]>
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 | ||
7bfec6f4 MG |
787 | /* |
788 | * A bad page could be due to a number of fields. Instead of multiple branches, | |
789 | * try and check multiple fields with one check. The caller must do a detailed | |
790 | * check if necessary. | |
791 | */ | |
792 | static inline bool page_expected_state(struct page *page, | |
793 | unsigned long check_flags) | |
794 | { | |
795 | if (unlikely(atomic_read(&page->_mapcount) != -1)) | |
796 | return false; | |
797 | ||
798 | if (unlikely((unsigned long)page->mapping | | |
799 | page_ref_count(page) | | |
800 | #ifdef CONFIG_MEMCG | |
801 | (unsigned long)page->mem_cgroup | | |
802 | #endif | |
803 | (page->flags & check_flags))) | |
804 | return false; | |
805 | ||
806 | return true; | |
807 | } | |
808 | ||
bb552ac6 | 809 | static void free_pages_check_bad(struct page *page) |
1da177e4 | 810 | { |
7bfec6f4 MG |
811 | const char *bad_reason; |
812 | unsigned long bad_flags; | |
813 | ||
7bfec6f4 MG |
814 | bad_reason = NULL; |
815 | bad_flags = 0; | |
f0b791a3 | 816 | |
53f9263b | 817 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
818 | bad_reason = "nonzero mapcount"; |
819 | if (unlikely(page->mapping != NULL)) | |
820 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 821 | if (unlikely(page_ref_count(page) != 0)) |
0139aa7b | 822 | bad_reason = "nonzero _refcount"; |
f0b791a3 DH |
823 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
824 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
825 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
826 | } | |
9edad6ea JW |
827 | #ifdef CONFIG_MEMCG |
828 | if (unlikely(page->mem_cgroup)) | |
829 | bad_reason = "page still charged to cgroup"; | |
830 | #endif | |
7bfec6f4 | 831 | bad_page(page, bad_reason, bad_flags); |
bb552ac6 MG |
832 | } |
833 | ||
834 | static inline int free_pages_check(struct page *page) | |
835 | { | |
836 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) { | |
837 | page_cpupid_reset_last(page); | |
838 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
839 | return 0; | |
840 | } | |
841 | ||
842 | /* Something has gone sideways, find it */ | |
843 | free_pages_check_bad(page); | |
7bfec6f4 | 844 | return 1; |
1da177e4 LT |
845 | } |
846 | ||
847 | /* | |
5f8dcc21 | 848 | * Frees a number of pages from the PCP lists |
1da177e4 | 849 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 850 | * count is the number of pages to free. |
1da177e4 LT |
851 | * |
852 | * If the zone was previously in an "all pages pinned" state then look to | |
853 | * see if this freeing clears that state. | |
854 | * | |
855 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
856 | * pinned" detection logic. | |
857 | */ | |
5f8dcc21 MG |
858 | static void free_pcppages_bulk(struct zone *zone, int count, |
859 | struct per_cpu_pages *pcp) | |
1da177e4 | 860 | { |
5f8dcc21 | 861 | int migratetype = 0; |
a6f9edd6 | 862 | int batch_free = 0; |
72853e29 | 863 | int to_free = count; |
0d5d823a | 864 | unsigned long nr_scanned; |
3777999d | 865 | bool isolated_pageblocks; |
5f8dcc21 | 866 | |
c54ad30c | 867 | spin_lock(&zone->lock); |
3777999d | 868 | isolated_pageblocks = has_isolate_pageblock(zone); |
0d5d823a MG |
869 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
870 | if (nr_scanned) | |
871 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 872 | |
72853e29 | 873 | while (to_free) { |
48db57f8 | 874 | struct page *page; |
5f8dcc21 MG |
875 | struct list_head *list; |
876 | ||
877 | /* | |
a6f9edd6 MG |
878 | * Remove pages from lists in a round-robin fashion. A |
879 | * batch_free count is maintained that is incremented when an | |
880 | * empty list is encountered. This is so more pages are freed | |
881 | * off fuller lists instead of spinning excessively around empty | |
882 | * lists | |
5f8dcc21 MG |
883 | */ |
884 | do { | |
a6f9edd6 | 885 | batch_free++; |
5f8dcc21 MG |
886 | if (++migratetype == MIGRATE_PCPTYPES) |
887 | migratetype = 0; | |
888 | list = &pcp->lists[migratetype]; | |
889 | } while (list_empty(list)); | |
48db57f8 | 890 | |
1d16871d NK |
891 | /* This is the only non-empty list. Free them all. */ |
892 | if (batch_free == MIGRATE_PCPTYPES) | |
893 | batch_free = to_free; | |
894 | ||
a6f9edd6 | 895 | do { |
770c8aaa BZ |
896 | int mt; /* migratetype of the to-be-freed page */ |
897 | ||
a16601c5 | 898 | page = list_last_entry(list, struct page, lru); |
a6f9edd6 MG |
899 | /* must delete as __free_one_page list manipulates */ |
900 | list_del(&page->lru); | |
aa016d14 | 901 | |
bb14c2c7 | 902 | mt = get_pcppage_migratetype(page); |
aa016d14 VB |
903 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
904 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
905 | /* Pageblock could have been isolated meanwhile */ | |
3777999d | 906 | if (unlikely(isolated_pageblocks)) |
51bb1a40 | 907 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 908 | |
dc4b0caf | 909 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 910 | trace_mm_page_pcpu_drain(page, 0, mt); |
72853e29 | 911 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 912 | } |
c54ad30c | 913 | spin_unlock(&zone->lock); |
1da177e4 LT |
914 | } |
915 | ||
dc4b0caf MG |
916 | static void free_one_page(struct zone *zone, |
917 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 918 | unsigned int order, |
ed0ae21d | 919 | int migratetype) |
1da177e4 | 920 | { |
0d5d823a | 921 | unsigned long nr_scanned; |
006d22d9 | 922 | spin_lock(&zone->lock); |
0d5d823a MG |
923 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
924 | if (nr_scanned) | |
925 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 926 | |
ad53f92e JK |
927 | if (unlikely(has_isolate_pageblock(zone) || |
928 | is_migrate_isolate(migratetype))) { | |
929 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 930 | } |
dc4b0caf | 931 | __free_one_page(page, pfn, zone, order, migratetype); |
006d22d9 | 932 | spin_unlock(&zone->lock); |
48db57f8 NP |
933 | } |
934 | ||
81422f29 KS |
935 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
936 | { | |
1d798ca3 KS |
937 | int ret = 1; |
938 | ||
939 | /* | |
940 | * We rely page->lru.next never has bit 0 set, unless the page | |
941 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
942 | */ | |
943 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
944 | ||
945 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
946 | ret = 0; | |
947 | goto out; | |
948 | } | |
9a982250 KS |
949 | switch (page - head_page) { |
950 | case 1: | |
951 | /* the first tail page: ->mapping is compound_mapcount() */ | |
53f9263b KS |
952 | if (unlikely(compound_mapcount(page))) { |
953 | bad_page(page, "nonzero compound_mapcount", 0); | |
954 | goto out; | |
955 | } | |
9a982250 KS |
956 | break; |
957 | case 2: | |
958 | /* | |
959 | * the second tail page: ->mapping is | |
960 | * page_deferred_list().next -- ignore value. | |
961 | */ | |
962 | break; | |
963 | default: | |
964 | if (page->mapping != TAIL_MAPPING) { | |
965 | bad_page(page, "corrupted mapping in tail page", 0); | |
966 | goto out; | |
967 | } | |
968 | break; | |
1c290f64 | 969 | } |
81422f29 KS |
970 | if (unlikely(!PageTail(page))) { |
971 | bad_page(page, "PageTail not set", 0); | |
1d798ca3 | 972 | goto out; |
81422f29 | 973 | } |
1d798ca3 KS |
974 | if (unlikely(compound_head(page) != head_page)) { |
975 | bad_page(page, "compound_head not consistent", 0); | |
976 | goto out; | |
81422f29 | 977 | } |
1d798ca3 KS |
978 | ret = 0; |
979 | out: | |
1c290f64 | 980 | page->mapping = NULL; |
1d798ca3 KS |
981 | clear_compound_head(page); |
982 | return ret; | |
81422f29 KS |
983 | } |
984 | ||
1e8ce83c RH |
985 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
986 | unsigned long zone, int nid) | |
987 | { | |
1e8ce83c | 988 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
989 | init_page_count(page); |
990 | page_mapcount_reset(page); | |
991 | page_cpupid_reset_last(page); | |
1e8ce83c | 992 | |
1e8ce83c RH |
993 | INIT_LIST_HEAD(&page->lru); |
994 | #ifdef WANT_PAGE_VIRTUAL | |
995 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
996 | if (!is_highmem_idx(zone)) | |
997 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
998 | #endif | |
999 | } | |
1000 | ||
1001 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
1002 | int nid) | |
1003 | { | |
1004 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
1005 | } | |
1006 | ||
7e18adb4 MG |
1007 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1008 | static void init_reserved_page(unsigned long pfn) | |
1009 | { | |
1010 | pg_data_t *pgdat; | |
1011 | int nid, zid; | |
1012 | ||
1013 | if (!early_page_uninitialised(pfn)) | |
1014 | return; | |
1015 | ||
1016 | nid = early_pfn_to_nid(pfn); | |
1017 | pgdat = NODE_DATA(nid); | |
1018 | ||
1019 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1020 | struct zone *zone = &pgdat->node_zones[zid]; | |
1021 | ||
1022 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
1023 | break; | |
1024 | } | |
1025 | __init_single_pfn(pfn, zid, nid); | |
1026 | } | |
1027 | #else | |
1028 | static inline void init_reserved_page(unsigned long pfn) | |
1029 | { | |
1030 | } | |
1031 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
1032 | ||
92923ca3 NZ |
1033 | /* |
1034 | * Initialised pages do not have PageReserved set. This function is | |
1035 | * called for each range allocated by the bootmem allocator and | |
1036 | * marks the pages PageReserved. The remaining valid pages are later | |
1037 | * sent to the buddy page allocator. | |
1038 | */ | |
7e18adb4 | 1039 | void __meminit reserve_bootmem_region(unsigned long start, unsigned long end) |
92923ca3 NZ |
1040 | { |
1041 | unsigned long start_pfn = PFN_DOWN(start); | |
1042 | unsigned long end_pfn = PFN_UP(end); | |
1043 | ||
7e18adb4 MG |
1044 | for (; start_pfn < end_pfn; start_pfn++) { |
1045 | if (pfn_valid(start_pfn)) { | |
1046 | struct page *page = pfn_to_page(start_pfn); | |
1047 | ||
1048 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
1049 | |
1050 | /* Avoid false-positive PageTail() */ | |
1051 | INIT_LIST_HEAD(&page->lru); | |
1052 | ||
7e18adb4 MG |
1053 | SetPageReserved(page); |
1054 | } | |
1055 | } | |
92923ca3 NZ |
1056 | } |
1057 | ||
ec95f53a | 1058 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 1059 | { |
d61f8590 | 1060 | int bad = 0; |
1da177e4 | 1061 | |
ab1f306f | 1062 | VM_BUG_ON_PAGE(PageTail(page), page); |
ab1f306f | 1063 | |
b413d48a | 1064 | trace_mm_page_free(page, order); |
b1eeab67 | 1065 | kmemcheck_free_shadow(page, order); |
b8c73fc2 | 1066 | kasan_free_pages(page, order); |
b1eeab67 | 1067 | |
d61f8590 MG |
1068 | /* |
1069 | * Check tail pages before head page information is cleared to | |
1070 | * avoid checking PageCompound for order-0 pages. | |
1071 | */ | |
1072 | if (unlikely(order)) { | |
1073 | bool compound = PageCompound(page); | |
1074 | int i; | |
1075 | ||
1076 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); | |
1077 | ||
1078 | for (i = 1; i < (1 << order); i++) { | |
1079 | if (compound) | |
1080 | bad += free_tail_pages_check(page, page + i); | |
1081 | bad += free_pages_check(page + i); | |
1082 | } | |
1083 | } | |
17514574 | 1084 | if (PageAnonHead(page)) |
8dd60a3a | 1085 | page->mapping = NULL; |
81422f29 | 1086 | bad += free_pages_check(page); |
8cc3b392 | 1087 | if (bad) |
ec95f53a | 1088 | return false; |
689bcebf | 1089 | |
48c96a36 JK |
1090 | reset_page_owner(page, order); |
1091 | ||
3ac7fe5a | 1092 | if (!PageHighMem(page)) { |
b8af2941 PK |
1093 | debug_check_no_locks_freed(page_address(page), |
1094 | PAGE_SIZE << order); | |
3ac7fe5a TG |
1095 | debug_check_no_obj_freed(page_address(page), |
1096 | PAGE_SIZE << order); | |
1097 | } | |
dafb1367 | 1098 | arch_free_page(page, order); |
8823b1db | 1099 | kernel_poison_pages(page, 1 << order, 0); |
48db57f8 | 1100 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 1101 | |
ec95f53a KM |
1102 | return true; |
1103 | } | |
1104 | ||
1105 | static void __free_pages_ok(struct page *page, unsigned int order) | |
1106 | { | |
1107 | unsigned long flags; | |
95e34412 | 1108 | int migratetype; |
dc4b0caf | 1109 | unsigned long pfn = page_to_pfn(page); |
ec95f53a KM |
1110 | |
1111 | if (!free_pages_prepare(page, order)) | |
1112 | return; | |
1113 | ||
cfc47a28 | 1114 | migratetype = get_pfnblock_migratetype(page, pfn); |
c54ad30c | 1115 | local_irq_save(flags); |
f8891e5e | 1116 | __count_vm_events(PGFREE, 1 << order); |
dc4b0caf | 1117 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
c54ad30c | 1118 | local_irq_restore(flags); |
1da177e4 LT |
1119 | } |
1120 | ||
949698a3 | 1121 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
a226f6c8 | 1122 | { |
c3993076 | 1123 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1124 | struct page *p = page; |
c3993076 | 1125 | unsigned int loop; |
a226f6c8 | 1126 | |
e2d0bd2b YL |
1127 | prefetchw(p); |
1128 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1129 | prefetchw(p + 1); | |
c3993076 JW |
1130 | __ClearPageReserved(p); |
1131 | set_page_count(p, 0); | |
a226f6c8 | 1132 | } |
e2d0bd2b YL |
1133 | __ClearPageReserved(p); |
1134 | set_page_count(p, 0); | |
c3993076 | 1135 | |
e2d0bd2b | 1136 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1137 | set_page_refcounted(page); |
1138 | __free_pages(page, order); | |
a226f6c8 DH |
1139 | } |
1140 | ||
75a592a4 MG |
1141 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1142 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1143 | |
75a592a4 MG |
1144 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1145 | ||
1146 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1147 | { | |
7ace9917 | 1148 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1149 | int nid; |
1150 | ||
7ace9917 | 1151 | spin_lock(&early_pfn_lock); |
75a592a4 | 1152 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 MG |
1153 | if (nid < 0) |
1154 | nid = 0; | |
1155 | spin_unlock(&early_pfn_lock); | |
1156 | ||
1157 | return nid; | |
75a592a4 MG |
1158 | } |
1159 | #endif | |
1160 | ||
1161 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
1162 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1163 | struct mminit_pfnnid_cache *state) | |
1164 | { | |
1165 | int nid; | |
1166 | ||
1167 | nid = __early_pfn_to_nid(pfn, state); | |
1168 | if (nid >= 0 && nid != node) | |
1169 | return false; | |
1170 | return true; | |
1171 | } | |
1172 | ||
1173 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1174 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1175 | { | |
1176 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1177 | } | |
1178 | ||
1179 | #else | |
1180 | ||
1181 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1182 | { | |
1183 | return true; | |
1184 | } | |
1185 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1186 | struct mminit_pfnnid_cache *state) | |
1187 | { | |
1188 | return true; | |
1189 | } | |
1190 | #endif | |
1191 | ||
1192 | ||
0e1cc95b | 1193 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1194 | unsigned int order) |
1195 | { | |
1196 | if (early_page_uninitialised(pfn)) | |
1197 | return; | |
949698a3 | 1198 | return __free_pages_boot_core(page, order); |
3a80a7fa MG |
1199 | } |
1200 | ||
7cf91a98 JK |
1201 | /* |
1202 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1203 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1204 | * with the migration of free compaction scanner. The scanners then need to | |
1205 | * use only pfn_valid_within() check for arches that allow holes within | |
1206 | * pageblocks. | |
1207 | * | |
1208 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1209 | * | |
1210 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1211 | * i.e. it's possible that all pages within a zones range of pages do not | |
1212 | * belong to a single zone. We assume that a border between node0 and node1 | |
1213 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1214 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1215 | * the first and last page of a pageblock and avoid checking each individual | |
1216 | * page in a pageblock. | |
1217 | */ | |
1218 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1219 | unsigned long end_pfn, struct zone *zone) | |
1220 | { | |
1221 | struct page *start_page; | |
1222 | struct page *end_page; | |
1223 | ||
1224 | /* end_pfn is one past the range we are checking */ | |
1225 | end_pfn--; | |
1226 | ||
1227 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1228 | return NULL; | |
1229 | ||
1230 | start_page = pfn_to_page(start_pfn); | |
1231 | ||
1232 | if (page_zone(start_page) != zone) | |
1233 | return NULL; | |
1234 | ||
1235 | end_page = pfn_to_page(end_pfn); | |
1236 | ||
1237 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1238 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1239 | return NULL; | |
1240 | ||
1241 | return start_page; | |
1242 | } | |
1243 | ||
1244 | void set_zone_contiguous(struct zone *zone) | |
1245 | { | |
1246 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1247 | unsigned long block_end_pfn; | |
1248 | ||
1249 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1250 | for (; block_start_pfn < zone_end_pfn(zone); | |
1251 | block_start_pfn = block_end_pfn, | |
1252 | block_end_pfn += pageblock_nr_pages) { | |
1253 | ||
1254 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1255 | ||
1256 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1257 | block_end_pfn, zone)) | |
1258 | return; | |
1259 | } | |
1260 | ||
1261 | /* We confirm that there is no hole */ | |
1262 | zone->contiguous = true; | |
1263 | } | |
1264 | ||
1265 | void clear_zone_contiguous(struct zone *zone) | |
1266 | { | |
1267 | zone->contiguous = false; | |
1268 | } | |
1269 | ||
7e18adb4 | 1270 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
0e1cc95b | 1271 | static void __init deferred_free_range(struct page *page, |
a4de83dd MG |
1272 | unsigned long pfn, int nr_pages) |
1273 | { | |
1274 | int i; | |
1275 | ||
1276 | if (!page) | |
1277 | return; | |
1278 | ||
1279 | /* Free a large naturally-aligned chunk if possible */ | |
1280 | if (nr_pages == MAX_ORDER_NR_PAGES && | |
1281 | (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) { | |
ac5d2539 | 1282 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
949698a3 | 1283 | __free_pages_boot_core(page, MAX_ORDER-1); |
a4de83dd MG |
1284 | return; |
1285 | } | |
1286 | ||
949698a3 LZ |
1287 | for (i = 0; i < nr_pages; i++, page++) |
1288 | __free_pages_boot_core(page, 0); | |
a4de83dd MG |
1289 | } |
1290 | ||
d3cd131d NS |
1291 | /* Completion tracking for deferred_init_memmap() threads */ |
1292 | static atomic_t pgdat_init_n_undone __initdata; | |
1293 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1294 | ||
1295 | static inline void __init pgdat_init_report_one_done(void) | |
1296 | { | |
1297 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1298 | complete(&pgdat_init_all_done_comp); | |
1299 | } | |
0e1cc95b | 1300 | |
7e18adb4 | 1301 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1302 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1303 | { |
0e1cc95b MG |
1304 | pg_data_t *pgdat = data; |
1305 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1306 | struct mminit_pfnnid_cache nid_init_state = { }; |
1307 | unsigned long start = jiffies; | |
1308 | unsigned long nr_pages = 0; | |
1309 | unsigned long walk_start, walk_end; | |
1310 | int i, zid; | |
1311 | struct zone *zone; | |
7e18adb4 | 1312 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
0e1cc95b | 1313 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
7e18adb4 | 1314 | |
0e1cc95b | 1315 | if (first_init_pfn == ULONG_MAX) { |
d3cd131d | 1316 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1317 | return 0; |
1318 | } | |
1319 | ||
1320 | /* Bind memory initialisation thread to a local node if possible */ | |
1321 | if (!cpumask_empty(cpumask)) | |
1322 | set_cpus_allowed_ptr(current, cpumask); | |
7e18adb4 MG |
1323 | |
1324 | /* Sanity check boundaries */ | |
1325 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1326 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1327 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1328 | ||
1329 | /* Only the highest zone is deferred so find it */ | |
1330 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1331 | zone = pgdat->node_zones + zid; | |
1332 | if (first_init_pfn < zone_end_pfn(zone)) | |
1333 | break; | |
1334 | } | |
1335 | ||
1336 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { | |
1337 | unsigned long pfn, end_pfn; | |
54608c3f | 1338 | struct page *page = NULL; |
a4de83dd MG |
1339 | struct page *free_base_page = NULL; |
1340 | unsigned long free_base_pfn = 0; | |
1341 | int nr_to_free = 0; | |
7e18adb4 MG |
1342 | |
1343 | end_pfn = min(walk_end, zone_end_pfn(zone)); | |
1344 | pfn = first_init_pfn; | |
1345 | if (pfn < walk_start) | |
1346 | pfn = walk_start; | |
1347 | if (pfn < zone->zone_start_pfn) | |
1348 | pfn = zone->zone_start_pfn; | |
1349 | ||
1350 | for (; pfn < end_pfn; pfn++) { | |
54608c3f | 1351 | if (!pfn_valid_within(pfn)) |
a4de83dd | 1352 | goto free_range; |
7e18adb4 | 1353 | |
54608c3f MG |
1354 | /* |
1355 | * Ensure pfn_valid is checked every | |
1356 | * MAX_ORDER_NR_PAGES for memory holes | |
1357 | */ | |
1358 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
1359 | if (!pfn_valid(pfn)) { | |
1360 | page = NULL; | |
a4de83dd | 1361 | goto free_range; |
54608c3f MG |
1362 | } |
1363 | } | |
1364 | ||
1365 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { | |
1366 | page = NULL; | |
a4de83dd | 1367 | goto free_range; |
54608c3f MG |
1368 | } |
1369 | ||
1370 | /* Minimise pfn page lookups and scheduler checks */ | |
1371 | if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) { | |
1372 | page++; | |
1373 | } else { | |
a4de83dd MG |
1374 | nr_pages += nr_to_free; |
1375 | deferred_free_range(free_base_page, | |
1376 | free_base_pfn, nr_to_free); | |
1377 | free_base_page = NULL; | |
1378 | free_base_pfn = nr_to_free = 0; | |
1379 | ||
54608c3f MG |
1380 | page = pfn_to_page(pfn); |
1381 | cond_resched(); | |
1382 | } | |
7e18adb4 MG |
1383 | |
1384 | if (page->flags) { | |
1385 | VM_BUG_ON(page_zone(page) != zone); | |
a4de83dd | 1386 | goto free_range; |
7e18adb4 MG |
1387 | } |
1388 | ||
1389 | __init_single_page(page, pfn, zid, nid); | |
a4de83dd MG |
1390 | if (!free_base_page) { |
1391 | free_base_page = page; | |
1392 | free_base_pfn = pfn; | |
1393 | nr_to_free = 0; | |
1394 | } | |
1395 | nr_to_free++; | |
1396 | ||
1397 | /* Where possible, batch up pages for a single free */ | |
1398 | continue; | |
1399 | free_range: | |
1400 | /* Free the current block of pages to allocator */ | |
1401 | nr_pages += nr_to_free; | |
1402 | deferred_free_range(free_base_page, free_base_pfn, | |
1403 | nr_to_free); | |
1404 | free_base_page = NULL; | |
1405 | free_base_pfn = nr_to_free = 0; | |
7e18adb4 | 1406 | } |
a4de83dd | 1407 | |
7e18adb4 MG |
1408 | first_init_pfn = max(end_pfn, first_init_pfn); |
1409 | } | |
1410 | ||
1411 | /* Sanity check that the next zone really is unpopulated */ | |
1412 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1413 | ||
0e1cc95b | 1414 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1415 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1416 | |
1417 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1418 | return 0; |
1419 | } | |
7cf91a98 | 1420 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1421 | |
1422 | void __init page_alloc_init_late(void) | |
1423 | { | |
7cf91a98 JK |
1424 | struct zone *zone; |
1425 | ||
1426 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1427 | int nid; |
1428 | ||
d3cd131d NS |
1429 | /* There will be num_node_state(N_MEMORY) threads */ |
1430 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1431 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1432 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1433 | } | |
1434 | ||
1435 | /* Block until all are initialised */ | |
d3cd131d | 1436 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da MG |
1437 | |
1438 | /* Reinit limits that are based on free pages after the kernel is up */ | |
1439 | files_maxfiles_init(); | |
7cf91a98 JK |
1440 | #endif |
1441 | ||
1442 | for_each_populated_zone(zone) | |
1443 | set_zone_contiguous(zone); | |
7e18adb4 | 1444 | } |
7e18adb4 | 1445 | |
47118af0 | 1446 | #ifdef CONFIG_CMA |
9cf510a5 | 1447 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1448 | void __init init_cma_reserved_pageblock(struct page *page) |
1449 | { | |
1450 | unsigned i = pageblock_nr_pages; | |
1451 | struct page *p = page; | |
1452 | ||
1453 | do { | |
1454 | __ClearPageReserved(p); | |
1455 | set_page_count(p, 0); | |
1456 | } while (++p, --i); | |
1457 | ||
47118af0 | 1458 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1459 | |
1460 | if (pageblock_order >= MAX_ORDER) { | |
1461 | i = pageblock_nr_pages; | |
1462 | p = page; | |
1463 | do { | |
1464 | set_page_refcounted(p); | |
1465 | __free_pages(p, MAX_ORDER - 1); | |
1466 | p += MAX_ORDER_NR_PAGES; | |
1467 | } while (i -= MAX_ORDER_NR_PAGES); | |
1468 | } else { | |
1469 | set_page_refcounted(page); | |
1470 | __free_pages(page, pageblock_order); | |
1471 | } | |
1472 | ||
3dcc0571 | 1473 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1474 | } |
1475 | #endif | |
1da177e4 LT |
1476 | |
1477 | /* | |
1478 | * The order of subdivision here is critical for the IO subsystem. | |
1479 | * Please do not alter this order without good reasons and regression | |
1480 | * testing. Specifically, as large blocks of memory are subdivided, | |
1481 | * the order in which smaller blocks are delivered depends on the order | |
1482 | * they're subdivided in this function. This is the primary factor | |
1483 | * influencing the order in which pages are delivered to the IO | |
1484 | * subsystem according to empirical testing, and this is also justified | |
1485 | * by considering the behavior of a buddy system containing a single | |
1486 | * large block of memory acted on by a series of small allocations. | |
1487 | * This behavior is a critical factor in sglist merging's success. | |
1488 | * | |
6d49e352 | 1489 | * -- nyc |
1da177e4 | 1490 | */ |
085cc7d5 | 1491 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1492 | int low, int high, struct free_area *area, |
1493 | int migratetype) | |
1da177e4 LT |
1494 | { |
1495 | unsigned long size = 1 << high; | |
1496 | ||
1497 | while (high > low) { | |
1498 | area--; | |
1499 | high--; | |
1500 | size >>= 1; | |
309381fe | 1501 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1502 | |
2847cf95 | 1503 | if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && |
e30825f1 | 1504 | debug_guardpage_enabled() && |
2847cf95 | 1505 | high < debug_guardpage_minorder()) { |
c0a32fc5 SG |
1506 | /* |
1507 | * Mark as guard pages (or page), that will allow to | |
1508 | * merge back to allocator when buddy will be freed. | |
1509 | * Corresponding page table entries will not be touched, | |
1510 | * pages will stay not present in virtual address space | |
1511 | */ | |
2847cf95 | 1512 | set_page_guard(zone, &page[size], high, migratetype); |
c0a32fc5 SG |
1513 | continue; |
1514 | } | |
b2a0ac88 | 1515 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1516 | area->nr_free++; |
1517 | set_page_order(&page[size], high); | |
1518 | } | |
1da177e4 LT |
1519 | } |
1520 | ||
1da177e4 LT |
1521 | /* |
1522 | * This page is about to be returned from the page allocator | |
1523 | */ | |
2a7684a2 | 1524 | static inline int check_new_page(struct page *page) |
1da177e4 | 1525 | { |
7bfec6f4 MG |
1526 | const char *bad_reason; |
1527 | unsigned long bad_flags; | |
1528 | ||
1529 | if (page_expected_state(page, PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)) | |
1530 | return 0; | |
f0b791a3 | 1531 | |
7bfec6f4 MG |
1532 | bad_reason = NULL; |
1533 | bad_flags = 0; | |
53f9263b | 1534 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1535 | bad_reason = "nonzero mapcount"; |
1536 | if (unlikely(page->mapping != NULL)) | |
1537 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1538 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1539 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1540 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1541 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1542 | bad_flags = __PG_HWPOISON; | |
1543 | } | |
f0b791a3 DH |
1544 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1545 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1546 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1547 | } | |
9edad6ea JW |
1548 | #ifdef CONFIG_MEMCG |
1549 | if (unlikely(page->mem_cgroup)) | |
1550 | bad_reason = "page still charged to cgroup"; | |
1551 | #endif | |
f0b791a3 DH |
1552 | if (unlikely(bad_reason)) { |
1553 | bad_page(page, bad_reason, bad_flags); | |
689bcebf | 1554 | return 1; |
8cc3b392 | 1555 | } |
2a7684a2 WF |
1556 | return 0; |
1557 | } | |
1558 | ||
1414c7f4 LA |
1559 | static inline bool free_pages_prezeroed(bool poisoned) |
1560 | { | |
1561 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
1562 | page_poisoning_enabled() && poisoned; | |
1563 | } | |
1564 | ||
75379191 | 1565 | static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
c603844b | 1566 | unsigned int alloc_flags) |
2a7684a2 WF |
1567 | { |
1568 | int i; | |
1414c7f4 | 1569 | bool poisoned = true; |
2a7684a2 WF |
1570 | |
1571 | for (i = 0; i < (1 << order); i++) { | |
1572 | struct page *p = page + i; | |
1573 | if (unlikely(check_new_page(p))) | |
1574 | return 1; | |
1414c7f4 LA |
1575 | if (poisoned) |
1576 | poisoned &= page_is_poisoned(p); | |
2a7684a2 | 1577 | } |
689bcebf | 1578 | |
4c21e2f2 | 1579 | set_page_private(page, 0); |
7835e98b | 1580 | set_page_refcounted(page); |
cc102509 NP |
1581 | |
1582 | arch_alloc_page(page, order); | |
1da177e4 | 1583 | kernel_map_pages(page, 1 << order, 1); |
8823b1db | 1584 | kernel_poison_pages(page, 1 << order, 1); |
b8c73fc2 | 1585 | kasan_alloc_pages(page, order); |
17cf4406 | 1586 | |
1414c7f4 | 1587 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1588 | for (i = 0; i < (1 << order); i++) |
1589 | clear_highpage(page + i); | |
17cf4406 NP |
1590 | |
1591 | if (order && (gfp_flags & __GFP_COMP)) | |
1592 | prep_compound_page(page, order); | |
1593 | ||
48c96a36 JK |
1594 | set_page_owner(page, order, gfp_flags); |
1595 | ||
75379191 | 1596 | /* |
2f064f34 | 1597 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1598 | * allocate the page. The expectation is that the caller is taking |
1599 | * steps that will free more memory. The caller should avoid the page | |
1600 | * being used for !PFMEMALLOC purposes. | |
1601 | */ | |
2f064f34 MH |
1602 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1603 | set_page_pfmemalloc(page); | |
1604 | else | |
1605 | clear_page_pfmemalloc(page); | |
75379191 | 1606 | |
689bcebf | 1607 | return 0; |
1da177e4 LT |
1608 | } |
1609 | ||
56fd56b8 MG |
1610 | /* |
1611 | * Go through the free lists for the given migratetype and remove | |
1612 | * the smallest available page from the freelists | |
1613 | */ | |
728ec980 MG |
1614 | static inline |
1615 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1616 | int migratetype) |
1617 | { | |
1618 | unsigned int current_order; | |
b8af2941 | 1619 | struct free_area *area; |
56fd56b8 MG |
1620 | struct page *page; |
1621 | ||
1622 | /* Find a page of the appropriate size in the preferred list */ | |
1623 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1624 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1625 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1626 | struct page, lru); |
a16601c5 GT |
1627 | if (!page) |
1628 | continue; | |
56fd56b8 MG |
1629 | list_del(&page->lru); |
1630 | rmv_page_order(page); | |
1631 | area->nr_free--; | |
56fd56b8 | 1632 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1633 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1634 | return page; |
1635 | } | |
1636 | ||
1637 | return NULL; | |
1638 | } | |
1639 | ||
1640 | ||
b2a0ac88 MG |
1641 | /* |
1642 | * This array describes the order lists are fallen back to when | |
1643 | * the free lists for the desirable migrate type are depleted | |
1644 | */ | |
47118af0 | 1645 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1646 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1647 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1648 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1649 | #ifdef CONFIG_CMA |
974a786e | 1650 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1651 | #endif |
194159fb | 1652 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1653 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1654 | #endif |
b2a0ac88 MG |
1655 | }; |
1656 | ||
dc67647b JK |
1657 | #ifdef CONFIG_CMA |
1658 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1659 | unsigned int order) | |
1660 | { | |
1661 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1662 | } | |
1663 | #else | |
1664 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1665 | unsigned int order) { return NULL; } | |
1666 | #endif | |
1667 | ||
c361be55 MG |
1668 | /* |
1669 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1670 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1671 | * boundary. If alignment is required, use move_freepages_block() |
1672 | */ | |
435b405c | 1673 | int move_freepages(struct zone *zone, |
b69a7288 AB |
1674 | struct page *start_page, struct page *end_page, |
1675 | int migratetype) | |
c361be55 MG |
1676 | { |
1677 | struct page *page; | |
d00181b9 | 1678 | unsigned int order; |
d100313f | 1679 | int pages_moved = 0; |
c361be55 MG |
1680 | |
1681 | #ifndef CONFIG_HOLES_IN_ZONE | |
1682 | /* | |
1683 | * page_zone is not safe to call in this context when | |
1684 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1685 | * anyway as we check zone boundaries in move_freepages_block(). | |
1686 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1687 | * grouping pages by mobility |
c361be55 | 1688 | */ |
97ee4ba7 | 1689 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1690 | #endif |
1691 | ||
1692 | for (page = start_page; page <= end_page;) { | |
344c790e | 1693 | /* Make sure we are not inadvertently changing nodes */ |
309381fe | 1694 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
344c790e | 1695 | |
c361be55 MG |
1696 | if (!pfn_valid_within(page_to_pfn(page))) { |
1697 | page++; | |
1698 | continue; | |
1699 | } | |
1700 | ||
1701 | if (!PageBuddy(page)) { | |
1702 | page++; | |
1703 | continue; | |
1704 | } | |
1705 | ||
1706 | order = page_order(page); | |
84be48d8 KS |
1707 | list_move(&page->lru, |
1708 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 1709 | page += 1 << order; |
d100313f | 1710 | pages_moved += 1 << order; |
c361be55 MG |
1711 | } |
1712 | ||
d100313f | 1713 | return pages_moved; |
c361be55 MG |
1714 | } |
1715 | ||
ee6f509c | 1716 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1717 | int migratetype) |
c361be55 MG |
1718 | { |
1719 | unsigned long start_pfn, end_pfn; | |
1720 | struct page *start_page, *end_page; | |
1721 | ||
1722 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1723 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1724 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1725 | end_page = start_page + pageblock_nr_pages - 1; |
1726 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1727 | |
1728 | /* Do not cross zone boundaries */ | |
108bcc96 | 1729 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1730 | start_page = page; |
108bcc96 | 1731 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1732 | return 0; |
1733 | ||
1734 | return move_freepages(zone, start_page, end_page, migratetype); | |
1735 | } | |
1736 | ||
2f66a68f MG |
1737 | static void change_pageblock_range(struct page *pageblock_page, |
1738 | int start_order, int migratetype) | |
1739 | { | |
1740 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1741 | ||
1742 | while (nr_pageblocks--) { | |
1743 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1744 | pageblock_page += pageblock_nr_pages; | |
1745 | } | |
1746 | } | |
1747 | ||
fef903ef | 1748 | /* |
9c0415eb VB |
1749 | * When we are falling back to another migratetype during allocation, try to |
1750 | * steal extra free pages from the same pageblocks to satisfy further | |
1751 | * allocations, instead of polluting multiple pageblocks. | |
1752 | * | |
1753 | * If we are stealing a relatively large buddy page, it is likely there will | |
1754 | * be more free pages in the pageblock, so try to steal them all. For | |
1755 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1756 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1757 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1758 | * pageblocks. | |
fef903ef | 1759 | */ |
4eb7dce6 JK |
1760 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1761 | { | |
1762 | /* | |
1763 | * Leaving this order check is intended, although there is | |
1764 | * relaxed order check in next check. The reason is that | |
1765 | * we can actually steal whole pageblock if this condition met, | |
1766 | * but, below check doesn't guarantee it and that is just heuristic | |
1767 | * so could be changed anytime. | |
1768 | */ | |
1769 | if (order >= pageblock_order) | |
1770 | return true; | |
1771 | ||
1772 | if (order >= pageblock_order / 2 || | |
1773 | start_mt == MIGRATE_RECLAIMABLE || | |
1774 | start_mt == MIGRATE_UNMOVABLE || | |
1775 | page_group_by_mobility_disabled) | |
1776 | return true; | |
1777 | ||
1778 | return false; | |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * This function implements actual steal behaviour. If order is large enough, | |
1783 | * we can steal whole pageblock. If not, we first move freepages in this | |
1784 | * pageblock and check whether half of pages are moved or not. If half of | |
1785 | * pages are moved, we can change migratetype of pageblock and permanently | |
1786 | * use it's pages as requested migratetype in the future. | |
1787 | */ | |
1788 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1789 | int start_type) | |
fef903ef | 1790 | { |
d00181b9 | 1791 | unsigned int current_order = page_order(page); |
4eb7dce6 | 1792 | int pages; |
fef903ef | 1793 | |
fef903ef SB |
1794 | /* Take ownership for orders >= pageblock_order */ |
1795 | if (current_order >= pageblock_order) { | |
1796 | change_pageblock_range(page, current_order, start_type); | |
3a1086fb | 1797 | return; |
fef903ef SB |
1798 | } |
1799 | ||
4eb7dce6 | 1800 | pages = move_freepages_block(zone, page, start_type); |
fef903ef | 1801 | |
4eb7dce6 JK |
1802 | /* Claim the whole block if over half of it is free */ |
1803 | if (pages >= (1 << (pageblock_order-1)) || | |
1804 | page_group_by_mobility_disabled) | |
1805 | set_pageblock_migratetype(page, start_type); | |
1806 | } | |
1807 | ||
2149cdae JK |
1808 | /* |
1809 | * Check whether there is a suitable fallback freepage with requested order. | |
1810 | * If only_stealable is true, this function returns fallback_mt only if | |
1811 | * we can steal other freepages all together. This would help to reduce | |
1812 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1813 | */ | |
1814 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1815 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1816 | { |
1817 | int i; | |
1818 | int fallback_mt; | |
1819 | ||
1820 | if (area->nr_free == 0) | |
1821 | return -1; | |
1822 | ||
1823 | *can_steal = false; | |
1824 | for (i = 0;; i++) { | |
1825 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 1826 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
1827 | break; |
1828 | ||
1829 | if (list_empty(&area->free_list[fallback_mt])) | |
1830 | continue; | |
fef903ef | 1831 | |
4eb7dce6 JK |
1832 | if (can_steal_fallback(order, migratetype)) |
1833 | *can_steal = true; | |
1834 | ||
2149cdae JK |
1835 | if (!only_stealable) |
1836 | return fallback_mt; | |
1837 | ||
1838 | if (*can_steal) | |
1839 | return fallback_mt; | |
fef903ef | 1840 | } |
4eb7dce6 JK |
1841 | |
1842 | return -1; | |
fef903ef SB |
1843 | } |
1844 | ||
0aaa29a5 MG |
1845 | /* |
1846 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
1847 | * there are no empty page blocks that contain a page with a suitable order | |
1848 | */ | |
1849 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
1850 | unsigned int alloc_order) | |
1851 | { | |
1852 | int mt; | |
1853 | unsigned long max_managed, flags; | |
1854 | ||
1855 | /* | |
1856 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
1857 | * Check is race-prone but harmless. | |
1858 | */ | |
1859 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
1860 | if (zone->nr_reserved_highatomic >= max_managed) | |
1861 | return; | |
1862 | ||
1863 | spin_lock_irqsave(&zone->lock, flags); | |
1864 | ||
1865 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
1866 | if (zone->nr_reserved_highatomic >= max_managed) | |
1867 | goto out_unlock; | |
1868 | ||
1869 | /* Yoink! */ | |
1870 | mt = get_pageblock_migratetype(page); | |
1871 | if (mt != MIGRATE_HIGHATOMIC && | |
1872 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { | |
1873 | zone->nr_reserved_highatomic += pageblock_nr_pages; | |
1874 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
1875 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); | |
1876 | } | |
1877 | ||
1878 | out_unlock: | |
1879 | spin_unlock_irqrestore(&zone->lock, flags); | |
1880 | } | |
1881 | ||
1882 | /* | |
1883 | * Used when an allocation is about to fail under memory pressure. This | |
1884 | * potentially hurts the reliability of high-order allocations when under | |
1885 | * intense memory pressure but failed atomic allocations should be easier | |
1886 | * to recover from than an OOM. | |
1887 | */ | |
1888 | static void unreserve_highatomic_pageblock(const struct alloc_context *ac) | |
1889 | { | |
1890 | struct zonelist *zonelist = ac->zonelist; | |
1891 | unsigned long flags; | |
1892 | struct zoneref *z; | |
1893 | struct zone *zone; | |
1894 | struct page *page; | |
1895 | int order; | |
1896 | ||
1897 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
1898 | ac->nodemask) { | |
1899 | /* Preserve at least one pageblock */ | |
1900 | if (zone->nr_reserved_highatomic <= pageblock_nr_pages) | |
1901 | continue; | |
1902 | ||
1903 | spin_lock_irqsave(&zone->lock, flags); | |
1904 | for (order = 0; order < MAX_ORDER; order++) { | |
1905 | struct free_area *area = &(zone->free_area[order]); | |
1906 | ||
a16601c5 GT |
1907 | page = list_first_entry_or_null( |
1908 | &area->free_list[MIGRATE_HIGHATOMIC], | |
1909 | struct page, lru); | |
1910 | if (!page) | |
0aaa29a5 MG |
1911 | continue; |
1912 | ||
0aaa29a5 MG |
1913 | /* |
1914 | * It should never happen but changes to locking could | |
1915 | * inadvertently allow a per-cpu drain to add pages | |
1916 | * to MIGRATE_HIGHATOMIC while unreserving so be safe | |
1917 | * and watch for underflows. | |
1918 | */ | |
1919 | zone->nr_reserved_highatomic -= min(pageblock_nr_pages, | |
1920 | zone->nr_reserved_highatomic); | |
1921 | ||
1922 | /* | |
1923 | * Convert to ac->migratetype and avoid the normal | |
1924 | * pageblock stealing heuristics. Minimally, the caller | |
1925 | * is doing the work and needs the pages. More | |
1926 | * importantly, if the block was always converted to | |
1927 | * MIGRATE_UNMOVABLE or another type then the number | |
1928 | * of pageblocks that cannot be completely freed | |
1929 | * may increase. | |
1930 | */ | |
1931 | set_pageblock_migratetype(page, ac->migratetype); | |
1932 | move_freepages_block(zone, page, ac->migratetype); | |
1933 | spin_unlock_irqrestore(&zone->lock, flags); | |
1934 | return; | |
1935 | } | |
1936 | spin_unlock_irqrestore(&zone->lock, flags); | |
1937 | } | |
1938 | } | |
1939 | ||
b2a0ac88 | 1940 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 1941 | static inline struct page * |
7aeb09f9 | 1942 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 1943 | { |
b8af2941 | 1944 | struct free_area *area; |
7aeb09f9 | 1945 | unsigned int current_order; |
b2a0ac88 | 1946 | struct page *page; |
4eb7dce6 JK |
1947 | int fallback_mt; |
1948 | bool can_steal; | |
b2a0ac88 MG |
1949 | |
1950 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
1951 | for (current_order = MAX_ORDER-1; |
1952 | current_order >= order && current_order <= MAX_ORDER-1; | |
1953 | --current_order) { | |
4eb7dce6 JK |
1954 | area = &(zone->free_area[current_order]); |
1955 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 1956 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
1957 | if (fallback_mt == -1) |
1958 | continue; | |
b2a0ac88 | 1959 | |
a16601c5 | 1960 | page = list_first_entry(&area->free_list[fallback_mt], |
4eb7dce6 JK |
1961 | struct page, lru); |
1962 | if (can_steal) | |
1963 | steal_suitable_fallback(zone, page, start_migratetype); | |
b2a0ac88 | 1964 | |
4eb7dce6 JK |
1965 | /* Remove the page from the freelists */ |
1966 | area->nr_free--; | |
1967 | list_del(&page->lru); | |
1968 | rmv_page_order(page); | |
3a1086fb | 1969 | |
4eb7dce6 JK |
1970 | expand(zone, page, order, current_order, area, |
1971 | start_migratetype); | |
1972 | /* | |
bb14c2c7 | 1973 | * The pcppage_migratetype may differ from pageblock's |
4eb7dce6 | 1974 | * migratetype depending on the decisions in |
bb14c2c7 VB |
1975 | * find_suitable_fallback(). This is OK as long as it does not |
1976 | * differ for MIGRATE_CMA pageblocks. Those can be used as | |
1977 | * fallback only via special __rmqueue_cma_fallback() function | |
4eb7dce6 | 1978 | */ |
bb14c2c7 | 1979 | set_pcppage_migratetype(page, start_migratetype); |
e0fff1bd | 1980 | |
4eb7dce6 JK |
1981 | trace_mm_page_alloc_extfrag(page, order, current_order, |
1982 | start_migratetype, fallback_mt); | |
e0fff1bd | 1983 | |
4eb7dce6 | 1984 | return page; |
b2a0ac88 MG |
1985 | } |
1986 | ||
728ec980 | 1987 | return NULL; |
b2a0ac88 MG |
1988 | } |
1989 | ||
56fd56b8 | 1990 | /* |
1da177e4 LT |
1991 | * Do the hard work of removing an element from the buddy allocator. |
1992 | * Call me with the zone->lock already held. | |
1993 | */ | |
b2a0ac88 | 1994 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
6ac0206b | 1995 | int migratetype) |
1da177e4 | 1996 | { |
1da177e4 LT |
1997 | struct page *page; |
1998 | ||
56fd56b8 | 1999 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 2000 | if (unlikely(!page)) { |
dc67647b JK |
2001 | if (migratetype == MIGRATE_MOVABLE) |
2002 | page = __rmqueue_cma_fallback(zone, order); | |
2003 | ||
2004 | if (!page) | |
2005 | page = __rmqueue_fallback(zone, order, migratetype); | |
728ec980 MG |
2006 | } |
2007 | ||
0d3d062a | 2008 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 2009 | return page; |
1da177e4 LT |
2010 | } |
2011 | ||
5f63b720 | 2012 | /* |
1da177e4 LT |
2013 | * Obtain a specified number of elements from the buddy allocator, all under |
2014 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
2015 | * Returns the number of new pages which were placed at *list. | |
2016 | */ | |
5f63b720 | 2017 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 2018 | unsigned long count, struct list_head *list, |
b745bc85 | 2019 | int migratetype, bool cold) |
1da177e4 | 2020 | { |
5bcc9f86 | 2021 | int i; |
5f63b720 | 2022 | |
c54ad30c | 2023 | spin_lock(&zone->lock); |
1da177e4 | 2024 | for (i = 0; i < count; ++i) { |
6ac0206b | 2025 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 2026 | if (unlikely(page == NULL)) |
1da177e4 | 2027 | break; |
81eabcbe MG |
2028 | |
2029 | /* | |
2030 | * Split buddy pages returned by expand() are received here | |
2031 | * in physical page order. The page is added to the callers and | |
2032 | * list and the list head then moves forward. From the callers | |
2033 | * perspective, the linked list is ordered by page number in | |
2034 | * some conditions. This is useful for IO devices that can | |
2035 | * merge IO requests if the physical pages are ordered | |
2036 | * properly. | |
2037 | */ | |
b745bc85 | 2038 | if (likely(!cold)) |
e084b2d9 MG |
2039 | list_add(&page->lru, list); |
2040 | else | |
2041 | list_add_tail(&page->lru, list); | |
81eabcbe | 2042 | list = &page->lru; |
bb14c2c7 | 2043 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
2044 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2045 | -(1 << order)); | |
1da177e4 | 2046 | } |
f2260e6b | 2047 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 2048 | spin_unlock(&zone->lock); |
085cc7d5 | 2049 | return i; |
1da177e4 LT |
2050 | } |
2051 | ||
4ae7c039 | 2052 | #ifdef CONFIG_NUMA |
8fce4d8e | 2053 | /* |
4037d452 CL |
2054 | * Called from the vmstat counter updater to drain pagesets of this |
2055 | * currently executing processor on remote nodes after they have | |
2056 | * expired. | |
2057 | * | |
879336c3 CL |
2058 | * Note that this function must be called with the thread pinned to |
2059 | * a single processor. | |
8fce4d8e | 2060 | */ |
4037d452 | 2061 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 2062 | { |
4ae7c039 | 2063 | unsigned long flags; |
7be12fc9 | 2064 | int to_drain, batch; |
4ae7c039 | 2065 | |
4037d452 | 2066 | local_irq_save(flags); |
4db0c3c2 | 2067 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 2068 | to_drain = min(pcp->count, batch); |
2a13515c KM |
2069 | if (to_drain > 0) { |
2070 | free_pcppages_bulk(zone, to_drain, pcp); | |
2071 | pcp->count -= to_drain; | |
2072 | } | |
4037d452 | 2073 | local_irq_restore(flags); |
4ae7c039 CL |
2074 | } |
2075 | #endif | |
2076 | ||
9f8f2172 | 2077 | /* |
93481ff0 | 2078 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2079 | * |
2080 | * The processor must either be the current processor and the | |
2081 | * thread pinned to the current processor or a processor that | |
2082 | * is not online. | |
2083 | */ | |
93481ff0 | 2084 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2085 | { |
c54ad30c | 2086 | unsigned long flags; |
93481ff0 VB |
2087 | struct per_cpu_pageset *pset; |
2088 | struct per_cpu_pages *pcp; | |
1da177e4 | 2089 | |
93481ff0 VB |
2090 | local_irq_save(flags); |
2091 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2092 | |
93481ff0 VB |
2093 | pcp = &pset->pcp; |
2094 | if (pcp->count) { | |
2095 | free_pcppages_bulk(zone, pcp->count, pcp); | |
2096 | pcp->count = 0; | |
2097 | } | |
2098 | local_irq_restore(flags); | |
2099 | } | |
3dfa5721 | 2100 | |
93481ff0 VB |
2101 | /* |
2102 | * Drain pcplists of all zones on the indicated processor. | |
2103 | * | |
2104 | * The processor must either be the current processor and the | |
2105 | * thread pinned to the current processor or a processor that | |
2106 | * is not online. | |
2107 | */ | |
2108 | static void drain_pages(unsigned int cpu) | |
2109 | { | |
2110 | struct zone *zone; | |
2111 | ||
2112 | for_each_populated_zone(zone) { | |
2113 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2114 | } |
2115 | } | |
1da177e4 | 2116 | |
9f8f2172 CL |
2117 | /* |
2118 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2119 | * |
2120 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2121 | * the single zone's pages. | |
9f8f2172 | 2122 | */ |
93481ff0 | 2123 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2124 | { |
93481ff0 VB |
2125 | int cpu = smp_processor_id(); |
2126 | ||
2127 | if (zone) | |
2128 | drain_pages_zone(cpu, zone); | |
2129 | else | |
2130 | drain_pages(cpu); | |
9f8f2172 CL |
2131 | } |
2132 | ||
2133 | /* | |
74046494 GBY |
2134 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2135 | * | |
93481ff0 VB |
2136 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2137 | * | |
74046494 GBY |
2138 | * Note that this code is protected against sending an IPI to an offline |
2139 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
2140 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
2141 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
2142 | * before the call to on_each_cpu_mask(). | |
9f8f2172 | 2143 | */ |
93481ff0 | 2144 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2145 | { |
74046494 | 2146 | int cpu; |
74046494 GBY |
2147 | |
2148 | /* | |
2149 | * Allocate in the BSS so we wont require allocation in | |
2150 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2151 | */ | |
2152 | static cpumask_t cpus_with_pcps; | |
2153 | ||
2154 | /* | |
2155 | * We don't care about racing with CPU hotplug event | |
2156 | * as offline notification will cause the notified | |
2157 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2158 | * disables preemption as part of its processing | |
2159 | */ | |
2160 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2161 | struct per_cpu_pageset *pcp; |
2162 | struct zone *z; | |
74046494 | 2163 | bool has_pcps = false; |
93481ff0 VB |
2164 | |
2165 | if (zone) { | |
74046494 | 2166 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2167 | if (pcp->pcp.count) |
74046494 | 2168 | has_pcps = true; |
93481ff0 VB |
2169 | } else { |
2170 | for_each_populated_zone(z) { | |
2171 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2172 | if (pcp->pcp.count) { | |
2173 | has_pcps = true; | |
2174 | break; | |
2175 | } | |
74046494 GBY |
2176 | } |
2177 | } | |
93481ff0 | 2178 | |
74046494 GBY |
2179 | if (has_pcps) |
2180 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2181 | else | |
2182 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2183 | } | |
93481ff0 VB |
2184 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
2185 | zone, 1); | |
9f8f2172 CL |
2186 | } |
2187 | ||
296699de | 2188 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
2189 | |
2190 | void mark_free_pages(struct zone *zone) | |
2191 | { | |
f623f0db RW |
2192 | unsigned long pfn, max_zone_pfn; |
2193 | unsigned long flags; | |
7aeb09f9 | 2194 | unsigned int order, t; |
86760a2c | 2195 | struct page *page; |
1da177e4 | 2196 | |
8080fc03 | 2197 | if (zone_is_empty(zone)) |
1da177e4 LT |
2198 | return; |
2199 | ||
2200 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2201 | |
108bcc96 | 2202 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2203 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2204 | if (pfn_valid(pfn)) { | |
86760a2c | 2205 | page = pfn_to_page(pfn); |
ba6b0979 JK |
2206 | |
2207 | if (page_zone(page) != zone) | |
2208 | continue; | |
2209 | ||
7be98234 RW |
2210 | if (!swsusp_page_is_forbidden(page)) |
2211 | swsusp_unset_page_free(page); | |
f623f0db | 2212 | } |
1da177e4 | 2213 | |
b2a0ac88 | 2214 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2215 | list_for_each_entry(page, |
2216 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2217 | unsigned long i; |
1da177e4 | 2218 | |
86760a2c | 2219 | pfn = page_to_pfn(page); |
f623f0db | 2220 | for (i = 0; i < (1UL << order); i++) |
7be98234 | 2221 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 2222 | } |
b2a0ac88 | 2223 | } |
1da177e4 LT |
2224 | spin_unlock_irqrestore(&zone->lock, flags); |
2225 | } | |
e2c55dc8 | 2226 | #endif /* CONFIG_PM */ |
1da177e4 | 2227 | |
1da177e4 LT |
2228 | /* |
2229 | * Free a 0-order page | |
b745bc85 | 2230 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 2231 | */ |
b745bc85 | 2232 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
2233 | { |
2234 | struct zone *zone = page_zone(page); | |
2235 | struct per_cpu_pages *pcp; | |
2236 | unsigned long flags; | |
dc4b0caf | 2237 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 2238 | int migratetype; |
1da177e4 | 2239 | |
ec95f53a | 2240 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
2241 | return; |
2242 | ||
dc4b0caf | 2243 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2244 | set_pcppage_migratetype(page, migratetype); |
1da177e4 | 2245 | local_irq_save(flags); |
f8891e5e | 2246 | __count_vm_event(PGFREE); |
da456f14 | 2247 | |
5f8dcc21 MG |
2248 | /* |
2249 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2250 | * Free ISOLATE pages back to the allocator because they are being | |
2251 | * offlined but treat RESERVE as movable pages so we can get those | |
2252 | * areas back if necessary. Otherwise, we may have to free | |
2253 | * excessively into the page allocator | |
2254 | */ | |
2255 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2256 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2257 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
2258 | goto out; |
2259 | } | |
2260 | migratetype = MIGRATE_MOVABLE; | |
2261 | } | |
2262 | ||
99dcc3e5 | 2263 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 2264 | if (!cold) |
5f8dcc21 | 2265 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
2266 | else |
2267 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 2268 | pcp->count++; |
48db57f8 | 2269 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2270 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
2271 | free_pcppages_bulk(zone, batch, pcp); |
2272 | pcp->count -= batch; | |
48db57f8 | 2273 | } |
5f8dcc21 MG |
2274 | |
2275 | out: | |
1da177e4 | 2276 | local_irq_restore(flags); |
1da177e4 LT |
2277 | } |
2278 | ||
cc59850e KK |
2279 | /* |
2280 | * Free a list of 0-order pages | |
2281 | */ | |
b745bc85 | 2282 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
2283 | { |
2284 | struct page *page, *next; | |
2285 | ||
2286 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 2287 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
2288 | free_hot_cold_page(page, cold); |
2289 | } | |
2290 | } | |
2291 | ||
8dfcc9ba NP |
2292 | /* |
2293 | * split_page takes a non-compound higher-order page, and splits it into | |
2294 | * n (1<<order) sub-pages: page[0..n] | |
2295 | * Each sub-page must be freed individually. | |
2296 | * | |
2297 | * Note: this is probably too low level an operation for use in drivers. | |
2298 | * Please consult with lkml before using this in your driver. | |
2299 | */ | |
2300 | void split_page(struct page *page, unsigned int order) | |
2301 | { | |
2302 | int i; | |
e2cfc911 | 2303 | gfp_t gfp_mask; |
8dfcc9ba | 2304 | |
309381fe SL |
2305 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2306 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
2307 | |
2308 | #ifdef CONFIG_KMEMCHECK | |
2309 | /* | |
2310 | * Split shadow pages too, because free(page[0]) would | |
2311 | * otherwise free the whole shadow. | |
2312 | */ | |
2313 | if (kmemcheck_page_is_tracked(page)) | |
2314 | split_page(virt_to_page(page[0].shadow), order); | |
2315 | #endif | |
2316 | ||
e2cfc911 JK |
2317 | gfp_mask = get_page_owner_gfp(page); |
2318 | set_page_owner(page, 0, gfp_mask); | |
48c96a36 | 2319 | for (i = 1; i < (1 << order); i++) { |
7835e98b | 2320 | set_page_refcounted(page + i); |
e2cfc911 | 2321 | set_page_owner(page + i, 0, gfp_mask); |
48c96a36 | 2322 | } |
8dfcc9ba | 2323 | } |
5853ff23 | 2324 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2325 | |
3c605096 | 2326 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2327 | { |
748446bb MG |
2328 | unsigned long watermark; |
2329 | struct zone *zone; | |
2139cbe6 | 2330 | int mt; |
748446bb MG |
2331 | |
2332 | BUG_ON(!PageBuddy(page)); | |
2333 | ||
2334 | zone = page_zone(page); | |
2e30abd1 | 2335 | mt = get_pageblock_migratetype(page); |
748446bb | 2336 | |
194159fb | 2337 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
2338 | /* Obey watermarks as if the page was being allocated */ |
2339 | watermark = low_wmark_pages(zone) + (1 << order); | |
2340 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
2341 | return 0; | |
2342 | ||
8fb74b9f | 2343 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2344 | } |
748446bb MG |
2345 | |
2346 | /* Remove page from free list */ | |
2347 | list_del(&page->lru); | |
2348 | zone->free_area[order].nr_free--; | |
2349 | rmv_page_order(page); | |
2139cbe6 | 2350 | |
e2cfc911 | 2351 | set_page_owner(page, order, __GFP_MOVABLE); |
f3a14ced | 2352 | |
8fb74b9f | 2353 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
2354 | if (order >= pageblock_order - 1) { |
2355 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2356 | for (; page < endpage; page += pageblock_nr_pages) { |
2357 | int mt = get_pageblock_migratetype(page); | |
194159fb | 2358 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
2359 | set_pageblock_migratetype(page, |
2360 | MIGRATE_MOVABLE); | |
2361 | } | |
748446bb MG |
2362 | } |
2363 | ||
f3a14ced | 2364 | |
8fb74b9f | 2365 | return 1UL << order; |
1fb3f8ca MG |
2366 | } |
2367 | ||
2368 | /* | |
2369 | * Similar to split_page except the page is already free. As this is only | |
2370 | * being used for migration, the migratetype of the block also changes. | |
2371 | * As this is called with interrupts disabled, the caller is responsible | |
2372 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
2373 | * are enabled. | |
2374 | * | |
2375 | * Note: this is probably too low level an operation for use in drivers. | |
2376 | * Please consult with lkml before using this in your driver. | |
2377 | */ | |
2378 | int split_free_page(struct page *page) | |
2379 | { | |
2380 | unsigned int order; | |
2381 | int nr_pages; | |
2382 | ||
1fb3f8ca MG |
2383 | order = page_order(page); |
2384 | ||
8fb74b9f | 2385 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
2386 | if (!nr_pages) |
2387 | return 0; | |
2388 | ||
2389 | /* Split into individual pages */ | |
2390 | set_page_refcounted(page); | |
2391 | split_page(page, order); | |
2392 | return nr_pages; | |
748446bb MG |
2393 | } |
2394 | ||
060e7417 MG |
2395 | /* |
2396 | * Update NUMA hit/miss statistics | |
2397 | * | |
2398 | * Must be called with interrupts disabled. | |
2399 | * | |
2400 | * When __GFP_OTHER_NODE is set assume the node of the preferred | |
2401 | * zone is the local node. This is useful for daemons who allocate | |
2402 | * memory on behalf of other processes. | |
2403 | */ | |
2404 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, | |
2405 | gfp_t flags) | |
2406 | { | |
2407 | #ifdef CONFIG_NUMA | |
2408 | int local_nid = numa_node_id(); | |
2409 | enum zone_stat_item local_stat = NUMA_LOCAL; | |
2410 | ||
2411 | if (unlikely(flags & __GFP_OTHER_NODE)) { | |
2412 | local_stat = NUMA_OTHER; | |
2413 | local_nid = preferred_zone->node; | |
2414 | } | |
2415 | ||
2416 | if (z->node == local_nid) { | |
2417 | __inc_zone_state(z, NUMA_HIT); | |
2418 | __inc_zone_state(z, local_stat); | |
2419 | } else { | |
2420 | __inc_zone_state(z, NUMA_MISS); | |
2421 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); | |
2422 | } | |
2423 | #endif | |
2424 | } | |
2425 | ||
1da177e4 | 2426 | /* |
75379191 | 2427 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2428 | */ |
0a15c3e9 MG |
2429 | static inline |
2430 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
7aeb09f9 | 2431 | struct zone *zone, unsigned int order, |
c603844b MG |
2432 | gfp_t gfp_flags, unsigned int alloc_flags, |
2433 | int migratetype) | |
1da177e4 LT |
2434 | { |
2435 | unsigned long flags; | |
689bcebf | 2436 | struct page *page; |
b745bc85 | 2437 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
1da177e4 | 2438 | |
48db57f8 | 2439 | if (likely(order == 0)) { |
1da177e4 | 2440 | struct per_cpu_pages *pcp; |
5f8dcc21 | 2441 | struct list_head *list; |
1da177e4 | 2442 | |
1da177e4 | 2443 | local_irq_save(flags); |
99dcc3e5 CL |
2444 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2445 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 2446 | if (list_empty(list)) { |
535131e6 | 2447 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 2448 | pcp->batch, list, |
e084b2d9 | 2449 | migratetype, cold); |
5f8dcc21 | 2450 | if (unlikely(list_empty(list))) |
6fb332fa | 2451 | goto failed; |
535131e6 | 2452 | } |
b92a6edd | 2453 | |
5f8dcc21 | 2454 | if (cold) |
a16601c5 | 2455 | page = list_last_entry(list, struct page, lru); |
5f8dcc21 | 2456 | else |
a16601c5 | 2457 | page = list_first_entry(list, struct page, lru); |
5f8dcc21 | 2458 | |
754078eb | 2459 | __dec_zone_state(zone, NR_ALLOC_BATCH); |
b92a6edd MG |
2460 | list_del(&page->lru); |
2461 | pcp->count--; | |
7fb1d9fc | 2462 | } else { |
0f352e53 MH |
2463 | /* |
2464 | * We most definitely don't want callers attempting to | |
2465 | * allocate greater than order-1 page units with __GFP_NOFAIL. | |
2466 | */ | |
2467 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); | |
1da177e4 | 2468 | spin_lock_irqsave(&zone->lock, flags); |
0aaa29a5 MG |
2469 | |
2470 | page = NULL; | |
2471 | if (alloc_flags & ALLOC_HARDER) { | |
2472 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2473 | if (page) | |
2474 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
2475 | } | |
2476 | if (!page) | |
6ac0206b | 2477 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
2478 | spin_unlock(&zone->lock); |
2479 | if (!page) | |
2480 | goto failed; | |
754078eb | 2481 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
d1ce749a | 2482 | __mod_zone_freepage_state(zone, -(1 << order), |
bb14c2c7 | 2483 | get_pcppage_migratetype(page)); |
1da177e4 LT |
2484 | } |
2485 | ||
abe5f972 | 2486 | if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 && |
57054651 JW |
2487 | !test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) |
2488 | set_bit(ZONE_FAIR_DEPLETED, &zone->flags); | |
27329369 | 2489 | |
f8891e5e | 2490 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 2491 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 2492 | local_irq_restore(flags); |
1da177e4 | 2493 | |
309381fe | 2494 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 2495 | return page; |
a74609fa NP |
2496 | |
2497 | failed: | |
2498 | local_irq_restore(flags); | |
a74609fa | 2499 | return NULL; |
1da177e4 LT |
2500 | } |
2501 | ||
933e312e AM |
2502 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2503 | ||
b2588c4b | 2504 | static struct { |
933e312e AM |
2505 | struct fault_attr attr; |
2506 | ||
621a5f7a | 2507 | bool ignore_gfp_highmem; |
71baba4b | 2508 | bool ignore_gfp_reclaim; |
54114994 | 2509 | u32 min_order; |
933e312e AM |
2510 | } fail_page_alloc = { |
2511 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 2512 | .ignore_gfp_reclaim = true, |
621a5f7a | 2513 | .ignore_gfp_highmem = true, |
54114994 | 2514 | .min_order = 1, |
933e312e AM |
2515 | }; |
2516 | ||
2517 | static int __init setup_fail_page_alloc(char *str) | |
2518 | { | |
2519 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
2520 | } | |
2521 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
2522 | ||
deaf386e | 2523 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2524 | { |
54114994 | 2525 | if (order < fail_page_alloc.min_order) |
deaf386e | 2526 | return false; |
933e312e | 2527 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 2528 | return false; |
933e312e | 2529 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 2530 | return false; |
71baba4b MG |
2531 | if (fail_page_alloc.ignore_gfp_reclaim && |
2532 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 2533 | return false; |
933e312e AM |
2534 | |
2535 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
2536 | } | |
2537 | ||
2538 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
2539 | ||
2540 | static int __init fail_page_alloc_debugfs(void) | |
2541 | { | |
f4ae40a6 | 2542 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 2543 | struct dentry *dir; |
933e312e | 2544 | |
dd48c085 AM |
2545 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2546 | &fail_page_alloc.attr); | |
2547 | if (IS_ERR(dir)) | |
2548 | return PTR_ERR(dir); | |
933e312e | 2549 | |
b2588c4b | 2550 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 2551 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
2552 | goto fail; |
2553 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
2554 | &fail_page_alloc.ignore_gfp_highmem)) | |
2555 | goto fail; | |
2556 | if (!debugfs_create_u32("min-order", mode, dir, | |
2557 | &fail_page_alloc.min_order)) | |
2558 | goto fail; | |
2559 | ||
2560 | return 0; | |
2561 | fail: | |
dd48c085 | 2562 | debugfs_remove_recursive(dir); |
933e312e | 2563 | |
b2588c4b | 2564 | return -ENOMEM; |
933e312e AM |
2565 | } |
2566 | ||
2567 | late_initcall(fail_page_alloc_debugfs); | |
2568 | ||
2569 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2570 | ||
2571 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2572 | ||
deaf386e | 2573 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2574 | { |
deaf386e | 2575 | return false; |
933e312e AM |
2576 | } |
2577 | ||
2578 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2579 | ||
1da177e4 | 2580 | /* |
97a16fc8 MG |
2581 | * Return true if free base pages are above 'mark'. For high-order checks it |
2582 | * will return true of the order-0 watermark is reached and there is at least | |
2583 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
2584 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 2585 | */ |
7aeb09f9 | 2586 | static bool __zone_watermark_ok(struct zone *z, unsigned int order, |
c603844b MG |
2587 | unsigned long mark, int classzone_idx, |
2588 | unsigned int alloc_flags, | |
7aeb09f9 | 2589 | long free_pages) |
1da177e4 | 2590 | { |
d23ad423 | 2591 | long min = mark; |
1da177e4 | 2592 | int o; |
c603844b | 2593 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
1da177e4 | 2594 | |
0aaa29a5 | 2595 | /* free_pages may go negative - that's OK */ |
df0a6daa | 2596 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 2597 | |
7fb1d9fc | 2598 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2599 | min -= min / 2; |
0aaa29a5 MG |
2600 | |
2601 | /* | |
2602 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
2603 | * the high-atomic reserves. This will over-estimate the size of the | |
2604 | * atomic reserve but it avoids a search. | |
2605 | */ | |
97a16fc8 | 2606 | if (likely(!alloc_harder)) |
0aaa29a5 MG |
2607 | free_pages -= z->nr_reserved_highatomic; |
2608 | else | |
1da177e4 | 2609 | min -= min / 4; |
e2b19197 | 2610 | |
d95ea5d1 BZ |
2611 | #ifdef CONFIG_CMA |
2612 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2613 | if (!(alloc_flags & ALLOC_CMA)) | |
97a16fc8 | 2614 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2615 | #endif |
026b0814 | 2616 | |
97a16fc8 MG |
2617 | /* |
2618 | * Check watermarks for an order-0 allocation request. If these | |
2619 | * are not met, then a high-order request also cannot go ahead | |
2620 | * even if a suitable page happened to be free. | |
2621 | */ | |
2622 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 2623 | return false; |
1da177e4 | 2624 | |
97a16fc8 MG |
2625 | /* If this is an order-0 request then the watermark is fine */ |
2626 | if (!order) | |
2627 | return true; | |
2628 | ||
2629 | /* For a high-order request, check at least one suitable page is free */ | |
2630 | for (o = order; o < MAX_ORDER; o++) { | |
2631 | struct free_area *area = &z->free_area[o]; | |
2632 | int mt; | |
2633 | ||
2634 | if (!area->nr_free) | |
2635 | continue; | |
2636 | ||
2637 | if (alloc_harder) | |
2638 | return true; | |
1da177e4 | 2639 | |
97a16fc8 MG |
2640 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
2641 | if (!list_empty(&area->free_list[mt])) | |
2642 | return true; | |
2643 | } | |
2644 | ||
2645 | #ifdef CONFIG_CMA | |
2646 | if ((alloc_flags & ALLOC_CMA) && | |
2647 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
2648 | return true; | |
2649 | } | |
2650 | #endif | |
1da177e4 | 2651 | } |
97a16fc8 | 2652 | return false; |
88f5acf8 MG |
2653 | } |
2654 | ||
7aeb09f9 | 2655 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
c603844b | 2656 | int classzone_idx, unsigned int alloc_flags) |
88f5acf8 MG |
2657 | { |
2658 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2659 | zone_page_state(z, NR_FREE_PAGES)); | |
2660 | } | |
2661 | ||
48ee5f36 MG |
2662 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
2663 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) | |
2664 | { | |
2665 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2666 | long cma_pages = 0; | |
2667 | ||
2668 | #ifdef CONFIG_CMA | |
2669 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2670 | if (!(alloc_flags & ALLOC_CMA)) | |
2671 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); | |
2672 | #endif | |
2673 | ||
2674 | /* | |
2675 | * Fast check for order-0 only. If this fails then the reserves | |
2676 | * need to be calculated. There is a corner case where the check | |
2677 | * passes but only the high-order atomic reserve are free. If | |
2678 | * the caller is !atomic then it'll uselessly search the free | |
2679 | * list. That corner case is then slower but it is harmless. | |
2680 | */ | |
2681 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) | |
2682 | return true; | |
2683 | ||
2684 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2685 | free_pages); | |
2686 | } | |
2687 | ||
7aeb09f9 | 2688 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 2689 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
2690 | { |
2691 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2692 | ||
2693 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2694 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2695 | ||
e2b19197 | 2696 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 2697 | free_pages); |
1da177e4 LT |
2698 | } |
2699 | ||
9276b1bc | 2700 | #ifdef CONFIG_NUMA |
81c0a2bb JW |
2701 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2702 | { | |
fff4068c | 2703 | return local_zone->node == zone->node; |
81c0a2bb JW |
2704 | } |
2705 | ||
957f822a DR |
2706 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2707 | { | |
5f7a75ac MG |
2708 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
2709 | RECLAIM_DISTANCE; | |
957f822a | 2710 | } |
9276b1bc | 2711 | #else /* CONFIG_NUMA */ |
81c0a2bb JW |
2712 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2713 | { | |
2714 | return true; | |
2715 | } | |
2716 | ||
957f822a DR |
2717 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2718 | { | |
2719 | return true; | |
2720 | } | |
9276b1bc PJ |
2721 | #endif /* CONFIG_NUMA */ |
2722 | ||
4ffeaf35 MG |
2723 | static void reset_alloc_batches(struct zone *preferred_zone) |
2724 | { | |
2725 | struct zone *zone = preferred_zone->zone_pgdat->node_zones; | |
2726 | ||
2727 | do { | |
2728 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
2729 | high_wmark_pages(zone) - low_wmark_pages(zone) - | |
2730 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
57054651 | 2731 | clear_bit(ZONE_FAIR_DEPLETED, &zone->flags); |
4ffeaf35 MG |
2732 | } while (zone++ != preferred_zone); |
2733 | } | |
2734 | ||
7fb1d9fc | 2735 | /* |
0798e519 | 2736 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
2737 | * a page. |
2738 | */ | |
2739 | static struct page * | |
a9263751 VB |
2740 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
2741 | const struct alloc_context *ac) | |
753ee728 | 2742 | { |
c33d6c06 | 2743 | struct zoneref *z = ac->preferred_zoneref; |
5117f45d | 2744 | struct zone *zone; |
30534755 MG |
2745 | bool fair_skipped = false; |
2746 | bool apply_fair = (alloc_flags & ALLOC_FAIR); | |
54a6eb5c | 2747 | |
9276b1bc | 2748 | zonelist_scan: |
7fb1d9fc | 2749 | /* |
9276b1bc | 2750 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 2751 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 2752 | */ |
c33d6c06 | 2753 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
a9263751 | 2754 | ac->nodemask) { |
be06af00 | 2755 | struct page *page; |
e085dbc5 JW |
2756 | unsigned long mark; |
2757 | ||
664eedde MG |
2758 | if (cpusets_enabled() && |
2759 | (alloc_flags & ALLOC_CPUSET) && | |
344736f2 | 2760 | !cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 2761 | continue; |
81c0a2bb JW |
2762 | /* |
2763 | * Distribute pages in proportion to the individual | |
2764 | * zone size to ensure fair page aging. The zone a | |
2765 | * page was allocated in should have no effect on the | |
2766 | * time the page has in memory before being reclaimed. | |
81c0a2bb | 2767 | */ |
30534755 | 2768 | if (apply_fair) { |
57054651 | 2769 | if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) { |
fa379b95 | 2770 | fair_skipped = true; |
3a025760 | 2771 | continue; |
4ffeaf35 | 2772 | } |
c33d6c06 | 2773 | if (!zone_local(ac->preferred_zoneref->zone, zone)) { |
30534755 MG |
2774 | if (fair_skipped) |
2775 | goto reset_fair; | |
2776 | apply_fair = false; | |
2777 | } | |
81c0a2bb | 2778 | } |
a756cf59 JW |
2779 | /* |
2780 | * When allocating a page cache page for writing, we | |
2781 | * want to get it from a zone that is within its dirty | |
2782 | * limit, such that no single zone holds more than its | |
2783 | * proportional share of globally allowed dirty pages. | |
2784 | * The dirty limits take into account the zone's | |
2785 | * lowmem reserves and high watermark so that kswapd | |
2786 | * should be able to balance it without having to | |
2787 | * write pages from its LRU list. | |
2788 | * | |
2789 | * This may look like it could increase pressure on | |
2790 | * lower zones by failing allocations in higher zones | |
2791 | * before they are full. But the pages that do spill | |
2792 | * over are limited as the lower zones are protected | |
2793 | * by this very same mechanism. It should not become | |
2794 | * a practical burden to them. | |
2795 | * | |
2796 | * XXX: For now, allow allocations to potentially | |
2797 | * exceed the per-zone dirty limit in the slowpath | |
c9ab0c4f | 2798 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 JW |
2799 | * which is important when on a NUMA setup the allowed |
2800 | * zones are together not big enough to reach the | |
2801 | * global limit. The proper fix for these situations | |
2802 | * will require awareness of zones in the | |
2803 | * dirty-throttling and the flusher threads. | |
2804 | */ | |
c9ab0c4f | 2805 | if (ac->spread_dirty_pages && !zone_dirty_ok(zone)) |
800a1e75 | 2806 | continue; |
7fb1d9fc | 2807 | |
e085dbc5 | 2808 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
48ee5f36 | 2809 | if (!zone_watermark_fast(zone, order, mark, |
93ea9964 | 2810 | ac_classzone_idx(ac), alloc_flags)) { |
fa5e084e MG |
2811 | int ret; |
2812 | ||
5dab2911 MG |
2813 | /* Checked here to keep the fast path fast */ |
2814 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
2815 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
2816 | goto try_this_zone; | |
2817 | ||
957f822a | 2818 | if (zone_reclaim_mode == 0 || |
c33d6c06 | 2819 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
cd38b115 MG |
2820 | continue; |
2821 | ||
fa5e084e MG |
2822 | ret = zone_reclaim(zone, gfp_mask, order); |
2823 | switch (ret) { | |
2824 | case ZONE_RECLAIM_NOSCAN: | |
2825 | /* did not scan */ | |
cd38b115 | 2826 | continue; |
fa5e084e MG |
2827 | case ZONE_RECLAIM_FULL: |
2828 | /* scanned but unreclaimable */ | |
cd38b115 | 2829 | continue; |
fa5e084e MG |
2830 | default: |
2831 | /* did we reclaim enough */ | |
fed2719e | 2832 | if (zone_watermark_ok(zone, order, mark, |
93ea9964 | 2833 | ac_classzone_idx(ac), alloc_flags)) |
fed2719e MG |
2834 | goto try_this_zone; |
2835 | ||
fed2719e | 2836 | continue; |
0798e519 | 2837 | } |
7fb1d9fc RS |
2838 | } |
2839 | ||
fa5e084e | 2840 | try_this_zone: |
c33d6c06 | 2841 | page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order, |
0aaa29a5 | 2842 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 VB |
2843 | if (page) { |
2844 | if (prep_new_page(page, order, gfp_mask, alloc_flags)) | |
2845 | goto try_this_zone; | |
0aaa29a5 MG |
2846 | |
2847 | /* | |
2848 | * If this is a high-order atomic allocation then check | |
2849 | * if the pageblock should be reserved for the future | |
2850 | */ | |
2851 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
2852 | reserve_highatomic_pageblock(page, zone, order); | |
2853 | ||
75379191 VB |
2854 | return page; |
2855 | } | |
54a6eb5c | 2856 | } |
9276b1bc | 2857 | |
4ffeaf35 MG |
2858 | /* |
2859 | * The first pass makes sure allocations are spread fairly within the | |
2860 | * local node. However, the local node might have free pages left | |
2861 | * after the fairness batches are exhausted, and remote zones haven't | |
2862 | * even been considered yet. Try once more without fairness, and | |
2863 | * include remote zones now, before entering the slowpath and waking | |
2864 | * kswapd: prefer spilling to a remote zone over swapping locally. | |
2865 | */ | |
30534755 MG |
2866 | if (fair_skipped) { |
2867 | reset_fair: | |
2868 | apply_fair = false; | |
2869 | fair_skipped = false; | |
c33d6c06 | 2870 | reset_alloc_batches(ac->preferred_zoneref->zone); |
4ffeaf35 | 2871 | goto zonelist_scan; |
30534755 | 2872 | } |
4ffeaf35 MG |
2873 | |
2874 | return NULL; | |
753ee728 MH |
2875 | } |
2876 | ||
29423e77 DR |
2877 | /* |
2878 | * Large machines with many possible nodes should not always dump per-node | |
2879 | * meminfo in irq context. | |
2880 | */ | |
2881 | static inline bool should_suppress_show_mem(void) | |
2882 | { | |
2883 | bool ret = false; | |
2884 | ||
2885 | #if NODES_SHIFT > 8 | |
2886 | ret = in_interrupt(); | |
2887 | #endif | |
2888 | return ret; | |
2889 | } | |
2890 | ||
a238ab5b DH |
2891 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2892 | DEFAULT_RATELIMIT_INTERVAL, | |
2893 | DEFAULT_RATELIMIT_BURST); | |
2894 | ||
d00181b9 | 2895 | void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...) |
a238ab5b | 2896 | { |
a238ab5b DH |
2897 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2898 | ||
c0a32fc5 SG |
2899 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2900 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2901 | return; |
2902 | ||
2903 | /* | |
2904 | * This documents exceptions given to allocations in certain | |
2905 | * contexts that are allowed to allocate outside current's set | |
2906 | * of allowed nodes. | |
2907 | */ | |
2908 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2909 | if (test_thread_flag(TIF_MEMDIE) || | |
2910 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2911 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 2912 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
2913 | filter &= ~SHOW_MEM_FILTER_NODES; |
2914 | ||
2915 | if (fmt) { | |
3ee9a4f0 JP |
2916 | struct va_format vaf; |
2917 | va_list args; | |
2918 | ||
a238ab5b | 2919 | va_start(args, fmt); |
3ee9a4f0 JP |
2920 | |
2921 | vaf.fmt = fmt; | |
2922 | vaf.va = &args; | |
2923 | ||
2924 | pr_warn("%pV", &vaf); | |
2925 | ||
a238ab5b DH |
2926 | va_end(args); |
2927 | } | |
2928 | ||
c5c990e8 VB |
2929 | pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n", |
2930 | current->comm, order, gfp_mask, &gfp_mask); | |
a238ab5b DH |
2931 | dump_stack(); |
2932 | if (!should_suppress_show_mem()) | |
2933 | show_mem(filter); | |
2934 | } | |
2935 | ||
11e33f6a MG |
2936 | static inline struct page * |
2937 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 2938 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 2939 | { |
6e0fc46d DR |
2940 | struct oom_control oc = { |
2941 | .zonelist = ac->zonelist, | |
2942 | .nodemask = ac->nodemask, | |
2943 | .gfp_mask = gfp_mask, | |
2944 | .order = order, | |
6e0fc46d | 2945 | }; |
11e33f6a MG |
2946 | struct page *page; |
2947 | ||
9879de73 JW |
2948 | *did_some_progress = 0; |
2949 | ||
9879de73 | 2950 | /* |
dc56401f JW |
2951 | * Acquire the oom lock. If that fails, somebody else is |
2952 | * making progress for us. | |
9879de73 | 2953 | */ |
dc56401f | 2954 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 2955 | *did_some_progress = 1; |
11e33f6a | 2956 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2957 | return NULL; |
2958 | } | |
6b1de916 | 2959 | |
11e33f6a MG |
2960 | /* |
2961 | * Go through the zonelist yet one more time, keep very high watermark | |
2962 | * here, this is only to catch a parallel oom killing, we must fail if | |
2963 | * we're still under heavy pressure. | |
2964 | */ | |
a9263751 VB |
2965 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
2966 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 2967 | if (page) |
11e33f6a MG |
2968 | goto out; |
2969 | ||
4365a567 | 2970 | if (!(gfp_mask & __GFP_NOFAIL)) { |
9879de73 JW |
2971 | /* Coredumps can quickly deplete all memory reserves */ |
2972 | if (current->flags & PF_DUMPCORE) | |
2973 | goto out; | |
4365a567 KH |
2974 | /* The OOM killer will not help higher order allocs */ |
2975 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2976 | goto out; | |
03668b3c | 2977 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
a9263751 | 2978 | if (ac->high_zoneidx < ZONE_NORMAL) |
03668b3c | 2979 | goto out; |
9083905a JW |
2980 | if (pm_suspended_storage()) |
2981 | goto out; | |
3da88fb3 MH |
2982 | /* |
2983 | * XXX: GFP_NOFS allocations should rather fail than rely on | |
2984 | * other request to make a forward progress. | |
2985 | * We are in an unfortunate situation where out_of_memory cannot | |
2986 | * do much for this context but let's try it to at least get | |
2987 | * access to memory reserved if the current task is killed (see | |
2988 | * out_of_memory). Once filesystems are ready to handle allocation | |
2989 | * failures more gracefully we should just bail out here. | |
2990 | */ | |
2991 | ||
4167e9b2 | 2992 | /* The OOM killer may not free memory on a specific node */ |
4365a567 KH |
2993 | if (gfp_mask & __GFP_THISNODE) |
2994 | goto out; | |
2995 | } | |
11e33f6a | 2996 | /* Exhausted what can be done so it's blamo time */ |
5020e285 | 2997 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 2998 | *did_some_progress = 1; |
5020e285 MH |
2999 | |
3000 | if (gfp_mask & __GFP_NOFAIL) { | |
3001 | page = get_page_from_freelist(gfp_mask, order, | |
3002 | ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac); | |
3003 | /* | |
3004 | * fallback to ignore cpuset restriction if our nodes | |
3005 | * are depleted | |
3006 | */ | |
3007 | if (!page) | |
3008 | page = get_page_from_freelist(gfp_mask, order, | |
3009 | ALLOC_NO_WATERMARKS, ac); | |
3010 | } | |
3011 | } | |
11e33f6a | 3012 | out: |
dc56401f | 3013 | mutex_unlock(&oom_lock); |
11e33f6a MG |
3014 | return page; |
3015 | } | |
3016 | ||
56de7263 MG |
3017 | #ifdef CONFIG_COMPACTION |
3018 | /* Try memory compaction for high-order allocations before reclaim */ | |
3019 | static struct page * | |
3020 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3021 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 VB |
3022 | enum migrate_mode mode, int *contended_compaction, |
3023 | bool *deferred_compaction) | |
56de7263 | 3024 | { |
53853e2d | 3025 | unsigned long compact_result; |
98dd3b48 | 3026 | struct page *page; |
53853e2d VB |
3027 | |
3028 | if (!order) | |
66199712 | 3029 | return NULL; |
66199712 | 3030 | |
c06b1fca | 3031 | current->flags |= PF_MEMALLOC; |
1a6d53a1 VB |
3032 | compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
3033 | mode, contended_compaction); | |
c06b1fca | 3034 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 3035 | |
98dd3b48 VB |
3036 | switch (compact_result) { |
3037 | case COMPACT_DEFERRED: | |
53853e2d | 3038 | *deferred_compaction = true; |
98dd3b48 VB |
3039 | /* fall-through */ |
3040 | case COMPACT_SKIPPED: | |
3041 | return NULL; | |
3042 | default: | |
3043 | break; | |
3044 | } | |
53853e2d | 3045 | |
98dd3b48 VB |
3046 | /* |
3047 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
3048 | * count a compaction stall | |
3049 | */ | |
3050 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 3051 | |
a9263751 VB |
3052 | page = get_page_from_freelist(gfp_mask, order, |
3053 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
53853e2d | 3054 | |
98dd3b48 VB |
3055 | if (page) { |
3056 | struct zone *zone = page_zone(page); | |
53853e2d | 3057 | |
98dd3b48 VB |
3058 | zone->compact_blockskip_flush = false; |
3059 | compaction_defer_reset(zone, order, true); | |
3060 | count_vm_event(COMPACTSUCCESS); | |
3061 | return page; | |
3062 | } | |
56de7263 | 3063 | |
98dd3b48 VB |
3064 | /* |
3065 | * It's bad if compaction run occurs and fails. The most likely reason | |
3066 | * is that pages exist, but not enough to satisfy watermarks. | |
3067 | */ | |
3068 | count_vm_event(COMPACTFAIL); | |
66199712 | 3069 | |
98dd3b48 | 3070 | cond_resched(); |
56de7263 MG |
3071 | |
3072 | return NULL; | |
3073 | } | |
3074 | #else | |
3075 | static inline struct page * | |
3076 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3077 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 VB |
3078 | enum migrate_mode mode, int *contended_compaction, |
3079 | bool *deferred_compaction) | |
56de7263 MG |
3080 | { |
3081 | return NULL; | |
3082 | } | |
3083 | #endif /* CONFIG_COMPACTION */ | |
3084 | ||
bba90710 MS |
3085 | /* Perform direct synchronous page reclaim */ |
3086 | static int | |
a9263751 VB |
3087 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3088 | const struct alloc_context *ac) | |
11e33f6a | 3089 | { |
11e33f6a | 3090 | struct reclaim_state reclaim_state; |
bba90710 | 3091 | int progress; |
11e33f6a MG |
3092 | |
3093 | cond_resched(); | |
3094 | ||
3095 | /* We now go into synchronous reclaim */ | |
3096 | cpuset_memory_pressure_bump(); | |
c06b1fca | 3097 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
3098 | lockdep_set_current_reclaim_state(gfp_mask); |
3099 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 3100 | current->reclaim_state = &reclaim_state; |
11e33f6a | 3101 | |
a9263751 VB |
3102 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3103 | ac->nodemask); | |
11e33f6a | 3104 | |
c06b1fca | 3105 | current->reclaim_state = NULL; |
11e33f6a | 3106 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 3107 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
3108 | |
3109 | cond_resched(); | |
3110 | ||
bba90710 MS |
3111 | return progress; |
3112 | } | |
3113 | ||
3114 | /* The really slow allocator path where we enter direct reclaim */ | |
3115 | static inline struct page * | |
3116 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
c603844b | 3117 | unsigned int alloc_flags, const struct alloc_context *ac, |
a9263751 | 3118 | unsigned long *did_some_progress) |
bba90710 MS |
3119 | { |
3120 | struct page *page = NULL; | |
3121 | bool drained = false; | |
3122 | ||
a9263751 | 3123 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3124 | if (unlikely(!(*did_some_progress))) |
3125 | return NULL; | |
11e33f6a | 3126 | |
9ee493ce | 3127 | retry: |
a9263751 VB |
3128 | page = get_page_from_freelist(gfp_mask, order, |
3129 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
9ee493ce MG |
3130 | |
3131 | /* | |
3132 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3133 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3134 | * Shrink them them and try again | |
9ee493ce MG |
3135 | */ |
3136 | if (!page && !drained) { | |
0aaa29a5 | 3137 | unreserve_highatomic_pageblock(ac); |
93481ff0 | 3138 | drain_all_pages(NULL); |
9ee493ce MG |
3139 | drained = true; |
3140 | goto retry; | |
3141 | } | |
3142 | ||
11e33f6a MG |
3143 | return page; |
3144 | } | |
3145 | ||
a9263751 | 3146 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
3147 | { |
3148 | struct zoneref *z; | |
3149 | struct zone *zone; | |
3150 | ||
a9263751 VB |
3151 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3152 | ac->high_zoneidx, ac->nodemask) | |
93ea9964 | 3153 | wakeup_kswapd(zone, order, ac_classzone_idx(ac)); |
3a025760 JW |
3154 | } |
3155 | ||
c603844b | 3156 | static inline unsigned int |
341ce06f PZ |
3157 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3158 | { | |
c603844b | 3159 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3160 | |
a56f57ff | 3161 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3162 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3163 | |
341ce06f PZ |
3164 | /* |
3165 | * The caller may dip into page reserves a bit more if the caller | |
3166 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3167 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3168 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3169 | */ |
e6223a3b | 3170 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3171 | |
d0164adc | 3172 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3173 | /* |
b104a35d DR |
3174 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3175 | * if it can't schedule. | |
5c3240d9 | 3176 | */ |
b104a35d | 3177 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3178 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3179 | /* |
b104a35d | 3180 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3181 | * comment for __cpuset_node_allowed(). |
523b9458 | 3182 | */ |
341ce06f | 3183 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3184 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3185 | alloc_flags |= ALLOC_HARDER; |
3186 | ||
b37f1dd0 MG |
3187 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
3188 | if (gfp_mask & __GFP_MEMALLOC) | |
3189 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
3190 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
3191 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
3192 | else if (!in_interrupt() && | |
3193 | ((current->flags & PF_MEMALLOC) || | |
3194 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 3195 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 3196 | } |
d95ea5d1 | 3197 | #ifdef CONFIG_CMA |
43e7a34d | 3198 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
3199 | alloc_flags |= ALLOC_CMA; |
3200 | #endif | |
341ce06f PZ |
3201 | return alloc_flags; |
3202 | } | |
3203 | ||
072bb0aa MG |
3204 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3205 | { | |
b37f1dd0 | 3206 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
3207 | } |
3208 | ||
d0164adc MG |
3209 | static inline bool is_thp_gfp_mask(gfp_t gfp_mask) |
3210 | { | |
3211 | return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE; | |
3212 | } | |
3213 | ||
11e33f6a MG |
3214 | static inline struct page * |
3215 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3216 | struct alloc_context *ac) |
11e33f6a | 3217 | { |
d0164adc | 3218 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
11e33f6a | 3219 | struct page *page = NULL; |
c603844b | 3220 | unsigned int alloc_flags; |
11e33f6a MG |
3221 | unsigned long pages_reclaimed = 0; |
3222 | unsigned long did_some_progress; | |
e0b9daeb | 3223 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
66199712 | 3224 | bool deferred_compaction = false; |
1f9efdef | 3225 | int contended_compaction = COMPACT_CONTENDED_NONE; |
1da177e4 | 3226 | |
72807a74 MG |
3227 | /* |
3228 | * In the slowpath, we sanity check order to avoid ever trying to | |
3229 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
3230 | * be using allocators in order of preference for an area that is | |
3231 | * too large. | |
3232 | */ | |
1fc28b70 MG |
3233 | if (order >= MAX_ORDER) { |
3234 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 3235 | return NULL; |
1fc28b70 | 3236 | } |
1da177e4 | 3237 | |
d0164adc MG |
3238 | /* |
3239 | * We also sanity check to catch abuse of atomic reserves being used by | |
3240 | * callers that are not in atomic context. | |
3241 | */ | |
3242 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
3243 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
3244 | gfp_mask &= ~__GFP_ATOMIC; | |
3245 | ||
9879de73 | 3246 | retry: |
d0164adc | 3247 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
a9263751 | 3248 | wake_all_kswapds(order, ac); |
1da177e4 | 3249 | |
9bf2229f | 3250 | /* |
7fb1d9fc RS |
3251 | * OK, we're below the kswapd watermark and have kicked background |
3252 | * reclaim. Now things get more complex, so set up alloc_flags according | |
3253 | * to how we want to proceed. | |
9bf2229f | 3254 | */ |
341ce06f | 3255 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 3256 | |
341ce06f | 3257 | /* This is the last chance, in general, before the goto nopage. */ |
a9263751 VB |
3258 | page = get_page_from_freelist(gfp_mask, order, |
3259 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
7fb1d9fc RS |
3260 | if (page) |
3261 | goto got_pg; | |
1da177e4 | 3262 | |
11e33f6a | 3263 | /* Allocate without watermarks if the context allows */ |
341ce06f | 3264 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
3265 | /* |
3266 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
3267 | * the allocation is high priority and these type of | |
3268 | * allocations are system rather than user orientated | |
3269 | */ | |
a9263751 | 3270 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
33d53103 MH |
3271 | page = get_page_from_freelist(gfp_mask, order, |
3272 | ALLOC_NO_WATERMARKS, ac); | |
3273 | if (page) | |
3274 | goto got_pg; | |
1da177e4 LT |
3275 | } |
3276 | ||
d0164adc MG |
3277 | /* Caller is not willing to reclaim, we can't balance anything */ |
3278 | if (!can_direct_reclaim) { | |
aed0a0e3 | 3279 | /* |
33d53103 MH |
3280 | * All existing users of the __GFP_NOFAIL are blockable, so warn |
3281 | * of any new users that actually allow this type of allocation | |
3282 | * to fail. | |
aed0a0e3 DR |
3283 | */ |
3284 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 3285 | goto nopage; |
aed0a0e3 | 3286 | } |
1da177e4 | 3287 | |
341ce06f | 3288 | /* Avoid recursion of direct reclaim */ |
33d53103 MH |
3289 | if (current->flags & PF_MEMALLOC) { |
3290 | /* | |
3291 | * __GFP_NOFAIL request from this context is rather bizarre | |
3292 | * because we cannot reclaim anything and only can loop waiting | |
3293 | * for somebody to do a work for us. | |
3294 | */ | |
3295 | if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { | |
3296 | cond_resched(); | |
3297 | goto retry; | |
3298 | } | |
341ce06f | 3299 | goto nopage; |
33d53103 | 3300 | } |
341ce06f | 3301 | |
6583bb64 DR |
3302 | /* Avoid allocations with no watermarks from looping endlessly */ |
3303 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
3304 | goto nopage; | |
3305 | ||
77f1fe6b MG |
3306 | /* |
3307 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
3308 | * attempts after direct reclaim are synchronous | |
3309 | */ | |
a9263751 VB |
3310 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
3311 | migration_mode, | |
3312 | &contended_compaction, | |
53853e2d | 3313 | &deferred_compaction); |
56de7263 MG |
3314 | if (page) |
3315 | goto got_pg; | |
75f30861 | 3316 | |
1f9efdef | 3317 | /* Checks for THP-specific high-order allocations */ |
d0164adc | 3318 | if (is_thp_gfp_mask(gfp_mask)) { |
1f9efdef VB |
3319 | /* |
3320 | * If compaction is deferred for high-order allocations, it is | |
3321 | * because sync compaction recently failed. If this is the case | |
3322 | * and the caller requested a THP allocation, we do not want | |
3323 | * to heavily disrupt the system, so we fail the allocation | |
3324 | * instead of entering direct reclaim. | |
3325 | */ | |
3326 | if (deferred_compaction) | |
3327 | goto nopage; | |
3328 | ||
3329 | /* | |
3330 | * In all zones where compaction was attempted (and not | |
3331 | * deferred or skipped), lock contention has been detected. | |
3332 | * For THP allocation we do not want to disrupt the others | |
3333 | * so we fallback to base pages instead. | |
3334 | */ | |
3335 | if (contended_compaction == COMPACT_CONTENDED_LOCK) | |
3336 | goto nopage; | |
3337 | ||
3338 | /* | |
3339 | * If compaction was aborted due to need_resched(), we do not | |
3340 | * want to further increase allocation latency, unless it is | |
3341 | * khugepaged trying to collapse. | |
3342 | */ | |
3343 | if (contended_compaction == COMPACT_CONTENDED_SCHED | |
3344 | && !(current->flags & PF_KTHREAD)) | |
3345 | goto nopage; | |
3346 | } | |
66199712 | 3347 | |
8fe78048 DR |
3348 | /* |
3349 | * It can become very expensive to allocate transparent hugepages at | |
3350 | * fault, so use asynchronous memory compaction for THP unless it is | |
3351 | * khugepaged trying to collapse. | |
3352 | */ | |
d0164adc | 3353 | if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD)) |
8fe78048 DR |
3354 | migration_mode = MIGRATE_SYNC_LIGHT; |
3355 | ||
11e33f6a | 3356 | /* Try direct reclaim and then allocating */ |
a9263751 VB |
3357 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, |
3358 | &did_some_progress); | |
11e33f6a MG |
3359 | if (page) |
3360 | goto got_pg; | |
1da177e4 | 3361 | |
9083905a JW |
3362 | /* Do not loop if specifically requested */ |
3363 | if (gfp_mask & __GFP_NORETRY) | |
3364 | goto noretry; | |
3365 | ||
3366 | /* Keep reclaiming pages as long as there is reasonable progress */ | |
a41f24ea | 3367 | pages_reclaimed += did_some_progress; |
9083905a JW |
3368 | if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) || |
3369 | ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) { | |
11e33f6a | 3370 | /* Wait for some write requests to complete then retry */ |
c33d6c06 | 3371 | wait_iff_congested(ac->preferred_zoneref->zone, BLK_RW_ASYNC, HZ/50); |
9879de73 | 3372 | goto retry; |
1da177e4 LT |
3373 | } |
3374 | ||
9083905a JW |
3375 | /* Reclaim has failed us, start killing things */ |
3376 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
3377 | if (page) | |
3378 | goto got_pg; | |
3379 | ||
3380 | /* Retry as long as the OOM killer is making progress */ | |
3381 | if (did_some_progress) | |
3382 | goto retry; | |
3383 | ||
3384 | noretry: | |
3385 | /* | |
3386 | * High-order allocations do not necessarily loop after | |
3387 | * direct reclaim and reclaim/compaction depends on compaction | |
3388 | * being called after reclaim so call directly if necessary | |
3389 | */ | |
3390 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, | |
3391 | ac, migration_mode, | |
3392 | &contended_compaction, | |
3393 | &deferred_compaction); | |
3394 | if (page) | |
3395 | goto got_pg; | |
1da177e4 | 3396 | nopage: |
a238ab5b | 3397 | warn_alloc_failed(gfp_mask, order, NULL); |
1da177e4 | 3398 | got_pg: |
072bb0aa | 3399 | return page; |
1da177e4 | 3400 | } |
11e33f6a MG |
3401 | |
3402 | /* | |
3403 | * This is the 'heart' of the zoned buddy allocator. | |
3404 | */ | |
3405 | struct page * | |
3406 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
3407 | struct zonelist *zonelist, nodemask_t *nodemask) | |
3408 | { | |
5bb1b169 | 3409 | struct page *page; |
cc9a6c87 | 3410 | unsigned int cpuset_mems_cookie; |
c603844b | 3411 | unsigned int alloc_flags = ALLOC_WMARK_LOW|ALLOC_FAIR; |
83d4ca81 | 3412 | gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */ |
a9263751 VB |
3413 | struct alloc_context ac = { |
3414 | .high_zoneidx = gfp_zone(gfp_mask), | |
682a3385 | 3415 | .zonelist = zonelist, |
a9263751 VB |
3416 | .nodemask = nodemask, |
3417 | .migratetype = gfpflags_to_migratetype(gfp_mask), | |
3418 | }; | |
11e33f6a | 3419 | |
682a3385 | 3420 | if (cpusets_enabled()) { |
83d4ca81 | 3421 | alloc_mask |= __GFP_HARDWALL; |
682a3385 MG |
3422 | alloc_flags |= ALLOC_CPUSET; |
3423 | if (!ac.nodemask) | |
3424 | ac.nodemask = &cpuset_current_mems_allowed; | |
3425 | } | |
3426 | ||
dcce284a BH |
3427 | gfp_mask &= gfp_allowed_mask; |
3428 | ||
11e33f6a MG |
3429 | lockdep_trace_alloc(gfp_mask); |
3430 | ||
d0164adc | 3431 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
3432 | |
3433 | if (should_fail_alloc_page(gfp_mask, order)) | |
3434 | return NULL; | |
3435 | ||
3436 | /* | |
3437 | * Check the zones suitable for the gfp_mask contain at least one | |
3438 | * valid zone. It's possible to have an empty zonelist as a result | |
4167e9b2 | 3439 | * of __GFP_THISNODE and a memoryless node |
11e33f6a MG |
3440 | */ |
3441 | if (unlikely(!zonelist->_zonerefs->zone)) | |
3442 | return NULL; | |
3443 | ||
a9263751 | 3444 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
21bb9bd1 VB |
3445 | alloc_flags |= ALLOC_CMA; |
3446 | ||
cc9a6c87 | 3447 | retry_cpuset: |
d26914d1 | 3448 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3449 | |
c9ab0c4f MG |
3450 | /* Dirty zone balancing only done in the fast path */ |
3451 | ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE); | |
3452 | ||
5117f45d | 3453 | /* The preferred zone is used for statistics later */ |
c33d6c06 MG |
3454 | ac.preferred_zoneref = first_zones_zonelist(ac.zonelist, |
3455 | ac.high_zoneidx, ac.nodemask); | |
3456 | if (!ac.preferred_zoneref) { | |
5bb1b169 | 3457 | page = NULL; |
4fcb0971 | 3458 | goto no_zone; |
5bb1b169 MG |
3459 | } |
3460 | ||
5117f45d | 3461 | /* First allocation attempt */ |
a9263751 | 3462 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4fcb0971 MG |
3463 | if (likely(page)) |
3464 | goto out; | |
11e33f6a | 3465 | |
4fcb0971 MG |
3466 | /* |
3467 | * Runtime PM, block IO and its error handling path can deadlock | |
3468 | * because I/O on the device might not complete. | |
3469 | */ | |
3470 | alloc_mask = memalloc_noio_flags(gfp_mask); | |
3471 | ac.spread_dirty_pages = false; | |
23f086f9 | 3472 | |
4fcb0971 | 3473 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
cc9a6c87 | 3474 | |
4fcb0971 | 3475 | no_zone: |
cc9a6c87 MG |
3476 | /* |
3477 | * When updating a task's mems_allowed, it is possible to race with | |
3478 | * parallel threads in such a way that an allocation can fail while | |
3479 | * the mask is being updated. If a page allocation is about to fail, | |
3480 | * check if the cpuset changed during allocation and if so, retry. | |
3481 | */ | |
83d4ca81 MG |
3482 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) { |
3483 | alloc_mask = gfp_mask; | |
cc9a6c87 | 3484 | goto retry_cpuset; |
83d4ca81 | 3485 | } |
cc9a6c87 | 3486 | |
4fcb0971 MG |
3487 | out: |
3488 | if (kmemcheck_enabled && page) | |
3489 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
3490 | ||
3491 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); | |
3492 | ||
11e33f6a | 3493 | return page; |
1da177e4 | 3494 | } |
d239171e | 3495 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
3496 | |
3497 | /* | |
3498 | * Common helper functions. | |
3499 | */ | |
920c7a5d | 3500 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 3501 | { |
945a1113 AM |
3502 | struct page *page; |
3503 | ||
3504 | /* | |
3505 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
3506 | * a highmem page | |
3507 | */ | |
3508 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
3509 | ||
1da177e4 LT |
3510 | page = alloc_pages(gfp_mask, order); |
3511 | if (!page) | |
3512 | return 0; | |
3513 | return (unsigned long) page_address(page); | |
3514 | } | |
1da177e4 LT |
3515 | EXPORT_SYMBOL(__get_free_pages); |
3516 | ||
920c7a5d | 3517 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 3518 | { |
945a1113 | 3519 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 3520 | } |
1da177e4 LT |
3521 | EXPORT_SYMBOL(get_zeroed_page); |
3522 | ||
920c7a5d | 3523 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 3524 | { |
b5810039 | 3525 | if (put_page_testzero(page)) { |
1da177e4 | 3526 | if (order == 0) |
b745bc85 | 3527 | free_hot_cold_page(page, false); |
1da177e4 LT |
3528 | else |
3529 | __free_pages_ok(page, order); | |
3530 | } | |
3531 | } | |
3532 | ||
3533 | EXPORT_SYMBOL(__free_pages); | |
3534 | ||
920c7a5d | 3535 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
3536 | { |
3537 | if (addr != 0) { | |
725d704e | 3538 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
3539 | __free_pages(virt_to_page((void *)addr), order); |
3540 | } | |
3541 | } | |
3542 | ||
3543 | EXPORT_SYMBOL(free_pages); | |
3544 | ||
b63ae8ca AD |
3545 | /* |
3546 | * Page Fragment: | |
3547 | * An arbitrary-length arbitrary-offset area of memory which resides | |
3548 | * within a 0 or higher order page. Multiple fragments within that page | |
3549 | * are individually refcounted, in the page's reference counter. | |
3550 | * | |
3551 | * The page_frag functions below provide a simple allocation framework for | |
3552 | * page fragments. This is used by the network stack and network device | |
3553 | * drivers to provide a backing region of memory for use as either an | |
3554 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
3555 | */ | |
3556 | static struct page *__page_frag_refill(struct page_frag_cache *nc, | |
3557 | gfp_t gfp_mask) | |
3558 | { | |
3559 | struct page *page = NULL; | |
3560 | gfp_t gfp = gfp_mask; | |
3561 | ||
3562 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3563 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
3564 | __GFP_NOMEMALLOC; | |
3565 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
3566 | PAGE_FRAG_CACHE_MAX_ORDER); | |
3567 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
3568 | #endif | |
3569 | if (unlikely(!page)) | |
3570 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
3571 | ||
3572 | nc->va = page ? page_address(page) : NULL; | |
3573 | ||
3574 | return page; | |
3575 | } | |
3576 | ||
3577 | void *__alloc_page_frag(struct page_frag_cache *nc, | |
3578 | unsigned int fragsz, gfp_t gfp_mask) | |
3579 | { | |
3580 | unsigned int size = PAGE_SIZE; | |
3581 | struct page *page; | |
3582 | int offset; | |
3583 | ||
3584 | if (unlikely(!nc->va)) { | |
3585 | refill: | |
3586 | page = __page_frag_refill(nc, gfp_mask); | |
3587 | if (!page) | |
3588 | return NULL; | |
3589 | ||
3590 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3591 | /* if size can vary use size else just use PAGE_SIZE */ | |
3592 | size = nc->size; | |
3593 | #endif | |
3594 | /* Even if we own the page, we do not use atomic_set(). | |
3595 | * This would break get_page_unless_zero() users. | |
3596 | */ | |
fe896d18 | 3597 | page_ref_add(page, size - 1); |
b63ae8ca AD |
3598 | |
3599 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 3600 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
3601 | nc->pagecnt_bias = size; |
3602 | nc->offset = size; | |
3603 | } | |
3604 | ||
3605 | offset = nc->offset - fragsz; | |
3606 | if (unlikely(offset < 0)) { | |
3607 | page = virt_to_page(nc->va); | |
3608 | ||
fe896d18 | 3609 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
3610 | goto refill; |
3611 | ||
3612 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3613 | /* if size can vary use size else just use PAGE_SIZE */ | |
3614 | size = nc->size; | |
3615 | #endif | |
3616 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 3617 | set_page_count(page, size); |
b63ae8ca AD |
3618 | |
3619 | /* reset page count bias and offset to start of new frag */ | |
3620 | nc->pagecnt_bias = size; | |
3621 | offset = size - fragsz; | |
3622 | } | |
3623 | ||
3624 | nc->pagecnt_bias--; | |
3625 | nc->offset = offset; | |
3626 | ||
3627 | return nc->va + offset; | |
3628 | } | |
3629 | EXPORT_SYMBOL(__alloc_page_frag); | |
3630 | ||
3631 | /* | |
3632 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
3633 | */ | |
3634 | void __free_page_frag(void *addr) | |
3635 | { | |
3636 | struct page *page = virt_to_head_page(addr); | |
3637 | ||
3638 | if (unlikely(put_page_testzero(page))) | |
3639 | __free_pages_ok(page, compound_order(page)); | |
3640 | } | |
3641 | EXPORT_SYMBOL(__free_page_frag); | |
3642 | ||
6a1a0d3b | 3643 | /* |
52383431 | 3644 | * alloc_kmem_pages charges newly allocated pages to the kmem resource counter |
a9bb7e62 VD |
3645 | * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is |
3646 | * equivalent to alloc_pages. | |
6a1a0d3b | 3647 | * |
52383431 VD |
3648 | * It should be used when the caller would like to use kmalloc, but since the |
3649 | * allocation is large, it has to fall back to the page allocator. | |
3650 | */ | |
3651 | struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order) | |
3652 | { | |
3653 | struct page *page; | |
52383431 | 3654 | |
52383431 | 3655 | page = alloc_pages(gfp_mask, order); |
d05e83a6 VD |
3656 | if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) { |
3657 | __free_pages(page, order); | |
3658 | page = NULL; | |
3659 | } | |
52383431 VD |
3660 | return page; |
3661 | } | |
3662 | ||
3663 | struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
3664 | { | |
3665 | struct page *page; | |
52383431 | 3666 | |
52383431 | 3667 | page = alloc_pages_node(nid, gfp_mask, order); |
d05e83a6 VD |
3668 | if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) { |
3669 | __free_pages(page, order); | |
3670 | page = NULL; | |
3671 | } | |
52383431 VD |
3672 | return page; |
3673 | } | |
3674 | ||
3675 | /* | |
3676 | * __free_kmem_pages and free_kmem_pages will free pages allocated with | |
3677 | * alloc_kmem_pages. | |
6a1a0d3b | 3678 | */ |
52383431 | 3679 | void __free_kmem_pages(struct page *page, unsigned int order) |
6a1a0d3b | 3680 | { |
d05e83a6 | 3681 | memcg_kmem_uncharge(page, order); |
6a1a0d3b GC |
3682 | __free_pages(page, order); |
3683 | } | |
3684 | ||
52383431 | 3685 | void free_kmem_pages(unsigned long addr, unsigned int order) |
6a1a0d3b GC |
3686 | { |
3687 | if (addr != 0) { | |
3688 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
52383431 | 3689 | __free_kmem_pages(virt_to_page((void *)addr), order); |
6a1a0d3b GC |
3690 | } |
3691 | } | |
3692 | ||
d00181b9 KS |
3693 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
3694 | size_t size) | |
ee85c2e1 AK |
3695 | { |
3696 | if (addr) { | |
3697 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
3698 | unsigned long used = addr + PAGE_ALIGN(size); | |
3699 | ||
3700 | split_page(virt_to_page((void *)addr), order); | |
3701 | while (used < alloc_end) { | |
3702 | free_page(used); | |
3703 | used += PAGE_SIZE; | |
3704 | } | |
3705 | } | |
3706 | return (void *)addr; | |
3707 | } | |
3708 | ||
2be0ffe2 TT |
3709 | /** |
3710 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
3711 | * @size: the number of bytes to allocate | |
3712 | * @gfp_mask: GFP flags for the allocation | |
3713 | * | |
3714 | * This function is similar to alloc_pages(), except that it allocates the | |
3715 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
3716 | * allocate memory in power-of-two pages. | |
3717 | * | |
3718 | * This function is also limited by MAX_ORDER. | |
3719 | * | |
3720 | * Memory allocated by this function must be released by free_pages_exact(). | |
3721 | */ | |
3722 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
3723 | { | |
3724 | unsigned int order = get_order(size); | |
3725 | unsigned long addr; | |
3726 | ||
3727 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 3728 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
3729 | } |
3730 | EXPORT_SYMBOL(alloc_pages_exact); | |
3731 | ||
ee85c2e1 AK |
3732 | /** |
3733 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
3734 | * pages on a node. | |
b5e6ab58 | 3735 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
3736 | * @size: the number of bytes to allocate |
3737 | * @gfp_mask: GFP flags for the allocation | |
3738 | * | |
3739 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
3740 | * back. | |
ee85c2e1 | 3741 | */ |
e1931811 | 3742 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 3743 | { |
d00181b9 | 3744 | unsigned int order = get_order(size); |
ee85c2e1 AK |
3745 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
3746 | if (!p) | |
3747 | return NULL; | |
3748 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
3749 | } | |
ee85c2e1 | 3750 | |
2be0ffe2 TT |
3751 | /** |
3752 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
3753 | * @virt: the value returned by alloc_pages_exact. | |
3754 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
3755 | * | |
3756 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
3757 | */ | |
3758 | void free_pages_exact(void *virt, size_t size) | |
3759 | { | |
3760 | unsigned long addr = (unsigned long)virt; | |
3761 | unsigned long end = addr + PAGE_ALIGN(size); | |
3762 | ||
3763 | while (addr < end) { | |
3764 | free_page(addr); | |
3765 | addr += PAGE_SIZE; | |
3766 | } | |
3767 | } | |
3768 | EXPORT_SYMBOL(free_pages_exact); | |
3769 | ||
e0fb5815 ZY |
3770 | /** |
3771 | * nr_free_zone_pages - count number of pages beyond high watermark | |
3772 | * @offset: The zone index of the highest zone | |
3773 | * | |
3774 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
3775 | * high watermark within all zones at or below a given zone index. For each | |
3776 | * zone, the number of pages is calculated as: | |
834405c3 | 3777 | * managed_pages - high_pages |
e0fb5815 | 3778 | */ |
ebec3862 | 3779 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 3780 | { |
dd1a239f | 3781 | struct zoneref *z; |
54a6eb5c MG |
3782 | struct zone *zone; |
3783 | ||
e310fd43 | 3784 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 3785 | unsigned long sum = 0; |
1da177e4 | 3786 | |
0e88460d | 3787 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 3788 | |
54a6eb5c | 3789 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 3790 | unsigned long size = zone->managed_pages; |
41858966 | 3791 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
3792 | if (size > high) |
3793 | sum += size - high; | |
1da177e4 LT |
3794 | } |
3795 | ||
3796 | return sum; | |
3797 | } | |
3798 | ||
e0fb5815 ZY |
3799 | /** |
3800 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
3801 | * | |
3802 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
3803 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 3804 | */ |
ebec3862 | 3805 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 3806 | { |
af4ca457 | 3807 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 3808 | } |
c2f1a551 | 3809 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 3810 | |
e0fb5815 ZY |
3811 | /** |
3812 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
3813 | * | |
3814 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
3815 | * high watermark within all zones. | |
1da177e4 | 3816 | */ |
ebec3862 | 3817 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 3818 | { |
2a1e274a | 3819 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 3820 | } |
08e0f6a9 CL |
3821 | |
3822 | static inline void show_node(struct zone *zone) | |
1da177e4 | 3823 | { |
e5adfffc | 3824 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 3825 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 3826 | } |
1da177e4 | 3827 | |
d02bd27b IR |
3828 | long si_mem_available(void) |
3829 | { | |
3830 | long available; | |
3831 | unsigned long pagecache; | |
3832 | unsigned long wmark_low = 0; | |
3833 | unsigned long pages[NR_LRU_LISTS]; | |
3834 | struct zone *zone; | |
3835 | int lru; | |
3836 | ||
3837 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
3838 | pages[lru] = global_page_state(NR_LRU_BASE + lru); | |
3839 | ||
3840 | for_each_zone(zone) | |
3841 | wmark_low += zone->watermark[WMARK_LOW]; | |
3842 | ||
3843 | /* | |
3844 | * Estimate the amount of memory available for userspace allocations, | |
3845 | * without causing swapping. | |
3846 | */ | |
3847 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; | |
3848 | ||
3849 | /* | |
3850 | * Not all the page cache can be freed, otherwise the system will | |
3851 | * start swapping. Assume at least half of the page cache, or the | |
3852 | * low watermark worth of cache, needs to stay. | |
3853 | */ | |
3854 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
3855 | pagecache -= min(pagecache / 2, wmark_low); | |
3856 | available += pagecache; | |
3857 | ||
3858 | /* | |
3859 | * Part of the reclaimable slab consists of items that are in use, | |
3860 | * and cannot be freed. Cap this estimate at the low watermark. | |
3861 | */ | |
3862 | available += global_page_state(NR_SLAB_RECLAIMABLE) - | |
3863 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); | |
3864 | ||
3865 | if (available < 0) | |
3866 | available = 0; | |
3867 | return available; | |
3868 | } | |
3869 | EXPORT_SYMBOL_GPL(si_mem_available); | |
3870 | ||
1da177e4 LT |
3871 | void si_meminfo(struct sysinfo *val) |
3872 | { | |
3873 | val->totalram = totalram_pages; | |
cc7452b6 | 3874 | val->sharedram = global_page_state(NR_SHMEM); |
d23ad423 | 3875 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 3876 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
3877 | val->totalhigh = totalhigh_pages; |
3878 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
3879 | val->mem_unit = PAGE_SIZE; |
3880 | } | |
3881 | ||
3882 | EXPORT_SYMBOL(si_meminfo); | |
3883 | ||
3884 | #ifdef CONFIG_NUMA | |
3885 | void si_meminfo_node(struct sysinfo *val, int nid) | |
3886 | { | |
cdd91a77 JL |
3887 | int zone_type; /* needs to be signed */ |
3888 | unsigned long managed_pages = 0; | |
fc2bd799 JK |
3889 | unsigned long managed_highpages = 0; |
3890 | unsigned long free_highpages = 0; | |
1da177e4 LT |
3891 | pg_data_t *pgdat = NODE_DATA(nid); |
3892 | ||
cdd91a77 JL |
3893 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
3894 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
3895 | val->totalram = managed_pages; | |
cc7452b6 | 3896 | val->sharedram = node_page_state(nid, NR_SHMEM); |
d23ad423 | 3897 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 3898 | #ifdef CONFIG_HIGHMEM |
fc2bd799 JK |
3899 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
3900 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
3901 | ||
3902 | if (is_highmem(zone)) { | |
3903 | managed_highpages += zone->managed_pages; | |
3904 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); | |
3905 | } | |
3906 | } | |
3907 | val->totalhigh = managed_highpages; | |
3908 | val->freehigh = free_highpages; | |
98d2b0eb | 3909 | #else |
fc2bd799 JK |
3910 | val->totalhigh = managed_highpages; |
3911 | val->freehigh = free_highpages; | |
98d2b0eb | 3912 | #endif |
1da177e4 LT |
3913 | val->mem_unit = PAGE_SIZE; |
3914 | } | |
3915 | #endif | |
3916 | ||
ddd588b5 | 3917 | /* |
7bf02ea2 DR |
3918 | * Determine whether the node should be displayed or not, depending on whether |
3919 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 3920 | */ |
7bf02ea2 | 3921 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
3922 | { |
3923 | bool ret = false; | |
cc9a6c87 | 3924 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
3925 | |
3926 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
3927 | goto out; | |
3928 | ||
cc9a6c87 | 3929 | do { |
d26914d1 | 3930 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3931 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 3932 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
3933 | out: |
3934 | return ret; | |
3935 | } | |
3936 | ||
1da177e4 LT |
3937 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
3938 | ||
377e4f16 RV |
3939 | static void show_migration_types(unsigned char type) |
3940 | { | |
3941 | static const char types[MIGRATE_TYPES] = { | |
3942 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 3943 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
3944 | [MIGRATE_RECLAIMABLE] = 'E', |
3945 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
3946 | #ifdef CONFIG_CMA |
3947 | [MIGRATE_CMA] = 'C', | |
3948 | #endif | |
194159fb | 3949 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 3950 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 3951 | #endif |
377e4f16 RV |
3952 | }; |
3953 | char tmp[MIGRATE_TYPES + 1]; | |
3954 | char *p = tmp; | |
3955 | int i; | |
3956 | ||
3957 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
3958 | if (type & (1 << i)) | |
3959 | *p++ = types[i]; | |
3960 | } | |
3961 | ||
3962 | *p = '\0'; | |
3963 | printk("(%s) ", tmp); | |
3964 | } | |
3965 | ||
1da177e4 LT |
3966 | /* |
3967 | * Show free area list (used inside shift_scroll-lock stuff) | |
3968 | * We also calculate the percentage fragmentation. We do this by counting the | |
3969 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
3970 | * |
3971 | * Bits in @filter: | |
3972 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
3973 | * cpuset. | |
1da177e4 | 3974 | */ |
7bf02ea2 | 3975 | void show_free_areas(unsigned int filter) |
1da177e4 | 3976 | { |
d1bfcdb8 | 3977 | unsigned long free_pcp = 0; |
c7241913 | 3978 | int cpu; |
1da177e4 LT |
3979 | struct zone *zone; |
3980 | ||
ee99c71c | 3981 | for_each_populated_zone(zone) { |
7bf02ea2 | 3982 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3983 | continue; |
d1bfcdb8 | 3984 | |
761b0677 KK |
3985 | for_each_online_cpu(cpu) |
3986 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
3987 | } |
3988 | ||
a731286d KM |
3989 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
3990 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
3991 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
3992 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 3993 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 3994 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
4f98a2fe | 3995 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 3996 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
3997 | global_page_state(NR_ISOLATED_ANON), |
3998 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 3999 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 4000 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 4001 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 4002 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 4003 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 4004 | global_page_state(NR_UNSTABLE_NFS), |
3701b033 KM |
4005 | global_page_state(NR_SLAB_RECLAIMABLE), |
4006 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 4007 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 4008 | global_page_state(NR_SHMEM), |
a25700a5 | 4009 | global_page_state(NR_PAGETABLE), |
d1ce749a | 4010 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
4011 | global_page_state(NR_FREE_PAGES), |
4012 | free_pcp, | |
d1ce749a | 4013 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 4014 | |
ee99c71c | 4015 | for_each_populated_zone(zone) { |
1da177e4 LT |
4016 | int i; |
4017 | ||
7bf02ea2 | 4018 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4019 | continue; |
d1bfcdb8 KK |
4020 | |
4021 | free_pcp = 0; | |
4022 | for_each_online_cpu(cpu) | |
4023 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
4024 | ||
1da177e4 LT |
4025 | show_node(zone); |
4026 | printk("%s" | |
4027 | " free:%lukB" | |
4028 | " min:%lukB" | |
4029 | " low:%lukB" | |
4030 | " high:%lukB" | |
4f98a2fe RR |
4031 | " active_anon:%lukB" |
4032 | " inactive_anon:%lukB" | |
4033 | " active_file:%lukB" | |
4034 | " inactive_file:%lukB" | |
7b854121 | 4035 | " unevictable:%lukB" |
a731286d KM |
4036 | " isolated(anon):%lukB" |
4037 | " isolated(file):%lukB" | |
1da177e4 | 4038 | " present:%lukB" |
9feedc9d | 4039 | " managed:%lukB" |
4a0aa73f KM |
4040 | " mlocked:%lukB" |
4041 | " dirty:%lukB" | |
4042 | " writeback:%lukB" | |
4043 | " mapped:%lukB" | |
4b02108a | 4044 | " shmem:%lukB" |
4a0aa73f KM |
4045 | " slab_reclaimable:%lukB" |
4046 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 4047 | " kernel_stack:%lukB" |
4a0aa73f KM |
4048 | " pagetables:%lukB" |
4049 | " unstable:%lukB" | |
4050 | " bounce:%lukB" | |
d1bfcdb8 KK |
4051 | " free_pcp:%lukB" |
4052 | " local_pcp:%ukB" | |
d1ce749a | 4053 | " free_cma:%lukB" |
4a0aa73f | 4054 | " writeback_tmp:%lukB" |
1da177e4 LT |
4055 | " pages_scanned:%lu" |
4056 | " all_unreclaimable? %s" | |
4057 | "\n", | |
4058 | zone->name, | |
88f5acf8 | 4059 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
4060 | K(min_wmark_pages(zone)), |
4061 | K(low_wmark_pages(zone)), | |
4062 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
4063 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
4064 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
4065 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
4066 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 4067 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
4068 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
4069 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 4070 | K(zone->present_pages), |
9feedc9d | 4071 | K(zone->managed_pages), |
4a0aa73f KM |
4072 | K(zone_page_state(zone, NR_MLOCK)), |
4073 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
4074 | K(zone_page_state(zone, NR_WRITEBACK)), | |
4075 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 4076 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
4077 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
4078 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
4079 | zone_page_state(zone, NR_KERNEL_STACK) * |
4080 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
4081 | K(zone_page_state(zone, NR_PAGETABLE)), |
4082 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
4083 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1bfcdb8 KK |
4084 | K(free_pcp), |
4085 | K(this_cpu_read(zone->pageset->pcp.count)), | |
d1ce749a | 4086 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 4087 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
0d5d823a | 4088 | K(zone_page_state(zone, NR_PAGES_SCANNED)), |
6e543d57 | 4089 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
4090 | ); |
4091 | printk("lowmem_reserve[]:"); | |
4092 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3484b2de | 4093 | printk(" %ld", zone->lowmem_reserve[i]); |
1da177e4 LT |
4094 | printk("\n"); |
4095 | } | |
4096 | ||
ee99c71c | 4097 | for_each_populated_zone(zone) { |
d00181b9 KS |
4098 | unsigned int order; |
4099 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 4100 | unsigned char types[MAX_ORDER]; |
1da177e4 | 4101 | |
7bf02ea2 | 4102 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 4103 | continue; |
1da177e4 LT |
4104 | show_node(zone); |
4105 | printk("%s: ", zone->name); | |
1da177e4 LT |
4106 | |
4107 | spin_lock_irqsave(&zone->lock, flags); | |
4108 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
4109 | struct free_area *area = &zone->free_area[order]; |
4110 | int type; | |
4111 | ||
4112 | nr[order] = area->nr_free; | |
8f9de51a | 4113 | total += nr[order] << order; |
377e4f16 RV |
4114 | |
4115 | types[order] = 0; | |
4116 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
4117 | if (!list_empty(&area->free_list[type])) | |
4118 | types[order] |= 1 << type; | |
4119 | } | |
1da177e4 LT |
4120 | } |
4121 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 4122 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 4123 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
4124 | if (nr[order]) |
4125 | show_migration_types(types[order]); | |
4126 | } | |
1da177e4 LT |
4127 | printk("= %lukB\n", K(total)); |
4128 | } | |
4129 | ||
949f7ec5 DR |
4130 | hugetlb_show_meminfo(); |
4131 | ||
e6f3602d LW |
4132 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
4133 | ||
1da177e4 LT |
4134 | show_swap_cache_info(); |
4135 | } | |
4136 | ||
19770b32 MG |
4137 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4138 | { | |
4139 | zoneref->zone = zone; | |
4140 | zoneref->zone_idx = zone_idx(zone); | |
4141 | } | |
4142 | ||
1da177e4 LT |
4143 | /* |
4144 | * Builds allocation fallback zone lists. | |
1a93205b CL |
4145 | * |
4146 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 4147 | */ |
f0c0b2b8 | 4148 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 4149 | int nr_zones) |
1da177e4 | 4150 | { |
1a93205b | 4151 | struct zone *zone; |
bc732f1d | 4152 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
4153 | |
4154 | do { | |
2f6726e5 | 4155 | zone_type--; |
070f8032 | 4156 | zone = pgdat->node_zones + zone_type; |
1a93205b | 4157 | if (populated_zone(zone)) { |
dd1a239f MG |
4158 | zoneref_set_zone(zone, |
4159 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 4160 | check_highest_zone(zone_type); |
1da177e4 | 4161 | } |
2f6726e5 | 4162 | } while (zone_type); |
bc732f1d | 4163 | |
070f8032 | 4164 | return nr_zones; |
1da177e4 LT |
4165 | } |
4166 | ||
f0c0b2b8 KH |
4167 | |
4168 | /* | |
4169 | * zonelist_order: | |
4170 | * 0 = automatic detection of better ordering. | |
4171 | * 1 = order by ([node] distance, -zonetype) | |
4172 | * 2 = order by (-zonetype, [node] distance) | |
4173 | * | |
4174 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
4175 | * the same zonelist. So only NUMA can configure this param. | |
4176 | */ | |
4177 | #define ZONELIST_ORDER_DEFAULT 0 | |
4178 | #define ZONELIST_ORDER_NODE 1 | |
4179 | #define ZONELIST_ORDER_ZONE 2 | |
4180 | ||
4181 | /* zonelist order in the kernel. | |
4182 | * set_zonelist_order() will set this to NODE or ZONE. | |
4183 | */ | |
4184 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4185 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
4186 | ||
4187 | ||
1da177e4 | 4188 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
4189 | /* The value user specified ....changed by config */ |
4190 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4191 | /* string for sysctl */ | |
4192 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
4193 | char numa_zonelist_order[16] = "default"; | |
4194 | ||
4195 | /* | |
4196 | * interface for configure zonelist ordering. | |
4197 | * command line option "numa_zonelist_order" | |
4198 | * = "[dD]efault - default, automatic configuration. | |
4199 | * = "[nN]ode - order by node locality, then by zone within node | |
4200 | * = "[zZ]one - order by zone, then by locality within zone | |
4201 | */ | |
4202 | ||
4203 | static int __parse_numa_zonelist_order(char *s) | |
4204 | { | |
4205 | if (*s == 'd' || *s == 'D') { | |
4206 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4207 | } else if (*s == 'n' || *s == 'N') { | |
4208 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
4209 | } else if (*s == 'z' || *s == 'Z') { | |
4210 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
4211 | } else { | |
1170532b | 4212 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
f0c0b2b8 KH |
4213 | return -EINVAL; |
4214 | } | |
4215 | return 0; | |
4216 | } | |
4217 | ||
4218 | static __init int setup_numa_zonelist_order(char *s) | |
4219 | { | |
ecb256f8 VL |
4220 | int ret; |
4221 | ||
4222 | if (!s) | |
4223 | return 0; | |
4224 | ||
4225 | ret = __parse_numa_zonelist_order(s); | |
4226 | if (ret == 0) | |
4227 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
4228 | ||
4229 | return ret; | |
f0c0b2b8 KH |
4230 | } |
4231 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
4232 | ||
4233 | /* | |
4234 | * sysctl handler for numa_zonelist_order | |
4235 | */ | |
cccad5b9 | 4236 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 4237 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
4238 | loff_t *ppos) |
4239 | { | |
4240 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
4241 | int ret; | |
443c6f14 | 4242 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 4243 | |
443c6f14 | 4244 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
4245 | if (write) { |
4246 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
4247 | ret = -EINVAL; | |
4248 | goto out; | |
4249 | } | |
4250 | strcpy(saved_string, (char *)table->data); | |
4251 | } | |
8d65af78 | 4252 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 4253 | if (ret) |
443c6f14 | 4254 | goto out; |
f0c0b2b8 KH |
4255 | if (write) { |
4256 | int oldval = user_zonelist_order; | |
dacbde09 CG |
4257 | |
4258 | ret = __parse_numa_zonelist_order((char *)table->data); | |
4259 | if (ret) { | |
f0c0b2b8 KH |
4260 | /* |
4261 | * bogus value. restore saved string | |
4262 | */ | |
dacbde09 | 4263 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
4264 | NUMA_ZONELIST_ORDER_LEN); |
4265 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
4266 | } else if (oldval != user_zonelist_order) { |
4267 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 4268 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
4269 | mutex_unlock(&zonelists_mutex); |
4270 | } | |
f0c0b2b8 | 4271 | } |
443c6f14 AK |
4272 | out: |
4273 | mutex_unlock(&zl_order_mutex); | |
4274 | return ret; | |
f0c0b2b8 KH |
4275 | } |
4276 | ||
4277 | ||
62bc62a8 | 4278 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
4279 | static int node_load[MAX_NUMNODES]; |
4280 | ||
1da177e4 | 4281 | /** |
4dc3b16b | 4282 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
4283 | * @node: node whose fallback list we're appending |
4284 | * @used_node_mask: nodemask_t of already used nodes | |
4285 | * | |
4286 | * We use a number of factors to determine which is the next node that should | |
4287 | * appear on a given node's fallback list. The node should not have appeared | |
4288 | * already in @node's fallback list, and it should be the next closest node | |
4289 | * according to the distance array (which contains arbitrary distance values | |
4290 | * from each node to each node in the system), and should also prefer nodes | |
4291 | * with no CPUs, since presumably they'll have very little allocation pressure | |
4292 | * on them otherwise. | |
4293 | * It returns -1 if no node is found. | |
4294 | */ | |
f0c0b2b8 | 4295 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 4296 | { |
4cf808eb | 4297 | int n, val; |
1da177e4 | 4298 | int min_val = INT_MAX; |
00ef2d2f | 4299 | int best_node = NUMA_NO_NODE; |
a70f7302 | 4300 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 4301 | |
4cf808eb LT |
4302 | /* Use the local node if we haven't already */ |
4303 | if (!node_isset(node, *used_node_mask)) { | |
4304 | node_set(node, *used_node_mask); | |
4305 | return node; | |
4306 | } | |
1da177e4 | 4307 | |
4b0ef1fe | 4308 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
4309 | |
4310 | /* Don't want a node to appear more than once */ | |
4311 | if (node_isset(n, *used_node_mask)) | |
4312 | continue; | |
4313 | ||
1da177e4 LT |
4314 | /* Use the distance array to find the distance */ |
4315 | val = node_distance(node, n); | |
4316 | ||
4cf808eb LT |
4317 | /* Penalize nodes under us ("prefer the next node") */ |
4318 | val += (n < node); | |
4319 | ||
1da177e4 | 4320 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
4321 | tmp = cpumask_of_node(n); |
4322 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
4323 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4324 | ||
4325 | /* Slight preference for less loaded node */ | |
4326 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
4327 | val += node_load[n]; | |
4328 | ||
4329 | if (val < min_val) { | |
4330 | min_val = val; | |
4331 | best_node = n; | |
4332 | } | |
4333 | } | |
4334 | ||
4335 | if (best_node >= 0) | |
4336 | node_set(best_node, *used_node_mask); | |
4337 | ||
4338 | return best_node; | |
4339 | } | |
4340 | ||
f0c0b2b8 KH |
4341 | |
4342 | /* | |
4343 | * Build zonelists ordered by node and zones within node. | |
4344 | * This results in maximum locality--normal zone overflows into local | |
4345 | * DMA zone, if any--but risks exhausting DMA zone. | |
4346 | */ | |
4347 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 4348 | { |
f0c0b2b8 | 4349 | int j; |
1da177e4 | 4350 | struct zonelist *zonelist; |
f0c0b2b8 | 4351 | |
54a6eb5c | 4352 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 4353 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 4354 | ; |
bc732f1d | 4355 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
4356 | zonelist->_zonerefs[j].zone = NULL; |
4357 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
4358 | } |
4359 | ||
523b9458 CL |
4360 | /* |
4361 | * Build gfp_thisnode zonelists | |
4362 | */ | |
4363 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
4364 | { | |
523b9458 CL |
4365 | int j; |
4366 | struct zonelist *zonelist; | |
4367 | ||
54a6eb5c | 4368 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 4369 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
4370 | zonelist->_zonerefs[j].zone = NULL; |
4371 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
4372 | } |
4373 | ||
f0c0b2b8 KH |
4374 | /* |
4375 | * Build zonelists ordered by zone and nodes within zones. | |
4376 | * This results in conserving DMA zone[s] until all Normal memory is | |
4377 | * exhausted, but results in overflowing to remote node while memory | |
4378 | * may still exist in local DMA zone. | |
4379 | */ | |
4380 | static int node_order[MAX_NUMNODES]; | |
4381 | ||
4382 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
4383 | { | |
f0c0b2b8 KH |
4384 | int pos, j, node; |
4385 | int zone_type; /* needs to be signed */ | |
4386 | struct zone *z; | |
4387 | struct zonelist *zonelist; | |
4388 | ||
54a6eb5c MG |
4389 | zonelist = &pgdat->node_zonelists[0]; |
4390 | pos = 0; | |
4391 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
4392 | for (j = 0; j < nr_nodes; j++) { | |
4393 | node = node_order[j]; | |
4394 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
4395 | if (populated_zone(z)) { | |
dd1a239f MG |
4396 | zoneref_set_zone(z, |
4397 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 4398 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
4399 | } |
4400 | } | |
f0c0b2b8 | 4401 | } |
dd1a239f MG |
4402 | zonelist->_zonerefs[pos].zone = NULL; |
4403 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
4404 | } |
4405 | ||
3193913c MG |
4406 | #if defined(CONFIG_64BIT) |
4407 | /* | |
4408 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
4409 | * penalty to every machine in case the specialised case applies. Default | |
4410 | * to Node-ordering on 64-bit NUMA machines | |
4411 | */ | |
4412 | static int default_zonelist_order(void) | |
4413 | { | |
4414 | return ZONELIST_ORDER_NODE; | |
4415 | } | |
4416 | #else | |
4417 | /* | |
4418 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
4419 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
4420 | * then reclaim will occur more frequency increasing stalls and potentially | |
4421 | * be easier to OOM if a large percentage of the zone is under writeback or | |
4422 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
4423 | * Hence, default to zone ordering on 32-bit. | |
4424 | */ | |
f0c0b2b8 KH |
4425 | static int default_zonelist_order(void) |
4426 | { | |
f0c0b2b8 KH |
4427 | return ZONELIST_ORDER_ZONE; |
4428 | } | |
3193913c | 4429 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
4430 | |
4431 | static void set_zonelist_order(void) | |
4432 | { | |
4433 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
4434 | current_zonelist_order = default_zonelist_order(); | |
4435 | else | |
4436 | current_zonelist_order = user_zonelist_order; | |
4437 | } | |
4438 | ||
4439 | static void build_zonelists(pg_data_t *pgdat) | |
4440 | { | |
c00eb15a | 4441 | int i, node, load; |
1da177e4 | 4442 | nodemask_t used_mask; |
f0c0b2b8 KH |
4443 | int local_node, prev_node; |
4444 | struct zonelist *zonelist; | |
d00181b9 | 4445 | unsigned int order = current_zonelist_order; |
1da177e4 LT |
4446 | |
4447 | /* initialize zonelists */ | |
523b9458 | 4448 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 4449 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
4450 | zonelist->_zonerefs[0].zone = NULL; |
4451 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
4452 | } |
4453 | ||
4454 | /* NUMA-aware ordering of nodes */ | |
4455 | local_node = pgdat->node_id; | |
62bc62a8 | 4456 | load = nr_online_nodes; |
1da177e4 LT |
4457 | prev_node = local_node; |
4458 | nodes_clear(used_mask); | |
f0c0b2b8 | 4459 | |
f0c0b2b8 | 4460 | memset(node_order, 0, sizeof(node_order)); |
c00eb15a | 4461 | i = 0; |
f0c0b2b8 | 4462 | |
1da177e4 LT |
4463 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
4464 | /* | |
4465 | * We don't want to pressure a particular node. | |
4466 | * So adding penalty to the first node in same | |
4467 | * distance group to make it round-robin. | |
4468 | */ | |
957f822a DR |
4469 | if (node_distance(local_node, node) != |
4470 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
4471 | node_load[node] = load; |
4472 | ||
1da177e4 LT |
4473 | prev_node = node; |
4474 | load--; | |
f0c0b2b8 KH |
4475 | if (order == ZONELIST_ORDER_NODE) |
4476 | build_zonelists_in_node_order(pgdat, node); | |
4477 | else | |
c00eb15a | 4478 | node_order[i++] = node; /* remember order */ |
f0c0b2b8 | 4479 | } |
1da177e4 | 4480 | |
f0c0b2b8 KH |
4481 | if (order == ZONELIST_ORDER_ZONE) { |
4482 | /* calculate node order -- i.e., DMA last! */ | |
c00eb15a | 4483 | build_zonelists_in_zone_order(pgdat, i); |
1da177e4 | 4484 | } |
523b9458 CL |
4485 | |
4486 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
4487 | } |
4488 | ||
7aac7898 LS |
4489 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4490 | /* | |
4491 | * Return node id of node used for "local" allocations. | |
4492 | * I.e., first node id of first zone in arg node's generic zonelist. | |
4493 | * Used for initializing percpu 'numa_mem', which is used primarily | |
4494 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
4495 | */ | |
4496 | int local_memory_node(int node) | |
4497 | { | |
c33d6c06 | 4498 | struct zoneref *z; |
7aac7898 | 4499 | |
c33d6c06 | 4500 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
7aac7898 | 4501 | gfp_zone(GFP_KERNEL), |
c33d6c06 MG |
4502 | NULL); |
4503 | return z->zone->node; | |
7aac7898 LS |
4504 | } |
4505 | #endif | |
f0c0b2b8 | 4506 | |
1da177e4 LT |
4507 | #else /* CONFIG_NUMA */ |
4508 | ||
f0c0b2b8 KH |
4509 | static void set_zonelist_order(void) |
4510 | { | |
4511 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
4512 | } | |
4513 | ||
4514 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 4515 | { |
19655d34 | 4516 | int node, local_node; |
54a6eb5c MG |
4517 | enum zone_type j; |
4518 | struct zonelist *zonelist; | |
1da177e4 LT |
4519 | |
4520 | local_node = pgdat->node_id; | |
1da177e4 | 4521 | |
54a6eb5c | 4522 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 4523 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 4524 | |
54a6eb5c MG |
4525 | /* |
4526 | * Now we build the zonelist so that it contains the zones | |
4527 | * of all the other nodes. | |
4528 | * We don't want to pressure a particular node, so when | |
4529 | * building the zones for node N, we make sure that the | |
4530 | * zones coming right after the local ones are those from | |
4531 | * node N+1 (modulo N) | |
4532 | */ | |
4533 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
4534 | if (!node_online(node)) | |
4535 | continue; | |
bc732f1d | 4536 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 4537 | } |
54a6eb5c MG |
4538 | for (node = 0; node < local_node; node++) { |
4539 | if (!node_online(node)) | |
4540 | continue; | |
bc732f1d | 4541 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
4542 | } |
4543 | ||
dd1a239f MG |
4544 | zonelist->_zonerefs[j].zone = NULL; |
4545 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
4546 | } |
4547 | ||
4548 | #endif /* CONFIG_NUMA */ | |
4549 | ||
99dcc3e5 CL |
4550 | /* |
4551 | * Boot pageset table. One per cpu which is going to be used for all | |
4552 | * zones and all nodes. The parameters will be set in such a way | |
4553 | * that an item put on a list will immediately be handed over to | |
4554 | * the buddy list. This is safe since pageset manipulation is done | |
4555 | * with interrupts disabled. | |
4556 | * | |
4557 | * The boot_pagesets must be kept even after bootup is complete for | |
4558 | * unused processors and/or zones. They do play a role for bootstrapping | |
4559 | * hotplugged processors. | |
4560 | * | |
4561 | * zoneinfo_show() and maybe other functions do | |
4562 | * not check if the processor is online before following the pageset pointer. | |
4563 | * Other parts of the kernel may not check if the zone is available. | |
4564 | */ | |
4565 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
4566 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 4567 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 4568 | |
4eaf3f64 HL |
4569 | /* |
4570 | * Global mutex to protect against size modification of zonelists | |
4571 | * as well as to serialize pageset setup for the new populated zone. | |
4572 | */ | |
4573 | DEFINE_MUTEX(zonelists_mutex); | |
4574 | ||
9b1a4d38 | 4575 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 4576 | static int __build_all_zonelists(void *data) |
1da177e4 | 4577 | { |
6811378e | 4578 | int nid; |
99dcc3e5 | 4579 | int cpu; |
9adb62a5 | 4580 | pg_data_t *self = data; |
9276b1bc | 4581 | |
7f9cfb31 BL |
4582 | #ifdef CONFIG_NUMA |
4583 | memset(node_load, 0, sizeof(node_load)); | |
4584 | #endif | |
9adb62a5 JL |
4585 | |
4586 | if (self && !node_online(self->node_id)) { | |
4587 | build_zonelists(self); | |
9adb62a5 JL |
4588 | } |
4589 | ||
9276b1bc | 4590 | for_each_online_node(nid) { |
7ea1530a CL |
4591 | pg_data_t *pgdat = NODE_DATA(nid); |
4592 | ||
4593 | build_zonelists(pgdat); | |
9276b1bc | 4594 | } |
99dcc3e5 CL |
4595 | |
4596 | /* | |
4597 | * Initialize the boot_pagesets that are going to be used | |
4598 | * for bootstrapping processors. The real pagesets for | |
4599 | * each zone will be allocated later when the per cpu | |
4600 | * allocator is available. | |
4601 | * | |
4602 | * boot_pagesets are used also for bootstrapping offline | |
4603 | * cpus if the system is already booted because the pagesets | |
4604 | * are needed to initialize allocators on a specific cpu too. | |
4605 | * F.e. the percpu allocator needs the page allocator which | |
4606 | * needs the percpu allocator in order to allocate its pagesets | |
4607 | * (a chicken-egg dilemma). | |
4608 | */ | |
7aac7898 | 4609 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
4610 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
4611 | ||
7aac7898 LS |
4612 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4613 | /* | |
4614 | * We now know the "local memory node" for each node-- | |
4615 | * i.e., the node of the first zone in the generic zonelist. | |
4616 | * Set up numa_mem percpu variable for on-line cpus. During | |
4617 | * boot, only the boot cpu should be on-line; we'll init the | |
4618 | * secondary cpus' numa_mem as they come on-line. During | |
4619 | * node/memory hotplug, we'll fixup all on-line cpus. | |
4620 | */ | |
4621 | if (cpu_online(cpu)) | |
4622 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
4623 | #endif | |
4624 | } | |
4625 | ||
6811378e YG |
4626 | return 0; |
4627 | } | |
4628 | ||
061f67bc RV |
4629 | static noinline void __init |
4630 | build_all_zonelists_init(void) | |
4631 | { | |
4632 | __build_all_zonelists(NULL); | |
4633 | mminit_verify_zonelist(); | |
4634 | cpuset_init_current_mems_allowed(); | |
4635 | } | |
4636 | ||
4eaf3f64 HL |
4637 | /* |
4638 | * Called with zonelists_mutex held always | |
4639 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
4640 | * |
4641 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
4642 | * [we're only called with non-NULL zone through __meminit paths] and | |
4643 | * (2) call of __init annotated helper build_all_zonelists_init | |
4644 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 4645 | */ |
9adb62a5 | 4646 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 4647 | { |
f0c0b2b8 KH |
4648 | set_zonelist_order(); |
4649 | ||
6811378e | 4650 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 4651 | build_all_zonelists_init(); |
6811378e | 4652 | } else { |
e9959f0f | 4653 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
4654 | if (zone) |
4655 | setup_zone_pageset(zone); | |
e9959f0f | 4656 | #endif |
dd1895e2 CS |
4657 | /* we have to stop all cpus to guarantee there is no user |
4658 | of zonelist */ | |
9adb62a5 | 4659 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
4660 | /* cpuset refresh routine should be here */ |
4661 | } | |
bd1e22b8 | 4662 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
4663 | /* |
4664 | * Disable grouping by mobility if the number of pages in the | |
4665 | * system is too low to allow the mechanism to work. It would be | |
4666 | * more accurate, but expensive to check per-zone. This check is | |
4667 | * made on memory-hotadd so a system can start with mobility | |
4668 | * disabled and enable it later | |
4669 | */ | |
d9c23400 | 4670 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
4671 | page_group_by_mobility_disabled = 1; |
4672 | else | |
4673 | page_group_by_mobility_disabled = 0; | |
4674 | ||
756a025f JP |
4675 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
4676 | nr_online_nodes, | |
4677 | zonelist_order_name[current_zonelist_order], | |
4678 | page_group_by_mobility_disabled ? "off" : "on", | |
4679 | vm_total_pages); | |
f0c0b2b8 | 4680 | #ifdef CONFIG_NUMA |
f88dfff5 | 4681 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 4682 | #endif |
1da177e4 LT |
4683 | } |
4684 | ||
4685 | /* | |
4686 | * Helper functions to size the waitqueue hash table. | |
4687 | * Essentially these want to choose hash table sizes sufficiently | |
4688 | * large so that collisions trying to wait on pages are rare. | |
4689 | * But in fact, the number of active page waitqueues on typical | |
4690 | * systems is ridiculously low, less than 200. So this is even | |
4691 | * conservative, even though it seems large. | |
4692 | * | |
4693 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
4694 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
4695 | */ | |
4696 | #define PAGES_PER_WAITQUEUE 256 | |
4697 | ||
cca448fe | 4698 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 4699 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
4700 | { |
4701 | unsigned long size = 1; | |
4702 | ||
4703 | pages /= PAGES_PER_WAITQUEUE; | |
4704 | ||
4705 | while (size < pages) | |
4706 | size <<= 1; | |
4707 | ||
4708 | /* | |
4709 | * Once we have dozens or even hundreds of threads sleeping | |
4710 | * on IO we've got bigger problems than wait queue collision. | |
4711 | * Limit the size of the wait table to a reasonable size. | |
4712 | */ | |
4713 | size = min(size, 4096UL); | |
4714 | ||
4715 | return max(size, 4UL); | |
4716 | } | |
cca448fe YG |
4717 | #else |
4718 | /* | |
4719 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
4720 | * a suitable size for its wait_table. So we use the maximum size now. | |
4721 | * | |
4722 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
4723 | * | |
4724 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
4725 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
4726 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
4727 | * | |
4728 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
4729 | * or more by the traditional way. (See above). It equals: | |
4730 | * | |
4731 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
4732 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
4733 | * powerpc (64K page size) : = (32G +16M)byte. | |
4734 | */ | |
4735 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
4736 | { | |
4737 | return 4096UL; | |
4738 | } | |
4739 | #endif | |
1da177e4 LT |
4740 | |
4741 | /* | |
4742 | * This is an integer logarithm so that shifts can be used later | |
4743 | * to extract the more random high bits from the multiplicative | |
4744 | * hash function before the remainder is taken. | |
4745 | */ | |
4746 | static inline unsigned long wait_table_bits(unsigned long size) | |
4747 | { | |
4748 | return ffz(~size); | |
4749 | } | |
4750 | ||
1da177e4 LT |
4751 | /* |
4752 | * Initially all pages are reserved - free ones are freed | |
4753 | * up by free_all_bootmem() once the early boot process is | |
4754 | * done. Non-atomic initialization, single-pass. | |
4755 | */ | |
c09b4240 | 4756 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 4757 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 4758 | { |
4b94ffdc | 4759 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
29751f69 | 4760 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 4761 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 4762 | unsigned long pfn; |
3a80a7fa | 4763 | unsigned long nr_initialised = 0; |
342332e6 TI |
4764 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
4765 | struct memblock_region *r = NULL, *tmp; | |
4766 | #endif | |
1da177e4 | 4767 | |
22b31eec HD |
4768 | if (highest_memmap_pfn < end_pfn - 1) |
4769 | highest_memmap_pfn = end_pfn - 1; | |
4770 | ||
4b94ffdc DW |
4771 | /* |
4772 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
4773 | * memory | |
4774 | */ | |
4775 | if (altmap && start_pfn == altmap->base_pfn) | |
4776 | start_pfn += altmap->reserve; | |
4777 | ||
cbe8dd4a | 4778 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 4779 | /* |
b72d0ffb AM |
4780 | * There can be holes in boot-time mem_map[]s handed to this |
4781 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 4782 | */ |
b72d0ffb AM |
4783 | if (context != MEMMAP_EARLY) |
4784 | goto not_early; | |
4785 | ||
4786 | if (!early_pfn_valid(pfn)) | |
4787 | continue; | |
4788 | if (!early_pfn_in_nid(pfn, nid)) | |
4789 | continue; | |
4790 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
4791 | break; | |
342332e6 TI |
4792 | |
4793 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
4794 | /* |
4795 | * If not mirrored_kernelcore and ZONE_MOVABLE exists, range | |
4796 | * from zone_movable_pfn[nid] to end of each node should be | |
4797 | * ZONE_MOVABLE not ZONE_NORMAL. skip it. | |
4798 | */ | |
4799 | if (!mirrored_kernelcore && zone_movable_pfn[nid]) | |
4800 | if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid]) | |
4801 | continue; | |
342332e6 | 4802 | |
b72d0ffb AM |
4803 | /* |
4804 | * Check given memblock attribute by firmware which can affect | |
4805 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
4806 | * mirrored, it's an overlapped memmap init. skip it. | |
4807 | */ | |
4808 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
4809 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
4810 | for_each_memblock(memory, tmp) | |
4811 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
4812 | break; | |
4813 | r = tmp; | |
4814 | } | |
4815 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
4816 | memblock_is_mirror(r)) { | |
4817 | /* already initialized as NORMAL */ | |
4818 | pfn = memblock_region_memory_end_pfn(r); | |
4819 | continue; | |
342332e6 | 4820 | } |
a2f3aa02 | 4821 | } |
b72d0ffb | 4822 | #endif |
ac5d2539 | 4823 | |
b72d0ffb | 4824 | not_early: |
ac5d2539 MG |
4825 | /* |
4826 | * Mark the block movable so that blocks are reserved for | |
4827 | * movable at startup. This will force kernel allocations | |
4828 | * to reserve their blocks rather than leaking throughout | |
4829 | * the address space during boot when many long-lived | |
974a786e | 4830 | * kernel allocations are made. |
ac5d2539 MG |
4831 | * |
4832 | * bitmap is created for zone's valid pfn range. but memmap | |
4833 | * can be created for invalid pages (for alignment) | |
4834 | * check here not to call set_pageblock_migratetype() against | |
4835 | * pfn out of zone. | |
4836 | */ | |
4837 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
4838 | struct page *page = pfn_to_page(pfn); | |
4839 | ||
4840 | __init_single_page(page, pfn, zone, nid); | |
4841 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4842 | } else { | |
4843 | __init_single_pfn(pfn, zone, nid); | |
4844 | } | |
1da177e4 LT |
4845 | } |
4846 | } | |
4847 | ||
1e548deb | 4848 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 4849 | { |
7aeb09f9 | 4850 | unsigned int order, t; |
b2a0ac88 MG |
4851 | for_each_migratetype_order(order, t) { |
4852 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
4853 | zone->free_area[order].nr_free = 0; |
4854 | } | |
4855 | } | |
4856 | ||
4857 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
4858 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 4859 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
4860 | #endif |
4861 | ||
7cd2b0a3 | 4862 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 4863 | { |
3a6be87f | 4864 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
4865 | int batch; |
4866 | ||
4867 | /* | |
4868 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 4869 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
4870 | * |
4871 | * OK, so we don't know how big the cache is. So guess. | |
4872 | */ | |
b40da049 | 4873 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
4874 | if (batch * PAGE_SIZE > 512 * 1024) |
4875 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
4876 | batch /= 4; /* We effectively *= 4 below */ |
4877 | if (batch < 1) | |
4878 | batch = 1; | |
4879 | ||
4880 | /* | |
0ceaacc9 NP |
4881 | * Clamp the batch to a 2^n - 1 value. Having a power |
4882 | * of 2 value was found to be more likely to have | |
4883 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 4884 | * |
0ceaacc9 NP |
4885 | * For example if 2 tasks are alternately allocating |
4886 | * batches of pages, one task can end up with a lot | |
4887 | * of pages of one half of the possible page colors | |
4888 | * and the other with pages of the other colors. | |
e7c8d5c9 | 4889 | */ |
9155203a | 4890 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 4891 | |
e7c8d5c9 | 4892 | return batch; |
3a6be87f DH |
4893 | |
4894 | #else | |
4895 | /* The deferral and batching of frees should be suppressed under NOMMU | |
4896 | * conditions. | |
4897 | * | |
4898 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4899 | * of contiguous memory as there's no hardware page translation to | |
4900 | * assemble apparent contiguous memory from discontiguous pages. | |
4901 | * | |
4902 | * Queueing large contiguous runs of pages for batching, however, | |
4903 | * causes the pages to actually be freed in smaller chunks. As there | |
4904 | * can be a significant delay between the individual batches being | |
4905 | * recycled, this leads to the once large chunks of space being | |
4906 | * fragmented and becoming unavailable for high-order allocations. | |
4907 | */ | |
4908 | return 0; | |
4909 | #endif | |
e7c8d5c9 CL |
4910 | } |
4911 | ||
8d7a8fa9 CS |
4912 | /* |
4913 | * pcp->high and pcp->batch values are related and dependent on one another: | |
4914 | * ->batch must never be higher then ->high. | |
4915 | * The following function updates them in a safe manner without read side | |
4916 | * locking. | |
4917 | * | |
4918 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
4919 | * those fields changing asynchronously (acording the the above rule). | |
4920 | * | |
4921 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
4922 | * outside of boot time (or some other assurance that no concurrent updaters | |
4923 | * exist). | |
4924 | */ | |
4925 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
4926 | unsigned long batch) | |
4927 | { | |
4928 | /* start with a fail safe value for batch */ | |
4929 | pcp->batch = 1; | |
4930 | smp_wmb(); | |
4931 | ||
4932 | /* Update high, then batch, in order */ | |
4933 | pcp->high = high; | |
4934 | smp_wmb(); | |
4935 | ||
4936 | pcp->batch = batch; | |
4937 | } | |
4938 | ||
3664033c | 4939 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
4940 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
4941 | { | |
8d7a8fa9 | 4942 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
4943 | } |
4944 | ||
88c90dbc | 4945 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
4946 | { |
4947 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4948 | int migratetype; |
2caaad41 | 4949 | |
1c6fe946 MD |
4950 | memset(p, 0, sizeof(*p)); |
4951 | ||
3dfa5721 | 4952 | pcp = &p->pcp; |
2caaad41 | 4953 | pcp->count = 0; |
5f8dcc21 MG |
4954 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4955 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4956 | } |
4957 | ||
88c90dbc CS |
4958 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
4959 | { | |
4960 | pageset_init(p); | |
4961 | pageset_set_batch(p, batch); | |
4962 | } | |
4963 | ||
8ad4b1fb | 4964 | /* |
3664033c | 4965 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
4966 | * to the value high for the pageset p. |
4967 | */ | |
3664033c | 4968 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
4969 | unsigned long high) |
4970 | { | |
8d7a8fa9 CS |
4971 | unsigned long batch = max(1UL, high / 4); |
4972 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
4973 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 4974 | |
8d7a8fa9 | 4975 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
4976 | } |
4977 | ||
7cd2b0a3 DR |
4978 | static void pageset_set_high_and_batch(struct zone *zone, |
4979 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 4980 | { |
56cef2b8 | 4981 | if (percpu_pagelist_fraction) |
3664033c | 4982 | pageset_set_high(pcp, |
56cef2b8 CS |
4983 | (zone->managed_pages / |
4984 | percpu_pagelist_fraction)); | |
4985 | else | |
4986 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
4987 | } | |
4988 | ||
169f6c19 CS |
4989 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
4990 | { | |
4991 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4992 | ||
4993 | pageset_init(pcp); | |
4994 | pageset_set_high_and_batch(zone, pcp); | |
4995 | } | |
4996 | ||
4ed7e022 | 4997 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4998 | { |
4999 | int cpu; | |
319774e2 | 5000 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
5001 | for_each_possible_cpu(cpu) |
5002 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
5003 | } |
5004 | ||
2caaad41 | 5005 | /* |
99dcc3e5 CL |
5006 | * Allocate per cpu pagesets and initialize them. |
5007 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 5008 | */ |
99dcc3e5 | 5009 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 5010 | { |
99dcc3e5 | 5011 | struct zone *zone; |
e7c8d5c9 | 5012 | |
319774e2 WF |
5013 | for_each_populated_zone(zone) |
5014 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
5015 | } |
5016 | ||
577a32f6 | 5017 | static noinline __init_refok |
cca448fe | 5018 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
5019 | { |
5020 | int i; | |
cca448fe | 5021 | size_t alloc_size; |
ed8ece2e DH |
5022 | |
5023 | /* | |
5024 | * The per-page waitqueue mechanism uses hashed waitqueues | |
5025 | * per zone. | |
5026 | */ | |
02b694de YG |
5027 | zone->wait_table_hash_nr_entries = |
5028 | wait_table_hash_nr_entries(zone_size_pages); | |
5029 | zone->wait_table_bits = | |
5030 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
5031 | alloc_size = zone->wait_table_hash_nr_entries |
5032 | * sizeof(wait_queue_head_t); | |
5033 | ||
cd94b9db | 5034 | if (!slab_is_available()) { |
cca448fe | 5035 | zone->wait_table = (wait_queue_head_t *) |
6782832e SS |
5036 | memblock_virt_alloc_node_nopanic( |
5037 | alloc_size, zone->zone_pgdat->node_id); | |
cca448fe YG |
5038 | } else { |
5039 | /* | |
5040 | * This case means that a zone whose size was 0 gets new memory | |
5041 | * via memory hot-add. | |
5042 | * But it may be the case that a new node was hot-added. In | |
5043 | * this case vmalloc() will not be able to use this new node's | |
5044 | * memory - this wait_table must be initialized to use this new | |
5045 | * node itself as well. | |
5046 | * To use this new node's memory, further consideration will be | |
5047 | * necessary. | |
5048 | */ | |
8691f3a7 | 5049 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
5050 | } |
5051 | if (!zone->wait_table) | |
5052 | return -ENOMEM; | |
ed8ece2e | 5053 | |
b8af2941 | 5054 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 5055 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
5056 | |
5057 | return 0; | |
ed8ece2e DH |
5058 | } |
5059 | ||
c09b4240 | 5060 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 5061 | { |
99dcc3e5 CL |
5062 | /* |
5063 | * per cpu subsystem is not up at this point. The following code | |
5064 | * relies on the ability of the linker to provide the | |
5065 | * offset of a (static) per cpu variable into the per cpu area. | |
5066 | */ | |
5067 | zone->pageset = &boot_pageset; | |
ed8ece2e | 5068 | |
b38a8725 | 5069 | if (populated_zone(zone)) |
99dcc3e5 CL |
5070 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5071 | zone->name, zone->present_pages, | |
5072 | zone_batchsize(zone)); | |
ed8ece2e DH |
5073 | } |
5074 | ||
4ed7e022 | 5075 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 5076 | unsigned long zone_start_pfn, |
b171e409 | 5077 | unsigned long size) |
ed8ece2e DH |
5078 | { |
5079 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
5080 | int ret; |
5081 | ret = zone_wait_table_init(zone, size); | |
5082 | if (ret) | |
5083 | return ret; | |
ed8ece2e DH |
5084 | pgdat->nr_zones = zone_idx(zone) + 1; |
5085 | ||
ed8ece2e DH |
5086 | zone->zone_start_pfn = zone_start_pfn; |
5087 | ||
708614e6 MG |
5088 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5089 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
5090 | pgdat->node_id, | |
5091 | (unsigned long)zone_idx(zone), | |
5092 | zone_start_pfn, (zone_start_pfn + size)); | |
5093 | ||
1e548deb | 5094 | zone_init_free_lists(zone); |
718127cc YG |
5095 | |
5096 | return 0; | |
ed8ece2e DH |
5097 | } |
5098 | ||
0ee332c1 | 5099 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 5100 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 5101 | |
c713216d MG |
5102 | /* |
5103 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 5104 | */ |
8a942fde MG |
5105 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5106 | struct mminit_pfnnid_cache *state) | |
c713216d | 5107 | { |
c13291a5 | 5108 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 5109 | int nid; |
7c243c71 | 5110 | |
8a942fde MG |
5111 | if (state->last_start <= pfn && pfn < state->last_end) |
5112 | return state->last_nid; | |
c713216d | 5113 | |
e76b63f8 YL |
5114 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5115 | if (nid != -1) { | |
8a942fde MG |
5116 | state->last_start = start_pfn; |
5117 | state->last_end = end_pfn; | |
5118 | state->last_nid = nid; | |
e76b63f8 YL |
5119 | } |
5120 | ||
5121 | return nid; | |
c713216d MG |
5122 | } |
5123 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5124 | ||
c713216d | 5125 | /** |
6782832e | 5126 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5127 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5128 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5129 | * |
7d018176 ZZ |
5130 | * If an architecture guarantees that all ranges registered contain no holes |
5131 | * and may be freed, this this function may be used instead of calling | |
5132 | * memblock_free_early_nid() manually. | |
c713216d | 5133 | */ |
c13291a5 | 5134 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5135 | { |
c13291a5 TH |
5136 | unsigned long start_pfn, end_pfn; |
5137 | int i, this_nid; | |
edbe7d23 | 5138 | |
c13291a5 TH |
5139 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5140 | start_pfn = min(start_pfn, max_low_pfn); | |
5141 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5142 | |
c13291a5 | 5143 | if (start_pfn < end_pfn) |
6782832e SS |
5144 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5145 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5146 | this_nid); | |
edbe7d23 | 5147 | } |
edbe7d23 | 5148 | } |
edbe7d23 | 5149 | |
c713216d MG |
5150 | /** |
5151 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5152 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5153 | * |
7d018176 ZZ |
5154 | * If an architecture guarantees that all ranges registered contain no holes and may |
5155 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5156 | */ |
5157 | void __init sparse_memory_present_with_active_regions(int nid) | |
5158 | { | |
c13291a5 TH |
5159 | unsigned long start_pfn, end_pfn; |
5160 | int i, this_nid; | |
c713216d | 5161 | |
c13291a5 TH |
5162 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5163 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5164 | } |
5165 | ||
5166 | /** | |
5167 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5168 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5169 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5170 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5171 | * |
5172 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5173 | * provided by memblock_set_node(). If called for a node |
c713216d | 5174 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5175 | * PFNs will be 0. |
c713216d | 5176 | */ |
a3142c8e | 5177 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5178 | unsigned long *start_pfn, unsigned long *end_pfn) |
5179 | { | |
c13291a5 | 5180 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5181 | int i; |
c13291a5 | 5182 | |
c713216d MG |
5183 | *start_pfn = -1UL; |
5184 | *end_pfn = 0; | |
5185 | ||
c13291a5 TH |
5186 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5187 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5188 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5189 | } |
5190 | ||
633c0666 | 5191 | if (*start_pfn == -1UL) |
c713216d | 5192 | *start_pfn = 0; |
c713216d MG |
5193 | } |
5194 | ||
2a1e274a MG |
5195 | /* |
5196 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5197 | * assumption is made that zones within a node are ordered in monotonic | |
5198 | * increasing memory addresses so that the "highest" populated zone is used | |
5199 | */ | |
b69a7288 | 5200 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5201 | { |
5202 | int zone_index; | |
5203 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5204 | if (zone_index == ZONE_MOVABLE) | |
5205 | continue; | |
5206 | ||
5207 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5208 | arch_zone_lowest_possible_pfn[zone_index]) | |
5209 | break; | |
5210 | } | |
5211 | ||
5212 | VM_BUG_ON(zone_index == -1); | |
5213 | movable_zone = zone_index; | |
5214 | } | |
5215 | ||
5216 | /* | |
5217 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5218 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5219 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5220 | * in each node depending on the size of each node and how evenly kernelcore | |
5221 | * is distributed. This helper function adjusts the zone ranges | |
5222 | * provided by the architecture for a given node by using the end of the | |
5223 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5224 | * zones within a node are in order of monotonic increases memory addresses | |
5225 | */ | |
b69a7288 | 5226 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5227 | unsigned long zone_type, |
5228 | unsigned long node_start_pfn, | |
5229 | unsigned long node_end_pfn, | |
5230 | unsigned long *zone_start_pfn, | |
5231 | unsigned long *zone_end_pfn) | |
5232 | { | |
5233 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5234 | if (zone_movable_pfn[nid]) { | |
5235 | /* Size ZONE_MOVABLE */ | |
5236 | if (zone_type == ZONE_MOVABLE) { | |
5237 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5238 | *zone_end_pfn = min(node_end_pfn, | |
5239 | arch_zone_highest_possible_pfn[movable_zone]); | |
5240 | ||
2a1e274a MG |
5241 | /* Check if this whole range is within ZONE_MOVABLE */ |
5242 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5243 | *zone_start_pfn = *zone_end_pfn; | |
5244 | } | |
5245 | } | |
5246 | ||
c713216d MG |
5247 | /* |
5248 | * Return the number of pages a zone spans in a node, including holes | |
5249 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5250 | */ | |
6ea6e688 | 5251 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5252 | unsigned long zone_type, |
7960aedd ZY |
5253 | unsigned long node_start_pfn, |
5254 | unsigned long node_end_pfn, | |
d91749c1 TI |
5255 | unsigned long *zone_start_pfn, |
5256 | unsigned long *zone_end_pfn, | |
c713216d MG |
5257 | unsigned long *ignored) |
5258 | { | |
b5685e92 | 5259 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5260 | if (!node_start_pfn && !node_end_pfn) |
5261 | return 0; | |
5262 | ||
7960aedd | 5263 | /* Get the start and end of the zone */ |
d91749c1 TI |
5264 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5265 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5266 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5267 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5268 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5269 | |
5270 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5271 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5272 | return 0; |
5273 | ||
5274 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5275 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5276 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5277 | |
5278 | /* Return the spanned pages */ | |
d91749c1 | 5279 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5280 | } |
5281 | ||
5282 | /* | |
5283 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5284 | * then all holes in the requested range will be accounted for. |
c713216d | 5285 | */ |
32996250 | 5286 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5287 | unsigned long range_start_pfn, |
5288 | unsigned long range_end_pfn) | |
5289 | { | |
96e907d1 TH |
5290 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5291 | unsigned long start_pfn, end_pfn; | |
5292 | int i; | |
c713216d | 5293 | |
96e907d1 TH |
5294 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5295 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5296 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5297 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5298 | } |
96e907d1 | 5299 | return nr_absent; |
c713216d MG |
5300 | } |
5301 | ||
5302 | /** | |
5303 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5304 | * @start_pfn: The start PFN to start searching for holes | |
5305 | * @end_pfn: The end PFN to stop searching for holes | |
5306 | * | |
88ca3b94 | 5307 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5308 | */ |
5309 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5310 | unsigned long end_pfn) | |
5311 | { | |
5312 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5313 | } | |
5314 | ||
5315 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5316 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5317 | unsigned long zone_type, |
7960aedd ZY |
5318 | unsigned long node_start_pfn, |
5319 | unsigned long node_end_pfn, | |
c713216d MG |
5320 | unsigned long *ignored) |
5321 | { | |
96e907d1 TH |
5322 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5323 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5324 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5325 | unsigned long nr_absent; |
9c7cd687 | 5326 | |
b5685e92 | 5327 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5328 | if (!node_start_pfn && !node_end_pfn) |
5329 | return 0; | |
5330 | ||
96e907d1 TH |
5331 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5332 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5333 | |
2a1e274a MG |
5334 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5335 | node_start_pfn, node_end_pfn, | |
5336 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5337 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5338 | ||
5339 | /* | |
5340 | * ZONE_MOVABLE handling. | |
5341 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
5342 | * and vice versa. | |
5343 | */ | |
5344 | if (zone_movable_pfn[nid]) { | |
5345 | if (mirrored_kernelcore) { | |
5346 | unsigned long start_pfn, end_pfn; | |
5347 | struct memblock_region *r; | |
5348 | ||
5349 | for_each_memblock(memory, r) { | |
5350 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
5351 | zone_start_pfn, zone_end_pfn); | |
5352 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
5353 | zone_start_pfn, zone_end_pfn); | |
5354 | ||
5355 | if (zone_type == ZONE_MOVABLE && | |
5356 | memblock_is_mirror(r)) | |
5357 | nr_absent += end_pfn - start_pfn; | |
5358 | ||
5359 | if (zone_type == ZONE_NORMAL && | |
5360 | !memblock_is_mirror(r)) | |
5361 | nr_absent += end_pfn - start_pfn; | |
5362 | } | |
5363 | } else { | |
5364 | if (zone_type == ZONE_NORMAL) | |
5365 | nr_absent += node_end_pfn - zone_movable_pfn[nid]; | |
5366 | } | |
5367 | } | |
5368 | ||
5369 | return nr_absent; | |
c713216d | 5370 | } |
0e0b864e | 5371 | |
0ee332c1 | 5372 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 5373 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5374 | unsigned long zone_type, |
7960aedd ZY |
5375 | unsigned long node_start_pfn, |
5376 | unsigned long node_end_pfn, | |
d91749c1 TI |
5377 | unsigned long *zone_start_pfn, |
5378 | unsigned long *zone_end_pfn, | |
c713216d MG |
5379 | unsigned long *zones_size) |
5380 | { | |
d91749c1 TI |
5381 | unsigned int zone; |
5382 | ||
5383 | *zone_start_pfn = node_start_pfn; | |
5384 | for (zone = 0; zone < zone_type; zone++) | |
5385 | *zone_start_pfn += zones_size[zone]; | |
5386 | ||
5387 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
5388 | ||
c713216d MG |
5389 | return zones_size[zone_type]; |
5390 | } | |
5391 | ||
6ea6e688 | 5392 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5393 | unsigned long zone_type, |
7960aedd ZY |
5394 | unsigned long node_start_pfn, |
5395 | unsigned long node_end_pfn, | |
c713216d MG |
5396 | unsigned long *zholes_size) |
5397 | { | |
5398 | if (!zholes_size) | |
5399 | return 0; | |
5400 | ||
5401 | return zholes_size[zone_type]; | |
5402 | } | |
20e6926d | 5403 | |
0ee332c1 | 5404 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5405 | |
a3142c8e | 5406 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
5407 | unsigned long node_start_pfn, |
5408 | unsigned long node_end_pfn, | |
5409 | unsigned long *zones_size, | |
5410 | unsigned long *zholes_size) | |
c713216d | 5411 | { |
febd5949 | 5412 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
5413 | enum zone_type i; |
5414 | ||
febd5949 GZ |
5415 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5416 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 5417 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 5418 | unsigned long size, real_size; |
c713216d | 5419 | |
febd5949 GZ |
5420 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5421 | node_start_pfn, | |
5422 | node_end_pfn, | |
d91749c1 TI |
5423 | &zone_start_pfn, |
5424 | &zone_end_pfn, | |
febd5949 GZ |
5425 | zones_size); |
5426 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
5427 | node_start_pfn, node_end_pfn, |
5428 | zholes_size); | |
d91749c1 TI |
5429 | if (size) |
5430 | zone->zone_start_pfn = zone_start_pfn; | |
5431 | else | |
5432 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
5433 | zone->spanned_pages = size; |
5434 | zone->present_pages = real_size; | |
5435 | ||
5436 | totalpages += size; | |
5437 | realtotalpages += real_size; | |
5438 | } | |
5439 | ||
5440 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
5441 | pgdat->node_present_pages = realtotalpages; |
5442 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
5443 | realtotalpages); | |
5444 | } | |
5445 | ||
835c134e MG |
5446 | #ifndef CONFIG_SPARSEMEM |
5447 | /* | |
5448 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
5449 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5450 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
5451 | * round what is now in bits to nearest long in bits, then return it in |
5452 | * bytes. | |
5453 | */ | |
7c45512d | 5454 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
5455 | { |
5456 | unsigned long usemapsize; | |
5457 | ||
7c45512d | 5458 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
5459 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
5460 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
5461 | usemapsize *= NR_PAGEBLOCK_BITS; |
5462 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
5463 | ||
5464 | return usemapsize / 8; | |
5465 | } | |
5466 | ||
5467 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
5468 | struct zone *zone, |
5469 | unsigned long zone_start_pfn, | |
5470 | unsigned long zonesize) | |
835c134e | 5471 | { |
7c45512d | 5472 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 5473 | zone->pageblock_flags = NULL; |
58a01a45 | 5474 | if (usemapsize) |
6782832e SS |
5475 | zone->pageblock_flags = |
5476 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5477 | pgdat->node_id); | |
835c134e MG |
5478 | } |
5479 | #else | |
7c45512d LT |
5480 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5481 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5482 | #endif /* CONFIG_SPARSEMEM */ |
5483 | ||
d9c23400 | 5484 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5485 | |
d9c23400 | 5486 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5487 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5488 | { |
955c1cd7 AM |
5489 | unsigned int order; |
5490 | ||
d9c23400 MG |
5491 | /* Check that pageblock_nr_pages has not already been setup */ |
5492 | if (pageblock_order) | |
5493 | return; | |
5494 | ||
955c1cd7 AM |
5495 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5496 | order = HUGETLB_PAGE_ORDER; | |
5497 | else | |
5498 | order = MAX_ORDER - 1; | |
5499 | ||
d9c23400 MG |
5500 | /* |
5501 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5502 | * This value may be variable depending on boot parameters on IA64 and |
5503 | * powerpc. | |
d9c23400 MG |
5504 | */ |
5505 | pageblock_order = order; | |
5506 | } | |
5507 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5508 | ||
ba72cb8c MG |
5509 | /* |
5510 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5511 | * is unused as pageblock_order is set at compile-time. See |
5512 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5513 | * the kernel config | |
ba72cb8c | 5514 | */ |
15ca220e | 5515 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5516 | { |
ba72cb8c | 5517 | } |
d9c23400 MG |
5518 | |
5519 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5520 | ||
01cefaef JL |
5521 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5522 | unsigned long present_pages) | |
5523 | { | |
5524 | unsigned long pages = spanned_pages; | |
5525 | ||
5526 | /* | |
5527 | * Provide a more accurate estimation if there are holes within | |
5528 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5529 | * zone, each populated memory region may cost us one or two extra | |
5530 | * memmap pages due to alignment because memmap pages for each | |
5531 | * populated regions may not naturally algined on page boundary. | |
5532 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
5533 | */ | |
5534 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5535 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5536 | pages = present_pages; | |
5537 | ||
5538 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5539 | } | |
5540 | ||
1da177e4 LT |
5541 | /* |
5542 | * Set up the zone data structures: | |
5543 | * - mark all pages reserved | |
5544 | * - mark all memory queues empty | |
5545 | * - clear the memory bitmaps | |
6527af5d MK |
5546 | * |
5547 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 5548 | */ |
7f3eb55b | 5549 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 5550 | { |
2f1b6248 | 5551 | enum zone_type j; |
ed8ece2e | 5552 | int nid = pgdat->node_id; |
718127cc | 5553 | int ret; |
1da177e4 | 5554 | |
208d54e5 | 5555 | pgdat_resize_init(pgdat); |
8177a420 AA |
5556 | #ifdef CONFIG_NUMA_BALANCING |
5557 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
5558 | pgdat->numabalancing_migrate_nr_pages = 0; | |
5559 | pgdat->numabalancing_migrate_next_window = jiffies; | |
a3d0a918 KS |
5560 | #endif |
5561 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5562 | spin_lock_init(&pgdat->split_queue_lock); | |
5563 | INIT_LIST_HEAD(&pgdat->split_queue); | |
5564 | pgdat->split_queue_len = 0; | |
8177a420 | 5565 | #endif |
1da177e4 | 5566 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 5567 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
698b1b30 VB |
5568 | #ifdef CONFIG_COMPACTION |
5569 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
5570 | #endif | |
eefa864b | 5571 | pgdat_page_ext_init(pgdat); |
5f63b720 | 5572 | |
1da177e4 LT |
5573 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5574 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 5575 | unsigned long size, realsize, freesize, memmap_pages; |
d91749c1 | 5576 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 5577 | |
febd5949 GZ |
5578 | size = zone->spanned_pages; |
5579 | realsize = freesize = zone->present_pages; | |
1da177e4 | 5580 | |
0e0b864e | 5581 | /* |
9feedc9d | 5582 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
5583 | * is used by this zone for memmap. This affects the watermark |
5584 | * and per-cpu initialisations | |
5585 | */ | |
01cefaef | 5586 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
5587 | if (!is_highmem_idx(j)) { |
5588 | if (freesize >= memmap_pages) { | |
5589 | freesize -= memmap_pages; | |
5590 | if (memmap_pages) | |
5591 | printk(KERN_DEBUG | |
5592 | " %s zone: %lu pages used for memmap\n", | |
5593 | zone_names[j], memmap_pages); | |
5594 | } else | |
1170532b | 5595 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
ba914f48 ZH |
5596 | zone_names[j], memmap_pages, freesize); |
5597 | } | |
0e0b864e | 5598 | |
6267276f | 5599 | /* Account for reserved pages */ |
9feedc9d JL |
5600 | if (j == 0 && freesize > dma_reserve) { |
5601 | freesize -= dma_reserve; | |
d903ef9f | 5602 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 5603 | zone_names[0], dma_reserve); |
0e0b864e MG |
5604 | } |
5605 | ||
98d2b0eb | 5606 | if (!is_highmem_idx(j)) |
9feedc9d | 5607 | nr_kernel_pages += freesize; |
01cefaef JL |
5608 | /* Charge for highmem memmap if there are enough kernel pages */ |
5609 | else if (nr_kernel_pages > memmap_pages * 2) | |
5610 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 5611 | nr_all_pages += freesize; |
1da177e4 | 5612 | |
9feedc9d JL |
5613 | /* |
5614 | * Set an approximate value for lowmem here, it will be adjusted | |
5615 | * when the bootmem allocator frees pages into the buddy system. | |
5616 | * And all highmem pages will be managed by the buddy system. | |
5617 | */ | |
5618 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 5619 | #ifdef CONFIG_NUMA |
d5f541ed | 5620 | zone->node = nid; |
9feedc9d | 5621 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 5622 | / 100; |
9feedc9d | 5623 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 5624 | #endif |
1da177e4 LT |
5625 | zone->name = zone_names[j]; |
5626 | spin_lock_init(&zone->lock); | |
5627 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 5628 | zone_seqlock_init(zone); |
1da177e4 | 5629 | zone->zone_pgdat = pgdat; |
ed8ece2e | 5630 | zone_pcp_init(zone); |
81c0a2bb JW |
5631 | |
5632 | /* For bootup, initialized properly in watermark setup */ | |
5633 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
5634 | ||
bea8c150 | 5635 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
5636 | if (!size) |
5637 | continue; | |
5638 | ||
955c1cd7 | 5639 | set_pageblock_order(); |
7c45512d | 5640 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
b171e409 | 5641 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
718127cc | 5642 | BUG_ON(ret); |
76cdd58e | 5643 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
5644 | } |
5645 | } | |
5646 | ||
577a32f6 | 5647 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 5648 | { |
b0aeba74 | 5649 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
5650 | unsigned long __maybe_unused offset = 0; |
5651 | ||
1da177e4 LT |
5652 | /* Skip empty nodes */ |
5653 | if (!pgdat->node_spanned_pages) | |
5654 | return; | |
5655 | ||
d41dee36 | 5656 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
b0aeba74 TL |
5657 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
5658 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
5659 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
5660 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 5661 | unsigned long size, end; |
d41dee36 AW |
5662 | struct page *map; |
5663 | ||
e984bb43 BP |
5664 | /* |
5665 | * The zone's endpoints aren't required to be MAX_ORDER | |
5666 | * aligned but the node_mem_map endpoints must be in order | |
5667 | * for the buddy allocator to function correctly. | |
5668 | */ | |
108bcc96 | 5669 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
5670 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
5671 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
5672 | map = alloc_remap(pgdat->node_id, size); |
5673 | if (!map) | |
6782832e SS |
5674 | map = memblock_virt_alloc_node_nopanic(size, |
5675 | pgdat->node_id); | |
a1c34a3b | 5676 | pgdat->node_mem_map = map + offset; |
1da177e4 | 5677 | } |
12d810c1 | 5678 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
5679 | /* |
5680 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
5681 | */ | |
c713216d | 5682 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 5683 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 5684 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 5685 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 5686 | mem_map -= offset; |
0ee332c1 | 5687 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5688 | } |
1da177e4 | 5689 | #endif |
d41dee36 | 5690 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
5691 | } |
5692 | ||
9109fb7b JW |
5693 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
5694 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 5695 | { |
9109fb7b | 5696 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
5697 | unsigned long start_pfn = 0; |
5698 | unsigned long end_pfn = 0; | |
9109fb7b | 5699 | |
88fdf75d | 5700 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 5701 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 5702 | |
3a80a7fa | 5703 | reset_deferred_meminit(pgdat); |
1da177e4 LT |
5704 | pgdat->node_id = nid; |
5705 | pgdat->node_start_pfn = node_start_pfn; | |
7960aedd ZY |
5706 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5707 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 5708 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
5709 | (u64)start_pfn << PAGE_SHIFT, |
5710 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
5711 | #else |
5712 | start_pfn = node_start_pfn; | |
7960aedd ZY |
5713 | #endif |
5714 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
5715 | zones_size, zholes_size); | |
1da177e4 LT |
5716 | |
5717 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
5718 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
5719 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
5720 | nid, (unsigned long)pgdat, | |
5721 | (unsigned long)pgdat->node_mem_map); | |
5722 | #endif | |
1da177e4 | 5723 | |
7f3eb55b | 5724 | free_area_init_core(pgdat); |
1da177e4 LT |
5725 | } |
5726 | ||
0ee332c1 | 5727 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
5728 | |
5729 | #if MAX_NUMNODES > 1 | |
5730 | /* | |
5731 | * Figure out the number of possible node ids. | |
5732 | */ | |
f9872caf | 5733 | void __init setup_nr_node_ids(void) |
418508c1 | 5734 | { |
904a9553 | 5735 | unsigned int highest; |
418508c1 | 5736 | |
904a9553 | 5737 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
5738 | nr_node_ids = highest + 1; |
5739 | } | |
418508c1 MS |
5740 | #endif |
5741 | ||
1e01979c TH |
5742 | /** |
5743 | * node_map_pfn_alignment - determine the maximum internode alignment | |
5744 | * | |
5745 | * This function should be called after node map is populated and sorted. | |
5746 | * It calculates the maximum power of two alignment which can distinguish | |
5747 | * all the nodes. | |
5748 | * | |
5749 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
5750 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
5751 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
5752 | * shifted, 1GiB is enough and this function will indicate so. | |
5753 | * | |
5754 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
5755 | * model has fine enough granularity to avoid incorrect mapping for the | |
5756 | * populated node map. | |
5757 | * | |
5758 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
5759 | * requirement (single node). | |
5760 | */ | |
5761 | unsigned long __init node_map_pfn_alignment(void) | |
5762 | { | |
5763 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 5764 | unsigned long start, end, mask; |
1e01979c | 5765 | int last_nid = -1; |
c13291a5 | 5766 | int i, nid; |
1e01979c | 5767 | |
c13291a5 | 5768 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5769 | if (!start || last_nid < 0 || last_nid == nid) { |
5770 | last_nid = nid; | |
5771 | last_end = end; | |
5772 | continue; | |
5773 | } | |
5774 | ||
5775 | /* | |
5776 | * Start with a mask granular enough to pin-point to the | |
5777 | * start pfn and tick off bits one-by-one until it becomes | |
5778 | * too coarse to separate the current node from the last. | |
5779 | */ | |
5780 | mask = ~((1 << __ffs(start)) - 1); | |
5781 | while (mask && last_end <= (start & (mask << 1))) | |
5782 | mask <<= 1; | |
5783 | ||
5784 | /* accumulate all internode masks */ | |
5785 | accl_mask |= mask; | |
5786 | } | |
5787 | ||
5788 | /* convert mask to number of pages */ | |
5789 | return ~accl_mask + 1; | |
5790 | } | |
5791 | ||
a6af2bc3 | 5792 | /* Find the lowest pfn for a node */ |
b69a7288 | 5793 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 5794 | { |
a6af2bc3 | 5795 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
5796 | unsigned long start_pfn; |
5797 | int i; | |
1abbfb41 | 5798 | |
c13291a5 TH |
5799 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
5800 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 5801 | |
a6af2bc3 | 5802 | if (min_pfn == ULONG_MAX) { |
1170532b | 5803 | pr_warn("Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
5804 | return 0; |
5805 | } | |
5806 | ||
5807 | return min_pfn; | |
c713216d MG |
5808 | } |
5809 | ||
5810 | /** | |
5811 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
5812 | * | |
5813 | * It returns the minimum PFN based on information provided via | |
7d018176 | 5814 | * memblock_set_node(). |
c713216d MG |
5815 | */ |
5816 | unsigned long __init find_min_pfn_with_active_regions(void) | |
5817 | { | |
5818 | return find_min_pfn_for_node(MAX_NUMNODES); | |
5819 | } | |
5820 | ||
37b07e41 LS |
5821 | /* |
5822 | * early_calculate_totalpages() | |
5823 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 5824 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 5825 | */ |
484f51f8 | 5826 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 5827 | { |
7e63efef | 5828 | unsigned long totalpages = 0; |
c13291a5 TH |
5829 | unsigned long start_pfn, end_pfn; |
5830 | int i, nid; | |
5831 | ||
5832 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
5833 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 5834 | |
37b07e41 LS |
5835 | totalpages += pages; |
5836 | if (pages) | |
4b0ef1fe | 5837 | node_set_state(nid, N_MEMORY); |
37b07e41 | 5838 | } |
b8af2941 | 5839 | return totalpages; |
7e63efef MG |
5840 | } |
5841 | ||
2a1e274a MG |
5842 | /* |
5843 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
5844 | * is spread evenly between nodes as long as the nodes have enough | |
5845 | * memory. When they don't, some nodes will have more kernelcore than | |
5846 | * others | |
5847 | */ | |
b224ef85 | 5848 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
5849 | { |
5850 | int i, nid; | |
5851 | unsigned long usable_startpfn; | |
5852 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 5853 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 5854 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 5855 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 5856 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 5857 | struct memblock_region *r; |
b2f3eebe TC |
5858 | |
5859 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
5860 | find_usable_zone_for_movable(); | |
5861 | ||
5862 | /* | |
5863 | * If movable_node is specified, ignore kernelcore and movablecore | |
5864 | * options. | |
5865 | */ | |
5866 | if (movable_node_is_enabled()) { | |
136199f0 EM |
5867 | for_each_memblock(memory, r) { |
5868 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
5869 | continue; |
5870 | ||
136199f0 | 5871 | nid = r->nid; |
b2f3eebe | 5872 | |
136199f0 | 5873 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
5874 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
5875 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5876 | usable_startpfn; | |
5877 | } | |
5878 | ||
5879 | goto out2; | |
5880 | } | |
2a1e274a | 5881 | |
342332e6 TI |
5882 | /* |
5883 | * If kernelcore=mirror is specified, ignore movablecore option | |
5884 | */ | |
5885 | if (mirrored_kernelcore) { | |
5886 | bool mem_below_4gb_not_mirrored = false; | |
5887 | ||
5888 | for_each_memblock(memory, r) { | |
5889 | if (memblock_is_mirror(r)) | |
5890 | continue; | |
5891 | ||
5892 | nid = r->nid; | |
5893 | ||
5894 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
5895 | ||
5896 | if (usable_startpfn < 0x100000) { | |
5897 | mem_below_4gb_not_mirrored = true; | |
5898 | continue; | |
5899 | } | |
5900 | ||
5901 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
5902 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5903 | usable_startpfn; | |
5904 | } | |
5905 | ||
5906 | if (mem_below_4gb_not_mirrored) | |
5907 | pr_warn("This configuration results in unmirrored kernel memory."); | |
5908 | ||
5909 | goto out2; | |
5910 | } | |
5911 | ||
7e63efef | 5912 | /* |
b2f3eebe | 5913 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
5914 | * kernelcore that corresponds so that memory usable for |
5915 | * any allocation type is evenly spread. If both kernelcore | |
5916 | * and movablecore are specified, then the value of kernelcore | |
5917 | * will be used for required_kernelcore if it's greater than | |
5918 | * what movablecore would have allowed. | |
5919 | */ | |
5920 | if (required_movablecore) { | |
7e63efef MG |
5921 | unsigned long corepages; |
5922 | ||
5923 | /* | |
5924 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
5925 | * was requested by the user | |
5926 | */ | |
5927 | required_movablecore = | |
5928 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 5929 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
5930 | corepages = totalpages - required_movablecore; |
5931 | ||
5932 | required_kernelcore = max(required_kernelcore, corepages); | |
5933 | } | |
5934 | ||
bde304bd XQ |
5935 | /* |
5936 | * If kernelcore was not specified or kernelcore size is larger | |
5937 | * than totalpages, there is no ZONE_MOVABLE. | |
5938 | */ | |
5939 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 5940 | goto out; |
2a1e274a MG |
5941 | |
5942 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
5943 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
5944 | ||
5945 | restart: | |
5946 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
5947 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 5948 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
5949 | unsigned long start_pfn, end_pfn; |
5950 | ||
2a1e274a MG |
5951 | /* |
5952 | * Recalculate kernelcore_node if the division per node | |
5953 | * now exceeds what is necessary to satisfy the requested | |
5954 | * amount of memory for the kernel | |
5955 | */ | |
5956 | if (required_kernelcore < kernelcore_node) | |
5957 | kernelcore_node = required_kernelcore / usable_nodes; | |
5958 | ||
5959 | /* | |
5960 | * As the map is walked, we track how much memory is usable | |
5961 | * by the kernel using kernelcore_remaining. When it is | |
5962 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
5963 | */ | |
5964 | kernelcore_remaining = kernelcore_node; | |
5965 | ||
5966 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 5967 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
5968 | unsigned long size_pages; |
5969 | ||
c13291a5 | 5970 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
5971 | if (start_pfn >= end_pfn) |
5972 | continue; | |
5973 | ||
5974 | /* Account for what is only usable for kernelcore */ | |
5975 | if (start_pfn < usable_startpfn) { | |
5976 | unsigned long kernel_pages; | |
5977 | kernel_pages = min(end_pfn, usable_startpfn) | |
5978 | - start_pfn; | |
5979 | ||
5980 | kernelcore_remaining -= min(kernel_pages, | |
5981 | kernelcore_remaining); | |
5982 | required_kernelcore -= min(kernel_pages, | |
5983 | required_kernelcore); | |
5984 | ||
5985 | /* Continue if range is now fully accounted */ | |
5986 | if (end_pfn <= usable_startpfn) { | |
5987 | ||
5988 | /* | |
5989 | * Push zone_movable_pfn to the end so | |
5990 | * that if we have to rebalance | |
5991 | * kernelcore across nodes, we will | |
5992 | * not double account here | |
5993 | */ | |
5994 | zone_movable_pfn[nid] = end_pfn; | |
5995 | continue; | |
5996 | } | |
5997 | start_pfn = usable_startpfn; | |
5998 | } | |
5999 | ||
6000 | /* | |
6001 | * The usable PFN range for ZONE_MOVABLE is from | |
6002 | * start_pfn->end_pfn. Calculate size_pages as the | |
6003 | * number of pages used as kernelcore | |
6004 | */ | |
6005 | size_pages = end_pfn - start_pfn; | |
6006 | if (size_pages > kernelcore_remaining) | |
6007 | size_pages = kernelcore_remaining; | |
6008 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
6009 | ||
6010 | /* | |
6011 | * Some kernelcore has been met, update counts and | |
6012 | * break if the kernelcore for this node has been | |
b8af2941 | 6013 | * satisfied |
2a1e274a MG |
6014 | */ |
6015 | required_kernelcore -= min(required_kernelcore, | |
6016 | size_pages); | |
6017 | kernelcore_remaining -= size_pages; | |
6018 | if (!kernelcore_remaining) | |
6019 | break; | |
6020 | } | |
6021 | } | |
6022 | ||
6023 | /* | |
6024 | * If there is still required_kernelcore, we do another pass with one | |
6025 | * less node in the count. This will push zone_movable_pfn[nid] further | |
6026 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 6027 | * satisfied |
2a1e274a MG |
6028 | */ |
6029 | usable_nodes--; | |
6030 | if (usable_nodes && required_kernelcore > usable_nodes) | |
6031 | goto restart; | |
6032 | ||
b2f3eebe | 6033 | out2: |
2a1e274a MG |
6034 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6035 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
6036 | zone_movable_pfn[nid] = | |
6037 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 6038 | |
20e6926d | 6039 | out: |
66918dcd | 6040 | /* restore the node_state */ |
4b0ef1fe | 6041 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
6042 | } |
6043 | ||
4b0ef1fe LJ |
6044 | /* Any regular or high memory on that node ? */ |
6045 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 6046 | { |
37b07e41 LS |
6047 | enum zone_type zone_type; |
6048 | ||
4b0ef1fe LJ |
6049 | if (N_MEMORY == N_NORMAL_MEMORY) |
6050 | return; | |
6051 | ||
6052 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 6053 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 6054 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
6055 | node_set_state(nid, N_HIGH_MEMORY); |
6056 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
6057 | zone_type <= ZONE_NORMAL) | |
6058 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
6059 | break; |
6060 | } | |
37b07e41 | 6061 | } |
37b07e41 LS |
6062 | } |
6063 | ||
c713216d MG |
6064 | /** |
6065 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 6066 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
6067 | * |
6068 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 6069 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
6070 | * zone in each node and their holes is calculated. If the maximum PFN |
6071 | * between two adjacent zones match, it is assumed that the zone is empty. | |
6072 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
6073 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
6074 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
6075 | * at arch_max_dma_pfn. | |
6076 | */ | |
6077 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
6078 | { | |
c13291a5 TH |
6079 | unsigned long start_pfn, end_pfn; |
6080 | int i, nid; | |
a6af2bc3 | 6081 | |
c713216d MG |
6082 | /* Record where the zone boundaries are */ |
6083 | memset(arch_zone_lowest_possible_pfn, 0, | |
6084 | sizeof(arch_zone_lowest_possible_pfn)); | |
6085 | memset(arch_zone_highest_possible_pfn, 0, | |
6086 | sizeof(arch_zone_highest_possible_pfn)); | |
6087 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
6088 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
6089 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
6090 | if (i == ZONE_MOVABLE) |
6091 | continue; | |
c713216d MG |
6092 | arch_zone_lowest_possible_pfn[i] = |
6093 | arch_zone_highest_possible_pfn[i-1]; | |
6094 | arch_zone_highest_possible_pfn[i] = | |
6095 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
6096 | } | |
2a1e274a MG |
6097 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
6098 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
6099 | ||
6100 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
6101 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 6102 | find_zone_movable_pfns_for_nodes(); |
c713216d | 6103 | |
c713216d | 6104 | /* Print out the zone ranges */ |
f88dfff5 | 6105 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
6106 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6107 | if (i == ZONE_MOVABLE) | |
6108 | continue; | |
f88dfff5 | 6109 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
6110 | if (arch_zone_lowest_possible_pfn[i] == |
6111 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 6112 | pr_cont("empty\n"); |
72f0ba02 | 6113 | else |
8d29e18a JG |
6114 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6115 | (u64)arch_zone_lowest_possible_pfn[i] | |
6116 | << PAGE_SHIFT, | |
6117 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 6118 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
6119 | } |
6120 | ||
6121 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6122 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6123 | for (i = 0; i < MAX_NUMNODES; i++) { |
6124 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6125 | pr_info(" Node %d: %#018Lx\n", i, |
6126 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6127 | } |
c713216d | 6128 | |
f2d52fe5 | 6129 | /* Print out the early node map */ |
f88dfff5 | 6130 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6131 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6132 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6133 | (u64)start_pfn << PAGE_SHIFT, | |
6134 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6135 | |
6136 | /* Initialise every node */ | |
708614e6 | 6137 | mminit_verify_pageflags_layout(); |
8ef82866 | 6138 | setup_nr_node_ids(); |
c713216d MG |
6139 | for_each_online_node(nid) { |
6140 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6141 | free_area_init_node(nid, NULL, |
c713216d | 6142 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6143 | |
6144 | /* Any memory on that node */ | |
6145 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6146 | node_set_state(nid, N_MEMORY); |
6147 | check_for_memory(pgdat, nid); | |
c713216d MG |
6148 | } |
6149 | } | |
2a1e274a | 6150 | |
7e63efef | 6151 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
6152 | { |
6153 | unsigned long long coremem; | |
6154 | if (!p) | |
6155 | return -EINVAL; | |
6156 | ||
6157 | coremem = memparse(p, &p); | |
7e63efef | 6158 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 6159 | |
7e63efef | 6160 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
6161 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6162 | ||
6163 | return 0; | |
6164 | } | |
ed7ed365 | 6165 | |
7e63efef MG |
6166 | /* |
6167 | * kernelcore=size sets the amount of memory for use for allocations that | |
6168 | * cannot be reclaimed or migrated. | |
6169 | */ | |
6170 | static int __init cmdline_parse_kernelcore(char *p) | |
6171 | { | |
342332e6 TI |
6172 | /* parse kernelcore=mirror */ |
6173 | if (parse_option_str(p, "mirror")) { | |
6174 | mirrored_kernelcore = true; | |
6175 | return 0; | |
6176 | } | |
6177 | ||
7e63efef MG |
6178 | return cmdline_parse_core(p, &required_kernelcore); |
6179 | } | |
6180 | ||
6181 | /* | |
6182 | * movablecore=size sets the amount of memory for use for allocations that | |
6183 | * can be reclaimed or migrated. | |
6184 | */ | |
6185 | static int __init cmdline_parse_movablecore(char *p) | |
6186 | { | |
6187 | return cmdline_parse_core(p, &required_movablecore); | |
6188 | } | |
6189 | ||
ed7ed365 | 6190 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6191 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6192 | |
0ee332c1 | 6193 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6194 | |
c3d5f5f0 JL |
6195 | void adjust_managed_page_count(struct page *page, long count) |
6196 | { | |
6197 | spin_lock(&managed_page_count_lock); | |
6198 | page_zone(page)->managed_pages += count; | |
6199 | totalram_pages += count; | |
3dcc0571 JL |
6200 | #ifdef CONFIG_HIGHMEM |
6201 | if (PageHighMem(page)) | |
6202 | totalhigh_pages += count; | |
6203 | #endif | |
c3d5f5f0 JL |
6204 | spin_unlock(&managed_page_count_lock); |
6205 | } | |
3dcc0571 | 6206 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6207 | |
11199692 | 6208 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6209 | { |
11199692 JL |
6210 | void *pos; |
6211 | unsigned long pages = 0; | |
69afade7 | 6212 | |
11199692 JL |
6213 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6214 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6215 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 6216 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
6217 | memset(pos, poison, PAGE_SIZE); |
6218 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
6219 | } |
6220 | ||
6221 | if (pages && s) | |
11199692 | 6222 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
6223 | s, pages << (PAGE_SHIFT - 10), start, end); |
6224 | ||
6225 | return pages; | |
6226 | } | |
11199692 | 6227 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6228 | |
cfa11e08 JL |
6229 | #ifdef CONFIG_HIGHMEM |
6230 | void free_highmem_page(struct page *page) | |
6231 | { | |
6232 | __free_reserved_page(page); | |
6233 | totalram_pages++; | |
7b4b2a0d | 6234 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6235 | totalhigh_pages++; |
6236 | } | |
6237 | #endif | |
6238 | ||
7ee3d4e8 JL |
6239 | |
6240 | void __init mem_init_print_info(const char *str) | |
6241 | { | |
6242 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
6243 | unsigned long init_code_size, init_data_size; | |
6244 | ||
6245 | physpages = get_num_physpages(); | |
6246 | codesize = _etext - _stext; | |
6247 | datasize = _edata - _sdata; | |
6248 | rosize = __end_rodata - __start_rodata; | |
6249 | bss_size = __bss_stop - __bss_start; | |
6250 | init_data_size = __init_end - __init_begin; | |
6251 | init_code_size = _einittext - _sinittext; | |
6252 | ||
6253 | /* | |
6254 | * Detect special cases and adjust section sizes accordingly: | |
6255 | * 1) .init.* may be embedded into .data sections | |
6256 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
6257 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
6258 | * 3) .rodata.* may be embedded into .text or .data sections. | |
6259 | */ | |
6260 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
6261 | do { \ |
6262 | if (start <= pos && pos < end && size > adj) \ | |
6263 | size -= adj; \ | |
6264 | } while (0) | |
7ee3d4e8 JL |
6265 | |
6266 | adj_init_size(__init_begin, __init_end, init_data_size, | |
6267 | _sinittext, init_code_size); | |
6268 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
6269 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
6270 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
6271 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
6272 | ||
6273 | #undef adj_init_size | |
6274 | ||
756a025f | 6275 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 | 6276 | #ifdef CONFIG_HIGHMEM |
756a025f | 6277 | ", %luK highmem" |
7ee3d4e8 | 6278 | #endif |
756a025f JP |
6279 | "%s%s)\n", |
6280 | nr_free_pages() << (PAGE_SHIFT - 10), | |
6281 | physpages << (PAGE_SHIFT - 10), | |
6282 | codesize >> 10, datasize >> 10, rosize >> 10, | |
6283 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
6284 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), | |
6285 | totalcma_pages << (PAGE_SHIFT - 10), | |
7ee3d4e8 | 6286 | #ifdef CONFIG_HIGHMEM |
756a025f | 6287 | totalhigh_pages << (PAGE_SHIFT - 10), |
7ee3d4e8 | 6288 | #endif |
756a025f | 6289 | str ? ", " : "", str ? str : ""); |
7ee3d4e8 JL |
6290 | } |
6291 | ||
0e0b864e | 6292 | /** |
88ca3b94 RD |
6293 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6294 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 6295 | * |
013110a7 | 6296 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
6297 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6298 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
6299 | * function may optionally be used to account for unfreeable pages in the |
6300 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
6301 | * smaller per-cpu batchsize. | |
0e0b864e MG |
6302 | */ |
6303 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
6304 | { | |
6305 | dma_reserve = new_dma_reserve; | |
6306 | } | |
6307 | ||
1da177e4 LT |
6308 | void __init free_area_init(unsigned long *zones_size) |
6309 | { | |
9109fb7b | 6310 | free_area_init_node(0, zones_size, |
1da177e4 LT |
6311 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6312 | } | |
1da177e4 | 6313 | |
1da177e4 LT |
6314 | static int page_alloc_cpu_notify(struct notifier_block *self, |
6315 | unsigned long action, void *hcpu) | |
6316 | { | |
6317 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 6318 | |
8bb78442 | 6319 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 6320 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
6321 | drain_pages(cpu); |
6322 | ||
6323 | /* | |
6324 | * Spill the event counters of the dead processor | |
6325 | * into the current processors event counters. | |
6326 | * This artificially elevates the count of the current | |
6327 | * processor. | |
6328 | */ | |
f8891e5e | 6329 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
6330 | |
6331 | /* | |
6332 | * Zero the differential counters of the dead processor | |
6333 | * so that the vm statistics are consistent. | |
6334 | * | |
6335 | * This is only okay since the processor is dead and cannot | |
6336 | * race with what we are doing. | |
6337 | */ | |
2bb921e5 | 6338 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
6339 | } |
6340 | return NOTIFY_OK; | |
6341 | } | |
1da177e4 LT |
6342 | |
6343 | void __init page_alloc_init(void) | |
6344 | { | |
6345 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
6346 | } | |
6347 | ||
cb45b0e9 | 6348 | /* |
34b10060 | 6349 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
6350 | * or min_free_kbytes changes. |
6351 | */ | |
6352 | static void calculate_totalreserve_pages(void) | |
6353 | { | |
6354 | struct pglist_data *pgdat; | |
6355 | unsigned long reserve_pages = 0; | |
2f6726e5 | 6356 | enum zone_type i, j; |
cb45b0e9 HA |
6357 | |
6358 | for_each_online_pgdat(pgdat) { | |
6359 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
6360 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 6361 | long max = 0; |
cb45b0e9 HA |
6362 | |
6363 | /* Find valid and maximum lowmem_reserve in the zone */ | |
6364 | for (j = i; j < MAX_NR_ZONES; j++) { | |
6365 | if (zone->lowmem_reserve[j] > max) | |
6366 | max = zone->lowmem_reserve[j]; | |
6367 | } | |
6368 | ||
41858966 MG |
6369 | /* we treat the high watermark as reserved pages. */ |
6370 | max += high_wmark_pages(zone); | |
cb45b0e9 | 6371 | |
b40da049 JL |
6372 | if (max > zone->managed_pages) |
6373 | max = zone->managed_pages; | |
a8d01437 JW |
6374 | |
6375 | zone->totalreserve_pages = max; | |
6376 | ||
cb45b0e9 HA |
6377 | reserve_pages += max; |
6378 | } | |
6379 | } | |
6380 | totalreserve_pages = reserve_pages; | |
6381 | } | |
6382 | ||
1da177e4 LT |
6383 | /* |
6384 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 6385 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
6386 | * has a correct pages reserved value, so an adequate number of |
6387 | * pages are left in the zone after a successful __alloc_pages(). | |
6388 | */ | |
6389 | static void setup_per_zone_lowmem_reserve(void) | |
6390 | { | |
6391 | struct pglist_data *pgdat; | |
2f6726e5 | 6392 | enum zone_type j, idx; |
1da177e4 | 6393 | |
ec936fc5 | 6394 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
6395 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6396 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 6397 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
6398 | |
6399 | zone->lowmem_reserve[j] = 0; | |
6400 | ||
2f6726e5 CL |
6401 | idx = j; |
6402 | while (idx) { | |
1da177e4 LT |
6403 | struct zone *lower_zone; |
6404 | ||
2f6726e5 CL |
6405 | idx--; |
6406 | ||
1da177e4 LT |
6407 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6408 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
6409 | ||
6410 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 6411 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 6412 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 6413 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
6414 | } |
6415 | } | |
6416 | } | |
cb45b0e9 HA |
6417 | |
6418 | /* update totalreserve_pages */ | |
6419 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6420 | } |
6421 | ||
cfd3da1e | 6422 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
6423 | { |
6424 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
6425 | unsigned long lowmem_pages = 0; | |
6426 | struct zone *zone; | |
6427 | unsigned long flags; | |
6428 | ||
6429 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
6430 | for_each_zone(zone) { | |
6431 | if (!is_highmem(zone)) | |
b40da049 | 6432 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
6433 | } |
6434 | ||
6435 | for_each_zone(zone) { | |
ac924c60 AM |
6436 | u64 tmp; |
6437 | ||
1125b4e3 | 6438 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 6439 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 6440 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
6441 | if (is_highmem(zone)) { |
6442 | /* | |
669ed175 NP |
6443 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6444 | * need highmem pages, so cap pages_min to a small | |
6445 | * value here. | |
6446 | * | |
41858966 | 6447 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 6448 | * deltas control asynch page reclaim, and so should |
669ed175 | 6449 | * not be capped for highmem. |
1da177e4 | 6450 | */ |
90ae8d67 | 6451 | unsigned long min_pages; |
1da177e4 | 6452 | |
b40da049 | 6453 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 6454 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 6455 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 6456 | } else { |
669ed175 NP |
6457 | /* |
6458 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
6459 | * proportionate to the zone's size. |
6460 | */ | |
41858966 | 6461 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
6462 | } |
6463 | ||
795ae7a0 JW |
6464 | /* |
6465 | * Set the kswapd watermarks distance according to the | |
6466 | * scale factor in proportion to available memory, but | |
6467 | * ensure a minimum size on small systems. | |
6468 | */ | |
6469 | tmp = max_t(u64, tmp >> 2, | |
6470 | mult_frac(zone->managed_pages, | |
6471 | watermark_scale_factor, 10000)); | |
6472 | ||
6473 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
6474 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 6475 | |
81c0a2bb | 6476 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
abe5f972 JW |
6477 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
6478 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
81c0a2bb | 6479 | |
1125b4e3 | 6480 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 6481 | } |
cb45b0e9 HA |
6482 | |
6483 | /* update totalreserve_pages */ | |
6484 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6485 | } |
6486 | ||
cfd3da1e MG |
6487 | /** |
6488 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
6489 | * or when memory is hot-{added|removed} | |
6490 | * | |
6491 | * Ensures that the watermark[min,low,high] values for each zone are set | |
6492 | * correctly with respect to min_free_kbytes. | |
6493 | */ | |
6494 | void setup_per_zone_wmarks(void) | |
6495 | { | |
6496 | mutex_lock(&zonelists_mutex); | |
6497 | __setup_per_zone_wmarks(); | |
6498 | mutex_unlock(&zonelists_mutex); | |
6499 | } | |
6500 | ||
55a4462a | 6501 | /* |
556adecb RR |
6502 | * The inactive anon list should be small enough that the VM never has to |
6503 | * do too much work, but large enough that each inactive page has a chance | |
6504 | * to be referenced again before it is swapped out. | |
6505 | * | |
6506 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
6507 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
6508 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
6509 | * the anonymous pages are kept on the inactive list. | |
6510 | * | |
6511 | * total target max | |
6512 | * memory ratio inactive anon | |
6513 | * ------------------------------------- | |
6514 | * 10MB 1 5MB | |
6515 | * 100MB 1 50MB | |
6516 | * 1GB 3 250MB | |
6517 | * 10GB 10 0.9GB | |
6518 | * 100GB 31 3GB | |
6519 | * 1TB 101 10GB | |
6520 | * 10TB 320 32GB | |
6521 | */ | |
1b79acc9 | 6522 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 6523 | { |
96cb4df5 | 6524 | unsigned int gb, ratio; |
556adecb | 6525 | |
96cb4df5 | 6526 | /* Zone size in gigabytes */ |
b40da049 | 6527 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 6528 | if (gb) |
556adecb | 6529 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
6530 | else |
6531 | ratio = 1; | |
556adecb | 6532 | |
96cb4df5 MK |
6533 | zone->inactive_ratio = ratio; |
6534 | } | |
556adecb | 6535 | |
839a4fcc | 6536 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
6537 | { |
6538 | struct zone *zone; | |
6539 | ||
6540 | for_each_zone(zone) | |
6541 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
6542 | } |
6543 | ||
1da177e4 LT |
6544 | /* |
6545 | * Initialise min_free_kbytes. | |
6546 | * | |
6547 | * For small machines we want it small (128k min). For large machines | |
6548 | * we want it large (64MB max). But it is not linear, because network | |
6549 | * bandwidth does not increase linearly with machine size. We use | |
6550 | * | |
b8af2941 | 6551 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6552 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6553 | * | |
6554 | * which yields | |
6555 | * | |
6556 | * 16MB: 512k | |
6557 | * 32MB: 724k | |
6558 | * 64MB: 1024k | |
6559 | * 128MB: 1448k | |
6560 | * 256MB: 2048k | |
6561 | * 512MB: 2896k | |
6562 | * 1024MB: 4096k | |
6563 | * 2048MB: 5792k | |
6564 | * 4096MB: 8192k | |
6565 | * 8192MB: 11584k | |
6566 | * 16384MB: 16384k | |
6567 | */ | |
1b79acc9 | 6568 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6569 | { |
6570 | unsigned long lowmem_kbytes; | |
5f12733e | 6571 | int new_min_free_kbytes; |
1da177e4 LT |
6572 | |
6573 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6574 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6575 | ||
6576 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6577 | min_free_kbytes = new_min_free_kbytes; | |
6578 | if (min_free_kbytes < 128) | |
6579 | min_free_kbytes = 128; | |
6580 | if (min_free_kbytes > 65536) | |
6581 | min_free_kbytes = 65536; | |
6582 | } else { | |
6583 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6584 | new_min_free_kbytes, user_min_free_kbytes); | |
6585 | } | |
bc75d33f | 6586 | setup_per_zone_wmarks(); |
a6cccdc3 | 6587 | refresh_zone_stat_thresholds(); |
1da177e4 | 6588 | setup_per_zone_lowmem_reserve(); |
556adecb | 6589 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
6590 | return 0; |
6591 | } | |
bc22af74 | 6592 | core_initcall(init_per_zone_wmark_min) |
1da177e4 LT |
6593 | |
6594 | /* | |
b8af2941 | 6595 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6596 | * that we can call two helper functions whenever min_free_kbytes |
6597 | * changes. | |
6598 | */ | |
cccad5b9 | 6599 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6600 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6601 | { |
da8c757b HP |
6602 | int rc; |
6603 | ||
6604 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6605 | if (rc) | |
6606 | return rc; | |
6607 | ||
5f12733e MH |
6608 | if (write) { |
6609 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6610 | setup_per_zone_wmarks(); |
5f12733e | 6611 | } |
1da177e4 LT |
6612 | return 0; |
6613 | } | |
6614 | ||
795ae7a0 JW |
6615 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
6616 | void __user *buffer, size_t *length, loff_t *ppos) | |
6617 | { | |
6618 | int rc; | |
6619 | ||
6620 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6621 | if (rc) | |
6622 | return rc; | |
6623 | ||
6624 | if (write) | |
6625 | setup_per_zone_wmarks(); | |
6626 | ||
6627 | return 0; | |
6628 | } | |
6629 | ||
9614634f | 6630 | #ifdef CONFIG_NUMA |
cccad5b9 | 6631 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6632 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
6633 | { |
6634 | struct zone *zone; | |
6635 | int rc; | |
6636 | ||
8d65af78 | 6637 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
6638 | if (rc) |
6639 | return rc; | |
6640 | ||
6641 | for_each_zone(zone) | |
b40da049 | 6642 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
6643 | sysctl_min_unmapped_ratio) / 100; |
6644 | return 0; | |
6645 | } | |
0ff38490 | 6646 | |
cccad5b9 | 6647 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6648 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
6649 | { |
6650 | struct zone *zone; | |
6651 | int rc; | |
6652 | ||
8d65af78 | 6653 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
6654 | if (rc) |
6655 | return rc; | |
6656 | ||
6657 | for_each_zone(zone) | |
b40da049 | 6658 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
6659 | sysctl_min_slab_ratio) / 100; |
6660 | return 0; | |
6661 | } | |
9614634f CL |
6662 | #endif |
6663 | ||
1da177e4 LT |
6664 | /* |
6665 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
6666 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
6667 | * whenever sysctl_lowmem_reserve_ratio changes. | |
6668 | * | |
6669 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 6670 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
6671 | * if in function of the boot time zone sizes. |
6672 | */ | |
cccad5b9 | 6673 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6674 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6675 | { |
8d65af78 | 6676 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
6677 | setup_per_zone_lowmem_reserve(); |
6678 | return 0; | |
6679 | } | |
6680 | ||
8ad4b1fb RS |
6681 | /* |
6682 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
6683 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
6684 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 6685 | */ |
cccad5b9 | 6686 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6687 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
6688 | { |
6689 | struct zone *zone; | |
7cd2b0a3 | 6690 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
6691 | int ret; |
6692 | ||
7cd2b0a3 DR |
6693 | mutex_lock(&pcp_batch_high_lock); |
6694 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
6695 | ||
8d65af78 | 6696 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
6697 | if (!write || ret < 0) |
6698 | goto out; | |
6699 | ||
6700 | /* Sanity checking to avoid pcp imbalance */ | |
6701 | if (percpu_pagelist_fraction && | |
6702 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
6703 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
6704 | ret = -EINVAL; | |
6705 | goto out; | |
6706 | } | |
6707 | ||
6708 | /* No change? */ | |
6709 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
6710 | goto out; | |
c8e251fa | 6711 | |
364df0eb | 6712 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
6713 | unsigned int cpu; |
6714 | ||
22a7f12b | 6715 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
6716 | pageset_set_high_and_batch(zone, |
6717 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 6718 | } |
7cd2b0a3 | 6719 | out: |
c8e251fa | 6720 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6721 | return ret; |
8ad4b1fb RS |
6722 | } |
6723 | ||
a9919c79 | 6724 | #ifdef CONFIG_NUMA |
f034b5d4 | 6725 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 6726 | |
1da177e4 LT |
6727 | static int __init set_hashdist(char *str) |
6728 | { | |
6729 | if (!str) | |
6730 | return 0; | |
6731 | hashdist = simple_strtoul(str, &str, 0); | |
6732 | return 1; | |
6733 | } | |
6734 | __setup("hashdist=", set_hashdist); | |
6735 | #endif | |
6736 | ||
6737 | /* | |
6738 | * allocate a large system hash table from bootmem | |
6739 | * - it is assumed that the hash table must contain an exact power-of-2 | |
6740 | * quantity of entries | |
6741 | * - limit is the number of hash buckets, not the total allocation size | |
6742 | */ | |
6743 | void *__init alloc_large_system_hash(const char *tablename, | |
6744 | unsigned long bucketsize, | |
6745 | unsigned long numentries, | |
6746 | int scale, | |
6747 | int flags, | |
6748 | unsigned int *_hash_shift, | |
6749 | unsigned int *_hash_mask, | |
31fe62b9 TB |
6750 | unsigned long low_limit, |
6751 | unsigned long high_limit) | |
1da177e4 | 6752 | { |
31fe62b9 | 6753 | unsigned long long max = high_limit; |
1da177e4 LT |
6754 | unsigned long log2qty, size; |
6755 | void *table = NULL; | |
6756 | ||
6757 | /* allow the kernel cmdline to have a say */ | |
6758 | if (!numentries) { | |
6759 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 6760 | numentries = nr_kernel_pages; |
a7e83318 JZ |
6761 | |
6762 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
6763 | if (PAGE_SHIFT < 20) | |
6764 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
6765 | |
6766 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
6767 | if (scale > PAGE_SHIFT) | |
6768 | numentries >>= (scale - PAGE_SHIFT); | |
6769 | else | |
6770 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
6771 | |
6772 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
6773 | if (unlikely(flags & HASH_SMALL)) { |
6774 | /* Makes no sense without HASH_EARLY */ | |
6775 | WARN_ON(!(flags & HASH_EARLY)); | |
6776 | if (!(numentries >> *_hash_shift)) { | |
6777 | numentries = 1UL << *_hash_shift; | |
6778 | BUG_ON(!numentries); | |
6779 | } | |
6780 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 6781 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 6782 | } |
6e692ed3 | 6783 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
6784 | |
6785 | /* limit allocation size to 1/16 total memory by default */ | |
6786 | if (max == 0) { | |
6787 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
6788 | do_div(max, bucketsize); | |
6789 | } | |
074b8517 | 6790 | max = min(max, 0x80000000ULL); |
1da177e4 | 6791 | |
31fe62b9 TB |
6792 | if (numentries < low_limit) |
6793 | numentries = low_limit; | |
1da177e4 LT |
6794 | if (numentries > max) |
6795 | numentries = max; | |
6796 | ||
f0d1b0b3 | 6797 | log2qty = ilog2(numentries); |
1da177e4 LT |
6798 | |
6799 | do { | |
6800 | size = bucketsize << log2qty; | |
6801 | if (flags & HASH_EARLY) | |
6782832e | 6802 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
6803 | else if (hashdist) |
6804 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
6805 | else { | |
1037b83b ED |
6806 | /* |
6807 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
6808 | * some pages at the end of hash table which |
6809 | * alloc_pages_exact() automatically does | |
1037b83b | 6810 | */ |
264ef8a9 | 6811 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 6812 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
6813 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
6814 | } | |
1da177e4 LT |
6815 | } |
6816 | } while (!table && size > PAGE_SIZE && --log2qty); | |
6817 | ||
6818 | if (!table) | |
6819 | panic("Failed to allocate %s hash table\n", tablename); | |
6820 | ||
1170532b JP |
6821 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
6822 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); | |
1da177e4 LT |
6823 | |
6824 | if (_hash_shift) | |
6825 | *_hash_shift = log2qty; | |
6826 | if (_hash_mask) | |
6827 | *_hash_mask = (1 << log2qty) - 1; | |
6828 | ||
6829 | return table; | |
6830 | } | |
a117e66e | 6831 | |
835c134e | 6832 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
f75fb889 | 6833 | static inline unsigned long *get_pageblock_bitmap(struct page *page, |
835c134e MG |
6834 | unsigned long pfn) |
6835 | { | |
6836 | #ifdef CONFIG_SPARSEMEM | |
6837 | return __pfn_to_section(pfn)->pageblock_flags; | |
6838 | #else | |
f75fb889 | 6839 | return page_zone(page)->pageblock_flags; |
835c134e MG |
6840 | #endif /* CONFIG_SPARSEMEM */ |
6841 | } | |
6842 | ||
f75fb889 | 6843 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) |
835c134e MG |
6844 | { |
6845 | #ifdef CONFIG_SPARSEMEM | |
6846 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 6847 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 6848 | #else |
f75fb889 | 6849 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 6850 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
6851 | #endif /* CONFIG_SPARSEMEM */ |
6852 | } | |
6853 | ||
6854 | /** | |
1aab4d77 | 6855 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e | 6856 | * @page: The page within the block of interest |
1aab4d77 RD |
6857 | * @pfn: The target page frame number |
6858 | * @end_bitidx: The last bit of interest to retrieve | |
6859 | * @mask: mask of bits that the caller is interested in | |
6860 | * | |
6861 | * Return: pageblock_bits flags | |
835c134e | 6862 | */ |
dc4b0caf | 6863 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
e58469ba MG |
6864 | unsigned long end_bitidx, |
6865 | unsigned long mask) | |
835c134e | 6866 | { |
835c134e | 6867 | unsigned long *bitmap; |
dc4b0caf | 6868 | unsigned long bitidx, word_bitidx; |
e58469ba | 6869 | unsigned long word; |
835c134e | 6870 | |
f75fb889 MG |
6871 | bitmap = get_pageblock_bitmap(page, pfn); |
6872 | bitidx = pfn_to_bitidx(page, pfn); | |
e58469ba MG |
6873 | word_bitidx = bitidx / BITS_PER_LONG; |
6874 | bitidx &= (BITS_PER_LONG-1); | |
835c134e | 6875 | |
e58469ba MG |
6876 | word = bitmap[word_bitidx]; |
6877 | bitidx += end_bitidx; | |
6878 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
835c134e MG |
6879 | } |
6880 | ||
6881 | /** | |
dc4b0caf | 6882 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e | 6883 | * @page: The page within the block of interest |
835c134e | 6884 | * @flags: The flags to set |
1aab4d77 RD |
6885 | * @pfn: The target page frame number |
6886 | * @end_bitidx: The last bit of interest | |
6887 | * @mask: mask of bits that the caller is interested in | |
835c134e | 6888 | */ |
dc4b0caf MG |
6889 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
6890 | unsigned long pfn, | |
e58469ba MG |
6891 | unsigned long end_bitidx, |
6892 | unsigned long mask) | |
835c134e | 6893 | { |
835c134e | 6894 | unsigned long *bitmap; |
dc4b0caf | 6895 | unsigned long bitidx, word_bitidx; |
e58469ba MG |
6896 | unsigned long old_word, word; |
6897 | ||
6898 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
835c134e | 6899 | |
f75fb889 MG |
6900 | bitmap = get_pageblock_bitmap(page, pfn); |
6901 | bitidx = pfn_to_bitidx(page, pfn); | |
e58469ba MG |
6902 | word_bitidx = bitidx / BITS_PER_LONG; |
6903 | bitidx &= (BITS_PER_LONG-1); | |
6904 | ||
f75fb889 | 6905 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); |
835c134e | 6906 | |
e58469ba MG |
6907 | bitidx += end_bitidx; |
6908 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
6909 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
6910 | ||
4db0c3c2 | 6911 | word = READ_ONCE(bitmap[word_bitidx]); |
e58469ba MG |
6912 | for (;;) { |
6913 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
6914 | if (word == old_word) | |
6915 | break; | |
6916 | word = old_word; | |
6917 | } | |
835c134e | 6918 | } |
a5d76b54 KH |
6919 | |
6920 | /* | |
80934513 MK |
6921 | * This function checks whether pageblock includes unmovable pages or not. |
6922 | * If @count is not zero, it is okay to include less @count unmovable pages | |
6923 | * | |
b8af2941 | 6924 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
6925 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
6926 | * expect this function should be exact. | |
a5d76b54 | 6927 | */ |
b023f468 WC |
6928 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
6929 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
6930 | { |
6931 | unsigned long pfn, iter, found; | |
47118af0 MN |
6932 | int mt; |
6933 | ||
49ac8255 KH |
6934 | /* |
6935 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 6936 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
6937 | */ |
6938 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 6939 | return false; |
47118af0 MN |
6940 | mt = get_pageblock_migratetype(page); |
6941 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 6942 | return false; |
49ac8255 KH |
6943 | |
6944 | pfn = page_to_pfn(page); | |
6945 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
6946 | unsigned long check = pfn + iter; | |
6947 | ||
29723fcc | 6948 | if (!pfn_valid_within(check)) |
49ac8255 | 6949 | continue; |
29723fcc | 6950 | |
49ac8255 | 6951 | page = pfn_to_page(check); |
c8721bbb NH |
6952 | |
6953 | /* | |
6954 | * Hugepages are not in LRU lists, but they're movable. | |
6955 | * We need not scan over tail pages bacause we don't | |
6956 | * handle each tail page individually in migration. | |
6957 | */ | |
6958 | if (PageHuge(page)) { | |
6959 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
6960 | continue; | |
6961 | } | |
6962 | ||
97d255c8 MK |
6963 | /* |
6964 | * We can't use page_count without pin a page | |
6965 | * because another CPU can free compound page. | |
6966 | * This check already skips compound tails of THP | |
0139aa7b | 6967 | * because their page->_refcount is zero at all time. |
97d255c8 | 6968 | */ |
fe896d18 | 6969 | if (!page_ref_count(page)) { |
49ac8255 KH |
6970 | if (PageBuddy(page)) |
6971 | iter += (1 << page_order(page)) - 1; | |
6972 | continue; | |
6973 | } | |
97d255c8 | 6974 | |
b023f468 WC |
6975 | /* |
6976 | * The HWPoisoned page may be not in buddy system, and | |
6977 | * page_count() is not 0. | |
6978 | */ | |
6979 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6980 | continue; | |
6981 | ||
49ac8255 KH |
6982 | if (!PageLRU(page)) |
6983 | found++; | |
6984 | /* | |
6b4f7799 JW |
6985 | * If there are RECLAIMABLE pages, we need to check |
6986 | * it. But now, memory offline itself doesn't call | |
6987 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
6988 | */ |
6989 | /* | |
6990 | * If the page is not RAM, page_count()should be 0. | |
6991 | * we don't need more check. This is an _used_ not-movable page. | |
6992 | * | |
6993 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6994 | * is set to both of a memory hole page and a _used_ kernel | |
6995 | * page at boot. | |
6996 | */ | |
6997 | if (found > count) | |
80934513 | 6998 | return true; |
49ac8255 | 6999 | } |
80934513 | 7000 | return false; |
49ac8255 KH |
7001 | } |
7002 | ||
7003 | bool is_pageblock_removable_nolock(struct page *page) | |
7004 | { | |
656a0706 MH |
7005 | struct zone *zone; |
7006 | unsigned long pfn; | |
687875fb MH |
7007 | |
7008 | /* | |
7009 | * We have to be careful here because we are iterating over memory | |
7010 | * sections which are not zone aware so we might end up outside of | |
7011 | * the zone but still within the section. | |
656a0706 MH |
7012 | * We have to take care about the node as well. If the node is offline |
7013 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 7014 | */ |
656a0706 MH |
7015 | if (!node_online(page_to_nid(page))) |
7016 | return false; | |
7017 | ||
7018 | zone = page_zone(page); | |
7019 | pfn = page_to_pfn(page); | |
108bcc96 | 7020 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
7021 | return false; |
7022 | ||
b023f468 | 7023 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 7024 | } |
0c0e6195 | 7025 | |
080fe206 | 7026 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
7027 | |
7028 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
7029 | { | |
7030 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7031 | pageblock_nr_pages) - 1); | |
7032 | } | |
7033 | ||
7034 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
7035 | { | |
7036 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
7037 | pageblock_nr_pages)); | |
7038 | } | |
7039 | ||
041d3a8c | 7040 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
7041 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7042 | unsigned long start, unsigned long end) | |
041d3a8c MN |
7043 | { |
7044 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 7045 | unsigned long nr_reclaimed; |
041d3a8c MN |
7046 | unsigned long pfn = start; |
7047 | unsigned int tries = 0; | |
7048 | int ret = 0; | |
7049 | ||
be49a6e1 | 7050 | migrate_prep(); |
041d3a8c | 7051 | |
bb13ffeb | 7052 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
7053 | if (fatal_signal_pending(current)) { |
7054 | ret = -EINTR; | |
7055 | break; | |
7056 | } | |
7057 | ||
bb13ffeb MG |
7058 | if (list_empty(&cc->migratepages)) { |
7059 | cc->nr_migratepages = 0; | |
edc2ca61 | 7060 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
7061 | if (!pfn) { |
7062 | ret = -EINTR; | |
7063 | break; | |
7064 | } | |
7065 | tries = 0; | |
7066 | } else if (++tries == 5) { | |
7067 | ret = ret < 0 ? ret : -EBUSY; | |
7068 | break; | |
7069 | } | |
7070 | ||
beb51eaa MK |
7071 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7072 | &cc->migratepages); | |
7073 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 7074 | |
9c620e2b | 7075 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 7076 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 7077 | } |
2a6f5124 SP |
7078 | if (ret < 0) { |
7079 | putback_movable_pages(&cc->migratepages); | |
7080 | return ret; | |
7081 | } | |
7082 | return 0; | |
041d3a8c MN |
7083 | } |
7084 | ||
7085 | /** | |
7086 | * alloc_contig_range() -- tries to allocate given range of pages | |
7087 | * @start: start PFN to allocate | |
7088 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
7089 | * @migratetype: migratetype of the underlaying pageblocks (either |
7090 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
7091 | * in range must have the same migratetype and it must | |
7092 | * be either of the two. | |
041d3a8c MN |
7093 | * |
7094 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
7095 | * aligned, however it's the caller's responsibility to guarantee that | |
7096 | * we are the only thread that changes migrate type of pageblocks the | |
7097 | * pages fall in. | |
7098 | * | |
7099 | * The PFN range must belong to a single zone. | |
7100 | * | |
7101 | * Returns zero on success or negative error code. On success all | |
7102 | * pages which PFN is in [start, end) are allocated for the caller and | |
7103 | * need to be freed with free_contig_range(). | |
7104 | */ | |
0815f3d8 MN |
7105 | int alloc_contig_range(unsigned long start, unsigned long end, |
7106 | unsigned migratetype) | |
041d3a8c | 7107 | { |
041d3a8c | 7108 | unsigned long outer_start, outer_end; |
d00181b9 KS |
7109 | unsigned int order; |
7110 | int ret = 0; | |
041d3a8c | 7111 | |
bb13ffeb MG |
7112 | struct compact_control cc = { |
7113 | .nr_migratepages = 0, | |
7114 | .order = -1, | |
7115 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7116 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
7117 | .ignore_skip_hint = true, |
7118 | }; | |
7119 | INIT_LIST_HEAD(&cc.migratepages); | |
7120 | ||
041d3a8c MN |
7121 | /* |
7122 | * What we do here is we mark all pageblocks in range as | |
7123 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7124 | * have different sizes, and due to the way page allocator | |
7125 | * work, we align the range to biggest of the two pages so | |
7126 | * that page allocator won't try to merge buddies from | |
7127 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7128 | * other migration type. | |
7129 | * | |
7130 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7131 | * migrate the pages from an unaligned range (ie. pages that | |
7132 | * we are interested in). This will put all the pages in | |
7133 | * range back to page allocator as MIGRATE_ISOLATE. | |
7134 | * | |
7135 | * When this is done, we take the pages in range from page | |
7136 | * allocator removing them from the buddy system. This way | |
7137 | * page allocator will never consider using them. | |
7138 | * | |
7139 | * This lets us mark the pageblocks back as | |
7140 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7141 | * aligned range but not in the unaligned, original range are | |
7142 | * put back to page allocator so that buddy can use them. | |
7143 | */ | |
7144 | ||
7145 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7146 | pfn_max_align_up(end), migratetype, |
7147 | false); | |
041d3a8c | 7148 | if (ret) |
86a595f9 | 7149 | return ret; |
041d3a8c | 7150 | |
8ef5849f JK |
7151 | /* |
7152 | * In case of -EBUSY, we'd like to know which page causes problem. | |
7153 | * So, just fall through. We will check it in test_pages_isolated(). | |
7154 | */ | |
bb13ffeb | 7155 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7156 | if (ret && ret != -EBUSY) |
041d3a8c MN |
7157 | goto done; |
7158 | ||
7159 | /* | |
7160 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7161 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7162 | * more, all pages in [start, end) are free in page allocator. | |
7163 | * What we are going to do is to allocate all pages from | |
7164 | * [start, end) (that is remove them from page allocator). | |
7165 | * | |
7166 | * The only problem is that pages at the beginning and at the | |
7167 | * end of interesting range may be not aligned with pages that | |
7168 | * page allocator holds, ie. they can be part of higher order | |
7169 | * pages. Because of this, we reserve the bigger range and | |
7170 | * once this is done free the pages we are not interested in. | |
7171 | * | |
7172 | * We don't have to hold zone->lock here because the pages are | |
7173 | * isolated thus they won't get removed from buddy. | |
7174 | */ | |
7175 | ||
7176 | lru_add_drain_all(); | |
510f5507 | 7177 | drain_all_pages(cc.zone); |
041d3a8c MN |
7178 | |
7179 | order = 0; | |
7180 | outer_start = start; | |
7181 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7182 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7183 | outer_start = start; |
7184 | break; | |
041d3a8c MN |
7185 | } |
7186 | outer_start &= ~0UL << order; | |
7187 | } | |
7188 | ||
8ef5849f JK |
7189 | if (outer_start != start) { |
7190 | order = page_order(pfn_to_page(outer_start)); | |
7191 | ||
7192 | /* | |
7193 | * outer_start page could be small order buddy page and | |
7194 | * it doesn't include start page. Adjust outer_start | |
7195 | * in this case to report failed page properly | |
7196 | * on tracepoint in test_pages_isolated() | |
7197 | */ | |
7198 | if (outer_start + (1UL << order) <= start) | |
7199 | outer_start = start; | |
7200 | } | |
7201 | ||
041d3a8c | 7202 | /* Make sure the range is really isolated. */ |
b023f468 | 7203 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
7204 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
7205 | __func__, outer_start, end); | |
041d3a8c MN |
7206 | ret = -EBUSY; |
7207 | goto done; | |
7208 | } | |
7209 | ||
49f223a9 | 7210 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7211 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7212 | if (!outer_end) { |
7213 | ret = -EBUSY; | |
7214 | goto done; | |
7215 | } | |
7216 | ||
7217 | /* Free head and tail (if any) */ | |
7218 | if (start != outer_start) | |
7219 | free_contig_range(outer_start, start - outer_start); | |
7220 | if (end != outer_end) | |
7221 | free_contig_range(end, outer_end - end); | |
7222 | ||
7223 | done: | |
7224 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7225 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7226 | return ret; |
7227 | } | |
7228 | ||
7229 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7230 | { | |
bcc2b02f MS |
7231 | unsigned int count = 0; |
7232 | ||
7233 | for (; nr_pages--; pfn++) { | |
7234 | struct page *page = pfn_to_page(pfn); | |
7235 | ||
7236 | count += page_count(page) != 1; | |
7237 | __free_page(page); | |
7238 | } | |
7239 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7240 | } |
7241 | #endif | |
7242 | ||
4ed7e022 | 7243 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7244 | /* |
7245 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7246 | * page high values need to be recalulated. | |
7247 | */ | |
4ed7e022 JL |
7248 | void __meminit zone_pcp_update(struct zone *zone) |
7249 | { | |
0a647f38 | 7250 | unsigned cpu; |
c8e251fa | 7251 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7252 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7253 | pageset_set_high_and_batch(zone, |
7254 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7255 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7256 | } |
7257 | #endif | |
7258 | ||
340175b7 JL |
7259 | void zone_pcp_reset(struct zone *zone) |
7260 | { | |
7261 | unsigned long flags; | |
5a883813 MK |
7262 | int cpu; |
7263 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7264 | |
7265 | /* avoid races with drain_pages() */ | |
7266 | local_irq_save(flags); | |
7267 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7268 | for_each_online_cpu(cpu) { |
7269 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7270 | drain_zonestat(zone, pset); | |
7271 | } | |
340175b7 JL |
7272 | free_percpu(zone->pageset); |
7273 | zone->pageset = &boot_pageset; | |
7274 | } | |
7275 | local_irq_restore(flags); | |
7276 | } | |
7277 | ||
6dcd73d7 | 7278 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 | 7279 | /* |
b9eb6319 JK |
7280 | * All pages in the range must be in a single zone and isolated |
7281 | * before calling this. | |
0c0e6195 KH |
7282 | */ |
7283 | void | |
7284 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7285 | { | |
7286 | struct page *page; | |
7287 | struct zone *zone; | |
7aeb09f9 | 7288 | unsigned int order, i; |
0c0e6195 KH |
7289 | unsigned long pfn; |
7290 | unsigned long flags; | |
7291 | /* find the first valid pfn */ | |
7292 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
7293 | if (pfn_valid(pfn)) | |
7294 | break; | |
7295 | if (pfn == end_pfn) | |
7296 | return; | |
7297 | zone = page_zone(pfn_to_page(pfn)); | |
7298 | spin_lock_irqsave(&zone->lock, flags); | |
7299 | pfn = start_pfn; | |
7300 | while (pfn < end_pfn) { | |
7301 | if (!pfn_valid(pfn)) { | |
7302 | pfn++; | |
7303 | continue; | |
7304 | } | |
7305 | page = pfn_to_page(pfn); | |
b023f468 WC |
7306 | /* |
7307 | * The HWPoisoned page may be not in buddy system, and | |
7308 | * page_count() is not 0. | |
7309 | */ | |
7310 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
7311 | pfn++; | |
7312 | SetPageReserved(page); | |
7313 | continue; | |
7314 | } | |
7315 | ||
0c0e6195 KH |
7316 | BUG_ON(page_count(page)); |
7317 | BUG_ON(!PageBuddy(page)); | |
7318 | order = page_order(page); | |
7319 | #ifdef CONFIG_DEBUG_VM | |
1170532b JP |
7320 | pr_info("remove from free list %lx %d %lx\n", |
7321 | pfn, 1 << order, end_pfn); | |
0c0e6195 KH |
7322 | #endif |
7323 | list_del(&page->lru); | |
7324 | rmv_page_order(page); | |
7325 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
7326 | for (i = 0; i < (1 << order); i++) |
7327 | SetPageReserved((page+i)); | |
7328 | pfn += (1 << order); | |
7329 | } | |
7330 | spin_unlock_irqrestore(&zone->lock, flags); | |
7331 | } | |
7332 | #endif | |
8d22ba1b | 7333 | |
8d22ba1b WF |
7334 | bool is_free_buddy_page(struct page *page) |
7335 | { | |
7336 | struct zone *zone = page_zone(page); | |
7337 | unsigned long pfn = page_to_pfn(page); | |
7338 | unsigned long flags; | |
7aeb09f9 | 7339 | unsigned int order; |
8d22ba1b WF |
7340 | |
7341 | spin_lock_irqsave(&zone->lock, flags); | |
7342 | for (order = 0; order < MAX_ORDER; order++) { | |
7343 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
7344 | ||
7345 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
7346 | break; | |
7347 | } | |
7348 | spin_unlock_irqrestore(&zone->lock, flags); | |
7349 | ||
7350 | return order < MAX_ORDER; | |
7351 | } |