmp; |
8774 | ||
1125b4e3 | 8775 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 8776 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 8777 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
8778 | if (is_highmem(zone)) { |
8779 | /* | |
669ed175 NP |
8780 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
8781 | * need highmem pages, so cap pages_min to a small | |
8782 | * value here. | |
8783 | * | |
41858966 | 8784 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 8785 | * deltas control async page reclaim, and so should |
669ed175 | 8786 | * not be capped for highmem. |
1da177e4 | 8787 | */ |
90ae8d67 | 8788 | unsigned long min_pages; |
1da177e4 | 8789 | |
9705bea5 | 8790 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 8791 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 8792 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 8793 | } else { |
669ed175 NP |
8794 | /* |
8795 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
8796 | * proportionate to the zone's size. |
8797 | */ | |
a9214443 | 8798 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
8799 | } |
8800 | ||
795ae7a0 JW |
8801 | /* |
8802 | * Set the kswapd watermarks distance according to the | |
8803 | * scale factor in proportion to available memory, but | |
8804 | * ensure a minimum size on small systems. | |
8805 | */ | |
8806 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 8807 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
8808 | watermark_scale_factor, 10000)); |
8809 | ||
aa092591 | 8810 | zone->watermark_boost = 0; |
a9214443 | 8811 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
c574bbe9 HY |
8812 | zone->_watermark[WMARK_HIGH] = low_wmark_pages(zone) + tmp; |
8813 | zone->_watermark[WMARK_PROMO] = high_wmark_pages(zone) + tmp; | |
49f223a9 | 8814 | |
1125b4e3 | 8815 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 8816 | } |
cb45b0e9 HA |
8817 | |
8818 | /* update totalreserve_pages */ | |
8819 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8820 | } |
8821 | ||
cfd3da1e MG |
8822 | /** |
8823 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
8824 | * or when memory is hot-{added|removed} | |
8825 | * | |
8826 | * Ensures that the watermark[min,low,high] values for each zone are set | |
8827 | * correctly with respect to min_free_kbytes. | |
8828 | */ | |
8829 | void setup_per_zone_wmarks(void) | |
8830 | { | |
b92ca18e | 8831 | struct zone *zone; |
b93e0f32 MH |
8832 | static DEFINE_SPINLOCK(lock); |
8833 | ||
8834 | spin_lock(&lock); | |
cfd3da1e | 8835 | __setup_per_zone_wmarks(); |
b93e0f32 | 8836 | spin_unlock(&lock); |
b92ca18e MG |
8837 | |
8838 | /* | |
8839 | * The watermark size have changed so update the pcpu batch | |
8840 | * and high limits or the limits may be inappropriate. | |
8841 | */ | |
8842 | for_each_zone(zone) | |
04f8cfea | 8843 | zone_pcp_update(zone, 0); |
cfd3da1e MG |
8844 | } |
8845 | ||
1da177e4 LT |
8846 | /* |
8847 | * Initialise min_free_kbytes. | |
8848 | * | |
8849 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 8850 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
8851 | * bandwidth does not increase linearly with machine size. We use |
8852 | * | |
b8af2941 | 8853 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
8854 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
8855 | * | |
8856 | * which yields | |
8857 | * | |
8858 | * 16MB: 512k | |
8859 | * 32MB: 724k | |
8860 | * 64MB: 1024k | |
8861 | * 128MB: 1448k | |
8862 | * 256MB: 2048k | |
8863 | * 512MB: 2896k | |
8864 | * 1024MB: 4096k | |
8865 | * 2048MB: 5792k | |
8866 | * 4096MB: 8192k | |
8867 | * 8192MB: 11584k | |
8868 | * 16384MB: 16384k | |
8869 | */ | |
bd3400ea | 8870 | void calculate_min_free_kbytes(void) |
1da177e4 LT |
8871 | { |
8872 | unsigned long lowmem_kbytes; | |
5f12733e | 8873 | int new_min_free_kbytes; |
1da177e4 LT |
8874 | |
8875 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
8876 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
8877 | ||
59d336bd WS |
8878 | if (new_min_free_kbytes > user_min_free_kbytes) |
8879 | min_free_kbytes = clamp(new_min_free_kbytes, 128, 262144); | |
8880 | else | |
5f12733e MH |
8881 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", |
8882 | new_min_free_kbytes, user_min_free_kbytes); | |
59d336bd | 8883 | |
bd3400ea LF |
8884 | } |
8885 | ||
8886 | int __meminit init_per_zone_wmark_min(void) | |
8887 | { | |
8888 | calculate_min_free_kbytes(); | |
bc75d33f | 8889 | setup_per_zone_wmarks(); |
a6cccdc3 | 8890 | refresh_zone_stat_thresholds(); |
1da177e4 | 8891 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
8892 | |
8893 | #ifdef CONFIG_NUMA | |
8894 | setup_min_unmapped_ratio(); | |
8895 | setup_min_slab_ratio(); | |
8896 | #endif | |
8897 | ||
4aab2be0 VB |
8898 | khugepaged_min_free_kbytes_update(); |
8899 | ||
1da177e4 LT |
8900 | return 0; |
8901 | } | |
e08d3fdf | 8902 | postcore_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
8903 | |
8904 | /* | |
b8af2941 | 8905 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
8906 | * that we can call two helper functions whenever min_free_kbytes |
8907 | * changes. | |
8908 | */ | |
cccad5b9 | 8909 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8910 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8911 | { |
da8c757b HP |
8912 | int rc; |
8913 | ||
8914 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8915 | if (rc) | |
8916 | return rc; | |
8917 | ||
5f12733e MH |
8918 | if (write) { |
8919 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 8920 | setup_per_zone_wmarks(); |
5f12733e | 8921 | } |
1da177e4 LT |
8922 | return 0; |
8923 | } | |
8924 | ||
795ae7a0 | 8925 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8926 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
8927 | { |
8928 | int rc; | |
8929 | ||
8930 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8931 | if (rc) | |
8932 | return rc; | |
8933 | ||
8934 | if (write) | |
8935 | setup_per_zone_wmarks(); | |
8936 | ||
8937 | return 0; | |
8938 | } | |
8939 | ||
9614634f | 8940 | #ifdef CONFIG_NUMA |
6423aa81 | 8941 | static void setup_min_unmapped_ratio(void) |
9614634f | 8942 | { |
6423aa81 | 8943 | pg_data_t *pgdat; |
9614634f | 8944 | struct zone *zone; |
9614634f | 8945 | |
a5f5f91d | 8946 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 8947 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 8948 | |
9614634f | 8949 | for_each_zone(zone) |
9705bea5 AK |
8950 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
8951 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 8952 | } |
0ff38490 | 8953 | |
6423aa81 JK |
8954 | |
8955 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8956 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 8957 | { |
0ff38490 CL |
8958 | int rc; |
8959 | ||
8d65af78 | 8960 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
8961 | if (rc) |
8962 | return rc; | |
8963 | ||
6423aa81 JK |
8964 | setup_min_unmapped_ratio(); |
8965 | ||
8966 | return 0; | |
8967 | } | |
8968 | ||
8969 | static void setup_min_slab_ratio(void) | |
8970 | { | |
8971 | pg_data_t *pgdat; | |
8972 | struct zone *zone; | |
8973 | ||
a5f5f91d MG |
8974 | for_each_online_pgdat(pgdat) |
8975 | pgdat->min_slab_pages = 0; | |
8976 | ||
0ff38490 | 8977 | for_each_zone(zone) |
9705bea5 AK |
8978 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
8979 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
8980 | } |
8981 | ||
8982 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8983 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
8984 | { |
8985 | int rc; | |
8986 | ||
8987 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8988 | if (rc) | |
8989 | return rc; | |
8990 | ||
8991 | setup_min_slab_ratio(); | |
8992 | ||
0ff38490 CL |
8993 | return 0; |
8994 | } | |
9614634f CL |
8995 | #endif |
8996 | ||
1da177e4 LT |
8997 | /* |
8998 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
8999 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
9000 | * whenever sysctl_lowmem_reserve_ratio changes. | |
9001 | * | |
9002 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 9003 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
9004 | * if in function of the boot time zone sizes. |
9005 | */ | |
cccad5b9 | 9006 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 9007 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 9008 | { |
86aaf255 BH |
9009 | int i; |
9010 | ||
8d65af78 | 9011 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
9012 | |
9013 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
9014 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
9015 | sysctl_lowmem_reserve_ratio[i] = 0; | |
9016 | } | |
9017 | ||
1da177e4 LT |
9018 | setup_per_zone_lowmem_reserve(); |
9019 | return 0; | |
9020 | } | |
9021 | ||
8ad4b1fb | 9022 | /* |
74f44822 MG |
9023 | * percpu_pagelist_high_fraction - changes the pcp->high for each zone on each |
9024 | * cpu. It is the fraction of total pages in each zone that a hot per cpu | |
b8af2941 | 9025 | * pagelist can have before it gets flushed back to buddy allocator. |
8ad4b1fb | 9026 | */ |
74f44822 MG |
9027 | int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *table, |
9028 | int write, void *buffer, size_t *length, loff_t *ppos) | |
8ad4b1fb RS |
9029 | { |
9030 | struct zone *zone; | |
74f44822 | 9031 | int old_percpu_pagelist_high_fraction; |
8ad4b1fb RS |
9032 | int ret; |
9033 | ||
7cd2b0a3 | 9034 | mutex_lock(&pcp_batch_high_lock); |
74f44822 | 9035 | old_percpu_pagelist_high_fraction = percpu_pagelist_high_fraction; |
7cd2b0a3 | 9036 | |
8d65af78 | 9037 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
9038 | if (!write || ret < 0) |
9039 | goto out; | |
9040 | ||
9041 | /* Sanity checking to avoid pcp imbalance */ | |
74f44822 MG |
9042 | if (percpu_pagelist_high_fraction && |
9043 | percpu_pagelist_high_fraction < MIN_PERCPU_PAGELIST_HIGH_FRACTION) { | |
9044 | percpu_pagelist_high_fraction = old_percpu_pagelist_high_fraction; | |
7cd2b0a3 DR |
9045 | ret = -EINVAL; |
9046 | goto out; | |
9047 | } | |
9048 | ||
9049 | /* No change? */ | |
74f44822 | 9050 | if (percpu_pagelist_high_fraction == old_percpu_pagelist_high_fraction) |
7cd2b0a3 | 9051 | goto out; |
c8e251fa | 9052 | |
cb1ef534 | 9053 | for_each_populated_zone(zone) |
74f44822 | 9054 | zone_set_pageset_high_and_batch(zone, 0); |
7cd2b0a3 | 9055 | out: |
c8e251fa | 9056 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 9057 | return ret; |
8ad4b1fb RS |
9058 | } |
9059 | ||
f6f34b43 SD |
9060 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
9061 | /* | |
9062 | * Returns the number of pages that arch has reserved but | |
9063 | * is not known to alloc_large_system_hash(). | |
9064 | */ | |
9065 | static unsigned long __init arch_reserved_kernel_pages(void) | |
9066 | { | |
9067 | return 0; | |
9068 | } | |
9069 | #endif | |
9070 | ||
9017217b PT |
9071 | /* |
9072 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
9073 | * machines. As memory size is increased the scale is also increased but at | |
9074 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
9075 | * quadruples the scale is increased by one, which means the size of hash table | |
9076 | * only doubles, instead of quadrupling as well. | |
9077 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
9078 | * makes sense, it is disabled on such platforms. | |
9079 | */ | |
9080 | #if __BITS_PER_LONG > 32 | |
9081 | #define ADAPT_SCALE_BASE (64ul << 30) | |
9082 | #define ADAPT_SCALE_SHIFT 2 | |
9083 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
9084 | #endif | |
9085 | ||
1da177e4 LT |
9086 | /* |
9087 | * allocate a large system hash table from bootmem | |
9088 | * - it is assumed that the hash table must contain an exact power-of-2 | |
9089 | * quantity of entries | |
9090 | * - limit is the number of hash buckets, not the total allocation size | |
9091 | */ | |
9092 | void *__init alloc_large_system_hash(const char *tablename, | |
9093 | unsigned long bucketsize, | |
9094 | unsigned long numentries, | |
9095 | int scale, | |
9096 | int flags, | |
9097 | unsigned int *_hash_shift, | |
9098 | unsigned int *_hash_mask, | |
31fe62b9 TB |
9099 | unsigned long low_limit, |
9100 | unsigned long high_limit) | |
1da177e4 | 9101 | { |
31fe62b9 | 9102 | unsigned long long max = high_limit; |
1da177e4 | 9103 | unsigned long log2qty, size; |
97bab178 | 9104 | void *table; |
3749a8f0 | 9105 | gfp_t gfp_flags; |
ec11408a | 9106 | bool virt; |
121e6f32 | 9107 | bool huge; |
1da177e4 LT |
9108 | |
9109 | /* allow the kernel cmdline to have a say */ | |
9110 | if (!numentries) { | |
9111 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 9112 | numentries = nr_kernel_pages; |
f6f34b43 | 9113 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
9114 | |
9115 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
c940e020 ML |
9116 | if (PAGE_SIZE < SZ_1M) |
9117 | numentries = round_up(numentries, SZ_1M / PAGE_SIZE); | |
1da177e4 | 9118 | |
9017217b PT |
9119 | #if __BITS_PER_LONG > 32 |
9120 | if (!high_limit) { | |
9121 | unsigned long adapt; | |
9122 | ||
9123 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
9124 | adapt <<= ADAPT_SCALE_SHIFT) | |
9125 | scale++; | |
9126 | } | |
9127 | #endif | |
9128 | ||
1da177e4 LT |
9129 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
9130 | if (scale > PAGE_SHIFT) | |
9131 | numentries >>= (scale - PAGE_SHIFT); | |
9132 | else | |
9133 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
9134 | |
9135 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
9136 | if (unlikely(flags & HASH_SMALL)) { |
9137 | /* Makes no sense without HASH_EARLY */ | |
9138 | WARN_ON(!(flags & HASH_EARLY)); | |
9139 | if (!(numentries >> *_hash_shift)) { | |
9140 | numentries = 1UL << *_hash_shift; | |
9141 | BUG_ON(!numentries); | |
9142 | } | |
9143 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 9144 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 9145 | } |
6e692ed3 | 9146 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
9147 | |
9148 | /* limit allocation size to 1/16 total memory by default */ | |
9149 | if (max == 0) { | |
9150 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
9151 | do_div(max, bucketsize); | |
9152 | } | |
074b8517 | 9153 | max = min(max, 0x80000000ULL); |
1da177e4 | 9154 | |
31fe62b9 TB |
9155 | if (numentries < low_limit) |
9156 | numentries = low_limit; | |
1da177e4 LT |
9157 | if (numentries > max) |
9158 | numentries = max; | |
9159 | ||
f0d1b0b3 | 9160 | log2qty = ilog2(numentries); |
1da177e4 | 9161 | |
3749a8f0 | 9162 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 9163 | do { |
ec11408a | 9164 | virt = false; |
1da177e4 | 9165 | size = bucketsize << log2qty; |
ea1f5f37 PT |
9166 | if (flags & HASH_EARLY) { |
9167 | if (flags & HASH_ZERO) | |
26fb3dae | 9168 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 9169 | else |
7e1c4e27 MR |
9170 | table = memblock_alloc_raw(size, |
9171 | SMP_CACHE_BYTES); | |
ec11408a | 9172 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
f2edd118 | 9173 | table = vmalloc_huge(size, gfp_flags); |
ec11408a | 9174 | virt = true; |
084f7e23 ED |
9175 | if (table) |
9176 | huge = is_vm_area_hugepages(table); | |
ea1f5f37 | 9177 | } else { |
1037b83b ED |
9178 | /* |
9179 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
9180 | * some pages at the end of hash table which |
9181 | * alloc_pages_exact() automatically does | |
1037b83b | 9182 | */ |
ec11408a NP |
9183 | table = alloc_pages_exact(size, gfp_flags); |
9184 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
9185 | } |
9186 | } while (!table && size > PAGE_SIZE && --log2qty); | |
9187 | ||
9188 | if (!table) | |
9189 | panic("Failed to allocate %s hash table\n", tablename); | |
9190 | ||
ec11408a NP |
9191 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
9192 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
121e6f32 | 9193 | virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear"); |
1da177e4 LT |
9194 | |
9195 | if (_hash_shift) | |
9196 | *_hash_shift = log2qty; | |
9197 | if (_hash_mask) | |
9198 | *_hash_mask = (1 << log2qty) - 1; | |
9199 | ||
9200 | return table; | |
9201 | } | |
a117e66e | 9202 | |
8df995f6 | 9203 | #ifdef CONFIG_CONTIG_ALLOC |
a1394bdd MK |
9204 | #if defined(CONFIG_DYNAMIC_DEBUG) || \ |
9205 | (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) | |
9206 | /* Usage: See admin-guide/dynamic-debug-howto.rst */ | |
9207 | static void alloc_contig_dump_pages(struct list_head *page_list) | |
9208 | { | |
9209 | DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure"); | |
9210 | ||
9211 | if (DYNAMIC_DEBUG_BRANCH(descriptor)) { | |
9212 | struct page *page; | |
9213 | ||
9214 | dump_stack(); | |
9215 | list_for_each_entry(page, page_list, lru) | |
9216 | dump_page(page, "migration failure"); | |
9217 | } | |
9218 | } | |
9219 | #else | |
9220 | static inline void alloc_contig_dump_pages(struct list_head *page_list) | |
9221 | { | |
9222 | } | |
9223 | #endif | |
9224 | ||
041d3a8c | 9225 | /* [start, end) must belong to a single zone. */ |
b2c9e2fb | 9226 | int __alloc_contig_migrate_range(struct compact_control *cc, |
bb13ffeb | 9227 | unsigned long start, unsigned long end) |
041d3a8c MN |
9228 | { |
9229 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 9230 | unsigned int nr_reclaimed; |
041d3a8c MN |
9231 | unsigned long pfn = start; |
9232 | unsigned int tries = 0; | |
9233 | int ret = 0; | |
8b94e0b8 JK |
9234 | struct migration_target_control mtc = { |
9235 | .nid = zone_to_nid(cc->zone), | |
9236 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
9237 | }; | |
041d3a8c | 9238 | |
361a2a22 | 9239 | lru_cache_disable(); |
041d3a8c | 9240 | |
bb13ffeb | 9241 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
9242 | if (fatal_signal_pending(current)) { |
9243 | ret = -EINTR; | |
9244 | break; | |
9245 | } | |
9246 | ||
bb13ffeb MG |
9247 | if (list_empty(&cc->migratepages)) { |
9248 | cc->nr_migratepages = 0; | |
c2ad7a1f OS |
9249 | ret = isolate_migratepages_range(cc, pfn, end); |
9250 | if (ret && ret != -EAGAIN) | |
041d3a8c | 9251 | break; |
c2ad7a1f | 9252 | pfn = cc->migrate_pfn; |
041d3a8c MN |
9253 | tries = 0; |
9254 | } else if (++tries == 5) { | |
c8e28b47 | 9255 | ret = -EBUSY; |
041d3a8c MN |
9256 | break; |
9257 | } | |
9258 | ||
beb51eaa MK |
9259 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
9260 | &cc->migratepages); | |
9261 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 9262 | |
8b94e0b8 | 9263 | ret = migrate_pages(&cc->migratepages, alloc_migration_target, |
5ac95884 | 9264 | NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE, NULL); |
c8e28b47 OS |
9265 | |
9266 | /* | |
9267 | * On -ENOMEM, migrate_pages() bails out right away. It is pointless | |
9268 | * to retry again over this error, so do the same here. | |
9269 | */ | |
9270 | if (ret == -ENOMEM) | |
9271 | break; | |
041d3a8c | 9272 | } |
d479960e | 9273 | |
361a2a22 | 9274 | lru_cache_enable(); |
2a6f5124 | 9275 | if (ret < 0) { |
3f913fc5 | 9276 | if (!(cc->gfp_mask & __GFP_NOWARN) && ret == -EBUSY) |
151e084a | 9277 | alloc_contig_dump_pages(&cc->migratepages); |
2a6f5124 SP |
9278 | putback_movable_pages(&cc->migratepages); |
9279 | return ret; | |
9280 | } | |
9281 | return 0; | |
041d3a8c MN |
9282 | } |
9283 | ||
9284 | /** | |
9285 | * alloc_contig_range() -- tries to allocate given range of pages | |
9286 | * @start: start PFN to allocate | |
9287 | * @end: one-past-the-last PFN to allocate | |
f0953a1b | 9288 | * @migratetype: migratetype of the underlying pageblocks (either |
0815f3d8 MN |
9289 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
9290 | * in range must have the same migratetype and it must | |
9291 | * be either of the two. | |
ca96b625 | 9292 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c | 9293 | * |
11ac3e87 ZY |
9294 | * The PFN range does not have to be pageblock aligned. The PFN range must |
9295 | * belong to a single zone. | |
041d3a8c | 9296 | * |
2c7452a0 MK |
9297 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
9298 | * pageblocks in the range. Once isolated, the pageblocks should not | |
9299 | * be modified by others. | |
041d3a8c | 9300 | * |
a862f68a | 9301 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
9302 | * pages which PFN is in [start, end) are allocated for the caller and |
9303 | * need to be freed with free_contig_range(). | |
9304 | */ | |
0815f3d8 | 9305 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 9306 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 9307 | { |
041d3a8c | 9308 | unsigned long outer_start, outer_end; |
b2c9e2fb | 9309 | int order; |
d00181b9 | 9310 | int ret = 0; |
041d3a8c | 9311 | |
bb13ffeb MG |
9312 | struct compact_control cc = { |
9313 | .nr_migratepages = 0, | |
9314 | .order = -1, | |
9315 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 9316 | .mode = MIGRATE_SYNC, |
bb13ffeb | 9317 | .ignore_skip_hint = true, |
2583d671 | 9318 | .no_set_skip_hint = true, |
7dea19f9 | 9319 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 9320 | .alloc_contig = true, |
bb13ffeb MG |
9321 | }; |
9322 | INIT_LIST_HEAD(&cc.migratepages); | |
9323 | ||
041d3a8c MN |
9324 | /* |
9325 | * What we do here is we mark all pageblocks in range as | |
9326 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
9327 | * have different sizes, and due to the way page allocator | |
b2c9e2fb | 9328 | * work, start_isolate_page_range() has special handlings for this. |
041d3a8c MN |
9329 | * |
9330 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
9331 | * migrate the pages from an unaligned range (ie. pages that | |
b2c9e2fb | 9332 | * we are interested in). This will put all the pages in |
041d3a8c MN |
9333 | * range back to page allocator as MIGRATE_ISOLATE. |
9334 | * | |
9335 | * When this is done, we take the pages in range from page | |
9336 | * allocator removing them from the buddy system. This way | |
9337 | * page allocator will never consider using them. | |
9338 | * | |
9339 | * This lets us mark the pageblocks back as | |
9340 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
9341 | * aligned range but not in the unaligned, original range are | |
9342 | * put back to page allocator so that buddy can use them. | |
9343 | */ | |
9344 | ||
6e263fff | 9345 | ret = start_isolate_page_range(start, end, migratetype, 0, gfp_mask); |
3fa0c7c7 | 9346 | if (ret) |
b2c9e2fb | 9347 | goto done; |
041d3a8c | 9348 | |
7612921f VB |
9349 | drain_all_pages(cc.zone); |
9350 | ||
8ef5849f JK |
9351 | /* |
9352 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
9353 | * So, just fall through. test_pages_isolated() has a tracepoint |
9354 | * which will report the busy page. | |
9355 | * | |
9356 | * It is possible that busy pages could become available before | |
9357 | * the call to test_pages_isolated, and the range will actually be | |
9358 | * allocated. So, if we fall through be sure to clear ret so that | |
9359 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 9360 | */ |
bb13ffeb | 9361 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 9362 | if (ret && ret != -EBUSY) |
041d3a8c | 9363 | goto done; |
68d68ff6 | 9364 | ret = 0; |
041d3a8c MN |
9365 | |
9366 | /* | |
b2c9e2fb | 9367 | * Pages from [start, end) are within a pageblock_nr_pages |
041d3a8c MN |
9368 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's |
9369 | * more, all pages in [start, end) are free in page allocator. | |
9370 | * What we are going to do is to allocate all pages from | |
9371 | * [start, end) (that is remove them from page allocator). | |
9372 | * | |
9373 | * The only problem is that pages at the beginning and at the | |
9374 | * end of interesting range may be not aligned with pages that | |
9375 | * page allocator holds, ie. they can be part of higher order | |
9376 | * pages. Because of this, we reserve the bigger range and | |
9377 | * once this is done free the pages we are not interested in. | |
9378 | * | |
9379 | * We don't have to hold zone->lock here because the pages are | |
9380 | * isolated thus they won't get removed from buddy. | |
9381 | */ | |
9382 | ||
041d3a8c MN |
9383 | order = 0; |
9384 | outer_start = start; | |
9385 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
9386 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
9387 | outer_start = start; |
9388 | break; | |
041d3a8c MN |
9389 | } |
9390 | outer_start &= ~0UL << order; | |
9391 | } | |
9392 | ||
8ef5849f | 9393 | if (outer_start != start) { |
ab130f91 | 9394 | order = buddy_order(pfn_to_page(outer_start)); |
8ef5849f JK |
9395 | |
9396 | /* | |
9397 | * outer_start page could be small order buddy page and | |
9398 | * it doesn't include start page. Adjust outer_start | |
9399 | * in this case to report failed page properly | |
9400 | * on tracepoint in test_pages_isolated() | |
9401 | */ | |
9402 | if (outer_start + (1UL << order) <= start) | |
9403 | outer_start = start; | |
9404 | } | |
9405 | ||
041d3a8c | 9406 | /* Make sure the range is really isolated. */ |
756d25be | 9407 | if (test_pages_isolated(outer_start, end, 0)) { |
041d3a8c MN |
9408 | ret = -EBUSY; |
9409 | goto done; | |
9410 | } | |
9411 | ||
49f223a9 | 9412 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 9413 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
9414 | if (!outer_end) { |
9415 | ret = -EBUSY; | |
9416 | goto done; | |
9417 | } | |
9418 | ||
9419 | /* Free head and tail (if any) */ | |
9420 | if (start != outer_start) | |
9421 | free_contig_range(outer_start, start - outer_start); | |
9422 | if (end != outer_end) | |
9423 | free_contig_range(end, outer_end - end); | |
9424 | ||
9425 | done: | |
6e263fff | 9426 | undo_isolate_page_range(start, end, migratetype); |
041d3a8c MN |
9427 | return ret; |
9428 | } | |
255f5985 | 9429 | EXPORT_SYMBOL(alloc_contig_range); |
5e27a2df AK |
9430 | |
9431 | static int __alloc_contig_pages(unsigned long start_pfn, | |
9432 | unsigned long nr_pages, gfp_t gfp_mask) | |
9433 | { | |
9434 | unsigned long end_pfn = start_pfn + nr_pages; | |
9435 | ||
9436 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
9437 | gfp_mask); | |
9438 | } | |
9439 | ||
9440 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
9441 | unsigned long nr_pages) | |
9442 | { | |
9443 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
9444 | struct page *page; | |
9445 | ||
9446 | for (i = start_pfn; i < end_pfn; i++) { | |
9447 | page = pfn_to_online_page(i); | |
9448 | if (!page) | |
9449 | return false; | |
9450 | ||
9451 | if (page_zone(page) != z) | |
9452 | return false; | |
9453 | ||
9454 | if (PageReserved(page)) | |
9455 | return false; | |
5e27a2df AK |
9456 | } |
9457 | return true; | |
9458 | } | |
9459 | ||
9460 | static bool zone_spans_last_pfn(const struct zone *zone, | |
9461 | unsigned long start_pfn, unsigned long nr_pages) | |
9462 | { | |
9463 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
9464 | ||
9465 | return zone_spans_pfn(zone, last_pfn); | |
9466 | } | |
9467 | ||
9468 | /** | |
9469 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
9470 | * @nr_pages: Number of contiguous pages to allocate | |
9471 | * @gfp_mask: GFP mask to limit search and used during compaction | |
9472 | * @nid: Target node | |
9473 | * @nodemask: Mask for other possible nodes | |
9474 | * | |
9475 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
9476 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
9477 | * tried for allocation with alloc_contig_range(). This routine is intended | |
9478 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
9479 | * | |
9480 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
eaab8e75 AK |
9481 | * power of two, then allocated range is also guaranteed to be aligned to same |
9482 | * nr_pages (e.g. 1GB request would be aligned to 1GB). | |
5e27a2df AK |
9483 | * |
9484 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
9485 | * __free_page() on each allocated page. | |
9486 | * | |
9487 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
9488 | */ | |
9489 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
9490 | int nid, nodemask_t *nodemask) | |
9491 | { | |
9492 | unsigned long ret, pfn, flags; | |
9493 | struct zonelist *zonelist; | |
9494 | struct zone *zone; | |
9495 | struct zoneref *z; | |
9496 | ||
9497 | zonelist = node_zonelist(nid, gfp_mask); | |
9498 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
9499 | gfp_zone(gfp_mask), nodemask) { | |
9500 | spin_lock_irqsave(&zone->lock, flags); | |
9501 | ||
9502 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
9503 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
9504 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
9505 | /* | |
9506 | * We release the zone lock here because | |
9507 | * alloc_contig_range() will also lock the zone | |
9508 | * at some point. If there's an allocation | |
9509 | * spinning on this lock, it may win the race | |
9510 | * and cause alloc_contig_range() to fail... | |
9511 | */ | |
9512 | spin_unlock_irqrestore(&zone->lock, flags); | |
9513 | ret = __alloc_contig_pages(pfn, nr_pages, | |
9514 | gfp_mask); | |
9515 | if (!ret) | |
9516 | return pfn_to_page(pfn); | |
9517 | spin_lock_irqsave(&zone->lock, flags); | |
9518 | } | |
9519 | pfn += nr_pages; | |
9520 | } | |
9521 | spin_unlock_irqrestore(&zone->lock, flags); | |
9522 | } | |
9523 | return NULL; | |
9524 | } | |
4eb0716e | 9525 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 9526 | |
78fa5150 | 9527 | void free_contig_range(unsigned long pfn, unsigned long nr_pages) |
041d3a8c | 9528 | { |
78fa5150 | 9529 | unsigned long count = 0; |
bcc2b02f MS |
9530 | |
9531 | for (; nr_pages--; pfn++) { | |
9532 | struct page *page = pfn_to_page(pfn); | |
9533 | ||
9534 | count += page_count(page) != 1; | |
9535 | __free_page(page); | |
9536 | } | |
78fa5150 | 9537 | WARN(count != 0, "%lu pages are still in use!\n", count); |
041d3a8c | 9538 | } |
255f5985 | 9539 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 9540 | |
ec6e8c7e VB |
9541 | /* |
9542 | * Effectively disable pcplists for the zone by setting the high limit to 0 | |
9543 | * and draining all cpus. A concurrent page freeing on another CPU that's about | |
9544 | * to put the page on pcplist will either finish before the drain and the page | |
9545 | * will be drained, or observe the new high limit and skip the pcplist. | |
9546 | * | |
9547 | * Must be paired with a call to zone_pcp_enable(). | |
9548 | */ | |
9549 | void zone_pcp_disable(struct zone *zone) | |
9550 | { | |
9551 | mutex_lock(&pcp_batch_high_lock); | |
9552 | __zone_set_pageset_high_and_batch(zone, 0, 1); | |
9553 | __drain_all_pages(zone, true); | |
9554 | } | |
9555 | ||
9556 | void zone_pcp_enable(struct zone *zone) | |
9557 | { | |
9558 | __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch); | |
9559 | mutex_unlock(&pcp_batch_high_lock); | |
9560 | } | |
9561 | ||
340175b7 JL |
9562 | void zone_pcp_reset(struct zone *zone) |
9563 | { | |
5a883813 | 9564 | int cpu; |
28f836b6 | 9565 | struct per_cpu_zonestat *pzstats; |
340175b7 | 9566 | |
28f836b6 | 9567 | if (zone->per_cpu_pageset != &boot_pageset) { |
5a883813 | 9568 | for_each_online_cpu(cpu) { |
28f836b6 MG |
9569 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); |
9570 | drain_zonestat(zone, pzstats); | |
5a883813 | 9571 | } |
28f836b6 | 9572 | free_percpu(zone->per_cpu_pageset); |
28f836b6 | 9573 | zone->per_cpu_pageset = &boot_pageset; |
022e7fa0 ML |
9574 | if (zone->per_cpu_zonestats != &boot_zonestats) { |
9575 | free_percpu(zone->per_cpu_zonestats); | |
9576 | zone->per_cpu_zonestats = &boot_zonestats; | |
9577 | } | |
340175b7 | 9578 | } |
340175b7 JL |
9579 | } |
9580 | ||
6dcd73d7 | 9581 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 9582 | /* |
257bea71 DH |
9583 | * All pages in the range must be in a single zone, must not contain holes, |
9584 | * must span full sections, and must be isolated before calling this function. | |
0c0e6195 | 9585 | */ |
257bea71 | 9586 | void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 9587 | { |
257bea71 | 9588 | unsigned long pfn = start_pfn; |
0c0e6195 KH |
9589 | struct page *page; |
9590 | struct zone *zone; | |
0ee5f4f3 | 9591 | unsigned int order; |
0c0e6195 | 9592 | unsigned long flags; |
5557c766 | 9593 | |
2d070eab | 9594 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
9595 | zone = page_zone(pfn_to_page(pfn)); |
9596 | spin_lock_irqsave(&zone->lock, flags); | |
0c0e6195 | 9597 | while (pfn < end_pfn) { |
0c0e6195 | 9598 | page = pfn_to_page(pfn); |
b023f468 WC |
9599 | /* |
9600 | * The HWPoisoned page may be not in buddy system, and | |
9601 | * page_count() is not 0. | |
9602 | */ | |
9603 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
9604 | pfn++; | |
b023f468 WC |
9605 | continue; |
9606 | } | |
aa218795 DH |
9607 | /* |
9608 | * At this point all remaining PageOffline() pages have a | |
9609 | * reference count of 0 and can simply be skipped. | |
9610 | */ | |
9611 | if (PageOffline(page)) { | |
9612 | BUG_ON(page_count(page)); | |
9613 | BUG_ON(PageBuddy(page)); | |
9614 | pfn++; | |
aa218795 DH |
9615 | continue; |
9616 | } | |
b023f468 | 9617 | |
0c0e6195 KH |
9618 | BUG_ON(page_count(page)); |
9619 | BUG_ON(!PageBuddy(page)); | |
ab130f91 | 9620 | order = buddy_order(page); |
6ab01363 | 9621 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
9622 | pfn += (1 << order); |
9623 | } | |
9624 | spin_unlock_irqrestore(&zone->lock, flags); | |
9625 | } | |
9626 | #endif | |
8d22ba1b | 9627 | |
8446b59b ED |
9628 | /* |
9629 | * This function returns a stable result only if called under zone lock. | |
9630 | */ | |
8d22ba1b WF |
9631 | bool is_free_buddy_page(struct page *page) |
9632 | { | |
8d22ba1b | 9633 | unsigned long pfn = page_to_pfn(page); |
7aeb09f9 | 9634 | unsigned int order; |
8d22ba1b | 9635 | |
8d22ba1b WF |
9636 | for (order = 0; order < MAX_ORDER; order++) { |
9637 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
9638 | ||
8446b59b ED |
9639 | if (PageBuddy(page_head) && |
9640 | buddy_order_unsafe(page_head) >= order) | |
8d22ba1b WF |
9641 | break; |
9642 | } | |
8d22ba1b WF |
9643 | |
9644 | return order < MAX_ORDER; | |
9645 | } | |
a581865e | 9646 | EXPORT_SYMBOL(is_free_buddy_page); |
d4ae9916 NH |
9647 | |
9648 | #ifdef CONFIG_MEMORY_FAILURE | |
9649 | /* | |
06be6ff3 OS |
9650 | * Break down a higher-order page in sub-pages, and keep our target out of |
9651 | * buddy allocator. | |
d4ae9916 | 9652 | */ |
06be6ff3 OS |
9653 | static void break_down_buddy_pages(struct zone *zone, struct page *page, |
9654 | struct page *target, int low, int high, | |
9655 | int migratetype) | |
9656 | { | |
9657 | unsigned long size = 1 << high; | |
9658 | struct page *current_buddy, *next_page; | |
9659 | ||
9660 | while (high > low) { | |
9661 | high--; | |
9662 | size >>= 1; | |
9663 | ||
9664 | if (target >= &page[size]) { | |
9665 | next_page = page + size; | |
9666 | current_buddy = page; | |
9667 | } else { | |
9668 | next_page = page; | |
9669 | current_buddy = page + size; | |
9670 | } | |
9671 | ||
9672 | if (set_page_guard(zone, current_buddy, high, migratetype)) | |
9673 | continue; | |
9674 | ||
9675 | if (current_buddy != target) { | |
9676 | add_to_free_list(current_buddy, zone, high, migratetype); | |
ab130f91 | 9677 | set_buddy_order(current_buddy, high); |
06be6ff3 OS |
9678 | page = next_page; |
9679 | } | |
9680 | } | |
9681 | } | |
9682 | ||
9683 | /* | |
9684 | * Take a page that will be marked as poisoned off the buddy allocator. | |
9685 | */ | |
9686 | bool take_page_off_buddy(struct page *page) | |
d4ae9916 NH |
9687 | { |
9688 | struct zone *zone = page_zone(page); | |
9689 | unsigned long pfn = page_to_pfn(page); | |
9690 | unsigned long flags; | |
9691 | unsigned int order; | |
06be6ff3 | 9692 | bool ret = false; |
d4ae9916 NH |
9693 | |
9694 | spin_lock_irqsave(&zone->lock, flags); | |
9695 | for (order = 0; order < MAX_ORDER; order++) { | |
9696 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
ab130f91 | 9697 | int page_order = buddy_order(page_head); |
d4ae9916 | 9698 | |
ab130f91 | 9699 | if (PageBuddy(page_head) && page_order >= order) { |
06be6ff3 OS |
9700 | unsigned long pfn_head = page_to_pfn(page_head); |
9701 | int migratetype = get_pfnblock_migratetype(page_head, | |
9702 | pfn_head); | |
9703 | ||
ab130f91 | 9704 | del_page_from_free_list(page_head, zone, page_order); |
06be6ff3 | 9705 | break_down_buddy_pages(zone, page_head, page, 0, |
ab130f91 | 9706 | page_order, migratetype); |
bf181c58 | 9707 | SetPageHWPoisonTakenOff(page); |
bac9c6fa DH |
9708 | if (!is_migrate_isolate(migratetype)) |
9709 | __mod_zone_freepage_state(zone, -1, migratetype); | |
06be6ff3 | 9710 | ret = true; |
d4ae9916 NH |
9711 | break; |
9712 | } | |
06be6ff3 OS |
9713 | if (page_count(page_head) > 0) |
9714 | break; | |
d4ae9916 NH |
9715 | } |
9716 | spin_unlock_irqrestore(&zone->lock, flags); | |
06be6ff3 | 9717 | return ret; |
d4ae9916 | 9718 | } |
bf181c58 NH |
9719 | |
9720 | /* | |
9721 | * Cancel takeoff done by take_page_off_buddy(). | |
9722 | */ | |
9723 | bool put_page_back_buddy(struct page *page) | |
9724 | { | |
9725 | struct zone *zone = page_zone(page); | |
9726 | unsigned long pfn = page_to_pfn(page); | |
9727 | unsigned long flags; | |
9728 | int migratetype = get_pfnblock_migratetype(page, pfn); | |
9729 | bool ret = false; | |
9730 | ||
9731 | spin_lock_irqsave(&zone->lock, flags); | |
9732 | if (put_page_testzero(page)) { | |
9733 | ClearPageHWPoisonTakenOff(page); | |
9734 | __free_one_page(page, pfn, zone, 0, migratetype, FPI_NONE); | |
9735 | if (TestClearPageHWPoison(page)) { | |
bf181c58 NH |
9736 | ret = true; |
9737 | } | |
9738 | } | |
9739 | spin_unlock_irqrestore(&zone->lock, flags); | |
9740 | ||
9741 | return ret; | |
9742 | } | |
d4ae9916 | 9743 | #endif |
62b31070 BH |
9744 | |
9745 | #ifdef CONFIG_ZONE_DMA | |
9746 | bool has_managed_dma(void) | |
9747 | { | |
9748 | struct pglist_data *pgdat; | |
9749 | ||
9750 | for_each_online_pgdat(pgdat) { | |
9751 | struct zone *zone = &pgdat->node_zones[ZONE_DMA]; | |
9752 | ||
9753 | if (managed_zone(zone)) | |
9754 | return true; | |
9755 | } | |
9756 | return false; | |
9757 | } | |
9758 | #endif /* CONFIG_ZONE_DMA */ |