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> | |
a238ab5b | 29 | #include <linux/ratelimit.h> |
5a3135c2 | 30 | #include <linux/oom.h> |
1da177e4 LT |
31 | #include <linux/topology.h> |
32 | #include <linux/sysctl.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/cpuset.h> | |
90491d87 | 35 | #include <linux/pagevec.h> |
bdc8cb98 | 36 | #include <linux/memory_hotplug.h> |
1da177e4 | 37 | #include <linux/nodemask.h> |
a6cccdc3 | 38 | #include <linux/vmstat.h> |
933e312e | 39 | #include <linux/fault-inject.h> |
56de7263 | 40 | #include <linux/compaction.h> |
0d3d062a | 41 | #include <trace/events/kmem.h> |
d379f01d | 42 | #include <trace/events/oom.h> |
268bb0ce | 43 | #include <linux/prefetch.h> |
6e543d57 | 44 | #include <linux/mm_inline.h> |
f920e413 | 45 | #include <linux/mmu_notifier.h> |
041d3a8c | 46 | #include <linux/migrate.h> |
5b3cc15a | 47 | #include <linux/sched/mm.h> |
48c96a36 | 48 | #include <linux/page_owner.h> |
df4e817b | 49 | #include <linux/page_table_check.h> |
4949148a | 50 | #include <linux/memcontrol.h> |
42c269c8 | 51 | #include <linux/ftrace.h> |
d92a8cfc | 52 | #include <linux/lockdep.h> |
eb414681 | 53 | #include <linux/psi.h> |
4aab2be0 | 54 | #include <linux/khugepaged.h> |
5bf18281 | 55 | #include <linux/delayacct.h> |
362d37a1 | 56 | #include <linux/cacheinfo.h> |
dcfe378c | 57 | #include <linux/pgalloc_tag.h> |
ac924c60 | 58 | #include <asm/div64.h> |
1da177e4 | 59 | #include "internal.h" |
e900a918 | 60 | #include "shuffle.h" |
36e66c55 | 61 | #include "page_reporting.h" |
1da177e4 | 62 | |
f04a5d5d DH |
63 | /* Free Page Internal flags: for internal, non-pcp variants of free_pages(). */ |
64 | typedef int __bitwise fpi_t; | |
65 | ||
66 | /* No special request */ | |
67 | #define FPI_NONE ((__force fpi_t)0) | |
68 | ||
69 | /* | |
70 | * Skip free page reporting notification for the (possibly merged) page. | |
71 | * This does not hinder free page reporting from grabbing the page, | |
72 | * reporting it and marking it "reported" - it only skips notifying | |
73 | * the free page reporting infrastructure about a newly freed page. For | |
74 | * example, used when temporarily pulling a page from a freelist and | |
75 | * putting it back unmodified. | |
76 | */ | |
77 | #define FPI_SKIP_REPORT_NOTIFY ((__force fpi_t)BIT(0)) | |
78 | ||
47b6a24a DH |
79 | /* |
80 | * Place the (possibly merged) page to the tail of the freelist. Will ignore | |
81 | * page shuffling (relevant code - e.g., memory onlining - is expected to | |
82 | * shuffle the whole zone). | |
83 | * | |
84 | * Note: No code should rely on this flag for correctness - it's purely | |
85 | * to allow for optimizations when handing back either fresh pages | |
86 | * (memory onlining) or untouched pages (page isolation, free page | |
87 | * reporting). | |
88 | */ | |
89 | #define FPI_TO_TAIL ((__force fpi_t)BIT(1)) | |
90 | ||
c8e251fa CS |
91 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
92 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
74f44822 | 93 | #define MIN_PERCPU_PAGELIST_HIGH_FRACTION (8) |
c8e251fa | 94 | |
4b23a68f MG |
95 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT) |
96 | /* | |
97 | * On SMP, spin_trylock is sufficient protection. | |
98 | * On PREEMPT_RT, spin_trylock is equivalent on both SMP and UP. | |
99 | */ | |
100 | #define pcp_trylock_prepare(flags) do { } while (0) | |
101 | #define pcp_trylock_finish(flag) do { } while (0) | |
102 | #else | |
103 | ||
104 | /* UP spin_trylock always succeeds so disable IRQs to prevent re-entrancy. */ | |
105 | #define pcp_trylock_prepare(flags) local_irq_save(flags) | |
106 | #define pcp_trylock_finish(flags) local_irq_restore(flags) | |
107 | #endif | |
108 | ||
01b44456 MG |
109 | /* |
110 | * Locking a pcp requires a PCP lookup followed by a spinlock. To avoid | |
111 | * a migration causing the wrong PCP to be locked and remote memory being | |
112 | * potentially allocated, pin the task to the CPU for the lookup+lock. | |
113 | * preempt_disable is used on !RT because it is faster than migrate_disable. | |
114 | * migrate_disable is used on RT because otherwise RT spinlock usage is | |
115 | * interfered with and a high priority task cannot preempt the allocator. | |
116 | */ | |
117 | #ifndef CONFIG_PREEMPT_RT | |
118 | #define pcpu_task_pin() preempt_disable() | |
119 | #define pcpu_task_unpin() preempt_enable() | |
120 | #else | |
121 | #define pcpu_task_pin() migrate_disable() | |
122 | #define pcpu_task_unpin() migrate_enable() | |
123 | #endif | |
c8e251fa | 124 | |
01b44456 MG |
125 | /* |
126 | * Generic helper to lookup and a per-cpu variable with an embedded spinlock. | |
127 | * Return value should be used with equivalent unlock helper. | |
128 | */ | |
129 | #define pcpu_spin_lock(type, member, ptr) \ | |
130 | ({ \ | |
131 | type *_ret; \ | |
132 | pcpu_task_pin(); \ | |
133 | _ret = this_cpu_ptr(ptr); \ | |
134 | spin_lock(&_ret->member); \ | |
135 | _ret; \ | |
136 | }) | |
137 | ||
57490774 | 138 | #define pcpu_spin_trylock(type, member, ptr) \ |
01b44456 MG |
139 | ({ \ |
140 | type *_ret; \ | |
141 | pcpu_task_pin(); \ | |
142 | _ret = this_cpu_ptr(ptr); \ | |
57490774 | 143 | if (!spin_trylock(&_ret->member)) { \ |
01b44456 MG |
144 | pcpu_task_unpin(); \ |
145 | _ret = NULL; \ | |
146 | } \ | |
147 | _ret; \ | |
148 | }) | |
149 | ||
150 | #define pcpu_spin_unlock(member, ptr) \ | |
151 | ({ \ | |
152 | spin_unlock(&ptr->member); \ | |
153 | pcpu_task_unpin(); \ | |
154 | }) | |
155 | ||
01b44456 MG |
156 | /* struct per_cpu_pages specific helpers. */ |
157 | #define pcp_spin_lock(ptr) \ | |
158 | pcpu_spin_lock(struct per_cpu_pages, lock, ptr) | |
159 | ||
57490774 MG |
160 | #define pcp_spin_trylock(ptr) \ |
161 | pcpu_spin_trylock(struct per_cpu_pages, lock, ptr) | |
01b44456 MG |
162 | |
163 | #define pcp_spin_unlock(ptr) \ | |
164 | pcpu_spin_unlock(lock, ptr) | |
165 | ||
72812019 LS |
166 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
167 | DEFINE_PER_CPU(int, numa_node); | |
168 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
169 | #endif | |
170 | ||
4518085e KW |
171 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
172 | ||
7aac7898 LS |
173 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
174 | /* | |
175 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
176 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
177 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
178 | * defined in <linux/topology.h>. | |
179 | */ | |
180 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
181 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
182 | #endif | |
183 | ||
8b885f53 | 184 | static DEFINE_MUTEX(pcpu_drain_mutex); |
bd233f53 | 185 | |
38addce8 | 186 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 187 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
188 | EXPORT_SYMBOL(latent_entropy); |
189 | #endif | |
190 | ||
1da177e4 | 191 | /* |
13808910 | 192 | * Array of node states. |
1da177e4 | 193 | */ |
13808910 CL |
194 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
195 | [N_POSSIBLE] = NODE_MASK_ALL, | |
196 | [N_ONLINE] = { { [0] = 1UL } }, | |
197 | #ifndef CONFIG_NUMA | |
198 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
199 | #ifdef CONFIG_HIGHMEM | |
200 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 201 | #endif |
20b2f52b | 202 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
203 | [N_CPU] = { { [0] = 1UL } }, |
204 | #endif /* NUMA */ | |
205 | }; | |
206 | EXPORT_SYMBOL(node_states); | |
207 | ||
dcce284a | 208 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
6471384a | 209 | |
d9c23400 | 210 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 211 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
212 | #endif |
213 | ||
7fef431b DH |
214 | static void __free_pages_ok(struct page *page, unsigned int order, |
215 | fpi_t fpi_flags); | |
a226f6c8 | 216 | |
1da177e4 LT |
217 | /* |
218 | * results with 256, 32 in the lowmem_reserve sysctl: | |
219 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
220 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
221 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
222 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 223 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
224 | * |
225 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
226 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 227 | */ |
62069aac | 228 | static int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 229 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 230 | [ZONE_DMA] = 256, |
4b51d669 | 231 | #endif |
fb0e7942 | 232 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 233 | [ZONE_DMA32] = 256, |
fb0e7942 | 234 | #endif |
d3cda233 | 235 | [ZONE_NORMAL] = 32, |
e53ef38d | 236 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 237 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 238 | #endif |
d3cda233 | 239 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 240 | }; |
1da177e4 | 241 | |
9420f89d | 242 | char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 243 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 244 | "DMA", |
4b51d669 | 245 | #endif |
fb0e7942 | 246 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 247 | "DMA32", |
fb0e7942 | 248 | #endif |
2f1b6248 | 249 | "Normal", |
e53ef38d | 250 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 251 | "HighMem", |
e53ef38d | 252 | #endif |
2a1e274a | 253 | "Movable", |
033fbae9 DW |
254 | #ifdef CONFIG_ZONE_DEVICE |
255 | "Device", | |
256 | #endif | |
2f1b6248 CL |
257 | }; |
258 | ||
c999fbd3 | 259 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
260 | "Unmovable", |
261 | "Movable", | |
262 | "Reclaimable", | |
263 | "HighAtomic", | |
264 | #ifdef CONFIG_CMA | |
265 | "CMA", | |
266 | #endif | |
267 | #ifdef CONFIG_MEMORY_ISOLATION | |
268 | "Isolate", | |
269 | #endif | |
270 | }; | |
271 | ||
1da177e4 | 272 | int min_free_kbytes = 1024; |
42aa83cb | 273 | int user_min_free_kbytes = -1; |
e95d372c KW |
274 | static int watermark_boost_factor __read_mostly = 15000; |
275 | static int watermark_scale_factor = 10; | |
0ee332c1 TH |
276 | |
277 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
278 | int movable_zone; | |
279 | EXPORT_SYMBOL(movable_zone); | |
c713216d | 280 | |
418508c1 | 281 | #if MAX_NUMNODES > 1 |
b9726c26 | 282 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 283 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 284 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 285 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
286 | #endif |
287 | ||
dcdfdd40 KS |
288 | static bool page_contains_unaccepted(struct page *page, unsigned int order); |
289 | static void accept_page(struct page *page, unsigned int order); | |
290 | static bool try_to_accept_memory(struct zone *zone, unsigned int order); | |
291 | static inline bool has_unaccepted_memory(void); | |
292 | static bool __free_unaccepted(struct page *page); | |
293 | ||
9ef9acb0 MG |
294 | int page_group_by_mobility_disabled __read_mostly; |
295 | ||
3a80a7fa | 296 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
297 | /* |
298 | * During boot we initialize deferred pages on-demand, as needed, but once | |
299 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
300 | * and we can permanently disable that path. | |
301 | */ | |
9420f89d | 302 | DEFINE_STATIC_KEY_TRUE(deferred_pages); |
3c0c12cc | 303 | |
94ae8b83 | 304 | static inline bool deferred_pages_enabled(void) |
3c0c12cc | 305 | { |
94ae8b83 | 306 | return static_branch_unlikely(&deferred_pages); |
3c0c12cc WL |
307 | } |
308 | ||
3a80a7fa | 309 | /* |
9420f89d MRI |
310 | * deferred_grow_zone() is __init, but it is called from |
311 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
312 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
313 | * and to ensure that the function body gets unloaded. | |
3a80a7fa | 314 | */ |
9420f89d MRI |
315 | static bool __ref |
316 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
3a80a7fa | 317 | { |
96a5c186 | 318 | return deferred_grow_zone(zone, order); |
3a80a7fa MG |
319 | } |
320 | #else | |
94ae8b83 | 321 | static inline bool deferred_pages_enabled(void) |
2c335680 | 322 | { |
94ae8b83 | 323 | return false; |
2c335680 | 324 | } |
9420f89d | 325 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
3a80a7fa | 326 | |
0b423ca2 | 327 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
ca891f41 | 328 | static inline unsigned long *get_pageblock_bitmap(const struct page *page, |
0b423ca2 MG |
329 | unsigned long pfn) |
330 | { | |
331 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 332 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
333 | #else |
334 | return page_zone(page)->pageblock_flags; | |
335 | #endif /* CONFIG_SPARSEMEM */ | |
336 | } | |
337 | ||
ca891f41 | 338 | static inline int pfn_to_bitidx(const struct page *page, unsigned long pfn) |
0b423ca2 MG |
339 | { |
340 | #ifdef CONFIG_SPARSEMEM | |
341 | pfn &= (PAGES_PER_SECTION-1); | |
0b423ca2 | 342 | #else |
4f9bc69a | 343 | pfn = pfn - pageblock_start_pfn(page_zone(page)->zone_start_pfn); |
0b423ca2 | 344 | #endif /* CONFIG_SPARSEMEM */ |
399b795b | 345 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
0b423ca2 MG |
346 | } |
347 | ||
a04d12c2 KS |
348 | /** |
349 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
350 | * @page: The page within the block of interest | |
351 | * @pfn: The target page frame number | |
352 | * @mask: mask of bits that the caller is interested in | |
353 | * | |
354 | * Return: pageblock_bits flags | |
355 | */ | |
356 | unsigned long get_pfnblock_flags_mask(const struct page *page, | |
357 | unsigned long pfn, unsigned long mask) | |
0b423ca2 MG |
358 | { |
359 | unsigned long *bitmap; | |
360 | unsigned long bitidx, word_bitidx; | |
361 | unsigned long word; | |
362 | ||
363 | bitmap = get_pageblock_bitmap(page, pfn); | |
364 | bitidx = pfn_to_bitidx(page, pfn); | |
365 | word_bitidx = bitidx / BITS_PER_LONG; | |
366 | bitidx &= (BITS_PER_LONG-1); | |
1c563432 MK |
367 | /* |
368 | * This races, without locks, with set_pfnblock_flags_mask(). Ensure | |
369 | * a consistent read of the memory array, so that results, even though | |
370 | * racy, are not corrupted. | |
371 | */ | |
372 | word = READ_ONCE(bitmap[word_bitidx]); | |
d93d5ab9 | 373 | return (word >> bitidx) & mask; |
0b423ca2 MG |
374 | } |
375 | ||
ca891f41 MWO |
376 | static __always_inline int get_pfnblock_migratetype(const struct page *page, |
377 | unsigned long pfn) | |
0b423ca2 | 378 | { |
a04d12c2 | 379 | return get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK); |
0b423ca2 MG |
380 | } |
381 | ||
382 | /** | |
383 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
384 | * @page: The page within the block of interest | |
385 | * @flags: The flags to set | |
386 | * @pfn: The target page frame number | |
0b423ca2 MG |
387 | * @mask: mask of bits that the caller is interested in |
388 | */ | |
389 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
390 | unsigned long pfn, | |
0b423ca2 MG |
391 | unsigned long mask) |
392 | { | |
393 | unsigned long *bitmap; | |
394 | unsigned long bitidx, word_bitidx; | |
04ec0061 | 395 | unsigned long word; |
0b423ca2 MG |
396 | |
397 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 398 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
399 | |
400 | bitmap = get_pageblock_bitmap(page, pfn); | |
401 | bitidx = pfn_to_bitidx(page, pfn); | |
402 | word_bitidx = bitidx / BITS_PER_LONG; | |
403 | bitidx &= (BITS_PER_LONG-1); | |
404 | ||
405 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
406 | ||
d93d5ab9 WY |
407 | mask <<= bitidx; |
408 | flags <<= bitidx; | |
0b423ca2 MG |
409 | |
410 | word = READ_ONCE(bitmap[word_bitidx]); | |
04ec0061 UB |
411 | do { |
412 | } while (!try_cmpxchg(&bitmap[word_bitidx], &word, (word & ~mask) | flags)); | |
0b423ca2 | 413 | } |
3a80a7fa | 414 | |
ee6f509c | 415 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 416 | { |
5d0f3f72 KM |
417 | if (unlikely(page_group_by_mobility_disabled && |
418 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
419 | migratetype = MIGRATE_UNMOVABLE; |
420 | ||
d93d5ab9 | 421 | set_pfnblock_flags_mask(page, (unsigned long)migratetype, |
535b81e2 | 422 | page_to_pfn(page), MIGRATETYPE_MASK); |
b2a0ac88 MG |
423 | } |
424 | ||
13e7444b | 425 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 426 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 427 | { |
82d9b8c8 | 428 | int ret; |
bdc8cb98 DH |
429 | unsigned seq; |
430 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 431 | unsigned long sp, start_pfn; |
c6a57e19 | 432 | |
bdc8cb98 DH |
433 | do { |
434 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
435 | start_pfn = zone->zone_start_pfn; |
436 | sp = zone->spanned_pages; | |
82d9b8c8 | 437 | ret = !zone_spans_pfn(zone, pfn); |
bdc8cb98 DH |
438 | } while (zone_span_seqretry(zone, seq)); |
439 | ||
b5e6a5a2 | 440 | if (ret) |
613813e8 DH |
441 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
442 | pfn, zone_to_nid(zone), zone->name, | |
443 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 444 | |
bdc8cb98 | 445 | return ret; |
c6a57e19 DH |
446 | } |
447 | ||
c6a57e19 DH |
448 | /* |
449 | * Temporary debugging check for pages not lying within a given zone. | |
450 | */ | |
5bb14214 | 451 | static bool __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
452 | { |
453 | if (page_outside_zone_boundaries(zone, page)) | |
5bb14214 | 454 | return true; |
5b855aa3 | 455 | if (zone != page_zone(page)) |
5bb14214 | 456 | return true; |
c6a57e19 | 457 | |
5bb14214 | 458 | return false; |
1da177e4 | 459 | } |
13e7444b | 460 | #else |
5bb14214 | 461 | static inline bool __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b | 462 | { |
5bb14214 | 463 | return false; |
13e7444b NP |
464 | } |
465 | #endif | |
466 | ||
82a3241a | 467 | static void bad_page(struct page *page, const char *reason) |
1da177e4 | 468 | { |
d936cf9b HD |
469 | static unsigned long resume; |
470 | static unsigned long nr_shown; | |
471 | static unsigned long nr_unshown; | |
472 | ||
473 | /* | |
474 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
475 | * or allow a steady drip of one report per second. | |
476 | */ | |
477 | if (nr_shown == 60) { | |
478 | if (time_before(jiffies, resume)) { | |
479 | nr_unshown++; | |
480 | goto out; | |
481 | } | |
482 | if (nr_unshown) { | |
ff8e8116 | 483 | pr_alert( |
1e9e6365 | 484 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
485 | nr_unshown); |
486 | nr_unshown = 0; | |
487 | } | |
488 | nr_shown = 0; | |
489 | } | |
490 | if (nr_shown++ == 0) | |
491 | resume = jiffies + 60 * HZ; | |
492 | ||
ff8e8116 | 493 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 494 | current->comm, page_to_pfn(page)); |
d2f07ec0 | 495 | dump_page(page, reason); |
3dc14741 | 496 | |
4f31888c | 497 | print_modules(); |
1da177e4 | 498 | dump_stack(); |
d936cf9b | 499 | out: |
8cc3b392 | 500 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 501 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 502 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
503 | } |
504 | ||
44042b44 MG |
505 | static inline unsigned int order_to_pindex(int migratetype, int order) |
506 | { | |
44042b44 MG |
507 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
508 | if (order > PAGE_ALLOC_COSTLY_ORDER) { | |
6303d1c5 | 509 | VM_BUG_ON(order != HPAGE_PMD_ORDER); |
5d0a661d | 510 | return NR_LOWORDER_PCP_LISTS; |
44042b44 MG |
511 | } |
512 | #else | |
513 | VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER); | |
514 | #endif | |
515 | ||
c1dc69e6 | 516 | return (MIGRATE_PCPTYPES * order) + migratetype; |
44042b44 MG |
517 | } |
518 | ||
519 | static inline int pindex_to_order(unsigned int pindex) | |
520 | { | |
521 | int order = pindex / MIGRATE_PCPTYPES; | |
522 | ||
523 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5d0a661d | 524 | if (pindex == NR_LOWORDER_PCP_LISTS) |
6303d1c5 | 525 | order = HPAGE_PMD_ORDER; |
44042b44 MG |
526 | #else |
527 | VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER); | |
528 | #endif | |
529 | ||
530 | return order; | |
531 | } | |
532 | ||
533 | static inline bool pcp_allowed_order(unsigned int order) | |
534 | { | |
535 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
536 | return true; | |
537 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6303d1c5 | 538 | if (order == HPAGE_PMD_ORDER) |
44042b44 MG |
539 | return true; |
540 | #endif | |
541 | return false; | |
542 | } | |
543 | ||
1da177e4 LT |
544 | /* |
545 | * Higher-order pages are called "compound pages". They are structured thusly: | |
546 | * | |
1d798ca3 | 547 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 548 | * |
1d798ca3 KS |
549 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
550 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 551 | * |
1d798ca3 | 552 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 553 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 554 | */ |
d98c7a09 | 555 | |
d00181b9 | 556 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
557 | { |
558 | int i; | |
559 | int nr_pages = 1 << order; | |
560 | ||
18229df5 | 561 | __SetPageHead(page); |
5b24eeef JM |
562 | for (i = 1; i < nr_pages; i++) |
563 | prep_compound_tail(page, i); | |
1378a5ee | 564 | |
5b24eeef | 565 | prep_compound_head(page, order); |
18229df5 AW |
566 | } |
567 | ||
ab130f91 | 568 | static inline void set_buddy_order(struct page *page, unsigned int order) |
6aa3001b | 569 | { |
4c21e2f2 | 570 | set_page_private(page, order); |
676165a8 | 571 | __SetPageBuddy(page); |
1da177e4 LT |
572 | } |
573 | ||
5e1f0f09 MG |
574 | #ifdef CONFIG_COMPACTION |
575 | static inline struct capture_control *task_capc(struct zone *zone) | |
576 | { | |
577 | struct capture_control *capc = current->capture_control; | |
578 | ||
deba0487 | 579 | return unlikely(capc) && |
5e1f0f09 MG |
580 | !(current->flags & PF_KTHREAD) && |
581 | !capc->page && | |
deba0487 | 582 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
583 | } |
584 | ||
585 | static inline bool | |
586 | compaction_capture(struct capture_control *capc, struct page *page, | |
587 | int order, int migratetype) | |
588 | { | |
589 | if (!capc || order != capc->cc->order) | |
590 | return false; | |
591 | ||
592 | /* Do not accidentally pollute CMA or isolated regions*/ | |
593 | if (is_migrate_cma(migratetype) || | |
594 | is_migrate_isolate(migratetype)) | |
595 | return false; | |
596 | ||
597 | /* | |
231f8c71 BW |
598 | * Do not let lower order allocations pollute a movable pageblock |
599 | * unless compaction is also requesting movable pages. | |
5e1f0f09 MG |
600 | * This might let an unmovable request use a reclaimable pageblock |
601 | * and vice-versa but no more than normal fallback logic which can | |
602 | * have trouble finding a high-order free page. | |
603 | */ | |
231f8c71 BW |
604 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE && |
605 | capc->cc->migratetype != MIGRATE_MOVABLE) | |
5e1f0f09 MG |
606 | return false; |
607 | ||
608 | capc->page = page; | |
609 | return true; | |
610 | } | |
611 | ||
612 | #else | |
613 | static inline struct capture_control *task_capc(struct zone *zone) | |
614 | { | |
615 | return NULL; | |
616 | } | |
617 | ||
618 | static inline bool | |
619 | compaction_capture(struct capture_control *capc, struct page *page, | |
620 | int order, int migratetype) | |
621 | { | |
622 | return false; | |
623 | } | |
624 | #endif /* CONFIG_COMPACTION */ | |
625 | ||
e0932b6c JW |
626 | static inline void account_freepages(struct zone *zone, int nr_pages, |
627 | int migratetype) | |
6ab01363 | 628 | { |
e0932b6c JW |
629 | if (is_migrate_isolate(migratetype)) |
630 | return; | |
6ab01363 | 631 | |
e0932b6c JW |
632 | __mod_zone_page_state(zone, NR_FREE_PAGES, nr_pages); |
633 | ||
634 | if (is_migrate_cma(migratetype)) | |
635 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, nr_pages); | |
6ab01363 AD |
636 | } |
637 | ||
638 | /* Used for pages not on another list */ | |
e0932b6c JW |
639 | static inline void __add_to_free_list(struct page *page, struct zone *zone, |
640 | unsigned int order, int migratetype, | |
641 | bool tail) | |
6ab01363 AD |
642 | { |
643 | struct free_area *area = &zone->free_area[order]; | |
644 | ||
e0932b6c JW |
645 | VM_WARN_ONCE(get_pageblock_migratetype(page) != migratetype, |
646 | "page type is %lu, passed migratetype is %d (nr=%d)\n", | |
647 | get_pageblock_migratetype(page), migratetype, 1 << order); | |
648 | ||
649 | if (tail) | |
650 | list_add_tail(&page->buddy_list, &area->free_list[migratetype]); | |
651 | else | |
652 | list_add(&page->buddy_list, &area->free_list[migratetype]); | |
6ab01363 AD |
653 | area->nr_free++; |
654 | } | |
655 | ||
293ffa5e DH |
656 | /* |
657 | * Used for pages which are on another list. Move the pages to the tail | |
658 | * of the list - so the moved pages won't immediately be considered for | |
659 | * allocation again (e.g., optimization for memory onlining). | |
660 | */ | |
6ab01363 | 661 | static inline void move_to_free_list(struct page *page, struct zone *zone, |
e0932b6c | 662 | unsigned int order, int old_mt, int new_mt) |
6ab01363 AD |
663 | { |
664 | struct free_area *area = &zone->free_area[order]; | |
665 | ||
e0932b6c JW |
666 | /* Free page moving can fail, so it happens before the type update */ |
667 | VM_WARN_ONCE(get_pageblock_migratetype(page) != old_mt, | |
668 | "page type is %lu, passed migratetype is %d (nr=%d)\n", | |
669 | get_pageblock_migratetype(page), old_mt, 1 << order); | |
670 | ||
671 | list_move_tail(&page->buddy_list, &area->free_list[new_mt]); | |
672 | ||
673 | account_freepages(zone, -(1 << order), old_mt); | |
674 | account_freepages(zone, 1 << order, new_mt); | |
6ab01363 AD |
675 | } |
676 | ||
e0932b6c JW |
677 | static inline void __del_page_from_free_list(struct page *page, struct zone *zone, |
678 | unsigned int order, int migratetype) | |
6ab01363 | 679 | { |
e0932b6c JW |
680 | VM_WARN_ONCE(get_pageblock_migratetype(page) != migratetype, |
681 | "page type is %lu, passed migratetype is %d (nr=%d)\n", | |
682 | get_pageblock_migratetype(page), migratetype, 1 << order); | |
683 | ||
36e66c55 AD |
684 | /* clear reported state and update reported page count */ |
685 | if (page_reported(page)) | |
686 | __ClearPageReported(page); | |
687 | ||
bf75f200 | 688 | list_del(&page->buddy_list); |
6ab01363 AD |
689 | __ClearPageBuddy(page); |
690 | set_page_private(page, 0); | |
691 | zone->free_area[order].nr_free--; | |
692 | } | |
693 | ||
e0932b6c JW |
694 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, |
695 | unsigned int order, int migratetype) | |
696 | { | |
697 | __del_page_from_free_list(page, zone, order, migratetype); | |
698 | account_freepages(zone, -(1 << order), migratetype); | |
699 | } | |
700 | ||
5d671eb4 MRI |
701 | static inline struct page *get_page_from_free_area(struct free_area *area, |
702 | int migratetype) | |
703 | { | |
704 | return list_first_entry_or_null(&area->free_list[migratetype], | |
1bf61092 | 705 | struct page, buddy_list); |
5d671eb4 MRI |
706 | } |
707 | ||
a2129f24 AD |
708 | /* |
709 | * If this is not the largest possible page, check if the buddy | |
710 | * of the next-highest order is free. If it is, it's possible | |
711 | * that pages are being freed that will coalesce soon. In case, | |
712 | * that is happening, add the free page to the tail of the list | |
713 | * so it's less likely to be used soon and more likely to be merged | |
714 | * as a higher order page | |
715 | */ | |
716 | static inline bool | |
717 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
718 | struct page *page, unsigned int order) | |
719 | { | |
8170ac47 ZY |
720 | unsigned long higher_page_pfn; |
721 | struct page *higher_page; | |
a2129f24 | 722 | |
5e0a760b | 723 | if (order >= MAX_PAGE_ORDER - 1) |
a2129f24 AD |
724 | return false; |
725 | ||
8170ac47 ZY |
726 | higher_page_pfn = buddy_pfn & pfn; |
727 | higher_page = page + (higher_page_pfn - pfn); | |
a2129f24 | 728 | |
8170ac47 ZY |
729 | return find_buddy_page_pfn(higher_page, higher_page_pfn, order + 1, |
730 | NULL) != NULL; | |
a2129f24 AD |
731 | } |
732 | ||
1da177e4 LT |
733 | /* |
734 | * Freeing function for a buddy system allocator. | |
735 | * | |
736 | * The concept of a buddy system is to maintain direct-mapped table | |
737 | * (containing bit values) for memory blocks of various "orders". | |
738 | * The bottom level table contains the map for the smallest allocatable | |
739 | * units of memory (here, pages), and each level above it describes | |
740 | * pairs of units from the levels below, hence, "buddies". | |
741 | * At a high level, all that happens here is marking the table entry | |
742 | * at the bottom level available, and propagating the changes upward | |
743 | * as necessary, plus some accounting needed to play nicely with other | |
744 | * parts of the VM system. | |
745 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
746 | * free pages of length of (1 << order) and marked with PageBuddy. |
747 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 748 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
749 | * other. That is, if we allocate a small block, and both were |
750 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 751 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 752 | * triggers coalescing into a block of larger size. |
1da177e4 | 753 | * |
6d49e352 | 754 | * -- nyc |
1da177e4 LT |
755 | */ |
756 | ||
48db57f8 | 757 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 758 | unsigned long pfn, |
ed0ae21d | 759 | struct zone *zone, unsigned int order, |
f04a5d5d | 760 | int migratetype, fpi_t fpi_flags) |
1da177e4 | 761 | { |
a2129f24 | 762 | struct capture_control *capc = task_capc(zone); |
dae37a5d | 763 | unsigned long buddy_pfn = 0; |
a2129f24 | 764 | unsigned long combined_pfn; |
a2129f24 AD |
765 | struct page *buddy; |
766 | bool to_tail; | |
d9dddbf5 | 767 | |
d29bb978 | 768 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 769 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 770 | |
ed0ae21d | 771 | VM_BUG_ON(migratetype == -1); |
76741e77 | 772 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 773 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 774 | |
e0932b6c JW |
775 | account_freepages(zone, 1 << order, migratetype); |
776 | ||
5e0a760b | 777 | while (order < MAX_PAGE_ORDER) { |
e0932b6c JW |
778 | int buddy_mt = migratetype; |
779 | ||
5e1f0f09 | 780 | if (compaction_capture(capc, page, order, migratetype)) { |
e0932b6c | 781 | account_freepages(zone, -(1 << order), migratetype); |
5e1f0f09 MG |
782 | return; |
783 | } | |
13ad59df | 784 | |
8170ac47 ZY |
785 | buddy = find_buddy_page_pfn(page, pfn, order, &buddy_pfn); |
786 | if (!buddy) | |
d9dddbf5 | 787 | goto done_merging; |
bb0e28eb ZY |
788 | |
789 | if (unlikely(order >= pageblock_order)) { | |
790 | /* | |
791 | * We want to prevent merge between freepages on pageblock | |
792 | * without fallbacks and normal pageblock. Without this, | |
793 | * pageblock isolation could cause incorrect freepage or CMA | |
794 | * accounting or HIGHATOMIC accounting. | |
795 | */ | |
e0932b6c | 796 | buddy_mt = get_pfnblock_migratetype(buddy, buddy_pfn); |
bb0e28eb | 797 | |
e0932b6c JW |
798 | if (migratetype != buddy_mt && |
799 | (!migratetype_is_mergeable(migratetype) || | |
800 | !migratetype_is_mergeable(buddy_mt))) | |
801 | goto done_merging; | |
bb0e28eb ZY |
802 | } |
803 | ||
c0a32fc5 SG |
804 | /* |
805 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
806 | * merge with it and move up one order. | |
807 | */ | |
b03641af | 808 | if (page_is_guard(buddy)) |
e0932b6c | 809 | clear_page_guard(zone, buddy, order); |
b03641af | 810 | else |
e0932b6c JW |
811 | __del_page_from_free_list(buddy, zone, order, buddy_mt); |
812 | ||
813 | if (unlikely(buddy_mt != migratetype)) { | |
814 | /* | |
815 | * Match buddy type. This ensures that an | |
816 | * expand() down the line puts the sub-blocks | |
817 | * on the right freelists. | |
818 | */ | |
819 | set_pageblock_migratetype(buddy, migratetype); | |
820 | } | |
821 | ||
76741e77 VB |
822 | combined_pfn = buddy_pfn & pfn; |
823 | page = page + (combined_pfn - pfn); | |
824 | pfn = combined_pfn; | |
1da177e4 LT |
825 | order++; |
826 | } | |
d9dddbf5 VB |
827 | |
828 | done_merging: | |
ab130f91 | 829 | set_buddy_order(page, order); |
6dda9d55 | 830 | |
47b6a24a DH |
831 | if (fpi_flags & FPI_TO_TAIL) |
832 | to_tail = true; | |
833 | else if (is_shuffle_order(order)) | |
a2129f24 | 834 | to_tail = shuffle_pick_tail(); |
97500a4a | 835 | else |
a2129f24 | 836 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 837 | |
e0932b6c | 838 | __add_to_free_list(page, zone, order, migratetype, to_tail); |
36e66c55 AD |
839 | |
840 | /* Notify page reporting subsystem of freed page */ | |
f04a5d5d | 841 | if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY)) |
36e66c55 | 842 | page_reporting_notify_free(order); |
1da177e4 LT |
843 | } |
844 | ||
7bfec6f4 MG |
845 | /* |
846 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
847 | * try and check multiple fields with one check. The caller must do a detailed | |
848 | * check if necessary. | |
849 | */ | |
850 | static inline bool page_expected_state(struct page *page, | |
851 | unsigned long check_flags) | |
852 | { | |
853 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
854 | return false; | |
855 | ||
856 | if (unlikely((unsigned long)page->mapping | | |
857 | page_ref_count(page) | | |
858 | #ifdef CONFIG_MEMCG | |
48060834 | 859 | page->memcg_data | |
dba1b8a7 JDB |
860 | #endif |
861 | #ifdef CONFIG_PAGE_POOL | |
862 | ((page->pp_magic & ~0x3UL) == PP_SIGNATURE) | | |
7bfec6f4 MG |
863 | #endif |
864 | (page->flags & check_flags))) | |
865 | return false; | |
866 | ||
867 | return true; | |
868 | } | |
869 | ||
58b7f119 | 870 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 871 | { |
82a3241a | 872 | const char *bad_reason = NULL; |
f0b791a3 | 873 | |
53f9263b | 874 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
875 | bad_reason = "nonzero mapcount"; |
876 | if (unlikely(page->mapping != NULL)) | |
877 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 878 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 879 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
880 | if (unlikely(page->flags & flags)) { |
881 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
882 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
883 | else | |
884 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 885 | } |
9edad6ea | 886 | #ifdef CONFIG_MEMCG |
48060834 | 887 | if (unlikely(page->memcg_data)) |
9edad6ea | 888 | bad_reason = "page still charged to cgroup"; |
dba1b8a7 JDB |
889 | #endif |
890 | #ifdef CONFIG_PAGE_POOL | |
891 | if (unlikely((page->pp_magic & ~0x3UL) == PP_SIGNATURE)) | |
892 | bad_reason = "page_pool leak"; | |
9edad6ea | 893 | #endif |
58b7f119 WY |
894 | return bad_reason; |
895 | } | |
896 | ||
a8368cd8 | 897 | static void free_page_is_bad_report(struct page *page) |
58b7f119 WY |
898 | { |
899 | bad_page(page, | |
900 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
901 | } |
902 | ||
a8368cd8 | 903 | static inline bool free_page_is_bad(struct page *page) |
bb552ac6 | 904 | { |
da838d4f | 905 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
a8368cd8 | 906 | return false; |
bb552ac6 MG |
907 | |
908 | /* Something has gone sideways, find it */ | |
a8368cd8 AM |
909 | free_page_is_bad_report(page); |
910 | return true; | |
1da177e4 LT |
911 | } |
912 | ||
ecbb490d KW |
913 | static inline bool is_check_pages_enabled(void) |
914 | { | |
915 | return static_branch_unlikely(&check_pages_enabled); | |
916 | } | |
917 | ||
8666925c | 918 | static int free_tail_page_prepare(struct page *head_page, struct page *page) |
4db7548c | 919 | { |
94688e8e | 920 | struct folio *folio = (struct folio *)head_page; |
4db7548c MG |
921 | int ret = 1; |
922 | ||
923 | /* | |
924 | * We rely page->lru.next never has bit 0 set, unless the page | |
925 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
926 | */ | |
927 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
928 | ||
ecbb490d | 929 | if (!is_check_pages_enabled()) { |
4db7548c MG |
930 | ret = 0; |
931 | goto out; | |
932 | } | |
933 | switch (page - head_page) { | |
934 | case 1: | |
cb67f428 | 935 | /* the first tail page: these may be in place of ->mapping */ |
65a689f3 MWO |
936 | if (unlikely(folio_entire_mapcount(folio))) { |
937 | bad_page(page, "nonzero entire_mapcount"); | |
4db7548c MG |
938 | goto out; |
939 | } | |
05c5323b DH |
940 | if (unlikely(folio_large_mapcount(folio))) { |
941 | bad_page(page, "nonzero large_mapcount"); | |
942 | goto out; | |
943 | } | |
65a689f3 MWO |
944 | if (unlikely(atomic_read(&folio->_nr_pages_mapped))) { |
945 | bad_page(page, "nonzero nr_pages_mapped"); | |
cb67f428 HD |
946 | goto out; |
947 | } | |
94688e8e MWO |
948 | if (unlikely(atomic_read(&folio->_pincount))) { |
949 | bad_page(page, "nonzero pincount"); | |
cb67f428 HD |
950 | goto out; |
951 | } | |
4db7548c MG |
952 | break; |
953 | case 2: | |
b7b098cf MWO |
954 | /* the second tail page: deferred_list overlaps ->mapping */ |
955 | if (unlikely(!list_empty(&folio->_deferred_list))) { | |
956 | bad_page(page, "on deferred list"); | |
957 | goto out; | |
958 | } | |
4db7548c MG |
959 | break; |
960 | default: | |
961 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 962 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
963 | goto out; |
964 | } | |
965 | break; | |
966 | } | |
967 | if (unlikely(!PageTail(page))) { | |
82a3241a | 968 | bad_page(page, "PageTail not set"); |
4db7548c MG |
969 | goto out; |
970 | } | |
971 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 972 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
973 | goto out; |
974 | } | |
975 | ret = 0; | |
976 | out: | |
977 | page->mapping = NULL; | |
978 | clear_compound_head(page); | |
979 | return ret; | |
980 | } | |
981 | ||
94ae8b83 AK |
982 | /* |
983 | * Skip KASAN memory poisoning when either: | |
984 | * | |
0a54864f PC |
985 | * 1. For generic KASAN: deferred memory initialization has not yet completed. |
986 | * Tag-based KASAN modes skip pages freed via deferred memory initialization | |
987 | * using page tags instead (see below). | |
988 | * 2. For tag-based KASAN modes: the page has a match-all KASAN tag, indicating | |
989 | * that error detection is disabled for accesses via the page address. | |
990 | * | |
991 | * Pages will have match-all tags in the following circumstances: | |
992 | * | |
993 | * 1. Pages are being initialized for the first time, including during deferred | |
994 | * memory init; see the call to page_kasan_tag_reset in __init_single_page. | |
995 | * 2. The allocation was not unpoisoned due to __GFP_SKIP_KASAN, with the | |
996 | * exception of pages unpoisoned by kasan_unpoison_vmalloc. | |
997 | * 3. The allocation was excluded from being checked due to sampling, | |
44383cef | 998 | * see the call to kasan_unpoison_pages. |
94ae8b83 AK |
999 | * |
1000 | * Poisoning pages during deferred memory init will greatly lengthen the | |
1001 | * process and cause problem in large memory systems as the deferred pages | |
1002 | * initialization is done with interrupt disabled. | |
1003 | * | |
1004 | * Assuming that there will be no reference to those newly initialized | |
1005 | * pages before they are ever allocated, this should have no effect on | |
1006 | * KASAN memory tracking as the poison will be properly inserted at page | |
1007 | * allocation time. The only corner case is when pages are allocated by | |
1008 | * on-demand allocation and then freed again before the deferred pages | |
1009 | * initialization is done, but this is not likely to happen. | |
1010 | */ | |
5267fe5d | 1011 | static inline bool should_skip_kasan_poison(struct page *page) |
94ae8b83 | 1012 | { |
0a54864f PC |
1013 | if (IS_ENABLED(CONFIG_KASAN_GENERIC)) |
1014 | return deferred_pages_enabled(); | |
1015 | ||
5cb6674b | 1016 | return page_kasan_tag(page) == KASAN_TAG_KERNEL; |
94ae8b83 AK |
1017 | } |
1018 | ||
ba42b524 | 1019 | void kernel_init_pages(struct page *page, int numpages) |
6471384a AP |
1020 | { |
1021 | int i; | |
1022 | ||
9e15afa5 QC |
1023 | /* s390's use of memset() could override KASAN redzones. */ |
1024 | kasan_disable_current(); | |
d9da8f6c AK |
1025 | for (i = 0; i < numpages; i++) |
1026 | clear_highpage_kasan_tagged(page + i); | |
9e15afa5 | 1027 | kasan_enable_current(); |
6471384a AP |
1028 | } |
1029 | ||
733aea0b | 1030 | __always_inline bool free_pages_prepare(struct page *page, |
5267fe5d | 1031 | unsigned int order) |
4db7548c | 1032 | { |
e2769dbd | 1033 | int bad = 0; |
5267fe5d | 1034 | bool skip_kasan_poison = should_skip_kasan_poison(page); |
c3525330 | 1035 | bool init = want_init_on_free(); |
76f26535 | 1036 | bool compound = PageCompound(page); |
4db7548c | 1037 | |
4db7548c MG |
1038 | VM_BUG_ON_PAGE(PageTail(page), page); |
1039 | ||
e2769dbd | 1040 | trace_mm_page_free(page, order); |
b073d7f8 | 1041 | kmsan_free_page(page, order); |
e2769dbd | 1042 | |
17b46e7b BJ |
1043 | if (memcg_kmem_online() && PageMemcgKmem(page)) |
1044 | __memcg_kmem_uncharge_page(page, order); | |
1045 | ||
79f5f8fa | 1046 | if (unlikely(PageHWPoison(page)) && !order) { |
17b46e7b | 1047 | /* Do not let hwpoison pages hit pcplists/buddy */ |
79f5f8fa | 1048 | reset_page_owner(page, order); |
df4e817b | 1049 | page_table_check_free(page, order); |
dcfe378c | 1050 | pgalloc_tag_sub(page, 1 << order); |
79f5f8fa OS |
1051 | return false; |
1052 | } | |
1053 | ||
76f26535 HY |
1054 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); |
1055 | ||
e2769dbd MG |
1056 | /* |
1057 | * Check tail pages before head page information is cleared to | |
1058 | * avoid checking PageCompound for order-0 pages. | |
1059 | */ | |
1060 | if (unlikely(order)) { | |
e2769dbd MG |
1061 | int i; |
1062 | ||
cb67f428 | 1063 | if (compound) |
9c5ccf2d | 1064 | page[1].flags &= ~PAGE_FLAGS_SECOND; |
e2769dbd MG |
1065 | for (i = 1; i < (1 << order); i++) { |
1066 | if (compound) | |
8666925c | 1067 | bad += free_tail_page_prepare(page, page + i); |
fce0b421 | 1068 | if (is_check_pages_enabled()) { |
8666925c | 1069 | if (free_page_is_bad(page + i)) { |
700d2e9a VB |
1070 | bad++; |
1071 | continue; | |
1072 | } | |
e2769dbd MG |
1073 | } |
1074 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1075 | } | |
1076 | } | |
bda807d4 | 1077 | if (PageMappingFlags(page)) |
4db7548c | 1078 | page->mapping = NULL; |
fce0b421 | 1079 | if (is_check_pages_enabled()) { |
700d2e9a VB |
1080 | if (free_page_is_bad(page)) |
1081 | bad++; | |
1082 | if (bad) | |
1083 | return false; | |
1084 | } | |
4db7548c | 1085 | |
e2769dbd MG |
1086 | page_cpupid_reset_last(page); |
1087 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1088 | reset_page_owner(page, order); | |
df4e817b | 1089 | page_table_check_free(page, order); |
dcfe378c | 1090 | pgalloc_tag_sub(page, 1 << order); |
4db7548c MG |
1091 | |
1092 | if (!PageHighMem(page)) { | |
1093 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1094 | PAGE_SIZE << order); |
4db7548c | 1095 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1096 | PAGE_SIZE << order); |
4db7548c | 1097 | } |
6471384a | 1098 | |
8db26a3d VB |
1099 | kernel_poison_pages(page, 1 << order); |
1100 | ||
f9d79e8d | 1101 | /* |
1bb5eab3 | 1102 | * As memory initialization might be integrated into KASAN, |
7c13c163 | 1103 | * KASAN poisoning and memory initialization code must be |
1bb5eab3 AK |
1104 | * kept together to avoid discrepancies in behavior. |
1105 | * | |
f9d79e8d AK |
1106 | * With hardware tag-based KASAN, memory tags must be set before the |
1107 | * page becomes unavailable via debug_pagealloc or arch_free_page. | |
1108 | */ | |
f446883d | 1109 | if (!skip_kasan_poison) { |
c3525330 | 1110 | kasan_poison_pages(page, order, init); |
f9d79e8d | 1111 | |
db8a0477 AK |
1112 | /* Memory is already initialized if KASAN did it internally. */ |
1113 | if (kasan_has_integrated_init()) | |
1114 | init = false; | |
1115 | } | |
1116 | if (init) | |
aeaec8e2 | 1117 | kernel_init_pages(page, 1 << order); |
db8a0477 | 1118 | |
234fdce8 QC |
1119 | /* |
1120 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1121 | * does this. So nothing which can access the page's contents should | |
1122 | * happen after this. | |
1123 | */ | |
1124 | arch_free_page(page, order); | |
1125 | ||
77bc7fd6 | 1126 | debug_pagealloc_unmap_pages(page, 1 << order); |
d6332692 | 1127 | |
4db7548c MG |
1128 | return true; |
1129 | } | |
1130 | ||
1da177e4 | 1131 | /* |
5f8dcc21 | 1132 | * Frees a number of pages from the PCP lists |
7cba630b | 1133 | * Assumes all pages on list are in same zone. |
207f36ee | 1134 | * count is the number of pages to free. |
1da177e4 | 1135 | */ |
5f8dcc21 | 1136 | static void free_pcppages_bulk(struct zone *zone, int count, |
fd56eef2 MG |
1137 | struct per_cpu_pages *pcp, |
1138 | int pindex) | |
1da177e4 | 1139 | { |
57490774 | 1140 | unsigned long flags; |
44042b44 | 1141 | unsigned int order; |
8b10b465 | 1142 | struct page *page; |
f2260e6b | 1143 | |
88e8ac11 CTR |
1144 | /* |
1145 | * Ensure proper count is passed which otherwise would stuck in the | |
1146 | * below while (list_empty(list)) loop. | |
1147 | */ | |
1148 | count = min(pcp->count, count); | |
d61372bc MG |
1149 | |
1150 | /* Ensure requested pindex is drained first. */ | |
1151 | pindex = pindex - 1; | |
1152 | ||
57490774 | 1153 | spin_lock_irqsave(&zone->lock, flags); |
8b10b465 | 1154 | |
44042b44 | 1155 | while (count > 0) { |
5f8dcc21 | 1156 | struct list_head *list; |
fd56eef2 | 1157 | int nr_pages; |
5f8dcc21 | 1158 | |
fd56eef2 | 1159 | /* Remove pages from lists in a round-robin fashion. */ |
5f8dcc21 | 1160 | do { |
f142b2c2 KS |
1161 | if (++pindex > NR_PCP_LISTS - 1) |
1162 | pindex = 0; | |
44042b44 | 1163 | list = &pcp->lists[pindex]; |
f142b2c2 | 1164 | } while (list_empty(list)); |
48db57f8 | 1165 | |
44042b44 | 1166 | order = pindex_to_order(pindex); |
fd56eef2 | 1167 | nr_pages = 1 << order; |
a6f9edd6 | 1168 | do { |
17edeb5d | 1169 | unsigned long pfn; |
8b10b465 MG |
1170 | int mt; |
1171 | ||
bf75f200 | 1172 | page = list_last_entry(list, struct page, pcp_list); |
17edeb5d JW |
1173 | pfn = page_to_pfn(page); |
1174 | mt = get_pfnblock_migratetype(page, pfn); | |
8b10b465 | 1175 | |
0a5f4e5b | 1176 | /* must delete to avoid corrupting pcp list */ |
bf75f200 | 1177 | list_del(&page->pcp_list); |
fd56eef2 MG |
1178 | count -= nr_pages; |
1179 | pcp->count -= nr_pages; | |
aa016d14 | 1180 | |
17edeb5d | 1181 | __free_one_page(page, pfn, zone, order, mt, FPI_NONE); |
8b10b465 MG |
1182 | trace_mm_page_pcpu_drain(page, order, mt); |
1183 | } while (count > 0 && !list_empty(list)); | |
0a5f4e5b | 1184 | } |
8b10b465 | 1185 | |
57490774 | 1186 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 LT |
1187 | } |
1188 | ||
55612e80 JW |
1189 | static void free_one_page(struct zone *zone, struct page *page, |
1190 | unsigned long pfn, unsigned int order, | |
1191 | fpi_t fpi_flags) | |
1da177e4 | 1192 | { |
df1acc85 | 1193 | unsigned long flags; |
55612e80 | 1194 | int migratetype; |
df1acc85 MG |
1195 | |
1196 | spin_lock_irqsave(&zone->lock, flags); | |
55612e80 | 1197 | migratetype = get_pfnblock_migratetype(page, pfn); |
7fef431b | 1198 | __free_one_page(page, pfn, zone, order, migratetype, fpi_flags); |
df1acc85 | 1199 | spin_unlock_irqrestore(&zone->lock, flags); |
48db57f8 NP |
1200 | } |
1201 | ||
7fef431b DH |
1202 | static void __free_pages_ok(struct page *page, unsigned int order, |
1203 | fpi_t fpi_flags) | |
ec95f53a | 1204 | { |
dc4b0caf | 1205 | unsigned long pfn = page_to_pfn(page); |
56f0e661 | 1206 | struct zone *zone = page_zone(page); |
ec95f53a | 1207 | |
5267fe5d | 1208 | if (!free_pages_prepare(page, order)) |
ec95f53a KM |
1209 | return; |
1210 | ||
55612e80 | 1211 | free_one_page(zone, page, pfn, order, fpi_flags); |
90249993 | 1212 | |
d34b0733 | 1213 | __count_vm_events(PGFREE, 1 << order); |
1da177e4 LT |
1214 | } |
1215 | ||
a9cd410a | 1216 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1217 | { |
c3993076 | 1218 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1219 | struct page *p = page; |
c3993076 | 1220 | unsigned int loop; |
a226f6c8 | 1221 | |
7fef431b DH |
1222 | /* |
1223 | * When initializing the memmap, __init_single_page() sets the refcount | |
1224 | * of all pages to 1 ("allocated"/"not free"). We have to set the | |
1225 | * refcount of all involved pages to 0. | |
1226 | */ | |
e2d0bd2b YL |
1227 | prefetchw(p); |
1228 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1229 | prefetchw(p + 1); | |
c3993076 JW |
1230 | __ClearPageReserved(p); |
1231 | set_page_count(p, 0); | |
a226f6c8 | 1232 | } |
e2d0bd2b YL |
1233 | __ClearPageReserved(p); |
1234 | set_page_count(p, 0); | |
c3993076 | 1235 | |
9705bea5 | 1236 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
7fef431b | 1237 | |
dcdfdd40 | 1238 | if (page_contains_unaccepted(page, order)) { |
5e0a760b | 1239 | if (order == MAX_PAGE_ORDER && __free_unaccepted(page)) |
dcdfdd40 KS |
1240 | return; |
1241 | ||
1242 | accept_page(page, order); | |
1243 | } | |
1244 | ||
7fef431b DH |
1245 | /* |
1246 | * Bypass PCP and place fresh pages right to the tail, primarily | |
1247 | * relevant for memory onlining. | |
1248 | */ | |
0a54864f | 1249 | __free_pages_ok(page, order, FPI_TO_TAIL); |
a226f6c8 DH |
1250 | } |
1251 | ||
7cf91a98 JK |
1252 | /* |
1253 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1254 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
859a85dd | 1255 | * with the migration of free compaction scanner. |
7cf91a98 JK |
1256 | * |
1257 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1258 | * | |
1259 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1260 | * i.e. it's possible that all pages within a zones range of pages do not | |
1261 | * belong to a single zone. We assume that a border between node0 and node1 | |
1262 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1263 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1264 | * the first and last page of a pageblock and avoid checking each individual | |
1265 | * page in a pageblock. | |
65f67a3e BW |
1266 | * |
1267 | * Note: the function may return non-NULL struct page even for a page block | |
1268 | * which contains a memory hole (i.e. there is no physical memory for a subset | |
5e0a760b | 1269 | * of the pfn range). For example, if the pageblock order is MAX_PAGE_ORDER, which |
65f67a3e BW |
1270 | * will fall into 2 sub-sections, and the end pfn of the pageblock may be hole |
1271 | * even though the start pfn is online and valid. This should be safe most of | |
1272 | * the time because struct pages are still initialized via init_unavailable_range() | |
1273 | * and pfn walkers shouldn't touch any physical memory range for which they do | |
1274 | * not recognize any specific metadata in struct pages. | |
7cf91a98 JK |
1275 | */ |
1276 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1277 | unsigned long end_pfn, struct zone *zone) | |
1278 | { | |
1279 | struct page *start_page; | |
1280 | struct page *end_page; | |
1281 | ||
1282 | /* end_pfn is one past the range we are checking */ | |
1283 | end_pfn--; | |
1284 | ||
3c4322c9 | 1285 | if (!pfn_valid(end_pfn)) |
7cf91a98 JK |
1286 | return NULL; |
1287 | ||
2d070eab MH |
1288 | start_page = pfn_to_online_page(start_pfn); |
1289 | if (!start_page) | |
1290 | return NULL; | |
7cf91a98 JK |
1291 | |
1292 | if (page_zone(start_page) != zone) | |
1293 | return NULL; | |
1294 | ||
1295 | end_page = pfn_to_page(end_pfn); | |
1296 | ||
1297 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1298 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1299 | return NULL; | |
1300 | ||
1301 | return start_page; | |
1302 | } | |
1303 | ||
2f47a91f | 1304 | /* |
9420f89d MRI |
1305 | * The order of subdivision here is critical for the IO subsystem. |
1306 | * Please do not alter this order without good reasons and regression | |
1307 | * testing. Specifically, as large blocks of memory are subdivided, | |
1308 | * the order in which smaller blocks are delivered depends on the order | |
1309 | * they're subdivided in this function. This is the primary factor | |
1310 | * influencing the order in which pages are delivered to the IO | |
1311 | * subsystem according to empirical testing, and this is also justified | |
1312 | * by considering the behavior of a buddy system containing a single | |
1313 | * large block of memory acted on by a series of small allocations. | |
1314 | * This behavior is a critical factor in sglist merging's success. | |
80b1f41c | 1315 | * |
9420f89d | 1316 | * -- nyc |
2f47a91f | 1317 | */ |
9420f89d MRI |
1318 | static inline void expand(struct zone *zone, struct page *page, |
1319 | int low, int high, int migratetype) | |
2f47a91f | 1320 | { |
9420f89d | 1321 | unsigned long size = 1 << high; |
883dd161 | 1322 | unsigned long nr_added = 0; |
2f47a91f | 1323 | |
9420f89d MRI |
1324 | while (high > low) { |
1325 | high--; | |
1326 | size >>= 1; | |
1327 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); | |
2f47a91f | 1328 | |
9420f89d MRI |
1329 | /* |
1330 | * Mark as guard pages (or page), that will allow to | |
1331 | * merge back to allocator when buddy will be freed. | |
1332 | * Corresponding page table entries will not be touched, | |
1333 | * pages will stay not present in virtual address space | |
1334 | */ | |
e0932b6c | 1335 | if (set_page_guard(zone, &page[size], high)) |
2f47a91f | 1336 | continue; |
9420f89d | 1337 | |
883dd161 | 1338 | __add_to_free_list(&page[size], zone, high, migratetype, false); |
9420f89d | 1339 | set_buddy_order(&page[size], high); |
883dd161 | 1340 | nr_added += size; |
2f47a91f | 1341 | } |
883dd161 | 1342 | account_freepages(zone, nr_added, migratetype); |
2f47a91f PT |
1343 | } |
1344 | ||
9420f89d | 1345 | static void check_new_page_bad(struct page *page) |
0e56acae | 1346 | { |
9420f89d MRI |
1347 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1348 | /* Don't complain about hwpoisoned pages */ | |
1349 | page_mapcount_reset(page); /* remove PageBuddy */ | |
1350 | return; | |
0e56acae AD |
1351 | } |
1352 | ||
9420f89d MRI |
1353 | bad_page(page, |
1354 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
0e56acae AD |
1355 | } |
1356 | ||
1357 | /* | |
9420f89d | 1358 | * This page is about to be returned from the page allocator |
0e56acae | 1359 | */ |
77c7a095 | 1360 | static bool check_new_page(struct page *page) |
0e56acae | 1361 | { |
9420f89d MRI |
1362 | if (likely(page_expected_state(page, |
1363 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
77c7a095 | 1364 | return false; |
0e56acae | 1365 | |
9420f89d | 1366 | check_new_page_bad(page); |
77c7a095 | 1367 | return true; |
9420f89d | 1368 | } |
0e56acae | 1369 | |
9420f89d MRI |
1370 | static inline bool check_new_pages(struct page *page, unsigned int order) |
1371 | { | |
1372 | if (is_check_pages_enabled()) { | |
1373 | for (int i = 0; i < (1 << order); i++) { | |
1374 | struct page *p = page + i; | |
0e56acae | 1375 | |
8666925c | 1376 | if (check_new_page(p)) |
9420f89d | 1377 | return true; |
0e56acae AD |
1378 | } |
1379 | } | |
1380 | ||
9420f89d | 1381 | return false; |
0e56acae AD |
1382 | } |
1383 | ||
9420f89d | 1384 | static inline bool should_skip_kasan_unpoison(gfp_t flags) |
e4443149 | 1385 | { |
9420f89d MRI |
1386 | /* Don't skip if a software KASAN mode is enabled. */ |
1387 | if (IS_ENABLED(CONFIG_KASAN_GENERIC) || | |
1388 | IS_ENABLED(CONFIG_KASAN_SW_TAGS)) | |
1389 | return false; | |
e4443149 | 1390 | |
9420f89d MRI |
1391 | /* Skip, if hardware tag-based KASAN is not enabled. */ |
1392 | if (!kasan_hw_tags_enabled()) | |
1393 | return true; | |
e4443149 DJ |
1394 | |
1395 | /* | |
9420f89d MRI |
1396 | * With hardware tag-based KASAN enabled, skip if this has been |
1397 | * requested via __GFP_SKIP_KASAN. | |
e4443149 | 1398 | */ |
9420f89d | 1399 | return flags & __GFP_SKIP_KASAN; |
e4443149 DJ |
1400 | } |
1401 | ||
9420f89d | 1402 | static inline bool should_skip_init(gfp_t flags) |
ecd09650 | 1403 | { |
9420f89d MRI |
1404 | /* Don't skip, if hardware tag-based KASAN is not enabled. */ |
1405 | if (!kasan_hw_tags_enabled()) | |
1406 | return false; | |
1407 | ||
1408 | /* For hardware tag-based KASAN, skip if requested. */ | |
1409 | return (flags & __GFP_SKIP_ZERO); | |
ecd09650 DJ |
1410 | } |
1411 | ||
9420f89d MRI |
1412 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1413 | gfp_t gfp_flags) | |
7e18adb4 | 1414 | { |
9420f89d MRI |
1415 | bool init = !want_init_on_free() && want_init_on_alloc(gfp_flags) && |
1416 | !should_skip_init(gfp_flags); | |
1417 | bool zero_tags = init && (gfp_flags & __GFP_ZEROTAGS); | |
1418 | int i; | |
1419 | ||
1420 | set_page_private(page, 0); | |
1421 | set_page_refcounted(page); | |
0e1cc95b | 1422 | |
9420f89d MRI |
1423 | arch_alloc_page(page, order); |
1424 | debug_pagealloc_map_pages(page, 1 << order); | |
7e18adb4 | 1425 | |
3d060856 | 1426 | /* |
9420f89d MRI |
1427 | * Page unpoisoning must happen before memory initialization. |
1428 | * Otherwise, the poison pattern will be overwritten for __GFP_ZERO | |
1429 | * allocations and the page unpoisoning code will complain. | |
3d060856 | 1430 | */ |
9420f89d | 1431 | kernel_unpoison_pages(page, 1 << order); |
862b6dee | 1432 | |
1bb5eab3 AK |
1433 | /* |
1434 | * As memory initialization might be integrated into KASAN, | |
b42090ae | 1435 | * KASAN unpoisoning and memory initializion code must be |
1bb5eab3 AK |
1436 | * kept together to avoid discrepancies in behavior. |
1437 | */ | |
9294b128 AK |
1438 | |
1439 | /* | |
44383cef AK |
1440 | * If memory tags should be zeroed |
1441 | * (which happens only when memory should be initialized as well). | |
9294b128 | 1442 | */ |
44383cef | 1443 | if (zero_tags) { |
420ef683 | 1444 | /* Initialize both memory and memory tags. */ |
9294b128 AK |
1445 | for (i = 0; i != 1 << order; ++i) |
1446 | tag_clear_highpage(page + i); | |
1447 | ||
44383cef | 1448 | /* Take note that memory was initialized by the loop above. */ |
9294b128 AK |
1449 | init = false; |
1450 | } | |
0a54864f PC |
1451 | if (!should_skip_kasan_unpoison(gfp_flags) && |
1452 | kasan_unpoison_pages(page, order, init)) { | |
1453 | /* Take note that memory was initialized by KASAN. */ | |
1454 | if (kasan_has_integrated_init()) | |
1455 | init = false; | |
1456 | } else { | |
1457 | /* | |
1458 | * If memory tags have not been set by KASAN, reset the page | |
1459 | * tags to ensure page_address() dereferencing does not fault. | |
1460 | */ | |
70c248ac CM |
1461 | for (i = 0; i != 1 << order; ++i) |
1462 | page_kasan_tag_reset(page + i); | |
7a3b8353 | 1463 | } |
44383cef | 1464 | /* If memory is still not initialized, initialize it now. */ |
7e3cbba6 | 1465 | if (init) |
aeaec8e2 | 1466 | kernel_init_pages(page, 1 << order); |
1bb5eab3 AK |
1467 | |
1468 | set_page_owner(page, order, gfp_flags); | |
df4e817b | 1469 | page_table_check_alloc(page, order); |
dcfe378c | 1470 | pgalloc_tag_add(page, current, 1 << order); |
46f24fd8 JK |
1471 | } |
1472 | ||
479f854a | 1473 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1474 | unsigned int alloc_flags) |
2a7684a2 | 1475 | { |
46f24fd8 | 1476 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 1477 | |
17cf4406 NP |
1478 | if (order && (gfp_flags & __GFP_COMP)) |
1479 | prep_compound_page(page, order); | |
1480 | ||
75379191 | 1481 | /* |
2f064f34 | 1482 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1483 | * allocate the page. The expectation is that the caller is taking |
1484 | * steps that will free more memory. The caller should avoid the page | |
1485 | * being used for !PFMEMALLOC purposes. | |
1486 | */ | |
2f064f34 MH |
1487 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1488 | set_page_pfmemalloc(page); | |
1489 | else | |
1490 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
1491 | } |
1492 | ||
56fd56b8 MG |
1493 | /* |
1494 | * Go through the free lists for the given migratetype and remove | |
1495 | * the smallest available page from the freelists | |
1496 | */ | |
85ccc8fa | 1497 | static __always_inline |
728ec980 | 1498 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
1499 | int migratetype) |
1500 | { | |
1501 | unsigned int current_order; | |
b8af2941 | 1502 | struct free_area *area; |
56fd56b8 MG |
1503 | struct page *page; |
1504 | ||
1505 | /* Find a page of the appropriate size in the preferred list */ | |
fd377218 | 1506 | for (current_order = order; current_order < NR_PAGE_ORDERS; ++current_order) { |
56fd56b8 | 1507 | area = &(zone->free_area[current_order]); |
b03641af | 1508 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
1509 | if (!page) |
1510 | continue; | |
e0932b6c | 1511 | del_page_from_free_list(page, zone, current_order, migratetype); |
6ab01363 | 1512 | expand(zone, page, order, current_order, migratetype); |
10e0f753 WY |
1513 | trace_mm_page_alloc_zone_locked(page, order, migratetype, |
1514 | pcp_allowed_order(order) && | |
1515 | migratetype < MIGRATE_PCPTYPES); | |
56fd56b8 MG |
1516 | return page; |
1517 | } | |
1518 | ||
1519 | return NULL; | |
1520 | } | |
1521 | ||
1522 | ||
b2a0ac88 MG |
1523 | /* |
1524 | * This array describes the order lists are fallen back to when | |
1525 | * the free lists for the desirable migrate type are depleted | |
1dd214b8 ZY |
1526 | * |
1527 | * The other migratetypes do not have fallbacks. | |
b2a0ac88 | 1528 | */ |
0aac4566 | 1529 | static int fallbacks[MIGRATE_PCPTYPES][MIGRATE_PCPTYPES - 1] = { |
aa02d3c1 YD |
1530 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE }, |
1531 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE }, | |
1532 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE }, | |
b2a0ac88 MG |
1533 | }; |
1534 | ||
dc67647b | 1535 | #ifdef CONFIG_CMA |
85ccc8fa | 1536 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
1537 | unsigned int order) |
1538 | { | |
1539 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1540 | } | |
1541 | #else | |
1542 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1543 | unsigned int order) { return NULL; } | |
1544 | #endif | |
1545 | ||
c361be55 | 1546 | /* |
f37c0f68 ZY |
1547 | * Change the type of a block and move all its free pages to that |
1548 | * type's freelist. | |
c361be55 | 1549 | */ |
e1f42a57 VB |
1550 | static int __move_freepages_block(struct zone *zone, unsigned long start_pfn, |
1551 | int old_mt, int new_mt) | |
c361be55 MG |
1552 | { |
1553 | struct page *page; | |
e1f42a57 | 1554 | unsigned long pfn, end_pfn; |
d00181b9 | 1555 | unsigned int order; |
d100313f | 1556 | int pages_moved = 0; |
c361be55 | 1557 | |
f37c0f68 | 1558 | VM_WARN_ON(start_pfn & (pageblock_nr_pages - 1)); |
e1f42a57 | 1559 | end_pfn = pageblock_end_pfn(start_pfn); |
f37c0f68 | 1560 | |
e1f42a57 | 1561 | for (pfn = start_pfn; pfn < end_pfn;) { |
39ddb991 | 1562 | page = pfn_to_page(pfn); |
c361be55 | 1563 | if (!PageBuddy(page)) { |
39ddb991 | 1564 | pfn++; |
c361be55 MG |
1565 | continue; |
1566 | } | |
1567 | ||
cd961038 DR |
1568 | /* Make sure we are not inadvertently changing nodes */ |
1569 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
1570 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
1571 | ||
ab130f91 | 1572 | order = buddy_order(page); |
e0932b6c JW |
1573 | |
1574 | move_to_free_list(page, zone, order, old_mt, new_mt); | |
1575 | ||
39ddb991 | 1576 | pfn += 1 << order; |
d100313f | 1577 | pages_moved += 1 << order; |
c361be55 MG |
1578 | } |
1579 | ||
e0932b6c | 1580 | set_pageblock_migratetype(pfn_to_page(start_pfn), new_mt); |
f37c0f68 | 1581 | |
d100313f | 1582 | return pages_moved; |
c361be55 MG |
1583 | } |
1584 | ||
c0cd6f55 JW |
1585 | static bool prep_move_freepages_block(struct zone *zone, struct page *page, |
1586 | unsigned long *start_pfn, | |
c0cd6f55 | 1587 | int *num_free, int *num_movable) |
c361be55 | 1588 | { |
c0cd6f55 | 1589 | unsigned long pfn, start, end; |
4a222127 | 1590 | |
39ddb991 | 1591 | pfn = page_to_pfn(page); |
c0cd6f55 | 1592 | start = pageblock_start_pfn(pfn); |
e1f42a57 | 1593 | end = pageblock_end_pfn(pfn); |
c361be55 | 1594 | |
2dd482ba JW |
1595 | /* |
1596 | * The caller only has the lock for @zone, don't touch ranges | |
1597 | * that straddle into other zones. While we could move part of | |
1598 | * the range that's inside the zone, this call is usually | |
1599 | * accompanied by other operations such as migratetype updates | |
1600 | * which also should be locked. | |
1601 | */ | |
c0cd6f55 JW |
1602 | if (!zone_spans_pfn(zone, start)) |
1603 | return false; | |
e1f42a57 | 1604 | if (!zone_spans_pfn(zone, end - 1)) |
c0cd6f55 JW |
1605 | return false; |
1606 | ||
1607 | *start_pfn = start; | |
c0cd6f55 JW |
1608 | |
1609 | if (num_free) { | |
1610 | *num_free = 0; | |
1611 | *num_movable = 0; | |
e1f42a57 | 1612 | for (pfn = start; pfn < end;) { |
c0cd6f55 JW |
1613 | page = pfn_to_page(pfn); |
1614 | if (PageBuddy(page)) { | |
1615 | int nr = 1 << buddy_order(page); | |
1616 | ||
1617 | *num_free += nr; | |
1618 | pfn += nr; | |
1619 | continue; | |
1620 | } | |
1621 | /* | |
1622 | * We assume that pages that could be isolated for | |
1623 | * migration are movable. But we don't actually try | |
1624 | * isolating, as that would be expensive. | |
1625 | */ | |
1626 | if (PageLRU(page) || __PageMovable(page)) | |
1627 | (*num_movable)++; | |
1628 | pfn++; | |
1629 | } | |
1630 | } | |
c361be55 | 1631 | |
c0cd6f55 JW |
1632 | return true; |
1633 | } | |
1634 | ||
fd919a85 | 1635 | static int move_freepages_block(struct zone *zone, struct page *page, |
e0932b6c | 1636 | int old_mt, int new_mt) |
c0cd6f55 | 1637 | { |
e1f42a57 | 1638 | unsigned long start_pfn; |
c0cd6f55 | 1639 | |
e1f42a57 | 1640 | if (!prep_move_freepages_block(zone, page, &start_pfn, NULL, NULL)) |
c0cd6f55 JW |
1641 | return -1; |
1642 | ||
e1f42a57 | 1643 | return __move_freepages_block(zone, start_pfn, old_mt, new_mt); |
c361be55 MG |
1644 | } |
1645 | ||
fd919a85 JW |
1646 | #ifdef CONFIG_MEMORY_ISOLATION |
1647 | /* Look for a buddy that straddles start_pfn */ | |
1648 | static unsigned long find_large_buddy(unsigned long start_pfn) | |
1649 | { | |
1650 | int order = 0; | |
1651 | struct page *page; | |
1652 | unsigned long pfn = start_pfn; | |
1653 | ||
1654 | while (!PageBuddy(page = pfn_to_page(pfn))) { | |
1655 | /* Nothing found */ | |
1656 | if (++order > MAX_PAGE_ORDER) | |
1657 | return start_pfn; | |
1658 | pfn &= ~0UL << order; | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * Found a preceding buddy, but does it straddle? | |
1663 | */ | |
1664 | if (pfn + (1 << buddy_order(page)) > start_pfn) | |
1665 | return pfn; | |
1666 | ||
1667 | /* Nothing found */ | |
1668 | return start_pfn; | |
1669 | } | |
1670 | ||
1671 | /* Split a multi-block free page into its individual pageblocks */ | |
1672 | static void split_large_buddy(struct zone *zone, struct page *page, | |
1673 | unsigned long pfn, int order) | |
1674 | { | |
1675 | unsigned long end_pfn = pfn + (1 << order); | |
1676 | ||
1677 | VM_WARN_ON_ONCE(order <= pageblock_order); | |
1678 | VM_WARN_ON_ONCE(pfn & (pageblock_nr_pages - 1)); | |
1679 | ||
1680 | /* Caller removed page from freelist, buddy info cleared! */ | |
1681 | VM_WARN_ON_ONCE(PageBuddy(page)); | |
1682 | ||
1683 | while (pfn != end_pfn) { | |
1684 | int mt = get_pfnblock_migratetype(page, pfn); | |
1685 | ||
1686 | __free_one_page(page, pfn, zone, pageblock_order, mt, FPI_NONE); | |
1687 | pfn += pageblock_nr_pages; | |
1688 | page = pfn_to_page(pfn); | |
1689 | } | |
1690 | } | |
1691 | ||
1692 | /** | |
1693 | * move_freepages_block_isolate - move free pages in block for page isolation | |
1694 | * @zone: the zone | |
1695 | * @page: the pageblock page | |
1696 | * @migratetype: migratetype to set on the pageblock | |
1697 | * | |
1698 | * This is similar to move_freepages_block(), but handles the special | |
1699 | * case encountered in page isolation, where the block of interest | |
1700 | * might be part of a larger buddy spanning multiple pageblocks. | |
1701 | * | |
1702 | * Unlike the regular page allocator path, which moves pages while | |
1703 | * stealing buddies off the freelist, page isolation is interested in | |
1704 | * arbitrary pfn ranges that may have overlapping buddies on both ends. | |
1705 | * | |
1706 | * This function handles that. Straddling buddies are split into | |
1707 | * individual pageblocks. Only the block of interest is moved. | |
1708 | * | |
1709 | * Returns %true if pages could be moved, %false otherwise. | |
1710 | */ | |
1711 | bool move_freepages_block_isolate(struct zone *zone, struct page *page, | |
1712 | int migratetype) | |
1713 | { | |
e1f42a57 | 1714 | unsigned long start_pfn, pfn; |
fd919a85 | 1715 | |
e1f42a57 | 1716 | if (!prep_move_freepages_block(zone, page, &start_pfn, NULL, NULL)) |
fd919a85 JW |
1717 | return false; |
1718 | ||
1719 | /* No splits needed if buddies can't span multiple blocks */ | |
1720 | if (pageblock_order == MAX_PAGE_ORDER) | |
1721 | goto move; | |
1722 | ||
1723 | /* We're a tail block in a larger buddy */ | |
1724 | pfn = find_large_buddy(start_pfn); | |
1725 | if (pfn != start_pfn) { | |
1726 | struct page *buddy = pfn_to_page(pfn); | |
1727 | int order = buddy_order(buddy); | |
fd919a85 | 1728 | |
e0932b6c JW |
1729 | del_page_from_free_list(buddy, zone, order, |
1730 | get_pfnblock_migratetype(buddy, pfn)); | |
fd919a85 JW |
1731 | set_pageblock_migratetype(page, migratetype); |
1732 | split_large_buddy(zone, buddy, pfn, order); | |
1733 | return true; | |
1734 | } | |
1735 | ||
1736 | /* We're the starting block of a larger buddy */ | |
1737 | if (PageBuddy(page) && buddy_order(page) > pageblock_order) { | |
fd919a85 JW |
1738 | int order = buddy_order(page); |
1739 | ||
e0932b6c JW |
1740 | del_page_from_free_list(page, zone, order, |
1741 | get_pfnblock_migratetype(page, pfn)); | |
fd919a85 JW |
1742 | set_pageblock_migratetype(page, migratetype); |
1743 | split_large_buddy(zone, page, pfn, order); | |
1744 | return true; | |
1745 | } | |
1746 | move: | |
e1f42a57 VB |
1747 | __move_freepages_block(zone, start_pfn, |
1748 | get_pfnblock_migratetype(page, start_pfn), | |
1749 | migratetype); | |
fd919a85 JW |
1750 | return true; |
1751 | } | |
1752 | #endif /* CONFIG_MEMORY_ISOLATION */ | |
1753 | ||
2f66a68f MG |
1754 | static void change_pageblock_range(struct page *pageblock_page, |
1755 | int start_order, int migratetype) | |
1756 | { | |
1757 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1758 | ||
1759 | while (nr_pageblocks--) { | |
1760 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1761 | pageblock_page += pageblock_nr_pages; | |
1762 | } | |
1763 | } | |
1764 | ||
fef903ef | 1765 | /* |
9c0415eb VB |
1766 | * When we are falling back to another migratetype during allocation, try to |
1767 | * steal extra free pages from the same pageblocks to satisfy further | |
1768 | * allocations, instead of polluting multiple pageblocks. | |
1769 | * | |
1770 | * If we are stealing a relatively large buddy page, it is likely there will | |
1771 | * be more free pages in the pageblock, so try to steal them all. For | |
1772 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1773 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1774 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1775 | * pageblocks. | |
fef903ef | 1776 | */ |
4eb7dce6 JK |
1777 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1778 | { | |
1779 | /* | |
1780 | * Leaving this order check is intended, although there is | |
1781 | * relaxed order check in next check. The reason is that | |
1782 | * we can actually steal whole pageblock if this condition met, | |
1783 | * but, below check doesn't guarantee it and that is just heuristic | |
1784 | * so could be changed anytime. | |
1785 | */ | |
1786 | if (order >= pageblock_order) | |
1787 | return true; | |
1788 | ||
1789 | if (order >= pageblock_order / 2 || | |
1790 | start_mt == MIGRATE_RECLAIMABLE || | |
1791 | start_mt == MIGRATE_UNMOVABLE || | |
1792 | page_group_by_mobility_disabled) | |
1793 | return true; | |
1794 | ||
1795 | return false; | |
1796 | } | |
1797 | ||
597c8920 | 1798 | static inline bool boost_watermark(struct zone *zone) |
1c30844d MG |
1799 | { |
1800 | unsigned long max_boost; | |
1801 | ||
1802 | if (!watermark_boost_factor) | |
597c8920 | 1803 | return false; |
14f69140 HW |
1804 | /* |
1805 | * Don't bother in zones that are unlikely to produce results. | |
1806 | * On small machines, including kdump capture kernels running | |
1807 | * in a small area, boosting the watermark can cause an out of | |
1808 | * memory situation immediately. | |
1809 | */ | |
1810 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
597c8920 | 1811 | return false; |
1c30844d MG |
1812 | |
1813 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
1814 | watermark_boost_factor, 10000); | |
94b3334c MG |
1815 | |
1816 | /* | |
1817 | * high watermark may be uninitialised if fragmentation occurs | |
1818 | * very early in boot so do not boost. We do not fall | |
1819 | * through and boost by pageblock_nr_pages as failing | |
1820 | * allocations that early means that reclaim is not going | |
1821 | * to help and it may even be impossible to reclaim the | |
1822 | * boosted watermark resulting in a hang. | |
1823 | */ | |
1824 | if (!max_boost) | |
597c8920 | 1825 | return false; |
94b3334c | 1826 | |
1c30844d MG |
1827 | max_boost = max(pageblock_nr_pages, max_boost); |
1828 | ||
1829 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
1830 | max_boost); | |
597c8920 JW |
1831 | |
1832 | return true; | |
1c30844d MG |
1833 | } |
1834 | ||
4eb7dce6 | 1835 | /* |
c0cd6f55 JW |
1836 | * This function implements actual steal behaviour. If order is large enough, we |
1837 | * can claim the whole pageblock for the requested migratetype. If not, we check | |
1838 | * the pageblock for constituent pages; if at least half of the pages are free | |
1839 | * or compatible, we can still claim the whole block, so pages freed in the | |
1840 | * future will be put on the correct free list. Otherwise, we isolate exactly | |
1841 | * the order we need from the fallback block and leave its migratetype alone. | |
4eb7dce6 | 1842 | */ |
c0cd6f55 JW |
1843 | static struct page * |
1844 | steal_suitable_fallback(struct zone *zone, struct page *page, | |
1845 | int current_order, int order, int start_type, | |
1846 | unsigned int alloc_flags, bool whole_block) | |
fef903ef | 1847 | { |
02aa0cdd | 1848 | int free_pages, movable_pages, alike_pages; |
e1f42a57 | 1849 | unsigned long start_pfn; |
c0cd6f55 | 1850 | int block_type; |
02aa0cdd | 1851 | |
c0cd6f55 | 1852 | block_type = get_pageblock_migratetype(page); |
fef903ef | 1853 | |
3bc48f96 VB |
1854 | /* |
1855 | * This can happen due to races and we want to prevent broken | |
1856 | * highatomic accounting. | |
1857 | */ | |
c0cd6f55 | 1858 | if (is_migrate_highatomic(block_type)) |
3bc48f96 VB |
1859 | goto single_page; |
1860 | ||
fef903ef SB |
1861 | /* Take ownership for orders >= pageblock_order */ |
1862 | if (current_order >= pageblock_order) { | |
e0932b6c | 1863 | del_page_from_free_list(page, zone, current_order, block_type); |
fef903ef | 1864 | change_pageblock_range(page, current_order, start_type); |
c0cd6f55 JW |
1865 | expand(zone, page, order, current_order, start_type); |
1866 | return page; | |
fef903ef SB |
1867 | } |
1868 | ||
1c30844d MG |
1869 | /* |
1870 | * Boost watermarks to increase reclaim pressure to reduce the | |
1871 | * likelihood of future fallbacks. Wake kswapd now as the node | |
1872 | * may be balanced overall and kswapd will not wake naturally. | |
1873 | */ | |
597c8920 | 1874 | if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD)) |
73444bc4 | 1875 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 1876 | |
3bc48f96 VB |
1877 | /* We are not allowed to try stealing from the whole block */ |
1878 | if (!whole_block) | |
1879 | goto single_page; | |
1880 | ||
ebddd111 | 1881 | /* moving whole block can fail due to zone boundary conditions */ |
e1f42a57 VB |
1882 | if (!prep_move_freepages_block(zone, page, &start_pfn, &free_pages, |
1883 | &movable_pages)) | |
ebddd111 ML |
1884 | goto single_page; |
1885 | ||
02aa0cdd VB |
1886 | /* |
1887 | * Determine how many pages are compatible with our allocation. | |
1888 | * For movable allocation, it's the number of movable pages which | |
1889 | * we just obtained. For other types it's a bit more tricky. | |
1890 | */ | |
1891 | if (start_type == MIGRATE_MOVABLE) { | |
1892 | alike_pages = movable_pages; | |
1893 | } else { | |
1894 | /* | |
1895 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
1896 | * to MOVABLE pageblock, consider all non-movable pages as | |
1897 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
1898 | * vice versa, be conservative since we can't distinguish the | |
1899 | * exact migratetype of non-movable pages. | |
1900 | */ | |
c0cd6f55 | 1901 | if (block_type == MIGRATE_MOVABLE) |
02aa0cdd VB |
1902 | alike_pages = pageblock_nr_pages |
1903 | - (free_pages + movable_pages); | |
1904 | else | |
1905 | alike_pages = 0; | |
1906 | } | |
02aa0cdd VB |
1907 | /* |
1908 | * If a sufficient number of pages in the block are either free or of | |
ebddd111 | 1909 | * compatible migratability as our allocation, claim the whole block. |
02aa0cdd VB |
1910 | */ |
1911 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
c0cd6f55 | 1912 | page_group_by_mobility_disabled) { |
e1f42a57 | 1913 | __move_freepages_block(zone, start_pfn, block_type, start_type); |
c0cd6f55 JW |
1914 | return __rmqueue_smallest(zone, order, start_type); |
1915 | } | |
3bc48f96 VB |
1916 | |
1917 | single_page: | |
e0932b6c | 1918 | del_page_from_free_list(page, zone, current_order, block_type); |
c0cd6f55 JW |
1919 | expand(zone, page, order, current_order, block_type); |
1920 | return page; | |
4eb7dce6 JK |
1921 | } |
1922 | ||
2149cdae JK |
1923 | /* |
1924 | * Check whether there is a suitable fallback freepage with requested order. | |
1925 | * If only_stealable is true, this function returns fallback_mt only if | |
1926 | * we can steal other freepages all together. This would help to reduce | |
1927 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1928 | */ | |
1929 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1930 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1931 | { |
1932 | int i; | |
1933 | int fallback_mt; | |
1934 | ||
1935 | if (area->nr_free == 0) | |
1936 | return -1; | |
1937 | ||
1938 | *can_steal = false; | |
aa02d3c1 | 1939 | for (i = 0; i < MIGRATE_PCPTYPES - 1 ; i++) { |
4eb7dce6 | 1940 | fallback_mt = fallbacks[migratetype][i]; |
b03641af | 1941 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 1942 | continue; |
fef903ef | 1943 | |
4eb7dce6 JK |
1944 | if (can_steal_fallback(order, migratetype)) |
1945 | *can_steal = true; | |
1946 | ||
2149cdae JK |
1947 | if (!only_stealable) |
1948 | return fallback_mt; | |
1949 | ||
1950 | if (*can_steal) | |
1951 | return fallback_mt; | |
fef903ef | 1952 | } |
4eb7dce6 JK |
1953 | |
1954 | return -1; | |
fef903ef SB |
1955 | } |
1956 | ||
0aaa29a5 MG |
1957 | /* |
1958 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
1959 | * there are no empty page blocks that contain a page with a suitable order | |
1960 | */ | |
368d983b | 1961 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone) |
0aaa29a5 MG |
1962 | { |
1963 | int mt; | |
1964 | unsigned long max_managed, flags; | |
1965 | ||
1966 | /* | |
d68e39fc | 1967 | * The number reserved as: minimum is 1 pageblock, maximum is |
9cd20f3f CTK |
1968 | * roughly 1% of a zone. But if 1% of a zone falls below a |
1969 | * pageblock size, then don't reserve any pageblocks. | |
0aaa29a5 MG |
1970 | * Check is race-prone but harmless. |
1971 | */ | |
9cd20f3f CTK |
1972 | if ((zone_managed_pages(zone) / 100) < pageblock_nr_pages) |
1973 | return; | |
d68e39fc | 1974 | max_managed = ALIGN((zone_managed_pages(zone) / 100), pageblock_nr_pages); |
0aaa29a5 MG |
1975 | if (zone->nr_reserved_highatomic >= max_managed) |
1976 | return; | |
1977 | ||
1978 | spin_lock_irqsave(&zone->lock, flags); | |
1979 | ||
1980 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
1981 | if (zone->nr_reserved_highatomic >= max_managed) | |
1982 | goto out_unlock; | |
1983 | ||
1984 | /* Yoink! */ | |
1985 | mt = get_pageblock_migratetype(page); | |
1dd214b8 | 1986 | /* Only reserve normal pageblocks (i.e., they can merge with others) */ |
f37c0f68 | 1987 | if (migratetype_is_mergeable(mt)) |
e0932b6c | 1988 | if (move_freepages_block(zone, page, mt, |
f37c0f68 | 1989 | MIGRATE_HIGHATOMIC) != -1) |
c0cd6f55 | 1990 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
0aaa29a5 MG |
1991 | |
1992 | out_unlock: | |
1993 | spin_unlock_irqrestore(&zone->lock, flags); | |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * Used when an allocation is about to fail under memory pressure. This | |
1998 | * potentially hurts the reliability of high-order allocations when under | |
1999 | * intense memory pressure but failed atomic allocations should be easier | |
2000 | * to recover from than an OOM. | |
29fac03b MK |
2001 | * |
2002 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2003 | * pageblock is exhausted. | |
0aaa29a5 | 2004 | */ |
29fac03b MK |
2005 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2006 | bool force) | |
0aaa29a5 MG |
2007 | { |
2008 | struct zonelist *zonelist = ac->zonelist; | |
2009 | unsigned long flags; | |
2010 | struct zoneref *z; | |
2011 | struct zone *zone; | |
2012 | struct page *page; | |
2013 | int order; | |
c0cd6f55 | 2014 | int ret; |
0aaa29a5 | 2015 | |
97a225e6 | 2016 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2017 | ac->nodemask) { |
29fac03b MK |
2018 | /* |
2019 | * Preserve at least one pageblock unless memory pressure | |
2020 | * is really high. | |
2021 | */ | |
2022 | if (!force && zone->nr_reserved_highatomic <= | |
2023 | pageblock_nr_pages) | |
0aaa29a5 MG |
2024 | continue; |
2025 | ||
2026 | spin_lock_irqsave(&zone->lock, flags); | |
fd377218 | 2027 | for (order = 0; order < NR_PAGE_ORDERS; order++) { |
0aaa29a5 | 2028 | struct free_area *area = &(zone->free_area[order]); |
e0932b6c | 2029 | int mt; |
0aaa29a5 | 2030 | |
b03641af | 2031 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2032 | if (!page) |
0aaa29a5 MG |
2033 | continue; |
2034 | ||
e0932b6c | 2035 | mt = get_pageblock_migratetype(page); |
0aaa29a5 | 2036 | /* |
4855e4a7 MK |
2037 | * In page freeing path, migratetype change is racy so |
2038 | * we can counter several free pages in a pageblock | |
f0953a1b | 2039 | * in this loop although we changed the pageblock type |
4855e4a7 MK |
2040 | * from highatomic to ac->migratetype. So we should |
2041 | * adjust the count once. | |
0aaa29a5 | 2042 | */ |
e0932b6c | 2043 | if (is_migrate_highatomic(mt)) { |
4855e4a7 MK |
2044 | /* |
2045 | * It should never happen but changes to | |
2046 | * locking could inadvertently allow a per-cpu | |
2047 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2048 | * while unreserving so be safe and watch for | |
2049 | * underflows. | |
2050 | */ | |
2051 | zone->nr_reserved_highatomic -= min( | |
2052 | pageblock_nr_pages, | |
2053 | zone->nr_reserved_highatomic); | |
2054 | } | |
0aaa29a5 MG |
2055 | |
2056 | /* | |
2057 | * Convert to ac->migratetype and avoid the normal | |
2058 | * pageblock stealing heuristics. Minimally, the caller | |
2059 | * is doing the work and needs the pages. More | |
2060 | * importantly, if the block was always converted to | |
2061 | * MIGRATE_UNMOVABLE or another type then the number | |
2062 | * of pageblocks that cannot be completely freed | |
2063 | * may increase. | |
2064 | */ | |
e0932b6c JW |
2065 | ret = move_freepages_block(zone, page, mt, |
2066 | ac->migratetype); | |
c0cd6f55 JW |
2067 | /* |
2068 | * Reserving this block already succeeded, so this should | |
2069 | * not fail on zone boundaries. | |
2070 | */ | |
2071 | WARN_ON_ONCE(ret == -1); | |
c0cd6f55 | 2072 | if (ret > 0) { |
29fac03b MK |
2073 | spin_unlock_irqrestore(&zone->lock, flags); |
2074 | return ret; | |
2075 | } | |
0aaa29a5 MG |
2076 | } |
2077 | spin_unlock_irqrestore(&zone->lock, flags); | |
2078 | } | |
04c8716f MK |
2079 | |
2080 | return false; | |
0aaa29a5 MG |
2081 | } |
2082 | ||
3bc48f96 VB |
2083 | /* |
2084 | * Try finding a free buddy page on the fallback list and put it on the free | |
2085 | * list of requested migratetype, possibly along with other pages from the same | |
2086 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2087 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2088 | * |
2089 | * The use of signed ints for order and current_order is a deliberate | |
2090 | * deviation from the rest of this file, to make the for loop | |
2091 | * condition simpler. | |
3bc48f96 | 2092 | */ |
c0cd6f55 | 2093 | static __always_inline struct page * |
6bb15450 MG |
2094 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2095 | unsigned int alloc_flags) | |
b2a0ac88 | 2096 | { |
b8af2941 | 2097 | struct free_area *area; |
b002529d | 2098 | int current_order; |
6bb15450 | 2099 | int min_order = order; |
b2a0ac88 | 2100 | struct page *page; |
4eb7dce6 JK |
2101 | int fallback_mt; |
2102 | bool can_steal; | |
b2a0ac88 | 2103 | |
6bb15450 MG |
2104 | /* |
2105 | * Do not steal pages from freelists belonging to other pageblocks | |
2106 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2107 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2108 | */ | |
e933dc4a | 2109 | if (order < pageblock_order && alloc_flags & ALLOC_NOFRAGMENT) |
6bb15450 MG |
2110 | min_order = pageblock_order; |
2111 | ||
7a8f58f3 VB |
2112 | /* |
2113 | * Find the largest available free page in the other list. This roughly | |
2114 | * approximates finding the pageblock with the most free pages, which | |
2115 | * would be too costly to do exactly. | |
2116 | */ | |
5e0a760b | 2117 | for (current_order = MAX_PAGE_ORDER; current_order >= min_order; |
7aeb09f9 | 2118 | --current_order) { |
4eb7dce6 JK |
2119 | area = &(zone->free_area[current_order]); |
2120 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2121 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2122 | if (fallback_mt == -1) |
2123 | continue; | |
b2a0ac88 | 2124 | |
7a8f58f3 VB |
2125 | /* |
2126 | * We cannot steal all free pages from the pageblock and the | |
2127 | * requested migratetype is movable. In that case it's better to | |
2128 | * steal and split the smallest available page instead of the | |
2129 | * largest available page, because even if the next movable | |
2130 | * allocation falls back into a different pageblock than this | |
2131 | * one, it won't cause permanent fragmentation. | |
2132 | */ | |
2133 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2134 | && current_order > order) | |
2135 | goto find_smallest; | |
b2a0ac88 | 2136 | |
7a8f58f3 VB |
2137 | goto do_steal; |
2138 | } | |
e0fff1bd | 2139 | |
c0cd6f55 | 2140 | return NULL; |
e0fff1bd | 2141 | |
7a8f58f3 | 2142 | find_smallest: |
fd377218 | 2143 | for (current_order = order; current_order < NR_PAGE_ORDERS; current_order++) { |
7a8f58f3 VB |
2144 | area = &(zone->free_area[current_order]); |
2145 | fallback_mt = find_suitable_fallback(area, current_order, | |
2146 | start_migratetype, false, &can_steal); | |
2147 | if (fallback_mt != -1) | |
2148 | break; | |
b2a0ac88 MG |
2149 | } |
2150 | ||
7a8f58f3 VB |
2151 | /* |
2152 | * This should not happen - we already found a suitable fallback | |
2153 | * when looking for the largest page. | |
2154 | */ | |
5e0a760b | 2155 | VM_BUG_ON(current_order > MAX_PAGE_ORDER); |
7a8f58f3 VB |
2156 | |
2157 | do_steal: | |
b03641af | 2158 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2159 | |
c0cd6f55 JW |
2160 | /* take off list, maybe claim block, expand remainder */ |
2161 | page = steal_suitable_fallback(zone, page, current_order, order, | |
2162 | start_migratetype, alloc_flags, can_steal); | |
7a8f58f3 VB |
2163 | |
2164 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2165 | start_migratetype, fallback_mt); | |
2166 | ||
c0cd6f55 | 2167 | return page; |
b2a0ac88 MG |
2168 | } |
2169 | ||
56fd56b8 | 2170 | /* |
1da177e4 LT |
2171 | * Do the hard work of removing an element from the buddy allocator. |
2172 | * Call me with the zone->lock already held. | |
2173 | */ | |
85ccc8fa | 2174 | static __always_inline struct page * |
6bb15450 MG |
2175 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2176 | unsigned int alloc_flags) | |
1da177e4 | 2177 | { |
1da177e4 LT |
2178 | struct page *page; |
2179 | ||
ce8f86ee H |
2180 | if (IS_ENABLED(CONFIG_CMA)) { |
2181 | /* | |
2182 | * Balance movable allocations between regular and CMA areas by | |
2183 | * allocating from CMA when over half of the zone's free memory | |
2184 | * is in the CMA area. | |
2185 | */ | |
2186 | if (alloc_flags & ALLOC_CMA && | |
2187 | zone_page_state(zone, NR_FREE_CMA_PAGES) > | |
2188 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
2189 | page = __rmqueue_cma_fallback(zone, order); | |
2190 | if (page) | |
10e0f753 | 2191 | return page; |
ce8f86ee | 2192 | } |
16867664 | 2193 | } |
c0cd6f55 | 2194 | |
56fd56b8 | 2195 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2196 | if (unlikely(!page)) { |
8510e69c | 2197 | if (alloc_flags & ALLOC_CMA) |
dc67647b JK |
2198 | page = __rmqueue_cma_fallback(zone, order); |
2199 | ||
c0cd6f55 JW |
2200 | if (!page) |
2201 | page = __rmqueue_fallback(zone, order, migratetype, | |
2202 | alloc_flags); | |
728ec980 | 2203 | } |
b2a0ac88 | 2204 | return page; |
1da177e4 LT |
2205 | } |
2206 | ||
5f63b720 | 2207 | /* |
1da177e4 LT |
2208 | * Obtain a specified number of elements from the buddy allocator, all under |
2209 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2210 | * Returns the number of new pages which were placed at *list. | |
2211 | */ | |
5f63b720 | 2212 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2213 | unsigned long count, struct list_head *list, |
6bb15450 | 2214 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2215 | { |
57490774 | 2216 | unsigned long flags; |
700d2e9a | 2217 | int i; |
5f63b720 | 2218 | |
57490774 | 2219 | spin_lock_irqsave(&zone->lock, flags); |
1da177e4 | 2220 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2221 | struct page *page = __rmqueue(zone, order, migratetype, |
2222 | alloc_flags); | |
085cc7d5 | 2223 | if (unlikely(page == NULL)) |
1da177e4 | 2224 | break; |
81eabcbe MG |
2225 | |
2226 | /* | |
0fac3ba5 VB |
2227 | * Split buddy pages returned by expand() are received here in |
2228 | * physical page order. The page is added to the tail of | |
2229 | * caller's list. From the callers perspective, the linked list | |
2230 | * is ordered by page number under some conditions. This is | |
2231 | * useful for IO devices that can forward direction from the | |
2232 | * head, thus also in the physical page order. This is useful | |
2233 | * for IO devices that can merge IO requests if the physical | |
2234 | * pages are ordered properly. | |
81eabcbe | 2235 | */ |
bf75f200 | 2236 | list_add_tail(&page->pcp_list, list); |
1da177e4 | 2237 | } |
57490774 | 2238 | spin_unlock_irqrestore(&zone->lock, flags); |
2ede3c13 | 2239 | |
700d2e9a | 2240 | return i; |
1da177e4 LT |
2241 | } |
2242 | ||
51a755c5 HY |
2243 | /* |
2244 | * Called from the vmstat counter updater to decay the PCP high. | |
2245 | * Return whether there are addition works to do. | |
2246 | */ | |
2247 | int decay_pcp_high(struct zone *zone, struct per_cpu_pages *pcp) | |
2248 | { | |
2249 | int high_min, to_drain, batch; | |
2250 | int todo = 0; | |
2251 | ||
2252 | high_min = READ_ONCE(pcp->high_min); | |
2253 | batch = READ_ONCE(pcp->batch); | |
2254 | /* | |
2255 | * Decrease pcp->high periodically to try to free possible | |
2256 | * idle PCP pages. And, avoid to free too many pages to | |
2257 | * control latency. This caps pcp->high decrement too. | |
2258 | */ | |
2259 | if (pcp->high > high_min) { | |
2260 | pcp->high = max3(pcp->count - (batch << CONFIG_PCP_BATCH_SCALE_MAX), | |
2261 | pcp->high - (pcp->high >> 3), high_min); | |
2262 | if (pcp->high > high_min) | |
2263 | todo++; | |
2264 | } | |
2265 | ||
2266 | to_drain = pcp->count - pcp->high; | |
2267 | if (to_drain > 0) { | |
2268 | spin_lock(&pcp->lock); | |
2269 | free_pcppages_bulk(zone, to_drain, pcp, 0); | |
2270 | spin_unlock(&pcp->lock); | |
2271 | todo++; | |
2272 | } | |
2273 | ||
2274 | return todo; | |
2275 | } | |
2276 | ||
4ae7c039 | 2277 | #ifdef CONFIG_NUMA |
8fce4d8e | 2278 | /* |
4037d452 CL |
2279 | * Called from the vmstat counter updater to drain pagesets of this |
2280 | * currently executing processor on remote nodes after they have | |
2281 | * expired. | |
8fce4d8e | 2282 | */ |
4037d452 | 2283 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2284 | { |
7be12fc9 | 2285 | int to_drain, batch; |
4ae7c039 | 2286 | |
4db0c3c2 | 2287 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2288 | to_drain = min(pcp->count, batch); |
4b23a68f | 2289 | if (to_drain > 0) { |
57490774 | 2290 | spin_lock(&pcp->lock); |
fd56eef2 | 2291 | free_pcppages_bulk(zone, to_drain, pcp, 0); |
57490774 | 2292 | spin_unlock(&pcp->lock); |
4b23a68f | 2293 | } |
4ae7c039 CL |
2294 | } |
2295 | #endif | |
2296 | ||
9f8f2172 | 2297 | /* |
93481ff0 | 2298 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 | 2299 | */ |
93481ff0 | 2300 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2301 | { |
55f77df7 LS |
2302 | struct per_cpu_pages *pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
2303 | int count = READ_ONCE(pcp->count); | |
2304 | ||
2305 | while (count) { | |
2306 | int to_drain = min(count, pcp->batch << CONFIG_PCP_BATCH_SCALE_MAX); | |
2307 | count -= to_drain; | |
1da177e4 | 2308 | |
57490774 | 2309 | spin_lock(&pcp->lock); |
55f77df7 | 2310 | free_pcppages_bulk(zone, to_drain, pcp, 0); |
57490774 | 2311 | spin_unlock(&pcp->lock); |
4b23a68f | 2312 | } |
93481ff0 | 2313 | } |
3dfa5721 | 2314 | |
93481ff0 VB |
2315 | /* |
2316 | * Drain pcplists of all zones on the indicated processor. | |
93481ff0 VB |
2317 | */ |
2318 | static void drain_pages(unsigned int cpu) | |
2319 | { | |
2320 | struct zone *zone; | |
2321 | ||
2322 | for_each_populated_zone(zone) { | |
2323 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2324 | } |
2325 | } | |
1da177e4 | 2326 | |
9f8f2172 CL |
2327 | /* |
2328 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
2329 | */ | |
93481ff0 | 2330 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2331 | { |
93481ff0 VB |
2332 | int cpu = smp_processor_id(); |
2333 | ||
2334 | if (zone) | |
2335 | drain_pages_zone(cpu, zone); | |
2336 | else | |
2337 | drain_pages(cpu); | |
9f8f2172 CL |
2338 | } |
2339 | ||
2340 | /* | |
ec6e8c7e VB |
2341 | * The implementation of drain_all_pages(), exposing an extra parameter to |
2342 | * drain on all cpus. | |
93481ff0 | 2343 | * |
ec6e8c7e VB |
2344 | * drain_all_pages() is optimized to only execute on cpus where pcplists are |
2345 | * not empty. The check for non-emptiness can however race with a free to | |
2346 | * pcplist that has not yet increased the pcp->count from 0 to 1. Callers | |
2347 | * that need the guarantee that every CPU has drained can disable the | |
2348 | * optimizing racy check. | |
9f8f2172 | 2349 | */ |
3b1f3658 | 2350 | static void __drain_all_pages(struct zone *zone, bool force_all_cpus) |
9f8f2172 | 2351 | { |
74046494 | 2352 | int cpu; |
74046494 GBY |
2353 | |
2354 | /* | |
041711ce | 2355 | * Allocate in the BSS so we won't require allocation in |
74046494 GBY |
2356 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y |
2357 | */ | |
2358 | static cpumask_t cpus_with_pcps; | |
2359 | ||
bd233f53 MG |
2360 | /* |
2361 | * Do not drain if one is already in progress unless it's specific to | |
2362 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2363 | * the drain to be complete when the call returns. | |
2364 | */ | |
2365 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2366 | if (!zone) | |
2367 | return; | |
2368 | mutex_lock(&pcpu_drain_mutex); | |
2369 | } | |
0ccce3b9 | 2370 | |
74046494 GBY |
2371 | /* |
2372 | * We don't care about racing with CPU hotplug event | |
2373 | * as offline notification will cause the notified | |
2374 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2375 | * disables preemption as part of its processing | |
2376 | */ | |
2377 | for_each_online_cpu(cpu) { | |
28f836b6 | 2378 | struct per_cpu_pages *pcp; |
93481ff0 | 2379 | struct zone *z; |
74046494 | 2380 | bool has_pcps = false; |
93481ff0 | 2381 | |
ec6e8c7e VB |
2382 | if (force_all_cpus) { |
2383 | /* | |
2384 | * The pcp.count check is racy, some callers need a | |
2385 | * guarantee that no cpu is missed. | |
2386 | */ | |
2387 | has_pcps = true; | |
2388 | } else if (zone) { | |
28f836b6 MG |
2389 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
2390 | if (pcp->count) | |
74046494 | 2391 | has_pcps = true; |
93481ff0 VB |
2392 | } else { |
2393 | for_each_populated_zone(z) { | |
28f836b6 MG |
2394 | pcp = per_cpu_ptr(z->per_cpu_pageset, cpu); |
2395 | if (pcp->count) { | |
93481ff0 VB |
2396 | has_pcps = true; |
2397 | break; | |
2398 | } | |
74046494 GBY |
2399 | } |
2400 | } | |
93481ff0 | 2401 | |
74046494 GBY |
2402 | if (has_pcps) |
2403 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2404 | else | |
2405 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2406 | } | |
0ccce3b9 | 2407 | |
bd233f53 | 2408 | for_each_cpu(cpu, &cpus_with_pcps) { |
443c2acc NSJ |
2409 | if (zone) |
2410 | drain_pages_zone(cpu, zone); | |
2411 | else | |
2412 | drain_pages(cpu); | |
0ccce3b9 | 2413 | } |
bd233f53 MG |
2414 | |
2415 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2416 | } |
2417 | ||
ec6e8c7e VB |
2418 | /* |
2419 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. | |
2420 | * | |
2421 | * When zone parameter is non-NULL, spill just the single zone's pages. | |
ec6e8c7e VB |
2422 | */ |
2423 | void drain_all_pages(struct zone *zone) | |
2424 | { | |
2425 | __drain_all_pages(zone, false); | |
2426 | } | |
2427 | ||
51a755c5 | 2428 | static int nr_pcp_free(struct per_cpu_pages *pcp, int batch, int high, bool free_high) |
3b12e7e9 MG |
2429 | { |
2430 | int min_nr_free, max_nr_free; | |
2431 | ||
51a755c5 | 2432 | /* Free as much as possible if batch freeing high-order pages. */ |
f26b3fa0 | 2433 | if (unlikely(free_high)) |
51a755c5 | 2434 | return min(pcp->count, batch << CONFIG_PCP_BATCH_SCALE_MAX); |
f26b3fa0 | 2435 | |
3b12e7e9 MG |
2436 | /* Check for PCP disabled or boot pageset */ |
2437 | if (unlikely(high < batch)) | |
2438 | return 1; | |
2439 | ||
2440 | /* Leave at least pcp->batch pages on the list */ | |
2441 | min_nr_free = batch; | |
2442 | max_nr_free = high - batch; | |
2443 | ||
2444 | /* | |
6ccdcb6d HY |
2445 | * Increase the batch number to the number of the consecutive |
2446 | * freed pages to reduce zone lock contention. | |
3b12e7e9 | 2447 | */ |
6ccdcb6d | 2448 | batch = clamp_t(int, pcp->free_count, min_nr_free, max_nr_free); |
3b12e7e9 MG |
2449 | |
2450 | return batch; | |
2451 | } | |
2452 | ||
f26b3fa0 | 2453 | static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone, |
51a755c5 | 2454 | int batch, bool free_high) |
c49c2c47 | 2455 | { |
51a755c5 | 2456 | int high, high_min, high_max; |
c49c2c47 | 2457 | |
51a755c5 HY |
2458 | high_min = READ_ONCE(pcp->high_min); |
2459 | high_max = READ_ONCE(pcp->high_max); | |
2460 | high = pcp->high = clamp(pcp->high, high_min, high_max); | |
2461 | ||
2462 | if (unlikely(!high)) | |
c49c2c47 MG |
2463 | return 0; |
2464 | ||
51a755c5 HY |
2465 | if (unlikely(free_high)) { |
2466 | pcp->high = max(high - (batch << CONFIG_PCP_BATCH_SCALE_MAX), | |
2467 | high_min); | |
2468 | return 0; | |
2469 | } | |
c49c2c47 MG |
2470 | |
2471 | /* | |
2472 | * If reclaim is active, limit the number of pages that can be | |
2473 | * stored on pcp lists | |
2474 | */ | |
51a755c5 | 2475 | if (test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags)) { |
6ccdcb6d HY |
2476 | int free_count = max_t(int, pcp->free_count, batch); |
2477 | ||
2478 | pcp->high = max(high - free_count, high_min); | |
51a755c5 HY |
2479 | return min(batch << 2, pcp->high); |
2480 | } | |
2481 | ||
57c0419c HY |
2482 | if (high_min == high_max) |
2483 | return high; | |
2484 | ||
2485 | if (test_bit(ZONE_BELOW_HIGH, &zone->flags)) { | |
6ccdcb6d HY |
2486 | int free_count = max_t(int, pcp->free_count, batch); |
2487 | ||
2488 | pcp->high = max(high - free_count, high_min); | |
57c0419c HY |
2489 | high = max(pcp->count, high_min); |
2490 | } else if (pcp->count >= high) { | |
6ccdcb6d | 2491 | int need_high = pcp->free_count + batch; |
51a755c5 HY |
2492 | |
2493 | /* pcp->high should be large enough to hold batch freed pages */ | |
2494 | if (pcp->high < need_high) | |
2495 | pcp->high = clamp(need_high, high_min, high_max); | |
2496 | } | |
2497 | ||
2498 | return high; | |
c49c2c47 MG |
2499 | } |
2500 | ||
4b23a68f MG |
2501 | static void free_unref_page_commit(struct zone *zone, struct per_cpu_pages *pcp, |
2502 | struct page *page, int migratetype, | |
56651377 | 2503 | unsigned int order) |
9cca35d4 | 2504 | { |
51a755c5 | 2505 | int high, batch; |
44042b44 | 2506 | int pindex; |
ca71fe1a | 2507 | bool free_high = false; |
9cca35d4 | 2508 | |
c0a24239 HY |
2509 | /* |
2510 | * On freeing, reduce the number of pages that are batch allocated. | |
2511 | * See nr_pcp_alloc() where alloc_factor is increased for subsequent | |
2512 | * allocations. | |
2513 | */ | |
2514 | pcp->alloc_factor >>= 1; | |
15cd9004 | 2515 | __count_vm_events(PGFREE, 1 << order); |
44042b44 | 2516 | pindex = order_to_pindex(migratetype, order); |
bf75f200 | 2517 | list_add(&page->pcp_list, &pcp->lists[pindex]); |
44042b44 | 2518 | pcp->count += 1 << order; |
f26b3fa0 | 2519 | |
51a755c5 | 2520 | batch = READ_ONCE(pcp->batch); |
f26b3fa0 MG |
2521 | /* |
2522 | * As high-order pages other than THP's stored on PCP can contribute | |
2523 | * to fragmentation, limit the number stored when PCP is heavily | |
2524 | * freeing without allocation. The remainder after bulk freeing | |
2525 | * stops will be drained from vmstat refresh context. | |
2526 | */ | |
ca71fe1a | 2527 | if (order && order <= PAGE_ALLOC_COSTLY_ORDER) { |
6ccdcb6d | 2528 | free_high = (pcp->free_count >= batch && |
362d37a1 HY |
2529 | (pcp->flags & PCPF_PREV_FREE_HIGH_ORDER) && |
2530 | (!(pcp->flags & PCPF_FREE_HIGH_BATCH) || | |
51a755c5 | 2531 | pcp->count >= READ_ONCE(batch))); |
ca71fe1a HY |
2532 | pcp->flags |= PCPF_PREV_FREE_HIGH_ORDER; |
2533 | } else if (pcp->flags & PCPF_PREV_FREE_HIGH_ORDER) { | |
2534 | pcp->flags &= ~PCPF_PREV_FREE_HIGH_ORDER; | |
2535 | } | |
6ccdcb6d HY |
2536 | if (pcp->free_count < (batch << CONFIG_PCP_BATCH_SCALE_MAX)) |
2537 | pcp->free_count += (1 << order); | |
51a755c5 | 2538 | high = nr_pcp_high(pcp, zone, batch, free_high); |
3b12e7e9 | 2539 | if (pcp->count >= high) { |
51a755c5 HY |
2540 | free_pcppages_bulk(zone, nr_pcp_free(pcp, batch, high, free_high), |
2541 | pcp, pindex); | |
57c0419c HY |
2542 | if (test_bit(ZONE_BELOW_HIGH, &zone->flags) && |
2543 | zone_watermark_ok(zone, 0, high_wmark_pages(zone), | |
2544 | ZONE_MOVABLE, 0)) | |
2545 | clear_bit(ZONE_BELOW_HIGH, &zone->flags); | |
3b12e7e9 | 2546 | } |
9cca35d4 | 2547 | } |
5f8dcc21 | 2548 | |
9cca35d4 | 2549 | /* |
44042b44 | 2550 | * Free a pcp page |
9cca35d4 | 2551 | */ |
44042b44 | 2552 | void free_unref_page(struct page *page, unsigned int order) |
9cca35d4 | 2553 | { |
4b23a68f MG |
2554 | unsigned long __maybe_unused UP_flags; |
2555 | struct per_cpu_pages *pcp; | |
2556 | struct zone *zone; | |
9cca35d4 | 2557 | unsigned long pfn = page_to_pfn(page); |
55612e80 | 2558 | int migratetype; |
9cca35d4 | 2559 | |
5b8d7591 MWO |
2560 | if (!pcp_allowed_order(order)) { |
2561 | __free_pages_ok(page, order, FPI_NONE); | |
2562 | return; | |
2563 | } | |
2564 | ||
17edeb5d | 2565 | if (!free_pages_prepare(page, order)) |
9cca35d4 | 2566 | return; |
da456f14 | 2567 | |
5f8dcc21 MG |
2568 | /* |
2569 | * We only track unmovable, reclaimable and movable on pcp lists. | |
df1acc85 | 2570 | * Place ISOLATE pages on the isolated list because they are being |
7b086755 JW |
2571 | * offlined but treat HIGHATOMIC and CMA as movable pages so we can |
2572 | * get those areas back if necessary. Otherwise, we may have to free | |
5f8dcc21 MG |
2573 | * excessively into the page allocator |
2574 | */ | |
55612e80 | 2575 | migratetype = get_pfnblock_migratetype(page, pfn); |
df1acc85 | 2576 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) { |
194159fb | 2577 | if (unlikely(is_migrate_isolate(migratetype))) { |
55612e80 | 2578 | free_one_page(page_zone(page), page, pfn, order, FPI_NONE); |
9cca35d4 | 2579 | return; |
5f8dcc21 | 2580 | } |
55612e80 | 2581 | migratetype = MIGRATE_MOVABLE; |
5f8dcc21 MG |
2582 | } |
2583 | ||
4b23a68f MG |
2584 | zone = page_zone(page); |
2585 | pcp_trylock_prepare(UP_flags); | |
57490774 | 2586 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 2587 | if (pcp) { |
55612e80 | 2588 | free_unref_page_commit(zone, pcp, page, migratetype, order); |
57490774 | 2589 | pcp_spin_unlock(pcp); |
4b23a68f | 2590 | } else { |
55612e80 | 2591 | free_one_page(zone, page, pfn, order, FPI_NONE); |
4b23a68f MG |
2592 | } |
2593 | pcp_trylock_finish(UP_flags); | |
1da177e4 LT |
2594 | } |
2595 | ||
cc59850e | 2596 | /* |
31b2ff82 | 2597 | * Free a batch of folios |
cc59850e | 2598 | */ |
90491d87 | 2599 | void free_unref_folios(struct folio_batch *folios) |
cc59850e | 2600 | { |
57490774 | 2601 | unsigned long __maybe_unused UP_flags; |
4b23a68f MG |
2602 | struct per_cpu_pages *pcp = NULL; |
2603 | struct zone *locked_zone = NULL; | |
9cbe97ba | 2604 | int i, j; |
9cca35d4 | 2605 | |
90491d87 MWO |
2606 | /* Prepare folios for freeing */ |
2607 | for (i = 0, j = 0; i < folios->nr; i++) { | |
2608 | struct folio *folio = folios->folios[i]; | |
7c76d922 | 2609 | unsigned long pfn = folio_pfn(folio); |
31b2ff82 | 2610 | unsigned int order = folio_order(folio); |
9cca35d4 | 2611 | |
31b2ff82 MWO |
2612 | if (order > 0 && folio_test_large_rmappable(folio)) |
2613 | folio_undo_large_rmappable(folio); | |
17edeb5d | 2614 | if (!free_pages_prepare(&folio->page, order)) |
053cfda1 | 2615 | continue; |
df1acc85 | 2616 | /* |
9cbe97ba JW |
2617 | * Free orders not handled on the PCP directly to the |
2618 | * allocator. | |
df1acc85 | 2619 | */ |
9cbe97ba | 2620 | if (!pcp_allowed_order(order)) { |
55612e80 JW |
2621 | free_one_page(folio_zone(folio), &folio->page, |
2622 | pfn, order, FPI_NONE); | |
47aef601 | 2623 | continue; |
df1acc85 | 2624 | } |
31b2ff82 | 2625 | folio->private = (void *)(unsigned long)order; |
90491d87 MWO |
2626 | if (j != i) |
2627 | folios->folios[j] = folio; | |
2628 | j++; | |
9cca35d4 | 2629 | } |
90491d87 | 2630 | folios->nr = j; |
cc59850e | 2631 | |
90491d87 MWO |
2632 | for (i = 0; i < folios->nr; i++) { |
2633 | struct folio *folio = folios->folios[i]; | |
7c76d922 | 2634 | struct zone *zone = folio_zone(folio); |
17edeb5d | 2635 | unsigned long pfn = folio_pfn(folio); |
31b2ff82 | 2636 | unsigned int order = (unsigned long)folio->private; |
9cbe97ba | 2637 | int migratetype; |
4b23a68f | 2638 | |
31b2ff82 | 2639 | folio->private = NULL; |
17edeb5d | 2640 | migratetype = get_pfnblock_migratetype(&folio->page, pfn); |
c3e58a70 | 2641 | |
90491d87 | 2642 | /* Different zone requires a different pcp lock */ |
9cbe97ba JW |
2643 | if (zone != locked_zone || |
2644 | is_migrate_isolate(migratetype)) { | |
57490774 MG |
2645 | if (pcp) { |
2646 | pcp_spin_unlock(pcp); | |
2647 | pcp_trylock_finish(UP_flags); | |
9cbe97ba JW |
2648 | locked_zone = NULL; |
2649 | pcp = NULL; | |
2650 | } | |
2651 | ||
2652 | /* | |
2653 | * Free isolated pages directly to the | |
2654 | * allocator, see comment in free_unref_page. | |
2655 | */ | |
2656 | if (is_migrate_isolate(migratetype)) { | |
2657 | free_one_page(zone, &folio->page, pfn, | |
55612e80 | 2658 | order, FPI_NONE); |
9cbe97ba | 2659 | continue; |
57490774 | 2660 | } |
01b44456 | 2661 | |
57490774 | 2662 | /* |
7c76d922 | 2663 | * trylock is necessary as folios may be getting freed |
57490774 MG |
2664 | * from IRQ or SoftIRQ context after an IO completion. |
2665 | */ | |
2666 | pcp_trylock_prepare(UP_flags); | |
2667 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); | |
2668 | if (unlikely(!pcp)) { | |
2669 | pcp_trylock_finish(UP_flags); | |
17edeb5d | 2670 | free_one_page(zone, &folio->page, pfn, |
55612e80 | 2671 | order, FPI_NONE); |
57490774 MG |
2672 | continue; |
2673 | } | |
4b23a68f | 2674 | locked_zone = zone; |
4b23a68f MG |
2675 | } |
2676 | ||
47aef601 DB |
2677 | /* |
2678 | * Non-isolated types over MIGRATE_PCPTYPES get added | |
2679 | * to the MIGRATE_MOVABLE pcp list. | |
2680 | */ | |
47aef601 DB |
2681 | if (unlikely(migratetype >= MIGRATE_PCPTYPES)) |
2682 | migratetype = MIGRATE_MOVABLE; | |
2683 | ||
7c76d922 | 2684 | trace_mm_page_free_batched(&folio->page); |
31b2ff82 MWO |
2685 | free_unref_page_commit(zone, pcp, &folio->page, migratetype, |
2686 | order); | |
cc59850e | 2687 | } |
4b23a68f | 2688 | |
57490774 MG |
2689 | if (pcp) { |
2690 | pcp_spin_unlock(pcp); | |
2691 | pcp_trylock_finish(UP_flags); | |
2692 | } | |
90491d87 | 2693 | folio_batch_reinit(folios); |
cc59850e KK |
2694 | } |
2695 | ||
8dfcc9ba NP |
2696 | /* |
2697 | * split_page takes a non-compound higher-order page, and splits it into | |
2698 | * n (1<<order) sub-pages: page[0..n] | |
2699 | * Each sub-page must be freed individually. | |
2700 | * | |
2701 | * Note: this is probably too low level an operation for use in drivers. | |
2702 | * Please consult with lkml before using this in your driver. | |
2703 | */ | |
2704 | void split_page(struct page *page, unsigned int order) | |
2705 | { | |
2706 | int i; | |
2707 | ||
309381fe SL |
2708 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2709 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 2710 | |
a9627bc5 | 2711 | for (i = 1; i < (1 << order); i++) |
7835e98b | 2712 | set_page_refcounted(page + i); |
46d44d09 | 2713 | split_page_owner(page, order, 0); |
be25d1d4 | 2714 | pgalloc_tag_split(page, 1 << order); |
b8791381 | 2715 | split_page_memcg(page, order, 0); |
8dfcc9ba | 2716 | } |
5853ff23 | 2717 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2718 | |
3c605096 | 2719 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2720 | { |
9a157dd8 KW |
2721 | struct zone *zone = page_zone(page); |
2722 | int mt = get_pageblock_migratetype(page); | |
748446bb | 2723 | |
194159fb | 2724 | if (!is_migrate_isolate(mt)) { |
9a157dd8 | 2725 | unsigned long watermark; |
8348faf9 VB |
2726 | /* |
2727 | * Obey watermarks as if the page was being allocated. We can | |
2728 | * emulate a high-order watermark check with a raised order-0 | |
2729 | * watermark, because we already know our high-order page | |
2730 | * exists. | |
2731 | */ | |
fd1444b2 | 2732 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 2733 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 | 2734 | return 0; |
2e30abd1 | 2735 | } |
748446bb | 2736 | |
e0932b6c | 2737 | del_page_from_free_list(page, zone, order, mt); |
2139cbe6 | 2738 | |
400bc7fd | 2739 | /* |
2740 | * Set the pageblock if the isolated page is at least half of a | |
2741 | * pageblock | |
2742 | */ | |
748446bb MG |
2743 | if (order >= pageblock_order - 1) { |
2744 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2745 | for (; page < endpage; page += pageblock_nr_pages) { |
2746 | int mt = get_pageblock_migratetype(page); | |
1dd214b8 ZY |
2747 | /* |
2748 | * Only change normal pageblocks (i.e., they can merge | |
2749 | * with others) | |
2750 | */ | |
f37c0f68 | 2751 | if (migratetype_is_mergeable(mt)) |
e0932b6c | 2752 | move_freepages_block(zone, page, mt, |
f37c0f68 | 2753 | MIGRATE_MOVABLE); |
47118af0 | 2754 | } |
748446bb MG |
2755 | } |
2756 | ||
8fb74b9f | 2757 | return 1UL << order; |
1fb3f8ca MG |
2758 | } |
2759 | ||
624f58d8 AD |
2760 | /** |
2761 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
2762 | * @page: Page that was isolated | |
2763 | * @order: Order of the isolated page | |
e6a0a7ad | 2764 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
2765 | * |
2766 | * This function is meant to return a page pulled from the free lists via | |
2767 | * __isolate_free_page back to the free lists they were pulled from. | |
2768 | */ | |
2769 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
2770 | { | |
2771 | struct zone *zone = page_zone(page); | |
2772 | ||
2773 | /* zone lock should be held when this function is called */ | |
2774 | lockdep_assert_held(&zone->lock); | |
2775 | ||
2776 | /* Return isolated page to tail of freelist. */ | |
f04a5d5d | 2777 | __free_one_page(page, page_to_pfn(page), zone, order, mt, |
47b6a24a | 2778 | FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL); |
624f58d8 AD |
2779 | } |
2780 | ||
060e7417 MG |
2781 | /* |
2782 | * Update NUMA hit/miss statistics | |
060e7417 | 2783 | */ |
3e23060b MG |
2784 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, |
2785 | long nr_account) | |
060e7417 MG |
2786 | { |
2787 | #ifdef CONFIG_NUMA | |
3a321d2a | 2788 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 2789 | |
4518085e KW |
2790 | /* skip numa counters update if numa stats is disabled */ |
2791 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2792 | return; | |
2793 | ||
c1093b74 | 2794 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 2795 | local_stat = NUMA_OTHER; |
060e7417 | 2796 | |
c1093b74 | 2797 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3e23060b | 2798 | __count_numa_events(z, NUMA_HIT, nr_account); |
2df26639 | 2799 | else { |
3e23060b MG |
2800 | __count_numa_events(z, NUMA_MISS, nr_account); |
2801 | __count_numa_events(preferred_zone, NUMA_FOREIGN, nr_account); | |
060e7417 | 2802 | } |
3e23060b | 2803 | __count_numa_events(z, local_stat, nr_account); |
060e7417 MG |
2804 | #endif |
2805 | } | |
2806 | ||
589d9973 MG |
2807 | static __always_inline |
2808 | struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, | |
2809 | unsigned int order, unsigned int alloc_flags, | |
2810 | int migratetype) | |
2811 | { | |
2812 | struct page *page; | |
2813 | unsigned long flags; | |
2814 | ||
2815 | do { | |
2816 | page = NULL; | |
2817 | spin_lock_irqsave(&zone->lock, flags); | |
eb2e2b42 | 2818 | if (alloc_flags & ALLOC_HIGHATOMIC) |
589d9973 MG |
2819 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
2820 | if (!page) { | |
2821 | page = __rmqueue(zone, order, migratetype, alloc_flags); | |
eb2e2b42 MG |
2822 | |
2823 | /* | |
2824 | * If the allocation fails, allow OOM handling access | |
2825 | * to HIGHATOMIC reserves as failing now is worse than | |
2826 | * failing a high-order atomic allocation in the | |
2827 | * future. | |
2828 | */ | |
2829 | if (!page && (alloc_flags & ALLOC_OOM)) | |
2830 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2831 | ||
589d9973 MG |
2832 | if (!page) { |
2833 | spin_unlock_irqrestore(&zone->lock, flags); | |
2834 | return NULL; | |
2835 | } | |
2836 | } | |
589d9973 MG |
2837 | spin_unlock_irqrestore(&zone->lock, flags); |
2838 | } while (check_new_pages(page, order)); | |
2839 | ||
2840 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); | |
2841 | zone_statistics(preferred_zone, zone, 1); | |
2842 | ||
2843 | return page; | |
2844 | } | |
2845 | ||
51a755c5 | 2846 | static int nr_pcp_alloc(struct per_cpu_pages *pcp, struct zone *zone, int order) |
c0a24239 | 2847 | { |
51a755c5 HY |
2848 | int high, base_batch, batch, max_nr_alloc; |
2849 | int high_max, high_min; | |
c0a24239 | 2850 | |
51a755c5 HY |
2851 | base_batch = READ_ONCE(pcp->batch); |
2852 | high_min = READ_ONCE(pcp->high_min); | |
2853 | high_max = READ_ONCE(pcp->high_max); | |
2854 | high = pcp->high = clamp(pcp->high, high_min, high_max); | |
c0a24239 HY |
2855 | |
2856 | /* Check for PCP disabled or boot pageset */ | |
51a755c5 | 2857 | if (unlikely(high < base_batch)) |
c0a24239 HY |
2858 | return 1; |
2859 | ||
51a755c5 HY |
2860 | if (order) |
2861 | batch = base_batch; | |
2862 | else | |
2863 | batch = (base_batch << pcp->alloc_factor); | |
2864 | ||
c0a24239 | 2865 | /* |
51a755c5 HY |
2866 | * If we had larger pcp->high, we could avoid to allocate from |
2867 | * zone. | |
c0a24239 | 2868 | */ |
57c0419c | 2869 | if (high_min != high_max && !test_bit(ZONE_BELOW_HIGH, &zone->flags)) |
51a755c5 HY |
2870 | high = pcp->high = min(high + batch, high_max); |
2871 | ||
c0a24239 | 2872 | if (!order) { |
51a755c5 HY |
2873 | max_nr_alloc = max(high - pcp->count - base_batch, base_batch); |
2874 | /* | |
2875 | * Double the number of pages allocated each time there is | |
2876 | * subsequent allocation of order-0 pages without any freeing. | |
2877 | */ | |
c0a24239 HY |
2878 | if (batch <= max_nr_alloc && |
2879 | pcp->alloc_factor < CONFIG_PCP_BATCH_SCALE_MAX) | |
2880 | pcp->alloc_factor++; | |
2881 | batch = min(batch, max_nr_alloc); | |
2882 | } | |
2883 | ||
2884 | /* | |
2885 | * Scale batch relative to order if batch implies free pages | |
2886 | * can be stored on the PCP. Batch can be 1 for small zones or | |
2887 | * for boot pagesets which should never store free pages as | |
2888 | * the pages may belong to arbitrary zones. | |
2889 | */ | |
2890 | if (batch > 1) | |
2891 | batch = max(batch >> order, 2); | |
2892 | ||
2893 | return batch; | |
2894 | } | |
2895 | ||
066b2393 | 2896 | /* Remove page from the per-cpu list, caller must protect the list */ |
3b822017 | 2897 | static inline |
44042b44 MG |
2898 | struct page *__rmqueue_pcplist(struct zone *zone, unsigned int order, |
2899 | int migratetype, | |
6bb15450 | 2900 | unsigned int alloc_flags, |
453f85d4 | 2901 | struct per_cpu_pages *pcp, |
066b2393 MG |
2902 | struct list_head *list) |
2903 | { | |
2904 | struct page *page; | |
2905 | ||
2906 | do { | |
2907 | if (list_empty(list)) { | |
51a755c5 | 2908 | int batch = nr_pcp_alloc(pcp, zone, order); |
44042b44 MG |
2909 | int alloced; |
2910 | ||
44042b44 MG |
2911 | alloced = rmqueue_bulk(zone, order, |
2912 | batch, list, | |
6bb15450 | 2913 | migratetype, alloc_flags); |
44042b44 MG |
2914 | |
2915 | pcp->count += alloced << order; | |
066b2393 MG |
2916 | if (unlikely(list_empty(list))) |
2917 | return NULL; | |
2918 | } | |
2919 | ||
bf75f200 MG |
2920 | page = list_first_entry(list, struct page, pcp_list); |
2921 | list_del(&page->pcp_list); | |
44042b44 | 2922 | pcp->count -= 1 << order; |
700d2e9a | 2923 | } while (check_new_pages(page, order)); |
066b2393 MG |
2924 | |
2925 | return page; | |
2926 | } | |
2927 | ||
2928 | /* Lock and remove page from the per-cpu list */ | |
2929 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
44042b44 | 2930 | struct zone *zone, unsigned int order, |
663d0cfd | 2931 | int migratetype, unsigned int alloc_flags) |
066b2393 MG |
2932 | { |
2933 | struct per_cpu_pages *pcp; | |
2934 | struct list_head *list; | |
066b2393 | 2935 | struct page *page; |
4b23a68f | 2936 | unsigned long __maybe_unused UP_flags; |
066b2393 | 2937 | |
57490774 | 2938 | /* spin_trylock may fail due to a parallel drain or IRQ reentrancy. */ |
4b23a68f | 2939 | pcp_trylock_prepare(UP_flags); |
57490774 | 2940 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 2941 | if (!pcp) { |
4b23a68f | 2942 | pcp_trylock_finish(UP_flags); |
4b23a68f MG |
2943 | return NULL; |
2944 | } | |
3b12e7e9 MG |
2945 | |
2946 | /* | |
2947 | * On allocation, reduce the number of pages that are batch freed. | |
2948 | * See nr_pcp_free() where free_factor is increased for subsequent | |
2949 | * frees. | |
2950 | */ | |
6ccdcb6d | 2951 | pcp->free_count >>= 1; |
44042b44 MG |
2952 | list = &pcp->lists[order_to_pindex(migratetype, order)]; |
2953 | page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list); | |
57490774 | 2954 | pcp_spin_unlock(pcp); |
4b23a68f | 2955 | pcp_trylock_finish(UP_flags); |
066b2393 | 2956 | if (page) { |
15cd9004 | 2957 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
3e23060b | 2958 | zone_statistics(preferred_zone, zone, 1); |
066b2393 | 2959 | } |
066b2393 MG |
2960 | return page; |
2961 | } | |
2962 | ||
1da177e4 | 2963 | /* |
a57ae9ef RX |
2964 | * Allocate a page from the given zone. |
2965 | * Use pcplists for THP or "cheap" high-order allocations. | |
1da177e4 | 2966 | */ |
b073d7f8 AP |
2967 | |
2968 | /* | |
2969 | * Do not instrument rmqueue() with KMSAN. This function may call | |
2970 | * __msan_poison_alloca() through a call to set_pfnblock_flags_mask(). | |
2971 | * If __msan_poison_alloca() attempts to allocate pages for the stack depot, it | |
2972 | * may call rmqueue() again, which will result in a deadlock. | |
1da177e4 | 2973 | */ |
b073d7f8 | 2974 | __no_sanitize_memory |
0a15c3e9 | 2975 | static inline |
066b2393 | 2976 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 2977 | struct zone *zone, unsigned int order, |
c603844b MG |
2978 | gfp_t gfp_flags, unsigned int alloc_flags, |
2979 | int migratetype) | |
1da177e4 | 2980 | { |
689bcebf | 2981 | struct page *page; |
1da177e4 | 2982 | |
589d9973 MG |
2983 | /* |
2984 | * We most definitely don't want callers attempting to | |
2985 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2986 | */ | |
2987 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
2988 | ||
44042b44 | 2989 | if (likely(pcp_allowed_order(order))) { |
f945116e JW |
2990 | page = rmqueue_pcplist(preferred_zone, zone, order, |
2991 | migratetype, alloc_flags); | |
2992 | if (likely(page)) | |
2993 | goto out; | |
066b2393 | 2994 | } |
83b9355b | 2995 | |
589d9973 MG |
2996 | page = rmqueue_buddy(preferred_zone, zone, order, alloc_flags, |
2997 | migratetype); | |
1da177e4 | 2998 | |
066b2393 | 2999 | out: |
73444bc4 | 3000 | /* Separate test+clear to avoid unnecessary atomics */ |
3b11edf1 TH |
3001 | if ((alloc_flags & ALLOC_KSWAPD) && |
3002 | unlikely(test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags))) { | |
73444bc4 MG |
3003 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
3004 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3005 | } | |
3006 | ||
066b2393 | 3007 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 LT |
3008 | return page; |
3009 | } | |
3010 | ||
54aa3866 | 3011 | noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
af3b8544 BP |
3012 | { |
3013 | return __should_fail_alloc_page(gfp_mask, order); | |
3014 | } | |
3015 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3016 | ||
f27ce0e1 JK |
3017 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3018 | unsigned int order, unsigned int alloc_flags) | |
3019 | { | |
f27ce0e1 JK |
3020 | long unusable_free = (1 << order) - 1; |
3021 | ||
3022 | /* | |
ab350885 MG |
3023 | * If the caller does not have rights to reserves below the min |
3024 | * watermark then subtract the high-atomic reserves. This will | |
3025 | * over-estimate the size of the atomic reserve but it avoids a search. | |
f27ce0e1 | 3026 | */ |
ab350885 | 3027 | if (likely(!(alloc_flags & ALLOC_RESERVES))) |
f27ce0e1 JK |
3028 | unusable_free += z->nr_reserved_highatomic; |
3029 | ||
3030 | #ifdef CONFIG_CMA | |
3031 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3032 | if (!(alloc_flags & ALLOC_CMA)) | |
3033 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3034 | #endif | |
dcdfdd40 KS |
3035 | #ifdef CONFIG_UNACCEPTED_MEMORY |
3036 | unusable_free += zone_page_state(z, NR_UNACCEPTED); | |
3037 | #endif | |
f27ce0e1 JK |
3038 | |
3039 | return unusable_free; | |
3040 | } | |
3041 | ||
1da177e4 | 3042 | /* |
97a16fc8 MG |
3043 | * Return true if free base pages are above 'mark'. For high-order checks it |
3044 | * will return true of the order-0 watermark is reached and there is at least | |
3045 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3046 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3047 | */ |
86a294a8 | 3048 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3049 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3050 | long free_pages) |
1da177e4 | 3051 | { |
d23ad423 | 3052 | long min = mark; |
1da177e4 LT |
3053 | int o; |
3054 | ||
0aaa29a5 | 3055 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 3056 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 3057 | |
ab350885 MG |
3058 | if (unlikely(alloc_flags & ALLOC_RESERVES)) { |
3059 | /* | |
3060 | * __GFP_HIGH allows access to 50% of the min reserve as well | |
3061 | * as OOM. | |
3062 | */ | |
1ebbb218 | 3063 | if (alloc_flags & ALLOC_MIN_RESERVE) { |
ab350885 | 3064 | min -= min / 2; |
0aaa29a5 | 3065 | |
1ebbb218 MG |
3066 | /* |
3067 | * Non-blocking allocations (e.g. GFP_ATOMIC) can | |
3068 | * access more reserves than just __GFP_HIGH. Other | |
3069 | * non-blocking allocations requests such as GFP_NOWAIT | |
3070 | * or (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) do not get | |
3071 | * access to the min reserve. | |
3072 | */ | |
3073 | if (alloc_flags & ALLOC_NON_BLOCK) | |
3074 | min -= min / 4; | |
3075 | } | |
0aaa29a5 | 3076 | |
cd04ae1e | 3077 | /* |
ab350885 | 3078 | * OOM victims can try even harder than the normal reserve |
cd04ae1e MH |
3079 | * users on the grounds that it's definitely going to be in |
3080 | * the exit path shortly and free memory. Any allocation it | |
3081 | * makes during the free path will be small and short-lived. | |
3082 | */ | |
3083 | if (alloc_flags & ALLOC_OOM) | |
3084 | min -= min / 2; | |
cd04ae1e MH |
3085 | } |
3086 | ||
97a16fc8 MG |
3087 | /* |
3088 | * Check watermarks for an order-0 allocation request. If these | |
3089 | * are not met, then a high-order request also cannot go ahead | |
3090 | * even if a suitable page happened to be free. | |
3091 | */ | |
97a225e6 | 3092 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 3093 | return false; |
1da177e4 | 3094 | |
97a16fc8 MG |
3095 | /* If this is an order-0 request then the watermark is fine */ |
3096 | if (!order) | |
3097 | return true; | |
3098 | ||
3099 | /* For a high-order request, check at least one suitable page is free */ | |
fd377218 | 3100 | for (o = order; o < NR_PAGE_ORDERS; o++) { |
97a16fc8 MG |
3101 | struct free_area *area = &z->free_area[o]; |
3102 | int mt; | |
3103 | ||
3104 | if (!area->nr_free) | |
3105 | continue; | |
3106 | ||
97a16fc8 | 3107 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3108 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3109 | return true; |
3110 | } | |
3111 | ||
3112 | #ifdef CONFIG_CMA | |
d883c6cf | 3113 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3114 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3115 | return true; |
d883c6cf | 3116 | } |
97a16fc8 | 3117 | #endif |
eb2e2b42 MG |
3118 | if ((alloc_flags & (ALLOC_HIGHATOMIC|ALLOC_OOM)) && |
3119 | !free_area_empty(area, MIGRATE_HIGHATOMIC)) { | |
b050e376 | 3120 | return true; |
eb2e2b42 | 3121 | } |
1da177e4 | 3122 | } |
97a16fc8 | 3123 | return false; |
88f5acf8 MG |
3124 | } |
3125 | ||
7aeb09f9 | 3126 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3127 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 3128 | { |
97a225e6 | 3129 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
3130 | zone_page_state(z, NR_FREE_PAGES)); |
3131 | } | |
3132 | ||
48ee5f36 | 3133 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 3134 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 3135 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 3136 | { |
f27ce0e1 | 3137 | long free_pages; |
d883c6cf | 3138 | |
f27ce0e1 | 3139 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
3140 | |
3141 | /* | |
3142 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 3143 | * need to be calculated. |
48ee5f36 | 3144 | */ |
f27ce0e1 | 3145 | if (!order) { |
9282012f JK |
3146 | long usable_free; |
3147 | long reserved; | |
f27ce0e1 | 3148 | |
9282012f JK |
3149 | usable_free = free_pages; |
3150 | reserved = __zone_watermark_unusable_free(z, 0, alloc_flags); | |
3151 | ||
3152 | /* reserved may over estimate high-atomic reserves. */ | |
3153 | usable_free -= min(usable_free, reserved); | |
3154 | if (usable_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
f27ce0e1 JK |
3155 | return true; |
3156 | } | |
48ee5f36 | 3157 | |
f80b08fc CTR |
3158 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
3159 | free_pages)) | |
3160 | return true; | |
2973d822 | 3161 | |
f80b08fc | 3162 | /* |
2973d822 | 3163 | * Ignore watermark boosting for __GFP_HIGH order-0 allocations |
f80b08fc CTR |
3164 | * when checking the min watermark. The min watermark is the |
3165 | * point where boosting is ignored so that kswapd is woken up | |
3166 | * when below the low watermark. | |
3167 | */ | |
2973d822 | 3168 | if (unlikely(!order && (alloc_flags & ALLOC_MIN_RESERVE) && z->watermark_boost |
f80b08fc CTR |
3169 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { |
3170 | mark = z->_watermark[WMARK_MIN]; | |
3171 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
3172 | alloc_flags, free_pages); | |
3173 | } | |
3174 | ||
3175 | return false; | |
48ee5f36 MG |
3176 | } |
3177 | ||
7aeb09f9 | 3178 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 3179 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
3180 | { |
3181 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3182 | ||
3183 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3184 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3185 | ||
97a225e6 | 3186 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 3187 | free_pages); |
1da177e4 LT |
3188 | } |
3189 | ||
9276b1bc | 3190 | #ifdef CONFIG_NUMA |
61bb6cd2 GU |
3191 | int __read_mostly node_reclaim_distance = RECLAIM_DISTANCE; |
3192 | ||
957f822a DR |
3193 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3194 | { | |
e02dc017 | 3195 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3196 | node_reclaim_distance; |
957f822a | 3197 | } |
9276b1bc | 3198 | #else /* CONFIG_NUMA */ |
957f822a DR |
3199 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3200 | { | |
3201 | return true; | |
3202 | } | |
9276b1bc PJ |
3203 | #endif /* CONFIG_NUMA */ |
3204 | ||
6bb15450 MG |
3205 | /* |
3206 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3207 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3208 | * premature use of a lower zone may cause lowmem pressure problems that | |
3209 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3210 | * probably too small. It only makes sense to spread allocations to avoid | |
3211 | * fragmentation between the Normal and DMA32 zones. | |
3212 | */ | |
3213 | static inline unsigned int | |
0a79cdad | 3214 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3215 | { |
736838e9 | 3216 | unsigned int alloc_flags; |
0a79cdad | 3217 | |
736838e9 MN |
3218 | /* |
3219 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
3220 | * to save a branch. | |
3221 | */ | |
3222 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
3223 | |
3224 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3225 | if (!zone) |
3226 | return alloc_flags; | |
3227 | ||
6bb15450 | 3228 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3229 | return alloc_flags; |
6bb15450 MG |
3230 | |
3231 | /* | |
3232 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3233 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3234 | * on UMA that if Normal is populated then so is DMA32. | |
3235 | */ | |
3236 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3237 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3238 | return alloc_flags; |
6bb15450 | 3239 | |
8118b82e | 3240 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3241 | #endif /* CONFIG_ZONE_DMA32 */ |
3242 | return alloc_flags; | |
6bb15450 | 3243 | } |
6bb15450 | 3244 | |
8e3560d9 PT |
3245 | /* Must be called after current_gfp_context() which can change gfp_mask */ |
3246 | static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask, | |
3247 | unsigned int alloc_flags) | |
8510e69c JK |
3248 | { |
3249 | #ifdef CONFIG_CMA | |
8e3560d9 | 3250 | if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
8510e69c | 3251 | alloc_flags |= ALLOC_CMA; |
8510e69c JK |
3252 | #endif |
3253 | return alloc_flags; | |
3254 | } | |
3255 | ||
7fb1d9fc | 3256 | /* |
0798e519 | 3257 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3258 | * a page. |
3259 | */ | |
3260 | static struct page * | |
a9263751 VB |
3261 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3262 | const struct alloc_context *ac) | |
753ee728 | 3263 | { |
6bb15450 | 3264 | struct zoneref *z; |
5117f45d | 3265 | struct zone *zone; |
8a87d695 WY |
3266 | struct pglist_data *last_pgdat = NULL; |
3267 | bool last_pgdat_dirty_ok = false; | |
6bb15450 | 3268 | bool no_fallback; |
3b8c0be4 | 3269 | |
6bb15450 | 3270 | retry: |
7fb1d9fc | 3271 | /* |
9276b1bc | 3272 | * Scan zonelist, looking for a zone with enough free. |
8e464522 | 3273 | * See also cpuset_node_allowed() comment in kernel/cgroup/cpuset.c. |
7fb1d9fc | 3274 | */ |
6bb15450 MG |
3275 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3276 | z = ac->preferred_zoneref; | |
30d8ec73 MN |
3277 | for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx, |
3278 | ac->nodemask) { | |
be06af00 | 3279 | struct page *page; |
e085dbc5 JW |
3280 | unsigned long mark; |
3281 | ||
664eedde MG |
3282 | if (cpusets_enabled() && |
3283 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3284 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3285 | continue; |
a756cf59 JW |
3286 | /* |
3287 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3288 | * want to get it from a node that is within its dirty |
3289 | * limit, such that no single node holds more than its | |
a756cf59 | 3290 | * proportional share of globally allowed dirty pages. |
281e3726 | 3291 | * The dirty limits take into account the node's |
a756cf59 JW |
3292 | * lowmem reserves and high watermark so that kswapd |
3293 | * should be able to balance it without having to | |
3294 | * write pages from its LRU list. | |
3295 | * | |
a756cf59 | 3296 | * XXX: For now, allow allocations to potentially |
281e3726 | 3297 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3298 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3299 | * which is important when on a NUMA setup the allowed |
281e3726 | 3300 | * nodes are together not big enough to reach the |
a756cf59 | 3301 | * global limit. The proper fix for these situations |
281e3726 | 3302 | * will require awareness of nodes in the |
a756cf59 JW |
3303 | * dirty-throttling and the flusher threads. |
3304 | */ | |
3b8c0be4 | 3305 | if (ac->spread_dirty_pages) { |
8a87d695 WY |
3306 | if (last_pgdat != zone->zone_pgdat) { |
3307 | last_pgdat = zone->zone_pgdat; | |
3308 | last_pgdat_dirty_ok = node_dirty_ok(zone->zone_pgdat); | |
3309 | } | |
3b8c0be4 | 3310 | |
8a87d695 | 3311 | if (!last_pgdat_dirty_ok) |
3b8c0be4 | 3312 | continue; |
3b8c0be4 | 3313 | } |
7fb1d9fc | 3314 | |
6bb15450 MG |
3315 | if (no_fallback && nr_online_nodes > 1 && |
3316 | zone != ac->preferred_zoneref->zone) { | |
3317 | int local_nid; | |
3318 | ||
3319 | /* | |
3320 | * If moving to a remote node, retry but allow | |
3321 | * fragmenting fallbacks. Locality is more important | |
3322 | * than fragmentation avoidance. | |
3323 | */ | |
3324 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3325 | if (zone_to_nid(zone) != local_nid) { | |
3326 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3327 | goto retry; | |
3328 | } | |
3329 | } | |
3330 | ||
57c0419c HY |
3331 | /* |
3332 | * Detect whether the number of free pages is below high | |
3333 | * watermark. If so, we will decrease pcp->high and free | |
3334 | * PCP pages in free path to reduce the possibility of | |
3335 | * premature page reclaiming. Detection is done here to | |
3336 | * avoid to do that in hotter free path. | |
3337 | */ | |
3338 | if (test_bit(ZONE_BELOW_HIGH, &zone->flags)) | |
3339 | goto check_alloc_wmark; | |
3340 | ||
3341 | mark = high_wmark_pages(zone); | |
3342 | if (zone_watermark_fast(zone, order, mark, | |
3343 | ac->highest_zoneidx, alloc_flags, | |
3344 | gfp_mask)) | |
3345 | goto try_this_zone; | |
3346 | else | |
3347 | set_bit(ZONE_BELOW_HIGH, &zone->flags); | |
3348 | ||
3349 | check_alloc_wmark: | |
a9214443 | 3350 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 3351 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
3352 | ac->highest_zoneidx, alloc_flags, |
3353 | gfp_mask)) { | |
fa5e084e MG |
3354 | int ret; |
3355 | ||
dcdfdd40 KS |
3356 | if (has_unaccepted_memory()) { |
3357 | if (try_to_accept_memory(zone, order)) | |
3358 | goto try_this_zone; | |
3359 | } | |
3360 | ||
c9e97a19 PT |
3361 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3362 | /* | |
3363 | * Watermark failed for this zone, but see if we can | |
3364 | * grow this zone if it contains deferred pages. | |
3365 | */ | |
076cf7ea | 3366 | if (deferred_pages_enabled()) { |
c9e97a19 PT |
3367 | if (_deferred_grow_zone(zone, order)) |
3368 | goto try_this_zone; | |
3369 | } | |
3370 | #endif | |
5dab2911 MG |
3371 | /* Checked here to keep the fast path fast */ |
3372 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3373 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3374 | goto try_this_zone; | |
3375 | ||
202e35db | 3376 | if (!node_reclaim_enabled() || |
c33d6c06 | 3377 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3378 | continue; |
3379 | ||
a5f5f91d | 3380 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3381 | switch (ret) { |
a5f5f91d | 3382 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3383 | /* did not scan */ |
cd38b115 | 3384 | continue; |
a5f5f91d | 3385 | case NODE_RECLAIM_FULL: |
fa5e084e | 3386 | /* scanned but unreclaimable */ |
cd38b115 | 3387 | continue; |
fa5e084e MG |
3388 | default: |
3389 | /* did we reclaim enough */ | |
fed2719e | 3390 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 3391 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
3392 | goto try_this_zone; |
3393 | ||
fed2719e | 3394 | continue; |
0798e519 | 3395 | } |
7fb1d9fc RS |
3396 | } |
3397 | ||
fa5e084e | 3398 | try_this_zone: |
066b2393 | 3399 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3400 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3401 | if (page) { |
479f854a | 3402 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3403 | |
3404 | /* | |
3405 | * If this is a high-order atomic allocation then check | |
3406 | * if the pageblock should be reserved for the future | |
3407 | */ | |
eb2e2b42 | 3408 | if (unlikely(alloc_flags & ALLOC_HIGHATOMIC)) |
368d983b | 3409 | reserve_highatomic_pageblock(page, zone); |
0aaa29a5 | 3410 | |
75379191 | 3411 | return page; |
c9e97a19 | 3412 | } else { |
dcdfdd40 KS |
3413 | if (has_unaccepted_memory()) { |
3414 | if (try_to_accept_memory(zone, order)) | |
3415 | goto try_this_zone; | |
3416 | } | |
3417 | ||
c9e97a19 PT |
3418 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3419 | /* Try again if zone has deferred pages */ | |
076cf7ea | 3420 | if (deferred_pages_enabled()) { |
c9e97a19 PT |
3421 | if (_deferred_grow_zone(zone, order)) |
3422 | goto try_this_zone; | |
3423 | } | |
3424 | #endif | |
75379191 | 3425 | } |
54a6eb5c | 3426 | } |
9276b1bc | 3427 | |
6bb15450 MG |
3428 | /* |
3429 | * It's possible on a UMA machine to get through all zones that are | |
3430 | * fragmented. If avoiding fragmentation, reset and try again. | |
3431 | */ | |
3432 | if (no_fallback) { | |
3433 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3434 | goto retry; | |
3435 | } | |
3436 | ||
4ffeaf35 | 3437 | return NULL; |
753ee728 MH |
3438 | } |
3439 | ||
9af744d7 | 3440 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3441 | { |
a238ab5b | 3442 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
3443 | |
3444 | /* | |
3445 | * This documents exceptions given to allocations in certain | |
3446 | * contexts that are allowed to allocate outside current's set | |
3447 | * of allowed nodes. | |
3448 | */ | |
3449 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3450 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3451 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3452 | filter &= ~SHOW_MEM_FILTER_NODES; | |
88dc6f20 | 3453 | if (!in_task() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3454 | filter &= ~SHOW_MEM_FILTER_NODES; |
3455 | ||
974f4367 | 3456 | __show_mem(filter, nodemask, gfp_zone(gfp_mask)); |
aa187507 MH |
3457 | } |
3458 | ||
a8e99259 | 3459 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3460 | { |
3461 | struct va_format vaf; | |
3462 | va_list args; | |
1be334e5 | 3463 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 3464 | |
c4dc63f0 BH |
3465 | if ((gfp_mask & __GFP_NOWARN) || |
3466 | !__ratelimit(&nopage_rs) || | |
3467 | ((gfp_mask & __GFP_DMA) && !has_managed_dma())) | |
aa187507 MH |
3468 | return; |
3469 | ||
7877cdcc MH |
3470 | va_start(args, fmt); |
3471 | vaf.fmt = fmt; | |
3472 | vaf.va = &args; | |
ef8444ea | 3473 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
3474 | current->comm, &vaf, gfp_mask, &gfp_mask, |
3475 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3476 | va_end(args); |
3ee9a4f0 | 3477 | |
a8e99259 | 3478 | cpuset_print_current_mems_allowed(); |
ef8444ea | 3479 | pr_cont("\n"); |
a238ab5b | 3480 | dump_stack(); |
685dbf6f | 3481 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3482 | } |
3483 | ||
6c18ba7a MH |
3484 | static inline struct page * |
3485 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3486 | unsigned int alloc_flags, | |
3487 | const struct alloc_context *ac) | |
3488 | { | |
3489 | struct page *page; | |
3490 | ||
3491 | page = get_page_from_freelist(gfp_mask, order, | |
3492 | alloc_flags|ALLOC_CPUSET, ac); | |
3493 | /* | |
3494 | * fallback to ignore cpuset restriction if our nodes | |
3495 | * are depleted | |
3496 | */ | |
3497 | if (!page) | |
3498 | page = get_page_from_freelist(gfp_mask, order, | |
3499 | alloc_flags, ac); | |
3500 | ||
3501 | return page; | |
3502 | } | |
3503 | ||
11e33f6a MG |
3504 | static inline struct page * |
3505 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3506 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3507 | { |
6e0fc46d DR |
3508 | struct oom_control oc = { |
3509 | .zonelist = ac->zonelist, | |
3510 | .nodemask = ac->nodemask, | |
2a966b77 | 3511 | .memcg = NULL, |
6e0fc46d DR |
3512 | .gfp_mask = gfp_mask, |
3513 | .order = order, | |
6e0fc46d | 3514 | }; |
11e33f6a MG |
3515 | struct page *page; |
3516 | ||
9879de73 JW |
3517 | *did_some_progress = 0; |
3518 | ||
9879de73 | 3519 | /* |
dc56401f JW |
3520 | * Acquire the oom lock. If that fails, somebody else is |
3521 | * making progress for us. | |
9879de73 | 3522 | */ |
dc56401f | 3523 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3524 | *did_some_progress = 1; |
11e33f6a | 3525 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3526 | return NULL; |
3527 | } | |
6b1de916 | 3528 | |
11e33f6a MG |
3529 | /* |
3530 | * Go through the zonelist yet one more time, keep very high watermark | |
3531 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3532 | * we're still under heavy pressure. But make sure that this reclaim |
3533 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3534 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3535 | */ |
e746bf73 TH |
3536 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3537 | ~__GFP_DIRECT_RECLAIM, order, | |
3538 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3539 | if (page) |
11e33f6a MG |
3540 | goto out; |
3541 | ||
06ad276a MH |
3542 | /* Coredumps can quickly deplete all memory reserves */ |
3543 | if (current->flags & PF_DUMPCORE) | |
3544 | goto out; | |
3545 | /* The OOM killer will not help higher order allocs */ | |
3546 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3547 | goto out; | |
dcda9b04 MH |
3548 | /* |
3549 | * We have already exhausted all our reclaim opportunities without any | |
3550 | * success so it is time to admit defeat. We will skip the OOM killer | |
3551 | * because it is very likely that the caller has a more reasonable | |
3552 | * fallback than shooting a random task. | |
cfb4a541 MN |
3553 | * |
3554 | * The OOM killer may not free memory on a specific node. | |
dcda9b04 | 3555 | */ |
cfb4a541 | 3556 | if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE)) |
dcda9b04 | 3557 | goto out; |
06ad276a | 3558 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 3559 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
3560 | goto out; |
3561 | if (pm_suspended_storage()) | |
3562 | goto out; | |
3563 | /* | |
3564 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3565 | * other request to make a forward progress. | |
3566 | * We are in an unfortunate situation where out_of_memory cannot | |
3567 | * do much for this context but let's try it to at least get | |
3568 | * access to memory reserved if the current task is killed (see | |
3569 | * out_of_memory). Once filesystems are ready to handle allocation | |
3570 | * failures more gracefully we should just bail out here. | |
3571 | */ | |
3572 | ||
3c2c6488 | 3573 | /* Exhausted what can be done so it's blame time */ |
3f913fc5 QZ |
3574 | if (out_of_memory(&oc) || |
3575 | WARN_ON_ONCE_GFP(gfp_mask & __GFP_NOFAIL, gfp_mask)) { | |
c32b3cbe | 3576 | *did_some_progress = 1; |
5020e285 | 3577 | |
6c18ba7a MH |
3578 | /* |
3579 | * Help non-failing allocations by giving them access to memory | |
3580 | * reserves | |
3581 | */ | |
3582 | if (gfp_mask & __GFP_NOFAIL) | |
3583 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3584 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3585 | } |
11e33f6a | 3586 | out: |
dc56401f | 3587 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3588 | return page; |
3589 | } | |
3590 | ||
33c2d214 | 3591 | /* |
baf2f90b | 3592 | * Maximum number of compaction retries with a progress before OOM |
33c2d214 MH |
3593 | * killer is consider as the only way to move forward. |
3594 | */ | |
3595 | #define MAX_COMPACT_RETRIES 16 | |
3596 | ||
56de7263 MG |
3597 | #ifdef CONFIG_COMPACTION |
3598 | /* Try memory compaction for high-order allocations before reclaim */ | |
3599 | static struct page * | |
3600 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3601 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3602 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3603 | { |
5e1f0f09 | 3604 | struct page *page = NULL; |
eb414681 | 3605 | unsigned long pflags; |
499118e9 | 3606 | unsigned int noreclaim_flag; |
53853e2d VB |
3607 | |
3608 | if (!order) | |
66199712 | 3609 | return NULL; |
66199712 | 3610 | |
eb414681 | 3611 | psi_memstall_enter(&pflags); |
5bf18281 | 3612 | delayacct_compact_start(); |
499118e9 | 3613 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 3614 | |
c5d01d0d | 3615 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 3616 | prio, &page); |
eb414681 | 3617 | |
499118e9 | 3618 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 3619 | psi_memstall_leave(&pflags); |
5bf18281 | 3620 | delayacct_compact_end(); |
56de7263 | 3621 | |
06dac2f4 CTR |
3622 | if (*compact_result == COMPACT_SKIPPED) |
3623 | return NULL; | |
98dd3b48 VB |
3624 | /* |
3625 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3626 | * count a compaction stall | |
3627 | */ | |
3628 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3629 | |
5e1f0f09 MG |
3630 | /* Prep a captured page if available */ |
3631 | if (page) | |
3632 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
3633 | ||
3634 | /* Try get a page from the freelist if available */ | |
3635 | if (!page) | |
3636 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 3637 | |
98dd3b48 VB |
3638 | if (page) { |
3639 | struct zone *zone = page_zone(page); | |
53853e2d | 3640 | |
98dd3b48 VB |
3641 | zone->compact_blockskip_flush = false; |
3642 | compaction_defer_reset(zone, order, true); | |
3643 | count_vm_event(COMPACTSUCCESS); | |
3644 | return page; | |
3645 | } | |
56de7263 | 3646 | |
98dd3b48 VB |
3647 | /* |
3648 | * It's bad if compaction run occurs and fails. The most likely reason | |
3649 | * is that pages exist, but not enough to satisfy watermarks. | |
3650 | */ | |
3651 | count_vm_event(COMPACTFAIL); | |
66199712 | 3652 | |
98dd3b48 | 3653 | cond_resched(); |
56de7263 MG |
3654 | |
3655 | return NULL; | |
3656 | } | |
33c2d214 | 3657 | |
3250845d VB |
3658 | static inline bool |
3659 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3660 | enum compact_result compact_result, | |
3661 | enum compact_priority *compact_priority, | |
d9436498 | 3662 | int *compaction_retries) |
3250845d VB |
3663 | { |
3664 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3665 | int min_priority; |
65190cff MH |
3666 | bool ret = false; |
3667 | int retries = *compaction_retries; | |
3668 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3669 | |
3670 | if (!order) | |
3671 | return false; | |
3672 | ||
691d9497 AT |
3673 | if (fatal_signal_pending(current)) |
3674 | return false; | |
3675 | ||
49433085 | 3676 | /* |
ecd8b292 JW |
3677 | * Compaction was skipped due to a lack of free order-0 |
3678 | * migration targets. Continue if reclaim can help. | |
49433085 | 3679 | */ |
ecd8b292 | 3680 | if (compact_result == COMPACT_SKIPPED) { |
49433085 VB |
3681 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); |
3682 | goto out; | |
3683 | } | |
3684 | ||
3250845d | 3685 | /* |
511a69b2 JW |
3686 | * Compaction managed to coalesce some page blocks, but the |
3687 | * allocation failed presumably due to a race. Retry some. | |
3250845d | 3688 | */ |
511a69b2 JW |
3689 | if (compact_result == COMPACT_SUCCESS) { |
3690 | /* | |
3691 | * !costly requests are much more important than | |
3692 | * __GFP_RETRY_MAYFAIL costly ones because they are de | |
3693 | * facto nofail and invoke OOM killer to move on while | |
3694 | * costly can fail and users are ready to cope with | |
3695 | * that. 1/4 retries is rather arbitrary but we would | |
3696 | * need much more detailed feedback from compaction to | |
3697 | * make a better decision. | |
3698 | */ | |
3699 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3700 | max_retries /= 4; | |
3250845d | 3701 | |
511a69b2 JW |
3702 | if (++(*compaction_retries) <= max_retries) { |
3703 | ret = true; | |
3704 | goto out; | |
3705 | } | |
65190cff | 3706 | } |
3250845d | 3707 | |
d9436498 | 3708 | /* |
511a69b2 | 3709 | * Compaction failed. Retry with increasing priority. |
d9436498 | 3710 | */ |
c2033b00 VB |
3711 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3712 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3713 | |
c2033b00 | 3714 | if (*compact_priority > min_priority) { |
d9436498 VB |
3715 | (*compact_priority)--; |
3716 | *compaction_retries = 0; | |
65190cff | 3717 | ret = true; |
d9436498 | 3718 | } |
65190cff MH |
3719 | out: |
3720 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
3721 | return ret; | |
3250845d | 3722 | } |
56de7263 MG |
3723 | #else |
3724 | static inline struct page * | |
3725 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3726 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3727 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3728 | { |
33c2d214 | 3729 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
3730 | return NULL; |
3731 | } | |
33c2d214 MH |
3732 | |
3733 | static inline bool | |
86a294a8 MH |
3734 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3735 | enum compact_result compact_result, | |
a5508cd8 | 3736 | enum compact_priority *compact_priority, |
d9436498 | 3737 | int *compaction_retries) |
33c2d214 | 3738 | { |
31e49bfd MH |
3739 | struct zone *zone; |
3740 | struct zoneref *z; | |
3741 | ||
3742 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
3743 | return false; | |
3744 | ||
3745 | /* | |
3746 | * There are setups with compaction disabled which would prefer to loop | |
3747 | * inside the allocator rather than hit the oom killer prematurely. | |
3748 | * Let's give them a good hope and keep retrying while the order-0 | |
3749 | * watermarks are OK. | |
3750 | */ | |
97a225e6 JK |
3751 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3752 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 3753 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 3754 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
3755 | return true; |
3756 | } | |
33c2d214 MH |
3757 | return false; |
3758 | } | |
3250845d | 3759 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 3760 | |
d92a8cfc | 3761 | #ifdef CONFIG_LOCKDEP |
93781325 | 3762 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
3763 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
3764 | ||
f920e413 | 3765 | static bool __need_reclaim(gfp_t gfp_mask) |
d92a8cfc | 3766 | { |
d92a8cfc PZ |
3767 | /* no reclaim without waiting on it */ |
3768 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
3769 | return false; | |
3770 | ||
3771 | /* this guy won't enter reclaim */ | |
2e517d68 | 3772 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
3773 | return false; |
3774 | ||
d92a8cfc PZ |
3775 | if (gfp_mask & __GFP_NOLOCKDEP) |
3776 | return false; | |
3777 | ||
3778 | return true; | |
3779 | } | |
3780 | ||
4f3eaf45 | 3781 | void __fs_reclaim_acquire(unsigned long ip) |
93781325 | 3782 | { |
4f3eaf45 | 3783 | lock_acquire_exclusive(&__fs_reclaim_map, 0, 0, NULL, ip); |
93781325 OS |
3784 | } |
3785 | ||
4f3eaf45 | 3786 | void __fs_reclaim_release(unsigned long ip) |
93781325 | 3787 | { |
4f3eaf45 | 3788 | lock_release(&__fs_reclaim_map, ip); |
93781325 OS |
3789 | } |
3790 | ||
d92a8cfc PZ |
3791 | void fs_reclaim_acquire(gfp_t gfp_mask) |
3792 | { | |
f920e413 DV |
3793 | gfp_mask = current_gfp_context(gfp_mask); |
3794 | ||
3795 | if (__need_reclaim(gfp_mask)) { | |
3796 | if (gfp_mask & __GFP_FS) | |
4f3eaf45 | 3797 | __fs_reclaim_acquire(_RET_IP_); |
f920e413 DV |
3798 | |
3799 | #ifdef CONFIG_MMU_NOTIFIER | |
3800 | lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); | |
3801 | lock_map_release(&__mmu_notifier_invalidate_range_start_map); | |
3802 | #endif | |
3803 | ||
3804 | } | |
d92a8cfc PZ |
3805 | } |
3806 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
3807 | ||
3808 | void fs_reclaim_release(gfp_t gfp_mask) | |
3809 | { | |
f920e413 DV |
3810 | gfp_mask = current_gfp_context(gfp_mask); |
3811 | ||
3812 | if (__need_reclaim(gfp_mask)) { | |
3813 | if (gfp_mask & __GFP_FS) | |
4f3eaf45 | 3814 | __fs_reclaim_release(_RET_IP_); |
f920e413 | 3815 | } |
d92a8cfc PZ |
3816 | } |
3817 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
3818 | #endif | |
3819 | ||
3d36424b MG |
3820 | /* |
3821 | * Zonelists may change due to hotplug during allocation. Detect when zonelists | |
3822 | * have been rebuilt so allocation retries. Reader side does not lock and | |
3823 | * retries the allocation if zonelist changes. Writer side is protected by the | |
3824 | * embedded spin_lock. | |
3825 | */ | |
3826 | static DEFINE_SEQLOCK(zonelist_update_seq); | |
3827 | ||
3828 | static unsigned int zonelist_iter_begin(void) | |
3829 | { | |
3830 | if (IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) | |
3831 | return read_seqbegin(&zonelist_update_seq); | |
3832 | ||
3833 | return 0; | |
3834 | } | |
3835 | ||
3836 | static unsigned int check_retry_zonelist(unsigned int seq) | |
3837 | { | |
3838 | if (IS_ENABLED(CONFIG_MEMORY_HOTREMOVE)) | |
3839 | return read_seqretry(&zonelist_update_seq, seq); | |
3840 | ||
3841 | return seq; | |
3842 | } | |
3843 | ||
bba90710 | 3844 | /* Perform direct synchronous page reclaim */ |
2187e17b | 3845 | static unsigned long |
a9263751 VB |
3846 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3847 | const struct alloc_context *ac) | |
11e33f6a | 3848 | { |
499118e9 | 3849 | unsigned int noreclaim_flag; |
fa7fc75f | 3850 | unsigned long progress; |
11e33f6a MG |
3851 | |
3852 | cond_resched(); | |
3853 | ||
3854 | /* We now go into synchronous reclaim */ | |
3855 | cpuset_memory_pressure_bump(); | |
d92a8cfc | 3856 | fs_reclaim_acquire(gfp_mask); |
93781325 | 3857 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 3858 | |
a9263751 VB |
3859 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3860 | ac->nodemask); | |
11e33f6a | 3861 | |
499118e9 | 3862 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 3863 | fs_reclaim_release(gfp_mask); |
11e33f6a MG |
3864 | |
3865 | cond_resched(); | |
3866 | ||
bba90710 MS |
3867 | return progress; |
3868 | } | |
3869 | ||
3870 | /* The really slow allocator path where we enter direct reclaim */ | |
3871 | static inline struct page * | |
3872 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3873 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3874 | unsigned long *did_some_progress) |
bba90710 MS |
3875 | { |
3876 | struct page *page = NULL; | |
fa7fc75f | 3877 | unsigned long pflags; |
bba90710 MS |
3878 | bool drained = false; |
3879 | ||
fa7fc75f | 3880 | psi_memstall_enter(&pflags); |
a9263751 | 3881 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce | 3882 | if (unlikely(!(*did_some_progress))) |
fa7fc75f | 3883 | goto out; |
11e33f6a | 3884 | |
9ee493ce | 3885 | retry: |
31a6c190 | 3886 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
3887 | |
3888 | /* | |
3889 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 | 3890 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
047b9967 | 3891 | * Shrink them and try again |
9ee493ce MG |
3892 | */ |
3893 | if (!page && !drained) { | |
29fac03b | 3894 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 3895 | drain_all_pages(NULL); |
9ee493ce MG |
3896 | drained = true; |
3897 | goto retry; | |
3898 | } | |
fa7fc75f SB |
3899 | out: |
3900 | psi_memstall_leave(&pflags); | |
9ee493ce | 3901 | |
11e33f6a MG |
3902 | return page; |
3903 | } | |
3904 | ||
5ecd9d40 DR |
3905 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
3906 | const struct alloc_context *ac) | |
3a025760 JW |
3907 | { |
3908 | struct zoneref *z; | |
3909 | struct zone *zone; | |
e1a55637 | 3910 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 3911 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 3912 | |
97a225e6 | 3913 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 3914 | ac->nodemask) { |
bc53008e WY |
3915 | if (!managed_zone(zone)) |
3916 | continue; | |
d137a7cb | 3917 | if (last_pgdat != zone->zone_pgdat) { |
97a225e6 | 3918 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
d137a7cb CW |
3919 | last_pgdat = zone->zone_pgdat; |
3920 | } | |
e1a55637 | 3921 | } |
3a025760 JW |
3922 | } |
3923 | ||
c603844b | 3924 | static inline unsigned int |
eb2e2b42 | 3925 | gfp_to_alloc_flags(gfp_t gfp_mask, unsigned int order) |
341ce06f | 3926 | { |
c603844b | 3927 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3928 | |
736838e9 | 3929 | /* |
524c4807 | 3930 | * __GFP_HIGH is assumed to be the same as ALLOC_MIN_RESERVE |
736838e9 MN |
3931 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD |
3932 | * to save two branches. | |
3933 | */ | |
524c4807 | 3934 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_MIN_RESERVE); |
736838e9 | 3935 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 3936 | |
341ce06f PZ |
3937 | /* |
3938 | * The caller may dip into page reserves a bit more if the caller | |
3939 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3940 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1ebbb218 | 3941 | * set both ALLOC_NON_BLOCK and ALLOC_MIN_RESERVE(__GFP_HIGH). |
341ce06f | 3942 | */ |
736838e9 MN |
3943 | alloc_flags |= (__force int) |
3944 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 3945 | |
1ebbb218 | 3946 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) { |
5c3240d9 | 3947 | /* |
b104a35d DR |
3948 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3949 | * if it can't schedule. | |
5c3240d9 | 3950 | */ |
eb2e2b42 | 3951 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { |
1ebbb218 | 3952 | alloc_flags |= ALLOC_NON_BLOCK; |
eb2e2b42 MG |
3953 | |
3954 | if (order > 0) | |
3955 | alloc_flags |= ALLOC_HIGHATOMIC; | |
3956 | } | |
3957 | ||
523b9458 | 3958 | /* |
1ebbb218 MG |
3959 | * Ignore cpuset mems for non-blocking __GFP_HIGH (probably |
3960 | * GFP_ATOMIC) rather than fail, see the comment for | |
8e464522 | 3961 | * cpuset_node_allowed(). |
523b9458 | 3962 | */ |
1ebbb218 MG |
3963 | if (alloc_flags & ALLOC_MIN_RESERVE) |
3964 | alloc_flags &= ~ALLOC_CPUSET; | |
88dc6f20 | 3965 | } else if (unlikely(rt_task(current)) && in_task()) |
c988dcbe | 3966 | alloc_flags |= ALLOC_MIN_RESERVE; |
341ce06f | 3967 | |
8e3560d9 | 3968 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags); |
8510e69c | 3969 | |
341ce06f PZ |
3970 | return alloc_flags; |
3971 | } | |
3972 | ||
cd04ae1e | 3973 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 3974 | { |
cd04ae1e MH |
3975 | if (!tsk_is_oom_victim(tsk)) |
3976 | return false; | |
3977 | ||
3978 | /* | |
3979 | * !MMU doesn't have oom reaper so give access to memory reserves | |
3980 | * only to the thread with TIF_MEMDIE set | |
3981 | */ | |
3982 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
3983 | return false; |
3984 | ||
cd04ae1e MH |
3985 | return true; |
3986 | } | |
3987 | ||
3988 | /* | |
3989 | * Distinguish requests which really need access to full memory | |
3990 | * reserves from oom victims which can live with a portion of it | |
3991 | */ | |
3992 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
3993 | { | |
3994 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
3995 | return 0; | |
31a6c190 | 3996 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 3997 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 3998 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
3999 | return ALLOC_NO_WATERMARKS; |
4000 | if (!in_interrupt()) { | |
4001 | if (current->flags & PF_MEMALLOC) | |
4002 | return ALLOC_NO_WATERMARKS; | |
4003 | else if (oom_reserves_allowed(current)) | |
4004 | return ALLOC_OOM; | |
4005 | } | |
31a6c190 | 4006 | |
cd04ae1e MH |
4007 | return 0; |
4008 | } | |
4009 | ||
4010 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4011 | { | |
4012 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4013 | } |
4014 | ||
0a0337e0 MH |
4015 | /* |
4016 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4017 | * for the given allocation request. | |
491d79ae JW |
4018 | * |
4019 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4020 | * without success, or when we couldn't even meet the watermark if we | |
4021 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4022 | * |
4023 | * Returns true if a retry is viable or false to enter the oom path. | |
4024 | */ | |
4025 | static inline bool | |
4026 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4027 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4028 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4029 | { |
4030 | struct zone *zone; | |
4031 | struct zoneref *z; | |
15f570bf | 4032 | bool ret = false; |
0a0337e0 | 4033 | |
423b452e VB |
4034 | /* |
4035 | * Costly allocations might have made a progress but this doesn't mean | |
4036 | * their order will become available due to high fragmentation so | |
4037 | * always increment the no progress counter for them | |
4038 | */ | |
4039 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4040 | *no_progress_loops = 0; | |
4041 | else | |
4042 | (*no_progress_loops)++; | |
4043 | ||
ac3f3b0a CTK |
4044 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) |
4045 | goto out; | |
4046 | ||
0a0337e0 | 4047 | |
bca67592 MG |
4048 | /* |
4049 | * Keep reclaiming pages while there is a chance this will lead | |
4050 | * somewhere. If none of the target zones can satisfy our allocation | |
4051 | * request even if all reclaimable pages are considered then we are | |
4052 | * screwed and have to go OOM. | |
0a0337e0 | 4053 | */ |
97a225e6 JK |
4054 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4055 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4056 | unsigned long available; |
ede37713 | 4057 | unsigned long reclaimable; |
d379f01d MH |
4058 | unsigned long min_wmark = min_wmark_pages(zone); |
4059 | bool wmark; | |
0a0337e0 | 4060 | |
5a1c84b4 | 4061 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4062 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4063 | |
4064 | /* | |
491d79ae JW |
4065 | * Would the allocation succeed if we reclaimed all |
4066 | * reclaimable pages? | |
0a0337e0 | 4067 | */ |
d379f01d | 4068 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 4069 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
4070 | trace_reclaim_retry_zone(z, order, reclaimable, |
4071 | available, min_wmark, *no_progress_loops, wmark); | |
4072 | if (wmark) { | |
15f570bf | 4073 | ret = true; |
132b0d21 | 4074 | break; |
0a0337e0 MH |
4075 | } |
4076 | } | |
4077 | ||
15f570bf MH |
4078 | /* |
4079 | * Memory allocation/reclaim might be called from a WQ context and the | |
4080 | * current implementation of the WQ concurrency control doesn't | |
4081 | * recognize that a particular WQ is congested if the worker thread is | |
4082 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4083 | * here rather than calling cond_resched(). | |
4084 | */ | |
4085 | if (current->flags & PF_WQ_WORKER) | |
4086 | schedule_timeout_uninterruptible(1); | |
4087 | else | |
4088 | cond_resched(); | |
ac3f3b0a CTK |
4089 | out: |
4090 | /* Before OOM, exhaust highatomic_reserve */ | |
4091 | if (!ret) | |
4092 | return unreserve_highatomic_pageblock(ac, true); | |
4093 | ||
15f570bf | 4094 | return ret; |
0a0337e0 MH |
4095 | } |
4096 | ||
902b6281 VB |
4097 | static inline bool |
4098 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4099 | { | |
4100 | /* | |
4101 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4102 | * mempolicy don't intersect. This should be normally dealt with by | |
4103 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4104 | * such a way the check therein was true, and then it became false | |
4105 | * before we got our cpuset_mems_cookie here. | |
4106 | * This assumes that for all allocations, ac->nodemask can come only | |
4107 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4108 | * when it does not intersect with the cpuset restrictions) or the | |
4109 | * caller can deal with a violated nodemask. | |
4110 | */ | |
4111 | if (cpusets_enabled() && ac->nodemask && | |
4112 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4113 | ac->nodemask = NULL; | |
4114 | return true; | |
4115 | } | |
4116 | ||
4117 | /* | |
4118 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4119 | * possible to race with parallel threads in such a way that our | |
4120 | * allocation can fail while the mask is being updated. If we are about | |
4121 | * to fail, check if the cpuset changed during allocation and if so, | |
4122 | * retry. | |
4123 | */ | |
4124 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4125 | return true; | |
4126 | ||
4127 | return false; | |
4128 | } | |
4129 | ||
11e33f6a MG |
4130 | static inline struct page * |
4131 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4132 | struct alloc_context *ac) |
11e33f6a | 4133 | { |
d0164adc | 4134 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
803de900 | 4135 | bool can_compact = gfp_compaction_allowed(gfp_mask); |
282722b0 | 4136 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4137 | struct page *page = NULL; |
c603844b | 4138 | unsigned int alloc_flags; |
11e33f6a | 4139 | unsigned long did_some_progress; |
5ce9bfef | 4140 | enum compact_priority compact_priority; |
c5d01d0d | 4141 | enum compact_result compact_result; |
5ce9bfef VB |
4142 | int compaction_retries; |
4143 | int no_progress_loops; | |
5ce9bfef | 4144 | unsigned int cpuset_mems_cookie; |
3d36424b | 4145 | unsigned int zonelist_iter_cookie; |
cd04ae1e | 4146 | int reserve_flags; |
1da177e4 | 4147 | |
3d36424b | 4148 | restart: |
5ce9bfef VB |
4149 | compaction_retries = 0; |
4150 | no_progress_loops = 0; | |
4151 | compact_priority = DEF_COMPACT_PRIORITY; | |
4152 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
3d36424b | 4153 | zonelist_iter_cookie = zonelist_iter_begin(); |
9a67f648 MH |
4154 | |
4155 | /* | |
4156 | * The fast path uses conservative alloc_flags to succeed only until | |
4157 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4158 | * alloc_flags precisely. So we do that now. | |
4159 | */ | |
eb2e2b42 | 4160 | alloc_flags = gfp_to_alloc_flags(gfp_mask, order); |
9a67f648 | 4161 | |
e47483bc VB |
4162 | /* |
4163 | * We need to recalculate the starting point for the zonelist iterator | |
4164 | * because we might have used different nodemask in the fast path, or | |
4165 | * there was a cpuset modification and we are retrying - otherwise we | |
4166 | * could end up iterating over non-eligible zones endlessly. | |
4167 | */ | |
4168 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 4169 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
4170 | if (!ac->preferred_zoneref->zone) |
4171 | goto nopage; | |
4172 | ||
8ca1b5a4 FT |
4173 | /* |
4174 | * Check for insane configurations where the cpuset doesn't contain | |
4175 | * any suitable zone to satisfy the request - e.g. non-movable | |
4176 | * GFP_HIGHUSER allocations from MOVABLE nodes only. | |
4177 | */ | |
4178 | if (cpusets_insane_config() && (gfp_mask & __GFP_HARDWALL)) { | |
4179 | struct zoneref *z = first_zones_zonelist(ac->zonelist, | |
4180 | ac->highest_zoneidx, | |
4181 | &cpuset_current_mems_allowed); | |
4182 | if (!z->zone) | |
4183 | goto nopage; | |
4184 | } | |
4185 | ||
0a79cdad | 4186 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4187 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4188 | |
4189 | /* | |
4190 | * The adjusted alloc_flags might result in immediate success, so try | |
4191 | * that first | |
4192 | */ | |
4193 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4194 | if (page) | |
4195 | goto got_pg; | |
4196 | ||
a8161d1e VB |
4197 | /* |
4198 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4199 | * that we have enough base pages and don't need to reclaim. For non- |
4200 | * movable high-order allocations, do that as well, as compaction will | |
4201 | * try prevent permanent fragmentation by migrating from blocks of the | |
4202 | * same migratetype. | |
4203 | * Don't try this for allocations that are allowed to ignore | |
4204 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4205 | */ |
803de900 | 4206 | if (can_direct_reclaim && can_compact && |
282722b0 VB |
4207 | (costly_order || |
4208 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4209 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4210 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4211 | alloc_flags, ac, | |
a5508cd8 | 4212 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4213 | &compact_result); |
4214 | if (page) | |
4215 | goto got_pg; | |
4216 | ||
cc638f32 VB |
4217 | /* |
4218 | * Checks for costly allocations with __GFP_NORETRY, which | |
4219 | * includes some THP page fault allocations | |
4220 | */ | |
4221 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4222 | /* |
4223 | * If allocating entire pageblock(s) and compaction | |
4224 | * failed because all zones are below low watermarks | |
4225 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4226 | * order, fail immediately unless the allocator has |
4227 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4228 | * |
4229 | * Reclaim is | |
4230 | * - potentially very expensive because zones are far | |
4231 | * below their low watermarks or this is part of very | |
4232 | * bursty high order allocations, | |
4233 | * - not guaranteed to help because isolate_freepages() | |
4234 | * may not iterate over freed pages as part of its | |
4235 | * linear scan, and | |
4236 | * - unlikely to make entire pageblocks free on its | |
4237 | * own. | |
4238 | */ | |
4239 | if (compact_result == COMPACT_SKIPPED || | |
4240 | compact_result == COMPACT_DEFERRED) | |
4241 | goto nopage; | |
a8161d1e | 4242 | |
a8161d1e | 4243 | /* |
3eb2771b VB |
4244 | * Looks like reclaim/compaction is worth trying, but |
4245 | * sync compaction could be very expensive, so keep | |
25160354 | 4246 | * using async compaction. |
a8161d1e | 4247 | */ |
a5508cd8 | 4248 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4249 | } |
4250 | } | |
23771235 | 4251 | |
31a6c190 | 4252 | retry: |
23771235 | 4253 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4254 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4255 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4256 | |
cd04ae1e MH |
4257 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4258 | if (reserve_flags) | |
ce96fa62 ML |
4259 | alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags) | |
4260 | (alloc_flags & ALLOC_KSWAPD); | |
23771235 | 4261 | |
e46e7b77 | 4262 | /* |
d6a24df0 VB |
4263 | * Reset the nodemask and zonelist iterators if memory policies can be |
4264 | * ignored. These allocations are high priority and system rather than | |
4265 | * user oriented. | |
e46e7b77 | 4266 | */ |
cd04ae1e | 4267 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4268 | ac->nodemask = NULL; |
e46e7b77 | 4269 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4270 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
4271 | } |
4272 | ||
23771235 | 4273 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4274 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4275 | if (page) |
4276 | goto got_pg; | |
1da177e4 | 4277 | |
d0164adc | 4278 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4279 | if (!can_direct_reclaim) |
1da177e4 LT |
4280 | goto nopage; |
4281 | ||
9a67f648 MH |
4282 | /* Avoid recursion of direct reclaim */ |
4283 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4284 | goto nopage; |
4285 | ||
a8161d1e VB |
4286 | /* Try direct reclaim and then allocating */ |
4287 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4288 | &did_some_progress); | |
4289 | if (page) | |
4290 | goto got_pg; | |
4291 | ||
4292 | /* Try direct compaction and then allocating */ | |
a9263751 | 4293 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4294 | compact_priority, &compact_result); |
56de7263 MG |
4295 | if (page) |
4296 | goto got_pg; | |
75f30861 | 4297 | |
9083905a JW |
4298 | /* Do not loop if specifically requested */ |
4299 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4300 | goto nopage; |
9083905a | 4301 | |
0a0337e0 MH |
4302 | /* |
4303 | * Do not retry costly high order allocations unless they are | |
803de900 | 4304 | * __GFP_RETRY_MAYFAIL and we can compact |
0a0337e0 | 4305 | */ |
803de900 VB |
4306 | if (costly_order && (!can_compact || |
4307 | !(gfp_mask & __GFP_RETRY_MAYFAIL))) | |
a8161d1e | 4308 | goto nopage; |
0a0337e0 | 4309 | |
0a0337e0 | 4310 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4311 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4312 | goto retry; |
4313 | ||
33c2d214 MH |
4314 | /* |
4315 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4316 | * reclaim is not able to make any progress because the current | |
4317 | * implementation of the compaction depends on the sufficient amount | |
4318 | * of free memory (see __compaction_suitable) | |
4319 | */ | |
803de900 | 4320 | if (did_some_progress > 0 && can_compact && |
86a294a8 | 4321 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4322 | compact_result, &compact_priority, |
d9436498 | 4323 | &compaction_retries)) |
33c2d214 MH |
4324 | goto retry; |
4325 | ||
902b6281 | 4326 | |
3d36424b MG |
4327 | /* |
4328 | * Deal with possible cpuset update races or zonelist updates to avoid | |
4329 | * a unnecessary OOM kill. | |
4330 | */ | |
4331 | if (check_retry_cpuset(cpuset_mems_cookie, ac) || | |
4332 | check_retry_zonelist(zonelist_iter_cookie)) | |
4333 | goto restart; | |
e47483bc | 4334 | |
9083905a JW |
4335 | /* Reclaim has failed us, start killing things */ |
4336 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4337 | if (page) | |
4338 | goto got_pg; | |
4339 | ||
9a67f648 | 4340 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e | 4341 | if (tsk_is_oom_victim(current) && |
8510e69c | 4342 | (alloc_flags & ALLOC_OOM || |
c288983d | 4343 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4344 | goto nopage; |
4345 | ||
9083905a | 4346 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4347 | if (did_some_progress) { |
4348 | no_progress_loops = 0; | |
9083905a | 4349 | goto retry; |
0a0337e0 | 4350 | } |
9083905a | 4351 | |
1da177e4 | 4352 | nopage: |
3d36424b MG |
4353 | /* |
4354 | * Deal with possible cpuset update races or zonelist updates to avoid | |
4355 | * a unnecessary OOM kill. | |
4356 | */ | |
4357 | if (check_retry_cpuset(cpuset_mems_cookie, ac) || | |
4358 | check_retry_zonelist(zonelist_iter_cookie)) | |
4359 | goto restart; | |
5ce9bfef | 4360 | |
9a67f648 MH |
4361 | /* |
4362 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4363 | * we always retry | |
4364 | */ | |
4365 | if (gfp_mask & __GFP_NOFAIL) { | |
4366 | /* | |
4367 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4368 | * of any new users that actually require GFP_NOWAIT | |
4369 | */ | |
3f913fc5 | 4370 | if (WARN_ON_ONCE_GFP(!can_direct_reclaim, gfp_mask)) |
9a67f648 MH |
4371 | goto fail; |
4372 | ||
4373 | /* | |
4374 | * PF_MEMALLOC request from this context is rather bizarre | |
4375 | * because we cannot reclaim anything and only can loop waiting | |
4376 | * for somebody to do a work for us | |
4377 | */ | |
3f913fc5 | 4378 | WARN_ON_ONCE_GFP(current->flags & PF_MEMALLOC, gfp_mask); |
9a67f648 MH |
4379 | |
4380 | /* | |
4381 | * non failing costly orders are a hard requirement which we | |
4382 | * are not prepared for much so let's warn about these users | |
4383 | * so that we can identify them and convert them to something | |
4384 | * else. | |
4385 | */ | |
896c4d52 | 4386 | WARN_ON_ONCE_GFP(costly_order, gfp_mask); |
9a67f648 | 4387 | |
6c18ba7a | 4388 | /* |
1ebbb218 MG |
4389 | * Help non-failing allocations by giving some access to memory |
4390 | * reserves normally used for high priority non-blocking | |
4391 | * allocations but do not use ALLOC_NO_WATERMARKS because this | |
6c18ba7a | 4392 | * could deplete whole memory reserves which would just make |
1ebbb218 | 4393 | * the situation worse. |
6c18ba7a | 4394 | */ |
1ebbb218 | 4395 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_MIN_RESERVE, ac); |
6c18ba7a MH |
4396 | if (page) |
4397 | goto got_pg; | |
4398 | ||
9a67f648 MH |
4399 | cond_resched(); |
4400 | goto retry; | |
4401 | } | |
4402 | fail: | |
a8e99259 | 4403 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4404 | "page allocation failure: order:%u", order); |
1da177e4 | 4405 | got_pg: |
072bb0aa | 4406 | return page; |
1da177e4 | 4407 | } |
11e33f6a | 4408 | |
9cd75558 | 4409 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4410 | int preferred_nid, nodemask_t *nodemask, |
8e6a930b | 4411 | struct alloc_context *ac, gfp_t *alloc_gfp, |
9cd75558 | 4412 | unsigned int *alloc_flags) |
11e33f6a | 4413 | { |
97a225e6 | 4414 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4415 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 4416 | ac->nodemask = nodemask; |
01c0bfe0 | 4417 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 4418 | |
682a3385 | 4419 | if (cpusets_enabled()) { |
8e6a930b | 4420 | *alloc_gfp |= __GFP_HARDWALL; |
182f3d7a MS |
4421 | /* |
4422 | * When we are in the interrupt context, it is irrelevant | |
4423 | * to the current task context. It means that any node ok. | |
4424 | */ | |
88dc6f20 | 4425 | if (in_task() && !ac->nodemask) |
9cd75558 | 4426 | ac->nodemask = &cpuset_current_mems_allowed; |
51047820 VB |
4427 | else |
4428 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4429 | } |
4430 | ||
446ec838 | 4431 | might_alloc(gfp_mask); |
11e33f6a MG |
4432 | |
4433 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4434 | return false; |
11e33f6a | 4435 | |
8e3560d9 | 4436 | *alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags); |
d883c6cf | 4437 | |
c9ab0c4f | 4438 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4439 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4440 | |
e46e7b77 MG |
4441 | /* |
4442 | * The preferred zone is used for statistics but crucially it is | |
4443 | * also used as the starting point for the zonelist iterator. It | |
4444 | * may get reset for allocations that ignore memory policies. | |
4445 | */ | |
9cd75558 | 4446 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4447 | ac->highest_zoneidx, ac->nodemask); |
a0622d05 MN |
4448 | |
4449 | return true; | |
9cd75558 MG |
4450 | } |
4451 | ||
387ba26f | 4452 | /* |
0f87d9d3 | 4453 | * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array |
387ba26f MG |
4454 | * @gfp: GFP flags for the allocation |
4455 | * @preferred_nid: The preferred NUMA node ID to allocate from | |
4456 | * @nodemask: Set of nodes to allocate from, may be NULL | |
0f87d9d3 MG |
4457 | * @nr_pages: The number of pages desired on the list or array |
4458 | * @page_list: Optional list to store the allocated pages | |
4459 | * @page_array: Optional array to store the pages | |
387ba26f MG |
4460 | * |
4461 | * This is a batched version of the page allocator that attempts to | |
0f87d9d3 MG |
4462 | * allocate nr_pages quickly. Pages are added to page_list if page_list |
4463 | * is not NULL, otherwise it is assumed that the page_array is valid. | |
387ba26f | 4464 | * |
0f87d9d3 MG |
4465 | * For lists, nr_pages is the number of pages that should be allocated. |
4466 | * | |
4467 | * For arrays, only NULL elements are populated with pages and nr_pages | |
4468 | * is the maximum number of pages that will be stored in the array. | |
4469 | * | |
4470 | * Returns the number of pages on the list or array. | |
387ba26f | 4471 | */ |
b951aaff | 4472 | unsigned long alloc_pages_bulk_noprof(gfp_t gfp, int preferred_nid, |
387ba26f | 4473 | nodemask_t *nodemask, int nr_pages, |
0f87d9d3 MG |
4474 | struct list_head *page_list, |
4475 | struct page **page_array) | |
387ba26f MG |
4476 | { |
4477 | struct page *page; | |
4b23a68f | 4478 | unsigned long __maybe_unused UP_flags; |
387ba26f MG |
4479 | struct zone *zone; |
4480 | struct zoneref *z; | |
4481 | struct per_cpu_pages *pcp; | |
4482 | struct list_head *pcp_list; | |
4483 | struct alloc_context ac; | |
4484 | gfp_t alloc_gfp; | |
4485 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
3e23060b | 4486 | int nr_populated = 0, nr_account = 0; |
387ba26f | 4487 | |
0f87d9d3 MG |
4488 | /* |
4489 | * Skip populated array elements to determine if any pages need | |
4490 | * to be allocated before disabling IRQs. | |
4491 | */ | |
b08e50dd | 4492 | while (page_array && nr_populated < nr_pages && page_array[nr_populated]) |
0f87d9d3 MG |
4493 | nr_populated++; |
4494 | ||
06147843 CL |
4495 | /* No pages requested? */ |
4496 | if (unlikely(nr_pages <= 0)) | |
4497 | goto out; | |
4498 | ||
b3b64ebd MG |
4499 | /* Already populated array? */ |
4500 | if (unlikely(page_array && nr_pages - nr_populated == 0)) | |
06147843 | 4501 | goto out; |
b3b64ebd | 4502 | |
8dcb3060 | 4503 | /* Bulk allocator does not support memcg accounting. */ |
f7a449f7 | 4504 | if (memcg_kmem_online() && (gfp & __GFP_ACCOUNT)) |
8dcb3060 SB |
4505 | goto failed; |
4506 | ||
387ba26f | 4507 | /* Use the single page allocator for one page. */ |
0f87d9d3 | 4508 | if (nr_pages - nr_populated == 1) |
387ba26f MG |
4509 | goto failed; |
4510 | ||
187ad460 MG |
4511 | #ifdef CONFIG_PAGE_OWNER |
4512 | /* | |
4513 | * PAGE_OWNER may recurse into the allocator to allocate space to | |
4514 | * save the stack with pagesets.lock held. Releasing/reacquiring | |
4515 | * removes much of the performance benefit of bulk allocation so | |
4516 | * force the caller to allocate one page at a time as it'll have | |
4517 | * similar performance to added complexity to the bulk allocator. | |
4518 | */ | |
4519 | if (static_branch_unlikely(&page_owner_inited)) | |
4520 | goto failed; | |
4521 | #endif | |
4522 | ||
387ba26f MG |
4523 | /* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */ |
4524 | gfp &= gfp_allowed_mask; | |
4525 | alloc_gfp = gfp; | |
4526 | if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags)) | |
06147843 | 4527 | goto out; |
387ba26f MG |
4528 | gfp = alloc_gfp; |
4529 | ||
4530 | /* Find an allowed local zone that meets the low watermark. */ | |
4531 | for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) { | |
4532 | unsigned long mark; | |
4533 | ||
4534 | if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) && | |
4535 | !__cpuset_zone_allowed(zone, gfp)) { | |
4536 | continue; | |
4537 | } | |
4538 | ||
4539 | if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone && | |
4540 | zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) { | |
4541 | goto failed; | |
4542 | } | |
4543 | ||
4544 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages; | |
4545 | if (zone_watermark_fast(zone, 0, mark, | |
4546 | zonelist_zone_idx(ac.preferred_zoneref), | |
4547 | alloc_flags, gfp)) { | |
4548 | break; | |
4549 | } | |
4550 | } | |
4551 | ||
4552 | /* | |
4553 | * If there are no allowed local zones that meets the watermarks then | |
4554 | * try to allocate a single page and reclaim if necessary. | |
4555 | */ | |
ce76f9a1 | 4556 | if (unlikely(!zone)) |
387ba26f MG |
4557 | goto failed; |
4558 | ||
57490774 | 4559 | /* spin_trylock may fail due to a parallel drain or IRQ reentrancy. */ |
4b23a68f | 4560 | pcp_trylock_prepare(UP_flags); |
57490774 | 4561 | pcp = pcp_spin_trylock(zone->per_cpu_pageset); |
01b44456 | 4562 | if (!pcp) |
4b23a68f | 4563 | goto failed_irq; |
387ba26f | 4564 | |
387ba26f | 4565 | /* Attempt the batch allocation */ |
44042b44 | 4566 | pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)]; |
0f87d9d3 MG |
4567 | while (nr_populated < nr_pages) { |
4568 | ||
4569 | /* Skip existing pages */ | |
4570 | if (page_array && page_array[nr_populated]) { | |
4571 | nr_populated++; | |
4572 | continue; | |
4573 | } | |
4574 | ||
44042b44 | 4575 | page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags, |
387ba26f | 4576 | pcp, pcp_list); |
ce76f9a1 | 4577 | if (unlikely(!page)) { |
c572e488 | 4578 | /* Try and allocate at least one page */ |
4b23a68f | 4579 | if (!nr_account) { |
57490774 | 4580 | pcp_spin_unlock(pcp); |
387ba26f | 4581 | goto failed_irq; |
4b23a68f | 4582 | } |
387ba26f MG |
4583 | break; |
4584 | } | |
3e23060b | 4585 | nr_account++; |
387ba26f MG |
4586 | |
4587 | prep_new_page(page, 0, gfp, 0); | |
0f87d9d3 MG |
4588 | if (page_list) |
4589 | list_add(&page->lru, page_list); | |
4590 | else | |
4591 | page_array[nr_populated] = page; | |
4592 | nr_populated++; | |
387ba26f MG |
4593 | } |
4594 | ||
57490774 | 4595 | pcp_spin_unlock(pcp); |
4b23a68f | 4596 | pcp_trylock_finish(UP_flags); |
43c95bcc | 4597 | |
3e23060b MG |
4598 | __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account); |
4599 | zone_statistics(ac.preferred_zoneref->zone, zone, nr_account); | |
387ba26f | 4600 | |
06147843 | 4601 | out: |
0f87d9d3 | 4602 | return nr_populated; |
387ba26f MG |
4603 | |
4604 | failed_irq: | |
4b23a68f | 4605 | pcp_trylock_finish(UP_flags); |
387ba26f MG |
4606 | |
4607 | failed: | |
b951aaff | 4608 | page = __alloc_pages_noprof(gfp, 0, preferred_nid, nodemask); |
387ba26f | 4609 | if (page) { |
0f87d9d3 MG |
4610 | if (page_list) |
4611 | list_add(&page->lru, page_list); | |
4612 | else | |
4613 | page_array[nr_populated] = page; | |
4614 | nr_populated++; | |
387ba26f MG |
4615 | } |
4616 | ||
06147843 | 4617 | goto out; |
387ba26f | 4618 | } |
b951aaff | 4619 | EXPORT_SYMBOL_GPL(alloc_pages_bulk_noprof); |
387ba26f | 4620 | |
9cd75558 MG |
4621 | /* |
4622 | * This is the 'heart' of the zoned buddy allocator. | |
4623 | */ | |
b951aaff SB |
4624 | struct page *__alloc_pages_noprof(gfp_t gfp, unsigned int order, |
4625 | int preferred_nid, nodemask_t *nodemask) | |
9cd75558 MG |
4626 | { |
4627 | struct page *page; | |
4628 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
8e6a930b | 4629 | gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4630 | struct alloc_context ac = { }; |
4631 | ||
c63ae43b MH |
4632 | /* |
4633 | * There are several places where we assume that the order value is sane | |
4634 | * so bail out early if the request is out of bound. | |
4635 | */ | |
5e0a760b | 4636 | if (WARN_ON_ONCE_GFP(order > MAX_PAGE_ORDER, gfp)) |
c63ae43b | 4637 | return NULL; |
c63ae43b | 4638 | |
6e5e0f28 | 4639 | gfp &= gfp_allowed_mask; |
da6df1b0 PT |
4640 | /* |
4641 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS | |
4642 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4643 | * from a particular context which has been marked by | |
8e3560d9 PT |
4644 | * memalloc_no{fs,io}_{save,restore}. And PF_MEMALLOC_PIN which ensures |
4645 | * movable zones are not used during allocation. | |
da6df1b0 PT |
4646 | */ |
4647 | gfp = current_gfp_context(gfp); | |
6e5e0f28 MWO |
4648 | alloc_gfp = gfp; |
4649 | if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac, | |
8e6a930b | 4650 | &alloc_gfp, &alloc_flags)) |
9cd75558 MG |
4651 | return NULL; |
4652 | ||
6bb15450 MG |
4653 | /* |
4654 | * Forbid the first pass from falling back to types that fragment | |
4655 | * memory until all local zones are considered. | |
4656 | */ | |
6e5e0f28 | 4657 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp); |
6bb15450 | 4658 | |
5117f45d | 4659 | /* First allocation attempt */ |
8e6a930b | 4660 | page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac); |
4fcb0971 MG |
4661 | if (likely(page)) |
4662 | goto out; | |
11e33f6a | 4663 | |
da6df1b0 | 4664 | alloc_gfp = gfp; |
4fcb0971 | 4665 | ac.spread_dirty_pages = false; |
23f086f9 | 4666 | |
4741526b MG |
4667 | /* |
4668 | * Restore the original nodemask if it was potentially replaced with | |
4669 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4670 | */ | |
97ce86f9 | 4671 | ac.nodemask = nodemask; |
16096c25 | 4672 | |
8e6a930b | 4673 | page = __alloc_pages_slowpath(alloc_gfp, order, &ac); |
cc9a6c87 | 4674 | |
4fcb0971 | 4675 | out: |
f7a449f7 | 4676 | if (memcg_kmem_online() && (gfp & __GFP_ACCOUNT) && page && |
6e5e0f28 | 4677 | unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) { |
c4159a75 VD |
4678 | __free_pages(page, order); |
4679 | page = NULL; | |
4949148a VD |
4680 | } |
4681 | ||
8e6a930b | 4682 | trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype); |
b073d7f8 | 4683 | kmsan_alloc_page(page, order, alloc_gfp); |
4fcb0971 | 4684 | |
11e33f6a | 4685 | return page; |
1da177e4 | 4686 | } |
b951aaff | 4687 | EXPORT_SYMBOL(__alloc_pages_noprof); |
1da177e4 | 4688 | |
b951aaff | 4689 | struct folio *__folio_alloc_noprof(gfp_t gfp, unsigned int order, int preferred_nid, |
cc09cb13 MWO |
4690 | nodemask_t *nodemask) |
4691 | { | |
b951aaff | 4692 | struct page *page = __alloc_pages_noprof(gfp | __GFP_COMP, order, |
23e48832 HD |
4693 | preferred_nid, nodemask); |
4694 | return page_rmappable_folio(page); | |
cc09cb13 | 4695 | } |
b951aaff | 4696 | EXPORT_SYMBOL(__folio_alloc_noprof); |
cc09cb13 | 4697 | |
1da177e4 | 4698 | /* |
9ea9a680 MH |
4699 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4700 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4701 | * you need to access high mem. | |
1da177e4 | 4702 | */ |
b951aaff | 4703 | unsigned long get_free_pages_noprof(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4704 | { |
945a1113 AM |
4705 | struct page *page; |
4706 | ||
b951aaff | 4707 | page = alloc_pages_noprof(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4708 | if (!page) |
4709 | return 0; | |
4710 | return (unsigned long) page_address(page); | |
4711 | } | |
b951aaff | 4712 | EXPORT_SYMBOL(get_free_pages_noprof); |
1da177e4 | 4713 | |
b951aaff | 4714 | unsigned long get_zeroed_page_noprof(gfp_t gfp_mask) |
1da177e4 | 4715 | { |
b951aaff | 4716 | return get_free_pages_noprof(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4717 | } |
b951aaff | 4718 | EXPORT_SYMBOL(get_zeroed_page_noprof); |
1da177e4 | 4719 | |
7f194fbb MWO |
4720 | /** |
4721 | * __free_pages - Free pages allocated with alloc_pages(). | |
4722 | * @page: The page pointer returned from alloc_pages(). | |
4723 | * @order: The order of the allocation. | |
4724 | * | |
4725 | * This function can free multi-page allocations that are not compound | |
4726 | * pages. It does not check that the @order passed in matches that of | |
4727 | * the allocation, so it is easy to leak memory. Freeing more memory | |
4728 | * than was allocated will probably emit a warning. | |
4729 | * | |
4730 | * If the last reference to this page is speculative, it will be released | |
4731 | * by put_page() which only frees the first page of a non-compound | |
4732 | * allocation. To prevent the remaining pages from being leaked, we free | |
4733 | * the subsequent pages here. If you want to use the page's reference | |
4734 | * count to decide when to free the allocation, you should allocate a | |
4735 | * compound page, and use put_page() instead of __free_pages(). | |
4736 | * | |
4737 | * Context: May be called in interrupt context or while holding a normal | |
4738 | * spinlock, but not in NMI context or while holding a raw spinlock. | |
4739 | */ | |
742aa7fb AL |
4740 | void __free_pages(struct page *page, unsigned int order) |
4741 | { | |
462a8e08 DC |
4742 | /* get PageHead before we drop reference */ |
4743 | int head = PageHead(page); | |
cc92eba1 | 4744 | struct alloc_tag *tag = pgalloc_tag_get(page); |
462a8e08 | 4745 | |
742aa7fb | 4746 | if (put_page_testzero(page)) |
5b8d7591 | 4747 | free_unref_page(page, order); |
cc92eba1 SB |
4748 | else if (!head) { |
4749 | pgalloc_tag_sub_pages(tag, (1 << order) - 1); | |
e320d301 | 4750 | while (order-- > 0) |
5b8d7591 | 4751 | free_unref_page(page + (1 << order), order); |
cc92eba1 | 4752 | } |
742aa7fb | 4753 | } |
1da177e4 LT |
4754 | EXPORT_SYMBOL(__free_pages); |
4755 | ||
920c7a5d | 4756 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4757 | { |
4758 | if (addr != 0) { | |
725d704e | 4759 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4760 | __free_pages(virt_to_page((void *)addr), order); |
4761 | } | |
4762 | } | |
4763 | ||
4764 | EXPORT_SYMBOL(free_pages); | |
4765 | ||
b63ae8ca AD |
4766 | /* |
4767 | * Page Fragment: | |
4768 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4769 | * within a 0 or higher order page. Multiple fragments within that page | |
4770 | * are individually refcounted, in the page's reference counter. | |
4771 | * | |
4772 | * The page_frag functions below provide a simple allocation framework for | |
4773 | * page fragments. This is used by the network stack and network device | |
4774 | * drivers to provide a backing region of memory for use as either an | |
4775 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4776 | */ | |
2976db80 AD |
4777 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4778 | gfp_t gfp_mask) | |
b63ae8ca AD |
4779 | { |
4780 | struct page *page = NULL; | |
4781 | gfp_t gfp = gfp_mask; | |
4782 | ||
4783 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4bc0d63a YL |
4784 | gfp_mask = (gfp_mask & ~__GFP_DIRECT_RECLAIM) | __GFP_COMP | |
4785 | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC; | |
b63ae8ca AD |
4786 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, |
4787 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4788 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4789 | #endif | |
4790 | if (unlikely(!page)) | |
4791 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4792 | ||
4793 | nc->va = page ? page_address(page) : NULL; | |
4794 | ||
4795 | return page; | |
4796 | } | |
4797 | ||
a0727489 YL |
4798 | void page_frag_cache_drain(struct page_frag_cache *nc) |
4799 | { | |
4800 | if (!nc->va) | |
4801 | return; | |
4802 | ||
4803 | __page_frag_cache_drain(virt_to_head_page(nc->va), nc->pagecnt_bias); | |
4804 | nc->va = NULL; | |
4805 | } | |
4806 | EXPORT_SYMBOL(page_frag_cache_drain); | |
4807 | ||
2976db80 | 4808 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4809 | { |
4810 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4811 | ||
742aa7fb | 4812 | if (page_ref_sub_and_test(page, count)) |
5b8d7591 | 4813 | free_unref_page(page, compound_order(page)); |
44fdffd7 | 4814 | } |
2976db80 | 4815 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4816 | |
411c5f36 YL |
4817 | void *__page_frag_alloc_align(struct page_frag_cache *nc, |
4818 | unsigned int fragsz, gfp_t gfp_mask, | |
4819 | unsigned int align_mask) | |
b63ae8ca AD |
4820 | { |
4821 | unsigned int size = PAGE_SIZE; | |
4822 | struct page *page; | |
4823 | int offset; | |
4824 | ||
4825 | if (unlikely(!nc->va)) { | |
4826 | refill: | |
2976db80 | 4827 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4828 | if (!page) |
4829 | return NULL; | |
4830 | ||
4831 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4832 | /* if size can vary use size else just use PAGE_SIZE */ | |
4833 | size = nc->size; | |
4834 | #endif | |
4835 | /* Even if we own the page, we do not use atomic_set(). | |
4836 | * This would break get_page_unless_zero() users. | |
4837 | */ | |
86447726 | 4838 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
4839 | |
4840 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4841 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 4842 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
4843 | nc->offset = size; |
4844 | } | |
4845 | ||
4846 | offset = nc->offset - fragsz; | |
4847 | if (unlikely(offset < 0)) { | |
4848 | page = virt_to_page(nc->va); | |
4849 | ||
fe896d18 | 4850 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4851 | goto refill; |
4852 | ||
d8c19014 | 4853 | if (unlikely(nc->pfmemalloc)) { |
5b8d7591 | 4854 | free_unref_page(page, compound_order(page)); |
d8c19014 DZ |
4855 | goto refill; |
4856 | } | |
4857 | ||
b63ae8ca AD |
4858 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
4859 | /* if size can vary use size else just use PAGE_SIZE */ | |
4860 | size = nc->size; | |
4861 | #endif | |
4862 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 4863 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
4864 | |
4865 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 4866 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca | 4867 | offset = size - fragsz; |
dac22531 ML |
4868 | if (unlikely(offset < 0)) { |
4869 | /* | |
4870 | * The caller is trying to allocate a fragment | |
4871 | * with fragsz > PAGE_SIZE but the cache isn't big | |
4872 | * enough to satisfy the request, this may | |
4873 | * happen in low memory conditions. | |
4874 | * We don't release the cache page because | |
4875 | * it could make memory pressure worse | |
4876 | * so we simply return NULL here. | |
4877 | */ | |
4878 | return NULL; | |
4879 | } | |
b63ae8ca AD |
4880 | } |
4881 | ||
4882 | nc->pagecnt_bias--; | |
b358e212 | 4883 | offset &= align_mask; |
b63ae8ca AD |
4884 | nc->offset = offset; |
4885 | ||
4886 | return nc->va + offset; | |
4887 | } | |
411c5f36 | 4888 | EXPORT_SYMBOL(__page_frag_alloc_align); |
b63ae8ca AD |
4889 | |
4890 | /* | |
4891 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4892 | */ | |
8c2dd3e4 | 4893 | void page_frag_free(void *addr) |
b63ae8ca AD |
4894 | { |
4895 | struct page *page = virt_to_head_page(addr); | |
4896 | ||
742aa7fb | 4897 | if (unlikely(put_page_testzero(page))) |
5b8d7591 | 4898 | free_unref_page(page, compound_order(page)); |
b63ae8ca | 4899 | } |
8c2dd3e4 | 4900 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4901 | |
d00181b9 KS |
4902 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4903 | size_t size) | |
ee85c2e1 AK |
4904 | { |
4905 | if (addr) { | |
df48a5f7 LH |
4906 | unsigned long nr = DIV_ROUND_UP(size, PAGE_SIZE); |
4907 | struct page *page = virt_to_page((void *)addr); | |
4908 | struct page *last = page + nr; | |
4909 | ||
46d44d09 | 4910 | split_page_owner(page, order, 0); |
be25d1d4 | 4911 | pgalloc_tag_split(page, 1 << order); |
b8791381 | 4912 | split_page_memcg(page, order, 0); |
df48a5f7 LH |
4913 | while (page < --last) |
4914 | set_page_refcounted(last); | |
4915 | ||
4916 | last = page + (1UL << order); | |
4917 | for (page += nr; page < last; page++) | |
4918 | __free_pages_ok(page, 0, FPI_TO_TAIL); | |
ee85c2e1 AK |
4919 | } |
4920 | return (void *)addr; | |
4921 | } | |
4922 | ||
2be0ffe2 TT |
4923 | /** |
4924 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4925 | * @size: the number of bytes to allocate | |
63931eb9 | 4926 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
4927 | * |
4928 | * This function is similar to alloc_pages(), except that it allocates the | |
4929 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4930 | * allocate memory in power-of-two pages. | |
4931 | * | |
5e0a760b | 4932 | * This function is also limited by MAX_PAGE_ORDER. |
2be0ffe2 TT |
4933 | * |
4934 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
4935 | * |
4936 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 | 4937 | */ |
b951aaff | 4938 | void *alloc_pages_exact_noprof(size_t size, gfp_t gfp_mask) |
2be0ffe2 TT |
4939 | { |
4940 | unsigned int order = get_order(size); | |
4941 | unsigned long addr; | |
4942 | ||
ba7f1b9e ML |
4943 | if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM))) |
4944 | gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM); | |
63931eb9 | 4945 | |
b951aaff | 4946 | addr = get_free_pages_noprof(gfp_mask, order); |
ee85c2e1 | 4947 | return make_alloc_exact(addr, order, size); |
2be0ffe2 | 4948 | } |
b951aaff | 4949 | EXPORT_SYMBOL(alloc_pages_exact_noprof); |
2be0ffe2 | 4950 | |
ee85c2e1 AK |
4951 | /** |
4952 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4953 | * pages on a node. | |
b5e6ab58 | 4954 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 4955 | * @size: the number of bytes to allocate |
63931eb9 | 4956 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
4957 | * |
4958 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4959 | * back. | |
a862f68a MR |
4960 | * |
4961 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 4962 | */ |
b951aaff | 4963 | void * __meminit alloc_pages_exact_nid_noprof(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4964 | { |
d00181b9 | 4965 | unsigned int order = get_order(size); |
63931eb9 VB |
4966 | struct page *p; |
4967 | ||
ba7f1b9e ML |
4968 | if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM))) |
4969 | gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM); | |
63931eb9 | 4970 | |
b951aaff | 4971 | p = alloc_pages_node_noprof(nid, gfp_mask, order); |
ee85c2e1 AK |
4972 | if (!p) |
4973 | return NULL; | |
4974 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4975 | } | |
ee85c2e1 | 4976 | |
2be0ffe2 TT |
4977 | /** |
4978 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4979 | * @virt: the value returned by alloc_pages_exact. | |
4980 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4981 | * | |
4982 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4983 | */ | |
4984 | void free_pages_exact(void *virt, size_t size) | |
4985 | { | |
4986 | unsigned long addr = (unsigned long)virt; | |
4987 | unsigned long end = addr + PAGE_ALIGN(size); | |
4988 | ||
4989 | while (addr < end) { | |
4990 | free_page(addr); | |
4991 | addr += PAGE_SIZE; | |
4992 | } | |
4993 | } | |
4994 | EXPORT_SYMBOL(free_pages_exact); | |
4995 | ||
e0fb5815 ZY |
4996 | /** |
4997 | * nr_free_zone_pages - count number of pages beyond high watermark | |
4998 | * @offset: The zone index of the highest zone | |
4999 | * | |
a862f68a | 5000 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5001 | * high watermark within all zones at or below a given zone index. For each |
5002 | * zone, the number of pages is calculated as: | |
0e056eb5 | 5003 | * |
5004 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5005 | * |
5006 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5007 | */ |
ebec3862 | 5008 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5009 | { |
dd1a239f | 5010 | struct zoneref *z; |
54a6eb5c MG |
5011 | struct zone *zone; |
5012 | ||
e310fd43 | 5013 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5014 | unsigned long sum = 0; |
1da177e4 | 5015 | |
0e88460d | 5016 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5017 | |
54a6eb5c | 5018 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5019 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5020 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5021 | if (size > high) |
5022 | sum += size - high; | |
1da177e4 LT |
5023 | } |
5024 | ||
5025 | return sum; | |
5026 | } | |
5027 | ||
e0fb5815 ZY |
5028 | /** |
5029 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5030 | * | |
5031 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5032 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5033 | * |
5034 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5035 | * ZONE_NORMAL. | |
1da177e4 | 5036 | */ |
ebec3862 | 5037 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5038 | { |
af4ca457 | 5039 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5040 | } |
c2f1a551 | 5041 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5042 | |
19770b32 MG |
5043 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5044 | { | |
5045 | zoneref->zone = zone; | |
5046 | zoneref->zone_idx = zone_idx(zone); | |
5047 | } | |
5048 | ||
1da177e4 LT |
5049 | /* |
5050 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5051 | * |
5052 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5053 | */ |
9d3be21b | 5054 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5055 | { |
1a93205b | 5056 | struct zone *zone; |
bc732f1d | 5057 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5058 | int nr_zones = 0; |
02a68a5e CL |
5059 | |
5060 | do { | |
2f6726e5 | 5061 | zone_type--; |
070f8032 | 5062 | zone = pgdat->node_zones + zone_type; |
e553f62f | 5063 | if (populated_zone(zone)) { |
9d3be21b | 5064 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5065 | check_highest_zone(zone_type); |
1da177e4 | 5066 | } |
2f6726e5 | 5067 | } while (zone_type); |
bc732f1d | 5068 | |
070f8032 | 5069 | return nr_zones; |
1da177e4 LT |
5070 | } |
5071 | ||
5072 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5073 | |
5074 | static int __parse_numa_zonelist_order(char *s) | |
5075 | { | |
c9bff3ee | 5076 | /* |
f0953a1b | 5077 | * We used to support different zonelists modes but they turned |
c9bff3ee MH |
5078 | * out to be just not useful. Let's keep the warning in place |
5079 | * if somebody still use the cmd line parameter so that we do | |
5080 | * not fail it silently | |
5081 | */ | |
5082 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5083 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5084 | return -EINVAL; |
5085 | } | |
5086 | return 0; | |
5087 | } | |
5088 | ||
e95d372c KW |
5089 | static char numa_zonelist_order[] = "Node"; |
5090 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
f0c0b2b8 KH |
5091 | /* |
5092 | * sysctl handler for numa_zonelist_order | |
5093 | */ | |
e95d372c | 5094 | static int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 5095 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 5096 | { |
32927393 CH |
5097 | if (write) |
5098 | return __parse_numa_zonelist_order(buffer); | |
5099 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
5100 | } |
5101 | ||
f0c0b2b8 KH |
5102 | static int node_load[MAX_NUMNODES]; |
5103 | ||
1da177e4 | 5104 | /** |
4dc3b16b | 5105 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5106 | * @node: node whose fallback list we're appending |
5107 | * @used_node_mask: nodemask_t of already used nodes | |
5108 | * | |
5109 | * We use a number of factors to determine which is the next node that should | |
5110 | * appear on a given node's fallback list. The node should not have appeared | |
5111 | * already in @node's fallback list, and it should be the next closest node | |
5112 | * according to the distance array (which contains arbitrary distance values | |
5113 | * from each node to each node in the system), and should also prefer nodes | |
5114 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5115 | * on them otherwise. | |
a862f68a MR |
5116 | * |
5117 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 5118 | */ |
79c28a41 | 5119 | int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5120 | { |
4cf808eb | 5121 | int n, val; |
1da177e4 | 5122 | int min_val = INT_MAX; |
00ef2d2f | 5123 | int best_node = NUMA_NO_NODE; |
1da177e4 | 5124 | |
c2baef39 QZ |
5125 | /* |
5126 | * Use the local node if we haven't already, but for memoryless local | |
5127 | * node, we should skip it and fall back to other nodes. | |
5128 | */ | |
5129 | if (!node_isset(node, *used_node_mask) && node_state(node, N_MEMORY)) { | |
4cf808eb LT |
5130 | node_set(node, *used_node_mask); |
5131 | return node; | |
5132 | } | |
1da177e4 | 5133 | |
4b0ef1fe | 5134 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5135 | |
5136 | /* Don't want a node to appear more than once */ | |
5137 | if (node_isset(n, *used_node_mask)) | |
5138 | continue; | |
5139 | ||
1da177e4 LT |
5140 | /* Use the distance array to find the distance */ |
5141 | val = node_distance(node, n); | |
5142 | ||
4cf808eb LT |
5143 | /* Penalize nodes under us ("prefer the next node") */ |
5144 | val += (n < node); | |
5145 | ||
1da177e4 | 5146 | /* Give preference to headless and unused nodes */ |
b630749f | 5147 | if (!cpumask_empty(cpumask_of_node(n))) |
1da177e4 LT |
5148 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5149 | ||
5150 | /* Slight preference for less loaded node */ | |
37931324 | 5151 | val *= MAX_NUMNODES; |
1da177e4 LT |
5152 | val += node_load[n]; |
5153 | ||
5154 | if (val < min_val) { | |
5155 | min_val = val; | |
5156 | best_node = n; | |
5157 | } | |
5158 | } | |
5159 | ||
5160 | if (best_node >= 0) | |
5161 | node_set(best_node, *used_node_mask); | |
5162 | ||
5163 | return best_node; | |
5164 | } | |
5165 | ||
f0c0b2b8 KH |
5166 | |
5167 | /* | |
5168 | * Build zonelists ordered by node and zones within node. | |
5169 | * This results in maximum locality--normal zone overflows into local | |
5170 | * DMA zone, if any--but risks exhausting DMA zone. | |
5171 | */ | |
9d3be21b MH |
5172 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5173 | unsigned nr_nodes) | |
1da177e4 | 5174 | { |
9d3be21b MH |
5175 | struct zoneref *zonerefs; |
5176 | int i; | |
5177 | ||
5178 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5179 | ||
5180 | for (i = 0; i < nr_nodes; i++) { | |
5181 | int nr_zones; | |
5182 | ||
5183 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5184 | |
9d3be21b MH |
5185 | nr_zones = build_zonerefs_node(node, zonerefs); |
5186 | zonerefs += nr_zones; | |
5187 | } | |
5188 | zonerefs->zone = NULL; | |
5189 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5190 | } |
5191 | ||
523b9458 CL |
5192 | /* |
5193 | * Build gfp_thisnode zonelists | |
5194 | */ | |
5195 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5196 | { | |
9d3be21b MH |
5197 | struct zoneref *zonerefs; |
5198 | int nr_zones; | |
523b9458 | 5199 | |
9d3be21b MH |
5200 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5201 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5202 | zonerefs += nr_zones; | |
5203 | zonerefs->zone = NULL; | |
5204 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5205 | } |
5206 | ||
f0c0b2b8 KH |
5207 | /* |
5208 | * Build zonelists ordered by zone and nodes within zones. | |
5209 | * This results in conserving DMA zone[s] until all Normal memory is | |
5210 | * exhausted, but results in overflowing to remote node while memory | |
5211 | * may still exist in local DMA zone. | |
5212 | */ | |
f0c0b2b8 | 5213 | |
f0c0b2b8 KH |
5214 | static void build_zonelists(pg_data_t *pgdat) |
5215 | { | |
9d3be21b | 5216 | static int node_order[MAX_NUMNODES]; |
37931324 | 5217 | int node, nr_nodes = 0; |
d0ddf49b | 5218 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 5219 | int local_node, prev_node; |
1da177e4 LT |
5220 | |
5221 | /* NUMA-aware ordering of nodes */ | |
5222 | local_node = pgdat->node_id; | |
1da177e4 | 5223 | prev_node = local_node; |
f0c0b2b8 | 5224 | |
f0c0b2b8 | 5225 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5226 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5227 | /* | |
5228 | * We don't want to pressure a particular node. | |
5229 | * So adding penalty to the first node in same | |
5230 | * distance group to make it round-robin. | |
5231 | */ | |
957f822a DR |
5232 | if (node_distance(local_node, node) != |
5233 | node_distance(local_node, prev_node)) | |
37931324 | 5234 | node_load[node] += 1; |
f0c0b2b8 | 5235 | |
9d3be21b | 5236 | node_order[nr_nodes++] = node; |
1da177e4 | 5237 | prev_node = node; |
1da177e4 | 5238 | } |
523b9458 | 5239 | |
9d3be21b | 5240 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5241 | build_thisnode_zonelists(pgdat); |
6cf25392 BR |
5242 | pr_info("Fallback order for Node %d: ", local_node); |
5243 | for (node = 0; node < nr_nodes; node++) | |
5244 | pr_cont("%d ", node_order[node]); | |
5245 | pr_cont("\n"); | |
1da177e4 LT |
5246 | } |
5247 | ||
7aac7898 LS |
5248 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5249 | /* | |
5250 | * Return node id of node used for "local" allocations. | |
5251 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5252 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5253 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5254 | */ | |
5255 | int local_memory_node(int node) | |
5256 | { | |
c33d6c06 | 5257 | struct zoneref *z; |
7aac7898 | 5258 | |
c33d6c06 | 5259 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5260 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 5261 | NULL); |
c1093b74 | 5262 | return zone_to_nid(z->zone); |
7aac7898 LS |
5263 | } |
5264 | #endif | |
f0c0b2b8 | 5265 | |
6423aa81 JK |
5266 | static void setup_min_unmapped_ratio(void); |
5267 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5268 | #else /* CONFIG_NUMA */ |
5269 | ||
f0c0b2b8 | 5270 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5271 | { |
9d3be21b MH |
5272 | struct zoneref *zonerefs; |
5273 | int nr_zones; | |
1da177e4 | 5274 | |
9d3be21b MH |
5275 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5276 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5277 | zonerefs += nr_zones; | |
1da177e4 | 5278 | |
9d3be21b MH |
5279 | zonerefs->zone = NULL; |
5280 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5281 | } |
5282 | ||
5283 | #endif /* CONFIG_NUMA */ | |
5284 | ||
99dcc3e5 CL |
5285 | /* |
5286 | * Boot pageset table. One per cpu which is going to be used for all | |
5287 | * zones and all nodes. The parameters will be set in such a way | |
5288 | * that an item put on a list will immediately be handed over to | |
5289 | * the buddy list. This is safe since pageset manipulation is done | |
5290 | * with interrupts disabled. | |
5291 | * | |
5292 | * The boot_pagesets must be kept even after bootup is complete for | |
5293 | * unused processors and/or zones. They do play a role for bootstrapping | |
5294 | * hotplugged processors. | |
5295 | * | |
5296 | * zoneinfo_show() and maybe other functions do | |
5297 | * not check if the processor is online before following the pageset pointer. | |
5298 | * Other parts of the kernel may not check if the zone is available. | |
5299 | */ | |
28f836b6 | 5300 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats); |
952eaf81 VB |
5301 | /* These effectively disable the pcplists in the boot pageset completely */ |
5302 | #define BOOT_PAGESET_HIGH 0 | |
5303 | #define BOOT_PAGESET_BATCH 1 | |
28f836b6 MG |
5304 | static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset); |
5305 | static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats); | |
99dcc3e5 | 5306 | |
11cd8638 | 5307 | static void __build_all_zonelists(void *data) |
1da177e4 | 5308 | { |
6811378e | 5309 | int nid; |
afb6ebb3 | 5310 | int __maybe_unused cpu; |
9adb62a5 | 5311 | pg_data_t *self = data; |
1007843a | 5312 | unsigned long flags; |
b93e0f32 | 5313 | |
1007843a | 5314 | /* |
a2ebb515 SAS |
5315 | * The zonelist_update_seq must be acquired with irqsave because the |
5316 | * reader can be invoked from IRQ with GFP_ATOMIC. | |
1007843a | 5317 | */ |
a2ebb515 | 5318 | write_seqlock_irqsave(&zonelist_update_seq, flags); |
1007843a | 5319 | /* |
a2ebb515 SAS |
5320 | * Also disable synchronous printk() to prevent any printk() from |
5321 | * trying to hold port->lock, for | |
1007843a TH |
5322 | * tty_insert_flip_string_and_push_buffer() on other CPU might be |
5323 | * calling kmalloc(GFP_ATOMIC | __GFP_NOWARN) with port->lock held. | |
5324 | */ | |
5325 | printk_deferred_enter(); | |
9276b1bc | 5326 | |
7f9cfb31 BL |
5327 | #ifdef CONFIG_NUMA |
5328 | memset(node_load, 0, sizeof(node_load)); | |
5329 | #endif | |
9adb62a5 | 5330 | |
c1152583 WY |
5331 | /* |
5332 | * This node is hotadded and no memory is yet present. So just | |
5333 | * building zonelists is fine - no need to touch other nodes. | |
5334 | */ | |
9adb62a5 JL |
5335 | if (self && !node_online(self->node_id)) { |
5336 | build_zonelists(self); | |
c1152583 | 5337 | } else { |
09f49dca MH |
5338 | /* |
5339 | * All possible nodes have pgdat preallocated | |
5340 | * in free_area_init | |
5341 | */ | |
5342 | for_each_node(nid) { | |
c1152583 | 5343 | pg_data_t *pgdat = NODE_DATA(nid); |
7ea1530a | 5344 | |
c1152583 WY |
5345 | build_zonelists(pgdat); |
5346 | } | |
99dcc3e5 | 5347 | |
7aac7898 LS |
5348 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5349 | /* | |
5350 | * We now know the "local memory node" for each node-- | |
5351 | * i.e., the node of the first zone in the generic zonelist. | |
5352 | * Set up numa_mem percpu variable for on-line cpus. During | |
5353 | * boot, only the boot cpu should be on-line; we'll init the | |
5354 | * secondary cpus' numa_mem as they come on-line. During | |
5355 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5356 | */ | |
d9c9a0b9 | 5357 | for_each_online_cpu(cpu) |
7aac7898 | 5358 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5359 | #endif |
d9c9a0b9 | 5360 | } |
b93e0f32 | 5361 | |
1007843a | 5362 | printk_deferred_exit(); |
a2ebb515 | 5363 | write_sequnlock_irqrestore(&zonelist_update_seq, flags); |
6811378e YG |
5364 | } |
5365 | ||
061f67bc RV |
5366 | static noinline void __init |
5367 | build_all_zonelists_init(void) | |
5368 | { | |
afb6ebb3 MH |
5369 | int cpu; |
5370 | ||
061f67bc | 5371 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5372 | |
5373 | /* | |
5374 | * Initialize the boot_pagesets that are going to be used | |
5375 | * for bootstrapping processors. The real pagesets for | |
5376 | * each zone will be allocated later when the per cpu | |
5377 | * allocator is available. | |
5378 | * | |
5379 | * boot_pagesets are used also for bootstrapping offline | |
5380 | * cpus if the system is already booted because the pagesets | |
5381 | * are needed to initialize allocators on a specific cpu too. | |
5382 | * F.e. the percpu allocator needs the page allocator which | |
5383 | * needs the percpu allocator in order to allocate its pagesets | |
5384 | * (a chicken-egg dilemma). | |
5385 | */ | |
5386 | for_each_possible_cpu(cpu) | |
28f836b6 | 5387 | per_cpu_pages_init(&per_cpu(boot_pageset, cpu), &per_cpu(boot_zonestats, cpu)); |
afb6ebb3 | 5388 | |
061f67bc RV |
5389 | mminit_verify_zonelist(); |
5390 | cpuset_init_current_mems_allowed(); | |
5391 | } | |
5392 | ||
4eaf3f64 | 5393 | /* |
4eaf3f64 | 5394 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5395 | * |
72675e13 | 5396 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5397 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5398 | */ |
72675e13 | 5399 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 5400 | { |
0a18e607 DH |
5401 | unsigned long vm_total_pages; |
5402 | ||
6811378e | 5403 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 5404 | build_all_zonelists_init(); |
6811378e | 5405 | } else { |
11cd8638 | 5406 | __build_all_zonelists(pgdat); |
6811378e YG |
5407 | /* cpuset refresh routine should be here */ |
5408 | } | |
56b9413b DH |
5409 | /* Get the number of free pages beyond high watermark in all zones. */ |
5410 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
5411 | /* |
5412 | * Disable grouping by mobility if the number of pages in the | |
5413 | * system is too low to allow the mechanism to work. It would be | |
5414 | * more accurate, but expensive to check per-zone. This check is | |
5415 | * made on memory-hotadd so a system can start with mobility | |
5416 | * disabled and enable it later | |
5417 | */ | |
d9c23400 | 5418 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5419 | page_group_by_mobility_disabled = 1; |
5420 | else | |
5421 | page_group_by_mobility_disabled = 0; | |
5422 | ||
ce0725f7 | 5423 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5424 | nr_online_nodes, |
756a025f JP |
5425 | page_group_by_mobility_disabled ? "off" : "on", |
5426 | vm_total_pages); | |
f0c0b2b8 | 5427 | #ifdef CONFIG_NUMA |
f88dfff5 | 5428 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5429 | #endif |
1da177e4 LT |
5430 | } |
5431 | ||
9420f89d | 5432 | static int zone_batchsize(struct zone *zone) |
1da177e4 | 5433 | { |
9420f89d MRI |
5434 | #ifdef CONFIG_MMU |
5435 | int batch; | |
1da177e4 | 5436 | |
9420f89d MRI |
5437 | /* |
5438 | * The number of pages to batch allocate is either ~0.1% | |
5439 | * of the zone or 1MB, whichever is smaller. The batch | |
5440 | * size is striking a balance between allocation latency | |
5441 | * and zone lock contention. | |
5442 | */ | |
5443 | batch = min(zone_managed_pages(zone) >> 10, SZ_1M / PAGE_SIZE); | |
5444 | batch /= 4; /* We effectively *= 4 below */ | |
5445 | if (batch < 1) | |
5446 | batch = 1; | |
22b31eec | 5447 | |
4b94ffdc | 5448 | /* |
9420f89d MRI |
5449 | * Clamp the batch to a 2^n - 1 value. Having a power |
5450 | * of 2 value was found to be more likely to have | |
5451 | * suboptimal cache aliasing properties in some cases. | |
5452 | * | |
5453 | * For example if 2 tasks are alternately allocating | |
5454 | * batches of pages, one task can end up with a lot | |
5455 | * of pages of one half of the possible page colors | |
5456 | * and the other with pages of the other colors. | |
4b94ffdc | 5457 | */ |
9420f89d | 5458 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
966cf44f | 5459 | |
9420f89d | 5460 | return batch; |
3a6be87f DH |
5461 | |
5462 | #else | |
5463 | /* The deferral and batching of frees should be suppressed under NOMMU | |
5464 | * conditions. | |
5465 | * | |
5466 | * The problem is that NOMMU needs to be able to allocate large chunks | |
5467 | * of contiguous memory as there's no hardware page translation to | |
5468 | * assemble apparent contiguous memory from discontiguous pages. | |
5469 | * | |
5470 | * Queueing large contiguous runs of pages for batching, however, | |
5471 | * causes the pages to actually be freed in smaller chunks. As there | |
5472 | * can be a significant delay between the individual batches being | |
5473 | * recycled, this leads to the once large chunks of space being | |
5474 | * fragmented and becoming unavailable for high-order allocations. | |
5475 | */ | |
5476 | return 0; | |
5477 | #endif | |
e7c8d5c9 CL |
5478 | } |
5479 | ||
e95d372c | 5480 | static int percpu_pagelist_high_fraction; |
90b41691 HY |
5481 | static int zone_highsize(struct zone *zone, int batch, int cpu_online, |
5482 | int high_fraction) | |
b92ca18e | 5483 | { |
9420f89d MRI |
5484 | #ifdef CONFIG_MMU |
5485 | int high; | |
5486 | int nr_split_cpus; | |
5487 | unsigned long total_pages; | |
c13291a5 | 5488 | |
90b41691 | 5489 | if (!high_fraction) { |
2a1e274a | 5490 | /* |
9420f89d MRI |
5491 | * By default, the high value of the pcp is based on the zone |
5492 | * low watermark so that if they are full then background | |
5493 | * reclaim will not be started prematurely. | |
2a1e274a | 5494 | */ |
9420f89d MRI |
5495 | total_pages = low_wmark_pages(zone); |
5496 | } else { | |
2a1e274a | 5497 | /* |
9420f89d MRI |
5498 | * If percpu_pagelist_high_fraction is configured, the high |
5499 | * value is based on a fraction of the managed pages in the | |
5500 | * zone. | |
2a1e274a | 5501 | */ |
90b41691 | 5502 | total_pages = zone_managed_pages(zone) / high_fraction; |
2a1e274a MG |
5503 | } |
5504 | ||
5505 | /* | |
9420f89d MRI |
5506 | * Split the high value across all online CPUs local to the zone. Note |
5507 | * that early in boot that CPUs may not be online yet and that during | |
5508 | * CPU hotplug that the cpumask is not yet updated when a CPU is being | |
90b41691 HY |
5509 | * onlined. For memory nodes that have no CPUs, split the high value |
5510 | * across all online CPUs to mitigate the risk that reclaim is triggered | |
9420f89d | 5511 | * prematurely due to pages stored on pcp lists. |
2a1e274a | 5512 | */ |
9420f89d MRI |
5513 | nr_split_cpus = cpumask_weight(cpumask_of_node(zone_to_nid(zone))) + cpu_online; |
5514 | if (!nr_split_cpus) | |
5515 | nr_split_cpus = num_online_cpus(); | |
5516 | high = total_pages / nr_split_cpus; | |
2a1e274a | 5517 | |
9420f89d MRI |
5518 | /* |
5519 | * Ensure high is at least batch*4. The multiple is based on the | |
5520 | * historical relationship between high and batch. | |
5521 | */ | |
5522 | high = max(high, batch << 2); | |
37b07e41 | 5523 | |
9420f89d MRI |
5524 | return high; |
5525 | #else | |
5526 | return 0; | |
5527 | #endif | |
37b07e41 LS |
5528 | } |
5529 | ||
51930df5 | 5530 | /* |
9420f89d MRI |
5531 | * pcp->high and pcp->batch values are related and generally batch is lower |
5532 | * than high. They are also related to pcp->count such that count is lower | |
5533 | * than high, and as soon as it reaches high, the pcplist is flushed. | |
5534 | * | |
5535 | * However, guaranteeing these relations at all times would require e.g. write | |
5536 | * barriers here but also careful usage of read barriers at the read side, and | |
5537 | * thus be prone to error and bad for performance. Thus the update only prevents | |
90b41691 HY |
5538 | * store tearing. Any new users of pcp->batch, pcp->high_min and pcp->high_max |
5539 | * should ensure they can cope with those fields changing asynchronously, and | |
5540 | * fully trust only the pcp->count field on the local CPU with interrupts | |
5541 | * disabled. | |
9420f89d MRI |
5542 | * |
5543 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
5544 | * outside of boot time (or some other assurance that no concurrent updaters | |
5545 | * exist). | |
51930df5 | 5546 | */ |
90b41691 HY |
5547 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high_min, |
5548 | unsigned long high_max, unsigned long batch) | |
51930df5 | 5549 | { |
9420f89d | 5550 | WRITE_ONCE(pcp->batch, batch); |
90b41691 HY |
5551 | WRITE_ONCE(pcp->high_min, high_min); |
5552 | WRITE_ONCE(pcp->high_max, high_max); | |
51930df5 MR |
5553 | } |
5554 | ||
9420f89d | 5555 | static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats) |
c713216d | 5556 | { |
9420f89d | 5557 | int pindex; |
90cae1fe | 5558 | |
9420f89d MRI |
5559 | memset(pcp, 0, sizeof(*pcp)); |
5560 | memset(pzstats, 0, sizeof(*pzstats)); | |
90cae1fe | 5561 | |
9420f89d MRI |
5562 | spin_lock_init(&pcp->lock); |
5563 | for (pindex = 0; pindex < NR_PCP_LISTS; pindex++) | |
5564 | INIT_LIST_HEAD(&pcp->lists[pindex]); | |
2a1e274a | 5565 | |
9420f89d MRI |
5566 | /* |
5567 | * Set batch and high values safe for a boot pageset. A true percpu | |
5568 | * pageset's initialization will update them subsequently. Here we don't | |
5569 | * need to be as careful as pageset_update() as nobody can access the | |
5570 | * pageset yet. | |
5571 | */ | |
90b41691 HY |
5572 | pcp->high_min = BOOT_PAGESET_HIGH; |
5573 | pcp->high_max = BOOT_PAGESET_HIGH; | |
9420f89d | 5574 | pcp->batch = BOOT_PAGESET_BATCH; |
6ccdcb6d | 5575 | pcp->free_count = 0; |
9420f89d | 5576 | } |
c713216d | 5577 | |
90b41691 HY |
5578 | static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high_min, |
5579 | unsigned long high_max, unsigned long batch) | |
9420f89d MRI |
5580 | { |
5581 | struct per_cpu_pages *pcp; | |
5582 | int cpu; | |
2a1e274a | 5583 | |
9420f89d MRI |
5584 | for_each_possible_cpu(cpu) { |
5585 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); | |
90b41691 | 5586 | pageset_update(pcp, high_min, high_max, batch); |
2a1e274a | 5587 | } |
9420f89d | 5588 | } |
c713216d | 5589 | |
9420f89d MRI |
5590 | /* |
5591 | * Calculate and set new high and batch values for all per-cpu pagesets of a | |
5592 | * zone based on the zone's size. | |
5593 | */ | |
5594 | static void zone_set_pageset_high_and_batch(struct zone *zone, int cpu_online) | |
5595 | { | |
90b41691 | 5596 | int new_high_min, new_high_max, new_batch; |
09f49dca | 5597 | |
9420f89d | 5598 | new_batch = max(1, zone_batchsize(zone)); |
90b41691 HY |
5599 | if (percpu_pagelist_high_fraction) { |
5600 | new_high_min = zone_highsize(zone, new_batch, cpu_online, | |
5601 | percpu_pagelist_high_fraction); | |
5602 | /* | |
5603 | * PCP high is tuned manually, disable auto-tuning via | |
5604 | * setting high_min and high_max to the manual value. | |
5605 | */ | |
5606 | new_high_max = new_high_min; | |
5607 | } else { | |
5608 | new_high_min = zone_highsize(zone, new_batch, cpu_online, 0); | |
5609 | new_high_max = zone_highsize(zone, new_batch, cpu_online, | |
5610 | MIN_PERCPU_PAGELIST_HIGH_FRACTION); | |
5611 | } | |
09f49dca | 5612 | |
90b41691 HY |
5613 | if (zone->pageset_high_min == new_high_min && |
5614 | zone->pageset_high_max == new_high_max && | |
9420f89d MRI |
5615 | zone->pageset_batch == new_batch) |
5616 | return; | |
37b07e41 | 5617 | |
90b41691 HY |
5618 | zone->pageset_high_min = new_high_min; |
5619 | zone->pageset_high_max = new_high_max; | |
9420f89d | 5620 | zone->pageset_batch = new_batch; |
122e093c | 5621 | |
90b41691 HY |
5622 | __zone_set_pageset_high_and_batch(zone, new_high_min, new_high_max, |
5623 | new_batch); | |
c713216d | 5624 | } |
2a1e274a | 5625 | |
9420f89d | 5626 | void __meminit setup_zone_pageset(struct zone *zone) |
2a1e274a | 5627 | { |
9420f89d | 5628 | int cpu; |
2a1e274a | 5629 | |
9420f89d MRI |
5630 | /* Size may be 0 on !SMP && !NUMA */ |
5631 | if (sizeof(struct per_cpu_zonestat) > 0) | |
5632 | zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat); | |
2a1e274a | 5633 | |
9420f89d MRI |
5634 | zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages); |
5635 | for_each_possible_cpu(cpu) { | |
5636 | struct per_cpu_pages *pcp; | |
5637 | struct per_cpu_zonestat *pzstats; | |
2a1e274a | 5638 | |
9420f89d MRI |
5639 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
5640 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); | |
5641 | per_cpu_pages_init(pcp, pzstats); | |
a5c6d650 | 5642 | } |
9420f89d MRI |
5643 | |
5644 | zone_set_pageset_high_and_batch(zone, 0); | |
2a1e274a | 5645 | } |
ed7ed365 | 5646 | |
7e63efef | 5647 | /* |
9420f89d MRI |
5648 | * The zone indicated has a new number of managed_pages; batch sizes and percpu |
5649 | * page high values need to be recalculated. | |
7e63efef | 5650 | */ |
9420f89d | 5651 | static void zone_pcp_update(struct zone *zone, int cpu_online) |
7e63efef | 5652 | { |
9420f89d MRI |
5653 | mutex_lock(&pcp_batch_high_lock); |
5654 | zone_set_pageset_high_and_batch(zone, cpu_online); | |
5655 | mutex_unlock(&pcp_batch_high_lock); | |
7e63efef MG |
5656 | } |
5657 | ||
5cec4eb7 | 5658 | static void zone_pcp_update_cacheinfo(struct zone *zone, unsigned int cpu) |
362d37a1 | 5659 | { |
362d37a1 HY |
5660 | struct per_cpu_pages *pcp; |
5661 | struct cpu_cacheinfo *cci; | |
5662 | ||
5cec4eb7 HY |
5663 | pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); |
5664 | cci = get_cpu_cacheinfo(cpu); | |
5665 | /* | |
5666 | * If data cache slice of CPU is large enough, "pcp->batch" | |
5667 | * pages can be preserved in PCP before draining PCP for | |
5668 | * consecutive high-order pages freeing without allocation. | |
5669 | * This can reduce zone lock contention without hurting | |
5670 | * cache-hot pages sharing. | |
5671 | */ | |
5672 | spin_lock(&pcp->lock); | |
5673 | if ((cci->per_cpu_data_slice_size >> PAGE_SHIFT) > 3 * pcp->batch) | |
5674 | pcp->flags |= PCPF_FREE_HIGH_BATCH; | |
5675 | else | |
5676 | pcp->flags &= ~PCPF_FREE_HIGH_BATCH; | |
5677 | spin_unlock(&pcp->lock); | |
362d37a1 HY |
5678 | } |
5679 | ||
5cec4eb7 | 5680 | void setup_pcp_cacheinfo(unsigned int cpu) |
362d37a1 HY |
5681 | { |
5682 | struct zone *zone; | |
5683 | ||
5684 | for_each_populated_zone(zone) | |
5cec4eb7 | 5685 | zone_pcp_update_cacheinfo(zone, cpu); |
362d37a1 HY |
5686 | } |
5687 | ||
7e63efef | 5688 | /* |
9420f89d MRI |
5689 | * Allocate per cpu pagesets and initialize them. |
5690 | * Before this call only boot pagesets were available. | |
7e63efef | 5691 | */ |
9420f89d | 5692 | void __init setup_per_cpu_pageset(void) |
7e63efef | 5693 | { |
9420f89d MRI |
5694 | struct pglist_data *pgdat; |
5695 | struct zone *zone; | |
5696 | int __maybe_unused cpu; | |
5697 | ||
5698 | for_each_populated_zone(zone) | |
5699 | setup_zone_pageset(zone); | |
5700 | ||
5701 | #ifdef CONFIG_NUMA | |
5702 | /* | |
5703 | * Unpopulated zones continue using the boot pagesets. | |
5704 | * The numa stats for these pagesets need to be reset. | |
5705 | * Otherwise, they will end up skewing the stats of | |
5706 | * the nodes these zones are associated with. | |
5707 | */ | |
5708 | for_each_possible_cpu(cpu) { | |
5709 | struct per_cpu_zonestat *pzstats = &per_cpu(boot_zonestats, cpu); | |
5710 | memset(pzstats->vm_numa_event, 0, | |
5711 | sizeof(pzstats->vm_numa_event)); | |
5712 | } | |
5713 | #endif | |
5714 | ||
5715 | for_each_online_pgdat(pgdat) | |
5716 | pgdat->per_cpu_nodestats = | |
5717 | alloc_percpu(struct per_cpu_nodestat); | |
7e63efef MG |
5718 | } |
5719 | ||
9420f89d MRI |
5720 | __meminit void zone_pcp_init(struct zone *zone) |
5721 | { | |
5722 | /* | |
5723 | * per cpu subsystem is not up at this point. The following code | |
5724 | * relies on the ability of the linker to provide the | |
5725 | * offset of a (static) per cpu variable into the per cpu area. | |
5726 | */ | |
5727 | zone->per_cpu_pageset = &boot_pageset; | |
5728 | zone->per_cpu_zonestats = &boot_zonestats; | |
90b41691 HY |
5729 | zone->pageset_high_min = BOOT_PAGESET_HIGH; |
5730 | zone->pageset_high_max = BOOT_PAGESET_HIGH; | |
9420f89d MRI |
5731 | zone->pageset_batch = BOOT_PAGESET_BATCH; |
5732 | ||
5733 | if (populated_zone(zone)) | |
5734 | pr_debug(" %s zone: %lu pages, LIFO batch:%u\n", zone->name, | |
5735 | zone->present_pages, zone_batchsize(zone)); | |
5736 | } | |
ed7ed365 | 5737 | |
c3d5f5f0 JL |
5738 | void adjust_managed_page_count(struct page *page, long count) |
5739 | { | |
9705bea5 | 5740 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 5741 | totalram_pages_add(count); |
3dcc0571 JL |
5742 | #ifdef CONFIG_HIGHMEM |
5743 | if (PageHighMem(page)) | |
ca79b0c2 | 5744 | totalhigh_pages_add(count); |
3dcc0571 | 5745 | #endif |
c3d5f5f0 | 5746 | } |
3dcc0571 | 5747 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 5748 | |
e5cb113f | 5749 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 5750 | { |
11199692 JL |
5751 | void *pos; |
5752 | unsigned long pages = 0; | |
69afade7 | 5753 | |
11199692 JL |
5754 | start = (void *)PAGE_ALIGN((unsigned long)start); |
5755 | end = (void *)((unsigned long)end & PAGE_MASK); | |
5756 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
5757 | struct page *page = virt_to_page(pos); |
5758 | void *direct_map_addr; | |
5759 | ||
5760 | /* | |
5761 | * 'direct_map_addr' might be different from 'pos' | |
5762 | * because some architectures' virt_to_page() | |
5763 | * work with aliases. Getting the direct map | |
5764 | * address ensures that we get a _writeable_ | |
5765 | * alias for the memset(). | |
5766 | */ | |
5767 | direct_map_addr = page_address(page); | |
c746170d VF |
5768 | /* |
5769 | * Perform a kasan-unchecked memset() since this memory | |
5770 | * has not been initialized. | |
5771 | */ | |
5772 | direct_map_addr = kasan_reset_tag(direct_map_addr); | |
dbe67df4 | 5773 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
5774 | memset(direct_map_addr, poison, PAGE_SIZE); |
5775 | ||
5776 | free_reserved_page(page); | |
69afade7 JL |
5777 | } |
5778 | ||
5779 | if (pages && s) | |
ff7ed9e4 | 5780 | pr_info("Freeing %s memory: %ldK\n", s, K(pages)); |
69afade7 JL |
5781 | |
5782 | return pages; | |
5783 | } | |
5784 | ||
005fd4bb | 5785 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 5786 | { |
04f8cfea | 5787 | struct zone *zone; |
1da177e4 | 5788 | |
005fd4bb | 5789 | lru_add_drain_cpu(cpu); |
96f97c43 | 5790 | mlock_drain_remote(cpu); |
005fd4bb | 5791 | drain_pages(cpu); |
9f8f2172 | 5792 | |
005fd4bb SAS |
5793 | /* |
5794 | * Spill the event counters of the dead processor | |
5795 | * into the current processors event counters. | |
5796 | * This artificially elevates the count of the current | |
5797 | * processor. | |
5798 | */ | |
5799 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 5800 | |
005fd4bb SAS |
5801 | /* |
5802 | * Zero the differential counters of the dead processor | |
5803 | * so that the vm statistics are consistent. | |
5804 | * | |
5805 | * This is only okay since the processor is dead and cannot | |
5806 | * race with what we are doing. | |
5807 | */ | |
5808 | cpu_vm_stats_fold(cpu); | |
04f8cfea MG |
5809 | |
5810 | for_each_populated_zone(zone) | |
5811 | zone_pcp_update(zone, 0); | |
5812 | ||
5813 | return 0; | |
5814 | } | |
5815 | ||
5816 | static int page_alloc_cpu_online(unsigned int cpu) | |
5817 | { | |
5818 | struct zone *zone; | |
5819 | ||
5820 | for_each_populated_zone(zone) | |
5821 | zone_pcp_update(zone, 1); | |
005fd4bb | 5822 | return 0; |
1da177e4 | 5823 | } |
1da177e4 | 5824 | |
c4fbed4b | 5825 | void __init page_alloc_init_cpuhp(void) |
1da177e4 | 5826 | { |
005fd4bb SAS |
5827 | int ret; |
5828 | ||
04f8cfea MG |
5829 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC, |
5830 | "mm/page_alloc:pcp", | |
5831 | page_alloc_cpu_online, | |
005fd4bb SAS |
5832 | page_alloc_cpu_dead); |
5833 | WARN_ON(ret < 0); | |
1da177e4 LT |
5834 | } |
5835 | ||
cb45b0e9 | 5836 | /* |
34b10060 | 5837 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
5838 | * or min_free_kbytes changes. |
5839 | */ | |
5840 | static void calculate_totalreserve_pages(void) | |
5841 | { | |
5842 | struct pglist_data *pgdat; | |
5843 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5844 | enum zone_type i, j; |
cb45b0e9 HA |
5845 | |
5846 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
5847 | |
5848 | pgdat->totalreserve_pages = 0; | |
5849 | ||
cb45b0e9 HA |
5850 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5851 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 5852 | long max = 0; |
9705bea5 | 5853 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
5854 | |
5855 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5856 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5857 | if (zone->lowmem_reserve[j] > max) | |
5858 | max = zone->lowmem_reserve[j]; | |
5859 | } | |
5860 | ||
41858966 MG |
5861 | /* we treat the high watermark as reserved pages. */ |
5862 | max += high_wmark_pages(zone); | |
cb45b0e9 | 5863 | |
3d6357de AK |
5864 | if (max > managed_pages) |
5865 | max = managed_pages; | |
a8d01437 | 5866 | |
281e3726 | 5867 | pgdat->totalreserve_pages += max; |
a8d01437 | 5868 | |
cb45b0e9 HA |
5869 | reserve_pages += max; |
5870 | } | |
5871 | } | |
5872 | totalreserve_pages = reserve_pages; | |
5873 | } | |
5874 | ||
1da177e4 LT |
5875 | /* |
5876 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 5877 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
5878 | * has a correct pages reserved value, so an adequate number of |
5879 | * pages are left in the zone after a successful __alloc_pages(). | |
5880 | */ | |
5881 | static void setup_per_zone_lowmem_reserve(void) | |
5882 | { | |
5883 | struct pglist_data *pgdat; | |
470c61d7 | 5884 | enum zone_type i, j; |
1da177e4 | 5885 | |
ec936fc5 | 5886 | for_each_online_pgdat(pgdat) { |
470c61d7 LS |
5887 | for (i = 0; i < MAX_NR_ZONES - 1; i++) { |
5888 | struct zone *zone = &pgdat->node_zones[i]; | |
5889 | int ratio = sysctl_lowmem_reserve_ratio[i]; | |
5890 | bool clear = !ratio || !zone_managed_pages(zone); | |
5891 | unsigned long managed_pages = 0; | |
5892 | ||
5893 | for (j = i + 1; j < MAX_NR_ZONES; j++) { | |
f7ec1044 | 5894 | struct zone *upper_zone = &pgdat->node_zones[j]; |
96a5c186 | 5895 | bool empty = !zone_managed_pages(upper_zone); |
f7ec1044 LS |
5896 | |
5897 | managed_pages += zone_managed_pages(upper_zone); | |
470c61d7 | 5898 | |
96a5c186 | 5899 | if (clear || empty) |
f7ec1044 LS |
5900 | zone->lowmem_reserve[j] = 0; |
5901 | else | |
470c61d7 | 5902 | zone->lowmem_reserve[j] = managed_pages / ratio; |
1da177e4 LT |
5903 | } |
5904 | } | |
5905 | } | |
cb45b0e9 HA |
5906 | |
5907 | /* update totalreserve_pages */ | |
5908 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5909 | } |
5910 | ||
cfd3da1e | 5911 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5912 | { |
5913 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5914 | unsigned long lowmem_pages = 0; | |
5915 | struct zone *zone; | |
5916 | unsigned long flags; | |
5917 | ||
416ef04f | 5918 | /* Calculate total number of !ZONE_HIGHMEM and !ZONE_MOVABLE pages */ |
1da177e4 | 5919 | for_each_zone(zone) { |
416ef04f | 5920 | if (!is_highmem(zone) && zone_idx(zone) != ZONE_MOVABLE) |
9705bea5 | 5921 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
5922 | } |
5923 | ||
5924 | for_each_zone(zone) { | |
ac924c60 AM |
5925 | u64 tmp; |
5926 | ||
1125b4e3 | 5927 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 5928 | tmp = (u64)pages_min * zone_managed_pages(zone); |
72741db6 | 5929 | tmp = div64_ul(tmp, lowmem_pages); |
416ef04f | 5930 | if (is_highmem(zone) || zone_idx(zone) == ZONE_MOVABLE) { |
1da177e4 | 5931 | /* |
669ed175 | 5932 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
416ef04f | 5933 | * need highmem and movable zones pages, so cap pages_min |
5934 | * to a small value here. | |
669ed175 | 5935 | * |
41858966 | 5936 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 5937 | * deltas control async page reclaim, and so should |
416ef04f | 5938 | * not be capped for highmem and movable zones. |
1da177e4 | 5939 | */ |
90ae8d67 | 5940 | unsigned long min_pages; |
1da177e4 | 5941 | |
9705bea5 | 5942 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 5943 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 5944 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5945 | } else { |
669ed175 NP |
5946 | /* |
5947 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5948 | * proportionate to the zone's size. |
5949 | */ | |
a9214443 | 5950 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5951 | } |
5952 | ||
795ae7a0 JW |
5953 | /* |
5954 | * Set the kswapd watermarks distance according to the | |
5955 | * scale factor in proportion to available memory, but | |
5956 | * ensure a minimum size on small systems. | |
5957 | */ | |
5958 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 5959 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
5960 | watermark_scale_factor, 10000)); |
5961 | ||
aa092591 | 5962 | zone->watermark_boost = 0; |
a9214443 | 5963 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
c574bbe9 HY |
5964 | zone->_watermark[WMARK_HIGH] = low_wmark_pages(zone) + tmp; |
5965 | zone->_watermark[WMARK_PROMO] = high_wmark_pages(zone) + tmp; | |
49f223a9 | 5966 | |
1125b4e3 | 5967 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5968 | } |
cb45b0e9 HA |
5969 | |
5970 | /* update totalreserve_pages */ | |
5971 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5972 | } |
5973 | ||
cfd3da1e MG |
5974 | /** |
5975 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5976 | * or when memory is hot-{added|removed} | |
5977 | * | |
5978 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5979 | * correctly with respect to min_free_kbytes. | |
5980 | */ | |
5981 | void setup_per_zone_wmarks(void) | |
5982 | { | |
b92ca18e | 5983 | struct zone *zone; |
b93e0f32 MH |
5984 | static DEFINE_SPINLOCK(lock); |
5985 | ||
5986 | spin_lock(&lock); | |
cfd3da1e | 5987 | __setup_per_zone_wmarks(); |
b93e0f32 | 5988 | spin_unlock(&lock); |
b92ca18e MG |
5989 | |
5990 | /* | |
5991 | * The watermark size have changed so update the pcpu batch | |
5992 | * and high limits or the limits may be inappropriate. | |
5993 | */ | |
5994 | for_each_zone(zone) | |
04f8cfea | 5995 | zone_pcp_update(zone, 0); |
cfd3da1e MG |
5996 | } |
5997 | ||
1da177e4 LT |
5998 | /* |
5999 | * Initialise min_free_kbytes. | |
6000 | * | |
6001 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 6002 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
6003 | * bandwidth does not increase linearly with machine size. We use |
6004 | * | |
b8af2941 | 6005 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6006 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6007 | * | |
6008 | * which yields | |
6009 | * | |
6010 | * 16MB: 512k | |
6011 | * 32MB: 724k | |
6012 | * 64MB: 1024k | |
6013 | * 128MB: 1448k | |
6014 | * 256MB: 2048k | |
6015 | * 512MB: 2896k | |
6016 | * 1024MB: 4096k | |
6017 | * 2048MB: 5792k | |
6018 | * 4096MB: 8192k | |
6019 | * 8192MB: 11584k | |
6020 | * 16384MB: 16384k | |
6021 | */ | |
bd3400ea | 6022 | void calculate_min_free_kbytes(void) |
1da177e4 LT |
6023 | { |
6024 | unsigned long lowmem_kbytes; | |
5f12733e | 6025 | int new_min_free_kbytes; |
1da177e4 LT |
6026 | |
6027 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6028 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6029 | ||
59d336bd WS |
6030 | if (new_min_free_kbytes > user_min_free_kbytes) |
6031 | min_free_kbytes = clamp(new_min_free_kbytes, 128, 262144); | |
6032 | else | |
5f12733e MH |
6033 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", |
6034 | new_min_free_kbytes, user_min_free_kbytes); | |
59d336bd | 6035 | |
bd3400ea LF |
6036 | } |
6037 | ||
6038 | int __meminit init_per_zone_wmark_min(void) | |
6039 | { | |
6040 | calculate_min_free_kbytes(); | |
bc75d33f | 6041 | setup_per_zone_wmarks(); |
a6cccdc3 | 6042 | refresh_zone_stat_thresholds(); |
1da177e4 | 6043 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
6044 | |
6045 | #ifdef CONFIG_NUMA | |
6046 | setup_min_unmapped_ratio(); | |
6047 | setup_min_slab_ratio(); | |
6048 | #endif | |
6049 | ||
4aab2be0 VB |
6050 | khugepaged_min_free_kbytes_update(); |
6051 | ||
1da177e4 LT |
6052 | return 0; |
6053 | } | |
e08d3fdf | 6054 | postcore_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
6055 | |
6056 | /* | |
b8af2941 | 6057 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6058 | * that we can call two helper functions whenever min_free_kbytes |
6059 | * changes. | |
6060 | */ | |
e95d372c | 6061 | static int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 6062 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6063 | { |
da8c757b HP |
6064 | int rc; |
6065 | ||
6066 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6067 | if (rc) | |
6068 | return rc; | |
6069 | ||
5f12733e MH |
6070 | if (write) { |
6071 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6072 | setup_per_zone_wmarks(); |
5f12733e | 6073 | } |
1da177e4 LT |
6074 | return 0; |
6075 | } | |
6076 | ||
e95d372c | 6077 | static int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 6078 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
6079 | { |
6080 | int rc; | |
6081 | ||
6082 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6083 | if (rc) | |
6084 | return rc; | |
6085 | ||
6086 | if (write) | |
6087 | setup_per_zone_wmarks(); | |
6088 | ||
6089 | return 0; | |
6090 | } | |
6091 | ||
9614634f | 6092 | #ifdef CONFIG_NUMA |
6423aa81 | 6093 | static void setup_min_unmapped_ratio(void) |
9614634f | 6094 | { |
6423aa81 | 6095 | pg_data_t *pgdat; |
9614634f | 6096 | struct zone *zone; |
9614634f | 6097 | |
a5f5f91d | 6098 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 6099 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 6100 | |
9614634f | 6101 | for_each_zone(zone) |
9705bea5 AK |
6102 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
6103 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 6104 | } |
0ff38490 | 6105 | |
6423aa81 | 6106 | |
e95d372c | 6107 | static int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 6108 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 6109 | { |
0ff38490 CL |
6110 | int rc; |
6111 | ||
8d65af78 | 6112 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
6113 | if (rc) |
6114 | return rc; | |
6115 | ||
6423aa81 JK |
6116 | setup_min_unmapped_ratio(); |
6117 | ||
6118 | return 0; | |
6119 | } | |
6120 | ||
6121 | static void setup_min_slab_ratio(void) | |
6122 | { | |
6123 | pg_data_t *pgdat; | |
6124 | struct zone *zone; | |
6125 | ||
a5f5f91d MG |
6126 | for_each_online_pgdat(pgdat) |
6127 | pgdat->min_slab_pages = 0; | |
6128 | ||
0ff38490 | 6129 | for_each_zone(zone) |
9705bea5 AK |
6130 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
6131 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
6132 | } |
6133 | ||
e95d372c | 6134 | static int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 6135 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
6136 | { |
6137 | int rc; | |
6138 | ||
6139 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6140 | if (rc) | |
6141 | return rc; | |
6142 | ||
6143 | setup_min_slab_ratio(); | |
6144 | ||
0ff38490 CL |
6145 | return 0; |
6146 | } | |
9614634f CL |
6147 | #endif |
6148 | ||
1da177e4 LT |
6149 | /* |
6150 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
6151 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
6152 | * whenever sysctl_lowmem_reserve_ratio changes. | |
6153 | * | |
6154 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 6155 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
6156 | * if in function of the boot time zone sizes. |
6157 | */ | |
e95d372c KW |
6158 | static int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, |
6159 | int write, void *buffer, size_t *length, loff_t *ppos) | |
1da177e4 | 6160 | { |
86aaf255 BH |
6161 | int i; |
6162 | ||
8d65af78 | 6163 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
6164 | |
6165 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
6166 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
6167 | sysctl_lowmem_reserve_ratio[i] = 0; | |
6168 | } | |
6169 | ||
1da177e4 LT |
6170 | setup_per_zone_lowmem_reserve(); |
6171 | return 0; | |
6172 | } | |
6173 | ||
8ad4b1fb | 6174 | /* |
74f44822 MG |
6175 | * percpu_pagelist_high_fraction - changes the pcp->high for each zone on each |
6176 | * cpu. It is the fraction of total pages in each zone that a hot per cpu | |
b8af2941 | 6177 | * pagelist can have before it gets flushed back to buddy allocator. |
8ad4b1fb | 6178 | */ |
e95d372c | 6179 | static int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *table, |
74f44822 | 6180 | int write, void *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
6181 | { |
6182 | struct zone *zone; | |
74f44822 | 6183 | int old_percpu_pagelist_high_fraction; |
8ad4b1fb RS |
6184 | int ret; |
6185 | ||
7cd2b0a3 | 6186 | mutex_lock(&pcp_batch_high_lock); |
74f44822 | 6187 | old_percpu_pagelist_high_fraction = percpu_pagelist_high_fraction; |
7cd2b0a3 | 6188 | |
8d65af78 | 6189 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
6190 | if (!write || ret < 0) |
6191 | goto out; | |
6192 | ||
6193 | /* Sanity checking to avoid pcp imbalance */ | |
74f44822 MG |
6194 | if (percpu_pagelist_high_fraction && |
6195 | percpu_pagelist_high_fraction < MIN_PERCPU_PAGELIST_HIGH_FRACTION) { | |
6196 | percpu_pagelist_high_fraction = old_percpu_pagelist_high_fraction; | |
7cd2b0a3 DR |
6197 | ret = -EINVAL; |
6198 | goto out; | |
6199 | } | |
6200 | ||
6201 | /* No change? */ | |
74f44822 | 6202 | if (percpu_pagelist_high_fraction == old_percpu_pagelist_high_fraction) |
7cd2b0a3 | 6203 | goto out; |
c8e251fa | 6204 | |
cb1ef534 | 6205 | for_each_populated_zone(zone) |
74f44822 | 6206 | zone_set_pageset_high_and_batch(zone, 0); |
7cd2b0a3 | 6207 | out: |
c8e251fa | 6208 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6209 | return ret; |
8ad4b1fb RS |
6210 | } |
6211 | ||
e95d372c KW |
6212 | static struct ctl_table page_alloc_sysctl_table[] = { |
6213 | { | |
6214 | .procname = "min_free_kbytes", | |
6215 | .data = &min_free_kbytes, | |
6216 | .maxlen = sizeof(min_free_kbytes), | |
6217 | .mode = 0644, | |
6218 | .proc_handler = min_free_kbytes_sysctl_handler, | |
6219 | .extra1 = SYSCTL_ZERO, | |
6220 | }, | |
6221 | { | |
6222 | .procname = "watermark_boost_factor", | |
6223 | .data = &watermark_boost_factor, | |
6224 | .maxlen = sizeof(watermark_boost_factor), | |
6225 | .mode = 0644, | |
6226 | .proc_handler = proc_dointvec_minmax, | |
6227 | .extra1 = SYSCTL_ZERO, | |
6228 | }, | |
6229 | { | |
6230 | .procname = "watermark_scale_factor", | |
6231 | .data = &watermark_scale_factor, | |
6232 | .maxlen = sizeof(watermark_scale_factor), | |
6233 | .mode = 0644, | |
6234 | .proc_handler = watermark_scale_factor_sysctl_handler, | |
6235 | .extra1 = SYSCTL_ONE, | |
6236 | .extra2 = SYSCTL_THREE_THOUSAND, | |
6237 | }, | |
6238 | { | |
6239 | .procname = "percpu_pagelist_high_fraction", | |
6240 | .data = &percpu_pagelist_high_fraction, | |
6241 | .maxlen = sizeof(percpu_pagelist_high_fraction), | |
6242 | .mode = 0644, | |
6243 | .proc_handler = percpu_pagelist_high_fraction_sysctl_handler, | |
6244 | .extra1 = SYSCTL_ZERO, | |
6245 | }, | |
6246 | { | |
6247 | .procname = "lowmem_reserve_ratio", | |
6248 | .data = &sysctl_lowmem_reserve_ratio, | |
6249 | .maxlen = sizeof(sysctl_lowmem_reserve_ratio), | |
6250 | .mode = 0644, | |
6251 | .proc_handler = lowmem_reserve_ratio_sysctl_handler, | |
6252 | }, | |
6253 | #ifdef CONFIG_NUMA | |
6254 | { | |
6255 | .procname = "numa_zonelist_order", | |
6256 | .data = &numa_zonelist_order, | |
6257 | .maxlen = NUMA_ZONELIST_ORDER_LEN, | |
6258 | .mode = 0644, | |
6259 | .proc_handler = numa_zonelist_order_handler, | |
6260 | }, | |
6261 | { | |
6262 | .procname = "min_unmapped_ratio", | |
6263 | .data = &sysctl_min_unmapped_ratio, | |
6264 | .maxlen = sizeof(sysctl_min_unmapped_ratio), | |
6265 | .mode = 0644, | |
6266 | .proc_handler = sysctl_min_unmapped_ratio_sysctl_handler, | |
6267 | .extra1 = SYSCTL_ZERO, | |
6268 | .extra2 = SYSCTL_ONE_HUNDRED, | |
6269 | }, | |
6270 | { | |
6271 | .procname = "min_slab_ratio", | |
6272 | .data = &sysctl_min_slab_ratio, | |
6273 | .maxlen = sizeof(sysctl_min_slab_ratio), | |
6274 | .mode = 0644, | |
6275 | .proc_handler = sysctl_min_slab_ratio_sysctl_handler, | |
6276 | .extra1 = SYSCTL_ZERO, | |
6277 | .extra2 = SYSCTL_ONE_HUNDRED, | |
6278 | }, | |
6279 | #endif | |
e95d372c KW |
6280 | }; |
6281 | ||
6282 | void __init page_alloc_sysctl_init(void) | |
6283 | { | |
6284 | register_sysctl_init("vm", page_alloc_sysctl_table); | |
6285 | } | |
6286 | ||
8df995f6 | 6287 | #ifdef CONFIG_CONTIG_ALLOC |
a1394bdd MK |
6288 | /* Usage: See admin-guide/dynamic-debug-howto.rst */ |
6289 | static void alloc_contig_dump_pages(struct list_head *page_list) | |
6290 | { | |
6291 | DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure"); | |
6292 | ||
6293 | if (DYNAMIC_DEBUG_BRANCH(descriptor)) { | |
6294 | struct page *page; | |
6295 | ||
6296 | dump_stack(); | |
6297 | list_for_each_entry(page, page_list, lru) | |
6298 | dump_page(page, "migration failure"); | |
6299 | } | |
6300 | } | |
a1394bdd | 6301 | |
c8b36003 RC |
6302 | /* |
6303 | * [start, end) must belong to a single zone. | |
6304 | * @migratetype: using migratetype to filter the type of migration in | |
6305 | * trace_mm_alloc_contig_migrate_range_info. | |
6306 | */ | |
b2c9e2fb | 6307 | int __alloc_contig_migrate_range(struct compact_control *cc, |
c8b36003 RC |
6308 | unsigned long start, unsigned long end, |
6309 | int migratetype) | |
041d3a8c MN |
6310 | { |
6311 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 6312 | unsigned int nr_reclaimed; |
041d3a8c MN |
6313 | unsigned long pfn = start; |
6314 | unsigned int tries = 0; | |
6315 | int ret = 0; | |
8b94e0b8 JK |
6316 | struct migration_target_control mtc = { |
6317 | .nid = zone_to_nid(cc->zone), | |
6318 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
e42dfe4e | 6319 | .reason = MR_CONTIG_RANGE, |
8b94e0b8 | 6320 | }; |
c8b36003 RC |
6321 | struct page *page; |
6322 | unsigned long total_mapped = 0; | |
6323 | unsigned long total_migrated = 0; | |
6324 | unsigned long total_reclaimed = 0; | |
041d3a8c | 6325 | |
361a2a22 | 6326 | lru_cache_disable(); |
041d3a8c | 6327 | |
bb13ffeb | 6328 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6329 | if (fatal_signal_pending(current)) { |
6330 | ret = -EINTR; | |
6331 | break; | |
6332 | } | |
6333 | ||
bb13ffeb MG |
6334 | if (list_empty(&cc->migratepages)) { |
6335 | cc->nr_migratepages = 0; | |
c2ad7a1f OS |
6336 | ret = isolate_migratepages_range(cc, pfn, end); |
6337 | if (ret && ret != -EAGAIN) | |
041d3a8c | 6338 | break; |
c2ad7a1f | 6339 | pfn = cc->migrate_pfn; |
041d3a8c MN |
6340 | tries = 0; |
6341 | } else if (++tries == 5) { | |
c8e28b47 | 6342 | ret = -EBUSY; |
041d3a8c MN |
6343 | break; |
6344 | } | |
6345 | ||
beb51eaa MK |
6346 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6347 | &cc->migratepages); | |
6348 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6349 | |
c8b36003 RC |
6350 | if (trace_mm_alloc_contig_migrate_range_info_enabled()) { |
6351 | total_reclaimed += nr_reclaimed; | |
7115936a DH |
6352 | list_for_each_entry(page, &cc->migratepages, lru) { |
6353 | struct folio *folio = page_folio(page); | |
6354 | ||
6355 | total_mapped += folio_mapped(folio) * | |
6356 | folio_nr_pages(folio); | |
6357 | } | |
c8b36003 RC |
6358 | } |
6359 | ||
8b94e0b8 | 6360 | ret = migrate_pages(&cc->migratepages, alloc_migration_target, |
5ac95884 | 6361 | NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE, NULL); |
c8e28b47 | 6362 | |
c8b36003 RC |
6363 | if (trace_mm_alloc_contig_migrate_range_info_enabled() && !ret) |
6364 | total_migrated += cc->nr_migratepages; | |
6365 | ||
c8e28b47 OS |
6366 | /* |
6367 | * On -ENOMEM, migrate_pages() bails out right away. It is pointless | |
6368 | * to retry again over this error, so do the same here. | |
6369 | */ | |
6370 | if (ret == -ENOMEM) | |
6371 | break; | |
041d3a8c | 6372 | } |
d479960e | 6373 | |
361a2a22 | 6374 | lru_cache_enable(); |
2a6f5124 | 6375 | if (ret < 0) { |
3f913fc5 | 6376 | if (!(cc->gfp_mask & __GFP_NOWARN) && ret == -EBUSY) |
151e084a | 6377 | alloc_contig_dump_pages(&cc->migratepages); |
2a6f5124 | 6378 | putback_movable_pages(&cc->migratepages); |
2a6f5124 | 6379 | } |
c8b36003 RC |
6380 | |
6381 | trace_mm_alloc_contig_migrate_range_info(start, end, migratetype, | |
6382 | total_migrated, | |
6383 | total_reclaimed, | |
6384 | total_mapped); | |
6385 | return (ret < 0) ? ret : 0; | |
041d3a8c MN |
6386 | } |
6387 | ||
6388 | /** | |
6389 | * alloc_contig_range() -- tries to allocate given range of pages | |
6390 | * @start: start PFN to allocate | |
6391 | * @end: one-past-the-last PFN to allocate | |
f0953a1b | 6392 | * @migratetype: migratetype of the underlying pageblocks (either |
0815f3d8 MN |
6393 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
6394 | * in range must have the same migratetype and it must | |
6395 | * be either of the two. | |
ca96b625 | 6396 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c | 6397 | * |
11ac3e87 ZY |
6398 | * The PFN range does not have to be pageblock aligned. The PFN range must |
6399 | * belong to a single zone. | |
041d3a8c | 6400 | * |
2c7452a0 MK |
6401 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
6402 | * pageblocks in the range. Once isolated, the pageblocks should not | |
6403 | * be modified by others. | |
041d3a8c | 6404 | * |
a862f68a | 6405 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
6406 | * pages which PFN is in [start, end) are allocated for the caller and |
6407 | * need to be freed with free_contig_range(). | |
6408 | */ | |
b951aaff | 6409 | int alloc_contig_range_noprof(unsigned long start, unsigned long end, |
ca96b625 | 6410 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 6411 | { |
041d3a8c | 6412 | unsigned long outer_start, outer_end; |
d00181b9 | 6413 | int ret = 0; |
041d3a8c | 6414 | |
bb13ffeb MG |
6415 | struct compact_control cc = { |
6416 | .nr_migratepages = 0, | |
6417 | .order = -1, | |
6418 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 6419 | .mode = MIGRATE_SYNC, |
bb13ffeb | 6420 | .ignore_skip_hint = true, |
2583d671 | 6421 | .no_set_skip_hint = true, |
7dea19f9 | 6422 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 6423 | .alloc_contig = true, |
bb13ffeb MG |
6424 | }; |
6425 | INIT_LIST_HEAD(&cc.migratepages); | |
6426 | ||
041d3a8c MN |
6427 | /* |
6428 | * What we do here is we mark all pageblocks in range as | |
6429 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6430 | * have different sizes, and due to the way page allocator | |
b2c9e2fb | 6431 | * work, start_isolate_page_range() has special handlings for this. |
041d3a8c MN |
6432 | * |
6433 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6434 | * migrate the pages from an unaligned range (ie. pages that | |
b2c9e2fb | 6435 | * we are interested in). This will put all the pages in |
041d3a8c MN |
6436 | * range back to page allocator as MIGRATE_ISOLATE. |
6437 | * | |
6438 | * When this is done, we take the pages in range from page | |
6439 | * allocator removing them from the buddy system. This way | |
6440 | * page allocator will never consider using them. | |
6441 | * | |
6442 | * This lets us mark the pageblocks back as | |
6443 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6444 | * aligned range but not in the unaligned, original range are | |
6445 | * put back to page allocator so that buddy can use them. | |
6446 | */ | |
6447 | ||
6e263fff | 6448 | ret = start_isolate_page_range(start, end, migratetype, 0, gfp_mask); |
3fa0c7c7 | 6449 | if (ret) |
b2c9e2fb | 6450 | goto done; |
041d3a8c | 6451 | |
7612921f VB |
6452 | drain_all_pages(cc.zone); |
6453 | ||
8ef5849f JK |
6454 | /* |
6455 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
6456 | * So, just fall through. test_pages_isolated() has a tracepoint |
6457 | * which will report the busy page. | |
6458 | * | |
6459 | * It is possible that busy pages could become available before | |
6460 | * the call to test_pages_isolated, and the range will actually be | |
6461 | * allocated. So, if we fall through be sure to clear ret so that | |
6462 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 6463 | */ |
c8b36003 | 6464 | ret = __alloc_contig_migrate_range(&cc, start, end, migratetype); |
8ef5849f | 6465 | if (ret && ret != -EBUSY) |
041d3a8c | 6466 | goto done; |
68d68ff6 | 6467 | ret = 0; |
041d3a8c MN |
6468 | |
6469 | /* | |
b2c9e2fb | 6470 | * Pages from [start, end) are within a pageblock_nr_pages |
041d3a8c MN |
6471 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's |
6472 | * more, all pages in [start, end) are free in page allocator. | |
6473 | * What we are going to do is to allocate all pages from | |
6474 | * [start, end) (that is remove them from page allocator). | |
6475 | * | |
6476 | * The only problem is that pages at the beginning and at the | |
6477 | * end of interesting range may be not aligned with pages that | |
6478 | * page allocator holds, ie. they can be part of higher order | |
6479 | * pages. Because of this, we reserve the bigger range and | |
6480 | * once this is done free the pages we are not interested in. | |
6481 | * | |
6482 | * We don't have to hold zone->lock here because the pages are | |
6483 | * isolated thus they won't get removed from buddy. | |
6484 | */ | |
fd919a85 | 6485 | outer_start = find_large_buddy(start); |
8ef5849f | 6486 | |
041d3a8c | 6487 | /* Make sure the range is really isolated. */ |
756d25be | 6488 | if (test_pages_isolated(outer_start, end, 0)) { |
041d3a8c MN |
6489 | ret = -EBUSY; |
6490 | goto done; | |
6491 | } | |
6492 | ||
49f223a9 | 6493 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 6494 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6495 | if (!outer_end) { |
6496 | ret = -EBUSY; | |
6497 | goto done; | |
6498 | } | |
6499 | ||
6500 | /* Free head and tail (if any) */ | |
6501 | if (start != outer_start) | |
6502 | free_contig_range(outer_start, start - outer_start); | |
6503 | if (end != outer_end) | |
6504 | free_contig_range(end, outer_end - end); | |
6505 | ||
6506 | done: | |
6e263fff | 6507 | undo_isolate_page_range(start, end, migratetype); |
041d3a8c MN |
6508 | return ret; |
6509 | } | |
b951aaff | 6510 | EXPORT_SYMBOL(alloc_contig_range_noprof); |
5e27a2df AK |
6511 | |
6512 | static int __alloc_contig_pages(unsigned long start_pfn, | |
6513 | unsigned long nr_pages, gfp_t gfp_mask) | |
6514 | { | |
6515 | unsigned long end_pfn = start_pfn + nr_pages; | |
6516 | ||
b951aaff SB |
6517 | return alloc_contig_range_noprof(start_pfn, end_pfn, MIGRATE_MOVABLE, |
6518 | gfp_mask); | |
5e27a2df AK |
6519 | } |
6520 | ||
6521 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
6522 | unsigned long nr_pages) | |
6523 | { | |
6524 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
6525 | struct page *page; | |
6526 | ||
6527 | for (i = start_pfn; i < end_pfn; i++) { | |
6528 | page = pfn_to_online_page(i); | |
6529 | if (!page) | |
6530 | return false; | |
6531 | ||
6532 | if (page_zone(page) != z) | |
6533 | return false; | |
6534 | ||
6535 | if (PageReserved(page)) | |
4d73ba5f MG |
6536 | return false; |
6537 | ||
6538 | if (PageHuge(page)) | |
5e27a2df | 6539 | return false; |
5e27a2df AK |
6540 | } |
6541 | return true; | |
6542 | } | |
6543 | ||
6544 | static bool zone_spans_last_pfn(const struct zone *zone, | |
6545 | unsigned long start_pfn, unsigned long nr_pages) | |
6546 | { | |
6547 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
6548 | ||
6549 | return zone_spans_pfn(zone, last_pfn); | |
6550 | } | |
6551 | ||
6552 | /** | |
6553 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
6554 | * @nr_pages: Number of contiguous pages to allocate | |
6555 | * @gfp_mask: GFP mask to limit search and used during compaction | |
6556 | * @nid: Target node | |
6557 | * @nodemask: Mask for other possible nodes | |
6558 | * | |
6559 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
6560 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
6561 | * tried for allocation with alloc_contig_range(). This routine is intended | |
6562 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
6563 | * | |
6564 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
eaab8e75 AK |
6565 | * power of two, then allocated range is also guaranteed to be aligned to same |
6566 | * nr_pages (e.g. 1GB request would be aligned to 1GB). | |
5e27a2df AK |
6567 | * |
6568 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
6569 | * __free_page() on each allocated page. | |
6570 | * | |
6571 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
6572 | */ | |
b951aaff SB |
6573 | struct page *alloc_contig_pages_noprof(unsigned long nr_pages, gfp_t gfp_mask, |
6574 | int nid, nodemask_t *nodemask) | |
5e27a2df AK |
6575 | { |
6576 | unsigned long ret, pfn, flags; | |
6577 | struct zonelist *zonelist; | |
6578 | struct zone *zone; | |
6579 | struct zoneref *z; | |
6580 | ||
6581 | zonelist = node_zonelist(nid, gfp_mask); | |
6582 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
6583 | gfp_zone(gfp_mask), nodemask) { | |
6584 | spin_lock_irqsave(&zone->lock, flags); | |
6585 | ||
6586 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
6587 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
6588 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
6589 | /* | |
6590 | * We release the zone lock here because | |
6591 | * alloc_contig_range() will also lock the zone | |
6592 | * at some point. If there's an allocation | |
6593 | * spinning on this lock, it may win the race | |
6594 | * and cause alloc_contig_range() to fail... | |
6595 | */ | |
6596 | spin_unlock_irqrestore(&zone->lock, flags); | |
6597 | ret = __alloc_contig_pages(pfn, nr_pages, | |
6598 | gfp_mask); | |
6599 | if (!ret) | |
6600 | return pfn_to_page(pfn); | |
6601 | spin_lock_irqsave(&zone->lock, flags); | |
6602 | } | |
6603 | pfn += nr_pages; | |
6604 | } | |
6605 | spin_unlock_irqrestore(&zone->lock, flags); | |
6606 | } | |
6607 | return NULL; | |
6608 | } | |
4eb0716e | 6609 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 6610 | |
78fa5150 | 6611 | void free_contig_range(unsigned long pfn, unsigned long nr_pages) |
041d3a8c | 6612 | { |
78fa5150 | 6613 | unsigned long count = 0; |
bcc2b02f MS |
6614 | |
6615 | for (; nr_pages--; pfn++) { | |
6616 | struct page *page = pfn_to_page(pfn); | |
6617 | ||
6618 | count += page_count(page) != 1; | |
6619 | __free_page(page); | |
6620 | } | |
78fa5150 | 6621 | WARN(count != 0, "%lu pages are still in use!\n", count); |
041d3a8c | 6622 | } |
255f5985 | 6623 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 6624 | |
ec6e8c7e VB |
6625 | /* |
6626 | * Effectively disable pcplists for the zone by setting the high limit to 0 | |
6627 | * and draining all cpus. A concurrent page freeing on another CPU that's about | |
6628 | * to put the page on pcplist will either finish before the drain and the page | |
6629 | * will be drained, or observe the new high limit and skip the pcplist. | |
6630 | * | |
6631 | * Must be paired with a call to zone_pcp_enable(). | |
6632 | */ | |
6633 | void zone_pcp_disable(struct zone *zone) | |
6634 | { | |
6635 | mutex_lock(&pcp_batch_high_lock); | |
90b41691 | 6636 | __zone_set_pageset_high_and_batch(zone, 0, 0, 1); |
ec6e8c7e VB |
6637 | __drain_all_pages(zone, true); |
6638 | } | |
6639 | ||
6640 | void zone_pcp_enable(struct zone *zone) | |
6641 | { | |
90b41691 HY |
6642 | __zone_set_pageset_high_and_batch(zone, zone->pageset_high_min, |
6643 | zone->pageset_high_max, zone->pageset_batch); | |
ec6e8c7e VB |
6644 | mutex_unlock(&pcp_batch_high_lock); |
6645 | } | |
6646 | ||
340175b7 JL |
6647 | void zone_pcp_reset(struct zone *zone) |
6648 | { | |
5a883813 | 6649 | int cpu; |
28f836b6 | 6650 | struct per_cpu_zonestat *pzstats; |
340175b7 | 6651 | |
28f836b6 | 6652 | if (zone->per_cpu_pageset != &boot_pageset) { |
5a883813 | 6653 | for_each_online_cpu(cpu) { |
28f836b6 MG |
6654 | pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); |
6655 | drain_zonestat(zone, pzstats); | |
5a883813 | 6656 | } |
28f836b6 | 6657 | free_percpu(zone->per_cpu_pageset); |
28f836b6 | 6658 | zone->per_cpu_pageset = &boot_pageset; |
022e7fa0 ML |
6659 | if (zone->per_cpu_zonestats != &boot_zonestats) { |
6660 | free_percpu(zone->per_cpu_zonestats); | |
6661 | zone->per_cpu_zonestats = &boot_zonestats; | |
6662 | } | |
340175b7 | 6663 | } |
340175b7 JL |
6664 | } |
6665 | ||
6dcd73d7 | 6666 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 6667 | /* |
257bea71 DH |
6668 | * All pages in the range must be in a single zone, must not contain holes, |
6669 | * must span full sections, and must be isolated before calling this function. | |
0c0e6195 | 6670 | */ |
257bea71 | 6671 | void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
0c0e6195 | 6672 | { |
257bea71 | 6673 | unsigned long pfn = start_pfn; |
0c0e6195 KH |
6674 | struct page *page; |
6675 | struct zone *zone; | |
0ee5f4f3 | 6676 | unsigned int order; |
0c0e6195 | 6677 | unsigned long flags; |
5557c766 | 6678 | |
2d070eab | 6679 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
6680 | zone = page_zone(pfn_to_page(pfn)); |
6681 | spin_lock_irqsave(&zone->lock, flags); | |
0c0e6195 | 6682 | while (pfn < end_pfn) { |
0c0e6195 | 6683 | page = pfn_to_page(pfn); |
b023f468 WC |
6684 | /* |
6685 | * The HWPoisoned page may be not in buddy system, and | |
6686 | * page_count() is not 0. | |
6687 | */ | |
6688 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6689 | pfn++; | |
b023f468 WC |
6690 | continue; |
6691 | } | |
aa218795 DH |
6692 | /* |
6693 | * At this point all remaining PageOffline() pages have a | |
6694 | * reference count of 0 and can simply be skipped. | |
6695 | */ | |
6696 | if (PageOffline(page)) { | |
6697 | BUG_ON(page_count(page)); | |
6698 | BUG_ON(PageBuddy(page)); | |
6699 | pfn++; | |
aa218795 DH |
6700 | continue; |
6701 | } | |
b023f468 | 6702 | |
0c0e6195 KH |
6703 | BUG_ON(page_count(page)); |
6704 | BUG_ON(!PageBuddy(page)); | |
e0932b6c | 6705 | VM_WARN_ON(get_pageblock_migratetype(page) != MIGRATE_ISOLATE); |
ab130f91 | 6706 | order = buddy_order(page); |
e0932b6c | 6707 | del_page_from_free_list(page, zone, order, MIGRATE_ISOLATE); |
0c0e6195 KH |
6708 | pfn += (1 << order); |
6709 | } | |
6710 | spin_unlock_irqrestore(&zone->lock, flags); | |
6711 | } | |
6712 | #endif | |
8d22ba1b | 6713 | |
8446b59b ED |
6714 | /* |
6715 | * This function returns a stable result only if called under zone lock. | |
6716 | */ | |
2ace5a67 | 6717 | bool is_free_buddy_page(const struct page *page) |
8d22ba1b | 6718 | { |
8d22ba1b | 6719 | unsigned long pfn = page_to_pfn(page); |
7aeb09f9 | 6720 | unsigned int order; |
8d22ba1b | 6721 | |
fd377218 | 6722 | for (order = 0; order < NR_PAGE_ORDERS; order++) { |
2ace5a67 | 6723 | const struct page *head = page - (pfn & ((1 << order) - 1)); |
8d22ba1b | 6724 | |
2ace5a67 MWO |
6725 | if (PageBuddy(head) && |
6726 | buddy_order_unsafe(head) >= order) | |
8d22ba1b WF |
6727 | break; |
6728 | } | |
8d22ba1b | 6729 | |
5e0a760b | 6730 | return order <= MAX_PAGE_ORDER; |
8d22ba1b | 6731 | } |
a581865e | 6732 | EXPORT_SYMBOL(is_free_buddy_page); |
d4ae9916 NH |
6733 | |
6734 | #ifdef CONFIG_MEMORY_FAILURE | |
e0932b6c JW |
6735 | static inline void add_to_free_list(struct page *page, struct zone *zone, |
6736 | unsigned int order, int migratetype, | |
6737 | bool tail) | |
6738 | { | |
6739 | __add_to_free_list(page, zone, order, migratetype, tail); | |
6740 | account_freepages(zone, 1 << order, migratetype); | |
6741 | } | |
6742 | ||
d4ae9916 | 6743 | /* |
06be6ff3 OS |
6744 | * Break down a higher-order page in sub-pages, and keep our target out of |
6745 | * buddy allocator. | |
d4ae9916 | 6746 | */ |
06be6ff3 OS |
6747 | static void break_down_buddy_pages(struct zone *zone, struct page *page, |
6748 | struct page *target, int low, int high, | |
6749 | int migratetype) | |
6750 | { | |
6751 | unsigned long size = 1 << high; | |
0dfca313 | 6752 | struct page *current_buddy; |
06be6ff3 OS |
6753 | |
6754 | while (high > low) { | |
6755 | high--; | |
6756 | size >>= 1; | |
6757 | ||
6758 | if (target >= &page[size]) { | |
06be6ff3 | 6759 | current_buddy = page; |
0dfca313 | 6760 | page = page + size; |
06be6ff3 | 6761 | } else { |
06be6ff3 OS |
6762 | current_buddy = page + size; |
6763 | } | |
6764 | ||
e0932b6c | 6765 | if (set_page_guard(zone, current_buddy, high)) |
06be6ff3 OS |
6766 | continue; |
6767 | ||
e0932b6c | 6768 | add_to_free_list(current_buddy, zone, high, migratetype, false); |
27e0db3c | 6769 | set_buddy_order(current_buddy, high); |
06be6ff3 OS |
6770 | } |
6771 | } | |
6772 | ||
6773 | /* | |
6774 | * Take a page that will be marked as poisoned off the buddy allocator. | |
6775 | */ | |
6776 | bool take_page_off_buddy(struct page *page) | |
d4ae9916 NH |
6777 | { |
6778 | struct zone *zone = page_zone(page); | |
6779 | unsigned long pfn = page_to_pfn(page); | |
6780 | unsigned long flags; | |
6781 | unsigned int order; | |
06be6ff3 | 6782 | bool ret = false; |
d4ae9916 NH |
6783 | |
6784 | spin_lock_irqsave(&zone->lock, flags); | |
fd377218 | 6785 | for (order = 0; order < NR_PAGE_ORDERS; order++) { |
d4ae9916 | 6786 | struct page *page_head = page - (pfn & ((1 << order) - 1)); |
ab130f91 | 6787 | int page_order = buddy_order(page_head); |
d4ae9916 | 6788 | |
ab130f91 | 6789 | if (PageBuddy(page_head) && page_order >= order) { |
06be6ff3 OS |
6790 | unsigned long pfn_head = page_to_pfn(page_head); |
6791 | int migratetype = get_pfnblock_migratetype(page_head, | |
6792 | pfn_head); | |
6793 | ||
e0932b6c JW |
6794 | del_page_from_free_list(page_head, zone, page_order, |
6795 | migratetype); | |
06be6ff3 | 6796 | break_down_buddy_pages(zone, page_head, page, 0, |
ab130f91 | 6797 | page_order, migratetype); |
bf181c58 | 6798 | SetPageHWPoisonTakenOff(page); |
06be6ff3 | 6799 | ret = true; |
d4ae9916 NH |
6800 | break; |
6801 | } | |
06be6ff3 OS |
6802 | if (page_count(page_head) > 0) |
6803 | break; | |
d4ae9916 NH |
6804 | } |
6805 | spin_unlock_irqrestore(&zone->lock, flags); | |
06be6ff3 | 6806 | return ret; |
d4ae9916 | 6807 | } |
bf181c58 NH |
6808 | |
6809 | /* | |
6810 | * Cancel takeoff done by take_page_off_buddy(). | |
6811 | */ | |
6812 | bool put_page_back_buddy(struct page *page) | |
6813 | { | |
6814 | struct zone *zone = page_zone(page); | |
bf181c58 | 6815 | unsigned long flags; |
bf181c58 NH |
6816 | bool ret = false; |
6817 | ||
6818 | spin_lock_irqsave(&zone->lock, flags); | |
6819 | if (put_page_testzero(page)) { | |
55612e80 JW |
6820 | unsigned long pfn = page_to_pfn(page); |
6821 | int migratetype = get_pfnblock_migratetype(page, pfn); | |
6822 | ||
bf181c58 NH |
6823 | ClearPageHWPoisonTakenOff(page); |
6824 | __free_one_page(page, pfn, zone, 0, migratetype, FPI_NONE); | |
6825 | if (TestClearPageHWPoison(page)) { | |
bf181c58 NH |
6826 | ret = true; |
6827 | } | |
6828 | } | |
6829 | spin_unlock_irqrestore(&zone->lock, flags); | |
6830 | ||
6831 | return ret; | |
6832 | } | |
d4ae9916 | 6833 | #endif |
62b31070 BH |
6834 | |
6835 | #ifdef CONFIG_ZONE_DMA | |
6836 | bool has_managed_dma(void) | |
6837 | { | |
6838 | struct pglist_data *pgdat; | |
6839 | ||
6840 | for_each_online_pgdat(pgdat) { | |
6841 | struct zone *zone = &pgdat->node_zones[ZONE_DMA]; | |
6842 | ||
6843 | if (managed_zone(zone)) | |
6844 | return true; | |
6845 | } | |
6846 | return false; | |
6847 | } | |
6848 | #endif /* CONFIG_ZONE_DMA */ | |
dcdfdd40 KS |
6849 | |
6850 | #ifdef CONFIG_UNACCEPTED_MEMORY | |
6851 | ||
6852 | /* Counts number of zones with unaccepted pages. */ | |
6853 | static DEFINE_STATIC_KEY_FALSE(zones_with_unaccepted_pages); | |
6854 | ||
6855 | static bool lazy_accept = true; | |
6856 | ||
6857 | static int __init accept_memory_parse(char *p) | |
6858 | { | |
6859 | if (!strcmp(p, "lazy")) { | |
6860 | lazy_accept = true; | |
6861 | return 0; | |
6862 | } else if (!strcmp(p, "eager")) { | |
6863 | lazy_accept = false; | |
6864 | return 0; | |
6865 | } else { | |
6866 | return -EINVAL; | |
6867 | } | |
6868 | } | |
6869 | early_param("accept_memory", accept_memory_parse); | |
6870 | ||
6871 | static bool page_contains_unaccepted(struct page *page, unsigned int order) | |
6872 | { | |
6873 | phys_addr_t start = page_to_phys(page); | |
6874 | phys_addr_t end = start + (PAGE_SIZE << order); | |
6875 | ||
6876 | return range_contains_unaccepted_memory(start, end); | |
6877 | } | |
6878 | ||
6879 | static void accept_page(struct page *page, unsigned int order) | |
6880 | { | |
6881 | phys_addr_t start = page_to_phys(page); | |
6882 | ||
6883 | accept_memory(start, start + (PAGE_SIZE << order)); | |
6884 | } | |
6885 | ||
6886 | static bool try_to_accept_memory_one(struct zone *zone) | |
6887 | { | |
6888 | unsigned long flags; | |
6889 | struct page *page; | |
6890 | bool last; | |
6891 | ||
6892 | if (list_empty(&zone->unaccepted_pages)) | |
6893 | return false; | |
6894 | ||
6895 | spin_lock_irqsave(&zone->lock, flags); | |
6896 | page = list_first_entry_or_null(&zone->unaccepted_pages, | |
6897 | struct page, lru); | |
6898 | if (!page) { | |
6899 | spin_unlock_irqrestore(&zone->lock, flags); | |
6900 | return false; | |
6901 | } | |
6902 | ||
6903 | list_del(&page->lru); | |
6904 | last = list_empty(&zone->unaccepted_pages); | |
6905 | ||
e0932b6c | 6906 | account_freepages(zone, -MAX_ORDER_NR_PAGES, MIGRATE_MOVABLE); |
dcdfdd40 KS |
6907 | __mod_zone_page_state(zone, NR_UNACCEPTED, -MAX_ORDER_NR_PAGES); |
6908 | spin_unlock_irqrestore(&zone->lock, flags); | |
6909 | ||
5e0a760b | 6910 | accept_page(page, MAX_PAGE_ORDER); |
dcdfdd40 | 6911 | |
5e0a760b | 6912 | __free_pages_ok(page, MAX_PAGE_ORDER, FPI_TO_TAIL); |
dcdfdd40 KS |
6913 | |
6914 | if (last) | |
6915 | static_branch_dec(&zones_with_unaccepted_pages); | |
6916 | ||
6917 | return true; | |
6918 | } | |
6919 | ||
6920 | static bool try_to_accept_memory(struct zone *zone, unsigned int order) | |
6921 | { | |
6922 | long to_accept; | |
6923 | int ret = false; | |
6924 | ||
6925 | /* How much to accept to get to high watermark? */ | |
6926 | to_accept = high_wmark_pages(zone) - | |
6927 | (zone_page_state(zone, NR_FREE_PAGES) - | |
6928 | __zone_watermark_unusable_free(zone, order, 0)); | |
6929 | ||
6930 | /* Accept at least one page */ | |
6931 | do { | |
6932 | if (!try_to_accept_memory_one(zone)) | |
6933 | break; | |
6934 | ret = true; | |
6935 | to_accept -= MAX_ORDER_NR_PAGES; | |
6936 | } while (to_accept > 0); | |
6937 | ||
6938 | return ret; | |
6939 | } | |
6940 | ||
6941 | static inline bool has_unaccepted_memory(void) | |
6942 | { | |
6943 | return static_branch_unlikely(&zones_with_unaccepted_pages); | |
6944 | } | |
6945 | ||
6946 | static bool __free_unaccepted(struct page *page) | |
6947 | { | |
6948 | struct zone *zone = page_zone(page); | |
6949 | unsigned long flags; | |
6950 | bool first = false; | |
6951 | ||
6952 | if (!lazy_accept) | |
6953 | return false; | |
6954 | ||
6955 | spin_lock_irqsave(&zone->lock, flags); | |
6956 | first = list_empty(&zone->unaccepted_pages); | |
6957 | list_add_tail(&page->lru, &zone->unaccepted_pages); | |
e0932b6c | 6958 | account_freepages(zone, MAX_ORDER_NR_PAGES, MIGRATE_MOVABLE); |
dcdfdd40 KS |
6959 | __mod_zone_page_state(zone, NR_UNACCEPTED, MAX_ORDER_NR_PAGES); |
6960 | spin_unlock_irqrestore(&zone->lock, flags); | |
6961 | ||
6962 | if (first) | |
6963 | static_branch_inc(&zones_with_unaccepted_pages); | |
6964 | ||
6965 | return true; | |
6966 | } | |
6967 | ||
6968 | #else | |
6969 | ||
6970 | static bool page_contains_unaccepted(struct page *page, unsigned int order) | |
6971 | { | |
6972 | return false; | |
6973 | } | |
6974 | ||
6975 | static void accept_page(struct page *page, unsigned int order) | |
6976 | { | |
6977 | } | |
6978 | ||
6979 | static bool try_to_accept_memory(struct zone *zone, unsigned int order) | |
6980 | { | |
6981 | return false; | |
6982 | } | |
6983 | ||
6984 | static inline bool has_unaccepted_memory(void) | |
6985 | { | |
6986 | return false; | |
6987 | } | |
6988 | ||
6989 | static bool __free_unaccepted(struct page *page) | |
6990 | { | |
6991 | BUILD_BUG(); | |
6992 | return false; | |
6993 | } | |
6994 | ||
6995 | #endif /* CONFIG_UNACCEPTED_MEMORY */ |