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> |
e4443149 | 71 | #include <linux/padata.h> |
1da177e4 | 72 | |
7ee3d4e8 | 73 | #include <asm/sections.h> |
1da177e4 | 74 | #include <asm/tlbflush.h> |
ac924c60 | 75 | #include <asm/div64.h> |
1da177e4 | 76 | #include "internal.h" |
e900a918 | 77 | #include "shuffle.h" |
36e66c55 | 78 | #include "page_reporting.h" |
1da177e4 | 79 | |
c8e251fa CS |
80 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
81 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 82 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 83 | |
72812019 LS |
84 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
85 | DEFINE_PER_CPU(int, numa_node); | |
86 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
87 | #endif | |
88 | ||
4518085e KW |
89 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
90 | ||
7aac7898 LS |
91 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
92 | /* | |
93 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
94 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
95 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
96 | * defined in <linux/topology.h>. | |
97 | */ | |
98 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
99 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
100 | #endif | |
101 | ||
bd233f53 | 102 | /* work_structs for global per-cpu drains */ |
d9367bd0 WY |
103 | struct pcpu_drain { |
104 | struct zone *zone; | |
105 | struct work_struct work; | |
106 | }; | |
8b885f53 JY |
107 | static DEFINE_MUTEX(pcpu_drain_mutex); |
108 | static DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain); | |
bd233f53 | 109 | |
38addce8 | 110 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 111 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
112 | EXPORT_SYMBOL(latent_entropy); |
113 | #endif | |
114 | ||
1da177e4 | 115 | /* |
13808910 | 116 | * Array of node states. |
1da177e4 | 117 | */ |
13808910 CL |
118 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
119 | [N_POSSIBLE] = NODE_MASK_ALL, | |
120 | [N_ONLINE] = { { [0] = 1UL } }, | |
121 | #ifndef CONFIG_NUMA | |
122 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
123 | #ifdef CONFIG_HIGHMEM | |
124 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 125 | #endif |
20b2f52b | 126 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
127 | [N_CPU] = { { [0] = 1UL } }, |
128 | #endif /* NUMA */ | |
129 | }; | |
130 | EXPORT_SYMBOL(node_states); | |
131 | ||
ca79b0c2 AK |
132 | atomic_long_t _totalram_pages __read_mostly; |
133 | EXPORT_SYMBOL(_totalram_pages); | |
cb45b0e9 | 134 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 135 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 136 | |
1b76b02f | 137 | int percpu_pagelist_fraction; |
dcce284a | 138 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
6471384a AP |
139 | #ifdef CONFIG_INIT_ON_ALLOC_DEFAULT_ON |
140 | DEFINE_STATIC_KEY_TRUE(init_on_alloc); | |
141 | #else | |
142 | DEFINE_STATIC_KEY_FALSE(init_on_alloc); | |
143 | #endif | |
144 | EXPORT_SYMBOL(init_on_alloc); | |
145 | ||
146 | #ifdef CONFIG_INIT_ON_FREE_DEFAULT_ON | |
147 | DEFINE_STATIC_KEY_TRUE(init_on_free); | |
148 | #else | |
149 | DEFINE_STATIC_KEY_FALSE(init_on_free); | |
150 | #endif | |
151 | EXPORT_SYMBOL(init_on_free); | |
152 | ||
153 | static int __init early_init_on_alloc(char *buf) | |
154 | { | |
155 | int ret; | |
156 | bool bool_result; | |
157 | ||
158 | if (!buf) | |
159 | return -EINVAL; | |
160 | ret = kstrtobool(buf, &bool_result); | |
161 | if (bool_result && page_poisoning_enabled()) | |
162 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_alloc\n"); | |
163 | if (bool_result) | |
164 | static_branch_enable(&init_on_alloc); | |
165 | else | |
166 | static_branch_disable(&init_on_alloc); | |
167 | return ret; | |
168 | } | |
169 | early_param("init_on_alloc", early_init_on_alloc); | |
170 | ||
171 | static int __init early_init_on_free(char *buf) | |
172 | { | |
173 | int ret; | |
174 | bool bool_result; | |
175 | ||
176 | if (!buf) | |
177 | return -EINVAL; | |
178 | ret = kstrtobool(buf, &bool_result); | |
179 | if (bool_result && page_poisoning_enabled()) | |
180 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_free\n"); | |
181 | if (bool_result) | |
182 | static_branch_enable(&init_on_free); | |
183 | else | |
184 | static_branch_disable(&init_on_free); | |
185 | return ret; | |
186 | } | |
187 | early_param("init_on_free", early_init_on_free); | |
1da177e4 | 188 | |
bb14c2c7 VB |
189 | /* |
190 | * A cached value of the page's pageblock's migratetype, used when the page is | |
191 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
192 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
193 | * Also the migratetype set in the page does not necessarily match the pcplist | |
194 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
195 | * other index - this ensures that it will be put on the correct CMA freelist. | |
196 | */ | |
197 | static inline int get_pcppage_migratetype(struct page *page) | |
198 | { | |
199 | return page->index; | |
200 | } | |
201 | ||
202 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
203 | { | |
204 | page->index = migratetype; | |
205 | } | |
206 | ||
452aa699 RW |
207 | #ifdef CONFIG_PM_SLEEP |
208 | /* | |
209 | * The following functions are used by the suspend/hibernate code to temporarily | |
210 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
211 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
212 | * they should always be called with system_transition_mutex held |
213 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
214 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
215 | * with that modification). | |
452aa699 | 216 | */ |
c9e664f1 RW |
217 | |
218 | static gfp_t saved_gfp_mask; | |
219 | ||
220 | void pm_restore_gfp_mask(void) | |
452aa699 | 221 | { |
55f2503c | 222 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
223 | if (saved_gfp_mask) { |
224 | gfp_allowed_mask = saved_gfp_mask; | |
225 | saved_gfp_mask = 0; | |
226 | } | |
452aa699 RW |
227 | } |
228 | ||
c9e664f1 | 229 | void pm_restrict_gfp_mask(void) |
452aa699 | 230 | { |
55f2503c | 231 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
232 | WARN_ON(saved_gfp_mask); |
233 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 234 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 235 | } |
f90ac398 MG |
236 | |
237 | bool pm_suspended_storage(void) | |
238 | { | |
d0164adc | 239 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
240 | return false; |
241 | return true; | |
242 | } | |
452aa699 RW |
243 | #endif /* CONFIG_PM_SLEEP */ |
244 | ||
d9c23400 | 245 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 246 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
247 | #endif |
248 | ||
d98c7a09 | 249 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 250 | |
1da177e4 LT |
251 | /* |
252 | * results with 256, 32 in the lowmem_reserve sysctl: | |
253 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
254 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
255 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
256 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 257 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
258 | * |
259 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
260 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 261 | */ |
d3cda233 | 262 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 263 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 264 | [ZONE_DMA] = 256, |
4b51d669 | 265 | #endif |
fb0e7942 | 266 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 267 | [ZONE_DMA32] = 256, |
fb0e7942 | 268 | #endif |
d3cda233 | 269 | [ZONE_NORMAL] = 32, |
e53ef38d | 270 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 271 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 272 | #endif |
d3cda233 | 273 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 274 | }; |
1da177e4 | 275 | |
15ad7cdc | 276 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 277 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 278 | "DMA", |
4b51d669 | 279 | #endif |
fb0e7942 | 280 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 281 | "DMA32", |
fb0e7942 | 282 | #endif |
2f1b6248 | 283 | "Normal", |
e53ef38d | 284 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 285 | "HighMem", |
e53ef38d | 286 | #endif |
2a1e274a | 287 | "Movable", |
033fbae9 DW |
288 | #ifdef CONFIG_ZONE_DEVICE |
289 | "Device", | |
290 | #endif | |
2f1b6248 CL |
291 | }; |
292 | ||
c999fbd3 | 293 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
294 | "Unmovable", |
295 | "Movable", | |
296 | "Reclaimable", | |
297 | "HighAtomic", | |
298 | #ifdef CONFIG_CMA | |
299 | "CMA", | |
300 | #endif | |
301 | #ifdef CONFIG_MEMORY_ISOLATION | |
302 | "Isolate", | |
303 | #endif | |
304 | }; | |
305 | ||
ae70eddd AK |
306 | compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = { |
307 | [NULL_COMPOUND_DTOR] = NULL, | |
308 | [COMPOUND_PAGE_DTOR] = free_compound_page, | |
f1e61557 | 309 | #ifdef CONFIG_HUGETLB_PAGE |
ae70eddd | 310 | [HUGETLB_PAGE_DTOR] = free_huge_page, |
f1e61557 | 311 | #endif |
9a982250 | 312 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
ae70eddd | 313 | [TRANSHUGE_PAGE_DTOR] = free_transhuge_page, |
9a982250 | 314 | #endif |
f1e61557 KS |
315 | }; |
316 | ||
1da177e4 | 317 | int min_free_kbytes = 1024; |
42aa83cb | 318 | int user_min_free_kbytes = -1; |
24512228 MG |
319 | #ifdef CONFIG_DISCONTIGMEM |
320 | /* | |
321 | * DiscontigMem defines memory ranges as separate pg_data_t even if the ranges | |
322 | * are not on separate NUMA nodes. Functionally this works but with | |
323 | * watermark_boost_factor, it can reclaim prematurely as the ranges can be | |
324 | * quite small. By default, do not boost watermarks on discontigmem as in | |
325 | * many cases very high-order allocations like THP are likely to be | |
326 | * unsupported and the premature reclaim offsets the advantage of long-term | |
327 | * fragmentation avoidance. | |
328 | */ | |
329 | int watermark_boost_factor __read_mostly; | |
330 | #else | |
1c30844d | 331 | int watermark_boost_factor __read_mostly = 15000; |
24512228 | 332 | #endif |
795ae7a0 | 333 | int watermark_scale_factor = 10; |
1da177e4 | 334 | |
bbe5d993 OS |
335 | static unsigned long nr_kernel_pages __initdata; |
336 | static unsigned long nr_all_pages __initdata; | |
337 | static unsigned long dma_reserve __initdata; | |
1da177e4 | 338 | |
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); | |
c713216d | 351 | |
418508c1 | 352 | #if MAX_NUMNODES > 1 |
b9726c26 | 353 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 354 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 355 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 356 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
357 | #endif |
358 | ||
9ef9acb0 MG |
359 | int page_group_by_mobility_disabled __read_mostly; |
360 | ||
3a80a7fa | 361 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
362 | /* |
363 | * During boot we initialize deferred pages on-demand, as needed, but once | |
364 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
365 | * and we can permanently disable that path. | |
366 | */ | |
367 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
368 | ||
369 | /* | |
370 | * Calling kasan_free_pages() only after deferred memory initialization | |
371 | * has completed. Poisoning pages during deferred memory init will greatly | |
372 | * lengthen the process and cause problem in large memory systems as the | |
373 | * deferred pages initialization is done with interrupt disabled. | |
374 | * | |
375 | * Assuming that there will be no reference to those newly initialized | |
376 | * pages before they are ever allocated, this should have no effect on | |
377 | * KASAN memory tracking as the poison will be properly inserted at page | |
378 | * allocation time. The only corner case is when pages are allocated by | |
379 | * on-demand allocation and then freed again before the deferred pages | |
380 | * initialization is done, but this is not likely to happen. | |
381 | */ | |
382 | static inline void kasan_free_nondeferred_pages(struct page *page, int order) | |
383 | { | |
384 | if (!static_branch_unlikely(&deferred_pages)) | |
385 | kasan_free_pages(page, order); | |
386 | } | |
387 | ||
3a80a7fa | 388 | /* Returns true if the struct page for the pfn is uninitialised */ |
0e1cc95b | 389 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 390 | { |
ef70b6f4 MG |
391 | int nid = early_pfn_to_nid(pfn); |
392 | ||
393 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
394 | return true; |
395 | ||
396 | return false; | |
397 | } | |
398 | ||
399 | /* | |
d3035be4 | 400 | * Returns true when the remaining initialisation should be deferred until |
3a80a7fa MG |
401 | * later in the boot cycle when it can be parallelised. |
402 | */ | |
d3035be4 PT |
403 | static bool __meminit |
404 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn) | |
3a80a7fa | 405 | { |
d3035be4 PT |
406 | static unsigned long prev_end_pfn, nr_initialised; |
407 | ||
408 | /* | |
409 | * prev_end_pfn static that contains the end of previous zone | |
410 | * No need to protect because called very early in boot before smp_init. | |
411 | */ | |
412 | if (prev_end_pfn != end_pfn) { | |
413 | prev_end_pfn = end_pfn; | |
414 | nr_initialised = 0; | |
415 | } | |
416 | ||
3c2c6488 | 417 | /* Always populate low zones for address-constrained allocations */ |
d3035be4 | 418 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid))) |
3a80a7fa | 419 | return false; |
23b68cfa WY |
420 | |
421 | /* | |
422 | * We start only with one section of pages, more pages are added as | |
423 | * needed until the rest of deferred pages are initialized. | |
424 | */ | |
d3035be4 | 425 | nr_initialised++; |
23b68cfa | 426 | if ((nr_initialised > PAGES_PER_SECTION) && |
d3035be4 PT |
427 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
428 | NODE_DATA(nid)->first_deferred_pfn = pfn; | |
429 | return true; | |
3a80a7fa | 430 | } |
d3035be4 | 431 | return false; |
3a80a7fa MG |
432 | } |
433 | #else | |
3c0c12cc WL |
434 | #define kasan_free_nondeferred_pages(p, o) kasan_free_pages(p, o) |
435 | ||
3a80a7fa MG |
436 | static inline bool early_page_uninitialised(unsigned long pfn) |
437 | { | |
438 | return false; | |
439 | } | |
440 | ||
d3035be4 | 441 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn) |
3a80a7fa | 442 | { |
d3035be4 | 443 | return false; |
3a80a7fa MG |
444 | } |
445 | #endif | |
446 | ||
0b423ca2 MG |
447 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
448 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
449 | unsigned long pfn) | |
450 | { | |
451 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 452 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
453 | #else |
454 | return page_zone(page)->pageblock_flags; | |
455 | #endif /* CONFIG_SPARSEMEM */ | |
456 | } | |
457 | ||
458 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
459 | { | |
460 | #ifdef CONFIG_SPARSEMEM | |
461 | pfn &= (PAGES_PER_SECTION-1); | |
0b423ca2 MG |
462 | #else |
463 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
0b423ca2 | 464 | #endif /* CONFIG_SPARSEMEM */ |
399b795b | 465 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
0b423ca2 MG |
466 | } |
467 | ||
468 | /** | |
469 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
470 | * @page: The page within the block of interest | |
471 | * @pfn: The target page frame number | |
472 | * @end_bitidx: The last bit of interest to retrieve | |
473 | * @mask: mask of bits that the caller is interested in | |
474 | * | |
475 | * Return: pageblock_bits flags | |
476 | */ | |
477 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
478 | unsigned long pfn, | |
479 | unsigned long end_bitidx, | |
480 | unsigned long mask) | |
481 | { | |
482 | unsigned long *bitmap; | |
483 | unsigned long bitidx, word_bitidx; | |
484 | unsigned long word; | |
485 | ||
486 | bitmap = get_pageblock_bitmap(page, pfn); | |
487 | bitidx = pfn_to_bitidx(page, pfn); | |
488 | word_bitidx = bitidx / BITS_PER_LONG; | |
489 | bitidx &= (BITS_PER_LONG-1); | |
490 | ||
491 | word = bitmap[word_bitidx]; | |
492 | bitidx += end_bitidx; | |
493 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
494 | } | |
495 | ||
496 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
497 | unsigned long end_bitidx, | |
498 | unsigned long mask) | |
499 | { | |
500 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
501 | } | |
502 | ||
503 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
504 | { | |
505 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
506 | } | |
507 | ||
508 | /** | |
509 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
510 | * @page: The page within the block of interest | |
511 | * @flags: The flags to set | |
512 | * @pfn: The target page frame number | |
513 | * @end_bitidx: The last bit of interest | |
514 | * @mask: mask of bits that the caller is interested in | |
515 | */ | |
516 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
517 | unsigned long pfn, | |
518 | unsigned long end_bitidx, | |
519 | unsigned long mask) | |
520 | { | |
521 | unsigned long *bitmap; | |
522 | unsigned long bitidx, word_bitidx; | |
523 | unsigned long old_word, word; | |
524 | ||
525 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 526 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
527 | |
528 | bitmap = get_pageblock_bitmap(page, pfn); | |
529 | bitidx = pfn_to_bitidx(page, pfn); | |
530 | word_bitidx = bitidx / BITS_PER_LONG; | |
531 | bitidx &= (BITS_PER_LONG-1); | |
532 | ||
533 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
534 | ||
535 | bitidx += end_bitidx; | |
536 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
537 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
538 | ||
539 | word = READ_ONCE(bitmap[word_bitidx]); | |
540 | for (;;) { | |
541 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
542 | if (word == old_word) | |
543 | break; | |
544 | word = old_word; | |
545 | } | |
546 | } | |
3a80a7fa | 547 | |
ee6f509c | 548 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 549 | { |
5d0f3f72 KM |
550 | if (unlikely(page_group_by_mobility_disabled && |
551 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
552 | migratetype = MIGRATE_UNMOVABLE; |
553 | ||
b2a0ac88 MG |
554 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
555 | PB_migrate, PB_migrate_end); | |
556 | } | |
557 | ||
13e7444b | 558 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 559 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 560 | { |
bdc8cb98 DH |
561 | int ret = 0; |
562 | unsigned seq; | |
563 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 564 | unsigned long sp, start_pfn; |
c6a57e19 | 565 | |
bdc8cb98 DH |
566 | do { |
567 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
568 | start_pfn = zone->zone_start_pfn; |
569 | sp = zone->spanned_pages; | |
108bcc96 | 570 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
571 | ret = 1; |
572 | } while (zone_span_seqretry(zone, seq)); | |
573 | ||
b5e6a5a2 | 574 | if (ret) |
613813e8 DH |
575 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
576 | pfn, zone_to_nid(zone), zone->name, | |
577 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 578 | |
bdc8cb98 | 579 | return ret; |
c6a57e19 DH |
580 | } |
581 | ||
582 | static int page_is_consistent(struct zone *zone, struct page *page) | |
583 | { | |
14e07298 | 584 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 585 | return 0; |
1da177e4 | 586 | if (zone != page_zone(page)) |
c6a57e19 DH |
587 | return 0; |
588 | ||
589 | return 1; | |
590 | } | |
591 | /* | |
592 | * Temporary debugging check for pages not lying within a given zone. | |
593 | */ | |
d73d3c9f | 594 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
595 | { |
596 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 597 | return 1; |
c6a57e19 DH |
598 | if (!page_is_consistent(zone, page)) |
599 | return 1; | |
600 | ||
1da177e4 LT |
601 | return 0; |
602 | } | |
13e7444b | 603 | #else |
d73d3c9f | 604 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
605 | { |
606 | return 0; | |
607 | } | |
608 | #endif | |
609 | ||
82a3241a | 610 | static void bad_page(struct page *page, const char *reason) |
1da177e4 | 611 | { |
d936cf9b HD |
612 | static unsigned long resume; |
613 | static unsigned long nr_shown; | |
614 | static unsigned long nr_unshown; | |
615 | ||
616 | /* | |
617 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
618 | * or allow a steady drip of one report per second. | |
619 | */ | |
620 | if (nr_shown == 60) { | |
621 | if (time_before(jiffies, resume)) { | |
622 | nr_unshown++; | |
623 | goto out; | |
624 | } | |
625 | if (nr_unshown) { | |
ff8e8116 | 626 | pr_alert( |
1e9e6365 | 627 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
628 | nr_unshown); |
629 | nr_unshown = 0; | |
630 | } | |
631 | nr_shown = 0; | |
632 | } | |
633 | if (nr_shown++ == 0) | |
634 | resume = jiffies + 60 * HZ; | |
635 | ||
ff8e8116 | 636 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 637 | current->comm, page_to_pfn(page)); |
ff8e8116 | 638 | __dump_page(page, reason); |
4e462112 | 639 | dump_page_owner(page); |
3dc14741 | 640 | |
4f31888c | 641 | print_modules(); |
1da177e4 | 642 | dump_stack(); |
d936cf9b | 643 | out: |
8cc3b392 | 644 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 645 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 646 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
647 | } |
648 | ||
1da177e4 LT |
649 | /* |
650 | * Higher-order pages are called "compound pages". They are structured thusly: | |
651 | * | |
1d798ca3 | 652 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 653 | * |
1d798ca3 KS |
654 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
655 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 656 | * |
1d798ca3 KS |
657 | * The first tail page's ->compound_dtor holds the offset in array of compound |
658 | * page destructors. See compound_page_dtors. | |
1da177e4 | 659 | * |
1d798ca3 | 660 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 661 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 662 | */ |
d98c7a09 | 663 | |
9a982250 | 664 | void free_compound_page(struct page *page) |
d98c7a09 | 665 | { |
7ae88534 | 666 | mem_cgroup_uncharge(page); |
d85f3385 | 667 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
668 | } |
669 | ||
d00181b9 | 670 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
671 | { |
672 | int i; | |
673 | int nr_pages = 1 << order; | |
674 | ||
f1e61557 | 675 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
676 | set_compound_order(page, order); |
677 | __SetPageHead(page); | |
678 | for (i = 1; i < nr_pages; i++) { | |
679 | struct page *p = page + i; | |
58a84aa9 | 680 | set_page_count(p, 0); |
1c290f64 | 681 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 682 | set_compound_head(p, page); |
18229df5 | 683 | } |
53f9263b | 684 | atomic_set(compound_mapcount_ptr(page), -1); |
47e29d32 JH |
685 | if (hpage_pincount_available(page)) |
686 | atomic_set(compound_pincount_ptr(page), 0); | |
18229df5 AW |
687 | } |
688 | ||
c0a32fc5 SG |
689 | #ifdef CONFIG_DEBUG_PAGEALLOC |
690 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 691 | |
8e57f8ac VB |
692 | bool _debug_pagealloc_enabled_early __read_mostly |
693 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
694 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 695 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 696 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
697 | |
698 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 699 | |
031bc574 JK |
700 | static int __init early_debug_pagealloc(char *buf) |
701 | { | |
8e57f8ac | 702 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
703 | } |
704 | early_param("debug_pagealloc", early_debug_pagealloc); | |
705 | ||
8e57f8ac | 706 | void init_debug_pagealloc(void) |
e30825f1 | 707 | { |
031bc574 JK |
708 | if (!debug_pagealloc_enabled()) |
709 | return; | |
710 | ||
8e57f8ac VB |
711 | static_branch_enable(&_debug_pagealloc_enabled); |
712 | ||
f1c1e9f7 JK |
713 | if (!debug_guardpage_minorder()) |
714 | return; | |
715 | ||
96a2b03f | 716 | static_branch_enable(&_debug_guardpage_enabled); |
e30825f1 JK |
717 | } |
718 | ||
c0a32fc5 SG |
719 | static int __init debug_guardpage_minorder_setup(char *buf) |
720 | { | |
721 | unsigned long res; | |
722 | ||
723 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 724 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
725 | return 0; |
726 | } | |
727 | _debug_guardpage_minorder = res; | |
1170532b | 728 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
729 | return 0; |
730 | } | |
f1c1e9f7 | 731 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 732 | |
acbc15a4 | 733 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 734 | unsigned int order, int migratetype) |
c0a32fc5 | 735 | { |
e30825f1 | 736 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
737 | return false; |
738 | ||
739 | if (order >= debug_guardpage_minorder()) | |
740 | return false; | |
e30825f1 | 741 | |
3972f6bb | 742 | __SetPageGuard(page); |
2847cf95 JK |
743 | INIT_LIST_HEAD(&page->lru); |
744 | set_page_private(page, order); | |
745 | /* Guard pages are not available for any usage */ | |
746 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
747 | |
748 | return true; | |
c0a32fc5 SG |
749 | } |
750 | ||
2847cf95 JK |
751 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
752 | unsigned int order, int migratetype) | |
c0a32fc5 | 753 | { |
e30825f1 JK |
754 | if (!debug_guardpage_enabled()) |
755 | return; | |
756 | ||
3972f6bb | 757 | __ClearPageGuard(page); |
e30825f1 | 758 | |
2847cf95 JK |
759 | set_page_private(page, 0); |
760 | if (!is_migrate_isolate(migratetype)) | |
761 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
762 | } |
763 | #else | |
acbc15a4 JK |
764 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
765 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
766 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
767 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
768 | #endif |
769 | ||
7aeb09f9 | 770 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 771 | { |
4c21e2f2 | 772 | set_page_private(page, order); |
676165a8 | 773 | __SetPageBuddy(page); |
1da177e4 LT |
774 | } |
775 | ||
1da177e4 LT |
776 | /* |
777 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 778 | * we can coalesce a page and its buddy if |
13ad59df | 779 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 780 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
781 | * (c) a page and its buddy have the same order && |
782 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 783 | * |
6e292b9b MW |
784 | * For recording whether a page is in the buddy system, we set PageBuddy. |
785 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 786 | * |
676165a8 | 787 | * For recording page's order, we use page_private(page). |
1da177e4 | 788 | */ |
fe925c0c | 789 | static inline bool page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 790 | unsigned int order) |
1da177e4 | 791 | { |
fe925c0c | 792 | if (!page_is_guard(buddy) && !PageBuddy(buddy)) |
793 | return false; | |
4c5018ce | 794 | |
fe925c0c | 795 | if (page_order(buddy) != order) |
796 | return false; | |
c0a32fc5 | 797 | |
fe925c0c | 798 | /* |
799 | * zone check is done late to avoid uselessly calculating | |
800 | * zone/node ids for pages that could never merge. | |
801 | */ | |
802 | if (page_zone_id(page) != page_zone_id(buddy)) | |
803 | return false; | |
d34c5fa0 | 804 | |
fe925c0c | 805 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
4c5018ce | 806 | |
fe925c0c | 807 | return true; |
1da177e4 LT |
808 | } |
809 | ||
5e1f0f09 MG |
810 | #ifdef CONFIG_COMPACTION |
811 | static inline struct capture_control *task_capc(struct zone *zone) | |
812 | { | |
813 | struct capture_control *capc = current->capture_control; | |
814 | ||
deba0487 | 815 | return unlikely(capc) && |
5e1f0f09 MG |
816 | !(current->flags & PF_KTHREAD) && |
817 | !capc->page && | |
deba0487 | 818 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
819 | } |
820 | ||
821 | static inline bool | |
822 | compaction_capture(struct capture_control *capc, struct page *page, | |
823 | int order, int migratetype) | |
824 | { | |
825 | if (!capc || order != capc->cc->order) | |
826 | return false; | |
827 | ||
828 | /* Do not accidentally pollute CMA or isolated regions*/ | |
829 | if (is_migrate_cma(migratetype) || | |
830 | is_migrate_isolate(migratetype)) | |
831 | return false; | |
832 | ||
833 | /* | |
834 | * Do not let lower order allocations polluate a movable pageblock. | |
835 | * This might let an unmovable request use a reclaimable pageblock | |
836 | * and vice-versa but no more than normal fallback logic which can | |
837 | * have trouble finding a high-order free page. | |
838 | */ | |
839 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
840 | return false; | |
841 | ||
842 | capc->page = page; | |
843 | return true; | |
844 | } | |
845 | ||
846 | #else | |
847 | static inline struct capture_control *task_capc(struct zone *zone) | |
848 | { | |
849 | return NULL; | |
850 | } | |
851 | ||
852 | static inline bool | |
853 | compaction_capture(struct capture_control *capc, struct page *page, | |
854 | int order, int migratetype) | |
855 | { | |
856 | return false; | |
857 | } | |
858 | #endif /* CONFIG_COMPACTION */ | |
859 | ||
6ab01363 AD |
860 | /* Used for pages not on another list */ |
861 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
862 | unsigned int order, int migratetype) | |
863 | { | |
864 | struct free_area *area = &zone->free_area[order]; | |
865 | ||
866 | list_add(&page->lru, &area->free_list[migratetype]); | |
867 | area->nr_free++; | |
868 | } | |
869 | ||
870 | /* Used for pages not on another list */ | |
871 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
872 | unsigned int order, int migratetype) | |
873 | { | |
874 | struct free_area *area = &zone->free_area[order]; | |
875 | ||
876 | list_add_tail(&page->lru, &area->free_list[migratetype]); | |
877 | area->nr_free++; | |
878 | } | |
879 | ||
880 | /* Used for pages which are on another list */ | |
881 | static inline void move_to_free_list(struct page *page, struct zone *zone, | |
882 | unsigned int order, int migratetype) | |
883 | { | |
884 | struct free_area *area = &zone->free_area[order]; | |
885 | ||
886 | list_move(&page->lru, &area->free_list[migratetype]); | |
887 | } | |
888 | ||
889 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
890 | unsigned int order) | |
891 | { | |
36e66c55 AD |
892 | /* clear reported state and update reported page count */ |
893 | if (page_reported(page)) | |
894 | __ClearPageReported(page); | |
895 | ||
6ab01363 AD |
896 | list_del(&page->lru); |
897 | __ClearPageBuddy(page); | |
898 | set_page_private(page, 0); | |
899 | zone->free_area[order].nr_free--; | |
900 | } | |
901 | ||
a2129f24 AD |
902 | /* |
903 | * If this is not the largest possible page, check if the buddy | |
904 | * of the next-highest order is free. If it is, it's possible | |
905 | * that pages are being freed that will coalesce soon. In case, | |
906 | * that is happening, add the free page to the tail of the list | |
907 | * so it's less likely to be used soon and more likely to be merged | |
908 | * as a higher order page | |
909 | */ | |
910 | static inline bool | |
911 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
912 | struct page *page, unsigned int order) | |
913 | { | |
914 | struct page *higher_page, *higher_buddy; | |
915 | unsigned long combined_pfn; | |
916 | ||
917 | if (order >= MAX_ORDER - 2) | |
918 | return false; | |
919 | ||
920 | if (!pfn_valid_within(buddy_pfn)) | |
921 | return false; | |
922 | ||
923 | combined_pfn = buddy_pfn & pfn; | |
924 | higher_page = page + (combined_pfn - pfn); | |
925 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
926 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
927 | ||
928 | return pfn_valid_within(buddy_pfn) && | |
929 | page_is_buddy(higher_page, higher_buddy, order + 1); | |
930 | } | |
931 | ||
1da177e4 LT |
932 | /* |
933 | * Freeing function for a buddy system allocator. | |
934 | * | |
935 | * The concept of a buddy system is to maintain direct-mapped table | |
936 | * (containing bit values) for memory blocks of various "orders". | |
937 | * The bottom level table contains the map for the smallest allocatable | |
938 | * units of memory (here, pages), and each level above it describes | |
939 | * pairs of units from the levels below, hence, "buddies". | |
940 | * At a high level, all that happens here is marking the table entry | |
941 | * at the bottom level available, and propagating the changes upward | |
942 | * as necessary, plus some accounting needed to play nicely with other | |
943 | * parts of the VM system. | |
944 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
945 | * free pages of length of (1 << order) and marked with PageBuddy. |
946 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 947 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
948 | * other. That is, if we allocate a small block, and both were |
949 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 950 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 951 | * triggers coalescing into a block of larger size. |
1da177e4 | 952 | * |
6d49e352 | 953 | * -- nyc |
1da177e4 LT |
954 | */ |
955 | ||
48db57f8 | 956 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 957 | unsigned long pfn, |
ed0ae21d | 958 | struct zone *zone, unsigned int order, |
36e66c55 | 959 | int migratetype, bool report) |
1da177e4 | 960 | { |
a2129f24 | 961 | struct capture_control *capc = task_capc(zone); |
3f649ab7 | 962 | unsigned long buddy_pfn; |
a2129f24 | 963 | unsigned long combined_pfn; |
d9dddbf5 | 964 | unsigned int max_order; |
a2129f24 AD |
965 | struct page *buddy; |
966 | bool to_tail; | |
d9dddbf5 VB |
967 | |
968 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 969 | |
d29bb978 | 970 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 971 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 972 | |
ed0ae21d | 973 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 974 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 975 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 976 | |
76741e77 | 977 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 978 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 979 | |
d9dddbf5 | 980 | continue_merging: |
3c605096 | 981 | while (order < max_order - 1) { |
5e1f0f09 MG |
982 | if (compaction_capture(capc, page, order, migratetype)) { |
983 | __mod_zone_freepage_state(zone, -(1 << order), | |
984 | migratetype); | |
985 | return; | |
986 | } | |
76741e77 VB |
987 | buddy_pfn = __find_buddy_pfn(pfn, order); |
988 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
989 | |
990 | if (!pfn_valid_within(buddy_pfn)) | |
991 | goto done_merging; | |
cb2b95e1 | 992 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 993 | goto done_merging; |
c0a32fc5 SG |
994 | /* |
995 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
996 | * merge with it and move up one order. | |
997 | */ | |
b03641af | 998 | if (page_is_guard(buddy)) |
2847cf95 | 999 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 1000 | else |
6ab01363 | 1001 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
1002 | combined_pfn = buddy_pfn & pfn; |
1003 | page = page + (combined_pfn - pfn); | |
1004 | pfn = combined_pfn; | |
1da177e4 LT |
1005 | order++; |
1006 | } | |
d9dddbf5 VB |
1007 | if (max_order < MAX_ORDER) { |
1008 | /* If we are here, it means order is >= pageblock_order. | |
1009 | * We want to prevent merge between freepages on isolate | |
1010 | * pageblock and normal pageblock. Without this, pageblock | |
1011 | * isolation could cause incorrect freepage or CMA accounting. | |
1012 | * | |
1013 | * We don't want to hit this code for the more frequent | |
1014 | * low-order merging. | |
1015 | */ | |
1016 | if (unlikely(has_isolate_pageblock(zone))) { | |
1017 | int buddy_mt; | |
1018 | ||
76741e77 VB |
1019 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1020 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
1021 | buddy_mt = get_pageblock_migratetype(buddy); |
1022 | ||
1023 | if (migratetype != buddy_mt | |
1024 | && (is_migrate_isolate(migratetype) || | |
1025 | is_migrate_isolate(buddy_mt))) | |
1026 | goto done_merging; | |
1027 | } | |
1028 | max_order++; | |
1029 | goto continue_merging; | |
1030 | } | |
1031 | ||
1032 | done_merging: | |
1da177e4 | 1033 | set_page_order(page, order); |
6dda9d55 | 1034 | |
97500a4a | 1035 | if (is_shuffle_order(order)) |
a2129f24 | 1036 | to_tail = shuffle_pick_tail(); |
97500a4a | 1037 | else |
a2129f24 | 1038 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 1039 | |
a2129f24 | 1040 | if (to_tail) |
6ab01363 | 1041 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 1042 | else |
6ab01363 | 1043 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
1044 | |
1045 | /* Notify page reporting subsystem of freed page */ | |
1046 | if (report) | |
1047 | page_reporting_notify_free(order); | |
1da177e4 LT |
1048 | } |
1049 | ||
7bfec6f4 MG |
1050 | /* |
1051 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1052 | * try and check multiple fields with one check. The caller must do a detailed | |
1053 | * check if necessary. | |
1054 | */ | |
1055 | static inline bool page_expected_state(struct page *page, | |
1056 | unsigned long check_flags) | |
1057 | { | |
1058 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1059 | return false; | |
1060 | ||
1061 | if (unlikely((unsigned long)page->mapping | | |
1062 | page_ref_count(page) | | |
1063 | #ifdef CONFIG_MEMCG | |
1064 | (unsigned long)page->mem_cgroup | | |
1065 | #endif | |
1066 | (page->flags & check_flags))) | |
1067 | return false; | |
1068 | ||
1069 | return true; | |
1070 | } | |
1071 | ||
58b7f119 | 1072 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 1073 | { |
82a3241a | 1074 | const char *bad_reason = NULL; |
f0b791a3 | 1075 | |
53f9263b | 1076 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1077 | bad_reason = "nonzero mapcount"; |
1078 | if (unlikely(page->mapping != NULL)) | |
1079 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1080 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1081 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
1082 | if (unlikely(page->flags & flags)) { |
1083 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
1084 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
1085 | else | |
1086 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 1087 | } |
9edad6ea JW |
1088 | #ifdef CONFIG_MEMCG |
1089 | if (unlikely(page->mem_cgroup)) | |
1090 | bad_reason = "page still charged to cgroup"; | |
1091 | #endif | |
58b7f119 WY |
1092 | return bad_reason; |
1093 | } | |
1094 | ||
1095 | static void check_free_page_bad(struct page *page) | |
1096 | { | |
1097 | bad_page(page, | |
1098 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
1099 | } |
1100 | ||
534fe5e3 | 1101 | static inline int check_free_page(struct page *page) |
bb552ac6 | 1102 | { |
da838d4f | 1103 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 1104 | return 0; |
bb552ac6 MG |
1105 | |
1106 | /* Something has gone sideways, find it */ | |
0d0c48a2 | 1107 | check_free_page_bad(page); |
7bfec6f4 | 1108 | return 1; |
1da177e4 LT |
1109 | } |
1110 | ||
4db7548c MG |
1111 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1112 | { | |
1113 | int ret = 1; | |
1114 | ||
1115 | /* | |
1116 | * We rely page->lru.next never has bit 0 set, unless the page | |
1117 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1118 | */ | |
1119 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1120 | ||
1121 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1122 | ret = 0; | |
1123 | goto out; | |
1124 | } | |
1125 | switch (page - head_page) { | |
1126 | case 1: | |
4da1984e | 1127 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c | 1128 | if (unlikely(compound_mapcount(page))) { |
82a3241a | 1129 | bad_page(page, "nonzero compound_mapcount"); |
4db7548c MG |
1130 | goto out; |
1131 | } | |
1132 | break; | |
1133 | case 2: | |
1134 | /* | |
1135 | * the second tail page: ->mapping is | |
fa3015b7 | 1136 | * deferred_list.next -- ignore value. |
4db7548c MG |
1137 | */ |
1138 | break; | |
1139 | default: | |
1140 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 1141 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
1142 | goto out; |
1143 | } | |
1144 | break; | |
1145 | } | |
1146 | if (unlikely(!PageTail(page))) { | |
82a3241a | 1147 | bad_page(page, "PageTail not set"); |
4db7548c MG |
1148 | goto out; |
1149 | } | |
1150 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 1151 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
1152 | goto out; |
1153 | } | |
1154 | ret = 0; | |
1155 | out: | |
1156 | page->mapping = NULL; | |
1157 | clear_compound_head(page); | |
1158 | return ret; | |
1159 | } | |
1160 | ||
6471384a AP |
1161 | static void kernel_init_free_pages(struct page *page, int numpages) |
1162 | { | |
1163 | int i; | |
1164 | ||
1165 | for (i = 0; i < numpages; i++) | |
1166 | clear_highpage(page + i); | |
1167 | } | |
1168 | ||
e2769dbd MG |
1169 | static __always_inline bool free_pages_prepare(struct page *page, |
1170 | unsigned int order, bool check_free) | |
4db7548c | 1171 | { |
e2769dbd | 1172 | int bad = 0; |
4db7548c | 1173 | |
4db7548c MG |
1174 | VM_BUG_ON_PAGE(PageTail(page), page); |
1175 | ||
e2769dbd | 1176 | trace_mm_page_free(page, order); |
e2769dbd MG |
1177 | |
1178 | /* | |
1179 | * Check tail pages before head page information is cleared to | |
1180 | * avoid checking PageCompound for order-0 pages. | |
1181 | */ | |
1182 | if (unlikely(order)) { | |
1183 | bool compound = PageCompound(page); | |
1184 | int i; | |
1185 | ||
1186 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1187 | |
9a73f61b KS |
1188 | if (compound) |
1189 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1190 | for (i = 1; i < (1 << order); i++) { |
1191 | if (compound) | |
1192 | bad += free_tail_pages_check(page, page + i); | |
534fe5e3 | 1193 | if (unlikely(check_free_page(page + i))) { |
e2769dbd MG |
1194 | bad++; |
1195 | continue; | |
1196 | } | |
1197 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1198 | } | |
1199 | } | |
bda807d4 | 1200 | if (PageMappingFlags(page)) |
4db7548c | 1201 | page->mapping = NULL; |
c4159a75 | 1202 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
f4b00eab | 1203 | __memcg_kmem_uncharge_page(page, order); |
e2769dbd | 1204 | if (check_free) |
534fe5e3 | 1205 | bad += check_free_page(page); |
e2769dbd MG |
1206 | if (bad) |
1207 | return false; | |
4db7548c | 1208 | |
e2769dbd MG |
1209 | page_cpupid_reset_last(page); |
1210 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1211 | reset_page_owner(page, order); | |
4db7548c MG |
1212 | |
1213 | if (!PageHighMem(page)) { | |
1214 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1215 | PAGE_SIZE << order); |
4db7548c | 1216 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1217 | PAGE_SIZE << order); |
4db7548c | 1218 | } |
6471384a AP |
1219 | if (want_init_on_free()) |
1220 | kernel_init_free_pages(page, 1 << order); | |
1221 | ||
e2769dbd | 1222 | kernel_poison_pages(page, 1 << order, 0); |
234fdce8 QC |
1223 | /* |
1224 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1225 | * does this. So nothing which can access the page's contents should | |
1226 | * happen after this. | |
1227 | */ | |
1228 | arch_free_page(page, order); | |
1229 | ||
8e57f8ac | 1230 | if (debug_pagealloc_enabled_static()) |
d6332692 RE |
1231 | kernel_map_pages(page, 1 << order, 0); |
1232 | ||
3c0c12cc | 1233 | kasan_free_nondeferred_pages(page, order); |
4db7548c | 1234 | |
4db7548c MG |
1235 | return true; |
1236 | } | |
1237 | ||
e2769dbd | 1238 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1239 | /* |
1240 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1241 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1242 | * moved from pcp lists to free lists. | |
1243 | */ | |
1244 | static bool free_pcp_prepare(struct page *page) | |
e2769dbd MG |
1245 | { |
1246 | return free_pages_prepare(page, 0, true); | |
1247 | } | |
1248 | ||
4462b32c | 1249 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1250 | { |
8e57f8ac | 1251 | if (debug_pagealloc_enabled_static()) |
534fe5e3 | 1252 | return check_free_page(page); |
4462b32c VB |
1253 | else |
1254 | return false; | |
e2769dbd MG |
1255 | } |
1256 | #else | |
4462b32c VB |
1257 | /* |
1258 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1259 | * moving from pcp lists to free list in order to reduce overhead. With | |
1260 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1261 | * to the pcp lists. | |
1262 | */ | |
e2769dbd MG |
1263 | static bool free_pcp_prepare(struct page *page) |
1264 | { | |
8e57f8ac | 1265 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
1266 | return free_pages_prepare(page, 0, true); |
1267 | else | |
1268 | return free_pages_prepare(page, 0, false); | |
e2769dbd MG |
1269 | } |
1270 | ||
4db7548c MG |
1271 | static bool bulkfree_pcp_prepare(struct page *page) |
1272 | { | |
534fe5e3 | 1273 | return check_free_page(page); |
4db7548c MG |
1274 | } |
1275 | #endif /* CONFIG_DEBUG_VM */ | |
1276 | ||
97334162 AL |
1277 | static inline void prefetch_buddy(struct page *page) |
1278 | { | |
1279 | unsigned long pfn = page_to_pfn(page); | |
1280 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1281 | struct page *buddy = page + (buddy_pfn - pfn); | |
1282 | ||
1283 | prefetch(buddy); | |
1284 | } | |
1285 | ||
1da177e4 | 1286 | /* |
5f8dcc21 | 1287 | * Frees a number of pages from the PCP lists |
1da177e4 | 1288 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1289 | * count is the number of pages to free. |
1da177e4 LT |
1290 | * |
1291 | * If the zone was previously in an "all pages pinned" state then look to | |
1292 | * see if this freeing clears that state. | |
1293 | * | |
1294 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1295 | * pinned" detection logic. | |
1296 | */ | |
5f8dcc21 MG |
1297 | static void free_pcppages_bulk(struct zone *zone, int count, |
1298 | struct per_cpu_pages *pcp) | |
1da177e4 | 1299 | { |
5f8dcc21 | 1300 | int migratetype = 0; |
a6f9edd6 | 1301 | int batch_free = 0; |
97334162 | 1302 | int prefetch_nr = 0; |
3777999d | 1303 | bool isolated_pageblocks; |
0a5f4e5b AL |
1304 | struct page *page, *tmp; |
1305 | LIST_HEAD(head); | |
f2260e6b | 1306 | |
e5b31ac2 | 1307 | while (count) { |
5f8dcc21 MG |
1308 | struct list_head *list; |
1309 | ||
1310 | /* | |
a6f9edd6 MG |
1311 | * Remove pages from lists in a round-robin fashion. A |
1312 | * batch_free count is maintained that is incremented when an | |
1313 | * empty list is encountered. This is so more pages are freed | |
1314 | * off fuller lists instead of spinning excessively around empty | |
1315 | * lists | |
5f8dcc21 MG |
1316 | */ |
1317 | do { | |
a6f9edd6 | 1318 | batch_free++; |
5f8dcc21 MG |
1319 | if (++migratetype == MIGRATE_PCPTYPES) |
1320 | migratetype = 0; | |
1321 | list = &pcp->lists[migratetype]; | |
1322 | } while (list_empty(list)); | |
48db57f8 | 1323 | |
1d16871d NK |
1324 | /* This is the only non-empty list. Free them all. */ |
1325 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1326 | batch_free = count; |
1d16871d | 1327 | |
a6f9edd6 | 1328 | do { |
a16601c5 | 1329 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1330 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1331 | list_del(&page->lru); |
77ba9062 | 1332 | pcp->count--; |
aa016d14 | 1333 | |
4db7548c MG |
1334 | if (bulkfree_pcp_prepare(page)) |
1335 | continue; | |
1336 | ||
0a5f4e5b | 1337 | list_add_tail(&page->lru, &head); |
97334162 AL |
1338 | |
1339 | /* | |
1340 | * We are going to put the page back to the global | |
1341 | * pool, prefetch its buddy to speed up later access | |
1342 | * under zone->lock. It is believed the overhead of | |
1343 | * an additional test and calculating buddy_pfn here | |
1344 | * can be offset by reduced memory latency later. To | |
1345 | * avoid excessive prefetching due to large count, only | |
1346 | * prefetch buddy for the first pcp->batch nr of pages. | |
1347 | */ | |
1348 | if (prefetch_nr++ < pcp->batch) | |
1349 | prefetch_buddy(page); | |
e5b31ac2 | 1350 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1351 | } |
0a5f4e5b AL |
1352 | |
1353 | spin_lock(&zone->lock); | |
1354 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1355 | ||
1356 | /* | |
1357 | * Use safe version since after __free_one_page(), | |
1358 | * page->lru.next will not point to original list. | |
1359 | */ | |
1360 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1361 | int mt = get_pcppage_migratetype(page); | |
1362 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1363 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1364 | /* Pageblock could have been isolated meanwhile */ | |
1365 | if (unlikely(isolated_pageblocks)) | |
1366 | mt = get_pageblock_migratetype(page); | |
1367 | ||
36e66c55 | 1368 | __free_one_page(page, page_to_pfn(page), zone, 0, mt, true); |
0a5f4e5b AL |
1369 | trace_mm_page_pcpu_drain(page, 0, mt); |
1370 | } | |
d34b0733 | 1371 | spin_unlock(&zone->lock); |
1da177e4 LT |
1372 | } |
1373 | ||
dc4b0caf MG |
1374 | static void free_one_page(struct zone *zone, |
1375 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1376 | unsigned int order, |
ed0ae21d | 1377 | int migratetype) |
1da177e4 | 1378 | { |
d34b0733 | 1379 | spin_lock(&zone->lock); |
ad53f92e JK |
1380 | if (unlikely(has_isolate_pageblock(zone) || |
1381 | is_migrate_isolate(migratetype))) { | |
1382 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1383 | } |
36e66c55 | 1384 | __free_one_page(page, pfn, zone, order, migratetype, true); |
d34b0733 | 1385 | spin_unlock(&zone->lock); |
48db57f8 NP |
1386 | } |
1387 | ||
1e8ce83c | 1388 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1389 | unsigned long zone, int nid) |
1e8ce83c | 1390 | { |
d0dc12e8 | 1391 | mm_zero_struct_page(page); |
1e8ce83c | 1392 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1393 | init_page_count(page); |
1394 | page_mapcount_reset(page); | |
1395 | page_cpupid_reset_last(page); | |
2813b9c0 | 1396 | page_kasan_tag_reset(page); |
1e8ce83c | 1397 | |
1e8ce83c RH |
1398 | INIT_LIST_HEAD(&page->lru); |
1399 | #ifdef WANT_PAGE_VIRTUAL | |
1400 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1401 | if (!is_highmem_idx(zone)) | |
1402 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1403 | #endif | |
1404 | } | |
1405 | ||
7e18adb4 | 1406 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1407 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1408 | { |
1409 | pg_data_t *pgdat; | |
1410 | int nid, zid; | |
1411 | ||
1412 | if (!early_page_uninitialised(pfn)) | |
1413 | return; | |
1414 | ||
1415 | nid = early_pfn_to_nid(pfn); | |
1416 | pgdat = NODE_DATA(nid); | |
1417 | ||
1418 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1419 | struct zone *zone = &pgdat->node_zones[zid]; | |
1420 | ||
1421 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1422 | break; | |
1423 | } | |
d0dc12e8 | 1424 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1425 | } |
1426 | #else | |
1427 | static inline void init_reserved_page(unsigned long pfn) | |
1428 | { | |
1429 | } | |
1430 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1431 | ||
92923ca3 NZ |
1432 | /* |
1433 | * Initialised pages do not have PageReserved set. This function is | |
1434 | * called for each range allocated by the bootmem allocator and | |
1435 | * marks the pages PageReserved. The remaining valid pages are later | |
1436 | * sent to the buddy page allocator. | |
1437 | */ | |
4b50bcc7 | 1438 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1439 | { |
1440 | unsigned long start_pfn = PFN_DOWN(start); | |
1441 | unsigned long end_pfn = PFN_UP(end); | |
1442 | ||
7e18adb4 MG |
1443 | for (; start_pfn < end_pfn; start_pfn++) { |
1444 | if (pfn_valid(start_pfn)) { | |
1445 | struct page *page = pfn_to_page(start_pfn); | |
1446 | ||
1447 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1448 | |
1449 | /* Avoid false-positive PageTail() */ | |
1450 | INIT_LIST_HEAD(&page->lru); | |
1451 | ||
d483da5b AD |
1452 | /* |
1453 | * no need for atomic set_bit because the struct | |
1454 | * page is not visible yet so nobody should | |
1455 | * access it yet. | |
1456 | */ | |
1457 | __SetPageReserved(page); | |
7e18adb4 MG |
1458 | } |
1459 | } | |
92923ca3 NZ |
1460 | } |
1461 | ||
ec95f53a KM |
1462 | static void __free_pages_ok(struct page *page, unsigned int order) |
1463 | { | |
d34b0733 | 1464 | unsigned long flags; |
95e34412 | 1465 | int migratetype; |
dc4b0caf | 1466 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1467 | |
e2769dbd | 1468 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1469 | return; |
1470 | ||
cfc47a28 | 1471 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1472 | local_irq_save(flags); |
1473 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1474 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1475 | local_irq_restore(flags); |
1da177e4 LT |
1476 | } |
1477 | ||
a9cd410a | 1478 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1479 | { |
c3993076 | 1480 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1481 | struct page *p = page; |
c3993076 | 1482 | unsigned int loop; |
a226f6c8 | 1483 | |
e2d0bd2b YL |
1484 | prefetchw(p); |
1485 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1486 | prefetchw(p + 1); | |
c3993076 JW |
1487 | __ClearPageReserved(p); |
1488 | set_page_count(p, 0); | |
a226f6c8 | 1489 | } |
e2d0bd2b YL |
1490 | __ClearPageReserved(p); |
1491 | set_page_count(p, 0); | |
c3993076 | 1492 | |
9705bea5 | 1493 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
c3993076 JW |
1494 | set_page_refcounted(page); |
1495 | __free_pages(page, order); | |
a226f6c8 DH |
1496 | } |
1497 | ||
3f08a302 | 1498 | #ifdef CONFIG_NEED_MULTIPLE_NODES |
7ace9917 | 1499 | |
75a592a4 MG |
1500 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1501 | ||
6f24fbd3 MR |
1502 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
1503 | ||
1504 | /* | |
1505 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
1506 | */ | |
1507 | int __meminit __early_pfn_to_nid(unsigned long pfn, | |
1508 | struct mminit_pfnnid_cache *state) | |
75a592a4 | 1509 | { |
6f24fbd3 | 1510 | unsigned long start_pfn, end_pfn; |
75a592a4 MG |
1511 | int nid; |
1512 | ||
6f24fbd3 MR |
1513 | if (state->last_start <= pfn && pfn < state->last_end) |
1514 | return state->last_nid; | |
1515 | ||
1516 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); | |
1517 | if (nid != NUMA_NO_NODE) { | |
1518 | state->last_start = start_pfn; | |
1519 | state->last_end = end_pfn; | |
1520 | state->last_nid = nid; | |
1521 | } | |
7ace9917 MG |
1522 | |
1523 | return nid; | |
75a592a4 | 1524 | } |
6f24fbd3 | 1525 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ |
75a592a4 | 1526 | |
75a592a4 | 1527 | int __meminit early_pfn_to_nid(unsigned long pfn) |
75a592a4 | 1528 | { |
7ace9917 | 1529 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1530 | int nid; |
1531 | ||
7ace9917 | 1532 | spin_lock(&early_pfn_lock); |
56ec43d8 | 1533 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1534 | if (nid < 0) |
e4568d38 | 1535 | nid = first_online_node; |
7ace9917 | 1536 | spin_unlock(&early_pfn_lock); |
75a592a4 | 1537 | |
7ace9917 | 1538 | return nid; |
75a592a4 | 1539 | } |
3f08a302 | 1540 | #endif /* CONFIG_NEED_MULTIPLE_NODES */ |
75a592a4 | 1541 | |
7c2ee349 | 1542 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1543 | unsigned int order) |
1544 | { | |
1545 | if (early_page_uninitialised(pfn)) | |
1546 | return; | |
a9cd410a | 1547 | __free_pages_core(page, order); |
3a80a7fa MG |
1548 | } |
1549 | ||
7cf91a98 JK |
1550 | /* |
1551 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1552 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1553 | * with the migration of free compaction scanner. The scanners then need to | |
1554 | * use only pfn_valid_within() check for arches that allow holes within | |
1555 | * pageblocks. | |
1556 | * | |
1557 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1558 | * | |
1559 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1560 | * i.e. it's possible that all pages within a zones range of pages do not | |
1561 | * belong to a single zone. We assume that a border between node0 and node1 | |
1562 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1563 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1564 | * the first and last page of a pageblock and avoid checking each individual | |
1565 | * page in a pageblock. | |
1566 | */ | |
1567 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1568 | unsigned long end_pfn, struct zone *zone) | |
1569 | { | |
1570 | struct page *start_page; | |
1571 | struct page *end_page; | |
1572 | ||
1573 | /* end_pfn is one past the range we are checking */ | |
1574 | end_pfn--; | |
1575 | ||
1576 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1577 | return NULL; | |
1578 | ||
2d070eab MH |
1579 | start_page = pfn_to_online_page(start_pfn); |
1580 | if (!start_page) | |
1581 | return NULL; | |
7cf91a98 JK |
1582 | |
1583 | if (page_zone(start_page) != zone) | |
1584 | return NULL; | |
1585 | ||
1586 | end_page = pfn_to_page(end_pfn); | |
1587 | ||
1588 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1589 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1590 | return NULL; | |
1591 | ||
1592 | return start_page; | |
1593 | } | |
1594 | ||
1595 | void set_zone_contiguous(struct zone *zone) | |
1596 | { | |
1597 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1598 | unsigned long block_end_pfn; | |
1599 | ||
1600 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1601 | for (; block_start_pfn < zone_end_pfn(zone); | |
1602 | block_start_pfn = block_end_pfn, | |
1603 | block_end_pfn += pageblock_nr_pages) { | |
1604 | ||
1605 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1606 | ||
1607 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1608 | block_end_pfn, zone)) | |
1609 | return; | |
e84fe99b | 1610 | cond_resched(); |
7cf91a98 JK |
1611 | } |
1612 | ||
1613 | /* We confirm that there is no hole */ | |
1614 | zone->contiguous = true; | |
1615 | } | |
1616 | ||
1617 | void clear_zone_contiguous(struct zone *zone) | |
1618 | { | |
1619 | zone->contiguous = false; | |
1620 | } | |
1621 | ||
7e18adb4 | 1622 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1623 | static void __init deferred_free_range(unsigned long pfn, |
1624 | unsigned long nr_pages) | |
a4de83dd | 1625 | { |
2f47a91f PT |
1626 | struct page *page; |
1627 | unsigned long i; | |
a4de83dd | 1628 | |
2f47a91f | 1629 | if (!nr_pages) |
a4de83dd MG |
1630 | return; |
1631 | ||
2f47a91f PT |
1632 | page = pfn_to_page(pfn); |
1633 | ||
a4de83dd | 1634 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1635 | if (nr_pages == pageblock_nr_pages && |
1636 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1637 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1638 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1639 | return; |
1640 | } | |
1641 | ||
e780149b XQ |
1642 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1643 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1644 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
a9cd410a | 1645 | __free_pages_core(page, 0); |
e780149b | 1646 | } |
a4de83dd MG |
1647 | } |
1648 | ||
d3cd131d NS |
1649 | /* Completion tracking for deferred_init_memmap() threads */ |
1650 | static atomic_t pgdat_init_n_undone __initdata; | |
1651 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1652 | ||
1653 | static inline void __init pgdat_init_report_one_done(void) | |
1654 | { | |
1655 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1656 | complete(&pgdat_init_all_done_comp); | |
1657 | } | |
0e1cc95b | 1658 | |
2f47a91f | 1659 | /* |
80b1f41c PT |
1660 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1661 | * | |
1662 | * First we check if pfn is valid on architectures where it is possible to have | |
1663 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1664 | * function is optimized out. | |
1665 | * | |
1666 | * Then, we check if a current large page is valid by only checking the validity | |
1667 | * of the head pfn. | |
2f47a91f | 1668 | */ |
56ec43d8 | 1669 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1670 | { |
80b1f41c PT |
1671 | if (!pfn_valid_within(pfn)) |
1672 | return false; | |
1673 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1674 | return false; | |
80b1f41c PT |
1675 | return true; |
1676 | } | |
2f47a91f | 1677 | |
80b1f41c PT |
1678 | /* |
1679 | * Free pages to buddy allocator. Try to free aligned pages in | |
1680 | * pageblock_nr_pages sizes. | |
1681 | */ | |
56ec43d8 | 1682 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1683 | unsigned long end_pfn) |
1684 | { | |
80b1f41c PT |
1685 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
1686 | unsigned long nr_free = 0; | |
2f47a91f | 1687 | |
80b1f41c | 1688 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1689 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1690 | deferred_free_range(pfn - nr_free, nr_free); |
1691 | nr_free = 0; | |
1692 | } else if (!(pfn & nr_pgmask)) { | |
1693 | deferred_free_range(pfn - nr_free, nr_free); | |
1694 | nr_free = 1; | |
80b1f41c PT |
1695 | } else { |
1696 | nr_free++; | |
1697 | } | |
1698 | } | |
1699 | /* Free the last block of pages to allocator */ | |
1700 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1701 | } |
1702 | ||
80b1f41c PT |
1703 | /* |
1704 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1705 | * by performing it only once every pageblock_nr_pages. | |
1706 | * Return number of pages initialized. | |
1707 | */ | |
56ec43d8 | 1708 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1709 | unsigned long pfn, |
1710 | unsigned long end_pfn) | |
2f47a91f | 1711 | { |
2f47a91f | 1712 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
56ec43d8 | 1713 | int nid = zone_to_nid(zone); |
2f47a91f | 1714 | unsigned long nr_pages = 0; |
56ec43d8 | 1715 | int zid = zone_idx(zone); |
2f47a91f | 1716 | struct page *page = NULL; |
2f47a91f | 1717 | |
80b1f41c | 1718 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1719 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1720 | page = NULL; |
2f47a91f | 1721 | continue; |
80b1f41c | 1722 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1723 | page = pfn_to_page(pfn); |
80b1f41c PT |
1724 | } else { |
1725 | page++; | |
2f47a91f | 1726 | } |
d0dc12e8 | 1727 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1728 | nr_pages++; |
2f47a91f | 1729 | } |
80b1f41c | 1730 | return (nr_pages); |
2f47a91f PT |
1731 | } |
1732 | ||
0e56acae AD |
1733 | /* |
1734 | * This function is meant to pre-load the iterator for the zone init. | |
1735 | * Specifically it walks through the ranges until we are caught up to the | |
1736 | * first_init_pfn value and exits there. If we never encounter the value we | |
1737 | * return false indicating there are no valid ranges left. | |
1738 | */ | |
1739 | static bool __init | |
1740 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
1741 | unsigned long *spfn, unsigned long *epfn, | |
1742 | unsigned long first_init_pfn) | |
1743 | { | |
1744 | u64 j; | |
1745 | ||
1746 | /* | |
1747 | * Start out by walking through the ranges in this zone that have | |
1748 | * already been initialized. We don't need to do anything with them | |
1749 | * so we just need to flush them out of the system. | |
1750 | */ | |
1751 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
1752 | if (*epfn <= first_init_pfn) | |
1753 | continue; | |
1754 | if (*spfn < first_init_pfn) | |
1755 | *spfn = first_init_pfn; | |
1756 | *i = j; | |
1757 | return true; | |
1758 | } | |
1759 | ||
1760 | return false; | |
1761 | } | |
1762 | ||
1763 | /* | |
1764 | * Initialize and free pages. We do it in two loops: first we initialize | |
1765 | * struct page, then free to buddy allocator, because while we are | |
1766 | * freeing pages we can access pages that are ahead (computing buddy | |
1767 | * page in __free_one_page()). | |
1768 | * | |
1769 | * In order to try and keep some memory in the cache we have the loop | |
1770 | * broken along max page order boundaries. This way we will not cause | |
1771 | * any issues with the buddy page computation. | |
1772 | */ | |
1773 | static unsigned long __init | |
1774 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
1775 | unsigned long *end_pfn) | |
1776 | { | |
1777 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
1778 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
1779 | unsigned long nr_pages = 0; | |
1780 | u64 j = *i; | |
1781 | ||
1782 | /* First we loop through and initialize the page values */ | |
1783 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
1784 | unsigned long t; | |
1785 | ||
1786 | if (mo_pfn <= *start_pfn) | |
1787 | break; | |
1788 | ||
1789 | t = min(mo_pfn, *end_pfn); | |
1790 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
1791 | ||
1792 | if (mo_pfn < *end_pfn) { | |
1793 | *start_pfn = mo_pfn; | |
1794 | break; | |
1795 | } | |
1796 | } | |
1797 | ||
1798 | /* Reset values and now loop through freeing pages as needed */ | |
1799 | swap(j, *i); | |
1800 | ||
1801 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
1802 | unsigned long t; | |
1803 | ||
1804 | if (mo_pfn <= spfn) | |
1805 | break; | |
1806 | ||
1807 | t = min(mo_pfn, epfn); | |
1808 | deferred_free_pages(spfn, t); | |
1809 | ||
1810 | if (mo_pfn <= epfn) | |
1811 | break; | |
1812 | } | |
1813 | ||
1814 | return nr_pages; | |
1815 | } | |
1816 | ||
e4443149 DJ |
1817 | static void __init |
1818 | deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, | |
1819 | void *arg) | |
1820 | { | |
1821 | unsigned long spfn, epfn; | |
1822 | struct zone *zone = arg; | |
1823 | u64 i; | |
1824 | ||
1825 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn); | |
1826 | ||
1827 | /* | |
1828 | * Initialize and free pages in MAX_ORDER sized increments so that we | |
1829 | * can avoid introducing any issues with the buddy allocator. | |
1830 | */ | |
1831 | while (spfn < end_pfn) { | |
1832 | deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
1833 | cond_resched(); | |
1834 | } | |
1835 | } | |
1836 | ||
ecd09650 DJ |
1837 | /* An arch may override for more concurrency. */ |
1838 | __weak int __init | |
1839 | deferred_page_init_max_threads(const struct cpumask *node_cpumask) | |
1840 | { | |
1841 | return 1; | |
1842 | } | |
1843 | ||
7e18adb4 | 1844 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1845 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1846 | { |
0e1cc95b | 1847 | pg_data_t *pgdat = data; |
0e56acae | 1848 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
89c7c402 | 1849 | unsigned long spfn = 0, epfn = 0; |
0e56acae | 1850 | unsigned long first_init_pfn, flags; |
7e18adb4 | 1851 | unsigned long start = jiffies; |
7e18adb4 | 1852 | struct zone *zone; |
e4443149 | 1853 | int zid, max_threads; |
2f47a91f | 1854 | u64 i; |
7e18adb4 | 1855 | |
3a2d7fa8 PT |
1856 | /* Bind memory initialisation thread to a local node if possible */ |
1857 | if (!cpumask_empty(cpumask)) | |
1858 | set_cpus_allowed_ptr(current, cpumask); | |
1859 | ||
1860 | pgdat_resize_lock(pgdat, &flags); | |
1861 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1862 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1863 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1864 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1865 | return 0; |
1866 | } | |
1867 | ||
7e18adb4 MG |
1868 | /* Sanity check boundaries */ |
1869 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1870 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1871 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1872 | ||
3d060856 PT |
1873 | /* |
1874 | * Once we unlock here, the zone cannot be grown anymore, thus if an | |
1875 | * interrupt thread must allocate this early in boot, zone must be | |
1876 | * pre-grown prior to start of deferred page initialization. | |
1877 | */ | |
1878 | pgdat_resize_unlock(pgdat, &flags); | |
1879 | ||
7e18adb4 MG |
1880 | /* Only the highest zone is deferred so find it */ |
1881 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1882 | zone = pgdat->node_zones + zid; | |
1883 | if (first_init_pfn < zone_end_pfn(zone)) | |
1884 | break; | |
1885 | } | |
0e56acae AD |
1886 | |
1887 | /* If the zone is empty somebody else may have cleared out the zone */ | |
1888 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1889 | first_init_pfn)) | |
1890 | goto zone_empty; | |
7e18adb4 | 1891 | |
ecd09650 | 1892 | max_threads = deferred_page_init_max_threads(cpumask); |
7e18adb4 | 1893 | |
117003c3 | 1894 | while (spfn < epfn) { |
e4443149 DJ |
1895 | unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION); |
1896 | struct padata_mt_job job = { | |
1897 | .thread_fn = deferred_init_memmap_chunk, | |
1898 | .fn_arg = zone, | |
1899 | .start = spfn, | |
1900 | .size = epfn_align - spfn, | |
1901 | .align = PAGES_PER_SECTION, | |
1902 | .min_chunk = PAGES_PER_SECTION, | |
1903 | .max_threads = max_threads, | |
1904 | }; | |
1905 | ||
1906 | padata_do_multithreaded(&job); | |
1907 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1908 | epfn_align); | |
117003c3 | 1909 | } |
0e56acae | 1910 | zone_empty: |
7e18adb4 MG |
1911 | /* Sanity check that the next zone really is unpopulated */ |
1912 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1913 | ||
89c7c402 DJ |
1914 | pr_info("node %d deferred pages initialised in %ums\n", |
1915 | pgdat->node_id, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
1916 | |
1917 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1918 | return 0; |
1919 | } | |
c9e97a19 | 1920 | |
c9e97a19 PT |
1921 | /* |
1922 | * If this zone has deferred pages, try to grow it by initializing enough | |
1923 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
1924 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
1925 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
1926 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
1927 | * | |
1928 | * Return true when zone was grown, otherwise return false. We return true even | |
1929 | * when we grow less than requested, to let the caller decide if there are | |
1930 | * enough pages to satisfy the allocation. | |
1931 | * | |
1932 | * Note: We use noinline because this function is needed only during boot, and | |
1933 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
1934 | * making sure that it is not inlined into permanent text section. | |
1935 | */ | |
1936 | static noinline bool __init | |
1937 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
1938 | { | |
c9e97a19 | 1939 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 1940 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 1941 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
1942 | unsigned long spfn, epfn, flags; |
1943 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
1944 | u64 i; |
1945 | ||
1946 | /* Only the last zone may have deferred pages */ | |
1947 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
1948 | return false; | |
1949 | ||
1950 | pgdat_resize_lock(pgdat, &flags); | |
1951 | ||
c9e97a19 PT |
1952 | /* |
1953 | * If someone grew this zone while we were waiting for spinlock, return | |
1954 | * true, as there might be enough pages already. | |
1955 | */ | |
1956 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
1957 | pgdat_resize_unlock(pgdat, &flags); | |
1958 | return true; | |
1959 | } | |
1960 | ||
0e56acae AD |
1961 | /* If the zone is empty somebody else may have cleared out the zone */ |
1962 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1963 | first_deferred_pfn)) { | |
1964 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 1965 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
1966 | /* Retry only once. */ |
1967 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
1968 | } |
1969 | ||
0e56acae AD |
1970 | /* |
1971 | * Initialize and free pages in MAX_ORDER sized increments so | |
1972 | * that we can avoid introducing any issues with the buddy | |
1973 | * allocator. | |
1974 | */ | |
1975 | while (spfn < epfn) { | |
1976 | /* update our first deferred PFN for this section */ | |
1977 | first_deferred_pfn = spfn; | |
1978 | ||
1979 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
117003c3 | 1980 | touch_nmi_watchdog(); |
c9e97a19 | 1981 | |
0e56acae AD |
1982 | /* We should only stop along section boundaries */ |
1983 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
1984 | continue; | |
c9e97a19 | 1985 | |
0e56acae | 1986 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
1987 | if (nr_pages >= nr_pages_needed) |
1988 | break; | |
1989 | } | |
1990 | ||
0e56acae | 1991 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
1992 | pgdat_resize_unlock(pgdat, &flags); |
1993 | ||
1994 | return nr_pages > 0; | |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * deferred_grow_zone() is __init, but it is called from | |
1999 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
2000 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
2001 | * and to ensure that the function body gets unloaded. | |
2002 | */ | |
2003 | static bool __ref | |
2004 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
2005 | { | |
2006 | return deferred_grow_zone(zone, order); | |
2007 | } | |
2008 | ||
7cf91a98 | 2009 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
2010 | |
2011 | void __init page_alloc_init_late(void) | |
2012 | { | |
7cf91a98 | 2013 | struct zone *zone; |
e900a918 | 2014 | int nid; |
7cf91a98 JK |
2015 | |
2016 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 2017 | |
d3cd131d NS |
2018 | /* There will be num_node_state(N_MEMORY) threads */ |
2019 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 2020 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
2021 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
2022 | } | |
2023 | ||
2024 | /* Block until all are initialised */ | |
d3cd131d | 2025 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 2026 | |
3e8fc007 MG |
2027 | /* |
2028 | * The number of managed pages has changed due to the initialisation | |
2029 | * so the pcpu batch and high limits needs to be updated or the limits | |
2030 | * will be artificially small. | |
2031 | */ | |
2032 | for_each_populated_zone(zone) | |
2033 | zone_pcp_update(zone); | |
2034 | ||
c9e97a19 PT |
2035 | /* |
2036 | * We initialized the rest of the deferred pages. Permanently disable | |
2037 | * on-demand struct page initialization. | |
2038 | */ | |
2039 | static_branch_disable(&deferred_pages); | |
2040 | ||
4248b0da MG |
2041 | /* Reinit limits that are based on free pages after the kernel is up */ |
2042 | files_maxfiles_init(); | |
7cf91a98 | 2043 | #endif |
350e88ba | 2044 | |
3010f876 PT |
2045 | /* Discard memblock private memory */ |
2046 | memblock_discard(); | |
7cf91a98 | 2047 | |
e900a918 DW |
2048 | for_each_node_state(nid, N_MEMORY) |
2049 | shuffle_free_memory(NODE_DATA(nid)); | |
2050 | ||
7cf91a98 JK |
2051 | for_each_populated_zone(zone) |
2052 | set_zone_contiguous(zone); | |
7e18adb4 | 2053 | } |
7e18adb4 | 2054 | |
47118af0 | 2055 | #ifdef CONFIG_CMA |
9cf510a5 | 2056 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
2057 | void __init init_cma_reserved_pageblock(struct page *page) |
2058 | { | |
2059 | unsigned i = pageblock_nr_pages; | |
2060 | struct page *p = page; | |
2061 | ||
2062 | do { | |
2063 | __ClearPageReserved(p); | |
2064 | set_page_count(p, 0); | |
d883c6cf | 2065 | } while (++p, --i); |
47118af0 | 2066 | |
47118af0 | 2067 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
2068 | |
2069 | if (pageblock_order >= MAX_ORDER) { | |
2070 | i = pageblock_nr_pages; | |
2071 | p = page; | |
2072 | do { | |
2073 | set_page_refcounted(p); | |
2074 | __free_pages(p, MAX_ORDER - 1); | |
2075 | p += MAX_ORDER_NR_PAGES; | |
2076 | } while (i -= MAX_ORDER_NR_PAGES); | |
2077 | } else { | |
2078 | set_page_refcounted(page); | |
2079 | __free_pages(page, pageblock_order); | |
2080 | } | |
2081 | ||
3dcc0571 | 2082 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
2083 | } |
2084 | #endif | |
1da177e4 LT |
2085 | |
2086 | /* | |
2087 | * The order of subdivision here is critical for the IO subsystem. | |
2088 | * Please do not alter this order without good reasons and regression | |
2089 | * testing. Specifically, as large blocks of memory are subdivided, | |
2090 | * the order in which smaller blocks are delivered depends on the order | |
2091 | * they're subdivided in this function. This is the primary factor | |
2092 | * influencing the order in which pages are delivered to the IO | |
2093 | * subsystem according to empirical testing, and this is also justified | |
2094 | * by considering the behavior of a buddy system containing a single | |
2095 | * large block of memory acted on by a series of small allocations. | |
2096 | * This behavior is a critical factor in sglist merging's success. | |
2097 | * | |
6d49e352 | 2098 | * -- nyc |
1da177e4 | 2099 | */ |
085cc7d5 | 2100 | static inline void expand(struct zone *zone, struct page *page, |
6ab01363 | 2101 | int low, int high, int migratetype) |
1da177e4 LT |
2102 | { |
2103 | unsigned long size = 1 << high; | |
2104 | ||
2105 | while (high > low) { | |
1da177e4 LT |
2106 | high--; |
2107 | size >>= 1; | |
309381fe | 2108 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2109 | |
acbc15a4 JK |
2110 | /* |
2111 | * Mark as guard pages (or page), that will allow to | |
2112 | * merge back to allocator when buddy will be freed. | |
2113 | * Corresponding page table entries will not be touched, | |
2114 | * pages will stay not present in virtual address space | |
2115 | */ | |
2116 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2117 | continue; |
acbc15a4 | 2118 | |
6ab01363 | 2119 | add_to_free_list(&page[size], zone, high, migratetype); |
1da177e4 LT |
2120 | set_page_order(&page[size], high); |
2121 | } | |
1da177e4 LT |
2122 | } |
2123 | ||
4e611801 | 2124 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2125 | { |
f4c18e6f | 2126 | if (unlikely(page->flags & __PG_HWPOISON)) { |
e570f56c NH |
2127 | /* Don't complain about hwpoisoned pages */ |
2128 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2129 | return; | |
f4c18e6f | 2130 | } |
58b7f119 WY |
2131 | |
2132 | bad_page(page, | |
2133 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
4e611801 VB |
2134 | } |
2135 | ||
2136 | /* | |
2137 | * This page is about to be returned from the page allocator | |
2138 | */ | |
2139 | static inline int check_new_page(struct page *page) | |
2140 | { | |
2141 | if (likely(page_expected_state(page, | |
2142 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2143 | return 0; | |
2144 | ||
2145 | check_new_page_bad(page); | |
2146 | return 1; | |
2a7684a2 WF |
2147 | } |
2148 | ||
bd33ef36 | 2149 | static inline bool free_pages_prezeroed(void) |
1414c7f4 | 2150 | { |
6471384a AP |
2151 | return (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && |
2152 | page_poisoning_enabled()) || want_init_on_free(); | |
1414c7f4 LA |
2153 | } |
2154 | ||
479f854a | 2155 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2156 | /* |
2157 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2158 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2159 | * also checked when pcp lists are refilled from the free lists. | |
2160 | */ | |
2161 | static inline bool check_pcp_refill(struct page *page) | |
479f854a | 2162 | { |
8e57f8ac | 2163 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2164 | return check_new_page(page); |
2165 | else | |
2166 | return false; | |
479f854a MG |
2167 | } |
2168 | ||
4462b32c | 2169 | static inline bool check_new_pcp(struct page *page) |
479f854a MG |
2170 | { |
2171 | return check_new_page(page); | |
2172 | } | |
2173 | #else | |
4462b32c VB |
2174 | /* |
2175 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2176 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2177 | * enabled, they are also checked when being allocated from the pcp lists. | |
2178 | */ | |
2179 | static inline bool check_pcp_refill(struct page *page) | |
479f854a MG |
2180 | { |
2181 | return check_new_page(page); | |
2182 | } | |
4462b32c | 2183 | static inline bool check_new_pcp(struct page *page) |
479f854a | 2184 | { |
8e57f8ac | 2185 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2186 | return check_new_page(page); |
2187 | else | |
2188 | return false; | |
479f854a MG |
2189 | } |
2190 | #endif /* CONFIG_DEBUG_VM */ | |
2191 | ||
2192 | static bool check_new_pages(struct page *page, unsigned int order) | |
2193 | { | |
2194 | int i; | |
2195 | for (i = 0; i < (1 << order); i++) { | |
2196 | struct page *p = page + i; | |
2197 | ||
2198 | if (unlikely(check_new_page(p))) | |
2199 | return true; | |
2200 | } | |
2201 | ||
2202 | return false; | |
2203 | } | |
2204 | ||
46f24fd8 JK |
2205 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2206 | gfp_t gfp_flags) | |
2207 | { | |
2208 | set_page_private(page, 0); | |
2209 | set_page_refcounted(page); | |
2210 | ||
2211 | arch_alloc_page(page, order); | |
8e57f8ac | 2212 | if (debug_pagealloc_enabled_static()) |
d6332692 | 2213 | kernel_map_pages(page, 1 << order, 1); |
46f24fd8 | 2214 | kasan_alloc_pages(page, order); |
4117992d | 2215 | kernel_poison_pages(page, 1 << order, 1); |
46f24fd8 JK |
2216 | set_page_owner(page, order, gfp_flags); |
2217 | } | |
2218 | ||
479f854a | 2219 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2220 | unsigned int alloc_flags) |
2a7684a2 | 2221 | { |
46f24fd8 | 2222 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2223 | |
6471384a AP |
2224 | if (!free_pages_prezeroed() && want_init_on_alloc(gfp_flags)) |
2225 | kernel_init_free_pages(page, 1 << order); | |
17cf4406 NP |
2226 | |
2227 | if (order && (gfp_flags & __GFP_COMP)) | |
2228 | prep_compound_page(page, order); | |
2229 | ||
75379191 | 2230 | /* |
2f064f34 | 2231 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2232 | * allocate the page. The expectation is that the caller is taking |
2233 | * steps that will free more memory. The caller should avoid the page | |
2234 | * being used for !PFMEMALLOC purposes. | |
2235 | */ | |
2f064f34 MH |
2236 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2237 | set_page_pfmemalloc(page); | |
2238 | else | |
2239 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2240 | } |
2241 | ||
56fd56b8 MG |
2242 | /* |
2243 | * Go through the free lists for the given migratetype and remove | |
2244 | * the smallest available page from the freelists | |
2245 | */ | |
85ccc8fa | 2246 | static __always_inline |
728ec980 | 2247 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2248 | int migratetype) |
2249 | { | |
2250 | unsigned int current_order; | |
b8af2941 | 2251 | struct free_area *area; |
56fd56b8 MG |
2252 | struct page *page; |
2253 | ||
2254 | /* Find a page of the appropriate size in the preferred list */ | |
2255 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2256 | area = &(zone->free_area[current_order]); | |
b03641af | 2257 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2258 | if (!page) |
2259 | continue; | |
6ab01363 AD |
2260 | del_page_from_free_list(page, zone, current_order); |
2261 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 2262 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
2263 | return page; |
2264 | } | |
2265 | ||
2266 | return NULL; | |
2267 | } | |
2268 | ||
2269 | ||
b2a0ac88 MG |
2270 | /* |
2271 | * This array describes the order lists are fallen back to when | |
2272 | * the free lists for the desirable migrate type are depleted | |
2273 | */ | |
47118af0 | 2274 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e | 2275 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2276 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2277 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
47118af0 | 2278 | #ifdef CONFIG_CMA |
974a786e | 2279 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 2280 | #endif |
194159fb | 2281 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 2282 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 2283 | #endif |
b2a0ac88 MG |
2284 | }; |
2285 | ||
dc67647b | 2286 | #ifdef CONFIG_CMA |
85ccc8fa | 2287 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2288 | unsigned int order) |
2289 | { | |
2290 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2291 | } | |
2292 | #else | |
2293 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2294 | unsigned int order) { return NULL; } | |
2295 | #endif | |
2296 | ||
c361be55 MG |
2297 | /* |
2298 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 2299 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2300 | * boundary. If alignment is required, use move_freepages_block() |
2301 | */ | |
02aa0cdd | 2302 | static int move_freepages(struct zone *zone, |
b69a7288 | 2303 | struct page *start_page, struct page *end_page, |
02aa0cdd | 2304 | int migratetype, int *num_movable) |
c361be55 MG |
2305 | { |
2306 | struct page *page; | |
d00181b9 | 2307 | unsigned int order; |
d100313f | 2308 | int pages_moved = 0; |
c361be55 | 2309 | |
c361be55 MG |
2310 | for (page = start_page; page <= end_page;) { |
2311 | if (!pfn_valid_within(page_to_pfn(page))) { | |
2312 | page++; | |
2313 | continue; | |
2314 | } | |
2315 | ||
2316 | if (!PageBuddy(page)) { | |
02aa0cdd VB |
2317 | /* |
2318 | * We assume that pages that could be isolated for | |
2319 | * migration are movable. But we don't actually try | |
2320 | * isolating, as that would be expensive. | |
2321 | */ | |
2322 | if (num_movable && | |
2323 | (PageLRU(page) || __PageMovable(page))) | |
2324 | (*num_movable)++; | |
2325 | ||
c361be55 MG |
2326 | page++; |
2327 | continue; | |
2328 | } | |
2329 | ||
cd961038 DR |
2330 | /* Make sure we are not inadvertently changing nodes */ |
2331 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2332 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2333 | ||
c361be55 | 2334 | order = page_order(page); |
6ab01363 | 2335 | move_to_free_list(page, zone, order, migratetype); |
c361be55 | 2336 | page += 1 << order; |
d100313f | 2337 | pages_moved += 1 << order; |
c361be55 MG |
2338 | } |
2339 | ||
d100313f | 2340 | return pages_moved; |
c361be55 MG |
2341 | } |
2342 | ||
ee6f509c | 2343 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2344 | int migratetype, int *num_movable) |
c361be55 MG |
2345 | { |
2346 | unsigned long start_pfn, end_pfn; | |
2347 | struct page *start_page, *end_page; | |
2348 | ||
4a222127 DR |
2349 | if (num_movable) |
2350 | *num_movable = 0; | |
2351 | ||
c361be55 | 2352 | start_pfn = page_to_pfn(page); |
d9c23400 | 2353 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 2354 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
2355 | end_page = start_page + pageblock_nr_pages - 1; |
2356 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
2357 | |
2358 | /* Do not cross zone boundaries */ | |
108bcc96 | 2359 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 2360 | start_page = page; |
108bcc96 | 2361 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2362 | return 0; |
2363 | ||
02aa0cdd VB |
2364 | return move_freepages(zone, start_page, end_page, migratetype, |
2365 | num_movable); | |
c361be55 MG |
2366 | } |
2367 | ||
2f66a68f MG |
2368 | static void change_pageblock_range(struct page *pageblock_page, |
2369 | int start_order, int migratetype) | |
2370 | { | |
2371 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2372 | ||
2373 | while (nr_pageblocks--) { | |
2374 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2375 | pageblock_page += pageblock_nr_pages; | |
2376 | } | |
2377 | } | |
2378 | ||
fef903ef | 2379 | /* |
9c0415eb VB |
2380 | * When we are falling back to another migratetype during allocation, try to |
2381 | * steal extra free pages from the same pageblocks to satisfy further | |
2382 | * allocations, instead of polluting multiple pageblocks. | |
2383 | * | |
2384 | * If we are stealing a relatively large buddy page, it is likely there will | |
2385 | * be more free pages in the pageblock, so try to steal them all. For | |
2386 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2387 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2388 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2389 | * pageblocks. | |
fef903ef | 2390 | */ |
4eb7dce6 JK |
2391 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2392 | { | |
2393 | /* | |
2394 | * Leaving this order check is intended, although there is | |
2395 | * relaxed order check in next check. The reason is that | |
2396 | * we can actually steal whole pageblock if this condition met, | |
2397 | * but, below check doesn't guarantee it and that is just heuristic | |
2398 | * so could be changed anytime. | |
2399 | */ | |
2400 | if (order >= pageblock_order) | |
2401 | return true; | |
2402 | ||
2403 | if (order >= pageblock_order / 2 || | |
2404 | start_mt == MIGRATE_RECLAIMABLE || | |
2405 | start_mt == MIGRATE_UNMOVABLE || | |
2406 | page_group_by_mobility_disabled) | |
2407 | return true; | |
2408 | ||
2409 | return false; | |
2410 | } | |
2411 | ||
1c30844d MG |
2412 | static inline void boost_watermark(struct zone *zone) |
2413 | { | |
2414 | unsigned long max_boost; | |
2415 | ||
2416 | if (!watermark_boost_factor) | |
2417 | return; | |
14f69140 HW |
2418 | /* |
2419 | * Don't bother in zones that are unlikely to produce results. | |
2420 | * On small machines, including kdump capture kernels running | |
2421 | * in a small area, boosting the watermark can cause an out of | |
2422 | * memory situation immediately. | |
2423 | */ | |
2424 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
2425 | return; | |
1c30844d MG |
2426 | |
2427 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2428 | watermark_boost_factor, 10000); | |
94b3334c MG |
2429 | |
2430 | /* | |
2431 | * high watermark may be uninitialised if fragmentation occurs | |
2432 | * very early in boot so do not boost. We do not fall | |
2433 | * through and boost by pageblock_nr_pages as failing | |
2434 | * allocations that early means that reclaim is not going | |
2435 | * to help and it may even be impossible to reclaim the | |
2436 | * boosted watermark resulting in a hang. | |
2437 | */ | |
2438 | if (!max_boost) | |
2439 | return; | |
2440 | ||
1c30844d MG |
2441 | max_boost = max(pageblock_nr_pages, max_boost); |
2442 | ||
2443 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2444 | max_boost); | |
2445 | } | |
2446 | ||
4eb7dce6 JK |
2447 | /* |
2448 | * This function implements actual steal behaviour. If order is large enough, | |
2449 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2450 | * pageblock to our migratetype and determine how many already-allocated pages |
2451 | * are there in the pageblock with a compatible migratetype. If at least half | |
2452 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2453 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2454 | */ |
2455 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2456 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2457 | { |
d00181b9 | 2458 | unsigned int current_order = page_order(page); |
02aa0cdd VB |
2459 | int free_pages, movable_pages, alike_pages; |
2460 | int old_block_type; | |
2461 | ||
2462 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2463 | |
3bc48f96 VB |
2464 | /* |
2465 | * This can happen due to races and we want to prevent broken | |
2466 | * highatomic accounting. | |
2467 | */ | |
02aa0cdd | 2468 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2469 | goto single_page; |
2470 | ||
fef903ef SB |
2471 | /* Take ownership for orders >= pageblock_order */ |
2472 | if (current_order >= pageblock_order) { | |
2473 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2474 | goto single_page; |
fef903ef SB |
2475 | } |
2476 | ||
1c30844d MG |
2477 | /* |
2478 | * Boost watermarks to increase reclaim pressure to reduce the | |
2479 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2480 | * may be balanced overall and kswapd will not wake naturally. | |
2481 | */ | |
2482 | boost_watermark(zone); | |
2483 | if (alloc_flags & ALLOC_KSWAPD) | |
73444bc4 | 2484 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2485 | |
3bc48f96 VB |
2486 | /* We are not allowed to try stealing from the whole block */ |
2487 | if (!whole_block) | |
2488 | goto single_page; | |
2489 | ||
02aa0cdd VB |
2490 | free_pages = move_freepages_block(zone, page, start_type, |
2491 | &movable_pages); | |
2492 | /* | |
2493 | * Determine how many pages are compatible with our allocation. | |
2494 | * For movable allocation, it's the number of movable pages which | |
2495 | * we just obtained. For other types it's a bit more tricky. | |
2496 | */ | |
2497 | if (start_type == MIGRATE_MOVABLE) { | |
2498 | alike_pages = movable_pages; | |
2499 | } else { | |
2500 | /* | |
2501 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2502 | * to MOVABLE pageblock, consider all non-movable pages as | |
2503 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2504 | * vice versa, be conservative since we can't distinguish the | |
2505 | * exact migratetype of non-movable pages. | |
2506 | */ | |
2507 | if (old_block_type == MIGRATE_MOVABLE) | |
2508 | alike_pages = pageblock_nr_pages | |
2509 | - (free_pages + movable_pages); | |
2510 | else | |
2511 | alike_pages = 0; | |
2512 | } | |
2513 | ||
3bc48f96 | 2514 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2515 | if (!free_pages) |
3bc48f96 | 2516 | goto single_page; |
fef903ef | 2517 | |
02aa0cdd VB |
2518 | /* |
2519 | * If a sufficient number of pages in the block are either free or of | |
2520 | * comparable migratability as our allocation, claim the whole block. | |
2521 | */ | |
2522 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2523 | page_group_by_mobility_disabled) |
2524 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2525 | |
2526 | return; | |
2527 | ||
2528 | single_page: | |
6ab01363 | 2529 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
2530 | } |
2531 | ||
2149cdae JK |
2532 | /* |
2533 | * Check whether there is a suitable fallback freepage with requested order. | |
2534 | * If only_stealable is true, this function returns fallback_mt only if | |
2535 | * we can steal other freepages all together. This would help to reduce | |
2536 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2537 | */ | |
2538 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2539 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2540 | { |
2541 | int i; | |
2542 | int fallback_mt; | |
2543 | ||
2544 | if (area->nr_free == 0) | |
2545 | return -1; | |
2546 | ||
2547 | *can_steal = false; | |
2548 | for (i = 0;; i++) { | |
2549 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2550 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2551 | break; |
2552 | ||
b03641af | 2553 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2554 | continue; |
fef903ef | 2555 | |
4eb7dce6 JK |
2556 | if (can_steal_fallback(order, migratetype)) |
2557 | *can_steal = true; | |
2558 | ||
2149cdae JK |
2559 | if (!only_stealable) |
2560 | return fallback_mt; | |
2561 | ||
2562 | if (*can_steal) | |
2563 | return fallback_mt; | |
fef903ef | 2564 | } |
4eb7dce6 JK |
2565 | |
2566 | return -1; | |
fef903ef SB |
2567 | } |
2568 | ||
0aaa29a5 MG |
2569 | /* |
2570 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2571 | * there are no empty page blocks that contain a page with a suitable order | |
2572 | */ | |
2573 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2574 | unsigned int alloc_order) | |
2575 | { | |
2576 | int mt; | |
2577 | unsigned long max_managed, flags; | |
2578 | ||
2579 | /* | |
2580 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2581 | * Check is race-prone but harmless. | |
2582 | */ | |
9705bea5 | 2583 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2584 | if (zone->nr_reserved_highatomic >= max_managed) |
2585 | return; | |
2586 | ||
2587 | spin_lock_irqsave(&zone->lock, flags); | |
2588 | ||
2589 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2590 | if (zone->nr_reserved_highatomic >= max_managed) | |
2591 | goto out_unlock; | |
2592 | ||
2593 | /* Yoink! */ | |
2594 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2595 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2596 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2597 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2598 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2599 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2600 | } |
2601 | ||
2602 | out_unlock: | |
2603 | spin_unlock_irqrestore(&zone->lock, flags); | |
2604 | } | |
2605 | ||
2606 | /* | |
2607 | * Used when an allocation is about to fail under memory pressure. This | |
2608 | * potentially hurts the reliability of high-order allocations when under | |
2609 | * intense memory pressure but failed atomic allocations should be easier | |
2610 | * to recover from than an OOM. | |
29fac03b MK |
2611 | * |
2612 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2613 | * pageblock is exhausted. | |
0aaa29a5 | 2614 | */ |
29fac03b MK |
2615 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2616 | bool force) | |
0aaa29a5 MG |
2617 | { |
2618 | struct zonelist *zonelist = ac->zonelist; | |
2619 | unsigned long flags; | |
2620 | struct zoneref *z; | |
2621 | struct zone *zone; | |
2622 | struct page *page; | |
2623 | int order; | |
04c8716f | 2624 | bool ret; |
0aaa29a5 | 2625 | |
97a225e6 | 2626 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2627 | ac->nodemask) { |
29fac03b MK |
2628 | /* |
2629 | * Preserve at least one pageblock unless memory pressure | |
2630 | * is really high. | |
2631 | */ | |
2632 | if (!force && zone->nr_reserved_highatomic <= | |
2633 | pageblock_nr_pages) | |
0aaa29a5 MG |
2634 | continue; |
2635 | ||
2636 | spin_lock_irqsave(&zone->lock, flags); | |
2637 | for (order = 0; order < MAX_ORDER; order++) { | |
2638 | struct free_area *area = &(zone->free_area[order]); | |
2639 | ||
b03641af | 2640 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2641 | if (!page) |
0aaa29a5 MG |
2642 | continue; |
2643 | ||
0aaa29a5 | 2644 | /* |
4855e4a7 MK |
2645 | * In page freeing path, migratetype change is racy so |
2646 | * we can counter several free pages in a pageblock | |
2647 | * in this loop althoug we changed the pageblock type | |
2648 | * from highatomic to ac->migratetype. So we should | |
2649 | * adjust the count once. | |
0aaa29a5 | 2650 | */ |
a6ffdc07 | 2651 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2652 | /* |
2653 | * It should never happen but changes to | |
2654 | * locking could inadvertently allow a per-cpu | |
2655 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2656 | * while unreserving so be safe and watch for | |
2657 | * underflows. | |
2658 | */ | |
2659 | zone->nr_reserved_highatomic -= min( | |
2660 | pageblock_nr_pages, | |
2661 | zone->nr_reserved_highatomic); | |
2662 | } | |
0aaa29a5 MG |
2663 | |
2664 | /* | |
2665 | * Convert to ac->migratetype and avoid the normal | |
2666 | * pageblock stealing heuristics. Minimally, the caller | |
2667 | * is doing the work and needs the pages. More | |
2668 | * importantly, if the block was always converted to | |
2669 | * MIGRATE_UNMOVABLE or another type then the number | |
2670 | * of pageblocks that cannot be completely freed | |
2671 | * may increase. | |
2672 | */ | |
2673 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2674 | ret = move_freepages_block(zone, page, ac->migratetype, |
2675 | NULL); | |
29fac03b MK |
2676 | if (ret) { |
2677 | spin_unlock_irqrestore(&zone->lock, flags); | |
2678 | return ret; | |
2679 | } | |
0aaa29a5 MG |
2680 | } |
2681 | spin_unlock_irqrestore(&zone->lock, flags); | |
2682 | } | |
04c8716f MK |
2683 | |
2684 | return false; | |
0aaa29a5 MG |
2685 | } |
2686 | ||
3bc48f96 VB |
2687 | /* |
2688 | * Try finding a free buddy page on the fallback list and put it on the free | |
2689 | * list of requested migratetype, possibly along with other pages from the same | |
2690 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2691 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2692 | * |
2693 | * The use of signed ints for order and current_order is a deliberate | |
2694 | * deviation from the rest of this file, to make the for loop | |
2695 | * condition simpler. | |
3bc48f96 | 2696 | */ |
85ccc8fa | 2697 | static __always_inline bool |
6bb15450 MG |
2698 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2699 | unsigned int alloc_flags) | |
b2a0ac88 | 2700 | { |
b8af2941 | 2701 | struct free_area *area; |
b002529d | 2702 | int current_order; |
6bb15450 | 2703 | int min_order = order; |
b2a0ac88 | 2704 | struct page *page; |
4eb7dce6 JK |
2705 | int fallback_mt; |
2706 | bool can_steal; | |
b2a0ac88 | 2707 | |
6bb15450 MG |
2708 | /* |
2709 | * Do not steal pages from freelists belonging to other pageblocks | |
2710 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2711 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2712 | */ | |
2713 | if (alloc_flags & ALLOC_NOFRAGMENT) | |
2714 | min_order = pageblock_order; | |
2715 | ||
7a8f58f3 VB |
2716 | /* |
2717 | * Find the largest available free page in the other list. This roughly | |
2718 | * approximates finding the pageblock with the most free pages, which | |
2719 | * would be too costly to do exactly. | |
2720 | */ | |
6bb15450 | 2721 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 2722 | --current_order) { |
4eb7dce6 JK |
2723 | area = &(zone->free_area[current_order]); |
2724 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2725 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2726 | if (fallback_mt == -1) |
2727 | continue; | |
b2a0ac88 | 2728 | |
7a8f58f3 VB |
2729 | /* |
2730 | * We cannot steal all free pages from the pageblock and the | |
2731 | * requested migratetype is movable. In that case it's better to | |
2732 | * steal and split the smallest available page instead of the | |
2733 | * largest available page, because even if the next movable | |
2734 | * allocation falls back into a different pageblock than this | |
2735 | * one, it won't cause permanent fragmentation. | |
2736 | */ | |
2737 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2738 | && current_order > order) | |
2739 | goto find_smallest; | |
b2a0ac88 | 2740 | |
7a8f58f3 VB |
2741 | goto do_steal; |
2742 | } | |
e0fff1bd | 2743 | |
7a8f58f3 | 2744 | return false; |
e0fff1bd | 2745 | |
7a8f58f3 VB |
2746 | find_smallest: |
2747 | for (current_order = order; current_order < MAX_ORDER; | |
2748 | current_order++) { | |
2749 | area = &(zone->free_area[current_order]); | |
2750 | fallback_mt = find_suitable_fallback(area, current_order, | |
2751 | start_migratetype, false, &can_steal); | |
2752 | if (fallback_mt != -1) | |
2753 | break; | |
b2a0ac88 MG |
2754 | } |
2755 | ||
7a8f58f3 VB |
2756 | /* |
2757 | * This should not happen - we already found a suitable fallback | |
2758 | * when looking for the largest page. | |
2759 | */ | |
2760 | VM_BUG_ON(current_order == MAX_ORDER); | |
2761 | ||
2762 | do_steal: | |
b03641af | 2763 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2764 | |
1c30844d MG |
2765 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
2766 | can_steal); | |
7a8f58f3 VB |
2767 | |
2768 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2769 | start_migratetype, fallback_mt); | |
2770 | ||
2771 | return true; | |
2772 | ||
b2a0ac88 MG |
2773 | } |
2774 | ||
56fd56b8 | 2775 | /* |
1da177e4 LT |
2776 | * Do the hard work of removing an element from the buddy allocator. |
2777 | * Call me with the zone->lock already held. | |
2778 | */ | |
85ccc8fa | 2779 | static __always_inline struct page * |
6bb15450 MG |
2780 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2781 | unsigned int alloc_flags) | |
1da177e4 | 2782 | { |
1da177e4 LT |
2783 | struct page *page; |
2784 | ||
16867664 RG |
2785 | #ifdef CONFIG_CMA |
2786 | /* | |
2787 | * Balance movable allocations between regular and CMA areas by | |
2788 | * allocating from CMA when over half of the zone's free memory | |
2789 | * is in the CMA area. | |
2790 | */ | |
2791 | if (migratetype == MIGRATE_MOVABLE && | |
2792 | zone_page_state(zone, NR_FREE_CMA_PAGES) > | |
2793 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
2794 | page = __rmqueue_cma_fallback(zone, order); | |
2795 | if (page) | |
2796 | return page; | |
2797 | } | |
2798 | #endif | |
3bc48f96 | 2799 | retry: |
56fd56b8 | 2800 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2801 | if (unlikely(!page)) { |
dc67647b JK |
2802 | if (migratetype == MIGRATE_MOVABLE) |
2803 | page = __rmqueue_cma_fallback(zone, order); | |
2804 | ||
6bb15450 MG |
2805 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2806 | alloc_flags)) | |
3bc48f96 | 2807 | goto retry; |
728ec980 MG |
2808 | } |
2809 | ||
0d3d062a | 2810 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2811 | return page; |
1da177e4 LT |
2812 | } |
2813 | ||
5f63b720 | 2814 | /* |
1da177e4 LT |
2815 | * Obtain a specified number of elements from the buddy allocator, all under |
2816 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2817 | * Returns the number of new pages which were placed at *list. | |
2818 | */ | |
5f63b720 | 2819 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2820 | unsigned long count, struct list_head *list, |
6bb15450 | 2821 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2822 | { |
a6de734b | 2823 | int i, alloced = 0; |
5f63b720 | 2824 | |
d34b0733 | 2825 | spin_lock(&zone->lock); |
1da177e4 | 2826 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2827 | struct page *page = __rmqueue(zone, order, migratetype, |
2828 | alloc_flags); | |
085cc7d5 | 2829 | if (unlikely(page == NULL)) |
1da177e4 | 2830 | break; |
81eabcbe | 2831 | |
479f854a MG |
2832 | if (unlikely(check_pcp_refill(page))) |
2833 | continue; | |
2834 | ||
81eabcbe | 2835 | /* |
0fac3ba5 VB |
2836 | * Split buddy pages returned by expand() are received here in |
2837 | * physical page order. The page is added to the tail of | |
2838 | * caller's list. From the callers perspective, the linked list | |
2839 | * is ordered by page number under some conditions. This is | |
2840 | * useful for IO devices that can forward direction from the | |
2841 | * head, thus also in the physical page order. This is useful | |
2842 | * for IO devices that can merge IO requests if the physical | |
2843 | * pages are ordered properly. | |
81eabcbe | 2844 | */ |
0fac3ba5 | 2845 | list_add_tail(&page->lru, list); |
a6de734b | 2846 | alloced++; |
bb14c2c7 | 2847 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2848 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2849 | -(1 << order)); | |
1da177e4 | 2850 | } |
a6de734b MG |
2851 | |
2852 | /* | |
2853 | * i pages were removed from the buddy list even if some leak due | |
2854 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2855 | * on i. Do not confuse with 'alloced' which is the number of | |
2856 | * pages added to the pcp list. | |
2857 | */ | |
f2260e6b | 2858 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2859 | spin_unlock(&zone->lock); |
a6de734b | 2860 | return alloced; |
1da177e4 LT |
2861 | } |
2862 | ||
4ae7c039 | 2863 | #ifdef CONFIG_NUMA |
8fce4d8e | 2864 | /* |
4037d452 CL |
2865 | * Called from the vmstat counter updater to drain pagesets of this |
2866 | * currently executing processor on remote nodes after they have | |
2867 | * expired. | |
2868 | * | |
879336c3 CL |
2869 | * Note that this function must be called with the thread pinned to |
2870 | * a single processor. | |
8fce4d8e | 2871 | */ |
4037d452 | 2872 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2873 | { |
4ae7c039 | 2874 | unsigned long flags; |
7be12fc9 | 2875 | int to_drain, batch; |
4ae7c039 | 2876 | |
4037d452 | 2877 | local_irq_save(flags); |
4db0c3c2 | 2878 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2879 | to_drain = min(pcp->count, batch); |
77ba9062 | 2880 | if (to_drain > 0) |
2a13515c | 2881 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 2882 | local_irq_restore(flags); |
4ae7c039 CL |
2883 | } |
2884 | #endif | |
2885 | ||
9f8f2172 | 2886 | /* |
93481ff0 | 2887 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2888 | * |
2889 | * The processor must either be the current processor and the | |
2890 | * thread pinned to the current processor or a processor that | |
2891 | * is not online. | |
2892 | */ | |
93481ff0 | 2893 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2894 | { |
c54ad30c | 2895 | unsigned long flags; |
93481ff0 VB |
2896 | struct per_cpu_pageset *pset; |
2897 | struct per_cpu_pages *pcp; | |
1da177e4 | 2898 | |
93481ff0 VB |
2899 | local_irq_save(flags); |
2900 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2901 | |
93481ff0 | 2902 | pcp = &pset->pcp; |
77ba9062 | 2903 | if (pcp->count) |
93481ff0 | 2904 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
2905 | local_irq_restore(flags); |
2906 | } | |
3dfa5721 | 2907 | |
93481ff0 VB |
2908 | /* |
2909 | * Drain pcplists of all zones on the indicated processor. | |
2910 | * | |
2911 | * The processor must either be the current processor and the | |
2912 | * thread pinned to the current processor or a processor that | |
2913 | * is not online. | |
2914 | */ | |
2915 | static void drain_pages(unsigned int cpu) | |
2916 | { | |
2917 | struct zone *zone; | |
2918 | ||
2919 | for_each_populated_zone(zone) { | |
2920 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2921 | } |
2922 | } | |
1da177e4 | 2923 | |
9f8f2172 CL |
2924 | /* |
2925 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2926 | * |
2927 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2928 | * the single zone's pages. | |
9f8f2172 | 2929 | */ |
93481ff0 | 2930 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2931 | { |
93481ff0 VB |
2932 | int cpu = smp_processor_id(); |
2933 | ||
2934 | if (zone) | |
2935 | drain_pages_zone(cpu, zone); | |
2936 | else | |
2937 | drain_pages(cpu); | |
9f8f2172 CL |
2938 | } |
2939 | ||
0ccce3b9 MG |
2940 | static void drain_local_pages_wq(struct work_struct *work) |
2941 | { | |
d9367bd0 WY |
2942 | struct pcpu_drain *drain; |
2943 | ||
2944 | drain = container_of(work, struct pcpu_drain, work); | |
2945 | ||
a459eeb7 MH |
2946 | /* |
2947 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2948 | * we can race with cpu offline when the WQ can move this from | |
2949 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2950 | * cpu which is allright but we also have to make sure to not move to | |
2951 | * a different one. | |
2952 | */ | |
2953 | preempt_disable(); | |
d9367bd0 | 2954 | drain_local_pages(drain->zone); |
a459eeb7 | 2955 | preempt_enable(); |
0ccce3b9 MG |
2956 | } |
2957 | ||
9f8f2172 | 2958 | /* |
74046494 GBY |
2959 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2960 | * | |
93481ff0 VB |
2961 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2962 | * | |
0ccce3b9 | 2963 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2964 | */ |
93481ff0 | 2965 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2966 | { |
74046494 | 2967 | int cpu; |
74046494 GBY |
2968 | |
2969 | /* | |
2970 | * Allocate in the BSS so we wont require allocation in | |
2971 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2972 | */ | |
2973 | static cpumask_t cpus_with_pcps; | |
2974 | ||
ce612879 MH |
2975 | /* |
2976 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2977 | * initialized. | |
2978 | */ | |
2979 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2980 | return; | |
2981 | ||
bd233f53 MG |
2982 | /* |
2983 | * Do not drain if one is already in progress unless it's specific to | |
2984 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2985 | * the drain to be complete when the call returns. | |
2986 | */ | |
2987 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2988 | if (!zone) | |
2989 | return; | |
2990 | mutex_lock(&pcpu_drain_mutex); | |
2991 | } | |
0ccce3b9 | 2992 | |
74046494 GBY |
2993 | /* |
2994 | * We don't care about racing with CPU hotplug event | |
2995 | * as offline notification will cause the notified | |
2996 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2997 | * disables preemption as part of its processing | |
2998 | */ | |
2999 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
3000 | struct per_cpu_pageset *pcp; |
3001 | struct zone *z; | |
74046494 | 3002 | bool has_pcps = false; |
93481ff0 VB |
3003 | |
3004 | if (zone) { | |
74046494 | 3005 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 3006 | if (pcp->pcp.count) |
74046494 | 3007 | has_pcps = true; |
93481ff0 VB |
3008 | } else { |
3009 | for_each_populated_zone(z) { | |
3010 | pcp = per_cpu_ptr(z->pageset, cpu); | |
3011 | if (pcp->pcp.count) { | |
3012 | has_pcps = true; | |
3013 | break; | |
3014 | } | |
74046494 GBY |
3015 | } |
3016 | } | |
93481ff0 | 3017 | |
74046494 GBY |
3018 | if (has_pcps) |
3019 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
3020 | else | |
3021 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
3022 | } | |
0ccce3b9 | 3023 | |
bd233f53 | 3024 | for_each_cpu(cpu, &cpus_with_pcps) { |
d9367bd0 WY |
3025 | struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu); |
3026 | ||
3027 | drain->zone = zone; | |
3028 | INIT_WORK(&drain->work, drain_local_pages_wq); | |
3029 | queue_work_on(cpu, mm_percpu_wq, &drain->work); | |
0ccce3b9 | 3030 | } |
bd233f53 | 3031 | for_each_cpu(cpu, &cpus_with_pcps) |
d9367bd0 | 3032 | flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work); |
bd233f53 MG |
3033 | |
3034 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
3035 | } |
3036 | ||
296699de | 3037 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3038 | |
556b969a CY |
3039 | /* |
3040 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
3041 | */ | |
3042 | #define WD_PAGE_COUNT (128*1024) | |
3043 | ||
1da177e4 LT |
3044 | void mark_free_pages(struct zone *zone) |
3045 | { | |
556b969a | 3046 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 3047 | unsigned long flags; |
7aeb09f9 | 3048 | unsigned int order, t; |
86760a2c | 3049 | struct page *page; |
1da177e4 | 3050 | |
8080fc03 | 3051 | if (zone_is_empty(zone)) |
1da177e4 LT |
3052 | return; |
3053 | ||
3054 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 3055 | |
108bcc96 | 3056 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
3057 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
3058 | if (pfn_valid(pfn)) { | |
86760a2c | 3059 | page = pfn_to_page(pfn); |
ba6b0979 | 3060 | |
556b969a CY |
3061 | if (!--page_count) { |
3062 | touch_nmi_watchdog(); | |
3063 | page_count = WD_PAGE_COUNT; | |
3064 | } | |
3065 | ||
ba6b0979 JK |
3066 | if (page_zone(page) != zone) |
3067 | continue; | |
3068 | ||
7be98234 RW |
3069 | if (!swsusp_page_is_forbidden(page)) |
3070 | swsusp_unset_page_free(page); | |
f623f0db | 3071 | } |
1da177e4 | 3072 | |
b2a0ac88 | 3073 | for_each_migratetype_order(order, t) { |
86760a2c GT |
3074 | list_for_each_entry(page, |
3075 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 3076 | unsigned long i; |
1da177e4 | 3077 | |
86760a2c | 3078 | pfn = page_to_pfn(page); |
556b969a CY |
3079 | for (i = 0; i < (1UL << order); i++) { |
3080 | if (!--page_count) { | |
3081 | touch_nmi_watchdog(); | |
3082 | page_count = WD_PAGE_COUNT; | |
3083 | } | |
7be98234 | 3084 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 3085 | } |
f623f0db | 3086 | } |
b2a0ac88 | 3087 | } |
1da177e4 LT |
3088 | spin_unlock_irqrestore(&zone->lock, flags); |
3089 | } | |
e2c55dc8 | 3090 | #endif /* CONFIG_PM */ |
1da177e4 | 3091 | |
2d4894b5 | 3092 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 3093 | { |
5f8dcc21 | 3094 | int migratetype; |
1da177e4 | 3095 | |
4db7548c | 3096 | if (!free_pcp_prepare(page)) |
9cca35d4 | 3097 | return false; |
689bcebf | 3098 | |
dc4b0caf | 3099 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 3100 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
3101 | return true; |
3102 | } | |
3103 | ||
2d4894b5 | 3104 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
3105 | { |
3106 | struct zone *zone = page_zone(page); | |
3107 | struct per_cpu_pages *pcp; | |
3108 | int migratetype; | |
3109 | ||
3110 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 3111 | __count_vm_event(PGFREE); |
da456f14 | 3112 | |
5f8dcc21 MG |
3113 | /* |
3114 | * We only track unmovable, reclaimable and movable on pcp lists. | |
3115 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 3116 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
3117 | * areas back if necessary. Otherwise, we may have to free |
3118 | * excessively into the page allocator | |
3119 | */ | |
3120 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 3121 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 3122 | free_one_page(zone, page, pfn, 0, migratetype); |
9cca35d4 | 3123 | return; |
5f8dcc21 MG |
3124 | } |
3125 | migratetype = MIGRATE_MOVABLE; | |
3126 | } | |
3127 | ||
99dcc3e5 | 3128 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 3129 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 3130 | pcp->count++; |
48db57f8 | 3131 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 3132 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb | 3133 | free_pcppages_bulk(zone, batch, pcp); |
48db57f8 | 3134 | } |
9cca35d4 | 3135 | } |
5f8dcc21 | 3136 | |
9cca35d4 MG |
3137 | /* |
3138 | * Free a 0-order page | |
9cca35d4 | 3139 | */ |
2d4894b5 | 3140 | void free_unref_page(struct page *page) |
9cca35d4 MG |
3141 | { |
3142 | unsigned long flags; | |
3143 | unsigned long pfn = page_to_pfn(page); | |
3144 | ||
2d4894b5 | 3145 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3146 | return; |
3147 | ||
3148 | local_irq_save(flags); | |
2d4894b5 | 3149 | free_unref_page_commit(page, pfn); |
d34b0733 | 3150 | local_irq_restore(flags); |
1da177e4 LT |
3151 | } |
3152 | ||
cc59850e KK |
3153 | /* |
3154 | * Free a list of 0-order pages | |
3155 | */ | |
2d4894b5 | 3156 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
3157 | { |
3158 | struct page *page, *next; | |
9cca35d4 | 3159 | unsigned long flags, pfn; |
c24ad77d | 3160 | int batch_count = 0; |
9cca35d4 MG |
3161 | |
3162 | /* Prepare pages for freeing */ | |
3163 | list_for_each_entry_safe(page, next, list, lru) { | |
3164 | pfn = page_to_pfn(page); | |
2d4894b5 | 3165 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3166 | list_del(&page->lru); |
3167 | set_page_private(page, pfn); | |
3168 | } | |
cc59850e | 3169 | |
9cca35d4 | 3170 | local_irq_save(flags); |
cc59850e | 3171 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
3172 | unsigned long pfn = page_private(page); |
3173 | ||
3174 | set_page_private(page, 0); | |
2d4894b5 MG |
3175 | trace_mm_page_free_batched(page); |
3176 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
3177 | |
3178 | /* | |
3179 | * Guard against excessive IRQ disabled times when we get | |
3180 | * a large list of pages to free. | |
3181 | */ | |
3182 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
3183 | local_irq_restore(flags); | |
3184 | batch_count = 0; | |
3185 | local_irq_save(flags); | |
3186 | } | |
cc59850e | 3187 | } |
9cca35d4 | 3188 | local_irq_restore(flags); |
cc59850e KK |
3189 | } |
3190 | ||
8dfcc9ba NP |
3191 | /* |
3192 | * split_page takes a non-compound higher-order page, and splits it into | |
3193 | * n (1<<order) sub-pages: page[0..n] | |
3194 | * Each sub-page must be freed individually. | |
3195 | * | |
3196 | * Note: this is probably too low level an operation for use in drivers. | |
3197 | * Please consult with lkml before using this in your driver. | |
3198 | */ | |
3199 | void split_page(struct page *page, unsigned int order) | |
3200 | { | |
3201 | int i; | |
3202 | ||
309381fe SL |
3203 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3204 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3205 | |
a9627bc5 | 3206 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3207 | set_page_refcounted(page + i); |
a9627bc5 | 3208 | split_page_owner(page, order); |
8dfcc9ba | 3209 | } |
5853ff23 | 3210 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3211 | |
3c605096 | 3212 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3213 | { |
748446bb MG |
3214 | unsigned long watermark; |
3215 | struct zone *zone; | |
2139cbe6 | 3216 | int mt; |
748446bb MG |
3217 | |
3218 | BUG_ON(!PageBuddy(page)); | |
3219 | ||
3220 | zone = page_zone(page); | |
2e30abd1 | 3221 | mt = get_pageblock_migratetype(page); |
748446bb | 3222 | |
194159fb | 3223 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
3224 | /* |
3225 | * Obey watermarks as if the page was being allocated. We can | |
3226 | * emulate a high-order watermark check with a raised order-0 | |
3227 | * watermark, because we already know our high-order page | |
3228 | * exists. | |
3229 | */ | |
fd1444b2 | 3230 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3231 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3232 | return 0; |
3233 | ||
8fb74b9f | 3234 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3235 | } |
748446bb MG |
3236 | |
3237 | /* Remove page from free list */ | |
b03641af | 3238 | |
6ab01363 | 3239 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 3240 | |
400bc7fd | 3241 | /* |
3242 | * Set the pageblock if the isolated page is at least half of a | |
3243 | * pageblock | |
3244 | */ | |
748446bb MG |
3245 | if (order >= pageblock_order - 1) { |
3246 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3247 | for (; page < endpage; page += pageblock_nr_pages) { |
3248 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 3249 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 3250 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
3251 | set_pageblock_migratetype(page, |
3252 | MIGRATE_MOVABLE); | |
3253 | } | |
748446bb MG |
3254 | } |
3255 | ||
f3a14ced | 3256 | |
8fb74b9f | 3257 | return 1UL << order; |
1fb3f8ca MG |
3258 | } |
3259 | ||
624f58d8 AD |
3260 | /** |
3261 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
3262 | * @page: Page that was isolated | |
3263 | * @order: Order of the isolated page | |
e6a0a7ad | 3264 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
3265 | * |
3266 | * This function is meant to return a page pulled from the free lists via | |
3267 | * __isolate_free_page back to the free lists they were pulled from. | |
3268 | */ | |
3269 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
3270 | { | |
3271 | struct zone *zone = page_zone(page); | |
3272 | ||
3273 | /* zone lock should be held when this function is called */ | |
3274 | lockdep_assert_held(&zone->lock); | |
3275 | ||
3276 | /* Return isolated page to tail of freelist. */ | |
36e66c55 | 3277 | __free_one_page(page, page_to_pfn(page), zone, order, mt, false); |
624f58d8 AD |
3278 | } |
3279 | ||
060e7417 MG |
3280 | /* |
3281 | * Update NUMA hit/miss statistics | |
3282 | * | |
3283 | * Must be called with interrupts disabled. | |
060e7417 | 3284 | */ |
41b6167e | 3285 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
3286 | { |
3287 | #ifdef CONFIG_NUMA | |
3a321d2a | 3288 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3289 | |
4518085e KW |
3290 | /* skip numa counters update if numa stats is disabled */ |
3291 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3292 | return; | |
3293 | ||
c1093b74 | 3294 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3295 | local_stat = NUMA_OTHER; |
060e7417 | 3296 | |
c1093b74 | 3297 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3a321d2a | 3298 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 3299 | else { |
3a321d2a KW |
3300 | __inc_numa_state(z, NUMA_MISS); |
3301 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 3302 | } |
3a321d2a | 3303 | __inc_numa_state(z, local_stat); |
060e7417 MG |
3304 | #endif |
3305 | } | |
3306 | ||
066b2393 MG |
3307 | /* Remove page from the per-cpu list, caller must protect the list */ |
3308 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
6bb15450 | 3309 | unsigned int alloc_flags, |
453f85d4 | 3310 | struct per_cpu_pages *pcp, |
066b2393 MG |
3311 | struct list_head *list) |
3312 | { | |
3313 | struct page *page; | |
3314 | ||
3315 | do { | |
3316 | if (list_empty(list)) { | |
3317 | pcp->count += rmqueue_bulk(zone, 0, | |
3318 | pcp->batch, list, | |
6bb15450 | 3319 | migratetype, alloc_flags); |
066b2393 MG |
3320 | if (unlikely(list_empty(list))) |
3321 | return NULL; | |
3322 | } | |
3323 | ||
453f85d4 | 3324 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
3325 | list_del(&page->lru); |
3326 | pcp->count--; | |
3327 | } while (check_new_pcp(page)); | |
3328 | ||
3329 | return page; | |
3330 | } | |
3331 | ||
3332 | /* Lock and remove page from the per-cpu list */ | |
3333 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
1c52e6d0 YS |
3334 | struct zone *zone, gfp_t gfp_flags, |
3335 | int migratetype, unsigned int alloc_flags) | |
066b2393 MG |
3336 | { |
3337 | struct per_cpu_pages *pcp; | |
3338 | struct list_head *list; | |
066b2393 | 3339 | struct page *page; |
d34b0733 | 3340 | unsigned long flags; |
066b2393 | 3341 | |
d34b0733 | 3342 | local_irq_save(flags); |
066b2393 MG |
3343 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3344 | list = &pcp->lists[migratetype]; | |
6bb15450 | 3345 | page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list); |
066b2393 | 3346 | if (page) { |
1c52e6d0 | 3347 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1); |
066b2393 MG |
3348 | zone_statistics(preferred_zone, zone); |
3349 | } | |
d34b0733 | 3350 | local_irq_restore(flags); |
066b2393 MG |
3351 | return page; |
3352 | } | |
3353 | ||
1da177e4 | 3354 | /* |
75379191 | 3355 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 3356 | */ |
0a15c3e9 | 3357 | static inline |
066b2393 | 3358 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3359 | struct zone *zone, unsigned int order, |
c603844b MG |
3360 | gfp_t gfp_flags, unsigned int alloc_flags, |
3361 | int migratetype) | |
1da177e4 LT |
3362 | { |
3363 | unsigned long flags; | |
689bcebf | 3364 | struct page *page; |
1da177e4 | 3365 | |
d34b0733 | 3366 | if (likely(order == 0)) { |
1c52e6d0 YS |
3367 | page = rmqueue_pcplist(preferred_zone, zone, gfp_flags, |
3368 | migratetype, alloc_flags); | |
066b2393 MG |
3369 | goto out; |
3370 | } | |
83b9355b | 3371 | |
066b2393 MG |
3372 | /* |
3373 | * We most definitely don't want callers attempting to | |
3374 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3375 | */ | |
3376 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3377 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3378 | |
066b2393 MG |
3379 | do { |
3380 | page = NULL; | |
3381 | if (alloc_flags & ALLOC_HARDER) { | |
3382 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3383 | if (page) | |
3384 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3385 | } | |
a74609fa | 3386 | if (!page) |
6bb15450 | 3387 | page = __rmqueue(zone, order, migratetype, alloc_flags); |
066b2393 MG |
3388 | } while (page && check_new_pages(page, order)); |
3389 | spin_unlock(&zone->lock); | |
3390 | if (!page) | |
3391 | goto failed; | |
3392 | __mod_zone_freepage_state(zone, -(1 << order), | |
3393 | get_pcppage_migratetype(page)); | |
1da177e4 | 3394 | |
16709d1d | 3395 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3396 | zone_statistics(preferred_zone, zone); |
a74609fa | 3397 | local_irq_restore(flags); |
1da177e4 | 3398 | |
066b2393 | 3399 | out: |
73444bc4 MG |
3400 | /* Separate test+clear to avoid unnecessary atomics */ |
3401 | if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) { | |
3402 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); | |
3403 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3404 | } | |
3405 | ||
066b2393 | 3406 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 | 3407 | return page; |
a74609fa NP |
3408 | |
3409 | failed: | |
3410 | local_irq_restore(flags); | |
a74609fa | 3411 | return NULL; |
1da177e4 LT |
3412 | } |
3413 | ||
933e312e AM |
3414 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3415 | ||
b2588c4b | 3416 | static struct { |
933e312e AM |
3417 | struct fault_attr attr; |
3418 | ||
621a5f7a | 3419 | bool ignore_gfp_highmem; |
71baba4b | 3420 | bool ignore_gfp_reclaim; |
54114994 | 3421 | u32 min_order; |
933e312e AM |
3422 | } fail_page_alloc = { |
3423 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3424 | .ignore_gfp_reclaim = true, |
621a5f7a | 3425 | .ignore_gfp_highmem = true, |
54114994 | 3426 | .min_order = 1, |
933e312e AM |
3427 | }; |
3428 | ||
3429 | static int __init setup_fail_page_alloc(char *str) | |
3430 | { | |
3431 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3432 | } | |
3433 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3434 | ||
af3b8544 | 3435 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3436 | { |
54114994 | 3437 | if (order < fail_page_alloc.min_order) |
deaf386e | 3438 | return false; |
933e312e | 3439 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3440 | return false; |
933e312e | 3441 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3442 | return false; |
71baba4b MG |
3443 | if (fail_page_alloc.ignore_gfp_reclaim && |
3444 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3445 | return false; |
933e312e AM |
3446 | |
3447 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3448 | } | |
3449 | ||
3450 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3451 | ||
3452 | static int __init fail_page_alloc_debugfs(void) | |
3453 | { | |
0825a6f9 | 3454 | umode_t mode = S_IFREG | 0600; |
933e312e | 3455 | struct dentry *dir; |
933e312e | 3456 | |
dd48c085 AM |
3457 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3458 | &fail_page_alloc.attr); | |
b2588c4b | 3459 | |
d9f7979c GKH |
3460 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3461 | &fail_page_alloc.ignore_gfp_reclaim); | |
3462 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3463 | &fail_page_alloc.ignore_gfp_highmem); | |
3464 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3465 | |
d9f7979c | 3466 | return 0; |
933e312e AM |
3467 | } |
3468 | ||
3469 | late_initcall(fail_page_alloc_debugfs); | |
3470 | ||
3471 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3472 | ||
3473 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3474 | ||
af3b8544 | 3475 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3476 | { |
deaf386e | 3477 | return false; |
933e312e AM |
3478 | } |
3479 | ||
3480 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3481 | ||
af3b8544 BP |
3482 | static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
3483 | { | |
3484 | return __should_fail_alloc_page(gfp_mask, order); | |
3485 | } | |
3486 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3487 | ||
f27ce0e1 JK |
3488 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3489 | unsigned int order, unsigned int alloc_flags) | |
3490 | { | |
3491 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); | |
3492 | long unusable_free = (1 << order) - 1; | |
3493 | ||
3494 | /* | |
3495 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3496 | * the high-atomic reserves. This will over-estimate the size of the | |
3497 | * atomic reserve but it avoids a search. | |
3498 | */ | |
3499 | if (likely(!alloc_harder)) | |
3500 | unusable_free += z->nr_reserved_highatomic; | |
3501 | ||
3502 | #ifdef CONFIG_CMA | |
3503 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3504 | if (!(alloc_flags & ALLOC_CMA)) | |
3505 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3506 | #endif | |
3507 | ||
3508 | return unusable_free; | |
3509 | } | |
3510 | ||
1da177e4 | 3511 | /* |
97a16fc8 MG |
3512 | * Return true if free base pages are above 'mark'. For high-order checks it |
3513 | * will return true of the order-0 watermark is reached and there is at least | |
3514 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3515 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3516 | */ |
86a294a8 | 3517 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3518 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3519 | long free_pages) |
1da177e4 | 3520 | { |
d23ad423 | 3521 | long min = mark; |
1da177e4 | 3522 | int o; |
cd04ae1e | 3523 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3524 | |
0aaa29a5 | 3525 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 3526 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 3527 | |
7fb1d9fc | 3528 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3529 | min -= min / 2; |
0aaa29a5 | 3530 | |
f27ce0e1 | 3531 | if (unlikely(alloc_harder)) { |
cd04ae1e MH |
3532 | /* |
3533 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3534 | * users on the grounds that it's definitely going to be in | |
3535 | * the exit path shortly and free memory. Any allocation it | |
3536 | * makes during the free path will be small and short-lived. | |
3537 | */ | |
3538 | if (alloc_flags & ALLOC_OOM) | |
3539 | min -= min / 2; | |
3540 | else | |
3541 | min -= min / 4; | |
3542 | } | |
3543 | ||
97a16fc8 MG |
3544 | /* |
3545 | * Check watermarks for an order-0 allocation request. If these | |
3546 | * are not met, then a high-order request also cannot go ahead | |
3547 | * even if a suitable page happened to be free. | |
3548 | */ | |
97a225e6 | 3549 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 3550 | return false; |
1da177e4 | 3551 | |
97a16fc8 MG |
3552 | /* If this is an order-0 request then the watermark is fine */ |
3553 | if (!order) | |
3554 | return true; | |
3555 | ||
3556 | /* For a high-order request, check at least one suitable page is free */ | |
3557 | for (o = order; o < MAX_ORDER; o++) { | |
3558 | struct free_area *area = &z->free_area[o]; | |
3559 | int mt; | |
3560 | ||
3561 | if (!area->nr_free) | |
3562 | continue; | |
3563 | ||
97a16fc8 | 3564 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3565 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3566 | return true; |
3567 | } | |
3568 | ||
3569 | #ifdef CONFIG_CMA | |
d883c6cf | 3570 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3571 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3572 | return true; |
d883c6cf | 3573 | } |
97a16fc8 | 3574 | #endif |
76089d00 | 3575 | if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC)) |
b050e376 | 3576 | return true; |
1da177e4 | 3577 | } |
97a16fc8 | 3578 | return false; |
88f5acf8 MG |
3579 | } |
3580 | ||
7aeb09f9 | 3581 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3582 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 3583 | { |
97a225e6 | 3584 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
3585 | zone_page_state(z, NR_FREE_PAGES)); |
3586 | } | |
3587 | ||
48ee5f36 | 3588 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 3589 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 3590 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 3591 | { |
f27ce0e1 | 3592 | long free_pages; |
d883c6cf | 3593 | |
f27ce0e1 | 3594 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
3595 | |
3596 | /* | |
3597 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 3598 | * need to be calculated. |
48ee5f36 | 3599 | */ |
f27ce0e1 JK |
3600 | if (!order) { |
3601 | long fast_free; | |
3602 | ||
3603 | fast_free = free_pages; | |
3604 | fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags); | |
3605 | if (fast_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
3606 | return true; | |
3607 | } | |
48ee5f36 | 3608 | |
f80b08fc CTR |
3609 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
3610 | free_pages)) | |
3611 | return true; | |
3612 | /* | |
3613 | * Ignore watermark boosting for GFP_ATOMIC order-0 allocations | |
3614 | * when checking the min watermark. The min watermark is the | |
3615 | * point where boosting is ignored so that kswapd is woken up | |
3616 | * when below the low watermark. | |
3617 | */ | |
3618 | if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost | |
3619 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { | |
3620 | mark = z->_watermark[WMARK_MIN]; | |
3621 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
3622 | alloc_flags, free_pages); | |
3623 | } | |
3624 | ||
3625 | return false; | |
48ee5f36 MG |
3626 | } |
3627 | ||
7aeb09f9 | 3628 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 3629 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
3630 | { |
3631 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3632 | ||
3633 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3634 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3635 | ||
97a225e6 | 3636 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 3637 | free_pages); |
1da177e4 LT |
3638 | } |
3639 | ||
9276b1bc | 3640 | #ifdef CONFIG_NUMA |
957f822a DR |
3641 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3642 | { | |
e02dc017 | 3643 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3644 | node_reclaim_distance; |
957f822a | 3645 | } |
9276b1bc | 3646 | #else /* CONFIG_NUMA */ |
957f822a DR |
3647 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3648 | { | |
3649 | return true; | |
3650 | } | |
9276b1bc PJ |
3651 | #endif /* CONFIG_NUMA */ |
3652 | ||
6bb15450 MG |
3653 | /* |
3654 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3655 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3656 | * premature use of a lower zone may cause lowmem pressure problems that | |
3657 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3658 | * probably too small. It only makes sense to spread allocations to avoid | |
3659 | * fragmentation between the Normal and DMA32 zones. | |
3660 | */ | |
3661 | static inline unsigned int | |
0a79cdad | 3662 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3663 | { |
736838e9 | 3664 | unsigned int alloc_flags; |
0a79cdad | 3665 | |
736838e9 MN |
3666 | /* |
3667 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
3668 | * to save a branch. | |
3669 | */ | |
3670 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
3671 | |
3672 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3673 | if (!zone) |
3674 | return alloc_flags; | |
3675 | ||
6bb15450 | 3676 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3677 | return alloc_flags; |
6bb15450 MG |
3678 | |
3679 | /* | |
3680 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3681 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3682 | * on UMA that if Normal is populated then so is DMA32. | |
3683 | */ | |
3684 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3685 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3686 | return alloc_flags; |
6bb15450 | 3687 | |
8118b82e | 3688 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3689 | #endif /* CONFIG_ZONE_DMA32 */ |
3690 | return alloc_flags; | |
6bb15450 | 3691 | } |
6bb15450 | 3692 | |
7fb1d9fc | 3693 | /* |
0798e519 | 3694 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3695 | * a page. |
3696 | */ | |
3697 | static struct page * | |
a9263751 VB |
3698 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3699 | const struct alloc_context *ac) | |
753ee728 | 3700 | { |
6bb15450 | 3701 | struct zoneref *z; |
5117f45d | 3702 | struct zone *zone; |
3b8c0be4 | 3703 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
6bb15450 | 3704 | bool no_fallback; |
3b8c0be4 | 3705 | |
6bb15450 | 3706 | retry: |
7fb1d9fc | 3707 | /* |
9276b1bc | 3708 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3709 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3710 | */ |
6bb15450 MG |
3711 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3712 | z = ac->preferred_zoneref; | |
97a225e6 JK |
3713 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3714 | ac->highest_zoneidx, ac->nodemask) { | |
be06af00 | 3715 | struct page *page; |
e085dbc5 JW |
3716 | unsigned long mark; |
3717 | ||
664eedde MG |
3718 | if (cpusets_enabled() && |
3719 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3720 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3721 | continue; |
a756cf59 JW |
3722 | /* |
3723 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3724 | * want to get it from a node that is within its dirty |
3725 | * limit, such that no single node holds more than its | |
a756cf59 | 3726 | * proportional share of globally allowed dirty pages. |
281e3726 | 3727 | * The dirty limits take into account the node's |
a756cf59 JW |
3728 | * lowmem reserves and high watermark so that kswapd |
3729 | * should be able to balance it without having to | |
3730 | * write pages from its LRU list. | |
3731 | * | |
a756cf59 | 3732 | * XXX: For now, allow allocations to potentially |
281e3726 | 3733 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3734 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3735 | * which is important when on a NUMA setup the allowed |
281e3726 | 3736 | * nodes are together not big enough to reach the |
a756cf59 | 3737 | * global limit. The proper fix for these situations |
281e3726 | 3738 | * will require awareness of nodes in the |
a756cf59 JW |
3739 | * dirty-throttling and the flusher threads. |
3740 | */ | |
3b8c0be4 MG |
3741 | if (ac->spread_dirty_pages) { |
3742 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3743 | continue; | |
3744 | ||
3745 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3746 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3747 | continue; | |
3748 | } | |
3749 | } | |
7fb1d9fc | 3750 | |
6bb15450 MG |
3751 | if (no_fallback && nr_online_nodes > 1 && |
3752 | zone != ac->preferred_zoneref->zone) { | |
3753 | int local_nid; | |
3754 | ||
3755 | /* | |
3756 | * If moving to a remote node, retry but allow | |
3757 | * fragmenting fallbacks. Locality is more important | |
3758 | * than fragmentation avoidance. | |
3759 | */ | |
3760 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3761 | if (zone_to_nid(zone) != local_nid) { | |
3762 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3763 | goto retry; | |
3764 | } | |
3765 | } | |
3766 | ||
a9214443 | 3767 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 3768 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
3769 | ac->highest_zoneidx, alloc_flags, |
3770 | gfp_mask)) { | |
fa5e084e MG |
3771 | int ret; |
3772 | ||
c9e97a19 PT |
3773 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3774 | /* | |
3775 | * Watermark failed for this zone, but see if we can | |
3776 | * grow this zone if it contains deferred pages. | |
3777 | */ | |
3778 | if (static_branch_unlikely(&deferred_pages)) { | |
3779 | if (_deferred_grow_zone(zone, order)) | |
3780 | goto try_this_zone; | |
3781 | } | |
3782 | #endif | |
5dab2911 MG |
3783 | /* Checked here to keep the fast path fast */ |
3784 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3785 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3786 | goto try_this_zone; | |
3787 | ||
a5f5f91d | 3788 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3789 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3790 | continue; |
3791 | ||
a5f5f91d | 3792 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3793 | switch (ret) { |
a5f5f91d | 3794 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3795 | /* did not scan */ |
cd38b115 | 3796 | continue; |
a5f5f91d | 3797 | case NODE_RECLAIM_FULL: |
fa5e084e | 3798 | /* scanned but unreclaimable */ |
cd38b115 | 3799 | continue; |
fa5e084e MG |
3800 | default: |
3801 | /* did we reclaim enough */ | |
fed2719e | 3802 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 3803 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
3804 | goto try_this_zone; |
3805 | ||
fed2719e | 3806 | continue; |
0798e519 | 3807 | } |
7fb1d9fc RS |
3808 | } |
3809 | ||
fa5e084e | 3810 | try_this_zone: |
066b2393 | 3811 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3812 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3813 | if (page) { |
479f854a | 3814 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3815 | |
3816 | /* | |
3817 | * If this is a high-order atomic allocation then check | |
3818 | * if the pageblock should be reserved for the future | |
3819 | */ | |
3820 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3821 | reserve_highatomic_pageblock(page, zone, order); | |
3822 | ||
75379191 | 3823 | return page; |
c9e97a19 PT |
3824 | } else { |
3825 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
3826 | /* Try again if zone has deferred pages */ | |
3827 | if (static_branch_unlikely(&deferred_pages)) { | |
3828 | if (_deferred_grow_zone(zone, order)) | |
3829 | goto try_this_zone; | |
3830 | } | |
3831 | #endif | |
75379191 | 3832 | } |
54a6eb5c | 3833 | } |
9276b1bc | 3834 | |
6bb15450 MG |
3835 | /* |
3836 | * It's possible on a UMA machine to get through all zones that are | |
3837 | * fragmented. If avoiding fragmentation, reset and try again. | |
3838 | */ | |
3839 | if (no_fallback) { | |
3840 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3841 | goto retry; | |
3842 | } | |
3843 | ||
4ffeaf35 | 3844 | return NULL; |
753ee728 MH |
3845 | } |
3846 | ||
9af744d7 | 3847 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3848 | { |
a238ab5b | 3849 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
3850 | |
3851 | /* | |
3852 | * This documents exceptions given to allocations in certain | |
3853 | * contexts that are allowed to allocate outside current's set | |
3854 | * of allowed nodes. | |
3855 | */ | |
3856 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3857 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3858 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3859 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3860 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3861 | filter &= ~SHOW_MEM_FILTER_NODES; |
3862 | ||
9af744d7 | 3863 | show_mem(filter, nodemask); |
aa187507 MH |
3864 | } |
3865 | ||
a8e99259 | 3866 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3867 | { |
3868 | struct va_format vaf; | |
3869 | va_list args; | |
1be334e5 | 3870 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 3871 | |
0f7896f1 | 3872 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3873 | return; |
3874 | ||
7877cdcc MH |
3875 | va_start(args, fmt); |
3876 | vaf.fmt = fmt; | |
3877 | vaf.va = &args; | |
ef8444ea | 3878 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
3879 | current->comm, &vaf, gfp_mask, &gfp_mask, |
3880 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3881 | va_end(args); |
3ee9a4f0 | 3882 | |
a8e99259 | 3883 | cpuset_print_current_mems_allowed(); |
ef8444ea | 3884 | pr_cont("\n"); |
a238ab5b | 3885 | dump_stack(); |
685dbf6f | 3886 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3887 | } |
3888 | ||
6c18ba7a MH |
3889 | static inline struct page * |
3890 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3891 | unsigned int alloc_flags, | |
3892 | const struct alloc_context *ac) | |
3893 | { | |
3894 | struct page *page; | |
3895 | ||
3896 | page = get_page_from_freelist(gfp_mask, order, | |
3897 | alloc_flags|ALLOC_CPUSET, ac); | |
3898 | /* | |
3899 | * fallback to ignore cpuset restriction if our nodes | |
3900 | * are depleted | |
3901 | */ | |
3902 | if (!page) | |
3903 | page = get_page_from_freelist(gfp_mask, order, | |
3904 | alloc_flags, ac); | |
3905 | ||
3906 | return page; | |
3907 | } | |
3908 | ||
11e33f6a MG |
3909 | static inline struct page * |
3910 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3911 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3912 | { |
6e0fc46d DR |
3913 | struct oom_control oc = { |
3914 | .zonelist = ac->zonelist, | |
3915 | .nodemask = ac->nodemask, | |
2a966b77 | 3916 | .memcg = NULL, |
6e0fc46d DR |
3917 | .gfp_mask = gfp_mask, |
3918 | .order = order, | |
6e0fc46d | 3919 | }; |
11e33f6a MG |
3920 | struct page *page; |
3921 | ||
9879de73 JW |
3922 | *did_some_progress = 0; |
3923 | ||
9879de73 | 3924 | /* |
dc56401f JW |
3925 | * Acquire the oom lock. If that fails, somebody else is |
3926 | * making progress for us. | |
9879de73 | 3927 | */ |
dc56401f | 3928 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3929 | *did_some_progress = 1; |
11e33f6a | 3930 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3931 | return NULL; |
3932 | } | |
6b1de916 | 3933 | |
11e33f6a MG |
3934 | /* |
3935 | * Go through the zonelist yet one more time, keep very high watermark | |
3936 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3937 | * we're still under heavy pressure. But make sure that this reclaim |
3938 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3939 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3940 | */ |
e746bf73 TH |
3941 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3942 | ~__GFP_DIRECT_RECLAIM, order, | |
3943 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3944 | if (page) |
11e33f6a MG |
3945 | goto out; |
3946 | ||
06ad276a MH |
3947 | /* Coredumps can quickly deplete all memory reserves */ |
3948 | if (current->flags & PF_DUMPCORE) | |
3949 | goto out; | |
3950 | /* The OOM killer will not help higher order allocs */ | |
3951 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3952 | goto out; | |
dcda9b04 MH |
3953 | /* |
3954 | * We have already exhausted all our reclaim opportunities without any | |
3955 | * success so it is time to admit defeat. We will skip the OOM killer | |
3956 | * because it is very likely that the caller has a more reasonable | |
3957 | * fallback than shooting a random task. | |
3958 | */ | |
3959 | if (gfp_mask & __GFP_RETRY_MAYFAIL) | |
3960 | goto out; | |
06ad276a | 3961 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 3962 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
3963 | goto out; |
3964 | if (pm_suspended_storage()) | |
3965 | goto out; | |
3966 | /* | |
3967 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3968 | * other request to make a forward progress. | |
3969 | * We are in an unfortunate situation where out_of_memory cannot | |
3970 | * do much for this context but let's try it to at least get | |
3971 | * access to memory reserved if the current task is killed (see | |
3972 | * out_of_memory). Once filesystems are ready to handle allocation | |
3973 | * failures more gracefully we should just bail out here. | |
3974 | */ | |
3975 | ||
3976 | /* The OOM killer may not free memory on a specific node */ | |
3977 | if (gfp_mask & __GFP_THISNODE) | |
3978 | goto out; | |
3da88fb3 | 3979 | |
3c2c6488 | 3980 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 3981 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3982 | *did_some_progress = 1; |
5020e285 | 3983 | |
6c18ba7a MH |
3984 | /* |
3985 | * Help non-failing allocations by giving them access to memory | |
3986 | * reserves | |
3987 | */ | |
3988 | if (gfp_mask & __GFP_NOFAIL) | |
3989 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3990 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3991 | } |
11e33f6a | 3992 | out: |
dc56401f | 3993 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3994 | return page; |
3995 | } | |
3996 | ||
33c2d214 MH |
3997 | /* |
3998 | * Maximum number of compaction retries wit a progress before OOM | |
3999 | * killer is consider as the only way to move forward. | |
4000 | */ | |
4001 | #define MAX_COMPACT_RETRIES 16 | |
4002 | ||
56de7263 MG |
4003 | #ifdef CONFIG_COMPACTION |
4004 | /* Try memory compaction for high-order allocations before reclaim */ | |
4005 | static struct page * | |
4006 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4007 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4008 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4009 | { |
5e1f0f09 | 4010 | struct page *page = NULL; |
eb414681 | 4011 | unsigned long pflags; |
499118e9 | 4012 | unsigned int noreclaim_flag; |
53853e2d VB |
4013 | |
4014 | if (!order) | |
66199712 | 4015 | return NULL; |
66199712 | 4016 | |
eb414681 | 4017 | psi_memstall_enter(&pflags); |
499118e9 | 4018 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 4019 | |
c5d01d0d | 4020 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 4021 | prio, &page); |
eb414681 | 4022 | |
499118e9 | 4023 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 4024 | psi_memstall_leave(&pflags); |
56de7263 | 4025 | |
98dd3b48 VB |
4026 | /* |
4027 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
4028 | * count a compaction stall | |
4029 | */ | |
4030 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 4031 | |
5e1f0f09 MG |
4032 | /* Prep a captured page if available */ |
4033 | if (page) | |
4034 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
4035 | ||
4036 | /* Try get a page from the freelist if available */ | |
4037 | if (!page) | |
4038 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 4039 | |
98dd3b48 VB |
4040 | if (page) { |
4041 | struct zone *zone = page_zone(page); | |
53853e2d | 4042 | |
98dd3b48 VB |
4043 | zone->compact_blockskip_flush = false; |
4044 | compaction_defer_reset(zone, order, true); | |
4045 | count_vm_event(COMPACTSUCCESS); | |
4046 | return page; | |
4047 | } | |
56de7263 | 4048 | |
98dd3b48 VB |
4049 | /* |
4050 | * It's bad if compaction run occurs and fails. The most likely reason | |
4051 | * is that pages exist, but not enough to satisfy watermarks. | |
4052 | */ | |
4053 | count_vm_event(COMPACTFAIL); | |
66199712 | 4054 | |
98dd3b48 | 4055 | cond_resched(); |
56de7263 MG |
4056 | |
4057 | return NULL; | |
4058 | } | |
33c2d214 | 4059 | |
3250845d VB |
4060 | static inline bool |
4061 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
4062 | enum compact_result compact_result, | |
4063 | enum compact_priority *compact_priority, | |
d9436498 | 4064 | int *compaction_retries) |
3250845d VB |
4065 | { |
4066 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 4067 | int min_priority; |
65190cff MH |
4068 | bool ret = false; |
4069 | int retries = *compaction_retries; | |
4070 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
4071 | |
4072 | if (!order) | |
4073 | return false; | |
4074 | ||
d9436498 VB |
4075 | if (compaction_made_progress(compact_result)) |
4076 | (*compaction_retries)++; | |
4077 | ||
3250845d VB |
4078 | /* |
4079 | * compaction considers all the zone as desperately out of memory | |
4080 | * so it doesn't really make much sense to retry except when the | |
4081 | * failure could be caused by insufficient priority | |
4082 | */ | |
d9436498 VB |
4083 | if (compaction_failed(compact_result)) |
4084 | goto check_priority; | |
3250845d | 4085 | |
49433085 VB |
4086 | /* |
4087 | * compaction was skipped because there are not enough order-0 pages | |
4088 | * to work with, so we retry only if it looks like reclaim can help. | |
4089 | */ | |
4090 | if (compaction_needs_reclaim(compact_result)) { | |
4091 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
4092 | goto out; | |
4093 | } | |
4094 | ||
3250845d VB |
4095 | /* |
4096 | * make sure the compaction wasn't deferred or didn't bail out early | |
4097 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
4098 | * But the next retry should use a higher priority if allowed, so |
4099 | * we don't just keep bailing out endlessly. | |
3250845d | 4100 | */ |
65190cff | 4101 | if (compaction_withdrawn(compact_result)) { |
49433085 | 4102 | goto check_priority; |
65190cff | 4103 | } |
3250845d VB |
4104 | |
4105 | /* | |
dcda9b04 | 4106 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
4107 | * costly ones because they are de facto nofail and invoke OOM |
4108 | * killer to move on while costly can fail and users are ready | |
4109 | * to cope with that. 1/4 retries is rather arbitrary but we | |
4110 | * would need much more detailed feedback from compaction to | |
4111 | * make a better decision. | |
4112 | */ | |
4113 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4114 | max_retries /= 4; | |
65190cff MH |
4115 | if (*compaction_retries <= max_retries) { |
4116 | ret = true; | |
4117 | goto out; | |
4118 | } | |
3250845d | 4119 | |
d9436498 VB |
4120 | /* |
4121 | * Make sure there are attempts at the highest priority if we exhausted | |
4122 | * all retries or failed at the lower priorities. | |
4123 | */ | |
4124 | check_priority: | |
c2033b00 VB |
4125 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
4126 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 4127 | |
c2033b00 | 4128 | if (*compact_priority > min_priority) { |
d9436498 VB |
4129 | (*compact_priority)--; |
4130 | *compaction_retries = 0; | |
65190cff | 4131 | ret = true; |
d9436498 | 4132 | } |
65190cff MH |
4133 | out: |
4134 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4135 | return ret; | |
3250845d | 4136 | } |
56de7263 MG |
4137 | #else |
4138 | static inline struct page * | |
4139 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4140 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4141 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4142 | { |
33c2d214 | 4143 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4144 | return NULL; |
4145 | } | |
33c2d214 MH |
4146 | |
4147 | static inline bool | |
86a294a8 MH |
4148 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4149 | enum compact_result compact_result, | |
a5508cd8 | 4150 | enum compact_priority *compact_priority, |
d9436498 | 4151 | int *compaction_retries) |
33c2d214 | 4152 | { |
31e49bfd MH |
4153 | struct zone *zone; |
4154 | struct zoneref *z; | |
4155 | ||
4156 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4157 | return false; | |
4158 | ||
4159 | /* | |
4160 | * There are setups with compaction disabled which would prefer to loop | |
4161 | * inside the allocator rather than hit the oom killer prematurely. | |
4162 | * Let's give them a good hope and keep retrying while the order-0 | |
4163 | * watermarks are OK. | |
4164 | */ | |
97a225e6 JK |
4165 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4166 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 4167 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 4168 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
4169 | return true; |
4170 | } | |
33c2d214 MH |
4171 | return false; |
4172 | } | |
3250845d | 4173 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4174 | |
d92a8cfc | 4175 | #ifdef CONFIG_LOCKDEP |
93781325 | 4176 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4177 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4178 | ||
4179 | static bool __need_fs_reclaim(gfp_t gfp_mask) | |
4180 | { | |
4181 | gfp_mask = current_gfp_context(gfp_mask); | |
4182 | ||
4183 | /* no reclaim without waiting on it */ | |
4184 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4185 | return false; | |
4186 | ||
4187 | /* this guy won't enter reclaim */ | |
2e517d68 | 4188 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4189 | return false; |
4190 | ||
4191 | /* We're only interested __GFP_FS allocations for now */ | |
4192 | if (!(gfp_mask & __GFP_FS)) | |
4193 | return false; | |
4194 | ||
4195 | if (gfp_mask & __GFP_NOLOCKDEP) | |
4196 | return false; | |
4197 | ||
4198 | return true; | |
4199 | } | |
4200 | ||
93781325 OS |
4201 | void __fs_reclaim_acquire(void) |
4202 | { | |
4203 | lock_map_acquire(&__fs_reclaim_map); | |
4204 | } | |
4205 | ||
4206 | void __fs_reclaim_release(void) | |
4207 | { | |
4208 | lock_map_release(&__fs_reclaim_map); | |
4209 | } | |
4210 | ||
d92a8cfc PZ |
4211 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4212 | { | |
4213 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 4214 | __fs_reclaim_acquire(); |
d92a8cfc PZ |
4215 | } |
4216 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4217 | ||
4218 | void fs_reclaim_release(gfp_t gfp_mask) | |
4219 | { | |
4220 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 4221 | __fs_reclaim_release(); |
d92a8cfc PZ |
4222 | } |
4223 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4224 | #endif | |
4225 | ||
bba90710 MS |
4226 | /* Perform direct synchronous page reclaim */ |
4227 | static int | |
a9263751 VB |
4228 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4229 | const struct alloc_context *ac) | |
11e33f6a | 4230 | { |
bba90710 | 4231 | int progress; |
499118e9 | 4232 | unsigned int noreclaim_flag; |
eb414681 | 4233 | unsigned long pflags; |
11e33f6a MG |
4234 | |
4235 | cond_resched(); | |
4236 | ||
4237 | /* We now go into synchronous reclaim */ | |
4238 | cpuset_memory_pressure_bump(); | |
eb414681 | 4239 | psi_memstall_enter(&pflags); |
d92a8cfc | 4240 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4241 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4242 | |
a9263751 VB |
4243 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4244 | ac->nodemask); | |
11e33f6a | 4245 | |
499118e9 | 4246 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4247 | fs_reclaim_release(gfp_mask); |
eb414681 | 4248 | psi_memstall_leave(&pflags); |
11e33f6a MG |
4249 | |
4250 | cond_resched(); | |
4251 | ||
bba90710 MS |
4252 | return progress; |
4253 | } | |
4254 | ||
4255 | /* The really slow allocator path where we enter direct reclaim */ | |
4256 | static inline struct page * | |
4257 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4258 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4259 | unsigned long *did_some_progress) |
bba90710 MS |
4260 | { |
4261 | struct page *page = NULL; | |
4262 | bool drained = false; | |
4263 | ||
a9263751 | 4264 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
4265 | if (unlikely(!(*did_some_progress))) |
4266 | return NULL; | |
11e33f6a | 4267 | |
9ee493ce | 4268 | retry: |
31a6c190 | 4269 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4270 | |
4271 | /* | |
4272 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
4273 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
4274 | * Shrink them them and try again | |
9ee493ce MG |
4275 | */ |
4276 | if (!page && !drained) { | |
29fac03b | 4277 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4278 | drain_all_pages(NULL); |
9ee493ce MG |
4279 | drained = true; |
4280 | goto retry; | |
4281 | } | |
4282 | ||
11e33f6a MG |
4283 | return page; |
4284 | } | |
4285 | ||
5ecd9d40 DR |
4286 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4287 | const struct alloc_context *ac) | |
3a025760 JW |
4288 | { |
4289 | struct zoneref *z; | |
4290 | struct zone *zone; | |
e1a55637 | 4291 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 4292 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 4293 | |
97a225e6 | 4294 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 4295 | ac->nodemask) { |
e1a55637 | 4296 | if (last_pgdat != zone->zone_pgdat) |
97a225e6 | 4297 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
e1a55637 MG |
4298 | last_pgdat = zone->zone_pgdat; |
4299 | } | |
3a025760 JW |
4300 | } |
4301 | ||
c603844b | 4302 | static inline unsigned int |
341ce06f PZ |
4303 | gfp_to_alloc_flags(gfp_t gfp_mask) |
4304 | { | |
c603844b | 4305 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4306 | |
736838e9 MN |
4307 | /* |
4308 | * __GFP_HIGH is assumed to be the same as ALLOC_HIGH | |
4309 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4310 | * to save two branches. | |
4311 | */ | |
e6223a3b | 4312 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
736838e9 | 4313 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 4314 | |
341ce06f PZ |
4315 | /* |
4316 | * The caller may dip into page reserves a bit more if the caller | |
4317 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4318 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 4319 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 4320 | */ |
736838e9 MN |
4321 | alloc_flags |= (__force int) |
4322 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 4323 | |
d0164adc | 4324 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 4325 | /* |
b104a35d DR |
4326 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4327 | * if it can't schedule. | |
5c3240d9 | 4328 | */ |
b104a35d | 4329 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 4330 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 4331 | /* |
b104a35d | 4332 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 4333 | * comment for __cpuset_node_allowed(). |
523b9458 | 4334 | */ |
341ce06f | 4335 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 4336 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
4337 | alloc_flags |= ALLOC_HARDER; |
4338 | ||
d883c6cf | 4339 | #ifdef CONFIG_CMA |
01c0bfe0 | 4340 | if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d883c6cf JK |
4341 | alloc_flags |= ALLOC_CMA; |
4342 | #endif | |
341ce06f PZ |
4343 | return alloc_flags; |
4344 | } | |
4345 | ||
cd04ae1e | 4346 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4347 | { |
cd04ae1e MH |
4348 | if (!tsk_is_oom_victim(tsk)) |
4349 | return false; | |
4350 | ||
4351 | /* | |
4352 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4353 | * only to the thread with TIF_MEMDIE set | |
4354 | */ | |
4355 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4356 | return false; |
4357 | ||
cd04ae1e MH |
4358 | return true; |
4359 | } | |
4360 | ||
4361 | /* | |
4362 | * Distinguish requests which really need access to full memory | |
4363 | * reserves from oom victims which can live with a portion of it | |
4364 | */ | |
4365 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4366 | { | |
4367 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4368 | return 0; | |
31a6c190 | 4369 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4370 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4371 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4372 | return ALLOC_NO_WATERMARKS; |
4373 | if (!in_interrupt()) { | |
4374 | if (current->flags & PF_MEMALLOC) | |
4375 | return ALLOC_NO_WATERMARKS; | |
4376 | else if (oom_reserves_allowed(current)) | |
4377 | return ALLOC_OOM; | |
4378 | } | |
31a6c190 | 4379 | |
cd04ae1e MH |
4380 | return 0; |
4381 | } | |
4382 | ||
4383 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4384 | { | |
4385 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4386 | } |
4387 | ||
0a0337e0 MH |
4388 | /* |
4389 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4390 | * for the given allocation request. | |
491d79ae JW |
4391 | * |
4392 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4393 | * without success, or when we couldn't even meet the watermark if we | |
4394 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4395 | * |
4396 | * Returns true if a retry is viable or false to enter the oom path. | |
4397 | */ | |
4398 | static inline bool | |
4399 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4400 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4401 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4402 | { |
4403 | struct zone *zone; | |
4404 | struct zoneref *z; | |
15f570bf | 4405 | bool ret = false; |
0a0337e0 | 4406 | |
423b452e VB |
4407 | /* |
4408 | * Costly allocations might have made a progress but this doesn't mean | |
4409 | * their order will become available due to high fragmentation so | |
4410 | * always increment the no progress counter for them | |
4411 | */ | |
4412 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4413 | *no_progress_loops = 0; | |
4414 | else | |
4415 | (*no_progress_loops)++; | |
4416 | ||
0a0337e0 MH |
4417 | /* |
4418 | * Make sure we converge to OOM if we cannot make any progress | |
4419 | * several times in the row. | |
4420 | */ | |
04c8716f MK |
4421 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4422 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4423 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4424 | } |
0a0337e0 | 4425 | |
bca67592 MG |
4426 | /* |
4427 | * Keep reclaiming pages while there is a chance this will lead | |
4428 | * somewhere. If none of the target zones can satisfy our allocation | |
4429 | * request even if all reclaimable pages are considered then we are | |
4430 | * screwed and have to go OOM. | |
0a0337e0 | 4431 | */ |
97a225e6 JK |
4432 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4433 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4434 | unsigned long available; |
ede37713 | 4435 | unsigned long reclaimable; |
d379f01d MH |
4436 | unsigned long min_wmark = min_wmark_pages(zone); |
4437 | bool wmark; | |
0a0337e0 | 4438 | |
5a1c84b4 | 4439 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4440 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4441 | |
4442 | /* | |
491d79ae JW |
4443 | * Would the allocation succeed if we reclaimed all |
4444 | * reclaimable pages? | |
0a0337e0 | 4445 | */ |
d379f01d | 4446 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 4447 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
4448 | trace_reclaim_retry_zone(z, order, reclaimable, |
4449 | available, min_wmark, *no_progress_loops, wmark); | |
4450 | if (wmark) { | |
ede37713 MH |
4451 | /* |
4452 | * If we didn't make any progress and have a lot of | |
4453 | * dirty + writeback pages then we should wait for | |
4454 | * an IO to complete to slow down the reclaim and | |
4455 | * prevent from pre mature OOM | |
4456 | */ | |
4457 | if (!did_some_progress) { | |
11fb9989 | 4458 | unsigned long write_pending; |
ede37713 | 4459 | |
5a1c84b4 MG |
4460 | write_pending = zone_page_state_snapshot(zone, |
4461 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 4462 | |
11fb9989 | 4463 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
4464 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
4465 | return true; | |
4466 | } | |
4467 | } | |
5a1c84b4 | 4468 | |
15f570bf MH |
4469 | ret = true; |
4470 | goto out; | |
0a0337e0 MH |
4471 | } |
4472 | } | |
4473 | ||
15f570bf MH |
4474 | out: |
4475 | /* | |
4476 | * Memory allocation/reclaim might be called from a WQ context and the | |
4477 | * current implementation of the WQ concurrency control doesn't | |
4478 | * recognize that a particular WQ is congested if the worker thread is | |
4479 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4480 | * here rather than calling cond_resched(). | |
4481 | */ | |
4482 | if (current->flags & PF_WQ_WORKER) | |
4483 | schedule_timeout_uninterruptible(1); | |
4484 | else | |
4485 | cond_resched(); | |
4486 | return ret; | |
0a0337e0 MH |
4487 | } |
4488 | ||
902b6281 VB |
4489 | static inline bool |
4490 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4491 | { | |
4492 | /* | |
4493 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4494 | * mempolicy don't intersect. This should be normally dealt with by | |
4495 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4496 | * such a way the check therein was true, and then it became false | |
4497 | * before we got our cpuset_mems_cookie here. | |
4498 | * This assumes that for all allocations, ac->nodemask can come only | |
4499 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4500 | * when it does not intersect with the cpuset restrictions) or the | |
4501 | * caller can deal with a violated nodemask. | |
4502 | */ | |
4503 | if (cpusets_enabled() && ac->nodemask && | |
4504 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4505 | ac->nodemask = NULL; | |
4506 | return true; | |
4507 | } | |
4508 | ||
4509 | /* | |
4510 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4511 | * possible to race with parallel threads in such a way that our | |
4512 | * allocation can fail while the mask is being updated. If we are about | |
4513 | * to fail, check if the cpuset changed during allocation and if so, | |
4514 | * retry. | |
4515 | */ | |
4516 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4517 | return true; | |
4518 | ||
4519 | return false; | |
4520 | } | |
4521 | ||
11e33f6a MG |
4522 | static inline struct page * |
4523 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4524 | struct alloc_context *ac) |
11e33f6a | 4525 | { |
d0164adc | 4526 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4527 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4528 | struct page *page = NULL; |
c603844b | 4529 | unsigned int alloc_flags; |
11e33f6a | 4530 | unsigned long did_some_progress; |
5ce9bfef | 4531 | enum compact_priority compact_priority; |
c5d01d0d | 4532 | enum compact_result compact_result; |
5ce9bfef VB |
4533 | int compaction_retries; |
4534 | int no_progress_loops; | |
5ce9bfef | 4535 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4536 | int reserve_flags; |
1da177e4 | 4537 | |
d0164adc MG |
4538 | /* |
4539 | * We also sanity check to catch abuse of atomic reserves being used by | |
4540 | * callers that are not in atomic context. | |
4541 | */ | |
4542 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4543 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4544 | gfp_mask &= ~__GFP_ATOMIC; | |
4545 | ||
5ce9bfef VB |
4546 | retry_cpuset: |
4547 | compaction_retries = 0; | |
4548 | no_progress_loops = 0; | |
4549 | compact_priority = DEF_COMPACT_PRIORITY; | |
4550 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4551 | |
4552 | /* | |
4553 | * The fast path uses conservative alloc_flags to succeed only until | |
4554 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4555 | * alloc_flags precisely. So we do that now. | |
4556 | */ | |
4557 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4558 | ||
e47483bc VB |
4559 | /* |
4560 | * We need to recalculate the starting point for the zonelist iterator | |
4561 | * because we might have used different nodemask in the fast path, or | |
4562 | * there was a cpuset modification and we are retrying - otherwise we | |
4563 | * could end up iterating over non-eligible zones endlessly. | |
4564 | */ | |
4565 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 4566 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
4567 | if (!ac->preferred_zoneref->zone) |
4568 | goto nopage; | |
4569 | ||
0a79cdad | 4570 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4571 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4572 | |
4573 | /* | |
4574 | * The adjusted alloc_flags might result in immediate success, so try | |
4575 | * that first | |
4576 | */ | |
4577 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4578 | if (page) | |
4579 | goto got_pg; | |
4580 | ||
a8161d1e VB |
4581 | /* |
4582 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4583 | * that we have enough base pages and don't need to reclaim. For non- |
4584 | * movable high-order allocations, do that as well, as compaction will | |
4585 | * try prevent permanent fragmentation by migrating from blocks of the | |
4586 | * same migratetype. | |
4587 | * Don't try this for allocations that are allowed to ignore | |
4588 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4589 | */ |
282722b0 VB |
4590 | if (can_direct_reclaim && |
4591 | (costly_order || | |
4592 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4593 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4594 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4595 | alloc_flags, ac, | |
a5508cd8 | 4596 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4597 | &compact_result); |
4598 | if (page) | |
4599 | goto got_pg; | |
4600 | ||
cc638f32 VB |
4601 | /* |
4602 | * Checks for costly allocations with __GFP_NORETRY, which | |
4603 | * includes some THP page fault allocations | |
4604 | */ | |
4605 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4606 | /* |
4607 | * If allocating entire pageblock(s) and compaction | |
4608 | * failed because all zones are below low watermarks | |
4609 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4610 | * order, fail immediately unless the allocator has |
4611 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4612 | * |
4613 | * Reclaim is | |
4614 | * - potentially very expensive because zones are far | |
4615 | * below their low watermarks or this is part of very | |
4616 | * bursty high order allocations, | |
4617 | * - not guaranteed to help because isolate_freepages() | |
4618 | * may not iterate over freed pages as part of its | |
4619 | * linear scan, and | |
4620 | * - unlikely to make entire pageblocks free on its | |
4621 | * own. | |
4622 | */ | |
4623 | if (compact_result == COMPACT_SKIPPED || | |
4624 | compact_result == COMPACT_DEFERRED) | |
4625 | goto nopage; | |
a8161d1e | 4626 | |
a8161d1e | 4627 | /* |
3eb2771b VB |
4628 | * Looks like reclaim/compaction is worth trying, but |
4629 | * sync compaction could be very expensive, so keep | |
25160354 | 4630 | * using async compaction. |
a8161d1e | 4631 | */ |
a5508cd8 | 4632 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4633 | } |
4634 | } | |
23771235 | 4635 | |
31a6c190 | 4636 | retry: |
23771235 | 4637 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4638 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4639 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4640 | |
cd04ae1e MH |
4641 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4642 | if (reserve_flags) | |
4643 | alloc_flags = reserve_flags; | |
23771235 | 4644 | |
e46e7b77 | 4645 | /* |
d6a24df0 VB |
4646 | * Reset the nodemask and zonelist iterators if memory policies can be |
4647 | * ignored. These allocations are high priority and system rather than | |
4648 | * user oriented. | |
e46e7b77 | 4649 | */ |
cd04ae1e | 4650 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4651 | ac->nodemask = NULL; |
e46e7b77 | 4652 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4653 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
4654 | } |
4655 | ||
23771235 | 4656 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4657 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4658 | if (page) |
4659 | goto got_pg; | |
1da177e4 | 4660 | |
d0164adc | 4661 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4662 | if (!can_direct_reclaim) |
1da177e4 LT |
4663 | goto nopage; |
4664 | ||
9a67f648 MH |
4665 | /* Avoid recursion of direct reclaim */ |
4666 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4667 | goto nopage; |
4668 | ||
a8161d1e VB |
4669 | /* Try direct reclaim and then allocating */ |
4670 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4671 | &did_some_progress); | |
4672 | if (page) | |
4673 | goto got_pg; | |
4674 | ||
4675 | /* Try direct compaction and then allocating */ | |
a9263751 | 4676 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4677 | compact_priority, &compact_result); |
56de7263 MG |
4678 | if (page) |
4679 | goto got_pg; | |
75f30861 | 4680 | |
9083905a JW |
4681 | /* Do not loop if specifically requested */ |
4682 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4683 | goto nopage; |
9083905a | 4684 | |
0a0337e0 MH |
4685 | /* |
4686 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4687 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4688 | */ |
dcda9b04 | 4689 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4690 | goto nopage; |
0a0337e0 | 4691 | |
0a0337e0 | 4692 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4693 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4694 | goto retry; |
4695 | ||
33c2d214 MH |
4696 | /* |
4697 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4698 | * reclaim is not able to make any progress because the current | |
4699 | * implementation of the compaction depends on the sufficient amount | |
4700 | * of free memory (see __compaction_suitable) | |
4701 | */ | |
4702 | if (did_some_progress > 0 && | |
86a294a8 | 4703 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4704 | compact_result, &compact_priority, |
d9436498 | 4705 | &compaction_retries)) |
33c2d214 MH |
4706 | goto retry; |
4707 | ||
902b6281 VB |
4708 | |
4709 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4710 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4711 | goto retry_cpuset; |
4712 | ||
9083905a JW |
4713 | /* Reclaim has failed us, start killing things */ |
4714 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4715 | if (page) | |
4716 | goto got_pg; | |
4717 | ||
9a67f648 | 4718 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e MH |
4719 | if (tsk_is_oom_victim(current) && |
4720 | (alloc_flags == ALLOC_OOM || | |
c288983d | 4721 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4722 | goto nopage; |
4723 | ||
9083905a | 4724 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4725 | if (did_some_progress) { |
4726 | no_progress_loops = 0; | |
9083905a | 4727 | goto retry; |
0a0337e0 | 4728 | } |
9083905a | 4729 | |
1da177e4 | 4730 | nopage: |
902b6281 VB |
4731 | /* Deal with possible cpuset update races before we fail */ |
4732 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4733 | goto retry_cpuset; |
4734 | ||
9a67f648 MH |
4735 | /* |
4736 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4737 | * we always retry | |
4738 | */ | |
4739 | if (gfp_mask & __GFP_NOFAIL) { | |
4740 | /* | |
4741 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4742 | * of any new users that actually require GFP_NOWAIT | |
4743 | */ | |
4744 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4745 | goto fail; | |
4746 | ||
4747 | /* | |
4748 | * PF_MEMALLOC request from this context is rather bizarre | |
4749 | * because we cannot reclaim anything and only can loop waiting | |
4750 | * for somebody to do a work for us | |
4751 | */ | |
4752 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4753 | ||
4754 | /* | |
4755 | * non failing costly orders are a hard requirement which we | |
4756 | * are not prepared for much so let's warn about these users | |
4757 | * so that we can identify them and convert them to something | |
4758 | * else. | |
4759 | */ | |
4760 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4761 | ||
6c18ba7a MH |
4762 | /* |
4763 | * Help non-failing allocations by giving them access to memory | |
4764 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4765 | * could deplete whole memory reserves which would just make | |
4766 | * the situation worse | |
4767 | */ | |
4768 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4769 | if (page) | |
4770 | goto got_pg; | |
4771 | ||
9a67f648 MH |
4772 | cond_resched(); |
4773 | goto retry; | |
4774 | } | |
4775 | fail: | |
a8e99259 | 4776 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4777 | "page allocation failure: order:%u", order); |
1da177e4 | 4778 | got_pg: |
072bb0aa | 4779 | return page; |
1da177e4 | 4780 | } |
11e33f6a | 4781 | |
9cd75558 | 4782 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4783 | int preferred_nid, nodemask_t *nodemask, |
9cd75558 MG |
4784 | struct alloc_context *ac, gfp_t *alloc_mask, |
4785 | unsigned int *alloc_flags) | |
11e33f6a | 4786 | { |
97a225e6 | 4787 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4788 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 4789 | ac->nodemask = nodemask; |
01c0bfe0 | 4790 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 4791 | |
682a3385 | 4792 | if (cpusets_enabled()) { |
9cd75558 | 4793 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
4794 | if (!ac->nodemask) |
4795 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
4796 | else |
4797 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4798 | } |
4799 | ||
d92a8cfc PZ |
4800 | fs_reclaim_acquire(gfp_mask); |
4801 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4802 | |
d0164adc | 4803 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4804 | |
4805 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4806 | return false; |
11e33f6a | 4807 | |
d883c6cf JK |
4808 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
4809 | *alloc_flags |= ALLOC_CMA; | |
4810 | ||
9cd75558 MG |
4811 | return true; |
4812 | } | |
21bb9bd1 | 4813 | |
9cd75558 | 4814 | /* Determine whether to spread dirty pages and what the first usable zone */ |
a380b40a | 4815 | static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac) |
9cd75558 | 4816 | { |
c9ab0c4f | 4817 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4818 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4819 | |
e46e7b77 MG |
4820 | /* |
4821 | * The preferred zone is used for statistics but crucially it is | |
4822 | * also used as the starting point for the zonelist iterator. It | |
4823 | * may get reset for allocations that ignore memory policies. | |
4824 | */ | |
9cd75558 | 4825 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4826 | ac->highest_zoneidx, ac->nodemask); |
9cd75558 MG |
4827 | } |
4828 | ||
4829 | /* | |
4830 | * This is the 'heart' of the zoned buddy allocator. | |
4831 | */ | |
4832 | struct page * | |
04ec6264 VB |
4833 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
4834 | nodemask_t *nodemask) | |
9cd75558 MG |
4835 | { |
4836 | struct page *page; | |
4837 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
f19360f0 | 4838 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4839 | struct alloc_context ac = { }; |
4840 | ||
c63ae43b MH |
4841 | /* |
4842 | * There are several places where we assume that the order value is sane | |
4843 | * so bail out early if the request is out of bound. | |
4844 | */ | |
4845 | if (unlikely(order >= MAX_ORDER)) { | |
4846 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
4847 | return NULL; | |
4848 | } | |
4849 | ||
9cd75558 | 4850 | gfp_mask &= gfp_allowed_mask; |
f19360f0 | 4851 | alloc_mask = gfp_mask; |
04ec6264 | 4852 | if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags)) |
9cd75558 MG |
4853 | return NULL; |
4854 | ||
a380b40a | 4855 | finalise_ac(gfp_mask, &ac); |
5bb1b169 | 4856 | |
6bb15450 MG |
4857 | /* |
4858 | * Forbid the first pass from falling back to types that fragment | |
4859 | * memory until all local zones are considered. | |
4860 | */ | |
0a79cdad | 4861 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask); |
6bb15450 | 4862 | |
5117f45d | 4863 | /* First allocation attempt */ |
a9263751 | 4864 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4865 | if (likely(page)) |
4866 | goto out; | |
11e33f6a | 4867 | |
4fcb0971 | 4868 | /* |
7dea19f9 MH |
4869 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4870 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4871 | * from a particular context which has been marked by | |
4872 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4873 | */ |
7dea19f9 | 4874 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4875 | ac.spread_dirty_pages = false; |
23f086f9 | 4876 | |
4741526b MG |
4877 | /* |
4878 | * Restore the original nodemask if it was potentially replaced with | |
4879 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4880 | */ | |
97ce86f9 | 4881 | ac.nodemask = nodemask; |
16096c25 | 4882 | |
4fcb0971 | 4883 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4884 | |
4fcb0971 | 4885 | out: |
c4159a75 | 4886 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
f4b00eab | 4887 | unlikely(__memcg_kmem_charge_page(page, gfp_mask, order) != 0)) { |
c4159a75 VD |
4888 | __free_pages(page, order); |
4889 | page = NULL; | |
4949148a VD |
4890 | } |
4891 | ||
4fcb0971 MG |
4892 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
4893 | ||
11e33f6a | 4894 | return page; |
1da177e4 | 4895 | } |
d239171e | 4896 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4897 | |
4898 | /* | |
9ea9a680 MH |
4899 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4900 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4901 | * you need to access high mem. | |
1da177e4 | 4902 | */ |
920c7a5d | 4903 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4904 | { |
945a1113 AM |
4905 | struct page *page; |
4906 | ||
9ea9a680 | 4907 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4908 | if (!page) |
4909 | return 0; | |
4910 | return (unsigned long) page_address(page); | |
4911 | } | |
1da177e4 LT |
4912 | EXPORT_SYMBOL(__get_free_pages); |
4913 | ||
920c7a5d | 4914 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4915 | { |
945a1113 | 4916 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4917 | } |
1da177e4 LT |
4918 | EXPORT_SYMBOL(get_zeroed_page); |
4919 | ||
742aa7fb | 4920 | static inline void free_the_page(struct page *page, unsigned int order) |
1da177e4 | 4921 | { |
742aa7fb AL |
4922 | if (order == 0) /* Via pcp? */ |
4923 | free_unref_page(page); | |
4924 | else | |
4925 | __free_pages_ok(page, order); | |
1da177e4 LT |
4926 | } |
4927 | ||
742aa7fb AL |
4928 | void __free_pages(struct page *page, unsigned int order) |
4929 | { | |
4930 | if (put_page_testzero(page)) | |
4931 | free_the_page(page, order); | |
4932 | } | |
1da177e4 LT |
4933 | EXPORT_SYMBOL(__free_pages); |
4934 | ||
920c7a5d | 4935 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4936 | { |
4937 | if (addr != 0) { | |
725d704e | 4938 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4939 | __free_pages(virt_to_page((void *)addr), order); |
4940 | } | |
4941 | } | |
4942 | ||
4943 | EXPORT_SYMBOL(free_pages); | |
4944 | ||
b63ae8ca AD |
4945 | /* |
4946 | * Page Fragment: | |
4947 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4948 | * within a 0 or higher order page. Multiple fragments within that page | |
4949 | * are individually refcounted, in the page's reference counter. | |
4950 | * | |
4951 | * The page_frag functions below provide a simple allocation framework for | |
4952 | * page fragments. This is used by the network stack and network device | |
4953 | * drivers to provide a backing region of memory for use as either an | |
4954 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4955 | */ | |
2976db80 AD |
4956 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4957 | gfp_t gfp_mask) | |
b63ae8ca AD |
4958 | { |
4959 | struct page *page = NULL; | |
4960 | gfp_t gfp = gfp_mask; | |
4961 | ||
4962 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4963 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4964 | __GFP_NOMEMALLOC; | |
4965 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4966 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4967 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4968 | #endif | |
4969 | if (unlikely(!page)) | |
4970 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4971 | ||
4972 | nc->va = page ? page_address(page) : NULL; | |
4973 | ||
4974 | return page; | |
4975 | } | |
4976 | ||
2976db80 | 4977 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4978 | { |
4979 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4980 | ||
742aa7fb AL |
4981 | if (page_ref_sub_and_test(page, count)) |
4982 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 4983 | } |
2976db80 | 4984 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4985 | |
8c2dd3e4 AD |
4986 | void *page_frag_alloc(struct page_frag_cache *nc, |
4987 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4988 | { |
4989 | unsigned int size = PAGE_SIZE; | |
4990 | struct page *page; | |
4991 | int offset; | |
4992 | ||
4993 | if (unlikely(!nc->va)) { | |
4994 | refill: | |
2976db80 | 4995 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4996 | if (!page) |
4997 | return NULL; | |
4998 | ||
4999 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5000 | /* if size can vary use size else just use PAGE_SIZE */ | |
5001 | size = nc->size; | |
5002 | #endif | |
5003 | /* Even if we own the page, we do not use atomic_set(). | |
5004 | * This would break get_page_unless_zero() users. | |
5005 | */ | |
86447726 | 5006 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
5007 | |
5008 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 5009 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 5010 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5011 | nc->offset = size; |
5012 | } | |
5013 | ||
5014 | offset = nc->offset - fragsz; | |
5015 | if (unlikely(offset < 0)) { | |
5016 | page = virt_to_page(nc->va); | |
5017 | ||
fe896d18 | 5018 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
5019 | goto refill; |
5020 | ||
5021 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5022 | /* if size can vary use size else just use PAGE_SIZE */ | |
5023 | size = nc->size; | |
5024 | #endif | |
5025 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 5026 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
5027 | |
5028 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 5029 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5030 | offset = size - fragsz; |
5031 | } | |
5032 | ||
5033 | nc->pagecnt_bias--; | |
5034 | nc->offset = offset; | |
5035 | ||
5036 | return nc->va + offset; | |
5037 | } | |
8c2dd3e4 | 5038 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
5039 | |
5040 | /* | |
5041 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
5042 | */ | |
8c2dd3e4 | 5043 | void page_frag_free(void *addr) |
b63ae8ca AD |
5044 | { |
5045 | struct page *page = virt_to_head_page(addr); | |
5046 | ||
742aa7fb AL |
5047 | if (unlikely(put_page_testzero(page))) |
5048 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 5049 | } |
8c2dd3e4 | 5050 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 5051 | |
d00181b9 KS |
5052 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
5053 | size_t size) | |
ee85c2e1 AK |
5054 | { |
5055 | if (addr) { | |
5056 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
5057 | unsigned long used = addr + PAGE_ALIGN(size); | |
5058 | ||
5059 | split_page(virt_to_page((void *)addr), order); | |
5060 | while (used < alloc_end) { | |
5061 | free_page(used); | |
5062 | used += PAGE_SIZE; | |
5063 | } | |
5064 | } | |
5065 | return (void *)addr; | |
5066 | } | |
5067 | ||
2be0ffe2 TT |
5068 | /** |
5069 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
5070 | * @size: the number of bytes to allocate | |
63931eb9 | 5071 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
5072 | * |
5073 | * This function is similar to alloc_pages(), except that it allocates the | |
5074 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
5075 | * allocate memory in power-of-two pages. | |
5076 | * | |
5077 | * This function is also limited by MAX_ORDER. | |
5078 | * | |
5079 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
5080 | * |
5081 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
5082 | */ |
5083 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
5084 | { | |
5085 | unsigned int order = get_order(size); | |
5086 | unsigned long addr; | |
5087 | ||
63931eb9 VB |
5088 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) |
5089 | gfp_mask &= ~__GFP_COMP; | |
5090 | ||
2be0ffe2 | 5091 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 5092 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
5093 | } |
5094 | EXPORT_SYMBOL(alloc_pages_exact); | |
5095 | ||
ee85c2e1 AK |
5096 | /** |
5097 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
5098 | * pages on a node. | |
b5e6ab58 | 5099 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 5100 | * @size: the number of bytes to allocate |
63931eb9 | 5101 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
5102 | * |
5103 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
5104 | * back. | |
a862f68a MR |
5105 | * |
5106 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 5107 | */ |
e1931811 | 5108 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5109 | { |
d00181b9 | 5110 | unsigned int order = get_order(size); |
63931eb9 VB |
5111 | struct page *p; |
5112 | ||
5113 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) | |
5114 | gfp_mask &= ~__GFP_COMP; | |
5115 | ||
5116 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5117 | if (!p) |
5118 | return NULL; | |
5119 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5120 | } | |
ee85c2e1 | 5121 | |
2be0ffe2 TT |
5122 | /** |
5123 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5124 | * @virt: the value returned by alloc_pages_exact. | |
5125 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5126 | * | |
5127 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5128 | */ | |
5129 | void free_pages_exact(void *virt, size_t size) | |
5130 | { | |
5131 | unsigned long addr = (unsigned long)virt; | |
5132 | unsigned long end = addr + PAGE_ALIGN(size); | |
5133 | ||
5134 | while (addr < end) { | |
5135 | free_page(addr); | |
5136 | addr += PAGE_SIZE; | |
5137 | } | |
5138 | } | |
5139 | EXPORT_SYMBOL(free_pages_exact); | |
5140 | ||
e0fb5815 ZY |
5141 | /** |
5142 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5143 | * @offset: The zone index of the highest zone | |
5144 | * | |
a862f68a | 5145 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5146 | * high watermark within all zones at or below a given zone index. For each |
5147 | * zone, the number of pages is calculated as: | |
0e056eb5 | 5148 | * |
5149 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5150 | * |
5151 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5152 | */ |
ebec3862 | 5153 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5154 | { |
dd1a239f | 5155 | struct zoneref *z; |
54a6eb5c MG |
5156 | struct zone *zone; |
5157 | ||
e310fd43 | 5158 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5159 | unsigned long sum = 0; |
1da177e4 | 5160 | |
0e88460d | 5161 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5162 | |
54a6eb5c | 5163 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5164 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5165 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5166 | if (size > high) |
5167 | sum += size - high; | |
1da177e4 LT |
5168 | } |
5169 | ||
5170 | return sum; | |
5171 | } | |
5172 | ||
e0fb5815 ZY |
5173 | /** |
5174 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5175 | * | |
5176 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5177 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5178 | * |
5179 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5180 | * ZONE_NORMAL. | |
1da177e4 | 5181 | */ |
ebec3862 | 5182 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5183 | { |
af4ca457 | 5184 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5185 | } |
c2f1a551 | 5186 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5187 | |
08e0f6a9 | 5188 | static inline void show_node(struct zone *zone) |
1da177e4 | 5189 | { |
e5adfffc | 5190 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5191 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5192 | } |
1da177e4 | 5193 | |
d02bd27b IR |
5194 | long si_mem_available(void) |
5195 | { | |
5196 | long available; | |
5197 | unsigned long pagecache; | |
5198 | unsigned long wmark_low = 0; | |
5199 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5200 | unsigned long reclaimable; |
d02bd27b IR |
5201 | struct zone *zone; |
5202 | int lru; | |
5203 | ||
5204 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5205 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5206 | |
5207 | for_each_zone(zone) | |
a9214443 | 5208 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5209 | |
5210 | /* | |
5211 | * Estimate the amount of memory available for userspace allocations, | |
5212 | * without causing swapping. | |
5213 | */ | |
c41f012a | 5214 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5215 | |
5216 | /* | |
5217 | * Not all the page cache can be freed, otherwise the system will | |
5218 | * start swapping. Assume at least half of the page cache, or the | |
5219 | * low watermark worth of cache, needs to stay. | |
5220 | */ | |
5221 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
5222 | pagecache -= min(pagecache / 2, wmark_low); | |
5223 | available += pagecache; | |
5224 | ||
5225 | /* | |
b29940c1 VB |
5226 | * Part of the reclaimable slab and other kernel memory consists of |
5227 | * items that are in use, and cannot be freed. Cap this estimate at the | |
5228 | * low watermark. | |
d02bd27b | 5229 | */ |
d42f3245 RG |
5230 | reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) + |
5231 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
b29940c1 | 5232 | available += reclaimable - min(reclaimable / 2, wmark_low); |
034ebf65 | 5233 | |
d02bd27b IR |
5234 | if (available < 0) |
5235 | available = 0; | |
5236 | return available; | |
5237 | } | |
5238 | EXPORT_SYMBOL_GPL(si_mem_available); | |
5239 | ||
1da177e4 LT |
5240 | void si_meminfo(struct sysinfo *val) |
5241 | { | |
ca79b0c2 | 5242 | val->totalram = totalram_pages(); |
11fb9989 | 5243 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 5244 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 5245 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 5246 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 5247 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
5248 | val->mem_unit = PAGE_SIZE; |
5249 | } | |
5250 | ||
5251 | EXPORT_SYMBOL(si_meminfo); | |
5252 | ||
5253 | #ifdef CONFIG_NUMA | |
5254 | void si_meminfo_node(struct sysinfo *val, int nid) | |
5255 | { | |
cdd91a77 JL |
5256 | int zone_type; /* needs to be signed */ |
5257 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
5258 | unsigned long managed_highpages = 0; |
5259 | unsigned long free_highpages = 0; | |
1da177e4 LT |
5260 | pg_data_t *pgdat = NODE_DATA(nid); |
5261 | ||
cdd91a77 | 5262 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 5263 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 5264 | val->totalram = managed_pages; |
11fb9989 | 5265 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 5266 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 5267 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
5268 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
5269 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
5270 | ||
5271 | if (is_highmem(zone)) { | |
9705bea5 | 5272 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
5273 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
5274 | } | |
5275 | } | |
5276 | val->totalhigh = managed_highpages; | |
5277 | val->freehigh = free_highpages; | |
98d2b0eb | 5278 | #else |
fc2bd799 JK |
5279 | val->totalhigh = managed_highpages; |
5280 | val->freehigh = free_highpages; | |
98d2b0eb | 5281 | #endif |
1da177e4 LT |
5282 | val->mem_unit = PAGE_SIZE; |
5283 | } | |
5284 | #endif | |
5285 | ||
ddd588b5 | 5286 | /* |
7bf02ea2 DR |
5287 | * Determine whether the node should be displayed or not, depending on whether |
5288 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 5289 | */ |
9af744d7 | 5290 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 5291 | { |
ddd588b5 | 5292 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 5293 | return false; |
ddd588b5 | 5294 | |
9af744d7 MH |
5295 | /* |
5296 | * no node mask - aka implicit memory numa policy. Do not bother with | |
5297 | * the synchronization - read_mems_allowed_begin - because we do not | |
5298 | * have to be precise here. | |
5299 | */ | |
5300 | if (!nodemask) | |
5301 | nodemask = &cpuset_current_mems_allowed; | |
5302 | ||
5303 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
5304 | } |
5305 | ||
1da177e4 LT |
5306 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
5307 | ||
377e4f16 RV |
5308 | static void show_migration_types(unsigned char type) |
5309 | { | |
5310 | static const char types[MIGRATE_TYPES] = { | |
5311 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 5312 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
5313 | [MIGRATE_RECLAIMABLE] = 'E', |
5314 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
5315 | #ifdef CONFIG_CMA |
5316 | [MIGRATE_CMA] = 'C', | |
5317 | #endif | |
194159fb | 5318 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 5319 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 5320 | #endif |
377e4f16 RV |
5321 | }; |
5322 | char tmp[MIGRATE_TYPES + 1]; | |
5323 | char *p = tmp; | |
5324 | int i; | |
5325 | ||
5326 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
5327 | if (type & (1 << i)) | |
5328 | *p++ = types[i]; | |
5329 | } | |
5330 | ||
5331 | *p = '\0'; | |
1f84a18f | 5332 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
5333 | } |
5334 | ||
1da177e4 LT |
5335 | /* |
5336 | * Show free area list (used inside shift_scroll-lock stuff) | |
5337 | * We also calculate the percentage fragmentation. We do this by counting the | |
5338 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
5339 | * |
5340 | * Bits in @filter: | |
5341 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
5342 | * cpuset. | |
1da177e4 | 5343 | */ |
9af744d7 | 5344 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 5345 | { |
d1bfcdb8 | 5346 | unsigned long free_pcp = 0; |
c7241913 | 5347 | int cpu; |
1da177e4 | 5348 | struct zone *zone; |
599d0c95 | 5349 | pg_data_t *pgdat; |
1da177e4 | 5350 | |
ee99c71c | 5351 | for_each_populated_zone(zone) { |
9af744d7 | 5352 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5353 | continue; |
d1bfcdb8 | 5354 | |
761b0677 KK |
5355 | for_each_online_cpu(cpu) |
5356 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
5357 | } |
5358 | ||
a731286d KM |
5359 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
5360 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
8d92890b | 5361 | " unevictable:%lu dirty:%lu writeback:%lu\n" |
d1bfcdb8 | 5362 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a | 5363 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 5364 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
5365 | global_node_page_state(NR_ACTIVE_ANON), |
5366 | global_node_page_state(NR_INACTIVE_ANON), | |
5367 | global_node_page_state(NR_ISOLATED_ANON), | |
5368 | global_node_page_state(NR_ACTIVE_FILE), | |
5369 | global_node_page_state(NR_INACTIVE_FILE), | |
5370 | global_node_page_state(NR_ISOLATED_FILE), | |
5371 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
5372 | global_node_page_state(NR_FILE_DIRTY), |
5373 | global_node_page_state(NR_WRITEBACK), | |
d42f3245 RG |
5374 | global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B), |
5375 | global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B), | |
50658e2e | 5376 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 5377 | global_node_page_state(NR_SHMEM), |
c41f012a MH |
5378 | global_zone_page_state(NR_PAGETABLE), |
5379 | global_zone_page_state(NR_BOUNCE), | |
5380 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 5381 | free_pcp, |
c41f012a | 5382 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 5383 | |
599d0c95 | 5384 | for_each_online_pgdat(pgdat) { |
9af744d7 | 5385 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
5386 | continue; |
5387 | ||
599d0c95 MG |
5388 | printk("Node %d" |
5389 | " active_anon:%lukB" | |
5390 | " inactive_anon:%lukB" | |
5391 | " active_file:%lukB" | |
5392 | " inactive_file:%lukB" | |
5393 | " unevictable:%lukB" | |
5394 | " isolated(anon):%lukB" | |
5395 | " isolated(file):%lukB" | |
50658e2e | 5396 | " mapped:%lukB" |
11fb9989 MG |
5397 | " dirty:%lukB" |
5398 | " writeback:%lukB" | |
5399 | " shmem:%lukB" | |
5400 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5401 | " shmem_thp: %lukB" | |
5402 | " shmem_pmdmapped: %lukB" | |
5403 | " anon_thp: %lukB" | |
5404 | #endif | |
5405 | " writeback_tmp:%lukB" | |
991e7673 SB |
5406 | " kernel_stack:%lukB" |
5407 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5408 | " shadow_call_stack:%lukB" | |
5409 | #endif | |
599d0c95 MG |
5410 | " all_unreclaimable? %s" |
5411 | "\n", | |
5412 | pgdat->node_id, | |
5413 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
5414 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
5415 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
5416 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
5417 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
5418 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
5419 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 5420 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
5421 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
5422 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 5423 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
5424 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5425 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
5426 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
5427 | * HPAGE_PMD_NR), | |
5428 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
5429 | #endif | |
11fb9989 | 5430 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
991e7673 SB |
5431 | node_page_state(pgdat, NR_KERNEL_STACK_KB), |
5432 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5433 | node_page_state(pgdat, NR_KERNEL_SCS_KB), | |
5434 | #endif | |
c73322d0 JW |
5435 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
5436 | "yes" : "no"); | |
599d0c95 MG |
5437 | } |
5438 | ||
ee99c71c | 5439 | for_each_populated_zone(zone) { |
1da177e4 LT |
5440 | int i; |
5441 | ||
9af744d7 | 5442 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5443 | continue; |
d1bfcdb8 KK |
5444 | |
5445 | free_pcp = 0; | |
5446 | for_each_online_cpu(cpu) | |
5447 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
5448 | ||
1da177e4 | 5449 | show_node(zone); |
1f84a18f JP |
5450 | printk(KERN_CONT |
5451 | "%s" | |
1da177e4 LT |
5452 | " free:%lukB" |
5453 | " min:%lukB" | |
5454 | " low:%lukB" | |
5455 | " high:%lukB" | |
e47b346a | 5456 | " reserved_highatomic:%luKB" |
71c799f4 MK |
5457 | " active_anon:%lukB" |
5458 | " inactive_anon:%lukB" | |
5459 | " active_file:%lukB" | |
5460 | " inactive_file:%lukB" | |
5461 | " unevictable:%lukB" | |
5a1c84b4 | 5462 | " writepending:%lukB" |
1da177e4 | 5463 | " present:%lukB" |
9feedc9d | 5464 | " managed:%lukB" |
4a0aa73f | 5465 | " mlocked:%lukB" |
4a0aa73f | 5466 | " pagetables:%lukB" |
4a0aa73f | 5467 | " bounce:%lukB" |
d1bfcdb8 KK |
5468 | " free_pcp:%lukB" |
5469 | " local_pcp:%ukB" | |
d1ce749a | 5470 | " free_cma:%lukB" |
1da177e4 LT |
5471 | "\n", |
5472 | zone->name, | |
88f5acf8 | 5473 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
5474 | K(min_wmark_pages(zone)), |
5475 | K(low_wmark_pages(zone)), | |
5476 | K(high_wmark_pages(zone)), | |
e47b346a | 5477 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
5478 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
5479 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
5480 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
5481 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
5482 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 5483 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 5484 | K(zone->present_pages), |
9705bea5 | 5485 | K(zone_managed_pages(zone)), |
4a0aa73f | 5486 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f | 5487 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 5488 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
5489 | K(free_pcp), |
5490 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 5491 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
5492 | printk("lowmem_reserve[]:"); |
5493 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
5494 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5495 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5496 | } |
5497 | ||
ee99c71c | 5498 | for_each_populated_zone(zone) { |
d00181b9 KS |
5499 | unsigned int order; |
5500 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5501 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5502 | |
9af744d7 | 5503 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5504 | continue; |
1da177e4 | 5505 | show_node(zone); |
1f84a18f | 5506 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5507 | |
5508 | spin_lock_irqsave(&zone->lock, flags); | |
5509 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5510 | struct free_area *area = &zone->free_area[order]; |
5511 | int type; | |
5512 | ||
5513 | nr[order] = area->nr_free; | |
8f9de51a | 5514 | total += nr[order] << order; |
377e4f16 RV |
5515 | |
5516 | types[order] = 0; | |
5517 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 5518 | if (!free_area_empty(area, type)) |
377e4f16 RV |
5519 | types[order] |= 1 << type; |
5520 | } | |
1da177e4 LT |
5521 | } |
5522 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5523 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5524 | printk(KERN_CONT "%lu*%lukB ", |
5525 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5526 | if (nr[order]) |
5527 | show_migration_types(types[order]); | |
5528 | } | |
1f84a18f | 5529 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5530 | } |
5531 | ||
949f7ec5 DR |
5532 | hugetlb_show_meminfo(); |
5533 | ||
11fb9989 | 5534 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5535 | |
1da177e4 LT |
5536 | show_swap_cache_info(); |
5537 | } | |
5538 | ||
19770b32 MG |
5539 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5540 | { | |
5541 | zoneref->zone = zone; | |
5542 | zoneref->zone_idx = zone_idx(zone); | |
5543 | } | |
5544 | ||
1da177e4 LT |
5545 | /* |
5546 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5547 | * |
5548 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5549 | */ |
9d3be21b | 5550 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5551 | { |
1a93205b | 5552 | struct zone *zone; |
bc732f1d | 5553 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5554 | int nr_zones = 0; |
02a68a5e CL |
5555 | |
5556 | do { | |
2f6726e5 | 5557 | zone_type--; |
070f8032 | 5558 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5559 | if (managed_zone(zone)) { |
9d3be21b | 5560 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5561 | check_highest_zone(zone_type); |
1da177e4 | 5562 | } |
2f6726e5 | 5563 | } while (zone_type); |
bc732f1d | 5564 | |
070f8032 | 5565 | return nr_zones; |
1da177e4 LT |
5566 | } |
5567 | ||
5568 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5569 | |
5570 | static int __parse_numa_zonelist_order(char *s) | |
5571 | { | |
c9bff3ee MH |
5572 | /* |
5573 | * We used to support different zonlists modes but they turned | |
5574 | * out to be just not useful. Let's keep the warning in place | |
5575 | * if somebody still use the cmd line parameter so that we do | |
5576 | * not fail it silently | |
5577 | */ | |
5578 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5579 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5580 | return -EINVAL; |
5581 | } | |
5582 | return 0; | |
5583 | } | |
5584 | ||
c9bff3ee MH |
5585 | char numa_zonelist_order[] = "Node"; |
5586 | ||
f0c0b2b8 KH |
5587 | /* |
5588 | * sysctl handler for numa_zonelist_order | |
5589 | */ | |
cccad5b9 | 5590 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 5591 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 5592 | { |
32927393 CH |
5593 | if (write) |
5594 | return __parse_numa_zonelist_order(buffer); | |
5595 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
5596 | } |
5597 | ||
5598 | ||
62bc62a8 | 5599 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5600 | static int node_load[MAX_NUMNODES]; |
5601 | ||
1da177e4 | 5602 | /** |
4dc3b16b | 5603 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5604 | * @node: node whose fallback list we're appending |
5605 | * @used_node_mask: nodemask_t of already used nodes | |
5606 | * | |
5607 | * We use a number of factors to determine which is the next node that should | |
5608 | * appear on a given node's fallback list. The node should not have appeared | |
5609 | * already in @node's fallback list, and it should be the next closest node | |
5610 | * according to the distance array (which contains arbitrary distance values | |
5611 | * from each node to each node in the system), and should also prefer nodes | |
5612 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5613 | * on them otherwise. | |
a862f68a MR |
5614 | * |
5615 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 5616 | */ |
f0c0b2b8 | 5617 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5618 | { |
4cf808eb | 5619 | int n, val; |
1da177e4 | 5620 | int min_val = INT_MAX; |
00ef2d2f | 5621 | int best_node = NUMA_NO_NODE; |
a70f7302 | 5622 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 5623 | |
4cf808eb LT |
5624 | /* Use the local node if we haven't already */ |
5625 | if (!node_isset(node, *used_node_mask)) { | |
5626 | node_set(node, *used_node_mask); | |
5627 | return node; | |
5628 | } | |
1da177e4 | 5629 | |
4b0ef1fe | 5630 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5631 | |
5632 | /* Don't want a node to appear more than once */ | |
5633 | if (node_isset(n, *used_node_mask)) | |
5634 | continue; | |
5635 | ||
1da177e4 LT |
5636 | /* Use the distance array to find the distance */ |
5637 | val = node_distance(node, n); | |
5638 | ||
4cf808eb LT |
5639 | /* Penalize nodes under us ("prefer the next node") */ |
5640 | val += (n < node); | |
5641 | ||
1da177e4 | 5642 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
5643 | tmp = cpumask_of_node(n); |
5644 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
5645 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5646 | ||
5647 | /* Slight preference for less loaded node */ | |
5648 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
5649 | val += node_load[n]; | |
5650 | ||
5651 | if (val < min_val) { | |
5652 | min_val = val; | |
5653 | best_node = n; | |
5654 | } | |
5655 | } | |
5656 | ||
5657 | if (best_node >= 0) | |
5658 | node_set(best_node, *used_node_mask); | |
5659 | ||
5660 | return best_node; | |
5661 | } | |
5662 | ||
f0c0b2b8 KH |
5663 | |
5664 | /* | |
5665 | * Build zonelists ordered by node and zones within node. | |
5666 | * This results in maximum locality--normal zone overflows into local | |
5667 | * DMA zone, if any--but risks exhausting DMA zone. | |
5668 | */ | |
9d3be21b MH |
5669 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5670 | unsigned nr_nodes) | |
1da177e4 | 5671 | { |
9d3be21b MH |
5672 | struct zoneref *zonerefs; |
5673 | int i; | |
5674 | ||
5675 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5676 | ||
5677 | for (i = 0; i < nr_nodes; i++) { | |
5678 | int nr_zones; | |
5679 | ||
5680 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5681 | |
9d3be21b MH |
5682 | nr_zones = build_zonerefs_node(node, zonerefs); |
5683 | zonerefs += nr_zones; | |
5684 | } | |
5685 | zonerefs->zone = NULL; | |
5686 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5687 | } |
5688 | ||
523b9458 CL |
5689 | /* |
5690 | * Build gfp_thisnode zonelists | |
5691 | */ | |
5692 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5693 | { | |
9d3be21b MH |
5694 | struct zoneref *zonerefs; |
5695 | int nr_zones; | |
523b9458 | 5696 | |
9d3be21b MH |
5697 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5698 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5699 | zonerefs += nr_zones; | |
5700 | zonerefs->zone = NULL; | |
5701 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5702 | } |
5703 | ||
f0c0b2b8 KH |
5704 | /* |
5705 | * Build zonelists ordered by zone and nodes within zones. | |
5706 | * This results in conserving DMA zone[s] until all Normal memory is | |
5707 | * exhausted, but results in overflowing to remote node while memory | |
5708 | * may still exist in local DMA zone. | |
5709 | */ | |
f0c0b2b8 | 5710 | |
f0c0b2b8 KH |
5711 | static void build_zonelists(pg_data_t *pgdat) |
5712 | { | |
9d3be21b MH |
5713 | static int node_order[MAX_NUMNODES]; |
5714 | int node, load, nr_nodes = 0; | |
d0ddf49b | 5715 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 5716 | int local_node, prev_node; |
1da177e4 LT |
5717 | |
5718 | /* NUMA-aware ordering of nodes */ | |
5719 | local_node = pgdat->node_id; | |
62bc62a8 | 5720 | load = nr_online_nodes; |
1da177e4 | 5721 | prev_node = local_node; |
f0c0b2b8 | 5722 | |
f0c0b2b8 | 5723 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5724 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5725 | /* | |
5726 | * We don't want to pressure a particular node. | |
5727 | * So adding penalty to the first node in same | |
5728 | * distance group to make it round-robin. | |
5729 | */ | |
957f822a DR |
5730 | if (node_distance(local_node, node) != |
5731 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5732 | node_load[node] = load; |
5733 | ||
9d3be21b | 5734 | node_order[nr_nodes++] = node; |
1da177e4 LT |
5735 | prev_node = node; |
5736 | load--; | |
1da177e4 | 5737 | } |
523b9458 | 5738 | |
9d3be21b | 5739 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5740 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
5741 | } |
5742 | ||
7aac7898 LS |
5743 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5744 | /* | |
5745 | * Return node id of node used for "local" allocations. | |
5746 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5747 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5748 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5749 | */ | |
5750 | int local_memory_node(int node) | |
5751 | { | |
c33d6c06 | 5752 | struct zoneref *z; |
7aac7898 | 5753 | |
c33d6c06 | 5754 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5755 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 5756 | NULL); |
c1093b74 | 5757 | return zone_to_nid(z->zone); |
7aac7898 LS |
5758 | } |
5759 | #endif | |
f0c0b2b8 | 5760 | |
6423aa81 JK |
5761 | static void setup_min_unmapped_ratio(void); |
5762 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5763 | #else /* CONFIG_NUMA */ |
5764 | ||
f0c0b2b8 | 5765 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5766 | { |
19655d34 | 5767 | int node, local_node; |
9d3be21b MH |
5768 | struct zoneref *zonerefs; |
5769 | int nr_zones; | |
1da177e4 LT |
5770 | |
5771 | local_node = pgdat->node_id; | |
1da177e4 | 5772 | |
9d3be21b MH |
5773 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5774 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5775 | zonerefs += nr_zones; | |
1da177e4 | 5776 | |
54a6eb5c MG |
5777 | /* |
5778 | * Now we build the zonelist so that it contains the zones | |
5779 | * of all the other nodes. | |
5780 | * We don't want to pressure a particular node, so when | |
5781 | * building the zones for node N, we make sure that the | |
5782 | * zones coming right after the local ones are those from | |
5783 | * node N+1 (modulo N) | |
5784 | */ | |
5785 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5786 | if (!node_online(node)) | |
5787 | continue; | |
9d3be21b MH |
5788 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5789 | zonerefs += nr_zones; | |
1da177e4 | 5790 | } |
54a6eb5c MG |
5791 | for (node = 0; node < local_node; node++) { |
5792 | if (!node_online(node)) | |
5793 | continue; | |
9d3be21b MH |
5794 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5795 | zonerefs += nr_zones; | |
54a6eb5c MG |
5796 | } |
5797 | ||
9d3be21b MH |
5798 | zonerefs->zone = NULL; |
5799 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5800 | } |
5801 | ||
5802 | #endif /* CONFIG_NUMA */ | |
5803 | ||
99dcc3e5 CL |
5804 | /* |
5805 | * Boot pageset table. One per cpu which is going to be used for all | |
5806 | * zones and all nodes. The parameters will be set in such a way | |
5807 | * that an item put on a list will immediately be handed over to | |
5808 | * the buddy list. This is safe since pageset manipulation is done | |
5809 | * with interrupts disabled. | |
5810 | * | |
5811 | * The boot_pagesets must be kept even after bootup is complete for | |
5812 | * unused processors and/or zones. They do play a role for bootstrapping | |
5813 | * hotplugged processors. | |
5814 | * | |
5815 | * zoneinfo_show() and maybe other functions do | |
5816 | * not check if the processor is online before following the pageset pointer. | |
5817 | * Other parts of the kernel may not check if the zone is available. | |
5818 | */ | |
5819 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5820 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
385386cf | 5821 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 5822 | |
11cd8638 | 5823 | static void __build_all_zonelists(void *data) |
1da177e4 | 5824 | { |
6811378e | 5825 | int nid; |
afb6ebb3 | 5826 | int __maybe_unused cpu; |
9adb62a5 | 5827 | pg_data_t *self = data; |
b93e0f32 MH |
5828 | static DEFINE_SPINLOCK(lock); |
5829 | ||
5830 | spin_lock(&lock); | |
9276b1bc | 5831 | |
7f9cfb31 BL |
5832 | #ifdef CONFIG_NUMA |
5833 | memset(node_load, 0, sizeof(node_load)); | |
5834 | #endif | |
9adb62a5 | 5835 | |
c1152583 WY |
5836 | /* |
5837 | * This node is hotadded and no memory is yet present. So just | |
5838 | * building zonelists is fine - no need to touch other nodes. | |
5839 | */ | |
9adb62a5 JL |
5840 | if (self && !node_online(self->node_id)) { |
5841 | build_zonelists(self); | |
c1152583 WY |
5842 | } else { |
5843 | for_each_online_node(nid) { | |
5844 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 5845 | |
c1152583 WY |
5846 | build_zonelists(pgdat); |
5847 | } | |
99dcc3e5 | 5848 | |
7aac7898 LS |
5849 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5850 | /* | |
5851 | * We now know the "local memory node" for each node-- | |
5852 | * i.e., the node of the first zone in the generic zonelist. | |
5853 | * Set up numa_mem percpu variable for on-line cpus. During | |
5854 | * boot, only the boot cpu should be on-line; we'll init the | |
5855 | * secondary cpus' numa_mem as they come on-line. During | |
5856 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5857 | */ | |
d9c9a0b9 | 5858 | for_each_online_cpu(cpu) |
7aac7898 | 5859 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5860 | #endif |
d9c9a0b9 | 5861 | } |
b93e0f32 MH |
5862 | |
5863 | spin_unlock(&lock); | |
6811378e YG |
5864 | } |
5865 | ||
061f67bc RV |
5866 | static noinline void __init |
5867 | build_all_zonelists_init(void) | |
5868 | { | |
afb6ebb3 MH |
5869 | int cpu; |
5870 | ||
061f67bc | 5871 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5872 | |
5873 | /* | |
5874 | * Initialize the boot_pagesets that are going to be used | |
5875 | * for bootstrapping processors. The real pagesets for | |
5876 | * each zone will be allocated later when the per cpu | |
5877 | * allocator is available. | |
5878 | * | |
5879 | * boot_pagesets are used also for bootstrapping offline | |
5880 | * cpus if the system is already booted because the pagesets | |
5881 | * are needed to initialize allocators on a specific cpu too. | |
5882 | * F.e. the percpu allocator needs the page allocator which | |
5883 | * needs the percpu allocator in order to allocate its pagesets | |
5884 | * (a chicken-egg dilemma). | |
5885 | */ | |
5886 | for_each_possible_cpu(cpu) | |
5887 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); | |
5888 | ||
061f67bc RV |
5889 | mminit_verify_zonelist(); |
5890 | cpuset_init_current_mems_allowed(); | |
5891 | } | |
5892 | ||
4eaf3f64 | 5893 | /* |
4eaf3f64 | 5894 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5895 | * |
72675e13 | 5896 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5897 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5898 | */ |
72675e13 | 5899 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 5900 | { |
0a18e607 DH |
5901 | unsigned long vm_total_pages; |
5902 | ||
6811378e | 5903 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 5904 | build_all_zonelists_init(); |
6811378e | 5905 | } else { |
11cd8638 | 5906 | __build_all_zonelists(pgdat); |
6811378e YG |
5907 | /* cpuset refresh routine should be here */ |
5908 | } | |
56b9413b DH |
5909 | /* Get the number of free pages beyond high watermark in all zones. */ |
5910 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
5911 | /* |
5912 | * Disable grouping by mobility if the number of pages in the | |
5913 | * system is too low to allow the mechanism to work. It would be | |
5914 | * more accurate, but expensive to check per-zone. This check is | |
5915 | * made on memory-hotadd so a system can start with mobility | |
5916 | * disabled and enable it later | |
5917 | */ | |
d9c23400 | 5918 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5919 | page_group_by_mobility_disabled = 1; |
5920 | else | |
5921 | page_group_by_mobility_disabled = 0; | |
5922 | ||
ce0725f7 | 5923 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5924 | nr_online_nodes, |
756a025f JP |
5925 | page_group_by_mobility_disabled ? "off" : "on", |
5926 | vm_total_pages); | |
f0c0b2b8 | 5927 | #ifdef CONFIG_NUMA |
f88dfff5 | 5928 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5929 | #endif |
1da177e4 LT |
5930 | } |
5931 | ||
a9a9e77f PT |
5932 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
5933 | static bool __meminit | |
5934 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
5935 | { | |
a9a9e77f PT |
5936 | static struct memblock_region *r; |
5937 | ||
5938 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5939 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
5940 | for_each_memblock(memory, r) { | |
5941 | if (*pfn < memblock_region_memory_end_pfn(r)) | |
5942 | break; | |
5943 | } | |
5944 | } | |
5945 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
5946 | memblock_is_mirror(r)) { | |
5947 | *pfn = memblock_region_memory_end_pfn(r); | |
5948 | return true; | |
5949 | } | |
5950 | } | |
a9a9e77f PT |
5951 | return false; |
5952 | } | |
5953 | ||
1da177e4 LT |
5954 | /* |
5955 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 5956 | * up by memblock_free_all() once the early boot process is |
1da177e4 LT |
5957 | * done. Non-atomic initialization, single-pass. |
5958 | */ | |
c09b4240 | 5959 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a99583e7 CH |
5960 | unsigned long start_pfn, enum memmap_context context, |
5961 | struct vmem_altmap *altmap) | |
1da177e4 | 5962 | { |
a9a9e77f | 5963 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 5964 | struct page *page; |
1da177e4 | 5965 | |
22b31eec HD |
5966 | if (highest_memmap_pfn < end_pfn - 1) |
5967 | highest_memmap_pfn = end_pfn - 1; | |
5968 | ||
966cf44f | 5969 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
5970 | /* |
5971 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
5972 | * memory. We limit the total number of pages to initialize to just |
5973 | * those that might contain the memory mapping. We will defer the | |
5974 | * ZONE_DEVICE page initialization until after we have released | |
5975 | * the hotplug lock. | |
4b94ffdc | 5976 | */ |
966cf44f AD |
5977 | if (zone == ZONE_DEVICE) { |
5978 | if (!altmap) | |
5979 | return; | |
5980 | ||
5981 | if (start_pfn == altmap->base_pfn) | |
5982 | start_pfn += altmap->reserve; | |
5983 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
5984 | } | |
5985 | #endif | |
4b94ffdc | 5986 | |
948c436e | 5987 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 5988 | /* |
b72d0ffb AM |
5989 | * There can be holes in boot-time mem_map[]s handed to this |
5990 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5991 | */ |
a9a9e77f | 5992 | if (context == MEMMAP_EARLY) { |
a9a9e77f PT |
5993 | if (overlap_memmap_init(zone, &pfn)) |
5994 | continue; | |
5995 | if (defer_init(nid, pfn, end_pfn)) | |
5996 | break; | |
a2f3aa02 | 5997 | } |
ac5d2539 | 5998 | |
d0dc12e8 PT |
5999 | page = pfn_to_page(pfn); |
6000 | __init_single_page(page, pfn, zone, nid); | |
6001 | if (context == MEMMAP_HOTPLUG) | |
d483da5b | 6002 | __SetPageReserved(page); |
d0dc12e8 | 6003 | |
ac5d2539 MG |
6004 | /* |
6005 | * Mark the block movable so that blocks are reserved for | |
6006 | * movable at startup. This will force kernel allocations | |
6007 | * to reserve their blocks rather than leaking throughout | |
6008 | * the address space during boot when many long-lived | |
974a786e | 6009 | * kernel allocations are made. |
ac5d2539 MG |
6010 | * |
6011 | * bitmap is created for zone's valid pfn range. but memmap | |
6012 | * can be created for invalid pages (for alignment) | |
6013 | * check here not to call set_pageblock_migratetype() against | |
6014 | * pfn out of zone. | |
6015 | */ | |
6016 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
ac5d2539 | 6017 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
9b6e63cb | 6018 | cond_resched(); |
ac5d2539 | 6019 | } |
948c436e | 6020 | pfn++; |
1da177e4 LT |
6021 | } |
6022 | } | |
6023 | ||
966cf44f AD |
6024 | #ifdef CONFIG_ZONE_DEVICE |
6025 | void __ref memmap_init_zone_device(struct zone *zone, | |
6026 | unsigned long start_pfn, | |
1f8d75c1 | 6027 | unsigned long nr_pages, |
966cf44f AD |
6028 | struct dev_pagemap *pgmap) |
6029 | { | |
1f8d75c1 | 6030 | unsigned long pfn, end_pfn = start_pfn + nr_pages; |
966cf44f | 6031 | struct pglist_data *pgdat = zone->zone_pgdat; |
514caf23 | 6032 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
966cf44f AD |
6033 | unsigned long zone_idx = zone_idx(zone); |
6034 | unsigned long start = jiffies; | |
6035 | int nid = pgdat->node_id; | |
6036 | ||
46d945ae | 6037 | if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE)) |
966cf44f AD |
6038 | return; |
6039 | ||
6040 | /* | |
6041 | * The call to memmap_init_zone should have already taken care | |
6042 | * of the pages reserved for the memmap, so we can just jump to | |
6043 | * the end of that region and start processing the device pages. | |
6044 | */ | |
514caf23 | 6045 | if (altmap) { |
966cf44f | 6046 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
1f8d75c1 | 6047 | nr_pages = end_pfn - start_pfn; |
966cf44f AD |
6048 | } |
6049 | ||
6050 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
6051 | struct page *page = pfn_to_page(pfn); | |
6052 | ||
6053 | __init_single_page(page, pfn, zone_idx, nid); | |
6054 | ||
6055 | /* | |
6056 | * Mark page reserved as it will need to wait for onlining | |
6057 | * phase for it to be fully associated with a zone. | |
6058 | * | |
6059 | * We can use the non-atomic __set_bit operation for setting | |
6060 | * the flag as we are still initializing the pages. | |
6061 | */ | |
6062 | __SetPageReserved(page); | |
6063 | ||
6064 | /* | |
8a164fef CH |
6065 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer |
6066 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
6067 | * ever freed or placed on a driver-private list. | |
966cf44f AD |
6068 | */ |
6069 | page->pgmap = pgmap; | |
8a164fef | 6070 | page->zone_device_data = NULL; |
966cf44f AD |
6071 | |
6072 | /* | |
6073 | * Mark the block movable so that blocks are reserved for | |
6074 | * movable at startup. This will force kernel allocations | |
6075 | * to reserve their blocks rather than leaking throughout | |
6076 | * the address space during boot when many long-lived | |
6077 | * kernel allocations are made. | |
6078 | * | |
6079 | * bitmap is created for zone's valid pfn range. but memmap | |
6080 | * can be created for invalid pages (for alignment) | |
6081 | * check here not to call set_pageblock_migratetype() against | |
6082 | * pfn out of zone. | |
6083 | * | |
6084 | * Please note that MEMMAP_HOTPLUG path doesn't clear memmap | |
ba72b4c8 | 6085 | * because this is done early in section_activate() |
966cf44f AD |
6086 | */ |
6087 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
6088 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
6089 | cond_resched(); | |
6090 | } | |
6091 | } | |
6092 | ||
fdc029b1 | 6093 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
1f8d75c1 | 6094 | nr_pages, jiffies_to_msecs(jiffies - start)); |
966cf44f AD |
6095 | } |
6096 | ||
6097 | #endif | |
1e548deb | 6098 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6099 | { |
7aeb09f9 | 6100 | unsigned int order, t; |
b2a0ac88 MG |
6101 | for_each_migratetype_order(order, t) { |
6102 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6103 | zone->free_area[order].nr_free = 0; |
6104 | } | |
6105 | } | |
6106 | ||
dfb3ccd0 | 6107 | void __meminit __weak memmap_init(unsigned long size, int nid, |
73a6e474 BH |
6108 | unsigned long zone, |
6109 | unsigned long range_start_pfn) | |
dfb3ccd0 | 6110 | { |
73a6e474 BH |
6111 | unsigned long start_pfn, end_pfn; |
6112 | unsigned long range_end_pfn = range_start_pfn + size; | |
6113 | int i; | |
6114 | ||
6115 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { | |
6116 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6117 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6118 | ||
6119 | if (end_pfn > start_pfn) { | |
6120 | size = end_pfn - start_pfn; | |
6121 | memmap_init_zone(size, nid, zone, start_pfn, | |
6122 | MEMMAP_EARLY, NULL); | |
6123 | } | |
6124 | } | |
dfb3ccd0 | 6125 | } |
1da177e4 | 6126 | |
7cd2b0a3 | 6127 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 6128 | { |
3a6be87f | 6129 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
6130 | int batch; |
6131 | ||
6132 | /* | |
6133 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 6134 | * size of the zone. |
e7c8d5c9 | 6135 | */ |
9705bea5 | 6136 | batch = zone_managed_pages(zone) / 1024; |
d8a759b5 AL |
6137 | /* But no more than a meg. */ |
6138 | if (batch * PAGE_SIZE > 1024 * 1024) | |
6139 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
6140 | batch /= 4; /* We effectively *= 4 below */ |
6141 | if (batch < 1) | |
6142 | batch = 1; | |
6143 | ||
6144 | /* | |
0ceaacc9 NP |
6145 | * Clamp the batch to a 2^n - 1 value. Having a power |
6146 | * of 2 value was found to be more likely to have | |
6147 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 6148 | * |
0ceaacc9 NP |
6149 | * For example if 2 tasks are alternately allocating |
6150 | * batches of pages, one task can end up with a lot | |
6151 | * of pages of one half of the possible page colors | |
6152 | * and the other with pages of the other colors. | |
e7c8d5c9 | 6153 | */ |
9155203a | 6154 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 6155 | |
e7c8d5c9 | 6156 | return batch; |
3a6be87f DH |
6157 | |
6158 | #else | |
6159 | /* The deferral and batching of frees should be suppressed under NOMMU | |
6160 | * conditions. | |
6161 | * | |
6162 | * The problem is that NOMMU needs to be able to allocate large chunks | |
6163 | * of contiguous memory as there's no hardware page translation to | |
6164 | * assemble apparent contiguous memory from discontiguous pages. | |
6165 | * | |
6166 | * Queueing large contiguous runs of pages for batching, however, | |
6167 | * causes the pages to actually be freed in smaller chunks. As there | |
6168 | * can be a significant delay between the individual batches being | |
6169 | * recycled, this leads to the once large chunks of space being | |
6170 | * fragmented and becoming unavailable for high-order allocations. | |
6171 | */ | |
6172 | return 0; | |
6173 | #endif | |
e7c8d5c9 CL |
6174 | } |
6175 | ||
8d7a8fa9 CS |
6176 | /* |
6177 | * pcp->high and pcp->batch values are related and dependent on one another: | |
6178 | * ->batch must never be higher then ->high. | |
6179 | * The following function updates them in a safe manner without read side | |
6180 | * locking. | |
6181 | * | |
6182 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
6183 | * those fields changing asynchronously (acording the the above rule). | |
6184 | * | |
6185 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
6186 | * outside of boot time (or some other assurance that no concurrent updaters | |
6187 | * exist). | |
6188 | */ | |
6189 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
6190 | unsigned long batch) | |
6191 | { | |
6192 | /* start with a fail safe value for batch */ | |
6193 | pcp->batch = 1; | |
6194 | smp_wmb(); | |
6195 | ||
6196 | /* Update high, then batch, in order */ | |
6197 | pcp->high = high; | |
6198 | smp_wmb(); | |
6199 | ||
6200 | pcp->batch = batch; | |
6201 | } | |
6202 | ||
3664033c | 6203 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
6204 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
6205 | { | |
8d7a8fa9 | 6206 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
6207 | } |
6208 | ||
88c90dbc | 6209 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
6210 | { |
6211 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 6212 | int migratetype; |
2caaad41 | 6213 | |
1c6fe946 MD |
6214 | memset(p, 0, sizeof(*p)); |
6215 | ||
3dfa5721 | 6216 | pcp = &p->pcp; |
5f8dcc21 MG |
6217 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
6218 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
6219 | } |
6220 | ||
88c90dbc CS |
6221 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
6222 | { | |
6223 | pageset_init(p); | |
6224 | pageset_set_batch(p, batch); | |
6225 | } | |
6226 | ||
8ad4b1fb | 6227 | /* |
3664033c | 6228 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
6229 | * to the value high for the pageset p. |
6230 | */ | |
3664033c | 6231 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
6232 | unsigned long high) |
6233 | { | |
8d7a8fa9 CS |
6234 | unsigned long batch = max(1UL, high / 4); |
6235 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
6236 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 6237 | |
8d7a8fa9 | 6238 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
6239 | } |
6240 | ||
7cd2b0a3 DR |
6241 | static void pageset_set_high_and_batch(struct zone *zone, |
6242 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 6243 | { |
56cef2b8 | 6244 | if (percpu_pagelist_fraction) |
3664033c | 6245 | pageset_set_high(pcp, |
9705bea5 | 6246 | (zone_managed_pages(zone) / |
56cef2b8 CS |
6247 | percpu_pagelist_fraction)); |
6248 | else | |
6249 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
6250 | } | |
6251 | ||
169f6c19 CS |
6252 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
6253 | { | |
6254 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
6255 | ||
6256 | pageset_init(pcp); | |
6257 | pageset_set_high_and_batch(zone, pcp); | |
6258 | } | |
6259 | ||
72675e13 | 6260 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
6261 | { |
6262 | int cpu; | |
319774e2 | 6263 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
6264 | for_each_possible_cpu(cpu) |
6265 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
6266 | } |
6267 | ||
2caaad41 | 6268 | /* |
99dcc3e5 CL |
6269 | * Allocate per cpu pagesets and initialize them. |
6270 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 6271 | */ |
99dcc3e5 | 6272 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 6273 | { |
b4911ea2 | 6274 | struct pglist_data *pgdat; |
99dcc3e5 | 6275 | struct zone *zone; |
b418a0f9 | 6276 | int __maybe_unused cpu; |
e7c8d5c9 | 6277 | |
319774e2 WF |
6278 | for_each_populated_zone(zone) |
6279 | setup_zone_pageset(zone); | |
b4911ea2 | 6280 | |
b418a0f9 SD |
6281 | #ifdef CONFIG_NUMA |
6282 | /* | |
6283 | * Unpopulated zones continue using the boot pagesets. | |
6284 | * The numa stats for these pagesets need to be reset. | |
6285 | * Otherwise, they will end up skewing the stats of | |
6286 | * the nodes these zones are associated with. | |
6287 | */ | |
6288 | for_each_possible_cpu(cpu) { | |
6289 | struct per_cpu_pageset *pcp = &per_cpu(boot_pageset, cpu); | |
6290 | memset(pcp->vm_numa_stat_diff, 0, | |
6291 | sizeof(pcp->vm_numa_stat_diff)); | |
6292 | } | |
6293 | #endif | |
6294 | ||
b4911ea2 MG |
6295 | for_each_online_pgdat(pgdat) |
6296 | pgdat->per_cpu_nodestats = | |
6297 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
6298 | } |
6299 | ||
c09b4240 | 6300 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 6301 | { |
99dcc3e5 CL |
6302 | /* |
6303 | * per cpu subsystem is not up at this point. The following code | |
6304 | * relies on the ability of the linker to provide the | |
6305 | * offset of a (static) per cpu variable into the per cpu area. | |
6306 | */ | |
6307 | zone->pageset = &boot_pageset; | |
ed8ece2e | 6308 | |
b38a8725 | 6309 | if (populated_zone(zone)) |
99dcc3e5 CL |
6310 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
6311 | zone->name, zone->present_pages, | |
6312 | zone_batchsize(zone)); | |
ed8ece2e DH |
6313 | } |
6314 | ||
dc0bbf3b | 6315 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 6316 | unsigned long zone_start_pfn, |
b171e409 | 6317 | unsigned long size) |
ed8ece2e DH |
6318 | { |
6319 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 6320 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 6321 | |
8f416836 WY |
6322 | if (zone_idx > pgdat->nr_zones) |
6323 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 6324 | |
ed8ece2e DH |
6325 | zone->zone_start_pfn = zone_start_pfn; |
6326 | ||
708614e6 MG |
6327 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
6328 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
6329 | pgdat->node_id, | |
6330 | (unsigned long)zone_idx(zone), | |
6331 | zone_start_pfn, (zone_start_pfn + size)); | |
6332 | ||
1e548deb | 6333 | zone_init_free_lists(zone); |
9dcb8b68 | 6334 | zone->initialized = 1; |
ed8ece2e DH |
6335 | } |
6336 | ||
c713216d MG |
6337 | /** |
6338 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
6339 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
6340 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
6341 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
6342 | * |
6343 | * It returns the start and end page frame of a node based on information | |
7d018176 | 6344 | * provided by memblock_set_node(). If called for a node |
c713216d | 6345 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 6346 | * PFNs will be 0. |
c713216d | 6347 | */ |
bbe5d993 | 6348 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
6349 | unsigned long *start_pfn, unsigned long *end_pfn) |
6350 | { | |
c13291a5 | 6351 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 6352 | int i; |
c13291a5 | 6353 | |
c713216d MG |
6354 | *start_pfn = -1UL; |
6355 | *end_pfn = 0; | |
6356 | ||
c13291a5 TH |
6357 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
6358 | *start_pfn = min(*start_pfn, this_start_pfn); | |
6359 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
6360 | } |
6361 | ||
633c0666 | 6362 | if (*start_pfn == -1UL) |
c713216d | 6363 | *start_pfn = 0; |
c713216d MG |
6364 | } |
6365 | ||
2a1e274a MG |
6366 | /* |
6367 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
6368 | * assumption is made that zones within a node are ordered in monotonic | |
6369 | * increasing memory addresses so that the "highest" populated zone is used | |
6370 | */ | |
b69a7288 | 6371 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
6372 | { |
6373 | int zone_index; | |
6374 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
6375 | if (zone_index == ZONE_MOVABLE) | |
6376 | continue; | |
6377 | ||
6378 | if (arch_zone_highest_possible_pfn[zone_index] > | |
6379 | arch_zone_lowest_possible_pfn[zone_index]) | |
6380 | break; | |
6381 | } | |
6382 | ||
6383 | VM_BUG_ON(zone_index == -1); | |
6384 | movable_zone = zone_index; | |
6385 | } | |
6386 | ||
6387 | /* | |
6388 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 6389 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
6390 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
6391 | * in each node depending on the size of each node and how evenly kernelcore | |
6392 | * is distributed. This helper function adjusts the zone ranges | |
6393 | * provided by the architecture for a given node by using the end of the | |
6394 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
6395 | * zones within a node are in order of monotonic increases memory addresses | |
6396 | */ | |
bbe5d993 | 6397 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
6398 | unsigned long zone_type, |
6399 | unsigned long node_start_pfn, | |
6400 | unsigned long node_end_pfn, | |
6401 | unsigned long *zone_start_pfn, | |
6402 | unsigned long *zone_end_pfn) | |
6403 | { | |
6404 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
6405 | if (zone_movable_pfn[nid]) { | |
6406 | /* Size ZONE_MOVABLE */ | |
6407 | if (zone_type == ZONE_MOVABLE) { | |
6408 | *zone_start_pfn = zone_movable_pfn[nid]; | |
6409 | *zone_end_pfn = min(node_end_pfn, | |
6410 | arch_zone_highest_possible_pfn[movable_zone]); | |
6411 | ||
e506b996 XQ |
6412 | /* Adjust for ZONE_MOVABLE starting within this range */ |
6413 | } else if (!mirrored_kernelcore && | |
6414 | *zone_start_pfn < zone_movable_pfn[nid] && | |
6415 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
6416 | *zone_end_pfn = zone_movable_pfn[nid]; | |
6417 | ||
2a1e274a MG |
6418 | /* Check if this whole range is within ZONE_MOVABLE */ |
6419 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
6420 | *zone_start_pfn = *zone_end_pfn; | |
6421 | } | |
6422 | } | |
6423 | ||
c713216d MG |
6424 | /* |
6425 | * Return the number of pages a zone spans in a node, including holes | |
6426 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
6427 | */ | |
bbe5d993 | 6428 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 6429 | unsigned long zone_type, |
7960aedd ZY |
6430 | unsigned long node_start_pfn, |
6431 | unsigned long node_end_pfn, | |
d91749c1 | 6432 | unsigned long *zone_start_pfn, |
854e8848 | 6433 | unsigned long *zone_end_pfn) |
c713216d | 6434 | { |
299c83dc LF |
6435 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6436 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 6437 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6438 | if (!node_start_pfn && !node_end_pfn) |
6439 | return 0; | |
6440 | ||
7960aedd | 6441 | /* Get the start and end of the zone */ |
299c83dc LF |
6442 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6443 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
6444 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6445 | node_start_pfn, node_end_pfn, | |
d91749c1 | 6446 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
6447 | |
6448 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 6449 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
6450 | return 0; |
6451 | ||
6452 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
6453 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
6454 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
6455 | |
6456 | /* Return the spanned pages */ | |
d91749c1 | 6457 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
6458 | } |
6459 | ||
6460 | /* | |
6461 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 6462 | * then all holes in the requested range will be accounted for. |
c713216d | 6463 | */ |
bbe5d993 | 6464 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
6465 | unsigned long range_start_pfn, |
6466 | unsigned long range_end_pfn) | |
6467 | { | |
96e907d1 TH |
6468 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
6469 | unsigned long start_pfn, end_pfn; | |
6470 | int i; | |
c713216d | 6471 | |
96e907d1 TH |
6472 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6473 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6474 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6475 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 6476 | } |
96e907d1 | 6477 | return nr_absent; |
c713216d MG |
6478 | } |
6479 | ||
6480 | /** | |
6481 | * absent_pages_in_range - Return number of page frames in holes within a range | |
6482 | * @start_pfn: The start PFN to start searching for holes | |
6483 | * @end_pfn: The end PFN to stop searching for holes | |
6484 | * | |
a862f68a | 6485 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
6486 | */ |
6487 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
6488 | unsigned long end_pfn) | |
6489 | { | |
6490 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
6491 | } | |
6492 | ||
6493 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 6494 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 6495 | unsigned long zone_type, |
7960aedd | 6496 | unsigned long node_start_pfn, |
854e8848 | 6497 | unsigned long node_end_pfn) |
c713216d | 6498 | { |
96e907d1 TH |
6499 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6500 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 6501 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 6502 | unsigned long nr_absent; |
9c7cd687 | 6503 | |
b5685e92 | 6504 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6505 | if (!node_start_pfn && !node_end_pfn) |
6506 | return 0; | |
6507 | ||
96e907d1 TH |
6508 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6509 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 6510 | |
2a1e274a MG |
6511 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6512 | node_start_pfn, node_end_pfn, | |
6513 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
6514 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
6515 | ||
6516 | /* | |
6517 | * ZONE_MOVABLE handling. | |
6518 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6519 | * and vice versa. | |
6520 | */ | |
e506b996 XQ |
6521 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6522 | unsigned long start_pfn, end_pfn; | |
6523 | struct memblock_region *r; | |
6524 | ||
6525 | for_each_memblock(memory, r) { | |
6526 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
6527 | zone_start_pfn, zone_end_pfn); | |
6528 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6529 | zone_start_pfn, zone_end_pfn); | |
6530 | ||
6531 | if (zone_type == ZONE_MOVABLE && | |
6532 | memblock_is_mirror(r)) | |
6533 | nr_absent += end_pfn - start_pfn; | |
6534 | ||
6535 | if (zone_type == ZONE_NORMAL && | |
6536 | !memblock_is_mirror(r)) | |
6537 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6538 | } |
6539 | } | |
6540 | ||
6541 | return nr_absent; | |
c713216d | 6542 | } |
0e0b864e | 6543 | |
bbe5d993 | 6544 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd | 6545 | unsigned long node_start_pfn, |
854e8848 | 6546 | unsigned long node_end_pfn) |
c713216d | 6547 | { |
febd5949 | 6548 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6549 | enum zone_type i; |
6550 | ||
febd5949 GZ |
6551 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6552 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6553 | unsigned long zone_start_pfn, zone_end_pfn; |
3f08a302 | 6554 | unsigned long spanned, absent; |
febd5949 | 6555 | unsigned long size, real_size; |
c713216d | 6556 | |
854e8848 MR |
6557 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i, |
6558 | node_start_pfn, | |
6559 | node_end_pfn, | |
6560 | &zone_start_pfn, | |
6561 | &zone_end_pfn); | |
6562 | absent = zone_absent_pages_in_node(pgdat->node_id, i, | |
6563 | node_start_pfn, | |
6564 | node_end_pfn); | |
3f08a302 MR |
6565 | |
6566 | size = spanned; | |
6567 | real_size = size - absent; | |
6568 | ||
d91749c1 TI |
6569 | if (size) |
6570 | zone->zone_start_pfn = zone_start_pfn; | |
6571 | else | |
6572 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
6573 | zone->spanned_pages = size; |
6574 | zone->present_pages = real_size; | |
6575 | ||
6576 | totalpages += size; | |
6577 | realtotalpages += real_size; | |
6578 | } | |
6579 | ||
6580 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
6581 | pgdat->node_present_pages = realtotalpages; |
6582 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
6583 | realtotalpages); | |
6584 | } | |
6585 | ||
835c134e MG |
6586 | #ifndef CONFIG_SPARSEMEM |
6587 | /* | |
6588 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
6589 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
6590 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
6591 | * round what is now in bits to nearest long in bits, then return it in |
6592 | * bytes. | |
6593 | */ | |
7c45512d | 6594 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
6595 | { |
6596 | unsigned long usemapsize; | |
6597 | ||
7c45512d | 6598 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
6599 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
6600 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
6601 | usemapsize *= NR_PAGEBLOCK_BITS; |
6602 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
6603 | ||
6604 | return usemapsize / 8; | |
6605 | } | |
6606 | ||
7cc2a959 | 6607 | static void __ref setup_usemap(struct pglist_data *pgdat, |
7c45512d LT |
6608 | struct zone *zone, |
6609 | unsigned long zone_start_pfn, | |
6610 | unsigned long zonesize) | |
835c134e | 6611 | { |
7c45512d | 6612 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 6613 | zone->pageblock_flags = NULL; |
23a7052a | 6614 | if (usemapsize) { |
6782832e | 6615 | zone->pageblock_flags = |
26fb3dae MR |
6616 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
6617 | pgdat->node_id); | |
23a7052a MR |
6618 | if (!zone->pageblock_flags) |
6619 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
6620 | usemapsize, zone->name, pgdat->node_id); | |
6621 | } | |
835c134e MG |
6622 | } |
6623 | #else | |
7c45512d LT |
6624 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
6625 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
6626 | #endif /* CONFIG_SPARSEMEM */ |
6627 | ||
d9c23400 | 6628 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 6629 | |
d9c23400 | 6630 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 6631 | void __init set_pageblock_order(void) |
d9c23400 | 6632 | { |
955c1cd7 AM |
6633 | unsigned int order; |
6634 | ||
d9c23400 MG |
6635 | /* Check that pageblock_nr_pages has not already been setup */ |
6636 | if (pageblock_order) | |
6637 | return; | |
6638 | ||
955c1cd7 AM |
6639 | if (HPAGE_SHIFT > PAGE_SHIFT) |
6640 | order = HUGETLB_PAGE_ORDER; | |
6641 | else | |
6642 | order = MAX_ORDER - 1; | |
6643 | ||
d9c23400 MG |
6644 | /* |
6645 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
6646 | * This value may be variable depending on boot parameters on IA64 and |
6647 | * powerpc. | |
d9c23400 MG |
6648 | */ |
6649 | pageblock_order = order; | |
6650 | } | |
6651 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6652 | ||
ba72cb8c MG |
6653 | /* |
6654 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
6655 | * is unused as pageblock_order is set at compile-time. See |
6656 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
6657 | * the kernel config | |
ba72cb8c | 6658 | */ |
03e85f9d | 6659 | void __init set_pageblock_order(void) |
ba72cb8c | 6660 | { |
ba72cb8c | 6661 | } |
d9c23400 MG |
6662 | |
6663 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6664 | ||
03e85f9d | 6665 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 6666 | unsigned long present_pages) |
01cefaef JL |
6667 | { |
6668 | unsigned long pages = spanned_pages; | |
6669 | ||
6670 | /* | |
6671 | * Provide a more accurate estimation if there are holes within | |
6672 | * the zone and SPARSEMEM is in use. If there are holes within the | |
6673 | * zone, each populated memory region may cost us one or two extra | |
6674 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 6675 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
6676 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
6677 | */ | |
6678 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
6679 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
6680 | pages = present_pages; | |
6681 | ||
6682 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
6683 | } | |
6684 | ||
ace1db39 OS |
6685 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6686 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
6687 | { | |
364c1eeb YS |
6688 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
6689 | ||
6690 | spin_lock_init(&ds_queue->split_queue_lock); | |
6691 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
6692 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
6693 | } |
6694 | #else | |
6695 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
6696 | #endif | |
6697 | ||
6698 | #ifdef CONFIG_COMPACTION | |
6699 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
6700 | { | |
6701 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6702 | } | |
6703 | #else | |
6704 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
6705 | #endif | |
6706 | ||
03e85f9d | 6707 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 6708 | { |
208d54e5 | 6709 | pgdat_resize_init(pgdat); |
ace1db39 | 6710 | |
ace1db39 OS |
6711 | pgdat_init_split_queue(pgdat); |
6712 | pgdat_init_kcompactd(pgdat); | |
6713 | ||
1da177e4 | 6714 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6715 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 6716 | |
eefa864b | 6717 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6718 | spin_lock_init(&pgdat->lru_lock); |
867e5e1d | 6719 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
6720 | } |
6721 | ||
6722 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
6723 | unsigned long remaining_pages) | |
6724 | { | |
9705bea5 | 6725 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
6726 | zone_set_nid(zone, nid); |
6727 | zone->name = zone_names[idx]; | |
6728 | zone->zone_pgdat = NODE_DATA(nid); | |
6729 | spin_lock_init(&zone->lock); | |
6730 | zone_seqlock_init(zone); | |
6731 | zone_pcp_init(zone); | |
6732 | } | |
6733 | ||
6734 | /* | |
6735 | * Set up the zone data structures | |
6736 | * - init pgdat internals | |
6737 | * - init all zones belonging to this node | |
6738 | * | |
6739 | * NOTE: this function is only called during memory hotplug | |
6740 | */ | |
6741 | #ifdef CONFIG_MEMORY_HOTPLUG | |
6742 | void __ref free_area_init_core_hotplug(int nid) | |
6743 | { | |
6744 | enum zone_type z; | |
6745 | pg_data_t *pgdat = NODE_DATA(nid); | |
6746 | ||
6747 | pgdat_init_internals(pgdat); | |
6748 | for (z = 0; z < MAX_NR_ZONES; z++) | |
6749 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
6750 | } | |
6751 | #endif | |
6752 | ||
6753 | /* | |
6754 | * Set up the zone data structures: | |
6755 | * - mark all pages reserved | |
6756 | * - mark all memory queues empty | |
6757 | * - clear the memory bitmaps | |
6758 | * | |
6759 | * NOTE: pgdat should get zeroed by caller. | |
6760 | * NOTE: this function is only called during early init. | |
6761 | */ | |
6762 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
6763 | { | |
6764 | enum zone_type j; | |
6765 | int nid = pgdat->node_id; | |
5f63b720 | 6766 | |
03e85f9d | 6767 | pgdat_init_internals(pgdat); |
385386cf JW |
6768 | pgdat->per_cpu_nodestats = &boot_nodestats; |
6769 | ||
1da177e4 LT |
6770 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6771 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 6772 | unsigned long size, freesize, memmap_pages; |
d91749c1 | 6773 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6774 | |
febd5949 | 6775 | size = zone->spanned_pages; |
e6943859 | 6776 | freesize = zone->present_pages; |
1da177e4 | 6777 | |
0e0b864e | 6778 | /* |
9feedc9d | 6779 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6780 | * is used by this zone for memmap. This affects the watermark |
6781 | * and per-cpu initialisations | |
6782 | */ | |
e6943859 | 6783 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
6784 | if (!is_highmem_idx(j)) { |
6785 | if (freesize >= memmap_pages) { | |
6786 | freesize -= memmap_pages; | |
6787 | if (memmap_pages) | |
6788 | printk(KERN_DEBUG | |
6789 | " %s zone: %lu pages used for memmap\n", | |
6790 | zone_names[j], memmap_pages); | |
6791 | } else | |
1170532b | 6792 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6793 | zone_names[j], memmap_pages, freesize); |
6794 | } | |
0e0b864e | 6795 | |
6267276f | 6796 | /* Account for reserved pages */ |
9feedc9d JL |
6797 | if (j == 0 && freesize > dma_reserve) { |
6798 | freesize -= dma_reserve; | |
d903ef9f | 6799 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6800 | zone_names[0], dma_reserve); |
0e0b864e MG |
6801 | } |
6802 | ||
98d2b0eb | 6803 | if (!is_highmem_idx(j)) |
9feedc9d | 6804 | nr_kernel_pages += freesize; |
01cefaef JL |
6805 | /* Charge for highmem memmap if there are enough kernel pages */ |
6806 | else if (nr_kernel_pages > memmap_pages * 2) | |
6807 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6808 | nr_all_pages += freesize; |
1da177e4 | 6809 | |
9feedc9d JL |
6810 | /* |
6811 | * Set an approximate value for lowmem here, it will be adjusted | |
6812 | * when the bootmem allocator frees pages into the buddy system. | |
6813 | * And all highmem pages will be managed by the buddy system. | |
6814 | */ | |
03e85f9d | 6815 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 6816 | |
d883c6cf | 6817 | if (!size) |
1da177e4 LT |
6818 | continue; |
6819 | ||
955c1cd7 | 6820 | set_pageblock_order(); |
d883c6cf JK |
6821 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
6822 | init_currently_empty_zone(zone, zone_start_pfn, size); | |
76cdd58e | 6823 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6824 | } |
6825 | } | |
6826 | ||
0cd842f9 | 6827 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 6828 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6829 | { |
b0aeba74 | 6830 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6831 | unsigned long __maybe_unused offset = 0; |
6832 | ||
1da177e4 LT |
6833 | /* Skip empty nodes */ |
6834 | if (!pgdat->node_spanned_pages) | |
6835 | return; | |
6836 | ||
b0aeba74 TL |
6837 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6838 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6839 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6840 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6841 | unsigned long size, end; |
d41dee36 AW |
6842 | struct page *map; |
6843 | ||
e984bb43 BP |
6844 | /* |
6845 | * The zone's endpoints aren't required to be MAX_ORDER | |
6846 | * aligned but the node_mem_map endpoints must be in order | |
6847 | * for the buddy allocator to function correctly. | |
6848 | */ | |
108bcc96 | 6849 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6850 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6851 | size = (end - start) * sizeof(struct page); | |
26fb3dae MR |
6852 | map = memblock_alloc_node(size, SMP_CACHE_BYTES, |
6853 | pgdat->node_id); | |
23a7052a MR |
6854 | if (!map) |
6855 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
6856 | size, pgdat->node_id); | |
a1c34a3b | 6857 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6858 | } |
0cd842f9 OS |
6859 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
6860 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
6861 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 6862 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6863 | /* |
6864 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6865 | */ | |
c713216d | 6866 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6867 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d | 6868 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6869 | mem_map -= offset; |
c713216d | 6870 | } |
1da177e4 LT |
6871 | #endif |
6872 | } | |
0cd842f9 OS |
6873 | #else |
6874 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
6875 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 6876 | |
0188dc98 OS |
6877 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
6878 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
6879 | { | |
0188dc98 OS |
6880 | pgdat->first_deferred_pfn = ULONG_MAX; |
6881 | } | |
6882 | #else | |
6883 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
6884 | #endif | |
6885 | ||
854e8848 | 6886 | static void __init free_area_init_node(int nid) |
1da177e4 | 6887 | { |
9109fb7b | 6888 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6889 | unsigned long start_pfn = 0; |
6890 | unsigned long end_pfn = 0; | |
9109fb7b | 6891 | |
88fdf75d | 6892 | /* pg_data_t should be reset to zero when it's allocated */ |
97a225e6 | 6893 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx); |
88fdf75d | 6894 | |
854e8848 | 6895 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
88fdf75d | 6896 | |
1da177e4 | 6897 | pgdat->node_id = nid; |
854e8848 | 6898 | pgdat->node_start_pfn = start_pfn; |
75ef7184 | 6899 | pgdat->per_cpu_nodestats = NULL; |
854e8848 | 6900 | |
8d29e18a | 6901 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6902 | (u64)start_pfn << PAGE_SHIFT, |
6903 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
854e8848 | 6904 | calculate_node_totalpages(pgdat, start_pfn, end_pfn); |
1da177e4 LT |
6905 | |
6906 | alloc_node_mem_map(pgdat); | |
0188dc98 | 6907 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 6908 | |
7f3eb55b | 6909 | free_area_init_core(pgdat); |
1da177e4 LT |
6910 | } |
6911 | ||
bc9331a1 | 6912 | void __init free_area_init_memoryless_node(int nid) |
3f08a302 | 6913 | { |
854e8848 | 6914 | free_area_init_node(nid); |
3f08a302 MR |
6915 | } |
6916 | ||
aca52c39 | 6917 | #if !defined(CONFIG_FLAT_NODE_MEM_MAP) |
ec393a0f | 6918 | /* |
4b094b78 DH |
6919 | * Initialize all valid struct pages in the range [spfn, epfn) and mark them |
6920 | * PageReserved(). Return the number of struct pages that were initialized. | |
ec393a0f | 6921 | */ |
4b094b78 | 6922 | static u64 __init init_unavailable_range(unsigned long spfn, unsigned long epfn) |
ec393a0f PT |
6923 | { |
6924 | unsigned long pfn; | |
6925 | u64 pgcnt = 0; | |
6926 | ||
6927 | for (pfn = spfn; pfn < epfn; pfn++) { | |
6928 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { | |
6929 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6930 | + pageblock_nr_pages - 1; | |
6931 | continue; | |
6932 | } | |
4b094b78 DH |
6933 | /* |
6934 | * Use a fake node/zone (0) for now. Some of these pages | |
6935 | * (in memblock.reserved but not in memblock.memory) will | |
6936 | * get re-initialized via reserve_bootmem_region() later. | |
6937 | */ | |
6938 | __init_single_page(pfn_to_page(pfn), pfn, 0, 0); | |
6939 | __SetPageReserved(pfn_to_page(pfn)); | |
ec393a0f PT |
6940 | pgcnt++; |
6941 | } | |
6942 | ||
6943 | return pgcnt; | |
6944 | } | |
6945 | ||
a4a3ede2 PT |
6946 | /* |
6947 | * Only struct pages that are backed by physical memory are zeroed and | |
6948 | * initialized by going through __init_single_page(). But, there are some | |
6949 | * struct pages which are reserved in memblock allocator and their fields | |
6950 | * may be accessed (for example page_to_pfn() on some configuration accesses | |
4b094b78 | 6951 | * flags). We must explicitly initialize those struct pages. |
907ec5fc NH |
6952 | * |
6953 | * This function also addresses a similar issue where struct pages are left | |
6954 | * uninitialized because the physical address range is not covered by | |
6955 | * memblock.memory or memblock.reserved. That could happen when memblock | |
e822969c DH |
6956 | * layout is manually configured via memmap=, or when the highest physical |
6957 | * address (max_pfn) does not end on a section boundary. | |
a4a3ede2 | 6958 | */ |
4b094b78 | 6959 | static void __init init_unavailable_mem(void) |
a4a3ede2 PT |
6960 | { |
6961 | phys_addr_t start, end; | |
a4a3ede2 | 6962 | u64 i, pgcnt; |
907ec5fc | 6963 | phys_addr_t next = 0; |
a4a3ede2 PT |
6964 | |
6965 | /* | |
907ec5fc | 6966 | * Loop through unavailable ranges not covered by memblock.memory. |
a4a3ede2 PT |
6967 | */ |
6968 | pgcnt = 0; | |
907ec5fc NH |
6969 | for_each_mem_range(i, &memblock.memory, NULL, |
6970 | NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) { | |
ec393a0f | 6971 | if (next < start) |
4b094b78 DH |
6972 | pgcnt += init_unavailable_range(PFN_DOWN(next), |
6973 | PFN_UP(start)); | |
907ec5fc NH |
6974 | next = end; |
6975 | } | |
e822969c DH |
6976 | |
6977 | /* | |
6978 | * Early sections always have a fully populated memmap for the whole | |
6979 | * section - see pfn_valid(). If the last section has holes at the | |
6980 | * end and that section is marked "online", the memmap will be | |
6981 | * considered initialized. Make sure that memmap has a well defined | |
6982 | * state. | |
6983 | */ | |
4b094b78 DH |
6984 | pgcnt += init_unavailable_range(PFN_DOWN(next), |
6985 | round_up(max_pfn, PAGES_PER_SECTION)); | |
907ec5fc | 6986 | |
a4a3ede2 PT |
6987 | /* |
6988 | * Struct pages that do not have backing memory. This could be because | |
6989 | * firmware is using some of this memory, or for some other reasons. | |
a4a3ede2 PT |
6990 | */ |
6991 | if (pgcnt) | |
907ec5fc | 6992 | pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt); |
a4a3ede2 | 6993 | } |
4b094b78 DH |
6994 | #else |
6995 | static inline void __init init_unavailable_mem(void) | |
6996 | { | |
6997 | } | |
aca52c39 | 6998 | #endif /* !CONFIG_FLAT_NODE_MEM_MAP */ |
a4a3ede2 | 6999 | |
418508c1 MS |
7000 | #if MAX_NUMNODES > 1 |
7001 | /* | |
7002 | * Figure out the number of possible node ids. | |
7003 | */ | |
f9872caf | 7004 | void __init setup_nr_node_ids(void) |
418508c1 | 7005 | { |
904a9553 | 7006 | unsigned int highest; |
418508c1 | 7007 | |
904a9553 | 7008 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7009 | nr_node_ids = highest + 1; |
7010 | } | |
418508c1 MS |
7011 | #endif |
7012 | ||
1e01979c TH |
7013 | /** |
7014 | * node_map_pfn_alignment - determine the maximum internode alignment | |
7015 | * | |
7016 | * This function should be called after node map is populated and sorted. | |
7017 | * It calculates the maximum power of two alignment which can distinguish | |
7018 | * all the nodes. | |
7019 | * | |
7020 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
7021 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
7022 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
7023 | * shifted, 1GiB is enough and this function will indicate so. | |
7024 | * | |
7025 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
7026 | * model has fine enough granularity to avoid incorrect mapping for the | |
7027 | * populated node map. | |
7028 | * | |
a862f68a | 7029 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
7030 | * requirement (single node). |
7031 | */ | |
7032 | unsigned long __init node_map_pfn_alignment(void) | |
7033 | { | |
7034 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 7035 | unsigned long start, end, mask; |
98fa15f3 | 7036 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 7037 | int i, nid; |
1e01979c | 7038 | |
c13291a5 | 7039 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
7040 | if (!start || last_nid < 0 || last_nid == nid) { |
7041 | last_nid = nid; | |
7042 | last_end = end; | |
7043 | continue; | |
7044 | } | |
7045 | ||
7046 | /* | |
7047 | * Start with a mask granular enough to pin-point to the | |
7048 | * start pfn and tick off bits one-by-one until it becomes | |
7049 | * too coarse to separate the current node from the last. | |
7050 | */ | |
7051 | mask = ~((1 << __ffs(start)) - 1); | |
7052 | while (mask && last_end <= (start & (mask << 1))) | |
7053 | mask <<= 1; | |
7054 | ||
7055 | /* accumulate all internode masks */ | |
7056 | accl_mask |= mask; | |
7057 | } | |
7058 | ||
7059 | /* convert mask to number of pages */ | |
7060 | return ~accl_mask + 1; | |
7061 | } | |
7062 | ||
c713216d MG |
7063 | /** |
7064 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
7065 | * | |
a862f68a | 7066 | * Return: the minimum PFN based on information provided via |
7d018176 | 7067 | * memblock_set_node(). |
c713216d MG |
7068 | */ |
7069 | unsigned long __init find_min_pfn_with_active_regions(void) | |
7070 | { | |
8a1b25fe | 7071 | return PHYS_PFN(memblock_start_of_DRAM()); |
c713216d MG |
7072 | } |
7073 | ||
37b07e41 LS |
7074 | /* |
7075 | * early_calculate_totalpages() | |
7076 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 7077 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 7078 | */ |
484f51f8 | 7079 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 7080 | { |
7e63efef | 7081 | unsigned long totalpages = 0; |
c13291a5 TH |
7082 | unsigned long start_pfn, end_pfn; |
7083 | int i, nid; | |
7084 | ||
7085 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
7086 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 7087 | |
37b07e41 LS |
7088 | totalpages += pages; |
7089 | if (pages) | |
4b0ef1fe | 7090 | node_set_state(nid, N_MEMORY); |
37b07e41 | 7091 | } |
b8af2941 | 7092 | return totalpages; |
7e63efef MG |
7093 | } |
7094 | ||
2a1e274a MG |
7095 | /* |
7096 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
7097 | * is spread evenly between nodes as long as the nodes have enough | |
7098 | * memory. When they don't, some nodes will have more kernelcore than | |
7099 | * others | |
7100 | */ | |
b224ef85 | 7101 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
7102 | { |
7103 | int i, nid; | |
7104 | unsigned long usable_startpfn; | |
7105 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 7106 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 7107 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 7108 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 7109 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 7110 | struct memblock_region *r; |
b2f3eebe TC |
7111 | |
7112 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
7113 | find_usable_zone_for_movable(); | |
7114 | ||
7115 | /* | |
7116 | * If movable_node is specified, ignore kernelcore and movablecore | |
7117 | * options. | |
7118 | */ | |
7119 | if (movable_node_is_enabled()) { | |
136199f0 EM |
7120 | for_each_memblock(memory, r) { |
7121 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
7122 | continue; |
7123 | ||
d622abf7 | 7124 | nid = memblock_get_region_node(r); |
b2f3eebe | 7125 | |
136199f0 | 7126 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
7127 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
7128 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7129 | usable_startpfn; | |
7130 | } | |
7131 | ||
7132 | goto out2; | |
7133 | } | |
2a1e274a | 7134 | |
342332e6 TI |
7135 | /* |
7136 | * If kernelcore=mirror is specified, ignore movablecore option | |
7137 | */ | |
7138 | if (mirrored_kernelcore) { | |
7139 | bool mem_below_4gb_not_mirrored = false; | |
7140 | ||
7141 | for_each_memblock(memory, r) { | |
7142 | if (memblock_is_mirror(r)) | |
7143 | continue; | |
7144 | ||
d622abf7 | 7145 | nid = memblock_get_region_node(r); |
342332e6 TI |
7146 | |
7147 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
7148 | ||
7149 | if (usable_startpfn < 0x100000) { | |
7150 | mem_below_4gb_not_mirrored = true; | |
7151 | continue; | |
7152 | } | |
7153 | ||
7154 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
7155 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7156 | usable_startpfn; | |
7157 | } | |
7158 | ||
7159 | if (mem_below_4gb_not_mirrored) | |
633bf2fe | 7160 | pr_warn("This configuration results in unmirrored kernel memory.\n"); |
342332e6 TI |
7161 | |
7162 | goto out2; | |
7163 | } | |
7164 | ||
7e63efef | 7165 | /* |
a5c6d650 DR |
7166 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
7167 | * amount of necessary memory. | |
7168 | */ | |
7169 | if (required_kernelcore_percent) | |
7170 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
7171 | 10000UL; | |
7172 | if (required_movablecore_percent) | |
7173 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
7174 | 10000UL; | |
7175 | ||
7176 | /* | |
7177 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
7178 | * kernelcore that corresponds so that memory usable for |
7179 | * any allocation type is evenly spread. If both kernelcore | |
7180 | * and movablecore are specified, then the value of kernelcore | |
7181 | * will be used for required_kernelcore if it's greater than | |
7182 | * what movablecore would have allowed. | |
7183 | */ | |
7184 | if (required_movablecore) { | |
7e63efef MG |
7185 | unsigned long corepages; |
7186 | ||
7187 | /* | |
7188 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
7189 | * was requested by the user | |
7190 | */ | |
7191 | required_movablecore = | |
7192 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 7193 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
7194 | corepages = totalpages - required_movablecore; |
7195 | ||
7196 | required_kernelcore = max(required_kernelcore, corepages); | |
7197 | } | |
7198 | ||
bde304bd XQ |
7199 | /* |
7200 | * If kernelcore was not specified or kernelcore size is larger | |
7201 | * than totalpages, there is no ZONE_MOVABLE. | |
7202 | */ | |
7203 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 7204 | goto out; |
2a1e274a MG |
7205 | |
7206 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
7207 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
7208 | ||
7209 | restart: | |
7210 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
7211 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 7212 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
7213 | unsigned long start_pfn, end_pfn; |
7214 | ||
2a1e274a MG |
7215 | /* |
7216 | * Recalculate kernelcore_node if the division per node | |
7217 | * now exceeds what is necessary to satisfy the requested | |
7218 | * amount of memory for the kernel | |
7219 | */ | |
7220 | if (required_kernelcore < kernelcore_node) | |
7221 | kernelcore_node = required_kernelcore / usable_nodes; | |
7222 | ||
7223 | /* | |
7224 | * As the map is walked, we track how much memory is usable | |
7225 | * by the kernel using kernelcore_remaining. When it is | |
7226 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
7227 | */ | |
7228 | kernelcore_remaining = kernelcore_node; | |
7229 | ||
7230 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 7231 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
7232 | unsigned long size_pages; |
7233 | ||
c13291a5 | 7234 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
7235 | if (start_pfn >= end_pfn) |
7236 | continue; | |
7237 | ||
7238 | /* Account for what is only usable for kernelcore */ | |
7239 | if (start_pfn < usable_startpfn) { | |
7240 | unsigned long kernel_pages; | |
7241 | kernel_pages = min(end_pfn, usable_startpfn) | |
7242 | - start_pfn; | |
7243 | ||
7244 | kernelcore_remaining -= min(kernel_pages, | |
7245 | kernelcore_remaining); | |
7246 | required_kernelcore -= min(kernel_pages, | |
7247 | required_kernelcore); | |
7248 | ||
7249 | /* Continue if range is now fully accounted */ | |
7250 | if (end_pfn <= usable_startpfn) { | |
7251 | ||
7252 | /* | |
7253 | * Push zone_movable_pfn to the end so | |
7254 | * that if we have to rebalance | |
7255 | * kernelcore across nodes, we will | |
7256 | * not double account here | |
7257 | */ | |
7258 | zone_movable_pfn[nid] = end_pfn; | |
7259 | continue; | |
7260 | } | |
7261 | start_pfn = usable_startpfn; | |
7262 | } | |
7263 | ||
7264 | /* | |
7265 | * The usable PFN range for ZONE_MOVABLE is from | |
7266 | * start_pfn->end_pfn. Calculate size_pages as the | |
7267 | * number of pages used as kernelcore | |
7268 | */ | |
7269 | size_pages = end_pfn - start_pfn; | |
7270 | if (size_pages > kernelcore_remaining) | |
7271 | size_pages = kernelcore_remaining; | |
7272 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
7273 | ||
7274 | /* | |
7275 | * Some kernelcore has been met, update counts and | |
7276 | * break if the kernelcore for this node has been | |
b8af2941 | 7277 | * satisfied |
2a1e274a MG |
7278 | */ |
7279 | required_kernelcore -= min(required_kernelcore, | |
7280 | size_pages); | |
7281 | kernelcore_remaining -= size_pages; | |
7282 | if (!kernelcore_remaining) | |
7283 | break; | |
7284 | } | |
7285 | } | |
7286 | ||
7287 | /* | |
7288 | * If there is still required_kernelcore, we do another pass with one | |
7289 | * less node in the count. This will push zone_movable_pfn[nid] further | |
7290 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 7291 | * satisfied |
2a1e274a MG |
7292 | */ |
7293 | usable_nodes--; | |
7294 | if (usable_nodes && required_kernelcore > usable_nodes) | |
7295 | goto restart; | |
7296 | ||
b2f3eebe | 7297 | out2: |
2a1e274a MG |
7298 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
7299 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
7300 | zone_movable_pfn[nid] = | |
7301 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 7302 | |
20e6926d | 7303 | out: |
66918dcd | 7304 | /* restore the node_state */ |
4b0ef1fe | 7305 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
7306 | } |
7307 | ||
4b0ef1fe LJ |
7308 | /* Any regular or high memory on that node ? */ |
7309 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 7310 | { |
37b07e41 LS |
7311 | enum zone_type zone_type; |
7312 | ||
4b0ef1fe | 7313 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 7314 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 7315 | if (populated_zone(zone)) { |
7b0e0c0e OS |
7316 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
7317 | node_set_state(nid, N_HIGH_MEMORY); | |
7318 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 7319 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
7320 | break; |
7321 | } | |
37b07e41 | 7322 | } |
37b07e41 LS |
7323 | } |
7324 | ||
51930df5 MR |
7325 | /* |
7326 | * Some architecturs, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For | |
7327 | * such cases we allow max_zone_pfn sorted in the descending order | |
7328 | */ | |
7329 | bool __weak arch_has_descending_max_zone_pfns(void) | |
7330 | { | |
7331 | return false; | |
7332 | } | |
7333 | ||
c713216d | 7334 | /** |
9691a071 | 7335 | * free_area_init - Initialise all pg_data_t and zone data |
88ca3b94 | 7336 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
7337 | * |
7338 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 7339 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
7340 | * zone in each node and their holes is calculated. If the maximum PFN |
7341 | * between two adjacent zones match, it is assumed that the zone is empty. | |
7342 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
7343 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
7344 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
7345 | * at arch_max_dma_pfn. | |
7346 | */ | |
9691a071 | 7347 | void __init free_area_init(unsigned long *max_zone_pfn) |
c713216d | 7348 | { |
c13291a5 | 7349 | unsigned long start_pfn, end_pfn; |
51930df5 MR |
7350 | int i, nid, zone; |
7351 | bool descending; | |
a6af2bc3 | 7352 | |
c713216d MG |
7353 | /* Record where the zone boundaries are */ |
7354 | memset(arch_zone_lowest_possible_pfn, 0, | |
7355 | sizeof(arch_zone_lowest_possible_pfn)); | |
7356 | memset(arch_zone_highest_possible_pfn, 0, | |
7357 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
7358 | |
7359 | start_pfn = find_min_pfn_with_active_regions(); | |
51930df5 | 7360 | descending = arch_has_descending_max_zone_pfns(); |
90cae1fe OH |
7361 | |
7362 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
51930df5 MR |
7363 | if (descending) |
7364 | zone = MAX_NR_ZONES - i - 1; | |
7365 | else | |
7366 | zone = i; | |
7367 | ||
7368 | if (zone == ZONE_MOVABLE) | |
2a1e274a | 7369 | continue; |
90cae1fe | 7370 | |
51930df5 MR |
7371 | end_pfn = max(max_zone_pfn[zone], start_pfn); |
7372 | arch_zone_lowest_possible_pfn[zone] = start_pfn; | |
7373 | arch_zone_highest_possible_pfn[zone] = end_pfn; | |
90cae1fe OH |
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); | |
854e8848 | 7426 | free_area_init_node(nid); |
37b07e41 LS |
7427 | |
7428 | /* Any memory on that node */ | |
7429 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
7430 | node_set_state(nid, N_MEMORY); |
7431 | check_for_memory(pgdat, nid); | |
c713216d MG |
7432 | } |
7433 | } | |
2a1e274a | 7434 | |
a5c6d650 DR |
7435 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
7436 | unsigned long *percent) | |
2a1e274a MG |
7437 | { |
7438 | unsigned long long coremem; | |
a5c6d650 DR |
7439 | char *endptr; |
7440 | ||
2a1e274a MG |
7441 | if (!p) |
7442 | return -EINVAL; | |
7443 | ||
a5c6d650 DR |
7444 | /* Value may be a percentage of total memory, otherwise bytes */ |
7445 | coremem = simple_strtoull(p, &endptr, 0); | |
7446 | if (*endptr == '%') { | |
7447 | /* Paranoid check for percent values greater than 100 */ | |
7448 | WARN_ON(coremem > 100); | |
2a1e274a | 7449 | |
a5c6d650 DR |
7450 | *percent = coremem; |
7451 | } else { | |
7452 | coremem = memparse(p, &p); | |
7453 | /* Paranoid check that UL is enough for the coremem value */ | |
7454 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7455 | |
a5c6d650 DR |
7456 | *core = coremem >> PAGE_SHIFT; |
7457 | *percent = 0UL; | |
7458 | } | |
2a1e274a MG |
7459 | return 0; |
7460 | } | |
ed7ed365 | 7461 | |
7e63efef MG |
7462 | /* |
7463 | * kernelcore=size sets the amount of memory for use for allocations that | |
7464 | * cannot be reclaimed or migrated. | |
7465 | */ | |
7466 | static int __init cmdline_parse_kernelcore(char *p) | |
7467 | { | |
342332e6 TI |
7468 | /* parse kernelcore=mirror */ |
7469 | if (parse_option_str(p, "mirror")) { | |
7470 | mirrored_kernelcore = true; | |
7471 | return 0; | |
7472 | } | |
7473 | ||
a5c6d650 DR |
7474 | return cmdline_parse_core(p, &required_kernelcore, |
7475 | &required_kernelcore_percent); | |
7e63efef MG |
7476 | } |
7477 | ||
7478 | /* | |
7479 | * movablecore=size sets the amount of memory for use for allocations that | |
7480 | * can be reclaimed or migrated. | |
7481 | */ | |
7482 | static int __init cmdline_parse_movablecore(char *p) | |
7483 | { | |
a5c6d650 DR |
7484 | return cmdline_parse_core(p, &required_movablecore, |
7485 | &required_movablecore_percent); | |
7e63efef MG |
7486 | } |
7487 | ||
ed7ed365 | 7488 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7489 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7490 | |
c3d5f5f0 JL |
7491 | void adjust_managed_page_count(struct page *page, long count) |
7492 | { | |
9705bea5 | 7493 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 7494 | totalram_pages_add(count); |
3dcc0571 JL |
7495 | #ifdef CONFIG_HIGHMEM |
7496 | if (PageHighMem(page)) | |
ca79b0c2 | 7497 | totalhigh_pages_add(count); |
3dcc0571 | 7498 | #endif |
c3d5f5f0 | 7499 | } |
3dcc0571 | 7500 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7501 | |
e5cb113f | 7502 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 7503 | { |
11199692 JL |
7504 | void *pos; |
7505 | unsigned long pages = 0; | |
69afade7 | 7506 | |
11199692 JL |
7507 | start = (void *)PAGE_ALIGN((unsigned long)start); |
7508 | end = (void *)((unsigned long)end & PAGE_MASK); | |
7509 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
7510 | struct page *page = virt_to_page(pos); |
7511 | void *direct_map_addr; | |
7512 | ||
7513 | /* | |
7514 | * 'direct_map_addr' might be different from 'pos' | |
7515 | * because some architectures' virt_to_page() | |
7516 | * work with aliases. Getting the direct map | |
7517 | * address ensures that we get a _writeable_ | |
7518 | * alias for the memset(). | |
7519 | */ | |
7520 | direct_map_addr = page_address(page); | |
dbe67df4 | 7521 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
7522 | memset(direct_map_addr, poison, PAGE_SIZE); |
7523 | ||
7524 | free_reserved_page(page); | |
69afade7 JL |
7525 | } |
7526 | ||
7527 | if (pages && s) | |
adb1fe9a JP |
7528 | pr_info("Freeing %s memory: %ldK\n", |
7529 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
7530 | |
7531 | return pages; | |
7532 | } | |
7533 | ||
cfa11e08 JL |
7534 | #ifdef CONFIG_HIGHMEM |
7535 | void free_highmem_page(struct page *page) | |
7536 | { | |
7537 | __free_reserved_page(page); | |
ca79b0c2 | 7538 | totalram_pages_inc(); |
9705bea5 | 7539 | atomic_long_inc(&page_zone(page)->managed_pages); |
ca79b0c2 | 7540 | totalhigh_pages_inc(); |
cfa11e08 JL |
7541 | } |
7542 | #endif | |
7543 | ||
7ee3d4e8 JL |
7544 | |
7545 | void __init mem_init_print_info(const char *str) | |
7546 | { | |
7547 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7548 | unsigned long init_code_size, init_data_size; | |
7549 | ||
7550 | physpages = get_num_physpages(); | |
7551 | codesize = _etext - _stext; | |
7552 | datasize = _edata - _sdata; | |
7553 | rosize = __end_rodata - __start_rodata; | |
7554 | bss_size = __bss_stop - __bss_start; | |
7555 | init_data_size = __init_end - __init_begin; | |
7556 | init_code_size = _einittext - _sinittext; | |
7557 | ||
7558 | /* | |
7559 | * Detect special cases and adjust section sizes accordingly: | |
7560 | * 1) .init.* may be embedded into .data sections | |
7561 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7562 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7563 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7564 | */ | |
7565 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7566 | do { \ |
7567 | if (start <= pos && pos < end && size > adj) \ | |
7568 | size -= adj; \ | |
7569 | } while (0) | |
7ee3d4e8 JL |
7570 | |
7571 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7572 | _sinittext, init_code_size); | |
7573 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7574 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7575 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7576 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7577 | ||
7578 | #undef adj_init_size | |
7579 | ||
756a025f | 7580 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7581 | #ifdef CONFIG_HIGHMEM |
756a025f | 7582 | ", %luK highmem" |
7ee3d4e8 | 7583 | #endif |
756a025f JP |
7584 | "%s%s)\n", |
7585 | nr_free_pages() << (PAGE_SHIFT - 10), | |
7586 | physpages << (PAGE_SHIFT - 10), | |
7587 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7588 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
ca79b0c2 | 7589 | (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10), |
756a025f | 7590 | totalcma_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7591 | #ifdef CONFIG_HIGHMEM |
ca79b0c2 | 7592 | totalhigh_pages() << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7593 | #endif |
756a025f | 7594 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
7595 | } |
7596 | ||
0e0b864e | 7597 | /** |
88ca3b94 RD |
7598 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7599 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7600 | * |
013110a7 | 7601 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7602 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7603 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7604 | * function may optionally be used to account for unfreeable pages in the |
7605 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7606 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7607 | */ |
7608 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7609 | { | |
7610 | dma_reserve = new_dma_reserve; | |
7611 | } | |
7612 | ||
005fd4bb | 7613 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7614 | { |
1da177e4 | 7615 | |
005fd4bb SAS |
7616 | lru_add_drain_cpu(cpu); |
7617 | drain_pages(cpu); | |
9f8f2172 | 7618 | |
005fd4bb SAS |
7619 | /* |
7620 | * Spill the event counters of the dead processor | |
7621 | * into the current processors event counters. | |
7622 | * This artificially elevates the count of the current | |
7623 | * processor. | |
7624 | */ | |
7625 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7626 | |
005fd4bb SAS |
7627 | /* |
7628 | * Zero the differential counters of the dead processor | |
7629 | * so that the vm statistics are consistent. | |
7630 | * | |
7631 | * This is only okay since the processor is dead and cannot | |
7632 | * race with what we are doing. | |
7633 | */ | |
7634 | cpu_vm_stats_fold(cpu); | |
7635 | return 0; | |
1da177e4 | 7636 | } |
1da177e4 | 7637 | |
e03a5125 NP |
7638 | #ifdef CONFIG_NUMA |
7639 | int hashdist = HASHDIST_DEFAULT; | |
7640 | ||
7641 | static int __init set_hashdist(char *str) | |
7642 | { | |
7643 | if (!str) | |
7644 | return 0; | |
7645 | hashdist = simple_strtoul(str, &str, 0); | |
7646 | return 1; | |
7647 | } | |
7648 | __setup("hashdist=", set_hashdist); | |
7649 | #endif | |
7650 | ||
1da177e4 LT |
7651 | void __init page_alloc_init(void) |
7652 | { | |
005fd4bb SAS |
7653 | int ret; |
7654 | ||
e03a5125 NP |
7655 | #ifdef CONFIG_NUMA |
7656 | if (num_node_state(N_MEMORY) == 1) | |
7657 | hashdist = 0; | |
7658 | #endif | |
7659 | ||
005fd4bb SAS |
7660 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, |
7661 | "mm/page_alloc:dead", NULL, | |
7662 | page_alloc_cpu_dead); | |
7663 | WARN_ON(ret < 0); | |
1da177e4 LT |
7664 | } |
7665 | ||
cb45b0e9 | 7666 | /* |
34b10060 | 7667 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
7668 | * or min_free_kbytes changes. |
7669 | */ | |
7670 | static void calculate_totalreserve_pages(void) | |
7671 | { | |
7672 | struct pglist_data *pgdat; | |
7673 | unsigned long reserve_pages = 0; | |
2f6726e5 | 7674 | enum zone_type i, j; |
cb45b0e9 HA |
7675 | |
7676 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
7677 | |
7678 | pgdat->totalreserve_pages = 0; | |
7679 | ||
cb45b0e9 HA |
7680 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7681 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 7682 | long max = 0; |
9705bea5 | 7683 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
7684 | |
7685 | /* Find valid and maximum lowmem_reserve in the zone */ | |
7686 | for (j = i; j < MAX_NR_ZONES; j++) { | |
7687 | if (zone->lowmem_reserve[j] > max) | |
7688 | max = zone->lowmem_reserve[j]; | |
7689 | } | |
7690 | ||
41858966 MG |
7691 | /* we treat the high watermark as reserved pages. */ |
7692 | max += high_wmark_pages(zone); | |
cb45b0e9 | 7693 | |
3d6357de AK |
7694 | if (max > managed_pages) |
7695 | max = managed_pages; | |
a8d01437 | 7696 | |
281e3726 | 7697 | pgdat->totalreserve_pages += max; |
a8d01437 | 7698 | |
cb45b0e9 HA |
7699 | reserve_pages += max; |
7700 | } | |
7701 | } | |
7702 | totalreserve_pages = reserve_pages; | |
7703 | } | |
7704 | ||
1da177e4 LT |
7705 | /* |
7706 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 7707 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
7708 | * has a correct pages reserved value, so an adequate number of |
7709 | * pages are left in the zone after a successful __alloc_pages(). | |
7710 | */ | |
7711 | static void setup_per_zone_lowmem_reserve(void) | |
7712 | { | |
7713 | struct pglist_data *pgdat; | |
2f6726e5 | 7714 | enum zone_type j, idx; |
1da177e4 | 7715 | |
ec936fc5 | 7716 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
7717 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7718 | struct zone *zone = pgdat->node_zones + j; | |
9705bea5 | 7719 | unsigned long managed_pages = zone_managed_pages(zone); |
1da177e4 LT |
7720 | |
7721 | zone->lowmem_reserve[j] = 0; | |
7722 | ||
2f6726e5 CL |
7723 | idx = j; |
7724 | while (idx) { | |
1da177e4 LT |
7725 | struct zone *lower_zone; |
7726 | ||
2f6726e5 | 7727 | idx--; |
1da177e4 | 7728 | lower_zone = pgdat->node_zones + idx; |
d3cda233 | 7729 | |
f6366156 BH |
7730 | if (!sysctl_lowmem_reserve_ratio[idx] || |
7731 | !zone_managed_pages(lower_zone)) { | |
d3cda233 | 7732 | lower_zone->lowmem_reserve[j] = 0; |
f6366156 | 7733 | continue; |
d3cda233 JK |
7734 | } else { |
7735 | lower_zone->lowmem_reserve[j] = | |
7736 | managed_pages / sysctl_lowmem_reserve_ratio[idx]; | |
7737 | } | |
9705bea5 | 7738 | managed_pages += zone_managed_pages(lower_zone); |
1da177e4 LT |
7739 | } |
7740 | } | |
7741 | } | |
cb45b0e9 HA |
7742 | |
7743 | /* update totalreserve_pages */ | |
7744 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7745 | } |
7746 | ||
cfd3da1e | 7747 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
7748 | { |
7749 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
7750 | unsigned long lowmem_pages = 0; | |
7751 | struct zone *zone; | |
7752 | unsigned long flags; | |
7753 | ||
7754 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
7755 | for_each_zone(zone) { | |
7756 | if (!is_highmem(zone)) | |
9705bea5 | 7757 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
7758 | } |
7759 | ||
7760 | for_each_zone(zone) { | |
ac924c60 AM |
7761 | u64 tmp; |
7762 | ||
1125b4e3 | 7763 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 7764 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 7765 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
7766 | if (is_highmem(zone)) { |
7767 | /* | |
669ed175 NP |
7768 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
7769 | * need highmem pages, so cap pages_min to a small | |
7770 | * value here. | |
7771 | * | |
41858966 | 7772 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 7773 | * deltas control async page reclaim, and so should |
669ed175 | 7774 | * not be capped for highmem. |
1da177e4 | 7775 | */ |
90ae8d67 | 7776 | unsigned long min_pages; |
1da177e4 | 7777 | |
9705bea5 | 7778 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 7779 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 7780 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 7781 | } else { |
669ed175 NP |
7782 | /* |
7783 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
7784 | * proportionate to the zone's size. |
7785 | */ | |
a9214443 | 7786 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
7787 | } |
7788 | ||
795ae7a0 JW |
7789 | /* |
7790 | * Set the kswapd watermarks distance according to the | |
7791 | * scale factor in proportion to available memory, but | |
7792 | * ensure a minimum size on small systems. | |
7793 | */ | |
7794 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 7795 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
7796 | watermark_scale_factor, 10000)); |
7797 | ||
aa092591 | 7798 | zone->watermark_boost = 0; |
a9214443 MG |
7799 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
7800 | zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 7801 | |
1125b4e3 | 7802 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 7803 | } |
cb45b0e9 HA |
7804 | |
7805 | /* update totalreserve_pages */ | |
7806 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7807 | } |
7808 | ||
cfd3da1e MG |
7809 | /** |
7810 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
7811 | * or when memory is hot-{added|removed} | |
7812 | * | |
7813 | * Ensures that the watermark[min,low,high] values for each zone are set | |
7814 | * correctly with respect to min_free_kbytes. | |
7815 | */ | |
7816 | void setup_per_zone_wmarks(void) | |
7817 | { | |
b93e0f32 MH |
7818 | static DEFINE_SPINLOCK(lock); |
7819 | ||
7820 | spin_lock(&lock); | |
cfd3da1e | 7821 | __setup_per_zone_wmarks(); |
b93e0f32 | 7822 | spin_unlock(&lock); |
cfd3da1e MG |
7823 | } |
7824 | ||
1da177e4 LT |
7825 | /* |
7826 | * Initialise min_free_kbytes. | |
7827 | * | |
7828 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 7829 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
7830 | * bandwidth does not increase linearly with machine size. We use |
7831 | * | |
b8af2941 | 7832 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
7833 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
7834 | * | |
7835 | * which yields | |
7836 | * | |
7837 | * 16MB: 512k | |
7838 | * 32MB: 724k | |
7839 | * 64MB: 1024k | |
7840 | * 128MB: 1448k | |
7841 | * 256MB: 2048k | |
7842 | * 512MB: 2896k | |
7843 | * 1024MB: 4096k | |
7844 | * 2048MB: 5792k | |
7845 | * 4096MB: 8192k | |
7846 | * 8192MB: 11584k | |
7847 | * 16384MB: 16384k | |
7848 | */ | |
1b79acc9 | 7849 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
7850 | { |
7851 | unsigned long lowmem_kbytes; | |
5f12733e | 7852 | int new_min_free_kbytes; |
1da177e4 LT |
7853 | |
7854 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
7855 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
7856 | ||
7857 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
7858 | min_free_kbytes = new_min_free_kbytes; | |
7859 | if (min_free_kbytes < 128) | |
7860 | min_free_kbytes = 128; | |
ee8eb9a5 JS |
7861 | if (min_free_kbytes > 262144) |
7862 | min_free_kbytes = 262144; | |
5f12733e MH |
7863 | } else { |
7864 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
7865 | new_min_free_kbytes, user_min_free_kbytes); | |
7866 | } | |
bc75d33f | 7867 | setup_per_zone_wmarks(); |
a6cccdc3 | 7868 | refresh_zone_stat_thresholds(); |
1da177e4 | 7869 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
7870 | |
7871 | #ifdef CONFIG_NUMA | |
7872 | setup_min_unmapped_ratio(); | |
7873 | setup_min_slab_ratio(); | |
7874 | #endif | |
7875 | ||
1da177e4 LT |
7876 | return 0; |
7877 | } | |
bc22af74 | 7878 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
7879 | |
7880 | /* | |
b8af2941 | 7881 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
7882 | * that we can call two helper functions whenever min_free_kbytes |
7883 | * changes. | |
7884 | */ | |
cccad5b9 | 7885 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 7886 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7887 | { |
da8c757b HP |
7888 | int rc; |
7889 | ||
7890 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7891 | if (rc) | |
7892 | return rc; | |
7893 | ||
5f12733e MH |
7894 | if (write) { |
7895 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7896 | setup_per_zone_wmarks(); |
5f12733e | 7897 | } |
1da177e4 LT |
7898 | return 0; |
7899 | } | |
7900 | ||
795ae7a0 | 7901 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 7902 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
7903 | { |
7904 | int rc; | |
7905 | ||
7906 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7907 | if (rc) | |
7908 | return rc; | |
7909 | ||
7910 | if (write) | |
7911 | setup_per_zone_wmarks(); | |
7912 | ||
7913 | return 0; | |
7914 | } | |
7915 | ||
9614634f | 7916 | #ifdef CONFIG_NUMA |
6423aa81 | 7917 | static void setup_min_unmapped_ratio(void) |
9614634f | 7918 | { |
6423aa81 | 7919 | pg_data_t *pgdat; |
9614634f | 7920 | struct zone *zone; |
9614634f | 7921 | |
a5f5f91d | 7922 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7923 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7924 | |
9614634f | 7925 | for_each_zone(zone) |
9705bea5 AK |
7926 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
7927 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 7928 | } |
0ff38490 | 7929 | |
6423aa81 JK |
7930 | |
7931 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 7932 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7933 | { |
0ff38490 CL |
7934 | int rc; |
7935 | ||
8d65af78 | 7936 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7937 | if (rc) |
7938 | return rc; | |
7939 | ||
6423aa81 JK |
7940 | setup_min_unmapped_ratio(); |
7941 | ||
7942 | return 0; | |
7943 | } | |
7944 | ||
7945 | static void setup_min_slab_ratio(void) | |
7946 | { | |
7947 | pg_data_t *pgdat; | |
7948 | struct zone *zone; | |
7949 | ||
a5f5f91d MG |
7950 | for_each_online_pgdat(pgdat) |
7951 | pgdat->min_slab_pages = 0; | |
7952 | ||
0ff38490 | 7953 | for_each_zone(zone) |
9705bea5 AK |
7954 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
7955 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
7956 | } |
7957 | ||
7958 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 7959 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
7960 | { |
7961 | int rc; | |
7962 | ||
7963 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7964 | if (rc) | |
7965 | return rc; | |
7966 | ||
7967 | setup_min_slab_ratio(); | |
7968 | ||
0ff38490 CL |
7969 | return 0; |
7970 | } | |
9614634f CL |
7971 | #endif |
7972 | ||
1da177e4 LT |
7973 | /* |
7974 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7975 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7976 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7977 | * | |
7978 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7979 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7980 | * if in function of the boot time zone sizes. |
7981 | */ | |
cccad5b9 | 7982 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 7983 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7984 | { |
86aaf255 BH |
7985 | int i; |
7986 | ||
8d65af78 | 7987 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
7988 | |
7989 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
7990 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
7991 | sysctl_lowmem_reserve_ratio[i] = 0; | |
7992 | } | |
7993 | ||
1da177e4 LT |
7994 | setup_per_zone_lowmem_reserve(); |
7995 | return 0; | |
7996 | } | |
7997 | ||
cb1ef534 MG |
7998 | static void __zone_pcp_update(struct zone *zone) |
7999 | { | |
8000 | unsigned int cpu; | |
8001 | ||
8002 | for_each_possible_cpu(cpu) | |
8003 | pageset_set_high_and_batch(zone, | |
8004 | per_cpu_ptr(zone->pageset, cpu)); | |
8005 | } | |
8006 | ||
8ad4b1fb RS |
8007 | /* |
8008 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
8009 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
8010 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 8011 | */ |
cccad5b9 | 8012 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8013 | void *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
8014 | { |
8015 | struct zone *zone; | |
7cd2b0a3 | 8016 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
8017 | int ret; |
8018 | ||
7cd2b0a3 DR |
8019 | mutex_lock(&pcp_batch_high_lock); |
8020 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
8021 | ||
8d65af78 | 8022 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
8023 | if (!write || ret < 0) |
8024 | goto out; | |
8025 | ||
8026 | /* Sanity checking to avoid pcp imbalance */ | |
8027 | if (percpu_pagelist_fraction && | |
8028 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
8029 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
8030 | ret = -EINVAL; | |
8031 | goto out; | |
8032 | } | |
8033 | ||
8034 | /* No change? */ | |
8035 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
8036 | goto out; | |
c8e251fa | 8037 | |
cb1ef534 MG |
8038 | for_each_populated_zone(zone) |
8039 | __zone_pcp_update(zone); | |
7cd2b0a3 | 8040 | out: |
c8e251fa | 8041 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 8042 | return ret; |
8ad4b1fb RS |
8043 | } |
8044 | ||
f6f34b43 SD |
8045 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
8046 | /* | |
8047 | * Returns the number of pages that arch has reserved but | |
8048 | * is not known to alloc_large_system_hash(). | |
8049 | */ | |
8050 | static unsigned long __init arch_reserved_kernel_pages(void) | |
8051 | { | |
8052 | return 0; | |
8053 | } | |
8054 | #endif | |
8055 | ||
9017217b PT |
8056 | /* |
8057 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
8058 | * machines. As memory size is increased the scale is also increased but at | |
8059 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
8060 | * quadruples the scale is increased by one, which means the size of hash table | |
8061 | * only doubles, instead of quadrupling as well. | |
8062 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
8063 | * makes sense, it is disabled on such platforms. | |
8064 | */ | |
8065 | #if __BITS_PER_LONG > 32 | |
8066 | #define ADAPT_SCALE_BASE (64ul << 30) | |
8067 | #define ADAPT_SCALE_SHIFT 2 | |
8068 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
8069 | #endif | |
8070 | ||
1da177e4 LT |
8071 | /* |
8072 | * allocate a large system hash table from bootmem | |
8073 | * - it is assumed that the hash table must contain an exact power-of-2 | |
8074 | * quantity of entries | |
8075 | * - limit is the number of hash buckets, not the total allocation size | |
8076 | */ | |
8077 | void *__init alloc_large_system_hash(const char *tablename, | |
8078 | unsigned long bucketsize, | |
8079 | unsigned long numentries, | |
8080 | int scale, | |
8081 | int flags, | |
8082 | unsigned int *_hash_shift, | |
8083 | unsigned int *_hash_mask, | |
31fe62b9 TB |
8084 | unsigned long low_limit, |
8085 | unsigned long high_limit) | |
1da177e4 | 8086 | { |
31fe62b9 | 8087 | unsigned long long max = high_limit; |
1da177e4 LT |
8088 | unsigned long log2qty, size; |
8089 | void *table = NULL; | |
3749a8f0 | 8090 | gfp_t gfp_flags; |
ec11408a | 8091 | bool virt; |
1da177e4 LT |
8092 | |
8093 | /* allow the kernel cmdline to have a say */ | |
8094 | if (!numentries) { | |
8095 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 8096 | numentries = nr_kernel_pages; |
f6f34b43 | 8097 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
8098 | |
8099 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
8100 | if (PAGE_SHIFT < 20) | |
8101 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 8102 | |
9017217b PT |
8103 | #if __BITS_PER_LONG > 32 |
8104 | if (!high_limit) { | |
8105 | unsigned long adapt; | |
8106 | ||
8107 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
8108 | adapt <<= ADAPT_SCALE_SHIFT) | |
8109 | scale++; | |
8110 | } | |
8111 | #endif | |
8112 | ||
1da177e4 LT |
8113 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
8114 | if (scale > PAGE_SHIFT) | |
8115 | numentries >>= (scale - PAGE_SHIFT); | |
8116 | else | |
8117 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
8118 | |
8119 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
8120 | if (unlikely(flags & HASH_SMALL)) { |
8121 | /* Makes no sense without HASH_EARLY */ | |
8122 | WARN_ON(!(flags & HASH_EARLY)); | |
8123 | if (!(numentries >> *_hash_shift)) { | |
8124 | numentries = 1UL << *_hash_shift; | |
8125 | BUG_ON(!numentries); | |
8126 | } | |
8127 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 8128 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 8129 | } |
6e692ed3 | 8130 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
8131 | |
8132 | /* limit allocation size to 1/16 total memory by default */ | |
8133 | if (max == 0) { | |
8134 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
8135 | do_div(max, bucketsize); | |
8136 | } | |
074b8517 | 8137 | max = min(max, 0x80000000ULL); |
1da177e4 | 8138 | |
31fe62b9 TB |
8139 | if (numentries < low_limit) |
8140 | numentries = low_limit; | |
1da177e4 LT |
8141 | if (numentries > max) |
8142 | numentries = max; | |
8143 | ||
f0d1b0b3 | 8144 | log2qty = ilog2(numentries); |
1da177e4 | 8145 | |
3749a8f0 | 8146 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 8147 | do { |
ec11408a | 8148 | virt = false; |
1da177e4 | 8149 | size = bucketsize << log2qty; |
ea1f5f37 PT |
8150 | if (flags & HASH_EARLY) { |
8151 | if (flags & HASH_ZERO) | |
26fb3dae | 8152 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 8153 | else |
7e1c4e27 MR |
8154 | table = memblock_alloc_raw(size, |
8155 | SMP_CACHE_BYTES); | |
ec11408a | 8156 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
88dca4ca | 8157 | table = __vmalloc(size, gfp_flags); |
ec11408a | 8158 | virt = true; |
ea1f5f37 | 8159 | } else { |
1037b83b ED |
8160 | /* |
8161 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
8162 | * some pages at the end of hash table which |
8163 | * alloc_pages_exact() automatically does | |
1037b83b | 8164 | */ |
ec11408a NP |
8165 | table = alloc_pages_exact(size, gfp_flags); |
8166 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
8167 | } |
8168 | } while (!table && size > PAGE_SIZE && --log2qty); | |
8169 | ||
8170 | if (!table) | |
8171 | panic("Failed to allocate %s hash table\n", tablename); | |
8172 | ||
ec11408a NP |
8173 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
8174 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
8175 | virt ? "vmalloc" : "linear"); | |
1da177e4 LT |
8176 | |
8177 | if (_hash_shift) | |
8178 | *_hash_shift = log2qty; | |
8179 | if (_hash_mask) | |
8180 | *_hash_mask = (1 << log2qty) - 1; | |
8181 | ||
8182 | return table; | |
8183 | } | |
a117e66e | 8184 | |
a5d76b54 | 8185 | /* |
80934513 | 8186 | * This function checks whether pageblock includes unmovable pages or not. |
80934513 | 8187 | * |
b8af2941 | 8188 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
8189 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
8190 | * check without lock_page also may miss some movable non-lru pages at | |
8191 | * race condition. So you can't expect this function should be exact. | |
4a55c047 QC |
8192 | * |
8193 | * Returns a page without holding a reference. If the caller wants to | |
8194 | * dereference that page (e.g., dumping), it has to make sure that that it | |
8195 | * cannot get removed (e.g., via memory unplug) concurrently. | |
8196 | * | |
a5d76b54 | 8197 | */ |
4a55c047 QC |
8198 | struct page *has_unmovable_pages(struct zone *zone, struct page *page, |
8199 | int migratetype, int flags) | |
49ac8255 | 8200 | { |
1a9f2191 QC |
8201 | unsigned long iter = 0; |
8202 | unsigned long pfn = page_to_pfn(page); | |
47118af0 | 8203 | |
49ac8255 | 8204 | /* |
15c30bc0 MH |
8205 | * TODO we could make this much more efficient by not checking every |
8206 | * page in the range if we know all of them are in MOVABLE_ZONE and | |
8207 | * that the movable zone guarantees that pages are migratable but | |
8208 | * the later is not the case right now unfortunatelly. E.g. movablecore | |
8209 | * can still lead to having bootmem allocations in zone_movable. | |
49ac8255 | 8210 | */ |
49ac8255 | 8211 | |
1a9f2191 QC |
8212 | if (is_migrate_cma_page(page)) { |
8213 | /* | |
8214 | * CMA allocations (alloc_contig_range) really need to mark | |
8215 | * isolate CMA pageblocks even when they are not movable in fact | |
8216 | * so consider them movable here. | |
8217 | */ | |
8218 | if (is_migrate_cma(migratetype)) | |
4a55c047 | 8219 | return NULL; |
1a9f2191 | 8220 | |
3d680bdf | 8221 | return page; |
1a9f2191 | 8222 | } |
4da2ce25 | 8223 | |
fe4c86c9 DH |
8224 | for (; iter < pageblock_nr_pages; iter++) { |
8225 | if (!pfn_valid_within(pfn + iter)) | |
49ac8255 | 8226 | continue; |
29723fcc | 8227 | |
fe4c86c9 | 8228 | page = pfn_to_page(pfn + iter); |
c8721bbb | 8229 | |
d7ab3672 | 8230 | if (PageReserved(page)) |
3d680bdf | 8231 | return page; |
d7ab3672 | 8232 | |
9d789999 MH |
8233 | /* |
8234 | * If the zone is movable and we have ruled out all reserved | |
8235 | * pages then it should be reasonably safe to assume the rest | |
8236 | * is movable. | |
8237 | */ | |
8238 | if (zone_idx(zone) == ZONE_MOVABLE) | |
8239 | continue; | |
8240 | ||
c8721bbb NH |
8241 | /* |
8242 | * Hugepages are not in LRU lists, but they're movable. | |
1da2f328 | 8243 | * THPs are on the LRU, but need to be counted as #small pages. |
8bb4e7a2 | 8244 | * We need not scan over tail pages because we don't |
c8721bbb NH |
8245 | * handle each tail page individually in migration. |
8246 | */ | |
1da2f328 | 8247 | if (PageHuge(page) || PageTransCompound(page)) { |
17e2e7d7 OS |
8248 | struct page *head = compound_head(page); |
8249 | unsigned int skip_pages; | |
464c7ffb | 8250 | |
1da2f328 RR |
8251 | if (PageHuge(page)) { |
8252 | if (!hugepage_migration_supported(page_hstate(head))) | |
8253 | return page; | |
8254 | } else if (!PageLRU(head) && !__PageMovable(head)) { | |
3d680bdf | 8255 | return page; |
1da2f328 | 8256 | } |
464c7ffb | 8257 | |
d8c6546b | 8258 | skip_pages = compound_nr(head) - (page - head); |
17e2e7d7 | 8259 | iter += skip_pages - 1; |
c8721bbb NH |
8260 | continue; |
8261 | } | |
8262 | ||
97d255c8 MK |
8263 | /* |
8264 | * We can't use page_count without pin a page | |
8265 | * because another CPU can free compound page. | |
8266 | * This check already skips compound tails of THP | |
0139aa7b | 8267 | * because their page->_refcount is zero at all time. |
97d255c8 | 8268 | */ |
fe896d18 | 8269 | if (!page_ref_count(page)) { |
49ac8255 KH |
8270 | if (PageBuddy(page)) |
8271 | iter += (1 << page_order(page)) - 1; | |
8272 | continue; | |
8273 | } | |
97d255c8 | 8274 | |
b023f468 WC |
8275 | /* |
8276 | * The HWPoisoned page may be not in buddy system, and | |
8277 | * page_count() is not 0. | |
8278 | */ | |
756d25be | 8279 | if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
b023f468 WC |
8280 | continue; |
8281 | ||
aa218795 DH |
8282 | /* |
8283 | * We treat all PageOffline() pages as movable when offlining | |
8284 | * to give drivers a chance to decrement their reference count | |
8285 | * in MEM_GOING_OFFLINE in order to indicate that these pages | |
8286 | * can be offlined as there are no direct references anymore. | |
8287 | * For actually unmovable PageOffline() where the driver does | |
8288 | * not support this, we will fail later when trying to actually | |
8289 | * move these pages that still have a reference count > 0. | |
8290 | * (false negatives in this function only) | |
8291 | */ | |
8292 | if ((flags & MEMORY_OFFLINE) && PageOffline(page)) | |
8293 | continue; | |
8294 | ||
fe4c86c9 | 8295 | if (__PageMovable(page) || PageLRU(page)) |
0efadf48 YX |
8296 | continue; |
8297 | ||
49ac8255 | 8298 | /* |
6b4f7799 JW |
8299 | * If there are RECLAIMABLE pages, we need to check |
8300 | * it. But now, memory offline itself doesn't call | |
8301 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
8302 | */ |
8303 | /* | |
8304 | * If the page is not RAM, page_count()should be 0. | |
8305 | * we don't need more check. This is an _used_ not-movable page. | |
8306 | * | |
8307 | * The problematic thing here is PG_reserved pages. PG_reserved | |
8308 | * is set to both of a memory hole page and a _used_ kernel | |
8309 | * page at boot. | |
8310 | */ | |
3d680bdf | 8311 | return page; |
49ac8255 | 8312 | } |
4a55c047 | 8313 | return NULL; |
49ac8255 KH |
8314 | } |
8315 | ||
8df995f6 | 8316 | #ifdef CONFIG_CONTIG_ALLOC |
041d3a8c MN |
8317 | static unsigned long pfn_max_align_down(unsigned long pfn) |
8318 | { | |
8319 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8320 | pageblock_nr_pages) - 1); | |
8321 | } | |
8322 | ||
8323 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
8324 | { | |
8325 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8326 | pageblock_nr_pages)); | |
8327 | } | |
8328 | ||
041d3a8c | 8329 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
8330 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
8331 | unsigned long start, unsigned long end) | |
041d3a8c MN |
8332 | { |
8333 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 8334 | unsigned int nr_reclaimed; |
041d3a8c MN |
8335 | unsigned long pfn = start; |
8336 | unsigned int tries = 0; | |
8337 | int ret = 0; | |
8338 | ||
be49a6e1 | 8339 | migrate_prep(); |
041d3a8c | 8340 | |
bb13ffeb | 8341 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
8342 | if (fatal_signal_pending(current)) { |
8343 | ret = -EINTR; | |
8344 | break; | |
8345 | } | |
8346 | ||
bb13ffeb MG |
8347 | if (list_empty(&cc->migratepages)) { |
8348 | cc->nr_migratepages = 0; | |
edc2ca61 | 8349 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
8350 | if (!pfn) { |
8351 | ret = -EINTR; | |
8352 | break; | |
8353 | } | |
8354 | tries = 0; | |
8355 | } else if (++tries == 5) { | |
8356 | ret = ret < 0 ? ret : -EBUSY; | |
8357 | break; | |
8358 | } | |
8359 | ||
beb51eaa MK |
8360 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
8361 | &cc->migratepages); | |
8362 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 8363 | |
9c620e2b | 8364 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
31025351 | 8365 | NULL, 0, cc->mode, MR_CONTIG_RANGE); |
041d3a8c | 8366 | } |
2a6f5124 SP |
8367 | if (ret < 0) { |
8368 | putback_movable_pages(&cc->migratepages); | |
8369 | return ret; | |
8370 | } | |
8371 | return 0; | |
041d3a8c MN |
8372 | } |
8373 | ||
8374 | /** | |
8375 | * alloc_contig_range() -- tries to allocate given range of pages | |
8376 | * @start: start PFN to allocate | |
8377 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
8378 | * @migratetype: migratetype of the underlaying pageblocks (either |
8379 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
8380 | * in range must have the same migratetype and it must | |
8381 | * be either of the two. | |
ca96b625 | 8382 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
8383 | * |
8384 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 8385 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 8386 | * |
2c7452a0 MK |
8387 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
8388 | * pageblocks in the range. Once isolated, the pageblocks should not | |
8389 | * be modified by others. | |
041d3a8c | 8390 | * |
a862f68a | 8391 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
8392 | * pages which PFN is in [start, end) are allocated for the caller and |
8393 | * need to be freed with free_contig_range(). | |
8394 | */ | |
0815f3d8 | 8395 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 8396 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 8397 | { |
041d3a8c | 8398 | unsigned long outer_start, outer_end; |
d00181b9 KS |
8399 | unsigned int order; |
8400 | int ret = 0; | |
041d3a8c | 8401 | |
bb13ffeb MG |
8402 | struct compact_control cc = { |
8403 | .nr_migratepages = 0, | |
8404 | .order = -1, | |
8405 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 8406 | .mode = MIGRATE_SYNC, |
bb13ffeb | 8407 | .ignore_skip_hint = true, |
2583d671 | 8408 | .no_set_skip_hint = true, |
7dea19f9 | 8409 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 8410 | .alloc_contig = true, |
bb13ffeb MG |
8411 | }; |
8412 | INIT_LIST_HEAD(&cc.migratepages); | |
8413 | ||
041d3a8c MN |
8414 | /* |
8415 | * What we do here is we mark all pageblocks in range as | |
8416 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
8417 | * have different sizes, and due to the way page allocator | |
8418 | * work, we align the range to biggest of the two pages so | |
8419 | * that page allocator won't try to merge buddies from | |
8420 | * different pageblocks and change MIGRATE_ISOLATE to some | |
8421 | * other migration type. | |
8422 | * | |
8423 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
8424 | * migrate the pages from an unaligned range (ie. pages that | |
8425 | * we are interested in). This will put all the pages in | |
8426 | * range back to page allocator as MIGRATE_ISOLATE. | |
8427 | * | |
8428 | * When this is done, we take the pages in range from page | |
8429 | * allocator removing them from the buddy system. This way | |
8430 | * page allocator will never consider using them. | |
8431 | * | |
8432 | * This lets us mark the pageblocks back as | |
8433 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
8434 | * aligned range but not in the unaligned, original range are | |
8435 | * put back to page allocator so that buddy can use them. | |
8436 | */ | |
8437 | ||
8438 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
d381c547 | 8439 | pfn_max_align_up(end), migratetype, 0); |
9b7ea46a | 8440 | if (ret < 0) |
86a595f9 | 8441 | return ret; |
041d3a8c | 8442 | |
8ef5849f JK |
8443 | /* |
8444 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
8445 | * So, just fall through. test_pages_isolated() has a tracepoint |
8446 | * which will report the busy page. | |
8447 | * | |
8448 | * It is possible that busy pages could become available before | |
8449 | * the call to test_pages_isolated, and the range will actually be | |
8450 | * allocated. So, if we fall through be sure to clear ret so that | |
8451 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 8452 | */ |
bb13ffeb | 8453 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 8454 | if (ret && ret != -EBUSY) |
041d3a8c | 8455 | goto done; |
63cd4489 | 8456 | ret =0; |
041d3a8c MN |
8457 | |
8458 | /* | |
8459 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
8460 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
8461 | * more, all pages in [start, end) are free in page allocator. | |
8462 | * What we are going to do is to allocate all pages from | |
8463 | * [start, end) (that is remove them from page allocator). | |
8464 | * | |
8465 | * The only problem is that pages at the beginning and at the | |
8466 | * end of interesting range may be not aligned with pages that | |
8467 | * page allocator holds, ie. they can be part of higher order | |
8468 | * pages. Because of this, we reserve the bigger range and | |
8469 | * once this is done free the pages we are not interested in. | |
8470 | * | |
8471 | * We don't have to hold zone->lock here because the pages are | |
8472 | * isolated thus they won't get removed from buddy. | |
8473 | */ | |
8474 | ||
8475 | lru_add_drain_all(); | |
041d3a8c MN |
8476 | |
8477 | order = 0; | |
8478 | outer_start = start; | |
8479 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
8480 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
8481 | outer_start = start; |
8482 | break; | |
041d3a8c MN |
8483 | } |
8484 | outer_start &= ~0UL << order; | |
8485 | } | |
8486 | ||
8ef5849f JK |
8487 | if (outer_start != start) { |
8488 | order = page_order(pfn_to_page(outer_start)); | |
8489 | ||
8490 | /* | |
8491 | * outer_start page could be small order buddy page and | |
8492 | * it doesn't include start page. Adjust outer_start | |
8493 | * in this case to report failed page properly | |
8494 | * on tracepoint in test_pages_isolated() | |
8495 | */ | |
8496 | if (outer_start + (1UL << order) <= start) | |
8497 | outer_start = start; | |
8498 | } | |
8499 | ||
041d3a8c | 8500 | /* Make sure the range is really isolated. */ |
756d25be | 8501 | if (test_pages_isolated(outer_start, end, 0)) { |
75dddef3 | 8502 | pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n", |
dae803e1 | 8503 | __func__, outer_start, end); |
041d3a8c MN |
8504 | ret = -EBUSY; |
8505 | goto done; | |
8506 | } | |
8507 | ||
49f223a9 | 8508 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 8509 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
8510 | if (!outer_end) { |
8511 | ret = -EBUSY; | |
8512 | goto done; | |
8513 | } | |
8514 | ||
8515 | /* Free head and tail (if any) */ | |
8516 | if (start != outer_start) | |
8517 | free_contig_range(outer_start, start - outer_start); | |
8518 | if (end != outer_end) | |
8519 | free_contig_range(end, outer_end - end); | |
8520 | ||
8521 | done: | |
8522 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 8523 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
8524 | return ret; |
8525 | } | |
255f5985 | 8526 | EXPORT_SYMBOL(alloc_contig_range); |
5e27a2df AK |
8527 | |
8528 | static int __alloc_contig_pages(unsigned long start_pfn, | |
8529 | unsigned long nr_pages, gfp_t gfp_mask) | |
8530 | { | |
8531 | unsigned long end_pfn = start_pfn + nr_pages; | |
8532 | ||
8533 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
8534 | gfp_mask); | |
8535 | } | |
8536 | ||
8537 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
8538 | unsigned long nr_pages) | |
8539 | { | |
8540 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
8541 | struct page *page; | |
8542 | ||
8543 | for (i = start_pfn; i < end_pfn; i++) { | |
8544 | page = pfn_to_online_page(i); | |
8545 | if (!page) | |
8546 | return false; | |
8547 | ||
8548 | if (page_zone(page) != z) | |
8549 | return false; | |
8550 | ||
8551 | if (PageReserved(page)) | |
8552 | return false; | |
8553 | ||
8554 | if (page_count(page) > 0) | |
8555 | return false; | |
8556 | ||
8557 | if (PageHuge(page)) | |
8558 | return false; | |
8559 | } | |
8560 | return true; | |
8561 | } | |
8562 | ||
8563 | static bool zone_spans_last_pfn(const struct zone *zone, | |
8564 | unsigned long start_pfn, unsigned long nr_pages) | |
8565 | { | |
8566 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
8567 | ||
8568 | return zone_spans_pfn(zone, last_pfn); | |
8569 | } | |
8570 | ||
8571 | /** | |
8572 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
8573 | * @nr_pages: Number of contiguous pages to allocate | |
8574 | * @gfp_mask: GFP mask to limit search and used during compaction | |
8575 | * @nid: Target node | |
8576 | * @nodemask: Mask for other possible nodes | |
8577 | * | |
8578 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
8579 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
8580 | * tried for allocation with alloc_contig_range(). This routine is intended | |
8581 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
8582 | * | |
8583 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
8584 | * power of two then the alignment is guaranteed to be to the given nr_pages | |
8585 | * (e.g. 1GB request would be aligned to 1GB). | |
8586 | * | |
8587 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
8588 | * __free_page() on each allocated page. | |
8589 | * | |
8590 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
8591 | */ | |
8592 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
8593 | int nid, nodemask_t *nodemask) | |
8594 | { | |
8595 | unsigned long ret, pfn, flags; | |
8596 | struct zonelist *zonelist; | |
8597 | struct zone *zone; | |
8598 | struct zoneref *z; | |
8599 | ||
8600 | zonelist = node_zonelist(nid, gfp_mask); | |
8601 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
8602 | gfp_zone(gfp_mask), nodemask) { | |
8603 | spin_lock_irqsave(&zone->lock, flags); | |
8604 | ||
8605 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
8606 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
8607 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
8608 | /* | |
8609 | * We release the zone lock here because | |
8610 | * alloc_contig_range() will also lock the zone | |
8611 | * at some point. If there's an allocation | |
8612 | * spinning on this lock, it may win the race | |
8613 | * and cause alloc_contig_range() to fail... | |
8614 | */ | |
8615 | spin_unlock_irqrestore(&zone->lock, flags); | |
8616 | ret = __alloc_contig_pages(pfn, nr_pages, | |
8617 | gfp_mask); | |
8618 | if (!ret) | |
8619 | return pfn_to_page(pfn); | |
8620 | spin_lock_irqsave(&zone->lock, flags); | |
8621 | } | |
8622 | pfn += nr_pages; | |
8623 | } | |
8624 | spin_unlock_irqrestore(&zone->lock, flags); | |
8625 | } | |
8626 | return NULL; | |
8627 | } | |
4eb0716e | 8628 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 8629 | |
4eb0716e | 8630 | void free_contig_range(unsigned long pfn, unsigned int nr_pages) |
041d3a8c | 8631 | { |
bcc2b02f MS |
8632 | unsigned int count = 0; |
8633 | ||
8634 | for (; nr_pages--; pfn++) { | |
8635 | struct page *page = pfn_to_page(pfn); | |
8636 | ||
8637 | count += page_count(page) != 1; | |
8638 | __free_page(page); | |
8639 | } | |
8640 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c | 8641 | } |
255f5985 | 8642 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 8643 | |
0a647f38 CS |
8644 | /* |
8645 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
8646 | * page high values need to be recalulated. | |
8647 | */ | |
4ed7e022 JL |
8648 | void __meminit zone_pcp_update(struct zone *zone) |
8649 | { | |
c8e251fa | 8650 | mutex_lock(&pcp_batch_high_lock); |
cb1ef534 | 8651 | __zone_pcp_update(zone); |
c8e251fa | 8652 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 | 8653 | } |
4ed7e022 | 8654 | |
340175b7 JL |
8655 | void zone_pcp_reset(struct zone *zone) |
8656 | { | |
8657 | unsigned long flags; | |
5a883813 MK |
8658 | int cpu; |
8659 | struct per_cpu_pageset *pset; | |
340175b7 JL |
8660 | |
8661 | /* avoid races with drain_pages() */ | |
8662 | local_irq_save(flags); | |
8663 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
8664 | for_each_online_cpu(cpu) { |
8665 | pset = per_cpu_ptr(zone->pageset, cpu); | |
8666 | drain_zonestat(zone, pset); | |
8667 | } | |
340175b7 JL |
8668 | free_percpu(zone->pageset); |
8669 | zone->pageset = &boot_pageset; | |
8670 | } | |
8671 | local_irq_restore(flags); | |
8672 | } | |
8673 | ||
6dcd73d7 | 8674 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 8675 | /* |
b9eb6319 JK |
8676 | * All pages in the range must be in a single zone and isolated |
8677 | * before calling this. | |
0c0e6195 | 8678 | */ |
5557c766 | 8679 | unsigned long |
0c0e6195 KH |
8680 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
8681 | { | |
8682 | struct page *page; | |
8683 | struct zone *zone; | |
0ee5f4f3 | 8684 | unsigned int order; |
0c0e6195 KH |
8685 | unsigned long pfn; |
8686 | unsigned long flags; | |
5557c766 MH |
8687 | unsigned long offlined_pages = 0; |
8688 | ||
0c0e6195 KH |
8689 | /* find the first valid pfn */ |
8690 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
8691 | if (pfn_valid(pfn)) | |
8692 | break; | |
8693 | if (pfn == end_pfn) | |
5557c766 MH |
8694 | return offlined_pages; |
8695 | ||
2d070eab | 8696 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
8697 | zone = page_zone(pfn_to_page(pfn)); |
8698 | spin_lock_irqsave(&zone->lock, flags); | |
8699 | pfn = start_pfn; | |
8700 | while (pfn < end_pfn) { | |
8701 | if (!pfn_valid(pfn)) { | |
8702 | pfn++; | |
8703 | continue; | |
8704 | } | |
8705 | page = pfn_to_page(pfn); | |
b023f468 WC |
8706 | /* |
8707 | * The HWPoisoned page may be not in buddy system, and | |
8708 | * page_count() is not 0. | |
8709 | */ | |
8710 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
8711 | pfn++; | |
5557c766 | 8712 | offlined_pages++; |
b023f468 WC |
8713 | continue; |
8714 | } | |
aa218795 DH |
8715 | /* |
8716 | * At this point all remaining PageOffline() pages have a | |
8717 | * reference count of 0 and can simply be skipped. | |
8718 | */ | |
8719 | if (PageOffline(page)) { | |
8720 | BUG_ON(page_count(page)); | |
8721 | BUG_ON(PageBuddy(page)); | |
8722 | pfn++; | |
8723 | offlined_pages++; | |
8724 | continue; | |
8725 | } | |
b023f468 | 8726 | |
0c0e6195 KH |
8727 | BUG_ON(page_count(page)); |
8728 | BUG_ON(!PageBuddy(page)); | |
8729 | order = page_order(page); | |
5557c766 | 8730 | offlined_pages += 1 << order; |
6ab01363 | 8731 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
8732 | pfn += (1 << order); |
8733 | } | |
8734 | spin_unlock_irqrestore(&zone->lock, flags); | |
5557c766 MH |
8735 | |
8736 | return offlined_pages; | |
0c0e6195 KH |
8737 | } |
8738 | #endif | |
8d22ba1b | 8739 | |
8d22ba1b WF |
8740 | bool is_free_buddy_page(struct page *page) |
8741 | { | |
8742 | struct zone *zone = page_zone(page); | |
8743 | unsigned long pfn = page_to_pfn(page); | |
8744 | unsigned long flags; | |
7aeb09f9 | 8745 | unsigned int order; |
8d22ba1b WF |
8746 | |
8747 | spin_lock_irqsave(&zone->lock, flags); | |
8748 | for (order = 0; order < MAX_ORDER; order++) { | |
8749 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8750 | ||
8751 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
8752 | break; | |
8753 | } | |
8754 | spin_unlock_irqrestore(&zone->lock, flags); | |
8755 | ||
8756 | return order < MAX_ORDER; | |
8757 | } | |
d4ae9916 NH |
8758 | |
8759 | #ifdef CONFIG_MEMORY_FAILURE | |
8760 | /* | |
8761 | * Set PG_hwpoison flag if a given page is confirmed to be a free page. This | |
8762 | * test is performed under the zone lock to prevent a race against page | |
8763 | * allocation. | |
8764 | */ | |
8765 | bool set_hwpoison_free_buddy_page(struct page *page) | |
8766 | { | |
8767 | struct zone *zone = page_zone(page); | |
8768 | unsigned long pfn = page_to_pfn(page); | |
8769 | unsigned long flags; | |
8770 | unsigned int order; | |
8771 | bool hwpoisoned = false; | |
8772 | ||
8773 | spin_lock_irqsave(&zone->lock, flags); | |
8774 | for (order = 0; order < MAX_ORDER; order++) { | |
8775 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8776 | ||
8777 | if (PageBuddy(page_head) && page_order(page_head) >= order) { | |
8778 | if (!TestSetPageHWPoison(page)) | |
8779 | hwpoisoned = true; | |
8780 | break; | |
8781 | } | |
8782 | } | |
8783 | spin_unlock_irqrestore(&zone->lock, flags); | |
8784 | ||
8785 | return hwpoisoned; | |
8786 | } | |
8787 | #endif |