]>
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/interrupt.h> |
10ed273f | 22 | #include <linux/jiffies.h> |
1da177e4 | 23 | #include <linux/compiler.h> |
9f158333 | 24 | #include <linux/kernel.h> |
b8c73fc2 | 25 | #include <linux/kasan.h> |
b073d7f8 | 26 | #include <linux/kmsan.h> |
1da177e4 LT |
27 | #include <linux/module.h> |
28 | #include <linux/suspend.h> | |
29 | #include <linux/pagevec.h> | |
a238ab5b | 30 | #include <linux/ratelimit.h> |
5a3135c2 | 31 | #include <linux/oom.h> |
1da177e4 LT |
32 | #include <linux/topology.h> |
33 | #include <linux/sysctl.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/cpuset.h> | |
bdc8cb98 | 36 | #include <linux/memory_hotplug.h> |
1da177e4 | 37 | #include <linux/nodemask.h> |
a6cccdc3 | 38 | #include <linux/vmstat.h> |
c713216d MG |
39 | #include <linux/sort.h> |
40 | #include <linux/pfn.h> | |
933e312e | 41 | #include <linux/fault-inject.h> |
56de7263 | 42 | #include <linux/compaction.h> |
0d3d062a | 43 | #include <trace/events/kmem.h> |
d379f01d | 44 | #include <trace/events/oom.h> |
268bb0ce | 45 | #include <linux/prefetch.h> |
6e543d57 | 46 | #include <linux/mm_inline.h> |
f920e413 | 47 | #include <linux/mmu_notifier.h> |
041d3a8c | 48 | #include <linux/migrate.h> |
5b3cc15a | 49 | #include <linux/sched/mm.h> |
48c96a36 | 50 | #include <linux/page_owner.h> |
df4e817b | 51 | #include <linux/page_table_check.h> |
4949148a | 52 | #include <linux/memcontrol.h> |
42c269c8 | 53 | #include <linux/ftrace.h> |
d92a8cfc | 54 | #include <linux/lockdep.h> |
556b969a | 55 | #include <linux/nmi.h> |
eb414681 | 56 | #include <linux/psi.h> |
4aab2be0 | 57 | #include <linux/khugepaged.h> |
5bf18281 | 58 | #include <linux/delayacct.h> |
ac924c60 | 59 | #include <asm/div64.h> |
1da177e4 | 60 | #include "internal.h" |
e900a918 | 61 | #include "shuffle.h" |
36e66c55 | 62 | #include "page_reporting.h" |
1da177e4 | 63 | |
f04a5d5d DH |
64 | /* Free Page Internal flags: for internal, non-pcp variants of free_pages(). */ |
65 | typedef int __bitwise fpi_t; | |
66 | ||
67 | /* No special request */ | |
68 | #define FPI_NONE ((__force fpi_t)0) | |
69 | ||
70 | /* | |
71 | * Skip free page reporting notification for the (possibly merged) page. | |
72 | * This does not hinder free page reporting from grabbing the page, | |
73 | * reporting it and marking it "reported" - it only skips notifying | |
74 | * the free page reporting infrastructure about a newly freed page. For | |
75 | * example, used when temporarily pulling a page from a freelist and | |
76 | * putting it back unmodified. | |
77 | */ | |
78 | #define FPI_SKIP_REPORT_NOTIFY ((__force fpi_t)BIT(0)) | |
79 | ||
47b6a24a DH |
80 | /* |
81 | * Place the (possibly merged) page to the tail of the freelist. Will ignore | |
82 | * page shuffling (relevant code - e.g., memory onlining - is expected to | |
83 | * shuffle the whole zone). | |
84 | * | |
85 | * Note: No code should rely on this flag for correctness - it's purely | |
86 | * to allow for optimizations when handing back either fresh pages | |
87 | * (memory onlining) or untouched pages (page isolation, free page | |
88 | * reporting). | |
89 | */ | |
90 | #define FPI_TO_TAIL ((__force fpi_t)BIT(1)) | |
91 | ||
c8e251fa CS |
92 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
93 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
74f44822 | 94 | #define MIN_PERCPU_PAGELIST_HIGH_FRACTION (8) |
c8e251fa | 95 | |
4b23a68f MG |
96 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT) |
97 | /* | |
98 | * On SMP, spin_trylock is sufficient protection. | |
99 | * On PREEMPT_RT, spin_trylock is equivalent on both SMP and UP. | |
100 | */ | |
101 | #define pcp_trylock_prepare(flags) do { } while (0) | |
102 | #define pcp_trylock_finish(flag) do { } while (0) | |
103 | #else | |
104 | ||
105 | /* UP spin_trylock always succeeds so disable IRQs to prevent re-entrancy. */ | |
106 | #define pcp_trylock_prepare(flags) local_irq_save(flags) | |
107 | #define pcp_trylock_finish(flags) local_irq_restore(flags) | |
108 | #endif | |
109 | ||
01b44456 MG |
110 | /* |
111 | * Locking a pcp requires a PCP lookup followed by a spinlock. To avoid | |
112 | * a migration causing the wrong PCP to be locked and remote memory being | |
113 | * potentially allocated, pin the task to the CPU for the lookup+lock. | |
114 | * preempt_disable is used on !RT because it is faster than migrate_disable. | |
115 | * migrate_disable is used on RT because otherwise RT spinlock usage is | |
116 | * interfered with and a high priority task cannot preempt the allocator. | |
117 | */ | |
118 | #ifndef CONFIG_PREEMPT_RT | |
119 | #define pcpu_task_pin() preempt_disable() | |
120 | #define pcpu_task_unpin() preempt_enable() | |
121 | #else | |
122 | #define pcpu_task_pin() migrate_disable() | |
123 | #define pcpu_task_unpin() migrate_enable() | |
124 | #endif | |
c8e251fa | 125 | |
01b44456 MG |
126 | /* |
127 | * Generic helper to lookup and a per-cpu variable with an embedded spinlock. | |
128 | * Return value should be used with equivalent unlock helper. | |
129 | */ | |
130 | #define pcpu_spin_lock(type, member, ptr) \ | |
131 | ({ \ | |
132 | type *_ret; \ | |
133 | pcpu_task_pin(); \ | |
134 | _ret = this_cpu_ptr(ptr); \ | |
135 | spin_lock(&_ret->member); \ | |
136 | _ret; \ | |
137 | }) | |
138 | ||
57490774 | 139 | #define pcpu_spin_trylock(type, member, ptr) \ |
01b44456 MG |
140 | ({ \ |
141 | type *_ret; \ | |
142 | pcpu_task_pin(); \ | |
143 | _ret = this_cpu_ptr(ptr); \ | |
57490774 | 144 | if (!spin_trylock(&_ret->member)) { \ |
01b44456 MG |
145 | pcpu_task_unpin(); \ |
146 | _ret = NULL; \ | |
147 | } \ | |
148 | _ret; \ | |
149 | }) | |
150 | ||
151 | #define pcpu_spin_unlock(member, ptr) \ | |
152 | ({ \ | |
153 | spin_unlock(&ptr->member); \ | |
154 | pcpu_task_unpin(); \ | |
155 | }) | |
156 | ||
01b44456 MG |
157 | /* struct per_cpu_pages specific helpers. */ |
158 | #define pcp_spin_lock(ptr) \ | |
159 | pcpu_spin_lock(struct per_cpu_pages, lock, ptr) | |
160 | ||
57490774 MG |
161 | #define pcp_spin_trylock(ptr) \ |
162 | pcpu_spin_trylock(struct per_cpu_pages, lock, ptr) | |
01b44456 MG |
163 | |
164 | #define pcp_spin_unlock(ptr) \ | |
165 | pcpu_spin_unlock(lock, ptr) | |
166 | ||
72812019 LS |
167 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
168 | DEFINE_PER_CPU(int, numa_node); | |
169 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
170 | #endif | |
171 | ||
4518085e KW |
172 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
173 | ||
7aac7898 LS |
174 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
175 | /* | |
176 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
177 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
178 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
179 | * defined in <linux/topology.h>. | |
180 | */ | |
181 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
182 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
183 | #endif | |
184 | ||
8b885f53 | 185 | static DEFINE_MUTEX(pcpu_drain_mutex); |
bd233f53 | 186 | |
38addce8 | 187 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 188 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
189 | EXPORT_SYMBOL(latent_entropy); |
190 | #endif | |
191 | ||
1da177e4 | 192 | /* |
13808910 | 193 | * Array of node states. |
1da177e4 | 194 | */ |
13808910 CL |
195 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
196 | [N_POSSIBLE] = NODE_MASK_ALL, | |
197 | [N_ONLINE] = { { [0] = 1UL } }, | |
198 | #ifndef CONFIG_NUMA | |
199 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
200 | #ifdef CONFIG_HIGHMEM | |
201 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 202 | #endif |
20b2f52b | 203 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
204 | [N_CPU] = { { [0] = 1UL } }, |
205 | #endif /* NUMA */ | |
206 | }; | |
207 | EXPORT_SYMBOL(node_states); | |
208 | ||
74f44822 | 209 | int percpu_pagelist_high_fraction; |
dcce284a | 210 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
6471384a | 211 | |
bb14c2c7 VB |
212 | /* |
213 | * A cached value of the page's pageblock's migratetype, used when the page is | |
214 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
215 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
216 | * Also the migratetype set in the page does not necessarily match the pcplist | |
217 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
218 | * other index - this ensures that it will be put on the correct CMA freelist. | |
219 | */ | |
220 | static inline int get_pcppage_migratetype(struct page *page) | |
221 | { | |
222 | return page->index; | |
223 | } | |
224 | ||
225 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
226 | { | |
227 | page->index = migratetype; | |
228 | } | |
229 | ||
452aa699 RW |
230 | #ifdef CONFIG_PM_SLEEP |
231 | /* | |
232 | * The following functions are used by the suspend/hibernate code to temporarily | |
233 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
234 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
235 | * they should always be called with system_transition_mutex held |
236 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
237 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
238 | * with that modification). | |
452aa699 | 239 | */ |
c9e664f1 RW |
240 | |
241 | static gfp_t saved_gfp_mask; | |
242 | ||
243 | void pm_restore_gfp_mask(void) | |
452aa699 | 244 | { |
55f2503c | 245 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
246 | if (saved_gfp_mask) { |
247 | gfp_allowed_mask = saved_gfp_mask; | |
248 | saved_gfp_mask = 0; | |
249 | } | |
452aa699 RW |
250 | } |
251 | ||
c9e664f1 | 252 | void pm_restrict_gfp_mask(void) |
452aa699 | 253 | { |
55f2503c | 254 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
255 | WARN_ON(saved_gfp_mask); |
256 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 257 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 258 | } |
f90ac398 MG |
259 | |
260 | bool pm_suspended_storage(void) | |
261 | { | |
d0164adc | 262 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
263 | return false; |
264 | return true; | |
265 | } | |
452aa699 RW |
266 | #endif /* CONFIG_PM_SLEEP */ |
267 | ||
d9c23400 | 268 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 269 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
270 | #endif |
271 | ||
7fef431b DH |
272 | static void __free_pages_ok(struct page *page, unsigned int order, |
273 | fpi_t fpi_flags); | |
a226f6c8 | 274 | |
1da177e4 LT |
275 | /* |
276 | * results with 256, 32 in the lowmem_reserve sysctl: | |
277 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
278 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
279 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
280 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 281 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
282 | * |
283 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
284 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 285 | */ |
d3cda233 | 286 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 287 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 288 | [ZONE_DMA] = 256, |
4b51d669 | 289 | #endif |
fb0e7942 | 290 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 291 | [ZONE_DMA32] = 256, |
fb0e7942 | 292 | #endif |
d3cda233 | 293 | [ZONE_NORMAL] = 32, |
e53ef38d | 294 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 295 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 296 | #endif |
d3cda233 | 297 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 298 | }; |
1da177e4 | 299 | |
9420f89d | 300 | char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 301 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 302 | "DMA", |
4b51d669 | 303 | #endif |
fb0e7942 | 304 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 305 | "DMA32", |
fb0e7942 | 306 | #endif |
2f1b6248 | 307 | "Normal", |
e53ef38d | 308 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 309 | "HighMem", |
e53ef38d | 310 | #endif |
2a1e274a | 311 | "Movable", |
033fbae9 DW |
312 | #ifdef CONFIG_ZONE_DEVICE |
313 | "Device", | |
314 | #endif | |
2f1b6248 CL |
315 | }; |
316 | ||
c999fbd3 | 317 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
318 | "Unmovable", |
319 | "Movable", | |
320 | "Reclaimable", | |
321 | "HighAtomic", | |
322 | #ifdef CONFIG_CMA | |
323 | "CMA", | |
324 | #endif | |
325 | #ifdef CONFIG_MEMORY_ISOLATION | |
326 | "Isolate", | |
327 | #endif | |
328 | }; | |
329 | ||
ae70eddd AK |
330 | compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = { |
331 | [NULL_COMPOUND_DTOR] = NULL, | |
332 | [COMPOUND_PAGE_DTOR] = free_compound_page, | |
f1e61557 | 333 | #ifdef CONFIG_HUGETLB_PAGE |
ae70eddd | 334 | [HUGETLB_PAGE_DTOR] = free_huge_page, |
f1e61557 | 335 | #endif |
9a982250 | 336 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
ae70eddd | 337 | [TRANSHUGE_PAGE_DTOR] = free_transhuge_page, |
9a982250 | 338 | #endif |
f1e61557 KS |
339 | }; |
340 | ||
1da177e4 | 341 | int min_free_kbytes = 1024; |
42aa83cb | 342 | int user_min_free_kbytes = -1; |
1c30844d | 343 | int watermark_boost_factor __read_mostly = 15000; |
795ae7a0 | 344 | int watermark_scale_factor = 10; |
1da177e4 | 345 | |
0ee332c1 TH |
346 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ |
347 | int movable_zone; | |
348 | EXPORT_SYMBOL(movable_zone); | |
c713216d | 349 | |
418508c1 | 350 | #if MAX_NUMNODES > 1 |
b9726c26 | 351 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 352 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 353 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 354 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
355 | #endif |
356 | ||
9ef9acb0 MG |
357 | int page_group_by_mobility_disabled __read_mostly; |
358 | ||
3a80a7fa | 359 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
360 | /* |
361 | * During boot we initialize deferred pages on-demand, as needed, but once | |
362 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
363 | * and we can permanently disable that path. | |
364 | */ | |
9420f89d | 365 | DEFINE_STATIC_KEY_TRUE(deferred_pages); |
3c0c12cc | 366 | |
94ae8b83 | 367 | static inline bool deferred_pages_enabled(void) |
3c0c12cc | 368 | { |
94ae8b83 | 369 | return static_branch_unlikely(&deferred_pages); |
3c0c12cc WL |
370 | } |
371 | ||
3a80a7fa | 372 | /* |
9420f89d MRI |
373 | * deferred_grow_zone() is __init, but it is called from |
374 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
375 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
376 | * and to ensure that the function body gets unloaded. | |
3a80a7fa | 377 | */ |
9420f89d MRI |
378 | static bool __ref |
379 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
3a80a7fa | 380 | { |
9420f89d | 381 | return deferred_grow_zone(zone, order); |
3a80a7fa MG |
382 | } |
383 | #else | |
94ae8b83 | 384 | static inline bool deferred_pages_enabled(void) |
2c335680 | 385 | { |
94ae8b83 | 386 | return false; |
2c335680 | 387 | } |
9420f89d | 388 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
3a80a7fa | 389 | |
0b423ca2 | 390 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
ca891f41 | 391 | static inline unsigned long *get_pageblock_bitmap(const struct page *page, |
0b423ca2 MG |
392 | unsigned long pfn) |
393 | { | |
394 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 395 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
396 | #else |
397 | return page_zone(page)->pageblock_flags; | |
398 | #endif /* CONFIG_SPARSEMEM */ | |
399 | } | |
400 | ||
ca891f41 | 401 | static inline int pfn_to_bitidx(const struct page *page, unsigned long pfn) |
0b423ca2 MG |
402 | { |
403 | #ifdef CONFIG_SPARSEMEM | |
404 | pfn &= (PAGES_PER_SECTION-1); | |
0b423ca2 | 405 | #else |
4f9bc69a | 406 | pfn = pfn - pageblock_start_pfn(page_zone(page)->zone_start_pfn); |
0b423ca2 | 407 | #endif /* CONFIG_SPARSEMEM */ |
399b795b | 408 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
0b423ca2 MG |
409 | } |
410 | ||
535b81e2 | 411 | static __always_inline |
ca891f41 | 412 | unsigned long __get_pfnblock_flags_mask(const struct page *page, |
0b423ca2 | 413 | unsigned long pfn, |
0b423ca2 MG |
414 | unsigned long mask) |
415 | { | |
416 | unsigned long *bitmap; | |
417 | unsigned long bitidx, word_bitidx; | |
418 | unsigned long word; | |
419 | ||
420 | bitmap = get_pageblock_bitmap(page, pfn); | |
421 | bitidx = pfn_to_bitidx(page, pfn); | |
422 | word_bitidx = bitidx / BITS_PER_LONG; | |
423 | bitidx &= (BITS_PER_LONG-1); | |
1c563432 MK |
424 | /* |
425 | * This races, without locks, with set_pfnblock_flags_mask(). Ensure | |
426 | * a consistent read of the memory array, so that results, even though | |
427 | * racy, are not corrupted. | |
428 | */ | |
429 | word = READ_ONCE(bitmap[word_bitidx]); | |
d93d5ab9 | 430 | return (word >> bitidx) & mask; |
0b423ca2 MG |
431 | } |
432 | ||
a00cda3f MCC |
433 | /** |
434 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
435 | * @page: The page within the block of interest | |
436 | * @pfn: The target page frame number | |
437 | * @mask: mask of bits that the caller is interested in | |
438 | * | |
439 | * Return: pageblock_bits flags | |
440 | */ | |
ca891f41 MWO |
441 | unsigned long get_pfnblock_flags_mask(const struct page *page, |
442 | unsigned long pfn, unsigned long mask) | |
0b423ca2 | 443 | { |
535b81e2 | 444 | return __get_pfnblock_flags_mask(page, pfn, mask); |
0b423ca2 MG |
445 | } |
446 | ||
ca891f41 MWO |
447 | static __always_inline int get_pfnblock_migratetype(const struct page *page, |
448 | unsigned long pfn) | |
0b423ca2 | 449 | { |
535b81e2 | 450 | return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK); |
0b423ca2 MG |
451 | } |
452 | ||
453 | /** | |
454 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
455 | * @page: The page within the block of interest | |
456 | * @flags: The flags to set | |
457 | * @pfn: The target page frame number | |
0b423ca2 MG |
458 | * @mask: mask of bits that the caller is interested in |
459 | */ | |
460 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
461 | unsigned long pfn, | |
0b423ca2 MG |
462 | unsigned long mask) |
463 | { | |
464 | unsigned long *bitmap; | |
465 | unsigned long bitidx, word_bitidx; | |
04ec0061 | 466 | unsigned long word; |
0b423ca2 MG |
467 | |
468 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 469 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
470 | |
471 | bitmap = get_pageblock_bitmap(page, pfn); | |
472 | bitidx = pfn_to_bitidx(page, pfn); | |
473 | word_bitidx = bitidx / BITS_PER_LONG; | |
474 | bitidx &= (BITS_PER_LONG-1); | |
475 | ||
476 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
477 | ||
d93d5ab9 WY |
478 | mask <<= bitidx; |
479 | flags <<= bitidx; | |
0b423ca2 MG |
480 | |
481 | word = READ_ONCE(bitmap[word_bitidx]); | |
04ec0061 UB |
482 | do { |
483 | } while (!try_cmpxchg(&bitmap[word_bitidx], &word, (word & ~mask) | flags)); | |
0b423ca2 | 484 | } |
3a80a7fa | 485 | |
ee6f509c | 486 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 487 | { |
5d0f3f72 KM |
488 | if (unlikely(page_group_by_mobility_disabled && |
489 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
490 | migratetype = MIGRATE_UNMOVABLE; |
491 | ||
d93d5ab9 | 492 | set_pfnblock_flags_mask(page, (unsigned long)migratetype, |
535b81e2 | 493 | page_to_pfn(page), MIGRATETYPE_MASK); |
b2a0ac88 MG |
494 | } |
495 | ||
13e7444b | 496 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 497 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 498 | { |
bdc8cb98 DH |
499 | int ret = 0; |
500 | unsigned seq; | |
501 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 502 | unsigned long sp, start_pfn; |
c6a57e19 | 503 | |
bdc8cb98 DH |
504 | do { |
505 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
506 | start_pfn = zone->zone_start_pfn; |
507 | sp = zone->spanned_pages; | |
108bcc96 | 508 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
509 | ret = 1; |
510 | } while (zone_span_seqretry(zone, seq)); | |
511 | ||
b5e6a5a2 | 512 | if (ret) |
613813e8 DH |
513 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
514 | pfn, zone_to_nid(zone), zone->name, | |
515 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 516 | |
bdc8cb98 | 517 | return ret; |
c6a57e19 DH |
518 | } |
519 | ||
520 | static int page_is_consistent(struct zone *zone, struct page *page) | |
521 | { | |
1da177e4 | 522 | if (zone != page_zone(page)) |
c6a57e19 DH |
523 | return 0; |
524 | ||
525 | return 1; | |
526 | } | |
527 | /* | |
528 | * Temporary debugging check for pages not lying within a given zone. | |
529 | */ | |
d73d3c9f | 530 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
531 | { |
532 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 533 | return 1; |
c6a57e19 DH |
534 | if (!page_is_consistent(zone, page)) |
535 | return 1; | |
536 | ||
1da177e4 LT |
537 | return 0; |
538 | } | |
13e7444b | 539 | #else |
d73d3c9f | 540 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
541 | { |
542 | return 0; | |
543 | } | |
544 | #endif | |
545 | ||
82a3241a | 546 | static void bad_page(struct page *page, const char *reason) |
1da177e4 | 547 | { |
d936cf9b HD |
548 | static unsigned long resume; |
549 | static unsigned long nr_shown; | |
550 | static unsigned long nr_unshown; | |
551 | ||
552 | /* | |
553 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
554 | * or allow a steady drip of one report per second. | |
555 | */ | |
556 | if (nr_shown == 60) { | |
557 | if (time_before(jiffies, resume)) { | |
558 | nr_unshown++; | |
559 | goto out; | |
560 | } | |
561 | if (nr_unshown) { | |
ff8e8116 | 562 | pr_alert( |
1e9e6365 | 563 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
564 | nr_unshown); |
565 | nr_unshown = 0; | |
566 | } | |
567 | nr_shown = 0; | |
568 | } | |
569 | if (nr_shown++ == 0) | |
570 | resume = jiffies + 60 * HZ; | |
571 | ||
ff8e8116 | 572 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 573 | current->comm, page_to_pfn(page)); |
d2f07ec0 | 574 | dump_page(page, reason); |
3dc14741 | 575 | |
4f31888c | 576 | print_modules(); |
1da177e4 | 577 | dump_stack(); |
d936cf9b | 578 | out: |
8cc3b392 | 579 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 580 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 581 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
582 | } |
583 | ||
44042b44 MG |
584 | static inline unsigned int order_to_pindex(int migratetype, int order) |
585 | { | |
586 | int base = order; | |
587 | ||
588 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
589 | if (order > PAGE_ALLOC_COSTLY_ORDER) { | |
590 | VM_BUG_ON(order != pageblock_order); | |
5d0a661d | 591 | return NR_LOWORDER_PCP_LISTS; |
44042b44 MG |
592 | } |
593 | #else | |
594 | VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER); | |
595 | #endif | |
596 | ||
597 | return (MIGRATE_PCPTYPES * base) + migratetype; | |
598 | } | |
599 | ||
600 | static inline int pindex_to_order(unsigned int pindex) | |
601 | { | |
602 | int order = pindex / MIGRATE_PCPTYPES; | |
603 | ||
604 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5d0a661d | 605 | if (pindex == NR_LOWORDER_PCP_LISTS) |
44042b44 | 606 | order = pageblock_order; |
44042b44 MG |
607 | #else |
608 | VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER); | |
609 | #endif | |
610 | ||
611 | return order; | |
612 | } | |
613 | ||
614 | static inline bool pcp_allowed_order(unsigned int order) | |
615 | { | |
616 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
617 | return true; | |
618 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
619 | if (order == pageblock_order) | |
620 | return true; | |
621 | #endif | |
622 | return false; | |
623 | } | |
624 | ||
21d02f8f MG |
625 | static inline void free_the_page(struct page *page, unsigned int order) |
626 | { | |
44042b44 MG |
627 | if (pcp_allowed_order(order)) /* Via pcp? */ |
628 | free_unref_page(page, order); | |
21d02f8f MG |
629 | else |
630 | __free_pages_ok(page, order, FPI_NONE); | |
631 | } | |
632 | ||
1da177e4 LT |
633 | /* |
634 | * Higher-order pages are called "compound pages". They are structured thusly: | |
635 | * | |
1d798ca3 | 636 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 637 | * |
1d798ca3 KS |
638 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
639 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 640 | * |
1d798ca3 KS |
641 | * The first tail page's ->compound_dtor holds the offset in array of compound |
642 | * page destructors. See compound_page_dtors. | |
1da177e4 | 643 | * |
1d798ca3 | 644 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 645 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 646 | */ |
d98c7a09 | 647 | |
9a982250 | 648 | void free_compound_page(struct page *page) |
d98c7a09 | 649 | { |
bbc6b703 | 650 | mem_cgroup_uncharge(page_folio(page)); |
44042b44 | 651 | free_the_page(page, compound_order(page)); |
d98c7a09 HD |
652 | } |
653 | ||
d00181b9 | 654 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
655 | { |
656 | int i; | |
657 | int nr_pages = 1 << order; | |
658 | ||
18229df5 | 659 | __SetPageHead(page); |
5b24eeef JM |
660 | for (i = 1; i < nr_pages; i++) |
661 | prep_compound_tail(page, i); | |
1378a5ee | 662 | |
5b24eeef | 663 | prep_compound_head(page, order); |
18229df5 AW |
664 | } |
665 | ||
5375336c MWO |
666 | void destroy_large_folio(struct folio *folio) |
667 | { | |
a60d5942 | 668 | enum compound_dtor_id dtor = folio->_folio_dtor; |
5375336c MWO |
669 | |
670 | VM_BUG_ON_FOLIO(dtor >= NR_COMPOUND_DTORS, folio); | |
671 | compound_page_dtors[dtor](&folio->page); | |
672 | } | |
673 | ||
c0a32fc5 SG |
674 | #ifdef CONFIG_DEBUG_PAGEALLOC |
675 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 676 | |
8e57f8ac VB |
677 | bool _debug_pagealloc_enabled_early __read_mostly |
678 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
679 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 680 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 681 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
682 | |
683 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 684 | |
031bc574 JK |
685 | static int __init early_debug_pagealloc(char *buf) |
686 | { | |
8e57f8ac | 687 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
688 | } |
689 | early_param("debug_pagealloc", early_debug_pagealloc); | |
690 | ||
c0a32fc5 SG |
691 | static int __init debug_guardpage_minorder_setup(char *buf) |
692 | { | |
693 | unsigned long res; | |
694 | ||
695 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 696 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
697 | return 0; |
698 | } | |
699 | _debug_guardpage_minorder = res; | |
1170532b | 700 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
701 | return 0; |
702 | } | |
f1c1e9f7 | 703 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 704 | |
acbc15a4 | 705 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 706 | unsigned int order, int migratetype) |
c0a32fc5 | 707 | { |
e30825f1 | 708 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
709 | return false; |
710 | ||
711 | if (order >= debug_guardpage_minorder()) | |
712 | return false; | |
e30825f1 | 713 | |
3972f6bb | 714 | __SetPageGuard(page); |
bf75f200 | 715 | INIT_LIST_HEAD(&page->buddy_list); |
2847cf95 JK |
716 | set_page_private(page, order); |
717 | /* Guard pages are not available for any usage */ | |
b3618455 ML |
718 | if (!is_migrate_isolate(migratetype)) |
719 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
720 | |
721 | return true; | |
c0a32fc5 SG |
722 | } |
723 | ||
2847cf95 JK |
724 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
725 | unsigned int order, int migratetype) | |
c0a32fc5 | 726 | { |
e30825f1 JK |
727 | if (!debug_guardpage_enabled()) |
728 | return; | |
729 | ||
3972f6bb | 730 | __ClearPageGuard(page); |
e30825f1 | 731 | |
2847cf95 JK |
732 | set_page_private(page, 0); |
733 | if (!is_migrate_isolate(migratetype)) | |
734 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
735 | } |
736 | #else | |
acbc15a4 JK |
737 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
738 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
739 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
740 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
741 | #endif |
742 | ||
ab130f91 | 743 | static inline void set_buddy_order(struct page *page, unsigned int order) |
6aa3001b | 744 | { |
4c21e2f2 | 745 | set_page_private(page, order); |
676165a8 | 746 | __SetPageBuddy(page); |
1da177e4 LT |
747 | } |
748 | ||
5e1f0f09 MG |
749 | #ifdef CONFIG_COMPACTION |
750 | static inline struct capture_control *task_capc(struct zone *zone) | |
751 | { | |
752 | struct capture_control *capc = current->capture_control; | |
753 | ||
deba0487 | 754 | return unlikely(capc) && |
5e1f0f09 MG |
755 | !(current->flags & PF_KTHREAD) && |
756 | !capc->page && | |
deba0487 | 757 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
758 | } |
759 | ||
760 | static inline bool | |
761 | compaction_capture(struct capture_control *capc, struct page *page, | |
762 | int order, int migratetype) | |
763 | { | |
764 | if (!capc || order != capc->cc->order) | |
765 | return false; | |
766 | ||
767 | /* Do not accidentally pollute CMA or isolated regions*/ | |
768 | if (is_migrate_cma(migratetype) || | |
769 | is_migrate_isolate(migratetype)) | |
770 | return false; | |
771 | ||
772 | /* | |
f0953a1b | 773 | * Do not let lower order allocations pollute a movable pageblock. |
5e1f0f09 MG |
774 | * This might let an unmovable request use a reclaimable pageblock |
775 | * and vice-versa but no more than normal fallback logic which can | |
776 | * have trouble finding a high-order free page. | |
777 | */ | |
778 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
779 | return false; | |
780 | ||
781 | capc->page = page; | |
782 | return true; | |
783 | } | |
784 | ||
785 | #else | |
786 | static inline struct capture_control *task_capc(struct zone *zone) | |
787 | { | |
788 | return NULL; | |
789 | } | |
790 | ||
791 | static inline bool | |
792 | compaction_capture(struct capture_control *capc, struct page *page, | |
793 | int order, int migratetype) | |
794 | { | |
795 | return false; | |
796 | } | |
797 | #endif /* CONFIG_COMPACTION */ | |
798 | ||
6ab01363 AD |
799 | /* Used for pages not on another list */ |
800 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
801 | unsigned int order, int migratetype) | |
802 | { | |
803 | struct free_area *area = &zone->free_area[order]; | |
804 | ||
bf75f200 | 805 | list_add(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
806 | area->nr_free++; |
807 | } | |
808 | ||
809 | /* Used for pages not on another list */ | |
810 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
811 | unsigned int order, int migratetype) | |
812 | { | |
813 | struct free_area *area = &zone->free_area[order]; | |
814 | ||
bf75f200 | 815 | list_add_tail(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
816 | area->nr_free++; |
817 | } | |
818 | ||
293ffa5e DH |
819 | /* |
820 | * Used for pages which are on another list. Move the pages to the tail | |
821 | * of the list - so the moved pages won't immediately be considered for | |
822 | * allocation again (e.g., optimization for memory onlining). | |
823 | */ | |
6ab01363 AD |
824 | static inline void move_to_free_list(struct page *page, struct zone *zone, |
825 | unsigned int order, int migratetype) | |
826 | { | |
827 | struct free_area *area = &zone->free_area[order]; | |
828 | ||
bf75f200 | 829 | list_move_tail(&page->buddy_list, &area->free_list[migratetype]); |
6ab01363 AD |
830 | } |
831 | ||
832 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
833 | unsigned int order) | |
834 | { | |
36e66c55 AD |
835 | /* clear reported state and update reported page count */ |
836 | if (page_reported(page)) | |
837 | __ClearPageReported(page); | |
838 | ||
bf75f200 | 839 | list_del(&page->buddy_list); |
6ab01363 AD |
840 | __ClearPageBuddy(page); |
841 | set_page_private(page, 0); | |
842 | zone->free_area[order].nr_free--; | |
843 | } | |
844 | ||
5d671eb4 MRI |
845 | static inline struct page *get_page_from_free_area(struct free_area *area, |
846 | int migratetype) | |
847 | { | |
848 | return list_first_entry_or_null(&area->free_list[migratetype], | |
849 | struct page, lru); | |
850 | } | |
851 | ||
a2129f24 AD |
852 | /* |
853 | * If this is not the largest possible page, check if the buddy | |
854 | * of the next-highest order is free. If it is, it's possible | |
855 | * that pages are being freed that will coalesce soon. In case, | |
856 | * that is happening, add the free page to the tail of the list | |
857 | * so it's less likely to be used soon and more likely to be merged | |
858 | * as a higher order page | |
859 | */ | |
860 | static inline bool | |
861 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
862 | struct page *page, unsigned int order) | |
863 | { | |
8170ac47 ZY |
864 | unsigned long higher_page_pfn; |
865 | struct page *higher_page; | |
a2129f24 | 866 | |
23baf831 | 867 | if (order >= MAX_ORDER - 1) |
a2129f24 AD |
868 | return false; |
869 | ||
8170ac47 ZY |
870 | higher_page_pfn = buddy_pfn & pfn; |
871 | higher_page = page + (higher_page_pfn - pfn); | |
a2129f24 | 872 | |
8170ac47 ZY |
873 | return find_buddy_page_pfn(higher_page, higher_page_pfn, order + 1, |
874 | NULL) != NULL; | |
a2129f24 AD |
875 | } |
876 | ||
1da177e4 LT |
877 | /* |
878 | * Freeing function for a buddy system allocator. | |
879 | * | |
880 | * The concept of a buddy system is to maintain direct-mapped table | |
881 | * (containing bit values) for memory blocks of various "orders". | |
882 | * The bottom level table contains the map for the smallest allocatable | |
883 | * units of memory (here, pages), and each level above it describes | |
884 | * pairs of units from the levels below, hence, "buddies". | |
885 | * At a high level, all that happens here is marking the table entry | |
886 | * at the bottom level available, and propagating the changes upward | |
887 | * as necessary, plus some accounting needed to play nicely with other | |
888 | * parts of the VM system. | |
889 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
890 | * free pages of length of (1 << order) and marked with PageBuddy. |
891 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 892 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
893 | * other. That is, if we allocate a small block, and both were |
894 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 895 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 896 | * triggers coalescing into a block of larger size. |
1da177e4 | 897 | * |
6d49e352 | 898 | * -- nyc |
1da177e4 LT |
899 | */ |
900 | ||
48db57f8 | 901 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 902 | unsigned long pfn, |
ed0ae21d | 903 | struct zone *zone, unsigned int order, |
f04a5d5d | 904 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 905 | { |
a2129f24 | 906 | struct capture_control *capc = task_capc(zone); |
dae37a5d | 907 | unsigned long buddy_pfn = 0; |
a2129f24 | 908 | unsigned long combined_pfn; |
a2129f24 AD |
909 | struct page *buddy; |
910 | bool to_tail; | |
d9dddbf5 | 911 | |
d29bb978 | 912 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 913 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 914 | |
ed0ae21d | 915 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 916 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 917 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 918 | |
76741e77 | 919 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 920 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 921 | |
23baf831 | 922 | while (order < MAX_ORDER) { |
5e1f0f09 MG |
923 | if (compaction_capture(capc, page, order, migratetype)) { |
924 | __mod_zone_freepage_state(zone, -(1 << order), | |
925 | migratetype); | |
926 | return; | |
927 | } | |
13ad59df | 928 | |
8170ac47 ZY |
929 | buddy = find_buddy_page_pfn(page, pfn, order, &buddy_pfn); |
930 | if (!buddy) | |
d9dddbf5 | 931 | goto done_merging; |
bb0e28eb ZY |
932 | |
933 | if (unlikely(order >= pageblock_order)) { | |
934 | /* | |
935 | * We want to prevent merge between freepages on pageblock | |
936 | * without fallbacks and normal pageblock. Without this, | |
937 | * pageblock isolation could cause incorrect freepage or CMA | |
938 | * accounting or HIGHATOMIC accounting. | |
939 | */ | |
940 | int buddy_mt = get_pageblock_migratetype(buddy); | |
941 | ||
942 | if (migratetype != buddy_mt | |
943 | && (!migratetype_is_mergeable(migratetype) || | |
944 | !migratetype_is_mergeable(buddy_mt))) | |
945 | goto done_merging; | |
946 | } | |
947 | ||
c0a32fc5 SG |
948 | /* |
949 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
950 | * merge with it and move up one order. | |
951 | */ | |
b03641af | 952 | if (page_is_guard(buddy)) |
2847cf95 | 953 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 954 | else |
6ab01363 | 955 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
956 | combined_pfn = buddy_pfn & pfn; |
957 | page = page + (combined_pfn - pfn); | |
958 | pfn = combined_pfn; | |
1da177e4 LT |
959 | order++; |
960 | } | |
d9dddbf5 VB |
961 | |
962 | done_merging: | |
ab130f91 | 963 | set_buddy_order(page, order); |
6dda9d55 | 964 | |
47b6a24a DH |
965 | if (fpi_flags & FPI_TO_TAIL) |
966 | to_tail = true; | |
967 | else if (is_shuffle_order(order)) | |
a2129f24 | 968 | to_tail = shuffle_pick_tail(); |
97500a4a | 969 | else |
a2129f24 | 970 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 971 | |
a2129f24 | 972 | if (to_tail) |
6ab01363 | 973 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 974 | else |
6ab01363 | 975 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
976 | |
977 | /* Notify page reporting subsystem of freed page */ | |
f04a5d5d | 978 | if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY)) |
36e66c55 | 979 | page_reporting_notify_free(order); |
1da177e4 LT |
980 | } |
981 | ||
b2c9e2fb ZY |
982 | /** |
983 | * split_free_page() -- split a free page at split_pfn_offset | |
984 | * @free_page: the original free page | |
985 | * @order: the order of the page | |
986 | * @split_pfn_offset: split offset within the page | |
987 | * | |
86d28b07 ZY |
988 | * Return -ENOENT if the free page is changed, otherwise 0 |
989 | * | |
b2c9e2fb ZY |
990 | * It is used when the free page crosses two pageblocks with different migratetypes |
991 | * at split_pfn_offset within the page. The split free page will be put into | |
992 | * separate migratetype lists afterwards. Otherwise, the function achieves | |
993 | * nothing. | |
994 | */ | |
86d28b07 ZY |
995 | int split_free_page(struct page *free_page, |
996 | unsigned int order, unsigned long split_pfn_offset) | |
b2c9e2fb ZY |
997 | { |
998 | struct zone *zone = page_zone(free_page); | |
999 | unsigned long free_page_pfn = page_to_pfn(free_page); | |
1000 | unsigned long pfn; | |
1001 | unsigned long flags; | |
1002 | int free_page_order; | |
86d28b07 ZY |
1003 | int mt; |
1004 | int ret = 0; | |
b2c9e2fb | 1005 | |
88ee1343 | 1006 | if (split_pfn_offset == 0) |
86d28b07 | 1007 | return ret; |
88ee1343 | 1008 | |
b2c9e2fb | 1009 | spin_lock_irqsave(&zone->lock, flags); |
86d28b07 ZY |
1010 | |
1011 | if (!PageBuddy(free_page) || buddy_order(free_page) != order) { | |
1012 | ret = -ENOENT; | |
1013 | goto out; | |
1014 | } | |
1015 | ||
1016 | mt = get_pageblock_migratetype(free_page); | |
1017 | if (likely(!is_migrate_isolate(mt))) | |
1018 | __mod_zone_freepage_state(zone, -(1UL << order), mt); | |
1019 | ||
b2c9e2fb ZY |
1020 | del_page_from_free_list(free_page, zone, order); |
1021 | for (pfn = free_page_pfn; | |
1022 | pfn < free_page_pfn + (1UL << order);) { | |
1023 | int mt = get_pfnblock_migratetype(pfn_to_page(pfn), pfn); | |
1024 | ||
86d28b07 | 1025 | free_page_order = min_t(unsigned int, |
88ee1343 ZY |
1026 | pfn ? __ffs(pfn) : order, |
1027 | __fls(split_pfn_offset)); | |
b2c9e2fb ZY |
1028 | __free_one_page(pfn_to_page(pfn), pfn, zone, free_page_order, |
1029 | mt, FPI_NONE); | |
1030 | pfn += 1UL << free_page_order; | |
1031 | split_pfn_offset -= (1UL << free_page_order); | |
1032 | /* we have done the first part, now switch to second part */ | |
1033 | if (split_pfn_offset == 0) | |
1034 | split_pfn_offset = (1UL << order) - (pfn - free_page_pfn); | |
1035 | } | |
86d28b07 | 1036 | out: |
b2c9e2fb | 1037 | spin_unlock_irqrestore(&zone->lock, flags); |
86d28b07 | 1038 | return ret; |
b2c9e2fb | 1039 | } |
7bfec6f4 MG |
1040 | /* |
1041 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1042 | * try and check multiple fields with one check. The caller must do a detailed | |
1043 | * check if necessary. | |
1044 | */ | |
1045 | static inline bool page_expected_state(struct page *page, | |
1046 | unsigned long check_flags) | |
1047 | { | |
1048 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1049 | return false; | |
1050 | ||
1051 | if (unlikely((unsigned long)page->mapping | | |
1052 | page_ref_count(page) | | |
1053 | #ifdef CONFIG_MEMCG | |
48060834 | 1054 | page->memcg_data | |
7bfec6f4 MG |
1055 | #endif |
1056 | (page->flags & check_flags))) | |
1057 | return false; | |
1058 | ||
1059 | return true; | |
1060 | } | |
1061 | ||
58b7f119 | 1062 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 1063 | { |
82a3241a | 1064 | const char *bad_reason = NULL; |
f0b791a3 | 1065 | |
53f9263b | 1066 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1067 | bad_reason = "nonzero mapcount"; |
1068 | if (unlikely(page->mapping != NULL)) | |
1069 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1070 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1071 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
1072 | if (unlikely(page->flags & flags)) { |
1073 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
1074 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
1075 | else | |
1076 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 1077 | } |
9edad6ea | 1078 | #ifdef CONFIG_MEMCG |
48060834 | 1079 | if (unlikely(page->memcg_data)) |
9edad6ea JW |
1080 | bad_reason = "page still charged to cgroup"; |
1081 | #endif | |
58b7f119 WY |
1082 | return bad_reason; |
1083 | } | |
1084 | ||
a8368cd8 | 1085 | static void free_page_is_bad_report(struct page *page) |
58b7f119 WY |
1086 | { |
1087 | bad_page(page, | |
1088 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
1089 | } |
1090 | ||
a8368cd8 | 1091 | static inline bool free_page_is_bad(struct page *page) |
bb552ac6 | 1092 | { |
da838d4f | 1093 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
a8368cd8 | 1094 | return false; |
bb552ac6 MG |
1095 | |
1096 | /* Something has gone sideways, find it */ | |
a8368cd8 AM |
1097 | free_page_is_bad_report(page); |
1098 | return true; | |
1da177e4 LT |
1099 | } |
1100 | ||
8666925c | 1101 | static int free_tail_page_prepare(struct page *head_page, struct page *page) |
4db7548c | 1102 | { |
94688e8e | 1103 | struct folio *folio = (struct folio *)head_page; |
4db7548c MG |
1104 | int ret = 1; |
1105 | ||
1106 | /* | |
1107 | * We rely page->lru.next never has bit 0 set, unless the page | |
1108 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1109 | */ | |
1110 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1111 | ||
8666925c | 1112 | if (!static_branch_unlikely(&check_pages_enabled)) { |
4db7548c MG |
1113 | ret = 0; |
1114 | goto out; | |
1115 | } | |
1116 | switch (page - head_page) { | |
1117 | case 1: | |
cb67f428 | 1118 | /* the first tail page: these may be in place of ->mapping */ |
65a689f3 MWO |
1119 | if (unlikely(folio_entire_mapcount(folio))) { |
1120 | bad_page(page, "nonzero entire_mapcount"); | |
4db7548c MG |
1121 | goto out; |
1122 | } | |
65a689f3 MWO |
1123 | if (unlikely(atomic_read(&folio->_nr_pages_mapped))) { |
1124 | bad_page(page, "nonzero nr_pages_mapped"); | |
cb67f428 HD |
1125 | goto out; |
1126 | } | |
94688e8e MWO |
1127 | if (unlikely(atomic_read(&folio->_pincount))) { |
1128 | bad_page(page, "nonzero pincount"); | |
cb67f428 HD |
1129 | goto out; |
1130 | } | |
4db7548c MG |
1131 | break; |
1132 | case 2: | |
1133 | /* | |
1134 | * the second tail page: ->mapping is | |
fa3015b7 | 1135 | * deferred_list.next -- ignore value. |
4db7548c MG |
1136 | */ |
1137 | break; | |
1138 | default: | |
1139 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 1140 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
1141 | goto out; |
1142 | } | |
1143 | break; | |
1144 | } | |
1145 | if (unlikely(!PageTail(page))) { | |
82a3241a | 1146 | bad_page(page, "PageTail not set"); |
4db7548c MG |
1147 | goto out; |
1148 | } | |
1149 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 1150 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
1151 | goto out; |
1152 | } | |
1153 | ret = 0; | |
1154 | out: | |
1155 | page->mapping = NULL; | |
1156 | clear_compound_head(page); | |
1157 | return ret; | |
1158 | } | |
1159 | ||
94ae8b83 AK |
1160 | /* |
1161 | * Skip KASAN memory poisoning when either: | |
1162 | * | |
0a54864f PC |
1163 | * 1. For generic KASAN: deferred memory initialization has not yet completed. |
1164 | * Tag-based KASAN modes skip pages freed via deferred memory initialization | |
1165 | * using page tags instead (see below). | |
1166 | * 2. For tag-based KASAN modes: the page has a match-all KASAN tag, indicating | |
1167 | * that error detection is disabled for accesses via the page address. | |
1168 | * | |
1169 | * Pages will have match-all tags in the following circumstances: | |
1170 | * | |
1171 | * 1. Pages are being initialized for the first time, including during deferred | |
1172 | * memory init; see the call to page_kasan_tag_reset in __init_single_page. | |
1173 | * 2. The allocation was not unpoisoned due to __GFP_SKIP_KASAN, with the | |
1174 | * exception of pages unpoisoned by kasan_unpoison_vmalloc. | |
1175 | * 3. The allocation was excluded from being checked due to sampling, | |
44383cef | 1176 | * see the call to kasan_unpoison_pages. |
94ae8b83 AK |
1177 | * |
1178 | * Poisoning pages during deferred memory init will greatly lengthen the | |
1179 | * process and cause problem in large memory systems as the deferred pages | |
1180 | * initialization is done with interrupt disabled. | |
1181 | * | |
1182 | * Assuming that there will be no reference to those newly initialized | |
1183 | * pages before they are ever allocated, this should have no effect on | |
1184 | * KASAN memory tracking as the poison will be properly inserted at page | |
1185 | * allocation time. The only corner case is when pages are allocated by | |
1186 | * on-demand allocation and then freed again before the deferred pages | |
1187 | * initialization is done, but this is not likely to happen. | |
1188 | */ | |
1189 | static inline bool should_skip_kasan_poison(struct page *page, fpi_t fpi_flags) | |
1190 | { | |
0a54864f PC |
1191 | if (IS_ENABLED(CONFIG_KASAN_GENERIC)) |
1192 | return deferred_pages_enabled(); | |
1193 | ||
1194 | return page_kasan_tag(page) == 0xff; | |
94ae8b83 AK |
1195 | } |
1196 | ||
aeaec8e2 | 1197 | static void kernel_init_pages(struct page *page, int numpages) |
6471384a AP |
1198 | { |
1199 | int i; | |
1200 | ||
9e15afa5 QC |
1201 | /* s390's use of memset() could override KASAN redzones. */ |
1202 | kasan_disable_current(); | |
d9da8f6c AK |
1203 | for (i = 0; i < numpages; i++) |
1204 | clear_highpage_kasan_tagged(page + i); | |
9e15afa5 | 1205 | kasan_enable_current(); |
6471384a AP |
1206 | } |
1207 | ||
e2769dbd | 1208 | static __always_inline bool free_pages_prepare(struct page *page, |
700d2e9a | 1209 | unsigned int order, fpi_t fpi_flags) |
4db7548c | 1210 | { |
e2769dbd | 1211 | int bad = 0; |
f446883d | 1212 | bool skip_kasan_poison = should_skip_kasan_poison(page, fpi_flags); |
c3525330 | 1213 | bool init = want_init_on_free(); |
4db7548c | 1214 | |
4db7548c MG |
1215 | VM_BUG_ON_PAGE(PageTail(page), page); |
1216 | ||
e2769dbd | 1217 | trace_mm_page_free(page, order); |
b073d7f8 | 1218 | kmsan_free_page(page, order); |
e2769dbd | 1219 | |
79f5f8fa OS |
1220 | if (unlikely(PageHWPoison(page)) && !order) { |
1221 | /* | |
1222 | * Do not let hwpoison pages hit pcplists/buddy | |
1223 | * Untie memcg state and reset page's owner | |
1224 | */ | |
f7a449f7 | 1225 | if (memcg_kmem_online() && PageMemcgKmem(page)) |
79f5f8fa OS |
1226 | __memcg_kmem_uncharge_page(page, order); |
1227 | reset_page_owner(page, order); | |
df4e817b | 1228 | page_table_check_free(page, order); |
79f5f8fa OS |
1229 | return false; |
1230 | } | |
1231 | ||
e2769dbd MG |
1232 | /* |
1233 | * Check tail pages before head page information is cleared to | |
1234 | * avoid checking PageCompound for order-0 pages. | |
1235 | */ | |
1236 | if (unlikely(order)) { | |
1237 | bool compound = PageCompound(page); | |
1238 | int i; | |
1239 | ||
1240 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1241 | |
cb67f428 | 1242 | if (compound) |
eac96c3e | 1243 | ClearPageHasHWPoisoned(page); |
e2769dbd MG |
1244 | for (i = 1; i < (1 << order); i++) { |
1245 | if (compound) | |
8666925c | 1246 | bad += free_tail_page_prepare(page, page + i); |
fce0b421 | 1247 | if (is_check_pages_enabled()) { |
8666925c | 1248 | if (free_page_is_bad(page + i)) { |
700d2e9a VB |
1249 | bad++; |
1250 | continue; | |
1251 | } | |
e2769dbd MG |
1252 | } |
1253 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1254 | } | |
1255 | } | |
bda807d4 | 1256 | if (PageMappingFlags(page)) |
4db7548c | 1257 | page->mapping = NULL; |
f7a449f7 | 1258 | if (memcg_kmem_online() && PageMemcgKmem(page)) |
f4b00eab | 1259 | __memcg_kmem_uncharge_page(page, order); |
fce0b421 | 1260 | if (is_check_pages_enabled()) { |
700d2e9a VB |
1261 | if (free_page_is_bad(page)) |
1262 | bad++; | |
1263 | if (bad) | |
1264 | return false; | |
1265 | } | |
4db7548c | 1266 | |
e2769dbd MG |
1267 | page_cpupid_reset_last(page); |
1268 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1269 | reset_page_owner(page, order); | |
df4e817b | 1270 | page_table_check_free(page, order); |
4db7548c MG |
1271 | |
1272 | if (!PageHighMem(page)) { | |
1273 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1274 | PAGE_SIZE << order); |
4db7548c | 1275 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1276 | PAGE_SIZE << order); |
4db7548c | 1277 | } |
6471384a | 1278 | |
8db26a3d VB |
1279 | kernel_poison_pages(page, 1 << order); |
1280 | ||
f9d79e8d | 1281 | /* |
1bb5eab3 | 1282 | * As memory initialization might be integrated into KASAN, |
7c13c163 | 1283 | * KASAN poisoning and memory initialization code must be |
1bb5eab3 AK |
1284 | * kept together to avoid discrepancies in behavior. |
1285 | * | |
f9d79e8d AK |
1286 | * With hardware tag-based KASAN, memory tags must be set before the |
1287 | * page becomes unavailable via debug_pagealloc or arch_free_page. | |
1288 | */ | |
f446883d | 1289 | if (!skip_kasan_poison) { |
c3525330 | 1290 | kasan_poison_pages(page, order, init); |
f9d79e8d | 1291 | |
db8a0477 AK |
1292 | /* Memory is already initialized if KASAN did it internally. */ |
1293 | if (kasan_has_integrated_init()) | |
1294 | init = false; | |
1295 | } | |
1296 | if (init) | |
aeaec8e2 | 1297 | kernel_init_pages(page, 1 << order); |
db8a0477 | 1298 | |
234fdce8 QC |
1299 | /* |
1300 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1301 | * does this. So nothing which can access the page's contents should | |
1302 | * happen after this. | |
1303 | */ | |
1304 | arch_free_page(page, order); | |
1305 | ||
77bc7fd6 | 1306 | debug_pagealloc_unmap_pages(page, 1 << order); |
d6332692 | 1307 | |
4db7548c MG |
1308 | return true; |
1309 | } | |
1310 | ||
1da177e4 | 1311 | /* |
5f8dcc21 | 1312 | * Frees a number of pages from the PCP lists |
7cba630b | 1313 | * Assumes all pages on list are in same zone. |
207f36ee | 1314 | * count is the number of pages to free. |
1da177e4 | 1315 | */ |
5f8dcc21 | 1316 | static void free_pcppages_bulk(struct zone *zone, int count, |
fd56eef2 MG |
1317 | struct per_cpu_pages *pcp, |
1318 | int pindex) | |
1da177e4 | 1319 | { |
57490774 | 1320 | unsigned long flags; |
35b6d770 MG |
1321 | int min_pindex = 0; |
1322 | int max_pindex = NR_PCP_LISTS - 1; | |
44042b44 | 1323 | unsigned int order; |
3777999d | 1324 | bool isolated_pageblocks; |
8b10b465 | 1325 | struct page *page; |
f2260e6b | 1326 | |
88e8ac11 CTR |
1327 | /* |
1328 | * Ensure proper count is passed which otherwise would stuck in the | |
1329 | * below while (list_empty(list)) loop. | |
1330 | */ | |
1331 | count = min(pcp->count, count); | |
d61372bc MG |
1332 | |
1333 | /* Ensure requested pindex is drained first. */ | |
1334 | pindex = pindex - 1; | |
1335 | ||
57490774 | 1336 | spin_lock_irqsave(&zone->lock, flags); |
8b10b465 MG |
1337 | isolated_pageblocks = has_isolate_pageblock(zone); |
1338 | ||
44042b44 | 1339 | while (count > 0) { |
5f8dcc21 | 1340 | struct list_head *list; |
fd56eef2 | 1341 | int nr_pages; |
5f8dcc21 | 1342 | |
fd56eef2 | 1343 | /* Remove pages from lists in a round-robin fashion. */ |
5f8dcc21 | 1344 | do { |
35b6d770 MG |
1345 | if (++pindex > max_pindex) |
1346 | pindex = min_pindex; | |
44042b44 | 1347 | list = &pcp->lists[pindex]; |
35b6d770 MG |
1348 | if (!list_empty(list)) |
1349 | break; | |
1350 | ||
1351 | if (pindex == max_pindex) | |
1352 | max_pindex--; | |
1353 | if (pindex == min_pindex) | |
1354 | min_pindex++; | |
1355 | } while (1); | |
48db57f8 | 1356 | |
44042b44 | 1357 | order = pindex_to_order(pindex); |
fd56eef2 | 1358 | nr_pages = 1 << order; |
a6f9edd6 | 1359 | do { |
8b10b465 MG |
1360 | int mt; |
1361 | ||
bf75f200 | 1362 | page = list_last_entry(list, struct page, pcp_list); |
8b10b465 MG |
1363 | mt = get_pcppage_migratetype(page); |
1364 | ||
0a5f4e5b | 1365 | /* must delete to avoid corrupting pcp list */ |
bf75f200 | 1366 | list_del(&page->pcp_list); |
fd56eef2 MG |
1367 | count -= nr_pages; |
1368 | pcp->count -= nr_pages; | |
aa016d14 | 1369 | |
8b10b465 MG |
1370 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1371 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1372 | /* Pageblock could have been isolated meanwhile */ | |
1373 | if (unlikely(isolated_pageblocks)) | |
1374 | mt = get_pageblock_migratetype(page); | |
0a5f4e5b | 1375 | |
8b10b465 MG |
1376 | __free_one_page(page, page_to_pfn(page), zone, order, mt, FPI_NONE); |
1377 | trace_mm_page_pcpu_drain(page, order, mt); | |
1378 | } while (count > 0 && !list_empty(list)); | |
0a5f4e5b | 1379 | } |
8b10b465 | 1380 | |
57490774 | 1381 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 LT |
1382 | } |
1383 | ||
dc4b0caf MG |
1384 | static void free_one_page(struct zone *zone, |
1385 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1386 | unsigned int order, |
7fef431b | 1387 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 1388 | { |
df1acc85 MG |
1389 | unsigned long flags; |
1390 | ||
1391 | spin_lock_irqsave(&zone->lock, flags); | |
ad53f92e JK |
1392 | if (unlikely(has_isolate_pageblock(zone) || |
1393 | is_migrate_isolate(migratetype))) { | |
1394 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1395 | } |
7fef431b | 1396 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); |
df1acc85 | 1397 | spin_unlock_irqrestore(&zone->lock, flags); |
48db57f8 NP |
1398 | } |
1399 | ||
7fef431b DH |
1400 | static void __free_pages_ok(struct page *page, unsigned int order, |
1401 | fpi_t fpi_flags) | |
ec95f53a | 1402 | { |
d34b0733 | 1403 | unsigned long flags; |
95e34412 | 1404 | int migratetype; |
dc4b0caf | 1405 | unsigned long pfn = page_to_pfn(page); |
56f0e661 | 1406 | struct zone *zone = page_zone(page); |
ec95f53a | 1407 | |
700d2e9a | 1408 | if (!free_pages_prepare(page, order, fpi_flags)) |
ec95f53a KM |
1409 | return; |
1410 | ||
ac4b2901 DW |
1411 | /* |
1412 | * Calling get_pfnblock_migratetype() without spin_lock_irqsave() here | |
1413 | * is used to avoid calling get_pfnblock_migratetype() under the lock. | |
1414 | * This will reduce the lock holding time. | |
1415 | */ | |
cfc47a28 | 1416 | migratetype = get_pfnblock_migratetype(page, pfn); |
dbbee9d5 | 1417 | |
56f0e661 | 1418 | spin_lock_irqsave(&zone->lock, flags); |
56f0e661 MG |
1419 | if (unlikely(has_isolate_pageblock(zone) || |
1420 | is_migrate_isolate(migratetype))) { | |
1421 | migratetype = get_pfnblock_migratetype(page, pfn); | |
1422 | } | |
1423 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); | |
1424 | spin_unlock_irqrestore(&zone->lock, flags); | |
90249993 | 1425 | |
d34b0733 | 1426 | __count_vm_events(PGFREE, 1 << order); |
1da177e4 LT |
1427 | } |
1428 | ||
a9cd410a | 1429 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1430 | { |
c3993076 | 1431 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1432 | struct page *p = page; |
c3993076 | 1433 | unsigned int loop; |
a226f6c8 | 1434 | |
7fef431b DH |
1435 | /* |
1436 | * When initializing the memmap, __init_single_page() sets the refcount | |
1437 | * of all pages to 1 ("allocated"/"not free"). We have to set the | |
1438 | * refcount of all involved pages to 0. | |
1439 | */ | |
e2d0bd2b YL |
1440 | prefetchw(p); |
1441 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1442 | prefetchw(p + 1); | |
c3993076 JW |
1443 | __ClearPageReserved(p); |
1444 | set_page_count(p, 0); | |
a226f6c8 | 1445 | } |
e2d0bd2b YL |
1446 | __ClearPageReserved(p); |
1447 | set_page_count(p, 0); | |
c3993076 | 1448 | |
9705bea5 | 1449 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
7fef431b DH |
1450 | |
1451 | /* | |
1452 | * Bypass PCP and place fresh pages right to the tail, primarily | |
1453 | * relevant for memory onlining. | |
1454 | */ | |
0a54864f | 1455 | __free_pages_ok(page, order, FPI_TO_TAIL); |
a226f6c8 DH |
1456 | } |
1457 | ||
7cf91a98 JK |
1458 | /* |
1459 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1460 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
859a85dd | 1461 | * with the migration of free compaction scanner. |
7cf91a98 JK |
1462 | * |
1463 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1464 | * | |
1465 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1466 | * i.e. it's possible that all pages within a zones range of pages do not | |
1467 | * belong to a single zone. We assume that a border between node0 and node1 | |
1468 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1469 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1470 | * the first and last page of a pageblock and avoid checking each individual | |
1471 | * page in a pageblock. | |
65f67a3e BW |
1472 | * |
1473 | * Note: the function may return non-NULL struct page even for a page block | |
1474 | * which contains a memory hole (i.e. there is no physical memory for a subset | |
1475 | * of the pfn range). For example, if the pageblock order is MAX_ORDER, which | |
1476 | * will fall into 2 sub-sections, and the end pfn of the pageblock may be hole | |
1477 | * even though the start pfn is online and valid. This should be safe most of | |
1478 | * the time because struct pages are still initialized via init_unavailable_range() | |
1479 | * and pfn walkers shouldn't touch any physical memory range for which they do | |
1480 | * not recognize any specific metadata in struct pages. | |
7cf91a98 JK |
1481 | */ |
1482 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1483 | unsigned long end_pfn, struct zone *zone) | |
1484 | { | |
1485 | struct page *start_page; | |
1486 | struct page *end_page; | |
1487 | ||
1488 | /* end_pfn is one past the range we are checking */ | |
1489 | end_pfn--; | |
1490 | ||
3c4322c9 | 1491 | if (!pfn_valid(end_pfn)) |
7cf91a98 JK |
1492 | return NULL; |
1493 | ||
2d070eab MH |
1494 | start_page = pfn_to_online_page(start_pfn); |
1495 | if (!start_page) | |
1496 | return NULL; | |
7cf91a98 JK |
1497 | |
1498 | if (page_zone(start_page) != zone) | |
1499 | return NULL; | |
1500 | ||
1501 | end_page = pfn_to_page(end_pfn); | |
1502 | ||
1503 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1504 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1505 | return NULL; | |
1506 | ||
1507 | return start_page; | |
1508 | } | |
1509 | ||
2f47a91f | 1510 | /* |
9420f89d MRI |
1511 | * The order of subdivision here is critical for the IO subsystem. |
1512 | * Please do not alter this order without good reasons and regression | |
1513 | * testing. Specifically, as large blocks of memory are subdivided, | |
1514 | * the order in which smaller blocks are delivered depends on the order | |
1515 | * they're subdivided in this function. This is the primary factor | |
1516 | * influencing the order in which pages are delivered to the IO | |
1517 | * subsystem according to empirical testing, and this is also justified | |
1518 | * by considering the behavior of a buddy system containing a single | |
1519 | * large block of memory acted on by a series of small allocations. | |
1520 | * This behavior is a critical factor in sglist merging's success. | |
80b1f41c | 1521 | * |
9420f89d | 1522 | * -- nyc |
2f47a91f | 1523 | */ |
9420f89d MRI |
1524 | static inline void expand(struct zone *zone, struct page *page, |
1525 | int low, int high, int migratetype) | |
2f47a91f | 1526 | { |
9420f89d | 1527 | unsigned long size = 1 << high; |
2f47a91f | 1528 | |
9420f89d MRI |
1529 | while (high > low) { |
1530 | high--; | |
1531 | size >>= 1; | |
1532 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); | |
2f47a91f | 1533 | |
9420f89d MRI |
1534 | /* |
1535 | * Mark as guard pages (or page), that will allow to | |
1536 | * merge back to allocator when buddy will be freed. | |
1537 | * Corresponding page table entries will not be touched, | |
1538 | * pages will stay not present in virtual address space | |
1539 | */ | |
1540 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
2f47a91f | 1541 | continue; |
9420f89d MRI |
1542 | |
1543 | add_to_free_list(&page[size], zone, high, migratetype); | |
1544 | set_buddy_order(&page[size], high); | |
2f47a91f | 1545 | } |
2f47a91f PT |
1546 | } |
1547 | ||
9420f89d | 1548 | static void check_new_page_bad(struct page *page) |
0e56acae | 1549 | { |
9420f89d MRI |
1550 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1551 | /* Don't complain about hwpoisoned pages */ | |
1552 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1553 | return; | |
0e56acae AD |
1554 | } |
1555 | ||
9420f89d MRI |
1556 | bad_page(page, |
1557 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
0e56acae AD |
1558 | } |
1559 | ||
1560 | /* | |
9420f89d | 1561 | * This page is about to be returned from the page allocator |
0e56acae | 1562 | */ |
9420f89d | 1563 | static int check_new_page(struct page *page) |
0e56acae | 1564 | { |
9420f89d MRI |
1565 | if (likely(page_expected_state(page, |
1566 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
1567 | return 0; | |
0e56acae | 1568 | |
9420f89d MRI |
1569 | check_new_page_bad(page); |
1570 | return 1; | |
1571 | } | |
0e56acae | 1572 | |
9420f89d MRI |
1573 | static inline bool check_new_pages(struct page *page, unsigned int order) |
1574 | { | |
1575 | if (is_check_pages_enabled()) { | |
1576 | for (int i = 0; i < (1 << order); i++) { | |
1577 | struct page *p = page + i; | |
0e56acae | 1578 | |
8666925c | 1579 | if (check_new_page(p)) |
9420f89d | 1580 | return true; |
0e56acae AD |
1581 | } |
1582 | } | |
1583 | ||
9420f89d | 1584 | return false; |
0e56acae AD |
1585 | } |
1586 | ||
9420f89d | 1587 | static inline bool should_skip_kasan_unpoison(gfp_t flags) |
e4443149 | 1588 | { |
9420f89d MRI |
1589 | /* Don't skip if a software KASAN mode is enabled. */ |
1590 | if (IS_ENABLED(CONFIG_KASAN_GENERIC) || | |
1591 | IS_ENABLED(CONFIG_KASAN_SW_TAGS)) | |
1592 | return false; | |
e4443149 | 1593 | |
9420f89d MRI |
1594 | /* Skip, if hardware tag-based KASAN is not enabled. */ |
1595 | if (!kasan_hw_tags_enabled()) | |
1596 | return true; | |
e4443149 DJ |
1597 | |
1598 | /* | |
9420f89d MRI |
1599 | * With hardware tag-based KASAN enabled, skip if this has been |
1600 | * requested via __GFP_SKIP_KASAN. | |
e4443149 | 1601 | */ |
9420f89d | 1602 | return flags & __GFP_SKIP_KASAN; |
e4443149 DJ |
1603 | } |
1604 | ||
9420f89d | 1605 | static inline bool should_skip_init(gfp_t flags) |
ecd09650 | 1606 | { |
9420f89d MRI |
1607 | /* Don't skip, if hardware tag-based KASAN is not enabled. */ |
1608 | if (!kasan_hw_tags_enabled()) | |
1609 | return false; | |
1610 | ||
1611 | /* For hardware tag-based KASAN, skip if requested. */ | |
1612 | return (flags & __GFP_SKIP_ZERO); | |
ecd09650 DJ |
1613 | } |
1614 | ||
9420f89d MRI |
1615 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1616 | gfp_t gfp_flags) | |
7e18adb4 | 1617 | { |
9420f89d MRI |
1618 | bool init = !want_init_on_free() && want_init_on_alloc(gfp_flags) && |
1619 | !should_skip_init(gfp_flags); | |
1620 | bool zero_tags = init && (gfp_flags & __GFP_ZEROTAGS); | |
1621 | int i; | |
1622 | ||
1623 | set_page_private(page, 0); | |
1624 | set_page_refcounted(page); | |
0e1cc95b | 1625 | |
9420f89d MRI |
1626 | arch_alloc_page(page, order); |
1627 | debug_pagealloc_map_pages(page, 1 << order); | |
7e18adb4 | 1628 | |
3d060856 | 1629 | /* |
9420f89d MRI |
1630 | * Page unpoisoning must happen before memory initialization. |
1631 | * Otherwise, the poison pattern will be overwritten for __GFP_ZERO | |
1632 | * allocations and the page unpoisoning code will complain. | |
3d060856 | 1633 | */ |
9420f89d | 1634 | kernel_unpoison_pages(page, 1 << order); |
862b6dee | 1635 | |
1bb5eab3 AK |
1636 | /* |
1637 | * As memory initialization might be integrated into KASAN, | |
b42090ae | 1638 | * KASAN unpoisoning and memory initializion code must be |
1bb5eab3 AK |
1639 | * kept together to avoid discrepancies in behavior. |
1640 | */ | |
9294b128 AK |
1641 | |
1642 | /* | |
44383cef AK |
1643 | * If memory tags should be zeroed |
1644 | * (which happens only when memory should be initialized as well). | |
9294b128 | 1645 | */ |
44383cef | 1646 | if (zero_tags) { |
420ef683 | 1647 | /* Initialize both memory and memory tags. */ |
9294b128 AK |
1648 | for (i = 0; i != 1 << order; ++i) |
1649 | tag_clear_highpage(page + i); | |
1650 | ||
44383cef | 1651 | /* Take note that memory was initialized by the loop above. */ |
9294b128 AK |
1652 | init = false; |
1653 | } | |
0a54864f PC |
1654 | if (!should_skip_kasan_unpoison(gfp_flags) && |
1655 | kasan_unpoison_pages(page, order, init)) { | |
1656 | /* Take note that memory was initialized by KASAN. */ | |
1657 | if (kasan_has_integrated_init()) | |
1658 | init = false; | |
1659 | } else { | |
1660 | /* | |
1661 | * If memory tags have not been set by KASAN, reset the page | |
1662 | * tags to ensure page_address() dereferencing does not fault. | |
1663 | */ | |
70c248ac CM |
1664 | for (i = 0; i != 1 << order; ++i) |
1665 | page_kasan_tag_reset(page + i); | |
7a3b8353 | 1666 | } |
44383cef | 1667 | /* If memory is still not initialized, initialize it now. */ |
7e3cbba6 | 1668 | if (init) |
aeaec8e2 | 1669 | kernel_init_pages(page, 1 << order); |
1bb5eab3 AK |
1670 | |
1671 | set_page_owner(page, order, gfp_flags); | |
df4e817b | 1672 | page_table_check_alloc(page, order); |
46f24fd8 JK |
1673 | } |
1674 | ||
479f854a | 1675 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1676 | unsigned int alloc_flags) |
2a7684a2 | 1677 | { |
46f24fd8 | 1678 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1679 | |
17cf4406 NP |
1680 | if (order && (gfp_flags & __GFP_COMP)) |
1681 | prep_compound_page(page, order); | |
1682 | ||
75379191 | 1683 | /* |
2f064f34 | 1684 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1685 | * allocate the page. The expectation is that the caller is taking |
1686 | * steps that will free more memory. The caller should avoid the page | |
1687 | * being used for !PFMEMALLOC purposes. | |
1688 | */ | |
2f064f34 MH |
1689 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1690 | set_page_pfmemalloc(page); | |
1691 | else | |
1692 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1693 | } |
1694 | ||
56fd56b8 MG |
1695 | /* |
1696 | * Go through the free lists for the given migratetype and remove | |
1697 | * the smallest available page from the freelists | |
1698 | */ | |
85ccc8fa | 1699 | static __always_inline |
728ec980 | 1700 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
1701 | int migratetype) |
1702 | { | |
1703 | unsigned int current_order; | |
b8af2941 | 1704 | struct free_area *area; |
56fd56b8 MG |
1705 | struct page *page; |
1706 | ||
1707 | /* Find a page of the appropriate size in the preferred list */ | |
23baf831 | 1708 | for (current_order = order; current_order <= MAX_ORDER; ++current_order) { |
56fd56b8 | 1709 | area = &(zone->free_area[current_order]); |
b03641af | 1710 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
1711 | if (!page) |
1712 | continue; | |
6ab01363 AD |
1713 | del_page_from_free_list(page, zone, current_order); |
1714 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 1715 | set_pcppage_migratetype(page, migratetype); |
10e0f753 WY |
1716 | trace_mm_page_alloc_zone_locked(page, order, migratetype, |
1717 | pcp_allowed_order(order) && | |
1718 | migratetype < MIGRATE_PCPTYPES); | |
56fd56b8 MG |
1719 | return page; |
1720 | } | |
1721 | ||
1722 | return NULL; | |
1723 | } | |
1724 | ||
1725 | ||
b2a0ac88 MG |
1726 | /* |
1727 | * This array describes the order lists are fallen back to when | |
1728 | * the free lists for the desirable migrate type are depleted | |
1dd214b8 ZY |
1729 | * |
1730 | * The other migratetypes do not have fallbacks. | |
b2a0ac88 | 1731 | */ |
aa02d3c1 YD |
1732 | static int fallbacks[MIGRATE_TYPES][MIGRATE_PCPTYPES - 1] = { |
1733 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE }, | |
1734 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE }, | |
1735 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE }, | |
b2a0ac88 MG |
1736 | }; |
1737 | ||
dc67647b | 1738 | #ifdef CONFIG_CMA |
85ccc8fa | 1739 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
1740 | unsigned int order) |
1741 | { | |
1742 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1743 | } | |
1744 | #else | |
1745 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1746 | unsigned int order) { return NULL; } | |
1747 | #endif | |
1748 | ||
c361be55 | 1749 | /* |
293ffa5e | 1750 | * Move the free pages in a range to the freelist tail of the requested type. |
d9c23400 | 1751 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1752 | * boundary. If alignment is required, use move_freepages_block() |
1753 | */ | |
02aa0cdd | 1754 | static int move_freepages(struct zone *zone, |
39ddb991 | 1755 | unsigned long start_pfn, unsigned long end_pfn, |
02aa0cdd | 1756 | int migratetype, int *num_movable) |
c361be55 MG |
1757 | { |
1758 | struct page *page; | |
39ddb991 | 1759 | unsigned long pfn; |
d00181b9 | 1760 | unsigned int order; |
d100313f | 1761 | int pages_moved = 0; |
c361be55 | 1762 | |
39ddb991 | 1763 | for (pfn = start_pfn; pfn <= end_pfn;) { |
39ddb991 | 1764 | page = pfn_to_page(pfn); |
c361be55 | 1765 | if (!PageBuddy(page)) { |
02aa0cdd VB |
1766 | /* |
1767 | * We assume that pages that could be isolated for | |
1768 | * migration are movable. But we don't actually try | |
1769 | * isolating, as that would be expensive. | |
1770 | */ | |
1771 | if (num_movable && | |
1772 | (PageLRU(page) || __PageMovable(page))) | |
1773 | (*num_movable)++; | |
39ddb991 | 1774 | pfn++; |
c361be55 MG |
1775 | continue; |
1776 | } | |
1777 | ||
cd961038 DR |
1778 | /* Make sure we are not inadvertently changing nodes */ |
1779 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
1780 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
1781 | ||
ab130f91 | 1782 | order = buddy_order(page); |
6ab01363 | 1783 | move_to_free_list(page, zone, order, migratetype); |
39ddb991 | 1784 | pfn += 1 << order; |
d100313f | 1785 | pages_moved += 1 << order; |
c361be55 MG |
1786 | } |
1787 | ||
d100313f | 1788 | return pages_moved; |
c361be55 MG |
1789 | } |
1790 | ||
ee6f509c | 1791 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 1792 | int migratetype, int *num_movable) |
c361be55 | 1793 | { |
39ddb991 | 1794 | unsigned long start_pfn, end_pfn, pfn; |
c361be55 | 1795 | |
4a222127 DR |
1796 | if (num_movable) |
1797 | *num_movable = 0; | |
1798 | ||
39ddb991 | 1799 | pfn = page_to_pfn(page); |
4f9bc69a KW |
1800 | start_pfn = pageblock_start_pfn(pfn); |
1801 | end_pfn = pageblock_end_pfn(pfn) - 1; | |
c361be55 MG |
1802 | |
1803 | /* Do not cross zone boundaries */ | |
108bcc96 | 1804 | if (!zone_spans_pfn(zone, start_pfn)) |
39ddb991 | 1805 | start_pfn = pfn; |
108bcc96 | 1806 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1807 | return 0; |
1808 | ||
39ddb991 | 1809 | return move_freepages(zone, start_pfn, end_pfn, migratetype, |
02aa0cdd | 1810 | num_movable); |
c361be55 MG |
1811 | } |
1812 | ||
2f66a68f MG |
1813 | static void change_pageblock_range(struct page *pageblock_page, |
1814 | int start_order, int migratetype) | |
1815 | { | |
1816 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1817 | ||
1818 | while (nr_pageblocks--) { | |
1819 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1820 | pageblock_page += pageblock_nr_pages; | |
1821 | } | |
1822 | } | |
1823 | ||
fef903ef | 1824 | /* |
9c0415eb VB |
1825 | * When we are falling back to another migratetype during allocation, try to |
1826 | * steal extra free pages from the same pageblocks to satisfy further | |
1827 | * allocations, instead of polluting multiple pageblocks. | |
1828 | * | |
1829 | * If we are stealing a relatively large buddy page, it is likely there will | |
1830 | * be more free pages in the pageblock, so try to steal them all. For | |
1831 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1832 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1833 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1834 | * pageblocks. | |
fef903ef | 1835 | */ |
4eb7dce6 JK |
1836 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1837 | { | |
1838 | /* | |
1839 | * Leaving this order check is intended, although there is | |
1840 | * relaxed order check in next check. The reason is that | |
1841 | * we can actually steal whole pageblock if this condition met, | |
1842 | * but, below check doesn't guarantee it and that is just heuristic | |
1843 | * so could be changed anytime. | |
1844 | */ | |
1845 | if (order >= pageblock_order) | |
1846 | return true; | |
1847 | ||
1848 | if (order >= pageblock_order / 2 || | |
1849 | start_mt == MIGRATE_RECLAIMABLE || | |
1850 | start_mt == MIGRATE_UNMOVABLE || | |
1851 | page_group_by_mobility_disabled) | |
1852 | return true; | |
1853 | ||
1854 | return false; | |
1855 | } | |
1856 | ||
597c8920 | 1857 | static inline bool boost_watermark(struct zone *zone) |
1c30844d MG |
1858 | { |
1859 | unsigned long max_boost; | |
1860 | ||
1861 | if (!watermark_boost_factor) | |
597c8920 | 1862 | return false; |
14f69140 HW |
1863 | /* |
1864 | * Don't bother in zones that are unlikely to produce results. | |
1865 | * On small machines, including kdump capture kernels running | |
1866 | * in a small area, boosting the watermark can cause an out of | |
1867 | * memory situation immediately. | |
1868 | */ | |
1869 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
597c8920 | 1870 | return false; |
1c30844d MG |
1871 | |
1872 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
1873 | watermark_boost_factor, 10000); | |
94b3334c MG |
1874 | |
1875 | /* | |
1876 | * high watermark may be uninitialised if fragmentation occurs | |
1877 | * very early in boot so do not boost. We do not fall | |
1878 | * through and boost by pageblock_nr_pages as failing | |
1879 | * allocations that early means that reclaim is not going | |
1880 | * to help and it may even be impossible to reclaim the | |
1881 | * boosted watermark resulting in a hang. | |
1882 | */ | |
1883 | if (!max_boost) | |
597c8920 | 1884 | return false; |
94b3334c | 1885 | |
1c30844d MG |
1886 | max_boost = max(pageblock_nr_pages, max_boost); |
1887 | ||
1888 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
1889 | max_boost); | |
597c8920 JW |
1890 | |
1891 | return true; | |
1c30844d MG |
1892 | } |
1893 | ||
4eb7dce6 JK |
1894 | /* |
1895 | * This function implements actual steal behaviour. If order is large enough, | |
1896 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
1897 | * pageblock to our migratetype and determine how many already-allocated pages |
1898 | * are there in the pageblock with a compatible migratetype. If at least half | |
1899 | * of pages are free or compatible, we can change migratetype of the pageblock | |
1900 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
1901 | */ |
1902 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 1903 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 1904 | { |
ab130f91 | 1905 | unsigned int current_order = buddy_order(page); |
02aa0cdd VB |
1906 | int free_pages, movable_pages, alike_pages; |
1907 | int old_block_type; | |
1908 | ||
1909 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 1910 | |
3bc48f96 VB |
1911 | /* |
1912 | * This can happen due to races and we want to prevent broken | |
1913 | * highatomic accounting. | |
1914 | */ | |
02aa0cdd | 1915 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
1916 | goto single_page; |
1917 | ||
fef903ef SB |
1918 | /* Take ownership for orders >= pageblock_order */ |
1919 | if (current_order >= pageblock_order) { | |
1920 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 1921 | goto single_page; |
fef903ef SB |
1922 | } |
1923 | ||
1c30844d MG |
1924 | /* |
1925 | * Boost watermarks to increase reclaim pressure to reduce the | |
1926 | * likelihood of future fallbacks. Wake kswapd now as the node | |
1927 | * may be balanced overall and kswapd will not wake naturally. | |
1928 | */ | |
597c8920 | 1929 | if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD)) |
73444bc4 | 1930 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 1931 | |
3bc48f96 VB |
1932 | /* We are not allowed to try stealing from the whole block */ |
1933 | if (!whole_block) | |
1934 | goto single_page; | |
1935 | ||
02aa0cdd VB |
1936 | free_pages = move_freepages_block(zone, page, start_type, |
1937 | &movable_pages); | |
1938 | /* | |
1939 | * Determine how many pages are compatible with our allocation. | |
1940 | * For movable allocation, it's the number of movable pages which | |
1941 | * we just obtained. For other types it's a bit more tricky. | |
1942 | */ | |
1943 | if (start_type == MIGRATE_MOVABLE) { | |
1944 | alike_pages = movable_pages; | |
1945 | } else { | |
1946 | /* | |
1947 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
1948 | * to MOVABLE pageblock, consider all non-movable pages as | |
1949 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
1950 | * vice versa, be conservative since we can't distinguish the | |
1951 | * exact migratetype of non-movable pages. | |
1952 | */ | |
1953 | if (old_block_type == MIGRATE_MOVABLE) | |
1954 | alike_pages = pageblock_nr_pages | |
1955 | - (free_pages + movable_pages); | |
1956 | else | |
1957 | alike_pages = 0; | |
1958 | } | |
1959 | ||
3bc48f96 | 1960 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 1961 | if (!free_pages) |
3bc48f96 | 1962 | goto single_page; |
fef903ef | 1963 | |
02aa0cdd VB |
1964 | /* |
1965 | * If a sufficient number of pages in the block are either free or of | |
1966 | * comparable migratability as our allocation, claim the whole block. | |
1967 | */ | |
1968 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
1969 | page_group_by_mobility_disabled) |
1970 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
1971 | |
1972 | return; | |
1973 | ||
1974 | single_page: | |
6ab01363 | 1975 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
1976 | } |
1977 | ||
2149cdae JK |
1978 | /* |
1979 | * Check whether there is a suitable fallback freepage with requested order. | |
1980 | * If only_stealable is true, this function returns fallback_mt only if | |
1981 | * we can steal other freepages all together. This would help to reduce | |
1982 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1983 | */ | |
1984 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1985 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1986 | { |
1987 | int i; | |
1988 | int fallback_mt; | |
1989 | ||
1990 | if (area->nr_free == 0) | |
1991 | return -1; | |
1992 | ||
1993 | *can_steal = false; | |
aa02d3c1 | 1994 | for (i = 0; i < MIGRATE_PCPTYPES - 1 ; i++) { |
4eb7dce6 | 1995 | fallback_mt = fallbacks[migratetype][i]; |
b03641af | 1996 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 1997 | continue; |
fef903ef | 1998 | |
4eb7dce6 JK |
1999 | if (can_steal_fallback(order, migratetype)) |
2000 | *can_steal = true; | |
2001 | ||
2149cdae JK |
2002 | if (!only_stealable) |
2003 | return fallback_mt; | |
2004 | ||
2005 | if (*can_steal) | |
2006 | return fallback_mt; | |
fef903ef | 2007 | } |
4eb7dce6 JK |
2008 | |
2009 | return -1; | |
fef903ef SB |
2010 | } |
2011 | ||
0aaa29a5 MG |
2012 | /* |
2013 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2014 | * there are no empty page blocks that contain a page with a suitable order | |
2015 | */ | |
2016 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2017 | unsigned int alloc_order) | |
2018 | { | |
2019 | int mt; | |
2020 | unsigned long max_managed, flags; | |
2021 | ||
2022 | /* | |
2023 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2024 | * Check is race-prone but harmless. | |
2025 | */ | |
9705bea5 | 2026 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2027 | if (zone->nr_reserved_highatomic >= max_managed) |
2028 | return; | |
2029 | ||
2030 | spin_lock_irqsave(&zone->lock, flags); | |
2031 | ||
2032 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2033 | if (zone->nr_reserved_highatomic >= max_managed) | |
2034 | goto out_unlock; | |
2035 | ||
2036 | /* Yoink! */ | |
2037 | mt = get_pageblock_migratetype(page); | |
1dd214b8 ZY |
2038 | /* Only reserve normal pageblocks (i.e., they can merge with others) */ |
2039 | if (migratetype_is_mergeable(mt)) { | |
0aaa29a5 MG |
2040 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2041 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2042 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2043 | } |
2044 | ||
2045 | out_unlock: | |
2046 | spin_unlock_irqrestore(&zone->lock, flags); | |
2047 | } | |
2048 | ||
2049 | /* | |
2050 | * Used when an allocation is about to fail under memory pressure. This | |
2051 | * potentially hurts the reliability of high-order allocations when under | |
2052 | * intense memory pressure but failed atomic allocations should be easier | |
2053 | * to recover from than an OOM. | |
29fac03b MK |
2054 | * |
2055 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2056 | * pageblock is exhausted. | |
0aaa29a5 | 2057 | */ |
29fac03b MK |
2058 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2059 | bool force) | |
0aaa29a5 MG |
2060 | { |
2061 | struct zonelist *zonelist = ac->zonelist; | |
2062 | unsigned long flags; | |
2063 | struct zoneref *z; | |
2064 | struct zone *zone; | |
2065 | struct page *page; | |
2066 | int order; | |
04c8716f | 2067 | bool ret; |
0aaa29a5 | 2068 | |
97a225e6 | 2069 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2070 | ac->nodemask) { |
29fac03b MK |
2071 | /* |
2072 | * Preserve at least one pageblock unless memory pressure | |
2073 | * is really high. | |
2074 | */ | |
2075 | if (!force && zone->nr_reserved_highatomic <= | |
2076 | pageblock_nr_pages) | |
0aaa29a5 MG |
2077 | continue; |
2078 | ||
2079 | spin_lock_irqsave(&zone->lock, flags); | |
23baf831 | 2080 | for (order = 0; order <= MAX_ORDER; order++) { |
0aaa29a5 MG |
2081 | struct free_area *area = &(zone->free_area[order]); |
2082 | ||
b03641af | 2083 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2084 | if (!page) |
0aaa29a5 MG |
2085 | continue; |
2086 | ||
0aaa29a5 | 2087 | /* |
4855e4a7 MK |
2088 | * In page freeing path, migratetype change is racy so |
2089 | * we can counter several free pages in a pageblock | |
f0953a1b | 2090 | * in this loop although we changed the pageblock type |
4855e4a7 MK |
2091 | * from highatomic to ac->migratetype. So we should |
2092 | * adjust the count once. | |
0aaa29a5 | 2093 | */ |
a6ffdc07 | 2094 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2095 | /* |
2096 | * It should never happen but changes to | |
2097 | * locking could inadvertently allow a per-cpu | |
2098 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2099 | * while unreserving so be safe and watch for | |
2100 | * underflows. | |
2101 | */ | |
2102 | zone->nr_reserved_highatomic -= min( | |
2103 | pageblock_nr_pages, | |
2104 | zone->nr_reserved_highatomic); | |
2105 | } | |
0aaa29a5 MG |
2106 | |
2107 | /* | |
2108 | * Convert to ac->migratetype and avoid the normal | |
2109 | * pageblock stealing heuristics. Minimally, the caller | |
2110 | * is doing the work and needs the pages. More | |
2111 | * importantly, if the block was always converted to | |
2112 | * MIGRATE_UNMOVABLE or another type then the number | |
2113 | * of pageblocks that cannot be completely freed | |
2114 | * may increase. | |
2115 | */ | |
2116 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2117 | ret = move_freepages_block(zone, page, ac->migratetype, |
2118 | NULL); | |
29fac03b MK |
2119 | if (ret) { |
2120 | spin_unlock_irqrestore(&zone->lock, flags); | |
2121 | return ret; | |
2122 | } | |
0aaa29a5 MG |
2123 | } |
2124 | spin_unlock_irqrestore(&zone->lock, flags); | |
2125 | } | |
04c8716f MK |
2126 | |
2127 | return false; | |
0aaa29a5 MG |
2128 | } |
2129 | ||
3bc48f96 VB |
2130 | /* |
2131 | * Try finding a free buddy page on the fallback list and put it on the free | |
2132 | * list of requested migratetype, possibly along with other pages from the same | |
2133 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2134 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2135 | * |
2136 | * The use of signed ints for order and current_order is a deliberate | |
2137 | * deviation from the rest of this file, to make the for loop | |
2138 | * condition simpler. | |
3bc48f96 | 2139 | */ |
85ccc8fa | 2140 | static __always_inline bool |
6bb15450 MG |
2141 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2142 | unsigned int alloc_flags) | |
b2a0ac88 | 2143 | { |
b8af2941 | 2144 | struct free_area *area; |
b002529d | 2145 | int current_order; |
6bb15450 | 2146 | int min_order = order; |
b2a0ac88 | 2147 | struct page *page; |
4eb7dce6 JK |
2148 | int fallback_mt; |
2149 | bool can_steal; | |
b2a0ac88 | 2150 | |
6bb15450 MG |
2151 | /* |
2152 | * Do not steal pages from freelists belonging to other pageblocks | |
2153 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2154 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2155 | */ | |
e933dc4a | 2156 | if (order < pageblock_order && alloc_flags & ALLOC_NOFRAGMENT) |
6bb15450 MG |
2157 | min_order = pageblock_order; |
2158 | ||
7a8f58f3 VB |
2159 | /* |
2160 | * Find the largest available free page in the other list. This roughly | |
2161 | * approximates finding the pageblock with the most free pages, which | |
2162 | * would be too costly to do exactly. | |
2163 | */ | |
23baf831 | 2164 | for (current_order = MAX_ORDER; current_order >= min_order; |
7aeb09f9 | 2165 | --current_order) { |
4eb7dce6 JK |
2166 | area = &(zone->free_area[current_order]); |
2167 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2168 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2169 | if (fallback_mt == -1) |
2170 | continue; | |
b2a0ac88 | 2171 | |
7a8f58f3 VB |
2172 | /* |
2173 | * We cannot steal all free pages from the pageblock and the | |
2174 | * requested migratetype is movable. In that case it's better to | |
2175 | * steal and split the smallest available page instead of the | |
2176 | * largest available page, because even if the next movable | |
2177 | * allocation falls back into a different pageblock than this | |
2178 | * one, it won't cause permanent fragmentation. | |
2179 | */ | |
2180 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2181 | && current_order > order) | |
2182 | goto find_smallest; | |
b2a0ac88 | 2183 | |
7a8f58f3 VB |
2184 | goto do_steal; |
2185 | } | |
e0fff1bd | 2186 | |
7a8f58f3 | 2187 | return false; |
e0fff1bd | 2188 | |
7a8f58f3 | 2189 | find_smallest: |
23baf831 | 2190 | for (current_order = order; current_order <= MAX_ORDER; |
7a8f58f3 VB |
2191 | current_order++) { |
2192 | area = &(zone->free_area[current_order]); | |
2193 | fallback_mt = find_suitable_fallback(area, current_order, | |
2194 | start_migratetype, false, &can_steal); | |
2195 | if (fallback_mt != -1) | |
2196 | break; | |
b2a0ac88 MG |
2197 | } |
2198 | ||
7a8f58f3 VB |
2199 | /* |
2200 | * This should not happen - we already found a suitable fallback | |
2201 | * when looking for the largest page. | |
2202 | */ | |
23baf831 | 2203 | VM_BUG_ON(current_order > MAX_ORDER); |
7a8f58f3 VB |
2204 | |
2205 | do_steal: | |
b03641af | 2206 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2207 | |
1c30844d MG |
2208 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
2209 | can_steal); | |
7a8f58f3 VB |
2210 | |
2211 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2212 | start_migratetype, fallback_mt); | |
2213 | ||
2214 | return true; | |
2215 | ||
b2a0ac88 MG |
2216 | } |
2217 | ||
56fd56b8 | 2218 | /* |
1da177e4 LT |
2219 | * Do the hard work of removing an element from the buddy allocator. |
2220 | * Call me with the zone->lock already held. | |
2221 | */ | |
85ccc8fa | 2222 | static __always_inline struct page * |
6bb15450 MG |
2223 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2224 | unsigned int alloc_flags) | |
1da177e4 | 2225 | { |
1da177e4 LT |
2226 | struct page *page; |
2227 | ||
ce8f86ee H |
2228 | if (IS_ENABLED(CONFIG_CMA)) { |
2229 | /* | |
2230 | * Balance movable allocations between regular and CMA areas by | |
2231 | * allocating from CMA when over half of the zone's free memory | |
2232 | * is in the CMA area. | |
2233 | */ | |
2234 | if (alloc_flags & ALLOC_CMA && | |
2235 | zone_page_state(zone, NR_FREE_CMA_PAGES) > | |
2236 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
2237 | page = __rmqueue_cma_fallback(zone, order); | |
2238 | if (page) | |
10e0f753 | 2239 | return page; |
ce8f86ee | 2240 | } |
16867664 | 2241 | } |
3bc48f96 | 2242 | retry: |
56fd56b8 | 2243 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2244 | if (unlikely(!page)) { |
8510e69c | 2245 | if (alloc_flags & ALLOC_CMA) |
dc67647b JK |
2246 | page = __rmqueue_cma_fallback(zone, order); |
2247 | ||
6bb15450 MG |
2248 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2249 | alloc_flags)) | |
3bc48f96 | 2250 | goto retry; |
728ec980 | 2251 | } |
b2a0ac88 | 2252 | return page; |
1da177e4 LT |
2253 | } |
2254 | ||
5f63b720 | 2255 | /* |
1da177e4 LT |
2256 | * Obtain a specified number of elements from the buddy allocator, all under |
2257 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2258 | * Returns the number of new pages which were placed at *list. | |
2259 | */ | |
5f63b720 | 2260 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2261 | unsigned long count, struct list_head *list, |
6bb15450 | 2262 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2263 | { |
57490774 | 2264 | unsigned long flags; |
700d2e9a | 2265 | int i; |
5f63b720 | 2266 | |
57490774 | 2267 | spin_lock_irqsave(&zone->lock, flags); |
1da177e4 | 2268 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2269 | struct page *page = __rmqueue(zone, order, migratetype, |
2270 | alloc_flags); | |
085cc7d5 | 2271 | if (unlikely(page == NULL)) |
1da177e4 | 2272 | break; |
81eabcbe MG |
2273 | |
2274 | /* | |
0fac3ba5 VB |
2275 | * Split buddy pages returned by expand() are received here in |
2276 | * physical page order. The page is added to the tail of | |
2277 | * caller's list. From the callers perspective, the linked list | |
2278 | * is ordered by page number under some conditions. This is | |
2279 | * useful for IO devices that can forward direction from the | |
2280 | * head, thus also in the physical page order. This is useful | |
2281 | * for IO devices that can merge IO requests if the physical | |
2282 | * pages are ordered properly. | |
81eabcbe | 2283 | */ |
bf75f200 | 2284 | list_add_tail(&page->pcp_list, list); |
bb14c2c7 | 2285 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2286 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2287 | -(1 << order)); | |
1da177e4 | 2288 | } |
a6de734b | 2289 | |
f2260e6b | 2290 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
57490774 | 2291 | spin_unlock_irqrestore(&zone->lock, flags); |
2ede3c13 | 2292 | |
700d2e9a | 2293 | return i; |
1da177e4 LT |
2294 | } |
2295 | ||
4ae7c039 | 2296 | #ifdef CONFIG_NUMA |
8fce4d8e | 2297 | /* |
4037d452 CL |
2298 | * Called from the vmstat counter updater to drain pagesets of this |
2299 | * currently executing processor on remote nodes after they have | |
2300 | * expired. | |
8fce4d8e | 2301 | */ |
4037d452 | 2302 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2303 | { |
7be12fc9 | 2304 | int to_drain, batch; |
4ae7c039 | 2305 | |
4db0c3c2 | 2306 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2307 | to_drain = min(pcp->count, batch); |
4b23a68f | 2308 | if (to_drain > 0) { |
57490774 | 2309 | spin_lock(&pcp->lock); |
fd56eef2 | 2310 | free_pcppages_bulk(zone, to_drain, pcp, 0); |
57490774 | 2311 | spin_unlock(&pcp->lock); |
4b23a68f | 2312 | } |
4ae7c039 CL |
2313 | } |
2314 | #endif | |
2315 | ||
9f8f2172 | 2316 | /* |
93481ff0 | 2317 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 | 2318 | */ |
93481ff0 | 2319 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2320 | { |
93481ff0 | 2321 | struct per_cpu_pages *pcp; |
1da177e4 | 2322 | |
28f836b6 | 2323 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
4b23a68f | 2324 | if (pcp->count) { |
57490774 | 2325 | spin_lock(&pcp->lock); |
4b23a68f | 2326 | free_pcppages_bulk(zone, pcp->count, pcp, 0); |
57490774 | 2327 | spin_unlock(&pcp->lock); |
4b23a68f | 2328 | } |
93481ff0 | 2329 | } |
3dfa5721 | 2330 | |
93481ff0 VB |
2331 | /* |
2332 | * Drain pcplists of all zones on the indicated processor. | |
93481ff0 VB |
2333 | */ |
2334 | static void drain_pages(unsigned int cpu) | |
2335 | { | |
2336 | struct zone *zone; | |
2337 | ||
2338 | for_each_populated_zone(zone) { | |
2339 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2340 | } |
2341 | } | |
1da177e4 | 2342 | |
9f8f2172 CL |
2343 | /* |
2344 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
2345 | */ | |
93481ff0 | 2346 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2347 | { |
93481ff0 VB |
2348 | int cpu = smp_processor_id(); |
2349 | ||
2350 | if (zone) | |
2351 | drain_pages_zone(cpu, zone); | |
2352 | else | |
2353 | drain_pages(cpu); | |
9f8f2172 CL |
2354 | } |
2355 | ||
2356 | /* | |
ec6e8c7e VB |
2357 | * The implementation of drain_all_pages(), exposing an extra parameter to |
2358 | * drain on all cpus. | |
93481ff0 | 2359 | * |
ec6e8c7e VB |
2360 | * drain_all_pages() is optimized to only execute on cpus where pcplists are |
2361 | * not empty. The check for non-emptiness can however race with a free to | |
2362 | * pcplist that has not yet increased the pcp->count from 0 to 1. Callers | |
2363 | * that need the guarantee that every CPU has drained can disable the | |
2364 | * optimizing racy check. | |
9f8f2172 | 2365 | */ |
3b1f3658 | 2366 | static void __drain_all_pages(struct zone *zone, bool force_all_cpus) |
9f8f2172 | 2367 | { |
74046494 | 2368 | int cpu; |
74046494 GBY |
2369 | |
2370 | /* | |
041711ce | 2371 | * Allocate in the BSS so we won't require allocation in |
74046494 GBY |
2372 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y |
2373 | */ | |
2374 | static cpumask_t cpus_with_pcps; | |
2375 | ||
bd233f53 MG |
2376 | /* |
2377 | * Do not drain if one is already in progress unless it's specific to | |
2378 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2379 | * the drain to be complete when the call returns. | |
2380 | */ | |
2381 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2382 | if (!zone) | |
2383 | return; | |
2384 | mutex_lock(&pcpu_drain_mutex); | |
2385 | } | |
0ccce3b9 | 2386 | |
74046494 GBY |
2387 | /* |
2388 | * We don't care about racing with CPU hotplug event | |
2389 | * as offline notification will cause the notified | |
2390 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2391 | * disables preemption as part of its processing | |
2392 | */ | |
2393 | for_each_online_cpu(cpu) { | |
28f836b6 | 2394 | struct per_cpu_pages *pcp; |
93481ff0 | 2395 | struct zone *z; |
74046494 | 2396 | bool has_pcps = false; |
93481ff0 | 2397 | |
ec6e8c7e VB |
2398 | if (force_all_cpus) { |
2399 | /* | |
2400 | * The pcp.count check is racy, some callers need a | |
2401 | * guarantee that no cpu is missed. | |
2402 | */ | |
2403 | has_pcps = true; | |
2404 | } else if (zone) { | |
28f836b6 MG |
2405 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
2406 | if (pcp->count) | |
74046494 | 2407 | has_pcps = true; |
93481ff0 VB |
2408 | } else { |
2409 | for_each_populated_zone(z) { | |
28f836b6 MG |
2410 | pcp = per_cpu_ptr(z->per_cpu_pageset, cpu); |
2411 | if (pcp->count) { | |
93481ff0 VB |
2412 | has_pcps = true; |
2413 | break; | |
2414 | } | |
74046494 GBY |
2415 | } |
2416 | } | |
93481ff0 | 2417 | |
74046494 GBY |
2418 | if (has_pcps) |
2419 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2420 | else | |
2421 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2422 | } | |
0ccce3b9 | 2423 | |
bd233f53 | 2424 | for_each_cpu(cpu, &cpus_with_pcps) { |
443c2acc NSJ |
2425 | if (zone) |
2426 | drain_pages_zone(cpu, zone); | |
2427 | else | |
2428 | drain_pages(cpu); | |
0ccce3b9 | 2429 | } |
bd233f53 MG |
2430 | |
2431 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2432 | } |
2433 | ||
ec6e8c7e VB |
2434 | /* |
2435 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. | |
2436 | * | |
2437 | * When zone parameter is non-NULL, spill just the single zone's pages. | |
ec6e8c7e VB |
2438 | */ |
2439 | void drain_all_pages(struct zone *zone) | |
2440 | { | |
2441 | __drain_all_pages(zone, false); | |
2442 | } | |
2443 | ||
296699de | 2444 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 2445 | |
556b969a CY |
2446 | /* |
2447 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
2448 | */ | |
2449 | #define WD_PAGE_COUNT (128*1024) | |
2450 | ||
1da177e4 LT |
2451 | void mark_free_pages(struct zone *zone) |
2452 | { | |
556b969a | 2453 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 2454 | unsigned long flags; |
7aeb09f9 | 2455 | unsigned int order, t; |
86760a2c | 2456 | struct page *page; |
1da177e4 | 2457 | |
8080fc03 | 2458 | if (zone_is_empty(zone)) |
1da177e4 LT |
2459 | return; |
2460 | ||
2461 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2462 | |
108bcc96 | 2463 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2464 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2465 | if (pfn_valid(pfn)) { | |
86760a2c | 2466 | page = pfn_to_page(pfn); |
ba6b0979 | 2467 | |
556b969a CY |
2468 | if (!--page_count) { |
2469 | touch_nmi_watchdog(); | |
2470 | page_count = WD_PAGE_COUNT; | |
2471 | } | |
2472 | ||
ba6b0979 JK |
2473 | if (page_zone(page) != zone) |
2474 | continue; | |
2475 | ||
7be98234 RW |
2476 | if (!swsusp_page_is_forbidden(page)) |
2477 | swsusp_unset_page_free(page); | |
f623f0db | 2478 | } |
1da177e4 | 2479 | |
b2a0ac88 | 2480 | for_each_migratetype_order(order, t) { |
86760a2c | 2481 | list_for_each_entry(page, |
bf75f200 | 2482 | &zone->free_area[order].free_list[t], buddy_list) { |
f623f0db | 2483 | unsigned long i; |
1da177e4 | 2484 | |
86760a2c | 2485 | pfn = page_to_pfn(page); |
556b969a CY |
2486 | for (i = 0; i < (1UL << order); i++) { |
2487 | if (!--page_count) { | |
2488 | touch_nmi_watchdog(); | |
2489 | page_count = WD_PAGE_COUNT; | |
2490 | } | |
7be98234 | 2491 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 2492 | } |
f623f0db | 2493 | } |
b2a0ac88 | 2494 | } |
1da177e4 LT |
2495 | spin_unlock_irqrestore(&zone->lock, flags); |
2496 | } | |
e2c55dc8 | 2497 | #endif /* CONFIG_PM */ |
1da177e4 | 2498 | |
44042b44 MG |
2499 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn, |
2500 | unsigned int order) | |
1da177e4 | 2501 | { |
5f8dcc21 | 2502 | int migratetype; |
1da177e4 | 2503 | |
700d2e9a | 2504 | if (!free_pages_prepare(page, order, FPI_NONE)) |
9cca35d4 | 2505 | return false; |
689bcebf | 2506 | |
dc4b0caf | 2507 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2508 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
2509 | return true; |
2510 | } | |
2511 | ||
f26b3fa0 MG |
2512 | static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch, |
2513 | bool free_high) | |
3b12e7e9 MG |
2514 | { |
2515 | int min_nr_free, max_nr_free; | |
2516 | ||
f26b3fa0 MG |
2517 | /* Free everything if batch freeing high-order pages. */ |
2518 | if (unlikely(free_high)) | |
2519 | return pcp->count; | |
2520 | ||
3b12e7e9 MG |
2521 | /* Check for PCP disabled or boot pageset */ |
2522 | if (unlikely(high < batch)) | |
2523 | return 1; | |
2524 | ||
2525 | /* Leave at least pcp->batch pages on the list */ | |
2526 | min_nr_free = batch; | |
2527 | max_nr_free = high - batch; | |
2528 | ||
2529 | /* | |
2530 | * Double the number of pages freed each time there is subsequent | |
2531 | * freeing of pages without any allocation. | |
2532 | */ | |
2533 | batch <<= pcp->free_factor; | |
2534 | if (batch < max_nr_free) | |
2535 | pcp->free_factor++; | |
2536 | batch = clamp(batch, min_nr_free, max_nr_free); | |
2537 | ||
2538 | return batch; | |
2539 | } | |
2540 | ||
f26b3fa0 MG |
2541 | static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone, |
2542 | bool free_high) | |
c49c2c47 MG |
2543 | { |
2544 | int high = READ_ONCE(pcp->high); | |
2545 | ||
f26b3fa0 | 2546 | if (unlikely(!high || free_high)) |
c49c2c47 MG |
2547 | return 0; |
2548 | ||
2549 | if (!test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags)) | |
2550 | return high; | |
2551 | ||
2552 | /* | |
2553 | * If reclaim is active, limit the number of pages that can be | |
2554 | * stored on pcp lists | |
2555 | */ | |
2556 | return min(READ_ONCE(pcp->batch) << 2, high); | |
2557 | } | |
2558 | ||
4b23a68f MG |
2559 | static void free_unref_page_commit(struct zone *zone, struct per_cpu_pages *pcp, |
2560 | struct page *page, int migratetype, | |
56651377 | 2561 | unsigned int order) |
9cca35d4 | 2562 | { |
3b12e7e9 | 2563 | int high; |
44042b44 | 2564 | int pindex; |
f26b3fa0 | 2565 | bool free_high; |
9cca35d4 | 2566 | |
15cd9004 | 2567 | __count_vm_events(PGFREE, 1 << order); |
44042b44 | 2568 | pindex = order_to_pindex(migratetype, order); |
bf75f200 | 2569 | list_add(&page->pcp_list, &pcp->lists[pindex]); |
44042b44 | 2570 | pcp->count += 1 << order; |
f26b3fa0 MG |
2571 | |
2572 | /* | |
2573 | * As high-order pages other than THP's stored on PCP can contribute | |
2574 | * to fragmentation, limit the number stored when PCP is heavily | |
2575 | * freeing without allocation. The remainder after bulk freeing | |
2576 | * stops will be drained from vmstat refresh context. | |
2577 | */ | |
2578 | free_high = (pcp->free_factor && order && order <= PAGE_ALLOC_COSTLY_ORDER); | |
2579 | ||
2580 | high = nr_pcp_high(pcp, zone, free_high); | |
3b12e7e9 MG |
2581 | if (pcp->count >= high) { |
2582 | int batch = READ_ONCE(pcp->batch); | |
2583 | ||
f26b3fa0 | 2584 | free_pcppages_bulk(zone, nr_pcp_free(pcp, high, batch, free_high), pcp, pindex); |
3b12e7e9 | 2585 | } |
9cca35d4 | 2586 | } |
5f8dcc21 | 2587 | |
9cca35d4 | 2588 | /* |
44042b44 | 2589 | * Free a pcp page |
9cca35d4 | 2590 | */ |
44042b44 | 2591 | void free_unref_page(struct page *page, unsigned int order) |
9cca35d4 | 2592 | { |
4b23a68f MG |
2593 | unsigned long __maybe_unused UP_flags; |
2594 | struct per_cpu_pages *pcp; | |
2595 | struct zone *zone; | |
9cca35d4 | 2596 | unsigned long pfn = page_to_pfn(page); |
df1acc85 | 2597 | int migratetype; |
9cca35d4 | 2598 | |
44042b44 | 2599 | if (!free_unref_page_prepare(page, pfn, order)) |
9cca35d4 | 2600 | return; |
da456f14 | 2601 | |
5f8dcc21 MG |
2602 | /* |
2603 | * We only track unmovable, reclaimable and movable on pcp lists. | |
df1acc85 | 2604 | * Place ISOLATE pages on the isolated list because they are being |
a6ffdc07 | 2605 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
2606 | * areas back if necessary. Otherwise, we may have to free |
2607 | * excessively into the page allocator | |
2608 | */ | |
df1acc85 MG |
2609 | migratetype = get_pcppage_migratetype(page); |
2610 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) { | |
194159fb | 2611 | if (unlikely(is_migrate_isolate(migratetype))) { |
44042b44 | 2612 | free_one_page(page_zone(page), page, pfn, order, migratetype, FPI_NONE); |
9cca35d4 | 2613 | return; |
5f8dcc21 MG |
2614 | } |
2615 | migratetype = MIGRATE_MOVABLE; | |
2616 | } | |
2617 | ||
4b23a68f MG |
2618 | zone = page_zone(page); |
2619 | pcp_trylock_prepare(UP_flags); | |
57490774 | 2620 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 2621 | if (pcp) { |
4b23a68f | 2622 | free_unref_page_commit(zone, pcp, page, migratetype, order); |
57490774 | 2623 | pcp_spin_unlock(pcp); |
4b23a68f MG |
2624 | } else { |
2625 | free_one_page(zone, page, pfn, order, migratetype, FPI_NONE); | |
2626 | } | |
2627 | pcp_trylock_finish(UP_flags); | |
1da177e4 LT |
2628 | } |
2629 | ||
cc59850e KK |
2630 | /* |
2631 | * Free a list of 0-order pages | |
2632 | */ | |
2d4894b5 | 2633 | void free_unref_page_list(struct list_head *list) |
cc59850e | 2634 | { |
57490774 | 2635 | unsigned long __maybe_unused UP_flags; |
cc59850e | 2636 | struct page *page, *next; |
4b23a68f MG |
2637 | struct per_cpu_pages *pcp = NULL; |
2638 | struct zone *locked_zone = NULL; | |
c24ad77d | 2639 | int batch_count = 0; |
df1acc85 | 2640 | int migratetype; |
9cca35d4 MG |
2641 | |
2642 | /* Prepare pages for freeing */ | |
2643 | list_for_each_entry_safe(page, next, list, lru) { | |
56651377 | 2644 | unsigned long pfn = page_to_pfn(page); |
053cfda1 | 2645 | if (!free_unref_page_prepare(page, pfn, 0)) { |
9cca35d4 | 2646 | list_del(&page->lru); |
053cfda1 ML |
2647 | continue; |
2648 | } | |
df1acc85 MG |
2649 | |
2650 | /* | |
2651 | * Free isolated pages directly to the allocator, see | |
2652 | * comment in free_unref_page. | |
2653 | */ | |
2654 | migratetype = get_pcppage_migratetype(page); | |
47aef601 DB |
2655 | if (unlikely(is_migrate_isolate(migratetype))) { |
2656 | list_del(&page->lru); | |
2657 | free_one_page(page_zone(page), page, pfn, 0, migratetype, FPI_NONE); | |
2658 | continue; | |
df1acc85 | 2659 | } |
9cca35d4 | 2660 | } |
cc59850e KK |
2661 | |
2662 | list_for_each_entry_safe(page, next, list, lru) { | |
4b23a68f MG |
2663 | struct zone *zone = page_zone(page); |
2664 | ||
c3e58a70 | 2665 | list_del(&page->lru); |
57490774 | 2666 | migratetype = get_pcppage_migratetype(page); |
c3e58a70 | 2667 | |
a4bafffb MG |
2668 | /* |
2669 | * Either different zone requiring a different pcp lock or | |
2670 | * excessive lock hold times when freeing a large list of | |
2671 | * pages. | |
2672 | */ | |
2673 | if (zone != locked_zone || batch_count == SWAP_CLUSTER_MAX) { | |
57490774 MG |
2674 | if (pcp) { |
2675 | pcp_spin_unlock(pcp); | |
2676 | pcp_trylock_finish(UP_flags); | |
2677 | } | |
01b44456 | 2678 | |
a4bafffb MG |
2679 | batch_count = 0; |
2680 | ||
57490774 MG |
2681 | /* |
2682 | * trylock is necessary as pages may be getting freed | |
2683 | * from IRQ or SoftIRQ context after an IO completion. | |
2684 | */ | |
2685 | pcp_trylock_prepare(UP_flags); | |
2686 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); | |
2687 | if (unlikely(!pcp)) { | |
2688 | pcp_trylock_finish(UP_flags); | |
2689 | free_one_page(zone, page, page_to_pfn(page), | |
2690 | 0, migratetype, FPI_NONE); | |
2691 | locked_zone = NULL; | |
2692 | continue; | |
2693 | } | |
4b23a68f | 2694 | locked_zone = zone; |
4b23a68f MG |
2695 | } |
2696 | ||
47aef601 DB |
2697 | /* |
2698 | * Non-isolated types over MIGRATE_PCPTYPES get added | |
2699 | * to the MIGRATE_MOVABLE pcp list. | |
2700 | */ | |
47aef601 DB |
2701 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) |
2702 | migratetype = MIGRATE_MOVABLE; | |
2703 | ||
2d4894b5 | 2704 | trace_mm_page_free_batched(page); |
4b23a68f | 2705 | free_unref_page_commit(zone, pcp, page, migratetype, 0); |
a4bafffb | 2706 | batch_count++; |
cc59850e | 2707 | } |
4b23a68f | 2708 | |
57490774 MG |
2709 | if (pcp) { |
2710 | pcp_spin_unlock(pcp); | |
2711 | pcp_trylock_finish(UP_flags); | |
2712 | } | |
cc59850e KK |
2713 | } |
2714 | ||
8dfcc9ba NP |
2715 | /* |
2716 | * split_page takes a non-compound higher-order page, and splits it into | |
2717 | * n (1<<order) sub-pages: page[0..n] | |
2718 | * Each sub-page must be freed individually. | |
2719 | * | |
2720 | * Note: this is probably too low level an operation for use in drivers. | |
2721 | * Please consult with lkml before using this in your driver. | |
2722 | */ | |
2723 | void split_page(struct page *page, unsigned int order) | |
2724 | { | |
2725 | int i; | |
2726 | ||
309381fe SL |
2727 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2728 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 2729 | |
a9627bc5 | 2730 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2731 | set_page_refcounted(page + i); |
8fb156c9 | 2732 | split_page_owner(page, 1 << order); |
e1baddf8 | 2733 | split_page_memcg(page, 1 << order); |
8dfcc9ba | 2734 | } |
5853ff23 | 2735 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2736 | |
3c605096 | 2737 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2738 | { |
9a157dd8 KW |
2739 | struct zone *zone = page_zone(page); |
2740 | int mt = get_pageblock_migratetype(page); | |
748446bb | 2741 | |
194159fb | 2742 | if (!is_migrate_isolate(mt)) { |
9a157dd8 | 2743 | unsigned long watermark; |
8348faf9 VB |
2744 | /* |
2745 | * Obey watermarks as if the page was being allocated. We can | |
2746 | * emulate a high-order watermark check with a raised order-0 | |
2747 | * watermark, because we already know our high-order page | |
2748 | * exists. | |
2749 | */ | |
fd1444b2 | 2750 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 2751 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
2752 | return 0; |
2753 | ||
8fb74b9f | 2754 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2755 | } |
748446bb | 2756 | |
6ab01363 | 2757 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 2758 | |
400bc7fd | 2759 | /* |
2760 | * Set the pageblock if the isolated page is at least half of a | |
2761 | * pageblock | |
2762 | */ | |
748446bb MG |
2763 | if (order >= pageblock_order - 1) { |
2764 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2765 | for (; page < endpage; page += pageblock_nr_pages) { |
2766 | int mt = get_pageblock_migratetype(page); | |
1dd214b8 ZY |
2767 | /* |
2768 | * Only change normal pageblocks (i.e., they can merge | |
2769 | * with others) | |
2770 | */ | |
2771 | if (migratetype_is_mergeable(mt)) | |
47118af0 MN |
2772 | set_pageblock_migratetype(page, |
2773 | MIGRATE_MOVABLE); | |
2774 | } | |
748446bb MG |
2775 | } |
2776 | ||
8fb74b9f | 2777 | return 1UL << order; |
1fb3f8ca MG |
2778 | } |
2779 | ||
624f58d8 AD |
2780 | /** |
2781 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
2782 | * @page: Page that was isolated | |
2783 | * @order: Order of the isolated page | |
e6a0a7ad | 2784 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
2785 | * |
2786 | * This function is meant to return a page pulled from the free lists via | |
2787 | * __isolate_free_page back to the free lists they were pulled from. | |
2788 | */ | |
2789 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
2790 | { | |
2791 | struct zone *zone = page_zone(page); | |
2792 | ||
2793 | /* zone lock should be held when this function is called */ | |
2794 | lockdep_assert_held(&zone->lock); | |
2795 | ||
2796 | /* Return isolated page to tail of freelist. */ | |
f04a5d5d | 2797 | __free_one_page(page, page_to_pfn(page), zone, order, mt, |
47b6a24a | 2798 | FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL); |
624f58d8 AD |
2799 | } |
2800 | ||
060e7417 MG |
2801 | /* |
2802 | * Update NUMA hit/miss statistics | |
060e7417 | 2803 | */ |
3e23060b MG |
2804 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, |
2805 | long nr_account) | |
060e7417 MG |
2806 | { |
2807 | #ifdef CONFIG_NUMA | |
3a321d2a | 2808 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 2809 | |
4518085e KW |
2810 | /* skip numa counters update if numa stats is disabled */ |
2811 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2812 | return; | |
2813 | ||
c1093b74 | 2814 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 2815 | local_stat = NUMA_OTHER; |
060e7417 | 2816 | |
c1093b74 | 2817 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3e23060b | 2818 | __count_numa_events(z, NUMA_HIT, nr_account); |
2df26639 | 2819 | else { |
3e23060b MG |
2820 | __count_numa_events(z, NUMA_MISS, nr_account); |
2821 | __count_numa_events(preferred_zone, NUMA_FOREIGN, nr_account); | |
060e7417 | 2822 | } |
3e23060b | 2823 | __count_numa_events(z, local_stat, nr_account); |
060e7417 MG |
2824 | #endif |
2825 | } | |
2826 | ||
589d9973 MG |
2827 | static __always_inline |
2828 | struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, | |
2829 | unsigned int order, unsigned int alloc_flags, | |
2830 | int migratetype) | |
2831 | { | |
2832 | struct page *page; | |
2833 | unsigned long flags; | |
2834 | ||
2835 | do { | |
2836 | page = NULL; | |
2837 | spin_lock_irqsave(&zone->lock, flags); | |
2838 | /* | |
2839 | * order-0 request can reach here when the pcplist is skipped | |
2840 | * due to non-CMA allocation context. HIGHATOMIC area is | |
2841 | * reserved for high-order atomic allocation, so order-0 | |
2842 | * request should skip it. | |
2843 | */ | |
eb2e2b42 | 2844 | if (alloc_flags & ALLOC_HIGHATOMIC) |
589d9973 MG |
2845 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
2846 | if (!page) { | |
2847 | page = __rmqueue(zone, order, migratetype, alloc_flags); | |
eb2e2b42 MG |
2848 | |
2849 | /* | |
2850 | * If the allocation fails, allow OOM handling access | |
2851 | * to HIGHATOMIC reserves as failing now is worse than | |
2852 | * failing a high-order atomic allocation in the | |
2853 | * future. | |
2854 | */ | |
2855 | if (!page && (alloc_flags & ALLOC_OOM)) | |
2856 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2857 | ||
589d9973 MG |
2858 | if (!page) { |
2859 | spin_unlock_irqrestore(&zone->lock, flags); | |
2860 | return NULL; | |
2861 | } | |
2862 | } | |
2863 | __mod_zone_freepage_state(zone, -(1 << order), | |
2864 | get_pcppage_migratetype(page)); | |
2865 | spin_unlock_irqrestore(&zone->lock, flags); | |
2866 | } while (check_new_pages(page, order)); | |
2867 | ||
2868 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
2869 | zone_statistics(preferred_zone, zone, 1); | |
2870 | ||
2871 | return page; | |
2872 | } | |
2873 | ||
066b2393 | 2874 | /* Remove page from the per-cpu list, caller must protect the list */ |
3b822017 | 2875 | static inline |
44042b44 MG |
2876 | struct page *__rmqueue_pcplist(struct zone *zone, unsigned int order, |
2877 | int migratetype, | |
6bb15450 | 2878 | unsigned int alloc_flags, |
453f85d4 | 2879 | struct per_cpu_pages *pcp, |
066b2393 MG |
2880 | struct list_head *list) |
2881 | { | |
2882 | struct page *page; | |
2883 | ||
2884 | do { | |
2885 | if (list_empty(list)) { | |
44042b44 MG |
2886 | int batch = READ_ONCE(pcp->batch); |
2887 | int alloced; | |
2888 | ||
2889 | /* | |
2890 | * Scale batch relative to order if batch implies | |
2891 | * free pages can be stored on the PCP. Batch can | |
2892 | * be 1 for small zones or for boot pagesets which | |
2893 | * should never store free pages as the pages may | |
2894 | * belong to arbitrary zones. | |
2895 | */ | |
2896 | if (batch > 1) | |
2897 | batch = max(batch >> order, 2); | |
2898 | alloced = rmqueue_bulk(zone, order, | |
2899 | batch, list, | |
6bb15450 | 2900 | migratetype, alloc_flags); |
44042b44 MG |
2901 | |
2902 | pcp->count += alloced << order; | |
066b2393 MG |
2903 | if (unlikely(list_empty(list))) |
2904 | return NULL; | |
2905 | } | |
2906 | ||
bf75f200 MG |
2907 | page = list_first_entry(list, struct page, pcp_list); |
2908 | list_del(&page->pcp_list); | |
44042b44 | 2909 | pcp->count -= 1 << order; |
700d2e9a | 2910 | } while (check_new_pages(page, order)); |
066b2393 MG |
2911 | |
2912 | return page; | |
2913 | } | |
2914 | ||
2915 | /* Lock and remove page from the per-cpu list */ | |
2916 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
44042b44 | 2917 | struct zone *zone, unsigned int order, |
663d0cfd | 2918 | int migratetype, unsigned int alloc_flags) |
066b2393 MG |
2919 | { |
2920 | struct per_cpu_pages *pcp; | |
2921 | struct list_head *list; | |
066b2393 | 2922 | struct page *page; |
4b23a68f | 2923 | unsigned long __maybe_unused UP_flags; |
066b2393 | 2924 | |
57490774 | 2925 | /* spin_trylock may fail due to a parallel drain or IRQ reentrancy. */ |
4b23a68f | 2926 | pcp_trylock_prepare(UP_flags); |
57490774 | 2927 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 2928 | if (!pcp) { |
4b23a68f | 2929 | pcp_trylock_finish(UP_flags); |
4b23a68f MG |
2930 | return NULL; |
2931 | } | |
3b12e7e9 MG |
2932 | |
2933 | /* | |
2934 | * On allocation, reduce the number of pages that are batch freed. | |
2935 | * See nr_pcp_free() where free_factor is increased for subsequent | |
2936 | * frees. | |
2937 | */ | |
3b12e7e9 | 2938 | pcp->free_factor >>= 1; |
44042b44 MG |
2939 | list = &pcp->lists[order_to_pindex(migratetype, order)]; |
2940 | page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list); | |
57490774 | 2941 | pcp_spin_unlock(pcp); |
4b23a68f | 2942 | pcp_trylock_finish(UP_flags); |
066b2393 | 2943 | if (page) { |
15cd9004 | 2944 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
3e23060b | 2945 | zone_statistics(preferred_zone, zone, 1); |
066b2393 | 2946 | } |
066b2393 MG |
2947 | return page; |
2948 | } | |
2949 | ||
1da177e4 | 2950 | /* |
a57ae9ef RX |
2951 | * Allocate a page from the given zone. |
2952 | * Use pcplists for THP or "cheap" high-order allocations. | |
1da177e4 | 2953 | */ |
b073d7f8 AP |
2954 | |
2955 | /* | |
2956 | * Do not instrument rmqueue() with KMSAN. This function may call | |
2957 | * __msan_poison_alloca() through a call to set_pfnblock_flags_mask(). | |
2958 | * If __msan_poison_alloca() attempts to allocate pages for the stack depot, it | |
2959 | * may call rmqueue() again, which will result in a deadlock. | |
1da177e4 | 2960 | */ |
b073d7f8 | 2961 | __no_sanitize_memory |
0a15c3e9 | 2962 | static inline |
066b2393 | 2963 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 2964 | struct zone *zone, unsigned int order, |
c603844b MG |
2965 | gfp_t gfp_flags, unsigned int alloc_flags, |
2966 | int migratetype) | |
1da177e4 | 2967 | { |
689bcebf | 2968 | struct page *page; |
1da177e4 | 2969 | |
589d9973 MG |
2970 | /* |
2971 | * We most definitely don't want callers attempting to | |
2972 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2973 | */ | |
2974 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
2975 | ||
44042b44 | 2976 | if (likely(pcp_allowed_order(order))) { |
1d91df85 JK |
2977 | /* |
2978 | * MIGRATE_MOVABLE pcplist could have the pages on CMA area and | |
2979 | * we need to skip it when CMA area isn't allowed. | |
2980 | */ | |
2981 | if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA || | |
2982 | migratetype != MIGRATE_MOVABLE) { | |
44042b44 | 2983 | page = rmqueue_pcplist(preferred_zone, zone, order, |
663d0cfd | 2984 | migratetype, alloc_flags); |
4b23a68f MG |
2985 | if (likely(page)) |
2986 | goto out; | |
1d91df85 | 2987 | } |
066b2393 | 2988 | } |
83b9355b | 2989 | |
589d9973 MG |
2990 | page = rmqueue_buddy(preferred_zone, zone, order, alloc_flags, |
2991 | migratetype); | |
1da177e4 | 2992 | |
066b2393 | 2993 | out: |
73444bc4 | 2994 | /* Separate test+clear to avoid unnecessary atomics */ |
e2a66c21 | 2995 | if (unlikely(test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags))) { |
73444bc4 MG |
2996 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
2997 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
2998 | } | |
2999 | ||
066b2393 | 3000 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 LT |
3001 | return page; |
3002 | } | |
3003 | ||
933e312e AM |
3004 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3005 | ||
b2588c4b | 3006 | static struct { |
933e312e AM |
3007 | struct fault_attr attr; |
3008 | ||
621a5f7a | 3009 | bool ignore_gfp_highmem; |
71baba4b | 3010 | bool ignore_gfp_reclaim; |
54114994 | 3011 | u32 min_order; |
933e312e AM |
3012 | } fail_page_alloc = { |
3013 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3014 | .ignore_gfp_reclaim = true, |
621a5f7a | 3015 | .ignore_gfp_highmem = true, |
54114994 | 3016 | .min_order = 1, |
933e312e AM |
3017 | }; |
3018 | ||
3019 | static int __init setup_fail_page_alloc(char *str) | |
3020 | { | |
3021 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3022 | } | |
3023 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3024 | ||
af3b8544 | 3025 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3026 | { |
ea4452de QZ |
3027 | int flags = 0; |
3028 | ||
54114994 | 3029 | if (order < fail_page_alloc.min_order) |
deaf386e | 3030 | return false; |
933e312e | 3031 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3032 | return false; |
933e312e | 3033 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3034 | return false; |
71baba4b MG |
3035 | if (fail_page_alloc.ignore_gfp_reclaim && |
3036 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3037 | return false; |
933e312e | 3038 | |
ea4452de | 3039 | /* See comment in __should_failslab() */ |
3f913fc5 | 3040 | if (gfp_mask & __GFP_NOWARN) |
ea4452de | 3041 | flags |= FAULT_NOWARN; |
3f913fc5 | 3042 | |
ea4452de | 3043 | return should_fail_ex(&fail_page_alloc.attr, 1 << order, flags); |
933e312e AM |
3044 | } |
3045 | ||
3046 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3047 | ||
3048 | static int __init fail_page_alloc_debugfs(void) | |
3049 | { | |
0825a6f9 | 3050 | umode_t mode = S_IFREG | 0600; |
933e312e | 3051 | struct dentry *dir; |
933e312e | 3052 | |
dd48c085 AM |
3053 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3054 | &fail_page_alloc.attr); | |
b2588c4b | 3055 | |
d9f7979c GKH |
3056 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3057 | &fail_page_alloc.ignore_gfp_reclaim); | |
3058 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3059 | &fail_page_alloc.ignore_gfp_highmem); | |
3060 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3061 | |
d9f7979c | 3062 | return 0; |
933e312e AM |
3063 | } |
3064 | ||
3065 | late_initcall(fail_page_alloc_debugfs); | |
3066 | ||
3067 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3068 | ||
3069 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3070 | ||
af3b8544 | 3071 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3072 | { |
deaf386e | 3073 | return false; |
933e312e AM |
3074 | } |
3075 | ||
3076 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3077 | ||
54aa3866 | 3078 | noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
af3b8544 BP |
3079 | { |
3080 | return __should_fail_alloc_page(gfp_mask, order); | |
3081 | } | |
3082 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3083 | ||
f27ce0e1 JK |
3084 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3085 | unsigned int order, unsigned int alloc_flags) | |
3086 | { | |
f27ce0e1 JK |
3087 | long unusable_free = (1 << order) - 1; |
3088 | ||
3089 | /* | |
ab350885 MG |
3090 | * If the caller does not have rights to reserves below the min |
3091 | * watermark then subtract the high-atomic reserves. This will | |
3092 | * over-estimate the size of the atomic reserve but it avoids a search. | |
f27ce0e1 | 3093 | */ |
ab350885 | 3094 | if (likely(!(alloc_flags & ALLOC_RESERVES))) |
f27ce0e1 JK |
3095 | unusable_free += z->nr_reserved_highatomic; |
3096 | ||
3097 | #ifdef CONFIG_CMA | |
3098 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3099 | if (!(alloc_flags & ALLOC_CMA)) | |
3100 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3101 | #endif | |
3102 | ||
3103 | return unusable_free; | |
3104 | } | |
3105 | ||
1da177e4 | 3106 | /* |
97a16fc8 MG |
3107 | * Return true if free base pages are above 'mark'. For high-order checks it |
3108 | * will return true of the order-0 watermark is reached and there is at least | |
3109 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3110 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3111 | */ |
86a294a8 | 3112 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3113 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3114 | long free_pages) |
1da177e4 | 3115 | { |
d23ad423 | 3116 | long min = mark; |
1da177e4 LT |
3117 | int o; |
3118 | ||
0aaa29a5 | 3119 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 3120 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 3121 | |
ab350885 MG |
3122 | if (unlikely(alloc_flags & ALLOC_RESERVES)) { |
3123 | /* | |
3124 | * __GFP_HIGH allows access to 50% of the min reserve as well | |
3125 | * as OOM. | |
3126 | */ | |
1ebbb218 | 3127 | if (alloc_flags & ALLOC_MIN_RESERVE) { |
ab350885 | 3128 | min -= min / 2; |
0aaa29a5 | 3129 | |
1ebbb218 MG |
3130 | /* |
3131 | * Non-blocking allocations (e.g. GFP_ATOMIC) can | |
3132 | * access more reserves than just __GFP_HIGH. Other | |
3133 | * non-blocking allocations requests such as GFP_NOWAIT | |
3134 | * or (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) do not get | |
3135 | * access to the min reserve. | |
3136 | */ | |
3137 | if (alloc_flags & ALLOC_NON_BLOCK) | |
3138 | min -= min / 4; | |
3139 | } | |
0aaa29a5 | 3140 | |
cd04ae1e | 3141 | /* |
ab350885 | 3142 | * OOM victims can try even harder than the normal reserve |
cd04ae1e MH |
3143 | * users on the grounds that it's definitely going to be in |
3144 | * the exit path shortly and free memory. Any allocation it | |
3145 | * makes during the free path will be small and short-lived. | |
3146 | */ | |
3147 | if (alloc_flags & ALLOC_OOM) | |
3148 | min -= min / 2; | |
cd04ae1e MH |
3149 | } |
3150 | ||
97a16fc8 MG |
3151 | /* |
3152 | * Check watermarks for an order-0 allocation request. If these | |
3153 | * are not met, then a high-order request also cannot go ahead | |
3154 | * even if a suitable page happened to be free. | |
3155 | */ | |
97a225e6 | 3156 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 3157 | return false; |
1da177e4 | 3158 | |
97a16fc8 MG |
3159 | /* If this is an order-0 request then the watermark is fine */ |
3160 | if (!order) | |
3161 | return true; | |
3162 | ||
3163 | /* For a high-order request, check at least one suitable page is free */ | |
23baf831 | 3164 | for (o = order; o <= MAX_ORDER; o++) { |
97a16fc8 MG |
3165 | struct free_area *area = &z->free_area[o]; |
3166 | int mt; | |
3167 | ||
3168 | if (!area->nr_free) | |
3169 | continue; | |
3170 | ||
97a16fc8 | 3171 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3172 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3173 | return true; |
3174 | } | |
3175 | ||
3176 | #ifdef CONFIG_CMA | |
d883c6cf | 3177 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3178 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3179 | return true; |
d883c6cf | 3180 | } |
97a16fc8 | 3181 | #endif |
eb2e2b42 MG |
3182 | if ((alloc_flags & (ALLOC_HIGHATOMIC|ALLOC_OOM)) && |
3183 | !free_area_empty(area, MIGRATE_HIGHATOMIC)) { | |
b050e376 | 3184 | return true; |
eb2e2b42 | 3185 | } |
1da177e4 | 3186 | } |
97a16fc8 | 3187 | return false; |
88f5acf8 MG |
3188 | } |
3189 | ||
7aeb09f9 | 3190 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3191 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 3192 | { |
97a225e6 | 3193 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
3194 | zone_page_state(z, NR_FREE_PAGES)); |
3195 | } | |
3196 | ||
48ee5f36 | 3197 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 3198 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 3199 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 3200 | { |
f27ce0e1 | 3201 | long free_pages; |
d883c6cf | 3202 | |
f27ce0e1 | 3203 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
3204 | |
3205 | /* | |
3206 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 3207 | * need to be calculated. |
48ee5f36 | 3208 | */ |
f27ce0e1 | 3209 | if (!order) { |
9282012f JK |
3210 | long usable_free; |
3211 | long reserved; | |
f27ce0e1 | 3212 | |
9282012f JK |
3213 | usable_free = free_pages; |
3214 | reserved = __zone_watermark_unusable_free(z, 0, alloc_flags); | |
3215 | ||
3216 | /* reserved may over estimate high-atomic reserves. */ | |
3217 | usable_free -= min(usable_free, reserved); | |
3218 | if (usable_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
f27ce0e1 JK |
3219 | return true; |
3220 | } | |
48ee5f36 | 3221 | |
f80b08fc CTR |
3222 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
3223 | free_pages)) | |
3224 | return true; | |
2973d822 | 3225 | |
f80b08fc | 3226 | /* |
2973d822 | 3227 | * Ignore watermark boosting for __GFP_HIGH order-0 allocations |
f80b08fc CTR |
3228 | * when checking the min watermark. The min watermark is the |
3229 | * point where boosting is ignored so that kswapd is woken up | |
3230 | * when below the low watermark. | |
3231 | */ | |
2973d822 | 3232 | if (unlikely(!order && (alloc_flags & ALLOC_MIN_RESERVE) && z->watermark_boost |
f80b08fc CTR |
3233 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { |
3234 | mark = z->_watermark[WMARK_MIN]; | |
3235 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
3236 | alloc_flags, free_pages); | |
3237 | } | |
3238 | ||
3239 | return false; | |
48ee5f36 MG |
3240 | } |
3241 | ||
7aeb09f9 | 3242 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 3243 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
3244 | { |
3245 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3246 | ||
3247 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3248 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3249 | ||
97a225e6 | 3250 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 3251 | free_pages); |
1da177e4 LT |
3252 | } |
3253 | ||
9276b1bc | 3254 | #ifdef CONFIG_NUMA |
61bb6cd2 GU |
3255 | int __read_mostly node_reclaim_distance = RECLAIM_DISTANCE; |
3256 | ||
957f822a DR |
3257 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3258 | { | |
e02dc017 | 3259 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3260 | node_reclaim_distance; |
957f822a | 3261 | } |
9276b1bc | 3262 | #else /* CONFIG_NUMA */ |
957f822a DR |
3263 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3264 | { | |
3265 | return true; | |
3266 | } | |
9276b1bc PJ |
3267 | #endif /* CONFIG_NUMA */ |
3268 | ||
6bb15450 MG |
3269 | /* |
3270 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3271 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3272 | * premature use of a lower zone may cause lowmem pressure problems that | |
3273 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3274 | * probably too small. It only makes sense to spread allocations to avoid | |
3275 | * fragmentation between the Normal and DMA32 zones. | |
3276 | */ | |
3277 | static inline unsigned int | |
0a79cdad | 3278 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3279 | { |
736838e9 | 3280 | unsigned int alloc_flags; |
0a79cdad | 3281 | |
736838e9 MN |
3282 | /* |
3283 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
3284 | * to save a branch. | |
3285 | */ | |
3286 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
3287 | |
3288 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3289 | if (!zone) |
3290 | return alloc_flags; | |
3291 | ||
6bb15450 | 3292 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3293 | return alloc_flags; |
6bb15450 MG |
3294 | |
3295 | /* | |
3296 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3297 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3298 | * on UMA that if Normal is populated then so is DMA32. | |
3299 | */ | |
3300 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3301 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3302 | return alloc_flags; |
6bb15450 | 3303 | |
8118b82e | 3304 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3305 | #endif /* CONFIG_ZONE_DMA32 */ |
3306 | return alloc_flags; | |
6bb15450 | 3307 | } |
6bb15450 | 3308 | |
8e3560d9 PT |
3309 | /* Must be called after current_gfp_context() which can change gfp_mask */ |
3310 | static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask, | |
3311 | unsigned int alloc_flags) | |
8510e69c JK |
3312 | { |
3313 | #ifdef CONFIG_CMA | |
8e3560d9 | 3314 | if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
8510e69c | 3315 | alloc_flags |= ALLOC_CMA; |
8510e69c JK |
3316 | #endif |
3317 | return alloc_flags; | |
3318 | } | |
3319 | ||
7fb1d9fc | 3320 | /* |
0798e519 | 3321 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3322 | * a page. |
3323 | */ | |
3324 | static struct page * | |
a9263751 VB |
3325 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3326 | const struct alloc_context *ac) | |
753ee728 | 3327 | { |
6bb15450 | 3328 | struct zoneref *z; |
5117f45d | 3329 | struct zone *zone; |
8a87d695 WY |
3330 | struct pglist_data *last_pgdat = NULL; |
3331 | bool last_pgdat_dirty_ok = false; | |
6bb15450 | 3332 | bool no_fallback; |
3b8c0be4 | 3333 | |
6bb15450 | 3334 | retry: |
7fb1d9fc | 3335 | /* |
9276b1bc | 3336 | * Scan zonelist, looking for a zone with enough free. |
8e464522 | 3337 | * See also cpuset_node_allowed() comment in kernel/cgroup/cpuset.c. |
7fb1d9fc | 3338 | */ |
6bb15450 MG |
3339 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3340 | z = ac->preferred_zoneref; | |
30d8ec73 MN |
3341 | for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx, |
3342 | ac->nodemask) { | |
be06af00 | 3343 | struct page *page; |
e085dbc5 JW |
3344 | unsigned long mark; |
3345 | ||
664eedde MG |
3346 | if (cpusets_enabled() && |
3347 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3348 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3349 | continue; |
a756cf59 JW |
3350 | /* |
3351 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3352 | * want to get it from a node that is within its dirty |
3353 | * limit, such that no single node holds more than its | |
a756cf59 | 3354 | * proportional share of globally allowed dirty pages. |
281e3726 | 3355 | * The dirty limits take into account the node's |
a756cf59 JW |
3356 | * lowmem reserves and high watermark so that kswapd |
3357 | * should be able to balance it without having to | |
3358 | * write pages from its LRU list. | |
3359 | * | |
a756cf59 | 3360 | * XXX: For now, allow allocations to potentially |
281e3726 | 3361 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3362 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3363 | * which is important when on a NUMA setup the allowed |
281e3726 | 3364 | * nodes are together not big enough to reach the |
a756cf59 | 3365 | * global limit. The proper fix for these situations |
281e3726 | 3366 | * will require awareness of nodes in the |
a756cf59 JW |
3367 | * dirty-throttling and the flusher threads. |
3368 | */ | |
3b8c0be4 | 3369 | if (ac->spread_dirty_pages) { |
8a87d695 WY |
3370 | if (last_pgdat != zone->zone_pgdat) { |
3371 | last_pgdat = zone->zone_pgdat; | |
3372 | last_pgdat_dirty_ok = node_dirty_ok(zone->zone_pgdat); | |
3373 | } | |
3b8c0be4 | 3374 | |
8a87d695 | 3375 | if (!last_pgdat_dirty_ok) |
3b8c0be4 | 3376 | continue; |
3b8c0be4 | 3377 | } |
7fb1d9fc | 3378 | |
6bb15450 MG |
3379 | if (no_fallback && nr_online_nodes > 1 && |
3380 | zone != ac->preferred_zoneref->zone) { | |
3381 | int local_nid; | |
3382 | ||
3383 | /* | |
3384 | * If moving to a remote node, retry but allow | |
3385 | * fragmenting fallbacks. Locality is more important | |
3386 | * than fragmentation avoidance. | |
3387 | */ | |
3388 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3389 | if (zone_to_nid(zone) != local_nid) { | |
3390 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3391 | goto retry; | |
3392 | } | |
3393 | } | |
3394 | ||
a9214443 | 3395 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 3396 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
3397 | ac->highest_zoneidx, alloc_flags, |
3398 | gfp_mask)) { | |
fa5e084e MG |
3399 | int ret; |
3400 | ||
c9e97a19 PT |
3401 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3402 | /* | |
3403 | * Watermark failed for this zone, but see if we can | |
3404 | * grow this zone if it contains deferred pages. | |
3405 | */ | |
076cf7ea | 3406 | if (deferred_pages_enabled()) { |
c9e97a19 PT |
3407 | if (_deferred_grow_zone(zone, order)) |
3408 | goto try_this_zone; | |
3409 | } | |
3410 | #endif | |
5dab2911 MG |
3411 | /* Checked here to keep the fast path fast */ |
3412 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3413 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3414 | goto try_this_zone; | |
3415 | ||
202e35db | 3416 | if (!node_reclaim_enabled() || |
c33d6c06 | 3417 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3418 | continue; |
3419 | ||
a5f5f91d | 3420 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3421 | switch (ret) { |
a5f5f91d | 3422 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3423 | /* did not scan */ |
cd38b115 | 3424 | continue; |
a5f5f91d | 3425 | case NODE_RECLAIM_FULL: |
fa5e084e | 3426 | /* scanned but unreclaimable */ |
cd38b115 | 3427 | continue; |
fa5e084e MG |
3428 | default: |
3429 | /* did we reclaim enough */ | |
fed2719e | 3430 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 3431 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
3432 | goto try_this_zone; |
3433 | ||
fed2719e | 3434 | continue; |
0798e519 | 3435 | } |
7fb1d9fc RS |
3436 | } |
3437 | ||
fa5e084e | 3438 | try_this_zone: |
066b2393 | 3439 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3440 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3441 | if (page) { |
479f854a | 3442 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3443 | |
3444 | /* | |
3445 | * If this is a high-order atomic allocation then check | |
3446 | * if the pageblock should be reserved for the future | |
3447 | */ | |
eb2e2b42 | 3448 | if (unlikely(alloc_flags & ALLOC_HIGHATOMIC)) |
0aaa29a5 MG |
3449 | reserve_highatomic_pageblock(page, zone, order); |
3450 | ||
75379191 | 3451 | return page; |
c9e97a19 PT |
3452 | } else { |
3453 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
3454 | /* Try again if zone has deferred pages */ | |
076cf7ea | 3455 | if (deferred_pages_enabled()) { |
c9e97a19 PT |
3456 | if (_deferred_grow_zone(zone, order)) |
3457 | goto try_this_zone; | |
3458 | } | |
3459 | #endif | |
75379191 | 3460 | } |
54a6eb5c | 3461 | } |
9276b1bc | 3462 | |
6bb15450 MG |
3463 | /* |
3464 | * It's possible on a UMA machine to get through all zones that are | |
3465 | * fragmented. If avoiding fragmentation, reset and try again. | |
3466 | */ | |
3467 | if (no_fallback) { | |
3468 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3469 | goto retry; | |
3470 | } | |
3471 | ||
4ffeaf35 | 3472 | return NULL; |
753ee728 MH |
3473 | } |
3474 | ||
9af744d7 | 3475 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3476 | { |
a238ab5b | 3477 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
3478 | |
3479 | /* | |
3480 | * This documents exceptions given to allocations in certain | |
3481 | * contexts that are allowed to allocate outside current's set | |
3482 | * of allowed nodes. | |
3483 | */ | |
3484 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3485 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3486 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3487 | filter &= ~SHOW_MEM_FILTER_NODES; | |
88dc6f20 | 3488 | if (!in_task() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3489 | filter &= ~SHOW_MEM_FILTER_NODES; |
3490 | ||
974f4367 | 3491 | __show_mem(filter, nodemask, gfp_zone(gfp_mask)); |
aa187507 MH |
3492 | } |
3493 | ||
a8e99259 | 3494 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3495 | { |
3496 | struct va_format vaf; | |
3497 | va_list args; | |
1be334e5 | 3498 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 3499 | |
c4dc63f0 BH |
3500 | if ((gfp_mask & __GFP_NOWARN) || |
3501 | !__ratelimit(&nopage_rs) || | |
3502 | ((gfp_mask & __GFP_DMA) && !has_managed_dma())) | |
aa187507 MH |
3503 | return; |
3504 | ||
7877cdcc MH |
3505 | va_start(args, fmt); |
3506 | vaf.fmt = fmt; | |
3507 | vaf.va = &args; | |
ef8444ea | 3508 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
3509 | current->comm, &vaf, gfp_mask, &gfp_mask, |
3510 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3511 | va_end(args); |
3ee9a4f0 | 3512 | |
a8e99259 | 3513 | cpuset_print_current_mems_allowed(); |
ef8444ea | 3514 | pr_cont("\n"); |
a238ab5b | 3515 | dump_stack(); |
685dbf6f | 3516 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3517 | } |
3518 | ||
6c18ba7a MH |
3519 | static inline struct page * |
3520 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3521 | unsigned int alloc_flags, | |
3522 | const struct alloc_context *ac) | |
3523 | { | |
3524 | struct page *page; | |
3525 | ||
3526 | page = get_page_from_freelist(gfp_mask, order, | |
3527 | alloc_flags|ALLOC_CPUSET, ac); | |
3528 | /* | |
3529 | * fallback to ignore cpuset restriction if our nodes | |
3530 | * are depleted | |
3531 | */ | |
3532 | if (!page) | |
3533 | page = get_page_from_freelist(gfp_mask, order, | |
3534 | alloc_flags, ac); | |
3535 | ||
3536 | return page; | |
3537 | } | |
3538 | ||
11e33f6a MG |
3539 | static inline struct page * |
3540 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3541 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3542 | { |
6e0fc46d DR |
3543 | struct oom_control oc = { |
3544 | .zonelist = ac->zonelist, | |
3545 | .nodemask = ac->nodemask, | |
2a966b77 | 3546 | .memcg = NULL, |
6e0fc46d DR |
3547 | .gfp_mask = gfp_mask, |
3548 | .order = order, | |
6e0fc46d | 3549 | }; |
11e33f6a MG |
3550 | struct page *page; |
3551 | ||
9879de73 JW |
3552 | *did_some_progress = 0; |
3553 | ||
9879de73 | 3554 | /* |
dc56401f JW |
3555 | * Acquire the oom lock. If that fails, somebody else is |
3556 | * making progress for us. | |
9879de73 | 3557 | */ |
dc56401f | 3558 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3559 | *did_some_progress = 1; |
11e33f6a | 3560 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3561 | return NULL; |
3562 | } | |
6b1de916 | 3563 | |
11e33f6a MG |
3564 | /* |
3565 | * Go through the zonelist yet one more time, keep very high watermark | |
3566 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3567 | * we're still under heavy pressure. But make sure that this reclaim |
3568 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3569 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3570 | */ |
e746bf73 TH |
3571 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3572 | ~__GFP_DIRECT_RECLAIM, order, | |
3573 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3574 | if (page) |
11e33f6a MG |
3575 | goto out; |
3576 | ||
06ad276a MH |
3577 | /* Coredumps can quickly deplete all memory reserves */ |
3578 | if (current->flags & PF_DUMPCORE) | |
3579 | goto out; | |
3580 | /* The OOM killer will not help higher order allocs */ | |
3581 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3582 | goto out; | |
dcda9b04 MH |
3583 | /* |
3584 | * We have already exhausted all our reclaim opportunities without any | |
3585 | * success so it is time to admit defeat. We will skip the OOM killer | |
3586 | * because it is very likely that the caller has a more reasonable | |
3587 | * fallback than shooting a random task. | |
cfb4a541 MN |
3588 | * |
3589 | * The OOM killer may not free memory on a specific node. | |
dcda9b04 | 3590 | */ |
cfb4a541 | 3591 | if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE)) |
dcda9b04 | 3592 | goto out; |
06ad276a | 3593 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 3594 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
3595 | goto out; |
3596 | if (pm_suspended_storage()) | |
3597 | goto out; | |
3598 | /* | |
3599 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3600 | * other request to make a forward progress. | |
3601 | * We are in an unfortunate situation where out_of_memory cannot | |
3602 | * do much for this context but let's try it to at least get | |
3603 | * access to memory reserved if the current task is killed (see | |
3604 | * out_of_memory). Once filesystems are ready to handle allocation | |
3605 | * failures more gracefully we should just bail out here. | |
3606 | */ | |
3607 | ||
3c2c6488 | 3608 | /* Exhausted what can be done so it's blame time */ |
3f913fc5 QZ |
3609 | if (out_of_memory(&oc) || |
3610 | WARN_ON_ONCE_GFP(gfp_mask & __GFP_NOFAIL, gfp_mask)) { | |
c32b3cbe | 3611 | *did_some_progress = 1; |
5020e285 | 3612 | |
6c18ba7a MH |
3613 | /* |
3614 | * Help non-failing allocations by giving them access to memory | |
3615 | * reserves | |
3616 | */ | |
3617 | if (gfp_mask & __GFP_NOFAIL) | |
3618 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3619 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3620 | } |
11e33f6a | 3621 | out: |
dc56401f | 3622 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3623 | return page; |
3624 | } | |
3625 | ||
33c2d214 | 3626 | /* |
baf2f90b | 3627 | * Maximum number of compaction retries with a progress before OOM |
33c2d214 MH |
3628 | * killer is consider as the only way to move forward. |
3629 | */ | |
3630 | #define MAX_COMPACT_RETRIES 16 | |
3631 | ||
56de7263 MG |
3632 | #ifdef CONFIG_COMPACTION |
3633 | /* Try memory compaction for high-order allocations before reclaim */ | |
3634 | static struct page * | |
3635 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3636 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3637 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3638 | { |
5e1f0f09 | 3639 | struct page *page = NULL; |
eb414681 | 3640 | unsigned long pflags; |
499118e9 | 3641 | unsigned int noreclaim_flag; |
53853e2d VB |
3642 | |
3643 | if (!order) | |
66199712 | 3644 | return NULL; |
66199712 | 3645 | |
eb414681 | 3646 | psi_memstall_enter(&pflags); |
5bf18281 | 3647 | delayacct_compact_start(); |
499118e9 | 3648 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 3649 | |
c5d01d0d | 3650 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 3651 | prio, &page); |
eb414681 | 3652 | |
499118e9 | 3653 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 3654 | psi_memstall_leave(&pflags); |
5bf18281 | 3655 | delayacct_compact_end(); |
56de7263 | 3656 | |
06dac2f4 CTR |
3657 | if (*compact_result == COMPACT_SKIPPED) |
3658 | return NULL; | |
98dd3b48 VB |
3659 | /* |
3660 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3661 | * count a compaction stall | |
3662 | */ | |
3663 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3664 | |
5e1f0f09 MG |
3665 | /* Prep a captured page if available */ |
3666 | if (page) | |
3667 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
3668 | ||
3669 | /* Try get a page from the freelist if available */ | |
3670 | if (!page) | |
3671 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 3672 | |
98dd3b48 VB |
3673 | if (page) { |
3674 | struct zone *zone = page_zone(page); | |
53853e2d | 3675 | |
98dd3b48 VB |
3676 | zone->compact_blockskip_flush = false; |
3677 | compaction_defer_reset(zone, order, true); | |
3678 | count_vm_event(COMPACTSUCCESS); | |
3679 | return page; | |
3680 | } | |
56de7263 | 3681 | |
98dd3b48 VB |
3682 | /* |
3683 | * It's bad if compaction run occurs and fails. The most likely reason | |
3684 | * is that pages exist, but not enough to satisfy watermarks. | |
3685 | */ | |
3686 | count_vm_event(COMPACTFAIL); | |
66199712 | 3687 | |
98dd3b48 | 3688 | cond_resched(); |
56de7263 MG |
3689 | |
3690 | return NULL; | |
3691 | } | |
33c2d214 | 3692 | |
3250845d VB |
3693 | static inline bool |
3694 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3695 | enum compact_result compact_result, | |
3696 | enum compact_priority *compact_priority, | |
d9436498 | 3697 | int *compaction_retries) |
3250845d VB |
3698 | { |
3699 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3700 | int min_priority; |
65190cff MH |
3701 | bool ret = false; |
3702 | int retries = *compaction_retries; | |
3703 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3704 | |
3705 | if (!order) | |
3706 | return false; | |
3707 | ||
691d9497 AT |
3708 | if (fatal_signal_pending(current)) |
3709 | return false; | |
3710 | ||
d9436498 VB |
3711 | if (compaction_made_progress(compact_result)) |
3712 | (*compaction_retries)++; | |
3713 | ||
3250845d VB |
3714 | /* |
3715 | * compaction considers all the zone as desperately out of memory | |
3716 | * so it doesn't really make much sense to retry except when the | |
3717 | * failure could be caused by insufficient priority | |
3718 | */ | |
d9436498 VB |
3719 | if (compaction_failed(compact_result)) |
3720 | goto check_priority; | |
3250845d | 3721 | |
49433085 VB |
3722 | /* |
3723 | * compaction was skipped because there are not enough order-0 pages | |
3724 | * to work with, so we retry only if it looks like reclaim can help. | |
3725 | */ | |
3726 | if (compaction_needs_reclaim(compact_result)) { | |
3727 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
3728 | goto out; | |
3729 | } | |
3730 | ||
3250845d VB |
3731 | /* |
3732 | * make sure the compaction wasn't deferred or didn't bail out early | |
3733 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
3734 | * But the next retry should use a higher priority if allowed, so |
3735 | * we don't just keep bailing out endlessly. | |
3250845d | 3736 | */ |
65190cff | 3737 | if (compaction_withdrawn(compact_result)) { |
49433085 | 3738 | goto check_priority; |
65190cff | 3739 | } |
3250845d VB |
3740 | |
3741 | /* | |
dcda9b04 | 3742 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
3743 | * costly ones because they are de facto nofail and invoke OOM |
3744 | * killer to move on while costly can fail and users are ready | |
3745 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3746 | * would need much more detailed feedback from compaction to | |
3747 | * make a better decision. | |
3748 | */ | |
3749 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3750 | max_retries /= 4; | |
65190cff MH |
3751 | if (*compaction_retries <= max_retries) { |
3752 | ret = true; | |
3753 | goto out; | |
3754 | } | |
3250845d | 3755 | |
d9436498 VB |
3756 | /* |
3757 | * Make sure there are attempts at the highest priority if we exhausted | |
3758 | * all retries or failed at the lower priorities. | |
3759 | */ | |
3760 | check_priority: | |
c2033b00 VB |
3761 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3762 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3763 | |
c2033b00 | 3764 | if (*compact_priority > min_priority) { |
d9436498 VB |
3765 | (*compact_priority)--; |
3766 | *compaction_retries = 0; | |
65190cff | 3767 | ret = true; |
d9436498 | 3768 | } |
65190cff MH |
3769 | out: |
3770 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
3771 | return ret; | |
3250845d | 3772 | } |
56de7263 MG |
3773 | #else |
3774 | static inline struct page * | |
3775 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3776 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3777 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3778 | { |
33c2d214 | 3779 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3780 | return NULL; |
3781 | } | |
33c2d214 MH |
3782 | |
3783 | static inline bool | |
86a294a8 MH |
3784 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3785 | enum compact_result compact_result, | |
a5508cd8 | 3786 | enum compact_priority *compact_priority, |
d9436498 | 3787 | int *compaction_retries) |
33c2d214 | 3788 | { |
31e49bfd MH |
3789 | struct zone *zone; |
3790 | struct zoneref *z; | |
3791 | ||
3792 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3793 | return false; | |
3794 | ||
3795 | /* | |
3796 | * There are setups with compaction disabled which would prefer to loop | |
3797 | * inside the allocator rather than hit the oom killer prematurely. | |
3798 | * Let's give them a good hope and keep retrying while the order-0 | |
3799 | * watermarks are OK. | |
3800 | */ | |
97a225e6 JK |
3801 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3802 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 3803 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 3804 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
3805 | return true; |
3806 | } | |
33c2d214 MH |
3807 | return false; |
3808 | } | |
3250845d | 3809 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3810 | |
d92a8cfc | 3811 | #ifdef CONFIG_LOCKDEP |
93781325 | 3812 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
3813 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
3814 | ||
f920e413 | 3815 | static bool __need_reclaim(gfp_t gfp_mask) |
d92a8cfc | 3816 | { |
d92a8cfc PZ |
3817 | /* no reclaim without waiting on it */ |
3818 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
3819 | return false; | |
3820 | ||
3821 | /* this guy won't enter reclaim */ | |
2e517d68 | 3822 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
3823 | return false; |
3824 | ||
d92a8cfc PZ |
3825 | if (gfp_mask & __GFP_NOLOCKDEP) |
3826 | return false; | |
3827 | ||
3828 | return true; | |
3829 | } | |
3830 | ||
4f3eaf45 | 3831 | void __fs_reclaim_acquire(unsigned long ip) |
93781325 | 3832 | { |
4f3eaf45 | 3833 | lock_acquire_exclusive(&__fs_reclaim_map, 0, 0, NULL, ip); |
93781325 OS |
3834 | } |
3835 | ||
4f3eaf45 | 3836 | void __fs_reclaim_release(unsigned long ip) |
93781325 | 3837 | { |
4f3eaf45 | 3838 | lock_release(&__fs_reclaim_map, ip); |
93781325 OS |
3839 | } |
3840 | ||
d92a8cfc PZ |
3841 | void fs_reclaim_acquire(gfp_t gfp_mask) |
3842 | { | |
f920e413 DV |
3843 | gfp_mask = current_gfp_context(gfp_mask); |
3844 | ||
3845 | if (__need_reclaim(gfp_mask)) { | |
3846 | if (gfp_mask & __GFP_FS) | |
4f3eaf45 | 3847 | __fs_reclaim_acquire(_RET_IP_); |
f920e413 DV |
3848 | |
3849 | #ifdef CONFIG_MMU_NOTIFIER | |
3850 | lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); | |
3851 | lock_map_release(&__mmu_notifier_invalidate_range_start_map); | |
3852 | #endif | |
3853 | ||
3854 | } | |
d92a8cfc PZ |
3855 | } |
3856 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
3857 | ||
3858 | void fs_reclaim_release(gfp_t gfp_mask) | |
3859 | { | |
f920e413 DV |
3860 | gfp_mask = current_gfp_context(gfp_mask); |
3861 | ||
3862 | if (__need_reclaim(gfp_mask)) { | |
3863 | if (gfp_mask & __GFP_FS) | |
4f3eaf45 | 3864 | __fs_reclaim_release(_RET_IP_); |
f920e413 | 3865 | } |
d92a8cfc PZ |
3866 | } |
3867 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
3868 | #endif | |
3869 | ||
3d36424b MG |
3870 | /* |
3871 | * Zonelists may change due to hotplug during allocation. Detect when zonelists | |
3872 | * have been rebuilt so allocation retries. Reader side does not lock and | |
3873 | * retries the allocation if zonelist changes. Writer side is protected by the | |
3874 | * embedded spin_lock. | |
3875 | */ | |
3876 | static DEFINE_SEQLOCK(zonelist_update_seq); | |
3877 | ||
3878 | static unsigned int zonelist_iter_begin(void) | |
3879 | { | |
3880 | if (IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) | |
3881 | return read_seqbegin(&zonelist_update_seq); | |
3882 | ||
3883 | return 0; | |
3884 | } | |
3885 | ||
3886 | static unsigned int check_retry_zonelist(unsigned int seq) | |
3887 | { | |
3888 | if (IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) | |
3889 | return read_seqretry(&zonelist_update_seq, seq); | |
3890 | ||
3891 | return seq; | |
3892 | } | |
3893 | ||
bba90710 | 3894 | /* Perform direct synchronous page reclaim */ |
2187e17b | 3895 | static unsigned long |
a9263751 VB |
3896 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3897 | const struct alloc_context *ac) | |
11e33f6a | 3898 | { |
499118e9 | 3899 | unsigned int noreclaim_flag; |
fa7fc75f | 3900 | unsigned long progress; |
11e33f6a MG |
3901 | |
3902 | cond_resched(); | |
3903 | ||
3904 | /* We now go into synchronous reclaim */ | |
3905 | cpuset_memory_pressure_bump(); | |
d92a8cfc | 3906 | fs_reclaim_acquire(gfp_mask); |
93781325 | 3907 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 3908 | |
a9263751 VB |
3909 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3910 | ac->nodemask); | |
11e33f6a | 3911 | |
499118e9 | 3912 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 3913 | fs_reclaim_release(gfp_mask); |
11e33f6a MG |
3914 | |
3915 | cond_resched(); | |
3916 | ||
bba90710 MS |
3917 | return progress; |
3918 | } | |
3919 | ||
3920 | /* The really slow allocator path where we enter direct reclaim */ | |
3921 | static inline struct page * | |
3922 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3923 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3924 | unsigned long *did_some_progress) |
bba90710 MS |
3925 | { |
3926 | struct page *page = NULL; | |
fa7fc75f | 3927 | unsigned long pflags; |
bba90710 MS |
3928 | bool drained = false; |
3929 | ||
fa7fc75f | 3930 | psi_memstall_enter(&pflags); |
a9263751 | 3931 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce | 3932 | if (unlikely(!(*did_some_progress))) |
fa7fc75f | 3933 | goto out; |
11e33f6a | 3934 | |
9ee493ce | 3935 | retry: |
31a6c190 | 3936 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3937 | |
3938 | /* | |
3939 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 | 3940 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
047b9967 | 3941 | * Shrink them and try again |
9ee493ce MG |
3942 | */ |
3943 | if (!page && !drained) { | |
29fac03b | 3944 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3945 | drain_all_pages(NULL); |
9ee493ce MG |
3946 | drained = true; |
3947 | goto retry; | |
3948 | } | |
fa7fc75f SB |
3949 | out: |
3950 | psi_memstall_leave(&pflags); | |
9ee493ce | 3951 | |
11e33f6a MG |
3952 | return page; |
3953 | } | |
3954 | ||
5ecd9d40 DR |
3955 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
3956 | const struct alloc_context *ac) | |
3a025760 JW |
3957 | { |
3958 | struct zoneref *z; | |
3959 | struct zone *zone; | |
e1a55637 | 3960 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 3961 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 3962 | |
97a225e6 | 3963 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 3964 | ac->nodemask) { |
bc53008e WY |
3965 | if (!managed_zone(zone)) |
3966 | continue; | |
d137a7cb | 3967 | if (last_pgdat != zone->zone_pgdat) { |
97a225e6 | 3968 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
d137a7cb CW |
3969 | last_pgdat = zone->zone_pgdat; |
3970 | } | |
e1a55637 | 3971 | } |
3a025760 JW |
3972 | } |
3973 | ||
c603844b | 3974 | static inline unsigned int |
eb2e2b42 | 3975 | gfp_to_alloc_flags(gfp_t gfp_mask, unsigned int order) |
341ce06f | 3976 | { |
c603844b | 3977 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3978 | |
736838e9 | 3979 | /* |
524c4807 | 3980 | * __GFP_HIGH is assumed to be the same as ALLOC_MIN_RESERVE |
736838e9 MN |
3981 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD |
3982 | * to save two branches. | |
3983 | */ | |
524c4807 | 3984 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_MIN_RESERVE); |
736838e9 | 3985 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 3986 | |
341ce06f PZ |
3987 | /* |
3988 | * The caller may dip into page reserves a bit more if the caller | |
3989 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3990 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1ebbb218 | 3991 | * set both ALLOC_NON_BLOCK and ALLOC_MIN_RESERVE(__GFP_HIGH). |
341ce06f | 3992 | */ |
736838e9 MN |
3993 | alloc_flags |= (__force int) |
3994 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 3995 | |
1ebbb218 | 3996 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) { |
5c3240d9 | 3997 | /* |
b104a35d DR |
3998 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3999 | * if it can't schedule. | |
5c3240d9 | 4000 | */ |
eb2e2b42 | 4001 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { |
1ebbb218 | 4002 | alloc_flags |= ALLOC_NON_BLOCK; |
eb2e2b42 MG |
4003 | |
4004 | if (order > 0) | |
4005 | alloc_flags |= ALLOC_HIGHATOMIC; | |
4006 | } | |
4007 | ||
523b9458 | 4008 | /* |
1ebbb218 MG |
4009 | * Ignore cpuset mems for non-blocking __GFP_HIGH (probably |
4010 | * GFP_ATOMIC) rather than fail, see the comment for | |
8e464522 | 4011 | * cpuset_node_allowed(). |
523b9458 | 4012 | */ |
1ebbb218 MG |
4013 | if (alloc_flags & ALLOC_MIN_RESERVE) |
4014 | alloc_flags &= ~ALLOC_CPUSET; | |
88dc6f20 | 4015 | } else if (unlikely(rt_task(current)) && in_task()) |
c988dcbe | 4016 | alloc_flags |= ALLOC_MIN_RESERVE; |
341ce06f | 4017 | |
8e3560d9 | 4018 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags); |
8510e69c | 4019 | |
341ce06f PZ |
4020 | return alloc_flags; |
4021 | } | |
4022 | ||
cd04ae1e | 4023 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4024 | { |
cd04ae1e MH |
4025 | if (!tsk_is_oom_victim(tsk)) |
4026 | return false; | |
4027 | ||
4028 | /* | |
4029 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4030 | * only to the thread with TIF_MEMDIE set | |
4031 | */ | |
4032 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4033 | return false; |
4034 | ||
cd04ae1e MH |
4035 | return true; |
4036 | } | |
4037 | ||
4038 | /* | |
4039 | * Distinguish requests which really need access to full memory | |
4040 | * reserves from oom victims which can live with a portion of it | |
4041 | */ | |
4042 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4043 | { | |
4044 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4045 | return 0; | |
31a6c190 | 4046 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4047 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4048 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4049 | return ALLOC_NO_WATERMARKS; |
4050 | if (!in_interrupt()) { | |
4051 | if (current->flags & PF_MEMALLOC) | |
4052 | return ALLOC_NO_WATERMARKS; | |
4053 | else if (oom_reserves_allowed(current)) | |
4054 | return ALLOC_OOM; | |
4055 | } | |
31a6c190 | 4056 | |
cd04ae1e MH |
4057 | return 0; |
4058 | } | |
4059 | ||
4060 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4061 | { | |
4062 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4063 | } |
4064 | ||
0a0337e0 MH |
4065 | /* |
4066 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4067 | * for the given allocation request. | |
491d79ae JW |
4068 | * |
4069 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4070 | * without success, or when we couldn't even meet the watermark if we | |
4071 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4072 | * |
4073 | * Returns true if a retry is viable or false to enter the oom path. | |
4074 | */ | |
4075 | static inline bool | |
4076 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4077 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4078 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4079 | { |
4080 | struct zone *zone; | |
4081 | struct zoneref *z; | |
15f570bf | 4082 | bool ret = false; |
0a0337e0 | 4083 | |
423b452e VB |
4084 | /* |
4085 | * Costly allocations might have made a progress but this doesn't mean | |
4086 | * their order will become available due to high fragmentation so | |
4087 | * always increment the no progress counter for them | |
4088 | */ | |
4089 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4090 | *no_progress_loops = 0; | |
4091 | else | |
4092 | (*no_progress_loops)++; | |
4093 | ||
0a0337e0 MH |
4094 | /* |
4095 | * Make sure we converge to OOM if we cannot make any progress | |
4096 | * several times in the row. | |
4097 | */ | |
04c8716f MK |
4098 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4099 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4100 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4101 | } |
0a0337e0 | 4102 | |
bca67592 MG |
4103 | /* |
4104 | * Keep reclaiming pages while there is a chance this will lead | |
4105 | * somewhere. If none of the target zones can satisfy our allocation | |
4106 | * request even if all reclaimable pages are considered then we are | |
4107 | * screwed and have to go OOM. | |
0a0337e0 | 4108 | */ |
97a225e6 JK |
4109 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4110 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4111 | unsigned long available; |
ede37713 | 4112 | unsigned long reclaimable; |
d379f01d MH |
4113 | unsigned long min_wmark = min_wmark_pages(zone); |
4114 | bool wmark; | |
0a0337e0 | 4115 | |
5a1c84b4 | 4116 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4117 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4118 | |
4119 | /* | |
491d79ae JW |
4120 | * Would the allocation succeed if we reclaimed all |
4121 | * reclaimable pages? | |
0a0337e0 | 4122 | */ |
d379f01d | 4123 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 4124 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
4125 | trace_reclaim_retry_zone(z, order, reclaimable, |
4126 | available, min_wmark, *no_progress_loops, wmark); | |
4127 | if (wmark) { | |
15f570bf | 4128 | ret = true; |
132b0d21 | 4129 | break; |
0a0337e0 MH |
4130 | } |
4131 | } | |
4132 | ||
15f570bf MH |
4133 | /* |
4134 | * Memory allocation/reclaim might be called from a WQ context and the | |
4135 | * current implementation of the WQ concurrency control doesn't | |
4136 | * recognize that a particular WQ is congested if the worker thread is | |
4137 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4138 | * here rather than calling cond_resched(). | |
4139 | */ | |
4140 | if (current->flags & PF_WQ_WORKER) | |
4141 | schedule_timeout_uninterruptible(1); | |
4142 | else | |
4143 | cond_resched(); | |
4144 | return ret; | |
0a0337e0 MH |
4145 | } |
4146 | ||
902b6281 VB |
4147 | static inline bool |
4148 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4149 | { | |
4150 | /* | |
4151 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4152 | * mempolicy don't intersect. This should be normally dealt with by | |
4153 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4154 | * such a way the check therein was true, and then it became false | |
4155 | * before we got our cpuset_mems_cookie here. | |
4156 | * This assumes that for all allocations, ac->nodemask can come only | |
4157 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4158 | * when it does not intersect with the cpuset restrictions) or the | |
4159 | * caller can deal with a violated nodemask. | |
4160 | */ | |
4161 | if (cpusets_enabled() && ac->nodemask && | |
4162 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4163 | ac->nodemask = NULL; | |
4164 | return true; | |
4165 | } | |
4166 | ||
4167 | /* | |
4168 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4169 | * possible to race with parallel threads in such a way that our | |
4170 | * allocation can fail while the mask is being updated. If we are about | |
4171 | * to fail, check if the cpuset changed during allocation and if so, | |
4172 | * retry. | |
4173 | */ | |
4174 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4175 | return true; | |
4176 | ||
4177 | return false; | |
4178 | } | |
4179 | ||
11e33f6a MG |
4180 | static inline struct page * |
4181 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4182 | struct alloc_context *ac) |
11e33f6a | 4183 | { |
d0164adc | 4184 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4185 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4186 | struct page *page = NULL; |
c603844b | 4187 | unsigned int alloc_flags; |
11e33f6a | 4188 | unsigned long did_some_progress; |
5ce9bfef | 4189 | enum compact_priority compact_priority; |
c5d01d0d | 4190 | enum compact_result compact_result; |
5ce9bfef VB |
4191 | int compaction_retries; |
4192 | int no_progress_loops; | |
5ce9bfef | 4193 | unsigned int cpuset_mems_cookie; |
3d36424b | 4194 | unsigned int zonelist_iter_cookie; |
cd04ae1e | 4195 | int reserve_flags; |
1da177e4 | 4196 | |
3d36424b | 4197 | restart: |
5ce9bfef VB |
4198 | compaction_retries = 0; |
4199 | no_progress_loops = 0; | |
4200 | compact_priority = DEF_COMPACT_PRIORITY; | |
4201 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
3d36424b | 4202 | zonelist_iter_cookie = zonelist_iter_begin(); |
9a67f648 MH |
4203 | |
4204 | /* | |
4205 | * The fast path uses conservative alloc_flags to succeed only until | |
4206 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4207 | * alloc_flags precisely. So we do that now. | |
4208 | */ | |
eb2e2b42 | 4209 | alloc_flags = gfp_to_alloc_flags(gfp_mask, order); |
9a67f648 | 4210 | |
e47483bc VB |
4211 | /* |
4212 | * We need to recalculate the starting point for the zonelist iterator | |
4213 | * because we might have used different nodemask in the fast path, or | |
4214 | * there was a cpuset modification and we are retrying - otherwise we | |
4215 | * could end up iterating over non-eligible zones endlessly. | |
4216 | */ | |
4217 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 4218 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
4219 | if (!ac->preferred_zoneref->zone) |
4220 | goto nopage; | |
4221 | ||
8ca1b5a4 FT |
4222 | /* |
4223 | * Check for insane configurations where the cpuset doesn't contain | |
4224 | * any suitable zone to satisfy the request - e.g. non-movable | |
4225 | * GFP_HIGHUSER allocations from MOVABLE nodes only. | |
4226 | */ | |
4227 | if (cpusets_insane_config() && (gfp_mask & __GFP_HARDWALL)) { | |
4228 | struct zoneref *z = first_zones_zonelist(ac->zonelist, | |
4229 | ac->highest_zoneidx, | |
4230 | &cpuset_current_mems_allowed); | |
4231 | if (!z->zone) | |
4232 | goto nopage; | |
4233 | } | |
4234 | ||
0a79cdad | 4235 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4236 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4237 | |
4238 | /* | |
4239 | * The adjusted alloc_flags might result in immediate success, so try | |
4240 | * that first | |
4241 | */ | |
4242 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4243 | if (page) | |
4244 | goto got_pg; | |
4245 | ||
a8161d1e VB |
4246 | /* |
4247 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4248 | * that we have enough base pages and don't need to reclaim. For non- |
4249 | * movable high-order allocations, do that as well, as compaction will | |
4250 | * try prevent permanent fragmentation by migrating from blocks of the | |
4251 | * same migratetype. | |
4252 | * Don't try this for allocations that are allowed to ignore | |
4253 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4254 | */ |
282722b0 VB |
4255 | if (can_direct_reclaim && |
4256 | (costly_order || | |
4257 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4258 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4259 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4260 | alloc_flags, ac, | |
a5508cd8 | 4261 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4262 | &compact_result); |
4263 | if (page) | |
4264 | goto got_pg; | |
4265 | ||
cc638f32 VB |
4266 | /* |
4267 | * Checks for costly allocations with __GFP_NORETRY, which | |
4268 | * includes some THP page fault allocations | |
4269 | */ | |
4270 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4271 | /* |
4272 | * If allocating entire pageblock(s) and compaction | |
4273 | * failed because all zones are below low watermarks | |
4274 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4275 | * order, fail immediately unless the allocator has |
4276 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4277 | * |
4278 | * Reclaim is | |
4279 | * - potentially very expensive because zones are far | |
4280 | * below their low watermarks or this is part of very | |
4281 | * bursty high order allocations, | |
4282 | * - not guaranteed to help because isolate_freepages() | |
4283 | * may not iterate over freed pages as part of its | |
4284 | * linear scan, and | |
4285 | * - unlikely to make entire pageblocks free on its | |
4286 | * own. | |
4287 | */ | |
4288 | if (compact_result == COMPACT_SKIPPED || | |
4289 | compact_result == COMPACT_DEFERRED) | |
4290 | goto nopage; | |
a8161d1e | 4291 | |
a8161d1e | 4292 | /* |
3eb2771b VB |
4293 | * Looks like reclaim/compaction is worth trying, but |
4294 | * sync compaction could be very expensive, so keep | |
25160354 | 4295 | * using async compaction. |
a8161d1e | 4296 | */ |
a5508cd8 | 4297 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4298 | } |
4299 | } | |
23771235 | 4300 | |
31a6c190 | 4301 | retry: |
23771235 | 4302 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4303 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4304 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4305 | |
cd04ae1e MH |
4306 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4307 | if (reserve_flags) | |
ce96fa62 ML |
4308 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags) | |
4309 | (alloc_flags & ALLOC_KSWAPD); | |
23771235 | 4310 | |
e46e7b77 | 4311 | /* |
d6a24df0 VB |
4312 | * Reset the nodemask and zonelist iterators if memory policies can be |
4313 | * ignored. These allocations are high priority and system rather than | |
4314 | * user oriented. | |
e46e7b77 | 4315 | */ |
cd04ae1e | 4316 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4317 | ac->nodemask = NULL; |
e46e7b77 | 4318 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4319 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
4320 | } |
4321 | ||
23771235 | 4322 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4323 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4324 | if (page) |
4325 | goto got_pg; | |
1da177e4 | 4326 | |
d0164adc | 4327 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4328 | if (!can_direct_reclaim) |
1da177e4 LT |
4329 | goto nopage; |
4330 | ||
9a67f648 MH |
4331 | /* Avoid recursion of direct reclaim */ |
4332 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4333 | goto nopage; |
4334 | ||
a8161d1e VB |
4335 | /* Try direct reclaim and then allocating */ |
4336 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4337 | &did_some_progress); | |
4338 | if (page) | |
4339 | goto got_pg; | |
4340 | ||
4341 | /* Try direct compaction and then allocating */ | |
a9263751 | 4342 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4343 | compact_priority, &compact_result); |
56de7263 MG |
4344 | if (page) |
4345 | goto got_pg; | |
75f30861 | 4346 | |
9083905a JW |
4347 | /* Do not loop if specifically requested */ |
4348 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4349 | goto nopage; |
9083905a | 4350 | |
0a0337e0 MH |
4351 | /* |
4352 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4353 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4354 | */ |
dcda9b04 | 4355 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4356 | goto nopage; |
0a0337e0 | 4357 | |
0a0337e0 | 4358 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4359 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4360 | goto retry; |
4361 | ||
33c2d214 MH |
4362 | /* |
4363 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4364 | * reclaim is not able to make any progress because the current | |
4365 | * implementation of the compaction depends on the sufficient amount | |
4366 | * of free memory (see __compaction_suitable) | |
4367 | */ | |
4368 | if (did_some_progress > 0 && | |
86a294a8 | 4369 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4370 | compact_result, &compact_priority, |
d9436498 | 4371 | &compaction_retries)) |
33c2d214 MH |
4372 | goto retry; |
4373 | ||
902b6281 | 4374 | |
3d36424b MG |
4375 | /* |
4376 | * Deal with possible cpuset update races or zonelist updates to avoid | |
4377 | * a unnecessary OOM kill. | |
4378 | */ | |
4379 | if (check_retry_cpuset(cpuset_mems_cookie, ac) || | |
4380 | check_retry_zonelist(zonelist_iter_cookie)) | |
4381 | goto restart; | |
e47483bc | 4382 | |
9083905a JW |
4383 | /* Reclaim has failed us, start killing things */ |
4384 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4385 | if (page) | |
4386 | goto got_pg; | |
4387 | ||
9a67f648 | 4388 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e | 4389 | if (tsk_is_oom_victim(current) && |
8510e69c | 4390 | (alloc_flags & ALLOC_OOM || |
c288983d | 4391 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4392 | goto nopage; |
4393 | ||
9083905a | 4394 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4395 | if (did_some_progress) { |
4396 | no_progress_loops = 0; | |
9083905a | 4397 | goto retry; |
0a0337e0 | 4398 | } |
9083905a | 4399 | |
1da177e4 | 4400 | nopage: |
3d36424b MG |
4401 | /* |
4402 | * Deal with possible cpuset update races or zonelist updates to avoid | |
4403 | * a unnecessary OOM kill. | |
4404 | */ | |
4405 | if (check_retry_cpuset(cpuset_mems_cookie, ac) || | |
4406 | check_retry_zonelist(zonelist_iter_cookie)) | |
4407 | goto restart; | |
5ce9bfef | 4408 | |
9a67f648 MH |
4409 | /* |
4410 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4411 | * we always retry | |
4412 | */ | |
4413 | if (gfp_mask & __GFP_NOFAIL) { | |
4414 | /* | |
4415 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4416 | * of any new users that actually require GFP_NOWAIT | |
4417 | */ | |
3f913fc5 | 4418 | if (WARN_ON_ONCE_GFP(!can_direct_reclaim, gfp_mask)) |
9a67f648 MH |
4419 | goto fail; |
4420 | ||
4421 | /* | |
4422 | * PF_MEMALLOC request from this context is rather bizarre | |
4423 | * because we cannot reclaim anything and only can loop waiting | |
4424 | * for somebody to do a work for us | |
4425 | */ | |
3f913fc5 | 4426 | WARN_ON_ONCE_GFP(current->flags & PF_MEMALLOC, gfp_mask); |
9a67f648 MH |
4427 | |
4428 | /* | |
4429 | * non failing costly orders are a hard requirement which we | |
4430 | * are not prepared for much so let's warn about these users | |
4431 | * so that we can identify them and convert them to something | |
4432 | * else. | |
4433 | */ | |
896c4d52 | 4434 | WARN_ON_ONCE_GFP(costly_order, gfp_mask); |
9a67f648 | 4435 | |
6c18ba7a | 4436 | /* |
1ebbb218 MG |
4437 | * Help non-failing allocations by giving some access to memory |
4438 | * reserves normally used for high priority non-blocking | |
4439 | * allocations but do not use ALLOC_NO_WATERMARKS because this | |
6c18ba7a | 4440 | * could deplete whole memory reserves which would just make |
1ebbb218 | 4441 | * the situation worse. |
6c18ba7a | 4442 | */ |
1ebbb218 | 4443 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_MIN_RESERVE, ac); |
6c18ba7a MH |
4444 | if (page) |
4445 | goto got_pg; | |
4446 | ||
9a67f648 MH |
4447 | cond_resched(); |
4448 | goto retry; | |
4449 | } | |
4450 | fail: | |
a8e99259 | 4451 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4452 | "page allocation failure: order:%u", order); |
1da177e4 | 4453 | got_pg: |
072bb0aa | 4454 | return page; |
1da177e4 | 4455 | } |
11e33f6a | 4456 | |
9cd75558 | 4457 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4458 | int preferred_nid, nodemask_t *nodemask, |
8e6a930b | 4459 | struct alloc_context *ac, gfp_t *alloc_gfp, |
9cd75558 | 4460 | unsigned int *alloc_flags) |
11e33f6a | 4461 | { |
97a225e6 | 4462 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4463 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 4464 | ac->nodemask = nodemask; |
01c0bfe0 | 4465 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 4466 | |
682a3385 | 4467 | if (cpusets_enabled()) { |
8e6a930b | 4468 | *alloc_gfp |= __GFP_HARDWALL; |
182f3d7a MS |
4469 | /* |
4470 | * When we are in the interrupt context, it is irrelevant | |
4471 | * to the current task context. It means that any node ok. | |
4472 | */ | |
88dc6f20 | 4473 | if (in_task() && !ac->nodemask) |
9cd75558 | 4474 | ac->nodemask = &cpuset_current_mems_allowed; |
51047820 VB |
4475 | else |
4476 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4477 | } |
4478 | ||
446ec838 | 4479 | might_alloc(gfp_mask); |
11e33f6a MG |
4480 | |
4481 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4482 | return false; |
11e33f6a | 4483 | |
8e3560d9 | 4484 | *alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags); |
d883c6cf | 4485 | |
c9ab0c4f | 4486 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4487 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4488 | |
e46e7b77 MG |
4489 | /* |
4490 | * The preferred zone is used for statistics but crucially it is | |
4491 | * also used as the starting point for the zonelist iterator. It | |
4492 | * may get reset for allocations that ignore memory policies. | |
4493 | */ | |
9cd75558 | 4494 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4495 | ac->highest_zoneidx, ac->nodemask); |
a0622d05 MN |
4496 | |
4497 | return true; | |
9cd75558 MG |
4498 | } |
4499 | ||
387ba26f | 4500 | /* |
0f87d9d3 | 4501 | * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array |
387ba26f MG |
4502 | * @gfp: GFP flags for the allocation |
4503 | * @preferred_nid: The preferred NUMA node ID to allocate from | |
4504 | * @nodemask: Set of nodes to allocate from, may be NULL | |
0f87d9d3 MG |
4505 | * @nr_pages: The number of pages desired on the list or array |
4506 | * @page_list: Optional list to store the allocated pages | |
4507 | * @page_array: Optional array to store the pages | |
387ba26f MG |
4508 | * |
4509 | * This is a batched version of the page allocator that attempts to | |
0f87d9d3 MG |
4510 | * allocate nr_pages quickly. Pages are added to page_list if page_list |
4511 | * is not NULL, otherwise it is assumed that the page_array is valid. | |
387ba26f | 4512 | * |
0f87d9d3 MG |
4513 | * For lists, nr_pages is the number of pages that should be allocated. |
4514 | * | |
4515 | * For arrays, only NULL elements are populated with pages and nr_pages | |
4516 | * is the maximum number of pages that will be stored in the array. | |
4517 | * | |
4518 | * Returns the number of pages on the list or array. | |
387ba26f MG |
4519 | */ |
4520 | unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, | |
4521 | nodemask_t *nodemask, int nr_pages, | |
0f87d9d3 MG |
4522 | struct list_head *page_list, |
4523 | struct page **page_array) | |
387ba26f MG |
4524 | { |
4525 | struct page *page; | |
4b23a68f | 4526 | unsigned long __maybe_unused UP_flags; |
387ba26f MG |
4527 | struct zone *zone; |
4528 | struct zoneref *z; | |
4529 | struct per_cpu_pages *pcp; | |
4530 | struct list_head *pcp_list; | |
4531 | struct alloc_context ac; | |
4532 | gfp_t alloc_gfp; | |
4533 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
3e23060b | 4534 | int nr_populated = 0, nr_account = 0; |
387ba26f | 4535 | |
0f87d9d3 MG |
4536 | /* |
4537 | * Skip populated array elements to determine if any pages need | |
4538 | * to be allocated before disabling IRQs. | |
4539 | */ | |
b08e50dd | 4540 | while (page_array && nr_populated < nr_pages && page_array[nr_populated]) |
0f87d9d3 MG |
4541 | nr_populated++; |
4542 | ||
06147843 CL |
4543 | /* No pages requested? */ |
4544 | if (unlikely(nr_pages <= 0)) | |
4545 | goto out; | |
4546 | ||
b3b64ebd MG |
4547 | /* Already populated array? */ |
4548 | if (unlikely(page_array && nr_pages - nr_populated == 0)) | |
06147843 | 4549 | goto out; |
b3b64ebd | 4550 | |
8dcb3060 | 4551 | /* Bulk allocator does not support memcg accounting. */ |
f7a449f7 | 4552 | if (memcg_kmem_online() && (gfp & __GFP_ACCOUNT)) |
8dcb3060 SB |
4553 | goto failed; |
4554 | ||
387ba26f | 4555 | /* Use the single page allocator for one page. */ |
0f87d9d3 | 4556 | if (nr_pages - nr_populated == 1) |
387ba26f MG |
4557 | goto failed; |
4558 | ||
187ad460 MG |
4559 | #ifdef CONFIG_PAGE_OWNER |
4560 | /* | |
4561 | * PAGE_OWNER may recurse into the allocator to allocate space to | |
4562 | * save the stack with pagesets.lock held. Releasing/reacquiring | |
4563 | * removes much of the performance benefit of bulk allocation so | |
4564 | * force the caller to allocate one page at a time as it'll have | |
4565 | * similar performance to added complexity to the bulk allocator. | |
4566 | */ | |
4567 | if (static_branch_unlikely(&page_owner_inited)) | |
4568 | goto failed; | |
4569 | #endif | |
4570 | ||
387ba26f MG |
4571 | /* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */ |
4572 | gfp &= gfp_allowed_mask; | |
4573 | alloc_gfp = gfp; | |
4574 | if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags)) | |
06147843 | 4575 | goto out; |
387ba26f MG |
4576 | gfp = alloc_gfp; |
4577 | ||
4578 | /* Find an allowed local zone that meets the low watermark. */ | |
4579 | for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) { | |
4580 | unsigned long mark; | |
4581 | ||
4582 | if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) && | |
4583 | !__cpuset_zone_allowed(zone, gfp)) { | |
4584 | continue; | |
4585 | } | |
4586 | ||
4587 | if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone && | |
4588 | zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) { | |
4589 | goto failed; | |
4590 | } | |
4591 | ||
4592 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages; | |
4593 | if (zone_watermark_fast(zone, 0, mark, | |
4594 | zonelist_zone_idx(ac.preferred_zoneref), | |
4595 | alloc_flags, gfp)) { | |
4596 | break; | |
4597 | } | |
4598 | } | |
4599 | ||
4600 | /* | |
4601 | * If there are no allowed local zones that meets the watermarks then | |
4602 | * try to allocate a single page and reclaim if necessary. | |
4603 | */ | |
ce76f9a1 | 4604 | if (unlikely(!zone)) |
387ba26f MG |
4605 | goto failed; |
4606 | ||
57490774 | 4607 | /* spin_trylock may fail due to a parallel drain or IRQ reentrancy. */ |
4b23a68f | 4608 | pcp_trylock_prepare(UP_flags); |
57490774 | 4609 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 4610 | if (!pcp) |
4b23a68f | 4611 | goto failed_irq; |
387ba26f | 4612 | |
387ba26f | 4613 | /* Attempt the batch allocation */ |
44042b44 | 4614 | pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)]; |
0f87d9d3 MG |
4615 | while (nr_populated < nr_pages) { |
4616 | ||
4617 | /* Skip existing pages */ | |
4618 | if (page_array && page_array[nr_populated]) { | |
4619 | nr_populated++; | |
4620 | continue; | |
4621 | } | |
4622 | ||
44042b44 | 4623 | page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags, |
387ba26f | 4624 | pcp, pcp_list); |
ce76f9a1 | 4625 | if (unlikely(!page)) { |
c572e488 | 4626 | /* Try and allocate at least one page */ |
4b23a68f | 4627 | if (!nr_account) { |
57490774 | 4628 | pcp_spin_unlock(pcp); |
387ba26f | 4629 | goto failed_irq; |
4b23a68f | 4630 | } |
387ba26f MG |
4631 | break; |
4632 | } | |
3e23060b | 4633 | nr_account++; |
387ba26f MG |
4634 | |
4635 | prep_new_page(page, 0, gfp, 0); | |
0f87d9d3 MG |
4636 | if (page_list) |
4637 | list_add(&page->lru, page_list); | |
4638 | else | |
4639 | page_array[nr_populated] = page; | |
4640 | nr_populated++; | |
387ba26f MG |
4641 | } |
4642 | ||
57490774 | 4643 | pcp_spin_unlock(pcp); |
4b23a68f | 4644 | pcp_trylock_finish(UP_flags); |
43c95bcc | 4645 | |
3e23060b MG |
4646 | __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account); |
4647 | zone_statistics(ac.preferred_zoneref->zone, zone, nr_account); | |
387ba26f | 4648 | |
06147843 | 4649 | out: |
0f87d9d3 | 4650 | return nr_populated; |
387ba26f MG |
4651 | |
4652 | failed_irq: | |
4b23a68f | 4653 | pcp_trylock_finish(UP_flags); |
387ba26f MG |
4654 | |
4655 | failed: | |
4656 | page = __alloc_pages(gfp, 0, preferred_nid, nodemask); | |
4657 | if (page) { | |
0f87d9d3 MG |
4658 | if (page_list) |
4659 | list_add(&page->lru, page_list); | |
4660 | else | |
4661 | page_array[nr_populated] = page; | |
4662 | nr_populated++; | |
387ba26f MG |
4663 | } |
4664 | ||
06147843 | 4665 | goto out; |
387ba26f MG |
4666 | } |
4667 | EXPORT_SYMBOL_GPL(__alloc_pages_bulk); | |
4668 | ||
9cd75558 MG |
4669 | /* |
4670 | * This is the 'heart' of the zoned buddy allocator. | |
4671 | */ | |
84172f4b | 4672 | struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, |
04ec6264 | 4673 | nodemask_t *nodemask) |
9cd75558 MG |
4674 | { |
4675 | struct page *page; | |
4676 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
8e6a930b | 4677 | gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4678 | struct alloc_context ac = { }; |
4679 | ||
c63ae43b MH |
4680 | /* |
4681 | * There are several places where we assume that the order value is sane | |
4682 | * so bail out early if the request is out of bound. | |
4683 | */ | |
23baf831 | 4684 | if (WARN_ON_ONCE_GFP(order > MAX_ORDER, gfp)) |
c63ae43b | 4685 | return NULL; |
c63ae43b | 4686 | |
6e5e0f28 | 4687 | gfp &= gfp_allowed_mask; |
da6df1b0 PT |
4688 | /* |
4689 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS | |
4690 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4691 | * from a particular context which has been marked by | |
8e3560d9 PT |
4692 | * memalloc_no{fs,io}_{save,restore}. And PF_MEMALLOC_PIN which ensures |
4693 | * movable zones are not used during allocation. | |
da6df1b0 PT |
4694 | */ |
4695 | gfp = current_gfp_context(gfp); | |
6e5e0f28 MWO |
4696 | alloc_gfp = gfp; |
4697 | if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac, | |
8e6a930b | 4698 | &alloc_gfp, &alloc_flags)) |
9cd75558 MG |
4699 | return NULL; |
4700 | ||
6bb15450 MG |
4701 | /* |
4702 | * Forbid the first pass from falling back to types that fragment | |
4703 | * memory until all local zones are considered. | |
4704 | */ | |
6e5e0f28 | 4705 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp); |
6bb15450 | 4706 | |
5117f45d | 4707 | /* First allocation attempt */ |
8e6a930b | 4708 | page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac); |
4fcb0971 MG |
4709 | if (likely(page)) |
4710 | goto out; | |
11e33f6a | 4711 | |
da6df1b0 | 4712 | alloc_gfp = gfp; |
4fcb0971 | 4713 | ac.spread_dirty_pages = false; |
23f086f9 | 4714 | |
4741526b MG |
4715 | /* |
4716 | * Restore the original nodemask if it was potentially replaced with | |
4717 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4718 | */ | |
97ce86f9 | 4719 | ac.nodemask = nodemask; |
16096c25 | 4720 | |
8e6a930b | 4721 | page = __alloc_pages_slowpath(alloc_gfp, order, &ac); |
cc9a6c87 | 4722 | |
4fcb0971 | 4723 | out: |
f7a449f7 | 4724 | if (memcg_kmem_online() && (gfp & __GFP_ACCOUNT) && page && |
6e5e0f28 | 4725 | unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) { |
c4159a75 VD |
4726 | __free_pages(page, order); |
4727 | page = NULL; | |
4949148a VD |
4728 | } |
4729 | ||
8e6a930b | 4730 | trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype); |
b073d7f8 | 4731 | kmsan_alloc_page(page, order, alloc_gfp); |
4fcb0971 | 4732 | |
11e33f6a | 4733 | return page; |
1da177e4 | 4734 | } |
84172f4b | 4735 | EXPORT_SYMBOL(__alloc_pages); |
1da177e4 | 4736 | |
cc09cb13 MWO |
4737 | struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid, |
4738 | nodemask_t *nodemask) | |
4739 | { | |
4740 | struct page *page = __alloc_pages(gfp | __GFP_COMP, order, | |
4741 | preferred_nid, nodemask); | |
4742 | ||
4743 | if (page && order > 1) | |
4744 | prep_transhuge_page(page); | |
4745 | return (struct folio *)page; | |
4746 | } | |
4747 | EXPORT_SYMBOL(__folio_alloc); | |
4748 | ||
1da177e4 | 4749 | /* |
9ea9a680 MH |
4750 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4751 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4752 | * you need to access high mem. | |
1da177e4 | 4753 | */ |
920c7a5d | 4754 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4755 | { |
945a1113 AM |
4756 | struct page *page; |
4757 | ||
9ea9a680 | 4758 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4759 | if (!page) |
4760 | return 0; | |
4761 | return (unsigned long) page_address(page); | |
4762 | } | |
1da177e4 LT |
4763 | EXPORT_SYMBOL(__get_free_pages); |
4764 | ||
920c7a5d | 4765 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4766 | { |
dcc1be11 | 4767 | return __get_free_page(gfp_mask | __GFP_ZERO); |
1da177e4 | 4768 | } |
1da177e4 LT |
4769 | EXPORT_SYMBOL(get_zeroed_page); |
4770 | ||
7f194fbb MWO |
4771 | /** |
4772 | * __free_pages - Free pages allocated with alloc_pages(). | |
4773 | * @page: The page pointer returned from alloc_pages(). | |
4774 | * @order: The order of the allocation. | |
4775 | * | |
4776 | * This function can free multi-page allocations that are not compound | |
4777 | * pages. It does not check that the @order passed in matches that of | |
4778 | * the allocation, so it is easy to leak memory. Freeing more memory | |
4779 | * than was allocated will probably emit a warning. | |
4780 | * | |
4781 | * If the last reference to this page is speculative, it will be released | |
4782 | * by put_page() which only frees the first page of a non-compound | |
4783 | * allocation. To prevent the remaining pages from being leaked, we free | |
4784 | * the subsequent pages here. If you want to use the page's reference | |
4785 | * count to decide when to free the allocation, you should allocate a | |
4786 | * compound page, and use put_page() instead of __free_pages(). | |
4787 | * | |
4788 | * Context: May be called in interrupt context or while holding a normal | |
4789 | * spinlock, but not in NMI context or while holding a raw spinlock. | |
4790 | */ | |
742aa7fb AL |
4791 | void __free_pages(struct page *page, unsigned int order) |
4792 | { | |
462a8e08 DC |
4793 | /* get PageHead before we drop reference */ |
4794 | int head = PageHead(page); | |
4795 | ||
742aa7fb AL |
4796 | if (put_page_testzero(page)) |
4797 | free_the_page(page, order); | |
462a8e08 | 4798 | else if (!head) |
e320d301 MWO |
4799 | while (order-- > 0) |
4800 | free_the_page(page + (1 << order), order); | |
742aa7fb | 4801 | } |
1da177e4 LT |
4802 | EXPORT_SYMBOL(__free_pages); |
4803 | ||
920c7a5d | 4804 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4805 | { |
4806 | if (addr != 0) { | |
725d704e | 4807 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4808 | __free_pages(virt_to_page((void *)addr), order); |
4809 | } | |
4810 | } | |
4811 | ||
4812 | EXPORT_SYMBOL(free_pages); | |
4813 | ||
b63ae8ca AD |
4814 | /* |
4815 | * Page Fragment: | |
4816 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4817 | * within a 0 or higher order page. Multiple fragments within that page | |
4818 | * are individually refcounted, in the page's reference counter. | |
4819 | * | |
4820 | * The page_frag functions below provide a simple allocation framework for | |
4821 | * page fragments. This is used by the network stack and network device | |
4822 | * drivers to provide a backing region of memory for use as either an | |
4823 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4824 | */ | |
2976db80 AD |
4825 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4826 | gfp_t gfp_mask) | |
b63ae8ca AD |
4827 | { |
4828 | struct page *page = NULL; | |
4829 | gfp_t gfp = gfp_mask; | |
4830 | ||
4831 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4832 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4833 | __GFP_NOMEMALLOC; | |
4834 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4835 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4836 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4837 | #endif | |
4838 | if (unlikely(!page)) | |
4839 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4840 | ||
4841 | nc->va = page ? page_address(page) : NULL; | |
4842 | ||
4843 | return page; | |
4844 | } | |
4845 | ||
2976db80 | 4846 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4847 | { |
4848 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4849 | ||
742aa7fb AL |
4850 | if (page_ref_sub_and_test(page, count)) |
4851 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 4852 | } |
2976db80 | 4853 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4854 | |
b358e212 KH |
4855 | void *page_frag_alloc_align(struct page_frag_cache *nc, |
4856 | unsigned int fragsz, gfp_t gfp_mask, | |
4857 | unsigned int align_mask) | |
b63ae8ca AD |
4858 | { |
4859 | unsigned int size = PAGE_SIZE; | |
4860 | struct page *page; | |
4861 | int offset; | |
4862 | ||
4863 | if (unlikely(!nc->va)) { | |
4864 | refill: | |
2976db80 | 4865 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4866 | if (!page) |
4867 | return NULL; | |
4868 | ||
4869 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4870 | /* if size can vary use size else just use PAGE_SIZE */ | |
4871 | size = nc->size; | |
4872 | #endif | |
4873 | /* Even if we own the page, we do not use atomic_set(). | |
4874 | * This would break get_page_unless_zero() users. | |
4875 | */ | |
86447726 | 4876 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
4877 | |
4878 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4879 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 4880 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
4881 | nc->offset = size; |
4882 | } | |
4883 | ||
4884 | offset = nc->offset - fragsz; | |
4885 | if (unlikely(offset < 0)) { | |
4886 | page = virt_to_page(nc->va); | |
4887 | ||
fe896d18 | 4888 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4889 | goto refill; |
4890 | ||
d8c19014 DZ |
4891 | if (unlikely(nc->pfmemalloc)) { |
4892 | free_the_page(page, compound_order(page)); | |
4893 | goto refill; | |
4894 | } | |
4895 | ||
b63ae8ca AD |
4896 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
4897 | /* if size can vary use size else just use PAGE_SIZE */ | |
4898 | size = nc->size; | |
4899 | #endif | |
4900 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 4901 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
4902 | |
4903 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 4904 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca | 4905 | offset = size - fragsz; |
dac22531 ML |
4906 | if (unlikely(offset < 0)) { |
4907 | /* | |
4908 | * The caller is trying to allocate a fragment | |
4909 | * with fragsz > PAGE_SIZE but the cache isn't big | |
4910 | * enough to satisfy the request, this may | |
4911 | * happen in low memory conditions. | |
4912 | * We don't release the cache page because | |
4913 | * it could make memory pressure worse | |
4914 | * so we simply return NULL here. | |
4915 | */ | |
4916 | return NULL; | |
4917 | } | |
b63ae8ca AD |
4918 | } |
4919 | ||
4920 | nc->pagecnt_bias--; | |
b358e212 | 4921 | offset &= align_mask; |
b63ae8ca AD |
4922 | nc->offset = offset; |
4923 | ||
4924 | return nc->va + offset; | |
4925 | } | |
b358e212 | 4926 | EXPORT_SYMBOL(page_frag_alloc_align); |
b63ae8ca AD |
4927 | |
4928 | /* | |
4929 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4930 | */ | |
8c2dd3e4 | 4931 | void page_frag_free(void *addr) |
b63ae8ca AD |
4932 | { |
4933 | struct page *page = virt_to_head_page(addr); | |
4934 | ||
742aa7fb AL |
4935 | if (unlikely(put_page_testzero(page))) |
4936 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 4937 | } |
8c2dd3e4 | 4938 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4939 | |
d00181b9 KS |
4940 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4941 | size_t size) | |
ee85c2e1 AK |
4942 | { |
4943 | if (addr) { | |
df48a5f7 LH |
4944 | unsigned long nr = DIV_ROUND_UP(size, PAGE_SIZE); |
4945 | struct page *page = virt_to_page((void *)addr); | |
4946 | struct page *last = page + nr; | |
4947 | ||
4948 | split_page_owner(page, 1 << order); | |
4949 | split_page_memcg(page, 1 << order); | |
4950 | while (page < --last) | |
4951 | set_page_refcounted(last); | |
4952 | ||
4953 | last = page + (1UL << order); | |
4954 | for (page += nr; page < last; page++) | |
4955 | __free_pages_ok(page, 0, FPI_TO_TAIL); | |
ee85c2e1 AK |
4956 | } |
4957 | return (void *)addr; | |
4958 | } | |
4959 | ||
2be0ffe2 TT |
4960 | /** |
4961 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4962 | * @size: the number of bytes to allocate | |
63931eb9 | 4963 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
4964 | * |
4965 | * This function is similar to alloc_pages(), except that it allocates the | |
4966 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4967 | * allocate memory in power-of-two pages. | |
4968 | * | |
4969 | * This function is also limited by MAX_ORDER. | |
4970 | * | |
4971 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
4972 | * |
4973 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
4974 | */ |
4975 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4976 | { | |
4977 | unsigned int order = get_order(size); | |
4978 | unsigned long addr; | |
4979 | ||
ba7f1b9e ML |
4980 | if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM))) |
4981 | gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM); | |
63931eb9 | 4982 | |
2be0ffe2 | 4983 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 4984 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4985 | } |
4986 | EXPORT_SYMBOL(alloc_pages_exact); | |
4987 | ||
ee85c2e1 AK |
4988 | /** |
4989 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4990 | * pages on a node. | |
b5e6ab58 | 4991 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 4992 | * @size: the number of bytes to allocate |
63931eb9 | 4993 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
4994 | * |
4995 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4996 | * back. | |
a862f68a MR |
4997 | * |
4998 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 4999 | */ |
e1931811 | 5000 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5001 | { |
d00181b9 | 5002 | unsigned int order = get_order(size); |
63931eb9 VB |
5003 | struct page *p; |
5004 | ||
ba7f1b9e ML |
5005 | if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM))) |
5006 | gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM); | |
63931eb9 VB |
5007 | |
5008 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5009 | if (!p) |
5010 | return NULL; | |
5011 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5012 | } | |
ee85c2e1 | 5013 | |
2be0ffe2 TT |
5014 | /** |
5015 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5016 | * @virt: the value returned by alloc_pages_exact. | |
5017 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5018 | * | |
5019 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5020 | */ | |
5021 | void free_pages_exact(void *virt, size_t size) | |
5022 | { | |
5023 | unsigned long addr = (unsigned long)virt; | |
5024 | unsigned long end = addr + PAGE_ALIGN(size); | |
5025 | ||
5026 | while (addr < end) { | |
5027 | free_page(addr); | |
5028 | addr += PAGE_SIZE; | |
5029 | } | |
5030 | } | |
5031 | EXPORT_SYMBOL(free_pages_exact); | |
5032 | ||
e0fb5815 ZY |
5033 | /** |
5034 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5035 | * @offset: The zone index of the highest zone | |
5036 | * | |
a862f68a | 5037 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5038 | * high watermark within all zones at or below a given zone index. For each |
5039 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
5040 | * |
5041 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5042 | * |
5043 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5044 | */ |
ebec3862 | 5045 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5046 | { |
dd1a239f | 5047 | struct zoneref *z; |
54a6eb5c MG |
5048 | struct zone *zone; |
5049 | ||
e310fd43 | 5050 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5051 | unsigned long sum = 0; |
1da177e4 | 5052 | |
0e88460d | 5053 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5054 | |
54a6eb5c | 5055 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5056 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5057 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5058 | if (size > high) |
5059 | sum += size - high; | |
1da177e4 LT |
5060 | } |
5061 | ||
5062 | return sum; | |
5063 | } | |
5064 | ||
e0fb5815 ZY |
5065 | /** |
5066 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5067 | * | |
5068 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5069 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5070 | * |
5071 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5072 | * ZONE_NORMAL. | |
1da177e4 | 5073 | */ |
ebec3862 | 5074 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5075 | { |
af4ca457 | 5076 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5077 | } |
c2f1a551 | 5078 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5079 | |
19770b32 MG |
5080 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5081 | { | |
5082 | zoneref->zone = zone; | |
5083 | zoneref->zone_idx = zone_idx(zone); | |
5084 | } | |
5085 | ||
1da177e4 LT |
5086 | /* |
5087 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5088 | * |
5089 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5090 | */ |
9d3be21b | 5091 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5092 | { |
1a93205b | 5093 | struct zone *zone; |
bc732f1d | 5094 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5095 | int nr_zones = 0; |
02a68a5e CL |
5096 | |
5097 | do { | |
2f6726e5 | 5098 | zone_type--; |
070f8032 | 5099 | zone = pgdat->node_zones + zone_type; |
e553f62f | 5100 | if (populated_zone(zone)) { |
9d3be21b | 5101 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5102 | check_highest_zone(zone_type); |
1da177e4 | 5103 | } |
2f6726e5 | 5104 | } while (zone_type); |
bc732f1d | 5105 | |
070f8032 | 5106 | return nr_zones; |
1da177e4 LT |
5107 | } |
5108 | ||
5109 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5110 | |
5111 | static int __parse_numa_zonelist_order(char *s) | |
5112 | { | |
c9bff3ee | 5113 | /* |
f0953a1b | 5114 | * We used to support different zonelists modes but they turned |
c9bff3ee MH |
5115 | * out to be just not useful. Let's keep the warning in place |
5116 | * if somebody still use the cmd line parameter so that we do | |
5117 | * not fail it silently | |
5118 | */ | |
5119 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5120 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5121 | return -EINVAL; |
5122 | } | |
5123 | return 0; | |
5124 | } | |
5125 | ||
c9bff3ee MH |
5126 | char numa_zonelist_order[] = "Node"; |
5127 | ||
f0c0b2b8 KH |
5128 | /* |
5129 | * sysctl handler for numa_zonelist_order | |
5130 | */ | |
cccad5b9 | 5131 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 5132 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 5133 | { |
32927393 CH |
5134 | if (write) |
5135 | return __parse_numa_zonelist_order(buffer); | |
5136 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
5137 | } |
5138 | ||
5139 | ||
f0c0b2b8 KH |
5140 | static int node_load[MAX_NUMNODES]; |
5141 | ||
1da177e4 | 5142 | /** |
4dc3b16b | 5143 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5144 | * @node: node whose fallback list we're appending |
5145 | * @used_node_mask: nodemask_t of already used nodes | |
5146 | * | |
5147 | * We use a number of factors to determine which is the next node that should | |
5148 | * appear on a given node's fallback list. The node should not have appeared | |
5149 | * already in @node's fallback list, and it should be the next closest node | |
5150 | * according to the distance array (which contains arbitrary distance values | |
5151 | * from each node to each node in the system), and should also prefer nodes | |
5152 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5153 | * on them otherwise. | |
a862f68a MR |
5154 | * |
5155 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 5156 | */ |
79c28a41 | 5157 | int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5158 | { |
4cf808eb | 5159 | int n, val; |
1da177e4 | 5160 | int min_val = INT_MAX; |
00ef2d2f | 5161 | int best_node = NUMA_NO_NODE; |
1da177e4 | 5162 | |
4cf808eb LT |
5163 | /* Use the local node if we haven't already */ |
5164 | if (!node_isset(node, *used_node_mask)) { | |
5165 | node_set(node, *used_node_mask); | |
5166 | return node; | |
5167 | } | |
1da177e4 | 5168 | |
4b0ef1fe | 5169 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5170 | |
5171 | /* Don't want a node to appear more than once */ | |
5172 | if (node_isset(n, *used_node_mask)) | |
5173 | continue; | |
5174 | ||
1da177e4 LT |
5175 | /* Use the distance array to find the distance */ |
5176 | val = node_distance(node, n); | |
5177 | ||
4cf808eb LT |
5178 | /* Penalize nodes under us ("prefer the next node") */ |
5179 | val += (n < node); | |
5180 | ||
1da177e4 | 5181 | /* Give preference to headless and unused nodes */ |
b630749f | 5182 | if (!cpumask_empty(cpumask_of_node(n))) |
1da177e4 LT |
5183 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5184 | ||
5185 | /* Slight preference for less loaded node */ | |
37931324 | 5186 | val *= MAX_NUMNODES; |
1da177e4 LT |
5187 | val += node_load[n]; |
5188 | ||
5189 | if (val < min_val) { | |
5190 | min_val = val; | |
5191 | best_node = n; | |
5192 | } | |
5193 | } | |
5194 | ||
5195 | if (best_node >= 0) | |
5196 | node_set(best_node, *used_node_mask); | |
5197 | ||
5198 | return best_node; | |
5199 | } | |
5200 | ||
f0c0b2b8 KH |
5201 | |
5202 | /* | |
5203 | * Build zonelists ordered by node and zones within node. | |
5204 | * This results in maximum locality--normal zone overflows into local | |
5205 | * DMA zone, if any--but risks exhausting DMA zone. | |
5206 | */ | |
9d3be21b MH |
5207 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5208 | unsigned nr_nodes) | |
1da177e4 | 5209 | { |
9d3be21b MH |
5210 | struct zoneref *zonerefs; |
5211 | int i; | |
5212 | ||
5213 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5214 | ||
5215 | for (i = 0; i < nr_nodes; i++) { | |
5216 | int nr_zones; | |
5217 | ||
5218 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5219 | |
9d3be21b MH |
5220 | nr_zones = build_zonerefs_node(node, zonerefs); |
5221 | zonerefs += nr_zones; | |
5222 | } | |
5223 | zonerefs->zone = NULL; | |
5224 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5225 | } |
5226 | ||
523b9458 CL |
5227 | /* |
5228 | * Build gfp_thisnode zonelists | |
5229 | */ | |
5230 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5231 | { | |
9d3be21b MH |
5232 | struct zoneref *zonerefs; |
5233 | int nr_zones; | |
523b9458 | 5234 | |
9d3be21b MH |
5235 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5236 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5237 | zonerefs += nr_zones; | |
5238 | zonerefs->zone = NULL; | |
5239 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5240 | } |
5241 | ||
f0c0b2b8 KH |
5242 | /* |
5243 | * Build zonelists ordered by zone and nodes within zones. | |
5244 | * This results in conserving DMA zone[s] until all Normal memory is | |
5245 | * exhausted, but results in overflowing to remote node while memory | |
5246 | * may still exist in local DMA zone. | |
5247 | */ | |
f0c0b2b8 | 5248 | |
f0c0b2b8 KH |
5249 | static void build_zonelists(pg_data_t *pgdat) |
5250 | { | |
9d3be21b | 5251 | static int node_order[MAX_NUMNODES]; |
37931324 | 5252 | int node, nr_nodes = 0; |
d0ddf49b | 5253 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 5254 | int local_node, prev_node; |
1da177e4 LT |
5255 | |
5256 | /* NUMA-aware ordering of nodes */ | |
5257 | local_node = pgdat->node_id; | |
1da177e4 | 5258 | prev_node = local_node; |
f0c0b2b8 | 5259 | |
f0c0b2b8 | 5260 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5261 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5262 | /* | |
5263 | * We don't want to pressure a particular node. | |
5264 | * So adding penalty to the first node in same | |
5265 | * distance group to make it round-robin. | |
5266 | */ | |
957f822a DR |
5267 | if (node_distance(local_node, node) != |
5268 | node_distance(local_node, prev_node)) | |
37931324 | 5269 | node_load[node] += 1; |
f0c0b2b8 | 5270 | |
9d3be21b | 5271 | node_order[nr_nodes++] = node; |
1da177e4 | 5272 | prev_node = node; |
1da177e4 | 5273 | } |
523b9458 | 5274 | |
9d3be21b | 5275 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5276 | build_thisnode_zonelists(pgdat); |
6cf25392 BR |
5277 | pr_info("Fallback order for Node %d: ", local_node); |
5278 | for (node = 0; node < nr_nodes; node++) | |
5279 | pr_cont("%d ", node_order[node]); | |
5280 | pr_cont("\n"); | |
1da177e4 LT |
5281 | } |
5282 | ||
7aac7898 LS |
5283 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5284 | /* | |
5285 | * Return node id of node used for "local" allocations. | |
5286 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5287 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5288 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5289 | */ | |
5290 | int local_memory_node(int node) | |
5291 | { | |
c33d6c06 | 5292 | struct zoneref *z; |
7aac7898 | 5293 | |
c33d6c06 | 5294 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5295 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 5296 | NULL); |
c1093b74 | 5297 | return zone_to_nid(z->zone); |
7aac7898 LS |
5298 | } |
5299 | #endif | |
f0c0b2b8 | 5300 | |
6423aa81 JK |
5301 | static void setup_min_unmapped_ratio(void); |
5302 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5303 | #else /* CONFIG_NUMA */ |
5304 | ||
f0c0b2b8 | 5305 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5306 | { |
19655d34 | 5307 | int node, local_node; |
9d3be21b MH |
5308 | struct zoneref *zonerefs; |
5309 | int nr_zones; | |
1da177e4 LT |
5310 | |
5311 | local_node = pgdat->node_id; | |
1da177e4 | 5312 | |
9d3be21b MH |
5313 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5314 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5315 | zonerefs += nr_zones; | |
1da177e4 | 5316 | |
54a6eb5c MG |
5317 | /* |
5318 | * Now we build the zonelist so that it contains the zones | |
5319 | * of all the other nodes. | |
5320 | * We don't want to pressure a particular node, so when | |
5321 | * building the zones for node N, we make sure that the | |
5322 | * zones coming right after the local ones are those from | |
5323 | * node N+1 (modulo N) | |
5324 | */ | |
5325 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5326 | if (!node_online(node)) | |
5327 | continue; | |
9d3be21b MH |
5328 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5329 | zonerefs += nr_zones; | |
1da177e4 | 5330 | } |
54a6eb5c MG |
5331 | for (node = 0; node < local_node; node++) { |
5332 | if (!node_online(node)) | |
5333 | continue; | |
9d3be21b MH |
5334 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5335 | zonerefs += nr_zones; | |
54a6eb5c MG |
5336 | } |
5337 | ||
9d3be21b MH |
5338 | zonerefs->zone = NULL; |
5339 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5340 | } |
5341 | ||
5342 | #endif /* CONFIG_NUMA */ | |
5343 | ||
99dcc3e5 CL |
5344 | /* |
5345 | * Boot pageset table. One per cpu which is going to be used for all | |
5346 | * zones and all nodes. The parameters will be set in such a way | |
5347 | * that an item put on a list will immediately be handed over to | |
5348 | * the buddy list. This is safe since pageset manipulation is done | |
5349 | * with interrupts disabled. | |
5350 | * | |
5351 | * The boot_pagesets must be kept even after bootup is complete for | |
5352 | * unused processors and/or zones. They do play a role for bootstrapping | |
5353 | * hotplugged processors. | |
5354 | * | |
5355 | * zoneinfo_show() and maybe other functions do | |
5356 | * not check if the processor is online before following the pageset pointer. | |
5357 | * Other parts of the kernel may not check if the zone is available. | |
5358 | */ | |
28f836b6 | 5359 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats); |
952eaf81 VB |
5360 | /* These effectively disable the pcplists in the boot pageset completely */ |
5361 | #define BOOT_PAGESET_HIGH 0 | |
5362 | #define BOOT_PAGESET_BATCH 1 | |
28f836b6 MG |
5363 | static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset); |
5364 | static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats); | |
99dcc3e5 | 5365 | |
11cd8638 | 5366 | static void __build_all_zonelists(void *data) |
1da177e4 | 5367 | { |
6811378e | 5368 | int nid; |
afb6ebb3 | 5369 | int __maybe_unused cpu; |
9adb62a5 | 5370 | pg_data_t *self = data; |
1007843a | 5371 | unsigned long flags; |
b93e0f32 | 5372 | |
1007843a TH |
5373 | /* |
5374 | * Explicitly disable this CPU's interrupts before taking seqlock | |
5375 | * to prevent any IRQ handler from calling into the page allocator | |
5376 | * (e.g. GFP_ATOMIC) that could hit zonelist_iter_begin and livelock. | |
5377 | */ | |
5378 | local_irq_save(flags); | |
5379 | /* | |
5380 | * Explicitly disable this CPU's synchronous printk() before taking | |
5381 | * seqlock to prevent any printk() from trying to hold port->lock, for | |
5382 | * tty_insert_flip_string_and_push_buffer() on other CPU might be | |
5383 | * calling kmalloc(GFP_ATOMIC | __GFP_NOWARN) with port->lock held. | |
5384 | */ | |
5385 | printk_deferred_enter(); | |
3d36424b | 5386 | write_seqlock(&zonelist_update_seq); |
9276b1bc | 5387 | |
7f9cfb31 BL |
5388 | #ifdef CONFIG_NUMA |
5389 | memset(node_load, 0, sizeof(node_load)); | |
5390 | #endif | |
9adb62a5 | 5391 | |
c1152583 WY |
5392 | /* |
5393 | * This node is hotadded and no memory is yet present. So just | |
5394 | * building zonelists is fine - no need to touch other nodes. | |
5395 | */ | |
9adb62a5 JL |
5396 | if (self && !node_online(self->node_id)) { |
5397 | build_zonelists(self); | |
c1152583 | 5398 | } else { |
09f49dca MH |
5399 | /* |
5400 | * All possible nodes have pgdat preallocated | |
5401 | * in free_area_init | |
5402 | */ | |
5403 | for_each_node(nid) { | |
c1152583 | 5404 | pg_data_t *pgdat = NODE_DATA(nid); |
7ea1530a | 5405 | |
c1152583 WY |
5406 | build_zonelists(pgdat); |
5407 | } | |
99dcc3e5 | 5408 | |
7aac7898 LS |
5409 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5410 | /* | |
5411 | * We now know the "local memory node" for each node-- | |
5412 | * i.e., the node of the first zone in the generic zonelist. | |
5413 | * Set up numa_mem percpu variable for on-line cpus. During | |
5414 | * boot, only the boot cpu should be on-line; we'll init the | |
5415 | * secondary cpus' numa_mem as they come on-line. During | |
5416 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5417 | */ | |
d9c9a0b9 | 5418 | for_each_online_cpu(cpu) |
7aac7898 | 5419 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5420 | #endif |
d9c9a0b9 | 5421 | } |
b93e0f32 | 5422 | |
3d36424b | 5423 | write_sequnlock(&zonelist_update_seq); |
1007843a TH |
5424 | printk_deferred_exit(); |
5425 | local_irq_restore(flags); | |
6811378e YG |
5426 | } |
5427 | ||
061f67bc RV |
5428 | static noinline void __init |
5429 | build_all_zonelists_init(void) | |
5430 | { | |
afb6ebb3 MH |
5431 | int cpu; |
5432 | ||
061f67bc | 5433 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5434 | |
5435 | /* | |
5436 | * Initialize the boot_pagesets that are going to be used | |
5437 | * for bootstrapping processors. The real pagesets for | |
5438 | * each zone will be allocated later when the per cpu | |
5439 | * allocator is available. | |
5440 | * | |
5441 | * boot_pagesets are used also for bootstrapping offline | |
5442 | * cpus if the system is already booted because the pagesets | |
5443 | * are needed to initialize allocators on a specific cpu too. | |
5444 | * F.e. the percpu allocator needs the page allocator which | |
5445 | * needs the percpu allocator in order to allocate its pagesets | |
5446 | * (a chicken-egg dilemma). | |
5447 | */ | |
5448 | for_each_possible_cpu(cpu) | |
28f836b6 | 5449 | per_cpu_pages_init(&per_cpu(boot_pageset, cpu), &per_cpu(boot_zonestats, cpu)); |
afb6ebb3 | 5450 | |
061f67bc RV |
5451 | mminit_verify_zonelist(); |
5452 | cpuset_init_current_mems_allowed(); | |
5453 | } | |
5454 | ||
4eaf3f64 | 5455 | /* |
4eaf3f64 | 5456 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5457 | * |
72675e13 | 5458 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5459 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5460 | */ |
72675e13 | 5461 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 5462 | { |
0a18e607 DH |
5463 | unsigned long vm_total_pages; |
5464 | ||
6811378e | 5465 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 5466 | build_all_zonelists_init(); |
6811378e | 5467 | } else { |
11cd8638 | 5468 | __build_all_zonelists(pgdat); |
6811378e YG |
5469 | /* cpuset refresh routine should be here */ |
5470 | } | |
56b9413b DH |
5471 | /* Get the number of free pages beyond high watermark in all zones. */ |
5472 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
5473 | /* |
5474 | * Disable grouping by mobility if the number of pages in the | |
5475 | * system is too low to allow the mechanism to work. It would be | |
5476 | * more accurate, but expensive to check per-zone. This check is | |
5477 | * made on memory-hotadd so a system can start with mobility | |
5478 | * disabled and enable it later | |
5479 | */ | |
d9c23400 | 5480 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5481 | page_group_by_mobility_disabled = 1; |
5482 | else | |
5483 | page_group_by_mobility_disabled = 0; | |
5484 | ||
ce0725f7 | 5485 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5486 | nr_online_nodes, |
756a025f JP |
5487 | page_group_by_mobility_disabled ? "off" : "on", |
5488 | vm_total_pages); | |
f0c0b2b8 | 5489 | #ifdef CONFIG_NUMA |
f88dfff5 | 5490 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5491 | #endif |
1da177e4 LT |
5492 | } |
5493 | ||
9420f89d | 5494 | static int zone_batchsize(struct zone *zone) |
1da177e4 | 5495 | { |
9420f89d MRI |
5496 | #ifdef CONFIG_MMU |
5497 | int batch; | |
1da177e4 | 5498 | |
9420f89d MRI |
5499 | /* |
5500 | * The number of pages to batch allocate is either ~0.1% | |
5501 | * of the zone or 1MB, whichever is smaller. The batch | |
5502 | * size is striking a balance between allocation latency | |
5503 | * and zone lock contention. | |
5504 | */ | |
5505 | batch = min(zone_managed_pages(zone) >> 10, SZ_1M / PAGE_SIZE); | |
5506 | batch /= 4; /* We effectively *= 4 below */ | |
5507 | if (batch < 1) | |
5508 | batch = 1; | |
22b31eec | 5509 | |
4b94ffdc | 5510 | /* |
9420f89d MRI |
5511 | * Clamp the batch to a 2^n - 1 value. Having a power |
5512 | * of 2 value was found to be more likely to have | |
5513 | * suboptimal cache aliasing properties in some cases. | |
5514 | * | |
5515 | * For example if 2 tasks are alternately allocating | |
5516 | * batches of pages, one task can end up with a lot | |
5517 | * of pages of one half of the possible page colors | |
5518 | * and the other with pages of the other colors. | |
4b94ffdc | 5519 | */ |
9420f89d | 5520 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
966cf44f | 5521 | |
9420f89d | 5522 | return batch; |
3a6be87f DH |
5523 | |
5524 | #else | |
5525 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5526 | * conditions. | |
5527 | * | |
5528 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5529 | * of contiguous memory as there's no hardware page translation to | |
5530 | * assemble apparent contiguous memory from discontiguous pages. | |
5531 | * | |
5532 | * Queueing large contiguous runs of pages for batching, however, | |
5533 | * causes the pages to actually be freed in smaller chunks. As there | |
5534 | * can be a significant delay between the individual batches being | |
5535 | * recycled, this leads to the once large chunks of space being | |
5536 | * fragmented and becoming unavailable for high-order allocations. | |
5537 | */ | |
5538 | return 0; | |
5539 | #endif | |
e7c8d5c9 CL |
5540 | } |
5541 | ||
04f8cfea | 5542 | static int zone_highsize(struct zone *zone, int batch, int cpu_online) |
b92ca18e | 5543 | { |
9420f89d MRI |
5544 | #ifdef CONFIG_MMU |
5545 | int high; | |
5546 | int nr_split_cpus; | |
5547 | unsigned long total_pages; | |
c13291a5 | 5548 | |
9420f89d | 5549 | if (!percpu_pagelist_high_fraction) { |
2a1e274a | 5550 | /* |
9420f89d MRI |
5551 | * By default, the high value of the pcp is based on the zone |
5552 | * low watermark so that if they are full then background | |
5553 | * reclaim will not be started prematurely. | |
2a1e274a | 5554 | */ |
9420f89d MRI |
5555 | total_pages = low_wmark_pages(zone); |
5556 | } else { | |
2a1e274a | 5557 | /* |
9420f89d MRI |
5558 | * If percpu_pagelist_high_fraction is configured, the high |
5559 | * value is based on a fraction of the managed pages in the | |
5560 | * zone. | |
2a1e274a | 5561 | */ |
9420f89d | 5562 | total_pages = zone_managed_pages(zone) / percpu_pagelist_high_fraction; |
2a1e274a MG |
5563 | } |
5564 | ||
5565 | /* | |
9420f89d MRI |
5566 | * Split the high value across all online CPUs local to the zone. Note |
5567 | * that early in boot that CPUs may not be online yet and that during | |
5568 | * CPU hotplug that the cpumask is not yet updated when a CPU is being | |
5569 | * onlined. For memory nodes that have no CPUs, split pcp->high across | |
5570 | * all online CPUs to mitigate the risk that reclaim is triggered | |
5571 | * prematurely due to pages stored on pcp lists. | |
2a1e274a | 5572 | */ |
9420f89d MRI |
5573 | nr_split_cpus = cpumask_weight(cpumask_of_node(zone_to_nid(zone))) + cpu_online; |
5574 | if (!nr_split_cpus) | |
5575 | nr_split_cpus = num_online_cpus(); | |
5576 | high = total_pages / nr_split_cpus; | |
2a1e274a | 5577 | |
9420f89d MRI |
5578 | /* |
5579 | * Ensure high is at least batch*4. The multiple is based on the | |
5580 | * historical relationship between high and batch. | |
5581 | */ | |
5582 | high = max(high, batch << 2); | |
37b07e41 | 5583 | |
9420f89d MRI |
5584 | return high; |
5585 | #else | |
5586 | return 0; | |
5587 | #endif | |
37b07e41 LS |
5588 | } |
5589 | ||
51930df5 | 5590 | /* |
9420f89d MRI |
5591 | * pcp->high and pcp->batch values are related and generally batch is lower |
5592 | * than high. They are also related to pcp->count such that count is lower | |
5593 | * than high, and as soon as it reaches high, the pcplist is flushed. | |
5594 | * | |
5595 | * However, guaranteeing these relations at all times would require e.g. write | |
5596 | * barriers here but also careful usage of read barriers at the read side, and | |
5597 | * thus be prone to error and bad for performance. Thus the update only prevents | |
5598 | * store tearing. Any new users of pcp->batch and pcp->high should ensure they | |
5599 | * can cope with those fields changing asynchronously, and fully trust only the | |
5600 | * pcp->count field on the local CPU with interrupts disabled. | |
5601 | * | |
5602 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5603 | * outside of boot time (or some other assurance that no concurrent updaters | |
5604 | * exist). | |
51930df5 | 5605 | */ |
9420f89d MRI |
5606 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, |
5607 | unsigned long batch) | |
51930df5 | 5608 | { |
9420f89d MRI |
5609 | WRITE_ONCE(pcp->batch, batch); |
5610 | WRITE_ONCE(pcp->high, high); | |
51930df5 MR |
5611 | } |
5612 | ||
9420f89d | 5613 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats) |
c713216d | 5614 | { |
9420f89d | 5615 | int pindex; |
90cae1fe | 5616 | |
9420f89d MRI |
5617 | memset(pcp, 0, sizeof(*pcp)); |
5618 | memset(pzstats, 0, sizeof(*pzstats)); | |
90cae1fe | 5619 | |
9420f89d MRI |
5620 | spin_lock_init(&pcp->lock); |
5621 | for (pindex = 0; pindex < NR_PCP_LISTS; pindex++) | |
5622 | INIT_LIST_HEAD(&pcp->lists[pindex]); | |
2a1e274a | 5623 | |
9420f89d MRI |
5624 | /* |
5625 | * Set batch and high values safe for a boot pageset. A true percpu | |
5626 | * pageset's initialization will update them subsequently. Here we don't | |
5627 | * need to be as careful as pageset_update() as nobody can access the | |
5628 | * pageset yet. | |
5629 | */ | |
5630 | pcp->high = BOOT_PAGESET_HIGH; | |
5631 | pcp->batch = BOOT_PAGESET_BATCH; | |
5632 | pcp->free_factor = 0; | |
5633 | } | |
c713216d | 5634 | |
9420f89d MRI |
5635 | static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high, |
5636 | unsigned long batch) | |
5637 | { | |
5638 | struct per_cpu_pages *pcp; | |
5639 | int cpu; | |
2a1e274a | 5640 | |
9420f89d MRI |
5641 | for_each_possible_cpu(cpu) { |
5642 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); | |
5643 | pageset_update(pcp, high, batch); | |
2a1e274a | 5644 | } |
9420f89d | 5645 | } |
c713216d | 5646 | |
9420f89d MRI |
5647 | /* |
5648 | * Calculate and set new high and batch values for all per-cpu pagesets of a | |
5649 | * zone based on the zone's size. | |
5650 | */ | |
5651 | static void zone_set_pageset_high_and_batch(struct zone *zone, int cpu_online) | |
5652 | { | |
5653 | int new_high, new_batch; | |
09f49dca | 5654 | |
9420f89d MRI |
5655 | new_batch = max(1, zone_batchsize(zone)); |
5656 | new_high = zone_highsize(zone, new_batch, cpu_online); | |
09f49dca | 5657 | |
9420f89d MRI |
5658 | if (zone->pageset_high == new_high && |
5659 | zone->pageset_batch == new_batch) | |
5660 | return; | |
37b07e41 | 5661 | |
9420f89d MRI |
5662 | zone->pageset_high = new_high; |
5663 | zone->pageset_batch = new_batch; | |
122e093c | 5664 | |
9420f89d | 5665 | __zone_set_pageset_high_and_batch(zone, new_high, new_batch); |
c713216d | 5666 | } |
2a1e274a | 5667 | |
9420f89d | 5668 | void __meminit setup_zone_pageset(struct zone *zone) |
2a1e274a | 5669 | { |
9420f89d | 5670 | int cpu; |
2a1e274a | 5671 | |
9420f89d MRI |
5672 | /* Size may be 0 on !SMP && !NUMA */ |
5673 | if (sizeof(struct per_cpu_zonestat) > 0) | |
5674 | zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat); | |
2a1e274a | 5675 | |
9420f89d MRI |
5676 | zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages); |
5677 | for_each_possible_cpu(cpu) { | |
5678 | struct per_cpu_pages *pcp; | |
5679 | struct per_cpu_zonestat *pzstats; | |
2a1e274a | 5680 | |
9420f89d MRI |
5681 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
5682 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); | |
5683 | per_cpu_pages_init(pcp, pzstats); | |
a5c6d650 | 5684 | } |
9420f89d MRI |
5685 | |
5686 | zone_set_pageset_high_and_batch(zone, 0); | |
2a1e274a | 5687 | } |
ed7ed365 | 5688 | |
7e63efef | 5689 | /* |
9420f89d MRI |
5690 | * The zone indicated has a new number of managed_pages; batch sizes and percpu |
5691 | * page high values need to be recalculated. | |
7e63efef | 5692 | */ |
9420f89d | 5693 | static void zone_pcp_update(struct zone *zone, int cpu_online) |
7e63efef | 5694 | { |
9420f89d MRI |
5695 | mutex_lock(&pcp_batch_high_lock); |
5696 | zone_set_pageset_high_and_batch(zone, cpu_online); | |
5697 | mutex_unlock(&pcp_batch_high_lock); | |
7e63efef MG |
5698 | } |
5699 | ||
5700 | /* | |
9420f89d MRI |
5701 | * Allocate per cpu pagesets and initialize them. |
5702 | * Before this call only boot pagesets were available. | |
7e63efef | 5703 | */ |
9420f89d | 5704 | void __init setup_per_cpu_pageset(void) |
7e63efef | 5705 | { |
9420f89d MRI |
5706 | struct pglist_data *pgdat; |
5707 | struct zone *zone; | |
5708 | int __maybe_unused cpu; | |
5709 | ||
5710 | for_each_populated_zone(zone) | |
5711 | setup_zone_pageset(zone); | |
5712 | ||
5713 | #ifdef CONFIG_NUMA | |
5714 | /* | |
5715 | * Unpopulated zones continue using the boot pagesets. | |
5716 | * The numa stats for these pagesets need to be reset. | |
5717 | * Otherwise, they will end up skewing the stats of | |
5718 | * the nodes these zones are associated with. | |
5719 | */ | |
5720 | for_each_possible_cpu(cpu) { | |
5721 | struct per_cpu_zonestat *pzstats = &per_cpu(boot_zonestats, cpu); | |
5722 | memset(pzstats->vm_numa_event, 0, | |
5723 | sizeof(pzstats->vm_numa_event)); | |
5724 | } | |
5725 | #endif | |
5726 | ||
5727 | for_each_online_pgdat(pgdat) | |
5728 | pgdat->per_cpu_nodestats = | |
5729 | alloc_percpu(struct per_cpu_nodestat); | |
7e63efef MG |
5730 | } |
5731 | ||
9420f89d MRI |
5732 | __meminit void zone_pcp_init(struct zone *zone) |
5733 | { | |
5734 | /* | |
5735 | * per cpu subsystem is not up at this point. The following code | |
5736 | * relies on the ability of the linker to provide the | |
5737 | * offset of a (static) per cpu variable into the per cpu area. | |
5738 | */ | |
5739 | zone->per_cpu_pageset = &boot_pageset; | |
5740 | zone->per_cpu_zonestats = &boot_zonestats; | |
5741 | zone->pageset_high = BOOT_PAGESET_HIGH; | |
5742 | zone->pageset_batch = BOOT_PAGESET_BATCH; | |
5743 | ||
5744 | if (populated_zone(zone)) | |
5745 | pr_debug(" %s zone: %lu pages, LIFO batch:%u\n", zone->name, | |
5746 | zone->present_pages, zone_batchsize(zone)); | |
5747 | } | |
ed7ed365 | 5748 | |
c3d5f5f0 JL |
5749 | void adjust_managed_page_count(struct page *page, long count) |
5750 | { | |
9705bea5 | 5751 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 5752 | totalram_pages_add(count); |
3dcc0571 JL |
5753 | #ifdef CONFIG_HIGHMEM |
5754 | if (PageHighMem(page)) | |
ca79b0c2 | 5755 | totalhigh_pages_add(count); |
3dcc0571 | 5756 | #endif |
c3d5f5f0 | 5757 | } |
3dcc0571 | 5758 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 5759 | |
e5cb113f | 5760 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 5761 | { |
11199692 JL |
5762 | void *pos; |
5763 | unsigned long pages = 0; | |
69afade7 | 5764 | |
11199692 JL |
5765 | start = (void *)PAGE_ALIGN((unsigned long)start); |
5766 | end = (void *)((unsigned long)end & PAGE_MASK); | |
5767 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
5768 | struct page *page = virt_to_page(pos); |
5769 | void *direct_map_addr; | |
5770 | ||
5771 | /* | |
5772 | * 'direct_map_addr' might be different from 'pos' | |
5773 | * because some architectures' virt_to_page() | |
5774 | * work with aliases. Getting the direct map | |
5775 | * address ensures that we get a _writeable_ | |
5776 | * alias for the memset(). | |
5777 | */ | |
5778 | direct_map_addr = page_address(page); | |
c746170d VF |
5779 | /* |
5780 | * Perform a kasan-unchecked memset() since this memory | |
5781 | * has not been initialized. | |
5782 | */ | |
5783 | direct_map_addr = kasan_reset_tag(direct_map_addr); | |
dbe67df4 | 5784 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
5785 | memset(direct_map_addr, poison, PAGE_SIZE); |
5786 | ||
5787 | free_reserved_page(page); | |
69afade7 JL |
5788 | } |
5789 | ||
5790 | if (pages && s) | |
ff7ed9e4 | 5791 | pr_info("Freeing %s memory: %ldK\n", s, K(pages)); |
69afade7 JL |
5792 | |
5793 | return pages; | |
5794 | } | |
5795 | ||
005fd4bb | 5796 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 5797 | { |
04f8cfea | 5798 | struct zone *zone; |
1da177e4 | 5799 | |
005fd4bb | 5800 | lru_add_drain_cpu(cpu); |
96f97c43 | 5801 | mlock_drain_remote(cpu); |
005fd4bb | 5802 | drain_pages(cpu); |
9f8f2172 | 5803 | |
005fd4bb SAS |
5804 | /* |
5805 | * Spill the event counters of the dead processor | |
5806 | * into the current processors event counters. | |
5807 | * This artificially elevates the count of the current | |
5808 | * processor. | |
5809 | */ | |
5810 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 5811 | |
005fd4bb SAS |
5812 | /* |
5813 | * Zero the differential counters of the dead processor | |
5814 | * so that the vm statistics are consistent. | |
5815 | * | |
5816 | * This is only okay since the processor is dead and cannot | |
5817 | * race with what we are doing. | |
5818 | */ | |
5819 | cpu_vm_stats_fold(cpu); | |
04f8cfea MG |
5820 | |
5821 | for_each_populated_zone(zone) | |
5822 | zone_pcp_update(zone, 0); | |
5823 | ||
5824 | return 0; | |
5825 | } | |
5826 | ||
5827 | static int page_alloc_cpu_online(unsigned int cpu) | |
5828 | { | |
5829 | struct zone *zone; | |
5830 | ||
5831 | for_each_populated_zone(zone) | |
5832 | zone_pcp_update(zone, 1); | |
005fd4bb | 5833 | return 0; |
1da177e4 | 5834 | } |
1da177e4 | 5835 | |
c4fbed4b | 5836 | void __init page_alloc_init_cpuhp(void) |
1da177e4 | 5837 | { |
005fd4bb SAS |
5838 | int ret; |
5839 | ||
04f8cfea MG |
5840 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC, |
5841 | "mm/page_alloc:pcp", | |
5842 | page_alloc_cpu_online, | |
005fd4bb SAS |
5843 | page_alloc_cpu_dead); |
5844 | WARN_ON(ret < 0); | |
1da177e4 LT |
5845 | } |
5846 | ||
cb45b0e9 | 5847 | /* |
34b10060 | 5848 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
5849 | * or min_free_kbytes changes. |
5850 | */ | |
5851 | static void calculate_totalreserve_pages(void) | |
5852 | { | |
5853 | struct pglist_data *pgdat; | |
5854 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5855 | enum zone_type i, j; |
cb45b0e9 HA |
5856 | |
5857 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
5858 | |
5859 | pgdat->totalreserve_pages = 0; | |
5860 | ||
cb45b0e9 HA |
5861 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5862 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 5863 | long max = 0; |
9705bea5 | 5864 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
5865 | |
5866 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5867 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5868 | if (zone->lowmem_reserve[j] > max) | |
5869 | max = zone->lowmem_reserve[j]; | |
5870 | } | |
5871 | ||
41858966 MG |
5872 | /* we treat the high watermark as reserved pages. */ |
5873 | max += high_wmark_pages(zone); | |
cb45b0e9 | 5874 | |
3d6357de AK |
5875 | if (max > managed_pages) |
5876 | max = managed_pages; | |
a8d01437 | 5877 | |
281e3726 | 5878 | pgdat->totalreserve_pages += max; |
a8d01437 | 5879 | |
cb45b0e9 HA |
5880 | reserve_pages += max; |
5881 | } | |
5882 | } | |
5883 | totalreserve_pages = reserve_pages; | |
5884 | } | |
5885 | ||
1da177e4 LT |
5886 | /* |
5887 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 5888 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
5889 | * has a correct pages reserved value, so an adequate number of |
5890 | * pages are left in the zone after a successful __alloc_pages(). | |
5891 | */ | |
5892 | static void setup_per_zone_lowmem_reserve(void) | |
5893 | { | |
5894 | struct pglist_data *pgdat; | |
470c61d7 | 5895 | enum zone_type i, j; |
1da177e4 | 5896 | |
ec936fc5 | 5897 | for_each_online_pgdat(pgdat) { |
470c61d7 LS |
5898 | for (i = 0; i < MAX_NR_ZONES - 1; i++) { |
5899 | struct zone *zone = &pgdat->node_zones[i]; | |
5900 | int ratio = sysctl_lowmem_reserve_ratio[i]; | |
5901 | bool clear = !ratio || !zone_managed_pages(zone); | |
5902 | unsigned long managed_pages = 0; | |
5903 | ||
5904 | for (j = i + 1; j < MAX_NR_ZONES; j++) { | |
f7ec1044 LS |
5905 | struct zone *upper_zone = &pgdat->node_zones[j]; |
5906 | ||
5907 | managed_pages += zone_managed_pages(upper_zone); | |
470c61d7 | 5908 | |
f7ec1044 LS |
5909 | if (clear) |
5910 | zone->lowmem_reserve[j] = 0; | |
5911 | else | |
470c61d7 | 5912 | zone->lowmem_reserve[j] = managed_pages / ratio; |
1da177e4 LT |
5913 | } |
5914 | } | |
5915 | } | |
cb45b0e9 HA |
5916 | |
5917 | /* update totalreserve_pages */ | |
5918 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5919 | } |
5920 | ||
cfd3da1e | 5921 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5922 | { |
5923 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5924 | unsigned long lowmem_pages = 0; | |
5925 | struct zone *zone; | |
5926 | unsigned long flags; | |
5927 | ||
5928 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5929 | for_each_zone(zone) { | |
5930 | if (!is_highmem(zone)) | |
9705bea5 | 5931 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
5932 | } |
5933 | ||
5934 | for_each_zone(zone) { | |
ac924c60 AM |
5935 | u64 tmp; |
5936 | ||
1125b4e3 | 5937 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 5938 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 5939 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
5940 | if (is_highmem(zone)) { |
5941 | /* | |
669ed175 NP |
5942 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5943 | * need highmem pages, so cap pages_min to a small | |
5944 | * value here. | |
5945 | * | |
41858966 | 5946 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 5947 | * deltas control async page reclaim, and so should |
669ed175 | 5948 | * not be capped for highmem. |
1da177e4 | 5949 | */ |
90ae8d67 | 5950 | unsigned long min_pages; |
1da177e4 | 5951 | |
9705bea5 | 5952 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 5953 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 5954 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5955 | } else { |
669ed175 NP |
5956 | /* |
5957 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5958 | * proportionate to the zone's size. |
5959 | */ | |
a9214443 | 5960 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5961 | } |
5962 | ||
795ae7a0 JW |
5963 | /* |
5964 | * Set the kswapd watermarks distance according to the | |
5965 | * scale factor in proportion to available memory, but | |
5966 | * ensure a minimum size on small systems. | |
5967 | */ | |
5968 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 5969 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
5970 | watermark_scale_factor, 10000)); |
5971 | ||
aa092591 | 5972 | zone->watermark_boost = 0; |
a9214443 | 5973 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
c574bbe9 HY |
5974 | zone->_watermark[WMARK_HIGH] = low_wmark_pages(zone) + tmp; |
5975 | zone->_watermark[WMARK_PROMO] = high_wmark_pages(zone) + tmp; | |
49f223a9 | 5976 | |
1125b4e3 | 5977 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5978 | } |
cb45b0e9 HA |
5979 | |
5980 | /* update totalreserve_pages */ | |
5981 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5982 | } |
5983 | ||
cfd3da1e MG |
5984 | /** |
5985 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5986 | * or when memory is hot-{added|removed} | |
5987 | * | |
5988 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5989 | * correctly with respect to min_free_kbytes. | |
5990 | */ | |
5991 | void setup_per_zone_wmarks(void) | |
5992 | { | |
b92ca18e | 5993 | struct zone *zone; |
b93e0f32 MH |
5994 | static DEFINE_SPINLOCK(lock); |
5995 | ||
5996 | spin_lock(&lock); | |
cfd3da1e | 5997 | __setup_per_zone_wmarks(); |
b93e0f32 | 5998 | spin_unlock(&lock); |
b92ca18e MG |
5999 | |
6000 | /* | |
6001 | * The watermark size have changed so update the pcpu batch | |
6002 | * and high limits or the limits may be inappropriate. | |
6003 | */ | |
6004 | for_each_zone(zone) | |
04f8cfea | 6005 | zone_pcp_update(zone, 0); |
cfd3da1e MG |
6006 | } |
6007 | ||
1da177e4 LT |
6008 | /* |
6009 | * Initialise min_free_kbytes. | |
6010 | * | |
6011 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 6012 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
6013 | * bandwidth does not increase linearly with machine size. We use |
6014 | * | |
b8af2941 | 6015 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6016 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6017 | * | |
6018 | * which yields | |
6019 | * | |
6020 | * 16MB: 512k | |
6021 | * 32MB: 724k | |
6022 | * 64MB: 1024k | |
6023 | * 128MB: 1448k | |
6024 | * 256MB: 2048k | |
6025 | * 512MB: 2896k | |
6026 | * 1024MB: 4096k | |
6027 | * 2048MB: 5792k | |
6028 | * 4096MB: 8192k | |
6029 | * 8192MB: 11584k | |
6030 | * 16384MB: 16384k | |
6031 | */ | |
bd3400ea | 6032 | void calculate_min_free_kbytes(void) |
1da177e4 LT |
6033 | { |
6034 | unsigned long lowmem_kbytes; | |
5f12733e | 6035 | int new_min_free_kbytes; |
1da177e4 LT |
6036 | |
6037 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6038 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6039 | ||
59d336bd WS |
6040 | if (new_min_free_kbytes > user_min_free_kbytes) |
6041 | min_free_kbytes = clamp(new_min_free_kbytes, 128, 262144); | |
6042 | else | |
5f12733e MH |
6043 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", |
6044 | new_min_free_kbytes, user_min_free_kbytes); | |
59d336bd | 6045 | |
bd3400ea LF |
6046 | } |
6047 | ||
6048 | int __meminit init_per_zone_wmark_min(void) | |
6049 | { | |
6050 | calculate_min_free_kbytes(); | |
bc75d33f | 6051 | setup_per_zone_wmarks(); |
a6cccdc3 | 6052 | refresh_zone_stat_thresholds(); |
1da177e4 | 6053 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
6054 | |
6055 | #ifdef CONFIG_NUMA | |
6056 | setup_min_unmapped_ratio(); | |
6057 | setup_min_slab_ratio(); | |
6058 | #endif | |
6059 | ||
4aab2be0 VB |
6060 | khugepaged_min_free_kbytes_update(); |
6061 | ||
1da177e4 LT |
6062 | return 0; |
6063 | } | |
e08d3fdf | 6064 | postcore_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
6065 | |
6066 | /* | |
b8af2941 | 6067 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6068 | * that we can call two helper functions whenever min_free_kbytes |
6069 | * changes. | |
6070 | */ | |
cccad5b9 | 6071 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 6072 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6073 | { |
da8c757b HP |
6074 | int rc; |
6075 | ||
6076 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6077 | if (rc) | |
6078 | return rc; | |
6079 | ||
5f12733e MH |
6080 | if (write) { |
6081 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6082 | setup_per_zone_wmarks(); |
5f12733e | 6083 | } |
1da177e4 LT |
6084 | return 0; |
6085 | } | |
6086 | ||
795ae7a0 | 6087 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 6088 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
6089 | { |
6090 | int rc; | |
6091 | ||
6092 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6093 | if (rc) | |
6094 | return rc; | |
6095 | ||
6096 | if (write) | |
6097 | setup_per_zone_wmarks(); | |
6098 | ||
6099 | return 0; | |
6100 | } | |
6101 | ||
9614634f | 6102 | #ifdef CONFIG_NUMA |
6423aa81 | 6103 | static void setup_min_unmapped_ratio(void) |
9614634f | 6104 | { |
6423aa81 | 6105 | pg_data_t *pgdat; |
9614634f | 6106 | struct zone *zone; |
9614634f | 6107 | |
a5f5f91d | 6108 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 6109 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 6110 | |
9614634f | 6111 | for_each_zone(zone) |
9705bea5 AK |
6112 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
6113 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 6114 | } |
0ff38490 | 6115 | |
6423aa81 JK |
6116 | |
6117 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 6118 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 6119 | { |
0ff38490 CL |
6120 | int rc; |
6121 | ||
8d65af78 | 6122 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
6123 | if (rc) |
6124 | return rc; | |
6125 | ||
6423aa81 JK |
6126 | setup_min_unmapped_ratio(); |
6127 | ||
6128 | return 0; | |
6129 | } | |
6130 | ||
6131 | static void setup_min_slab_ratio(void) | |
6132 | { | |
6133 | pg_data_t *pgdat; | |
6134 | struct zone *zone; | |
6135 | ||
a5f5f91d MG |
6136 | for_each_online_pgdat(pgdat) |
6137 | pgdat->min_slab_pages = 0; | |
6138 | ||
0ff38490 | 6139 | for_each_zone(zone) |
9705bea5 AK |
6140 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
6141 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
6142 | } |
6143 | ||
6144 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 6145 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
6146 | { |
6147 | int rc; | |
6148 | ||
6149 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6150 | if (rc) | |
6151 | return rc; | |
6152 | ||
6153 | setup_min_slab_ratio(); | |
6154 | ||
0ff38490 CL |
6155 | return 0; |
6156 | } | |
9614634f CL |
6157 | #endif |
6158 | ||
1da177e4 LT |
6159 | /* |
6160 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
6161 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
6162 | * whenever sysctl_lowmem_reserve_ratio changes. | |
6163 | * | |
6164 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 6165 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
6166 | * if in function of the boot time zone sizes. |
6167 | */ | |
cccad5b9 | 6168 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 6169 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6170 | { |
86aaf255 BH |
6171 | int i; |
6172 | ||
8d65af78 | 6173 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
6174 | |
6175 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
6176 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
6177 | sysctl_lowmem_reserve_ratio[i] = 0; | |
6178 | } | |
6179 | ||
1da177e4 LT |
6180 | setup_per_zone_lowmem_reserve(); |
6181 | return 0; | |
6182 | } | |
6183 | ||
8ad4b1fb | 6184 | /* |
74f44822 MG |
6185 | * percpu_pagelist_high_fraction - changes the pcp->high for each zone on each |
6186 | * cpu. It is the fraction of total pages in each zone that a hot per cpu | |
b8af2941 | 6187 | * pagelist can have before it gets flushed back to buddy allocator. |
8ad4b1fb | 6188 | */ |
74f44822 MG |
6189 | int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *table, |
6190 | int write, void *buffer, size_t *length, loff_t *ppos) | |
8ad4b1fb RS |
6191 | { |
6192 | struct zone *zone; | |
74f44822 | 6193 | int old_percpu_pagelist_high_fraction; |
8ad4b1fb RS |
6194 | int ret; |
6195 | ||
7cd2b0a3 | 6196 | mutex_lock(&pcp_batch_high_lock); |
74f44822 | 6197 | old_percpu_pagelist_high_fraction = percpu_pagelist_high_fraction; |
7cd2b0a3 | 6198 | |
8d65af78 | 6199 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
6200 | if (!write || ret < 0) |
6201 | goto out; | |
6202 | ||
6203 | /* Sanity checking to avoid pcp imbalance */ | |
74f44822 MG |
6204 | if (percpu_pagelist_high_fraction && |
6205 | percpu_pagelist_high_fraction < MIN_PERCPU_PAGELIST_HIGH_FRACTION) { | |
6206 | percpu_pagelist_high_fraction = old_percpu_pagelist_high_fraction; | |
7cd2b0a3 DR |
6207 | ret = -EINVAL; |
6208 | goto out; | |
6209 | } | |
6210 | ||
6211 | /* No change? */ | |
74f44822 | 6212 | if (percpu_pagelist_high_fraction == old_percpu_pagelist_high_fraction) |
7cd2b0a3 | 6213 | goto out; |
c8e251fa | 6214 | |
cb1ef534 | 6215 | for_each_populated_zone(zone) |
74f44822 | 6216 | zone_set_pageset_high_and_batch(zone, 0); |
7cd2b0a3 | 6217 | out: |
c8e251fa | 6218 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6219 | return ret; |
8ad4b1fb RS |
6220 | } |
6221 | ||
8df995f6 | 6222 | #ifdef CONFIG_CONTIG_ALLOC |
a1394bdd MK |
6223 | #if defined(CONFIG_DYNAMIC_DEBUG) || \ |
6224 | (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) | |
6225 | /* Usage: See admin-guide/dynamic-debug-howto.rst */ | |
6226 | static void alloc_contig_dump_pages(struct list_head *page_list) | |
6227 | { | |
6228 | DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure"); | |
6229 | ||
6230 | if (DYNAMIC_DEBUG_BRANCH(descriptor)) { | |
6231 | struct page *page; | |
6232 | ||
6233 | dump_stack(); | |
6234 | list_for_each_entry(page, page_list, lru) | |
6235 | dump_page(page, "migration failure"); | |
6236 | } | |
6237 | } | |
6238 | #else | |
6239 | static inline void alloc_contig_dump_pages(struct list_head *page_list) | |
6240 | { | |
6241 | } | |
6242 | #endif | |
6243 | ||
041d3a8c | 6244 | /* [start, end) must belong to a single zone. */ |
b2c9e2fb | 6245 | int __alloc_contig_migrate_range(struct compact_control *cc, |
bb13ffeb | 6246 | unsigned long start, unsigned long end) |
041d3a8c MN |
6247 | { |
6248 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 6249 | unsigned int nr_reclaimed; |
041d3a8c MN |
6250 | unsigned long pfn = start; |
6251 | unsigned int tries = 0; | |
6252 | int ret = 0; | |
8b94e0b8 JK |
6253 | struct migration_target_control mtc = { |
6254 | .nid = zone_to_nid(cc->zone), | |
6255 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
6256 | }; | |
041d3a8c | 6257 | |
361a2a22 | 6258 | lru_cache_disable(); |
041d3a8c | 6259 | |
bb13ffeb | 6260 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6261 | if (fatal_signal_pending(current)) { |
6262 | ret = -EINTR; | |
6263 | break; | |
6264 | } | |
6265 | ||
bb13ffeb MG |
6266 | if (list_empty(&cc->migratepages)) { |
6267 | cc->nr_migratepages = 0; | |
c2ad7a1f OS |
6268 | ret = isolate_migratepages_range(cc, pfn, end); |
6269 | if (ret && ret != -EAGAIN) | |
041d3a8c | 6270 | break; |
c2ad7a1f | 6271 | pfn = cc->migrate_pfn; |
041d3a8c MN |
6272 | tries = 0; |
6273 | } else if (++tries == 5) { | |
c8e28b47 | 6274 | ret = -EBUSY; |
041d3a8c MN |
6275 | break; |
6276 | } | |
6277 | ||
beb51eaa MK |
6278 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6279 | &cc->migratepages); | |
6280 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6281 | |
8b94e0b8 | 6282 | ret = migrate_pages(&cc->migratepages, alloc_migration_target, |
5ac95884 | 6283 | NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE, NULL); |
c8e28b47 OS |
6284 | |
6285 | /* | |
6286 | * On -ENOMEM, migrate_pages() bails out right away. It is pointless | |
6287 | * to retry again over this error, so do the same here. | |
6288 | */ | |
6289 | if (ret == -ENOMEM) | |
6290 | break; | |
041d3a8c | 6291 | } |
d479960e | 6292 | |
361a2a22 | 6293 | lru_cache_enable(); |
2a6f5124 | 6294 | if (ret < 0) { |
3f913fc5 | 6295 | if (!(cc->gfp_mask & __GFP_NOWARN) && ret == -EBUSY) |
151e084a | 6296 | alloc_contig_dump_pages(&cc->migratepages); |
2a6f5124 SP |
6297 | putback_movable_pages(&cc->migratepages); |
6298 | return ret; | |
6299 | } | |
6300 | return 0; | |
041d3a8c MN |
6301 | } |
6302 | ||
6303 | /** | |
6304 | * alloc_contig_range() -- tries to allocate given range of pages | |
6305 | * @start: start PFN to allocate | |
6306 | * @end: one-past-the-last PFN to allocate | |
f0953a1b | 6307 | * @migratetype: migratetype of the underlying pageblocks (either |
0815f3d8 MN |
6308 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
6309 | * in range must have the same migratetype and it must | |
6310 | * be either of the two. | |
ca96b625 | 6311 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c | 6312 | * |
11ac3e87 ZY |
6313 | * The PFN range does not have to be pageblock aligned. The PFN range must |
6314 | * belong to a single zone. | |
041d3a8c | 6315 | * |
2c7452a0 MK |
6316 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
6317 | * pageblocks in the range. Once isolated, the pageblocks should not | |
6318 | * be modified by others. | |
041d3a8c | 6319 | * |
a862f68a | 6320 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
6321 | * pages which PFN is in [start, end) are allocated for the caller and |
6322 | * need to be freed with free_contig_range(). | |
6323 | */ | |
0815f3d8 | 6324 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 6325 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 6326 | { |
041d3a8c | 6327 | unsigned long outer_start, outer_end; |
b2c9e2fb | 6328 | int order; |
d00181b9 | 6329 | int ret = 0; |
041d3a8c | 6330 | |
bb13ffeb MG |
6331 | struct compact_control cc = { |
6332 | .nr_migratepages = 0, | |
6333 | .order = -1, | |
6334 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 6335 | .mode = MIGRATE_SYNC, |
bb13ffeb | 6336 | .ignore_skip_hint = true, |
2583d671 | 6337 | .no_set_skip_hint = true, |
7dea19f9 | 6338 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 6339 | .alloc_contig = true, |
bb13ffeb MG |
6340 | }; |
6341 | INIT_LIST_HEAD(&cc.migratepages); | |
6342 | ||
041d3a8c MN |
6343 | /* |
6344 | * What we do here is we mark all pageblocks in range as | |
6345 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6346 | * have different sizes, and due to the way page allocator | |
b2c9e2fb | 6347 | * work, start_isolate_page_range() has special handlings for this. |
041d3a8c MN |
6348 | * |
6349 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6350 | * migrate the pages from an unaligned range (ie. pages that | |
b2c9e2fb | 6351 | * we are interested in). This will put all the pages in |
041d3a8c MN |
6352 | * range back to page allocator as MIGRATE_ISOLATE. |
6353 | * | |
6354 | * When this is done, we take the pages in range from page | |
6355 | * allocator removing them from the buddy system. This way | |
6356 | * page allocator will never consider using them. | |
6357 | * | |
6358 | * This lets us mark the pageblocks back as | |
6359 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6360 | * aligned range but not in the unaligned, original range are | |
6361 | * put back to page allocator so that buddy can use them. | |
6362 | */ | |
6363 | ||
6e263fff | 6364 | ret = start_isolate_page_range(start, end, migratetype, 0, gfp_mask); |
3fa0c7c7 | 6365 | if (ret) |
b2c9e2fb | 6366 | goto done; |
041d3a8c | 6367 | |
7612921f VB |
6368 | drain_all_pages(cc.zone); |
6369 | ||
8ef5849f JK |
6370 | /* |
6371 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
6372 | * So, just fall through. test_pages_isolated() has a tracepoint |
6373 | * which will report the busy page. | |
6374 | * | |
6375 | * It is possible that busy pages could become available before | |
6376 | * the call to test_pages_isolated, and the range will actually be | |
6377 | * allocated. So, if we fall through be sure to clear ret so that | |
6378 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 6379 | */ |
bb13ffeb | 6380 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 6381 | if (ret && ret != -EBUSY) |
041d3a8c | 6382 | goto done; |
68d68ff6 | 6383 | ret = 0; |
041d3a8c MN |
6384 | |
6385 | /* | |
b2c9e2fb | 6386 | * Pages from [start, end) are within a pageblock_nr_pages |
041d3a8c MN |
6387 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's |
6388 | * more, all pages in [start, end) are free in page allocator. | |
6389 | * What we are going to do is to allocate all pages from | |
6390 | * [start, end) (that is remove them from page allocator). | |
6391 | * | |
6392 | * The only problem is that pages at the beginning and at the | |
6393 | * end of interesting range may be not aligned with pages that | |
6394 | * page allocator holds, ie. they can be part of higher order | |
6395 | * pages. Because of this, we reserve the bigger range and | |
6396 | * once this is done free the pages we are not interested in. | |
6397 | * | |
6398 | * We don't have to hold zone->lock here because the pages are | |
6399 | * isolated thus they won't get removed from buddy. | |
6400 | */ | |
6401 | ||
041d3a8c MN |
6402 | order = 0; |
6403 | outer_start = start; | |
6404 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
23baf831 | 6405 | if (++order > MAX_ORDER) { |
8ef5849f JK |
6406 | outer_start = start; |
6407 | break; | |
041d3a8c MN |
6408 | } |
6409 | outer_start &= ~0UL << order; | |
6410 | } | |
6411 | ||
8ef5849f | 6412 | if (outer_start != start) { |
ab130f91 | 6413 | order = buddy_order(pfn_to_page(outer_start)); |
8ef5849f JK |
6414 | |
6415 | /* | |
6416 | * outer_start page could be small order buddy page and | |
6417 | * it doesn't include start page. Adjust outer_start | |
6418 | * in this case to report failed page properly | |
6419 | * on tracepoint in test_pages_isolated() | |
6420 | */ | |
6421 | if (outer_start + (1UL << order) <= start) | |
6422 | outer_start = start; | |
6423 | } | |
6424 | ||
041d3a8c | 6425 | /* Make sure the range is really isolated. */ |
756d25be | 6426 | if (test_pages_isolated(outer_start, end, 0)) { |
041d3a8c MN |
6427 | ret = -EBUSY; |
6428 | goto done; | |
6429 | } | |
6430 | ||
49f223a9 | 6431 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 6432 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6433 | if (!outer_end) { |
6434 | ret = -EBUSY; | |
6435 | goto done; | |
6436 | } | |
6437 | ||
6438 | /* Free head and tail (if any) */ | |
6439 | if (start != outer_start) | |
6440 | free_contig_range(outer_start, start - outer_start); | |
6441 | if (end != outer_end) | |
6442 | free_contig_range(end, outer_end - end); | |
6443 | ||
6444 | done: | |
6e263fff | 6445 | undo_isolate_page_range(start, end, migratetype); |
041d3a8c MN |
6446 | return ret; |
6447 | } | |
255f5985 | 6448 | EXPORT_SYMBOL(alloc_contig_range); |
5e27a2df AK |
6449 | |
6450 | static int __alloc_contig_pages(unsigned long start_pfn, | |
6451 | unsigned long nr_pages, gfp_t gfp_mask) | |
6452 | { | |
6453 | unsigned long end_pfn = start_pfn + nr_pages; | |
6454 | ||
6455 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
6456 | gfp_mask); | |
6457 | } | |
6458 | ||
6459 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
6460 | unsigned long nr_pages) | |
6461 | { | |
6462 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
6463 | struct page *page; | |
6464 | ||
6465 | for (i = start_pfn; i < end_pfn; i++) { | |
6466 | page = pfn_to_online_page(i); | |
6467 | if (!page) | |
6468 | return false; | |
6469 | ||
6470 | if (page_zone(page) != z) | |
6471 | return false; | |
6472 | ||
6473 | if (PageReserved(page)) | |
4d73ba5f MG |
6474 | return false; |
6475 | ||
6476 | if (PageHuge(page)) | |
5e27a2df | 6477 | return false; |
5e27a2df AK |
6478 | } |
6479 | return true; | |
6480 | } | |
6481 | ||
6482 | static bool zone_spans_last_pfn(const struct zone *zone, | |
6483 | unsigned long start_pfn, unsigned long nr_pages) | |
6484 | { | |
6485 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
6486 | ||
6487 | return zone_spans_pfn(zone, last_pfn); | |
6488 | } | |
6489 | ||
6490 | /** | |
6491 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
6492 | * @nr_pages: Number of contiguous pages to allocate | |
6493 | * @gfp_mask: GFP mask to limit search and used during compaction | |
6494 | * @nid: Target node | |
6495 | * @nodemask: Mask for other possible nodes | |
6496 | * | |
6497 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
6498 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
6499 | * tried for allocation with alloc_contig_range(). This routine is intended | |
6500 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
6501 | * | |
6502 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
eaab8e75 AK |
6503 | * power of two, then allocated range is also guaranteed to be aligned to same |
6504 | * nr_pages (e.g. 1GB request would be aligned to 1GB). | |
5e27a2df AK |
6505 | * |
6506 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
6507 | * __free_page() on each allocated page. | |
6508 | * | |
6509 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
6510 | */ | |
6511 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
6512 | int nid, nodemask_t *nodemask) | |
6513 | { | |
6514 | unsigned long ret, pfn, flags; | |
6515 | struct zonelist *zonelist; | |
6516 | struct zone *zone; | |
6517 | struct zoneref *z; | |
6518 | ||
6519 | zonelist = node_zonelist(nid, gfp_mask); | |
6520 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
6521 | gfp_zone(gfp_mask), nodemask) { | |
6522 | spin_lock_irqsave(&zone->lock, flags); | |
6523 | ||
6524 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
6525 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
6526 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
6527 | /* | |
6528 | * We release the zone lock here because | |
6529 | * alloc_contig_range() will also lock the zone | |
6530 | * at some point. If there's an allocation | |
6531 | * spinning on this lock, it may win the race | |
6532 | * and cause alloc_contig_range() to fail... | |
6533 | */ | |
6534 | spin_unlock_irqrestore(&zone->lock, flags); | |
6535 | ret = __alloc_contig_pages(pfn, nr_pages, | |
6536 | gfp_mask); | |
6537 | if (!ret) | |
6538 | return pfn_to_page(pfn); | |
6539 | spin_lock_irqsave(&zone->lock, flags); | |
6540 | } | |
6541 | pfn += nr_pages; | |
6542 | } | |
6543 | spin_unlock_irqrestore(&zone->lock, flags); | |
6544 | } | |
6545 | return NULL; | |
6546 | } | |
4eb0716e | 6547 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 6548 | |
78fa5150 | 6549 | void free_contig_range(unsigned long pfn, unsigned long nr_pages) |
041d3a8c | 6550 | { |
78fa5150 | 6551 | unsigned long count = 0; |
bcc2b02f MS |
6552 | |
6553 | for (; nr_pages--; pfn++) { | |
6554 | struct page *page = pfn_to_page(pfn); | |
6555 | ||
6556 | count += page_count(page) != 1; | |
6557 | __free_page(page); | |
6558 | } | |
78fa5150 | 6559 | WARN(count != 0, "%lu pages are still in use!\n", count); |
041d3a8c | 6560 | } |
255f5985 | 6561 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 6562 | |
ec6e8c7e VB |
6563 | /* |
6564 | * Effectively disable pcplists for the zone by setting the high limit to 0 | |
6565 | * and draining all cpus. A concurrent page freeing on another CPU that's about | |
6566 | * to put the page on pcplist will either finish before the drain and the page | |
6567 | * will be drained, or observe the new high limit and skip the pcplist. | |
6568 | * | |
6569 | * Must be paired with a call to zone_pcp_enable(). | |
6570 | */ | |
6571 | void zone_pcp_disable(struct zone *zone) | |
6572 | { | |
6573 | mutex_lock(&pcp_batch_high_lock); | |
6574 | __zone_set_pageset_high_and_batch(zone, 0, 1); | |
6575 | __drain_all_pages(zone, true); | |
6576 | } | |
6577 | ||
6578 | void zone_pcp_enable(struct zone *zone) | |
6579 | { | |
6580 | __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch); | |
6581 | mutex_unlock(&pcp_batch_high_lock); | |
6582 | } | |
6583 | ||
340175b7 JL |
6584 | void zone_pcp_reset(struct zone *zone) |
6585 | { | |
5a883813 | 6586 | int cpu; |
28f836b6 | 6587 | struct per_cpu_zonestat *pzstats; |
340175b7 | 6588 | |
28f836b6 | 6589 | if (zone->per_cpu_pageset != &boot_pageset) { |
5a883813 | 6590 | for_each_online_cpu(cpu) { |
28f836b6 MG |
6591 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); |
6592 | drain_zonestat(zone, pzstats); | |
5a883813 | 6593 | } |
28f836b6 | 6594 | free_percpu(zone->per_cpu_pageset); |
28f836b6 | 6595 | zone->per_cpu_pageset = &boot_pageset; |
022e7fa0 ML |
6596 | if (zone->per_cpu_zonestats != &boot_zonestats) { |
6597 | free_percpu(zone->per_cpu_zonestats); | |
6598 | zone->per_cpu_zonestats = &boot_zonestats; | |
6599 | } | |
340175b7 | 6600 | } |
340175b7 JL |
6601 | } |
6602 | ||
6dcd73d7 | 6603 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 6604 | /* |
257bea71 DH |
6605 | * All pages in the range must be in a single zone, must not contain holes, |
6606 | * must span full sections, and must be isolated before calling this function. | |
0c0e6195 | 6607 | */ |
257bea71 | 6608 | void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 6609 | { |
257bea71 | 6610 | unsigned long pfn = start_pfn; |
0c0e6195 KH |
6611 | struct page *page; |
6612 | struct zone *zone; | |
0ee5f4f3 | 6613 | unsigned int order; |
0c0e6195 | 6614 | unsigned long flags; |
5557c766 | 6615 | |
2d070eab | 6616 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
6617 | zone = page_zone(pfn_to_page(pfn)); |
6618 | spin_lock_irqsave(&zone->lock, flags); | |
0c0e6195 | 6619 | while (pfn < end_pfn) { |
0c0e6195 | 6620 | page = pfn_to_page(pfn); |
b023f468 WC |
6621 | /* |
6622 | * The HWPoisoned page may be not in buddy system, and | |
6623 | * page_count() is not 0. | |
6624 | */ | |
6625 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6626 | pfn++; | |
b023f468 WC |
6627 | continue; |
6628 | } | |
aa218795 DH |
6629 | /* |
6630 | * At this point all remaining PageOffline() pages have a | |
6631 | * reference count of 0 and can simply be skipped. | |
6632 | */ | |
6633 | if (PageOffline(page)) { | |
6634 | BUG_ON(page_count(page)); | |
6635 | BUG_ON(PageBuddy(page)); | |
6636 | pfn++; | |
aa218795 DH |
6637 | continue; |
6638 | } | |
b023f468 | 6639 | |
0c0e6195 KH |
6640 | BUG_ON(page_count(page)); |
6641 | BUG_ON(!PageBuddy(page)); | |
ab130f91 | 6642 | order = buddy_order(page); |
6ab01363 | 6643 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
6644 | pfn += (1 << order); |
6645 | } | |
6646 | spin_unlock_irqrestore(&zone->lock, flags); | |
6647 | } | |
6648 | #endif | |
8d22ba1b | 6649 | |
8446b59b ED |
6650 | /* |
6651 | * This function returns a stable result only if called under zone lock. | |
6652 | */ | |
8d22ba1b WF |
6653 | bool is_free_buddy_page(struct page *page) |
6654 | { | |
8d22ba1b | 6655 | unsigned long pfn = page_to_pfn(page); |
7aeb09f9 | 6656 | unsigned int order; |
8d22ba1b | 6657 | |
23baf831 | 6658 | for (order = 0; order <= MAX_ORDER; order++) { |
8d22ba1b WF |
6659 | struct page *page_head = page - (pfn & ((1 << order) - 1)); |
6660 | ||
8446b59b ED |
6661 | if (PageBuddy(page_head) && |
6662 | buddy_order_unsafe(page_head) >= order) | |
8d22ba1b WF |
6663 | break; |
6664 | } | |
8d22ba1b | 6665 | |
23baf831 | 6666 | return order <= MAX_ORDER; |
8d22ba1b | 6667 | } |
a581865e | 6668 | EXPORT_SYMBOL(is_free_buddy_page); |
d4ae9916 NH |
6669 | |
6670 | #ifdef CONFIG_MEMORY_FAILURE | |
6671 | /* | |
06be6ff3 OS |
6672 | * Break down a higher-order page in sub-pages, and keep our target out of |
6673 | * buddy allocator. | |
d4ae9916 | 6674 | */ |
06be6ff3 OS |
6675 | static void break_down_buddy_pages(struct zone *zone, struct page *page, |
6676 | struct page *target, int low, int high, | |
6677 | int migratetype) | |
6678 | { | |
6679 | unsigned long size = 1 << high; | |
6680 | struct page *current_buddy, *next_page; | |
6681 | ||
6682 | while (high > low) { | |
6683 | high--; | |
6684 | size >>= 1; | |
6685 | ||
6686 | if (target >= &page[size]) { | |
6687 | next_page = page + size; | |
6688 | current_buddy = page; | |
6689 | } else { | |
6690 | next_page = page; | |
6691 | current_buddy = page + size; | |
6692 | } | |
6693 | ||
6694 | if (set_page_guard(zone, current_buddy, high, migratetype)) | |
6695 | continue; | |
6696 | ||
6697 | if (current_buddy != target) { | |
6698 | add_to_free_list(current_buddy, zone, high, migratetype); | |
ab130f91 | 6699 | set_buddy_order(current_buddy, high); |
06be6ff3 OS |
6700 | page = next_page; |
6701 | } | |
6702 | } | |
6703 | } | |
6704 | ||
6705 | /* | |
6706 | * Take a page that will be marked as poisoned off the buddy allocator. | |
6707 | */ | |
6708 | bool take_page_off_buddy(struct page *page) | |
d4ae9916 NH |
6709 | { |
6710 | struct zone *zone = page_zone(page); | |
6711 | unsigned long pfn = page_to_pfn(page); | |
6712 | unsigned long flags; | |
6713 | unsigned int order; | |
06be6ff3 | 6714 | bool ret = false; |
d4ae9916 NH |
6715 | |
6716 | spin_lock_irqsave(&zone->lock, flags); | |
23baf831 | 6717 | for (order = 0; order <= MAX_ORDER; order++) { |
d4ae9916 | 6718 | struct page *page_head = page - (pfn & ((1 << order) - 1)); |
ab130f91 | 6719 | int page_order = buddy_order(page_head); |
d4ae9916 | 6720 | |
ab130f91 | 6721 | if (PageBuddy(page_head) && page_order >= order) { |
06be6ff3 OS |
6722 | unsigned long pfn_head = page_to_pfn(page_head); |
6723 | int migratetype = get_pfnblock_migratetype(page_head, | |
6724 | pfn_head); | |
6725 | ||
ab130f91 | 6726 | del_page_from_free_list(page_head, zone, page_order); |
06be6ff3 | 6727 | break_down_buddy_pages(zone, page_head, page, 0, |
ab130f91 | 6728 | page_order, migratetype); |
bf181c58 | 6729 | SetPageHWPoisonTakenOff(page); |
bac9c6fa DH |
6730 | if (!is_migrate_isolate(migratetype)) |
6731 | __mod_zone_freepage_state(zone, -1, migratetype); | |
06be6ff3 | 6732 | ret = true; |
d4ae9916 NH |
6733 | break; |
6734 | } | |
06be6ff3 OS |
6735 | if (page_count(page_head) > 0) |
6736 | break; | |
d4ae9916 NH |
6737 | } |
6738 | spin_unlock_irqrestore(&zone->lock, flags); | |
06be6ff3 | 6739 | return ret; |
d4ae9916 | 6740 | } |
bf181c58 NH |
6741 | |
6742 | /* | |
6743 | * Cancel takeoff done by take_page_off_buddy(). | |
6744 | */ | |
6745 | bool put_page_back_buddy(struct page *page) | |
6746 | { | |
6747 | struct zone *zone = page_zone(page); | |
6748 | unsigned long pfn = page_to_pfn(page); | |
6749 | unsigned long flags; | |
6750 | int migratetype = get_pfnblock_migratetype(page, pfn); | |
6751 | bool ret = false; | |
6752 | ||
6753 | spin_lock_irqsave(&zone->lock, flags); | |
6754 | if (put_page_testzero(page)) { | |
6755 | ClearPageHWPoisonTakenOff(page); | |
6756 | __free_one_page(page, pfn, zone, 0, migratetype, FPI_NONE); | |
6757 | if (TestClearPageHWPoison(page)) { | |
bf181c58 NH |
6758 | ret = true; |
6759 | } | |
6760 | } | |
6761 | spin_unlock_irqrestore(&zone->lock, flags); | |
6762 | ||
6763 | return ret; | |
6764 | } | |
d4ae9916 | 6765 | #endif |
62b31070 BH |
6766 | |
6767 | #ifdef CONFIG_ZONE_DMA | |
6768 | bool has_managed_dma(void) | |
6769 | { | |
6770 | struct pglist_data *pgdat; | |
6771 | ||
6772 | for_each_online_pgdat(pgdat) { | |
6773 | struct zone *zone = &pgdat->node_zones[ZONE_DMA]; | |
6774 | ||
6775 | if (managed_zone(zone)) | |
6776 | return true; | |
6777 | } | |
6778 | return false; | |
6779 | } | |
6780 | #endif /* CONFIG_ZONE_DMA */ |