]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/stddef.h> |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
10ed273f | 22 | #include <linux/jiffies.h> |
1da177e4 | 23 | #include <linux/bootmem.h> |
edbe7d23 | 24 | #include <linux/memblock.h> |
1da177e4 | 25 | #include <linux/compiler.h> |
9f158333 | 26 | #include <linux/kernel.h> |
b1eeab67 | 27 | #include <linux/kmemcheck.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/notifier.h> |
37 | #include <linux/topology.h> | |
38 | #include <linux/sysctl.h> | |
39 | #include <linux/cpu.h> | |
40 | #include <linux/cpuset.h> | |
bdc8cb98 | 41 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
42 | #include <linux/nodemask.h> |
43 | #include <linux/vmalloc.h> | |
a6cccdc3 | 44 | #include <linux/vmstat.h> |
4be38e35 | 45 | #include <linux/mempolicy.h> |
4b94ffdc | 46 | #include <linux/memremap.h> |
6811378e | 47 | #include <linux/stop_machine.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> |
eefa864b | 53 | #include <linux/page_ext.h> |
3ac7fe5a | 54 | #include <linux/debugobjects.h> |
dbb1f81c | 55 | #include <linux/kmemleak.h> |
56de7263 | 56 | #include <linux/compaction.h> |
0d3d062a | 57 | #include <trace/events/kmem.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
041d3a8c | 60 | #include <linux/migrate.h> |
e30825f1 | 61 | #include <linux/page_ext.h> |
949f7ec5 | 62 | #include <linux/hugetlb.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
48c96a36 | 64 | #include <linux/page_owner.h> |
0e1cc95b | 65 | #include <linux/kthread.h> |
1da177e4 | 66 | |
7ee3d4e8 | 67 | #include <asm/sections.h> |
1da177e4 | 68 | #include <asm/tlbflush.h> |
ac924c60 | 69 | #include <asm/div64.h> |
1da177e4 LT |
70 | #include "internal.h" |
71 | ||
c8e251fa CS |
72 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
73 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
7cd2b0a3 | 74 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 75 | |
72812019 LS |
76 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
77 | DEFINE_PER_CPU(int, numa_node); | |
78 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
79 | #endif | |
80 | ||
7aac7898 LS |
81 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
82 | /* | |
83 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
84 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
85 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
86 | * defined in <linux/topology.h>. | |
87 | */ | |
88 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
89 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
ad2c8144 | 90 | int _node_numa_mem_[MAX_NUMNODES]; |
7aac7898 LS |
91 | #endif |
92 | ||
1da177e4 | 93 | /* |
13808910 | 94 | * Array of node states. |
1da177e4 | 95 | */ |
13808910 CL |
96 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
97 | [N_POSSIBLE] = NODE_MASK_ALL, | |
98 | [N_ONLINE] = { { [0] = 1UL } }, | |
99 | #ifndef CONFIG_NUMA | |
100 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
101 | #ifdef CONFIG_HIGHMEM | |
102 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
103 | #endif |
104 | #ifdef CONFIG_MOVABLE_NODE | |
105 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
106 | #endif |
107 | [N_CPU] = { { [0] = 1UL } }, | |
108 | #endif /* NUMA */ | |
109 | }; | |
110 | EXPORT_SYMBOL(node_states); | |
111 | ||
c3d5f5f0 JL |
112 | /* Protect totalram_pages and zone->managed_pages */ |
113 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
114 | ||
6c231b7b | 115 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 116 | unsigned long totalreserve_pages __read_mostly; |
e48322ab | 117 | unsigned long totalcma_pages __read_mostly; |
ab8fabd4 | 118 | |
1b76b02f | 119 | int percpu_pagelist_fraction; |
dcce284a | 120 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 121 | |
bb14c2c7 VB |
122 | /* |
123 | * A cached value of the page's pageblock's migratetype, used when the page is | |
124 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when | |
125 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. | |
126 | * Also the migratetype set in the page does not necessarily match the pcplist | |
127 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any | |
128 | * other index - this ensures that it will be put on the correct CMA freelist. | |
129 | */ | |
130 | static inline int get_pcppage_migratetype(struct page *page) | |
131 | { | |
132 | return page->index; | |
133 | } | |
134 | ||
135 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) | |
136 | { | |
137 | page->index = migratetype; | |
138 | } | |
139 | ||
452aa699 RW |
140 | #ifdef CONFIG_PM_SLEEP |
141 | /* | |
142 | * The following functions are used by the suspend/hibernate code to temporarily | |
143 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
144 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
145 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
146 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
147 | * guaranteed not to run in parallel with that modification). | |
148 | */ | |
c9e664f1 RW |
149 | |
150 | static gfp_t saved_gfp_mask; | |
151 | ||
152 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
153 | { |
154 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
155 | if (saved_gfp_mask) { |
156 | gfp_allowed_mask = saved_gfp_mask; | |
157 | saved_gfp_mask = 0; | |
158 | } | |
452aa699 RW |
159 | } |
160 | ||
c9e664f1 | 161 | void pm_restrict_gfp_mask(void) |
452aa699 | 162 | { |
452aa699 | 163 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
164 | WARN_ON(saved_gfp_mask); |
165 | saved_gfp_mask = gfp_allowed_mask; | |
d0164adc | 166 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
452aa699 | 167 | } |
f90ac398 MG |
168 | |
169 | bool pm_suspended_storage(void) | |
170 | { | |
d0164adc | 171 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
f90ac398 MG |
172 | return false; |
173 | return true; | |
174 | } | |
452aa699 RW |
175 | #endif /* CONFIG_PM_SLEEP */ |
176 | ||
d9c23400 | 177 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
d00181b9 | 178 | unsigned int pageblock_order __read_mostly; |
d9c23400 MG |
179 | #endif |
180 | ||
d98c7a09 | 181 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 182 | |
1da177e4 LT |
183 | /* |
184 | * results with 256, 32 in the lowmem_reserve sysctl: | |
185 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
186 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
187 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
188 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84109e15 | 189 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
a2f1b424 AK |
190 | * |
191 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
192 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 193 | */ |
2f1b6248 | 194 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 195 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 196 | 256, |
4b51d669 | 197 | #endif |
fb0e7942 | 198 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 199 | 256, |
fb0e7942 | 200 | #endif |
e53ef38d | 201 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 202 | 32, |
e53ef38d | 203 | #endif |
2a1e274a | 204 | 32, |
2f1b6248 | 205 | }; |
1da177e4 LT |
206 | |
207 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 208 | |
15ad7cdc | 209 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 210 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 211 | "DMA", |
4b51d669 | 212 | #endif |
fb0e7942 | 213 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 214 | "DMA32", |
fb0e7942 | 215 | #endif |
2f1b6248 | 216 | "Normal", |
e53ef38d | 217 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 218 | "HighMem", |
e53ef38d | 219 | #endif |
2a1e274a | 220 | "Movable", |
033fbae9 DW |
221 | #ifdef CONFIG_ZONE_DEVICE |
222 | "Device", | |
223 | #endif | |
2f1b6248 CL |
224 | }; |
225 | ||
60f30350 VB |
226 | char * const migratetype_names[MIGRATE_TYPES] = { |
227 | "Unmovable", | |
228 | "Movable", | |
229 | "Reclaimable", | |
230 | "HighAtomic", | |
231 | #ifdef CONFIG_CMA | |
232 | "CMA", | |
233 | #endif | |
234 | #ifdef CONFIG_MEMORY_ISOLATION | |
235 | "Isolate", | |
236 | #endif | |
237 | }; | |
238 | ||
f1e61557 KS |
239 | compound_page_dtor * const compound_page_dtors[] = { |
240 | NULL, | |
241 | free_compound_page, | |
242 | #ifdef CONFIG_HUGETLB_PAGE | |
243 | free_huge_page, | |
244 | #endif | |
9a982250 KS |
245 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
246 | free_transhuge_page, | |
247 | #endif | |
f1e61557 KS |
248 | }; |
249 | ||
1da177e4 | 250 | int min_free_kbytes = 1024; |
42aa83cb | 251 | int user_min_free_kbytes = -1; |
795ae7a0 | 252 | int watermark_scale_factor = 10; |
1da177e4 | 253 | |
2c85f51d JB |
254 | static unsigned long __meminitdata nr_kernel_pages; |
255 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 256 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 257 | |
0ee332c1 TH |
258 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
259 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
260 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
261 | static unsigned long __initdata required_kernelcore; | |
262 | static unsigned long __initdata required_movablecore; | |
263 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
342332e6 | 264 | static bool mirrored_kernelcore; |
0ee332c1 TH |
265 | |
266 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
267 | int movable_zone; | |
268 | EXPORT_SYMBOL(movable_zone); | |
269 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 270 | |
418508c1 MS |
271 | #if MAX_NUMNODES > 1 |
272 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 273 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 274 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 275 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
276 | #endif |
277 | ||
9ef9acb0 MG |
278 | int page_group_by_mobility_disabled __read_mostly; |
279 | ||
3a80a7fa MG |
280 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
281 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
282 | { | |
283 | pgdat->first_deferred_pfn = ULONG_MAX; | |
284 | } | |
285 | ||
286 | /* Returns true if the struct page for the pfn is uninitialised */ | |
0e1cc95b | 287 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
3a80a7fa | 288 | { |
ae026b2a | 289 | if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn) |
3a80a7fa MG |
290 | return true; |
291 | ||
292 | return false; | |
293 | } | |
294 | ||
7e18adb4 MG |
295 | static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid) |
296 | { | |
297 | if (pfn >= NODE_DATA(nid)->first_deferred_pfn) | |
298 | return true; | |
299 | ||
300 | return false; | |
301 | } | |
302 | ||
3a80a7fa MG |
303 | /* |
304 | * Returns false when the remaining initialisation should be deferred until | |
305 | * later in the boot cycle when it can be parallelised. | |
306 | */ | |
307 | static inline bool update_defer_init(pg_data_t *pgdat, | |
308 | unsigned long pfn, unsigned long zone_end, | |
309 | unsigned long *nr_initialised) | |
310 | { | |
987b3095 LZ |
311 | unsigned long max_initialise; |
312 | ||
3a80a7fa MG |
313 | /* Always populate low zones for address-contrained allocations */ |
314 | if (zone_end < pgdat_end_pfn(pgdat)) | |
315 | return true; | |
987b3095 LZ |
316 | /* |
317 | * Initialise at least 2G of a node but also take into account that | |
318 | * two large system hashes that can take up 1GB for 0.25TB/node. | |
319 | */ | |
320 | max_initialise = max(2UL << (30 - PAGE_SHIFT), | |
321 | (pgdat->node_spanned_pages >> 8)); | |
3a80a7fa | 322 | |
3a80a7fa | 323 | (*nr_initialised)++; |
987b3095 | 324 | if ((*nr_initialised > max_initialise) && |
3a80a7fa MG |
325 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
326 | pgdat->first_deferred_pfn = pfn; | |
327 | return false; | |
328 | } | |
329 | ||
330 | return true; | |
331 | } | |
332 | #else | |
333 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
334 | { | |
335 | } | |
336 | ||
337 | static inline bool early_page_uninitialised(unsigned long pfn) | |
338 | { | |
339 | return false; | |
340 | } | |
341 | ||
7e18adb4 MG |
342 | static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid) |
343 | { | |
344 | return false; | |
345 | } | |
346 | ||
3a80a7fa MG |
347 | static inline bool update_defer_init(pg_data_t *pgdat, |
348 | unsigned long pfn, unsigned long zone_end, | |
349 | unsigned long *nr_initialised) | |
350 | { | |
351 | return true; | |
352 | } | |
353 | #endif | |
354 | ||
355 | ||
ee6f509c | 356 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 357 | { |
5d0f3f72 KM |
358 | if (unlikely(page_group_by_mobility_disabled && |
359 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
360 | migratetype = MIGRATE_UNMOVABLE; |
361 | ||
b2a0ac88 MG |
362 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
363 | PB_migrate, PB_migrate_end); | |
364 | } | |
365 | ||
13e7444b | 366 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 367 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 368 | { |
bdc8cb98 DH |
369 | int ret = 0; |
370 | unsigned seq; | |
371 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 372 | unsigned long sp, start_pfn; |
c6a57e19 | 373 | |
bdc8cb98 DH |
374 | do { |
375 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
376 | start_pfn = zone->zone_start_pfn; |
377 | sp = zone->spanned_pages; | |
108bcc96 | 378 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
379 | ret = 1; |
380 | } while (zone_span_seqretry(zone, seq)); | |
381 | ||
b5e6a5a2 | 382 | if (ret) |
613813e8 DH |
383 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
384 | pfn, zone_to_nid(zone), zone->name, | |
385 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 386 | |
bdc8cb98 | 387 | return ret; |
c6a57e19 DH |
388 | } |
389 | ||
390 | static int page_is_consistent(struct zone *zone, struct page *page) | |
391 | { | |
14e07298 | 392 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 393 | return 0; |
1da177e4 | 394 | if (zone != page_zone(page)) |
c6a57e19 DH |
395 | return 0; |
396 | ||
397 | return 1; | |
398 | } | |
399 | /* | |
400 | * Temporary debugging check for pages not lying within a given zone. | |
401 | */ | |
402 | static int bad_range(struct zone *zone, struct page *page) | |
403 | { | |
404 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 405 | return 1; |
c6a57e19 DH |
406 | if (!page_is_consistent(zone, page)) |
407 | return 1; | |
408 | ||
1da177e4 LT |
409 | return 0; |
410 | } | |
13e7444b NP |
411 | #else |
412 | static inline int bad_range(struct zone *zone, struct page *page) | |
413 | { | |
414 | return 0; | |
415 | } | |
416 | #endif | |
417 | ||
d230dec1 KS |
418 | static void bad_page(struct page *page, const char *reason, |
419 | unsigned long bad_flags) | |
1da177e4 | 420 | { |
d936cf9b HD |
421 | static unsigned long resume; |
422 | static unsigned long nr_shown; | |
423 | static unsigned long nr_unshown; | |
424 | ||
2a7684a2 WF |
425 | /* Don't complain about poisoned pages */ |
426 | if (PageHWPoison(page)) { | |
22b751c3 | 427 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
428 | return; |
429 | } | |
430 | ||
d936cf9b HD |
431 | /* |
432 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
433 | * or allow a steady drip of one report per second. | |
434 | */ | |
435 | if (nr_shown == 60) { | |
436 | if (time_before(jiffies, resume)) { | |
437 | nr_unshown++; | |
438 | goto out; | |
439 | } | |
440 | if (nr_unshown) { | |
ff8e8116 | 441 | pr_alert( |
1e9e6365 | 442 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
443 | nr_unshown); |
444 | nr_unshown = 0; | |
445 | } | |
446 | nr_shown = 0; | |
447 | } | |
448 | if (nr_shown++ == 0) | |
449 | resume = jiffies + 60 * HZ; | |
450 | ||
ff8e8116 | 451 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 452 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
453 | __dump_page(page, reason); |
454 | bad_flags &= page->flags; | |
455 | if (bad_flags) | |
456 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
457 | bad_flags, &bad_flags); | |
4e462112 | 458 | dump_page_owner(page); |
3dc14741 | 459 | |
4f31888c | 460 | print_modules(); |
1da177e4 | 461 | dump_stack(); |
d936cf9b | 462 | out: |
8cc3b392 | 463 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 464 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 465 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
466 | } |
467 | ||
1da177e4 LT |
468 | /* |
469 | * Higher-order pages are called "compound pages". They are structured thusly: | |
470 | * | |
1d798ca3 | 471 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 472 | * |
1d798ca3 KS |
473 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
474 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 475 | * |
1d798ca3 KS |
476 | * The first tail page's ->compound_dtor holds the offset in array of compound |
477 | * page destructors. See compound_page_dtors. | |
1da177e4 | 478 | * |
1d798ca3 | 479 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 480 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 481 | */ |
d98c7a09 | 482 | |
9a982250 | 483 | void free_compound_page(struct page *page) |
d98c7a09 | 484 | { |
d85f3385 | 485 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
486 | } |
487 | ||
d00181b9 | 488 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
489 | { |
490 | int i; | |
491 | int nr_pages = 1 << order; | |
492 | ||
f1e61557 | 493 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
494 | set_compound_order(page, order); |
495 | __SetPageHead(page); | |
496 | for (i = 1; i < nr_pages; i++) { | |
497 | struct page *p = page + i; | |
58a84aa9 | 498 | set_page_count(p, 0); |
1c290f64 | 499 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 500 | set_compound_head(p, page); |
18229df5 | 501 | } |
53f9263b | 502 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
503 | } |
504 | ||
c0a32fc5 SG |
505 | #ifdef CONFIG_DEBUG_PAGEALLOC |
506 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
507 | bool _debug_pagealloc_enabled __read_mostly |
508 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 509 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
510 | bool _debug_guardpage_enabled __read_mostly; |
511 | ||
031bc574 JK |
512 | static int __init early_debug_pagealloc(char *buf) |
513 | { | |
514 | if (!buf) | |
515 | return -EINVAL; | |
516 | ||
517 | if (strcmp(buf, "on") == 0) | |
518 | _debug_pagealloc_enabled = true; | |
519 | ||
ea6eabb0 CB |
520 | if (strcmp(buf, "off") == 0) |
521 | _debug_pagealloc_enabled = false; | |
522 | ||
031bc574 JK |
523 | return 0; |
524 | } | |
525 | early_param("debug_pagealloc", early_debug_pagealloc); | |
526 | ||
e30825f1 JK |
527 | static bool need_debug_guardpage(void) |
528 | { | |
031bc574 JK |
529 | /* If we don't use debug_pagealloc, we don't need guard page */ |
530 | if (!debug_pagealloc_enabled()) | |
531 | return false; | |
532 | ||
e30825f1 JK |
533 | return true; |
534 | } | |
535 | ||
536 | static void init_debug_guardpage(void) | |
537 | { | |
031bc574 JK |
538 | if (!debug_pagealloc_enabled()) |
539 | return; | |
540 | ||
e30825f1 JK |
541 | _debug_guardpage_enabled = true; |
542 | } | |
543 | ||
544 | struct page_ext_operations debug_guardpage_ops = { | |
545 | .need = need_debug_guardpage, | |
546 | .init = init_debug_guardpage, | |
547 | }; | |
c0a32fc5 SG |
548 | |
549 | static int __init debug_guardpage_minorder_setup(char *buf) | |
550 | { | |
551 | unsigned long res; | |
552 | ||
553 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
1170532b | 554 | pr_err("Bad debug_guardpage_minorder value\n"); |
c0a32fc5 SG |
555 | return 0; |
556 | } | |
557 | _debug_guardpage_minorder = res; | |
1170532b | 558 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
c0a32fc5 SG |
559 | return 0; |
560 | } | |
561 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
562 | ||
2847cf95 JK |
563 | static inline void set_page_guard(struct zone *zone, struct page *page, |
564 | unsigned int order, int migratetype) | |
c0a32fc5 | 565 | { |
e30825f1 JK |
566 | struct page_ext *page_ext; |
567 | ||
568 | if (!debug_guardpage_enabled()) | |
569 | return; | |
570 | ||
571 | page_ext = lookup_page_ext(page); | |
572 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
573 | ||
2847cf95 JK |
574 | INIT_LIST_HEAD(&page->lru); |
575 | set_page_private(page, order); | |
576 | /* Guard pages are not available for any usage */ | |
577 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
c0a32fc5 SG |
578 | } |
579 | ||
2847cf95 JK |
580 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
581 | unsigned int order, int migratetype) | |
c0a32fc5 | 582 | { |
e30825f1 JK |
583 | struct page_ext *page_ext; |
584 | ||
585 | if (!debug_guardpage_enabled()) | |
586 | return; | |
587 | ||
588 | page_ext = lookup_page_ext(page); | |
589 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
590 | ||
2847cf95 JK |
591 | set_page_private(page, 0); |
592 | if (!is_migrate_isolate(migratetype)) | |
593 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
594 | } |
595 | #else | |
e30825f1 | 596 | struct page_ext_operations debug_guardpage_ops = { NULL, }; |
2847cf95 JK |
597 | static inline void set_page_guard(struct zone *zone, struct page *page, |
598 | unsigned int order, int migratetype) {} | |
599 | static inline void clear_page_guard(struct zone *zone, struct page *page, | |
600 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
601 | #endif |
602 | ||
7aeb09f9 | 603 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 604 | { |
4c21e2f2 | 605 | set_page_private(page, order); |
676165a8 | 606 | __SetPageBuddy(page); |
1da177e4 LT |
607 | } |
608 | ||
609 | static inline void rmv_page_order(struct page *page) | |
610 | { | |
676165a8 | 611 | __ClearPageBuddy(page); |
4c21e2f2 | 612 | set_page_private(page, 0); |
1da177e4 LT |
613 | } |
614 | ||
1da177e4 LT |
615 | /* |
616 | * This function checks whether a page is free && is the buddy | |
617 | * we can do coalesce a page and its buddy if | |
13e7444b | 618 | * (a) the buddy is not in a hole && |
676165a8 | 619 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
620 | * (c) a page and its buddy have the same order && |
621 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 622 | * |
cf6fe945 WSH |
623 | * For recording whether a page is in the buddy system, we set ->_mapcount |
624 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
625 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
626 | * serialized by zone->lock. | |
1da177e4 | 627 | * |
676165a8 | 628 | * For recording page's order, we use page_private(page). |
1da177e4 | 629 | */ |
cb2b95e1 | 630 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 631 | unsigned int order) |
1da177e4 | 632 | { |
14e07298 | 633 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 634 | return 0; |
13e7444b | 635 | |
c0a32fc5 | 636 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
637 | if (page_zone_id(page) != page_zone_id(buddy)) |
638 | return 0; | |
639 | ||
4c5018ce WY |
640 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
641 | ||
c0a32fc5 SG |
642 | return 1; |
643 | } | |
644 | ||
cb2b95e1 | 645 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
646 | /* |
647 | * zone check is done late to avoid uselessly | |
648 | * calculating zone/node ids for pages that could | |
649 | * never merge. | |
650 | */ | |
651 | if (page_zone_id(page) != page_zone_id(buddy)) | |
652 | return 0; | |
653 | ||
4c5018ce WY |
654 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
655 | ||
6aa3001b | 656 | return 1; |
676165a8 | 657 | } |
6aa3001b | 658 | return 0; |
1da177e4 LT |
659 | } |
660 | ||
661 | /* | |
662 | * Freeing function for a buddy system allocator. | |
663 | * | |
664 | * The concept of a buddy system is to maintain direct-mapped table | |
665 | * (containing bit values) for memory blocks of various "orders". | |
666 | * The bottom level table contains the map for the smallest allocatable | |
667 | * units of memory (here, pages), and each level above it describes | |
668 | * pairs of units from the levels below, hence, "buddies". | |
669 | * At a high level, all that happens here is marking the table entry | |
670 | * at the bottom level available, and propagating the changes upward | |
671 | * as necessary, plus some accounting needed to play nicely with other | |
672 | * parts of the VM system. | |
673 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
674 | * free pages of length of (1 << order) and marked with _mapcount |
675 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
676 | * field. | |
1da177e4 | 677 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
678 | * other. That is, if we allocate a small block, and both were |
679 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 680 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 681 | * triggers coalescing into a block of larger size. |
1da177e4 | 682 | * |
6d49e352 | 683 | * -- nyc |
1da177e4 LT |
684 | */ |
685 | ||
48db57f8 | 686 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 687 | unsigned long pfn, |
ed0ae21d MG |
688 | struct zone *zone, unsigned int order, |
689 | int migratetype) | |
1da177e4 LT |
690 | { |
691 | unsigned long page_idx; | |
6dda9d55 | 692 | unsigned long combined_idx; |
43506fad | 693 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 694 | struct page *buddy; |
d9dddbf5 VB |
695 | unsigned int max_order; |
696 | ||
697 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); | |
1da177e4 | 698 | |
d29bb978 | 699 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 700 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 701 | |
ed0ae21d | 702 | VM_BUG_ON(migratetype == -1); |
d9dddbf5 | 703 | if (likely(!is_migrate_isolate(migratetype))) |
8f82b55d | 704 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
ed0ae21d | 705 | |
d9dddbf5 | 706 | page_idx = pfn & ((1 << MAX_ORDER) - 1); |
1da177e4 | 707 | |
309381fe SL |
708 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
709 | VM_BUG_ON_PAGE(bad_range(zone, page), page); | |
1da177e4 | 710 | |
d9dddbf5 | 711 | continue_merging: |
3c605096 | 712 | while (order < max_order - 1) { |
43506fad KC |
713 | buddy_idx = __find_buddy_index(page_idx, order); |
714 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 715 | if (!page_is_buddy(page, buddy, order)) |
d9dddbf5 | 716 | goto done_merging; |
c0a32fc5 SG |
717 | /* |
718 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
719 | * merge with it and move up one order. | |
720 | */ | |
721 | if (page_is_guard(buddy)) { | |
2847cf95 | 722 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
723 | } else { |
724 | list_del(&buddy->lru); | |
725 | zone->free_area[order].nr_free--; | |
726 | rmv_page_order(buddy); | |
727 | } | |
43506fad | 728 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
729 | page = page + (combined_idx - page_idx); |
730 | page_idx = combined_idx; | |
731 | order++; | |
732 | } | |
d9dddbf5 VB |
733 | if (max_order < MAX_ORDER) { |
734 | /* If we are here, it means order is >= pageblock_order. | |
735 | * We want to prevent merge between freepages on isolate | |
736 | * pageblock and normal pageblock. Without this, pageblock | |
737 | * isolation could cause incorrect freepage or CMA accounting. | |
738 | * | |
739 | * We don't want to hit this code for the more frequent | |
740 | * low-order merging. | |
741 | */ | |
742 | if (unlikely(has_isolate_pageblock(zone))) { | |
743 | int buddy_mt; | |
744 | ||
745 | buddy_idx = __find_buddy_index(page_idx, order); | |
746 | buddy = page + (buddy_idx - page_idx); | |
747 | buddy_mt = get_pageblock_migratetype(buddy); | |
748 | ||
749 | if (migratetype != buddy_mt | |
750 | && (is_migrate_isolate(migratetype) || | |
751 | is_migrate_isolate(buddy_mt))) | |
752 | goto done_merging; | |
753 | } | |
754 | max_order++; | |
755 | goto continue_merging; | |
756 | } | |
757 | ||
758 | done_merging: | |
1da177e4 | 759 | set_page_order(page, order); |
6dda9d55 CZ |
760 | |
761 | /* | |
762 | * If this is not the largest possible page, check if the buddy | |
763 | * of the next-highest order is free. If it is, it's possible | |
764 | * that pages are being freed that will coalesce soon. In case, | |
765 | * that is happening, add the free page to the tail of the list | |
766 | * so it's less likely to be used soon and more likely to be merged | |
767 | * as a higher order page | |
768 | */ | |
b7f50cfa | 769 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 770 | struct page *higher_page, *higher_buddy; |
43506fad KC |
771 | combined_idx = buddy_idx & page_idx; |
772 | higher_page = page + (combined_idx - page_idx); | |
773 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 774 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
775 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
776 | list_add_tail(&page->lru, | |
777 | &zone->free_area[order].free_list[migratetype]); | |
778 | goto out; | |
779 | } | |
780 | } | |
781 | ||
782 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
783 | out: | |
1da177e4 LT |
784 | zone->free_area[order].nr_free++; |
785 | } | |
786 | ||
224abf92 | 787 | static inline int free_pages_check(struct page *page) |
1da177e4 | 788 | { |
d230dec1 | 789 | const char *bad_reason = NULL; |
f0b791a3 DH |
790 | unsigned long bad_flags = 0; |
791 | ||
53f9263b | 792 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
793 | bad_reason = "nonzero mapcount"; |
794 | if (unlikely(page->mapping != NULL)) | |
795 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 796 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 797 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
798 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
799 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
800 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
801 | } | |
9edad6ea JW |
802 | #ifdef CONFIG_MEMCG |
803 | if (unlikely(page->mem_cgroup)) | |
804 | bad_reason = "page still charged to cgroup"; | |
805 | #endif | |
f0b791a3 DH |
806 | if (unlikely(bad_reason)) { |
807 | bad_page(page, bad_reason, bad_flags); | |
79f4b7bf | 808 | return 1; |
8cc3b392 | 809 | } |
90572890 | 810 | page_cpupid_reset_last(page); |
79f4b7bf HD |
811 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
812 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
813 | return 0; | |
1da177e4 LT |
814 | } |
815 | ||
816 | /* | |
5f8dcc21 | 817 | * Frees a number of pages from the PCP lists |
1da177e4 | 818 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 819 | * count is the number of pages to free. |
1da177e4 LT |
820 | * |
821 | * If the zone was previously in an "all pages pinned" state then look to | |
822 | * see if this freeing clears that state. | |
823 | * | |
824 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
825 | * pinned" detection logic. | |
826 | */ | |
5f8dcc21 MG |
827 | static void free_pcppages_bulk(struct zone *zone, int count, |
828 | struct per_cpu_pages *pcp) | |
1da177e4 | 829 | { |
5f8dcc21 | 830 | int migratetype = 0; |
a6f9edd6 | 831 | int batch_free = 0; |
72853e29 | 832 | int to_free = count; |
0d5d823a | 833 | unsigned long nr_scanned; |
5f8dcc21 | 834 | |
c54ad30c | 835 | spin_lock(&zone->lock); |
0d5d823a MG |
836 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
837 | if (nr_scanned) | |
838 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 839 | |
72853e29 | 840 | while (to_free) { |
48db57f8 | 841 | struct page *page; |
5f8dcc21 MG |
842 | struct list_head *list; |
843 | ||
844 | /* | |
a6f9edd6 MG |
845 | * Remove pages from lists in a round-robin fashion. A |
846 | * batch_free count is maintained that is incremented when an | |
847 | * empty list is encountered. This is so more pages are freed | |
848 | * off fuller lists instead of spinning excessively around empty | |
849 | * lists | |
5f8dcc21 MG |
850 | */ |
851 | do { | |
a6f9edd6 | 852 | batch_free++; |
5f8dcc21 MG |
853 | if (++migratetype == MIGRATE_PCPTYPES) |
854 | migratetype = 0; | |
855 | list = &pcp->lists[migratetype]; | |
856 | } while (list_empty(list)); | |
48db57f8 | 857 | |
1d16871d NK |
858 | /* This is the only non-empty list. Free them all. */ |
859 | if (batch_free == MIGRATE_PCPTYPES) | |
860 | batch_free = to_free; | |
861 | ||
a6f9edd6 | 862 | do { |
770c8aaa BZ |
863 | int mt; /* migratetype of the to-be-freed page */ |
864 | ||
a16601c5 | 865 | page = list_last_entry(list, struct page, lru); |
a6f9edd6 MG |
866 | /* must delete as __free_one_page list manipulates */ |
867 | list_del(&page->lru); | |
aa016d14 | 868 | |
bb14c2c7 | 869 | mt = get_pcppage_migratetype(page); |
aa016d14 VB |
870 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
871 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
872 | /* Pageblock could have been isolated meanwhile */ | |
8f82b55d | 873 | if (unlikely(has_isolate_pageblock(zone))) |
51bb1a40 | 874 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 875 | |
dc4b0caf | 876 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 877 | trace_mm_page_pcpu_drain(page, 0, mt); |
72853e29 | 878 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 879 | } |
c54ad30c | 880 | spin_unlock(&zone->lock); |
1da177e4 LT |
881 | } |
882 | ||
dc4b0caf MG |
883 | static void free_one_page(struct zone *zone, |
884 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 885 | unsigned int order, |
ed0ae21d | 886 | int migratetype) |
1da177e4 | 887 | { |
0d5d823a | 888 | unsigned long nr_scanned; |
006d22d9 | 889 | spin_lock(&zone->lock); |
0d5d823a MG |
890 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
891 | if (nr_scanned) | |
892 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 893 | |
ad53f92e JK |
894 | if (unlikely(has_isolate_pageblock(zone) || |
895 | is_migrate_isolate(migratetype))) { | |
896 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 897 | } |
dc4b0caf | 898 | __free_one_page(page, pfn, zone, order, migratetype); |
006d22d9 | 899 | spin_unlock(&zone->lock); |
48db57f8 NP |
900 | } |
901 | ||
81422f29 KS |
902 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
903 | { | |
1d798ca3 KS |
904 | int ret = 1; |
905 | ||
906 | /* | |
907 | * We rely page->lru.next never has bit 0 set, unless the page | |
908 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
909 | */ | |
910 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
911 | ||
912 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
913 | ret = 0; | |
914 | goto out; | |
915 | } | |
9a982250 KS |
916 | switch (page - head_page) { |
917 | case 1: | |
918 | /* the first tail page: ->mapping is compound_mapcount() */ | |
53f9263b KS |
919 | if (unlikely(compound_mapcount(page))) { |
920 | bad_page(page, "nonzero compound_mapcount", 0); | |
921 | goto out; | |
922 | } | |
9a982250 KS |
923 | break; |
924 | case 2: | |
925 | /* | |
926 | * the second tail page: ->mapping is | |
927 | * page_deferred_list().next -- ignore value. | |
928 | */ | |
929 | break; | |
930 | default: | |
931 | if (page->mapping != TAIL_MAPPING) { | |
932 | bad_page(page, "corrupted mapping in tail page", 0); | |
933 | goto out; | |
934 | } | |
935 | break; | |
1c290f64 | 936 | } |
81422f29 KS |
937 | if (unlikely(!PageTail(page))) { |
938 | bad_page(page, "PageTail not set", 0); | |
1d798ca3 | 939 | goto out; |
81422f29 | 940 | } |
1d798ca3 KS |
941 | if (unlikely(compound_head(page) != head_page)) { |
942 | bad_page(page, "compound_head not consistent", 0); | |
943 | goto out; | |
81422f29 | 944 | } |
1d798ca3 KS |
945 | ret = 0; |
946 | out: | |
1c290f64 | 947 | page->mapping = NULL; |
1d798ca3 KS |
948 | clear_compound_head(page); |
949 | return ret; | |
81422f29 KS |
950 | } |
951 | ||
1e8ce83c RH |
952 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
953 | unsigned long zone, int nid) | |
954 | { | |
1e8ce83c | 955 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
956 | init_page_count(page); |
957 | page_mapcount_reset(page); | |
958 | page_cpupid_reset_last(page); | |
1e8ce83c | 959 | |
1e8ce83c RH |
960 | INIT_LIST_HEAD(&page->lru); |
961 | #ifdef WANT_PAGE_VIRTUAL | |
962 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
963 | if (!is_highmem_idx(zone)) | |
964 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
965 | #endif | |
966 | } | |
967 | ||
968 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
969 | int nid) | |
970 | { | |
971 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
972 | } | |
973 | ||
7e18adb4 MG |
974 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
975 | static void init_reserved_page(unsigned long pfn) | |
976 | { | |
977 | pg_data_t *pgdat; | |
978 | int nid, zid; | |
979 | ||
980 | if (!early_page_uninitialised(pfn)) | |
981 | return; | |
982 | ||
983 | nid = early_pfn_to_nid(pfn); | |
984 | pgdat = NODE_DATA(nid); | |
985 | ||
986 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
987 | struct zone *zone = &pgdat->node_zones[zid]; | |
988 | ||
989 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
990 | break; | |
991 | } | |
992 | __init_single_pfn(pfn, zid, nid); | |
993 | } | |
994 | #else | |
995 | static inline void init_reserved_page(unsigned long pfn) | |
996 | { | |
997 | } | |
998 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
999 | ||
92923ca3 NZ |
1000 | /* |
1001 | * Initialised pages do not have PageReserved set. This function is | |
1002 | * called for each range allocated by the bootmem allocator and | |
1003 | * marks the pages PageReserved. The remaining valid pages are later | |
1004 | * sent to the buddy page allocator. | |
1005 | */ | |
7e18adb4 | 1006 | void __meminit reserve_bootmem_region(unsigned long start, unsigned long end) |
92923ca3 NZ |
1007 | { |
1008 | unsigned long start_pfn = PFN_DOWN(start); | |
1009 | unsigned long end_pfn = PFN_UP(end); | |
1010 | ||
7e18adb4 MG |
1011 | for (; start_pfn < end_pfn; start_pfn++) { |
1012 | if (pfn_valid(start_pfn)) { | |
1013 | struct page *page = pfn_to_page(start_pfn); | |
1014 | ||
1015 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1016 | |
1017 | /* Avoid false-positive PageTail() */ | |
1018 | INIT_LIST_HEAD(&page->lru); | |
1019 | ||
7e18adb4 MG |
1020 | SetPageReserved(page); |
1021 | } | |
1022 | } | |
92923ca3 NZ |
1023 | } |
1024 | ||
ec95f53a | 1025 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 1026 | { |
d61f8590 | 1027 | int bad = 0; |
1da177e4 | 1028 | |
ab1f306f | 1029 | VM_BUG_ON_PAGE(PageTail(page), page); |
ab1f306f | 1030 | |
b413d48a | 1031 | trace_mm_page_free(page, order); |
b1eeab67 | 1032 | kmemcheck_free_shadow(page, order); |
b8c73fc2 | 1033 | kasan_free_pages(page, order); |
b1eeab67 | 1034 | |
d61f8590 MG |
1035 | /* |
1036 | * Check tail pages before head page information is cleared to | |
1037 | * avoid checking PageCompound for order-0 pages. | |
1038 | */ | |
1039 | if (unlikely(order)) { | |
1040 | bool compound = PageCompound(page); | |
1041 | int i; | |
1042 | ||
1043 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
1044 | ||
1045 | for (i = 1; i < (1 << order); i++) { | |
1046 | if (compound) | |
1047 | bad += free_tail_pages_check(page, page + i); | |
1048 | bad += free_pages_check(page + i); | |
1049 | } | |
1050 | } | |
17514574 | 1051 | if (PageAnonHead(page)) |
8dd60a3a | 1052 | page->mapping = NULL; |
81422f29 | 1053 | bad += free_pages_check(page); |
8cc3b392 | 1054 | if (bad) |
ec95f53a | 1055 | return false; |
689bcebf | 1056 | |
48c96a36 JK |
1057 | reset_page_owner(page, order); |
1058 | ||
3ac7fe5a | 1059 | if (!PageHighMem(page)) { |
b8af2941 PK |
1060 | debug_check_no_locks_freed(page_address(page), |
1061 | PAGE_SIZE << order); | |
3ac7fe5a TG |
1062 | debug_check_no_obj_freed(page_address(page), |
1063 | PAGE_SIZE << order); | |
1064 | } | |
dafb1367 | 1065 | arch_free_page(page, order); |
8823b1db | 1066 | kernel_poison_pages(page, 1 << order, 0); |
48db57f8 | 1067 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 1068 | |
ec95f53a KM |
1069 | return true; |
1070 | } | |
1071 | ||
1072 | static void __free_pages_ok(struct page *page, unsigned int order) | |
1073 | { | |
1074 | unsigned long flags; | |
95e34412 | 1075 | int migratetype; |
dc4b0caf | 1076 | unsigned long pfn = page_to_pfn(page); |
ec95f53a KM |
1077 | |
1078 | if (!free_pages_prepare(page, order)) | |
1079 | return; | |
1080 | ||
cfc47a28 | 1081 | migratetype = get_pfnblock_migratetype(page, pfn); |
c54ad30c | 1082 | local_irq_save(flags); |
f8891e5e | 1083 | __count_vm_events(PGFREE, 1 << order); |
dc4b0caf | 1084 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
c54ad30c | 1085 | local_irq_restore(flags); |
1da177e4 LT |
1086 | } |
1087 | ||
949698a3 | 1088 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1089 | { |
c3993076 | 1090 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1091 | struct page *p = page; |
c3993076 | 1092 | unsigned int loop; |
a226f6c8 | 1093 | |
e2d0bd2b YL |
1094 | prefetchw(p); |
1095 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1096 | prefetchw(p + 1); | |
c3993076 JW |
1097 | __ClearPageReserved(p); |
1098 | set_page_count(p, 0); | |
a226f6c8 | 1099 | } |
e2d0bd2b YL |
1100 | __ClearPageReserved(p); |
1101 | set_page_count(p, 0); | |
c3993076 | 1102 | |
e2d0bd2b | 1103 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1104 | set_page_refcounted(page); |
1105 | __free_pages(page, order); | |
a226f6c8 DH |
1106 | } |
1107 | ||
75a592a4 MG |
1108 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1109 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1110 | |
75a592a4 MG |
1111 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1112 | ||
1113 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1114 | { | |
7ace9917 | 1115 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1116 | int nid; |
1117 | ||
7ace9917 | 1118 | spin_lock(&early_pfn_lock); |
75a592a4 | 1119 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 MG |
1120 | if (nid < 0) |
1121 | nid = 0; | |
1122 | spin_unlock(&early_pfn_lock); | |
1123 | ||
1124 | return nid; | |
75a592a4 MG |
1125 | } |
1126 | #endif | |
1127 | ||
1128 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
1129 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1130 | struct mminit_pfnnid_cache *state) | |
1131 | { | |
1132 | int nid; | |
1133 | ||
1134 | nid = __early_pfn_to_nid(pfn, state); | |
1135 | if (nid >= 0 && nid != node) | |
1136 | return false; | |
1137 | return true; | |
1138 | } | |
1139 | ||
1140 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1141 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1142 | { | |
1143 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1144 | } | |
1145 | ||
1146 | #else | |
1147 | ||
1148 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1149 | { | |
1150 | return true; | |
1151 | } | |
1152 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1153 | struct mminit_pfnnid_cache *state) | |
1154 | { | |
1155 | return true; | |
1156 | } | |
1157 | #endif | |
1158 | ||
1159 | ||
0e1cc95b | 1160 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1161 | unsigned int order) |
1162 | { | |
1163 | if (early_page_uninitialised(pfn)) | |
1164 | return; | |
949698a3 | 1165 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1166 | } |
1167 | ||
7cf91a98 JK |
1168 | /* |
1169 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1170 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1171 | * with the migration of free compaction scanner. The scanners then need to | |
1172 | * use only pfn_valid_within() check for arches that allow holes within | |
1173 | * pageblocks. | |
1174 | * | |
1175 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1176 | * | |
1177 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1178 | * i.e. it's possible that all pages within a zones range of pages do not | |
1179 | * belong to a single zone. We assume that a border between node0 and node1 | |
1180 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1181 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1182 | * the first and last page of a pageblock and avoid checking each individual | |
1183 | * page in a pageblock. | |
1184 | */ | |
1185 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1186 | unsigned long end_pfn, struct zone *zone) | |
1187 | { | |
1188 | struct page *start_page; | |
1189 | struct page *end_page; | |
1190 | ||
1191 | /* end_pfn is one past the range we are checking */ | |
1192 | end_pfn--; | |
1193 | ||
1194 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1195 | return NULL; | |
1196 | ||
1197 | start_page = pfn_to_page(start_pfn); | |
1198 | ||
1199 | if (page_zone(start_page) != zone) | |
1200 | return NULL; | |
1201 | ||
1202 | end_page = pfn_to_page(end_pfn); | |
1203 | ||
1204 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1205 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1206 | return NULL; | |
1207 | ||
1208 | return start_page; | |
1209 | } | |
1210 | ||
1211 | void set_zone_contiguous(struct zone *zone) | |
1212 | { | |
1213 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1214 | unsigned long block_end_pfn; | |
1215 | ||
1216 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1217 | for (; block_start_pfn < zone_end_pfn(zone); | |
1218 | block_start_pfn = block_end_pfn, | |
1219 | block_end_pfn += pageblock_nr_pages) { | |
1220 | ||
1221 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1222 | ||
1223 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1224 | block_end_pfn, zone)) | |
1225 | return; | |
1226 | } | |
1227 | ||
1228 | /* We confirm that there is no hole */ | |
1229 | zone->contiguous = true; | |
1230 | } | |
1231 | ||
1232 | void clear_zone_contiguous(struct zone *zone) | |
1233 | { | |
1234 | zone->contiguous = false; | |
1235 | } | |
1236 | ||
7e18adb4 | 1237 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
0e1cc95b | 1238 | static void __init deferred_free_range(struct page *page, |
a4de83dd MG |
1239 | unsigned long pfn, int nr_pages) |
1240 | { | |
1241 | int i; | |
1242 | ||
1243 | if (!page) | |
1244 | return; | |
1245 | ||
1246 | /* Free a large naturally-aligned chunk if possible */ | |
1247 | if (nr_pages == MAX_ORDER_NR_PAGES && | |
1248 | (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) { | |
ac5d2539 | 1249 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
949698a3 | 1250 | __free_pages_boot_core(page, MAX_ORDER-1); |
a4de83dd MG |
1251 | return; |
1252 | } | |
1253 | ||
949698a3 LZ |
1254 | for (i = 0; i < nr_pages; i++, page++) |
1255 | __free_pages_boot_core(page, 0); | |
a4de83dd MG |
1256 | } |
1257 | ||
d3cd131d NS |
1258 | /* Completion tracking for deferred_init_memmap() threads */ |
1259 | static atomic_t pgdat_init_n_undone __initdata; | |
1260 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1261 | ||
1262 | static inline void __init pgdat_init_report_one_done(void) | |
1263 | { | |
1264 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1265 | complete(&pgdat_init_all_done_comp); | |
1266 | } | |
0e1cc95b | 1267 | |
7e18adb4 | 1268 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1269 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1270 | { |
0e1cc95b MG |
1271 | pg_data_t *pgdat = data; |
1272 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1273 | struct mminit_pfnnid_cache nid_init_state = { }; |
1274 | unsigned long start = jiffies; | |
1275 | unsigned long nr_pages = 0; | |
1276 | unsigned long walk_start, walk_end; | |
1277 | int i, zid; | |
1278 | struct zone *zone; | |
7e18adb4 | 1279 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
0e1cc95b | 1280 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
7e18adb4 | 1281 | |
0e1cc95b | 1282 | if (first_init_pfn == ULONG_MAX) { |
d3cd131d | 1283 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1284 | return 0; |
1285 | } | |
1286 | ||
1287 | /* Bind memory initialisation thread to a local node if possible */ | |
1288 | if (!cpumask_empty(cpumask)) | |
1289 | set_cpus_allowed_ptr(current, cpumask); | |
7e18adb4 MG |
1290 | |
1291 | /* Sanity check boundaries */ | |
1292 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1293 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1294 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1295 | ||
1296 | /* Only the highest zone is deferred so find it */ | |
1297 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1298 | zone = pgdat->node_zones + zid; | |
1299 | if (first_init_pfn < zone_end_pfn(zone)) | |
1300 | break; | |
1301 | } | |
1302 | ||
1303 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { | |
1304 | unsigned long pfn, end_pfn; | |
54608c3f | 1305 | struct page *page = NULL; |
a4de83dd MG |
1306 | struct page *free_base_page = NULL; |
1307 | unsigned long free_base_pfn = 0; | |
1308 | int nr_to_free = 0; | |
7e18adb4 MG |
1309 | |
1310 | end_pfn = min(walk_end, zone_end_pfn(zone)); | |
1311 | pfn = first_init_pfn; | |
1312 | if (pfn < walk_start) | |
1313 | pfn = walk_start; | |
1314 | if (pfn < zone->zone_start_pfn) | |
1315 | pfn = zone->zone_start_pfn; | |
1316 | ||
1317 | for (; pfn < end_pfn; pfn++) { | |
54608c3f | 1318 | if (!pfn_valid_within(pfn)) |
a4de83dd | 1319 | goto free_range; |
7e18adb4 | 1320 | |
54608c3f MG |
1321 | /* |
1322 | * Ensure pfn_valid is checked every | |
1323 | * MAX_ORDER_NR_PAGES for memory holes | |
1324 | */ | |
1325 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
1326 | if (!pfn_valid(pfn)) { | |
1327 | page = NULL; | |
a4de83dd | 1328 | goto free_range; |
54608c3f MG |
1329 | } |
1330 | } | |
1331 | ||
1332 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { | |
1333 | page = NULL; | |
a4de83dd | 1334 | goto free_range; |
54608c3f MG |
1335 | } |
1336 | ||
1337 | /* Minimise pfn page lookups and scheduler checks */ | |
1338 | if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) { | |
1339 | page++; | |
1340 | } else { | |
a4de83dd MG |
1341 | nr_pages += nr_to_free; |
1342 | deferred_free_range(free_base_page, | |
1343 | free_base_pfn, nr_to_free); | |
1344 | free_base_page = NULL; | |
1345 | free_base_pfn = nr_to_free = 0; | |
1346 | ||
54608c3f MG |
1347 | page = pfn_to_page(pfn); |
1348 | cond_resched(); | |
1349 | } | |
7e18adb4 MG |
1350 | |
1351 | if (page->flags) { | |
1352 | VM_BUG_ON(page_zone(page) != zone); | |
a4de83dd | 1353 | goto free_range; |
7e18adb4 MG |
1354 | } |
1355 | ||
1356 | __init_single_page(page, pfn, zid, nid); | |
a4de83dd MG |
1357 | if (!free_base_page) { |
1358 | free_base_page = page; | |
1359 | free_base_pfn = pfn; | |
1360 | nr_to_free = 0; | |
1361 | } | |
1362 | nr_to_free++; | |
1363 | ||
1364 | /* Where possible, batch up pages for a single free */ | |
1365 | continue; | |
1366 | free_range: | |
1367 | /* Free the current block of pages to allocator */ | |
1368 | nr_pages += nr_to_free; | |
1369 | deferred_free_range(free_base_page, free_base_pfn, | |
1370 | nr_to_free); | |
1371 | free_base_page = NULL; | |
1372 | free_base_pfn = nr_to_free = 0; | |
7e18adb4 | 1373 | } |
a4de83dd | 1374 | |
7e18adb4 MG |
1375 | first_init_pfn = max(end_pfn, first_init_pfn); |
1376 | } | |
1377 | ||
1378 | /* Sanity check that the next zone really is unpopulated */ | |
1379 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1380 | ||
0e1cc95b | 1381 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1382 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1383 | |
1384 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1385 | return 0; |
1386 | } | |
7cf91a98 | 1387 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1388 | |
1389 | void __init page_alloc_init_late(void) | |
1390 | { | |
7cf91a98 JK |
1391 | struct zone *zone; |
1392 | ||
1393 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1394 | int nid; |
1395 | ||
d3cd131d NS |
1396 | /* There will be num_node_state(N_MEMORY) threads */ |
1397 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1398 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1399 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1400 | } | |
1401 | ||
1402 | /* Block until all are initialised */ | |
d3cd131d | 1403 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da MG |
1404 | |
1405 | /* Reinit limits that are based on free pages after the kernel is up */ | |
1406 | files_maxfiles_init(); | |
7cf91a98 JK |
1407 | #endif |
1408 | ||
1409 | for_each_populated_zone(zone) | |
1410 | set_zone_contiguous(zone); | |
7e18adb4 | 1411 | } |
7e18adb4 | 1412 | |
47118af0 | 1413 | #ifdef CONFIG_CMA |
9cf510a5 | 1414 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1415 | void __init init_cma_reserved_pageblock(struct page *page) |
1416 | { | |
1417 | unsigned i = pageblock_nr_pages; | |
1418 | struct page *p = page; | |
1419 | ||
1420 | do { | |
1421 | __ClearPageReserved(p); | |
1422 | set_page_count(p, 0); | |
1423 | } while (++p, --i); | |
1424 | ||
47118af0 | 1425 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1426 | |
1427 | if (pageblock_order >= MAX_ORDER) { | |
1428 | i = pageblock_nr_pages; | |
1429 | p = page; | |
1430 | do { | |
1431 | set_page_refcounted(p); | |
1432 | __free_pages(p, MAX_ORDER - 1); | |
1433 | p += MAX_ORDER_NR_PAGES; | |
1434 | } while (i -= MAX_ORDER_NR_PAGES); | |
1435 | } else { | |
1436 | set_page_refcounted(page); | |
1437 | __free_pages(page, pageblock_order); | |
1438 | } | |
1439 | ||
3dcc0571 | 1440 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1441 | } |
1442 | #endif | |
1da177e4 LT |
1443 | |
1444 | /* | |
1445 | * The order of subdivision here is critical for the IO subsystem. | |
1446 | * Please do not alter this order without good reasons and regression | |
1447 | * testing. Specifically, as large blocks of memory are subdivided, | |
1448 | * the order in which smaller blocks are delivered depends on the order | |
1449 | * they're subdivided in this function. This is the primary factor | |
1450 | * influencing the order in which pages are delivered to the IO | |
1451 | * subsystem according to empirical testing, and this is also justified | |
1452 | * by considering the behavior of a buddy system containing a single | |
1453 | * large block of memory acted on by a series of small allocations. | |
1454 | * This behavior is a critical factor in sglist merging's success. | |
1455 | * | |
6d49e352 | 1456 | * -- nyc |
1da177e4 | 1457 | */ |
085cc7d5 | 1458 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1459 | int low, int high, struct free_area *area, |
1460 | int migratetype) | |
1da177e4 LT |
1461 | { |
1462 | unsigned long size = 1 << high; | |
1463 | ||
1464 | while (high > low) { | |
1465 | area--; | |
1466 | high--; | |
1467 | size >>= 1; | |
309381fe | 1468 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1469 | |
2847cf95 | 1470 | if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && |
e30825f1 | 1471 | debug_guardpage_enabled() && |
2847cf95 | 1472 | high < debug_guardpage_minorder()) { |
c0a32fc5 SG |
1473 | /* |
1474 | * Mark as guard pages (or page), that will allow to | |
1475 | * merge back to allocator when buddy will be freed. | |
1476 | * Corresponding page table entries will not be touched, | |
1477 | * pages will stay not present in virtual address space | |
1478 | */ | |
2847cf95 | 1479 | set_page_guard(zone, &page[size], high, migratetype); |
c0a32fc5 SG |
1480 | continue; |
1481 | } | |
b2a0ac88 | 1482 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1483 | area->nr_free++; |
1484 | set_page_order(&page[size], high); | |
1485 | } | |
1da177e4 LT |
1486 | } |
1487 | ||
1da177e4 LT |
1488 | /* |
1489 | * This page is about to be returned from the page allocator | |
1490 | */ | |
2a7684a2 | 1491 | static inline int check_new_page(struct page *page) |
1da177e4 | 1492 | { |
d230dec1 | 1493 | const char *bad_reason = NULL; |
f0b791a3 DH |
1494 | unsigned long bad_flags = 0; |
1495 | ||
53f9263b | 1496 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1497 | bad_reason = "nonzero mapcount"; |
1498 | if (unlikely(page->mapping != NULL)) | |
1499 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1500 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1501 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1502 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1503 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1504 | bad_flags = __PG_HWPOISON; | |
1505 | } | |
f0b791a3 DH |
1506 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1507 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1508 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1509 | } | |
9edad6ea JW |
1510 | #ifdef CONFIG_MEMCG |
1511 | if (unlikely(page->mem_cgroup)) | |
1512 | bad_reason = "page still charged to cgroup"; | |
1513 | #endif | |
f0b791a3 DH |
1514 | if (unlikely(bad_reason)) { |
1515 | bad_page(page, bad_reason, bad_flags); | |
689bcebf | 1516 | return 1; |
8cc3b392 | 1517 | } |
2a7684a2 WF |
1518 | return 0; |
1519 | } | |
1520 | ||
1414c7f4 LA |
1521 | static inline bool free_pages_prezeroed(bool poisoned) |
1522 | { | |
1523 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
1524 | page_poisoning_enabled() && poisoned; | |
1525 | } | |
1526 | ||
75379191 | 1527 | static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1528 | unsigned int alloc_flags) |
2a7684a2 WF |
1529 | { |
1530 | int i; | |
1414c7f4 | 1531 | bool poisoned = true; |
2a7684a2 WF |
1532 | |
1533 | for (i = 0; i < (1 << order); i++) { | |
1534 | struct page *p = page + i; | |
1535 | if (unlikely(check_new_page(p))) | |
1536 | return 1; | |
1414c7f4 LA |
1537 | if (poisoned) |
1538 | poisoned &= page_is_poisoned(p); | |
2a7684a2 | 1539 | } |
689bcebf | 1540 | |
4c21e2f2 | 1541 | set_page_private(page, 0); |
7835e98b | 1542 | set_page_refcounted(page); |
cc102509 NP |
1543 | |
1544 | arch_alloc_page(page, order); | |
1da177e4 | 1545 | kernel_map_pages(page, 1 << order, 1); |
8823b1db | 1546 | kernel_poison_pages(page, 1 << order, 1); |
b8c73fc2 | 1547 | kasan_alloc_pages(page, order); |
17cf4406 | 1548 | |
1414c7f4 | 1549 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1550 | for (i = 0; i < (1 << order); i++) |
1551 | clear_highpage(page + i); | |
17cf4406 NP |
1552 | |
1553 | if (order && (gfp_flags & __GFP_COMP)) | |
1554 | prep_compound_page(page, order); | |
1555 | ||
48c96a36 JK |
1556 | set_page_owner(page, order, gfp_flags); |
1557 | ||
75379191 | 1558 | /* |
2f064f34 | 1559 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1560 | * allocate the page. The expectation is that the caller is taking |
1561 | * steps that will free more memory. The caller should avoid the page | |
1562 | * being used for !PFMEMALLOC purposes. | |
1563 | */ | |
2f064f34 MH |
1564 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1565 | set_page_pfmemalloc(page); | |
1566 | else | |
1567 | clear_page_pfmemalloc(page); | |
75379191 | 1568 | |
689bcebf | 1569 | return 0; |
1da177e4 LT |
1570 | } |
1571 | ||
56fd56b8 MG |
1572 | /* |
1573 | * Go through the free lists for the given migratetype and remove | |
1574 | * the smallest available page from the freelists | |
1575 | */ | |
728ec980 MG |
1576 | static inline |
1577 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1578 | int migratetype) |
1579 | { | |
1580 | unsigned int current_order; | |
b8af2941 | 1581 | struct free_area *area; |
56fd56b8 MG |
1582 | struct page *page; |
1583 | ||
1584 | /* Find a page of the appropriate size in the preferred list */ | |
1585 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1586 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1587 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1588 | struct page, lru); |
a16601c5 GT |
1589 | if (!page) |
1590 | continue; | |
56fd56b8 MG |
1591 | list_del(&page->lru); |
1592 | rmv_page_order(page); | |
1593 | area->nr_free--; | |
56fd56b8 | 1594 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1595 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1596 | return page; |
1597 | } | |
1598 | ||
1599 | return NULL; | |
1600 | } | |
1601 | ||
1602 | ||
b2a0ac88 MG |
1603 | /* |
1604 | * This array describes the order lists are fallen back to when | |
1605 | * the free lists for the desirable migrate type are depleted | |
1606 | */ | |
47118af0 | 1607 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1608 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1609 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1610 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1611 | #ifdef CONFIG_CMA |
974a786e | 1612 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1613 | #endif |
194159fb | 1614 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1615 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1616 | #endif |
b2a0ac88 MG |
1617 | }; |
1618 | ||
dc67647b JK |
1619 | #ifdef CONFIG_CMA |
1620 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1621 | unsigned int order) | |
1622 | { | |
1623 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1624 | } | |
1625 | #else | |
1626 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1627 | unsigned int order) { return NULL; } | |
1628 | #endif | |
1629 | ||
c361be55 MG |
1630 | /* |
1631 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1632 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1633 | * boundary. If alignment is required, use move_freepages_block() |
1634 | */ | |
435b405c | 1635 | int move_freepages(struct zone *zone, |
b69a7288 AB |
1636 | struct page *start_page, struct page *end_page, |
1637 | int migratetype) | |
c361be55 MG |
1638 | { |
1639 | struct page *page; | |
d00181b9 | 1640 | unsigned int order; |
d100313f | 1641 | int pages_moved = 0; |
c361be55 MG |
1642 | |
1643 | #ifndef CONFIG_HOLES_IN_ZONE | |
1644 | /* | |
1645 | * page_zone is not safe to call in this context when | |
1646 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1647 | * anyway as we check zone boundaries in move_freepages_block(). | |
1648 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1649 | * grouping pages by mobility |
c361be55 | 1650 | */ |
97ee4ba7 | 1651 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1652 | #endif |
1653 | ||
1654 | for (page = start_page; page <= end_page;) { | |
344c790e | 1655 | /* Make sure we are not inadvertently changing nodes */ |
309381fe | 1656 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
344c790e | 1657 | |
c361be55 MG |
1658 | if (!pfn_valid_within(page_to_pfn(page))) { |
1659 | page++; | |
1660 | continue; | |
1661 | } | |
1662 | ||
1663 | if (!PageBuddy(page)) { | |
1664 | page++; | |
1665 | continue; | |
1666 | } | |
1667 | ||
1668 | order = page_order(page); | |
84be48d8 KS |
1669 | list_move(&page->lru, |
1670 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 1671 | page += 1 << order; |
d100313f | 1672 | pages_moved += 1 << order; |
c361be55 MG |
1673 | } |
1674 | ||
d100313f | 1675 | return pages_moved; |
c361be55 MG |
1676 | } |
1677 | ||
ee6f509c | 1678 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1679 | int migratetype) |
c361be55 MG |
1680 | { |
1681 | unsigned long start_pfn, end_pfn; | |
1682 | struct page *start_page, *end_page; | |
1683 | ||
1684 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1685 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1686 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1687 | end_page = start_page + pageblock_nr_pages - 1; |
1688 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1689 | |
1690 | /* Do not cross zone boundaries */ | |
108bcc96 | 1691 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1692 | start_page = page; |
108bcc96 | 1693 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1694 | return 0; |
1695 | ||
1696 | return move_freepages(zone, start_page, end_page, migratetype); | |
1697 | } | |
1698 | ||
2f66a68f MG |
1699 | static void change_pageblock_range(struct page *pageblock_page, |
1700 | int start_order, int migratetype) | |
1701 | { | |
1702 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1703 | ||
1704 | while (nr_pageblocks--) { | |
1705 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1706 | pageblock_page += pageblock_nr_pages; | |
1707 | } | |
1708 | } | |
1709 | ||
fef903ef | 1710 | /* |
9c0415eb VB |
1711 | * When we are falling back to another migratetype during allocation, try to |
1712 | * steal extra free pages from the same pageblocks to satisfy further | |
1713 | * allocations, instead of polluting multiple pageblocks. | |
1714 | * | |
1715 | * If we are stealing a relatively large buddy page, it is likely there will | |
1716 | * be more free pages in the pageblock, so try to steal them all. For | |
1717 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1718 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1719 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1720 | * pageblocks. | |
fef903ef | 1721 | */ |
4eb7dce6 JK |
1722 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1723 | { | |
1724 | /* | |
1725 | * Leaving this order check is intended, although there is | |
1726 | * relaxed order check in next check. The reason is that | |
1727 | * we can actually steal whole pageblock if this condition met, | |
1728 | * but, below check doesn't guarantee it and that is just heuristic | |
1729 | * so could be changed anytime. | |
1730 | */ | |
1731 | if (order >= pageblock_order) | |
1732 | return true; | |
1733 | ||
1734 | if (order >= pageblock_order / 2 || | |
1735 | start_mt == MIGRATE_RECLAIMABLE || | |
1736 | start_mt == MIGRATE_UNMOVABLE || | |
1737 | page_group_by_mobility_disabled) | |
1738 | return true; | |
1739 | ||
1740 | return false; | |
1741 | } | |
1742 | ||
1743 | /* | |
1744 | * This function implements actual steal behaviour. If order is large enough, | |
1745 | * we can steal whole pageblock. If not, we first move freepages in this | |
1746 | * pageblock and check whether half of pages are moved or not. If half of | |
1747 | * pages are moved, we can change migratetype of pageblock and permanently | |
1748 | * use it's pages as requested migratetype in the future. | |
1749 | */ | |
1750 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1751 | int start_type) | |
fef903ef | 1752 | { |
d00181b9 | 1753 | unsigned int current_order = page_order(page); |
4eb7dce6 | 1754 | int pages; |
fef903ef | 1755 | |
fef903ef SB |
1756 | /* Take ownership for orders >= pageblock_order */ |
1757 | if (current_order >= pageblock_order) { | |
1758 | change_pageblock_range(page, current_order, start_type); | |
3a1086fb | 1759 | return; |
fef903ef SB |
1760 | } |
1761 | ||
4eb7dce6 | 1762 | pages = move_freepages_block(zone, page, start_type); |
fef903ef | 1763 | |
4eb7dce6 JK |
1764 | /* Claim the whole block if over half of it is free */ |
1765 | if (pages >= (1 << (pageblock_order-1)) || | |
1766 | page_group_by_mobility_disabled) | |
1767 | set_pageblock_migratetype(page, start_type); | |
1768 | } | |
1769 | ||
2149cdae JK |
1770 | /* |
1771 | * Check whether there is a suitable fallback freepage with requested order. | |
1772 | * If only_stealable is true, this function returns fallback_mt only if | |
1773 | * we can steal other freepages all together. This would help to reduce | |
1774 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1775 | */ | |
1776 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1777 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1778 | { |
1779 | int i; | |
1780 | int fallback_mt; | |
1781 | ||
1782 | if (area->nr_free == 0) | |
1783 | return -1; | |
1784 | ||
1785 | *can_steal = false; | |
1786 | for (i = 0;; i++) { | |
1787 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 1788 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
1789 | break; |
1790 | ||
1791 | if (list_empty(&area->free_list[fallback_mt])) | |
1792 | continue; | |
fef903ef | 1793 | |
4eb7dce6 JK |
1794 | if (can_steal_fallback(order, migratetype)) |
1795 | *can_steal = true; | |
1796 | ||
2149cdae JK |
1797 | if (!only_stealable) |
1798 | return fallback_mt; | |
1799 | ||
1800 | if (*can_steal) | |
1801 | return fallback_mt; | |
fef903ef | 1802 | } |
4eb7dce6 JK |
1803 | |
1804 | return -1; | |
fef903ef SB |
1805 | } |
1806 | ||
0aaa29a5 MG |
1807 | /* |
1808 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
1809 | * there are no empty page blocks that contain a page with a suitable order | |
1810 | */ | |
1811 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
1812 | unsigned int alloc_order) | |
1813 | { | |
1814 | int mt; | |
1815 | unsigned long max_managed, flags; | |
1816 | ||
1817 | /* | |
1818 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
1819 | * Check is race-prone but harmless. | |
1820 | */ | |
1821 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
1822 | if (zone->nr_reserved_highatomic >= max_managed) | |
1823 | return; | |
1824 | ||
1825 | spin_lock_irqsave(&zone->lock, flags); | |
1826 | ||
1827 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
1828 | if (zone->nr_reserved_highatomic >= max_managed) | |
1829 | goto out_unlock; | |
1830 | ||
1831 | /* Yoink! */ | |
1832 | mt = get_pageblock_migratetype(page); | |
1833 | if (mt != MIGRATE_HIGHATOMIC && | |
1834 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { | |
1835 | zone->nr_reserved_highatomic += pageblock_nr_pages; | |
1836 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
1837 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); | |
1838 | } | |
1839 | ||
1840 | out_unlock: | |
1841 | spin_unlock_irqrestore(&zone->lock, flags); | |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * Used when an allocation is about to fail under memory pressure. This | |
1846 | * potentially hurts the reliability of high-order allocations when under | |
1847 | * intense memory pressure but failed atomic allocations should be easier | |
1848 | * to recover from than an OOM. | |
1849 | */ | |
1850 | static void unreserve_highatomic_pageblock(const struct alloc_context *ac) | |
1851 | { | |
1852 | struct zonelist *zonelist = ac->zonelist; | |
1853 | unsigned long flags; | |
1854 | struct zoneref *z; | |
1855 | struct zone *zone; | |
1856 | struct page *page; | |
1857 | int order; | |
1858 | ||
1859 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
1860 | ac->nodemask) { | |
1861 | /* Preserve at least one pageblock */ | |
1862 | if (zone->nr_reserved_highatomic <= pageblock_nr_pages) | |
1863 | continue; | |
1864 | ||
1865 | spin_lock_irqsave(&zone->lock, flags); | |
1866 | for (order = 0; order < MAX_ORDER; order++) { | |
1867 | struct free_area *area = &(zone->free_area[order]); | |
1868 | ||
a16601c5 GT |
1869 | page = list_first_entry_or_null( |
1870 | &area->free_list[MIGRATE_HIGHATOMIC], | |
1871 | struct page, lru); | |
1872 | if (!page) | |
0aaa29a5 MG |
1873 | continue; |
1874 | ||
0aaa29a5 MG |
1875 | /* |
1876 | * It should never happen but changes to locking could | |
1877 | * inadvertently allow a per-cpu drain to add pages | |
1878 | * to MIGRATE_HIGHATOMIC while unreserving so be safe | |
1879 | * and watch for underflows. | |
1880 | */ | |
1881 | zone->nr_reserved_highatomic -= min(pageblock_nr_pages, | |
1882 | zone->nr_reserved_highatomic); | |
1883 | ||
1884 | /* | |
1885 | * Convert to ac->migratetype and avoid the normal | |
1886 | * pageblock stealing heuristics. Minimally, the caller | |
1887 | * is doing the work and needs the pages. More | |
1888 | * importantly, if the block was always converted to | |
1889 | * MIGRATE_UNMOVABLE or another type then the number | |
1890 | * of pageblocks that cannot be completely freed | |
1891 | * may increase. | |
1892 | */ | |
1893 | set_pageblock_migratetype(page, ac->migratetype); | |
1894 | move_freepages_block(zone, page, ac->migratetype); | |
1895 | spin_unlock_irqrestore(&zone->lock, flags); | |
1896 | return; | |
1897 | } | |
1898 | spin_unlock_irqrestore(&zone->lock, flags); | |
1899 | } | |
1900 | } | |
1901 | ||
b2a0ac88 | 1902 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 1903 | static inline struct page * |
7aeb09f9 | 1904 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 1905 | { |
b8af2941 | 1906 | struct free_area *area; |
7aeb09f9 | 1907 | unsigned int current_order; |
b2a0ac88 | 1908 | struct page *page; |
4eb7dce6 JK |
1909 | int fallback_mt; |
1910 | bool can_steal; | |
b2a0ac88 MG |
1911 | |
1912 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
1913 | for (current_order = MAX_ORDER-1; |
1914 | current_order >= order && current_order <= MAX_ORDER-1; | |
1915 | --current_order) { | |
4eb7dce6 JK |
1916 | area = &(zone->free_area[current_order]); |
1917 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 1918 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
1919 | if (fallback_mt == -1) |
1920 | continue; | |
b2a0ac88 | 1921 | |
a16601c5 | 1922 | page = list_first_entry(&area->free_list[fallback_mt], |
4eb7dce6 JK |
1923 | struct page, lru); |
1924 | if (can_steal) | |
1925 | steal_suitable_fallback(zone, page, start_migratetype); | |
b2a0ac88 | 1926 | |
4eb7dce6 JK |
1927 | /* Remove the page from the freelists */ |
1928 | area->nr_free--; | |
1929 | list_del(&page->lru); | |
1930 | rmv_page_order(page); | |
3a1086fb | 1931 | |
4eb7dce6 JK |
1932 | expand(zone, page, order, current_order, area, |
1933 | start_migratetype); | |
1934 | /* | |
bb14c2c7 | 1935 | * The pcppage_migratetype may differ from pageblock's |
4eb7dce6 | 1936 | * migratetype depending on the decisions in |
bb14c2c7 VB |
1937 | * find_suitable_fallback(). This is OK as long as it does not |
1938 | * differ for MIGRATE_CMA pageblocks. Those can be used as | |
1939 | * fallback only via special __rmqueue_cma_fallback() function | |
4eb7dce6 | 1940 | */ |
bb14c2c7 | 1941 | set_pcppage_migratetype(page, start_migratetype); |
e0fff1bd | 1942 | |
4eb7dce6 JK |
1943 | trace_mm_page_alloc_extfrag(page, order, current_order, |
1944 | start_migratetype, fallback_mt); | |
e0fff1bd | 1945 | |
4eb7dce6 | 1946 | return page; |
b2a0ac88 MG |
1947 | } |
1948 | ||
728ec980 | 1949 | return NULL; |
b2a0ac88 MG |
1950 | } |
1951 | ||
56fd56b8 | 1952 | /* |
1da177e4 LT |
1953 | * Do the hard work of removing an element from the buddy allocator. |
1954 | * Call me with the zone->lock already held. | |
1955 | */ | |
b2a0ac88 | 1956 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
6ac0206b | 1957 | int migratetype) |
1da177e4 | 1958 | { |
1da177e4 LT |
1959 | struct page *page; |
1960 | ||
56fd56b8 | 1961 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 1962 | if (unlikely(!page)) { |
dc67647b JK |
1963 | if (migratetype == MIGRATE_MOVABLE) |
1964 | page = __rmqueue_cma_fallback(zone, order); | |
1965 | ||
1966 | if (!page) | |
1967 | page = __rmqueue_fallback(zone, order, migratetype); | |
728ec980 MG |
1968 | } |
1969 | ||
0d3d062a | 1970 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1971 | return page; |
1da177e4 LT |
1972 | } |
1973 | ||
5f63b720 | 1974 | /* |
1da177e4 LT |
1975 | * Obtain a specified number of elements from the buddy allocator, all under |
1976 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1977 | * Returns the number of new pages which were placed at *list. | |
1978 | */ | |
5f63b720 | 1979 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1980 | unsigned long count, struct list_head *list, |
b745bc85 | 1981 | int migratetype, bool cold) |
1da177e4 | 1982 | { |
5bcc9f86 | 1983 | int i; |
5f63b720 | 1984 | |
c54ad30c | 1985 | spin_lock(&zone->lock); |
1da177e4 | 1986 | for (i = 0; i < count; ++i) { |
6ac0206b | 1987 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1988 | if (unlikely(page == NULL)) |
1da177e4 | 1989 | break; |
81eabcbe MG |
1990 | |
1991 | /* | |
1992 | * Split buddy pages returned by expand() are received here | |
1993 | * in physical page order. The page is added to the callers and | |
1994 | * list and the list head then moves forward. From the callers | |
1995 | * perspective, the linked list is ordered by page number in | |
1996 | * some conditions. This is useful for IO devices that can | |
1997 | * merge IO requests if the physical pages are ordered | |
1998 | * properly. | |
1999 | */ | |
b745bc85 | 2000 | if (likely(!cold)) |
e084b2d9 MG |
2001 | list_add(&page->lru, list); |
2002 | else | |
2003 | list_add_tail(&page->lru, list); | |
81eabcbe | 2004 | list = &page->lru; |
bb14c2c7 | 2005 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2006 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2007 | -(1 << order)); | |
1da177e4 | 2008 | } |
f2260e6b | 2009 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 2010 | spin_unlock(&zone->lock); |
085cc7d5 | 2011 | return i; |
1da177e4 LT |
2012 | } |
2013 | ||
4ae7c039 | 2014 | #ifdef CONFIG_NUMA |
8fce4d8e | 2015 | /* |
4037d452 CL |
2016 | * Called from the vmstat counter updater to drain pagesets of this |
2017 | * currently executing processor on remote nodes after they have | |
2018 | * expired. | |
2019 | * | |
879336c3 CL |
2020 | * Note that this function must be called with the thread pinned to |
2021 | * a single processor. | |
8fce4d8e | 2022 | */ |
4037d452 | 2023 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2024 | { |
4ae7c039 | 2025 | unsigned long flags; |
7be12fc9 | 2026 | int to_drain, batch; |
4ae7c039 | 2027 | |
4037d452 | 2028 | local_irq_save(flags); |
4db0c3c2 | 2029 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2030 | to_drain = min(pcp->count, batch); |
2a13515c KM |
2031 | if (to_drain > 0) { |
2032 | free_pcppages_bulk(zone, to_drain, pcp); | |
2033 | pcp->count -= to_drain; | |
2034 | } | |
4037d452 | 2035 | local_irq_restore(flags); |
4ae7c039 CL |
2036 | } |
2037 | #endif | |
2038 | ||
9f8f2172 | 2039 | /* |
93481ff0 | 2040 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2041 | * |
2042 | * The processor must either be the current processor and the | |
2043 | * thread pinned to the current processor or a processor that | |
2044 | * is not online. | |
2045 | */ | |
93481ff0 | 2046 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2047 | { |
c54ad30c | 2048 | unsigned long flags; |
93481ff0 VB |
2049 | struct per_cpu_pageset *pset; |
2050 | struct per_cpu_pages *pcp; | |
1da177e4 | 2051 | |
93481ff0 VB |
2052 | local_irq_save(flags); |
2053 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2054 | |
93481ff0 VB |
2055 | pcp = &pset->pcp; |
2056 | if (pcp->count) { | |
2057 | free_pcppages_bulk(zone, pcp->count, pcp); | |
2058 | pcp->count = 0; | |
2059 | } | |
2060 | local_irq_restore(flags); | |
2061 | } | |
3dfa5721 | 2062 | |
93481ff0 VB |
2063 | /* |
2064 | * Drain pcplists of all zones on the indicated processor. | |
2065 | * | |
2066 | * The processor must either be the current processor and the | |
2067 | * thread pinned to the current processor or a processor that | |
2068 | * is not online. | |
2069 | */ | |
2070 | static void drain_pages(unsigned int cpu) | |
2071 | { | |
2072 | struct zone *zone; | |
2073 | ||
2074 | for_each_populated_zone(zone) { | |
2075 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2076 | } |
2077 | } | |
1da177e4 | 2078 | |
9f8f2172 CL |
2079 | /* |
2080 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2081 | * |
2082 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2083 | * the single zone's pages. | |
9f8f2172 | 2084 | */ |
93481ff0 | 2085 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2086 | { |
93481ff0 VB |
2087 | int cpu = smp_processor_id(); |
2088 | ||
2089 | if (zone) | |
2090 | drain_pages_zone(cpu, zone); | |
2091 | else | |
2092 | drain_pages(cpu); | |
9f8f2172 CL |
2093 | } |
2094 | ||
2095 | /* | |
74046494 GBY |
2096 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2097 | * | |
93481ff0 VB |
2098 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2099 | * | |
74046494 GBY |
2100 | * Note that this code is protected against sending an IPI to an offline |
2101 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
2102 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
2103 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
2104 | * before the call to on_each_cpu_mask(). | |
9f8f2172 | 2105 | */ |
93481ff0 | 2106 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2107 | { |
74046494 | 2108 | int cpu; |
74046494 GBY |
2109 | |
2110 | /* | |
2111 | * Allocate in the BSS so we wont require allocation in | |
2112 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2113 | */ | |
2114 | static cpumask_t cpus_with_pcps; | |
2115 | ||
2116 | /* | |
2117 | * We don't care about racing with CPU hotplug event | |
2118 | * as offline notification will cause the notified | |
2119 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2120 | * disables preemption as part of its processing | |
2121 | */ | |
2122 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2123 | struct per_cpu_pageset *pcp; |
2124 | struct zone *z; | |
74046494 | 2125 | bool has_pcps = false; |
93481ff0 VB |
2126 | |
2127 | if (zone) { | |
74046494 | 2128 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2129 | if (pcp->pcp.count) |
74046494 | 2130 | has_pcps = true; |
93481ff0 VB |
2131 | } else { |
2132 | for_each_populated_zone(z) { | |
2133 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2134 | if (pcp->pcp.count) { | |
2135 | has_pcps = true; | |
2136 | break; | |
2137 | } | |
74046494 GBY |
2138 | } |
2139 | } | |
93481ff0 | 2140 | |
74046494 GBY |
2141 | if (has_pcps) |
2142 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2143 | else | |
2144 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2145 | } | |
93481ff0 VB |
2146 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
2147 | zone, 1); | |
9f8f2172 CL |
2148 | } |
2149 | ||
296699de | 2150 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
2151 | |
2152 | void mark_free_pages(struct zone *zone) | |
2153 | { | |
f623f0db RW |
2154 | unsigned long pfn, max_zone_pfn; |
2155 | unsigned long flags; | |
7aeb09f9 | 2156 | unsigned int order, t; |
86760a2c | 2157 | struct page *page; |
1da177e4 | 2158 | |
8080fc03 | 2159 | if (zone_is_empty(zone)) |
1da177e4 LT |
2160 | return; |
2161 | ||
2162 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2163 | |
108bcc96 | 2164 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2165 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2166 | if (pfn_valid(pfn)) { | |
86760a2c | 2167 | page = pfn_to_page(pfn); |
ba6b0979 JK |
2168 | |
2169 | if (page_zone(page) != zone) | |
2170 | continue; | |
2171 | ||
7be98234 RW |
2172 | if (!swsusp_page_is_forbidden(page)) |
2173 | swsusp_unset_page_free(page); | |
f623f0db | 2174 | } |
1da177e4 | 2175 | |
b2a0ac88 | 2176 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2177 | list_for_each_entry(page, |
2178 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2179 | unsigned long i; |
1da177e4 | 2180 | |
86760a2c | 2181 | pfn = page_to_pfn(page); |
f623f0db | 2182 | for (i = 0; i < (1UL << order); i++) |
7be98234 | 2183 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 2184 | } |
b2a0ac88 | 2185 | } |
1da177e4 LT |
2186 | spin_unlock_irqrestore(&zone->lock, flags); |
2187 | } | |
e2c55dc8 | 2188 | #endif /* CONFIG_PM */ |
1da177e4 | 2189 | |
1da177e4 LT |
2190 | /* |
2191 | * Free a 0-order page | |
b745bc85 | 2192 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 2193 | */ |
b745bc85 | 2194 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
2195 | { |
2196 | struct zone *zone = page_zone(page); | |
2197 | struct per_cpu_pages *pcp; | |
2198 | unsigned long flags; | |
dc4b0caf | 2199 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 2200 | int migratetype; |
1da177e4 | 2201 | |
ec95f53a | 2202 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
2203 | return; |
2204 | ||
dc4b0caf | 2205 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2206 | set_pcppage_migratetype(page, migratetype); |
1da177e4 | 2207 | local_irq_save(flags); |
f8891e5e | 2208 | __count_vm_event(PGFREE); |
da456f14 | 2209 | |
5f8dcc21 MG |
2210 | /* |
2211 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2212 | * Free ISOLATE pages back to the allocator because they are being | |
2213 | * offlined but treat RESERVE as movable pages so we can get those | |
2214 | * areas back if necessary. Otherwise, we may have to free | |
2215 | * excessively into the page allocator | |
2216 | */ | |
2217 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2218 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2219 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
2220 | goto out; |
2221 | } | |
2222 | migratetype = MIGRATE_MOVABLE; | |
2223 | } | |
2224 | ||
99dcc3e5 | 2225 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 2226 | if (!cold) |
5f8dcc21 | 2227 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
2228 | else |
2229 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 2230 | pcp->count++; |
48db57f8 | 2231 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2232 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
2233 | free_pcppages_bulk(zone, batch, pcp); |
2234 | pcp->count -= batch; | |
48db57f8 | 2235 | } |
5f8dcc21 MG |
2236 | |
2237 | out: | |
1da177e4 | 2238 | local_irq_restore(flags); |
1da177e4 LT |
2239 | } |
2240 | ||
cc59850e KK |
2241 | /* |
2242 | * Free a list of 0-order pages | |
2243 | */ | |
b745bc85 | 2244 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
2245 | { |
2246 | struct page *page, *next; | |
2247 | ||
2248 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 2249 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
2250 | free_hot_cold_page(page, cold); |
2251 | } | |
2252 | } | |
2253 | ||
8dfcc9ba NP |
2254 | /* |
2255 | * split_page takes a non-compound higher-order page, and splits it into | |
2256 | * n (1<<order) sub-pages: page[0..n] | |
2257 | * Each sub-page must be freed individually. | |
2258 | * | |
2259 | * Note: this is probably too low level an operation for use in drivers. | |
2260 | * Please consult with lkml before using this in your driver. | |
2261 | */ | |
2262 | void split_page(struct page *page, unsigned int order) | |
2263 | { | |
2264 | int i; | |
e2cfc911 | 2265 | gfp_t gfp_mask; |
8dfcc9ba | 2266 | |
309381fe SL |
2267 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2268 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
2269 | |
2270 | #ifdef CONFIG_KMEMCHECK | |
2271 | /* | |
2272 | * Split shadow pages too, because free(page[0]) would | |
2273 | * otherwise free the whole shadow. | |
2274 | */ | |
2275 | if (kmemcheck_page_is_tracked(page)) | |
2276 | split_page(virt_to_page(page[0].shadow), order); | |
2277 | #endif | |
2278 | ||
e2cfc911 JK |
2279 | gfp_mask = get_page_owner_gfp(page); |
2280 | set_page_owner(page, 0, gfp_mask); | |
48c96a36 | 2281 | for (i = 1; i < (1 << order); i++) { |
7835e98b | 2282 | set_page_refcounted(page + i); |
e2cfc911 | 2283 | set_page_owner(page + i, 0, gfp_mask); |
48c96a36 | 2284 | } |
8dfcc9ba | 2285 | } |
5853ff23 | 2286 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2287 | |
3c605096 | 2288 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2289 | { |
748446bb MG |
2290 | unsigned long watermark; |
2291 | struct zone *zone; | |
2139cbe6 | 2292 | int mt; |
748446bb MG |
2293 | |
2294 | BUG_ON(!PageBuddy(page)); | |
2295 | ||
2296 | zone = page_zone(page); | |
2e30abd1 | 2297 | mt = get_pageblock_migratetype(page); |
748446bb | 2298 | |
194159fb | 2299 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
2300 | /* Obey watermarks as if the page was being allocated */ |
2301 | watermark = low_wmark_pages(zone) + (1 << order); | |
2302 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
2303 | return 0; | |
2304 | ||
8fb74b9f | 2305 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2306 | } |
748446bb MG |
2307 | |
2308 | /* Remove page from free list */ | |
2309 | list_del(&page->lru); | |
2310 | zone->free_area[order].nr_free--; | |
2311 | rmv_page_order(page); | |
2139cbe6 | 2312 | |
e2cfc911 | 2313 | set_page_owner(page, order, __GFP_MOVABLE); |
f3a14ced | 2314 | |
8fb74b9f | 2315 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
2316 | if (order >= pageblock_order - 1) { |
2317 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2318 | for (; page < endpage; page += pageblock_nr_pages) { |
2319 | int mt = get_pageblock_migratetype(page); | |
194159fb | 2320 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
2321 | set_pageblock_migratetype(page, |
2322 | MIGRATE_MOVABLE); | |
2323 | } | |
748446bb MG |
2324 | } |
2325 | ||
f3a14ced | 2326 | |
8fb74b9f | 2327 | return 1UL << order; |
1fb3f8ca MG |
2328 | } |
2329 | ||
2330 | /* | |
2331 | * Similar to split_page except the page is already free. As this is only | |
2332 | * being used for migration, the migratetype of the block also changes. | |
2333 | * As this is called with interrupts disabled, the caller is responsible | |
2334 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
2335 | * are enabled. | |
2336 | * | |
2337 | * Note: this is probably too low level an operation for use in drivers. | |
2338 | * Please consult with lkml before using this in your driver. | |
2339 | */ | |
2340 | int split_free_page(struct page *page) | |
2341 | { | |
2342 | unsigned int order; | |
2343 | int nr_pages; | |
2344 | ||
1fb3f8ca MG |
2345 | order = page_order(page); |
2346 | ||
8fb74b9f | 2347 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
2348 | if (!nr_pages) |
2349 | return 0; | |
2350 | ||
2351 | /* Split into individual pages */ | |
2352 | set_page_refcounted(page); | |
2353 | split_page(page, order); | |
2354 | return nr_pages; | |
748446bb MG |
2355 | } |
2356 | ||
060e7417 MG |
2357 | /* |
2358 | * Update NUMA hit/miss statistics | |
2359 | * | |
2360 | * Must be called with interrupts disabled. | |
2361 | * | |
2362 | * When __GFP_OTHER_NODE is set assume the node of the preferred | |
2363 | * zone is the local node. This is useful for daemons who allocate | |
2364 | * memory on behalf of other processes. | |
2365 | */ | |
2366 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, | |
2367 | gfp_t flags) | |
2368 | { | |
2369 | #ifdef CONFIG_NUMA | |
2370 | int local_nid = numa_node_id(); | |
2371 | enum zone_stat_item local_stat = NUMA_LOCAL; | |
2372 | ||
2373 | if (unlikely(flags & __GFP_OTHER_NODE)) { | |
2374 | local_stat = NUMA_OTHER; | |
2375 | local_nid = preferred_zone->node; | |
2376 | } | |
2377 | ||
2378 | if (z->node == local_nid) { | |
2379 | __inc_zone_state(z, NUMA_HIT); | |
2380 | __inc_zone_state(z, local_stat); | |
2381 | } else { | |
2382 | __inc_zone_state(z, NUMA_MISS); | |
2383 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); | |
2384 | } | |
2385 | #endif | |
2386 | } | |
2387 | ||
1da177e4 | 2388 | /* |
75379191 | 2389 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2390 | */ |
0a15c3e9 MG |
2391 | static inline |
2392 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
7aeb09f9 | 2393 | struct zone *zone, unsigned int order, |
c603844b MG |
2394 | gfp_t gfp_flags, unsigned int alloc_flags, |
2395 | int migratetype) | |
1da177e4 LT |
2396 | { |
2397 | unsigned long flags; | |
689bcebf | 2398 | struct page *page; |
b745bc85 | 2399 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
1da177e4 | 2400 | |
48db57f8 | 2401 | if (likely(order == 0)) { |
1da177e4 | 2402 | struct per_cpu_pages *pcp; |
5f8dcc21 | 2403 | struct list_head *list; |
1da177e4 | 2404 | |
1da177e4 | 2405 | local_irq_save(flags); |
99dcc3e5 CL |
2406 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2407 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 2408 | if (list_empty(list)) { |
535131e6 | 2409 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 2410 | pcp->batch, list, |
e084b2d9 | 2411 | migratetype, cold); |
5f8dcc21 | 2412 | if (unlikely(list_empty(list))) |
6fb332fa | 2413 | goto failed; |
535131e6 | 2414 | } |
b92a6edd | 2415 | |
5f8dcc21 | 2416 | if (cold) |
a16601c5 | 2417 | page = list_last_entry(list, struct page, lru); |
5f8dcc21 | 2418 | else |
a16601c5 | 2419 | page = list_first_entry(list, struct page, lru); |
5f8dcc21 | 2420 | |
754078eb | 2421 | __dec_zone_state(zone, NR_ALLOC_BATCH); |
b92a6edd MG |
2422 | list_del(&page->lru); |
2423 | pcp->count--; | |
7fb1d9fc | 2424 | } else { |
0f352e53 MH |
2425 | /* |
2426 | * We most definitely don't want callers attempting to | |
2427 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2428 | */ | |
2429 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
1da177e4 | 2430 | spin_lock_irqsave(&zone->lock, flags); |
0aaa29a5 MG |
2431 | |
2432 | page = NULL; | |
2433 | if (alloc_flags & ALLOC_HARDER) { | |
2434 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2435 | if (page) | |
2436 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
2437 | } | |
2438 | if (!page) | |
6ac0206b | 2439 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
2440 | spin_unlock(&zone->lock); |
2441 | if (!page) | |
2442 | goto failed; | |
754078eb | 2443 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
d1ce749a | 2444 | __mod_zone_freepage_state(zone, -(1 << order), |
bb14c2c7 | 2445 | get_pcppage_migratetype(page)); |
1da177e4 LT |
2446 | } |
2447 | ||
abe5f972 | 2448 | if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 && |
57054651 JW |
2449 | !test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) |
2450 | set_bit(ZONE_FAIR_DEPLETED, &zone->flags); | |
27329369 | 2451 | |
f8891e5e | 2452 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 2453 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 2454 | local_irq_restore(flags); |
1da177e4 | 2455 | |
309381fe | 2456 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 2457 | return page; |
a74609fa NP |
2458 | |
2459 | failed: | |
2460 | local_irq_restore(flags); | |
a74609fa | 2461 | return NULL; |
1da177e4 LT |
2462 | } |
2463 | ||
933e312e AM |
2464 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2465 | ||
b2588c4b | 2466 | static struct { |
933e312e AM |
2467 | struct fault_attr attr; |
2468 | ||
621a5f7a | 2469 | bool ignore_gfp_highmem; |
71baba4b | 2470 | bool ignore_gfp_reclaim; |
54114994 | 2471 | u32 min_order; |
933e312e AM |
2472 | } fail_page_alloc = { |
2473 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 2474 | .ignore_gfp_reclaim = true, |
621a5f7a | 2475 | .ignore_gfp_highmem = true, |
54114994 | 2476 | .min_order = 1, |
933e312e AM |
2477 | }; |
2478 | ||
2479 | static int __init setup_fail_page_alloc(char *str) | |
2480 | { | |
2481 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
2482 | } | |
2483 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
2484 | ||
deaf386e | 2485 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2486 | { |
54114994 | 2487 | if (order < fail_page_alloc.min_order) |
deaf386e | 2488 | return false; |
933e312e | 2489 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 2490 | return false; |
933e312e | 2491 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 2492 | return false; |
71baba4b MG |
2493 | if (fail_page_alloc.ignore_gfp_reclaim && |
2494 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 2495 | return false; |
933e312e AM |
2496 | |
2497 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
2498 | } | |
2499 | ||
2500 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
2501 | ||
2502 | static int __init fail_page_alloc_debugfs(void) | |
2503 | { | |
f4ae40a6 | 2504 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 2505 | struct dentry *dir; |
933e312e | 2506 | |
dd48c085 AM |
2507 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2508 | &fail_page_alloc.attr); | |
2509 | if (IS_ERR(dir)) | |
2510 | return PTR_ERR(dir); | |
933e312e | 2511 | |
b2588c4b | 2512 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 2513 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
2514 | goto fail; |
2515 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
2516 | &fail_page_alloc.ignore_gfp_highmem)) | |
2517 | goto fail; | |
2518 | if (!debugfs_create_u32("min-order", mode, dir, | |
2519 | &fail_page_alloc.min_order)) | |
2520 | goto fail; | |
2521 | ||
2522 | return 0; | |
2523 | fail: | |
dd48c085 | 2524 | debugfs_remove_recursive(dir); |
933e312e | 2525 | |
b2588c4b | 2526 | return -ENOMEM; |
933e312e AM |
2527 | } |
2528 | ||
2529 | late_initcall(fail_page_alloc_debugfs); | |
2530 | ||
2531 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2532 | ||
2533 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2534 | ||
deaf386e | 2535 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2536 | { |
deaf386e | 2537 | return false; |
933e312e AM |
2538 | } |
2539 | ||
2540 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2541 | ||
1da177e4 | 2542 | /* |
97a16fc8 MG |
2543 | * Return true if free base pages are above 'mark'. For high-order checks it |
2544 | * will return true of the order-0 watermark is reached and there is at least | |
2545 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
2546 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 2547 | */ |
7aeb09f9 | 2548 | static bool __zone_watermark_ok(struct zone *z, unsigned int order, |
c603844b MG |
2549 | unsigned long mark, int classzone_idx, |
2550 | unsigned int alloc_flags, | |
7aeb09f9 | 2551 | long free_pages) |
1da177e4 | 2552 | { |
d23ad423 | 2553 | long min = mark; |
1da177e4 | 2554 | int o; |
c603844b | 2555 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
1da177e4 | 2556 | |
0aaa29a5 | 2557 | /* free_pages may go negative - that's OK */ |
df0a6daa | 2558 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 2559 | |
7fb1d9fc | 2560 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2561 | min -= min / 2; |
0aaa29a5 MG |
2562 | |
2563 | /* | |
2564 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
2565 | * the high-atomic reserves. This will over-estimate the size of the | |
2566 | * atomic reserve but it avoids a search. | |
2567 | */ | |
97a16fc8 | 2568 | if (likely(!alloc_harder)) |
0aaa29a5 MG |
2569 | free_pages -= z->nr_reserved_highatomic; |
2570 | else | |
1da177e4 | 2571 | min -= min / 4; |
e2b19197 | 2572 | |
d95ea5d1 BZ |
2573 | #ifdef CONFIG_CMA |
2574 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2575 | if (!(alloc_flags & ALLOC_CMA)) | |
97a16fc8 | 2576 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2577 | #endif |
026b0814 | 2578 | |
97a16fc8 MG |
2579 | /* |
2580 | * Check watermarks for an order-0 allocation request. If these | |
2581 | * are not met, then a high-order request also cannot go ahead | |
2582 | * even if a suitable page happened to be free. | |
2583 | */ | |
2584 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 2585 | return false; |
1da177e4 | 2586 | |
97a16fc8 MG |
2587 | /* If this is an order-0 request then the watermark is fine */ |
2588 | if (!order) | |
2589 | return true; | |
2590 | ||
2591 | /* For a high-order request, check at least one suitable page is free */ | |
2592 | for (o = order; o < MAX_ORDER; o++) { | |
2593 | struct free_area *area = &z->free_area[o]; | |
2594 | int mt; | |
2595 | ||
2596 | if (!area->nr_free) | |
2597 | continue; | |
2598 | ||
2599 | if (alloc_harder) | |
2600 | return true; | |
1da177e4 | 2601 | |
97a16fc8 MG |
2602 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
2603 | if (!list_empty(&area->free_list[mt])) | |
2604 | return true; | |
2605 | } | |
2606 | ||
2607 | #ifdef CONFIG_CMA | |
2608 | if ((alloc_flags & ALLOC_CMA) && | |
2609 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
2610 | return true; | |
2611 | } | |
2612 | #endif | |
1da177e4 | 2613 | } |
97a16fc8 | 2614 | return false; |
88f5acf8 MG |
2615 | } |
2616 | ||
7aeb09f9 | 2617 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 2618 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
2619 | { |
2620 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2621 | zone_page_state(z, NR_FREE_PAGES)); | |
2622 | } | |
2623 | ||
7aeb09f9 | 2624 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 2625 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
2626 | { |
2627 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2628 | ||
2629 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2630 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2631 | ||
e2b19197 | 2632 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 2633 | free_pages); |
1da177e4 LT |
2634 | } |
2635 | ||
9276b1bc | 2636 | #ifdef CONFIG_NUMA |
81c0a2bb JW |
2637 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2638 | { | |
fff4068c | 2639 | return local_zone->node == zone->node; |
81c0a2bb JW |
2640 | } |
2641 | ||
957f822a DR |
2642 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2643 | { | |
5f7a75ac MG |
2644 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
2645 | RECLAIM_DISTANCE; | |
957f822a | 2646 | } |
9276b1bc | 2647 | #else /* CONFIG_NUMA */ |
81c0a2bb JW |
2648 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2649 | { | |
2650 | return true; | |
2651 | } | |
2652 | ||
957f822a DR |
2653 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2654 | { | |
2655 | return true; | |
2656 | } | |
9276b1bc PJ |
2657 | #endif /* CONFIG_NUMA */ |
2658 | ||
4ffeaf35 MG |
2659 | static void reset_alloc_batches(struct zone *preferred_zone) |
2660 | { | |
2661 | struct zone *zone = preferred_zone->zone_pgdat->node_zones; | |
2662 | ||
2663 | do { | |
2664 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
2665 | high_wmark_pages(zone) - low_wmark_pages(zone) - | |
2666 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
57054651 | 2667 | clear_bit(ZONE_FAIR_DEPLETED, &zone->flags); |
4ffeaf35 MG |
2668 | } while (zone++ != preferred_zone); |
2669 | } | |
2670 | ||
7fb1d9fc | 2671 | /* |
0798e519 | 2672 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
2673 | * a page. |
2674 | */ | |
2675 | static struct page * | |
a9263751 VB |
2676 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
2677 | const struct alloc_context *ac) | |
753ee728 | 2678 | { |
dd1a239f | 2679 | struct zoneref *z; |
5117f45d | 2680 | struct zone *zone; |
fa379b95 | 2681 | bool fair_skipped; |
4ffeaf35 | 2682 | bool zonelist_rescan; |
54a6eb5c | 2683 | |
9276b1bc | 2684 | zonelist_scan: |
4ffeaf35 MG |
2685 | zonelist_rescan = false; |
2686 | ||
7fb1d9fc | 2687 | /* |
9276b1bc | 2688 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 2689 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 2690 | */ |
4dfa6cd8 | 2691 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 2692 | ac->nodemask) { |
be06af00 | 2693 | struct page *page; |
e085dbc5 JW |
2694 | unsigned long mark; |
2695 | ||
664eedde MG |
2696 | if (cpusets_enabled() && |
2697 | (alloc_flags & ALLOC_CPUSET) && | |
344736f2 | 2698 | !cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 2699 | continue; |
81c0a2bb JW |
2700 | /* |
2701 | * Distribute pages in proportion to the individual | |
2702 | * zone size to ensure fair page aging. The zone a | |
2703 | * page was allocated in should have no effect on the | |
2704 | * time the page has in memory before being reclaimed. | |
81c0a2bb | 2705 | */ |
3a025760 | 2706 | if (alloc_flags & ALLOC_FAIR) { |
a9263751 | 2707 | if (!zone_local(ac->preferred_zone, zone)) |
f7b5d647 | 2708 | break; |
57054651 | 2709 | if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) { |
fa379b95 | 2710 | fair_skipped = true; |
3a025760 | 2711 | continue; |
4ffeaf35 | 2712 | } |
81c0a2bb | 2713 | } |
a756cf59 JW |
2714 | /* |
2715 | * When allocating a page cache page for writing, we | |
2716 | * want to get it from a zone that is within its dirty | |
2717 | * limit, such that no single zone holds more than its | |
2718 | * proportional share of globally allowed dirty pages. | |
2719 | * The dirty limits take into account the zone's | |
2720 | * lowmem reserves and high watermark so that kswapd | |
2721 | * should be able to balance it without having to | |
2722 | * write pages from its LRU list. | |
2723 | * | |
2724 | * This may look like it could increase pressure on | |
2725 | * lower zones by failing allocations in higher zones | |
2726 | * before they are full. But the pages that do spill | |
2727 | * over are limited as the lower zones are protected | |
2728 | * by this very same mechanism. It should not become | |
2729 | * a practical burden to them. | |
2730 | * | |
2731 | * XXX: For now, allow allocations to potentially | |
2732 | * exceed the per-zone dirty limit in the slowpath | |
c9ab0c4f | 2733 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 JW |
2734 | * which is important when on a NUMA setup the allowed |
2735 | * zones are together not big enough to reach the | |
2736 | * global limit. The proper fix for these situations | |
2737 | * will require awareness of zones in the | |
2738 | * dirty-throttling and the flusher threads. | |
2739 | */ | |
c9ab0c4f | 2740 | if (ac->spread_dirty_pages && !zone_dirty_ok(zone)) |
800a1e75 | 2741 | continue; |
7fb1d9fc | 2742 | |
e085dbc5 JW |
2743 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
2744 | if (!zone_watermark_ok(zone, order, mark, | |
a9263751 | 2745 | ac->classzone_idx, alloc_flags)) { |
fa5e084e MG |
2746 | int ret; |
2747 | ||
5dab2911 MG |
2748 | /* Checked here to keep the fast path fast */ |
2749 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
2750 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
2751 | goto try_this_zone; | |
2752 | ||
957f822a | 2753 | if (zone_reclaim_mode == 0 || |
a9263751 | 2754 | !zone_allows_reclaim(ac->preferred_zone, zone)) |
cd38b115 MG |
2755 | continue; |
2756 | ||
fa5e084e MG |
2757 | ret = zone_reclaim(zone, gfp_mask, order); |
2758 | switch (ret) { | |
2759 | case ZONE_RECLAIM_NOSCAN: | |
2760 | /* did not scan */ | |
cd38b115 | 2761 | continue; |
fa5e084e MG |
2762 | case ZONE_RECLAIM_FULL: |
2763 | /* scanned but unreclaimable */ | |
cd38b115 | 2764 | continue; |
fa5e084e MG |
2765 | default: |
2766 | /* did we reclaim enough */ | |
fed2719e | 2767 | if (zone_watermark_ok(zone, order, mark, |
a9263751 | 2768 | ac->classzone_idx, alloc_flags)) |
fed2719e MG |
2769 | goto try_this_zone; |
2770 | ||
fed2719e | 2771 | continue; |
0798e519 | 2772 | } |
7fb1d9fc RS |
2773 | } |
2774 | ||
fa5e084e | 2775 | try_this_zone: |
a9263751 | 2776 | page = buffered_rmqueue(ac->preferred_zone, zone, order, |
0aaa29a5 | 2777 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 VB |
2778 | if (page) { |
2779 | if (prep_new_page(page, order, gfp_mask, alloc_flags)) | |
2780 | goto try_this_zone; | |
0aaa29a5 MG |
2781 | |
2782 | /* | |
2783 | * If this is a high-order atomic allocation then check | |
2784 | * if the pageblock should be reserved for the future | |
2785 | */ | |
2786 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
2787 | reserve_highatomic_pageblock(page, zone, order); | |
2788 | ||
75379191 VB |
2789 | return page; |
2790 | } | |
54a6eb5c | 2791 | } |
9276b1bc | 2792 | |
4ffeaf35 MG |
2793 | /* |
2794 | * The first pass makes sure allocations are spread fairly within the | |
2795 | * local node. However, the local node might have free pages left | |
2796 | * after the fairness batches are exhausted, and remote zones haven't | |
2797 | * even been considered yet. Try once more without fairness, and | |
2798 | * include remote zones now, before entering the slowpath and waking | |
2799 | * kswapd: prefer spilling to a remote zone over swapping locally. | |
2800 | */ | |
2801 | if (alloc_flags & ALLOC_FAIR) { | |
2802 | alloc_flags &= ~ALLOC_FAIR; | |
fa379b95 | 2803 | if (fair_skipped) { |
4ffeaf35 | 2804 | zonelist_rescan = true; |
a9263751 | 2805 | reset_alloc_batches(ac->preferred_zone); |
4ffeaf35 MG |
2806 | } |
2807 | if (nr_online_nodes > 1) | |
2808 | zonelist_rescan = true; | |
2809 | } | |
2810 | ||
4ffeaf35 MG |
2811 | if (zonelist_rescan) |
2812 | goto zonelist_scan; | |
2813 | ||
2814 | return NULL; | |
753ee728 MH |
2815 | } |
2816 | ||
29423e77 DR |
2817 | /* |
2818 | * Large machines with many possible nodes should not always dump per-node | |
2819 | * meminfo in irq context. | |
2820 | */ | |
2821 | static inline bool should_suppress_show_mem(void) | |
2822 | { | |
2823 | bool ret = false; | |
2824 | ||
2825 | #if NODES_SHIFT > 8 | |
2826 | ret = in_interrupt(); | |
2827 | #endif | |
2828 | return ret; | |
2829 | } | |
2830 | ||
a238ab5b DH |
2831 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2832 | DEFAULT_RATELIMIT_INTERVAL, | |
2833 | DEFAULT_RATELIMIT_BURST); | |
2834 | ||
d00181b9 | 2835 | void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...) |
a238ab5b | 2836 | { |
a238ab5b DH |
2837 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2838 | ||
c0a32fc5 SG |
2839 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2840 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2841 | return; |
2842 | ||
2843 | /* | |
2844 | * This documents exceptions given to allocations in certain | |
2845 | * contexts that are allowed to allocate outside current's set | |
2846 | * of allowed nodes. | |
2847 | */ | |
2848 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2849 | if (test_thread_flag(TIF_MEMDIE) || | |
2850 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2851 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 2852 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
2853 | filter &= ~SHOW_MEM_FILTER_NODES; |
2854 | ||
2855 | if (fmt) { | |
3ee9a4f0 JP |
2856 | struct va_format vaf; |
2857 | va_list args; | |
2858 | ||
a238ab5b | 2859 | va_start(args, fmt); |
3ee9a4f0 JP |
2860 | |
2861 | vaf.fmt = fmt; | |
2862 | vaf.va = &args; | |
2863 | ||
2864 | pr_warn("%pV", &vaf); | |
2865 | ||
a238ab5b DH |
2866 | va_end(args); |
2867 | } | |
2868 | ||
c5c990e8 VB |
2869 | pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n", |
2870 | current->comm, order, gfp_mask, &gfp_mask); | |
a238ab5b DH |
2871 | dump_stack(); |
2872 | if (!should_suppress_show_mem()) | |
2873 | show_mem(filter); | |
2874 | } | |
2875 | ||
11e33f6a MG |
2876 | static inline struct page * |
2877 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 2878 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 2879 | { |
6e0fc46d DR |
2880 | struct oom_control oc = { |
2881 | .zonelist = ac->zonelist, | |
2882 | .nodemask = ac->nodemask, | |
2883 | .gfp_mask = gfp_mask, | |
2884 | .order = order, | |
6e0fc46d | 2885 | }; |
11e33f6a MG |
2886 | struct page *page; |
2887 | ||
9879de73 JW |
2888 | *did_some_progress = 0; |
2889 | ||
9879de73 | 2890 | /* |
dc56401f JW |
2891 | * Acquire the oom lock. If that fails, somebody else is |
2892 | * making progress for us. | |
9879de73 | 2893 | */ |
dc56401f | 2894 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 2895 | *did_some_progress = 1; |
11e33f6a | 2896 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2897 | return NULL; |
2898 | } | |
6b1de916 | 2899 | |
11e33f6a MG |
2900 | /* |
2901 | * Go through the zonelist yet one more time, keep very high watermark | |
2902 | * here, this is only to catch a parallel oom killing, we must fail if | |
2903 | * we're still under heavy pressure. | |
2904 | */ | |
a9263751 VB |
2905 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
2906 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 2907 | if (page) |
11e33f6a MG |
2908 | goto out; |
2909 | ||
4365a567 | 2910 | if (!(gfp_mask & __GFP_NOFAIL)) { |
9879de73 JW |
2911 | /* Coredumps can quickly deplete all memory reserves */ |
2912 | if (current->flags & PF_DUMPCORE) | |
2913 | goto out; | |
4365a567 KH |
2914 | /* The OOM killer will not help higher order allocs */ |
2915 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2916 | goto out; | |
03668b3c | 2917 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
a9263751 | 2918 | if (ac->high_zoneidx < ZONE_NORMAL) |
03668b3c | 2919 | goto out; |
9083905a JW |
2920 | if (pm_suspended_storage()) |
2921 | goto out; | |
3da88fb3 MH |
2922 | /* |
2923 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
2924 | * other request to make a forward progress. | |
2925 | * We are in an unfortunate situation where out_of_memory cannot | |
2926 | * do much for this context but let's try it to at least get | |
2927 | * access to memory reserved if the current task is killed (see | |
2928 | * out_of_memory). Once filesystems are ready to handle allocation | |
2929 | * failures more gracefully we should just bail out here. | |
2930 | */ | |
2931 | ||
4167e9b2 | 2932 | /* The OOM killer may not free memory on a specific node */ |
4365a567 KH |
2933 | if (gfp_mask & __GFP_THISNODE) |
2934 | goto out; | |
2935 | } | |
11e33f6a | 2936 | /* Exhausted what can be done so it's blamo time */ |
5020e285 | 2937 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 2938 | *did_some_progress = 1; |
5020e285 MH |
2939 | |
2940 | if (gfp_mask & __GFP_NOFAIL) { | |
2941 | page = get_page_from_freelist(gfp_mask, order, | |
2942 | ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac); | |
2943 | /* | |
2944 | * fallback to ignore cpuset restriction if our nodes | |
2945 | * are depleted | |
2946 | */ | |
2947 | if (!page) | |
2948 | page = get_page_from_freelist(gfp_mask, order, | |
2949 | ALLOC_NO_WATERMARKS, ac); | |
2950 | } | |
2951 | } | |
11e33f6a | 2952 | out: |
dc56401f | 2953 | mutex_unlock(&oom_lock); |
11e33f6a MG |
2954 | return page; |
2955 | } | |
2956 | ||
56de7263 MG |
2957 | #ifdef CONFIG_COMPACTION |
2958 | /* Try memory compaction for high-order allocations before reclaim */ | |
2959 | static struct page * | |
2960 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 2961 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 VB |
2962 | enum migrate_mode mode, int *contended_compaction, |
2963 | bool *deferred_compaction) | |
56de7263 | 2964 | { |
53853e2d | 2965 | unsigned long compact_result; |
98dd3b48 | 2966 | struct page *page; |
53853e2d VB |
2967 | |
2968 | if (!order) | |
66199712 | 2969 | return NULL; |
66199712 | 2970 | |
c06b1fca | 2971 | current->flags |= PF_MEMALLOC; |
1a6d53a1 VB |
2972 | compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
2973 | mode, contended_compaction); | |
c06b1fca | 2974 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2975 | |
98dd3b48 VB |
2976 | switch (compact_result) { |
2977 | case COMPACT_DEFERRED: | |
53853e2d | 2978 | *deferred_compaction = true; |
98dd3b48 VB |
2979 | /* fall-through */ |
2980 | case COMPACT_SKIPPED: | |
2981 | return NULL; | |
2982 | default: | |
2983 | break; | |
2984 | } | |
53853e2d | 2985 | |
98dd3b48 VB |
2986 | /* |
2987 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
2988 | * count a compaction stall | |
2989 | */ | |
2990 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 2991 | |
a9263751 VB |
2992 | page = get_page_from_freelist(gfp_mask, order, |
2993 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
53853e2d | 2994 | |
98dd3b48 VB |
2995 | if (page) { |
2996 | struct zone *zone = page_zone(page); | |
53853e2d | 2997 | |
98dd3b48 VB |
2998 | zone->compact_blockskip_flush = false; |
2999 | compaction_defer_reset(zone, order, true); | |
3000 | count_vm_event(COMPACTSUCCESS); | |
3001 | return page; | |
3002 | } | |
56de7263 | 3003 | |
98dd3b48 VB |
3004 | /* |
3005 | * It's bad if compaction run occurs and fails. The most likely reason | |
3006 | * is that pages exist, but not enough to satisfy watermarks. | |
3007 | */ | |
3008 | count_vm_event(COMPACTFAIL); | |
66199712 | 3009 | |
98dd3b48 | 3010 | cond_resched(); |
56de7263 MG |
3011 | |
3012 | return NULL; | |
3013 | } | |
3014 | #else | |
3015 | static inline struct page * | |
3016 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3017 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 VB |
3018 | enum migrate_mode mode, int *contended_compaction, |
3019 | bool *deferred_compaction) | |
56de7263 MG |
3020 | { |
3021 | return NULL; | |
3022 | } | |
3023 | #endif /* CONFIG_COMPACTION */ | |
3024 | ||
bba90710 MS |
3025 | /* Perform direct synchronous page reclaim */ |
3026 | static int | |
a9263751 VB |
3027 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3028 | const struct alloc_context *ac) | |
11e33f6a | 3029 | { |
11e33f6a | 3030 | struct reclaim_state reclaim_state; |
bba90710 | 3031 | int progress; |
11e33f6a MG |
3032 | |
3033 | cond_resched(); | |
3034 | ||
3035 | /* We now go into synchronous reclaim */ | |
3036 | cpuset_memory_pressure_bump(); | |
c06b1fca | 3037 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
3038 | lockdep_set_current_reclaim_state(gfp_mask); |
3039 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 3040 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3041 | |
a9263751 VB |
3042 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3043 | ac->nodemask); | |
11e33f6a | 3044 | |
c06b1fca | 3045 | current->reclaim_state = NULL; |
11e33f6a | 3046 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 3047 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
3048 | |
3049 | cond_resched(); | |
3050 | ||
bba90710 MS |
3051 | return progress; |
3052 | } | |
3053 | ||
3054 | /* The really slow allocator path where we enter direct reclaim */ | |
3055 | static inline struct page * | |
3056 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3057 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3058 | unsigned long *did_some_progress) |
bba90710 MS |
3059 | { |
3060 | struct page *page = NULL; | |
3061 | bool drained = false; | |
3062 | ||
a9263751 | 3063 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3064 | if (unlikely(!(*did_some_progress))) |
3065 | return NULL; | |
11e33f6a | 3066 | |
9ee493ce | 3067 | retry: |
a9263751 VB |
3068 | page = get_page_from_freelist(gfp_mask, order, |
3069 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
9ee493ce MG |
3070 | |
3071 | /* | |
3072 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3073 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3074 | * Shrink them them and try again | |
9ee493ce MG |
3075 | */ |
3076 | if (!page && !drained) { | |
0aaa29a5 | 3077 | unreserve_highatomic_pageblock(ac); |
93481ff0 | 3078 | drain_all_pages(NULL); |
9ee493ce MG |
3079 | drained = true; |
3080 | goto retry; | |
3081 | } | |
3082 | ||
11e33f6a MG |
3083 | return page; |
3084 | } | |
3085 | ||
a9263751 | 3086 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
3087 | { |
3088 | struct zoneref *z; | |
3089 | struct zone *zone; | |
3090 | ||
a9263751 VB |
3091 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3092 | ac->high_zoneidx, ac->nodemask) | |
3093 | wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone)); | |
3a025760 JW |
3094 | } |
3095 | ||
c603844b | 3096 | static inline unsigned int |
341ce06f PZ |
3097 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3098 | { | |
c603844b | 3099 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3100 | |
a56f57ff | 3101 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3102 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3103 | |
341ce06f PZ |
3104 | /* |
3105 | * The caller may dip into page reserves a bit more if the caller | |
3106 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3107 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3108 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3109 | */ |
e6223a3b | 3110 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3111 | |
d0164adc | 3112 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3113 | /* |
b104a35d DR |
3114 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3115 | * if it can't schedule. | |
5c3240d9 | 3116 | */ |
b104a35d | 3117 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3118 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3119 | /* |
b104a35d | 3120 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3121 | * comment for __cpuset_node_allowed(). |
523b9458 | 3122 | */ |
341ce06f | 3123 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3124 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3125 | alloc_flags |= ALLOC_HARDER; |
3126 | ||
b37f1dd0 MG |
3127 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
3128 | if (gfp_mask & __GFP_MEMALLOC) | |
3129 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
3130 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
3131 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
3132 | else if (!in_interrupt() && | |
3133 | ((current->flags & PF_MEMALLOC) || | |
3134 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 3135 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 3136 | } |
d95ea5d1 | 3137 | #ifdef CONFIG_CMA |
43e7a34d | 3138 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
3139 | alloc_flags |= ALLOC_CMA; |
3140 | #endif | |
341ce06f PZ |
3141 | return alloc_flags; |
3142 | } | |
3143 | ||
072bb0aa MG |
3144 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3145 | { | |
b37f1dd0 | 3146 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
3147 | } |
3148 | ||
d0164adc MG |
3149 | static inline bool is_thp_gfp_mask(gfp_t gfp_mask) |
3150 | { | |
3151 | return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE; | |
3152 | } | |
3153 | ||
11e33f6a MG |
3154 | static inline struct page * |
3155 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3156 | struct alloc_context *ac) |
11e33f6a | 3157 | { |
d0164adc | 3158 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
11e33f6a | 3159 | struct page *page = NULL; |
c603844b | 3160 | unsigned int alloc_flags; |
11e33f6a MG |
3161 | unsigned long pages_reclaimed = 0; |
3162 | unsigned long did_some_progress; | |
e0b9daeb | 3163 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
66199712 | 3164 | bool deferred_compaction = false; |
1f9efdef | 3165 | int contended_compaction = COMPACT_CONTENDED_NONE; |
1da177e4 | 3166 | |
72807a74 MG |
3167 | /* |
3168 | * In the slowpath, we sanity check order to avoid ever trying to | |
3169 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
3170 | * be using allocators in order of preference for an area that is | |
3171 | * too large. | |
3172 | */ | |
1fc28b70 MG |
3173 | if (order >= MAX_ORDER) { |
3174 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 3175 | return NULL; |
1fc28b70 | 3176 | } |
1da177e4 | 3177 | |
d0164adc MG |
3178 | /* |
3179 | * We also sanity check to catch abuse of atomic reserves being used by | |
3180 | * callers that are not in atomic context. | |
3181 | */ | |
3182 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
3183 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
3184 | gfp_mask &= ~__GFP_ATOMIC; | |
3185 | ||
9879de73 | 3186 | retry: |
d0164adc | 3187 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
a9263751 | 3188 | wake_all_kswapds(order, ac); |
1da177e4 | 3189 | |
9bf2229f | 3190 | /* |
7fb1d9fc RS |
3191 | * OK, we're below the kswapd watermark and have kicked background |
3192 | * reclaim. Now things get more complex, so set up alloc_flags according | |
3193 | * to how we want to proceed. | |
9bf2229f | 3194 | */ |
341ce06f | 3195 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 3196 | |
341ce06f | 3197 | /* This is the last chance, in general, before the goto nopage. */ |
a9263751 VB |
3198 | page = get_page_from_freelist(gfp_mask, order, |
3199 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
7fb1d9fc RS |
3200 | if (page) |
3201 | goto got_pg; | |
1da177e4 | 3202 | |
11e33f6a | 3203 | /* Allocate without watermarks if the context allows */ |
341ce06f | 3204 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
3205 | /* |
3206 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
3207 | * the allocation is high priority and these type of | |
3208 | * allocations are system rather than user orientated | |
3209 | */ | |
a9263751 | 3210 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
33d53103 MH |
3211 | page = get_page_from_freelist(gfp_mask, order, |
3212 | ALLOC_NO_WATERMARKS, ac); | |
3213 | if (page) | |
3214 | goto got_pg; | |
1da177e4 LT |
3215 | } |
3216 | ||
d0164adc MG |
3217 | /* Caller is not willing to reclaim, we can't balance anything */ |
3218 | if (!can_direct_reclaim) { | |
aed0a0e3 | 3219 | /* |
33d53103 MH |
3220 | * All existing users of the __GFP_NOFAIL are blockable, so warn |
3221 | * of any new users that actually allow this type of allocation | |
3222 | * to fail. | |
aed0a0e3 DR |
3223 | */ |
3224 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 3225 | goto nopage; |
aed0a0e3 | 3226 | } |
1da177e4 | 3227 | |
341ce06f | 3228 | /* Avoid recursion of direct reclaim */ |
33d53103 MH |
3229 | if (current->flags & PF_MEMALLOC) { |
3230 | /* | |
3231 | * __GFP_NOFAIL request from this context is rather bizarre | |
3232 | * because we cannot reclaim anything and only can loop waiting | |
3233 | * for somebody to do a work for us. | |
3234 | */ | |
3235 | if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { | |
3236 | cond_resched(); | |
3237 | goto retry; | |
3238 | } | |
341ce06f | 3239 | goto nopage; |
33d53103 | 3240 | } |
341ce06f | 3241 | |
6583bb64 DR |
3242 | /* Avoid allocations with no watermarks from looping endlessly */ |
3243 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
3244 | goto nopage; | |
3245 | ||
77f1fe6b MG |
3246 | /* |
3247 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
3248 | * attempts after direct reclaim are synchronous | |
3249 | */ | |
a9263751 VB |
3250 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
3251 | migration_mode, | |
3252 | &contended_compaction, | |
53853e2d | 3253 | &deferred_compaction); |
56de7263 MG |
3254 | if (page) |
3255 | goto got_pg; | |
75f30861 | 3256 | |
1f9efdef | 3257 | /* Checks for THP-specific high-order allocations */ |
d0164adc | 3258 | if (is_thp_gfp_mask(gfp_mask)) { |
1f9efdef VB |
3259 | /* |
3260 | * If compaction is deferred for high-order allocations, it is | |
3261 | * because sync compaction recently failed. If this is the case | |
3262 | * and the caller requested a THP allocation, we do not want | |
3263 | * to heavily disrupt the system, so we fail the allocation | |
3264 | * instead of entering direct reclaim. | |
3265 | */ | |
3266 | if (deferred_compaction) | |
3267 | goto nopage; | |
3268 | ||
3269 | /* | |
3270 | * In all zones where compaction was attempted (and not | |
3271 | * deferred or skipped), lock contention has been detected. | |
3272 | * For THP allocation we do not want to disrupt the others | |
3273 | * so we fallback to base pages instead. | |
3274 | */ | |
3275 | if (contended_compaction == COMPACT_CONTENDED_LOCK) | |
3276 | goto nopage; | |
3277 | ||
3278 | /* | |
3279 | * If compaction was aborted due to need_resched(), we do not | |
3280 | * want to further increase allocation latency, unless it is | |
3281 | * khugepaged trying to collapse. | |
3282 | */ | |
3283 | if (contended_compaction == COMPACT_CONTENDED_SCHED | |
3284 | && !(current->flags & PF_KTHREAD)) | |
3285 | goto nopage; | |
3286 | } | |
66199712 | 3287 | |
8fe78048 DR |
3288 | /* |
3289 | * It can become very expensive to allocate transparent hugepages at | |
3290 | * fault, so use asynchronous memory compaction for THP unless it is | |
3291 | * khugepaged trying to collapse. | |
3292 | */ | |
d0164adc | 3293 | if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD)) |
8fe78048 DR |
3294 | migration_mode = MIGRATE_SYNC_LIGHT; |
3295 | ||
11e33f6a | 3296 | /* Try direct reclaim and then allocating */ |
a9263751 VB |
3297 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, |
3298 | &did_some_progress); | |
11e33f6a MG |
3299 | if (page) |
3300 | goto got_pg; | |
1da177e4 | 3301 | |
9083905a JW |
3302 | /* Do not loop if specifically requested */ |
3303 | if (gfp_mask & __GFP_NORETRY) | |
3304 | goto noretry; | |
3305 | ||
3306 | /* Keep reclaiming pages as long as there is reasonable progress */ | |
a41f24ea | 3307 | pages_reclaimed += did_some_progress; |
9083905a JW |
3308 | if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) || |
3309 | ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) { | |
11e33f6a | 3310 | /* Wait for some write requests to complete then retry */ |
a9263751 | 3311 | wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50); |
9879de73 | 3312 | goto retry; |
1da177e4 LT |
3313 | } |
3314 | ||
9083905a JW |
3315 | /* Reclaim has failed us, start killing things */ |
3316 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
3317 | if (page) | |
3318 | goto got_pg; | |
3319 | ||
3320 | /* Retry as long as the OOM killer is making progress */ | |
3321 | if (did_some_progress) | |
3322 | goto retry; | |
3323 | ||
3324 | noretry: | |
3325 | /* | |
3326 | * High-order allocations do not necessarily loop after | |
3327 | * direct reclaim and reclaim/compaction depends on compaction | |
3328 | * being called after reclaim so call directly if necessary | |
3329 | */ | |
3330 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, | |
3331 | ac, migration_mode, | |
3332 | &contended_compaction, | |
3333 | &deferred_compaction); | |
3334 | if (page) | |
3335 | goto got_pg; | |
1da177e4 | 3336 | nopage: |
a238ab5b | 3337 | warn_alloc_failed(gfp_mask, order, NULL); |
1da177e4 | 3338 | got_pg: |
072bb0aa | 3339 | return page; |
1da177e4 | 3340 | } |
11e33f6a MG |
3341 | |
3342 | /* | |
3343 | * This is the 'heart' of the zoned buddy allocator. | |
3344 | */ | |
3345 | struct page * | |
3346 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
3347 | struct zonelist *zonelist, nodemask_t *nodemask) | |
3348 | { | |
d8846374 | 3349 | struct zoneref *preferred_zoneref; |
cc9a6c87 | 3350 | struct page *page = NULL; |
cc9a6c87 | 3351 | unsigned int cpuset_mems_cookie; |
c603844b | 3352 | unsigned int alloc_flags = ALLOC_WMARK_LOW|ALLOC_FAIR; |
91fbdc0f | 3353 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
a9263751 VB |
3354 | struct alloc_context ac = { |
3355 | .high_zoneidx = gfp_zone(gfp_mask), | |
682a3385 | 3356 | .zonelist = zonelist, |
a9263751 VB |
3357 | .nodemask = nodemask, |
3358 | .migratetype = gfpflags_to_migratetype(gfp_mask), | |
3359 | }; | |
11e33f6a | 3360 | |
682a3385 MG |
3361 | if (cpusets_enabled()) { |
3362 | alloc_flags |= ALLOC_CPUSET; | |
3363 | if (!ac.nodemask) | |
3364 | ac.nodemask = &cpuset_current_mems_allowed; | |
3365 | } | |
3366 | ||
dcce284a BH |
3367 | gfp_mask &= gfp_allowed_mask; |
3368 | ||
11e33f6a MG |
3369 | lockdep_trace_alloc(gfp_mask); |
3370 | ||
d0164adc | 3371 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
3372 | |
3373 | if (should_fail_alloc_page(gfp_mask, order)) | |
3374 | return NULL; | |
3375 | ||
3376 | /* | |
3377 | * Check the zones suitable for the gfp_mask contain at least one | |
3378 | * valid zone. It's possible to have an empty zonelist as a result | |
4167e9b2 | 3379 | * of __GFP_THISNODE and a memoryless node |
11e33f6a MG |
3380 | */ |
3381 | if (unlikely(!zonelist->_zonerefs->zone)) | |
3382 | return NULL; | |
3383 | ||
a9263751 | 3384 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
21bb9bd1 VB |
3385 | alloc_flags |= ALLOC_CMA; |
3386 | ||
cc9a6c87 | 3387 | retry_cpuset: |
d26914d1 | 3388 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3389 | |
c9ab0c4f MG |
3390 | /* Dirty zone balancing only done in the fast path */ |
3391 | ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE); | |
3392 | ||
5117f45d | 3393 | /* The preferred zone is used for statistics later */ |
a9263751 | 3394 | preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx, |
682a3385 | 3395 | ac.nodemask, &ac.preferred_zone); |
a9263751 | 3396 | if (!ac.preferred_zone) |
cc9a6c87 | 3397 | goto out; |
a9263751 | 3398 | ac.classzone_idx = zonelist_zone_idx(preferred_zoneref); |
5117f45d MG |
3399 | |
3400 | /* First allocation attempt */ | |
91fbdc0f | 3401 | alloc_mask = gfp_mask|__GFP_HARDWALL; |
a9263751 | 3402 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
21caf2fc ML |
3403 | if (unlikely(!page)) { |
3404 | /* | |
3405 | * Runtime PM, block IO and its error handling path | |
3406 | * can deadlock because I/O on the device might not | |
3407 | * complete. | |
3408 | */ | |
91fbdc0f | 3409 | alloc_mask = memalloc_noio_flags(gfp_mask); |
c9ab0c4f | 3410 | ac.spread_dirty_pages = false; |
91fbdc0f | 3411 | |
a9263751 | 3412 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
21caf2fc | 3413 | } |
11e33f6a | 3414 | |
23f086f9 XQ |
3415 | if (kmemcheck_enabled && page) |
3416 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
3417 | ||
a9263751 | 3418 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
cc9a6c87 MG |
3419 | |
3420 | out: | |
3421 | /* | |
3422 | * When updating a task's mems_allowed, it is possible to race with | |
3423 | * parallel threads in such a way that an allocation can fail while | |
3424 | * the mask is being updated. If a page allocation is about to fail, | |
3425 | * check if the cpuset changed during allocation and if so, retry. | |
3426 | */ | |
d26914d1 | 3427 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
3428 | goto retry_cpuset; |
3429 | ||
11e33f6a | 3430 | return page; |
1da177e4 | 3431 | } |
d239171e | 3432 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
3433 | |
3434 | /* | |
3435 | * Common helper functions. | |
3436 | */ | |
920c7a5d | 3437 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 3438 | { |
945a1113 AM |
3439 | struct page *page; |
3440 | ||
3441 | /* | |
3442 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
3443 | * a highmem page | |
3444 | */ | |
3445 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
3446 | ||
1da177e4 LT |
3447 | page = alloc_pages(gfp_mask, order); |
3448 | if (!page) | |
3449 | return 0; | |
3450 | return (unsigned long) page_address(page); | |
3451 | } | |
1da177e4 LT |
3452 | EXPORT_SYMBOL(__get_free_pages); |
3453 | ||
920c7a5d | 3454 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 3455 | { |
945a1113 | 3456 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 3457 | } |
1da177e4 LT |
3458 | EXPORT_SYMBOL(get_zeroed_page); |
3459 | ||
920c7a5d | 3460 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 3461 | { |
b5810039 | 3462 | if (put_page_testzero(page)) { |
1da177e4 | 3463 | if (order == 0) |
b745bc85 | 3464 | free_hot_cold_page(page, false); |
1da177e4 LT |
3465 | else |
3466 | __free_pages_ok(page, order); | |
3467 | } | |
3468 | } | |
3469 | ||
3470 | EXPORT_SYMBOL(__free_pages); | |
3471 | ||
920c7a5d | 3472 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
3473 | { |
3474 | if (addr != 0) { | |
725d704e | 3475 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
3476 | __free_pages(virt_to_page((void *)addr), order); |
3477 | } | |
3478 | } | |
3479 | ||
3480 | EXPORT_SYMBOL(free_pages); | |
3481 | ||
b63ae8ca AD |
3482 | /* |
3483 | * Page Fragment: | |
3484 | * An arbitrary-length arbitrary-offset area of memory which resides | |
3485 | * within a 0 or higher order page. Multiple fragments within that page | |
3486 | * are individually refcounted, in the page's reference counter. | |
3487 | * | |
3488 | * The page_frag functions below provide a simple allocation framework for | |
3489 | * page fragments. This is used by the network stack and network device | |
3490 | * drivers to provide a backing region of memory for use as either an | |
3491 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
3492 | */ | |
3493 | static struct page *__page_frag_refill(struct page_frag_cache *nc, | |
3494 | gfp_t gfp_mask) | |
3495 | { | |
3496 | struct page *page = NULL; | |
3497 | gfp_t gfp = gfp_mask; | |
3498 | ||
3499 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3500 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
3501 | __GFP_NOMEMALLOC; | |
3502 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
3503 | PAGE_FRAG_CACHE_MAX_ORDER); | |
3504 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
3505 | #endif | |
3506 | if (unlikely(!page)) | |
3507 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
3508 | ||
3509 | nc->va = page ? page_address(page) : NULL; | |
3510 | ||
3511 | return page; | |
3512 | } | |
3513 | ||
3514 | void *__alloc_page_frag(struct page_frag_cache *nc, | |
3515 | unsigned int fragsz, gfp_t gfp_mask) | |
3516 | { | |
3517 | unsigned int size = PAGE_SIZE; | |
3518 | struct page *page; | |
3519 | int offset; | |
3520 | ||
3521 | if (unlikely(!nc->va)) { | |
3522 | refill: | |
3523 | page = __page_frag_refill(nc, gfp_mask); | |
3524 | if (!page) | |
3525 | return NULL; | |
3526 | ||
3527 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3528 | /* if size can vary use size else just use PAGE_SIZE */ | |
3529 | size = nc->size; | |
3530 | #endif | |
3531 | /* Even if we own the page, we do not use atomic_set(). | |
3532 | * This would break get_page_unless_zero() users. | |
3533 | */ | |
fe896d18 | 3534 | page_ref_add(page, size - 1); |
b63ae8ca AD |
3535 | |
3536 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 3537 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
3538 | nc->pagecnt_bias = size; |
3539 | nc->offset = size; | |
3540 | } | |
3541 | ||
3542 | offset = nc->offset - fragsz; | |
3543 | if (unlikely(offset < 0)) { | |
3544 | page = virt_to_page(nc->va); | |
3545 | ||
fe896d18 | 3546 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
3547 | goto refill; |
3548 | ||
3549 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3550 | /* if size can vary use size else just use PAGE_SIZE */ | |
3551 | size = nc->size; | |
3552 | #endif | |
3553 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 3554 | set_page_count(page, size); |
b63ae8ca AD |
3555 | |
3556 | /* reset page count bias and offset to start of new frag */ | |
3557 | nc->pagecnt_bias = size; | |
3558 | offset = size - fragsz; | |
3559 | } | |
3560 | ||
3561 | nc->pagecnt_bias--; | |
3562 | nc->offset = offset; | |
3563 | ||
3564 | return nc->va + offset; | |
3565 | } | |
3566 | EXPORT_SYMBOL(__alloc_page_frag); | |
3567 | ||
3568 | /* | |
3569 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
3570 | */ | |
3571 | void __free_page_frag(void *addr) | |
3572 | { | |
3573 | struct page *page = virt_to_head_page(addr); | |
3574 | ||
3575 | if (unlikely(put_page_testzero(page))) | |
3576 | __free_pages_ok(page, compound_order(page)); | |
3577 | } | |
3578 | EXPORT_SYMBOL(__free_page_frag); | |
3579 | ||
6a1a0d3b | 3580 | /* |
52383431 | 3581 | * alloc_kmem_pages charges newly allocated pages to the kmem resource counter |
a9bb7e62 VD |
3582 | * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is |
3583 | * equivalent to alloc_pages. | |
6a1a0d3b | 3584 | * |
52383431 VD |
3585 | * It should be used when the caller would like to use kmalloc, but since the |
3586 | * allocation is large, it has to fall back to the page allocator. | |
3587 | */ | |
3588 | struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order) | |
3589 | { | |
3590 | struct page *page; | |
52383431 | 3591 | |
52383431 | 3592 | page = alloc_pages(gfp_mask, order); |
d05e83a6 VD |
3593 | if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) { |
3594 | __free_pages(page, order); | |
3595 | page = NULL; | |
3596 | } | |
52383431 VD |
3597 | return page; |
3598 | } | |
3599 | ||
3600 | struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
3601 | { | |
3602 | struct page *page; | |
52383431 | 3603 | |
52383431 | 3604 | page = alloc_pages_node(nid, gfp_mask, order); |
d05e83a6 VD |
3605 | if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) { |
3606 | __free_pages(page, order); | |
3607 | page = NULL; | |
3608 | } | |
52383431 VD |
3609 | return page; |
3610 | } | |
3611 | ||
3612 | /* | |
3613 | * __free_kmem_pages and free_kmem_pages will free pages allocated with | |
3614 | * alloc_kmem_pages. | |
6a1a0d3b | 3615 | */ |
52383431 | 3616 | void __free_kmem_pages(struct page *page, unsigned int order) |
6a1a0d3b | 3617 | { |
d05e83a6 | 3618 | memcg_kmem_uncharge(page, order); |
6a1a0d3b GC |
3619 | __free_pages(page, order); |
3620 | } | |
3621 | ||
52383431 | 3622 | void free_kmem_pages(unsigned long addr, unsigned int order) |
6a1a0d3b GC |
3623 | { |
3624 | if (addr != 0) { | |
3625 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
52383431 | 3626 | __free_kmem_pages(virt_to_page((void *)addr), order); |
6a1a0d3b GC |
3627 | } |
3628 | } | |
3629 | ||
d00181b9 KS |
3630 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
3631 | size_t size) | |
ee85c2e1 AK |
3632 | { |
3633 | if (addr) { | |
3634 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
3635 | unsigned long used = addr + PAGE_ALIGN(size); | |
3636 | ||
3637 | split_page(virt_to_page((void *)addr), order); | |
3638 | while (used < alloc_end) { | |
3639 | free_page(used); | |
3640 | used += PAGE_SIZE; | |
3641 | } | |
3642 | } | |
3643 | return (void *)addr; | |
3644 | } | |
3645 | ||
2be0ffe2 TT |
3646 | /** |
3647 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
3648 | * @size: the number of bytes to allocate | |
3649 | * @gfp_mask: GFP flags for the allocation | |
3650 | * | |
3651 | * This function is similar to alloc_pages(), except that it allocates the | |
3652 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
3653 | * allocate memory in power-of-two pages. | |
3654 | * | |
3655 | * This function is also limited by MAX_ORDER. | |
3656 | * | |
3657 | * Memory allocated by this function must be released by free_pages_exact(). | |
3658 | */ | |
3659 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
3660 | { | |
3661 | unsigned int order = get_order(size); | |
3662 | unsigned long addr; | |
3663 | ||
3664 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 3665 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
3666 | } |
3667 | EXPORT_SYMBOL(alloc_pages_exact); | |
3668 | ||
ee85c2e1 AK |
3669 | /** |
3670 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
3671 | * pages on a node. | |
b5e6ab58 | 3672 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
3673 | * @size: the number of bytes to allocate |
3674 | * @gfp_mask: GFP flags for the allocation | |
3675 | * | |
3676 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
3677 | * back. | |
ee85c2e1 | 3678 | */ |
e1931811 | 3679 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 3680 | { |
d00181b9 | 3681 | unsigned int order = get_order(size); |
ee85c2e1 AK |
3682 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
3683 | if (!p) | |
3684 | return NULL; | |
3685 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
3686 | } | |
ee85c2e1 | 3687 | |
2be0ffe2 TT |
3688 | /** |
3689 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
3690 | * @virt: the value returned by alloc_pages_exact. | |
3691 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
3692 | * | |
3693 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
3694 | */ | |
3695 | void free_pages_exact(void *virt, size_t size) | |
3696 | { | |
3697 | unsigned long addr = (unsigned long)virt; | |
3698 | unsigned long end = addr + PAGE_ALIGN(size); | |
3699 | ||
3700 | while (addr < end) { | |
3701 | free_page(addr); | |
3702 | addr += PAGE_SIZE; | |
3703 | } | |
3704 | } | |
3705 | EXPORT_SYMBOL(free_pages_exact); | |
3706 | ||
e0fb5815 ZY |
3707 | /** |
3708 | * nr_free_zone_pages - count number of pages beyond high watermark | |
3709 | * @offset: The zone index of the highest zone | |
3710 | * | |
3711 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
3712 | * high watermark within all zones at or below a given zone index. For each | |
3713 | * zone, the number of pages is calculated as: | |
834405c3 | 3714 | * managed_pages - high_pages |
e0fb5815 | 3715 | */ |
ebec3862 | 3716 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 3717 | { |
dd1a239f | 3718 | struct zoneref *z; |
54a6eb5c MG |
3719 | struct zone *zone; |
3720 | ||
e310fd43 | 3721 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 3722 | unsigned long sum = 0; |
1da177e4 | 3723 | |
0e88460d | 3724 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 3725 | |
54a6eb5c | 3726 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 3727 | unsigned long size = zone->managed_pages; |
41858966 | 3728 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
3729 | if (size > high) |
3730 | sum += size - high; | |
1da177e4 LT |
3731 | } |
3732 | ||
3733 | return sum; | |
3734 | } | |
3735 | ||
e0fb5815 ZY |
3736 | /** |
3737 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
3738 | * | |
3739 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
3740 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 3741 | */ |
ebec3862 | 3742 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 3743 | { |
af4ca457 | 3744 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 3745 | } |
c2f1a551 | 3746 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 3747 | |
e0fb5815 ZY |
3748 | /** |
3749 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
3750 | * | |
3751 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
3752 | * high watermark within all zones. | |
1da177e4 | 3753 | */ |
ebec3862 | 3754 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 3755 | { |
2a1e274a | 3756 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 3757 | } |
08e0f6a9 CL |
3758 | |
3759 | static inline void show_node(struct zone *zone) | |
1da177e4 | 3760 | { |
e5adfffc | 3761 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 3762 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 3763 | } |
1da177e4 | 3764 | |
d02bd27b IR |
3765 | long si_mem_available(void) |
3766 | { | |
3767 | long available; | |
3768 | unsigned long pagecache; | |
3769 | unsigned long wmark_low = 0; | |
3770 | unsigned long pages[NR_LRU_LISTS]; | |
3771 | struct zone *zone; | |
3772 | int lru; | |
3773 | ||
3774 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
3775 | pages[lru] = global_page_state(NR_LRU_BASE + lru); | |
3776 | ||
3777 | for_each_zone(zone) | |
3778 | wmark_low += zone->watermark[WMARK_LOW]; | |
3779 | ||
3780 | /* | |
3781 | * Estimate the amount of memory available for userspace allocations, | |
3782 | * without causing swapping. | |
3783 | */ | |
3784 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; | |
3785 | ||
3786 | /* | |
3787 | * Not all the page cache can be freed, otherwise the system will | |
3788 | * start swapping. Assume at least half of the page cache, or the | |
3789 | * low watermark worth of cache, needs to stay. | |
3790 | */ | |
3791 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
3792 | pagecache -= min(pagecache / 2, wmark_low); | |
3793 | available += pagecache; | |
3794 | ||
3795 | /* | |
3796 | * Part of the reclaimable slab consists of items that are in use, | |
3797 | * and cannot be freed. Cap this estimate at the low watermark. | |
3798 | */ | |
3799 | available += global_page_state(NR_SLAB_RECLAIMABLE) - | |
3800 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); | |
3801 | ||
3802 | if (available < 0) | |
3803 | available = 0; | |
3804 | return available; | |
3805 | } | |
3806 | EXPORT_SYMBOL_GPL(si_mem_available); | |
3807 | ||
1da177e4 LT |
3808 | void si_meminfo(struct sysinfo *val) |
3809 | { | |
3810 | val->totalram = totalram_pages; | |
cc7452b6 | 3811 | val->sharedram = global_page_state(NR_SHMEM); |
d23ad423 | 3812 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 3813 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
3814 | val->totalhigh = totalhigh_pages; |
3815 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
3816 | val->mem_unit = PAGE_SIZE; |
3817 | } | |
3818 | ||
3819 | EXPORT_SYMBOL(si_meminfo); | |
3820 | ||
3821 | #ifdef CONFIG_NUMA | |
3822 | void si_meminfo_node(struct sysinfo *val, int nid) | |
3823 | { | |
cdd91a77 JL |
3824 | int zone_type; /* needs to be signed */ |
3825 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
3826 | unsigned long managed_highpages = 0; |
3827 | unsigned long free_highpages = 0; | |
1da177e4 LT |
3828 | pg_data_t *pgdat = NODE_DATA(nid); |
3829 | ||
cdd91a77 JL |
3830 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
3831 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
3832 | val->totalram = managed_pages; | |
cc7452b6 | 3833 | val->sharedram = node_page_state(nid, NR_SHMEM); |
d23ad423 | 3834 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 3835 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
3836 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
3837 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
3838 | ||
3839 | if (is_highmem(zone)) { | |
3840 | managed_highpages += zone->managed_pages; | |
3841 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
3842 | } | |
3843 | } | |
3844 | val->totalhigh = managed_highpages; | |
3845 | val->freehigh = free_highpages; | |
98d2b0eb | 3846 | #else |
fc2bd799 JK |
3847 | val->totalhigh = managed_highpages; |
3848 | val->freehigh = free_highpages; | |
98d2b0eb | 3849 | #endif |
1da177e4 LT |
3850 | val->mem_unit = PAGE_SIZE; |
3851 | } | |
3852 | #endif | |
3853 | ||
ddd588b5 | 3854 | /* |
7bf02ea2 DR |
3855 | * Determine whether the node should be displayed or not, depending on whether |
3856 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 3857 | */ |
7bf02ea2 | 3858 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
3859 | { |
3860 | bool ret = false; | |
cc9a6c87 | 3861 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
3862 | |
3863 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
3864 | goto out; | |
3865 | ||
cc9a6c87 | 3866 | do { |
d26914d1 | 3867 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3868 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 3869 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
3870 | out: |
3871 | return ret; | |
3872 | } | |
3873 | ||
1da177e4 LT |
3874 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
3875 | ||
377e4f16 RV |
3876 | static void show_migration_types(unsigned char type) |
3877 | { | |
3878 | static const char types[MIGRATE_TYPES] = { | |
3879 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 3880 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
3881 | [MIGRATE_RECLAIMABLE] = 'E', |
3882 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
3883 | #ifdef CONFIG_CMA |
3884 | [MIGRATE_CMA] = 'C', | |
3885 | #endif | |
194159fb | 3886 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 3887 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 3888 | #endif |
377e4f16 RV |
3889 | }; |
3890 | char tmp[MIGRATE_TYPES + 1]; | |
3891 | char *p = tmp; | |
3892 | int i; | |
3893 | ||
3894 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
3895 | if (type & (1 << i)) | |
3896 | *p++ = types[i]; | |
3897 | } | |
3898 | ||
3899 | *p = '\0'; | |
3900 | printk("(%s) ", tmp); | |
3901 | } | |
3902 | ||
1da177e4 LT |
3903 | /* |
3904 | * Show free area list (used inside shift_scroll-lock stuff) | |
3905 | * We also calculate the percentage fragmentation. We do this by counting the | |
3906 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
3907 | * |
3908 | * Bits in @filter: | |
3909 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
3910 | * cpuset. | |
1da177e4 | 3911 | */ |
7bf02ea2 | 3912 | void show_free_areas(unsigned int filter) |
1da177e4 | 3913 | { |
d1bfcdb8 | 3914 | unsigned long free_pcp = 0; |
c7241913 | 3915 | int cpu; |
1da177e4 LT |
3916 | struct zone *zone; |
3917 | ||
ee99c71c | 3918 | for_each_populated_zone(zone) { |
7bf02ea2 | 3919 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3920 | continue; |
d1bfcdb8 | 3921 | |
761b0677 KK |
3922 | for_each_online_cpu(cpu) |
3923 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
3924 | } |
3925 | ||
a731286d KM |
3926 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
3927 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
3928 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
3929 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 3930 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 3931 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
4f98a2fe | 3932 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 3933 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
3934 | global_page_state(NR_ISOLATED_ANON), |
3935 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 3936 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 3937 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 3938 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 3939 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 3940 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 3941 | global_page_state(NR_UNSTABLE_NFS), |
3701b033 KM |
3942 | global_page_state(NR_SLAB_RECLAIMABLE), |
3943 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 3944 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 3945 | global_page_state(NR_SHMEM), |
a25700a5 | 3946 | global_page_state(NR_PAGETABLE), |
d1ce749a | 3947 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
3948 | global_page_state(NR_FREE_PAGES), |
3949 | free_pcp, | |
d1ce749a | 3950 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 3951 | |
ee99c71c | 3952 | for_each_populated_zone(zone) { |
1da177e4 LT |
3953 | int i; |
3954 | ||
7bf02ea2 | 3955 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3956 | continue; |
d1bfcdb8 KK |
3957 | |
3958 | free_pcp = 0; | |
3959 | for_each_online_cpu(cpu) | |
3960 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
3961 | ||
1da177e4 LT |
3962 | show_node(zone); |
3963 | printk("%s" | |
3964 | " free:%lukB" | |
3965 | " min:%lukB" | |
3966 | " low:%lukB" | |
3967 | " high:%lukB" | |
4f98a2fe RR |
3968 | " active_anon:%lukB" |
3969 | " inactive_anon:%lukB" | |
3970 | " active_file:%lukB" | |
3971 | " inactive_file:%lukB" | |
7b854121 | 3972 | " unevictable:%lukB" |
a731286d KM |
3973 | " isolated(anon):%lukB" |
3974 | " isolated(file):%lukB" | |
1da177e4 | 3975 | " present:%lukB" |
9feedc9d | 3976 | " managed:%lukB" |
4a0aa73f KM |
3977 | " mlocked:%lukB" |
3978 | " dirty:%lukB" | |
3979 | " writeback:%lukB" | |
3980 | " mapped:%lukB" | |
4b02108a | 3981 | " shmem:%lukB" |
4a0aa73f KM |
3982 | " slab_reclaimable:%lukB" |
3983 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3984 | " kernel_stack:%lukB" |
4a0aa73f KM |
3985 | " pagetables:%lukB" |
3986 | " unstable:%lukB" | |
3987 | " bounce:%lukB" | |
d1bfcdb8 KK |
3988 | " free_pcp:%lukB" |
3989 | " local_pcp:%ukB" | |
d1ce749a | 3990 | " free_cma:%lukB" |
4a0aa73f | 3991 | " writeback_tmp:%lukB" |
1da177e4 LT |
3992 | " pages_scanned:%lu" |
3993 | " all_unreclaimable? %s" | |
3994 | "\n", | |
3995 | zone->name, | |
88f5acf8 | 3996 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3997 | K(min_wmark_pages(zone)), |
3998 | K(low_wmark_pages(zone)), | |
3999 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
4000 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
4001 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
4002 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
4003 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 4004 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
4005 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
4006 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 4007 | K(zone->present_pages), |
9feedc9d | 4008 | K(zone->managed_pages), |
4a0aa73f KM |
4009 | K(zone_page_state(zone, NR_MLOCK)), |
4010 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
4011 | K(zone_page_state(zone, NR_WRITEBACK)), | |
4012 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 4013 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
4014 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
4015 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
4016 | zone_page_state(zone, NR_KERNEL_STACK) * |
4017 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
4018 | K(zone_page_state(zone, NR_PAGETABLE)), |
4019 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
4020 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1bfcdb8 KK |
4021 | K(free_pcp), |
4022 | K(this_cpu_read(zone->pageset->pcp.count)), | |
d1ce749a | 4023 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 4024 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
0d5d823a | 4025 | K(zone_page_state(zone, NR_PAGES_SCANNED)), |
6e543d57 | 4026 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
4027 | ); |
4028 | printk("lowmem_reserve[]:"); | |
4029 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3484b2de | 4030 | printk(" %ld", zone->lowmem_reserve[i]); |
1da177e4 LT |
4031 | printk("\n"); |
4032 | } | |
4033 | ||
ee99c71c | 4034 | for_each_populated_zone(zone) { |
d00181b9 KS |
4035 | unsigned int order; |
4036 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 4037 | unsigned char types[MAX_ORDER]; |
1da177e4 | 4038 | |
7bf02ea2 | 4039 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4040 | continue; |
1da177e4 LT |
4041 | show_node(zone); |
4042 | printk("%s: ", zone->name); | |
1da177e4 LT |
4043 | |
4044 | spin_lock_irqsave(&zone->lock, flags); | |
4045 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
4046 | struct free_area *area = &zone->free_area[order]; |
4047 | int type; | |
4048 | ||
4049 | nr[order] = area->nr_free; | |
8f9de51a | 4050 | total += nr[order] << order; |
377e4f16 RV |
4051 | |
4052 | types[order] = 0; | |
4053 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
4054 | if (!list_empty(&area->free_list[type])) | |
4055 | types[order] |= 1 << type; | |
4056 | } | |
1da177e4 LT |
4057 | } |
4058 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 4059 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 4060 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
4061 | if (nr[order]) |
4062 | show_migration_types(types[order]); | |
4063 | } | |
1da177e4 LT |
4064 | printk("= %lukB\n", K(total)); |
4065 | } | |
4066 | ||
949f7ec5 DR |
4067 | hugetlb_show_meminfo(); |
4068 | ||
e6f3602d LW |
4069 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
4070 | ||
1da177e4 LT |
4071 | show_swap_cache_info(); |
4072 | } | |
4073 | ||
19770b32 MG |
4074 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4075 | { | |
4076 | zoneref->zone = zone; | |
4077 | zoneref->zone_idx = zone_idx(zone); | |
4078 | } | |
4079 | ||
1da177e4 LT |
4080 | /* |
4081 | * Builds allocation fallback zone lists. | |
1a93205b CL |
4082 | * |
4083 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 4084 | */ |
f0c0b2b8 | 4085 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 4086 | int nr_zones) |
1da177e4 | 4087 | { |
1a93205b | 4088 | struct zone *zone; |
bc732f1d | 4089 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
4090 | |
4091 | do { | |
2f6726e5 | 4092 | zone_type--; |
070f8032 | 4093 | zone = pgdat->node_zones + zone_type; |
1a93205b | 4094 | if (populated_zone(zone)) { |
dd1a239f MG |
4095 | zoneref_set_zone(zone, |
4096 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 4097 | check_highest_zone(zone_type); |
1da177e4 | 4098 | } |
2f6726e5 | 4099 | } while (zone_type); |
bc732f1d | 4100 | |
070f8032 | 4101 | return nr_zones; |
1da177e4 LT |
4102 | } |
4103 | ||
f0c0b2b8 KH |
4104 | |
4105 | /* | |
4106 | * zonelist_order: | |
4107 | * 0 = automatic detection of better ordering. | |
4108 | * 1 = order by ([node] distance, -zonetype) | |
4109 | * 2 = order by (-zonetype, [node] distance) | |
4110 | * | |
4111 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
4112 | * the same zonelist. So only NUMA can configure this param. | |
4113 | */ | |
4114 | #define ZONELIST_ORDER_DEFAULT 0 | |
4115 | #define ZONELIST_ORDER_NODE 1 | |
4116 | #define ZONELIST_ORDER_ZONE 2 | |
4117 | ||
4118 | /* zonelist order in the kernel. | |
4119 | * set_zonelist_order() will set this to NODE or ZONE. | |
4120 | */ | |
4121 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4122 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
4123 | ||
4124 | ||
1da177e4 | 4125 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
4126 | /* The value user specified ....changed by config */ |
4127 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4128 | /* string for sysctl */ | |
4129 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
4130 | char numa_zonelist_order[16] = "default"; | |
4131 | ||
4132 | /* | |
4133 | * interface for configure zonelist ordering. | |
4134 | * command line option "numa_zonelist_order" | |
4135 | * = "[dD]efault - default, automatic configuration. | |
4136 | * = "[nN]ode - order by node locality, then by zone within node | |
4137 | * = "[zZ]one - order by zone, then by locality within zone | |
4138 | */ | |
4139 | ||
4140 | static int __parse_numa_zonelist_order(char *s) | |
4141 | { | |
4142 | if (*s == 'd' || *s == 'D') { | |
4143 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4144 | } else if (*s == 'n' || *s == 'N') { | |
4145 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
4146 | } else if (*s == 'z' || *s == 'Z') { | |
4147 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
4148 | } else { | |
1170532b | 4149 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
f0c0b2b8 KH |
4150 | return -EINVAL; |
4151 | } | |
4152 | return 0; | |
4153 | } | |
4154 | ||
4155 | static __init int setup_numa_zonelist_order(char *s) | |
4156 | { | |
ecb256f8 VL |
4157 | int ret; |
4158 | ||
4159 | if (!s) | |
4160 | return 0; | |
4161 | ||
4162 | ret = __parse_numa_zonelist_order(s); | |
4163 | if (ret == 0) | |
4164 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
4165 | ||
4166 | return ret; | |
f0c0b2b8 KH |
4167 | } |
4168 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
4169 | ||
4170 | /* | |
4171 | * sysctl handler for numa_zonelist_order | |
4172 | */ | |
cccad5b9 | 4173 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 4174 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
4175 | loff_t *ppos) |
4176 | { | |
4177 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
4178 | int ret; | |
443c6f14 | 4179 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 4180 | |
443c6f14 | 4181 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
4182 | if (write) { |
4183 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
4184 | ret = -EINVAL; | |
4185 | goto out; | |
4186 | } | |
4187 | strcpy(saved_string, (char *)table->data); | |
4188 | } | |
8d65af78 | 4189 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 4190 | if (ret) |
443c6f14 | 4191 | goto out; |
f0c0b2b8 KH |
4192 | if (write) { |
4193 | int oldval = user_zonelist_order; | |
dacbde09 CG |
4194 | |
4195 | ret = __parse_numa_zonelist_order((char *)table->data); | |
4196 | if (ret) { | |
f0c0b2b8 KH |
4197 | /* |
4198 | * bogus value. restore saved string | |
4199 | */ | |
dacbde09 | 4200 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
4201 | NUMA_ZONELIST_ORDER_LEN); |
4202 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
4203 | } else if (oldval != user_zonelist_order) { |
4204 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 4205 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
4206 | mutex_unlock(&zonelists_mutex); |
4207 | } | |
f0c0b2b8 | 4208 | } |
443c6f14 AK |
4209 | out: |
4210 | mutex_unlock(&zl_order_mutex); | |
4211 | return ret; | |
f0c0b2b8 KH |
4212 | } |
4213 | ||
4214 | ||
62bc62a8 | 4215 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
4216 | static int node_load[MAX_NUMNODES]; |
4217 | ||
1da177e4 | 4218 | /** |
4dc3b16b | 4219 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
4220 | * @node: node whose fallback list we're appending |
4221 | * @used_node_mask: nodemask_t of already used nodes | |
4222 | * | |
4223 | * We use a number of factors to determine which is the next node that should | |
4224 | * appear on a given node's fallback list. The node should not have appeared | |
4225 | * already in @node's fallback list, and it should be the next closest node | |
4226 | * according to the distance array (which contains arbitrary distance values | |
4227 | * from each node to each node in the system), and should also prefer nodes | |
4228 | * with no CPUs, since presumably they'll have very little allocation pressure | |
4229 | * on them otherwise. | |
4230 | * It returns -1 if no node is found. | |
4231 | */ | |
f0c0b2b8 | 4232 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 4233 | { |
4cf808eb | 4234 | int n, val; |
1da177e4 | 4235 | int min_val = INT_MAX; |
00ef2d2f | 4236 | int best_node = NUMA_NO_NODE; |
a70f7302 | 4237 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 4238 | |
4cf808eb LT |
4239 | /* Use the local node if we haven't already */ |
4240 | if (!node_isset(node, *used_node_mask)) { | |
4241 | node_set(node, *used_node_mask); | |
4242 | return node; | |
4243 | } | |
1da177e4 | 4244 | |
4b0ef1fe | 4245 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
4246 | |
4247 | /* Don't want a node to appear more than once */ | |
4248 | if (node_isset(n, *used_node_mask)) | |
4249 | continue; | |
4250 | ||
1da177e4 LT |
4251 | /* Use the distance array to find the distance */ |
4252 | val = node_distance(node, n); | |
4253 | ||
4cf808eb LT |
4254 | /* Penalize nodes under us ("prefer the next node") */ |
4255 | val += (n < node); | |
4256 | ||
1da177e4 | 4257 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
4258 | tmp = cpumask_of_node(n); |
4259 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
4260 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4261 | ||
4262 | /* Slight preference for less loaded node */ | |
4263 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
4264 | val += node_load[n]; | |
4265 | ||
4266 | if (val < min_val) { | |
4267 | min_val = val; | |
4268 | best_node = n; | |
4269 | } | |
4270 | } | |
4271 | ||
4272 | if (best_node >= 0) | |
4273 | node_set(best_node, *used_node_mask); | |
4274 | ||
4275 | return best_node; | |
4276 | } | |
4277 | ||
f0c0b2b8 KH |
4278 | |
4279 | /* | |
4280 | * Build zonelists ordered by node and zones within node. | |
4281 | * This results in maximum locality--normal zone overflows into local | |
4282 | * DMA zone, if any--but risks exhausting DMA zone. | |
4283 | */ | |
4284 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 4285 | { |
f0c0b2b8 | 4286 | int j; |
1da177e4 | 4287 | struct zonelist *zonelist; |
f0c0b2b8 | 4288 | |
54a6eb5c | 4289 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 4290 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 4291 | ; |
bc732f1d | 4292 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
4293 | zonelist->_zonerefs[j].zone = NULL; |
4294 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
4295 | } |
4296 | ||
523b9458 CL |
4297 | /* |
4298 | * Build gfp_thisnode zonelists | |
4299 | */ | |
4300 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
4301 | { | |
523b9458 CL |
4302 | int j; |
4303 | struct zonelist *zonelist; | |
4304 | ||
54a6eb5c | 4305 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 4306 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
4307 | zonelist->_zonerefs[j].zone = NULL; |
4308 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
4309 | } |
4310 | ||
f0c0b2b8 KH |
4311 | /* |
4312 | * Build zonelists ordered by zone and nodes within zones. | |
4313 | * This results in conserving DMA zone[s] until all Normal memory is | |
4314 | * exhausted, but results in overflowing to remote node while memory | |
4315 | * may still exist in local DMA zone. | |
4316 | */ | |
4317 | static int node_order[MAX_NUMNODES]; | |
4318 | ||
4319 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
4320 | { | |
f0c0b2b8 KH |
4321 | int pos, j, node; |
4322 | int zone_type; /* needs to be signed */ | |
4323 | struct zone *z; | |
4324 | struct zonelist *zonelist; | |
4325 | ||
54a6eb5c MG |
4326 | zonelist = &pgdat->node_zonelists[0]; |
4327 | pos = 0; | |
4328 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
4329 | for (j = 0; j < nr_nodes; j++) { | |
4330 | node = node_order[j]; | |
4331 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
4332 | if (populated_zone(z)) { | |
dd1a239f MG |
4333 | zoneref_set_zone(z, |
4334 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 4335 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
4336 | } |
4337 | } | |
f0c0b2b8 | 4338 | } |
dd1a239f MG |
4339 | zonelist->_zonerefs[pos].zone = NULL; |
4340 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
4341 | } |
4342 | ||
3193913c MG |
4343 | #if defined(CONFIG_64BIT) |
4344 | /* | |
4345 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
4346 | * penalty to every machine in case the specialised case applies. Default | |
4347 | * to Node-ordering on 64-bit NUMA machines | |
4348 | */ | |
4349 | static int default_zonelist_order(void) | |
4350 | { | |
4351 | return ZONELIST_ORDER_NODE; | |
4352 | } | |
4353 | #else | |
4354 | /* | |
4355 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
4356 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
4357 | * then reclaim will occur more frequency increasing stalls and potentially | |
4358 | * be easier to OOM if a large percentage of the zone is under writeback or | |
4359 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
4360 | * Hence, default to zone ordering on 32-bit. | |
4361 | */ | |
f0c0b2b8 KH |
4362 | static int default_zonelist_order(void) |
4363 | { | |
f0c0b2b8 KH |
4364 | return ZONELIST_ORDER_ZONE; |
4365 | } | |
3193913c | 4366 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
4367 | |
4368 | static void set_zonelist_order(void) | |
4369 | { | |
4370 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
4371 | current_zonelist_order = default_zonelist_order(); | |
4372 | else | |
4373 | current_zonelist_order = user_zonelist_order; | |
4374 | } | |
4375 | ||
4376 | static void build_zonelists(pg_data_t *pgdat) | |
4377 | { | |
c00eb15a | 4378 | int i, node, load; |
1da177e4 | 4379 | nodemask_t used_mask; |
f0c0b2b8 KH |
4380 | int local_node, prev_node; |
4381 | struct zonelist *zonelist; | |
d00181b9 | 4382 | unsigned int order = current_zonelist_order; |
1da177e4 LT |
4383 | |
4384 | /* initialize zonelists */ | |
523b9458 | 4385 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 4386 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
4387 | zonelist->_zonerefs[0].zone = NULL; |
4388 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
4389 | } |
4390 | ||
4391 | /* NUMA-aware ordering of nodes */ | |
4392 | local_node = pgdat->node_id; | |
62bc62a8 | 4393 | load = nr_online_nodes; |
1da177e4 LT |
4394 | prev_node = local_node; |
4395 | nodes_clear(used_mask); | |
f0c0b2b8 | 4396 | |
f0c0b2b8 | 4397 | memset(node_order, 0, sizeof(node_order)); |
c00eb15a | 4398 | i = 0; |
f0c0b2b8 | 4399 | |
1da177e4 LT |
4400 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
4401 | /* | |
4402 | * We don't want to pressure a particular node. | |
4403 | * So adding penalty to the first node in same | |
4404 | * distance group to make it round-robin. | |
4405 | */ | |
957f822a DR |
4406 | if (node_distance(local_node, node) != |
4407 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
4408 | node_load[node] = load; |
4409 | ||
1da177e4 LT |
4410 | prev_node = node; |
4411 | load--; | |
f0c0b2b8 KH |
4412 | if (order == ZONELIST_ORDER_NODE) |
4413 | build_zonelists_in_node_order(pgdat, node); | |
4414 | else | |
c00eb15a | 4415 | node_order[i++] = node; /* remember order */ |
f0c0b2b8 | 4416 | } |
1da177e4 | 4417 | |
f0c0b2b8 KH |
4418 | if (order == ZONELIST_ORDER_ZONE) { |
4419 | /* calculate node order -- i.e., DMA last! */ | |
c00eb15a | 4420 | build_zonelists_in_zone_order(pgdat, i); |
1da177e4 | 4421 | } |
523b9458 CL |
4422 | |
4423 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
4424 | } |
4425 | ||
7aac7898 LS |
4426 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4427 | /* | |
4428 | * Return node id of node used for "local" allocations. | |
4429 | * I.e., first node id of first zone in arg node's generic zonelist. | |
4430 | * Used for initializing percpu 'numa_mem', which is used primarily | |
4431 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
4432 | */ | |
4433 | int local_memory_node(int node) | |
4434 | { | |
4435 | struct zone *zone; | |
4436 | ||
4437 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
4438 | gfp_zone(GFP_KERNEL), | |
4439 | NULL, | |
4440 | &zone); | |
4441 | return zone->node; | |
4442 | } | |
4443 | #endif | |
f0c0b2b8 | 4444 | |
1da177e4 LT |
4445 | #else /* CONFIG_NUMA */ |
4446 | ||
f0c0b2b8 KH |
4447 | static void set_zonelist_order(void) |
4448 | { | |
4449 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
4450 | } | |
4451 | ||
4452 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 4453 | { |
19655d34 | 4454 | int node, local_node; |
54a6eb5c MG |
4455 | enum zone_type j; |
4456 | struct zonelist *zonelist; | |
1da177e4 LT |
4457 | |
4458 | local_node = pgdat->node_id; | |
1da177e4 | 4459 | |
54a6eb5c | 4460 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 4461 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 4462 | |
54a6eb5c MG |
4463 | /* |
4464 | * Now we build the zonelist so that it contains the zones | |
4465 | * of all the other nodes. | |
4466 | * We don't want to pressure a particular node, so when | |
4467 | * building the zones for node N, we make sure that the | |
4468 | * zones coming right after the local ones are those from | |
4469 | * node N+1 (modulo N) | |
4470 | */ | |
4471 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
4472 | if (!node_online(node)) | |
4473 | continue; | |
bc732f1d | 4474 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 4475 | } |
54a6eb5c MG |
4476 | for (node = 0; node < local_node; node++) { |
4477 | if (!node_online(node)) | |
4478 | continue; | |
bc732f1d | 4479 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
4480 | } |
4481 | ||
dd1a239f MG |
4482 | zonelist->_zonerefs[j].zone = NULL; |
4483 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
4484 | } |
4485 | ||
4486 | #endif /* CONFIG_NUMA */ | |
4487 | ||
99dcc3e5 CL |
4488 | /* |
4489 | * Boot pageset table. One per cpu which is going to be used for all | |
4490 | * zones and all nodes. The parameters will be set in such a way | |
4491 | * that an item put on a list will immediately be handed over to | |
4492 | * the buddy list. This is safe since pageset manipulation is done | |
4493 | * with interrupts disabled. | |
4494 | * | |
4495 | * The boot_pagesets must be kept even after bootup is complete for | |
4496 | * unused processors and/or zones. They do play a role for bootstrapping | |
4497 | * hotplugged processors. | |
4498 | * | |
4499 | * zoneinfo_show() and maybe other functions do | |
4500 | * not check if the processor is online before following the pageset pointer. | |
4501 | * Other parts of the kernel may not check if the zone is available. | |
4502 | */ | |
4503 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
4504 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 4505 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 4506 | |
4eaf3f64 HL |
4507 | /* |
4508 | * Global mutex to protect against size modification of zonelists | |
4509 | * as well as to serialize pageset setup for the new populated zone. | |
4510 | */ | |
4511 | DEFINE_MUTEX(zonelists_mutex); | |
4512 | ||
9b1a4d38 | 4513 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 4514 | static int __build_all_zonelists(void *data) |
1da177e4 | 4515 | { |
6811378e | 4516 | int nid; |
99dcc3e5 | 4517 | int cpu; |
9adb62a5 | 4518 | pg_data_t *self = data; |
9276b1bc | 4519 | |
7f9cfb31 BL |
4520 | #ifdef CONFIG_NUMA |
4521 | memset(node_load, 0, sizeof(node_load)); | |
4522 | #endif | |
9adb62a5 JL |
4523 | |
4524 | if (self && !node_online(self->node_id)) { | |
4525 | build_zonelists(self); | |
9adb62a5 JL |
4526 | } |
4527 | ||
9276b1bc | 4528 | for_each_online_node(nid) { |
7ea1530a CL |
4529 | pg_data_t *pgdat = NODE_DATA(nid); |
4530 | ||
4531 | build_zonelists(pgdat); | |
9276b1bc | 4532 | } |
99dcc3e5 CL |
4533 | |
4534 | /* | |
4535 | * Initialize the boot_pagesets that are going to be used | |
4536 | * for bootstrapping processors. The real pagesets for | |
4537 | * each zone will be allocated later when the per cpu | |
4538 | * allocator is available. | |
4539 | * | |
4540 | * boot_pagesets are used also for bootstrapping offline | |
4541 | * cpus if the system is already booted because the pagesets | |
4542 | * are needed to initialize allocators on a specific cpu too. | |
4543 | * F.e. the percpu allocator needs the page allocator which | |
4544 | * needs the percpu allocator in order to allocate its pagesets | |
4545 | * (a chicken-egg dilemma). | |
4546 | */ | |
7aac7898 | 4547 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
4548 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
4549 | ||
7aac7898 LS |
4550 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4551 | /* | |
4552 | * We now know the "local memory node" for each node-- | |
4553 | * i.e., the node of the first zone in the generic zonelist. | |
4554 | * Set up numa_mem percpu variable for on-line cpus. During | |
4555 | * boot, only the boot cpu should be on-line; we'll init the | |
4556 | * secondary cpus' numa_mem as they come on-line. During | |
4557 | * node/memory hotplug, we'll fixup all on-line cpus. | |
4558 | */ | |
4559 | if (cpu_online(cpu)) | |
4560 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
4561 | #endif | |
4562 | } | |
4563 | ||
6811378e YG |
4564 | return 0; |
4565 | } | |
4566 | ||
061f67bc RV |
4567 | static noinline void __init |
4568 | build_all_zonelists_init(void) | |
4569 | { | |
4570 | __build_all_zonelists(NULL); | |
4571 | mminit_verify_zonelist(); | |
4572 | cpuset_init_current_mems_allowed(); | |
4573 | } | |
4574 | ||
4eaf3f64 HL |
4575 | /* |
4576 | * Called with zonelists_mutex held always | |
4577 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
4578 | * |
4579 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
4580 | * [we're only called with non-NULL zone through __meminit paths] and | |
4581 | * (2) call of __init annotated helper build_all_zonelists_init | |
4582 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 4583 | */ |
9adb62a5 | 4584 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 4585 | { |
f0c0b2b8 KH |
4586 | set_zonelist_order(); |
4587 | ||
6811378e | 4588 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 4589 | build_all_zonelists_init(); |
6811378e | 4590 | } else { |
e9959f0f | 4591 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
4592 | if (zone) |
4593 | setup_zone_pageset(zone); | |
e9959f0f | 4594 | #endif |
dd1895e2 CS |
4595 | /* we have to stop all cpus to guarantee there is no user |
4596 | of zonelist */ | |
9adb62a5 | 4597 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
4598 | /* cpuset refresh routine should be here */ |
4599 | } | |
bd1e22b8 | 4600 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
4601 | /* |
4602 | * Disable grouping by mobility if the number of pages in the | |
4603 | * system is too low to allow the mechanism to work. It would be | |
4604 | * more accurate, but expensive to check per-zone. This check is | |
4605 | * made on memory-hotadd so a system can start with mobility | |
4606 | * disabled and enable it later | |
4607 | */ | |
d9c23400 | 4608 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
4609 | page_group_by_mobility_disabled = 1; |
4610 | else | |
4611 | page_group_by_mobility_disabled = 0; | |
4612 | ||
756a025f JP |
4613 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
4614 | nr_online_nodes, | |
4615 | zonelist_order_name[current_zonelist_order], | |
4616 | page_group_by_mobility_disabled ? "off" : "on", | |
4617 | vm_total_pages); | |
f0c0b2b8 | 4618 | #ifdef CONFIG_NUMA |
f88dfff5 | 4619 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 4620 | #endif |
1da177e4 LT |
4621 | } |
4622 | ||
4623 | /* | |
4624 | * Helper functions to size the waitqueue hash table. | |
4625 | * Essentially these want to choose hash table sizes sufficiently | |
4626 | * large so that collisions trying to wait on pages are rare. | |
4627 | * But in fact, the number of active page waitqueues on typical | |
4628 | * systems is ridiculously low, less than 200. So this is even | |
4629 | * conservative, even though it seems large. | |
4630 | * | |
4631 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
4632 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
4633 | */ | |
4634 | #define PAGES_PER_WAITQUEUE 256 | |
4635 | ||
cca448fe | 4636 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 4637 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
4638 | { |
4639 | unsigned long size = 1; | |
4640 | ||
4641 | pages /= PAGES_PER_WAITQUEUE; | |
4642 | ||
4643 | while (size < pages) | |
4644 | size <<= 1; | |
4645 | ||
4646 | /* | |
4647 | * Once we have dozens or even hundreds of threads sleeping | |
4648 | * on IO we've got bigger problems than wait queue collision. | |
4649 | * Limit the size of the wait table to a reasonable size. | |
4650 | */ | |
4651 | size = min(size, 4096UL); | |
4652 | ||
4653 | return max(size, 4UL); | |
4654 | } | |
cca448fe YG |
4655 | #else |
4656 | /* | |
4657 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
4658 | * a suitable size for its wait_table. So we use the maximum size now. | |
4659 | * | |
4660 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
4661 | * | |
4662 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
4663 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
4664 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
4665 | * | |
4666 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
4667 | * or more by the traditional way. (See above). It equals: | |
4668 | * | |
4669 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
4670 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
4671 | * powerpc (64K page size) : = (32G +16M)byte. | |
4672 | */ | |
4673 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
4674 | { | |
4675 | return 4096UL; | |
4676 | } | |
4677 | #endif | |
1da177e4 LT |
4678 | |
4679 | /* | |
4680 | * This is an integer logarithm so that shifts can be used later | |
4681 | * to extract the more random high bits from the multiplicative | |
4682 | * hash function before the remainder is taken. | |
4683 | */ | |
4684 | static inline unsigned long wait_table_bits(unsigned long size) | |
4685 | { | |
4686 | return ffz(~size); | |
4687 | } | |
4688 | ||
1da177e4 LT |
4689 | /* |
4690 | * Initially all pages are reserved - free ones are freed | |
4691 | * up by free_all_bootmem() once the early boot process is | |
4692 | * done. Non-atomic initialization, single-pass. | |
4693 | */ | |
c09b4240 | 4694 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 4695 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 4696 | { |
4b94ffdc | 4697 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
29751f69 | 4698 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 4699 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 4700 | unsigned long pfn; |
3a80a7fa | 4701 | unsigned long nr_initialised = 0; |
342332e6 TI |
4702 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
4703 | struct memblock_region *r = NULL, *tmp; | |
4704 | #endif | |
1da177e4 | 4705 | |
22b31eec HD |
4706 | if (highest_memmap_pfn < end_pfn - 1) |
4707 | highest_memmap_pfn = end_pfn - 1; | |
4708 | ||
4b94ffdc DW |
4709 | /* |
4710 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
4711 | * memory | |
4712 | */ | |
4713 | if (altmap && start_pfn == altmap->base_pfn) | |
4714 | start_pfn += altmap->reserve; | |
4715 | ||
cbe8dd4a | 4716 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 4717 | /* |
b72d0ffb AM |
4718 | * There can be holes in boot-time mem_map[]s handed to this |
4719 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 4720 | */ |
b72d0ffb AM |
4721 | if (context != MEMMAP_EARLY) |
4722 | goto not_early; | |
4723 | ||
4724 | if (!early_pfn_valid(pfn)) | |
4725 | continue; | |
4726 | if (!early_pfn_in_nid(pfn, nid)) | |
4727 | continue; | |
4728 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
4729 | break; | |
342332e6 TI |
4730 | |
4731 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
4732 | /* |
4733 | * If not mirrored_kernelcore and ZONE_MOVABLE exists, range | |
4734 | * from zone_movable_pfn[nid] to end of each node should be | |
4735 | * ZONE_MOVABLE not ZONE_NORMAL. skip it. | |
4736 | */ | |
4737 | if (!mirrored_kernelcore && zone_movable_pfn[nid]) | |
4738 | if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid]) | |
4739 | continue; | |
342332e6 | 4740 | |
b72d0ffb AM |
4741 | /* |
4742 | * Check given memblock attribute by firmware which can affect | |
4743 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
4744 | * mirrored, it's an overlapped memmap init. skip it. | |
4745 | */ | |
4746 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
4747 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
4748 | for_each_memblock(memory, tmp) | |
4749 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
4750 | break; | |
4751 | r = tmp; | |
4752 | } | |
4753 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
4754 | memblock_is_mirror(r)) { | |
4755 | /* already initialized as NORMAL */ | |
4756 | pfn = memblock_region_memory_end_pfn(r); | |
4757 | continue; | |
342332e6 | 4758 | } |
a2f3aa02 | 4759 | } |
b72d0ffb | 4760 | #endif |
ac5d2539 | 4761 | |
b72d0ffb | 4762 | not_early: |
ac5d2539 MG |
4763 | /* |
4764 | * Mark the block movable so that blocks are reserved for | |
4765 | * movable at startup. This will force kernel allocations | |
4766 | * to reserve their blocks rather than leaking throughout | |
4767 | * the address space during boot when many long-lived | |
974a786e | 4768 | * kernel allocations are made. |
ac5d2539 MG |
4769 | * |
4770 | * bitmap is created for zone's valid pfn range. but memmap | |
4771 | * can be created for invalid pages (for alignment) | |
4772 | * check here not to call set_pageblock_migratetype() against | |
4773 | * pfn out of zone. | |
4774 | */ | |
4775 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
4776 | struct page *page = pfn_to_page(pfn); | |
4777 | ||
4778 | __init_single_page(page, pfn, zone, nid); | |
4779 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4780 | } else { | |
4781 | __init_single_pfn(pfn, zone, nid); | |
4782 | } | |
1da177e4 LT |
4783 | } |
4784 | } | |
4785 | ||
1e548deb | 4786 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 4787 | { |
7aeb09f9 | 4788 | unsigned int order, t; |
b2a0ac88 MG |
4789 | for_each_migratetype_order(order, t) { |
4790 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
4791 | zone->free_area[order].nr_free = 0; |
4792 | } | |
4793 | } | |
4794 | ||
4795 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
4796 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 4797 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
4798 | #endif |
4799 | ||
7cd2b0a3 | 4800 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 4801 | { |
3a6be87f | 4802 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
4803 | int batch; |
4804 | ||
4805 | /* | |
4806 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 4807 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
4808 | * |
4809 | * OK, so we don't know how big the cache is. So guess. | |
4810 | */ | |
b40da049 | 4811 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
4812 | if (batch * PAGE_SIZE > 512 * 1024) |
4813 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
4814 | batch /= 4; /* We effectively *= 4 below */ |
4815 | if (batch < 1) | |
4816 | batch = 1; | |
4817 | ||
4818 | /* | |
0ceaacc9 NP |
4819 | * Clamp the batch to a 2^n - 1 value. Having a power |
4820 | * of 2 value was found to be more likely to have | |
4821 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 4822 | * |
0ceaacc9 NP |
4823 | * For example if 2 tasks are alternately allocating |
4824 | * batches of pages, one task can end up with a lot | |
4825 | * of pages of one half of the possible page colors | |
4826 | * and the other with pages of the other colors. | |
e7c8d5c9 | 4827 | */ |
9155203a | 4828 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 4829 | |
e7c8d5c9 | 4830 | return batch; |
3a6be87f DH |
4831 | |
4832 | #else | |
4833 | /* The deferral and batching of frees should be suppressed under NOMMU | |
4834 | * conditions. | |
4835 | * | |
4836 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4837 | * of contiguous memory as there's no hardware page translation to | |
4838 | * assemble apparent contiguous memory from discontiguous pages. | |
4839 | * | |
4840 | * Queueing large contiguous runs of pages for batching, however, | |
4841 | * causes the pages to actually be freed in smaller chunks. As there | |
4842 | * can be a significant delay between the individual batches being | |
4843 | * recycled, this leads to the once large chunks of space being | |
4844 | * fragmented and becoming unavailable for high-order allocations. | |
4845 | */ | |
4846 | return 0; | |
4847 | #endif | |
e7c8d5c9 CL |
4848 | } |
4849 | ||
8d7a8fa9 CS |
4850 | /* |
4851 | * pcp->high and pcp->batch values are related and dependent on one another: | |
4852 | * ->batch must never be higher then ->high. | |
4853 | * The following function updates them in a safe manner without read side | |
4854 | * locking. | |
4855 | * | |
4856 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
4857 | * those fields changing asynchronously (acording the the above rule). | |
4858 | * | |
4859 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
4860 | * outside of boot time (or some other assurance that no concurrent updaters | |
4861 | * exist). | |
4862 | */ | |
4863 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
4864 | unsigned long batch) | |
4865 | { | |
4866 | /* start with a fail safe value for batch */ | |
4867 | pcp->batch = 1; | |
4868 | smp_wmb(); | |
4869 | ||
4870 | /* Update high, then batch, in order */ | |
4871 | pcp->high = high; | |
4872 | smp_wmb(); | |
4873 | ||
4874 | pcp->batch = batch; | |
4875 | } | |
4876 | ||
3664033c | 4877 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
4878 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
4879 | { | |
8d7a8fa9 | 4880 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
4881 | } |
4882 | ||
88c90dbc | 4883 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
4884 | { |
4885 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4886 | int migratetype; |
2caaad41 | 4887 | |
1c6fe946 MD |
4888 | memset(p, 0, sizeof(*p)); |
4889 | ||
3dfa5721 | 4890 | pcp = &p->pcp; |
2caaad41 | 4891 | pcp->count = 0; |
5f8dcc21 MG |
4892 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4893 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4894 | } |
4895 | ||
88c90dbc CS |
4896 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
4897 | { | |
4898 | pageset_init(p); | |
4899 | pageset_set_batch(p, batch); | |
4900 | } | |
4901 | ||
8ad4b1fb | 4902 | /* |
3664033c | 4903 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
4904 | * to the value high for the pageset p. |
4905 | */ | |
3664033c | 4906 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
4907 | unsigned long high) |
4908 | { | |
8d7a8fa9 CS |
4909 | unsigned long batch = max(1UL, high / 4); |
4910 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
4911 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 4912 | |
8d7a8fa9 | 4913 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
4914 | } |
4915 | ||
7cd2b0a3 DR |
4916 | static void pageset_set_high_and_batch(struct zone *zone, |
4917 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 4918 | { |
56cef2b8 | 4919 | if (percpu_pagelist_fraction) |
3664033c | 4920 | pageset_set_high(pcp, |
56cef2b8 CS |
4921 | (zone->managed_pages / |
4922 | percpu_pagelist_fraction)); | |
4923 | else | |
4924 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
4925 | } | |
4926 | ||
169f6c19 CS |
4927 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
4928 | { | |
4929 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4930 | ||
4931 | pageset_init(pcp); | |
4932 | pageset_set_high_and_batch(zone, pcp); | |
4933 | } | |
4934 | ||
4ed7e022 | 4935 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4936 | { |
4937 | int cpu; | |
319774e2 | 4938 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
4939 | for_each_possible_cpu(cpu) |
4940 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
4941 | } |
4942 | ||
2caaad41 | 4943 | /* |
99dcc3e5 CL |
4944 | * Allocate per cpu pagesets and initialize them. |
4945 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4946 | */ |
99dcc3e5 | 4947 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4948 | { |
99dcc3e5 | 4949 | struct zone *zone; |
e7c8d5c9 | 4950 | |
319774e2 WF |
4951 | for_each_populated_zone(zone) |
4952 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4953 | } |
4954 | ||
577a32f6 | 4955 | static noinline __init_refok |
cca448fe | 4956 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4957 | { |
4958 | int i; | |
cca448fe | 4959 | size_t alloc_size; |
ed8ece2e DH |
4960 | |
4961 | /* | |
4962 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4963 | * per zone. | |
4964 | */ | |
02b694de YG |
4965 | zone->wait_table_hash_nr_entries = |
4966 | wait_table_hash_nr_entries(zone_size_pages); | |
4967 | zone->wait_table_bits = | |
4968 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4969 | alloc_size = zone->wait_table_hash_nr_entries |
4970 | * sizeof(wait_queue_head_t); | |
4971 | ||
cd94b9db | 4972 | if (!slab_is_available()) { |
cca448fe | 4973 | zone->wait_table = (wait_queue_head_t *) |
6782832e SS |
4974 | memblock_virt_alloc_node_nopanic( |
4975 | alloc_size, zone->zone_pgdat->node_id); | |
cca448fe YG |
4976 | } else { |
4977 | /* | |
4978 | * This case means that a zone whose size was 0 gets new memory | |
4979 | * via memory hot-add. | |
4980 | * But it may be the case that a new node was hot-added. In | |
4981 | * this case vmalloc() will not be able to use this new node's | |
4982 | * memory - this wait_table must be initialized to use this new | |
4983 | * node itself as well. | |
4984 | * To use this new node's memory, further consideration will be | |
4985 | * necessary. | |
4986 | */ | |
8691f3a7 | 4987 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4988 | } |
4989 | if (!zone->wait_table) | |
4990 | return -ENOMEM; | |
ed8ece2e | 4991 | |
b8af2941 | 4992 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4993 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4994 | |
4995 | return 0; | |
ed8ece2e DH |
4996 | } |
4997 | ||
c09b4240 | 4998 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4999 | { |
99dcc3e5 CL |
5000 | /* |
5001 | * per cpu subsystem is not up at this point. The following code | |
5002 | * relies on the ability of the linker to provide the | |
5003 | * offset of a (static) per cpu variable into the per cpu area. | |
5004 | */ | |
5005 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5006 | |
b38a8725 | 5007 | if (populated_zone(zone)) |
99dcc3e5 CL |
5008 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5009 | zone->name, zone->present_pages, | |
5010 | zone_batchsize(zone)); | |
ed8ece2e DH |
5011 | } |
5012 | ||
4ed7e022 | 5013 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5014 | unsigned long zone_start_pfn, |
b171e409 | 5015 | unsigned long size) |
ed8ece2e DH |
5016 | { |
5017 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
5018 | int ret; |
5019 | ret = zone_wait_table_init(zone, size); | |
5020 | if (ret) | |
5021 | return ret; | |
ed8ece2e DH |
5022 | pgdat->nr_zones = zone_idx(zone) + 1; |
5023 | ||
ed8ece2e DH |
5024 | zone->zone_start_pfn = zone_start_pfn; |
5025 | ||
708614e6 MG |
5026 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5027 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5028 | pgdat->node_id, | |
5029 | (unsigned long)zone_idx(zone), | |
5030 | zone_start_pfn, (zone_start_pfn + size)); | |
5031 | ||
1e548deb | 5032 | zone_init_free_lists(zone); |
718127cc YG |
5033 | |
5034 | return 0; | |
ed8ece2e DH |
5035 | } |
5036 | ||
0ee332c1 | 5037 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5038 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5039 | |
c713216d MG |
5040 | /* |
5041 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5042 | */ |
8a942fde MG |
5043 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5044 | struct mminit_pfnnid_cache *state) | |
c713216d | 5045 | { |
c13291a5 | 5046 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5047 | int nid; |
7c243c71 | 5048 | |
8a942fde MG |
5049 | if (state->last_start <= pfn && pfn < state->last_end) |
5050 | return state->last_nid; | |
c713216d | 5051 | |
e76b63f8 YL |
5052 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5053 | if (nid != -1) { | |
8a942fde MG |
5054 | state->last_start = start_pfn; |
5055 | state->last_end = end_pfn; | |
5056 | state->last_nid = nid; | |
e76b63f8 YL |
5057 | } |
5058 | ||
5059 | return nid; | |
c713216d MG |
5060 | } |
5061 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5062 | ||
c713216d | 5063 | /** |
6782832e | 5064 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5065 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5066 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5067 | * |
7d018176 ZZ |
5068 | * If an architecture guarantees that all ranges registered contain no holes |
5069 | * and may be freed, this this function may be used instead of calling | |
5070 | * memblock_free_early_nid() manually. | |
c713216d | 5071 | */ |
c13291a5 | 5072 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5073 | { |
c13291a5 TH |
5074 | unsigned long start_pfn, end_pfn; |
5075 | int i, this_nid; | |
edbe7d23 | 5076 | |
c13291a5 TH |
5077 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5078 | start_pfn = min(start_pfn, max_low_pfn); | |
5079 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5080 | |
c13291a5 | 5081 | if (start_pfn < end_pfn) |
6782832e SS |
5082 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5083 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5084 | this_nid); | |
edbe7d23 | 5085 | } |
edbe7d23 | 5086 | } |
edbe7d23 | 5087 | |
c713216d MG |
5088 | /** |
5089 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5090 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5091 | * |
7d018176 ZZ |
5092 | * If an architecture guarantees that all ranges registered contain no holes and may |
5093 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5094 | */ |
5095 | void __init sparse_memory_present_with_active_regions(int nid) | |
5096 | { | |
c13291a5 TH |
5097 | unsigned long start_pfn, end_pfn; |
5098 | int i, this_nid; | |
c713216d | 5099 | |
c13291a5 TH |
5100 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5101 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5102 | } |
5103 | ||
5104 | /** | |
5105 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5106 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5107 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5108 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5109 | * |
5110 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5111 | * provided by memblock_set_node(). If called for a node |
c713216d | 5112 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5113 | * PFNs will be 0. |
c713216d | 5114 | */ |
a3142c8e | 5115 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5116 | unsigned long *start_pfn, unsigned long *end_pfn) |
5117 | { | |
c13291a5 | 5118 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5119 | int i; |
c13291a5 | 5120 | |
c713216d MG |
5121 | *start_pfn = -1UL; |
5122 | *end_pfn = 0; | |
5123 | ||
c13291a5 TH |
5124 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5125 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5126 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5127 | } |
5128 | ||
633c0666 | 5129 | if (*start_pfn == -1UL) |
c713216d | 5130 | *start_pfn = 0; |
c713216d MG |
5131 | } |
5132 | ||
2a1e274a MG |
5133 | /* |
5134 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5135 | * assumption is made that zones within a node are ordered in monotonic | |
5136 | * increasing memory addresses so that the "highest" populated zone is used | |
5137 | */ | |
b69a7288 | 5138 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5139 | { |
5140 | int zone_index; | |
5141 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5142 | if (zone_index == ZONE_MOVABLE) | |
5143 | continue; | |
5144 | ||
5145 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5146 | arch_zone_lowest_possible_pfn[zone_index]) | |
5147 | break; | |
5148 | } | |
5149 | ||
5150 | VM_BUG_ON(zone_index == -1); | |
5151 | movable_zone = zone_index; | |
5152 | } | |
5153 | ||
5154 | /* | |
5155 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5156 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5157 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5158 | * in each node depending on the size of each node and how evenly kernelcore | |
5159 | * is distributed. This helper function adjusts the zone ranges | |
5160 | * provided by the architecture for a given node by using the end of the | |
5161 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5162 | * zones within a node are in order of monotonic increases memory addresses | |
5163 | */ | |
b69a7288 | 5164 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5165 | unsigned long zone_type, |
5166 | unsigned long node_start_pfn, | |
5167 | unsigned long node_end_pfn, | |
5168 | unsigned long *zone_start_pfn, | |
5169 | unsigned long *zone_end_pfn) | |
5170 | { | |
5171 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5172 | if (zone_movable_pfn[nid]) { | |
5173 | /* Size ZONE_MOVABLE */ | |
5174 | if (zone_type == ZONE_MOVABLE) { | |
5175 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5176 | *zone_end_pfn = min(node_end_pfn, | |
5177 | arch_zone_highest_possible_pfn[movable_zone]); | |
5178 | ||
2a1e274a MG |
5179 | /* Check if this whole range is within ZONE_MOVABLE */ |
5180 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5181 | *zone_start_pfn = *zone_end_pfn; | |
5182 | } | |
5183 | } | |
5184 | ||
c713216d MG |
5185 | /* |
5186 | * Return the number of pages a zone spans in a node, including holes | |
5187 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5188 | */ | |
6ea6e688 | 5189 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5190 | unsigned long zone_type, |
7960aedd ZY |
5191 | unsigned long node_start_pfn, |
5192 | unsigned long node_end_pfn, | |
d91749c1 TI |
5193 | unsigned long *zone_start_pfn, |
5194 | unsigned long *zone_end_pfn, | |
c713216d MG |
5195 | unsigned long *ignored) |
5196 | { | |
b5685e92 | 5197 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5198 | if (!node_start_pfn && !node_end_pfn) |
5199 | return 0; | |
5200 | ||
7960aedd | 5201 | /* Get the start and end of the zone */ |
d91749c1 TI |
5202 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5203 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5204 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5205 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5206 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5207 | |
5208 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5209 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5210 | return 0; |
5211 | ||
5212 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5213 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5214 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5215 | |
5216 | /* Return the spanned pages */ | |
d91749c1 | 5217 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5218 | } |
5219 | ||
5220 | /* | |
5221 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5222 | * then all holes in the requested range will be accounted for. |
c713216d | 5223 | */ |
32996250 | 5224 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5225 | unsigned long range_start_pfn, |
5226 | unsigned long range_end_pfn) | |
5227 | { | |
96e907d1 TH |
5228 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5229 | unsigned long start_pfn, end_pfn; | |
5230 | int i; | |
c713216d | 5231 | |
96e907d1 TH |
5232 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5233 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5234 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5235 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5236 | } |
96e907d1 | 5237 | return nr_absent; |
c713216d MG |
5238 | } |
5239 | ||
5240 | /** | |
5241 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5242 | * @start_pfn: The start PFN to start searching for holes | |
5243 | * @end_pfn: The end PFN to stop searching for holes | |
5244 | * | |
88ca3b94 | 5245 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5246 | */ |
5247 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5248 | unsigned long end_pfn) | |
5249 | { | |
5250 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5251 | } | |
5252 | ||
5253 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5254 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5255 | unsigned long zone_type, |
7960aedd ZY |
5256 | unsigned long node_start_pfn, |
5257 | unsigned long node_end_pfn, | |
c713216d MG |
5258 | unsigned long *ignored) |
5259 | { | |
96e907d1 TH |
5260 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5261 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5262 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5263 | unsigned long nr_absent; |
9c7cd687 | 5264 | |
b5685e92 | 5265 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5266 | if (!node_start_pfn && !node_end_pfn) |
5267 | return 0; | |
5268 | ||
96e907d1 TH |
5269 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5270 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5271 | |
2a1e274a MG |
5272 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5273 | node_start_pfn, node_end_pfn, | |
5274 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5275 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5276 | ||
5277 | /* | |
5278 | * ZONE_MOVABLE handling. | |
5279 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
5280 | * and vice versa. | |
5281 | */ | |
5282 | if (zone_movable_pfn[nid]) { | |
5283 | if (mirrored_kernelcore) { | |
5284 | unsigned long start_pfn, end_pfn; | |
5285 | struct memblock_region *r; | |
5286 | ||
5287 | for_each_memblock(memory, r) { | |
5288 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
5289 | zone_start_pfn, zone_end_pfn); | |
5290 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
5291 | zone_start_pfn, zone_end_pfn); | |
5292 | ||
5293 | if (zone_type == ZONE_MOVABLE && | |
5294 | memblock_is_mirror(r)) | |
5295 | nr_absent += end_pfn - start_pfn; | |
5296 | ||
5297 | if (zone_type == ZONE_NORMAL && | |
5298 | !memblock_is_mirror(r)) | |
5299 | nr_absent += end_pfn - start_pfn; | |
5300 | } | |
5301 | } else { | |
5302 | if (zone_type == ZONE_NORMAL) | |
5303 | nr_absent += node_end_pfn - zone_movable_pfn[nid]; | |
5304 | } | |
5305 | } | |
5306 | ||
5307 | return nr_absent; | |
c713216d | 5308 | } |
0e0b864e | 5309 | |
0ee332c1 | 5310 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 5311 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5312 | unsigned long zone_type, |
7960aedd ZY |
5313 | unsigned long node_start_pfn, |
5314 | unsigned long node_end_pfn, | |
d91749c1 TI |
5315 | unsigned long *zone_start_pfn, |
5316 | unsigned long *zone_end_pfn, | |
c713216d MG |
5317 | unsigned long *zones_size) |
5318 | { | |
d91749c1 TI |
5319 | unsigned int zone; |
5320 | ||
5321 | *zone_start_pfn = node_start_pfn; | |
5322 | for (zone = 0; zone < zone_type; zone++) | |
5323 | *zone_start_pfn += zones_size[zone]; | |
5324 | ||
5325 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
5326 | ||
c713216d MG |
5327 | return zones_size[zone_type]; |
5328 | } | |
5329 | ||
6ea6e688 | 5330 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5331 | unsigned long zone_type, |
7960aedd ZY |
5332 | unsigned long node_start_pfn, |
5333 | unsigned long node_end_pfn, | |
c713216d MG |
5334 | unsigned long *zholes_size) |
5335 | { | |
5336 | if (!zholes_size) | |
5337 | return 0; | |
5338 | ||
5339 | return zholes_size[zone_type]; | |
5340 | } | |
20e6926d | 5341 | |
0ee332c1 | 5342 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5343 | |
a3142c8e | 5344 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
5345 | unsigned long node_start_pfn, |
5346 | unsigned long node_end_pfn, | |
5347 | unsigned long *zones_size, | |
5348 | unsigned long *zholes_size) | |
c713216d | 5349 | { |
febd5949 | 5350 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
5351 | enum zone_type i; |
5352 | ||
febd5949 GZ |
5353 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5354 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 5355 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 5356 | unsigned long size, real_size; |
c713216d | 5357 | |
febd5949 GZ |
5358 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5359 | node_start_pfn, | |
5360 | node_end_pfn, | |
d91749c1 TI |
5361 | &zone_start_pfn, |
5362 | &zone_end_pfn, | |
febd5949 GZ |
5363 | zones_size); |
5364 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
5365 | node_start_pfn, node_end_pfn, |
5366 | zholes_size); | |
d91749c1 TI |
5367 | if (size) |
5368 | zone->zone_start_pfn = zone_start_pfn; | |
5369 | else | |
5370 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
5371 | zone->spanned_pages = size; |
5372 | zone->present_pages = real_size; | |
5373 | ||
5374 | totalpages += size; | |
5375 | realtotalpages += real_size; | |
5376 | } | |
5377 | ||
5378 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
5379 | pgdat->node_present_pages = realtotalpages; |
5380 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
5381 | realtotalpages); | |
5382 | } | |
5383 | ||
835c134e MG |
5384 | #ifndef CONFIG_SPARSEMEM |
5385 | /* | |
5386 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
5387 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5388 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
5389 | * round what is now in bits to nearest long in bits, then return it in |
5390 | * bytes. | |
5391 | */ | |
7c45512d | 5392 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
5393 | { |
5394 | unsigned long usemapsize; | |
5395 | ||
7c45512d | 5396 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
5397 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
5398 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
5399 | usemapsize *= NR_PAGEBLOCK_BITS; |
5400 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
5401 | ||
5402 | return usemapsize / 8; | |
5403 | } | |
5404 | ||
5405 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
5406 | struct zone *zone, |
5407 | unsigned long zone_start_pfn, | |
5408 | unsigned long zonesize) | |
835c134e | 5409 | { |
7c45512d | 5410 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 5411 | zone->pageblock_flags = NULL; |
58a01a45 | 5412 | if (usemapsize) |
6782832e SS |
5413 | zone->pageblock_flags = |
5414 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5415 | pgdat->node_id); | |
835c134e MG |
5416 | } |
5417 | #else | |
7c45512d LT |
5418 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5419 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5420 | #endif /* CONFIG_SPARSEMEM */ |
5421 | ||
d9c23400 | 5422 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5423 | |
d9c23400 | 5424 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5425 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5426 | { |
955c1cd7 AM |
5427 | unsigned int order; |
5428 | ||
d9c23400 MG |
5429 | /* Check that pageblock_nr_pages has not already been setup */ |
5430 | if (pageblock_order) | |
5431 | return; | |
5432 | ||
955c1cd7 AM |
5433 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5434 | order = HUGETLB_PAGE_ORDER; | |
5435 | else | |
5436 | order = MAX_ORDER - 1; | |
5437 | ||
d9c23400 MG |
5438 | /* |
5439 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5440 | * This value may be variable depending on boot parameters on IA64 and |
5441 | * powerpc. | |
d9c23400 MG |
5442 | */ |
5443 | pageblock_order = order; | |
5444 | } | |
5445 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5446 | ||
ba72cb8c MG |
5447 | /* |
5448 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5449 | * is unused as pageblock_order is set at compile-time. See |
5450 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5451 | * the kernel config | |
ba72cb8c | 5452 | */ |
15ca220e | 5453 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5454 | { |
ba72cb8c | 5455 | } |
d9c23400 MG |
5456 | |
5457 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5458 | ||
01cefaef JL |
5459 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5460 | unsigned long present_pages) | |
5461 | { | |
5462 | unsigned long pages = spanned_pages; | |
5463 | ||
5464 | /* | |
5465 | * Provide a more accurate estimation if there are holes within | |
5466 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5467 | * zone, each populated memory region may cost us one or two extra | |
5468 | * memmap pages due to alignment because memmap pages for each | |
5469 | * populated regions may not naturally algined on page boundary. | |
5470 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
5471 | */ | |
5472 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5473 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5474 | pages = present_pages; | |
5475 | ||
5476 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5477 | } | |
5478 | ||
1da177e4 LT |
5479 | /* |
5480 | * Set up the zone data structures: | |
5481 | * - mark all pages reserved | |
5482 | * - mark all memory queues empty | |
5483 | * - clear the memory bitmaps | |
6527af5d MK |
5484 | * |
5485 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 5486 | */ |
7f3eb55b | 5487 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 5488 | { |
2f1b6248 | 5489 | enum zone_type j; |
ed8ece2e | 5490 | int nid = pgdat->node_id; |
718127cc | 5491 | int ret; |
1da177e4 | 5492 | |
208d54e5 | 5493 | pgdat_resize_init(pgdat); |
8177a420 AA |
5494 | #ifdef CONFIG_NUMA_BALANCING |
5495 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
5496 | pgdat->numabalancing_migrate_nr_pages = 0; | |
5497 | pgdat->numabalancing_migrate_next_window = jiffies; | |
a3d0a918 KS |
5498 | #endif |
5499 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5500 | spin_lock_init(&pgdat->split_queue_lock); | |
5501 | INIT_LIST_HEAD(&pgdat->split_queue); | |
5502 | pgdat->split_queue_len = 0; | |
8177a420 | 5503 | #endif |
1da177e4 | 5504 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 5505 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
698b1b30 VB |
5506 | #ifdef CONFIG_COMPACTION |
5507 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
5508 | #endif | |
eefa864b | 5509 | pgdat_page_ext_init(pgdat); |
5f63b720 | 5510 | |
1da177e4 LT |
5511 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5512 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 5513 | unsigned long size, realsize, freesize, memmap_pages; |
d91749c1 | 5514 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 5515 | |
febd5949 GZ |
5516 | size = zone->spanned_pages; |
5517 | realsize = freesize = zone->present_pages; | |
1da177e4 | 5518 | |
0e0b864e | 5519 | /* |
9feedc9d | 5520 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
5521 | * is used by this zone for memmap. This affects the watermark |
5522 | * and per-cpu initialisations | |
5523 | */ | |
01cefaef | 5524 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
5525 | if (!is_highmem_idx(j)) { |
5526 | if (freesize >= memmap_pages) { | |
5527 | freesize -= memmap_pages; | |
5528 | if (memmap_pages) | |
5529 | printk(KERN_DEBUG | |
5530 | " %s zone: %lu pages used for memmap\n", | |
5531 | zone_names[j], memmap_pages); | |
5532 | } else | |
1170532b | 5533 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
5534 | zone_names[j], memmap_pages, freesize); |
5535 | } | |
0e0b864e | 5536 | |
6267276f | 5537 | /* Account for reserved pages */ |
9feedc9d JL |
5538 | if (j == 0 && freesize > dma_reserve) { |
5539 | freesize -= dma_reserve; | |
d903ef9f | 5540 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 5541 | zone_names[0], dma_reserve); |
0e0b864e MG |
5542 | } |
5543 | ||
98d2b0eb | 5544 | if (!is_highmem_idx(j)) |
9feedc9d | 5545 | nr_kernel_pages += freesize; |
01cefaef JL |
5546 | /* Charge for highmem memmap if there are enough kernel pages */ |
5547 | else if (nr_kernel_pages > memmap_pages * 2) | |
5548 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 5549 | nr_all_pages += freesize; |
1da177e4 | 5550 | |
9feedc9d JL |
5551 | /* |
5552 | * Set an approximate value for lowmem here, it will be adjusted | |
5553 | * when the bootmem allocator frees pages into the buddy system. | |
5554 | * And all highmem pages will be managed by the buddy system. | |
5555 | */ | |
5556 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 5557 | #ifdef CONFIG_NUMA |
d5f541ed | 5558 | zone->node = nid; |
9feedc9d | 5559 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 5560 | / 100; |
9feedc9d | 5561 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 5562 | #endif |
1da177e4 LT |
5563 | zone->name = zone_names[j]; |
5564 | spin_lock_init(&zone->lock); | |
5565 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 5566 | zone_seqlock_init(zone); |
1da177e4 | 5567 | zone->zone_pgdat = pgdat; |
ed8ece2e | 5568 | zone_pcp_init(zone); |
81c0a2bb JW |
5569 | |
5570 | /* For bootup, initialized properly in watermark setup */ | |
5571 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
5572 | ||
bea8c150 | 5573 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
5574 | if (!size) |
5575 | continue; | |
5576 | ||
955c1cd7 | 5577 | set_pageblock_order(); |
7c45512d | 5578 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
b171e409 | 5579 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
718127cc | 5580 | BUG_ON(ret); |
76cdd58e | 5581 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
5582 | } |
5583 | } | |
5584 | ||
577a32f6 | 5585 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 5586 | { |
b0aeba74 | 5587 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
5588 | unsigned long __maybe_unused offset = 0; |
5589 | ||
1da177e4 LT |
5590 | /* Skip empty nodes */ |
5591 | if (!pgdat->node_spanned_pages) | |
5592 | return; | |
5593 | ||
d41dee36 | 5594 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
b0aeba74 TL |
5595 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
5596 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
5597 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
5598 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 5599 | unsigned long size, end; |
d41dee36 AW |
5600 | struct page *map; |
5601 | ||
e984bb43 BP |
5602 | /* |
5603 | * The zone's endpoints aren't required to be MAX_ORDER | |
5604 | * aligned but the node_mem_map endpoints must be in order | |
5605 | * for the buddy allocator to function correctly. | |
5606 | */ | |
108bcc96 | 5607 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
5608 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
5609 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
5610 | map = alloc_remap(pgdat->node_id, size); |
5611 | if (!map) | |
6782832e SS |
5612 | map = memblock_virt_alloc_node_nopanic(size, |
5613 | pgdat->node_id); | |
a1c34a3b | 5614 | pgdat->node_mem_map = map + offset; |
1da177e4 | 5615 | } |
12d810c1 | 5616 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
5617 | /* |
5618 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
5619 | */ | |
c713216d | 5620 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 5621 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 5622 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 5623 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 5624 | mem_map -= offset; |
0ee332c1 | 5625 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5626 | } |
1da177e4 | 5627 | #endif |
d41dee36 | 5628 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
5629 | } |
5630 | ||
9109fb7b JW |
5631 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
5632 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 5633 | { |
9109fb7b | 5634 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
5635 | unsigned long start_pfn = 0; |
5636 | unsigned long end_pfn = 0; | |
9109fb7b | 5637 | |
88fdf75d | 5638 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 5639 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 5640 | |
3a80a7fa | 5641 | reset_deferred_meminit(pgdat); |
1da177e4 LT |
5642 | pgdat->node_id = nid; |
5643 | pgdat->node_start_pfn = node_start_pfn; | |
7960aedd ZY |
5644 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5645 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 5646 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
5647 | (u64)start_pfn << PAGE_SHIFT, |
5648 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
5649 | #else |
5650 | start_pfn = node_start_pfn; | |
7960aedd ZY |
5651 | #endif |
5652 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
5653 | zones_size, zholes_size); | |
1da177e4 LT |
5654 | |
5655 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
5656 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
5657 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
5658 | nid, (unsigned long)pgdat, | |
5659 | (unsigned long)pgdat->node_mem_map); | |
5660 | #endif | |
1da177e4 | 5661 | |
7f3eb55b | 5662 | free_area_init_core(pgdat); |
1da177e4 LT |
5663 | } |
5664 | ||
0ee332c1 | 5665 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
5666 | |
5667 | #if MAX_NUMNODES > 1 | |
5668 | /* | |
5669 | * Figure out the number of possible node ids. | |
5670 | */ | |
f9872caf | 5671 | void __init setup_nr_node_ids(void) |
418508c1 | 5672 | { |
904a9553 | 5673 | unsigned int highest; |
418508c1 | 5674 | |
904a9553 | 5675 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
5676 | nr_node_ids = highest + 1; |
5677 | } | |
418508c1 MS |
5678 | #endif |
5679 | ||
1e01979c TH |
5680 | /** |
5681 | * node_map_pfn_alignment - determine the maximum internode alignment | |
5682 | * | |
5683 | * This function should be called after node map is populated and sorted. | |
5684 | * It calculates the maximum power of two alignment which can distinguish | |
5685 | * all the nodes. | |
5686 | * | |
5687 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
5688 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
5689 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
5690 | * shifted, 1GiB is enough and this function will indicate so. | |
5691 | * | |
5692 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
5693 | * model has fine enough granularity to avoid incorrect mapping for the | |
5694 | * populated node map. | |
5695 | * | |
5696 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
5697 | * requirement (single node). | |
5698 | */ | |
5699 | unsigned long __init node_map_pfn_alignment(void) | |
5700 | { | |
5701 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 5702 | unsigned long start, end, mask; |
1e01979c | 5703 | int last_nid = -1; |
c13291a5 | 5704 | int i, nid; |
1e01979c | 5705 | |
c13291a5 | 5706 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5707 | if (!start || last_nid < 0 || last_nid == nid) { |
5708 | last_nid = nid; | |
5709 | last_end = end; | |
5710 | continue; | |
5711 | } | |
5712 | ||
5713 | /* | |
5714 | * Start with a mask granular enough to pin-point to the | |
5715 | * start pfn and tick off bits one-by-one until it becomes | |
5716 | * too coarse to separate the current node from the last. | |
5717 | */ | |
5718 | mask = ~((1 << __ffs(start)) - 1); | |
5719 | while (mask && last_end <= (start & (mask << 1))) | |
5720 | mask <<= 1; | |
5721 | ||
5722 | /* accumulate all internode masks */ | |
5723 | accl_mask |= mask; | |
5724 | } | |
5725 | ||
5726 | /* convert mask to number of pages */ | |
5727 | return ~accl_mask + 1; | |
5728 | } | |
5729 | ||
a6af2bc3 | 5730 | /* Find the lowest pfn for a node */ |
b69a7288 | 5731 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 5732 | { |
a6af2bc3 | 5733 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
5734 | unsigned long start_pfn; |
5735 | int i; | |
1abbfb41 | 5736 | |
c13291a5 TH |
5737 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
5738 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 5739 | |
a6af2bc3 | 5740 | if (min_pfn == ULONG_MAX) { |
1170532b | 5741 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
5742 | return 0; |
5743 | } | |
5744 | ||
5745 | return min_pfn; | |
c713216d MG |
5746 | } |
5747 | ||
5748 | /** | |
5749 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
5750 | * | |
5751 | * It returns the minimum PFN based on information provided via | |
7d018176 | 5752 | * memblock_set_node(). |
c713216d MG |
5753 | */ |
5754 | unsigned long __init find_min_pfn_with_active_regions(void) | |
5755 | { | |
5756 | return find_min_pfn_for_node(MAX_NUMNODES); | |
5757 | } | |
5758 | ||
37b07e41 LS |
5759 | /* |
5760 | * early_calculate_totalpages() | |
5761 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 5762 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 5763 | */ |
484f51f8 | 5764 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 5765 | { |
7e63efef | 5766 | unsigned long totalpages = 0; |
c13291a5 TH |
5767 | unsigned long start_pfn, end_pfn; |
5768 | int i, nid; | |
5769 | ||
5770 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
5771 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 5772 | |
37b07e41 LS |
5773 | totalpages += pages; |
5774 | if (pages) | |
4b0ef1fe | 5775 | node_set_state(nid, N_MEMORY); |
37b07e41 | 5776 | } |
b8af2941 | 5777 | return totalpages; |
7e63efef MG |
5778 | } |
5779 | ||
2a1e274a MG |
5780 | /* |
5781 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
5782 | * is spread evenly between nodes as long as the nodes have enough | |
5783 | * memory. When they don't, some nodes will have more kernelcore than | |
5784 | * others | |
5785 | */ | |
b224ef85 | 5786 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
5787 | { |
5788 | int i, nid; | |
5789 | unsigned long usable_startpfn; | |
5790 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 5791 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 5792 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 5793 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 5794 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 5795 | struct memblock_region *r; |
b2f3eebe TC |
5796 | |
5797 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
5798 | find_usable_zone_for_movable(); | |
5799 | ||
5800 | /* | |
5801 | * If movable_node is specified, ignore kernelcore and movablecore | |
5802 | * options. | |
5803 | */ | |
5804 | if (movable_node_is_enabled()) { | |
136199f0 EM |
5805 | for_each_memblock(memory, r) { |
5806 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
5807 | continue; |
5808 | ||
136199f0 | 5809 | nid = r->nid; |
b2f3eebe | 5810 | |
136199f0 | 5811 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
5812 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
5813 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5814 | usable_startpfn; | |
5815 | } | |
5816 | ||
5817 | goto out2; | |
5818 | } | |
2a1e274a | 5819 | |
342332e6 TI |
5820 | /* |
5821 | * If kernelcore=mirror is specified, ignore movablecore option | |
5822 | */ | |
5823 | if (mirrored_kernelcore) { | |
5824 | bool mem_below_4gb_not_mirrored = false; | |
5825 | ||
5826 | for_each_memblock(memory, r) { | |
5827 | if (memblock_is_mirror(r)) | |
5828 | continue; | |
5829 | ||
5830 | nid = r->nid; | |
5831 | ||
5832 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
5833 | ||
5834 | if (usable_startpfn < 0x100000) { | |
5835 | mem_below_4gb_not_mirrored = true; | |
5836 | continue; | |
5837 | } | |
5838 | ||
5839 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
5840 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5841 | usable_startpfn; | |
5842 | } | |
5843 | ||
5844 | if (mem_below_4gb_not_mirrored) | |
5845 | pr_warn("This configuration results in unmirrored kernel memory."); | |
5846 | ||
5847 | goto out2; | |
5848 | } | |
5849 | ||
7e63efef | 5850 | /* |
b2f3eebe | 5851 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
5852 | * kernelcore that corresponds so that memory usable for |
5853 | * any allocation type is evenly spread. If both kernelcore | |
5854 | * and movablecore are specified, then the value of kernelcore | |
5855 | * will be used for required_kernelcore if it's greater than | |
5856 | * what movablecore would have allowed. | |
5857 | */ | |
5858 | if (required_movablecore) { | |
7e63efef MG |
5859 | unsigned long corepages; |
5860 | ||
5861 | /* | |
5862 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
5863 | * was requested by the user | |
5864 | */ | |
5865 | required_movablecore = | |
5866 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 5867 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
5868 | corepages = totalpages - required_movablecore; |
5869 | ||
5870 | required_kernelcore = max(required_kernelcore, corepages); | |
5871 | } | |
5872 | ||
bde304bd XQ |
5873 | /* |
5874 | * If kernelcore was not specified or kernelcore size is larger | |
5875 | * than totalpages, there is no ZONE_MOVABLE. | |
5876 | */ | |
5877 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 5878 | goto out; |
2a1e274a MG |
5879 | |
5880 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
5881 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
5882 | ||
5883 | restart: | |
5884 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
5885 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 5886 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
5887 | unsigned long start_pfn, end_pfn; |
5888 | ||
2a1e274a MG |
5889 | /* |
5890 | * Recalculate kernelcore_node if the division per node | |
5891 | * now exceeds what is necessary to satisfy the requested | |
5892 | * amount of memory for the kernel | |
5893 | */ | |
5894 | if (required_kernelcore < kernelcore_node) | |
5895 | kernelcore_node = required_kernelcore / usable_nodes; | |
5896 | ||
5897 | /* | |
5898 | * As the map is walked, we track how much memory is usable | |
5899 | * by the kernel using kernelcore_remaining. When it is | |
5900 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
5901 | */ | |
5902 | kernelcore_remaining = kernelcore_node; | |
5903 | ||
5904 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 5905 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
5906 | unsigned long size_pages; |
5907 | ||
c13291a5 | 5908 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
5909 | if (start_pfn >= end_pfn) |
5910 | continue; | |
5911 | ||
5912 | /* Account for what is only usable for kernelcore */ | |
5913 | if (start_pfn < usable_startpfn) { | |
5914 | unsigned long kernel_pages; | |
5915 | kernel_pages = min(end_pfn, usable_startpfn) | |
5916 | - start_pfn; | |
5917 | ||
5918 | kernelcore_remaining -= min(kernel_pages, | |
5919 | kernelcore_remaining); | |
5920 | required_kernelcore -= min(kernel_pages, | |
5921 | required_kernelcore); | |
5922 | ||
5923 | /* Continue if range is now fully accounted */ | |
5924 | if (end_pfn <= usable_startpfn) { | |
5925 | ||
5926 | /* | |
5927 | * Push zone_movable_pfn to the end so | |
5928 | * that if we have to rebalance | |
5929 | * kernelcore across nodes, we will | |
5930 | * not double account here | |
5931 | */ | |
5932 | zone_movable_pfn[nid] = end_pfn; | |
5933 | continue; | |
5934 | } | |
5935 | start_pfn = usable_startpfn; | |
5936 | } | |
5937 | ||
5938 | /* | |
5939 | * The usable PFN range for ZONE_MOVABLE is from | |
5940 | * start_pfn->end_pfn. Calculate size_pages as the | |
5941 | * number of pages used as kernelcore | |
5942 | */ | |
5943 | size_pages = end_pfn - start_pfn; | |
5944 | if (size_pages > kernelcore_remaining) | |
5945 | size_pages = kernelcore_remaining; | |
5946 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5947 | ||
5948 | /* | |
5949 | * Some kernelcore has been met, update counts and | |
5950 | * break if the kernelcore for this node has been | |
b8af2941 | 5951 | * satisfied |
2a1e274a MG |
5952 | */ |
5953 | required_kernelcore -= min(required_kernelcore, | |
5954 | size_pages); | |
5955 | kernelcore_remaining -= size_pages; | |
5956 | if (!kernelcore_remaining) | |
5957 | break; | |
5958 | } | |
5959 | } | |
5960 | ||
5961 | /* | |
5962 | * If there is still required_kernelcore, we do another pass with one | |
5963 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5964 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 5965 | * satisfied |
2a1e274a MG |
5966 | */ |
5967 | usable_nodes--; | |
5968 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5969 | goto restart; | |
5970 | ||
b2f3eebe | 5971 | out2: |
2a1e274a MG |
5972 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
5973 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5974 | zone_movable_pfn[nid] = | |
5975 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5976 | |
20e6926d | 5977 | out: |
66918dcd | 5978 | /* restore the node_state */ |
4b0ef1fe | 5979 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5980 | } |
5981 | ||
4b0ef1fe LJ |
5982 | /* Any regular or high memory on that node ? */ |
5983 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5984 | { |
37b07e41 LS |
5985 | enum zone_type zone_type; |
5986 | ||
4b0ef1fe LJ |
5987 | if (N_MEMORY == N_NORMAL_MEMORY) |
5988 | return; | |
5989 | ||
5990 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5991 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 5992 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
5993 | node_set_state(nid, N_HIGH_MEMORY); |
5994 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5995 | zone_type <= ZONE_NORMAL) | |
5996 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5997 | break; |
5998 | } | |
37b07e41 | 5999 | } |
37b07e41 LS |
6000 | } |
6001 | ||
c713216d MG |
6002 | /** |
6003 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6004 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6005 | * |
6006 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6007 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6008 | * zone in each node and their holes is calculated. If the maximum PFN |
6009 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6010 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6011 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6012 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6013 | * at arch_max_dma_pfn. | |
6014 | */ | |
6015 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6016 | { | |
c13291a5 TH |
6017 | unsigned long start_pfn, end_pfn; |
6018 | int i, nid; | |
a6af2bc3 | 6019 | |
c713216d MG |
6020 | /* Record where the zone boundaries are */ |
6021 | memset(arch_zone_lowest_possible_pfn, 0, | |
6022 | sizeof(arch_zone_lowest_possible_pfn)); | |
6023 | memset(arch_zone_highest_possible_pfn, 0, | |
6024 | sizeof(arch_zone_highest_possible_pfn)); | |
6025 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
6026 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
6027 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6028 | if (i == ZONE_MOVABLE) |
6029 | continue; | |
c713216d MG |
6030 | arch_zone_lowest_possible_pfn[i] = |
6031 | arch_zone_highest_possible_pfn[i-1]; | |
6032 | arch_zone_highest_possible_pfn[i] = | |
6033 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
6034 | } | |
2a1e274a MG |
6035 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
6036 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
6037 | ||
6038 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6039 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6040 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6041 | |
c713216d | 6042 | /* Print out the zone ranges */ |
f88dfff5 | 6043 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6044 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6045 | if (i == ZONE_MOVABLE) | |
6046 | continue; | |
f88dfff5 | 6047 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6048 | if (arch_zone_lowest_possible_pfn[i] == |
6049 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6050 | pr_cont("empty\n"); |
72f0ba02 | 6051 | else |
8d29e18a JG |
6052 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6053 | (u64)arch_zone_lowest_possible_pfn[i] | |
6054 | << PAGE_SHIFT, | |
6055 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6056 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6057 | } |
6058 | ||
6059 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6060 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6061 | for (i = 0; i < MAX_NUMNODES; i++) { |
6062 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6063 | pr_info(" Node %d: %#018Lx\n", i, |
6064 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6065 | } |
c713216d | 6066 | |
f2d52fe5 | 6067 | /* Print out the early node map */ |
f88dfff5 | 6068 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6069 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6070 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6071 | (u64)start_pfn << PAGE_SHIFT, | |
6072 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6073 | |
6074 | /* Initialise every node */ | |
708614e6 | 6075 | mminit_verify_pageflags_layout(); |
8ef82866 | 6076 | setup_nr_node_ids(); |
c713216d MG |
6077 | for_each_online_node(nid) { |
6078 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6079 | free_area_init_node(nid, NULL, |
c713216d | 6080 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6081 | |
6082 | /* Any memory on that node */ | |
6083 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6084 | node_set_state(nid, N_MEMORY); |
6085 | check_for_memory(pgdat, nid); | |
c713216d MG |
6086 | } |
6087 | } | |
2a1e274a | 6088 | |
7e63efef | 6089 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
6090 | { |
6091 | unsigned long long coremem; | |
6092 | if (!p) | |
6093 | return -EINVAL; | |
6094 | ||
6095 | coremem = memparse(p, &p); | |
7e63efef | 6096 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 6097 | |
7e63efef | 6098 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
6099 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6100 | ||
6101 | return 0; | |
6102 | } | |
ed7ed365 | 6103 | |
7e63efef MG |
6104 | /* |
6105 | * kernelcore=size sets the amount of memory for use for allocations that | |
6106 | * cannot be reclaimed or migrated. | |
6107 | */ | |
6108 | static int __init cmdline_parse_kernelcore(char *p) | |
6109 | { | |
342332e6 TI |
6110 | /* parse kernelcore=mirror */ |
6111 | if (parse_option_str(p, "mirror")) { | |
6112 | mirrored_kernelcore = true; | |
6113 | return 0; | |
6114 | } | |
6115 | ||
7e63efef MG |
6116 | return cmdline_parse_core(p, &required_kernelcore); |
6117 | } | |
6118 | ||
6119 | /* | |
6120 | * movablecore=size sets the amount of memory for use for allocations that | |
6121 | * can be reclaimed or migrated. | |
6122 | */ | |
6123 | static int __init cmdline_parse_movablecore(char *p) | |
6124 | { | |
6125 | return cmdline_parse_core(p, &required_movablecore); | |
6126 | } | |
6127 | ||
ed7ed365 | 6128 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6129 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6130 | |
0ee332c1 | 6131 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6132 | |
c3d5f5f0 JL |
6133 | void adjust_managed_page_count(struct page *page, long count) |
6134 | { | |
6135 | spin_lock(&managed_page_count_lock); | |
6136 | page_zone(page)->managed_pages += count; | |
6137 | totalram_pages += count; | |
3dcc0571 JL |
6138 | #ifdef CONFIG_HIGHMEM |
6139 | if (PageHighMem(page)) | |
6140 | totalhigh_pages += count; | |
6141 | #endif | |
c3d5f5f0 JL |
6142 | spin_unlock(&managed_page_count_lock); |
6143 | } | |
3dcc0571 | 6144 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6145 | |
11199692 | 6146 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6147 | { |
11199692 JL |
6148 | void *pos; |
6149 | unsigned long pages = 0; | |
69afade7 | 6150 | |
11199692 JL |
6151 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6152 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6153 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 6154 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
6155 | memset(pos, poison, PAGE_SIZE); |
6156 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
6157 | } |
6158 | ||
6159 | if (pages && s) | |
11199692 | 6160 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
6161 | s, pages << (PAGE_SHIFT - 10), start, end); |
6162 | ||
6163 | return pages; | |
6164 | } | |
11199692 | 6165 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6166 | |
cfa11e08 JL |
6167 | #ifdef CONFIG_HIGHMEM |
6168 | void free_highmem_page(struct page *page) | |
6169 | { | |
6170 | __free_reserved_page(page); | |
6171 | totalram_pages++; | |
7b4b2a0d | 6172 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6173 | totalhigh_pages++; |
6174 | } | |
6175 | #endif | |
6176 | ||
7ee3d4e8 JL |
6177 | |
6178 | void __init mem_init_print_info(const char *str) | |
6179 | { | |
6180 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
6181 | unsigned long init_code_size, init_data_size; | |
6182 | ||
6183 | physpages = get_num_physpages(); | |
6184 | codesize = _etext - _stext; | |
6185 | datasize = _edata - _sdata; | |
6186 | rosize = __end_rodata - __start_rodata; | |
6187 | bss_size = __bss_stop - __bss_start; | |
6188 | init_data_size = __init_end - __init_begin; | |
6189 | init_code_size = _einittext - _sinittext; | |
6190 | ||
6191 | /* | |
6192 | * Detect special cases and adjust section sizes accordingly: | |
6193 | * 1) .init.* may be embedded into .data sections | |
6194 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
6195 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
6196 | * 3) .rodata.* may be embedded into .text or .data sections. | |
6197 | */ | |
6198 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
6199 | do { \ |
6200 | if (start <= pos && pos < end && size > adj) \ | |
6201 | size -= adj; \ | |
6202 | } while (0) | |
7ee3d4e8 JL |
6203 | |
6204 | adj_init_size(__init_begin, __init_end, init_data_size, | |
6205 | _sinittext, init_code_size); | |
6206 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
6207 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
6208 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
6209 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
6210 | ||
6211 | #undef adj_init_size | |
6212 | ||
756a025f | 6213 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 6214 | #ifdef CONFIG_HIGHMEM |
756a025f | 6215 | ", %luK highmem" |
7ee3d4e8 | 6216 | #endif |
756a025f JP |
6217 | "%s%s)\n", |
6218 | nr_free_pages() << (PAGE_SHIFT - 10), | |
6219 | physpages << (PAGE_SHIFT - 10), | |
6220 | codesize >> 10, datasize >> 10, rosize >> 10, | |
6221 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
6222 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
6223 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 6224 | #ifdef CONFIG_HIGHMEM |
756a025f | 6225 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 6226 | #endif |
756a025f | 6227 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
6228 | } |
6229 | ||
0e0b864e | 6230 | /** |
88ca3b94 RD |
6231 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6232 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 6233 | * |
013110a7 | 6234 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
6235 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6236 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
6237 | * function may optionally be used to account for unfreeable pages in the |
6238 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
6239 | * smaller per-cpu batchsize. | |
0e0b864e MG |
6240 | */ |
6241 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
6242 | { | |
6243 | dma_reserve = new_dma_reserve; | |
6244 | } | |
6245 | ||
1da177e4 LT |
6246 | void __init free_area_init(unsigned long *zones_size) |
6247 | { | |
9109fb7b | 6248 | free_area_init_node(0, zones_size, |
1da177e4 LT |
6249 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6250 | } | |
1da177e4 | 6251 | |
1da177e4 LT |
6252 | static int page_alloc_cpu_notify(struct notifier_block *self, |
6253 | unsigned long action, void *hcpu) | |
6254 | { | |
6255 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 6256 | |
8bb78442 | 6257 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 6258 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
6259 | drain_pages(cpu); |
6260 | ||
6261 | /* | |
6262 | * Spill the event counters of the dead processor | |
6263 | * into the current processors event counters. | |
6264 | * This artificially elevates the count of the current | |
6265 | * processor. | |
6266 | */ | |
f8891e5e | 6267 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
6268 | |
6269 | /* | |
6270 | * Zero the differential counters of the dead processor | |
6271 | * so that the vm statistics are consistent. | |
6272 | * | |
6273 | * This is only okay since the processor is dead and cannot | |
6274 | * race with what we are doing. | |
6275 | */ | |
2bb921e5 | 6276 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
6277 | } |
6278 | return NOTIFY_OK; | |
6279 | } | |
1da177e4 LT |
6280 | |
6281 | void __init page_alloc_init(void) | |
6282 | { | |
6283 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
6284 | } | |
6285 | ||
cb45b0e9 | 6286 | /* |
34b10060 | 6287 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
6288 | * or min_free_kbytes changes. |
6289 | */ | |
6290 | static void calculate_totalreserve_pages(void) | |
6291 | { | |
6292 | struct pglist_data *pgdat; | |
6293 | unsigned long reserve_pages = 0; | |
2f6726e5 | 6294 | enum zone_type i, j; |
cb45b0e9 HA |
6295 | |
6296 | for_each_online_pgdat(pgdat) { | |
6297 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
6298 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 6299 | long max = 0; |
cb45b0e9 HA |
6300 | |
6301 | /* Find valid and maximum lowmem_reserve in the zone */ | |
6302 | for (j = i; j < MAX_NR_ZONES; j++) { | |
6303 | if (zone->lowmem_reserve[j] > max) | |
6304 | max = zone->lowmem_reserve[j]; | |
6305 | } | |
6306 | ||
41858966 MG |
6307 | /* we treat the high watermark as reserved pages. */ |
6308 | max += high_wmark_pages(zone); | |
cb45b0e9 | 6309 | |
b40da049 JL |
6310 | if (max > zone->managed_pages) |
6311 | max = zone->managed_pages; | |
a8d01437 JW |
6312 | |
6313 | zone->totalreserve_pages = max; | |
6314 | ||
cb45b0e9 HA |
6315 | reserve_pages += max; |
6316 | } | |
6317 | } | |
6318 | totalreserve_pages = reserve_pages; | |
6319 | } | |
6320 | ||
1da177e4 LT |
6321 | /* |
6322 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 6323 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
6324 | * has a correct pages reserved value, so an adequate number of |
6325 | * pages are left in the zone after a successful __alloc_pages(). | |
6326 | */ | |
6327 | static void setup_per_zone_lowmem_reserve(void) | |
6328 | { | |
6329 | struct pglist_data *pgdat; | |
2f6726e5 | 6330 | enum zone_type j, idx; |
1da177e4 | 6331 | |
ec936fc5 | 6332 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
6333 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6334 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 6335 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
6336 | |
6337 | zone->lowmem_reserve[j] = 0; | |
6338 | ||
2f6726e5 CL |
6339 | idx = j; |
6340 | while (idx) { | |
1da177e4 LT |
6341 | struct zone *lower_zone; |
6342 | ||
2f6726e5 CL |
6343 | idx--; |
6344 | ||
1da177e4 LT |
6345 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6346 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
6347 | ||
6348 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 6349 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 6350 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 6351 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
6352 | } |
6353 | } | |
6354 | } | |
cb45b0e9 HA |
6355 | |
6356 | /* update totalreserve_pages */ | |
6357 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6358 | } |
6359 | ||
cfd3da1e | 6360 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
6361 | { |
6362 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
6363 | unsigned long lowmem_pages = 0; | |
6364 | struct zone *zone; | |
6365 | unsigned long flags; | |
6366 | ||
6367 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
6368 | for_each_zone(zone) { | |
6369 | if (!is_highmem(zone)) | |
b40da049 | 6370 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
6371 | } |
6372 | ||
6373 | for_each_zone(zone) { | |
ac924c60 AM |
6374 | u64 tmp; |
6375 | ||
1125b4e3 | 6376 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 6377 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 6378 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
6379 | if (is_highmem(zone)) { |
6380 | /* | |
669ed175 NP |
6381 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6382 | * need highmem pages, so cap pages_min to a small | |
6383 | * value here. | |
6384 | * | |
41858966 | 6385 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 6386 | * deltas control asynch page reclaim, and so should |
669ed175 | 6387 | * not be capped for highmem. |
1da177e4 | 6388 | */ |
90ae8d67 | 6389 | unsigned long min_pages; |
1da177e4 | 6390 | |
b40da049 | 6391 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 6392 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 6393 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 6394 | } else { |
669ed175 NP |
6395 | /* |
6396 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
6397 | * proportionate to the zone's size. |
6398 | */ | |
41858966 | 6399 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
6400 | } |
6401 | ||
795ae7a0 JW |
6402 | /* |
6403 | * Set the kswapd watermarks distance according to the | |
6404 | * scale factor in proportion to available memory, but | |
6405 | * ensure a minimum size on small systems. | |
6406 | */ | |
6407 | tmp = max_t(u64, tmp >> 2, | |
6408 | mult_frac(zone->managed_pages, | |
6409 | watermark_scale_factor, 10000)); | |
6410 | ||
6411 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
6412 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 6413 | |
81c0a2bb | 6414 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
abe5f972 JW |
6415 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
6416 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
81c0a2bb | 6417 | |
1125b4e3 | 6418 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 6419 | } |
cb45b0e9 HA |
6420 | |
6421 | /* update totalreserve_pages */ | |
6422 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6423 | } |
6424 | ||
cfd3da1e MG |
6425 | /** |
6426 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
6427 | * or when memory is hot-{added|removed} | |
6428 | * | |
6429 | * Ensures that the watermark[min,low,high] values for each zone are set | |
6430 | * correctly with respect to min_free_kbytes. | |
6431 | */ | |
6432 | void setup_per_zone_wmarks(void) | |
6433 | { | |
6434 | mutex_lock(&zonelists_mutex); | |
6435 | __setup_per_zone_wmarks(); | |
6436 | mutex_unlock(&zonelists_mutex); | |
6437 | } | |
6438 | ||
55a4462a | 6439 | /* |
556adecb RR |
6440 | * The inactive anon list should be small enough that the VM never has to |
6441 | * do too much work, but large enough that each inactive page has a chance | |
6442 | * to be referenced again before it is swapped out. | |
6443 | * | |
6444 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
6445 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
6446 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
6447 | * the anonymous pages are kept on the inactive list. | |
6448 | * | |
6449 | * total target max | |
6450 | * memory ratio inactive anon | |
6451 | * ------------------------------------- | |
6452 | * 10MB 1 5MB | |
6453 | * 100MB 1 50MB | |
6454 | * 1GB 3 250MB | |
6455 | * 10GB 10 0.9GB | |
6456 | * 100GB 31 3GB | |
6457 | * 1TB 101 10GB | |
6458 | * 10TB 320 32GB | |
6459 | */ | |
1b79acc9 | 6460 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 6461 | { |
96cb4df5 | 6462 | unsigned int gb, ratio; |
556adecb | 6463 | |
96cb4df5 | 6464 | /* Zone size in gigabytes */ |
b40da049 | 6465 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 6466 | if (gb) |
556adecb | 6467 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
6468 | else |
6469 | ratio = 1; | |
556adecb | 6470 | |
96cb4df5 MK |
6471 | zone->inactive_ratio = ratio; |
6472 | } | |
556adecb | 6473 | |
839a4fcc | 6474 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
6475 | { |
6476 | struct zone *zone; | |
6477 | ||
6478 | for_each_zone(zone) | |
6479 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
6480 | } |
6481 | ||
1da177e4 LT |
6482 | /* |
6483 | * Initialise min_free_kbytes. | |
6484 | * | |
6485 | * For small machines we want it small (128k min). For large machines | |
6486 | * we want it large (64MB max). But it is not linear, because network | |
6487 | * bandwidth does not increase linearly with machine size. We use | |
6488 | * | |
b8af2941 | 6489 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6490 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6491 | * | |
6492 | * which yields | |
6493 | * | |
6494 | * 16MB: 512k | |
6495 | * 32MB: 724k | |
6496 | * 64MB: 1024k | |
6497 | * 128MB: 1448k | |
6498 | * 256MB: 2048k | |
6499 | * 512MB: 2896k | |
6500 | * 1024MB: 4096k | |
6501 | * 2048MB: 5792k | |
6502 | * 4096MB: 8192k | |
6503 | * 8192MB: 11584k | |
6504 | * 16384MB: 16384k | |
6505 | */ | |
1b79acc9 | 6506 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6507 | { |
6508 | unsigned long lowmem_kbytes; | |
5f12733e | 6509 | int new_min_free_kbytes; |
1da177e4 LT |
6510 | |
6511 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6512 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6513 | ||
6514 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6515 | min_free_kbytes = new_min_free_kbytes; | |
6516 | if (min_free_kbytes < 128) | |
6517 | min_free_kbytes = 128; | |
6518 | if (min_free_kbytes > 65536) | |
6519 | min_free_kbytes = 65536; | |
6520 | } else { | |
6521 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6522 | new_min_free_kbytes, user_min_free_kbytes); | |
6523 | } | |
bc75d33f | 6524 | setup_per_zone_wmarks(); |
a6cccdc3 | 6525 | refresh_zone_stat_thresholds(); |
1da177e4 | 6526 | setup_per_zone_lowmem_reserve(); |
556adecb | 6527 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
6528 | return 0; |
6529 | } | |
bc22af74 | 6530 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
6531 | |
6532 | /* | |
b8af2941 | 6533 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6534 | * that we can call two helper functions whenever min_free_kbytes |
6535 | * changes. | |
6536 | */ | |
cccad5b9 | 6537 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6538 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6539 | { |
da8c757b HP |
6540 | int rc; |
6541 | ||
6542 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6543 | if (rc) | |
6544 | return rc; | |
6545 | ||
5f12733e MH |
6546 | if (write) { |
6547 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6548 | setup_per_zone_wmarks(); |
5f12733e | 6549 | } |
1da177e4 LT |
6550 | return 0; |
6551 | } | |
6552 | ||
795ae7a0 JW |
6553 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
6554 | void __user *buffer, size_t *length, loff_t *ppos) | |
6555 | { | |
6556 | int rc; | |
6557 | ||
6558 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6559 | if (rc) | |
6560 | return rc; | |
6561 | ||
6562 | if (write) | |
6563 | setup_per_zone_wmarks(); | |
6564 | ||
6565 | return 0; | |
6566 | } | |
6567 | ||
9614634f | 6568 | #ifdef CONFIG_NUMA |
cccad5b9 | 6569 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6570 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
6571 | { |
6572 | struct zone *zone; | |
6573 | int rc; | |
6574 | ||
8d65af78 | 6575 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
6576 | if (rc) |
6577 | return rc; | |
6578 | ||
6579 | for_each_zone(zone) | |
b40da049 | 6580 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
6581 | sysctl_min_unmapped_ratio) / 100; |
6582 | return 0; | |
6583 | } | |
0ff38490 | 6584 | |
cccad5b9 | 6585 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6586 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
6587 | { |
6588 | struct zone *zone; | |
6589 | int rc; | |
6590 | ||
8d65af78 | 6591 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
6592 | if (rc) |
6593 | return rc; | |
6594 | ||
6595 | for_each_zone(zone) | |
b40da049 | 6596 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
6597 | sysctl_min_slab_ratio) / 100; |
6598 | return 0; | |
6599 | } | |
9614634f CL |
6600 | #endif |
6601 | ||
1da177e4 LT |
6602 | /* |
6603 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
6604 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
6605 | * whenever sysctl_lowmem_reserve_ratio changes. | |
6606 | * | |
6607 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 6608 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
6609 | * if in function of the boot time zone sizes. |
6610 | */ | |
cccad5b9 | 6611 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6612 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6613 | { |
8d65af78 | 6614 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
6615 | setup_per_zone_lowmem_reserve(); |
6616 | return 0; | |
6617 | } | |
6618 | ||
8ad4b1fb RS |
6619 | /* |
6620 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
6621 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
6622 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 6623 | */ |
cccad5b9 | 6624 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6625 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
6626 | { |
6627 | struct zone *zone; | |
7cd2b0a3 | 6628 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
6629 | int ret; |
6630 | ||
7cd2b0a3 DR |
6631 | mutex_lock(&pcp_batch_high_lock); |
6632 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
6633 | ||
8d65af78 | 6634 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
6635 | if (!write || ret < 0) |
6636 | goto out; | |
6637 | ||
6638 | /* Sanity checking to avoid pcp imbalance */ | |
6639 | if (percpu_pagelist_fraction && | |
6640 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
6641 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
6642 | ret = -EINVAL; | |
6643 | goto out; | |
6644 | } | |
6645 | ||
6646 | /* No change? */ | |
6647 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
6648 | goto out; | |
c8e251fa | 6649 | |
364df0eb | 6650 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
6651 | unsigned int cpu; |
6652 | ||
22a7f12b | 6653 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
6654 | pageset_set_high_and_batch(zone, |
6655 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 6656 | } |
7cd2b0a3 | 6657 | out: |
c8e251fa | 6658 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6659 | return ret; |
8ad4b1fb RS |
6660 | } |
6661 | ||
a9919c79 | 6662 | #ifdef CONFIG_NUMA |
f034b5d4 | 6663 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 6664 | |
1da177e4 LT |
6665 | static int __init set_hashdist(char *str) |
6666 | { | |
6667 | if (!str) | |
6668 | return 0; | |
6669 | hashdist = simple_strtoul(str, &str, 0); | |
6670 | return 1; | |
6671 | } | |
6672 | __setup("hashdist=", set_hashdist); | |
6673 | #endif | |
6674 | ||
6675 | /* | |
6676 | * allocate a large system hash table from bootmem | |
6677 | * - it is assumed that the hash table must contain an exact power-of-2 | |
6678 | * quantity of entries | |
6679 | * - limit is the number of hash buckets, not the total allocation size | |
6680 | */ | |
6681 | void *__init alloc_large_system_hash(const char *tablename, | |
6682 | unsigned long bucketsize, | |
6683 | unsigned long numentries, | |
6684 | int scale, | |
6685 | int flags, | |
6686 | unsigned int *_hash_shift, | |
6687 | unsigned int *_hash_mask, | |
31fe62b9 TB |
6688 | unsigned long low_limit, |
6689 | unsigned long high_limit) | |
1da177e4 | 6690 | { |
31fe62b9 | 6691 | unsigned long long max = high_limit; |
1da177e4 LT |
6692 | unsigned long log2qty, size; |
6693 | void *table = NULL; | |
6694 | ||
6695 | /* allow the kernel cmdline to have a say */ | |
6696 | if (!numentries) { | |
6697 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 6698 | numentries = nr_kernel_pages; |
a7e83318 JZ |
6699 | |
6700 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
6701 | if (PAGE_SHIFT < 20) | |
6702 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
6703 | |
6704 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
6705 | if (scale > PAGE_SHIFT) | |
6706 | numentries >>= (scale - PAGE_SHIFT); | |
6707 | else | |
6708 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
6709 | |
6710 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
6711 | if (unlikely(flags & HASH_SMALL)) { |
6712 | /* Makes no sense without HASH_EARLY */ | |
6713 | WARN_ON(!(flags & HASH_EARLY)); | |
6714 | if (!(numentries >> *_hash_shift)) { | |
6715 | numentries = 1UL << *_hash_shift; | |
6716 | BUG_ON(!numentries); | |
6717 | } | |
6718 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 6719 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 6720 | } |
6e692ed3 | 6721 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
6722 | |
6723 | /* limit allocation size to 1/16 total memory by default */ | |
6724 | if (max == 0) { | |
6725 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
6726 | do_div(max, bucketsize); | |
6727 | } | |
074b8517 | 6728 | max = min(max, 0x80000000ULL); |
1da177e4 | 6729 | |
31fe62b9 TB |
6730 | if (numentries < low_limit) |
6731 | numentries = low_limit; | |
1da177e4 LT |
6732 | if (numentries > max) |
6733 | numentries = max; | |
6734 | ||
f0d1b0b3 | 6735 | log2qty = ilog2(numentries); |
1da177e4 LT |
6736 | |
6737 | do { | |
6738 | size = bucketsize << log2qty; | |
6739 | if (flags & HASH_EARLY) | |
6782832e | 6740 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
6741 | else if (hashdist) |
6742 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
6743 | else { | |
1037b83b ED |
6744 | /* |
6745 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
6746 | * some pages at the end of hash table which |
6747 | * alloc_pages_exact() automatically does | |
1037b83b | 6748 | */ |
264ef8a9 | 6749 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 6750 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
6751 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
6752 | } | |
1da177e4 LT |
6753 | } |
6754 | } while (!table && size > PAGE_SIZE && --log2qty); | |
6755 | ||
6756 | if (!table) | |
6757 | panic("Failed to allocate %s hash table\n", tablename); | |
6758 | ||
1170532b JP |
6759 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
6760 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
6761 | |
6762 | if (_hash_shift) | |
6763 | *_hash_shift = log2qty; | |
6764 | if (_hash_mask) | |
6765 | *_hash_mask = (1 << log2qty) - 1; | |
6766 | ||
6767 | return table; | |
6768 | } | |
a117e66e | 6769 | |
835c134e | 6770 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
f75fb889 | 6771 | static inline unsigned long *get_pageblock_bitmap(struct page *page, |
835c134e MG |
6772 | unsigned long pfn) |
6773 | { | |
6774 | #ifdef CONFIG_SPARSEMEM | |
6775 | return __pfn_to_section(pfn)->pageblock_flags; | |
6776 | #else | |
f75fb889 | 6777 | return page_zone(page)->pageblock_flags; |
835c134e MG |
6778 | #endif /* CONFIG_SPARSEMEM */ |
6779 | } | |
6780 | ||
f75fb889 | 6781 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) |
835c134e MG |
6782 | { |
6783 | #ifdef CONFIG_SPARSEMEM | |
6784 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 6785 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 6786 | #else |
f75fb889 | 6787 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 6788 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
6789 | #endif /* CONFIG_SPARSEMEM */ |
6790 | } | |
6791 | ||
6792 | /** | |
1aab4d77 | 6793 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e | 6794 | * @page: The page within the block of interest |
1aab4d77 RD |
6795 | * @pfn: The target page frame number |
6796 | * @end_bitidx: The last bit of interest to retrieve | |
6797 | * @mask: mask of bits that the caller is interested in | |
6798 | * | |
6799 | * Return: pageblock_bits flags | |
835c134e | 6800 | */ |
dc4b0caf | 6801 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
e58469ba MG |
6802 | unsigned long end_bitidx, |
6803 | unsigned long mask) | |
835c134e | 6804 | { |
835c134e | 6805 | unsigned long *bitmap; |
dc4b0caf | 6806 | unsigned long bitidx, word_bitidx; |
e58469ba | 6807 | unsigned long word; |
835c134e | 6808 | |
f75fb889 MG |
6809 | bitmap = get_pageblock_bitmap(page, pfn); |
6810 | bitidx = pfn_to_bitidx(page, pfn); | |
e58469ba MG |
6811 | word_bitidx = bitidx / BITS_PER_LONG; |
6812 | bitidx &= (BITS_PER_LONG-1); | |
835c134e | 6813 | |
e58469ba MG |
6814 | word = bitmap[word_bitidx]; |
6815 | bitidx += end_bitidx; | |
6816 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
835c134e MG |
6817 | } |
6818 | ||
6819 | /** | |
dc4b0caf | 6820 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e | 6821 | * @page: The page within the block of interest |
835c134e | 6822 | * @flags: The flags to set |
1aab4d77 RD |
6823 | * @pfn: The target page frame number |
6824 | * @end_bitidx: The last bit of interest | |
6825 | * @mask: mask of bits that the caller is interested in | |
835c134e | 6826 | */ |
dc4b0caf MG |
6827 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
6828 | unsigned long pfn, | |
e58469ba MG |
6829 | unsigned long end_bitidx, |
6830 | unsigned long mask) | |
835c134e | 6831 | { |
835c134e | 6832 | unsigned long *bitmap; |
dc4b0caf | 6833 | unsigned long bitidx, word_bitidx; |
e58469ba MG |
6834 | unsigned long old_word, word; |
6835 | ||
6836 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
835c134e | 6837 | |
f75fb889 MG |
6838 | bitmap = get_pageblock_bitmap(page, pfn); |
6839 | bitidx = pfn_to_bitidx(page, pfn); | |
e58469ba MG |
6840 | word_bitidx = bitidx / BITS_PER_LONG; |
6841 | bitidx &= (BITS_PER_LONG-1); | |
6842 | ||
f75fb889 | 6843 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); |
835c134e | 6844 | |
e58469ba MG |
6845 | bitidx += end_bitidx; |
6846 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
6847 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
6848 | ||
4db0c3c2 | 6849 | word = READ_ONCE(bitmap[word_bitidx]); |
e58469ba MG |
6850 | for (;;) { |
6851 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
6852 | if (word == old_word) | |
6853 | break; | |
6854 | word = old_word; | |
6855 | } | |
835c134e | 6856 | } |
a5d76b54 KH |
6857 | |
6858 | /* | |
80934513 MK |
6859 | * This function checks whether pageblock includes unmovable pages or not. |
6860 | * If @count is not zero, it is okay to include less @count unmovable pages | |
6861 | * | |
b8af2941 | 6862 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
6863 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
6864 | * expect this function should be exact. | |
a5d76b54 | 6865 | */ |
b023f468 WC |
6866 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
6867 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
6868 | { |
6869 | unsigned long pfn, iter, found; | |
47118af0 MN |
6870 | int mt; |
6871 | ||
49ac8255 KH |
6872 | /* |
6873 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 6874 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
6875 | */ |
6876 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 6877 | return false; |
47118af0 MN |
6878 | mt = get_pageblock_migratetype(page); |
6879 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 6880 | return false; |
49ac8255 KH |
6881 | |
6882 | pfn = page_to_pfn(page); | |
6883 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
6884 | unsigned long check = pfn + iter; | |
6885 | ||
29723fcc | 6886 | if (!pfn_valid_within(check)) |
49ac8255 | 6887 | continue; |
29723fcc | 6888 | |
49ac8255 | 6889 | page = pfn_to_page(check); |
c8721bbb NH |
6890 | |
6891 | /* | |
6892 | * Hugepages are not in LRU lists, but they're movable. | |
6893 | * We need not scan over tail pages bacause we don't | |
6894 | * handle each tail page individually in migration. | |
6895 | */ | |
6896 | if (PageHuge(page)) { | |
6897 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
6898 | continue; | |
6899 | } | |
6900 | ||
97d255c8 MK |
6901 | /* |
6902 | * We can't use page_count without pin a page | |
6903 | * because another CPU can free compound page. | |
6904 | * This check already skips compound tails of THP | |
0139aa7b | 6905 | * because their page->_refcount is zero at all time. |
97d255c8 | 6906 | */ |
fe896d18 | 6907 | if (!page_ref_count(page)) { |
49ac8255 KH |
6908 | if (PageBuddy(page)) |
6909 | iter += (1 << page_order(page)) - 1; | |
6910 | continue; | |
6911 | } | |
97d255c8 | 6912 | |
b023f468 WC |
6913 | /* |
6914 | * The HWPoisoned page may be not in buddy system, and | |
6915 | * page_count() is not 0. | |
6916 | */ | |
6917 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6918 | continue; | |
6919 | ||
49ac8255 KH |
6920 | if (!PageLRU(page)) |
6921 | found++; | |
6922 | /* | |
6b4f7799 JW |
6923 | * If there are RECLAIMABLE pages, we need to check |
6924 | * it. But now, memory offline itself doesn't call | |
6925 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
6926 | */ |
6927 | /* | |
6928 | * If the page is not RAM, page_count()should be 0. | |
6929 | * we don't need more check. This is an _used_ not-movable page. | |
6930 | * | |
6931 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6932 | * is set to both of a memory hole page and a _used_ kernel | |
6933 | * page at boot. | |
6934 | */ | |
6935 | if (found > count) | |
80934513 | 6936 | return true; |
49ac8255 | 6937 | } |
80934513 | 6938 | return false; |
49ac8255 KH |
6939 | } |
6940 | ||
6941 | bool is_pageblock_removable_nolock(struct page *page) | |
6942 | { | |
656a0706 MH |
6943 | struct zone *zone; |
6944 | unsigned long pfn; | |
687875fb MH |
6945 | |
6946 | /* | |
6947 | * We have to be careful here because we are iterating over memory | |
6948 | * sections which are not zone aware so we might end up outside of | |
6949 | * the zone but still within the section. | |
656a0706 MH |
6950 | * We have to take care about the node as well. If the node is offline |
6951 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 6952 | */ |
656a0706 MH |
6953 | if (!node_online(page_to_nid(page))) |
6954 | return false; | |
6955 | ||
6956 | zone = page_zone(page); | |
6957 | pfn = page_to_pfn(page); | |
108bcc96 | 6958 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
6959 | return false; |
6960 | ||
b023f468 | 6961 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6962 | } |
0c0e6195 | 6963 | |
080fe206 | 6964 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
6965 | |
6966 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6967 | { | |
6968 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6969 | pageblock_nr_pages) - 1); | |
6970 | } | |
6971 | ||
6972 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6973 | { | |
6974 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6975 | pageblock_nr_pages)); | |
6976 | } | |
6977 | ||
041d3a8c | 6978 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6979 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6980 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6981 | { |
6982 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6983 | unsigned long nr_reclaimed; |
041d3a8c MN |
6984 | unsigned long pfn = start; |
6985 | unsigned int tries = 0; | |
6986 | int ret = 0; | |
6987 | ||
be49a6e1 | 6988 | migrate_prep(); |
041d3a8c | 6989 | |
bb13ffeb | 6990 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6991 | if (fatal_signal_pending(current)) { |
6992 | ret = -EINTR; | |
6993 | break; | |
6994 | } | |
6995 | ||
bb13ffeb MG |
6996 | if (list_empty(&cc->migratepages)) { |
6997 | cc->nr_migratepages = 0; | |
edc2ca61 | 6998 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
6999 | if (!pfn) { |
7000 | ret = -EINTR; | |
7001 | break; | |
7002 | } | |
7003 | tries = 0; | |
7004 | } else if (++tries == 5) { | |
7005 | ret = ret < 0 ? ret : -EBUSY; | |
7006 | break; | |
7007 | } | |
7008 | ||
beb51eaa MK |
7009 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7010 | &cc->migratepages); | |
7011 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7012 | |
9c620e2b | 7013 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 7014 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 7015 | } |
2a6f5124 SP |
7016 | if (ret < 0) { |
7017 | putback_movable_pages(&cc->migratepages); | |
7018 | return ret; | |
7019 | } | |
7020 | return 0; | |
041d3a8c MN |
7021 | } |
7022 | ||
7023 | /** | |
7024 | * alloc_contig_range() -- tries to allocate given range of pages | |
7025 | * @start: start PFN to allocate | |
7026 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7027 | * @migratetype: migratetype of the underlaying pageblocks (either |
7028 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7029 | * in range must have the same migratetype and it must | |
7030 | * be either of the two. | |
041d3a8c MN |
7031 | * |
7032 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
7033 | * aligned, however it's the caller's responsibility to guarantee that | |
7034 | * we are the only thread that changes migrate type of pageblocks the | |
7035 | * pages fall in. | |
7036 | * | |
7037 | * The PFN range must belong to a single zone. | |
7038 | * | |
7039 | * Returns zero on success or negative error code. On success all | |
7040 | * pages which PFN is in [start, end) are allocated for the caller and | |
7041 | * need to be freed with free_contig_range(). | |
7042 | */ | |
0815f3d8 MN |
7043 | int alloc_contig_range(unsigned long start, unsigned long end, |
7044 | unsigned migratetype) | |
041d3a8c | 7045 | { |
041d3a8c | 7046 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7047 | unsigned int order; |
7048 | int ret = 0; | |
041d3a8c | 7049 | |
bb13ffeb MG |
7050 | struct compact_control cc = { |
7051 | .nr_migratepages = 0, | |
7052 | .order = -1, | |
7053 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7054 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
7055 | .ignore_skip_hint = true, |
7056 | }; | |
7057 | INIT_LIST_HEAD(&cc.migratepages); | |
7058 | ||
041d3a8c MN |
7059 | /* |
7060 | * What we do here is we mark all pageblocks in range as | |
7061 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7062 | * have different sizes, and due to the way page allocator | |
7063 | * work, we align the range to biggest of the two pages so | |
7064 | * that page allocator won't try to merge buddies from | |
7065 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7066 | * other migration type. | |
7067 | * | |
7068 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7069 | * migrate the pages from an unaligned range (ie. pages that | |
7070 | * we are interested in). This will put all the pages in | |
7071 | * range back to page allocator as MIGRATE_ISOLATE. | |
7072 | * | |
7073 | * When this is done, we take the pages in range from page | |
7074 | * allocator removing them from the buddy system. This way | |
7075 | * page allocator will never consider using them. | |
7076 | * | |
7077 | * This lets us mark the pageblocks back as | |
7078 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7079 | * aligned range but not in the unaligned, original range are | |
7080 | * put back to page allocator so that buddy can use them. | |
7081 | */ | |
7082 | ||
7083 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7084 | pfn_max_align_up(end), migratetype, |
7085 | false); | |
041d3a8c | 7086 | if (ret) |
86a595f9 | 7087 | return ret; |
041d3a8c | 7088 | |
8ef5849f JK |
7089 | /* |
7090 | * In case of -EBUSY, we'd like to know which page causes problem. | |
7091 | * So, just fall through. We will check it in test_pages_isolated(). | |
7092 | */ | |
bb13ffeb | 7093 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7094 | if (ret && ret != -EBUSY) |
041d3a8c MN |
7095 | goto done; |
7096 | ||
7097 | /* | |
7098 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7099 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7100 | * more, all pages in [start, end) are free in page allocator. | |
7101 | * What we are going to do is to allocate all pages from | |
7102 | * [start, end) (that is remove them from page allocator). | |
7103 | * | |
7104 | * The only problem is that pages at the beginning and at the | |
7105 | * end of interesting range may be not aligned with pages that | |
7106 | * page allocator holds, ie. they can be part of higher order | |
7107 | * pages. Because of this, we reserve the bigger range and | |
7108 | * once this is done free the pages we are not interested in. | |
7109 | * | |
7110 | * We don't have to hold zone->lock here because the pages are | |
7111 | * isolated thus they won't get removed from buddy. | |
7112 | */ | |
7113 | ||
7114 | lru_add_drain_all(); | |
510f5507 | 7115 | drain_all_pages(cc.zone); |
041d3a8c MN |
7116 | |
7117 | order = 0; | |
7118 | outer_start = start; | |
7119 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7120 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7121 | outer_start = start; |
7122 | break; | |
041d3a8c MN |
7123 | } |
7124 | outer_start &= ~0UL << order; | |
7125 | } | |
7126 | ||
8ef5849f JK |
7127 | if (outer_start != start) { |
7128 | order = page_order(pfn_to_page(outer_start)); | |
7129 | ||
7130 | /* | |
7131 | * outer_start page could be small order buddy page and | |
7132 | * it doesn't include start page. Adjust outer_start | |
7133 | * in this case to report failed page properly | |
7134 | * on tracepoint in test_pages_isolated() | |
7135 | */ | |
7136 | if (outer_start + (1UL << order) <= start) | |
7137 | outer_start = start; | |
7138 | } | |
7139 | ||
041d3a8c | 7140 | /* Make sure the range is really isolated. */ |
b023f468 | 7141 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
7142 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
7143 | __func__, outer_start, end); | |
041d3a8c MN |
7144 | ret = -EBUSY; |
7145 | goto done; | |
7146 | } | |
7147 | ||
49f223a9 | 7148 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7149 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7150 | if (!outer_end) { |
7151 | ret = -EBUSY; | |
7152 | goto done; | |
7153 | } | |
7154 | ||
7155 | /* Free head and tail (if any) */ | |
7156 | if (start != outer_start) | |
7157 | free_contig_range(outer_start, start - outer_start); | |
7158 | if (end != outer_end) | |
7159 | free_contig_range(end, outer_end - end); | |
7160 | ||
7161 | done: | |
7162 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7163 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7164 | return ret; |
7165 | } | |
7166 | ||
7167 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7168 | { | |
bcc2b02f MS |
7169 | unsigned int count = 0; |
7170 | ||
7171 | for (; nr_pages--; pfn++) { | |
7172 | struct page *page = pfn_to_page(pfn); | |
7173 | ||
7174 | count += page_count(page) != 1; | |
7175 | __free_page(page); | |
7176 | } | |
7177 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7178 | } |
7179 | #endif | |
7180 | ||
4ed7e022 | 7181 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7182 | /* |
7183 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7184 | * page high values need to be recalulated. | |
7185 | */ | |
4ed7e022 JL |
7186 | void __meminit zone_pcp_update(struct zone *zone) |
7187 | { | |
0a647f38 | 7188 | unsigned cpu; |
c8e251fa | 7189 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7190 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7191 | pageset_set_high_and_batch(zone, |
7192 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7193 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7194 | } |
7195 | #endif | |
7196 | ||
340175b7 JL |
7197 | void zone_pcp_reset(struct zone *zone) |
7198 | { | |
7199 | unsigned long flags; | |
5a883813 MK |
7200 | int cpu; |
7201 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7202 | |
7203 | /* avoid races with drain_pages() */ | |
7204 | local_irq_save(flags); | |
7205 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7206 | for_each_online_cpu(cpu) { |
7207 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7208 | drain_zonestat(zone, pset); | |
7209 | } | |
340175b7 JL |
7210 | free_percpu(zone->pageset); |
7211 | zone->pageset = &boot_pageset; | |
7212 | } | |
7213 | local_irq_restore(flags); | |
7214 | } | |
7215 | ||
6dcd73d7 | 7216 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 7217 | /* |
b9eb6319 JK |
7218 | * All pages in the range must be in a single zone and isolated |
7219 | * before calling this. | |
0c0e6195 KH |
7220 | */ |
7221 | void | |
7222 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7223 | { | |
7224 | struct page *page; | |
7225 | struct zone *zone; | |
7aeb09f9 | 7226 | unsigned int order, i; |
0c0e6195 KH |
7227 | unsigned long pfn; |
7228 | unsigned long flags; | |
7229 | /* find the first valid pfn */ | |
7230 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
7231 | if (pfn_valid(pfn)) | |
7232 | break; | |
7233 | if (pfn == end_pfn) | |
7234 | return; | |
7235 | zone = page_zone(pfn_to_page(pfn)); | |
7236 | spin_lock_irqsave(&zone->lock, flags); | |
7237 | pfn = start_pfn; | |
7238 | while (pfn < end_pfn) { | |
7239 | if (!pfn_valid(pfn)) { | |
7240 | pfn++; | |
7241 | continue; | |
7242 | } | |
7243 | page = pfn_to_page(pfn); | |
b023f468 WC |
7244 | /* |
7245 | * The HWPoisoned page may be not in buddy system, and | |
7246 | * page_count() is not 0. | |
7247 | */ | |
7248 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
7249 | pfn++; | |
7250 | SetPageReserved(page); | |
7251 | continue; | |
7252 | } | |
7253 | ||
0c0e6195 KH |
7254 | BUG_ON(page_count(page)); |
7255 | BUG_ON(!PageBuddy(page)); | |
7256 | order = page_order(page); | |
7257 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
7258 | pr_info("remove from free list %lx %d %lx\n", |
7259 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
7260 | #endif |
7261 | list_del(&page->lru); | |
7262 | rmv_page_order(page); | |
7263 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
7264 | for (i = 0; i < (1 << order); i++) |
7265 | SetPageReserved((page+i)); | |
7266 | pfn += (1 << order); | |
7267 | } | |
7268 | spin_unlock_irqrestore(&zone->lock, flags); | |
7269 | } | |
7270 | #endif | |
8d22ba1b | 7271 | |
8d22ba1b WF |
7272 | bool is_free_buddy_page(struct page *page) |
7273 | { | |
7274 | struct zone *zone = page_zone(page); | |
7275 | unsigned long pfn = page_to_pfn(page); | |
7276 | unsigned long flags; | |
7aeb09f9 | 7277 | unsigned int order; |
8d22ba1b WF |
7278 | |
7279 | spin_lock_irqsave(&zone->lock, flags); | |
7280 | for (order = 0; order < MAX_ORDER; order++) { | |
7281 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
7282 | ||
7283 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
7284 | break; | |
7285 | } | |
7286 | spin_unlock_irqrestore(&zone->lock, flags); | |
7287 | ||
7288 | return order < MAX_ORDER; | |
7289 | } |