]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/page_alloc.c | |
4 | * | |
5 | * Manages the free list, the system allocates free pages here. | |
6 | * Note that kmalloc() lives in slab.c | |
7 | * | |
8 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
9 | * Swap reorganised 29.12.95, Stephen Tweedie | |
10 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
11 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
12 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
13 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
14 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
15 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
16 | */ | |
17 | ||
1da177e4 LT |
18 | #include <linux/stddef.h> |
19 | #include <linux/mm.h> | |
ca79b0c2 | 20 | #include <linux/highmem.h> |
1da177e4 LT |
21 | #include <linux/swap.h> |
22 | #include <linux/interrupt.h> | |
23 | #include <linux/pagemap.h> | |
10ed273f | 24 | #include <linux/jiffies.h> |
edbe7d23 | 25 | #include <linux/memblock.h> |
1da177e4 | 26 | #include <linux/compiler.h> |
9f158333 | 27 | #include <linux/kernel.h> |
b8c73fc2 | 28 | #include <linux/kasan.h> |
1da177e4 LT |
29 | #include <linux/module.h> |
30 | #include <linux/suspend.h> | |
31 | #include <linux/pagevec.h> | |
32 | #include <linux/blkdev.h> | |
33 | #include <linux/slab.h> | |
a238ab5b | 34 | #include <linux/ratelimit.h> |
5a3135c2 | 35 | #include <linux/oom.h> |
1da177e4 LT |
36 | #include <linux/topology.h> |
37 | #include <linux/sysctl.h> | |
38 | #include <linux/cpu.h> | |
39 | #include <linux/cpuset.h> | |
bdc8cb98 | 40 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
41 | #include <linux/nodemask.h> |
42 | #include <linux/vmalloc.h> | |
a6cccdc3 | 43 | #include <linux/vmstat.h> |
4be38e35 | 44 | #include <linux/mempolicy.h> |
4b94ffdc | 45 | #include <linux/memremap.h> |
6811378e | 46 | #include <linux/stop_machine.h> |
97500a4a | 47 | #include <linux/random.h> |
c713216d MG |
48 | #include <linux/sort.h> |
49 | #include <linux/pfn.h> | |
3fcfab16 | 50 | #include <linux/backing-dev.h> |
933e312e | 51 | #include <linux/fault-inject.h> |
a5d76b54 | 52 | #include <linux/page-isolation.h> |
3ac7fe5a | 53 | #include <linux/debugobjects.h> |
dbb1f81c | 54 | #include <linux/kmemleak.h> |
56de7263 | 55 | #include <linux/compaction.h> |
0d3d062a | 56 | #include <trace/events/kmem.h> |
d379f01d | 57 | #include <trace/events/oom.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
041d3a8c | 60 | #include <linux/migrate.h> |
949f7ec5 | 61 | #include <linux/hugetlb.h> |
8bd75c77 | 62 | #include <linux/sched/rt.h> |
5b3cc15a | 63 | #include <linux/sched/mm.h> |
48c96a36 | 64 | #include <linux/page_owner.h> |
0e1cc95b | 65 | #include <linux/kthread.h> |
4949148a | 66 | #include <linux/memcontrol.h> |
42c269c8 | 67 | #include <linux/ftrace.h> |
d92a8cfc | 68 | #include <linux/lockdep.h> |
556b969a | 69 | #include <linux/nmi.h> |
eb414681 | 70 | #include <linux/psi.h> |
1da177e4 | 71 | |
7ee3d4e8 | 72 | #include <asm/sections.h> |
1da177e4 | 73 | #include <asm/tlbflush.h> |
ac924c60 | 74 | #include <asm/div64.h> |
1da177e4 | 75 | #include "internal.h" |
e900a918 | 76 | #include "shuffle.h" |
36e66c55 | 77 | #include "page_reporting.h" |
1da177e4 | 78 | |
c8e251fa CS |
79 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
80 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 81 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 82 | |
72812019 LS |
83 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
84 | DEFINE_PER_CPU(int, numa_node); | |
85 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
86 | #endif | |
87 | ||
4518085e KW |
88 | DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key); |
89 | ||
7aac7898 LS |
90 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
91 | /* | |
92 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
93 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
94 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
95 | * defined in <linux/topology.h>. | |
96 | */ | |
97 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
98 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
99 | #endif | |
100 | ||
bd233f53 | 101 | /* work_structs for global per-cpu drains */ |
d9367bd0 WY |
102 | struct pcpu_drain { |
103 | struct zone *zone; | |
104 | struct work_struct work; | |
105 | }; | |
8b885f53 JY |
106 | static DEFINE_MUTEX(pcpu_drain_mutex); |
107 | static DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain); | |
bd233f53 | 108 | |
38addce8 | 109 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
58bea414 | 110 | volatile unsigned long latent_entropy __latent_entropy; |
38addce8 ER |
111 | EXPORT_SYMBOL(latent_entropy); |
112 | #endif | |
113 | ||
1da177e4 | 114 | /* |
13808910 | 115 | * Array of node states. |
1da177e4 | 116 | */ |
13808910 CL |
117 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
118 | [N_POSSIBLE] = NODE_MASK_ALL, | |
119 | [N_ONLINE] = { { [0] = 1UL } }, | |
120 | #ifndef CONFIG_NUMA | |
121 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
122 | #ifdef CONFIG_HIGHMEM | |
123 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b | 124 | #endif |
20b2f52b | 125 | [N_MEMORY] = { { [0] = 1UL } }, |
13808910 CL |
126 | [N_CPU] = { { [0] = 1UL } }, |
127 | #endif /* NUMA */ | |
128 | }; | |
129 | EXPORT_SYMBOL(node_states); | |
130 | ||
ca79b0c2 AK |
131 | atomic_long_t _totalram_pages __read_mostly; |
132 | EXPORT_SYMBOL(_totalram_pages); | |
cb45b0e9 | 133 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 134 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 135 | |
1b76b02f | 136 | int percpu_pagelist_fraction; |
dcce284a | 137 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
6471384a AP |
138 | #ifdef CONFIG_INIT_ON_ALLOC_DEFAULT_ON |
139 | DEFINE_STATIC_KEY_TRUE(init_on_alloc); | |
140 | #else | |
141 | DEFINE_STATIC_KEY_FALSE(init_on_alloc); | |
142 | #endif | |
143 | EXPORT_SYMBOL(init_on_alloc); | |
144 | ||
145 | #ifdef CONFIG_INIT_ON_FREE_DEFAULT_ON | |
146 | DEFINE_STATIC_KEY_TRUE(init_on_free); | |
147 | #else | |
148 | DEFINE_STATIC_KEY_FALSE(init_on_free); | |
149 | #endif | |
150 | EXPORT_SYMBOL(init_on_free); | |
151 | ||
152 | static int __init early_init_on_alloc(char *buf) | |
153 | { | |
154 | int ret; | |
155 | bool bool_result; | |
156 | ||
157 | if (!buf) | |
158 | return -EINVAL; | |
159 | ret = kstrtobool(buf, &bool_result); | |
160 | if (bool_result && page_poisoning_enabled()) | |
161 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_alloc\n"); | |
162 | if (bool_result) | |
163 | static_branch_enable(&init_on_alloc); | |
164 | else | |
165 | static_branch_disable(&init_on_alloc); | |
166 | return ret; | |
167 | } | |
168 | early_param("init_on_alloc", early_init_on_alloc); | |
169 | ||
170 | static int __init early_init_on_free(char *buf) | |
171 | { | |
172 | int ret; | |
173 | bool bool_result; | |
174 | ||
175 | if (!buf) | |
176 | return -EINVAL; | |
177 | ret = kstrtobool(buf, &bool_result); | |
178 | if (bool_result && page_poisoning_enabled()) | |
179 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_free\n"); | |
180 | if (bool_result) | |
181 | static_branch_enable(&init_on_free); | |
182 | else | |
183 | static_branch_disable(&init_on_free); | |
184 | return ret; | |
185 | } | |
186 | early_param("init_on_free", early_init_on_free); | |
1da177e4 | 187 | |
bb14c2c7 VB |
188 | /* |
189 | * A cached value of the page's pageblock's migratetype, used when the page is | |
190 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
191 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
192 | * Also the migratetype set in the page does not necessarily match the pcplist | |
193 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
194 | * other index - this ensures that it will be put on the correct CMA freelist. | |
195 | */ | |
196 | static inline int get_pcppage_migratetype(struct page *page) | |
197 | { | |
198 | return page->index; | |
199 | } | |
200 | ||
201 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
202 | { | |
203 | page->index = migratetype; | |
204 | } | |
205 | ||
452aa699 RW |
206 | #ifdef CONFIG_PM_SLEEP |
207 | /* | |
208 | * The following functions are used by the suspend/hibernate code to temporarily | |
209 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
210 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
55f2503c PL |
211 | * they should always be called with system_transition_mutex held |
212 | * (gfp_allowed_mask also should only be modified with system_transition_mutex | |
213 | * held, unless the suspend/hibernate code is guaranteed not to run in parallel | |
214 | * with that modification). | |
452aa699 | 215 | */ |
c9e664f1 RW |
216 | |
217 | static gfp_t saved_gfp_mask; | |
218 | ||
219 | void pm_restore_gfp_mask(void) | |
452aa699 | 220 | { |
55f2503c | 221 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
222 | if (saved_gfp_mask) { |
223 | gfp_allowed_mask = saved_gfp_mask; | |
224 | saved_gfp_mask = 0; | |
225 | } | |
452aa699 RW |
226 | } |
227 | ||
c9e664f1 | 228 | void pm_restrict_gfp_mask(void) |
452aa699 | 229 | { |
55f2503c | 230 | WARN_ON(!mutex_is_locked(&system_transition_mutex)); |
c9e664f1 RW |
231 | WARN_ON(saved_gfp_mask); |
232 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 233 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 234 | } |
f90ac398 MG |
235 | |
236 | bool pm_suspended_storage(void) | |
237 | { | |
d0164adc | 238 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
239 | return false; |
240 | return true; | |
241 | } | |
452aa699 RW |
242 | #endif /* CONFIG_PM_SLEEP */ |
243 | ||
d9c23400 | 244 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 245 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
246 | #endif |
247 | ||
d98c7a09 | 248 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 249 | |
1da177e4 LT |
250 | /* |
251 | * results with 256, 32 in the lowmem_reserve sysctl: | |
252 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
253 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
254 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
255 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 256 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
257 | * |
258 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
259 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 260 | */ |
d3cda233 | 261 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = { |
4b51d669 | 262 | #ifdef CONFIG_ZONE_DMA |
d3cda233 | 263 | [ZONE_DMA] = 256, |
4b51d669 | 264 | #endif |
fb0e7942 | 265 | #ifdef CONFIG_ZONE_DMA32 |
d3cda233 | 266 | [ZONE_DMA32] = 256, |
fb0e7942 | 267 | #endif |
d3cda233 | 268 | [ZONE_NORMAL] = 32, |
e53ef38d | 269 | #ifdef CONFIG_HIGHMEM |
d3cda233 | 270 | [ZONE_HIGHMEM] = 0, |
e53ef38d | 271 | #endif |
d3cda233 | 272 | [ZONE_MOVABLE] = 0, |
2f1b6248 | 273 | }; |
1da177e4 | 274 | |
15ad7cdc | 275 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 276 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 277 | "DMA", |
4b51d669 | 278 | #endif |
fb0e7942 | 279 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 280 | "DMA32", |
fb0e7942 | 281 | #endif |
2f1b6248 | 282 | "Normal", |
e53ef38d | 283 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 284 | "HighMem", |
e53ef38d | 285 | #endif |
2a1e274a | 286 | "Movable", |
033fbae9 DW |
287 | #ifdef CONFIG_ZONE_DEVICE |
288 | "Device", | |
289 | #endif | |
2f1b6248 CL |
290 | }; |
291 | ||
c999fbd3 | 292 | const char * const migratetype_names[MIGRATE_TYPES] = { |
60f30350 VB |
293 | "Unmovable", |
294 | "Movable", | |
295 | "Reclaimable", | |
296 | "HighAtomic", | |
297 | #ifdef CONFIG_CMA | |
298 | "CMA", | |
299 | #endif | |
300 | #ifdef CONFIG_MEMORY_ISOLATION | |
301 | "Isolate", | |
302 | #endif | |
303 | }; | |
304 | ||
f1e61557 KS |
305 | compound_page_dtor * const compound_page_dtors[] = { |
306 | NULL, | |
307 | free_compound_page, | |
308 | #ifdef CONFIG_HUGETLB_PAGE | |
309 | free_huge_page, | |
310 | #endif | |
9a982250 KS |
311 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
312 | free_transhuge_page, | |
313 | #endif | |
f1e61557 KS |
314 | }; |
315 | ||
1da177e4 | 316 | int min_free_kbytes = 1024; |
42aa83cb | 317 | int user_min_free_kbytes = -1; |
24512228 MG |
318 | #ifdef CONFIG_DISCONTIGMEM |
319 | /* | |
320 | * DiscontigMem defines memory ranges as separate pg_data_t even if the ranges | |
321 | * are not on separate NUMA nodes. Functionally this works but with | |
322 | * watermark_boost_factor, it can reclaim prematurely as the ranges can be | |
323 | * quite small. By default, do not boost watermarks on discontigmem as in | |
324 | * many cases very high-order allocations like THP are likely to be | |
325 | * unsupported and the premature reclaim offsets the advantage of long-term | |
326 | * fragmentation avoidance. | |
327 | */ | |
328 | int watermark_boost_factor __read_mostly; | |
329 | #else | |
1c30844d | 330 | int watermark_boost_factor __read_mostly = 15000; |
24512228 | 331 | #endif |
795ae7a0 | 332 | int watermark_scale_factor = 10; |
1da177e4 | 333 | |
bbe5d993 OS |
334 | static unsigned long nr_kernel_pages __initdata; |
335 | static unsigned long nr_all_pages __initdata; | |
336 | static unsigned long dma_reserve __initdata; | |
1da177e4 | 337 | |
bbe5d993 OS |
338 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata; |
339 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata; | |
7f16f91f | 340 | static unsigned long required_kernelcore __initdata; |
a5c6d650 | 341 | static unsigned long required_kernelcore_percent __initdata; |
7f16f91f | 342 | static unsigned long required_movablecore __initdata; |
a5c6d650 | 343 | static unsigned long required_movablecore_percent __initdata; |
bbe5d993 | 344 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata; |
7f16f91f | 345 | static bool mirrored_kernelcore __meminitdata; |
0ee332c1 TH |
346 | |
347 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
348 | int movable_zone; | |
349 | EXPORT_SYMBOL(movable_zone); | |
c713216d | 350 | |
418508c1 | 351 | #if MAX_NUMNODES > 1 |
b9726c26 | 352 | unsigned int nr_node_ids __read_mostly = MAX_NUMNODES; |
ce0725f7 | 353 | unsigned int nr_online_nodes __read_mostly = 1; |
418508c1 | 354 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 355 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
356 | #endif |
357 | ||
9ef9acb0 MG |
358 | int page_group_by_mobility_disabled __read_mostly; |
359 | ||
3a80a7fa | 360 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3c0c12cc WL |
361 | /* |
362 | * During boot we initialize deferred pages on-demand, as needed, but once | |
363 | * page_alloc_init_late() has finished, the deferred pages are all initialized, | |
364 | * and we can permanently disable that path. | |
365 | */ | |
366 | static DEFINE_STATIC_KEY_TRUE(deferred_pages); | |
367 | ||
368 | /* | |
369 | * Calling kasan_free_pages() only after deferred memory initialization | |
370 | * has completed. Poisoning pages during deferred memory init will greatly | |
371 | * lengthen the process and cause problem in large memory systems as the | |
372 | * deferred pages initialization is done with interrupt disabled. | |
373 | * | |
374 | * Assuming that there will be no reference to those newly initialized | |
375 | * pages before they are ever allocated, this should have no effect on | |
376 | * KASAN memory tracking as the poison will be properly inserted at page | |
377 | * allocation time. The only corner case is when pages are allocated by | |
378 | * on-demand allocation and then freed again before the deferred pages | |
379 | * initialization is done, but this is not likely to happen. | |
380 | */ | |
381 | static inline void kasan_free_nondeferred_pages(struct page *page, int order) | |
382 | { | |
383 | if (!static_branch_unlikely(&deferred_pages)) | |
384 | kasan_free_pages(page, order); | |
385 | } | |
386 | ||
3a80a7fa | 387 | /* Returns true if the struct page for the pfn is uninitialised */ |
0e1cc95b | 388 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 389 | { |
ef70b6f4 MG |
390 | int nid = early_pfn_to_nid(pfn); |
391 | ||
392 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
3a80a7fa MG |
393 | return true; |
394 | ||
395 | return false; | |
396 | } | |
397 | ||
398 | /* | |
d3035be4 | 399 | * Returns true when the remaining initialisation should be deferred until |
3a80a7fa MG |
400 | * later in the boot cycle when it can be parallelised. |
401 | */ | |
d3035be4 PT |
402 | static bool __meminit |
403 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn) | |
3a80a7fa | 404 | { |
d3035be4 PT |
405 | static unsigned long prev_end_pfn, nr_initialised; |
406 | ||
407 | /* | |
408 | * prev_end_pfn static that contains the end of previous zone | |
409 | * No need to protect because called very early in boot before smp_init. | |
410 | */ | |
411 | if (prev_end_pfn != end_pfn) { | |
412 | prev_end_pfn = end_pfn; | |
413 | nr_initialised = 0; | |
414 | } | |
415 | ||
3c2c6488 | 416 | /* Always populate low zones for address-constrained allocations */ |
d3035be4 | 417 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid))) |
3a80a7fa | 418 | return false; |
23b68cfa WY |
419 | |
420 | /* | |
421 | * We start only with one section of pages, more pages are added as | |
422 | * needed until the rest of deferred pages are initialized. | |
423 | */ | |
d3035be4 | 424 | nr_initialised++; |
23b68cfa | 425 | if ((nr_initialised > PAGES_PER_SECTION) && |
d3035be4 PT |
426 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
427 | NODE_DATA(nid)->first_deferred_pfn = pfn; | |
428 | return true; | |
3a80a7fa | 429 | } |
d3035be4 | 430 | return false; |
3a80a7fa MG |
431 | } |
432 | #else | |
3c0c12cc WL |
433 | #define kasan_free_nondeferred_pages(p, o) kasan_free_pages(p, o) |
434 | ||
3a80a7fa MG |
435 | static inline bool early_page_uninitialised(unsigned long pfn) |
436 | { | |
437 | return false; | |
438 | } | |
439 | ||
d3035be4 | 440 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn) |
3a80a7fa | 441 | { |
d3035be4 | 442 | return false; |
3a80a7fa MG |
443 | } |
444 | #endif | |
445 | ||
0b423ca2 MG |
446 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
447 | static inline unsigned long *get_pageblock_bitmap(struct page *page, | |
448 | unsigned long pfn) | |
449 | { | |
450 | #ifdef CONFIG_SPARSEMEM | |
f1eca35a | 451 | return section_to_usemap(__pfn_to_section(pfn)); |
0b423ca2 MG |
452 | #else |
453 | return page_zone(page)->pageblock_flags; | |
454 | #endif /* CONFIG_SPARSEMEM */ | |
455 | } | |
456 | ||
457 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) | |
458 | { | |
459 | #ifdef CONFIG_SPARSEMEM | |
460 | pfn &= (PAGES_PER_SECTION-1); | |
461 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
462 | #else | |
463 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); | |
464 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
465 | #endif /* CONFIG_SPARSEMEM */ | |
466 | } | |
467 | ||
468 | /** | |
469 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages | |
470 | * @page: The page within the block of interest | |
471 | * @pfn: The target page frame number | |
472 | * @end_bitidx: The last bit of interest to retrieve | |
473 | * @mask: mask of bits that the caller is interested in | |
474 | * | |
475 | * Return: pageblock_bits flags | |
476 | */ | |
477 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, | |
478 | unsigned long pfn, | |
479 | unsigned long end_bitidx, | |
480 | unsigned long mask) | |
481 | { | |
482 | unsigned long *bitmap; | |
483 | unsigned long bitidx, word_bitidx; | |
484 | unsigned long word; | |
485 | ||
486 | bitmap = get_pageblock_bitmap(page, pfn); | |
487 | bitidx = pfn_to_bitidx(page, pfn); | |
488 | word_bitidx = bitidx / BITS_PER_LONG; | |
489 | bitidx &= (BITS_PER_LONG-1); | |
490 | ||
491 | word = bitmap[word_bitidx]; | |
492 | bitidx += end_bitidx; | |
493 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
494 | } | |
495 | ||
496 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, | |
497 | unsigned long end_bitidx, | |
498 | unsigned long mask) | |
499 | { | |
500 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); | |
501 | } | |
502 | ||
503 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
504 | { | |
505 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); | |
506 | } | |
507 | ||
508 | /** | |
509 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages | |
510 | * @page: The page within the block of interest | |
511 | * @flags: The flags to set | |
512 | * @pfn: The target page frame number | |
513 | * @end_bitidx: The last bit of interest | |
514 | * @mask: mask of bits that the caller is interested in | |
515 | */ | |
516 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, | |
517 | unsigned long pfn, | |
518 | unsigned long end_bitidx, | |
519 | unsigned long mask) | |
520 | { | |
521 | unsigned long *bitmap; | |
522 | unsigned long bitidx, word_bitidx; | |
523 | unsigned long old_word, word; | |
524 | ||
525 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
125b860b | 526 | BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits)); |
0b423ca2 MG |
527 | |
528 | bitmap = get_pageblock_bitmap(page, pfn); | |
529 | bitidx = pfn_to_bitidx(page, pfn); | |
530 | word_bitidx = bitidx / BITS_PER_LONG; | |
531 | bitidx &= (BITS_PER_LONG-1); | |
532 | ||
533 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); | |
534 | ||
535 | bitidx += end_bitidx; | |
536 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
537 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
538 | ||
539 | word = READ_ONCE(bitmap[word_bitidx]); | |
540 | for (;;) { | |
541 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
542 | if (word == old_word) | |
543 | break; | |
544 | word = old_word; | |
545 | } | |
546 | } | |
3a80a7fa | 547 | |
ee6f509c | 548 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 549 | { |
5d0f3f72 KM |
550 | if (unlikely(page_group_by_mobility_disabled && |
551 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
552 | migratetype = MIGRATE_UNMOVABLE; |
553 | ||
b2a0ac88 MG |
554 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
555 | PB_migrate, PB_migrate_end); | |
556 | } | |
557 | ||
13e7444b | 558 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 559 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 560 | { |
bdc8cb98 DH |
561 | int ret = 0; |
562 | unsigned seq; | |
563 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 564 | unsigned long sp, start_pfn; |
c6a57e19 | 565 | |
bdc8cb98 DH |
566 | do { |
567 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
568 | start_pfn = zone->zone_start_pfn; |
569 | sp = zone->spanned_pages; | |
108bcc96 | 570 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
571 | ret = 1; |
572 | } while (zone_span_seqretry(zone, seq)); | |
573 | ||
b5e6a5a2 | 574 | if (ret) |
613813e8 DH |
575 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
576 | pfn, zone_to_nid(zone), zone->name, | |
577 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 578 | |
bdc8cb98 | 579 | return ret; |
c6a57e19 DH |
580 | } |
581 | ||
582 | static int page_is_consistent(struct zone *zone, struct page *page) | |
583 | { | |
14e07298 | 584 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 585 | return 0; |
1da177e4 | 586 | if (zone != page_zone(page)) |
c6a57e19 DH |
587 | return 0; |
588 | ||
589 | return 1; | |
590 | } | |
591 | /* | |
592 | * Temporary debugging check for pages not lying within a given zone. | |
593 | */ | |
d73d3c9f | 594 | static int __maybe_unused bad_range(struct zone *zone, struct page *page) |
c6a57e19 DH |
595 | { |
596 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 597 | return 1; |
c6a57e19 DH |
598 | if (!page_is_consistent(zone, page)) |
599 | return 1; | |
600 | ||
1da177e4 LT |
601 | return 0; |
602 | } | |
13e7444b | 603 | #else |
d73d3c9f | 604 | static inline int __maybe_unused bad_range(struct zone *zone, struct page *page) |
13e7444b NP |
605 | { |
606 | return 0; | |
607 | } | |
608 | #endif | |
609 | ||
d230dec1 KS |
610 | static void bad_page(struct page *page, const char *reason, |
611 | unsigned long bad_flags) | |
1da177e4 | 612 | { |
d936cf9b HD |
613 | static unsigned long resume; |
614 | static unsigned long nr_shown; | |
615 | static unsigned long nr_unshown; | |
616 | ||
617 | /* | |
618 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
619 | * or allow a steady drip of one report per second. | |
620 | */ | |
621 | if (nr_shown == 60) { | |
622 | if (time_before(jiffies, resume)) { | |
623 | nr_unshown++; | |
624 | goto out; | |
625 | } | |
626 | if (nr_unshown) { | |
ff8e8116 | 627 | pr_alert( |
1e9e6365 | 628 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
629 | nr_unshown); |
630 | nr_unshown = 0; | |
631 | } | |
632 | nr_shown = 0; | |
633 | } | |
634 | if (nr_shown++ == 0) | |
635 | resume = jiffies + 60 * HZ; | |
636 | ||
ff8e8116 | 637 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 638 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
639 | __dump_page(page, reason); |
640 | bad_flags &= page->flags; | |
641 | if (bad_flags) | |
642 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
643 | bad_flags, &bad_flags); | |
4e462112 | 644 | dump_page_owner(page); |
3dc14741 | 645 | |
4f31888c | 646 | print_modules(); |
1da177e4 | 647 | dump_stack(); |
d936cf9b | 648 | out: |
8cc3b392 | 649 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 650 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 651 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
652 | } |
653 | ||
1da177e4 LT |
654 | /* |
655 | * Higher-order pages are called "compound pages". They are structured thusly: | |
656 | * | |
1d798ca3 | 657 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 658 | * |
1d798ca3 KS |
659 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
660 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 661 | * |
1d798ca3 KS |
662 | * The first tail page's ->compound_dtor holds the offset in array of compound |
663 | * page destructors. See compound_page_dtors. | |
1da177e4 | 664 | * |
1d798ca3 | 665 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 666 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 667 | */ |
d98c7a09 | 668 | |
9a982250 | 669 | void free_compound_page(struct page *page) |
d98c7a09 | 670 | { |
7ae88534 | 671 | mem_cgroup_uncharge(page); |
d85f3385 | 672 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
673 | } |
674 | ||
d00181b9 | 675 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
676 | { |
677 | int i; | |
678 | int nr_pages = 1 << order; | |
679 | ||
f1e61557 | 680 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
681 | set_compound_order(page, order); |
682 | __SetPageHead(page); | |
683 | for (i = 1; i < nr_pages; i++) { | |
684 | struct page *p = page + i; | |
58a84aa9 | 685 | set_page_count(p, 0); |
1c290f64 | 686 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 687 | set_compound_head(p, page); |
18229df5 | 688 | } |
53f9263b | 689 | atomic_set(compound_mapcount_ptr(page), -1); |
47e29d32 JH |
690 | if (hpage_pincount_available(page)) |
691 | atomic_set(compound_pincount_ptr(page), 0); | |
18229df5 AW |
692 | } |
693 | ||
c0a32fc5 SG |
694 | #ifdef CONFIG_DEBUG_PAGEALLOC |
695 | unsigned int _debug_guardpage_minorder; | |
96a2b03f | 696 | |
8e57f8ac VB |
697 | bool _debug_pagealloc_enabled_early __read_mostly |
698 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
699 | EXPORT_SYMBOL(_debug_pagealloc_enabled_early); | |
96a2b03f | 700 | DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled); |
505f6d22 | 701 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
96a2b03f VB |
702 | |
703 | DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled); | |
e30825f1 | 704 | |
031bc574 JK |
705 | static int __init early_debug_pagealloc(char *buf) |
706 | { | |
8e57f8ac | 707 | return kstrtobool(buf, &_debug_pagealloc_enabled_early); |
031bc574 JK |
708 | } |
709 | early_param("debug_pagealloc", early_debug_pagealloc); | |
710 | ||
8e57f8ac | 711 | void init_debug_pagealloc(void) |
e30825f1 | 712 | { |
031bc574 JK |
713 | if (!debug_pagealloc_enabled()) |
714 | return; | |
715 | ||
8e57f8ac VB |
716 | static_branch_enable(&_debug_pagealloc_enabled); |
717 | ||
f1c1e9f7 JK |
718 | if (!debug_guardpage_minorder()) |
719 | return; | |
720 | ||
96a2b03f | 721 | static_branch_enable(&_debug_guardpage_enabled); |
e30825f1 JK |
722 | } |
723 | ||
c0a32fc5 SG |
724 | static int __init debug_guardpage_minorder_setup(char *buf) |
725 | { | |
726 | unsigned long res; | |
727 | ||
728 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 729 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
730 | return 0; |
731 | } | |
732 | _debug_guardpage_minorder = res; | |
1170532b | 733 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
734 | return 0; |
735 | } | |
f1c1e9f7 | 736 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
c0a32fc5 | 737 | |
acbc15a4 | 738 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
2847cf95 | 739 | unsigned int order, int migratetype) |
c0a32fc5 | 740 | { |
e30825f1 | 741 | if (!debug_guardpage_enabled()) |
acbc15a4 JK |
742 | return false; |
743 | ||
744 | if (order >= debug_guardpage_minorder()) | |
745 | return false; | |
e30825f1 | 746 | |
3972f6bb | 747 | __SetPageGuard(page); |
2847cf95 JK |
748 | INIT_LIST_HEAD(&page->lru); |
749 | set_page_private(page, order); | |
750 | /* Guard pages are not available for any usage */ | |
751 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
acbc15a4 JK |
752 | |
753 | return true; | |
c0a32fc5 SG |
754 | } |
755 | ||
2847cf95 JK |
756 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
757 | unsigned int order, int migratetype) | |
c0a32fc5 | 758 | { |
e30825f1 JK |
759 | if (!debug_guardpage_enabled()) |
760 | return; | |
761 | ||
3972f6bb | 762 | __ClearPageGuard(page); |
e30825f1 | 763 | |
2847cf95 JK |
764 | set_page_private(page, 0); |
765 | if (!is_migrate_isolate(migratetype)) | |
766 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
767 | } |
768 | #else | |
acbc15a4 JK |
769 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
770 | unsigned int order, int migratetype) { return false; } | |
2847cf95 JK |
771 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
772 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
773 | #endif |
774 | ||
7aeb09f9 | 775 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 776 | { |
4c21e2f2 | 777 | set_page_private(page, order); |
676165a8 | 778 | __SetPageBuddy(page); |
1da177e4 LT |
779 | } |
780 | ||
1da177e4 LT |
781 | /* |
782 | * This function checks whether a page is free && is the buddy | |
6e292b9b | 783 | * we can coalesce a page and its buddy if |
13ad59df | 784 | * (a) the buddy is not in a hole (check before calling!) && |
676165a8 | 785 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
786 | * (c) a page and its buddy have the same order && |
787 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 788 | * |
6e292b9b MW |
789 | * For recording whether a page is in the buddy system, we set PageBuddy. |
790 | * Setting, clearing, and testing PageBuddy is serialized by zone->lock. | |
1da177e4 | 791 | * |
676165a8 | 792 | * For recording page's order, we use page_private(page). |
1da177e4 | 793 | */ |
fe925c0c | 794 | static inline bool page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 795 | unsigned int order) |
1da177e4 | 796 | { |
fe925c0c | 797 | if (!page_is_guard(buddy) && !PageBuddy(buddy)) |
798 | return false; | |
4c5018ce | 799 | |
fe925c0c | 800 | if (page_order(buddy) != order) |
801 | return false; | |
c0a32fc5 | 802 | |
fe925c0c | 803 | /* |
804 | * zone check is done late to avoid uselessly calculating | |
805 | * zone/node ids for pages that could never merge. | |
806 | */ | |
807 | if (page_zone_id(page) != page_zone_id(buddy)) | |
808 | return false; | |
d34c5fa0 | 809 | |
fe925c0c | 810 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
4c5018ce | 811 | |
fe925c0c | 812 | return true; |
1da177e4 LT |
813 | } |
814 | ||
5e1f0f09 MG |
815 | #ifdef CONFIG_COMPACTION |
816 | static inline struct capture_control *task_capc(struct zone *zone) | |
817 | { | |
818 | struct capture_control *capc = current->capture_control; | |
819 | ||
820 | return capc && | |
821 | !(current->flags & PF_KTHREAD) && | |
822 | !capc->page && | |
823 | capc->cc->zone == zone && | |
824 | capc->cc->direct_compaction ? capc : NULL; | |
825 | } | |
826 | ||
827 | static inline bool | |
828 | compaction_capture(struct capture_control *capc, struct page *page, | |
829 | int order, int migratetype) | |
830 | { | |
831 | if (!capc || order != capc->cc->order) | |
832 | return false; | |
833 | ||
834 | /* Do not accidentally pollute CMA or isolated regions*/ | |
835 | if (is_migrate_cma(migratetype) || | |
836 | is_migrate_isolate(migratetype)) | |
837 | return false; | |
838 | ||
839 | /* | |
840 | * Do not let lower order allocations polluate a movable pageblock. | |
841 | * This might let an unmovable request use a reclaimable pageblock | |
842 | * and vice-versa but no more than normal fallback logic which can | |
843 | * have trouble finding a high-order free page. | |
844 | */ | |
845 | if (order < pageblock_order && migratetype == MIGRATE_MOVABLE) | |
846 | return false; | |
847 | ||
848 | capc->page = page; | |
849 | return true; | |
850 | } | |
851 | ||
852 | #else | |
853 | static inline struct capture_control *task_capc(struct zone *zone) | |
854 | { | |
855 | return NULL; | |
856 | } | |
857 | ||
858 | static inline bool | |
859 | compaction_capture(struct capture_control *capc, struct page *page, | |
860 | int order, int migratetype) | |
861 | { | |
862 | return false; | |
863 | } | |
864 | #endif /* CONFIG_COMPACTION */ | |
865 | ||
6ab01363 AD |
866 | /* Used for pages not on another list */ |
867 | static inline void add_to_free_list(struct page *page, struct zone *zone, | |
868 | unsigned int order, int migratetype) | |
869 | { | |
870 | struct free_area *area = &zone->free_area[order]; | |
871 | ||
872 | list_add(&page->lru, &area->free_list[migratetype]); | |
873 | area->nr_free++; | |
874 | } | |
875 | ||
876 | /* Used for pages not on another list */ | |
877 | static inline void add_to_free_list_tail(struct page *page, struct zone *zone, | |
878 | unsigned int order, int migratetype) | |
879 | { | |
880 | struct free_area *area = &zone->free_area[order]; | |
881 | ||
882 | list_add_tail(&page->lru, &area->free_list[migratetype]); | |
883 | area->nr_free++; | |
884 | } | |
885 | ||
886 | /* Used for pages which are on another list */ | |
887 | static inline void move_to_free_list(struct page *page, struct zone *zone, | |
888 | unsigned int order, int migratetype) | |
889 | { | |
890 | struct free_area *area = &zone->free_area[order]; | |
891 | ||
892 | list_move(&page->lru, &area->free_list[migratetype]); | |
893 | } | |
894 | ||
895 | static inline void del_page_from_free_list(struct page *page, struct zone *zone, | |
896 | unsigned int order) | |
897 | { | |
36e66c55 AD |
898 | /* clear reported state and update reported page count */ |
899 | if (page_reported(page)) | |
900 | __ClearPageReported(page); | |
901 | ||
6ab01363 AD |
902 | list_del(&page->lru); |
903 | __ClearPageBuddy(page); | |
904 | set_page_private(page, 0); | |
905 | zone->free_area[order].nr_free--; | |
906 | } | |
907 | ||
a2129f24 AD |
908 | /* |
909 | * If this is not the largest possible page, check if the buddy | |
910 | * of the next-highest order is free. If it is, it's possible | |
911 | * that pages are being freed that will coalesce soon. In case, | |
912 | * that is happening, add the free page to the tail of the list | |
913 | * so it's less likely to be used soon and more likely to be merged | |
914 | * as a higher order page | |
915 | */ | |
916 | static inline bool | |
917 | buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn, | |
918 | struct page *page, unsigned int order) | |
919 | { | |
920 | struct page *higher_page, *higher_buddy; | |
921 | unsigned long combined_pfn; | |
922 | ||
923 | if (order >= MAX_ORDER - 2) | |
924 | return false; | |
925 | ||
926 | if (!pfn_valid_within(buddy_pfn)) | |
927 | return false; | |
928 | ||
929 | combined_pfn = buddy_pfn & pfn; | |
930 | higher_page = page + (combined_pfn - pfn); | |
931 | buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1); | |
932 | higher_buddy = higher_page + (buddy_pfn - combined_pfn); | |
933 | ||
934 | return pfn_valid_within(buddy_pfn) && | |
935 | page_is_buddy(higher_page, higher_buddy, order + 1); | |
936 | } | |
937 | ||
1da177e4 LT |
938 | /* |
939 | * Freeing function for a buddy system allocator. | |
940 | * | |
941 | * The concept of a buddy system is to maintain direct-mapped table | |
942 | * (containing bit values) for memory blocks of various "orders". | |
943 | * The bottom level table contains the map for the smallest allocatable | |
944 | * units of memory (here, pages), and each level above it describes | |
945 | * pairs of units from the levels below, hence, "buddies". | |
946 | * At a high level, all that happens here is marking the table entry | |
947 | * at the bottom level available, and propagating the changes upward | |
948 | * as necessary, plus some accounting needed to play nicely with other | |
949 | * parts of the VM system. | |
950 | * At each level, we keep a list of pages, which are heads of continuous | |
6e292b9b MW |
951 | * free pages of length of (1 << order) and marked with PageBuddy. |
952 | * Page's order is recorded in page_private(page) field. | |
1da177e4 | 953 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
954 | * other. That is, if we allocate a small block, and both were |
955 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 956 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 957 | * triggers coalescing into a block of larger size. |
1da177e4 | 958 | * |
6d49e352 | 959 | * -- nyc |
1da177e4 LT |
960 | */ |
961 | ||
48db57f8 | 962 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 963 | unsigned long pfn, |
ed0ae21d | 964 | struct zone *zone, unsigned int order, |
36e66c55 | 965 | int migratetype, bool report) |
1da177e4 | 966 | { |
a2129f24 | 967 | struct capture_control *capc = task_capc(zone); |
76741e77 | 968 | unsigned long uninitialized_var(buddy_pfn); |
a2129f24 | 969 | unsigned long combined_pfn; |
d9dddbf5 | 970 | unsigned int max_order; |
a2129f24 AD |
971 | struct page *buddy; |
972 | bool to_tail; | |
d9dddbf5 VB |
973 | |
974 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 975 | |
d29bb978 | 976 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 977 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 978 | |
ed0ae21d | 979 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 980 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 981 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 982 | |
76741e77 | 983 | VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); |
309381fe | 984 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 985 | |
d9dddbf5 | 986 | continue_merging: |
3c605096 | 987 | while (order < max_order - 1) { |
5e1f0f09 MG |
988 | if (compaction_capture(capc, page, order, migratetype)) { |
989 | __mod_zone_freepage_state(zone, -(1 << order), | |
990 | migratetype); | |
991 | return; | |
992 | } | |
76741e77 VB |
993 | buddy_pfn = __find_buddy_pfn(pfn, order); |
994 | buddy = page + (buddy_pfn - pfn); | |
13ad59df VB |
995 | |
996 | if (!pfn_valid_within(buddy_pfn)) | |
997 | goto done_merging; | |
cb2b95e1 | 998 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 999 | goto done_merging; |
c0a32fc5 SG |
1000 | /* |
1001 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
1002 | * merge with it and move up one order. | |
1003 | */ | |
b03641af | 1004 | if (page_is_guard(buddy)) |
2847cf95 | 1005 | clear_page_guard(zone, buddy, order, migratetype); |
b03641af | 1006 | else |
6ab01363 | 1007 | del_page_from_free_list(buddy, zone, order); |
76741e77 VB |
1008 | combined_pfn = buddy_pfn & pfn; |
1009 | page = page + (combined_pfn - pfn); | |
1010 | pfn = combined_pfn; | |
1da177e4 LT |
1011 | order++; |
1012 | } | |
d9dddbf5 VB |
1013 | if (max_order < MAX_ORDER) { |
1014 | /* If we are here, it means order is >= pageblock_order. | |
1015 | * We want to prevent merge between freepages on isolate | |
1016 | * pageblock and normal pageblock. Without this, pageblock | |
1017 | * isolation could cause incorrect freepage or CMA accounting. | |
1018 | * | |
1019 | * We don't want to hit this code for the more frequent | |
1020 | * low-order merging. | |
1021 | */ | |
1022 | if (unlikely(has_isolate_pageblock(zone))) { | |
1023 | int buddy_mt; | |
1024 | ||
76741e77 VB |
1025 | buddy_pfn = __find_buddy_pfn(pfn, order); |
1026 | buddy = page + (buddy_pfn - pfn); | |
d9dddbf5 VB |
1027 | buddy_mt = get_pageblock_migratetype(buddy); |
1028 | ||
1029 | if (migratetype != buddy_mt | |
1030 | && (is_migrate_isolate(migratetype) || | |
1031 | is_migrate_isolate(buddy_mt))) | |
1032 | goto done_merging; | |
1033 | } | |
1034 | max_order++; | |
1035 | goto continue_merging; | |
1036 | } | |
1037 | ||
1038 | done_merging: | |
1da177e4 | 1039 | set_page_order(page, order); |
6dda9d55 | 1040 | |
97500a4a | 1041 | if (is_shuffle_order(order)) |
a2129f24 | 1042 | to_tail = shuffle_pick_tail(); |
97500a4a | 1043 | else |
a2129f24 | 1044 | to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order); |
97500a4a | 1045 | |
a2129f24 | 1046 | if (to_tail) |
6ab01363 | 1047 | add_to_free_list_tail(page, zone, order, migratetype); |
a2129f24 | 1048 | else |
6ab01363 | 1049 | add_to_free_list(page, zone, order, migratetype); |
36e66c55 AD |
1050 | |
1051 | /* Notify page reporting subsystem of freed page */ | |
1052 | if (report) | |
1053 | page_reporting_notify_free(order); | |
1da177e4 LT |
1054 | } |
1055 | ||
7bfec6f4 MG |
1056 | /* |
1057 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
1058 | * try and check multiple fields with one check. The caller must do a detailed | |
1059 | * check if necessary. | |
1060 | */ | |
1061 | static inline bool page_expected_state(struct page *page, | |
1062 | unsigned long check_flags) | |
1063 | { | |
1064 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
1065 | return false; | |
1066 | ||
1067 | if (unlikely((unsigned long)page->mapping | | |
1068 | page_ref_count(page) | | |
1069 | #ifdef CONFIG_MEMCG | |
1070 | (unsigned long)page->mem_cgroup | | |
1071 | #endif | |
1072 | (page->flags & check_flags))) | |
1073 | return false; | |
1074 | ||
1075 | return true; | |
1076 | } | |
1077 | ||
bb552ac6 | 1078 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 1079 | { |
7bfec6f4 MG |
1080 | const char *bad_reason; |
1081 | unsigned long bad_flags; | |
1082 | ||
7bfec6f4 MG |
1083 | bad_reason = NULL; |
1084 | bad_flags = 0; | |
f0b791a3 | 1085 | |
53f9263b | 1086 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1087 | bad_reason = "nonzero mapcount"; |
1088 | if (unlikely(page->mapping != NULL)) | |
1089 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1090 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 1091 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
1092 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
1093 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
1094 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
1095 | } | |
9edad6ea JW |
1096 | #ifdef CONFIG_MEMCG |
1097 | if (unlikely(page->mem_cgroup)) | |
1098 | bad_reason = "page still charged to cgroup"; | |
1099 | #endif | |
7bfec6f4 | 1100 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
1101 | } |
1102 | ||
1103 | static inline int free_pages_check(struct page *page) | |
1104 | { | |
da838d4f | 1105 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
bb552ac6 | 1106 | return 0; |
bb552ac6 MG |
1107 | |
1108 | /* Something has gone sideways, find it */ | |
1109 | free_pages_check_bad(page); | |
7bfec6f4 | 1110 | return 1; |
1da177e4 LT |
1111 | } |
1112 | ||
4db7548c MG |
1113 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1114 | { | |
1115 | int ret = 1; | |
1116 | ||
1117 | /* | |
1118 | * We rely page->lru.next never has bit 0 set, unless the page | |
1119 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
1120 | */ | |
1121 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
1122 | ||
1123 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
1124 | ret = 0; | |
1125 | goto out; | |
1126 | } | |
1127 | switch (page - head_page) { | |
1128 | case 1: | |
4da1984e | 1129 | /* the first tail page: ->mapping may be compound_mapcount() */ |
4db7548c MG |
1130 | if (unlikely(compound_mapcount(page))) { |
1131 | bad_page(page, "nonzero compound_mapcount", 0); | |
1132 | goto out; | |
1133 | } | |
1134 | break; | |
1135 | case 2: | |
1136 | /* | |
1137 | * the second tail page: ->mapping is | |
fa3015b7 | 1138 | * deferred_list.next -- ignore value. |
4db7548c MG |
1139 | */ |
1140 | break; | |
1141 | default: | |
1142 | if (page->mapping != TAIL_MAPPING) { | |
1143 | bad_page(page, "corrupted mapping in tail page", 0); | |
1144 | goto out; | |
1145 | } | |
1146 | break; | |
1147 | } | |
1148 | if (unlikely(!PageTail(page))) { | |
1149 | bad_page(page, "PageTail not set", 0); | |
1150 | goto out; | |
1151 | } | |
1152 | if (unlikely(compound_head(page) != head_page)) { | |
1153 | bad_page(page, "compound_head not consistent", 0); | |
1154 | goto out; | |
1155 | } | |
1156 | ret = 0; | |
1157 | out: | |
1158 | page->mapping = NULL; | |
1159 | clear_compound_head(page); | |
1160 | return ret; | |
1161 | } | |
1162 | ||
6471384a AP |
1163 | static void kernel_init_free_pages(struct page *page, int numpages) |
1164 | { | |
1165 | int i; | |
1166 | ||
1167 | for (i = 0; i < numpages; i++) | |
1168 | clear_highpage(page + i); | |
1169 | } | |
1170 | ||
e2769dbd MG |
1171 | static __always_inline bool free_pages_prepare(struct page *page, |
1172 | unsigned int order, bool check_free) | |
4db7548c | 1173 | { |
e2769dbd | 1174 | int bad = 0; |
4db7548c | 1175 | |
4db7548c MG |
1176 | VM_BUG_ON_PAGE(PageTail(page), page); |
1177 | ||
e2769dbd | 1178 | trace_mm_page_free(page, order); |
e2769dbd MG |
1179 | |
1180 | /* | |
1181 | * Check tail pages before head page information is cleared to | |
1182 | * avoid checking PageCompound for order-0 pages. | |
1183 | */ | |
1184 | if (unlikely(order)) { | |
1185 | bool compound = PageCompound(page); | |
1186 | int i; | |
1187 | ||
1188 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
4db7548c | 1189 | |
9a73f61b KS |
1190 | if (compound) |
1191 | ClearPageDoubleMap(page); | |
e2769dbd MG |
1192 | for (i = 1; i < (1 << order); i++) { |
1193 | if (compound) | |
1194 | bad += free_tail_pages_check(page, page + i); | |
1195 | if (unlikely(free_pages_check(page + i))) { | |
1196 | bad++; | |
1197 | continue; | |
1198 | } | |
1199 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1200 | } | |
1201 | } | |
bda807d4 | 1202 | if (PageMappingFlags(page)) |
4db7548c | 1203 | page->mapping = NULL; |
c4159a75 | 1204 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
f4b00eab | 1205 | __memcg_kmem_uncharge_page(page, order); |
e2769dbd MG |
1206 | if (check_free) |
1207 | bad += free_pages_check(page); | |
1208 | if (bad) | |
1209 | return false; | |
4db7548c | 1210 | |
e2769dbd MG |
1211 | page_cpupid_reset_last(page); |
1212 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
1213 | reset_page_owner(page, order); | |
4db7548c MG |
1214 | |
1215 | if (!PageHighMem(page)) { | |
1216 | debug_check_no_locks_freed(page_address(page), | |
e2769dbd | 1217 | PAGE_SIZE << order); |
4db7548c | 1218 | debug_check_no_obj_freed(page_address(page), |
e2769dbd | 1219 | PAGE_SIZE << order); |
4db7548c | 1220 | } |
6471384a AP |
1221 | if (want_init_on_free()) |
1222 | kernel_init_free_pages(page, 1 << order); | |
1223 | ||
e2769dbd | 1224 | kernel_poison_pages(page, 1 << order, 0); |
234fdce8 QC |
1225 | /* |
1226 | * arch_free_page() can make the page's contents inaccessible. s390 | |
1227 | * does this. So nothing which can access the page's contents should | |
1228 | * happen after this. | |
1229 | */ | |
1230 | arch_free_page(page, order); | |
1231 | ||
8e57f8ac | 1232 | if (debug_pagealloc_enabled_static()) |
d6332692 RE |
1233 | kernel_map_pages(page, 1 << order, 0); |
1234 | ||
3c0c12cc | 1235 | kasan_free_nondeferred_pages(page, order); |
4db7548c | 1236 | |
4db7548c MG |
1237 | return true; |
1238 | } | |
1239 | ||
e2769dbd | 1240 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
1241 | /* |
1242 | * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed | |
1243 | * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when | |
1244 | * moved from pcp lists to free lists. | |
1245 | */ | |
1246 | static bool free_pcp_prepare(struct page *page) | |
e2769dbd MG |
1247 | { |
1248 | return free_pages_prepare(page, 0, true); | |
1249 | } | |
1250 | ||
4462b32c | 1251 | static bool bulkfree_pcp_prepare(struct page *page) |
e2769dbd | 1252 | { |
8e57f8ac | 1253 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
1254 | return free_pages_check(page); |
1255 | else | |
1256 | return false; | |
e2769dbd MG |
1257 | } |
1258 | #else | |
4462b32c VB |
1259 | /* |
1260 | * With DEBUG_VM disabled, order-0 pages being freed are checked only when | |
1261 | * moving from pcp lists to free list in order to reduce overhead. With | |
1262 | * debug_pagealloc enabled, they are checked also immediately when being freed | |
1263 | * to the pcp lists. | |
1264 | */ | |
e2769dbd MG |
1265 | static bool free_pcp_prepare(struct page *page) |
1266 | { | |
8e57f8ac | 1267 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
1268 | return free_pages_prepare(page, 0, true); |
1269 | else | |
1270 | return free_pages_prepare(page, 0, false); | |
e2769dbd MG |
1271 | } |
1272 | ||
4db7548c MG |
1273 | static bool bulkfree_pcp_prepare(struct page *page) |
1274 | { | |
1275 | return free_pages_check(page); | |
1276 | } | |
1277 | #endif /* CONFIG_DEBUG_VM */ | |
1278 | ||
97334162 AL |
1279 | static inline void prefetch_buddy(struct page *page) |
1280 | { | |
1281 | unsigned long pfn = page_to_pfn(page); | |
1282 | unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0); | |
1283 | struct page *buddy = page + (buddy_pfn - pfn); | |
1284 | ||
1285 | prefetch(buddy); | |
1286 | } | |
1287 | ||
1da177e4 | 1288 | /* |
5f8dcc21 | 1289 | * Frees a number of pages from the PCP lists |
1da177e4 | 1290 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 1291 | * count is the number of pages to free. |
1da177e4 LT |
1292 | * |
1293 | * If the zone was previously in an "all pages pinned" state then look to | |
1294 | * see if this freeing clears that state. | |
1295 | * | |
1296 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
1297 | * pinned" detection logic. | |
1298 | */ | |
5f8dcc21 MG |
1299 | static void free_pcppages_bulk(struct zone *zone, int count, |
1300 | struct per_cpu_pages *pcp) | |
1da177e4 | 1301 | { |
5f8dcc21 | 1302 | int migratetype = 0; |
a6f9edd6 | 1303 | int batch_free = 0; |
97334162 | 1304 | int prefetch_nr = 0; |
3777999d | 1305 | bool isolated_pageblocks; |
0a5f4e5b AL |
1306 | struct page *page, *tmp; |
1307 | LIST_HEAD(head); | |
f2260e6b | 1308 | |
e5b31ac2 | 1309 | while (count) { |
5f8dcc21 MG |
1310 | struct list_head *list; |
1311 | ||
1312 | /* | |
a6f9edd6 MG |
1313 | * Remove pages from lists in a round-robin fashion. A |
1314 | * batch_free count is maintained that is incremented when an | |
1315 | * empty list is encountered. This is so more pages are freed | |
1316 | * off fuller lists instead of spinning excessively around empty | |
1317 | * lists | |
5f8dcc21 MG |
1318 | */ |
1319 | do { | |
a6f9edd6 | 1320 | batch_free++; |
5f8dcc21 MG |
1321 | if (++migratetype == MIGRATE_PCPTYPES) |
1322 | migratetype = 0; | |
1323 | list = &pcp->lists[migratetype]; | |
1324 | } while (list_empty(list)); | |
48db57f8 | 1325 | |
1d16871d NK |
1326 | /* This is the only non-empty list. Free them all. */ |
1327 | if (batch_free == MIGRATE_PCPTYPES) | |
e5b31ac2 | 1328 | batch_free = count; |
1d16871d | 1329 | |
a6f9edd6 | 1330 | do { |
a16601c5 | 1331 | page = list_last_entry(list, struct page, lru); |
0a5f4e5b | 1332 | /* must delete to avoid corrupting pcp list */ |
a6f9edd6 | 1333 | list_del(&page->lru); |
77ba9062 | 1334 | pcp->count--; |
aa016d14 | 1335 | |
4db7548c MG |
1336 | if (bulkfree_pcp_prepare(page)) |
1337 | continue; | |
1338 | ||
0a5f4e5b | 1339 | list_add_tail(&page->lru, &head); |
97334162 AL |
1340 | |
1341 | /* | |
1342 | * We are going to put the page back to the global | |
1343 | * pool, prefetch its buddy to speed up later access | |
1344 | * under zone->lock. It is believed the overhead of | |
1345 | * an additional test and calculating buddy_pfn here | |
1346 | * can be offset by reduced memory latency later. To | |
1347 | * avoid excessive prefetching due to large count, only | |
1348 | * prefetch buddy for the first pcp->batch nr of pages. | |
1349 | */ | |
1350 | if (prefetch_nr++ < pcp->batch) | |
1351 | prefetch_buddy(page); | |
e5b31ac2 | 1352 | } while (--count && --batch_free && !list_empty(list)); |
1da177e4 | 1353 | } |
0a5f4e5b AL |
1354 | |
1355 | spin_lock(&zone->lock); | |
1356 | isolated_pageblocks = has_isolate_pageblock(zone); | |
1357 | ||
1358 | /* | |
1359 | * Use safe version since after __free_one_page(), | |
1360 | * page->lru.next will not point to original list. | |
1361 | */ | |
1362 | list_for_each_entry_safe(page, tmp, &head, lru) { | |
1363 | int mt = get_pcppage_migratetype(page); | |
1364 | /* MIGRATE_ISOLATE page should not go to pcplists */ | |
1365 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
1366 | /* Pageblock could have been isolated meanwhile */ | |
1367 | if (unlikely(isolated_pageblocks)) | |
1368 | mt = get_pageblock_migratetype(page); | |
1369 | ||
36e66c55 | 1370 | __free_one_page(page, page_to_pfn(page), zone, 0, mt, true); |
0a5f4e5b AL |
1371 | trace_mm_page_pcpu_drain(page, 0, mt); |
1372 | } | |
d34b0733 | 1373 | spin_unlock(&zone->lock); |
1da177e4 LT |
1374 | } |
1375 | ||
dc4b0caf MG |
1376 | static void free_one_page(struct zone *zone, |
1377 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 1378 | unsigned int order, |
ed0ae21d | 1379 | int migratetype) |
1da177e4 | 1380 | { |
d34b0733 | 1381 | spin_lock(&zone->lock); |
ad53f92e JK |
1382 | if (unlikely(has_isolate_pageblock(zone) || |
1383 | is_migrate_isolate(migratetype))) { | |
1384 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 1385 | } |
36e66c55 | 1386 | __free_one_page(page, pfn, zone, order, migratetype, true); |
d34b0733 | 1387 | spin_unlock(&zone->lock); |
48db57f8 NP |
1388 | } |
1389 | ||
1e8ce83c | 1390 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
d0dc12e8 | 1391 | unsigned long zone, int nid) |
1e8ce83c | 1392 | { |
d0dc12e8 | 1393 | mm_zero_struct_page(page); |
1e8ce83c | 1394 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
1395 | init_page_count(page); |
1396 | page_mapcount_reset(page); | |
1397 | page_cpupid_reset_last(page); | |
2813b9c0 | 1398 | page_kasan_tag_reset(page); |
1e8ce83c | 1399 | |
1e8ce83c RH |
1400 | INIT_LIST_HEAD(&page->lru); |
1401 | #ifdef WANT_PAGE_VIRTUAL | |
1402 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1403 | if (!is_highmem_idx(zone)) | |
1404 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1405 | #endif | |
1406 | } | |
1407 | ||
7e18adb4 | 1408 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
57148a64 | 1409 | static void __meminit init_reserved_page(unsigned long pfn) |
7e18adb4 MG |
1410 | { |
1411 | pg_data_t *pgdat; | |
1412 | int nid, zid; | |
1413 | ||
1414 | if (!early_page_uninitialised(pfn)) | |
1415 | return; | |
1416 | ||
1417 | nid = early_pfn_to_nid(pfn); | |
1418 | pgdat = NODE_DATA(nid); | |
1419 | ||
1420 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1421 | struct zone *zone = &pgdat->node_zones[zid]; | |
1422 | ||
1423 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1424 | break; | |
1425 | } | |
d0dc12e8 | 1426 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid); |
7e18adb4 MG |
1427 | } |
1428 | #else | |
1429 | static inline void init_reserved_page(unsigned long pfn) | |
1430 | { | |
1431 | } | |
1432 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1433 | ||
92923ca3 NZ |
1434 | /* |
1435 | * Initialised pages do not have PageReserved set. This function is | |
1436 | * called for each range allocated by the bootmem allocator and | |
1437 | * marks the pages PageReserved. The remaining valid pages are later | |
1438 | * sent to the buddy page allocator. | |
1439 | */ | |
4b50bcc7 | 1440 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
92923ca3 NZ |
1441 | { |
1442 | unsigned long start_pfn = PFN_DOWN(start); | |
1443 | unsigned long end_pfn = PFN_UP(end); | |
1444 | ||
7e18adb4 MG |
1445 | for (; start_pfn < end_pfn; start_pfn++) { |
1446 | if (pfn_valid(start_pfn)) { | |
1447 | struct page *page = pfn_to_page(start_pfn); | |
1448 | ||
1449 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1450 | |
1451 | /* Avoid false-positive PageTail() */ | |
1452 | INIT_LIST_HEAD(&page->lru); | |
1453 | ||
d483da5b AD |
1454 | /* |
1455 | * no need for atomic set_bit because the struct | |
1456 | * page is not visible yet so nobody should | |
1457 | * access it yet. | |
1458 | */ | |
1459 | __SetPageReserved(page); | |
7e18adb4 MG |
1460 | } |
1461 | } | |
92923ca3 NZ |
1462 | } |
1463 | ||
ec95f53a KM |
1464 | static void __free_pages_ok(struct page *page, unsigned int order) |
1465 | { | |
d34b0733 | 1466 | unsigned long flags; |
95e34412 | 1467 | int migratetype; |
dc4b0caf | 1468 | unsigned long pfn = page_to_pfn(page); |
ec95f53a | 1469 | |
e2769dbd | 1470 | if (!free_pages_prepare(page, order, true)) |
ec95f53a KM |
1471 | return; |
1472 | ||
cfc47a28 | 1473 | migratetype = get_pfnblock_migratetype(page, pfn); |
d34b0733 MG |
1474 | local_irq_save(flags); |
1475 | __count_vm_events(PGFREE, 1 << order); | |
dc4b0caf | 1476 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
d34b0733 | 1477 | local_irq_restore(flags); |
1da177e4 LT |
1478 | } |
1479 | ||
a9cd410a | 1480 | void __free_pages_core(struct page *page, unsigned int order) |
a226f6c8 | 1481 | { |
c3993076 | 1482 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1483 | struct page *p = page; |
c3993076 | 1484 | unsigned int loop; |
a226f6c8 | 1485 | |
e2d0bd2b YL |
1486 | prefetchw(p); |
1487 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1488 | prefetchw(p + 1); | |
c3993076 JW |
1489 | __ClearPageReserved(p); |
1490 | set_page_count(p, 0); | |
a226f6c8 | 1491 | } |
e2d0bd2b YL |
1492 | __ClearPageReserved(p); |
1493 | set_page_count(p, 0); | |
c3993076 | 1494 | |
9705bea5 | 1495 | atomic_long_add(nr_pages, &page_zone(page)->managed_pages); |
c3993076 JW |
1496 | set_page_refcounted(page); |
1497 | __free_pages(page, order); | |
a226f6c8 DH |
1498 | } |
1499 | ||
3f08a302 | 1500 | #ifdef CONFIG_NEED_MULTIPLE_NODES |
7ace9917 | 1501 | |
75a592a4 MG |
1502 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1503 | ||
6f24fbd3 MR |
1504 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
1505 | ||
1506 | /* | |
1507 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
1508 | */ | |
1509 | int __meminit __early_pfn_to_nid(unsigned long pfn, | |
1510 | struct mminit_pfnnid_cache *state) | |
1511 | { | |
1512 | unsigned long start_pfn, end_pfn; | |
1513 | int nid; | |
1514 | ||
1515 | if (state->last_start <= pfn && pfn < state->last_end) | |
1516 | return state->last_nid; | |
1517 | ||
1518 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); | |
1519 | if (nid != NUMA_NO_NODE) { | |
1520 | state->last_start = start_pfn; | |
1521 | state->last_end = end_pfn; | |
1522 | state->last_nid = nid; | |
1523 | } | |
1524 | ||
1525 | return nid; | |
1526 | } | |
1527 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
1528 | ||
75a592a4 MG |
1529 | int __meminit early_pfn_to_nid(unsigned long pfn) |
1530 | { | |
7ace9917 | 1531 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1532 | int nid; |
1533 | ||
7ace9917 | 1534 | spin_lock(&early_pfn_lock); |
75a592a4 | 1535 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 | 1536 | if (nid < 0) |
e4568d38 | 1537 | nid = first_online_node; |
7ace9917 MG |
1538 | spin_unlock(&early_pfn_lock); |
1539 | ||
1540 | return nid; | |
75a592a4 | 1541 | } |
3f08a302 | 1542 | #endif /* CONFIG_NEED_MULTIPLE_NODES */ |
75a592a4 | 1543 | |
7c2ee349 | 1544 | void __init memblock_free_pages(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1545 | unsigned int order) |
1546 | { | |
1547 | if (early_page_uninitialised(pfn)) | |
1548 | return; | |
a9cd410a | 1549 | __free_pages_core(page, order); |
3a80a7fa MG |
1550 | } |
1551 | ||
7cf91a98 JK |
1552 | /* |
1553 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1554 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1555 | * with the migration of free compaction scanner. The scanners then need to | |
1556 | * use only pfn_valid_within() check for arches that allow holes within | |
1557 | * pageblocks. | |
1558 | * | |
1559 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1560 | * | |
1561 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1562 | * i.e. it's possible that all pages within a zones range of pages do not | |
1563 | * belong to a single zone. We assume that a border between node0 and node1 | |
1564 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1565 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1566 | * the first and last page of a pageblock and avoid checking each individual | |
1567 | * page in a pageblock. | |
1568 | */ | |
1569 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1570 | unsigned long end_pfn, struct zone *zone) | |
1571 | { | |
1572 | struct page *start_page; | |
1573 | struct page *end_page; | |
1574 | ||
1575 | /* end_pfn is one past the range we are checking */ | |
1576 | end_pfn--; | |
1577 | ||
1578 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1579 | return NULL; | |
1580 | ||
2d070eab MH |
1581 | start_page = pfn_to_online_page(start_pfn); |
1582 | if (!start_page) | |
1583 | return NULL; | |
7cf91a98 JK |
1584 | |
1585 | if (page_zone(start_page) != zone) | |
1586 | return NULL; | |
1587 | ||
1588 | end_page = pfn_to_page(end_pfn); | |
1589 | ||
1590 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1591 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1592 | return NULL; | |
1593 | ||
1594 | return start_page; | |
1595 | } | |
1596 | ||
1597 | void set_zone_contiguous(struct zone *zone) | |
1598 | { | |
1599 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1600 | unsigned long block_end_pfn; | |
1601 | ||
1602 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1603 | for (; block_start_pfn < zone_end_pfn(zone); | |
1604 | block_start_pfn = block_end_pfn, | |
1605 | block_end_pfn += pageblock_nr_pages) { | |
1606 | ||
1607 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1608 | ||
1609 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1610 | block_end_pfn, zone)) | |
1611 | return; | |
e84fe99b | 1612 | cond_resched(); |
7cf91a98 JK |
1613 | } |
1614 | ||
1615 | /* We confirm that there is no hole */ | |
1616 | zone->contiguous = true; | |
1617 | } | |
1618 | ||
1619 | void clear_zone_contiguous(struct zone *zone) | |
1620 | { | |
1621 | zone->contiguous = false; | |
1622 | } | |
1623 | ||
7e18adb4 | 1624 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
2f47a91f PT |
1625 | static void __init deferred_free_range(unsigned long pfn, |
1626 | unsigned long nr_pages) | |
a4de83dd | 1627 | { |
2f47a91f PT |
1628 | struct page *page; |
1629 | unsigned long i; | |
a4de83dd | 1630 | |
2f47a91f | 1631 | if (!nr_pages) |
a4de83dd MG |
1632 | return; |
1633 | ||
2f47a91f PT |
1634 | page = pfn_to_page(pfn); |
1635 | ||
a4de83dd | 1636 | /* Free a large naturally-aligned chunk if possible */ |
e780149b XQ |
1637 | if (nr_pages == pageblock_nr_pages && |
1638 | (pfn & (pageblock_nr_pages - 1)) == 0) { | |
ac5d2539 | 1639 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a9cd410a | 1640 | __free_pages_core(page, pageblock_order); |
a4de83dd MG |
1641 | return; |
1642 | } | |
1643 | ||
e780149b XQ |
1644 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1645 | if ((pfn & (pageblock_nr_pages - 1)) == 0) | |
1646 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
a9cd410a | 1647 | __free_pages_core(page, 0); |
e780149b | 1648 | } |
a4de83dd MG |
1649 | } |
1650 | ||
d3cd131d NS |
1651 | /* Completion tracking for deferred_init_memmap() threads */ |
1652 | static atomic_t pgdat_init_n_undone __initdata; | |
1653 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1654 | ||
1655 | static inline void __init pgdat_init_report_one_done(void) | |
1656 | { | |
1657 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1658 | complete(&pgdat_init_all_done_comp); | |
1659 | } | |
0e1cc95b | 1660 | |
2f47a91f | 1661 | /* |
80b1f41c PT |
1662 | * Returns true if page needs to be initialized or freed to buddy allocator. |
1663 | * | |
1664 | * First we check if pfn is valid on architectures where it is possible to have | |
1665 | * holes within pageblock_nr_pages. On systems where it is not possible, this | |
1666 | * function is optimized out. | |
1667 | * | |
1668 | * Then, we check if a current large page is valid by only checking the validity | |
1669 | * of the head pfn. | |
2f47a91f | 1670 | */ |
56ec43d8 | 1671 | static inline bool __init deferred_pfn_valid(unsigned long pfn) |
2f47a91f | 1672 | { |
80b1f41c PT |
1673 | if (!pfn_valid_within(pfn)) |
1674 | return false; | |
1675 | if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn)) | |
1676 | return false; | |
80b1f41c PT |
1677 | return true; |
1678 | } | |
2f47a91f | 1679 | |
80b1f41c PT |
1680 | /* |
1681 | * Free pages to buddy allocator. Try to free aligned pages in | |
1682 | * pageblock_nr_pages sizes. | |
1683 | */ | |
56ec43d8 | 1684 | static void __init deferred_free_pages(unsigned long pfn, |
80b1f41c PT |
1685 | unsigned long end_pfn) |
1686 | { | |
80b1f41c PT |
1687 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
1688 | unsigned long nr_free = 0; | |
2f47a91f | 1689 | |
80b1f41c | 1690 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1691 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c PT |
1692 | deferred_free_range(pfn - nr_free, nr_free); |
1693 | nr_free = 0; | |
1694 | } else if (!(pfn & nr_pgmask)) { | |
1695 | deferred_free_range(pfn - nr_free, nr_free); | |
1696 | nr_free = 1; | |
3a2d7fa8 | 1697 | touch_nmi_watchdog(); |
80b1f41c PT |
1698 | } else { |
1699 | nr_free++; | |
1700 | } | |
1701 | } | |
1702 | /* Free the last block of pages to allocator */ | |
1703 | deferred_free_range(pfn - nr_free, nr_free); | |
2f47a91f PT |
1704 | } |
1705 | ||
80b1f41c PT |
1706 | /* |
1707 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks | |
1708 | * by performing it only once every pageblock_nr_pages. | |
1709 | * Return number of pages initialized. | |
1710 | */ | |
56ec43d8 | 1711 | static unsigned long __init deferred_init_pages(struct zone *zone, |
80b1f41c PT |
1712 | unsigned long pfn, |
1713 | unsigned long end_pfn) | |
2f47a91f | 1714 | { |
2f47a91f | 1715 | unsigned long nr_pgmask = pageblock_nr_pages - 1; |
56ec43d8 | 1716 | int nid = zone_to_nid(zone); |
2f47a91f | 1717 | unsigned long nr_pages = 0; |
56ec43d8 | 1718 | int zid = zone_idx(zone); |
2f47a91f | 1719 | struct page *page = NULL; |
2f47a91f | 1720 | |
80b1f41c | 1721 | for (; pfn < end_pfn; pfn++) { |
56ec43d8 | 1722 | if (!deferred_pfn_valid(pfn)) { |
80b1f41c | 1723 | page = NULL; |
2f47a91f | 1724 | continue; |
80b1f41c | 1725 | } else if (!page || !(pfn & nr_pgmask)) { |
2f47a91f | 1726 | page = pfn_to_page(pfn); |
3a2d7fa8 | 1727 | touch_nmi_watchdog(); |
80b1f41c PT |
1728 | } else { |
1729 | page++; | |
2f47a91f | 1730 | } |
d0dc12e8 | 1731 | __init_single_page(page, pfn, zid, nid); |
80b1f41c | 1732 | nr_pages++; |
2f47a91f | 1733 | } |
80b1f41c | 1734 | return (nr_pages); |
2f47a91f PT |
1735 | } |
1736 | ||
0e56acae AD |
1737 | /* |
1738 | * This function is meant to pre-load the iterator for the zone init. | |
1739 | * Specifically it walks through the ranges until we are caught up to the | |
1740 | * first_init_pfn value and exits there. If we never encounter the value we | |
1741 | * return false indicating there are no valid ranges left. | |
1742 | */ | |
1743 | static bool __init | |
1744 | deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone, | |
1745 | unsigned long *spfn, unsigned long *epfn, | |
1746 | unsigned long first_init_pfn) | |
1747 | { | |
1748 | u64 j; | |
1749 | ||
1750 | /* | |
1751 | * Start out by walking through the ranges in this zone that have | |
1752 | * already been initialized. We don't need to do anything with them | |
1753 | * so we just need to flush them out of the system. | |
1754 | */ | |
1755 | for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) { | |
1756 | if (*epfn <= first_init_pfn) | |
1757 | continue; | |
1758 | if (*spfn < first_init_pfn) | |
1759 | *spfn = first_init_pfn; | |
1760 | *i = j; | |
1761 | return true; | |
1762 | } | |
1763 | ||
1764 | return false; | |
1765 | } | |
1766 | ||
1767 | /* | |
1768 | * Initialize and free pages. We do it in two loops: first we initialize | |
1769 | * struct page, then free to buddy allocator, because while we are | |
1770 | * freeing pages we can access pages that are ahead (computing buddy | |
1771 | * page in __free_one_page()). | |
1772 | * | |
1773 | * In order to try and keep some memory in the cache we have the loop | |
1774 | * broken along max page order boundaries. This way we will not cause | |
1775 | * any issues with the buddy page computation. | |
1776 | */ | |
1777 | static unsigned long __init | |
1778 | deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn, | |
1779 | unsigned long *end_pfn) | |
1780 | { | |
1781 | unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES); | |
1782 | unsigned long spfn = *start_pfn, epfn = *end_pfn; | |
1783 | unsigned long nr_pages = 0; | |
1784 | u64 j = *i; | |
1785 | ||
1786 | /* First we loop through and initialize the page values */ | |
1787 | for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) { | |
1788 | unsigned long t; | |
1789 | ||
1790 | if (mo_pfn <= *start_pfn) | |
1791 | break; | |
1792 | ||
1793 | t = min(mo_pfn, *end_pfn); | |
1794 | nr_pages += deferred_init_pages(zone, *start_pfn, t); | |
1795 | ||
1796 | if (mo_pfn < *end_pfn) { | |
1797 | *start_pfn = mo_pfn; | |
1798 | break; | |
1799 | } | |
1800 | } | |
1801 | ||
1802 | /* Reset values and now loop through freeing pages as needed */ | |
1803 | swap(j, *i); | |
1804 | ||
1805 | for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) { | |
1806 | unsigned long t; | |
1807 | ||
1808 | if (mo_pfn <= spfn) | |
1809 | break; | |
1810 | ||
1811 | t = min(mo_pfn, epfn); | |
1812 | deferred_free_pages(spfn, t); | |
1813 | ||
1814 | if (mo_pfn <= epfn) | |
1815 | break; | |
1816 | } | |
1817 | ||
1818 | return nr_pages; | |
1819 | } | |
1820 | ||
7e18adb4 | 1821 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1822 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1823 | { |
0e1cc95b | 1824 | pg_data_t *pgdat = data; |
0e56acae AD |
1825 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
1826 | unsigned long spfn = 0, epfn = 0, nr_pages = 0; | |
1827 | unsigned long first_init_pfn, flags; | |
7e18adb4 | 1828 | unsigned long start = jiffies; |
7e18adb4 | 1829 | struct zone *zone; |
0e56acae | 1830 | int zid; |
2f47a91f | 1831 | u64 i; |
7e18adb4 | 1832 | |
3a2d7fa8 PT |
1833 | /* Bind memory initialisation thread to a local node if possible */ |
1834 | if (!cpumask_empty(cpumask)) | |
1835 | set_cpus_allowed_ptr(current, cpumask); | |
1836 | ||
1837 | pgdat_resize_lock(pgdat, &flags); | |
1838 | first_init_pfn = pgdat->first_deferred_pfn; | |
0e1cc95b | 1839 | if (first_init_pfn == ULONG_MAX) { |
3a2d7fa8 | 1840 | pgdat_resize_unlock(pgdat, &flags); |
d3cd131d | 1841 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1842 | return 0; |
1843 | } | |
1844 | ||
7e18adb4 MG |
1845 | /* Sanity check boundaries */ |
1846 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1847 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1848 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1849 | ||
1850 | /* Only the highest zone is deferred so find it */ | |
1851 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1852 | zone = pgdat->node_zones + zid; | |
1853 | if (first_init_pfn < zone_end_pfn(zone)) | |
1854 | break; | |
1855 | } | |
0e56acae AD |
1856 | |
1857 | /* If the zone is empty somebody else may have cleared out the zone */ | |
1858 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1859 | first_init_pfn)) | |
1860 | goto zone_empty; | |
7e18adb4 | 1861 | |
80b1f41c | 1862 | /* |
0e56acae AD |
1863 | * Initialize and free pages in MAX_ORDER sized increments so |
1864 | * that we can avoid introducing any issues with the buddy | |
1865 | * allocator. | |
80b1f41c | 1866 | */ |
0e56acae AD |
1867 | while (spfn < epfn) |
1868 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
1869 | zone_empty: | |
3a2d7fa8 | 1870 | pgdat_resize_unlock(pgdat, &flags); |
7e18adb4 MG |
1871 | |
1872 | /* Sanity check that the next zone really is unpopulated */ | |
1873 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1874 | ||
837566e7 AD |
1875 | pr_info("node %d initialised, %lu pages in %ums\n", |
1876 | pgdat->node_id, nr_pages, jiffies_to_msecs(jiffies - start)); | |
d3cd131d NS |
1877 | |
1878 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1879 | return 0; |
1880 | } | |
c9e97a19 | 1881 | |
c9e97a19 PT |
1882 | /* |
1883 | * If this zone has deferred pages, try to grow it by initializing enough | |
1884 | * deferred pages to satisfy the allocation specified by order, rounded up to | |
1885 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments | |
1886 | * of SECTION_SIZE bytes by initializing struct pages in increments of | |
1887 | * PAGES_PER_SECTION * sizeof(struct page) bytes. | |
1888 | * | |
1889 | * Return true when zone was grown, otherwise return false. We return true even | |
1890 | * when we grow less than requested, to let the caller decide if there are | |
1891 | * enough pages to satisfy the allocation. | |
1892 | * | |
1893 | * Note: We use noinline because this function is needed only during boot, and | |
1894 | * it is called from a __ref function _deferred_grow_zone. This way we are | |
1895 | * making sure that it is not inlined into permanent text section. | |
1896 | */ | |
1897 | static noinline bool __init | |
1898 | deferred_grow_zone(struct zone *zone, unsigned int order) | |
1899 | { | |
c9e97a19 | 1900 | unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION); |
837566e7 | 1901 | pg_data_t *pgdat = zone->zone_pgdat; |
c9e97a19 | 1902 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn; |
0e56acae AD |
1903 | unsigned long spfn, epfn, flags; |
1904 | unsigned long nr_pages = 0; | |
c9e97a19 PT |
1905 | u64 i; |
1906 | ||
1907 | /* Only the last zone may have deferred pages */ | |
1908 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat)) | |
1909 | return false; | |
1910 | ||
1911 | pgdat_resize_lock(pgdat, &flags); | |
1912 | ||
1913 | /* | |
1914 | * If deferred pages have been initialized while we were waiting for | |
1915 | * the lock, return true, as the zone was grown. The caller will retry | |
1916 | * this zone. We won't return to this function since the caller also | |
1917 | * has this static branch. | |
1918 | */ | |
1919 | if (!static_branch_unlikely(&deferred_pages)) { | |
1920 | pgdat_resize_unlock(pgdat, &flags); | |
1921 | return true; | |
1922 | } | |
1923 | ||
1924 | /* | |
1925 | * If someone grew this zone while we were waiting for spinlock, return | |
1926 | * true, as there might be enough pages already. | |
1927 | */ | |
1928 | if (first_deferred_pfn != pgdat->first_deferred_pfn) { | |
1929 | pgdat_resize_unlock(pgdat, &flags); | |
1930 | return true; | |
1931 | } | |
1932 | ||
0e56acae AD |
1933 | /* If the zone is empty somebody else may have cleared out the zone */ |
1934 | if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, | |
1935 | first_deferred_pfn)) { | |
1936 | pgdat->first_deferred_pfn = ULONG_MAX; | |
c9e97a19 | 1937 | pgdat_resize_unlock(pgdat, &flags); |
b9705d87 JG |
1938 | /* Retry only once. */ |
1939 | return first_deferred_pfn != ULONG_MAX; | |
c9e97a19 PT |
1940 | } |
1941 | ||
0e56acae AD |
1942 | /* |
1943 | * Initialize and free pages in MAX_ORDER sized increments so | |
1944 | * that we can avoid introducing any issues with the buddy | |
1945 | * allocator. | |
1946 | */ | |
1947 | while (spfn < epfn) { | |
1948 | /* update our first deferred PFN for this section */ | |
1949 | first_deferred_pfn = spfn; | |
1950 | ||
1951 | nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); | |
c9e97a19 | 1952 | |
0e56acae AD |
1953 | /* We should only stop along section boundaries */ |
1954 | if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION) | |
1955 | continue; | |
c9e97a19 | 1956 | |
0e56acae | 1957 | /* If our quota has been met we can stop here */ |
c9e97a19 PT |
1958 | if (nr_pages >= nr_pages_needed) |
1959 | break; | |
1960 | } | |
1961 | ||
0e56acae | 1962 | pgdat->first_deferred_pfn = spfn; |
c9e97a19 PT |
1963 | pgdat_resize_unlock(pgdat, &flags); |
1964 | ||
1965 | return nr_pages > 0; | |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * deferred_grow_zone() is __init, but it is called from | |
1970 | * get_page_from_freelist() during early boot until deferred_pages permanently | |
1971 | * disables this call. This is why we have refdata wrapper to avoid warning, | |
1972 | * and to ensure that the function body gets unloaded. | |
1973 | */ | |
1974 | static bool __ref | |
1975 | _deferred_grow_zone(struct zone *zone, unsigned int order) | |
1976 | { | |
1977 | return deferred_grow_zone(zone, order); | |
1978 | } | |
1979 | ||
7cf91a98 | 1980 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1981 | |
1982 | void __init page_alloc_init_late(void) | |
1983 | { | |
7cf91a98 | 1984 | struct zone *zone; |
e900a918 | 1985 | int nid; |
7cf91a98 JK |
1986 | |
1987 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b | 1988 | |
d3cd131d NS |
1989 | /* There will be num_node_state(N_MEMORY) threads */ |
1990 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1991 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1992 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1993 | } | |
1994 | ||
1995 | /* Block until all are initialised */ | |
d3cd131d | 1996 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da | 1997 | |
3e8fc007 MG |
1998 | /* |
1999 | * The number of managed pages has changed due to the initialisation | |
2000 | * so the pcpu batch and high limits needs to be updated or the limits | |
2001 | * will be artificially small. | |
2002 | */ | |
2003 | for_each_populated_zone(zone) | |
2004 | zone_pcp_update(zone); | |
2005 | ||
c9e97a19 PT |
2006 | /* |
2007 | * We initialized the rest of the deferred pages. Permanently disable | |
2008 | * on-demand struct page initialization. | |
2009 | */ | |
2010 | static_branch_disable(&deferred_pages); | |
2011 | ||
4248b0da MG |
2012 | /* Reinit limits that are based on free pages after the kernel is up */ |
2013 | files_maxfiles_init(); | |
7cf91a98 | 2014 | #endif |
350e88ba | 2015 | |
3010f876 PT |
2016 | /* Discard memblock private memory */ |
2017 | memblock_discard(); | |
7cf91a98 | 2018 | |
e900a918 DW |
2019 | for_each_node_state(nid, N_MEMORY) |
2020 | shuffle_free_memory(NODE_DATA(nid)); | |
2021 | ||
7cf91a98 JK |
2022 | for_each_populated_zone(zone) |
2023 | set_zone_contiguous(zone); | |
7e18adb4 | 2024 | } |
7e18adb4 | 2025 | |
47118af0 | 2026 | #ifdef CONFIG_CMA |
9cf510a5 | 2027 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
2028 | void __init init_cma_reserved_pageblock(struct page *page) |
2029 | { | |
2030 | unsigned i = pageblock_nr_pages; | |
2031 | struct page *p = page; | |
2032 | ||
2033 | do { | |
2034 | __ClearPageReserved(p); | |
2035 | set_page_count(p, 0); | |
d883c6cf | 2036 | } while (++p, --i); |
47118af0 | 2037 | |
47118af0 | 2038 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
2039 | |
2040 | if (pageblock_order >= MAX_ORDER) { | |
2041 | i = pageblock_nr_pages; | |
2042 | p = page; | |
2043 | do { | |
2044 | set_page_refcounted(p); | |
2045 | __free_pages(p, MAX_ORDER - 1); | |
2046 | p += MAX_ORDER_NR_PAGES; | |
2047 | } while (i -= MAX_ORDER_NR_PAGES); | |
2048 | } else { | |
2049 | set_page_refcounted(page); | |
2050 | __free_pages(page, pageblock_order); | |
2051 | } | |
2052 | ||
3dcc0571 | 2053 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
2054 | } |
2055 | #endif | |
1da177e4 LT |
2056 | |
2057 | /* | |
2058 | * The order of subdivision here is critical for the IO subsystem. | |
2059 | * Please do not alter this order without good reasons and regression | |
2060 | * testing. Specifically, as large blocks of memory are subdivided, | |
2061 | * the order in which smaller blocks are delivered depends on the order | |
2062 | * they're subdivided in this function. This is the primary factor | |
2063 | * influencing the order in which pages are delivered to the IO | |
2064 | * subsystem according to empirical testing, and this is also justified | |
2065 | * by considering the behavior of a buddy system containing a single | |
2066 | * large block of memory acted on by a series of small allocations. | |
2067 | * This behavior is a critical factor in sglist merging's success. | |
2068 | * | |
6d49e352 | 2069 | * -- nyc |
1da177e4 | 2070 | */ |
085cc7d5 | 2071 | static inline void expand(struct zone *zone, struct page *page, |
6ab01363 | 2072 | int low, int high, int migratetype) |
1da177e4 LT |
2073 | { |
2074 | unsigned long size = 1 << high; | |
2075 | ||
2076 | while (high > low) { | |
1da177e4 LT |
2077 | high--; |
2078 | size >>= 1; | |
309381fe | 2079 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 2080 | |
acbc15a4 JK |
2081 | /* |
2082 | * Mark as guard pages (or page), that will allow to | |
2083 | * merge back to allocator when buddy will be freed. | |
2084 | * Corresponding page table entries will not be touched, | |
2085 | * pages will stay not present in virtual address space | |
2086 | */ | |
2087 | if (set_page_guard(zone, &page[size], high, migratetype)) | |
c0a32fc5 | 2088 | continue; |
acbc15a4 | 2089 | |
6ab01363 | 2090 | add_to_free_list(&page[size], zone, high, migratetype); |
1da177e4 LT |
2091 | set_page_order(&page[size], high); |
2092 | } | |
1da177e4 LT |
2093 | } |
2094 | ||
4e611801 | 2095 | static void check_new_page_bad(struct page *page) |
1da177e4 | 2096 | { |
4e611801 VB |
2097 | const char *bad_reason = NULL; |
2098 | unsigned long bad_flags = 0; | |
7bfec6f4 | 2099 | |
833d8a42 WY |
2100 | if (unlikely(page->flags & __PG_HWPOISON)) { |
2101 | /* Don't complain about hwpoisoned pages */ | |
2102 | page_mapcount_reset(page); /* remove PageBuddy */ | |
2103 | return; | |
2104 | } | |
53f9263b | 2105 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
2106 | bad_reason = "nonzero mapcount"; |
2107 | if (unlikely(page->mapping != NULL)) | |
2108 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 2109 | if (unlikely(page_ref_count(page) != 0)) |
136ac591 | 2110 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
2111 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
2112 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
2113 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
2114 | } | |
9edad6ea JW |
2115 | #ifdef CONFIG_MEMCG |
2116 | if (unlikely(page->mem_cgroup)) | |
2117 | bad_reason = "page still charged to cgroup"; | |
2118 | #endif | |
4e611801 VB |
2119 | bad_page(page, bad_reason, bad_flags); |
2120 | } | |
2121 | ||
2122 | /* | |
2123 | * This page is about to be returned from the page allocator | |
2124 | */ | |
2125 | static inline int check_new_page(struct page *page) | |
2126 | { | |
2127 | if (likely(page_expected_state(page, | |
2128 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) | |
2129 | return 0; | |
2130 | ||
2131 | check_new_page_bad(page); | |
2132 | return 1; | |
2a7684a2 WF |
2133 | } |
2134 | ||
bd33ef36 | 2135 | static inline bool free_pages_prezeroed(void) |
1414c7f4 | 2136 | { |
6471384a AP |
2137 | return (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && |
2138 | page_poisoning_enabled()) || want_init_on_free(); | |
1414c7f4 LA |
2139 | } |
2140 | ||
479f854a | 2141 | #ifdef CONFIG_DEBUG_VM |
4462b32c VB |
2142 | /* |
2143 | * With DEBUG_VM enabled, order-0 pages are checked for expected state when | |
2144 | * being allocated from pcp lists. With debug_pagealloc also enabled, they are | |
2145 | * also checked when pcp lists are refilled from the free lists. | |
2146 | */ | |
2147 | static inline bool check_pcp_refill(struct page *page) | |
479f854a | 2148 | { |
8e57f8ac | 2149 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2150 | return check_new_page(page); |
2151 | else | |
2152 | return false; | |
479f854a MG |
2153 | } |
2154 | ||
4462b32c | 2155 | static inline bool check_new_pcp(struct page *page) |
479f854a MG |
2156 | { |
2157 | return check_new_page(page); | |
2158 | } | |
2159 | #else | |
4462b32c VB |
2160 | /* |
2161 | * With DEBUG_VM disabled, free order-0 pages are checked for expected state | |
2162 | * when pcp lists are being refilled from the free lists. With debug_pagealloc | |
2163 | * enabled, they are also checked when being allocated from the pcp lists. | |
2164 | */ | |
2165 | static inline bool check_pcp_refill(struct page *page) | |
479f854a MG |
2166 | { |
2167 | return check_new_page(page); | |
2168 | } | |
4462b32c | 2169 | static inline bool check_new_pcp(struct page *page) |
479f854a | 2170 | { |
8e57f8ac | 2171 | if (debug_pagealloc_enabled_static()) |
4462b32c VB |
2172 | return check_new_page(page); |
2173 | else | |
2174 | return false; | |
479f854a MG |
2175 | } |
2176 | #endif /* CONFIG_DEBUG_VM */ | |
2177 | ||
2178 | static bool check_new_pages(struct page *page, unsigned int order) | |
2179 | { | |
2180 | int i; | |
2181 | for (i = 0; i < (1 << order); i++) { | |
2182 | struct page *p = page + i; | |
2183 | ||
2184 | if (unlikely(check_new_page(p))) | |
2185 | return true; | |
2186 | } | |
2187 | ||
2188 | return false; | |
2189 | } | |
2190 | ||
46f24fd8 JK |
2191 | inline void post_alloc_hook(struct page *page, unsigned int order, |
2192 | gfp_t gfp_flags) | |
2193 | { | |
2194 | set_page_private(page, 0); | |
2195 | set_page_refcounted(page); | |
2196 | ||
2197 | arch_alloc_page(page, order); | |
8e57f8ac | 2198 | if (debug_pagealloc_enabled_static()) |
d6332692 | 2199 | kernel_map_pages(page, 1 << order, 1); |
46f24fd8 | 2200 | kasan_alloc_pages(page, order); |
4117992d | 2201 | kernel_poison_pages(page, 1 << order, 1); |
46f24fd8 JK |
2202 | set_page_owner(page, order, gfp_flags); |
2203 | } | |
2204 | ||
479f854a | 2205 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 2206 | unsigned int alloc_flags) |
2a7684a2 | 2207 | { |
46f24fd8 | 2208 | post_alloc_hook(page, order, gfp_flags); |
17cf4406 | 2209 | |
6471384a AP |
2210 | if (!free_pages_prezeroed() && want_init_on_alloc(gfp_flags)) |
2211 | kernel_init_free_pages(page, 1 << order); | |
17cf4406 NP |
2212 | |
2213 | if (order && (gfp_flags & __GFP_COMP)) | |
2214 | prep_compound_page(page, order); | |
2215 | ||
75379191 | 2216 | /* |
2f064f34 | 2217 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
2218 | * allocate the page. The expectation is that the caller is taking |
2219 | * steps that will free more memory. The caller should avoid the page | |
2220 | * being used for !PFMEMALLOC purposes. | |
2221 | */ | |
2f064f34 MH |
2222 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
2223 | set_page_pfmemalloc(page); | |
2224 | else | |
2225 | clear_page_pfmemalloc(page); | |
1da177e4 LT |
2226 | } |
2227 | ||
56fd56b8 MG |
2228 | /* |
2229 | * Go through the free lists for the given migratetype and remove | |
2230 | * the smallest available page from the freelists | |
2231 | */ | |
85ccc8fa | 2232 | static __always_inline |
728ec980 | 2233 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
56fd56b8 MG |
2234 | int migratetype) |
2235 | { | |
2236 | unsigned int current_order; | |
b8af2941 | 2237 | struct free_area *area; |
56fd56b8 MG |
2238 | struct page *page; |
2239 | ||
2240 | /* Find a page of the appropriate size in the preferred list */ | |
2241 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
2242 | area = &(zone->free_area[current_order]); | |
b03641af | 2243 | page = get_page_from_free_area(area, migratetype); |
a16601c5 GT |
2244 | if (!page) |
2245 | continue; | |
6ab01363 AD |
2246 | del_page_from_free_list(page, zone, current_order); |
2247 | expand(zone, page, order, current_order, migratetype); | |
bb14c2c7 | 2248 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
2249 | return page; |
2250 | } | |
2251 | ||
2252 | return NULL; | |
2253 | } | |
2254 | ||
2255 | ||
b2a0ac88 MG |
2256 | /* |
2257 | * This array describes the order lists are fallen back to when | |
2258 | * the free lists for the desirable migrate type are depleted | |
2259 | */ | |
47118af0 | 2260 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e | 2261 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
974a786e | 2262 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
7ead3342 | 2263 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
47118af0 | 2264 | #ifdef CONFIG_CMA |
974a786e | 2265 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 2266 | #endif |
194159fb | 2267 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 2268 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 2269 | #endif |
b2a0ac88 MG |
2270 | }; |
2271 | ||
dc67647b | 2272 | #ifdef CONFIG_CMA |
85ccc8fa | 2273 | static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
dc67647b JK |
2274 | unsigned int order) |
2275 | { | |
2276 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
2277 | } | |
2278 | #else | |
2279 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
2280 | unsigned int order) { return NULL; } | |
2281 | #endif | |
2282 | ||
c361be55 MG |
2283 | /* |
2284 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 2285 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
2286 | * boundary. If alignment is required, use move_freepages_block() |
2287 | */ | |
02aa0cdd | 2288 | static int move_freepages(struct zone *zone, |
b69a7288 | 2289 | struct page *start_page, struct page *end_page, |
02aa0cdd | 2290 | int migratetype, int *num_movable) |
c361be55 MG |
2291 | { |
2292 | struct page *page; | |
d00181b9 | 2293 | unsigned int order; |
d100313f | 2294 | int pages_moved = 0; |
c361be55 | 2295 | |
c361be55 MG |
2296 | for (page = start_page; page <= end_page;) { |
2297 | if (!pfn_valid_within(page_to_pfn(page))) { | |
2298 | page++; | |
2299 | continue; | |
2300 | } | |
2301 | ||
2302 | if (!PageBuddy(page)) { | |
02aa0cdd VB |
2303 | /* |
2304 | * We assume that pages that could be isolated for | |
2305 | * migration are movable. But we don't actually try | |
2306 | * isolating, as that would be expensive. | |
2307 | */ | |
2308 | if (num_movable && | |
2309 | (PageLRU(page) || __PageMovable(page))) | |
2310 | (*num_movable)++; | |
2311 | ||
c361be55 MG |
2312 | page++; |
2313 | continue; | |
2314 | } | |
2315 | ||
cd961038 DR |
2316 | /* Make sure we are not inadvertently changing nodes */ |
2317 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); | |
2318 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
2319 | ||
c361be55 | 2320 | order = page_order(page); |
6ab01363 | 2321 | move_to_free_list(page, zone, order, migratetype); |
c361be55 | 2322 | page += 1 << order; |
d100313f | 2323 | pages_moved += 1 << order; |
c361be55 MG |
2324 | } |
2325 | ||
d100313f | 2326 | return pages_moved; |
c361be55 MG |
2327 | } |
2328 | ||
ee6f509c | 2329 | int move_freepages_block(struct zone *zone, struct page *page, |
02aa0cdd | 2330 | int migratetype, int *num_movable) |
c361be55 MG |
2331 | { |
2332 | unsigned long start_pfn, end_pfn; | |
2333 | struct page *start_page, *end_page; | |
2334 | ||
4a222127 DR |
2335 | if (num_movable) |
2336 | *num_movable = 0; | |
2337 | ||
c361be55 | 2338 | start_pfn = page_to_pfn(page); |
d9c23400 | 2339 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 2340 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
2341 | end_page = start_page + pageblock_nr_pages - 1; |
2342 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
2343 | |
2344 | /* Do not cross zone boundaries */ | |
108bcc96 | 2345 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 2346 | start_page = page; |
108bcc96 | 2347 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
2348 | return 0; |
2349 | ||
02aa0cdd VB |
2350 | return move_freepages(zone, start_page, end_page, migratetype, |
2351 | num_movable); | |
c361be55 MG |
2352 | } |
2353 | ||
2f66a68f MG |
2354 | static void change_pageblock_range(struct page *pageblock_page, |
2355 | int start_order, int migratetype) | |
2356 | { | |
2357 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
2358 | ||
2359 | while (nr_pageblocks--) { | |
2360 | set_pageblock_migratetype(pageblock_page, migratetype); | |
2361 | pageblock_page += pageblock_nr_pages; | |
2362 | } | |
2363 | } | |
2364 | ||
fef903ef | 2365 | /* |
9c0415eb VB |
2366 | * When we are falling back to another migratetype during allocation, try to |
2367 | * steal extra free pages from the same pageblocks to satisfy further | |
2368 | * allocations, instead of polluting multiple pageblocks. | |
2369 | * | |
2370 | * If we are stealing a relatively large buddy page, it is likely there will | |
2371 | * be more free pages in the pageblock, so try to steal them all. For | |
2372 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
2373 | * as fragmentation caused by those allocations polluting movable pageblocks | |
2374 | * is worse than movable allocations stealing from unmovable and reclaimable | |
2375 | * pageblocks. | |
fef903ef | 2376 | */ |
4eb7dce6 JK |
2377 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2378 | { | |
2379 | /* | |
2380 | * Leaving this order check is intended, although there is | |
2381 | * relaxed order check in next check. The reason is that | |
2382 | * we can actually steal whole pageblock if this condition met, | |
2383 | * but, below check doesn't guarantee it and that is just heuristic | |
2384 | * so could be changed anytime. | |
2385 | */ | |
2386 | if (order >= pageblock_order) | |
2387 | return true; | |
2388 | ||
2389 | if (order >= pageblock_order / 2 || | |
2390 | start_mt == MIGRATE_RECLAIMABLE || | |
2391 | start_mt == MIGRATE_UNMOVABLE || | |
2392 | page_group_by_mobility_disabled) | |
2393 | return true; | |
2394 | ||
2395 | return false; | |
2396 | } | |
2397 | ||
1c30844d MG |
2398 | static inline void boost_watermark(struct zone *zone) |
2399 | { | |
2400 | unsigned long max_boost; | |
2401 | ||
2402 | if (!watermark_boost_factor) | |
2403 | return; | |
14f69140 HW |
2404 | /* |
2405 | * Don't bother in zones that are unlikely to produce results. | |
2406 | * On small machines, including kdump capture kernels running | |
2407 | * in a small area, boosting the watermark can cause an out of | |
2408 | * memory situation immediately. | |
2409 | */ | |
2410 | if ((pageblock_nr_pages * 4) > zone_managed_pages(zone)) | |
2411 | return; | |
1c30844d MG |
2412 | |
2413 | max_boost = mult_frac(zone->_watermark[WMARK_HIGH], | |
2414 | watermark_boost_factor, 10000); | |
94b3334c MG |
2415 | |
2416 | /* | |
2417 | * high watermark may be uninitialised if fragmentation occurs | |
2418 | * very early in boot so do not boost. We do not fall | |
2419 | * through and boost by pageblock_nr_pages as failing | |
2420 | * allocations that early means that reclaim is not going | |
2421 | * to help and it may even be impossible to reclaim the | |
2422 | * boosted watermark resulting in a hang. | |
2423 | */ | |
2424 | if (!max_boost) | |
2425 | return; | |
2426 | ||
1c30844d MG |
2427 | max_boost = max(pageblock_nr_pages, max_boost); |
2428 | ||
2429 | zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, | |
2430 | max_boost); | |
2431 | } | |
2432 | ||
4eb7dce6 JK |
2433 | /* |
2434 | * This function implements actual steal behaviour. If order is large enough, | |
2435 | * we can steal whole pageblock. If not, we first move freepages in this | |
02aa0cdd VB |
2436 | * pageblock to our migratetype and determine how many already-allocated pages |
2437 | * are there in the pageblock with a compatible migratetype. If at least half | |
2438 | * of pages are free or compatible, we can change migratetype of the pageblock | |
2439 | * itself, so pages freed in the future will be put on the correct free list. | |
4eb7dce6 JK |
2440 | */ |
2441 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1c30844d | 2442 | unsigned int alloc_flags, int start_type, bool whole_block) |
fef903ef | 2443 | { |
d00181b9 | 2444 | unsigned int current_order = page_order(page); |
02aa0cdd VB |
2445 | int free_pages, movable_pages, alike_pages; |
2446 | int old_block_type; | |
2447 | ||
2448 | old_block_type = get_pageblock_migratetype(page); | |
fef903ef | 2449 | |
3bc48f96 VB |
2450 | /* |
2451 | * This can happen due to races and we want to prevent broken | |
2452 | * highatomic accounting. | |
2453 | */ | |
02aa0cdd | 2454 | if (is_migrate_highatomic(old_block_type)) |
3bc48f96 VB |
2455 | goto single_page; |
2456 | ||
fef903ef SB |
2457 | /* Take ownership for orders >= pageblock_order */ |
2458 | if (current_order >= pageblock_order) { | |
2459 | change_pageblock_range(page, current_order, start_type); | |
3bc48f96 | 2460 | goto single_page; |
fef903ef SB |
2461 | } |
2462 | ||
1c30844d MG |
2463 | /* |
2464 | * Boost watermarks to increase reclaim pressure to reduce the | |
2465 | * likelihood of future fallbacks. Wake kswapd now as the node | |
2466 | * may be balanced overall and kswapd will not wake naturally. | |
2467 | */ | |
2468 | boost_watermark(zone); | |
2469 | if (alloc_flags & ALLOC_KSWAPD) | |
73444bc4 | 2470 | set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); |
1c30844d | 2471 | |
3bc48f96 VB |
2472 | /* We are not allowed to try stealing from the whole block */ |
2473 | if (!whole_block) | |
2474 | goto single_page; | |
2475 | ||
02aa0cdd VB |
2476 | free_pages = move_freepages_block(zone, page, start_type, |
2477 | &movable_pages); | |
2478 | /* | |
2479 | * Determine how many pages are compatible with our allocation. | |
2480 | * For movable allocation, it's the number of movable pages which | |
2481 | * we just obtained. For other types it's a bit more tricky. | |
2482 | */ | |
2483 | if (start_type == MIGRATE_MOVABLE) { | |
2484 | alike_pages = movable_pages; | |
2485 | } else { | |
2486 | /* | |
2487 | * If we are falling back a RECLAIMABLE or UNMOVABLE allocation | |
2488 | * to MOVABLE pageblock, consider all non-movable pages as | |
2489 | * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or | |
2490 | * vice versa, be conservative since we can't distinguish the | |
2491 | * exact migratetype of non-movable pages. | |
2492 | */ | |
2493 | if (old_block_type == MIGRATE_MOVABLE) | |
2494 | alike_pages = pageblock_nr_pages | |
2495 | - (free_pages + movable_pages); | |
2496 | else | |
2497 | alike_pages = 0; | |
2498 | } | |
2499 | ||
3bc48f96 | 2500 | /* moving whole block can fail due to zone boundary conditions */ |
02aa0cdd | 2501 | if (!free_pages) |
3bc48f96 | 2502 | goto single_page; |
fef903ef | 2503 | |
02aa0cdd VB |
2504 | /* |
2505 | * If a sufficient number of pages in the block are either free or of | |
2506 | * comparable migratability as our allocation, claim the whole block. | |
2507 | */ | |
2508 | if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || | |
4eb7dce6 JK |
2509 | page_group_by_mobility_disabled) |
2510 | set_pageblock_migratetype(page, start_type); | |
3bc48f96 VB |
2511 | |
2512 | return; | |
2513 | ||
2514 | single_page: | |
6ab01363 | 2515 | move_to_free_list(page, zone, current_order, start_type); |
4eb7dce6 JK |
2516 | } |
2517 | ||
2149cdae JK |
2518 | /* |
2519 | * Check whether there is a suitable fallback freepage with requested order. | |
2520 | * If only_stealable is true, this function returns fallback_mt only if | |
2521 | * we can steal other freepages all together. This would help to reduce | |
2522 | * fragmentation due to mixed migratetype pages in one pageblock. | |
2523 | */ | |
2524 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
2525 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
2526 | { |
2527 | int i; | |
2528 | int fallback_mt; | |
2529 | ||
2530 | if (area->nr_free == 0) | |
2531 | return -1; | |
2532 | ||
2533 | *can_steal = false; | |
2534 | for (i = 0;; i++) { | |
2535 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 2536 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
2537 | break; |
2538 | ||
b03641af | 2539 | if (free_area_empty(area, fallback_mt)) |
4eb7dce6 | 2540 | continue; |
fef903ef | 2541 | |
4eb7dce6 JK |
2542 | if (can_steal_fallback(order, migratetype)) |
2543 | *can_steal = true; | |
2544 | ||
2149cdae JK |
2545 | if (!only_stealable) |
2546 | return fallback_mt; | |
2547 | ||
2548 | if (*can_steal) | |
2549 | return fallback_mt; | |
fef903ef | 2550 | } |
4eb7dce6 JK |
2551 | |
2552 | return -1; | |
fef903ef SB |
2553 | } |
2554 | ||
0aaa29a5 MG |
2555 | /* |
2556 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
2557 | * there are no empty page blocks that contain a page with a suitable order | |
2558 | */ | |
2559 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
2560 | unsigned int alloc_order) | |
2561 | { | |
2562 | int mt; | |
2563 | unsigned long max_managed, flags; | |
2564 | ||
2565 | /* | |
2566 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
2567 | * Check is race-prone but harmless. | |
2568 | */ | |
9705bea5 | 2569 | max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages; |
0aaa29a5 MG |
2570 | if (zone->nr_reserved_highatomic >= max_managed) |
2571 | return; | |
2572 | ||
2573 | spin_lock_irqsave(&zone->lock, flags); | |
2574 | ||
2575 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
2576 | if (zone->nr_reserved_highatomic >= max_managed) | |
2577 | goto out_unlock; | |
2578 | ||
2579 | /* Yoink! */ | |
2580 | mt = get_pageblock_migratetype(page); | |
a6ffdc07 XQ |
2581 | if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt) |
2582 | && !is_migrate_cma(mt)) { | |
0aaa29a5 MG |
2583 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2584 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
02aa0cdd | 2585 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL); |
0aaa29a5 MG |
2586 | } |
2587 | ||
2588 | out_unlock: | |
2589 | spin_unlock_irqrestore(&zone->lock, flags); | |
2590 | } | |
2591 | ||
2592 | /* | |
2593 | * Used when an allocation is about to fail under memory pressure. This | |
2594 | * potentially hurts the reliability of high-order allocations when under | |
2595 | * intense memory pressure but failed atomic allocations should be easier | |
2596 | * to recover from than an OOM. | |
29fac03b MK |
2597 | * |
2598 | * If @force is true, try to unreserve a pageblock even though highatomic | |
2599 | * pageblock is exhausted. | |
0aaa29a5 | 2600 | */ |
29fac03b MK |
2601 | static bool unreserve_highatomic_pageblock(const struct alloc_context *ac, |
2602 | bool force) | |
0aaa29a5 MG |
2603 | { |
2604 | struct zonelist *zonelist = ac->zonelist; | |
2605 | unsigned long flags; | |
2606 | struct zoneref *z; | |
2607 | struct zone *zone; | |
2608 | struct page *page; | |
2609 | int order; | |
04c8716f | 2610 | bool ret; |
0aaa29a5 MG |
2611 | |
2612 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
2613 | ac->nodemask) { | |
29fac03b MK |
2614 | /* |
2615 | * Preserve at least one pageblock unless memory pressure | |
2616 | * is really high. | |
2617 | */ | |
2618 | if (!force && zone->nr_reserved_highatomic <= | |
2619 | pageblock_nr_pages) | |
0aaa29a5 MG |
2620 | continue; |
2621 | ||
2622 | spin_lock_irqsave(&zone->lock, flags); | |
2623 | for (order = 0; order < MAX_ORDER; order++) { | |
2624 | struct free_area *area = &(zone->free_area[order]); | |
2625 | ||
b03641af | 2626 | page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC); |
a16601c5 | 2627 | if (!page) |
0aaa29a5 MG |
2628 | continue; |
2629 | ||
0aaa29a5 | 2630 | /* |
4855e4a7 MK |
2631 | * In page freeing path, migratetype change is racy so |
2632 | * we can counter several free pages in a pageblock | |
2633 | * in this loop althoug we changed the pageblock type | |
2634 | * from highatomic to ac->migratetype. So we should | |
2635 | * adjust the count once. | |
0aaa29a5 | 2636 | */ |
a6ffdc07 | 2637 | if (is_migrate_highatomic_page(page)) { |
4855e4a7 MK |
2638 | /* |
2639 | * It should never happen but changes to | |
2640 | * locking could inadvertently allow a per-cpu | |
2641 | * drain to add pages to MIGRATE_HIGHATOMIC | |
2642 | * while unreserving so be safe and watch for | |
2643 | * underflows. | |
2644 | */ | |
2645 | zone->nr_reserved_highatomic -= min( | |
2646 | pageblock_nr_pages, | |
2647 | zone->nr_reserved_highatomic); | |
2648 | } | |
0aaa29a5 MG |
2649 | |
2650 | /* | |
2651 | * Convert to ac->migratetype and avoid the normal | |
2652 | * pageblock stealing heuristics. Minimally, the caller | |
2653 | * is doing the work and needs the pages. More | |
2654 | * importantly, if the block was always converted to | |
2655 | * MIGRATE_UNMOVABLE or another type then the number | |
2656 | * of pageblocks that cannot be completely freed | |
2657 | * may increase. | |
2658 | */ | |
2659 | set_pageblock_migratetype(page, ac->migratetype); | |
02aa0cdd VB |
2660 | ret = move_freepages_block(zone, page, ac->migratetype, |
2661 | NULL); | |
29fac03b MK |
2662 | if (ret) { |
2663 | spin_unlock_irqrestore(&zone->lock, flags); | |
2664 | return ret; | |
2665 | } | |
0aaa29a5 MG |
2666 | } |
2667 | spin_unlock_irqrestore(&zone->lock, flags); | |
2668 | } | |
04c8716f MK |
2669 | |
2670 | return false; | |
0aaa29a5 MG |
2671 | } |
2672 | ||
3bc48f96 VB |
2673 | /* |
2674 | * Try finding a free buddy page on the fallback list and put it on the free | |
2675 | * list of requested migratetype, possibly along with other pages from the same | |
2676 | * block, depending on fragmentation avoidance heuristics. Returns true if | |
2677 | * fallback was found so that __rmqueue_smallest() can grab it. | |
b002529d RV |
2678 | * |
2679 | * The use of signed ints for order and current_order is a deliberate | |
2680 | * deviation from the rest of this file, to make the for loop | |
2681 | * condition simpler. | |
3bc48f96 | 2682 | */ |
85ccc8fa | 2683 | static __always_inline bool |
6bb15450 MG |
2684 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype, |
2685 | unsigned int alloc_flags) | |
b2a0ac88 | 2686 | { |
b8af2941 | 2687 | struct free_area *area; |
b002529d | 2688 | int current_order; |
6bb15450 | 2689 | int min_order = order; |
b2a0ac88 | 2690 | struct page *page; |
4eb7dce6 JK |
2691 | int fallback_mt; |
2692 | bool can_steal; | |
b2a0ac88 | 2693 | |
6bb15450 MG |
2694 | /* |
2695 | * Do not steal pages from freelists belonging to other pageblocks | |
2696 | * i.e. orders < pageblock_order. If there are no local zones free, | |
2697 | * the zonelists will be reiterated without ALLOC_NOFRAGMENT. | |
2698 | */ | |
2699 | if (alloc_flags & ALLOC_NOFRAGMENT) | |
2700 | min_order = pageblock_order; | |
2701 | ||
7a8f58f3 VB |
2702 | /* |
2703 | * Find the largest available free page in the other list. This roughly | |
2704 | * approximates finding the pageblock with the most free pages, which | |
2705 | * would be too costly to do exactly. | |
2706 | */ | |
6bb15450 | 2707 | for (current_order = MAX_ORDER - 1; current_order >= min_order; |
7aeb09f9 | 2708 | --current_order) { |
4eb7dce6 JK |
2709 | area = &(zone->free_area[current_order]); |
2710 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 2711 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
2712 | if (fallback_mt == -1) |
2713 | continue; | |
b2a0ac88 | 2714 | |
7a8f58f3 VB |
2715 | /* |
2716 | * We cannot steal all free pages from the pageblock and the | |
2717 | * requested migratetype is movable. In that case it's better to | |
2718 | * steal and split the smallest available page instead of the | |
2719 | * largest available page, because even if the next movable | |
2720 | * allocation falls back into a different pageblock than this | |
2721 | * one, it won't cause permanent fragmentation. | |
2722 | */ | |
2723 | if (!can_steal && start_migratetype == MIGRATE_MOVABLE | |
2724 | && current_order > order) | |
2725 | goto find_smallest; | |
b2a0ac88 | 2726 | |
7a8f58f3 VB |
2727 | goto do_steal; |
2728 | } | |
e0fff1bd | 2729 | |
7a8f58f3 | 2730 | return false; |
e0fff1bd | 2731 | |
7a8f58f3 VB |
2732 | find_smallest: |
2733 | for (current_order = order; current_order < MAX_ORDER; | |
2734 | current_order++) { | |
2735 | area = &(zone->free_area[current_order]); | |
2736 | fallback_mt = find_suitable_fallback(area, current_order, | |
2737 | start_migratetype, false, &can_steal); | |
2738 | if (fallback_mt != -1) | |
2739 | break; | |
b2a0ac88 MG |
2740 | } |
2741 | ||
7a8f58f3 VB |
2742 | /* |
2743 | * This should not happen - we already found a suitable fallback | |
2744 | * when looking for the largest page. | |
2745 | */ | |
2746 | VM_BUG_ON(current_order == MAX_ORDER); | |
2747 | ||
2748 | do_steal: | |
b03641af | 2749 | page = get_page_from_free_area(area, fallback_mt); |
7a8f58f3 | 2750 | |
1c30844d MG |
2751 | steal_suitable_fallback(zone, page, alloc_flags, start_migratetype, |
2752 | can_steal); | |
7a8f58f3 VB |
2753 | |
2754 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
2755 | start_migratetype, fallback_mt); | |
2756 | ||
2757 | return true; | |
2758 | ||
b2a0ac88 MG |
2759 | } |
2760 | ||
56fd56b8 | 2761 | /* |
1da177e4 LT |
2762 | * Do the hard work of removing an element from the buddy allocator. |
2763 | * Call me with the zone->lock already held. | |
2764 | */ | |
85ccc8fa | 2765 | static __always_inline struct page * |
6bb15450 MG |
2766 | __rmqueue(struct zone *zone, unsigned int order, int migratetype, |
2767 | unsigned int alloc_flags) | |
1da177e4 | 2768 | { |
1da177e4 LT |
2769 | struct page *page; |
2770 | ||
3bc48f96 | 2771 | retry: |
56fd56b8 | 2772 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2773 | if (unlikely(!page)) { |
dc67647b JK |
2774 | if (migratetype == MIGRATE_MOVABLE) |
2775 | page = __rmqueue_cma_fallback(zone, order); | |
2776 | ||
6bb15450 MG |
2777 | if (!page && __rmqueue_fallback(zone, order, migratetype, |
2778 | alloc_flags)) | |
3bc48f96 | 2779 | goto retry; |
728ec980 MG |
2780 | } |
2781 | ||
0d3d062a | 2782 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2783 | return page; |
1da177e4 LT |
2784 | } |
2785 | ||
5f63b720 | 2786 | /* |
1da177e4 LT |
2787 | * Obtain a specified number of elements from the buddy allocator, all under |
2788 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2789 | * Returns the number of new pages which were placed at *list. | |
2790 | */ | |
5f63b720 | 2791 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2792 | unsigned long count, struct list_head *list, |
6bb15450 | 2793 | int migratetype, unsigned int alloc_flags) |
1da177e4 | 2794 | { |
a6de734b | 2795 | int i, alloced = 0; |
5f63b720 | 2796 | |
d34b0733 | 2797 | spin_lock(&zone->lock); |
1da177e4 | 2798 | for (i = 0; i < count; ++i) { |
6bb15450 MG |
2799 | struct page *page = __rmqueue(zone, order, migratetype, |
2800 | alloc_flags); | |
085cc7d5 | 2801 | if (unlikely(page == NULL)) |
1da177e4 | 2802 | break; |
81eabcbe | 2803 | |
479f854a MG |
2804 | if (unlikely(check_pcp_refill(page))) |
2805 | continue; | |
2806 | ||
81eabcbe | 2807 | /* |
0fac3ba5 VB |
2808 | * Split buddy pages returned by expand() are received here in |
2809 | * physical page order. The page is added to the tail of | |
2810 | * caller's list. From the callers perspective, the linked list | |
2811 | * is ordered by page number under some conditions. This is | |
2812 | * useful for IO devices that can forward direction from the | |
2813 | * head, thus also in the physical page order. This is useful | |
2814 | * for IO devices that can merge IO requests if the physical | |
2815 | * pages are ordered properly. | |
81eabcbe | 2816 | */ |
0fac3ba5 | 2817 | list_add_tail(&page->lru, list); |
a6de734b | 2818 | alloced++; |
bb14c2c7 | 2819 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2820 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2821 | -(1 << order)); | |
1da177e4 | 2822 | } |
a6de734b MG |
2823 | |
2824 | /* | |
2825 | * i pages were removed from the buddy list even if some leak due | |
2826 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based | |
2827 | * on i. Do not confuse with 'alloced' which is the number of | |
2828 | * pages added to the pcp list. | |
2829 | */ | |
f2260e6b | 2830 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
d34b0733 | 2831 | spin_unlock(&zone->lock); |
a6de734b | 2832 | return alloced; |
1da177e4 LT |
2833 | } |
2834 | ||
4ae7c039 | 2835 | #ifdef CONFIG_NUMA |
8fce4d8e | 2836 | /* |
4037d452 CL |
2837 | * Called from the vmstat counter updater to drain pagesets of this |
2838 | * currently executing processor on remote nodes after they have | |
2839 | * expired. | |
2840 | * | |
879336c3 CL |
2841 | * Note that this function must be called with the thread pinned to |
2842 | * a single processor. | |
8fce4d8e | 2843 | */ |
4037d452 | 2844 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2845 | { |
4ae7c039 | 2846 | unsigned long flags; |
7be12fc9 | 2847 | int to_drain, batch; |
4ae7c039 | 2848 | |
4037d452 | 2849 | local_irq_save(flags); |
4db0c3c2 | 2850 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2851 | to_drain = min(pcp->count, batch); |
77ba9062 | 2852 | if (to_drain > 0) |
2a13515c | 2853 | free_pcppages_bulk(zone, to_drain, pcp); |
4037d452 | 2854 | local_irq_restore(flags); |
4ae7c039 CL |
2855 | } |
2856 | #endif | |
2857 | ||
9f8f2172 | 2858 | /* |
93481ff0 | 2859 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2860 | * |
2861 | * The processor must either be the current processor and the | |
2862 | * thread pinned to the current processor or a processor that | |
2863 | * is not online. | |
2864 | */ | |
93481ff0 | 2865 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2866 | { |
c54ad30c | 2867 | unsigned long flags; |
93481ff0 VB |
2868 | struct per_cpu_pageset *pset; |
2869 | struct per_cpu_pages *pcp; | |
1da177e4 | 2870 | |
93481ff0 VB |
2871 | local_irq_save(flags); |
2872 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2873 | |
93481ff0 | 2874 | pcp = &pset->pcp; |
77ba9062 | 2875 | if (pcp->count) |
93481ff0 | 2876 | free_pcppages_bulk(zone, pcp->count, pcp); |
93481ff0 VB |
2877 | local_irq_restore(flags); |
2878 | } | |
3dfa5721 | 2879 | |
93481ff0 VB |
2880 | /* |
2881 | * Drain pcplists of all zones on the indicated processor. | |
2882 | * | |
2883 | * The processor must either be the current processor and the | |
2884 | * thread pinned to the current processor or a processor that | |
2885 | * is not online. | |
2886 | */ | |
2887 | static void drain_pages(unsigned int cpu) | |
2888 | { | |
2889 | struct zone *zone; | |
2890 | ||
2891 | for_each_populated_zone(zone) { | |
2892 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2893 | } |
2894 | } | |
1da177e4 | 2895 | |
9f8f2172 CL |
2896 | /* |
2897 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2898 | * |
2899 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2900 | * the single zone's pages. | |
9f8f2172 | 2901 | */ |
93481ff0 | 2902 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2903 | { |
93481ff0 VB |
2904 | int cpu = smp_processor_id(); |
2905 | ||
2906 | if (zone) | |
2907 | drain_pages_zone(cpu, zone); | |
2908 | else | |
2909 | drain_pages(cpu); | |
9f8f2172 CL |
2910 | } |
2911 | ||
0ccce3b9 MG |
2912 | static void drain_local_pages_wq(struct work_struct *work) |
2913 | { | |
d9367bd0 WY |
2914 | struct pcpu_drain *drain; |
2915 | ||
2916 | drain = container_of(work, struct pcpu_drain, work); | |
2917 | ||
a459eeb7 MH |
2918 | /* |
2919 | * drain_all_pages doesn't use proper cpu hotplug protection so | |
2920 | * we can race with cpu offline when the WQ can move this from | |
2921 | * a cpu pinned worker to an unbound one. We can operate on a different | |
2922 | * cpu which is allright but we also have to make sure to not move to | |
2923 | * a different one. | |
2924 | */ | |
2925 | preempt_disable(); | |
d9367bd0 | 2926 | drain_local_pages(drain->zone); |
a459eeb7 | 2927 | preempt_enable(); |
0ccce3b9 MG |
2928 | } |
2929 | ||
9f8f2172 | 2930 | /* |
74046494 GBY |
2931 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2932 | * | |
93481ff0 VB |
2933 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2934 | * | |
0ccce3b9 | 2935 | * Note that this can be extremely slow as the draining happens in a workqueue. |
9f8f2172 | 2936 | */ |
93481ff0 | 2937 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2938 | { |
74046494 | 2939 | int cpu; |
74046494 GBY |
2940 | |
2941 | /* | |
2942 | * Allocate in the BSS so we wont require allocation in | |
2943 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2944 | */ | |
2945 | static cpumask_t cpus_with_pcps; | |
2946 | ||
ce612879 MH |
2947 | /* |
2948 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
2949 | * initialized. | |
2950 | */ | |
2951 | if (WARN_ON_ONCE(!mm_percpu_wq)) | |
2952 | return; | |
2953 | ||
bd233f53 MG |
2954 | /* |
2955 | * Do not drain if one is already in progress unless it's specific to | |
2956 | * a zone. Such callers are primarily CMA and memory hotplug and need | |
2957 | * the drain to be complete when the call returns. | |
2958 | */ | |
2959 | if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) { | |
2960 | if (!zone) | |
2961 | return; | |
2962 | mutex_lock(&pcpu_drain_mutex); | |
2963 | } | |
0ccce3b9 | 2964 | |
74046494 GBY |
2965 | /* |
2966 | * We don't care about racing with CPU hotplug event | |
2967 | * as offline notification will cause the notified | |
2968 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2969 | * disables preemption as part of its processing | |
2970 | */ | |
2971 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2972 | struct per_cpu_pageset *pcp; |
2973 | struct zone *z; | |
74046494 | 2974 | bool has_pcps = false; |
93481ff0 VB |
2975 | |
2976 | if (zone) { | |
74046494 | 2977 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2978 | if (pcp->pcp.count) |
74046494 | 2979 | has_pcps = true; |
93481ff0 VB |
2980 | } else { |
2981 | for_each_populated_zone(z) { | |
2982 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2983 | if (pcp->pcp.count) { | |
2984 | has_pcps = true; | |
2985 | break; | |
2986 | } | |
74046494 GBY |
2987 | } |
2988 | } | |
93481ff0 | 2989 | |
74046494 GBY |
2990 | if (has_pcps) |
2991 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2992 | else | |
2993 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2994 | } | |
0ccce3b9 | 2995 | |
bd233f53 | 2996 | for_each_cpu(cpu, &cpus_with_pcps) { |
d9367bd0 WY |
2997 | struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu); |
2998 | ||
2999 | drain->zone = zone; | |
3000 | INIT_WORK(&drain->work, drain_local_pages_wq); | |
3001 | queue_work_on(cpu, mm_percpu_wq, &drain->work); | |
0ccce3b9 | 3002 | } |
bd233f53 | 3003 | for_each_cpu(cpu, &cpus_with_pcps) |
d9367bd0 | 3004 | flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work); |
bd233f53 MG |
3005 | |
3006 | mutex_unlock(&pcpu_drain_mutex); | |
9f8f2172 CL |
3007 | } |
3008 | ||
296699de | 3009 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3010 | |
556b969a CY |
3011 | /* |
3012 | * Touch the watchdog for every WD_PAGE_COUNT pages. | |
3013 | */ | |
3014 | #define WD_PAGE_COUNT (128*1024) | |
3015 | ||
1da177e4 LT |
3016 | void mark_free_pages(struct zone *zone) |
3017 | { | |
556b969a | 3018 | unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT; |
f623f0db | 3019 | unsigned long flags; |
7aeb09f9 | 3020 | unsigned int order, t; |
86760a2c | 3021 | struct page *page; |
1da177e4 | 3022 | |
8080fc03 | 3023 | if (zone_is_empty(zone)) |
1da177e4 LT |
3024 | return; |
3025 | ||
3026 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 3027 | |
108bcc96 | 3028 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
3029 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
3030 | if (pfn_valid(pfn)) { | |
86760a2c | 3031 | page = pfn_to_page(pfn); |
ba6b0979 | 3032 | |
556b969a CY |
3033 | if (!--page_count) { |
3034 | touch_nmi_watchdog(); | |
3035 | page_count = WD_PAGE_COUNT; | |
3036 | } | |
3037 | ||
ba6b0979 JK |
3038 | if (page_zone(page) != zone) |
3039 | continue; | |
3040 | ||
7be98234 RW |
3041 | if (!swsusp_page_is_forbidden(page)) |
3042 | swsusp_unset_page_free(page); | |
f623f0db | 3043 | } |
1da177e4 | 3044 | |
b2a0ac88 | 3045 | for_each_migratetype_order(order, t) { |
86760a2c GT |
3046 | list_for_each_entry(page, |
3047 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 3048 | unsigned long i; |
1da177e4 | 3049 | |
86760a2c | 3050 | pfn = page_to_pfn(page); |
556b969a CY |
3051 | for (i = 0; i < (1UL << order); i++) { |
3052 | if (!--page_count) { | |
3053 | touch_nmi_watchdog(); | |
3054 | page_count = WD_PAGE_COUNT; | |
3055 | } | |
7be98234 | 3056 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
556b969a | 3057 | } |
f623f0db | 3058 | } |
b2a0ac88 | 3059 | } |
1da177e4 LT |
3060 | spin_unlock_irqrestore(&zone->lock, flags); |
3061 | } | |
e2c55dc8 | 3062 | #endif /* CONFIG_PM */ |
1da177e4 | 3063 | |
2d4894b5 | 3064 | static bool free_unref_page_prepare(struct page *page, unsigned long pfn) |
1da177e4 | 3065 | { |
5f8dcc21 | 3066 | int migratetype; |
1da177e4 | 3067 | |
4db7548c | 3068 | if (!free_pcp_prepare(page)) |
9cca35d4 | 3069 | return false; |
689bcebf | 3070 | |
dc4b0caf | 3071 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 3072 | set_pcppage_migratetype(page, migratetype); |
9cca35d4 MG |
3073 | return true; |
3074 | } | |
3075 | ||
2d4894b5 | 3076 | static void free_unref_page_commit(struct page *page, unsigned long pfn) |
9cca35d4 MG |
3077 | { |
3078 | struct zone *zone = page_zone(page); | |
3079 | struct per_cpu_pages *pcp; | |
3080 | int migratetype; | |
3081 | ||
3082 | migratetype = get_pcppage_migratetype(page); | |
d34b0733 | 3083 | __count_vm_event(PGFREE); |
da456f14 | 3084 | |
5f8dcc21 MG |
3085 | /* |
3086 | * We only track unmovable, reclaimable and movable on pcp lists. | |
3087 | * Free ISOLATE pages back to the allocator because they are being | |
a6ffdc07 | 3088 | * offlined but treat HIGHATOMIC as movable pages so we can get those |
5f8dcc21 MG |
3089 | * areas back if necessary. Otherwise, we may have to free |
3090 | * excessively into the page allocator | |
3091 | */ | |
3092 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 3093 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 3094 | free_one_page(zone, page, pfn, 0, migratetype); |
9cca35d4 | 3095 | return; |
5f8dcc21 MG |
3096 | } |
3097 | migratetype = MIGRATE_MOVABLE; | |
3098 | } | |
3099 | ||
99dcc3e5 | 3100 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2d4894b5 | 3101 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 3102 | pcp->count++; |
48db57f8 | 3103 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 3104 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb | 3105 | free_pcppages_bulk(zone, batch, pcp); |
48db57f8 | 3106 | } |
9cca35d4 | 3107 | } |
5f8dcc21 | 3108 | |
9cca35d4 MG |
3109 | /* |
3110 | * Free a 0-order page | |
9cca35d4 | 3111 | */ |
2d4894b5 | 3112 | void free_unref_page(struct page *page) |
9cca35d4 MG |
3113 | { |
3114 | unsigned long flags; | |
3115 | unsigned long pfn = page_to_pfn(page); | |
3116 | ||
2d4894b5 | 3117 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3118 | return; |
3119 | ||
3120 | local_irq_save(flags); | |
2d4894b5 | 3121 | free_unref_page_commit(page, pfn); |
d34b0733 | 3122 | local_irq_restore(flags); |
1da177e4 LT |
3123 | } |
3124 | ||
cc59850e KK |
3125 | /* |
3126 | * Free a list of 0-order pages | |
3127 | */ | |
2d4894b5 | 3128 | void free_unref_page_list(struct list_head *list) |
cc59850e KK |
3129 | { |
3130 | struct page *page, *next; | |
9cca35d4 | 3131 | unsigned long flags, pfn; |
c24ad77d | 3132 | int batch_count = 0; |
9cca35d4 MG |
3133 | |
3134 | /* Prepare pages for freeing */ | |
3135 | list_for_each_entry_safe(page, next, list, lru) { | |
3136 | pfn = page_to_pfn(page); | |
2d4894b5 | 3137 | if (!free_unref_page_prepare(page, pfn)) |
9cca35d4 MG |
3138 | list_del(&page->lru); |
3139 | set_page_private(page, pfn); | |
3140 | } | |
cc59850e | 3141 | |
9cca35d4 | 3142 | local_irq_save(flags); |
cc59850e | 3143 | list_for_each_entry_safe(page, next, list, lru) { |
9cca35d4 MG |
3144 | unsigned long pfn = page_private(page); |
3145 | ||
3146 | set_page_private(page, 0); | |
2d4894b5 MG |
3147 | trace_mm_page_free_batched(page); |
3148 | free_unref_page_commit(page, pfn); | |
c24ad77d LS |
3149 | |
3150 | /* | |
3151 | * Guard against excessive IRQ disabled times when we get | |
3152 | * a large list of pages to free. | |
3153 | */ | |
3154 | if (++batch_count == SWAP_CLUSTER_MAX) { | |
3155 | local_irq_restore(flags); | |
3156 | batch_count = 0; | |
3157 | local_irq_save(flags); | |
3158 | } | |
cc59850e | 3159 | } |
9cca35d4 | 3160 | local_irq_restore(flags); |
cc59850e KK |
3161 | } |
3162 | ||
8dfcc9ba NP |
3163 | /* |
3164 | * split_page takes a non-compound higher-order page, and splits it into | |
3165 | * n (1<<order) sub-pages: page[0..n] | |
3166 | * Each sub-page must be freed individually. | |
3167 | * | |
3168 | * Note: this is probably too low level an operation for use in drivers. | |
3169 | * Please consult with lkml before using this in your driver. | |
3170 | */ | |
3171 | void split_page(struct page *page, unsigned int order) | |
3172 | { | |
3173 | int i; | |
3174 | ||
309381fe SL |
3175 | VM_BUG_ON_PAGE(PageCompound(page), page); |
3176 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 | 3177 | |
a9627bc5 | 3178 | for (i = 1; i < (1 << order); i++) |
7835e98b | 3179 | set_page_refcounted(page + i); |
a9627bc5 | 3180 | split_page_owner(page, order); |
8dfcc9ba | 3181 | } |
5853ff23 | 3182 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 3183 | |
3c605096 | 3184 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 3185 | { |
748446bb MG |
3186 | unsigned long watermark; |
3187 | struct zone *zone; | |
2139cbe6 | 3188 | int mt; |
748446bb MG |
3189 | |
3190 | BUG_ON(!PageBuddy(page)); | |
3191 | ||
3192 | zone = page_zone(page); | |
2e30abd1 | 3193 | mt = get_pageblock_migratetype(page); |
748446bb | 3194 | |
194159fb | 3195 | if (!is_migrate_isolate(mt)) { |
8348faf9 VB |
3196 | /* |
3197 | * Obey watermarks as if the page was being allocated. We can | |
3198 | * emulate a high-order watermark check with a raised order-0 | |
3199 | * watermark, because we already know our high-order page | |
3200 | * exists. | |
3201 | */ | |
fd1444b2 | 3202 | watermark = zone->_watermark[WMARK_MIN] + (1UL << order); |
d883c6cf | 3203 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2e30abd1 MS |
3204 | return 0; |
3205 | ||
8fb74b9f | 3206 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 3207 | } |
748446bb MG |
3208 | |
3209 | /* Remove page from free list */ | |
b03641af | 3210 | |
6ab01363 | 3211 | del_page_from_free_list(page, zone, order); |
2139cbe6 | 3212 | |
400bc7fd | 3213 | /* |
3214 | * Set the pageblock if the isolated page is at least half of a | |
3215 | * pageblock | |
3216 | */ | |
748446bb MG |
3217 | if (order >= pageblock_order - 1) { |
3218 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
3219 | for (; page < endpage; page += pageblock_nr_pages) { |
3220 | int mt = get_pageblock_migratetype(page); | |
88ed365e | 3221 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) |
a6ffdc07 | 3222 | && !is_migrate_highatomic(mt)) |
47118af0 MN |
3223 | set_pageblock_migratetype(page, |
3224 | MIGRATE_MOVABLE); | |
3225 | } | |
748446bb MG |
3226 | } |
3227 | ||
f3a14ced | 3228 | |
8fb74b9f | 3229 | return 1UL << order; |
1fb3f8ca MG |
3230 | } |
3231 | ||
624f58d8 AD |
3232 | /** |
3233 | * __putback_isolated_page - Return a now-isolated page back where we got it | |
3234 | * @page: Page that was isolated | |
3235 | * @order: Order of the isolated page | |
e6a0a7ad | 3236 | * @mt: The page's pageblock's migratetype |
624f58d8 AD |
3237 | * |
3238 | * This function is meant to return a page pulled from the free lists via | |
3239 | * __isolate_free_page back to the free lists they were pulled from. | |
3240 | */ | |
3241 | void __putback_isolated_page(struct page *page, unsigned int order, int mt) | |
3242 | { | |
3243 | struct zone *zone = page_zone(page); | |
3244 | ||
3245 | /* zone lock should be held when this function is called */ | |
3246 | lockdep_assert_held(&zone->lock); | |
3247 | ||
3248 | /* Return isolated page to tail of freelist. */ | |
36e66c55 | 3249 | __free_one_page(page, page_to_pfn(page), zone, order, mt, false); |
624f58d8 AD |
3250 | } |
3251 | ||
060e7417 MG |
3252 | /* |
3253 | * Update NUMA hit/miss statistics | |
3254 | * | |
3255 | * Must be called with interrupts disabled. | |
060e7417 | 3256 | */ |
41b6167e | 3257 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z) |
060e7417 MG |
3258 | { |
3259 | #ifdef CONFIG_NUMA | |
3a321d2a | 3260 | enum numa_stat_item local_stat = NUMA_LOCAL; |
060e7417 | 3261 | |
4518085e KW |
3262 | /* skip numa counters update if numa stats is disabled */ |
3263 | if (!static_branch_likely(&vm_numa_stat_key)) | |
3264 | return; | |
3265 | ||
c1093b74 | 3266 | if (zone_to_nid(z) != numa_node_id()) |
060e7417 | 3267 | local_stat = NUMA_OTHER; |
060e7417 | 3268 | |
c1093b74 | 3269 | if (zone_to_nid(z) == zone_to_nid(preferred_zone)) |
3a321d2a | 3270 | __inc_numa_state(z, NUMA_HIT); |
2df26639 | 3271 | else { |
3a321d2a KW |
3272 | __inc_numa_state(z, NUMA_MISS); |
3273 | __inc_numa_state(preferred_zone, NUMA_FOREIGN); | |
060e7417 | 3274 | } |
3a321d2a | 3275 | __inc_numa_state(z, local_stat); |
060e7417 MG |
3276 | #endif |
3277 | } | |
3278 | ||
066b2393 MG |
3279 | /* Remove page from the per-cpu list, caller must protect the list */ |
3280 | static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype, | |
6bb15450 | 3281 | unsigned int alloc_flags, |
453f85d4 | 3282 | struct per_cpu_pages *pcp, |
066b2393 MG |
3283 | struct list_head *list) |
3284 | { | |
3285 | struct page *page; | |
3286 | ||
3287 | do { | |
3288 | if (list_empty(list)) { | |
3289 | pcp->count += rmqueue_bulk(zone, 0, | |
3290 | pcp->batch, list, | |
6bb15450 | 3291 | migratetype, alloc_flags); |
066b2393 MG |
3292 | if (unlikely(list_empty(list))) |
3293 | return NULL; | |
3294 | } | |
3295 | ||
453f85d4 | 3296 | page = list_first_entry(list, struct page, lru); |
066b2393 MG |
3297 | list_del(&page->lru); |
3298 | pcp->count--; | |
3299 | } while (check_new_pcp(page)); | |
3300 | ||
3301 | return page; | |
3302 | } | |
3303 | ||
3304 | /* Lock and remove page from the per-cpu list */ | |
3305 | static struct page *rmqueue_pcplist(struct zone *preferred_zone, | |
1c52e6d0 YS |
3306 | struct zone *zone, gfp_t gfp_flags, |
3307 | int migratetype, unsigned int alloc_flags) | |
066b2393 MG |
3308 | { |
3309 | struct per_cpu_pages *pcp; | |
3310 | struct list_head *list; | |
066b2393 | 3311 | struct page *page; |
d34b0733 | 3312 | unsigned long flags; |
066b2393 | 3313 | |
d34b0733 | 3314 | local_irq_save(flags); |
066b2393 MG |
3315 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3316 | list = &pcp->lists[migratetype]; | |
6bb15450 | 3317 | page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list); |
066b2393 | 3318 | if (page) { |
1c52e6d0 | 3319 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1); |
066b2393 MG |
3320 | zone_statistics(preferred_zone, zone); |
3321 | } | |
d34b0733 | 3322 | local_irq_restore(flags); |
066b2393 MG |
3323 | return page; |
3324 | } | |
3325 | ||
1da177e4 | 3326 | /* |
75379191 | 3327 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 3328 | */ |
0a15c3e9 | 3329 | static inline |
066b2393 | 3330 | struct page *rmqueue(struct zone *preferred_zone, |
7aeb09f9 | 3331 | struct zone *zone, unsigned int order, |
c603844b MG |
3332 | gfp_t gfp_flags, unsigned int alloc_flags, |
3333 | int migratetype) | |
1da177e4 LT |
3334 | { |
3335 | unsigned long flags; | |
689bcebf | 3336 | struct page *page; |
1da177e4 | 3337 | |
d34b0733 | 3338 | if (likely(order == 0)) { |
1c52e6d0 YS |
3339 | page = rmqueue_pcplist(preferred_zone, zone, gfp_flags, |
3340 | migratetype, alloc_flags); | |
066b2393 MG |
3341 | goto out; |
3342 | } | |
83b9355b | 3343 | |
066b2393 MG |
3344 | /* |
3345 | * We most definitely don't want callers attempting to | |
3346 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
3347 | */ | |
3348 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
3349 | spin_lock_irqsave(&zone->lock, flags); | |
0aaa29a5 | 3350 | |
066b2393 MG |
3351 | do { |
3352 | page = NULL; | |
3353 | if (alloc_flags & ALLOC_HARDER) { | |
3354 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
3355 | if (page) | |
3356 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
3357 | } | |
a74609fa | 3358 | if (!page) |
6bb15450 | 3359 | page = __rmqueue(zone, order, migratetype, alloc_flags); |
066b2393 MG |
3360 | } while (page && check_new_pages(page, order)); |
3361 | spin_unlock(&zone->lock); | |
3362 | if (!page) | |
3363 | goto failed; | |
3364 | __mod_zone_freepage_state(zone, -(1 << order), | |
3365 | get_pcppage_migratetype(page)); | |
1da177e4 | 3366 | |
16709d1d | 3367 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
41b6167e | 3368 | zone_statistics(preferred_zone, zone); |
a74609fa | 3369 | local_irq_restore(flags); |
1da177e4 | 3370 | |
066b2393 | 3371 | out: |
73444bc4 MG |
3372 | /* Separate test+clear to avoid unnecessary atomics */ |
3373 | if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) { | |
3374 | clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); | |
3375 | wakeup_kswapd(zone, 0, 0, zone_idx(zone)); | |
3376 | } | |
3377 | ||
066b2393 | 3378 | VM_BUG_ON_PAGE(page && bad_range(zone, page), page); |
1da177e4 | 3379 | return page; |
a74609fa NP |
3380 | |
3381 | failed: | |
3382 | local_irq_restore(flags); | |
a74609fa | 3383 | return NULL; |
1da177e4 LT |
3384 | } |
3385 | ||
933e312e AM |
3386 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
3387 | ||
b2588c4b | 3388 | static struct { |
933e312e AM |
3389 | struct fault_attr attr; |
3390 | ||
621a5f7a | 3391 | bool ignore_gfp_highmem; |
71baba4b | 3392 | bool ignore_gfp_reclaim; |
54114994 | 3393 | u32 min_order; |
933e312e AM |
3394 | } fail_page_alloc = { |
3395 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 3396 | .ignore_gfp_reclaim = true, |
621a5f7a | 3397 | .ignore_gfp_highmem = true, |
54114994 | 3398 | .min_order = 1, |
933e312e AM |
3399 | }; |
3400 | ||
3401 | static int __init setup_fail_page_alloc(char *str) | |
3402 | { | |
3403 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
3404 | } | |
3405 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
3406 | ||
af3b8544 | 3407 | static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3408 | { |
54114994 | 3409 | if (order < fail_page_alloc.min_order) |
deaf386e | 3410 | return false; |
933e312e | 3411 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 3412 | return false; |
933e312e | 3413 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 3414 | return false; |
71baba4b MG |
3415 | if (fail_page_alloc.ignore_gfp_reclaim && |
3416 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 3417 | return false; |
933e312e AM |
3418 | |
3419 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
3420 | } | |
3421 | ||
3422 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
3423 | ||
3424 | static int __init fail_page_alloc_debugfs(void) | |
3425 | { | |
0825a6f9 | 3426 | umode_t mode = S_IFREG | 0600; |
933e312e | 3427 | struct dentry *dir; |
933e312e | 3428 | |
dd48c085 AM |
3429 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
3430 | &fail_page_alloc.attr); | |
b2588c4b | 3431 | |
d9f7979c GKH |
3432 | debugfs_create_bool("ignore-gfp-wait", mode, dir, |
3433 | &fail_page_alloc.ignore_gfp_reclaim); | |
3434 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
3435 | &fail_page_alloc.ignore_gfp_highmem); | |
3436 | debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order); | |
933e312e | 3437 | |
d9f7979c | 3438 | return 0; |
933e312e AM |
3439 | } |
3440 | ||
3441 | late_initcall(fail_page_alloc_debugfs); | |
3442 | ||
3443 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
3444 | ||
3445 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
3446 | ||
af3b8544 | 3447 | static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 3448 | { |
deaf386e | 3449 | return false; |
933e312e AM |
3450 | } |
3451 | ||
3452 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
3453 | ||
af3b8544 BP |
3454 | static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
3455 | { | |
3456 | return __should_fail_alloc_page(gfp_mask, order); | |
3457 | } | |
3458 | ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE); | |
3459 | ||
1da177e4 | 3460 | /* |
97a16fc8 MG |
3461 | * Return true if free base pages are above 'mark'. For high-order checks it |
3462 | * will return true of the order-0 watermark is reached and there is at least | |
3463 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
3464 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 3465 | */ |
86a294a8 MH |
3466 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
3467 | int classzone_idx, unsigned int alloc_flags, | |
3468 | long free_pages) | |
1da177e4 | 3469 | { |
d23ad423 | 3470 | long min = mark; |
1da177e4 | 3471 | int o; |
cd04ae1e | 3472 | const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM)); |
1da177e4 | 3473 | |
0aaa29a5 | 3474 | /* free_pages may go negative - that's OK */ |
df0a6daa | 3475 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 3476 | |
7fb1d9fc | 3477 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 3478 | min -= min / 2; |
0aaa29a5 MG |
3479 | |
3480 | /* | |
3481 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
3482 | * the high-atomic reserves. This will over-estimate the size of the | |
3483 | * atomic reserve but it avoids a search. | |
3484 | */ | |
cd04ae1e | 3485 | if (likely(!alloc_harder)) { |
0aaa29a5 | 3486 | free_pages -= z->nr_reserved_highatomic; |
cd04ae1e MH |
3487 | } else { |
3488 | /* | |
3489 | * OOM victims can try even harder than normal ALLOC_HARDER | |
3490 | * users on the grounds that it's definitely going to be in | |
3491 | * the exit path shortly and free memory. Any allocation it | |
3492 | * makes during the free path will be small and short-lived. | |
3493 | */ | |
3494 | if (alloc_flags & ALLOC_OOM) | |
3495 | min -= min / 2; | |
3496 | else | |
3497 | min -= min / 4; | |
3498 | } | |
3499 | ||
e2b19197 | 3500 | |
d883c6cf JK |
3501 | #ifdef CONFIG_CMA |
3502 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3503 | if (!(alloc_flags & ALLOC_CMA)) | |
3504 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); | |
3505 | #endif | |
3506 | ||
97a16fc8 MG |
3507 | /* |
3508 | * Check watermarks for an order-0 allocation request. If these | |
3509 | * are not met, then a high-order request also cannot go ahead | |
3510 | * even if a suitable page happened to be free. | |
3511 | */ | |
3512 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 3513 | return false; |
1da177e4 | 3514 | |
97a16fc8 MG |
3515 | /* If this is an order-0 request then the watermark is fine */ |
3516 | if (!order) | |
3517 | return true; | |
3518 | ||
3519 | /* For a high-order request, check at least one suitable page is free */ | |
3520 | for (o = order; o < MAX_ORDER; o++) { | |
3521 | struct free_area *area = &z->free_area[o]; | |
3522 | int mt; | |
3523 | ||
3524 | if (!area->nr_free) | |
3525 | continue; | |
3526 | ||
97a16fc8 | 3527 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
b03641af | 3528 | if (!free_area_empty(area, mt)) |
97a16fc8 MG |
3529 | return true; |
3530 | } | |
3531 | ||
3532 | #ifdef CONFIG_CMA | |
d883c6cf | 3533 | if ((alloc_flags & ALLOC_CMA) && |
b03641af | 3534 | !free_area_empty(area, MIGRATE_CMA)) { |
97a16fc8 | 3535 | return true; |
d883c6cf | 3536 | } |
97a16fc8 | 3537 | #endif |
76089d00 | 3538 | if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC)) |
b050e376 | 3539 | return true; |
1da177e4 | 3540 | } |
97a16fc8 | 3541 | return false; |
88f5acf8 MG |
3542 | } |
3543 | ||
7aeb09f9 | 3544 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 3545 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
3546 | { |
3547 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3548 | zone_page_state(z, NR_FREE_PAGES)); | |
3549 | } | |
3550 | ||
48ee5f36 MG |
3551 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
3552 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
3553 | { | |
3554 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
d883c6cf JK |
3555 | long cma_pages = 0; |
3556 | ||
3557 | #ifdef CONFIG_CMA | |
3558 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
3559 | if (!(alloc_flags & ALLOC_CMA)) | |
3560 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
3561 | #endif | |
48ee5f36 MG |
3562 | |
3563 | /* | |
3564 | * Fast check for order-0 only. If this fails then the reserves | |
3565 | * need to be calculated. There is a corner case where the check | |
3566 | * passes but only the high-order atomic reserve are free. If | |
3567 | * the caller is !atomic then it'll uselessly search the free | |
3568 | * list. That corner case is then slower but it is harmless. | |
3569 | */ | |
d883c6cf | 3570 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) |
48ee5f36 MG |
3571 | return true; |
3572 | ||
3573 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
3574 | free_pages); | |
3575 | } | |
3576 | ||
7aeb09f9 | 3577 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 3578 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
3579 | { |
3580 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
3581 | ||
3582 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
3583 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
3584 | ||
e2b19197 | 3585 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 3586 | free_pages); |
1da177e4 LT |
3587 | } |
3588 | ||
9276b1bc | 3589 | #ifdef CONFIG_NUMA |
957f822a DR |
3590 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3591 | { | |
e02dc017 | 3592 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
a55c7454 | 3593 | node_reclaim_distance; |
957f822a | 3594 | } |
9276b1bc | 3595 | #else /* CONFIG_NUMA */ |
957f822a DR |
3596 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3597 | { | |
3598 | return true; | |
3599 | } | |
9276b1bc PJ |
3600 | #endif /* CONFIG_NUMA */ |
3601 | ||
6bb15450 MG |
3602 | /* |
3603 | * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid | |
3604 | * fragmentation is subtle. If the preferred zone was HIGHMEM then | |
3605 | * premature use of a lower zone may cause lowmem pressure problems that | |
3606 | * are worse than fragmentation. If the next zone is ZONE_DMA then it is | |
3607 | * probably too small. It only makes sense to spread allocations to avoid | |
3608 | * fragmentation between the Normal and DMA32 zones. | |
3609 | */ | |
3610 | static inline unsigned int | |
0a79cdad | 3611 | alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask) |
6bb15450 | 3612 | { |
736838e9 | 3613 | unsigned int alloc_flags; |
0a79cdad | 3614 | |
736838e9 MN |
3615 | /* |
3616 | * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
3617 | * to save a branch. | |
3618 | */ | |
3619 | alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM); | |
0a79cdad MG |
3620 | |
3621 | #ifdef CONFIG_ZONE_DMA32 | |
8139ad04 AR |
3622 | if (!zone) |
3623 | return alloc_flags; | |
3624 | ||
6bb15450 | 3625 | if (zone_idx(zone) != ZONE_NORMAL) |
8118b82e | 3626 | return alloc_flags; |
6bb15450 MG |
3627 | |
3628 | /* | |
3629 | * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and | |
3630 | * the pointer is within zone->zone_pgdat->node_zones[]. Also assume | |
3631 | * on UMA that if Normal is populated then so is DMA32. | |
3632 | */ | |
3633 | BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); | |
3634 | if (nr_online_nodes > 1 && !populated_zone(--zone)) | |
8118b82e | 3635 | return alloc_flags; |
6bb15450 | 3636 | |
8118b82e | 3637 | alloc_flags |= ALLOC_NOFRAGMENT; |
0a79cdad MG |
3638 | #endif /* CONFIG_ZONE_DMA32 */ |
3639 | return alloc_flags; | |
6bb15450 | 3640 | } |
6bb15450 | 3641 | |
7fb1d9fc | 3642 | /* |
0798e519 | 3643 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
3644 | * a page. |
3645 | */ | |
3646 | static struct page * | |
a9263751 VB |
3647 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3648 | const struct alloc_context *ac) | |
753ee728 | 3649 | { |
6bb15450 | 3650 | struct zoneref *z; |
5117f45d | 3651 | struct zone *zone; |
3b8c0be4 | 3652 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
6bb15450 | 3653 | bool no_fallback; |
3b8c0be4 | 3654 | |
6bb15450 | 3655 | retry: |
7fb1d9fc | 3656 | /* |
9276b1bc | 3657 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 3658 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 3659 | */ |
6bb15450 MG |
3660 | no_fallback = alloc_flags & ALLOC_NOFRAGMENT; |
3661 | z = ac->preferred_zoneref; | |
c33d6c06 | 3662 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 3663 | ac->nodemask) { |
be06af00 | 3664 | struct page *page; |
e085dbc5 JW |
3665 | unsigned long mark; |
3666 | ||
664eedde MG |
3667 | if (cpusets_enabled() && |
3668 | (alloc_flags & ALLOC_CPUSET) && | |
002f2906 | 3669 | !__cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 3670 | continue; |
a756cf59 JW |
3671 | /* |
3672 | * When allocating a page cache page for writing, we | |
281e3726 MG |
3673 | * want to get it from a node that is within its dirty |
3674 | * limit, such that no single node holds more than its | |
a756cf59 | 3675 | * proportional share of globally allowed dirty pages. |
281e3726 | 3676 | * The dirty limits take into account the node's |
a756cf59 JW |
3677 | * lowmem reserves and high watermark so that kswapd |
3678 | * should be able to balance it without having to | |
3679 | * write pages from its LRU list. | |
3680 | * | |
a756cf59 | 3681 | * XXX: For now, allow allocations to potentially |
281e3726 | 3682 | * exceed the per-node dirty limit in the slowpath |
c9ab0c4f | 3683 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 | 3684 | * which is important when on a NUMA setup the allowed |
281e3726 | 3685 | * nodes are together not big enough to reach the |
a756cf59 | 3686 | * global limit. The proper fix for these situations |
281e3726 | 3687 | * will require awareness of nodes in the |
a756cf59 JW |
3688 | * dirty-throttling and the flusher threads. |
3689 | */ | |
3b8c0be4 MG |
3690 | if (ac->spread_dirty_pages) { |
3691 | if (last_pgdat_dirty_limit == zone->zone_pgdat) | |
3692 | continue; | |
3693 | ||
3694 | if (!node_dirty_ok(zone->zone_pgdat)) { | |
3695 | last_pgdat_dirty_limit = zone->zone_pgdat; | |
3696 | continue; | |
3697 | } | |
3698 | } | |
7fb1d9fc | 3699 | |
6bb15450 MG |
3700 | if (no_fallback && nr_online_nodes > 1 && |
3701 | zone != ac->preferred_zoneref->zone) { | |
3702 | int local_nid; | |
3703 | ||
3704 | /* | |
3705 | * If moving to a remote node, retry but allow | |
3706 | * fragmenting fallbacks. Locality is more important | |
3707 | * than fragmentation avoidance. | |
3708 | */ | |
3709 | local_nid = zone_to_nid(ac->preferred_zoneref->zone); | |
3710 | if (zone_to_nid(zone) != local_nid) { | |
3711 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3712 | goto retry; | |
3713 | } | |
3714 | } | |
3715 | ||
a9214443 | 3716 | mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
48ee5f36 | 3717 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 3718 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
3719 | int ret; |
3720 | ||
c9e97a19 PT |
3721 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
3722 | /* | |
3723 | * Watermark failed for this zone, but see if we can | |
3724 | * grow this zone if it contains deferred pages. | |
3725 | */ | |
3726 | if (static_branch_unlikely(&deferred_pages)) { | |
3727 | if (_deferred_grow_zone(zone, order)) | |
3728 | goto try_this_zone; | |
3729 | } | |
3730 | #endif | |
5dab2911 MG |
3731 | /* Checked here to keep the fast path fast */ |
3732 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
3733 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
3734 | goto try_this_zone; | |
3735 | ||
a5f5f91d | 3736 | if (node_reclaim_mode == 0 || |
c33d6c06 | 3737 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
3738 | continue; |
3739 | ||
a5f5f91d | 3740 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
fa5e084e | 3741 | switch (ret) { |
a5f5f91d | 3742 | case NODE_RECLAIM_NOSCAN: |
fa5e084e | 3743 | /* did not scan */ |
cd38b115 | 3744 | continue; |
a5f5f91d | 3745 | case NODE_RECLAIM_FULL: |
fa5e084e | 3746 | /* scanned but unreclaimable */ |
cd38b115 | 3747 | continue; |
fa5e084e MG |
3748 | default: |
3749 | /* did we reclaim enough */ | |
fed2719e | 3750 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 3751 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
3752 | goto try_this_zone; |
3753 | ||
fed2719e | 3754 | continue; |
0798e519 | 3755 | } |
7fb1d9fc RS |
3756 | } |
3757 | ||
fa5e084e | 3758 | try_this_zone: |
066b2393 | 3759 | page = rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 3760 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 | 3761 | if (page) { |
479f854a | 3762 | prep_new_page(page, order, gfp_mask, alloc_flags); |
0aaa29a5 MG |
3763 | |
3764 | /* | |
3765 | * If this is a high-order atomic allocation then check | |
3766 | * if the pageblock should be reserved for the future | |
3767 | */ | |
3768 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
3769 | reserve_highatomic_pageblock(page, zone, order); | |
3770 | ||
75379191 | 3771 | return page; |
c9e97a19 PT |
3772 | } else { |
3773 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
3774 | /* Try again if zone has deferred pages */ | |
3775 | if (static_branch_unlikely(&deferred_pages)) { | |
3776 | if (_deferred_grow_zone(zone, order)) | |
3777 | goto try_this_zone; | |
3778 | } | |
3779 | #endif | |
75379191 | 3780 | } |
54a6eb5c | 3781 | } |
9276b1bc | 3782 | |
6bb15450 MG |
3783 | /* |
3784 | * It's possible on a UMA machine to get through all zones that are | |
3785 | * fragmented. If avoiding fragmentation, reset and try again. | |
3786 | */ | |
3787 | if (no_fallback) { | |
3788 | alloc_flags &= ~ALLOC_NOFRAGMENT; | |
3789 | goto retry; | |
3790 | } | |
3791 | ||
4ffeaf35 | 3792 | return NULL; |
753ee728 MH |
3793 | } |
3794 | ||
9af744d7 | 3795 | static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask) |
a238ab5b | 3796 | { |
a238ab5b | 3797 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
a238ab5b DH |
3798 | |
3799 | /* | |
3800 | * This documents exceptions given to allocations in certain | |
3801 | * contexts that are allowed to allocate outside current's set | |
3802 | * of allowed nodes. | |
3803 | */ | |
3804 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
cd04ae1e | 3805 | if (tsk_is_oom_victim(current) || |
a238ab5b DH |
3806 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3807 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 3808 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
3809 | filter &= ~SHOW_MEM_FILTER_NODES; |
3810 | ||
9af744d7 | 3811 | show_mem(filter, nodemask); |
aa187507 MH |
3812 | } |
3813 | ||
a8e99259 | 3814 | void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) |
aa187507 MH |
3815 | { |
3816 | struct va_format vaf; | |
3817 | va_list args; | |
1be334e5 | 3818 | static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1); |
aa187507 | 3819 | |
0f7896f1 | 3820 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs)) |
aa187507 MH |
3821 | return; |
3822 | ||
7877cdcc MH |
3823 | va_start(args, fmt); |
3824 | vaf.fmt = fmt; | |
3825 | vaf.va = &args; | |
ef8444ea | 3826 | pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl", |
0205f755 MH |
3827 | current->comm, &vaf, gfp_mask, &gfp_mask, |
3828 | nodemask_pr_args(nodemask)); | |
7877cdcc | 3829 | va_end(args); |
3ee9a4f0 | 3830 | |
a8e99259 | 3831 | cpuset_print_current_mems_allowed(); |
ef8444ea | 3832 | pr_cont("\n"); |
a238ab5b | 3833 | dump_stack(); |
685dbf6f | 3834 | warn_alloc_show_mem(gfp_mask, nodemask); |
a238ab5b DH |
3835 | } |
3836 | ||
6c18ba7a MH |
3837 | static inline struct page * |
3838 | __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order, | |
3839 | unsigned int alloc_flags, | |
3840 | const struct alloc_context *ac) | |
3841 | { | |
3842 | struct page *page; | |
3843 | ||
3844 | page = get_page_from_freelist(gfp_mask, order, | |
3845 | alloc_flags|ALLOC_CPUSET, ac); | |
3846 | /* | |
3847 | * fallback to ignore cpuset restriction if our nodes | |
3848 | * are depleted | |
3849 | */ | |
3850 | if (!page) | |
3851 | page = get_page_from_freelist(gfp_mask, order, | |
3852 | alloc_flags, ac); | |
3853 | ||
3854 | return page; | |
3855 | } | |
3856 | ||
11e33f6a MG |
3857 | static inline struct page * |
3858 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3859 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 3860 | { |
6e0fc46d DR |
3861 | struct oom_control oc = { |
3862 | .zonelist = ac->zonelist, | |
3863 | .nodemask = ac->nodemask, | |
2a966b77 | 3864 | .memcg = NULL, |
6e0fc46d DR |
3865 | .gfp_mask = gfp_mask, |
3866 | .order = order, | |
6e0fc46d | 3867 | }; |
11e33f6a MG |
3868 | struct page *page; |
3869 | ||
9879de73 JW |
3870 | *did_some_progress = 0; |
3871 | ||
9879de73 | 3872 | /* |
dc56401f JW |
3873 | * Acquire the oom lock. If that fails, somebody else is |
3874 | * making progress for us. | |
9879de73 | 3875 | */ |
dc56401f | 3876 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 3877 | *did_some_progress = 1; |
11e33f6a | 3878 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
3879 | return NULL; |
3880 | } | |
6b1de916 | 3881 | |
11e33f6a MG |
3882 | /* |
3883 | * Go through the zonelist yet one more time, keep very high watermark | |
3884 | * here, this is only to catch a parallel oom killing, we must fail if | |
e746bf73 TH |
3885 | * we're still under heavy pressure. But make sure that this reclaim |
3886 | * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY | |
3887 | * allocation which will never fail due to oom_lock already held. | |
11e33f6a | 3888 | */ |
e746bf73 TH |
3889 | page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) & |
3890 | ~__GFP_DIRECT_RECLAIM, order, | |
3891 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 3892 | if (page) |
11e33f6a MG |
3893 | goto out; |
3894 | ||
06ad276a MH |
3895 | /* Coredumps can quickly deplete all memory reserves */ |
3896 | if (current->flags & PF_DUMPCORE) | |
3897 | goto out; | |
3898 | /* The OOM killer will not help higher order allocs */ | |
3899 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
3900 | goto out; | |
dcda9b04 MH |
3901 | /* |
3902 | * We have already exhausted all our reclaim opportunities without any | |
3903 | * success so it is time to admit defeat. We will skip the OOM killer | |
3904 | * because it is very likely that the caller has a more reasonable | |
3905 | * fallback than shooting a random task. | |
3906 | */ | |
3907 | if (gfp_mask & __GFP_RETRY_MAYFAIL) | |
3908 | goto out; | |
06ad276a MH |
3909 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
3910 | if (ac->high_zoneidx < ZONE_NORMAL) | |
3911 | goto out; | |
3912 | if (pm_suspended_storage()) | |
3913 | goto out; | |
3914 | /* | |
3915 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
3916 | * other request to make a forward progress. | |
3917 | * We are in an unfortunate situation where out_of_memory cannot | |
3918 | * do much for this context but let's try it to at least get | |
3919 | * access to memory reserved if the current task is killed (see | |
3920 | * out_of_memory). Once filesystems are ready to handle allocation | |
3921 | * failures more gracefully we should just bail out here. | |
3922 | */ | |
3923 | ||
3924 | /* The OOM killer may not free memory on a specific node */ | |
3925 | if (gfp_mask & __GFP_THISNODE) | |
3926 | goto out; | |
3da88fb3 | 3927 | |
3c2c6488 | 3928 | /* Exhausted what can be done so it's blame time */ |
5020e285 | 3929 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 3930 | *did_some_progress = 1; |
5020e285 | 3931 | |
6c18ba7a MH |
3932 | /* |
3933 | * Help non-failing allocations by giving them access to memory | |
3934 | * reserves | |
3935 | */ | |
3936 | if (gfp_mask & __GFP_NOFAIL) | |
3937 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, | |
5020e285 | 3938 | ALLOC_NO_WATERMARKS, ac); |
5020e285 | 3939 | } |
11e33f6a | 3940 | out: |
dc56401f | 3941 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3942 | return page; |
3943 | } | |
3944 | ||
33c2d214 MH |
3945 | /* |
3946 | * Maximum number of compaction retries wit a progress before OOM | |
3947 | * killer is consider as the only way to move forward. | |
3948 | */ | |
3949 | #define MAX_COMPACT_RETRIES 16 | |
3950 | ||
56de7263 MG |
3951 | #ifdef CONFIG_COMPACTION |
3952 | /* Try memory compaction for high-order allocations before reclaim */ | |
3953 | static struct page * | |
3954 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3955 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 3956 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 3957 | { |
5e1f0f09 | 3958 | struct page *page = NULL; |
eb414681 | 3959 | unsigned long pflags; |
499118e9 | 3960 | unsigned int noreclaim_flag; |
53853e2d VB |
3961 | |
3962 | if (!order) | |
66199712 | 3963 | return NULL; |
66199712 | 3964 | |
eb414681 | 3965 | psi_memstall_enter(&pflags); |
499118e9 | 3966 | noreclaim_flag = memalloc_noreclaim_save(); |
eb414681 | 3967 | |
c5d01d0d | 3968 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
5e1f0f09 | 3969 | prio, &page); |
eb414681 | 3970 | |
499118e9 | 3971 | memalloc_noreclaim_restore(noreclaim_flag); |
eb414681 | 3972 | psi_memstall_leave(&pflags); |
56de7263 | 3973 | |
98dd3b48 VB |
3974 | /* |
3975 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3976 | * count a compaction stall | |
3977 | */ | |
3978 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3979 | |
5e1f0f09 MG |
3980 | /* Prep a captured page if available */ |
3981 | if (page) | |
3982 | prep_new_page(page, order, gfp_mask, alloc_flags); | |
3983 | ||
3984 | /* Try get a page from the freelist if available */ | |
3985 | if (!page) | |
3986 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
53853e2d | 3987 | |
98dd3b48 VB |
3988 | if (page) { |
3989 | struct zone *zone = page_zone(page); | |
53853e2d | 3990 | |
98dd3b48 VB |
3991 | zone->compact_blockskip_flush = false; |
3992 | compaction_defer_reset(zone, order, true); | |
3993 | count_vm_event(COMPACTSUCCESS); | |
3994 | return page; | |
3995 | } | |
56de7263 | 3996 | |
98dd3b48 VB |
3997 | /* |
3998 | * It's bad if compaction run occurs and fails. The most likely reason | |
3999 | * is that pages exist, but not enough to satisfy watermarks. | |
4000 | */ | |
4001 | count_vm_event(COMPACTFAIL); | |
66199712 | 4002 | |
98dd3b48 | 4003 | cond_resched(); |
56de7263 MG |
4004 | |
4005 | return NULL; | |
4006 | } | |
33c2d214 | 4007 | |
3250845d VB |
4008 | static inline bool |
4009 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, | |
4010 | enum compact_result compact_result, | |
4011 | enum compact_priority *compact_priority, | |
d9436498 | 4012 | int *compaction_retries) |
3250845d VB |
4013 | { |
4014 | int max_retries = MAX_COMPACT_RETRIES; | |
c2033b00 | 4015 | int min_priority; |
65190cff MH |
4016 | bool ret = false; |
4017 | int retries = *compaction_retries; | |
4018 | enum compact_priority priority = *compact_priority; | |
3250845d VB |
4019 | |
4020 | if (!order) | |
4021 | return false; | |
4022 | ||
d9436498 VB |
4023 | if (compaction_made_progress(compact_result)) |
4024 | (*compaction_retries)++; | |
4025 | ||
3250845d VB |
4026 | /* |
4027 | * compaction considers all the zone as desperately out of memory | |
4028 | * so it doesn't really make much sense to retry except when the | |
4029 | * failure could be caused by insufficient priority | |
4030 | */ | |
d9436498 VB |
4031 | if (compaction_failed(compact_result)) |
4032 | goto check_priority; | |
3250845d | 4033 | |
49433085 VB |
4034 | /* |
4035 | * compaction was skipped because there are not enough order-0 pages | |
4036 | * to work with, so we retry only if it looks like reclaim can help. | |
4037 | */ | |
4038 | if (compaction_needs_reclaim(compact_result)) { | |
4039 | ret = compaction_zonelist_suitable(ac, order, alloc_flags); | |
4040 | goto out; | |
4041 | } | |
4042 | ||
3250845d VB |
4043 | /* |
4044 | * make sure the compaction wasn't deferred or didn't bail out early | |
4045 | * due to locks contention before we declare that we should give up. | |
49433085 VB |
4046 | * But the next retry should use a higher priority if allowed, so |
4047 | * we don't just keep bailing out endlessly. | |
3250845d | 4048 | */ |
65190cff | 4049 | if (compaction_withdrawn(compact_result)) { |
49433085 | 4050 | goto check_priority; |
65190cff | 4051 | } |
3250845d VB |
4052 | |
4053 | /* | |
dcda9b04 | 4054 | * !costly requests are much more important than __GFP_RETRY_MAYFAIL |
3250845d VB |
4055 | * costly ones because they are de facto nofail and invoke OOM |
4056 | * killer to move on while costly can fail and users are ready | |
4057 | * to cope with that. 1/4 retries is rather arbitrary but we | |
4058 | * would need much more detailed feedback from compaction to | |
4059 | * make a better decision. | |
4060 | */ | |
4061 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
4062 | max_retries /= 4; | |
65190cff MH |
4063 | if (*compaction_retries <= max_retries) { |
4064 | ret = true; | |
4065 | goto out; | |
4066 | } | |
3250845d | 4067 | |
d9436498 VB |
4068 | /* |
4069 | * Make sure there are attempts at the highest priority if we exhausted | |
4070 | * all retries or failed at the lower priorities. | |
4071 | */ | |
4072 | check_priority: | |
c2033b00 VB |
4073 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
4074 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; | |
65190cff | 4075 | |
c2033b00 | 4076 | if (*compact_priority > min_priority) { |
d9436498 VB |
4077 | (*compact_priority)--; |
4078 | *compaction_retries = 0; | |
65190cff | 4079 | ret = true; |
d9436498 | 4080 | } |
65190cff MH |
4081 | out: |
4082 | trace_compact_retry(order, priority, compact_result, retries, max_retries, ret); | |
4083 | return ret; | |
3250845d | 4084 | } |
56de7263 MG |
4085 | #else |
4086 | static inline struct page * | |
4087 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4088 | unsigned int alloc_flags, const struct alloc_context *ac, |
a5508cd8 | 4089 | enum compact_priority prio, enum compact_result *compact_result) |
56de7263 | 4090 | { |
33c2d214 | 4091 | *compact_result = COMPACT_SKIPPED; |
56de7263 MG |
4092 | return NULL; |
4093 | } | |
33c2d214 MH |
4094 | |
4095 | static inline bool | |
86a294a8 MH |
4096 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
4097 | enum compact_result compact_result, | |
a5508cd8 | 4098 | enum compact_priority *compact_priority, |
d9436498 | 4099 | int *compaction_retries) |
33c2d214 | 4100 | { |
31e49bfd MH |
4101 | struct zone *zone; |
4102 | struct zoneref *z; | |
4103 | ||
4104 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) | |
4105 | return false; | |
4106 | ||
4107 | /* | |
4108 | * There are setups with compaction disabled which would prefer to loop | |
4109 | * inside the allocator rather than hit the oom killer prematurely. | |
4110 | * Let's give them a good hope and keep retrying while the order-0 | |
4111 | * watermarks are OK. | |
4112 | */ | |
4113 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
4114 | ac->nodemask) { | |
4115 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), | |
4116 | ac_classzone_idx(ac), alloc_flags)) | |
4117 | return true; | |
4118 | } | |
33c2d214 MH |
4119 | return false; |
4120 | } | |
3250845d | 4121 | #endif /* CONFIG_COMPACTION */ |
56de7263 | 4122 | |
d92a8cfc | 4123 | #ifdef CONFIG_LOCKDEP |
93781325 | 4124 | static struct lockdep_map __fs_reclaim_map = |
d92a8cfc PZ |
4125 | STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map); |
4126 | ||
4127 | static bool __need_fs_reclaim(gfp_t gfp_mask) | |
4128 | { | |
4129 | gfp_mask = current_gfp_context(gfp_mask); | |
4130 | ||
4131 | /* no reclaim without waiting on it */ | |
4132 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) | |
4133 | return false; | |
4134 | ||
4135 | /* this guy won't enter reclaim */ | |
2e517d68 | 4136 | if (current->flags & PF_MEMALLOC) |
d92a8cfc PZ |
4137 | return false; |
4138 | ||
4139 | /* We're only interested __GFP_FS allocations for now */ | |
4140 | if (!(gfp_mask & __GFP_FS)) | |
4141 | return false; | |
4142 | ||
4143 | if (gfp_mask & __GFP_NOLOCKDEP) | |
4144 | return false; | |
4145 | ||
4146 | return true; | |
4147 | } | |
4148 | ||
93781325 OS |
4149 | void __fs_reclaim_acquire(void) |
4150 | { | |
4151 | lock_map_acquire(&__fs_reclaim_map); | |
4152 | } | |
4153 | ||
4154 | void __fs_reclaim_release(void) | |
4155 | { | |
4156 | lock_map_release(&__fs_reclaim_map); | |
4157 | } | |
4158 | ||
d92a8cfc PZ |
4159 | void fs_reclaim_acquire(gfp_t gfp_mask) |
4160 | { | |
4161 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 4162 | __fs_reclaim_acquire(); |
d92a8cfc PZ |
4163 | } |
4164 | EXPORT_SYMBOL_GPL(fs_reclaim_acquire); | |
4165 | ||
4166 | void fs_reclaim_release(gfp_t gfp_mask) | |
4167 | { | |
4168 | if (__need_fs_reclaim(gfp_mask)) | |
93781325 | 4169 | __fs_reclaim_release(); |
d92a8cfc PZ |
4170 | } |
4171 | EXPORT_SYMBOL_GPL(fs_reclaim_release); | |
4172 | #endif | |
4173 | ||
bba90710 MS |
4174 | /* Perform direct synchronous page reclaim */ |
4175 | static int | |
a9263751 VB |
4176 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
4177 | const struct alloc_context *ac) | |
11e33f6a | 4178 | { |
bba90710 | 4179 | int progress; |
499118e9 | 4180 | unsigned int noreclaim_flag; |
eb414681 | 4181 | unsigned long pflags; |
11e33f6a MG |
4182 | |
4183 | cond_resched(); | |
4184 | ||
4185 | /* We now go into synchronous reclaim */ | |
4186 | cpuset_memory_pressure_bump(); | |
eb414681 | 4187 | psi_memstall_enter(&pflags); |
d92a8cfc | 4188 | fs_reclaim_acquire(gfp_mask); |
93781325 | 4189 | noreclaim_flag = memalloc_noreclaim_save(); |
11e33f6a | 4190 | |
a9263751 VB |
4191 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
4192 | ac->nodemask); | |
11e33f6a | 4193 | |
499118e9 | 4194 | memalloc_noreclaim_restore(noreclaim_flag); |
93781325 | 4195 | fs_reclaim_release(gfp_mask); |
eb414681 | 4196 | psi_memstall_leave(&pflags); |
11e33f6a MG |
4197 | |
4198 | cond_resched(); | |
4199 | ||
bba90710 MS |
4200 | return progress; |
4201 | } | |
4202 | ||
4203 | /* The really slow allocator path where we enter direct reclaim */ | |
4204 | static inline struct page * | |
4205 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 4206 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 4207 | unsigned long *did_some_progress) |
bba90710 MS |
4208 | { |
4209 | struct page *page = NULL; | |
4210 | bool drained = false; | |
4211 | ||
a9263751 | 4212 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
4213 | if (unlikely(!(*did_some_progress))) |
4214 | return NULL; | |
11e33f6a | 4215 | |
9ee493ce | 4216 | retry: |
31a6c190 | 4217 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
9ee493ce MG |
4218 | |
4219 | /* | |
4220 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
4221 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
4222 | * Shrink them them and try again | |
9ee493ce MG |
4223 | */ |
4224 | if (!page && !drained) { | |
29fac03b | 4225 | unreserve_highatomic_pageblock(ac, false); |
93481ff0 | 4226 | drain_all_pages(NULL); |
9ee493ce MG |
4227 | drained = true; |
4228 | goto retry; | |
4229 | } | |
4230 | ||
11e33f6a MG |
4231 | return page; |
4232 | } | |
4233 | ||
5ecd9d40 DR |
4234 | static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask, |
4235 | const struct alloc_context *ac) | |
3a025760 JW |
4236 | { |
4237 | struct zoneref *z; | |
4238 | struct zone *zone; | |
e1a55637 | 4239 | pg_data_t *last_pgdat = NULL; |
5ecd9d40 | 4240 | enum zone_type high_zoneidx = ac->high_zoneidx; |
3a025760 | 4241 | |
5ecd9d40 DR |
4242 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx, |
4243 | ac->nodemask) { | |
e1a55637 | 4244 | if (last_pgdat != zone->zone_pgdat) |
5ecd9d40 | 4245 | wakeup_kswapd(zone, gfp_mask, order, high_zoneidx); |
e1a55637 MG |
4246 | last_pgdat = zone->zone_pgdat; |
4247 | } | |
3a025760 JW |
4248 | } |
4249 | ||
c603844b | 4250 | static inline unsigned int |
341ce06f PZ |
4251 | gfp_to_alloc_flags(gfp_t gfp_mask) |
4252 | { | |
c603844b | 4253 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 4254 | |
736838e9 MN |
4255 | /* |
4256 | * __GFP_HIGH is assumed to be the same as ALLOC_HIGH | |
4257 | * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD | |
4258 | * to save two branches. | |
4259 | */ | |
e6223a3b | 4260 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
736838e9 | 4261 | BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD); |
933e312e | 4262 | |
341ce06f PZ |
4263 | /* |
4264 | * The caller may dip into page reserves a bit more if the caller | |
4265 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4266 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 4267 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 4268 | */ |
736838e9 MN |
4269 | alloc_flags |= (__force int) |
4270 | (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)); | |
1da177e4 | 4271 | |
d0164adc | 4272 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 4273 | /* |
b104a35d DR |
4274 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
4275 | * if it can't schedule. | |
5c3240d9 | 4276 | */ |
b104a35d | 4277 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 4278 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 4279 | /* |
b104a35d | 4280 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 4281 | * comment for __cpuset_node_allowed(). |
523b9458 | 4282 | */ |
341ce06f | 4283 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 4284 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
4285 | alloc_flags |= ALLOC_HARDER; |
4286 | ||
d883c6cf JK |
4287 | #ifdef CONFIG_CMA |
4288 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
4289 | alloc_flags |= ALLOC_CMA; | |
4290 | #endif | |
341ce06f PZ |
4291 | return alloc_flags; |
4292 | } | |
4293 | ||
cd04ae1e | 4294 | static bool oom_reserves_allowed(struct task_struct *tsk) |
072bb0aa | 4295 | { |
cd04ae1e MH |
4296 | if (!tsk_is_oom_victim(tsk)) |
4297 | return false; | |
4298 | ||
4299 | /* | |
4300 | * !MMU doesn't have oom reaper so give access to memory reserves | |
4301 | * only to the thread with TIF_MEMDIE set | |
4302 | */ | |
4303 | if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE)) | |
31a6c190 VB |
4304 | return false; |
4305 | ||
cd04ae1e MH |
4306 | return true; |
4307 | } | |
4308 | ||
4309 | /* | |
4310 | * Distinguish requests which really need access to full memory | |
4311 | * reserves from oom victims which can live with a portion of it | |
4312 | */ | |
4313 | static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask) | |
4314 | { | |
4315 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) | |
4316 | return 0; | |
31a6c190 | 4317 | if (gfp_mask & __GFP_MEMALLOC) |
cd04ae1e | 4318 | return ALLOC_NO_WATERMARKS; |
31a6c190 | 4319 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
cd04ae1e MH |
4320 | return ALLOC_NO_WATERMARKS; |
4321 | if (!in_interrupt()) { | |
4322 | if (current->flags & PF_MEMALLOC) | |
4323 | return ALLOC_NO_WATERMARKS; | |
4324 | else if (oom_reserves_allowed(current)) | |
4325 | return ALLOC_OOM; | |
4326 | } | |
31a6c190 | 4327 | |
cd04ae1e MH |
4328 | return 0; |
4329 | } | |
4330 | ||
4331 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) | |
4332 | { | |
4333 | return !!__gfp_pfmemalloc_flags(gfp_mask); | |
072bb0aa MG |
4334 | } |
4335 | ||
0a0337e0 MH |
4336 | /* |
4337 | * Checks whether it makes sense to retry the reclaim to make a forward progress | |
4338 | * for the given allocation request. | |
491d79ae JW |
4339 | * |
4340 | * We give up when we either have tried MAX_RECLAIM_RETRIES in a row | |
4341 | * without success, or when we couldn't even meet the watermark if we | |
4342 | * reclaimed all remaining pages on the LRU lists. | |
0a0337e0 MH |
4343 | * |
4344 | * Returns true if a retry is viable or false to enter the oom path. | |
4345 | */ | |
4346 | static inline bool | |
4347 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
4348 | struct alloc_context *ac, int alloc_flags, | |
423b452e | 4349 | bool did_some_progress, int *no_progress_loops) |
0a0337e0 MH |
4350 | { |
4351 | struct zone *zone; | |
4352 | struct zoneref *z; | |
15f570bf | 4353 | bool ret = false; |
0a0337e0 | 4354 | |
423b452e VB |
4355 | /* |
4356 | * Costly allocations might have made a progress but this doesn't mean | |
4357 | * their order will become available due to high fragmentation so | |
4358 | * always increment the no progress counter for them | |
4359 | */ | |
4360 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) | |
4361 | *no_progress_loops = 0; | |
4362 | else | |
4363 | (*no_progress_loops)++; | |
4364 | ||
0a0337e0 MH |
4365 | /* |
4366 | * Make sure we converge to OOM if we cannot make any progress | |
4367 | * several times in the row. | |
4368 | */ | |
04c8716f MK |
4369 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) { |
4370 | /* Before OOM, exhaust highatomic_reserve */ | |
29fac03b | 4371 | return unreserve_highatomic_pageblock(ac, true); |
04c8716f | 4372 | } |
0a0337e0 | 4373 | |
bca67592 MG |
4374 | /* |
4375 | * Keep reclaiming pages while there is a chance this will lead | |
4376 | * somewhere. If none of the target zones can satisfy our allocation | |
4377 | * request even if all reclaimable pages are considered then we are | |
4378 | * screwed and have to go OOM. | |
0a0337e0 MH |
4379 | */ |
4380 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
4381 | ac->nodemask) { | |
4382 | unsigned long available; | |
ede37713 | 4383 | unsigned long reclaimable; |
d379f01d MH |
4384 | unsigned long min_wmark = min_wmark_pages(zone); |
4385 | bool wmark; | |
0a0337e0 | 4386 | |
5a1c84b4 | 4387 | available = reclaimable = zone_reclaimable_pages(zone); |
5a1c84b4 | 4388 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
0a0337e0 MH |
4389 | |
4390 | /* | |
491d79ae JW |
4391 | * Would the allocation succeed if we reclaimed all |
4392 | * reclaimable pages? | |
0a0337e0 | 4393 | */ |
d379f01d MH |
4394 | wmark = __zone_watermark_ok(zone, order, min_wmark, |
4395 | ac_classzone_idx(ac), alloc_flags, available); | |
4396 | trace_reclaim_retry_zone(z, order, reclaimable, | |
4397 | available, min_wmark, *no_progress_loops, wmark); | |
4398 | if (wmark) { | |
ede37713 MH |
4399 | /* |
4400 | * If we didn't make any progress and have a lot of | |
4401 | * dirty + writeback pages then we should wait for | |
4402 | * an IO to complete to slow down the reclaim and | |
4403 | * prevent from pre mature OOM | |
4404 | */ | |
4405 | if (!did_some_progress) { | |
11fb9989 | 4406 | unsigned long write_pending; |
ede37713 | 4407 | |
5a1c84b4 MG |
4408 | write_pending = zone_page_state_snapshot(zone, |
4409 | NR_ZONE_WRITE_PENDING); | |
ede37713 | 4410 | |
11fb9989 | 4411 | if (2 * write_pending > reclaimable) { |
ede37713 MH |
4412 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
4413 | return true; | |
4414 | } | |
4415 | } | |
5a1c84b4 | 4416 | |
15f570bf MH |
4417 | ret = true; |
4418 | goto out; | |
0a0337e0 MH |
4419 | } |
4420 | } | |
4421 | ||
15f570bf MH |
4422 | out: |
4423 | /* | |
4424 | * Memory allocation/reclaim might be called from a WQ context and the | |
4425 | * current implementation of the WQ concurrency control doesn't | |
4426 | * recognize that a particular WQ is congested if the worker thread is | |
4427 | * looping without ever sleeping. Therefore we have to do a short sleep | |
4428 | * here rather than calling cond_resched(). | |
4429 | */ | |
4430 | if (current->flags & PF_WQ_WORKER) | |
4431 | schedule_timeout_uninterruptible(1); | |
4432 | else | |
4433 | cond_resched(); | |
4434 | return ret; | |
0a0337e0 MH |
4435 | } |
4436 | ||
902b6281 VB |
4437 | static inline bool |
4438 | check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac) | |
4439 | { | |
4440 | /* | |
4441 | * It's possible that cpuset's mems_allowed and the nodemask from | |
4442 | * mempolicy don't intersect. This should be normally dealt with by | |
4443 | * policy_nodemask(), but it's possible to race with cpuset update in | |
4444 | * such a way the check therein was true, and then it became false | |
4445 | * before we got our cpuset_mems_cookie here. | |
4446 | * This assumes that for all allocations, ac->nodemask can come only | |
4447 | * from MPOL_BIND mempolicy (whose documented semantics is to be ignored | |
4448 | * when it does not intersect with the cpuset restrictions) or the | |
4449 | * caller can deal with a violated nodemask. | |
4450 | */ | |
4451 | if (cpusets_enabled() && ac->nodemask && | |
4452 | !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { | |
4453 | ac->nodemask = NULL; | |
4454 | return true; | |
4455 | } | |
4456 | ||
4457 | /* | |
4458 | * When updating a task's mems_allowed or mempolicy nodemask, it is | |
4459 | * possible to race with parallel threads in such a way that our | |
4460 | * allocation can fail while the mask is being updated. If we are about | |
4461 | * to fail, check if the cpuset changed during allocation and if so, | |
4462 | * retry. | |
4463 | */ | |
4464 | if (read_mems_allowed_retry(cpuset_mems_cookie)) | |
4465 | return true; | |
4466 | ||
4467 | return false; | |
4468 | } | |
4469 | ||
11e33f6a MG |
4470 | static inline struct page * |
4471 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 4472 | struct alloc_context *ac) |
11e33f6a | 4473 | { |
d0164adc | 4474 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
282722b0 | 4475 | const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER; |
11e33f6a | 4476 | struct page *page = NULL; |
c603844b | 4477 | unsigned int alloc_flags; |
11e33f6a | 4478 | unsigned long did_some_progress; |
5ce9bfef | 4479 | enum compact_priority compact_priority; |
c5d01d0d | 4480 | enum compact_result compact_result; |
5ce9bfef VB |
4481 | int compaction_retries; |
4482 | int no_progress_loops; | |
5ce9bfef | 4483 | unsigned int cpuset_mems_cookie; |
cd04ae1e | 4484 | int reserve_flags; |
1da177e4 | 4485 | |
d0164adc MG |
4486 | /* |
4487 | * We also sanity check to catch abuse of atomic reserves being used by | |
4488 | * callers that are not in atomic context. | |
4489 | */ | |
4490 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
4491 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
4492 | gfp_mask &= ~__GFP_ATOMIC; | |
4493 | ||
5ce9bfef VB |
4494 | retry_cpuset: |
4495 | compaction_retries = 0; | |
4496 | no_progress_loops = 0; | |
4497 | compact_priority = DEF_COMPACT_PRIORITY; | |
4498 | cpuset_mems_cookie = read_mems_allowed_begin(); | |
9a67f648 MH |
4499 | |
4500 | /* | |
4501 | * The fast path uses conservative alloc_flags to succeed only until | |
4502 | * kswapd needs to be woken up, and to avoid the cost of setting up | |
4503 | * alloc_flags precisely. So we do that now. | |
4504 | */ | |
4505 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
4506 | ||
e47483bc VB |
4507 | /* |
4508 | * We need to recalculate the starting point for the zonelist iterator | |
4509 | * because we might have used different nodemask in the fast path, or | |
4510 | * there was a cpuset modification and we are retrying - otherwise we | |
4511 | * could end up iterating over non-eligible zones endlessly. | |
4512 | */ | |
4513 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, | |
4514 | ac->high_zoneidx, ac->nodemask); | |
4515 | if (!ac->preferred_zoneref->zone) | |
4516 | goto nopage; | |
4517 | ||
0a79cdad | 4518 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4519 | wake_all_kswapds(order, gfp_mask, ac); |
23771235 VB |
4520 | |
4521 | /* | |
4522 | * The adjusted alloc_flags might result in immediate success, so try | |
4523 | * that first | |
4524 | */ | |
4525 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); | |
4526 | if (page) | |
4527 | goto got_pg; | |
4528 | ||
a8161d1e VB |
4529 | /* |
4530 | * For costly allocations, try direct compaction first, as it's likely | |
282722b0 VB |
4531 | * that we have enough base pages and don't need to reclaim. For non- |
4532 | * movable high-order allocations, do that as well, as compaction will | |
4533 | * try prevent permanent fragmentation by migrating from blocks of the | |
4534 | * same migratetype. | |
4535 | * Don't try this for allocations that are allowed to ignore | |
4536 | * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen. | |
a8161d1e | 4537 | */ |
282722b0 VB |
4538 | if (can_direct_reclaim && |
4539 | (costly_order || | |
4540 | (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) | |
4541 | && !gfp_pfmemalloc_allowed(gfp_mask)) { | |
a8161d1e VB |
4542 | page = __alloc_pages_direct_compact(gfp_mask, order, |
4543 | alloc_flags, ac, | |
a5508cd8 | 4544 | INIT_COMPACT_PRIORITY, |
a8161d1e VB |
4545 | &compact_result); |
4546 | if (page) | |
4547 | goto got_pg; | |
4548 | ||
cc638f32 VB |
4549 | /* |
4550 | * Checks for costly allocations with __GFP_NORETRY, which | |
4551 | * includes some THP page fault allocations | |
4552 | */ | |
4553 | if (costly_order && (gfp_mask & __GFP_NORETRY)) { | |
b39d0ee2 DR |
4554 | /* |
4555 | * If allocating entire pageblock(s) and compaction | |
4556 | * failed because all zones are below low watermarks | |
4557 | * or is prohibited because it recently failed at this | |
3f36d866 DR |
4558 | * order, fail immediately unless the allocator has |
4559 | * requested compaction and reclaim retry. | |
b39d0ee2 DR |
4560 | * |
4561 | * Reclaim is | |
4562 | * - potentially very expensive because zones are far | |
4563 | * below their low watermarks or this is part of very | |
4564 | * bursty high order allocations, | |
4565 | * - not guaranteed to help because isolate_freepages() | |
4566 | * may not iterate over freed pages as part of its | |
4567 | * linear scan, and | |
4568 | * - unlikely to make entire pageblocks free on its | |
4569 | * own. | |
4570 | */ | |
4571 | if (compact_result == COMPACT_SKIPPED || | |
4572 | compact_result == COMPACT_DEFERRED) | |
4573 | goto nopage; | |
a8161d1e | 4574 | |
a8161d1e | 4575 | /* |
3eb2771b VB |
4576 | * Looks like reclaim/compaction is worth trying, but |
4577 | * sync compaction could be very expensive, so keep | |
25160354 | 4578 | * using async compaction. |
a8161d1e | 4579 | */ |
a5508cd8 | 4580 | compact_priority = INIT_COMPACT_PRIORITY; |
a8161d1e VB |
4581 | } |
4582 | } | |
23771235 | 4583 | |
31a6c190 | 4584 | retry: |
23771235 | 4585 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
0a79cdad | 4586 | if (alloc_flags & ALLOC_KSWAPD) |
5ecd9d40 | 4587 | wake_all_kswapds(order, gfp_mask, ac); |
31a6c190 | 4588 | |
cd04ae1e MH |
4589 | reserve_flags = __gfp_pfmemalloc_flags(gfp_mask); |
4590 | if (reserve_flags) | |
4591 | alloc_flags = reserve_flags; | |
23771235 | 4592 | |
e46e7b77 | 4593 | /* |
d6a24df0 VB |
4594 | * Reset the nodemask and zonelist iterators if memory policies can be |
4595 | * ignored. These allocations are high priority and system rather than | |
4596 | * user oriented. | |
e46e7b77 | 4597 | */ |
cd04ae1e | 4598 | if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) { |
d6a24df0 | 4599 | ac->nodemask = NULL; |
e46e7b77 MG |
4600 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4601 | ac->high_zoneidx, ac->nodemask); | |
4602 | } | |
4603 | ||
23771235 | 4604 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
31a6c190 | 4605 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
7fb1d9fc RS |
4606 | if (page) |
4607 | goto got_pg; | |
1da177e4 | 4608 | |
d0164adc | 4609 | /* Caller is not willing to reclaim, we can't balance anything */ |
9a67f648 | 4610 | if (!can_direct_reclaim) |
1da177e4 LT |
4611 | goto nopage; |
4612 | ||
9a67f648 MH |
4613 | /* Avoid recursion of direct reclaim */ |
4614 | if (current->flags & PF_MEMALLOC) | |
6583bb64 DR |
4615 | goto nopage; |
4616 | ||
a8161d1e VB |
4617 | /* Try direct reclaim and then allocating */ |
4618 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, | |
4619 | &did_some_progress); | |
4620 | if (page) | |
4621 | goto got_pg; | |
4622 | ||
4623 | /* Try direct compaction and then allocating */ | |
a9263751 | 4624 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
a5508cd8 | 4625 | compact_priority, &compact_result); |
56de7263 MG |
4626 | if (page) |
4627 | goto got_pg; | |
75f30861 | 4628 | |
9083905a JW |
4629 | /* Do not loop if specifically requested */ |
4630 | if (gfp_mask & __GFP_NORETRY) | |
a8161d1e | 4631 | goto nopage; |
9083905a | 4632 | |
0a0337e0 MH |
4633 | /* |
4634 | * Do not retry costly high order allocations unless they are | |
dcda9b04 | 4635 | * __GFP_RETRY_MAYFAIL |
0a0337e0 | 4636 | */ |
dcda9b04 | 4637 | if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL)) |
a8161d1e | 4638 | goto nopage; |
0a0337e0 | 4639 | |
0a0337e0 | 4640 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
423b452e | 4641 | did_some_progress > 0, &no_progress_loops)) |
0a0337e0 MH |
4642 | goto retry; |
4643 | ||
33c2d214 MH |
4644 | /* |
4645 | * It doesn't make any sense to retry for the compaction if the order-0 | |
4646 | * reclaim is not able to make any progress because the current | |
4647 | * implementation of the compaction depends on the sufficient amount | |
4648 | * of free memory (see __compaction_suitable) | |
4649 | */ | |
4650 | if (did_some_progress > 0 && | |
86a294a8 | 4651 | should_compact_retry(ac, order, alloc_flags, |
a5508cd8 | 4652 | compact_result, &compact_priority, |
d9436498 | 4653 | &compaction_retries)) |
33c2d214 MH |
4654 | goto retry; |
4655 | ||
902b6281 VB |
4656 | |
4657 | /* Deal with possible cpuset update races before we start OOM killing */ | |
4658 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
e47483bc VB |
4659 | goto retry_cpuset; |
4660 | ||
9083905a JW |
4661 | /* Reclaim has failed us, start killing things */ |
4662 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
4663 | if (page) | |
4664 | goto got_pg; | |
4665 | ||
9a67f648 | 4666 | /* Avoid allocations with no watermarks from looping endlessly */ |
cd04ae1e MH |
4667 | if (tsk_is_oom_victim(current) && |
4668 | (alloc_flags == ALLOC_OOM || | |
c288983d | 4669 | (gfp_mask & __GFP_NOMEMALLOC))) |
9a67f648 MH |
4670 | goto nopage; |
4671 | ||
9083905a | 4672 | /* Retry as long as the OOM killer is making progress */ |
0a0337e0 MH |
4673 | if (did_some_progress) { |
4674 | no_progress_loops = 0; | |
9083905a | 4675 | goto retry; |
0a0337e0 | 4676 | } |
9083905a | 4677 | |
1da177e4 | 4678 | nopage: |
902b6281 VB |
4679 | /* Deal with possible cpuset update races before we fail */ |
4680 | if (check_retry_cpuset(cpuset_mems_cookie, ac)) | |
5ce9bfef VB |
4681 | goto retry_cpuset; |
4682 | ||
9a67f648 MH |
4683 | /* |
4684 | * Make sure that __GFP_NOFAIL request doesn't leak out and make sure | |
4685 | * we always retry | |
4686 | */ | |
4687 | if (gfp_mask & __GFP_NOFAIL) { | |
4688 | /* | |
4689 | * All existing users of the __GFP_NOFAIL are blockable, so warn | |
4690 | * of any new users that actually require GFP_NOWAIT | |
4691 | */ | |
4692 | if (WARN_ON_ONCE(!can_direct_reclaim)) | |
4693 | goto fail; | |
4694 | ||
4695 | /* | |
4696 | * PF_MEMALLOC request from this context is rather bizarre | |
4697 | * because we cannot reclaim anything and only can loop waiting | |
4698 | * for somebody to do a work for us | |
4699 | */ | |
4700 | WARN_ON_ONCE(current->flags & PF_MEMALLOC); | |
4701 | ||
4702 | /* | |
4703 | * non failing costly orders are a hard requirement which we | |
4704 | * are not prepared for much so let's warn about these users | |
4705 | * so that we can identify them and convert them to something | |
4706 | * else. | |
4707 | */ | |
4708 | WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER); | |
4709 | ||
6c18ba7a MH |
4710 | /* |
4711 | * Help non-failing allocations by giving them access to memory | |
4712 | * reserves but do not use ALLOC_NO_WATERMARKS because this | |
4713 | * could deplete whole memory reserves which would just make | |
4714 | * the situation worse | |
4715 | */ | |
4716 | page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac); | |
4717 | if (page) | |
4718 | goto got_pg; | |
4719 | ||
9a67f648 MH |
4720 | cond_resched(); |
4721 | goto retry; | |
4722 | } | |
4723 | fail: | |
a8e99259 | 4724 | warn_alloc(gfp_mask, ac->nodemask, |
7877cdcc | 4725 | "page allocation failure: order:%u", order); |
1da177e4 | 4726 | got_pg: |
072bb0aa | 4727 | return page; |
1da177e4 | 4728 | } |
11e33f6a | 4729 | |
9cd75558 | 4730 | static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order, |
04ec6264 | 4731 | int preferred_nid, nodemask_t *nodemask, |
9cd75558 MG |
4732 | struct alloc_context *ac, gfp_t *alloc_mask, |
4733 | unsigned int *alloc_flags) | |
11e33f6a | 4734 | { |
9cd75558 | 4735 | ac->high_zoneidx = gfp_zone(gfp_mask); |
04ec6264 | 4736 | ac->zonelist = node_zonelist(preferred_nid, gfp_mask); |
9cd75558 MG |
4737 | ac->nodemask = nodemask; |
4738 | ac->migratetype = gfpflags_to_migratetype(gfp_mask); | |
11e33f6a | 4739 | |
682a3385 | 4740 | if (cpusets_enabled()) { |
9cd75558 | 4741 | *alloc_mask |= __GFP_HARDWALL; |
9cd75558 MG |
4742 | if (!ac->nodemask) |
4743 | ac->nodemask = &cpuset_current_mems_allowed; | |
51047820 VB |
4744 | else |
4745 | *alloc_flags |= ALLOC_CPUSET; | |
682a3385 MG |
4746 | } |
4747 | ||
d92a8cfc PZ |
4748 | fs_reclaim_acquire(gfp_mask); |
4749 | fs_reclaim_release(gfp_mask); | |
11e33f6a | 4750 | |
d0164adc | 4751 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
4752 | |
4753 | if (should_fail_alloc_page(gfp_mask, order)) | |
9cd75558 | 4754 | return false; |
11e33f6a | 4755 | |
d883c6cf JK |
4756 | if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE) |
4757 | *alloc_flags |= ALLOC_CMA; | |
4758 | ||
9cd75558 MG |
4759 | return true; |
4760 | } | |
21bb9bd1 | 4761 | |
9cd75558 | 4762 | /* Determine whether to spread dirty pages and what the first usable zone */ |
a380b40a | 4763 | static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac) |
9cd75558 | 4764 | { |
c9ab0c4f | 4765 | /* Dirty zone balancing only done in the fast path */ |
9cd75558 | 4766 | ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
c9ab0c4f | 4767 | |
e46e7b77 MG |
4768 | /* |
4769 | * The preferred zone is used for statistics but crucially it is | |
4770 | * also used as the starting point for the zonelist iterator. It | |
4771 | * may get reset for allocations that ignore memory policies. | |
4772 | */ | |
9cd75558 MG |
4773 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
4774 | ac->high_zoneidx, ac->nodemask); | |
4775 | } | |
4776 | ||
4777 | /* | |
4778 | * This is the 'heart' of the zoned buddy allocator. | |
4779 | */ | |
4780 | struct page * | |
04ec6264 VB |
4781 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
4782 | nodemask_t *nodemask) | |
9cd75558 MG |
4783 | { |
4784 | struct page *page; | |
4785 | unsigned int alloc_flags = ALLOC_WMARK_LOW; | |
f19360f0 | 4786 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
9cd75558 MG |
4787 | struct alloc_context ac = { }; |
4788 | ||
c63ae43b MH |
4789 | /* |
4790 | * There are several places where we assume that the order value is sane | |
4791 | * so bail out early if the request is out of bound. | |
4792 | */ | |
4793 | if (unlikely(order >= MAX_ORDER)) { | |
4794 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
4795 | return NULL; | |
4796 | } | |
4797 | ||
9cd75558 | 4798 | gfp_mask &= gfp_allowed_mask; |
f19360f0 | 4799 | alloc_mask = gfp_mask; |
04ec6264 | 4800 | if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags)) |
9cd75558 MG |
4801 | return NULL; |
4802 | ||
a380b40a | 4803 | finalise_ac(gfp_mask, &ac); |
5bb1b169 | 4804 | |
6bb15450 MG |
4805 | /* |
4806 | * Forbid the first pass from falling back to types that fragment | |
4807 | * memory until all local zones are considered. | |
4808 | */ | |
0a79cdad | 4809 | alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask); |
6bb15450 | 4810 | |
5117f45d | 4811 | /* First allocation attempt */ |
a9263751 | 4812 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
4813 | if (likely(page)) |
4814 | goto out; | |
11e33f6a | 4815 | |
4fcb0971 | 4816 | /* |
7dea19f9 MH |
4817 | * Apply scoped allocation constraints. This is mainly about GFP_NOFS |
4818 | * resp. GFP_NOIO which has to be inherited for all allocation requests | |
4819 | * from a particular context which has been marked by | |
4820 | * memalloc_no{fs,io}_{save,restore}. | |
4fcb0971 | 4821 | */ |
7dea19f9 | 4822 | alloc_mask = current_gfp_context(gfp_mask); |
4fcb0971 | 4823 | ac.spread_dirty_pages = false; |
23f086f9 | 4824 | |
4741526b MG |
4825 | /* |
4826 | * Restore the original nodemask if it was potentially replaced with | |
4827 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. | |
4828 | */ | |
97ce86f9 | 4829 | ac.nodemask = nodemask; |
16096c25 | 4830 | |
4fcb0971 | 4831 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 4832 | |
4fcb0971 | 4833 | out: |
c4159a75 | 4834 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
f4b00eab | 4835 | unlikely(__memcg_kmem_charge_page(page, gfp_mask, order) != 0)) { |
c4159a75 VD |
4836 | __free_pages(page, order); |
4837 | page = NULL; | |
4949148a VD |
4838 | } |
4839 | ||
4fcb0971 MG |
4840 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
4841 | ||
11e33f6a | 4842 | return page; |
1da177e4 | 4843 | } |
d239171e | 4844 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
4845 | |
4846 | /* | |
9ea9a680 MH |
4847 | * Common helper functions. Never use with __GFP_HIGHMEM because the returned |
4848 | * address cannot represent highmem pages. Use alloc_pages and then kmap if | |
4849 | * you need to access high mem. | |
1da177e4 | 4850 | */ |
920c7a5d | 4851 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 4852 | { |
945a1113 AM |
4853 | struct page *page; |
4854 | ||
9ea9a680 | 4855 | page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order); |
1da177e4 LT |
4856 | if (!page) |
4857 | return 0; | |
4858 | return (unsigned long) page_address(page); | |
4859 | } | |
1da177e4 LT |
4860 | EXPORT_SYMBOL(__get_free_pages); |
4861 | ||
920c7a5d | 4862 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 4863 | { |
945a1113 | 4864 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 4865 | } |
1da177e4 LT |
4866 | EXPORT_SYMBOL(get_zeroed_page); |
4867 | ||
742aa7fb | 4868 | static inline void free_the_page(struct page *page, unsigned int order) |
1da177e4 | 4869 | { |
742aa7fb AL |
4870 | if (order == 0) /* Via pcp? */ |
4871 | free_unref_page(page); | |
4872 | else | |
4873 | __free_pages_ok(page, order); | |
1da177e4 LT |
4874 | } |
4875 | ||
742aa7fb AL |
4876 | void __free_pages(struct page *page, unsigned int order) |
4877 | { | |
4878 | if (put_page_testzero(page)) | |
4879 | free_the_page(page, order); | |
4880 | } | |
1da177e4 LT |
4881 | EXPORT_SYMBOL(__free_pages); |
4882 | ||
920c7a5d | 4883 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
4884 | { |
4885 | if (addr != 0) { | |
725d704e | 4886 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
4887 | __free_pages(virt_to_page((void *)addr), order); |
4888 | } | |
4889 | } | |
4890 | ||
4891 | EXPORT_SYMBOL(free_pages); | |
4892 | ||
b63ae8ca AD |
4893 | /* |
4894 | * Page Fragment: | |
4895 | * An arbitrary-length arbitrary-offset area of memory which resides | |
4896 | * within a 0 or higher order page. Multiple fragments within that page | |
4897 | * are individually refcounted, in the page's reference counter. | |
4898 | * | |
4899 | * The page_frag functions below provide a simple allocation framework for | |
4900 | * page fragments. This is used by the network stack and network device | |
4901 | * drivers to provide a backing region of memory for use as either an | |
4902 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
4903 | */ | |
2976db80 AD |
4904 | static struct page *__page_frag_cache_refill(struct page_frag_cache *nc, |
4905 | gfp_t gfp_mask) | |
b63ae8ca AD |
4906 | { |
4907 | struct page *page = NULL; | |
4908 | gfp_t gfp = gfp_mask; | |
4909 | ||
4910 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4911 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
4912 | __GFP_NOMEMALLOC; | |
4913 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
4914 | PAGE_FRAG_CACHE_MAX_ORDER); | |
4915 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
4916 | #endif | |
4917 | if (unlikely(!page)) | |
4918 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
4919 | ||
4920 | nc->va = page ? page_address(page) : NULL; | |
4921 | ||
4922 | return page; | |
4923 | } | |
4924 | ||
2976db80 | 4925 | void __page_frag_cache_drain(struct page *page, unsigned int count) |
44fdffd7 AD |
4926 | { |
4927 | VM_BUG_ON_PAGE(page_ref_count(page) == 0, page); | |
4928 | ||
742aa7fb AL |
4929 | if (page_ref_sub_and_test(page, count)) |
4930 | free_the_page(page, compound_order(page)); | |
44fdffd7 | 4931 | } |
2976db80 | 4932 | EXPORT_SYMBOL(__page_frag_cache_drain); |
44fdffd7 | 4933 | |
8c2dd3e4 AD |
4934 | void *page_frag_alloc(struct page_frag_cache *nc, |
4935 | unsigned int fragsz, gfp_t gfp_mask) | |
b63ae8ca AD |
4936 | { |
4937 | unsigned int size = PAGE_SIZE; | |
4938 | struct page *page; | |
4939 | int offset; | |
4940 | ||
4941 | if (unlikely(!nc->va)) { | |
4942 | refill: | |
2976db80 | 4943 | page = __page_frag_cache_refill(nc, gfp_mask); |
b63ae8ca AD |
4944 | if (!page) |
4945 | return NULL; | |
4946 | ||
4947 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4948 | /* if size can vary use size else just use PAGE_SIZE */ | |
4949 | size = nc->size; | |
4950 | #endif | |
4951 | /* Even if we own the page, we do not use atomic_set(). | |
4952 | * This would break get_page_unless_zero() users. | |
4953 | */ | |
86447726 | 4954 | page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE); |
b63ae8ca AD |
4955 | |
4956 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 4957 | nc->pfmemalloc = page_is_pfmemalloc(page); |
86447726 | 4958 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
4959 | nc->offset = size; |
4960 | } | |
4961 | ||
4962 | offset = nc->offset - fragsz; | |
4963 | if (unlikely(offset < 0)) { | |
4964 | page = virt_to_page(nc->va); | |
4965 | ||
fe896d18 | 4966 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
4967 | goto refill; |
4968 | ||
4969 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
4970 | /* if size can vary use size else just use PAGE_SIZE */ | |
4971 | size = nc->size; | |
4972 | #endif | |
4973 | /* OK, page count is 0, we can safely set it */ | |
86447726 | 4974 | set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1); |
b63ae8ca AD |
4975 | |
4976 | /* reset page count bias and offset to start of new frag */ | |
86447726 | 4977 | nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; |
b63ae8ca AD |
4978 | offset = size - fragsz; |
4979 | } | |
4980 | ||
4981 | nc->pagecnt_bias--; | |
4982 | nc->offset = offset; | |
4983 | ||
4984 | return nc->va + offset; | |
4985 | } | |
8c2dd3e4 | 4986 | EXPORT_SYMBOL(page_frag_alloc); |
b63ae8ca AD |
4987 | |
4988 | /* | |
4989 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
4990 | */ | |
8c2dd3e4 | 4991 | void page_frag_free(void *addr) |
b63ae8ca AD |
4992 | { |
4993 | struct page *page = virt_to_head_page(addr); | |
4994 | ||
742aa7fb AL |
4995 | if (unlikely(put_page_testzero(page))) |
4996 | free_the_page(page, compound_order(page)); | |
b63ae8ca | 4997 | } |
8c2dd3e4 | 4998 | EXPORT_SYMBOL(page_frag_free); |
b63ae8ca | 4999 | |
d00181b9 KS |
5000 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
5001 | size_t size) | |
ee85c2e1 AK |
5002 | { |
5003 | if (addr) { | |
5004 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
5005 | unsigned long used = addr + PAGE_ALIGN(size); | |
5006 | ||
5007 | split_page(virt_to_page((void *)addr), order); | |
5008 | while (used < alloc_end) { | |
5009 | free_page(used); | |
5010 | used += PAGE_SIZE; | |
5011 | } | |
5012 | } | |
5013 | return (void *)addr; | |
5014 | } | |
5015 | ||
2be0ffe2 TT |
5016 | /** |
5017 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
5018 | * @size: the number of bytes to allocate | |
63931eb9 | 5019 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
2be0ffe2 TT |
5020 | * |
5021 | * This function is similar to alloc_pages(), except that it allocates the | |
5022 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
5023 | * allocate memory in power-of-two pages. | |
5024 | * | |
5025 | * This function is also limited by MAX_ORDER. | |
5026 | * | |
5027 | * Memory allocated by this function must be released by free_pages_exact(). | |
a862f68a MR |
5028 | * |
5029 | * Return: pointer to the allocated area or %NULL in case of error. | |
2be0ffe2 TT |
5030 | */ |
5031 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
5032 | { | |
5033 | unsigned int order = get_order(size); | |
5034 | unsigned long addr; | |
5035 | ||
63931eb9 VB |
5036 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) |
5037 | gfp_mask &= ~__GFP_COMP; | |
5038 | ||
2be0ffe2 | 5039 | addr = __get_free_pages(gfp_mask, order); |
ee85c2e1 | 5040 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
5041 | } |
5042 | EXPORT_SYMBOL(alloc_pages_exact); | |
5043 | ||
ee85c2e1 AK |
5044 | /** |
5045 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
5046 | * pages on a node. | |
b5e6ab58 | 5047 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 | 5048 | * @size: the number of bytes to allocate |
63931eb9 | 5049 | * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP |
ee85c2e1 AK |
5050 | * |
5051 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
5052 | * back. | |
a862f68a MR |
5053 | * |
5054 | * Return: pointer to the allocated area or %NULL in case of error. | |
ee85c2e1 | 5055 | */ |
e1931811 | 5056 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 5057 | { |
d00181b9 | 5058 | unsigned int order = get_order(size); |
63931eb9 VB |
5059 | struct page *p; |
5060 | ||
5061 | if (WARN_ON_ONCE(gfp_mask & __GFP_COMP)) | |
5062 | gfp_mask &= ~__GFP_COMP; | |
5063 | ||
5064 | p = alloc_pages_node(nid, gfp_mask, order); | |
ee85c2e1 AK |
5065 | if (!p) |
5066 | return NULL; | |
5067 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
5068 | } | |
ee85c2e1 | 5069 | |
2be0ffe2 TT |
5070 | /** |
5071 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
5072 | * @virt: the value returned by alloc_pages_exact. | |
5073 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
5074 | * | |
5075 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
5076 | */ | |
5077 | void free_pages_exact(void *virt, size_t size) | |
5078 | { | |
5079 | unsigned long addr = (unsigned long)virt; | |
5080 | unsigned long end = addr + PAGE_ALIGN(size); | |
5081 | ||
5082 | while (addr < end) { | |
5083 | free_page(addr); | |
5084 | addr += PAGE_SIZE; | |
5085 | } | |
5086 | } | |
5087 | EXPORT_SYMBOL(free_pages_exact); | |
5088 | ||
e0fb5815 ZY |
5089 | /** |
5090 | * nr_free_zone_pages - count number of pages beyond high watermark | |
5091 | * @offset: The zone index of the highest zone | |
5092 | * | |
a862f68a | 5093 | * nr_free_zone_pages() counts the number of pages which are beyond the |
e0fb5815 ZY |
5094 | * high watermark within all zones at or below a given zone index. For each |
5095 | * zone, the number of pages is calculated as: | |
0e056eb5 MCC |
5096 | * |
5097 | * nr_free_zone_pages = managed_pages - high_pages | |
a862f68a MR |
5098 | * |
5099 | * Return: number of pages beyond high watermark. | |
e0fb5815 | 5100 | */ |
ebec3862 | 5101 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 5102 | { |
dd1a239f | 5103 | struct zoneref *z; |
54a6eb5c MG |
5104 | struct zone *zone; |
5105 | ||
e310fd43 | 5106 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 5107 | unsigned long sum = 0; |
1da177e4 | 5108 | |
0e88460d | 5109 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 5110 | |
54a6eb5c | 5111 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
9705bea5 | 5112 | unsigned long size = zone_managed_pages(zone); |
41858966 | 5113 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
5114 | if (size > high) |
5115 | sum += size - high; | |
1da177e4 LT |
5116 | } |
5117 | ||
5118 | return sum; | |
5119 | } | |
5120 | ||
e0fb5815 ZY |
5121 | /** |
5122 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
5123 | * | |
5124 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
5125 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
a862f68a MR |
5126 | * |
5127 | * Return: number of pages beyond high watermark within ZONE_DMA and | |
5128 | * ZONE_NORMAL. | |
1da177e4 | 5129 | */ |
ebec3862 | 5130 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 5131 | { |
af4ca457 | 5132 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 5133 | } |
c2f1a551 | 5134 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 5135 | |
e0fb5815 ZY |
5136 | /** |
5137 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
5138 | * | |
5139 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
5140 | * high watermark within all zones. | |
a862f68a MR |
5141 | * |
5142 | * Return: number of pages beyond high watermark within all zones. | |
1da177e4 | 5143 | */ |
ebec3862 | 5144 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 5145 | { |
2a1e274a | 5146 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 5147 | } |
08e0f6a9 CL |
5148 | |
5149 | static inline void show_node(struct zone *zone) | |
1da177e4 | 5150 | { |
e5adfffc | 5151 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 5152 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 5153 | } |
1da177e4 | 5154 | |
d02bd27b IR |
5155 | long si_mem_available(void) |
5156 | { | |
5157 | long available; | |
5158 | unsigned long pagecache; | |
5159 | unsigned long wmark_low = 0; | |
5160 | unsigned long pages[NR_LRU_LISTS]; | |
b29940c1 | 5161 | unsigned long reclaimable; |
d02bd27b IR |
5162 | struct zone *zone; |
5163 | int lru; | |
5164 | ||
5165 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
2f95ff90 | 5166 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
d02bd27b IR |
5167 | |
5168 | for_each_zone(zone) | |
a9214443 | 5169 | wmark_low += low_wmark_pages(zone); |
d02bd27b IR |
5170 | |
5171 | /* | |
5172 | * Estimate the amount of memory available for userspace allocations, | |
5173 | * without causing swapping. | |
5174 | */ | |
c41f012a | 5175 | available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages; |
d02bd27b IR |
5176 | |
5177 | /* | |
5178 | * Not all the page cache can be freed, otherwise the system will | |
5179 | * start swapping. Assume at least half of the page cache, or the | |
5180 | * low watermark worth of cache, needs to stay. | |
5181 | */ | |
5182 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
5183 | pagecache -= min(pagecache / 2, wmark_low); | |
5184 | available += pagecache; | |
5185 | ||
5186 | /* | |
b29940c1 VB |
5187 | * Part of the reclaimable slab and other kernel memory consists of |
5188 | * items that are in use, and cannot be freed. Cap this estimate at the | |
5189 | * low watermark. | |
d02bd27b | 5190 | */ |
b29940c1 VB |
5191 | reclaimable = global_node_page_state(NR_SLAB_RECLAIMABLE) + |
5192 | global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE); | |
5193 | available += reclaimable - min(reclaimable / 2, wmark_low); | |
034ebf65 | 5194 | |
d02bd27b IR |
5195 | if (available < 0) |
5196 | available = 0; | |
5197 | return available; | |
5198 | } | |
5199 | EXPORT_SYMBOL_GPL(si_mem_available); | |
5200 | ||
1da177e4 LT |
5201 | void si_meminfo(struct sysinfo *val) |
5202 | { | |
ca79b0c2 | 5203 | val->totalram = totalram_pages(); |
11fb9989 | 5204 | val->sharedram = global_node_page_state(NR_SHMEM); |
c41f012a | 5205 | val->freeram = global_zone_page_state(NR_FREE_PAGES); |
1da177e4 | 5206 | val->bufferram = nr_blockdev_pages(); |
ca79b0c2 | 5207 | val->totalhigh = totalhigh_pages(); |
1da177e4 | 5208 | val->freehigh = nr_free_highpages(); |
1da177e4 LT |
5209 | val->mem_unit = PAGE_SIZE; |
5210 | } | |
5211 | ||
5212 | EXPORT_SYMBOL(si_meminfo); | |
5213 | ||
5214 | #ifdef CONFIG_NUMA | |
5215 | void si_meminfo_node(struct sysinfo *val, int nid) | |
5216 | { | |
cdd91a77 JL |
5217 | int zone_type; /* needs to be signed */ |
5218 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
5219 | unsigned long managed_highpages = 0; |
5220 | unsigned long free_highpages = 0; | |
1da177e4 LT |
5221 | pg_data_t *pgdat = NODE_DATA(nid); |
5222 | ||
cdd91a77 | 5223 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
9705bea5 | 5224 | managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]); |
cdd91a77 | 5225 | val->totalram = managed_pages; |
11fb9989 | 5226 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
75ef7184 | 5227 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 5228 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
5229 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
5230 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
5231 | ||
5232 | if (is_highmem(zone)) { | |
9705bea5 | 5233 | managed_highpages += zone_managed_pages(zone); |
fc2bd799 JK |
5234 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
5235 | } | |
5236 | } | |
5237 | val->totalhigh = managed_highpages; | |
5238 | val->freehigh = free_highpages; | |
98d2b0eb | 5239 | #else |
fc2bd799 JK |
5240 | val->totalhigh = managed_highpages; |
5241 | val->freehigh = free_highpages; | |
98d2b0eb | 5242 | #endif |
1da177e4 LT |
5243 | val->mem_unit = PAGE_SIZE; |
5244 | } | |
5245 | #endif | |
5246 | ||
ddd588b5 | 5247 | /* |
7bf02ea2 DR |
5248 | * Determine whether the node should be displayed or not, depending on whether |
5249 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 5250 | */ |
9af744d7 | 5251 | static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask) |
ddd588b5 | 5252 | { |
ddd588b5 | 5253 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
9af744d7 | 5254 | return false; |
ddd588b5 | 5255 | |
9af744d7 MH |
5256 | /* |
5257 | * no node mask - aka implicit memory numa policy. Do not bother with | |
5258 | * the synchronization - read_mems_allowed_begin - because we do not | |
5259 | * have to be precise here. | |
5260 | */ | |
5261 | if (!nodemask) | |
5262 | nodemask = &cpuset_current_mems_allowed; | |
5263 | ||
5264 | return !node_isset(nid, *nodemask); | |
ddd588b5 DR |
5265 | } |
5266 | ||
1da177e4 LT |
5267 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
5268 | ||
377e4f16 RV |
5269 | static void show_migration_types(unsigned char type) |
5270 | { | |
5271 | static const char types[MIGRATE_TYPES] = { | |
5272 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 5273 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
5274 | [MIGRATE_RECLAIMABLE] = 'E', |
5275 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
5276 | #ifdef CONFIG_CMA |
5277 | [MIGRATE_CMA] = 'C', | |
5278 | #endif | |
194159fb | 5279 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 5280 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 5281 | #endif |
377e4f16 RV |
5282 | }; |
5283 | char tmp[MIGRATE_TYPES + 1]; | |
5284 | char *p = tmp; | |
5285 | int i; | |
5286 | ||
5287 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
5288 | if (type & (1 << i)) | |
5289 | *p++ = types[i]; | |
5290 | } | |
5291 | ||
5292 | *p = '\0'; | |
1f84a18f | 5293 | printk(KERN_CONT "(%s) ", tmp); |
377e4f16 RV |
5294 | } |
5295 | ||
1da177e4 LT |
5296 | /* |
5297 | * Show free area list (used inside shift_scroll-lock stuff) | |
5298 | * We also calculate the percentage fragmentation. We do this by counting the | |
5299 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
5300 | * |
5301 | * Bits in @filter: | |
5302 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
5303 | * cpuset. | |
1da177e4 | 5304 | */ |
9af744d7 | 5305 | void show_free_areas(unsigned int filter, nodemask_t *nodemask) |
1da177e4 | 5306 | { |
d1bfcdb8 | 5307 | unsigned long free_pcp = 0; |
c7241913 | 5308 | int cpu; |
1da177e4 | 5309 | struct zone *zone; |
599d0c95 | 5310 | pg_data_t *pgdat; |
1da177e4 | 5311 | |
ee99c71c | 5312 | for_each_populated_zone(zone) { |
9af744d7 | 5313 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5314 | continue; |
d1bfcdb8 | 5315 | |
761b0677 KK |
5316 | for_each_online_cpu(cpu) |
5317 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
5318 | } |
5319 | ||
a731286d KM |
5320 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
5321 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
8d92890b | 5322 | " unevictable:%lu dirty:%lu writeback:%lu\n" |
d1bfcdb8 | 5323 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a | 5324 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 5325 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
599d0c95 MG |
5326 | global_node_page_state(NR_ACTIVE_ANON), |
5327 | global_node_page_state(NR_INACTIVE_ANON), | |
5328 | global_node_page_state(NR_ISOLATED_ANON), | |
5329 | global_node_page_state(NR_ACTIVE_FILE), | |
5330 | global_node_page_state(NR_INACTIVE_FILE), | |
5331 | global_node_page_state(NR_ISOLATED_FILE), | |
5332 | global_node_page_state(NR_UNEVICTABLE), | |
11fb9989 MG |
5333 | global_node_page_state(NR_FILE_DIRTY), |
5334 | global_node_page_state(NR_WRITEBACK), | |
d507e2eb JW |
5335 | global_node_page_state(NR_SLAB_RECLAIMABLE), |
5336 | global_node_page_state(NR_SLAB_UNRECLAIMABLE), | |
50658e2e | 5337 | global_node_page_state(NR_FILE_MAPPED), |
11fb9989 | 5338 | global_node_page_state(NR_SHMEM), |
c41f012a MH |
5339 | global_zone_page_state(NR_PAGETABLE), |
5340 | global_zone_page_state(NR_BOUNCE), | |
5341 | global_zone_page_state(NR_FREE_PAGES), | |
d1bfcdb8 | 5342 | free_pcp, |
c41f012a | 5343 | global_zone_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 5344 | |
599d0c95 | 5345 | for_each_online_pgdat(pgdat) { |
9af744d7 | 5346 | if (show_mem_node_skip(filter, pgdat->node_id, nodemask)) |
c02e50bb MH |
5347 | continue; |
5348 | ||
599d0c95 MG |
5349 | printk("Node %d" |
5350 | " active_anon:%lukB" | |
5351 | " inactive_anon:%lukB" | |
5352 | " active_file:%lukB" | |
5353 | " inactive_file:%lukB" | |
5354 | " unevictable:%lukB" | |
5355 | " isolated(anon):%lukB" | |
5356 | " isolated(file):%lukB" | |
50658e2e | 5357 | " mapped:%lukB" |
11fb9989 MG |
5358 | " dirty:%lukB" |
5359 | " writeback:%lukB" | |
5360 | " shmem:%lukB" | |
5361 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5362 | " shmem_thp: %lukB" | |
5363 | " shmem_pmdmapped: %lukB" | |
5364 | " anon_thp: %lukB" | |
5365 | #endif | |
5366 | " writeback_tmp:%lukB" | |
599d0c95 MG |
5367 | " all_unreclaimable? %s" |
5368 | "\n", | |
5369 | pgdat->node_id, | |
5370 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), | |
5371 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), | |
5372 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), | |
5373 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), | |
5374 | K(node_page_state(pgdat, NR_UNEVICTABLE)), | |
5375 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), | |
5376 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), | |
50658e2e | 5377 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
11fb9989 MG |
5378 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
5379 | K(node_page_state(pgdat, NR_WRITEBACK)), | |
1f06b81a | 5380 | K(node_page_state(pgdat, NR_SHMEM)), |
11fb9989 MG |
5381 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5382 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), | |
5383 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) | |
5384 | * HPAGE_PMD_NR), | |
5385 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), | |
5386 | #endif | |
11fb9989 | 5387 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
c73322d0 JW |
5388 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
5389 | "yes" : "no"); | |
599d0c95 MG |
5390 | } |
5391 | ||
ee99c71c | 5392 | for_each_populated_zone(zone) { |
1da177e4 LT |
5393 | int i; |
5394 | ||
9af744d7 | 5395 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5396 | continue; |
d1bfcdb8 KK |
5397 | |
5398 | free_pcp = 0; | |
5399 | for_each_online_cpu(cpu) | |
5400 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
5401 | ||
1da177e4 | 5402 | show_node(zone); |
1f84a18f JP |
5403 | printk(KERN_CONT |
5404 | "%s" | |
1da177e4 LT |
5405 | " free:%lukB" |
5406 | " min:%lukB" | |
5407 | " low:%lukB" | |
5408 | " high:%lukB" | |
e47b346a | 5409 | " reserved_highatomic:%luKB" |
71c799f4 MK |
5410 | " active_anon:%lukB" |
5411 | " inactive_anon:%lukB" | |
5412 | " active_file:%lukB" | |
5413 | " inactive_file:%lukB" | |
5414 | " unevictable:%lukB" | |
5a1c84b4 | 5415 | " writepending:%lukB" |
1da177e4 | 5416 | " present:%lukB" |
9feedc9d | 5417 | " managed:%lukB" |
4a0aa73f | 5418 | " mlocked:%lukB" |
c6a7f572 | 5419 | " kernel_stack:%lukB" |
628d06a4 ST |
5420 | #ifdef CONFIG_SHADOW_CALL_STACK |
5421 | " shadow_call_stack:%lukB" | |
5422 | #endif | |
4a0aa73f | 5423 | " pagetables:%lukB" |
4a0aa73f | 5424 | " bounce:%lukB" |
d1bfcdb8 KK |
5425 | " free_pcp:%lukB" |
5426 | " local_pcp:%ukB" | |
d1ce749a | 5427 | " free_cma:%lukB" |
1da177e4 LT |
5428 | "\n", |
5429 | zone->name, | |
88f5acf8 | 5430 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
5431 | K(min_wmark_pages(zone)), |
5432 | K(low_wmark_pages(zone)), | |
5433 | K(high_wmark_pages(zone)), | |
e47b346a | 5434 | K(zone->nr_reserved_highatomic), |
71c799f4 MK |
5435 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
5436 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), | |
5437 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), | |
5438 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), | |
5439 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), | |
5a1c84b4 | 5440 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
1da177e4 | 5441 | K(zone->present_pages), |
9705bea5 | 5442 | K(zone_managed_pages(zone)), |
4a0aa73f | 5443 | K(zone_page_state(zone, NR_MLOCK)), |
d30dd8be | 5444 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
628d06a4 ST |
5445 | #ifdef CONFIG_SHADOW_CALL_STACK |
5446 | zone_page_state(zone, NR_KERNEL_SCS_KB), | |
5447 | #endif | |
4a0aa73f | 5448 | K(zone_page_state(zone, NR_PAGETABLE)), |
4a0aa73f | 5449 | K(zone_page_state(zone, NR_BOUNCE)), |
d1bfcdb8 KK |
5450 | K(free_pcp), |
5451 | K(this_cpu_read(zone->pageset->pcp.count)), | |
33e077bd | 5452 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
1da177e4 LT |
5453 | printk("lowmem_reserve[]:"); |
5454 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1f84a18f JP |
5455 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
5456 | printk(KERN_CONT "\n"); | |
1da177e4 LT |
5457 | } |
5458 | ||
ee99c71c | 5459 | for_each_populated_zone(zone) { |
d00181b9 KS |
5460 | unsigned int order; |
5461 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 5462 | unsigned char types[MAX_ORDER]; |
1da177e4 | 5463 | |
9af744d7 | 5464 | if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask)) |
ddd588b5 | 5465 | continue; |
1da177e4 | 5466 | show_node(zone); |
1f84a18f | 5467 | printk(KERN_CONT "%s: ", zone->name); |
1da177e4 LT |
5468 | |
5469 | spin_lock_irqsave(&zone->lock, flags); | |
5470 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
5471 | struct free_area *area = &zone->free_area[order]; |
5472 | int type; | |
5473 | ||
5474 | nr[order] = area->nr_free; | |
8f9de51a | 5475 | total += nr[order] << order; |
377e4f16 RV |
5476 | |
5477 | types[order] = 0; | |
5478 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
b03641af | 5479 | if (!free_area_empty(area, type)) |
377e4f16 RV |
5480 | types[order] |= 1 << type; |
5481 | } | |
1da177e4 LT |
5482 | } |
5483 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 5484 | for (order = 0; order < MAX_ORDER; order++) { |
1f84a18f JP |
5485 | printk(KERN_CONT "%lu*%lukB ", |
5486 | nr[order], K(1UL) << order); | |
377e4f16 RV |
5487 | if (nr[order]) |
5488 | show_migration_types(types[order]); | |
5489 | } | |
1f84a18f | 5490 | printk(KERN_CONT "= %lukB\n", K(total)); |
1da177e4 LT |
5491 | } |
5492 | ||
949f7ec5 DR |
5493 | hugetlb_show_meminfo(); |
5494 | ||
11fb9989 | 5495 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
e6f3602d | 5496 | |
1da177e4 LT |
5497 | show_swap_cache_info(); |
5498 | } | |
5499 | ||
19770b32 MG |
5500 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
5501 | { | |
5502 | zoneref->zone = zone; | |
5503 | zoneref->zone_idx = zone_idx(zone); | |
5504 | } | |
5505 | ||
1da177e4 LT |
5506 | /* |
5507 | * Builds allocation fallback zone lists. | |
1a93205b CL |
5508 | * |
5509 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 5510 | */ |
9d3be21b | 5511 | static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs) |
1da177e4 | 5512 | { |
1a93205b | 5513 | struct zone *zone; |
bc732f1d | 5514 | enum zone_type zone_type = MAX_NR_ZONES; |
9d3be21b | 5515 | int nr_zones = 0; |
02a68a5e CL |
5516 | |
5517 | do { | |
2f6726e5 | 5518 | zone_type--; |
070f8032 | 5519 | zone = pgdat->node_zones + zone_type; |
6aa303de | 5520 | if (managed_zone(zone)) { |
9d3be21b | 5521 | zoneref_set_zone(zone, &zonerefs[nr_zones++]); |
070f8032 | 5522 | check_highest_zone(zone_type); |
1da177e4 | 5523 | } |
2f6726e5 | 5524 | } while (zone_type); |
bc732f1d | 5525 | |
070f8032 | 5526 | return nr_zones; |
1da177e4 LT |
5527 | } |
5528 | ||
5529 | #ifdef CONFIG_NUMA | |
f0c0b2b8 KH |
5530 | |
5531 | static int __parse_numa_zonelist_order(char *s) | |
5532 | { | |
c9bff3ee MH |
5533 | /* |
5534 | * We used to support different zonlists modes but they turned | |
5535 | * out to be just not useful. Let's keep the warning in place | |
5536 | * if somebody still use the cmd line parameter so that we do | |
5537 | * not fail it silently | |
5538 | */ | |
5539 | if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) { | |
5540 | pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s); | |
f0c0b2b8 KH |
5541 | return -EINVAL; |
5542 | } | |
5543 | return 0; | |
5544 | } | |
5545 | ||
5546 | static __init int setup_numa_zonelist_order(char *s) | |
5547 | { | |
ecb256f8 VL |
5548 | if (!s) |
5549 | return 0; | |
5550 | ||
c9bff3ee | 5551 | return __parse_numa_zonelist_order(s); |
f0c0b2b8 KH |
5552 | } |
5553 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
5554 | ||
c9bff3ee MH |
5555 | char numa_zonelist_order[] = "Node"; |
5556 | ||
f0c0b2b8 KH |
5557 | /* |
5558 | * sysctl handler for numa_zonelist_order | |
5559 | */ | |
cccad5b9 | 5560 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 5561 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
5562 | loff_t *ppos) |
5563 | { | |
c9bff3ee | 5564 | char *str; |
f0c0b2b8 KH |
5565 | int ret; |
5566 | ||
c9bff3ee MH |
5567 | if (!write) |
5568 | return proc_dostring(table, write, buffer, length, ppos); | |
5569 | str = memdup_user_nul(buffer, 16); | |
5570 | if (IS_ERR(str)) | |
5571 | return PTR_ERR(str); | |
dacbde09 | 5572 | |
c9bff3ee MH |
5573 | ret = __parse_numa_zonelist_order(str); |
5574 | kfree(str); | |
443c6f14 | 5575 | return ret; |
f0c0b2b8 KH |
5576 | } |
5577 | ||
5578 | ||
62bc62a8 | 5579 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
5580 | static int node_load[MAX_NUMNODES]; |
5581 | ||
1da177e4 | 5582 | /** |
4dc3b16b | 5583 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
5584 | * @node: node whose fallback list we're appending |
5585 | * @used_node_mask: nodemask_t of already used nodes | |
5586 | * | |
5587 | * We use a number of factors to determine which is the next node that should | |
5588 | * appear on a given node's fallback list. The node should not have appeared | |
5589 | * already in @node's fallback list, and it should be the next closest node | |
5590 | * according to the distance array (which contains arbitrary distance values | |
5591 | * from each node to each node in the system), and should also prefer nodes | |
5592 | * with no CPUs, since presumably they'll have very little allocation pressure | |
5593 | * on them otherwise. | |
a862f68a MR |
5594 | * |
5595 | * Return: node id of the found node or %NUMA_NO_NODE if no node is found. | |
1da177e4 | 5596 | */ |
f0c0b2b8 | 5597 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 5598 | { |
4cf808eb | 5599 | int n, val; |
1da177e4 | 5600 | int min_val = INT_MAX; |
00ef2d2f | 5601 | int best_node = NUMA_NO_NODE; |
a70f7302 | 5602 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 5603 | |
4cf808eb LT |
5604 | /* Use the local node if we haven't already */ |
5605 | if (!node_isset(node, *used_node_mask)) { | |
5606 | node_set(node, *used_node_mask); | |
5607 | return node; | |
5608 | } | |
1da177e4 | 5609 | |
4b0ef1fe | 5610 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
5611 | |
5612 | /* Don't want a node to appear more than once */ | |
5613 | if (node_isset(n, *used_node_mask)) | |
5614 | continue; | |
5615 | ||
1da177e4 LT |
5616 | /* Use the distance array to find the distance */ |
5617 | val = node_distance(node, n); | |
5618 | ||
4cf808eb LT |
5619 | /* Penalize nodes under us ("prefer the next node") */ |
5620 | val += (n < node); | |
5621 | ||
1da177e4 | 5622 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
5623 | tmp = cpumask_of_node(n); |
5624 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
5625 | val += PENALTY_FOR_NODE_WITH_CPUS; |
5626 | ||
5627 | /* Slight preference for less loaded node */ | |
5628 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
5629 | val += node_load[n]; | |
5630 | ||
5631 | if (val < min_val) { | |
5632 | min_val = val; | |
5633 | best_node = n; | |
5634 | } | |
5635 | } | |
5636 | ||
5637 | if (best_node >= 0) | |
5638 | node_set(best_node, *used_node_mask); | |
5639 | ||
5640 | return best_node; | |
5641 | } | |
5642 | ||
f0c0b2b8 KH |
5643 | |
5644 | /* | |
5645 | * Build zonelists ordered by node and zones within node. | |
5646 | * This results in maximum locality--normal zone overflows into local | |
5647 | * DMA zone, if any--but risks exhausting DMA zone. | |
5648 | */ | |
9d3be21b MH |
5649 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order, |
5650 | unsigned nr_nodes) | |
1da177e4 | 5651 | { |
9d3be21b MH |
5652 | struct zoneref *zonerefs; |
5653 | int i; | |
5654 | ||
5655 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; | |
5656 | ||
5657 | for (i = 0; i < nr_nodes; i++) { | |
5658 | int nr_zones; | |
5659 | ||
5660 | pg_data_t *node = NODE_DATA(node_order[i]); | |
f0c0b2b8 | 5661 | |
9d3be21b MH |
5662 | nr_zones = build_zonerefs_node(node, zonerefs); |
5663 | zonerefs += nr_zones; | |
5664 | } | |
5665 | zonerefs->zone = NULL; | |
5666 | zonerefs->zone_idx = 0; | |
f0c0b2b8 KH |
5667 | } |
5668 | ||
523b9458 CL |
5669 | /* |
5670 | * Build gfp_thisnode zonelists | |
5671 | */ | |
5672 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
5673 | { | |
9d3be21b MH |
5674 | struct zoneref *zonerefs; |
5675 | int nr_zones; | |
523b9458 | 5676 | |
9d3be21b MH |
5677 | zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; |
5678 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5679 | zonerefs += nr_zones; | |
5680 | zonerefs->zone = NULL; | |
5681 | zonerefs->zone_idx = 0; | |
523b9458 CL |
5682 | } |
5683 | ||
f0c0b2b8 KH |
5684 | /* |
5685 | * Build zonelists ordered by zone and nodes within zones. | |
5686 | * This results in conserving DMA zone[s] until all Normal memory is | |
5687 | * exhausted, but results in overflowing to remote node while memory | |
5688 | * may still exist in local DMA zone. | |
5689 | */ | |
f0c0b2b8 | 5690 | |
f0c0b2b8 KH |
5691 | static void build_zonelists(pg_data_t *pgdat) |
5692 | { | |
9d3be21b MH |
5693 | static int node_order[MAX_NUMNODES]; |
5694 | int node, load, nr_nodes = 0; | |
1da177e4 | 5695 | nodemask_t used_mask; |
f0c0b2b8 | 5696 | int local_node, prev_node; |
1da177e4 LT |
5697 | |
5698 | /* NUMA-aware ordering of nodes */ | |
5699 | local_node = pgdat->node_id; | |
62bc62a8 | 5700 | load = nr_online_nodes; |
1da177e4 LT |
5701 | prev_node = local_node; |
5702 | nodes_clear(used_mask); | |
f0c0b2b8 | 5703 | |
f0c0b2b8 | 5704 | memset(node_order, 0, sizeof(node_order)); |
1da177e4 LT |
5705 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5706 | /* | |
5707 | * We don't want to pressure a particular node. | |
5708 | * So adding penalty to the first node in same | |
5709 | * distance group to make it round-robin. | |
5710 | */ | |
957f822a DR |
5711 | if (node_distance(local_node, node) != |
5712 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
5713 | node_load[node] = load; |
5714 | ||
9d3be21b | 5715 | node_order[nr_nodes++] = node; |
1da177e4 LT |
5716 | prev_node = node; |
5717 | load--; | |
1da177e4 | 5718 | } |
523b9458 | 5719 | |
9d3be21b | 5720 | build_zonelists_in_node_order(pgdat, node_order, nr_nodes); |
523b9458 | 5721 | build_thisnode_zonelists(pgdat); |
1da177e4 LT |
5722 | } |
5723 | ||
7aac7898 LS |
5724 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5725 | /* | |
5726 | * Return node id of node used for "local" allocations. | |
5727 | * I.e., first node id of first zone in arg node's generic zonelist. | |
5728 | * Used for initializing percpu 'numa_mem', which is used primarily | |
5729 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
5730 | */ | |
5731 | int local_memory_node(int node) | |
5732 | { | |
c33d6c06 | 5733 | struct zoneref *z; |
7aac7898 | 5734 | |
c33d6c06 | 5735 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 5736 | gfp_zone(GFP_KERNEL), |
c33d6c06 | 5737 | NULL); |
c1093b74 | 5738 | return zone_to_nid(z->zone); |
7aac7898 LS |
5739 | } |
5740 | #endif | |
f0c0b2b8 | 5741 | |
6423aa81 JK |
5742 | static void setup_min_unmapped_ratio(void); |
5743 | static void setup_min_slab_ratio(void); | |
1da177e4 LT |
5744 | #else /* CONFIG_NUMA */ |
5745 | ||
f0c0b2b8 | 5746 | static void build_zonelists(pg_data_t *pgdat) |
1da177e4 | 5747 | { |
19655d34 | 5748 | int node, local_node; |
9d3be21b MH |
5749 | struct zoneref *zonerefs; |
5750 | int nr_zones; | |
1da177e4 LT |
5751 | |
5752 | local_node = pgdat->node_id; | |
1da177e4 | 5753 | |
9d3be21b MH |
5754 | zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; |
5755 | nr_zones = build_zonerefs_node(pgdat, zonerefs); | |
5756 | zonerefs += nr_zones; | |
1da177e4 | 5757 | |
54a6eb5c MG |
5758 | /* |
5759 | * Now we build the zonelist so that it contains the zones | |
5760 | * of all the other nodes. | |
5761 | * We don't want to pressure a particular node, so when | |
5762 | * building the zones for node N, we make sure that the | |
5763 | * zones coming right after the local ones are those from | |
5764 | * node N+1 (modulo N) | |
5765 | */ | |
5766 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
5767 | if (!node_online(node)) | |
5768 | continue; | |
9d3be21b MH |
5769 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5770 | zonerefs += nr_zones; | |
1da177e4 | 5771 | } |
54a6eb5c MG |
5772 | for (node = 0; node < local_node; node++) { |
5773 | if (!node_online(node)) | |
5774 | continue; | |
9d3be21b MH |
5775 | nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs); |
5776 | zonerefs += nr_zones; | |
54a6eb5c MG |
5777 | } |
5778 | ||
9d3be21b MH |
5779 | zonerefs->zone = NULL; |
5780 | zonerefs->zone_idx = 0; | |
1da177e4 LT |
5781 | } |
5782 | ||
5783 | #endif /* CONFIG_NUMA */ | |
5784 | ||
99dcc3e5 CL |
5785 | /* |
5786 | * Boot pageset table. One per cpu which is going to be used for all | |
5787 | * zones and all nodes. The parameters will be set in such a way | |
5788 | * that an item put on a list will immediately be handed over to | |
5789 | * the buddy list. This is safe since pageset manipulation is done | |
5790 | * with interrupts disabled. | |
5791 | * | |
5792 | * The boot_pagesets must be kept even after bootup is complete for | |
5793 | * unused processors and/or zones. They do play a role for bootstrapping | |
5794 | * hotplugged processors. | |
5795 | * | |
5796 | * zoneinfo_show() and maybe other functions do | |
5797 | * not check if the processor is online before following the pageset pointer. | |
5798 | * Other parts of the kernel may not check if the zone is available. | |
5799 | */ | |
5800 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
5801 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
385386cf | 5802 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); |
99dcc3e5 | 5803 | |
11cd8638 | 5804 | static void __build_all_zonelists(void *data) |
1da177e4 | 5805 | { |
6811378e | 5806 | int nid; |
afb6ebb3 | 5807 | int __maybe_unused cpu; |
9adb62a5 | 5808 | pg_data_t *self = data; |
b93e0f32 MH |
5809 | static DEFINE_SPINLOCK(lock); |
5810 | ||
5811 | spin_lock(&lock); | |
9276b1bc | 5812 | |
7f9cfb31 BL |
5813 | #ifdef CONFIG_NUMA |
5814 | memset(node_load, 0, sizeof(node_load)); | |
5815 | #endif | |
9adb62a5 | 5816 | |
c1152583 WY |
5817 | /* |
5818 | * This node is hotadded and no memory is yet present. So just | |
5819 | * building zonelists is fine - no need to touch other nodes. | |
5820 | */ | |
9adb62a5 JL |
5821 | if (self && !node_online(self->node_id)) { |
5822 | build_zonelists(self); | |
c1152583 WY |
5823 | } else { |
5824 | for_each_online_node(nid) { | |
5825 | pg_data_t *pgdat = NODE_DATA(nid); | |
7ea1530a | 5826 | |
c1152583 WY |
5827 | build_zonelists(pgdat); |
5828 | } | |
99dcc3e5 | 5829 | |
7aac7898 LS |
5830 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5831 | /* | |
5832 | * We now know the "local memory node" for each node-- | |
5833 | * i.e., the node of the first zone in the generic zonelist. | |
5834 | * Set up numa_mem percpu variable for on-line cpus. During | |
5835 | * boot, only the boot cpu should be on-line; we'll init the | |
5836 | * secondary cpus' numa_mem as they come on-line. During | |
5837 | * node/memory hotplug, we'll fixup all on-line cpus. | |
5838 | */ | |
d9c9a0b9 | 5839 | for_each_online_cpu(cpu) |
7aac7898 | 5840 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
afb6ebb3 | 5841 | #endif |
d9c9a0b9 | 5842 | } |
b93e0f32 MH |
5843 | |
5844 | spin_unlock(&lock); | |
6811378e YG |
5845 | } |
5846 | ||
061f67bc RV |
5847 | static noinline void __init |
5848 | build_all_zonelists_init(void) | |
5849 | { | |
afb6ebb3 MH |
5850 | int cpu; |
5851 | ||
061f67bc | 5852 | __build_all_zonelists(NULL); |
afb6ebb3 MH |
5853 | |
5854 | /* | |
5855 | * Initialize the boot_pagesets that are going to be used | |
5856 | * for bootstrapping processors. The real pagesets for | |
5857 | * each zone will be allocated later when the per cpu | |
5858 | * allocator is available. | |
5859 | * | |
5860 | * boot_pagesets are used also for bootstrapping offline | |
5861 | * cpus if the system is already booted because the pagesets | |
5862 | * are needed to initialize allocators on a specific cpu too. | |
5863 | * F.e. the percpu allocator needs the page allocator which | |
5864 | * needs the percpu allocator in order to allocate its pagesets | |
5865 | * (a chicken-egg dilemma). | |
5866 | */ | |
5867 | for_each_possible_cpu(cpu) | |
5868 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); | |
5869 | ||
061f67bc RV |
5870 | mminit_verify_zonelist(); |
5871 | cpuset_init_current_mems_allowed(); | |
5872 | } | |
5873 | ||
4eaf3f64 | 5874 | /* |
4eaf3f64 | 5875 | * unless system_state == SYSTEM_BOOTING. |
061f67bc | 5876 | * |
72675e13 | 5877 | * __ref due to call of __init annotated helper build_all_zonelists_init |
061f67bc | 5878 | * [protected by SYSTEM_BOOTING]. |
4eaf3f64 | 5879 | */ |
72675e13 | 5880 | void __ref build_all_zonelists(pg_data_t *pgdat) |
6811378e YG |
5881 | { |
5882 | if (system_state == SYSTEM_BOOTING) { | |
061f67bc | 5883 | build_all_zonelists_init(); |
6811378e | 5884 | } else { |
11cd8638 | 5885 | __build_all_zonelists(pgdat); |
6811378e YG |
5886 | /* cpuset refresh routine should be here */ |
5887 | } | |
bd1e22b8 | 5888 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
5889 | /* |
5890 | * Disable grouping by mobility if the number of pages in the | |
5891 | * system is too low to allow the mechanism to work. It would be | |
5892 | * more accurate, but expensive to check per-zone. This check is | |
5893 | * made on memory-hotadd so a system can start with mobility | |
5894 | * disabled and enable it later | |
5895 | */ | |
d9c23400 | 5896 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
5897 | page_group_by_mobility_disabled = 1; |
5898 | else | |
5899 | page_group_by_mobility_disabled = 0; | |
5900 | ||
ce0725f7 | 5901 | pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n", |
756a025f | 5902 | nr_online_nodes, |
756a025f JP |
5903 | page_group_by_mobility_disabled ? "off" : "on", |
5904 | vm_total_pages); | |
f0c0b2b8 | 5905 | #ifdef CONFIG_NUMA |
f88dfff5 | 5906 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 5907 | #endif |
1da177e4 LT |
5908 | } |
5909 | ||
a9a9e77f PT |
5910 | /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */ |
5911 | static bool __meminit | |
5912 | overlap_memmap_init(unsigned long zone, unsigned long *pfn) | |
5913 | { | |
a9a9e77f PT |
5914 | static struct memblock_region *r; |
5915 | ||
5916 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
5917 | if (!r || *pfn >= memblock_region_memory_end_pfn(r)) { | |
5918 | for_each_memblock(memory, r) { | |
5919 | if (*pfn < memblock_region_memory_end_pfn(r)) | |
5920 | break; | |
5921 | } | |
5922 | } | |
5923 | if (*pfn >= memblock_region_memory_base_pfn(r) && | |
5924 | memblock_is_mirror(r)) { | |
5925 | *pfn = memblock_region_memory_end_pfn(r); | |
5926 | return true; | |
5927 | } | |
5928 | } | |
a9a9e77f PT |
5929 | return false; |
5930 | } | |
5931 | ||
1da177e4 LT |
5932 | /* |
5933 | * Initially all pages are reserved - free ones are freed | |
c6ffc5ca | 5934 | * up by memblock_free_all() once the early boot process is |
1da177e4 LT |
5935 | * done. Non-atomic initialization, single-pass. |
5936 | */ | |
c09b4240 | 5937 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a99583e7 CH |
5938 | unsigned long start_pfn, enum memmap_context context, |
5939 | struct vmem_altmap *altmap) | |
1da177e4 | 5940 | { |
a9a9e77f | 5941 | unsigned long pfn, end_pfn = start_pfn + size; |
d0dc12e8 | 5942 | struct page *page; |
1da177e4 | 5943 | |
22b31eec HD |
5944 | if (highest_memmap_pfn < end_pfn - 1) |
5945 | highest_memmap_pfn = end_pfn - 1; | |
5946 | ||
966cf44f | 5947 | #ifdef CONFIG_ZONE_DEVICE |
4b94ffdc DW |
5948 | /* |
5949 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
966cf44f AD |
5950 | * memory. We limit the total number of pages to initialize to just |
5951 | * those that might contain the memory mapping. We will defer the | |
5952 | * ZONE_DEVICE page initialization until after we have released | |
5953 | * the hotplug lock. | |
4b94ffdc | 5954 | */ |
966cf44f AD |
5955 | if (zone == ZONE_DEVICE) { |
5956 | if (!altmap) | |
5957 | return; | |
5958 | ||
5959 | if (start_pfn == altmap->base_pfn) | |
5960 | start_pfn += altmap->reserve; | |
5961 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); | |
5962 | } | |
5963 | #endif | |
4b94ffdc | 5964 | |
948c436e | 5965 | for (pfn = start_pfn; pfn < end_pfn; ) { |
a2f3aa02 | 5966 | /* |
b72d0ffb AM |
5967 | * There can be holes in boot-time mem_map[]s handed to this |
5968 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 5969 | */ |
a9a9e77f | 5970 | if (context == MEMMAP_EARLY) { |
a9a9e77f PT |
5971 | if (overlap_memmap_init(zone, &pfn)) |
5972 | continue; | |
5973 | if (defer_init(nid, pfn, end_pfn)) | |
5974 | break; | |
a2f3aa02 | 5975 | } |
ac5d2539 | 5976 | |
d0dc12e8 PT |
5977 | page = pfn_to_page(pfn); |
5978 | __init_single_page(page, pfn, zone, nid); | |
5979 | if (context == MEMMAP_HOTPLUG) | |
d483da5b | 5980 | __SetPageReserved(page); |
d0dc12e8 | 5981 | |
ac5d2539 MG |
5982 | /* |
5983 | * Mark the block movable so that blocks are reserved for | |
5984 | * movable at startup. This will force kernel allocations | |
5985 | * to reserve their blocks rather than leaking throughout | |
5986 | * the address space during boot when many long-lived | |
974a786e | 5987 | * kernel allocations are made. |
ac5d2539 MG |
5988 | * |
5989 | * bitmap is created for zone's valid pfn range. but memmap | |
5990 | * can be created for invalid pages (for alignment) | |
5991 | * check here not to call set_pageblock_migratetype() against | |
5992 | * pfn out of zone. | |
5993 | */ | |
5994 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
ac5d2539 | 5995 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
9b6e63cb | 5996 | cond_resched(); |
ac5d2539 | 5997 | } |
948c436e | 5998 | pfn++; |
1da177e4 LT |
5999 | } |
6000 | } | |
6001 | ||
966cf44f AD |
6002 | #ifdef CONFIG_ZONE_DEVICE |
6003 | void __ref memmap_init_zone_device(struct zone *zone, | |
6004 | unsigned long start_pfn, | |
1f8d75c1 | 6005 | unsigned long nr_pages, |
966cf44f AD |
6006 | struct dev_pagemap *pgmap) |
6007 | { | |
1f8d75c1 | 6008 | unsigned long pfn, end_pfn = start_pfn + nr_pages; |
966cf44f | 6009 | struct pglist_data *pgdat = zone->zone_pgdat; |
514caf23 | 6010 | struct vmem_altmap *altmap = pgmap_altmap(pgmap); |
966cf44f AD |
6011 | unsigned long zone_idx = zone_idx(zone); |
6012 | unsigned long start = jiffies; | |
6013 | int nid = pgdat->node_id; | |
6014 | ||
46d945ae | 6015 | if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE)) |
966cf44f AD |
6016 | return; |
6017 | ||
6018 | /* | |
6019 | * The call to memmap_init_zone should have already taken care | |
6020 | * of the pages reserved for the memmap, so we can just jump to | |
6021 | * the end of that region and start processing the device pages. | |
6022 | */ | |
514caf23 | 6023 | if (altmap) { |
966cf44f | 6024 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap); |
1f8d75c1 | 6025 | nr_pages = end_pfn - start_pfn; |
966cf44f AD |
6026 | } |
6027 | ||
6028 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
6029 | struct page *page = pfn_to_page(pfn); | |
6030 | ||
6031 | __init_single_page(page, pfn, zone_idx, nid); | |
6032 | ||
6033 | /* | |
6034 | * Mark page reserved as it will need to wait for onlining | |
6035 | * phase for it to be fully associated with a zone. | |
6036 | * | |
6037 | * We can use the non-atomic __set_bit operation for setting | |
6038 | * the flag as we are still initializing the pages. | |
6039 | */ | |
6040 | __SetPageReserved(page); | |
6041 | ||
6042 | /* | |
8a164fef CH |
6043 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer |
6044 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is | |
6045 | * ever freed or placed on a driver-private list. | |
966cf44f AD |
6046 | */ |
6047 | page->pgmap = pgmap; | |
8a164fef | 6048 | page->zone_device_data = NULL; |
966cf44f AD |
6049 | |
6050 | /* | |
6051 | * Mark the block movable so that blocks are reserved for | |
6052 | * movable at startup. This will force kernel allocations | |
6053 | * to reserve their blocks rather than leaking throughout | |
6054 | * the address space during boot when many long-lived | |
6055 | * kernel allocations are made. | |
6056 | * | |
6057 | * bitmap is created for zone's valid pfn range. but memmap | |
6058 | * can be created for invalid pages (for alignment) | |
6059 | * check here not to call set_pageblock_migratetype() against | |
6060 | * pfn out of zone. | |
6061 | * | |
6062 | * Please note that MEMMAP_HOTPLUG path doesn't clear memmap | |
ba72b4c8 | 6063 | * because this is done early in section_activate() |
966cf44f AD |
6064 | */ |
6065 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
6066 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
6067 | cond_resched(); | |
6068 | } | |
6069 | } | |
6070 | ||
fdc029b1 | 6071 | pr_info("%s initialised %lu pages in %ums\n", __func__, |
1f8d75c1 | 6072 | nr_pages, jiffies_to_msecs(jiffies - start)); |
966cf44f AD |
6073 | } |
6074 | ||
6075 | #endif | |
1e548deb | 6076 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 6077 | { |
7aeb09f9 | 6078 | unsigned int order, t; |
b2a0ac88 MG |
6079 | for_each_migratetype_order(order, t) { |
6080 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
6081 | zone->free_area[order].nr_free = 0; |
6082 | } | |
6083 | } | |
6084 | ||
dfb3ccd0 | 6085 | void __meminit __weak memmap_init(unsigned long size, int nid, |
73a6e474 BH |
6086 | unsigned long zone, |
6087 | unsigned long range_start_pfn) | |
dfb3ccd0 | 6088 | { |
73a6e474 BH |
6089 | unsigned long start_pfn, end_pfn; |
6090 | unsigned long range_end_pfn = range_start_pfn + size; | |
6091 | int i; | |
6092 | ||
6093 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { | |
6094 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6095 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6096 | ||
6097 | if (end_pfn > start_pfn) { | |
6098 | size = end_pfn - start_pfn; | |
6099 | memmap_init_zone(size, nid, zone, start_pfn, | |
6100 | MEMMAP_EARLY, NULL); | |
6101 | } | |
6102 | } | |
dfb3ccd0 | 6103 | } |
1da177e4 | 6104 | |
7cd2b0a3 | 6105 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 6106 | { |
3a6be87f | 6107 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
6108 | int batch; |
6109 | ||
6110 | /* | |
6111 | * The per-cpu-pages pools are set to around 1000th of the | |
d8a759b5 | 6112 | * size of the zone. |
e7c8d5c9 | 6113 | */ |
9705bea5 | 6114 | batch = zone_managed_pages(zone) / 1024; |
d8a759b5 AL |
6115 | /* But no more than a meg. */ |
6116 | if (batch * PAGE_SIZE > 1024 * 1024) | |
6117 | batch = (1024 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
6118 | batch /= 4; /* We effectively *= 4 below */ |
6119 | if (batch < 1) | |
6120 | batch = 1; | |
6121 | ||
6122 | /* | |
0ceaacc9 NP |
6123 | * Clamp the batch to a 2^n - 1 value. Having a power |
6124 | * of 2 value was found to be more likely to have | |
6125 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 6126 | * |
0ceaacc9 NP |
6127 | * For example if 2 tasks are alternately allocating |
6128 | * batches of pages, one task can end up with a lot | |
6129 | * of pages of one half of the possible page colors | |
6130 | * and the other with pages of the other colors. | |
e7c8d5c9 | 6131 | */ |
9155203a | 6132 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 6133 | |
e7c8d5c9 | 6134 | return batch; |
3a6be87f DH |
6135 | |
6136 | #else | |
6137 | /* The deferral and batching of frees should be suppressed under NOMMU | |
6138 | * conditions. | |
6139 | * | |
6140 | * The problem is that NOMMU needs to be able to allocate large chunks | |
6141 | * of contiguous memory as there's no hardware page translation to | |
6142 | * assemble apparent contiguous memory from discontiguous pages. | |
6143 | * | |
6144 | * Queueing large contiguous runs of pages for batching, however, | |
6145 | * causes the pages to actually be freed in smaller chunks. As there | |
6146 | * can be a significant delay between the individual batches being | |
6147 | * recycled, this leads to the once large chunks of space being | |
6148 | * fragmented and becoming unavailable for high-order allocations. | |
6149 | */ | |
6150 | return 0; | |
6151 | #endif | |
e7c8d5c9 CL |
6152 | } |
6153 | ||
8d7a8fa9 CS |
6154 | /* |
6155 | * pcp->high and pcp->batch values are related and dependent on one another: | |
6156 | * ->batch must never be higher then ->high. | |
6157 | * The following function updates them in a safe manner without read side | |
6158 | * locking. | |
6159 | * | |
6160 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
6161 | * those fields changing asynchronously (acording the the above rule). | |
6162 | * | |
6163 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
6164 | * outside of boot time (or some other assurance that no concurrent updaters | |
6165 | * exist). | |
6166 | */ | |
6167 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
6168 | unsigned long batch) | |
6169 | { | |
6170 | /* start with a fail safe value for batch */ | |
6171 | pcp->batch = 1; | |
6172 | smp_wmb(); | |
6173 | ||
6174 | /* Update high, then batch, in order */ | |
6175 | pcp->high = high; | |
6176 | smp_wmb(); | |
6177 | ||
6178 | pcp->batch = batch; | |
6179 | } | |
6180 | ||
3664033c | 6181 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
6182 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
6183 | { | |
8d7a8fa9 | 6184 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
6185 | } |
6186 | ||
88c90dbc | 6187 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
6188 | { |
6189 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 6190 | int migratetype; |
2caaad41 | 6191 | |
1c6fe946 MD |
6192 | memset(p, 0, sizeof(*p)); |
6193 | ||
3dfa5721 | 6194 | pcp = &p->pcp; |
5f8dcc21 MG |
6195 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
6196 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
6197 | } |
6198 | ||
88c90dbc CS |
6199 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
6200 | { | |
6201 | pageset_init(p); | |
6202 | pageset_set_batch(p, batch); | |
6203 | } | |
6204 | ||
8ad4b1fb | 6205 | /* |
3664033c | 6206 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
6207 | * to the value high for the pageset p. |
6208 | */ | |
3664033c | 6209 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
6210 | unsigned long high) |
6211 | { | |
8d7a8fa9 CS |
6212 | unsigned long batch = max(1UL, high / 4); |
6213 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
6214 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 6215 | |
8d7a8fa9 | 6216 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
6217 | } |
6218 | ||
7cd2b0a3 DR |
6219 | static void pageset_set_high_and_batch(struct zone *zone, |
6220 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 6221 | { |
56cef2b8 | 6222 | if (percpu_pagelist_fraction) |
3664033c | 6223 | pageset_set_high(pcp, |
9705bea5 | 6224 | (zone_managed_pages(zone) / |
56cef2b8 CS |
6225 | percpu_pagelist_fraction)); |
6226 | else | |
6227 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
6228 | } | |
6229 | ||
169f6c19 CS |
6230 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
6231 | { | |
6232 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
6233 | ||
6234 | pageset_init(pcp); | |
6235 | pageset_set_high_and_batch(zone, pcp); | |
6236 | } | |
6237 | ||
72675e13 | 6238 | void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
6239 | { |
6240 | int cpu; | |
319774e2 | 6241 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
6242 | for_each_possible_cpu(cpu) |
6243 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
6244 | } |
6245 | ||
2caaad41 | 6246 | /* |
99dcc3e5 CL |
6247 | * Allocate per cpu pagesets and initialize them. |
6248 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 6249 | */ |
99dcc3e5 | 6250 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 6251 | { |
b4911ea2 | 6252 | struct pglist_data *pgdat; |
99dcc3e5 | 6253 | struct zone *zone; |
e7c8d5c9 | 6254 | |
319774e2 WF |
6255 | for_each_populated_zone(zone) |
6256 | setup_zone_pageset(zone); | |
b4911ea2 MG |
6257 | |
6258 | for_each_online_pgdat(pgdat) | |
6259 | pgdat->per_cpu_nodestats = | |
6260 | alloc_percpu(struct per_cpu_nodestat); | |
e7c8d5c9 CL |
6261 | } |
6262 | ||
c09b4240 | 6263 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 6264 | { |
99dcc3e5 CL |
6265 | /* |
6266 | * per cpu subsystem is not up at this point. The following code | |
6267 | * relies on the ability of the linker to provide the | |
6268 | * offset of a (static) per cpu variable into the per cpu area. | |
6269 | */ | |
6270 | zone->pageset = &boot_pageset; | |
ed8ece2e | 6271 | |
b38a8725 | 6272 | if (populated_zone(zone)) |
99dcc3e5 CL |
6273 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
6274 | zone->name, zone->present_pages, | |
6275 | zone_batchsize(zone)); | |
ed8ece2e DH |
6276 | } |
6277 | ||
dc0bbf3b | 6278 | void __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 6279 | unsigned long zone_start_pfn, |
b171e409 | 6280 | unsigned long size) |
ed8ece2e DH |
6281 | { |
6282 | struct pglist_data *pgdat = zone->zone_pgdat; | |
8f416836 | 6283 | int zone_idx = zone_idx(zone) + 1; |
9dcb8b68 | 6284 | |
8f416836 WY |
6285 | if (zone_idx > pgdat->nr_zones) |
6286 | pgdat->nr_zones = zone_idx; | |
ed8ece2e | 6287 | |
ed8ece2e DH |
6288 | zone->zone_start_pfn = zone_start_pfn; |
6289 | ||
708614e6 MG |
6290 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
6291 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
6292 | pgdat->node_id, | |
6293 | (unsigned long)zone_idx(zone), | |
6294 | zone_start_pfn, (zone_start_pfn + size)); | |
6295 | ||
1e548deb | 6296 | zone_init_free_lists(zone); |
9dcb8b68 | 6297 | zone->initialized = 1; |
ed8ece2e DH |
6298 | } |
6299 | ||
c713216d | 6300 | /** |
6782832e | 6301 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 6302 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 6303 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 6304 | * |
7d018176 ZZ |
6305 | * If an architecture guarantees that all ranges registered contain no holes |
6306 | * and may be freed, this this function may be used instead of calling | |
6307 | * memblock_free_early_nid() manually. | |
c713216d | 6308 | */ |
c13291a5 | 6309 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 6310 | { |
c13291a5 TH |
6311 | unsigned long start_pfn, end_pfn; |
6312 | int i, this_nid; | |
edbe7d23 | 6313 | |
c13291a5 TH |
6314 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
6315 | start_pfn = min(start_pfn, max_low_pfn); | |
6316 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 6317 | |
c13291a5 | 6318 | if (start_pfn < end_pfn) |
6782832e SS |
6319 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
6320 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
6321 | this_nid); | |
edbe7d23 | 6322 | } |
edbe7d23 | 6323 | } |
edbe7d23 | 6324 | |
c713216d MG |
6325 | /** |
6326 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 6327 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 6328 | * |
7d018176 ZZ |
6329 | * If an architecture guarantees that all ranges registered contain no holes and may |
6330 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
6331 | */ |
6332 | void __init sparse_memory_present_with_active_regions(int nid) | |
6333 | { | |
c13291a5 TH |
6334 | unsigned long start_pfn, end_pfn; |
6335 | int i, this_nid; | |
c713216d | 6336 | |
c13291a5 TH |
6337 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
6338 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
6339 | } |
6340 | ||
6341 | /** | |
6342 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
6343 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
6344 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
6345 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
6346 | * |
6347 | * It returns the start and end page frame of a node based on information | |
7d018176 | 6348 | * provided by memblock_set_node(). If called for a node |
c713216d | 6349 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 6350 | * PFNs will be 0. |
c713216d | 6351 | */ |
bbe5d993 | 6352 | void __init get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
6353 | unsigned long *start_pfn, unsigned long *end_pfn) |
6354 | { | |
c13291a5 | 6355 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 6356 | int i; |
c13291a5 | 6357 | |
c713216d MG |
6358 | *start_pfn = -1UL; |
6359 | *end_pfn = 0; | |
6360 | ||
c13291a5 TH |
6361 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
6362 | *start_pfn = min(*start_pfn, this_start_pfn); | |
6363 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
6364 | } |
6365 | ||
633c0666 | 6366 | if (*start_pfn == -1UL) |
c713216d | 6367 | *start_pfn = 0; |
c713216d MG |
6368 | } |
6369 | ||
2a1e274a MG |
6370 | /* |
6371 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
6372 | * assumption is made that zones within a node are ordered in monotonic | |
6373 | * increasing memory addresses so that the "highest" populated zone is used | |
6374 | */ | |
b69a7288 | 6375 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
6376 | { |
6377 | int zone_index; | |
6378 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
6379 | if (zone_index == ZONE_MOVABLE) | |
6380 | continue; | |
6381 | ||
6382 | if (arch_zone_highest_possible_pfn[zone_index] > | |
6383 | arch_zone_lowest_possible_pfn[zone_index]) | |
6384 | break; | |
6385 | } | |
6386 | ||
6387 | VM_BUG_ON(zone_index == -1); | |
6388 | movable_zone = zone_index; | |
6389 | } | |
6390 | ||
6391 | /* | |
6392 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 6393 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
6394 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
6395 | * in each node depending on the size of each node and how evenly kernelcore | |
6396 | * is distributed. This helper function adjusts the zone ranges | |
6397 | * provided by the architecture for a given node by using the end of the | |
6398 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
6399 | * zones within a node are in order of monotonic increases memory addresses | |
6400 | */ | |
bbe5d993 | 6401 | static void __init adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
6402 | unsigned long zone_type, |
6403 | unsigned long node_start_pfn, | |
6404 | unsigned long node_end_pfn, | |
6405 | unsigned long *zone_start_pfn, | |
6406 | unsigned long *zone_end_pfn) | |
6407 | { | |
6408 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
6409 | if (zone_movable_pfn[nid]) { | |
6410 | /* Size ZONE_MOVABLE */ | |
6411 | if (zone_type == ZONE_MOVABLE) { | |
6412 | *zone_start_pfn = zone_movable_pfn[nid]; | |
6413 | *zone_end_pfn = min(node_end_pfn, | |
6414 | arch_zone_highest_possible_pfn[movable_zone]); | |
6415 | ||
e506b996 XQ |
6416 | /* Adjust for ZONE_MOVABLE starting within this range */ |
6417 | } else if (!mirrored_kernelcore && | |
6418 | *zone_start_pfn < zone_movable_pfn[nid] && | |
6419 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
6420 | *zone_end_pfn = zone_movable_pfn[nid]; | |
6421 | ||
2a1e274a MG |
6422 | /* Check if this whole range is within ZONE_MOVABLE */ |
6423 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
6424 | *zone_start_pfn = *zone_end_pfn; | |
6425 | } | |
6426 | } | |
6427 | ||
c713216d MG |
6428 | /* |
6429 | * Return the number of pages a zone spans in a node, including holes | |
6430 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
6431 | */ | |
bbe5d993 | 6432 | static unsigned long __init zone_spanned_pages_in_node(int nid, |
c713216d | 6433 | unsigned long zone_type, |
7960aedd ZY |
6434 | unsigned long node_start_pfn, |
6435 | unsigned long node_end_pfn, | |
d91749c1 | 6436 | unsigned long *zone_start_pfn, |
854e8848 | 6437 | unsigned long *zone_end_pfn) |
c713216d | 6438 | { |
299c83dc LF |
6439 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6440 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
b5685e92 | 6441 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6442 | if (!node_start_pfn && !node_end_pfn) |
6443 | return 0; | |
6444 | ||
7960aedd | 6445 | /* Get the start and end of the zone */ |
299c83dc LF |
6446 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6447 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
2a1e274a MG |
6448 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6449 | node_start_pfn, node_end_pfn, | |
d91749c1 | 6450 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
6451 | |
6452 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 6453 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
6454 | return 0; |
6455 | ||
6456 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
6457 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
6458 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
6459 | |
6460 | /* Return the spanned pages */ | |
d91749c1 | 6461 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
6462 | } |
6463 | ||
6464 | /* | |
6465 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 6466 | * then all holes in the requested range will be accounted for. |
c713216d | 6467 | */ |
bbe5d993 | 6468 | unsigned long __init __absent_pages_in_range(int nid, |
c713216d MG |
6469 | unsigned long range_start_pfn, |
6470 | unsigned long range_end_pfn) | |
6471 | { | |
96e907d1 TH |
6472 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
6473 | unsigned long start_pfn, end_pfn; | |
6474 | int i; | |
c713216d | 6475 | |
96e907d1 TH |
6476 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6477 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
6478 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
6479 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 6480 | } |
96e907d1 | 6481 | return nr_absent; |
c713216d MG |
6482 | } |
6483 | ||
6484 | /** | |
6485 | * absent_pages_in_range - Return number of page frames in holes within a range | |
6486 | * @start_pfn: The start PFN to start searching for holes | |
6487 | * @end_pfn: The end PFN to stop searching for holes | |
6488 | * | |
a862f68a | 6489 | * Return: the number of pages frames in memory holes within a range. |
c713216d MG |
6490 | */ |
6491 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
6492 | unsigned long end_pfn) | |
6493 | { | |
6494 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
6495 | } | |
6496 | ||
6497 | /* Return the number of page frames in holes in a zone on a node */ | |
bbe5d993 | 6498 | static unsigned long __init zone_absent_pages_in_node(int nid, |
c713216d | 6499 | unsigned long zone_type, |
7960aedd | 6500 | unsigned long node_start_pfn, |
854e8848 | 6501 | unsigned long node_end_pfn) |
c713216d | 6502 | { |
96e907d1 TH |
6503 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
6504 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 6505 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 6506 | unsigned long nr_absent; |
9c7cd687 | 6507 | |
b5685e92 | 6508 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
6509 | if (!node_start_pfn && !node_end_pfn) |
6510 | return 0; | |
6511 | ||
96e907d1 TH |
6512 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
6513 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 6514 | |
2a1e274a MG |
6515 | adjust_zone_range_for_zone_movable(nid, zone_type, |
6516 | node_start_pfn, node_end_pfn, | |
6517 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
6518 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
6519 | ||
6520 | /* | |
6521 | * ZONE_MOVABLE handling. | |
6522 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
6523 | * and vice versa. | |
6524 | */ | |
e506b996 XQ |
6525 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
6526 | unsigned long start_pfn, end_pfn; | |
6527 | struct memblock_region *r; | |
6528 | ||
6529 | for_each_memblock(memory, r) { | |
6530 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
6531 | zone_start_pfn, zone_end_pfn); | |
6532 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
6533 | zone_start_pfn, zone_end_pfn); | |
6534 | ||
6535 | if (zone_type == ZONE_MOVABLE && | |
6536 | memblock_is_mirror(r)) | |
6537 | nr_absent += end_pfn - start_pfn; | |
6538 | ||
6539 | if (zone_type == ZONE_NORMAL && | |
6540 | !memblock_is_mirror(r)) | |
6541 | nr_absent += end_pfn - start_pfn; | |
342332e6 TI |
6542 | } |
6543 | } | |
6544 | ||
6545 | return nr_absent; | |
c713216d | 6546 | } |
0e0b864e | 6547 | |
bbe5d993 | 6548 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd | 6549 | unsigned long node_start_pfn, |
854e8848 | 6550 | unsigned long node_end_pfn) |
c713216d | 6551 | { |
febd5949 | 6552 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
6553 | enum zone_type i; |
6554 | ||
febd5949 GZ |
6555 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6556 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 6557 | unsigned long zone_start_pfn, zone_end_pfn; |
3f08a302 | 6558 | unsigned long spanned, absent; |
febd5949 | 6559 | unsigned long size, real_size; |
c713216d | 6560 | |
854e8848 MR |
6561 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i, |
6562 | node_start_pfn, | |
6563 | node_end_pfn, | |
6564 | &zone_start_pfn, | |
6565 | &zone_end_pfn); | |
6566 | absent = zone_absent_pages_in_node(pgdat->node_id, i, | |
6567 | node_start_pfn, | |
6568 | node_end_pfn); | |
3f08a302 MR |
6569 | |
6570 | size = spanned; | |
6571 | real_size = size - absent; | |
6572 | ||
d91749c1 TI |
6573 | if (size) |
6574 | zone->zone_start_pfn = zone_start_pfn; | |
6575 | else | |
6576 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
6577 | zone->spanned_pages = size; |
6578 | zone->present_pages = real_size; | |
6579 | ||
6580 | totalpages += size; | |
6581 | realtotalpages += real_size; | |
6582 | } | |
6583 | ||
6584 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
6585 | pgdat->node_present_pages = realtotalpages; |
6586 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
6587 | realtotalpages); | |
6588 | } | |
6589 | ||
835c134e MG |
6590 | #ifndef CONFIG_SPARSEMEM |
6591 | /* | |
6592 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
6593 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
6594 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
6595 | * round what is now in bits to nearest long in bits, then return it in |
6596 | * bytes. | |
6597 | */ | |
7c45512d | 6598 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
6599 | { |
6600 | unsigned long usemapsize; | |
6601 | ||
7c45512d | 6602 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
6603 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
6604 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
6605 | usemapsize *= NR_PAGEBLOCK_BITS; |
6606 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
6607 | ||
6608 | return usemapsize / 8; | |
6609 | } | |
6610 | ||
7cc2a959 | 6611 | static void __ref setup_usemap(struct pglist_data *pgdat, |
7c45512d LT |
6612 | struct zone *zone, |
6613 | unsigned long zone_start_pfn, | |
6614 | unsigned long zonesize) | |
835c134e | 6615 | { |
7c45512d | 6616 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 6617 | zone->pageblock_flags = NULL; |
23a7052a | 6618 | if (usemapsize) { |
6782832e | 6619 | zone->pageblock_flags = |
26fb3dae MR |
6620 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES, |
6621 | pgdat->node_id); | |
23a7052a MR |
6622 | if (!zone->pageblock_flags) |
6623 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n", | |
6624 | usemapsize, zone->name, pgdat->node_id); | |
6625 | } | |
835c134e MG |
6626 | } |
6627 | #else | |
7c45512d LT |
6628 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
6629 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
6630 | #endif /* CONFIG_SPARSEMEM */ |
6631 | ||
d9c23400 | 6632 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 6633 | |
d9c23400 | 6634 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
03e85f9d | 6635 | void __init set_pageblock_order(void) |
d9c23400 | 6636 | { |
955c1cd7 AM |
6637 | unsigned int order; |
6638 | ||
d9c23400 MG |
6639 | /* Check that pageblock_nr_pages has not already been setup */ |
6640 | if (pageblock_order) | |
6641 | return; | |
6642 | ||
955c1cd7 AM |
6643 | if (HPAGE_SHIFT > PAGE_SHIFT) |
6644 | order = HUGETLB_PAGE_ORDER; | |
6645 | else | |
6646 | order = MAX_ORDER - 1; | |
6647 | ||
d9c23400 MG |
6648 | /* |
6649 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
6650 | * This value may be variable depending on boot parameters on IA64 and |
6651 | * powerpc. | |
d9c23400 MG |
6652 | */ |
6653 | pageblock_order = order; | |
6654 | } | |
6655 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6656 | ||
ba72cb8c MG |
6657 | /* |
6658 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
6659 | * is unused as pageblock_order is set at compile-time. See |
6660 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
6661 | * the kernel config | |
ba72cb8c | 6662 | */ |
03e85f9d | 6663 | void __init set_pageblock_order(void) |
ba72cb8c | 6664 | { |
ba72cb8c | 6665 | } |
d9c23400 MG |
6666 | |
6667 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
6668 | ||
03e85f9d | 6669 | static unsigned long __init calc_memmap_size(unsigned long spanned_pages, |
7cc2a959 | 6670 | unsigned long present_pages) |
01cefaef JL |
6671 | { |
6672 | unsigned long pages = spanned_pages; | |
6673 | ||
6674 | /* | |
6675 | * Provide a more accurate estimation if there are holes within | |
6676 | * the zone and SPARSEMEM is in use. If there are holes within the | |
6677 | * zone, each populated memory region may cost us one or two extra | |
6678 | * memmap pages due to alignment because memmap pages for each | |
89d790ab | 6679 | * populated regions may not be naturally aligned on page boundary. |
01cefaef JL |
6680 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
6681 | */ | |
6682 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
6683 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
6684 | pages = present_pages; | |
6685 | ||
6686 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
6687 | } | |
6688 | ||
ace1db39 OS |
6689 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6690 | static void pgdat_init_split_queue(struct pglist_data *pgdat) | |
6691 | { | |
364c1eeb YS |
6692 | struct deferred_split *ds_queue = &pgdat->deferred_split_queue; |
6693 | ||
6694 | spin_lock_init(&ds_queue->split_queue_lock); | |
6695 | INIT_LIST_HEAD(&ds_queue->split_queue); | |
6696 | ds_queue->split_queue_len = 0; | |
ace1db39 OS |
6697 | } |
6698 | #else | |
6699 | static void pgdat_init_split_queue(struct pglist_data *pgdat) {} | |
6700 | #endif | |
6701 | ||
6702 | #ifdef CONFIG_COMPACTION | |
6703 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) | |
6704 | { | |
6705 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
6706 | } | |
6707 | #else | |
6708 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {} | |
6709 | #endif | |
6710 | ||
03e85f9d | 6711 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat) |
1da177e4 | 6712 | { |
208d54e5 | 6713 | pgdat_resize_init(pgdat); |
ace1db39 | 6714 | |
ace1db39 OS |
6715 | pgdat_init_split_queue(pgdat); |
6716 | pgdat_init_kcompactd(pgdat); | |
6717 | ||
1da177e4 | 6718 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 6719 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
ace1db39 | 6720 | |
eefa864b | 6721 | pgdat_page_ext_init(pgdat); |
a52633d8 | 6722 | spin_lock_init(&pgdat->lru_lock); |
867e5e1d | 6723 | lruvec_init(&pgdat->__lruvec); |
03e85f9d OS |
6724 | } |
6725 | ||
6726 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid, | |
6727 | unsigned long remaining_pages) | |
6728 | { | |
9705bea5 | 6729 | atomic_long_set(&zone->managed_pages, remaining_pages); |
03e85f9d OS |
6730 | zone_set_nid(zone, nid); |
6731 | zone->name = zone_names[idx]; | |
6732 | zone->zone_pgdat = NODE_DATA(nid); | |
6733 | spin_lock_init(&zone->lock); | |
6734 | zone_seqlock_init(zone); | |
6735 | zone_pcp_init(zone); | |
6736 | } | |
6737 | ||
6738 | /* | |
6739 | * Set up the zone data structures | |
6740 | * - init pgdat internals | |
6741 | * - init all zones belonging to this node | |
6742 | * | |
6743 | * NOTE: this function is only called during memory hotplug | |
6744 | */ | |
6745 | #ifdef CONFIG_MEMORY_HOTPLUG | |
6746 | void __ref free_area_init_core_hotplug(int nid) | |
6747 | { | |
6748 | enum zone_type z; | |
6749 | pg_data_t *pgdat = NODE_DATA(nid); | |
6750 | ||
6751 | pgdat_init_internals(pgdat); | |
6752 | for (z = 0; z < MAX_NR_ZONES; z++) | |
6753 | zone_init_internals(&pgdat->node_zones[z], z, nid, 0); | |
6754 | } | |
6755 | #endif | |
6756 | ||
6757 | /* | |
6758 | * Set up the zone data structures: | |
6759 | * - mark all pages reserved | |
6760 | * - mark all memory queues empty | |
6761 | * - clear the memory bitmaps | |
6762 | * | |
6763 | * NOTE: pgdat should get zeroed by caller. | |
6764 | * NOTE: this function is only called during early init. | |
6765 | */ | |
6766 | static void __init free_area_init_core(struct pglist_data *pgdat) | |
6767 | { | |
6768 | enum zone_type j; | |
6769 | int nid = pgdat->node_id; | |
5f63b720 | 6770 | |
03e85f9d | 6771 | pgdat_init_internals(pgdat); |
385386cf JW |
6772 | pgdat->per_cpu_nodestats = &boot_nodestats; |
6773 | ||
1da177e4 LT |
6774 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6775 | struct zone *zone = pgdat->node_zones + j; | |
e6943859 | 6776 | unsigned long size, freesize, memmap_pages; |
d91749c1 | 6777 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 6778 | |
febd5949 | 6779 | size = zone->spanned_pages; |
e6943859 | 6780 | freesize = zone->present_pages; |
1da177e4 | 6781 | |
0e0b864e | 6782 | /* |
9feedc9d | 6783 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
6784 | * is used by this zone for memmap. This affects the watermark |
6785 | * and per-cpu initialisations | |
6786 | */ | |
e6943859 | 6787 | memmap_pages = calc_memmap_size(size, freesize); |
ba914f48 ZH |
6788 | if (!is_highmem_idx(j)) { |
6789 | if (freesize >= memmap_pages) { | |
6790 | freesize -= memmap_pages; | |
6791 | if (memmap_pages) | |
6792 | printk(KERN_DEBUG | |
6793 | " %s zone: %lu pages used for memmap\n", | |
6794 | zone_names[j], memmap_pages); | |
6795 | } else | |
1170532b | 6796 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
6797 | zone_names[j], memmap_pages, freesize); |
6798 | } | |
0e0b864e | 6799 | |
6267276f | 6800 | /* Account for reserved pages */ |
9feedc9d JL |
6801 | if (j == 0 && freesize > dma_reserve) { |
6802 | freesize -= dma_reserve; | |
d903ef9f | 6803 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 6804 | zone_names[0], dma_reserve); |
0e0b864e MG |
6805 | } |
6806 | ||
98d2b0eb | 6807 | if (!is_highmem_idx(j)) |
9feedc9d | 6808 | nr_kernel_pages += freesize; |
01cefaef JL |
6809 | /* Charge for highmem memmap if there are enough kernel pages */ |
6810 | else if (nr_kernel_pages > memmap_pages * 2) | |
6811 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 6812 | nr_all_pages += freesize; |
1da177e4 | 6813 | |
9feedc9d JL |
6814 | /* |
6815 | * Set an approximate value for lowmem here, it will be adjusted | |
6816 | * when the bootmem allocator frees pages into the buddy system. | |
6817 | * And all highmem pages will be managed by the buddy system. | |
6818 | */ | |
03e85f9d | 6819 | zone_init_internals(zone, j, nid, freesize); |
81c0a2bb | 6820 | |
d883c6cf | 6821 | if (!size) |
1da177e4 LT |
6822 | continue; |
6823 | ||
955c1cd7 | 6824 | set_pageblock_order(); |
d883c6cf JK |
6825 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
6826 | init_currently_empty_zone(zone, zone_start_pfn, size); | |
76cdd58e | 6827 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
6828 | } |
6829 | } | |
6830 | ||
0cd842f9 | 6831 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bd721ea7 | 6832 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 6833 | { |
b0aeba74 | 6834 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
6835 | unsigned long __maybe_unused offset = 0; |
6836 | ||
1da177e4 LT |
6837 | /* Skip empty nodes */ |
6838 | if (!pgdat->node_spanned_pages) | |
6839 | return; | |
6840 | ||
b0aeba74 TL |
6841 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6842 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
6843 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6844 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 6845 | unsigned long size, end; |
d41dee36 AW |
6846 | struct page *map; |
6847 | ||
e984bb43 BP |
6848 | /* |
6849 | * The zone's endpoints aren't required to be MAX_ORDER | |
6850 | * aligned but the node_mem_map endpoints must be in order | |
6851 | * for the buddy allocator to function correctly. | |
6852 | */ | |
108bcc96 | 6853 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
6854 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6855 | size = (end - start) * sizeof(struct page); | |
26fb3dae MR |
6856 | map = memblock_alloc_node(size, SMP_CACHE_BYTES, |
6857 | pgdat->node_id); | |
23a7052a MR |
6858 | if (!map) |
6859 | panic("Failed to allocate %ld bytes for node %d memory map\n", | |
6860 | size, pgdat->node_id); | |
a1c34a3b | 6861 | pgdat->node_mem_map = map + offset; |
1da177e4 | 6862 | } |
0cd842f9 OS |
6863 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n", |
6864 | __func__, pgdat->node_id, (unsigned long)pgdat, | |
6865 | (unsigned long)pgdat->node_mem_map); | |
12d810c1 | 6866 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
6867 | /* |
6868 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
6869 | */ | |
c713216d | 6870 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 6871 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d | 6872 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 6873 | mem_map -= offset; |
c713216d | 6874 | } |
1da177e4 LT |
6875 | #endif |
6876 | } | |
0cd842f9 OS |
6877 | #else |
6878 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { } | |
6879 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
1da177e4 | 6880 | |
0188dc98 OS |
6881 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
6882 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) | |
6883 | { | |
0188dc98 OS |
6884 | pgdat->first_deferred_pfn = ULONG_MAX; |
6885 | } | |
6886 | #else | |
6887 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {} | |
6888 | #endif | |
6889 | ||
854e8848 | 6890 | static void __init free_area_init_node(int nid) |
1da177e4 | 6891 | { |
9109fb7b | 6892 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
6893 | unsigned long start_pfn = 0; |
6894 | unsigned long end_pfn = 0; | |
9109fb7b | 6895 | |
88fdf75d | 6896 | /* pg_data_t should be reset to zero when it's allocated */ |
38087d9b | 6897 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
88fdf75d | 6898 | |
854e8848 MR |
6899 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
6900 | ||
1da177e4 | 6901 | pgdat->node_id = nid; |
854e8848 | 6902 | pgdat->node_start_pfn = start_pfn; |
75ef7184 | 6903 | pgdat->per_cpu_nodestats = NULL; |
854e8848 MR |
6904 | |
6905 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, | |
6906 | (u64)start_pfn << PAGE_SHIFT, | |
6907 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
6908 | calculate_node_totalpages(pgdat, start_pfn, end_pfn); | |
1da177e4 LT |
6909 | |
6910 | alloc_node_mem_map(pgdat); | |
0188dc98 | 6911 | pgdat_set_deferred_range(pgdat); |
1da177e4 | 6912 | |
7f3eb55b | 6913 | free_area_init_core(pgdat); |
1da177e4 LT |
6914 | } |
6915 | ||
bc9331a1 | 6916 | void __init free_area_init_memoryless_node(int nid) |
3f08a302 | 6917 | { |
854e8848 | 6918 | free_area_init_node(nid); |
3f08a302 MR |
6919 | } |
6920 | ||
aca52c39 | 6921 | #if !defined(CONFIG_FLAT_NODE_MEM_MAP) |
ec393a0f | 6922 | /* |
4b094b78 DH |
6923 | * Initialize all valid struct pages in the range [spfn, epfn) and mark them |
6924 | * PageReserved(). Return the number of struct pages that were initialized. | |
ec393a0f | 6925 | */ |
4b094b78 | 6926 | static u64 __init init_unavailable_range(unsigned long spfn, unsigned long epfn) |
ec393a0f PT |
6927 | { |
6928 | unsigned long pfn; | |
6929 | u64 pgcnt = 0; | |
6930 | ||
6931 | for (pfn = spfn; pfn < epfn; pfn++) { | |
6932 | if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) { | |
6933 | pfn = ALIGN_DOWN(pfn, pageblock_nr_pages) | |
6934 | + pageblock_nr_pages - 1; | |
6935 | continue; | |
6936 | } | |
4b094b78 DH |
6937 | /* |
6938 | * Use a fake node/zone (0) for now. Some of these pages | |
6939 | * (in memblock.reserved but not in memblock.memory) will | |
6940 | * get re-initialized via reserve_bootmem_region() later. | |
6941 | */ | |
6942 | __init_single_page(pfn_to_page(pfn), pfn, 0, 0); | |
6943 | __SetPageReserved(pfn_to_page(pfn)); | |
ec393a0f PT |
6944 | pgcnt++; |
6945 | } | |
6946 | ||
6947 | return pgcnt; | |
6948 | } | |
6949 | ||
a4a3ede2 PT |
6950 | /* |
6951 | * Only struct pages that are backed by physical memory are zeroed and | |
6952 | * initialized by going through __init_single_page(). But, there are some | |
6953 | * struct pages which are reserved in memblock allocator and their fields | |
6954 | * may be accessed (for example page_to_pfn() on some configuration accesses | |
4b094b78 | 6955 | * flags). We must explicitly initialize those struct pages. |
907ec5fc NH |
6956 | * |
6957 | * This function also addresses a similar issue where struct pages are left | |
6958 | * uninitialized because the physical address range is not covered by | |
6959 | * memblock.memory or memblock.reserved. That could happen when memblock | |
e822969c DH |
6960 | * layout is manually configured via memmap=, or when the highest physical |
6961 | * address (max_pfn) does not end on a section boundary. | |
a4a3ede2 | 6962 | */ |
4b094b78 | 6963 | static void __init init_unavailable_mem(void) |
a4a3ede2 PT |
6964 | { |
6965 | phys_addr_t start, end; | |
a4a3ede2 | 6966 | u64 i, pgcnt; |
907ec5fc | 6967 | phys_addr_t next = 0; |
a4a3ede2 PT |
6968 | |
6969 | /* | |
907ec5fc | 6970 | * Loop through unavailable ranges not covered by memblock.memory. |
a4a3ede2 PT |
6971 | */ |
6972 | pgcnt = 0; | |
907ec5fc NH |
6973 | for_each_mem_range(i, &memblock.memory, NULL, |
6974 | NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) { | |
ec393a0f | 6975 | if (next < start) |
4b094b78 DH |
6976 | pgcnt += init_unavailable_range(PFN_DOWN(next), |
6977 | PFN_UP(start)); | |
907ec5fc NH |
6978 | next = end; |
6979 | } | |
e822969c DH |
6980 | |
6981 | /* | |
6982 | * Early sections always have a fully populated memmap for the whole | |
6983 | * section - see pfn_valid(). If the last section has holes at the | |
6984 | * end and that section is marked "online", the memmap will be | |
6985 | * considered initialized. Make sure that memmap has a well defined | |
6986 | * state. | |
6987 | */ | |
4b094b78 DH |
6988 | pgcnt += init_unavailable_range(PFN_DOWN(next), |
6989 | round_up(max_pfn, PAGES_PER_SECTION)); | |
907ec5fc | 6990 | |
a4a3ede2 PT |
6991 | /* |
6992 | * Struct pages that do not have backing memory. This could be because | |
6993 | * firmware is using some of this memory, or for some other reasons. | |
a4a3ede2 PT |
6994 | */ |
6995 | if (pgcnt) | |
907ec5fc | 6996 | pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt); |
a4a3ede2 | 6997 | } |
4b094b78 DH |
6998 | #else |
6999 | static inline void __init init_unavailable_mem(void) | |
7000 | { | |
7001 | } | |
aca52c39 | 7002 | #endif /* !CONFIG_FLAT_NODE_MEM_MAP */ |
a4a3ede2 | 7003 | |
418508c1 MS |
7004 | #if MAX_NUMNODES > 1 |
7005 | /* | |
7006 | * Figure out the number of possible node ids. | |
7007 | */ | |
f9872caf | 7008 | void __init setup_nr_node_ids(void) |
418508c1 | 7009 | { |
904a9553 | 7010 | unsigned int highest; |
418508c1 | 7011 | |
904a9553 | 7012 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
7013 | nr_node_ids = highest + 1; |
7014 | } | |
418508c1 MS |
7015 | #endif |
7016 | ||
1e01979c TH |
7017 | /** |
7018 | * node_map_pfn_alignment - determine the maximum internode alignment | |
7019 | * | |
7020 | * This function should be called after node map is populated and sorted. | |
7021 | * It calculates the maximum power of two alignment which can distinguish | |
7022 | * all the nodes. | |
7023 | * | |
7024 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
7025 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
7026 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
7027 | * shifted, 1GiB is enough and this function will indicate so. | |
7028 | * | |
7029 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
7030 | * model has fine enough granularity to avoid incorrect mapping for the | |
7031 | * populated node map. | |
7032 | * | |
a862f68a | 7033 | * Return: the determined alignment in pfn's. 0 if there is no alignment |
1e01979c TH |
7034 | * requirement (single node). |
7035 | */ | |
7036 | unsigned long __init node_map_pfn_alignment(void) | |
7037 | { | |
7038 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 7039 | unsigned long start, end, mask; |
98fa15f3 | 7040 | int last_nid = NUMA_NO_NODE; |
c13291a5 | 7041 | int i, nid; |
1e01979c | 7042 | |
c13291a5 | 7043 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
7044 | if (!start || last_nid < 0 || last_nid == nid) { |
7045 | last_nid = nid; | |
7046 | last_end = end; | |
7047 | continue; | |
7048 | } | |
7049 | ||
7050 | /* | |
7051 | * Start with a mask granular enough to pin-point to the | |
7052 | * start pfn and tick off bits one-by-one until it becomes | |
7053 | * too coarse to separate the current node from the last. | |
7054 | */ | |
7055 | mask = ~((1 << __ffs(start)) - 1); | |
7056 | while (mask && last_end <= (start & (mask << 1))) | |
7057 | mask <<= 1; | |
7058 | ||
7059 | /* accumulate all internode masks */ | |
7060 | accl_mask |= mask; | |
7061 | } | |
7062 | ||
7063 | /* convert mask to number of pages */ | |
7064 | return ~accl_mask + 1; | |
7065 | } | |
7066 | ||
c713216d MG |
7067 | /** |
7068 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
7069 | * | |
a862f68a | 7070 | * Return: the minimum PFN based on information provided via |
7d018176 | 7071 | * memblock_set_node(). |
c713216d MG |
7072 | */ |
7073 | unsigned long __init find_min_pfn_with_active_regions(void) | |
7074 | { | |
8a1b25fe | 7075 | return PHYS_PFN(memblock_start_of_DRAM()); |
c713216d MG |
7076 | } |
7077 | ||
37b07e41 LS |
7078 | /* |
7079 | * early_calculate_totalpages() | |
7080 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 7081 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 7082 | */ |
484f51f8 | 7083 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 7084 | { |
7e63efef | 7085 | unsigned long totalpages = 0; |
c13291a5 TH |
7086 | unsigned long start_pfn, end_pfn; |
7087 | int i, nid; | |
7088 | ||
7089 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
7090 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 7091 | |
37b07e41 LS |
7092 | totalpages += pages; |
7093 | if (pages) | |
4b0ef1fe | 7094 | node_set_state(nid, N_MEMORY); |
37b07e41 | 7095 | } |
b8af2941 | 7096 | return totalpages; |
7e63efef MG |
7097 | } |
7098 | ||
2a1e274a MG |
7099 | /* |
7100 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
7101 | * is spread evenly between nodes as long as the nodes have enough | |
7102 | * memory. When they don't, some nodes will have more kernelcore than | |
7103 | * others | |
7104 | */ | |
b224ef85 | 7105 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
7106 | { |
7107 | int i, nid; | |
7108 | unsigned long usable_startpfn; | |
7109 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 7110 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 7111 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 7112 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 7113 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 7114 | struct memblock_region *r; |
b2f3eebe TC |
7115 | |
7116 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
7117 | find_usable_zone_for_movable(); | |
7118 | ||
7119 | /* | |
7120 | * If movable_node is specified, ignore kernelcore and movablecore | |
7121 | * options. | |
7122 | */ | |
7123 | if (movable_node_is_enabled()) { | |
136199f0 EM |
7124 | for_each_memblock(memory, r) { |
7125 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
7126 | continue; |
7127 | ||
d622abf7 | 7128 | nid = memblock_get_region_node(r); |
b2f3eebe | 7129 | |
136199f0 | 7130 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
7131 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
7132 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7133 | usable_startpfn; | |
7134 | } | |
7135 | ||
7136 | goto out2; | |
7137 | } | |
2a1e274a | 7138 | |
342332e6 TI |
7139 | /* |
7140 | * If kernelcore=mirror is specified, ignore movablecore option | |
7141 | */ | |
7142 | if (mirrored_kernelcore) { | |
7143 | bool mem_below_4gb_not_mirrored = false; | |
7144 | ||
7145 | for_each_memblock(memory, r) { | |
7146 | if (memblock_is_mirror(r)) | |
7147 | continue; | |
7148 | ||
d622abf7 | 7149 | nid = memblock_get_region_node(r); |
342332e6 TI |
7150 | |
7151 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
7152 | ||
7153 | if (usable_startpfn < 0x100000) { | |
7154 | mem_below_4gb_not_mirrored = true; | |
7155 | continue; | |
7156 | } | |
7157 | ||
7158 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
7159 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
7160 | usable_startpfn; | |
7161 | } | |
7162 | ||
7163 | if (mem_below_4gb_not_mirrored) | |
7164 | pr_warn("This configuration results in unmirrored kernel memory."); | |
7165 | ||
7166 | goto out2; | |
7167 | } | |
7168 | ||
7e63efef | 7169 | /* |
a5c6d650 DR |
7170 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the |
7171 | * amount of necessary memory. | |
7172 | */ | |
7173 | if (required_kernelcore_percent) | |
7174 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) / | |
7175 | 10000UL; | |
7176 | if (required_movablecore_percent) | |
7177 | required_movablecore = (totalpages * 100 * required_movablecore_percent) / | |
7178 | 10000UL; | |
7179 | ||
7180 | /* | |
7181 | * If movablecore= was specified, calculate what size of | |
7e63efef MG |
7182 | * kernelcore that corresponds so that memory usable for |
7183 | * any allocation type is evenly spread. If both kernelcore | |
7184 | * and movablecore are specified, then the value of kernelcore | |
7185 | * will be used for required_kernelcore if it's greater than | |
7186 | * what movablecore would have allowed. | |
7187 | */ | |
7188 | if (required_movablecore) { | |
7e63efef MG |
7189 | unsigned long corepages; |
7190 | ||
7191 | /* | |
7192 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
7193 | * was requested by the user | |
7194 | */ | |
7195 | required_movablecore = | |
7196 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 7197 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
7198 | corepages = totalpages - required_movablecore; |
7199 | ||
7200 | required_kernelcore = max(required_kernelcore, corepages); | |
7201 | } | |
7202 | ||
bde304bd XQ |
7203 | /* |
7204 | * If kernelcore was not specified or kernelcore size is larger | |
7205 | * than totalpages, there is no ZONE_MOVABLE. | |
7206 | */ | |
7207 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 7208 | goto out; |
2a1e274a MG |
7209 | |
7210 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
7211 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
7212 | ||
7213 | restart: | |
7214 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
7215 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 7216 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
7217 | unsigned long start_pfn, end_pfn; |
7218 | ||
2a1e274a MG |
7219 | /* |
7220 | * Recalculate kernelcore_node if the division per node | |
7221 | * now exceeds what is necessary to satisfy the requested | |
7222 | * amount of memory for the kernel | |
7223 | */ | |
7224 | if (required_kernelcore < kernelcore_node) | |
7225 | kernelcore_node = required_kernelcore / usable_nodes; | |
7226 | ||
7227 | /* | |
7228 | * As the map is walked, we track how much memory is usable | |
7229 | * by the kernel using kernelcore_remaining. When it is | |
7230 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
7231 | */ | |
7232 | kernelcore_remaining = kernelcore_node; | |
7233 | ||
7234 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 7235 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
7236 | unsigned long size_pages; |
7237 | ||
c13291a5 | 7238 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
7239 | if (start_pfn >= end_pfn) |
7240 | continue; | |
7241 | ||
7242 | /* Account for what is only usable for kernelcore */ | |
7243 | if (start_pfn < usable_startpfn) { | |
7244 | unsigned long kernel_pages; | |
7245 | kernel_pages = min(end_pfn, usable_startpfn) | |
7246 | - start_pfn; | |
7247 | ||
7248 | kernelcore_remaining -= min(kernel_pages, | |
7249 | kernelcore_remaining); | |
7250 | required_kernelcore -= min(kernel_pages, | |
7251 | required_kernelcore); | |
7252 | ||
7253 | /* Continue if range is now fully accounted */ | |
7254 | if (end_pfn <= usable_startpfn) { | |
7255 | ||
7256 | /* | |
7257 | * Push zone_movable_pfn to the end so | |
7258 | * that if we have to rebalance | |
7259 | * kernelcore across nodes, we will | |
7260 | * not double account here | |
7261 | */ | |
7262 | zone_movable_pfn[nid] = end_pfn; | |
7263 | continue; | |
7264 | } | |
7265 | start_pfn = usable_startpfn; | |
7266 | } | |
7267 | ||
7268 | /* | |
7269 | * The usable PFN range for ZONE_MOVABLE is from | |
7270 | * start_pfn->end_pfn. Calculate size_pages as the | |
7271 | * number of pages used as kernelcore | |
7272 | */ | |
7273 | size_pages = end_pfn - start_pfn; | |
7274 | if (size_pages > kernelcore_remaining) | |
7275 | size_pages = kernelcore_remaining; | |
7276 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
7277 | ||
7278 | /* | |
7279 | * Some kernelcore has been met, update counts and | |
7280 | * break if the kernelcore for this node has been | |
b8af2941 | 7281 | * satisfied |
2a1e274a MG |
7282 | */ |
7283 | required_kernelcore -= min(required_kernelcore, | |
7284 | size_pages); | |
7285 | kernelcore_remaining -= size_pages; | |
7286 | if (!kernelcore_remaining) | |
7287 | break; | |
7288 | } | |
7289 | } | |
7290 | ||
7291 | /* | |
7292 | * If there is still required_kernelcore, we do another pass with one | |
7293 | * less node in the count. This will push zone_movable_pfn[nid] further | |
7294 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 7295 | * satisfied |
2a1e274a MG |
7296 | */ |
7297 | usable_nodes--; | |
7298 | if (usable_nodes && required_kernelcore > usable_nodes) | |
7299 | goto restart; | |
7300 | ||
b2f3eebe | 7301 | out2: |
2a1e274a MG |
7302 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
7303 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
7304 | zone_movable_pfn[nid] = | |
7305 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 7306 | |
20e6926d | 7307 | out: |
66918dcd | 7308 | /* restore the node_state */ |
4b0ef1fe | 7309 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
7310 | } |
7311 | ||
4b0ef1fe LJ |
7312 | /* Any regular or high memory on that node ? */ |
7313 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 7314 | { |
37b07e41 LS |
7315 | enum zone_type zone_type; |
7316 | ||
4b0ef1fe | 7317 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
37b07e41 | 7318 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 7319 | if (populated_zone(zone)) { |
7b0e0c0e OS |
7320 | if (IS_ENABLED(CONFIG_HIGHMEM)) |
7321 | node_set_state(nid, N_HIGH_MEMORY); | |
7322 | if (zone_type <= ZONE_NORMAL) | |
4b0ef1fe | 7323 | node_set_state(nid, N_NORMAL_MEMORY); |
d0048b0e BL |
7324 | break; |
7325 | } | |
37b07e41 | 7326 | } |
37b07e41 LS |
7327 | } |
7328 | ||
51930df5 MR |
7329 | /* |
7330 | * Some architecturs, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For | |
7331 | * such cases we allow max_zone_pfn sorted in the descending order | |
7332 | */ | |
7333 | bool __weak arch_has_descending_max_zone_pfns(void) | |
7334 | { | |
7335 | return false; | |
7336 | } | |
7337 | ||
c713216d | 7338 | /** |
9691a071 | 7339 | * free_area_init - Initialise all pg_data_t and zone data |
88ca3b94 | 7340 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
7341 | * |
7342 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 7343 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
7344 | * zone in each node and their holes is calculated. If the maximum PFN |
7345 | * between two adjacent zones match, it is assumed that the zone is empty. | |
7346 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
7347 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
7348 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
7349 | * at arch_max_dma_pfn. | |
7350 | */ | |
9691a071 | 7351 | void __init free_area_init(unsigned long *max_zone_pfn) |
c713216d | 7352 | { |
c13291a5 | 7353 | unsigned long start_pfn, end_pfn; |
51930df5 MR |
7354 | int i, nid, zone; |
7355 | bool descending; | |
a6af2bc3 | 7356 | |
c713216d MG |
7357 | /* Record where the zone boundaries are */ |
7358 | memset(arch_zone_lowest_possible_pfn, 0, | |
7359 | sizeof(arch_zone_lowest_possible_pfn)); | |
7360 | memset(arch_zone_highest_possible_pfn, 0, | |
7361 | sizeof(arch_zone_highest_possible_pfn)); | |
90cae1fe OH |
7362 | |
7363 | start_pfn = find_min_pfn_with_active_regions(); | |
51930df5 | 7364 | descending = arch_has_descending_max_zone_pfns(); |
90cae1fe OH |
7365 | |
7366 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
51930df5 MR |
7367 | if (descending) |
7368 | zone = MAX_NR_ZONES - i - 1; | |
7369 | else | |
7370 | zone = i; | |
7371 | ||
7372 | if (zone == ZONE_MOVABLE) | |
2a1e274a | 7373 | continue; |
90cae1fe | 7374 | |
51930df5 MR |
7375 | end_pfn = max(max_zone_pfn[zone], start_pfn); |
7376 | arch_zone_lowest_possible_pfn[zone] = start_pfn; | |
7377 | arch_zone_highest_possible_pfn[zone] = end_pfn; | |
90cae1fe OH |
7378 | |
7379 | start_pfn = end_pfn; | |
c713216d | 7380 | } |
2a1e274a MG |
7381 | |
7382 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
7383 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 7384 | find_zone_movable_pfns_for_nodes(); |
c713216d | 7385 | |
c713216d | 7386 | /* Print out the zone ranges */ |
f88dfff5 | 7387 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
7388 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7389 | if (i == ZONE_MOVABLE) | |
7390 | continue; | |
f88dfff5 | 7391 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
7392 | if (arch_zone_lowest_possible_pfn[i] == |
7393 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 7394 | pr_cont("empty\n"); |
72f0ba02 | 7395 | else |
8d29e18a JG |
7396 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
7397 | (u64)arch_zone_lowest_possible_pfn[i] | |
7398 | << PAGE_SHIFT, | |
7399 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 7400 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
7401 | } |
7402 | ||
7403 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 7404 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
7405 | for (i = 0; i < MAX_NUMNODES; i++) { |
7406 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
7407 | pr_info(" Node %d: %#018Lx\n", i, |
7408 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 7409 | } |
c713216d | 7410 | |
f46edbd1 DW |
7411 | /* |
7412 | * Print out the early node map, and initialize the | |
7413 | * subsection-map relative to active online memory ranges to | |
7414 | * enable future "sub-section" extensions of the memory map. | |
7415 | */ | |
f88dfff5 | 7416 | pr_info("Early memory node ranges\n"); |
f46edbd1 | 7417 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
8d29e18a JG |
7418 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
7419 | (u64)start_pfn << PAGE_SHIFT, | |
7420 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
f46edbd1 DW |
7421 | subsection_map_init(start_pfn, end_pfn - start_pfn); |
7422 | } | |
c713216d MG |
7423 | |
7424 | /* Initialise every node */ | |
708614e6 | 7425 | mminit_verify_pageflags_layout(); |
8ef82866 | 7426 | setup_nr_node_ids(); |
4b094b78 | 7427 | init_unavailable_mem(); |
c713216d MG |
7428 | for_each_online_node(nid) { |
7429 | pg_data_t *pgdat = NODE_DATA(nid); | |
854e8848 | 7430 | free_area_init_node(nid); |
37b07e41 LS |
7431 | |
7432 | /* Any memory on that node */ | |
7433 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
7434 | node_set_state(nid, N_MEMORY); |
7435 | check_for_memory(pgdat, nid); | |
c713216d MG |
7436 | } |
7437 | } | |
2a1e274a | 7438 | |
a5c6d650 DR |
7439 | static int __init cmdline_parse_core(char *p, unsigned long *core, |
7440 | unsigned long *percent) | |
2a1e274a MG |
7441 | { |
7442 | unsigned long long coremem; | |
a5c6d650 DR |
7443 | char *endptr; |
7444 | ||
2a1e274a MG |
7445 | if (!p) |
7446 | return -EINVAL; | |
7447 | ||
a5c6d650 DR |
7448 | /* Value may be a percentage of total memory, otherwise bytes */ |
7449 | coremem = simple_strtoull(p, &endptr, 0); | |
7450 | if (*endptr == '%') { | |
7451 | /* Paranoid check for percent values greater than 100 */ | |
7452 | WARN_ON(coremem > 100); | |
2a1e274a | 7453 | |
a5c6d650 DR |
7454 | *percent = coremem; |
7455 | } else { | |
7456 | coremem = memparse(p, &p); | |
7457 | /* Paranoid check that UL is enough for the coremem value */ | |
7458 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
2a1e274a | 7459 | |
a5c6d650 DR |
7460 | *core = coremem >> PAGE_SHIFT; |
7461 | *percent = 0UL; | |
7462 | } | |
2a1e274a MG |
7463 | return 0; |
7464 | } | |
ed7ed365 | 7465 | |
7e63efef MG |
7466 | /* |
7467 | * kernelcore=size sets the amount of memory for use for allocations that | |
7468 | * cannot be reclaimed or migrated. | |
7469 | */ | |
7470 | static int __init cmdline_parse_kernelcore(char *p) | |
7471 | { | |
342332e6 TI |
7472 | /* parse kernelcore=mirror */ |
7473 | if (parse_option_str(p, "mirror")) { | |
7474 | mirrored_kernelcore = true; | |
7475 | return 0; | |
7476 | } | |
7477 | ||
a5c6d650 DR |
7478 | return cmdline_parse_core(p, &required_kernelcore, |
7479 | &required_kernelcore_percent); | |
7e63efef MG |
7480 | } |
7481 | ||
7482 | /* | |
7483 | * movablecore=size sets the amount of memory for use for allocations that | |
7484 | * can be reclaimed or migrated. | |
7485 | */ | |
7486 | static int __init cmdline_parse_movablecore(char *p) | |
7487 | { | |
a5c6d650 DR |
7488 | return cmdline_parse_core(p, &required_movablecore, |
7489 | &required_movablecore_percent); | |
7e63efef MG |
7490 | } |
7491 | ||
ed7ed365 | 7492 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 7493 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 7494 | |
c3d5f5f0 JL |
7495 | void adjust_managed_page_count(struct page *page, long count) |
7496 | { | |
9705bea5 | 7497 | atomic_long_add(count, &page_zone(page)->managed_pages); |
ca79b0c2 | 7498 | totalram_pages_add(count); |
3dcc0571 JL |
7499 | #ifdef CONFIG_HIGHMEM |
7500 | if (PageHighMem(page)) | |
ca79b0c2 | 7501 | totalhigh_pages_add(count); |
3dcc0571 | 7502 | #endif |
c3d5f5f0 | 7503 | } |
3dcc0571 | 7504 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 7505 | |
e5cb113f | 7506 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s) |
69afade7 | 7507 | { |
11199692 JL |
7508 | void *pos; |
7509 | unsigned long pages = 0; | |
69afade7 | 7510 | |
11199692 JL |
7511 | start = (void *)PAGE_ALIGN((unsigned long)start); |
7512 | end = (void *)((unsigned long)end & PAGE_MASK); | |
7513 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
0d834328 DH |
7514 | struct page *page = virt_to_page(pos); |
7515 | void *direct_map_addr; | |
7516 | ||
7517 | /* | |
7518 | * 'direct_map_addr' might be different from 'pos' | |
7519 | * because some architectures' virt_to_page() | |
7520 | * work with aliases. Getting the direct map | |
7521 | * address ensures that we get a _writeable_ | |
7522 | * alias for the memset(). | |
7523 | */ | |
7524 | direct_map_addr = page_address(page); | |
dbe67df4 | 7525 | if ((unsigned int)poison <= 0xFF) |
0d834328 DH |
7526 | memset(direct_map_addr, poison, PAGE_SIZE); |
7527 | ||
7528 | free_reserved_page(page); | |
69afade7 JL |
7529 | } |
7530 | ||
7531 | if (pages && s) | |
adb1fe9a JP |
7532 | pr_info("Freeing %s memory: %ldK\n", |
7533 | s, pages << (PAGE_SHIFT - 10)); | |
69afade7 JL |
7534 | |
7535 | return pages; | |
7536 | } | |
7537 | ||
cfa11e08 JL |
7538 | #ifdef CONFIG_HIGHMEM |
7539 | void free_highmem_page(struct page *page) | |
7540 | { | |
7541 | __free_reserved_page(page); | |
ca79b0c2 | 7542 | totalram_pages_inc(); |
9705bea5 | 7543 | atomic_long_inc(&page_zone(page)->managed_pages); |
ca79b0c2 | 7544 | totalhigh_pages_inc(); |
cfa11e08 JL |
7545 | } |
7546 | #endif | |
7547 | ||
7ee3d4e8 JL |
7548 | |
7549 | void __init mem_init_print_info(const char *str) | |
7550 | { | |
7551 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
7552 | unsigned long init_code_size, init_data_size; | |
7553 | ||
7554 | physpages = get_num_physpages(); | |
7555 | codesize = _etext - _stext; | |
7556 | datasize = _edata - _sdata; | |
7557 | rosize = __end_rodata - __start_rodata; | |
7558 | bss_size = __bss_stop - __bss_start; | |
7559 | init_data_size = __init_end - __init_begin; | |
7560 | init_code_size = _einittext - _sinittext; | |
7561 | ||
7562 | /* | |
7563 | * Detect special cases and adjust section sizes accordingly: | |
7564 | * 1) .init.* may be embedded into .data sections | |
7565 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
7566 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
7567 | * 3) .rodata.* may be embedded into .text or .data sections. | |
7568 | */ | |
7569 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
7570 | do { \ |
7571 | if (start <= pos && pos < end && size > adj) \ | |
7572 | size -= adj; \ | |
7573 | } while (0) | |
7ee3d4e8 JL |
7574 | |
7575 | adj_init_size(__init_begin, __init_end, init_data_size, | |
7576 | _sinittext, init_code_size); | |
7577 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
7578 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
7579 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
7580 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
7581 | ||
7582 | #undef adj_init_size | |
7583 | ||
756a025f | 7584 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 7585 | #ifdef CONFIG_HIGHMEM |
756a025f | 7586 | ", %luK highmem" |
7ee3d4e8 | 7587 | #endif |
756a025f JP |
7588 | "%s%s)\n", |
7589 | nr_free_pages() << (PAGE_SHIFT - 10), | |
7590 | physpages << (PAGE_SHIFT - 10), | |
7591 | codesize >> 10, datasize >> 10, rosize >> 10, | |
7592 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
ca79b0c2 | 7593 | (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10), |
756a025f | 7594 | totalcma_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7595 | #ifdef CONFIG_HIGHMEM |
ca79b0c2 | 7596 | totalhigh_pages() << (PAGE_SHIFT - 10), |
7ee3d4e8 | 7597 | #endif |
756a025f | 7598 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
7599 | } |
7600 | ||
0e0b864e | 7601 | /** |
88ca3b94 RD |
7602 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
7603 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 7604 | * |
013110a7 | 7605 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
7606 | * In the DMA zone, a significant percentage may be consumed by kernel image |
7607 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
7608 | * function may optionally be used to account for unfreeable pages in the |
7609 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
7610 | * smaller per-cpu batchsize. | |
0e0b864e MG |
7611 | */ |
7612 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
7613 | { | |
7614 | dma_reserve = new_dma_reserve; | |
7615 | } | |
7616 | ||
005fd4bb | 7617 | static int page_alloc_cpu_dead(unsigned int cpu) |
1da177e4 | 7618 | { |
1da177e4 | 7619 | |
005fd4bb SAS |
7620 | lru_add_drain_cpu(cpu); |
7621 | drain_pages(cpu); | |
9f8f2172 | 7622 | |
005fd4bb SAS |
7623 | /* |
7624 | * Spill the event counters of the dead processor | |
7625 | * into the current processors event counters. | |
7626 | * This artificially elevates the count of the current | |
7627 | * processor. | |
7628 | */ | |
7629 | vm_events_fold_cpu(cpu); | |
9f8f2172 | 7630 | |
005fd4bb SAS |
7631 | /* |
7632 | * Zero the differential counters of the dead processor | |
7633 | * so that the vm statistics are consistent. | |
7634 | * | |
7635 | * This is only okay since the processor is dead and cannot | |
7636 | * race with what we are doing. | |
7637 | */ | |
7638 | cpu_vm_stats_fold(cpu); | |
7639 | return 0; | |
1da177e4 | 7640 | } |
1da177e4 | 7641 | |
e03a5125 NP |
7642 | #ifdef CONFIG_NUMA |
7643 | int hashdist = HASHDIST_DEFAULT; | |
7644 | ||
7645 | static int __init set_hashdist(char *str) | |
7646 | { | |
7647 | if (!str) | |
7648 | return 0; | |
7649 | hashdist = simple_strtoul(str, &str, 0); | |
7650 | return 1; | |
7651 | } | |
7652 | __setup("hashdist=", set_hashdist); | |
7653 | #endif | |
7654 | ||
1da177e4 LT |
7655 | void __init page_alloc_init(void) |
7656 | { | |
005fd4bb SAS |
7657 | int ret; |
7658 | ||
e03a5125 NP |
7659 | #ifdef CONFIG_NUMA |
7660 | if (num_node_state(N_MEMORY) == 1) | |
7661 | hashdist = 0; | |
7662 | #endif | |
7663 | ||
005fd4bb SAS |
7664 | ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, |
7665 | "mm/page_alloc:dead", NULL, | |
7666 | page_alloc_cpu_dead); | |
7667 | WARN_ON(ret < 0); | |
1da177e4 LT |
7668 | } |
7669 | ||
cb45b0e9 | 7670 | /* |
34b10060 | 7671 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
7672 | * or min_free_kbytes changes. |
7673 | */ | |
7674 | static void calculate_totalreserve_pages(void) | |
7675 | { | |
7676 | struct pglist_data *pgdat; | |
7677 | unsigned long reserve_pages = 0; | |
2f6726e5 | 7678 | enum zone_type i, j; |
cb45b0e9 HA |
7679 | |
7680 | for_each_online_pgdat(pgdat) { | |
281e3726 MG |
7681 | |
7682 | pgdat->totalreserve_pages = 0; | |
7683 | ||
cb45b0e9 HA |
7684 | for (i = 0; i < MAX_NR_ZONES; i++) { |
7685 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 7686 | long max = 0; |
9705bea5 | 7687 | unsigned long managed_pages = zone_managed_pages(zone); |
cb45b0e9 HA |
7688 | |
7689 | /* Find valid and maximum lowmem_reserve in the zone */ | |
7690 | for (j = i; j < MAX_NR_ZONES; j++) { | |
7691 | if (zone->lowmem_reserve[j] > max) | |
7692 | max = zone->lowmem_reserve[j]; | |
7693 | } | |
7694 | ||
41858966 MG |
7695 | /* we treat the high watermark as reserved pages. */ |
7696 | max += high_wmark_pages(zone); | |
cb45b0e9 | 7697 | |
3d6357de AK |
7698 | if (max > managed_pages) |
7699 | max = managed_pages; | |
a8d01437 | 7700 | |
281e3726 | 7701 | pgdat->totalreserve_pages += max; |
a8d01437 | 7702 | |
cb45b0e9 HA |
7703 | reserve_pages += max; |
7704 | } | |
7705 | } | |
7706 | totalreserve_pages = reserve_pages; | |
7707 | } | |
7708 | ||
1da177e4 LT |
7709 | /* |
7710 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 7711 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
7712 | * has a correct pages reserved value, so an adequate number of |
7713 | * pages are left in the zone after a successful __alloc_pages(). | |
7714 | */ | |
7715 | static void setup_per_zone_lowmem_reserve(void) | |
7716 | { | |
7717 | struct pglist_data *pgdat; | |
2f6726e5 | 7718 | enum zone_type j, idx; |
1da177e4 | 7719 | |
ec936fc5 | 7720 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
7721 | for (j = 0; j < MAX_NR_ZONES; j++) { |
7722 | struct zone *zone = pgdat->node_zones + j; | |
9705bea5 | 7723 | unsigned long managed_pages = zone_managed_pages(zone); |
1da177e4 LT |
7724 | |
7725 | zone->lowmem_reserve[j] = 0; | |
7726 | ||
2f6726e5 CL |
7727 | idx = j; |
7728 | while (idx) { | |
1da177e4 LT |
7729 | struct zone *lower_zone; |
7730 | ||
2f6726e5 | 7731 | idx--; |
1da177e4 | 7732 | lower_zone = pgdat->node_zones + idx; |
d3cda233 JK |
7733 | |
7734 | if (sysctl_lowmem_reserve_ratio[idx] < 1) { | |
7735 | sysctl_lowmem_reserve_ratio[idx] = 0; | |
7736 | lower_zone->lowmem_reserve[j] = 0; | |
7737 | } else { | |
7738 | lower_zone->lowmem_reserve[j] = | |
7739 | managed_pages / sysctl_lowmem_reserve_ratio[idx]; | |
7740 | } | |
9705bea5 | 7741 | managed_pages += zone_managed_pages(lower_zone); |
1da177e4 LT |
7742 | } |
7743 | } | |
7744 | } | |
cb45b0e9 HA |
7745 | |
7746 | /* update totalreserve_pages */ | |
7747 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7748 | } |
7749 | ||
cfd3da1e | 7750 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
7751 | { |
7752 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
7753 | unsigned long lowmem_pages = 0; | |
7754 | struct zone *zone; | |
7755 | unsigned long flags; | |
7756 | ||
7757 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
7758 | for_each_zone(zone) { | |
7759 | if (!is_highmem(zone)) | |
9705bea5 | 7760 | lowmem_pages += zone_managed_pages(zone); |
1da177e4 LT |
7761 | } |
7762 | ||
7763 | for_each_zone(zone) { | |
ac924c60 AM |
7764 | u64 tmp; |
7765 | ||
1125b4e3 | 7766 | spin_lock_irqsave(&zone->lock, flags); |
9705bea5 | 7767 | tmp = (u64)pages_min * zone_managed_pages(zone); |
ac924c60 | 7768 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
7769 | if (is_highmem(zone)) { |
7770 | /* | |
669ed175 NP |
7771 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
7772 | * need highmem pages, so cap pages_min to a small | |
7773 | * value here. | |
7774 | * | |
41858966 | 7775 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
8bb4e7a2 | 7776 | * deltas control async page reclaim, and so should |
669ed175 | 7777 | * not be capped for highmem. |
1da177e4 | 7778 | */ |
90ae8d67 | 7779 | unsigned long min_pages; |
1da177e4 | 7780 | |
9705bea5 | 7781 | min_pages = zone_managed_pages(zone) / 1024; |
90ae8d67 | 7782 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
a9214443 | 7783 | zone->_watermark[WMARK_MIN] = min_pages; |
1da177e4 | 7784 | } else { |
669ed175 NP |
7785 | /* |
7786 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
7787 | * proportionate to the zone's size. |
7788 | */ | |
a9214443 | 7789 | zone->_watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
7790 | } |
7791 | ||
795ae7a0 JW |
7792 | /* |
7793 | * Set the kswapd watermarks distance according to the | |
7794 | * scale factor in proportion to available memory, but | |
7795 | * ensure a minimum size on small systems. | |
7796 | */ | |
7797 | tmp = max_t(u64, tmp >> 2, | |
9705bea5 | 7798 | mult_frac(zone_managed_pages(zone), |
795ae7a0 JW |
7799 | watermark_scale_factor, 10000)); |
7800 | ||
a9214443 MG |
7801 | zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; |
7802 | zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
1c30844d | 7803 | zone->watermark_boost = 0; |
49f223a9 | 7804 | |
1125b4e3 | 7805 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 7806 | } |
cb45b0e9 HA |
7807 | |
7808 | /* update totalreserve_pages */ | |
7809 | calculate_totalreserve_pages(); | |
1da177e4 LT |
7810 | } |
7811 | ||
cfd3da1e MG |
7812 | /** |
7813 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
7814 | * or when memory is hot-{added|removed} | |
7815 | * | |
7816 | * Ensures that the watermark[min,low,high] values for each zone are set | |
7817 | * correctly with respect to min_free_kbytes. | |
7818 | */ | |
7819 | void setup_per_zone_wmarks(void) | |
7820 | { | |
b93e0f32 MH |
7821 | static DEFINE_SPINLOCK(lock); |
7822 | ||
7823 | spin_lock(&lock); | |
cfd3da1e | 7824 | __setup_per_zone_wmarks(); |
b93e0f32 | 7825 | spin_unlock(&lock); |
cfd3da1e MG |
7826 | } |
7827 | ||
1da177e4 LT |
7828 | /* |
7829 | * Initialise min_free_kbytes. | |
7830 | * | |
7831 | * For small machines we want it small (128k min). For large machines | |
7832 | * we want it large (64MB max). But it is not linear, because network | |
7833 | * bandwidth does not increase linearly with machine size. We use | |
7834 | * | |
b8af2941 | 7835 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
7836 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
7837 | * | |
7838 | * which yields | |
7839 | * | |
7840 | * 16MB: 512k | |
7841 | * 32MB: 724k | |
7842 | * 64MB: 1024k | |
7843 | * 128MB: 1448k | |
7844 | * 256MB: 2048k | |
7845 | * 512MB: 2896k | |
7846 | * 1024MB: 4096k | |
7847 | * 2048MB: 5792k | |
7848 | * 4096MB: 8192k | |
7849 | * 8192MB: 11584k | |
7850 | * 16384MB: 16384k | |
7851 | */ | |
1b79acc9 | 7852 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
7853 | { |
7854 | unsigned long lowmem_kbytes; | |
5f12733e | 7855 | int new_min_free_kbytes; |
1da177e4 LT |
7856 | |
7857 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
7858 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
7859 | ||
7860 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
7861 | min_free_kbytes = new_min_free_kbytes; | |
7862 | if (min_free_kbytes < 128) | |
7863 | min_free_kbytes = 128; | |
ee8eb9a5 JS |
7864 | if (min_free_kbytes > 262144) |
7865 | min_free_kbytes = 262144; | |
5f12733e MH |
7866 | } else { |
7867 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
7868 | new_min_free_kbytes, user_min_free_kbytes); | |
7869 | } | |
bc75d33f | 7870 | setup_per_zone_wmarks(); |
a6cccdc3 | 7871 | refresh_zone_stat_thresholds(); |
1da177e4 | 7872 | setup_per_zone_lowmem_reserve(); |
6423aa81 JK |
7873 | |
7874 | #ifdef CONFIG_NUMA | |
7875 | setup_min_unmapped_ratio(); | |
7876 | setup_min_slab_ratio(); | |
7877 | #endif | |
7878 | ||
1da177e4 LT |
7879 | return 0; |
7880 | } | |
bc22af74 | 7881 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
7882 | |
7883 | /* | |
b8af2941 | 7884 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
7885 | * that we can call two helper functions whenever min_free_kbytes |
7886 | * changes. | |
7887 | */ | |
cccad5b9 | 7888 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7889 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7890 | { |
da8c757b HP |
7891 | int rc; |
7892 | ||
7893 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7894 | if (rc) | |
7895 | return rc; | |
7896 | ||
5f12733e MH |
7897 | if (write) { |
7898 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 7899 | setup_per_zone_wmarks(); |
5f12733e | 7900 | } |
1da177e4 LT |
7901 | return 0; |
7902 | } | |
7903 | ||
1c30844d MG |
7904 | int watermark_boost_factor_sysctl_handler(struct ctl_table *table, int write, |
7905 | void __user *buffer, size_t *length, loff_t *ppos) | |
7906 | { | |
7907 | int rc; | |
7908 | ||
7909 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7910 | if (rc) | |
7911 | return rc; | |
7912 | ||
7913 | return 0; | |
7914 | } | |
7915 | ||
795ae7a0 JW |
7916 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
7917 | void __user *buffer, size_t *length, loff_t *ppos) | |
7918 | { | |
7919 | int rc; | |
7920 | ||
7921 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7922 | if (rc) | |
7923 | return rc; | |
7924 | ||
7925 | if (write) | |
7926 | setup_per_zone_wmarks(); | |
7927 | ||
7928 | return 0; | |
7929 | } | |
7930 | ||
9614634f | 7931 | #ifdef CONFIG_NUMA |
6423aa81 | 7932 | static void setup_min_unmapped_ratio(void) |
9614634f | 7933 | { |
6423aa81 | 7934 | pg_data_t *pgdat; |
9614634f | 7935 | struct zone *zone; |
9614634f | 7936 | |
a5f5f91d | 7937 | for_each_online_pgdat(pgdat) |
81cbcbc2 | 7938 | pgdat->min_unmapped_pages = 0; |
a5f5f91d | 7939 | |
9614634f | 7940 | for_each_zone(zone) |
9705bea5 AK |
7941 | zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * |
7942 | sysctl_min_unmapped_ratio) / 100; | |
9614634f | 7943 | } |
0ff38490 | 7944 | |
6423aa81 JK |
7945 | |
7946 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, | |
8d65af78 | 7947 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 | 7948 | { |
0ff38490 CL |
7949 | int rc; |
7950 | ||
8d65af78 | 7951 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
7952 | if (rc) |
7953 | return rc; | |
7954 | ||
6423aa81 JK |
7955 | setup_min_unmapped_ratio(); |
7956 | ||
7957 | return 0; | |
7958 | } | |
7959 | ||
7960 | static void setup_min_slab_ratio(void) | |
7961 | { | |
7962 | pg_data_t *pgdat; | |
7963 | struct zone *zone; | |
7964 | ||
a5f5f91d MG |
7965 | for_each_online_pgdat(pgdat) |
7966 | pgdat->min_slab_pages = 0; | |
7967 | ||
0ff38490 | 7968 | for_each_zone(zone) |
9705bea5 AK |
7969 | zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * |
7970 | sysctl_min_slab_ratio) / 100; | |
6423aa81 JK |
7971 | } |
7972 | ||
7973 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, | |
7974 | void __user *buffer, size_t *length, loff_t *ppos) | |
7975 | { | |
7976 | int rc; | |
7977 | ||
7978 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
7979 | if (rc) | |
7980 | return rc; | |
7981 | ||
7982 | setup_min_slab_ratio(); | |
7983 | ||
0ff38490 CL |
7984 | return 0; |
7985 | } | |
9614634f CL |
7986 | #endif |
7987 | ||
1da177e4 LT |
7988 | /* |
7989 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
7990 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
7991 | * whenever sysctl_lowmem_reserve_ratio changes. | |
7992 | * | |
7993 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 7994 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
7995 | * if in function of the boot time zone sizes. |
7996 | */ | |
cccad5b9 | 7997 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 7998 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 7999 | { |
8d65af78 | 8000 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
8001 | setup_per_zone_lowmem_reserve(); |
8002 | return 0; | |
8003 | } | |
8004 | ||
cb1ef534 MG |
8005 | static void __zone_pcp_update(struct zone *zone) |
8006 | { | |
8007 | unsigned int cpu; | |
8008 | ||
8009 | for_each_possible_cpu(cpu) | |
8010 | pageset_set_high_and_batch(zone, | |
8011 | per_cpu_ptr(zone->pageset, cpu)); | |
8012 | } | |
8013 | ||
8ad4b1fb RS |
8014 | /* |
8015 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
8016 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
8017 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 8018 | */ |
cccad5b9 | 8019 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 8020 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
8021 | { |
8022 | struct zone *zone; | |
7cd2b0a3 | 8023 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
8024 | int ret; |
8025 | ||
7cd2b0a3 DR |
8026 | mutex_lock(&pcp_batch_high_lock); |
8027 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
8028 | ||
8d65af78 | 8029 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
8030 | if (!write || ret < 0) |
8031 | goto out; | |
8032 | ||
8033 | /* Sanity checking to avoid pcp imbalance */ | |
8034 | if (percpu_pagelist_fraction && | |
8035 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
8036 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
8037 | ret = -EINVAL; | |
8038 | goto out; | |
8039 | } | |
8040 | ||
8041 | /* No change? */ | |
8042 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
8043 | goto out; | |
c8e251fa | 8044 | |
cb1ef534 MG |
8045 | for_each_populated_zone(zone) |
8046 | __zone_pcp_update(zone); | |
7cd2b0a3 | 8047 | out: |
c8e251fa | 8048 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 8049 | return ret; |
8ad4b1fb RS |
8050 | } |
8051 | ||
f6f34b43 SD |
8052 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
8053 | /* | |
8054 | * Returns the number of pages that arch has reserved but | |
8055 | * is not known to alloc_large_system_hash(). | |
8056 | */ | |
8057 | static unsigned long __init arch_reserved_kernel_pages(void) | |
8058 | { | |
8059 | return 0; | |
8060 | } | |
8061 | #endif | |
8062 | ||
9017217b PT |
8063 | /* |
8064 | * Adaptive scale is meant to reduce sizes of hash tables on large memory | |
8065 | * machines. As memory size is increased the scale is also increased but at | |
8066 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory | |
8067 | * quadruples the scale is increased by one, which means the size of hash table | |
8068 | * only doubles, instead of quadrupling as well. | |
8069 | * Because 32-bit systems cannot have large physical memory, where this scaling | |
8070 | * makes sense, it is disabled on such platforms. | |
8071 | */ | |
8072 | #if __BITS_PER_LONG > 32 | |
8073 | #define ADAPT_SCALE_BASE (64ul << 30) | |
8074 | #define ADAPT_SCALE_SHIFT 2 | |
8075 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT) | |
8076 | #endif | |
8077 | ||
1da177e4 LT |
8078 | /* |
8079 | * allocate a large system hash table from bootmem | |
8080 | * - it is assumed that the hash table must contain an exact power-of-2 | |
8081 | * quantity of entries | |
8082 | * - limit is the number of hash buckets, not the total allocation size | |
8083 | */ | |
8084 | void *__init alloc_large_system_hash(const char *tablename, | |
8085 | unsigned long bucketsize, | |
8086 | unsigned long numentries, | |
8087 | int scale, | |
8088 | int flags, | |
8089 | unsigned int *_hash_shift, | |
8090 | unsigned int *_hash_mask, | |
31fe62b9 TB |
8091 | unsigned long low_limit, |
8092 | unsigned long high_limit) | |
1da177e4 | 8093 | { |
31fe62b9 | 8094 | unsigned long long max = high_limit; |
1da177e4 LT |
8095 | unsigned long log2qty, size; |
8096 | void *table = NULL; | |
3749a8f0 | 8097 | gfp_t gfp_flags; |
ec11408a | 8098 | bool virt; |
1da177e4 LT |
8099 | |
8100 | /* allow the kernel cmdline to have a say */ | |
8101 | if (!numentries) { | |
8102 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 8103 | numentries = nr_kernel_pages; |
f6f34b43 | 8104 | numentries -= arch_reserved_kernel_pages(); |
a7e83318 JZ |
8105 | |
8106 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
8107 | if (PAGE_SHIFT < 20) | |
8108 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 | 8109 | |
9017217b PT |
8110 | #if __BITS_PER_LONG > 32 |
8111 | if (!high_limit) { | |
8112 | unsigned long adapt; | |
8113 | ||
8114 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries; | |
8115 | adapt <<= ADAPT_SCALE_SHIFT) | |
8116 | scale++; | |
8117 | } | |
8118 | #endif | |
8119 | ||
1da177e4 LT |
8120 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
8121 | if (scale > PAGE_SHIFT) | |
8122 | numentries >>= (scale - PAGE_SHIFT); | |
8123 | else | |
8124 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
8125 | |
8126 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
8127 | if (unlikely(flags & HASH_SMALL)) { |
8128 | /* Makes no sense without HASH_EARLY */ | |
8129 | WARN_ON(!(flags & HASH_EARLY)); | |
8130 | if (!(numentries >> *_hash_shift)) { | |
8131 | numentries = 1UL << *_hash_shift; | |
8132 | BUG_ON(!numentries); | |
8133 | } | |
8134 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 8135 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 8136 | } |
6e692ed3 | 8137 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
8138 | |
8139 | /* limit allocation size to 1/16 total memory by default */ | |
8140 | if (max == 0) { | |
8141 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
8142 | do_div(max, bucketsize); | |
8143 | } | |
074b8517 | 8144 | max = min(max, 0x80000000ULL); |
1da177e4 | 8145 | |
31fe62b9 TB |
8146 | if (numentries < low_limit) |
8147 | numentries = low_limit; | |
1da177e4 LT |
8148 | if (numentries > max) |
8149 | numentries = max; | |
8150 | ||
f0d1b0b3 | 8151 | log2qty = ilog2(numentries); |
1da177e4 | 8152 | |
3749a8f0 | 8153 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC; |
1da177e4 | 8154 | do { |
ec11408a | 8155 | virt = false; |
1da177e4 | 8156 | size = bucketsize << log2qty; |
ea1f5f37 PT |
8157 | if (flags & HASH_EARLY) { |
8158 | if (flags & HASH_ZERO) | |
26fb3dae | 8159 | table = memblock_alloc(size, SMP_CACHE_BYTES); |
ea1f5f37 | 8160 | else |
7e1c4e27 MR |
8161 | table = memblock_alloc_raw(size, |
8162 | SMP_CACHE_BYTES); | |
ec11408a | 8163 | } else if (get_order(size) >= MAX_ORDER || hashdist) { |
88dca4ca | 8164 | table = __vmalloc(size, gfp_flags); |
ec11408a | 8165 | virt = true; |
ea1f5f37 | 8166 | } else { |
1037b83b ED |
8167 | /* |
8168 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
8169 | * some pages at the end of hash table which |
8170 | * alloc_pages_exact() automatically does | |
1037b83b | 8171 | */ |
ec11408a NP |
8172 | table = alloc_pages_exact(size, gfp_flags); |
8173 | kmemleak_alloc(table, size, 1, gfp_flags); | |
1da177e4 LT |
8174 | } |
8175 | } while (!table && size > PAGE_SIZE && --log2qty); | |
8176 | ||
8177 | if (!table) | |
8178 | panic("Failed to allocate %s hash table\n", tablename); | |
8179 | ||
ec11408a NP |
8180 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n", |
8181 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size, | |
8182 | virt ? "vmalloc" : "linear"); | |
1da177e4 LT |
8183 | |
8184 | if (_hash_shift) | |
8185 | *_hash_shift = log2qty; | |
8186 | if (_hash_mask) | |
8187 | *_hash_mask = (1 << log2qty) - 1; | |
8188 | ||
8189 | return table; | |
8190 | } | |
a117e66e | 8191 | |
a5d76b54 | 8192 | /* |
80934513 | 8193 | * This function checks whether pageblock includes unmovable pages or not. |
80934513 | 8194 | * |
b8af2941 | 8195 | * PageLRU check without isolation or lru_lock could race so that |
0efadf48 YX |
8196 | * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable |
8197 | * check without lock_page also may miss some movable non-lru pages at | |
8198 | * race condition. So you can't expect this function should be exact. | |
4a55c047 QC |
8199 | * |
8200 | * Returns a page without holding a reference. If the caller wants to | |
8201 | * dereference that page (e.g., dumping), it has to make sure that that it | |
8202 | * cannot get removed (e.g., via memory unplug) concurrently. | |
8203 | * | |
a5d76b54 | 8204 | */ |
4a55c047 QC |
8205 | struct page *has_unmovable_pages(struct zone *zone, struct page *page, |
8206 | int migratetype, int flags) | |
49ac8255 | 8207 | { |
1a9f2191 QC |
8208 | unsigned long iter = 0; |
8209 | unsigned long pfn = page_to_pfn(page); | |
47118af0 | 8210 | |
49ac8255 | 8211 | /* |
15c30bc0 MH |
8212 | * TODO we could make this much more efficient by not checking every |
8213 | * page in the range if we know all of them are in MOVABLE_ZONE and | |
8214 | * that the movable zone guarantees that pages are migratable but | |
8215 | * the later is not the case right now unfortunatelly. E.g. movablecore | |
8216 | * can still lead to having bootmem allocations in zone_movable. | |
49ac8255 | 8217 | */ |
49ac8255 | 8218 | |
1a9f2191 QC |
8219 | if (is_migrate_cma_page(page)) { |
8220 | /* | |
8221 | * CMA allocations (alloc_contig_range) really need to mark | |
8222 | * isolate CMA pageblocks even when they are not movable in fact | |
8223 | * so consider them movable here. | |
8224 | */ | |
8225 | if (is_migrate_cma(migratetype)) | |
4a55c047 | 8226 | return NULL; |
1a9f2191 | 8227 | |
3d680bdf | 8228 | return page; |
1a9f2191 | 8229 | } |
4da2ce25 | 8230 | |
fe4c86c9 DH |
8231 | for (; iter < pageblock_nr_pages; iter++) { |
8232 | if (!pfn_valid_within(pfn + iter)) | |
49ac8255 | 8233 | continue; |
29723fcc | 8234 | |
fe4c86c9 | 8235 | page = pfn_to_page(pfn + iter); |
c8721bbb | 8236 | |
d7ab3672 | 8237 | if (PageReserved(page)) |
3d680bdf | 8238 | return page; |
d7ab3672 | 8239 | |
9d789999 MH |
8240 | /* |
8241 | * If the zone is movable and we have ruled out all reserved | |
8242 | * pages then it should be reasonably safe to assume the rest | |
8243 | * is movable. | |
8244 | */ | |
8245 | if (zone_idx(zone) == ZONE_MOVABLE) | |
8246 | continue; | |
8247 | ||
c8721bbb NH |
8248 | /* |
8249 | * Hugepages are not in LRU lists, but they're movable. | |
1da2f328 | 8250 | * THPs are on the LRU, but need to be counted as #small pages. |
8bb4e7a2 | 8251 | * We need not scan over tail pages because we don't |
c8721bbb NH |
8252 | * handle each tail page individually in migration. |
8253 | */ | |
1da2f328 | 8254 | if (PageHuge(page) || PageTransCompound(page)) { |
17e2e7d7 OS |
8255 | struct page *head = compound_head(page); |
8256 | unsigned int skip_pages; | |
464c7ffb | 8257 | |
1da2f328 RR |
8258 | if (PageHuge(page)) { |
8259 | if (!hugepage_migration_supported(page_hstate(head))) | |
8260 | return page; | |
8261 | } else if (!PageLRU(head) && !__PageMovable(head)) { | |
3d680bdf | 8262 | return page; |
1da2f328 | 8263 | } |
464c7ffb | 8264 | |
d8c6546b | 8265 | skip_pages = compound_nr(head) - (page - head); |
17e2e7d7 | 8266 | iter += skip_pages - 1; |
c8721bbb NH |
8267 | continue; |
8268 | } | |
8269 | ||
97d255c8 MK |
8270 | /* |
8271 | * We can't use page_count without pin a page | |
8272 | * because another CPU can free compound page. | |
8273 | * This check already skips compound tails of THP | |
0139aa7b | 8274 | * because their page->_refcount is zero at all time. |
97d255c8 | 8275 | */ |
fe896d18 | 8276 | if (!page_ref_count(page)) { |
49ac8255 KH |
8277 | if (PageBuddy(page)) |
8278 | iter += (1 << page_order(page)) - 1; | |
8279 | continue; | |
8280 | } | |
97d255c8 | 8281 | |
b023f468 WC |
8282 | /* |
8283 | * The HWPoisoned page may be not in buddy system, and | |
8284 | * page_count() is not 0. | |
8285 | */ | |
756d25be | 8286 | if ((flags & MEMORY_OFFLINE) && PageHWPoison(page)) |
b023f468 WC |
8287 | continue; |
8288 | ||
fe4c86c9 | 8289 | if (__PageMovable(page) || PageLRU(page)) |
0efadf48 YX |
8290 | continue; |
8291 | ||
49ac8255 | 8292 | /* |
6b4f7799 JW |
8293 | * If there are RECLAIMABLE pages, we need to check |
8294 | * it. But now, memory offline itself doesn't call | |
8295 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
8296 | */ |
8297 | /* | |
8298 | * If the page is not RAM, page_count()should be 0. | |
8299 | * we don't need more check. This is an _used_ not-movable page. | |
8300 | * | |
8301 | * The problematic thing here is PG_reserved pages. PG_reserved | |
8302 | * is set to both of a memory hole page and a _used_ kernel | |
8303 | * page at boot. | |
8304 | */ | |
3d680bdf | 8305 | return page; |
49ac8255 | 8306 | } |
4a55c047 | 8307 | return NULL; |
49ac8255 KH |
8308 | } |
8309 | ||
8df995f6 | 8310 | #ifdef CONFIG_CONTIG_ALLOC |
041d3a8c MN |
8311 | static unsigned long pfn_max_align_down(unsigned long pfn) |
8312 | { | |
8313 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8314 | pageblock_nr_pages) - 1); | |
8315 | } | |
8316 | ||
8317 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
8318 | { | |
8319 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
8320 | pageblock_nr_pages)); | |
8321 | } | |
8322 | ||
041d3a8c | 8323 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
8324 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
8325 | unsigned long start, unsigned long end) | |
041d3a8c MN |
8326 | { |
8327 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 8328 | unsigned long nr_reclaimed; |
041d3a8c MN |
8329 | unsigned long pfn = start; |
8330 | unsigned int tries = 0; | |
8331 | int ret = 0; | |
8332 | ||
be49a6e1 | 8333 | migrate_prep(); |
041d3a8c | 8334 | |
bb13ffeb | 8335 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
8336 | if (fatal_signal_pending(current)) { |
8337 | ret = -EINTR; | |
8338 | break; | |
8339 | } | |
8340 | ||
bb13ffeb MG |
8341 | if (list_empty(&cc->migratepages)) { |
8342 | cc->nr_migratepages = 0; | |
edc2ca61 | 8343 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
8344 | if (!pfn) { |
8345 | ret = -EINTR; | |
8346 | break; | |
8347 | } | |
8348 | tries = 0; | |
8349 | } else if (++tries == 5) { | |
8350 | ret = ret < 0 ? ret : -EBUSY; | |
8351 | break; | |
8352 | } | |
8353 | ||
beb51eaa MK |
8354 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
8355 | &cc->migratepages); | |
8356 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 8357 | |
9c620e2b | 8358 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
31025351 | 8359 | NULL, 0, cc->mode, MR_CONTIG_RANGE); |
041d3a8c | 8360 | } |
2a6f5124 SP |
8361 | if (ret < 0) { |
8362 | putback_movable_pages(&cc->migratepages); | |
8363 | return ret; | |
8364 | } | |
8365 | return 0; | |
041d3a8c MN |
8366 | } |
8367 | ||
8368 | /** | |
8369 | * alloc_contig_range() -- tries to allocate given range of pages | |
8370 | * @start: start PFN to allocate | |
8371 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
8372 | * @migratetype: migratetype of the underlaying pageblocks (either |
8373 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
8374 | * in range must have the same migratetype and it must | |
8375 | * be either of the two. | |
ca96b625 | 8376 | * @gfp_mask: GFP mask to use during compaction |
041d3a8c MN |
8377 | * |
8378 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
2c7452a0 | 8379 | * aligned. The PFN range must belong to a single zone. |
041d3a8c | 8380 | * |
2c7452a0 MK |
8381 | * The first thing this routine does is attempt to MIGRATE_ISOLATE all |
8382 | * pageblocks in the range. Once isolated, the pageblocks should not | |
8383 | * be modified by others. | |
041d3a8c | 8384 | * |
a862f68a | 8385 | * Return: zero on success or negative error code. On success all |
041d3a8c MN |
8386 | * pages which PFN is in [start, end) are allocated for the caller and |
8387 | * need to be freed with free_contig_range(). | |
8388 | */ | |
0815f3d8 | 8389 | int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 8390 | unsigned migratetype, gfp_t gfp_mask) |
041d3a8c | 8391 | { |
041d3a8c | 8392 | unsigned long outer_start, outer_end; |
d00181b9 KS |
8393 | unsigned int order; |
8394 | int ret = 0; | |
041d3a8c | 8395 | |
bb13ffeb MG |
8396 | struct compact_control cc = { |
8397 | .nr_migratepages = 0, | |
8398 | .order = -1, | |
8399 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 8400 | .mode = MIGRATE_SYNC, |
bb13ffeb | 8401 | .ignore_skip_hint = true, |
2583d671 | 8402 | .no_set_skip_hint = true, |
7dea19f9 | 8403 | .gfp_mask = current_gfp_context(gfp_mask), |
b06eda09 | 8404 | .alloc_contig = true, |
bb13ffeb MG |
8405 | }; |
8406 | INIT_LIST_HEAD(&cc.migratepages); | |
8407 | ||
041d3a8c MN |
8408 | /* |
8409 | * What we do here is we mark all pageblocks in range as | |
8410 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
8411 | * have different sizes, and due to the way page allocator | |
8412 | * work, we align the range to biggest of the two pages so | |
8413 | * that page allocator won't try to merge buddies from | |
8414 | * different pageblocks and change MIGRATE_ISOLATE to some | |
8415 | * other migration type. | |
8416 | * | |
8417 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
8418 | * migrate the pages from an unaligned range (ie. pages that | |
8419 | * we are interested in). This will put all the pages in | |
8420 | * range back to page allocator as MIGRATE_ISOLATE. | |
8421 | * | |
8422 | * When this is done, we take the pages in range from page | |
8423 | * allocator removing them from the buddy system. This way | |
8424 | * page allocator will never consider using them. | |
8425 | * | |
8426 | * This lets us mark the pageblocks back as | |
8427 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
8428 | * aligned range but not in the unaligned, original range are | |
8429 | * put back to page allocator so that buddy can use them. | |
8430 | */ | |
8431 | ||
8432 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
d381c547 | 8433 | pfn_max_align_up(end), migratetype, 0); |
9b7ea46a | 8434 | if (ret < 0) |
86a595f9 | 8435 | return ret; |
041d3a8c | 8436 | |
8ef5849f JK |
8437 | /* |
8438 | * In case of -EBUSY, we'd like to know which page causes problem. | |
63cd4489 MK |
8439 | * So, just fall through. test_pages_isolated() has a tracepoint |
8440 | * which will report the busy page. | |
8441 | * | |
8442 | * It is possible that busy pages could become available before | |
8443 | * the call to test_pages_isolated, and the range will actually be | |
8444 | * allocated. So, if we fall through be sure to clear ret so that | |
8445 | * -EBUSY is not accidentally used or returned to caller. | |
8ef5849f | 8446 | */ |
bb13ffeb | 8447 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 8448 | if (ret && ret != -EBUSY) |
041d3a8c | 8449 | goto done; |
63cd4489 | 8450 | ret =0; |
041d3a8c MN |
8451 | |
8452 | /* | |
8453 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
8454 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
8455 | * more, all pages in [start, end) are free in page allocator. | |
8456 | * What we are going to do is to allocate all pages from | |
8457 | * [start, end) (that is remove them from page allocator). | |
8458 | * | |
8459 | * The only problem is that pages at the beginning and at the | |
8460 | * end of interesting range may be not aligned with pages that | |
8461 | * page allocator holds, ie. they can be part of higher order | |
8462 | * pages. Because of this, we reserve the bigger range and | |
8463 | * once this is done free the pages we are not interested in. | |
8464 | * | |
8465 | * We don't have to hold zone->lock here because the pages are | |
8466 | * isolated thus they won't get removed from buddy. | |
8467 | */ | |
8468 | ||
8469 | lru_add_drain_all(); | |
041d3a8c MN |
8470 | |
8471 | order = 0; | |
8472 | outer_start = start; | |
8473 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
8474 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
8475 | outer_start = start; |
8476 | break; | |
041d3a8c MN |
8477 | } |
8478 | outer_start &= ~0UL << order; | |
8479 | } | |
8480 | ||
8ef5849f JK |
8481 | if (outer_start != start) { |
8482 | order = page_order(pfn_to_page(outer_start)); | |
8483 | ||
8484 | /* | |
8485 | * outer_start page could be small order buddy page and | |
8486 | * it doesn't include start page. Adjust outer_start | |
8487 | * in this case to report failed page properly | |
8488 | * on tracepoint in test_pages_isolated() | |
8489 | */ | |
8490 | if (outer_start + (1UL << order) <= start) | |
8491 | outer_start = start; | |
8492 | } | |
8493 | ||
041d3a8c | 8494 | /* Make sure the range is really isolated. */ |
756d25be | 8495 | if (test_pages_isolated(outer_start, end, 0)) { |
75dddef3 | 8496 | pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n", |
dae803e1 | 8497 | __func__, outer_start, end); |
041d3a8c MN |
8498 | ret = -EBUSY; |
8499 | goto done; | |
8500 | } | |
8501 | ||
49f223a9 | 8502 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 8503 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
8504 | if (!outer_end) { |
8505 | ret = -EBUSY; | |
8506 | goto done; | |
8507 | } | |
8508 | ||
8509 | /* Free head and tail (if any) */ | |
8510 | if (start != outer_start) | |
8511 | free_contig_range(outer_start, start - outer_start); | |
8512 | if (end != outer_end) | |
8513 | free_contig_range(end, outer_end - end); | |
8514 | ||
8515 | done: | |
8516 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 8517 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
8518 | return ret; |
8519 | } | |
5e27a2df AK |
8520 | |
8521 | static int __alloc_contig_pages(unsigned long start_pfn, | |
8522 | unsigned long nr_pages, gfp_t gfp_mask) | |
8523 | { | |
8524 | unsigned long end_pfn = start_pfn + nr_pages; | |
8525 | ||
8526 | return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE, | |
8527 | gfp_mask); | |
8528 | } | |
8529 | ||
8530 | static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn, | |
8531 | unsigned long nr_pages) | |
8532 | { | |
8533 | unsigned long i, end_pfn = start_pfn + nr_pages; | |
8534 | struct page *page; | |
8535 | ||
8536 | for (i = start_pfn; i < end_pfn; i++) { | |
8537 | page = pfn_to_online_page(i); | |
8538 | if (!page) | |
8539 | return false; | |
8540 | ||
8541 | if (page_zone(page) != z) | |
8542 | return false; | |
8543 | ||
8544 | if (PageReserved(page)) | |
8545 | return false; | |
8546 | ||
8547 | if (page_count(page) > 0) | |
8548 | return false; | |
8549 | ||
8550 | if (PageHuge(page)) | |
8551 | return false; | |
8552 | } | |
8553 | return true; | |
8554 | } | |
8555 | ||
8556 | static bool zone_spans_last_pfn(const struct zone *zone, | |
8557 | unsigned long start_pfn, unsigned long nr_pages) | |
8558 | { | |
8559 | unsigned long last_pfn = start_pfn + nr_pages - 1; | |
8560 | ||
8561 | return zone_spans_pfn(zone, last_pfn); | |
8562 | } | |
8563 | ||
8564 | /** | |
8565 | * alloc_contig_pages() -- tries to find and allocate contiguous range of pages | |
8566 | * @nr_pages: Number of contiguous pages to allocate | |
8567 | * @gfp_mask: GFP mask to limit search and used during compaction | |
8568 | * @nid: Target node | |
8569 | * @nodemask: Mask for other possible nodes | |
8570 | * | |
8571 | * This routine is a wrapper around alloc_contig_range(). It scans over zones | |
8572 | * on an applicable zonelist to find a contiguous pfn range which can then be | |
8573 | * tried for allocation with alloc_contig_range(). This routine is intended | |
8574 | * for allocation requests which can not be fulfilled with the buddy allocator. | |
8575 | * | |
8576 | * The allocated memory is always aligned to a page boundary. If nr_pages is a | |
8577 | * power of two then the alignment is guaranteed to be to the given nr_pages | |
8578 | * (e.g. 1GB request would be aligned to 1GB). | |
8579 | * | |
8580 | * Allocated pages can be freed with free_contig_range() or by manually calling | |
8581 | * __free_page() on each allocated page. | |
8582 | * | |
8583 | * Return: pointer to contiguous pages on success, or NULL if not successful. | |
8584 | */ | |
8585 | struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, | |
8586 | int nid, nodemask_t *nodemask) | |
8587 | { | |
8588 | unsigned long ret, pfn, flags; | |
8589 | struct zonelist *zonelist; | |
8590 | struct zone *zone; | |
8591 | struct zoneref *z; | |
8592 | ||
8593 | zonelist = node_zonelist(nid, gfp_mask); | |
8594 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
8595 | gfp_zone(gfp_mask), nodemask) { | |
8596 | spin_lock_irqsave(&zone->lock, flags); | |
8597 | ||
8598 | pfn = ALIGN(zone->zone_start_pfn, nr_pages); | |
8599 | while (zone_spans_last_pfn(zone, pfn, nr_pages)) { | |
8600 | if (pfn_range_valid_contig(zone, pfn, nr_pages)) { | |
8601 | /* | |
8602 | * We release the zone lock here because | |
8603 | * alloc_contig_range() will also lock the zone | |
8604 | * at some point. If there's an allocation | |
8605 | * spinning on this lock, it may win the race | |
8606 | * and cause alloc_contig_range() to fail... | |
8607 | */ | |
8608 | spin_unlock_irqrestore(&zone->lock, flags); | |
8609 | ret = __alloc_contig_pages(pfn, nr_pages, | |
8610 | gfp_mask); | |
8611 | if (!ret) | |
8612 | return pfn_to_page(pfn); | |
8613 | spin_lock_irqsave(&zone->lock, flags); | |
8614 | } | |
8615 | pfn += nr_pages; | |
8616 | } | |
8617 | spin_unlock_irqrestore(&zone->lock, flags); | |
8618 | } | |
8619 | return NULL; | |
8620 | } | |
4eb0716e | 8621 | #endif /* CONFIG_CONTIG_ALLOC */ |
041d3a8c | 8622 | |
4eb0716e | 8623 | void free_contig_range(unsigned long pfn, unsigned int nr_pages) |
041d3a8c | 8624 | { |
bcc2b02f MS |
8625 | unsigned int count = 0; |
8626 | ||
8627 | for (; nr_pages--; pfn++) { | |
8628 | struct page *page = pfn_to_page(pfn); | |
8629 | ||
8630 | count += page_count(page) != 1; | |
8631 | __free_page(page); | |
8632 | } | |
8633 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c | 8634 | } |
041d3a8c | 8635 | |
0a647f38 CS |
8636 | /* |
8637 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
8638 | * page high values need to be recalulated. | |
8639 | */ | |
4ed7e022 JL |
8640 | void __meminit zone_pcp_update(struct zone *zone) |
8641 | { | |
c8e251fa | 8642 | mutex_lock(&pcp_batch_high_lock); |
cb1ef534 | 8643 | __zone_pcp_update(zone); |
c8e251fa | 8644 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 | 8645 | } |
4ed7e022 | 8646 | |
340175b7 JL |
8647 | void zone_pcp_reset(struct zone *zone) |
8648 | { | |
8649 | unsigned long flags; | |
5a883813 MK |
8650 | int cpu; |
8651 | struct per_cpu_pageset *pset; | |
340175b7 JL |
8652 | |
8653 | /* avoid races with drain_pages() */ | |
8654 | local_irq_save(flags); | |
8655 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
8656 | for_each_online_cpu(cpu) { |
8657 | pset = per_cpu_ptr(zone->pageset, cpu); | |
8658 | drain_zonestat(zone, pset); | |
8659 | } | |
340175b7 JL |
8660 | free_percpu(zone->pageset); |
8661 | zone->pageset = &boot_pageset; | |
8662 | } | |
8663 | local_irq_restore(flags); | |
8664 | } | |
8665 | ||
6dcd73d7 | 8666 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 8667 | /* |
b9eb6319 JK |
8668 | * All pages in the range must be in a single zone and isolated |
8669 | * before calling this. | |
0c0e6195 | 8670 | */ |
5557c766 | 8671 | unsigned long |
0c0e6195 KH |
8672 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
8673 | { | |
8674 | struct page *page; | |
8675 | struct zone *zone; | |
0ee5f4f3 | 8676 | unsigned int order; |
0c0e6195 KH |
8677 | unsigned long pfn; |
8678 | unsigned long flags; | |
5557c766 MH |
8679 | unsigned long offlined_pages = 0; |
8680 | ||
0c0e6195 KH |
8681 | /* find the first valid pfn */ |
8682 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
8683 | if (pfn_valid(pfn)) | |
8684 | break; | |
8685 | if (pfn == end_pfn) | |
5557c766 MH |
8686 | return offlined_pages; |
8687 | ||
2d070eab | 8688 | offline_mem_sections(pfn, end_pfn); |
0c0e6195 KH |
8689 | zone = page_zone(pfn_to_page(pfn)); |
8690 | spin_lock_irqsave(&zone->lock, flags); | |
8691 | pfn = start_pfn; | |
8692 | while (pfn < end_pfn) { | |
8693 | if (!pfn_valid(pfn)) { | |
8694 | pfn++; | |
8695 | continue; | |
8696 | } | |
8697 | page = pfn_to_page(pfn); | |
b023f468 WC |
8698 | /* |
8699 | * The HWPoisoned page may be not in buddy system, and | |
8700 | * page_count() is not 0. | |
8701 | */ | |
8702 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
8703 | pfn++; | |
5557c766 | 8704 | offlined_pages++; |
b023f468 WC |
8705 | continue; |
8706 | } | |
8707 | ||
0c0e6195 KH |
8708 | BUG_ON(page_count(page)); |
8709 | BUG_ON(!PageBuddy(page)); | |
8710 | order = page_order(page); | |
5557c766 | 8711 | offlined_pages += 1 << order; |
6ab01363 | 8712 | del_page_from_free_list(page, zone, order); |
0c0e6195 KH |
8713 | pfn += (1 << order); |
8714 | } | |
8715 | spin_unlock_irqrestore(&zone->lock, flags); | |
5557c766 MH |
8716 | |
8717 | return offlined_pages; | |
0c0e6195 KH |
8718 | } |
8719 | #endif | |
8d22ba1b | 8720 | |
8d22ba1b WF |
8721 | bool is_free_buddy_page(struct page *page) |
8722 | { | |
8723 | struct zone *zone = page_zone(page); | |
8724 | unsigned long pfn = page_to_pfn(page); | |
8725 | unsigned long flags; | |
7aeb09f9 | 8726 | unsigned int order; |
8d22ba1b WF |
8727 | |
8728 | spin_lock_irqsave(&zone->lock, flags); | |
8729 | for (order = 0; order < MAX_ORDER; order++) { | |
8730 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8731 | ||
8732 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
8733 | break; | |
8734 | } | |
8735 | spin_unlock_irqrestore(&zone->lock, flags); | |
8736 | ||
8737 | return order < MAX_ORDER; | |
8738 | } | |
d4ae9916 NH |
8739 | |
8740 | #ifdef CONFIG_MEMORY_FAILURE | |
8741 | /* | |
8742 | * Set PG_hwpoison flag if a given page is confirmed to be a free page. This | |
8743 | * test is performed under the zone lock to prevent a race against page | |
8744 | * allocation. | |
8745 | */ | |
8746 | bool set_hwpoison_free_buddy_page(struct page *page) | |
8747 | { | |
8748 | struct zone *zone = page_zone(page); | |
8749 | unsigned long pfn = page_to_pfn(page); | |
8750 | unsigned long flags; | |
8751 | unsigned int order; | |
8752 | bool hwpoisoned = false; | |
8753 | ||
8754 | spin_lock_irqsave(&zone->lock, flags); | |
8755 | for (order = 0; order < MAX_ORDER; order++) { | |
8756 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
8757 | ||
8758 | if (PageBuddy(page_head) && page_order(page_head) >= order) { | |
8759 | if (!TestSetPageHWPoison(page)) | |
8760 | hwpoisoned = true; | |
8761 | break; | |
8762 | } | |
8763 | } | |
8764 | spin_unlock_irqrestore(&zone->lock, flags); | |
8765 | ||
8766 | return hwpoisoned; | |
8767 | } | |
8768 | #endif |