Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/page_alloc.c | |
4 | * | |
5 | * Manages the free list, the system allocates free pages here. | |
6 | * Note that kmalloc() lives in slab.c | |
7 | * | |
8 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
9 | * Swap reorganised 29.12.95, Stephen Tweedie | |
10 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
11 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
12 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
13 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
14 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
15 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
16 | */ | |
17 | ||
1da177e4 LT |
18 | #include <linux/stddef.h> |
19 | #include <linux/mm.h> | |
ca79b0c2 | 20 | #include <linux/highmem.h> |
1da177e4 LT |
21 | #include <linux/swap.h> |
22 | #include <linux/interrupt.h> | |
23 | #include <linux/pagemap.h> | |
10ed273f | 24 | #include <linux/jiffies.h> |
edbe7d23 | 25 | #include <linux/memblock.h> |
1da177e4 | 26 | #include <linux/compiler.h> |
9f158333 | 27 | #include <linux/kernel.h> |
b8c73fc2 | 28 | #include <linux/kasan.h> |
1da177e4 LT |
29 | #include <linux/module.h> |
30 | #include <linux/suspend.h> | |
31 | #include <linux/pagevec.h> | |
32 | #include <linux/blkdev.h> | |
33 | #include <linux/slab.h> | |
a238ab5b | 34 | #include <linux/ratelimit.h> |
5a3135c2 | 35 | #include <linux/oom.h> |
1da177e4 LT |
36 | #include <linux/topology.h> |
37 | #include <linux/sysctl.h> | |
38 | #include <linux/cpu.h> | |
39 | #include <linux/cpuset.h> | |
bdc8cb98 | 40 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
41 | #include <linux/nodemask.h> |
42 | #include <linux/vmalloc.h> | |
a6cccdc3 | 43 | #include <linux/vmstat.h> |
4be38e35 | 44 | #include <linux/mempolicy.h> |
4b94ffdc | 45 | #include <linux/memremap.h> |
6811378e | 46 | #include <linux/stop_machine.h> |
97500a4a | 47 | #include <linux/random.h> |
c713216d MG |
48 | #include <linux/sort.h> |
49 | #include <linux/pfn.h> | |
3fcfab16 | 50 | #include <linux/backing-dev.h> |
933e312e | 51 | #include <linux/fault-inject.h> |
a5d76b54 | 52 | #include <linux/page-isolation.h> |
3ac7fe5a | 53 | #include <linux/debugobjects.h> |
dbb1f81c | 54 | #include <linux/kmemleak.h> |
56de7263 | 55 | #include <linux/compaction.h> |
0d3d062a | 56 | #include <trace/events/kmem.h> |
d379f01d | 57 | #include <trace/events/oom.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
041d3a8c | 60 | #include <linux/migrate.h> |
949f7ec5 | 61 | #include <linux/hugetlb.h> |
8bd75c77 | 62 | #include <linux/sched/rt.h> |
5b3cc15a | 63 | #include <linux/sched/mm.h> |
48c96a36 | 64 | #include <linux/page_owner.h> |
0e1cc95b | 65 | #include <linux/kthread.h> |
4949148a | 66 | #include <linux/memcontrol.h> |
42c269c8 | 67 | #include <linux/ftrace.h> |
d92a8cfc | 68 | #include <linux/lockdep.h> |
556b969a | 69 | #include <linux/nmi.h> |
eb414681 | 70 | #include <linux/psi.h> |
1da177e4 | 71 | |
7ee3d4e8 | 72 | #include <asm/sections.h> |
1da177e4 | 73 | #include <asm/tlbflush.h> |
ac924c60 | 74 | #include <asm/div64.h> |
1da177e4 | 75 | #include "internal.h" |
e900a918 | 76 | #include "shuffle.h" |
1da177e4 | 77 | |
c8e251fa CS |
78 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
79 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 80 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 81 | |
72812019 LS |
82 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
83 | DEFINE_PER_CPU(int, numa_node); | |
84 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
85 | #endif | |
86 | ||
4518085e KW |
87 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
88 | ||
7aac7898 LS |
89 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
90 | /* | |
91 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
92 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
93 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
94 | * defined in <linux/topology.h>. | |
95 | */ | |
96 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
97 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 98 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
99 | #endif |
100 | ||
bd233f53 | 101 | /* work_structs for global per-cpu drains */ |
d9367bd0 WY |
102 | struct pcpu_drain { |
103 | struct zone *zone; | |
104 | struct work_struct work; | |
105 | }; | |
bd233f53 | 106 | DEFINE_MUTEX(pcpu_drain_mutex); |
d9367bd0 | 107 | DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain); |
bd233f53 | 108 | |
38addce8 | 109 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 110 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
111 | EXPORT_SYMBOL(latent_entropy); |
112 | #endif | |
113 | ||
1da177e4 | 114 | /* |
13808910 | 115 | * Array of node states. |
1da177e4 | 116 | */ |
13808910 CL |
117 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
118 | [N_POSSIBLE] = NODE_MASK_ALL, | |
119 | [N_ONLINE] = { { [0] = 1UL } }, | |
120 | #ifndef CONFIG_NUMA | |
121 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
122 | #ifdef CONFIG_HIGHMEM | |
123 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 124 | #endif |
20b2f52b | 125 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
126 | [N_CPU] = { { [0] = 1UL } }, |
127 | #endif /* NUMA */ | |
128 | }; | |
129 | EXPORT_SYMBOL(node_states); | |
130 | ||
ca79b0c2 AK |
131 | atomic_long_t _totalram_pages __read_mostly; |
132 | EXPORT_SYMBOL(_totalram_pages); | |
cb45b0e9 | 133 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 134 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 135 | |
1b76b02f | 136 | int percpu_pagelist_fraction; |
dcce284a | 137 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
6471384a AP |
138 | #ifdef CONFIG_INIT_ON_ALLOC_DEFAULT_ON |
139 | DEFINE_STATIC_KEY_TRUE(init_on_alloc); | |
140 | #else | |
141 | DEFINE_STATIC_KEY_FALSE(init_on_alloc); | |
142 | #endif | |
143 | EXPORT_SYMBOL(init_on_alloc); | |
144 | ||
145 | #ifdef CONFIG_INIT_ON_FREE_DEFAULT_ON | |
146 | DEFINE_STATIC_KEY_TRUE(init_on_free); | |
147 | #else | |
148 | DEFINE_STATIC_KEY_FALSE(init_on_free); | |
149 | #endif | |
150 | EXPORT_SYMBOL(init_on_free); | |
151 | ||
152 | static int __init early_init_on_alloc(char *buf) | |
153 | { | |
154 | int ret; | |
155 | bool bool_result; | |
156 | ||
157 | if (!buf) | |
158 | return -EINVAL; | |
159 | ret = kstrtobool(buf, &bool_result); | |
160 | if (bool_result && page_poisoning_enabled()) | |
161 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_alloc\n"); | |
162 | if (bool_result) | |
163 | static_branch_enable(&init_on_alloc); | |
164 | else | |
165 | static_branch_disable(&init_on_alloc); | |
166 | return ret; | |
167 | } | |
168 | early_param("init_on_alloc", early_init_on_alloc); | |
169 | ||
170 | static int __init early_init_on_free(char *buf) | |
171 | { | |
172 | int ret; | |
173 | bool bool_result; | |
174 | ||
175 | if (!buf) | |
176 | return -EINVAL; | |
177 | ret = kstrtobool(buf, &bool_result); | |
178 | if (bool_result && page_poisoning_enabled()) | |
179 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_free\n"); | |
180 | if (bool_result) | |
181 | static_branch_enable(&init_on_free); | |
182 | else | |
183 | static_branch_disable(&init_on_free); | |
184 | return ret; | |
185 | } | |
186 | early_param("init_on_free", early_init_on_free); | |
1da177e4 | 187 | |
bb14c2c7 VB |
188 | /* |
189 | * A cached value of the page's pageblock's migratetype, used when the page is | |
190 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
191 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
192 | * Also the migratetype set in the page does not necessarily match the pcplist | |
193 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
194 | * other index - this ensures that it will be put on the correct CMA freelist. | |
195 | */ | |
196 | static inline int get_pcppage_migratetype(struct page *page) | |
197 | { | |
198 | return page->index; | |
199 | } | |
200 | ||
201 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
202 | { | |
203 | page->index = migratetype; | |
204 | } | |
205 | ||
452aa699 RW |
206 | #ifdef CONFIG_PM_SLEEP |
207 | /* | |
208 | * The following functions are used by the suspend/hibernate code to temporarily | |
209 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
210 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
211 | * they should always be called with system_transition_mutex held |
212 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
213 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
214 | * with that modification). | |
452aa699 | 215 | */ |
c9e664f1 RW |
216 | |
217 | static gfp_t saved_gfp_mask; | |
218 | ||
219 | void pm_restore_gfp_mask(void) | |
452aa699 | 220 | { |
55f2503c | 221 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
222 | if (saved_gfp_mask) { |
223 | gfp_allowed_mask = saved_gfp_mask; | |
224 | saved_gfp_mask = 0; | |
225 | } | |
452aa699 RW |
226 | } |
227 | ||
c9e664f1 | 228 | void pm_restrict_gfp_mask(void) |
452aa699 | 229 | { |
55f2503c | 230 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
231 | WARN_ON(saved_gfp_mask); |
232 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 233 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 234 | } |
f90ac398 MG |
235 | |
236 | bool pm_suspended_storage(void) | |
237 | { | |
d0164adc | 238 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
239 | return false; |
240 | return true; | |
241 | } | |
452aa699 RW |
242 | #endif /* CONFIG_PM_SLEEP */ |
243 | ||
d9c23400 | 244 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 245 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
246 | #endif |
247 | ||
d98c7a09 | 248 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 249 | |
1da177e4 LT |
250 | /* |
251 | * results with 256, 32 in the lowmem_reserve sysctl: | |
252 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
253 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
254 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
255 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 256 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
257 | * |
258 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
259 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 260 | */ |
d3cda233 | 261 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 262 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 263 | [ZONE_DMA] = 256, |
4b51d669 | 264 | #endif |
fb0e7942 | 265 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 266 | [ZONE_DMA32] = 256, |
fb0e7942 | 267 | #endif |
d3cda233 | 268 | [ZONE_NORMAL] = 32, |
e53ef38d | 269 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 270 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 271 | #endif |
d3cda233 | 272 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 273 | }; |
1da177e4 | 274 | |
15ad7cdc | 275 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 276 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 277 | "DMA", |
4b51d669 | 278 | #endif |
fb0e7942 | 279 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 280 | "DMA32", |
fb0e7942 | 281 | #endif |
2f1b6248 | 282 | "Normal", |
e53ef38d | 283 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 284 | "HighMem", |
e53ef38d | 285 | #endif |
2a1e274a | 286 | "Movable", |
033fbae9 DW |
287 | #ifdef CONFIG_ZONE_DEVICE |
288 | "Device", | |
289 | #endif | |
2f1b6248 CL |
290 | }; |
291 | ||
c999fbd3 | 292 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
293 | "Unmovable", |
294 | "Movable", | |
295 | "Reclaimable", | |
296 | "HighAtomic", | |
297 | #ifdef CONFIG_CMA | |
298 | "CMA", | |
299 | #endif | |
300 | #ifdef CONFIG_MEMORY_ISOLATION | |
301 | "Isolate", | |
302 | #endif | |
303 | }; | |
304 | ||
f1e61557 KS |
305 | compound_page_dtor * const compound_page_dtors[] = { |
306 | NULL, | |
307 | free_compound_page, | |
308 | #ifdef CONFIG_HUGETLB_PAGE | |
309 | free_huge_page, | |
310 | #endif | |
9a982250 KS |
311 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
312 | free_transhuge_page, | |
313 | #endif | |
f1e61557 KS |
314 | }; |
315 | ||
1da177e4 | 316 | int min_free_kbytes = 1024; |
42aa83cb | 317 | int user_min_free_kbytes = -1; |
24512228 MG |
318 | #ifdef CONFIG_DISCONTIGMEM |
319 | /* | |
320 | * DiscontigMem defines memory ranges as separate pg_data_t even if the ranges | |
321 | * are not on separate NUMA nodes. Functionally this works but with | |
322 | * watermark_boost_factor, it can reclaim prematurely as the ranges can be | |
323 | * quite small. By default, do not boost watermarks on discontigmem as in | |
324 | * many cases very high-order allocations like THP are likely to be | |
325 | * unsupported and the premature reclaim offsets the advantage of long-term | |
326 | * fragmentation avoidance. | |
327 | */ | |
328 | int watermark_boost_factor __read_mostly; | |
329 | #else | |
1c30844d | 330 | int watermark_boost_factor __read_mostly = 15000; |
24512228 | 331 | #endif |
795ae7a0 | 332 | int watermark_scale_factor = 10; |
1da177e4 | 333 | |
bbe5d993 OS |
334 | static unsigned long nr_kernel_pages __initdata; |
335 | static unsigned long nr_all_pages __initdata; | |
336 | static unsigned long dma_reserve __initdata; | |
1da177e4 | 337 | |
0ee332c1 | 338 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
bbe5d993 OS |
339 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata; |
340 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata; | |
7f16f91f | 341 | static unsigned long required_kernelcore __initdata; |
a5c6d650 | 342 | static unsigned long required_kernelcore_percent __initdata; |
7f16f91f | 343 | static unsigned long required_movablecore __initdata; |
a5c6d650 | 344 | static unsigned long required_movablecore_percent __initdata; |
bbe5d993 | 345 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata; |
7f16f91f | 346 | static bool mirrored_kernelcore __meminitdata; |
0ee332c1 TH |
347 | |
348 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
349 | int movable_zone; | |
350 | EXPORT_SYMBOL(movable_zone); | |
351 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 352 | |
418508c1 | 353 | #if MAX_NUMNODES > 1 |
b9726c26 | 354 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 355 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 356 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 357 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
358 | #endif |
359 | ||
9ef9acb0 MG |
360 | int page_group_by_mobility_disabled __read_mostly; |
361 | ||
3a80a7fa | 362 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
363 | /* |
364 | * During boot we initialize deferred pages on-demand, as needed, but once | |
365 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
366 | * and we can permanently disable that path. | |
367 | */ | |
368 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
369 | ||
370 | /* | |
371 | * Calling kasan_free_pages() only after deferred memory initialization | |
372 | * has completed. Poisoning pages during deferred memory init will greatly | |
373 | * lengthen the process and cause problem in large memory systems as the | |
374 | * deferred pages initialization is done with interrupt disabled. | |
375 | * | |
376 | * Assuming that there will be no reference to those newly initialized | |
377 | * pages before they are ever allocated, this should have no effect on | |
378 | * KASAN memory tracking as the poison will be properly inserted at page | |
379 | * allocation time. The only corner case is when pages are allocated by | |
380 | * on-demand allocation and then freed again before the deferred pages | |
381 | * initialization is done, but this is not likely to happen. | |
382 | */ | |
383 | static inline void kasan_free_nondeferred_pages(struct page *page, int order) | |
384 | { | |
385 | if (!static_branch_unlikely(&deferred_pages)) | |
386 | kasan_free_pages(page, order); | |
387 | } | |
388 | ||
3a80a7fa | 389 | /* Returns true if the struct page for the pfn is uninitialised */ |
0e1cc95b | 390 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 391 | { |
ef70b6f4 MG |
392 | int nid = early_pfn_to_nid(pfn); |
393 | ||
394 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
395 | return true; |
396 | ||
397 | return false; | |
398 | } | |
399 | ||
400 | /* | |
d3035be4 | 401 | * Returns true when the remaining initialisation should be deferred until |
3a80a7fa MG |
402 | * later in the boot cycle when it can be parallelised. |
403 | */ | |
d3035be4 PT |
404 | static bool __meminit |
405 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn) | |
3a80a7fa | 406 | { |
d3035be4 PT |
407 | static unsigned long prev_end_pfn, nr_initialised; |
408 | ||
409 | /* | |
410 | * prev_end_pfn static that contains the end of previous zone | |
411 | * No need to protect because called very early in boot before smp_init. | |
412 | */ | |
413 | if (prev_end_pfn != end_pfn) { | |
414 | prev_end_pfn = end_pfn; | |
415 | nr_initialised = 0; | |
416 | } | |
417 | ||
3c2c6488 | 418 | /* Always populate low zones for address-constrained allocations */ |
d3035be4 | 419 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid))) |
3a80a7fa | 420 | return false; |
23b68cfa WY |
421 | |
422 | /* | |
423 | * We start only with one section of pages, more pages are added as | |
424 | * needed until the rest of deferred pages are initialized. | |
425 | */ | |
d3035be4 | 426 | nr_initialised++; |
23b68cfa | 427 | if ((nr_initialised > PAGES_PER_SECTION) && |
d3035be4 PT |
428 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
429 | NODE_DATA(nid)->first_deferred_pfn = pfn; | |
430 | return true; | |
3a80a7fa | 431 | } |
d3035be4 | 432 | return false; |
3a80a7fa MG |
433 | } |
434 | #else | |
3c0c12cc WL |
435 | #define kasan_free_nondeferred_pages(p, o) kasan_free_pages(p, o) |
436 | ||
3a80a7fa MG |
437 | static inline bool early_page_uninitialised(unsigned long pfn) |
438 | { | |
439 | return false; | |
440 | } | |
441 | ||
d3035be4 | 442 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn) |
3a80a7fa | 443 | { |
d3035be4 | 444 | return false; |
3a80a7fa MG |
445 | } |
446 | #endif | |
447 | ||
0b423ca2 MG |
448 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
449 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
450 | unsigned long pfn) | |
451 | { | |
452 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 453 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
454 | #else |
455 | return page_zone(page)->pageblock_flags; | |
456 | #endif /* CONFIG_SPARSEMEM */ | |
457 | } | |
458 | ||
459 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
460 | { | |
461 | #ifdef CONFIG_SPARSEMEM | |
462 | pfn &= (PAGES_PER_SECTION-1); | |
463 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
464 | #else | |
465 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
466 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
467 | #endif /* CONFIG_SPARSEMEM */ | |
468 | } | |
469 | ||
470 | /** | |
471 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
472 | * @page: The page within the block of interest | |
473 | * @pfn: The target page frame number | |
474 | * @end_bitidx: The last bit of interest to retrieve | |
475 | * @mask: mask of bits that the caller is interested in | |
476 | * | |
477 | * Return: pageblock_bits flags | |
478 | */ | |
479 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
480 | unsigned long pfn, | |
481 | unsigned long end_bitidx, | |
482 | unsigned long mask) | |
483 | { | |
484 | unsigned long *bitmap; | |
485 | unsigned long bitidx, word_bitidx; | |
486 | unsigned long word; | |
487 | ||
488 | bitmap = get_pageblock_bitmap(page, pfn); | |
489 | bitidx = pfn_to_bitidx(page, pfn); | |
490 | word_bitidx = bitidx / BITS_PER_LONG; | |
491 | bitidx &= (BITS_PER_LONG-1); | |
492 | ||
493 | word = bitmap[word_bitidx]; | |
494 | bitidx += end_bitidx; | |
495 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
496 | } | |
497 | ||
498 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
499 | unsigned long end_bitidx, | |
500 | unsigned long mask) | |
501 | { | |
502 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
503 | } | |
504 | ||
505 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
506 | { | |
507 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
508 | } | |
509 | ||
510 | /** | |
511 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
512 | * @page: The page within the block of interest | |
513 | * @flags: The flags to set | |
514 | * @pfn: The target page frame number | |
515 | * @end_bitidx: The last bit of interest | |
516 | * @mask: mask of bits that the caller is interested in | |
517 | */ | |
518 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
519 | unsigned long pfn, | |
520 | unsigned long end_bitidx, | |
521 | unsigned long mask) | |
522 | { | |
523 | unsigned long *bitmap; | |
524 | unsigned long bitidx, word_bitidx; | |
525 | unsigned long old_word, word; | |
526 | ||
527 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 528 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
529 | |
530 | bitmap = get_pageblock_bitmap(page, pfn); | |
531 | bitidx = pfn_to_bitidx(page, pfn); | |
532 | word_bitidx = bitidx / BITS_PER_LONG; | |
533 | bitidx &= (BITS_PER_LONG-1); | |
534 | ||
535 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
536 | ||
537 | bitidx += end_bitidx; | |
538 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
539 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
540 | ||
541 | word = READ_ONCE(bitmap[word_bitidx]); | |
542 | for (;;) { | |
543 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
544 | if (word == old_word) | |
545 | break; | |
546 | word = old_word; | |
547 | } | |
548 | } | |
3a80a7fa | 549 | |
ee6f509c | 550 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 551 | { |
5d0f3f72 KM |
552 | if (unlikely(page_group_by_mobility_disabled && |
553 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
554 | migratetype = MIGRATE_UNMOVABLE; |
555 | ||
b2a0ac88 MG |
556 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
557 | PB_migrate, PB_migrate_end); | |
558 | } | |
559 | ||
13e7444b | 560 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 561 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 562 | { |
bdc8cb98 DH |
563 | int ret = 0; |
564 | unsigned seq; | |
565 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 566 | unsigned long sp, start_pfn; |
c6a57e19 | 567 | |
bdc8cb98 DH |
568 | do { |
569 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
570 | start_pfn = zone->zone_start_pfn; |
571 | sp = zone->spanned_pages; | |
108bcc96 | 572 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
573 | ret = 1; |
574 | } while (zone_span_seqretry(zone, seq)); | |
575 | ||
b5e6a5a2 | 576 | if (ret) |
613813e8 DH |
577 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
578 | pfn, zone_to_nid(zone), zone->name, | |
579 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 580 | |
bdc8cb98 | 581 | return ret; |
c6a57e19 DH |
582 | } |
583 | ||
584 | static int page_is_consistent(struct zone *zone, struct page *page) | |
585 | { | |
14e07298 | 586 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 587 | return 0; |
1da177e4 | 588 | if (zone != page_zone(page)) |
c6a57e19 DH |
589 | return 0; |
590 | ||
591 | return 1; | |
592 | } | |
593 | /* | |
594 | * Temporary debugging check for pages not lying within a given zone. | |
595 | */ | |
d73d3c9f | 596 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
597 | { |
598 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 599 | return 1; |
c6a57e19 DH |
600 | if (!page_is_consistent(zone, page)) |
601 | return 1; | |
602 | ||
1da177e4 LT |
603 | return 0; |
604 | } | |
13e7444b | 605 | #else |
d73d3c9f | 606 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
607 | { |
608 | return 0; | |
609 | } | |
610 | #endif | |
611 | ||
d230dec1 KS |
612 | static void bad_page(struct page *page, const char *reason, |
613 | unsigned long bad_flags) | |
1da177e4 | 614 | { |
d936cf9b HD |
615 | static unsigned long resume; |
616 | static unsigned long nr_shown; | |
617 | static unsigned long nr_unshown; | |
618 | ||
619 | /* | |
620 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
621 | * or allow a steady drip of one report per second. | |
622 | */ | |
623 | if (nr_shown == 60) { | |
624 | if (time_before(jiffies, resume)) { | |
625 | nr_unshown++; | |
626 | goto out; | |
627 | } | |
628 | if (nr_unshown) { | |
ff8e8116 | 629 | pr_alert( |
1e9e6365 | 630 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
631 | nr_unshown); |
632 | nr_unshown = 0; | |
633 | } | |
634 | nr_shown = 0; | |
635 | } | |
636 | if (nr_shown++ == 0) | |
637 | resume = jiffies + 60 * HZ; | |
638 | ||
ff8e8116 | 639 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 640 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
641 | __dump_page(page, reason); |
642 | bad_flags &= page->flags; | |
643 | if (bad_flags) | |
644 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
645 | bad_flags, &bad_flags); | |
4e462112 | 646 | dump_page_owner(page); |
3dc14741 | 647 | |
4f31888c | 648 | print_modules(); |
1da177e4 | 649 | dump_stack(); |
d936cf9b | 650 | out: |
8cc3b392 | 651 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 652 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 653 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
654 | } |
655 | ||
1da177e4 LT |
656 | /* |
657 | * Higher-order pages are called "compound pages". They are structured thusly: | |
658 | * | |
1d798ca3 | 659 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 660 | * |
1d798ca3 KS |
661 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
662 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 663 | * |
1d798ca3 KS |
664 | * The first tail page's ->compound_dtor holds the offset in array of compound |
665 | * page destructors. See compound_page_dtors. | |
1da177e4 | 666 | * |
1d798ca3 | 667 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 668 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 669 | */ |
d98c7a09 | 670 | |
9a982250 | 671 | void free_compound_page(struct page *page) |
d98c7a09 | 672 | { |
7ae88534 | 673 | mem_cgroup_uncharge(page); |
d85f3385 | 674 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
675 | } |
676 | ||
d00181b9 | 677 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
678 | { |
679 | int i; | |
680 | int nr_pages = 1 << order; | |
681 | ||
f1e61557 | 682 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
683 | set_compound_order(page, order); |
684 | __SetPageHead(page); | |
685 | for (i = 1; i < nr_pages; i++) { | |
686 | struct page *p = page + i; | |
58a84aa9 | 687 | set_page_count(p, 0); |
1c290f64 | 688 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 689 | set_compound_head(p, page); |
18229df5 | 690 | } |
53f9263b | 691 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
692 | } |
693 | ||
c0a32fc5 SG |
694 | #ifdef CONFIG_DEBUG_PAGEALLOC |
695 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 696 | |
8e57f8ac VB |
697 | bool _debug_pagealloc_enabled_early __read_mostly |
698 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
699 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 700 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 701 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
702 | |
703 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 704 | |
031bc574 JK |
705 | static int __init early_debug_pagealloc(char *buf) |
706 | { | |
8e57f8ac | 707 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
708 | } |
709 | early_param("debug_pagealloc", early_debug_pagealloc); | |
710 | ||
8e57f8ac | 711 | void init_debug_pagealloc(void) |
e30825f1 | 712 | { |
031bc574 JK |
713 | if (!debug_pagealloc_enabled()) |
714 | return; | |
715 | ||
8e57f8ac VB |
716 | static_branch_enable(&_debug_pagealloc_enabled); |
717 | ||
f1c1e9f7 JK |
718 | if (!debug_guardpage_minorder()) |
719 | return; | |
720 | ||
96a2b03f | 721 | static_branch_enable(&_debug_guardpage_enabled); |
e30825f1 JK |
722 | } |
723 | ||
c0a32fc5 SG |
724 | static int __init debug_guardpage_minorder_setup(char *buf) |
725 | { | |
726 | unsigned long res; | |
727 | ||
728 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 729 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
730 | return 0; |
731 | } | |
732 | _debug_guardpage_minorder = res; | |
1170532b | 733 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
734 | return 0; |
735 | } | |
f1c1e9f7 | 736 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 737 | |
acbc15a4 | 738 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 739 | unsigned int order, int migratetype) |
c0a32fc5 | 740 | { |
e30825f1 | 741 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
742 | return false; |
743 | ||
744 | if (order >= debug_guardpage_minorder()) | |
745 | return false; | |
e30825f1 | 746 | |
3972f6bb | 747 | __SetPageGuard(page); |
2847cf95 JK |
748 | INIT_LIST_HEAD(&page->lru); |
749 | set_page_private(page, order); | |
750 | /* Guard pages are not available for any usage */ | |
751 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
752 | |
753 | return true; | |
c0a32fc5 SG |
754 | } |
755 | ||
2847cf95 JK |
756 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
757 | unsigned int order, int migratetype) | |
c0a32fc5 | 758 | { |
e30825f1 JK |
759 | if (!debug_guardpage_enabled()) |
760 | return; | |
761 | ||
3972f6bb | 762 | __ClearPageGuard(page); |
e30825f1 | 763 | |
2847cf95 JK |
764 | set_page_private(page, 0); |
765 | if (!is_migrate_isolate(migratetype)) | |
766 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
767 | } |
768 | #else | |
acbc15a4 JK |
769 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
770 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
771 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
772 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
773 | #endif |
774 | ||
7aeb09f9 | 775 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 776 | { |
4c21e2f2 | 777 | set_page_private(page, order); |
676165a8 | 778 | __SetPageBuddy(page); |
1da177e4 LT |
779 | } |
780 | ||
1da177e4 LT |
781 | /* |
782 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 783 | * we can coalesce a page and its buddy if |
13ad59df | 784 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 785 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
786 | * (c) a page and its buddy have the same order && |
787 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 788 | * |
6e292b9b MW |
789 | * For recording whether a page is in the buddy system, we set PageBuddy. |
790 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 791 | * |
676165a8 | 792 | * For recording page's order, we use page_private(page). |
1da177e4 | 793 | */ |
cb2b95e1 | 794 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 795 | unsigned int order) |
1da177e4 | 796 | { |
c0a32fc5 | 797 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
798 | if (page_zone_id(page) != page_zone_id(buddy)) |
799 | return 0; | |
800 | ||
4c5018ce WY |
801 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
802 | ||
c0a32fc5 SG |
803 | return 1; |
804 | } | |
805 | ||
cb2b95e1 | 806 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
807 | /* |
808 | * zone check is done late to avoid uselessly | |
809 | * calculating zone/node ids for pages that could | |
810 | * never merge. | |
811 | */ | |
812 | if (page_zone_id(page) != page_zone_id(buddy)) | |
813 | return 0; | |
814 | ||
4c5018ce WY |
815 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
816 | ||
6aa3001b | 817 | return 1; |
676165a8 | 818 | } |
6aa3001b | 819 | return 0; |
1da177e4 LT |
820 | } |
821 | ||
5e1f0f09 MG |
822 | #ifdef CONFIG_COMPACTION |
823 | static inline struct capture_control *task_capc(struct zone *zone) | |
824 | { | |
825 | struct capture_control *capc = current->capture_control; | |
826 | ||
827 | return capc && | |
828 | !(current->flags & PF_KTHREAD) && | |
829 | !capc->page && | |
830 | capc->cc->zone == zone && | |
831 | capc->cc->direct_compaction ? capc : NULL; | |
832 | } | |
833 | ||
834 | static inline bool | |
835 | compaction_capture(struct capture_control *capc, struct page *page, | |
836 | int order, int migratetype) | |
837 | { | |
838 | if (!capc || order != capc->cc->order) | |
839 | return false; | |
840 | ||
841 | /* Do not accidentally pollute CMA or isolated regions*/ | |
842 | if (is_migrate_cma(migratetype) || | |
843 | is_migrate_isolate(migratetype)) | |
844 | return false; | |
845 | ||
846 | /* | |
847 | * Do not let lower order allocations polluate a movable pageblock. | |
848 | * This might let an unmovable request use a reclaimable pageblock | |
849 | * and vice-versa but no more than normal fallback logic which can | |
850 | * have trouble finding a high-order free page. | |
851 | */ | |
852 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
853 | return false; | |
854 | ||
855 | capc->page = page; | |
856 | return true; | |
857 | } | |
858 | ||
859 | #else | |
860 | static inline struct capture_control *task_capc(struct zone *zone) | |
861 | { | |
862 | return NULL; | |
863 | } | |
864 | ||
865 | static inline bool | |
866 | compaction_capture(struct capture_control *capc, struct page *page, | |
867 | int order, int migratetype) | |
868 | { | |
869 | return false; | |
870 | } | |
871 | #endif /* CONFIG_COMPACTION */ | |
872 | ||
1da177e4 LT |
873 | /* |
874 | * Freeing function for a buddy system allocator. | |
875 | * | |
876 | * The concept of a buddy system is to maintain direct-mapped table | |
877 | * (containing bit values) for memory blocks of various "orders". | |
878 | * The bottom level table contains the map for the smallest allocatable | |
879 | * units of memory (here, pages), and each level above it describes | |
880 | * pairs of units from the levels below, hence, "buddies". | |
881 | * At a high level, all that happens here is marking the table entry | |
882 | * at the bottom level available, and propagating the changes upward | |
883 | * as necessary, plus some accounting needed to play nicely with other | |
884 | * parts of the VM system. | |
885 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
886 | * free pages of length of (1 << order) and marked with PageBuddy. |
887 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 888 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
889 | * other. That is, if we allocate a small block, and both were |
890 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 891 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 892 | * triggers coalescing into a block of larger size. |
1da177e4 | 893 | * |
6d49e352 | 894 | * -- nyc |
1da177e4 LT |
895 | */ |
896 | ||
48db57f8 | 897 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 898 | unsigned long pfn, |
ed0ae21d MG |
899 | struct zone *zone, unsigned int order, |
900 | int migratetype) | |
1da177e4 | 901 | { |
76741e77 VB |
902 | unsigned long combined_pfn; |
903 | unsigned long uninitialized_var(buddy_pfn); | |
6dda9d55 | 904 | struct page *buddy; |
d9dddbf5 | 905 | unsigned int max_order; |
5e1f0f09 | 906 | struct capture_control *capc = task_capc(zone); |
d9dddbf5 VB |
907 | |
908 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 909 | |
d29bb978 | 910 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 911 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 912 | |
ed0ae21d | 913 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 914 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 915 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 916 | |
76741e77 | 917 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 918 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 919 | |
d9dddbf5 | 920 | continue_merging: |
3c605096 | 921 | while (order < max_order - 1) { |
5e1f0f09 MG |
922 | if (compaction_capture(capc, page, order, migratetype)) { |
923 | __mod_zone_freepage_state(zone, -(1 << order), | |
924 | migratetype); | |
925 | return; | |
926 | } | |
76741e77 VB |
927 | buddy_pfn = __find_buddy_pfn(pfn, order); |
928 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
929 | |
930 | if (!pfn_valid_within(buddy_pfn)) | |
931 | goto done_merging; | |
cb2b95e1 | 932 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 933 | goto done_merging; |
c0a32fc5 SG |
934 | /* |
935 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
936 | * merge with it and move up one order. | |
937 | */ | |
b03641af | 938 | if (page_is_guard(buddy)) |
2847cf95 | 939 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af DW |
940 | else |
941 | del_page_from_free_area(buddy, &zone->free_area[order]); | |
76741e77 VB |
942 | combined_pfn = buddy_pfn & pfn; |
943 | page = page + (combined_pfn - pfn); | |
944 | pfn = combined_pfn; | |
1da177e4 LT |
945 | order++; |
946 | } | |
d9dddbf5 VB |
947 | if (max_order < MAX_ORDER) { |
948 | /* If we are here, it means order is >= pageblock_order. | |
949 | * We want to prevent merge between freepages on isolate | |
950 | * pageblock and normal pageblock. Without this, pageblock | |
951 | * isolation could cause incorrect freepage or CMA accounting. | |
952 | * | |
953 | * We don't want to hit this code for the more frequent | |
954 | * low-order merging. | |
955 | */ | |
956 | if (unlikely(has_isolate_pageblock(zone))) { | |
957 | int buddy_mt; | |
958 | ||
76741e77 VB |
959 | buddy_pfn = __find_buddy_pfn(pfn, order); |
960 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
961 | buddy_mt = get_pageblock_migratetype(buddy); |
962 | ||
963 | if (migratetype != buddy_mt | |
964 | && (is_migrate_isolate(migratetype) || | |
965 | is_migrate_isolate(buddy_mt))) | |
966 | goto done_merging; | |
967 | } | |
968 | max_order++; | |
969 | goto continue_merging; | |
970 | } | |
971 | ||
972 | done_merging: | |
1da177e4 | 973 | set_page_order(page, order); |
6dda9d55 CZ |
974 | |
975 | /* | |
976 | * If this is not the largest possible page, check if the buddy | |
977 | * of the next-highest order is free. If it is, it's possible | |
978 | * that pages are being freed that will coalesce soon. In case, | |
979 | * that is happening, add the free page to the tail of the list | |
980 | * so it's less likely to be used soon and more likely to be merged | |
981 | * as a higher order page | |
982 | */ | |
97500a4a DW |
983 | if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn) |
984 | && !is_shuffle_order(order)) { | |
6dda9d55 | 985 | struct page *higher_page, *higher_buddy; |
76741e77 VB |
986 | combined_pfn = buddy_pfn & pfn; |
987 | higher_page = page + (combined_pfn - pfn); | |
988 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
989 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
b4fb8f66 TL |
990 | if (pfn_valid_within(buddy_pfn) && |
991 | page_is_buddy(higher_page, higher_buddy, order + 1)) { | |
b03641af DW |
992 | add_to_free_area_tail(page, &zone->free_area[order], |
993 | migratetype); | |
994 | return; | |
6dda9d55 CZ |
995 | } |
996 | } | |
997 | ||
97500a4a DW |
998 | if (is_shuffle_order(order)) |
999 | add_to_free_area_random(page, &zone->free_area[order], | |
1000 | migratetype); | |
1001 | else | |
1002 | add_to_free_area(page, &zone->free_area[order], migratetype); | |
1003 | ||
1da177e4 LT |
1004 | } |
1005 | ||
7bfec6f4 MG |
1006 | /* |
1007 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1008 | * try and check multiple fields with one check. The caller must do a detailed | |
1009 | * check if necessary. | |
1010 | */ | |
1011 | static inline bool page_expected_state(struct page *page, | |
1012 | unsigned long check_flags) | |
1013 | { | |
1014 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1015 | return false; | |
1016 | ||
1017 | if (unlikely((unsigned long)page->mapping | | |
1018 | page_ref_count(page) | | |
1019 | #ifdef CONFIG_MEMCG | |
1020 | (unsigned long)page->mem_cgroup | | |
1021 | #endif | |
1022 | (page->flags & check_flags))) | |
1023 | return false; | |
1024 | ||
1025 | return true; | |
1026 | } | |
1027 | ||
bb552ac6 | 1028 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 1029 | { |
7bfec6f4 MG |
1030 | const char *bad_reason; |
1031 | unsigned long bad_flags; | |
1032 | ||
7bfec6f4 MG |
1033 | bad_reason = NULL; |
1034 | bad_flags = 0; | |
f0b791a3 | 1035 | |
53f9263b | 1036 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1037 | bad_reason = "nonzero mapcount"; |
1038 | if (unlikely(page->mapping != NULL)) | |
1039 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1040 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1041 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
1042 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
1043 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
1044 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
1045 | } | |
9edad6ea JW |
1046 | #ifdef CONFIG_MEMCG |
1047 | if (unlikely(page->mem_cgroup)) | |
1048 | bad_reason = "page still charged to cgroup"; | |
1049 | #endif | |
7bfec6f4 | 1050 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
1051 | } |
1052 | ||
1053 | static inline int free_pages_check(struct page *page) | |
1054 | { | |
da838d4f | 1055 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 1056 | return 0; |
bb552ac6 MG |
1057 | |
1058 | /* Something has gone sideways, find it */ | |
1059 | free_pages_check_bad(page); | |
7bfec6f4 | 1060 | return 1; |
1da177e4 LT |
1061 | } |
1062 | ||
4db7548c MG |
1063 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1064 | { | |
1065 | int ret = 1; | |
1066 | ||
1067 | /* | |
1068 | * We rely page->lru.next never has bit 0 set, unless the page | |
1069 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1070 | */ | |
1071 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1072 | ||
1073 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1074 | ret = 0; | |
1075 | goto out; | |
1076 | } | |
1077 | switch (page - head_page) { | |
1078 | case 1: | |
4da1984e | 1079 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c MG |
1080 | if (unlikely(compound_mapcount(page))) { |
1081 | bad_page(page, "nonzero compound_mapcount", 0); | |
1082 | goto out; | |
1083 | } | |
1084 | break; | |
1085 | case 2: | |
1086 | /* | |
1087 | * the second tail page: ->mapping is | |
fa3015b7 | 1088 | * deferred_list.next -- ignore value. |
4db7548c MG |
1089 | */ |
1090 | break; | |
1091 | default: | |
1092 | if (page->mapping != TAIL_MAPPING) { | |
1093 | bad_page(page, "corrupted mapping in tail page", 0); | |
1094 | goto out; | |
1095 | } | |
1096 | break; | |
1097 | } | |
1098 | if (unlikely(!PageTail(page))) { | |
1099 | bad_page(page, "PageTail not set", 0); | |
1100 | goto out; | |
1101 | } | |
1102 | if (unlikely(compound_head(page) != head_page)) { | |
1103 | bad_page(page, "compound_head not consistent", 0); | |
1104 | goto out; | |
1105 | } | |
1106 | ret = 0; | |
1107 | out: | |
1108 | page->mapping = NULL; | |
1109 | clear_compound_head(page); | |
1110 | return ret; | |
1111 | } | |
1112 | ||
6471384a AP |
1113 | static void kernel_init_free_pages(struct page *page, int numpages) |
1114 | { | |
1115 | int i; | |
1116 | ||
1117 | for (i = 0; i < numpages; i++) | |
1118 | clear_highpage(page + i); | |
1119 | } | |
1120 | ||
e2769dbd MG |
1121 | static __always_inline bool free_pages_prepare(struct page *page, |
1122 | unsigned int order, bool check_free) | |
4db7548c | 1123 | { |
e2769dbd | 1124 | int bad = 0; |
4db7548c | 1125 | |
4db7548c MG |
1126 | VM_BUG_ON_PAGE(PageTail(page), page); |
1127 | ||
e2769dbd | 1128 | trace_mm_page_free(page, order); |
e2769dbd MG |
1129 | |
1130 | /* | |
1131 | * Check tail pages before head page information is cleared to | |
1132 | * avoid checking PageCompound for order-0 pages. | |
1133 | */ | |
1134 | if (unlikely(order)) { | |
1135 | bool compound = PageCompound(page); | |
1136 | int i; | |
1137 | ||
1138 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1139 | |
9a73f61b KS |
1140 | if (compound) |
1141 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1142 | for (i = 1; i < (1 << order); i++) { |
1143 | if (compound) | |
1144 | bad += free_tail_pages_check(page, page + i); | |
1145 | if (unlikely(free_pages_check(page + i))) { | |
1146 | bad++; | |
1147 | continue; | |
1148 | } | |
1149 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1150 | } | |
1151 | } | |
bda807d4 | 1152 | if (PageMappingFlags(page)) |
4db7548c | 1153 | page->mapping = NULL; |
c4159a75 | 1154 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
60cd4bcd | 1155 | __memcg_kmem_uncharge(page, order); |
e2769dbd MG |
1156 | if (check_free) |
1157 | bad += free_pages_check(page); | |
1158 | if (bad) | |
1159 | return false; | |
4db7548c | 1160 | |
e2769dbd MG |
1161 | page_cpupid_reset_last(page); |
1162 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1163 | reset_page_owner(page, order); | |
4db7548c MG |
1164 | |
1165 | if (!PageHighMem(page)) { | |
1166 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1167 | PAGE_SIZE << order); |
4db7548c | 1168 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1169 | PAGE_SIZE << order); |
4db7548c | 1170 | } |
6471384a AP |
1171 | if (want_init_on_free()) |
1172 | kernel_init_free_pages(page, 1 << order); | |
1173 | ||
e2769dbd | 1174 | kernel_poison_pages(page, 1 << order, 0); |
234fdce8 QC |
1175 | /* |
1176 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1177 | * does this. So nothing which can access the page's contents should | |
1178 | * happen after this. | |
1179 | */ | |
1180 | arch_free_page(page, order); | |
1181 | ||
8e57f8ac | 1182 | if (debug_pagealloc_enabled_static()) |
d6332692 RE |
1183 | kernel_map_pages(page, 1 << order, 0); |
1184 | ||
3c0c12cc | 1185 | kasan_free_nondeferred_pages(page, order); |
4db7548c | 1186 | |
4db7548c MG |
1187 | return true; |
1188 | } | |
1189 | ||
e2769dbd | 1190 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1191 | /* |
1192 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1193 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1194 | * moved from pcp lists to free lists. | |
1195 | */ | |
1196 | static bool free_pcp_prepare(struct page *page) | |
e2769dbd MG |
1197 | { |
1198 | return free_pages_prepare(page, 0, true); | |
1199 | } | |
1200 | ||
4462b32c | 1201 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1202 | { |
8e57f8ac | 1203 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
1204 | return free_pages_check(page); |
1205 | else | |
1206 | return false; | |
e2769dbd MG |
1207 | } |
1208 | #else | |
4462b32c VB |
1209 | /* |
1210 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1211 | * moving from pcp lists to free list in order to reduce overhead. With | |
1212 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1213 | * to the pcp lists. | |
1214 | */ | |
e2769dbd MG |
1215 | static bool free_pcp_prepare(struct page *page) |
1216 | { | |
8e57f8ac | 1217 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
1218 | return free_pages_prepare(page, 0, true); |
1219 | else | |
1220 | return free_pages_prepare(page, 0, false); | |
e2769dbd MG |
1221 | } |
1222 | ||
4db7548c MG |
1223 | static bool bulkfree_pcp_prepare(struct page *page) |
1224 | { | |
1225 | return free_pages_check(page); | |
1226 | } | |
1227 | #endif /* CONFIG_DEBUG_VM */ | |
1228 | ||
97334162 AL |
1229 | static inline void prefetch_buddy(struct page *page) |
1230 | { | |
1231 | unsigned long pfn = page_to_pfn(page); | |
1232 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1233 | struct page *buddy = page + (buddy_pfn - pfn); | |
1234 | ||
1235 | prefetch(buddy); | |
1236 | } | |
1237 | ||
1da177e4 | 1238 | /* |
5f8dcc21 | 1239 | * Frees a number of pages from the PCP lists |
1da177e4 | 1240 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1241 | * count is the number of pages to free. |
1da177e4 LT |
1242 | * |
1243 | * If the zone was previously in an "all pages pinned" state then look to | |
1244 | * see if this freeing clears that state. | |
1245 | * | |
1246 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1247 | * pinned" detection logic. | |
1248 | */ | |
5f8dcc21 MG |
1249 | static void free_pcppages_bulk(struct zone *zone, int count, |
1250 | struct per_cpu_pages *pcp) | |
1da177e4 | 1251 | { |
5f8dcc21 | 1252 | int migratetype = 0; |
a6f9edd6 | 1253 | int batch_free = 0; |
97334162 | 1254 | int prefetch_nr = 0; |
3777999d | 1255 | bool isolated_pageblocks; |
0a5f4e5b AL |
1256 | struct page *page, *tmp; |
1257 | LIST_HEAD(head); | |
f2260e6b | 1258 | |
e5b31ac2 | 1259 | while (count) { |
5f8dcc21 MG |
1260 | struct list_head *list; |
1261 | ||
1262 | /* | |
a6f9edd6 MG |
1263 | * Remove pages from lists in a round-robin fashion. A |
1264 | * batch_free count is maintained that is incremented when an | |
1265 | * empty list is encountered. This is so more pages are freed | |
1266 | * off fuller lists instead of spinning excessively around empty | |
1267 | * lists | |
5f8dcc21 MG |
1268 | */ |
1269 | do { | |
a6f9edd6 | 1270 | batch_free++; |
5f8dcc21 MG |
1271 | if (++migratetype == MIGRATE_PCPTYPES) |
1272 | migratetype = 0; | |
1273 | list = &pcp->lists[migratetype]; | |
1274 | } while (list_empty(list)); | |
48db57f8 | 1275 | |
1d16871d NK |
1276 | /* This is the only non-empty list. Free them all. */ |
1277 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1278 | batch_free = count; |
1d16871d | 1279 | |
a6f9edd6 | 1280 | do { |
a16601c5 | 1281 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1282 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1283 | list_del(&page->lru); |
77ba9062 | 1284 | pcp->count--; |
aa016d14 | 1285 | |
4db7548c MG |
1286 | if (bulkfree_pcp_prepare(page)) |
1287 | continue; | |
1288 | ||
0a5f4e5b | 1289 | list_add_tail(&page->lru, &head); |
97334162 AL |
1290 | |
1291 | /* | |
1292 | * We are going to put the page back to the global | |
1293 | * pool, prefetch its buddy to speed up later access | |
1294 | * under zone->lock. It is believed the overhead of | |
1295 | * an additional test and calculating buddy_pfn here | |
1296 | * can be offset by reduced memory latency later. To | |
1297 | * avoid excessive prefetching due to large count, only | |
1298 | * prefetch buddy for the first pcp->batch nr of pages. | |
1299 | */ | |
1300 | if (prefetch_nr++ < pcp->batch) | |
1301 | prefetch_buddy(page); | |
e5b31ac2 | 1302 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1303 | } |
0a5f4e5b AL |
1304 | |
1305 | spin_lock(&zone->lock); | |
1306 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1307 | ||
1308 | /* | |
1309 | * Use safe version since after __free_one_page(), | |
1310 | * page->lru.next will not point to original list. | |
1311 | */ | |
1312 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1313 | int mt = get_pcppage_migratetype(page); | |
1314 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1315 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1316 | /* Pageblock could have been isolated meanwhile */ | |
1317 | if (unlikely(isolated_pageblocks)) | |
1318 | mt = get_pageblock_migratetype(page); | |
1319 | ||
1320 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); | |
1321 | trace_mm_page_pcpu_drain(page, 0, mt); | |
1322 | } | |
d34b0733 | 1323 | spin_unlock(&zone->lock); |
1da177e4 LT |
1324 | } |
1325 | ||
dc4b0caf MG |
1326 | static void free_one_page(struct zone *zone, |
1327 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1328 | unsigned int order, |
ed0ae21d | 1329 | int migratetype) |
1da177e4 | 1330 | { |
d34b0733 | 1331 | spin_lock(&zone->lock); |
ad53f92e JK |
1332 | if (unlikely(has_isolate_pageblock(zone) || |
1333 | is_migrate_isolate(migratetype))) { | |
1334 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1335 | } |
dc4b0caf | 1336 | __free_one_page(page, pfn, zone, order, migratetype); |
d34b0733 | 1337 | spin_unlock(&zone->lock); |
48db57f8 NP |
1338 | } |
1339 | ||
1e8ce83c | 1340 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1341 | unsigned long zone, int nid) |
1e8ce83c | 1342 | { |
d0dc12e8 | 1343 | mm_zero_struct_page(page); |
1e8ce83c | 1344 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1345 | init_page_count(page); |
1346 | page_mapcount_reset(page); | |
1347 | page_cpupid_reset_last(page); | |
2813b9c0 | 1348 | page_kasan_tag_reset(page); |
1e8ce83c | 1349 | |
1e8ce83c RH |
1350 | INIT_LIST_HEAD(&page->lru); |
1351 | #ifdef WANT_PAGE_VIRTUAL | |
1352 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1353 | if (!is_highmem_idx(zone)) | |
1354 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1355 | #endif | |
1356 | } | |
1357 | ||
7e18adb4 | 1358 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1359 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1360 | { |
1361 | pg_data_t *pgdat; | |
1362 | int nid, zid; | |
1363 | ||
1364 | if (!early_page_uninitialised(pfn)) | |
1365 | return; | |
1366 | ||
1367 | nid = early_pfn_to_nid(pfn); | |
1368 | pgdat = NODE_DATA(nid); | |
1369 | ||
1370 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1371 | struct zone *zone = &pgdat->node_zones[zid]; | |
1372 | ||
1373 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1374 | break; | |
1375 | } | |
d0dc12e8 | 1376 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1377 | } |
1378 | #else | |
1379 | static inline void init_reserved_page(unsigned long pfn) | |
1380 | { | |
1381 | } | |
1382 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1383 | ||
92923ca3 NZ |
1384 | /* |
1385 | * Initialised pages do not have PageReserved set. This function is | |
1386 | * called for each range allocated by the bootmem allocator and | |
1387 | * marks the pages PageReserved. The remaining valid pages are later | |
1388 | * sent to the buddy page allocator. | |
1389 | */ | |
4b50bcc7 | 1390 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1391 | { |
1392 | unsigned long start_pfn = PFN_DOWN(start); | |
1393 | unsigned long end_pfn = PFN_UP(end); | |
1394 | ||
7e18adb4 MG |
1395 | for (; start_pfn < end_pfn; start_pfn++) { |
1396 | if (pfn_valid(start_pfn)) { | |
1397 | struct page *page = pfn_to_page(start_pfn); | |
1398 | ||
1399 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1400 | |
1401 | /* Avoid false-positive PageTail() */ | |
1402 | INIT_LIST_HEAD(&page->lru); | |
1403 | ||
d483da5b AD |
1404 | /* |
1405 | * no need for atomic set_bit because the struct | |
1406 | * page is not visible yet so nobody should | |
1407 | * access it yet. | |
1408 | */ | |
1409 | __SetPageReserved(page); | |
7e18adb4 MG |
1410 | } |
1411 | } | |
92923ca3 NZ |
1412 | } |
1413 | ||
ec95f53a KM |
1414 | static void __free_pages_ok(struct page *page, unsigned int order) |
1415 | { | |
d34b0733 | 1416 | unsigned long flags; |
95e34412 | 1417 | int migratetype; |
dc4b0caf | 1418 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1419 | |
e2769dbd | 1420 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1421 | return; |
1422 | ||
cfc47a28 | 1423 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1424 | local_irq_save(flags); |
1425 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1426 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1427 | local_irq_restore(flags); |
1da177e4 LT |
1428 | } |
1429 | ||
a9cd410a | 1430 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1431 | { |
c3993076 | 1432 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1433 | struct page *p = page; |
c3993076 | 1434 | unsigned int loop; |
a226f6c8 | 1435 | |
e2d0bd2b YL |
1436 | prefetchw(p); |
1437 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1438 | prefetchw(p + 1); | |
c3993076 JW |
1439 | __ClearPageReserved(p); |
1440 | set_page_count(p, 0); | |
a226f6c8 | 1441 | } |
e2d0bd2b YL |
1442 | __ClearPageReserved(p); |
1443 | set_page_count(p, 0); | |
c3993076 | 1444 | |
9705bea5 | 1445 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
c3993076 JW |
1446 | set_page_refcounted(page); |
1447 | __free_pages(page, order); | |
a226f6c8 DH |
1448 | } |
1449 | ||
75a592a4 MG |
1450 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1451 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1452 | |
75a592a4 MG |
1453 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1454 | ||
1455 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1456 | { | |
7ace9917 | 1457 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1458 | int nid; |
1459 | ||
7ace9917 | 1460 | spin_lock(&early_pfn_lock); |
75a592a4 | 1461 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1462 | if (nid < 0) |
e4568d38 | 1463 | nid = first_online_node; |
7ace9917 MG |
1464 | spin_unlock(&early_pfn_lock); |
1465 | ||
1466 | return nid; | |
75a592a4 MG |
1467 | } |
1468 | #endif | |
1469 | ||
1470 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
56ec43d8 AD |
1471 | /* Only safe to use early in boot when initialisation is single-threaded */ |
1472 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
75a592a4 MG |
1473 | { |
1474 | int nid; | |
1475 | ||
56ec43d8 | 1476 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
75a592a4 MG |
1477 | if (nid >= 0 && nid != node) |
1478 | return false; | |
1479 | return true; | |
1480 | } | |
1481 | ||
75a592a4 | 1482 | #else |
75a592a4 MG |
1483 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
1484 | { | |
1485 | return true; | |
1486 | } | |
75a592a4 MG |
1487 | #endif |
1488 | ||
1489 | ||
7c2ee349 | 1490 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1491 | unsigned int order) |
1492 | { | |
1493 | if (early_page_uninitialised(pfn)) | |
1494 | return; | |
a9cd410a | 1495 | __free_pages_core(page, order); |
3a80a7fa MG |
1496 | } |
1497 | ||
7cf91a98 JK |
1498 | /* |
1499 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1500 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1501 | * with the migration of free compaction scanner. The scanners then need to | |
1502 | * use only pfn_valid_within() check for arches that allow holes within | |
1503 | * pageblocks. | |
1504 | * | |
1505 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1506 | * | |
1507 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1508 | * i.e. it's possible that all pages within a zones range of pages do not | |
1509 | * belong to a single zone. We assume that a border between node0 and node1 | |
1510 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1511 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1512 | * the first and last page of a pageblock and avoid checking each individual | |
1513 | * page in a pageblock. | |
1514 | */ | |
1515 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1516 | unsigned long end_pfn, struct zone *zone) | |
1517 | { | |
1518 | struct page *start_page; | |
1519 | struct page *end_page; | |
1520 | ||
1521 | /* end_pfn is one past the range we are checking */ | |
1522 | end_pfn--; | |
1523 | ||
1524 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1525 | return NULL; | |
1526 | ||
2d070eab MH |
1527 | start_page = pfn_to_online_page(start_pfn); |
1528 | if (!start_page) | |
1529 | return NULL; | |
7cf91a98 JK |
1530 | |
1531 | if (page_zone(start_page) != zone) | |
1532 | return NULL; | |
1533 | ||
1534 | end_page = pfn_to_page(end_pfn); | |
1535 | ||
1536 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1537 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1538 | return NULL; | |
1539 | ||
1540 | return start_page; | |
1541 | } | |
1542 | ||
1543 | void set_zone_contiguous(struct zone *zone) | |
1544 | { | |
1545 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1546 | unsigned long block_end_pfn; | |
1547 | ||
1548 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1549 | for (; block_start_pfn < zone_end_pfn(zone); | |
1550 | block_start_pfn = block_end_pfn, | |
1551 | block_end_pfn += pageblock_nr_pages) { | |
1552 | ||
1553 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1554 | ||
1555 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1556 | block_end_pfn, zone)) | |
1557 | return; | |
1558 | } | |
1559 | ||
1560 | /* We confirm that there is no hole */ | |
1561 | zone->contiguous = true; | |
1562 | } | |
1563 | ||
1564 | void clear_zone_contiguous(struct zone *zone) | |
1565 | { | |
1566 | zone->contiguous = false; | |
1567 | } | |
1568 | ||
7e18adb4 | 1569 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1570 | static void __init deferred_free_range(unsigned long pfn, |
1571 | unsigned long nr_pages) | |
a4de83dd | 1572 | { |
2f47a91f PT |
1573 | struct page *page; |
1574 | unsigned long i; | |
a4de83dd | 1575 | |
2f47a91f | 1576 | if (!nr_pages) |
a4de83dd MG |
1577 | return; |
1578 | ||
2f47a91f PT |
1579 | page = pfn_to_page(pfn); |
1580 | ||
a4de83dd | 1581 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1582 | if (nr_pages == pageblock_nr_pages && |
1583 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1584 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1585 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1586 | return; |
1587 | } | |
1588 | ||
e780149b XQ |
1589 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1590 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1591 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
a9cd410a | 1592 | __free_pages_core(page, 0); |
e780149b | 1593 | } |
a4de83dd MG |
1594 | } |
1595 | ||
d3cd131d NS |
1596 | /* Completion tracking for deferred_init_memmap() threads */ |
1597 | static atomic_t pgdat_init_n_undone __initdata; | |
1598 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1599 | ||
1600 | static inline void __init pgdat_init_report_one_done(void) | |
1601 | { | |
1602 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1603 | complete(&pgdat_init_all_done_comp); | |
1604 | } | |
0e1cc95b | 1605 | |
2f47a91f | 1606 | /* |
80b1f41c PT |
1607 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1608 | * | |
1609 | * First we check if pfn is valid on architectures where it is possible to have | |
1610 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1611 | * function is optimized out. | |
1612 | * | |
1613 | * Then, we check if a current large page is valid by only checking the validity | |
1614 | * of the head pfn. | |
2f47a91f | 1615 | */ |
56ec43d8 | 1616 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1617 | { |
80b1f41c PT |
1618 | if (!pfn_valid_within(pfn)) |
1619 | return false; | |
1620 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1621 | return false; | |
80b1f41c PT |
1622 | return true; |
1623 | } | |
2f47a91f | 1624 | |
80b1f41c PT |
1625 | /* |
1626 | * Free pages to buddy allocator. Try to free aligned pages in | |
1627 | * pageblock_nr_pages sizes. | |
1628 | */ | |
56ec43d8 | 1629 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1630 | unsigned long end_pfn) |
1631 | { | |
80b1f41c PT |
1632 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
1633 | unsigned long nr_free = 0; | |
2f47a91f | 1634 | |
80b1f41c | 1635 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1636 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1637 | deferred_free_range(pfn - nr_free, nr_free); |
1638 | nr_free = 0; | |
1639 | } else if (!(pfn & nr_pgmask)) { | |
1640 | deferred_free_range(pfn - nr_free, nr_free); | |
1641 | nr_free = 1; | |
3a2d7fa8 | 1642 | touch_nmi_watchdog(); |
80b1f41c PT |
1643 | } else { |
1644 | nr_free++; | |
1645 | } | |
1646 | } | |
1647 | /* Free the last block of pages to allocator */ | |
1648 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1649 | } |
1650 | ||
80b1f41c PT |
1651 | /* |
1652 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1653 | * by performing it only once every pageblock_nr_pages. | |
1654 | * Return number of pages initialized. | |
1655 | */ | |
56ec43d8 | 1656 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1657 | unsigned long pfn, |
1658 | unsigned long end_pfn) | |
2f47a91f | 1659 | { |
2f47a91f | 1660 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
56ec43d8 | 1661 | int nid = zone_to_nid(zone); |
2f47a91f | 1662 | unsigned long nr_pages = 0; |
56ec43d8 | 1663 | int zid = zone_idx(zone); |
2f47a91f | 1664 | struct page *page = NULL; |
2f47a91f | 1665 | |
80b1f41c | 1666 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1667 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1668 | page = NULL; |
2f47a91f | 1669 | continue; |
80b1f41c | 1670 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1671 | page = pfn_to_page(pfn); |
3a2d7fa8 | 1672 | touch_nmi_watchdog(); |
80b1f41c PT |
1673 | } else { |
1674 | page++; | |
2f47a91f | 1675 | } |
d0dc12e8 | 1676 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1677 | nr_pages++; |
2f47a91f | 1678 | } |
80b1f41c | 1679 | return (nr_pages); |
2f47a91f PT |
1680 | } |
1681 | ||
0e56acae AD |
1682 | /* |
1683 | * This function is meant to pre-load the iterator for the zone init. | |
1684 | * Specifically it walks through the ranges until we are caught up to the | |
1685 | * first_init_pfn value and exits there. If we never encounter the value we | |
1686 | * return false indicating there are no valid ranges left. | |
1687 | */ | |
1688 | static bool __init | |
1689 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
1690 | unsigned long *spfn, unsigned long *epfn, | |
1691 | unsigned long first_init_pfn) | |
1692 | { | |
1693 | u64 j; | |
1694 | ||
1695 | /* | |
1696 | * Start out by walking through the ranges in this zone that have | |
1697 | * already been initialized. We don't need to do anything with them | |
1698 | * so we just need to flush them out of the system. | |
1699 | */ | |
1700 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
1701 | if (*epfn <= first_init_pfn) | |
1702 | continue; | |
1703 | if (*spfn < first_init_pfn) | |
1704 | *spfn = first_init_pfn; | |
1705 | *i = j; | |
1706 | return true; | |
1707 | } | |
1708 | ||
1709 | return false; | |
1710 | } | |
1711 | ||
1712 | /* | |
1713 | * Initialize and free pages. We do it in two loops: first we initialize | |
1714 | * struct page, then free to buddy allocator, because while we are | |
1715 | * freeing pages we can access pages that are ahead (computing buddy | |
1716 | * page in __free_one_page()). | |
1717 | * | |
1718 | * In order to try and keep some memory in the cache we have the loop | |
1719 | * broken along max page order boundaries. This way we will not cause | |
1720 | * any issues with the buddy page computation. | |
1721 | */ | |
1722 | static unsigned long __init | |
1723 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
1724 | unsigned long *end_pfn) | |
1725 | { | |
1726 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
1727 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
1728 | unsigned long nr_pages = 0; | |
1729 | u64 j = *i; | |
1730 | ||
1731 | /* First we loop through and initialize the page values */ | |
1732 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
1733 | unsigned long t; | |
1734 | ||
1735 | if (mo_pfn <= *start_pfn) | |
1736 | break; | |
1737 | ||
1738 | t = min(mo_pfn, *end_pfn); | |
1739 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
1740 | ||
1741 | if (mo_pfn < *end_pfn) { | |
1742 | *start_pfn = mo_pfn; | |
1743 | break; | |
1744 | } | |
1745 | } | |
1746 | ||
1747 | /* Reset values and now loop through freeing pages as needed */ | |
1748 | swap(j, *i); | |
1749 | ||
1750 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
1751 | unsigned long t; | |
1752 | ||
1753 | if (mo_pfn <= spfn) | |
1754 | break; | |
1755 | ||
1756 | t = min(mo_pfn, epfn); | |
1757 | deferred_free_pages(spfn, t); | |
1758 | ||
1759 | if (mo_pfn <= epfn) | |
1760 | break; | |
1761 | } | |
1762 | ||
1763 | return nr_pages; | |
1764 | } | |
1765 | ||
7e18adb4 | 1766 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1767 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1768 | { |
0e1cc95b | 1769 | pg_data_t *pgdat = data; |
0e56acae AD |
1770 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
1771 | unsigned long spfn = 0, epfn = 0, nr_pages = 0; | |
1772 | unsigned long first_init_pfn, flags; | |
7e18adb4 | 1773 | unsigned long start = jiffies; |
7e18adb4 | 1774 | struct zone *zone; |
0e56acae | 1775 | int zid; |
2f47a91f | 1776 | u64 i; |
7e18adb4 | 1777 | |
3a2d7fa8 PT |
1778 | /* Bind memory initialisation thread to a local node if possible */ |
1779 | if (!cpumask_empty(cpumask)) | |
1780 | set_cpus_allowed_ptr(current, cpumask); | |
1781 | ||
1782 | pgdat_resize_lock(pgdat, &flags); | |
1783 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1784 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1785 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1786 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1787 | return 0; |
1788 | } | |
1789 | ||
7e18adb4 MG |
1790 | /* Sanity check boundaries */ |
1791 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1792 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1793 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1794 | ||
1795 | /* Only the highest zone is deferred so find it */ | |
1796 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1797 | zone = pgdat->node_zones + zid; | |
1798 | if (first_init_pfn < zone_end_pfn(zone)) | |
1799 | break; | |
1800 | } | |
0e56acae AD |
1801 | |
1802 | /* If the zone is empty somebody else may have cleared out the zone */ | |
1803 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1804 | first_init_pfn)) | |
1805 | goto zone_empty; | |
7e18adb4 | 1806 | |
80b1f41c | 1807 | /* |
0e56acae AD |
1808 | * Initialize and free pages in MAX_ORDER sized increments so |
1809 | * that we can avoid introducing any issues with the buddy | |
1810 | * allocator. | |
80b1f41c | 1811 | */ |
0e56acae AD |
1812 | while (spfn < epfn) |
1813 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
1814 | zone_empty: | |
3a2d7fa8 | 1815 | pgdat_resize_unlock(pgdat, &flags); |
7e18adb4 MG |
1816 | |
1817 | /* Sanity check that the next zone really is unpopulated */ | |
1818 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1819 | ||
837566e7 AD |
1820 | pr_info("node %d initialised, %lu pages in %ums\n", |
1821 | pgdat->node_id, nr_pages, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
1822 | |
1823 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1824 | return 0; |
1825 | } | |
c9e97a19 | 1826 | |
c9e97a19 PT |
1827 | /* |
1828 | * If this zone has deferred pages, try to grow it by initializing enough | |
1829 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
1830 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
1831 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
1832 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
1833 | * | |
1834 | * Return true when zone was grown, otherwise return false. We return true even | |
1835 | * when we grow less than requested, to let the caller decide if there are | |
1836 | * enough pages to satisfy the allocation. | |
1837 | * | |
1838 | * Note: We use noinline because this function is needed only during boot, and | |
1839 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
1840 | * making sure that it is not inlined into permanent text section. | |
1841 | */ | |
1842 | static noinline bool __init | |
1843 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
1844 | { | |
c9e97a19 | 1845 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 1846 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 1847 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
1848 | unsigned long spfn, epfn, flags; |
1849 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
1850 | u64 i; |
1851 | ||
1852 | /* Only the last zone may have deferred pages */ | |
1853 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
1854 | return false; | |
1855 | ||
1856 | pgdat_resize_lock(pgdat, &flags); | |
1857 | ||
1858 | /* | |
1859 | * If deferred pages have been initialized while we were waiting for | |
1860 | * the lock, return true, as the zone was grown. The caller will retry | |
1861 | * this zone. We won't return to this function since the caller also | |
1862 | * has this static branch. | |
1863 | */ | |
1864 | if (!static_branch_unlikely(&deferred_pages)) { | |
1865 | pgdat_resize_unlock(pgdat, &flags); | |
1866 | return true; | |
1867 | } | |
1868 | ||
1869 | /* | |
1870 | * If someone grew this zone while we were waiting for spinlock, return | |
1871 | * true, as there might be enough pages already. | |
1872 | */ | |
1873 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
1874 | pgdat_resize_unlock(pgdat, &flags); | |
1875 | return true; | |
1876 | } | |
1877 | ||
0e56acae AD |
1878 | /* If the zone is empty somebody else may have cleared out the zone */ |
1879 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1880 | first_deferred_pfn)) { | |
1881 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 1882 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
1883 | /* Retry only once. */ |
1884 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
1885 | } |
1886 | ||
0e56acae AD |
1887 | /* |
1888 | * Initialize and free pages in MAX_ORDER sized increments so | |
1889 | * that we can avoid introducing any issues with the buddy | |
1890 | * allocator. | |
1891 | */ | |
1892 | while (spfn < epfn) { | |
1893 | /* update our first deferred PFN for this section */ | |
1894 | first_deferred_pfn = spfn; | |
1895 | ||
1896 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
c9e97a19 | 1897 | |
0e56acae AD |
1898 | /* We should only stop along section boundaries */ |
1899 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
1900 | continue; | |
c9e97a19 | 1901 | |
0e56acae | 1902 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
1903 | if (nr_pages >= nr_pages_needed) |
1904 | break; | |
1905 | } | |
1906 | ||
0e56acae | 1907 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
1908 | pgdat_resize_unlock(pgdat, &flags); |
1909 | ||
1910 | return nr_pages > 0; | |
1911 | } | |
1912 | ||
1913 | /* | |
1914 | * deferred_grow_zone() is __init, but it is called from | |
1915 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
1916 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
1917 | * and to ensure that the function body gets unloaded. | |
1918 | */ | |
1919 | static bool __ref | |
1920 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
1921 | { | |
1922 | return deferred_grow_zone(zone, order); | |
1923 | } | |
1924 | ||
7cf91a98 | 1925 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1926 | |
1927 | void __init page_alloc_init_late(void) | |
1928 | { | |
7cf91a98 | 1929 | struct zone *zone; |
e900a918 | 1930 | int nid; |
7cf91a98 JK |
1931 | |
1932 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 1933 | |
d3cd131d NS |
1934 | /* There will be num_node_state(N_MEMORY) threads */ |
1935 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1936 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1937 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1938 | } | |
1939 | ||
1940 | /* Block until all are initialised */ | |
d3cd131d | 1941 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 1942 | |
3e8fc007 MG |
1943 | /* |
1944 | * The number of managed pages has changed due to the initialisation | |
1945 | * so the pcpu batch and high limits needs to be updated or the limits | |
1946 | * will be artificially small. | |
1947 | */ | |
1948 | for_each_populated_zone(zone) | |
1949 | zone_pcp_update(zone); | |
1950 | ||
c9e97a19 PT |
1951 | /* |
1952 | * We initialized the rest of the deferred pages. Permanently disable | |
1953 | * on-demand struct page initialization. | |
1954 | */ | |
1955 | static_branch_disable(&deferred_pages); | |
1956 | ||
4248b0da MG |
1957 | /* Reinit limits that are based on free pages after the kernel is up */ |
1958 | files_maxfiles_init(); | |
7cf91a98 | 1959 | #endif |
350e88ba | 1960 | |
3010f876 PT |
1961 | /* Discard memblock private memory */ |
1962 | memblock_discard(); | |
7cf91a98 | 1963 | |
e900a918 DW |
1964 | for_each_node_state(nid, N_MEMORY) |
1965 | shuffle_free_memory(NODE_DATA(nid)); | |
1966 | ||
7cf91a98 JK |
1967 | for_each_populated_zone(zone) |
1968 | set_zone_contiguous(zone); | |
7e18adb4 | 1969 | } |
7e18adb4 | 1970 | |
47118af0 | 1971 | #ifdef CONFIG_CMA |
9cf510a5 | 1972 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1973 | void __init init_cma_reserved_pageblock(struct page *page) |
1974 | { | |
1975 | unsigned i = pageblock_nr_pages; | |
1976 | struct page *p = page; | |
1977 | ||
1978 | do { | |
1979 | __ClearPageReserved(p); | |
1980 | set_page_count(p, 0); | |
d883c6cf | 1981 | } while (++p, --i); |
47118af0 | 1982 | |
47118af0 | 1983 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1984 | |
1985 | if (pageblock_order >= MAX_ORDER) { | |
1986 | i = pageblock_nr_pages; | |
1987 | p = page; | |
1988 | do { | |
1989 | set_page_refcounted(p); | |
1990 | __free_pages(p, MAX_ORDER - 1); | |
1991 | p += MAX_ORDER_NR_PAGES; | |
1992 | } while (i -= MAX_ORDER_NR_PAGES); | |
1993 | } else { | |
1994 | set_page_refcounted(page); | |
1995 | __free_pages(page, pageblock_order); | |
1996 | } | |
1997 | ||
3dcc0571 | 1998 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1999 | } |
2000 | #endif | |
1da177e4 LT |
2001 | |
2002 | /* | |
2003 | * The order of subdivision here is critical for the IO subsystem. | |
2004 | * Please do not alter this order without good reasons and regression | |
2005 | * testing. Specifically, as large blocks of memory are subdivided, | |
2006 | * the order in which smaller blocks are delivered depends on the order | |
2007 | * they're subdivided in this function. This is the primary factor | |
2008 | * influencing the order in which pages are delivered to the IO | |
2009 | * subsystem according to empirical testing, and this is also justified | |
2010 | * by considering the behavior of a buddy system containing a single | |
2011 | * large block of memory acted on by a series of small allocations. | |
2012 | * This behavior is a critical factor in sglist merging's success. | |
2013 | * | |
6d49e352 | 2014 | * -- nyc |
1da177e4 | 2015 | */ |
085cc7d5 | 2016 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
2017 | int low, int high, struct free_area *area, |
2018 | int migratetype) | |
1da177e4 LT |
2019 | { |
2020 | unsigned long size = 1 << high; | |
2021 | ||
2022 | while (high > low) { | |
2023 | area--; | |
2024 | high--; | |
2025 | size >>= 1; | |
309381fe | 2026 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2027 | |
acbc15a4 JK |
2028 | /* |
2029 | * Mark as guard pages (or page), that will allow to | |
2030 | * merge back to allocator when buddy will be freed. | |
2031 | * Corresponding page table entries will not be touched, | |
2032 | * pages will stay not present in virtual address space | |
2033 | */ | |
2034 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2035 | continue; |
acbc15a4 | 2036 | |
b03641af | 2037 | add_to_free_area(&page[size], area, migratetype); |
1da177e4 LT |
2038 | set_page_order(&page[size], high); |
2039 | } | |
1da177e4 LT |
2040 | } |
2041 | ||
4e611801 | 2042 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2043 | { |
4e611801 VB |
2044 | const char *bad_reason = NULL; |
2045 | unsigned long bad_flags = 0; | |
7bfec6f4 | 2046 | |
53f9263b | 2047 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
2048 | bad_reason = "nonzero mapcount"; |
2049 | if (unlikely(page->mapping != NULL)) | |
2050 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 2051 | if (unlikely(page_ref_count(page) != 0)) |
136ac591 | 2052 | bad_reason = "nonzero _refcount"; |
f4c18e6f NH |
2053 | if (unlikely(page->flags & __PG_HWPOISON)) { |
2054 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
2055 | bad_flags = __PG_HWPOISON; | |
e570f56c NH |
2056 | /* Don't complain about hwpoisoned pages */ |
2057 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2058 | return; | |
f4c18e6f | 2059 | } |
f0b791a3 DH |
2060 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
2061 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
2062 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
2063 | } | |
9edad6ea JW |
2064 | #ifdef CONFIG_MEMCG |
2065 | if (unlikely(page->mem_cgroup)) | |
2066 | bad_reason = "page still charged to cgroup"; | |
2067 | #endif | |
4e611801 VB |
2068 | bad_page(page, bad_reason, bad_flags); |
2069 | } | |
2070 | ||
2071 | /* | |
2072 | * This page is about to be returned from the page allocator | |
2073 | */ | |
2074 | static inline int check_new_page(struct page *page) | |
2075 | { | |
2076 | if (likely(page_expected_state(page, | |
2077 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2078 | return 0; | |
2079 | ||
2080 | check_new_page_bad(page); | |
2081 | return 1; | |
2a7684a2 WF |
2082 | } |
2083 | ||
bd33ef36 | 2084 | static inline bool free_pages_prezeroed(void) |
1414c7f4 | 2085 | { |
6471384a AP |
2086 | return (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && |
2087 | page_poisoning_enabled()) || want_init_on_free(); | |
1414c7f4 LA |
2088 | } |
2089 | ||
479f854a | 2090 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2091 | /* |
2092 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2093 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2094 | * also checked when pcp lists are refilled from the free lists. | |
2095 | */ | |
2096 | static inline bool check_pcp_refill(struct page *page) | |
479f854a | 2097 | { |
8e57f8ac | 2098 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2099 | return check_new_page(page); |
2100 | else | |
2101 | return false; | |
479f854a MG |
2102 | } |
2103 | ||
4462b32c | 2104 | static inline bool check_new_pcp(struct page *page) |
479f854a MG |
2105 | { |
2106 | return check_new_page(page); | |
2107 | } | |
2108 | #else | |
4462b32c VB |
2109 | /* |
2110 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2111 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2112 | * enabled, they are also checked when being allocated from the pcp lists. | |
2113 | */ | |
2114 | static inline bool check_pcp_refill(struct page *page) | |
479f854a MG |
2115 | { |
2116 | return check_new_page(page); | |
2117 | } | |
4462b32c | 2118 | static inline bool check_new_pcp(struct page *page) |
479f854a | 2119 | { |
8e57f8ac | 2120 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2121 | return check_new_page(page); |
2122 | else | |
2123 | return false; | |
479f854a MG |
2124 | } |
2125 | #endif /* CONFIG_DEBUG_VM */ | |
2126 | ||
2127 | static bool check_new_pages(struct page *page, unsigned int order) | |
2128 | { | |
2129 | int i; | |
2130 | for (i = 0; i < (1 << order); i++) { | |
2131 | struct page *p = page + i; | |
2132 | ||
2133 | if (unlikely(check_new_page(p))) | |
2134 | return true; | |
2135 | } | |
2136 | ||
2137 | return false; | |
2138 | } | |
2139 | ||
46f24fd8 JK |
2140 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2141 | gfp_t gfp_flags) | |
2142 | { | |
2143 | set_page_private(page, 0); | |
2144 | set_page_refcounted(page); | |
2145 | ||
2146 | arch_alloc_page(page, order); | |
8e57f8ac | 2147 | if (debug_pagealloc_enabled_static()) |
d6332692 | 2148 | kernel_map_pages(page, 1 << order, 1); |
46f24fd8 | 2149 | kasan_alloc_pages(page, order); |
4117992d | 2150 | kernel_poison_pages(page, 1 << order, 1); |
46f24fd8 JK |
2151 | set_page_owner(page, order, gfp_flags); |
2152 | } | |
2153 | ||
479f854a | 2154 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2155 | unsigned int alloc_flags) |
2a7684a2 | 2156 | { |
46f24fd8 | 2157 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2158 | |
6471384a AP |
2159 | if (!free_pages_prezeroed() && want_init_on_alloc(gfp_flags)) |
2160 | kernel_init_free_pages(page, 1 << order); | |
17cf4406 NP |
2161 | |
2162 | if (order && (gfp_flags & __GFP_COMP)) | |
2163 | prep_compound_page(page, order); | |
2164 | ||
75379191 | 2165 | /* |
2f064f34 | 2166 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2167 | * allocate the page. The expectation is that the caller is taking |
2168 | * steps that will free more memory. The caller should avoid the page | |
2169 | * being used for !PFMEMALLOC purposes. | |
2170 | */ | |
2f064f34 MH |
2171 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2172 | set_page_pfmemalloc(page); | |
2173 | else | |
2174 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2175 | } |
2176 | ||
56fd56b8 MG |
2177 | /* |
2178 | * Go through the free lists for the given migratetype and remove | |
2179 | * the smallest available page from the freelists | |
2180 | */ | |
85ccc8fa | 2181 | static __always_inline |
728ec980 | 2182 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2183 | int migratetype) |
2184 | { | |
2185 | unsigned int current_order; | |
b8af2941 | 2186 | struct free_area *area; |
56fd56b8 MG |
2187 | struct page *page; |
2188 | ||
2189 | /* Find a page of the appropriate size in the preferred list */ | |
2190 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2191 | area = &(zone->free_area[current_order]); | |
b03641af | 2192 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2193 | if (!page) |
2194 | continue; | |
b03641af | 2195 | del_page_from_free_area(page, area); |
56fd56b8 | 2196 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 2197 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
2198 | return page; |
2199 | } | |
2200 | ||
2201 | return NULL; | |
2202 | } | |
2203 | ||
2204 | ||
b2a0ac88 MG |
2205 | /* |
2206 | * This array describes the order lists are fallen back to when | |
2207 | * the free lists for the desirable migrate type are depleted | |
2208 | */ | |
47118af0 | 2209 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e | 2210 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2211 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2212 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
47118af0 | 2213 | #ifdef CONFIG_CMA |
974a786e | 2214 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 2215 | #endif |
194159fb | 2216 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 2217 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 2218 | #endif |
b2a0ac88 MG |
2219 | }; |
2220 | ||
dc67647b | 2221 | #ifdef CONFIG_CMA |
85ccc8fa | 2222 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2223 | unsigned int order) |
2224 | { | |
2225 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2226 | } | |
2227 | #else | |
2228 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2229 | unsigned int order) { return NULL; } | |
2230 | #endif | |
2231 | ||
c361be55 MG |
2232 | /* |
2233 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 2234 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2235 | * boundary. If alignment is required, use move_freepages_block() |
2236 | */ | |
02aa0cdd | 2237 | static int move_freepages(struct zone *zone, |
b69a7288 | 2238 | struct page *start_page, struct page *end_page, |
02aa0cdd | 2239 | int migratetype, int *num_movable) |
c361be55 MG |
2240 | { |
2241 | struct page *page; | |
d00181b9 | 2242 | unsigned int order; |
d100313f | 2243 | int pages_moved = 0; |
c361be55 | 2244 | |
c361be55 MG |
2245 | for (page = start_page; page <= end_page;) { |
2246 | if (!pfn_valid_within(page_to_pfn(page))) { | |
2247 | page++; | |
2248 | continue; | |
2249 | } | |
2250 | ||
2251 | if (!PageBuddy(page)) { | |
02aa0cdd VB |
2252 | /* |
2253 | * We assume that pages that could be isolated for | |
2254 | * migration are movable. But we don't actually try | |
2255 | * isolating, as that would be expensive. | |
2256 | */ | |
2257 | if (num_movable && | |
2258 | (PageLRU(page) || __PageMovable(page))) | |
2259 | (*num_movable)++; | |
2260 | ||
c361be55 MG |
2261 | page++; |
2262 | continue; | |
2263 | } | |
2264 | ||
cd961038 DR |
2265 | /* Make sure we are not inadvertently changing nodes */ |
2266 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2267 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2268 | ||
c361be55 | 2269 | order = page_order(page); |
b03641af | 2270 | move_to_free_area(page, &zone->free_area[order], migratetype); |
c361be55 | 2271 | page += 1 << order; |
d100313f | 2272 | pages_moved += 1 << order; |
c361be55 MG |
2273 | } |
2274 | ||
d100313f | 2275 | return pages_moved; |
c361be55 MG |
2276 | } |
2277 | ||
ee6f509c | 2278 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2279 | int migratetype, int *num_movable) |
c361be55 MG |
2280 | { |
2281 | unsigned long start_pfn, end_pfn; | |
2282 | struct page *start_page, *end_page; | |
2283 | ||
4a222127 DR |
2284 | if (num_movable) |
2285 | *num_movable = 0; | |
2286 | ||
c361be55 | 2287 | start_pfn = page_to_pfn(page); |
d9c23400 | 2288 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 2289 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
2290 | end_page = start_page + pageblock_nr_pages - 1; |
2291 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
2292 | |
2293 | /* Do not cross zone boundaries */ | |
108bcc96 | 2294 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 2295 | start_page = page; |
108bcc96 | 2296 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2297 | return 0; |
2298 | ||
02aa0cdd VB |
2299 | return move_freepages(zone, start_page, end_page, migratetype, |
2300 | num_movable); | |
c361be55 MG |
2301 | } |
2302 | ||
2f66a68f MG |
2303 | static void change_pageblock_range(struct page *pageblock_page, |
2304 | int start_order, int migratetype) | |
2305 | { | |
2306 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2307 | ||
2308 | while (nr_pageblocks--) { | |
2309 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2310 | pageblock_page += pageblock_nr_pages; | |
2311 | } | |
2312 | } | |
2313 | ||
fef903ef | 2314 | /* |
9c0415eb VB |
2315 | * When we are falling back to another migratetype during allocation, try to |
2316 | * steal extra free pages from the same pageblocks to satisfy further | |
2317 | * allocations, instead of polluting multiple pageblocks. | |
2318 | * | |
2319 | * If we are stealing a relatively large buddy page, it is likely there will | |
2320 | * be more free pages in the pageblock, so try to steal them all. For | |
2321 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2322 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2323 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2324 | * pageblocks. | |
fef903ef | 2325 | */ |
4eb7dce6 JK |
2326 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2327 | { | |
2328 | /* | |
2329 | * Leaving this order check is intended, although there is | |
2330 | * relaxed order check in next check. The reason is that | |
2331 | * we can actually steal whole pageblock if this condition met, | |
2332 | * but, below check doesn't guarantee it and that is just heuristic | |
2333 | * so could be changed anytime. | |
2334 | */ | |
2335 | if (order >= pageblock_order) | |
2336 | return true; | |
2337 | ||
2338 | if (order >= pageblock_order / 2 || | |
2339 | start_mt == MIGRATE_RECLAIMABLE || | |
2340 | start_mt == MIGRATE_UNMOVABLE || | |
2341 | page_group_by_mobility_disabled) | |
2342 | return true; | |
2343 | ||
2344 | return false; | |
2345 | } | |
2346 | ||
1c30844d MG |
2347 | static inline void boost_watermark(struct zone *zone) |
2348 | { | |
2349 | unsigned long max_boost; | |
2350 | ||
2351 | if (!watermark_boost_factor) | |
2352 | return; | |
2353 | ||
2354 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2355 | watermark_boost_factor, 10000); | |
94b3334c MG |
2356 | |
2357 | /* | |
2358 | * high watermark may be uninitialised if fragmentation occurs | |
2359 | * very early in boot so do not boost. We do not fall | |
2360 | * through and boost by pageblock_nr_pages as failing | |
2361 | * allocations that early means that reclaim is not going | |
2362 | * to help and it may even be impossible to reclaim the | |
2363 | * boosted watermark resulting in a hang. | |
2364 | */ | |
2365 | if (!max_boost) | |
2366 | return; | |
2367 | ||
1c30844d MG |
2368 | max_boost = max(pageblock_nr_pages, max_boost); |
2369 | ||
2370 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2371 | max_boost); | |
2372 | } | |
2373 | ||
4eb7dce6 JK |
2374 | /* |
2375 | * This function implements actual steal behaviour. If order is large enough, | |
2376 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2377 | * pageblock to our migratetype and determine how many already-allocated pages |
2378 | * are there in the pageblock with a compatible migratetype. If at least half | |
2379 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2380 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2381 | */ |
2382 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2383 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2384 | { |
d00181b9 | 2385 | unsigned int current_order = page_order(page); |
3bc48f96 | 2386 | struct free_area *area; |
02aa0cdd VB |
2387 | int free_pages, movable_pages, alike_pages; |
2388 | int old_block_type; | |
2389 | ||
2390 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2391 | |
3bc48f96 VB |
2392 | /* |
2393 | * This can happen due to races and we want to prevent broken | |
2394 | * highatomic accounting. | |
2395 | */ | |
02aa0cdd | 2396 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2397 | goto single_page; |
2398 | ||
fef903ef SB |
2399 | /* Take ownership for orders >= pageblock_order */ |
2400 | if (current_order >= pageblock_order) { | |
2401 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2402 | goto single_page; |
fef903ef SB |
2403 | } |
2404 | ||
1c30844d MG |
2405 | /* |
2406 | * Boost watermarks to increase reclaim pressure to reduce the | |
2407 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2408 | * may be balanced overall and kswapd will not wake naturally. | |
2409 | */ | |
2410 | boost_watermark(zone); | |
2411 | if (alloc_flags & ALLOC_KSWAPD) | |
73444bc4 | 2412 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2413 | |
3bc48f96 VB |
2414 | /* We are not allowed to try stealing from the whole block */ |
2415 | if (!whole_block) | |
2416 | goto single_page; | |
2417 | ||
02aa0cdd VB |
2418 | free_pages = move_freepages_block(zone, page, start_type, |
2419 | &movable_pages); | |
2420 | /* | |
2421 | * Determine how many pages are compatible with our allocation. | |
2422 | * For movable allocation, it's the number of movable pages which | |
2423 | * we just obtained. For other types it's a bit more tricky. | |
2424 | */ | |
2425 | if (start_type == MIGRATE_MOVABLE) { | |
2426 | alike_pages = movable_pages; | |
2427 | } else { | |
2428 | /* | |
2429 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2430 | * to MOVABLE pageblock, consider all non-movable pages as | |
2431 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2432 | * vice versa, be conservative since we can't distinguish the | |
2433 | * exact migratetype of non-movable pages. | |
2434 | */ | |
2435 | if (old_block_type == MIGRATE_MOVABLE) | |
2436 | alike_pages = pageblock_nr_pages | |
2437 | - (free_pages + movable_pages); | |
2438 | else | |
2439 | alike_pages = 0; | |
2440 | } | |
2441 | ||
3bc48f96 | 2442 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2443 | if (!free_pages) |
3bc48f96 | 2444 | goto single_page; |
fef903ef | 2445 | |
02aa0cdd VB |
2446 | /* |
2447 | * If a sufficient number of pages in the block are either free or of | |
2448 | * comparable migratability as our allocation, claim the whole block. | |
2449 | */ | |
2450 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2451 | page_group_by_mobility_disabled) |
2452 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2453 | |
2454 | return; | |
2455 | ||
2456 | single_page: | |
2457 | area = &zone->free_area[current_order]; | |
b03641af | 2458 | move_to_free_area(page, area, start_type); |
4eb7dce6 JK |
2459 | } |
2460 | ||
2149cdae JK |
2461 | /* |
2462 | * Check whether there is a suitable fallback freepage with requested order. | |
2463 | * If only_stealable is true, this function returns fallback_mt only if | |
2464 | * we can steal other freepages all together. This would help to reduce | |
2465 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2466 | */ | |
2467 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2468 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2469 | { |
2470 | int i; | |
2471 | int fallback_mt; | |
2472 | ||
2473 | if (area->nr_free == 0) | |
2474 | return -1; | |
2475 | ||
2476 | *can_steal = false; | |
2477 | for (i = 0;; i++) { | |
2478 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2479 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2480 | break; |
2481 | ||
b03641af | 2482 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2483 | continue; |
fef903ef | 2484 | |
4eb7dce6 JK |
2485 | if (can_steal_fallback(order, migratetype)) |
2486 | *can_steal = true; | |
2487 | ||
2149cdae JK |
2488 | if (!only_stealable) |
2489 | return fallback_mt; | |
2490 | ||
2491 | if (*can_steal) | |
2492 | return fallback_mt; | |
fef903ef | 2493 | } |
4eb7dce6 JK |
2494 | |
2495 | return -1; | |
fef903ef SB |
2496 | } |
2497 | ||
0aaa29a5 MG |
2498 | /* |
2499 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2500 | * there are no empty page blocks that contain a page with a suitable order | |
2501 | */ | |
2502 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2503 | unsigned int alloc_order) | |
2504 | { | |
2505 | int mt; | |
2506 | unsigned long max_managed, flags; | |
2507 | ||
2508 | /* | |
2509 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2510 | * Check is race-prone but harmless. | |
2511 | */ | |
9705bea5 | 2512 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2513 | if (zone->nr_reserved_highatomic >= max_managed) |
2514 | return; | |
2515 | ||
2516 | spin_lock_irqsave(&zone->lock, flags); | |
2517 | ||
2518 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2519 | if (zone->nr_reserved_highatomic >= max_managed) | |
2520 | goto out_unlock; | |
2521 | ||
2522 | /* Yoink! */ | |
2523 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2524 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2525 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2526 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2527 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2528 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2529 | } |
2530 | ||
2531 | out_unlock: | |
2532 | spin_unlock_irqrestore(&zone->lock, flags); | |
2533 | } | |
2534 | ||
2535 | /* | |
2536 | * Used when an allocation is about to fail under memory pressure. This | |
2537 | * potentially hurts the reliability of high-order allocations when under | |
2538 | * intense memory pressure but failed atomic allocations should be easier | |
2539 | * to recover from than an OOM. | |
29fac03b MK |
2540 | * |
2541 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2542 | * pageblock is exhausted. | |
0aaa29a5 | 2543 | */ |
29fac03b MK |
2544 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2545 | bool force) | |
0aaa29a5 MG |
2546 | { |
2547 | struct zonelist *zonelist = ac->zonelist; | |
2548 | unsigned long flags; | |
2549 | struct zoneref *z; | |
2550 | struct zone *zone; | |
2551 | struct page *page; | |
2552 | int order; | |
04c8716f | 2553 | bool ret; |
0aaa29a5 MG |
2554 | |
2555 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2556 | ac->nodemask) { | |
29fac03b MK |
2557 | /* |
2558 | * Preserve at least one pageblock unless memory pressure | |
2559 | * is really high. | |
2560 | */ | |
2561 | if (!force && zone->nr_reserved_highatomic <= | |
2562 | pageblock_nr_pages) | |
0aaa29a5 MG |
2563 | continue; |
2564 | ||
2565 | spin_lock_irqsave(&zone->lock, flags); | |
2566 | for (order = 0; order < MAX_ORDER; order++) { | |
2567 | struct free_area *area = &(zone->free_area[order]); | |
2568 | ||
b03641af | 2569 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2570 | if (!page) |
0aaa29a5 MG |
2571 | continue; |
2572 | ||
0aaa29a5 | 2573 | /* |
4855e4a7 MK |
2574 | * In page freeing path, migratetype change is racy so |
2575 | * we can counter several free pages in a pageblock | |
2576 | * in this loop althoug we changed the pageblock type | |
2577 | * from highatomic to ac->migratetype. So we should | |
2578 | * adjust the count once. | |
0aaa29a5 | 2579 | */ |
a6ffdc07 | 2580 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2581 | /* |
2582 | * It should never happen but changes to | |
2583 | * locking could inadvertently allow a per-cpu | |
2584 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2585 | * while unreserving so be safe and watch for | |
2586 | * underflows. | |
2587 | */ | |
2588 | zone->nr_reserved_highatomic -= min( | |
2589 | pageblock_nr_pages, | |
2590 | zone->nr_reserved_highatomic); | |
2591 | } | |
0aaa29a5 MG |
2592 | |
2593 | /* | |
2594 | * Convert to ac->migratetype and avoid the normal | |
2595 | * pageblock stealing heuristics. Minimally, the caller | |
2596 | * is doing the work and needs the pages. More | |
2597 | * importantly, if the block was always converted to | |
2598 | * MIGRATE_UNMOVABLE or another type then the number | |
2599 | * of pageblocks that cannot be completely freed | |
2600 | * may increase. | |
2601 | */ | |
2602 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2603 | ret = move_freepages_block(zone, page, ac->migratetype, |
2604 | NULL); | |
29fac03b MK |
2605 | if (ret) { |
2606 | spin_unlock_irqrestore(&zone->lock, flags); | |
2607 | return ret; | |
2608 | } | |
0aaa29a5 MG |
2609 | } |
2610 | spin_unlock_irqrestore(&zone->lock, flags); | |
2611 | } | |
04c8716f MK |
2612 | |
2613 | return false; | |
0aaa29a5 MG |
2614 | } |
2615 | ||
3bc48f96 VB |
2616 | /* |
2617 | * Try finding a free buddy page on the fallback list and put it on the free | |
2618 | * list of requested migratetype, possibly along with other pages from the same | |
2619 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2620 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2621 | * |
2622 | * The use of signed ints for order and current_order is a deliberate | |
2623 | * deviation from the rest of this file, to make the for loop | |
2624 | * condition simpler. | |
3bc48f96 | 2625 | */ |
85ccc8fa | 2626 | static __always_inline bool |
6bb15450 MG |
2627 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2628 | unsigned int alloc_flags) | |
b2a0ac88 | 2629 | { |
b8af2941 | 2630 | struct free_area *area; |
b002529d | 2631 | int current_order; |
6bb15450 | 2632 | int min_order = order; |
b2a0ac88 | 2633 | struct page *page; |
4eb7dce6 JK |
2634 | int fallback_mt; |
2635 | bool can_steal; | |
b2a0ac88 | 2636 | |
6bb15450 MG |
2637 | /* |
2638 | * Do not steal pages from freelists belonging to other pageblocks | |
2639 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2640 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2641 | */ | |
2642 | if (alloc_flags & ALLOC_NOFRAGMENT) | |
2643 | min_order = pageblock_order; | |
2644 | ||
7a8f58f3 VB |
2645 | /* |
2646 | * Find the largest available free page in the other list. This roughly | |
2647 | * approximates finding the pageblock with the most free pages, which | |
2648 | * would be too costly to do exactly. | |
2649 | */ | |
6bb15450 | 2650 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 2651 | --current_order) { |
4eb7dce6 JK |
2652 | area = &(zone->free_area[current_order]); |
2653 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2654 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2655 | if (fallback_mt == -1) |
2656 | continue; | |
b2a0ac88 | 2657 | |
7a8f58f3 VB |
2658 | /* |
2659 | * We cannot steal all free pages from the pageblock and the | |
2660 | * requested migratetype is movable. In that case it's better to | |
2661 | * steal and split the smallest available page instead of the | |
2662 | * largest available page, because even if the next movable | |
2663 | * allocation falls back into a different pageblock than this | |
2664 | * one, it won't cause permanent fragmentation. | |
2665 | */ | |
2666 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2667 | && current_order > order) | |
2668 | goto find_smallest; | |
b2a0ac88 | 2669 | |
7a8f58f3 VB |
2670 | goto do_steal; |
2671 | } | |
e0fff1bd | 2672 | |
7a8f58f3 | 2673 | return false; |
e0fff1bd | 2674 | |
7a8f58f3 VB |
2675 | find_smallest: |
2676 | for (current_order = order; current_order < MAX_ORDER; | |
2677 | current_order++) { | |
2678 | area = &(zone->free_area[current_order]); | |
2679 | fallback_mt = find_suitable_fallback(area, current_order, | |
2680 | start_migratetype, false, &can_steal); | |
2681 | if (fallback_mt != -1) | |
2682 | break; | |
b2a0ac88 MG |
2683 | } |
2684 | ||
7a8f58f3 VB |
2685 | /* |
2686 | * This should not happen - we already found a suitable fallback | |
2687 | * when looking for the largest page. | |
2688 | */ | |
2689 | VM_BUG_ON(current_order == MAX_ORDER); | |
2690 | ||
2691 | do_steal: | |
b03641af | 2692 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2693 | |
1c30844d MG |
2694 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
2695 | can_steal); | |
7a8f58f3 VB |
2696 | |
2697 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2698 | start_migratetype, fallback_mt); | |
2699 | ||
2700 | return true; | |
2701 | ||
b2a0ac88 MG |
2702 | } |
2703 | ||
56fd56b8 | 2704 | /* |
1da177e4 LT |
2705 | * Do the hard work of removing an element from the buddy allocator. |
2706 | * Call me with the zone->lock already held. | |
2707 | */ | |
85ccc8fa | 2708 | static __always_inline struct page * |
6bb15450 MG |
2709 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2710 | unsigned int alloc_flags) | |
1da177e4 | 2711 | { |
1da177e4 LT |
2712 | struct page *page; |
2713 | ||
3bc48f96 | 2714 | retry: |
56fd56b8 | 2715 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2716 | if (unlikely(!page)) { |
dc67647b JK |
2717 | if (migratetype == MIGRATE_MOVABLE) |
2718 | page = __rmqueue_cma_fallback(zone, order); | |
2719 | ||
6bb15450 MG |
2720 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2721 | alloc_flags)) | |
3bc48f96 | 2722 | goto retry; |
728ec980 MG |
2723 | } |
2724 | ||
0d3d062a | 2725 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2726 | return page; |
1da177e4 LT |
2727 | } |
2728 | ||
5f63b720 | 2729 | /* |
1da177e4 LT |
2730 | * Obtain a specified number of elements from the buddy allocator, all under |
2731 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2732 | * Returns the number of new pages which were placed at *list. | |
2733 | */ | |
5f63b720 | 2734 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2735 | unsigned long count, struct list_head *list, |
6bb15450 | 2736 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2737 | { |
a6de734b | 2738 | int i, alloced = 0; |
5f63b720 | 2739 | |
d34b0733 | 2740 | spin_lock(&zone->lock); |
1da177e4 | 2741 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2742 | struct page *page = __rmqueue(zone, order, migratetype, |
2743 | alloc_flags); | |
085cc7d5 | 2744 | if (unlikely(page == NULL)) |
1da177e4 | 2745 | break; |
81eabcbe | 2746 | |
479f854a MG |
2747 | if (unlikely(check_pcp_refill(page))) |
2748 | continue; | |
2749 | ||
81eabcbe | 2750 | /* |
0fac3ba5 VB |
2751 | * Split buddy pages returned by expand() are received here in |
2752 | * physical page order. The page is added to the tail of | |
2753 | * caller's list. From the callers perspective, the linked list | |
2754 | * is ordered by page number under some conditions. This is | |
2755 | * useful for IO devices that can forward direction from the | |
2756 | * head, thus also in the physical page order. This is useful | |
2757 | * for IO devices that can merge IO requests if the physical | |
2758 | * pages are ordered properly. | |
81eabcbe | 2759 | */ |
0fac3ba5 | 2760 | list_add_tail(&page->lru, list); |
a6de734b | 2761 | alloced++; |
bb14c2c7 | 2762 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2763 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2764 | -(1 << order)); | |
1da177e4 | 2765 | } |
a6de734b MG |
2766 | |
2767 | /* | |
2768 | * i pages were removed from the buddy list even if some leak due | |
2769 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2770 | * on i. Do not confuse with 'alloced' which is the number of | |
2771 | * pages added to the pcp list. | |
2772 | */ | |
f2260e6b | 2773 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2774 | spin_unlock(&zone->lock); |
a6de734b | 2775 | return alloced; |
1da177e4 LT |
2776 | } |
2777 | ||
4ae7c039 | 2778 | #ifdef CONFIG_NUMA |
8fce4d8e | 2779 | /* |
4037d452 CL |
2780 | * Called from the vmstat counter updater to drain pagesets of this |
2781 | * currently executing processor on remote nodes after they have | |
2782 | * expired. | |
2783 | * | |
879336c3 CL |
2784 | * Note that this function must be called with the thread pinned to |
2785 | * a single processor. | |
8fce4d8e | 2786 | */ |
4037d452 | 2787 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2788 | { |
4ae7c039 | 2789 | unsigned long flags; |
7be12fc9 | 2790 | int to_drain, batch; |
4ae7c039 | 2791 | |
4037d452 | 2792 | local_irq_save(flags); |
4db0c3c2 | 2793 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2794 | to_drain = min(pcp->count, batch); |
77ba9062 | 2795 | if (to_drain > 0) |
2a13515c | 2796 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 2797 | local_irq_restore(flags); |
4ae7c039 CL |
2798 | } |
2799 | #endif | |
2800 | ||
9f8f2172 | 2801 | /* |
93481ff0 | 2802 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2803 | * |
2804 | * The processor must either be the current processor and the | |
2805 | * thread pinned to the current processor or a processor that | |
2806 | * is not online. | |
2807 | */ | |
93481ff0 | 2808 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2809 | { |
c54ad30c | 2810 | unsigned long flags; |
93481ff0 VB |
2811 | struct per_cpu_pageset *pset; |
2812 | struct per_cpu_pages *pcp; | |
1da177e4 | 2813 | |
93481ff0 VB |
2814 | local_irq_save(flags); |
2815 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2816 | |
93481ff0 | 2817 | pcp = &pset->pcp; |
77ba9062 | 2818 | if (pcp->count) |
93481ff0 | 2819 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
2820 | local_irq_restore(flags); |
2821 | } | |
3dfa5721 | 2822 | |
93481ff0 VB |
2823 | /* |
2824 | * Drain pcplists of all zones on the indicated processor. | |
2825 | * | |
2826 | * The processor must either be the current processor and the | |
2827 | * thread pinned to the current processor or a processor that | |
2828 | * is not online. | |
2829 | */ | |
2830 | static void drain_pages(unsigned int cpu) | |
2831 | { | |
2832 | struct zone *zone; | |
2833 | ||
2834 | for_each_populated_zone(zone) { | |
2835 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2836 | } |
2837 | } | |
1da177e4 | 2838 | |
9f8f2172 CL |
2839 | /* |
2840 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2841 | * |
2842 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2843 | * the single zone's pages. | |
9f8f2172 | 2844 | */ |
93481ff0 | 2845 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2846 | { |
93481ff0 VB |
2847 | int cpu = smp_processor_id(); |
2848 | ||
2849 | if (zone) | |
2850 | drain_pages_zone(cpu, zone); | |
2851 | else | |
2852 | drain_pages(cpu); | |
9f8f2172 CL |
2853 | } |
2854 | ||
0ccce3b9 MG |
2855 | static void drain_local_pages_wq(struct work_struct *work) |
2856 | { | |
d9367bd0 WY |
2857 | struct pcpu_drain *drain; |
2858 | ||
2859 | drain = container_of(work, struct pcpu_drain, work); | |
2860 | ||
a459eeb7 MH |
2861 | /* |
2862 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2863 | * we can race with cpu offline when the WQ can move this from | |
2864 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2865 | * cpu which is allright but we also have to make sure to not move to | |
2866 | * a different one. | |
2867 | */ | |
2868 | preempt_disable(); | |
d9367bd0 | 2869 | drain_local_pages(drain->zone); |
a459eeb7 | 2870 | preempt_enable(); |
0ccce3b9 MG |
2871 | } |
2872 | ||
9f8f2172 | 2873 | /* |
74046494 GBY |
2874 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2875 | * | |
93481ff0 VB |
2876 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2877 | * | |
0ccce3b9 | 2878 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2879 | */ |
93481ff0 | 2880 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2881 | { |
74046494 | 2882 | int cpu; |
74046494 GBY |
2883 | |
2884 | /* | |
2885 | * Allocate in the BSS so we wont require allocation in | |
2886 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2887 | */ | |
2888 | static cpumask_t cpus_with_pcps; | |
2889 | ||
ce612879 MH |
2890 | /* |
2891 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2892 | * initialized. | |
2893 | */ | |
2894 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2895 | return; | |
2896 | ||
bd233f53 MG |
2897 | /* |
2898 | * Do not drain if one is already in progress unless it's specific to | |
2899 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2900 | * the drain to be complete when the call returns. | |
2901 | */ | |
2902 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2903 | if (!zone) | |
2904 | return; | |
2905 | mutex_lock(&pcpu_drain_mutex); | |
2906 | } | |
0ccce3b9 | 2907 | |
74046494 GBY |
2908 | /* |
2909 | * We don't care about racing with CPU hotplug event | |
2910 | * as offline notification will cause the notified | |
2911 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2912 | * disables preemption as part of its processing | |
2913 | */ | |
2914 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2915 | struct per_cpu_pageset *pcp; |
2916 | struct zone *z; | |
74046494 | 2917 | bool has_pcps = false; |
93481ff0 VB |
2918 | |
2919 | if (zone) { | |
74046494 | 2920 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2921 | if (pcp->pcp.count) |
74046494 | 2922 | has_pcps = true; |
93481ff0 VB |
2923 | } else { |
2924 | for_each_populated_zone(z) { | |
2925 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2926 | if (pcp->pcp.count) { | |
2927 | has_pcps = true; | |
2928 | break; | |
2929 | } | |
74046494 GBY |
2930 | } |
2931 | } | |
93481ff0 | 2932 | |
74046494 GBY |
2933 | if (has_pcps) |
2934 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2935 | else | |
2936 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2937 | } | |
0ccce3b9 | 2938 | |
bd233f53 | 2939 | for_each_cpu(cpu, &cpus_with_pcps) { |
d9367bd0 WY |
2940 | struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu); |
2941 | ||
2942 | drain->zone = zone; | |
2943 | INIT_WORK(&drain->work, drain_local_pages_wq); | |
2944 | queue_work_on(cpu, mm_percpu_wq, &drain->work); | |
0ccce3b9 | 2945 | } |
bd233f53 | 2946 | for_each_cpu(cpu, &cpus_with_pcps) |
d9367bd0 | 2947 | flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work); |
bd233f53 MG |
2948 | |
2949 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
2950 | } |
2951 | ||
296699de | 2952 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 2953 | |
556b969a CY |
2954 | /* |
2955 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
2956 | */ | |
2957 | #define WD_PAGE_COUNT (128*1024) | |
2958 | ||
1da177e4 LT |
2959 | void mark_free_pages(struct zone *zone) |
2960 | { | |
556b969a | 2961 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 2962 | unsigned long flags; |
7aeb09f9 | 2963 | unsigned int order, t; |
86760a2c | 2964 | struct page *page; |
1da177e4 | 2965 | |
8080fc03 | 2966 | if (zone_is_empty(zone)) |
1da177e4 LT |
2967 | return; |
2968 | ||
2969 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2970 | |
108bcc96 | 2971 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2972 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2973 | if (pfn_valid(pfn)) { | |
86760a2c | 2974 | page = pfn_to_page(pfn); |
ba6b0979 | 2975 | |
556b969a CY |
2976 | if (!--page_count) { |
2977 | touch_nmi_watchdog(); | |
2978 | page_count = WD_PAGE_COUNT; | |
2979 | } | |
2980 | ||
ba6b0979 JK |
2981 | if (page_zone(page) != zone) |
2982 | continue; | |
2983 | ||
7be98234 RW |
2984 | if (!swsusp_page_is_forbidden(page)) |
2985 | swsusp_unset_page_free(page); | |
f623f0db | 2986 | } |
1da177e4 | 2987 | |
b2a0ac88 | 2988 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2989 | list_for_each_entry(page, |
2990 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2991 | unsigned long i; |
1da177e4 | 2992 | |
86760a2c | 2993 | pfn = page_to_pfn(page); |
556b969a CY |
2994 | for (i = 0; i < (1UL << order); i++) { |
2995 | if (!--page_count) { | |
2996 | touch_nmi_watchdog(); | |
2997 | page_count = WD_PAGE_COUNT; | |
2998 | } | |
7be98234 | 2999 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 3000 | } |
f623f0db | 3001 | } |
b2a0ac88 | 3002 | } |
1da177e4 LT |
3003 | spin_unlock_irqrestore(&zone->lock, flags); |
3004 | } | |
e2c55dc8 | 3005 | #endif /* CONFIG_PM */ |
1da177e4 | 3006 | |
2d4894b5 | 3007 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 3008 | { |
5f8dcc21 | 3009 | int migratetype; |
1da177e4 | 3010 | |
4db7548c | 3011 | if (!free_pcp_prepare(page)) |
9cca35d4 | 3012 | return false; |
689bcebf | 3013 | |
dc4b0caf | 3014 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 3015 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
3016 | return true; |
3017 | } | |
3018 | ||
2d4894b5 | 3019 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
3020 | { |
3021 | struct zone *zone = page_zone(page); | |
3022 | struct per_cpu_pages *pcp; | |
3023 | int migratetype; | |
3024 | ||
3025 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 3026 | __count_vm_event(PGFREE); |
da456f14 | 3027 | |
5f8dcc21 MG |
3028 | /* |
3029 | * We only track unmovable, reclaimable and movable on pcp lists. | |
3030 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 3031 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
3032 | * areas back if necessary. Otherwise, we may have to free |
3033 | * excessively into the page allocator | |
3034 | */ | |
3035 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 3036 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 3037 | free_one_page(zone, page, pfn, 0, migratetype); |
9cca35d4 | 3038 | return; |
5f8dcc21 MG |
3039 | } |
3040 | migratetype = MIGRATE_MOVABLE; | |
3041 | } | |
3042 | ||
99dcc3e5 | 3043 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 3044 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 3045 | pcp->count++; |
48db57f8 | 3046 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 3047 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb | 3048 | free_pcppages_bulk(zone, batch, pcp); |
48db57f8 | 3049 | } |
9cca35d4 | 3050 | } |
5f8dcc21 | 3051 | |
9cca35d4 MG |
3052 | /* |
3053 | * Free a 0-order page | |
9cca35d4 | 3054 | */ |
2d4894b5 | 3055 | void free_unref_page(struct page *page) |
9cca35d4 MG |
3056 | { |
3057 | unsigned long flags; | |
3058 | unsigned long pfn = page_to_pfn(page); | |
3059 | ||
2d4894b5 | 3060 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3061 | return; |
3062 | ||
3063 | local_irq_save(flags); | |
2d4894b5 | 3064 | free_unref_page_commit(page, pfn); |
d34b0733 | 3065 | local_irq_restore(flags); |
1da177e4 LT |
3066 | } |
3067 | ||
cc59850e KK |
3068 | /* |
3069 | * Free a list of 0-order pages | |
3070 | */ | |
2d4894b5 | 3071 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
3072 | { |
3073 | struct page *page, *next; | |
9cca35d4 | 3074 | unsigned long flags, pfn; |
c24ad77d | 3075 | int batch_count = 0; |
9cca35d4 MG |
3076 | |
3077 | /* Prepare pages for freeing */ | |
3078 | list_for_each_entry_safe(page, next, list, lru) { | |
3079 | pfn = page_to_pfn(page); | |
2d4894b5 | 3080 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3081 | list_del(&page->lru); |
3082 | set_page_private(page, pfn); | |
3083 | } | |
cc59850e | 3084 | |
9cca35d4 | 3085 | local_irq_save(flags); |
cc59850e | 3086 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
3087 | unsigned long pfn = page_private(page); |
3088 | ||
3089 | set_page_private(page, 0); | |
2d4894b5 MG |
3090 | trace_mm_page_free_batched(page); |
3091 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
3092 | |
3093 | /* | |
3094 | * Guard against excessive IRQ disabled times when we get | |
3095 | * a large list of pages to free. | |
3096 | */ | |
3097 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
3098 | local_irq_restore(flags); | |
3099 | batch_count = 0; | |
3100 | local_irq_save(flags); | |
3101 | } | |
cc59850e | 3102 | } |
9cca35d4 | 3103 | local_irq_restore(flags); |
cc59850e KK |
3104 | } |
3105 | ||
8dfcc9ba NP |
3106 | /* |
3107 | * split_page takes a non-compound higher-order page, and splits it into | |
3108 | * n (1<<order) sub-pages: page[0..n] | |
3109 | * Each sub-page must be freed individually. | |
3110 | * | |
3111 | * Note: this is probably too low level an operation for use in drivers. | |
3112 | * Please consult with lkml before using this in your driver. | |
3113 | */ | |
3114 | void split_page(struct page *page, unsigned int order) | |
3115 | { | |
3116 | int i; | |
3117 | ||
309381fe SL |
3118 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3119 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3120 | |
a9627bc5 | 3121 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3122 | set_page_refcounted(page + i); |
a9627bc5 | 3123 | split_page_owner(page, order); |
8dfcc9ba | 3124 | } |
5853ff23 | 3125 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3126 | |
3c605096 | 3127 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3128 | { |
b03641af | 3129 | struct free_area *area = &page_zone(page)->free_area[order]; |
748446bb MG |
3130 | unsigned long watermark; |
3131 | struct zone *zone; | |
2139cbe6 | 3132 | int mt; |
748446bb MG |
3133 | |
3134 | BUG_ON(!PageBuddy(page)); | |
3135 | ||
3136 | zone = page_zone(page); | |
2e30abd1 | 3137 | mt = get_pageblock_migratetype(page); |
748446bb | 3138 | |
194159fb | 3139 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
3140 | /* |
3141 | * Obey watermarks as if the page was being allocated. We can | |
3142 | * emulate a high-order watermark check with a raised order-0 | |
3143 | * watermark, because we already know our high-order page | |
3144 | * exists. | |
3145 | */ | |
fd1444b2 | 3146 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3147 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3148 | return 0; |
3149 | ||
8fb74b9f | 3150 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3151 | } |
748446bb MG |
3152 | |
3153 | /* Remove page from free list */ | |
b03641af DW |
3154 | |
3155 | del_page_from_free_area(page, area); | |
2139cbe6 | 3156 | |
400bc7fd | 3157 | /* |
3158 | * Set the pageblock if the isolated page is at least half of a | |
3159 | * pageblock | |
3160 | */ | |
748446bb MG |
3161 | if (order >= pageblock_order - 1) { |
3162 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3163 | for (; page < endpage; page += pageblock_nr_pages) { |
3164 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 3165 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 3166 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
3167 | set_pageblock_migratetype(page, |
3168 | MIGRATE_MOVABLE); | |
3169 | } | |
748446bb MG |
3170 | } |
3171 | ||
f3a14ced | 3172 | |
8fb74b9f | 3173 | return 1UL << order; |
1fb3f8ca MG |
3174 | } |
3175 | ||
060e7417 MG |
3176 | /* |
3177 | * Update NUMA hit/miss statistics | |
3178 | * | |
3179 | * Must be called with interrupts disabled. | |
060e7417 | 3180 | */ |
41b6167e | 3181 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
3182 | { |
3183 | #ifdef CONFIG_NUMA | |
3a321d2a | 3184 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3185 | |
4518085e KW |
3186 | /* skip numa counters update if numa stats is disabled */ |
3187 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3188 | return; | |
3189 | ||
c1093b74 | 3190 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3191 | local_stat = NUMA_OTHER; |
060e7417 | 3192 | |
c1093b74 | 3193 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3a321d2a | 3194 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 3195 | else { |
3a321d2a KW |
3196 | __inc_numa_state(z, NUMA_MISS); |
3197 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 3198 | } |
3a321d2a | 3199 | __inc_numa_state(z, local_stat); |
060e7417 MG |
3200 | #endif |
3201 | } | |
3202 | ||
066b2393 MG |
3203 | /* Remove page from the per-cpu list, caller must protect the list */ |
3204 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
6bb15450 | 3205 | unsigned int alloc_flags, |
453f85d4 | 3206 | struct per_cpu_pages *pcp, |
066b2393 MG |
3207 | struct list_head *list) |
3208 | { | |
3209 | struct page *page; | |
3210 | ||
3211 | do { | |
3212 | if (list_empty(list)) { | |
3213 | pcp->count += rmqueue_bulk(zone, 0, | |
3214 | pcp->batch, list, | |
6bb15450 | 3215 | migratetype, alloc_flags); |
066b2393 MG |
3216 | if (unlikely(list_empty(list))) |
3217 | return NULL; | |
3218 | } | |
3219 | ||
453f85d4 | 3220 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
3221 | list_del(&page->lru); |
3222 | pcp->count--; | |
3223 | } while (check_new_pcp(page)); | |
3224 | ||
3225 | return page; | |
3226 | } | |
3227 | ||
3228 | /* Lock and remove page from the per-cpu list */ | |
3229 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
1c52e6d0 YS |
3230 | struct zone *zone, gfp_t gfp_flags, |
3231 | int migratetype, unsigned int alloc_flags) | |
066b2393 MG |
3232 | { |
3233 | struct per_cpu_pages *pcp; | |
3234 | struct list_head *list; | |
066b2393 | 3235 | struct page *page; |
d34b0733 | 3236 | unsigned long flags; |
066b2393 | 3237 | |
d34b0733 | 3238 | local_irq_save(flags); |
066b2393 MG |
3239 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3240 | list = &pcp->lists[migratetype]; | |
6bb15450 | 3241 | page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list); |
066b2393 | 3242 | if (page) { |
1c52e6d0 | 3243 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1); |
066b2393 MG |
3244 | zone_statistics(preferred_zone, zone); |
3245 | } | |
d34b0733 | 3246 | local_irq_restore(flags); |
066b2393 MG |
3247 | return page; |
3248 | } | |
3249 | ||
1da177e4 | 3250 | /* |
75379191 | 3251 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 3252 | */ |
0a15c3e9 | 3253 | static inline |
066b2393 | 3254 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3255 | struct zone *zone, unsigned int order, |
c603844b MG |
3256 | gfp_t gfp_flags, unsigned int alloc_flags, |
3257 | int migratetype) | |
1da177e4 LT |
3258 | { |
3259 | unsigned long flags; | |
689bcebf | 3260 | struct page *page; |
1da177e4 | 3261 | |
d34b0733 | 3262 | if (likely(order == 0)) { |
1c52e6d0 YS |
3263 | page = rmqueue_pcplist(preferred_zone, zone, gfp_flags, |
3264 | migratetype, alloc_flags); | |
066b2393 MG |
3265 | goto out; |
3266 | } | |
83b9355b | 3267 | |
066b2393 MG |
3268 | /* |
3269 | * We most definitely don't want callers attempting to | |
3270 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3271 | */ | |
3272 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3273 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3274 | |
066b2393 MG |
3275 | do { |
3276 | page = NULL; | |
3277 | if (alloc_flags & ALLOC_HARDER) { | |
3278 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3279 | if (page) | |
3280 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3281 | } | |
a74609fa | 3282 | if (!page) |
6bb15450 | 3283 | page = __rmqueue(zone, order, migratetype, alloc_flags); |
066b2393 MG |
3284 | } while (page && check_new_pages(page, order)); |
3285 | spin_unlock(&zone->lock); | |
3286 | if (!page) | |
3287 | goto failed; | |
3288 | __mod_zone_freepage_state(zone, -(1 << order), | |
3289 | get_pcppage_migratetype(page)); | |
1da177e4 | 3290 | |
16709d1d | 3291 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3292 | zone_statistics(preferred_zone, zone); |
a74609fa | 3293 | local_irq_restore(flags); |
1da177e4 | 3294 | |
066b2393 | 3295 | out: |
73444bc4 MG |
3296 | /* Separate test+clear to avoid unnecessary atomics */ |
3297 | if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) { | |
3298 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); | |
3299 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3300 | } | |
3301 | ||
066b2393 | 3302 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 | 3303 | return page; |
a74609fa NP |
3304 | |
3305 | failed: | |
3306 | local_irq_restore(flags); | |
a74609fa | 3307 | return NULL; |
1da177e4 LT |
3308 | } |
3309 | ||
933e312e AM |
3310 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3311 | ||
b2588c4b | 3312 | static struct { |
933e312e AM |
3313 | struct fault_attr attr; |
3314 | ||
621a5f7a | 3315 | bool ignore_gfp_highmem; |
71baba4b | 3316 | bool ignore_gfp_reclaim; |
54114994 | 3317 | u32 min_order; |
933e312e AM |
3318 | } fail_page_alloc = { |
3319 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3320 | .ignore_gfp_reclaim = true, |
621a5f7a | 3321 | .ignore_gfp_highmem = true, |
54114994 | 3322 | .min_order = 1, |
933e312e AM |
3323 | }; |
3324 | ||
3325 | static int __init setup_fail_page_alloc(char *str) | |
3326 | { | |
3327 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3328 | } | |
3329 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3330 | ||
af3b8544 | 3331 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3332 | { |
54114994 | 3333 | if (order < fail_page_alloc.min_order) |
deaf386e | 3334 | return false; |
933e312e | 3335 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3336 | return false; |
933e312e | 3337 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3338 | return false; |
71baba4b MG |
3339 | if (fail_page_alloc.ignore_gfp_reclaim && |
3340 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3341 | return false; |
933e312e AM |
3342 | |
3343 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3344 | } | |
3345 | ||
3346 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3347 | ||
3348 | static int __init fail_page_alloc_debugfs(void) | |
3349 | { | |
0825a6f9 | 3350 | umode_t mode = S_IFREG | 0600; |
933e312e | 3351 | struct dentry *dir; |
933e312e | 3352 | |
dd48c085 AM |
3353 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3354 | &fail_page_alloc.attr); | |
b2588c4b | 3355 | |
d9f7979c GKH |
3356 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3357 | &fail_page_alloc.ignore_gfp_reclaim); | |
3358 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3359 | &fail_page_alloc.ignore_gfp_highmem); | |
3360 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3361 | |
d9f7979c | 3362 | return 0; |
933e312e AM |
3363 | } |
3364 | ||
3365 | late_initcall(fail_page_alloc_debugfs); | |
3366 | ||
3367 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3368 | ||
3369 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3370 | ||
af3b8544 | 3371 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3372 | { |
deaf386e | 3373 | return false; |
933e312e AM |
3374 | } |
3375 | ||
3376 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3377 | ||
af3b8544 BP |
3378 | static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
3379 | { | |
3380 | return __should_fail_alloc_page(gfp_mask, order); | |
3381 | } | |
3382 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3383 | ||
1da177e4 | 3384 | /* |
97a16fc8 MG |
3385 | * Return true if free base pages are above 'mark'. For high-order checks it |
3386 | * will return true of the order-0 watermark is reached and there is at least | |
3387 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3388 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3389 | */ |
86a294a8 MH |
3390 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
3391 | int classzone_idx, unsigned int alloc_flags, | |
3392 | long free_pages) | |
1da177e4 | 3393 | { |
d23ad423 | 3394 | long min = mark; |
1da177e4 | 3395 | int o; |
cd04ae1e | 3396 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3397 | |
0aaa29a5 | 3398 | /* free_pages may go negative - that's OK */ |
df0a6daa | 3399 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 3400 | |
7fb1d9fc | 3401 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3402 | min -= min / 2; |
0aaa29a5 MG |
3403 | |
3404 | /* | |
3405 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3406 | * the high-atomic reserves. This will over-estimate the size of the | |
3407 | * atomic reserve but it avoids a search. | |
3408 | */ | |
cd04ae1e | 3409 | if (likely(!alloc_harder)) { |
0aaa29a5 | 3410 | free_pages -= z->nr_reserved_highatomic; |
cd04ae1e MH |
3411 | } else { |
3412 | /* | |
3413 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3414 | * users on the grounds that it's definitely going to be in | |
3415 | * the exit path shortly and free memory. Any allocation it | |
3416 | * makes during the free path will be small and short-lived. | |
3417 | */ | |
3418 | if (alloc_flags & ALLOC_OOM) | |
3419 | min -= min / 2; | |
3420 | else | |
3421 | min -= min / 4; | |
3422 | } | |
3423 | ||
e2b19197 | 3424 | |
d883c6cf JK |
3425 | #ifdef CONFIG_CMA |
3426 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3427 | if (!(alloc_flags & ALLOC_CMA)) | |
3428 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); | |
3429 | #endif | |
3430 | ||
97a16fc8 MG |
3431 | /* |
3432 | * Check watermarks for an order-0 allocation request. If these | |
3433 | * are not met, then a high-order request also cannot go ahead | |
3434 | * even if a suitable page happened to be free. | |
3435 | */ | |
3436 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 3437 | return false; |
1da177e4 | 3438 | |
97a16fc8 MG |
3439 | /* If this is an order-0 request then the watermark is fine */ |
3440 | if (!order) | |
3441 | return true; | |
3442 | ||
3443 | /* For a high-order request, check at least one suitable page is free */ | |
3444 | for (o = order; o < MAX_ORDER; o++) { | |
3445 | struct free_area *area = &z->free_area[o]; | |
3446 | int mt; | |
3447 | ||
3448 | if (!area->nr_free) | |
3449 | continue; | |
3450 | ||
97a16fc8 | 3451 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3452 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3453 | return true; |
3454 | } | |
3455 | ||
3456 | #ifdef CONFIG_CMA | |
d883c6cf | 3457 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3458 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3459 | return true; |
d883c6cf | 3460 | } |
97a16fc8 | 3461 | #endif |
b050e376 VB |
3462 | if (alloc_harder && |
3463 | !list_empty(&area->free_list[MIGRATE_HIGHATOMIC])) | |
3464 | return true; | |
1da177e4 | 3465 | } |
97a16fc8 | 3466 | return false; |
88f5acf8 MG |
3467 | } |
3468 | ||
7aeb09f9 | 3469 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 3470 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
3471 | { |
3472 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3473 | zone_page_state(z, NR_FREE_PAGES)); | |
3474 | } | |
3475 | ||
48ee5f36 MG |
3476 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
3477 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
3478 | { | |
3479 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
d883c6cf JK |
3480 | long cma_pages = 0; |
3481 | ||
3482 | #ifdef CONFIG_CMA | |
3483 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3484 | if (!(alloc_flags & ALLOC_CMA)) | |
3485 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
3486 | #endif | |
48ee5f36 MG |
3487 | |
3488 | /* | |
3489 | * Fast check for order-0 only. If this fails then the reserves | |
3490 | * need to be calculated. There is a corner case where the check | |
3491 | * passes but only the high-order atomic reserve are free. If | |
3492 | * the caller is !atomic then it'll uselessly search the free | |
3493 | * list. That corner case is then slower but it is harmless. | |
3494 | */ | |
d883c6cf | 3495 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) |
48ee5f36 MG |
3496 | return true; |
3497 | ||
3498 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3499 | free_pages); | |
3500 | } | |
3501 | ||
7aeb09f9 | 3502 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 3503 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
3504 | { |
3505 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3506 | ||
3507 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3508 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3509 | ||
e2b19197 | 3510 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 3511 | free_pages); |
1da177e4 LT |
3512 | } |
3513 | ||
9276b1bc | 3514 | #ifdef CONFIG_NUMA |
957f822a DR |
3515 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3516 | { | |
e02dc017 | 3517 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3518 | node_reclaim_distance; |
957f822a | 3519 | } |
9276b1bc | 3520 | #else /* CONFIG_NUMA */ |
957f822a DR |
3521 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3522 | { | |
3523 | return true; | |
3524 | } | |
9276b1bc PJ |
3525 | #endif /* CONFIG_NUMA */ |
3526 | ||
6bb15450 MG |
3527 | /* |
3528 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3529 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3530 | * premature use of a lower zone may cause lowmem pressure problems that | |
3531 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3532 | * probably too small. It only makes sense to spread allocations to avoid | |
3533 | * fragmentation between the Normal and DMA32 zones. | |
3534 | */ | |
3535 | static inline unsigned int | |
0a79cdad | 3536 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3537 | { |
0a79cdad MG |
3538 | unsigned int alloc_flags = 0; |
3539 | ||
3540 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) | |
3541 | alloc_flags |= ALLOC_KSWAPD; | |
3542 | ||
3543 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3544 | if (!zone) |
3545 | return alloc_flags; | |
3546 | ||
6bb15450 | 3547 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3548 | return alloc_flags; |
6bb15450 MG |
3549 | |
3550 | /* | |
3551 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3552 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3553 | * on UMA that if Normal is populated then so is DMA32. | |
3554 | */ | |
3555 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3556 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3557 | return alloc_flags; |
6bb15450 | 3558 | |
8118b82e | 3559 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3560 | #endif /* CONFIG_ZONE_DMA32 */ |
3561 | return alloc_flags; | |
6bb15450 | 3562 | } |
6bb15450 | 3563 | |
7fb1d9fc | 3564 | /* |
0798e519 | 3565 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3566 | * a page. |
3567 | */ | |
3568 | static struct page * | |
a9263751 VB |
3569 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3570 | const struct alloc_context *ac) | |
753ee728 | 3571 | { |
6bb15450 | 3572 | struct zoneref *z; |
5117f45d | 3573 | struct zone *zone; |
3b8c0be4 | 3574 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
6bb15450 | 3575 | bool no_fallback; |
3b8c0be4 | 3576 | |
6bb15450 | 3577 | retry: |
7fb1d9fc | 3578 | /* |
9276b1bc | 3579 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3580 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3581 | */ |
6bb15450 MG |
3582 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3583 | z = ac->preferred_zoneref; | |
c33d6c06 | 3584 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 3585 | ac->nodemask) { |
be06af00 | 3586 | struct page *page; |
e085dbc5 JW |
3587 | unsigned long mark; |
3588 | ||
664eedde MG |
3589 | if (cpusets_enabled() && |
3590 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3591 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3592 | continue; |
a756cf59 JW |
3593 | /* |
3594 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3595 | * want to get it from a node that is within its dirty |
3596 | * limit, such that no single node holds more than its | |
a756cf59 | 3597 | * proportional share of globally allowed dirty pages. |
281e3726 | 3598 | * The dirty limits take into account the node's |
a756cf59 JW |
3599 | * lowmem reserves and high watermark so that kswapd |
3600 | * should be able to balance it without having to | |
3601 | * write pages from its LRU list. | |
3602 | * | |
a756cf59 | 3603 | * XXX: For now, allow allocations to potentially |
281e3726 | 3604 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3605 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3606 | * which is important when on a NUMA setup the allowed |
281e3726 | 3607 | * nodes are together not big enough to reach the |
a756cf59 | 3608 | * global limit. The proper fix for these situations |
281e3726 | 3609 | * will require awareness of nodes in the |
a756cf59 JW |
3610 | * dirty-throttling and the flusher threads. |
3611 | */ | |
3b8c0be4 MG |
3612 | if (ac->spread_dirty_pages) { |
3613 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3614 | continue; | |
3615 | ||
3616 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3617 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3618 | continue; | |
3619 | } | |
3620 | } | |
7fb1d9fc | 3621 | |
6bb15450 MG |
3622 | if (no_fallback && nr_online_nodes > 1 && |
3623 | zone != ac->preferred_zoneref->zone) { | |
3624 | int local_nid; | |
3625 | ||
3626 | /* | |
3627 | * If moving to a remote node, retry but allow | |
3628 | * fragmenting fallbacks. Locality is more important | |
3629 | * than fragmentation avoidance. | |
3630 | */ | |
3631 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3632 | if (zone_to_nid(zone) != local_nid) { | |
3633 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3634 | goto retry; | |
3635 | } | |
3636 | } | |
3637 | ||
a9214443 | 3638 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 3639 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 3640 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
3641 | int ret; |
3642 | ||
c9e97a19 PT |
3643 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3644 | /* | |
3645 | * Watermark failed for this zone, but see if we can | |
3646 | * grow this zone if it contains deferred pages. | |
3647 | */ | |
3648 | if (static_branch_unlikely(&deferred_pages)) { | |
3649 | if (_deferred_grow_zone(zone, order)) | |
3650 | goto try_this_zone; | |
3651 | } | |
3652 | #endif | |
5dab2911 MG |
3653 | /* Checked here to keep the fast path fast */ |
3654 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3655 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3656 | goto try_this_zone; | |
3657 | ||
a5f5f91d | 3658 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3659 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3660 | continue; |
3661 | ||
a5f5f91d | 3662 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3663 | switch (ret) { |
a5f5f91d | 3664 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3665 | /* did not scan */ |
cd38b115 | 3666 | continue; |
a5f5f91d | 3667 | case NODE_RECLAIM_FULL: |
fa5e084e | 3668 | /* scanned but unreclaimable */ |
cd38b115 | 3669 | continue; |
fa5e084e MG |
3670 | default: |
3671 | /* did we reclaim enough */ | |
fed2719e | 3672 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3673 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3674 | goto try_this_zone; |
3675 | ||
fed2719e | 3676 | continue; |
0798e519 | 3677 | } |
7fb1d9fc RS |
3678 | } |
3679 | ||
fa5e084e | 3680 | try_this_zone: |
066b2393 | 3681 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3682 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3683 | if (page) { |
479f854a | 3684 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3685 | |
3686 | /* | |
3687 | * If this is a high-order atomic allocation then check | |
3688 | * if the pageblock should be reserved for the future | |
3689 | */ | |
3690 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3691 | reserve_highatomic_pageblock(page, zone, order); | |
3692 | ||
75379191 | 3693 | return page; |
c9e97a19 PT |
3694 | } else { |
3695 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
3696 | /* Try again if zone has deferred pages */ | |
3697 | if (static_branch_unlikely(&deferred_pages)) { | |
3698 | if (_deferred_grow_zone(zone, order)) | |
3699 | goto try_this_zone; | |
3700 | } | |
3701 | #endif | |
75379191 | 3702 | } |
54a6eb5c | 3703 | } |
9276b1bc | 3704 | |
6bb15450 MG |
3705 | /* |
3706 | * It's possible on a UMA machine to get through all zones that are | |
3707 | * fragmented. If avoiding fragmentation, reset and try again. | |
3708 | */ | |
3709 | if (no_fallback) { | |
3710 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3711 | goto retry; | |
3712 | } | |
3713 | ||
4ffeaf35 | 3714 | return NULL; |
753ee728 MH |
3715 | } |
3716 | ||
9af744d7 | 3717 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3718 | { |
a238ab5b | 3719 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
3720 | |
3721 | /* | |
3722 | * This documents exceptions given to allocations in certain | |
3723 | * contexts that are allowed to allocate outside current's set | |
3724 | * of allowed nodes. | |
3725 | */ | |
3726 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3727 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3728 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3729 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3730 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3731 | filter &= ~SHOW_MEM_FILTER_NODES; |
3732 | ||
9af744d7 | 3733 | show_mem(filter, nodemask); |
aa187507 MH |
3734 | } |
3735 | ||
a8e99259 | 3736 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3737 | { |
3738 | struct va_format vaf; | |
3739 | va_list args; | |
1be334e5 | 3740 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 3741 | |
0f7896f1 | 3742 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3743 | return; |
3744 | ||
7877cdcc MH |
3745 | va_start(args, fmt); |
3746 | vaf.fmt = fmt; | |
3747 | vaf.va = &args; | |
ef8444ea | 3748 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
3749 | current->comm, &vaf, gfp_mask, &gfp_mask, |
3750 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3751 | va_end(args); |
3ee9a4f0 | 3752 | |
a8e99259 | 3753 | cpuset_print_current_mems_allowed(); |
ef8444ea | 3754 | pr_cont("\n"); |
a238ab5b | 3755 | dump_stack(); |
685dbf6f | 3756 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3757 | } |
3758 | ||
6c18ba7a MH |
3759 | static inline struct page * |
3760 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3761 | unsigned int alloc_flags, | |
3762 | const struct alloc_context *ac) | |
3763 | { | |
3764 | struct page *page; | |
3765 | ||
3766 | page = get_page_from_freelist(gfp_mask, order, | |
3767 | alloc_flags|ALLOC_CPUSET, ac); | |
3768 | /* | |
3769 | * fallback to ignore cpuset restriction if our nodes | |
3770 | * are depleted | |
3771 | */ | |
3772 | if (!page) | |
3773 | page = get_page_from_freelist(gfp_mask, order, | |
3774 | alloc_flags, ac); | |
3775 | ||
3776 | return page; | |
3777 | } | |
3778 | ||
11e33f6a MG |
3779 | static inline struct page * |
3780 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3781 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3782 | { |
6e0fc46d DR |
3783 | struct oom_control oc = { |
3784 | .zonelist = ac->zonelist, | |
3785 | .nodemask = ac->nodemask, | |
2a966b77 | 3786 | .memcg = NULL, |
6e0fc46d DR |
3787 | .gfp_mask = gfp_mask, |
3788 | .order = order, | |
6e0fc46d | 3789 | }; |
11e33f6a MG |
3790 | struct page *page; |
3791 | ||
9879de73 JW |
3792 | *did_some_progress = 0; |
3793 | ||
9879de73 | 3794 | /* |
dc56401f JW |
3795 | * Acquire the oom lock. If that fails, somebody else is |
3796 | * making progress for us. | |
9879de73 | 3797 | */ |
dc56401f | 3798 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3799 | *did_some_progress = 1; |
11e33f6a | 3800 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3801 | return NULL; |
3802 | } | |
6b1de916 | 3803 | |
11e33f6a MG |
3804 | /* |
3805 | * Go through the zonelist yet one more time, keep very high watermark | |
3806 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3807 | * we're still under heavy pressure. But make sure that this reclaim |
3808 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3809 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3810 | */ |
e746bf73 TH |
3811 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3812 | ~__GFP_DIRECT_RECLAIM, order, | |
3813 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3814 | if (page) |
11e33f6a MG |
3815 | goto out; |
3816 | ||
06ad276a MH |
3817 | /* Coredumps can quickly deplete all memory reserves */ |
3818 | if (current->flags & PF_DUMPCORE) | |
3819 | goto out; | |
3820 | /* The OOM killer will not help higher order allocs */ | |
3821 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3822 | goto out; | |
dcda9b04 MH |
3823 | /* |
3824 | * We have already exhausted all our reclaim opportunities without any | |
3825 | * success so it is time to admit defeat. We will skip the OOM killer | |
3826 | * because it is very likely that the caller has a more reasonable | |
3827 | * fallback than shooting a random task. | |
3828 | */ | |
3829 | if (gfp_mask & __GFP_RETRY_MAYFAIL) | |
3830 | goto out; | |
06ad276a MH |
3831 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
3832 | if (ac->high_zoneidx < ZONE_NORMAL) | |
3833 | goto out; | |
3834 | if (pm_suspended_storage()) | |
3835 | goto out; | |
3836 | /* | |
3837 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3838 | * other request to make a forward progress. | |
3839 | * We are in an unfortunate situation where out_of_memory cannot | |
3840 | * do much for this context but let's try it to at least get | |
3841 | * access to memory reserved if the current task is killed (see | |
3842 | * out_of_memory). Once filesystems are ready to handle allocation | |
3843 | * failures more gracefully we should just bail out here. | |
3844 | */ | |
3845 | ||
3846 | /* The OOM killer may not free memory on a specific node */ | |
3847 | if (gfp_mask & __GFP_THISNODE) | |
3848 | goto out; | |
3da88fb3 | 3849 | |
3c2c6488 | 3850 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 3851 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3852 | *did_some_progress = 1; |
5020e285 | 3853 | |
6c18ba7a MH |
3854 | /* |
3855 | * Help non-failing allocations by giving them access to memory | |
3856 | * reserves | |
3857 | */ | |
3858 | if (gfp_mask & __GFP_NOFAIL) | |
3859 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3860 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3861 | } |
11e33f6a | 3862 | out: |
dc56401f | 3863 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3864 | return page; |
3865 | } | |
3866 | ||
33c2d214 MH |
3867 | /* |
3868 | * Maximum number of compaction retries wit a progress before OOM | |
3869 | * killer is consider as the only way to move forward. | |
3870 | */ | |
3871 | #define MAX_COMPACT_RETRIES 16 | |
3872 | ||
56de7263 MG |
3873 | #ifdef CONFIG_COMPACTION |
3874 | /* Try memory compaction for high-order allocations before reclaim */ | |
3875 | static struct page * | |
3876 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3877 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3878 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3879 | { |
5e1f0f09 | 3880 | struct page *page = NULL; |
eb414681 | 3881 | unsigned long pflags; |
499118e9 | 3882 | unsigned int noreclaim_flag; |
53853e2d VB |
3883 | |
3884 | if (!order) | |
66199712 | 3885 | return NULL; |
66199712 | 3886 | |
eb414681 | 3887 | psi_memstall_enter(&pflags); |
499118e9 | 3888 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 3889 | |
c5d01d0d | 3890 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 3891 | prio, &page); |
eb414681 | 3892 | |
499118e9 | 3893 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 3894 | psi_memstall_leave(&pflags); |
56de7263 | 3895 | |
98dd3b48 VB |
3896 | /* |
3897 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3898 | * count a compaction stall | |
3899 | */ | |
3900 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3901 | |
5e1f0f09 MG |
3902 | /* Prep a captured page if available */ |
3903 | if (page) | |
3904 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
3905 | ||
3906 | /* Try get a page from the freelist if available */ | |
3907 | if (!page) | |
3908 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 3909 | |
98dd3b48 VB |
3910 | if (page) { |
3911 | struct zone *zone = page_zone(page); | |
53853e2d | 3912 | |
98dd3b48 VB |
3913 | zone->compact_blockskip_flush = false; |
3914 | compaction_defer_reset(zone, order, true); | |
3915 | count_vm_event(COMPACTSUCCESS); | |
3916 | return page; | |
3917 | } | |
56de7263 | 3918 | |
98dd3b48 VB |
3919 | /* |
3920 | * It's bad if compaction run occurs and fails. The most likely reason | |
3921 | * is that pages exist, but not enough to satisfy watermarks. | |
3922 | */ | |
3923 | count_vm_event(COMPACTFAIL); | |
66199712 | 3924 | |
98dd3b48 | 3925 | cond_resched(); |
56de7263 MG |
3926 | |
3927 | return NULL; | |
3928 | } | |
33c2d214 | 3929 | |
3250845d VB |
3930 | static inline bool |
3931 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
3932 | enum compact_result compact_result, | |
3933 | enum compact_priority *compact_priority, | |
d9436498 | 3934 | int *compaction_retries) |
3250845d VB |
3935 | { |
3936 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 3937 | int min_priority; |
65190cff MH |
3938 | bool ret = false; |
3939 | int retries = *compaction_retries; | |
3940 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
3941 | |
3942 | if (!order) | |
3943 | return false; | |
3944 | ||
d9436498 VB |
3945 | if (compaction_made_progress(compact_result)) |
3946 | (*compaction_retries)++; | |
3947 | ||
3250845d VB |
3948 | /* |
3949 | * compaction considers all the zone as desperately out of memory | |
3950 | * so it doesn't really make much sense to retry except when the | |
3951 | * failure could be caused by insufficient priority | |
3952 | */ | |
d9436498 VB |
3953 | if (compaction_failed(compact_result)) |
3954 | goto check_priority; | |
3250845d | 3955 | |
49433085 VB |
3956 | /* |
3957 | * compaction was skipped because there are not enough order-0 pages | |
3958 | * to work with, so we retry only if it looks like reclaim can help. | |
3959 | */ | |
3960 | if (compaction_needs_reclaim(compact_result)) { | |
3961 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
3962 | goto out; | |
3963 | } | |
3964 | ||
3250845d VB |
3965 | /* |
3966 | * make sure the compaction wasn't deferred or didn't bail out early | |
3967 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
3968 | * But the next retry should use a higher priority if allowed, so |
3969 | * we don't just keep bailing out endlessly. | |
3250845d | 3970 | */ |
65190cff | 3971 | if (compaction_withdrawn(compact_result)) { |
49433085 | 3972 | goto check_priority; |
65190cff | 3973 | } |
3250845d VB |
3974 | |
3975 | /* | |
dcda9b04 | 3976 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
3977 | * costly ones because they are de facto nofail and invoke OOM |
3978 | * killer to move on while costly can fail and users are ready | |
3979 | * to cope with that. 1/4 retries is rather arbitrary but we | |
3980 | * would need much more detailed feedback from compaction to | |
3981 | * make a better decision. | |
3982 | */ | |
3983 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3984 | max_retries /= 4; | |
65190cff MH |
3985 | if (*compaction_retries <= max_retries) { |
3986 | ret = true; | |
3987 | goto out; | |
3988 | } | |
3250845d | 3989 | |
d9436498 VB |
3990 | /* |
3991 | * Make sure there are attempts at the highest priority if we exhausted | |
3992 | * all retries or failed at the lower priorities. | |
3993 | */ | |
3994 | check_priority: | |
c2033b00 VB |
3995 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3996 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 3997 | |
c2033b00 | 3998 | if (*compact_priority > min_priority) { |
d9436498 VB |
3999 | (*compact_priority)--; |
4000 | *compaction_retries = 0; | |
65190cff | 4001 | ret = true; |
d9436498 | 4002 | } |
65190cff MH |
4003 | out: |
4004 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4005 | return ret; | |
3250845d | 4006 | } |
56de7263 MG |
4007 | #else |
4008 | static inline struct page * | |
4009 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4010 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4011 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4012 | { |
33c2d214 | 4013 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4014 | return NULL; |
4015 | } | |
33c2d214 MH |
4016 | |
4017 | static inline bool | |
86a294a8 MH |
4018 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4019 | enum compact_result compact_result, | |
a5508cd8 | 4020 | enum compact_priority *compact_priority, |
d9436498 | 4021 | int *compaction_retries) |
33c2d214 | 4022 | { |
31e49bfd MH |
4023 | struct zone *zone; |
4024 | struct zoneref *z; | |
4025 | ||
4026 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4027 | return false; | |
4028 | ||
4029 | /* | |
4030 | * There are setups with compaction disabled which would prefer to loop | |
4031 | * inside the allocator rather than hit the oom killer prematurely. | |
4032 | * Let's give them a good hope and keep retrying while the order-0 | |
4033 | * watermarks are OK. | |
4034 | */ | |
4035 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
4036 | ac->nodemask) { | |
4037 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
4038 | ac_classzone_idx(ac), alloc_flags)) | |
4039 | return true; | |
4040 | } | |
33c2d214 MH |
4041 | return false; |
4042 | } | |
3250845d | 4043 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4044 | |
d92a8cfc | 4045 | #ifdef CONFIG_LOCKDEP |
93781325 | 4046 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4047 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4048 | ||
4049 | static bool __need_fs_reclaim(gfp_t gfp_mask) | |
4050 | { | |
4051 | gfp_mask = current_gfp_context(gfp_mask); | |
4052 | ||
4053 | /* no reclaim without waiting on it */ | |
4054 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4055 | return false; | |
4056 | ||
4057 | /* this guy won't enter reclaim */ | |
2e517d68 | 4058 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4059 | return false; |
4060 | ||
4061 | /* We're only interested __GFP_FS allocations for now */ | |
4062 | if (!(gfp_mask & __GFP_FS)) | |
4063 | return false; | |
4064 | ||
4065 | if (gfp_mask & __GFP_NOLOCKDEP) | |
4066 | return false; | |
4067 | ||
4068 | return true; | |
4069 | } | |
4070 | ||
93781325 OS |
4071 | void __fs_reclaim_acquire(void) |
4072 | { | |
4073 | lock_map_acquire(&__fs_reclaim_map); | |
4074 | } | |
4075 | ||
4076 | void __fs_reclaim_release(void) | |
4077 | { | |
4078 | lock_map_release(&__fs_reclaim_map); | |
4079 | } | |
4080 | ||
d92a8cfc PZ |
4081 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4082 | { | |
4083 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 4084 | __fs_reclaim_acquire(); |
d92a8cfc PZ |
4085 | } |
4086 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4087 | ||
4088 | void fs_reclaim_release(gfp_t gfp_mask) | |
4089 | { | |
4090 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 4091 | __fs_reclaim_release(); |
d92a8cfc PZ |
4092 | } |
4093 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4094 | #endif | |
4095 | ||
bba90710 MS |
4096 | /* Perform direct synchronous page reclaim */ |
4097 | static int | |
a9263751 VB |
4098 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4099 | const struct alloc_context *ac) | |
11e33f6a | 4100 | { |
bba90710 | 4101 | int progress; |
499118e9 | 4102 | unsigned int noreclaim_flag; |
eb414681 | 4103 | unsigned long pflags; |
11e33f6a MG |
4104 | |
4105 | cond_resched(); | |
4106 | ||
4107 | /* We now go into synchronous reclaim */ | |
4108 | cpuset_memory_pressure_bump(); | |
eb414681 | 4109 | psi_memstall_enter(&pflags); |
d92a8cfc | 4110 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4111 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4112 | |
a9263751 VB |
4113 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4114 | ac->nodemask); | |
11e33f6a | 4115 | |
499118e9 | 4116 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4117 | fs_reclaim_release(gfp_mask); |
eb414681 | 4118 | psi_memstall_leave(&pflags); |
11e33f6a MG |
4119 | |
4120 | cond_resched(); | |
4121 | ||
bba90710 MS |
4122 | return progress; |
4123 | } | |
4124 | ||
4125 | /* The really slow allocator path where we enter direct reclaim */ | |
4126 | static inline struct page * | |
4127 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4128 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4129 | unsigned long *did_some_progress) |
bba90710 MS |
4130 | { |
4131 | struct page *page = NULL; | |
4132 | bool drained = false; | |
4133 | ||
a9263751 | 4134 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
4135 | if (unlikely(!(*did_some_progress))) |
4136 | return NULL; | |
11e33f6a | 4137 | |
9ee493ce | 4138 | retry: |
31a6c190 | 4139 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4140 | |
4141 | /* | |
4142 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
4143 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
4144 | * Shrink them them and try again | |
9ee493ce MG |
4145 | */ |
4146 | if (!page && !drained) { | |
29fac03b | 4147 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4148 | drain_all_pages(NULL); |
9ee493ce MG |
4149 | drained = true; |
4150 | goto retry; | |
4151 | } | |
4152 | ||
11e33f6a MG |
4153 | return page; |
4154 | } | |
4155 | ||
5ecd9d40 DR |
4156 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4157 | const struct alloc_context *ac) | |
3a025760 JW |
4158 | { |
4159 | struct zoneref *z; | |
4160 | struct zone *zone; | |
e1a55637 | 4161 | pg_data_t *last_pgdat = NULL; |
5ecd9d40 | 4162 | enum zone_type high_zoneidx = ac->high_zoneidx; |
3a025760 | 4163 | |
5ecd9d40 DR |
4164 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx, |
4165 | ac->nodemask) { | |
e1a55637 | 4166 | if (last_pgdat != zone->zone_pgdat) |
5ecd9d40 | 4167 | wakeup_kswapd(zone, gfp_mask, order, high_zoneidx); |
e1a55637 MG |
4168 | last_pgdat = zone->zone_pgdat; |
4169 | } | |
3a025760 JW |
4170 | } |
4171 | ||
c603844b | 4172 | static inline unsigned int |
341ce06f PZ |
4173 | gfp_to_alloc_flags(gfp_t gfp_mask) |
4174 | { | |
c603844b | 4175 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4176 | |
a56f57ff | 4177 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 4178 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 4179 | |
341ce06f PZ |
4180 | /* |
4181 | * The caller may dip into page reserves a bit more if the caller | |
4182 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4183 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 4184 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 4185 | */ |
e6223a3b | 4186 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 4187 | |
d0164adc | 4188 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 4189 | /* |
b104a35d DR |
4190 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4191 | * if it can't schedule. | |
5c3240d9 | 4192 | */ |
b104a35d | 4193 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 4194 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 4195 | /* |
b104a35d | 4196 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 4197 | * comment for __cpuset_node_allowed(). |
523b9458 | 4198 | */ |
341ce06f | 4199 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 4200 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
4201 | alloc_flags |= ALLOC_HARDER; |
4202 | ||
0a79cdad MG |
4203 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
4204 | alloc_flags |= ALLOC_KSWAPD; | |
4205 | ||
d883c6cf JK |
4206 | #ifdef CONFIG_CMA |
4207 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
4208 | alloc_flags |= ALLOC_CMA; | |
4209 | #endif | |
341ce06f PZ |
4210 | return alloc_flags; |
4211 | } | |
4212 | ||
cd04ae1e | 4213 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4214 | { |
cd04ae1e MH |
4215 | if (!tsk_is_oom_victim(tsk)) |
4216 | return false; | |
4217 | ||
4218 | /* | |
4219 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4220 | * only to the thread with TIF_MEMDIE set | |
4221 | */ | |
4222 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4223 | return false; |
4224 | ||
cd04ae1e MH |
4225 | return true; |
4226 | } | |
4227 | ||
4228 | /* | |
4229 | * Distinguish requests which really need access to full memory | |
4230 | * reserves from oom victims which can live with a portion of it | |
4231 | */ | |
4232 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4233 | { | |
4234 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4235 | return 0; | |
31a6c190 | 4236 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4237 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4238 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4239 | return ALLOC_NO_WATERMARKS; |
4240 | if (!in_interrupt()) { | |
4241 | if (current->flags & PF_MEMALLOC) | |
4242 | return ALLOC_NO_WATERMARKS; | |
4243 | else if (oom_reserves_allowed(current)) | |
4244 | return ALLOC_OOM; | |
4245 | } | |
31a6c190 | 4246 | |
cd04ae1e MH |
4247 | return 0; |
4248 | } | |
4249 | ||
4250 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4251 | { | |
4252 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4253 | } |
4254 | ||
0a0337e0 MH |
4255 | /* |
4256 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4257 | * for the given allocation request. | |
491d79ae JW |
4258 | * |
4259 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4260 | * without success, or when we couldn't even meet the watermark if we | |
4261 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4262 | * |
4263 | * Returns true if a retry is viable or false to enter the oom path. | |
4264 | */ | |
4265 | static inline bool | |
4266 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4267 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4268 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4269 | { |
4270 | struct zone *zone; | |
4271 | struct zoneref *z; | |
15f570bf | 4272 | bool ret = false; |
0a0337e0 | 4273 | |
423b452e VB |
4274 | /* |
4275 | * Costly allocations might have made a progress but this doesn't mean | |
4276 | * their order will become available due to high fragmentation so | |
4277 | * always increment the no progress counter for them | |
4278 | */ | |
4279 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4280 | *no_progress_loops = 0; | |
4281 | else | |
4282 | (*no_progress_loops)++; | |
4283 | ||
0a0337e0 MH |
4284 | /* |
4285 | * Make sure we converge to OOM if we cannot make any progress | |
4286 | * several times in the row. | |
4287 | */ | |
04c8716f MK |
4288 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4289 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4290 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4291 | } |
0a0337e0 | 4292 | |
bca67592 MG |
4293 | /* |
4294 | * Keep reclaiming pages while there is a chance this will lead | |
4295 | * somewhere. If none of the target zones can satisfy our allocation | |
4296 | * request even if all reclaimable pages are considered then we are | |
4297 | * screwed and have to go OOM. | |
0a0337e0 MH |
4298 | */ |
4299 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
4300 | ac->nodemask) { | |
4301 | unsigned long available; | |
ede37713 | 4302 | unsigned long reclaimable; |
d379f01d MH |
4303 | unsigned long min_wmark = min_wmark_pages(zone); |
4304 | bool wmark; | |
0a0337e0 | 4305 | |
5a1c84b4 | 4306 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4307 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4308 | |
4309 | /* | |
491d79ae JW |
4310 | * Would the allocation succeed if we reclaimed all |
4311 | * reclaimable pages? | |
0a0337e0 | 4312 | */ |
d379f01d MH |
4313 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
4314 | ac_classzone_idx(ac), alloc_flags, available); | |
4315 | trace_reclaim_retry_zone(z, order, reclaimable, | |
4316 | available, min_wmark, *no_progress_loops, wmark); | |
4317 | if (wmark) { | |
ede37713 MH |
4318 | /* |
4319 | * If we didn't make any progress and have a lot of | |
4320 | * dirty + writeback pages then we should wait for | |
4321 | * an IO to complete to slow down the reclaim and | |
4322 | * prevent from pre mature OOM | |
4323 | */ | |
4324 | if (!did_some_progress) { | |
11fb9989 | 4325 | unsigned long write_pending; |
ede37713 | 4326 | |
5a1c84b4 MG |
4327 | write_pending = zone_page_state_snapshot(zone, |
4328 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 4329 | |
11fb9989 | 4330 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
4331 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
4332 | return true; | |
4333 | } | |
4334 | } | |
5a1c84b4 | 4335 | |
15f570bf MH |
4336 | ret = true; |
4337 | goto out; | |
0a0337e0 MH |
4338 | } |
4339 | } | |
4340 | ||
15f570bf MH |
4341 | out: |
4342 | /* | |
4343 | * Memory allocation/reclaim might be called from a WQ context and the | |
4344 | * current implementation of the WQ concurrency control doesn't | |
4345 | * recognize that a particular WQ is congested if the worker thread is | |
4346 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4347 | * here rather than calling cond_resched(). | |
4348 | */ | |
4349 | if (current->flags & PF_WQ_WORKER) | |
4350 | schedule_timeout_uninterruptible(1); | |
4351 | else | |
4352 | cond_resched(); | |
4353 | return ret; | |
0a0337e0 MH |
4354 | } |
4355 | ||
902b6281 VB |
4356 | static inline bool |
4357 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4358 | { | |
4359 | /* | |
4360 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4361 | * mempolicy don't intersect. This should be normally dealt with by | |
4362 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4363 | * such a way the check therein was true, and then it became false | |
4364 | * before we got our cpuset_mems_cookie here. | |
4365 | * This assumes that for all allocations, ac->nodemask can come only | |
4366 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4367 | * when it does not intersect with the cpuset restrictions) or the | |
4368 | * caller can deal with a violated nodemask. | |
4369 | */ | |
4370 | if (cpusets_enabled() && ac->nodemask && | |
4371 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4372 | ac->nodemask = NULL; | |
4373 | return true; | |
4374 | } | |
4375 | ||
4376 | /* | |
4377 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4378 | * possible to race with parallel threads in such a way that our | |
4379 | * allocation can fail while the mask is being updated. If we are about | |
4380 | * to fail, check if the cpuset changed during allocation and if so, | |
4381 | * retry. | |
4382 | */ | |
4383 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4384 | return true; | |
4385 | ||
4386 | return false; | |
4387 | } | |
4388 | ||
11e33f6a MG |
4389 | static inline struct page * |
4390 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4391 | struct alloc_context *ac) |
11e33f6a | 4392 | { |
d0164adc | 4393 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4394 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4395 | struct page *page = NULL; |
c603844b | 4396 | unsigned int alloc_flags; |
11e33f6a | 4397 | unsigned long did_some_progress; |
5ce9bfef | 4398 | enum compact_priority compact_priority; |
c5d01d0d | 4399 | enum compact_result compact_result; |
5ce9bfef VB |
4400 | int compaction_retries; |
4401 | int no_progress_loops; | |
5ce9bfef | 4402 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4403 | int reserve_flags; |
1da177e4 | 4404 | |
d0164adc MG |
4405 | /* |
4406 | * We also sanity check to catch abuse of atomic reserves being used by | |
4407 | * callers that are not in atomic context. | |
4408 | */ | |
4409 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4410 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4411 | gfp_mask &= ~__GFP_ATOMIC; | |
4412 | ||
5ce9bfef VB |
4413 | retry_cpuset: |
4414 | compaction_retries = 0; | |
4415 | no_progress_loops = 0; | |
4416 | compact_priority = DEF_COMPACT_PRIORITY; | |
4417 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4418 | |
4419 | /* | |
4420 | * The fast path uses conservative alloc_flags to succeed only until | |
4421 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4422 | * alloc_flags precisely. So we do that now. | |
4423 | */ | |
4424 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4425 | ||
e47483bc VB |
4426 | /* |
4427 | * We need to recalculate the starting point for the zonelist iterator | |
4428 | * because we might have used different nodemask in the fast path, or | |
4429 | * there was a cpuset modification and we are retrying - otherwise we | |
4430 | * could end up iterating over non-eligible zones endlessly. | |
4431 | */ | |
4432 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
4433 | ac->high_zoneidx, ac->nodemask); | |
4434 | if (!ac->preferred_zoneref->zone) | |
4435 | goto nopage; | |
4436 | ||
0a79cdad | 4437 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4438 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4439 | |
4440 | /* | |
4441 | * The adjusted alloc_flags might result in immediate success, so try | |
4442 | * that first | |
4443 | */ | |
4444 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4445 | if (page) | |
4446 | goto got_pg; | |
4447 | ||
a8161d1e VB |
4448 | /* |
4449 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4450 | * that we have enough base pages and don't need to reclaim. For non- |
4451 | * movable high-order allocations, do that as well, as compaction will | |
4452 | * try prevent permanent fragmentation by migrating from blocks of the | |
4453 | * same migratetype. | |
4454 | * Don't try this for allocations that are allowed to ignore | |
4455 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4456 | */ |
282722b0 VB |
4457 | if (can_direct_reclaim && |
4458 | (costly_order || | |
4459 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4460 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4461 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4462 | alloc_flags, ac, | |
a5508cd8 | 4463 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4464 | &compact_result); |
4465 | if (page) | |
4466 | goto got_pg; | |
4467 | ||
cc638f32 VB |
4468 | /* |
4469 | * Checks for costly allocations with __GFP_NORETRY, which | |
4470 | * includes some THP page fault allocations | |
4471 | */ | |
4472 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4473 | /* |
4474 | * If allocating entire pageblock(s) and compaction | |
4475 | * failed because all zones are below low watermarks | |
4476 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4477 | * order, fail immediately unless the allocator has |
4478 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4479 | * |
4480 | * Reclaim is | |
4481 | * - potentially very expensive because zones are far | |
4482 | * below their low watermarks or this is part of very | |
4483 | * bursty high order allocations, | |
4484 | * - not guaranteed to help because isolate_freepages() | |
4485 | * may not iterate over freed pages as part of its | |
4486 | * linear scan, and | |
4487 | * - unlikely to make entire pageblocks free on its | |
4488 | * own. | |
4489 | */ | |
4490 | if (compact_result == COMPACT_SKIPPED || | |
4491 | compact_result == COMPACT_DEFERRED) | |
4492 | goto nopage; | |
a8161d1e | 4493 | |
a8161d1e | 4494 | /* |
3eb2771b VB |
4495 | * Looks like reclaim/compaction is worth trying, but |
4496 | * sync compaction could be very expensive, so keep | |
25160354 | 4497 | * using async compaction. |
a8161d1e | 4498 | */ |
a5508cd8 | 4499 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4500 | } |
4501 | } | |
23771235 | 4502 | |
31a6c190 | 4503 | retry: |
23771235 | 4504 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4505 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4506 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4507 | |
cd04ae1e MH |
4508 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4509 | if (reserve_flags) | |
4510 | alloc_flags = reserve_flags; | |
23771235 | 4511 | |
e46e7b77 | 4512 | /* |
d6a24df0 VB |
4513 | * Reset the nodemask and zonelist iterators if memory policies can be |
4514 | * ignored. These allocations are high priority and system rather than | |
4515 | * user oriented. | |
e46e7b77 | 4516 | */ |
cd04ae1e | 4517 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4518 | ac->nodemask = NULL; |
e46e7b77 MG |
4519 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4520 | ac->high_zoneidx, ac->nodemask); | |
4521 | } | |
4522 | ||
23771235 | 4523 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4524 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4525 | if (page) |
4526 | goto got_pg; | |
1da177e4 | 4527 | |
d0164adc | 4528 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4529 | if (!can_direct_reclaim) |
1da177e4 LT |
4530 | goto nopage; |
4531 | ||
9a67f648 MH |
4532 | /* Avoid recursion of direct reclaim */ |
4533 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4534 | goto nopage; |
4535 | ||
a8161d1e VB |
4536 | /* Try direct reclaim and then allocating */ |
4537 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4538 | &did_some_progress); | |
4539 | if (page) | |
4540 | goto got_pg; | |
4541 | ||
4542 | /* Try direct compaction and then allocating */ | |
a9263751 | 4543 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4544 | compact_priority, &compact_result); |
56de7263 MG |
4545 | if (page) |
4546 | goto got_pg; | |
75f30861 | 4547 | |
9083905a JW |
4548 | /* Do not loop if specifically requested */ |
4549 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4550 | goto nopage; |
9083905a | 4551 | |
0a0337e0 MH |
4552 | /* |
4553 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4554 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4555 | */ |
dcda9b04 | 4556 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4557 | goto nopage; |
0a0337e0 | 4558 | |
0a0337e0 | 4559 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4560 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4561 | goto retry; |
4562 | ||
33c2d214 MH |
4563 | /* |
4564 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4565 | * reclaim is not able to make any progress because the current | |
4566 | * implementation of the compaction depends on the sufficient amount | |
4567 | * of free memory (see __compaction_suitable) | |
4568 | */ | |
4569 | if (did_some_progress > 0 && | |
86a294a8 | 4570 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4571 | compact_result, &compact_priority, |
d9436498 | 4572 | &compaction_retries)) |
33c2d214 MH |
4573 | goto retry; |
4574 | ||
902b6281 VB |
4575 | |
4576 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4577 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4578 | goto retry_cpuset; |
4579 | ||
9083905a JW |
4580 | /* Reclaim has failed us, start killing things */ |
4581 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4582 | if (page) | |
4583 | goto got_pg; | |
4584 | ||
9a67f648 | 4585 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e MH |
4586 | if (tsk_is_oom_victim(current) && |
4587 | (alloc_flags == ALLOC_OOM || | |
c288983d | 4588 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4589 | goto nopage; |
4590 | ||
9083905a | 4591 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4592 | if (did_some_progress) { |
4593 | no_progress_loops = 0; | |
9083905a | 4594 | goto retry; |
0a0337e0 | 4595 | } |
9083905a | 4596 | |
1da177e4 | 4597 | nopage: |
902b6281 VB |
4598 | /* Deal with possible cpuset update races before we fail */ |
4599 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4600 | goto retry_cpuset; |
4601 | ||
9a67f648 MH |
4602 | /* |
4603 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4604 | * we always retry | |
4605 | */ | |
4606 | if (gfp_mask & __GFP_NOFAIL) { | |
4607 | /* | |
4608 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4609 | * of any new users that actually require GFP_NOWAIT | |
4610 | */ | |
4611 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4612 | goto fail; | |
4613 | ||
4614 | /* | |
4615 | * PF_MEMALLOC request from this context is rather bizarre | |
4616 | * because we cannot reclaim anything and only can loop waiting | |
4617 | * for somebody to do a work for us | |
4618 | */ | |
4619 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4620 | ||
4621 | /* | |
4622 | * non failing costly orders are a hard requirement which we | |
4623 | * are not prepared for much so let's warn about these users | |
4624 | * so that we can identify them and convert them to something | |
4625 | * else. | |
4626 | */ | |
4627 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4628 | ||
6c18ba7a MH |
4629 | /* |
4630 | * Help non-failing allocations by giving them access to memory | |
4631 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4632 | * could deplete whole memory reserves which would just make | |
4633 | * the situation worse | |
4634 | */ | |
4635 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4636 | if (page) | |
4637 | goto got_pg; | |
4638 | ||
9a67f648 MH |
4639 | cond_resched(); |
4640 | goto retry; | |
4641 | } | |
4642 | fail: | |
a8e99259 | 4643 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4644 | "page allocation failure: order:%u", order); |
1da177e4 | 4645 | got_pg: |
072bb0aa | 4646 | return page; |
1da177e4 | 4647 | } |
11e33f6a | 4648 | |
9cd75558 | 4649 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4650 | int preferred_nid, nodemask_t *nodemask, |
9cd75558 MG |
4651 | struct alloc_context *ac, gfp_t *alloc_mask, |
4652 | unsigned int *alloc_flags) | |
11e33f6a | 4653 | { |
9cd75558 | 4654 | ac->high_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4655 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 MG |
4656 | ac->nodemask = nodemask; |
4657 | ac->migratetype = gfpflags_to_migratetype(gfp_mask); | |
11e33f6a | 4658 | |
682a3385 | 4659 | if (cpusets_enabled()) { |
9cd75558 | 4660 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
4661 | if (!ac->nodemask) |
4662 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
4663 | else |
4664 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4665 | } |
4666 | ||
d92a8cfc PZ |
4667 | fs_reclaim_acquire(gfp_mask); |
4668 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4669 | |
d0164adc | 4670 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4671 | |
4672 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4673 | return false; |
11e33f6a | 4674 | |
d883c6cf JK |
4675 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
4676 | *alloc_flags |= ALLOC_CMA; | |
4677 | ||
9cd75558 MG |
4678 | return true; |
4679 | } | |
21bb9bd1 | 4680 | |
9cd75558 | 4681 | /* Determine whether to spread dirty pages and what the first usable zone */ |
a380b40a | 4682 | static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac) |
9cd75558 | 4683 | { |
c9ab0c4f | 4684 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4685 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4686 | |
e46e7b77 MG |
4687 | /* |
4688 | * The preferred zone is used for statistics but crucially it is | |
4689 | * also used as the starting point for the zonelist iterator. It | |
4690 | * may get reset for allocations that ignore memory policies. | |
4691 | */ | |
9cd75558 MG |
4692 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4693 | ac->high_zoneidx, ac->nodemask); | |
4694 | } | |
4695 | ||
4696 | /* | |
4697 | * This is the 'heart' of the zoned buddy allocator. | |
4698 | */ | |
4699 | struct page * | |
04ec6264 VB |
4700 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
4701 | nodemask_t *nodemask) | |
9cd75558 MG |
4702 | { |
4703 | struct page *page; | |
4704 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
f19360f0 | 4705 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4706 | struct alloc_context ac = { }; |
4707 | ||
c63ae43b MH |
4708 | /* |
4709 | * There are several places where we assume that the order value is sane | |
4710 | * so bail out early if the request is out of bound. | |
4711 | */ | |
4712 | if (unlikely(order >= MAX_ORDER)) { | |
4713 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
4714 | return NULL; | |
4715 | } | |
4716 | ||
9cd75558 | 4717 | gfp_mask &= gfp_allowed_mask; |
f19360f0 | 4718 | alloc_mask = gfp_mask; |
04ec6264 | 4719 | if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags)) |
9cd75558 MG |
4720 | return NULL; |
4721 | ||
a380b40a | 4722 | finalise_ac(gfp_mask, &ac); |
5bb1b169 | 4723 | |
6bb15450 MG |
4724 | /* |
4725 | * Forbid the first pass from falling back to types that fragment | |
4726 | * memory until all local zones are considered. | |
4727 | */ | |
0a79cdad | 4728 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask); |
6bb15450 | 4729 | |
5117f45d | 4730 | /* First allocation attempt */ |
a9263751 | 4731 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4732 | if (likely(page)) |
4733 | goto out; | |
11e33f6a | 4734 | |
4fcb0971 | 4735 | /* |
7dea19f9 MH |
4736 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4737 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4738 | * from a particular context which has been marked by | |
4739 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4740 | */ |
7dea19f9 | 4741 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4742 | ac.spread_dirty_pages = false; |
23f086f9 | 4743 | |
4741526b MG |
4744 | /* |
4745 | * Restore the original nodemask if it was potentially replaced with | |
4746 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4747 | */ | |
e47483bc | 4748 | if (unlikely(ac.nodemask != nodemask)) |
4741526b | 4749 | ac.nodemask = nodemask; |
16096c25 | 4750 | |
4fcb0971 | 4751 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4752 | |
4fcb0971 | 4753 | out: |
c4159a75 | 4754 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
60cd4bcd | 4755 | unlikely(__memcg_kmem_charge(page, gfp_mask, order) != 0)) { |
c4159a75 VD |
4756 | __free_pages(page, order); |
4757 | page = NULL; | |
4949148a VD |
4758 | } |
4759 | ||
4fcb0971 MG |
4760 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
4761 | ||
11e33f6a | 4762 | return page; |
1da177e4 | 4763 | } |
d239171e | 4764 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4765 | |
4766 | /* | |
9ea9a680 MH |
4767 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4768 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4769 | * you need to access high mem. | |
1da177e4 | 4770 | */ |
920c7a5d | 4771 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4772 | { |
945a1113 AM |
4773 | struct page *page; |
4774 | ||
9ea9a680 | 4775 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4776 | if (!page) |
4777 | return 0; | |
4778 | return (unsigned long) page_address(page); | |
4779 | } | |
1da177e4 LT |
4780 | EXPORT_SYMBOL(__get_free_pages); |
4781 | ||
920c7a5d | 4782 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4783 | { |
945a1113 | 4784 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4785 | } |
1da177e4 LT |
4786 | EXPORT_SYMBOL(get_zeroed_page); |
4787 | ||
742aa7fb | 4788 | static inline void free_the_page(struct page *page, unsigned int order) |
1da177e4 | 4789 | { |
742aa7fb AL |
4790 | if (order == 0) /* Via pcp? */ |
4791 | free_unref_page(page); | |
4792 | else | |
4793 | __free_pages_ok(page, order); | |
1da177e4 LT |
4794 | } |
4795 | ||
742aa7fb AL |
4796 | void __free_pages(struct page *page, unsigned int order) |
4797 | { | |
4798 | if (put_page_testzero(page)) | |
4799 | free_the_page(page, order); | |
4800 | } | |
1da177e4 LT |
4801 | EXPORT_SYMBOL(__free_pages); |
4802 | ||
920c7a5d | 4803 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4804 | { |
4805 | if (addr != 0) { | |
725d704e | 4806 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4807 | __free_pages(virt_to_page((void *)addr), order); |
4808 | } | |
4809 | } | |
4810 | ||
4811 | EXPORT_SYMBOL(free_pages); | |
4812 | ||
b63ae8ca AD |
4813 | /* |
4814 | * Page Fragment: | |
4815 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4816 | * within a 0 or higher order page. Multiple fragments within that page | |
4817 | * are individually refcounted, in the page's reference counter. | |
4818 | * | |
4819 | * The page_frag functions below provide a simple allocation framework for | |
4820 | * page fragments. This is used by the network stack and network device | |
4821 | * drivers to provide a backing region of memory for use as either an | |
4822 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4823 | */ | |
2976db80 AD |
4824 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4825 | gfp_t gfp_mask) | |
b63ae8ca AD |
4826 | { |
4827 | struct page *page = NULL; | |
4828 | gfp_t gfp = gfp_mask; | |
4829 | ||
4830 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4831 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4832 | __GFP_NOMEMALLOC; | |
4833 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4834 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4835 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4836 | #endif | |
4837 | if (unlikely(!page)) | |
4838 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4839 | ||
4840 | nc->va = page ? page_address(page) : NULL; | |
4841 | ||
4842 | return page; | |
4843 | } | |
4844 | ||
2976db80 | 4845 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4846 | { |
4847 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4848 | ||
742aa7fb AL |
4849 | if (page_ref_sub_and_test(page, count)) |
4850 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 4851 | } |
2976db80 | 4852 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4853 | |
8c2dd3e4 AD |
4854 | void *page_frag_alloc(struct page_frag_cache *nc, |
4855 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4856 | { |
4857 | unsigned int size = PAGE_SIZE; | |
4858 | struct page *page; | |
4859 | int offset; | |
4860 | ||
4861 | if (unlikely(!nc->va)) { | |
4862 | refill: | |
2976db80 | 4863 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4864 | if (!page) |
4865 | return NULL; | |
4866 | ||
4867 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4868 | /* if size can vary use size else just use PAGE_SIZE */ | |
4869 | size = nc->size; | |
4870 | #endif | |
4871 | /* Even if we own the page, we do not use atomic_set(). | |
4872 | * This would break get_page_unless_zero() users. | |
4873 | */ | |
86447726 | 4874 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
4875 | |
4876 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4877 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 4878 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
4879 | nc->offset = size; |
4880 | } | |
4881 | ||
4882 | offset = nc->offset - fragsz; | |
4883 | if (unlikely(offset < 0)) { | |
4884 | page = virt_to_page(nc->va); | |
4885 | ||
fe896d18 | 4886 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4887 | goto refill; |
4888 | ||
4889 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4890 | /* if size can vary use size else just use PAGE_SIZE */ | |
4891 | size = nc->size; | |
4892 | #endif | |
4893 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 4894 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
4895 | |
4896 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 4897 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
4898 | offset = size - fragsz; |
4899 | } | |
4900 | ||
4901 | nc->pagecnt_bias--; | |
4902 | nc->offset = offset; | |
4903 | ||
4904 | return nc->va + offset; | |
4905 | } | |
8c2dd3e4 | 4906 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
4907 | |
4908 | /* | |
4909 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4910 | */ | |
8c2dd3e4 | 4911 | void page_frag_free(void *addr) |
b63ae8ca AD |
4912 | { |
4913 | struct page *page = virt_to_head_page(addr); | |
4914 | ||
742aa7fb AL |
4915 | if (unlikely(put_page_testzero(page))) |
4916 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 4917 | } |
8c2dd3e4 | 4918 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4919 | |
d00181b9 KS |
4920 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4921 | size_t size) | |
ee85c2e1 AK |
4922 | { |
4923 | if (addr) { | |
4924 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
4925 | unsigned long used = addr + PAGE_ALIGN(size); | |
4926 | ||
4927 | split_page(virt_to_page((void *)addr), order); | |
4928 | while (used < alloc_end) { | |
4929 | free_page(used); | |
4930 | used += PAGE_SIZE; | |
4931 | } | |
4932 | } | |
4933 | return (void *)addr; | |
4934 | } | |
4935 | ||
2be0ffe2 TT |
4936 | /** |
4937 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
4938 | * @size: the number of bytes to allocate | |
63931eb9 | 4939 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
4940 | * |
4941 | * This function is similar to alloc_pages(), except that it allocates the | |
4942 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
4943 | * allocate memory in power-of-two pages. | |
4944 | * | |
4945 | * This function is also limited by MAX_ORDER. | |
4946 | * | |
4947 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
4948 | * |
4949 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
4950 | */ |
4951 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
4952 | { | |
4953 | unsigned int order = get_order(size); | |
4954 | unsigned long addr; | |
4955 | ||
63931eb9 VB |
4956 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) |
4957 | gfp_mask &= ~__GFP_COMP; | |
4958 | ||
2be0ffe2 | 4959 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 4960 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
4961 | } |
4962 | EXPORT_SYMBOL(alloc_pages_exact); | |
4963 | ||
ee85c2e1 AK |
4964 | /** |
4965 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
4966 | * pages on a node. | |
b5e6ab58 | 4967 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 4968 | * @size: the number of bytes to allocate |
63931eb9 | 4969 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
4970 | * |
4971 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
4972 | * back. | |
a862f68a MR |
4973 | * |
4974 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 4975 | */ |
e1931811 | 4976 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 4977 | { |
d00181b9 | 4978 | unsigned int order = get_order(size); |
63931eb9 VB |
4979 | struct page *p; |
4980 | ||
4981 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) | |
4982 | gfp_mask &= ~__GFP_COMP; | |
4983 | ||
4984 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
4985 | if (!p) |
4986 | return NULL; | |
4987 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
4988 | } | |
ee85c2e1 | 4989 | |
2be0ffe2 TT |
4990 | /** |
4991 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
4992 | * @virt: the value returned by alloc_pages_exact. | |
4993 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
4994 | * | |
4995 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
4996 | */ | |
4997 | void free_pages_exact(void *virt, size_t size) | |
4998 | { | |
4999 | unsigned long addr = (unsigned long)virt; | |
5000 | unsigned long end = addr + PAGE_ALIGN(size); | |
5001 | ||
5002 | while (addr < end) { | |
5003 | free_page(addr); | |
5004 | addr += PAGE_SIZE; | |
5005 | } | |
5006 | } | |
5007 | EXPORT_SYMBOL(free_pages_exact); | |
5008 | ||
e0fb5815 ZY |
5009 | /** |
5010 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5011 | * @offset: The zone index of the highest zone | |
5012 | * | |
a862f68a | 5013 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5014 | * high watermark within all zones at or below a given zone index. For each |
5015 | * zone, the number of pages is calculated as: | |
0e056eb5 | 5016 | * |
5017 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5018 | * |
5019 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5020 | */ |
ebec3862 | 5021 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5022 | { |
dd1a239f | 5023 | struct zoneref *z; |
54a6eb5c MG |
5024 | struct zone *zone; |
5025 | ||
e310fd43 | 5026 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5027 | unsigned long sum = 0; |
1da177e4 | 5028 | |
0e88460d | 5029 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5030 | |
54a6eb5c | 5031 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5032 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5033 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5034 | if (size > high) |
5035 | sum += size - high; | |
1da177e4 LT |
5036 | } |
5037 | ||
5038 | return sum; | |
5039 | } | |
5040 | ||
e0fb5815 ZY |
5041 | /** |
5042 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5043 | * | |
5044 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5045 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5046 | * |
5047 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5048 | * ZONE_NORMAL. | |
1da177e4 | 5049 | */ |
ebec3862 | 5050 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5051 | { |
af4ca457 | 5052 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5053 | } |
c2f1a551 | 5054 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5055 | |
e0fb5815 ZY |
5056 | /** |
5057 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
5058 | * | |
5059 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
5060 | * high watermark within all zones. | |
a862f68a MR |
5061 | * |
5062 | * Return: number of pages beyond high watermark within all zones. | |
1da177e4 | 5063 | */ |
ebec3862 | 5064 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 5065 | { |
2a1e274a | 5066 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 5067 | } |
08e0f6a9 CL |
5068 | |
5069 | static inline void show_node(struct zone *zone) | |
1da177e4 | 5070 | { |
e5adfffc | 5071 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5072 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5073 | } |
1da177e4 | 5074 | |
d02bd27b IR |
5075 | long si_mem_available(void) |
5076 | { | |
5077 | long available; | |
5078 | unsigned long pagecache; | |
5079 | unsigned long wmark_low = 0; | |
5080 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5081 | unsigned long reclaimable; |
d02bd27b IR |
5082 | struct zone *zone; |
5083 | int lru; | |
5084 | ||
5085 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5086 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5087 | |
5088 | for_each_zone(zone) | |
a9214443 | 5089 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5090 | |
5091 | /* | |
5092 | * Estimate the amount of memory available for userspace allocations, | |
5093 | * without causing swapping. | |
5094 | */ | |
c41f012a | 5095 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5096 | |
5097 | /* | |
5098 | * Not all the page cache can be freed, otherwise the system will | |
5099 | * start swapping. Assume at least half of the page cache, or the | |
5100 | * low watermark worth of cache, needs to stay. | |
5101 | */ | |
5102 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
5103 | pagecache -= min(pagecache / 2, wmark_low); | |
5104 | available += pagecache; | |
5105 | ||
5106 | /* | |
b29940c1 VB |
5107 | * Part of the reclaimable slab and other kernel memory consists of |
5108 | * items that are in use, and cannot be freed. Cap this estimate at the | |
5109 | * low watermark. | |
d02bd27b | 5110 | */ |
b29940c1 VB |
5111 | reclaimable = global_node_page_state(NR_SLAB_RECLAIMABLE) + |
5112 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
5113 | available += reclaimable - min(reclaimable / 2, wmark_low); | |
034ebf65 | 5114 | |
d02bd27b IR |
5115 | if (available < 0) |
5116 | available = 0; | |
5117 | return available; | |
5118 | } | |
5119 | EXPORT_SYMBOL_GPL(si_mem_available); | |
5120 | ||
1da177e4 LT |
5121 | void si_meminfo(struct sysinfo *val) |
5122 | { | |
ca79b0c2 | 5123 | val->totalram = totalram_pages(); |
11fb9989 | 5124 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 5125 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 5126 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 5127 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 5128 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
5129 | val->mem_unit = PAGE_SIZE; |
5130 | } | |
5131 | ||
5132 | EXPORT_SYMBOL(si_meminfo); | |
5133 | ||
5134 | #ifdef CONFIG_NUMA | |
5135 | void si_meminfo_node(struct sysinfo *val, int nid) | |
5136 | { | |
cdd91a77 JL |
5137 | int zone_type; /* needs to be signed */ |
5138 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
5139 | unsigned long managed_highpages = 0; |
5140 | unsigned long free_highpages = 0; | |
1da177e4 LT |
5141 | pg_data_t *pgdat = NODE_DATA(nid); |
5142 | ||
cdd91a77 | 5143 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 5144 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 5145 | val->totalram = managed_pages; |
11fb9989 | 5146 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 5147 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 5148 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
5149 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
5150 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
5151 | ||
5152 | if (is_highmem(zone)) { | |
9705bea5 | 5153 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
5154 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
5155 | } | |
5156 | } | |
5157 | val->totalhigh = managed_highpages; | |
5158 | val->freehigh = free_highpages; | |
98d2b0eb | 5159 | #else |
fc2bd799 JK |
5160 | val->totalhigh = managed_highpages; |
5161 | val->freehigh = free_highpages; | |
98d2b0eb | 5162 | #endif |
1da177e4 LT |
5163 | val->mem_unit = PAGE_SIZE; |
5164 | } | |
5165 | #endif | |
5166 | ||
ddd588b5 | 5167 | /* |
7bf02ea2 DR |
5168 | * Determine whether the node should be displayed or not, depending on whether |
5169 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 5170 | */ |
9af744d7 | 5171 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 5172 | { |
ddd588b5 | 5173 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 5174 | return false; |
ddd588b5 | 5175 | |
9af744d7 MH |
5176 | /* |
5177 | * no node mask - aka implicit memory numa policy. Do not bother with | |
5178 | * the synchronization - read_mems_allowed_begin - because we do not | |
5179 | * have to be precise here. | |
5180 | */ | |
5181 | if (!nodemask) | |
5182 | nodemask = &cpuset_current_mems_allowed; | |
5183 | ||
5184 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
5185 | } |
5186 | ||
1da177e4 LT |
5187 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
5188 | ||
377e4f16 RV |
5189 | static void show_migration_types(unsigned char type) |
5190 | { | |
5191 | static const char types[MIGRATE_TYPES] = { | |
5192 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 5193 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
5194 | [MIGRATE_RECLAIMABLE] = 'E', |
5195 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
5196 | #ifdef CONFIG_CMA |
5197 | [MIGRATE_CMA] = 'C', | |
5198 | #endif | |
194159fb | 5199 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 5200 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 5201 | #endif |
377e4f16 RV |
5202 | }; |
5203 | char tmp[MIGRATE_TYPES + 1]; | |
5204 | char *p = tmp; | |
5205 | int i; | |
5206 | ||
5207 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
5208 | if (type & (1 << i)) | |
5209 | *p++ = types[i]; | |
5210 | } | |
5211 | ||
5212 | *p = '\0'; | |
1f84a18f | 5213 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
5214 | } |
5215 | ||
1da177e4 LT |
5216 | /* |
5217 | * Show free area list (used inside shift_scroll-lock stuff) | |
5218 | * We also calculate the percentage fragmentation. We do this by counting the | |
5219 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
5220 | * |
5221 | * Bits in @filter: | |
5222 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
5223 | * cpuset. | |
1da177e4 | 5224 | */ |
9af744d7 | 5225 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 5226 | { |
d1bfcdb8 | 5227 | unsigned long free_pcp = 0; |
c7241913 | 5228 | int cpu; |
1da177e4 | 5229 | struct zone *zone; |
599d0c95 | 5230 | pg_data_t *pgdat; |
1da177e4 | 5231 | |
ee99c71c | 5232 | for_each_populated_zone(zone) { |
9af744d7 | 5233 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5234 | continue; |
d1bfcdb8 | 5235 | |
761b0677 KK |
5236 | for_each_online_cpu(cpu) |
5237 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
5238 | } |
5239 | ||
a731286d KM |
5240 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
5241 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
5242 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
5243 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 5244 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 5245 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
5246 | global_node_page_state(NR_ACTIVE_ANON), |
5247 | global_node_page_state(NR_INACTIVE_ANON), | |
5248 | global_node_page_state(NR_ISOLATED_ANON), | |
5249 | global_node_page_state(NR_ACTIVE_FILE), | |
5250 | global_node_page_state(NR_INACTIVE_FILE), | |
5251 | global_node_page_state(NR_ISOLATED_FILE), | |
5252 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
5253 | global_node_page_state(NR_FILE_DIRTY), |
5254 | global_node_page_state(NR_WRITEBACK), | |
5255 | global_node_page_state(NR_UNSTABLE_NFS), | |
d507e2eb JW |
5256 | global_node_page_state(NR_SLAB_RECLAIMABLE), |
5257 | global_node_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 5258 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 5259 | global_node_page_state(NR_SHMEM), |
c41f012a MH |
5260 | global_zone_page_state(NR_PAGETABLE), |
5261 | global_zone_page_state(NR_BOUNCE), | |
5262 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 5263 | free_pcp, |
c41f012a | 5264 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 5265 | |
599d0c95 | 5266 | for_each_online_pgdat(pgdat) { |
9af744d7 | 5267 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
5268 | continue; |
5269 | ||
599d0c95 MG |
5270 | printk("Node %d" |
5271 | " active_anon:%lukB" | |
5272 | " inactive_anon:%lukB" | |
5273 | " active_file:%lukB" | |
5274 | " inactive_file:%lukB" | |
5275 | " unevictable:%lukB" | |
5276 | " isolated(anon):%lukB" | |
5277 | " isolated(file):%lukB" | |
50658e2e | 5278 | " mapped:%lukB" |
11fb9989 MG |
5279 | " dirty:%lukB" |
5280 | " writeback:%lukB" | |
5281 | " shmem:%lukB" | |
5282 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5283 | " shmem_thp: %lukB" | |
5284 | " shmem_pmdmapped: %lukB" | |
5285 | " anon_thp: %lukB" | |
5286 | #endif | |
5287 | " writeback_tmp:%lukB" | |
5288 | " unstable:%lukB" | |
599d0c95 MG |
5289 | " all_unreclaimable? %s" |
5290 | "\n", | |
5291 | pgdat->node_id, | |
5292 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
5293 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
5294 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
5295 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
5296 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
5297 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
5298 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 5299 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
5300 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
5301 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 5302 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
5303 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5304 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
5305 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
5306 | * HPAGE_PMD_NR), | |
5307 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
5308 | #endif | |
11fb9989 MG |
5309 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
5310 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), | |
c73322d0 JW |
5311 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
5312 | "yes" : "no"); | |
599d0c95 MG |
5313 | } |
5314 | ||
ee99c71c | 5315 | for_each_populated_zone(zone) { |
1da177e4 LT |
5316 | int i; |
5317 | ||
9af744d7 | 5318 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5319 | continue; |
d1bfcdb8 KK |
5320 | |
5321 | free_pcp = 0; | |
5322 | for_each_online_cpu(cpu) | |
5323 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
5324 | ||
1da177e4 | 5325 | show_node(zone); |
1f84a18f JP |
5326 | printk(KERN_CONT |
5327 | "%s" | |
1da177e4 LT |
5328 | " free:%lukB" |
5329 | " min:%lukB" | |
5330 | " low:%lukB" | |
5331 | " high:%lukB" | |
e47b346a | 5332 | " reserved_highatomic:%luKB" |
71c799f4 MK |
5333 | " active_anon:%lukB" |
5334 | " inactive_anon:%lukB" | |
5335 | " active_file:%lukB" | |
5336 | " inactive_file:%lukB" | |
5337 | " unevictable:%lukB" | |
5a1c84b4 | 5338 | " writepending:%lukB" |
1da177e4 | 5339 | " present:%lukB" |
9feedc9d | 5340 | " managed:%lukB" |
4a0aa73f | 5341 | " mlocked:%lukB" |
c6a7f572 | 5342 | " kernel_stack:%lukB" |
4a0aa73f | 5343 | " pagetables:%lukB" |
4a0aa73f | 5344 | " bounce:%lukB" |
d1bfcdb8 KK |
5345 | " free_pcp:%lukB" |
5346 | " local_pcp:%ukB" | |
d1ce749a | 5347 | " free_cma:%lukB" |
1da177e4 LT |
5348 | "\n", |
5349 | zone->name, | |
88f5acf8 | 5350 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
5351 | K(min_wmark_pages(zone)), |
5352 | K(low_wmark_pages(zone)), | |
5353 | K(high_wmark_pages(zone)), | |
e47b346a | 5354 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
5355 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
5356 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
5357 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
5358 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
5359 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 5360 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 5361 | K(zone->present_pages), |
9705bea5 | 5362 | K(zone_managed_pages(zone)), |
4a0aa73f | 5363 | K(zone_page_state(zone, NR_MLOCK)), |
d30dd8be | 5364 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4a0aa73f | 5365 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 5366 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
5367 | K(free_pcp), |
5368 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 5369 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
5370 | printk("lowmem_reserve[]:"); |
5371 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
5372 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5373 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5374 | } |
5375 | ||
ee99c71c | 5376 | for_each_populated_zone(zone) { |
d00181b9 KS |
5377 | unsigned int order; |
5378 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5379 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5380 | |
9af744d7 | 5381 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5382 | continue; |
1da177e4 | 5383 | show_node(zone); |
1f84a18f | 5384 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5385 | |
5386 | spin_lock_irqsave(&zone->lock, flags); | |
5387 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5388 | struct free_area *area = &zone->free_area[order]; |
5389 | int type; | |
5390 | ||
5391 | nr[order] = area->nr_free; | |
8f9de51a | 5392 | total += nr[order] << order; |
377e4f16 RV |
5393 | |
5394 | types[order] = 0; | |
5395 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 5396 | if (!free_area_empty(area, type)) |
377e4f16 RV |
5397 | types[order] |= 1 << type; |
5398 | } | |
1da177e4 LT |
5399 | } |
5400 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5401 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5402 | printk(KERN_CONT "%lu*%lukB ", |
5403 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5404 | if (nr[order]) |
5405 | show_migration_types(types[order]); | |
5406 | } | |
1f84a18f | 5407 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5408 | } |
5409 | ||
949f7ec5 DR |
5410 | hugetlb_show_meminfo(); |
5411 | ||
11fb9989 | 5412 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5413 | |
1da177e4 LT |
5414 | show_swap_cache_info(); |
5415 | } | |
5416 | ||
19770b32 MG |
5417 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5418 | { | |
5419 | zoneref->zone = zone; | |
5420 | zoneref->zone_idx = zone_idx(zone); | |
5421 | } | |
5422 | ||
1da177e4 LT |
5423 | /* |
5424 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5425 | * |
5426 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5427 | */ |
9d3be21b | 5428 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5429 | { |
1a93205b | 5430 | struct zone *zone; |
bc732f1d | 5431 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5432 | int nr_zones = 0; |
02a68a5e CL |
5433 | |
5434 | do { | |
2f6726e5 | 5435 | zone_type--; |
070f8032 | 5436 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5437 | if (managed_zone(zone)) { |
9d3be21b | 5438 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5439 | check_highest_zone(zone_type); |
1da177e4 | 5440 | } |
2f6726e5 | 5441 | } while (zone_type); |
bc732f1d | 5442 | |
070f8032 | 5443 | return nr_zones; |
1da177e4 LT |
5444 | } |
5445 | ||
5446 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5447 | |
5448 | static int __parse_numa_zonelist_order(char *s) | |
5449 | { | |
c9bff3ee MH |
5450 | /* |
5451 | * We used to support different zonlists modes but they turned | |
5452 | * out to be just not useful. Let's keep the warning in place | |
5453 | * if somebody still use the cmd line parameter so that we do | |
5454 | * not fail it silently | |
5455 | */ | |
5456 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5457 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5458 | return -EINVAL; |
5459 | } | |
5460 | return 0; | |
5461 | } | |
5462 | ||
5463 | static __init int setup_numa_zonelist_order(char *s) | |
5464 | { | |
ecb256f8 VL |
5465 | if (!s) |
5466 | return 0; | |
5467 | ||
c9bff3ee | 5468 | return __parse_numa_zonelist_order(s); |
f0c0b2b8 KH |
5469 | } |
5470 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
5471 | ||
c9bff3ee MH |
5472 | char numa_zonelist_order[] = "Node"; |
5473 | ||
f0c0b2b8 KH |
5474 | /* |
5475 | * sysctl handler for numa_zonelist_order | |
5476 | */ | |
cccad5b9 | 5477 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 5478 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
5479 | loff_t *ppos) |
5480 | { | |
c9bff3ee | 5481 | char *str; |
f0c0b2b8 KH |
5482 | int ret; |
5483 | ||
c9bff3ee MH |
5484 | if (!write) |
5485 | return proc_dostring(table, write, buffer, length, ppos); | |
5486 | str = memdup_user_nul(buffer, 16); | |
5487 | if (IS_ERR(str)) | |
5488 | return PTR_ERR(str); | |
dacbde09 | 5489 | |
c9bff3ee MH |
5490 | ret = __parse_numa_zonelist_order(str); |
5491 | kfree(str); | |
443c6f14 | 5492 | return ret; |
f0c0b2b8 KH |
5493 | } |
5494 | ||
5495 | ||
62bc62a8 | 5496 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5497 | static int node_load[MAX_NUMNODES]; |
5498 | ||
1da177e4 | 5499 | /** |
4dc3b16b | 5500 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5501 | * @node: node whose fallback list we're appending |
5502 | * @used_node_mask: nodemask_t of already used nodes | |
5503 | * | |
5504 | * We use a number of factors to determine which is the next node that should | |
5505 | * appear on a given node's fallback list. The node should not have appeared | |
5506 | * already in @node's fallback list, and it should be the next closest node | |
5507 | * according to the distance array (which contains arbitrary distance values | |
5508 | * from each node to each node in the system), and should also prefer nodes | |
5509 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5510 | * on them otherwise. | |
a862f68a MR |
5511 | * |
5512 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 5513 | */ |
f0c0b2b8 | 5514 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5515 | { |
4cf808eb | 5516 | int n, val; |
1da177e4 | 5517 | int min_val = INT_MAX; |
00ef2d2f | 5518 | int best_node = NUMA_NO_NODE; |
a70f7302 | 5519 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 5520 | |
4cf808eb LT |
5521 | /* Use the local node if we haven't already */ |
5522 | if (!node_isset(node, *used_node_mask)) { | |
5523 | node_set(node, *used_node_mask); | |
5524 | return node; | |
5525 | } | |
1da177e4 | 5526 | |
4b0ef1fe | 5527 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5528 | |
5529 | /* Don't want a node to appear more than once */ | |
5530 | if (node_isset(n, *used_node_mask)) | |
5531 | continue; | |
5532 | ||
1da177e4 LT |
5533 | /* Use the distance array to find the distance */ |
5534 | val = node_distance(node, n); | |
5535 | ||
4cf808eb LT |
5536 | /* Penalize nodes under us ("prefer the next node") */ |
5537 | val += (n < node); | |
5538 | ||
1da177e4 | 5539 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
5540 | tmp = cpumask_of_node(n); |
5541 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
5542 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5543 | ||
5544 | /* Slight preference for less loaded node */ | |
5545 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
5546 | val += node_load[n]; | |
5547 | ||
5548 | if (val < min_val) { | |
5549 | min_val = val; | |
5550 | best_node = n; | |
5551 | } | |
5552 | } | |
5553 | ||
5554 | if (best_node >= 0) | |
5555 | node_set(best_node, *used_node_mask); | |
5556 | ||
5557 | return best_node; | |
5558 | } | |
5559 | ||
f0c0b2b8 KH |
5560 | |
5561 | /* | |
5562 | * Build zonelists ordered by node and zones within node. | |
5563 | * This results in maximum locality--normal zone overflows into local | |
5564 | * DMA zone, if any--but risks exhausting DMA zone. | |
5565 | */ | |
9d3be21b MH |
5566 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5567 | unsigned nr_nodes) | |
1da177e4 | 5568 | { |
9d3be21b MH |
5569 | struct zoneref *zonerefs; |
5570 | int i; | |
5571 | ||
5572 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5573 | ||
5574 | for (i = 0; i < nr_nodes; i++) { | |
5575 | int nr_zones; | |
5576 | ||
5577 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5578 | |
9d3be21b MH |
5579 | nr_zones = build_zonerefs_node(node, zonerefs); |
5580 | zonerefs += nr_zones; | |
5581 | } | |
5582 | zonerefs->zone = NULL; | |
5583 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5584 | } |
5585 | ||
523b9458 CL |
5586 | /* |
5587 | * Build gfp_thisnode zonelists | |
5588 | */ | |
5589 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5590 | { | |
9d3be21b MH |
5591 | struct zoneref *zonerefs; |
5592 | int nr_zones; | |
523b9458 | 5593 | |
9d3be21b MH |
5594 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5595 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5596 | zonerefs += nr_zones; | |
5597 | zonerefs->zone = NULL; | |
5598 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5599 | } |
5600 | ||
f0c0b2b8 KH |
5601 | /* |
5602 | * Build zonelists ordered by zone and nodes within zones. | |
5603 | * This results in conserving DMA zone[s] until all Normal memory is | |
5604 | * exhausted, but results in overflowing to remote node while memory | |
5605 | * may still exist in local DMA zone. | |
5606 | */ | |
f0c0b2b8 | 5607 | |
f0c0b2b8 KH |
5608 | static void build_zonelists(pg_data_t *pgdat) |
5609 | { | |
9d3be21b MH |
5610 | static int node_order[MAX_NUMNODES]; |
5611 | int node, load, nr_nodes = 0; | |
1da177e4 | 5612 | nodemask_t used_mask; |
f0c0b2b8 | 5613 | int local_node, prev_node; |
1da177e4 LT |
5614 | |
5615 | /* NUMA-aware ordering of nodes */ | |
5616 | local_node = pgdat->node_id; | |
62bc62a8 | 5617 | load = nr_online_nodes; |
1da177e4 LT |
5618 | prev_node = local_node; |
5619 | nodes_clear(used_mask); | |
f0c0b2b8 | 5620 | |
f0c0b2b8 | 5621 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5622 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5623 | /* | |
5624 | * We don't want to pressure a particular node. | |
5625 | * So adding penalty to the first node in same | |
5626 | * distance group to make it round-robin. | |
5627 | */ | |
957f822a DR |
5628 | if (node_distance(local_node, node) != |
5629 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5630 | node_load[node] = load; |
5631 | ||
9d3be21b | 5632 | node_order[nr_nodes++] = node; |
1da177e4 LT |
5633 | prev_node = node; |
5634 | load--; | |
1da177e4 | 5635 | } |
523b9458 | 5636 | |
9d3be21b | 5637 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5638 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
5639 | } |
5640 | ||
7aac7898 LS |
5641 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5642 | /* | |
5643 | * Return node id of node used for "local" allocations. | |
5644 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5645 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5646 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5647 | */ | |
5648 | int local_memory_node(int node) | |
5649 | { | |
c33d6c06 | 5650 | struct zoneref *z; |
7aac7898 | 5651 | |
c33d6c06 | 5652 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5653 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 5654 | NULL); |
c1093b74 | 5655 | return zone_to_nid(z->zone); |
7aac7898 LS |
5656 | } |
5657 | #endif | |
f0c0b2b8 | 5658 | |
6423aa81 JK |
5659 | static void setup_min_unmapped_ratio(void); |
5660 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5661 | #else /* CONFIG_NUMA */ |
5662 | ||
f0c0b2b8 | 5663 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5664 | { |
19655d34 | 5665 | int node, local_node; |
9d3be21b MH |
5666 | struct zoneref *zonerefs; |
5667 | int nr_zones; | |
1da177e4 LT |
5668 | |
5669 | local_node = pgdat->node_id; | |
1da177e4 | 5670 | |
9d3be21b MH |
5671 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5672 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5673 | zonerefs += nr_zones; | |
1da177e4 | 5674 | |
54a6eb5c MG |
5675 | /* |
5676 | * Now we build the zonelist so that it contains the zones | |
5677 | * of all the other nodes. | |
5678 | * We don't want to pressure a particular node, so when | |
5679 | * building the zones for node N, we make sure that the | |
5680 | * zones coming right after the local ones are those from | |
5681 | * node N+1 (modulo N) | |
5682 | */ | |
5683 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5684 | if (!node_online(node)) | |
5685 | continue; | |
9d3be21b MH |
5686 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5687 | zonerefs += nr_zones; | |
1da177e4 | 5688 | } |
54a6eb5c MG |
5689 | for (node = 0; node < local_node; node++) { |
5690 | if (!node_online(node)) | |
5691 | continue; | |
9d3be21b MH |
5692 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5693 | zonerefs += nr_zones; | |
54a6eb5c MG |
5694 | } |
5695 | ||
9d3be21b MH |
5696 | zonerefs->zone = NULL; |
5697 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5698 | } |
5699 | ||
5700 | #endif /* CONFIG_NUMA */ | |
5701 | ||
99dcc3e5 CL |
5702 | /* |
5703 | * Boot pageset table. One per cpu which is going to be used for all | |
5704 | * zones and all nodes. The parameters will be set in such a way | |
5705 | * that an item put on a list will immediately be handed over to | |
5706 | * the buddy list. This is safe since pageset manipulation is done | |
5707 | * with interrupts disabled. | |
5708 | * | |
5709 | * The boot_pagesets must be kept even after bootup is complete for | |
5710 | * unused processors and/or zones. They do play a role for bootstrapping | |
5711 | * hotplugged processors. | |
5712 | * | |
5713 | * zoneinfo_show() and maybe other functions do | |
5714 | * not check if the processor is online before following the pageset pointer. | |
5715 | * Other parts of the kernel may not check if the zone is available. | |
5716 | */ | |
5717 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5718 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
385386cf | 5719 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 5720 | |
11cd8638 | 5721 | static void __build_all_zonelists(void *data) |
1da177e4 | 5722 | { |
6811378e | 5723 | int nid; |
afb6ebb3 | 5724 | int __maybe_unused cpu; |
9adb62a5 | 5725 | pg_data_t *self = data; |
b93e0f32 MH |
5726 | static DEFINE_SPINLOCK(lock); |
5727 | ||
5728 | spin_lock(&lock); | |
9276b1bc | 5729 | |
7f9cfb31 BL |
5730 | #ifdef CONFIG_NUMA |
5731 | memset(node_load, 0, sizeof(node_load)); | |
5732 | #endif | |
9adb62a5 | 5733 | |
c1152583 WY |
5734 | /* |
5735 | * This node is hotadded and no memory is yet present. So just | |
5736 | * building zonelists is fine - no need to touch other nodes. | |
5737 | */ | |
9adb62a5 JL |
5738 | if (self && !node_online(self->node_id)) { |
5739 | build_zonelists(self); | |
c1152583 WY |
5740 | } else { |
5741 | for_each_online_node(nid) { | |
5742 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 5743 | |
c1152583 WY |
5744 | build_zonelists(pgdat); |
5745 | } | |
99dcc3e5 | 5746 | |
7aac7898 LS |
5747 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5748 | /* | |
5749 | * We now know the "local memory node" for each node-- | |
5750 | * i.e., the node of the first zone in the generic zonelist. | |
5751 | * Set up numa_mem percpu variable for on-line cpus. During | |
5752 | * boot, only the boot cpu should be on-line; we'll init the | |
5753 | * secondary cpus' numa_mem as they come on-line. During | |
5754 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5755 | */ | |
d9c9a0b9 | 5756 | for_each_online_cpu(cpu) |
7aac7898 | 5757 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5758 | #endif |
d9c9a0b9 | 5759 | } |
b93e0f32 MH |
5760 | |
5761 | spin_unlock(&lock); | |
6811378e YG |
5762 | } |
5763 | ||
061f67bc RV |
5764 | static noinline void __init |
5765 | build_all_zonelists_init(void) | |
5766 | { | |
afb6ebb3 MH |
5767 | int cpu; |
5768 | ||
061f67bc | 5769 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5770 | |
5771 | /* | |
5772 | * Initialize the boot_pagesets that are going to be used | |
5773 | * for bootstrapping processors. The real pagesets for | |
5774 | * each zone will be allocated later when the per cpu | |
5775 | * allocator is available. | |
5776 | * | |
5777 | * boot_pagesets are used also for bootstrapping offline | |
5778 | * cpus if the system is already booted because the pagesets | |
5779 | * are needed to initialize allocators on a specific cpu too. | |
5780 | * F.e. the percpu allocator needs the page allocator which | |
5781 | * needs the percpu allocator in order to allocate its pagesets | |
5782 | * (a chicken-egg dilemma). | |
5783 | */ | |
5784 | for_each_possible_cpu(cpu) | |
5785 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); | |
5786 | ||
061f67bc RV |
5787 | mminit_verify_zonelist(); |
5788 | cpuset_init_current_mems_allowed(); | |
5789 | } | |
5790 | ||
4eaf3f64 | 5791 | /* |
4eaf3f64 | 5792 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5793 | * |
72675e13 | 5794 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5795 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5796 | */ |
72675e13 | 5797 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e YG |
5798 | { |
5799 | if (system_state == SYSTEM_BOOTING) { | |
061f67bc | 5800 | build_all_zonelists_init(); |
6811378e | 5801 | } else { |
11cd8638 | 5802 | __build_all_zonelists(pgdat); |
6811378e YG |
5803 | /* cpuset refresh routine should be here */ |
5804 | } | |
bd1e22b8 | 5805 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5806 | /* |
5807 | * Disable grouping by mobility if the number of pages in the | |
5808 | * system is too low to allow the mechanism to work. It would be | |
5809 | * more accurate, but expensive to check per-zone. This check is | |
5810 | * made on memory-hotadd so a system can start with mobility | |
5811 | * disabled and enable it later | |
5812 | */ | |
d9c23400 | 5813 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5814 | page_group_by_mobility_disabled = 1; |
5815 | else | |
5816 | page_group_by_mobility_disabled = 0; | |
5817 | ||
ce0725f7 | 5818 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5819 | nr_online_nodes, |
756a025f JP |
5820 | page_group_by_mobility_disabled ? "off" : "on", |
5821 | vm_total_pages); | |
f0c0b2b8 | 5822 | #ifdef CONFIG_NUMA |
f88dfff5 | 5823 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5824 | #endif |
1da177e4 LT |
5825 | } |
5826 | ||
a9a9e77f PT |
5827 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
5828 | static bool __meminit | |
5829 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
5830 | { | |
5831 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
5832 | static struct memblock_region *r; | |
5833 | ||
5834 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5835 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
5836 | for_each_memblock(memory, r) { | |
5837 | if (*pfn < memblock_region_memory_end_pfn(r)) | |
5838 | break; | |
5839 | } | |
5840 | } | |
5841 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
5842 | memblock_is_mirror(r)) { | |
5843 | *pfn = memblock_region_memory_end_pfn(r); | |
5844 | return true; | |
5845 | } | |
5846 | } | |
5847 | #endif | |
5848 | return false; | |
5849 | } | |
5850 | ||
3f135355 KS |
5851 | #ifdef CONFIG_SPARSEMEM |
5852 | /* Skip PFNs that belong to non-present sections */ | |
5853 | static inline __meminit unsigned long next_pfn(unsigned long pfn) | |
5854 | { | |
4c605881 | 5855 | const unsigned long section_nr = pfn_to_section_nr(++pfn); |
3f135355 | 5856 | |
3f135355 KS |
5857 | if (present_section_nr(section_nr)) |
5858 | return pfn; | |
4c605881 | 5859 | return section_nr_to_pfn(next_present_section_nr(section_nr)); |
3f135355 KS |
5860 | } |
5861 | #else | |
5862 | static inline __meminit unsigned long next_pfn(unsigned long pfn) | |
5863 | { | |
5864 | return pfn++; | |
5865 | } | |
5866 | #endif | |
5867 | ||
1da177e4 LT |
5868 | /* |
5869 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 5870 | * up by memblock_free_all() once the early boot process is |
1da177e4 LT |
5871 | * done. Non-atomic initialization, single-pass. |
5872 | */ | |
c09b4240 | 5873 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a99583e7 CH |
5874 | unsigned long start_pfn, enum memmap_context context, |
5875 | struct vmem_altmap *altmap) | |
1da177e4 | 5876 | { |
a9a9e77f | 5877 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 5878 | struct page *page; |
1da177e4 | 5879 | |
22b31eec HD |
5880 | if (highest_memmap_pfn < end_pfn - 1) |
5881 | highest_memmap_pfn = end_pfn - 1; | |
5882 | ||
966cf44f | 5883 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
5884 | /* |
5885 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
5886 | * memory. We limit the total number of pages to initialize to just |
5887 | * those that might contain the memory mapping. We will defer the | |
5888 | * ZONE_DEVICE page initialization until after we have released | |
5889 | * the hotplug lock. | |
4b94ffdc | 5890 | */ |
966cf44f AD |
5891 | if (zone == ZONE_DEVICE) { |
5892 | if (!altmap) | |
5893 | return; | |
5894 | ||
5895 | if (start_pfn == altmap->base_pfn) | |
5896 | start_pfn += altmap->reserve; | |
5897 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
5898 | } | |
5899 | #endif | |
4b94ffdc | 5900 | |
948c436e | 5901 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 5902 | /* |
b72d0ffb AM |
5903 | * There can be holes in boot-time mem_map[]s handed to this |
5904 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5905 | */ |
a9a9e77f | 5906 | if (context == MEMMAP_EARLY) { |
3f135355 | 5907 | if (!early_pfn_valid(pfn)) { |
948c436e | 5908 | pfn = next_pfn(pfn); |
b72d0ffb | 5909 | continue; |
3f135355 | 5910 | } |
948c436e DH |
5911 | if (!early_pfn_in_nid(pfn, nid)) { |
5912 | pfn++; | |
a9a9e77f | 5913 | continue; |
948c436e | 5914 | } |
a9a9e77f PT |
5915 | if (overlap_memmap_init(zone, &pfn)) |
5916 | continue; | |
5917 | if (defer_init(nid, pfn, end_pfn)) | |
5918 | break; | |
a2f3aa02 | 5919 | } |
ac5d2539 | 5920 | |
d0dc12e8 PT |
5921 | page = pfn_to_page(pfn); |
5922 | __init_single_page(page, pfn, zone, nid); | |
5923 | if (context == MEMMAP_HOTPLUG) | |
d483da5b | 5924 | __SetPageReserved(page); |
d0dc12e8 | 5925 | |
ac5d2539 MG |
5926 | /* |
5927 | * Mark the block movable so that blocks are reserved for | |
5928 | * movable at startup. This will force kernel allocations | |
5929 | * to reserve their blocks rather than leaking throughout | |
5930 | * the address space during boot when many long-lived | |
974a786e | 5931 | * kernel allocations are made. |
ac5d2539 MG |
5932 | * |
5933 | * bitmap is created for zone's valid pfn range. but memmap | |
5934 | * can be created for invalid pages (for alignment) | |
5935 | * check here not to call set_pageblock_migratetype() against | |
5936 | * pfn out of zone. | |
5937 | */ | |
5938 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
ac5d2539 | 5939 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
9b6e63cb | 5940 | cond_resched(); |
ac5d2539 | 5941 | } |
948c436e | 5942 | pfn++; |
1da177e4 LT |
5943 | } |
5944 | } | |
5945 | ||
966cf44f AD |
5946 | #ifdef CONFIG_ZONE_DEVICE |
5947 | void __ref memmap_init_zone_device(struct zone *zone, | |
5948 | unsigned long start_pfn, | |
5949 | unsigned long size, | |
5950 | struct dev_pagemap *pgmap) | |
5951 | { | |
5952 | unsigned long pfn, end_pfn = start_pfn + size; | |
5953 | struct pglist_data *pgdat = zone->zone_pgdat; | |
514caf23 | 5954 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
966cf44f AD |
5955 | unsigned long zone_idx = zone_idx(zone); |
5956 | unsigned long start = jiffies; | |
5957 | int nid = pgdat->node_id; | |
5958 | ||
46d945ae | 5959 | if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE)) |
966cf44f AD |
5960 | return; |
5961 | ||
5962 | /* | |
5963 | * The call to memmap_init_zone should have already taken care | |
5964 | * of the pages reserved for the memmap, so we can just jump to | |
5965 | * the end of that region and start processing the device pages. | |
5966 | */ | |
514caf23 | 5967 | if (altmap) { |
966cf44f AD |
5968 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
5969 | size = end_pfn - start_pfn; | |
5970 | } | |
5971 | ||
5972 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
5973 | struct page *page = pfn_to_page(pfn); | |
5974 | ||
5975 | __init_single_page(page, pfn, zone_idx, nid); | |
5976 | ||
5977 | /* | |
5978 | * Mark page reserved as it will need to wait for onlining | |
5979 | * phase for it to be fully associated with a zone. | |
5980 | * | |
5981 | * We can use the non-atomic __set_bit operation for setting | |
5982 | * the flag as we are still initializing the pages. | |
5983 | */ | |
5984 | __SetPageReserved(page); | |
5985 | ||
5986 | /* | |
8a164fef CH |
5987 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer |
5988 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
5989 | * ever freed or placed on a driver-private list. | |
966cf44f AD |
5990 | */ |
5991 | page->pgmap = pgmap; | |
8a164fef | 5992 | page->zone_device_data = NULL; |
966cf44f AD |
5993 | |
5994 | /* | |
5995 | * Mark the block movable so that blocks are reserved for | |
5996 | * movable at startup. This will force kernel allocations | |
5997 | * to reserve their blocks rather than leaking throughout | |
5998 | * the address space during boot when many long-lived | |
5999 | * kernel allocations are made. | |
6000 | * | |
6001 | * bitmap is created for zone's valid pfn range. but memmap | |
6002 | * can be created for invalid pages (for alignment) | |
6003 | * check here not to call set_pageblock_migratetype() against | |
6004 | * pfn out of zone. | |
6005 | * | |
6006 | * Please note that MEMMAP_HOTPLUG path doesn't clear memmap | |
ba72b4c8 | 6007 | * because this is done early in section_activate() |
966cf44f AD |
6008 | */ |
6009 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
6010 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
6011 | cond_resched(); | |
6012 | } | |
6013 | } | |
6014 | ||
fdc029b1 | 6015 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
966cf44f AD |
6016 | size, jiffies_to_msecs(jiffies - start)); |
6017 | } | |
6018 | ||
6019 | #endif | |
1e548deb | 6020 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6021 | { |
7aeb09f9 | 6022 | unsigned int order, t; |
b2a0ac88 MG |
6023 | for_each_migratetype_order(order, t) { |
6024 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6025 | zone->free_area[order].nr_free = 0; |
6026 | } | |
6027 | } | |
6028 | ||
dfb3ccd0 PT |
6029 | void __meminit __weak memmap_init(unsigned long size, int nid, |
6030 | unsigned long zone, unsigned long start_pfn) | |
6031 | { | |
6032 | memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY, NULL); | |
6033 | } | |
1da177e4 | 6034 | |
7cd2b0a3 | 6035 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 6036 | { |
3a6be87f | 6037 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
6038 | int batch; |
6039 | ||
6040 | /* | |
6041 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 6042 | * size of the zone. |
e7c8d5c9 | 6043 | */ |
9705bea5 | 6044 | batch = zone_managed_pages(zone) / 1024; |
d8a759b5 AL |
6045 | /* But no more than a meg. */ |
6046 | if (batch * PAGE_SIZE > 1024 * 1024) | |
6047 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
6048 | batch /= 4; /* We effectively *= 4 below */ |
6049 | if (batch < 1) | |
6050 | batch = 1; | |
6051 | ||
6052 | /* | |
0ceaacc9 NP |
6053 | * Clamp the batch to a 2^n - 1 value. Having a power |
6054 | * of 2 value was found to be more likely to have | |
6055 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 6056 | * |
0ceaacc9 NP |
6057 | * For example if 2 tasks are alternately allocating |
6058 | * batches of pages, one task can end up with a lot | |
6059 | * of pages of one half of the possible page colors | |
6060 | * and the other with pages of the other colors. | |
e7c8d5c9 | 6061 | */ |
9155203a | 6062 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 6063 | |
e7c8d5c9 | 6064 | return batch; |
3a6be87f DH |
6065 | |
6066 | #else | |
6067 | /* The deferral and batching of frees should be suppressed under NOMMU | |
6068 | * conditions. | |
6069 | * | |
6070 | * The problem is that NOMMU needs to be able to allocate large chunks | |
6071 | * of contiguous memory as there's no hardware page translation to | |
6072 | * assemble apparent contiguous memory from discontiguous pages. | |
6073 | * | |
6074 | * Queueing large contiguous runs of pages for batching, however, | |
6075 | * causes the pages to actually be freed in smaller chunks. As there | |
6076 | * can be a significant delay between the individual batches being | |
6077 | * recycled, this leads to the once large chunks of space being | |
6078 | * fragmented and becoming unavailable for high-order allocations. | |
6079 | */ | |
6080 | return 0; | |
6081 | #endif | |
e7c8d5c9 CL |
6082 | } |
6083 | ||
8d7a8fa9 CS |
6084 | /* |
6085 | * pcp->high and pcp->batch values are related and dependent on one another: | |
6086 | * ->batch must never be higher then ->high. | |
6087 | * The following function updates them in a safe manner without read side | |
6088 | * locking. | |
6089 | * | |
6090 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
6091 | * those fields changing asynchronously (acording the the above rule). | |
6092 | * | |
6093 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
6094 | * outside of boot time (or some other assurance that no concurrent updaters | |
6095 | * exist). | |
6096 | */ | |
6097 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
6098 | unsigned long batch) | |
6099 | { | |
6100 | /* start with a fail safe value for batch */ | |
6101 | pcp->batch = 1; | |
6102 | smp_wmb(); | |
6103 | ||
6104 | /* Update high, then batch, in order */ | |
6105 | pcp->high = high; | |
6106 | smp_wmb(); | |
6107 | ||
6108 | pcp->batch = batch; | |
6109 | } | |
6110 | ||
3664033c | 6111 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
6112 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
6113 | { | |
8d7a8fa9 | 6114 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
6115 | } |
6116 | ||
88c90dbc | 6117 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
6118 | { |
6119 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 6120 | int migratetype; |
2caaad41 | 6121 | |
1c6fe946 MD |
6122 | memset(p, 0, sizeof(*p)); |
6123 | ||
3dfa5721 | 6124 | pcp = &p->pcp; |
5f8dcc21 MG |
6125 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
6126 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
6127 | } |
6128 | ||
88c90dbc CS |
6129 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
6130 | { | |
6131 | pageset_init(p); | |
6132 | pageset_set_batch(p, batch); | |
6133 | } | |
6134 | ||
8ad4b1fb | 6135 | /* |
3664033c | 6136 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
6137 | * to the value high for the pageset p. |
6138 | */ | |
3664033c | 6139 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
6140 | unsigned long high) |
6141 | { | |
8d7a8fa9 CS |
6142 | unsigned long batch = max(1UL, high / 4); |
6143 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
6144 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 6145 | |
8d7a8fa9 | 6146 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
6147 | } |
6148 | ||
7cd2b0a3 DR |
6149 | static void pageset_set_high_and_batch(struct zone *zone, |
6150 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 6151 | { |
56cef2b8 | 6152 | if (percpu_pagelist_fraction) |
3664033c | 6153 | pageset_set_high(pcp, |
9705bea5 | 6154 | (zone_managed_pages(zone) / |
56cef2b8 CS |
6155 | percpu_pagelist_fraction)); |
6156 | else | |
6157 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
6158 | } | |
6159 | ||
169f6c19 CS |
6160 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
6161 | { | |
6162 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
6163 | ||
6164 | pageset_init(pcp); | |
6165 | pageset_set_high_and_batch(zone, pcp); | |
6166 | } | |
6167 | ||
72675e13 | 6168 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
6169 | { |
6170 | int cpu; | |
319774e2 | 6171 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
6172 | for_each_possible_cpu(cpu) |
6173 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
6174 | } |
6175 | ||
2caaad41 | 6176 | /* |
99dcc3e5 CL |
6177 | * Allocate per cpu pagesets and initialize them. |
6178 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 6179 | */ |
99dcc3e5 | 6180 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 6181 | { |
b4911ea2 | 6182 | struct pglist_data *pgdat; |
99dcc3e5 | 6183 | struct zone *zone; |
e7c8d5c9 | 6184 | |
319774e2 WF |
6185 | for_each_populated_zone(zone) |
6186 | setup_zone_pageset(zone); | |
b4911ea2 MG |
6187 | |
6188 | for_each_online_pgdat(pgdat) | |
6189 | pgdat->per_cpu_nodestats = | |
6190 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
6191 | } |
6192 | ||
c09b4240 | 6193 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 6194 | { |
99dcc3e5 CL |
6195 | /* |
6196 | * per cpu subsystem is not up at this point. The following code | |
6197 | * relies on the ability of the linker to provide the | |
6198 | * offset of a (static) per cpu variable into the per cpu area. | |
6199 | */ | |
6200 | zone->pageset = &boot_pageset; | |
ed8ece2e | 6201 | |
b38a8725 | 6202 | if (populated_zone(zone)) |
99dcc3e5 CL |
6203 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
6204 | zone->name, zone->present_pages, | |
6205 | zone_batchsize(zone)); | |
ed8ece2e DH |
6206 | } |
6207 | ||
dc0bbf3b | 6208 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 6209 | unsigned long zone_start_pfn, |
b171e409 | 6210 | unsigned long size) |
ed8ece2e DH |
6211 | { |
6212 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 6213 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 6214 | |
8f416836 WY |
6215 | if (zone_idx > pgdat->nr_zones) |
6216 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 6217 | |
ed8ece2e DH |
6218 | zone->zone_start_pfn = zone_start_pfn; |
6219 | ||
708614e6 MG |
6220 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
6221 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
6222 | pgdat->node_id, | |
6223 | (unsigned long)zone_idx(zone), | |
6224 | zone_start_pfn, (zone_start_pfn + size)); | |
6225 | ||
1e548deb | 6226 | zone_init_free_lists(zone); |
9dcb8b68 | 6227 | zone->initialized = 1; |
ed8ece2e DH |
6228 | } |
6229 | ||
0ee332c1 | 6230 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 6231 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 6232 | |
c713216d MG |
6233 | /* |
6234 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 6235 | */ |
8a942fde MG |
6236 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
6237 | struct mminit_pfnnid_cache *state) | |
c713216d | 6238 | { |
c13291a5 | 6239 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 6240 | int nid; |
7c243c71 | 6241 | |
8a942fde MG |
6242 | if (state->last_start <= pfn && pfn < state->last_end) |
6243 | return state->last_nid; | |
c713216d | 6244 | |
e76b63f8 | 6245 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
98fa15f3 | 6246 | if (nid != NUMA_NO_NODE) { |
8a942fde MG |
6247 | state->last_start = start_pfn; |
6248 | state->last_end = end_pfn; | |
6249 | state->last_nid = nid; | |
e76b63f8 YL |
6250 | } |
6251 | ||
6252 | return nid; | |
c713216d MG |
6253 | } |
6254 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
6255 | ||
c713216d | 6256 | /** |
6782832e | 6257 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 6258 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 6259 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 6260 | * |
7d018176 ZZ |
6261 | * If an architecture guarantees that all ranges registered contain no holes |
6262 | * and may be freed, this this function may be used instead of calling | |
6263 | * memblock_free_early_nid() manually. | |
c713216d | 6264 | */ |
c13291a5 | 6265 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 6266 | { |
c13291a5 TH |
6267 | unsigned long start_pfn, end_pfn; |
6268 | int i, this_nid; | |
edbe7d23 | 6269 | |
c13291a5 TH |
6270 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
6271 | start_pfn = min(start_pfn, max_low_pfn); | |
6272 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 6273 | |
c13291a5 | 6274 | if (start_pfn < end_pfn) |
6782832e SS |
6275 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
6276 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
6277 | this_nid); | |
edbe7d23 | 6278 | } |
edbe7d23 | 6279 | } |
edbe7d23 | 6280 | |
c713216d MG |
6281 | /** |
6282 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 6283 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 6284 | * |
7d018176 ZZ |
6285 | * If an architecture guarantees that all ranges registered contain no holes and may |
6286 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
6287 | */ |
6288 | void __init sparse_memory_present_with_active_regions(int nid) | |
6289 | { | |
c13291a5 TH |
6290 | unsigned long start_pfn, end_pfn; |
6291 | int i, this_nid; | |
c713216d | 6292 | |
c13291a5 TH |
6293 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
6294 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
6295 | } |
6296 | ||
6297 | /** | |
6298 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
6299 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
6300 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
6301 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
6302 | * |
6303 | * It returns the start and end page frame of a node based on information | |
7d018176 | 6304 | * provided by memblock_set_node(). If called for a node |
c713216d | 6305 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 6306 | * PFNs will be 0. |
c713216d | 6307 | */ |
bbe5d993 | 6308 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
6309 | unsigned long *start_pfn, unsigned long *end_pfn) |
6310 | { | |
c13291a5 | 6311 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 6312 | int i; |
c13291a5 | 6313 | |
c713216d MG |
6314 | *start_pfn = -1UL; |
6315 | *end_pfn = 0; | |
6316 | ||
c13291a5 TH |
6317 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
6318 | *start_pfn = min(*start_pfn, this_start_pfn); | |
6319 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
6320 | } |
6321 | ||
633c0666 | 6322 | if (*start_pfn == -1UL) |
c713216d | 6323 | *start_pfn = 0; |
c713216d MG |
6324 | } |
6325 | ||
2a1e274a MG |
6326 | /* |
6327 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
6328 | * assumption is made that zones within a node are ordered in monotonic | |
6329 | * increasing memory addresses so that the "highest" populated zone is used | |
6330 | */ | |
b69a7288 | 6331 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
6332 | { |
6333 | int zone_index; | |
6334 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
6335 | if (zone_index == ZONE_MOVABLE) | |
6336 | continue; | |
6337 | ||
6338 | if (arch_zone_highest_possible_pfn[zone_index] > | |
6339 | arch_zone_lowest_possible_pfn[zone_index]) | |
6340 | break; | |
6341 | } | |
6342 | ||
6343 | VM_BUG_ON(zone_index == -1); | |
6344 | movable_zone = zone_index; | |
6345 | } | |
6346 | ||
6347 | /* | |
6348 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 6349 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
6350 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
6351 | * in each node depending on the size of each node and how evenly kernelcore | |
6352 | * is distributed. This helper function adjusts the zone ranges | |
6353 | * provided by the architecture for a given node by using the end of the | |
6354 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
6355 | * zones within a node are in order of monotonic increases memory addresses | |
6356 | */ | |
bbe5d993 | 6357 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
6358 | unsigned long zone_type, |
6359 | unsigned long node_start_pfn, | |
6360 | unsigned long node_end_pfn, | |
6361 | unsigned long *zone_start_pfn, | |
6362 | unsigned long *zone_end_pfn) | |
6363 | { | |
6364 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
6365 | if (zone_movable_pfn[nid]) { | |
6366 | /* Size ZONE_MOVABLE */ | |
6367 | if (zone_type == ZONE_MOVABLE) { | |
6368 | *zone_start_pfn = zone_movable_pfn[nid]; | |
6369 | *zone_end_pfn = min(node_end_pfn, | |
6370 | arch_zone_highest_possible_pfn[movable_zone]); | |
6371 | ||
e506b996 XQ |
6372 | /* Adjust for ZONE_MOVABLE starting within this range */ |
6373 | } else if (!mirrored_kernelcore && | |
6374 | *zone_start_pfn < zone_movable_pfn[nid] && | |
6375 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
6376 | *zone_end_pfn = zone_movable_pfn[nid]; | |
6377 | ||
2a1e274a MG |
6378 | /* Check if this whole range is within ZONE_MOVABLE */ |
6379 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
6380 | *zone_start_pfn = *zone_end_pfn; | |
6381 | } | |
6382 | } | |
6383 | ||
c713216d MG |
6384 | /* |
6385 | * Return the number of pages a zone spans in a node, including holes | |
6386 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
6387 | */ | |
bbe5d993 | 6388 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 6389 | unsigned long zone_type, |
7960aedd ZY |
6390 | unsigned long node_start_pfn, |
6391 | unsigned long node_end_pfn, | |
d91749c1 TI |
6392 | unsigned long *zone_start_pfn, |
6393 | unsigned long *zone_end_pfn, | |
c713216d MG |
6394 | unsigned long *ignored) |
6395 | { | |
299c83dc LF |
6396 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6397 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 6398 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6399 | if (!node_start_pfn && !node_end_pfn) |
6400 | return 0; | |
6401 | ||
7960aedd | 6402 | /* Get the start and end of the zone */ |
299c83dc LF |
6403 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6404 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
6405 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6406 | node_start_pfn, node_end_pfn, | |
d91749c1 | 6407 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
6408 | |
6409 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 6410 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
6411 | return 0; |
6412 | ||
6413 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
6414 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
6415 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
6416 | |
6417 | /* Return the spanned pages */ | |
d91749c1 | 6418 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
6419 | } |
6420 | ||
6421 | /* | |
6422 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 6423 | * then all holes in the requested range will be accounted for. |
c713216d | 6424 | */ |
bbe5d993 | 6425 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
6426 | unsigned long range_start_pfn, |
6427 | unsigned long range_end_pfn) | |
6428 | { | |
96e907d1 TH |
6429 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
6430 | unsigned long start_pfn, end_pfn; | |
6431 | int i; | |
c713216d | 6432 | |
96e907d1 TH |
6433 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6434 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6435 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6436 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 6437 | } |
96e907d1 | 6438 | return nr_absent; |
c713216d MG |
6439 | } |
6440 | ||
6441 | /** | |
6442 | * absent_pages_in_range - Return number of page frames in holes within a range | |
6443 | * @start_pfn: The start PFN to start searching for holes | |
6444 | * @end_pfn: The end PFN to stop searching for holes | |
6445 | * | |
a862f68a | 6446 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
6447 | */ |
6448 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
6449 | unsigned long end_pfn) | |
6450 | { | |
6451 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
6452 | } | |
6453 | ||
6454 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 6455 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 6456 | unsigned long zone_type, |
7960aedd ZY |
6457 | unsigned long node_start_pfn, |
6458 | unsigned long node_end_pfn, | |
c713216d MG |
6459 | unsigned long *ignored) |
6460 | { | |
96e907d1 TH |
6461 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6462 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 6463 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 6464 | unsigned long nr_absent; |
9c7cd687 | 6465 | |
b5685e92 | 6466 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6467 | if (!node_start_pfn && !node_end_pfn) |
6468 | return 0; | |
6469 | ||
96e907d1 TH |
6470 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6471 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 6472 | |
2a1e274a MG |
6473 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6474 | node_start_pfn, node_end_pfn, | |
6475 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
6476 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
6477 | ||
6478 | /* | |
6479 | * ZONE_MOVABLE handling. | |
6480 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6481 | * and vice versa. | |
6482 | */ | |
e506b996 XQ |
6483 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6484 | unsigned long start_pfn, end_pfn; | |
6485 | struct memblock_region *r; | |
6486 | ||
6487 | for_each_memblock(memory, r) { | |
6488 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
6489 | zone_start_pfn, zone_end_pfn); | |
6490 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6491 | zone_start_pfn, zone_end_pfn); | |
6492 | ||
6493 | if (zone_type == ZONE_MOVABLE && | |
6494 | memblock_is_mirror(r)) | |
6495 | nr_absent += end_pfn - start_pfn; | |
6496 | ||
6497 | if (zone_type == ZONE_NORMAL && | |
6498 | !memblock_is_mirror(r)) | |
6499 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6500 | } |
6501 | } | |
6502 | ||
6503 | return nr_absent; | |
c713216d | 6504 | } |
0e0b864e | 6505 | |
0ee332c1 | 6506 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
bbe5d993 | 6507 | static inline unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 6508 | unsigned long zone_type, |
7960aedd ZY |
6509 | unsigned long node_start_pfn, |
6510 | unsigned long node_end_pfn, | |
d91749c1 TI |
6511 | unsigned long *zone_start_pfn, |
6512 | unsigned long *zone_end_pfn, | |
c713216d MG |
6513 | unsigned long *zones_size) |
6514 | { | |
d91749c1 TI |
6515 | unsigned int zone; |
6516 | ||
6517 | *zone_start_pfn = node_start_pfn; | |
6518 | for (zone = 0; zone < zone_type; zone++) | |
6519 | *zone_start_pfn += zones_size[zone]; | |
6520 | ||
6521 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
6522 | ||
c713216d MG |
6523 | return zones_size[zone_type]; |
6524 | } | |
6525 | ||
bbe5d993 | 6526 | static inline unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 6527 | unsigned long zone_type, |
7960aedd ZY |
6528 | unsigned long node_start_pfn, |
6529 | unsigned long node_end_pfn, | |
c713216d MG |
6530 | unsigned long *zholes_size) |
6531 | { | |
6532 | if (!zholes_size) | |
6533 | return 0; | |
6534 | ||
6535 | return zholes_size[zone_type]; | |
6536 | } | |
20e6926d | 6537 | |
0ee332c1 | 6538 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6539 | |
bbe5d993 | 6540 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
6541 | unsigned long node_start_pfn, |
6542 | unsigned long node_end_pfn, | |
6543 | unsigned long *zones_size, | |
6544 | unsigned long *zholes_size) | |
c713216d | 6545 | { |
febd5949 | 6546 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6547 | enum zone_type i; |
6548 | ||
febd5949 GZ |
6549 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6550 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6551 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 6552 | unsigned long size, real_size; |
c713216d | 6553 | |
febd5949 GZ |
6554 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
6555 | node_start_pfn, | |
6556 | node_end_pfn, | |
d91749c1 TI |
6557 | &zone_start_pfn, |
6558 | &zone_end_pfn, | |
febd5949 GZ |
6559 | zones_size); |
6560 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
6561 | node_start_pfn, node_end_pfn, |
6562 | zholes_size); | |
d91749c1 TI |
6563 | if (size) |
6564 | zone->zone_start_pfn = zone_start_pfn; | |
6565 | else | |
6566 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
6567 | zone->spanned_pages = size; |
6568 | zone->present_pages = real_size; | |
6569 | ||
6570 | totalpages += size; | |
6571 | realtotalpages += real_size; | |
6572 | } | |
6573 | ||
6574 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
6575 | pgdat->node_present_pages = realtotalpages; |
6576 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
6577 | realtotalpages); | |
6578 | } | |
6579 | ||
835c134e MG |
6580 | #ifndef CONFIG_SPARSEMEM |
6581 | /* | |
6582 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
6583 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
6584 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
6585 | * round what is now in bits to nearest long in bits, then return it in |
6586 | * bytes. | |
6587 | */ | |
7c45512d | 6588 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
6589 | { |
6590 | unsigned long usemapsize; | |
6591 | ||
7c45512d | 6592 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
6593 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
6594 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
6595 | usemapsize *= NR_PAGEBLOCK_BITS; |
6596 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
6597 | ||
6598 | return usemapsize / 8; | |
6599 | } | |
6600 | ||
7cc2a959 | 6601 | static void __ref setup_usemap(struct pglist_data *pgdat, |
7c45512d LT |
6602 | struct zone *zone, |
6603 | unsigned long zone_start_pfn, | |
6604 | unsigned long zonesize) | |
835c134e | 6605 | { |
7c45512d | 6606 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 6607 | zone->pageblock_flags = NULL; |
23a7052a | 6608 | if (usemapsize) { |
6782832e | 6609 | zone->pageblock_flags = |
26fb3dae MR |
6610 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
6611 | pgdat->node_id); | |
23a7052a MR |
6612 | if (!zone->pageblock_flags) |
6613 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
6614 | usemapsize, zone->name, pgdat->node_id); | |
6615 | } | |
835c134e MG |
6616 | } |
6617 | #else | |
7c45512d LT |
6618 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
6619 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
6620 | #endif /* CONFIG_SPARSEMEM */ |
6621 | ||
d9c23400 | 6622 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 6623 | |
d9c23400 | 6624 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 6625 | void __init set_pageblock_order(void) |
d9c23400 | 6626 | { |
955c1cd7 AM |
6627 | unsigned int order; |
6628 | ||
d9c23400 MG |
6629 | /* Check that pageblock_nr_pages has not already been setup */ |
6630 | if (pageblock_order) | |
6631 | return; | |
6632 | ||
955c1cd7 AM |
6633 | if (HPAGE_SHIFT > PAGE_SHIFT) |
6634 | order = HUGETLB_PAGE_ORDER; | |
6635 | else | |
6636 | order = MAX_ORDER - 1; | |
6637 | ||
d9c23400 MG |
6638 | /* |
6639 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
6640 | * This value may be variable depending on boot parameters on IA64 and |
6641 | * powerpc. | |
d9c23400 MG |
6642 | */ |
6643 | pageblock_order = order; | |
6644 | } | |
6645 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6646 | ||
ba72cb8c MG |
6647 | /* |
6648 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
6649 | * is unused as pageblock_order is set at compile-time. See |
6650 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
6651 | * the kernel config | |
ba72cb8c | 6652 | */ |
03e85f9d | 6653 | void __init set_pageblock_order(void) |
ba72cb8c | 6654 | { |
ba72cb8c | 6655 | } |
d9c23400 MG |
6656 | |
6657 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6658 | ||
03e85f9d | 6659 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 6660 | unsigned long present_pages) |
01cefaef JL |
6661 | { |
6662 | unsigned long pages = spanned_pages; | |
6663 | ||
6664 | /* | |
6665 | * Provide a more accurate estimation if there are holes within | |
6666 | * the zone and SPARSEMEM is in use. If there are holes within the | |
6667 | * zone, each populated memory region may cost us one or two extra | |
6668 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 6669 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
6670 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
6671 | */ | |
6672 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
6673 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
6674 | pages = present_pages; | |
6675 | ||
6676 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
6677 | } | |
6678 | ||
ace1db39 OS |
6679 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6680 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
6681 | { | |
364c1eeb YS |
6682 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
6683 | ||
6684 | spin_lock_init(&ds_queue->split_queue_lock); | |
6685 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
6686 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
6687 | } |
6688 | #else | |
6689 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
6690 | #endif | |
6691 | ||
6692 | #ifdef CONFIG_COMPACTION | |
6693 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
6694 | { | |
6695 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6696 | } | |
6697 | #else | |
6698 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
6699 | #endif | |
6700 | ||
03e85f9d | 6701 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 6702 | { |
208d54e5 | 6703 | pgdat_resize_init(pgdat); |
ace1db39 | 6704 | |
ace1db39 OS |
6705 | pgdat_init_split_queue(pgdat); |
6706 | pgdat_init_kcompactd(pgdat); | |
6707 | ||
1da177e4 | 6708 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6709 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 6710 | |
eefa864b | 6711 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6712 | spin_lock_init(&pgdat->lru_lock); |
867e5e1d | 6713 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
6714 | } |
6715 | ||
6716 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
6717 | unsigned long remaining_pages) | |
6718 | { | |
9705bea5 | 6719 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
6720 | zone_set_nid(zone, nid); |
6721 | zone->name = zone_names[idx]; | |
6722 | zone->zone_pgdat = NODE_DATA(nid); | |
6723 | spin_lock_init(&zone->lock); | |
6724 | zone_seqlock_init(zone); | |
6725 | zone_pcp_init(zone); | |
6726 | } | |
6727 | ||
6728 | /* | |
6729 | * Set up the zone data structures | |
6730 | * - init pgdat internals | |
6731 | * - init all zones belonging to this node | |
6732 | * | |
6733 | * NOTE: this function is only called during memory hotplug | |
6734 | */ | |
6735 | #ifdef CONFIG_MEMORY_HOTPLUG | |
6736 | void __ref free_area_init_core_hotplug(int nid) | |
6737 | { | |
6738 | enum zone_type z; | |
6739 | pg_data_t *pgdat = NODE_DATA(nid); | |
6740 | ||
6741 | pgdat_init_internals(pgdat); | |
6742 | for (z = 0; z < MAX_NR_ZONES; z++) | |
6743 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
6744 | } | |
6745 | #endif | |
6746 | ||
6747 | /* | |
6748 | * Set up the zone data structures: | |
6749 | * - mark all pages reserved | |
6750 | * - mark all memory queues empty | |
6751 | * - clear the memory bitmaps | |
6752 | * | |
6753 | * NOTE: pgdat should get zeroed by caller. | |
6754 | * NOTE: this function is only called during early init. | |
6755 | */ | |
6756 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
6757 | { | |
6758 | enum zone_type j; | |
6759 | int nid = pgdat->node_id; | |
5f63b720 | 6760 | |
03e85f9d | 6761 | pgdat_init_internals(pgdat); |
385386cf JW |
6762 | pgdat->per_cpu_nodestats = &boot_nodestats; |
6763 | ||
1da177e4 LT |
6764 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6765 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 6766 | unsigned long size, freesize, memmap_pages; |
d91749c1 | 6767 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6768 | |
febd5949 | 6769 | size = zone->spanned_pages; |
e6943859 | 6770 | freesize = zone->present_pages; |
1da177e4 | 6771 | |
0e0b864e | 6772 | /* |
9feedc9d | 6773 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6774 | * is used by this zone for memmap. This affects the watermark |
6775 | * and per-cpu initialisations | |
6776 | */ | |
e6943859 | 6777 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
6778 | if (!is_highmem_idx(j)) { |
6779 | if (freesize >= memmap_pages) { | |
6780 | freesize -= memmap_pages; | |
6781 | if (memmap_pages) | |
6782 | printk(KERN_DEBUG | |
6783 | " %s zone: %lu pages used for memmap\n", | |
6784 | zone_names[j], memmap_pages); | |
6785 | } else | |
1170532b | 6786 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6787 | zone_names[j], memmap_pages, freesize); |
6788 | } | |
0e0b864e | 6789 | |
6267276f | 6790 | /* Account for reserved pages */ |
9feedc9d JL |
6791 | if (j == 0 && freesize > dma_reserve) { |
6792 | freesize -= dma_reserve; | |
d903ef9f | 6793 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6794 | zone_names[0], dma_reserve); |
0e0b864e MG |
6795 | } |
6796 | ||
98d2b0eb | 6797 | if (!is_highmem_idx(j)) |
9feedc9d | 6798 | nr_kernel_pages += freesize; |
01cefaef JL |
6799 | /* Charge for highmem memmap if there are enough kernel pages */ |
6800 | else if (nr_kernel_pages > memmap_pages * 2) | |
6801 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6802 | nr_all_pages += freesize; |
1da177e4 | 6803 | |
9feedc9d JL |
6804 | /* |
6805 | * Set an approximate value for lowmem here, it will be adjusted | |
6806 | * when the bootmem allocator frees pages into the buddy system. | |
6807 | * And all highmem pages will be managed by the buddy system. | |
6808 | */ | |
03e85f9d | 6809 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 6810 | |
d883c6cf | 6811 | if (!size) |
1da177e4 LT |
6812 | continue; |
6813 | ||
955c1cd7 | 6814 | set_pageblock_order(); |
d883c6cf JK |
6815 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
6816 | init_currently_empty_zone(zone, zone_start_pfn, size); | |
76cdd58e | 6817 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6818 | } |
6819 | } | |
6820 | ||
0cd842f9 | 6821 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 6822 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6823 | { |
b0aeba74 | 6824 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6825 | unsigned long __maybe_unused offset = 0; |
6826 | ||
1da177e4 LT |
6827 | /* Skip empty nodes */ |
6828 | if (!pgdat->node_spanned_pages) | |
6829 | return; | |
6830 | ||
b0aeba74 TL |
6831 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6832 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6833 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6834 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6835 | unsigned long size, end; |
d41dee36 AW |
6836 | struct page *map; |
6837 | ||
e984bb43 BP |
6838 | /* |
6839 | * The zone's endpoints aren't required to be MAX_ORDER | |
6840 | * aligned but the node_mem_map endpoints must be in order | |
6841 | * for the buddy allocator to function correctly. | |
6842 | */ | |
108bcc96 | 6843 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6844 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6845 | size = (end - start) * sizeof(struct page); | |
26fb3dae MR |
6846 | map = memblock_alloc_node(size, SMP_CACHE_BYTES, |
6847 | pgdat->node_id); | |
23a7052a MR |
6848 | if (!map) |
6849 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
6850 | size, pgdat->node_id); | |
a1c34a3b | 6851 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6852 | } |
0cd842f9 OS |
6853 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
6854 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
6855 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 6856 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6857 | /* |
6858 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6859 | */ | |
c713216d | 6860 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6861 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 6862 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 6863 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6864 | mem_map -= offset; |
0ee332c1 | 6865 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6866 | } |
1da177e4 LT |
6867 | #endif |
6868 | } | |
0cd842f9 OS |
6869 | #else |
6870 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
6871 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 6872 | |
0188dc98 OS |
6873 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
6874 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
6875 | { | |
0188dc98 OS |
6876 | pgdat->first_deferred_pfn = ULONG_MAX; |
6877 | } | |
6878 | #else | |
6879 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
6880 | #endif | |
6881 | ||
03e85f9d | 6882 | void __init free_area_init_node(int nid, unsigned long *zones_size, |
7cc2a959 PT |
6883 | unsigned long node_start_pfn, |
6884 | unsigned long *zholes_size) | |
1da177e4 | 6885 | { |
9109fb7b | 6886 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6887 | unsigned long start_pfn = 0; |
6888 | unsigned long end_pfn = 0; | |
9109fb7b | 6889 | |
88fdf75d | 6890 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 6891 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 6892 | |
1da177e4 LT |
6893 | pgdat->node_id = nid; |
6894 | pgdat->node_start_pfn = node_start_pfn; | |
75ef7184 | 6895 | pgdat->per_cpu_nodestats = NULL; |
7960aedd ZY |
6896 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6897 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 6898 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6899 | (u64)start_pfn << PAGE_SHIFT, |
6900 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
6901 | #else |
6902 | start_pfn = node_start_pfn; | |
7960aedd ZY |
6903 | #endif |
6904 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
6905 | zones_size, zholes_size); | |
1da177e4 LT |
6906 | |
6907 | alloc_node_mem_map(pgdat); | |
0188dc98 | 6908 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 6909 | |
7f3eb55b | 6910 | free_area_init_core(pgdat); |
1da177e4 LT |
6911 | } |
6912 | ||
aca52c39 | 6913 | #if !defined(CONFIG_FLAT_NODE_MEM_MAP) |
ec393a0f | 6914 | /* |
4b094b78 DH |
6915 | * Initialize all valid struct pages in the range [spfn, epfn) and mark them |
6916 | * PageReserved(). Return the number of struct pages that were initialized. | |
ec393a0f | 6917 | */ |
4b094b78 | 6918 | static u64 __init init_unavailable_range(unsigned long spfn, unsigned long epfn) |
ec393a0f PT |
6919 | { |
6920 | unsigned long pfn; | |
6921 | u64 pgcnt = 0; | |
6922 | ||
6923 | for (pfn = spfn; pfn < epfn; pfn++) { | |
6924 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { | |
6925 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6926 | + pageblock_nr_pages - 1; | |
6927 | continue; | |
6928 | } | |
4b094b78 DH |
6929 | /* |
6930 | * Use a fake node/zone (0) for now. Some of these pages | |
6931 | * (in memblock.reserved but not in memblock.memory) will | |
6932 | * get re-initialized via reserve_bootmem_region() later. | |
6933 | */ | |
6934 | __init_single_page(pfn_to_page(pfn), pfn, 0, 0); | |
6935 | __SetPageReserved(pfn_to_page(pfn)); | |
ec393a0f PT |
6936 | pgcnt++; |
6937 | } | |
6938 | ||
6939 | return pgcnt; | |
6940 | } | |
6941 | ||
a4a3ede2 PT |
6942 | /* |
6943 | * Only struct pages that are backed by physical memory are zeroed and | |
6944 | * initialized by going through __init_single_page(). But, there are some | |
6945 | * struct pages which are reserved in memblock allocator and their fields | |
6946 | * may be accessed (for example page_to_pfn() on some configuration accesses | |
4b094b78 | 6947 | * flags). We must explicitly initialize those struct pages. |
907ec5fc NH |
6948 | * |
6949 | * This function also addresses a similar issue where struct pages are left | |
6950 | * uninitialized because the physical address range is not covered by | |
6951 | * memblock.memory or memblock.reserved. That could happen when memblock | |
e822969c DH |
6952 | * layout is manually configured via memmap=, or when the highest physical |
6953 | * address (max_pfn) does not end on a section boundary. | |
a4a3ede2 | 6954 | */ |
4b094b78 | 6955 | static void __init init_unavailable_mem(void) |
a4a3ede2 PT |
6956 | { |
6957 | phys_addr_t start, end; | |
a4a3ede2 | 6958 | u64 i, pgcnt; |
907ec5fc | 6959 | phys_addr_t next = 0; |
a4a3ede2 PT |
6960 | |
6961 | /* | |
907ec5fc | 6962 | * Loop through unavailable ranges not covered by memblock.memory. |
a4a3ede2 PT |
6963 | */ |
6964 | pgcnt = 0; | |
907ec5fc NH |
6965 | for_each_mem_range(i, &memblock.memory, NULL, |
6966 | NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) { | |
ec393a0f | 6967 | if (next < start) |
4b094b78 DH |
6968 | pgcnt += init_unavailable_range(PFN_DOWN(next), |
6969 | PFN_UP(start)); | |
907ec5fc NH |
6970 | next = end; |
6971 | } | |
e822969c DH |
6972 | |
6973 | /* | |
6974 | * Early sections always have a fully populated memmap for the whole | |
6975 | * section - see pfn_valid(). If the last section has holes at the | |
6976 | * end and that section is marked "online", the memmap will be | |
6977 | * considered initialized. Make sure that memmap has a well defined | |
6978 | * state. | |
6979 | */ | |
4b094b78 DH |
6980 | pgcnt += init_unavailable_range(PFN_DOWN(next), |
6981 | round_up(max_pfn, PAGES_PER_SECTION)); | |
907ec5fc | 6982 | |
a4a3ede2 PT |
6983 | /* |
6984 | * Struct pages that do not have backing memory. This could be because | |
6985 | * firmware is using some of this memory, or for some other reasons. | |
a4a3ede2 PT |
6986 | */ |
6987 | if (pgcnt) | |
907ec5fc | 6988 | pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt); |
a4a3ede2 | 6989 | } |
4b094b78 DH |
6990 | #else |
6991 | static inline void __init init_unavailable_mem(void) | |
6992 | { | |
6993 | } | |
aca52c39 | 6994 | #endif /* !CONFIG_FLAT_NODE_MEM_MAP */ |
a4a3ede2 | 6995 | |
0ee332c1 | 6996 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
6997 | |
6998 | #if MAX_NUMNODES > 1 | |
6999 | /* | |
7000 | * Figure out the number of possible node ids. | |
7001 | */ | |
f9872caf | 7002 | void __init setup_nr_node_ids(void) |
418508c1 | 7003 | { |
904a9553 | 7004 | unsigned int highest; |
418508c1 | 7005 | |
904a9553 | 7006 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7007 | nr_node_ids = highest + 1; |
7008 | } | |
418508c1 MS |
7009 | #endif |
7010 | ||
1e01979c TH |
7011 | /** |
7012 | * node_map_pfn_alignment - determine the maximum internode alignment | |
7013 | * | |
7014 | * This function should be called after node map is populated and sorted. | |
7015 | * It calculates the maximum power of two alignment which can distinguish | |
7016 | * all the nodes. | |
7017 | * | |
7018 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
7019 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
7020 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
7021 | * shifted, 1GiB is enough and this function will indicate so. | |
7022 | * | |
7023 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
7024 | * model has fine enough granularity to avoid incorrect mapping for the | |
7025 | * populated node map. | |
7026 | * | |
a862f68a | 7027 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
7028 | * requirement (single node). |
7029 | */ | |
7030 | unsigned long __init node_map_pfn_alignment(void) | |
7031 | { | |
7032 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 7033 | unsigned long start, end, mask; |
98fa15f3 | 7034 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 7035 | int i, nid; |
1e01979c | 7036 | |
c13291a5 | 7037 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
7038 | if (!start || last_nid < 0 || last_nid == nid) { |
7039 | last_nid = nid; | |
7040 | last_end = end; | |
7041 | continue; | |
7042 | } | |
7043 | ||
7044 | /* | |
7045 | * Start with a mask granular enough to pin-point to the | |
7046 | * start pfn and tick off bits one-by-one until it becomes | |
7047 | * too coarse to separate the current node from the last. | |
7048 | */ | |
7049 | mask = ~((1 << __ffs(start)) - 1); | |
7050 | while (mask && last_end <= (start & (mask << 1))) | |
7051 | mask <<= 1; | |
7052 | ||
7053 | /* accumulate all internode masks */ | |
7054 | accl_mask |= mask; | |
7055 | } | |
7056 | ||
7057 | /* convert mask to number of pages */ | |
7058 | return ~accl_mask + 1; | |
7059 | } | |
7060 | ||
a6af2bc3 | 7061 | /* Find the lowest pfn for a node */ |
b69a7288 | 7062 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 7063 | { |
a6af2bc3 | 7064 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
7065 | unsigned long start_pfn; |
7066 | int i; | |
1abbfb41 | 7067 | |
c13291a5 TH |
7068 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
7069 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 7070 | |
a6af2bc3 | 7071 | if (min_pfn == ULONG_MAX) { |
1170532b | 7072 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
7073 | return 0; |
7074 | } | |
7075 | ||
7076 | return min_pfn; | |
c713216d MG |
7077 | } |
7078 | ||
7079 | /** | |
7080 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
7081 | * | |
a862f68a | 7082 | * Return: the minimum PFN based on information provided via |
7d018176 | 7083 | * memblock_set_node(). |
c713216d MG |
7084 | */ |
7085 | unsigned long __init find_min_pfn_with_active_regions(void) | |
7086 | { | |
7087 | return find_min_pfn_for_node(MAX_NUMNODES); | |
7088 | } | |
7089 | ||
37b07e41 LS |
7090 | /* |
7091 | * early_calculate_totalpages() | |
7092 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 7093 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 7094 | */ |
484f51f8 | 7095 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 7096 | { |
7e63efef | 7097 | unsigned long totalpages = 0; |
c13291a5 TH |
7098 | unsigned long start_pfn, end_pfn; |
7099 | int i, nid; | |
7100 | ||
7101 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
7102 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 7103 | |
37b07e41 LS |
7104 | totalpages += pages; |
7105 | if (pages) | |
4b0ef1fe | 7106 | node_set_state(nid, N_MEMORY); |
37b07e41 | 7107 | } |
b8af2941 | 7108 | return totalpages; |
7e63efef MG |
7109 | } |
7110 | ||
2a1e274a MG |
7111 | /* |
7112 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
7113 | * is spread evenly between nodes as long as the nodes have enough | |
7114 | * memory. When they don't, some nodes will have more kernelcore than | |
7115 | * others | |
7116 | */ | |
b224ef85 | 7117 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
7118 | { |
7119 | int i, nid; | |
7120 | unsigned long usable_startpfn; | |
7121 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 7122 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 7123 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 7124 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 7125 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 7126 | struct memblock_region *r; |
b2f3eebe TC |
7127 | |
7128 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
7129 | find_usable_zone_for_movable(); | |
7130 | ||
7131 | /* | |
7132 | * If movable_node is specified, ignore kernelcore and movablecore | |
7133 | * options. | |
7134 | */ | |
7135 | if (movable_node_is_enabled()) { | |
136199f0 EM |
7136 | for_each_memblock(memory, r) { |
7137 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
7138 | continue; |
7139 | ||
136199f0 | 7140 | nid = r->nid; |
b2f3eebe | 7141 | |
136199f0 | 7142 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
7143 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
7144 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7145 | usable_startpfn; | |
7146 | } | |
7147 | ||
7148 | goto out2; | |
7149 | } | |
2a1e274a | 7150 | |
342332e6 TI |
7151 | /* |
7152 | * If kernelcore=mirror is specified, ignore movablecore option | |
7153 | */ | |
7154 | if (mirrored_kernelcore) { | |
7155 | bool mem_below_4gb_not_mirrored = false; | |
7156 | ||
7157 | for_each_memblock(memory, r) { | |
7158 | if (memblock_is_mirror(r)) | |
7159 | continue; | |
7160 | ||
7161 | nid = r->nid; | |
7162 | ||
7163 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
7164 | ||
7165 | if (usable_startpfn < 0x100000) { | |
7166 | mem_below_4gb_not_mirrored = true; | |
7167 | continue; | |
7168 | } | |
7169 | ||
7170 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
7171 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7172 | usable_startpfn; | |
7173 | } | |
7174 | ||
7175 | if (mem_below_4gb_not_mirrored) | |
7176 | pr_warn("This configuration results in unmirrored kernel memory."); | |
7177 | ||
7178 | goto out2; | |
7179 | } | |
7180 | ||
7e63efef | 7181 | /* |
a5c6d650 DR |
7182 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
7183 | * amount of necessary memory. | |
7184 | */ | |
7185 | if (required_kernelcore_percent) | |
7186 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
7187 | 10000UL; | |
7188 | if (required_movablecore_percent) | |
7189 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
7190 | 10000UL; | |
7191 | ||
7192 | /* | |
7193 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
7194 | * kernelcore that corresponds so that memory usable for |
7195 | * any allocation type is evenly spread. If both kernelcore | |
7196 | * and movablecore are specified, then the value of kernelcore | |
7197 | * will be used for required_kernelcore if it's greater than | |
7198 | * what movablecore would have allowed. | |
7199 | */ | |
7200 | if (required_movablecore) { | |
7e63efef MG |
7201 | unsigned long corepages; |
7202 | ||
7203 | /* | |
7204 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
7205 | * was requested by the user | |
7206 | */ | |
7207 | required_movablecore = | |
7208 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 7209 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
7210 | corepages = totalpages - required_movablecore; |
7211 | ||
7212 | required_kernelcore = max(required_kernelcore, corepages); | |
7213 | } | |
7214 | ||
bde304bd XQ |
7215 | /* |
7216 | * If kernelcore was not specified or kernelcore size is larger | |
7217 | * than totalpages, there is no ZONE_MOVABLE. | |
7218 | */ | |
7219 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 7220 | goto out; |
2a1e274a MG |
7221 | |
7222 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
7223 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
7224 | ||
7225 | restart: | |
7226 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
7227 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 7228 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
7229 | unsigned long start_pfn, end_pfn; |
7230 | ||
2a1e274a MG |
7231 | /* |
7232 | * Recalculate kernelcore_node if the division per node | |
7233 | * now exceeds what is necessary to satisfy the requested | |
7234 | * amount of memory for the kernel | |
7235 | */ | |
7236 | if (required_kernelcore < kernelcore_node) | |
7237 | kernelcore_node = required_kernelcore / usable_nodes; | |
7238 | ||
7239 | /* | |
7240 | * As the map is walked, we track how much memory is usable | |
7241 | * by the kernel using kernelcore_remaining. When it is | |
7242 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
7243 | */ | |
7244 | kernelcore_remaining = kernelcore_node; | |
7245 | ||
7246 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 7247 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
7248 | unsigned long size_pages; |
7249 | ||
c13291a5 | 7250 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
7251 | if (start_pfn >= end_pfn) |
7252 | continue; | |
7253 | ||
7254 | /* Account for what is only usable for kernelcore */ | |
7255 | if (start_pfn < usable_startpfn) { | |
7256 | unsigned long kernel_pages; | |
7257 | kernel_pages = min(end_pfn, usable_startpfn) | |
7258 | - start_pfn; | |
7259 | ||
7260 | kernelcore_remaining -= min(kernel_pages, | |
7261 | kernelcore_remaining); | |
7262 | required_kernelcore -= min(kernel_pages, | |
7263 | required_kernelcore); | |
7264 | ||
7265 | /* Continue if range is now fully accounted */ | |
7266 | if (end_pfn <= usable_startpfn) { | |
7267 | ||
7268 | /* | |
7269 | * Push zone_movable_pfn to the end so | |
7270 | * that if we have to rebalance | |
7271 | * kernelcore across nodes, we will | |
7272 | * not double account here | |
7273 | */ | |
7274 | zone_movable_pfn[nid] = end_pfn; | |
7275 | continue; | |
7276 | } | |
7277 | start_pfn = usable_startpfn; | |
7278 | } | |
7279 | ||
7280 | /* | |
7281 | * The usable PFN range for ZONE_MOVABLE is from | |
7282 | * start_pfn->end_pfn. Calculate size_pages as the | |
7283 | * number of pages used as kernelcore | |
7284 | */ | |
7285 | size_pages = end_pfn - start_pfn; | |
7286 | if (size_pages > kernelcore_remaining) | |
7287 | size_pages = kernelcore_remaining; | |
7288 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
7289 | ||
7290 | /* | |
7291 | * Some kernelcore has been met, update counts and | |
7292 | * break if the kernelcore for this node has been | |
b8af2941 | 7293 | * satisfied |
2a1e274a MG |
7294 | */ |
7295 | required_kernelcore -= min(required_kernelcore, | |
7296 | size_pages); | |
7297 | kernelcore_remaining -= size_pages; | |
7298 | if (!kernelcore_remaining) | |
7299 | break; | |
7300 | } | |
7301 | } | |
7302 | ||
7303 | /* | |
7304 | * If there is still required_kernelcore, we do another pass with one | |
7305 | * less node in the count. This will push zone_movable_pfn[nid] further | |
7306 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 7307 | * satisfied |
2a1e274a MG |
7308 | */ |
7309 | usable_nodes--; | |
7310 | if (usable_nodes && required_kernelcore > usable_nodes) | |
7311 | goto restart; | |
7312 | ||
b2f3eebe | 7313 | out2: |
2a1e274a MG |
7314 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
7315 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
7316 | zone_movable_pfn[nid] = | |
7317 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 7318 | |
20e6926d | 7319 | out: |
66918dcd | 7320 | /* restore the node_state */ |
4b0ef1fe | 7321 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
7322 | } |
7323 | ||
4b0ef1fe LJ |
7324 | /* Any regular or high memory on that node ? */ |
7325 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 7326 | { |
37b07e41 LS |
7327 | enum zone_type zone_type; |
7328 | ||
4b0ef1fe | 7329 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 7330 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 7331 | if (populated_zone(zone)) { |
7b0e0c0e OS |
7332 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
7333 | node_set_state(nid, N_HIGH_MEMORY); | |
7334 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 7335 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
7336 | break; |
7337 | } | |
37b07e41 | 7338 | } |
37b07e41 LS |
7339 | } |
7340 | ||
c713216d MG |
7341 | /** |
7342 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 7343 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
7344 | * |
7345 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 7346 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
7347 | * zone in each node and their holes is calculated. If the maximum PFN |
7348 | * between two adjacent zones match, it is assumed that the zone is empty. | |
7349 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
7350 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
7351 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
7352 | * at arch_max_dma_pfn. | |
7353 | */ | |
7354 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
7355 | { | |
c13291a5 TH |
7356 | unsigned long start_pfn, end_pfn; |
7357 | int i, nid; | |
a6af2bc3 | 7358 | |
c713216d MG |
7359 | /* Record where the zone boundaries are */ |
7360 | memset(arch_zone_lowest_possible_pfn, 0, | |
7361 | sizeof(arch_zone_lowest_possible_pfn)); | |
7362 | memset(arch_zone_highest_possible_pfn, 0, | |
7363 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
7364 | |
7365 | start_pfn = find_min_pfn_with_active_regions(); | |
7366 | ||
7367 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
7368 | if (i == ZONE_MOVABLE) |
7369 | continue; | |
90cae1fe OH |
7370 | |
7371 | end_pfn = max(max_zone_pfn[i], start_pfn); | |
7372 | arch_zone_lowest_possible_pfn[i] = start_pfn; | |
7373 | arch_zone_highest_possible_pfn[i] = end_pfn; | |
7374 | ||
7375 | start_pfn = end_pfn; | |
c713216d | 7376 | } |
2a1e274a MG |
7377 | |
7378 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
7379 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 7380 | find_zone_movable_pfns_for_nodes(); |
c713216d | 7381 | |
c713216d | 7382 | /* Print out the zone ranges */ |
f88dfff5 | 7383 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
7384 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7385 | if (i == ZONE_MOVABLE) | |
7386 | continue; | |
f88dfff5 | 7387 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
7388 | if (arch_zone_lowest_possible_pfn[i] == |
7389 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 7390 | pr_cont("empty\n"); |
72f0ba02 | 7391 | else |
8d29e18a JG |
7392 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
7393 | (u64)arch_zone_lowest_possible_pfn[i] | |
7394 | << PAGE_SHIFT, | |
7395 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 7396 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
7397 | } |
7398 | ||
7399 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 7400 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
7401 | for (i = 0; i < MAX_NUMNODES; i++) { |
7402 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
7403 | pr_info(" Node %d: %#018Lx\n", i, |
7404 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 7405 | } |
c713216d | 7406 | |
f46edbd1 DW |
7407 | /* |
7408 | * Print out the early node map, and initialize the | |
7409 | * subsection-map relative to active online memory ranges to | |
7410 | * enable future "sub-section" extensions of the memory map. | |
7411 | */ | |
f88dfff5 | 7412 | pr_info("Early memory node ranges\n"); |
f46edbd1 | 7413 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
8d29e18a JG |
7414 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
7415 | (u64)start_pfn << PAGE_SHIFT, | |
7416 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
f46edbd1 DW |
7417 | subsection_map_init(start_pfn, end_pfn - start_pfn); |
7418 | } | |
c713216d MG |
7419 | |
7420 | /* Initialise every node */ | |
708614e6 | 7421 | mminit_verify_pageflags_layout(); |
8ef82866 | 7422 | setup_nr_node_ids(); |
4b094b78 | 7423 | init_unavailable_mem(); |
c713216d MG |
7424 | for_each_online_node(nid) { |
7425 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 7426 | free_area_init_node(nid, NULL, |
c713216d | 7427 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
7428 | |
7429 | /* Any memory on that node */ | |
7430 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
7431 | node_set_state(nid, N_MEMORY); |
7432 | check_for_memory(pgdat, nid); | |
c713216d MG |
7433 | } |
7434 | } | |
2a1e274a | 7435 | |
a5c6d650 DR |
7436 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
7437 | unsigned long *percent) | |
2a1e274a MG |
7438 | { |
7439 | unsigned long long coremem; | |
a5c6d650 DR |
7440 | char *endptr; |
7441 | ||
2a1e274a MG |
7442 | if (!p) |
7443 | return -EINVAL; | |
7444 | ||
a5c6d650 DR |
7445 | /* Value may be a percentage of total memory, otherwise bytes */ |
7446 | coremem = simple_strtoull(p, &endptr, 0); | |
7447 | if (*endptr == '%') { | |
7448 | /* Paranoid check for percent values greater than 100 */ | |
7449 | WARN_ON(coremem > 100); | |
2a1e274a | 7450 | |
a5c6d650 DR |
7451 | *percent = coremem; |
7452 | } else { | |
7453 | coremem = memparse(p, &p); | |
7454 | /* Paranoid check that UL is enough for the coremem value */ | |
7455 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7456 | |
a5c6d650 DR |
7457 | *core = coremem >> PAGE_SHIFT; |
7458 | *percent = 0UL; | |
7459 | } | |
2a1e274a MG |
7460 | return 0; |
7461 | } | |
ed7ed365 | 7462 | |
7e63efef MG |
7463 | /* |
7464 | * kernelcore=size sets the amount of memory for use for allocations that | |
7465 | * cannot be reclaimed or migrated. | |
7466 | */ | |
7467 | static int __init cmdline_parse_kernelcore(char *p) | |
7468 | { | |
342332e6 TI |
7469 | /* parse kernelcore=mirror */ |
7470 | if (parse_option_str(p, "mirror")) { | |
7471 | mirrored_kernelcore = true; | |
7472 | return 0; | |
7473 | } | |
7474 | ||
a5c6d650 DR |
7475 | return cmdline_parse_core(p, &required_kernelcore, |
7476 | &required_kernelcore_percent); | |
7e63efef MG |
7477 | } |
7478 | ||
7479 | /* | |
7480 | * movablecore=size sets the amount of memory for use for allocations that | |
7481 | * can be reclaimed or migrated. | |
7482 | */ | |
7483 | static int __init cmdline_parse_movablecore(char *p) | |
7484 | { | |
a5c6d650 DR |
7485 | return cmdline_parse_core(p, &required_movablecore, |
7486 | &required_movablecore_percent); | |
7e63efef MG |
7487 | } |
7488 | ||
ed7ed365 | 7489 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7490 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7491 | |
0ee332c1 | 7492 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 7493 | |
c3d5f5f0 JL |
7494 | void adjust_managed_page_count(struct page *page, long count) |
7495 | { | |
9705bea5 | 7496 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 7497 | totalram_pages_add(count); |
3dcc0571 JL |
7498 | #ifdef CONFIG_HIGHMEM |
7499 | if (PageHighMem(page)) | |
ca79b0c2 | 7500 | totalhigh_pages_add(count); |
3dcc0571 | 7501 | #endif |
c3d5f5f0 | 7502 | } |
3dcc0571 | 7503 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7504 | |
e5cb113f | 7505 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 7506 | { |
11199692 JL |
7507 | void *pos; |
7508 | unsigned long pages = 0; | |
69afade7 | 7509 | |
11199692 JL |
7510 | start = (void *)PAGE_ALIGN((unsigned long)start); |
7511 | end = (void *)((unsigned long)end & PAGE_MASK); | |
7512 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
7513 | struct page *page = virt_to_page(pos); |
7514 | void *direct_map_addr; | |
7515 | ||
7516 | /* | |
7517 | * 'direct_map_addr' might be different from 'pos' | |
7518 | * because some architectures' virt_to_page() | |
7519 | * work with aliases. Getting the direct map | |
7520 | * address ensures that we get a _writeable_ | |
7521 | * alias for the memset(). | |
7522 | */ | |
7523 | direct_map_addr = page_address(page); | |
dbe67df4 | 7524 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
7525 | memset(direct_map_addr, poison, PAGE_SIZE); |
7526 | ||
7527 | free_reserved_page(page); | |
69afade7 JL |
7528 | } |
7529 | ||
7530 | if (pages && s) | |
adb1fe9a JP |
7531 | pr_info("Freeing %s memory: %ldK\n", |
7532 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
7533 | |
7534 | return pages; | |
7535 | } | |
7536 | ||
cfa11e08 JL |
7537 | #ifdef CONFIG_HIGHMEM |
7538 | void free_highmem_page(struct page *page) | |
7539 | { | |
7540 | __free_reserved_page(page); | |
ca79b0c2 | 7541 | totalram_pages_inc(); |
9705bea5 | 7542 | atomic_long_inc(&page_zone(page)->managed_pages); |
ca79b0c2 | 7543 | totalhigh_pages_inc(); |
cfa11e08 JL |
7544 | } |
7545 | #endif | |
7546 | ||
7ee3d4e8 JL |
7547 | |
7548 | void __init mem_init_print_info(const char *str) | |
7549 | { | |
7550 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7551 | unsigned long init_code_size, init_data_size; | |
7552 | ||
7553 | physpages = get_num_physpages(); | |
7554 | codesize = _etext - _stext; | |
7555 | datasize = _edata - _sdata; | |
7556 | rosize = __end_rodata - __start_rodata; | |
7557 | bss_size = __bss_stop - __bss_start; | |
7558 | init_data_size = __init_end - __init_begin; | |
7559 | init_code_size = _einittext - _sinittext; | |
7560 | ||
7561 | /* | |
7562 | * Detect special cases and adjust section sizes accordingly: | |
7563 | * 1) .init.* may be embedded into .data sections | |
7564 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7565 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7566 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7567 | */ | |
7568 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7569 | do { \ |
7570 | if (start <= pos && pos < end && size > adj) \ | |
7571 | size -= adj; \ | |
7572 | } while (0) | |
7ee3d4e8 JL |
7573 | |
7574 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7575 | _sinittext, init_code_size); | |
7576 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7577 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7578 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7579 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7580 | ||
7581 | #undef adj_init_size | |
7582 | ||
756a025f | 7583 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7584 | #ifdef CONFIG_HIGHMEM |
756a025f | 7585 | ", %luK highmem" |
7ee3d4e8 | 7586 | #endif |
756a025f JP |
7587 | "%s%s)\n", |
7588 | nr_free_pages() << (PAGE_SHIFT - 10), | |
7589 | physpages << (PAGE_SHIFT - 10), | |
7590 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7591 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
ca79b0c2 | 7592 | (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10), |
756a025f | 7593 | totalcma_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7594 | #ifdef CONFIG_HIGHMEM |
ca79b0c2 | 7595 | totalhigh_pages() << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7596 | #endif |
756a025f | 7597 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
7598 | } |
7599 | ||
0e0b864e | 7600 | /** |
88ca3b94 RD |
7601 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7602 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7603 | * |
013110a7 | 7604 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7605 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7606 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7607 | * function may optionally be used to account for unfreeable pages in the |
7608 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7609 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7610 | */ |
7611 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7612 | { | |
7613 | dma_reserve = new_dma_reserve; | |
7614 | } | |
7615 | ||
1da177e4 LT |
7616 | void __init free_area_init(unsigned long *zones_size) |
7617 | { | |
4b094b78 | 7618 | init_unavailable_mem(); |
9109fb7b | 7619 | free_area_init_node(0, zones_size, |
1da177e4 LT |
7620 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
7621 | } | |
1da177e4 | 7622 | |
005fd4bb | 7623 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7624 | { |
1da177e4 | 7625 | |
005fd4bb SAS |
7626 | lru_add_drain_cpu(cpu); |
7627 | drain_pages(cpu); | |
9f8f2172 | 7628 | |
005fd4bb SAS |
7629 | /* |
7630 | * Spill the event counters of the dead processor | |
7631 | * into the current processors event counters. | |
7632 | * This artificially elevates the count of the current | |
7633 | * processor. | |
7634 | */ | |
7635 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7636 | |
005fd4bb SAS |
7637 | /* |
7638 | * Zero the differential counters of the dead processor | |
7639 | * so that the vm statistics are consistent. | |
7640 | * | |
7641 | * This is only okay since the processor is dead and cannot | |
7642 | * race with what we are doing. | |
7643 | */ | |
7644 | cpu_vm_stats_fold(cpu); | |
7645 | return 0; | |
1da177e4 | 7646 | } |
1da177e4 | 7647 | |
e03a5125 NP |
7648 | #ifdef CONFIG_NUMA |
7649 | int hashdist = HASHDIST_DEFAULT; | |
7650 | ||
7651 | static int __init set_hashdist(char *str) | |
7652 | { | |
7653 | if (!str) | |
7654 | return 0; | |
7655 | hashdist = simple_strtoul(str, &str, 0); | |
7656 | return 1; | |
7657 | } | |
7658 | __setup("hashdist=", set_hashdist); | |
7659 | #endif | |
7660 | ||
1da177e4 LT |
7661 | void __init page_alloc_init(void) |
7662 | { | |
005fd4bb SAS |
7663 | int ret; |
7664 | ||
e03a5125 NP |
7665 | #ifdef CONFIG_NUMA |
7666 | if (num_node_state(N_MEMORY) == 1) | |
7667 | hashdist = 0; | |
7668 | #endif | |
7669 | ||
005fd4bb SAS |
7670 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, |
7671 | "mm/page_alloc:dead", NULL, | |
7672 | page_alloc_cpu_dead); | |
7673 | WARN_ON(ret < 0); | |
1da177e4 LT |
7674 | } |
7675 | ||
cb45b0e9 | 7676 | /* |
34b10060 | 7677 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
7678 | * or min_free_kbytes changes. |
7679 | */ | |
7680 | static void calculate_totalreserve_pages(void) | |
7681 | { | |
7682 | struct pglist_data *pgdat; | |
7683 | unsigned long reserve_pages = 0; | |
2f6726e5 | 7684 | enum zone_type i, j; |
cb45b0e9 HA |
7685 | |
7686 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
7687 | |
7688 | pgdat->totalreserve_pages = 0; | |
7689 | ||
cb45b0e9 HA |
7690 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7691 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 7692 | long max = 0; |
9705bea5 | 7693 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
7694 | |
7695 | /* Find valid and maximum lowmem_reserve in the zone */ | |
7696 | for (j = i; j < MAX_NR_ZONES; j++) { | |
7697 | if (zone->lowmem_reserve[j] > max) | |
7698 | max = zone->lowmem_reserve[j]; | |
7699 | } | |
7700 | ||
41858966 MG |
7701 | /* we treat the high watermark as reserved pages. */ |
7702 | max += high_wmark_pages(zone); | |
cb45b0e9 | 7703 | |
3d6357de AK |
7704 | if (max > managed_pages) |
7705 | max = managed_pages; | |
a8d01437 | 7706 | |
281e3726 | 7707 | pgdat->totalreserve_pages += max; |
a8d01437 | 7708 | |
cb45b0e9 HA |
7709 | reserve_pages += max; |
7710 | } | |
7711 | } | |
7712 | totalreserve_pages = reserve_pages; | |
7713 | } | |
7714 | ||
1da177e4 LT |
7715 | /* |
7716 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 7717 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
7718 | * has a correct pages reserved value, so an adequate number of |
7719 | * pages are left in the zone after a successful __alloc_pages(). | |
7720 | */ | |
7721 | static void setup_per_zone_lowmem_reserve(void) | |
7722 | { | |
7723 | struct pglist_data *pgdat; | |
2f6726e5 | 7724 | enum zone_type j, idx; |
1da177e4 | 7725 | |
ec936fc5 | 7726 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
7727 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7728 | struct zone *zone = pgdat->node_zones + j; | |
9705bea5 | 7729 | unsigned long managed_pages = zone_managed_pages(zone); |
1da177e4 LT |
7730 | |
7731 | zone->lowmem_reserve[j] = 0; | |
7732 | ||
2f6726e5 CL |
7733 | idx = j; |
7734 | while (idx) { | |
1da177e4 LT |
7735 | struct zone *lower_zone; |
7736 | ||
2f6726e5 | 7737 | idx--; |
1da177e4 | 7738 | lower_zone = pgdat->node_zones + idx; |
d3cda233 JK |
7739 | |
7740 | if (sysctl_lowmem_reserve_ratio[idx] < 1) { | |
7741 | sysctl_lowmem_reserve_ratio[idx] = 0; | |
7742 | lower_zone->lowmem_reserve[j] = 0; | |
7743 | } else { | |
7744 | lower_zone->lowmem_reserve[j] = | |
7745 | managed_pages / sysctl_lowmem_reserve_ratio[idx]; | |
7746 | } | |
9705bea5 | 7747 | managed_pages += zone_managed_pages(lower_zone); |
1da177e4 LT |
7748 | } |
7749 | } | |
7750 | } | |
cb45b0e9 HA |
7751 | |
7752 | /* update totalreserve_pages */ | |
7753 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7754 | } |
7755 | ||
cfd3da1e | 7756 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
7757 | { |
7758 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
7759 | unsigned long lowmem_pages = 0; | |
7760 | struct zone *zone; | |
7761 | unsigned long flags; | |
7762 | ||
7763 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
7764 | for_each_zone(zone) { | |
7765 | if (!is_highmem(zone)) | |
9705bea5 | 7766 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
7767 | } |
7768 | ||
7769 | for_each_zone(zone) { | |
ac924c60 AM |
7770 | u64 tmp; |
7771 | ||
1125b4e3 | 7772 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 7773 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 7774 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
7775 | if (is_highmem(zone)) { |
7776 | /* | |
669ed175 NP |
7777 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
7778 | * need highmem pages, so cap pages_min to a small | |
7779 | * value here. | |
7780 | * | |
41858966 | 7781 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 7782 | * deltas control async page reclaim, and so should |
669ed175 | 7783 | * not be capped for highmem. |
1da177e4 | 7784 | */ |
90ae8d67 | 7785 | unsigned long min_pages; |
1da177e4 | 7786 | |
9705bea5 | 7787 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 7788 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 7789 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 7790 | } else { |
669ed175 NP |
7791 | /* |
7792 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
7793 | * proportionate to the zone's size. |
7794 | */ | |
a9214443 | 7795 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
7796 | } |
7797 | ||
795ae7a0 JW |
7798 | /* |
7799 | * Set the kswapd watermarks distance according to the | |
7800 | * scale factor in proportion to available memory, but | |
7801 | * ensure a minimum size on small systems. | |
7802 | */ | |
7803 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 7804 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
7805 | watermark_scale_factor, 10000)); |
7806 | ||
a9214443 MG |
7807 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
7808 | zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
1c30844d | 7809 | zone->watermark_boost = 0; |
49f223a9 | 7810 | |
1125b4e3 | 7811 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 7812 | } |
cb45b0e9 HA |
7813 | |
7814 | /* update totalreserve_pages */ | |
7815 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7816 | } |
7817 | ||
cfd3da1e MG |
7818 | /** |
7819 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
7820 | * or when memory is hot-{added|removed} | |
7821 | * | |
7822 | * Ensures that the watermark[min,low,high] values for each zone are set | |
7823 | * correctly with respect to min_free_kbytes. | |
7824 | */ | |
7825 | void setup_per_zone_wmarks(void) | |
7826 | { | |
b93e0f32 MH |
7827 | static DEFINE_SPINLOCK(lock); |
7828 | ||
7829 | spin_lock(&lock); | |
cfd3da1e | 7830 | __setup_per_zone_wmarks(); |
b93e0f32 | 7831 | spin_unlock(&lock); |
cfd3da1e MG |
7832 | } |
7833 | ||
1da177e4 LT |
7834 | /* |
7835 | * Initialise min_free_kbytes. | |
7836 | * | |
7837 | * For small machines we want it small (128k min). For large machines | |
7838 | * we want it large (64MB max). But it is not linear, because network | |
7839 | * bandwidth does not increase linearly with machine size. We use | |
7840 | * | |
b8af2941 | 7841 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
7842 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
7843 | * | |
7844 | * which yields | |
7845 | * | |
7846 | * 16MB: 512k | |
7847 | * 32MB: 724k | |
7848 | * 64MB: 1024k | |
7849 | * 128MB: 1448k | |
7850 | * 256MB: 2048k | |
7851 | * 512MB: 2896k | |
7852 | * 1024MB: 4096k | |
7853 | * 2048MB: 5792k | |
7854 | * 4096MB: 8192k | |
7855 | * 8192MB: 11584k | |
7856 | * 16384MB: 16384k | |
7857 | */ | |
1b79acc9 | 7858 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
7859 | { |
7860 | unsigned long lowmem_kbytes; | |
5f12733e | 7861 | int new_min_free_kbytes; |
1da177e4 LT |
7862 | |
7863 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
7864 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
7865 | ||
7866 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
7867 | min_free_kbytes = new_min_free_kbytes; | |
7868 | if (min_free_kbytes < 128) | |
7869 | min_free_kbytes = 128; | |
7870 | if (min_free_kbytes > 65536) | |
7871 | min_free_kbytes = 65536; | |
7872 | } else { | |
7873 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
7874 | new_min_free_kbytes, user_min_free_kbytes); | |
7875 | } | |
bc75d33f | 7876 | setup_per_zone_wmarks(); |
a6cccdc3 | 7877 | refresh_zone_stat_thresholds(); |
1da177e4 | 7878 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
7879 | |
7880 | #ifdef CONFIG_NUMA | |
7881 | setup_min_unmapped_ratio(); | |
7882 | setup_min_slab_ratio(); | |
7883 | #endif | |
7884 | ||
1da177e4 LT |
7885 | return 0; |
7886 | } | |
bc22af74 | 7887 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
7888 | |
7889 | /* | |
b8af2941 | 7890 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
7891 | * that we can call two helper functions whenever min_free_kbytes |
7892 | * changes. | |
7893 | */ | |
cccad5b9 | 7894 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7895 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7896 | { |
da8c757b HP |
7897 | int rc; |
7898 | ||
7899 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7900 | if (rc) | |
7901 | return rc; | |
7902 | ||
5f12733e MH |
7903 | if (write) { |
7904 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7905 | setup_per_zone_wmarks(); |
5f12733e | 7906 | } |
1da177e4 LT |
7907 | return 0; |
7908 | } | |
7909 | ||
1c30844d MG |
7910 | int watermark_boost_factor_sysctl_handler(struct ctl_table *table, int write, |
7911 | void __user *buffer, size_t *length, loff_t *ppos) | |
7912 | { | |
7913 | int rc; | |
7914 | ||
7915 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7916 | if (rc) | |
7917 | return rc; | |
7918 | ||
7919 | return 0; | |
7920 | } | |
7921 | ||
795ae7a0 JW |
7922 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
7923 | void __user *buffer, size_t *length, loff_t *ppos) | |
7924 | { | |
7925 | int rc; | |
7926 | ||
7927 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7928 | if (rc) | |
7929 | return rc; | |
7930 | ||
7931 | if (write) | |
7932 | setup_per_zone_wmarks(); | |
7933 | ||
7934 | return 0; | |
7935 | } | |
7936 | ||
9614634f | 7937 | #ifdef CONFIG_NUMA |
6423aa81 | 7938 | static void setup_min_unmapped_ratio(void) |
9614634f | 7939 | { |
6423aa81 | 7940 | pg_data_t *pgdat; |
9614634f | 7941 | struct zone *zone; |
9614634f | 7942 | |
a5f5f91d | 7943 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7944 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7945 | |
9614634f | 7946 | for_each_zone(zone) |
9705bea5 AK |
7947 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
7948 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 7949 | } |
0ff38490 | 7950 | |
6423aa81 JK |
7951 | |
7952 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 7953 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7954 | { |
0ff38490 CL |
7955 | int rc; |
7956 | ||
8d65af78 | 7957 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7958 | if (rc) |
7959 | return rc; | |
7960 | ||
6423aa81 JK |
7961 | setup_min_unmapped_ratio(); |
7962 | ||
7963 | return 0; | |
7964 | } | |
7965 | ||
7966 | static void setup_min_slab_ratio(void) | |
7967 | { | |
7968 | pg_data_t *pgdat; | |
7969 | struct zone *zone; | |
7970 | ||
a5f5f91d MG |
7971 | for_each_online_pgdat(pgdat) |
7972 | pgdat->min_slab_pages = 0; | |
7973 | ||
0ff38490 | 7974 | for_each_zone(zone) |
9705bea5 AK |
7975 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
7976 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
7977 | } |
7978 | ||
7979 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
7980 | void __user *buffer, size_t *length, loff_t *ppos) | |
7981 | { | |
7982 | int rc; | |
7983 | ||
7984 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7985 | if (rc) | |
7986 | return rc; | |
7987 | ||
7988 | setup_min_slab_ratio(); | |
7989 | ||
0ff38490 CL |
7990 | return 0; |
7991 | } | |
9614634f CL |
7992 | #endif |
7993 | ||
1da177e4 LT |
7994 | /* |
7995 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7996 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7997 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7998 | * | |
7999 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 8000 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
8001 | * if in function of the boot time zone sizes. |
8002 | */ | |
cccad5b9 | 8003 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 8004 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 8005 | { |
8d65af78 | 8006 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
8007 | setup_per_zone_lowmem_reserve(); |
8008 | return 0; | |
8009 | } | |
8010 | ||
cb1ef534 MG |
8011 | static void __zone_pcp_update(struct zone *zone) |
8012 | { | |
8013 | unsigned int cpu; | |
8014 | ||
8015 | for_each_possible_cpu(cpu) | |
8016 | pageset_set_high_and_batch(zone, | |
8017 | per_cpu_ptr(zone->pageset, cpu)); | |
8018 | } | |
8019 | ||
8ad4b1fb RS |
8020 | /* |
8021 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
8022 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
8023 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 8024 | */ |
cccad5b9 | 8025 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 8026 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
8027 | { |
8028 | struct zone *zone; | |
7cd2b0a3 | 8029 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
8030 | int ret; |
8031 | ||
7cd2b0a3 DR |
8032 | mutex_lock(&pcp_batch_high_lock); |
8033 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
8034 | ||
8d65af78 | 8035 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
8036 | if (!write || ret < 0) |
8037 | goto out; | |
8038 | ||
8039 | /* Sanity checking to avoid pcp imbalance */ | |
8040 | if (percpu_pagelist_fraction && | |
8041 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
8042 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
8043 | ret = -EINVAL; | |
8044 | goto out; | |
8045 | } | |
8046 | ||
8047 | /* No change? */ | |
8048 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
8049 | goto out; | |
c8e251fa | 8050 | |
cb1ef534 MG |
8051 | for_each_populated_zone(zone) |
8052 | __zone_pcp_update(zone); | |
7cd2b0a3 | 8053 | out: |
c8e251fa | 8054 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 8055 | return ret; |
8ad4b1fb RS |
8056 | } |
8057 | ||
f6f34b43 SD |
8058 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
8059 | /* | |
8060 | * Returns the number of pages that arch has reserved but | |
8061 | * is not known to alloc_large_system_hash(). | |
8062 | */ | |
8063 | static unsigned long __init arch_reserved_kernel_pages(void) | |
8064 | { | |
8065 | return 0; | |
8066 | } | |
8067 | #endif | |
8068 | ||
9017217b PT |
8069 | /* |
8070 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
8071 | * machines. As memory size is increased the scale is also increased but at | |
8072 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
8073 | * quadruples the scale is increased by one, which means the size of hash table | |
8074 | * only doubles, instead of quadrupling as well. | |
8075 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
8076 | * makes sense, it is disabled on such platforms. | |
8077 | */ | |
8078 | #if __BITS_PER_LONG > 32 | |
8079 | #define ADAPT_SCALE_BASE (64ul << 30) | |
8080 | #define ADAPT_SCALE_SHIFT 2 | |
8081 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
8082 | #endif | |
8083 | ||
1da177e4 LT |
8084 | /* |
8085 | * allocate a large system hash table from bootmem | |
8086 | * - it is assumed that the hash table must contain an exact power-of-2 | |
8087 | * quantity of entries | |
8088 | * - limit is the number of hash buckets, not the total allocation size | |
8089 | */ | |
8090 | void *__init alloc_large_system_hash(const char *tablename, | |
8091 | unsigned long bucketsize, | |
8092 | unsigned long numentries, | |
8093 | int scale, | |
8094 | int flags, | |
8095 | unsigned int *_hash_shift, | |
8096 | unsigned int *_hash_mask, | |
31fe62b9 TB |
8097 | unsigned long low_limit, |
8098 | unsigned long high_limit) | |
1da177e4 | 8099 | { |
31fe62b9 | 8100 | unsigned long long max = high_limit; |
1da177e4 LT |
8101 | unsigned long log2qty, size; |
8102 | void *table = NULL; | |
3749a8f0 | 8103 | gfp_t gfp_flags; |
ec11408a | 8104 | bool virt; |
1da177e4 LT |
8105 | |
8106 | /* allow the kernel cmdline to have a say */ | |
8107 | if (!numentries) { | |
8108 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 8109 | numentries = nr_kernel_pages; |
f6f34b43 | 8110 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
8111 | |
8112 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
8113 | if (PAGE_SHIFT < 20) | |
8114 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 8115 | |
9017217b PT |
8116 | #if __BITS_PER_LONG > 32 |
8117 | if (!high_limit) { | |
8118 | unsigned long adapt; | |
8119 | ||
8120 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
8121 | adapt <<= ADAPT_SCALE_SHIFT) | |
8122 | scale++; | |
8123 | } | |
8124 | #endif | |
8125 | ||
1da177e4 LT |
8126 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
8127 | if (scale > PAGE_SHIFT) | |
8128 | numentries >>= (scale - PAGE_SHIFT); | |
8129 | else | |
8130 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
8131 | |
8132 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
8133 | if (unlikely(flags & HASH_SMALL)) { |
8134 | /* Makes no sense without HASH_EARLY */ | |
8135 | WARN_ON(!(flags & HASH_EARLY)); | |
8136 | if (!(numentries >> *_hash_shift)) { | |
8137 | numentries = 1UL << *_hash_shift; | |
8138 | BUG_ON(!numentries); | |
8139 | } | |
8140 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 8141 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 8142 | } |
6e692ed3 | 8143 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
8144 | |
8145 | /* limit allocation size to 1/16 total memory by default */ | |
8146 | if (max == 0) { | |
8147 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
8148 | do_div(max, bucketsize); | |
8149 | } | |
074b8517 | 8150 | max = min(max, 0x80000000ULL); |
1da177e4 | 8151 | |
31fe62b9 TB |
8152 | if (numentries < low_limit) |
8153 | numentries = low_limit; | |
1da177e4 LT |
8154 | if (numentries > max) |
8155 | numentries = max; | |
8156 | ||
f0d1b0b3 | 8157 | log2qty = ilog2(numentries); |
1da177e4 | 8158 | |
3749a8f0 | 8159 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 8160 | do { |
ec11408a | 8161 | virt = false; |
1da177e4 | 8162 | size = bucketsize << log2qty; |
ea1f5f37 PT |
8163 | if (flags & HASH_EARLY) { |
8164 | if (flags & HASH_ZERO) | |
26fb3dae | 8165 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 8166 | else |
7e1c4e27 MR |
8167 | table = memblock_alloc_raw(size, |
8168 | SMP_CACHE_BYTES); | |
ec11408a | 8169 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
3749a8f0 | 8170 | table = __vmalloc(size, gfp_flags, PAGE_KERNEL); |
ec11408a | 8171 | virt = true; |
ea1f5f37 | 8172 | } else { |
1037b83b ED |
8173 | /* |
8174 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
8175 | * some pages at the end of hash table which |
8176 | * alloc_pages_exact() automatically does | |
1037b83b | 8177 | */ |
ec11408a NP |
8178 | table = alloc_pages_exact(size, gfp_flags); |
8179 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
8180 | } |
8181 | } while (!table && size > PAGE_SIZE && --log2qty); | |
8182 | ||
8183 | if (!table) | |
8184 | panic("Failed to allocate %s hash table\n", tablename); | |
8185 | ||
ec11408a NP |
8186 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
8187 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
8188 | virt ? "vmalloc" : "linear"); | |
1da177e4 LT |
8189 | |
8190 | if (_hash_shift) | |
8191 | *_hash_shift = log2qty; | |
8192 | if (_hash_mask) | |
8193 | *_hash_mask = (1 << log2qty) - 1; | |
8194 | ||
8195 | return table; | |
8196 | } | |
a117e66e | 8197 | |
a5d76b54 | 8198 | /* |
80934513 | 8199 | * This function checks whether pageblock includes unmovable pages or not. |
80934513 | 8200 | * |
b8af2941 | 8201 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
8202 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
8203 | * check without lock_page also may miss some movable non-lru pages at | |
8204 | * race condition. So you can't expect this function should be exact. | |
4a55c047 QC |
8205 | * |
8206 | * Returns a page without holding a reference. If the caller wants to | |
8207 | * dereference that page (e.g., dumping), it has to make sure that that it | |
8208 | * cannot get removed (e.g., via memory unplug) concurrently. | |
8209 | * | |
a5d76b54 | 8210 | */ |
4a55c047 QC |
8211 | struct page *has_unmovable_pages(struct zone *zone, struct page *page, |
8212 | int migratetype, int flags) | |
49ac8255 | 8213 | { |
1a9f2191 QC |
8214 | unsigned long iter = 0; |
8215 | unsigned long pfn = page_to_pfn(page); | |
47118af0 | 8216 | |
49ac8255 | 8217 | /* |
15c30bc0 MH |
8218 | * TODO we could make this much more efficient by not checking every |
8219 | * page in the range if we know all of them are in MOVABLE_ZONE and | |
8220 | * that the movable zone guarantees that pages are migratable but | |
8221 | * the later is not the case right now unfortunatelly. E.g. movablecore | |
8222 | * can still lead to having bootmem allocations in zone_movable. | |
49ac8255 | 8223 | */ |
49ac8255 | 8224 | |
1a9f2191 QC |
8225 | if (is_migrate_cma_page(page)) { |
8226 | /* | |
8227 | * CMA allocations (alloc_contig_range) really need to mark | |
8228 | * isolate CMA pageblocks even when they are not movable in fact | |
8229 | * so consider them movable here. | |
8230 | */ | |
8231 | if (is_migrate_cma(migratetype)) | |
4a55c047 | 8232 | return NULL; |
1a9f2191 | 8233 | |
3d680bdf | 8234 | return page; |
1a9f2191 | 8235 | } |
4da2ce25 | 8236 | |
fe4c86c9 DH |
8237 | for (; iter < pageblock_nr_pages; iter++) { |
8238 | if (!pfn_valid_within(pfn + iter)) | |
49ac8255 | 8239 | continue; |
29723fcc | 8240 | |
fe4c86c9 | 8241 | page = pfn_to_page(pfn + iter); |
c8721bbb | 8242 | |
d7ab3672 | 8243 | if (PageReserved(page)) |
3d680bdf | 8244 | return page; |
d7ab3672 | 8245 | |
9d789999 MH |
8246 | /* |
8247 | * If the zone is movable and we have ruled out all reserved | |
8248 | * pages then it should be reasonably safe to assume the rest | |
8249 | * is movable. | |
8250 | */ | |
8251 | if (zone_idx(zone) == ZONE_MOVABLE) | |
8252 | continue; | |
8253 | ||
c8721bbb NH |
8254 | /* |
8255 | * Hugepages are not in LRU lists, but they're movable. | |
8bb4e7a2 | 8256 | * We need not scan over tail pages because we don't |
c8721bbb NH |
8257 | * handle each tail page individually in migration. |
8258 | */ | |
8259 | if (PageHuge(page)) { | |
17e2e7d7 OS |
8260 | struct page *head = compound_head(page); |
8261 | unsigned int skip_pages; | |
464c7ffb | 8262 | |
17e2e7d7 | 8263 | if (!hugepage_migration_supported(page_hstate(head))) |
3d680bdf | 8264 | return page; |
464c7ffb | 8265 | |
d8c6546b | 8266 | skip_pages = compound_nr(head) - (page - head); |
17e2e7d7 | 8267 | iter += skip_pages - 1; |
c8721bbb NH |
8268 | continue; |
8269 | } | |
8270 | ||
97d255c8 MK |
8271 | /* |
8272 | * We can't use page_count without pin a page | |
8273 | * because another CPU can free compound page. | |
8274 | * This check already skips compound tails of THP | |
0139aa7b | 8275 | * because their page->_refcount is zero at all time. |
97d255c8 | 8276 | */ |
fe896d18 | 8277 | if (!page_ref_count(page)) { |
49ac8255 KH |
8278 | if (PageBuddy(page)) |
8279 | iter += (1 << page_order(page)) - 1; | |
8280 | continue; | |
8281 | } | |
97d255c8 | 8282 | |
b023f468 WC |
8283 | /* |
8284 | * The HWPoisoned page may be not in buddy system, and | |
8285 | * page_count() is not 0. | |
8286 | */ | |
756d25be | 8287 | if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
b023f468 WC |
8288 | continue; |
8289 | ||
fe4c86c9 | 8290 | if (__PageMovable(page) || PageLRU(page)) |
0efadf48 YX |
8291 | continue; |
8292 | ||
49ac8255 | 8293 | /* |
6b4f7799 JW |
8294 | * If there are RECLAIMABLE pages, we need to check |
8295 | * it. But now, memory offline itself doesn't call | |
8296 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
8297 | */ |
8298 | /* | |
8299 | * If the page is not RAM, page_count()should be 0. | |
8300 | * we don't need more check. This is an _used_ not-movable page. | |
8301 | * | |
8302 | * The problematic thing here is PG_reserved pages. PG_reserved | |
8303 | * is set to both of a memory hole page and a _used_ kernel | |
8304 | * page at boot. | |
8305 | */ | |
3d680bdf | 8306 | return page; |
49ac8255 | 8307 | } |
4a55c047 | 8308 | return NULL; |
49ac8255 KH |
8309 | } |
8310 | ||
8df995f6 | 8311 | #ifdef CONFIG_CONTIG_ALLOC |
041d3a8c MN |
8312 | static unsigned long pfn_max_align_down(unsigned long pfn) |
8313 | { | |
8314 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8315 | pageblock_nr_pages) - 1); | |
8316 | } | |
8317 | ||
8318 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
8319 | { | |
8320 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8321 | pageblock_nr_pages)); | |
8322 | } | |
8323 | ||
041d3a8c | 8324 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
8325 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
8326 | unsigned long start, unsigned long end) | |
041d3a8c MN |
8327 | { |
8328 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 8329 | unsigned long nr_reclaimed; |
041d3a8c MN |
8330 | unsigned long pfn = start; |
8331 | unsigned int tries = 0; | |
8332 | int ret = 0; | |
8333 | ||
be49a6e1 | 8334 | migrate_prep(); |
041d3a8c | 8335 | |
bb13ffeb | 8336 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
8337 | if (fatal_signal_pending(current)) { |
8338 | ret = -EINTR; | |
8339 | break; | |
8340 | } | |
8341 | ||
bb13ffeb MG |
8342 | if (list_empty(&cc->migratepages)) { |
8343 | cc->nr_migratepages = 0; | |
edc2ca61 | 8344 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
8345 | if (!pfn) { |
8346 | ret = -EINTR; | |
8347 | break; | |
8348 | } | |
8349 | tries = 0; | |
8350 | } else if (++tries == 5) { | |
8351 | ret = ret < 0 ? ret : -EBUSY; | |
8352 | break; | |
8353 | } | |
8354 | ||
beb51eaa MK |
8355 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
8356 | &cc->migratepages); | |
8357 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 8358 | |
9c620e2b | 8359 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
31025351 | 8360 | NULL, 0, cc->mode, MR_CONTIG_RANGE); |
041d3a8c | 8361 | } |
2a6f5124 SP |
8362 | if (ret < 0) { |
8363 | putback_movable_pages(&cc->migratepages); | |
8364 | return ret; | |
8365 | } | |
8366 | return 0; | |
041d3a8c MN |
8367 | } |
8368 | ||
8369 | /** | |
8370 | * alloc_contig_range() -- tries to allocate given range of pages | |
8371 | * @start: start PFN to allocate | |
8372 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
8373 | * @migratetype: migratetype of the underlaying pageblocks (either |
8374 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
8375 | * in range must have the same migratetype and it must | |
8376 | * be either of the two. | |
ca96b625 | 8377 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
8378 | * |
8379 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 8380 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 8381 | * |
2c7452a0 MK |
8382 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
8383 | * pageblocks in the range. Once isolated, the pageblocks should not | |
8384 | * be modified by others. | |
041d3a8c | 8385 | * |
a862f68a | 8386 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
8387 | * pages which PFN is in [start, end) are allocated for the caller and |
8388 | * need to be freed with free_contig_range(). | |
8389 | */ | |
0815f3d8 | 8390 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 8391 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 8392 | { |
041d3a8c | 8393 | unsigned long outer_start, outer_end; |
d00181b9 KS |
8394 | unsigned int order; |
8395 | int ret = 0; | |
041d3a8c | 8396 | |
bb13ffeb MG |
8397 | struct compact_control cc = { |
8398 | .nr_migratepages = 0, | |
8399 | .order = -1, | |
8400 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 8401 | .mode = MIGRATE_SYNC, |
bb13ffeb | 8402 | .ignore_skip_hint = true, |
2583d671 | 8403 | .no_set_skip_hint = true, |
7dea19f9 | 8404 | .gfp_mask = current_gfp_context(gfp_mask), |
bb13ffeb MG |
8405 | }; |
8406 | INIT_LIST_HEAD(&cc.migratepages); | |
8407 | ||
041d3a8c MN |
8408 | /* |
8409 | * What we do here is we mark all pageblocks in range as | |
8410 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
8411 | * have different sizes, and due to the way page allocator | |
8412 | * work, we align the range to biggest of the two pages so | |
8413 | * that page allocator won't try to merge buddies from | |
8414 | * different pageblocks and change MIGRATE_ISOLATE to some | |
8415 | * other migration type. | |
8416 | * | |
8417 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
8418 | * migrate the pages from an unaligned range (ie. pages that | |
8419 | * we are interested in). This will put all the pages in | |
8420 | * range back to page allocator as MIGRATE_ISOLATE. | |
8421 | * | |
8422 | * When this is done, we take the pages in range from page | |
8423 | * allocator removing them from the buddy system. This way | |
8424 | * page allocator will never consider using them. | |
8425 | * | |
8426 | * This lets us mark the pageblocks back as | |
8427 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
8428 | * aligned range but not in the unaligned, original range are | |
8429 | * put back to page allocator so that buddy can use them. | |
8430 | */ | |
8431 | ||
8432 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
d381c547 | 8433 | pfn_max_align_up(end), migratetype, 0); |
9b7ea46a | 8434 | if (ret < 0) |
86a595f9 | 8435 | return ret; |
041d3a8c | 8436 | |
8ef5849f JK |
8437 | /* |
8438 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
8439 | * So, just fall through. test_pages_isolated() has a tracepoint |
8440 | * which will report the busy page. | |
8441 | * | |
8442 | * It is possible that busy pages could become available before | |
8443 | * the call to test_pages_isolated, and the range will actually be | |
8444 | * allocated. So, if we fall through be sure to clear ret so that | |
8445 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 8446 | */ |
bb13ffeb | 8447 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 8448 | if (ret && ret != -EBUSY) |
041d3a8c | 8449 | goto done; |
63cd4489 | 8450 | ret =0; |
041d3a8c MN |
8451 | |
8452 | /* | |
8453 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
8454 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
8455 | * more, all pages in [start, end) are free in page allocator. | |
8456 | * What we are going to do is to allocate all pages from | |
8457 | * [start, end) (that is remove them from page allocator). | |
8458 | * | |
8459 | * The only problem is that pages at the beginning and at the | |
8460 | * end of interesting range may be not aligned with pages that | |
8461 | * page allocator holds, ie. they can be part of higher order | |
8462 | * pages. Because of this, we reserve the bigger range and | |
8463 | * once this is done free the pages we are not interested in. | |
8464 | * | |
8465 | * We don't have to hold zone->lock here because the pages are | |
8466 | * isolated thus they won't get removed from buddy. | |
8467 | */ | |
8468 | ||
8469 | lru_add_drain_all(); | |
041d3a8c MN |
8470 | |
8471 | order = 0; | |
8472 | outer_start = start; | |
8473 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
8474 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
8475 | outer_start = start; |
8476 | break; | |
041d3a8c MN |
8477 | } |
8478 | outer_start &= ~0UL << order; | |
8479 | } | |
8480 | ||
8ef5849f JK |
8481 | if (outer_start != start) { |
8482 | order = page_order(pfn_to_page(outer_start)); | |
8483 | ||
8484 | /* | |
8485 | * outer_start page could be small order buddy page and | |
8486 | * it doesn't include start page. Adjust outer_start | |
8487 | * in this case to report failed page properly | |
8488 | * on tracepoint in test_pages_isolated() | |
8489 | */ | |
8490 | if (outer_start + (1UL << order) <= start) | |
8491 | outer_start = start; | |
8492 | } | |
8493 | ||
041d3a8c | 8494 | /* Make sure the range is really isolated. */ |
756d25be | 8495 | if (test_pages_isolated(outer_start, end, 0)) { |
75dddef3 | 8496 | pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n", |
dae803e1 | 8497 | __func__, outer_start, end); |
041d3a8c MN |
8498 | ret = -EBUSY; |
8499 | goto done; | |
8500 | } | |
8501 | ||
49f223a9 | 8502 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 8503 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
8504 | if (!outer_end) { |
8505 | ret = -EBUSY; | |
8506 | goto done; | |
8507 | } | |
8508 | ||
8509 | /* Free head and tail (if any) */ | |
8510 | if (start != outer_start) | |
8511 | free_contig_range(outer_start, start - outer_start); | |
8512 | if (end != outer_end) | |
8513 | free_contig_range(end, outer_end - end); | |
8514 | ||
8515 | done: | |
8516 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 8517 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
8518 | return ret; |
8519 | } | |
5e27a2df AK |
8520 | |
8521 | static int __alloc_contig_pages(unsigned long start_pfn, | |
8522 | unsigned long nr_pages, gfp_t gfp_mask) | |
8523 | { | |
8524 | unsigned long end_pfn = start_pfn + nr_pages; | |
8525 | ||
8526 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
8527 | gfp_mask); | |
8528 | } | |
8529 | ||
8530 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
8531 | unsigned long nr_pages) | |
8532 | { | |
8533 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
8534 | struct page *page; | |
8535 | ||
8536 | for (i = start_pfn; i < end_pfn; i++) { | |
8537 | page = pfn_to_online_page(i); | |
8538 | if (!page) | |
8539 | return false; | |
8540 | ||
8541 | if (page_zone(page) != z) | |
8542 | return false; | |
8543 | ||
8544 | if (PageReserved(page)) | |
8545 | return false; | |
8546 | ||
8547 | if (page_count(page) > 0) | |
8548 | return false; | |
8549 | ||
8550 | if (PageHuge(page)) | |
8551 | return false; | |
8552 | } | |
8553 | return true; | |
8554 | } | |
8555 | ||
8556 | static bool zone_spans_last_pfn(const struct zone *zone, | |
8557 | unsigned long start_pfn, unsigned long nr_pages) | |
8558 | { | |
8559 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
8560 | ||
8561 | return zone_spans_pfn(zone, last_pfn); | |
8562 | } | |
8563 | ||
8564 | /** | |
8565 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
8566 | * @nr_pages: Number of contiguous pages to allocate | |
8567 | * @gfp_mask: GFP mask to limit search and used during compaction | |
8568 | * @nid: Target node | |
8569 | * @nodemask: Mask for other possible nodes | |
8570 | * | |
8571 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
8572 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
8573 | * tried for allocation with alloc_contig_range(). This routine is intended | |
8574 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
8575 | * | |
8576 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
8577 | * power of two then the alignment is guaranteed to be to the given nr_pages | |
8578 | * (e.g. 1GB request would be aligned to 1GB). | |
8579 | * | |
8580 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
8581 | * __free_page() on each allocated page. | |
8582 | * | |
8583 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
8584 | */ | |
8585 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
8586 | int nid, nodemask_t *nodemask) | |
8587 | { | |
8588 | unsigned long ret, pfn, flags; | |
8589 | struct zonelist *zonelist; | |
8590 | struct zone *zone; | |
8591 | struct zoneref *z; | |
8592 | ||
8593 | zonelist = node_zonelist(nid, gfp_mask); | |
8594 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
8595 | gfp_zone(gfp_mask), nodemask) { | |
8596 | spin_lock_irqsave(&zone->lock, flags); | |
8597 | ||
8598 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
8599 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
8600 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
8601 | /* | |
8602 | * We release the zone lock here because | |
8603 | * alloc_contig_range() will also lock the zone | |
8604 | * at some point. If there's an allocation | |
8605 | * spinning on this lock, it may win the race | |
8606 | * and cause alloc_contig_range() to fail... | |
8607 | */ | |
8608 | spin_unlock_irqrestore(&zone->lock, flags); | |
8609 | ret = __alloc_contig_pages(pfn, nr_pages, | |
8610 | gfp_mask); | |
8611 | if (!ret) | |
8612 | return pfn_to_page(pfn); | |
8613 | spin_lock_irqsave(&zone->lock, flags); | |
8614 | } | |
8615 | pfn += nr_pages; | |
8616 | } | |
8617 | spin_unlock_irqrestore(&zone->lock, flags); | |
8618 | } | |
8619 | return NULL; | |
8620 | } | |
4eb0716e | 8621 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 8622 | |
4eb0716e | 8623 | void free_contig_range(unsigned long pfn, unsigned int nr_pages) |
041d3a8c | 8624 | { |
bcc2b02f MS |
8625 | unsigned int count = 0; |
8626 | ||
8627 | for (; nr_pages--; pfn++) { | |
8628 | struct page *page = pfn_to_page(pfn); | |
8629 | ||
8630 | count += page_count(page) != 1; | |
8631 | __free_page(page); | |
8632 | } | |
8633 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c | 8634 | } |
041d3a8c | 8635 | |
0a647f38 CS |
8636 | /* |
8637 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
8638 | * page high values need to be recalulated. | |
8639 | */ | |
4ed7e022 JL |
8640 | void __meminit zone_pcp_update(struct zone *zone) |
8641 | { | |
c8e251fa | 8642 | mutex_lock(&pcp_batch_high_lock); |
cb1ef534 | 8643 | __zone_pcp_update(zone); |
c8e251fa | 8644 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 | 8645 | } |
4ed7e022 | 8646 | |
340175b7 JL |
8647 | void zone_pcp_reset(struct zone *zone) |
8648 | { | |
8649 | unsigned long flags; | |
5a883813 MK |
8650 | int cpu; |
8651 | struct per_cpu_pageset *pset; | |
340175b7 JL |
8652 | |
8653 | /* avoid races with drain_pages() */ | |
8654 | local_irq_save(flags); | |
8655 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
8656 | for_each_online_cpu(cpu) { |
8657 | pset = per_cpu_ptr(zone->pageset, cpu); | |
8658 | drain_zonestat(zone, pset); | |
8659 | } | |
340175b7 JL |
8660 | free_percpu(zone->pageset); |
8661 | zone->pageset = &boot_pageset; | |
8662 | } | |
8663 | local_irq_restore(flags); | |
8664 | } | |
8665 | ||
6dcd73d7 | 8666 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 8667 | /* |
b9eb6319 JK |
8668 | * All pages in the range must be in a single zone and isolated |
8669 | * before calling this. | |
0c0e6195 | 8670 | */ |
5557c766 | 8671 | unsigned long |
0c0e6195 KH |
8672 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
8673 | { | |
8674 | struct page *page; | |
8675 | struct zone *zone; | |
0ee5f4f3 | 8676 | unsigned int order; |
0c0e6195 KH |
8677 | unsigned long pfn; |
8678 | unsigned long flags; | |
5557c766 MH |
8679 | unsigned long offlined_pages = 0; |
8680 | ||
0c0e6195 KH |
8681 | /* find the first valid pfn */ |
8682 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
8683 | if (pfn_valid(pfn)) | |
8684 | break; | |
8685 | if (pfn == end_pfn) | |
5557c766 MH |
8686 | return offlined_pages; |
8687 | ||
2d070eab | 8688 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
8689 | zone = page_zone(pfn_to_page(pfn)); |
8690 | spin_lock_irqsave(&zone->lock, flags); | |
8691 | pfn = start_pfn; | |
8692 | while (pfn < end_pfn) { | |
8693 | if (!pfn_valid(pfn)) { | |
8694 | pfn++; | |
8695 | continue; | |
8696 | } | |
8697 | page = pfn_to_page(pfn); | |
b023f468 WC |
8698 | /* |
8699 | * The HWPoisoned page may be not in buddy system, and | |
8700 | * page_count() is not 0. | |
8701 | */ | |
8702 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
8703 | pfn++; | |
5557c766 | 8704 | offlined_pages++; |
b023f468 WC |
8705 | continue; |
8706 | } | |
8707 | ||
0c0e6195 KH |
8708 | BUG_ON(page_count(page)); |
8709 | BUG_ON(!PageBuddy(page)); | |
8710 | order = page_order(page); | |
5557c766 | 8711 | offlined_pages += 1 << order; |
b03641af | 8712 | del_page_from_free_area(page, &zone->free_area[order]); |
0c0e6195 KH |
8713 | pfn += (1 << order); |
8714 | } | |
8715 | spin_unlock_irqrestore(&zone->lock, flags); | |
5557c766 MH |
8716 | |
8717 | return offlined_pages; | |
0c0e6195 KH |
8718 | } |
8719 | #endif | |
8d22ba1b | 8720 | |
8d22ba1b WF |
8721 | bool is_free_buddy_page(struct page *page) |
8722 | { | |
8723 | struct zone *zone = page_zone(page); | |
8724 | unsigned long pfn = page_to_pfn(page); | |
8725 | unsigned long flags; | |
7aeb09f9 | 8726 | unsigned int order; |
8d22ba1b WF |
8727 | |
8728 | spin_lock_irqsave(&zone->lock, flags); | |
8729 | for (order = 0; order < MAX_ORDER; order++) { | |
8730 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8731 | ||
8732 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
8733 | break; | |
8734 | } | |
8735 | spin_unlock_irqrestore(&zone->lock, flags); | |
8736 | ||
8737 | return order < MAX_ORDER; | |
8738 | } | |
d4ae9916 NH |
8739 | |
8740 | #ifdef CONFIG_MEMORY_FAILURE | |
8741 | /* | |
8742 | * Set PG_hwpoison flag if a given page is confirmed to be a free page. This | |
8743 | * test is performed under the zone lock to prevent a race against page | |
8744 | * allocation. | |
8745 | */ | |
8746 | bool set_hwpoison_free_buddy_page(struct page *page) | |
8747 | { | |
8748 | struct zone *zone = page_zone(page); | |
8749 | unsigned long pfn = page_to_pfn(page); | |
8750 | unsigned long flags; | |
8751 | unsigned int order; | |
8752 | bool hwpoisoned = false; | |
8753 | ||
8754 | spin_lock_irqsave(&zone->lock, flags); | |
8755 | for (order = 0; order < MAX_ORDER; order++) { | |
8756 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8757 | ||
8758 | if (PageBuddy(page_head) && page_order(page_head) >= order) { | |
8759 | if (!TestSetPageHWPoison(page)) | |
8760 | hwpoisoned = true; | |
8761 | break; | |
8762 | } | |
8763 | } | |
8764 | spin_unlock_irqrestore(&zone->lock, flags); | |
8765 | ||
8766 | return hwpoisoned; | |
8767 | } | |
8768 | #endif |