]>
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 | |
0b423ca2 MG |
472 | * @mask: mask of bits that the caller is interested in |
473 | * | |
474 | * Return: pageblock_bits flags | |
475 | */ | |
535b81e2 WY |
476 | static __always_inline |
477 | unsigned long __get_pfnblock_flags_mask(struct page *page, | |
0b423ca2 | 478 | unsigned long pfn, |
0b423ca2 MG |
479 | unsigned long mask) |
480 | { | |
481 | unsigned long *bitmap; | |
482 | unsigned long bitidx, word_bitidx; | |
483 | unsigned long word; | |
484 | ||
485 | bitmap = get_pageblock_bitmap(page, pfn); | |
486 | bitidx = pfn_to_bitidx(page, pfn); | |
487 | word_bitidx = bitidx / BITS_PER_LONG; | |
488 | bitidx &= (BITS_PER_LONG-1); | |
489 | ||
490 | word = bitmap[word_bitidx]; | |
d93d5ab9 | 491 | return (word >> bitidx) & mask; |
0b423ca2 MG |
492 | } |
493 | ||
494 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
0b423ca2 MG |
495 | unsigned long mask) |
496 | { | |
535b81e2 | 497 | return __get_pfnblock_flags_mask(page, pfn, mask); |
0b423ca2 MG |
498 | } |
499 | ||
500 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
501 | { | |
535b81e2 | 502 | return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK); |
0b423ca2 MG |
503 | } |
504 | ||
505 | /** | |
506 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
507 | * @page: The page within the block of interest | |
508 | * @flags: The flags to set | |
509 | * @pfn: The target page frame number | |
0b423ca2 MG |
510 | * @mask: mask of bits that the caller is interested in |
511 | */ | |
512 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
513 | unsigned long pfn, | |
0b423ca2 MG |
514 | unsigned long mask) |
515 | { | |
516 | unsigned long *bitmap; | |
517 | unsigned long bitidx, word_bitidx; | |
518 | unsigned long old_word, word; | |
519 | ||
520 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 521 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
522 | |
523 | bitmap = get_pageblock_bitmap(page, pfn); | |
524 | bitidx = pfn_to_bitidx(page, pfn); | |
525 | word_bitidx = bitidx / BITS_PER_LONG; | |
526 | bitidx &= (BITS_PER_LONG-1); | |
527 | ||
528 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
529 | ||
d93d5ab9 WY |
530 | mask <<= bitidx; |
531 | flags <<= bitidx; | |
0b423ca2 MG |
532 | |
533 | word = READ_ONCE(bitmap[word_bitidx]); | |
534 | for (;;) { | |
535 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
536 | if (word == old_word) | |
537 | break; | |
538 | word = old_word; | |
539 | } | |
540 | } | |
3a80a7fa | 541 | |
ee6f509c | 542 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 543 | { |
5d0f3f72 KM |
544 | if (unlikely(page_group_by_mobility_disabled && |
545 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
546 | migratetype = MIGRATE_UNMOVABLE; |
547 | ||
d93d5ab9 | 548 | set_pfnblock_flags_mask(page, (unsigned long)migratetype, |
535b81e2 | 549 | page_to_pfn(page), MIGRATETYPE_MASK); |
b2a0ac88 MG |
550 | } |
551 | ||
13e7444b | 552 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 553 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 554 | { |
bdc8cb98 DH |
555 | int ret = 0; |
556 | unsigned seq; | |
557 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 558 | unsigned long sp, start_pfn; |
c6a57e19 | 559 | |
bdc8cb98 DH |
560 | do { |
561 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
562 | start_pfn = zone->zone_start_pfn; |
563 | sp = zone->spanned_pages; | |
108bcc96 | 564 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
565 | ret = 1; |
566 | } while (zone_span_seqretry(zone, seq)); | |
567 | ||
b5e6a5a2 | 568 | if (ret) |
613813e8 DH |
569 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
570 | pfn, zone_to_nid(zone), zone->name, | |
571 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 572 | |
bdc8cb98 | 573 | return ret; |
c6a57e19 DH |
574 | } |
575 | ||
576 | static int page_is_consistent(struct zone *zone, struct page *page) | |
577 | { | |
14e07298 | 578 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 579 | return 0; |
1da177e4 | 580 | if (zone != page_zone(page)) |
c6a57e19 DH |
581 | return 0; |
582 | ||
583 | return 1; | |
584 | } | |
585 | /* | |
586 | * Temporary debugging check for pages not lying within a given zone. | |
587 | */ | |
d73d3c9f | 588 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
589 | { |
590 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 591 | return 1; |
c6a57e19 DH |
592 | if (!page_is_consistent(zone, page)) |
593 | return 1; | |
594 | ||
1da177e4 LT |
595 | return 0; |
596 | } | |
13e7444b | 597 | #else |
d73d3c9f | 598 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
599 | { |
600 | return 0; | |
601 | } | |
602 | #endif | |
603 | ||
82a3241a | 604 | static void bad_page(struct page *page, const char *reason) |
1da177e4 | 605 | { |
d936cf9b HD |
606 | static unsigned long resume; |
607 | static unsigned long nr_shown; | |
608 | static unsigned long nr_unshown; | |
609 | ||
610 | /* | |
611 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
612 | * or allow a steady drip of one report per second. | |
613 | */ | |
614 | if (nr_shown == 60) { | |
615 | if (time_before(jiffies, resume)) { | |
616 | nr_unshown++; | |
617 | goto out; | |
618 | } | |
619 | if (nr_unshown) { | |
ff8e8116 | 620 | pr_alert( |
1e9e6365 | 621 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
622 | nr_unshown); |
623 | nr_unshown = 0; | |
624 | } | |
625 | nr_shown = 0; | |
626 | } | |
627 | if (nr_shown++ == 0) | |
628 | resume = jiffies + 60 * HZ; | |
629 | ||
ff8e8116 | 630 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 631 | current->comm, page_to_pfn(page)); |
ff8e8116 | 632 | __dump_page(page, reason); |
4e462112 | 633 | dump_page_owner(page); |
3dc14741 | 634 | |
4f31888c | 635 | print_modules(); |
1da177e4 | 636 | dump_stack(); |
d936cf9b | 637 | out: |
8cc3b392 | 638 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 639 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 640 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
641 | } |
642 | ||
1da177e4 LT |
643 | /* |
644 | * Higher-order pages are called "compound pages". They are structured thusly: | |
645 | * | |
1d798ca3 | 646 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 647 | * |
1d798ca3 KS |
648 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
649 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 650 | * |
1d798ca3 KS |
651 | * The first tail page's ->compound_dtor holds the offset in array of compound |
652 | * page destructors. See compound_page_dtors. | |
1da177e4 | 653 | * |
1d798ca3 | 654 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 655 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 656 | */ |
d98c7a09 | 657 | |
9a982250 | 658 | void free_compound_page(struct page *page) |
d98c7a09 | 659 | { |
7ae88534 | 660 | mem_cgroup_uncharge(page); |
d85f3385 | 661 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
662 | } |
663 | ||
d00181b9 | 664 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
665 | { |
666 | int i; | |
667 | int nr_pages = 1 << order; | |
668 | ||
18229df5 AW |
669 | __SetPageHead(page); |
670 | for (i = 1; i < nr_pages; i++) { | |
671 | struct page *p = page + i; | |
58a84aa9 | 672 | set_page_count(p, 0); |
1c290f64 | 673 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 674 | set_compound_head(p, page); |
18229df5 | 675 | } |
1378a5ee MWO |
676 | |
677 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); | |
678 | set_compound_order(page, order); | |
53f9263b | 679 | atomic_set(compound_mapcount_ptr(page), -1); |
47e29d32 JH |
680 | if (hpage_pincount_available(page)) |
681 | atomic_set(compound_pincount_ptr(page), 0); | |
18229df5 AW |
682 | } |
683 | ||
c0a32fc5 SG |
684 | #ifdef CONFIG_DEBUG_PAGEALLOC |
685 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 686 | |
8e57f8ac VB |
687 | bool _debug_pagealloc_enabled_early __read_mostly |
688 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
689 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 690 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 691 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
692 | |
693 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 694 | |
031bc574 JK |
695 | static int __init early_debug_pagealloc(char *buf) |
696 | { | |
8e57f8ac | 697 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
698 | } |
699 | early_param("debug_pagealloc", early_debug_pagealloc); | |
700 | ||
8e57f8ac | 701 | void init_debug_pagealloc(void) |
e30825f1 | 702 | { |
031bc574 JK |
703 | if (!debug_pagealloc_enabled()) |
704 | return; | |
705 | ||
8e57f8ac VB |
706 | static_branch_enable(&_debug_pagealloc_enabled); |
707 | ||
f1c1e9f7 JK |
708 | if (!debug_guardpage_minorder()) |
709 | return; | |
710 | ||
96a2b03f | 711 | static_branch_enable(&_debug_guardpage_enabled); |
e30825f1 JK |
712 | } |
713 | ||
c0a32fc5 SG |
714 | static int __init debug_guardpage_minorder_setup(char *buf) |
715 | { | |
716 | unsigned long res; | |
717 | ||
718 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 719 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
720 | return 0; |
721 | } | |
722 | _debug_guardpage_minorder = res; | |
1170532b | 723 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
724 | return 0; |
725 | } | |
f1c1e9f7 | 726 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 727 | |
acbc15a4 | 728 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 729 | unsigned int order, int migratetype) |
c0a32fc5 | 730 | { |
e30825f1 | 731 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
732 | return false; |
733 | ||
734 | if (order >= debug_guardpage_minorder()) | |
735 | return false; | |
e30825f1 | 736 | |
3972f6bb | 737 | __SetPageGuard(page); |
2847cf95 JK |
738 | INIT_LIST_HEAD(&page->lru); |
739 | set_page_private(page, order); | |
740 | /* Guard pages are not available for any usage */ | |
741 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
742 | |
743 | return true; | |
c0a32fc5 SG |
744 | } |
745 | ||
2847cf95 JK |
746 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
747 | unsigned int order, int migratetype) | |
c0a32fc5 | 748 | { |
e30825f1 JK |
749 | if (!debug_guardpage_enabled()) |
750 | return; | |
751 | ||
3972f6bb | 752 | __ClearPageGuard(page); |
e30825f1 | 753 | |
2847cf95 JK |
754 | set_page_private(page, 0); |
755 | if (!is_migrate_isolate(migratetype)) | |
756 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
757 | } |
758 | #else | |
acbc15a4 JK |
759 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
760 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
761 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
762 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
763 | #endif |
764 | ||
7aeb09f9 | 765 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 766 | { |
4c21e2f2 | 767 | set_page_private(page, order); |
676165a8 | 768 | __SetPageBuddy(page); |
1da177e4 LT |
769 | } |
770 | ||
1da177e4 LT |
771 | /* |
772 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 773 | * we can coalesce a page and its buddy if |
13ad59df | 774 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 775 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
776 | * (c) a page and its buddy have the same order && |
777 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 778 | * |
6e292b9b MW |
779 | * For recording whether a page is in the buddy system, we set PageBuddy. |
780 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 781 | * |
676165a8 | 782 | * For recording page's order, we use page_private(page). |
1da177e4 | 783 | */ |
fe925c0c | 784 | static inline bool page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 785 | unsigned int order) |
1da177e4 | 786 | { |
fe925c0c | 787 | if (!page_is_guard(buddy) && !PageBuddy(buddy)) |
788 | return false; | |
4c5018ce | 789 | |
fe925c0c | 790 | if (page_order(buddy) != order) |
791 | return false; | |
c0a32fc5 | 792 | |
fe925c0c | 793 | /* |
794 | * zone check is done late to avoid uselessly calculating | |
795 | * zone/node ids for pages that could never merge. | |
796 | */ | |
797 | if (page_zone_id(page) != page_zone_id(buddy)) | |
798 | return false; | |
d34c5fa0 | 799 | |
fe925c0c | 800 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
4c5018ce | 801 | |
fe925c0c | 802 | return true; |
1da177e4 LT |
803 | } |
804 | ||
5e1f0f09 MG |
805 | #ifdef CONFIG_COMPACTION |
806 | static inline struct capture_control *task_capc(struct zone *zone) | |
807 | { | |
808 | struct capture_control *capc = current->capture_control; | |
809 | ||
deba0487 | 810 | return unlikely(capc) && |
5e1f0f09 MG |
811 | !(current->flags & PF_KTHREAD) && |
812 | !capc->page && | |
deba0487 | 813 | capc->cc->zone == zone ? capc : NULL; |
5e1f0f09 MG |
814 | } |
815 | ||
816 | static inline bool | |
817 | compaction_capture(struct capture_control *capc, struct page *page, | |
818 | int order, int migratetype) | |
819 | { | |
820 | if (!capc || order != capc->cc->order) | |
821 | return false; | |
822 | ||
823 | /* Do not accidentally pollute CMA or isolated regions*/ | |
824 | if (is_migrate_cma(migratetype) || | |
825 | is_migrate_isolate(migratetype)) | |
826 | return false; | |
827 | ||
828 | /* | |
829 | * Do not let lower order allocations polluate a movable pageblock. | |
830 | * This might let an unmovable request use a reclaimable pageblock | |
831 | * and vice-versa but no more than normal fallback logic which can | |
832 | * have trouble finding a high-order free page. | |
833 | */ | |
834 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
835 | return false; | |
836 | ||
837 | capc->page = page; | |
838 | return true; | |
839 | } | |
840 | ||
841 | #else | |
842 | static inline struct capture_control *task_capc(struct zone *zone) | |
843 | { | |
844 | return NULL; | |
845 | } | |
846 | ||
847 | static inline bool | |
848 | compaction_capture(struct capture_control *capc, struct page *page, | |
849 | int order, int migratetype) | |
850 | { | |
851 | return false; | |
852 | } | |
853 | #endif /* CONFIG_COMPACTION */ | |
854 | ||
6ab01363 AD |
855 | /* Used for pages not on another list */ |
856 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
857 | unsigned int order, int migratetype) | |
858 | { | |
859 | struct free_area *area = &zone->free_area[order]; | |
860 | ||
861 | list_add(&page->lru, &area->free_list[migratetype]); | |
862 | area->nr_free++; | |
863 | } | |
864 | ||
865 | /* Used for pages not on another list */ | |
866 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
867 | unsigned int order, int migratetype) | |
868 | { | |
869 | struct free_area *area = &zone->free_area[order]; | |
870 | ||
871 | list_add_tail(&page->lru, &area->free_list[migratetype]); | |
872 | area->nr_free++; | |
873 | } | |
874 | ||
875 | /* Used for pages which are on another list */ | |
876 | static inline void move_to_free_list(struct page *page, struct zone *zone, | |
877 | unsigned int order, int migratetype) | |
878 | { | |
879 | struct free_area *area = &zone->free_area[order]; | |
880 | ||
881 | list_move(&page->lru, &area->free_list[migratetype]); | |
882 | } | |
883 | ||
884 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
885 | unsigned int order) | |
886 | { | |
36e66c55 AD |
887 | /* clear reported state and update reported page count */ |
888 | if (page_reported(page)) | |
889 | __ClearPageReported(page); | |
890 | ||
6ab01363 AD |
891 | list_del(&page->lru); |
892 | __ClearPageBuddy(page); | |
893 | set_page_private(page, 0); | |
894 | zone->free_area[order].nr_free--; | |
895 | } | |
896 | ||
a2129f24 AD |
897 | /* |
898 | * If this is not the largest possible page, check if the buddy | |
899 | * of the next-highest order is free. If it is, it's possible | |
900 | * that pages are being freed that will coalesce soon. In case, | |
901 | * that is happening, add the free page to the tail of the list | |
902 | * so it's less likely to be used soon and more likely to be merged | |
903 | * as a higher order page | |
904 | */ | |
905 | static inline bool | |
906 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
907 | struct page *page, unsigned int order) | |
908 | { | |
909 | struct page *higher_page, *higher_buddy; | |
910 | unsigned long combined_pfn; | |
911 | ||
912 | if (order >= MAX_ORDER - 2) | |
913 | return false; | |
914 | ||
915 | if (!pfn_valid_within(buddy_pfn)) | |
916 | return false; | |
917 | ||
918 | combined_pfn = buddy_pfn & pfn; | |
919 | higher_page = page + (combined_pfn - pfn); | |
920 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
921 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
922 | ||
923 | return pfn_valid_within(buddy_pfn) && | |
924 | page_is_buddy(higher_page, higher_buddy, order + 1); | |
925 | } | |
926 | ||
1da177e4 LT |
927 | /* |
928 | * Freeing function for a buddy system allocator. | |
929 | * | |
930 | * The concept of a buddy system is to maintain direct-mapped table | |
931 | * (containing bit values) for memory blocks of various "orders". | |
932 | * The bottom level table contains the map for the smallest allocatable | |
933 | * units of memory (here, pages), and each level above it describes | |
934 | * pairs of units from the levels below, hence, "buddies". | |
935 | * At a high level, all that happens here is marking the table entry | |
936 | * at the bottom level available, and propagating the changes upward | |
937 | * as necessary, plus some accounting needed to play nicely with other | |
938 | * parts of the VM system. | |
939 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
940 | * free pages of length of (1 << order) and marked with PageBuddy. |
941 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 942 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
943 | * other. That is, if we allocate a small block, and both were |
944 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 945 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 946 | * triggers coalescing into a block of larger size. |
1da177e4 | 947 | * |
6d49e352 | 948 | * -- nyc |
1da177e4 LT |
949 | */ |
950 | ||
48db57f8 | 951 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 952 | unsigned long pfn, |
ed0ae21d | 953 | struct zone *zone, unsigned int order, |
36e66c55 | 954 | int migratetype, bool report) |
1da177e4 | 955 | { |
a2129f24 | 956 | struct capture_control *capc = task_capc(zone); |
3f649ab7 | 957 | unsigned long buddy_pfn; |
a2129f24 | 958 | unsigned long combined_pfn; |
d9dddbf5 | 959 | unsigned int max_order; |
a2129f24 AD |
960 | struct page *buddy; |
961 | bool to_tail; | |
d9dddbf5 VB |
962 | |
963 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 964 | |
d29bb978 | 965 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 966 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 967 | |
ed0ae21d | 968 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 969 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 970 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 971 | |
76741e77 | 972 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 973 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 974 | |
d9dddbf5 | 975 | continue_merging: |
3c605096 | 976 | while (order < max_order - 1) { |
5e1f0f09 MG |
977 | if (compaction_capture(capc, page, order, migratetype)) { |
978 | __mod_zone_freepage_state(zone, -(1 << order), | |
979 | migratetype); | |
980 | return; | |
981 | } | |
76741e77 VB |
982 | buddy_pfn = __find_buddy_pfn(pfn, order); |
983 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
984 | |
985 | if (!pfn_valid_within(buddy_pfn)) | |
986 | goto done_merging; | |
cb2b95e1 | 987 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 988 | goto done_merging; |
c0a32fc5 SG |
989 | /* |
990 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
991 | * merge with it and move up one order. | |
992 | */ | |
b03641af | 993 | if (page_is_guard(buddy)) |
2847cf95 | 994 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 995 | else |
6ab01363 | 996 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
997 | combined_pfn = buddy_pfn & pfn; |
998 | page = page + (combined_pfn - pfn); | |
999 | pfn = combined_pfn; | |
1da177e4 LT |
1000 | order++; |
1001 | } | |
d9dddbf5 VB |
1002 | if (max_order < MAX_ORDER) { |
1003 | /* If we are here, it means order is >= pageblock_order. | |
1004 | * We want to prevent merge between freepages on isolate | |
1005 | * pageblock and normal pageblock. Without this, pageblock | |
1006 | * isolation could cause incorrect freepage or CMA accounting. | |
1007 | * | |
1008 | * We don't want to hit this code for the more frequent | |
1009 | * low-order merging. | |
1010 | */ | |
1011 | if (unlikely(has_isolate_pageblock(zone))) { | |
1012 | int buddy_mt; | |
1013 | ||
76741e77 VB |
1014 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1015 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
1016 | buddy_mt = get_pageblock_migratetype(buddy); |
1017 | ||
1018 | if (migratetype != buddy_mt | |
1019 | && (is_migrate_isolate(migratetype) || | |
1020 | is_migrate_isolate(buddy_mt))) | |
1021 | goto done_merging; | |
1022 | } | |
1023 | max_order++; | |
1024 | goto continue_merging; | |
1025 | } | |
1026 | ||
1027 | done_merging: | |
1da177e4 | 1028 | set_page_order(page, order); |
6dda9d55 | 1029 | |
97500a4a | 1030 | if (is_shuffle_order(order)) |
a2129f24 | 1031 | to_tail = shuffle_pick_tail(); |
97500a4a | 1032 | else |
a2129f24 | 1033 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 1034 | |
a2129f24 | 1035 | if (to_tail) |
6ab01363 | 1036 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 1037 | else |
6ab01363 | 1038 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
1039 | |
1040 | /* Notify page reporting subsystem of freed page */ | |
1041 | if (report) | |
1042 | page_reporting_notify_free(order); | |
1da177e4 LT |
1043 | } |
1044 | ||
7bfec6f4 MG |
1045 | /* |
1046 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1047 | * try and check multiple fields with one check. The caller must do a detailed | |
1048 | * check if necessary. | |
1049 | */ | |
1050 | static inline bool page_expected_state(struct page *page, | |
1051 | unsigned long check_flags) | |
1052 | { | |
1053 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1054 | return false; | |
1055 | ||
1056 | if (unlikely((unsigned long)page->mapping | | |
1057 | page_ref_count(page) | | |
1058 | #ifdef CONFIG_MEMCG | |
1059 | (unsigned long)page->mem_cgroup | | |
1060 | #endif | |
1061 | (page->flags & check_flags))) | |
1062 | return false; | |
1063 | ||
1064 | return true; | |
1065 | } | |
1066 | ||
58b7f119 | 1067 | static const char *page_bad_reason(struct page *page, unsigned long flags) |
1da177e4 | 1068 | { |
82a3241a | 1069 | const char *bad_reason = NULL; |
f0b791a3 | 1070 | |
53f9263b | 1071 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1072 | bad_reason = "nonzero mapcount"; |
1073 | if (unlikely(page->mapping != NULL)) | |
1074 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1075 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1076 | bad_reason = "nonzero _refcount"; |
58b7f119 WY |
1077 | if (unlikely(page->flags & flags)) { |
1078 | if (flags == PAGE_FLAGS_CHECK_AT_PREP) | |
1079 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set"; | |
1080 | else | |
1081 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
f0b791a3 | 1082 | } |
9edad6ea JW |
1083 | #ifdef CONFIG_MEMCG |
1084 | if (unlikely(page->mem_cgroup)) | |
1085 | bad_reason = "page still charged to cgroup"; | |
1086 | #endif | |
58b7f119 WY |
1087 | return bad_reason; |
1088 | } | |
1089 | ||
1090 | static void check_free_page_bad(struct page *page) | |
1091 | { | |
1092 | bad_page(page, | |
1093 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE)); | |
bb552ac6 MG |
1094 | } |
1095 | ||
534fe5e3 | 1096 | static inline int check_free_page(struct page *page) |
bb552ac6 | 1097 | { |
da838d4f | 1098 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 1099 | return 0; |
bb552ac6 MG |
1100 | |
1101 | /* Something has gone sideways, find it */ | |
0d0c48a2 | 1102 | check_free_page_bad(page); |
7bfec6f4 | 1103 | return 1; |
1da177e4 LT |
1104 | } |
1105 | ||
4db7548c MG |
1106 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1107 | { | |
1108 | int ret = 1; | |
1109 | ||
1110 | /* | |
1111 | * We rely page->lru.next never has bit 0 set, unless the page | |
1112 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1113 | */ | |
1114 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1115 | ||
1116 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1117 | ret = 0; | |
1118 | goto out; | |
1119 | } | |
1120 | switch (page - head_page) { | |
1121 | case 1: | |
4da1984e | 1122 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c | 1123 | if (unlikely(compound_mapcount(page))) { |
82a3241a | 1124 | bad_page(page, "nonzero compound_mapcount"); |
4db7548c MG |
1125 | goto out; |
1126 | } | |
1127 | break; | |
1128 | case 2: | |
1129 | /* | |
1130 | * the second tail page: ->mapping is | |
fa3015b7 | 1131 | * deferred_list.next -- ignore value. |
4db7548c MG |
1132 | */ |
1133 | break; | |
1134 | default: | |
1135 | if (page->mapping != TAIL_MAPPING) { | |
82a3241a | 1136 | bad_page(page, "corrupted mapping in tail page"); |
4db7548c MG |
1137 | goto out; |
1138 | } | |
1139 | break; | |
1140 | } | |
1141 | if (unlikely(!PageTail(page))) { | |
82a3241a | 1142 | bad_page(page, "PageTail not set"); |
4db7548c MG |
1143 | goto out; |
1144 | } | |
1145 | if (unlikely(compound_head(page) != head_page)) { | |
82a3241a | 1146 | bad_page(page, "compound_head not consistent"); |
4db7548c MG |
1147 | goto out; |
1148 | } | |
1149 | ret = 0; | |
1150 | out: | |
1151 | page->mapping = NULL; | |
1152 | clear_compound_head(page); | |
1153 | return ret; | |
1154 | } | |
1155 | ||
6471384a AP |
1156 | static void kernel_init_free_pages(struct page *page, int numpages) |
1157 | { | |
1158 | int i; | |
1159 | ||
9e15afa5 QC |
1160 | /* s390's use of memset() could override KASAN redzones. */ |
1161 | kasan_disable_current(); | |
6471384a AP |
1162 | for (i = 0; i < numpages; i++) |
1163 | clear_highpage(page + i); | |
9e15afa5 | 1164 | kasan_enable_current(); |
6471384a AP |
1165 | } |
1166 | ||
e2769dbd MG |
1167 | static __always_inline bool free_pages_prepare(struct page *page, |
1168 | unsigned int order, bool check_free) | |
4db7548c | 1169 | { |
e2769dbd | 1170 | int bad = 0; |
4db7548c | 1171 | |
4db7548c MG |
1172 | VM_BUG_ON_PAGE(PageTail(page), page); |
1173 | ||
e2769dbd | 1174 | trace_mm_page_free(page, order); |
e2769dbd MG |
1175 | |
1176 | /* | |
1177 | * Check tail pages before head page information is cleared to | |
1178 | * avoid checking PageCompound for order-0 pages. | |
1179 | */ | |
1180 | if (unlikely(order)) { | |
1181 | bool compound = PageCompound(page); | |
1182 | int i; | |
1183 | ||
1184 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1185 | |
9a73f61b KS |
1186 | if (compound) |
1187 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1188 | for (i = 1; i < (1 << order); i++) { |
1189 | if (compound) | |
1190 | bad += free_tail_pages_check(page, page + i); | |
534fe5e3 | 1191 | if (unlikely(check_free_page(page + i))) { |
e2769dbd MG |
1192 | bad++; |
1193 | continue; | |
1194 | } | |
1195 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1196 | } | |
1197 | } | |
bda807d4 | 1198 | if (PageMappingFlags(page)) |
4db7548c | 1199 | page->mapping = NULL; |
c4159a75 | 1200 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
f4b00eab | 1201 | __memcg_kmem_uncharge_page(page, order); |
e2769dbd | 1202 | if (check_free) |
534fe5e3 | 1203 | bad += check_free_page(page); |
e2769dbd MG |
1204 | if (bad) |
1205 | return false; | |
4db7548c | 1206 | |
e2769dbd MG |
1207 | page_cpupid_reset_last(page); |
1208 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1209 | reset_page_owner(page, order); | |
4db7548c MG |
1210 | |
1211 | if (!PageHighMem(page)) { | |
1212 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1213 | PAGE_SIZE << order); |
4db7548c | 1214 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1215 | PAGE_SIZE << order); |
4db7548c | 1216 | } |
6471384a AP |
1217 | if (want_init_on_free()) |
1218 | kernel_init_free_pages(page, 1 << order); | |
1219 | ||
e2769dbd | 1220 | kernel_poison_pages(page, 1 << order, 0); |
234fdce8 QC |
1221 | /* |
1222 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1223 | * does this. So nothing which can access the page's contents should | |
1224 | * happen after this. | |
1225 | */ | |
1226 | arch_free_page(page, order); | |
1227 | ||
8e57f8ac | 1228 | if (debug_pagealloc_enabled_static()) |
d6332692 RE |
1229 | kernel_map_pages(page, 1 << order, 0); |
1230 | ||
3c0c12cc | 1231 | kasan_free_nondeferred_pages(page, order); |
4db7548c | 1232 | |
4db7548c MG |
1233 | return true; |
1234 | } | |
1235 | ||
e2769dbd | 1236 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1237 | /* |
1238 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1239 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1240 | * moved from pcp lists to free lists. | |
1241 | */ | |
1242 | static bool free_pcp_prepare(struct page *page) | |
e2769dbd MG |
1243 | { |
1244 | return free_pages_prepare(page, 0, true); | |
1245 | } | |
1246 | ||
4462b32c | 1247 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1248 | { |
8e57f8ac | 1249 | if (debug_pagealloc_enabled_static()) |
534fe5e3 | 1250 | return check_free_page(page); |
4462b32c VB |
1251 | else |
1252 | return false; | |
e2769dbd MG |
1253 | } |
1254 | #else | |
4462b32c VB |
1255 | /* |
1256 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1257 | * moving from pcp lists to free list in order to reduce overhead. With | |
1258 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1259 | * to the pcp lists. | |
1260 | */ | |
e2769dbd MG |
1261 | static bool free_pcp_prepare(struct page *page) |
1262 | { | |
8e57f8ac | 1263 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
1264 | return free_pages_prepare(page, 0, true); |
1265 | else | |
1266 | return free_pages_prepare(page, 0, false); | |
e2769dbd MG |
1267 | } |
1268 | ||
4db7548c MG |
1269 | static bool bulkfree_pcp_prepare(struct page *page) |
1270 | { | |
534fe5e3 | 1271 | return check_free_page(page); |
4db7548c MG |
1272 | } |
1273 | #endif /* CONFIG_DEBUG_VM */ | |
1274 | ||
97334162 AL |
1275 | static inline void prefetch_buddy(struct page *page) |
1276 | { | |
1277 | unsigned long pfn = page_to_pfn(page); | |
1278 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1279 | struct page *buddy = page + (buddy_pfn - pfn); | |
1280 | ||
1281 | prefetch(buddy); | |
1282 | } | |
1283 | ||
1da177e4 | 1284 | /* |
5f8dcc21 | 1285 | * Frees a number of pages from the PCP lists |
1da177e4 | 1286 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1287 | * count is the number of pages to free. |
1da177e4 LT |
1288 | * |
1289 | * If the zone was previously in an "all pages pinned" state then look to | |
1290 | * see if this freeing clears that state. | |
1291 | * | |
1292 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1293 | * pinned" detection logic. | |
1294 | */ | |
5f8dcc21 MG |
1295 | static void free_pcppages_bulk(struct zone *zone, int count, |
1296 | struct per_cpu_pages *pcp) | |
1da177e4 | 1297 | { |
5f8dcc21 | 1298 | int migratetype = 0; |
a6f9edd6 | 1299 | int batch_free = 0; |
97334162 | 1300 | int prefetch_nr = 0; |
3777999d | 1301 | bool isolated_pageblocks; |
0a5f4e5b AL |
1302 | struct page *page, *tmp; |
1303 | LIST_HEAD(head); | |
f2260e6b | 1304 | |
e5b31ac2 | 1305 | while (count) { |
5f8dcc21 MG |
1306 | struct list_head *list; |
1307 | ||
1308 | /* | |
a6f9edd6 MG |
1309 | * Remove pages from lists in a round-robin fashion. A |
1310 | * batch_free count is maintained that is incremented when an | |
1311 | * empty list is encountered. This is so more pages are freed | |
1312 | * off fuller lists instead of spinning excessively around empty | |
1313 | * lists | |
5f8dcc21 MG |
1314 | */ |
1315 | do { | |
a6f9edd6 | 1316 | batch_free++; |
5f8dcc21 MG |
1317 | if (++migratetype == MIGRATE_PCPTYPES) |
1318 | migratetype = 0; | |
1319 | list = &pcp->lists[migratetype]; | |
1320 | } while (list_empty(list)); | |
48db57f8 | 1321 | |
1d16871d NK |
1322 | /* This is the only non-empty list. Free them all. */ |
1323 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1324 | batch_free = count; |
1d16871d | 1325 | |
a6f9edd6 | 1326 | do { |
a16601c5 | 1327 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1328 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1329 | list_del(&page->lru); |
77ba9062 | 1330 | pcp->count--; |
aa016d14 | 1331 | |
4db7548c MG |
1332 | if (bulkfree_pcp_prepare(page)) |
1333 | continue; | |
1334 | ||
0a5f4e5b | 1335 | list_add_tail(&page->lru, &head); |
97334162 AL |
1336 | |
1337 | /* | |
1338 | * We are going to put the page back to the global | |
1339 | * pool, prefetch its buddy to speed up later access | |
1340 | * under zone->lock. It is believed the overhead of | |
1341 | * an additional test and calculating buddy_pfn here | |
1342 | * can be offset by reduced memory latency later. To | |
1343 | * avoid excessive prefetching due to large count, only | |
1344 | * prefetch buddy for the first pcp->batch nr of pages. | |
1345 | */ | |
1346 | if (prefetch_nr++ < pcp->batch) | |
1347 | prefetch_buddy(page); | |
e5b31ac2 | 1348 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1349 | } |
0a5f4e5b AL |
1350 | |
1351 | spin_lock(&zone->lock); | |
1352 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1353 | ||
1354 | /* | |
1355 | * Use safe version since after __free_one_page(), | |
1356 | * page->lru.next will not point to original list. | |
1357 | */ | |
1358 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1359 | int mt = get_pcppage_migratetype(page); | |
1360 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1361 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1362 | /* Pageblock could have been isolated meanwhile */ | |
1363 | if (unlikely(isolated_pageblocks)) | |
1364 | mt = get_pageblock_migratetype(page); | |
1365 | ||
36e66c55 | 1366 | __free_one_page(page, page_to_pfn(page), zone, 0, mt, true); |
0a5f4e5b AL |
1367 | trace_mm_page_pcpu_drain(page, 0, mt); |
1368 | } | |
d34b0733 | 1369 | spin_unlock(&zone->lock); |
1da177e4 LT |
1370 | } |
1371 | ||
dc4b0caf MG |
1372 | static void free_one_page(struct zone *zone, |
1373 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1374 | unsigned int order, |
ed0ae21d | 1375 | int migratetype) |
1da177e4 | 1376 | { |
d34b0733 | 1377 | spin_lock(&zone->lock); |
ad53f92e JK |
1378 | if (unlikely(has_isolate_pageblock(zone) || |
1379 | is_migrate_isolate(migratetype))) { | |
1380 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1381 | } |
36e66c55 | 1382 | __free_one_page(page, pfn, zone, order, migratetype, true); |
d34b0733 | 1383 | spin_unlock(&zone->lock); |
48db57f8 NP |
1384 | } |
1385 | ||
1e8ce83c | 1386 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1387 | unsigned long zone, int nid) |
1e8ce83c | 1388 | { |
d0dc12e8 | 1389 | mm_zero_struct_page(page); |
1e8ce83c | 1390 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1391 | init_page_count(page); |
1392 | page_mapcount_reset(page); | |
1393 | page_cpupid_reset_last(page); | |
2813b9c0 | 1394 | page_kasan_tag_reset(page); |
1e8ce83c | 1395 | |
1e8ce83c RH |
1396 | INIT_LIST_HEAD(&page->lru); |
1397 | #ifdef WANT_PAGE_VIRTUAL | |
1398 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1399 | if (!is_highmem_idx(zone)) | |
1400 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1401 | #endif | |
1402 | } | |
1403 | ||
7e18adb4 | 1404 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1405 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1406 | { |
1407 | pg_data_t *pgdat; | |
1408 | int nid, zid; | |
1409 | ||
1410 | if (!early_page_uninitialised(pfn)) | |
1411 | return; | |
1412 | ||
1413 | nid = early_pfn_to_nid(pfn); | |
1414 | pgdat = NODE_DATA(nid); | |
1415 | ||
1416 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1417 | struct zone *zone = &pgdat->node_zones[zid]; | |
1418 | ||
1419 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1420 | break; | |
1421 | } | |
d0dc12e8 | 1422 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1423 | } |
1424 | #else | |
1425 | static inline void init_reserved_page(unsigned long pfn) | |
1426 | { | |
1427 | } | |
1428 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1429 | ||
92923ca3 NZ |
1430 | /* |
1431 | * Initialised pages do not have PageReserved set. This function is | |
1432 | * called for each range allocated by the bootmem allocator and | |
1433 | * marks the pages PageReserved. The remaining valid pages are later | |
1434 | * sent to the buddy page allocator. | |
1435 | */ | |
4b50bcc7 | 1436 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1437 | { |
1438 | unsigned long start_pfn = PFN_DOWN(start); | |
1439 | unsigned long end_pfn = PFN_UP(end); | |
1440 | ||
7e18adb4 MG |
1441 | for (; start_pfn < end_pfn; start_pfn++) { |
1442 | if (pfn_valid(start_pfn)) { | |
1443 | struct page *page = pfn_to_page(start_pfn); | |
1444 | ||
1445 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1446 | |
1447 | /* Avoid false-positive PageTail() */ | |
1448 | INIT_LIST_HEAD(&page->lru); | |
1449 | ||
d483da5b AD |
1450 | /* |
1451 | * no need for atomic set_bit because the struct | |
1452 | * page is not visible yet so nobody should | |
1453 | * access it yet. | |
1454 | */ | |
1455 | __SetPageReserved(page); | |
7e18adb4 MG |
1456 | } |
1457 | } | |
92923ca3 NZ |
1458 | } |
1459 | ||
ec95f53a KM |
1460 | static void __free_pages_ok(struct page *page, unsigned int order) |
1461 | { | |
d34b0733 | 1462 | unsigned long flags; |
95e34412 | 1463 | int migratetype; |
dc4b0caf | 1464 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1465 | |
e2769dbd | 1466 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1467 | return; |
1468 | ||
cfc47a28 | 1469 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1470 | local_irq_save(flags); |
1471 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1472 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1473 | local_irq_restore(flags); |
1da177e4 LT |
1474 | } |
1475 | ||
a9cd410a | 1476 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1477 | { |
c3993076 | 1478 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1479 | struct page *p = page; |
c3993076 | 1480 | unsigned int loop; |
a226f6c8 | 1481 | |
e2d0bd2b YL |
1482 | prefetchw(p); |
1483 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1484 | prefetchw(p + 1); | |
c3993076 JW |
1485 | __ClearPageReserved(p); |
1486 | set_page_count(p, 0); | |
a226f6c8 | 1487 | } |
e2d0bd2b YL |
1488 | __ClearPageReserved(p); |
1489 | set_page_count(p, 0); | |
c3993076 | 1490 | |
9705bea5 | 1491 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
c3993076 JW |
1492 | set_page_refcounted(page); |
1493 | __free_pages(page, order); | |
a226f6c8 DH |
1494 | } |
1495 | ||
3f08a302 | 1496 | #ifdef CONFIG_NEED_MULTIPLE_NODES |
7ace9917 | 1497 | |
75a592a4 MG |
1498 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1499 | ||
6f24fbd3 MR |
1500 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
1501 | ||
1502 | /* | |
1503 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
1504 | */ | |
1505 | int __meminit __early_pfn_to_nid(unsigned long pfn, | |
1506 | struct mminit_pfnnid_cache *state) | |
75a592a4 | 1507 | { |
6f24fbd3 | 1508 | unsigned long start_pfn, end_pfn; |
75a592a4 MG |
1509 | int nid; |
1510 | ||
6f24fbd3 MR |
1511 | if (state->last_start <= pfn && pfn < state->last_end) |
1512 | return state->last_nid; | |
1513 | ||
1514 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); | |
1515 | if (nid != NUMA_NO_NODE) { | |
1516 | state->last_start = start_pfn; | |
1517 | state->last_end = end_pfn; | |
1518 | state->last_nid = nid; | |
1519 | } | |
7ace9917 MG |
1520 | |
1521 | return nid; | |
75a592a4 | 1522 | } |
6f24fbd3 | 1523 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ |
75a592a4 | 1524 | |
75a592a4 | 1525 | int __meminit early_pfn_to_nid(unsigned long pfn) |
75a592a4 | 1526 | { |
7ace9917 | 1527 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1528 | int nid; |
1529 | ||
7ace9917 | 1530 | spin_lock(&early_pfn_lock); |
56ec43d8 | 1531 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1532 | if (nid < 0) |
e4568d38 | 1533 | nid = first_online_node; |
7ace9917 | 1534 | spin_unlock(&early_pfn_lock); |
75a592a4 | 1535 | |
7ace9917 | 1536 | return nid; |
75a592a4 | 1537 | } |
3f08a302 | 1538 | #endif /* CONFIG_NEED_MULTIPLE_NODES */ |
75a592a4 | 1539 | |
7c2ee349 | 1540 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1541 | unsigned int order) |
1542 | { | |
1543 | if (early_page_uninitialised(pfn)) | |
1544 | return; | |
a9cd410a | 1545 | __free_pages_core(page, order); |
3a80a7fa MG |
1546 | } |
1547 | ||
7cf91a98 JK |
1548 | /* |
1549 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1550 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1551 | * with the migration of free compaction scanner. The scanners then need to | |
1552 | * use only pfn_valid_within() check for arches that allow holes within | |
1553 | * pageblocks. | |
1554 | * | |
1555 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1556 | * | |
1557 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1558 | * i.e. it's possible that all pages within a zones range of pages do not | |
1559 | * belong to a single zone. We assume that a border between node0 and node1 | |
1560 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1561 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1562 | * the first and last page of a pageblock and avoid checking each individual | |
1563 | * page in a pageblock. | |
1564 | */ | |
1565 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1566 | unsigned long end_pfn, struct zone *zone) | |
1567 | { | |
1568 | struct page *start_page; | |
1569 | struct page *end_page; | |
1570 | ||
1571 | /* end_pfn is one past the range we are checking */ | |
1572 | end_pfn--; | |
1573 | ||
1574 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1575 | return NULL; | |
1576 | ||
2d070eab MH |
1577 | start_page = pfn_to_online_page(start_pfn); |
1578 | if (!start_page) | |
1579 | return NULL; | |
7cf91a98 JK |
1580 | |
1581 | if (page_zone(start_page) != zone) | |
1582 | return NULL; | |
1583 | ||
1584 | end_page = pfn_to_page(end_pfn); | |
1585 | ||
1586 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1587 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1588 | return NULL; | |
1589 | ||
1590 | return start_page; | |
1591 | } | |
1592 | ||
1593 | void set_zone_contiguous(struct zone *zone) | |
1594 | { | |
1595 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1596 | unsigned long block_end_pfn; | |
1597 | ||
1598 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1599 | for (; block_start_pfn < zone_end_pfn(zone); | |
1600 | block_start_pfn = block_end_pfn, | |
1601 | block_end_pfn += pageblock_nr_pages) { | |
1602 | ||
1603 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1604 | ||
1605 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1606 | block_end_pfn, zone)) | |
1607 | return; | |
e84fe99b | 1608 | cond_resched(); |
7cf91a98 JK |
1609 | } |
1610 | ||
1611 | /* We confirm that there is no hole */ | |
1612 | zone->contiguous = true; | |
1613 | } | |
1614 | ||
1615 | void clear_zone_contiguous(struct zone *zone) | |
1616 | { | |
1617 | zone->contiguous = false; | |
1618 | } | |
1619 | ||
7e18adb4 | 1620 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1621 | static void __init deferred_free_range(unsigned long pfn, |
1622 | unsigned long nr_pages) | |
a4de83dd | 1623 | { |
2f47a91f PT |
1624 | struct page *page; |
1625 | unsigned long i; | |
a4de83dd | 1626 | |
2f47a91f | 1627 | if (!nr_pages) |
a4de83dd MG |
1628 | return; |
1629 | ||
2f47a91f PT |
1630 | page = pfn_to_page(pfn); |
1631 | ||
a4de83dd | 1632 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1633 | if (nr_pages == pageblock_nr_pages && |
1634 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1635 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1636 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1637 | return; |
1638 | } | |
1639 | ||
e780149b XQ |
1640 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1641 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1642 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
a9cd410a | 1643 | __free_pages_core(page, 0); |
e780149b | 1644 | } |
a4de83dd MG |
1645 | } |
1646 | ||
d3cd131d NS |
1647 | /* Completion tracking for deferred_init_memmap() threads */ |
1648 | static atomic_t pgdat_init_n_undone __initdata; | |
1649 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1650 | ||
1651 | static inline void __init pgdat_init_report_one_done(void) | |
1652 | { | |
1653 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1654 | complete(&pgdat_init_all_done_comp); | |
1655 | } | |
0e1cc95b | 1656 | |
2f47a91f | 1657 | /* |
80b1f41c PT |
1658 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1659 | * | |
1660 | * First we check if pfn is valid on architectures where it is possible to have | |
1661 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1662 | * function is optimized out. | |
1663 | * | |
1664 | * Then, we check if a current large page is valid by only checking the validity | |
1665 | * of the head pfn. | |
2f47a91f | 1666 | */ |
56ec43d8 | 1667 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1668 | { |
80b1f41c PT |
1669 | if (!pfn_valid_within(pfn)) |
1670 | return false; | |
1671 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1672 | return false; | |
80b1f41c PT |
1673 | return true; |
1674 | } | |
2f47a91f | 1675 | |
80b1f41c PT |
1676 | /* |
1677 | * Free pages to buddy allocator. Try to free aligned pages in | |
1678 | * pageblock_nr_pages sizes. | |
1679 | */ | |
56ec43d8 | 1680 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1681 | unsigned long end_pfn) |
1682 | { | |
80b1f41c PT |
1683 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
1684 | unsigned long nr_free = 0; | |
2f47a91f | 1685 | |
80b1f41c | 1686 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1687 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1688 | deferred_free_range(pfn - nr_free, nr_free); |
1689 | nr_free = 0; | |
1690 | } else if (!(pfn & nr_pgmask)) { | |
1691 | deferred_free_range(pfn - nr_free, nr_free); | |
1692 | nr_free = 1; | |
80b1f41c PT |
1693 | } else { |
1694 | nr_free++; | |
1695 | } | |
1696 | } | |
1697 | /* Free the last block of pages to allocator */ | |
1698 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1699 | } |
1700 | ||
80b1f41c PT |
1701 | /* |
1702 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1703 | * by performing it only once every pageblock_nr_pages. | |
1704 | * Return number of pages initialized. | |
1705 | */ | |
56ec43d8 | 1706 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1707 | unsigned long pfn, |
1708 | unsigned long end_pfn) | |
2f47a91f | 1709 | { |
2f47a91f | 1710 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
56ec43d8 | 1711 | int nid = zone_to_nid(zone); |
2f47a91f | 1712 | unsigned long nr_pages = 0; |
56ec43d8 | 1713 | int zid = zone_idx(zone); |
2f47a91f | 1714 | struct page *page = NULL; |
2f47a91f | 1715 | |
80b1f41c | 1716 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1717 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1718 | page = NULL; |
2f47a91f | 1719 | continue; |
80b1f41c | 1720 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1721 | page = pfn_to_page(pfn); |
80b1f41c PT |
1722 | } else { |
1723 | page++; | |
2f47a91f | 1724 | } |
d0dc12e8 | 1725 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1726 | nr_pages++; |
2f47a91f | 1727 | } |
80b1f41c | 1728 | return (nr_pages); |
2f47a91f PT |
1729 | } |
1730 | ||
0e56acae AD |
1731 | /* |
1732 | * This function is meant to pre-load the iterator for the zone init. | |
1733 | * Specifically it walks through the ranges until we are caught up to the | |
1734 | * first_init_pfn value and exits there. If we never encounter the value we | |
1735 | * return false indicating there are no valid ranges left. | |
1736 | */ | |
1737 | static bool __init | |
1738 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
1739 | unsigned long *spfn, unsigned long *epfn, | |
1740 | unsigned long first_init_pfn) | |
1741 | { | |
1742 | u64 j; | |
1743 | ||
1744 | /* | |
1745 | * Start out by walking through the ranges in this zone that have | |
1746 | * already been initialized. We don't need to do anything with them | |
1747 | * so we just need to flush them out of the system. | |
1748 | */ | |
1749 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
1750 | if (*epfn <= first_init_pfn) | |
1751 | continue; | |
1752 | if (*spfn < first_init_pfn) | |
1753 | *spfn = first_init_pfn; | |
1754 | *i = j; | |
1755 | return true; | |
1756 | } | |
1757 | ||
1758 | return false; | |
1759 | } | |
1760 | ||
1761 | /* | |
1762 | * Initialize and free pages. We do it in two loops: first we initialize | |
1763 | * struct page, then free to buddy allocator, because while we are | |
1764 | * freeing pages we can access pages that are ahead (computing buddy | |
1765 | * page in __free_one_page()). | |
1766 | * | |
1767 | * In order to try and keep some memory in the cache we have the loop | |
1768 | * broken along max page order boundaries. This way we will not cause | |
1769 | * any issues with the buddy page computation. | |
1770 | */ | |
1771 | static unsigned long __init | |
1772 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
1773 | unsigned long *end_pfn) | |
1774 | { | |
1775 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
1776 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
1777 | unsigned long nr_pages = 0; | |
1778 | u64 j = *i; | |
1779 | ||
1780 | /* First we loop through and initialize the page values */ | |
1781 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
1782 | unsigned long t; | |
1783 | ||
1784 | if (mo_pfn <= *start_pfn) | |
1785 | break; | |
1786 | ||
1787 | t = min(mo_pfn, *end_pfn); | |
1788 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
1789 | ||
1790 | if (mo_pfn < *end_pfn) { | |
1791 | *start_pfn = mo_pfn; | |
1792 | break; | |
1793 | } | |
1794 | } | |
1795 | ||
1796 | /* Reset values and now loop through freeing pages as needed */ | |
1797 | swap(j, *i); | |
1798 | ||
1799 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
1800 | unsigned long t; | |
1801 | ||
1802 | if (mo_pfn <= spfn) | |
1803 | break; | |
1804 | ||
1805 | t = min(mo_pfn, epfn); | |
1806 | deferred_free_pages(spfn, t); | |
1807 | ||
1808 | if (mo_pfn <= epfn) | |
1809 | break; | |
1810 | } | |
1811 | ||
1812 | return nr_pages; | |
1813 | } | |
1814 | ||
e4443149 DJ |
1815 | static void __init |
1816 | deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, | |
1817 | void *arg) | |
1818 | { | |
1819 | unsigned long spfn, epfn; | |
1820 | struct zone *zone = arg; | |
1821 | u64 i; | |
1822 | ||
1823 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn); | |
1824 | ||
1825 | /* | |
1826 | * Initialize and free pages in MAX_ORDER sized increments so that we | |
1827 | * can avoid introducing any issues with the buddy allocator. | |
1828 | */ | |
1829 | while (spfn < end_pfn) { | |
1830 | deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
1831 | cond_resched(); | |
1832 | } | |
1833 | } | |
1834 | ||
ecd09650 DJ |
1835 | /* An arch may override for more concurrency. */ |
1836 | __weak int __init | |
1837 | deferred_page_init_max_threads(const struct cpumask *node_cpumask) | |
1838 | { | |
1839 | return 1; | |
1840 | } | |
1841 | ||
7e18adb4 | 1842 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1843 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1844 | { |
0e1cc95b | 1845 | pg_data_t *pgdat = data; |
0e56acae | 1846 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
89c7c402 | 1847 | unsigned long spfn = 0, epfn = 0; |
0e56acae | 1848 | unsigned long first_init_pfn, flags; |
7e18adb4 | 1849 | unsigned long start = jiffies; |
7e18adb4 | 1850 | struct zone *zone; |
e4443149 | 1851 | int zid, max_threads; |
2f47a91f | 1852 | u64 i; |
7e18adb4 | 1853 | |
3a2d7fa8 PT |
1854 | /* Bind memory initialisation thread to a local node if possible */ |
1855 | if (!cpumask_empty(cpumask)) | |
1856 | set_cpus_allowed_ptr(current, cpumask); | |
1857 | ||
1858 | pgdat_resize_lock(pgdat, &flags); | |
1859 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1860 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1861 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1862 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1863 | return 0; |
1864 | } | |
1865 | ||
7e18adb4 MG |
1866 | /* Sanity check boundaries */ |
1867 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1868 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1869 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1870 | ||
3d060856 PT |
1871 | /* |
1872 | * Once we unlock here, the zone cannot be grown anymore, thus if an | |
1873 | * interrupt thread must allocate this early in boot, zone must be | |
1874 | * pre-grown prior to start of deferred page initialization. | |
1875 | */ | |
1876 | pgdat_resize_unlock(pgdat, &flags); | |
1877 | ||
7e18adb4 MG |
1878 | /* Only the highest zone is deferred so find it */ |
1879 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1880 | zone = pgdat->node_zones + zid; | |
1881 | if (first_init_pfn < zone_end_pfn(zone)) | |
1882 | break; | |
1883 | } | |
0e56acae AD |
1884 | |
1885 | /* If the zone is empty somebody else may have cleared out the zone */ | |
1886 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1887 | first_init_pfn)) | |
1888 | goto zone_empty; | |
7e18adb4 | 1889 | |
ecd09650 | 1890 | max_threads = deferred_page_init_max_threads(cpumask); |
7e18adb4 | 1891 | |
117003c3 | 1892 | while (spfn < epfn) { |
e4443149 DJ |
1893 | unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION); |
1894 | struct padata_mt_job job = { | |
1895 | .thread_fn = deferred_init_memmap_chunk, | |
1896 | .fn_arg = zone, | |
1897 | .start = spfn, | |
1898 | .size = epfn_align - spfn, | |
1899 | .align = PAGES_PER_SECTION, | |
1900 | .min_chunk = PAGES_PER_SECTION, | |
1901 | .max_threads = max_threads, | |
1902 | }; | |
1903 | ||
1904 | padata_do_multithreaded(&job); | |
1905 | deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1906 | epfn_align); | |
117003c3 | 1907 | } |
0e56acae | 1908 | zone_empty: |
7e18adb4 MG |
1909 | /* Sanity check that the next zone really is unpopulated */ |
1910 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1911 | ||
89c7c402 DJ |
1912 | pr_info("node %d deferred pages initialised in %ums\n", |
1913 | pgdat->node_id, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
1914 | |
1915 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1916 | return 0; |
1917 | } | |
c9e97a19 | 1918 | |
c9e97a19 PT |
1919 | /* |
1920 | * If this zone has deferred pages, try to grow it by initializing enough | |
1921 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
1922 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
1923 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
1924 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
1925 | * | |
1926 | * Return true when zone was grown, otherwise return false. We return true even | |
1927 | * when we grow less than requested, to let the caller decide if there are | |
1928 | * enough pages to satisfy the allocation. | |
1929 | * | |
1930 | * Note: We use noinline because this function is needed only during boot, and | |
1931 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
1932 | * making sure that it is not inlined into permanent text section. | |
1933 | */ | |
1934 | static noinline bool __init | |
1935 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
1936 | { | |
c9e97a19 | 1937 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 1938 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 1939 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
1940 | unsigned long spfn, epfn, flags; |
1941 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
1942 | u64 i; |
1943 | ||
1944 | /* Only the last zone may have deferred pages */ | |
1945 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
1946 | return false; | |
1947 | ||
1948 | pgdat_resize_lock(pgdat, &flags); | |
1949 | ||
c9e97a19 PT |
1950 | /* |
1951 | * If someone grew this zone while we were waiting for spinlock, return | |
1952 | * true, as there might be enough pages already. | |
1953 | */ | |
1954 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
1955 | pgdat_resize_unlock(pgdat, &flags); | |
1956 | return true; | |
1957 | } | |
1958 | ||
0e56acae AD |
1959 | /* If the zone is empty somebody else may have cleared out the zone */ |
1960 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1961 | first_deferred_pfn)) { | |
1962 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 1963 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
1964 | /* Retry only once. */ |
1965 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
1966 | } |
1967 | ||
0e56acae AD |
1968 | /* |
1969 | * Initialize and free pages in MAX_ORDER sized increments so | |
1970 | * that we can avoid introducing any issues with the buddy | |
1971 | * allocator. | |
1972 | */ | |
1973 | while (spfn < epfn) { | |
1974 | /* update our first deferred PFN for this section */ | |
1975 | first_deferred_pfn = spfn; | |
1976 | ||
1977 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
117003c3 | 1978 | touch_nmi_watchdog(); |
c9e97a19 | 1979 | |
0e56acae AD |
1980 | /* We should only stop along section boundaries */ |
1981 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
1982 | continue; | |
c9e97a19 | 1983 | |
0e56acae | 1984 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
1985 | if (nr_pages >= nr_pages_needed) |
1986 | break; | |
1987 | } | |
1988 | ||
0e56acae | 1989 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
1990 | pgdat_resize_unlock(pgdat, &flags); |
1991 | ||
1992 | return nr_pages > 0; | |
1993 | } | |
1994 | ||
1995 | /* | |
1996 | * deferred_grow_zone() is __init, but it is called from | |
1997 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
1998 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
1999 | * and to ensure that the function body gets unloaded. | |
2000 | */ | |
2001 | static bool __ref | |
2002 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
2003 | { | |
2004 | return deferred_grow_zone(zone, order); | |
2005 | } | |
2006 | ||
7cf91a98 | 2007 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
2008 | |
2009 | void __init page_alloc_init_late(void) | |
2010 | { | |
7cf91a98 | 2011 | struct zone *zone; |
e900a918 | 2012 | int nid; |
7cf91a98 JK |
2013 | |
2014 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 2015 | |
d3cd131d NS |
2016 | /* There will be num_node_state(N_MEMORY) threads */ |
2017 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 2018 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
2019 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
2020 | } | |
2021 | ||
2022 | /* Block until all are initialised */ | |
d3cd131d | 2023 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 2024 | |
3e8fc007 MG |
2025 | /* |
2026 | * The number of managed pages has changed due to the initialisation | |
2027 | * so the pcpu batch and high limits needs to be updated or the limits | |
2028 | * will be artificially small. | |
2029 | */ | |
2030 | for_each_populated_zone(zone) | |
2031 | zone_pcp_update(zone); | |
2032 | ||
c9e97a19 PT |
2033 | /* |
2034 | * We initialized the rest of the deferred pages. Permanently disable | |
2035 | * on-demand struct page initialization. | |
2036 | */ | |
2037 | static_branch_disable(&deferred_pages); | |
2038 | ||
4248b0da MG |
2039 | /* Reinit limits that are based on free pages after the kernel is up */ |
2040 | files_maxfiles_init(); | |
7cf91a98 | 2041 | #endif |
350e88ba | 2042 | |
3010f876 PT |
2043 | /* Discard memblock private memory */ |
2044 | memblock_discard(); | |
7cf91a98 | 2045 | |
e900a918 DW |
2046 | for_each_node_state(nid, N_MEMORY) |
2047 | shuffle_free_memory(NODE_DATA(nid)); | |
2048 | ||
7cf91a98 JK |
2049 | for_each_populated_zone(zone) |
2050 | set_zone_contiguous(zone); | |
7e18adb4 | 2051 | } |
7e18adb4 | 2052 | |
47118af0 | 2053 | #ifdef CONFIG_CMA |
9cf510a5 | 2054 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
2055 | void __init init_cma_reserved_pageblock(struct page *page) |
2056 | { | |
2057 | unsigned i = pageblock_nr_pages; | |
2058 | struct page *p = page; | |
2059 | ||
2060 | do { | |
2061 | __ClearPageReserved(p); | |
2062 | set_page_count(p, 0); | |
d883c6cf | 2063 | } while (++p, --i); |
47118af0 | 2064 | |
47118af0 | 2065 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
2066 | |
2067 | if (pageblock_order >= MAX_ORDER) { | |
2068 | i = pageblock_nr_pages; | |
2069 | p = page; | |
2070 | do { | |
2071 | set_page_refcounted(p); | |
2072 | __free_pages(p, MAX_ORDER - 1); | |
2073 | p += MAX_ORDER_NR_PAGES; | |
2074 | } while (i -= MAX_ORDER_NR_PAGES); | |
2075 | } else { | |
2076 | set_page_refcounted(page); | |
2077 | __free_pages(page, pageblock_order); | |
2078 | } | |
2079 | ||
3dcc0571 | 2080 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
2081 | } |
2082 | #endif | |
1da177e4 LT |
2083 | |
2084 | /* | |
2085 | * The order of subdivision here is critical for the IO subsystem. | |
2086 | * Please do not alter this order without good reasons and regression | |
2087 | * testing. Specifically, as large blocks of memory are subdivided, | |
2088 | * the order in which smaller blocks are delivered depends on the order | |
2089 | * they're subdivided in this function. This is the primary factor | |
2090 | * influencing the order in which pages are delivered to the IO | |
2091 | * subsystem according to empirical testing, and this is also justified | |
2092 | * by considering the behavior of a buddy system containing a single | |
2093 | * large block of memory acted on by a series of small allocations. | |
2094 | * This behavior is a critical factor in sglist merging's success. | |
2095 | * | |
6d49e352 | 2096 | * -- nyc |
1da177e4 | 2097 | */ |
085cc7d5 | 2098 | static inline void expand(struct zone *zone, struct page *page, |
6ab01363 | 2099 | int low, int high, int migratetype) |
1da177e4 LT |
2100 | { |
2101 | unsigned long size = 1 << high; | |
2102 | ||
2103 | while (high > low) { | |
1da177e4 LT |
2104 | high--; |
2105 | size >>= 1; | |
309381fe | 2106 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2107 | |
acbc15a4 JK |
2108 | /* |
2109 | * Mark as guard pages (or page), that will allow to | |
2110 | * merge back to allocator when buddy will be freed. | |
2111 | * Corresponding page table entries will not be touched, | |
2112 | * pages will stay not present in virtual address space | |
2113 | */ | |
2114 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2115 | continue; |
acbc15a4 | 2116 | |
6ab01363 | 2117 | add_to_free_list(&page[size], zone, high, migratetype); |
1da177e4 LT |
2118 | set_page_order(&page[size], high); |
2119 | } | |
1da177e4 LT |
2120 | } |
2121 | ||
4e611801 | 2122 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2123 | { |
f4c18e6f | 2124 | if (unlikely(page->flags & __PG_HWPOISON)) { |
e570f56c NH |
2125 | /* Don't complain about hwpoisoned pages */ |
2126 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2127 | return; | |
f4c18e6f | 2128 | } |
58b7f119 WY |
2129 | |
2130 | bad_page(page, | |
2131 | page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP)); | |
4e611801 VB |
2132 | } |
2133 | ||
2134 | /* | |
2135 | * This page is about to be returned from the page allocator | |
2136 | */ | |
2137 | static inline int check_new_page(struct page *page) | |
2138 | { | |
2139 | if (likely(page_expected_state(page, | |
2140 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2141 | return 0; | |
2142 | ||
2143 | check_new_page_bad(page); | |
2144 | return 1; | |
2a7684a2 WF |
2145 | } |
2146 | ||
bd33ef36 | 2147 | static inline bool free_pages_prezeroed(void) |
1414c7f4 | 2148 | { |
6471384a AP |
2149 | return (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && |
2150 | page_poisoning_enabled()) || want_init_on_free(); | |
1414c7f4 LA |
2151 | } |
2152 | ||
479f854a | 2153 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2154 | /* |
2155 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2156 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2157 | * also checked when pcp lists are refilled from the free lists. | |
2158 | */ | |
2159 | static inline bool check_pcp_refill(struct page *page) | |
479f854a | 2160 | { |
8e57f8ac | 2161 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2162 | return check_new_page(page); |
2163 | else | |
2164 | return false; | |
479f854a MG |
2165 | } |
2166 | ||
4462b32c | 2167 | static inline bool check_new_pcp(struct page *page) |
479f854a MG |
2168 | { |
2169 | return check_new_page(page); | |
2170 | } | |
2171 | #else | |
4462b32c VB |
2172 | /* |
2173 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2174 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2175 | * enabled, they are also checked when being allocated from the pcp lists. | |
2176 | */ | |
2177 | static inline bool check_pcp_refill(struct page *page) | |
479f854a MG |
2178 | { |
2179 | return check_new_page(page); | |
2180 | } | |
4462b32c | 2181 | static inline bool check_new_pcp(struct page *page) |
479f854a | 2182 | { |
8e57f8ac | 2183 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2184 | return check_new_page(page); |
2185 | else | |
2186 | return false; | |
479f854a MG |
2187 | } |
2188 | #endif /* CONFIG_DEBUG_VM */ | |
2189 | ||
2190 | static bool check_new_pages(struct page *page, unsigned int order) | |
2191 | { | |
2192 | int i; | |
2193 | for (i = 0; i < (1 << order); i++) { | |
2194 | struct page *p = page + i; | |
2195 | ||
2196 | if (unlikely(check_new_page(p))) | |
2197 | return true; | |
2198 | } | |
2199 | ||
2200 | return false; | |
2201 | } | |
2202 | ||
46f24fd8 JK |
2203 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2204 | gfp_t gfp_flags) | |
2205 | { | |
2206 | set_page_private(page, 0); | |
2207 | set_page_refcounted(page); | |
2208 | ||
2209 | arch_alloc_page(page, order); | |
8e57f8ac | 2210 | if (debug_pagealloc_enabled_static()) |
d6332692 | 2211 | kernel_map_pages(page, 1 << order, 1); |
46f24fd8 | 2212 | kasan_alloc_pages(page, order); |
4117992d | 2213 | kernel_poison_pages(page, 1 << order, 1); |
46f24fd8 JK |
2214 | set_page_owner(page, order, gfp_flags); |
2215 | } | |
2216 | ||
479f854a | 2217 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2218 | unsigned int alloc_flags) |
2a7684a2 | 2219 | { |
46f24fd8 | 2220 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2221 | |
6471384a AP |
2222 | if (!free_pages_prezeroed() && want_init_on_alloc(gfp_flags)) |
2223 | kernel_init_free_pages(page, 1 << order); | |
17cf4406 NP |
2224 | |
2225 | if (order && (gfp_flags & __GFP_COMP)) | |
2226 | prep_compound_page(page, order); | |
2227 | ||
75379191 | 2228 | /* |
2f064f34 | 2229 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2230 | * allocate the page. The expectation is that the caller is taking |
2231 | * steps that will free more memory. The caller should avoid the page | |
2232 | * being used for !PFMEMALLOC purposes. | |
2233 | */ | |
2f064f34 MH |
2234 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2235 | set_page_pfmemalloc(page); | |
2236 | else | |
2237 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2238 | } |
2239 | ||
56fd56b8 MG |
2240 | /* |
2241 | * Go through the free lists for the given migratetype and remove | |
2242 | * the smallest available page from the freelists | |
2243 | */ | |
85ccc8fa | 2244 | static __always_inline |
728ec980 | 2245 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2246 | int migratetype) |
2247 | { | |
2248 | unsigned int current_order; | |
b8af2941 | 2249 | struct free_area *area; |
56fd56b8 MG |
2250 | struct page *page; |
2251 | ||
2252 | /* Find a page of the appropriate size in the preferred list */ | |
2253 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2254 | area = &(zone->free_area[current_order]); | |
b03641af | 2255 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2256 | if (!page) |
2257 | continue; | |
6ab01363 AD |
2258 | del_page_from_free_list(page, zone, current_order); |
2259 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 2260 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
2261 | return page; |
2262 | } | |
2263 | ||
2264 | return NULL; | |
2265 | } | |
2266 | ||
2267 | ||
b2a0ac88 MG |
2268 | /* |
2269 | * This array describes the order lists are fallen back to when | |
2270 | * the free lists for the desirable migrate type are depleted | |
2271 | */ | |
da415663 | 2272 | static int fallbacks[MIGRATE_TYPES][3] = { |
974a786e | 2273 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2274 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2275 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
47118af0 | 2276 | #ifdef CONFIG_CMA |
974a786e | 2277 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 2278 | #endif |
194159fb | 2279 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 2280 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 2281 | #endif |
b2a0ac88 MG |
2282 | }; |
2283 | ||
dc67647b | 2284 | #ifdef CONFIG_CMA |
85ccc8fa | 2285 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2286 | unsigned int order) |
2287 | { | |
2288 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2289 | } | |
2290 | #else | |
2291 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2292 | unsigned int order) { return NULL; } | |
2293 | #endif | |
2294 | ||
c361be55 MG |
2295 | /* |
2296 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 2297 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2298 | * boundary. If alignment is required, use move_freepages_block() |
2299 | */ | |
02aa0cdd | 2300 | static int move_freepages(struct zone *zone, |
b69a7288 | 2301 | struct page *start_page, struct page *end_page, |
02aa0cdd | 2302 | int migratetype, int *num_movable) |
c361be55 MG |
2303 | { |
2304 | struct page *page; | |
d00181b9 | 2305 | unsigned int order; |
d100313f | 2306 | int pages_moved = 0; |
c361be55 | 2307 | |
c361be55 MG |
2308 | for (page = start_page; page <= end_page;) { |
2309 | if (!pfn_valid_within(page_to_pfn(page))) { | |
2310 | page++; | |
2311 | continue; | |
2312 | } | |
2313 | ||
2314 | if (!PageBuddy(page)) { | |
02aa0cdd VB |
2315 | /* |
2316 | * We assume that pages that could be isolated for | |
2317 | * migration are movable. But we don't actually try | |
2318 | * isolating, as that would be expensive. | |
2319 | */ | |
2320 | if (num_movable && | |
2321 | (PageLRU(page) || __PageMovable(page))) | |
2322 | (*num_movable)++; | |
2323 | ||
c361be55 MG |
2324 | page++; |
2325 | continue; | |
2326 | } | |
2327 | ||
cd961038 DR |
2328 | /* Make sure we are not inadvertently changing nodes */ |
2329 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2330 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2331 | ||
c361be55 | 2332 | order = page_order(page); |
6ab01363 | 2333 | move_to_free_list(page, zone, order, migratetype); |
c361be55 | 2334 | page += 1 << order; |
d100313f | 2335 | pages_moved += 1 << order; |
c361be55 MG |
2336 | } |
2337 | ||
d100313f | 2338 | return pages_moved; |
c361be55 MG |
2339 | } |
2340 | ||
ee6f509c | 2341 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2342 | int migratetype, int *num_movable) |
c361be55 MG |
2343 | { |
2344 | unsigned long start_pfn, end_pfn; | |
2345 | struct page *start_page, *end_page; | |
2346 | ||
4a222127 DR |
2347 | if (num_movable) |
2348 | *num_movable = 0; | |
2349 | ||
c361be55 | 2350 | start_pfn = page_to_pfn(page); |
d9c23400 | 2351 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 2352 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
2353 | end_page = start_page + pageblock_nr_pages - 1; |
2354 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
2355 | |
2356 | /* Do not cross zone boundaries */ | |
108bcc96 | 2357 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 2358 | start_page = page; |
108bcc96 | 2359 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2360 | return 0; |
2361 | ||
02aa0cdd VB |
2362 | return move_freepages(zone, start_page, end_page, migratetype, |
2363 | num_movable); | |
c361be55 MG |
2364 | } |
2365 | ||
2f66a68f MG |
2366 | static void change_pageblock_range(struct page *pageblock_page, |
2367 | int start_order, int migratetype) | |
2368 | { | |
2369 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2370 | ||
2371 | while (nr_pageblocks--) { | |
2372 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2373 | pageblock_page += pageblock_nr_pages; | |
2374 | } | |
2375 | } | |
2376 | ||
fef903ef | 2377 | /* |
9c0415eb VB |
2378 | * When we are falling back to another migratetype during allocation, try to |
2379 | * steal extra free pages from the same pageblocks to satisfy further | |
2380 | * allocations, instead of polluting multiple pageblocks. | |
2381 | * | |
2382 | * If we are stealing a relatively large buddy page, it is likely there will | |
2383 | * be more free pages in the pageblock, so try to steal them all. For | |
2384 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2385 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2386 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2387 | * pageblocks. | |
fef903ef | 2388 | */ |
4eb7dce6 JK |
2389 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2390 | { | |
2391 | /* | |
2392 | * Leaving this order check is intended, although there is | |
2393 | * relaxed order check in next check. The reason is that | |
2394 | * we can actually steal whole pageblock if this condition met, | |
2395 | * but, below check doesn't guarantee it and that is just heuristic | |
2396 | * so could be changed anytime. | |
2397 | */ | |
2398 | if (order >= pageblock_order) | |
2399 | return true; | |
2400 | ||
2401 | if (order >= pageblock_order / 2 || | |
2402 | start_mt == MIGRATE_RECLAIMABLE || | |
2403 | start_mt == MIGRATE_UNMOVABLE || | |
2404 | page_group_by_mobility_disabled) | |
2405 | return true; | |
2406 | ||
2407 | return false; | |
2408 | } | |
2409 | ||
1c30844d MG |
2410 | static inline void boost_watermark(struct zone *zone) |
2411 | { | |
2412 | unsigned long max_boost; | |
2413 | ||
2414 | if (!watermark_boost_factor) | |
2415 | return; | |
14f69140 HW |
2416 | /* |
2417 | * Don't bother in zones that are unlikely to produce results. | |
2418 | * On small machines, including kdump capture kernels running | |
2419 | * in a small area, boosting the watermark can cause an out of | |
2420 | * memory situation immediately. | |
2421 | */ | |
2422 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
2423 | return; | |
1c30844d MG |
2424 | |
2425 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2426 | watermark_boost_factor, 10000); | |
94b3334c MG |
2427 | |
2428 | /* | |
2429 | * high watermark may be uninitialised if fragmentation occurs | |
2430 | * very early in boot so do not boost. We do not fall | |
2431 | * through and boost by pageblock_nr_pages as failing | |
2432 | * allocations that early means that reclaim is not going | |
2433 | * to help and it may even be impossible to reclaim the | |
2434 | * boosted watermark resulting in a hang. | |
2435 | */ | |
2436 | if (!max_boost) | |
2437 | return; | |
2438 | ||
1c30844d MG |
2439 | max_boost = max(pageblock_nr_pages, max_boost); |
2440 | ||
2441 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2442 | max_boost); | |
2443 | } | |
2444 | ||
4eb7dce6 JK |
2445 | /* |
2446 | * This function implements actual steal behaviour. If order is large enough, | |
2447 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2448 | * pageblock to our migratetype and determine how many already-allocated pages |
2449 | * are there in the pageblock with a compatible migratetype. If at least half | |
2450 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2451 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2452 | */ |
2453 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2454 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2455 | { |
d00181b9 | 2456 | unsigned int current_order = page_order(page); |
02aa0cdd VB |
2457 | int free_pages, movable_pages, alike_pages; |
2458 | int old_block_type; | |
2459 | ||
2460 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2461 | |
3bc48f96 VB |
2462 | /* |
2463 | * This can happen due to races and we want to prevent broken | |
2464 | * highatomic accounting. | |
2465 | */ | |
02aa0cdd | 2466 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2467 | goto single_page; |
2468 | ||
fef903ef SB |
2469 | /* Take ownership for orders >= pageblock_order */ |
2470 | if (current_order >= pageblock_order) { | |
2471 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2472 | goto single_page; |
fef903ef SB |
2473 | } |
2474 | ||
1c30844d MG |
2475 | /* |
2476 | * Boost watermarks to increase reclaim pressure to reduce the | |
2477 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2478 | * may be balanced overall and kswapd will not wake naturally. | |
2479 | */ | |
2480 | boost_watermark(zone); | |
2481 | if (alloc_flags & ALLOC_KSWAPD) | |
73444bc4 | 2482 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2483 | |
3bc48f96 VB |
2484 | /* We are not allowed to try stealing from the whole block */ |
2485 | if (!whole_block) | |
2486 | goto single_page; | |
2487 | ||
02aa0cdd VB |
2488 | free_pages = move_freepages_block(zone, page, start_type, |
2489 | &movable_pages); | |
2490 | /* | |
2491 | * Determine how many pages are compatible with our allocation. | |
2492 | * For movable allocation, it's the number of movable pages which | |
2493 | * we just obtained. For other types it's a bit more tricky. | |
2494 | */ | |
2495 | if (start_type == MIGRATE_MOVABLE) { | |
2496 | alike_pages = movable_pages; | |
2497 | } else { | |
2498 | /* | |
2499 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2500 | * to MOVABLE pageblock, consider all non-movable pages as | |
2501 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2502 | * vice versa, be conservative since we can't distinguish the | |
2503 | * exact migratetype of non-movable pages. | |
2504 | */ | |
2505 | if (old_block_type == MIGRATE_MOVABLE) | |
2506 | alike_pages = pageblock_nr_pages | |
2507 | - (free_pages + movable_pages); | |
2508 | else | |
2509 | alike_pages = 0; | |
2510 | } | |
2511 | ||
3bc48f96 | 2512 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2513 | if (!free_pages) |
3bc48f96 | 2514 | goto single_page; |
fef903ef | 2515 | |
02aa0cdd VB |
2516 | /* |
2517 | * If a sufficient number of pages in the block are either free or of | |
2518 | * comparable migratability as our allocation, claim the whole block. | |
2519 | */ | |
2520 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2521 | page_group_by_mobility_disabled) |
2522 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2523 | |
2524 | return; | |
2525 | ||
2526 | single_page: | |
6ab01363 | 2527 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
2528 | } |
2529 | ||
2149cdae JK |
2530 | /* |
2531 | * Check whether there is a suitable fallback freepage with requested order. | |
2532 | * If only_stealable is true, this function returns fallback_mt only if | |
2533 | * we can steal other freepages all together. This would help to reduce | |
2534 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2535 | */ | |
2536 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2537 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2538 | { |
2539 | int i; | |
2540 | int fallback_mt; | |
2541 | ||
2542 | if (area->nr_free == 0) | |
2543 | return -1; | |
2544 | ||
2545 | *can_steal = false; | |
2546 | for (i = 0;; i++) { | |
2547 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2548 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2549 | break; |
2550 | ||
b03641af | 2551 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2552 | continue; |
fef903ef | 2553 | |
4eb7dce6 JK |
2554 | if (can_steal_fallback(order, migratetype)) |
2555 | *can_steal = true; | |
2556 | ||
2149cdae JK |
2557 | if (!only_stealable) |
2558 | return fallback_mt; | |
2559 | ||
2560 | if (*can_steal) | |
2561 | return fallback_mt; | |
fef903ef | 2562 | } |
4eb7dce6 JK |
2563 | |
2564 | return -1; | |
fef903ef SB |
2565 | } |
2566 | ||
0aaa29a5 MG |
2567 | /* |
2568 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2569 | * there are no empty page blocks that contain a page with a suitable order | |
2570 | */ | |
2571 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2572 | unsigned int alloc_order) | |
2573 | { | |
2574 | int mt; | |
2575 | unsigned long max_managed, flags; | |
2576 | ||
2577 | /* | |
2578 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2579 | * Check is race-prone but harmless. | |
2580 | */ | |
9705bea5 | 2581 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2582 | if (zone->nr_reserved_highatomic >= max_managed) |
2583 | return; | |
2584 | ||
2585 | spin_lock_irqsave(&zone->lock, flags); | |
2586 | ||
2587 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2588 | if (zone->nr_reserved_highatomic >= max_managed) | |
2589 | goto out_unlock; | |
2590 | ||
2591 | /* Yoink! */ | |
2592 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2593 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2594 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2595 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2596 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2597 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2598 | } |
2599 | ||
2600 | out_unlock: | |
2601 | spin_unlock_irqrestore(&zone->lock, flags); | |
2602 | } | |
2603 | ||
2604 | /* | |
2605 | * Used when an allocation is about to fail under memory pressure. This | |
2606 | * potentially hurts the reliability of high-order allocations when under | |
2607 | * intense memory pressure but failed atomic allocations should be easier | |
2608 | * to recover from than an OOM. | |
29fac03b MK |
2609 | * |
2610 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2611 | * pageblock is exhausted. | |
0aaa29a5 | 2612 | */ |
29fac03b MK |
2613 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2614 | bool force) | |
0aaa29a5 MG |
2615 | { |
2616 | struct zonelist *zonelist = ac->zonelist; | |
2617 | unsigned long flags; | |
2618 | struct zoneref *z; | |
2619 | struct zone *zone; | |
2620 | struct page *page; | |
2621 | int order; | |
04c8716f | 2622 | bool ret; |
0aaa29a5 | 2623 | |
97a225e6 | 2624 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, |
0aaa29a5 | 2625 | ac->nodemask) { |
29fac03b MK |
2626 | /* |
2627 | * Preserve at least one pageblock unless memory pressure | |
2628 | * is really high. | |
2629 | */ | |
2630 | if (!force && zone->nr_reserved_highatomic <= | |
2631 | pageblock_nr_pages) | |
0aaa29a5 MG |
2632 | continue; |
2633 | ||
2634 | spin_lock_irqsave(&zone->lock, flags); | |
2635 | for (order = 0; order < MAX_ORDER; order++) { | |
2636 | struct free_area *area = &(zone->free_area[order]); | |
2637 | ||
b03641af | 2638 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2639 | if (!page) |
0aaa29a5 MG |
2640 | continue; |
2641 | ||
0aaa29a5 | 2642 | /* |
4855e4a7 MK |
2643 | * In page freeing path, migratetype change is racy so |
2644 | * we can counter several free pages in a pageblock | |
2645 | * in this loop althoug we changed the pageblock type | |
2646 | * from highatomic to ac->migratetype. So we should | |
2647 | * adjust the count once. | |
0aaa29a5 | 2648 | */ |
a6ffdc07 | 2649 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2650 | /* |
2651 | * It should never happen but changes to | |
2652 | * locking could inadvertently allow a per-cpu | |
2653 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2654 | * while unreserving so be safe and watch for | |
2655 | * underflows. | |
2656 | */ | |
2657 | zone->nr_reserved_highatomic -= min( | |
2658 | pageblock_nr_pages, | |
2659 | zone->nr_reserved_highatomic); | |
2660 | } | |
0aaa29a5 MG |
2661 | |
2662 | /* | |
2663 | * Convert to ac->migratetype and avoid the normal | |
2664 | * pageblock stealing heuristics. Minimally, the caller | |
2665 | * is doing the work and needs the pages. More | |
2666 | * importantly, if the block was always converted to | |
2667 | * MIGRATE_UNMOVABLE or another type then the number | |
2668 | * of pageblocks that cannot be completely freed | |
2669 | * may increase. | |
2670 | */ | |
2671 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2672 | ret = move_freepages_block(zone, page, ac->migratetype, |
2673 | NULL); | |
29fac03b MK |
2674 | if (ret) { |
2675 | spin_unlock_irqrestore(&zone->lock, flags); | |
2676 | return ret; | |
2677 | } | |
0aaa29a5 MG |
2678 | } |
2679 | spin_unlock_irqrestore(&zone->lock, flags); | |
2680 | } | |
04c8716f MK |
2681 | |
2682 | return false; | |
0aaa29a5 MG |
2683 | } |
2684 | ||
3bc48f96 VB |
2685 | /* |
2686 | * Try finding a free buddy page on the fallback list and put it on the free | |
2687 | * list of requested migratetype, possibly along with other pages from the same | |
2688 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2689 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2690 | * |
2691 | * The use of signed ints for order and current_order is a deliberate | |
2692 | * deviation from the rest of this file, to make the for loop | |
2693 | * condition simpler. | |
3bc48f96 | 2694 | */ |
85ccc8fa | 2695 | static __always_inline bool |
6bb15450 MG |
2696 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2697 | unsigned int alloc_flags) | |
b2a0ac88 | 2698 | { |
b8af2941 | 2699 | struct free_area *area; |
b002529d | 2700 | int current_order; |
6bb15450 | 2701 | int min_order = order; |
b2a0ac88 | 2702 | struct page *page; |
4eb7dce6 JK |
2703 | int fallback_mt; |
2704 | bool can_steal; | |
b2a0ac88 | 2705 | |
6bb15450 MG |
2706 | /* |
2707 | * Do not steal pages from freelists belonging to other pageblocks | |
2708 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2709 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2710 | */ | |
2711 | if (alloc_flags & ALLOC_NOFRAGMENT) | |
2712 | min_order = pageblock_order; | |
2713 | ||
7a8f58f3 VB |
2714 | /* |
2715 | * Find the largest available free page in the other list. This roughly | |
2716 | * approximates finding the pageblock with the most free pages, which | |
2717 | * would be too costly to do exactly. | |
2718 | */ | |
6bb15450 | 2719 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 2720 | --current_order) { |
4eb7dce6 JK |
2721 | area = &(zone->free_area[current_order]); |
2722 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2723 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2724 | if (fallback_mt == -1) |
2725 | continue; | |
b2a0ac88 | 2726 | |
7a8f58f3 VB |
2727 | /* |
2728 | * We cannot steal all free pages from the pageblock and the | |
2729 | * requested migratetype is movable. In that case it's better to | |
2730 | * steal and split the smallest available page instead of the | |
2731 | * largest available page, because even if the next movable | |
2732 | * allocation falls back into a different pageblock than this | |
2733 | * one, it won't cause permanent fragmentation. | |
2734 | */ | |
2735 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2736 | && current_order > order) | |
2737 | goto find_smallest; | |
b2a0ac88 | 2738 | |
7a8f58f3 VB |
2739 | goto do_steal; |
2740 | } | |
e0fff1bd | 2741 | |
7a8f58f3 | 2742 | return false; |
e0fff1bd | 2743 | |
7a8f58f3 VB |
2744 | find_smallest: |
2745 | for (current_order = order; current_order < MAX_ORDER; | |
2746 | current_order++) { | |
2747 | area = &(zone->free_area[current_order]); | |
2748 | fallback_mt = find_suitable_fallback(area, current_order, | |
2749 | start_migratetype, false, &can_steal); | |
2750 | if (fallback_mt != -1) | |
2751 | break; | |
b2a0ac88 MG |
2752 | } |
2753 | ||
7a8f58f3 VB |
2754 | /* |
2755 | * This should not happen - we already found a suitable fallback | |
2756 | * when looking for the largest page. | |
2757 | */ | |
2758 | VM_BUG_ON(current_order == MAX_ORDER); | |
2759 | ||
2760 | do_steal: | |
b03641af | 2761 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2762 | |
1c30844d MG |
2763 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
2764 | can_steal); | |
7a8f58f3 VB |
2765 | |
2766 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2767 | start_migratetype, fallback_mt); | |
2768 | ||
2769 | return true; | |
2770 | ||
b2a0ac88 MG |
2771 | } |
2772 | ||
56fd56b8 | 2773 | /* |
1da177e4 LT |
2774 | * Do the hard work of removing an element from the buddy allocator. |
2775 | * Call me with the zone->lock already held. | |
2776 | */ | |
85ccc8fa | 2777 | static __always_inline struct page * |
6bb15450 MG |
2778 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2779 | unsigned int alloc_flags) | |
1da177e4 | 2780 | { |
1da177e4 LT |
2781 | struct page *page; |
2782 | ||
16867664 RG |
2783 | #ifdef CONFIG_CMA |
2784 | /* | |
2785 | * Balance movable allocations between regular and CMA areas by | |
2786 | * allocating from CMA when over half of the zone's free memory | |
2787 | * is in the CMA area. | |
2788 | */ | |
8510e69c | 2789 | if (alloc_flags & ALLOC_CMA && |
16867664 RG |
2790 | zone_page_state(zone, NR_FREE_CMA_PAGES) > |
2791 | zone_page_state(zone, NR_FREE_PAGES) / 2) { | |
2792 | page = __rmqueue_cma_fallback(zone, order); | |
2793 | if (page) | |
2794 | return page; | |
2795 | } | |
2796 | #endif | |
3bc48f96 | 2797 | retry: |
56fd56b8 | 2798 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2799 | if (unlikely(!page)) { |
8510e69c | 2800 | if (alloc_flags & ALLOC_CMA) |
dc67647b JK |
2801 | page = __rmqueue_cma_fallback(zone, order); |
2802 | ||
6bb15450 MG |
2803 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2804 | alloc_flags)) | |
3bc48f96 | 2805 | goto retry; |
728ec980 MG |
2806 | } |
2807 | ||
0d3d062a | 2808 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2809 | return page; |
1da177e4 LT |
2810 | } |
2811 | ||
5f63b720 | 2812 | /* |
1da177e4 LT |
2813 | * Obtain a specified number of elements from the buddy allocator, all under |
2814 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2815 | * Returns the number of new pages which were placed at *list. | |
2816 | */ | |
5f63b720 | 2817 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2818 | unsigned long count, struct list_head *list, |
6bb15450 | 2819 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2820 | { |
a6de734b | 2821 | int i, alloced = 0; |
5f63b720 | 2822 | |
d34b0733 | 2823 | spin_lock(&zone->lock); |
1da177e4 | 2824 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2825 | struct page *page = __rmqueue(zone, order, migratetype, |
2826 | alloc_flags); | |
085cc7d5 | 2827 | if (unlikely(page == NULL)) |
1da177e4 | 2828 | break; |
81eabcbe | 2829 | |
479f854a MG |
2830 | if (unlikely(check_pcp_refill(page))) |
2831 | continue; | |
2832 | ||
81eabcbe | 2833 | /* |
0fac3ba5 VB |
2834 | * Split buddy pages returned by expand() are received here in |
2835 | * physical page order. The page is added to the tail of | |
2836 | * caller's list. From the callers perspective, the linked list | |
2837 | * is ordered by page number under some conditions. This is | |
2838 | * useful for IO devices that can forward direction from the | |
2839 | * head, thus also in the physical page order. This is useful | |
2840 | * for IO devices that can merge IO requests if the physical | |
2841 | * pages are ordered properly. | |
81eabcbe | 2842 | */ |
0fac3ba5 | 2843 | list_add_tail(&page->lru, list); |
a6de734b | 2844 | alloced++; |
bb14c2c7 | 2845 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2846 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2847 | -(1 << order)); | |
1da177e4 | 2848 | } |
a6de734b MG |
2849 | |
2850 | /* | |
2851 | * i pages were removed from the buddy list even if some leak due | |
2852 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2853 | * on i. Do not confuse with 'alloced' which is the number of | |
2854 | * pages added to the pcp list. | |
2855 | */ | |
f2260e6b | 2856 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2857 | spin_unlock(&zone->lock); |
a6de734b | 2858 | return alloced; |
1da177e4 LT |
2859 | } |
2860 | ||
4ae7c039 | 2861 | #ifdef CONFIG_NUMA |
8fce4d8e | 2862 | /* |
4037d452 CL |
2863 | * Called from the vmstat counter updater to drain pagesets of this |
2864 | * currently executing processor on remote nodes after they have | |
2865 | * expired. | |
2866 | * | |
879336c3 CL |
2867 | * Note that this function must be called with the thread pinned to |
2868 | * a single processor. | |
8fce4d8e | 2869 | */ |
4037d452 | 2870 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2871 | { |
4ae7c039 | 2872 | unsigned long flags; |
7be12fc9 | 2873 | int to_drain, batch; |
4ae7c039 | 2874 | |
4037d452 | 2875 | local_irq_save(flags); |
4db0c3c2 | 2876 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2877 | to_drain = min(pcp->count, batch); |
77ba9062 | 2878 | if (to_drain > 0) |
2a13515c | 2879 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 2880 | local_irq_restore(flags); |
4ae7c039 CL |
2881 | } |
2882 | #endif | |
2883 | ||
9f8f2172 | 2884 | /* |
93481ff0 | 2885 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2886 | * |
2887 | * The processor must either be the current processor and the | |
2888 | * thread pinned to the current processor or a processor that | |
2889 | * is not online. | |
2890 | */ | |
93481ff0 | 2891 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2892 | { |
c54ad30c | 2893 | unsigned long flags; |
93481ff0 VB |
2894 | struct per_cpu_pageset *pset; |
2895 | struct per_cpu_pages *pcp; | |
1da177e4 | 2896 | |
93481ff0 VB |
2897 | local_irq_save(flags); |
2898 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2899 | |
93481ff0 | 2900 | pcp = &pset->pcp; |
77ba9062 | 2901 | if (pcp->count) |
93481ff0 | 2902 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
2903 | local_irq_restore(flags); |
2904 | } | |
3dfa5721 | 2905 | |
93481ff0 VB |
2906 | /* |
2907 | * Drain pcplists of all zones on the indicated processor. | |
2908 | * | |
2909 | * The processor must either be the current processor and the | |
2910 | * thread pinned to the current processor or a processor that | |
2911 | * is not online. | |
2912 | */ | |
2913 | static void drain_pages(unsigned int cpu) | |
2914 | { | |
2915 | struct zone *zone; | |
2916 | ||
2917 | for_each_populated_zone(zone) { | |
2918 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2919 | } |
2920 | } | |
1da177e4 | 2921 | |
9f8f2172 CL |
2922 | /* |
2923 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2924 | * |
2925 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2926 | * the single zone's pages. | |
9f8f2172 | 2927 | */ |
93481ff0 | 2928 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2929 | { |
93481ff0 VB |
2930 | int cpu = smp_processor_id(); |
2931 | ||
2932 | if (zone) | |
2933 | drain_pages_zone(cpu, zone); | |
2934 | else | |
2935 | drain_pages(cpu); | |
9f8f2172 CL |
2936 | } |
2937 | ||
0ccce3b9 MG |
2938 | static void drain_local_pages_wq(struct work_struct *work) |
2939 | { | |
d9367bd0 WY |
2940 | struct pcpu_drain *drain; |
2941 | ||
2942 | drain = container_of(work, struct pcpu_drain, work); | |
2943 | ||
a459eeb7 MH |
2944 | /* |
2945 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2946 | * we can race with cpu offline when the WQ can move this from | |
2947 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2948 | * cpu which is allright but we also have to make sure to not move to | |
2949 | * a different one. | |
2950 | */ | |
2951 | preempt_disable(); | |
d9367bd0 | 2952 | drain_local_pages(drain->zone); |
a459eeb7 | 2953 | preempt_enable(); |
0ccce3b9 MG |
2954 | } |
2955 | ||
9f8f2172 | 2956 | /* |
74046494 GBY |
2957 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2958 | * | |
93481ff0 VB |
2959 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2960 | * | |
0ccce3b9 | 2961 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2962 | */ |
93481ff0 | 2963 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2964 | { |
74046494 | 2965 | int cpu; |
74046494 GBY |
2966 | |
2967 | /* | |
2968 | * Allocate in the BSS so we wont require allocation in | |
2969 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2970 | */ | |
2971 | static cpumask_t cpus_with_pcps; | |
2972 | ||
ce612879 MH |
2973 | /* |
2974 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2975 | * initialized. | |
2976 | */ | |
2977 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2978 | return; | |
2979 | ||
bd233f53 MG |
2980 | /* |
2981 | * Do not drain if one is already in progress unless it's specific to | |
2982 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2983 | * the drain to be complete when the call returns. | |
2984 | */ | |
2985 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2986 | if (!zone) | |
2987 | return; | |
2988 | mutex_lock(&pcpu_drain_mutex); | |
2989 | } | |
0ccce3b9 | 2990 | |
74046494 GBY |
2991 | /* |
2992 | * We don't care about racing with CPU hotplug event | |
2993 | * as offline notification will cause the notified | |
2994 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2995 | * disables preemption as part of its processing | |
2996 | */ | |
2997 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2998 | struct per_cpu_pageset *pcp; |
2999 | struct zone *z; | |
74046494 | 3000 | bool has_pcps = false; |
93481ff0 VB |
3001 | |
3002 | if (zone) { | |
74046494 | 3003 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 3004 | if (pcp->pcp.count) |
74046494 | 3005 | has_pcps = true; |
93481ff0 VB |
3006 | } else { |
3007 | for_each_populated_zone(z) { | |
3008 | pcp = per_cpu_ptr(z->pageset, cpu); | |
3009 | if (pcp->pcp.count) { | |
3010 | has_pcps = true; | |
3011 | break; | |
3012 | } | |
74046494 GBY |
3013 | } |
3014 | } | |
93481ff0 | 3015 | |
74046494 GBY |
3016 | if (has_pcps) |
3017 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
3018 | else | |
3019 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
3020 | } | |
0ccce3b9 | 3021 | |
bd233f53 | 3022 | for_each_cpu(cpu, &cpus_with_pcps) { |
d9367bd0 WY |
3023 | struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu); |
3024 | ||
3025 | drain->zone = zone; | |
3026 | INIT_WORK(&drain->work, drain_local_pages_wq); | |
3027 | queue_work_on(cpu, mm_percpu_wq, &drain->work); | |
0ccce3b9 | 3028 | } |
bd233f53 | 3029 | for_each_cpu(cpu, &cpus_with_pcps) |
d9367bd0 | 3030 | flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work); |
bd233f53 MG |
3031 | |
3032 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
3033 | } |
3034 | ||
296699de | 3035 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3036 | |
556b969a CY |
3037 | /* |
3038 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
3039 | */ | |
3040 | #define WD_PAGE_COUNT (128*1024) | |
3041 | ||
1da177e4 LT |
3042 | void mark_free_pages(struct zone *zone) |
3043 | { | |
556b969a | 3044 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 3045 | unsigned long flags; |
7aeb09f9 | 3046 | unsigned int order, t; |
86760a2c | 3047 | struct page *page; |
1da177e4 | 3048 | |
8080fc03 | 3049 | if (zone_is_empty(zone)) |
1da177e4 LT |
3050 | return; |
3051 | ||
3052 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 3053 | |
108bcc96 | 3054 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
3055 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
3056 | if (pfn_valid(pfn)) { | |
86760a2c | 3057 | page = pfn_to_page(pfn); |
ba6b0979 | 3058 | |
556b969a CY |
3059 | if (!--page_count) { |
3060 | touch_nmi_watchdog(); | |
3061 | page_count = WD_PAGE_COUNT; | |
3062 | } | |
3063 | ||
ba6b0979 JK |
3064 | if (page_zone(page) != zone) |
3065 | continue; | |
3066 | ||
7be98234 RW |
3067 | if (!swsusp_page_is_forbidden(page)) |
3068 | swsusp_unset_page_free(page); | |
f623f0db | 3069 | } |
1da177e4 | 3070 | |
b2a0ac88 | 3071 | for_each_migratetype_order(order, t) { |
86760a2c GT |
3072 | list_for_each_entry(page, |
3073 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 3074 | unsigned long i; |
1da177e4 | 3075 | |
86760a2c | 3076 | pfn = page_to_pfn(page); |
556b969a CY |
3077 | for (i = 0; i < (1UL << order); i++) { |
3078 | if (!--page_count) { | |
3079 | touch_nmi_watchdog(); | |
3080 | page_count = WD_PAGE_COUNT; | |
3081 | } | |
7be98234 | 3082 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 3083 | } |
f623f0db | 3084 | } |
b2a0ac88 | 3085 | } |
1da177e4 LT |
3086 | spin_unlock_irqrestore(&zone->lock, flags); |
3087 | } | |
e2c55dc8 | 3088 | #endif /* CONFIG_PM */ |
1da177e4 | 3089 | |
2d4894b5 | 3090 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 3091 | { |
5f8dcc21 | 3092 | int migratetype; |
1da177e4 | 3093 | |
4db7548c | 3094 | if (!free_pcp_prepare(page)) |
9cca35d4 | 3095 | return false; |
689bcebf | 3096 | |
dc4b0caf | 3097 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 3098 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
3099 | return true; |
3100 | } | |
3101 | ||
2d4894b5 | 3102 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
3103 | { |
3104 | struct zone *zone = page_zone(page); | |
3105 | struct per_cpu_pages *pcp; | |
3106 | int migratetype; | |
3107 | ||
3108 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 3109 | __count_vm_event(PGFREE); |
da456f14 | 3110 | |
5f8dcc21 MG |
3111 | /* |
3112 | * We only track unmovable, reclaimable and movable on pcp lists. | |
3113 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 3114 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
3115 | * areas back if necessary. Otherwise, we may have to free |
3116 | * excessively into the page allocator | |
3117 | */ | |
3118 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 3119 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 3120 | free_one_page(zone, page, pfn, 0, migratetype); |
9cca35d4 | 3121 | return; |
5f8dcc21 MG |
3122 | } |
3123 | migratetype = MIGRATE_MOVABLE; | |
3124 | } | |
3125 | ||
99dcc3e5 | 3126 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 3127 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 3128 | pcp->count++; |
48db57f8 | 3129 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 3130 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb | 3131 | free_pcppages_bulk(zone, batch, pcp); |
48db57f8 | 3132 | } |
9cca35d4 | 3133 | } |
5f8dcc21 | 3134 | |
9cca35d4 MG |
3135 | /* |
3136 | * Free a 0-order page | |
9cca35d4 | 3137 | */ |
2d4894b5 | 3138 | void free_unref_page(struct page *page) |
9cca35d4 MG |
3139 | { |
3140 | unsigned long flags; | |
3141 | unsigned long pfn = page_to_pfn(page); | |
3142 | ||
2d4894b5 | 3143 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3144 | return; |
3145 | ||
3146 | local_irq_save(flags); | |
2d4894b5 | 3147 | free_unref_page_commit(page, pfn); |
d34b0733 | 3148 | local_irq_restore(flags); |
1da177e4 LT |
3149 | } |
3150 | ||
cc59850e KK |
3151 | /* |
3152 | * Free a list of 0-order pages | |
3153 | */ | |
2d4894b5 | 3154 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
3155 | { |
3156 | struct page *page, *next; | |
9cca35d4 | 3157 | unsigned long flags, pfn; |
c24ad77d | 3158 | int batch_count = 0; |
9cca35d4 MG |
3159 | |
3160 | /* Prepare pages for freeing */ | |
3161 | list_for_each_entry_safe(page, next, list, lru) { | |
3162 | pfn = page_to_pfn(page); | |
2d4894b5 | 3163 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3164 | list_del(&page->lru); |
3165 | set_page_private(page, pfn); | |
3166 | } | |
cc59850e | 3167 | |
9cca35d4 | 3168 | local_irq_save(flags); |
cc59850e | 3169 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
3170 | unsigned long pfn = page_private(page); |
3171 | ||
3172 | set_page_private(page, 0); | |
2d4894b5 MG |
3173 | trace_mm_page_free_batched(page); |
3174 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
3175 | |
3176 | /* | |
3177 | * Guard against excessive IRQ disabled times when we get | |
3178 | * a large list of pages to free. | |
3179 | */ | |
3180 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
3181 | local_irq_restore(flags); | |
3182 | batch_count = 0; | |
3183 | local_irq_save(flags); | |
3184 | } | |
cc59850e | 3185 | } |
9cca35d4 | 3186 | local_irq_restore(flags); |
cc59850e KK |
3187 | } |
3188 | ||
8dfcc9ba NP |
3189 | /* |
3190 | * split_page takes a non-compound higher-order page, and splits it into | |
3191 | * n (1<<order) sub-pages: page[0..n] | |
3192 | * Each sub-page must be freed individually. | |
3193 | * | |
3194 | * Note: this is probably too low level an operation for use in drivers. | |
3195 | * Please consult with lkml before using this in your driver. | |
3196 | */ | |
3197 | void split_page(struct page *page, unsigned int order) | |
3198 | { | |
3199 | int i; | |
3200 | ||
309381fe SL |
3201 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3202 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3203 | |
a9627bc5 | 3204 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3205 | set_page_refcounted(page + i); |
a9627bc5 | 3206 | split_page_owner(page, order); |
8dfcc9ba | 3207 | } |
5853ff23 | 3208 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3209 | |
3c605096 | 3210 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3211 | { |
748446bb MG |
3212 | unsigned long watermark; |
3213 | struct zone *zone; | |
2139cbe6 | 3214 | int mt; |
748446bb MG |
3215 | |
3216 | BUG_ON(!PageBuddy(page)); | |
3217 | ||
3218 | zone = page_zone(page); | |
2e30abd1 | 3219 | mt = get_pageblock_migratetype(page); |
748446bb | 3220 | |
194159fb | 3221 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
3222 | /* |
3223 | * Obey watermarks as if the page was being allocated. We can | |
3224 | * emulate a high-order watermark check with a raised order-0 | |
3225 | * watermark, because we already know our high-order page | |
3226 | * exists. | |
3227 | */ | |
fd1444b2 | 3228 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3229 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3230 | return 0; |
3231 | ||
8fb74b9f | 3232 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3233 | } |
748446bb MG |
3234 | |
3235 | /* Remove page from free list */ | |
b03641af | 3236 | |
6ab01363 | 3237 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 3238 | |
400bc7fd | 3239 | /* |
3240 | * Set the pageblock if the isolated page is at least half of a | |
3241 | * pageblock | |
3242 | */ | |
748446bb MG |
3243 | if (order >= pageblock_order - 1) { |
3244 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3245 | for (; page < endpage; page += pageblock_nr_pages) { |
3246 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 3247 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 3248 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
3249 | set_pageblock_migratetype(page, |
3250 | MIGRATE_MOVABLE); | |
3251 | } | |
748446bb MG |
3252 | } |
3253 | ||
f3a14ced | 3254 | |
8fb74b9f | 3255 | return 1UL << order; |
1fb3f8ca MG |
3256 | } |
3257 | ||
624f58d8 AD |
3258 | /** |
3259 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
3260 | * @page: Page that was isolated | |
3261 | * @order: Order of the isolated page | |
e6a0a7ad | 3262 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
3263 | * |
3264 | * This function is meant to return a page pulled from the free lists via | |
3265 | * __isolate_free_page back to the free lists they were pulled from. | |
3266 | */ | |
3267 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
3268 | { | |
3269 | struct zone *zone = page_zone(page); | |
3270 | ||
3271 | /* zone lock should be held when this function is called */ | |
3272 | lockdep_assert_held(&zone->lock); | |
3273 | ||
3274 | /* Return isolated page to tail of freelist. */ | |
36e66c55 | 3275 | __free_one_page(page, page_to_pfn(page), zone, order, mt, false); |
624f58d8 AD |
3276 | } |
3277 | ||
060e7417 MG |
3278 | /* |
3279 | * Update NUMA hit/miss statistics | |
3280 | * | |
3281 | * Must be called with interrupts disabled. | |
060e7417 | 3282 | */ |
41b6167e | 3283 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
3284 | { |
3285 | #ifdef CONFIG_NUMA | |
3a321d2a | 3286 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3287 | |
4518085e KW |
3288 | /* skip numa counters update if numa stats is disabled */ |
3289 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3290 | return; | |
3291 | ||
c1093b74 | 3292 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3293 | local_stat = NUMA_OTHER; |
060e7417 | 3294 | |
c1093b74 | 3295 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3a321d2a | 3296 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 3297 | else { |
3a321d2a KW |
3298 | __inc_numa_state(z, NUMA_MISS); |
3299 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 3300 | } |
3a321d2a | 3301 | __inc_numa_state(z, local_stat); |
060e7417 MG |
3302 | #endif |
3303 | } | |
3304 | ||
066b2393 MG |
3305 | /* Remove page from the per-cpu list, caller must protect the list */ |
3306 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
6bb15450 | 3307 | unsigned int alloc_flags, |
453f85d4 | 3308 | struct per_cpu_pages *pcp, |
066b2393 MG |
3309 | struct list_head *list) |
3310 | { | |
3311 | struct page *page; | |
3312 | ||
3313 | do { | |
3314 | if (list_empty(list)) { | |
3315 | pcp->count += rmqueue_bulk(zone, 0, | |
3316 | pcp->batch, list, | |
6bb15450 | 3317 | migratetype, alloc_flags); |
066b2393 MG |
3318 | if (unlikely(list_empty(list))) |
3319 | return NULL; | |
3320 | } | |
3321 | ||
453f85d4 | 3322 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
3323 | list_del(&page->lru); |
3324 | pcp->count--; | |
3325 | } while (check_new_pcp(page)); | |
3326 | ||
3327 | return page; | |
3328 | } | |
3329 | ||
3330 | /* Lock and remove page from the per-cpu list */ | |
3331 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
1c52e6d0 YS |
3332 | struct zone *zone, gfp_t gfp_flags, |
3333 | int migratetype, unsigned int alloc_flags) | |
066b2393 MG |
3334 | { |
3335 | struct per_cpu_pages *pcp; | |
3336 | struct list_head *list; | |
066b2393 | 3337 | struct page *page; |
d34b0733 | 3338 | unsigned long flags; |
066b2393 | 3339 | |
d34b0733 | 3340 | local_irq_save(flags); |
066b2393 MG |
3341 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3342 | list = &pcp->lists[migratetype]; | |
6bb15450 | 3343 | page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list); |
066b2393 | 3344 | if (page) { |
1c52e6d0 | 3345 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1); |
066b2393 MG |
3346 | zone_statistics(preferred_zone, zone); |
3347 | } | |
d34b0733 | 3348 | local_irq_restore(flags); |
066b2393 MG |
3349 | return page; |
3350 | } | |
3351 | ||
1da177e4 | 3352 | /* |
75379191 | 3353 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 3354 | */ |
0a15c3e9 | 3355 | static inline |
066b2393 | 3356 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3357 | struct zone *zone, unsigned int order, |
c603844b MG |
3358 | gfp_t gfp_flags, unsigned int alloc_flags, |
3359 | int migratetype) | |
1da177e4 LT |
3360 | { |
3361 | unsigned long flags; | |
689bcebf | 3362 | struct page *page; |
1da177e4 | 3363 | |
d34b0733 | 3364 | if (likely(order == 0)) { |
1c52e6d0 YS |
3365 | page = rmqueue_pcplist(preferred_zone, zone, gfp_flags, |
3366 | migratetype, alloc_flags); | |
066b2393 MG |
3367 | goto out; |
3368 | } | |
83b9355b | 3369 | |
066b2393 MG |
3370 | /* |
3371 | * We most definitely don't want callers attempting to | |
3372 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3373 | */ | |
3374 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3375 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3376 | |
066b2393 MG |
3377 | do { |
3378 | page = NULL; | |
3379 | if (alloc_flags & ALLOC_HARDER) { | |
3380 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3381 | if (page) | |
3382 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3383 | } | |
a74609fa | 3384 | if (!page) |
6bb15450 | 3385 | page = __rmqueue(zone, order, migratetype, alloc_flags); |
066b2393 MG |
3386 | } while (page && check_new_pages(page, order)); |
3387 | spin_unlock(&zone->lock); | |
3388 | if (!page) | |
3389 | goto failed; | |
3390 | __mod_zone_freepage_state(zone, -(1 << order), | |
3391 | get_pcppage_migratetype(page)); | |
1da177e4 | 3392 | |
16709d1d | 3393 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3394 | zone_statistics(preferred_zone, zone); |
a74609fa | 3395 | local_irq_restore(flags); |
1da177e4 | 3396 | |
066b2393 | 3397 | out: |
73444bc4 MG |
3398 | /* Separate test+clear to avoid unnecessary atomics */ |
3399 | if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) { | |
3400 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); | |
3401 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3402 | } | |
3403 | ||
066b2393 | 3404 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 | 3405 | return page; |
a74609fa NP |
3406 | |
3407 | failed: | |
3408 | local_irq_restore(flags); | |
a74609fa | 3409 | return NULL; |
1da177e4 LT |
3410 | } |
3411 | ||
933e312e AM |
3412 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3413 | ||
b2588c4b | 3414 | static struct { |
933e312e AM |
3415 | struct fault_attr attr; |
3416 | ||
621a5f7a | 3417 | bool ignore_gfp_highmem; |
71baba4b | 3418 | bool ignore_gfp_reclaim; |
54114994 | 3419 | u32 min_order; |
933e312e AM |
3420 | } fail_page_alloc = { |
3421 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3422 | .ignore_gfp_reclaim = true, |
621a5f7a | 3423 | .ignore_gfp_highmem = true, |
54114994 | 3424 | .min_order = 1, |
933e312e AM |
3425 | }; |
3426 | ||
3427 | static int __init setup_fail_page_alloc(char *str) | |
3428 | { | |
3429 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3430 | } | |
3431 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3432 | ||
af3b8544 | 3433 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3434 | { |
54114994 | 3435 | if (order < fail_page_alloc.min_order) |
deaf386e | 3436 | return false; |
933e312e | 3437 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3438 | return false; |
933e312e | 3439 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3440 | return false; |
71baba4b MG |
3441 | if (fail_page_alloc.ignore_gfp_reclaim && |
3442 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3443 | return false; |
933e312e AM |
3444 | |
3445 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3446 | } | |
3447 | ||
3448 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3449 | ||
3450 | static int __init fail_page_alloc_debugfs(void) | |
3451 | { | |
0825a6f9 | 3452 | umode_t mode = S_IFREG | 0600; |
933e312e | 3453 | struct dentry *dir; |
933e312e | 3454 | |
dd48c085 AM |
3455 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3456 | &fail_page_alloc.attr); | |
b2588c4b | 3457 | |
d9f7979c GKH |
3458 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3459 | &fail_page_alloc.ignore_gfp_reclaim); | |
3460 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3461 | &fail_page_alloc.ignore_gfp_highmem); | |
3462 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3463 | |
d9f7979c | 3464 | return 0; |
933e312e AM |
3465 | } |
3466 | ||
3467 | late_initcall(fail_page_alloc_debugfs); | |
3468 | ||
3469 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3470 | ||
3471 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3472 | ||
af3b8544 | 3473 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3474 | { |
deaf386e | 3475 | return false; |
933e312e AM |
3476 | } |
3477 | ||
3478 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3479 | ||
af3b8544 BP |
3480 | static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
3481 | { | |
3482 | return __should_fail_alloc_page(gfp_mask, order); | |
3483 | } | |
3484 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3485 | ||
f27ce0e1 JK |
3486 | static inline long __zone_watermark_unusable_free(struct zone *z, |
3487 | unsigned int order, unsigned int alloc_flags) | |
3488 | { | |
3489 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); | |
3490 | long unusable_free = (1 << order) - 1; | |
3491 | ||
3492 | /* | |
3493 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3494 | * the high-atomic reserves. This will over-estimate the size of the | |
3495 | * atomic reserve but it avoids a search. | |
3496 | */ | |
3497 | if (likely(!alloc_harder)) | |
3498 | unusable_free += z->nr_reserved_highatomic; | |
3499 | ||
3500 | #ifdef CONFIG_CMA | |
3501 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3502 | if (!(alloc_flags & ALLOC_CMA)) | |
3503 | unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES); | |
3504 | #endif | |
3505 | ||
3506 | return unusable_free; | |
3507 | } | |
3508 | ||
1da177e4 | 3509 | /* |
97a16fc8 MG |
3510 | * Return true if free base pages are above 'mark'. For high-order checks it |
3511 | * will return true of the order-0 watermark is reached and there is at least | |
3512 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3513 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3514 | */ |
86a294a8 | 3515 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3516 | int highest_zoneidx, unsigned int alloc_flags, |
86a294a8 | 3517 | long free_pages) |
1da177e4 | 3518 | { |
d23ad423 | 3519 | long min = mark; |
1da177e4 | 3520 | int o; |
cd04ae1e | 3521 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3522 | |
0aaa29a5 | 3523 | /* free_pages may go negative - that's OK */ |
f27ce0e1 | 3524 | free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); |
0aaa29a5 | 3525 | |
7fb1d9fc | 3526 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3527 | min -= min / 2; |
0aaa29a5 | 3528 | |
f27ce0e1 | 3529 | if (unlikely(alloc_harder)) { |
cd04ae1e MH |
3530 | /* |
3531 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3532 | * users on the grounds that it's definitely going to be in | |
3533 | * the exit path shortly and free memory. Any allocation it | |
3534 | * makes during the free path will be small and short-lived. | |
3535 | */ | |
3536 | if (alloc_flags & ALLOC_OOM) | |
3537 | min -= min / 2; | |
3538 | else | |
3539 | min -= min / 4; | |
3540 | } | |
3541 | ||
97a16fc8 MG |
3542 | /* |
3543 | * Check watermarks for an order-0 allocation request. If these | |
3544 | * are not met, then a high-order request also cannot go ahead | |
3545 | * even if a suitable page happened to be free. | |
3546 | */ | |
97a225e6 | 3547 | if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) |
88f5acf8 | 3548 | return false; |
1da177e4 | 3549 | |
97a16fc8 MG |
3550 | /* If this is an order-0 request then the watermark is fine */ |
3551 | if (!order) | |
3552 | return true; | |
3553 | ||
3554 | /* For a high-order request, check at least one suitable page is free */ | |
3555 | for (o = order; o < MAX_ORDER; o++) { | |
3556 | struct free_area *area = &z->free_area[o]; | |
3557 | int mt; | |
3558 | ||
3559 | if (!area->nr_free) | |
3560 | continue; | |
3561 | ||
97a16fc8 | 3562 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3563 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3564 | return true; |
3565 | } | |
3566 | ||
3567 | #ifdef CONFIG_CMA | |
d883c6cf | 3568 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3569 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3570 | return true; |
d883c6cf | 3571 | } |
97a16fc8 | 3572 | #endif |
76089d00 | 3573 | if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC)) |
b050e376 | 3574 | return true; |
1da177e4 | 3575 | } |
97a16fc8 | 3576 | return false; |
88f5acf8 MG |
3577 | } |
3578 | ||
7aeb09f9 | 3579 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
97a225e6 | 3580 | int highest_zoneidx, unsigned int alloc_flags) |
88f5acf8 | 3581 | { |
97a225e6 | 3582 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
88f5acf8 MG |
3583 | zone_page_state(z, NR_FREE_PAGES)); |
3584 | } | |
3585 | ||
48ee5f36 | 3586 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
97a225e6 | 3587 | unsigned long mark, int highest_zoneidx, |
f80b08fc | 3588 | unsigned int alloc_flags, gfp_t gfp_mask) |
48ee5f36 | 3589 | { |
f27ce0e1 | 3590 | long free_pages; |
d883c6cf | 3591 | |
f27ce0e1 | 3592 | free_pages = zone_page_state(z, NR_FREE_PAGES); |
48ee5f36 MG |
3593 | |
3594 | /* | |
3595 | * Fast check for order-0 only. If this fails then the reserves | |
f27ce0e1 | 3596 | * need to be calculated. |
48ee5f36 | 3597 | */ |
f27ce0e1 JK |
3598 | if (!order) { |
3599 | long fast_free; | |
3600 | ||
3601 | fast_free = free_pages; | |
3602 | fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags); | |
3603 | if (fast_free > mark + z->lowmem_reserve[highest_zoneidx]) | |
3604 | return true; | |
3605 | } | |
48ee5f36 | 3606 | |
f80b08fc CTR |
3607 | if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags, |
3608 | free_pages)) | |
3609 | return true; | |
3610 | /* | |
3611 | * Ignore watermark boosting for GFP_ATOMIC order-0 allocations | |
3612 | * when checking the min watermark. The min watermark is the | |
3613 | * point where boosting is ignored so that kswapd is woken up | |
3614 | * when below the low watermark. | |
3615 | */ | |
3616 | if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost | |
3617 | && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) { | |
3618 | mark = z->_watermark[WMARK_MIN]; | |
3619 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, | |
3620 | alloc_flags, free_pages); | |
3621 | } | |
3622 | ||
3623 | return false; | |
48ee5f36 MG |
3624 | } |
3625 | ||
7aeb09f9 | 3626 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
97a225e6 | 3627 | unsigned long mark, int highest_zoneidx) |
88f5acf8 MG |
3628 | { |
3629 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3630 | ||
3631 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3632 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3633 | ||
97a225e6 | 3634 | return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0, |
88f5acf8 | 3635 | free_pages); |
1da177e4 LT |
3636 | } |
3637 | ||
9276b1bc | 3638 | #ifdef CONFIG_NUMA |
957f822a DR |
3639 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3640 | { | |
e02dc017 | 3641 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3642 | node_reclaim_distance; |
957f822a | 3643 | } |
9276b1bc | 3644 | #else /* CONFIG_NUMA */ |
957f822a DR |
3645 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3646 | { | |
3647 | return true; | |
3648 | } | |
9276b1bc PJ |
3649 | #endif /* CONFIG_NUMA */ |
3650 | ||
6bb15450 MG |
3651 | /* |
3652 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3653 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3654 | * premature use of a lower zone may cause lowmem pressure problems that | |
3655 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3656 | * probably too small. It only makes sense to spread allocations to avoid | |
3657 | * fragmentation between the Normal and DMA32 zones. | |
3658 | */ | |
3659 | static inline unsigned int | |
0a79cdad | 3660 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3661 | { |
736838e9 | 3662 | unsigned int alloc_flags; |
0a79cdad | 3663 | |
736838e9 MN |
3664 | /* |
3665 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
3666 | * to save a branch. | |
3667 | */ | |
3668 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
3669 | |
3670 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3671 | if (!zone) |
3672 | return alloc_flags; | |
3673 | ||
6bb15450 | 3674 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3675 | return alloc_flags; |
6bb15450 MG |
3676 | |
3677 | /* | |
3678 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3679 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3680 | * on UMA that if Normal is populated then so is DMA32. | |
3681 | */ | |
3682 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3683 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3684 | return alloc_flags; |
6bb15450 | 3685 | |
8118b82e | 3686 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3687 | #endif /* CONFIG_ZONE_DMA32 */ |
3688 | return alloc_flags; | |
6bb15450 | 3689 | } |
6bb15450 | 3690 | |
8510e69c JK |
3691 | static inline unsigned int current_alloc_flags(gfp_t gfp_mask, |
3692 | unsigned int alloc_flags) | |
3693 | { | |
3694 | #ifdef CONFIG_CMA | |
3695 | unsigned int pflags = current->flags; | |
3696 | ||
3697 | if (!(pflags & PF_MEMALLOC_NOCMA) && | |
3698 | gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
3699 | alloc_flags |= ALLOC_CMA; | |
3700 | ||
3701 | #endif | |
3702 | return alloc_flags; | |
3703 | } | |
3704 | ||
7fb1d9fc | 3705 | /* |
0798e519 | 3706 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3707 | * a page. |
3708 | */ | |
3709 | static struct page * | |
a9263751 VB |
3710 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3711 | const struct alloc_context *ac) | |
753ee728 | 3712 | { |
6bb15450 | 3713 | struct zoneref *z; |
5117f45d | 3714 | struct zone *zone; |
3b8c0be4 | 3715 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
6bb15450 | 3716 | bool no_fallback; |
3b8c0be4 | 3717 | |
6bb15450 | 3718 | retry: |
7fb1d9fc | 3719 | /* |
9276b1bc | 3720 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3721 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3722 | */ |
6bb15450 MG |
3723 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3724 | z = ac->preferred_zoneref; | |
97a225e6 JK |
3725 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3726 | ac->highest_zoneidx, ac->nodemask) { | |
be06af00 | 3727 | struct page *page; |
e085dbc5 JW |
3728 | unsigned long mark; |
3729 | ||
664eedde MG |
3730 | if (cpusets_enabled() && |
3731 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3732 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3733 | continue; |
a756cf59 JW |
3734 | /* |
3735 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3736 | * want to get it from a node that is within its dirty |
3737 | * limit, such that no single node holds more than its | |
a756cf59 | 3738 | * proportional share of globally allowed dirty pages. |
281e3726 | 3739 | * The dirty limits take into account the node's |
a756cf59 JW |
3740 | * lowmem reserves and high watermark so that kswapd |
3741 | * should be able to balance it without having to | |
3742 | * write pages from its LRU list. | |
3743 | * | |
a756cf59 | 3744 | * XXX: For now, allow allocations to potentially |
281e3726 | 3745 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3746 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3747 | * which is important when on a NUMA setup the allowed |
281e3726 | 3748 | * nodes are together not big enough to reach the |
a756cf59 | 3749 | * global limit. The proper fix for these situations |
281e3726 | 3750 | * will require awareness of nodes in the |
a756cf59 JW |
3751 | * dirty-throttling and the flusher threads. |
3752 | */ | |
3b8c0be4 MG |
3753 | if (ac->spread_dirty_pages) { |
3754 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3755 | continue; | |
3756 | ||
3757 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3758 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3759 | continue; | |
3760 | } | |
3761 | } | |
7fb1d9fc | 3762 | |
6bb15450 MG |
3763 | if (no_fallback && nr_online_nodes > 1 && |
3764 | zone != ac->preferred_zoneref->zone) { | |
3765 | int local_nid; | |
3766 | ||
3767 | /* | |
3768 | * If moving to a remote node, retry but allow | |
3769 | * fragmenting fallbacks. Locality is more important | |
3770 | * than fragmentation avoidance. | |
3771 | */ | |
3772 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3773 | if (zone_to_nid(zone) != local_nid) { | |
3774 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3775 | goto retry; | |
3776 | } | |
3777 | } | |
3778 | ||
a9214443 | 3779 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 3780 | if (!zone_watermark_fast(zone, order, mark, |
f80b08fc CTR |
3781 | ac->highest_zoneidx, alloc_flags, |
3782 | gfp_mask)) { | |
fa5e084e MG |
3783 | int ret; |
3784 | ||
c9e97a19 PT |
3785 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3786 | /* | |
3787 | * Watermark failed for this zone, but see if we can | |
3788 | * grow this zone if it contains deferred pages. | |
3789 | */ | |
3790 | if (static_branch_unlikely(&deferred_pages)) { | |
3791 | if (_deferred_grow_zone(zone, order)) | |
3792 | goto try_this_zone; | |
3793 | } | |
3794 | #endif | |
5dab2911 MG |
3795 | /* Checked here to keep the fast path fast */ |
3796 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3797 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3798 | goto try_this_zone; | |
3799 | ||
a5f5f91d | 3800 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3801 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3802 | continue; |
3803 | ||
a5f5f91d | 3804 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3805 | switch (ret) { |
a5f5f91d | 3806 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3807 | /* did not scan */ |
cd38b115 | 3808 | continue; |
a5f5f91d | 3809 | case NODE_RECLAIM_FULL: |
fa5e084e | 3810 | /* scanned but unreclaimable */ |
cd38b115 | 3811 | continue; |
fa5e084e MG |
3812 | default: |
3813 | /* did we reclaim enough */ | |
fed2719e | 3814 | if (zone_watermark_ok(zone, order, mark, |
97a225e6 | 3815 | ac->highest_zoneidx, alloc_flags)) |
fed2719e MG |
3816 | goto try_this_zone; |
3817 | ||
fed2719e | 3818 | continue; |
0798e519 | 3819 | } |
7fb1d9fc RS |
3820 | } |
3821 | ||
fa5e084e | 3822 | try_this_zone: |
066b2393 | 3823 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3824 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3825 | if (page) { |
479f854a | 3826 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3827 | |
3828 | /* | |
3829 | * If this is a high-order atomic allocation then check | |
3830 | * if the pageblock should be reserved for the future | |
3831 | */ | |
3832 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3833 | reserve_highatomic_pageblock(page, zone, order); | |
3834 | ||
75379191 | 3835 | return page; |
c9e97a19 PT |
3836 | } else { |
3837 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
3838 | /* Try again if zone has deferred pages */ | |
3839 | if (static_branch_unlikely(&deferred_pages)) { | |
3840 | if (_deferred_grow_zone(zone, order)) | |
3841 | goto try_this_zone; | |
3842 | } | |
3843 | #endif | |
75379191 | 3844 | } |
54a6eb5c | 3845 | } |
9276b1bc | 3846 | |
6bb15450 MG |
3847 | /* |
3848 | * It's possible on a UMA machine to get through all zones that are | |
3849 | * fragmented. If avoiding fragmentation, reset and try again. | |
3850 | */ | |
3851 | if (no_fallback) { | |
3852 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3853 | goto retry; | |
3854 | } | |
3855 | ||
4ffeaf35 | 3856 | return NULL; |
753ee728 MH |
3857 | } |
3858 | ||
9af744d7 | 3859 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3860 | { |
a238ab5b | 3861 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
3862 | |
3863 | /* | |
3864 | * This documents exceptions given to allocations in certain | |
3865 | * contexts that are allowed to allocate outside current's set | |
3866 | * of allowed nodes. | |
3867 | */ | |
3868 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3869 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3870 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3871 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3872 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3873 | filter &= ~SHOW_MEM_FILTER_NODES; |
3874 | ||
9af744d7 | 3875 | show_mem(filter, nodemask); |
aa187507 MH |
3876 | } |
3877 | ||
a8e99259 | 3878 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3879 | { |
3880 | struct va_format vaf; | |
3881 | va_list args; | |
1be334e5 | 3882 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 3883 | |
0f7896f1 | 3884 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3885 | return; |
3886 | ||
7877cdcc MH |
3887 | va_start(args, fmt); |
3888 | vaf.fmt = fmt; | |
3889 | vaf.va = &args; | |
ef8444ea | 3890 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
3891 | current->comm, &vaf, gfp_mask, &gfp_mask, |
3892 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3893 | va_end(args); |
3ee9a4f0 | 3894 | |
a8e99259 | 3895 | cpuset_print_current_mems_allowed(); |
ef8444ea | 3896 | pr_cont("\n"); |
a238ab5b | 3897 | dump_stack(); |
685dbf6f | 3898 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3899 | } |
3900 | ||
6c18ba7a MH |
3901 | static inline struct page * |
3902 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3903 | unsigned int alloc_flags, | |
3904 | const struct alloc_context *ac) | |
3905 | { | |
3906 | struct page *page; | |
3907 | ||
3908 | page = get_page_from_freelist(gfp_mask, order, | |
3909 | alloc_flags|ALLOC_CPUSET, ac); | |
3910 | /* | |
3911 | * fallback to ignore cpuset restriction if our nodes | |
3912 | * are depleted | |
3913 | */ | |
3914 | if (!page) | |
3915 | page = get_page_from_freelist(gfp_mask, order, | |
3916 | alloc_flags, ac); | |
3917 | ||
3918 | return page; | |
3919 | } | |
3920 | ||
11e33f6a MG |
3921 | static inline struct page * |
3922 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3923 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3924 | { |
6e0fc46d DR |
3925 | struct oom_control oc = { |
3926 | .zonelist = ac->zonelist, | |
3927 | .nodemask = ac->nodemask, | |
2a966b77 | 3928 | .memcg = NULL, |
6e0fc46d DR |
3929 | .gfp_mask = gfp_mask, |
3930 | .order = order, | |
6e0fc46d | 3931 | }; |
11e33f6a MG |
3932 | struct page *page; |
3933 | ||
9879de73 JW |
3934 | *did_some_progress = 0; |
3935 | ||
9879de73 | 3936 | /* |
dc56401f JW |
3937 | * Acquire the oom lock. If that fails, somebody else is |
3938 | * making progress for us. | |
9879de73 | 3939 | */ |
dc56401f | 3940 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3941 | *did_some_progress = 1; |
11e33f6a | 3942 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3943 | return NULL; |
3944 | } | |
6b1de916 | 3945 | |
11e33f6a MG |
3946 | /* |
3947 | * Go through the zonelist yet one more time, keep very high watermark | |
3948 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3949 | * we're still under heavy pressure. But make sure that this reclaim |
3950 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3951 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3952 | */ |
e746bf73 TH |
3953 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3954 | ~__GFP_DIRECT_RECLAIM, order, | |
3955 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3956 | if (page) |
11e33f6a MG |
3957 | goto out; |
3958 | ||
06ad276a MH |
3959 | /* Coredumps can quickly deplete all memory reserves */ |
3960 | if (current->flags & PF_DUMPCORE) | |
3961 | goto out; | |
3962 | /* The OOM killer will not help higher order allocs */ | |
3963 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3964 | goto out; | |
dcda9b04 MH |
3965 | /* |
3966 | * We have already exhausted all our reclaim opportunities without any | |
3967 | * success so it is time to admit defeat. We will skip the OOM killer | |
3968 | * because it is very likely that the caller has a more reasonable | |
3969 | * fallback than shooting a random task. | |
3970 | */ | |
3971 | if (gfp_mask & __GFP_RETRY_MAYFAIL) | |
3972 | goto out; | |
06ad276a | 3973 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
97a225e6 | 3974 | if (ac->highest_zoneidx < ZONE_NORMAL) |
06ad276a MH |
3975 | goto out; |
3976 | if (pm_suspended_storage()) | |
3977 | goto out; | |
3978 | /* | |
3979 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3980 | * other request to make a forward progress. | |
3981 | * We are in an unfortunate situation where out_of_memory cannot | |
3982 | * do much for this context but let's try it to at least get | |
3983 | * access to memory reserved if the current task is killed (see | |
3984 | * out_of_memory). Once filesystems are ready to handle allocation | |
3985 | * failures more gracefully we should just bail out here. | |
3986 | */ | |
3987 | ||
3988 | /* The OOM killer may not free memory on a specific node */ | |
3989 | if (gfp_mask & __GFP_THISNODE) | |
3990 | goto out; | |
3da88fb3 | 3991 | |
3c2c6488 | 3992 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 3993 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3994 | *did_some_progress = 1; |
5020e285 | 3995 | |
6c18ba7a MH |
3996 | /* |
3997 | * Help non-failing allocations by giving them access to memory | |
3998 | * reserves | |
3999 | */ | |
4000 | if (gfp_mask & __GFP_NOFAIL) | |
4001 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 4002 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 4003 | } |
11e33f6a | 4004 | out: |
dc56401f | 4005 | mutex_unlock(&oom_lock); |
11e33f6a MG |
4006 | return page; |
4007 | } | |
4008 | ||
33c2d214 MH |
4009 | /* |
4010 | * Maximum number of compaction retries wit a progress before OOM | |
4011 | * killer is consider as the only way to move forward. | |
4012 | */ | |
4013 | #define MAX_COMPACT_RETRIES 16 | |
4014 | ||
56de7263 MG |
4015 | #ifdef CONFIG_COMPACTION |
4016 | /* Try memory compaction for high-order allocations before reclaim */ | |
4017 | static struct page * | |
4018 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4019 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4020 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4021 | { |
5e1f0f09 | 4022 | struct page *page = NULL; |
eb414681 | 4023 | unsigned long pflags; |
499118e9 | 4024 | unsigned int noreclaim_flag; |
53853e2d VB |
4025 | |
4026 | if (!order) | |
66199712 | 4027 | return NULL; |
66199712 | 4028 | |
eb414681 | 4029 | psi_memstall_enter(&pflags); |
499118e9 | 4030 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 4031 | |
c5d01d0d | 4032 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 4033 | prio, &page); |
eb414681 | 4034 | |
499118e9 | 4035 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 4036 | psi_memstall_leave(&pflags); |
56de7263 | 4037 | |
98dd3b48 VB |
4038 | /* |
4039 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
4040 | * count a compaction stall | |
4041 | */ | |
4042 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 4043 | |
5e1f0f09 MG |
4044 | /* Prep a captured page if available */ |
4045 | if (page) | |
4046 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
4047 | ||
4048 | /* Try get a page from the freelist if available */ | |
4049 | if (!page) | |
4050 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 4051 | |
98dd3b48 VB |
4052 | if (page) { |
4053 | struct zone *zone = page_zone(page); | |
53853e2d | 4054 | |
98dd3b48 VB |
4055 | zone->compact_blockskip_flush = false; |
4056 | compaction_defer_reset(zone, order, true); | |
4057 | count_vm_event(COMPACTSUCCESS); | |
4058 | return page; | |
4059 | } | |
56de7263 | 4060 | |
98dd3b48 VB |
4061 | /* |
4062 | * It's bad if compaction run occurs and fails. The most likely reason | |
4063 | * is that pages exist, but not enough to satisfy watermarks. | |
4064 | */ | |
4065 | count_vm_event(COMPACTFAIL); | |
66199712 | 4066 | |
98dd3b48 | 4067 | cond_resched(); |
56de7263 MG |
4068 | |
4069 | return NULL; | |
4070 | } | |
33c2d214 | 4071 | |
3250845d VB |
4072 | static inline bool |
4073 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
4074 | enum compact_result compact_result, | |
4075 | enum compact_priority *compact_priority, | |
d9436498 | 4076 | int *compaction_retries) |
3250845d VB |
4077 | { |
4078 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 4079 | int min_priority; |
65190cff MH |
4080 | bool ret = false; |
4081 | int retries = *compaction_retries; | |
4082 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
4083 | |
4084 | if (!order) | |
4085 | return false; | |
4086 | ||
d9436498 VB |
4087 | if (compaction_made_progress(compact_result)) |
4088 | (*compaction_retries)++; | |
4089 | ||
3250845d VB |
4090 | /* |
4091 | * compaction considers all the zone as desperately out of memory | |
4092 | * so it doesn't really make much sense to retry except when the | |
4093 | * failure could be caused by insufficient priority | |
4094 | */ | |
d9436498 VB |
4095 | if (compaction_failed(compact_result)) |
4096 | goto check_priority; | |
3250845d | 4097 | |
49433085 VB |
4098 | /* |
4099 | * compaction was skipped because there are not enough order-0 pages | |
4100 | * to work with, so we retry only if it looks like reclaim can help. | |
4101 | */ | |
4102 | if (compaction_needs_reclaim(compact_result)) { | |
4103 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
4104 | goto out; | |
4105 | } | |
4106 | ||
3250845d VB |
4107 | /* |
4108 | * make sure the compaction wasn't deferred or didn't bail out early | |
4109 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
4110 | * But the next retry should use a higher priority if allowed, so |
4111 | * we don't just keep bailing out endlessly. | |
3250845d | 4112 | */ |
65190cff | 4113 | if (compaction_withdrawn(compact_result)) { |
49433085 | 4114 | goto check_priority; |
65190cff | 4115 | } |
3250845d VB |
4116 | |
4117 | /* | |
dcda9b04 | 4118 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
4119 | * costly ones because they are de facto nofail and invoke OOM |
4120 | * killer to move on while costly can fail and users are ready | |
4121 | * to cope with that. 1/4 retries is rather arbitrary but we | |
4122 | * would need much more detailed feedback from compaction to | |
4123 | * make a better decision. | |
4124 | */ | |
4125 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4126 | max_retries /= 4; | |
65190cff MH |
4127 | if (*compaction_retries <= max_retries) { |
4128 | ret = true; | |
4129 | goto out; | |
4130 | } | |
3250845d | 4131 | |
d9436498 VB |
4132 | /* |
4133 | * Make sure there are attempts at the highest priority if we exhausted | |
4134 | * all retries or failed at the lower priorities. | |
4135 | */ | |
4136 | check_priority: | |
c2033b00 VB |
4137 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
4138 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 4139 | |
c2033b00 | 4140 | if (*compact_priority > min_priority) { |
d9436498 VB |
4141 | (*compact_priority)--; |
4142 | *compaction_retries = 0; | |
65190cff | 4143 | ret = true; |
d9436498 | 4144 | } |
65190cff MH |
4145 | out: |
4146 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4147 | return ret; | |
3250845d | 4148 | } |
56de7263 MG |
4149 | #else |
4150 | static inline struct page * | |
4151 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4152 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4153 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4154 | { |
33c2d214 | 4155 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4156 | return NULL; |
4157 | } | |
33c2d214 MH |
4158 | |
4159 | static inline bool | |
86a294a8 MH |
4160 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4161 | enum compact_result compact_result, | |
a5508cd8 | 4162 | enum compact_priority *compact_priority, |
d9436498 | 4163 | int *compaction_retries) |
33c2d214 | 4164 | { |
31e49bfd MH |
4165 | struct zone *zone; |
4166 | struct zoneref *z; | |
4167 | ||
4168 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4169 | return false; | |
4170 | ||
4171 | /* | |
4172 | * There are setups with compaction disabled which would prefer to loop | |
4173 | * inside the allocator rather than hit the oom killer prematurely. | |
4174 | * Let's give them a good hope and keep retrying while the order-0 | |
4175 | * watermarks are OK. | |
4176 | */ | |
97a225e6 JK |
4177 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4178 | ac->highest_zoneidx, ac->nodemask) { | |
31e49bfd | 4179 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
97a225e6 | 4180 | ac->highest_zoneidx, alloc_flags)) |
31e49bfd MH |
4181 | return true; |
4182 | } | |
33c2d214 MH |
4183 | return false; |
4184 | } | |
3250845d | 4185 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4186 | |
d92a8cfc | 4187 | #ifdef CONFIG_LOCKDEP |
93781325 | 4188 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4189 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4190 | ||
4191 | static bool __need_fs_reclaim(gfp_t gfp_mask) | |
4192 | { | |
4193 | gfp_mask = current_gfp_context(gfp_mask); | |
4194 | ||
4195 | /* no reclaim without waiting on it */ | |
4196 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4197 | return false; | |
4198 | ||
4199 | /* this guy won't enter reclaim */ | |
2e517d68 | 4200 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4201 | return false; |
4202 | ||
4203 | /* We're only interested __GFP_FS allocations for now */ | |
4204 | if (!(gfp_mask & __GFP_FS)) | |
4205 | return false; | |
4206 | ||
4207 | if (gfp_mask & __GFP_NOLOCKDEP) | |
4208 | return false; | |
4209 | ||
4210 | return true; | |
4211 | } | |
4212 | ||
93781325 OS |
4213 | void __fs_reclaim_acquire(void) |
4214 | { | |
4215 | lock_map_acquire(&__fs_reclaim_map); | |
4216 | } | |
4217 | ||
4218 | void __fs_reclaim_release(void) | |
4219 | { | |
4220 | lock_map_release(&__fs_reclaim_map); | |
4221 | } | |
4222 | ||
d92a8cfc PZ |
4223 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4224 | { | |
4225 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 4226 | __fs_reclaim_acquire(); |
d92a8cfc PZ |
4227 | } |
4228 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4229 | ||
4230 | void fs_reclaim_release(gfp_t gfp_mask) | |
4231 | { | |
4232 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 4233 | __fs_reclaim_release(); |
d92a8cfc PZ |
4234 | } |
4235 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4236 | #endif | |
4237 | ||
bba90710 MS |
4238 | /* Perform direct synchronous page reclaim */ |
4239 | static int | |
a9263751 VB |
4240 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4241 | const struct alloc_context *ac) | |
11e33f6a | 4242 | { |
bba90710 | 4243 | int progress; |
499118e9 | 4244 | unsigned int noreclaim_flag; |
eb414681 | 4245 | unsigned long pflags; |
11e33f6a MG |
4246 | |
4247 | cond_resched(); | |
4248 | ||
4249 | /* We now go into synchronous reclaim */ | |
4250 | cpuset_memory_pressure_bump(); | |
eb414681 | 4251 | psi_memstall_enter(&pflags); |
d92a8cfc | 4252 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4253 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4254 | |
a9263751 VB |
4255 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4256 | ac->nodemask); | |
11e33f6a | 4257 | |
499118e9 | 4258 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4259 | fs_reclaim_release(gfp_mask); |
eb414681 | 4260 | psi_memstall_leave(&pflags); |
11e33f6a MG |
4261 | |
4262 | cond_resched(); | |
4263 | ||
bba90710 MS |
4264 | return progress; |
4265 | } | |
4266 | ||
4267 | /* The really slow allocator path where we enter direct reclaim */ | |
4268 | static inline struct page * | |
4269 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4270 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4271 | unsigned long *did_some_progress) |
bba90710 MS |
4272 | { |
4273 | struct page *page = NULL; | |
4274 | bool drained = false; | |
4275 | ||
a9263751 | 4276 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
4277 | if (unlikely(!(*did_some_progress))) |
4278 | return NULL; | |
11e33f6a | 4279 | |
9ee493ce | 4280 | retry: |
31a6c190 | 4281 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4282 | |
4283 | /* | |
4284 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 | 4285 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
047b9967 | 4286 | * Shrink them and try again |
9ee493ce MG |
4287 | */ |
4288 | if (!page && !drained) { | |
29fac03b | 4289 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4290 | drain_all_pages(NULL); |
9ee493ce MG |
4291 | drained = true; |
4292 | goto retry; | |
4293 | } | |
4294 | ||
11e33f6a MG |
4295 | return page; |
4296 | } | |
4297 | ||
5ecd9d40 DR |
4298 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4299 | const struct alloc_context *ac) | |
3a025760 JW |
4300 | { |
4301 | struct zoneref *z; | |
4302 | struct zone *zone; | |
e1a55637 | 4303 | pg_data_t *last_pgdat = NULL; |
97a225e6 | 4304 | enum zone_type highest_zoneidx = ac->highest_zoneidx; |
3a025760 | 4305 | |
97a225e6 | 4306 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, |
5ecd9d40 | 4307 | ac->nodemask) { |
e1a55637 | 4308 | if (last_pgdat != zone->zone_pgdat) |
97a225e6 | 4309 | wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx); |
e1a55637 MG |
4310 | last_pgdat = zone->zone_pgdat; |
4311 | } | |
3a025760 JW |
4312 | } |
4313 | ||
c603844b | 4314 | static inline unsigned int |
341ce06f PZ |
4315 | gfp_to_alloc_flags(gfp_t gfp_mask) |
4316 | { | |
c603844b | 4317 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4318 | |
736838e9 MN |
4319 | /* |
4320 | * __GFP_HIGH is assumed to be the same as ALLOC_HIGH | |
4321 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4322 | * to save two branches. | |
4323 | */ | |
e6223a3b | 4324 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
736838e9 | 4325 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 4326 | |
341ce06f PZ |
4327 | /* |
4328 | * The caller may dip into page reserves a bit more if the caller | |
4329 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4330 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 4331 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 4332 | */ |
736838e9 MN |
4333 | alloc_flags |= (__force int) |
4334 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 4335 | |
d0164adc | 4336 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 4337 | /* |
b104a35d DR |
4338 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4339 | * if it can't schedule. | |
5c3240d9 | 4340 | */ |
b104a35d | 4341 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 4342 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 4343 | /* |
b104a35d | 4344 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 4345 | * comment for __cpuset_node_allowed(). |
523b9458 | 4346 | */ |
341ce06f | 4347 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 4348 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
4349 | alloc_flags |= ALLOC_HARDER; |
4350 | ||
8510e69c JK |
4351 | alloc_flags = current_alloc_flags(gfp_mask, alloc_flags); |
4352 | ||
341ce06f PZ |
4353 | return alloc_flags; |
4354 | } | |
4355 | ||
cd04ae1e | 4356 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4357 | { |
cd04ae1e MH |
4358 | if (!tsk_is_oom_victim(tsk)) |
4359 | return false; | |
4360 | ||
4361 | /* | |
4362 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4363 | * only to the thread with TIF_MEMDIE set | |
4364 | */ | |
4365 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4366 | return false; |
4367 | ||
cd04ae1e MH |
4368 | return true; |
4369 | } | |
4370 | ||
4371 | /* | |
4372 | * Distinguish requests which really need access to full memory | |
4373 | * reserves from oom victims which can live with a portion of it | |
4374 | */ | |
4375 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4376 | { | |
4377 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4378 | return 0; | |
31a6c190 | 4379 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4380 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4381 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4382 | return ALLOC_NO_WATERMARKS; |
4383 | if (!in_interrupt()) { | |
4384 | if (current->flags & PF_MEMALLOC) | |
4385 | return ALLOC_NO_WATERMARKS; | |
4386 | else if (oom_reserves_allowed(current)) | |
4387 | return ALLOC_OOM; | |
4388 | } | |
31a6c190 | 4389 | |
cd04ae1e MH |
4390 | return 0; |
4391 | } | |
4392 | ||
4393 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4394 | { | |
4395 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4396 | } |
4397 | ||
0a0337e0 MH |
4398 | /* |
4399 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4400 | * for the given allocation request. | |
491d79ae JW |
4401 | * |
4402 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4403 | * without success, or when we couldn't even meet the watermark if we | |
4404 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4405 | * |
4406 | * Returns true if a retry is viable or false to enter the oom path. | |
4407 | */ | |
4408 | static inline bool | |
4409 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4410 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4411 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4412 | { |
4413 | struct zone *zone; | |
4414 | struct zoneref *z; | |
15f570bf | 4415 | bool ret = false; |
0a0337e0 | 4416 | |
423b452e VB |
4417 | /* |
4418 | * Costly allocations might have made a progress but this doesn't mean | |
4419 | * their order will become available due to high fragmentation so | |
4420 | * always increment the no progress counter for them | |
4421 | */ | |
4422 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4423 | *no_progress_loops = 0; | |
4424 | else | |
4425 | (*no_progress_loops)++; | |
4426 | ||
0a0337e0 MH |
4427 | /* |
4428 | * Make sure we converge to OOM if we cannot make any progress | |
4429 | * several times in the row. | |
4430 | */ | |
04c8716f MK |
4431 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4432 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4433 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4434 | } |
0a0337e0 | 4435 | |
bca67592 MG |
4436 | /* |
4437 | * Keep reclaiming pages while there is a chance this will lead | |
4438 | * somewhere. If none of the target zones can satisfy our allocation | |
4439 | * request even if all reclaimable pages are considered then we are | |
4440 | * screwed and have to go OOM. | |
0a0337e0 | 4441 | */ |
97a225e6 JK |
4442 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
4443 | ac->highest_zoneidx, ac->nodemask) { | |
0a0337e0 | 4444 | unsigned long available; |
ede37713 | 4445 | unsigned long reclaimable; |
d379f01d MH |
4446 | unsigned long min_wmark = min_wmark_pages(zone); |
4447 | bool wmark; | |
0a0337e0 | 4448 | |
5a1c84b4 | 4449 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4450 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4451 | |
4452 | /* | |
491d79ae JW |
4453 | * Would the allocation succeed if we reclaimed all |
4454 | * reclaimable pages? | |
0a0337e0 | 4455 | */ |
d379f01d | 4456 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
97a225e6 | 4457 | ac->highest_zoneidx, alloc_flags, available); |
d379f01d MH |
4458 | trace_reclaim_retry_zone(z, order, reclaimable, |
4459 | available, min_wmark, *no_progress_loops, wmark); | |
4460 | if (wmark) { | |
ede37713 MH |
4461 | /* |
4462 | * If we didn't make any progress and have a lot of | |
4463 | * dirty + writeback pages then we should wait for | |
4464 | * an IO to complete to slow down the reclaim and | |
4465 | * prevent from pre mature OOM | |
4466 | */ | |
4467 | if (!did_some_progress) { | |
11fb9989 | 4468 | unsigned long write_pending; |
ede37713 | 4469 | |
5a1c84b4 MG |
4470 | write_pending = zone_page_state_snapshot(zone, |
4471 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 4472 | |
11fb9989 | 4473 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
4474 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
4475 | return true; | |
4476 | } | |
4477 | } | |
5a1c84b4 | 4478 | |
15f570bf MH |
4479 | ret = true; |
4480 | goto out; | |
0a0337e0 MH |
4481 | } |
4482 | } | |
4483 | ||
15f570bf MH |
4484 | out: |
4485 | /* | |
4486 | * Memory allocation/reclaim might be called from a WQ context and the | |
4487 | * current implementation of the WQ concurrency control doesn't | |
4488 | * recognize that a particular WQ is congested if the worker thread is | |
4489 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4490 | * here rather than calling cond_resched(). | |
4491 | */ | |
4492 | if (current->flags & PF_WQ_WORKER) | |
4493 | schedule_timeout_uninterruptible(1); | |
4494 | else | |
4495 | cond_resched(); | |
4496 | return ret; | |
0a0337e0 MH |
4497 | } |
4498 | ||
902b6281 VB |
4499 | static inline bool |
4500 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4501 | { | |
4502 | /* | |
4503 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4504 | * mempolicy don't intersect. This should be normally dealt with by | |
4505 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4506 | * such a way the check therein was true, and then it became false | |
4507 | * before we got our cpuset_mems_cookie here. | |
4508 | * This assumes that for all allocations, ac->nodemask can come only | |
4509 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4510 | * when it does not intersect with the cpuset restrictions) or the | |
4511 | * caller can deal with a violated nodemask. | |
4512 | */ | |
4513 | if (cpusets_enabled() && ac->nodemask && | |
4514 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4515 | ac->nodemask = NULL; | |
4516 | return true; | |
4517 | } | |
4518 | ||
4519 | /* | |
4520 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4521 | * possible to race with parallel threads in such a way that our | |
4522 | * allocation can fail while the mask is being updated. If we are about | |
4523 | * to fail, check if the cpuset changed during allocation and if so, | |
4524 | * retry. | |
4525 | */ | |
4526 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4527 | return true; | |
4528 | ||
4529 | return false; | |
4530 | } | |
4531 | ||
11e33f6a MG |
4532 | static inline struct page * |
4533 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4534 | struct alloc_context *ac) |
11e33f6a | 4535 | { |
d0164adc | 4536 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4537 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4538 | struct page *page = NULL; |
c603844b | 4539 | unsigned int alloc_flags; |
11e33f6a | 4540 | unsigned long did_some_progress; |
5ce9bfef | 4541 | enum compact_priority compact_priority; |
c5d01d0d | 4542 | enum compact_result compact_result; |
5ce9bfef VB |
4543 | int compaction_retries; |
4544 | int no_progress_loops; | |
5ce9bfef | 4545 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4546 | int reserve_flags; |
1da177e4 | 4547 | |
d0164adc MG |
4548 | /* |
4549 | * We also sanity check to catch abuse of atomic reserves being used by | |
4550 | * callers that are not in atomic context. | |
4551 | */ | |
4552 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4553 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4554 | gfp_mask &= ~__GFP_ATOMIC; | |
4555 | ||
5ce9bfef VB |
4556 | retry_cpuset: |
4557 | compaction_retries = 0; | |
4558 | no_progress_loops = 0; | |
4559 | compact_priority = DEF_COMPACT_PRIORITY; | |
4560 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4561 | |
4562 | /* | |
4563 | * The fast path uses conservative alloc_flags to succeed only until | |
4564 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4565 | * alloc_flags precisely. So we do that now. | |
4566 | */ | |
4567 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4568 | ||
e47483bc VB |
4569 | /* |
4570 | * We need to recalculate the starting point for the zonelist iterator | |
4571 | * because we might have used different nodemask in the fast path, or | |
4572 | * there was a cpuset modification and we are retrying - otherwise we | |
4573 | * could end up iterating over non-eligible zones endlessly. | |
4574 | */ | |
4575 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
97a225e6 | 4576 | ac->highest_zoneidx, ac->nodemask); |
e47483bc VB |
4577 | if (!ac->preferred_zoneref->zone) |
4578 | goto nopage; | |
4579 | ||
0a79cdad | 4580 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4581 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4582 | |
4583 | /* | |
4584 | * The adjusted alloc_flags might result in immediate success, so try | |
4585 | * that first | |
4586 | */ | |
4587 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4588 | if (page) | |
4589 | goto got_pg; | |
4590 | ||
a8161d1e VB |
4591 | /* |
4592 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4593 | * that we have enough base pages and don't need to reclaim. For non- |
4594 | * movable high-order allocations, do that as well, as compaction will | |
4595 | * try prevent permanent fragmentation by migrating from blocks of the | |
4596 | * same migratetype. | |
4597 | * Don't try this for allocations that are allowed to ignore | |
4598 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4599 | */ |
282722b0 VB |
4600 | if (can_direct_reclaim && |
4601 | (costly_order || | |
4602 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4603 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4604 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4605 | alloc_flags, ac, | |
a5508cd8 | 4606 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4607 | &compact_result); |
4608 | if (page) | |
4609 | goto got_pg; | |
4610 | ||
cc638f32 VB |
4611 | /* |
4612 | * Checks for costly allocations with __GFP_NORETRY, which | |
4613 | * includes some THP page fault allocations | |
4614 | */ | |
4615 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4616 | /* |
4617 | * If allocating entire pageblock(s) and compaction | |
4618 | * failed because all zones are below low watermarks | |
4619 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4620 | * order, fail immediately unless the allocator has |
4621 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4622 | * |
4623 | * Reclaim is | |
4624 | * - potentially very expensive because zones are far | |
4625 | * below their low watermarks or this is part of very | |
4626 | * bursty high order allocations, | |
4627 | * - not guaranteed to help because isolate_freepages() | |
4628 | * may not iterate over freed pages as part of its | |
4629 | * linear scan, and | |
4630 | * - unlikely to make entire pageblocks free on its | |
4631 | * own. | |
4632 | */ | |
4633 | if (compact_result == COMPACT_SKIPPED || | |
4634 | compact_result == COMPACT_DEFERRED) | |
4635 | goto nopage; | |
a8161d1e | 4636 | |
a8161d1e | 4637 | /* |
3eb2771b VB |
4638 | * Looks like reclaim/compaction is worth trying, but |
4639 | * sync compaction could be very expensive, so keep | |
25160354 | 4640 | * using async compaction. |
a8161d1e | 4641 | */ |
a5508cd8 | 4642 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4643 | } |
4644 | } | |
23771235 | 4645 | |
31a6c190 | 4646 | retry: |
23771235 | 4647 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4648 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4649 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4650 | |
cd04ae1e MH |
4651 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4652 | if (reserve_flags) | |
8510e69c | 4653 | alloc_flags = current_alloc_flags(gfp_mask, reserve_flags); |
23771235 | 4654 | |
e46e7b77 | 4655 | /* |
d6a24df0 VB |
4656 | * Reset the nodemask and zonelist iterators if memory policies can be |
4657 | * ignored. These allocations are high priority and system rather than | |
4658 | * user oriented. | |
e46e7b77 | 4659 | */ |
cd04ae1e | 4660 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4661 | ac->nodemask = NULL; |
e46e7b77 | 4662 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4663 | ac->highest_zoneidx, ac->nodemask); |
e46e7b77 MG |
4664 | } |
4665 | ||
23771235 | 4666 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4667 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4668 | if (page) |
4669 | goto got_pg; | |
1da177e4 | 4670 | |
d0164adc | 4671 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4672 | if (!can_direct_reclaim) |
1da177e4 LT |
4673 | goto nopage; |
4674 | ||
9a67f648 MH |
4675 | /* Avoid recursion of direct reclaim */ |
4676 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4677 | goto nopage; |
4678 | ||
a8161d1e VB |
4679 | /* Try direct reclaim and then allocating */ |
4680 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4681 | &did_some_progress); | |
4682 | if (page) | |
4683 | goto got_pg; | |
4684 | ||
4685 | /* Try direct compaction and then allocating */ | |
a9263751 | 4686 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4687 | compact_priority, &compact_result); |
56de7263 MG |
4688 | if (page) |
4689 | goto got_pg; | |
75f30861 | 4690 | |
9083905a JW |
4691 | /* Do not loop if specifically requested */ |
4692 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4693 | goto nopage; |
9083905a | 4694 | |
0a0337e0 MH |
4695 | /* |
4696 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4697 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4698 | */ |
dcda9b04 | 4699 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4700 | goto nopage; |
0a0337e0 | 4701 | |
0a0337e0 | 4702 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4703 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4704 | goto retry; |
4705 | ||
33c2d214 MH |
4706 | /* |
4707 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4708 | * reclaim is not able to make any progress because the current | |
4709 | * implementation of the compaction depends on the sufficient amount | |
4710 | * of free memory (see __compaction_suitable) | |
4711 | */ | |
4712 | if (did_some_progress > 0 && | |
86a294a8 | 4713 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4714 | compact_result, &compact_priority, |
d9436498 | 4715 | &compaction_retries)) |
33c2d214 MH |
4716 | goto retry; |
4717 | ||
902b6281 VB |
4718 | |
4719 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4720 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4721 | goto retry_cpuset; |
4722 | ||
9083905a JW |
4723 | /* Reclaim has failed us, start killing things */ |
4724 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4725 | if (page) | |
4726 | goto got_pg; | |
4727 | ||
9a67f648 | 4728 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e | 4729 | if (tsk_is_oom_victim(current) && |
8510e69c | 4730 | (alloc_flags & ALLOC_OOM || |
c288983d | 4731 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4732 | goto nopage; |
4733 | ||
9083905a | 4734 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4735 | if (did_some_progress) { |
4736 | no_progress_loops = 0; | |
9083905a | 4737 | goto retry; |
0a0337e0 | 4738 | } |
9083905a | 4739 | |
1da177e4 | 4740 | nopage: |
902b6281 VB |
4741 | /* Deal with possible cpuset update races before we fail */ |
4742 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4743 | goto retry_cpuset; |
4744 | ||
9a67f648 MH |
4745 | /* |
4746 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4747 | * we always retry | |
4748 | */ | |
4749 | if (gfp_mask & __GFP_NOFAIL) { | |
4750 | /* | |
4751 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4752 | * of any new users that actually require GFP_NOWAIT | |
4753 | */ | |
4754 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4755 | goto fail; | |
4756 | ||
4757 | /* | |
4758 | * PF_MEMALLOC request from this context is rather bizarre | |
4759 | * because we cannot reclaim anything and only can loop waiting | |
4760 | * for somebody to do a work for us | |
4761 | */ | |
4762 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4763 | ||
4764 | /* | |
4765 | * non failing costly orders are a hard requirement which we | |
4766 | * are not prepared for much so let's warn about these users | |
4767 | * so that we can identify them and convert them to something | |
4768 | * else. | |
4769 | */ | |
4770 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4771 | ||
6c18ba7a MH |
4772 | /* |
4773 | * Help non-failing allocations by giving them access to memory | |
4774 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4775 | * could deplete whole memory reserves which would just make | |
4776 | * the situation worse | |
4777 | */ | |
4778 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4779 | if (page) | |
4780 | goto got_pg; | |
4781 | ||
9a67f648 MH |
4782 | cond_resched(); |
4783 | goto retry; | |
4784 | } | |
4785 | fail: | |
a8e99259 | 4786 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4787 | "page allocation failure: order:%u", order); |
1da177e4 | 4788 | got_pg: |
072bb0aa | 4789 | return page; |
1da177e4 | 4790 | } |
11e33f6a | 4791 | |
9cd75558 | 4792 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4793 | int preferred_nid, nodemask_t *nodemask, |
9cd75558 MG |
4794 | struct alloc_context *ac, gfp_t *alloc_mask, |
4795 | unsigned int *alloc_flags) | |
11e33f6a | 4796 | { |
97a225e6 | 4797 | ac->highest_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4798 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 | 4799 | ac->nodemask = nodemask; |
01c0bfe0 | 4800 | ac->migratetype = gfp_migratetype(gfp_mask); |
11e33f6a | 4801 | |
682a3385 | 4802 | if (cpusets_enabled()) { |
9cd75558 | 4803 | *alloc_mask |= __GFP_HARDWALL; |
182f3d7a MS |
4804 | /* |
4805 | * When we are in the interrupt context, it is irrelevant | |
4806 | * to the current task context. It means that any node ok. | |
4807 | */ | |
4808 | if (!in_interrupt() && !ac->nodemask) | |
9cd75558 | 4809 | ac->nodemask = &cpuset_current_mems_allowed; |
51047820 VB |
4810 | else |
4811 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4812 | } |
4813 | ||
d92a8cfc PZ |
4814 | fs_reclaim_acquire(gfp_mask); |
4815 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4816 | |
d0164adc | 4817 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4818 | |
4819 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4820 | return false; |
11e33f6a | 4821 | |
8510e69c | 4822 | *alloc_flags = current_alloc_flags(gfp_mask, *alloc_flags); |
d883c6cf | 4823 | |
9cd75558 MG |
4824 | return true; |
4825 | } | |
21bb9bd1 | 4826 | |
9cd75558 | 4827 | /* Determine whether to spread dirty pages and what the first usable zone */ |
a380b40a | 4828 | static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac) |
9cd75558 | 4829 | { |
c9ab0c4f | 4830 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4831 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4832 | |
e46e7b77 MG |
4833 | /* |
4834 | * The preferred zone is used for statistics but crucially it is | |
4835 | * also used as the starting point for the zonelist iterator. It | |
4836 | * may get reset for allocations that ignore memory policies. | |
4837 | */ | |
9cd75558 | 4838 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
97a225e6 | 4839 | ac->highest_zoneidx, ac->nodemask); |
9cd75558 MG |
4840 | } |
4841 | ||
4842 | /* | |
4843 | * This is the 'heart' of the zoned buddy allocator. | |
4844 | */ | |
4845 | struct page * | |
04ec6264 VB |
4846 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
4847 | nodemask_t *nodemask) | |
9cd75558 MG |
4848 | { |
4849 | struct page *page; | |
4850 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
f19360f0 | 4851 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4852 | struct alloc_context ac = { }; |
4853 | ||
c63ae43b MH |
4854 | /* |
4855 | * There are several places where we assume that the order value is sane | |
4856 | * so bail out early if the request is out of bound. | |
4857 | */ | |
4858 | if (unlikely(order >= MAX_ORDER)) { | |
4859 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
4860 | return NULL; | |
4861 | } | |
4862 | ||
9cd75558 | 4863 | gfp_mask &= gfp_allowed_mask; |
f19360f0 | 4864 | alloc_mask = gfp_mask; |
04ec6264 | 4865 | if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags)) |
9cd75558 MG |
4866 | return NULL; |
4867 | ||
a380b40a | 4868 | finalise_ac(gfp_mask, &ac); |
5bb1b169 | 4869 | |
6bb15450 MG |
4870 | /* |
4871 | * Forbid the first pass from falling back to types that fragment | |
4872 | * memory until all local zones are considered. | |
4873 | */ | |
0a79cdad | 4874 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask); |
6bb15450 | 4875 | |
5117f45d | 4876 | /* First allocation attempt */ |
a9263751 | 4877 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4878 | if (likely(page)) |
4879 | goto out; | |
11e33f6a | 4880 | |
4fcb0971 | 4881 | /* |
7dea19f9 MH |
4882 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4883 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4884 | * from a particular context which has been marked by | |
4885 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4886 | */ |
7dea19f9 | 4887 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4888 | ac.spread_dirty_pages = false; |
23f086f9 | 4889 | |
4741526b MG |
4890 | /* |
4891 | * Restore the original nodemask if it was potentially replaced with | |
4892 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4893 | */ | |
97ce86f9 | 4894 | ac.nodemask = nodemask; |
16096c25 | 4895 | |
4fcb0971 | 4896 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4897 | |
4fcb0971 | 4898 | out: |
c4159a75 | 4899 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
f4b00eab | 4900 | unlikely(__memcg_kmem_charge_page(page, gfp_mask, order) != 0)) { |
c4159a75 VD |
4901 | __free_pages(page, order); |
4902 | page = NULL; | |
4949148a VD |
4903 | } |
4904 | ||
4fcb0971 MG |
4905 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
4906 | ||
11e33f6a | 4907 | return page; |
1da177e4 | 4908 | } |
d239171e | 4909 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4910 | |
4911 | /* | |
9ea9a680 MH |
4912 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4913 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4914 | * you need to access high mem. | |
1da177e4 | 4915 | */ |
920c7a5d | 4916 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4917 | { |
945a1113 AM |
4918 | struct page *page; |
4919 | ||
9ea9a680 | 4920 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4921 | if (!page) |
4922 | return 0; | |
4923 | return (unsigned long) page_address(page); | |
4924 | } | |
1da177e4 LT |
4925 | EXPORT_SYMBOL(__get_free_pages); |
4926 | ||
920c7a5d | 4927 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4928 | { |
945a1113 | 4929 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4930 | } |
1da177e4 LT |
4931 | EXPORT_SYMBOL(get_zeroed_page); |
4932 | ||
742aa7fb | 4933 | static inline void free_the_page(struct page *page, unsigned int order) |
1da177e4 | 4934 | { |
742aa7fb AL |
4935 | if (order == 0) /* Via pcp? */ |
4936 | free_unref_page(page); | |
4937 | else | |
4938 | __free_pages_ok(page, order); | |
1da177e4 LT |
4939 | } |
4940 | ||
742aa7fb AL |
4941 | void __free_pages(struct page *page, unsigned int order) |
4942 | { | |
4943 | if (put_page_testzero(page)) | |
4944 | free_the_page(page, order); | |
4945 | } | |
1da177e4 LT |
4946 | EXPORT_SYMBOL(__free_pages); |
4947 | ||
920c7a5d | 4948 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4949 | { |
4950 | if (addr != 0) { | |
725d704e | 4951 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4952 | __free_pages(virt_to_page((void *)addr), order); |
4953 | } | |
4954 | } | |
4955 | ||
4956 | EXPORT_SYMBOL(free_pages); | |
4957 | ||
b63ae8ca AD |
4958 | /* |
4959 | * Page Fragment: | |
4960 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4961 | * within a 0 or higher order page. Multiple fragments within that page | |
4962 | * are individually refcounted, in the page's reference counter. | |
4963 | * | |
4964 | * The page_frag functions below provide a simple allocation framework for | |
4965 | * page fragments. This is used by the network stack and network device | |
4966 | * drivers to provide a backing region of memory for use as either an | |
4967 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4968 | */ | |
2976db80 AD |
4969 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4970 | gfp_t gfp_mask) | |
b63ae8ca AD |
4971 | { |
4972 | struct page *page = NULL; | |
4973 | gfp_t gfp = gfp_mask; | |
4974 | ||
4975 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4976 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4977 | __GFP_NOMEMALLOC; | |
4978 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4979 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4980 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4981 | #endif | |
4982 | if (unlikely(!page)) | |
4983 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4984 | ||
4985 | nc->va = page ? page_address(page) : NULL; | |
4986 | ||
4987 | return page; | |
4988 | } | |
4989 | ||
2976db80 | 4990 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4991 | { |
4992 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4993 | ||
742aa7fb AL |
4994 | if (page_ref_sub_and_test(page, count)) |
4995 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 4996 | } |
2976db80 | 4997 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4998 | |
8c2dd3e4 AD |
4999 | void *page_frag_alloc(struct page_frag_cache *nc, |
5000 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
5001 | { |
5002 | unsigned int size = PAGE_SIZE; | |
5003 | struct page *page; | |
5004 | int offset; | |
5005 | ||
5006 | if (unlikely(!nc->va)) { | |
5007 | refill: | |
2976db80 | 5008 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
5009 | if (!page) |
5010 | return NULL; | |
5011 | ||
5012 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5013 | /* if size can vary use size else just use PAGE_SIZE */ | |
5014 | size = nc->size; | |
5015 | #endif | |
5016 | /* Even if we own the page, we do not use atomic_set(). | |
5017 | * This would break get_page_unless_zero() users. | |
5018 | */ | |
86447726 | 5019 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
5020 | |
5021 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 5022 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 5023 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5024 | nc->offset = size; |
5025 | } | |
5026 | ||
5027 | offset = nc->offset - fragsz; | |
5028 | if (unlikely(offset < 0)) { | |
5029 | page = virt_to_page(nc->va); | |
5030 | ||
fe896d18 | 5031 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
5032 | goto refill; |
5033 | ||
5034 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
5035 | /* if size can vary use size else just use PAGE_SIZE */ | |
5036 | size = nc->size; | |
5037 | #endif | |
5038 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 5039 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
5040 | |
5041 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 5042 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
5043 | offset = size - fragsz; |
5044 | } | |
5045 | ||
5046 | nc->pagecnt_bias--; | |
5047 | nc->offset = offset; | |
5048 | ||
5049 | return nc->va + offset; | |
5050 | } | |
8c2dd3e4 | 5051 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
5052 | |
5053 | /* | |
5054 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
5055 | */ | |
8c2dd3e4 | 5056 | void page_frag_free(void *addr) |
b63ae8ca AD |
5057 | { |
5058 | struct page *page = virt_to_head_page(addr); | |
5059 | ||
742aa7fb AL |
5060 | if (unlikely(put_page_testzero(page))) |
5061 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 5062 | } |
8c2dd3e4 | 5063 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 5064 | |
d00181b9 KS |
5065 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
5066 | size_t size) | |
ee85c2e1 AK |
5067 | { |
5068 | if (addr) { | |
5069 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
5070 | unsigned long used = addr + PAGE_ALIGN(size); | |
5071 | ||
5072 | split_page(virt_to_page((void *)addr), order); | |
5073 | while (used < alloc_end) { | |
5074 | free_page(used); | |
5075 | used += PAGE_SIZE; | |
5076 | } | |
5077 | } | |
5078 | return (void *)addr; | |
5079 | } | |
5080 | ||
2be0ffe2 TT |
5081 | /** |
5082 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
5083 | * @size: the number of bytes to allocate | |
63931eb9 | 5084 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
5085 | * |
5086 | * This function is similar to alloc_pages(), except that it allocates the | |
5087 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
5088 | * allocate memory in power-of-two pages. | |
5089 | * | |
5090 | * This function is also limited by MAX_ORDER. | |
5091 | * | |
5092 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
5093 | * |
5094 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
5095 | */ |
5096 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
5097 | { | |
5098 | unsigned int order = get_order(size); | |
5099 | unsigned long addr; | |
5100 | ||
63931eb9 VB |
5101 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) |
5102 | gfp_mask &= ~__GFP_COMP; | |
5103 | ||
2be0ffe2 | 5104 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 5105 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
5106 | } |
5107 | EXPORT_SYMBOL(alloc_pages_exact); | |
5108 | ||
ee85c2e1 AK |
5109 | /** |
5110 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
5111 | * pages on a node. | |
b5e6ab58 | 5112 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 5113 | * @size: the number of bytes to allocate |
63931eb9 | 5114 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
5115 | * |
5116 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
5117 | * back. | |
a862f68a MR |
5118 | * |
5119 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 5120 | */ |
e1931811 | 5121 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5122 | { |
d00181b9 | 5123 | unsigned int order = get_order(size); |
63931eb9 VB |
5124 | struct page *p; |
5125 | ||
5126 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) | |
5127 | gfp_mask &= ~__GFP_COMP; | |
5128 | ||
5129 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5130 | if (!p) |
5131 | return NULL; | |
5132 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5133 | } | |
ee85c2e1 | 5134 | |
2be0ffe2 TT |
5135 | /** |
5136 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5137 | * @virt: the value returned by alloc_pages_exact. | |
5138 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5139 | * | |
5140 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5141 | */ | |
5142 | void free_pages_exact(void *virt, size_t size) | |
5143 | { | |
5144 | unsigned long addr = (unsigned long)virt; | |
5145 | unsigned long end = addr + PAGE_ALIGN(size); | |
5146 | ||
5147 | while (addr < end) { | |
5148 | free_page(addr); | |
5149 | addr += PAGE_SIZE; | |
5150 | } | |
5151 | } | |
5152 | EXPORT_SYMBOL(free_pages_exact); | |
5153 | ||
e0fb5815 ZY |
5154 | /** |
5155 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5156 | * @offset: The zone index of the highest zone | |
5157 | * | |
a862f68a | 5158 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5159 | * high watermark within all zones at or below a given zone index. For each |
5160 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
5161 | * |
5162 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5163 | * |
5164 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5165 | */ |
ebec3862 | 5166 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5167 | { |
dd1a239f | 5168 | struct zoneref *z; |
54a6eb5c MG |
5169 | struct zone *zone; |
5170 | ||
e310fd43 | 5171 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5172 | unsigned long sum = 0; |
1da177e4 | 5173 | |
0e88460d | 5174 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5175 | |
54a6eb5c | 5176 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5177 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5178 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5179 | if (size > high) |
5180 | sum += size - high; | |
1da177e4 LT |
5181 | } |
5182 | ||
5183 | return sum; | |
5184 | } | |
5185 | ||
e0fb5815 ZY |
5186 | /** |
5187 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5188 | * | |
5189 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5190 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5191 | * |
5192 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5193 | * ZONE_NORMAL. | |
1da177e4 | 5194 | */ |
ebec3862 | 5195 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5196 | { |
af4ca457 | 5197 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5198 | } |
c2f1a551 | 5199 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5200 | |
08e0f6a9 | 5201 | static inline void show_node(struct zone *zone) |
1da177e4 | 5202 | { |
e5adfffc | 5203 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5204 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5205 | } |
1da177e4 | 5206 | |
d02bd27b IR |
5207 | long si_mem_available(void) |
5208 | { | |
5209 | long available; | |
5210 | unsigned long pagecache; | |
5211 | unsigned long wmark_low = 0; | |
5212 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5213 | unsigned long reclaimable; |
d02bd27b IR |
5214 | struct zone *zone; |
5215 | int lru; | |
5216 | ||
5217 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5218 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5219 | |
5220 | for_each_zone(zone) | |
a9214443 | 5221 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5222 | |
5223 | /* | |
5224 | * Estimate the amount of memory available for userspace allocations, | |
5225 | * without causing swapping. | |
5226 | */ | |
c41f012a | 5227 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5228 | |
5229 | /* | |
5230 | * Not all the page cache can be freed, otherwise the system will | |
5231 | * start swapping. Assume at least half of the page cache, or the | |
5232 | * low watermark worth of cache, needs to stay. | |
5233 | */ | |
5234 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
5235 | pagecache -= min(pagecache / 2, wmark_low); | |
5236 | available += pagecache; | |
5237 | ||
5238 | /* | |
b29940c1 VB |
5239 | * Part of the reclaimable slab and other kernel memory consists of |
5240 | * items that are in use, and cannot be freed. Cap this estimate at the | |
5241 | * low watermark. | |
d02bd27b | 5242 | */ |
d42f3245 RG |
5243 | reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) + |
5244 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
b29940c1 | 5245 | available += reclaimable - min(reclaimable / 2, wmark_low); |
034ebf65 | 5246 | |
d02bd27b IR |
5247 | if (available < 0) |
5248 | available = 0; | |
5249 | return available; | |
5250 | } | |
5251 | EXPORT_SYMBOL_GPL(si_mem_available); | |
5252 | ||
1da177e4 LT |
5253 | void si_meminfo(struct sysinfo *val) |
5254 | { | |
ca79b0c2 | 5255 | val->totalram = totalram_pages(); |
11fb9989 | 5256 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 5257 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 5258 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 5259 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 5260 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
5261 | val->mem_unit = PAGE_SIZE; |
5262 | } | |
5263 | ||
5264 | EXPORT_SYMBOL(si_meminfo); | |
5265 | ||
5266 | #ifdef CONFIG_NUMA | |
5267 | void si_meminfo_node(struct sysinfo *val, int nid) | |
5268 | { | |
cdd91a77 JL |
5269 | int zone_type; /* needs to be signed */ |
5270 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
5271 | unsigned long managed_highpages = 0; |
5272 | unsigned long free_highpages = 0; | |
1da177e4 LT |
5273 | pg_data_t *pgdat = NODE_DATA(nid); |
5274 | ||
cdd91a77 | 5275 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 5276 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 5277 | val->totalram = managed_pages; |
11fb9989 | 5278 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 5279 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 5280 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
5281 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
5282 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
5283 | ||
5284 | if (is_highmem(zone)) { | |
9705bea5 | 5285 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
5286 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
5287 | } | |
5288 | } | |
5289 | val->totalhigh = managed_highpages; | |
5290 | val->freehigh = free_highpages; | |
98d2b0eb | 5291 | #else |
fc2bd799 JK |
5292 | val->totalhigh = managed_highpages; |
5293 | val->freehigh = free_highpages; | |
98d2b0eb | 5294 | #endif |
1da177e4 LT |
5295 | val->mem_unit = PAGE_SIZE; |
5296 | } | |
5297 | #endif | |
5298 | ||
ddd588b5 | 5299 | /* |
7bf02ea2 DR |
5300 | * Determine whether the node should be displayed or not, depending on whether |
5301 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 5302 | */ |
9af744d7 | 5303 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 5304 | { |
ddd588b5 | 5305 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 5306 | return false; |
ddd588b5 | 5307 | |
9af744d7 MH |
5308 | /* |
5309 | * no node mask - aka implicit memory numa policy. Do not bother with | |
5310 | * the synchronization - read_mems_allowed_begin - because we do not | |
5311 | * have to be precise here. | |
5312 | */ | |
5313 | if (!nodemask) | |
5314 | nodemask = &cpuset_current_mems_allowed; | |
5315 | ||
5316 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
5317 | } |
5318 | ||
1da177e4 LT |
5319 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
5320 | ||
377e4f16 RV |
5321 | static void show_migration_types(unsigned char type) |
5322 | { | |
5323 | static const char types[MIGRATE_TYPES] = { | |
5324 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 5325 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
5326 | [MIGRATE_RECLAIMABLE] = 'E', |
5327 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
5328 | #ifdef CONFIG_CMA |
5329 | [MIGRATE_CMA] = 'C', | |
5330 | #endif | |
194159fb | 5331 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 5332 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 5333 | #endif |
377e4f16 RV |
5334 | }; |
5335 | char tmp[MIGRATE_TYPES + 1]; | |
5336 | char *p = tmp; | |
5337 | int i; | |
5338 | ||
5339 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
5340 | if (type & (1 << i)) | |
5341 | *p++ = types[i]; | |
5342 | } | |
5343 | ||
5344 | *p = '\0'; | |
1f84a18f | 5345 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
5346 | } |
5347 | ||
1da177e4 LT |
5348 | /* |
5349 | * Show free area list (used inside shift_scroll-lock stuff) | |
5350 | * We also calculate the percentage fragmentation. We do this by counting the | |
5351 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
5352 | * |
5353 | * Bits in @filter: | |
5354 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
5355 | * cpuset. | |
1da177e4 | 5356 | */ |
9af744d7 | 5357 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 5358 | { |
d1bfcdb8 | 5359 | unsigned long free_pcp = 0; |
c7241913 | 5360 | int cpu; |
1da177e4 | 5361 | struct zone *zone; |
599d0c95 | 5362 | pg_data_t *pgdat; |
1da177e4 | 5363 | |
ee99c71c | 5364 | for_each_populated_zone(zone) { |
9af744d7 | 5365 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5366 | continue; |
d1bfcdb8 | 5367 | |
761b0677 KK |
5368 | for_each_online_cpu(cpu) |
5369 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
5370 | } |
5371 | ||
a731286d KM |
5372 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
5373 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
8d92890b | 5374 | " unevictable:%lu dirty:%lu writeback:%lu\n" |
d1bfcdb8 | 5375 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a | 5376 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 5377 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
5378 | global_node_page_state(NR_ACTIVE_ANON), |
5379 | global_node_page_state(NR_INACTIVE_ANON), | |
5380 | global_node_page_state(NR_ISOLATED_ANON), | |
5381 | global_node_page_state(NR_ACTIVE_FILE), | |
5382 | global_node_page_state(NR_INACTIVE_FILE), | |
5383 | global_node_page_state(NR_ISOLATED_FILE), | |
5384 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
5385 | global_node_page_state(NR_FILE_DIRTY), |
5386 | global_node_page_state(NR_WRITEBACK), | |
d42f3245 RG |
5387 | global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B), |
5388 | global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B), | |
50658e2e | 5389 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 5390 | global_node_page_state(NR_SHMEM), |
c41f012a MH |
5391 | global_zone_page_state(NR_PAGETABLE), |
5392 | global_zone_page_state(NR_BOUNCE), | |
5393 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 5394 | free_pcp, |
c41f012a | 5395 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 5396 | |
599d0c95 | 5397 | for_each_online_pgdat(pgdat) { |
9af744d7 | 5398 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
5399 | continue; |
5400 | ||
599d0c95 MG |
5401 | printk("Node %d" |
5402 | " active_anon:%lukB" | |
5403 | " inactive_anon:%lukB" | |
5404 | " active_file:%lukB" | |
5405 | " inactive_file:%lukB" | |
5406 | " unevictable:%lukB" | |
5407 | " isolated(anon):%lukB" | |
5408 | " isolated(file):%lukB" | |
50658e2e | 5409 | " mapped:%lukB" |
11fb9989 MG |
5410 | " dirty:%lukB" |
5411 | " writeback:%lukB" | |
5412 | " shmem:%lukB" | |
5413 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5414 | " shmem_thp: %lukB" | |
5415 | " shmem_pmdmapped: %lukB" | |
5416 | " anon_thp: %lukB" | |
5417 | #endif | |
5418 | " writeback_tmp:%lukB" | |
991e7673 SB |
5419 | " kernel_stack:%lukB" |
5420 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5421 | " shadow_call_stack:%lukB" | |
5422 | #endif | |
599d0c95 MG |
5423 | " all_unreclaimable? %s" |
5424 | "\n", | |
5425 | pgdat->node_id, | |
5426 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
5427 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
5428 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
5429 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
5430 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
5431 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
5432 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 5433 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
5434 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
5435 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 5436 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
5437 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5438 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
5439 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
5440 | * HPAGE_PMD_NR), | |
5441 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
5442 | #endif | |
11fb9989 | 5443 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
991e7673 SB |
5444 | node_page_state(pgdat, NR_KERNEL_STACK_KB), |
5445 | #ifdef CONFIG_SHADOW_CALL_STACK | |
5446 | node_page_state(pgdat, NR_KERNEL_SCS_KB), | |
5447 | #endif | |
c73322d0 JW |
5448 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
5449 | "yes" : "no"); | |
599d0c95 MG |
5450 | } |
5451 | ||
ee99c71c | 5452 | for_each_populated_zone(zone) { |
1da177e4 LT |
5453 | int i; |
5454 | ||
9af744d7 | 5455 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5456 | continue; |
d1bfcdb8 KK |
5457 | |
5458 | free_pcp = 0; | |
5459 | for_each_online_cpu(cpu) | |
5460 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
5461 | ||
1da177e4 | 5462 | show_node(zone); |
1f84a18f JP |
5463 | printk(KERN_CONT |
5464 | "%s" | |
1da177e4 LT |
5465 | " free:%lukB" |
5466 | " min:%lukB" | |
5467 | " low:%lukB" | |
5468 | " high:%lukB" | |
e47b346a | 5469 | " reserved_highatomic:%luKB" |
71c799f4 MK |
5470 | " active_anon:%lukB" |
5471 | " inactive_anon:%lukB" | |
5472 | " active_file:%lukB" | |
5473 | " inactive_file:%lukB" | |
5474 | " unevictable:%lukB" | |
5a1c84b4 | 5475 | " writepending:%lukB" |
1da177e4 | 5476 | " present:%lukB" |
9feedc9d | 5477 | " managed:%lukB" |
4a0aa73f | 5478 | " mlocked:%lukB" |
4a0aa73f | 5479 | " pagetables:%lukB" |
4a0aa73f | 5480 | " bounce:%lukB" |
d1bfcdb8 KK |
5481 | " free_pcp:%lukB" |
5482 | " local_pcp:%ukB" | |
d1ce749a | 5483 | " free_cma:%lukB" |
1da177e4 LT |
5484 | "\n", |
5485 | zone->name, | |
88f5acf8 | 5486 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
5487 | K(min_wmark_pages(zone)), |
5488 | K(low_wmark_pages(zone)), | |
5489 | K(high_wmark_pages(zone)), | |
e47b346a | 5490 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
5491 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
5492 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
5493 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
5494 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
5495 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 5496 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 5497 | K(zone->present_pages), |
9705bea5 | 5498 | K(zone_managed_pages(zone)), |
4a0aa73f | 5499 | K(zone_page_state(zone, NR_MLOCK)), |
4a0aa73f | 5500 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 5501 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
5502 | K(free_pcp), |
5503 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 5504 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
5505 | printk("lowmem_reserve[]:"); |
5506 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
5507 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5508 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5509 | } |
5510 | ||
ee99c71c | 5511 | for_each_populated_zone(zone) { |
d00181b9 KS |
5512 | unsigned int order; |
5513 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5514 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5515 | |
9af744d7 | 5516 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5517 | continue; |
1da177e4 | 5518 | show_node(zone); |
1f84a18f | 5519 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5520 | |
5521 | spin_lock_irqsave(&zone->lock, flags); | |
5522 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5523 | struct free_area *area = &zone->free_area[order]; |
5524 | int type; | |
5525 | ||
5526 | nr[order] = area->nr_free; | |
8f9de51a | 5527 | total += nr[order] << order; |
377e4f16 RV |
5528 | |
5529 | types[order] = 0; | |
5530 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 5531 | if (!free_area_empty(area, type)) |
377e4f16 RV |
5532 | types[order] |= 1 << type; |
5533 | } | |
1da177e4 LT |
5534 | } |
5535 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5536 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5537 | printk(KERN_CONT "%lu*%lukB ", |
5538 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5539 | if (nr[order]) |
5540 | show_migration_types(types[order]); | |
5541 | } | |
1f84a18f | 5542 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5543 | } |
5544 | ||
949f7ec5 DR |
5545 | hugetlb_show_meminfo(); |
5546 | ||
11fb9989 | 5547 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5548 | |
1da177e4 LT |
5549 | show_swap_cache_info(); |
5550 | } | |
5551 | ||
19770b32 MG |
5552 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5553 | { | |
5554 | zoneref->zone = zone; | |
5555 | zoneref->zone_idx = zone_idx(zone); | |
5556 | } | |
5557 | ||
1da177e4 LT |
5558 | /* |
5559 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5560 | * |
5561 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5562 | */ |
9d3be21b | 5563 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5564 | { |
1a93205b | 5565 | struct zone *zone; |
bc732f1d | 5566 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5567 | int nr_zones = 0; |
02a68a5e CL |
5568 | |
5569 | do { | |
2f6726e5 | 5570 | zone_type--; |
070f8032 | 5571 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5572 | if (managed_zone(zone)) { |
9d3be21b | 5573 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5574 | check_highest_zone(zone_type); |
1da177e4 | 5575 | } |
2f6726e5 | 5576 | } while (zone_type); |
bc732f1d | 5577 | |
070f8032 | 5578 | return nr_zones; |
1da177e4 LT |
5579 | } |
5580 | ||
5581 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5582 | |
5583 | static int __parse_numa_zonelist_order(char *s) | |
5584 | { | |
c9bff3ee MH |
5585 | /* |
5586 | * We used to support different zonlists modes but they turned | |
5587 | * out to be just not useful. Let's keep the warning in place | |
5588 | * if somebody still use the cmd line parameter so that we do | |
5589 | * not fail it silently | |
5590 | */ | |
5591 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5592 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5593 | return -EINVAL; |
5594 | } | |
5595 | return 0; | |
5596 | } | |
5597 | ||
c9bff3ee MH |
5598 | char numa_zonelist_order[] = "Node"; |
5599 | ||
f0c0b2b8 KH |
5600 | /* |
5601 | * sysctl handler for numa_zonelist_order | |
5602 | */ | |
cccad5b9 | 5603 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
32927393 | 5604 | void *buffer, size_t *length, loff_t *ppos) |
f0c0b2b8 | 5605 | { |
32927393 CH |
5606 | if (write) |
5607 | return __parse_numa_zonelist_order(buffer); | |
5608 | return proc_dostring(table, write, buffer, length, ppos); | |
f0c0b2b8 KH |
5609 | } |
5610 | ||
5611 | ||
62bc62a8 | 5612 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5613 | static int node_load[MAX_NUMNODES]; |
5614 | ||
1da177e4 | 5615 | /** |
4dc3b16b | 5616 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5617 | * @node: node whose fallback list we're appending |
5618 | * @used_node_mask: nodemask_t of already used nodes | |
5619 | * | |
5620 | * We use a number of factors to determine which is the next node that should | |
5621 | * appear on a given node's fallback list. The node should not have appeared | |
5622 | * already in @node's fallback list, and it should be the next closest node | |
5623 | * according to the distance array (which contains arbitrary distance values | |
5624 | * from each node to each node in the system), and should also prefer nodes | |
5625 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5626 | * on them otherwise. | |
a862f68a MR |
5627 | * |
5628 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 5629 | */ |
f0c0b2b8 | 5630 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5631 | { |
4cf808eb | 5632 | int n, val; |
1da177e4 | 5633 | int min_val = INT_MAX; |
00ef2d2f | 5634 | int best_node = NUMA_NO_NODE; |
a70f7302 | 5635 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 5636 | |
4cf808eb LT |
5637 | /* Use the local node if we haven't already */ |
5638 | if (!node_isset(node, *used_node_mask)) { | |
5639 | node_set(node, *used_node_mask); | |
5640 | return node; | |
5641 | } | |
1da177e4 | 5642 | |
4b0ef1fe | 5643 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5644 | |
5645 | /* Don't want a node to appear more than once */ | |
5646 | if (node_isset(n, *used_node_mask)) | |
5647 | continue; | |
5648 | ||
1da177e4 LT |
5649 | /* Use the distance array to find the distance */ |
5650 | val = node_distance(node, n); | |
5651 | ||
4cf808eb LT |
5652 | /* Penalize nodes under us ("prefer the next node") */ |
5653 | val += (n < node); | |
5654 | ||
1da177e4 | 5655 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
5656 | tmp = cpumask_of_node(n); |
5657 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
5658 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5659 | ||
5660 | /* Slight preference for less loaded node */ | |
5661 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
5662 | val += node_load[n]; | |
5663 | ||
5664 | if (val < min_val) { | |
5665 | min_val = val; | |
5666 | best_node = n; | |
5667 | } | |
5668 | } | |
5669 | ||
5670 | if (best_node >= 0) | |
5671 | node_set(best_node, *used_node_mask); | |
5672 | ||
5673 | return best_node; | |
5674 | } | |
5675 | ||
f0c0b2b8 KH |
5676 | |
5677 | /* | |
5678 | * Build zonelists ordered by node and zones within node. | |
5679 | * This results in maximum locality--normal zone overflows into local | |
5680 | * DMA zone, if any--but risks exhausting DMA zone. | |
5681 | */ | |
9d3be21b MH |
5682 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5683 | unsigned nr_nodes) | |
1da177e4 | 5684 | { |
9d3be21b MH |
5685 | struct zoneref *zonerefs; |
5686 | int i; | |
5687 | ||
5688 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5689 | ||
5690 | for (i = 0; i < nr_nodes; i++) { | |
5691 | int nr_zones; | |
5692 | ||
5693 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5694 | |
9d3be21b MH |
5695 | nr_zones = build_zonerefs_node(node, zonerefs); |
5696 | zonerefs += nr_zones; | |
5697 | } | |
5698 | zonerefs->zone = NULL; | |
5699 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5700 | } |
5701 | ||
523b9458 CL |
5702 | /* |
5703 | * Build gfp_thisnode zonelists | |
5704 | */ | |
5705 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5706 | { | |
9d3be21b MH |
5707 | struct zoneref *zonerefs; |
5708 | int nr_zones; | |
523b9458 | 5709 | |
9d3be21b MH |
5710 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5711 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5712 | zonerefs += nr_zones; | |
5713 | zonerefs->zone = NULL; | |
5714 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5715 | } |
5716 | ||
f0c0b2b8 KH |
5717 | /* |
5718 | * Build zonelists ordered by zone and nodes within zones. | |
5719 | * This results in conserving DMA zone[s] until all Normal memory is | |
5720 | * exhausted, but results in overflowing to remote node while memory | |
5721 | * may still exist in local DMA zone. | |
5722 | */ | |
f0c0b2b8 | 5723 | |
f0c0b2b8 KH |
5724 | static void build_zonelists(pg_data_t *pgdat) |
5725 | { | |
9d3be21b MH |
5726 | static int node_order[MAX_NUMNODES]; |
5727 | int node, load, nr_nodes = 0; | |
d0ddf49b | 5728 | nodemask_t used_mask = NODE_MASK_NONE; |
f0c0b2b8 | 5729 | int local_node, prev_node; |
1da177e4 LT |
5730 | |
5731 | /* NUMA-aware ordering of nodes */ | |
5732 | local_node = pgdat->node_id; | |
62bc62a8 | 5733 | load = nr_online_nodes; |
1da177e4 | 5734 | prev_node = local_node; |
f0c0b2b8 | 5735 | |
f0c0b2b8 | 5736 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5737 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5738 | /* | |
5739 | * We don't want to pressure a particular node. | |
5740 | * So adding penalty to the first node in same | |
5741 | * distance group to make it round-robin. | |
5742 | */ | |
957f822a DR |
5743 | if (node_distance(local_node, node) != |
5744 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5745 | node_load[node] = load; |
5746 | ||
9d3be21b | 5747 | node_order[nr_nodes++] = node; |
1da177e4 LT |
5748 | prev_node = node; |
5749 | load--; | |
1da177e4 | 5750 | } |
523b9458 | 5751 | |
9d3be21b | 5752 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5753 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
5754 | } |
5755 | ||
7aac7898 LS |
5756 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5757 | /* | |
5758 | * Return node id of node used for "local" allocations. | |
5759 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5760 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5761 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5762 | */ | |
5763 | int local_memory_node(int node) | |
5764 | { | |
c33d6c06 | 5765 | struct zoneref *z; |
7aac7898 | 5766 | |
c33d6c06 | 5767 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5768 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 5769 | NULL); |
c1093b74 | 5770 | return zone_to_nid(z->zone); |
7aac7898 LS |
5771 | } |
5772 | #endif | |
f0c0b2b8 | 5773 | |
6423aa81 JK |
5774 | static void setup_min_unmapped_ratio(void); |
5775 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5776 | #else /* CONFIG_NUMA */ |
5777 | ||
f0c0b2b8 | 5778 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5779 | { |
19655d34 | 5780 | int node, local_node; |
9d3be21b MH |
5781 | struct zoneref *zonerefs; |
5782 | int nr_zones; | |
1da177e4 LT |
5783 | |
5784 | local_node = pgdat->node_id; | |
1da177e4 | 5785 | |
9d3be21b MH |
5786 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5787 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5788 | zonerefs += nr_zones; | |
1da177e4 | 5789 | |
54a6eb5c MG |
5790 | /* |
5791 | * Now we build the zonelist so that it contains the zones | |
5792 | * of all the other nodes. | |
5793 | * We don't want to pressure a particular node, so when | |
5794 | * building the zones for node N, we make sure that the | |
5795 | * zones coming right after the local ones are those from | |
5796 | * node N+1 (modulo N) | |
5797 | */ | |
5798 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5799 | if (!node_online(node)) | |
5800 | continue; | |
9d3be21b MH |
5801 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5802 | zonerefs += nr_zones; | |
1da177e4 | 5803 | } |
54a6eb5c MG |
5804 | for (node = 0; node < local_node; node++) { |
5805 | if (!node_online(node)) | |
5806 | continue; | |
9d3be21b MH |
5807 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5808 | zonerefs += nr_zones; | |
54a6eb5c MG |
5809 | } |
5810 | ||
9d3be21b MH |
5811 | zonerefs->zone = NULL; |
5812 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5813 | } |
5814 | ||
5815 | #endif /* CONFIG_NUMA */ | |
5816 | ||
99dcc3e5 CL |
5817 | /* |
5818 | * Boot pageset table. One per cpu which is going to be used for all | |
5819 | * zones and all nodes. The parameters will be set in such a way | |
5820 | * that an item put on a list will immediately be handed over to | |
5821 | * the buddy list. This is safe since pageset manipulation is done | |
5822 | * with interrupts disabled. | |
5823 | * | |
5824 | * The boot_pagesets must be kept even after bootup is complete for | |
5825 | * unused processors and/or zones. They do play a role for bootstrapping | |
5826 | * hotplugged processors. | |
5827 | * | |
5828 | * zoneinfo_show() and maybe other functions do | |
5829 | * not check if the processor is online before following the pageset pointer. | |
5830 | * Other parts of the kernel may not check if the zone is available. | |
5831 | */ | |
5832 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5833 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
385386cf | 5834 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 5835 | |
11cd8638 | 5836 | static void __build_all_zonelists(void *data) |
1da177e4 | 5837 | { |
6811378e | 5838 | int nid; |
afb6ebb3 | 5839 | int __maybe_unused cpu; |
9adb62a5 | 5840 | pg_data_t *self = data; |
b93e0f32 MH |
5841 | static DEFINE_SPINLOCK(lock); |
5842 | ||
5843 | spin_lock(&lock); | |
9276b1bc | 5844 | |
7f9cfb31 BL |
5845 | #ifdef CONFIG_NUMA |
5846 | memset(node_load, 0, sizeof(node_load)); | |
5847 | #endif | |
9adb62a5 | 5848 | |
c1152583 WY |
5849 | /* |
5850 | * This node is hotadded and no memory is yet present. So just | |
5851 | * building zonelists is fine - no need to touch other nodes. | |
5852 | */ | |
9adb62a5 JL |
5853 | if (self && !node_online(self->node_id)) { |
5854 | build_zonelists(self); | |
c1152583 WY |
5855 | } else { |
5856 | for_each_online_node(nid) { | |
5857 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 5858 | |
c1152583 WY |
5859 | build_zonelists(pgdat); |
5860 | } | |
99dcc3e5 | 5861 | |
7aac7898 LS |
5862 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5863 | /* | |
5864 | * We now know the "local memory node" for each node-- | |
5865 | * i.e., the node of the first zone in the generic zonelist. | |
5866 | * Set up numa_mem percpu variable for on-line cpus. During | |
5867 | * boot, only the boot cpu should be on-line; we'll init the | |
5868 | * secondary cpus' numa_mem as they come on-line. During | |
5869 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5870 | */ | |
d9c9a0b9 | 5871 | for_each_online_cpu(cpu) |
7aac7898 | 5872 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5873 | #endif |
d9c9a0b9 | 5874 | } |
b93e0f32 MH |
5875 | |
5876 | spin_unlock(&lock); | |
6811378e YG |
5877 | } |
5878 | ||
061f67bc RV |
5879 | static noinline void __init |
5880 | build_all_zonelists_init(void) | |
5881 | { | |
afb6ebb3 MH |
5882 | int cpu; |
5883 | ||
061f67bc | 5884 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5885 | |
5886 | /* | |
5887 | * Initialize the boot_pagesets that are going to be used | |
5888 | * for bootstrapping processors. The real pagesets for | |
5889 | * each zone will be allocated later when the per cpu | |
5890 | * allocator is available. | |
5891 | * | |
5892 | * boot_pagesets are used also for bootstrapping offline | |
5893 | * cpus if the system is already booted because the pagesets | |
5894 | * are needed to initialize allocators on a specific cpu too. | |
5895 | * F.e. the percpu allocator needs the page allocator which | |
5896 | * needs the percpu allocator in order to allocate its pagesets | |
5897 | * (a chicken-egg dilemma). | |
5898 | */ | |
5899 | for_each_possible_cpu(cpu) | |
5900 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); | |
5901 | ||
061f67bc RV |
5902 | mminit_verify_zonelist(); |
5903 | cpuset_init_current_mems_allowed(); | |
5904 | } | |
5905 | ||
4eaf3f64 | 5906 | /* |
4eaf3f64 | 5907 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5908 | * |
72675e13 | 5909 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5910 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5911 | */ |
72675e13 | 5912 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e | 5913 | { |
0a18e607 DH |
5914 | unsigned long vm_total_pages; |
5915 | ||
6811378e | 5916 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 5917 | build_all_zonelists_init(); |
6811378e | 5918 | } else { |
11cd8638 | 5919 | __build_all_zonelists(pgdat); |
6811378e YG |
5920 | /* cpuset refresh routine should be here */ |
5921 | } | |
56b9413b DH |
5922 | /* Get the number of free pages beyond high watermark in all zones. */ |
5923 | vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
9ef9acb0 MG |
5924 | /* |
5925 | * Disable grouping by mobility if the number of pages in the | |
5926 | * system is too low to allow the mechanism to work. It would be | |
5927 | * more accurate, but expensive to check per-zone. This check is | |
5928 | * made on memory-hotadd so a system can start with mobility | |
5929 | * disabled and enable it later | |
5930 | */ | |
d9c23400 | 5931 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5932 | page_group_by_mobility_disabled = 1; |
5933 | else | |
5934 | page_group_by_mobility_disabled = 0; | |
5935 | ||
ce0725f7 | 5936 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5937 | nr_online_nodes, |
756a025f JP |
5938 | page_group_by_mobility_disabled ? "off" : "on", |
5939 | vm_total_pages); | |
f0c0b2b8 | 5940 | #ifdef CONFIG_NUMA |
f88dfff5 | 5941 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5942 | #endif |
1da177e4 LT |
5943 | } |
5944 | ||
a9a9e77f PT |
5945 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
5946 | static bool __meminit | |
5947 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
5948 | { | |
a9a9e77f PT |
5949 | static struct memblock_region *r; |
5950 | ||
5951 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5952 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
5953 | for_each_memblock(memory, r) { | |
5954 | if (*pfn < memblock_region_memory_end_pfn(r)) | |
5955 | break; | |
5956 | } | |
5957 | } | |
5958 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
5959 | memblock_is_mirror(r)) { | |
5960 | *pfn = memblock_region_memory_end_pfn(r); | |
5961 | return true; | |
5962 | } | |
5963 | } | |
a9a9e77f PT |
5964 | return false; |
5965 | } | |
5966 | ||
1da177e4 LT |
5967 | /* |
5968 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 5969 | * up by memblock_free_all() once the early boot process is |
1da177e4 LT |
5970 | * done. Non-atomic initialization, single-pass. |
5971 | */ | |
c09b4240 | 5972 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a99583e7 CH |
5973 | unsigned long start_pfn, enum memmap_context context, |
5974 | struct vmem_altmap *altmap) | |
1da177e4 | 5975 | { |
a9a9e77f | 5976 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 5977 | struct page *page; |
1da177e4 | 5978 | |
22b31eec HD |
5979 | if (highest_memmap_pfn < end_pfn - 1) |
5980 | highest_memmap_pfn = end_pfn - 1; | |
5981 | ||
966cf44f | 5982 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
5983 | /* |
5984 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
5985 | * memory. We limit the total number of pages to initialize to just |
5986 | * those that might contain the memory mapping. We will defer the | |
5987 | * ZONE_DEVICE page initialization until after we have released | |
5988 | * the hotplug lock. | |
4b94ffdc | 5989 | */ |
966cf44f AD |
5990 | if (zone == ZONE_DEVICE) { |
5991 | if (!altmap) | |
5992 | return; | |
5993 | ||
5994 | if (start_pfn == altmap->base_pfn) | |
5995 | start_pfn += altmap->reserve; | |
5996 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
5997 | } | |
5998 | #endif | |
4b94ffdc | 5999 | |
948c436e | 6000 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 6001 | /* |
b72d0ffb AM |
6002 | * There can be holes in boot-time mem_map[]s handed to this |
6003 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 6004 | */ |
a9a9e77f | 6005 | if (context == MEMMAP_EARLY) { |
a9a9e77f PT |
6006 | if (overlap_memmap_init(zone, &pfn)) |
6007 | continue; | |
6008 | if (defer_init(nid, pfn, end_pfn)) | |
6009 | break; | |
a2f3aa02 | 6010 | } |
ac5d2539 | 6011 | |
d0dc12e8 PT |
6012 | page = pfn_to_page(pfn); |
6013 | __init_single_page(page, pfn, zone, nid); | |
6014 | if (context == MEMMAP_HOTPLUG) | |
d483da5b | 6015 | __SetPageReserved(page); |
d0dc12e8 | 6016 | |
ac5d2539 MG |
6017 | /* |
6018 | * Mark the block movable so that blocks are reserved for | |
6019 | * movable at startup. This will force kernel allocations | |
6020 | * to reserve their blocks rather than leaking throughout | |
6021 | * the address space during boot when many long-lived | |
974a786e | 6022 | * kernel allocations are made. |
ac5d2539 MG |
6023 | * |
6024 | * bitmap is created for zone's valid pfn range. but memmap | |
6025 | * can be created for invalid pages (for alignment) | |
6026 | * check here not to call set_pageblock_migratetype() against | |
6027 | * pfn out of zone. | |
6028 | */ | |
6029 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
ac5d2539 | 6030 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
9b6e63cb | 6031 | cond_resched(); |
ac5d2539 | 6032 | } |
948c436e | 6033 | pfn++; |
1da177e4 LT |
6034 | } |
6035 | } | |
6036 | ||
966cf44f AD |
6037 | #ifdef CONFIG_ZONE_DEVICE |
6038 | void __ref memmap_init_zone_device(struct zone *zone, | |
6039 | unsigned long start_pfn, | |
1f8d75c1 | 6040 | unsigned long nr_pages, |
966cf44f AD |
6041 | struct dev_pagemap *pgmap) |
6042 | { | |
1f8d75c1 | 6043 | unsigned long pfn, end_pfn = start_pfn + nr_pages; |
966cf44f | 6044 | struct pglist_data *pgdat = zone->zone_pgdat; |
514caf23 | 6045 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
966cf44f AD |
6046 | unsigned long zone_idx = zone_idx(zone); |
6047 | unsigned long start = jiffies; | |
6048 | int nid = pgdat->node_id; | |
6049 | ||
46d945ae | 6050 | if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE)) |
966cf44f AD |
6051 | return; |
6052 | ||
6053 | /* | |
6054 | * The call to memmap_init_zone should have already taken care | |
6055 | * of the pages reserved for the memmap, so we can just jump to | |
6056 | * the end of that region and start processing the device pages. | |
6057 | */ | |
514caf23 | 6058 | if (altmap) { |
966cf44f | 6059 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
1f8d75c1 | 6060 | nr_pages = end_pfn - start_pfn; |
966cf44f AD |
6061 | } |
6062 | ||
6063 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
6064 | struct page *page = pfn_to_page(pfn); | |
6065 | ||
6066 | __init_single_page(page, pfn, zone_idx, nid); | |
6067 | ||
6068 | /* | |
6069 | * Mark page reserved as it will need to wait for onlining | |
6070 | * phase for it to be fully associated with a zone. | |
6071 | * | |
6072 | * We can use the non-atomic __set_bit operation for setting | |
6073 | * the flag as we are still initializing the pages. | |
6074 | */ | |
6075 | __SetPageReserved(page); | |
6076 | ||
6077 | /* | |
8a164fef CH |
6078 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer |
6079 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
6080 | * ever freed or placed on a driver-private list. | |
966cf44f AD |
6081 | */ |
6082 | page->pgmap = pgmap; | |
8a164fef | 6083 | page->zone_device_data = NULL; |
966cf44f AD |
6084 | |
6085 | /* | |
6086 | * Mark the block movable so that blocks are reserved for | |
6087 | * movable at startup. This will force kernel allocations | |
6088 | * to reserve their blocks rather than leaking throughout | |
6089 | * the address space during boot when many long-lived | |
6090 | * kernel allocations are made. | |
6091 | * | |
6092 | * bitmap is created for zone's valid pfn range. but memmap | |
6093 | * can be created for invalid pages (for alignment) | |
6094 | * check here not to call set_pageblock_migratetype() against | |
6095 | * pfn out of zone. | |
6096 | * | |
6097 | * Please note that MEMMAP_HOTPLUG path doesn't clear memmap | |
ba72b4c8 | 6098 | * because this is done early in section_activate() |
966cf44f AD |
6099 | */ |
6100 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
6101 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
6102 | cond_resched(); | |
6103 | } | |
6104 | } | |
6105 | ||
fdc029b1 | 6106 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
1f8d75c1 | 6107 | nr_pages, jiffies_to_msecs(jiffies - start)); |
966cf44f AD |
6108 | } |
6109 | ||
6110 | #endif | |
1e548deb | 6111 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6112 | { |
7aeb09f9 | 6113 | unsigned int order, t; |
b2a0ac88 MG |
6114 | for_each_migratetype_order(order, t) { |
6115 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6116 | zone->free_area[order].nr_free = 0; |
6117 | } | |
6118 | } | |
6119 | ||
dfb3ccd0 | 6120 | void __meminit __weak memmap_init(unsigned long size, int nid, |
73a6e474 BH |
6121 | unsigned long zone, |
6122 | unsigned long range_start_pfn) | |
dfb3ccd0 | 6123 | { |
73a6e474 BH |
6124 | unsigned long start_pfn, end_pfn; |
6125 | unsigned long range_end_pfn = range_start_pfn + size; | |
6126 | int i; | |
6127 | ||
6128 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { | |
6129 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6130 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6131 | ||
6132 | if (end_pfn > start_pfn) { | |
6133 | size = end_pfn - start_pfn; | |
6134 | memmap_init_zone(size, nid, zone, start_pfn, | |
6135 | MEMMAP_EARLY, NULL); | |
6136 | } | |
6137 | } | |
dfb3ccd0 | 6138 | } |
1da177e4 | 6139 | |
7cd2b0a3 | 6140 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 6141 | { |
3a6be87f | 6142 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
6143 | int batch; |
6144 | ||
6145 | /* | |
6146 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 6147 | * size of the zone. |
e7c8d5c9 | 6148 | */ |
9705bea5 | 6149 | batch = zone_managed_pages(zone) / 1024; |
d8a759b5 AL |
6150 | /* But no more than a meg. */ |
6151 | if (batch * PAGE_SIZE > 1024 * 1024) | |
6152 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
6153 | batch /= 4; /* We effectively *= 4 below */ |
6154 | if (batch < 1) | |
6155 | batch = 1; | |
6156 | ||
6157 | /* | |
0ceaacc9 NP |
6158 | * Clamp the batch to a 2^n - 1 value. Having a power |
6159 | * of 2 value was found to be more likely to have | |
6160 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 6161 | * |
0ceaacc9 NP |
6162 | * For example if 2 tasks are alternately allocating |
6163 | * batches of pages, one task can end up with a lot | |
6164 | * of pages of one half of the possible page colors | |
6165 | * and the other with pages of the other colors. | |
e7c8d5c9 | 6166 | */ |
9155203a | 6167 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 6168 | |
e7c8d5c9 | 6169 | return batch; |
3a6be87f DH |
6170 | |
6171 | #else | |
6172 | /* The deferral and batching of frees should be suppressed under NOMMU | |
6173 | * conditions. | |
6174 | * | |
6175 | * The problem is that NOMMU needs to be able to allocate large chunks | |
6176 | * of contiguous memory as there's no hardware page translation to | |
6177 | * assemble apparent contiguous memory from discontiguous pages. | |
6178 | * | |
6179 | * Queueing large contiguous runs of pages for batching, however, | |
6180 | * causes the pages to actually be freed in smaller chunks. As there | |
6181 | * can be a significant delay between the individual batches being | |
6182 | * recycled, this leads to the once large chunks of space being | |
6183 | * fragmented and becoming unavailable for high-order allocations. | |
6184 | */ | |
6185 | return 0; | |
6186 | #endif | |
e7c8d5c9 CL |
6187 | } |
6188 | ||
8d7a8fa9 CS |
6189 | /* |
6190 | * pcp->high and pcp->batch values are related and dependent on one another: | |
6191 | * ->batch must never be higher then ->high. | |
6192 | * The following function updates them in a safe manner without read side | |
6193 | * locking. | |
6194 | * | |
6195 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
047b9967 | 6196 | * those fields changing asynchronously (acording to the above rule). |
8d7a8fa9 CS |
6197 | * |
6198 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
6199 | * outside of boot time (or some other assurance that no concurrent updaters | |
6200 | * exist). | |
6201 | */ | |
6202 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
6203 | unsigned long batch) | |
6204 | { | |
6205 | /* start with a fail safe value for batch */ | |
6206 | pcp->batch = 1; | |
6207 | smp_wmb(); | |
6208 | ||
6209 | /* Update high, then batch, in order */ | |
6210 | pcp->high = high; | |
6211 | smp_wmb(); | |
6212 | ||
6213 | pcp->batch = batch; | |
6214 | } | |
6215 | ||
3664033c | 6216 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
6217 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
6218 | { | |
8d7a8fa9 | 6219 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
6220 | } |
6221 | ||
88c90dbc | 6222 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
6223 | { |
6224 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 6225 | int migratetype; |
2caaad41 | 6226 | |
1c6fe946 MD |
6227 | memset(p, 0, sizeof(*p)); |
6228 | ||
3dfa5721 | 6229 | pcp = &p->pcp; |
5f8dcc21 MG |
6230 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
6231 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
6232 | } |
6233 | ||
88c90dbc CS |
6234 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
6235 | { | |
6236 | pageset_init(p); | |
6237 | pageset_set_batch(p, batch); | |
6238 | } | |
6239 | ||
8ad4b1fb | 6240 | /* |
3664033c | 6241 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
6242 | * to the value high for the pageset p. |
6243 | */ | |
3664033c | 6244 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
6245 | unsigned long high) |
6246 | { | |
8d7a8fa9 CS |
6247 | unsigned long batch = max(1UL, high / 4); |
6248 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
6249 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 6250 | |
8d7a8fa9 | 6251 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
6252 | } |
6253 | ||
7cd2b0a3 DR |
6254 | static void pageset_set_high_and_batch(struct zone *zone, |
6255 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 6256 | { |
56cef2b8 | 6257 | if (percpu_pagelist_fraction) |
3664033c | 6258 | pageset_set_high(pcp, |
9705bea5 | 6259 | (zone_managed_pages(zone) / |
56cef2b8 CS |
6260 | percpu_pagelist_fraction)); |
6261 | else | |
6262 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
6263 | } | |
6264 | ||
169f6c19 CS |
6265 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
6266 | { | |
6267 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
6268 | ||
6269 | pageset_init(pcp); | |
6270 | pageset_set_high_and_batch(zone, pcp); | |
6271 | } | |
6272 | ||
72675e13 | 6273 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
6274 | { |
6275 | int cpu; | |
319774e2 | 6276 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
6277 | for_each_possible_cpu(cpu) |
6278 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
6279 | } |
6280 | ||
2caaad41 | 6281 | /* |
99dcc3e5 CL |
6282 | * Allocate per cpu pagesets and initialize them. |
6283 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 6284 | */ |
99dcc3e5 | 6285 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 6286 | { |
b4911ea2 | 6287 | struct pglist_data *pgdat; |
99dcc3e5 | 6288 | struct zone *zone; |
b418a0f9 | 6289 | int __maybe_unused cpu; |
e7c8d5c9 | 6290 | |
319774e2 WF |
6291 | for_each_populated_zone(zone) |
6292 | setup_zone_pageset(zone); | |
b4911ea2 | 6293 | |
b418a0f9 SD |
6294 | #ifdef CONFIG_NUMA |
6295 | /* | |
6296 | * Unpopulated zones continue using the boot pagesets. | |
6297 | * The numa stats for these pagesets need to be reset. | |
6298 | * Otherwise, they will end up skewing the stats of | |
6299 | * the nodes these zones are associated with. | |
6300 | */ | |
6301 | for_each_possible_cpu(cpu) { | |
6302 | struct per_cpu_pageset *pcp = &per_cpu(boot_pageset, cpu); | |
6303 | memset(pcp->vm_numa_stat_diff, 0, | |
6304 | sizeof(pcp->vm_numa_stat_diff)); | |
6305 | } | |
6306 | #endif | |
6307 | ||
b4911ea2 MG |
6308 | for_each_online_pgdat(pgdat) |
6309 | pgdat->per_cpu_nodestats = | |
6310 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
6311 | } |
6312 | ||
c09b4240 | 6313 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 6314 | { |
99dcc3e5 CL |
6315 | /* |
6316 | * per cpu subsystem is not up at this point. The following code | |
6317 | * relies on the ability of the linker to provide the | |
6318 | * offset of a (static) per cpu variable into the per cpu area. | |
6319 | */ | |
6320 | zone->pageset = &boot_pageset; | |
ed8ece2e | 6321 | |
b38a8725 | 6322 | if (populated_zone(zone)) |
99dcc3e5 CL |
6323 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
6324 | zone->name, zone->present_pages, | |
6325 | zone_batchsize(zone)); | |
ed8ece2e DH |
6326 | } |
6327 | ||
dc0bbf3b | 6328 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 6329 | unsigned long zone_start_pfn, |
b171e409 | 6330 | unsigned long size) |
ed8ece2e DH |
6331 | { |
6332 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 6333 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 6334 | |
8f416836 WY |
6335 | if (zone_idx > pgdat->nr_zones) |
6336 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 6337 | |
ed8ece2e DH |
6338 | zone->zone_start_pfn = zone_start_pfn; |
6339 | ||
708614e6 MG |
6340 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
6341 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
6342 | pgdat->node_id, | |
6343 | (unsigned long)zone_idx(zone), | |
6344 | zone_start_pfn, (zone_start_pfn + size)); | |
6345 | ||
1e548deb | 6346 | zone_init_free_lists(zone); |
9dcb8b68 | 6347 | zone->initialized = 1; |
ed8ece2e DH |
6348 | } |
6349 | ||
c713216d MG |
6350 | /** |
6351 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
6352 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
6353 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
6354 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
6355 | * |
6356 | * It returns the start and end page frame of a node based on information | |
7d018176 | 6357 | * provided by memblock_set_node(). If called for a node |
c713216d | 6358 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 6359 | * PFNs will be 0. |
c713216d | 6360 | */ |
bbe5d993 | 6361 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
6362 | unsigned long *start_pfn, unsigned long *end_pfn) |
6363 | { | |
c13291a5 | 6364 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 6365 | int i; |
c13291a5 | 6366 | |
c713216d MG |
6367 | *start_pfn = -1UL; |
6368 | *end_pfn = 0; | |
6369 | ||
c13291a5 TH |
6370 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
6371 | *start_pfn = min(*start_pfn, this_start_pfn); | |
6372 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
6373 | } |
6374 | ||
633c0666 | 6375 | if (*start_pfn == -1UL) |
c713216d | 6376 | *start_pfn = 0; |
c713216d MG |
6377 | } |
6378 | ||
2a1e274a MG |
6379 | /* |
6380 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
6381 | * assumption is made that zones within a node are ordered in monotonic | |
6382 | * increasing memory addresses so that the "highest" populated zone is used | |
6383 | */ | |
b69a7288 | 6384 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
6385 | { |
6386 | int zone_index; | |
6387 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
6388 | if (zone_index == ZONE_MOVABLE) | |
6389 | continue; | |
6390 | ||
6391 | if (arch_zone_highest_possible_pfn[zone_index] > | |
6392 | arch_zone_lowest_possible_pfn[zone_index]) | |
6393 | break; | |
6394 | } | |
6395 | ||
6396 | VM_BUG_ON(zone_index == -1); | |
6397 | movable_zone = zone_index; | |
6398 | } | |
6399 | ||
6400 | /* | |
6401 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 6402 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
6403 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
6404 | * in each node depending on the size of each node and how evenly kernelcore | |
6405 | * is distributed. This helper function adjusts the zone ranges | |
6406 | * provided by the architecture for a given node by using the end of the | |
6407 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
6408 | * zones within a node are in order of monotonic increases memory addresses | |
6409 | */ | |
bbe5d993 | 6410 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
6411 | unsigned long zone_type, |
6412 | unsigned long node_start_pfn, | |
6413 | unsigned long node_end_pfn, | |
6414 | unsigned long *zone_start_pfn, | |
6415 | unsigned long *zone_end_pfn) | |
6416 | { | |
6417 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
6418 | if (zone_movable_pfn[nid]) { | |
6419 | /* Size ZONE_MOVABLE */ | |
6420 | if (zone_type == ZONE_MOVABLE) { | |
6421 | *zone_start_pfn = zone_movable_pfn[nid]; | |
6422 | *zone_end_pfn = min(node_end_pfn, | |
6423 | arch_zone_highest_possible_pfn[movable_zone]); | |
6424 | ||
e506b996 XQ |
6425 | /* Adjust for ZONE_MOVABLE starting within this range */ |
6426 | } else if (!mirrored_kernelcore && | |
6427 | *zone_start_pfn < zone_movable_pfn[nid] && | |
6428 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
6429 | *zone_end_pfn = zone_movable_pfn[nid]; | |
6430 | ||
2a1e274a MG |
6431 | /* Check if this whole range is within ZONE_MOVABLE */ |
6432 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
6433 | *zone_start_pfn = *zone_end_pfn; | |
6434 | } | |
6435 | } | |
6436 | ||
c713216d MG |
6437 | /* |
6438 | * Return the number of pages a zone spans in a node, including holes | |
6439 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
6440 | */ | |
bbe5d993 | 6441 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 6442 | unsigned long zone_type, |
7960aedd ZY |
6443 | unsigned long node_start_pfn, |
6444 | unsigned long node_end_pfn, | |
d91749c1 | 6445 | unsigned long *zone_start_pfn, |
854e8848 | 6446 | unsigned long *zone_end_pfn) |
c713216d | 6447 | { |
299c83dc LF |
6448 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6449 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 6450 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6451 | if (!node_start_pfn && !node_end_pfn) |
6452 | return 0; | |
6453 | ||
7960aedd | 6454 | /* Get the start and end of the zone */ |
299c83dc LF |
6455 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6456 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
6457 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6458 | node_start_pfn, node_end_pfn, | |
d91749c1 | 6459 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
6460 | |
6461 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 6462 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
6463 | return 0; |
6464 | ||
6465 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
6466 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
6467 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
6468 | |
6469 | /* Return the spanned pages */ | |
d91749c1 | 6470 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
6471 | } |
6472 | ||
6473 | /* | |
6474 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 6475 | * then all holes in the requested range will be accounted for. |
c713216d | 6476 | */ |
bbe5d993 | 6477 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
6478 | unsigned long range_start_pfn, |
6479 | unsigned long range_end_pfn) | |
6480 | { | |
96e907d1 TH |
6481 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
6482 | unsigned long start_pfn, end_pfn; | |
6483 | int i; | |
c713216d | 6484 | |
96e907d1 TH |
6485 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6486 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6487 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6488 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 6489 | } |
96e907d1 | 6490 | return nr_absent; |
c713216d MG |
6491 | } |
6492 | ||
6493 | /** | |
6494 | * absent_pages_in_range - Return number of page frames in holes within a range | |
6495 | * @start_pfn: The start PFN to start searching for holes | |
6496 | * @end_pfn: The end PFN to stop searching for holes | |
6497 | * | |
a862f68a | 6498 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
6499 | */ |
6500 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
6501 | unsigned long end_pfn) | |
6502 | { | |
6503 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
6504 | } | |
6505 | ||
6506 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 6507 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 6508 | unsigned long zone_type, |
7960aedd | 6509 | unsigned long node_start_pfn, |
854e8848 | 6510 | unsigned long node_end_pfn) |
c713216d | 6511 | { |
96e907d1 TH |
6512 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6513 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 6514 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 6515 | unsigned long nr_absent; |
9c7cd687 | 6516 | |
b5685e92 | 6517 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6518 | if (!node_start_pfn && !node_end_pfn) |
6519 | return 0; | |
6520 | ||
96e907d1 TH |
6521 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6522 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 6523 | |
2a1e274a MG |
6524 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6525 | node_start_pfn, node_end_pfn, | |
6526 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
6527 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
6528 | ||
6529 | /* | |
6530 | * ZONE_MOVABLE handling. | |
6531 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6532 | * and vice versa. | |
6533 | */ | |
e506b996 XQ |
6534 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6535 | unsigned long start_pfn, end_pfn; | |
6536 | struct memblock_region *r; | |
6537 | ||
6538 | for_each_memblock(memory, r) { | |
6539 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
6540 | zone_start_pfn, zone_end_pfn); | |
6541 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6542 | zone_start_pfn, zone_end_pfn); | |
6543 | ||
6544 | if (zone_type == ZONE_MOVABLE && | |
6545 | memblock_is_mirror(r)) | |
6546 | nr_absent += end_pfn - start_pfn; | |
6547 | ||
6548 | if (zone_type == ZONE_NORMAL && | |
6549 | !memblock_is_mirror(r)) | |
6550 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6551 | } |
6552 | } | |
6553 | ||
6554 | return nr_absent; | |
c713216d | 6555 | } |
0e0b864e | 6556 | |
bbe5d993 | 6557 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd | 6558 | unsigned long node_start_pfn, |
854e8848 | 6559 | unsigned long node_end_pfn) |
c713216d | 6560 | { |
febd5949 | 6561 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6562 | enum zone_type i; |
6563 | ||
febd5949 GZ |
6564 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6565 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6566 | unsigned long zone_start_pfn, zone_end_pfn; |
3f08a302 | 6567 | unsigned long spanned, absent; |
febd5949 | 6568 | unsigned long size, real_size; |
c713216d | 6569 | |
854e8848 MR |
6570 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i, |
6571 | node_start_pfn, | |
6572 | node_end_pfn, | |
6573 | &zone_start_pfn, | |
6574 | &zone_end_pfn); | |
6575 | absent = zone_absent_pages_in_node(pgdat->node_id, i, | |
6576 | node_start_pfn, | |
6577 | node_end_pfn); | |
3f08a302 MR |
6578 | |
6579 | size = spanned; | |
6580 | real_size = size - absent; | |
6581 | ||
d91749c1 TI |
6582 | if (size) |
6583 | zone->zone_start_pfn = zone_start_pfn; | |
6584 | else | |
6585 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
6586 | zone->spanned_pages = size; |
6587 | zone->present_pages = real_size; | |
6588 | ||
6589 | totalpages += size; | |
6590 | realtotalpages += real_size; | |
6591 | } | |
6592 | ||
6593 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
6594 | pgdat->node_present_pages = realtotalpages; |
6595 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
6596 | realtotalpages); | |
6597 | } | |
6598 | ||
835c134e MG |
6599 | #ifndef CONFIG_SPARSEMEM |
6600 | /* | |
6601 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
6602 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
6603 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
6604 | * round what is now in bits to nearest long in bits, then return it in |
6605 | * bytes. | |
6606 | */ | |
7c45512d | 6607 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
6608 | { |
6609 | unsigned long usemapsize; | |
6610 | ||
7c45512d | 6611 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
6612 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
6613 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
6614 | usemapsize *= NR_PAGEBLOCK_BITS; |
6615 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
6616 | ||
6617 | return usemapsize / 8; | |
6618 | } | |
6619 | ||
7cc2a959 | 6620 | static void __ref setup_usemap(struct pglist_data *pgdat, |
7c45512d LT |
6621 | struct zone *zone, |
6622 | unsigned long zone_start_pfn, | |
6623 | unsigned long zonesize) | |
835c134e | 6624 | { |
7c45512d | 6625 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 6626 | zone->pageblock_flags = NULL; |
23a7052a | 6627 | if (usemapsize) { |
6782832e | 6628 | zone->pageblock_flags = |
26fb3dae MR |
6629 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
6630 | pgdat->node_id); | |
23a7052a MR |
6631 | if (!zone->pageblock_flags) |
6632 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
6633 | usemapsize, zone->name, pgdat->node_id); | |
6634 | } | |
835c134e MG |
6635 | } |
6636 | #else | |
7c45512d LT |
6637 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
6638 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
6639 | #endif /* CONFIG_SPARSEMEM */ |
6640 | ||
d9c23400 | 6641 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 6642 | |
d9c23400 | 6643 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 6644 | void __init set_pageblock_order(void) |
d9c23400 | 6645 | { |
955c1cd7 AM |
6646 | unsigned int order; |
6647 | ||
d9c23400 MG |
6648 | /* Check that pageblock_nr_pages has not already been setup */ |
6649 | if (pageblock_order) | |
6650 | return; | |
6651 | ||
955c1cd7 AM |
6652 | if (HPAGE_SHIFT > PAGE_SHIFT) |
6653 | order = HUGETLB_PAGE_ORDER; | |
6654 | else | |
6655 | order = MAX_ORDER - 1; | |
6656 | ||
d9c23400 MG |
6657 | /* |
6658 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
6659 | * This value may be variable depending on boot parameters on IA64 and |
6660 | * powerpc. | |
d9c23400 MG |
6661 | */ |
6662 | pageblock_order = order; | |
6663 | } | |
6664 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6665 | ||
ba72cb8c MG |
6666 | /* |
6667 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
6668 | * is unused as pageblock_order is set at compile-time. See |
6669 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
6670 | * the kernel config | |
ba72cb8c | 6671 | */ |
03e85f9d | 6672 | void __init set_pageblock_order(void) |
ba72cb8c | 6673 | { |
ba72cb8c | 6674 | } |
d9c23400 MG |
6675 | |
6676 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6677 | ||
03e85f9d | 6678 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 6679 | unsigned long present_pages) |
01cefaef JL |
6680 | { |
6681 | unsigned long pages = spanned_pages; | |
6682 | ||
6683 | /* | |
6684 | * Provide a more accurate estimation if there are holes within | |
6685 | * the zone and SPARSEMEM is in use. If there are holes within the | |
6686 | * zone, each populated memory region may cost us one or two extra | |
6687 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 6688 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
6689 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
6690 | */ | |
6691 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
6692 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
6693 | pages = present_pages; | |
6694 | ||
6695 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
6696 | } | |
6697 | ||
ace1db39 OS |
6698 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6699 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
6700 | { | |
364c1eeb YS |
6701 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
6702 | ||
6703 | spin_lock_init(&ds_queue->split_queue_lock); | |
6704 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
6705 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
6706 | } |
6707 | #else | |
6708 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
6709 | #endif | |
6710 | ||
6711 | #ifdef CONFIG_COMPACTION | |
6712 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
6713 | { | |
6714 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6715 | } | |
6716 | #else | |
6717 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
6718 | #endif | |
6719 | ||
03e85f9d | 6720 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 6721 | { |
208d54e5 | 6722 | pgdat_resize_init(pgdat); |
ace1db39 | 6723 | |
ace1db39 OS |
6724 | pgdat_init_split_queue(pgdat); |
6725 | pgdat_init_kcompactd(pgdat); | |
6726 | ||
1da177e4 | 6727 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6728 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 6729 | |
eefa864b | 6730 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6731 | spin_lock_init(&pgdat->lru_lock); |
867e5e1d | 6732 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
6733 | } |
6734 | ||
6735 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
6736 | unsigned long remaining_pages) | |
6737 | { | |
9705bea5 | 6738 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
6739 | zone_set_nid(zone, nid); |
6740 | zone->name = zone_names[idx]; | |
6741 | zone->zone_pgdat = NODE_DATA(nid); | |
6742 | spin_lock_init(&zone->lock); | |
6743 | zone_seqlock_init(zone); | |
6744 | zone_pcp_init(zone); | |
6745 | } | |
6746 | ||
6747 | /* | |
6748 | * Set up the zone data structures | |
6749 | * - init pgdat internals | |
6750 | * - init all zones belonging to this node | |
6751 | * | |
6752 | * NOTE: this function is only called during memory hotplug | |
6753 | */ | |
6754 | #ifdef CONFIG_MEMORY_HOTPLUG | |
6755 | void __ref free_area_init_core_hotplug(int nid) | |
6756 | { | |
6757 | enum zone_type z; | |
6758 | pg_data_t *pgdat = NODE_DATA(nid); | |
6759 | ||
6760 | pgdat_init_internals(pgdat); | |
6761 | for (z = 0; z < MAX_NR_ZONES; z++) | |
6762 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
6763 | } | |
6764 | #endif | |
6765 | ||
6766 | /* | |
6767 | * Set up the zone data structures: | |
6768 | * - mark all pages reserved | |
6769 | * - mark all memory queues empty | |
6770 | * - clear the memory bitmaps | |
6771 | * | |
6772 | * NOTE: pgdat should get zeroed by caller. | |
6773 | * NOTE: this function is only called during early init. | |
6774 | */ | |
6775 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
6776 | { | |
6777 | enum zone_type j; | |
6778 | int nid = pgdat->node_id; | |
5f63b720 | 6779 | |
03e85f9d | 6780 | pgdat_init_internals(pgdat); |
385386cf JW |
6781 | pgdat->per_cpu_nodestats = &boot_nodestats; |
6782 | ||
1da177e4 LT |
6783 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6784 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 6785 | unsigned long size, freesize, memmap_pages; |
d91749c1 | 6786 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6787 | |
febd5949 | 6788 | size = zone->spanned_pages; |
e6943859 | 6789 | freesize = zone->present_pages; |
1da177e4 | 6790 | |
0e0b864e | 6791 | /* |
9feedc9d | 6792 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6793 | * is used by this zone for memmap. This affects the watermark |
6794 | * and per-cpu initialisations | |
6795 | */ | |
e6943859 | 6796 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
6797 | if (!is_highmem_idx(j)) { |
6798 | if (freesize >= memmap_pages) { | |
6799 | freesize -= memmap_pages; | |
6800 | if (memmap_pages) | |
6801 | printk(KERN_DEBUG | |
6802 | " %s zone: %lu pages used for memmap\n", | |
6803 | zone_names[j], memmap_pages); | |
6804 | } else | |
1170532b | 6805 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6806 | zone_names[j], memmap_pages, freesize); |
6807 | } | |
0e0b864e | 6808 | |
6267276f | 6809 | /* Account for reserved pages */ |
9feedc9d JL |
6810 | if (j == 0 && freesize > dma_reserve) { |
6811 | freesize -= dma_reserve; | |
d903ef9f | 6812 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6813 | zone_names[0], dma_reserve); |
0e0b864e MG |
6814 | } |
6815 | ||
98d2b0eb | 6816 | if (!is_highmem_idx(j)) |
9feedc9d | 6817 | nr_kernel_pages += freesize; |
01cefaef JL |
6818 | /* Charge for highmem memmap if there are enough kernel pages */ |
6819 | else if (nr_kernel_pages > memmap_pages * 2) | |
6820 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6821 | nr_all_pages += freesize; |
1da177e4 | 6822 | |
9feedc9d JL |
6823 | /* |
6824 | * Set an approximate value for lowmem here, it will be adjusted | |
6825 | * when the bootmem allocator frees pages into the buddy system. | |
6826 | * And all highmem pages will be managed by the buddy system. | |
6827 | */ | |
03e85f9d | 6828 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 6829 | |
d883c6cf | 6830 | if (!size) |
1da177e4 LT |
6831 | continue; |
6832 | ||
955c1cd7 | 6833 | set_pageblock_order(); |
d883c6cf JK |
6834 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
6835 | init_currently_empty_zone(zone, zone_start_pfn, size); | |
76cdd58e | 6836 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6837 | } |
6838 | } | |
6839 | ||
0cd842f9 | 6840 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 6841 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6842 | { |
b0aeba74 | 6843 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6844 | unsigned long __maybe_unused offset = 0; |
6845 | ||
1da177e4 LT |
6846 | /* Skip empty nodes */ |
6847 | if (!pgdat->node_spanned_pages) | |
6848 | return; | |
6849 | ||
b0aeba74 TL |
6850 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6851 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6852 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6853 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6854 | unsigned long size, end; |
d41dee36 AW |
6855 | struct page *map; |
6856 | ||
e984bb43 BP |
6857 | /* |
6858 | * The zone's endpoints aren't required to be MAX_ORDER | |
6859 | * aligned but the node_mem_map endpoints must be in order | |
6860 | * for the buddy allocator to function correctly. | |
6861 | */ | |
108bcc96 | 6862 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6863 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6864 | size = (end - start) * sizeof(struct page); | |
26fb3dae MR |
6865 | map = memblock_alloc_node(size, SMP_CACHE_BYTES, |
6866 | pgdat->node_id); | |
23a7052a MR |
6867 | if (!map) |
6868 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
6869 | size, pgdat->node_id); | |
a1c34a3b | 6870 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6871 | } |
0cd842f9 OS |
6872 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
6873 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
6874 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 6875 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6876 | /* |
6877 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6878 | */ | |
c713216d | 6879 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6880 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d | 6881 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6882 | mem_map -= offset; |
c713216d | 6883 | } |
1da177e4 LT |
6884 | #endif |
6885 | } | |
0cd842f9 OS |
6886 | #else |
6887 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
6888 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 6889 | |
0188dc98 OS |
6890 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
6891 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
6892 | { | |
0188dc98 OS |
6893 | pgdat->first_deferred_pfn = ULONG_MAX; |
6894 | } | |
6895 | #else | |
6896 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
6897 | #endif | |
6898 | ||
854e8848 | 6899 | static void __init free_area_init_node(int nid) |
1da177e4 | 6900 | { |
9109fb7b | 6901 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6902 | unsigned long start_pfn = 0; |
6903 | unsigned long end_pfn = 0; | |
9109fb7b | 6904 | |
88fdf75d | 6905 | /* pg_data_t should be reset to zero when it's allocated */ |
97a225e6 | 6906 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx); |
88fdf75d | 6907 | |
854e8848 | 6908 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
88fdf75d | 6909 | |
1da177e4 | 6910 | pgdat->node_id = nid; |
854e8848 | 6911 | pgdat->node_start_pfn = start_pfn; |
75ef7184 | 6912 | pgdat->per_cpu_nodestats = NULL; |
854e8848 | 6913 | |
8d29e18a | 6914 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
6915 | (u64)start_pfn << PAGE_SHIFT, |
6916 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
854e8848 | 6917 | calculate_node_totalpages(pgdat, start_pfn, end_pfn); |
1da177e4 LT |
6918 | |
6919 | alloc_node_mem_map(pgdat); | |
0188dc98 | 6920 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 6921 | |
7f3eb55b | 6922 | free_area_init_core(pgdat); |
1da177e4 LT |
6923 | } |
6924 | ||
bc9331a1 | 6925 | void __init free_area_init_memoryless_node(int nid) |
3f08a302 | 6926 | { |
854e8848 | 6927 | free_area_init_node(nid); |
3f08a302 MR |
6928 | } |
6929 | ||
aca52c39 | 6930 | #if !defined(CONFIG_FLAT_NODE_MEM_MAP) |
ec393a0f | 6931 | /* |
4b094b78 DH |
6932 | * Initialize all valid struct pages in the range [spfn, epfn) and mark them |
6933 | * PageReserved(). Return the number of struct pages that were initialized. | |
ec393a0f | 6934 | */ |
4b094b78 | 6935 | static u64 __init init_unavailable_range(unsigned long spfn, unsigned long epfn) |
ec393a0f PT |
6936 | { |
6937 | unsigned long pfn; | |
6938 | u64 pgcnt = 0; | |
6939 | ||
6940 | for (pfn = spfn; pfn < epfn; pfn++) { | |
6941 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { | |
6942 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6943 | + pageblock_nr_pages - 1; | |
6944 | continue; | |
6945 | } | |
4b094b78 DH |
6946 | /* |
6947 | * Use a fake node/zone (0) for now. Some of these pages | |
6948 | * (in memblock.reserved but not in memblock.memory) will | |
6949 | * get re-initialized via reserve_bootmem_region() later. | |
6950 | */ | |
6951 | __init_single_page(pfn_to_page(pfn), pfn, 0, 0); | |
6952 | __SetPageReserved(pfn_to_page(pfn)); | |
ec393a0f PT |
6953 | pgcnt++; |
6954 | } | |
6955 | ||
6956 | return pgcnt; | |
6957 | } | |
6958 | ||
a4a3ede2 PT |
6959 | /* |
6960 | * Only struct pages that are backed by physical memory are zeroed and | |
6961 | * initialized by going through __init_single_page(). But, there are some | |
6962 | * struct pages which are reserved in memblock allocator and their fields | |
6963 | * may be accessed (for example page_to_pfn() on some configuration accesses | |
4b094b78 | 6964 | * flags). We must explicitly initialize those struct pages. |
907ec5fc NH |
6965 | * |
6966 | * This function also addresses a similar issue where struct pages are left | |
6967 | * uninitialized because the physical address range is not covered by | |
6968 | * memblock.memory or memblock.reserved. That could happen when memblock | |
e822969c DH |
6969 | * layout is manually configured via memmap=, or when the highest physical |
6970 | * address (max_pfn) does not end on a section boundary. | |
a4a3ede2 | 6971 | */ |
4b094b78 | 6972 | static void __init init_unavailable_mem(void) |
a4a3ede2 PT |
6973 | { |
6974 | phys_addr_t start, end; | |
a4a3ede2 | 6975 | u64 i, pgcnt; |
907ec5fc | 6976 | phys_addr_t next = 0; |
a4a3ede2 PT |
6977 | |
6978 | /* | |
907ec5fc | 6979 | * Loop through unavailable ranges not covered by memblock.memory. |
a4a3ede2 PT |
6980 | */ |
6981 | pgcnt = 0; | |
907ec5fc NH |
6982 | for_each_mem_range(i, &memblock.memory, NULL, |
6983 | NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) { | |
ec393a0f | 6984 | if (next < start) |
4b094b78 DH |
6985 | pgcnt += init_unavailable_range(PFN_DOWN(next), |
6986 | PFN_UP(start)); | |
907ec5fc NH |
6987 | next = end; |
6988 | } | |
e822969c DH |
6989 | |
6990 | /* | |
6991 | * Early sections always have a fully populated memmap for the whole | |
6992 | * section - see pfn_valid(). If the last section has holes at the | |
6993 | * end and that section is marked "online", the memmap will be | |
6994 | * considered initialized. Make sure that memmap has a well defined | |
6995 | * state. | |
6996 | */ | |
4b094b78 DH |
6997 | pgcnt += init_unavailable_range(PFN_DOWN(next), |
6998 | round_up(max_pfn, PAGES_PER_SECTION)); | |
907ec5fc | 6999 | |
a4a3ede2 PT |
7000 | /* |
7001 | * Struct pages that do not have backing memory. This could be because | |
7002 | * firmware is using some of this memory, or for some other reasons. | |
a4a3ede2 PT |
7003 | */ |
7004 | if (pgcnt) | |
907ec5fc | 7005 | pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt); |
a4a3ede2 | 7006 | } |
4b094b78 DH |
7007 | #else |
7008 | static inline void __init init_unavailable_mem(void) | |
7009 | { | |
7010 | } | |
aca52c39 | 7011 | #endif /* !CONFIG_FLAT_NODE_MEM_MAP */ |
a4a3ede2 | 7012 | |
418508c1 MS |
7013 | #if MAX_NUMNODES > 1 |
7014 | /* | |
7015 | * Figure out the number of possible node ids. | |
7016 | */ | |
f9872caf | 7017 | void __init setup_nr_node_ids(void) |
418508c1 | 7018 | { |
904a9553 | 7019 | unsigned int highest; |
418508c1 | 7020 | |
904a9553 | 7021 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7022 | nr_node_ids = highest + 1; |
7023 | } | |
418508c1 MS |
7024 | #endif |
7025 | ||
1e01979c TH |
7026 | /** |
7027 | * node_map_pfn_alignment - determine the maximum internode alignment | |
7028 | * | |
7029 | * This function should be called after node map is populated and sorted. | |
7030 | * It calculates the maximum power of two alignment which can distinguish | |
7031 | * all the nodes. | |
7032 | * | |
7033 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
7034 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
7035 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
7036 | * shifted, 1GiB is enough and this function will indicate so. | |
7037 | * | |
7038 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
7039 | * model has fine enough granularity to avoid incorrect mapping for the | |
7040 | * populated node map. | |
7041 | * | |
a862f68a | 7042 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
7043 | * requirement (single node). |
7044 | */ | |
7045 | unsigned long __init node_map_pfn_alignment(void) | |
7046 | { | |
7047 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 7048 | unsigned long start, end, mask; |
98fa15f3 | 7049 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 7050 | int i, nid; |
1e01979c | 7051 | |
c13291a5 | 7052 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
7053 | if (!start || last_nid < 0 || last_nid == nid) { |
7054 | last_nid = nid; | |
7055 | last_end = end; | |
7056 | continue; | |
7057 | } | |
7058 | ||
7059 | /* | |
7060 | * Start with a mask granular enough to pin-point to the | |
7061 | * start pfn and tick off bits one-by-one until it becomes | |
7062 | * too coarse to separate the current node from the last. | |
7063 | */ | |
7064 | mask = ~((1 << __ffs(start)) - 1); | |
7065 | while (mask && last_end <= (start & (mask << 1))) | |
7066 | mask <<= 1; | |
7067 | ||
7068 | /* accumulate all internode masks */ | |
7069 | accl_mask |= mask; | |
7070 | } | |
7071 | ||
7072 | /* convert mask to number of pages */ | |
7073 | return ~accl_mask + 1; | |
7074 | } | |
7075 | ||
c713216d MG |
7076 | /** |
7077 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
7078 | * | |
a862f68a | 7079 | * Return: the minimum PFN based on information provided via |
7d018176 | 7080 | * memblock_set_node(). |
c713216d MG |
7081 | */ |
7082 | unsigned long __init find_min_pfn_with_active_regions(void) | |
7083 | { | |
8a1b25fe | 7084 | return PHYS_PFN(memblock_start_of_DRAM()); |
c713216d MG |
7085 | } |
7086 | ||
37b07e41 LS |
7087 | /* |
7088 | * early_calculate_totalpages() | |
7089 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 7090 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 7091 | */ |
484f51f8 | 7092 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 7093 | { |
7e63efef | 7094 | unsigned long totalpages = 0; |
c13291a5 TH |
7095 | unsigned long start_pfn, end_pfn; |
7096 | int i, nid; | |
7097 | ||
7098 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
7099 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 7100 | |
37b07e41 LS |
7101 | totalpages += pages; |
7102 | if (pages) | |
4b0ef1fe | 7103 | node_set_state(nid, N_MEMORY); |
37b07e41 | 7104 | } |
b8af2941 | 7105 | return totalpages; |
7e63efef MG |
7106 | } |
7107 | ||
2a1e274a MG |
7108 | /* |
7109 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
7110 | * is spread evenly between nodes as long as the nodes have enough | |
7111 | * memory. When they don't, some nodes will have more kernelcore than | |
7112 | * others | |
7113 | */ | |
b224ef85 | 7114 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
7115 | { |
7116 | int i, nid; | |
7117 | unsigned long usable_startpfn; | |
7118 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 7119 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 7120 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 7121 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 7122 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 7123 | struct memblock_region *r; |
b2f3eebe TC |
7124 | |
7125 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
7126 | find_usable_zone_for_movable(); | |
7127 | ||
7128 | /* | |
7129 | * If movable_node is specified, ignore kernelcore and movablecore | |
7130 | * options. | |
7131 | */ | |
7132 | if (movable_node_is_enabled()) { | |
136199f0 EM |
7133 | for_each_memblock(memory, r) { |
7134 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
7135 | continue; |
7136 | ||
d622abf7 | 7137 | nid = memblock_get_region_node(r); |
b2f3eebe | 7138 | |
136199f0 | 7139 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
7140 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
7141 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7142 | usable_startpfn; | |
7143 | } | |
7144 | ||
7145 | goto out2; | |
7146 | } | |
2a1e274a | 7147 | |
342332e6 TI |
7148 | /* |
7149 | * If kernelcore=mirror is specified, ignore movablecore option | |
7150 | */ | |
7151 | if (mirrored_kernelcore) { | |
7152 | bool mem_below_4gb_not_mirrored = false; | |
7153 | ||
7154 | for_each_memblock(memory, r) { | |
7155 | if (memblock_is_mirror(r)) | |
7156 | continue; | |
7157 | ||
d622abf7 | 7158 | nid = memblock_get_region_node(r); |
342332e6 TI |
7159 | |
7160 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
7161 | ||
7162 | if (usable_startpfn < 0x100000) { | |
7163 | mem_below_4gb_not_mirrored = true; | |
7164 | continue; | |
7165 | } | |
7166 | ||
7167 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
7168 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7169 | usable_startpfn; | |
7170 | } | |
7171 | ||
7172 | if (mem_below_4gb_not_mirrored) | |
633bf2fe | 7173 | pr_warn("This configuration results in unmirrored kernel memory.\n"); |
342332e6 TI |
7174 | |
7175 | goto out2; | |
7176 | } | |
7177 | ||
7e63efef | 7178 | /* |
a5c6d650 DR |
7179 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
7180 | * amount of necessary memory. | |
7181 | */ | |
7182 | if (required_kernelcore_percent) | |
7183 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
7184 | 10000UL; | |
7185 | if (required_movablecore_percent) | |
7186 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
7187 | 10000UL; | |
7188 | ||
7189 | /* | |
7190 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
7191 | * kernelcore that corresponds so that memory usable for |
7192 | * any allocation type is evenly spread. If both kernelcore | |
7193 | * and movablecore are specified, then the value of kernelcore | |
7194 | * will be used for required_kernelcore if it's greater than | |
7195 | * what movablecore would have allowed. | |
7196 | */ | |
7197 | if (required_movablecore) { | |
7e63efef MG |
7198 | unsigned long corepages; |
7199 | ||
7200 | /* | |
7201 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
7202 | * was requested by the user | |
7203 | */ | |
7204 | required_movablecore = | |
7205 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 7206 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
7207 | corepages = totalpages - required_movablecore; |
7208 | ||
7209 | required_kernelcore = max(required_kernelcore, corepages); | |
7210 | } | |
7211 | ||
bde304bd XQ |
7212 | /* |
7213 | * If kernelcore was not specified or kernelcore size is larger | |
7214 | * than totalpages, there is no ZONE_MOVABLE. | |
7215 | */ | |
7216 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 7217 | goto out; |
2a1e274a MG |
7218 | |
7219 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
7220 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
7221 | ||
7222 | restart: | |
7223 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
7224 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 7225 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
7226 | unsigned long start_pfn, end_pfn; |
7227 | ||
2a1e274a MG |
7228 | /* |
7229 | * Recalculate kernelcore_node if the division per node | |
7230 | * now exceeds what is necessary to satisfy the requested | |
7231 | * amount of memory for the kernel | |
7232 | */ | |
7233 | if (required_kernelcore < kernelcore_node) | |
7234 | kernelcore_node = required_kernelcore / usable_nodes; | |
7235 | ||
7236 | /* | |
7237 | * As the map is walked, we track how much memory is usable | |
7238 | * by the kernel using kernelcore_remaining. When it is | |
7239 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
7240 | */ | |
7241 | kernelcore_remaining = kernelcore_node; | |
7242 | ||
7243 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 7244 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
7245 | unsigned long size_pages; |
7246 | ||
c13291a5 | 7247 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
7248 | if (start_pfn >= end_pfn) |
7249 | continue; | |
7250 | ||
7251 | /* Account for what is only usable for kernelcore */ | |
7252 | if (start_pfn < usable_startpfn) { | |
7253 | unsigned long kernel_pages; | |
7254 | kernel_pages = min(end_pfn, usable_startpfn) | |
7255 | - start_pfn; | |
7256 | ||
7257 | kernelcore_remaining -= min(kernel_pages, | |
7258 | kernelcore_remaining); | |
7259 | required_kernelcore -= min(kernel_pages, | |
7260 | required_kernelcore); | |
7261 | ||
7262 | /* Continue if range is now fully accounted */ | |
7263 | if (end_pfn <= usable_startpfn) { | |
7264 | ||
7265 | /* | |
7266 | * Push zone_movable_pfn to the end so | |
7267 | * that if we have to rebalance | |
7268 | * kernelcore across nodes, we will | |
7269 | * not double account here | |
7270 | */ | |
7271 | zone_movable_pfn[nid] = end_pfn; | |
7272 | continue; | |
7273 | } | |
7274 | start_pfn = usable_startpfn; | |
7275 | } | |
7276 | ||
7277 | /* | |
7278 | * The usable PFN range for ZONE_MOVABLE is from | |
7279 | * start_pfn->end_pfn. Calculate size_pages as the | |
7280 | * number of pages used as kernelcore | |
7281 | */ | |
7282 | size_pages = end_pfn - start_pfn; | |
7283 | if (size_pages > kernelcore_remaining) | |
7284 | size_pages = kernelcore_remaining; | |
7285 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
7286 | ||
7287 | /* | |
7288 | * Some kernelcore has been met, update counts and | |
7289 | * break if the kernelcore for this node has been | |
b8af2941 | 7290 | * satisfied |
2a1e274a MG |
7291 | */ |
7292 | required_kernelcore -= min(required_kernelcore, | |
7293 | size_pages); | |
7294 | kernelcore_remaining -= size_pages; | |
7295 | if (!kernelcore_remaining) | |
7296 | break; | |
7297 | } | |
7298 | } | |
7299 | ||
7300 | /* | |
7301 | * If there is still required_kernelcore, we do another pass with one | |
7302 | * less node in the count. This will push zone_movable_pfn[nid] further | |
7303 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 7304 | * satisfied |
2a1e274a MG |
7305 | */ |
7306 | usable_nodes--; | |
7307 | if (usable_nodes && required_kernelcore > usable_nodes) | |
7308 | goto restart; | |
7309 | ||
b2f3eebe | 7310 | out2: |
2a1e274a MG |
7311 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
7312 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
7313 | zone_movable_pfn[nid] = | |
7314 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 7315 | |
20e6926d | 7316 | out: |
66918dcd | 7317 | /* restore the node_state */ |
4b0ef1fe | 7318 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
7319 | } |
7320 | ||
4b0ef1fe LJ |
7321 | /* Any regular or high memory on that node ? */ |
7322 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 7323 | { |
37b07e41 LS |
7324 | enum zone_type zone_type; |
7325 | ||
4b0ef1fe | 7326 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 7327 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 7328 | if (populated_zone(zone)) { |
7b0e0c0e OS |
7329 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
7330 | node_set_state(nid, N_HIGH_MEMORY); | |
7331 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 7332 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
7333 | break; |
7334 | } | |
37b07e41 | 7335 | } |
37b07e41 LS |
7336 | } |
7337 | ||
51930df5 MR |
7338 | /* |
7339 | * Some architecturs, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For | |
7340 | * such cases we allow max_zone_pfn sorted in the descending order | |
7341 | */ | |
7342 | bool __weak arch_has_descending_max_zone_pfns(void) | |
7343 | { | |
7344 | return false; | |
7345 | } | |
7346 | ||
c713216d | 7347 | /** |
9691a071 | 7348 | * free_area_init - Initialise all pg_data_t and zone data |
88ca3b94 | 7349 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
7350 | * |
7351 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 7352 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
7353 | * zone in each node and their holes is calculated. If the maximum PFN |
7354 | * between two adjacent zones match, it is assumed that the zone is empty. | |
7355 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
7356 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
7357 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
7358 | * at arch_max_dma_pfn. | |
7359 | */ | |
9691a071 | 7360 | void __init free_area_init(unsigned long *max_zone_pfn) |
c713216d | 7361 | { |
c13291a5 | 7362 | unsigned long start_pfn, end_pfn; |
51930df5 MR |
7363 | int i, nid, zone; |
7364 | bool descending; | |
a6af2bc3 | 7365 | |
c713216d MG |
7366 | /* Record where the zone boundaries are */ |
7367 | memset(arch_zone_lowest_possible_pfn, 0, | |
7368 | sizeof(arch_zone_lowest_possible_pfn)); | |
7369 | memset(arch_zone_highest_possible_pfn, 0, | |
7370 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
7371 | |
7372 | start_pfn = find_min_pfn_with_active_regions(); | |
51930df5 | 7373 | descending = arch_has_descending_max_zone_pfns(); |
90cae1fe OH |
7374 | |
7375 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
51930df5 MR |
7376 | if (descending) |
7377 | zone = MAX_NR_ZONES - i - 1; | |
7378 | else | |
7379 | zone = i; | |
7380 | ||
7381 | if (zone == ZONE_MOVABLE) | |
2a1e274a | 7382 | continue; |
90cae1fe | 7383 | |
51930df5 MR |
7384 | end_pfn = max(max_zone_pfn[zone], start_pfn); |
7385 | arch_zone_lowest_possible_pfn[zone] = start_pfn; | |
7386 | arch_zone_highest_possible_pfn[zone] = end_pfn; | |
90cae1fe OH |
7387 | |
7388 | start_pfn = end_pfn; | |
c713216d | 7389 | } |
2a1e274a MG |
7390 | |
7391 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
7392 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 7393 | find_zone_movable_pfns_for_nodes(); |
c713216d | 7394 | |
c713216d | 7395 | /* Print out the zone ranges */ |
f88dfff5 | 7396 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
7397 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7398 | if (i == ZONE_MOVABLE) | |
7399 | continue; | |
f88dfff5 | 7400 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
7401 | if (arch_zone_lowest_possible_pfn[i] == |
7402 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 7403 | pr_cont("empty\n"); |
72f0ba02 | 7404 | else |
8d29e18a JG |
7405 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
7406 | (u64)arch_zone_lowest_possible_pfn[i] | |
7407 | << PAGE_SHIFT, | |
7408 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 7409 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
7410 | } |
7411 | ||
7412 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 7413 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
7414 | for (i = 0; i < MAX_NUMNODES; i++) { |
7415 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
7416 | pr_info(" Node %d: %#018Lx\n", i, |
7417 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 7418 | } |
c713216d | 7419 | |
f46edbd1 DW |
7420 | /* |
7421 | * Print out the early node map, and initialize the | |
7422 | * subsection-map relative to active online memory ranges to | |
7423 | * enable future "sub-section" extensions of the memory map. | |
7424 | */ | |
f88dfff5 | 7425 | pr_info("Early memory node ranges\n"); |
f46edbd1 | 7426 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
8d29e18a JG |
7427 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
7428 | (u64)start_pfn << PAGE_SHIFT, | |
7429 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
f46edbd1 DW |
7430 | subsection_map_init(start_pfn, end_pfn - start_pfn); |
7431 | } | |
c713216d MG |
7432 | |
7433 | /* Initialise every node */ | |
708614e6 | 7434 | mminit_verify_pageflags_layout(); |
8ef82866 | 7435 | setup_nr_node_ids(); |
4b094b78 | 7436 | init_unavailable_mem(); |
c713216d MG |
7437 | for_each_online_node(nid) { |
7438 | pg_data_t *pgdat = NODE_DATA(nid); | |
854e8848 | 7439 | free_area_init_node(nid); |
37b07e41 LS |
7440 | |
7441 | /* Any memory on that node */ | |
7442 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
7443 | node_set_state(nid, N_MEMORY); |
7444 | check_for_memory(pgdat, nid); | |
c713216d MG |
7445 | } |
7446 | } | |
2a1e274a | 7447 | |
a5c6d650 DR |
7448 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
7449 | unsigned long *percent) | |
2a1e274a MG |
7450 | { |
7451 | unsigned long long coremem; | |
a5c6d650 DR |
7452 | char *endptr; |
7453 | ||
2a1e274a MG |
7454 | if (!p) |
7455 | return -EINVAL; | |
7456 | ||
a5c6d650 DR |
7457 | /* Value may be a percentage of total memory, otherwise bytes */ |
7458 | coremem = simple_strtoull(p, &endptr, 0); | |
7459 | if (*endptr == '%') { | |
7460 | /* Paranoid check for percent values greater than 100 */ | |
7461 | WARN_ON(coremem > 100); | |
2a1e274a | 7462 | |
a5c6d650 DR |
7463 | *percent = coremem; |
7464 | } else { | |
7465 | coremem = memparse(p, &p); | |
7466 | /* Paranoid check that UL is enough for the coremem value */ | |
7467 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7468 | |
a5c6d650 DR |
7469 | *core = coremem >> PAGE_SHIFT; |
7470 | *percent = 0UL; | |
7471 | } | |
2a1e274a MG |
7472 | return 0; |
7473 | } | |
ed7ed365 | 7474 | |
7e63efef MG |
7475 | /* |
7476 | * kernelcore=size sets the amount of memory for use for allocations that | |
7477 | * cannot be reclaimed or migrated. | |
7478 | */ | |
7479 | static int __init cmdline_parse_kernelcore(char *p) | |
7480 | { | |
342332e6 TI |
7481 | /* parse kernelcore=mirror */ |
7482 | if (parse_option_str(p, "mirror")) { | |
7483 | mirrored_kernelcore = true; | |
7484 | return 0; | |
7485 | } | |
7486 | ||
a5c6d650 DR |
7487 | return cmdline_parse_core(p, &required_kernelcore, |
7488 | &required_kernelcore_percent); | |
7e63efef MG |
7489 | } |
7490 | ||
7491 | /* | |
7492 | * movablecore=size sets the amount of memory for use for allocations that | |
7493 | * can be reclaimed or migrated. | |
7494 | */ | |
7495 | static int __init cmdline_parse_movablecore(char *p) | |
7496 | { | |
a5c6d650 DR |
7497 | return cmdline_parse_core(p, &required_movablecore, |
7498 | &required_movablecore_percent); | |
7e63efef MG |
7499 | } |
7500 | ||
ed7ed365 | 7501 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7502 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7503 | |
c3d5f5f0 JL |
7504 | void adjust_managed_page_count(struct page *page, long count) |
7505 | { | |
9705bea5 | 7506 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 7507 | totalram_pages_add(count); |
3dcc0571 JL |
7508 | #ifdef CONFIG_HIGHMEM |
7509 | if (PageHighMem(page)) | |
ca79b0c2 | 7510 | totalhigh_pages_add(count); |
3dcc0571 | 7511 | #endif |
c3d5f5f0 | 7512 | } |
3dcc0571 | 7513 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7514 | |
e5cb113f | 7515 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 7516 | { |
11199692 JL |
7517 | void *pos; |
7518 | unsigned long pages = 0; | |
69afade7 | 7519 | |
11199692 JL |
7520 | start = (void *)PAGE_ALIGN((unsigned long)start); |
7521 | end = (void *)((unsigned long)end & PAGE_MASK); | |
7522 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
7523 | struct page *page = virt_to_page(pos); |
7524 | void *direct_map_addr; | |
7525 | ||
7526 | /* | |
7527 | * 'direct_map_addr' might be different from 'pos' | |
7528 | * because some architectures' virt_to_page() | |
7529 | * work with aliases. Getting the direct map | |
7530 | * address ensures that we get a _writeable_ | |
7531 | * alias for the memset(). | |
7532 | */ | |
7533 | direct_map_addr = page_address(page); | |
dbe67df4 | 7534 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
7535 | memset(direct_map_addr, poison, PAGE_SIZE); |
7536 | ||
7537 | free_reserved_page(page); | |
69afade7 JL |
7538 | } |
7539 | ||
7540 | if (pages && s) | |
adb1fe9a JP |
7541 | pr_info("Freeing %s memory: %ldK\n", |
7542 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
7543 | |
7544 | return pages; | |
7545 | } | |
7546 | ||
cfa11e08 JL |
7547 | #ifdef CONFIG_HIGHMEM |
7548 | void free_highmem_page(struct page *page) | |
7549 | { | |
7550 | __free_reserved_page(page); | |
ca79b0c2 | 7551 | totalram_pages_inc(); |
9705bea5 | 7552 | atomic_long_inc(&page_zone(page)->managed_pages); |
ca79b0c2 | 7553 | totalhigh_pages_inc(); |
cfa11e08 JL |
7554 | } |
7555 | #endif | |
7556 | ||
7ee3d4e8 JL |
7557 | |
7558 | void __init mem_init_print_info(const char *str) | |
7559 | { | |
7560 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7561 | unsigned long init_code_size, init_data_size; | |
7562 | ||
7563 | physpages = get_num_physpages(); | |
7564 | codesize = _etext - _stext; | |
7565 | datasize = _edata - _sdata; | |
7566 | rosize = __end_rodata - __start_rodata; | |
7567 | bss_size = __bss_stop - __bss_start; | |
7568 | init_data_size = __init_end - __init_begin; | |
7569 | init_code_size = _einittext - _sinittext; | |
7570 | ||
7571 | /* | |
7572 | * Detect special cases and adjust section sizes accordingly: | |
7573 | * 1) .init.* may be embedded into .data sections | |
7574 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7575 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7576 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7577 | */ | |
7578 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7579 | do { \ |
7580 | if (start <= pos && pos < end && size > adj) \ | |
7581 | size -= adj; \ | |
7582 | } while (0) | |
7ee3d4e8 JL |
7583 | |
7584 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7585 | _sinittext, init_code_size); | |
7586 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7587 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7588 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7589 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7590 | ||
7591 | #undef adj_init_size | |
7592 | ||
756a025f | 7593 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7594 | #ifdef CONFIG_HIGHMEM |
756a025f | 7595 | ", %luK highmem" |
7ee3d4e8 | 7596 | #endif |
756a025f JP |
7597 | "%s%s)\n", |
7598 | nr_free_pages() << (PAGE_SHIFT - 10), | |
7599 | physpages << (PAGE_SHIFT - 10), | |
7600 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7601 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
ca79b0c2 | 7602 | (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10), |
756a025f | 7603 | totalcma_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7604 | #ifdef CONFIG_HIGHMEM |
ca79b0c2 | 7605 | totalhigh_pages() << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7606 | #endif |
756a025f | 7607 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
7608 | } |
7609 | ||
0e0b864e | 7610 | /** |
88ca3b94 RD |
7611 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7612 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7613 | * |
013110a7 | 7614 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7615 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7616 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7617 | * function may optionally be used to account for unfreeable pages in the |
7618 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7619 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7620 | */ |
7621 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7622 | { | |
7623 | dma_reserve = new_dma_reserve; | |
7624 | } | |
7625 | ||
005fd4bb | 7626 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7627 | { |
1da177e4 | 7628 | |
005fd4bb SAS |
7629 | lru_add_drain_cpu(cpu); |
7630 | drain_pages(cpu); | |
9f8f2172 | 7631 | |
005fd4bb SAS |
7632 | /* |
7633 | * Spill the event counters of the dead processor | |
7634 | * into the current processors event counters. | |
7635 | * This artificially elevates the count of the current | |
7636 | * processor. | |
7637 | */ | |
7638 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7639 | |
005fd4bb SAS |
7640 | /* |
7641 | * Zero the differential counters of the dead processor | |
7642 | * so that the vm statistics are consistent. | |
7643 | * | |
7644 | * This is only okay since the processor is dead and cannot | |
7645 | * race with what we are doing. | |
7646 | */ | |
7647 | cpu_vm_stats_fold(cpu); | |
7648 | return 0; | |
1da177e4 | 7649 | } |
1da177e4 | 7650 | |
e03a5125 NP |
7651 | #ifdef CONFIG_NUMA |
7652 | int hashdist = HASHDIST_DEFAULT; | |
7653 | ||
7654 | static int __init set_hashdist(char *str) | |
7655 | { | |
7656 | if (!str) | |
7657 | return 0; | |
7658 | hashdist = simple_strtoul(str, &str, 0); | |
7659 | return 1; | |
7660 | } | |
7661 | __setup("hashdist=", set_hashdist); | |
7662 | #endif | |
7663 | ||
1da177e4 LT |
7664 | void __init page_alloc_init(void) |
7665 | { | |
005fd4bb SAS |
7666 | int ret; |
7667 | ||
e03a5125 NP |
7668 | #ifdef CONFIG_NUMA |
7669 | if (num_node_state(N_MEMORY) == 1) | |
7670 | hashdist = 0; | |
7671 | #endif | |
7672 | ||
005fd4bb SAS |
7673 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, |
7674 | "mm/page_alloc:dead", NULL, | |
7675 | page_alloc_cpu_dead); | |
7676 | WARN_ON(ret < 0); | |
1da177e4 LT |
7677 | } |
7678 | ||
cb45b0e9 | 7679 | /* |
34b10060 | 7680 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
7681 | * or min_free_kbytes changes. |
7682 | */ | |
7683 | static void calculate_totalreserve_pages(void) | |
7684 | { | |
7685 | struct pglist_data *pgdat; | |
7686 | unsigned long reserve_pages = 0; | |
2f6726e5 | 7687 | enum zone_type i, j; |
cb45b0e9 HA |
7688 | |
7689 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
7690 | |
7691 | pgdat->totalreserve_pages = 0; | |
7692 | ||
cb45b0e9 HA |
7693 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7694 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 7695 | long max = 0; |
9705bea5 | 7696 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
7697 | |
7698 | /* Find valid and maximum lowmem_reserve in the zone */ | |
7699 | for (j = i; j < MAX_NR_ZONES; j++) { | |
7700 | if (zone->lowmem_reserve[j] > max) | |
7701 | max = zone->lowmem_reserve[j]; | |
7702 | } | |
7703 | ||
41858966 MG |
7704 | /* we treat the high watermark as reserved pages. */ |
7705 | max += high_wmark_pages(zone); | |
cb45b0e9 | 7706 | |
3d6357de AK |
7707 | if (max > managed_pages) |
7708 | max = managed_pages; | |
a8d01437 | 7709 | |
281e3726 | 7710 | pgdat->totalreserve_pages += max; |
a8d01437 | 7711 | |
cb45b0e9 HA |
7712 | reserve_pages += max; |
7713 | } | |
7714 | } | |
7715 | totalreserve_pages = reserve_pages; | |
7716 | } | |
7717 | ||
1da177e4 LT |
7718 | /* |
7719 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 7720 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
7721 | * has a correct pages reserved value, so an adequate number of |
7722 | * pages are left in the zone after a successful __alloc_pages(). | |
7723 | */ | |
7724 | static void setup_per_zone_lowmem_reserve(void) | |
7725 | { | |
7726 | struct pglist_data *pgdat; | |
2f6726e5 | 7727 | enum zone_type j, idx; |
1da177e4 | 7728 | |
ec936fc5 | 7729 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
7730 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7731 | struct zone *zone = pgdat->node_zones + j; | |
9705bea5 | 7732 | unsigned long managed_pages = zone_managed_pages(zone); |
1da177e4 LT |
7733 | |
7734 | zone->lowmem_reserve[j] = 0; | |
7735 | ||
2f6726e5 CL |
7736 | idx = j; |
7737 | while (idx) { | |
1da177e4 LT |
7738 | struct zone *lower_zone; |
7739 | ||
2f6726e5 | 7740 | idx--; |
1da177e4 | 7741 | lower_zone = pgdat->node_zones + idx; |
d3cda233 | 7742 | |
f6366156 BH |
7743 | if (!sysctl_lowmem_reserve_ratio[idx] || |
7744 | !zone_managed_pages(lower_zone)) { | |
d3cda233 | 7745 | lower_zone->lowmem_reserve[j] = 0; |
f6366156 | 7746 | continue; |
d3cda233 JK |
7747 | } else { |
7748 | lower_zone->lowmem_reserve[j] = | |
7749 | managed_pages / sysctl_lowmem_reserve_ratio[idx]; | |
7750 | } | |
9705bea5 | 7751 | managed_pages += zone_managed_pages(lower_zone); |
1da177e4 LT |
7752 | } |
7753 | } | |
7754 | } | |
cb45b0e9 HA |
7755 | |
7756 | /* update totalreserve_pages */ | |
7757 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7758 | } |
7759 | ||
cfd3da1e | 7760 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
7761 | { |
7762 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
7763 | unsigned long lowmem_pages = 0; | |
7764 | struct zone *zone; | |
7765 | unsigned long flags; | |
7766 | ||
7767 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
7768 | for_each_zone(zone) { | |
7769 | if (!is_highmem(zone)) | |
9705bea5 | 7770 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
7771 | } |
7772 | ||
7773 | for_each_zone(zone) { | |
ac924c60 AM |
7774 | u64 tmp; |
7775 | ||
1125b4e3 | 7776 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 7777 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 7778 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
7779 | if (is_highmem(zone)) { |
7780 | /* | |
669ed175 NP |
7781 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
7782 | * need highmem pages, so cap pages_min to a small | |
7783 | * value here. | |
7784 | * | |
41858966 | 7785 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 7786 | * deltas control async page reclaim, and so should |
669ed175 | 7787 | * not be capped for highmem. |
1da177e4 | 7788 | */ |
90ae8d67 | 7789 | unsigned long min_pages; |
1da177e4 | 7790 | |
9705bea5 | 7791 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 7792 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 7793 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 7794 | } else { |
669ed175 NP |
7795 | /* |
7796 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
7797 | * proportionate to the zone's size. |
7798 | */ | |
a9214443 | 7799 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
7800 | } |
7801 | ||
795ae7a0 JW |
7802 | /* |
7803 | * Set the kswapd watermarks distance according to the | |
7804 | * scale factor in proportion to available memory, but | |
7805 | * ensure a minimum size on small systems. | |
7806 | */ | |
7807 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 7808 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
7809 | watermark_scale_factor, 10000)); |
7810 | ||
aa092591 | 7811 | zone->watermark_boost = 0; |
a9214443 MG |
7812 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
7813 | zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 7814 | |
1125b4e3 | 7815 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 7816 | } |
cb45b0e9 HA |
7817 | |
7818 | /* update totalreserve_pages */ | |
7819 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7820 | } |
7821 | ||
cfd3da1e MG |
7822 | /** |
7823 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
7824 | * or when memory is hot-{added|removed} | |
7825 | * | |
7826 | * Ensures that the watermark[min,low,high] values for each zone are set | |
7827 | * correctly with respect to min_free_kbytes. | |
7828 | */ | |
7829 | void setup_per_zone_wmarks(void) | |
7830 | { | |
b93e0f32 MH |
7831 | static DEFINE_SPINLOCK(lock); |
7832 | ||
7833 | spin_lock(&lock); | |
cfd3da1e | 7834 | __setup_per_zone_wmarks(); |
b93e0f32 | 7835 | spin_unlock(&lock); |
cfd3da1e MG |
7836 | } |
7837 | ||
1da177e4 LT |
7838 | /* |
7839 | * Initialise min_free_kbytes. | |
7840 | * | |
7841 | * For small machines we want it small (128k min). For large machines | |
8beeae86 | 7842 | * we want it large (256MB max). But it is not linear, because network |
1da177e4 LT |
7843 | * bandwidth does not increase linearly with machine size. We use |
7844 | * | |
b8af2941 | 7845 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
7846 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
7847 | * | |
7848 | * which yields | |
7849 | * | |
7850 | * 16MB: 512k | |
7851 | * 32MB: 724k | |
7852 | * 64MB: 1024k | |
7853 | * 128MB: 1448k | |
7854 | * 256MB: 2048k | |
7855 | * 512MB: 2896k | |
7856 | * 1024MB: 4096k | |
7857 | * 2048MB: 5792k | |
7858 | * 4096MB: 8192k | |
7859 | * 8192MB: 11584k | |
7860 | * 16384MB: 16384k | |
7861 | */ | |
1b79acc9 | 7862 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
7863 | { |
7864 | unsigned long lowmem_kbytes; | |
5f12733e | 7865 | int new_min_free_kbytes; |
1da177e4 LT |
7866 | |
7867 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
7868 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
7869 | ||
7870 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
7871 | min_free_kbytes = new_min_free_kbytes; | |
7872 | if (min_free_kbytes < 128) | |
7873 | min_free_kbytes = 128; | |
ee8eb9a5 JS |
7874 | if (min_free_kbytes > 262144) |
7875 | min_free_kbytes = 262144; | |
5f12733e MH |
7876 | } else { |
7877 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
7878 | new_min_free_kbytes, user_min_free_kbytes); | |
7879 | } | |
bc75d33f | 7880 | setup_per_zone_wmarks(); |
a6cccdc3 | 7881 | refresh_zone_stat_thresholds(); |
1da177e4 | 7882 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
7883 | |
7884 | #ifdef CONFIG_NUMA | |
7885 | setup_min_unmapped_ratio(); | |
7886 | setup_min_slab_ratio(); | |
7887 | #endif | |
7888 | ||
1da177e4 LT |
7889 | return 0; |
7890 | } | |
bc22af74 | 7891 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
7892 | |
7893 | /* | |
b8af2941 | 7894 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
7895 | * that we can call two helper functions whenever min_free_kbytes |
7896 | * changes. | |
7897 | */ | |
cccad5b9 | 7898 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 7899 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7900 | { |
da8c757b HP |
7901 | int rc; |
7902 | ||
7903 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7904 | if (rc) | |
7905 | return rc; | |
7906 | ||
5f12733e MH |
7907 | if (write) { |
7908 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7909 | setup_per_zone_wmarks(); |
5f12733e | 7910 | } |
1da177e4 LT |
7911 | return 0; |
7912 | } | |
7913 | ||
795ae7a0 | 7914 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 7915 | void *buffer, size_t *length, loff_t *ppos) |
795ae7a0 JW |
7916 | { |
7917 | int rc; | |
7918 | ||
7919 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7920 | if (rc) | |
7921 | return rc; | |
7922 | ||
7923 | if (write) | |
7924 | setup_per_zone_wmarks(); | |
7925 | ||
7926 | return 0; | |
7927 | } | |
7928 | ||
9614634f | 7929 | #ifdef CONFIG_NUMA |
6423aa81 | 7930 | static void setup_min_unmapped_ratio(void) |
9614634f | 7931 | { |
6423aa81 | 7932 | pg_data_t *pgdat; |
9614634f | 7933 | struct zone *zone; |
9614634f | 7934 | |
a5f5f91d | 7935 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7936 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7937 | |
9614634f | 7938 | for_each_zone(zone) |
9705bea5 AK |
7939 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
7940 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 7941 | } |
0ff38490 | 7942 | |
6423aa81 JK |
7943 | |
7944 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 7945 | void *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7946 | { |
0ff38490 CL |
7947 | int rc; |
7948 | ||
8d65af78 | 7949 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7950 | if (rc) |
7951 | return rc; | |
7952 | ||
6423aa81 JK |
7953 | setup_min_unmapped_ratio(); |
7954 | ||
7955 | return 0; | |
7956 | } | |
7957 | ||
7958 | static void setup_min_slab_ratio(void) | |
7959 | { | |
7960 | pg_data_t *pgdat; | |
7961 | struct zone *zone; | |
7962 | ||
a5f5f91d MG |
7963 | for_each_online_pgdat(pgdat) |
7964 | pgdat->min_slab_pages = 0; | |
7965 | ||
0ff38490 | 7966 | for_each_zone(zone) |
9705bea5 AK |
7967 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
7968 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
7969 | } |
7970 | ||
7971 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
32927393 | 7972 | void *buffer, size_t *length, loff_t *ppos) |
6423aa81 JK |
7973 | { |
7974 | int rc; | |
7975 | ||
7976 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7977 | if (rc) | |
7978 | return rc; | |
7979 | ||
7980 | setup_min_slab_ratio(); | |
7981 | ||
0ff38490 CL |
7982 | return 0; |
7983 | } | |
9614634f CL |
7984 | #endif |
7985 | ||
1da177e4 LT |
7986 | /* |
7987 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7988 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7989 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7990 | * | |
7991 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7992 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7993 | * if in function of the boot time zone sizes. |
7994 | */ | |
cccad5b9 | 7995 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 7996 | void *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7997 | { |
86aaf255 BH |
7998 | int i; |
7999 | ||
8d65af78 | 8000 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
86aaf255 BH |
8001 | |
8002 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
8003 | if (sysctl_lowmem_reserve_ratio[i] < 1) | |
8004 | sysctl_lowmem_reserve_ratio[i] = 0; | |
8005 | } | |
8006 | ||
1da177e4 LT |
8007 | setup_per_zone_lowmem_reserve(); |
8008 | return 0; | |
8009 | } | |
8010 | ||
cb1ef534 MG |
8011 | static void __zone_pcp_update(struct zone *zone) |
8012 | { | |
8013 | unsigned int cpu; | |
8014 | ||
8015 | for_each_possible_cpu(cpu) | |
8016 | pageset_set_high_and_batch(zone, | |
8017 | per_cpu_ptr(zone->pageset, cpu)); | |
8018 | } | |
8019 | ||
8ad4b1fb RS |
8020 | /* |
8021 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
8022 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
8023 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 8024 | */ |
cccad5b9 | 8025 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
32927393 | 8026 | void *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
8027 | { |
8028 | struct zone *zone; | |
7cd2b0a3 | 8029 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
8030 | int ret; |
8031 | ||
7cd2b0a3 DR |
8032 | mutex_lock(&pcp_batch_high_lock); |
8033 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
8034 | ||
8d65af78 | 8035 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
8036 | if (!write || ret < 0) |
8037 | goto out; | |
8038 | ||
8039 | /* Sanity checking to avoid pcp imbalance */ | |
8040 | if (percpu_pagelist_fraction && | |
8041 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
8042 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
8043 | ret = -EINVAL; | |
8044 | goto out; | |
8045 | } | |
8046 | ||
8047 | /* No change? */ | |
8048 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
8049 | goto out; | |
c8e251fa | 8050 | |
cb1ef534 MG |
8051 | for_each_populated_zone(zone) |
8052 | __zone_pcp_update(zone); | |
7cd2b0a3 | 8053 | out: |
c8e251fa | 8054 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 8055 | return ret; |
8ad4b1fb RS |
8056 | } |
8057 | ||
f6f34b43 SD |
8058 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
8059 | /* | |
8060 | * Returns the number of pages that arch has reserved but | |
8061 | * is not known to alloc_large_system_hash(). | |
8062 | */ | |
8063 | static unsigned long __init arch_reserved_kernel_pages(void) | |
8064 | { | |
8065 | return 0; | |
8066 | } | |
8067 | #endif | |
8068 | ||
9017217b PT |
8069 | /* |
8070 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
8071 | * machines. As memory size is increased the scale is also increased but at | |
8072 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
8073 | * quadruples the scale is increased by one, which means the size of hash table | |
8074 | * only doubles, instead of quadrupling as well. | |
8075 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
8076 | * makes sense, it is disabled on such platforms. | |
8077 | */ | |
8078 | #if __BITS_PER_LONG > 32 | |
8079 | #define ADAPT_SCALE_BASE (64ul << 30) | |
8080 | #define ADAPT_SCALE_SHIFT 2 | |
8081 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
8082 | #endif | |
8083 | ||
1da177e4 LT |
8084 | /* |
8085 | * allocate a large system hash table from bootmem | |
8086 | * - it is assumed that the hash table must contain an exact power-of-2 | |
8087 | * quantity of entries | |
8088 | * - limit is the number of hash buckets, not the total allocation size | |
8089 | */ | |
8090 | void *__init alloc_large_system_hash(const char *tablename, | |
8091 | unsigned long bucketsize, | |
8092 | unsigned long numentries, | |
8093 | int scale, | |
8094 | int flags, | |
8095 | unsigned int *_hash_shift, | |
8096 | unsigned int *_hash_mask, | |
31fe62b9 TB |
8097 | unsigned long low_limit, |
8098 | unsigned long high_limit) | |
1da177e4 | 8099 | { |
31fe62b9 | 8100 | unsigned long long max = high_limit; |
1da177e4 LT |
8101 | unsigned long log2qty, size; |
8102 | void *table = NULL; | |
3749a8f0 | 8103 | gfp_t gfp_flags; |
ec11408a | 8104 | bool virt; |
1da177e4 LT |
8105 | |
8106 | /* allow the kernel cmdline to have a say */ | |
8107 | if (!numentries) { | |
8108 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 8109 | numentries = nr_kernel_pages; |
f6f34b43 | 8110 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
8111 | |
8112 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
8113 | if (PAGE_SHIFT < 20) | |
8114 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 8115 | |
9017217b PT |
8116 | #if __BITS_PER_LONG > 32 |
8117 | if (!high_limit) { | |
8118 | unsigned long adapt; | |
8119 | ||
8120 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
8121 | adapt <<= ADAPT_SCALE_SHIFT) | |
8122 | scale++; | |
8123 | } | |
8124 | #endif | |
8125 | ||
1da177e4 LT |
8126 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
8127 | if (scale > PAGE_SHIFT) | |
8128 | numentries >>= (scale - PAGE_SHIFT); | |
8129 | else | |
8130 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
8131 | |
8132 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
8133 | if (unlikely(flags & HASH_SMALL)) { |
8134 | /* Makes no sense without HASH_EARLY */ | |
8135 | WARN_ON(!(flags & HASH_EARLY)); | |
8136 | if (!(numentries >> *_hash_shift)) { | |
8137 | numentries = 1UL << *_hash_shift; | |
8138 | BUG_ON(!numentries); | |
8139 | } | |
8140 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 8141 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 8142 | } |
6e692ed3 | 8143 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
8144 | |
8145 | /* limit allocation size to 1/16 total memory by default */ | |
8146 | if (max == 0) { | |
8147 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
8148 | do_div(max, bucketsize); | |
8149 | } | |
074b8517 | 8150 | max = min(max, 0x80000000ULL); |
1da177e4 | 8151 | |
31fe62b9 TB |
8152 | if (numentries < low_limit) |
8153 | numentries = low_limit; | |
1da177e4 LT |
8154 | if (numentries > max) |
8155 | numentries = max; | |
8156 | ||
f0d1b0b3 | 8157 | log2qty = ilog2(numentries); |
1da177e4 | 8158 | |
3749a8f0 | 8159 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 8160 | do { |
ec11408a | 8161 | virt = false; |
1da177e4 | 8162 | size = bucketsize << log2qty; |
ea1f5f37 PT |
8163 | if (flags & HASH_EARLY) { |
8164 | if (flags & HASH_ZERO) | |
26fb3dae | 8165 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 8166 | else |
7e1c4e27 MR |
8167 | table = memblock_alloc_raw(size, |
8168 | SMP_CACHE_BYTES); | |
ec11408a | 8169 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
88dca4ca | 8170 | table = __vmalloc(size, gfp_flags); |
ec11408a | 8171 | virt = true; |
ea1f5f37 | 8172 | } else { |
1037b83b ED |
8173 | /* |
8174 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
8175 | * some pages at the end of hash table which |
8176 | * alloc_pages_exact() automatically does | |
1037b83b | 8177 | */ |
ec11408a NP |
8178 | table = alloc_pages_exact(size, gfp_flags); |
8179 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
8180 | } |
8181 | } while (!table && size > PAGE_SIZE && --log2qty); | |
8182 | ||
8183 | if (!table) | |
8184 | panic("Failed to allocate %s hash table\n", tablename); | |
8185 | ||
ec11408a NP |
8186 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
8187 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
8188 | virt ? "vmalloc" : "linear"); | |
1da177e4 LT |
8189 | |
8190 | if (_hash_shift) | |
8191 | *_hash_shift = log2qty; | |
8192 | if (_hash_mask) | |
8193 | *_hash_mask = (1 << log2qty) - 1; | |
8194 | ||
8195 | return table; | |
8196 | } | |
a117e66e | 8197 | |
a5d76b54 | 8198 | /* |
80934513 | 8199 | * This function checks whether pageblock includes unmovable pages or not. |
80934513 | 8200 | * |
b8af2941 | 8201 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
8202 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
8203 | * check without lock_page also may miss some movable non-lru pages at | |
8204 | * race condition. So you can't expect this function should be exact. | |
4a55c047 QC |
8205 | * |
8206 | * Returns a page without holding a reference. If the caller wants to | |
047b9967 | 8207 | * dereference that page (e.g., dumping), it has to make sure that it |
4a55c047 QC |
8208 | * cannot get removed (e.g., via memory unplug) concurrently. |
8209 | * | |
a5d76b54 | 8210 | */ |
4a55c047 QC |
8211 | struct page *has_unmovable_pages(struct zone *zone, struct page *page, |
8212 | int migratetype, int flags) | |
49ac8255 | 8213 | { |
1a9f2191 QC |
8214 | unsigned long iter = 0; |
8215 | unsigned long pfn = page_to_pfn(page); | |
47118af0 | 8216 | |
49ac8255 | 8217 | /* |
15c30bc0 MH |
8218 | * TODO we could make this much more efficient by not checking every |
8219 | * page in the range if we know all of them are in MOVABLE_ZONE and | |
8220 | * that the movable zone guarantees that pages are migratable but | |
8221 | * the later is not the case right now unfortunatelly. E.g. movablecore | |
8222 | * can still lead to having bootmem allocations in zone_movable. | |
49ac8255 | 8223 | */ |
49ac8255 | 8224 | |
1a9f2191 QC |
8225 | if (is_migrate_cma_page(page)) { |
8226 | /* | |
8227 | * CMA allocations (alloc_contig_range) really need to mark | |
8228 | * isolate CMA pageblocks even when they are not movable in fact | |
8229 | * so consider them movable here. | |
8230 | */ | |
8231 | if (is_migrate_cma(migratetype)) | |
4a55c047 | 8232 | return NULL; |
1a9f2191 | 8233 | |
3d680bdf | 8234 | return page; |
1a9f2191 | 8235 | } |
4da2ce25 | 8236 | |
fe4c86c9 DH |
8237 | for (; iter < pageblock_nr_pages; iter++) { |
8238 | if (!pfn_valid_within(pfn + iter)) | |
49ac8255 | 8239 | continue; |
29723fcc | 8240 | |
fe4c86c9 | 8241 | page = pfn_to_page(pfn + iter); |
c8721bbb | 8242 | |
d7ab3672 | 8243 | if (PageReserved(page)) |
3d680bdf | 8244 | return page; |
d7ab3672 | 8245 | |
9d789999 MH |
8246 | /* |
8247 | * If the zone is movable and we have ruled out all reserved | |
8248 | * pages then it should be reasonably safe to assume the rest | |
8249 | * is movable. | |
8250 | */ | |
8251 | if (zone_idx(zone) == ZONE_MOVABLE) | |
8252 | continue; | |
8253 | ||
c8721bbb NH |
8254 | /* |
8255 | * Hugepages are not in LRU lists, but they're movable. | |
1da2f328 | 8256 | * THPs are on the LRU, but need to be counted as #small pages. |
8bb4e7a2 | 8257 | * We need not scan over tail pages because we don't |
c8721bbb NH |
8258 | * handle each tail page individually in migration. |
8259 | */ | |
1da2f328 | 8260 | if (PageHuge(page) || PageTransCompound(page)) { |
17e2e7d7 OS |
8261 | struct page *head = compound_head(page); |
8262 | unsigned int skip_pages; | |
464c7ffb | 8263 | |
1da2f328 RR |
8264 | if (PageHuge(page)) { |
8265 | if (!hugepage_migration_supported(page_hstate(head))) | |
8266 | return page; | |
8267 | } else if (!PageLRU(head) && !__PageMovable(head)) { | |
3d680bdf | 8268 | return page; |
1da2f328 | 8269 | } |
464c7ffb | 8270 | |
d8c6546b | 8271 | skip_pages = compound_nr(head) - (page - head); |
17e2e7d7 | 8272 | iter += skip_pages - 1; |
c8721bbb NH |
8273 | continue; |
8274 | } | |
8275 | ||
97d255c8 MK |
8276 | /* |
8277 | * We can't use page_count without pin a page | |
8278 | * because another CPU can free compound page. | |
8279 | * This check already skips compound tails of THP | |
0139aa7b | 8280 | * because their page->_refcount is zero at all time. |
97d255c8 | 8281 | */ |
fe896d18 | 8282 | if (!page_ref_count(page)) { |
49ac8255 KH |
8283 | if (PageBuddy(page)) |
8284 | iter += (1 << page_order(page)) - 1; | |
8285 | continue; | |
8286 | } | |
97d255c8 | 8287 | |
b023f468 WC |
8288 | /* |
8289 | * The HWPoisoned page may be not in buddy system, and | |
8290 | * page_count() is not 0. | |
8291 | */ | |
756d25be | 8292 | if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
b023f468 WC |
8293 | continue; |
8294 | ||
aa218795 DH |
8295 | /* |
8296 | * We treat all PageOffline() pages as movable when offlining | |
8297 | * to give drivers a chance to decrement their reference count | |
8298 | * in MEM_GOING_OFFLINE in order to indicate that these pages | |
8299 | * can be offlined as there are no direct references anymore. | |
8300 | * For actually unmovable PageOffline() where the driver does | |
8301 | * not support this, we will fail later when trying to actually | |
8302 | * move these pages that still have a reference count > 0. | |
8303 | * (false negatives in this function only) | |
8304 | */ | |
8305 | if ((flags & MEMORY_OFFLINE) && PageOffline(page)) | |
8306 | continue; | |
8307 | ||
fe4c86c9 | 8308 | if (__PageMovable(page) || PageLRU(page)) |
0efadf48 YX |
8309 | continue; |
8310 | ||
49ac8255 | 8311 | /* |
6b4f7799 JW |
8312 | * If there are RECLAIMABLE pages, we need to check |
8313 | * it. But now, memory offline itself doesn't call | |
8314 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
8315 | */ |
8316 | /* | |
8317 | * If the page is not RAM, page_count()should be 0. | |
8318 | * we don't need more check. This is an _used_ not-movable page. | |
8319 | * | |
8320 | * The problematic thing here is PG_reserved pages. PG_reserved | |
8321 | * is set to both of a memory hole page and a _used_ kernel | |
8322 | * page at boot. | |
8323 | */ | |
3d680bdf | 8324 | return page; |
49ac8255 | 8325 | } |
4a55c047 | 8326 | return NULL; |
49ac8255 KH |
8327 | } |
8328 | ||
8df995f6 | 8329 | #ifdef CONFIG_CONTIG_ALLOC |
041d3a8c MN |
8330 | static unsigned long pfn_max_align_down(unsigned long pfn) |
8331 | { | |
8332 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8333 | pageblock_nr_pages) - 1); | |
8334 | } | |
8335 | ||
8336 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
8337 | { | |
8338 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8339 | pageblock_nr_pages)); | |
8340 | } | |
8341 | ||
041d3a8c | 8342 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
8343 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
8344 | unsigned long start, unsigned long end) | |
041d3a8c MN |
8345 | { |
8346 | /* This function is based on compact_zone() from compaction.c. */ | |
730ec8c0 | 8347 | unsigned int nr_reclaimed; |
041d3a8c MN |
8348 | unsigned long pfn = start; |
8349 | unsigned int tries = 0; | |
8350 | int ret = 0; | |
8b94e0b8 JK |
8351 | struct migration_target_control mtc = { |
8352 | .nid = zone_to_nid(cc->zone), | |
8353 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL, | |
8354 | }; | |
041d3a8c | 8355 | |
be49a6e1 | 8356 | migrate_prep(); |
041d3a8c | 8357 | |
bb13ffeb | 8358 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
8359 | if (fatal_signal_pending(current)) { |
8360 | ret = -EINTR; | |
8361 | break; | |
8362 | } | |
8363 | ||
bb13ffeb MG |
8364 | if (list_empty(&cc->migratepages)) { |
8365 | cc->nr_migratepages = 0; | |
edc2ca61 | 8366 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
8367 | if (!pfn) { |
8368 | ret = -EINTR; | |
8369 | break; | |
8370 | } | |
8371 | tries = 0; | |
8372 | } else if (++tries == 5) { | |
8373 | ret = ret < 0 ? ret : -EBUSY; | |
8374 | break; | |
8375 | } | |
8376 | ||
beb51eaa MK |
8377 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
8378 | &cc->migratepages); | |
8379 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 8380 | |
8b94e0b8 JK |
8381 | ret = migrate_pages(&cc->migratepages, alloc_migration_target, |
8382 | NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE); | |
041d3a8c | 8383 | } |
2a6f5124 SP |
8384 | if (ret < 0) { |
8385 | putback_movable_pages(&cc->migratepages); | |
8386 | return ret; | |
8387 | } | |
8388 | return 0; | |
041d3a8c MN |
8389 | } |
8390 | ||
8391 | /** | |
8392 | * alloc_contig_range() -- tries to allocate given range of pages | |
8393 | * @start: start PFN to allocate | |
8394 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
8395 | * @migratetype: migratetype of the underlaying pageblocks (either |
8396 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
8397 | * in range must have the same migratetype and it must | |
8398 | * be either of the two. | |
ca96b625 | 8399 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
8400 | * |
8401 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 8402 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 8403 | * |
2c7452a0 MK |
8404 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
8405 | * pageblocks in the range. Once isolated, the pageblocks should not | |
8406 | * be modified by others. | |
041d3a8c | 8407 | * |
a862f68a | 8408 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
8409 | * pages which PFN is in [start, end) are allocated for the caller and |
8410 | * need to be freed with free_contig_range(). | |
8411 | */ | |
0815f3d8 | 8412 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 8413 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 8414 | { |
041d3a8c | 8415 | unsigned long outer_start, outer_end; |
d00181b9 KS |
8416 | unsigned int order; |
8417 | int ret = 0; | |
041d3a8c | 8418 | |
bb13ffeb MG |
8419 | struct compact_control cc = { |
8420 | .nr_migratepages = 0, | |
8421 | .order = -1, | |
8422 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 8423 | .mode = MIGRATE_SYNC, |
bb13ffeb | 8424 | .ignore_skip_hint = true, |
2583d671 | 8425 | .no_set_skip_hint = true, |
7dea19f9 | 8426 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 8427 | .alloc_contig = true, |
bb13ffeb MG |
8428 | }; |
8429 | INIT_LIST_HEAD(&cc.migratepages); | |
8430 | ||
041d3a8c MN |
8431 | /* |
8432 | * What we do here is we mark all pageblocks in range as | |
8433 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
8434 | * have different sizes, and due to the way page allocator | |
8435 | * work, we align the range to biggest of the two pages so | |
8436 | * that page allocator won't try to merge buddies from | |
8437 | * different pageblocks and change MIGRATE_ISOLATE to some | |
8438 | * other migration type. | |
8439 | * | |
8440 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
8441 | * migrate the pages from an unaligned range (ie. pages that | |
8442 | * we are interested in). This will put all the pages in | |
8443 | * range back to page allocator as MIGRATE_ISOLATE. | |
8444 | * | |
8445 | * When this is done, we take the pages in range from page | |
8446 | * allocator removing them from the buddy system. This way | |
8447 | * page allocator will never consider using them. | |
8448 | * | |
8449 | * This lets us mark the pageblocks back as | |
8450 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
8451 | * aligned range but not in the unaligned, original range are | |
8452 | * put back to page allocator so that buddy can use them. | |
8453 | */ | |
8454 | ||
8455 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
d381c547 | 8456 | pfn_max_align_up(end), migratetype, 0); |
9b7ea46a | 8457 | if (ret < 0) |
86a595f9 | 8458 | return ret; |
041d3a8c | 8459 | |
8ef5849f JK |
8460 | /* |
8461 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
8462 | * So, just fall through. test_pages_isolated() has a tracepoint |
8463 | * which will report the busy page. | |
8464 | * | |
8465 | * It is possible that busy pages could become available before | |
8466 | * the call to test_pages_isolated, and the range will actually be | |
8467 | * allocated. So, if we fall through be sure to clear ret so that | |
8468 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 8469 | */ |
bb13ffeb | 8470 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 8471 | if (ret && ret != -EBUSY) |
041d3a8c | 8472 | goto done; |
63cd4489 | 8473 | ret =0; |
041d3a8c MN |
8474 | |
8475 | /* | |
8476 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
8477 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
8478 | * more, all pages in [start, end) are free in page allocator. | |
8479 | * What we are going to do is to allocate all pages from | |
8480 | * [start, end) (that is remove them from page allocator). | |
8481 | * | |
8482 | * The only problem is that pages at the beginning and at the | |
8483 | * end of interesting range may be not aligned with pages that | |
8484 | * page allocator holds, ie. they can be part of higher order | |
8485 | * pages. Because of this, we reserve the bigger range and | |
8486 | * once this is done free the pages we are not interested in. | |
8487 | * | |
8488 | * We don't have to hold zone->lock here because the pages are | |
8489 | * isolated thus they won't get removed from buddy. | |
8490 | */ | |
8491 | ||
8492 | lru_add_drain_all(); | |
041d3a8c MN |
8493 | |
8494 | order = 0; | |
8495 | outer_start = start; | |
8496 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
8497 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
8498 | outer_start = start; |
8499 | break; | |
041d3a8c MN |
8500 | } |
8501 | outer_start &= ~0UL << order; | |
8502 | } | |
8503 | ||
8ef5849f JK |
8504 | if (outer_start != start) { |
8505 | order = page_order(pfn_to_page(outer_start)); | |
8506 | ||
8507 | /* | |
8508 | * outer_start page could be small order buddy page and | |
8509 | * it doesn't include start page. Adjust outer_start | |
8510 | * in this case to report failed page properly | |
8511 | * on tracepoint in test_pages_isolated() | |
8512 | */ | |
8513 | if (outer_start + (1UL << order) <= start) | |
8514 | outer_start = start; | |
8515 | } | |
8516 | ||
041d3a8c | 8517 | /* Make sure the range is really isolated. */ |
756d25be | 8518 | if (test_pages_isolated(outer_start, end, 0)) { |
75dddef3 | 8519 | pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n", |
dae803e1 | 8520 | __func__, outer_start, end); |
041d3a8c MN |
8521 | ret = -EBUSY; |
8522 | goto done; | |
8523 | } | |
8524 | ||
49f223a9 | 8525 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 8526 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
8527 | if (!outer_end) { |
8528 | ret = -EBUSY; | |
8529 | goto done; | |
8530 | } | |
8531 | ||
8532 | /* Free head and tail (if any) */ | |
8533 | if (start != outer_start) | |
8534 | free_contig_range(outer_start, start - outer_start); | |
8535 | if (end != outer_end) | |
8536 | free_contig_range(end, outer_end - end); | |
8537 | ||
8538 | done: | |
8539 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 8540 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
8541 | return ret; |
8542 | } | |
255f5985 | 8543 | EXPORT_SYMBOL(alloc_contig_range); |
5e27a2df AK |
8544 | |
8545 | static int __alloc_contig_pages(unsigned long start_pfn, | |
8546 | unsigned long nr_pages, gfp_t gfp_mask) | |
8547 | { | |
8548 | unsigned long end_pfn = start_pfn + nr_pages; | |
8549 | ||
8550 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
8551 | gfp_mask); | |
8552 | } | |
8553 | ||
8554 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
8555 | unsigned long nr_pages) | |
8556 | { | |
8557 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
8558 | struct page *page; | |
8559 | ||
8560 | for (i = start_pfn; i < end_pfn; i++) { | |
8561 | page = pfn_to_online_page(i); | |
8562 | if (!page) | |
8563 | return false; | |
8564 | ||
8565 | if (page_zone(page) != z) | |
8566 | return false; | |
8567 | ||
8568 | if (PageReserved(page)) | |
8569 | return false; | |
8570 | ||
8571 | if (page_count(page) > 0) | |
8572 | return false; | |
8573 | ||
8574 | if (PageHuge(page)) | |
8575 | return false; | |
8576 | } | |
8577 | return true; | |
8578 | } | |
8579 | ||
8580 | static bool zone_spans_last_pfn(const struct zone *zone, | |
8581 | unsigned long start_pfn, unsigned long nr_pages) | |
8582 | { | |
8583 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
8584 | ||
8585 | return zone_spans_pfn(zone, last_pfn); | |
8586 | } | |
8587 | ||
8588 | /** | |
8589 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
8590 | * @nr_pages: Number of contiguous pages to allocate | |
8591 | * @gfp_mask: GFP mask to limit search and used during compaction | |
8592 | * @nid: Target node | |
8593 | * @nodemask: Mask for other possible nodes | |
8594 | * | |
8595 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
8596 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
8597 | * tried for allocation with alloc_contig_range(). This routine is intended | |
8598 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
8599 | * | |
8600 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
8601 | * power of two then the alignment is guaranteed to be to the given nr_pages | |
8602 | * (e.g. 1GB request would be aligned to 1GB). | |
8603 | * | |
8604 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
8605 | * __free_page() on each allocated page. | |
8606 | * | |
8607 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
8608 | */ | |
8609 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
8610 | int nid, nodemask_t *nodemask) | |
8611 | { | |
8612 | unsigned long ret, pfn, flags; | |
8613 | struct zonelist *zonelist; | |
8614 | struct zone *zone; | |
8615 | struct zoneref *z; | |
8616 | ||
8617 | zonelist = node_zonelist(nid, gfp_mask); | |
8618 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
8619 | gfp_zone(gfp_mask), nodemask) { | |
8620 | spin_lock_irqsave(&zone->lock, flags); | |
8621 | ||
8622 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
8623 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
8624 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
8625 | /* | |
8626 | * We release the zone lock here because | |
8627 | * alloc_contig_range() will also lock the zone | |
8628 | * at some point. If there's an allocation | |
8629 | * spinning on this lock, it may win the race | |
8630 | * and cause alloc_contig_range() to fail... | |
8631 | */ | |
8632 | spin_unlock_irqrestore(&zone->lock, flags); | |
8633 | ret = __alloc_contig_pages(pfn, nr_pages, | |
8634 | gfp_mask); | |
8635 | if (!ret) | |
8636 | return pfn_to_page(pfn); | |
8637 | spin_lock_irqsave(&zone->lock, flags); | |
8638 | } | |
8639 | pfn += nr_pages; | |
8640 | } | |
8641 | spin_unlock_irqrestore(&zone->lock, flags); | |
8642 | } | |
8643 | return NULL; | |
8644 | } | |
4eb0716e | 8645 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 8646 | |
4eb0716e | 8647 | void free_contig_range(unsigned long pfn, unsigned int nr_pages) |
041d3a8c | 8648 | { |
bcc2b02f MS |
8649 | unsigned int count = 0; |
8650 | ||
8651 | for (; nr_pages--; pfn++) { | |
8652 | struct page *page = pfn_to_page(pfn); | |
8653 | ||
8654 | count += page_count(page) != 1; | |
8655 | __free_page(page); | |
8656 | } | |
8657 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c | 8658 | } |
255f5985 | 8659 | EXPORT_SYMBOL(free_contig_range); |
041d3a8c | 8660 | |
0a647f38 CS |
8661 | /* |
8662 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
8663 | * page high values need to be recalulated. | |
8664 | */ | |
4ed7e022 JL |
8665 | void __meminit zone_pcp_update(struct zone *zone) |
8666 | { | |
c8e251fa | 8667 | mutex_lock(&pcp_batch_high_lock); |
cb1ef534 | 8668 | __zone_pcp_update(zone); |
c8e251fa | 8669 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 | 8670 | } |
4ed7e022 | 8671 | |
340175b7 JL |
8672 | void zone_pcp_reset(struct zone *zone) |
8673 | { | |
8674 | unsigned long flags; | |
5a883813 MK |
8675 | int cpu; |
8676 | struct per_cpu_pageset *pset; | |
340175b7 JL |
8677 | |
8678 | /* avoid races with drain_pages() */ | |
8679 | local_irq_save(flags); | |
8680 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
8681 | for_each_online_cpu(cpu) { |
8682 | pset = per_cpu_ptr(zone->pageset, cpu); | |
8683 | drain_zonestat(zone, pset); | |
8684 | } | |
340175b7 JL |
8685 | free_percpu(zone->pageset); |
8686 | zone->pageset = &boot_pageset; | |
8687 | } | |
8688 | local_irq_restore(flags); | |
8689 | } | |
8690 | ||
6dcd73d7 | 8691 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 8692 | /* |
b9eb6319 JK |
8693 | * All pages in the range must be in a single zone and isolated |
8694 | * before calling this. | |
0c0e6195 | 8695 | */ |
5557c766 | 8696 | unsigned long |
0c0e6195 KH |
8697 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
8698 | { | |
8699 | struct page *page; | |
8700 | struct zone *zone; | |
0ee5f4f3 | 8701 | unsigned int order; |
0c0e6195 KH |
8702 | unsigned long pfn; |
8703 | unsigned long flags; | |
5557c766 MH |
8704 | unsigned long offlined_pages = 0; |
8705 | ||
0c0e6195 KH |
8706 | /* find the first valid pfn */ |
8707 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
8708 | if (pfn_valid(pfn)) | |
8709 | break; | |
8710 | if (pfn == end_pfn) | |
5557c766 MH |
8711 | return offlined_pages; |
8712 | ||
2d070eab | 8713 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
8714 | zone = page_zone(pfn_to_page(pfn)); |
8715 | spin_lock_irqsave(&zone->lock, flags); | |
8716 | pfn = start_pfn; | |
8717 | while (pfn < end_pfn) { | |
8718 | if (!pfn_valid(pfn)) { | |
8719 | pfn++; | |
8720 | continue; | |
8721 | } | |
8722 | page = pfn_to_page(pfn); | |
b023f468 WC |
8723 | /* |
8724 | * The HWPoisoned page may be not in buddy system, and | |
8725 | * page_count() is not 0. | |
8726 | */ | |
8727 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
8728 | pfn++; | |
5557c766 | 8729 | offlined_pages++; |
b023f468 WC |
8730 | continue; |
8731 | } | |
aa218795 DH |
8732 | /* |
8733 | * At this point all remaining PageOffline() pages have a | |
8734 | * reference count of 0 and can simply be skipped. | |
8735 | */ | |
8736 | if (PageOffline(page)) { | |
8737 | BUG_ON(page_count(page)); | |
8738 | BUG_ON(PageBuddy(page)); | |
8739 | pfn++; | |
8740 | offlined_pages++; | |
8741 | continue; | |
8742 | } | |
b023f468 | 8743 | |
0c0e6195 KH |
8744 | BUG_ON(page_count(page)); |
8745 | BUG_ON(!PageBuddy(page)); | |
8746 | order = page_order(page); | |
5557c766 | 8747 | offlined_pages += 1 << order; |
6ab01363 | 8748 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
8749 | pfn += (1 << order); |
8750 | } | |
8751 | spin_unlock_irqrestore(&zone->lock, flags); | |
5557c766 MH |
8752 | |
8753 | return offlined_pages; | |
0c0e6195 KH |
8754 | } |
8755 | #endif | |
8d22ba1b | 8756 | |
8d22ba1b WF |
8757 | bool is_free_buddy_page(struct page *page) |
8758 | { | |
8759 | struct zone *zone = page_zone(page); | |
8760 | unsigned long pfn = page_to_pfn(page); | |
8761 | unsigned long flags; | |
7aeb09f9 | 8762 | unsigned int order; |
8d22ba1b WF |
8763 | |
8764 | spin_lock_irqsave(&zone->lock, flags); | |
8765 | for (order = 0; order < MAX_ORDER; order++) { | |
8766 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8767 | ||
8768 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
8769 | break; | |
8770 | } | |
8771 | spin_unlock_irqrestore(&zone->lock, flags); | |
8772 | ||
8773 | return order < MAX_ORDER; | |
8774 | } | |
d4ae9916 NH |
8775 | |
8776 | #ifdef CONFIG_MEMORY_FAILURE | |
8777 | /* | |
8778 | * Set PG_hwpoison flag if a given page is confirmed to be a free page. This | |
8779 | * test is performed under the zone lock to prevent a race against page | |
8780 | * allocation. | |
8781 | */ | |
8782 | bool set_hwpoison_free_buddy_page(struct page *page) | |
8783 | { | |
8784 | struct zone *zone = page_zone(page); | |
8785 | unsigned long pfn = page_to_pfn(page); | |
8786 | unsigned long flags; | |
8787 | unsigned int order; | |
8788 | bool hwpoisoned = false; | |
8789 | ||
8790 | spin_lock_irqsave(&zone->lock, flags); | |
8791 | for (order = 0; order < MAX_ORDER; order++) { | |
8792 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8793 | ||
8794 | if (PageBuddy(page_head) && page_order(page_head) >= order) { | |
8795 | if (!TestSetPageHWPoison(page)) | |
8796 | hwpoisoned = true; | |
8797 | break; | |
8798 | } | |
8799 | } | |
8800 | spin_unlock_irqrestore(&zone->lock, flags); | |
8801 | ||
8802 | return hwpoisoned; | |
8803 | } | |
8804 | #endif |