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 | ||
173 | #define pcpu_spin_lock_irqsave(type, member, ptr, flags) \ | |
174 | ({ \ | |
175 | type *_ret; \ | |
176 | pcpu_task_pin(); \ | |
177 | _ret = this_cpu_ptr(ptr); \ | |
178 | spin_lock_irqsave(&_ret->member, flags); \ | |
179 | _ret; \ | |
180 | }) | |
181 | ||
182 | #define pcpu_spin_trylock_irqsave(type, member, ptr, flags) \ | |
183 | ({ \ | |
184 | type *_ret; \ | |
185 | pcpu_task_pin(); \ | |
186 | _ret = this_cpu_ptr(ptr); \ | |
187 | if (!spin_trylock_irqsave(&_ret->member, flags)) { \ | |
188 | pcpu_task_unpin(); \ | |
189 | _ret = NULL; \ | |
190 | } \ | |
191 | _ret; \ | |
192 | }) | |
193 | ||
194 | #define pcpu_spin_unlock(member, ptr) \ | |
195 | ({ \ | |
196 | spin_unlock(&ptr->member); \ | |
197 | pcpu_task_unpin(); \ | |
198 | }) | |
199 | ||
200 | #define pcpu_spin_unlock_irqrestore(member, ptr, flags) \ | |
201 | ({ \ | |
202 | spin_unlock_irqrestore(&ptr->member, flags); \ | |
203 | pcpu_task_unpin(); \ | |
204 | }) | |
205 | ||
206 | /* struct per_cpu_pages specific helpers. */ | |
207 | #define pcp_spin_lock(ptr) \ | |
208 | pcpu_spin_lock(struct per_cpu_pages, lock, ptr) | |
209 | ||
210 | #define pcp_spin_lock_irqsave(ptr, flags) \ | |
211 | pcpu_spin_lock_irqsave(struct per_cpu_pages, lock, ptr, flags) | |
212 | ||
213 | #define pcp_spin_trylock_irqsave(ptr, flags) \ | |
214 | pcpu_spin_trylock_irqsave(struct per_cpu_pages, lock, ptr, flags) | |
215 | ||
216 | #define pcp_spin_unlock(ptr) \ | |
217 | pcpu_spin_unlock(lock, ptr) | |
218 | ||
219 | #define pcp_spin_unlock_irqrestore(ptr, flags) \ | |
220 | pcpu_spin_unlock_irqrestore(lock, ptr, flags) | |
72812019 LS |
221 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
222 | DEFINE_PER_CPU(int, numa_node); | |
223 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
224 | #endif | |
225 | ||
4518085e KW |
226 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
227 | ||
7aac7898 LS |
228 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
229 | /* | |
230 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
231 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
232 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
233 | * defined in <linux/topology.h>. | |
234 | */ | |
235 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
236 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
237 | #endif | |
238 | ||
8b885f53 | 239 | static DEFINE_MUTEX(pcpu_drain_mutex); |
bd233f53 | 240 | |
38addce8 | 241 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 242 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
243 | EXPORT_SYMBOL(latent_entropy); |
244 | #endif | |
245 | ||
1da177e4 | 246 | /* |
13808910 | 247 | * Array of node states. |
1da177e4 | 248 | */ |
13808910 CL |
249 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
250 | [N_POSSIBLE] = NODE_MASK_ALL, | |
251 | [N_ONLINE] = { { [0] = 1UL } }, | |
252 | #ifndef CONFIG_NUMA | |
253 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
254 | #ifdef CONFIG_HIGHMEM | |
255 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 256 | #endif |
20b2f52b | 257 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
258 | [N_CPU] = { { [0] = 1UL } }, |
259 | #endif /* NUMA */ | |
260 | }; | |
261 | EXPORT_SYMBOL(node_states); | |
262 | ||
ca79b0c2 AK |
263 | atomic_long_t _totalram_pages __read_mostly; |
264 | EXPORT_SYMBOL(_totalram_pages); | |
cb45b0e9 | 265 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 266 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 267 | |
74f44822 | 268 | int percpu_pagelist_high_fraction; |
dcce284a | 269 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
51cba1eb | 270 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc); |
6471384a AP |
271 | EXPORT_SYMBOL(init_on_alloc); |
272 | ||
51cba1eb | 273 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free); |
6471384a AP |
274 | EXPORT_SYMBOL(init_on_free); |
275 | ||
04013513 VB |
276 | static bool _init_on_alloc_enabled_early __read_mostly |
277 | = IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON); | |
6471384a AP |
278 | static int __init early_init_on_alloc(char *buf) |
279 | { | |
6471384a | 280 | |
04013513 | 281 | return kstrtobool(buf, &_init_on_alloc_enabled_early); |
6471384a AP |
282 | } |
283 | early_param("init_on_alloc", early_init_on_alloc); | |
284 | ||
04013513 VB |
285 | static bool _init_on_free_enabled_early __read_mostly |
286 | = IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON); | |
6471384a AP |
287 | static int __init early_init_on_free(char *buf) |
288 | { | |
04013513 | 289 | return kstrtobool(buf, &_init_on_free_enabled_early); |
6471384a AP |
290 | } |
291 | early_param("init_on_free", early_init_on_free); | |
1da177e4 | 292 | |
bb14c2c7 VB |
293 | /* |
294 | * A cached value of the page's pageblock's migratetype, used when the page is | |
295 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
296 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
297 | * Also the migratetype set in the page does not necessarily match the pcplist | |
298 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
299 | * other index - this ensures that it will be put on the correct CMA freelist. | |
300 | */ | |
301 | static inline int get_pcppage_migratetype(struct page *page) | |
302 | { | |
303 | return page->index; | |
304 | } | |
305 | ||
306 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
307 | { | |
308 | page->index = migratetype; | |
309 | } | |
310 | ||
452aa699 RW |
311 | #ifdef CONFIG_PM_SLEEP |
312 | /* | |
313 | * The following functions are used by the suspend/hibernate code to temporarily | |
314 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
315 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
316 | * they should always be called with system_transition_mutex held |
317 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
318 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
319 | * with that modification). | |
452aa699 | 320 | */ |
c9e664f1 RW |
321 | |
322 | static gfp_t saved_gfp_mask; | |
323 | ||
324 | void pm_restore_gfp_mask(void) | |
452aa699 | 325 | { |
55f2503c | 326 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
327 | if (saved_gfp_mask) { |
328 | gfp_allowed_mask = saved_gfp_mask; | |
329 | saved_gfp_mask = 0; | |
330 | } | |
452aa699 RW |
331 | } |
332 | ||
c9e664f1 | 333 | void pm_restrict_gfp_mask(void) |
452aa699 | 334 | { |
55f2503c | 335 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
336 | WARN_ON(saved_gfp_mask); |
337 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 338 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 339 | } |
f90ac398 MG |
340 | |
341 | bool pm_suspended_storage(void) | |
342 | { | |
d0164adc | 343 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
344 | return false; |
345 | return true; | |
346 | } | |
452aa699 RW |
347 | #endif /* CONFIG_PM_SLEEP */ |
348 | ||
d9c23400 | 349 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 350 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
351 | #endif |
352 | ||
7fef431b DH |
353 | static void __free_pages_ok(struct page *page, unsigned int order, |
354 | fpi_t fpi_flags); | |
a226f6c8 | 355 | |
1da177e4 LT |
356 | /* |
357 | * results with 256, 32 in the lowmem_reserve sysctl: | |
358 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
359 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
360 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
361 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 362 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
363 | * |
364 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
365 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 366 | */ |
d3cda233 | 367 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 368 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 369 | [ZONE_DMA] = 256, |
4b51d669 | 370 | #endif |
fb0e7942 | 371 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 372 | [ZONE_DMA32] = 256, |
fb0e7942 | 373 | #endif |
d3cda233 | 374 | [ZONE_NORMAL] = 32, |
e53ef38d | 375 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 376 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 377 | #endif |
d3cda233 | 378 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 379 | }; |
1da177e4 | 380 | |
15ad7cdc | 381 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 382 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 383 | "DMA", |
4b51d669 | 384 | #endif |
fb0e7942 | 385 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 386 | "DMA32", |
fb0e7942 | 387 | #endif |
2f1b6248 | 388 | "Normal", |
e53ef38d | 389 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 390 | "HighMem", |
e53ef38d | 391 | #endif |
2a1e274a | 392 | "Movable", |
033fbae9 DW |
393 | #ifdef CONFIG_ZONE_DEVICE |
394 | "Device", | |
395 | #endif | |
2f1b6248 CL |
396 | }; |
397 | ||
c999fbd3 | 398 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
399 | "Unmovable", |
400 | "Movable", | |
401 | "Reclaimable", | |
402 | "HighAtomic", | |
403 | #ifdef CONFIG_CMA | |
404 | "CMA", | |
405 | #endif | |
406 | #ifdef CONFIG_MEMORY_ISOLATION | |
407 | "Isolate", | |
408 | #endif | |
409 | }; | |
410 | ||
ae70eddd AK |
411 | compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = { |
412 | [NULL_COMPOUND_DTOR] = NULL, | |
413 | [COMPOUND_PAGE_DTOR] = free_compound_page, | |
f1e61557 | 414 | #ifdef CONFIG_HUGETLB_PAGE |
ae70eddd | 415 | [HUGETLB_PAGE_DTOR] = free_huge_page, |
f1e61557 | 416 | #endif |
9a982250 | 417 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
ae70eddd | 418 | [TRANSHUGE_PAGE_DTOR] = free_transhuge_page, |
9a982250 | 419 | #endif |
f1e61557 KS |
420 | }; |
421 | ||
1da177e4 | 422 | int min_free_kbytes = 1024; |
42aa83cb | 423 | int user_min_free_kbytes = -1; |
1c30844d | 424 | int watermark_boost_factor __read_mostly = 15000; |
795ae7a0 | 425 | int watermark_scale_factor = 10; |
1da177e4 | 426 | |
bbe5d993 OS |
427 | static unsigned long nr_kernel_pages __initdata; |
428 | static unsigned long nr_all_pages __initdata; | |
429 | static unsigned long dma_reserve __initdata; | |
1da177e4 | 430 | |
bbe5d993 OS |
431 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata; |
432 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata; | |
7f16f91f | 433 | static unsigned long required_kernelcore __initdata; |
a5c6d650 | 434 | static unsigned long required_kernelcore_percent __initdata; |
7f16f91f | 435 | static unsigned long required_movablecore __initdata; |
a5c6d650 | 436 | static unsigned long required_movablecore_percent __initdata; |
bbe5d993 | 437 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata; |
902c2d91 | 438 | bool mirrored_kernelcore __initdata_memblock; |
0ee332c1 TH |
439 | |
440 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
441 | int movable_zone; | |
442 | EXPORT_SYMBOL(movable_zone); | |
c713216d | 443 | |
418508c1 | 444 | #if MAX_NUMNODES > 1 |
b9726c26 | 445 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 446 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 447 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 448 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
449 | #endif |
450 | ||
9ef9acb0 MG |
451 | int page_group_by_mobility_disabled __read_mostly; |
452 | ||
3a80a7fa | 453 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
454 | /* |
455 | * During boot we initialize deferred pages on-demand, as needed, but once | |
456 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
457 | * and we can permanently disable that path. | |
458 | */ | |
459 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
460 | ||
94ae8b83 | 461 | static inline bool deferred_pages_enabled(void) |
3c0c12cc | 462 | { |
94ae8b83 | 463 | return static_branch_unlikely(&deferred_pages); |
3c0c12cc WL |
464 | } |
465 | ||
3a80a7fa | 466 | /* Returns true if the struct page for the pfn is uninitialised */ |
0e1cc95b | 467 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 468 | { |
ef70b6f4 MG |
469 | int nid = early_pfn_to_nid(pfn); |
470 | ||
471 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
472 | return true; |
473 | ||
474 | return false; | |
475 | } | |
476 | ||
477 | /* | |
d3035be4 | 478 | * Returns true when the remaining initialisation should be deferred until |
3a80a7fa MG |
479 | * later in the boot cycle when it can be parallelised. |
480 | */ | |
d3035be4 PT |
481 | static bool __meminit |
482 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn) | |
3a80a7fa | 483 | { |
d3035be4 PT |
484 | static unsigned long prev_end_pfn, nr_initialised; |
485 | ||
c4f20f14 LZ |
486 | if (early_page_ext_enabled()) |
487 | return false; | |
d3035be4 PT |
488 | /* |
489 | * prev_end_pfn static that contains the end of previous zone | |
490 | * No need to protect because called very early in boot before smp_init. | |
491 | */ | |
492 | if (prev_end_pfn != end_pfn) { | |
493 | prev_end_pfn = end_pfn; | |
494 | nr_initialised = 0; | |
495 | } | |
496 | ||
3c2c6488 | 497 | /* Always populate low zones for address-constrained allocations */ |
d3035be4 | 498 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid))) |
3a80a7fa | 499 | return false; |
23b68cfa | 500 | |
dc2da7b4 BH |
501 | if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX) |
502 | return true; | |
23b68cfa WY |
503 | /* |
504 | * We start only with one section of pages, more pages are added as | |
505 | * needed until the rest of deferred pages are initialized. | |
506 | */ | |
d3035be4 | 507 | nr_initialised++; |
23b68cfa | 508 | if ((nr_initialised > PAGES_PER_SECTION) && |
d3035be4 PT |
509 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
510 | NODE_DATA(nid)->first_deferred_pfn = pfn; | |
511 | return true; | |
3a80a7fa | 512 | } |
d3035be4 | 513 | return false; |
3a80a7fa MG |
514 | } |
515 | #else | |
94ae8b83 | 516 | static inline bool deferred_pages_enabled(void) |
2c335680 | 517 | { |
94ae8b83 | 518 | return false; |
2c335680 | 519 | } |
3c0c12cc | 520 | |
3a80a7fa MG |
521 | static inline bool early_page_uninitialised(unsigned long pfn) |
522 | { | |
523 | return false; | |
524 | } | |
525 | ||
d3035be4 | 526 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn) |
3a80a7fa | 527 | { |
d3035be4 | 528 | return false; |
3a80a7fa MG |
529 | } |
530 | #endif | |
531 | ||
0b423ca2 | 532 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
ca891f41 | 533 | static inline unsigned long *get_pageblock_bitmap(const struct page *page, |
0b423ca2 MG |
534 | unsigned long pfn) |
535 | { | |
536 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 537 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
538 | #else |
539 | return page_zone(page)->pageblock_flags; | |
540 | #endif /* CONFIG_SPARSEMEM */ | |
541 | } | |
542 | ||
ca891f41 | 543 | static inline int pfn_to_bitidx(const struct page *page, unsigned long pfn) |
0b423ca2 MG |
544 | { |
545 | #ifdef CONFIG_SPARSEMEM | |
546 | pfn &= (PAGES_PER_SECTION-1); | |
0b423ca2 | 547 | #else |
4f9bc69a | 548 | pfn = pfn - pageblock_start_pfn(page_zone(page)->zone_start_pfn); |
0b423ca2 | 549 | #endif /* CONFIG_SPARSEMEM */ |
399b795b | 550 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
0b423ca2 MG |
551 | } |
552 | ||
535b81e2 | 553 | static __always_inline |
ca891f41 | 554 | unsigned long __get_pfnblock_flags_mask(const struct page *page, |
0b423ca2 | 555 | unsigned long pfn, |
0b423ca2 MG |
556 | unsigned long mask) |
557 | { | |
558 | unsigned long *bitmap; | |
559 | unsigned long bitidx, word_bitidx; | |
560 | unsigned long word; | |
561 | ||
562 | bitmap = get_pageblock_bitmap(page, pfn); | |
563 | bitidx = pfn_to_bitidx(page, pfn); | |
564 | word_bitidx = bitidx / BITS_PER_LONG; | |
565 | bitidx &= (BITS_PER_LONG-1); | |
1c563432 MK |
566 | /* |
567 | * This races, without locks, with set_pfnblock_flags_mask(). Ensure | |
568 | * a consistent read of the memory array, so that results, even though | |
569 | * racy, are not corrupted. | |
570 | */ | |
571 | word = READ_ONCE(bitmap[word_bitidx]); | |
d93d5ab9 | 572 | return (word >> bitidx) & mask; |
0b423ca2 MG |
573 | } |
574 | ||
a00cda3f MCC |
575 | /** |
576 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
577 | * @page: The page within the block of interest | |
578 | * @pfn: The target page frame number | |
579 | * @mask: mask of bits that the caller is interested in | |
580 | * | |
581 | * Return: pageblock_bits flags | |
582 | */ | |
ca891f41 MWO |
583 | unsigned long get_pfnblock_flags_mask(const struct page *page, |
584 | unsigned long pfn, unsigned long mask) | |
0b423ca2 | 585 | { |
535b81e2 | 586 | return __get_pfnblock_flags_mask(page, pfn, mask); |
0b423ca2 MG |
587 | } |
588 | ||
ca891f41 MWO |
589 | static __always_inline int get_pfnblock_migratetype(const struct page *page, |
590 | unsigned long pfn) | |
0b423ca2 | 591 | { |
535b81e2 | 592 | return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK); |
0b423ca2 MG |
593 | } |
594 | ||
595 | /** | |
596 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
597 | * @page: The page within the block of interest | |
598 | * @flags: The flags to set | |
599 | * @pfn: The target page frame number | |
0b423ca2 MG |
600 | * @mask: mask of bits that the caller is interested in |
601 | */ | |
602 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
603 | unsigned long pfn, | |
0b423ca2 MG |
604 | unsigned long mask) |
605 | { | |
606 | unsigned long *bitmap; | |
607 | unsigned long bitidx, word_bitidx; | |
04ec0061 | 608 | unsigned long word; |
0b423ca2 MG |
609 | |
610 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 611 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
612 | |
613 | bitmap = get_pageblock_bitmap(page, pfn); | |
614 | bitidx = pfn_to_bitidx(page, pfn); | |
615 | word_bitidx = bitidx / BITS_PER_LONG; | |
616 | bitidx &= (BITS_PER_LONG-1); | |
617 | ||
618 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
619 | ||
d93d5ab9 WY |
620 | mask <<= bitidx; |
621 | flags <<= bitidx; | |
0b423ca2 MG |
622 | |
623 | word = READ_ONCE(bitmap[word_bitidx]); | |
04ec0061 UB |
624 | do { |
625 | } while (!try_cmpxchg(&bitmap[word_bitidx], &word, (word & ~mask) | flags)); | |
0b423ca2 | 626 | } |
3a80a7fa | 627 | |
ee6f509c | 628 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 629 | { |
5d0f3f72 KM |
630 | if (unlikely(page_group_by_mobility_disabled && |
631 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
632 | migratetype = MIGRATE_UNMOVABLE; |
633 | ||
d93d5ab9 | 634 | set_pfnblock_flags_mask(page, (unsigned long)migratetype, |
535b81e2 | 635 | page_to_pfn(page), MIGRATETYPE_MASK); |
b2a0ac88 MG |
636 | } |
637 | ||
13e7444b | 638 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 639 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 640 | { |
bdc8cb98 DH |
641 | int ret = 0; |
642 | unsigned seq; | |
643 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 644 | unsigned long sp, start_pfn; |
c6a57e19 | 645 | |
bdc8cb98 DH |
646 | do { |
647 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
648 | start_pfn = zone->zone_start_pfn; |
649 | sp = zone->spanned_pages; | |
108bcc96 | 650 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
651 | ret = 1; |
652 | } while (zone_span_seqretry(zone, seq)); | |
653 | ||
b5e6a5a2 | 654 | if (ret) |
613813e8 DH |
655 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
656 | pfn, zone_to_nid(zone), zone->name, | |
657 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 658 | |
bdc8cb98 | 659 | return ret; |
c6a57e19 DH |
660 | } |
661 | ||
662 | static int page_is_consistent(struct zone *zone, struct page *page) | |
663 | { | |
1da177e4 | 664 | if (zone != page_zone(page)) |
c6a57e19 DH |
665 | return 0; |
666 | ||
667 | return 1; | |
668 | } | |
669 | /* | |
670 | * Temporary debugging check for pages not lying within a given zone. | |
671 | */ | |
d73d3c9f | 672 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
673 | { |
674 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 675 | return 1; |
c6a57e19 DH |
676 | if (!page_is_consistent(zone, page)) |
677 | return 1; | |
678 | ||
1da177e4 LT |
679 | return 0; |
680 | } | |
13e7444b | 681 | #else |
d73d3c9f | 682 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
683 | { |
684 | return 0; | |
685 | } | |
686 | #endif | |
687 | ||
82a3241a | 688 | static void bad_page(struct page *page, const char *reason) |
1da177e4 | 689 | { |
d936cf9b HD |
690 | static unsigned long resume; |
691 | static unsigned long nr_shown; | |
692 | static unsigned long nr_unshown; | |
693 | ||
694 | /* | |
695 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
696 | * or allow a steady drip of one report per second. | |
697 | */ | |
698 | if (nr_shown == 60) { | |
699 | if (time_before(jiffies, resume)) { | |
700 | nr_unshown++; | |
701 | goto out; | |
702 | } | |
703 | if (nr_unshown) { | |
ff8e8116 | 704 | pr_alert( |
1e9e6365 | 705 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
706 | nr_unshown); |
707 | nr_unshown = 0; | |
708 | } | |
709 | nr_shown = 0; | |
710 | } | |
711 | if (nr_shown++ == 0) | |
712 | resume = jiffies + 60 * HZ; | |
713 | ||
ff8e8116 | 714 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 715 | current->comm, page_to_pfn(page)); |
d2f07ec0 | 716 | dump_page(page, reason); |
3dc14741 | 717 | |
4f31888c | 718 | print_modules(); |
1da177e4 | 719 | dump_stack(); |
d936cf9b | 720 | out: |
8cc3b392 | 721 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 722 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 723 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
724 | } |
725 | ||
44042b44 MG |
726 | static inline unsigned int order_to_pindex(int migratetype, int order) |
727 | { | |
728 | int base = order; | |
729 | ||
730 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
731 | if (order > PAGE_ALLOC_COSTLY_ORDER) { | |
732 | VM_BUG_ON(order != pageblock_order); | |
5d0a661d | 733 | return NR_LOWORDER_PCP_LISTS; |
44042b44 MG |
734 | } |
735 | #else | |
736 | VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER); | |
737 | #endif | |
738 | ||
739 | return (MIGRATE_PCPTYPES * base) + migratetype; | |
740 | } | |
741 | ||
742 | static inline int pindex_to_order(unsigned int pindex) | |
743 | { | |
744 | int order = pindex / MIGRATE_PCPTYPES; | |
745 | ||
746 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5d0a661d | 747 | if (pindex == NR_LOWORDER_PCP_LISTS) |
44042b44 | 748 | order = pageblock_order; |
44042b44 MG |
749 | #else |
750 | VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER); | |
751 | #endif | |
752 | ||
753 | return order; | |
754 | } | |
755 | ||
756 | static inline bool pcp_allowed_order(unsigned int order) | |
757 | { | |
758 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
759 | return true; | |
760 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
761 | if (order == pageblock_order) | |
762 | return true; | |
763 | #endif | |
764 | return false; | |
765 | } | |
766 | ||
21d02f8f MG |
767 | static inline void free_the_page(struct page *page, unsigned int order) |
768 | { | |
44042b44 MG |
769 | if (pcp_allowed_order(order)) /* Via pcp? */ |
770 | free_unref_page(page, order); | |
21d02f8f MG |
771 | else |
772 | __free_pages_ok(page, order, FPI_NONE); | |
773 | } | |
774 | ||
1da177e4 LT |
775 | /* |
776 | * Higher-order pages are called "compound pages". They are structured thusly: | |
777 | * | |
1d798ca3 | 778 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 779 | * |
1d798ca3 KS |
780 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
781 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 782 | * |
1d798ca3 KS |
783 | * The first tail page's ->compound_dtor holds the offset in array of compound |
784 | * page destructors. See compound_page_dtors. | |
1da177e4 | 785 | * |
1d798ca3 | 786 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 787 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 788 | */ |
d98c7a09 | 789 | |
9a982250 | 790 | void free_compound_page(struct page *page) |
d98c7a09 | 791 | { |
bbc6b703 | 792 | mem_cgroup_uncharge(page_folio(page)); |
44042b44 | 793 | free_the_page(page, compound_order(page)); |
d98c7a09 HD |
794 | } |
795 | ||
5b24eeef JM |
796 | static void prep_compound_head(struct page *page, unsigned int order) |
797 | { | |
798 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); | |
799 | set_compound_order(page, order); | |
800 | atomic_set(compound_mapcount_ptr(page), -1); | |
5232c63f | 801 | atomic_set(compound_pincount_ptr(page), 0); |
5b24eeef JM |
802 | } |
803 | ||
804 | static void prep_compound_tail(struct page *head, int tail_idx) | |
805 | { | |
806 | struct page *p = head + tail_idx; | |
807 | ||
808 | p->mapping = TAIL_MAPPING; | |
809 | set_compound_head(p, head); | |
5aae9265 | 810 | set_page_private(p, 0); |
5b24eeef JM |
811 | } |
812 | ||
d00181b9 | 813 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
814 | { |
815 | int i; | |
816 | int nr_pages = 1 << order; | |
817 | ||
18229df5 | 818 | __SetPageHead(page); |
5b24eeef JM |
819 | for (i = 1; i < nr_pages; i++) |
820 | prep_compound_tail(page, i); | |
1378a5ee | 821 | |
5b24eeef | 822 | prep_compound_head(page, order); |
18229df5 AW |
823 | } |
824 | ||
5375336c MWO |
825 | void destroy_large_folio(struct folio *folio) |
826 | { | |
827 | enum compound_dtor_id dtor = folio_page(folio, 1)->compound_dtor; | |
828 | ||
829 | VM_BUG_ON_FOLIO(dtor >= NR_COMPOUND_DTORS, folio); | |
830 | compound_page_dtors[dtor](&folio->page); | |
831 | } | |
832 | ||
c0a32fc5 SG |
833 | #ifdef CONFIG_DEBUG_PAGEALLOC |
834 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 835 | |
8e57f8ac VB |
836 | bool _debug_pagealloc_enabled_early __read_mostly |
837 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
838 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 839 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 840 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
841 | |
842 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 843 | |
031bc574 JK |
844 | static int __init early_debug_pagealloc(char *buf) |
845 | { | |
8e57f8ac | 846 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
847 | } |
848 | early_param("debug_pagealloc", early_debug_pagealloc); | |
849 | ||
c0a32fc5 SG |
850 | static int __init debug_guardpage_minorder_setup(char *buf) |
851 | { | |
852 | unsigned long res; | |
853 | ||
854 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 855 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
856 | return 0; |
857 | } | |
858 | _debug_guardpage_minorder = res; | |
1170532b | 859 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
860 | return 0; |
861 | } | |
f1c1e9f7 | 862 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 863 | |
acbc15a4 | 864 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 865 | unsigned int order, int migratetype) |
c0a32fc5 | 866 | { |
e30825f1 | 867 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
868 | return false; |
869 | ||
870 | if (order >= debug_guardpage_minorder()) | |
871 | return false; | |
e30825f1 | 872 | |
3972f6bb | 873 | __SetPageGuard(page); |
bf75f200 | 874 | INIT_LIST_HEAD(&page->buddy_list); |
2847cf95 JK |
875 | set_page_private(page, order); |
876 | /* Guard pages are not available for any usage */ | |
b3618455 ML |
877 | if (!is_migrate_isolate(migratetype)) |
878 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
879 | |
880 | return true; | |
c0a32fc5 SG |
881 | } |
882 | ||
2847cf95 JK |
883 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
884 | unsigned int order, int migratetype) | |
c0a32fc5 | 885 | { |
e30825f1 JK |
886 | if (!debug_guardpage_enabled()) |
887 | return; | |
888 | ||
3972f6bb | 889 | __ClearPageGuard(page); |
e30825f1 | 890 | |
2847cf95 JK |
891 | set_page_private(page, 0); |
892 | if (!is_migrate_isolate(migratetype)) | |
893 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
894 | } |
895 | #else | |
acbc15a4 JK |
896 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
897 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
898 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
899 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
900 | #endif |
901 | ||
04013513 VB |
902 | /* |
903 | * Enable static keys related to various memory debugging and hardening options. | |
904 | * Some override others, and depend on early params that are evaluated in the | |
905 | * order of appearance. So we need to first gather the full picture of what was | |
906 | * enabled, and then make decisions. | |
907 | */ | |
5749fcc5 | 908 | void __init init_mem_debugging_and_hardening(void) |
04013513 | 909 | { |
9df65f52 ST |
910 | bool page_poisoning_requested = false; |
911 | ||
912 | #ifdef CONFIG_PAGE_POISONING | |
913 | /* | |
914 | * Page poisoning is debug page alloc for some arches. If | |
915 | * either of those options are enabled, enable poisoning. | |
916 | */ | |
917 | if (page_poisoning_enabled() || | |
918 | (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) && | |
919 | debug_pagealloc_enabled())) { | |
920 | static_branch_enable(&_page_poisoning_enabled); | |
921 | page_poisoning_requested = true; | |
922 | } | |
923 | #endif | |
924 | ||
69e5d322 ST |
925 | if ((_init_on_alloc_enabled_early || _init_on_free_enabled_early) && |
926 | page_poisoning_requested) { | |
927 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, " | |
928 | "will take precedence over init_on_alloc and init_on_free\n"); | |
929 | _init_on_alloc_enabled_early = false; | |
930 | _init_on_free_enabled_early = false; | |
04013513 VB |
931 | } |
932 | ||
69e5d322 ST |
933 | if (_init_on_alloc_enabled_early) |
934 | static_branch_enable(&init_on_alloc); | |
935 | else | |
936 | static_branch_disable(&init_on_alloc); | |
937 | ||
938 | if (_init_on_free_enabled_early) | |
939 | static_branch_enable(&init_on_free); | |
940 | else | |
941 | static_branch_disable(&init_on_free); | |
942 | ||
42eaa27d AP |
943 | if (IS_ENABLED(CONFIG_KMSAN) && |
944 | (_init_on_alloc_enabled_early || _init_on_free_enabled_early)) | |
945 | pr_info("mem auto-init: please make sure init_on_alloc and init_on_free are disabled when running KMSAN\n"); | |
946 | ||
04013513 VB |
947 | #ifdef CONFIG_DEBUG_PAGEALLOC |
948 | if (!debug_pagealloc_enabled()) | |
949 | return; | |
950 | ||
951 | static_branch_enable(&_debug_pagealloc_enabled); | |
952 | ||
953 | if (!debug_guardpage_minorder()) | |
954 | return; | |
955 | ||
956 | static_branch_enable(&_debug_guardpage_enabled); | |
957 | #endif | |
958 | } | |
959 | ||
ab130f91 | 960 | static inline void set_buddy_order(struct page *page, unsigned int order) |
6aa3001b | 961 | { |
4c21e2f2 | 962 | set_page_private(page, order); |
676165a8 | 963 | __SetPageBuddy(page); |
1da177e4 LT |
964 | } |
965 | ||
5e1f0f09 MG |
966 | #ifdef CONFIG_COMPACTION |
967 | static inline struct capture_control *task_capc(struct zone *zone) | |
968 | { | |
969 | struct capture_control *capc = current->capture_control; | |
970 | ||
deba0487 | 971 | return unlikely(capc) && |
5e1f0f09 MG |
972 | !(current->flags & PF_KTHREAD) && |
973 | !capc->page && | |
deba0487 | 974 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
975 | } |
976 | ||
977 | static inline bool | |
978 | compaction_capture(struct capture_control *capc, struct page *page, | |
979 | int order, int migratetype) | |
980 | { | |
981 | if (!capc || order != capc->cc->order) | |
982 | return false; | |
983 | ||
984 | /* Do not accidentally pollute CMA or isolated regions*/ | |
985 | if (is_migrate_cma(migratetype) || | |
986 | is_migrate_isolate(migratetype)) | |
987 | return false; | |
988 | ||
989 | /* | |
f0953a1b | 990 | * Do not let lower order allocations pollute a movable pageblock. |
5e1f0f09 MG |
991 | * This might let an unmovable request use a reclaimable pageblock |
992 | * and vice-versa but no more than normal fallback logic which can | |
993 | * have trouble finding a high-order free page. | |
994 | */ | |
995 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
996 | return false; | |
997 | ||
998 | capc->page = page; | |
999 | return true; | |
1000 | } | |
1001 | ||
1002 | #else | |
1003 | static inline struct capture_control *task_capc(struct zone *zone) | |
1004 | { | |
1005 | return NULL; | |
1006 | } | |
1007 | ||
1008 | static inline bool | |
1009 | compaction_capture(struct capture_control *capc, struct page *page, | |
1010 | int order, int migratetype) | |
1011 | { | |
1012 | return false; | |
1013 | } | |
1014 | #endif /* CONFIG_COMPACTION */ | |
1015 | ||
6ab01363 AD |
1016 | /* Used for pages not on another list */ |
1017 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
1018 | unsigned int order, int migratetype) | |
1019 | { | |
1020 | struct free_area *area = &zone->free_area[order]; | |
1021 | ||
bf75f200 | 1022 | list_add(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
1023 | area->nr_free++; |
1024 | } | |
1025 | ||
1026 | /* Used for pages not on another list */ | |
1027 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
1028 | unsigned int order, int migratetype) | |
1029 | { | |
1030 | struct free_area *area = &zone->free_area[order]; | |
1031 | ||
bf75f200 | 1032 | list_add_tail(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
1033 | area->nr_free++; |
1034 | } | |
1035 | ||
293ffa5e DH |
1036 | /* |
1037 | * Used for pages which are on another list. Move the pages to the tail | |
1038 | * of the list - so the moved pages won't immediately be considered for | |
1039 | * allocation again (e.g., optimization for memory onlining). | |
1040 | */ | |
6ab01363 AD |
1041 | static inline void move_to_free_list(struct page *page, struct zone *zone, |
1042 | unsigned int order, int migratetype) | |
1043 | { | |
1044 | struct free_area *area = &zone->free_area[order]; | |
1045 | ||
bf75f200 | 1046 | list_move_tail(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
1047 | } |
1048 | ||
1049 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
1050 | unsigned int order) | |
1051 | { | |
36e66c55 AD |
1052 | /* clear reported state and update reported page count */ |
1053 | if (page_reported(page)) | |
1054 | __ClearPageReported(page); | |
1055 | ||
bf75f200 | 1056 | list_del(&page->buddy_list); |
6ab01363 AD |
1057 | __ClearPageBuddy(page); |
1058 | set_page_private(page, 0); | |
1059 | zone->free_area[order].nr_free--; | |
1060 | } | |
1061 | ||
a2129f24 AD |
1062 | /* |
1063 | * If this is not the largest possible page, check if the buddy | |
1064 | * of the next-highest order is free. If it is, it's possible | |
1065 | * that pages are being freed that will coalesce soon. In case, | |
1066 | * that is happening, add the free page to the tail of the list | |
1067 | * so it's less likely to be used soon and more likely to be merged | |
1068 | * as a higher order page | |
1069 | */ | |
1070 | static inline bool | |
1071 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
1072 | struct page *page, unsigned int order) | |
1073 | { | |
8170ac47 ZY |
1074 | unsigned long higher_page_pfn; |
1075 | struct page *higher_page; | |
a2129f24 AD |
1076 | |
1077 | if (order >= MAX_ORDER - 2) | |
1078 | return false; | |
1079 | ||
8170ac47 ZY |
1080 | higher_page_pfn = buddy_pfn & pfn; |
1081 | higher_page = page + (higher_page_pfn - pfn); | |
a2129f24 | 1082 | |
8170ac47 ZY |
1083 | return find_buddy_page_pfn(higher_page, higher_page_pfn, order + 1, |
1084 | NULL) != NULL; | |
a2129f24 AD |
1085 | } |
1086 | ||
1da177e4 LT |
1087 | /* |
1088 | * Freeing function for a buddy system allocator. | |
1089 | * | |
1090 | * The concept of a buddy system is to maintain direct-mapped table | |
1091 | * (containing bit values) for memory blocks of various "orders". | |
1092 | * The bottom level table contains the map for the smallest allocatable | |
1093 | * units of memory (here, pages), and each level above it describes | |
1094 | * pairs of units from the levels below, hence, "buddies". | |
1095 | * At a high level, all that happens here is marking the table entry | |
1096 | * at the bottom level available, and propagating the changes upward | |
1097 | * as necessary, plus some accounting needed to play nicely with other | |
1098 | * parts of the VM system. | |
1099 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
1100 | * free pages of length of (1 << order) and marked with PageBuddy. |
1101 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 1102 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
1103 | * other. That is, if we allocate a small block, and both were |
1104 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 1105 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 1106 | * triggers coalescing into a block of larger size. |
1da177e4 | 1107 | * |
6d49e352 | 1108 | * -- nyc |
1da177e4 LT |
1109 | */ |
1110 | ||
48db57f8 | 1111 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 1112 | unsigned long pfn, |
ed0ae21d | 1113 | struct zone *zone, unsigned int order, |
f04a5d5d | 1114 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1115 | { |
a2129f24 | 1116 | struct capture_control *capc = task_capc(zone); |
dae37a5d | 1117 | unsigned long buddy_pfn = 0; |
a2129f24 | 1118 | unsigned long combined_pfn; |
a2129f24 AD |
1119 | struct page *buddy; |
1120 | bool to_tail; | |
d9dddbf5 | 1121 | |
d29bb978 | 1122 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 1123 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 1124 | |
ed0ae21d | 1125 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 1126 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 1127 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 1128 | |
76741e77 | 1129 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 1130 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 1131 | |
bb0e28eb | 1132 | while (order < MAX_ORDER - 1) { |
5e1f0f09 MG |
1133 | if (compaction_capture(capc, page, order, migratetype)) { |
1134 | __mod_zone_freepage_state(zone, -(1 << order), | |
1135 | migratetype); | |
1136 | return; | |
1137 | } | |
13ad59df | 1138 | |
8170ac47 ZY |
1139 | buddy = find_buddy_page_pfn(page, pfn, order, &buddy_pfn); |
1140 | if (!buddy) | |
d9dddbf5 | 1141 | goto done_merging; |
bb0e28eb ZY |
1142 | |
1143 | if (unlikely(order >= pageblock_order)) { | |
1144 | /* | |
1145 | * We want to prevent merge between freepages on pageblock | |
1146 | * without fallbacks and normal pageblock. Without this, | |
1147 | * pageblock isolation could cause incorrect freepage or CMA | |
1148 | * accounting or HIGHATOMIC accounting. | |
1149 | */ | |
1150 | int buddy_mt = get_pageblock_migratetype(buddy); | |
1151 | ||
1152 | if (migratetype != buddy_mt | |
1153 | && (!migratetype_is_mergeable(migratetype) || | |
1154 | !migratetype_is_mergeable(buddy_mt))) | |
1155 | goto done_merging; | |
1156 | } | |
1157 | ||
c0a32fc5 SG |
1158 | /* |
1159 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
1160 | * merge with it and move up one order. | |
1161 | */ | |
b03641af | 1162 | if (page_is_guard(buddy)) |
2847cf95 | 1163 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 1164 | else |
6ab01363 | 1165 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
1166 | combined_pfn = buddy_pfn & pfn; |
1167 | page = page + (combined_pfn - pfn); | |
1168 | pfn = combined_pfn; | |
1da177e4 LT |
1169 | order++; |
1170 | } | |
d9dddbf5 VB |
1171 | |
1172 | done_merging: | |
ab130f91 | 1173 | set_buddy_order(page, order); |
6dda9d55 | 1174 | |
47b6a24a DH |
1175 | if (fpi_flags & FPI_TO_TAIL) |
1176 | to_tail = true; | |
1177 | else if (is_shuffle_order(order)) | |
a2129f24 | 1178 | to_tail = shuffle_pick_tail(); |
97500a4a | 1179 | else |
a2129f24 | 1180 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 1181 | |
a2129f24 | 1182 | if (to_tail) |
6ab01363 | 1183 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 1184 | else |
6ab01363 | 1185 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
1186 | |
1187 | /* Notify page reporting subsystem of freed page */ | |
f04a5d5d | 1188 | if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY)) |
36e66c55 | 1189 | page_reporting_notify_free(order); |
1da177e4 LT |
1190 | } |
1191 | ||
b2c9e2fb ZY |
1192 | /** |
1193 | * split_free_page() -- split a free page at split_pfn_offset | |
1194 | * @free_page: the original free page | |
1195 | * @order: the order of the page | |
1196 | * @split_pfn_offset: split offset within the page | |
1197 | * | |
86d28b07 ZY |
1198 | * Return -ENOENT if the free page is changed, otherwise 0 |
1199 | * | |
b2c9e2fb ZY |
1200 | * It is used when the free page crosses two pageblocks with different migratetypes |
1201 | * at split_pfn_offset within the page. The split free page will be put into | |
1202 | * separate migratetype lists afterwards. Otherwise, the function achieves | |
1203 | * nothing. | |
1204 | */ | |
86d28b07 ZY |
1205 | int split_free_page(struct page *free_page, |
1206 | unsigned int order, unsigned long split_pfn_offset) | |
b2c9e2fb ZY |
1207 | { |
1208 | struct zone *zone = page_zone(free_page); | |
1209 | unsigned long free_page_pfn = page_to_pfn(free_page); | |
1210 | unsigned long pfn; | |
1211 | unsigned long flags; | |
1212 | int free_page_order; | |
86d28b07 ZY |
1213 | int mt; |
1214 | int ret = 0; | |
b2c9e2fb | 1215 | |
88ee1343 | 1216 | if (split_pfn_offset == 0) |
86d28b07 | 1217 | return ret; |
88ee1343 | 1218 | |
b2c9e2fb | 1219 | spin_lock_irqsave(&zone->lock, flags); |
86d28b07 ZY |
1220 | |
1221 | if (!PageBuddy(free_page) || buddy_order(free_page) != order) { | |
1222 | ret = -ENOENT; | |
1223 | goto out; | |
1224 | } | |
1225 | ||
1226 | mt = get_pageblock_migratetype(free_page); | |
1227 | if (likely(!is_migrate_isolate(mt))) | |
1228 | __mod_zone_freepage_state(zone, -(1UL << order), mt); | |
1229 | ||
b2c9e2fb ZY |
1230 | del_page_from_free_list(free_page, zone, order); |
1231 | for (pfn = free_page_pfn; | |
1232 | pfn < free_page_pfn + (1UL << order);) { | |
1233 | int mt = get_pfnblock_migratetype(pfn_to_page(pfn), pfn); | |
1234 | ||
86d28b07 | 1235 | free_page_order = min_t(unsigned int, |
88ee1343 ZY |
1236 | pfn ? __ffs(pfn) : order, |
1237 | __fls(split_pfn_offset)); | |
b2c9e2fb ZY |
1238 | __free_one_page(pfn_to_page(pfn), pfn, zone, free_page_order, |
1239 | mt, FPI_NONE); | |
1240 | pfn += 1UL << free_page_order; | |
1241 | split_pfn_offset -= (1UL << free_page_order); | |
1242 | /* we have done the first part, now switch to second part */ | |
1243 | if (split_pfn_offset == 0) | |
1244 | split_pfn_offset = (1UL << order) - (pfn - free_page_pfn); | |
1245 | } | |
86d28b07 | 1246 | out: |
b2c9e2fb | 1247 | spin_unlock_irqrestore(&zone->lock, flags); |
86d28b07 | 1248 | return ret; |
b2c9e2fb | 1249 | } |
7bfec6f4 MG |
1250 | /* |
1251 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1252 | * try and check multiple fields with one check. The caller must do a detailed | |
1253 | * check if necessary. | |
1254 | */ | |
1255 | static inline bool page_expected_state(struct page *page, | |
1256 | unsigned long check_flags) | |
1257 | { | |
1258 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1259 | return false; | |
1260 | ||
1261 | if (unlikely((unsigned long)page->mapping | | |
1262 | page_ref_count(page) | | |
1263 | #ifdef CONFIG_MEMCG | |
48060834 | 1264 | page->memcg_data | |
7bfec6f4 MG |
1265 | #endif |
1266 | (page->flags & check_flags))) | |
1267 | return false; | |
1268 | ||
1269 | return true; | |
1270 | } | |
1271 | ||
58b7f119 | 1272 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 1273 | { |
82a3241a | 1274 | const char *bad_reason = NULL; |
f0b791a3 | 1275 | |
53f9263b | 1276 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1277 | bad_reason = "nonzero mapcount"; |
1278 | if (unlikely(page->mapping != NULL)) | |
1279 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1280 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1281 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
1282 | if (unlikely(page->flags & flags)) { |
1283 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
1284 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
1285 | else | |
1286 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 1287 | } |
9edad6ea | 1288 | #ifdef CONFIG_MEMCG |
48060834 | 1289 | if (unlikely(page->memcg_data)) |
9edad6ea JW |
1290 | bad_reason = "page still charged to cgroup"; |
1291 | #endif | |
58b7f119 WY |
1292 | return bad_reason; |
1293 | } | |
1294 | ||
a8368cd8 | 1295 | static void free_page_is_bad_report(struct page *page) |
58b7f119 WY |
1296 | { |
1297 | bad_page(page, | |
1298 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
1299 | } |
1300 | ||
a8368cd8 | 1301 | static inline bool free_page_is_bad(struct page *page) |
bb552ac6 | 1302 | { |
da838d4f | 1303 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
a8368cd8 | 1304 | return false; |
bb552ac6 MG |
1305 | |
1306 | /* Something has gone sideways, find it */ | |
a8368cd8 AM |
1307 | free_page_is_bad_report(page); |
1308 | return true; | |
1da177e4 LT |
1309 | } |
1310 | ||
4db7548c MG |
1311 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1312 | { | |
1313 | int ret = 1; | |
1314 | ||
1315 | /* | |
1316 | * We rely page->lru.next never has bit 0 set, unless the page | |
1317 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1318 | */ | |
1319 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1320 | ||
1321 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1322 | ret = 0; | |
1323 | goto out; | |
1324 | } | |
1325 | switch (page - head_page) { | |
1326 | case 1: | |
4da1984e | 1327 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c | 1328 | if (unlikely(compound_mapcount(page))) { |
82a3241a | 1329 | bad_page(page, "nonzero compound_mapcount"); |
4db7548c MG |
1330 | goto out; |
1331 | } | |
1332 | break; | |
1333 | case 2: | |
1334 | /* | |
1335 | * the second tail page: ->mapping is | |
fa3015b7 | 1336 | * deferred_list.next -- ignore value. |
4db7548c MG |
1337 | */ |
1338 | break; | |
1339 | default: | |
1340 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 1341 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
1342 | goto out; |
1343 | } | |
1344 | break; | |
1345 | } | |
1346 | if (unlikely(!PageTail(page))) { | |
82a3241a | 1347 | bad_page(page, "PageTail not set"); |
4db7548c MG |
1348 | goto out; |
1349 | } | |
1350 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 1351 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
1352 | goto out; |
1353 | } | |
1354 | ret = 0; | |
1355 | out: | |
1356 | page->mapping = NULL; | |
1357 | clear_compound_head(page); | |
1358 | return ret; | |
1359 | } | |
1360 | ||
94ae8b83 AK |
1361 | /* |
1362 | * Skip KASAN memory poisoning when either: | |
1363 | * | |
1364 | * 1. Deferred memory initialization has not yet completed, | |
1365 | * see the explanation below. | |
1366 | * 2. Skipping poisoning is requested via FPI_SKIP_KASAN_POISON, | |
1367 | * see the comment next to it. | |
1368 | * 3. Skipping poisoning is requested via __GFP_SKIP_KASAN_POISON, | |
1369 | * see the comment next to it. | |
1370 | * | |
1371 | * Poisoning pages during deferred memory init will greatly lengthen the | |
1372 | * process and cause problem in large memory systems as the deferred pages | |
1373 | * initialization is done with interrupt disabled. | |
1374 | * | |
1375 | * Assuming that there will be no reference to those newly initialized | |
1376 | * pages before they are ever allocated, this should have no effect on | |
1377 | * KASAN memory tracking as the poison will be properly inserted at page | |
1378 | * allocation time. The only corner case is when pages are allocated by | |
1379 | * on-demand allocation and then freed again before the deferred pages | |
1380 | * initialization is done, but this is not likely to happen. | |
1381 | */ | |
1382 | static inline bool should_skip_kasan_poison(struct page *page, fpi_t fpi_flags) | |
1383 | { | |
1384 | return deferred_pages_enabled() || | |
1385 | (!IS_ENABLED(CONFIG_KASAN_GENERIC) && | |
1386 | (fpi_flags & FPI_SKIP_KASAN_POISON)) || | |
1387 | PageSkipKASanPoison(page); | |
1388 | } | |
1389 | ||
aeaec8e2 | 1390 | static void kernel_init_pages(struct page *page, int numpages) |
6471384a AP |
1391 | { |
1392 | int i; | |
1393 | ||
9e15afa5 QC |
1394 | /* s390's use of memset() could override KASAN redzones. */ |
1395 | kasan_disable_current(); | |
d9da8f6c AK |
1396 | for (i = 0; i < numpages; i++) |
1397 | clear_highpage_kasan_tagged(page + i); | |
9e15afa5 | 1398 | kasan_enable_current(); |
6471384a AP |
1399 | } |
1400 | ||
e2769dbd | 1401 | static __always_inline bool free_pages_prepare(struct page *page, |
2c335680 | 1402 | unsigned int order, bool check_free, fpi_t fpi_flags) |
4db7548c | 1403 | { |
e2769dbd | 1404 | int bad = 0; |
c3525330 | 1405 | bool init = want_init_on_free(); |
4db7548c | 1406 | |
4db7548c MG |
1407 | VM_BUG_ON_PAGE(PageTail(page), page); |
1408 | ||
e2769dbd | 1409 | trace_mm_page_free(page, order); |
b073d7f8 | 1410 | kmsan_free_page(page, order); |
e2769dbd | 1411 | |
79f5f8fa OS |
1412 | if (unlikely(PageHWPoison(page)) && !order) { |
1413 | /* | |
1414 | * Do not let hwpoison pages hit pcplists/buddy | |
1415 | * Untie memcg state and reset page's owner | |
1416 | */ | |
18b2db3b | 1417 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
79f5f8fa OS |
1418 | __memcg_kmem_uncharge_page(page, order); |
1419 | reset_page_owner(page, order); | |
df4e817b | 1420 | page_table_check_free(page, order); |
79f5f8fa OS |
1421 | return false; |
1422 | } | |
1423 | ||
e2769dbd MG |
1424 | /* |
1425 | * Check tail pages before head page information is cleared to | |
1426 | * avoid checking PageCompound for order-0 pages. | |
1427 | */ | |
1428 | if (unlikely(order)) { | |
1429 | bool compound = PageCompound(page); | |
1430 | int i; | |
1431 | ||
1432 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1433 | |
eac96c3e | 1434 | if (compound) { |
9a73f61b | 1435 | ClearPageDoubleMap(page); |
eac96c3e YS |
1436 | ClearPageHasHWPoisoned(page); |
1437 | } | |
e2769dbd MG |
1438 | for (i = 1; i < (1 << order); i++) { |
1439 | if (compound) | |
1440 | bad += free_tail_pages_check(page, page + i); | |
a8368cd8 | 1441 | if (unlikely(free_page_is_bad(page + i))) { |
e2769dbd MG |
1442 | bad++; |
1443 | continue; | |
1444 | } | |
1445 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1446 | } | |
1447 | } | |
bda807d4 | 1448 | if (PageMappingFlags(page)) |
4db7548c | 1449 | page->mapping = NULL; |
18b2db3b | 1450 | if (memcg_kmem_enabled() && PageMemcgKmem(page)) |
f4b00eab | 1451 | __memcg_kmem_uncharge_page(page, order); |
a8368cd8 AM |
1452 | if (check_free && free_page_is_bad(page)) |
1453 | bad++; | |
e2769dbd MG |
1454 | if (bad) |
1455 | return false; | |
4db7548c | 1456 | |
e2769dbd MG |
1457 | page_cpupid_reset_last(page); |
1458 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1459 | reset_page_owner(page, order); | |
df4e817b | 1460 | page_table_check_free(page, order); |
4db7548c MG |
1461 | |
1462 | if (!PageHighMem(page)) { | |
1463 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1464 | PAGE_SIZE << order); |
4db7548c | 1465 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1466 | PAGE_SIZE << order); |
4db7548c | 1467 | } |
6471384a | 1468 | |
8db26a3d VB |
1469 | kernel_poison_pages(page, 1 << order); |
1470 | ||
f9d79e8d | 1471 | /* |
1bb5eab3 | 1472 | * As memory initialization might be integrated into KASAN, |
7c13c163 | 1473 | * KASAN poisoning and memory initialization code must be |
1bb5eab3 AK |
1474 | * kept together to avoid discrepancies in behavior. |
1475 | * | |
f9d79e8d AK |
1476 | * With hardware tag-based KASAN, memory tags must be set before the |
1477 | * page becomes unavailable via debug_pagealloc or arch_free_page. | |
1478 | */ | |
487a32ec | 1479 | if (!should_skip_kasan_poison(page, fpi_flags)) { |
c3525330 | 1480 | kasan_poison_pages(page, order, init); |
f9d79e8d | 1481 | |
db8a0477 AK |
1482 | /* Memory is already initialized if KASAN did it internally. */ |
1483 | if (kasan_has_integrated_init()) | |
1484 | init = false; | |
1485 | } | |
1486 | if (init) | |
aeaec8e2 | 1487 | kernel_init_pages(page, 1 << order); |
db8a0477 | 1488 | |
234fdce8 QC |
1489 | /* |
1490 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1491 | * does this. So nothing which can access the page's contents should | |
1492 | * happen after this. | |
1493 | */ | |
1494 | arch_free_page(page, order); | |
1495 | ||
77bc7fd6 | 1496 | debug_pagealloc_unmap_pages(page, 1 << order); |
d6332692 | 1497 | |
4db7548c MG |
1498 | return true; |
1499 | } | |
1500 | ||
e2769dbd | 1501 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1502 | /* |
1503 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1504 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1505 | * moved from pcp lists to free lists. | |
1506 | */ | |
44042b44 | 1507 | static bool free_pcp_prepare(struct page *page, unsigned int order) |
e2769dbd | 1508 | { |
44042b44 | 1509 | return free_pages_prepare(page, order, true, FPI_NONE); |
e2769dbd MG |
1510 | } |
1511 | ||
d452289f | 1512 | /* return true if this page has an inappropriate state */ |
4462b32c | 1513 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1514 | { |
8e57f8ac | 1515 | if (debug_pagealloc_enabled_static()) |
a8368cd8 | 1516 | return free_page_is_bad(page); |
4462b32c VB |
1517 | else |
1518 | return false; | |
e2769dbd MG |
1519 | } |
1520 | #else | |
4462b32c VB |
1521 | /* |
1522 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1523 | * moving from pcp lists to free list in order to reduce overhead. With | |
1524 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1525 | * to the pcp lists. | |
1526 | */ | |
44042b44 | 1527 | static bool free_pcp_prepare(struct page *page, unsigned int order) |
e2769dbd | 1528 | { |
8e57f8ac | 1529 | if (debug_pagealloc_enabled_static()) |
44042b44 | 1530 | return free_pages_prepare(page, order, true, FPI_NONE); |
4462b32c | 1531 | else |
44042b44 | 1532 | return free_pages_prepare(page, order, false, FPI_NONE); |
e2769dbd MG |
1533 | } |
1534 | ||
4db7548c MG |
1535 | static bool bulkfree_pcp_prepare(struct page *page) |
1536 | { | |
a8368cd8 | 1537 | return free_page_is_bad(page); |
4db7548c MG |
1538 | } |
1539 | #endif /* CONFIG_DEBUG_VM */ | |
1540 | ||
1da177e4 | 1541 | /* |
5f8dcc21 | 1542 | * Frees a number of pages from the PCP lists |
7cba630b | 1543 | * Assumes all pages on list are in same zone. |
207f36ee | 1544 | * count is the number of pages to free. |
1da177e4 | 1545 | */ |
5f8dcc21 | 1546 | static void free_pcppages_bulk(struct zone *zone, int count, |
fd56eef2 MG |
1547 | struct per_cpu_pages *pcp, |
1548 | int pindex) | |
1da177e4 | 1549 | { |
35b6d770 MG |
1550 | int min_pindex = 0; |
1551 | int max_pindex = NR_PCP_LISTS - 1; | |
44042b44 | 1552 | unsigned int order; |
3777999d | 1553 | bool isolated_pageblocks; |
8b10b465 | 1554 | struct page *page; |
f2260e6b | 1555 | |
88e8ac11 CTR |
1556 | /* |
1557 | * Ensure proper count is passed which otherwise would stuck in the | |
1558 | * below while (list_empty(list)) loop. | |
1559 | */ | |
1560 | count = min(pcp->count, count); | |
d61372bc MG |
1561 | |
1562 | /* Ensure requested pindex is drained first. */ | |
1563 | pindex = pindex - 1; | |
1564 | ||
01b44456 | 1565 | /* Caller must hold IRQ-safe pcp->lock so IRQs are disabled. */ |
8b10b465 MG |
1566 | spin_lock(&zone->lock); |
1567 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1568 | ||
44042b44 | 1569 | while (count > 0) { |
5f8dcc21 | 1570 | struct list_head *list; |
fd56eef2 | 1571 | int nr_pages; |
5f8dcc21 | 1572 | |
fd56eef2 | 1573 | /* Remove pages from lists in a round-robin fashion. */ |
5f8dcc21 | 1574 | do { |
35b6d770 MG |
1575 | if (++pindex > max_pindex) |
1576 | pindex = min_pindex; | |
44042b44 | 1577 | list = &pcp->lists[pindex]; |
35b6d770 MG |
1578 | if (!list_empty(list)) |
1579 | break; | |
1580 | ||
1581 | if (pindex == max_pindex) | |
1582 | max_pindex--; | |
1583 | if (pindex == min_pindex) | |
1584 | min_pindex++; | |
1585 | } while (1); | |
48db57f8 | 1586 | |
44042b44 | 1587 | order = pindex_to_order(pindex); |
fd56eef2 | 1588 | nr_pages = 1 << order; |
a6f9edd6 | 1589 | do { |
8b10b465 MG |
1590 | int mt; |
1591 | ||
bf75f200 | 1592 | page = list_last_entry(list, struct page, pcp_list); |
8b10b465 MG |
1593 | mt = get_pcppage_migratetype(page); |
1594 | ||
0a5f4e5b | 1595 | /* must delete to avoid corrupting pcp list */ |
bf75f200 | 1596 | list_del(&page->pcp_list); |
fd56eef2 MG |
1597 | count -= nr_pages; |
1598 | pcp->count -= nr_pages; | |
aa016d14 | 1599 | |
4db7548c MG |
1600 | if (bulkfree_pcp_prepare(page)) |
1601 | continue; | |
1602 | ||
8b10b465 MG |
1603 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1604 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1605 | /* Pageblock could have been isolated meanwhile */ | |
1606 | if (unlikely(isolated_pageblocks)) | |
1607 | mt = get_pageblock_migratetype(page); | |
0a5f4e5b | 1608 | |
8b10b465 MG |
1609 | __free_one_page(page, page_to_pfn(page), zone, order, mt, FPI_NONE); |
1610 | trace_mm_page_pcpu_drain(page, order, mt); | |
1611 | } while (count > 0 && !list_empty(list)); | |
0a5f4e5b | 1612 | } |
8b10b465 | 1613 | |
d34b0733 | 1614 | spin_unlock(&zone->lock); |
1da177e4 LT |
1615 | } |
1616 | ||
dc4b0caf MG |
1617 | static void free_one_page(struct zone *zone, |
1618 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1619 | unsigned int order, |
7fef431b | 1620 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1621 | { |
df1acc85 MG |
1622 | unsigned long flags; |
1623 | ||
1624 | spin_lock_irqsave(&zone->lock, flags); | |
ad53f92e JK |
1625 | if (unlikely(has_isolate_pageblock(zone) || |
1626 | is_migrate_isolate(migratetype))) { | |
1627 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1628 | } |
7fef431b | 1629 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); |
df1acc85 | 1630 | spin_unlock_irqrestore(&zone->lock, flags); |
48db57f8 NP |
1631 | } |
1632 | ||
1e8ce83c | 1633 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1634 | unsigned long zone, int nid) |
1e8ce83c | 1635 | { |
d0dc12e8 | 1636 | mm_zero_struct_page(page); |
1e8ce83c | 1637 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1638 | init_page_count(page); |
1639 | page_mapcount_reset(page); | |
1640 | page_cpupid_reset_last(page); | |
2813b9c0 | 1641 | page_kasan_tag_reset(page); |
1e8ce83c | 1642 | |
1e8ce83c RH |
1643 | INIT_LIST_HEAD(&page->lru); |
1644 | #ifdef WANT_PAGE_VIRTUAL | |
1645 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1646 | if (!is_highmem_idx(zone)) | |
1647 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1648 | #endif | |
1649 | } | |
1650 | ||
7e18adb4 | 1651 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1652 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1653 | { |
1654 | pg_data_t *pgdat; | |
1655 | int nid, zid; | |
1656 | ||
1657 | if (!early_page_uninitialised(pfn)) | |
1658 | return; | |
1659 | ||
1660 | nid = early_pfn_to_nid(pfn); | |
1661 | pgdat = NODE_DATA(nid); | |
1662 | ||
1663 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1664 | struct zone *zone = &pgdat->node_zones[zid]; | |
1665 | ||
86fb05b9 | 1666 | if (zone_spans_pfn(zone, pfn)) |
7e18adb4 MG |
1667 | break; |
1668 | } | |
d0dc12e8 | 1669 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1670 | } |
1671 | #else | |
1672 | static inline void init_reserved_page(unsigned long pfn) | |
1673 | { | |
1674 | } | |
1675 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1676 | ||
92923ca3 NZ |
1677 | /* |
1678 | * Initialised pages do not have PageReserved set. This function is | |
1679 | * called for each range allocated by the bootmem allocator and | |
1680 | * marks the pages PageReserved. The remaining valid pages are later | |
1681 | * sent to the buddy page allocator. | |
1682 | */ | |
4b50bcc7 | 1683 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1684 | { |
1685 | unsigned long start_pfn = PFN_DOWN(start); | |
1686 | unsigned long end_pfn = PFN_UP(end); | |
1687 | ||
7e18adb4 MG |
1688 | for (; start_pfn < end_pfn; start_pfn++) { |
1689 | if (pfn_valid(start_pfn)) { | |
1690 | struct page *page = pfn_to_page(start_pfn); | |
1691 | ||
1692 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1693 | |
1694 | /* Avoid false-positive PageTail() */ | |
1695 | INIT_LIST_HEAD(&page->lru); | |
1696 | ||
d483da5b AD |
1697 | /* |
1698 | * no need for atomic set_bit because the struct | |
1699 | * page is not visible yet so nobody should | |
1700 | * access it yet. | |
1701 | */ | |
1702 | __SetPageReserved(page); | |
7e18adb4 MG |
1703 | } |
1704 | } | |
92923ca3 NZ |
1705 | } |
1706 | ||
7fef431b DH |
1707 | static void __free_pages_ok(struct page *page, unsigned int order, |
1708 | fpi_t fpi_flags) | |
ec95f53a | 1709 | { |
d34b0733 | 1710 | unsigned long flags; |
95e34412 | 1711 | int migratetype; |
dc4b0caf | 1712 | unsigned long pfn = page_to_pfn(page); |
56f0e661 | 1713 | struct zone *zone = page_zone(page); |
ec95f53a | 1714 | |
2c335680 | 1715 | if (!free_pages_prepare(page, order, true, fpi_flags)) |
ec95f53a KM |
1716 | return; |
1717 | ||
cfc47a28 | 1718 | migratetype = get_pfnblock_migratetype(page, pfn); |
dbbee9d5 | 1719 | |
56f0e661 | 1720 | spin_lock_irqsave(&zone->lock, flags); |
56f0e661 MG |
1721 | if (unlikely(has_isolate_pageblock(zone) || |
1722 | is_migrate_isolate(migratetype))) { | |
1723 | migratetype = get_pfnblock_migratetype(page, pfn); | |
1724 | } | |
1725 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); | |
1726 | spin_unlock_irqrestore(&zone->lock, flags); | |
90249993 | 1727 | |
d34b0733 | 1728 | __count_vm_events(PGFREE, 1 << order); |
1da177e4 LT |
1729 | } |
1730 | ||
a9cd410a | 1731 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1732 | { |
c3993076 | 1733 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1734 | struct page *p = page; |
c3993076 | 1735 | unsigned int loop; |
a226f6c8 | 1736 | |
7fef431b DH |
1737 | /* |
1738 | * When initializing the memmap, __init_single_page() sets the refcount | |
1739 | * of all pages to 1 ("allocated"/"not free"). We have to set the | |
1740 | * refcount of all involved pages to 0. | |
1741 | */ | |
e2d0bd2b YL |
1742 | prefetchw(p); |
1743 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1744 | prefetchw(p + 1); | |
c3993076 JW |
1745 | __ClearPageReserved(p); |
1746 | set_page_count(p, 0); | |
a226f6c8 | 1747 | } |
e2d0bd2b YL |
1748 | __ClearPageReserved(p); |
1749 | set_page_count(p, 0); | |
c3993076 | 1750 | |
9705bea5 | 1751 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
7fef431b DH |
1752 | |
1753 | /* | |
1754 | * Bypass PCP and place fresh pages right to the tail, primarily | |
1755 | * relevant for memory onlining. | |
1756 | */ | |
2c335680 | 1757 | __free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON); |
a226f6c8 DH |
1758 | } |
1759 | ||
a9ee6cf5 | 1760 | #ifdef CONFIG_NUMA |
7ace9917 | 1761 | |
03e92a5e MR |
1762 | /* |
1763 | * During memory init memblocks map pfns to nids. The search is expensive and | |
1764 | * this caches recent lookups. The implementation of __early_pfn_to_nid | |
1765 | * treats start/end as pfns. | |
1766 | */ | |
1767 | struct mminit_pfnnid_cache { | |
1768 | unsigned long last_start; | |
1769 | unsigned long last_end; | |
1770 | int last_nid; | |
1771 | }; | |
75a592a4 | 1772 | |
03e92a5e | 1773 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
6f24fbd3 MR |
1774 | |
1775 | /* | |
1776 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
1777 | */ | |
03e92a5e | 1778 | static int __meminit __early_pfn_to_nid(unsigned long pfn, |
6f24fbd3 | 1779 | struct mminit_pfnnid_cache *state) |
75a592a4 | 1780 | { |
6f24fbd3 | 1781 | unsigned long start_pfn, end_pfn; |
75a592a4 MG |
1782 | int nid; |
1783 | ||
6f24fbd3 MR |
1784 | if (state->last_start <= pfn && pfn < state->last_end) |
1785 | return state->last_nid; | |
1786 | ||
1787 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); | |
1788 | if (nid != NUMA_NO_NODE) { | |
1789 | state->last_start = start_pfn; | |
1790 | state->last_end = end_pfn; | |
1791 | state->last_nid = nid; | |
1792 | } | |
7ace9917 MG |
1793 | |
1794 | return nid; | |
75a592a4 | 1795 | } |
75a592a4 | 1796 | |
75a592a4 | 1797 | int __meminit early_pfn_to_nid(unsigned long pfn) |
75a592a4 | 1798 | { |
7ace9917 | 1799 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1800 | int nid; |
1801 | ||
7ace9917 | 1802 | spin_lock(&early_pfn_lock); |
56ec43d8 | 1803 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1804 | if (nid < 0) |
e4568d38 | 1805 | nid = first_online_node; |
7ace9917 | 1806 | spin_unlock(&early_pfn_lock); |
75a592a4 | 1807 | |
7ace9917 | 1808 | return nid; |
75a592a4 | 1809 | } |
a9ee6cf5 | 1810 | #endif /* CONFIG_NUMA */ |
75a592a4 | 1811 | |
7c2ee349 | 1812 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1813 | unsigned int order) |
1814 | { | |
1815 | if (early_page_uninitialised(pfn)) | |
1816 | return; | |
3c206509 AP |
1817 | if (!kmsan_memblock_free_pages(page, order)) { |
1818 | /* KMSAN will take care of these pages. */ | |
1819 | return; | |
1820 | } | |
a9cd410a | 1821 | __free_pages_core(page, order); |
3a80a7fa MG |
1822 | } |
1823 | ||
7cf91a98 JK |
1824 | /* |
1825 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1826 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
859a85dd | 1827 | * with the migration of free compaction scanner. |
7cf91a98 JK |
1828 | * |
1829 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1830 | * | |
1831 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1832 | * i.e. it's possible that all pages within a zones range of pages do not | |
1833 | * belong to a single zone. We assume that a border between node0 and node1 | |
1834 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1835 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1836 | * the first and last page of a pageblock and avoid checking each individual | |
1837 | * page in a pageblock. | |
1838 | */ | |
1839 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1840 | unsigned long end_pfn, struct zone *zone) | |
1841 | { | |
1842 | struct page *start_page; | |
1843 | struct page *end_page; | |
1844 | ||
1845 | /* end_pfn is one past the range we are checking */ | |
1846 | end_pfn--; | |
1847 | ||
1848 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1849 | return NULL; | |
1850 | ||
2d070eab MH |
1851 | start_page = pfn_to_online_page(start_pfn); |
1852 | if (!start_page) | |
1853 | return NULL; | |
7cf91a98 JK |
1854 | |
1855 | if (page_zone(start_page) != zone) | |
1856 | return NULL; | |
1857 | ||
1858 | end_page = pfn_to_page(end_pfn); | |
1859 | ||
1860 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1861 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1862 | return NULL; | |
1863 | ||
1864 | return start_page; | |
1865 | } | |
1866 | ||
1867 | void set_zone_contiguous(struct zone *zone) | |
1868 | { | |
1869 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1870 | unsigned long block_end_pfn; | |
1871 | ||
4f9bc69a | 1872 | block_end_pfn = pageblock_end_pfn(block_start_pfn); |
7cf91a98 JK |
1873 | for (; block_start_pfn < zone_end_pfn(zone); |
1874 | block_start_pfn = block_end_pfn, | |
1875 | block_end_pfn += pageblock_nr_pages) { | |
1876 | ||
1877 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1878 | ||
1879 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1880 | block_end_pfn, zone)) | |
1881 | return; | |
e84fe99b | 1882 | cond_resched(); |
7cf91a98 JK |
1883 | } |
1884 | ||
1885 | /* We confirm that there is no hole */ | |
1886 | zone->contiguous = true; | |
1887 | } | |
1888 | ||
1889 | void clear_zone_contiguous(struct zone *zone) | |
1890 | { | |
1891 | zone->contiguous = false; | |
1892 | } | |
1893 | ||
7e18adb4 | 1894 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1895 | static void __init deferred_free_range(unsigned long pfn, |
1896 | unsigned long nr_pages) | |
a4de83dd | 1897 | { |
2f47a91f PT |
1898 | struct page *page; |
1899 | unsigned long i; | |
a4de83dd | 1900 | |
2f47a91f | 1901 | if (!nr_pages) |
a4de83dd MG |
1902 | return; |
1903 | ||
2f47a91f PT |
1904 | page = pfn_to_page(pfn); |
1905 | ||
a4de83dd | 1906 | /* Free a large naturally-aligned chunk if possible */ |
ee0913c4 | 1907 | if (nr_pages == pageblock_nr_pages && pageblock_aligned(pfn)) { |
ac5d2539 | 1908 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1909 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1910 | return; |
1911 | } | |
1912 | ||
e780149b | 1913 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
ee0913c4 | 1914 | if (pageblock_aligned(pfn)) |
e780149b | 1915 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1916 | __free_pages_core(page, 0); |
e780149b | 1917 | } |
a4de83dd MG |
1918 | } |
1919 | ||
d3cd131d NS |
1920 | /* Completion tracking for deferred_init_memmap() threads */ |
1921 | static atomic_t pgdat_init_n_undone __initdata; | |
1922 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1923 | ||
1924 | static inline void __init pgdat_init_report_one_done(void) | |
1925 | { | |
1926 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1927 | complete(&pgdat_init_all_done_comp); | |
1928 | } | |
0e1cc95b | 1929 | |
2f47a91f | 1930 | /* |
80b1f41c PT |
1931 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1932 | * | |
c9b3637f | 1933 | * We check if a current large page is valid by only checking the validity |
80b1f41c | 1934 | * of the head pfn. |
2f47a91f | 1935 | */ |
56ec43d8 | 1936 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1937 | { |
ee0913c4 | 1938 | if (pageblock_aligned(pfn) && !pfn_valid(pfn)) |
80b1f41c | 1939 | return false; |
80b1f41c PT |
1940 | return true; |
1941 | } | |
2f47a91f | 1942 | |
80b1f41c PT |
1943 | /* |
1944 | * Free pages to buddy allocator. Try to free aligned pages in | |
1945 | * pageblock_nr_pages sizes. | |
1946 | */ | |
56ec43d8 | 1947 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1948 | unsigned long end_pfn) |
1949 | { | |
80b1f41c | 1950 | unsigned long nr_free = 0; |
2f47a91f | 1951 | |
80b1f41c | 1952 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1953 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1954 | deferred_free_range(pfn - nr_free, nr_free); |
1955 | nr_free = 0; | |
ee0913c4 | 1956 | } else if (pageblock_aligned(pfn)) { |
80b1f41c PT |
1957 | deferred_free_range(pfn - nr_free, nr_free); |
1958 | nr_free = 1; | |
80b1f41c PT |
1959 | } else { |
1960 | nr_free++; | |
1961 | } | |
1962 | } | |
1963 | /* Free the last block of pages to allocator */ | |
1964 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1965 | } |
1966 | ||
80b1f41c PT |
1967 | /* |
1968 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1969 | * by performing it only once every pageblock_nr_pages. | |
1970 | * Return number of pages initialized. | |
1971 | */ | |
56ec43d8 | 1972 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1973 | unsigned long pfn, |
1974 | unsigned long end_pfn) | |
2f47a91f | 1975 | { |
56ec43d8 | 1976 | int nid = zone_to_nid(zone); |
2f47a91f | 1977 | unsigned long nr_pages = 0; |
56ec43d8 | 1978 | int zid = zone_idx(zone); |
2f47a91f | 1979 | struct page *page = NULL; |
2f47a91f | 1980 | |
80b1f41c | 1981 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1982 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1983 | page = NULL; |
2f47a91f | 1984 | continue; |
ee0913c4 | 1985 | } else if (!page || pageblock_aligned(pfn)) { |
2f47a91f | 1986 | page = pfn_to_page(pfn); |
80b1f41c PT |
1987 | } else { |
1988 | page++; | |
2f47a91f | 1989 | } |
d0dc12e8 | 1990 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1991 | nr_pages++; |
2f47a91f | 1992 | } |
80b1f41c | 1993 | return (nr_pages); |
2f47a91f PT |
1994 | } |
1995 | ||
0e56acae AD |
1996 | /* |
1997 | * This function is meant to pre-load the iterator for the zone init. | |
1998 | * Specifically it walks through the ranges until we are caught up to the | |
1999 | * first_init_pfn value and exits there. If we never encounter the value we | |
2000 | * return false indicating there are no valid ranges left. | |
2001 | */ | |
2002 | static bool __init | |
2003 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
2004 | unsigned long *spfn, unsigned long *epfn, | |
2005 | unsigned long first_init_pfn) | |
2006 | { | |
2007 | u64 j; | |
2008 | ||
2009 | /* | |
2010 | * Start out by walking through the ranges in this zone that have | |
2011 | * already been initialized. We don't need to do anything with them | |
2012 | * so we just need to flush them out of the system. | |
2013 | */ | |
2014 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
2015 | if (*epfn <= first_init_pfn) | |
2016 | continue; | |
2017 | if (*spfn < first_init_pfn) | |
2018 | *spfn = first_init_pfn; | |
2019 | *i = j; | |
2020 | return true; | |
2021 | } | |
2022 | ||
2023 | return false; | |
2024 | } | |
2025 | ||
2026 | /* | |
2027 | * Initialize and free pages. We do it in two loops: first we initialize | |
2028 | * struct page, then free to buddy allocator, because while we are | |
2029 | * freeing pages we can access pages that are ahead (computing buddy | |
2030 | * page in __free_one_page()). | |
2031 | * | |
2032 | * In order to try and keep some memory in the cache we have the loop | |
2033 | * broken along max page order boundaries. This way we will not cause | |
2034 | * any issues with the buddy page computation. | |
2035 | */ | |
2036 | static unsigned long __init | |
2037 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
2038 | unsigned long *end_pfn) | |
2039 | { | |
2040 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
2041 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
2042 | unsigned long nr_pages = 0; | |
2043 | u64 j = *i; | |
2044 | ||
2045 | /* First we loop through and initialize the page values */ | |
2046 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
2047 | unsigned long t; | |
2048 | ||
2049 | if (mo_pfn <= *start_pfn) | |
2050 | break; | |
2051 | ||
2052 | t = min(mo_pfn, *end_pfn); | |
2053 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
2054 | ||
2055 | if (mo_pfn < *end_pfn) { | |
2056 | *start_pfn = mo_pfn; | |
2057 | break; | |
2058 | } | |
2059 | } | |
2060 | ||
2061 | /* Reset values and now loop through freeing pages as needed */ | |
2062 | swap(j, *i); | |
2063 | ||
2064 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
2065 | unsigned long t; | |
2066 | ||
2067 | if (mo_pfn <= spfn) | |
2068 | break; | |
2069 | ||
2070 | t = min(mo_pfn, epfn); | |
2071 | deferred_free_pages(spfn, t); | |
2072 | ||
2073 | if (mo_pfn <= epfn) | |
2074 | break; | |
2075 | } | |
2076 | ||
2077 | return nr_pages; | |
2078 | } | |
2079 | ||
e4443149 DJ |
2080 | static void __init |
2081 | deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, | |
2082 | void *arg) | |
2083 | { | |
2084 | unsigned long spfn, epfn; | |
2085 | struct zone *zone = arg; | |
2086 | u64 i; | |
2087 | ||
2088 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn); | |
2089 | ||
2090 | /* | |
2091 | * Initialize and free pages in MAX_ORDER sized increments so that we | |
2092 | * can avoid introducing any issues with the buddy allocator. | |
2093 | */ | |
2094 | while (spfn < end_pfn) { | |
2095 | deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
2096 | cond_resched(); | |
2097 | } | |
2098 | } | |
2099 | ||
ecd09650 DJ |
2100 | /* An arch may override for more concurrency. */ |
2101 | __weak int __init | |
2102 | deferred_page_init_max_threads(const struct cpumask *node_cpumask) | |
2103 | { | |
2104 | return 1; | |
2105 | } | |
2106 | ||
7e18adb4 | 2107 | /* Initialise remaining memory on a node */ |
0e1cc95b | 2108 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 2109 | { |
0e1cc95b | 2110 | pg_data_t *pgdat = data; |
0e56acae | 2111 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
89c7c402 | 2112 | unsigned long spfn = 0, epfn = 0; |
0e56acae | 2113 | unsigned long first_init_pfn, flags; |
7e18adb4 | 2114 | unsigned long start = jiffies; |
7e18adb4 | 2115 | struct zone *zone; |
e4443149 | 2116 | int zid, max_threads; |
2f47a91f | 2117 | u64 i; |
7e18adb4 | 2118 | |
3a2d7fa8 PT |
2119 | /* Bind memory initialisation thread to a local node if possible */ |
2120 | if (!cpumask_empty(cpumask)) | |
2121 | set_cpus_allowed_ptr(current, cpumask); | |
2122 | ||
2123 | pgdat_resize_lock(pgdat, &flags); | |
2124 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 2125 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 2126 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 2127 | pgdat_init_report_one_done(); |
0e1cc95b MG |
2128 | return 0; |
2129 | } | |
2130 | ||
7e18adb4 MG |
2131 | /* Sanity check boundaries */ |
2132 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
2133 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
2134 | pgdat->first_deferred_pfn = ULONG_MAX; | |
2135 | ||
3d060856 PT |
2136 | /* |
2137 | * Once we unlock here, the zone cannot be grown anymore, thus if an | |
2138 | * interrupt thread must allocate this early in boot, zone must be | |
2139 | * pre-grown prior to start of deferred page initialization. | |
2140 | */ | |
2141 | pgdat_resize_unlock(pgdat, &flags); | |
2142 | ||
7e18adb4 MG |
2143 | /* Only the highest zone is deferred so find it */ |
2144 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
2145 | zone = pgdat->node_zones + zid; | |
2146 | if (first_init_pfn < zone_end_pfn(zone)) | |
2147 | break; | |
2148 | } | |
0e56acae AD |
2149 | |
2150 | /* If the zone is empty somebody else may have cleared out the zone */ | |
2151 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2152 | first_init_pfn)) | |
2153 | goto zone_empty; | |
7e18adb4 | 2154 | |
ecd09650 | 2155 | max_threads = deferred_page_init_max_threads(cpumask); |
7e18adb4 | 2156 | |
117003c3 | 2157 | while (spfn < epfn) { |
e4443149 DJ |
2158 | unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION); |
2159 | struct padata_mt_job job = { | |
2160 | .thread_fn = deferred_init_memmap_chunk, | |
2161 | .fn_arg = zone, | |
2162 | .start = spfn, | |
2163 | .size = epfn_align - spfn, | |
2164 | .align = PAGES_PER_SECTION, | |
2165 | .min_chunk = PAGES_PER_SECTION, | |
2166 | .max_threads = max_threads, | |
2167 | }; | |
2168 | ||
2169 | padata_do_multithreaded(&job); | |
2170 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2171 | epfn_align); | |
117003c3 | 2172 | } |
0e56acae | 2173 | zone_empty: |
7e18adb4 MG |
2174 | /* Sanity check that the next zone really is unpopulated */ |
2175 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
2176 | ||
89c7c402 DJ |
2177 | pr_info("node %d deferred pages initialised in %ums\n", |
2178 | pgdat->node_id, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
2179 | |
2180 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
2181 | return 0; |
2182 | } | |
c9e97a19 | 2183 | |
c9e97a19 PT |
2184 | /* |
2185 | * If this zone has deferred pages, try to grow it by initializing enough | |
2186 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
2187 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
2188 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
2189 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
2190 | * | |
2191 | * Return true when zone was grown, otherwise return false. We return true even | |
2192 | * when we grow less than requested, to let the caller decide if there are | |
2193 | * enough pages to satisfy the allocation. | |
2194 | * | |
2195 | * Note: We use noinline because this function is needed only during boot, and | |
2196 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
2197 | * making sure that it is not inlined into permanent text section. | |
2198 | */ | |
2199 | static noinline bool __init | |
2200 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
2201 | { | |
c9e97a19 | 2202 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 2203 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 2204 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
2205 | unsigned long spfn, epfn, flags; |
2206 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
2207 | u64 i; |
2208 | ||
2209 | /* Only the last zone may have deferred pages */ | |
2210 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
2211 | return false; | |
2212 | ||
2213 | pgdat_resize_lock(pgdat, &flags); | |
2214 | ||
c9e97a19 PT |
2215 | /* |
2216 | * If someone grew this zone while we were waiting for spinlock, return | |
2217 | * true, as there might be enough pages already. | |
2218 | */ | |
2219 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
2220 | pgdat_resize_unlock(pgdat, &flags); | |
2221 | return true; | |
2222 | } | |
2223 | ||
0e56acae AD |
2224 | /* If the zone is empty somebody else may have cleared out the zone */ |
2225 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
2226 | first_deferred_pfn)) { | |
2227 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 2228 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
2229 | /* Retry only once. */ |
2230 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
2231 | } |
2232 | ||
0e56acae AD |
2233 | /* |
2234 | * Initialize and free pages in MAX_ORDER sized increments so | |
2235 | * that we can avoid introducing any issues with the buddy | |
2236 | * allocator. | |
2237 | */ | |
2238 | while (spfn < epfn) { | |
2239 | /* update our first deferred PFN for this section */ | |
2240 | first_deferred_pfn = spfn; | |
2241 | ||
2242 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
117003c3 | 2243 | touch_nmi_watchdog(); |
c9e97a19 | 2244 | |
0e56acae AD |
2245 | /* We should only stop along section boundaries */ |
2246 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
2247 | continue; | |
c9e97a19 | 2248 | |
0e56acae | 2249 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
2250 | if (nr_pages >= nr_pages_needed) |
2251 | break; | |
2252 | } | |
2253 | ||
0e56acae | 2254 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
2255 | pgdat_resize_unlock(pgdat, &flags); |
2256 | ||
2257 | return nr_pages > 0; | |
2258 | } | |
2259 | ||
2260 | /* | |
2261 | * deferred_grow_zone() is __init, but it is called from | |
2262 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
2263 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
2264 | * and to ensure that the function body gets unloaded. | |
2265 | */ | |
2266 | static bool __ref | |
2267 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
2268 | { | |
2269 | return deferred_grow_zone(zone, order); | |
2270 | } | |
2271 | ||
7cf91a98 | 2272 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
2273 | |
2274 | void __init page_alloc_init_late(void) | |
2275 | { | |
7cf91a98 | 2276 | struct zone *zone; |
e900a918 | 2277 | int nid; |
7cf91a98 JK |
2278 | |
2279 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 2280 | |
d3cd131d NS |
2281 | /* There will be num_node_state(N_MEMORY) threads */ |
2282 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 2283 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
2284 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
2285 | } | |
2286 | ||
2287 | /* Block until all are initialised */ | |
d3cd131d | 2288 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 2289 | |
c9e97a19 PT |
2290 | /* |
2291 | * We initialized the rest of the deferred pages. Permanently disable | |
2292 | * on-demand struct page initialization. | |
2293 | */ | |
2294 | static_branch_disable(&deferred_pages); | |
2295 | ||
4248b0da MG |
2296 | /* Reinit limits that are based on free pages after the kernel is up */ |
2297 | files_maxfiles_init(); | |
7cf91a98 | 2298 | #endif |
350e88ba | 2299 | |
ba8f3587 LF |
2300 | buffer_init(); |
2301 | ||
3010f876 PT |
2302 | /* Discard memblock private memory */ |
2303 | memblock_discard(); | |
7cf91a98 | 2304 | |
e900a918 DW |
2305 | for_each_node_state(nid, N_MEMORY) |
2306 | shuffle_free_memory(NODE_DATA(nid)); | |
2307 | ||
7cf91a98 JK |
2308 | for_each_populated_zone(zone) |
2309 | set_zone_contiguous(zone); | |
7e18adb4 | 2310 | } |
7e18adb4 | 2311 | |
47118af0 | 2312 | #ifdef CONFIG_CMA |
9cf510a5 | 2313 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
2314 | void __init init_cma_reserved_pageblock(struct page *page) |
2315 | { | |
2316 | unsigned i = pageblock_nr_pages; | |
2317 | struct page *p = page; | |
2318 | ||
2319 | do { | |
2320 | __ClearPageReserved(p); | |
2321 | set_page_count(p, 0); | |
d883c6cf | 2322 | } while (++p, --i); |
47118af0 | 2323 | |
47118af0 | 2324 | set_pageblock_migratetype(page, MIGRATE_CMA); |
b3d40a2b DH |
2325 | set_page_refcounted(page); |
2326 | __free_pages(page, pageblock_order); | |
dc78327c | 2327 | |
3dcc0571 | 2328 | adjust_managed_page_count(page, pageblock_nr_pages); |
3c381db1 | 2329 | page_zone(page)->cma_pages += pageblock_nr_pages; |
47118af0 MN |
2330 | } |
2331 | #endif | |
1da177e4 LT |
2332 | |
2333 | /* | |
2334 | * The order of subdivision here is critical for the IO subsystem. | |
2335 | * Please do not alter this order without good reasons and regression | |
2336 | * testing. Specifically, as large blocks of memory are subdivided, | |
2337 | * the order in which smaller blocks are delivered depends on the order | |
2338 | * they're subdivided in this function. This is the primary factor | |
2339 | * influencing the order in which pages are delivered to the IO | |
2340 | * subsystem according to empirical testing, and this is also justified | |
2341 | * by considering the behavior of a buddy system containing a single | |
2342 | * large block of memory acted on by a series of small allocations. | |
2343 | * This behavior is a critical factor in sglist merging's success. | |
2344 | * | |
6d49e352 | 2345 | * -- nyc |
1da177e4 | 2346 | */ |
085cc7d5 | 2347 | static inline void expand(struct zone *zone, struct page *page, |
6ab01363 | 2348 | int low, int high, int migratetype) |
1da177e4 LT |
2349 | { |
2350 | unsigned long size = 1 << high; | |
2351 | ||
2352 | while (high > low) { | |
1da177e4 LT |
2353 | high--; |
2354 | size >>= 1; | |
309381fe | 2355 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2356 | |
acbc15a4 JK |
2357 | /* |
2358 | * Mark as guard pages (or page), that will allow to | |
2359 | * merge back to allocator when buddy will be freed. | |
2360 | * Corresponding page table entries will not be touched, | |
2361 | * pages will stay not present in virtual address space | |
2362 | */ | |
2363 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2364 | continue; |
acbc15a4 | 2365 | |
6ab01363 | 2366 | add_to_free_list(&page[size], zone, high, migratetype); |
ab130f91 | 2367 | set_buddy_order(&page[size], high); |
1da177e4 | 2368 | } |
1da177e4 LT |
2369 | } |
2370 | ||
4e611801 | 2371 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2372 | { |
f4c18e6f | 2373 | if (unlikely(page->flags & __PG_HWPOISON)) { |
e570f56c NH |
2374 | /* Don't complain about hwpoisoned pages */ |
2375 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2376 | return; | |
f4c18e6f | 2377 | } |
58b7f119 WY |
2378 | |
2379 | bad_page(page, | |
2380 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
4e611801 VB |
2381 | } |
2382 | ||
2383 | /* | |
2384 | * This page is about to be returned from the page allocator | |
2385 | */ | |
2386 | static inline int check_new_page(struct page *page) | |
2387 | { | |
2388 | if (likely(page_expected_state(page, | |
2389 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2390 | return 0; | |
2391 | ||
2392 | check_new_page_bad(page); | |
2393 | return 1; | |
2a7684a2 WF |
2394 | } |
2395 | ||
77fe7f13 MG |
2396 | static bool check_new_pages(struct page *page, unsigned int order) |
2397 | { | |
2398 | int i; | |
2399 | for (i = 0; i < (1 << order); i++) { | |
2400 | struct page *p = page + i; | |
2401 | ||
2402 | if (unlikely(check_new_page(p))) | |
2403 | return true; | |
2404 | } | |
2405 | ||
2406 | return false; | |
2407 | } | |
2408 | ||
479f854a | 2409 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2410 | /* |
2411 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2412 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2413 | * also checked when pcp lists are refilled from the free lists. | |
2414 | */ | |
77fe7f13 | 2415 | static inline bool check_pcp_refill(struct page *page, unsigned int order) |
479f854a | 2416 | { |
8e57f8ac | 2417 | if (debug_pagealloc_enabled_static()) |
77fe7f13 | 2418 | return check_new_pages(page, order); |
4462b32c VB |
2419 | else |
2420 | return false; | |
479f854a MG |
2421 | } |
2422 | ||
77fe7f13 | 2423 | static inline bool check_new_pcp(struct page *page, unsigned int order) |
479f854a | 2424 | { |
77fe7f13 | 2425 | return check_new_pages(page, order); |
479f854a MG |
2426 | } |
2427 | #else | |
4462b32c VB |
2428 | /* |
2429 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2430 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2431 | * enabled, they are also checked when being allocated from the pcp lists. | |
2432 | */ | |
77fe7f13 | 2433 | static inline bool check_pcp_refill(struct page *page, unsigned int order) |
479f854a | 2434 | { |
77fe7f13 | 2435 | return check_new_pages(page, order); |
479f854a | 2436 | } |
77fe7f13 | 2437 | static inline bool check_new_pcp(struct page *page, unsigned int order) |
479f854a | 2438 | { |
8e57f8ac | 2439 | if (debug_pagealloc_enabled_static()) |
77fe7f13 | 2440 | return check_new_pages(page, order); |
4462b32c VB |
2441 | else |
2442 | return false; | |
479f854a MG |
2443 | } |
2444 | #endif /* CONFIG_DEBUG_VM */ | |
2445 | ||
6d05141a | 2446 | static inline bool should_skip_kasan_unpoison(gfp_t flags) |
53ae233c AK |
2447 | { |
2448 | /* Don't skip if a software KASAN mode is enabled. */ | |
2449 | if (IS_ENABLED(CONFIG_KASAN_GENERIC) || | |
2450 | IS_ENABLED(CONFIG_KASAN_SW_TAGS)) | |
2451 | return false; | |
2452 | ||
2453 | /* Skip, if hardware tag-based KASAN is not enabled. */ | |
2454 | if (!kasan_hw_tags_enabled()) | |
2455 | return true; | |
2456 | ||
2457 | /* | |
6d05141a CM |
2458 | * With hardware tag-based KASAN enabled, skip if this has been |
2459 | * requested via __GFP_SKIP_KASAN_UNPOISON. | |
53ae233c | 2460 | */ |
6d05141a | 2461 | return flags & __GFP_SKIP_KASAN_UNPOISON; |
53ae233c AK |
2462 | } |
2463 | ||
9353ffa6 AK |
2464 | static inline bool should_skip_init(gfp_t flags) |
2465 | { | |
2466 | /* Don't skip, if hardware tag-based KASAN is not enabled. */ | |
2467 | if (!kasan_hw_tags_enabled()) | |
2468 | return false; | |
2469 | ||
2470 | /* For hardware tag-based KASAN, skip if requested. */ | |
2471 | return (flags & __GFP_SKIP_ZERO); | |
2472 | } | |
2473 | ||
46f24fd8 JK |
2474 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2475 | gfp_t gfp_flags) | |
2476 | { | |
9353ffa6 AK |
2477 | bool init = !want_init_on_free() && want_init_on_alloc(gfp_flags) && |
2478 | !should_skip_init(gfp_flags); | |
b42090ae | 2479 | bool init_tags = init && (gfp_flags & __GFP_ZEROTAGS); |
70c248ac | 2480 | int i; |
b42090ae | 2481 | |
46f24fd8 JK |
2482 | set_page_private(page, 0); |
2483 | set_page_refcounted(page); | |
2484 | ||
2485 | arch_alloc_page(page, order); | |
77bc7fd6 | 2486 | debug_pagealloc_map_pages(page, 1 << order); |
1bb5eab3 AK |
2487 | |
2488 | /* | |
2489 | * Page unpoisoning must happen before memory initialization. | |
2490 | * Otherwise, the poison pattern will be overwritten for __GFP_ZERO | |
2491 | * allocations and the page unpoisoning code will complain. | |
2492 | */ | |
8db26a3d | 2493 | kernel_unpoison_pages(page, 1 << order); |
862b6dee | 2494 | |
1bb5eab3 AK |
2495 | /* |
2496 | * As memory initialization might be integrated into KASAN, | |
b42090ae | 2497 | * KASAN unpoisoning and memory initializion code must be |
1bb5eab3 AK |
2498 | * kept together to avoid discrepancies in behavior. |
2499 | */ | |
9294b128 AK |
2500 | |
2501 | /* | |
2502 | * If memory tags should be zeroed (which happens only when memory | |
2503 | * should be initialized as well). | |
2504 | */ | |
2505 | if (init_tags) { | |
9294b128 AK |
2506 | /* Initialize both memory and tags. */ |
2507 | for (i = 0; i != 1 << order; ++i) | |
2508 | tag_clear_highpage(page + i); | |
2509 | ||
2510 | /* Note that memory is already initialized by the loop above. */ | |
2511 | init = false; | |
2512 | } | |
6d05141a | 2513 | if (!should_skip_kasan_unpoison(gfp_flags)) { |
53ae233c | 2514 | /* Unpoison shadow memory or set memory tags. */ |
e9d0ca92 | 2515 | kasan_unpoison_pages(page, order, init); |
7e3cbba6 | 2516 | |
e9d0ca92 AK |
2517 | /* Note that memory is already initialized by KASAN. */ |
2518 | if (kasan_has_integrated_init()) | |
7e3cbba6 | 2519 | init = false; |
70c248ac CM |
2520 | } else { |
2521 | /* Ensure page_address() dereferencing does not fault. */ | |
2522 | for (i = 0; i != 1 << order; ++i) | |
2523 | page_kasan_tag_reset(page + i); | |
7a3b8353 | 2524 | } |
7e3cbba6 AK |
2525 | /* If memory is still not initialized, do it now. */ |
2526 | if (init) | |
aeaec8e2 | 2527 | kernel_init_pages(page, 1 << order); |
89b27116 AK |
2528 | /* Propagate __GFP_SKIP_KASAN_POISON to page flags. */ |
2529 | if (kasan_hw_tags_enabled() && (gfp_flags & __GFP_SKIP_KASAN_POISON)) | |
2530 | SetPageSkipKASanPoison(page); | |
1bb5eab3 AK |
2531 | |
2532 | set_page_owner(page, order, gfp_flags); | |
df4e817b | 2533 | page_table_check_alloc(page, order); |
46f24fd8 JK |
2534 | } |
2535 | ||
479f854a | 2536 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2537 | unsigned int alloc_flags) |
2a7684a2 | 2538 | { |
46f24fd8 | 2539 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2540 | |
17cf4406 NP |
2541 | if (order && (gfp_flags & __GFP_COMP)) |
2542 | prep_compound_page(page, order); | |
2543 | ||
75379191 | 2544 | /* |
2f064f34 | 2545 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2546 | * allocate the page. The expectation is that the caller is taking |
2547 | * steps that will free more memory. The caller should avoid the page | |
2548 | * being used for !PFMEMALLOC purposes. | |
2549 | */ | |
2f064f34 MH |
2550 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2551 | set_page_pfmemalloc(page); | |
2552 | else | |
2553 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2554 | } |
2555 | ||
56fd56b8 MG |
2556 | /* |
2557 | * Go through the free lists for the given migratetype and remove | |
2558 | * the smallest available page from the freelists | |
2559 | */ | |
85ccc8fa | 2560 | static __always_inline |
728ec980 | 2561 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2562 | int migratetype) |
2563 | { | |
2564 | unsigned int current_order; | |
b8af2941 | 2565 | struct free_area *area; |
56fd56b8 MG |
2566 | struct page *page; |
2567 | ||
2568 | /* Find a page of the appropriate size in the preferred list */ | |
2569 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2570 | area = &(zone->free_area[current_order]); | |
b03641af | 2571 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2572 | if (!page) |
2573 | continue; | |
6ab01363 AD |
2574 | del_page_from_free_list(page, zone, current_order); |
2575 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 2576 | set_pcppage_migratetype(page, migratetype); |
10e0f753 WY |
2577 | trace_mm_page_alloc_zone_locked(page, order, migratetype, |
2578 | pcp_allowed_order(order) && | |
2579 | migratetype < MIGRATE_PCPTYPES); | |
56fd56b8 MG |
2580 | return page; |
2581 | } | |
2582 | ||
2583 | return NULL; | |
2584 | } | |
2585 | ||
2586 | ||
b2a0ac88 MG |
2587 | /* |
2588 | * This array describes the order lists are fallen back to when | |
2589 | * the free lists for the desirable migrate type are depleted | |
1dd214b8 ZY |
2590 | * |
2591 | * The other migratetypes do not have fallbacks. | |
b2a0ac88 | 2592 | */ |
da415663 | 2593 | static int fallbacks[MIGRATE_TYPES][3] = { |
974a786e | 2594 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2595 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2596 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
b2a0ac88 MG |
2597 | }; |
2598 | ||
dc67647b | 2599 | #ifdef CONFIG_CMA |
85ccc8fa | 2600 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2601 | unsigned int order) |
2602 | { | |
2603 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2604 | } | |
2605 | #else | |
2606 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2607 | unsigned int order) { return NULL; } | |
2608 | #endif | |
2609 | ||
c361be55 | 2610 | /* |
293ffa5e | 2611 | * Move the free pages in a range to the freelist tail of the requested type. |
d9c23400 | 2612 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2613 | * boundary. If alignment is required, use move_freepages_block() |
2614 | */ | |
02aa0cdd | 2615 | static int move_freepages(struct zone *zone, |
39ddb991 | 2616 | unsigned long start_pfn, unsigned long end_pfn, |
02aa0cdd | 2617 | int migratetype, int *num_movable) |
c361be55 MG |
2618 | { |
2619 | struct page *page; | |
39ddb991 | 2620 | unsigned long pfn; |
d00181b9 | 2621 | unsigned int order; |
d100313f | 2622 | int pages_moved = 0; |
c361be55 | 2623 | |
39ddb991 | 2624 | for (pfn = start_pfn; pfn <= end_pfn;) { |
39ddb991 | 2625 | page = pfn_to_page(pfn); |
c361be55 | 2626 | if (!PageBuddy(page)) { |
02aa0cdd VB |
2627 | /* |
2628 | * We assume that pages that could be isolated for | |
2629 | * migration are movable. But we don't actually try | |
2630 | * isolating, as that would be expensive. | |
2631 | */ | |
2632 | if (num_movable && | |
2633 | (PageLRU(page) || __PageMovable(page))) | |
2634 | (*num_movable)++; | |
39ddb991 | 2635 | pfn++; |
c361be55 MG |
2636 | continue; |
2637 | } | |
2638 | ||
cd961038 DR |
2639 | /* Make sure we are not inadvertently changing nodes */ |
2640 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2641 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2642 | ||
ab130f91 | 2643 | order = buddy_order(page); |
6ab01363 | 2644 | move_to_free_list(page, zone, order, migratetype); |
39ddb991 | 2645 | pfn += 1 << order; |
d100313f | 2646 | pages_moved += 1 << order; |
c361be55 MG |
2647 | } |
2648 | ||
d100313f | 2649 | return pages_moved; |
c361be55 MG |
2650 | } |
2651 | ||
ee6f509c | 2652 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2653 | int migratetype, int *num_movable) |
c361be55 | 2654 | { |
39ddb991 | 2655 | unsigned long start_pfn, end_pfn, pfn; |
c361be55 | 2656 | |
4a222127 DR |
2657 | if (num_movable) |
2658 | *num_movable = 0; | |
2659 | ||
39ddb991 | 2660 | pfn = page_to_pfn(page); |
4f9bc69a KW |
2661 | start_pfn = pageblock_start_pfn(pfn); |
2662 | end_pfn = pageblock_end_pfn(pfn) - 1; | |
c361be55 MG |
2663 | |
2664 | /* Do not cross zone boundaries */ | |
108bcc96 | 2665 | if (!zone_spans_pfn(zone, start_pfn)) |
39ddb991 | 2666 | start_pfn = pfn; |
108bcc96 | 2667 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2668 | return 0; |
2669 | ||
39ddb991 | 2670 | return move_freepages(zone, start_pfn, end_pfn, migratetype, |
02aa0cdd | 2671 | num_movable); |
c361be55 MG |
2672 | } |
2673 | ||
2f66a68f MG |
2674 | static void change_pageblock_range(struct page *pageblock_page, |
2675 | int start_order, int migratetype) | |
2676 | { | |
2677 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2678 | ||
2679 | while (nr_pageblocks--) { | |
2680 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2681 | pageblock_page += pageblock_nr_pages; | |
2682 | } | |
2683 | } | |
2684 | ||
fef903ef | 2685 | /* |
9c0415eb VB |
2686 | * When we are falling back to another migratetype during allocation, try to |
2687 | * steal extra free pages from the same pageblocks to satisfy further | |
2688 | * allocations, instead of polluting multiple pageblocks. | |
2689 | * | |
2690 | * If we are stealing a relatively large buddy page, it is likely there will | |
2691 | * be more free pages in the pageblock, so try to steal them all. For | |
2692 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2693 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2694 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2695 | * pageblocks. | |
fef903ef | 2696 | */ |
4eb7dce6 JK |
2697 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2698 | { | |
2699 | /* | |
2700 | * Leaving this order check is intended, although there is | |
2701 | * relaxed order check in next check. The reason is that | |
2702 | * we can actually steal whole pageblock if this condition met, | |
2703 | * but, below check doesn't guarantee it and that is just heuristic | |
2704 | * so could be changed anytime. | |
2705 | */ | |
2706 | if (order >= pageblock_order) | |
2707 | return true; | |
2708 | ||
2709 | if (order >= pageblock_order / 2 || | |
2710 | start_mt == MIGRATE_RECLAIMABLE || | |
2711 | start_mt == MIGRATE_UNMOVABLE || | |
2712 | page_group_by_mobility_disabled) | |
2713 | return true; | |
2714 | ||
2715 | return false; | |
2716 | } | |
2717 | ||
597c8920 | 2718 | static inline bool boost_watermark(struct zone *zone) |
1c30844d MG |
2719 | { |
2720 | unsigned long max_boost; | |
2721 | ||
2722 | if (!watermark_boost_factor) | |
597c8920 | 2723 | return false; |
14f69140 HW |
2724 | /* |
2725 | * Don't bother in zones that are unlikely to produce results. | |
2726 | * On small machines, including kdump capture kernels running | |
2727 | * in a small area, boosting the watermark can cause an out of | |
2728 | * memory situation immediately. | |
2729 | */ | |
2730 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
597c8920 | 2731 | return false; |
1c30844d MG |
2732 | |
2733 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2734 | watermark_boost_factor, 10000); | |
94b3334c MG |
2735 | |
2736 | /* | |
2737 | * high watermark may be uninitialised if fragmentation occurs | |
2738 | * very early in boot so do not boost. We do not fall | |
2739 | * through and boost by pageblock_nr_pages as failing | |
2740 | * allocations that early means that reclaim is not going | |
2741 | * to help and it may even be impossible to reclaim the | |
2742 | * boosted watermark resulting in a hang. | |
2743 | */ | |
2744 | if (!max_boost) | |
597c8920 | 2745 | return false; |
94b3334c | 2746 | |
1c30844d MG |
2747 | max_boost = max(pageblock_nr_pages, max_boost); |
2748 | ||
2749 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2750 | max_boost); | |
597c8920 JW |
2751 | |
2752 | return true; | |
1c30844d MG |
2753 | } |
2754 | ||
4eb7dce6 JK |
2755 | /* |
2756 | * This function implements actual steal behaviour. If order is large enough, | |
2757 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2758 | * pageblock to our migratetype and determine how many already-allocated pages |
2759 | * are there in the pageblock with a compatible migratetype. If at least half | |
2760 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2761 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2762 | */ |
2763 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2764 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2765 | { |
ab130f91 | 2766 | unsigned int current_order = buddy_order(page); |
02aa0cdd VB |
2767 | int free_pages, movable_pages, alike_pages; |
2768 | int old_block_type; | |
2769 | ||
2770 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2771 | |
3bc48f96 VB |
2772 | /* |
2773 | * This can happen due to races and we want to prevent broken | |
2774 | * highatomic accounting. | |
2775 | */ | |
02aa0cdd | 2776 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2777 | goto single_page; |
2778 | ||
fef903ef SB |
2779 | /* Take ownership for orders >= pageblock_order */ |
2780 | if (current_order >= pageblock_order) { | |
2781 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2782 | goto single_page; |
fef903ef SB |
2783 | } |
2784 | ||
1c30844d MG |
2785 | /* |
2786 | * Boost watermarks to increase reclaim pressure to reduce the | |
2787 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2788 | * may be balanced overall and kswapd will not wake naturally. | |
2789 | */ | |
597c8920 | 2790 | if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD)) |
73444bc4 | 2791 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2792 | |
3bc48f96 VB |
2793 | /* We are not allowed to try stealing from the whole block */ |
2794 | if (!whole_block) | |
2795 | goto single_page; | |
2796 | ||
02aa0cdd VB |
2797 | free_pages = move_freepages_block(zone, page, start_type, |
2798 | &movable_pages); | |
2799 | /* | |
2800 | * Determine how many pages are compatible with our allocation. | |
2801 | * For movable allocation, it's the number of movable pages which | |
2802 | * we just obtained. For other types it's a bit more tricky. | |
2803 | */ | |
2804 | if (start_type == MIGRATE_MOVABLE) { | |
2805 | alike_pages = movable_pages; | |
2806 | } else { | |
2807 | /* | |
2808 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2809 | * to MOVABLE pageblock, consider all non-movable pages as | |
2810 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2811 | * vice versa, be conservative since we can't distinguish the | |
2812 | * exact migratetype of non-movable pages. | |
2813 | */ | |
2814 | if (old_block_type == MIGRATE_MOVABLE) | |
2815 | alike_pages = pageblock_nr_pages | |
2816 | - (free_pages + movable_pages); | |
2817 | else | |
2818 | alike_pages = 0; | |
2819 | } | |
2820 | ||
3bc48f96 | 2821 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2822 | if (!free_pages) |
3bc48f96 | 2823 | goto single_page; |
fef903ef | 2824 | |
02aa0cdd VB |
2825 | /* |
2826 | * If a sufficient number of pages in the block are either free or of | |
2827 | * comparable migratability as our allocation, claim the whole block. | |
2828 | */ | |
2829 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2830 | page_group_by_mobility_disabled) |
2831 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2832 | |
2833 | return; | |
2834 | ||
2835 | single_page: | |
6ab01363 | 2836 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
2837 | } |
2838 | ||
2149cdae JK |
2839 | /* |
2840 | * Check whether there is a suitable fallback freepage with requested order. | |
2841 | * If only_stealable is true, this function returns fallback_mt only if | |
2842 | * we can steal other freepages all together. This would help to reduce | |
2843 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2844 | */ | |
2845 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2846 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2847 | { |
2848 | int i; | |
2849 | int fallback_mt; | |
2850 | ||
2851 | if (area->nr_free == 0) | |
2852 | return -1; | |
2853 | ||
2854 | *can_steal = false; | |
2855 | for (i = 0;; i++) { | |
2856 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2857 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2858 | break; |
2859 | ||
b03641af | 2860 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2861 | continue; |
fef903ef | 2862 | |
4eb7dce6 JK |
2863 | if (can_steal_fallback(order, migratetype)) |
2864 | *can_steal = true; | |
2865 | ||
2149cdae JK |
2866 | if (!only_stealable) |
2867 | return fallback_mt; | |
2868 | ||
2869 | if (*can_steal) | |
2870 | return fallback_mt; | |
fef903ef | 2871 | } |
4eb7dce6 JK |
2872 | |
2873 | return -1; | |
fef903ef SB |
2874 | } |
2875 | ||
0aaa29a5 MG |
2876 | /* |
2877 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2878 | * there are no empty page blocks that contain a page with a suitable order | |
2879 | */ | |
2880 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2881 | unsigned int alloc_order) | |
2882 | { | |
2883 | int mt; | |
2884 | unsigned long max_managed, flags; | |
2885 | ||
2886 | /* | |
2887 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2888 | * Check is race-prone but harmless. | |
2889 | */ | |
9705bea5 | 2890 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2891 | if (zone->nr_reserved_highatomic >= max_managed) |
2892 | return; | |
2893 | ||
2894 | spin_lock_irqsave(&zone->lock, flags); | |
2895 | ||
2896 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2897 | if (zone->nr_reserved_highatomic >= max_managed) | |
2898 | goto out_unlock; | |
2899 | ||
2900 | /* Yoink! */ | |
2901 | mt = get_pageblock_migratetype(page); | |
1dd214b8 ZY |
2902 | /* Only reserve normal pageblocks (i.e., they can merge with others) */ |
2903 | if (migratetype_is_mergeable(mt)) { | |
0aaa29a5 MG |
2904 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2905 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2906 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2907 | } |
2908 | ||
2909 | out_unlock: | |
2910 | spin_unlock_irqrestore(&zone->lock, flags); | |
2911 | } | |
2912 | ||
2913 | /* | |
2914 | * Used when an allocation is about to fail under memory pressure. This | |
2915 | * potentially hurts the reliability of high-order allocations when under | |
2916 | * intense memory pressure but failed atomic allocations should be easier | |
2917 | * to recover from than an OOM. | |
29fac03b MK |
2918 | * |
2919 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2920 | * pageblock is exhausted. | |
0aaa29a5 | 2921 | */ |
29fac03b MK |
2922 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2923 | bool force) | |
0aaa29a5 MG |
2924 | { |
2925 | struct zonelist *zonelist = ac->zonelist; | |
2926 | unsigned long flags; | |
2927 | struct zoneref *z; | |
2928 | struct zone *zone; | |
2929 | struct page *page; | |
2930 | int order; | |
04c8716f | 2931 | bool ret; |
0aaa29a5 | 2932 | |
97a225e6 | 2933 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2934 | ac->nodemask) { |
29fac03b MK |
2935 | /* |
2936 | * Preserve at least one pageblock unless memory pressure | |
2937 | * is really high. | |
2938 | */ | |
2939 | if (!force && zone->nr_reserved_highatomic <= | |
2940 | pageblock_nr_pages) | |
0aaa29a5 MG |
2941 | continue; |
2942 | ||
2943 | spin_lock_irqsave(&zone->lock, flags); | |
2944 | for (order = 0; order < MAX_ORDER; order++) { | |
2945 | struct free_area *area = &(zone->free_area[order]); | |
2946 | ||
b03641af | 2947 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2948 | if (!page) |
0aaa29a5 MG |
2949 | continue; |
2950 | ||
0aaa29a5 | 2951 | /* |
4855e4a7 MK |
2952 | * In page freeing path, migratetype change is racy so |
2953 | * we can counter several free pages in a pageblock | |
f0953a1b | 2954 | * in this loop although we changed the pageblock type |
4855e4a7 MK |
2955 | * from highatomic to ac->migratetype. So we should |
2956 | * adjust the count once. | |
0aaa29a5 | 2957 | */ |
a6ffdc07 | 2958 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2959 | /* |
2960 | * It should never happen but changes to | |
2961 | * locking could inadvertently allow a per-cpu | |
2962 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2963 | * while unreserving so be safe and watch for | |
2964 | * underflows. | |
2965 | */ | |
2966 | zone->nr_reserved_highatomic -= min( | |
2967 | pageblock_nr_pages, | |
2968 | zone->nr_reserved_highatomic); | |
2969 | } | |
0aaa29a5 MG |
2970 | |
2971 | /* | |
2972 | * Convert to ac->migratetype and avoid the normal | |
2973 | * pageblock stealing heuristics. Minimally, the caller | |
2974 | * is doing the work and needs the pages. More | |
2975 | * importantly, if the block was always converted to | |
2976 | * MIGRATE_UNMOVABLE or another type then the number | |
2977 | * of pageblocks that cannot be completely freed | |
2978 | * may increase. | |
2979 | */ | |
2980 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2981 | ret = move_freepages_block(zone, page, ac->migratetype, |
2982 | NULL); | |
29fac03b MK |
2983 | if (ret) { |
2984 | spin_unlock_irqrestore(&zone->lock, flags); | |
2985 | return ret; | |
2986 | } | |
0aaa29a5 MG |
2987 | } |
2988 | spin_unlock_irqrestore(&zone->lock, flags); | |
2989 | } | |
04c8716f MK |
2990 | |
2991 | return false; | |
0aaa29a5 MG |
2992 | } |
2993 | ||
3bc48f96 VB |
2994 | /* |
2995 | * Try finding a free buddy page on the fallback list and put it on the free | |
2996 | * list of requested migratetype, possibly along with other pages from the same | |
2997 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2998 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2999 | * |
3000 | * The use of signed ints for order and current_order is a deliberate | |
3001 | * deviation from the rest of this file, to make the for loop | |
3002 | * condition simpler. | |
3bc48f96 | 3003 | */ |
85ccc8fa | 3004 | static __always_inline bool |
6bb15450 MG |
3005 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
3006 | unsigned int alloc_flags) | |
b2a0ac88 | 3007 | { |
b8af2941 | 3008 | struct free_area *area; |
b002529d | 3009 | int current_order; |
6bb15450 | 3010 | int min_order = order; |
b2a0ac88 | 3011 | struct page *page; |
4eb7dce6 JK |
3012 | int fallback_mt; |
3013 | bool can_steal; | |
b2a0ac88 | 3014 | |
6bb15450 MG |
3015 | /* |
3016 | * Do not steal pages from freelists belonging to other pageblocks | |
3017 | * i.e. orders < pageblock_order. If there are no local zones free, | |
3018 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
3019 | */ | |
e933dc4a | 3020 | if (order < pageblock_order && alloc_flags & ALLOC_NOFRAGMENT) |
6bb15450 MG |
3021 | min_order = pageblock_order; |
3022 | ||
7a8f58f3 VB |
3023 | /* |
3024 | * Find the largest available free page in the other list. This roughly | |
3025 | * approximates finding the pageblock with the most free pages, which | |
3026 | * would be too costly to do exactly. | |
3027 | */ | |
6bb15450 | 3028 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 3029 | --current_order) { |
4eb7dce6 JK |
3030 | area = &(zone->free_area[current_order]); |
3031 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 3032 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
3033 | if (fallback_mt == -1) |
3034 | continue; | |
b2a0ac88 | 3035 | |
7a8f58f3 VB |
3036 | /* |
3037 | * We cannot steal all free pages from the pageblock and the | |
3038 | * requested migratetype is movable. In that case it's better to | |
3039 | * steal and split the smallest available page instead of the | |
3040 | * largest available page, because even if the next movable | |
3041 | * allocation falls back into a different pageblock than this | |
3042 | * one, it won't cause permanent fragmentation. | |
3043 | */ | |
3044 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
3045 | && current_order > order) | |
3046 | goto find_smallest; | |
b2a0ac88 | 3047 | |
7a8f58f3 VB |
3048 | goto do_steal; |
3049 | } | |
e0fff1bd | 3050 | |
7a8f58f3 | 3051 | return false; |
e0fff1bd | 3052 | |
7a8f58f3 VB |
3053 | find_smallest: |
3054 | for (current_order = order; current_order < MAX_ORDER; | |
3055 | current_order++) { | |
3056 | area = &(zone->free_area[current_order]); | |
3057 | fallback_mt = find_suitable_fallback(area, current_order, | |
3058 | start_migratetype, false, &can_steal); | |
3059 | if (fallback_mt != -1) | |
3060 | break; | |
b2a0ac88 MG |
3061 | } |
3062 | ||
7a8f58f3 VB |
3063 | /* |
3064 | * This should not happen - we already found a suitable fallback | |
3065 | * when looking for the largest page. | |
3066 | */ | |
3067 | VM_BUG_ON(current_order == MAX_ORDER); | |
3068 | ||
3069 | do_steal: | |
b03641af | 3070 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 3071 | |
1c30844d MG |
3072 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
3073 | can_steal); | |
7a8f58f3 VB |
3074 | |
3075 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
3076 | start_migratetype, fallback_mt); | |
3077 | ||
3078 | return true; | |
3079 | ||
b2a0ac88 MG |
3080 | } |
3081 | ||
56fd56b8 | 3082 | /* |
1da177e4 LT |
3083 | * Do the hard work of removing an element from the buddy allocator. |
3084 | * Call me with the zone->lock already held. | |
3085 | */ | |
85ccc8fa | 3086 | static __always_inline struct page * |
6bb15450 MG |
3087 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
3088 | unsigned int alloc_flags) | |
1da177e4 | 3089 | { |
1da177e4 LT |
3090 | struct page *page; |
3091 | ||
ce8f86ee H |
3092 | if (IS_ENABLED(CONFIG_CMA)) { |
3093 | /* | |
3094 | * Balance movable allocations between regular and CMA areas by | |
3095 | * allocating from CMA when over half of the zone's free memory | |
3096 | * is in the CMA area. | |
3097 | */ | |
3098 | if (alloc_flags & ALLOC_CMA && | |
3099 | zone_page_state(zone, NR_FREE_CMA_PAGES) > | |
3100 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
3101 | page = __rmqueue_cma_fallback(zone, order); | |
3102 | if (page) | |
10e0f753 | 3103 | return page; |
ce8f86ee | 3104 | } |
16867664 | 3105 | } |
3bc48f96 | 3106 | retry: |
56fd56b8 | 3107 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 3108 | if (unlikely(!page)) { |
8510e69c | 3109 | if (alloc_flags & ALLOC_CMA) |
dc67647b JK |
3110 | page = __rmqueue_cma_fallback(zone, order); |
3111 | ||
6bb15450 MG |
3112 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
3113 | alloc_flags)) | |
3bc48f96 | 3114 | goto retry; |
728ec980 | 3115 | } |
b2a0ac88 | 3116 | return page; |
1da177e4 LT |
3117 | } |
3118 | ||
5f63b720 | 3119 | /* |
1da177e4 LT |
3120 | * Obtain a specified number of elements from the buddy allocator, all under |
3121 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
3122 | * Returns the number of new pages which were placed at *list. | |
3123 | */ | |
5f63b720 | 3124 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 3125 | unsigned long count, struct list_head *list, |
6bb15450 | 3126 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 3127 | { |
cb66bede | 3128 | int i, allocated = 0; |
5f63b720 | 3129 | |
01b44456 | 3130 | /* Caller must hold IRQ-safe pcp->lock so IRQs are disabled. */ |
d34b0733 | 3131 | spin_lock(&zone->lock); |
1da177e4 | 3132 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
3133 | struct page *page = __rmqueue(zone, order, migratetype, |
3134 | alloc_flags); | |
085cc7d5 | 3135 | if (unlikely(page == NULL)) |
1da177e4 | 3136 | break; |
81eabcbe | 3137 | |
77fe7f13 | 3138 | if (unlikely(check_pcp_refill(page, order))) |
479f854a MG |
3139 | continue; |
3140 | ||
81eabcbe | 3141 | /* |
0fac3ba5 VB |
3142 | * Split buddy pages returned by expand() are received here in |
3143 | * physical page order. The page is added to the tail of | |
3144 | * caller's list. From the callers perspective, the linked list | |
3145 | * is ordered by page number under some conditions. This is | |
3146 | * useful for IO devices that can forward direction from the | |
3147 | * head, thus also in the physical page order. This is useful | |
3148 | * for IO devices that can merge IO requests if the physical | |
3149 | * pages are ordered properly. | |
81eabcbe | 3150 | */ |
bf75f200 | 3151 | list_add_tail(&page->pcp_list, list); |
cb66bede | 3152 | allocated++; |
bb14c2c7 | 3153 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
3154 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
3155 | -(1 << order)); | |
1da177e4 | 3156 | } |
a6de734b MG |
3157 | |
3158 | /* | |
3159 | * i pages were removed from the buddy list even if some leak due | |
3160 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
cb66bede | 3161 | * on i. Do not confuse with 'allocated' which is the number of |
a6de734b MG |
3162 | * pages added to the pcp list. |
3163 | */ | |
f2260e6b | 3164 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 3165 | spin_unlock(&zone->lock); |
cb66bede | 3166 | return allocated; |
1da177e4 LT |
3167 | } |
3168 | ||
4ae7c039 | 3169 | #ifdef CONFIG_NUMA |
8fce4d8e | 3170 | /* |
4037d452 CL |
3171 | * Called from the vmstat counter updater to drain pagesets of this |
3172 | * currently executing processor on remote nodes after they have | |
3173 | * expired. | |
8fce4d8e | 3174 | */ |
4037d452 | 3175 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 3176 | { |
7be12fc9 | 3177 | int to_drain, batch; |
4ae7c039 | 3178 | |
4db0c3c2 | 3179 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 3180 | to_drain = min(pcp->count, batch); |
4b23a68f MG |
3181 | if (to_drain > 0) { |
3182 | unsigned long flags; | |
3183 | ||
3184 | /* | |
3185 | * free_pcppages_bulk expects IRQs disabled for zone->lock | |
3186 | * so even though pcp->lock is not intended to be IRQ-safe, | |
3187 | * it's needed in this context. | |
3188 | */ | |
3189 | spin_lock_irqsave(&pcp->lock, flags); | |
fd56eef2 | 3190 | free_pcppages_bulk(zone, to_drain, pcp, 0); |
4b23a68f MG |
3191 | spin_unlock_irqrestore(&pcp->lock, flags); |
3192 | } | |
4ae7c039 CL |
3193 | } |
3194 | #endif | |
3195 | ||
9f8f2172 | 3196 | /* |
93481ff0 | 3197 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 | 3198 | */ |
93481ff0 | 3199 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 3200 | { |
93481ff0 | 3201 | struct per_cpu_pages *pcp; |
1da177e4 | 3202 | |
28f836b6 | 3203 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
4b23a68f MG |
3204 | if (pcp->count) { |
3205 | unsigned long flags; | |
28f836b6 | 3206 | |
4b23a68f MG |
3207 | /* See drain_zone_pages on why this is disabling IRQs */ |
3208 | spin_lock_irqsave(&pcp->lock, flags); | |
3209 | free_pcppages_bulk(zone, pcp->count, pcp, 0); | |
3210 | spin_unlock_irqrestore(&pcp->lock, flags); | |
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 MG |
3475 | { |
3476 | unsigned long flags; | |
4b23a68f MG |
3477 | unsigned long __maybe_unused UP_flags; |
3478 | struct per_cpu_pages *pcp; | |
3479 | struct zone *zone; | |
9cca35d4 | 3480 | unsigned long pfn = page_to_pfn(page); |
df1acc85 | 3481 | int migratetype; |
9cca35d4 | 3482 | |
44042b44 | 3483 | if (!free_unref_page_prepare(page, pfn, order)) |
9cca35d4 | 3484 | return; |
da456f14 | 3485 | |
5f8dcc21 MG |
3486 | /* |
3487 | * We only track unmovable, reclaimable and movable on pcp lists. | |
df1acc85 | 3488 | * Place ISOLATE pages on the isolated list because they are being |
a6ffdc07 | 3489 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
3490 | * areas back if necessary. Otherwise, we may have to free |
3491 | * excessively into the page allocator | |
3492 | */ | |
df1acc85 MG |
3493 | migratetype = get_pcppage_migratetype(page); |
3494 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) { | |
194159fb | 3495 | if (unlikely(is_migrate_isolate(migratetype))) { |
44042b44 | 3496 | free_one_page(page_zone(page), page, pfn, order, migratetype, FPI_NONE); |
9cca35d4 | 3497 | return; |
5f8dcc21 MG |
3498 | } |
3499 | migratetype = MIGRATE_MOVABLE; | |
3500 | } | |
3501 | ||
4b23a68f MG |
3502 | zone = page_zone(page); |
3503 | pcp_trylock_prepare(UP_flags); | |
01b44456 MG |
3504 | pcp = pcp_spin_trylock_irqsave(zone->per_cpu_pageset, flags); |
3505 | if (pcp) { | |
4b23a68f | 3506 | free_unref_page_commit(zone, pcp, page, migratetype, order); |
01b44456 | 3507 | pcp_spin_unlock_irqrestore(pcp, flags); |
4b23a68f MG |
3508 | } else { |
3509 | free_one_page(zone, page, pfn, order, migratetype, FPI_NONE); | |
3510 | } | |
3511 | pcp_trylock_finish(UP_flags); | |
1da177e4 LT |
3512 | } |
3513 | ||
cc59850e KK |
3514 | /* |
3515 | * Free a list of 0-order pages | |
3516 | */ | |
2d4894b5 | 3517 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
3518 | { |
3519 | struct page *page, *next; | |
4b23a68f MG |
3520 | struct per_cpu_pages *pcp = NULL; |
3521 | struct zone *locked_zone = NULL; | |
56651377 | 3522 | unsigned long flags; |
c24ad77d | 3523 | int batch_count = 0; |
df1acc85 | 3524 | int migratetype; |
9cca35d4 MG |
3525 | |
3526 | /* Prepare pages for freeing */ | |
3527 | list_for_each_entry_safe(page, next, list, lru) { | |
56651377 | 3528 | unsigned long pfn = page_to_pfn(page); |
053cfda1 | 3529 | if (!free_unref_page_prepare(page, pfn, 0)) { |
9cca35d4 | 3530 | list_del(&page->lru); |
053cfda1 ML |
3531 | continue; |
3532 | } | |
df1acc85 MG |
3533 | |
3534 | /* | |
3535 | * Free isolated pages directly to the allocator, see | |
3536 | * comment in free_unref_page. | |
3537 | */ | |
3538 | migratetype = get_pcppage_migratetype(page); | |
47aef601 DB |
3539 | if (unlikely(is_migrate_isolate(migratetype))) { |
3540 | list_del(&page->lru); | |
3541 | free_one_page(page_zone(page), page, pfn, 0, migratetype, FPI_NONE); | |
3542 | continue; | |
df1acc85 | 3543 | } |
9cca35d4 | 3544 | } |
cc59850e KK |
3545 | |
3546 | list_for_each_entry_safe(page, next, list, lru) { | |
4b23a68f MG |
3547 | struct zone *zone = page_zone(page); |
3548 | ||
3549 | /* Different zone, different pcp lock. */ | |
3550 | if (zone != locked_zone) { | |
3551 | if (pcp) | |
01b44456 MG |
3552 | pcp_spin_unlock_irqrestore(pcp, flags); |
3553 | ||
4b23a68f | 3554 | locked_zone = zone; |
01b44456 | 3555 | pcp = pcp_spin_lock_irqsave(locked_zone->per_cpu_pageset, flags); |
4b23a68f MG |
3556 | } |
3557 | ||
47aef601 DB |
3558 | /* |
3559 | * Non-isolated types over MIGRATE_PCPTYPES get added | |
3560 | * to the MIGRATE_MOVABLE pcp list. | |
3561 | */ | |
df1acc85 | 3562 | migratetype = get_pcppage_migratetype(page); |
47aef601 DB |
3563 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) |
3564 | migratetype = MIGRATE_MOVABLE; | |
3565 | ||
2d4894b5 | 3566 | trace_mm_page_free_batched(page); |
4b23a68f | 3567 | free_unref_page_commit(zone, pcp, page, migratetype, 0); |
c24ad77d LS |
3568 | |
3569 | /* | |
3570 | * Guard against excessive IRQ disabled times when we get | |
3571 | * a large list of pages to free. | |
3572 | */ | |
3573 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
01b44456 | 3574 | pcp_spin_unlock_irqrestore(pcp, flags); |
c24ad77d | 3575 | batch_count = 0; |
01b44456 | 3576 | pcp = pcp_spin_lock_irqsave(locked_zone->per_cpu_pageset, flags); |
c24ad77d | 3577 | } |
cc59850e | 3578 | } |
4b23a68f MG |
3579 | |
3580 | if (pcp) | |
01b44456 | 3581 | pcp_spin_unlock_irqrestore(pcp, flags); |
cc59850e KK |
3582 | } |
3583 | ||
8dfcc9ba NP |
3584 | /* |
3585 | * split_page takes a non-compound higher-order page, and splits it into | |
3586 | * n (1<<order) sub-pages: page[0..n] | |
3587 | * Each sub-page must be freed individually. | |
3588 | * | |
3589 | * Note: this is probably too low level an operation for use in drivers. | |
3590 | * Please consult with lkml before using this in your driver. | |
3591 | */ | |
3592 | void split_page(struct page *page, unsigned int order) | |
3593 | { | |
3594 | int i; | |
3595 | ||
309381fe SL |
3596 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3597 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3598 | |
a9627bc5 | 3599 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3600 | set_page_refcounted(page + i); |
8fb156c9 | 3601 | split_page_owner(page, 1 << order); |
e1baddf8 | 3602 | split_page_memcg(page, 1 << order); |
8dfcc9ba | 3603 | } |
5853ff23 | 3604 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3605 | |
3c605096 | 3606 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3607 | { |
9a157dd8 KW |
3608 | struct zone *zone = page_zone(page); |
3609 | int mt = get_pageblock_migratetype(page); | |
748446bb | 3610 | |
194159fb | 3611 | if (!is_migrate_isolate(mt)) { |
9a157dd8 | 3612 | unsigned long watermark; |
8348faf9 VB |
3613 | /* |
3614 | * Obey watermarks as if the page was being allocated. We can | |
3615 | * emulate a high-order watermark check with a raised order-0 | |
3616 | * watermark, because we already know our high-order page | |
3617 | * exists. | |
3618 | */ | |
fd1444b2 | 3619 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3620 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3621 | return 0; |
3622 | ||
8fb74b9f | 3623 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3624 | } |
748446bb | 3625 | |
6ab01363 | 3626 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 3627 | |
400bc7fd | 3628 | /* |
3629 | * Set the pageblock if the isolated page is at least half of a | |
3630 | * pageblock | |
3631 | */ | |
748446bb MG |
3632 | if (order >= pageblock_order - 1) { |
3633 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3634 | for (; page < endpage; page += pageblock_nr_pages) { |
3635 | int mt = get_pageblock_migratetype(page); | |
1dd214b8 ZY |
3636 | /* |
3637 | * Only change normal pageblocks (i.e., they can merge | |
3638 | * with others) | |
3639 | */ | |
3640 | if (migratetype_is_mergeable(mt)) | |
47118af0 MN |
3641 | set_pageblock_migratetype(page, |
3642 | MIGRATE_MOVABLE); | |
3643 | } | |
748446bb MG |
3644 | } |
3645 | ||
8fb74b9f | 3646 | return 1UL << order; |
1fb3f8ca MG |
3647 | } |
3648 | ||
624f58d8 AD |
3649 | /** |
3650 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
3651 | * @page: Page that was isolated | |
3652 | * @order: Order of the isolated page | |
e6a0a7ad | 3653 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
3654 | * |
3655 | * This function is meant to return a page pulled from the free lists via | |
3656 | * __isolate_free_page back to the free lists they were pulled from. | |
3657 | */ | |
3658 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
3659 | { | |
3660 | struct zone *zone = page_zone(page); | |
3661 | ||
3662 | /* zone lock should be held when this function is called */ | |
3663 | lockdep_assert_held(&zone->lock); | |
3664 | ||
3665 | /* Return isolated page to tail of freelist. */ | |
f04a5d5d | 3666 | __free_one_page(page, page_to_pfn(page), zone, order, mt, |
47b6a24a | 3667 | FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL); |
624f58d8 AD |
3668 | } |
3669 | ||
060e7417 MG |
3670 | /* |
3671 | * Update NUMA hit/miss statistics | |
060e7417 | 3672 | */ |
3e23060b MG |
3673 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, |
3674 | long nr_account) | |
060e7417 MG |
3675 | { |
3676 | #ifdef CONFIG_NUMA | |
3a321d2a | 3677 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3678 | |
4518085e KW |
3679 | /* skip numa counters update if numa stats is disabled */ |
3680 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3681 | return; | |
3682 | ||
c1093b74 | 3683 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3684 | local_stat = NUMA_OTHER; |
060e7417 | 3685 | |
c1093b74 | 3686 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3e23060b | 3687 | __count_numa_events(z, NUMA_HIT, nr_account); |
2df26639 | 3688 | else { |
3e23060b MG |
3689 | __count_numa_events(z, NUMA_MISS, nr_account); |
3690 | __count_numa_events(preferred_zone, NUMA_FOREIGN, nr_account); | |
060e7417 | 3691 | } |
3e23060b | 3692 | __count_numa_events(z, local_stat, nr_account); |
060e7417 MG |
3693 | #endif |
3694 | } | |
3695 | ||
589d9973 MG |
3696 | static __always_inline |
3697 | struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, | |
3698 | unsigned int order, unsigned int alloc_flags, | |
3699 | int migratetype) | |
3700 | { | |
3701 | struct page *page; | |
3702 | unsigned long flags; | |
3703 | ||
3704 | do { | |
3705 | page = NULL; | |
3706 | spin_lock_irqsave(&zone->lock, flags); | |
3707 | /* | |
3708 | * order-0 request can reach here when the pcplist is skipped | |
3709 | * due to non-CMA allocation context. HIGHATOMIC area is | |
3710 | * reserved for high-order atomic allocation, so order-0 | |
3711 | * request should skip it. | |
3712 | */ | |
3713 | if (order > 0 && alloc_flags & ALLOC_HARDER) | |
3714 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3715 | if (!page) { | |
3716 | page = __rmqueue(zone, order, migratetype, alloc_flags); | |
3717 | if (!page) { | |
3718 | spin_unlock_irqrestore(&zone->lock, flags); | |
3719 | return NULL; | |
3720 | } | |
3721 | } | |
3722 | __mod_zone_freepage_state(zone, -(1 << order), | |
3723 | get_pcppage_migratetype(page)); | |
3724 | spin_unlock_irqrestore(&zone->lock, flags); | |
3725 | } while (check_new_pages(page, order)); | |
3726 | ||
3727 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
3728 | zone_statistics(preferred_zone, zone, 1); | |
3729 | ||
3730 | return page; | |
3731 | } | |
3732 | ||
066b2393 | 3733 | /* Remove page from the per-cpu list, caller must protect the list */ |
3b822017 | 3734 | static inline |
44042b44 MG |
3735 | struct page *__rmqueue_pcplist(struct zone *zone, unsigned int order, |
3736 | int migratetype, | |
6bb15450 | 3737 | unsigned int alloc_flags, |
453f85d4 | 3738 | struct per_cpu_pages *pcp, |
066b2393 MG |
3739 | struct list_head *list) |
3740 | { | |
3741 | struct page *page; | |
3742 | ||
3743 | do { | |
3744 | if (list_empty(list)) { | |
44042b44 MG |
3745 | int batch = READ_ONCE(pcp->batch); |
3746 | int alloced; | |
3747 | ||
3748 | /* | |
3749 | * Scale batch relative to order if batch implies | |
3750 | * free pages can be stored on the PCP. Batch can | |
3751 | * be 1 for small zones or for boot pagesets which | |
3752 | * should never store free pages as the pages may | |
3753 | * belong to arbitrary zones. | |
3754 | */ | |
3755 | if (batch > 1) | |
3756 | batch = max(batch >> order, 2); | |
3757 | alloced = rmqueue_bulk(zone, order, | |
3758 | batch, list, | |
6bb15450 | 3759 | migratetype, alloc_flags); |
44042b44 MG |
3760 | |
3761 | pcp->count += alloced << order; | |
066b2393 MG |
3762 | if (unlikely(list_empty(list))) |
3763 | return NULL; | |
3764 | } | |
3765 | ||
bf75f200 MG |
3766 | page = list_first_entry(list, struct page, pcp_list); |
3767 | list_del(&page->pcp_list); | |
44042b44 | 3768 | pcp->count -= 1 << order; |
77fe7f13 | 3769 | } while (check_new_pcp(page, order)); |
066b2393 MG |
3770 | |
3771 | return page; | |
3772 | } | |
3773 | ||
3774 | /* Lock and remove page from the per-cpu list */ | |
3775 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
44042b44 | 3776 | struct zone *zone, unsigned int order, |
663d0cfd | 3777 | int migratetype, unsigned int alloc_flags) |
066b2393 MG |
3778 | { |
3779 | struct per_cpu_pages *pcp; | |
3780 | struct list_head *list; | |
066b2393 | 3781 | struct page *page; |
d34b0733 | 3782 | unsigned long flags; |
4b23a68f | 3783 | unsigned long __maybe_unused UP_flags; |
066b2393 | 3784 | |
4b23a68f MG |
3785 | /* |
3786 | * spin_trylock may fail due to a parallel drain. In the future, the | |
3787 | * trylock will also protect against IRQ reentrancy. | |
3788 | */ | |
4b23a68f | 3789 | pcp_trylock_prepare(UP_flags); |
01b44456 MG |
3790 | pcp = pcp_spin_trylock_irqsave(zone->per_cpu_pageset, flags); |
3791 | if (!pcp) { | |
4b23a68f | 3792 | pcp_trylock_finish(UP_flags); |
4b23a68f MG |
3793 | return NULL; |
3794 | } | |
3b12e7e9 MG |
3795 | |
3796 | /* | |
3797 | * On allocation, reduce the number of pages that are batch freed. | |
3798 | * See nr_pcp_free() where free_factor is increased for subsequent | |
3799 | * frees. | |
3800 | */ | |
3b12e7e9 | 3801 | pcp->free_factor >>= 1; |
44042b44 MG |
3802 | list = &pcp->lists[order_to_pindex(migratetype, order)]; |
3803 | page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list); | |
01b44456 | 3804 | pcp_spin_unlock_irqrestore(pcp, flags); |
4b23a68f | 3805 | pcp_trylock_finish(UP_flags); |
066b2393 | 3806 | if (page) { |
15cd9004 | 3807 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
3e23060b | 3808 | zone_statistics(preferred_zone, zone, 1); |
066b2393 | 3809 | } |
066b2393 MG |
3810 | return page; |
3811 | } | |
3812 | ||
1da177e4 | 3813 | /* |
a57ae9ef RX |
3814 | * Allocate a page from the given zone. |
3815 | * Use pcplists for THP or "cheap" high-order allocations. | |
1da177e4 | 3816 | */ |
b073d7f8 AP |
3817 | |
3818 | /* | |
3819 | * Do not instrument rmqueue() with KMSAN. This function may call | |
3820 | * __msan_poison_alloca() through a call to set_pfnblock_flags_mask(). | |
3821 | * If __msan_poison_alloca() attempts to allocate pages for the stack depot, it | |
3822 | * may call rmqueue() again, which will result in a deadlock. | |
1da177e4 | 3823 | */ |
b073d7f8 | 3824 | __no_sanitize_memory |
0a15c3e9 | 3825 | static inline |
066b2393 | 3826 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3827 | struct zone *zone, unsigned int order, |
c603844b MG |
3828 | gfp_t gfp_flags, unsigned int alloc_flags, |
3829 | int migratetype) | |
1da177e4 | 3830 | { |
689bcebf | 3831 | struct page *page; |
1da177e4 | 3832 | |
589d9973 MG |
3833 | /* |
3834 | * We most definitely don't want callers attempting to | |
3835 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3836 | */ | |
3837 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3838 | ||
44042b44 | 3839 | if (likely(pcp_allowed_order(order))) { |
1d91df85 JK |
3840 | /* |
3841 | * MIGRATE_MOVABLE pcplist could have the pages on CMA area and | |
3842 | * we need to skip it when CMA area isn't allowed. | |
3843 | */ | |
3844 | if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA || | |
3845 | migratetype != MIGRATE_MOVABLE) { | |
44042b44 | 3846 | page = rmqueue_pcplist(preferred_zone, zone, order, |
663d0cfd | 3847 | migratetype, alloc_flags); |
4b23a68f MG |
3848 | if (likely(page)) |
3849 | goto out; | |
1d91df85 | 3850 | } |
066b2393 | 3851 | } |
83b9355b | 3852 | |
589d9973 MG |
3853 | page = rmqueue_buddy(preferred_zone, zone, order, alloc_flags, |
3854 | migratetype); | |
1da177e4 | 3855 | |
066b2393 | 3856 | out: |
73444bc4 | 3857 | /* Separate test+clear to avoid unnecessary atomics */ |
e2a66c21 | 3858 | if (unlikely(test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags))) { |
73444bc4 MG |
3859 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
3860 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3861 | } | |
3862 | ||
066b2393 | 3863 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 LT |
3864 | return page; |
3865 | } | |
3866 | ||
933e312e AM |
3867 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3868 | ||
b2588c4b | 3869 | static struct { |
933e312e AM |
3870 | struct fault_attr attr; |
3871 | ||
621a5f7a | 3872 | bool ignore_gfp_highmem; |
71baba4b | 3873 | bool ignore_gfp_reclaim; |
54114994 | 3874 | u32 min_order; |
933e312e AM |
3875 | } fail_page_alloc = { |
3876 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3877 | .ignore_gfp_reclaim = true, |
621a5f7a | 3878 | .ignore_gfp_highmem = true, |
54114994 | 3879 | .min_order = 1, |
933e312e AM |
3880 | }; |
3881 | ||
3882 | static int __init setup_fail_page_alloc(char *str) | |
3883 | { | |
3884 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3885 | } | |
3886 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3887 | ||
af3b8544 | 3888 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3889 | { |
54114994 | 3890 | if (order < fail_page_alloc.min_order) |
deaf386e | 3891 | return false; |
933e312e | 3892 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3893 | return false; |
933e312e | 3894 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3895 | return false; |
71baba4b MG |
3896 | if (fail_page_alloc.ignore_gfp_reclaim && |
3897 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3898 | return false; |
933e312e | 3899 | |
3f913fc5 QZ |
3900 | if (gfp_mask & __GFP_NOWARN) |
3901 | fail_page_alloc.attr.no_warn = true; | |
3902 | ||
933e312e AM |
3903 | return should_fail(&fail_page_alloc.attr, 1 << order); |
3904 | } | |
3905 | ||
3906 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3907 | ||
3908 | static int __init fail_page_alloc_debugfs(void) | |
3909 | { | |
0825a6f9 | 3910 | umode_t mode = S_IFREG | 0600; |
933e312e | 3911 | struct dentry *dir; |
933e312e | 3912 | |
dd48c085 AM |
3913 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3914 | &fail_page_alloc.attr); | |
b2588c4b | 3915 | |
d9f7979c GKH |
3916 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3917 | &fail_page_alloc.ignore_gfp_reclaim); | |
3918 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3919 | &fail_page_alloc.ignore_gfp_highmem); | |
3920 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3921 | |
d9f7979c | 3922 | return 0; |
933e312e AM |
3923 | } |
3924 | ||
3925 | late_initcall(fail_page_alloc_debugfs); | |
3926 | ||
3927 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3928 | ||
3929 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3930 | ||
af3b8544 | 3931 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3932 | { |
deaf386e | 3933 | return false; |
933e312e AM |
3934 | } |
3935 | ||
3936 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3937 | ||
54aa3866 | 3938 | noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
af3b8544 BP |
3939 | { |
3940 | return __should_fail_alloc_page(gfp_mask, order); | |
3941 | } | |
3942 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3943 | ||
f27ce0e1 JK |
3944 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3945 | unsigned int order, unsigned int alloc_flags) | |
3946 | { | |
3947 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); | |
3948 | long unusable_free = (1 << order) - 1; | |
3949 | ||
3950 | /* | |
3951 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3952 | * the high-atomic reserves. This will over-estimate the size of the | |
3953 | * atomic reserve but it avoids a search. | |
3954 | */ | |
3955 | if (likely(!alloc_harder)) | |
3956 | unusable_free += z->nr_reserved_highatomic; | |
3957 | ||
3958 | #ifdef CONFIG_CMA | |
3959 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3960 | if (!(alloc_flags & ALLOC_CMA)) | |
3961 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3962 | #endif | |
3963 | ||
3964 | return unusable_free; | |
3965 | } | |
3966 | ||
1da177e4 | 3967 | /* |
97a16fc8 MG |
3968 | * Return true if free base pages are above 'mark'. For high-order checks it |
3969 | * will return true of the order-0 watermark is reached and there is at least | |
3970 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3971 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3972 | */ |
86a294a8 | 3973 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3974 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3975 | long free_pages) |
1da177e4 | 3976 | { |
d23ad423 | 3977 | long min = mark; |
1da177e4 | 3978 | int o; |
cd04ae1e | 3979 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3980 | |
0aaa29a5 | 3981 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 3982 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 3983 | |
7fb1d9fc | 3984 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3985 | min -= min / 2; |
0aaa29a5 | 3986 | |
f27ce0e1 | 3987 | if (unlikely(alloc_harder)) { |
cd04ae1e MH |
3988 | /* |
3989 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3990 | * users on the grounds that it's definitely going to be in | |
3991 | * the exit path shortly and free memory. Any allocation it | |
3992 | * makes during the free path will be small and short-lived. | |
3993 | */ | |
3994 | if (alloc_flags & ALLOC_OOM) | |
3995 | min -= min / 2; | |
3996 | else | |
3997 | min -= min / 4; | |
3998 | } | |
3999 | ||
97a16fc8 MG |
4000 | /* |
4001 | * Check watermarks for an order-0 allocation request. If these | |
4002 | * are not met, then a high-order request also cannot go ahead | |
4003 | * even if a suitable page happened to be free. | |
4004 | */ | |
97a225e6 | 4005 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 4006 | return false; |
1da177e4 | 4007 | |
97a16fc8 MG |
4008 | /* If this is an order-0 request then the watermark is fine */ |
4009 | if (!order) | |
4010 | return true; | |
4011 | ||
4012 | /* For a high-order request, check at least one suitable page is free */ | |
4013 | for (o = order; o < MAX_ORDER; o++) { | |
4014 | struct free_area *area = &z->free_area[o]; | |
4015 | int mt; | |
4016 | ||
4017 | if (!area->nr_free) | |
4018 | continue; | |
4019 | ||
97a16fc8 | 4020 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 4021 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
4022 | return true; |
4023 | } | |
4024 | ||
4025 | #ifdef CONFIG_CMA | |
d883c6cf | 4026 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 4027 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 4028 | return true; |
d883c6cf | 4029 | } |
97a16fc8 | 4030 | #endif |
76089d00 | 4031 | if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC)) |
b050e376 | 4032 | return true; |
1da177e4 | 4033 | } |
97a16fc8 | 4034 | return false; |
88f5acf8 MG |
4035 | } |
4036 | ||
7aeb09f9 | 4037 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 4038 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 4039 | { |
97a225e6 | 4040 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
4041 | zone_page_state(z, NR_FREE_PAGES)); |
4042 | } | |
4043 | ||
48ee5f36 | 4044 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 4045 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 4046 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 4047 | { |
f27ce0e1 | 4048 | long free_pages; |
d883c6cf | 4049 | |
f27ce0e1 | 4050 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
4051 | |
4052 | /* | |
4053 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 4054 | * need to be calculated. |
48ee5f36 | 4055 | */ |
f27ce0e1 | 4056 | if (!order) { |
9282012f JK |
4057 | long usable_free; |
4058 | long reserved; | |
f27ce0e1 | 4059 | |
9282012f JK |
4060 | usable_free = free_pages; |
4061 | reserved = __zone_watermark_unusable_free(z, 0, alloc_flags); | |
4062 | ||
4063 | /* reserved may over estimate high-atomic reserves. */ | |
4064 | usable_free -= min(usable_free, reserved); | |
4065 | if (usable_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
f27ce0e1 JK |
4066 | return true; |
4067 | } | |
48ee5f36 | 4068 | |
f80b08fc CTR |
4069 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
4070 | free_pages)) | |
4071 | return true; | |
4072 | /* | |
4073 | * Ignore watermark boosting for GFP_ATOMIC order-0 allocations | |
4074 | * when checking the min watermark. The min watermark is the | |
4075 | * point where boosting is ignored so that kswapd is woken up | |
4076 | * when below the low watermark. | |
4077 | */ | |
4078 | if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost | |
4079 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { | |
4080 | mark = z->_watermark[WMARK_MIN]; | |
4081 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
4082 | alloc_flags, free_pages); | |
4083 | } | |
4084 | ||
4085 | return false; | |
48ee5f36 MG |
4086 | } |
4087 | ||
7aeb09f9 | 4088 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 4089 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
4090 | { |
4091 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
4092 | ||
4093 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
4094 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
4095 | ||
97a225e6 | 4096 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 4097 | free_pages); |
1da177e4 LT |
4098 | } |
4099 | ||
9276b1bc | 4100 | #ifdef CONFIG_NUMA |
61bb6cd2 GU |
4101 | int __read_mostly node_reclaim_distance = RECLAIM_DISTANCE; |
4102 | ||
957f822a DR |
4103 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
4104 | { | |
e02dc017 | 4105 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 4106 | node_reclaim_distance; |
957f822a | 4107 | } |
9276b1bc | 4108 | #else /* CONFIG_NUMA */ |
957f822a DR |
4109 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
4110 | { | |
4111 | return true; | |
4112 | } | |
9276b1bc PJ |
4113 | #endif /* CONFIG_NUMA */ |
4114 | ||
6bb15450 MG |
4115 | /* |
4116 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
4117 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
4118 | * premature use of a lower zone may cause lowmem pressure problems that | |
4119 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
4120 | * probably too small. It only makes sense to spread allocations to avoid | |
4121 | * fragmentation between the Normal and DMA32 zones. | |
4122 | */ | |
4123 | static inline unsigned int | |
0a79cdad | 4124 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 4125 | { |
736838e9 | 4126 | unsigned int alloc_flags; |
0a79cdad | 4127 | |
736838e9 MN |
4128 | /* |
4129 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4130 | * to save a branch. | |
4131 | */ | |
4132 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
4133 | |
4134 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
4135 | if (!zone) |
4136 | return alloc_flags; | |
4137 | ||
6bb15450 | 4138 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 4139 | return alloc_flags; |
6bb15450 MG |
4140 | |
4141 | /* | |
4142 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
4143 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
4144 | * on UMA that if Normal is populated then so is DMA32. | |
4145 | */ | |
4146 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
4147 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 4148 | return alloc_flags; |
6bb15450 | 4149 | |
8118b82e | 4150 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
4151 | #endif /* CONFIG_ZONE_DMA32 */ |
4152 | return alloc_flags; | |
6bb15450 | 4153 | } |
6bb15450 | 4154 | |
8e3560d9 PT |
4155 | /* Must be called after current_gfp_context() which can change gfp_mask */ |
4156 | static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask, | |
4157 | unsigned int alloc_flags) | |
8510e69c JK |
4158 | { |
4159 | #ifdef CONFIG_CMA | |
8e3560d9 | 4160 | if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
8510e69c | 4161 | alloc_flags |= ALLOC_CMA; |
8510e69c JK |
4162 | #endif |
4163 | return alloc_flags; | |
4164 | } | |
4165 | ||
7fb1d9fc | 4166 | /* |
0798e519 | 4167 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
4168 | * a page. |
4169 | */ | |
4170 | static struct page * | |
a9263751 VB |
4171 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
4172 | const struct alloc_context *ac) | |
753ee728 | 4173 | { |
6bb15450 | 4174 | struct zoneref *z; |
5117f45d | 4175 | struct zone *zone; |
8a87d695 WY |
4176 | struct pglist_data *last_pgdat = NULL; |
4177 | bool last_pgdat_dirty_ok = false; | |
6bb15450 | 4178 | bool no_fallback; |
3b8c0be4 | 4179 | |
6bb15450 | 4180 | retry: |
7fb1d9fc | 4181 | /* |
9276b1bc | 4182 | * Scan zonelist, looking for a zone with enough free. |
189cdcfe | 4183 | * See also __cpuset_node_allowed() comment in kernel/cgroup/cpuset.c. |
7fb1d9fc | 4184 | */ |
6bb15450 MG |
4185 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
4186 | z = ac->preferred_zoneref; | |
30d8ec73 MN |
4187 | for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx, |
4188 | ac->nodemask) { | |
be06af00 | 4189 | struct page *page; |
e085dbc5 JW |
4190 | unsigned long mark; |
4191 | ||
664eedde MG |
4192 | if (cpusets_enabled() && |
4193 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 4194 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 4195 | continue; |
a756cf59 JW |
4196 | /* |
4197 | * When allocating a page cache page for writing, we | |
281e3726 MG |
4198 | * want to get it from a node that is within its dirty |
4199 | * limit, such that no single node holds more than its | |
a756cf59 | 4200 | * proportional share of globally allowed dirty pages. |
281e3726 | 4201 | * The dirty limits take into account the node's |
a756cf59 JW |
4202 | * lowmem reserves and high watermark so that kswapd |
4203 | * should be able to balance it without having to | |
4204 | * write pages from its LRU list. | |
4205 | * | |
a756cf59 | 4206 | * XXX: For now, allow allocations to potentially |
281e3726 | 4207 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 4208 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 4209 | * which is important when on a NUMA setup the allowed |
281e3726 | 4210 | * nodes are together not big enough to reach the |
a756cf59 | 4211 | * global limit. The proper fix for these situations |
281e3726 | 4212 | * will require awareness of nodes in the |
a756cf59 JW |
4213 | * dirty-throttling and the flusher threads. |
4214 | */ | |
3b8c0be4 | 4215 | if (ac->spread_dirty_pages) { |
8a87d695 WY |
4216 | if (last_pgdat != zone->zone_pgdat) { |
4217 | last_pgdat = zone->zone_pgdat; | |
4218 | last_pgdat_dirty_ok = node_dirty_ok(zone->zone_pgdat); | |
4219 | } | |
3b8c0be4 | 4220 | |
8a87d695 | 4221 | if (!last_pgdat_dirty_ok) |
3b8c0be4 | 4222 | continue; |
3b8c0be4 | 4223 | } |
7fb1d9fc | 4224 | |
6bb15450 MG |
4225 | if (no_fallback && nr_online_nodes > 1 && |
4226 | zone != ac->preferred_zoneref->zone) { | |
4227 | int local_nid; | |
4228 | ||
4229 | /* | |
4230 | * If moving to a remote node, retry but allow | |
4231 | * fragmenting fallbacks. Locality is more important | |
4232 | * than fragmentation avoidance. | |
4233 | */ | |
4234 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
4235 | if (zone_to_nid(zone) != local_nid) { | |
4236 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
4237 | goto retry; | |
4238 | } | |
4239 | } | |
4240 | ||
a9214443 | 4241 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 4242 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
4243 | ac->highest_zoneidx, alloc_flags, |
4244 | gfp_mask)) { | |
fa5e084e MG |
4245 | int ret; |
4246 | ||
c9e97a19 PT |
4247 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
4248 | /* | |
4249 | * Watermark failed for this zone, but see if we can | |
4250 | * grow this zone if it contains deferred pages. | |
4251 | */ | |
4252 | if (static_branch_unlikely(&deferred_pages)) { | |
4253 | if (_deferred_grow_zone(zone, order)) | |
4254 | goto try_this_zone; | |
4255 | } | |
4256 | #endif | |
5dab2911 MG |
4257 | /* Checked here to keep the fast path fast */ |
4258 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
4259 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
4260 | goto try_this_zone; | |
4261 | ||
202e35db | 4262 | if (!node_reclaim_enabled() || |
c33d6c06 | 4263 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
4264 | continue; |
4265 | ||
a5f5f91d | 4266 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 4267 | switch (ret) { |
a5f5f91d | 4268 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 4269 | /* did not scan */ |
cd38b115 | 4270 | continue; |
a5f5f91d | 4271 | case NODE_RECLAIM_FULL: |
fa5e084e | 4272 | /* scanned but unreclaimable */ |
cd38b115 | 4273 | continue; |
fa5e084e MG |
4274 | default: |
4275 | /* did we reclaim enough */ | |
fed2719e | 4276 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 4277 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
4278 | goto try_this_zone; |
4279 | ||
fed2719e | 4280 | continue; |
0798e519 | 4281 | } |
7fb1d9fc RS |
4282 | } |
4283 | ||
fa5e084e | 4284 | try_this_zone: |
066b2393 | 4285 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 4286 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 4287 | if (page) { |
479f854a | 4288 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
4289 | |
4290 | /* | |
4291 | * If this is a high-order atomic allocation then check | |
4292 | * if the pageblock should be reserved for the future | |
4293 | */ | |
4294 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
4295 | reserve_highatomic_pageblock(page, zone, order); | |
4296 | ||
75379191 | 4297 | return page; |
c9e97a19 PT |
4298 | } else { |
4299 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
4300 | /* Try again if zone has deferred pages */ | |
4301 | if (static_branch_unlikely(&deferred_pages)) { | |
4302 | if (_deferred_grow_zone(zone, order)) | |
4303 | goto try_this_zone; | |
4304 | } | |
4305 | #endif | |
75379191 | 4306 | } |
54a6eb5c | 4307 | } |
9276b1bc | 4308 | |
6bb15450 MG |
4309 | /* |
4310 | * It's possible on a UMA machine to get through all zones that are | |
4311 | * fragmented. If avoiding fragmentation, reset and try again. | |
4312 | */ | |
4313 | if (no_fallback) { | |
4314 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
4315 | goto retry; | |
4316 | } | |
4317 | ||
4ffeaf35 | 4318 | return NULL; |
753ee728 MH |
4319 | } |
4320 | ||
9af744d7 | 4321 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 4322 | { |
a238ab5b | 4323 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
4324 | |
4325 | /* | |
4326 | * This documents exceptions given to allocations in certain | |
4327 | * contexts that are allowed to allocate outside current's set | |
4328 | * of allowed nodes. | |
4329 | */ | |
4330 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 4331 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
4332 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
4333 | filter &= ~SHOW_MEM_FILTER_NODES; | |
88dc6f20 | 4334 | if (!in_task() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
4335 | filter &= ~SHOW_MEM_FILTER_NODES; |
4336 | ||
974f4367 | 4337 | __show_mem(filter, nodemask, gfp_zone(gfp_mask)); |
aa187507 MH |
4338 | } |
4339 | ||
a8e99259 | 4340 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
4341 | { |
4342 | struct va_format vaf; | |
4343 | va_list args; | |
1be334e5 | 4344 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 4345 | |
c4dc63f0 BH |
4346 | if ((gfp_mask & __GFP_NOWARN) || |
4347 | !__ratelimit(&nopage_rs) || | |
4348 | ((gfp_mask & __GFP_DMA) && !has_managed_dma())) | |
aa187507 MH |
4349 | return; |
4350 | ||
7877cdcc MH |
4351 | va_start(args, fmt); |
4352 | vaf.fmt = fmt; | |
4353 | vaf.va = &args; | |
ef8444ea | 4354 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
4355 | current->comm, &vaf, gfp_mask, &gfp_mask, |
4356 | nodemask_pr_args(nodemask)); | |
7877cdcc | 4357 | va_end(args); |
3ee9a4f0 | 4358 | |
a8e99259 | 4359 | cpuset_print_current_mems_allowed(); |
ef8444ea | 4360 | pr_cont("\n"); |
a238ab5b | 4361 | dump_stack(); |
685dbf6f | 4362 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
4363 | } |
4364 | ||
6c18ba7a MH |
4365 | static inline struct page * |
4366 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
4367 | unsigned int alloc_flags, | |
4368 | const struct alloc_context *ac) | |
4369 | { | |
4370 | struct page *page; | |
4371 | ||
4372 | page = get_page_from_freelist(gfp_mask, order, | |
4373 | alloc_flags|ALLOC_CPUSET, ac); | |
4374 | /* | |
4375 | * fallback to ignore cpuset restriction if our nodes | |
4376 | * are depleted | |
4377 | */ | |
4378 | if (!page) | |
4379 | page = get_page_from_freelist(gfp_mask, order, | |
4380 | alloc_flags, ac); | |
4381 | ||
4382 | return page; | |
4383 | } | |
4384 | ||
11e33f6a MG |
4385 | static inline struct page * |
4386 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4387 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 4388 | { |
6e0fc46d DR |
4389 | struct oom_control oc = { |
4390 | .zonelist = ac->zonelist, | |
4391 | .nodemask = ac->nodemask, | |
2a966b77 | 4392 | .memcg = NULL, |
6e0fc46d DR |
4393 | .gfp_mask = gfp_mask, |
4394 | .order = order, | |
6e0fc46d | 4395 | }; |
11e33f6a MG |
4396 | struct page *page; |
4397 | ||
9879de73 JW |
4398 | *did_some_progress = 0; |
4399 | ||
9879de73 | 4400 | /* |
dc56401f JW |
4401 | * Acquire the oom lock. If that fails, somebody else is |
4402 | * making progress for us. | |
9879de73 | 4403 | */ |
dc56401f | 4404 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 4405 | *did_some_progress = 1; |
11e33f6a | 4406 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
4407 | return NULL; |
4408 | } | |
6b1de916 | 4409 | |
11e33f6a MG |
4410 | /* |
4411 | * Go through the zonelist yet one more time, keep very high watermark | |
4412 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
4413 | * we're still under heavy pressure. But make sure that this reclaim |
4414 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
4415 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 4416 | */ |
e746bf73 TH |
4417 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
4418 | ~__GFP_DIRECT_RECLAIM, order, | |
4419 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 4420 | if (page) |
11e33f6a MG |
4421 | goto out; |
4422 | ||
06ad276a MH |
4423 | /* Coredumps can quickly deplete all memory reserves */ |
4424 | if (current->flags & PF_DUMPCORE) | |
4425 | goto out; | |
4426 | /* The OOM killer will not help higher order allocs */ | |
4427 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4428 | goto out; | |
dcda9b04 MH |
4429 | /* |
4430 | * We have already exhausted all our reclaim opportunities without any | |
4431 | * success so it is time to admit defeat. We will skip the OOM killer | |
4432 | * because it is very likely that the caller has a more reasonable | |
4433 | * fallback than shooting a random task. | |
cfb4a541 MN |
4434 | * |
4435 | * The OOM killer may not free memory on a specific node. | |
dcda9b04 | 4436 | */ |
cfb4a541 | 4437 | if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE)) |
dcda9b04 | 4438 | goto out; |
06ad276a | 4439 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 4440 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
4441 | goto out; |
4442 | if (pm_suspended_storage()) | |
4443 | goto out; | |
4444 | /* | |
4445 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
4446 | * other request to make a forward progress. | |
4447 | * We are in an unfortunate situation where out_of_memory cannot | |
4448 | * do much for this context but let's try it to at least get | |
4449 | * access to memory reserved if the current task is killed (see | |
4450 | * out_of_memory). Once filesystems are ready to handle allocation | |
4451 | * failures more gracefully we should just bail out here. | |
4452 | */ | |
4453 | ||
3c2c6488 | 4454 | /* Exhausted what can be done so it's blame time */ |
3f913fc5 QZ |
4455 | if (out_of_memory(&oc) || |
4456 | WARN_ON_ONCE_GFP(gfp_mask & __GFP_NOFAIL, gfp_mask)) { | |
c32b3cbe | 4457 | *did_some_progress = 1; |
5020e285 | 4458 | |
6c18ba7a MH |
4459 | /* |
4460 | * Help non-failing allocations by giving them access to memory | |
4461 | * reserves | |
4462 | */ | |
4463 | if (gfp_mask & __GFP_NOFAIL) | |
4464 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 4465 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 4466 | } |
11e33f6a | 4467 | out: |
dc56401f | 4468 | mutex_unlock(&oom_lock); |
11e33f6a MG |
4469 | return page; |
4470 | } | |
4471 | ||
33c2d214 | 4472 | /* |
baf2f90b | 4473 | * Maximum number of compaction retries with a progress before OOM |
33c2d214 MH |
4474 | * killer is consider as the only way to move forward. |
4475 | */ | |
4476 | #define MAX_COMPACT_RETRIES 16 | |
4477 | ||
56de7263 MG |
4478 | #ifdef CONFIG_COMPACTION |
4479 | /* Try memory compaction for high-order allocations before reclaim */ | |
4480 | static struct page * | |
4481 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4482 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4483 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4484 | { |
5e1f0f09 | 4485 | struct page *page = NULL; |
eb414681 | 4486 | unsigned long pflags; |
499118e9 | 4487 | unsigned int noreclaim_flag; |
53853e2d VB |
4488 | |
4489 | if (!order) | |
66199712 | 4490 | return NULL; |
66199712 | 4491 | |
eb414681 | 4492 | psi_memstall_enter(&pflags); |
5bf18281 | 4493 | delayacct_compact_start(); |
499118e9 | 4494 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 4495 | |
c5d01d0d | 4496 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 4497 | prio, &page); |
eb414681 | 4498 | |
499118e9 | 4499 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 4500 | psi_memstall_leave(&pflags); |
5bf18281 | 4501 | delayacct_compact_end(); |
56de7263 | 4502 | |
06dac2f4 CTR |
4503 | if (*compact_result == COMPACT_SKIPPED) |
4504 | return NULL; | |
98dd3b48 VB |
4505 | /* |
4506 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
4507 | * count a compaction stall | |
4508 | */ | |
4509 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 4510 | |
5e1f0f09 MG |
4511 | /* Prep a captured page if available */ |
4512 | if (page) | |
4513 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
4514 | ||
4515 | /* Try get a page from the freelist if available */ | |
4516 | if (!page) | |
4517 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 4518 | |
98dd3b48 VB |
4519 | if (page) { |
4520 | struct zone *zone = page_zone(page); | |
53853e2d | 4521 | |
98dd3b48 VB |
4522 | zone->compact_blockskip_flush = false; |
4523 | compaction_defer_reset(zone, order, true); | |
4524 | count_vm_event(COMPACTSUCCESS); | |
4525 | return page; | |
4526 | } | |
56de7263 | 4527 | |
98dd3b48 VB |
4528 | /* |
4529 | * It's bad if compaction run occurs and fails. The most likely reason | |
4530 | * is that pages exist, but not enough to satisfy watermarks. | |
4531 | */ | |
4532 | count_vm_event(COMPACTFAIL); | |
66199712 | 4533 | |
98dd3b48 | 4534 | cond_resched(); |
56de7263 MG |
4535 | |
4536 | return NULL; | |
4537 | } | |
33c2d214 | 4538 | |
3250845d VB |
4539 | static inline bool |
4540 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
4541 | enum compact_result compact_result, | |
4542 | enum compact_priority *compact_priority, | |
d9436498 | 4543 | int *compaction_retries) |
3250845d VB |
4544 | { |
4545 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 4546 | int min_priority; |
65190cff MH |
4547 | bool ret = false; |
4548 | int retries = *compaction_retries; | |
4549 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
4550 | |
4551 | if (!order) | |
4552 | return false; | |
4553 | ||
691d9497 AT |
4554 | if (fatal_signal_pending(current)) |
4555 | return false; | |
4556 | ||
d9436498 VB |
4557 | if (compaction_made_progress(compact_result)) |
4558 | (*compaction_retries)++; | |
4559 | ||
3250845d VB |
4560 | /* |
4561 | * compaction considers all the zone as desperately out of memory | |
4562 | * so it doesn't really make much sense to retry except when the | |
4563 | * failure could be caused by insufficient priority | |
4564 | */ | |
d9436498 VB |
4565 | if (compaction_failed(compact_result)) |
4566 | goto check_priority; | |
3250845d | 4567 | |
49433085 VB |
4568 | /* |
4569 | * compaction was skipped because there are not enough order-0 pages | |
4570 | * to work with, so we retry only if it looks like reclaim can help. | |
4571 | */ | |
4572 | if (compaction_needs_reclaim(compact_result)) { | |
4573 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
4574 | goto out; | |
4575 | } | |
4576 | ||
3250845d VB |
4577 | /* |
4578 | * make sure the compaction wasn't deferred or didn't bail out early | |
4579 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
4580 | * But the next retry should use a higher priority if allowed, so |
4581 | * we don't just keep bailing out endlessly. | |
3250845d | 4582 | */ |
65190cff | 4583 | if (compaction_withdrawn(compact_result)) { |
49433085 | 4584 | goto check_priority; |
65190cff | 4585 | } |
3250845d VB |
4586 | |
4587 | /* | |
dcda9b04 | 4588 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
4589 | * costly ones because they are de facto nofail and invoke OOM |
4590 | * killer to move on while costly can fail and users are ready | |
4591 | * to cope with that. 1/4 retries is rather arbitrary but we | |
4592 | * would need much more detailed feedback from compaction to | |
4593 | * make a better decision. | |
4594 | */ | |
4595 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4596 | max_retries /= 4; | |
65190cff MH |
4597 | if (*compaction_retries <= max_retries) { |
4598 | ret = true; | |
4599 | goto out; | |
4600 | } | |
3250845d | 4601 | |
d9436498 VB |
4602 | /* |
4603 | * Make sure there are attempts at the highest priority if we exhausted | |
4604 | * all retries or failed at the lower priorities. | |
4605 | */ | |
4606 | check_priority: | |
c2033b00 VB |
4607 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
4608 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 4609 | |
c2033b00 | 4610 | if (*compact_priority > min_priority) { |
d9436498 VB |
4611 | (*compact_priority)--; |
4612 | *compaction_retries = 0; | |
65190cff | 4613 | ret = true; |
d9436498 | 4614 | } |
65190cff MH |
4615 | out: |
4616 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4617 | return ret; | |
3250845d | 4618 | } |
56de7263 MG |
4619 | #else |
4620 | static inline struct page * | |
4621 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4622 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4623 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4624 | { |
33c2d214 | 4625 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4626 | return NULL; |
4627 | } | |
33c2d214 MH |
4628 | |
4629 | static inline bool | |
86a294a8 MH |
4630 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4631 | enum compact_result compact_result, | |
a5508cd8 | 4632 | enum compact_priority *compact_priority, |
d9436498 | 4633 | int *compaction_retries) |
33c2d214 | 4634 | { |
31e49bfd MH |
4635 | struct zone *zone; |
4636 | struct zoneref *z; | |
4637 | ||
4638 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4639 | return false; | |
4640 | ||
4641 | /* | |
4642 | * There are setups with compaction disabled which would prefer to loop | |
4643 | * inside the allocator rather than hit the oom killer prematurely. | |
4644 | * Let's give them a good hope and keep retrying while the order-0 | |
4645 | * watermarks are OK. | |
4646 | */ | |
97a225e6 JK |
4647 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4648 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 4649 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 4650 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
4651 | return true; |
4652 | } | |
33c2d214 MH |
4653 | return false; |
4654 | } | |
3250845d | 4655 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4656 | |
d92a8cfc | 4657 | #ifdef CONFIG_LOCKDEP |
93781325 | 4658 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4659 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4660 | ||
f920e413 | 4661 | static bool __need_reclaim(gfp_t gfp_mask) |
d92a8cfc | 4662 | { |
d92a8cfc PZ |
4663 | /* no reclaim without waiting on it */ |
4664 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4665 | return false; | |
4666 | ||
4667 | /* this guy won't enter reclaim */ | |
2e517d68 | 4668 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4669 | return false; |
4670 | ||
d92a8cfc PZ |
4671 | if (gfp_mask & __GFP_NOLOCKDEP) |
4672 | return false; | |
4673 | ||
4674 | return true; | |
4675 | } | |
4676 | ||
4f3eaf45 | 4677 | void __fs_reclaim_acquire(unsigned long ip) |
93781325 | 4678 | { |
4f3eaf45 | 4679 | lock_acquire_exclusive(&__fs_reclaim_map, 0, 0, NULL, ip); |
93781325 OS |
4680 | } |
4681 | ||
4f3eaf45 | 4682 | void __fs_reclaim_release(unsigned long ip) |
93781325 | 4683 | { |
4f3eaf45 | 4684 | lock_release(&__fs_reclaim_map, ip); |
93781325 OS |
4685 | } |
4686 | ||
d92a8cfc PZ |
4687 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4688 | { | |
f920e413 DV |
4689 | gfp_mask = current_gfp_context(gfp_mask); |
4690 | ||
4691 | if (__need_reclaim(gfp_mask)) { | |
4692 | if (gfp_mask & __GFP_FS) | |
4f3eaf45 | 4693 | __fs_reclaim_acquire(_RET_IP_); |
f920e413 DV |
4694 | |
4695 | #ifdef CONFIG_MMU_NOTIFIER | |
4696 | lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); | |
4697 | lock_map_release(&__mmu_notifier_invalidate_range_start_map); | |
4698 | #endif | |
4699 | ||
4700 | } | |
d92a8cfc PZ |
4701 | } |
4702 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4703 | ||
4704 | void fs_reclaim_release(gfp_t gfp_mask) | |
4705 | { | |
f920e413 DV |
4706 | gfp_mask = current_gfp_context(gfp_mask); |
4707 | ||
4708 | if (__need_reclaim(gfp_mask)) { | |
4709 | if (gfp_mask & __GFP_FS) | |
4f3eaf45 | 4710 | __fs_reclaim_release(_RET_IP_); |
f920e413 | 4711 | } |
d92a8cfc PZ |
4712 | } |
4713 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4714 | #endif | |
4715 | ||
3d36424b MG |
4716 | /* |
4717 | * Zonelists may change due to hotplug during allocation. Detect when zonelists | |
4718 | * have been rebuilt so allocation retries. Reader side does not lock and | |
4719 | * retries the allocation if zonelist changes. Writer side is protected by the | |
4720 | * embedded spin_lock. | |
4721 | */ | |
4722 | static DEFINE_SEQLOCK(zonelist_update_seq); | |
4723 | ||
4724 | static unsigned int zonelist_iter_begin(void) | |
4725 | { | |
4726 | if (IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) | |
4727 | return read_seqbegin(&zonelist_update_seq); | |
4728 | ||
4729 | return 0; | |
4730 | } | |
4731 | ||
4732 | static unsigned int check_retry_zonelist(unsigned int seq) | |
4733 | { | |
4734 | if (IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) | |
4735 | return read_seqretry(&zonelist_update_seq, seq); | |
4736 | ||
4737 | return seq; | |
4738 | } | |
4739 | ||
bba90710 | 4740 | /* Perform direct synchronous page reclaim */ |
2187e17b | 4741 | static unsigned long |
a9263751 VB |
4742 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4743 | const struct alloc_context *ac) | |
11e33f6a | 4744 | { |
499118e9 | 4745 | unsigned int noreclaim_flag; |
fa7fc75f | 4746 | unsigned long progress; |
11e33f6a MG |
4747 | |
4748 | cond_resched(); | |
4749 | ||
4750 | /* We now go into synchronous reclaim */ | |
4751 | cpuset_memory_pressure_bump(); | |
d92a8cfc | 4752 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4753 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4754 | |
a9263751 VB |
4755 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4756 | ac->nodemask); | |
11e33f6a | 4757 | |
499118e9 | 4758 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4759 | fs_reclaim_release(gfp_mask); |
11e33f6a MG |
4760 | |
4761 | cond_resched(); | |
4762 | ||
bba90710 MS |
4763 | return progress; |
4764 | } | |
4765 | ||
4766 | /* The really slow allocator path where we enter direct reclaim */ | |
4767 | static inline struct page * | |
4768 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4769 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4770 | unsigned long *did_some_progress) |
bba90710 MS |
4771 | { |
4772 | struct page *page = NULL; | |
fa7fc75f | 4773 | unsigned long pflags; |
bba90710 MS |
4774 | bool drained = false; |
4775 | ||
fa7fc75f | 4776 | psi_memstall_enter(&pflags); |
a9263751 | 4777 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce | 4778 | if (unlikely(!(*did_some_progress))) |
fa7fc75f | 4779 | goto out; |
11e33f6a | 4780 | |
9ee493ce | 4781 | retry: |
31a6c190 | 4782 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4783 | |
4784 | /* | |
4785 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 | 4786 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
047b9967 | 4787 | * Shrink them and try again |
9ee493ce MG |
4788 | */ |
4789 | if (!page && !drained) { | |
29fac03b | 4790 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4791 | drain_all_pages(NULL); |
9ee493ce MG |
4792 | drained = true; |
4793 | goto retry; | |
4794 | } | |
fa7fc75f SB |
4795 | out: |
4796 | psi_memstall_leave(&pflags); | |
9ee493ce | 4797 | |
11e33f6a MG |
4798 | return page; |
4799 | } | |
4800 | ||
5ecd9d40 DR |
4801 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4802 | const struct alloc_context *ac) | |
3a025760 JW |
4803 | { |
4804 | struct zoneref *z; | |
4805 | struct zone *zone; | |
e1a55637 | 4806 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 4807 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 4808 | |
97a225e6 | 4809 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 4810 | ac->nodemask) { |
bc53008e WY |
4811 | if (!managed_zone(zone)) |
4812 | continue; | |
d137a7cb | 4813 | if (last_pgdat != zone->zone_pgdat) { |
97a225e6 | 4814 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
d137a7cb CW |
4815 | last_pgdat = zone->zone_pgdat; |
4816 | } | |
e1a55637 | 4817 | } |
3a025760 JW |
4818 | } |
4819 | ||
c603844b | 4820 | static inline unsigned int |
341ce06f PZ |
4821 | gfp_to_alloc_flags(gfp_t gfp_mask) |
4822 | { | |
c603844b | 4823 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4824 | |
736838e9 MN |
4825 | /* |
4826 | * __GFP_HIGH is assumed to be the same as ALLOC_HIGH | |
4827 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4828 | * to save two branches. | |
4829 | */ | |
e6223a3b | 4830 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
736838e9 | 4831 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 4832 | |
341ce06f PZ |
4833 | /* |
4834 | * The caller may dip into page reserves a bit more if the caller | |
4835 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4836 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 4837 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 4838 | */ |
736838e9 MN |
4839 | alloc_flags |= (__force int) |
4840 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 4841 | |
d0164adc | 4842 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 4843 | /* |
b104a35d DR |
4844 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4845 | * if it can't schedule. | |
5c3240d9 | 4846 | */ |
b104a35d | 4847 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 4848 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 4849 | /* |
b104a35d | 4850 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 4851 | * comment for __cpuset_node_allowed(). |
523b9458 | 4852 | */ |
341ce06f | 4853 | alloc_flags &= ~ALLOC_CPUSET; |
88dc6f20 | 4854 | } else if (unlikely(rt_task(current)) && in_task()) |
341ce06f PZ |
4855 | alloc_flags |= ALLOC_HARDER; |
4856 | ||
8e3560d9 | 4857 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags); |
8510e69c | 4858 | |
341ce06f PZ |
4859 | return alloc_flags; |
4860 | } | |
4861 | ||
cd04ae1e | 4862 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4863 | { |
cd04ae1e MH |
4864 | if (!tsk_is_oom_victim(tsk)) |
4865 | return false; | |
4866 | ||
4867 | /* | |
4868 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4869 | * only to the thread with TIF_MEMDIE set | |
4870 | */ | |
4871 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4872 | return false; |
4873 | ||
cd04ae1e MH |
4874 | return true; |
4875 | } | |
4876 | ||
4877 | /* | |
4878 | * Distinguish requests which really need access to full memory | |
4879 | * reserves from oom victims which can live with a portion of it | |
4880 | */ | |
4881 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4882 | { | |
4883 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4884 | return 0; | |
31a6c190 | 4885 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4886 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4887 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4888 | return ALLOC_NO_WATERMARKS; |
4889 | if (!in_interrupt()) { | |
4890 | if (current->flags & PF_MEMALLOC) | |
4891 | return ALLOC_NO_WATERMARKS; | |
4892 | else if (oom_reserves_allowed(current)) | |
4893 | return ALLOC_OOM; | |
4894 | } | |
31a6c190 | 4895 | |
cd04ae1e MH |
4896 | return 0; |
4897 | } | |
4898 | ||
4899 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4900 | { | |
4901 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4902 | } |
4903 | ||
0a0337e0 MH |
4904 | /* |
4905 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4906 | * for the given allocation request. | |
491d79ae JW |
4907 | * |
4908 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4909 | * without success, or when we couldn't even meet the watermark if we | |
4910 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4911 | * |
4912 | * Returns true if a retry is viable or false to enter the oom path. | |
4913 | */ | |
4914 | static inline bool | |
4915 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4916 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4917 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4918 | { |
4919 | struct zone *zone; | |
4920 | struct zoneref *z; | |
15f570bf | 4921 | bool ret = false; |
0a0337e0 | 4922 | |
423b452e VB |
4923 | /* |
4924 | * Costly allocations might have made a progress but this doesn't mean | |
4925 | * their order will become available due to high fragmentation so | |
4926 | * always increment the no progress counter for them | |
4927 | */ | |
4928 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4929 | *no_progress_loops = 0; | |
4930 | else | |
4931 | (*no_progress_loops)++; | |
4932 | ||
0a0337e0 MH |
4933 | /* |
4934 | * Make sure we converge to OOM if we cannot make any progress | |
4935 | * several times in the row. | |
4936 | */ | |
04c8716f MK |
4937 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4938 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4939 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4940 | } |
0a0337e0 | 4941 | |
bca67592 MG |
4942 | /* |
4943 | * Keep reclaiming pages while there is a chance this will lead | |
4944 | * somewhere. If none of the target zones can satisfy our allocation | |
4945 | * request even if all reclaimable pages are considered then we are | |
4946 | * screwed and have to go OOM. | |
0a0337e0 | 4947 | */ |
97a225e6 JK |
4948 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4949 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4950 | unsigned long available; |
ede37713 | 4951 | unsigned long reclaimable; |
d379f01d MH |
4952 | unsigned long min_wmark = min_wmark_pages(zone); |
4953 | bool wmark; | |
0a0337e0 | 4954 | |
5a1c84b4 | 4955 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4956 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4957 | |
4958 | /* | |
491d79ae JW |
4959 | * Would the allocation succeed if we reclaimed all |
4960 | * reclaimable pages? | |
0a0337e0 | 4961 | */ |
d379f01d | 4962 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 4963 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
4964 | trace_reclaim_retry_zone(z, order, reclaimable, |
4965 | available, min_wmark, *no_progress_loops, wmark); | |
4966 | if (wmark) { | |
15f570bf | 4967 | ret = true; |
132b0d21 | 4968 | break; |
0a0337e0 MH |
4969 | } |
4970 | } | |
4971 | ||
15f570bf MH |
4972 | /* |
4973 | * Memory allocation/reclaim might be called from a WQ context and the | |
4974 | * current implementation of the WQ concurrency control doesn't | |
4975 | * recognize that a particular WQ is congested if the worker thread is | |
4976 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4977 | * here rather than calling cond_resched(). | |
4978 | */ | |
4979 | if (current->flags & PF_WQ_WORKER) | |
4980 | schedule_timeout_uninterruptible(1); | |
4981 | else | |
4982 | cond_resched(); | |
4983 | return ret; | |
0a0337e0 MH |
4984 | } |
4985 | ||
902b6281 VB |
4986 | static inline bool |
4987 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4988 | { | |
4989 | /* | |
4990 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4991 | * mempolicy don't intersect. This should be normally dealt with by | |
4992 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4993 | * such a way the check therein was true, and then it became false | |
4994 | * before we got our cpuset_mems_cookie here. | |
4995 | * This assumes that for all allocations, ac->nodemask can come only | |
4996 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4997 | * when it does not intersect with the cpuset restrictions) or the | |
4998 | * caller can deal with a violated nodemask. | |
4999 | */ | |
5000 | if (cpusets_enabled() && ac->nodemask && | |
5001 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
5002 | ac->nodemask = NULL; | |
5003 | return true; | |
5004 | } | |
5005 | ||
5006 | /* | |
5007 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
5008 | * possible to race with parallel threads in such a way that our | |
5009 | * allocation can fail while the mask is being updated. If we are about | |
5010 | * to fail, check if the cpuset changed during allocation and if so, | |
5011 | * retry. | |
5012 | */ | |
5013 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
5014 | return true; | |
5015 | ||
5016 | return false; | |
5017 | } | |
5018 | ||
11e33f6a MG |
5019 | static inline struct page * |
5020 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 5021 | struct alloc_context *ac) |
11e33f6a | 5022 | { |
d0164adc | 5023 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 5024 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 5025 | struct page *page = NULL; |
c603844b | 5026 | unsigned int alloc_flags; |
11e33f6a | 5027 | unsigned long did_some_progress; |
5ce9bfef | 5028 | enum compact_priority compact_priority; |
c5d01d0d | 5029 | enum compact_result compact_result; |
5ce9bfef VB |
5030 | int compaction_retries; |
5031 | int no_progress_loops; | |
5ce9bfef | 5032 | unsigned int cpuset_mems_cookie; |
3d36424b | 5033 | unsigned int zonelist_iter_cookie; |
cd04ae1e | 5034 | int reserve_flags; |
1da177e4 | 5035 | |
d0164adc MG |
5036 | /* |
5037 | * We also sanity check to catch abuse of atomic reserves being used by | |
5038 | * callers that are not in atomic context. | |
5039 | */ | |
5040 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
5041 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
5042 | gfp_mask &= ~__GFP_ATOMIC; | |
5043 | ||
3d36424b | 5044 | restart: |
5ce9bfef VB |
5045 | compaction_retries = 0; |
5046 | no_progress_loops = 0; | |
5047 | compact_priority = DEF_COMPACT_PRIORITY; | |
5048 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
3d36424b | 5049 | zonelist_iter_cookie = zonelist_iter_begin(); |
9a67f648 MH |
5050 | |
5051 | /* | |
5052 | * The fast path uses conservative alloc_flags to succeed only until | |
5053 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
5054 | * alloc_flags precisely. So we do that now. | |
5055 | */ | |
5056 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
5057 | ||
e47483bc VB |
5058 | /* |
5059 | * We need to recalculate the starting point for the zonelist iterator | |
5060 | * because we might have used different nodemask in the fast path, or | |
5061 | * there was a cpuset modification and we are retrying - otherwise we | |
5062 | * could end up iterating over non-eligible zones endlessly. | |
5063 | */ | |
5064 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 5065 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
5066 | if (!ac->preferred_zoneref->zone) |
5067 | goto nopage; | |
5068 | ||
8ca1b5a4 FT |
5069 | /* |
5070 | * Check for insane configurations where the cpuset doesn't contain | |
5071 | * any suitable zone to satisfy the request - e.g. non-movable | |
5072 | * GFP_HIGHUSER allocations from MOVABLE nodes only. | |
5073 | */ | |
5074 | if (cpusets_insane_config() && (gfp_mask & __GFP_HARDWALL)) { | |
5075 | struct zoneref *z = first_zones_zonelist(ac->zonelist, | |
5076 | ac->highest_zoneidx, | |
5077 | &cpuset_current_mems_allowed); | |
5078 | if (!z->zone) | |
5079 | goto nopage; | |
5080 | } | |
5081 | ||
0a79cdad | 5082 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 5083 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
5084 | |
5085 | /* | |
5086 | * The adjusted alloc_flags might result in immediate success, so try | |
5087 | * that first | |
5088 | */ | |
5089 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
5090 | if (page) | |
5091 | goto got_pg; | |
5092 | ||
a8161d1e VB |
5093 | /* |
5094 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
5095 | * that we have enough base pages and don't need to reclaim. For non- |
5096 | * movable high-order allocations, do that as well, as compaction will | |
5097 | * try prevent permanent fragmentation by migrating from blocks of the | |
5098 | * same migratetype. | |
5099 | * Don't try this for allocations that are allowed to ignore | |
5100 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 5101 | */ |
282722b0 VB |
5102 | if (can_direct_reclaim && |
5103 | (costly_order || | |
5104 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
5105 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
5106 | page = __alloc_pages_direct_compact(gfp_mask, order, |
5107 | alloc_flags, ac, | |
a5508cd8 | 5108 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
5109 | &compact_result); |
5110 | if (page) | |
5111 | goto got_pg; | |
5112 | ||
cc638f32 VB |
5113 | /* |
5114 | * Checks for costly allocations with __GFP_NORETRY, which | |
5115 | * includes some THP page fault allocations | |
5116 | */ | |
5117 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
5118 | /* |
5119 | * If allocating entire pageblock(s) and compaction | |
5120 | * failed because all zones are below low watermarks | |
5121 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
5122 | * order, fail immediately unless the allocator has |
5123 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
5124 | * |
5125 | * Reclaim is | |
5126 | * - potentially very expensive because zones are far | |
5127 | * below their low watermarks or this is part of very | |
5128 | * bursty high order allocations, | |
5129 | * - not guaranteed to help because isolate_freepages() | |
5130 | * may not iterate over freed pages as part of its | |
5131 | * linear scan, and | |
5132 | * - unlikely to make entire pageblocks free on its | |
5133 | * own. | |
5134 | */ | |
5135 | if (compact_result == COMPACT_SKIPPED || | |
5136 | compact_result == COMPACT_DEFERRED) | |
5137 | goto nopage; | |
a8161d1e | 5138 | |
a8161d1e | 5139 | /* |
3eb2771b VB |
5140 | * Looks like reclaim/compaction is worth trying, but |
5141 | * sync compaction could be very expensive, so keep | |
25160354 | 5142 | * using async compaction. |
a8161d1e | 5143 | */ |
a5508cd8 | 5144 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
5145 | } |
5146 | } | |
23771235 | 5147 | |
31a6c190 | 5148 | retry: |
23771235 | 5149 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 5150 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 5151 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 5152 | |
cd04ae1e MH |
5153 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
5154 | if (reserve_flags) | |
ce96fa62 ML |
5155 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags) | |
5156 | (alloc_flags & ALLOC_KSWAPD); | |
23771235 | 5157 | |
e46e7b77 | 5158 | /* |
d6a24df0 VB |
5159 | * Reset the nodemask and zonelist iterators if memory policies can be |
5160 | * ignored. These allocations are high priority and system rather than | |
5161 | * user oriented. | |
e46e7b77 | 5162 | */ |
cd04ae1e | 5163 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 5164 | ac->nodemask = NULL; |
e46e7b77 | 5165 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 5166 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
5167 | } |
5168 | ||
23771235 | 5169 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 5170 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
5171 | if (page) |
5172 | goto got_pg; | |
1da177e4 | 5173 | |
d0164adc | 5174 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 5175 | if (!can_direct_reclaim) |
1da177e4 LT |
5176 | goto nopage; |
5177 | ||
9a67f648 MH |
5178 | /* Avoid recursion of direct reclaim */ |
5179 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
5180 | goto nopage; |
5181 | ||
a8161d1e VB |
5182 | /* Try direct reclaim and then allocating */ |
5183 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
5184 | &did_some_progress); | |
5185 | if (page) | |
5186 | goto got_pg; | |
5187 | ||
5188 | /* Try direct compaction and then allocating */ | |
a9263751 | 5189 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 5190 | compact_priority, &compact_result); |
56de7263 MG |
5191 | if (page) |
5192 | goto got_pg; | |
75f30861 | 5193 | |
9083905a JW |
5194 | /* Do not loop if specifically requested */ |
5195 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 5196 | goto nopage; |
9083905a | 5197 | |
0a0337e0 MH |
5198 | /* |
5199 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 5200 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 5201 | */ |
dcda9b04 | 5202 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 5203 | goto nopage; |
0a0337e0 | 5204 | |
0a0337e0 | 5205 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 5206 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
5207 | goto retry; |
5208 | ||
33c2d214 MH |
5209 | /* |
5210 | * It doesn't make any sense to retry for the compaction if the order-0 | |
5211 | * reclaim is not able to make any progress because the current | |
5212 | * implementation of the compaction depends on the sufficient amount | |
5213 | * of free memory (see __compaction_suitable) | |
5214 | */ | |
5215 | if (did_some_progress > 0 && | |
86a294a8 | 5216 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 5217 | compact_result, &compact_priority, |
d9436498 | 5218 | &compaction_retries)) |
33c2d214 MH |
5219 | goto retry; |
5220 | ||
902b6281 | 5221 | |
3d36424b MG |
5222 | /* |
5223 | * Deal with possible cpuset update races or zonelist updates to avoid | |
5224 | * a unnecessary OOM kill. | |
5225 | */ | |
5226 | if (check_retry_cpuset(cpuset_mems_cookie, ac) || | |
5227 | check_retry_zonelist(zonelist_iter_cookie)) | |
5228 | goto restart; | |
e47483bc | 5229 | |
9083905a JW |
5230 | /* Reclaim has failed us, start killing things */ |
5231 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
5232 | if (page) | |
5233 | goto got_pg; | |
5234 | ||
9a67f648 | 5235 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e | 5236 | if (tsk_is_oom_victim(current) && |
8510e69c | 5237 | (alloc_flags & ALLOC_OOM || |
c288983d | 5238 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
5239 | goto nopage; |
5240 | ||
9083905a | 5241 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
5242 | if (did_some_progress) { |
5243 | no_progress_loops = 0; | |
9083905a | 5244 | goto retry; |
0a0337e0 | 5245 | } |
9083905a | 5246 | |
1da177e4 | 5247 | nopage: |
3d36424b MG |
5248 | /* |
5249 | * Deal with possible cpuset update races or zonelist updates to avoid | |
5250 | * a unnecessary OOM kill. | |
5251 | */ | |
5252 | if (check_retry_cpuset(cpuset_mems_cookie, ac) || | |
5253 | check_retry_zonelist(zonelist_iter_cookie)) | |
5254 | goto restart; | |
5ce9bfef | 5255 | |
9a67f648 MH |
5256 | /* |
5257 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
5258 | * we always retry | |
5259 | */ | |
5260 | if (gfp_mask & __GFP_NOFAIL) { | |
5261 | /* | |
5262 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
5263 | * of any new users that actually require GFP_NOWAIT | |
5264 | */ | |
3f913fc5 | 5265 | if (WARN_ON_ONCE_GFP(!can_direct_reclaim, gfp_mask)) |
9a67f648 MH |
5266 | goto fail; |
5267 | ||
5268 | /* | |
5269 | * PF_MEMALLOC request from this context is rather bizarre | |
5270 | * because we cannot reclaim anything and only can loop waiting | |
5271 | * for somebody to do a work for us | |
5272 | */ | |
3f913fc5 | 5273 | WARN_ON_ONCE_GFP(current->flags & PF_MEMALLOC, gfp_mask); |
9a67f648 MH |
5274 | |
5275 | /* | |
5276 | * non failing costly orders are a hard requirement which we | |
5277 | * are not prepared for much so let's warn about these users | |
5278 | * so that we can identify them and convert them to something | |
5279 | * else. | |
5280 | */ | |
896c4d52 | 5281 | WARN_ON_ONCE_GFP(costly_order, gfp_mask); |
9a67f648 | 5282 | |
6c18ba7a MH |
5283 | /* |
5284 | * Help non-failing allocations by giving them access to memory | |
5285 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
5286 | * could deplete whole memory reserves which would just make | |
5287 | * the situation worse | |
5288 | */ | |
5289 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
5290 | if (page) | |
5291 | goto got_pg; | |
5292 | ||
9a67f648 MH |
5293 | cond_resched(); |
5294 | goto retry; | |
5295 | } | |
5296 | fail: | |
a8e99259 | 5297 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 5298 | "page allocation failure: order:%u", order); |
1da177e4 | 5299 | got_pg: |
072bb0aa | 5300 | return page; |
1da177e4 | 5301 | } |
11e33f6a | 5302 | |
9cd75558 | 5303 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 5304 | int preferred_nid, nodemask_t *nodemask, |
8e6a930b | 5305 | struct alloc_context *ac, gfp_t *alloc_gfp, |
9cd75558 | 5306 | unsigned int *alloc_flags) |
11e33f6a | 5307 | { |
97a225e6 | 5308 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 5309 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 5310 | ac->nodemask = nodemask; |
01c0bfe0 | 5311 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 5312 | |
682a3385 | 5313 | if (cpusets_enabled()) { |
8e6a930b | 5314 | *alloc_gfp |= __GFP_HARDWALL; |
182f3d7a MS |
5315 | /* |
5316 | * When we are in the interrupt context, it is irrelevant | |
5317 | * to the current task context. It means that any node ok. | |
5318 | */ | |
88dc6f20 | 5319 | if (in_task() && !ac->nodemask) |
9cd75558 | 5320 | ac->nodemask = &cpuset_current_mems_allowed; |
51047820 VB |
5321 | else |
5322 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
5323 | } |
5324 | ||
446ec838 | 5325 | might_alloc(gfp_mask); |
11e33f6a MG |
5326 | |
5327 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 5328 | return false; |
11e33f6a | 5329 | |
8e3560d9 | 5330 | *alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags); |
d883c6cf | 5331 | |
c9ab0c4f | 5332 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 5333 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 5334 | |
e46e7b77 MG |
5335 | /* |
5336 | * The preferred zone is used for statistics but crucially it is | |
5337 | * also used as the starting point for the zonelist iterator. It | |
5338 | * may get reset for allocations that ignore memory policies. | |
5339 | */ | |
9cd75558 | 5340 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 5341 | ac->highest_zoneidx, ac->nodemask); |
a0622d05 MN |
5342 | |
5343 | return true; | |
9cd75558 MG |
5344 | } |
5345 | ||
387ba26f | 5346 | /* |
0f87d9d3 | 5347 | * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array |
387ba26f MG |
5348 | * @gfp: GFP flags for the allocation |
5349 | * @preferred_nid: The preferred NUMA node ID to allocate from | |
5350 | * @nodemask: Set of nodes to allocate from, may be NULL | |
0f87d9d3 MG |
5351 | * @nr_pages: The number of pages desired on the list or array |
5352 | * @page_list: Optional list to store the allocated pages | |
5353 | * @page_array: Optional array to store the pages | |
387ba26f MG |
5354 | * |
5355 | * This is a batched version of the page allocator that attempts to | |
0f87d9d3 MG |
5356 | * allocate nr_pages quickly. Pages are added to page_list if page_list |
5357 | * is not NULL, otherwise it is assumed that the page_array is valid. | |
387ba26f | 5358 | * |
0f87d9d3 MG |
5359 | * For lists, nr_pages is the number of pages that should be allocated. |
5360 | * | |
5361 | * For arrays, only NULL elements are populated with pages and nr_pages | |
5362 | * is the maximum number of pages that will be stored in the array. | |
5363 | * | |
5364 | * Returns the number of pages on the list or array. | |
387ba26f MG |
5365 | */ |
5366 | unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, | |
5367 | nodemask_t *nodemask, int nr_pages, | |
0f87d9d3 MG |
5368 | struct list_head *page_list, |
5369 | struct page **page_array) | |
387ba26f MG |
5370 | { |
5371 | struct page *page; | |
5372 | unsigned long flags; | |
4b23a68f | 5373 | unsigned long __maybe_unused UP_flags; |
387ba26f MG |
5374 | struct zone *zone; |
5375 | struct zoneref *z; | |
5376 | struct per_cpu_pages *pcp; | |
5377 | struct list_head *pcp_list; | |
5378 | struct alloc_context ac; | |
5379 | gfp_t alloc_gfp; | |
5380 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
3e23060b | 5381 | int nr_populated = 0, nr_account = 0; |
387ba26f | 5382 | |
0f87d9d3 MG |
5383 | /* |
5384 | * Skip populated array elements to determine if any pages need | |
5385 | * to be allocated before disabling IRQs. | |
5386 | */ | |
b08e50dd | 5387 | while (page_array && nr_populated < nr_pages && page_array[nr_populated]) |
0f87d9d3 MG |
5388 | nr_populated++; |
5389 | ||
06147843 CL |
5390 | /* No pages requested? */ |
5391 | if (unlikely(nr_pages <= 0)) | |
5392 | goto out; | |
5393 | ||
b3b64ebd MG |
5394 | /* Already populated array? */ |
5395 | if (unlikely(page_array && nr_pages - nr_populated == 0)) | |
06147843 | 5396 | goto out; |
b3b64ebd | 5397 | |
8dcb3060 SB |
5398 | /* Bulk allocator does not support memcg accounting. */ |
5399 | if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT)) | |
5400 | goto failed; | |
5401 | ||
387ba26f | 5402 | /* Use the single page allocator for one page. */ |
0f87d9d3 | 5403 | if (nr_pages - nr_populated == 1) |
387ba26f MG |
5404 | goto failed; |
5405 | ||
187ad460 MG |
5406 | #ifdef CONFIG_PAGE_OWNER |
5407 | /* | |
5408 | * PAGE_OWNER may recurse into the allocator to allocate space to | |
5409 | * save the stack with pagesets.lock held. Releasing/reacquiring | |
5410 | * removes much of the performance benefit of bulk allocation so | |
5411 | * force the caller to allocate one page at a time as it'll have | |
5412 | * similar performance to added complexity to the bulk allocator. | |
5413 | */ | |
5414 | if (static_branch_unlikely(&page_owner_inited)) | |
5415 | goto failed; | |
5416 | #endif | |
5417 | ||
387ba26f MG |
5418 | /* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */ |
5419 | gfp &= gfp_allowed_mask; | |
5420 | alloc_gfp = gfp; | |
5421 | if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags)) | |
06147843 | 5422 | goto out; |
387ba26f MG |
5423 | gfp = alloc_gfp; |
5424 | ||
5425 | /* Find an allowed local zone that meets the low watermark. */ | |
5426 | for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) { | |
5427 | unsigned long mark; | |
5428 | ||
5429 | if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) && | |
5430 | !__cpuset_zone_allowed(zone, gfp)) { | |
5431 | continue; | |
5432 | } | |
5433 | ||
5434 | if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone && | |
5435 | zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) { | |
5436 | goto failed; | |
5437 | } | |
5438 | ||
5439 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages; | |
5440 | if (zone_watermark_fast(zone, 0, mark, | |
5441 | zonelist_zone_idx(ac.preferred_zoneref), | |
5442 | alloc_flags, gfp)) { | |
5443 | break; | |
5444 | } | |
5445 | } | |
5446 | ||
5447 | /* | |
5448 | * If there are no allowed local zones that meets the watermarks then | |
5449 | * try to allocate a single page and reclaim if necessary. | |
5450 | */ | |
ce76f9a1 | 5451 | if (unlikely(!zone)) |
387ba26f MG |
5452 | goto failed; |
5453 | ||
4b23a68f | 5454 | /* Is a parallel drain in progress? */ |
4b23a68f | 5455 | pcp_trylock_prepare(UP_flags); |
01b44456 MG |
5456 | pcp = pcp_spin_trylock_irqsave(zone->per_cpu_pageset, flags); |
5457 | if (!pcp) | |
4b23a68f | 5458 | goto failed_irq; |
387ba26f | 5459 | |
387ba26f | 5460 | /* Attempt the batch allocation */ |
44042b44 | 5461 | pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)]; |
0f87d9d3 MG |
5462 | while (nr_populated < nr_pages) { |
5463 | ||
5464 | /* Skip existing pages */ | |
5465 | if (page_array && page_array[nr_populated]) { | |
5466 | nr_populated++; | |
5467 | continue; | |
5468 | } | |
5469 | ||
44042b44 | 5470 | page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags, |
387ba26f | 5471 | pcp, pcp_list); |
ce76f9a1 | 5472 | if (unlikely(!page)) { |
c572e488 | 5473 | /* Try and allocate at least one page */ |
4b23a68f | 5474 | if (!nr_account) { |
01b44456 | 5475 | pcp_spin_unlock_irqrestore(pcp, flags); |
387ba26f | 5476 | goto failed_irq; |
4b23a68f | 5477 | } |
387ba26f MG |
5478 | break; |
5479 | } | |
3e23060b | 5480 | nr_account++; |
387ba26f MG |
5481 | |
5482 | prep_new_page(page, 0, gfp, 0); | |
0f87d9d3 MG |
5483 | if (page_list) |
5484 | list_add(&page->lru, page_list); | |
5485 | else | |
5486 | page_array[nr_populated] = page; | |
5487 | nr_populated++; | |
387ba26f MG |
5488 | } |
5489 | ||
01b44456 | 5490 | pcp_spin_unlock_irqrestore(pcp, flags); |
4b23a68f | 5491 | pcp_trylock_finish(UP_flags); |
43c95bcc | 5492 | |
3e23060b MG |
5493 | __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account); |
5494 | zone_statistics(ac.preferred_zoneref->zone, zone, nr_account); | |
387ba26f | 5495 | |
06147843 | 5496 | out: |
0f87d9d3 | 5497 | return nr_populated; |
387ba26f MG |
5498 | |
5499 | failed_irq: | |
4b23a68f | 5500 | pcp_trylock_finish(UP_flags); |
387ba26f MG |
5501 | |
5502 | failed: | |
5503 | page = __alloc_pages(gfp, 0, preferred_nid, nodemask); | |
5504 | if (page) { | |
0f87d9d3 MG |
5505 | if (page_list) |
5506 | list_add(&page->lru, page_list); | |
5507 | else | |
5508 | page_array[nr_populated] = page; | |
5509 | nr_populated++; | |
387ba26f MG |
5510 | } |
5511 | ||
06147843 | 5512 | goto out; |
387ba26f MG |
5513 | } |
5514 | EXPORT_SYMBOL_GPL(__alloc_pages_bulk); | |
5515 | ||
9cd75558 MG |
5516 | /* |
5517 | * This is the 'heart' of the zoned buddy allocator. | |
5518 | */ | |
84172f4b | 5519 | struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, |
04ec6264 | 5520 | nodemask_t *nodemask) |
9cd75558 MG |
5521 | { |
5522 | struct page *page; | |
5523 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
8e6a930b | 5524 | gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
5525 | struct alloc_context ac = { }; |
5526 | ||
c63ae43b MH |
5527 | /* |
5528 | * There are several places where we assume that the order value is sane | |
5529 | * so bail out early if the request is out of bound. | |
5530 | */ | |
3f913fc5 | 5531 | if (WARN_ON_ONCE_GFP(order >= MAX_ORDER, gfp)) |
c63ae43b | 5532 | return NULL; |
c63ae43b | 5533 | |
6e5e0f28 | 5534 | gfp &= gfp_allowed_mask; |
da6df1b0 PT |
5535 | /* |
5536 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS | |
5537 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
5538 | * from a particular context which has been marked by | |
8e3560d9 PT |
5539 | * memalloc_no{fs,io}_{save,restore}. And PF_MEMALLOC_PIN which ensures |
5540 | * movable zones are not used during allocation. | |
da6df1b0 PT |
5541 | */ |
5542 | gfp = current_gfp_context(gfp); | |
6e5e0f28 MWO |
5543 | alloc_gfp = gfp; |
5544 | if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac, | |
8e6a930b | 5545 | &alloc_gfp, &alloc_flags)) |
9cd75558 MG |
5546 | return NULL; |
5547 | ||
6bb15450 MG |
5548 | /* |
5549 | * Forbid the first pass from falling back to types that fragment | |
5550 | * memory until all local zones are considered. | |
5551 | */ | |
6e5e0f28 | 5552 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp); |
6bb15450 | 5553 | |
5117f45d | 5554 | /* First allocation attempt */ |
8e6a930b | 5555 | page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac); |
4fcb0971 MG |
5556 | if (likely(page)) |
5557 | goto out; | |
11e33f6a | 5558 | |
da6df1b0 | 5559 | alloc_gfp = gfp; |
4fcb0971 | 5560 | ac.spread_dirty_pages = false; |
23f086f9 | 5561 | |
4741526b MG |
5562 | /* |
5563 | * Restore the original nodemask if it was potentially replaced with | |
5564 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
5565 | */ | |
97ce86f9 | 5566 | ac.nodemask = nodemask; |
16096c25 | 5567 | |
8e6a930b | 5568 | page = __alloc_pages_slowpath(alloc_gfp, order, &ac); |
cc9a6c87 | 5569 | |
4fcb0971 | 5570 | out: |
6e5e0f28 MWO |
5571 | if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT) && page && |
5572 | unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) { | |
c4159a75 VD |
5573 | __free_pages(page, order); |
5574 | page = NULL; | |
4949148a VD |
5575 | } |
5576 | ||
8e6a930b | 5577 | trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype); |
b073d7f8 | 5578 | kmsan_alloc_page(page, order, alloc_gfp); |
4fcb0971 | 5579 | |
11e33f6a | 5580 | return page; |
1da177e4 | 5581 | } |
84172f4b | 5582 | EXPORT_SYMBOL(__alloc_pages); |
1da177e4 | 5583 | |
cc09cb13 MWO |
5584 | struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid, |
5585 | nodemask_t *nodemask) | |
5586 | { | |
5587 | struct page *page = __alloc_pages(gfp | __GFP_COMP, order, | |
5588 | preferred_nid, nodemask); | |
5589 | ||
5590 | if (page && order > 1) | |
5591 | prep_transhuge_page(page); | |
5592 | return (struct folio *)page; | |
5593 | } | |
5594 | EXPORT_SYMBOL(__folio_alloc); | |
5595 | ||
1da177e4 | 5596 | /* |
9ea9a680 MH |
5597 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
5598 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
5599 | * you need to access high mem. | |
1da177e4 | 5600 | */ |
920c7a5d | 5601 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 5602 | { |
945a1113 AM |
5603 | struct page *page; |
5604 | ||
9ea9a680 | 5605 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
5606 | if (!page) |
5607 | return 0; | |
5608 | return (unsigned long) page_address(page); | |
5609 | } | |
1da177e4 LT |
5610 | EXPORT_SYMBOL(__get_free_pages); |
5611 | ||
920c7a5d | 5612 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 5613 | { |
945a1113 | 5614 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 5615 | } |
1da177e4 LT |
5616 | EXPORT_SYMBOL(get_zeroed_page); |
5617 | ||
7f194fbb MWO |
5618 | /** |
5619 | * __free_pages - Free pages allocated with alloc_pages(). | |
5620 | * @page: The page pointer returned from alloc_pages(). | |
5621 | * @order: The order of the allocation. | |
5622 | * | |
5623 | * This function can free multi-page allocations that are not compound | |
5624 | * pages. It does not check that the @order passed in matches that of | |
5625 | * the allocation, so it is easy to leak memory. Freeing more memory | |
5626 | * than was allocated will probably emit a warning. | |
5627 | * | |
5628 | * If the last reference to this page is speculative, it will be released | |
5629 | * by put_page() which only frees the first page of a non-compound | |
5630 | * allocation. To prevent the remaining pages from being leaked, we free | |
5631 | * the subsequent pages here. If you want to use the page's reference | |
5632 | * count to decide when to free the allocation, you should allocate a | |
5633 | * compound page, and use put_page() instead of __free_pages(). | |
5634 | * | |
5635 | * Context: May be called in interrupt context or while holding a normal | |
5636 | * spinlock, but not in NMI context or while holding a raw spinlock. | |
5637 | */ | |
742aa7fb AL |
5638 | void __free_pages(struct page *page, unsigned int order) |
5639 | { | |
5640 | if (put_page_testzero(page)) | |
5641 | free_the_page(page, order); | |
e320d301 MWO |
5642 | else if (!PageHead(page)) |
5643 | while (order-- > 0) | |
5644 | free_the_page(page + (1 << order), order); | |
742aa7fb | 5645 | } |
1da177e4 LT |
5646 | EXPORT_SYMBOL(__free_pages); |
5647 | ||
920c7a5d | 5648 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
5649 | { |
5650 | if (addr != 0) { | |
725d704e | 5651 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
5652 | __free_pages(virt_to_page((void *)addr), order); |
5653 | } | |
5654 | } | |
5655 | ||
5656 | EXPORT_SYMBOL(free_pages); | |
5657 | ||
b63ae8ca AD |
5658 | /* |
5659 | * Page Fragment: | |
5660 | * An arbitrary-length arbitrary-offset area of memory which resides | |
5661 | * within a 0 or higher order page. Multiple fragments within that page | |
5662 | * are individually refcounted, in the page's reference counter. | |
5663 | * | |
5664 | * The page_frag functions below provide a simple allocation framework for | |
5665 | * page fragments. This is used by the network stack and network device | |
5666 | * drivers to provide a backing region of memory for use as either an | |
5667 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
5668 | */ | |
2976db80 AD |
5669 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
5670 | gfp_t gfp_mask) | |
b63ae8ca AD |
5671 | { |
5672 | struct page *page = NULL; | |
5673 | gfp_t gfp = gfp_mask; | |
5674 | ||
5675 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5676 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
5677 | __GFP_NOMEMALLOC; | |
5678 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
5679 | PAGE_FRAG_CACHE_MAX_ORDER); | |
5680 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
5681 | #endif | |
5682 | if (unlikely(!page)) | |
5683 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
5684 | ||
5685 | nc->va = page ? page_address(page) : NULL; | |
5686 | ||
5687 | return page; | |
5688 | } | |
5689 | ||
2976db80 | 5690 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
5691 | { |
5692 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
5693 | ||
742aa7fb AL |
5694 | if (page_ref_sub_and_test(page, count)) |
5695 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 5696 | } |
2976db80 | 5697 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 5698 | |
b358e212 KH |
5699 | void *page_frag_alloc_align(struct page_frag_cache *nc, |
5700 | unsigned int fragsz, gfp_t gfp_mask, | |
5701 | unsigned int align_mask) | |
b63ae8ca AD |
5702 | { |
5703 | unsigned int size = PAGE_SIZE; | |
5704 | struct page *page; | |
5705 | int offset; | |
5706 | ||
5707 | if (unlikely(!nc->va)) { | |
5708 | refill: | |
2976db80 | 5709 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
5710 | if (!page) |
5711 | return NULL; | |
5712 | ||
5713 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5714 | /* if size can vary use size else just use PAGE_SIZE */ | |
5715 | size = nc->size; | |
5716 | #endif | |
5717 | /* Even if we own the page, we do not use atomic_set(). | |
5718 | * This would break get_page_unless_zero() users. | |
5719 | */ | |
86447726 | 5720 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
5721 | |
5722 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 5723 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 5724 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5725 | nc->offset = size; |
5726 | } | |
5727 | ||
5728 | offset = nc->offset - fragsz; | |
5729 | if (unlikely(offset < 0)) { | |
5730 | page = virt_to_page(nc->va); | |
5731 | ||
fe896d18 | 5732 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
5733 | goto refill; |
5734 | ||
d8c19014 DZ |
5735 | if (unlikely(nc->pfmemalloc)) { |
5736 | free_the_page(page, compound_order(page)); | |
5737 | goto refill; | |
5738 | } | |
5739 | ||
b63ae8ca AD |
5740 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
5741 | /* if size can vary use size else just use PAGE_SIZE */ | |
5742 | size = nc->size; | |
5743 | #endif | |
5744 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 5745 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
5746 | |
5747 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 5748 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca | 5749 | offset = size - fragsz; |
dac22531 ML |
5750 | if (unlikely(offset < 0)) { |
5751 | /* | |
5752 | * The caller is trying to allocate a fragment | |
5753 | * with fragsz > PAGE_SIZE but the cache isn't big | |
5754 | * enough to satisfy the request, this may | |
5755 | * happen in low memory conditions. | |
5756 | * We don't release the cache page because | |
5757 | * it could make memory pressure worse | |
5758 | * so we simply return NULL here. | |
5759 | */ | |
5760 | return NULL; | |
5761 | } | |
b63ae8ca AD |
5762 | } |
5763 | ||
5764 | nc->pagecnt_bias--; | |
b358e212 | 5765 | offset &= align_mask; |
b63ae8ca AD |
5766 | nc->offset = offset; |
5767 | ||
5768 | return nc->va + offset; | |
5769 | } | |
b358e212 | 5770 | EXPORT_SYMBOL(page_frag_alloc_align); |
b63ae8ca AD |
5771 | |
5772 | /* | |
5773 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
5774 | */ | |
8c2dd3e4 | 5775 | void page_frag_free(void *addr) |
b63ae8ca AD |
5776 | { |
5777 | struct page *page = virt_to_head_page(addr); | |
5778 | ||
742aa7fb AL |
5779 | if (unlikely(put_page_testzero(page))) |
5780 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 5781 | } |
8c2dd3e4 | 5782 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 5783 | |
d00181b9 KS |
5784 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
5785 | size_t size) | |
ee85c2e1 AK |
5786 | { |
5787 | if (addr) { | |
df48a5f7 LH |
5788 | unsigned long nr = DIV_ROUND_UP(size, PAGE_SIZE); |
5789 | struct page *page = virt_to_page((void *)addr); | |
5790 | struct page *last = page + nr; | |
5791 | ||
5792 | split_page_owner(page, 1 << order); | |
5793 | split_page_memcg(page, 1 << order); | |
5794 | while (page < --last) | |
5795 | set_page_refcounted(last); | |
5796 | ||
5797 | last = page + (1UL << order); | |
5798 | for (page += nr; page < last; page++) | |
5799 | __free_pages_ok(page, 0, FPI_TO_TAIL); | |
ee85c2e1 AK |
5800 | } |
5801 | return (void *)addr; | |
5802 | } | |
5803 | ||
2be0ffe2 TT |
5804 | /** |
5805 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
5806 | * @size: the number of bytes to allocate | |
63931eb9 | 5807 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
5808 | * |
5809 | * This function is similar to alloc_pages(), except that it allocates the | |
5810 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
5811 | * allocate memory in power-of-two pages. | |
5812 | * | |
5813 | * This function is also limited by MAX_ORDER. | |
5814 | * | |
5815 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
5816 | * |
5817 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
5818 | */ |
5819 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
5820 | { | |
5821 | unsigned int order = get_order(size); | |
5822 | unsigned long addr; | |
5823 | ||
ba7f1b9e ML |
5824 | if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM))) |
5825 | gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM); | |
63931eb9 | 5826 | |
2be0ffe2 | 5827 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 5828 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
5829 | } |
5830 | EXPORT_SYMBOL(alloc_pages_exact); | |
5831 | ||
ee85c2e1 AK |
5832 | /** |
5833 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
5834 | * pages on a node. | |
b5e6ab58 | 5835 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 5836 | * @size: the number of bytes to allocate |
63931eb9 | 5837 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
5838 | * |
5839 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
5840 | * back. | |
a862f68a MR |
5841 | * |
5842 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 5843 | */ |
e1931811 | 5844 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5845 | { |
d00181b9 | 5846 | unsigned int order = get_order(size); |
63931eb9 VB |
5847 | struct page *p; |
5848 | ||
ba7f1b9e ML |
5849 | if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM))) |
5850 | gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM); | |
63931eb9 VB |
5851 | |
5852 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5853 | if (!p) |
5854 | return NULL; | |
5855 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5856 | } | |
ee85c2e1 | 5857 | |
2be0ffe2 TT |
5858 | /** |
5859 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5860 | * @virt: the value returned by alloc_pages_exact. | |
5861 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5862 | * | |
5863 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5864 | */ | |
5865 | void free_pages_exact(void *virt, size_t size) | |
5866 | { | |
5867 | unsigned long addr = (unsigned long)virt; | |
5868 | unsigned long end = addr + PAGE_ALIGN(size); | |
5869 | ||
5870 | while (addr < end) { | |
5871 | free_page(addr); | |
5872 | addr += PAGE_SIZE; | |
5873 | } | |
5874 | } | |
5875 | EXPORT_SYMBOL(free_pages_exact); | |
5876 | ||
e0fb5815 ZY |
5877 | /** |
5878 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5879 | * @offset: The zone index of the highest zone | |
5880 | * | |
a862f68a | 5881 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5882 | * high watermark within all zones at or below a given zone index. For each |
5883 | * zone, the number of pages is calculated as: | |
0e056eb5 | 5884 | * |
5885 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5886 | * |
5887 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5888 | */ |
ebec3862 | 5889 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5890 | { |
dd1a239f | 5891 | struct zoneref *z; |
54a6eb5c MG |
5892 | struct zone *zone; |
5893 | ||
e310fd43 | 5894 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5895 | unsigned long sum = 0; |
1da177e4 | 5896 | |
0e88460d | 5897 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5898 | |
54a6eb5c | 5899 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5900 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5901 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5902 | if (size > high) |
5903 | sum += size - high; | |
1da177e4 LT |
5904 | } |
5905 | ||
5906 | return sum; | |
5907 | } | |
5908 | ||
e0fb5815 ZY |
5909 | /** |
5910 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5911 | * | |
5912 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5913 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5914 | * |
5915 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5916 | * ZONE_NORMAL. | |
1da177e4 | 5917 | */ |
ebec3862 | 5918 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5919 | { |
af4ca457 | 5920 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5921 | } |
c2f1a551 | 5922 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5923 | |
08e0f6a9 | 5924 | static inline void show_node(struct zone *zone) |
1da177e4 | 5925 | { |
e5adfffc | 5926 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5927 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5928 | } |
1da177e4 | 5929 | |
d02bd27b IR |
5930 | long si_mem_available(void) |
5931 | { | |
5932 | long available; | |
5933 | unsigned long pagecache; | |
5934 | unsigned long wmark_low = 0; | |
5935 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5936 | unsigned long reclaimable; |
d02bd27b IR |
5937 | struct zone *zone; |
5938 | int lru; | |
5939 | ||
5940 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5941 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5942 | |
5943 | for_each_zone(zone) | |
a9214443 | 5944 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5945 | |
5946 | /* | |
5947 | * Estimate the amount of memory available for userspace allocations, | |
ade63b41 | 5948 | * without causing swapping or OOM. |
d02bd27b | 5949 | */ |
c41f012a | 5950 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5951 | |
5952 | /* | |
5953 | * Not all the page cache can be freed, otherwise the system will | |
ade63b41 YY |
5954 | * start swapping or thrashing. Assume at least half of the page |
5955 | * cache, or the low watermark worth of cache, needs to stay. | |
d02bd27b IR |
5956 | */ |
5957 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
5958 | pagecache -= min(pagecache / 2, wmark_low); | |
5959 | available += pagecache; | |
5960 | ||
5961 | /* | |
b29940c1 VB |
5962 | * Part of the reclaimable slab and other kernel memory consists of |
5963 | * items that are in use, and cannot be freed. Cap this estimate at the | |
5964 | * low watermark. | |
d02bd27b | 5965 | */ |
d42f3245 RG |
5966 | reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) + |
5967 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
b29940c1 | 5968 | available += reclaimable - min(reclaimable / 2, wmark_low); |
034ebf65 | 5969 | |
d02bd27b IR |
5970 | if (available < 0) |
5971 | available = 0; | |
5972 | return available; | |
5973 | } | |
5974 | EXPORT_SYMBOL_GPL(si_mem_available); | |
5975 | ||
1da177e4 LT |
5976 | void si_meminfo(struct sysinfo *val) |
5977 | { | |
ca79b0c2 | 5978 | val->totalram = totalram_pages(); |
11fb9989 | 5979 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 5980 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 5981 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 5982 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 5983 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
5984 | val->mem_unit = PAGE_SIZE; |
5985 | } | |
5986 | ||
5987 | EXPORT_SYMBOL(si_meminfo); | |
5988 | ||
5989 | #ifdef CONFIG_NUMA | |
5990 | void si_meminfo_node(struct sysinfo *val, int nid) | |
5991 | { | |
cdd91a77 JL |
5992 | int zone_type; /* needs to be signed */ |
5993 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
5994 | unsigned long managed_highpages = 0; |
5995 | unsigned long free_highpages = 0; | |
1da177e4 LT |
5996 | pg_data_t *pgdat = NODE_DATA(nid); |
5997 | ||
cdd91a77 | 5998 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 5999 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 6000 | val->totalram = managed_pages; |
11fb9989 | 6001 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 6002 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 6003 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
6004 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
6005 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
6006 | ||
6007 | if (is_highmem(zone)) { | |
9705bea5 | 6008 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
6009 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
6010 | } | |
6011 | } | |
6012 | val->totalhigh = managed_highpages; | |
6013 | val->freehigh = free_highpages; | |
98d2b0eb | 6014 | #else |
fc2bd799 JK |
6015 | val->totalhigh = managed_highpages; |
6016 | val->freehigh = free_highpages; | |
98d2b0eb | 6017 | #endif |
1da177e4 LT |
6018 | val->mem_unit = PAGE_SIZE; |
6019 | } | |
6020 | #endif | |
6021 | ||
ddd588b5 | 6022 | /* |
7bf02ea2 DR |
6023 | * Determine whether the node should be displayed or not, depending on whether |
6024 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 6025 | */ |
9af744d7 | 6026 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 6027 | { |
ddd588b5 | 6028 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 6029 | return false; |
ddd588b5 | 6030 | |
9af744d7 MH |
6031 | /* |
6032 | * no node mask - aka implicit memory numa policy. Do not bother with | |
6033 | * the synchronization - read_mems_allowed_begin - because we do not | |
6034 | * have to be precise here. | |
6035 | */ | |
6036 | if (!nodemask) | |
6037 | nodemask = &cpuset_current_mems_allowed; | |
6038 | ||
6039 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
6040 | } |
6041 | ||
1da177e4 LT |
6042 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
6043 | ||
377e4f16 RV |
6044 | static void show_migration_types(unsigned char type) |
6045 | { | |
6046 | static const char types[MIGRATE_TYPES] = { | |
6047 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 6048 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
6049 | [MIGRATE_RECLAIMABLE] = 'E', |
6050 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
6051 | #ifdef CONFIG_CMA |
6052 | [MIGRATE_CMA] = 'C', | |
6053 | #endif | |
194159fb | 6054 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 6055 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 6056 | #endif |
377e4f16 RV |
6057 | }; |
6058 | char tmp[MIGRATE_TYPES + 1]; | |
6059 | char *p = tmp; | |
6060 | int i; | |
6061 | ||
6062 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
6063 | if (type & (1 << i)) | |
6064 | *p++ = types[i]; | |
6065 | } | |
6066 | ||
6067 | *p = '\0'; | |
1f84a18f | 6068 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
6069 | } |
6070 | ||
974f4367 MH |
6071 | static bool node_has_managed_zones(pg_data_t *pgdat, int max_zone_idx) |
6072 | { | |
6073 | int zone_idx; | |
6074 | for (zone_idx = 0; zone_idx <= max_zone_idx; zone_idx++) | |
6075 | if (zone_managed_pages(pgdat->node_zones + zone_idx)) | |
6076 | return true; | |
6077 | return false; | |
6078 | } | |
6079 | ||
1da177e4 LT |
6080 | /* |
6081 | * Show free area list (used inside shift_scroll-lock stuff) | |
6082 | * We also calculate the percentage fragmentation. We do this by counting the | |
6083 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
6084 | * |
6085 | * Bits in @filter: | |
6086 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
6087 | * cpuset. | |
1da177e4 | 6088 | */ |
974f4367 | 6089 | void __show_free_areas(unsigned int filter, nodemask_t *nodemask, int max_zone_idx) |
1da177e4 | 6090 | { |
d1bfcdb8 | 6091 | unsigned long free_pcp = 0; |
dcadcf1c | 6092 | int cpu, nid; |
1da177e4 | 6093 | struct zone *zone; |
599d0c95 | 6094 | pg_data_t *pgdat; |
1da177e4 | 6095 | |
ee99c71c | 6096 | for_each_populated_zone(zone) { |
974f4367 MH |
6097 | if (zone_idx(zone) > max_zone_idx) |
6098 | continue; | |
9af744d7 | 6099 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 6100 | continue; |
d1bfcdb8 | 6101 | |
761b0677 | 6102 | for_each_online_cpu(cpu) |
28f836b6 | 6103 | free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count; |
1da177e4 LT |
6104 | } |
6105 | ||
a731286d KM |
6106 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
6107 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
8d92890b | 6108 | " unevictable:%lu dirty:%lu writeback:%lu\n" |
d1bfcdb8 | 6109 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
ebc97a52 YA |
6110 | " mapped:%lu shmem:%lu pagetables:%lu\n" |
6111 | " sec_pagetables:%lu bounce:%lu\n" | |
eb2169ce | 6112 | " kernel_misc_reclaimable:%lu\n" |
d1bfcdb8 | 6113 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
6114 | global_node_page_state(NR_ACTIVE_ANON), |
6115 | global_node_page_state(NR_INACTIVE_ANON), | |
6116 | global_node_page_state(NR_ISOLATED_ANON), | |
6117 | global_node_page_state(NR_ACTIVE_FILE), | |
6118 | global_node_page_state(NR_INACTIVE_FILE), | |
6119 | global_node_page_state(NR_ISOLATED_FILE), | |
6120 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
6121 | global_node_page_state(NR_FILE_DIRTY), |
6122 | global_node_page_state(NR_WRITEBACK), | |
d42f3245 RG |
6123 | global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B), |
6124 | global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B), | |
50658e2e | 6125 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 6126 | global_node_page_state(NR_SHMEM), |
f0c0c115 | 6127 | global_node_page_state(NR_PAGETABLE), |
ebc97a52 | 6128 | global_node_page_state(NR_SECONDARY_PAGETABLE), |
c41f012a | 6129 | global_zone_page_state(NR_BOUNCE), |
eb2169ce | 6130 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE), |
c41f012a | 6131 | global_zone_page_state(NR_FREE_PAGES), |
d1bfcdb8 | 6132 | free_pcp, |
c41f012a | 6133 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 6134 | |
599d0c95 | 6135 | for_each_online_pgdat(pgdat) { |
9af744d7 | 6136 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb | 6137 | continue; |
974f4367 MH |
6138 | if (!node_has_managed_zones(pgdat, max_zone_idx)) |
6139 | continue; | |
c02e50bb | 6140 | |
599d0c95 MG |
6141 | printk("Node %d" |
6142 | " active_anon:%lukB" | |
6143 | " inactive_anon:%lukB" | |
6144 | " active_file:%lukB" | |
6145 | " inactive_file:%lukB" | |
6146 | " unevictable:%lukB" | |
6147 | " isolated(anon):%lukB" | |
6148 | " isolated(file):%lukB" | |
50658e2e | 6149 | " mapped:%lukB" |
11fb9989 MG |
6150 | " dirty:%lukB" |
6151 | " writeback:%lukB" | |
6152 | " shmem:%lukB" | |
6153 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6154 | " shmem_thp: %lukB" | |
6155 | " shmem_pmdmapped: %lukB" | |
6156 | " anon_thp: %lukB" | |
6157 | #endif | |
6158 | " writeback_tmp:%lukB" | |
991e7673 SB |
6159 | " kernel_stack:%lukB" |
6160 | #ifdef CONFIG_SHADOW_CALL_STACK | |
6161 | " shadow_call_stack:%lukB" | |
6162 | #endif | |
f0c0c115 | 6163 | " pagetables:%lukB" |
ebc97a52 | 6164 | " sec_pagetables:%lukB" |
599d0c95 MG |
6165 | " all_unreclaimable? %s" |
6166 | "\n", | |
6167 | pgdat->node_id, | |
6168 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
6169 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
6170 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
6171 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
6172 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
6173 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
6174 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 6175 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
6176 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
6177 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 6178 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 | 6179 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
57b2847d | 6180 | K(node_page_state(pgdat, NR_SHMEM_THPS)), |
a1528e21 | 6181 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)), |
69473e5d | 6182 | K(node_page_state(pgdat, NR_ANON_THPS)), |
11fb9989 | 6183 | #endif |
11fb9989 | 6184 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
991e7673 SB |
6185 | node_page_state(pgdat, NR_KERNEL_STACK_KB), |
6186 | #ifdef CONFIG_SHADOW_CALL_STACK | |
6187 | node_page_state(pgdat, NR_KERNEL_SCS_KB), | |
6188 | #endif | |
f0c0c115 | 6189 | K(node_page_state(pgdat, NR_PAGETABLE)), |
ebc97a52 | 6190 | K(node_page_state(pgdat, NR_SECONDARY_PAGETABLE)), |
c73322d0 JW |
6191 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
6192 | "yes" : "no"); | |
599d0c95 MG |
6193 | } |
6194 | ||
ee99c71c | 6195 | for_each_populated_zone(zone) { |
1da177e4 LT |
6196 | int i; |
6197 | ||
974f4367 MH |
6198 | if (zone_idx(zone) > max_zone_idx) |
6199 | continue; | |
9af744d7 | 6200 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 6201 | continue; |
d1bfcdb8 KK |
6202 | |
6203 | free_pcp = 0; | |
6204 | for_each_online_cpu(cpu) | |
28f836b6 | 6205 | free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count; |
d1bfcdb8 | 6206 | |
1da177e4 | 6207 | show_node(zone); |
1f84a18f JP |
6208 | printk(KERN_CONT |
6209 | "%s" | |
1da177e4 | 6210 | " free:%lukB" |
a6ea8b5b | 6211 | " boost:%lukB" |
1da177e4 LT |
6212 | " min:%lukB" |
6213 | " low:%lukB" | |
6214 | " high:%lukB" | |
e47b346a | 6215 | " reserved_highatomic:%luKB" |
71c799f4 MK |
6216 | " active_anon:%lukB" |
6217 | " inactive_anon:%lukB" | |
6218 | " active_file:%lukB" | |
6219 | " inactive_file:%lukB" | |
6220 | " unevictable:%lukB" | |
5a1c84b4 | 6221 | " writepending:%lukB" |
1da177e4 | 6222 | " present:%lukB" |
9feedc9d | 6223 | " managed:%lukB" |
4a0aa73f | 6224 | " mlocked:%lukB" |
4a0aa73f | 6225 | " bounce:%lukB" |
d1bfcdb8 KK |
6226 | " free_pcp:%lukB" |
6227 | " local_pcp:%ukB" | |
d1ce749a | 6228 | " free_cma:%lukB" |
1da177e4 LT |
6229 | "\n", |
6230 | zone->name, | |
88f5acf8 | 6231 | K(zone_page_state(zone, NR_FREE_PAGES)), |
a6ea8b5b | 6232 | K(zone->watermark_boost), |
41858966 MG |
6233 | K(min_wmark_pages(zone)), |
6234 | K(low_wmark_pages(zone)), | |
6235 | K(high_wmark_pages(zone)), | |
e47b346a | 6236 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
6237 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
6238 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
6239 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
6240 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
6241 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 6242 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 6243 | K(zone->present_pages), |
9705bea5 | 6244 | K(zone_managed_pages(zone)), |
4a0aa73f | 6245 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f | 6246 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 | 6247 | K(free_pcp), |
28f836b6 | 6248 | K(this_cpu_read(zone->per_cpu_pageset->count)), |
33e077bd | 6249 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
6250 | printk("lowmem_reserve[]:"); |
6251 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
6252 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
6253 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
6254 | } |
6255 | ||
ee99c71c | 6256 | for_each_populated_zone(zone) { |
d00181b9 KS |
6257 | unsigned int order; |
6258 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 6259 | unsigned char types[MAX_ORDER]; |
1da177e4 | 6260 | |
974f4367 MH |
6261 | if (zone_idx(zone) > max_zone_idx) |
6262 | continue; | |
9af744d7 | 6263 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 6264 | continue; |
1da177e4 | 6265 | show_node(zone); |
1f84a18f | 6266 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
6267 | |
6268 | spin_lock_irqsave(&zone->lock, flags); | |
6269 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
6270 | struct free_area *area = &zone->free_area[order]; |
6271 | int type; | |
6272 | ||
6273 | nr[order] = area->nr_free; | |
8f9de51a | 6274 | total += nr[order] << order; |
377e4f16 RV |
6275 | |
6276 | types[order] = 0; | |
6277 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 6278 | if (!free_area_empty(area, type)) |
377e4f16 RV |
6279 | types[order] |= 1 << type; |
6280 | } | |
1da177e4 LT |
6281 | } |
6282 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 6283 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
6284 | printk(KERN_CONT "%lu*%lukB ", |
6285 | nr[order], K(1UL) << order); | |
377e4f16 RV |
6286 | if (nr[order]) |
6287 | show_migration_types(types[order]); | |
6288 | } | |
1f84a18f | 6289 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
6290 | } |
6291 | ||
dcadcf1c GL |
6292 | for_each_online_node(nid) { |
6293 | if (show_mem_node_skip(filter, nid, nodemask)) | |
6294 | continue; | |
6295 | hugetlb_show_meminfo_node(nid); | |
6296 | } | |
949f7ec5 | 6297 | |
11fb9989 | 6298 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 6299 | |
1da177e4 LT |
6300 | show_swap_cache_info(); |
6301 | } | |
6302 | ||
19770b32 MG |
6303 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
6304 | { | |
6305 | zoneref->zone = zone; | |
6306 | zoneref->zone_idx = zone_idx(zone); | |
6307 | } | |
6308 | ||
1da177e4 LT |
6309 | /* |
6310 | * Builds allocation fallback zone lists. | |
1a93205b CL |
6311 | * |
6312 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 6313 | */ |
9d3be21b | 6314 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 6315 | { |
1a93205b | 6316 | struct zone *zone; |
bc732f1d | 6317 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 6318 | int nr_zones = 0; |
02a68a5e CL |
6319 | |
6320 | do { | |
2f6726e5 | 6321 | zone_type--; |
070f8032 | 6322 | zone = pgdat->node_zones + zone_type; |
e553f62f | 6323 | if (populated_zone(zone)) { |
9d3be21b | 6324 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 6325 | check_highest_zone(zone_type); |
1da177e4 | 6326 | } |
2f6726e5 | 6327 | } while (zone_type); |
bc732f1d | 6328 | |
070f8032 | 6329 | return nr_zones; |
1da177e4 LT |
6330 | } |
6331 | ||
6332 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
6333 | |
6334 | static int __parse_numa_zonelist_order(char *s) | |
6335 | { | |
c9bff3ee | 6336 | /* |
f0953a1b | 6337 | * We used to support different zonelists modes but they turned |
c9bff3ee MH |
6338 | * out to be just not useful. Let's keep the warning in place |
6339 | * if somebody still use the cmd line parameter so that we do | |
6340 | * not fail it silently | |
6341 | */ | |
6342 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
6343 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
6344 | return -EINVAL; |
6345 | } | |
6346 | return 0; | |
6347 | } | |
6348 | ||
c9bff3ee MH |
6349 | char numa_zonelist_order[] = "Node"; |
6350 | ||
f0c0b2b8 KH |
6351 | /* |
6352 | * sysctl handler for numa_zonelist_order | |
6353 | */ | |
cccad5b9 | 6354 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 6355 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 6356 | { |
32927393 CH |
6357 | if (write) |
6358 | return __parse_numa_zonelist_order(buffer); | |
6359 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
6360 | } |
6361 | ||
6362 | ||
f0c0b2b8 KH |
6363 | static int node_load[MAX_NUMNODES]; |
6364 | ||
1da177e4 | 6365 | /** |
4dc3b16b | 6366 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
6367 | * @node: node whose fallback list we're appending |
6368 | * @used_node_mask: nodemask_t of already used nodes | |
6369 | * | |
6370 | * We use a number of factors to determine which is the next node that should | |
6371 | * appear on a given node's fallback list. The node should not have appeared | |
6372 | * already in @node's fallback list, and it should be the next closest node | |
6373 | * according to the distance array (which contains arbitrary distance values | |
6374 | * from each node to each node in the system), and should also prefer nodes | |
6375 | * with no CPUs, since presumably they'll have very little allocation pressure | |
6376 | * on them otherwise. | |
a862f68a MR |
6377 | * |
6378 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 6379 | */ |
79c28a41 | 6380 | int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 6381 | { |
4cf808eb | 6382 | int n, val; |
1da177e4 | 6383 | int min_val = INT_MAX; |
00ef2d2f | 6384 | int best_node = NUMA_NO_NODE; |
1da177e4 | 6385 | |
4cf808eb LT |
6386 | /* Use the local node if we haven't already */ |
6387 | if (!node_isset(node, *used_node_mask)) { | |
6388 | node_set(node, *used_node_mask); | |
6389 | return node; | |
6390 | } | |
1da177e4 | 6391 | |
4b0ef1fe | 6392 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
6393 | |
6394 | /* Don't want a node to appear more than once */ | |
6395 | if (node_isset(n, *used_node_mask)) | |
6396 | continue; | |
6397 | ||
1da177e4 LT |
6398 | /* Use the distance array to find the distance */ |
6399 | val = node_distance(node, n); | |
6400 | ||
4cf808eb LT |
6401 | /* Penalize nodes under us ("prefer the next node") */ |
6402 | val += (n < node); | |
6403 | ||
1da177e4 | 6404 | /* Give preference to headless and unused nodes */ |
b630749f | 6405 | if (!cpumask_empty(cpumask_of_node(n))) |
1da177e4 LT |
6406 | val += PENALTY_FOR_NODE_WITH_CPUS; |
6407 | ||
6408 | /* Slight preference for less loaded node */ | |
37931324 | 6409 | val *= MAX_NUMNODES; |
1da177e4 LT |
6410 | val += node_load[n]; |
6411 | ||
6412 | if (val < min_val) { | |
6413 | min_val = val; | |
6414 | best_node = n; | |
6415 | } | |
6416 | } | |
6417 | ||
6418 | if (best_node >= 0) | |
6419 | node_set(best_node, *used_node_mask); | |
6420 | ||
6421 | return best_node; | |
6422 | } | |
6423 | ||
f0c0b2b8 KH |
6424 | |
6425 | /* | |
6426 | * Build zonelists ordered by node and zones within node. | |
6427 | * This results in maximum locality--normal zone overflows into local | |
6428 | * DMA zone, if any--but risks exhausting DMA zone. | |
6429 | */ | |
9d3be21b MH |
6430 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
6431 | unsigned nr_nodes) | |
1da177e4 | 6432 | { |
9d3be21b MH |
6433 | struct zoneref *zonerefs; |
6434 | int i; | |
6435 | ||
6436 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
6437 | ||
6438 | for (i = 0; i < nr_nodes; i++) { | |
6439 | int nr_zones; | |
6440 | ||
6441 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 6442 | |
9d3be21b MH |
6443 | nr_zones = build_zonerefs_node(node, zonerefs); |
6444 | zonerefs += nr_zones; | |
6445 | } | |
6446 | zonerefs->zone = NULL; | |
6447 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
6448 | } |
6449 | ||
523b9458 CL |
6450 | /* |
6451 | * Build gfp_thisnode zonelists | |
6452 | */ | |
6453 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
6454 | { | |
9d3be21b MH |
6455 | struct zoneref *zonerefs; |
6456 | int nr_zones; | |
523b9458 | 6457 | |
9d3be21b MH |
6458 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
6459 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6460 | zonerefs += nr_zones; | |
6461 | zonerefs->zone = NULL; | |
6462 | zonerefs->zone_idx = 0; | |
523b9458 CL |
6463 | } |
6464 | ||
f0c0b2b8 KH |
6465 | /* |
6466 | * Build zonelists ordered by zone and nodes within zones. | |
6467 | * This results in conserving DMA zone[s] until all Normal memory is | |
6468 | * exhausted, but results in overflowing to remote node while memory | |
6469 | * may still exist in local DMA zone. | |
6470 | */ | |
f0c0b2b8 | 6471 | |
f0c0b2b8 KH |
6472 | static void build_zonelists(pg_data_t *pgdat) |
6473 | { | |
9d3be21b | 6474 | static int node_order[MAX_NUMNODES]; |
37931324 | 6475 | int node, nr_nodes = 0; |
d0ddf49b | 6476 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 6477 | int local_node, prev_node; |
1da177e4 LT |
6478 | |
6479 | /* NUMA-aware ordering of nodes */ | |
6480 | local_node = pgdat->node_id; | |
1da177e4 | 6481 | prev_node = local_node; |
f0c0b2b8 | 6482 | |
f0c0b2b8 | 6483 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
6484 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
6485 | /* | |
6486 | * We don't want to pressure a particular node. | |
6487 | * So adding penalty to the first node in same | |
6488 | * distance group to make it round-robin. | |
6489 | */ | |
957f822a DR |
6490 | if (node_distance(local_node, node) != |
6491 | node_distance(local_node, prev_node)) | |
37931324 | 6492 | node_load[node] += 1; |
f0c0b2b8 | 6493 | |
9d3be21b | 6494 | node_order[nr_nodes++] = node; |
1da177e4 | 6495 | prev_node = node; |
1da177e4 | 6496 | } |
523b9458 | 6497 | |
9d3be21b | 6498 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 6499 | build_thisnode_zonelists(pgdat); |
6cf25392 BR |
6500 | pr_info("Fallback order for Node %d: ", local_node); |
6501 | for (node = 0; node < nr_nodes; node++) | |
6502 | pr_cont("%d ", node_order[node]); | |
6503 | pr_cont("\n"); | |
1da177e4 LT |
6504 | } |
6505 | ||
7aac7898 LS |
6506 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6507 | /* | |
6508 | * Return node id of node used for "local" allocations. | |
6509 | * I.e., first node id of first zone in arg node's generic zonelist. | |
6510 | * Used for initializing percpu 'numa_mem', which is used primarily | |
6511 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
6512 | */ | |
6513 | int local_memory_node(int node) | |
6514 | { | |
c33d6c06 | 6515 | struct zoneref *z; |
7aac7898 | 6516 | |
c33d6c06 | 6517 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 6518 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 6519 | NULL); |
c1093b74 | 6520 | return zone_to_nid(z->zone); |
7aac7898 LS |
6521 | } |
6522 | #endif | |
f0c0b2b8 | 6523 | |
6423aa81 JK |
6524 | static void setup_min_unmapped_ratio(void); |
6525 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
6526 | #else /* CONFIG_NUMA */ |
6527 | ||
f0c0b2b8 | 6528 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 6529 | { |
19655d34 | 6530 | int node, local_node; |
9d3be21b MH |
6531 | struct zoneref *zonerefs; |
6532 | int nr_zones; | |
1da177e4 LT |
6533 | |
6534 | local_node = pgdat->node_id; | |
1da177e4 | 6535 | |
9d3be21b MH |
6536 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
6537 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
6538 | zonerefs += nr_zones; | |
1da177e4 | 6539 | |
54a6eb5c MG |
6540 | /* |
6541 | * Now we build the zonelist so that it contains the zones | |
6542 | * of all the other nodes. | |
6543 | * We don't want to pressure a particular node, so when | |
6544 | * building the zones for node N, we make sure that the | |
6545 | * zones coming right after the local ones are those from | |
6546 | * node N+1 (modulo N) | |
6547 | */ | |
6548 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
6549 | if (!node_online(node)) | |
6550 | continue; | |
9d3be21b MH |
6551 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6552 | zonerefs += nr_zones; | |
1da177e4 | 6553 | } |
54a6eb5c MG |
6554 | for (node = 0; node < local_node; node++) { |
6555 | if (!node_online(node)) | |
6556 | continue; | |
9d3be21b MH |
6557 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
6558 | zonerefs += nr_zones; | |
54a6eb5c MG |
6559 | } |
6560 | ||
9d3be21b MH |
6561 | zonerefs->zone = NULL; |
6562 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
6563 | } |
6564 | ||
6565 | #endif /* CONFIG_NUMA */ | |
6566 | ||
99dcc3e5 CL |
6567 | /* |
6568 | * Boot pageset table. One per cpu which is going to be used for all | |
6569 | * zones and all nodes. The parameters will be set in such a way | |
6570 | * that an item put on a list will immediately be handed over to | |
6571 | * the buddy list. This is safe since pageset manipulation is done | |
6572 | * with interrupts disabled. | |
6573 | * | |
6574 | * The boot_pagesets must be kept even after bootup is complete for | |
6575 | * unused processors and/or zones. They do play a role for bootstrapping | |
6576 | * hotplugged processors. | |
6577 | * | |
6578 | * zoneinfo_show() and maybe other functions do | |
6579 | * not check if the processor is online before following the pageset pointer. | |
6580 | * Other parts of the kernel may not check if the zone is available. | |
6581 | */ | |
28f836b6 | 6582 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats); |
952eaf81 VB |
6583 | /* These effectively disable the pcplists in the boot pageset completely */ |
6584 | #define BOOT_PAGESET_HIGH 0 | |
6585 | #define BOOT_PAGESET_BATCH 1 | |
28f836b6 MG |
6586 | static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset); |
6587 | static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats); | |
6dc2c87a | 6588 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 6589 | |
11cd8638 | 6590 | static void __build_all_zonelists(void *data) |
1da177e4 | 6591 | { |
6811378e | 6592 | int nid; |
afb6ebb3 | 6593 | int __maybe_unused cpu; |
9adb62a5 | 6594 | pg_data_t *self = data; |
b93e0f32 | 6595 | |
3d36424b | 6596 | write_seqlock(&zonelist_update_seq); |
9276b1bc | 6597 | |
7f9cfb31 BL |
6598 | #ifdef CONFIG_NUMA |
6599 | memset(node_load, 0, sizeof(node_load)); | |
6600 | #endif | |
9adb62a5 | 6601 | |
c1152583 WY |
6602 | /* |
6603 | * This node is hotadded and no memory is yet present. So just | |
6604 | * building zonelists is fine - no need to touch other nodes. | |
6605 | */ | |
9adb62a5 JL |
6606 | if (self && !node_online(self->node_id)) { |
6607 | build_zonelists(self); | |
c1152583 | 6608 | } else { |
09f49dca MH |
6609 | /* |
6610 | * All possible nodes have pgdat preallocated | |
6611 | * in free_area_init | |
6612 | */ | |
6613 | for_each_node(nid) { | |
c1152583 | 6614 | pg_data_t *pgdat = NODE_DATA(nid); |
7ea1530a | 6615 | |
c1152583 WY |
6616 | build_zonelists(pgdat); |
6617 | } | |
99dcc3e5 | 6618 | |
7aac7898 LS |
6619 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
6620 | /* | |
6621 | * We now know the "local memory node" for each node-- | |
6622 | * i.e., the node of the first zone in the generic zonelist. | |
6623 | * Set up numa_mem percpu variable for on-line cpus. During | |
6624 | * boot, only the boot cpu should be on-line; we'll init the | |
6625 | * secondary cpus' numa_mem as they come on-line. During | |
6626 | * node/memory hotplug, we'll fixup all on-line cpus. | |
6627 | */ | |
d9c9a0b9 | 6628 | for_each_online_cpu(cpu) |
7aac7898 | 6629 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 6630 | #endif |
d9c9a0b9 | 6631 | } |
b93e0f32 | 6632 | |
3d36424b | 6633 | write_sequnlock(&zonelist_update_seq); |
6811378e YG |
6634 | } |
6635 | ||
061f67bc RV |
6636 | static noinline void __init |
6637 | build_all_zonelists_init(void) | |
6638 | { | |
afb6ebb3 MH |
6639 | int cpu; |
6640 | ||
061f67bc | 6641 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
6642 | |
6643 | /* | |
6644 | * Initialize the boot_pagesets that are going to be used | |
6645 | * for bootstrapping processors. The real pagesets for | |
6646 | * each zone will be allocated later when the per cpu | |
6647 | * allocator is available. | |
6648 | * | |
6649 | * boot_pagesets are used also for bootstrapping offline | |
6650 | * cpus if the system is already booted because the pagesets | |
6651 | * are needed to initialize allocators on a specific cpu too. | |
6652 | * F.e. the percpu allocator needs the page allocator which | |
6653 | * needs the percpu allocator in order to allocate its pagesets | |
6654 | * (a chicken-egg dilemma). | |
6655 | */ | |
6656 | for_each_possible_cpu(cpu) | |
28f836b6 | 6657 | per_cpu_pages_init(&per_cpu(boot_pageset, cpu), &per_cpu(boot_zonestats, cpu)); |
afb6ebb3 | 6658 | |
061f67bc RV |
6659 | mminit_verify_zonelist(); |
6660 | cpuset_init_current_mems_allowed(); | |
6661 | } | |
6662 | ||
4eaf3f64 | 6663 | /* |
4eaf3f64 | 6664 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 6665 | * |
72675e13 | 6666 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 6667 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 6668 | */ |
72675e13 | 6669 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 6670 | { |
0a18e607 DH |
6671 | unsigned long vm_total_pages; |
6672 | ||
6811378e | 6673 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 6674 | build_all_zonelists_init(); |
6811378e | 6675 | } else { |
11cd8638 | 6676 | __build_all_zonelists(pgdat); |
6811378e YG |
6677 | /* cpuset refresh routine should be here */ |
6678 | } | |
56b9413b DH |
6679 | /* Get the number of free pages beyond high watermark in all zones. */ |
6680 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
6681 | /* |
6682 | * Disable grouping by mobility if the number of pages in the | |
6683 | * system is too low to allow the mechanism to work. It would be | |
6684 | * more accurate, but expensive to check per-zone. This check is | |
6685 | * made on memory-hotadd so a system can start with mobility | |
6686 | * disabled and enable it later | |
6687 | */ | |
d9c23400 | 6688 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
6689 | page_group_by_mobility_disabled = 1; |
6690 | else | |
6691 | page_group_by_mobility_disabled = 0; | |
6692 | ||
ce0725f7 | 6693 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 6694 | nr_online_nodes, |
756a025f JP |
6695 | page_group_by_mobility_disabled ? "off" : "on", |
6696 | vm_total_pages); | |
f0c0b2b8 | 6697 | #ifdef CONFIG_NUMA |
f88dfff5 | 6698 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 6699 | #endif |
1da177e4 LT |
6700 | } |
6701 | ||
a9a9e77f PT |
6702 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
6703 | static bool __meminit | |
6704 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
6705 | { | |
a9a9e77f PT |
6706 | static struct memblock_region *r; |
6707 | ||
6708 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
6709 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
cc6de168 | 6710 | for_each_mem_region(r) { |
a9a9e77f PT |
6711 | if (*pfn < memblock_region_memory_end_pfn(r)) |
6712 | break; | |
6713 | } | |
6714 | } | |
6715 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
6716 | memblock_is_mirror(r)) { | |
6717 | *pfn = memblock_region_memory_end_pfn(r); | |
6718 | return true; | |
6719 | } | |
6720 | } | |
a9a9e77f PT |
6721 | return false; |
6722 | } | |
6723 | ||
1da177e4 LT |
6724 | /* |
6725 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 6726 | * up by memblock_free_all() once the early boot process is |
1da177e4 | 6727 | * done. Non-atomic initialization, single-pass. |
d882c006 DH |
6728 | * |
6729 | * All aligned pageblocks are initialized to the specified migratetype | |
6730 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related | |
6731 | * zone stats (e.g., nr_isolate_pageblock) are touched. | |
1da177e4 | 6732 | */ |
ab28cb6e | 6733 | void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone, |
dc2da7b4 | 6734 | unsigned long start_pfn, unsigned long zone_end_pfn, |
d882c006 DH |
6735 | enum meminit_context context, |
6736 | struct vmem_altmap *altmap, int migratetype) | |
1da177e4 | 6737 | { |
a9a9e77f | 6738 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 6739 | struct page *page; |
1da177e4 | 6740 | |
22b31eec HD |
6741 | if (highest_memmap_pfn < end_pfn - 1) |
6742 | highest_memmap_pfn = end_pfn - 1; | |
6743 | ||
966cf44f | 6744 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
6745 | /* |
6746 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
6747 | * memory. We limit the total number of pages to initialize to just |
6748 | * those that might contain the memory mapping. We will defer the | |
6749 | * ZONE_DEVICE page initialization until after we have released | |
6750 | * the hotplug lock. | |
4b94ffdc | 6751 | */ |
966cf44f AD |
6752 | if (zone == ZONE_DEVICE) { |
6753 | if (!altmap) | |
6754 | return; | |
6755 | ||
6756 | if (start_pfn == altmap->base_pfn) | |
6757 | start_pfn += altmap->reserve; | |
6758 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
6759 | } | |
6760 | #endif | |
4b94ffdc | 6761 | |
948c436e | 6762 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 6763 | /* |
b72d0ffb AM |
6764 | * There can be holes in boot-time mem_map[]s handed to this |
6765 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 6766 | */ |
c1d0da83 | 6767 | if (context == MEMINIT_EARLY) { |
a9a9e77f PT |
6768 | if (overlap_memmap_init(zone, &pfn)) |
6769 | continue; | |
dc2da7b4 | 6770 | if (defer_init(nid, pfn, zone_end_pfn)) |
a9a9e77f | 6771 | break; |
a2f3aa02 | 6772 | } |
ac5d2539 | 6773 | |
d0dc12e8 PT |
6774 | page = pfn_to_page(pfn); |
6775 | __init_single_page(page, pfn, zone, nid); | |
c1d0da83 | 6776 | if (context == MEMINIT_HOTPLUG) |
d483da5b | 6777 | __SetPageReserved(page); |
d0dc12e8 | 6778 | |
ac5d2539 | 6779 | /* |
d882c006 DH |
6780 | * Usually, we want to mark the pageblock MIGRATE_MOVABLE, |
6781 | * such that unmovable allocations won't be scattered all | |
6782 | * over the place during system boot. | |
ac5d2539 | 6783 | */ |
ee0913c4 | 6784 | if (pageblock_aligned(pfn)) { |
d882c006 | 6785 | set_pageblock_migratetype(page, migratetype); |
9b6e63cb | 6786 | cond_resched(); |
ac5d2539 | 6787 | } |
948c436e | 6788 | pfn++; |
1da177e4 LT |
6789 | } |
6790 | } | |
6791 | ||
966cf44f | 6792 | #ifdef CONFIG_ZONE_DEVICE |
46487e00 JM |
6793 | static void __ref __init_zone_device_page(struct page *page, unsigned long pfn, |
6794 | unsigned long zone_idx, int nid, | |
6795 | struct dev_pagemap *pgmap) | |
6796 | { | |
6797 | ||
6798 | __init_single_page(page, pfn, zone_idx, nid); | |
6799 | ||
6800 | /* | |
6801 | * Mark page reserved as it will need to wait for onlining | |
6802 | * phase for it to be fully associated with a zone. | |
6803 | * | |
6804 | * We can use the non-atomic __set_bit operation for setting | |
6805 | * the flag as we are still initializing the pages. | |
6806 | */ | |
6807 | __SetPageReserved(page); | |
6808 | ||
6809 | /* | |
6810 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer | |
6811 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
6812 | * ever freed or placed on a driver-private list. | |
6813 | */ | |
6814 | page->pgmap = pgmap; | |
6815 | page->zone_device_data = NULL; | |
6816 | ||
6817 | /* | |
6818 | * Mark the block movable so that blocks are reserved for | |
6819 | * movable at startup. This will force kernel allocations | |
6820 | * to reserve their blocks rather than leaking throughout | |
6821 | * the address space during boot when many long-lived | |
6822 | * kernel allocations are made. | |
6823 | * | |
6824 | * Please note that MEMINIT_HOTPLUG path doesn't clear memmap | |
6825 | * because this is done early in section_activate() | |
6826 | */ | |
ee0913c4 | 6827 | if (pageblock_aligned(pfn)) { |
46487e00 JM |
6828 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
6829 | cond_resched(); | |
6830 | } | |
ef233450 AP |
6831 | |
6832 | /* | |
6833 | * ZONE_DEVICE pages are released directly to the driver page allocator | |
6834 | * which will set the page count to 1 when allocating the page. | |
6835 | */ | |
6836 | if (pgmap->type == MEMORY_DEVICE_PRIVATE || | |
6837 | pgmap->type == MEMORY_DEVICE_COHERENT) | |
6838 | set_page_count(page, 0); | |
46487e00 JM |
6839 | } |
6840 | ||
6fd3620b JM |
6841 | /* |
6842 | * With compound page geometry and when struct pages are stored in ram most | |
6843 | * tail pages are reused. Consequently, the amount of unique struct pages to | |
6844 | * initialize is a lot smaller that the total amount of struct pages being | |
6845 | * mapped. This is a paired / mild layering violation with explicit knowledge | |
6846 | * of how the sparse_vmemmap internals handle compound pages in the lack | |
6847 | * of an altmap. See vmemmap_populate_compound_pages(). | |
6848 | */ | |
6849 | static inline unsigned long compound_nr_pages(struct vmem_altmap *altmap, | |
6850 | unsigned long nr_pages) | |
6851 | { | |
6852 | return is_power_of_2(sizeof(struct page)) && | |
6853 | !altmap ? 2 * (PAGE_SIZE / sizeof(struct page)) : nr_pages; | |
6854 | } | |
6855 | ||
c4386bd8 JM |
6856 | static void __ref memmap_init_compound(struct page *head, |
6857 | unsigned long head_pfn, | |
6858 | unsigned long zone_idx, int nid, | |
6859 | struct dev_pagemap *pgmap, | |
6860 | unsigned long nr_pages) | |
6861 | { | |
6862 | unsigned long pfn, end_pfn = head_pfn + nr_pages; | |
6863 | unsigned int order = pgmap->vmemmap_shift; | |
6864 | ||
6865 | __SetPageHead(head); | |
6866 | for (pfn = head_pfn + 1; pfn < end_pfn; pfn++) { | |
6867 | struct page *page = pfn_to_page(pfn); | |
6868 | ||
6869 | __init_zone_device_page(page, pfn, zone_idx, nid, pgmap); | |
6870 | prep_compound_tail(head, pfn - head_pfn); | |
6871 | set_page_count(page, 0); | |
6872 | ||
6873 | /* | |
6874 | * The first tail page stores compound_mapcount_ptr() and | |
6875 | * compound_order() and the second tail page stores | |
6876 | * compound_pincount_ptr(). Call prep_compound_head() after | |
6877 | * the first and second tail pages have been initialized to | |
6878 | * not have the data overwritten. | |
6879 | */ | |
6880 | if (pfn == head_pfn + 2) | |
6881 | prep_compound_head(head, order); | |
6882 | } | |
6883 | } | |
6884 | ||
966cf44f AD |
6885 | void __ref memmap_init_zone_device(struct zone *zone, |
6886 | unsigned long start_pfn, | |
1f8d75c1 | 6887 | unsigned long nr_pages, |
966cf44f AD |
6888 | struct dev_pagemap *pgmap) |
6889 | { | |
1f8d75c1 | 6890 | unsigned long pfn, end_pfn = start_pfn + nr_pages; |
966cf44f | 6891 | struct pglist_data *pgdat = zone->zone_pgdat; |
514caf23 | 6892 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
c4386bd8 | 6893 | unsigned int pfns_per_compound = pgmap_vmemmap_nr(pgmap); |
966cf44f AD |
6894 | unsigned long zone_idx = zone_idx(zone); |
6895 | unsigned long start = jiffies; | |
6896 | int nid = pgdat->node_id; | |
6897 | ||
c0352904 | 6898 | if (WARN_ON_ONCE(!pgmap || zone_idx != ZONE_DEVICE)) |
966cf44f AD |
6899 | return; |
6900 | ||
6901 | /* | |
122e093c | 6902 | * The call to memmap_init should have already taken care |
966cf44f AD |
6903 | * of the pages reserved for the memmap, so we can just jump to |
6904 | * the end of that region and start processing the device pages. | |
6905 | */ | |
514caf23 | 6906 | if (altmap) { |
966cf44f | 6907 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
1f8d75c1 | 6908 | nr_pages = end_pfn - start_pfn; |
966cf44f AD |
6909 | } |
6910 | ||
c4386bd8 | 6911 | for (pfn = start_pfn; pfn < end_pfn; pfn += pfns_per_compound) { |
966cf44f AD |
6912 | struct page *page = pfn_to_page(pfn); |
6913 | ||
46487e00 | 6914 | __init_zone_device_page(page, pfn, zone_idx, nid, pgmap); |
c4386bd8 JM |
6915 | |
6916 | if (pfns_per_compound == 1) | |
6917 | continue; | |
6918 | ||
6919 | memmap_init_compound(page, pfn, zone_idx, nid, pgmap, | |
6fd3620b | 6920 | compound_nr_pages(altmap, pfns_per_compound)); |
966cf44f AD |
6921 | } |
6922 | ||
fdc029b1 | 6923 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
1f8d75c1 | 6924 | nr_pages, jiffies_to_msecs(jiffies - start)); |
966cf44f AD |
6925 | } |
6926 | ||
6927 | #endif | |
1e548deb | 6928 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6929 | { |
7aeb09f9 | 6930 | unsigned int order, t; |
b2a0ac88 MG |
6931 | for_each_migratetype_order(order, t) { |
6932 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6933 | zone->free_area[order].nr_free = 0; |
6934 | } | |
6935 | } | |
6936 | ||
0740a50b MR |
6937 | /* |
6938 | * Only struct pages that correspond to ranges defined by memblock.memory | |
6939 | * are zeroed and initialized by going through __init_single_page() during | |
122e093c | 6940 | * memmap_init_zone_range(). |
0740a50b MR |
6941 | * |
6942 | * But, there could be struct pages that correspond to holes in | |
6943 | * memblock.memory. This can happen because of the following reasons: | |
6944 | * - physical memory bank size is not necessarily the exact multiple of the | |
6945 | * arbitrary section size | |
6946 | * - early reserved memory may not be listed in memblock.memory | |
6947 | * - memory layouts defined with memmap= kernel parameter may not align | |
6948 | * nicely with memmap sections | |
6949 | * | |
6950 | * Explicitly initialize those struct pages so that: | |
6951 | * - PG_Reserved is set | |
6952 | * - zone and node links point to zone and node that span the page if the | |
6953 | * hole is in the middle of a zone | |
6954 | * - zone and node links point to adjacent zone/node if the hole falls on | |
6955 | * the zone boundary; the pages in such holes will be prepended to the | |
6956 | * zone/node above the hole except for the trailing pages in the last | |
6957 | * section that will be appended to the zone/node below. | |
6958 | */ | |
122e093c MR |
6959 | static void __init init_unavailable_range(unsigned long spfn, |
6960 | unsigned long epfn, | |
6961 | int zone, int node) | |
0740a50b MR |
6962 | { |
6963 | unsigned long pfn; | |
6964 | u64 pgcnt = 0; | |
6965 | ||
6966 | for (pfn = spfn; pfn < epfn; pfn++) { | |
4f9bc69a KW |
6967 | if (!pfn_valid(pageblock_start_pfn(pfn))) { |
6968 | pfn = pageblock_end_pfn(pfn) - 1; | |
0740a50b MR |
6969 | continue; |
6970 | } | |
6971 | __init_single_page(pfn_to_page(pfn), pfn, zone, node); | |
6972 | __SetPageReserved(pfn_to_page(pfn)); | |
6973 | pgcnt++; | |
6974 | } | |
6975 | ||
122e093c MR |
6976 | if (pgcnt) |
6977 | pr_info("On node %d, zone %s: %lld pages in unavailable ranges", | |
6978 | node, zone_names[zone], pgcnt); | |
0740a50b | 6979 | } |
0740a50b | 6980 | |
122e093c MR |
6981 | static void __init memmap_init_zone_range(struct zone *zone, |
6982 | unsigned long start_pfn, | |
6983 | unsigned long end_pfn, | |
6984 | unsigned long *hole_pfn) | |
dfb3ccd0 | 6985 | { |
3256ff83 BH |
6986 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
6987 | unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages; | |
122e093c MR |
6988 | int nid = zone_to_nid(zone), zone_id = zone_idx(zone); |
6989 | ||
6990 | start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn); | |
6991 | end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn); | |
6992 | ||
6993 | if (start_pfn >= end_pfn) | |
6994 | return; | |
6995 | ||
6996 | memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn, | |
6997 | zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE); | |
6998 | ||
6999 | if (*hole_pfn < start_pfn) | |
7000 | init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid); | |
7001 | ||
7002 | *hole_pfn = end_pfn; | |
7003 | } | |
7004 | ||
7005 | static void __init memmap_init(void) | |
7006 | { | |
73a6e474 | 7007 | unsigned long start_pfn, end_pfn; |
122e093c | 7008 | unsigned long hole_pfn = 0; |
b346075f | 7009 | int i, j, zone_id = 0, nid; |
73a6e474 | 7010 | |
122e093c MR |
7011 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
7012 | struct pglist_data *node = NODE_DATA(nid); | |
73a6e474 | 7013 | |
122e093c MR |
7014 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7015 | struct zone *zone = node->node_zones + j; | |
0740a50b | 7016 | |
122e093c MR |
7017 | if (!populated_zone(zone)) |
7018 | continue; | |
0740a50b | 7019 | |
122e093c MR |
7020 | memmap_init_zone_range(zone, start_pfn, end_pfn, |
7021 | &hole_pfn); | |
7022 | zone_id = j; | |
7023 | } | |
73a6e474 | 7024 | } |
0740a50b MR |
7025 | |
7026 | #ifdef CONFIG_SPARSEMEM | |
7027 | /* | |
122e093c MR |
7028 | * Initialize the memory map for hole in the range [memory_end, |
7029 | * section_end]. | |
7030 | * Append the pages in this hole to the highest zone in the last | |
7031 | * node. | |
7032 | * The call to init_unavailable_range() is outside the ifdef to | |
7033 | * silence the compiler warining about zone_id set but not used; | |
7034 | * for FLATMEM it is a nop anyway | |
0740a50b | 7035 | */ |
122e093c | 7036 | end_pfn = round_up(end_pfn, PAGES_PER_SECTION); |
0740a50b | 7037 | if (hole_pfn < end_pfn) |
0740a50b | 7038 | #endif |
122e093c | 7039 | init_unavailable_range(hole_pfn, end_pfn, zone_id, nid); |
dfb3ccd0 | 7040 | } |
1da177e4 | 7041 | |
c803b3c8 MR |
7042 | void __init *memmap_alloc(phys_addr_t size, phys_addr_t align, |
7043 | phys_addr_t min_addr, int nid, bool exact_nid) | |
7044 | { | |
7045 | void *ptr; | |
7046 | ||
7047 | if (exact_nid) | |
7048 | ptr = memblock_alloc_exact_nid_raw(size, align, min_addr, | |
7049 | MEMBLOCK_ALLOC_ACCESSIBLE, | |
7050 | nid); | |
7051 | else | |
7052 | ptr = memblock_alloc_try_nid_raw(size, align, min_addr, | |
7053 | MEMBLOCK_ALLOC_ACCESSIBLE, | |
7054 | nid); | |
7055 | ||
7056 | if (ptr && size > 0) | |
7057 | page_init_poison(ptr, size); | |
7058 | ||
7059 | return ptr; | |
7060 | } | |
7061 | ||
7cd2b0a3 | 7062 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 7063 | { |
3a6be87f | 7064 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
7065 | int batch; |
7066 | ||
7067 | /* | |
b92ca18e MG |
7068 | * The number of pages to batch allocate is either ~0.1% |
7069 | * of the zone or 1MB, whichever is smaller. The batch | |
7070 | * size is striking a balance between allocation latency | |
7071 | * and zone lock contention. | |
e7c8d5c9 | 7072 | */ |
c940e020 | 7073 | batch = min(zone_managed_pages(zone) >> 10, SZ_1M / PAGE_SIZE); |
e7c8d5c9 CL |
7074 | batch /= 4; /* We effectively *= 4 below */ |
7075 | if (batch < 1) | |
7076 | batch = 1; | |
7077 | ||
7078 | /* | |
0ceaacc9 NP |
7079 | * Clamp the batch to a 2^n - 1 value. Having a power |
7080 | * of 2 value was found to be more likely to have | |
7081 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 7082 | * |
0ceaacc9 NP |
7083 | * For example if 2 tasks are alternately allocating |
7084 | * batches of pages, one task can end up with a lot | |
7085 | * of pages of one half of the possible page colors | |
7086 | * and the other with pages of the other colors. | |
e7c8d5c9 | 7087 | */ |
9155203a | 7088 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 7089 | |
e7c8d5c9 | 7090 | return batch; |
3a6be87f DH |
7091 | |
7092 | #else | |
7093 | /* The deferral and batching of frees should be suppressed under NOMMU | |
7094 | * conditions. | |
7095 | * | |
7096 | * The problem is that NOMMU needs to be able to allocate large chunks | |
7097 | * of contiguous memory as there's no hardware page translation to | |
7098 | * assemble apparent contiguous memory from discontiguous pages. | |
7099 | * | |
7100 | * Queueing large contiguous runs of pages for batching, however, | |
7101 | * causes the pages to actually be freed in smaller chunks. As there | |
7102 | * can be a significant delay between the individual batches being | |
7103 | * recycled, this leads to the once large chunks of space being | |
7104 | * fragmented and becoming unavailable for high-order allocations. | |
7105 | */ | |
7106 | return 0; | |
7107 | #endif | |
e7c8d5c9 CL |
7108 | } |
7109 | ||
04f8cfea | 7110 | static int zone_highsize(struct zone *zone, int batch, int cpu_online) |
b92ca18e MG |
7111 | { |
7112 | #ifdef CONFIG_MMU | |
7113 | int high; | |
203c06ee | 7114 | int nr_split_cpus; |
74f44822 MG |
7115 | unsigned long total_pages; |
7116 | ||
7117 | if (!percpu_pagelist_high_fraction) { | |
7118 | /* | |
7119 | * By default, the high value of the pcp is based on the zone | |
7120 | * low watermark so that if they are full then background | |
7121 | * reclaim will not be started prematurely. | |
7122 | */ | |
7123 | total_pages = low_wmark_pages(zone); | |
7124 | } else { | |
7125 | /* | |
7126 | * If percpu_pagelist_high_fraction is configured, the high | |
7127 | * value is based on a fraction of the managed pages in the | |
7128 | * zone. | |
7129 | */ | |
7130 | total_pages = zone_managed_pages(zone) / percpu_pagelist_high_fraction; | |
7131 | } | |
b92ca18e MG |
7132 | |
7133 | /* | |
74f44822 MG |
7134 | * Split the high value across all online CPUs local to the zone. Note |
7135 | * that early in boot that CPUs may not be online yet and that during | |
7136 | * CPU hotplug that the cpumask is not yet updated when a CPU is being | |
203c06ee MG |
7137 | * onlined. For memory nodes that have no CPUs, split pcp->high across |
7138 | * all online CPUs to mitigate the risk that reclaim is triggered | |
7139 | * prematurely due to pages stored on pcp lists. | |
b92ca18e | 7140 | */ |
203c06ee MG |
7141 | nr_split_cpus = cpumask_weight(cpumask_of_node(zone_to_nid(zone))) + cpu_online; |
7142 | if (!nr_split_cpus) | |
7143 | nr_split_cpus = num_online_cpus(); | |
7144 | high = total_pages / nr_split_cpus; | |
b92ca18e MG |
7145 | |
7146 | /* | |
7147 | * Ensure high is at least batch*4. The multiple is based on the | |
7148 | * historical relationship between high and batch. | |
7149 | */ | |
7150 | high = max(high, batch << 2); | |
7151 | ||
7152 | return high; | |
7153 | #else | |
7154 | return 0; | |
7155 | #endif | |
7156 | } | |
7157 | ||
8d7a8fa9 | 7158 | /* |
5c3ad2eb VB |
7159 | * pcp->high and pcp->batch values are related and generally batch is lower |
7160 | * than high. They are also related to pcp->count such that count is lower | |
7161 | * than high, and as soon as it reaches high, the pcplist is flushed. | |
8d7a8fa9 | 7162 | * |
5c3ad2eb VB |
7163 | * However, guaranteeing these relations at all times would require e.g. write |
7164 | * barriers here but also careful usage of read barriers at the read side, and | |
7165 | * thus be prone to error and bad for performance. Thus the update only prevents | |
7166 | * store tearing. Any new users of pcp->batch and pcp->high should ensure they | |
7167 | * can cope with those fields changing asynchronously, and fully trust only the | |
7168 | * pcp->count field on the local CPU with interrupts disabled. | |
8d7a8fa9 CS |
7169 | * |
7170 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
7171 | * outside of boot time (or some other assurance that no concurrent updaters | |
7172 | * exist). | |
7173 | */ | |
7174 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
7175 | unsigned long batch) | |
7176 | { | |
5c3ad2eb VB |
7177 | WRITE_ONCE(pcp->batch, batch); |
7178 | WRITE_ONCE(pcp->high, high); | |
8d7a8fa9 CS |
7179 | } |
7180 | ||
28f836b6 | 7181 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats) |
2caaad41 | 7182 | { |
44042b44 | 7183 | int pindex; |
2caaad41 | 7184 | |
28f836b6 MG |
7185 | memset(pcp, 0, sizeof(*pcp)); |
7186 | memset(pzstats, 0, sizeof(*pzstats)); | |
1c6fe946 | 7187 | |
4b23a68f | 7188 | spin_lock_init(&pcp->lock); |
44042b44 MG |
7189 | for (pindex = 0; pindex < NR_PCP_LISTS; pindex++) |
7190 | INIT_LIST_HEAD(&pcp->lists[pindex]); | |
2caaad41 | 7191 | |
69a8396a VB |
7192 | /* |
7193 | * Set batch and high values safe for a boot pageset. A true percpu | |
7194 | * pageset's initialization will update them subsequently. Here we don't | |
7195 | * need to be as careful as pageset_update() as nobody can access the | |
7196 | * pageset yet. | |
7197 | */ | |
952eaf81 VB |
7198 | pcp->high = BOOT_PAGESET_HIGH; |
7199 | pcp->batch = BOOT_PAGESET_BATCH; | |
3b12e7e9 | 7200 | pcp->free_factor = 0; |
88c90dbc CS |
7201 | } |
7202 | ||
3b1f3658 | 7203 | static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high, |
ec6e8c7e VB |
7204 | unsigned long batch) |
7205 | { | |
28f836b6 | 7206 | struct per_cpu_pages *pcp; |
ec6e8c7e VB |
7207 | int cpu; |
7208 | ||
7209 | for_each_possible_cpu(cpu) { | |
28f836b6 MG |
7210 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
7211 | pageset_update(pcp, high, batch); | |
ec6e8c7e VB |
7212 | } |
7213 | } | |
7214 | ||
8ad4b1fb | 7215 | /* |
0a8b4f1d | 7216 | * Calculate and set new high and batch values for all per-cpu pagesets of a |
bbbecb35 | 7217 | * zone based on the zone's size. |
8ad4b1fb | 7218 | */ |
04f8cfea | 7219 | static void zone_set_pageset_high_and_batch(struct zone *zone, int cpu_online) |
56cef2b8 | 7220 | { |
b92ca18e | 7221 | int new_high, new_batch; |
7115ac6e | 7222 | |
b92ca18e | 7223 | new_batch = max(1, zone_batchsize(zone)); |
04f8cfea | 7224 | new_high = zone_highsize(zone, new_batch, cpu_online); |
169f6c19 | 7225 | |
952eaf81 VB |
7226 | if (zone->pageset_high == new_high && |
7227 | zone->pageset_batch == new_batch) | |
7228 | return; | |
7229 | ||
7230 | zone->pageset_high = new_high; | |
7231 | zone->pageset_batch = new_batch; | |
7232 | ||
ec6e8c7e | 7233 | __zone_set_pageset_high_and_batch(zone, new_high, new_batch); |
169f6c19 CS |
7234 | } |
7235 | ||
72675e13 | 7236 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
7237 | { |
7238 | int cpu; | |
0a8b4f1d | 7239 | |
28f836b6 MG |
7240 | /* Size may be 0 on !SMP && !NUMA */ |
7241 | if (sizeof(struct per_cpu_zonestat) > 0) | |
7242 | zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat); | |
7243 | ||
7244 | zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages); | |
0a8b4f1d | 7245 | for_each_possible_cpu(cpu) { |
28f836b6 MG |
7246 | struct per_cpu_pages *pcp; |
7247 | struct per_cpu_zonestat *pzstats; | |
7248 | ||
7249 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); | |
7250 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); | |
7251 | per_cpu_pages_init(pcp, pzstats); | |
0a8b4f1d VB |
7252 | } |
7253 | ||
04f8cfea | 7254 | zone_set_pageset_high_and_batch(zone, 0); |
319774e2 WF |
7255 | } |
7256 | ||
b89f1735 ML |
7257 | /* |
7258 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7259 | * page high values need to be recalculated. | |
7260 | */ | |
7261 | static void zone_pcp_update(struct zone *zone, int cpu_online) | |
7262 | { | |
7263 | mutex_lock(&pcp_batch_high_lock); | |
7264 | zone_set_pageset_high_and_batch(zone, cpu_online); | |
7265 | mutex_unlock(&pcp_batch_high_lock); | |
7266 | } | |
7267 | ||
2caaad41 | 7268 | /* |
99dcc3e5 CL |
7269 | * Allocate per cpu pagesets and initialize them. |
7270 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 7271 | */ |
99dcc3e5 | 7272 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 7273 | { |
b4911ea2 | 7274 | struct pglist_data *pgdat; |
99dcc3e5 | 7275 | struct zone *zone; |
b418a0f9 | 7276 | int __maybe_unused cpu; |
e7c8d5c9 | 7277 | |
319774e2 WF |
7278 | for_each_populated_zone(zone) |
7279 | setup_zone_pageset(zone); | |
b4911ea2 | 7280 | |
b418a0f9 SD |
7281 | #ifdef CONFIG_NUMA |
7282 | /* | |
7283 | * Unpopulated zones continue using the boot pagesets. | |
7284 | * The numa stats for these pagesets need to be reset. | |
7285 | * Otherwise, they will end up skewing the stats of | |
7286 | * the nodes these zones are associated with. | |
7287 | */ | |
7288 | for_each_possible_cpu(cpu) { | |
28f836b6 | 7289 | struct per_cpu_zonestat *pzstats = &per_cpu(boot_zonestats, cpu); |
f19298b9 MG |
7290 | memset(pzstats->vm_numa_event, 0, |
7291 | sizeof(pzstats->vm_numa_event)); | |
b418a0f9 SD |
7292 | } |
7293 | #endif | |
7294 | ||
b4911ea2 MG |
7295 | for_each_online_pgdat(pgdat) |
7296 | pgdat->per_cpu_nodestats = | |
7297 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
7298 | } |
7299 | ||
c09b4240 | 7300 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 7301 | { |
99dcc3e5 CL |
7302 | /* |
7303 | * per cpu subsystem is not up at this point. The following code | |
7304 | * relies on the ability of the linker to provide the | |
7305 | * offset of a (static) per cpu variable into the per cpu area. | |
7306 | */ | |
28f836b6 MG |
7307 | zone->per_cpu_pageset = &boot_pageset; |
7308 | zone->per_cpu_zonestats = &boot_zonestats; | |
952eaf81 VB |
7309 | zone->pageset_high = BOOT_PAGESET_HIGH; |
7310 | zone->pageset_batch = BOOT_PAGESET_BATCH; | |
ed8ece2e | 7311 | |
b38a8725 | 7312 | if (populated_zone(zone)) |
9660ecaa HK |
7313 | pr_debug(" %s zone: %lu pages, LIFO batch:%u\n", zone->name, |
7314 | zone->present_pages, zone_batchsize(zone)); | |
ed8ece2e DH |
7315 | } |
7316 | ||
dc0bbf3b | 7317 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 7318 | unsigned long zone_start_pfn, |
b171e409 | 7319 | unsigned long size) |
ed8ece2e DH |
7320 | { |
7321 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 7322 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 7323 | |
8f416836 WY |
7324 | if (zone_idx > pgdat->nr_zones) |
7325 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 7326 | |
ed8ece2e DH |
7327 | zone->zone_start_pfn = zone_start_pfn; |
7328 | ||
708614e6 MG |
7329 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
7330 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
7331 | pgdat->node_id, | |
7332 | (unsigned long)zone_idx(zone), | |
7333 | zone_start_pfn, (zone_start_pfn + size)); | |
7334 | ||
1e548deb | 7335 | zone_init_free_lists(zone); |
9dcb8b68 | 7336 | zone->initialized = 1; |
ed8ece2e DH |
7337 | } |
7338 | ||
c713216d MG |
7339 | /** |
7340 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
7341 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
7342 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
7343 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
7344 | * |
7345 | * It returns the start and end page frame of a node based on information | |
7d018176 | 7346 | * provided by memblock_set_node(). If called for a node |
c713216d | 7347 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 7348 | * PFNs will be 0. |
c713216d | 7349 | */ |
bbe5d993 | 7350 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
7351 | unsigned long *start_pfn, unsigned long *end_pfn) |
7352 | { | |
c13291a5 | 7353 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 7354 | int i; |
c13291a5 | 7355 | |
c713216d MG |
7356 | *start_pfn = -1UL; |
7357 | *end_pfn = 0; | |
7358 | ||
c13291a5 TH |
7359 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
7360 | *start_pfn = min(*start_pfn, this_start_pfn); | |
7361 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
7362 | } |
7363 | ||
633c0666 | 7364 | if (*start_pfn == -1UL) |
c713216d | 7365 | *start_pfn = 0; |
c713216d MG |
7366 | } |
7367 | ||
2a1e274a MG |
7368 | /* |
7369 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
7370 | * assumption is made that zones within a node are ordered in monotonic | |
7371 | * increasing memory addresses so that the "highest" populated zone is used | |
7372 | */ | |
b69a7288 | 7373 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
7374 | { |
7375 | int zone_index; | |
7376 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
7377 | if (zone_index == ZONE_MOVABLE) | |
7378 | continue; | |
7379 | ||
7380 | if (arch_zone_highest_possible_pfn[zone_index] > | |
7381 | arch_zone_lowest_possible_pfn[zone_index]) | |
7382 | break; | |
7383 | } | |
7384 | ||
7385 | VM_BUG_ON(zone_index == -1); | |
7386 | movable_zone = zone_index; | |
7387 | } | |
7388 | ||
7389 | /* | |
7390 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 7391 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
7392 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
7393 | * in each node depending on the size of each node and how evenly kernelcore | |
7394 | * is distributed. This helper function adjusts the zone ranges | |
7395 | * provided by the architecture for a given node by using the end of the | |
7396 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
7397 | * zones within a node are in order of monotonic increases memory addresses | |
7398 | */ | |
bbe5d993 | 7399 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
7400 | unsigned long zone_type, |
7401 | unsigned long node_start_pfn, | |
7402 | unsigned long node_end_pfn, | |
7403 | unsigned long *zone_start_pfn, | |
7404 | unsigned long *zone_end_pfn) | |
7405 | { | |
7406 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
7407 | if (zone_movable_pfn[nid]) { | |
7408 | /* Size ZONE_MOVABLE */ | |
7409 | if (zone_type == ZONE_MOVABLE) { | |
7410 | *zone_start_pfn = zone_movable_pfn[nid]; | |
7411 | *zone_end_pfn = min(node_end_pfn, | |
7412 | arch_zone_highest_possible_pfn[movable_zone]); | |
7413 | ||
e506b996 XQ |
7414 | /* Adjust for ZONE_MOVABLE starting within this range */ |
7415 | } else if (!mirrored_kernelcore && | |
7416 | *zone_start_pfn < zone_movable_pfn[nid] && | |
7417 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
7418 | *zone_end_pfn = zone_movable_pfn[nid]; | |
7419 | ||
2a1e274a MG |
7420 | /* Check if this whole range is within ZONE_MOVABLE */ |
7421 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
7422 | *zone_start_pfn = *zone_end_pfn; | |
7423 | } | |
7424 | } | |
7425 | ||
c713216d MG |
7426 | /* |
7427 | * Return the number of pages a zone spans in a node, including holes | |
7428 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
7429 | */ | |
bbe5d993 | 7430 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 7431 | unsigned long zone_type, |
7960aedd ZY |
7432 | unsigned long node_start_pfn, |
7433 | unsigned long node_end_pfn, | |
d91749c1 | 7434 | unsigned long *zone_start_pfn, |
854e8848 | 7435 | unsigned long *zone_end_pfn) |
c713216d | 7436 | { |
299c83dc LF |
7437 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
7438 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 7439 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
7440 | if (!node_start_pfn && !node_end_pfn) |
7441 | return 0; | |
7442 | ||
7960aedd | 7443 | /* Get the start and end of the zone */ |
299c83dc LF |
7444 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
7445 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
7446 | adjust_zone_range_for_zone_movable(nid, zone_type, |
7447 | node_start_pfn, node_end_pfn, | |
d91749c1 | 7448 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
7449 | |
7450 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 7451 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
7452 | return 0; |
7453 | ||
7454 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
7455 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
7456 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
7457 | |
7458 | /* Return the spanned pages */ | |
d91749c1 | 7459 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
7460 | } |
7461 | ||
7462 | /* | |
7463 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 7464 | * then all holes in the requested range will be accounted for. |
c713216d | 7465 | */ |
bbe5d993 | 7466 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
7467 | unsigned long range_start_pfn, |
7468 | unsigned long range_end_pfn) | |
7469 | { | |
96e907d1 TH |
7470 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
7471 | unsigned long start_pfn, end_pfn; | |
7472 | int i; | |
c713216d | 7473 | |
96e907d1 TH |
7474 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
7475 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
7476 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
7477 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 7478 | } |
96e907d1 | 7479 | return nr_absent; |
c713216d MG |
7480 | } |
7481 | ||
7482 | /** | |
7483 | * absent_pages_in_range - Return number of page frames in holes within a range | |
7484 | * @start_pfn: The start PFN to start searching for holes | |
7485 | * @end_pfn: The end PFN to stop searching for holes | |
7486 | * | |
a862f68a | 7487 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
7488 | */ |
7489 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
7490 | unsigned long end_pfn) | |
7491 | { | |
7492 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
7493 | } | |
7494 | ||
7495 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 7496 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 7497 | unsigned long zone_type, |
7960aedd | 7498 | unsigned long node_start_pfn, |
854e8848 | 7499 | unsigned long node_end_pfn) |
c713216d | 7500 | { |
96e907d1 TH |
7501 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
7502 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 7503 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 7504 | unsigned long nr_absent; |
9c7cd687 | 7505 | |
b5685e92 | 7506 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
7507 | if (!node_start_pfn && !node_end_pfn) |
7508 | return 0; | |
7509 | ||
96e907d1 TH |
7510 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
7511 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 7512 | |
2a1e274a MG |
7513 | adjust_zone_range_for_zone_movable(nid, zone_type, |
7514 | node_start_pfn, node_end_pfn, | |
7515 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
7516 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
7517 | ||
7518 | /* | |
7519 | * ZONE_MOVABLE handling. | |
7520 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
7521 | * and vice versa. | |
7522 | */ | |
e506b996 XQ |
7523 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
7524 | unsigned long start_pfn, end_pfn; | |
7525 | struct memblock_region *r; | |
7526 | ||
cc6de168 | 7527 | for_each_mem_region(r) { |
e506b996 XQ |
7528 | start_pfn = clamp(memblock_region_memory_base_pfn(r), |
7529 | zone_start_pfn, zone_end_pfn); | |
7530 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
7531 | zone_start_pfn, zone_end_pfn); | |
7532 | ||
7533 | if (zone_type == ZONE_MOVABLE && | |
7534 | memblock_is_mirror(r)) | |
7535 | nr_absent += end_pfn - start_pfn; | |
7536 | ||
7537 | if (zone_type == ZONE_NORMAL && | |
7538 | !memblock_is_mirror(r)) | |
7539 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
7540 | } |
7541 | } | |
7542 | ||
7543 | return nr_absent; | |
c713216d | 7544 | } |
0e0b864e | 7545 | |
bbe5d993 | 7546 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd | 7547 | unsigned long node_start_pfn, |
854e8848 | 7548 | unsigned long node_end_pfn) |
c713216d | 7549 | { |
febd5949 | 7550 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
7551 | enum zone_type i; |
7552 | ||
febd5949 GZ |
7553 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7554 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 7555 | unsigned long zone_start_pfn, zone_end_pfn; |
3f08a302 | 7556 | unsigned long spanned, absent; |
febd5949 | 7557 | unsigned long size, real_size; |
c713216d | 7558 | |
854e8848 MR |
7559 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i, |
7560 | node_start_pfn, | |
7561 | node_end_pfn, | |
7562 | &zone_start_pfn, | |
7563 | &zone_end_pfn); | |
7564 | absent = zone_absent_pages_in_node(pgdat->node_id, i, | |
7565 | node_start_pfn, | |
7566 | node_end_pfn); | |
3f08a302 MR |
7567 | |
7568 | size = spanned; | |
7569 | real_size = size - absent; | |
7570 | ||
d91749c1 TI |
7571 | if (size) |
7572 | zone->zone_start_pfn = zone_start_pfn; | |
7573 | else | |
7574 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
7575 | zone->spanned_pages = size; |
7576 | zone->present_pages = real_size; | |
4b097002 DH |
7577 | #if defined(CONFIG_MEMORY_HOTPLUG) |
7578 | zone->present_early_pages = real_size; | |
7579 | #endif | |
febd5949 GZ |
7580 | |
7581 | totalpages += size; | |
7582 | realtotalpages += real_size; | |
7583 | } | |
7584 | ||
7585 | pgdat->node_spanned_pages = totalpages; | |
c713216d | 7586 | pgdat->node_present_pages = realtotalpages; |
9660ecaa | 7587 | pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); |
c713216d MG |
7588 | } |
7589 | ||
835c134e MG |
7590 | #ifndef CONFIG_SPARSEMEM |
7591 | /* | |
7592 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
7593 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
7594 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
7595 | * round what is now in bits to nearest long in bits, then return it in |
7596 | * bytes. | |
7597 | */ | |
7c45512d | 7598 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
7599 | { |
7600 | unsigned long usemapsize; | |
7601 | ||
7c45512d | 7602 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
7603 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
7604 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
7605 | usemapsize *= NR_PAGEBLOCK_BITS; |
7606 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
7607 | ||
7608 | return usemapsize / 8; | |
7609 | } | |
7610 | ||
7010a6ec | 7611 | static void __ref setup_usemap(struct zone *zone) |
835c134e | 7612 | { |
7010a6ec BH |
7613 | unsigned long usemapsize = usemap_size(zone->zone_start_pfn, |
7614 | zone->spanned_pages); | |
835c134e | 7615 | zone->pageblock_flags = NULL; |
23a7052a | 7616 | if (usemapsize) { |
6782832e | 7617 | zone->pageblock_flags = |
26fb3dae | 7618 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
7010a6ec | 7619 | zone_to_nid(zone)); |
23a7052a MR |
7620 | if (!zone->pageblock_flags) |
7621 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
7010a6ec | 7622 | usemapsize, zone->name, zone_to_nid(zone)); |
23a7052a | 7623 | } |
835c134e MG |
7624 | } |
7625 | #else | |
7010a6ec | 7626 | static inline void setup_usemap(struct zone *zone) {} |
835c134e MG |
7627 | #endif /* CONFIG_SPARSEMEM */ |
7628 | ||
d9c23400 | 7629 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 7630 | |
d9c23400 | 7631 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 7632 | void __init set_pageblock_order(void) |
d9c23400 | 7633 | { |
b3d40a2b | 7634 | unsigned int order = MAX_ORDER - 1; |
955c1cd7 | 7635 | |
d9c23400 MG |
7636 | /* Check that pageblock_nr_pages has not already been setup */ |
7637 | if (pageblock_order) | |
7638 | return; | |
7639 | ||
b3d40a2b DH |
7640 | /* Don't let pageblocks exceed the maximum allocation granularity. */ |
7641 | if (HPAGE_SHIFT > PAGE_SHIFT && HUGETLB_PAGE_ORDER < order) | |
955c1cd7 | 7642 | order = HUGETLB_PAGE_ORDER; |
955c1cd7 | 7643 | |
d9c23400 MG |
7644 | /* |
7645 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
7646 | * This value may be variable depending on boot parameters on IA64 and |
7647 | * powerpc. | |
d9c23400 MG |
7648 | */ |
7649 | pageblock_order = order; | |
7650 | } | |
7651 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7652 | ||
ba72cb8c MG |
7653 | /* |
7654 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
7655 | * is unused as pageblock_order is set at compile-time. See |
7656 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
7657 | * the kernel config | |
ba72cb8c | 7658 | */ |
03e85f9d | 7659 | void __init set_pageblock_order(void) |
ba72cb8c | 7660 | { |
ba72cb8c | 7661 | } |
d9c23400 MG |
7662 | |
7663 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
7664 | ||
03e85f9d | 7665 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 7666 | unsigned long present_pages) |
01cefaef JL |
7667 | { |
7668 | unsigned long pages = spanned_pages; | |
7669 | ||
7670 | /* | |
7671 | * Provide a more accurate estimation if there are holes within | |
7672 | * the zone and SPARSEMEM is in use. If there are holes within the | |
7673 | * zone, each populated memory region may cost us one or two extra | |
7674 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 7675 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
7676 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
7677 | */ | |
7678 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
7679 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
7680 | pages = present_pages; | |
7681 | ||
7682 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
7683 | } | |
7684 | ||
ace1db39 OS |
7685 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
7686 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
7687 | { | |
364c1eeb YS |
7688 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
7689 | ||
7690 | spin_lock_init(&ds_queue->split_queue_lock); | |
7691 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
7692 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
7693 | } |
7694 | #else | |
7695 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
7696 | #endif | |
7697 | ||
7698 | #ifdef CONFIG_COMPACTION | |
7699 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
7700 | { | |
7701 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
7702 | } | |
7703 | #else | |
7704 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
7705 | #endif | |
7706 | ||
03e85f9d | 7707 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 7708 | { |
8cd7c588 MG |
7709 | int i; |
7710 | ||
208d54e5 | 7711 | pgdat_resize_init(pgdat); |
b4a0215e | 7712 | pgdat_kswapd_lock_init(pgdat); |
ace1db39 | 7713 | |
ace1db39 OS |
7714 | pgdat_init_split_queue(pgdat); |
7715 | pgdat_init_kcompactd(pgdat); | |
7716 | ||
1da177e4 | 7717 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 7718 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 7719 | |
8cd7c588 MG |
7720 | for (i = 0; i < NR_VMSCAN_THROTTLE; i++) |
7721 | init_waitqueue_head(&pgdat->reclaim_wait[i]); | |
7722 | ||
eefa864b | 7723 | pgdat_page_ext_init(pgdat); |
867e5e1d | 7724 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
7725 | } |
7726 | ||
7727 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
7728 | unsigned long remaining_pages) | |
7729 | { | |
9705bea5 | 7730 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
7731 | zone_set_nid(zone, nid); |
7732 | zone->name = zone_names[idx]; | |
7733 | zone->zone_pgdat = NODE_DATA(nid); | |
7734 | spin_lock_init(&zone->lock); | |
7735 | zone_seqlock_init(zone); | |
7736 | zone_pcp_init(zone); | |
7737 | } | |
7738 | ||
7739 | /* | |
7740 | * Set up the zone data structures | |
7741 | * - init pgdat internals | |
7742 | * - init all zones belonging to this node | |
7743 | * | |
7744 | * NOTE: this function is only called during memory hotplug | |
7745 | */ | |
7746 | #ifdef CONFIG_MEMORY_HOTPLUG | |
70b5b46a | 7747 | void __ref free_area_init_core_hotplug(struct pglist_data *pgdat) |
03e85f9d | 7748 | { |
70b5b46a | 7749 | int nid = pgdat->node_id; |
03e85f9d | 7750 | enum zone_type z; |
70b5b46a | 7751 | int cpu; |
03e85f9d OS |
7752 | |
7753 | pgdat_init_internals(pgdat); | |
70b5b46a MH |
7754 | |
7755 | if (pgdat->per_cpu_nodestats == &boot_nodestats) | |
7756 | pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); | |
7757 | ||
7758 | /* | |
7759 | * Reset the nr_zones, order and highest_zoneidx before reuse. | |
7760 | * Note that kswapd will init kswapd_highest_zoneidx properly | |
7761 | * when it starts in the near future. | |
7762 | */ | |
7763 | pgdat->nr_zones = 0; | |
7764 | pgdat->kswapd_order = 0; | |
7765 | pgdat->kswapd_highest_zoneidx = 0; | |
7766 | pgdat->node_start_pfn = 0; | |
7767 | for_each_online_cpu(cpu) { | |
7768 | struct per_cpu_nodestat *p; | |
7769 | ||
7770 | p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu); | |
7771 | memset(p, 0, sizeof(*p)); | |
7772 | } | |
7773 | ||
03e85f9d OS |
7774 | for (z = 0; z < MAX_NR_ZONES; z++) |
7775 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
7776 | } | |
7777 | #endif | |
7778 | ||
7779 | /* | |
7780 | * Set up the zone data structures: | |
7781 | * - mark all pages reserved | |
7782 | * - mark all memory queues empty | |
7783 | * - clear the memory bitmaps | |
7784 | * | |
7785 | * NOTE: pgdat should get zeroed by caller. | |
7786 | * NOTE: this function is only called during early init. | |
7787 | */ | |
7788 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
7789 | { | |
7790 | enum zone_type j; | |
7791 | int nid = pgdat->node_id; | |
5f63b720 | 7792 | |
03e85f9d | 7793 | pgdat_init_internals(pgdat); |
385386cf JW |
7794 | pgdat->per_cpu_nodestats = &boot_nodestats; |
7795 | ||
1da177e4 LT |
7796 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7797 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 7798 | unsigned long size, freesize, memmap_pages; |
1da177e4 | 7799 | |
febd5949 | 7800 | size = zone->spanned_pages; |
e6943859 | 7801 | freesize = zone->present_pages; |
1da177e4 | 7802 | |
0e0b864e | 7803 | /* |
9feedc9d | 7804 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
7805 | * is used by this zone for memmap. This affects the watermark |
7806 | * and per-cpu initialisations | |
7807 | */ | |
e6943859 | 7808 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
7809 | if (!is_highmem_idx(j)) { |
7810 | if (freesize >= memmap_pages) { | |
7811 | freesize -= memmap_pages; | |
7812 | if (memmap_pages) | |
9660ecaa HK |
7813 | pr_debug(" %s zone: %lu pages used for memmap\n", |
7814 | zone_names[j], memmap_pages); | |
ba914f48 | 7815 | } else |
e47aa905 | 7816 | pr_warn(" %s zone: %lu memmap pages exceeds freesize %lu\n", |
ba914f48 ZH |
7817 | zone_names[j], memmap_pages, freesize); |
7818 | } | |
0e0b864e | 7819 | |
6267276f | 7820 | /* Account for reserved pages */ |
9feedc9d JL |
7821 | if (j == 0 && freesize > dma_reserve) { |
7822 | freesize -= dma_reserve; | |
9660ecaa | 7823 | pr_debug(" %s zone: %lu pages reserved\n", zone_names[0], dma_reserve); |
0e0b864e MG |
7824 | } |
7825 | ||
98d2b0eb | 7826 | if (!is_highmem_idx(j)) |
9feedc9d | 7827 | nr_kernel_pages += freesize; |
01cefaef JL |
7828 | /* Charge for highmem memmap if there are enough kernel pages */ |
7829 | else if (nr_kernel_pages > memmap_pages * 2) | |
7830 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 7831 | nr_all_pages += freesize; |
1da177e4 | 7832 | |
9feedc9d JL |
7833 | /* |
7834 | * Set an approximate value for lowmem here, it will be adjusted | |
7835 | * when the bootmem allocator frees pages into the buddy system. | |
7836 | * And all highmem pages will be managed by the buddy system. | |
7837 | */ | |
03e85f9d | 7838 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 7839 | |
d883c6cf | 7840 | if (!size) |
1da177e4 LT |
7841 | continue; |
7842 | ||
955c1cd7 | 7843 | set_pageblock_order(); |
7010a6ec | 7844 | setup_usemap(zone); |
9699ee7b | 7845 | init_currently_empty_zone(zone, zone->zone_start_pfn, size); |
1da177e4 LT |
7846 | } |
7847 | } | |
7848 | ||
43b02ba9 | 7849 | #ifdef CONFIG_FLATMEM |
3b446da6 | 7850 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 7851 | { |
b0aeba74 | 7852 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
7853 | unsigned long __maybe_unused offset = 0; |
7854 | ||
1da177e4 LT |
7855 | /* Skip empty nodes */ |
7856 | if (!pgdat->node_spanned_pages) | |
7857 | return; | |
7858 | ||
b0aeba74 TL |
7859 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
7860 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
7861 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
7862 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 7863 | unsigned long size, end; |
d41dee36 AW |
7864 | struct page *map; |
7865 | ||
e984bb43 BP |
7866 | /* |
7867 | * The zone's endpoints aren't required to be MAX_ORDER | |
7868 | * aligned but the node_mem_map endpoints must be in order | |
7869 | * for the buddy allocator to function correctly. | |
7870 | */ | |
108bcc96 | 7871 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
7872 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
7873 | size = (end - start) * sizeof(struct page); | |
c803b3c8 MR |
7874 | map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT, |
7875 | pgdat->node_id, false); | |
23a7052a MR |
7876 | if (!map) |
7877 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
7878 | size, pgdat->node_id); | |
a1c34a3b | 7879 | pgdat->node_mem_map = map + offset; |
1da177e4 | 7880 | } |
0cd842f9 OS |
7881 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
7882 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
7883 | (unsigned long)pgdat->node_mem_map); | |
a9ee6cf5 | 7884 | #ifndef CONFIG_NUMA |
1da177e4 LT |
7885 | /* |
7886 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
7887 | */ | |
c713216d | 7888 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 7889 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d | 7890 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 7891 | mem_map -= offset; |
c713216d | 7892 | } |
1da177e4 LT |
7893 | #endif |
7894 | } | |
0cd842f9 | 7895 | #else |
3b446da6 | 7896 | static inline void alloc_node_mem_map(struct pglist_data *pgdat) { } |
43b02ba9 | 7897 | #endif /* CONFIG_FLATMEM */ |
1da177e4 | 7898 | |
0188dc98 OS |
7899 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
7900 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
7901 | { | |
0188dc98 OS |
7902 | pgdat->first_deferred_pfn = ULONG_MAX; |
7903 | } | |
7904 | #else | |
7905 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
7906 | #endif | |
7907 | ||
854e8848 | 7908 | static void __init free_area_init_node(int nid) |
1da177e4 | 7909 | { |
9109fb7b | 7910 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
7911 | unsigned long start_pfn = 0; |
7912 | unsigned long end_pfn = 0; | |
9109fb7b | 7913 | |
88fdf75d | 7914 | /* pg_data_t should be reset to zero when it's allocated */ |
97a225e6 | 7915 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx); |
88fdf75d | 7916 | |
854e8848 | 7917 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
88fdf75d | 7918 | |
1da177e4 | 7919 | pgdat->node_id = nid; |
854e8848 | 7920 | pgdat->node_start_pfn = start_pfn; |
75ef7184 | 7921 | pgdat->per_cpu_nodestats = NULL; |
854e8848 | 7922 | |
7c30daac MH |
7923 | if (start_pfn != end_pfn) { |
7924 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, | |
7925 | (u64)start_pfn << PAGE_SHIFT, | |
7926 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
7927 | } else { | |
7928 | pr_info("Initmem setup node %d as memoryless\n", nid); | |
7929 | } | |
7930 | ||
854e8848 | 7931 | calculate_node_totalpages(pgdat, start_pfn, end_pfn); |
1da177e4 LT |
7932 | |
7933 | alloc_node_mem_map(pgdat); | |
0188dc98 | 7934 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 7935 | |
7f3eb55b | 7936 | free_area_init_core(pgdat); |
1da177e4 LT |
7937 | } |
7938 | ||
1ca75fa7 | 7939 | static void __init free_area_init_memoryless_node(int nid) |
3f08a302 | 7940 | { |
854e8848 | 7941 | free_area_init_node(nid); |
3f08a302 MR |
7942 | } |
7943 | ||
418508c1 MS |
7944 | #if MAX_NUMNODES > 1 |
7945 | /* | |
7946 | * Figure out the number of possible node ids. | |
7947 | */ | |
f9872caf | 7948 | void __init setup_nr_node_ids(void) |
418508c1 | 7949 | { |
904a9553 | 7950 | unsigned int highest; |
418508c1 | 7951 | |
904a9553 | 7952 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7953 | nr_node_ids = highest + 1; |
7954 | } | |
418508c1 MS |
7955 | #endif |
7956 | ||
1e01979c TH |
7957 | /** |
7958 | * node_map_pfn_alignment - determine the maximum internode alignment | |
7959 | * | |
7960 | * This function should be called after node map is populated and sorted. | |
7961 | * It calculates the maximum power of two alignment which can distinguish | |
7962 | * all the nodes. | |
7963 | * | |
7964 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
7965 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
7966 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
7967 | * shifted, 1GiB is enough and this function will indicate so. | |
7968 | * | |
7969 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
7970 | * model has fine enough granularity to avoid incorrect mapping for the | |
7971 | * populated node map. | |
7972 | * | |
a862f68a | 7973 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
7974 | * requirement (single node). |
7975 | */ | |
7976 | unsigned long __init node_map_pfn_alignment(void) | |
7977 | { | |
7978 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 7979 | unsigned long start, end, mask; |
98fa15f3 | 7980 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 7981 | int i, nid; |
1e01979c | 7982 | |
c13291a5 | 7983 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
7984 | if (!start || last_nid < 0 || last_nid == nid) { |
7985 | last_nid = nid; | |
7986 | last_end = end; | |
7987 | continue; | |
7988 | } | |
7989 | ||
7990 | /* | |
7991 | * Start with a mask granular enough to pin-point to the | |
7992 | * start pfn and tick off bits one-by-one until it becomes | |
7993 | * too coarse to separate the current node from the last. | |
7994 | */ | |
7995 | mask = ~((1 << __ffs(start)) - 1); | |
7996 | while (mask && last_end <= (start & (mask << 1))) | |
7997 | mask <<= 1; | |
7998 | ||
7999 | /* accumulate all internode masks */ | |
8000 | accl_mask |= mask; | |
8001 | } | |
8002 | ||
8003 | /* convert mask to number of pages */ | |
8004 | return ~accl_mask + 1; | |
8005 | } | |
8006 | ||
37b07e41 LS |
8007 | /* |
8008 | * early_calculate_totalpages() | |
8009 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 8010 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 8011 | */ |
484f51f8 | 8012 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 8013 | { |
7e63efef | 8014 | unsigned long totalpages = 0; |
c13291a5 TH |
8015 | unsigned long start_pfn, end_pfn; |
8016 | int i, nid; | |
8017 | ||
8018 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
8019 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 8020 | |
37b07e41 LS |
8021 | totalpages += pages; |
8022 | if (pages) | |
4b0ef1fe | 8023 | node_set_state(nid, N_MEMORY); |
37b07e41 | 8024 | } |
b8af2941 | 8025 | return totalpages; |
7e63efef MG |
8026 | } |
8027 | ||
2a1e274a MG |
8028 | /* |
8029 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
8030 | * is spread evenly between nodes as long as the nodes have enough | |
8031 | * memory. When they don't, some nodes will have more kernelcore than | |
8032 | * others | |
8033 | */ | |
b224ef85 | 8034 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
8035 | { |
8036 | int i, nid; | |
8037 | unsigned long usable_startpfn; | |
8038 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 8039 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 8040 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 8041 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 8042 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 8043 | struct memblock_region *r; |
b2f3eebe TC |
8044 | |
8045 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
8046 | find_usable_zone_for_movable(); | |
8047 | ||
8048 | /* | |
8049 | * If movable_node is specified, ignore kernelcore and movablecore | |
8050 | * options. | |
8051 | */ | |
8052 | if (movable_node_is_enabled()) { | |
cc6de168 | 8053 | for_each_mem_region(r) { |
136199f0 | 8054 | if (!memblock_is_hotpluggable(r)) |
b2f3eebe TC |
8055 | continue; |
8056 | ||
d622abf7 | 8057 | nid = memblock_get_region_node(r); |
b2f3eebe | 8058 | |
136199f0 | 8059 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
8060 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
8061 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
8062 | usable_startpfn; | |
8063 | } | |
8064 | ||
8065 | goto out2; | |
8066 | } | |
2a1e274a | 8067 | |
342332e6 TI |
8068 | /* |
8069 | * If kernelcore=mirror is specified, ignore movablecore option | |
8070 | */ | |
8071 | if (mirrored_kernelcore) { | |
8072 | bool mem_below_4gb_not_mirrored = false; | |
8073 | ||
cc6de168 | 8074 | for_each_mem_region(r) { |
342332e6 TI |
8075 | if (memblock_is_mirror(r)) |
8076 | continue; | |
8077 | ||
d622abf7 | 8078 | nid = memblock_get_region_node(r); |
342332e6 TI |
8079 | |
8080 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
8081 | ||
aa282a15 | 8082 | if (usable_startpfn < PHYS_PFN(SZ_4G)) { |
342332e6 TI |
8083 | mem_below_4gb_not_mirrored = true; |
8084 | continue; | |
8085 | } | |
8086 | ||
8087 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
8088 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
8089 | usable_startpfn; | |
8090 | } | |
8091 | ||
8092 | if (mem_below_4gb_not_mirrored) | |
633bf2fe | 8093 | pr_warn("This configuration results in unmirrored kernel memory.\n"); |
342332e6 TI |
8094 | |
8095 | goto out2; | |
8096 | } | |
8097 | ||
7e63efef | 8098 | /* |
a5c6d650 DR |
8099 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
8100 | * amount of necessary memory. | |
8101 | */ | |
8102 | if (required_kernelcore_percent) | |
8103 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
8104 | 10000UL; | |
8105 | if (required_movablecore_percent) | |
8106 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
8107 | 10000UL; | |
8108 | ||
8109 | /* | |
8110 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
8111 | * kernelcore that corresponds so that memory usable for |
8112 | * any allocation type is evenly spread. If both kernelcore | |
8113 | * and movablecore are specified, then the value of kernelcore | |
8114 | * will be used for required_kernelcore if it's greater than | |
8115 | * what movablecore would have allowed. | |
8116 | */ | |
8117 | if (required_movablecore) { | |
7e63efef MG |
8118 | unsigned long corepages; |
8119 | ||
8120 | /* | |
8121 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
8122 | * was requested by the user | |
8123 | */ | |
8124 | required_movablecore = | |
8125 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 8126 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
8127 | corepages = totalpages - required_movablecore; |
8128 | ||
8129 | required_kernelcore = max(required_kernelcore, corepages); | |
8130 | } | |
8131 | ||
bde304bd XQ |
8132 | /* |
8133 | * If kernelcore was not specified or kernelcore size is larger | |
8134 | * than totalpages, there is no ZONE_MOVABLE. | |
8135 | */ | |
8136 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 8137 | goto out; |
2a1e274a MG |
8138 | |
8139 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
8140 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
8141 | ||
8142 | restart: | |
8143 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
8144 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 8145 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
8146 | unsigned long start_pfn, end_pfn; |
8147 | ||
2a1e274a MG |
8148 | /* |
8149 | * Recalculate kernelcore_node if the division per node | |
8150 | * now exceeds what is necessary to satisfy the requested | |
8151 | * amount of memory for the kernel | |
8152 | */ | |
8153 | if (required_kernelcore < kernelcore_node) | |
8154 | kernelcore_node = required_kernelcore / usable_nodes; | |
8155 | ||
8156 | /* | |
8157 | * As the map is walked, we track how much memory is usable | |
8158 | * by the kernel using kernelcore_remaining. When it is | |
8159 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
8160 | */ | |
8161 | kernelcore_remaining = kernelcore_node; | |
8162 | ||
8163 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 8164 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
8165 | unsigned long size_pages; |
8166 | ||
c13291a5 | 8167 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
8168 | if (start_pfn >= end_pfn) |
8169 | continue; | |
8170 | ||
8171 | /* Account for what is only usable for kernelcore */ | |
8172 | if (start_pfn < usable_startpfn) { | |
8173 | unsigned long kernel_pages; | |
8174 | kernel_pages = min(end_pfn, usable_startpfn) | |
8175 | - start_pfn; | |
8176 | ||
8177 | kernelcore_remaining -= min(kernel_pages, | |
8178 | kernelcore_remaining); | |
8179 | required_kernelcore -= min(kernel_pages, | |
8180 | required_kernelcore); | |
8181 | ||
8182 | /* Continue if range is now fully accounted */ | |
8183 | if (end_pfn <= usable_startpfn) { | |
8184 | ||
8185 | /* | |
8186 | * Push zone_movable_pfn to the end so | |
8187 | * that if we have to rebalance | |
8188 | * kernelcore across nodes, we will | |
8189 | * not double account here | |
8190 | */ | |
8191 | zone_movable_pfn[nid] = end_pfn; | |
8192 | continue; | |
8193 | } | |
8194 | start_pfn = usable_startpfn; | |
8195 | } | |
8196 | ||
8197 | /* | |
8198 | * The usable PFN range for ZONE_MOVABLE is from | |
8199 | * start_pfn->end_pfn. Calculate size_pages as the | |
8200 | * number of pages used as kernelcore | |
8201 | */ | |
8202 | size_pages = end_pfn - start_pfn; | |
8203 | if (size_pages > kernelcore_remaining) | |
8204 | size_pages = kernelcore_remaining; | |
8205 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
8206 | ||
8207 | /* | |
8208 | * Some kernelcore has been met, update counts and | |
8209 | * break if the kernelcore for this node has been | |
b8af2941 | 8210 | * satisfied |
2a1e274a MG |
8211 | */ |
8212 | required_kernelcore -= min(required_kernelcore, | |
8213 | size_pages); | |
8214 | kernelcore_remaining -= size_pages; | |
8215 | if (!kernelcore_remaining) | |
8216 | break; | |
8217 | } | |
8218 | } | |
8219 | ||
8220 | /* | |
8221 | * If there is still required_kernelcore, we do another pass with one | |
8222 | * less node in the count. This will push zone_movable_pfn[nid] further | |
8223 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 8224 | * satisfied |
2a1e274a MG |
8225 | */ |
8226 | usable_nodes--; | |
8227 | if (usable_nodes && required_kernelcore > usable_nodes) | |
8228 | goto restart; | |
8229 | ||
b2f3eebe | 8230 | out2: |
2a1e274a | 8231 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
ddbc84f3 AP |
8232 | for (nid = 0; nid < MAX_NUMNODES; nid++) { |
8233 | unsigned long start_pfn, end_pfn; | |
8234 | ||
2a1e274a MG |
8235 | zone_movable_pfn[nid] = |
8236 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 8237 | |
ddbc84f3 AP |
8238 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
8239 | if (zone_movable_pfn[nid] >= end_pfn) | |
8240 | zone_movable_pfn[nid] = 0; | |
8241 | } | |
8242 | ||
20e6926d | 8243 | out: |
66918dcd | 8244 | /* restore the node_state */ |
4b0ef1fe | 8245 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
8246 | } |
8247 | ||
4b0ef1fe LJ |
8248 | /* Any regular or high memory on that node ? */ |
8249 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 8250 | { |
37b07e41 LS |
8251 | enum zone_type zone_type; |
8252 | ||
4b0ef1fe | 8253 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 8254 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 8255 | if (populated_zone(zone)) { |
7b0e0c0e OS |
8256 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
8257 | node_set_state(nid, N_HIGH_MEMORY); | |
8258 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 8259 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
8260 | break; |
8261 | } | |
37b07e41 | 8262 | } |
37b07e41 LS |
8263 | } |
8264 | ||
51930df5 | 8265 | /* |
f0953a1b | 8266 | * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For |
51930df5 MR |
8267 | * such cases we allow max_zone_pfn sorted in the descending order |
8268 | */ | |
8269 | bool __weak arch_has_descending_max_zone_pfns(void) | |
8270 | { | |
8271 | return false; | |
8272 | } | |
8273 | ||
c713216d | 8274 | /** |
9691a071 | 8275 | * free_area_init - Initialise all pg_data_t and zone data |
88ca3b94 | 8276 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
8277 | * |
8278 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 8279 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
8280 | * zone in each node and their holes is calculated. If the maximum PFN |
8281 | * between two adjacent zones match, it is assumed that the zone is empty. | |
8282 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
8283 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
8284 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
8285 | * at arch_max_dma_pfn. | |
8286 | */ | |
9691a071 | 8287 | void __init free_area_init(unsigned long *max_zone_pfn) |
c713216d | 8288 | { |
c13291a5 | 8289 | unsigned long start_pfn, end_pfn; |
51930df5 MR |
8290 | int i, nid, zone; |
8291 | bool descending; | |
a6af2bc3 | 8292 | |
c713216d MG |
8293 | /* Record where the zone boundaries are */ |
8294 | memset(arch_zone_lowest_possible_pfn, 0, | |
8295 | sizeof(arch_zone_lowest_possible_pfn)); | |
8296 | memset(arch_zone_highest_possible_pfn, 0, | |
8297 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe | 8298 | |
fb70c487 | 8299 | start_pfn = PHYS_PFN(memblock_start_of_DRAM()); |
51930df5 | 8300 | descending = arch_has_descending_max_zone_pfns(); |
90cae1fe OH |
8301 | |
8302 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
51930df5 MR |
8303 | if (descending) |
8304 | zone = MAX_NR_ZONES - i - 1; | |
8305 | else | |
8306 | zone = i; | |
8307 | ||
8308 | if (zone == ZONE_MOVABLE) | |
2a1e274a | 8309 | continue; |
90cae1fe | 8310 | |
51930df5 MR |
8311 | end_pfn = max(max_zone_pfn[zone], start_pfn); |
8312 | arch_zone_lowest_possible_pfn[zone] = start_pfn; | |
8313 | arch_zone_highest_possible_pfn[zone] = end_pfn; | |
90cae1fe OH |
8314 | |
8315 | start_pfn = end_pfn; | |
c713216d | 8316 | } |
2a1e274a MG |
8317 | |
8318 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
8319 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 8320 | find_zone_movable_pfns_for_nodes(); |
c713216d | 8321 | |
c713216d | 8322 | /* Print out the zone ranges */ |
f88dfff5 | 8323 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
8324 | for (i = 0; i < MAX_NR_ZONES; i++) { |
8325 | if (i == ZONE_MOVABLE) | |
8326 | continue; | |
f88dfff5 | 8327 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
8328 | if (arch_zone_lowest_possible_pfn[i] == |
8329 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 8330 | pr_cont("empty\n"); |
72f0ba02 | 8331 | else |
8d29e18a JG |
8332 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
8333 | (u64)arch_zone_lowest_possible_pfn[i] | |
8334 | << PAGE_SHIFT, | |
8335 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 8336 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
8337 | } |
8338 | ||
8339 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 8340 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
8341 | for (i = 0; i < MAX_NUMNODES; i++) { |
8342 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
8343 | pr_info(" Node %d: %#018Lx\n", i, |
8344 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 8345 | } |
c713216d | 8346 | |
f46edbd1 DW |
8347 | /* |
8348 | * Print out the early node map, and initialize the | |
8349 | * subsection-map relative to active online memory ranges to | |
8350 | * enable future "sub-section" extensions of the memory map. | |
8351 | */ | |
f88dfff5 | 8352 | pr_info("Early memory node ranges\n"); |
f46edbd1 | 8353 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
8d29e18a JG |
8354 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
8355 | (u64)start_pfn << PAGE_SHIFT, | |
8356 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
f46edbd1 DW |
8357 | subsection_map_init(start_pfn, end_pfn - start_pfn); |
8358 | } | |
c713216d MG |
8359 | |
8360 | /* Initialise every node */ | |
708614e6 | 8361 | mminit_verify_pageflags_layout(); |
8ef82866 | 8362 | setup_nr_node_ids(); |
09f49dca MH |
8363 | for_each_node(nid) { |
8364 | pg_data_t *pgdat; | |
8365 | ||
8366 | if (!node_online(nid)) { | |
8367 | pr_info("Initializing node %d as memoryless\n", nid); | |
8368 | ||
8369 | /* Allocator not initialized yet */ | |
8370 | pgdat = arch_alloc_nodedata(nid); | |
8371 | if (!pgdat) { | |
8372 | pr_err("Cannot allocate %zuB for node %d.\n", | |
8373 | sizeof(*pgdat), nid); | |
8374 | continue; | |
8375 | } | |
8376 | arch_refresh_nodedata(nid, pgdat); | |
8377 | free_area_init_memoryless_node(nid); | |
8378 | ||
8379 | /* | |
8380 | * We do not want to confuse userspace by sysfs | |
8381 | * files/directories for node without any memory | |
8382 | * attached to it, so this node is not marked as | |
8383 | * N_MEMORY and not marked online so that no sysfs | |
8384 | * hierarchy will be created via register_one_node for | |
8385 | * it. The pgdat will get fully initialized by | |
8386 | * hotadd_init_pgdat() when memory is hotplugged into | |
8387 | * this node. | |
8388 | */ | |
8389 | continue; | |
8390 | } | |
8391 | ||
8392 | pgdat = NODE_DATA(nid); | |
854e8848 | 8393 | free_area_init_node(nid); |
37b07e41 LS |
8394 | |
8395 | /* Any memory on that node */ | |
8396 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
8397 | node_set_state(nid, N_MEMORY); |
8398 | check_for_memory(pgdat, nid); | |
c713216d | 8399 | } |
122e093c MR |
8400 | |
8401 | memmap_init(); | |
c713216d | 8402 | } |
2a1e274a | 8403 | |
a5c6d650 DR |
8404 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
8405 | unsigned long *percent) | |
2a1e274a MG |
8406 | { |
8407 | unsigned long long coremem; | |
a5c6d650 DR |
8408 | char *endptr; |
8409 | ||
2a1e274a MG |
8410 | if (!p) |
8411 | return -EINVAL; | |
8412 | ||
a5c6d650 DR |
8413 | /* Value may be a percentage of total memory, otherwise bytes */ |
8414 | coremem = simple_strtoull(p, &endptr, 0); | |
8415 | if (*endptr == '%') { | |
8416 | /* Paranoid check for percent values greater than 100 */ | |
8417 | WARN_ON(coremem > 100); | |
2a1e274a | 8418 | |
a5c6d650 DR |
8419 | *percent = coremem; |
8420 | } else { | |
8421 | coremem = memparse(p, &p); | |
8422 | /* Paranoid check that UL is enough for the coremem value */ | |
8423 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 8424 | |
a5c6d650 DR |
8425 | *core = coremem >> PAGE_SHIFT; |
8426 | *percent = 0UL; | |
8427 | } | |
2a1e274a MG |
8428 | return 0; |
8429 | } | |
ed7ed365 | 8430 | |
7e63efef MG |
8431 | /* |
8432 | * kernelcore=size sets the amount of memory for use for allocations that | |
8433 | * cannot be reclaimed or migrated. | |
8434 | */ | |
8435 | static int __init cmdline_parse_kernelcore(char *p) | |
8436 | { | |
342332e6 TI |
8437 | /* parse kernelcore=mirror */ |
8438 | if (parse_option_str(p, "mirror")) { | |
8439 | mirrored_kernelcore = true; | |
8440 | return 0; | |
8441 | } | |
8442 | ||
a5c6d650 DR |
8443 | return cmdline_parse_core(p, &required_kernelcore, |
8444 | &required_kernelcore_percent); | |
7e63efef MG |
8445 | } |
8446 | ||
8447 | /* | |
8448 | * movablecore=size sets the amount of memory for use for allocations that | |
8449 | * can be reclaimed or migrated. | |
8450 | */ | |
8451 | static int __init cmdline_parse_movablecore(char *p) | |
8452 | { | |
a5c6d650 DR |
8453 | return cmdline_parse_core(p, &required_movablecore, |
8454 | &required_movablecore_percent); | |
7e63efef MG |
8455 | } |
8456 | ||
ed7ed365 | 8457 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 8458 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 8459 | |
c3d5f5f0 JL |
8460 | void adjust_managed_page_count(struct page *page, long count) |
8461 | { | |
9705bea5 | 8462 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 8463 | totalram_pages_add(count); |
3dcc0571 JL |
8464 | #ifdef CONFIG_HIGHMEM |
8465 | if (PageHighMem(page)) | |
ca79b0c2 | 8466 | totalhigh_pages_add(count); |
3dcc0571 | 8467 | #endif |
c3d5f5f0 | 8468 | } |
3dcc0571 | 8469 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 8470 | |
e5cb113f | 8471 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 8472 | { |
11199692 JL |
8473 | void *pos; |
8474 | unsigned long pages = 0; | |
69afade7 | 8475 | |
11199692 JL |
8476 | start = (void *)PAGE_ALIGN((unsigned long)start); |
8477 | end = (void *)((unsigned long)end & PAGE_MASK); | |
8478 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
8479 | struct page *page = virt_to_page(pos); |
8480 | void *direct_map_addr; | |
8481 | ||
8482 | /* | |
8483 | * 'direct_map_addr' might be different from 'pos' | |
8484 | * because some architectures' virt_to_page() | |
8485 | * work with aliases. Getting the direct map | |
8486 | * address ensures that we get a _writeable_ | |
8487 | * alias for the memset(). | |
8488 | */ | |
8489 | direct_map_addr = page_address(page); | |
c746170d VF |
8490 | /* |
8491 | * Perform a kasan-unchecked memset() since this memory | |
8492 | * has not been initialized. | |
8493 | */ | |
8494 | direct_map_addr = kasan_reset_tag(direct_map_addr); | |
dbe67df4 | 8495 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
8496 | memset(direct_map_addr, poison, PAGE_SIZE); |
8497 | ||
8498 | free_reserved_page(page); | |
69afade7 JL |
8499 | } |
8500 | ||
8501 | if (pages && s) | |
ff7ed9e4 | 8502 | pr_info("Freeing %s memory: %ldK\n", s, K(pages)); |
69afade7 JL |
8503 | |
8504 | return pages; | |
8505 | } | |
8506 | ||
1f9d03c5 | 8507 | void __init mem_init_print_info(void) |
7ee3d4e8 JL |
8508 | { |
8509 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
8510 | unsigned long init_code_size, init_data_size; | |
8511 | ||
8512 | physpages = get_num_physpages(); | |
8513 | codesize = _etext - _stext; | |
8514 | datasize = _edata - _sdata; | |
8515 | rosize = __end_rodata - __start_rodata; | |
8516 | bss_size = __bss_stop - __bss_start; | |
8517 | init_data_size = __init_end - __init_begin; | |
8518 | init_code_size = _einittext - _sinittext; | |
8519 | ||
8520 | /* | |
8521 | * Detect special cases and adjust section sizes accordingly: | |
8522 | * 1) .init.* may be embedded into .data sections | |
8523 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
8524 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
8525 | * 3) .rodata.* may be embedded into .text or .data sections. | |
8526 | */ | |
8527 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 | 8528 | do { \ |
ca831f29 | 8529 | if (&start[0] <= &pos[0] && &pos[0] < &end[0] && size > adj) \ |
b8af2941 PK |
8530 | size -= adj; \ |
8531 | } while (0) | |
7ee3d4e8 JL |
8532 | |
8533 | adj_init_size(__init_begin, __init_end, init_data_size, | |
8534 | _sinittext, init_code_size); | |
8535 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
8536 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
8537 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
8538 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
8539 | ||
8540 | #undef adj_init_size | |
8541 | ||
756a025f | 8542 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 8543 | #ifdef CONFIG_HIGHMEM |
756a025f | 8544 | ", %luK highmem" |
7ee3d4e8 | 8545 | #endif |
1f9d03c5 | 8546 | ")\n", |
ff7ed9e4 | 8547 | K(nr_free_pages()), K(physpages), |
c940e020 ML |
8548 | codesize / SZ_1K, datasize / SZ_1K, rosize / SZ_1K, |
8549 | (init_data_size + init_code_size) / SZ_1K, bss_size / SZ_1K, | |
ff7ed9e4 ML |
8550 | K(physpages - totalram_pages() - totalcma_pages), |
8551 | K(totalcma_pages) | |
7ee3d4e8 | 8552 | #ifdef CONFIG_HIGHMEM |
ff7ed9e4 | 8553 | , K(totalhigh_pages()) |
7ee3d4e8 | 8554 | #endif |
1f9d03c5 | 8555 | ); |
7ee3d4e8 JL |
8556 | } |
8557 | ||
0e0b864e | 8558 | /** |
88ca3b94 RD |
8559 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
8560 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 8561 | * |
013110a7 | 8562 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
8563 | * In the DMA zone, a significant percentage may be consumed by kernel image |
8564 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
8565 | * function may optionally be used to account for unfreeable pages in the |
8566 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
8567 | * smaller per-cpu batchsize. | |
0e0b864e MG |
8568 | */ |
8569 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
8570 | { | |
8571 | dma_reserve = new_dma_reserve; | |
8572 | } | |
8573 | ||
005fd4bb | 8574 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 8575 | { |
04f8cfea | 8576 | struct zone *zone; |
1da177e4 | 8577 | |
005fd4bb | 8578 | lru_add_drain_cpu(cpu); |
adb11e78 | 8579 | mlock_page_drain_remote(cpu); |
005fd4bb | 8580 | drain_pages(cpu); |
9f8f2172 | 8581 | |
005fd4bb SAS |
8582 | /* |
8583 | * Spill the event counters of the dead processor | |
8584 | * into the current processors event counters. | |
8585 | * This artificially elevates the count of the current | |
8586 | * processor. | |
8587 | */ | |
8588 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 8589 | |
005fd4bb SAS |
8590 | /* |
8591 | * Zero the differential counters of the dead processor | |
8592 | * so that the vm statistics are consistent. | |
8593 | * | |
8594 | * This is only okay since the processor is dead and cannot | |
8595 | * race with what we are doing. | |
8596 | */ | |
8597 | cpu_vm_stats_fold(cpu); | |
04f8cfea MG |
8598 | |
8599 | for_each_populated_zone(zone) | |
8600 | zone_pcp_update(zone, 0); | |
8601 | ||
8602 | return 0; | |
8603 | } | |
8604 | ||
8605 | static int page_alloc_cpu_online(unsigned int cpu) | |
8606 | { | |
8607 | struct zone *zone; | |
8608 | ||
8609 | for_each_populated_zone(zone) | |
8610 | zone_pcp_update(zone, 1); | |
005fd4bb | 8611 | return 0; |
1da177e4 | 8612 | } |
1da177e4 | 8613 | |
e03a5125 NP |
8614 | #ifdef CONFIG_NUMA |
8615 | int hashdist = HASHDIST_DEFAULT; | |
8616 | ||
8617 | static int __init set_hashdist(char *str) | |
8618 | { | |
8619 | if (!str) | |
8620 | return 0; | |
8621 | hashdist = simple_strtoul(str, &str, 0); | |
8622 | return 1; | |
8623 | } | |
8624 | __setup("hashdist=", set_hashdist); | |
8625 | #endif | |
8626 | ||
1da177e4 LT |
8627 | void __init page_alloc_init(void) |
8628 | { | |
005fd4bb SAS |
8629 | int ret; |
8630 | ||
e03a5125 NP |
8631 | #ifdef CONFIG_NUMA |
8632 | if (num_node_state(N_MEMORY) == 1) | |
8633 | hashdist = 0; | |
8634 | #endif | |
8635 | ||
04f8cfea MG |
8636 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC, |
8637 | "mm/page_alloc:pcp", | |
8638 | page_alloc_cpu_online, | |
005fd4bb SAS |
8639 | page_alloc_cpu_dead); |
8640 | WARN_ON(ret < 0); | |
1da177e4 LT |
8641 | } |
8642 | ||
cb45b0e9 | 8643 | /* |
34b10060 | 8644 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
8645 | * or min_free_kbytes changes. |
8646 | */ | |
8647 | static void calculate_totalreserve_pages(void) | |
8648 | { | |
8649 | struct pglist_data *pgdat; | |
8650 | unsigned long reserve_pages = 0; | |
2f6726e5 | 8651 | enum zone_type i, j; |
cb45b0e9 HA |
8652 | |
8653 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
8654 | |
8655 | pgdat->totalreserve_pages = 0; | |
8656 | ||
cb45b0e9 HA |
8657 | for (i = 0; i < MAX_NR_ZONES; i++) { |
8658 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 8659 | long max = 0; |
9705bea5 | 8660 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
8661 | |
8662 | /* Find valid and maximum lowmem_reserve in the zone */ | |
8663 | for (j = i; j < MAX_NR_ZONES; j++) { | |
8664 | if (zone->lowmem_reserve[j] > max) | |
8665 | max = zone->lowmem_reserve[j]; | |
8666 | } | |
8667 | ||
41858966 MG |
8668 | /* we treat the high watermark as reserved pages. */ |
8669 | max += high_wmark_pages(zone); | |
cb45b0e9 | 8670 | |
3d6357de AK |
8671 | if (max > managed_pages) |
8672 | max = managed_pages; | |
a8d01437 | 8673 | |
281e3726 | 8674 | pgdat->totalreserve_pages += max; |
a8d01437 | 8675 | |
cb45b0e9 HA |
8676 | reserve_pages += max; |
8677 | } | |
8678 | } | |
8679 | totalreserve_pages = reserve_pages; | |
8680 | } | |
8681 | ||
1da177e4 LT |
8682 | /* |
8683 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 8684 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
8685 | * has a correct pages reserved value, so an adequate number of |
8686 | * pages are left in the zone after a successful __alloc_pages(). | |
8687 | */ | |
8688 | static void setup_per_zone_lowmem_reserve(void) | |
8689 | { | |
8690 | struct pglist_data *pgdat; | |
470c61d7 | 8691 | enum zone_type i, j; |
1da177e4 | 8692 | |
ec936fc5 | 8693 | for_each_online_pgdat(pgdat) { |
470c61d7 LS |
8694 | for (i = 0; i < MAX_NR_ZONES - 1; i++) { |
8695 | struct zone *zone = &pgdat->node_zones[i]; | |
8696 | int ratio = sysctl_lowmem_reserve_ratio[i]; | |
8697 | bool clear = !ratio || !zone_managed_pages(zone); | |
8698 | unsigned long managed_pages = 0; | |
8699 | ||
8700 | for (j = i + 1; j < MAX_NR_ZONES; j++) { | |
f7ec1044 LS |
8701 | struct zone *upper_zone = &pgdat->node_zones[j]; |
8702 | ||
8703 | managed_pages += zone_managed_pages(upper_zone); | |
470c61d7 | 8704 | |
f7ec1044 LS |
8705 | if (clear) |
8706 | zone->lowmem_reserve[j] = 0; | |
8707 | else | |
470c61d7 | 8708 | zone->lowmem_reserve[j] = managed_pages / ratio; |
1da177e4 LT |
8709 | } |
8710 | } | |
8711 | } | |
cb45b0e9 HA |
8712 | |
8713 | /* update totalreserve_pages */ | |
8714 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8715 | } |
8716 | ||
cfd3da1e | 8717 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
8718 | { |
8719 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
8720 | unsigned long lowmem_pages = 0; | |
8721 | struct zone *zone; | |
8722 | unsigned long flags; | |
8723 | ||
8724 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
8725 | for_each_zone(zone) { | |
8726 | if (!is_highmem(zone)) | |
9705bea5 | 8727 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
8728 | } |
8729 | ||
8730 | for_each_zone(zone) { | |
ac924c60 AM |
8731 | u64 tmp; |
8732 | ||
1125b4e3 | 8733 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 8734 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 8735 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
8736 | if (is_highmem(zone)) { |
8737 | /* | |
669ed175 NP |
8738 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
8739 | * need highmem pages, so cap pages_min to a small | |
8740 | * value here. | |
8741 | * | |
41858966 | 8742 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 8743 | * deltas control async page reclaim, and so should |
669ed175 | 8744 | * not be capped for highmem. |
1da177e4 | 8745 | */ |
90ae8d67 | 8746 | unsigned long min_pages; |
1da177e4 | 8747 | |
9705bea5 | 8748 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 8749 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 8750 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 8751 | } else { |
669ed175 NP |
8752 | /* |
8753 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
8754 | * proportionate to the zone's size. |
8755 | */ | |
a9214443 | 8756 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
8757 | } |
8758 | ||
795ae7a0 JW |
8759 | /* |
8760 | * Set the kswapd watermarks distance according to the | |
8761 | * scale factor in proportion to available memory, but | |
8762 | * ensure a minimum size on small systems. | |
8763 | */ | |
8764 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 8765 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
8766 | watermark_scale_factor, 10000)); |
8767 | ||
aa092591 | 8768 | zone->watermark_boost = 0; |
a9214443 | 8769 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
c574bbe9 HY |
8770 | zone->_watermark[WMARK_HIGH] = low_wmark_pages(zone) + tmp; |
8771 | zone->_watermark[WMARK_PROMO] = high_wmark_pages(zone) + tmp; | |
49f223a9 | 8772 | |
1125b4e3 | 8773 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 8774 | } |
cb45b0e9 HA |
8775 | |
8776 | /* update totalreserve_pages */ | |
8777 | calculate_totalreserve_pages(); | |
1da177e4 LT |
8778 | } |
8779 | ||
cfd3da1e MG |
8780 | /** |
8781 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
8782 | * or when memory is hot-{added|removed} | |
8783 | * | |
8784 | * Ensures that the watermark[min,low,high] values for each zone are set | |
8785 | * correctly with respect to min_free_kbytes. | |
8786 | */ | |
8787 | void setup_per_zone_wmarks(void) | |
8788 | { | |
b92ca18e | 8789 | struct zone *zone; |
b93e0f32 MH |
8790 | static DEFINE_SPINLOCK(lock); |
8791 | ||
8792 | spin_lock(&lock); | |
cfd3da1e | 8793 | __setup_per_zone_wmarks(); |
b93e0f32 | 8794 | spin_unlock(&lock); |
b92ca18e MG |
8795 | |
8796 | /* | |
8797 | * The watermark size have changed so update the pcpu batch | |
8798 | * and high limits or the limits may be inappropriate. | |
8799 | */ | |
8800 | for_each_zone(zone) | |
04f8cfea | 8801 | zone_pcp_update(zone, 0); |
cfd3da1e MG |
8802 | } |
8803 | ||
1da177e4 LT |
8804 | /* |
8805 | * Initialise min_free_kbytes. | |
8806 | * | |
8807 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 8808 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
8809 | * bandwidth does not increase linearly with machine size. We use |
8810 | * | |
b8af2941 | 8811 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
8812 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
8813 | * | |
8814 | * which yields | |
8815 | * | |
8816 | * 16MB: 512k | |
8817 | * 32MB: 724k | |
8818 | * 64MB: 1024k | |
8819 | * 128MB: 1448k | |
8820 | * 256MB: 2048k | |
8821 | * 512MB: 2896k | |
8822 | * 1024MB: 4096k | |
8823 | * 2048MB: 5792k | |
8824 | * 4096MB: 8192k | |
8825 | * 8192MB: 11584k | |
8826 | * 16384MB: 16384k | |
8827 | */ | |
bd3400ea | 8828 | void calculate_min_free_kbytes(void) |
1da177e4 LT |
8829 | { |
8830 | unsigned long lowmem_kbytes; | |
5f12733e | 8831 | int new_min_free_kbytes; |
1da177e4 LT |
8832 | |
8833 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
8834 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
8835 | ||
59d336bd WS |
8836 | if (new_min_free_kbytes > user_min_free_kbytes) |
8837 | min_free_kbytes = clamp(new_min_free_kbytes, 128, 262144); | |
8838 | else | |
5f12733e MH |
8839 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", |
8840 | new_min_free_kbytes, user_min_free_kbytes); | |
59d336bd | 8841 | |
bd3400ea LF |
8842 | } |
8843 | ||
8844 | int __meminit init_per_zone_wmark_min(void) | |
8845 | { | |
8846 | calculate_min_free_kbytes(); | |
bc75d33f | 8847 | setup_per_zone_wmarks(); |
a6cccdc3 | 8848 | refresh_zone_stat_thresholds(); |
1da177e4 | 8849 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
8850 | |
8851 | #ifdef CONFIG_NUMA | |
8852 | setup_min_unmapped_ratio(); | |
8853 | setup_min_slab_ratio(); | |
8854 | #endif | |
8855 | ||
4aab2be0 VB |
8856 | khugepaged_min_free_kbytes_update(); |
8857 | ||
1da177e4 LT |
8858 | return 0; |
8859 | } | |
e08d3fdf | 8860 | postcore_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
8861 | |
8862 | /* | |
b8af2941 | 8863 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
8864 | * that we can call two helper functions whenever min_free_kbytes |
8865 | * changes. | |
8866 | */ | |
cccad5b9 | 8867 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8868 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8869 | { |
da8c757b HP |
8870 | int rc; |
8871 | ||
8872 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8873 | if (rc) | |
8874 | return rc; | |
8875 | ||
5f12733e MH |
8876 | if (write) { |
8877 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 8878 | setup_per_zone_wmarks(); |
5f12733e | 8879 | } |
1da177e4 LT |
8880 | return 0; |
8881 | } | |
8882 | ||
795ae7a0 | 8883 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8884 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
8885 | { |
8886 | int rc; | |
8887 | ||
8888 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8889 | if (rc) | |
8890 | return rc; | |
8891 | ||
8892 | if (write) | |
8893 | setup_per_zone_wmarks(); | |
8894 | ||
8895 | return 0; | |
8896 | } | |
8897 | ||
9614634f | 8898 | #ifdef CONFIG_NUMA |
6423aa81 | 8899 | static void setup_min_unmapped_ratio(void) |
9614634f | 8900 | { |
6423aa81 | 8901 | pg_data_t *pgdat; |
9614634f | 8902 | struct zone *zone; |
9614634f | 8903 | |
a5f5f91d | 8904 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 8905 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 8906 | |
9614634f | 8907 | for_each_zone(zone) |
9705bea5 AK |
8908 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
8909 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 8910 | } |
0ff38490 | 8911 | |
6423aa81 JK |
8912 | |
8913 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8914 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 8915 | { |
0ff38490 CL |
8916 | int rc; |
8917 | ||
8d65af78 | 8918 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
8919 | if (rc) |
8920 | return rc; | |
8921 | ||
6423aa81 JK |
8922 | setup_min_unmapped_ratio(); |
8923 | ||
8924 | return 0; | |
8925 | } | |
8926 | ||
8927 | static void setup_min_slab_ratio(void) | |
8928 | { | |
8929 | pg_data_t *pgdat; | |
8930 | struct zone *zone; | |
8931 | ||
a5f5f91d MG |
8932 | for_each_online_pgdat(pgdat) |
8933 | pgdat->min_slab_pages = 0; | |
8934 | ||
0ff38490 | 8935 | for_each_zone(zone) |
9705bea5 AK |
8936 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
8937 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
8938 | } |
8939 | ||
8940 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 8941 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
8942 | { |
8943 | int rc; | |
8944 | ||
8945 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
8946 | if (rc) | |
8947 | return rc; | |
8948 | ||
8949 | setup_min_slab_ratio(); | |
8950 | ||
0ff38490 CL |
8951 | return 0; |
8952 | } | |
9614634f CL |
8953 | #endif |
8954 | ||
1da177e4 LT |
8955 | /* |
8956 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
8957 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
8958 | * whenever sysctl_lowmem_reserve_ratio changes. | |
8959 | * | |
8960 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 8961 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
8962 | * if in function of the boot time zone sizes. |
8963 | */ | |
cccad5b9 | 8964 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8965 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8966 | { |
86aaf255 BH |
8967 | int i; |
8968 | ||
8d65af78 | 8969 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
8970 | |
8971 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
8972 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
8973 | sysctl_lowmem_reserve_ratio[i] = 0; | |
8974 | } | |
8975 | ||
1da177e4 LT |
8976 | setup_per_zone_lowmem_reserve(); |
8977 | return 0; | |
8978 | } | |
8979 | ||
8ad4b1fb | 8980 | /* |
74f44822 MG |
8981 | * percpu_pagelist_high_fraction - changes the pcp->high for each zone on each |
8982 | * cpu. It is the fraction of total pages in each zone that a hot per cpu | |
b8af2941 | 8983 | * pagelist can have before it gets flushed back to buddy allocator. |
8ad4b1fb | 8984 | */ |
74f44822 MG |
8985 | int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *table, |
8986 | int write, void *buffer, size_t *length, loff_t *ppos) | |
8ad4b1fb RS |
8987 | { |
8988 | struct zone *zone; | |
74f44822 | 8989 | int old_percpu_pagelist_high_fraction; |
8ad4b1fb RS |
8990 | int ret; |
8991 | ||
7cd2b0a3 | 8992 | mutex_lock(&pcp_batch_high_lock); |
74f44822 | 8993 | old_percpu_pagelist_high_fraction = percpu_pagelist_high_fraction; |
7cd2b0a3 | 8994 | |
8d65af78 | 8995 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
8996 | if (!write || ret < 0) |
8997 | goto out; | |
8998 | ||
8999 | /* Sanity checking to avoid pcp imbalance */ | |
74f44822 MG |
9000 | if (percpu_pagelist_high_fraction && |
9001 | percpu_pagelist_high_fraction < MIN_PERCPU_PAGELIST_HIGH_FRACTION) { | |
9002 | percpu_pagelist_high_fraction = old_percpu_pagelist_high_fraction; | |
7cd2b0a3 DR |
9003 | ret = -EINVAL; |
9004 | goto out; | |
9005 | } | |
9006 | ||
9007 | /* No change? */ | |
74f44822 | 9008 | if (percpu_pagelist_high_fraction == old_percpu_pagelist_high_fraction) |
7cd2b0a3 | 9009 | goto out; |
c8e251fa | 9010 | |
cb1ef534 | 9011 | for_each_populated_zone(zone) |
74f44822 | 9012 | zone_set_pageset_high_and_batch(zone, 0); |
7cd2b0a3 | 9013 | out: |
c8e251fa | 9014 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 9015 | return ret; |
8ad4b1fb RS |
9016 | } |
9017 | ||
f6f34b43 SD |
9018 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
9019 | /* | |
9020 | * Returns the number of pages that arch has reserved but | |
9021 | * is not known to alloc_large_system_hash(). | |
9022 | */ | |
9023 | static unsigned long __init arch_reserved_kernel_pages(void) | |
9024 | { | |
9025 | return 0; | |
9026 | } | |
9027 | #endif | |
9028 | ||
9017217b PT |
9029 | /* |
9030 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
9031 | * machines. As memory size is increased the scale is also increased but at | |
9032 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
9033 | * quadruples the scale is increased by one, which means the size of hash table | |
9034 | * only doubles, instead of quadrupling as well. | |
9035 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
9036 | * makes sense, it is disabled on such platforms. | |
9037 | */ | |
9038 | #if __BITS_PER_LONG > 32 | |
9039 | #define ADAPT_SCALE_BASE (64ul << 30) | |
9040 | #define ADAPT_SCALE_SHIFT 2 | |
9041 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
9042 | #endif | |
9043 | ||
1da177e4 LT |
9044 | /* |
9045 | * allocate a large system hash table from bootmem | |
9046 | * - it is assumed that the hash table must contain an exact power-of-2 | |
9047 | * quantity of entries | |
9048 | * - limit is the number of hash buckets, not the total allocation size | |
9049 | */ | |
9050 | void *__init alloc_large_system_hash(const char *tablename, | |
9051 | unsigned long bucketsize, | |
9052 | unsigned long numentries, | |
9053 | int scale, | |
9054 | int flags, | |
9055 | unsigned int *_hash_shift, | |
9056 | unsigned int *_hash_mask, | |
31fe62b9 TB |
9057 | unsigned long low_limit, |
9058 | unsigned long high_limit) | |
1da177e4 | 9059 | { |
31fe62b9 | 9060 | unsigned long long max = high_limit; |
1da177e4 | 9061 | unsigned long log2qty, size; |
97bab178 | 9062 | void *table; |
3749a8f0 | 9063 | gfp_t gfp_flags; |
ec11408a | 9064 | bool virt; |
121e6f32 | 9065 | bool huge; |
1da177e4 LT |
9066 | |
9067 | /* allow the kernel cmdline to have a say */ | |
9068 | if (!numentries) { | |
9069 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 9070 | numentries = nr_kernel_pages; |
f6f34b43 | 9071 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
9072 | |
9073 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
c940e020 ML |
9074 | if (PAGE_SIZE < SZ_1M) |
9075 | numentries = round_up(numentries, SZ_1M / PAGE_SIZE); | |
1da177e4 | 9076 | |
9017217b PT |
9077 | #if __BITS_PER_LONG > 32 |
9078 | if (!high_limit) { | |
9079 | unsigned long adapt; | |
9080 | ||
9081 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
9082 | adapt <<= ADAPT_SCALE_SHIFT) | |
9083 | scale++; | |
9084 | } | |
9085 | #endif | |
9086 | ||
1da177e4 LT |
9087 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
9088 | if (scale > PAGE_SHIFT) | |
9089 | numentries >>= (scale - PAGE_SHIFT); | |
9090 | else | |
9091 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
9092 | |
9093 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
9094 | if (unlikely(flags & HASH_SMALL)) { |
9095 | /* Makes no sense without HASH_EARLY */ | |
9096 | WARN_ON(!(flags & HASH_EARLY)); | |
9097 | if (!(numentries >> *_hash_shift)) { | |
9098 | numentries = 1UL << *_hash_shift; | |
9099 | BUG_ON(!numentries); | |
9100 | } | |
9101 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 9102 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 9103 | } |
6e692ed3 | 9104 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
9105 | |
9106 | /* limit allocation size to 1/16 total memory by default */ | |
9107 | if (max == 0) { | |
9108 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
9109 | do_div(max, bucketsize); | |
9110 | } | |
074b8517 | 9111 | max = min(max, 0x80000000ULL); |
1da177e4 | 9112 | |
31fe62b9 TB |
9113 | if (numentries < low_limit) |
9114 | numentries = low_limit; | |
1da177e4 LT |
9115 | if (numentries > max) |
9116 | numentries = max; | |
9117 | ||
f0d1b0b3 | 9118 | log2qty = ilog2(numentries); |
1da177e4 | 9119 | |
3749a8f0 | 9120 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 9121 | do { |
ec11408a | 9122 | virt = false; |
1da177e4 | 9123 | size = bucketsize << log2qty; |
ea1f5f37 PT |
9124 | if (flags & HASH_EARLY) { |
9125 | if (flags & HASH_ZERO) | |
26fb3dae | 9126 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 9127 | else |
7e1c4e27 MR |
9128 | table = memblock_alloc_raw(size, |
9129 | SMP_CACHE_BYTES); | |
ec11408a | 9130 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
f2edd118 | 9131 | table = vmalloc_huge(size, gfp_flags); |
ec11408a | 9132 | virt = true; |
084f7e23 ED |
9133 | if (table) |
9134 | huge = is_vm_area_hugepages(table); | |
ea1f5f37 | 9135 | } else { |
1037b83b ED |
9136 | /* |
9137 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
9138 | * some pages at the end of hash table which |
9139 | * alloc_pages_exact() automatically does | |
1037b83b | 9140 | */ |
ec11408a NP |
9141 | table = alloc_pages_exact(size, gfp_flags); |
9142 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
9143 | } |
9144 | } while (!table && size > PAGE_SIZE && --log2qty); | |
9145 | ||
9146 | if (!table) | |
9147 | panic("Failed to allocate %s hash table\n", tablename); | |
9148 | ||
ec11408a NP |
9149 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
9150 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
121e6f32 | 9151 | virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear"); |
1da177e4 LT |
9152 | |
9153 | if (_hash_shift) | |
9154 | *_hash_shift = log2qty; | |
9155 | if (_hash_mask) | |
9156 | *_hash_mask = (1 << log2qty) - 1; | |
9157 | ||
9158 | return table; | |
9159 | } | |
a117e66e | 9160 | |
8df995f6 | 9161 | #ifdef CONFIG_CONTIG_ALLOC |
a1394bdd MK |
9162 | #if defined(CONFIG_DYNAMIC_DEBUG) || \ |
9163 | (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) | |
9164 | /* Usage: See admin-guide/dynamic-debug-howto.rst */ | |
9165 | static void alloc_contig_dump_pages(struct list_head *page_list) | |
9166 | { | |
9167 | DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure"); | |
9168 | ||
9169 | if (DYNAMIC_DEBUG_BRANCH(descriptor)) { | |
9170 | struct page *page; | |
9171 | ||
9172 | dump_stack(); | |
9173 | list_for_each_entry(page, page_list, lru) | |
9174 | dump_page(page, "migration failure"); | |
9175 | } | |
9176 | } | |
9177 | #else | |
9178 | static inline void alloc_contig_dump_pages(struct list_head *page_list) | |
9179 | { | |
9180 | } | |
9181 | #endif | |
9182 | ||
041d3a8c | 9183 | /* [start, end) must belong to a single zone. */ |
b2c9e2fb | 9184 | int __alloc_contig_migrate_range(struct compact_control *cc, |
bb13ffeb | 9185 | unsigned long start, unsigned long end) |
041d3a8c MN |
9186 | { |
9187 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 9188 | unsigned int nr_reclaimed; |
041d3a8c MN |
9189 | unsigned long pfn = start; |
9190 | unsigned int tries = 0; | |
9191 | int ret = 0; | |
8b94e0b8 JK |
9192 | struct migration_target_control mtc = { |
9193 | .nid = zone_to_nid(cc->zone), | |
9194 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
9195 | }; | |
041d3a8c | 9196 | |
361a2a22 | 9197 | lru_cache_disable(); |
041d3a8c | 9198 | |
bb13ffeb | 9199 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
9200 | if (fatal_signal_pending(current)) { |
9201 | ret = -EINTR; | |
9202 | break; | |
9203 | } | |
9204 | ||
bb13ffeb MG |
9205 | if (list_empty(&cc->migratepages)) { |
9206 | cc->nr_migratepages = 0; | |
c2ad7a1f OS |
9207 | ret = isolate_migratepages_range(cc, pfn, end); |
9208 | if (ret && ret != -EAGAIN) | |
041d3a8c | 9209 | break; |
c2ad7a1f | 9210 | pfn = cc->migrate_pfn; |
041d3a8c MN |
9211 | tries = 0; |
9212 | } else if (++tries == 5) { | |
c8e28b47 | 9213 | ret = -EBUSY; |
041d3a8c MN |
9214 | break; |
9215 | } | |
9216 | ||
beb51eaa MK |
9217 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
9218 | &cc->migratepages); | |
9219 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 9220 | |
8b94e0b8 | 9221 | ret = migrate_pages(&cc->migratepages, alloc_migration_target, |
5ac95884 | 9222 | NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE, NULL); |
c8e28b47 OS |
9223 | |
9224 | /* | |
9225 | * On -ENOMEM, migrate_pages() bails out right away. It is pointless | |
9226 | * to retry again over this error, so do the same here. | |
9227 | */ | |
9228 | if (ret == -ENOMEM) | |
9229 | break; | |
041d3a8c | 9230 | } |
d479960e | 9231 | |
361a2a22 | 9232 | lru_cache_enable(); |
2a6f5124 | 9233 | if (ret < 0) { |
3f913fc5 | 9234 | if (!(cc->gfp_mask & __GFP_NOWARN) && ret == -EBUSY) |
151e084a | 9235 | alloc_contig_dump_pages(&cc->migratepages); |
2a6f5124 SP |
9236 | putback_movable_pages(&cc->migratepages); |
9237 | return ret; | |
9238 | } | |
9239 | return 0; | |
041d3a8c MN |
9240 | } |
9241 | ||
9242 | /** | |
9243 | * alloc_contig_range() -- tries to allocate given range of pages | |
9244 | * @start: start PFN to allocate | |
9245 | * @end: one-past-the-last PFN to allocate | |
f0953a1b | 9246 | * @migratetype: migratetype of the underlying pageblocks (either |
0815f3d8 MN |
9247 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
9248 | * in range must have the same migratetype and it must | |
9249 | * be either of the two. | |
ca96b625 | 9250 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c | 9251 | * |
11ac3e87 ZY |
9252 | * The PFN range does not have to be pageblock aligned. The PFN range must |
9253 | * belong to a single zone. | |
041d3a8c | 9254 | * |
2c7452a0 MK |
9255 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
9256 | * pageblocks in the range. Once isolated, the pageblocks should not | |
9257 | * be modified by others. | |
041d3a8c | 9258 | * |
a862f68a | 9259 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
9260 | * pages which PFN is in [start, end) are allocated for the caller and |
9261 | * need to be freed with free_contig_range(). | |
9262 | */ | |
0815f3d8 | 9263 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 9264 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 9265 | { |
041d3a8c | 9266 | unsigned long outer_start, outer_end; |
b2c9e2fb | 9267 | int order; |
d00181b9 | 9268 | int ret = 0; |
041d3a8c | 9269 | |
bb13ffeb MG |
9270 | struct compact_control cc = { |
9271 | .nr_migratepages = 0, | |
9272 | .order = -1, | |
9273 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 9274 | .mode = MIGRATE_SYNC, |
bb13ffeb | 9275 | .ignore_skip_hint = true, |
2583d671 | 9276 | .no_set_skip_hint = true, |
7dea19f9 | 9277 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 9278 | .alloc_contig = true, |
bb13ffeb MG |
9279 | }; |
9280 | INIT_LIST_HEAD(&cc.migratepages); | |
9281 | ||
041d3a8c MN |
9282 | /* |
9283 | * What we do here is we mark all pageblocks in range as | |
9284 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
9285 | * have different sizes, and due to the way page allocator | |
b2c9e2fb | 9286 | * work, start_isolate_page_range() has special handlings for this. |
041d3a8c MN |
9287 | * |
9288 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
9289 | * migrate the pages from an unaligned range (ie. pages that | |
b2c9e2fb | 9290 | * we are interested in). This will put all the pages in |
041d3a8c MN |
9291 | * range back to page allocator as MIGRATE_ISOLATE. |
9292 | * | |
9293 | * When this is done, we take the pages in range from page | |
9294 | * allocator removing them from the buddy system. This way | |
9295 | * page allocator will never consider using them. | |
9296 | * | |
9297 | * This lets us mark the pageblocks back as | |
9298 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
9299 | * aligned range but not in the unaligned, original range are | |
9300 | * put back to page allocator so that buddy can use them. | |
9301 | */ | |
9302 | ||
6e263fff | 9303 | ret = start_isolate_page_range(start, end, migratetype, 0, gfp_mask); |
3fa0c7c7 | 9304 | if (ret) |
b2c9e2fb | 9305 | goto done; |
041d3a8c | 9306 | |
7612921f VB |
9307 | drain_all_pages(cc.zone); |
9308 | ||
8ef5849f JK |
9309 | /* |
9310 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
9311 | * So, just fall through. test_pages_isolated() has a tracepoint |
9312 | * which will report the busy page. | |
9313 | * | |
9314 | * It is possible that busy pages could become available before | |
9315 | * the call to test_pages_isolated, and the range will actually be | |
9316 | * allocated. So, if we fall through be sure to clear ret so that | |
9317 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 9318 | */ |
bb13ffeb | 9319 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 9320 | if (ret && ret != -EBUSY) |
041d3a8c | 9321 | goto done; |
68d68ff6 | 9322 | ret = 0; |
041d3a8c MN |
9323 | |
9324 | /* | |
b2c9e2fb | 9325 | * Pages from [start, end) are within a pageblock_nr_pages |
041d3a8c MN |
9326 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's |
9327 | * more, all pages in [start, end) are free in page allocator. | |
9328 | * What we are going to do is to allocate all pages from | |
9329 | * [start, end) (that is remove them from page allocator). | |
9330 | * | |
9331 | * The only problem is that pages at the beginning and at the | |
9332 | * end of interesting range may be not aligned with pages that | |
9333 | * page allocator holds, ie. they can be part of higher order | |
9334 | * pages. Because of this, we reserve the bigger range and | |
9335 | * once this is done free the pages we are not interested in. | |
9336 | * | |
9337 | * We don't have to hold zone->lock here because the pages are | |
9338 | * isolated thus they won't get removed from buddy. | |
9339 | */ | |
9340 | ||
041d3a8c MN |
9341 | order = 0; |
9342 | outer_start = start; | |
9343 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
9344 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
9345 | outer_start = start; |
9346 | break; | |
041d3a8c MN |
9347 | } |
9348 | outer_start &= ~0UL << order; | |
9349 | } | |
9350 | ||
8ef5849f | 9351 | if (outer_start != start) { |
ab130f91 | 9352 | order = buddy_order(pfn_to_page(outer_start)); |
8ef5849f JK |
9353 | |
9354 | /* | |
9355 | * outer_start page could be small order buddy page and | |
9356 | * it doesn't include start page. Adjust outer_start | |
9357 | * in this case to report failed page properly | |
9358 | * on tracepoint in test_pages_isolated() | |
9359 | */ | |
9360 | if (outer_start + (1UL << order) <= start) | |
9361 | outer_start = start; | |
9362 | } | |
9363 | ||
041d3a8c | 9364 | /* Make sure the range is really isolated. */ |
756d25be | 9365 | if (test_pages_isolated(outer_start, end, 0)) { |
041d3a8c MN |
9366 | ret = -EBUSY; |
9367 | goto done; | |
9368 | } | |
9369 | ||
49f223a9 | 9370 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 9371 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
9372 | if (!outer_end) { |
9373 | ret = -EBUSY; | |
9374 | goto done; | |
9375 | } | |
9376 | ||
9377 | /* Free head and tail (if any) */ | |
9378 | if (start != outer_start) | |
9379 | free_contig_range(outer_start, start - outer_start); | |
9380 | if (end != outer_end) | |
9381 | free_contig_range(end, outer_end - end); | |
9382 | ||
9383 | done: | |
6e263fff | 9384 | undo_isolate_page_range(start, end, migratetype); |
041d3a8c MN |
9385 | return ret; |
9386 | } | |
255f5985 | 9387 | EXPORT_SYMBOL(alloc_contig_range); |
5e27a2df AK |
9388 | |
9389 | static int __alloc_contig_pages(unsigned long start_pfn, | |
9390 | unsigned long nr_pages, gfp_t gfp_mask) | |
9391 | { | |
9392 | unsigned long end_pfn = start_pfn + nr_pages; | |
9393 | ||
9394 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
9395 | gfp_mask); | |
9396 | } | |
9397 | ||
9398 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
9399 | unsigned long nr_pages) | |
9400 | { | |
9401 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
9402 | struct page *page; | |
9403 | ||
9404 | for (i = start_pfn; i < end_pfn; i++) { | |
9405 | page = pfn_to_online_page(i); | |
9406 | if (!page) | |
9407 | return false; | |
9408 | ||
9409 | if (page_zone(page) != z) | |
9410 | return false; | |
9411 | ||
9412 | if (PageReserved(page)) | |
9413 | return false; | |
5e27a2df AK |
9414 | } |
9415 | return true; | |
9416 | } | |
9417 | ||
9418 | static bool zone_spans_last_pfn(const struct zone *zone, | |
9419 | unsigned long start_pfn, unsigned long nr_pages) | |
9420 | { | |
9421 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
9422 | ||
9423 | return zone_spans_pfn(zone, last_pfn); | |
9424 | } | |
9425 | ||
9426 | /** | |
9427 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
9428 | * @nr_pages: Number of contiguous pages to allocate | |
9429 | * @gfp_mask: GFP mask to limit search and used during compaction | |
9430 | * @nid: Target node | |
9431 | * @nodemask: Mask for other possible nodes | |
9432 | * | |
9433 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
9434 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
9435 | * tried for allocation with alloc_contig_range(). This routine is intended | |
9436 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
9437 | * | |
9438 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
eaab8e75 AK |
9439 | * power of two, then allocated range is also guaranteed to be aligned to same |
9440 | * nr_pages (e.g. 1GB request would be aligned to 1GB). | |
5e27a2df AK |
9441 | * |
9442 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
9443 | * __free_page() on each allocated page. | |
9444 | * | |
9445 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
9446 | */ | |
9447 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
9448 | int nid, nodemask_t *nodemask) | |
9449 | { | |
9450 | unsigned long ret, pfn, flags; | |
9451 | struct zonelist *zonelist; | |
9452 | struct zone *zone; | |
9453 | struct zoneref *z; | |
9454 | ||
9455 | zonelist = node_zonelist(nid, gfp_mask); | |
9456 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
9457 | gfp_zone(gfp_mask), nodemask) { | |
9458 | spin_lock_irqsave(&zone->lock, flags); | |
9459 | ||
9460 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
9461 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
9462 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
9463 | /* | |
9464 | * We release the zone lock here because | |
9465 | * alloc_contig_range() will also lock the zone | |
9466 | * at some point. If there's an allocation | |
9467 | * spinning on this lock, it may win the race | |
9468 | * and cause alloc_contig_range() to fail... | |
9469 | */ | |
9470 | spin_unlock_irqrestore(&zone->lock, flags); | |
9471 | ret = __alloc_contig_pages(pfn, nr_pages, | |
9472 | gfp_mask); | |
9473 | if (!ret) | |
9474 | return pfn_to_page(pfn); | |
9475 | spin_lock_irqsave(&zone->lock, flags); | |
9476 | } | |
9477 | pfn += nr_pages; | |
9478 | } | |
9479 | spin_unlock_irqrestore(&zone->lock, flags); | |
9480 | } | |
9481 | return NULL; | |
9482 | } | |
4eb0716e | 9483 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 9484 | |
78fa5150 | 9485 | void free_contig_range(unsigned long pfn, unsigned long nr_pages) |
041d3a8c | 9486 | { |
78fa5150 | 9487 | unsigned long count = 0; |
bcc2b02f MS |
9488 | |
9489 | for (; nr_pages--; pfn++) { | |
9490 | struct page *page = pfn_to_page(pfn); | |
9491 | ||
9492 | count += page_count(page) != 1; | |
9493 | __free_page(page); | |
9494 | } | |
78fa5150 | 9495 | WARN(count != 0, "%lu pages are still in use!\n", count); |
041d3a8c | 9496 | } |
255f5985 | 9497 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 9498 | |
ec6e8c7e VB |
9499 | /* |
9500 | * Effectively disable pcplists for the zone by setting the high limit to 0 | |
9501 | * and draining all cpus. A concurrent page freeing on another CPU that's about | |
9502 | * to put the page on pcplist will either finish before the drain and the page | |
9503 | * will be drained, or observe the new high limit and skip the pcplist. | |
9504 | * | |
9505 | * Must be paired with a call to zone_pcp_enable(). | |
9506 | */ | |
9507 | void zone_pcp_disable(struct zone *zone) | |
9508 | { | |
9509 | mutex_lock(&pcp_batch_high_lock); | |
9510 | __zone_set_pageset_high_and_batch(zone, 0, 1); | |
9511 | __drain_all_pages(zone, true); | |
9512 | } | |
9513 | ||
9514 | void zone_pcp_enable(struct zone *zone) | |
9515 | { | |
9516 | __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch); | |
9517 | mutex_unlock(&pcp_batch_high_lock); | |
9518 | } | |
9519 | ||
340175b7 JL |
9520 | void zone_pcp_reset(struct zone *zone) |
9521 | { | |
5a883813 | 9522 | int cpu; |
28f836b6 | 9523 | struct per_cpu_zonestat *pzstats; |
340175b7 | 9524 | |
28f836b6 | 9525 | if (zone->per_cpu_pageset != &boot_pageset) { |
5a883813 | 9526 | for_each_online_cpu(cpu) { |
28f836b6 MG |
9527 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); |
9528 | drain_zonestat(zone, pzstats); | |
5a883813 | 9529 | } |
28f836b6 | 9530 | free_percpu(zone->per_cpu_pageset); |
28f836b6 | 9531 | zone->per_cpu_pageset = &boot_pageset; |
022e7fa0 ML |
9532 | if (zone->per_cpu_zonestats != &boot_zonestats) { |
9533 | free_percpu(zone->per_cpu_zonestats); | |
9534 | zone->per_cpu_zonestats = &boot_zonestats; | |
9535 | } | |
340175b7 | 9536 | } |
340175b7 JL |
9537 | } |
9538 | ||
6dcd73d7 | 9539 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 9540 | /* |
257bea71 DH |
9541 | * All pages in the range must be in a single zone, must not contain holes, |
9542 | * must span full sections, and must be isolated before calling this function. | |
0c0e6195 | 9543 | */ |
257bea71 | 9544 | void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 9545 | { |
257bea71 | 9546 | unsigned long pfn = start_pfn; |
0c0e6195 KH |
9547 | struct page *page; |
9548 | struct zone *zone; | |
0ee5f4f3 | 9549 | unsigned int order; |
0c0e6195 | 9550 | unsigned long flags; |
5557c766 | 9551 | |
2d070eab | 9552 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
9553 | zone = page_zone(pfn_to_page(pfn)); |
9554 | spin_lock_irqsave(&zone->lock, flags); | |
0c0e6195 | 9555 | while (pfn < end_pfn) { |
0c0e6195 | 9556 | page = pfn_to_page(pfn); |
b023f468 WC |
9557 | /* |
9558 | * The HWPoisoned page may be not in buddy system, and | |
9559 | * page_count() is not 0. | |
9560 | */ | |
9561 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
9562 | pfn++; | |
b023f468 WC |
9563 | continue; |
9564 | } | |
aa218795 DH |
9565 | /* |
9566 | * At this point all remaining PageOffline() pages have a | |
9567 | * reference count of 0 and can simply be skipped. | |
9568 | */ | |
9569 | if (PageOffline(page)) { | |
9570 | BUG_ON(page_count(page)); | |
9571 | BUG_ON(PageBuddy(page)); | |
9572 | pfn++; | |
aa218795 DH |
9573 | continue; |
9574 | } | |
b023f468 | 9575 | |
0c0e6195 KH |
9576 | BUG_ON(page_count(page)); |
9577 | BUG_ON(!PageBuddy(page)); | |
ab130f91 | 9578 | order = buddy_order(page); |
6ab01363 | 9579 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
9580 | pfn += (1 << order); |
9581 | } | |
9582 | spin_unlock_irqrestore(&zone->lock, flags); | |
9583 | } | |
9584 | #endif | |
8d22ba1b | 9585 | |
8446b59b ED |
9586 | /* |
9587 | * This function returns a stable result only if called under zone lock. | |
9588 | */ | |
8d22ba1b WF |
9589 | bool is_free_buddy_page(struct page *page) |
9590 | { | |
8d22ba1b | 9591 | unsigned long pfn = page_to_pfn(page); |
7aeb09f9 | 9592 | unsigned int order; |
8d22ba1b | 9593 | |
8d22ba1b WF |
9594 | for (order = 0; order < MAX_ORDER; order++) { |
9595 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
9596 | ||
8446b59b ED |
9597 | if (PageBuddy(page_head) && |
9598 | buddy_order_unsafe(page_head) >= order) | |
8d22ba1b WF |
9599 | break; |
9600 | } | |
8d22ba1b WF |
9601 | |
9602 | return order < MAX_ORDER; | |
9603 | } | |
a581865e | 9604 | EXPORT_SYMBOL(is_free_buddy_page); |
d4ae9916 NH |
9605 | |
9606 | #ifdef CONFIG_MEMORY_FAILURE | |
9607 | /* | |
06be6ff3 OS |
9608 | * Break down a higher-order page in sub-pages, and keep our target out of |
9609 | * buddy allocator. | |
d4ae9916 | 9610 | */ |
06be6ff3 OS |
9611 | static void break_down_buddy_pages(struct zone *zone, struct page *page, |
9612 | struct page *target, int low, int high, | |
9613 | int migratetype) | |
9614 | { | |
9615 | unsigned long size = 1 << high; | |
9616 | struct page *current_buddy, *next_page; | |
9617 | ||
9618 | while (high > low) { | |
9619 | high--; | |
9620 | size >>= 1; | |
9621 | ||
9622 | if (target >= &page[size]) { | |
9623 | next_page = page + size; | |
9624 | current_buddy = page; | |
9625 | } else { | |
9626 | next_page = page; | |
9627 | current_buddy = page + size; | |
9628 | } | |
9629 | ||
9630 | if (set_page_guard(zone, current_buddy, high, migratetype)) | |
9631 | continue; | |
9632 | ||
9633 | if (current_buddy != target) { | |
9634 | add_to_free_list(current_buddy, zone, high, migratetype); | |
ab130f91 | 9635 | set_buddy_order(current_buddy, high); |
06be6ff3 OS |
9636 | page = next_page; |
9637 | } | |
9638 | } | |
9639 | } | |
9640 | ||
9641 | /* | |
9642 | * Take a page that will be marked as poisoned off the buddy allocator. | |
9643 | */ | |
9644 | bool take_page_off_buddy(struct page *page) | |
d4ae9916 NH |
9645 | { |
9646 | struct zone *zone = page_zone(page); | |
9647 | unsigned long pfn = page_to_pfn(page); | |
9648 | unsigned long flags; | |
9649 | unsigned int order; | |
06be6ff3 | 9650 | bool ret = false; |
d4ae9916 NH |
9651 | |
9652 | spin_lock_irqsave(&zone->lock, flags); | |
9653 | for (order = 0; order < MAX_ORDER; order++) { | |
9654 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
ab130f91 | 9655 | int page_order = buddy_order(page_head); |
d4ae9916 | 9656 | |
ab130f91 | 9657 | if (PageBuddy(page_head) && page_order >= order) { |
06be6ff3 OS |
9658 | unsigned long pfn_head = page_to_pfn(page_head); |
9659 | int migratetype = get_pfnblock_migratetype(page_head, | |
9660 | pfn_head); | |
9661 | ||
ab130f91 | 9662 | del_page_from_free_list(page_head, zone, page_order); |
06be6ff3 | 9663 | break_down_buddy_pages(zone, page_head, page, 0, |
ab130f91 | 9664 | page_order, migratetype); |
bf181c58 | 9665 | SetPageHWPoisonTakenOff(page); |
bac9c6fa DH |
9666 | if (!is_migrate_isolate(migratetype)) |
9667 | __mod_zone_freepage_state(zone, -1, migratetype); | |
06be6ff3 | 9668 | ret = true; |
d4ae9916 NH |
9669 | break; |
9670 | } | |
06be6ff3 OS |
9671 | if (page_count(page_head) > 0) |
9672 | break; | |
d4ae9916 NH |
9673 | } |
9674 | spin_unlock_irqrestore(&zone->lock, flags); | |
06be6ff3 | 9675 | return ret; |
d4ae9916 | 9676 | } |
bf181c58 NH |
9677 | |
9678 | /* | |
9679 | * Cancel takeoff done by take_page_off_buddy(). | |
9680 | */ | |
9681 | bool put_page_back_buddy(struct page *page) | |
9682 | { | |
9683 | struct zone *zone = page_zone(page); | |
9684 | unsigned long pfn = page_to_pfn(page); | |
9685 | unsigned long flags; | |
9686 | int migratetype = get_pfnblock_migratetype(page, pfn); | |
9687 | bool ret = false; | |
9688 | ||
9689 | spin_lock_irqsave(&zone->lock, flags); | |
9690 | if (put_page_testzero(page)) { | |
9691 | ClearPageHWPoisonTakenOff(page); | |
9692 | __free_one_page(page, pfn, zone, 0, migratetype, FPI_NONE); | |
9693 | if (TestClearPageHWPoison(page)) { | |
bf181c58 NH |
9694 | ret = true; |
9695 | } | |
9696 | } | |
9697 | spin_unlock_irqrestore(&zone->lock, flags); | |
9698 | ||
9699 | return ret; | |
9700 | } | |
d4ae9916 | 9701 | #endif |
62b31070 BH |
9702 | |
9703 | #ifdef CONFIG_ZONE_DMA | |
9704 | bool has_managed_dma(void) | |
9705 | { | |
9706 | struct pglist_data *pgdat; | |
9707 | ||
9708 | for_each_online_pgdat(pgdat) { | |
9709 | struct zone *zone = &pgdat->node_zones[ZONE_DMA]; | |
9710 | ||
9711 | if (managed_zone(zone)) | |
9712 | return true; | |
9713 | } | |
9714 | return false; | |
9715 | } | |
9716 | #endif /* CONFIG_ZONE_DMA */ |