]>
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 | { | |
311 | /* Always populate low zones for address-contrained allocations */ | |
312 | if (zone_end < pgdat_end_pfn(pgdat)) | |
313 | return true; | |
314 | ||
315 | /* Initialise at least 2G of the highest zone */ | |
316 | (*nr_initialised)++; | |
317 | if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) && | |
318 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { | |
319 | pgdat->first_deferred_pfn = pfn; | |
320 | return false; | |
321 | } | |
322 | ||
323 | return true; | |
324 | } | |
325 | #else | |
326 | static inline void reset_deferred_meminit(pg_data_t *pgdat) | |
327 | { | |
328 | } | |
329 | ||
330 | static inline bool early_page_uninitialised(unsigned long pfn) | |
331 | { | |
332 | return false; | |
333 | } | |
334 | ||
7e18adb4 MG |
335 | static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid) |
336 | { | |
337 | return false; | |
338 | } | |
339 | ||
3a80a7fa MG |
340 | static inline bool update_defer_init(pg_data_t *pgdat, |
341 | unsigned long pfn, unsigned long zone_end, | |
342 | unsigned long *nr_initialised) | |
343 | { | |
344 | return true; | |
345 | } | |
346 | #endif | |
347 | ||
348 | ||
ee6f509c | 349 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 350 | { |
5d0f3f72 KM |
351 | if (unlikely(page_group_by_mobility_disabled && |
352 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
353 | migratetype = MIGRATE_UNMOVABLE; |
354 | ||
b2a0ac88 MG |
355 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
356 | PB_migrate, PB_migrate_end); | |
357 | } | |
358 | ||
13e7444b | 359 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 360 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 361 | { |
bdc8cb98 DH |
362 | int ret = 0; |
363 | unsigned seq; | |
364 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 365 | unsigned long sp, start_pfn; |
c6a57e19 | 366 | |
bdc8cb98 DH |
367 | do { |
368 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
369 | start_pfn = zone->zone_start_pfn; |
370 | sp = zone->spanned_pages; | |
108bcc96 | 371 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
372 | ret = 1; |
373 | } while (zone_span_seqretry(zone, seq)); | |
374 | ||
b5e6a5a2 | 375 | if (ret) |
613813e8 DH |
376 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
377 | pfn, zone_to_nid(zone), zone->name, | |
378 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 379 | |
bdc8cb98 | 380 | return ret; |
c6a57e19 DH |
381 | } |
382 | ||
383 | static int page_is_consistent(struct zone *zone, struct page *page) | |
384 | { | |
14e07298 | 385 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 386 | return 0; |
1da177e4 | 387 | if (zone != page_zone(page)) |
c6a57e19 DH |
388 | return 0; |
389 | ||
390 | return 1; | |
391 | } | |
392 | /* | |
393 | * Temporary debugging check for pages not lying within a given zone. | |
394 | */ | |
395 | static int bad_range(struct zone *zone, struct page *page) | |
396 | { | |
397 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 398 | return 1; |
c6a57e19 DH |
399 | if (!page_is_consistent(zone, page)) |
400 | return 1; | |
401 | ||
1da177e4 LT |
402 | return 0; |
403 | } | |
13e7444b NP |
404 | #else |
405 | static inline int bad_range(struct zone *zone, struct page *page) | |
406 | { | |
407 | return 0; | |
408 | } | |
409 | #endif | |
410 | ||
d230dec1 KS |
411 | static void bad_page(struct page *page, const char *reason, |
412 | unsigned long bad_flags) | |
1da177e4 | 413 | { |
d936cf9b HD |
414 | static unsigned long resume; |
415 | static unsigned long nr_shown; | |
416 | static unsigned long nr_unshown; | |
417 | ||
2a7684a2 WF |
418 | /* Don't complain about poisoned pages */ |
419 | if (PageHWPoison(page)) { | |
22b751c3 | 420 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
421 | return; |
422 | } | |
423 | ||
d936cf9b HD |
424 | /* |
425 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
426 | * or allow a steady drip of one report per second. | |
427 | */ | |
428 | if (nr_shown == 60) { | |
429 | if (time_before(jiffies, resume)) { | |
430 | nr_unshown++; | |
431 | goto out; | |
432 | } | |
433 | if (nr_unshown) { | |
ff8e8116 | 434 | pr_alert( |
1e9e6365 | 435 | "BUG: Bad page state: %lu messages suppressed\n", |
d936cf9b HD |
436 | nr_unshown); |
437 | nr_unshown = 0; | |
438 | } | |
439 | nr_shown = 0; | |
440 | } | |
441 | if (nr_shown++ == 0) | |
442 | resume = jiffies + 60 * HZ; | |
443 | ||
ff8e8116 | 444 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 445 | current->comm, page_to_pfn(page)); |
ff8e8116 VB |
446 | __dump_page(page, reason); |
447 | bad_flags &= page->flags; | |
448 | if (bad_flags) | |
449 | pr_alert("bad because of flags: %#lx(%pGp)\n", | |
450 | bad_flags, &bad_flags); | |
4e462112 | 451 | dump_page_owner(page); |
3dc14741 | 452 | |
4f31888c | 453 | print_modules(); |
1da177e4 | 454 | dump_stack(); |
d936cf9b | 455 | out: |
8cc3b392 | 456 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 457 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 458 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
459 | } |
460 | ||
1da177e4 LT |
461 | /* |
462 | * Higher-order pages are called "compound pages". They are structured thusly: | |
463 | * | |
1d798ca3 | 464 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
1da177e4 | 465 | * |
1d798ca3 KS |
466 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
467 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. | |
1da177e4 | 468 | * |
1d798ca3 KS |
469 | * The first tail page's ->compound_dtor holds the offset in array of compound |
470 | * page destructors. See compound_page_dtors. | |
1da177e4 | 471 | * |
1d798ca3 | 472 | * The first tail page's ->compound_order holds the order of allocation. |
41d78ba5 | 473 | * This usage means that zero-order pages may not be compound. |
1da177e4 | 474 | */ |
d98c7a09 | 475 | |
9a982250 | 476 | void free_compound_page(struct page *page) |
d98c7a09 | 477 | { |
d85f3385 | 478 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
479 | } |
480 | ||
d00181b9 | 481 | void prep_compound_page(struct page *page, unsigned int order) |
18229df5 AW |
482 | { |
483 | int i; | |
484 | int nr_pages = 1 << order; | |
485 | ||
f1e61557 | 486 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
18229df5 AW |
487 | set_compound_order(page, order); |
488 | __SetPageHead(page); | |
489 | for (i = 1; i < nr_pages; i++) { | |
490 | struct page *p = page + i; | |
58a84aa9 | 491 | set_page_count(p, 0); |
1c290f64 | 492 | p->mapping = TAIL_MAPPING; |
1d798ca3 | 493 | set_compound_head(p, page); |
18229df5 | 494 | } |
53f9263b | 495 | atomic_set(compound_mapcount_ptr(page), -1); |
18229df5 AW |
496 | } |
497 | ||
c0a32fc5 SG |
498 | #ifdef CONFIG_DEBUG_PAGEALLOC |
499 | unsigned int _debug_guardpage_minorder; | |
ea6eabb0 CB |
500 | bool _debug_pagealloc_enabled __read_mostly |
501 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); | |
505f6d22 | 502 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
e30825f1 JK |
503 | bool _debug_guardpage_enabled __read_mostly; |
504 | ||
031bc574 JK |
505 | static int __init early_debug_pagealloc(char *buf) |
506 | { | |
507 | if (!buf) | |
508 | return -EINVAL; | |
509 | ||
510 | if (strcmp(buf, "on") == 0) | |
511 | _debug_pagealloc_enabled = true; | |
512 | ||
ea6eabb0 CB |
513 | if (strcmp(buf, "off") == 0) |
514 | _debug_pagealloc_enabled = false; | |
515 | ||
031bc574 JK |
516 | return 0; |
517 | } | |
518 | early_param("debug_pagealloc", early_debug_pagealloc); | |
519 | ||
e30825f1 JK |
520 | static bool need_debug_guardpage(void) |
521 | { | |
031bc574 JK |
522 | /* If we don't use debug_pagealloc, we don't need guard page */ |
523 | if (!debug_pagealloc_enabled()) | |
524 | return false; | |
525 | ||
e30825f1 JK |
526 | return true; |
527 | } | |
528 | ||
529 | static void init_debug_guardpage(void) | |
530 | { | |
031bc574 JK |
531 | if (!debug_pagealloc_enabled()) |
532 | return; | |
533 | ||
e30825f1 JK |
534 | _debug_guardpage_enabled = true; |
535 | } | |
536 | ||
537 | struct page_ext_operations debug_guardpage_ops = { | |
538 | .need = need_debug_guardpage, | |
539 | .init = init_debug_guardpage, | |
540 | }; | |
c0a32fc5 SG |
541 | |
542 | static int __init debug_guardpage_minorder_setup(char *buf) | |
543 | { | |
544 | unsigned long res; | |
545 | ||
546 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
547 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
548 | return 0; | |
549 | } | |
550 | _debug_guardpage_minorder = res; | |
551 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
552 | return 0; | |
553 | } | |
554 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
555 | ||
2847cf95 JK |
556 | static inline void set_page_guard(struct zone *zone, struct page *page, |
557 | unsigned int order, int migratetype) | |
c0a32fc5 | 558 | { |
e30825f1 JK |
559 | struct page_ext *page_ext; |
560 | ||
561 | if (!debug_guardpage_enabled()) | |
562 | return; | |
563 | ||
564 | page_ext = lookup_page_ext(page); | |
565 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
566 | ||
2847cf95 JK |
567 | INIT_LIST_HEAD(&page->lru); |
568 | set_page_private(page, order); | |
569 | /* Guard pages are not available for any usage */ | |
570 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); | |
c0a32fc5 SG |
571 | } |
572 | ||
2847cf95 JK |
573 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
574 | unsigned int order, int migratetype) | |
c0a32fc5 | 575 | { |
e30825f1 JK |
576 | struct page_ext *page_ext; |
577 | ||
578 | if (!debug_guardpage_enabled()) | |
579 | return; | |
580 | ||
581 | page_ext = lookup_page_ext(page); | |
582 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); | |
583 | ||
2847cf95 JK |
584 | set_page_private(page, 0); |
585 | if (!is_migrate_isolate(migratetype)) | |
586 | __mod_zone_freepage_state(zone, (1 << order), migratetype); | |
c0a32fc5 SG |
587 | } |
588 | #else | |
e30825f1 | 589 | struct page_ext_operations debug_guardpage_ops = { NULL, }; |
2847cf95 JK |
590 | static inline void set_page_guard(struct zone *zone, struct page *page, |
591 | unsigned int order, int migratetype) {} | |
592 | static inline void clear_page_guard(struct zone *zone, struct page *page, | |
593 | unsigned int order, int migratetype) {} | |
c0a32fc5 SG |
594 | #endif |
595 | ||
7aeb09f9 | 596 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 597 | { |
4c21e2f2 | 598 | set_page_private(page, order); |
676165a8 | 599 | __SetPageBuddy(page); |
1da177e4 LT |
600 | } |
601 | ||
602 | static inline void rmv_page_order(struct page *page) | |
603 | { | |
676165a8 | 604 | __ClearPageBuddy(page); |
4c21e2f2 | 605 | set_page_private(page, 0); |
1da177e4 LT |
606 | } |
607 | ||
1da177e4 LT |
608 | /* |
609 | * This function checks whether a page is free && is the buddy | |
610 | * we can do coalesce a page and its buddy if | |
13e7444b | 611 | * (a) the buddy is not in a hole && |
676165a8 | 612 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
613 | * (c) a page and its buddy have the same order && |
614 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 615 | * |
cf6fe945 WSH |
616 | * For recording whether a page is in the buddy system, we set ->_mapcount |
617 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
618 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
619 | * serialized by zone->lock. | |
1da177e4 | 620 | * |
676165a8 | 621 | * For recording page's order, we use page_private(page). |
1da177e4 | 622 | */ |
cb2b95e1 | 623 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 624 | unsigned int order) |
1da177e4 | 625 | { |
14e07298 | 626 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 627 | return 0; |
13e7444b | 628 | |
c0a32fc5 | 629 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
630 | if (page_zone_id(page) != page_zone_id(buddy)) |
631 | return 0; | |
632 | ||
4c5018ce WY |
633 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
634 | ||
c0a32fc5 SG |
635 | return 1; |
636 | } | |
637 | ||
cb2b95e1 | 638 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
d34c5fa0 MG |
639 | /* |
640 | * zone check is done late to avoid uselessly | |
641 | * calculating zone/node ids for pages that could | |
642 | * never merge. | |
643 | */ | |
644 | if (page_zone_id(page) != page_zone_id(buddy)) | |
645 | return 0; | |
646 | ||
4c5018ce WY |
647 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
648 | ||
6aa3001b | 649 | return 1; |
676165a8 | 650 | } |
6aa3001b | 651 | return 0; |
1da177e4 LT |
652 | } |
653 | ||
654 | /* | |
655 | * Freeing function for a buddy system allocator. | |
656 | * | |
657 | * The concept of a buddy system is to maintain direct-mapped table | |
658 | * (containing bit values) for memory blocks of various "orders". | |
659 | * The bottom level table contains the map for the smallest allocatable | |
660 | * units of memory (here, pages), and each level above it describes | |
661 | * pairs of units from the levels below, hence, "buddies". | |
662 | * At a high level, all that happens here is marking the table entry | |
663 | * at the bottom level available, and propagating the changes upward | |
664 | * as necessary, plus some accounting needed to play nicely with other | |
665 | * parts of the VM system. | |
666 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
667 | * free pages of length of (1 << order) and marked with _mapcount |
668 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
669 | * field. | |
1da177e4 | 670 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
671 | * other. That is, if we allocate a small block, and both were |
672 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 673 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 674 | * triggers coalescing into a block of larger size. |
1da177e4 | 675 | * |
6d49e352 | 676 | * -- nyc |
1da177e4 LT |
677 | */ |
678 | ||
48db57f8 | 679 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 680 | unsigned long pfn, |
ed0ae21d MG |
681 | struct zone *zone, unsigned int order, |
682 | int migratetype) | |
1da177e4 LT |
683 | { |
684 | unsigned long page_idx; | |
6dda9d55 | 685 | unsigned long combined_idx; |
43506fad | 686 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 687 | struct page *buddy; |
d00181b9 | 688 | unsigned int max_order = MAX_ORDER; |
1da177e4 | 689 | |
d29bb978 | 690 | VM_BUG_ON(!zone_is_initialized(zone)); |
6e9f0d58 | 691 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
1da177e4 | 692 | |
ed0ae21d | 693 | VM_BUG_ON(migratetype == -1); |
3c605096 JK |
694 | if (is_migrate_isolate(migratetype)) { |
695 | /* | |
696 | * We restrict max order of merging to prevent merge | |
697 | * between freepages on isolate pageblock and normal | |
698 | * pageblock. Without this, pageblock isolation | |
699 | * could cause incorrect freepage accounting. | |
700 | */ | |
d00181b9 | 701 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); |
3c605096 | 702 | } else { |
8f82b55d | 703 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
3c605096 | 704 | } |
ed0ae21d | 705 | |
3c605096 | 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 | |
3c605096 | 711 | while (order < max_order - 1) { |
43506fad KC |
712 | buddy_idx = __find_buddy_index(page_idx, order); |
713 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 714 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 715 | break; |
c0a32fc5 SG |
716 | /* |
717 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
718 | * merge with it and move up one order. | |
719 | */ | |
720 | if (page_is_guard(buddy)) { | |
2847cf95 | 721 | clear_page_guard(zone, buddy, order, migratetype); |
c0a32fc5 SG |
722 | } else { |
723 | list_del(&buddy->lru); | |
724 | zone->free_area[order].nr_free--; | |
725 | rmv_page_order(buddy); | |
726 | } | |
43506fad | 727 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
728 | page = page + (combined_idx - page_idx); |
729 | page_idx = combined_idx; | |
730 | order++; | |
731 | } | |
732 | set_page_order(page, order); | |
6dda9d55 CZ |
733 | |
734 | /* | |
735 | * If this is not the largest possible page, check if the buddy | |
736 | * of the next-highest order is free. If it is, it's possible | |
737 | * that pages are being freed that will coalesce soon. In case, | |
738 | * that is happening, add the free page to the tail of the list | |
739 | * so it's less likely to be used soon and more likely to be merged | |
740 | * as a higher order page | |
741 | */ | |
b7f50cfa | 742 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 743 | struct page *higher_page, *higher_buddy; |
43506fad KC |
744 | combined_idx = buddy_idx & page_idx; |
745 | higher_page = page + (combined_idx - page_idx); | |
746 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 747 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
748 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
749 | list_add_tail(&page->lru, | |
750 | &zone->free_area[order].free_list[migratetype]); | |
751 | goto out; | |
752 | } | |
753 | } | |
754 | ||
755 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
756 | out: | |
1da177e4 LT |
757 | zone->free_area[order].nr_free++; |
758 | } | |
759 | ||
224abf92 | 760 | static inline int free_pages_check(struct page *page) |
1da177e4 | 761 | { |
d230dec1 | 762 | const char *bad_reason = NULL; |
f0b791a3 DH |
763 | unsigned long bad_flags = 0; |
764 | ||
53f9263b | 765 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
766 | bad_reason = "nonzero mapcount"; |
767 | if (unlikely(page->mapping != NULL)) | |
768 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 769 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 DH |
770 | bad_reason = "nonzero _count"; |
771 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { | |
772 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
773 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
774 | } | |
9edad6ea JW |
775 | #ifdef CONFIG_MEMCG |
776 | if (unlikely(page->mem_cgroup)) | |
777 | bad_reason = "page still charged to cgroup"; | |
778 | #endif | |
f0b791a3 DH |
779 | if (unlikely(bad_reason)) { |
780 | bad_page(page, bad_reason, bad_flags); | |
79f4b7bf | 781 | return 1; |
8cc3b392 | 782 | } |
90572890 | 783 | page_cpupid_reset_last(page); |
79f4b7bf HD |
784 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
785 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
786 | return 0; | |
1da177e4 LT |
787 | } |
788 | ||
789 | /* | |
5f8dcc21 | 790 | * Frees a number of pages from the PCP lists |
1da177e4 | 791 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 792 | * count is the number of pages to free. |
1da177e4 LT |
793 | * |
794 | * If the zone was previously in an "all pages pinned" state then look to | |
795 | * see if this freeing clears that state. | |
796 | * | |
797 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
798 | * pinned" detection logic. | |
799 | */ | |
5f8dcc21 MG |
800 | static void free_pcppages_bulk(struct zone *zone, int count, |
801 | struct per_cpu_pages *pcp) | |
1da177e4 | 802 | { |
5f8dcc21 | 803 | int migratetype = 0; |
a6f9edd6 | 804 | int batch_free = 0; |
72853e29 | 805 | int to_free = count; |
0d5d823a | 806 | unsigned long nr_scanned; |
5f8dcc21 | 807 | |
c54ad30c | 808 | spin_lock(&zone->lock); |
0d5d823a MG |
809 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
810 | if (nr_scanned) | |
811 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 812 | |
72853e29 | 813 | while (to_free) { |
48db57f8 | 814 | struct page *page; |
5f8dcc21 MG |
815 | struct list_head *list; |
816 | ||
817 | /* | |
a6f9edd6 MG |
818 | * Remove pages from lists in a round-robin fashion. A |
819 | * batch_free count is maintained that is incremented when an | |
820 | * empty list is encountered. This is so more pages are freed | |
821 | * off fuller lists instead of spinning excessively around empty | |
822 | * lists | |
5f8dcc21 MG |
823 | */ |
824 | do { | |
a6f9edd6 | 825 | batch_free++; |
5f8dcc21 MG |
826 | if (++migratetype == MIGRATE_PCPTYPES) |
827 | migratetype = 0; | |
828 | list = &pcp->lists[migratetype]; | |
829 | } while (list_empty(list)); | |
48db57f8 | 830 | |
1d16871d NK |
831 | /* This is the only non-empty list. Free them all. */ |
832 | if (batch_free == MIGRATE_PCPTYPES) | |
833 | batch_free = to_free; | |
834 | ||
a6f9edd6 | 835 | do { |
770c8aaa BZ |
836 | int mt; /* migratetype of the to-be-freed page */ |
837 | ||
a16601c5 | 838 | page = list_last_entry(list, struct page, lru); |
a6f9edd6 MG |
839 | /* must delete as __free_one_page list manipulates */ |
840 | list_del(&page->lru); | |
aa016d14 | 841 | |
bb14c2c7 | 842 | mt = get_pcppage_migratetype(page); |
aa016d14 VB |
843 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
844 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); | |
845 | /* Pageblock could have been isolated meanwhile */ | |
8f82b55d | 846 | if (unlikely(has_isolate_pageblock(zone))) |
51bb1a40 | 847 | mt = get_pageblock_migratetype(page); |
51bb1a40 | 848 | |
dc4b0caf | 849 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 850 | trace_mm_page_pcpu_drain(page, 0, mt); |
72853e29 | 851 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 852 | } |
c54ad30c | 853 | spin_unlock(&zone->lock); |
1da177e4 LT |
854 | } |
855 | ||
dc4b0caf MG |
856 | static void free_one_page(struct zone *zone, |
857 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 858 | unsigned int order, |
ed0ae21d | 859 | int migratetype) |
1da177e4 | 860 | { |
0d5d823a | 861 | unsigned long nr_scanned; |
006d22d9 | 862 | spin_lock(&zone->lock); |
0d5d823a MG |
863 | nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED); |
864 | if (nr_scanned) | |
865 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned); | |
f2260e6b | 866 | |
ad53f92e JK |
867 | if (unlikely(has_isolate_pageblock(zone) || |
868 | is_migrate_isolate(migratetype))) { | |
869 | migratetype = get_pfnblock_migratetype(page, pfn); | |
ad53f92e | 870 | } |
dc4b0caf | 871 | __free_one_page(page, pfn, zone, order, migratetype); |
006d22d9 | 872 | spin_unlock(&zone->lock); |
48db57f8 NP |
873 | } |
874 | ||
81422f29 KS |
875 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
876 | { | |
1d798ca3 KS |
877 | int ret = 1; |
878 | ||
879 | /* | |
880 | * We rely page->lru.next never has bit 0 set, unless the page | |
881 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. | |
882 | */ | |
883 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); | |
884 | ||
885 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { | |
886 | ret = 0; | |
887 | goto out; | |
888 | } | |
9a982250 KS |
889 | switch (page - head_page) { |
890 | case 1: | |
891 | /* the first tail page: ->mapping is compound_mapcount() */ | |
53f9263b KS |
892 | if (unlikely(compound_mapcount(page))) { |
893 | bad_page(page, "nonzero compound_mapcount", 0); | |
894 | goto out; | |
895 | } | |
9a982250 KS |
896 | break; |
897 | case 2: | |
898 | /* | |
899 | * the second tail page: ->mapping is | |
900 | * page_deferred_list().next -- ignore value. | |
901 | */ | |
902 | break; | |
903 | default: | |
904 | if (page->mapping != TAIL_MAPPING) { | |
905 | bad_page(page, "corrupted mapping in tail page", 0); | |
906 | goto out; | |
907 | } | |
908 | break; | |
1c290f64 | 909 | } |
81422f29 KS |
910 | if (unlikely(!PageTail(page))) { |
911 | bad_page(page, "PageTail not set", 0); | |
1d798ca3 | 912 | goto out; |
81422f29 | 913 | } |
1d798ca3 KS |
914 | if (unlikely(compound_head(page) != head_page)) { |
915 | bad_page(page, "compound_head not consistent", 0); | |
916 | goto out; | |
81422f29 | 917 | } |
1d798ca3 KS |
918 | ret = 0; |
919 | out: | |
1c290f64 | 920 | page->mapping = NULL; |
1d798ca3 KS |
921 | clear_compound_head(page); |
922 | return ret; | |
81422f29 KS |
923 | } |
924 | ||
1e8ce83c RH |
925 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
926 | unsigned long zone, int nid) | |
927 | { | |
1e8ce83c | 928 | set_page_links(page, zone, nid, pfn); |
1e8ce83c RH |
929 | init_page_count(page); |
930 | page_mapcount_reset(page); | |
931 | page_cpupid_reset_last(page); | |
1e8ce83c | 932 | |
1e8ce83c RH |
933 | INIT_LIST_HEAD(&page->lru); |
934 | #ifdef WANT_PAGE_VIRTUAL | |
935 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
936 | if (!is_highmem_idx(zone)) | |
937 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
938 | #endif | |
939 | } | |
940 | ||
941 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, | |
942 | int nid) | |
943 | { | |
944 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); | |
945 | } | |
946 | ||
7e18adb4 MG |
947 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
948 | static void init_reserved_page(unsigned long pfn) | |
949 | { | |
950 | pg_data_t *pgdat; | |
951 | int nid, zid; | |
952 | ||
953 | if (!early_page_uninitialised(pfn)) | |
954 | return; | |
955 | ||
956 | nid = early_pfn_to_nid(pfn); | |
957 | pgdat = NODE_DATA(nid); | |
958 | ||
959 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
960 | struct zone *zone = &pgdat->node_zones[zid]; | |
961 | ||
962 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) | |
963 | break; | |
964 | } | |
965 | __init_single_pfn(pfn, zid, nid); | |
966 | } | |
967 | #else | |
968 | static inline void init_reserved_page(unsigned long pfn) | |
969 | { | |
970 | } | |
971 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ | |
972 | ||
92923ca3 NZ |
973 | /* |
974 | * Initialised pages do not have PageReserved set. This function is | |
975 | * called for each range allocated by the bootmem allocator and | |
976 | * marks the pages PageReserved. The remaining valid pages are later | |
977 | * sent to the buddy page allocator. | |
978 | */ | |
7e18adb4 | 979 | void __meminit reserve_bootmem_region(unsigned long start, unsigned long end) |
92923ca3 NZ |
980 | { |
981 | unsigned long start_pfn = PFN_DOWN(start); | |
982 | unsigned long end_pfn = PFN_UP(end); | |
983 | ||
7e18adb4 MG |
984 | for (; start_pfn < end_pfn; start_pfn++) { |
985 | if (pfn_valid(start_pfn)) { | |
986 | struct page *page = pfn_to_page(start_pfn); | |
987 | ||
988 | init_reserved_page(start_pfn); | |
1d798ca3 KS |
989 | |
990 | /* Avoid false-positive PageTail() */ | |
991 | INIT_LIST_HEAD(&page->lru); | |
992 | ||
7e18adb4 MG |
993 | SetPageReserved(page); |
994 | } | |
995 | } | |
92923ca3 NZ |
996 | } |
997 | ||
ec95f53a | 998 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 999 | { |
81422f29 KS |
1000 | bool compound = PageCompound(page); |
1001 | int i, bad = 0; | |
1da177e4 | 1002 | |
ab1f306f | 1003 | VM_BUG_ON_PAGE(PageTail(page), page); |
81422f29 | 1004 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); |
ab1f306f | 1005 | |
b413d48a | 1006 | trace_mm_page_free(page, order); |
b1eeab67 | 1007 | kmemcheck_free_shadow(page, order); |
b8c73fc2 | 1008 | kasan_free_pages(page, order); |
b1eeab67 | 1009 | |
8dd60a3a AA |
1010 | if (PageAnon(page)) |
1011 | page->mapping = NULL; | |
81422f29 KS |
1012 | bad += free_pages_check(page); |
1013 | for (i = 1; i < (1 << order); i++) { | |
1014 | if (compound) | |
1015 | bad += free_tail_pages_check(page, page + i); | |
8dd60a3a | 1016 | bad += free_pages_check(page + i); |
81422f29 | 1017 | } |
8cc3b392 | 1018 | if (bad) |
ec95f53a | 1019 | return false; |
689bcebf | 1020 | |
48c96a36 JK |
1021 | reset_page_owner(page, order); |
1022 | ||
3ac7fe5a | 1023 | if (!PageHighMem(page)) { |
b8af2941 PK |
1024 | debug_check_no_locks_freed(page_address(page), |
1025 | PAGE_SIZE << order); | |
3ac7fe5a TG |
1026 | debug_check_no_obj_freed(page_address(page), |
1027 | PAGE_SIZE << order); | |
1028 | } | |
dafb1367 | 1029 | arch_free_page(page, order); |
8823b1db | 1030 | kernel_poison_pages(page, 1 << order, 0); |
48db57f8 | 1031 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 1032 | |
ec95f53a KM |
1033 | return true; |
1034 | } | |
1035 | ||
1036 | static void __free_pages_ok(struct page *page, unsigned int order) | |
1037 | { | |
1038 | unsigned long flags; | |
95e34412 | 1039 | int migratetype; |
dc4b0caf | 1040 | unsigned long pfn = page_to_pfn(page); |
ec95f53a KM |
1041 | |
1042 | if (!free_pages_prepare(page, order)) | |
1043 | return; | |
1044 | ||
cfc47a28 | 1045 | migratetype = get_pfnblock_migratetype(page, pfn); |
c54ad30c | 1046 | local_irq_save(flags); |
f8891e5e | 1047 | __count_vm_events(PGFREE, 1 << order); |
dc4b0caf | 1048 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
c54ad30c | 1049 | local_irq_restore(flags); |
1da177e4 LT |
1050 | } |
1051 | ||
0e1cc95b | 1052 | static void __init __free_pages_boot_core(struct page *page, |
3a80a7fa | 1053 | unsigned long pfn, unsigned int order) |
a226f6c8 | 1054 | { |
c3993076 | 1055 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 1056 | struct page *p = page; |
c3993076 | 1057 | unsigned int loop; |
a226f6c8 | 1058 | |
e2d0bd2b YL |
1059 | prefetchw(p); |
1060 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
1061 | prefetchw(p + 1); | |
c3993076 JW |
1062 | __ClearPageReserved(p); |
1063 | set_page_count(p, 0); | |
a226f6c8 | 1064 | } |
e2d0bd2b YL |
1065 | __ClearPageReserved(p); |
1066 | set_page_count(p, 0); | |
c3993076 | 1067 | |
e2d0bd2b | 1068 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
1069 | set_page_refcounted(page); |
1070 | __free_pages(page, order); | |
a226f6c8 DH |
1071 | } |
1072 | ||
75a592a4 MG |
1073 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1074 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) | |
7ace9917 | 1075 | |
75a592a4 MG |
1076 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1077 | ||
1078 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
1079 | { | |
7ace9917 | 1080 | static DEFINE_SPINLOCK(early_pfn_lock); |
75a592a4 MG |
1081 | int nid; |
1082 | ||
7ace9917 | 1083 | spin_lock(&early_pfn_lock); |
75a592a4 | 1084 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
7ace9917 MG |
1085 | if (nid < 0) |
1086 | nid = 0; | |
1087 | spin_unlock(&early_pfn_lock); | |
1088 | ||
1089 | return nid; | |
75a592a4 MG |
1090 | } |
1091 | #endif | |
1092 | ||
1093 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
1094 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1095 | struct mminit_pfnnid_cache *state) | |
1096 | { | |
1097 | int nid; | |
1098 | ||
1099 | nid = __early_pfn_to_nid(pfn, state); | |
1100 | if (nid >= 0 && nid != node) | |
1101 | return false; | |
1102 | return true; | |
1103 | } | |
1104 | ||
1105 | /* Only safe to use early in boot when initialisation is single-threaded */ | |
1106 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1107 | { | |
1108 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); | |
1109 | } | |
1110 | ||
1111 | #else | |
1112 | ||
1113 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
1114 | { | |
1115 | return true; | |
1116 | } | |
1117 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, | |
1118 | struct mminit_pfnnid_cache *state) | |
1119 | { | |
1120 | return true; | |
1121 | } | |
1122 | #endif | |
1123 | ||
1124 | ||
0e1cc95b | 1125 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
3a80a7fa MG |
1126 | unsigned int order) |
1127 | { | |
1128 | if (early_page_uninitialised(pfn)) | |
1129 | return; | |
1130 | return __free_pages_boot_core(page, pfn, order); | |
1131 | } | |
1132 | ||
7cf91a98 JK |
1133 | /* |
1134 | * Check that the whole (or subset of) a pageblock given by the interval of | |
1135 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
1136 | * with the migration of free compaction scanner. The scanners then need to | |
1137 | * use only pfn_valid_within() check for arches that allow holes within | |
1138 | * pageblocks. | |
1139 | * | |
1140 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
1141 | * | |
1142 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
1143 | * i.e. it's possible that all pages within a zones range of pages do not | |
1144 | * belong to a single zone. We assume that a border between node0 and node1 | |
1145 | * can occur within a single pageblock, but not a node0 node1 node0 | |
1146 | * interleaving within a single pageblock. It is therefore sufficient to check | |
1147 | * the first and last page of a pageblock and avoid checking each individual | |
1148 | * page in a pageblock. | |
1149 | */ | |
1150 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, | |
1151 | unsigned long end_pfn, struct zone *zone) | |
1152 | { | |
1153 | struct page *start_page; | |
1154 | struct page *end_page; | |
1155 | ||
1156 | /* end_pfn is one past the range we are checking */ | |
1157 | end_pfn--; | |
1158 | ||
1159 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
1160 | return NULL; | |
1161 | ||
1162 | start_page = pfn_to_page(start_pfn); | |
1163 | ||
1164 | if (page_zone(start_page) != zone) | |
1165 | return NULL; | |
1166 | ||
1167 | end_page = pfn_to_page(end_pfn); | |
1168 | ||
1169 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
1170 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
1171 | return NULL; | |
1172 | ||
1173 | return start_page; | |
1174 | } | |
1175 | ||
1176 | void set_zone_contiguous(struct zone *zone) | |
1177 | { | |
1178 | unsigned long block_start_pfn = zone->zone_start_pfn; | |
1179 | unsigned long block_end_pfn; | |
1180 | ||
1181 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); | |
1182 | for (; block_start_pfn < zone_end_pfn(zone); | |
1183 | block_start_pfn = block_end_pfn, | |
1184 | block_end_pfn += pageblock_nr_pages) { | |
1185 | ||
1186 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); | |
1187 | ||
1188 | if (!__pageblock_pfn_to_page(block_start_pfn, | |
1189 | block_end_pfn, zone)) | |
1190 | return; | |
1191 | } | |
1192 | ||
1193 | /* We confirm that there is no hole */ | |
1194 | zone->contiguous = true; | |
1195 | } | |
1196 | ||
1197 | void clear_zone_contiguous(struct zone *zone) | |
1198 | { | |
1199 | zone->contiguous = false; | |
1200 | } | |
1201 | ||
7e18adb4 | 1202 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
0e1cc95b | 1203 | static void __init deferred_free_range(struct page *page, |
a4de83dd MG |
1204 | unsigned long pfn, int nr_pages) |
1205 | { | |
1206 | int i; | |
1207 | ||
1208 | if (!page) | |
1209 | return; | |
1210 | ||
1211 | /* Free a large naturally-aligned chunk if possible */ | |
1212 | if (nr_pages == MAX_ORDER_NR_PAGES && | |
1213 | (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) { | |
ac5d2539 | 1214 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
a4de83dd MG |
1215 | __free_pages_boot_core(page, pfn, MAX_ORDER-1); |
1216 | return; | |
1217 | } | |
1218 | ||
1219 | for (i = 0; i < nr_pages; i++, page++, pfn++) | |
1220 | __free_pages_boot_core(page, pfn, 0); | |
1221 | } | |
1222 | ||
d3cd131d NS |
1223 | /* Completion tracking for deferred_init_memmap() threads */ |
1224 | static atomic_t pgdat_init_n_undone __initdata; | |
1225 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); | |
1226 | ||
1227 | static inline void __init pgdat_init_report_one_done(void) | |
1228 | { | |
1229 | if (atomic_dec_and_test(&pgdat_init_n_undone)) | |
1230 | complete(&pgdat_init_all_done_comp); | |
1231 | } | |
0e1cc95b | 1232 | |
7e18adb4 | 1233 | /* Initialise remaining memory on a node */ |
0e1cc95b | 1234 | static int __init deferred_init_memmap(void *data) |
7e18adb4 | 1235 | { |
0e1cc95b MG |
1236 | pg_data_t *pgdat = data; |
1237 | int nid = pgdat->node_id; | |
7e18adb4 MG |
1238 | struct mminit_pfnnid_cache nid_init_state = { }; |
1239 | unsigned long start = jiffies; | |
1240 | unsigned long nr_pages = 0; | |
1241 | unsigned long walk_start, walk_end; | |
1242 | int i, zid; | |
1243 | struct zone *zone; | |
7e18adb4 | 1244 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
0e1cc95b | 1245 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
7e18adb4 | 1246 | |
0e1cc95b | 1247 | if (first_init_pfn == ULONG_MAX) { |
d3cd131d | 1248 | pgdat_init_report_one_done(); |
0e1cc95b MG |
1249 | return 0; |
1250 | } | |
1251 | ||
1252 | /* Bind memory initialisation thread to a local node if possible */ | |
1253 | if (!cpumask_empty(cpumask)) | |
1254 | set_cpus_allowed_ptr(current, cpumask); | |
7e18adb4 MG |
1255 | |
1256 | /* Sanity check boundaries */ | |
1257 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); | |
1258 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); | |
1259 | pgdat->first_deferred_pfn = ULONG_MAX; | |
1260 | ||
1261 | /* Only the highest zone is deferred so find it */ | |
1262 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1263 | zone = pgdat->node_zones + zid; | |
1264 | if (first_init_pfn < zone_end_pfn(zone)) | |
1265 | break; | |
1266 | } | |
1267 | ||
1268 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { | |
1269 | unsigned long pfn, end_pfn; | |
54608c3f | 1270 | struct page *page = NULL; |
a4de83dd MG |
1271 | struct page *free_base_page = NULL; |
1272 | unsigned long free_base_pfn = 0; | |
1273 | int nr_to_free = 0; | |
7e18adb4 MG |
1274 | |
1275 | end_pfn = min(walk_end, zone_end_pfn(zone)); | |
1276 | pfn = first_init_pfn; | |
1277 | if (pfn < walk_start) | |
1278 | pfn = walk_start; | |
1279 | if (pfn < zone->zone_start_pfn) | |
1280 | pfn = zone->zone_start_pfn; | |
1281 | ||
1282 | for (; pfn < end_pfn; pfn++) { | |
54608c3f | 1283 | if (!pfn_valid_within(pfn)) |
a4de83dd | 1284 | goto free_range; |
7e18adb4 | 1285 | |
54608c3f MG |
1286 | /* |
1287 | * Ensure pfn_valid is checked every | |
1288 | * MAX_ORDER_NR_PAGES for memory holes | |
1289 | */ | |
1290 | if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
1291 | if (!pfn_valid(pfn)) { | |
1292 | page = NULL; | |
a4de83dd | 1293 | goto free_range; |
54608c3f MG |
1294 | } |
1295 | } | |
1296 | ||
1297 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { | |
1298 | page = NULL; | |
a4de83dd | 1299 | goto free_range; |
54608c3f MG |
1300 | } |
1301 | ||
1302 | /* Minimise pfn page lookups and scheduler checks */ | |
1303 | if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) { | |
1304 | page++; | |
1305 | } else { | |
a4de83dd MG |
1306 | nr_pages += nr_to_free; |
1307 | deferred_free_range(free_base_page, | |
1308 | free_base_pfn, nr_to_free); | |
1309 | free_base_page = NULL; | |
1310 | free_base_pfn = nr_to_free = 0; | |
1311 | ||
54608c3f MG |
1312 | page = pfn_to_page(pfn); |
1313 | cond_resched(); | |
1314 | } | |
7e18adb4 MG |
1315 | |
1316 | if (page->flags) { | |
1317 | VM_BUG_ON(page_zone(page) != zone); | |
a4de83dd | 1318 | goto free_range; |
7e18adb4 MG |
1319 | } |
1320 | ||
1321 | __init_single_page(page, pfn, zid, nid); | |
a4de83dd MG |
1322 | if (!free_base_page) { |
1323 | free_base_page = page; | |
1324 | free_base_pfn = pfn; | |
1325 | nr_to_free = 0; | |
1326 | } | |
1327 | nr_to_free++; | |
1328 | ||
1329 | /* Where possible, batch up pages for a single free */ | |
1330 | continue; | |
1331 | free_range: | |
1332 | /* Free the current block of pages to allocator */ | |
1333 | nr_pages += nr_to_free; | |
1334 | deferred_free_range(free_base_page, free_base_pfn, | |
1335 | nr_to_free); | |
1336 | free_base_page = NULL; | |
1337 | free_base_pfn = nr_to_free = 0; | |
7e18adb4 | 1338 | } |
a4de83dd | 1339 | |
7e18adb4 MG |
1340 | first_init_pfn = max(end_pfn, first_init_pfn); |
1341 | } | |
1342 | ||
1343 | /* Sanity check that the next zone really is unpopulated */ | |
1344 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); | |
1345 | ||
0e1cc95b | 1346 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
7e18adb4 | 1347 | jiffies_to_msecs(jiffies - start)); |
d3cd131d NS |
1348 | |
1349 | pgdat_init_report_one_done(); | |
0e1cc95b MG |
1350 | return 0; |
1351 | } | |
7cf91a98 | 1352 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
0e1cc95b MG |
1353 | |
1354 | void __init page_alloc_init_late(void) | |
1355 | { | |
7cf91a98 JK |
1356 | struct zone *zone; |
1357 | ||
1358 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT | |
0e1cc95b MG |
1359 | int nid; |
1360 | ||
d3cd131d NS |
1361 | /* There will be num_node_state(N_MEMORY) threads */ |
1362 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); | |
0e1cc95b | 1363 | for_each_node_state(nid, N_MEMORY) { |
0e1cc95b MG |
1364 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1365 | } | |
1366 | ||
1367 | /* Block until all are initialised */ | |
d3cd131d | 1368 | wait_for_completion(&pgdat_init_all_done_comp); |
4248b0da MG |
1369 | |
1370 | /* Reinit limits that are based on free pages after the kernel is up */ | |
1371 | files_maxfiles_init(); | |
7cf91a98 JK |
1372 | #endif |
1373 | ||
1374 | for_each_populated_zone(zone) | |
1375 | set_zone_contiguous(zone); | |
7e18adb4 | 1376 | } |
7e18adb4 | 1377 | |
47118af0 | 1378 | #ifdef CONFIG_CMA |
9cf510a5 | 1379 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
1380 | void __init init_cma_reserved_pageblock(struct page *page) |
1381 | { | |
1382 | unsigned i = pageblock_nr_pages; | |
1383 | struct page *p = page; | |
1384 | ||
1385 | do { | |
1386 | __ClearPageReserved(p); | |
1387 | set_page_count(p, 0); | |
1388 | } while (++p, --i); | |
1389 | ||
47118af0 | 1390 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
1391 | |
1392 | if (pageblock_order >= MAX_ORDER) { | |
1393 | i = pageblock_nr_pages; | |
1394 | p = page; | |
1395 | do { | |
1396 | set_page_refcounted(p); | |
1397 | __free_pages(p, MAX_ORDER - 1); | |
1398 | p += MAX_ORDER_NR_PAGES; | |
1399 | } while (i -= MAX_ORDER_NR_PAGES); | |
1400 | } else { | |
1401 | set_page_refcounted(page); | |
1402 | __free_pages(page, pageblock_order); | |
1403 | } | |
1404 | ||
3dcc0571 | 1405 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
1406 | } |
1407 | #endif | |
1da177e4 LT |
1408 | |
1409 | /* | |
1410 | * The order of subdivision here is critical for the IO subsystem. | |
1411 | * Please do not alter this order without good reasons and regression | |
1412 | * testing. Specifically, as large blocks of memory are subdivided, | |
1413 | * the order in which smaller blocks are delivered depends on the order | |
1414 | * they're subdivided in this function. This is the primary factor | |
1415 | * influencing the order in which pages are delivered to the IO | |
1416 | * subsystem according to empirical testing, and this is also justified | |
1417 | * by considering the behavior of a buddy system containing a single | |
1418 | * large block of memory acted on by a series of small allocations. | |
1419 | * This behavior is a critical factor in sglist merging's success. | |
1420 | * | |
6d49e352 | 1421 | * -- nyc |
1da177e4 | 1422 | */ |
085cc7d5 | 1423 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
1424 | int low, int high, struct free_area *area, |
1425 | int migratetype) | |
1da177e4 LT |
1426 | { |
1427 | unsigned long size = 1 << high; | |
1428 | ||
1429 | while (high > low) { | |
1430 | area--; | |
1431 | high--; | |
1432 | size >>= 1; | |
309381fe | 1433 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 | 1434 | |
2847cf95 | 1435 | if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) && |
e30825f1 | 1436 | debug_guardpage_enabled() && |
2847cf95 | 1437 | high < debug_guardpage_minorder()) { |
c0a32fc5 SG |
1438 | /* |
1439 | * Mark as guard pages (or page), that will allow to | |
1440 | * merge back to allocator when buddy will be freed. | |
1441 | * Corresponding page table entries will not be touched, | |
1442 | * pages will stay not present in virtual address space | |
1443 | */ | |
2847cf95 | 1444 | set_page_guard(zone, &page[size], high, migratetype); |
c0a32fc5 SG |
1445 | continue; |
1446 | } | |
b2a0ac88 | 1447 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
1448 | area->nr_free++; |
1449 | set_page_order(&page[size], high); | |
1450 | } | |
1da177e4 LT |
1451 | } |
1452 | ||
1da177e4 LT |
1453 | /* |
1454 | * This page is about to be returned from the page allocator | |
1455 | */ | |
2a7684a2 | 1456 | static inline int check_new_page(struct page *page) |
1da177e4 | 1457 | { |
d230dec1 | 1458 | const char *bad_reason = NULL; |
f0b791a3 DH |
1459 | unsigned long bad_flags = 0; |
1460 | ||
53f9263b | 1461 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
f0b791a3 DH |
1462 | bad_reason = "nonzero mapcount"; |
1463 | if (unlikely(page->mapping != NULL)) | |
1464 | bad_reason = "non-NULL mapping"; | |
fe896d18 | 1465 | if (unlikely(page_ref_count(page) != 0)) |
f0b791a3 | 1466 | bad_reason = "nonzero _count"; |
f4c18e6f NH |
1467 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1468 | bad_reason = "HWPoisoned (hardware-corrupted)"; | |
1469 | bad_flags = __PG_HWPOISON; | |
1470 | } | |
f0b791a3 DH |
1471 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1472 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
1473 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
1474 | } | |
9edad6ea JW |
1475 | #ifdef CONFIG_MEMCG |
1476 | if (unlikely(page->mem_cgroup)) | |
1477 | bad_reason = "page still charged to cgroup"; | |
1478 | #endif | |
f0b791a3 DH |
1479 | if (unlikely(bad_reason)) { |
1480 | bad_page(page, bad_reason, bad_flags); | |
689bcebf | 1481 | return 1; |
8cc3b392 | 1482 | } |
2a7684a2 WF |
1483 | return 0; |
1484 | } | |
1485 | ||
1414c7f4 LA |
1486 | static inline bool free_pages_prezeroed(bool poisoned) |
1487 | { | |
1488 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && | |
1489 | page_poisoning_enabled() && poisoned; | |
1490 | } | |
1491 | ||
75379191 VB |
1492 | static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
1493 | int alloc_flags) | |
2a7684a2 WF |
1494 | { |
1495 | int i; | |
1414c7f4 | 1496 | bool poisoned = true; |
2a7684a2 WF |
1497 | |
1498 | for (i = 0; i < (1 << order); i++) { | |
1499 | struct page *p = page + i; | |
1500 | if (unlikely(check_new_page(p))) | |
1501 | return 1; | |
1414c7f4 LA |
1502 | if (poisoned) |
1503 | poisoned &= page_is_poisoned(p); | |
2a7684a2 | 1504 | } |
689bcebf | 1505 | |
4c21e2f2 | 1506 | set_page_private(page, 0); |
7835e98b | 1507 | set_page_refcounted(page); |
cc102509 NP |
1508 | |
1509 | arch_alloc_page(page, order); | |
1da177e4 | 1510 | kernel_map_pages(page, 1 << order, 1); |
8823b1db | 1511 | kernel_poison_pages(page, 1 << order, 1); |
b8c73fc2 | 1512 | kasan_alloc_pages(page, order); |
17cf4406 | 1513 | |
1414c7f4 | 1514 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
f4d2897b AA |
1515 | for (i = 0; i < (1 << order); i++) |
1516 | clear_highpage(page + i); | |
17cf4406 NP |
1517 | |
1518 | if (order && (gfp_flags & __GFP_COMP)) | |
1519 | prep_compound_page(page, order); | |
1520 | ||
48c96a36 JK |
1521 | set_page_owner(page, order, gfp_flags); |
1522 | ||
75379191 | 1523 | /* |
2f064f34 | 1524 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
75379191 VB |
1525 | * allocate the page. The expectation is that the caller is taking |
1526 | * steps that will free more memory. The caller should avoid the page | |
1527 | * being used for !PFMEMALLOC purposes. | |
1528 | */ | |
2f064f34 MH |
1529 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1530 | set_page_pfmemalloc(page); | |
1531 | else | |
1532 | clear_page_pfmemalloc(page); | |
75379191 | 1533 | |
689bcebf | 1534 | return 0; |
1da177e4 LT |
1535 | } |
1536 | ||
56fd56b8 MG |
1537 | /* |
1538 | * Go through the free lists for the given migratetype and remove | |
1539 | * the smallest available page from the freelists | |
1540 | */ | |
728ec980 MG |
1541 | static inline |
1542 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
1543 | int migratetype) |
1544 | { | |
1545 | unsigned int current_order; | |
b8af2941 | 1546 | struct free_area *area; |
56fd56b8 MG |
1547 | struct page *page; |
1548 | ||
1549 | /* Find a page of the appropriate size in the preferred list */ | |
1550 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
1551 | area = &(zone->free_area[current_order]); | |
a16601c5 | 1552 | page = list_first_entry_or_null(&area->free_list[migratetype], |
56fd56b8 | 1553 | struct page, lru); |
a16601c5 GT |
1554 | if (!page) |
1555 | continue; | |
56fd56b8 MG |
1556 | list_del(&page->lru); |
1557 | rmv_page_order(page); | |
1558 | area->nr_free--; | |
56fd56b8 | 1559 | expand(zone, page, order, current_order, area, migratetype); |
bb14c2c7 | 1560 | set_pcppage_migratetype(page, migratetype); |
56fd56b8 MG |
1561 | return page; |
1562 | } | |
1563 | ||
1564 | return NULL; | |
1565 | } | |
1566 | ||
1567 | ||
b2a0ac88 MG |
1568 | /* |
1569 | * This array describes the order lists are fallen back to when | |
1570 | * the free lists for the desirable migrate type are depleted | |
1571 | */ | |
47118af0 | 1572 | static int fallbacks[MIGRATE_TYPES][4] = { |
974a786e MG |
1573 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1574 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, | |
1575 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, | |
47118af0 | 1576 | #ifdef CONFIG_CMA |
974a786e | 1577 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
47118af0 | 1578 | #endif |
194159fb | 1579 | #ifdef CONFIG_MEMORY_ISOLATION |
974a786e | 1580 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
194159fb | 1581 | #endif |
b2a0ac88 MG |
1582 | }; |
1583 | ||
dc67647b JK |
1584 | #ifdef CONFIG_CMA |
1585 | static struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1586 | unsigned int order) | |
1587 | { | |
1588 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); | |
1589 | } | |
1590 | #else | |
1591 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, | |
1592 | unsigned int order) { return NULL; } | |
1593 | #endif | |
1594 | ||
c361be55 MG |
1595 | /* |
1596 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 1597 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
1598 | * boundary. If alignment is required, use move_freepages_block() |
1599 | */ | |
435b405c | 1600 | int move_freepages(struct zone *zone, |
b69a7288 AB |
1601 | struct page *start_page, struct page *end_page, |
1602 | int migratetype) | |
c361be55 MG |
1603 | { |
1604 | struct page *page; | |
d00181b9 | 1605 | unsigned int order; |
d100313f | 1606 | int pages_moved = 0; |
c361be55 MG |
1607 | |
1608 | #ifndef CONFIG_HOLES_IN_ZONE | |
1609 | /* | |
1610 | * page_zone is not safe to call in this context when | |
1611 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1612 | * anyway as we check zone boundaries in move_freepages_block(). | |
1613 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1614 | * grouping pages by mobility |
c361be55 | 1615 | */ |
97ee4ba7 | 1616 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
c361be55 MG |
1617 | #endif |
1618 | ||
1619 | for (page = start_page; page <= end_page;) { | |
344c790e | 1620 | /* Make sure we are not inadvertently changing nodes */ |
309381fe | 1621 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
344c790e | 1622 | |
c361be55 MG |
1623 | if (!pfn_valid_within(page_to_pfn(page))) { |
1624 | page++; | |
1625 | continue; | |
1626 | } | |
1627 | ||
1628 | if (!PageBuddy(page)) { | |
1629 | page++; | |
1630 | continue; | |
1631 | } | |
1632 | ||
1633 | order = page_order(page); | |
84be48d8 KS |
1634 | list_move(&page->lru, |
1635 | &zone->free_area[order].free_list[migratetype]); | |
c361be55 | 1636 | page += 1 << order; |
d100313f | 1637 | pages_moved += 1 << order; |
c361be55 MG |
1638 | } |
1639 | ||
d100313f | 1640 | return pages_moved; |
c361be55 MG |
1641 | } |
1642 | ||
ee6f509c | 1643 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1644 | int migratetype) |
c361be55 MG |
1645 | { |
1646 | unsigned long start_pfn, end_pfn; | |
1647 | struct page *start_page, *end_page; | |
1648 | ||
1649 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1650 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1651 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1652 | end_page = start_page + pageblock_nr_pages - 1; |
1653 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1654 | |
1655 | /* Do not cross zone boundaries */ | |
108bcc96 | 1656 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1657 | start_page = page; |
108bcc96 | 1658 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1659 | return 0; |
1660 | ||
1661 | return move_freepages(zone, start_page, end_page, migratetype); | |
1662 | } | |
1663 | ||
2f66a68f MG |
1664 | static void change_pageblock_range(struct page *pageblock_page, |
1665 | int start_order, int migratetype) | |
1666 | { | |
1667 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1668 | ||
1669 | while (nr_pageblocks--) { | |
1670 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1671 | pageblock_page += pageblock_nr_pages; | |
1672 | } | |
1673 | } | |
1674 | ||
fef903ef | 1675 | /* |
9c0415eb VB |
1676 | * When we are falling back to another migratetype during allocation, try to |
1677 | * steal extra free pages from the same pageblocks to satisfy further | |
1678 | * allocations, instead of polluting multiple pageblocks. | |
1679 | * | |
1680 | * If we are stealing a relatively large buddy page, it is likely there will | |
1681 | * be more free pages in the pageblock, so try to steal them all. For | |
1682 | * reclaimable and unmovable allocations, we steal regardless of page size, | |
1683 | * as fragmentation caused by those allocations polluting movable pageblocks | |
1684 | * is worse than movable allocations stealing from unmovable and reclaimable | |
1685 | * pageblocks. | |
fef903ef | 1686 | */ |
4eb7dce6 JK |
1687 | static bool can_steal_fallback(unsigned int order, int start_mt) |
1688 | { | |
1689 | /* | |
1690 | * Leaving this order check is intended, although there is | |
1691 | * relaxed order check in next check. The reason is that | |
1692 | * we can actually steal whole pageblock if this condition met, | |
1693 | * but, below check doesn't guarantee it and that is just heuristic | |
1694 | * so could be changed anytime. | |
1695 | */ | |
1696 | if (order >= pageblock_order) | |
1697 | return true; | |
1698 | ||
1699 | if (order >= pageblock_order / 2 || | |
1700 | start_mt == MIGRATE_RECLAIMABLE || | |
1701 | start_mt == MIGRATE_UNMOVABLE || | |
1702 | page_group_by_mobility_disabled) | |
1703 | return true; | |
1704 | ||
1705 | return false; | |
1706 | } | |
1707 | ||
1708 | /* | |
1709 | * This function implements actual steal behaviour. If order is large enough, | |
1710 | * we can steal whole pageblock. If not, we first move freepages in this | |
1711 | * pageblock and check whether half of pages are moved or not. If half of | |
1712 | * pages are moved, we can change migratetype of pageblock and permanently | |
1713 | * use it's pages as requested migratetype in the future. | |
1714 | */ | |
1715 | static void steal_suitable_fallback(struct zone *zone, struct page *page, | |
1716 | int start_type) | |
fef903ef | 1717 | { |
d00181b9 | 1718 | unsigned int current_order = page_order(page); |
4eb7dce6 | 1719 | int pages; |
fef903ef | 1720 | |
fef903ef SB |
1721 | /* Take ownership for orders >= pageblock_order */ |
1722 | if (current_order >= pageblock_order) { | |
1723 | change_pageblock_range(page, current_order, start_type); | |
3a1086fb | 1724 | return; |
fef903ef SB |
1725 | } |
1726 | ||
4eb7dce6 | 1727 | pages = move_freepages_block(zone, page, start_type); |
fef903ef | 1728 | |
4eb7dce6 JK |
1729 | /* Claim the whole block if over half of it is free */ |
1730 | if (pages >= (1 << (pageblock_order-1)) || | |
1731 | page_group_by_mobility_disabled) | |
1732 | set_pageblock_migratetype(page, start_type); | |
1733 | } | |
1734 | ||
2149cdae JK |
1735 | /* |
1736 | * Check whether there is a suitable fallback freepage with requested order. | |
1737 | * If only_stealable is true, this function returns fallback_mt only if | |
1738 | * we can steal other freepages all together. This would help to reduce | |
1739 | * fragmentation due to mixed migratetype pages in one pageblock. | |
1740 | */ | |
1741 | int find_suitable_fallback(struct free_area *area, unsigned int order, | |
1742 | int migratetype, bool only_stealable, bool *can_steal) | |
4eb7dce6 JK |
1743 | { |
1744 | int i; | |
1745 | int fallback_mt; | |
1746 | ||
1747 | if (area->nr_free == 0) | |
1748 | return -1; | |
1749 | ||
1750 | *can_steal = false; | |
1751 | for (i = 0;; i++) { | |
1752 | fallback_mt = fallbacks[migratetype][i]; | |
974a786e | 1753 | if (fallback_mt == MIGRATE_TYPES) |
4eb7dce6 JK |
1754 | break; |
1755 | ||
1756 | if (list_empty(&area->free_list[fallback_mt])) | |
1757 | continue; | |
fef903ef | 1758 | |
4eb7dce6 JK |
1759 | if (can_steal_fallback(order, migratetype)) |
1760 | *can_steal = true; | |
1761 | ||
2149cdae JK |
1762 | if (!only_stealable) |
1763 | return fallback_mt; | |
1764 | ||
1765 | if (*can_steal) | |
1766 | return fallback_mt; | |
fef903ef | 1767 | } |
4eb7dce6 JK |
1768 | |
1769 | return -1; | |
fef903ef SB |
1770 | } |
1771 | ||
0aaa29a5 MG |
1772 | /* |
1773 | * Reserve a pageblock for exclusive use of high-order atomic allocations if | |
1774 | * there are no empty page blocks that contain a page with a suitable order | |
1775 | */ | |
1776 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, | |
1777 | unsigned int alloc_order) | |
1778 | { | |
1779 | int mt; | |
1780 | unsigned long max_managed, flags; | |
1781 | ||
1782 | /* | |
1783 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. | |
1784 | * Check is race-prone but harmless. | |
1785 | */ | |
1786 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; | |
1787 | if (zone->nr_reserved_highatomic >= max_managed) | |
1788 | return; | |
1789 | ||
1790 | spin_lock_irqsave(&zone->lock, flags); | |
1791 | ||
1792 | /* Recheck the nr_reserved_highatomic limit under the lock */ | |
1793 | if (zone->nr_reserved_highatomic >= max_managed) | |
1794 | goto out_unlock; | |
1795 | ||
1796 | /* Yoink! */ | |
1797 | mt = get_pageblock_migratetype(page); | |
1798 | if (mt != MIGRATE_HIGHATOMIC && | |
1799 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { | |
1800 | zone->nr_reserved_highatomic += pageblock_nr_pages; | |
1801 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); | |
1802 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); | |
1803 | } | |
1804 | ||
1805 | out_unlock: | |
1806 | spin_unlock_irqrestore(&zone->lock, flags); | |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * Used when an allocation is about to fail under memory pressure. This | |
1811 | * potentially hurts the reliability of high-order allocations when under | |
1812 | * intense memory pressure but failed atomic allocations should be easier | |
1813 | * to recover from than an OOM. | |
1814 | */ | |
1815 | static void unreserve_highatomic_pageblock(const struct alloc_context *ac) | |
1816 | { | |
1817 | struct zonelist *zonelist = ac->zonelist; | |
1818 | unsigned long flags; | |
1819 | struct zoneref *z; | |
1820 | struct zone *zone; | |
1821 | struct page *page; | |
1822 | int order; | |
1823 | ||
1824 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, | |
1825 | ac->nodemask) { | |
1826 | /* Preserve at least one pageblock */ | |
1827 | if (zone->nr_reserved_highatomic <= pageblock_nr_pages) | |
1828 | continue; | |
1829 | ||
1830 | spin_lock_irqsave(&zone->lock, flags); | |
1831 | for (order = 0; order < MAX_ORDER; order++) { | |
1832 | struct free_area *area = &(zone->free_area[order]); | |
1833 | ||
a16601c5 GT |
1834 | page = list_first_entry_or_null( |
1835 | &area->free_list[MIGRATE_HIGHATOMIC], | |
1836 | struct page, lru); | |
1837 | if (!page) | |
0aaa29a5 MG |
1838 | continue; |
1839 | ||
0aaa29a5 MG |
1840 | /* |
1841 | * It should never happen but changes to locking could | |
1842 | * inadvertently allow a per-cpu drain to add pages | |
1843 | * to MIGRATE_HIGHATOMIC while unreserving so be safe | |
1844 | * and watch for underflows. | |
1845 | */ | |
1846 | zone->nr_reserved_highatomic -= min(pageblock_nr_pages, | |
1847 | zone->nr_reserved_highatomic); | |
1848 | ||
1849 | /* | |
1850 | * Convert to ac->migratetype and avoid the normal | |
1851 | * pageblock stealing heuristics. Minimally, the caller | |
1852 | * is doing the work and needs the pages. More | |
1853 | * importantly, if the block was always converted to | |
1854 | * MIGRATE_UNMOVABLE or another type then the number | |
1855 | * of pageblocks that cannot be completely freed | |
1856 | * may increase. | |
1857 | */ | |
1858 | set_pageblock_migratetype(page, ac->migratetype); | |
1859 | move_freepages_block(zone, page, ac->migratetype); | |
1860 | spin_unlock_irqrestore(&zone->lock, flags); | |
1861 | return; | |
1862 | } | |
1863 | spin_unlock_irqrestore(&zone->lock, flags); | |
1864 | } | |
1865 | } | |
1866 | ||
b2a0ac88 | 1867 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 1868 | static inline struct page * |
7aeb09f9 | 1869 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 1870 | { |
b8af2941 | 1871 | struct free_area *area; |
7aeb09f9 | 1872 | unsigned int current_order; |
b2a0ac88 | 1873 | struct page *page; |
4eb7dce6 JK |
1874 | int fallback_mt; |
1875 | bool can_steal; | |
b2a0ac88 MG |
1876 | |
1877 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
1878 | for (current_order = MAX_ORDER-1; |
1879 | current_order >= order && current_order <= MAX_ORDER-1; | |
1880 | --current_order) { | |
4eb7dce6 JK |
1881 | area = &(zone->free_area[current_order]); |
1882 | fallback_mt = find_suitable_fallback(area, current_order, | |
2149cdae | 1883 | start_migratetype, false, &can_steal); |
4eb7dce6 JK |
1884 | if (fallback_mt == -1) |
1885 | continue; | |
b2a0ac88 | 1886 | |
a16601c5 | 1887 | page = list_first_entry(&area->free_list[fallback_mt], |
4eb7dce6 JK |
1888 | struct page, lru); |
1889 | if (can_steal) | |
1890 | steal_suitable_fallback(zone, page, start_migratetype); | |
b2a0ac88 | 1891 | |
4eb7dce6 JK |
1892 | /* Remove the page from the freelists */ |
1893 | area->nr_free--; | |
1894 | list_del(&page->lru); | |
1895 | rmv_page_order(page); | |
3a1086fb | 1896 | |
4eb7dce6 JK |
1897 | expand(zone, page, order, current_order, area, |
1898 | start_migratetype); | |
1899 | /* | |
bb14c2c7 | 1900 | * The pcppage_migratetype may differ from pageblock's |
4eb7dce6 | 1901 | * migratetype depending on the decisions in |
bb14c2c7 VB |
1902 | * find_suitable_fallback(). This is OK as long as it does not |
1903 | * differ for MIGRATE_CMA pageblocks. Those can be used as | |
1904 | * fallback only via special __rmqueue_cma_fallback() function | |
4eb7dce6 | 1905 | */ |
bb14c2c7 | 1906 | set_pcppage_migratetype(page, start_migratetype); |
e0fff1bd | 1907 | |
4eb7dce6 JK |
1908 | trace_mm_page_alloc_extfrag(page, order, current_order, |
1909 | start_migratetype, fallback_mt); | |
e0fff1bd | 1910 | |
4eb7dce6 | 1911 | return page; |
b2a0ac88 MG |
1912 | } |
1913 | ||
728ec980 | 1914 | return NULL; |
b2a0ac88 MG |
1915 | } |
1916 | ||
56fd56b8 | 1917 | /* |
1da177e4 LT |
1918 | * Do the hard work of removing an element from the buddy allocator. |
1919 | * Call me with the zone->lock already held. | |
1920 | */ | |
b2a0ac88 | 1921 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
6ac0206b | 1922 | int migratetype) |
1da177e4 | 1923 | { |
1da177e4 LT |
1924 | struct page *page; |
1925 | ||
56fd56b8 | 1926 | page = __rmqueue_smallest(zone, order, migratetype); |
974a786e | 1927 | if (unlikely(!page)) { |
dc67647b JK |
1928 | if (migratetype == MIGRATE_MOVABLE) |
1929 | page = __rmqueue_cma_fallback(zone, order); | |
1930 | ||
1931 | if (!page) | |
1932 | page = __rmqueue_fallback(zone, order, migratetype); | |
728ec980 MG |
1933 | } |
1934 | ||
0d3d062a | 1935 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1936 | return page; |
1da177e4 LT |
1937 | } |
1938 | ||
5f63b720 | 1939 | /* |
1da177e4 LT |
1940 | * Obtain a specified number of elements from the buddy allocator, all under |
1941 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1942 | * Returns the number of new pages which were placed at *list. | |
1943 | */ | |
5f63b720 | 1944 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1945 | unsigned long count, struct list_head *list, |
b745bc85 | 1946 | int migratetype, bool cold) |
1da177e4 | 1947 | { |
5bcc9f86 | 1948 | int i; |
5f63b720 | 1949 | |
c54ad30c | 1950 | spin_lock(&zone->lock); |
1da177e4 | 1951 | for (i = 0; i < count; ++i) { |
6ac0206b | 1952 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1953 | if (unlikely(page == NULL)) |
1da177e4 | 1954 | break; |
81eabcbe MG |
1955 | |
1956 | /* | |
1957 | * Split buddy pages returned by expand() are received here | |
1958 | * in physical page order. The page is added to the callers and | |
1959 | * list and the list head then moves forward. From the callers | |
1960 | * perspective, the linked list is ordered by page number in | |
1961 | * some conditions. This is useful for IO devices that can | |
1962 | * merge IO requests if the physical pages are ordered | |
1963 | * properly. | |
1964 | */ | |
b745bc85 | 1965 | if (likely(!cold)) |
e084b2d9 MG |
1966 | list_add(&page->lru, list); |
1967 | else | |
1968 | list_add_tail(&page->lru, list); | |
81eabcbe | 1969 | list = &page->lru; |
bb14c2c7 | 1970 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
d1ce749a BZ |
1971 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
1972 | -(1 << order)); | |
1da177e4 | 1973 | } |
f2260e6b | 1974 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1975 | spin_unlock(&zone->lock); |
085cc7d5 | 1976 | return i; |
1da177e4 LT |
1977 | } |
1978 | ||
4ae7c039 | 1979 | #ifdef CONFIG_NUMA |
8fce4d8e | 1980 | /* |
4037d452 CL |
1981 | * Called from the vmstat counter updater to drain pagesets of this |
1982 | * currently executing processor on remote nodes after they have | |
1983 | * expired. | |
1984 | * | |
879336c3 CL |
1985 | * Note that this function must be called with the thread pinned to |
1986 | * a single processor. | |
8fce4d8e | 1987 | */ |
4037d452 | 1988 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1989 | { |
4ae7c039 | 1990 | unsigned long flags; |
7be12fc9 | 1991 | int to_drain, batch; |
4ae7c039 | 1992 | |
4037d452 | 1993 | local_irq_save(flags); |
4db0c3c2 | 1994 | batch = READ_ONCE(pcp->batch); |
7be12fc9 | 1995 | to_drain = min(pcp->count, batch); |
2a13515c KM |
1996 | if (to_drain > 0) { |
1997 | free_pcppages_bulk(zone, to_drain, pcp); | |
1998 | pcp->count -= to_drain; | |
1999 | } | |
4037d452 | 2000 | local_irq_restore(flags); |
4ae7c039 CL |
2001 | } |
2002 | #endif | |
2003 | ||
9f8f2172 | 2004 | /* |
93481ff0 | 2005 | * Drain pcplists of the indicated processor and zone. |
9f8f2172 CL |
2006 | * |
2007 | * The processor must either be the current processor and the | |
2008 | * thread pinned to the current processor or a processor that | |
2009 | * is not online. | |
2010 | */ | |
93481ff0 | 2011 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
1da177e4 | 2012 | { |
c54ad30c | 2013 | unsigned long flags; |
93481ff0 VB |
2014 | struct per_cpu_pageset *pset; |
2015 | struct per_cpu_pages *pcp; | |
1da177e4 | 2016 | |
93481ff0 VB |
2017 | local_irq_save(flags); |
2018 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1da177e4 | 2019 | |
93481ff0 VB |
2020 | pcp = &pset->pcp; |
2021 | if (pcp->count) { | |
2022 | free_pcppages_bulk(zone, pcp->count, pcp); | |
2023 | pcp->count = 0; | |
2024 | } | |
2025 | local_irq_restore(flags); | |
2026 | } | |
3dfa5721 | 2027 | |
93481ff0 VB |
2028 | /* |
2029 | * Drain pcplists of all zones on the indicated processor. | |
2030 | * | |
2031 | * The processor must either be the current processor and the | |
2032 | * thread pinned to the current processor or a processor that | |
2033 | * is not online. | |
2034 | */ | |
2035 | static void drain_pages(unsigned int cpu) | |
2036 | { | |
2037 | struct zone *zone; | |
2038 | ||
2039 | for_each_populated_zone(zone) { | |
2040 | drain_pages_zone(cpu, zone); | |
1da177e4 LT |
2041 | } |
2042 | } | |
1da177e4 | 2043 | |
9f8f2172 CL |
2044 | /* |
2045 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
93481ff0 VB |
2046 | * |
2047 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just | |
2048 | * the single zone's pages. | |
9f8f2172 | 2049 | */ |
93481ff0 | 2050 | void drain_local_pages(struct zone *zone) |
9f8f2172 | 2051 | { |
93481ff0 VB |
2052 | int cpu = smp_processor_id(); |
2053 | ||
2054 | if (zone) | |
2055 | drain_pages_zone(cpu, zone); | |
2056 | else | |
2057 | drain_pages(cpu); | |
9f8f2172 CL |
2058 | } |
2059 | ||
2060 | /* | |
74046494 GBY |
2061 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2062 | * | |
93481ff0 VB |
2063 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2064 | * | |
74046494 GBY |
2065 | * Note that this code is protected against sending an IPI to an offline |
2066 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
2067 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
2068 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
2069 | * before the call to on_each_cpu_mask(). | |
9f8f2172 | 2070 | */ |
93481ff0 | 2071 | void drain_all_pages(struct zone *zone) |
9f8f2172 | 2072 | { |
74046494 | 2073 | int cpu; |
74046494 GBY |
2074 | |
2075 | /* | |
2076 | * Allocate in the BSS so we wont require allocation in | |
2077 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
2078 | */ | |
2079 | static cpumask_t cpus_with_pcps; | |
2080 | ||
2081 | /* | |
2082 | * We don't care about racing with CPU hotplug event | |
2083 | * as offline notification will cause the notified | |
2084 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
2085 | * disables preemption as part of its processing | |
2086 | */ | |
2087 | for_each_online_cpu(cpu) { | |
93481ff0 VB |
2088 | struct per_cpu_pageset *pcp; |
2089 | struct zone *z; | |
74046494 | 2090 | bool has_pcps = false; |
93481ff0 VB |
2091 | |
2092 | if (zone) { | |
74046494 | 2093 | pcp = per_cpu_ptr(zone->pageset, cpu); |
93481ff0 | 2094 | if (pcp->pcp.count) |
74046494 | 2095 | has_pcps = true; |
93481ff0 VB |
2096 | } else { |
2097 | for_each_populated_zone(z) { | |
2098 | pcp = per_cpu_ptr(z->pageset, cpu); | |
2099 | if (pcp->pcp.count) { | |
2100 | has_pcps = true; | |
2101 | break; | |
2102 | } | |
74046494 GBY |
2103 | } |
2104 | } | |
93481ff0 | 2105 | |
74046494 GBY |
2106 | if (has_pcps) |
2107 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
2108 | else | |
2109 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
2110 | } | |
93481ff0 VB |
2111 | on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages, |
2112 | zone, 1); | |
9f8f2172 CL |
2113 | } |
2114 | ||
296699de | 2115 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
2116 | |
2117 | void mark_free_pages(struct zone *zone) | |
2118 | { | |
f623f0db RW |
2119 | unsigned long pfn, max_zone_pfn; |
2120 | unsigned long flags; | |
7aeb09f9 | 2121 | unsigned int order, t; |
86760a2c | 2122 | struct page *page; |
1da177e4 | 2123 | |
8080fc03 | 2124 | if (zone_is_empty(zone)) |
1da177e4 LT |
2125 | return; |
2126 | ||
2127 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 2128 | |
108bcc96 | 2129 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
2130 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2131 | if (pfn_valid(pfn)) { | |
86760a2c | 2132 | page = pfn_to_page(pfn); |
7be98234 RW |
2133 | if (!swsusp_page_is_forbidden(page)) |
2134 | swsusp_unset_page_free(page); | |
f623f0db | 2135 | } |
1da177e4 | 2136 | |
b2a0ac88 | 2137 | for_each_migratetype_order(order, t) { |
86760a2c GT |
2138 | list_for_each_entry(page, |
2139 | &zone->free_area[order].free_list[t], lru) { | |
f623f0db | 2140 | unsigned long i; |
1da177e4 | 2141 | |
86760a2c | 2142 | pfn = page_to_pfn(page); |
f623f0db | 2143 | for (i = 0; i < (1UL << order); i++) |
7be98234 | 2144 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 2145 | } |
b2a0ac88 | 2146 | } |
1da177e4 LT |
2147 | spin_unlock_irqrestore(&zone->lock, flags); |
2148 | } | |
e2c55dc8 | 2149 | #endif /* CONFIG_PM */ |
1da177e4 | 2150 | |
1da177e4 LT |
2151 | /* |
2152 | * Free a 0-order page | |
b745bc85 | 2153 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 2154 | */ |
b745bc85 | 2155 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
2156 | { |
2157 | struct zone *zone = page_zone(page); | |
2158 | struct per_cpu_pages *pcp; | |
2159 | unsigned long flags; | |
dc4b0caf | 2160 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 2161 | int migratetype; |
1da177e4 | 2162 | |
ec95f53a | 2163 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
2164 | return; |
2165 | ||
dc4b0caf | 2166 | migratetype = get_pfnblock_migratetype(page, pfn); |
bb14c2c7 | 2167 | set_pcppage_migratetype(page, migratetype); |
1da177e4 | 2168 | local_irq_save(flags); |
f8891e5e | 2169 | __count_vm_event(PGFREE); |
da456f14 | 2170 | |
5f8dcc21 MG |
2171 | /* |
2172 | * We only track unmovable, reclaimable and movable on pcp lists. | |
2173 | * Free ISOLATE pages back to the allocator because they are being | |
2174 | * offlined but treat RESERVE as movable pages so we can get those | |
2175 | * areas back if necessary. Otherwise, we may have to free | |
2176 | * excessively into the page allocator | |
2177 | */ | |
2178 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 2179 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 2180 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
2181 | goto out; |
2182 | } | |
2183 | migratetype = MIGRATE_MOVABLE; | |
2184 | } | |
2185 | ||
99dcc3e5 | 2186 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 2187 | if (!cold) |
5f8dcc21 | 2188 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
2189 | else |
2190 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 2191 | pcp->count++; |
48db57f8 | 2192 | if (pcp->count >= pcp->high) { |
4db0c3c2 | 2193 | unsigned long batch = READ_ONCE(pcp->batch); |
998d39cb CS |
2194 | free_pcppages_bulk(zone, batch, pcp); |
2195 | pcp->count -= batch; | |
48db57f8 | 2196 | } |
5f8dcc21 MG |
2197 | |
2198 | out: | |
1da177e4 | 2199 | local_irq_restore(flags); |
1da177e4 LT |
2200 | } |
2201 | ||
cc59850e KK |
2202 | /* |
2203 | * Free a list of 0-order pages | |
2204 | */ | |
b745bc85 | 2205 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
2206 | { |
2207 | struct page *page, *next; | |
2208 | ||
2209 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 2210 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
2211 | free_hot_cold_page(page, cold); |
2212 | } | |
2213 | } | |
2214 | ||
8dfcc9ba NP |
2215 | /* |
2216 | * split_page takes a non-compound higher-order page, and splits it into | |
2217 | * n (1<<order) sub-pages: page[0..n] | |
2218 | * Each sub-page must be freed individually. | |
2219 | * | |
2220 | * Note: this is probably too low level an operation for use in drivers. | |
2221 | * Please consult with lkml before using this in your driver. | |
2222 | */ | |
2223 | void split_page(struct page *page, unsigned int order) | |
2224 | { | |
2225 | int i; | |
e2cfc911 | 2226 | gfp_t gfp_mask; |
8dfcc9ba | 2227 | |
309381fe SL |
2228 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2229 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
2230 | |
2231 | #ifdef CONFIG_KMEMCHECK | |
2232 | /* | |
2233 | * Split shadow pages too, because free(page[0]) would | |
2234 | * otherwise free the whole shadow. | |
2235 | */ | |
2236 | if (kmemcheck_page_is_tracked(page)) | |
2237 | split_page(virt_to_page(page[0].shadow), order); | |
2238 | #endif | |
2239 | ||
e2cfc911 JK |
2240 | gfp_mask = get_page_owner_gfp(page); |
2241 | set_page_owner(page, 0, gfp_mask); | |
48c96a36 | 2242 | for (i = 1; i < (1 << order); i++) { |
7835e98b | 2243 | set_page_refcounted(page + i); |
e2cfc911 | 2244 | set_page_owner(page + i, 0, gfp_mask); |
48c96a36 | 2245 | } |
8dfcc9ba | 2246 | } |
5853ff23 | 2247 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 2248 | |
3c605096 | 2249 | int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 2250 | { |
748446bb MG |
2251 | unsigned long watermark; |
2252 | struct zone *zone; | |
2139cbe6 | 2253 | int mt; |
748446bb MG |
2254 | |
2255 | BUG_ON(!PageBuddy(page)); | |
2256 | ||
2257 | zone = page_zone(page); | |
2e30abd1 | 2258 | mt = get_pageblock_migratetype(page); |
748446bb | 2259 | |
194159fb | 2260 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
2261 | /* Obey watermarks as if the page was being allocated */ |
2262 | watermark = low_wmark_pages(zone) + (1 << order); | |
2263 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
2264 | return 0; | |
2265 | ||
8fb74b9f | 2266 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 2267 | } |
748446bb MG |
2268 | |
2269 | /* Remove page from free list */ | |
2270 | list_del(&page->lru); | |
2271 | zone->free_area[order].nr_free--; | |
2272 | rmv_page_order(page); | |
2139cbe6 | 2273 | |
e2cfc911 | 2274 | set_page_owner(page, order, __GFP_MOVABLE); |
f3a14ced | 2275 | |
8fb74b9f | 2276 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
2277 | if (order >= pageblock_order - 1) { |
2278 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
2279 | for (; page < endpage; page += pageblock_nr_pages) { |
2280 | int mt = get_pageblock_migratetype(page); | |
194159fb | 2281 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
2282 | set_pageblock_migratetype(page, |
2283 | MIGRATE_MOVABLE); | |
2284 | } | |
748446bb MG |
2285 | } |
2286 | ||
f3a14ced | 2287 | |
8fb74b9f | 2288 | return 1UL << order; |
1fb3f8ca MG |
2289 | } |
2290 | ||
2291 | /* | |
2292 | * Similar to split_page except the page is already free. As this is only | |
2293 | * being used for migration, the migratetype of the block also changes. | |
2294 | * As this is called with interrupts disabled, the caller is responsible | |
2295 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
2296 | * are enabled. | |
2297 | * | |
2298 | * Note: this is probably too low level an operation for use in drivers. | |
2299 | * Please consult with lkml before using this in your driver. | |
2300 | */ | |
2301 | int split_free_page(struct page *page) | |
2302 | { | |
2303 | unsigned int order; | |
2304 | int nr_pages; | |
2305 | ||
1fb3f8ca MG |
2306 | order = page_order(page); |
2307 | ||
8fb74b9f | 2308 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
2309 | if (!nr_pages) |
2310 | return 0; | |
2311 | ||
2312 | /* Split into individual pages */ | |
2313 | set_page_refcounted(page); | |
2314 | split_page(page, order); | |
2315 | return nr_pages; | |
748446bb MG |
2316 | } |
2317 | ||
1da177e4 | 2318 | /* |
75379191 | 2319 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
1da177e4 | 2320 | */ |
0a15c3e9 MG |
2321 | static inline |
2322 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
7aeb09f9 | 2323 | struct zone *zone, unsigned int order, |
0aaa29a5 | 2324 | gfp_t gfp_flags, int alloc_flags, int migratetype) |
1da177e4 LT |
2325 | { |
2326 | unsigned long flags; | |
689bcebf | 2327 | struct page *page; |
b745bc85 | 2328 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
1da177e4 | 2329 | |
48db57f8 | 2330 | if (likely(order == 0)) { |
1da177e4 | 2331 | struct per_cpu_pages *pcp; |
5f8dcc21 | 2332 | struct list_head *list; |
1da177e4 | 2333 | |
1da177e4 | 2334 | local_irq_save(flags); |
99dcc3e5 CL |
2335 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2336 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 2337 | if (list_empty(list)) { |
535131e6 | 2338 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 2339 | pcp->batch, list, |
e084b2d9 | 2340 | migratetype, cold); |
5f8dcc21 | 2341 | if (unlikely(list_empty(list))) |
6fb332fa | 2342 | goto failed; |
535131e6 | 2343 | } |
b92a6edd | 2344 | |
5f8dcc21 | 2345 | if (cold) |
a16601c5 | 2346 | page = list_last_entry(list, struct page, lru); |
5f8dcc21 | 2347 | else |
a16601c5 | 2348 | page = list_first_entry(list, struct page, lru); |
5f8dcc21 | 2349 | |
b92a6edd MG |
2350 | list_del(&page->lru); |
2351 | pcp->count--; | |
7fb1d9fc | 2352 | } else { |
dab48dab AM |
2353 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
2354 | /* | |
2355 | * __GFP_NOFAIL is not to be used in new code. | |
2356 | * | |
2357 | * All __GFP_NOFAIL callers should be fixed so that they | |
2358 | * properly detect and handle allocation failures. | |
2359 | * | |
2360 | * We most definitely don't want callers attempting to | |
4923abf9 | 2361 | * allocate greater than order-1 page units with |
dab48dab AM |
2362 | * __GFP_NOFAIL. |
2363 | */ | |
4923abf9 | 2364 | WARN_ON_ONCE(order > 1); |
dab48dab | 2365 | } |
1da177e4 | 2366 | spin_lock_irqsave(&zone->lock, flags); |
0aaa29a5 MG |
2367 | |
2368 | page = NULL; | |
2369 | if (alloc_flags & ALLOC_HARDER) { | |
2370 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); | |
2371 | if (page) | |
2372 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
2373 | } | |
2374 | if (!page) | |
6ac0206b | 2375 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
2376 | spin_unlock(&zone->lock); |
2377 | if (!page) | |
2378 | goto failed; | |
d1ce749a | 2379 | __mod_zone_freepage_state(zone, -(1 << order), |
bb14c2c7 | 2380 | get_pcppage_migratetype(page)); |
1da177e4 LT |
2381 | } |
2382 | ||
3a025760 | 2383 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
abe5f972 | 2384 | if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 && |
57054651 JW |
2385 | !test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) |
2386 | set_bit(ZONE_FAIR_DEPLETED, &zone->flags); | |
27329369 | 2387 | |
f8891e5e | 2388 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 2389 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 2390 | local_irq_restore(flags); |
1da177e4 | 2391 | |
309381fe | 2392 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
1da177e4 | 2393 | return page; |
a74609fa NP |
2394 | |
2395 | failed: | |
2396 | local_irq_restore(flags); | |
a74609fa | 2397 | return NULL; |
1da177e4 LT |
2398 | } |
2399 | ||
933e312e AM |
2400 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2401 | ||
b2588c4b | 2402 | static struct { |
933e312e AM |
2403 | struct fault_attr attr; |
2404 | ||
621a5f7a | 2405 | bool ignore_gfp_highmem; |
71baba4b | 2406 | bool ignore_gfp_reclaim; |
54114994 | 2407 | u32 min_order; |
933e312e AM |
2408 | } fail_page_alloc = { |
2409 | .attr = FAULT_ATTR_INITIALIZER, | |
71baba4b | 2410 | .ignore_gfp_reclaim = true, |
621a5f7a | 2411 | .ignore_gfp_highmem = true, |
54114994 | 2412 | .min_order = 1, |
933e312e AM |
2413 | }; |
2414 | ||
2415 | static int __init setup_fail_page_alloc(char *str) | |
2416 | { | |
2417 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
2418 | } | |
2419 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
2420 | ||
deaf386e | 2421 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2422 | { |
54114994 | 2423 | if (order < fail_page_alloc.min_order) |
deaf386e | 2424 | return false; |
933e312e | 2425 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 2426 | return false; |
933e312e | 2427 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 2428 | return false; |
71baba4b MG |
2429 | if (fail_page_alloc.ignore_gfp_reclaim && |
2430 | (gfp_mask & __GFP_DIRECT_RECLAIM)) | |
deaf386e | 2431 | return false; |
933e312e AM |
2432 | |
2433 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
2434 | } | |
2435 | ||
2436 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
2437 | ||
2438 | static int __init fail_page_alloc_debugfs(void) | |
2439 | { | |
f4ae40a6 | 2440 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 2441 | struct dentry *dir; |
933e312e | 2442 | |
dd48c085 AM |
2443 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2444 | &fail_page_alloc.attr); | |
2445 | if (IS_ERR(dir)) | |
2446 | return PTR_ERR(dir); | |
933e312e | 2447 | |
b2588c4b | 2448 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
71baba4b | 2449 | &fail_page_alloc.ignore_gfp_reclaim)) |
b2588c4b AM |
2450 | goto fail; |
2451 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
2452 | &fail_page_alloc.ignore_gfp_highmem)) | |
2453 | goto fail; | |
2454 | if (!debugfs_create_u32("min-order", mode, dir, | |
2455 | &fail_page_alloc.min_order)) | |
2456 | goto fail; | |
2457 | ||
2458 | return 0; | |
2459 | fail: | |
dd48c085 | 2460 | debugfs_remove_recursive(dir); |
933e312e | 2461 | |
b2588c4b | 2462 | return -ENOMEM; |
933e312e AM |
2463 | } |
2464 | ||
2465 | late_initcall(fail_page_alloc_debugfs); | |
2466 | ||
2467 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
2468 | ||
2469 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
2470 | ||
deaf386e | 2471 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 2472 | { |
deaf386e | 2473 | return false; |
933e312e AM |
2474 | } |
2475 | ||
2476 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
2477 | ||
1da177e4 | 2478 | /* |
97a16fc8 MG |
2479 | * Return true if free base pages are above 'mark'. For high-order checks it |
2480 | * will return true of the order-0 watermark is reached and there is at least | |
2481 | * one free page of a suitable size. Checking now avoids taking the zone lock | |
2482 | * to check in the allocation paths if no pages are free. | |
1da177e4 | 2483 | */ |
7aeb09f9 MG |
2484 | static bool __zone_watermark_ok(struct zone *z, unsigned int order, |
2485 | unsigned long mark, int classzone_idx, int alloc_flags, | |
2486 | long free_pages) | |
1da177e4 | 2487 | { |
d23ad423 | 2488 | long min = mark; |
1da177e4 | 2489 | int o; |
97a16fc8 | 2490 | const int alloc_harder = (alloc_flags & ALLOC_HARDER); |
1da177e4 | 2491 | |
0aaa29a5 | 2492 | /* free_pages may go negative - that's OK */ |
df0a6daa | 2493 | free_pages -= (1 << order) - 1; |
0aaa29a5 | 2494 | |
7fb1d9fc | 2495 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 2496 | min -= min / 2; |
0aaa29a5 MG |
2497 | |
2498 | /* | |
2499 | * If the caller does not have rights to ALLOC_HARDER then subtract | |
2500 | * the high-atomic reserves. This will over-estimate the size of the | |
2501 | * atomic reserve but it avoids a search. | |
2502 | */ | |
97a16fc8 | 2503 | if (likely(!alloc_harder)) |
0aaa29a5 MG |
2504 | free_pages -= z->nr_reserved_highatomic; |
2505 | else | |
1da177e4 | 2506 | min -= min / 4; |
e2b19197 | 2507 | |
d95ea5d1 BZ |
2508 | #ifdef CONFIG_CMA |
2509 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
2510 | if (!(alloc_flags & ALLOC_CMA)) | |
97a16fc8 | 2511 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 2512 | #endif |
026b0814 | 2513 | |
97a16fc8 MG |
2514 | /* |
2515 | * Check watermarks for an order-0 allocation request. If these | |
2516 | * are not met, then a high-order request also cannot go ahead | |
2517 | * even if a suitable page happened to be free. | |
2518 | */ | |
2519 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
88f5acf8 | 2520 | return false; |
1da177e4 | 2521 | |
97a16fc8 MG |
2522 | /* If this is an order-0 request then the watermark is fine */ |
2523 | if (!order) | |
2524 | return true; | |
2525 | ||
2526 | /* For a high-order request, check at least one suitable page is free */ | |
2527 | for (o = order; o < MAX_ORDER; o++) { | |
2528 | struct free_area *area = &z->free_area[o]; | |
2529 | int mt; | |
2530 | ||
2531 | if (!area->nr_free) | |
2532 | continue; | |
2533 | ||
2534 | if (alloc_harder) | |
2535 | return true; | |
1da177e4 | 2536 | |
97a16fc8 MG |
2537 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
2538 | if (!list_empty(&area->free_list[mt])) | |
2539 | return true; | |
2540 | } | |
2541 | ||
2542 | #ifdef CONFIG_CMA | |
2543 | if ((alloc_flags & ALLOC_CMA) && | |
2544 | !list_empty(&area->free_list[MIGRATE_CMA])) { | |
2545 | return true; | |
2546 | } | |
2547 | #endif | |
1da177e4 | 2548 | } |
97a16fc8 | 2549 | return false; |
88f5acf8 MG |
2550 | } |
2551 | ||
7aeb09f9 | 2552 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
88f5acf8 MG |
2553 | int classzone_idx, int alloc_flags) |
2554 | { | |
2555 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
2556 | zone_page_state(z, NR_FREE_PAGES)); | |
2557 | } | |
2558 | ||
7aeb09f9 | 2559 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
e2b19197 | 2560 | unsigned long mark, int classzone_idx) |
88f5acf8 MG |
2561 | { |
2562 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
2563 | ||
2564 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
2565 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
2566 | ||
e2b19197 | 2567 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
88f5acf8 | 2568 | free_pages); |
1da177e4 LT |
2569 | } |
2570 | ||
9276b1bc | 2571 | #ifdef CONFIG_NUMA |
81c0a2bb JW |
2572 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2573 | { | |
fff4068c | 2574 | return local_zone->node == zone->node; |
81c0a2bb JW |
2575 | } |
2576 | ||
957f822a DR |
2577 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2578 | { | |
5f7a75ac MG |
2579 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
2580 | RECLAIM_DISTANCE; | |
957f822a | 2581 | } |
9276b1bc | 2582 | #else /* CONFIG_NUMA */ |
81c0a2bb JW |
2583 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
2584 | { | |
2585 | return true; | |
2586 | } | |
2587 | ||
957f822a DR |
2588 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
2589 | { | |
2590 | return true; | |
2591 | } | |
9276b1bc PJ |
2592 | #endif /* CONFIG_NUMA */ |
2593 | ||
4ffeaf35 MG |
2594 | static void reset_alloc_batches(struct zone *preferred_zone) |
2595 | { | |
2596 | struct zone *zone = preferred_zone->zone_pgdat->node_zones; | |
2597 | ||
2598 | do { | |
2599 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
2600 | high_wmark_pages(zone) - low_wmark_pages(zone) - | |
2601 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
57054651 | 2602 | clear_bit(ZONE_FAIR_DEPLETED, &zone->flags); |
4ffeaf35 MG |
2603 | } while (zone++ != preferred_zone); |
2604 | } | |
2605 | ||
7fb1d9fc | 2606 | /* |
0798e519 | 2607 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
2608 | * a page. |
2609 | */ | |
2610 | static struct page * | |
a9263751 VB |
2611 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
2612 | const struct alloc_context *ac) | |
753ee728 | 2613 | { |
a9263751 | 2614 | struct zonelist *zonelist = ac->zonelist; |
dd1a239f | 2615 | struct zoneref *z; |
7fb1d9fc | 2616 | struct page *page = NULL; |
5117f45d | 2617 | struct zone *zone; |
4ffeaf35 MG |
2618 | int nr_fair_skipped = 0; |
2619 | bool zonelist_rescan; | |
54a6eb5c | 2620 | |
9276b1bc | 2621 | zonelist_scan: |
4ffeaf35 MG |
2622 | zonelist_rescan = false; |
2623 | ||
7fb1d9fc | 2624 | /* |
9276b1bc | 2625 | * Scan zonelist, looking for a zone with enough free. |
344736f2 | 2626 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
7fb1d9fc | 2627 | */ |
a9263751 VB |
2628 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, |
2629 | ac->nodemask) { | |
e085dbc5 JW |
2630 | unsigned long mark; |
2631 | ||
664eedde MG |
2632 | if (cpusets_enabled() && |
2633 | (alloc_flags & ALLOC_CPUSET) && | |
344736f2 | 2634 | !cpuset_zone_allowed(zone, gfp_mask)) |
cd38b115 | 2635 | continue; |
81c0a2bb JW |
2636 | /* |
2637 | * Distribute pages in proportion to the individual | |
2638 | * zone size to ensure fair page aging. The zone a | |
2639 | * page was allocated in should have no effect on the | |
2640 | * time the page has in memory before being reclaimed. | |
81c0a2bb | 2641 | */ |
3a025760 | 2642 | if (alloc_flags & ALLOC_FAIR) { |
a9263751 | 2643 | if (!zone_local(ac->preferred_zone, zone)) |
f7b5d647 | 2644 | break; |
57054651 | 2645 | if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) { |
4ffeaf35 | 2646 | nr_fair_skipped++; |
3a025760 | 2647 | continue; |
4ffeaf35 | 2648 | } |
81c0a2bb | 2649 | } |
a756cf59 JW |
2650 | /* |
2651 | * When allocating a page cache page for writing, we | |
2652 | * want to get it from a zone that is within its dirty | |
2653 | * limit, such that no single zone holds more than its | |
2654 | * proportional share of globally allowed dirty pages. | |
2655 | * The dirty limits take into account the zone's | |
2656 | * lowmem reserves and high watermark so that kswapd | |
2657 | * should be able to balance it without having to | |
2658 | * write pages from its LRU list. | |
2659 | * | |
2660 | * This may look like it could increase pressure on | |
2661 | * lower zones by failing allocations in higher zones | |
2662 | * before they are full. But the pages that do spill | |
2663 | * over are limited as the lower zones are protected | |
2664 | * by this very same mechanism. It should not become | |
2665 | * a practical burden to them. | |
2666 | * | |
2667 | * XXX: For now, allow allocations to potentially | |
2668 | * exceed the per-zone dirty limit in the slowpath | |
c9ab0c4f | 2669 | * (spread_dirty_pages unset) before going into reclaim, |
a756cf59 JW |
2670 | * which is important when on a NUMA setup the allowed |
2671 | * zones are together not big enough to reach the | |
2672 | * global limit. The proper fix for these situations | |
2673 | * will require awareness of zones in the | |
2674 | * dirty-throttling and the flusher threads. | |
2675 | */ | |
c9ab0c4f | 2676 | if (ac->spread_dirty_pages && !zone_dirty_ok(zone)) |
800a1e75 | 2677 | continue; |
7fb1d9fc | 2678 | |
e085dbc5 JW |
2679 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
2680 | if (!zone_watermark_ok(zone, order, mark, | |
a9263751 | 2681 | ac->classzone_idx, alloc_flags)) { |
fa5e084e MG |
2682 | int ret; |
2683 | ||
5dab2911 MG |
2684 | /* Checked here to keep the fast path fast */ |
2685 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
2686 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
2687 | goto try_this_zone; | |
2688 | ||
957f822a | 2689 | if (zone_reclaim_mode == 0 || |
a9263751 | 2690 | !zone_allows_reclaim(ac->preferred_zone, zone)) |
cd38b115 MG |
2691 | continue; |
2692 | ||
fa5e084e MG |
2693 | ret = zone_reclaim(zone, gfp_mask, order); |
2694 | switch (ret) { | |
2695 | case ZONE_RECLAIM_NOSCAN: | |
2696 | /* did not scan */ | |
cd38b115 | 2697 | continue; |
fa5e084e MG |
2698 | case ZONE_RECLAIM_FULL: |
2699 | /* scanned but unreclaimable */ | |
cd38b115 | 2700 | continue; |
fa5e084e MG |
2701 | default: |
2702 | /* did we reclaim enough */ | |
fed2719e | 2703 | if (zone_watermark_ok(zone, order, mark, |
a9263751 | 2704 | ac->classzone_idx, alloc_flags)) |
fed2719e MG |
2705 | goto try_this_zone; |
2706 | ||
fed2719e | 2707 | continue; |
0798e519 | 2708 | } |
7fb1d9fc RS |
2709 | } |
2710 | ||
fa5e084e | 2711 | try_this_zone: |
a9263751 | 2712 | page = buffered_rmqueue(ac->preferred_zone, zone, order, |
0aaa29a5 | 2713 | gfp_mask, alloc_flags, ac->migratetype); |
75379191 VB |
2714 | if (page) { |
2715 | if (prep_new_page(page, order, gfp_mask, alloc_flags)) | |
2716 | goto try_this_zone; | |
0aaa29a5 MG |
2717 | |
2718 | /* | |
2719 | * If this is a high-order atomic allocation then check | |
2720 | * if the pageblock should be reserved for the future | |
2721 | */ | |
2722 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) | |
2723 | reserve_highatomic_pageblock(page, zone, order); | |
2724 | ||
75379191 VB |
2725 | return page; |
2726 | } | |
54a6eb5c | 2727 | } |
9276b1bc | 2728 | |
4ffeaf35 MG |
2729 | /* |
2730 | * The first pass makes sure allocations are spread fairly within the | |
2731 | * local node. However, the local node might have free pages left | |
2732 | * after the fairness batches are exhausted, and remote zones haven't | |
2733 | * even been considered yet. Try once more without fairness, and | |
2734 | * include remote zones now, before entering the slowpath and waking | |
2735 | * kswapd: prefer spilling to a remote zone over swapping locally. | |
2736 | */ | |
2737 | if (alloc_flags & ALLOC_FAIR) { | |
2738 | alloc_flags &= ~ALLOC_FAIR; | |
2739 | if (nr_fair_skipped) { | |
2740 | zonelist_rescan = true; | |
a9263751 | 2741 | reset_alloc_batches(ac->preferred_zone); |
4ffeaf35 MG |
2742 | } |
2743 | if (nr_online_nodes > 1) | |
2744 | zonelist_rescan = true; | |
2745 | } | |
2746 | ||
4ffeaf35 MG |
2747 | if (zonelist_rescan) |
2748 | goto zonelist_scan; | |
2749 | ||
2750 | return NULL; | |
753ee728 MH |
2751 | } |
2752 | ||
29423e77 DR |
2753 | /* |
2754 | * Large machines with many possible nodes should not always dump per-node | |
2755 | * meminfo in irq context. | |
2756 | */ | |
2757 | static inline bool should_suppress_show_mem(void) | |
2758 | { | |
2759 | bool ret = false; | |
2760 | ||
2761 | #if NODES_SHIFT > 8 | |
2762 | ret = in_interrupt(); | |
2763 | #endif | |
2764 | return ret; | |
2765 | } | |
2766 | ||
a238ab5b DH |
2767 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2768 | DEFAULT_RATELIMIT_INTERVAL, | |
2769 | DEFAULT_RATELIMIT_BURST); | |
2770 | ||
d00181b9 | 2771 | void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...) |
a238ab5b | 2772 | { |
a238ab5b DH |
2773 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2774 | ||
c0a32fc5 SG |
2775 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2776 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2777 | return; |
2778 | ||
2779 | /* | |
2780 | * This documents exceptions given to allocations in certain | |
2781 | * contexts that are allowed to allocate outside current's set | |
2782 | * of allowed nodes. | |
2783 | */ | |
2784 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2785 | if (test_thread_flag(TIF_MEMDIE) || | |
2786 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2787 | filter &= ~SHOW_MEM_FILTER_NODES; | |
d0164adc | 2788 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
a238ab5b DH |
2789 | filter &= ~SHOW_MEM_FILTER_NODES; |
2790 | ||
2791 | if (fmt) { | |
3ee9a4f0 JP |
2792 | struct va_format vaf; |
2793 | va_list args; | |
2794 | ||
a238ab5b | 2795 | va_start(args, fmt); |
3ee9a4f0 JP |
2796 | |
2797 | vaf.fmt = fmt; | |
2798 | vaf.va = &args; | |
2799 | ||
2800 | pr_warn("%pV", &vaf); | |
2801 | ||
a238ab5b DH |
2802 | va_end(args); |
2803 | } | |
2804 | ||
c5c990e8 VB |
2805 | pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n", |
2806 | current->comm, order, gfp_mask, &gfp_mask); | |
a238ab5b DH |
2807 | dump_stack(); |
2808 | if (!should_suppress_show_mem()) | |
2809 | show_mem(filter); | |
2810 | } | |
2811 | ||
11e33f6a MG |
2812 | static inline struct page * |
2813 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 2814 | const struct alloc_context *ac, unsigned long *did_some_progress) |
11e33f6a | 2815 | { |
6e0fc46d DR |
2816 | struct oom_control oc = { |
2817 | .zonelist = ac->zonelist, | |
2818 | .nodemask = ac->nodemask, | |
2819 | .gfp_mask = gfp_mask, | |
2820 | .order = order, | |
6e0fc46d | 2821 | }; |
11e33f6a MG |
2822 | struct page *page; |
2823 | ||
9879de73 JW |
2824 | *did_some_progress = 0; |
2825 | ||
9879de73 | 2826 | /* |
dc56401f JW |
2827 | * Acquire the oom lock. If that fails, somebody else is |
2828 | * making progress for us. | |
9879de73 | 2829 | */ |
dc56401f | 2830 | if (!mutex_trylock(&oom_lock)) { |
9879de73 | 2831 | *did_some_progress = 1; |
11e33f6a | 2832 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2833 | return NULL; |
2834 | } | |
6b1de916 | 2835 | |
11e33f6a MG |
2836 | /* |
2837 | * Go through the zonelist yet one more time, keep very high watermark | |
2838 | * here, this is only to catch a parallel oom killing, we must fail if | |
2839 | * we're still under heavy pressure. | |
2840 | */ | |
a9263751 VB |
2841 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
2842 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); | |
7fb1d9fc | 2843 | if (page) |
11e33f6a MG |
2844 | goto out; |
2845 | ||
4365a567 | 2846 | if (!(gfp_mask & __GFP_NOFAIL)) { |
9879de73 JW |
2847 | /* Coredumps can quickly deplete all memory reserves */ |
2848 | if (current->flags & PF_DUMPCORE) | |
2849 | goto out; | |
4365a567 KH |
2850 | /* The OOM killer will not help higher order allocs */ |
2851 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2852 | goto out; | |
03668b3c | 2853 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
a9263751 | 2854 | if (ac->high_zoneidx < ZONE_NORMAL) |
03668b3c | 2855 | goto out; |
9083905a | 2856 | /* The OOM killer does not compensate for IO-less reclaim */ |
cc873177 JW |
2857 | if (!(gfp_mask & __GFP_FS)) { |
2858 | /* | |
2859 | * XXX: Page reclaim didn't yield anything, | |
2860 | * and the OOM killer can't be invoked, but | |
9083905a | 2861 | * keep looping as per tradition. |
cc873177 JW |
2862 | */ |
2863 | *did_some_progress = 1; | |
9879de73 | 2864 | goto out; |
cc873177 | 2865 | } |
9083905a JW |
2866 | if (pm_suspended_storage()) |
2867 | goto out; | |
4167e9b2 | 2868 | /* The OOM killer may not free memory on a specific node */ |
4365a567 KH |
2869 | if (gfp_mask & __GFP_THISNODE) |
2870 | goto out; | |
2871 | } | |
11e33f6a | 2872 | /* Exhausted what can be done so it's blamo time */ |
5020e285 | 2873 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
c32b3cbe | 2874 | *did_some_progress = 1; |
5020e285 MH |
2875 | |
2876 | if (gfp_mask & __GFP_NOFAIL) { | |
2877 | page = get_page_from_freelist(gfp_mask, order, | |
2878 | ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac); | |
2879 | /* | |
2880 | * fallback to ignore cpuset restriction if our nodes | |
2881 | * are depleted | |
2882 | */ | |
2883 | if (!page) | |
2884 | page = get_page_from_freelist(gfp_mask, order, | |
2885 | ALLOC_NO_WATERMARKS, ac); | |
2886 | } | |
2887 | } | |
11e33f6a | 2888 | out: |
dc56401f | 2889 | mutex_unlock(&oom_lock); |
11e33f6a MG |
2890 | return page; |
2891 | } | |
2892 | ||
56de7263 MG |
2893 | #ifdef CONFIG_COMPACTION |
2894 | /* Try memory compaction for high-order allocations before reclaim */ | |
2895 | static struct page * | |
2896 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
a9263751 VB |
2897 | int alloc_flags, const struct alloc_context *ac, |
2898 | enum migrate_mode mode, int *contended_compaction, | |
2899 | bool *deferred_compaction) | |
56de7263 | 2900 | { |
53853e2d | 2901 | unsigned long compact_result; |
98dd3b48 | 2902 | struct page *page; |
53853e2d VB |
2903 | |
2904 | if (!order) | |
66199712 | 2905 | return NULL; |
66199712 | 2906 | |
c06b1fca | 2907 | current->flags |= PF_MEMALLOC; |
1a6d53a1 VB |
2908 | compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
2909 | mode, contended_compaction); | |
c06b1fca | 2910 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2911 | |
98dd3b48 VB |
2912 | switch (compact_result) { |
2913 | case COMPACT_DEFERRED: | |
53853e2d | 2914 | *deferred_compaction = true; |
98dd3b48 VB |
2915 | /* fall-through */ |
2916 | case COMPACT_SKIPPED: | |
2917 | return NULL; | |
2918 | default: | |
2919 | break; | |
2920 | } | |
53853e2d | 2921 | |
98dd3b48 VB |
2922 | /* |
2923 | * At least in one zone compaction wasn't deferred or skipped, so let's | |
2924 | * count a compaction stall | |
2925 | */ | |
2926 | count_vm_event(COMPACTSTALL); | |
8fb74b9f | 2927 | |
a9263751 VB |
2928 | page = get_page_from_freelist(gfp_mask, order, |
2929 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
53853e2d | 2930 | |
98dd3b48 VB |
2931 | if (page) { |
2932 | struct zone *zone = page_zone(page); | |
53853e2d | 2933 | |
98dd3b48 VB |
2934 | zone->compact_blockskip_flush = false; |
2935 | compaction_defer_reset(zone, order, true); | |
2936 | count_vm_event(COMPACTSUCCESS); | |
2937 | return page; | |
2938 | } | |
56de7263 | 2939 | |
98dd3b48 VB |
2940 | /* |
2941 | * It's bad if compaction run occurs and fails. The most likely reason | |
2942 | * is that pages exist, but not enough to satisfy watermarks. | |
2943 | */ | |
2944 | count_vm_event(COMPACTFAIL); | |
66199712 | 2945 | |
98dd3b48 | 2946 | cond_resched(); |
56de7263 MG |
2947 | |
2948 | return NULL; | |
2949 | } | |
2950 | #else | |
2951 | static inline struct page * | |
2952 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
a9263751 VB |
2953 | int alloc_flags, const struct alloc_context *ac, |
2954 | enum migrate_mode mode, int *contended_compaction, | |
2955 | bool *deferred_compaction) | |
56de7263 MG |
2956 | { |
2957 | return NULL; | |
2958 | } | |
2959 | #endif /* CONFIG_COMPACTION */ | |
2960 | ||
bba90710 MS |
2961 | /* Perform direct synchronous page reclaim */ |
2962 | static int | |
a9263751 VB |
2963 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
2964 | const struct alloc_context *ac) | |
11e33f6a | 2965 | { |
11e33f6a | 2966 | struct reclaim_state reclaim_state; |
bba90710 | 2967 | int progress; |
11e33f6a MG |
2968 | |
2969 | cond_resched(); | |
2970 | ||
2971 | /* We now go into synchronous reclaim */ | |
2972 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2973 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2974 | lockdep_set_current_reclaim_state(gfp_mask); |
2975 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2976 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2977 | |
a9263751 VB |
2978 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
2979 | ac->nodemask); | |
11e33f6a | 2980 | |
c06b1fca | 2981 | current->reclaim_state = NULL; |
11e33f6a | 2982 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2983 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2984 | |
2985 | cond_resched(); | |
2986 | ||
bba90710 MS |
2987 | return progress; |
2988 | } | |
2989 | ||
2990 | /* The really slow allocator path where we enter direct reclaim */ | |
2991 | static inline struct page * | |
2992 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
a9263751 VB |
2993 | int alloc_flags, const struct alloc_context *ac, |
2994 | unsigned long *did_some_progress) | |
bba90710 MS |
2995 | { |
2996 | struct page *page = NULL; | |
2997 | bool drained = false; | |
2998 | ||
a9263751 | 2999 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
9ee493ce MG |
3000 | if (unlikely(!(*did_some_progress))) |
3001 | return NULL; | |
11e33f6a | 3002 | |
9ee493ce | 3003 | retry: |
a9263751 VB |
3004 | page = get_page_from_freelist(gfp_mask, order, |
3005 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
9ee493ce MG |
3006 | |
3007 | /* | |
3008 | * If an allocation failed after direct reclaim, it could be because | |
0aaa29a5 MG |
3009 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3010 | * Shrink them them and try again | |
9ee493ce MG |
3011 | */ |
3012 | if (!page && !drained) { | |
0aaa29a5 | 3013 | unreserve_highatomic_pageblock(ac); |
93481ff0 | 3014 | drain_all_pages(NULL); |
9ee493ce MG |
3015 | drained = true; |
3016 | goto retry; | |
3017 | } | |
3018 | ||
11e33f6a MG |
3019 | return page; |
3020 | } | |
3021 | ||
a9263751 | 3022 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3a025760 JW |
3023 | { |
3024 | struct zoneref *z; | |
3025 | struct zone *zone; | |
3026 | ||
a9263751 VB |
3027 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3028 | ac->high_zoneidx, ac->nodemask) | |
3029 | wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone)); | |
3a025760 JW |
3030 | } |
3031 | ||
341ce06f PZ |
3032 | static inline int |
3033 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
3034 | { | |
341ce06f | 3035 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
1da177e4 | 3036 | |
a56f57ff | 3037 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 3038 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 3039 | |
341ce06f PZ |
3040 | /* |
3041 | * The caller may dip into page reserves a bit more if the caller | |
3042 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
3043 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
d0164adc | 3044 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 3045 | */ |
e6223a3b | 3046 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 3047 | |
d0164adc | 3048 | if (gfp_mask & __GFP_ATOMIC) { |
5c3240d9 | 3049 | /* |
b104a35d DR |
3050 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3051 | * if it can't schedule. | |
5c3240d9 | 3052 | */ |
b104a35d | 3053 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 3054 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 3055 | /* |
b104a35d | 3056 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
344736f2 | 3057 | * comment for __cpuset_node_allowed(). |
523b9458 | 3058 | */ |
341ce06f | 3059 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 3060 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
3061 | alloc_flags |= ALLOC_HARDER; |
3062 | ||
b37f1dd0 MG |
3063 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
3064 | if (gfp_mask & __GFP_MEMALLOC) | |
3065 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
3066 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
3067 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
3068 | else if (!in_interrupt() && | |
3069 | ((current->flags & PF_MEMALLOC) || | |
3070 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 3071 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 3072 | } |
d95ea5d1 | 3073 | #ifdef CONFIG_CMA |
43e7a34d | 3074 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
d95ea5d1 BZ |
3075 | alloc_flags |= ALLOC_CMA; |
3076 | #endif | |
341ce06f PZ |
3077 | return alloc_flags; |
3078 | } | |
3079 | ||
072bb0aa MG |
3080 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3081 | { | |
b37f1dd0 | 3082 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
3083 | } |
3084 | ||
d0164adc MG |
3085 | static inline bool is_thp_gfp_mask(gfp_t gfp_mask) |
3086 | { | |
3087 | return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE; | |
3088 | } | |
3089 | ||
11e33f6a MG |
3090 | static inline struct page * |
3091 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
a9263751 | 3092 | struct alloc_context *ac) |
11e33f6a | 3093 | { |
d0164adc | 3094 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
11e33f6a MG |
3095 | struct page *page = NULL; |
3096 | int alloc_flags; | |
3097 | unsigned long pages_reclaimed = 0; | |
3098 | unsigned long did_some_progress; | |
e0b9daeb | 3099 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
66199712 | 3100 | bool deferred_compaction = false; |
1f9efdef | 3101 | int contended_compaction = COMPACT_CONTENDED_NONE; |
1da177e4 | 3102 | |
72807a74 MG |
3103 | /* |
3104 | * In the slowpath, we sanity check order to avoid ever trying to | |
3105 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
3106 | * be using allocators in order of preference for an area that is | |
3107 | * too large. | |
3108 | */ | |
1fc28b70 MG |
3109 | if (order >= MAX_ORDER) { |
3110 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 3111 | return NULL; |
1fc28b70 | 3112 | } |
1da177e4 | 3113 | |
d0164adc MG |
3114 | /* |
3115 | * We also sanity check to catch abuse of atomic reserves being used by | |
3116 | * callers that are not in atomic context. | |
3117 | */ | |
3118 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == | |
3119 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) | |
3120 | gfp_mask &= ~__GFP_ATOMIC; | |
3121 | ||
9879de73 | 3122 | retry: |
d0164adc | 3123 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
a9263751 | 3124 | wake_all_kswapds(order, ac); |
1da177e4 | 3125 | |
9bf2229f | 3126 | /* |
7fb1d9fc RS |
3127 | * OK, we're below the kswapd watermark and have kicked background |
3128 | * reclaim. Now things get more complex, so set up alloc_flags according | |
3129 | * to how we want to proceed. | |
9bf2229f | 3130 | */ |
341ce06f | 3131 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 3132 | |
f33261d7 DR |
3133 | /* |
3134 | * Find the true preferred zone if the allocation is unconstrained by | |
3135 | * cpusets. | |
3136 | */ | |
a9263751 | 3137 | if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) { |
d8846374 | 3138 | struct zoneref *preferred_zoneref; |
a9263751 VB |
3139 | preferred_zoneref = first_zones_zonelist(ac->zonelist, |
3140 | ac->high_zoneidx, NULL, &ac->preferred_zone); | |
3141 | ac->classzone_idx = zonelist_zone_idx(preferred_zoneref); | |
d8846374 | 3142 | } |
f33261d7 | 3143 | |
341ce06f | 3144 | /* This is the last chance, in general, before the goto nopage. */ |
a9263751 VB |
3145 | page = get_page_from_freelist(gfp_mask, order, |
3146 | alloc_flags & ~ALLOC_NO_WATERMARKS, ac); | |
7fb1d9fc RS |
3147 | if (page) |
3148 | goto got_pg; | |
1da177e4 | 3149 | |
11e33f6a | 3150 | /* Allocate without watermarks if the context allows */ |
341ce06f | 3151 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
3152 | /* |
3153 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
3154 | * the allocation is high priority and these type of | |
3155 | * allocations are system rather than user orientated | |
3156 | */ | |
a9263751 | 3157 | ac->zonelist = node_zonelist(numa_node_id(), gfp_mask); |
33d53103 MH |
3158 | page = get_page_from_freelist(gfp_mask, order, |
3159 | ALLOC_NO_WATERMARKS, ac); | |
3160 | if (page) | |
3161 | goto got_pg; | |
1da177e4 LT |
3162 | } |
3163 | ||
d0164adc MG |
3164 | /* Caller is not willing to reclaim, we can't balance anything */ |
3165 | if (!can_direct_reclaim) { | |
aed0a0e3 | 3166 | /* |
33d53103 MH |
3167 | * All existing users of the __GFP_NOFAIL are blockable, so warn |
3168 | * of any new users that actually allow this type of allocation | |
3169 | * to fail. | |
aed0a0e3 DR |
3170 | */ |
3171 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 3172 | goto nopage; |
aed0a0e3 | 3173 | } |
1da177e4 | 3174 | |
341ce06f | 3175 | /* Avoid recursion of direct reclaim */ |
33d53103 MH |
3176 | if (current->flags & PF_MEMALLOC) { |
3177 | /* | |
3178 | * __GFP_NOFAIL request from this context is rather bizarre | |
3179 | * because we cannot reclaim anything and only can loop waiting | |
3180 | * for somebody to do a work for us. | |
3181 | */ | |
3182 | if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { | |
3183 | cond_resched(); | |
3184 | goto retry; | |
3185 | } | |
341ce06f | 3186 | goto nopage; |
33d53103 | 3187 | } |
341ce06f | 3188 | |
6583bb64 DR |
3189 | /* Avoid allocations with no watermarks from looping endlessly */ |
3190 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
3191 | goto nopage; | |
3192 | ||
77f1fe6b MG |
3193 | /* |
3194 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
3195 | * attempts after direct reclaim are synchronous | |
3196 | */ | |
a9263751 VB |
3197 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
3198 | migration_mode, | |
3199 | &contended_compaction, | |
53853e2d | 3200 | &deferred_compaction); |
56de7263 MG |
3201 | if (page) |
3202 | goto got_pg; | |
75f30861 | 3203 | |
1f9efdef | 3204 | /* Checks for THP-specific high-order allocations */ |
d0164adc | 3205 | if (is_thp_gfp_mask(gfp_mask)) { |
1f9efdef VB |
3206 | /* |
3207 | * If compaction is deferred for high-order allocations, it is | |
3208 | * because sync compaction recently failed. If this is the case | |
3209 | * and the caller requested a THP allocation, we do not want | |
3210 | * to heavily disrupt the system, so we fail the allocation | |
3211 | * instead of entering direct reclaim. | |
3212 | */ | |
3213 | if (deferred_compaction) | |
3214 | goto nopage; | |
3215 | ||
3216 | /* | |
3217 | * In all zones where compaction was attempted (and not | |
3218 | * deferred or skipped), lock contention has been detected. | |
3219 | * For THP allocation we do not want to disrupt the others | |
3220 | * so we fallback to base pages instead. | |
3221 | */ | |
3222 | if (contended_compaction == COMPACT_CONTENDED_LOCK) | |
3223 | goto nopage; | |
3224 | ||
3225 | /* | |
3226 | * If compaction was aborted due to need_resched(), we do not | |
3227 | * want to further increase allocation latency, unless it is | |
3228 | * khugepaged trying to collapse. | |
3229 | */ | |
3230 | if (contended_compaction == COMPACT_CONTENDED_SCHED | |
3231 | && !(current->flags & PF_KTHREAD)) | |
3232 | goto nopage; | |
3233 | } | |
66199712 | 3234 | |
8fe78048 DR |
3235 | /* |
3236 | * It can become very expensive to allocate transparent hugepages at | |
3237 | * fault, so use asynchronous memory compaction for THP unless it is | |
3238 | * khugepaged trying to collapse. | |
3239 | */ | |
d0164adc | 3240 | if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD)) |
8fe78048 DR |
3241 | migration_mode = MIGRATE_SYNC_LIGHT; |
3242 | ||
11e33f6a | 3243 | /* Try direct reclaim and then allocating */ |
a9263751 VB |
3244 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, |
3245 | &did_some_progress); | |
11e33f6a MG |
3246 | if (page) |
3247 | goto got_pg; | |
1da177e4 | 3248 | |
9083905a JW |
3249 | /* Do not loop if specifically requested */ |
3250 | if (gfp_mask & __GFP_NORETRY) | |
3251 | goto noretry; | |
3252 | ||
3253 | /* Keep reclaiming pages as long as there is reasonable progress */ | |
a41f24ea | 3254 | pages_reclaimed += did_some_progress; |
9083905a JW |
3255 | if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) || |
3256 | ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) { | |
11e33f6a | 3257 | /* Wait for some write requests to complete then retry */ |
a9263751 | 3258 | wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50); |
9879de73 | 3259 | goto retry; |
1da177e4 LT |
3260 | } |
3261 | ||
9083905a JW |
3262 | /* Reclaim has failed us, start killing things */ |
3263 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); | |
3264 | if (page) | |
3265 | goto got_pg; | |
3266 | ||
3267 | /* Retry as long as the OOM killer is making progress */ | |
3268 | if (did_some_progress) | |
3269 | goto retry; | |
3270 | ||
3271 | noretry: | |
3272 | /* | |
3273 | * High-order allocations do not necessarily loop after | |
3274 | * direct reclaim and reclaim/compaction depends on compaction | |
3275 | * being called after reclaim so call directly if necessary | |
3276 | */ | |
3277 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, | |
3278 | ac, migration_mode, | |
3279 | &contended_compaction, | |
3280 | &deferred_compaction); | |
3281 | if (page) | |
3282 | goto got_pg; | |
1da177e4 | 3283 | nopage: |
a238ab5b | 3284 | warn_alloc_failed(gfp_mask, order, NULL); |
1da177e4 | 3285 | got_pg: |
072bb0aa | 3286 | return page; |
1da177e4 | 3287 | } |
11e33f6a MG |
3288 | |
3289 | /* | |
3290 | * This is the 'heart' of the zoned buddy allocator. | |
3291 | */ | |
3292 | struct page * | |
3293 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
3294 | struct zonelist *zonelist, nodemask_t *nodemask) | |
3295 | { | |
d8846374 | 3296 | struct zoneref *preferred_zoneref; |
cc9a6c87 | 3297 | struct page *page = NULL; |
cc9a6c87 | 3298 | unsigned int cpuset_mems_cookie; |
3a025760 | 3299 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR; |
91fbdc0f | 3300 | gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */ |
a9263751 VB |
3301 | struct alloc_context ac = { |
3302 | .high_zoneidx = gfp_zone(gfp_mask), | |
3303 | .nodemask = nodemask, | |
3304 | .migratetype = gfpflags_to_migratetype(gfp_mask), | |
3305 | }; | |
11e33f6a | 3306 | |
dcce284a BH |
3307 | gfp_mask &= gfp_allowed_mask; |
3308 | ||
11e33f6a MG |
3309 | lockdep_trace_alloc(gfp_mask); |
3310 | ||
d0164adc | 3311 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
11e33f6a MG |
3312 | |
3313 | if (should_fail_alloc_page(gfp_mask, order)) | |
3314 | return NULL; | |
3315 | ||
3316 | /* | |
3317 | * Check the zones suitable for the gfp_mask contain at least one | |
3318 | * valid zone. It's possible to have an empty zonelist as a result | |
4167e9b2 | 3319 | * of __GFP_THISNODE and a memoryless node |
11e33f6a MG |
3320 | */ |
3321 | if (unlikely(!zonelist->_zonerefs->zone)) | |
3322 | return NULL; | |
3323 | ||
a9263751 | 3324 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
21bb9bd1 VB |
3325 | alloc_flags |= ALLOC_CMA; |
3326 | ||
cc9a6c87 | 3327 | retry_cpuset: |
d26914d1 | 3328 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3329 | |
a9263751 VB |
3330 | /* We set it here, as __alloc_pages_slowpath might have changed it */ |
3331 | ac.zonelist = zonelist; | |
c9ab0c4f MG |
3332 | |
3333 | /* Dirty zone balancing only done in the fast path */ | |
3334 | ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE); | |
3335 | ||
5117f45d | 3336 | /* The preferred zone is used for statistics later */ |
a9263751 VB |
3337 | preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx, |
3338 | ac.nodemask ? : &cpuset_current_mems_allowed, | |
3339 | &ac.preferred_zone); | |
3340 | if (!ac.preferred_zone) | |
cc9a6c87 | 3341 | goto out; |
a9263751 | 3342 | ac.classzone_idx = zonelist_zone_idx(preferred_zoneref); |
5117f45d MG |
3343 | |
3344 | /* First allocation attempt */ | |
91fbdc0f | 3345 | alloc_mask = gfp_mask|__GFP_HARDWALL; |
a9263751 | 3346 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
21caf2fc ML |
3347 | if (unlikely(!page)) { |
3348 | /* | |
3349 | * Runtime PM, block IO and its error handling path | |
3350 | * can deadlock because I/O on the device might not | |
3351 | * complete. | |
3352 | */ | |
91fbdc0f | 3353 | alloc_mask = memalloc_noio_flags(gfp_mask); |
c9ab0c4f | 3354 | ac.spread_dirty_pages = false; |
91fbdc0f | 3355 | |
a9263751 | 3356 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
21caf2fc | 3357 | } |
11e33f6a | 3358 | |
23f086f9 XQ |
3359 | if (kmemcheck_enabled && page) |
3360 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
3361 | ||
a9263751 | 3362 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
cc9a6c87 MG |
3363 | |
3364 | out: | |
3365 | /* | |
3366 | * When updating a task's mems_allowed, it is possible to race with | |
3367 | * parallel threads in such a way that an allocation can fail while | |
3368 | * the mask is being updated. If a page allocation is about to fail, | |
3369 | * check if the cpuset changed during allocation and if so, retry. | |
3370 | */ | |
d26914d1 | 3371 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
3372 | goto retry_cpuset; |
3373 | ||
11e33f6a | 3374 | return page; |
1da177e4 | 3375 | } |
d239171e | 3376 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
3377 | |
3378 | /* | |
3379 | * Common helper functions. | |
3380 | */ | |
920c7a5d | 3381 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 3382 | { |
945a1113 AM |
3383 | struct page *page; |
3384 | ||
3385 | /* | |
3386 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
3387 | * a highmem page | |
3388 | */ | |
3389 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
3390 | ||
1da177e4 LT |
3391 | page = alloc_pages(gfp_mask, order); |
3392 | if (!page) | |
3393 | return 0; | |
3394 | return (unsigned long) page_address(page); | |
3395 | } | |
1da177e4 LT |
3396 | EXPORT_SYMBOL(__get_free_pages); |
3397 | ||
920c7a5d | 3398 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 3399 | { |
945a1113 | 3400 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 3401 | } |
1da177e4 LT |
3402 | EXPORT_SYMBOL(get_zeroed_page); |
3403 | ||
920c7a5d | 3404 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 3405 | { |
b5810039 | 3406 | if (put_page_testzero(page)) { |
1da177e4 | 3407 | if (order == 0) |
b745bc85 | 3408 | free_hot_cold_page(page, false); |
1da177e4 LT |
3409 | else |
3410 | __free_pages_ok(page, order); | |
3411 | } | |
3412 | } | |
3413 | ||
3414 | EXPORT_SYMBOL(__free_pages); | |
3415 | ||
920c7a5d | 3416 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
3417 | { |
3418 | if (addr != 0) { | |
725d704e | 3419 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
3420 | __free_pages(virt_to_page((void *)addr), order); |
3421 | } | |
3422 | } | |
3423 | ||
3424 | EXPORT_SYMBOL(free_pages); | |
3425 | ||
b63ae8ca AD |
3426 | /* |
3427 | * Page Fragment: | |
3428 | * An arbitrary-length arbitrary-offset area of memory which resides | |
3429 | * within a 0 or higher order page. Multiple fragments within that page | |
3430 | * are individually refcounted, in the page's reference counter. | |
3431 | * | |
3432 | * The page_frag functions below provide a simple allocation framework for | |
3433 | * page fragments. This is used by the network stack and network device | |
3434 | * drivers to provide a backing region of memory for use as either an | |
3435 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. | |
3436 | */ | |
3437 | static struct page *__page_frag_refill(struct page_frag_cache *nc, | |
3438 | gfp_t gfp_mask) | |
3439 | { | |
3440 | struct page *page = NULL; | |
3441 | gfp_t gfp = gfp_mask; | |
3442 | ||
3443 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3444 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | | |
3445 | __GFP_NOMEMALLOC; | |
3446 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, | |
3447 | PAGE_FRAG_CACHE_MAX_ORDER); | |
3448 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; | |
3449 | #endif | |
3450 | if (unlikely(!page)) | |
3451 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); | |
3452 | ||
3453 | nc->va = page ? page_address(page) : NULL; | |
3454 | ||
3455 | return page; | |
3456 | } | |
3457 | ||
3458 | void *__alloc_page_frag(struct page_frag_cache *nc, | |
3459 | unsigned int fragsz, gfp_t gfp_mask) | |
3460 | { | |
3461 | unsigned int size = PAGE_SIZE; | |
3462 | struct page *page; | |
3463 | int offset; | |
3464 | ||
3465 | if (unlikely(!nc->va)) { | |
3466 | refill: | |
3467 | page = __page_frag_refill(nc, gfp_mask); | |
3468 | if (!page) | |
3469 | return NULL; | |
3470 | ||
3471 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3472 | /* if size can vary use size else just use PAGE_SIZE */ | |
3473 | size = nc->size; | |
3474 | #endif | |
3475 | /* Even if we own the page, we do not use atomic_set(). | |
3476 | * This would break get_page_unless_zero() users. | |
3477 | */ | |
fe896d18 | 3478 | page_ref_add(page, size - 1); |
b63ae8ca AD |
3479 | |
3480 | /* reset page count bias and offset to start of new frag */ | |
2f064f34 | 3481 | nc->pfmemalloc = page_is_pfmemalloc(page); |
b63ae8ca AD |
3482 | nc->pagecnt_bias = size; |
3483 | nc->offset = size; | |
3484 | } | |
3485 | ||
3486 | offset = nc->offset - fragsz; | |
3487 | if (unlikely(offset < 0)) { | |
3488 | page = virt_to_page(nc->va); | |
3489 | ||
fe896d18 | 3490 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
b63ae8ca AD |
3491 | goto refill; |
3492 | ||
3493 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
3494 | /* if size can vary use size else just use PAGE_SIZE */ | |
3495 | size = nc->size; | |
3496 | #endif | |
3497 | /* OK, page count is 0, we can safely set it */ | |
fe896d18 | 3498 | set_page_count(page, size); |
b63ae8ca AD |
3499 | |
3500 | /* reset page count bias and offset to start of new frag */ | |
3501 | nc->pagecnt_bias = size; | |
3502 | offset = size - fragsz; | |
3503 | } | |
3504 | ||
3505 | nc->pagecnt_bias--; | |
3506 | nc->offset = offset; | |
3507 | ||
3508 | return nc->va + offset; | |
3509 | } | |
3510 | EXPORT_SYMBOL(__alloc_page_frag); | |
3511 | ||
3512 | /* | |
3513 | * Frees a page fragment allocated out of either a compound or order 0 page. | |
3514 | */ | |
3515 | void __free_page_frag(void *addr) | |
3516 | { | |
3517 | struct page *page = virt_to_head_page(addr); | |
3518 | ||
3519 | if (unlikely(put_page_testzero(page))) | |
3520 | __free_pages_ok(page, compound_order(page)); | |
3521 | } | |
3522 | EXPORT_SYMBOL(__free_page_frag); | |
3523 | ||
6a1a0d3b | 3524 | /* |
52383431 | 3525 | * alloc_kmem_pages charges newly allocated pages to the kmem resource counter |
a9bb7e62 VD |
3526 | * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is |
3527 | * equivalent to alloc_pages. | |
6a1a0d3b | 3528 | * |
52383431 VD |
3529 | * It should be used when the caller would like to use kmalloc, but since the |
3530 | * allocation is large, it has to fall back to the page allocator. | |
3531 | */ | |
3532 | struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order) | |
3533 | { | |
3534 | struct page *page; | |
52383431 | 3535 | |
52383431 | 3536 | page = alloc_pages(gfp_mask, order); |
d05e83a6 VD |
3537 | if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) { |
3538 | __free_pages(page, order); | |
3539 | page = NULL; | |
3540 | } | |
52383431 VD |
3541 | return page; |
3542 | } | |
3543 | ||
3544 | struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
3545 | { | |
3546 | struct page *page; | |
52383431 | 3547 | |
52383431 | 3548 | page = alloc_pages_node(nid, gfp_mask, order); |
d05e83a6 VD |
3549 | if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) { |
3550 | __free_pages(page, order); | |
3551 | page = NULL; | |
3552 | } | |
52383431 VD |
3553 | return page; |
3554 | } | |
3555 | ||
3556 | /* | |
3557 | * __free_kmem_pages and free_kmem_pages will free pages allocated with | |
3558 | * alloc_kmem_pages. | |
6a1a0d3b | 3559 | */ |
52383431 | 3560 | void __free_kmem_pages(struct page *page, unsigned int order) |
6a1a0d3b | 3561 | { |
d05e83a6 | 3562 | memcg_kmem_uncharge(page, order); |
6a1a0d3b GC |
3563 | __free_pages(page, order); |
3564 | } | |
3565 | ||
52383431 | 3566 | void free_kmem_pages(unsigned long addr, unsigned int order) |
6a1a0d3b GC |
3567 | { |
3568 | if (addr != 0) { | |
3569 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
52383431 | 3570 | __free_kmem_pages(virt_to_page((void *)addr), order); |
6a1a0d3b GC |
3571 | } |
3572 | } | |
3573 | ||
d00181b9 KS |
3574 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
3575 | size_t size) | |
ee85c2e1 AK |
3576 | { |
3577 | if (addr) { | |
3578 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
3579 | unsigned long used = addr + PAGE_ALIGN(size); | |
3580 | ||
3581 | split_page(virt_to_page((void *)addr), order); | |
3582 | while (used < alloc_end) { | |
3583 | free_page(used); | |
3584 | used += PAGE_SIZE; | |
3585 | } | |
3586 | } | |
3587 | return (void *)addr; | |
3588 | } | |
3589 | ||
2be0ffe2 TT |
3590 | /** |
3591 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
3592 | * @size: the number of bytes to allocate | |
3593 | * @gfp_mask: GFP flags for the allocation | |
3594 | * | |
3595 | * This function is similar to alloc_pages(), except that it allocates the | |
3596 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
3597 | * allocate memory in power-of-two pages. | |
3598 | * | |
3599 | * This function is also limited by MAX_ORDER. | |
3600 | * | |
3601 | * Memory allocated by this function must be released by free_pages_exact(). | |
3602 | */ | |
3603 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
3604 | { | |
3605 | unsigned int order = get_order(size); | |
3606 | unsigned long addr; | |
3607 | ||
3608 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 3609 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
3610 | } |
3611 | EXPORT_SYMBOL(alloc_pages_exact); | |
3612 | ||
ee85c2e1 AK |
3613 | /** |
3614 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
3615 | * pages on a node. | |
b5e6ab58 | 3616 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
3617 | * @size: the number of bytes to allocate |
3618 | * @gfp_mask: GFP flags for the allocation | |
3619 | * | |
3620 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
3621 | * back. | |
ee85c2e1 | 3622 | */ |
e1931811 | 3623 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 | 3624 | { |
d00181b9 | 3625 | unsigned int order = get_order(size); |
ee85c2e1 AK |
3626 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
3627 | if (!p) | |
3628 | return NULL; | |
3629 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
3630 | } | |
ee85c2e1 | 3631 | |
2be0ffe2 TT |
3632 | /** |
3633 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
3634 | * @virt: the value returned by alloc_pages_exact. | |
3635 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
3636 | * | |
3637 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
3638 | */ | |
3639 | void free_pages_exact(void *virt, size_t size) | |
3640 | { | |
3641 | unsigned long addr = (unsigned long)virt; | |
3642 | unsigned long end = addr + PAGE_ALIGN(size); | |
3643 | ||
3644 | while (addr < end) { | |
3645 | free_page(addr); | |
3646 | addr += PAGE_SIZE; | |
3647 | } | |
3648 | } | |
3649 | EXPORT_SYMBOL(free_pages_exact); | |
3650 | ||
e0fb5815 ZY |
3651 | /** |
3652 | * nr_free_zone_pages - count number of pages beyond high watermark | |
3653 | * @offset: The zone index of the highest zone | |
3654 | * | |
3655 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
3656 | * high watermark within all zones at or below a given zone index. For each | |
3657 | * zone, the number of pages is calculated as: | |
834405c3 | 3658 | * managed_pages - high_pages |
e0fb5815 | 3659 | */ |
ebec3862 | 3660 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 3661 | { |
dd1a239f | 3662 | struct zoneref *z; |
54a6eb5c MG |
3663 | struct zone *zone; |
3664 | ||
e310fd43 | 3665 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 3666 | unsigned long sum = 0; |
1da177e4 | 3667 | |
0e88460d | 3668 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 3669 | |
54a6eb5c | 3670 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 3671 | unsigned long size = zone->managed_pages; |
41858966 | 3672 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
3673 | if (size > high) |
3674 | sum += size - high; | |
1da177e4 LT |
3675 | } |
3676 | ||
3677 | return sum; | |
3678 | } | |
3679 | ||
e0fb5815 ZY |
3680 | /** |
3681 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
3682 | * | |
3683 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
3684 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 3685 | */ |
ebec3862 | 3686 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 3687 | { |
af4ca457 | 3688 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 3689 | } |
c2f1a551 | 3690 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 3691 | |
e0fb5815 ZY |
3692 | /** |
3693 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
3694 | * | |
3695 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
3696 | * high watermark within all zones. | |
1da177e4 | 3697 | */ |
ebec3862 | 3698 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 3699 | { |
2a1e274a | 3700 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 3701 | } |
08e0f6a9 CL |
3702 | |
3703 | static inline void show_node(struct zone *zone) | |
1da177e4 | 3704 | { |
e5adfffc | 3705 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 3706 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 3707 | } |
1da177e4 | 3708 | |
d02bd27b IR |
3709 | long si_mem_available(void) |
3710 | { | |
3711 | long available; | |
3712 | unsigned long pagecache; | |
3713 | unsigned long wmark_low = 0; | |
3714 | unsigned long pages[NR_LRU_LISTS]; | |
3715 | struct zone *zone; | |
3716 | int lru; | |
3717 | ||
3718 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) | |
3719 | pages[lru] = global_page_state(NR_LRU_BASE + lru); | |
3720 | ||
3721 | for_each_zone(zone) | |
3722 | wmark_low += zone->watermark[WMARK_LOW]; | |
3723 | ||
3724 | /* | |
3725 | * Estimate the amount of memory available for userspace allocations, | |
3726 | * without causing swapping. | |
3727 | */ | |
3728 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; | |
3729 | ||
3730 | /* | |
3731 | * Not all the page cache can be freed, otherwise the system will | |
3732 | * start swapping. Assume at least half of the page cache, or the | |
3733 | * low watermark worth of cache, needs to stay. | |
3734 | */ | |
3735 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; | |
3736 | pagecache -= min(pagecache / 2, wmark_low); | |
3737 | available += pagecache; | |
3738 | ||
3739 | /* | |
3740 | * Part of the reclaimable slab consists of items that are in use, | |
3741 | * and cannot be freed. Cap this estimate at the low watermark. | |
3742 | */ | |
3743 | available += global_page_state(NR_SLAB_RECLAIMABLE) - | |
3744 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); | |
3745 | ||
3746 | if (available < 0) | |
3747 | available = 0; | |
3748 | return available; | |
3749 | } | |
3750 | EXPORT_SYMBOL_GPL(si_mem_available); | |
3751 | ||
1da177e4 LT |
3752 | void si_meminfo(struct sysinfo *val) |
3753 | { | |
3754 | val->totalram = totalram_pages; | |
cc7452b6 | 3755 | val->sharedram = global_page_state(NR_SHMEM); |
d23ad423 | 3756 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 3757 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
3758 | val->totalhigh = totalhigh_pages; |
3759 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
3760 | val->mem_unit = PAGE_SIZE; |
3761 | } | |
3762 | ||
3763 | EXPORT_SYMBOL(si_meminfo); | |
3764 | ||
3765 | #ifdef CONFIG_NUMA | |
3766 | void si_meminfo_node(struct sysinfo *val, int nid) | |
3767 | { | |
cdd91a77 JL |
3768 | int zone_type; /* needs to be signed */ |
3769 | unsigned long managed_pages = 0; | |
1da177e4 LT |
3770 | pg_data_t *pgdat = NODE_DATA(nid); |
3771 | ||
cdd91a77 JL |
3772 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
3773 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
3774 | val->totalram = managed_pages; | |
cc7452b6 | 3775 | val->sharedram = node_page_state(nid, NR_SHMEM); |
d23ad423 | 3776 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 3777 | #ifdef CONFIG_HIGHMEM |
b40da049 | 3778 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages; |
d23ad423 CL |
3779 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
3780 | NR_FREE_PAGES); | |
98d2b0eb CL |
3781 | #else |
3782 | val->totalhigh = 0; | |
3783 | val->freehigh = 0; | |
3784 | #endif | |
1da177e4 LT |
3785 | val->mem_unit = PAGE_SIZE; |
3786 | } | |
3787 | #endif | |
3788 | ||
ddd588b5 | 3789 | /* |
7bf02ea2 DR |
3790 | * Determine whether the node should be displayed or not, depending on whether |
3791 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 3792 | */ |
7bf02ea2 | 3793 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
3794 | { |
3795 | bool ret = false; | |
cc9a6c87 | 3796 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
3797 | |
3798 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
3799 | goto out; | |
3800 | ||
cc9a6c87 | 3801 | do { |
d26914d1 | 3802 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3803 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 3804 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
3805 | out: |
3806 | return ret; | |
3807 | } | |
3808 | ||
1da177e4 LT |
3809 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
3810 | ||
377e4f16 RV |
3811 | static void show_migration_types(unsigned char type) |
3812 | { | |
3813 | static const char types[MIGRATE_TYPES] = { | |
3814 | [MIGRATE_UNMOVABLE] = 'U', | |
377e4f16 | 3815 | [MIGRATE_MOVABLE] = 'M', |
475a2f90 VB |
3816 | [MIGRATE_RECLAIMABLE] = 'E', |
3817 | [MIGRATE_HIGHATOMIC] = 'H', | |
377e4f16 RV |
3818 | #ifdef CONFIG_CMA |
3819 | [MIGRATE_CMA] = 'C', | |
3820 | #endif | |
194159fb | 3821 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 3822 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 3823 | #endif |
377e4f16 RV |
3824 | }; |
3825 | char tmp[MIGRATE_TYPES + 1]; | |
3826 | char *p = tmp; | |
3827 | int i; | |
3828 | ||
3829 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
3830 | if (type & (1 << i)) | |
3831 | *p++ = types[i]; | |
3832 | } | |
3833 | ||
3834 | *p = '\0'; | |
3835 | printk("(%s) ", tmp); | |
3836 | } | |
3837 | ||
1da177e4 LT |
3838 | /* |
3839 | * Show free area list (used inside shift_scroll-lock stuff) | |
3840 | * We also calculate the percentage fragmentation. We do this by counting the | |
3841 | * memory on each free list with the exception of the first item on the list. | |
d1bfcdb8 KK |
3842 | * |
3843 | * Bits in @filter: | |
3844 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's | |
3845 | * cpuset. | |
1da177e4 | 3846 | */ |
7bf02ea2 | 3847 | void show_free_areas(unsigned int filter) |
1da177e4 | 3848 | { |
d1bfcdb8 | 3849 | unsigned long free_pcp = 0; |
c7241913 | 3850 | int cpu; |
1da177e4 LT |
3851 | struct zone *zone; |
3852 | ||
ee99c71c | 3853 | for_each_populated_zone(zone) { |
7bf02ea2 | 3854 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3855 | continue; |
d1bfcdb8 | 3856 | |
761b0677 KK |
3857 | for_each_online_cpu(cpu) |
3858 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
1da177e4 LT |
3859 | } |
3860 | ||
a731286d KM |
3861 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
3862 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
d1bfcdb8 KK |
3863 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
3864 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
d1ce749a | 3865 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
d1bfcdb8 | 3866 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
4f98a2fe | 3867 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 3868 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
3869 | global_page_state(NR_ISOLATED_ANON), |
3870 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 3871 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 3872 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 3873 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 3874 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 3875 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 3876 | global_page_state(NR_UNSTABLE_NFS), |
3701b033 KM |
3877 | global_page_state(NR_SLAB_RECLAIMABLE), |
3878 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 3879 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 3880 | global_page_state(NR_SHMEM), |
a25700a5 | 3881 | global_page_state(NR_PAGETABLE), |
d1ce749a | 3882 | global_page_state(NR_BOUNCE), |
d1bfcdb8 KK |
3883 | global_page_state(NR_FREE_PAGES), |
3884 | free_pcp, | |
d1ce749a | 3885 | global_page_state(NR_FREE_CMA_PAGES)); |
1da177e4 | 3886 | |
ee99c71c | 3887 | for_each_populated_zone(zone) { |
1da177e4 LT |
3888 | int i; |
3889 | ||
7bf02ea2 | 3890 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3891 | continue; |
d1bfcdb8 KK |
3892 | |
3893 | free_pcp = 0; | |
3894 | for_each_online_cpu(cpu) | |
3895 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; | |
3896 | ||
1da177e4 LT |
3897 | show_node(zone); |
3898 | printk("%s" | |
3899 | " free:%lukB" | |
3900 | " min:%lukB" | |
3901 | " low:%lukB" | |
3902 | " high:%lukB" | |
4f98a2fe RR |
3903 | " active_anon:%lukB" |
3904 | " inactive_anon:%lukB" | |
3905 | " active_file:%lukB" | |
3906 | " inactive_file:%lukB" | |
7b854121 | 3907 | " unevictable:%lukB" |
a731286d KM |
3908 | " isolated(anon):%lukB" |
3909 | " isolated(file):%lukB" | |
1da177e4 | 3910 | " present:%lukB" |
9feedc9d | 3911 | " managed:%lukB" |
4a0aa73f KM |
3912 | " mlocked:%lukB" |
3913 | " dirty:%lukB" | |
3914 | " writeback:%lukB" | |
3915 | " mapped:%lukB" | |
4b02108a | 3916 | " shmem:%lukB" |
4a0aa73f KM |
3917 | " slab_reclaimable:%lukB" |
3918 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3919 | " kernel_stack:%lukB" |
4a0aa73f KM |
3920 | " pagetables:%lukB" |
3921 | " unstable:%lukB" | |
3922 | " bounce:%lukB" | |
d1bfcdb8 KK |
3923 | " free_pcp:%lukB" |
3924 | " local_pcp:%ukB" | |
d1ce749a | 3925 | " free_cma:%lukB" |
4a0aa73f | 3926 | " writeback_tmp:%lukB" |
1da177e4 LT |
3927 | " pages_scanned:%lu" |
3928 | " all_unreclaimable? %s" | |
3929 | "\n", | |
3930 | zone->name, | |
88f5acf8 | 3931 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3932 | K(min_wmark_pages(zone)), |
3933 | K(low_wmark_pages(zone)), | |
3934 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3935 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3936 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3937 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3938 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3939 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3940 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3941 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3942 | K(zone->present_pages), |
9feedc9d | 3943 | K(zone->managed_pages), |
4a0aa73f KM |
3944 | K(zone_page_state(zone, NR_MLOCK)), |
3945 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3946 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3947 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3948 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3949 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3950 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3951 | zone_page_state(zone, NR_KERNEL_STACK) * |
3952 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3953 | K(zone_page_state(zone, NR_PAGETABLE)), |
3954 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3955 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1bfcdb8 KK |
3956 | K(free_pcp), |
3957 | K(this_cpu_read(zone->pageset->pcp.count)), | |
d1ce749a | 3958 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3959 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
0d5d823a | 3960 | K(zone_page_state(zone, NR_PAGES_SCANNED)), |
6e543d57 | 3961 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
3962 | ); |
3963 | printk("lowmem_reserve[]:"); | |
3964 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3484b2de | 3965 | printk(" %ld", zone->lowmem_reserve[i]); |
1da177e4 LT |
3966 | printk("\n"); |
3967 | } | |
3968 | ||
ee99c71c | 3969 | for_each_populated_zone(zone) { |
d00181b9 KS |
3970 | unsigned int order; |
3971 | unsigned long nr[MAX_ORDER], flags, total = 0; | |
377e4f16 | 3972 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3973 | |
7bf02ea2 | 3974 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3975 | continue; |
1da177e4 LT |
3976 | show_node(zone); |
3977 | printk("%s: ", zone->name); | |
1da177e4 LT |
3978 | |
3979 | spin_lock_irqsave(&zone->lock, flags); | |
3980 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3981 | struct free_area *area = &zone->free_area[order]; |
3982 | int type; | |
3983 | ||
3984 | nr[order] = area->nr_free; | |
8f9de51a | 3985 | total += nr[order] << order; |
377e4f16 RV |
3986 | |
3987 | types[order] = 0; | |
3988 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3989 | if (!list_empty(&area->free_list[type])) | |
3990 | types[order] |= 1 << type; | |
3991 | } | |
1da177e4 LT |
3992 | } |
3993 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3994 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3995 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3996 | if (nr[order]) |
3997 | show_migration_types(types[order]); | |
3998 | } | |
1da177e4 LT |
3999 | printk("= %lukB\n", K(total)); |
4000 | } | |
4001 | ||
949f7ec5 DR |
4002 | hugetlb_show_meminfo(); |
4003 | ||
e6f3602d LW |
4004 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
4005 | ||
1da177e4 LT |
4006 | show_swap_cache_info(); |
4007 | } | |
4008 | ||
19770b32 MG |
4009 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4010 | { | |
4011 | zoneref->zone = zone; | |
4012 | zoneref->zone_idx = zone_idx(zone); | |
4013 | } | |
4014 | ||
1da177e4 LT |
4015 | /* |
4016 | * Builds allocation fallback zone lists. | |
1a93205b CL |
4017 | * |
4018 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 4019 | */ |
f0c0b2b8 | 4020 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 4021 | int nr_zones) |
1da177e4 | 4022 | { |
1a93205b | 4023 | struct zone *zone; |
bc732f1d | 4024 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
4025 | |
4026 | do { | |
2f6726e5 | 4027 | zone_type--; |
070f8032 | 4028 | zone = pgdat->node_zones + zone_type; |
1a93205b | 4029 | if (populated_zone(zone)) { |
dd1a239f MG |
4030 | zoneref_set_zone(zone, |
4031 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 4032 | check_highest_zone(zone_type); |
1da177e4 | 4033 | } |
2f6726e5 | 4034 | } while (zone_type); |
bc732f1d | 4035 | |
070f8032 | 4036 | return nr_zones; |
1da177e4 LT |
4037 | } |
4038 | ||
f0c0b2b8 KH |
4039 | |
4040 | /* | |
4041 | * zonelist_order: | |
4042 | * 0 = automatic detection of better ordering. | |
4043 | * 1 = order by ([node] distance, -zonetype) | |
4044 | * 2 = order by (-zonetype, [node] distance) | |
4045 | * | |
4046 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
4047 | * the same zonelist. So only NUMA can configure this param. | |
4048 | */ | |
4049 | #define ZONELIST_ORDER_DEFAULT 0 | |
4050 | #define ZONELIST_ORDER_NODE 1 | |
4051 | #define ZONELIST_ORDER_ZONE 2 | |
4052 | ||
4053 | /* zonelist order in the kernel. | |
4054 | * set_zonelist_order() will set this to NODE or ZONE. | |
4055 | */ | |
4056 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4057 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
4058 | ||
4059 | ||
1da177e4 | 4060 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
4061 | /* The value user specified ....changed by config */ |
4062 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4063 | /* string for sysctl */ | |
4064 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
4065 | char numa_zonelist_order[16] = "default"; | |
4066 | ||
4067 | /* | |
4068 | * interface for configure zonelist ordering. | |
4069 | * command line option "numa_zonelist_order" | |
4070 | * = "[dD]efault - default, automatic configuration. | |
4071 | * = "[nN]ode - order by node locality, then by zone within node | |
4072 | * = "[zZ]one - order by zone, then by locality within zone | |
4073 | */ | |
4074 | ||
4075 | static int __parse_numa_zonelist_order(char *s) | |
4076 | { | |
4077 | if (*s == 'd' || *s == 'D') { | |
4078 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
4079 | } else if (*s == 'n' || *s == 'N') { | |
4080 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
4081 | } else if (*s == 'z' || *s == 'Z') { | |
4082 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
4083 | } else { | |
4084 | printk(KERN_WARNING | |
4085 | "Ignoring invalid numa_zonelist_order value: " | |
4086 | "%s\n", s); | |
4087 | return -EINVAL; | |
4088 | } | |
4089 | return 0; | |
4090 | } | |
4091 | ||
4092 | static __init int setup_numa_zonelist_order(char *s) | |
4093 | { | |
ecb256f8 VL |
4094 | int ret; |
4095 | ||
4096 | if (!s) | |
4097 | return 0; | |
4098 | ||
4099 | ret = __parse_numa_zonelist_order(s); | |
4100 | if (ret == 0) | |
4101 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
4102 | ||
4103 | return ret; | |
f0c0b2b8 KH |
4104 | } |
4105 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
4106 | ||
4107 | /* | |
4108 | * sysctl handler for numa_zonelist_order | |
4109 | */ | |
cccad5b9 | 4110 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 4111 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
4112 | loff_t *ppos) |
4113 | { | |
4114 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
4115 | int ret; | |
443c6f14 | 4116 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 4117 | |
443c6f14 | 4118 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
4119 | if (write) { |
4120 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
4121 | ret = -EINVAL; | |
4122 | goto out; | |
4123 | } | |
4124 | strcpy(saved_string, (char *)table->data); | |
4125 | } | |
8d65af78 | 4126 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 4127 | if (ret) |
443c6f14 | 4128 | goto out; |
f0c0b2b8 KH |
4129 | if (write) { |
4130 | int oldval = user_zonelist_order; | |
dacbde09 CG |
4131 | |
4132 | ret = __parse_numa_zonelist_order((char *)table->data); | |
4133 | if (ret) { | |
f0c0b2b8 KH |
4134 | /* |
4135 | * bogus value. restore saved string | |
4136 | */ | |
dacbde09 | 4137 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
4138 | NUMA_ZONELIST_ORDER_LEN); |
4139 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
4140 | } else if (oldval != user_zonelist_order) { |
4141 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 4142 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
4143 | mutex_unlock(&zonelists_mutex); |
4144 | } | |
f0c0b2b8 | 4145 | } |
443c6f14 AK |
4146 | out: |
4147 | mutex_unlock(&zl_order_mutex); | |
4148 | return ret; | |
f0c0b2b8 KH |
4149 | } |
4150 | ||
4151 | ||
62bc62a8 | 4152 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
4153 | static int node_load[MAX_NUMNODES]; |
4154 | ||
1da177e4 | 4155 | /** |
4dc3b16b | 4156 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
4157 | * @node: node whose fallback list we're appending |
4158 | * @used_node_mask: nodemask_t of already used nodes | |
4159 | * | |
4160 | * We use a number of factors to determine which is the next node that should | |
4161 | * appear on a given node's fallback list. The node should not have appeared | |
4162 | * already in @node's fallback list, and it should be the next closest node | |
4163 | * according to the distance array (which contains arbitrary distance values | |
4164 | * from each node to each node in the system), and should also prefer nodes | |
4165 | * with no CPUs, since presumably they'll have very little allocation pressure | |
4166 | * on them otherwise. | |
4167 | * It returns -1 if no node is found. | |
4168 | */ | |
f0c0b2b8 | 4169 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 4170 | { |
4cf808eb | 4171 | int n, val; |
1da177e4 | 4172 | int min_val = INT_MAX; |
00ef2d2f | 4173 | int best_node = NUMA_NO_NODE; |
a70f7302 | 4174 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 4175 | |
4cf808eb LT |
4176 | /* Use the local node if we haven't already */ |
4177 | if (!node_isset(node, *used_node_mask)) { | |
4178 | node_set(node, *used_node_mask); | |
4179 | return node; | |
4180 | } | |
1da177e4 | 4181 | |
4b0ef1fe | 4182 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
4183 | |
4184 | /* Don't want a node to appear more than once */ | |
4185 | if (node_isset(n, *used_node_mask)) | |
4186 | continue; | |
4187 | ||
1da177e4 LT |
4188 | /* Use the distance array to find the distance */ |
4189 | val = node_distance(node, n); | |
4190 | ||
4cf808eb LT |
4191 | /* Penalize nodes under us ("prefer the next node") */ |
4192 | val += (n < node); | |
4193 | ||
1da177e4 | 4194 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
4195 | tmp = cpumask_of_node(n); |
4196 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
4197 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4198 | ||
4199 | /* Slight preference for less loaded node */ | |
4200 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
4201 | val += node_load[n]; | |
4202 | ||
4203 | if (val < min_val) { | |
4204 | min_val = val; | |
4205 | best_node = n; | |
4206 | } | |
4207 | } | |
4208 | ||
4209 | if (best_node >= 0) | |
4210 | node_set(best_node, *used_node_mask); | |
4211 | ||
4212 | return best_node; | |
4213 | } | |
4214 | ||
f0c0b2b8 KH |
4215 | |
4216 | /* | |
4217 | * Build zonelists ordered by node and zones within node. | |
4218 | * This results in maximum locality--normal zone overflows into local | |
4219 | * DMA zone, if any--but risks exhausting DMA zone. | |
4220 | */ | |
4221 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 4222 | { |
f0c0b2b8 | 4223 | int j; |
1da177e4 | 4224 | struct zonelist *zonelist; |
f0c0b2b8 | 4225 | |
54a6eb5c | 4226 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 4227 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 4228 | ; |
bc732f1d | 4229 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
4230 | zonelist->_zonerefs[j].zone = NULL; |
4231 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
4232 | } |
4233 | ||
523b9458 CL |
4234 | /* |
4235 | * Build gfp_thisnode zonelists | |
4236 | */ | |
4237 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
4238 | { | |
523b9458 CL |
4239 | int j; |
4240 | struct zonelist *zonelist; | |
4241 | ||
54a6eb5c | 4242 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 4243 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
4244 | zonelist->_zonerefs[j].zone = NULL; |
4245 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
4246 | } |
4247 | ||
f0c0b2b8 KH |
4248 | /* |
4249 | * Build zonelists ordered by zone and nodes within zones. | |
4250 | * This results in conserving DMA zone[s] until all Normal memory is | |
4251 | * exhausted, but results in overflowing to remote node while memory | |
4252 | * may still exist in local DMA zone. | |
4253 | */ | |
4254 | static int node_order[MAX_NUMNODES]; | |
4255 | ||
4256 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
4257 | { | |
f0c0b2b8 KH |
4258 | int pos, j, node; |
4259 | int zone_type; /* needs to be signed */ | |
4260 | struct zone *z; | |
4261 | struct zonelist *zonelist; | |
4262 | ||
54a6eb5c MG |
4263 | zonelist = &pgdat->node_zonelists[0]; |
4264 | pos = 0; | |
4265 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
4266 | for (j = 0; j < nr_nodes; j++) { | |
4267 | node = node_order[j]; | |
4268 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
4269 | if (populated_zone(z)) { | |
dd1a239f MG |
4270 | zoneref_set_zone(z, |
4271 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 4272 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
4273 | } |
4274 | } | |
f0c0b2b8 | 4275 | } |
dd1a239f MG |
4276 | zonelist->_zonerefs[pos].zone = NULL; |
4277 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
4278 | } |
4279 | ||
3193913c MG |
4280 | #if defined(CONFIG_64BIT) |
4281 | /* | |
4282 | * Devices that require DMA32/DMA are relatively rare and do not justify a | |
4283 | * penalty to every machine in case the specialised case applies. Default | |
4284 | * to Node-ordering on 64-bit NUMA machines | |
4285 | */ | |
4286 | static int default_zonelist_order(void) | |
4287 | { | |
4288 | return ZONELIST_ORDER_NODE; | |
4289 | } | |
4290 | #else | |
4291 | /* | |
4292 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
4293 | * by the kernel. If processes running on node 0 deplete the low memory zone | |
4294 | * then reclaim will occur more frequency increasing stalls and potentially | |
4295 | * be easier to OOM if a large percentage of the zone is under writeback or | |
4296 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. | |
4297 | * Hence, default to zone ordering on 32-bit. | |
4298 | */ | |
f0c0b2b8 KH |
4299 | static int default_zonelist_order(void) |
4300 | { | |
f0c0b2b8 KH |
4301 | return ZONELIST_ORDER_ZONE; |
4302 | } | |
3193913c | 4303 | #endif /* CONFIG_64BIT */ |
f0c0b2b8 KH |
4304 | |
4305 | static void set_zonelist_order(void) | |
4306 | { | |
4307 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
4308 | current_zonelist_order = default_zonelist_order(); | |
4309 | else | |
4310 | current_zonelist_order = user_zonelist_order; | |
4311 | } | |
4312 | ||
4313 | static void build_zonelists(pg_data_t *pgdat) | |
4314 | { | |
c00eb15a | 4315 | int i, node, load; |
1da177e4 | 4316 | nodemask_t used_mask; |
f0c0b2b8 KH |
4317 | int local_node, prev_node; |
4318 | struct zonelist *zonelist; | |
d00181b9 | 4319 | unsigned int order = current_zonelist_order; |
1da177e4 LT |
4320 | |
4321 | /* initialize zonelists */ | |
523b9458 | 4322 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 4323 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
4324 | zonelist->_zonerefs[0].zone = NULL; |
4325 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
4326 | } |
4327 | ||
4328 | /* NUMA-aware ordering of nodes */ | |
4329 | local_node = pgdat->node_id; | |
62bc62a8 | 4330 | load = nr_online_nodes; |
1da177e4 LT |
4331 | prev_node = local_node; |
4332 | nodes_clear(used_mask); | |
f0c0b2b8 | 4333 | |
f0c0b2b8 | 4334 | memset(node_order, 0, sizeof(node_order)); |
c00eb15a | 4335 | i = 0; |
f0c0b2b8 | 4336 | |
1da177e4 LT |
4337 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
4338 | /* | |
4339 | * We don't want to pressure a particular node. | |
4340 | * So adding penalty to the first node in same | |
4341 | * distance group to make it round-robin. | |
4342 | */ | |
957f822a DR |
4343 | if (node_distance(local_node, node) != |
4344 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
4345 | node_load[node] = load; |
4346 | ||
1da177e4 LT |
4347 | prev_node = node; |
4348 | load--; | |
f0c0b2b8 KH |
4349 | if (order == ZONELIST_ORDER_NODE) |
4350 | build_zonelists_in_node_order(pgdat, node); | |
4351 | else | |
c00eb15a | 4352 | node_order[i++] = node; /* remember order */ |
f0c0b2b8 | 4353 | } |
1da177e4 | 4354 | |
f0c0b2b8 KH |
4355 | if (order == ZONELIST_ORDER_ZONE) { |
4356 | /* calculate node order -- i.e., DMA last! */ | |
c00eb15a | 4357 | build_zonelists_in_zone_order(pgdat, i); |
1da177e4 | 4358 | } |
523b9458 CL |
4359 | |
4360 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
4361 | } |
4362 | ||
7aac7898 LS |
4363 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4364 | /* | |
4365 | * Return node id of node used for "local" allocations. | |
4366 | * I.e., first node id of first zone in arg node's generic zonelist. | |
4367 | * Used for initializing percpu 'numa_mem', which is used primarily | |
4368 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
4369 | */ | |
4370 | int local_memory_node(int node) | |
4371 | { | |
4372 | struct zone *zone; | |
4373 | ||
4374 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
4375 | gfp_zone(GFP_KERNEL), | |
4376 | NULL, | |
4377 | &zone); | |
4378 | return zone->node; | |
4379 | } | |
4380 | #endif | |
f0c0b2b8 | 4381 | |
1da177e4 LT |
4382 | #else /* CONFIG_NUMA */ |
4383 | ||
f0c0b2b8 KH |
4384 | static void set_zonelist_order(void) |
4385 | { | |
4386 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
4387 | } | |
4388 | ||
4389 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 4390 | { |
19655d34 | 4391 | int node, local_node; |
54a6eb5c MG |
4392 | enum zone_type j; |
4393 | struct zonelist *zonelist; | |
1da177e4 LT |
4394 | |
4395 | local_node = pgdat->node_id; | |
1da177e4 | 4396 | |
54a6eb5c | 4397 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 4398 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 4399 | |
54a6eb5c MG |
4400 | /* |
4401 | * Now we build the zonelist so that it contains the zones | |
4402 | * of all the other nodes. | |
4403 | * We don't want to pressure a particular node, so when | |
4404 | * building the zones for node N, we make sure that the | |
4405 | * zones coming right after the local ones are those from | |
4406 | * node N+1 (modulo N) | |
4407 | */ | |
4408 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
4409 | if (!node_online(node)) | |
4410 | continue; | |
bc732f1d | 4411 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 4412 | } |
54a6eb5c MG |
4413 | for (node = 0; node < local_node; node++) { |
4414 | if (!node_online(node)) | |
4415 | continue; | |
bc732f1d | 4416 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
4417 | } |
4418 | ||
dd1a239f MG |
4419 | zonelist->_zonerefs[j].zone = NULL; |
4420 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
4421 | } |
4422 | ||
4423 | #endif /* CONFIG_NUMA */ | |
4424 | ||
99dcc3e5 CL |
4425 | /* |
4426 | * Boot pageset table. One per cpu which is going to be used for all | |
4427 | * zones and all nodes. The parameters will be set in such a way | |
4428 | * that an item put on a list will immediately be handed over to | |
4429 | * the buddy list. This is safe since pageset manipulation is done | |
4430 | * with interrupts disabled. | |
4431 | * | |
4432 | * The boot_pagesets must be kept even after bootup is complete for | |
4433 | * unused processors and/or zones. They do play a role for bootstrapping | |
4434 | * hotplugged processors. | |
4435 | * | |
4436 | * zoneinfo_show() and maybe other functions do | |
4437 | * not check if the processor is online before following the pageset pointer. | |
4438 | * Other parts of the kernel may not check if the zone is available. | |
4439 | */ | |
4440 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
4441 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 4442 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 4443 | |
4eaf3f64 HL |
4444 | /* |
4445 | * Global mutex to protect against size modification of zonelists | |
4446 | * as well as to serialize pageset setup for the new populated zone. | |
4447 | */ | |
4448 | DEFINE_MUTEX(zonelists_mutex); | |
4449 | ||
9b1a4d38 | 4450 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 4451 | static int __build_all_zonelists(void *data) |
1da177e4 | 4452 | { |
6811378e | 4453 | int nid; |
99dcc3e5 | 4454 | int cpu; |
9adb62a5 | 4455 | pg_data_t *self = data; |
9276b1bc | 4456 | |
7f9cfb31 BL |
4457 | #ifdef CONFIG_NUMA |
4458 | memset(node_load, 0, sizeof(node_load)); | |
4459 | #endif | |
9adb62a5 JL |
4460 | |
4461 | if (self && !node_online(self->node_id)) { | |
4462 | build_zonelists(self); | |
9adb62a5 JL |
4463 | } |
4464 | ||
9276b1bc | 4465 | for_each_online_node(nid) { |
7ea1530a CL |
4466 | pg_data_t *pgdat = NODE_DATA(nid); |
4467 | ||
4468 | build_zonelists(pgdat); | |
9276b1bc | 4469 | } |
99dcc3e5 CL |
4470 | |
4471 | /* | |
4472 | * Initialize the boot_pagesets that are going to be used | |
4473 | * for bootstrapping processors. The real pagesets for | |
4474 | * each zone will be allocated later when the per cpu | |
4475 | * allocator is available. | |
4476 | * | |
4477 | * boot_pagesets are used also for bootstrapping offline | |
4478 | * cpus if the system is already booted because the pagesets | |
4479 | * are needed to initialize allocators on a specific cpu too. | |
4480 | * F.e. the percpu allocator needs the page allocator which | |
4481 | * needs the percpu allocator in order to allocate its pagesets | |
4482 | * (a chicken-egg dilemma). | |
4483 | */ | |
7aac7898 | 4484 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
4485 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
4486 | ||
7aac7898 LS |
4487 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
4488 | /* | |
4489 | * We now know the "local memory node" for each node-- | |
4490 | * i.e., the node of the first zone in the generic zonelist. | |
4491 | * Set up numa_mem percpu variable for on-line cpus. During | |
4492 | * boot, only the boot cpu should be on-line; we'll init the | |
4493 | * secondary cpus' numa_mem as they come on-line. During | |
4494 | * node/memory hotplug, we'll fixup all on-line cpus. | |
4495 | */ | |
4496 | if (cpu_online(cpu)) | |
4497 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
4498 | #endif | |
4499 | } | |
4500 | ||
6811378e YG |
4501 | return 0; |
4502 | } | |
4503 | ||
061f67bc RV |
4504 | static noinline void __init |
4505 | build_all_zonelists_init(void) | |
4506 | { | |
4507 | __build_all_zonelists(NULL); | |
4508 | mminit_verify_zonelist(); | |
4509 | cpuset_init_current_mems_allowed(); | |
4510 | } | |
4511 | ||
4eaf3f64 HL |
4512 | /* |
4513 | * Called with zonelists_mutex held always | |
4514 | * unless system_state == SYSTEM_BOOTING. | |
061f67bc RV |
4515 | * |
4516 | * __ref due to (1) call of __meminit annotated setup_zone_pageset | |
4517 | * [we're only called with non-NULL zone through __meminit paths] and | |
4518 | * (2) call of __init annotated helper build_all_zonelists_init | |
4519 | * [protected by SYSTEM_BOOTING]. | |
4eaf3f64 | 4520 | */ |
9adb62a5 | 4521 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 4522 | { |
f0c0b2b8 KH |
4523 | set_zonelist_order(); |
4524 | ||
6811378e | 4525 | if (system_state == SYSTEM_BOOTING) { |
061f67bc | 4526 | build_all_zonelists_init(); |
6811378e | 4527 | } else { |
e9959f0f | 4528 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
4529 | if (zone) |
4530 | setup_zone_pageset(zone); | |
e9959f0f | 4531 | #endif |
dd1895e2 CS |
4532 | /* we have to stop all cpus to guarantee there is no user |
4533 | of zonelist */ | |
9adb62a5 | 4534 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
4535 | /* cpuset refresh routine should be here */ |
4536 | } | |
bd1e22b8 | 4537 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
4538 | /* |
4539 | * Disable grouping by mobility if the number of pages in the | |
4540 | * system is too low to allow the mechanism to work. It would be | |
4541 | * more accurate, but expensive to check per-zone. This check is | |
4542 | * made on memory-hotadd so a system can start with mobility | |
4543 | * disabled and enable it later | |
4544 | */ | |
d9c23400 | 4545 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
4546 | page_group_by_mobility_disabled = 1; |
4547 | else | |
4548 | page_group_by_mobility_disabled = 0; | |
4549 | ||
f88dfff5 | 4550 | pr_info("Built %i zonelists in %s order, mobility grouping %s. " |
9ef9acb0 | 4551 | "Total pages: %ld\n", |
62bc62a8 | 4552 | nr_online_nodes, |
f0c0b2b8 | 4553 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 4554 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
4555 | vm_total_pages); |
4556 | #ifdef CONFIG_NUMA | |
f88dfff5 | 4557 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
f0c0b2b8 | 4558 | #endif |
1da177e4 LT |
4559 | } |
4560 | ||
4561 | /* | |
4562 | * Helper functions to size the waitqueue hash table. | |
4563 | * Essentially these want to choose hash table sizes sufficiently | |
4564 | * large so that collisions trying to wait on pages are rare. | |
4565 | * But in fact, the number of active page waitqueues on typical | |
4566 | * systems is ridiculously low, less than 200. So this is even | |
4567 | * conservative, even though it seems large. | |
4568 | * | |
4569 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
4570 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
4571 | */ | |
4572 | #define PAGES_PER_WAITQUEUE 256 | |
4573 | ||
cca448fe | 4574 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 4575 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
4576 | { |
4577 | unsigned long size = 1; | |
4578 | ||
4579 | pages /= PAGES_PER_WAITQUEUE; | |
4580 | ||
4581 | while (size < pages) | |
4582 | size <<= 1; | |
4583 | ||
4584 | /* | |
4585 | * Once we have dozens or even hundreds of threads sleeping | |
4586 | * on IO we've got bigger problems than wait queue collision. | |
4587 | * Limit the size of the wait table to a reasonable size. | |
4588 | */ | |
4589 | size = min(size, 4096UL); | |
4590 | ||
4591 | return max(size, 4UL); | |
4592 | } | |
cca448fe YG |
4593 | #else |
4594 | /* | |
4595 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
4596 | * a suitable size for its wait_table. So we use the maximum size now. | |
4597 | * | |
4598 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
4599 | * | |
4600 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
4601 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
4602 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
4603 | * | |
4604 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
4605 | * or more by the traditional way. (See above). It equals: | |
4606 | * | |
4607 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
4608 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
4609 | * powerpc (64K page size) : = (32G +16M)byte. | |
4610 | */ | |
4611 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
4612 | { | |
4613 | return 4096UL; | |
4614 | } | |
4615 | #endif | |
1da177e4 LT |
4616 | |
4617 | /* | |
4618 | * This is an integer logarithm so that shifts can be used later | |
4619 | * to extract the more random high bits from the multiplicative | |
4620 | * hash function before the remainder is taken. | |
4621 | */ | |
4622 | static inline unsigned long wait_table_bits(unsigned long size) | |
4623 | { | |
4624 | return ffz(~size); | |
4625 | } | |
4626 | ||
1da177e4 LT |
4627 | /* |
4628 | * Initially all pages are reserved - free ones are freed | |
4629 | * up by free_all_bootmem() once the early boot process is | |
4630 | * done. Non-atomic initialization, single-pass. | |
4631 | */ | |
c09b4240 | 4632 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 4633 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 4634 | { |
4b94ffdc | 4635 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
29751f69 | 4636 | unsigned long end_pfn = start_pfn + size; |
4b94ffdc | 4637 | pg_data_t *pgdat = NODE_DATA(nid); |
29751f69 | 4638 | unsigned long pfn; |
3a80a7fa | 4639 | unsigned long nr_initialised = 0; |
342332e6 TI |
4640 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
4641 | struct memblock_region *r = NULL, *tmp; | |
4642 | #endif | |
1da177e4 | 4643 | |
22b31eec HD |
4644 | if (highest_memmap_pfn < end_pfn - 1) |
4645 | highest_memmap_pfn = end_pfn - 1; | |
4646 | ||
4b94ffdc DW |
4647 | /* |
4648 | * Honor reservation requested by the driver for this ZONE_DEVICE | |
4649 | * memory | |
4650 | */ | |
4651 | if (altmap && start_pfn == altmap->base_pfn) | |
4652 | start_pfn += altmap->reserve; | |
4653 | ||
cbe8dd4a | 4654 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 | 4655 | /* |
b72d0ffb AM |
4656 | * There can be holes in boot-time mem_map[]s handed to this |
4657 | * function. They do not exist on hotplugged memory. | |
a2f3aa02 | 4658 | */ |
b72d0ffb AM |
4659 | if (context != MEMMAP_EARLY) |
4660 | goto not_early; | |
4661 | ||
4662 | if (!early_pfn_valid(pfn)) | |
4663 | continue; | |
4664 | if (!early_pfn_in_nid(pfn, nid)) | |
4665 | continue; | |
4666 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) | |
4667 | break; | |
342332e6 TI |
4668 | |
4669 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
b72d0ffb AM |
4670 | /* |
4671 | * If not mirrored_kernelcore and ZONE_MOVABLE exists, range | |
4672 | * from zone_movable_pfn[nid] to end of each node should be | |
4673 | * ZONE_MOVABLE not ZONE_NORMAL. skip it. | |
4674 | */ | |
4675 | if (!mirrored_kernelcore && zone_movable_pfn[nid]) | |
4676 | if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid]) | |
4677 | continue; | |
342332e6 | 4678 | |
b72d0ffb AM |
4679 | /* |
4680 | * Check given memblock attribute by firmware which can affect | |
4681 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is | |
4682 | * mirrored, it's an overlapped memmap init. skip it. | |
4683 | */ | |
4684 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { | |
4685 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { | |
4686 | for_each_memblock(memory, tmp) | |
4687 | if (pfn < memblock_region_memory_end_pfn(tmp)) | |
4688 | break; | |
4689 | r = tmp; | |
4690 | } | |
4691 | if (pfn >= memblock_region_memory_base_pfn(r) && | |
4692 | memblock_is_mirror(r)) { | |
4693 | /* already initialized as NORMAL */ | |
4694 | pfn = memblock_region_memory_end_pfn(r); | |
4695 | continue; | |
342332e6 | 4696 | } |
a2f3aa02 | 4697 | } |
b72d0ffb | 4698 | #endif |
ac5d2539 | 4699 | |
b72d0ffb | 4700 | not_early: |
ac5d2539 MG |
4701 | /* |
4702 | * Mark the block movable so that blocks are reserved for | |
4703 | * movable at startup. This will force kernel allocations | |
4704 | * to reserve their blocks rather than leaking throughout | |
4705 | * the address space during boot when many long-lived | |
974a786e | 4706 | * kernel allocations are made. |
ac5d2539 MG |
4707 | * |
4708 | * bitmap is created for zone's valid pfn range. but memmap | |
4709 | * can be created for invalid pages (for alignment) | |
4710 | * check here not to call set_pageblock_migratetype() against | |
4711 | * pfn out of zone. | |
4712 | */ | |
4713 | if (!(pfn & (pageblock_nr_pages - 1))) { | |
4714 | struct page *page = pfn_to_page(pfn); | |
4715 | ||
4716 | __init_single_page(page, pfn, zone, nid); | |
4717 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4718 | } else { | |
4719 | __init_single_pfn(pfn, zone, nid); | |
4720 | } | |
1da177e4 LT |
4721 | } |
4722 | } | |
4723 | ||
1e548deb | 4724 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 4725 | { |
7aeb09f9 | 4726 | unsigned int order, t; |
b2a0ac88 MG |
4727 | for_each_migratetype_order(order, t) { |
4728 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
4729 | zone->free_area[order].nr_free = 0; |
4730 | } | |
4731 | } | |
4732 | ||
4733 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
4734 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 4735 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
4736 | #endif |
4737 | ||
7cd2b0a3 | 4738 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 4739 | { |
3a6be87f | 4740 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
4741 | int batch; |
4742 | ||
4743 | /* | |
4744 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 4745 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
4746 | * |
4747 | * OK, so we don't know how big the cache is. So guess. | |
4748 | */ | |
b40da049 | 4749 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
4750 | if (batch * PAGE_SIZE > 512 * 1024) |
4751 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
4752 | batch /= 4; /* We effectively *= 4 below */ |
4753 | if (batch < 1) | |
4754 | batch = 1; | |
4755 | ||
4756 | /* | |
0ceaacc9 NP |
4757 | * Clamp the batch to a 2^n - 1 value. Having a power |
4758 | * of 2 value was found to be more likely to have | |
4759 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 4760 | * |
0ceaacc9 NP |
4761 | * For example if 2 tasks are alternately allocating |
4762 | * batches of pages, one task can end up with a lot | |
4763 | * of pages of one half of the possible page colors | |
4764 | * and the other with pages of the other colors. | |
e7c8d5c9 | 4765 | */ |
9155203a | 4766 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 4767 | |
e7c8d5c9 | 4768 | return batch; |
3a6be87f DH |
4769 | |
4770 | #else | |
4771 | /* The deferral and batching of frees should be suppressed under NOMMU | |
4772 | * conditions. | |
4773 | * | |
4774 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4775 | * of contiguous memory as there's no hardware page translation to | |
4776 | * assemble apparent contiguous memory from discontiguous pages. | |
4777 | * | |
4778 | * Queueing large contiguous runs of pages for batching, however, | |
4779 | * causes the pages to actually be freed in smaller chunks. As there | |
4780 | * can be a significant delay between the individual batches being | |
4781 | * recycled, this leads to the once large chunks of space being | |
4782 | * fragmented and becoming unavailable for high-order allocations. | |
4783 | */ | |
4784 | return 0; | |
4785 | #endif | |
e7c8d5c9 CL |
4786 | } |
4787 | ||
8d7a8fa9 CS |
4788 | /* |
4789 | * pcp->high and pcp->batch values are related and dependent on one another: | |
4790 | * ->batch must never be higher then ->high. | |
4791 | * The following function updates them in a safe manner without read side | |
4792 | * locking. | |
4793 | * | |
4794 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
4795 | * those fields changing asynchronously (acording the the above rule). | |
4796 | * | |
4797 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
4798 | * outside of boot time (or some other assurance that no concurrent updaters | |
4799 | * exist). | |
4800 | */ | |
4801 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
4802 | unsigned long batch) | |
4803 | { | |
4804 | /* start with a fail safe value for batch */ | |
4805 | pcp->batch = 1; | |
4806 | smp_wmb(); | |
4807 | ||
4808 | /* Update high, then batch, in order */ | |
4809 | pcp->high = high; | |
4810 | smp_wmb(); | |
4811 | ||
4812 | pcp->batch = batch; | |
4813 | } | |
4814 | ||
3664033c | 4815 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
4816 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
4817 | { | |
8d7a8fa9 | 4818 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
4819 | } |
4820 | ||
88c90dbc | 4821 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
4822 | { |
4823 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4824 | int migratetype; |
2caaad41 | 4825 | |
1c6fe946 MD |
4826 | memset(p, 0, sizeof(*p)); |
4827 | ||
3dfa5721 | 4828 | pcp = &p->pcp; |
2caaad41 | 4829 | pcp->count = 0; |
5f8dcc21 MG |
4830 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4831 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4832 | } |
4833 | ||
88c90dbc CS |
4834 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
4835 | { | |
4836 | pageset_init(p); | |
4837 | pageset_set_batch(p, batch); | |
4838 | } | |
4839 | ||
8ad4b1fb | 4840 | /* |
3664033c | 4841 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
4842 | * to the value high for the pageset p. |
4843 | */ | |
3664033c | 4844 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
4845 | unsigned long high) |
4846 | { | |
8d7a8fa9 CS |
4847 | unsigned long batch = max(1UL, high / 4); |
4848 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
4849 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 4850 | |
8d7a8fa9 | 4851 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
4852 | } |
4853 | ||
7cd2b0a3 DR |
4854 | static void pageset_set_high_and_batch(struct zone *zone, |
4855 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 4856 | { |
56cef2b8 | 4857 | if (percpu_pagelist_fraction) |
3664033c | 4858 | pageset_set_high(pcp, |
56cef2b8 CS |
4859 | (zone->managed_pages / |
4860 | percpu_pagelist_fraction)); | |
4861 | else | |
4862 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
4863 | } | |
4864 | ||
169f6c19 CS |
4865 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
4866 | { | |
4867 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4868 | ||
4869 | pageset_init(pcp); | |
4870 | pageset_set_high_and_batch(zone, pcp); | |
4871 | } | |
4872 | ||
4ed7e022 | 4873 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4874 | { |
4875 | int cpu; | |
319774e2 | 4876 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
4877 | for_each_possible_cpu(cpu) |
4878 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
4879 | } |
4880 | ||
2caaad41 | 4881 | /* |
99dcc3e5 CL |
4882 | * Allocate per cpu pagesets and initialize them. |
4883 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4884 | */ |
99dcc3e5 | 4885 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4886 | { |
99dcc3e5 | 4887 | struct zone *zone; |
e7c8d5c9 | 4888 | |
319774e2 WF |
4889 | for_each_populated_zone(zone) |
4890 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4891 | } |
4892 | ||
577a32f6 | 4893 | static noinline __init_refok |
cca448fe | 4894 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4895 | { |
4896 | int i; | |
cca448fe | 4897 | size_t alloc_size; |
ed8ece2e DH |
4898 | |
4899 | /* | |
4900 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4901 | * per zone. | |
4902 | */ | |
02b694de YG |
4903 | zone->wait_table_hash_nr_entries = |
4904 | wait_table_hash_nr_entries(zone_size_pages); | |
4905 | zone->wait_table_bits = | |
4906 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4907 | alloc_size = zone->wait_table_hash_nr_entries |
4908 | * sizeof(wait_queue_head_t); | |
4909 | ||
cd94b9db | 4910 | if (!slab_is_available()) { |
cca448fe | 4911 | zone->wait_table = (wait_queue_head_t *) |
6782832e SS |
4912 | memblock_virt_alloc_node_nopanic( |
4913 | alloc_size, zone->zone_pgdat->node_id); | |
cca448fe YG |
4914 | } else { |
4915 | /* | |
4916 | * This case means that a zone whose size was 0 gets new memory | |
4917 | * via memory hot-add. | |
4918 | * But it may be the case that a new node was hot-added. In | |
4919 | * this case vmalloc() will not be able to use this new node's | |
4920 | * memory - this wait_table must be initialized to use this new | |
4921 | * node itself as well. | |
4922 | * To use this new node's memory, further consideration will be | |
4923 | * necessary. | |
4924 | */ | |
8691f3a7 | 4925 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4926 | } |
4927 | if (!zone->wait_table) | |
4928 | return -ENOMEM; | |
ed8ece2e | 4929 | |
b8af2941 | 4930 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4931 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4932 | |
4933 | return 0; | |
ed8ece2e DH |
4934 | } |
4935 | ||
c09b4240 | 4936 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4937 | { |
99dcc3e5 CL |
4938 | /* |
4939 | * per cpu subsystem is not up at this point. The following code | |
4940 | * relies on the ability of the linker to provide the | |
4941 | * offset of a (static) per cpu variable into the per cpu area. | |
4942 | */ | |
4943 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4944 | |
b38a8725 | 4945 | if (populated_zone(zone)) |
99dcc3e5 CL |
4946 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4947 | zone->name, zone->present_pages, | |
4948 | zone_batchsize(zone)); | |
ed8ece2e DH |
4949 | } |
4950 | ||
4ed7e022 | 4951 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4952 | unsigned long zone_start_pfn, |
b171e409 | 4953 | unsigned long size) |
ed8ece2e DH |
4954 | { |
4955 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4956 | int ret; |
4957 | ret = zone_wait_table_init(zone, size); | |
4958 | if (ret) | |
4959 | return ret; | |
ed8ece2e DH |
4960 | pgdat->nr_zones = zone_idx(zone) + 1; |
4961 | ||
ed8ece2e DH |
4962 | zone->zone_start_pfn = zone_start_pfn; |
4963 | ||
708614e6 MG |
4964 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4965 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4966 | pgdat->node_id, | |
4967 | (unsigned long)zone_idx(zone), | |
4968 | zone_start_pfn, (zone_start_pfn + size)); | |
4969 | ||
1e548deb | 4970 | zone_init_free_lists(zone); |
718127cc YG |
4971 | |
4972 | return 0; | |
ed8ece2e DH |
4973 | } |
4974 | ||
0ee332c1 | 4975 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4976 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
8a942fde | 4977 | |
c713216d MG |
4978 | /* |
4979 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 4980 | */ |
8a942fde MG |
4981 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
4982 | struct mminit_pfnnid_cache *state) | |
c713216d | 4983 | { |
c13291a5 | 4984 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 4985 | int nid; |
7c243c71 | 4986 | |
8a942fde MG |
4987 | if (state->last_start <= pfn && pfn < state->last_end) |
4988 | return state->last_nid; | |
c713216d | 4989 | |
e76b63f8 YL |
4990 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
4991 | if (nid != -1) { | |
8a942fde MG |
4992 | state->last_start = start_pfn; |
4993 | state->last_end = end_pfn; | |
4994 | state->last_nid = nid; | |
e76b63f8 YL |
4995 | } |
4996 | ||
4997 | return nid; | |
c713216d MG |
4998 | } |
4999 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
5000 | ||
c713216d | 5001 | /** |
6782832e | 5002 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 5003 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 5004 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 5005 | * |
7d018176 ZZ |
5006 | * If an architecture guarantees that all ranges registered contain no holes |
5007 | * and may be freed, this this function may be used instead of calling | |
5008 | * memblock_free_early_nid() manually. | |
c713216d | 5009 | */ |
c13291a5 | 5010 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 5011 | { |
c13291a5 TH |
5012 | unsigned long start_pfn, end_pfn; |
5013 | int i, this_nid; | |
edbe7d23 | 5014 | |
c13291a5 TH |
5015 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5016 | start_pfn = min(start_pfn, max_low_pfn); | |
5017 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 5018 | |
c13291a5 | 5019 | if (start_pfn < end_pfn) |
6782832e SS |
5020 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5021 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
5022 | this_nid); | |
edbe7d23 | 5023 | } |
edbe7d23 | 5024 | } |
edbe7d23 | 5025 | |
c713216d MG |
5026 | /** |
5027 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 5028 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 5029 | * |
7d018176 ZZ |
5030 | * If an architecture guarantees that all ranges registered contain no holes and may |
5031 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
5032 | */ |
5033 | void __init sparse_memory_present_with_active_regions(int nid) | |
5034 | { | |
c13291a5 TH |
5035 | unsigned long start_pfn, end_pfn; |
5036 | int i, this_nid; | |
c713216d | 5037 | |
c13291a5 TH |
5038 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5039 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
5040 | } |
5041 | ||
5042 | /** | |
5043 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
5044 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5045 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
5046 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
5047 | * |
5048 | * It returns the start and end page frame of a node based on information | |
7d018176 | 5049 | * provided by memblock_set_node(). If called for a node |
c713216d | 5050 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 5051 | * PFNs will be 0. |
c713216d | 5052 | */ |
a3142c8e | 5053 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
5054 | unsigned long *start_pfn, unsigned long *end_pfn) |
5055 | { | |
c13291a5 | 5056 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 5057 | int i; |
c13291a5 | 5058 | |
c713216d MG |
5059 | *start_pfn = -1UL; |
5060 | *end_pfn = 0; | |
5061 | ||
c13291a5 TH |
5062 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5063 | *start_pfn = min(*start_pfn, this_start_pfn); | |
5064 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
5065 | } |
5066 | ||
633c0666 | 5067 | if (*start_pfn == -1UL) |
c713216d | 5068 | *start_pfn = 0; |
c713216d MG |
5069 | } |
5070 | ||
2a1e274a MG |
5071 | /* |
5072 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
5073 | * assumption is made that zones within a node are ordered in monotonic | |
5074 | * increasing memory addresses so that the "highest" populated zone is used | |
5075 | */ | |
b69a7288 | 5076 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
5077 | { |
5078 | int zone_index; | |
5079 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
5080 | if (zone_index == ZONE_MOVABLE) | |
5081 | continue; | |
5082 | ||
5083 | if (arch_zone_highest_possible_pfn[zone_index] > | |
5084 | arch_zone_lowest_possible_pfn[zone_index]) | |
5085 | break; | |
5086 | } | |
5087 | ||
5088 | VM_BUG_ON(zone_index == -1); | |
5089 | movable_zone = zone_index; | |
5090 | } | |
5091 | ||
5092 | /* | |
5093 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 5094 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
5095 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5096 | * in each node depending on the size of each node and how evenly kernelcore | |
5097 | * is distributed. This helper function adjusts the zone ranges | |
5098 | * provided by the architecture for a given node by using the end of the | |
5099 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
5100 | * zones within a node are in order of monotonic increases memory addresses | |
5101 | */ | |
b69a7288 | 5102 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
5103 | unsigned long zone_type, |
5104 | unsigned long node_start_pfn, | |
5105 | unsigned long node_end_pfn, | |
5106 | unsigned long *zone_start_pfn, | |
5107 | unsigned long *zone_end_pfn) | |
5108 | { | |
5109 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
5110 | if (zone_movable_pfn[nid]) { | |
5111 | /* Size ZONE_MOVABLE */ | |
5112 | if (zone_type == ZONE_MOVABLE) { | |
5113 | *zone_start_pfn = zone_movable_pfn[nid]; | |
5114 | *zone_end_pfn = min(node_end_pfn, | |
5115 | arch_zone_highest_possible_pfn[movable_zone]); | |
5116 | ||
2a1e274a MG |
5117 | /* Check if this whole range is within ZONE_MOVABLE */ |
5118 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
5119 | *zone_start_pfn = *zone_end_pfn; | |
5120 | } | |
5121 | } | |
5122 | ||
c713216d MG |
5123 | /* |
5124 | * Return the number of pages a zone spans in a node, including holes | |
5125 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
5126 | */ | |
6ea6e688 | 5127 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5128 | unsigned long zone_type, |
7960aedd ZY |
5129 | unsigned long node_start_pfn, |
5130 | unsigned long node_end_pfn, | |
d91749c1 TI |
5131 | unsigned long *zone_start_pfn, |
5132 | unsigned long *zone_end_pfn, | |
c713216d MG |
5133 | unsigned long *ignored) |
5134 | { | |
b5685e92 | 5135 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5136 | if (!node_start_pfn && !node_end_pfn) |
5137 | return 0; | |
5138 | ||
7960aedd | 5139 | /* Get the start and end of the zone */ |
d91749c1 TI |
5140 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5141 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
5142 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5143 | node_start_pfn, node_end_pfn, | |
d91749c1 | 5144 | zone_start_pfn, zone_end_pfn); |
c713216d MG |
5145 | |
5146 | /* Check that this node has pages within the zone's required range */ | |
d91749c1 | 5147 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
c713216d MG |
5148 | return 0; |
5149 | ||
5150 | /* Move the zone boundaries inside the node if necessary */ | |
d91749c1 TI |
5151 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5152 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); | |
c713216d MG |
5153 | |
5154 | /* Return the spanned pages */ | |
d91749c1 | 5155 | return *zone_end_pfn - *zone_start_pfn; |
c713216d MG |
5156 | } |
5157 | ||
5158 | /* | |
5159 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 5160 | * then all holes in the requested range will be accounted for. |
c713216d | 5161 | */ |
32996250 | 5162 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
5163 | unsigned long range_start_pfn, |
5164 | unsigned long range_end_pfn) | |
5165 | { | |
96e907d1 TH |
5166 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5167 | unsigned long start_pfn, end_pfn; | |
5168 | int i; | |
c713216d | 5169 | |
96e907d1 TH |
5170 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5171 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
5172 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
5173 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 5174 | } |
96e907d1 | 5175 | return nr_absent; |
c713216d MG |
5176 | } |
5177 | ||
5178 | /** | |
5179 | * absent_pages_in_range - Return number of page frames in holes within a range | |
5180 | * @start_pfn: The start PFN to start searching for holes | |
5181 | * @end_pfn: The end PFN to stop searching for holes | |
5182 | * | |
88ca3b94 | 5183 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
5184 | */ |
5185 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
5186 | unsigned long end_pfn) | |
5187 | { | |
5188 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
5189 | } | |
5190 | ||
5191 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 5192 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5193 | unsigned long zone_type, |
7960aedd ZY |
5194 | unsigned long node_start_pfn, |
5195 | unsigned long node_end_pfn, | |
c713216d MG |
5196 | unsigned long *ignored) |
5197 | { | |
96e907d1 TH |
5198 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5199 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 | 5200 | unsigned long zone_start_pfn, zone_end_pfn; |
342332e6 | 5201 | unsigned long nr_absent; |
9c7cd687 | 5202 | |
b5685e92 | 5203 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
f9126ab9 XQ |
5204 | if (!node_start_pfn && !node_end_pfn) |
5205 | return 0; | |
5206 | ||
96e907d1 TH |
5207 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5208 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 5209 | |
2a1e274a MG |
5210 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5211 | node_start_pfn, node_end_pfn, | |
5212 | &zone_start_pfn, &zone_end_pfn); | |
342332e6 TI |
5213 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5214 | ||
5215 | /* | |
5216 | * ZONE_MOVABLE handling. | |
5217 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages | |
5218 | * and vice versa. | |
5219 | */ | |
5220 | if (zone_movable_pfn[nid]) { | |
5221 | if (mirrored_kernelcore) { | |
5222 | unsigned long start_pfn, end_pfn; | |
5223 | struct memblock_region *r; | |
5224 | ||
5225 | for_each_memblock(memory, r) { | |
5226 | start_pfn = clamp(memblock_region_memory_base_pfn(r), | |
5227 | zone_start_pfn, zone_end_pfn); | |
5228 | end_pfn = clamp(memblock_region_memory_end_pfn(r), | |
5229 | zone_start_pfn, zone_end_pfn); | |
5230 | ||
5231 | if (zone_type == ZONE_MOVABLE && | |
5232 | memblock_is_mirror(r)) | |
5233 | nr_absent += end_pfn - start_pfn; | |
5234 | ||
5235 | if (zone_type == ZONE_NORMAL && | |
5236 | !memblock_is_mirror(r)) | |
5237 | nr_absent += end_pfn - start_pfn; | |
5238 | } | |
5239 | } else { | |
5240 | if (zone_type == ZONE_NORMAL) | |
5241 | nr_absent += node_end_pfn - zone_movable_pfn[nid]; | |
5242 | } | |
5243 | } | |
5244 | ||
5245 | return nr_absent; | |
c713216d | 5246 | } |
0e0b864e | 5247 | |
0ee332c1 | 5248 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 5249 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 5250 | unsigned long zone_type, |
7960aedd ZY |
5251 | unsigned long node_start_pfn, |
5252 | unsigned long node_end_pfn, | |
d91749c1 TI |
5253 | unsigned long *zone_start_pfn, |
5254 | unsigned long *zone_end_pfn, | |
c713216d MG |
5255 | unsigned long *zones_size) |
5256 | { | |
d91749c1 TI |
5257 | unsigned int zone; |
5258 | ||
5259 | *zone_start_pfn = node_start_pfn; | |
5260 | for (zone = 0; zone < zone_type; zone++) | |
5261 | *zone_start_pfn += zones_size[zone]; | |
5262 | ||
5263 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; | |
5264 | ||
c713216d MG |
5265 | return zones_size[zone_type]; |
5266 | } | |
5267 | ||
6ea6e688 | 5268 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 5269 | unsigned long zone_type, |
7960aedd ZY |
5270 | unsigned long node_start_pfn, |
5271 | unsigned long node_end_pfn, | |
c713216d MG |
5272 | unsigned long *zholes_size) |
5273 | { | |
5274 | if (!zholes_size) | |
5275 | return 0; | |
5276 | ||
5277 | return zholes_size[zone_type]; | |
5278 | } | |
20e6926d | 5279 | |
0ee332c1 | 5280 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5281 | |
a3142c8e | 5282 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
5283 | unsigned long node_start_pfn, |
5284 | unsigned long node_end_pfn, | |
5285 | unsigned long *zones_size, | |
5286 | unsigned long *zholes_size) | |
c713216d | 5287 | { |
febd5949 | 5288 | unsigned long realtotalpages = 0, totalpages = 0; |
c713216d MG |
5289 | enum zone_type i; |
5290 | ||
febd5949 GZ |
5291 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5292 | struct zone *zone = pgdat->node_zones + i; | |
d91749c1 | 5293 | unsigned long zone_start_pfn, zone_end_pfn; |
febd5949 | 5294 | unsigned long size, real_size; |
c713216d | 5295 | |
febd5949 GZ |
5296 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5297 | node_start_pfn, | |
5298 | node_end_pfn, | |
d91749c1 TI |
5299 | &zone_start_pfn, |
5300 | &zone_end_pfn, | |
febd5949 GZ |
5301 | zones_size); |
5302 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
5303 | node_start_pfn, node_end_pfn, |
5304 | zholes_size); | |
d91749c1 TI |
5305 | if (size) |
5306 | zone->zone_start_pfn = zone_start_pfn; | |
5307 | else | |
5308 | zone->zone_start_pfn = 0; | |
febd5949 GZ |
5309 | zone->spanned_pages = size; |
5310 | zone->present_pages = real_size; | |
5311 | ||
5312 | totalpages += size; | |
5313 | realtotalpages += real_size; | |
5314 | } | |
5315 | ||
5316 | pgdat->node_spanned_pages = totalpages; | |
c713216d MG |
5317 | pgdat->node_present_pages = realtotalpages; |
5318 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
5319 | realtotalpages); | |
5320 | } | |
5321 | ||
835c134e MG |
5322 | #ifndef CONFIG_SPARSEMEM |
5323 | /* | |
5324 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
5325 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5326 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
5327 | * round what is now in bits to nearest long in bits, then return it in |
5328 | * bytes. | |
5329 | */ | |
7c45512d | 5330 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
5331 | { |
5332 | unsigned long usemapsize; | |
5333 | ||
7c45512d | 5334 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
5335 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
5336 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
5337 | usemapsize *= NR_PAGEBLOCK_BITS; |
5338 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
5339 | ||
5340 | return usemapsize / 8; | |
5341 | } | |
5342 | ||
5343 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
5344 | struct zone *zone, |
5345 | unsigned long zone_start_pfn, | |
5346 | unsigned long zonesize) | |
835c134e | 5347 | { |
7c45512d | 5348 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 5349 | zone->pageblock_flags = NULL; |
58a01a45 | 5350 | if (usemapsize) |
6782832e SS |
5351 | zone->pageblock_flags = |
5352 | memblock_virt_alloc_node_nopanic(usemapsize, | |
5353 | pgdat->node_id); | |
835c134e MG |
5354 | } |
5355 | #else | |
7c45512d LT |
5356 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
5357 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
5358 | #endif /* CONFIG_SPARSEMEM */ |
5359 | ||
d9c23400 | 5360 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 5361 | |
d9c23400 | 5362 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 5363 | void __paginginit set_pageblock_order(void) |
d9c23400 | 5364 | { |
955c1cd7 AM |
5365 | unsigned int order; |
5366 | ||
d9c23400 MG |
5367 | /* Check that pageblock_nr_pages has not already been setup */ |
5368 | if (pageblock_order) | |
5369 | return; | |
5370 | ||
955c1cd7 AM |
5371 | if (HPAGE_SHIFT > PAGE_SHIFT) |
5372 | order = HUGETLB_PAGE_ORDER; | |
5373 | else | |
5374 | order = MAX_ORDER - 1; | |
5375 | ||
d9c23400 MG |
5376 | /* |
5377 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
5378 | * This value may be variable depending on boot parameters on IA64 and |
5379 | * powerpc. | |
d9c23400 MG |
5380 | */ |
5381 | pageblock_order = order; | |
5382 | } | |
5383 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5384 | ||
ba72cb8c MG |
5385 | /* |
5386 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
5387 | * is unused as pageblock_order is set at compile-time. See |
5388 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
5389 | * the kernel config | |
ba72cb8c | 5390 | */ |
15ca220e | 5391 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 5392 | { |
ba72cb8c | 5393 | } |
d9c23400 MG |
5394 | |
5395 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
5396 | ||
01cefaef JL |
5397 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
5398 | unsigned long present_pages) | |
5399 | { | |
5400 | unsigned long pages = spanned_pages; | |
5401 | ||
5402 | /* | |
5403 | * Provide a more accurate estimation if there are holes within | |
5404 | * the zone and SPARSEMEM is in use. If there are holes within the | |
5405 | * zone, each populated memory region may cost us one or two extra | |
5406 | * memmap pages due to alignment because memmap pages for each | |
5407 | * populated regions may not naturally algined on page boundary. | |
5408 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
5409 | */ | |
5410 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
5411 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
5412 | pages = present_pages; | |
5413 | ||
5414 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
5415 | } | |
5416 | ||
1da177e4 LT |
5417 | /* |
5418 | * Set up the zone data structures: | |
5419 | * - mark all pages reserved | |
5420 | * - mark all memory queues empty | |
5421 | * - clear the memory bitmaps | |
6527af5d MK |
5422 | * |
5423 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 5424 | */ |
7f3eb55b | 5425 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
1da177e4 | 5426 | { |
2f1b6248 | 5427 | enum zone_type j; |
ed8ece2e | 5428 | int nid = pgdat->node_id; |
718127cc | 5429 | int ret; |
1da177e4 | 5430 | |
208d54e5 | 5431 | pgdat_resize_init(pgdat); |
8177a420 AA |
5432 | #ifdef CONFIG_NUMA_BALANCING |
5433 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
5434 | pgdat->numabalancing_migrate_nr_pages = 0; | |
5435 | pgdat->numabalancing_migrate_next_window = jiffies; | |
a3d0a918 KS |
5436 | #endif |
5437 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5438 | spin_lock_init(&pgdat->split_queue_lock); | |
5439 | INIT_LIST_HEAD(&pgdat->split_queue); | |
5440 | pgdat->split_queue_len = 0; | |
8177a420 | 5441 | #endif |
1da177e4 | 5442 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 5443 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
698b1b30 VB |
5444 | #ifdef CONFIG_COMPACTION |
5445 | init_waitqueue_head(&pgdat->kcompactd_wait); | |
5446 | #endif | |
eefa864b | 5447 | pgdat_page_ext_init(pgdat); |
5f63b720 | 5448 | |
1da177e4 LT |
5449 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5450 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 5451 | unsigned long size, realsize, freesize, memmap_pages; |
d91749c1 | 5452 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
1da177e4 | 5453 | |
febd5949 GZ |
5454 | size = zone->spanned_pages; |
5455 | realsize = freesize = zone->present_pages; | |
1da177e4 | 5456 | |
0e0b864e | 5457 | /* |
9feedc9d | 5458 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
5459 | * is used by this zone for memmap. This affects the watermark |
5460 | * and per-cpu initialisations | |
5461 | */ | |
01cefaef | 5462 | memmap_pages = calc_memmap_size(size, realsize); |
ba914f48 ZH |
5463 | if (!is_highmem_idx(j)) { |
5464 | if (freesize >= memmap_pages) { | |
5465 | freesize -= memmap_pages; | |
5466 | if (memmap_pages) | |
5467 | printk(KERN_DEBUG | |
5468 | " %s zone: %lu pages used for memmap\n", | |
5469 | zone_names[j], memmap_pages); | |
5470 | } else | |
5471 | printk(KERN_WARNING | |
5472 | " %s zone: %lu pages exceeds freesize %lu\n", | |
5473 | zone_names[j], memmap_pages, freesize); | |
5474 | } | |
0e0b864e | 5475 | |
6267276f | 5476 | /* Account for reserved pages */ |
9feedc9d JL |
5477 | if (j == 0 && freesize > dma_reserve) { |
5478 | freesize -= dma_reserve; | |
d903ef9f | 5479 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 5480 | zone_names[0], dma_reserve); |
0e0b864e MG |
5481 | } |
5482 | ||
98d2b0eb | 5483 | if (!is_highmem_idx(j)) |
9feedc9d | 5484 | nr_kernel_pages += freesize; |
01cefaef JL |
5485 | /* Charge for highmem memmap if there are enough kernel pages */ |
5486 | else if (nr_kernel_pages > memmap_pages * 2) | |
5487 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 5488 | nr_all_pages += freesize; |
1da177e4 | 5489 | |
9feedc9d JL |
5490 | /* |
5491 | * Set an approximate value for lowmem here, it will be adjusted | |
5492 | * when the bootmem allocator frees pages into the buddy system. | |
5493 | * And all highmem pages will be managed by the buddy system. | |
5494 | */ | |
5495 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 5496 | #ifdef CONFIG_NUMA |
d5f541ed | 5497 | zone->node = nid; |
9feedc9d | 5498 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 5499 | / 100; |
9feedc9d | 5500 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 5501 | #endif |
1da177e4 LT |
5502 | zone->name = zone_names[j]; |
5503 | spin_lock_init(&zone->lock); | |
5504 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 5505 | zone_seqlock_init(zone); |
1da177e4 | 5506 | zone->zone_pgdat = pgdat; |
ed8ece2e | 5507 | zone_pcp_init(zone); |
81c0a2bb JW |
5508 | |
5509 | /* For bootup, initialized properly in watermark setup */ | |
5510 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
5511 | ||
bea8c150 | 5512 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
5513 | if (!size) |
5514 | continue; | |
5515 | ||
955c1cd7 | 5516 | set_pageblock_order(); |
7c45512d | 5517 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
b171e409 | 5518 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
718127cc | 5519 | BUG_ON(ret); |
76cdd58e | 5520 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 LT |
5521 | } |
5522 | } | |
5523 | ||
577a32f6 | 5524 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 5525 | { |
b0aeba74 | 5526 | unsigned long __maybe_unused start = 0; |
a1c34a3b LA |
5527 | unsigned long __maybe_unused offset = 0; |
5528 | ||
1da177e4 LT |
5529 | /* Skip empty nodes */ |
5530 | if (!pgdat->node_spanned_pages) | |
5531 | return; | |
5532 | ||
d41dee36 | 5533 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
b0aeba74 TL |
5534 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
5535 | offset = pgdat->node_start_pfn - start; | |
1da177e4 LT |
5536 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
5537 | if (!pgdat->node_mem_map) { | |
b0aeba74 | 5538 | unsigned long size, end; |
d41dee36 AW |
5539 | struct page *map; |
5540 | ||
e984bb43 BP |
5541 | /* |
5542 | * The zone's endpoints aren't required to be MAX_ORDER | |
5543 | * aligned but the node_mem_map endpoints must be in order | |
5544 | * for the buddy allocator to function correctly. | |
5545 | */ | |
108bcc96 | 5546 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
5547 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
5548 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
5549 | map = alloc_remap(pgdat->node_id, size); |
5550 | if (!map) | |
6782832e SS |
5551 | map = memblock_virt_alloc_node_nopanic(size, |
5552 | pgdat->node_id); | |
a1c34a3b | 5553 | pgdat->node_mem_map = map + offset; |
1da177e4 | 5554 | } |
12d810c1 | 5555 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
5556 | /* |
5557 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
5558 | */ | |
c713216d | 5559 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 5560 | mem_map = NODE_DATA(0)->node_mem_map; |
a1c34a3b | 5561 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
c713216d | 5562 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
a1c34a3b | 5563 | mem_map -= offset; |
0ee332c1 | 5564 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5565 | } |
1da177e4 | 5566 | #endif |
d41dee36 | 5567 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
5568 | } |
5569 | ||
9109fb7b JW |
5570 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
5571 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 5572 | { |
9109fb7b | 5573 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
5574 | unsigned long start_pfn = 0; |
5575 | unsigned long end_pfn = 0; | |
9109fb7b | 5576 | |
88fdf75d | 5577 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 5578 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 5579 | |
3a80a7fa | 5580 | reset_deferred_meminit(pgdat); |
1da177e4 LT |
5581 | pgdat->node_id = nid; |
5582 | pgdat->node_start_pfn = node_start_pfn; | |
7960aedd ZY |
5583 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5584 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
8d29e18a | 5585 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
4ada0c5a ZL |
5586 | (u64)start_pfn << PAGE_SHIFT, |
5587 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); | |
d91749c1 TI |
5588 | #else |
5589 | start_pfn = node_start_pfn; | |
7960aedd ZY |
5590 | #endif |
5591 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
5592 | zones_size, zholes_size); | |
1da177e4 LT |
5593 | |
5594 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
5595 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
5596 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
5597 | nid, (unsigned long)pgdat, | |
5598 | (unsigned long)pgdat->node_mem_map); | |
5599 | #endif | |
1da177e4 | 5600 | |
7f3eb55b | 5601 | free_area_init_core(pgdat); |
1da177e4 LT |
5602 | } |
5603 | ||
0ee332c1 | 5604 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
5605 | |
5606 | #if MAX_NUMNODES > 1 | |
5607 | /* | |
5608 | * Figure out the number of possible node ids. | |
5609 | */ | |
f9872caf | 5610 | void __init setup_nr_node_ids(void) |
418508c1 | 5611 | { |
904a9553 | 5612 | unsigned int highest; |
418508c1 | 5613 | |
904a9553 | 5614 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
418508c1 MS |
5615 | nr_node_ids = highest + 1; |
5616 | } | |
418508c1 MS |
5617 | #endif |
5618 | ||
1e01979c TH |
5619 | /** |
5620 | * node_map_pfn_alignment - determine the maximum internode alignment | |
5621 | * | |
5622 | * This function should be called after node map is populated and sorted. | |
5623 | * It calculates the maximum power of two alignment which can distinguish | |
5624 | * all the nodes. | |
5625 | * | |
5626 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
5627 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
5628 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
5629 | * shifted, 1GiB is enough and this function will indicate so. | |
5630 | * | |
5631 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
5632 | * model has fine enough granularity to avoid incorrect mapping for the | |
5633 | * populated node map. | |
5634 | * | |
5635 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
5636 | * requirement (single node). | |
5637 | */ | |
5638 | unsigned long __init node_map_pfn_alignment(void) | |
5639 | { | |
5640 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 5641 | unsigned long start, end, mask; |
1e01979c | 5642 | int last_nid = -1; |
c13291a5 | 5643 | int i, nid; |
1e01979c | 5644 | |
c13291a5 | 5645 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5646 | if (!start || last_nid < 0 || last_nid == nid) { |
5647 | last_nid = nid; | |
5648 | last_end = end; | |
5649 | continue; | |
5650 | } | |
5651 | ||
5652 | /* | |
5653 | * Start with a mask granular enough to pin-point to the | |
5654 | * start pfn and tick off bits one-by-one until it becomes | |
5655 | * too coarse to separate the current node from the last. | |
5656 | */ | |
5657 | mask = ~((1 << __ffs(start)) - 1); | |
5658 | while (mask && last_end <= (start & (mask << 1))) | |
5659 | mask <<= 1; | |
5660 | ||
5661 | /* accumulate all internode masks */ | |
5662 | accl_mask |= mask; | |
5663 | } | |
5664 | ||
5665 | /* convert mask to number of pages */ | |
5666 | return ~accl_mask + 1; | |
5667 | } | |
5668 | ||
a6af2bc3 | 5669 | /* Find the lowest pfn for a node */ |
b69a7288 | 5670 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 5671 | { |
a6af2bc3 | 5672 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
5673 | unsigned long start_pfn; |
5674 | int i; | |
1abbfb41 | 5675 | |
c13291a5 TH |
5676 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
5677 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 5678 | |
a6af2bc3 MG |
5679 | if (min_pfn == ULONG_MAX) { |
5680 | printk(KERN_WARNING | |
2bc0d261 | 5681 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
5682 | return 0; |
5683 | } | |
5684 | ||
5685 | return min_pfn; | |
c713216d MG |
5686 | } |
5687 | ||
5688 | /** | |
5689 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
5690 | * | |
5691 | * It returns the minimum PFN based on information provided via | |
7d018176 | 5692 | * memblock_set_node(). |
c713216d MG |
5693 | */ |
5694 | unsigned long __init find_min_pfn_with_active_regions(void) | |
5695 | { | |
5696 | return find_min_pfn_for_node(MAX_NUMNODES); | |
5697 | } | |
5698 | ||
37b07e41 LS |
5699 | /* |
5700 | * early_calculate_totalpages() | |
5701 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 5702 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 5703 | */ |
484f51f8 | 5704 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 5705 | { |
7e63efef | 5706 | unsigned long totalpages = 0; |
c13291a5 TH |
5707 | unsigned long start_pfn, end_pfn; |
5708 | int i, nid; | |
5709 | ||
5710 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
5711 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 5712 | |
37b07e41 LS |
5713 | totalpages += pages; |
5714 | if (pages) | |
4b0ef1fe | 5715 | node_set_state(nid, N_MEMORY); |
37b07e41 | 5716 | } |
b8af2941 | 5717 | return totalpages; |
7e63efef MG |
5718 | } |
5719 | ||
2a1e274a MG |
5720 | /* |
5721 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
5722 | * is spread evenly between nodes as long as the nodes have enough | |
5723 | * memory. When they don't, some nodes will have more kernelcore than | |
5724 | * others | |
5725 | */ | |
b224ef85 | 5726 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
5727 | { |
5728 | int i, nid; | |
5729 | unsigned long usable_startpfn; | |
5730 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 5731 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 5732 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 5733 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 5734 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 5735 | struct memblock_region *r; |
b2f3eebe TC |
5736 | |
5737 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
5738 | find_usable_zone_for_movable(); | |
5739 | ||
5740 | /* | |
5741 | * If movable_node is specified, ignore kernelcore and movablecore | |
5742 | * options. | |
5743 | */ | |
5744 | if (movable_node_is_enabled()) { | |
136199f0 EM |
5745 | for_each_memblock(memory, r) { |
5746 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
5747 | continue; |
5748 | ||
136199f0 | 5749 | nid = r->nid; |
b2f3eebe | 5750 | |
136199f0 | 5751 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
5752 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
5753 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5754 | usable_startpfn; | |
5755 | } | |
5756 | ||
5757 | goto out2; | |
5758 | } | |
2a1e274a | 5759 | |
342332e6 TI |
5760 | /* |
5761 | * If kernelcore=mirror is specified, ignore movablecore option | |
5762 | */ | |
5763 | if (mirrored_kernelcore) { | |
5764 | bool mem_below_4gb_not_mirrored = false; | |
5765 | ||
5766 | for_each_memblock(memory, r) { | |
5767 | if (memblock_is_mirror(r)) | |
5768 | continue; | |
5769 | ||
5770 | nid = r->nid; | |
5771 | ||
5772 | usable_startpfn = memblock_region_memory_base_pfn(r); | |
5773 | ||
5774 | if (usable_startpfn < 0x100000) { | |
5775 | mem_below_4gb_not_mirrored = true; | |
5776 | continue; | |
5777 | } | |
5778 | ||
5779 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? | |
5780 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5781 | usable_startpfn; | |
5782 | } | |
5783 | ||
5784 | if (mem_below_4gb_not_mirrored) | |
5785 | pr_warn("This configuration results in unmirrored kernel memory."); | |
5786 | ||
5787 | goto out2; | |
5788 | } | |
5789 | ||
7e63efef | 5790 | /* |
b2f3eebe | 5791 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
5792 | * kernelcore that corresponds so that memory usable for |
5793 | * any allocation type is evenly spread. If both kernelcore | |
5794 | * and movablecore are specified, then the value of kernelcore | |
5795 | * will be used for required_kernelcore if it's greater than | |
5796 | * what movablecore would have allowed. | |
5797 | */ | |
5798 | if (required_movablecore) { | |
7e63efef MG |
5799 | unsigned long corepages; |
5800 | ||
5801 | /* | |
5802 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
5803 | * was requested by the user | |
5804 | */ | |
5805 | required_movablecore = | |
5806 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
9fd745d4 | 5807 | required_movablecore = min(totalpages, required_movablecore); |
7e63efef MG |
5808 | corepages = totalpages - required_movablecore; |
5809 | ||
5810 | required_kernelcore = max(required_kernelcore, corepages); | |
5811 | } | |
5812 | ||
bde304bd XQ |
5813 | /* |
5814 | * If kernelcore was not specified or kernelcore size is larger | |
5815 | * than totalpages, there is no ZONE_MOVABLE. | |
5816 | */ | |
5817 | if (!required_kernelcore || required_kernelcore >= totalpages) | |
66918dcd | 5818 | goto out; |
2a1e274a MG |
5819 | |
5820 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
5821 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
5822 | ||
5823 | restart: | |
5824 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
5825 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 5826 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
5827 | unsigned long start_pfn, end_pfn; |
5828 | ||
2a1e274a MG |
5829 | /* |
5830 | * Recalculate kernelcore_node if the division per node | |
5831 | * now exceeds what is necessary to satisfy the requested | |
5832 | * amount of memory for the kernel | |
5833 | */ | |
5834 | if (required_kernelcore < kernelcore_node) | |
5835 | kernelcore_node = required_kernelcore / usable_nodes; | |
5836 | ||
5837 | /* | |
5838 | * As the map is walked, we track how much memory is usable | |
5839 | * by the kernel using kernelcore_remaining. When it is | |
5840 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
5841 | */ | |
5842 | kernelcore_remaining = kernelcore_node; | |
5843 | ||
5844 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 5845 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
5846 | unsigned long size_pages; |
5847 | ||
c13291a5 | 5848 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
5849 | if (start_pfn >= end_pfn) |
5850 | continue; | |
5851 | ||
5852 | /* Account for what is only usable for kernelcore */ | |
5853 | if (start_pfn < usable_startpfn) { | |
5854 | unsigned long kernel_pages; | |
5855 | kernel_pages = min(end_pfn, usable_startpfn) | |
5856 | - start_pfn; | |
5857 | ||
5858 | kernelcore_remaining -= min(kernel_pages, | |
5859 | kernelcore_remaining); | |
5860 | required_kernelcore -= min(kernel_pages, | |
5861 | required_kernelcore); | |
5862 | ||
5863 | /* Continue if range is now fully accounted */ | |
5864 | if (end_pfn <= usable_startpfn) { | |
5865 | ||
5866 | /* | |
5867 | * Push zone_movable_pfn to the end so | |
5868 | * that if we have to rebalance | |
5869 | * kernelcore across nodes, we will | |
5870 | * not double account here | |
5871 | */ | |
5872 | zone_movable_pfn[nid] = end_pfn; | |
5873 | continue; | |
5874 | } | |
5875 | start_pfn = usable_startpfn; | |
5876 | } | |
5877 | ||
5878 | /* | |
5879 | * The usable PFN range for ZONE_MOVABLE is from | |
5880 | * start_pfn->end_pfn. Calculate size_pages as the | |
5881 | * number of pages used as kernelcore | |
5882 | */ | |
5883 | size_pages = end_pfn - start_pfn; | |
5884 | if (size_pages > kernelcore_remaining) | |
5885 | size_pages = kernelcore_remaining; | |
5886 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5887 | ||
5888 | /* | |
5889 | * Some kernelcore has been met, update counts and | |
5890 | * break if the kernelcore for this node has been | |
b8af2941 | 5891 | * satisfied |
2a1e274a MG |
5892 | */ |
5893 | required_kernelcore -= min(required_kernelcore, | |
5894 | size_pages); | |
5895 | kernelcore_remaining -= size_pages; | |
5896 | if (!kernelcore_remaining) | |
5897 | break; | |
5898 | } | |
5899 | } | |
5900 | ||
5901 | /* | |
5902 | * If there is still required_kernelcore, we do another pass with one | |
5903 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5904 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 5905 | * satisfied |
2a1e274a MG |
5906 | */ |
5907 | usable_nodes--; | |
5908 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5909 | goto restart; | |
5910 | ||
b2f3eebe | 5911 | out2: |
2a1e274a MG |
5912 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
5913 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5914 | zone_movable_pfn[nid] = | |
5915 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5916 | |
20e6926d | 5917 | out: |
66918dcd | 5918 | /* restore the node_state */ |
4b0ef1fe | 5919 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5920 | } |
5921 | ||
4b0ef1fe LJ |
5922 | /* Any regular or high memory on that node ? */ |
5923 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5924 | { |
37b07e41 LS |
5925 | enum zone_type zone_type; |
5926 | ||
4b0ef1fe LJ |
5927 | if (N_MEMORY == N_NORMAL_MEMORY) |
5928 | return; | |
5929 | ||
5930 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5931 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 5932 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
5933 | node_set_state(nid, N_HIGH_MEMORY); |
5934 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5935 | zone_type <= ZONE_NORMAL) | |
5936 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5937 | break; |
5938 | } | |
37b07e41 | 5939 | } |
37b07e41 LS |
5940 | } |
5941 | ||
c713216d MG |
5942 | /** |
5943 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 5944 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
5945 | * |
5946 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 5947 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
5948 | * zone in each node and their holes is calculated. If the maximum PFN |
5949 | * between two adjacent zones match, it is assumed that the zone is empty. | |
5950 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
5951 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
5952 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
5953 | * at arch_max_dma_pfn. | |
5954 | */ | |
5955 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
5956 | { | |
c13291a5 TH |
5957 | unsigned long start_pfn, end_pfn; |
5958 | int i, nid; | |
a6af2bc3 | 5959 | |
c713216d MG |
5960 | /* Record where the zone boundaries are */ |
5961 | memset(arch_zone_lowest_possible_pfn, 0, | |
5962 | sizeof(arch_zone_lowest_possible_pfn)); | |
5963 | memset(arch_zone_highest_possible_pfn, 0, | |
5964 | sizeof(arch_zone_highest_possible_pfn)); | |
5965 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
5966 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
5967 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
5968 | if (i == ZONE_MOVABLE) |
5969 | continue; | |
c713216d MG |
5970 | arch_zone_lowest_possible_pfn[i] = |
5971 | arch_zone_highest_possible_pfn[i-1]; | |
5972 | arch_zone_highest_possible_pfn[i] = | |
5973 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
5974 | } | |
2a1e274a MG |
5975 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
5976 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
5977 | ||
5978 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
5979 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 5980 | find_zone_movable_pfns_for_nodes(); |
c713216d | 5981 | |
c713216d | 5982 | /* Print out the zone ranges */ |
f88dfff5 | 5983 | pr_info("Zone ranges:\n"); |
2a1e274a MG |
5984 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5985 | if (i == ZONE_MOVABLE) | |
5986 | continue; | |
f88dfff5 | 5987 | pr_info(" %-8s ", zone_names[i]); |
72f0ba02 DR |
5988 | if (arch_zone_lowest_possible_pfn[i] == |
5989 | arch_zone_highest_possible_pfn[i]) | |
f88dfff5 | 5990 | pr_cont("empty\n"); |
72f0ba02 | 5991 | else |
8d29e18a JG |
5992 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
5993 | (u64)arch_zone_lowest_possible_pfn[i] | |
5994 | << PAGE_SHIFT, | |
5995 | ((u64)arch_zone_highest_possible_pfn[i] | |
a62e2f4f | 5996 | << PAGE_SHIFT) - 1); |
2a1e274a MG |
5997 | } |
5998 | ||
5999 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
f88dfff5 | 6000 | pr_info("Movable zone start for each node\n"); |
2a1e274a MG |
6001 | for (i = 0; i < MAX_NUMNODES; i++) { |
6002 | if (zone_movable_pfn[i]) | |
8d29e18a JG |
6003 | pr_info(" Node %d: %#018Lx\n", i, |
6004 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 6005 | } |
c713216d | 6006 | |
f2d52fe5 | 6007 | /* Print out the early node map */ |
f88dfff5 | 6008 | pr_info("Early memory node ranges\n"); |
c13291a5 | 6009 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
8d29e18a JG |
6010 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6011 | (u64)start_pfn << PAGE_SHIFT, | |
6012 | ((u64)end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
6013 | |
6014 | /* Initialise every node */ | |
708614e6 | 6015 | mminit_verify_pageflags_layout(); |
8ef82866 | 6016 | setup_nr_node_ids(); |
c713216d MG |
6017 | for_each_online_node(nid) { |
6018 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 6019 | free_area_init_node(nid, NULL, |
c713216d | 6020 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
6021 | |
6022 | /* Any memory on that node */ | |
6023 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
6024 | node_set_state(nid, N_MEMORY); |
6025 | check_for_memory(pgdat, nid); | |
c713216d MG |
6026 | } |
6027 | } | |
2a1e274a | 6028 | |
7e63efef | 6029 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
6030 | { |
6031 | unsigned long long coremem; | |
6032 | if (!p) | |
6033 | return -EINVAL; | |
6034 | ||
6035 | coremem = memparse(p, &p); | |
7e63efef | 6036 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 6037 | |
7e63efef | 6038 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
6039 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6040 | ||
6041 | return 0; | |
6042 | } | |
ed7ed365 | 6043 | |
7e63efef MG |
6044 | /* |
6045 | * kernelcore=size sets the amount of memory for use for allocations that | |
6046 | * cannot be reclaimed or migrated. | |
6047 | */ | |
6048 | static int __init cmdline_parse_kernelcore(char *p) | |
6049 | { | |
342332e6 TI |
6050 | /* parse kernelcore=mirror */ |
6051 | if (parse_option_str(p, "mirror")) { | |
6052 | mirrored_kernelcore = true; | |
6053 | return 0; | |
6054 | } | |
6055 | ||
7e63efef MG |
6056 | return cmdline_parse_core(p, &required_kernelcore); |
6057 | } | |
6058 | ||
6059 | /* | |
6060 | * movablecore=size sets the amount of memory for use for allocations that | |
6061 | * can be reclaimed or migrated. | |
6062 | */ | |
6063 | static int __init cmdline_parse_movablecore(char *p) | |
6064 | { | |
6065 | return cmdline_parse_core(p, &required_movablecore); | |
6066 | } | |
6067 | ||
ed7ed365 | 6068 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 6069 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 6070 | |
0ee332c1 | 6071 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 6072 | |
c3d5f5f0 JL |
6073 | void adjust_managed_page_count(struct page *page, long count) |
6074 | { | |
6075 | spin_lock(&managed_page_count_lock); | |
6076 | page_zone(page)->managed_pages += count; | |
6077 | totalram_pages += count; | |
3dcc0571 JL |
6078 | #ifdef CONFIG_HIGHMEM |
6079 | if (PageHighMem(page)) | |
6080 | totalhigh_pages += count; | |
6081 | #endif | |
c3d5f5f0 JL |
6082 | spin_unlock(&managed_page_count_lock); |
6083 | } | |
3dcc0571 | 6084 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 6085 | |
11199692 | 6086 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 6087 | { |
11199692 JL |
6088 | void *pos; |
6089 | unsigned long pages = 0; | |
69afade7 | 6090 | |
11199692 JL |
6091 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6092 | end = (void *)((unsigned long)end & PAGE_MASK); | |
6093 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 6094 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
6095 | memset(pos, poison, PAGE_SIZE); |
6096 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
6097 | } |
6098 | ||
6099 | if (pages && s) | |
11199692 | 6100 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
6101 | s, pages << (PAGE_SHIFT - 10), start, end); |
6102 | ||
6103 | return pages; | |
6104 | } | |
11199692 | 6105 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 6106 | |
cfa11e08 JL |
6107 | #ifdef CONFIG_HIGHMEM |
6108 | void free_highmem_page(struct page *page) | |
6109 | { | |
6110 | __free_reserved_page(page); | |
6111 | totalram_pages++; | |
7b4b2a0d | 6112 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
6113 | totalhigh_pages++; |
6114 | } | |
6115 | #endif | |
6116 | ||
7ee3d4e8 JL |
6117 | |
6118 | void __init mem_init_print_info(const char *str) | |
6119 | { | |
6120 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
6121 | unsigned long init_code_size, init_data_size; | |
6122 | ||
6123 | physpages = get_num_physpages(); | |
6124 | codesize = _etext - _stext; | |
6125 | datasize = _edata - _sdata; | |
6126 | rosize = __end_rodata - __start_rodata; | |
6127 | bss_size = __bss_stop - __bss_start; | |
6128 | init_data_size = __init_end - __init_begin; | |
6129 | init_code_size = _einittext - _sinittext; | |
6130 | ||
6131 | /* | |
6132 | * Detect special cases and adjust section sizes accordingly: | |
6133 | * 1) .init.* may be embedded into .data sections | |
6134 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
6135 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
6136 | * 3) .rodata.* may be embedded into .text or .data sections. | |
6137 | */ | |
6138 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
6139 | do { \ |
6140 | if (start <= pos && pos < end && size > adj) \ | |
6141 | size -= adj; \ | |
6142 | } while (0) | |
7ee3d4e8 JL |
6143 | |
6144 | adj_init_size(__init_begin, __init_end, init_data_size, | |
6145 | _sinittext, init_code_size); | |
6146 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
6147 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
6148 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
6149 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
6150 | ||
6151 | #undef adj_init_size | |
6152 | ||
f88dfff5 | 6153 | pr_info("Memory: %luK/%luK available " |
7ee3d4e8 | 6154 | "(%luK kernel code, %luK rwdata, %luK rodata, " |
e48322ab | 6155 | "%luK init, %luK bss, %luK reserved, %luK cma-reserved" |
7ee3d4e8 JL |
6156 | #ifdef CONFIG_HIGHMEM |
6157 | ", %luK highmem" | |
6158 | #endif | |
6159 | "%s%s)\n", | |
6160 | nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10), | |
6161 | codesize >> 10, datasize >> 10, rosize >> 10, | |
6162 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
e48322ab PK |
6163 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10), |
6164 | totalcma_pages << (PAGE_SHIFT-10), | |
7ee3d4e8 JL |
6165 | #ifdef CONFIG_HIGHMEM |
6166 | totalhigh_pages << (PAGE_SHIFT-10), | |
6167 | #endif | |
6168 | str ? ", " : "", str ? str : ""); | |
6169 | } | |
6170 | ||
0e0b864e | 6171 | /** |
88ca3b94 RD |
6172 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6173 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e | 6174 | * |
013110a7 | 6175 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
0e0b864e MG |
6176 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6177 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
6178 | * function may optionally be used to account for unfreeable pages in the |
6179 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
6180 | * smaller per-cpu batchsize. | |
0e0b864e MG |
6181 | */ |
6182 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
6183 | { | |
6184 | dma_reserve = new_dma_reserve; | |
6185 | } | |
6186 | ||
1da177e4 LT |
6187 | void __init free_area_init(unsigned long *zones_size) |
6188 | { | |
9109fb7b | 6189 | free_area_init_node(0, zones_size, |
1da177e4 LT |
6190 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6191 | } | |
1da177e4 | 6192 | |
1da177e4 LT |
6193 | static int page_alloc_cpu_notify(struct notifier_block *self, |
6194 | unsigned long action, void *hcpu) | |
6195 | { | |
6196 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 6197 | |
8bb78442 | 6198 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 6199 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
6200 | drain_pages(cpu); |
6201 | ||
6202 | /* | |
6203 | * Spill the event counters of the dead processor | |
6204 | * into the current processors event counters. | |
6205 | * This artificially elevates the count of the current | |
6206 | * processor. | |
6207 | */ | |
f8891e5e | 6208 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
6209 | |
6210 | /* | |
6211 | * Zero the differential counters of the dead processor | |
6212 | * so that the vm statistics are consistent. | |
6213 | * | |
6214 | * This is only okay since the processor is dead and cannot | |
6215 | * race with what we are doing. | |
6216 | */ | |
2bb921e5 | 6217 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
6218 | } |
6219 | return NOTIFY_OK; | |
6220 | } | |
1da177e4 LT |
6221 | |
6222 | void __init page_alloc_init(void) | |
6223 | { | |
6224 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
6225 | } | |
6226 | ||
cb45b0e9 | 6227 | /* |
34b10060 | 6228 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
cb45b0e9 HA |
6229 | * or min_free_kbytes changes. |
6230 | */ | |
6231 | static void calculate_totalreserve_pages(void) | |
6232 | { | |
6233 | struct pglist_data *pgdat; | |
6234 | unsigned long reserve_pages = 0; | |
2f6726e5 | 6235 | enum zone_type i, j; |
cb45b0e9 HA |
6236 | |
6237 | for_each_online_pgdat(pgdat) { | |
6238 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
6239 | struct zone *zone = pgdat->node_zones + i; | |
3484b2de | 6240 | long max = 0; |
cb45b0e9 HA |
6241 | |
6242 | /* Find valid and maximum lowmem_reserve in the zone */ | |
6243 | for (j = i; j < MAX_NR_ZONES; j++) { | |
6244 | if (zone->lowmem_reserve[j] > max) | |
6245 | max = zone->lowmem_reserve[j]; | |
6246 | } | |
6247 | ||
41858966 MG |
6248 | /* we treat the high watermark as reserved pages. */ |
6249 | max += high_wmark_pages(zone); | |
cb45b0e9 | 6250 | |
b40da049 JL |
6251 | if (max > zone->managed_pages) |
6252 | max = zone->managed_pages; | |
a8d01437 JW |
6253 | |
6254 | zone->totalreserve_pages = max; | |
6255 | ||
cb45b0e9 HA |
6256 | reserve_pages += max; |
6257 | } | |
6258 | } | |
6259 | totalreserve_pages = reserve_pages; | |
6260 | } | |
6261 | ||
1da177e4 LT |
6262 | /* |
6263 | * setup_per_zone_lowmem_reserve - called whenever | |
34b10060 | 6264 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
1da177e4 LT |
6265 | * has a correct pages reserved value, so an adequate number of |
6266 | * pages are left in the zone after a successful __alloc_pages(). | |
6267 | */ | |
6268 | static void setup_per_zone_lowmem_reserve(void) | |
6269 | { | |
6270 | struct pglist_data *pgdat; | |
2f6726e5 | 6271 | enum zone_type j, idx; |
1da177e4 | 6272 | |
ec936fc5 | 6273 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
6274 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6275 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 6276 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
6277 | |
6278 | zone->lowmem_reserve[j] = 0; | |
6279 | ||
2f6726e5 CL |
6280 | idx = j; |
6281 | while (idx) { | |
1da177e4 LT |
6282 | struct zone *lower_zone; |
6283 | ||
2f6726e5 CL |
6284 | idx--; |
6285 | ||
1da177e4 LT |
6286 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6287 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
6288 | ||
6289 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 6290 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 6291 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 6292 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
6293 | } |
6294 | } | |
6295 | } | |
cb45b0e9 HA |
6296 | |
6297 | /* update totalreserve_pages */ | |
6298 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6299 | } |
6300 | ||
cfd3da1e | 6301 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
6302 | { |
6303 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
6304 | unsigned long lowmem_pages = 0; | |
6305 | struct zone *zone; | |
6306 | unsigned long flags; | |
6307 | ||
6308 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
6309 | for_each_zone(zone) { | |
6310 | if (!is_highmem(zone)) | |
b40da049 | 6311 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
6312 | } |
6313 | ||
6314 | for_each_zone(zone) { | |
ac924c60 AM |
6315 | u64 tmp; |
6316 | ||
1125b4e3 | 6317 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 6318 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 6319 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
6320 | if (is_highmem(zone)) { |
6321 | /* | |
669ed175 NP |
6322 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6323 | * need highmem pages, so cap pages_min to a small | |
6324 | * value here. | |
6325 | * | |
41858966 | 6326 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
42ff2703 | 6327 | * deltas control asynch page reclaim, and so should |
669ed175 | 6328 | * not be capped for highmem. |
1da177e4 | 6329 | */ |
90ae8d67 | 6330 | unsigned long min_pages; |
1da177e4 | 6331 | |
b40da049 | 6332 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 6333 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 6334 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 6335 | } else { |
669ed175 NP |
6336 | /* |
6337 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
6338 | * proportionate to the zone's size. |
6339 | */ | |
41858966 | 6340 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
6341 | } |
6342 | ||
795ae7a0 JW |
6343 | /* |
6344 | * Set the kswapd watermarks distance according to the | |
6345 | * scale factor in proportion to available memory, but | |
6346 | * ensure a minimum size on small systems. | |
6347 | */ | |
6348 | tmp = max_t(u64, tmp >> 2, | |
6349 | mult_frac(zone->managed_pages, | |
6350 | watermark_scale_factor, 10000)); | |
6351 | ||
6352 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; | |
6353 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2; | |
49f223a9 | 6354 | |
81c0a2bb | 6355 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
abe5f972 JW |
6356 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
6357 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
81c0a2bb | 6358 | |
1125b4e3 | 6359 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 6360 | } |
cb45b0e9 HA |
6361 | |
6362 | /* update totalreserve_pages */ | |
6363 | calculate_totalreserve_pages(); | |
1da177e4 LT |
6364 | } |
6365 | ||
cfd3da1e MG |
6366 | /** |
6367 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
6368 | * or when memory is hot-{added|removed} | |
6369 | * | |
6370 | * Ensures that the watermark[min,low,high] values for each zone are set | |
6371 | * correctly with respect to min_free_kbytes. | |
6372 | */ | |
6373 | void setup_per_zone_wmarks(void) | |
6374 | { | |
6375 | mutex_lock(&zonelists_mutex); | |
6376 | __setup_per_zone_wmarks(); | |
6377 | mutex_unlock(&zonelists_mutex); | |
6378 | } | |
6379 | ||
55a4462a | 6380 | /* |
556adecb RR |
6381 | * The inactive anon list should be small enough that the VM never has to |
6382 | * do too much work, but large enough that each inactive page has a chance | |
6383 | * to be referenced again before it is swapped out. | |
6384 | * | |
6385 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
6386 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
6387 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
6388 | * the anonymous pages are kept on the inactive list. | |
6389 | * | |
6390 | * total target max | |
6391 | * memory ratio inactive anon | |
6392 | * ------------------------------------- | |
6393 | * 10MB 1 5MB | |
6394 | * 100MB 1 50MB | |
6395 | * 1GB 3 250MB | |
6396 | * 10GB 10 0.9GB | |
6397 | * 100GB 31 3GB | |
6398 | * 1TB 101 10GB | |
6399 | * 10TB 320 32GB | |
6400 | */ | |
1b79acc9 | 6401 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 6402 | { |
96cb4df5 | 6403 | unsigned int gb, ratio; |
556adecb | 6404 | |
96cb4df5 | 6405 | /* Zone size in gigabytes */ |
b40da049 | 6406 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 6407 | if (gb) |
556adecb | 6408 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
6409 | else |
6410 | ratio = 1; | |
556adecb | 6411 | |
96cb4df5 MK |
6412 | zone->inactive_ratio = ratio; |
6413 | } | |
556adecb | 6414 | |
839a4fcc | 6415 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
6416 | { |
6417 | struct zone *zone; | |
6418 | ||
6419 | for_each_zone(zone) | |
6420 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
6421 | } |
6422 | ||
1da177e4 LT |
6423 | /* |
6424 | * Initialise min_free_kbytes. | |
6425 | * | |
6426 | * For small machines we want it small (128k min). For large machines | |
6427 | * we want it large (64MB max). But it is not linear, because network | |
6428 | * bandwidth does not increase linearly with machine size. We use | |
6429 | * | |
b8af2941 | 6430 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
6431 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
6432 | * | |
6433 | * which yields | |
6434 | * | |
6435 | * 16MB: 512k | |
6436 | * 32MB: 724k | |
6437 | * 64MB: 1024k | |
6438 | * 128MB: 1448k | |
6439 | * 256MB: 2048k | |
6440 | * 512MB: 2896k | |
6441 | * 1024MB: 4096k | |
6442 | * 2048MB: 5792k | |
6443 | * 4096MB: 8192k | |
6444 | * 8192MB: 11584k | |
6445 | * 16384MB: 16384k | |
6446 | */ | |
1b79acc9 | 6447 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
6448 | { |
6449 | unsigned long lowmem_kbytes; | |
5f12733e | 6450 | int new_min_free_kbytes; |
1da177e4 LT |
6451 | |
6452 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
6453 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
6454 | ||
6455 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
6456 | min_free_kbytes = new_min_free_kbytes; | |
6457 | if (min_free_kbytes < 128) | |
6458 | min_free_kbytes = 128; | |
6459 | if (min_free_kbytes > 65536) | |
6460 | min_free_kbytes = 65536; | |
6461 | } else { | |
6462 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
6463 | new_min_free_kbytes, user_min_free_kbytes); | |
6464 | } | |
bc75d33f | 6465 | setup_per_zone_wmarks(); |
a6cccdc3 | 6466 | refresh_zone_stat_thresholds(); |
1da177e4 | 6467 | setup_per_zone_lowmem_reserve(); |
556adecb | 6468 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
6469 | return 0; |
6470 | } | |
bc75d33f | 6471 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
6472 | |
6473 | /* | |
b8af2941 | 6474 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
6475 | * that we can call two helper functions whenever min_free_kbytes |
6476 | * changes. | |
6477 | */ | |
cccad5b9 | 6478 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6479 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6480 | { |
da8c757b HP |
6481 | int rc; |
6482 | ||
6483 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6484 | if (rc) | |
6485 | return rc; | |
6486 | ||
5f12733e MH |
6487 | if (write) { |
6488 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 6489 | setup_per_zone_wmarks(); |
5f12733e | 6490 | } |
1da177e4 LT |
6491 | return 0; |
6492 | } | |
6493 | ||
795ae7a0 JW |
6494 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
6495 | void __user *buffer, size_t *length, loff_t *ppos) | |
6496 | { | |
6497 | int rc; | |
6498 | ||
6499 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
6500 | if (rc) | |
6501 | return rc; | |
6502 | ||
6503 | if (write) | |
6504 | setup_per_zone_wmarks(); | |
6505 | ||
6506 | return 0; | |
6507 | } | |
6508 | ||
9614634f | 6509 | #ifdef CONFIG_NUMA |
cccad5b9 | 6510 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6511 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
6512 | { |
6513 | struct zone *zone; | |
6514 | int rc; | |
6515 | ||
8d65af78 | 6516 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
6517 | if (rc) |
6518 | return rc; | |
6519 | ||
6520 | for_each_zone(zone) | |
b40da049 | 6521 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
6522 | sysctl_min_unmapped_ratio) / 100; |
6523 | return 0; | |
6524 | } | |
0ff38490 | 6525 | |
cccad5b9 | 6526 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6527 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
6528 | { |
6529 | struct zone *zone; | |
6530 | int rc; | |
6531 | ||
8d65af78 | 6532 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
6533 | if (rc) |
6534 | return rc; | |
6535 | ||
6536 | for_each_zone(zone) | |
b40da049 | 6537 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
6538 | sysctl_min_slab_ratio) / 100; |
6539 | return 0; | |
6540 | } | |
9614634f CL |
6541 | #endif |
6542 | ||
1da177e4 LT |
6543 | /* |
6544 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
6545 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
6546 | * whenever sysctl_lowmem_reserve_ratio changes. | |
6547 | * | |
6548 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 6549 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
6550 | * if in function of the boot time zone sizes. |
6551 | */ | |
cccad5b9 | 6552 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6553 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 6554 | { |
8d65af78 | 6555 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
6556 | setup_per_zone_lowmem_reserve(); |
6557 | return 0; | |
6558 | } | |
6559 | ||
8ad4b1fb RS |
6560 | /* |
6561 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
6562 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
6563 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 6564 | */ |
cccad5b9 | 6565 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 6566 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
6567 | { |
6568 | struct zone *zone; | |
7cd2b0a3 | 6569 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
6570 | int ret; |
6571 | ||
7cd2b0a3 DR |
6572 | mutex_lock(&pcp_batch_high_lock); |
6573 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
6574 | ||
8d65af78 | 6575 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
6576 | if (!write || ret < 0) |
6577 | goto out; | |
6578 | ||
6579 | /* Sanity checking to avoid pcp imbalance */ | |
6580 | if (percpu_pagelist_fraction && | |
6581 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
6582 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
6583 | ret = -EINVAL; | |
6584 | goto out; | |
6585 | } | |
6586 | ||
6587 | /* No change? */ | |
6588 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
6589 | goto out; | |
c8e251fa | 6590 | |
364df0eb | 6591 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
6592 | unsigned int cpu; |
6593 | ||
22a7f12b | 6594 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
6595 | pageset_set_high_and_batch(zone, |
6596 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 6597 | } |
7cd2b0a3 | 6598 | out: |
c8e251fa | 6599 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 6600 | return ret; |
8ad4b1fb RS |
6601 | } |
6602 | ||
a9919c79 | 6603 | #ifdef CONFIG_NUMA |
f034b5d4 | 6604 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 | 6605 | |
1da177e4 LT |
6606 | static int __init set_hashdist(char *str) |
6607 | { | |
6608 | if (!str) | |
6609 | return 0; | |
6610 | hashdist = simple_strtoul(str, &str, 0); | |
6611 | return 1; | |
6612 | } | |
6613 | __setup("hashdist=", set_hashdist); | |
6614 | #endif | |
6615 | ||
6616 | /* | |
6617 | * allocate a large system hash table from bootmem | |
6618 | * - it is assumed that the hash table must contain an exact power-of-2 | |
6619 | * quantity of entries | |
6620 | * - limit is the number of hash buckets, not the total allocation size | |
6621 | */ | |
6622 | void *__init alloc_large_system_hash(const char *tablename, | |
6623 | unsigned long bucketsize, | |
6624 | unsigned long numentries, | |
6625 | int scale, | |
6626 | int flags, | |
6627 | unsigned int *_hash_shift, | |
6628 | unsigned int *_hash_mask, | |
31fe62b9 TB |
6629 | unsigned long low_limit, |
6630 | unsigned long high_limit) | |
1da177e4 | 6631 | { |
31fe62b9 | 6632 | unsigned long long max = high_limit; |
1da177e4 LT |
6633 | unsigned long log2qty, size; |
6634 | void *table = NULL; | |
6635 | ||
6636 | /* allow the kernel cmdline to have a say */ | |
6637 | if (!numentries) { | |
6638 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 6639 | numentries = nr_kernel_pages; |
a7e83318 JZ |
6640 | |
6641 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
6642 | if (PAGE_SHIFT < 20) | |
6643 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
6644 | |
6645 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
6646 | if (scale > PAGE_SHIFT) | |
6647 | numentries >>= (scale - PAGE_SHIFT); | |
6648 | else | |
6649 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
6650 | |
6651 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
6652 | if (unlikely(flags & HASH_SMALL)) { |
6653 | /* Makes no sense without HASH_EARLY */ | |
6654 | WARN_ON(!(flags & HASH_EARLY)); | |
6655 | if (!(numentries >> *_hash_shift)) { | |
6656 | numentries = 1UL << *_hash_shift; | |
6657 | BUG_ON(!numentries); | |
6658 | } | |
6659 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 6660 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 6661 | } |
6e692ed3 | 6662 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
6663 | |
6664 | /* limit allocation size to 1/16 total memory by default */ | |
6665 | if (max == 0) { | |
6666 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
6667 | do_div(max, bucketsize); | |
6668 | } | |
074b8517 | 6669 | max = min(max, 0x80000000ULL); |
1da177e4 | 6670 | |
31fe62b9 TB |
6671 | if (numentries < low_limit) |
6672 | numentries = low_limit; | |
1da177e4 LT |
6673 | if (numentries > max) |
6674 | numentries = max; | |
6675 | ||
f0d1b0b3 | 6676 | log2qty = ilog2(numentries); |
1da177e4 LT |
6677 | |
6678 | do { | |
6679 | size = bucketsize << log2qty; | |
6680 | if (flags & HASH_EARLY) | |
6782832e | 6681 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
6682 | else if (hashdist) |
6683 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
6684 | else { | |
1037b83b ED |
6685 | /* |
6686 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
6687 | * some pages at the end of hash table which |
6688 | * alloc_pages_exact() automatically does | |
1037b83b | 6689 | */ |
264ef8a9 | 6690 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 6691 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
6692 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
6693 | } | |
1da177e4 LT |
6694 | } |
6695 | } while (!table && size > PAGE_SIZE && --log2qty); | |
6696 | ||
6697 | if (!table) | |
6698 | panic("Failed to allocate %s hash table\n", tablename); | |
6699 | ||
f241e660 | 6700 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 6701 | tablename, |
f241e660 | 6702 | (1UL << log2qty), |
f0d1b0b3 | 6703 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
6704 | size); |
6705 | ||
6706 | if (_hash_shift) | |
6707 | *_hash_shift = log2qty; | |
6708 | if (_hash_mask) | |
6709 | *_hash_mask = (1 << log2qty) - 1; | |
6710 | ||
6711 | return table; | |
6712 | } | |
a117e66e | 6713 | |
835c134e MG |
6714 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
6715 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
6716 | unsigned long pfn) | |
6717 | { | |
6718 | #ifdef CONFIG_SPARSEMEM | |
6719 | return __pfn_to_section(pfn)->pageblock_flags; | |
6720 | #else | |
6721 | return zone->pageblock_flags; | |
6722 | #endif /* CONFIG_SPARSEMEM */ | |
6723 | } | |
6724 | ||
6725 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
6726 | { | |
6727 | #ifdef CONFIG_SPARSEMEM | |
6728 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 6729 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 6730 | #else |
c060f943 | 6731 | pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 6732 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
6733 | #endif /* CONFIG_SPARSEMEM */ |
6734 | } | |
6735 | ||
6736 | /** | |
1aab4d77 | 6737 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e | 6738 | * @page: The page within the block of interest |
1aab4d77 RD |
6739 | * @pfn: The target page frame number |
6740 | * @end_bitidx: The last bit of interest to retrieve | |
6741 | * @mask: mask of bits that the caller is interested in | |
6742 | * | |
6743 | * Return: pageblock_bits flags | |
835c134e | 6744 | */ |
dc4b0caf | 6745 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
e58469ba MG |
6746 | unsigned long end_bitidx, |
6747 | unsigned long mask) | |
835c134e MG |
6748 | { |
6749 | struct zone *zone; | |
6750 | unsigned long *bitmap; | |
dc4b0caf | 6751 | unsigned long bitidx, word_bitidx; |
e58469ba | 6752 | unsigned long word; |
835c134e MG |
6753 | |
6754 | zone = page_zone(page); | |
835c134e MG |
6755 | bitmap = get_pageblock_bitmap(zone, pfn); |
6756 | bitidx = pfn_to_bitidx(zone, pfn); | |
e58469ba MG |
6757 | word_bitidx = bitidx / BITS_PER_LONG; |
6758 | bitidx &= (BITS_PER_LONG-1); | |
835c134e | 6759 | |
e58469ba MG |
6760 | word = bitmap[word_bitidx]; |
6761 | bitidx += end_bitidx; | |
6762 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
835c134e MG |
6763 | } |
6764 | ||
6765 | /** | |
dc4b0caf | 6766 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e | 6767 | * @page: The page within the block of interest |
835c134e | 6768 | * @flags: The flags to set |
1aab4d77 RD |
6769 | * @pfn: The target page frame number |
6770 | * @end_bitidx: The last bit of interest | |
6771 | * @mask: mask of bits that the caller is interested in | |
835c134e | 6772 | */ |
dc4b0caf MG |
6773 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
6774 | unsigned long pfn, | |
e58469ba MG |
6775 | unsigned long end_bitidx, |
6776 | unsigned long mask) | |
835c134e MG |
6777 | { |
6778 | struct zone *zone; | |
6779 | unsigned long *bitmap; | |
dc4b0caf | 6780 | unsigned long bitidx, word_bitidx; |
e58469ba MG |
6781 | unsigned long old_word, word; |
6782 | ||
6783 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
835c134e MG |
6784 | |
6785 | zone = page_zone(page); | |
835c134e MG |
6786 | bitmap = get_pageblock_bitmap(zone, pfn); |
6787 | bitidx = pfn_to_bitidx(zone, pfn); | |
e58469ba MG |
6788 | word_bitidx = bitidx / BITS_PER_LONG; |
6789 | bitidx &= (BITS_PER_LONG-1); | |
6790 | ||
309381fe | 6791 | VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page); |
835c134e | 6792 | |
e58469ba MG |
6793 | bitidx += end_bitidx; |
6794 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
6795 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
6796 | ||
4db0c3c2 | 6797 | word = READ_ONCE(bitmap[word_bitidx]); |
e58469ba MG |
6798 | for (;;) { |
6799 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
6800 | if (word == old_word) | |
6801 | break; | |
6802 | word = old_word; | |
6803 | } | |
835c134e | 6804 | } |
a5d76b54 KH |
6805 | |
6806 | /* | |
80934513 MK |
6807 | * This function checks whether pageblock includes unmovable pages or not. |
6808 | * If @count is not zero, it is okay to include less @count unmovable pages | |
6809 | * | |
b8af2941 | 6810 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
6811 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
6812 | * expect this function should be exact. | |
a5d76b54 | 6813 | */ |
b023f468 WC |
6814 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
6815 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
6816 | { |
6817 | unsigned long pfn, iter, found; | |
47118af0 MN |
6818 | int mt; |
6819 | ||
49ac8255 KH |
6820 | /* |
6821 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 6822 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
6823 | */ |
6824 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 6825 | return false; |
47118af0 MN |
6826 | mt = get_pageblock_migratetype(page); |
6827 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 6828 | return false; |
49ac8255 KH |
6829 | |
6830 | pfn = page_to_pfn(page); | |
6831 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
6832 | unsigned long check = pfn + iter; | |
6833 | ||
29723fcc | 6834 | if (!pfn_valid_within(check)) |
49ac8255 | 6835 | continue; |
29723fcc | 6836 | |
49ac8255 | 6837 | page = pfn_to_page(check); |
c8721bbb NH |
6838 | |
6839 | /* | |
6840 | * Hugepages are not in LRU lists, but they're movable. | |
6841 | * We need not scan over tail pages bacause we don't | |
6842 | * handle each tail page individually in migration. | |
6843 | */ | |
6844 | if (PageHuge(page)) { | |
6845 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
6846 | continue; | |
6847 | } | |
6848 | ||
97d255c8 MK |
6849 | /* |
6850 | * We can't use page_count without pin a page | |
6851 | * because another CPU can free compound page. | |
6852 | * This check already skips compound tails of THP | |
6853 | * because their page->_count is zero at all time. | |
6854 | */ | |
fe896d18 | 6855 | if (!page_ref_count(page)) { |
49ac8255 KH |
6856 | if (PageBuddy(page)) |
6857 | iter += (1 << page_order(page)) - 1; | |
6858 | continue; | |
6859 | } | |
97d255c8 | 6860 | |
b023f468 WC |
6861 | /* |
6862 | * The HWPoisoned page may be not in buddy system, and | |
6863 | * page_count() is not 0. | |
6864 | */ | |
6865 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6866 | continue; | |
6867 | ||
49ac8255 KH |
6868 | if (!PageLRU(page)) |
6869 | found++; | |
6870 | /* | |
6b4f7799 JW |
6871 | * If there are RECLAIMABLE pages, we need to check |
6872 | * it. But now, memory offline itself doesn't call | |
6873 | * shrink_node_slabs() and it still to be fixed. | |
49ac8255 KH |
6874 | */ |
6875 | /* | |
6876 | * If the page is not RAM, page_count()should be 0. | |
6877 | * we don't need more check. This is an _used_ not-movable page. | |
6878 | * | |
6879 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6880 | * is set to both of a memory hole page and a _used_ kernel | |
6881 | * page at boot. | |
6882 | */ | |
6883 | if (found > count) | |
80934513 | 6884 | return true; |
49ac8255 | 6885 | } |
80934513 | 6886 | return false; |
49ac8255 KH |
6887 | } |
6888 | ||
6889 | bool is_pageblock_removable_nolock(struct page *page) | |
6890 | { | |
656a0706 MH |
6891 | struct zone *zone; |
6892 | unsigned long pfn; | |
687875fb MH |
6893 | |
6894 | /* | |
6895 | * We have to be careful here because we are iterating over memory | |
6896 | * sections which are not zone aware so we might end up outside of | |
6897 | * the zone but still within the section. | |
656a0706 MH |
6898 | * We have to take care about the node as well. If the node is offline |
6899 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 6900 | */ |
656a0706 MH |
6901 | if (!node_online(page_to_nid(page))) |
6902 | return false; | |
6903 | ||
6904 | zone = page_zone(page); | |
6905 | pfn = page_to_pfn(page); | |
108bcc96 | 6906 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
6907 | return false; |
6908 | ||
b023f468 | 6909 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6910 | } |
0c0e6195 | 6911 | |
080fe206 | 6912 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c MN |
6913 | |
6914 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6915 | { | |
6916 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6917 | pageblock_nr_pages) - 1); | |
6918 | } | |
6919 | ||
6920 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6921 | { | |
6922 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6923 | pageblock_nr_pages)); | |
6924 | } | |
6925 | ||
041d3a8c | 6926 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6927 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6928 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6929 | { |
6930 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6931 | unsigned long nr_reclaimed; |
041d3a8c MN |
6932 | unsigned long pfn = start; |
6933 | unsigned int tries = 0; | |
6934 | int ret = 0; | |
6935 | ||
be49a6e1 | 6936 | migrate_prep(); |
041d3a8c | 6937 | |
bb13ffeb | 6938 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6939 | if (fatal_signal_pending(current)) { |
6940 | ret = -EINTR; | |
6941 | break; | |
6942 | } | |
6943 | ||
bb13ffeb MG |
6944 | if (list_empty(&cc->migratepages)) { |
6945 | cc->nr_migratepages = 0; | |
edc2ca61 | 6946 | pfn = isolate_migratepages_range(cc, pfn, end); |
041d3a8c MN |
6947 | if (!pfn) { |
6948 | ret = -EINTR; | |
6949 | break; | |
6950 | } | |
6951 | tries = 0; | |
6952 | } else if (++tries == 5) { | |
6953 | ret = ret < 0 ? ret : -EBUSY; | |
6954 | break; | |
6955 | } | |
6956 | ||
beb51eaa MK |
6957 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6958 | &cc->migratepages); | |
6959 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6960 | |
9c620e2b | 6961 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 6962 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 6963 | } |
2a6f5124 SP |
6964 | if (ret < 0) { |
6965 | putback_movable_pages(&cc->migratepages); | |
6966 | return ret; | |
6967 | } | |
6968 | return 0; | |
041d3a8c MN |
6969 | } |
6970 | ||
6971 | /** | |
6972 | * alloc_contig_range() -- tries to allocate given range of pages | |
6973 | * @start: start PFN to allocate | |
6974 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
6975 | * @migratetype: migratetype of the underlaying pageblocks (either |
6976 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
6977 | * in range must have the same migratetype and it must | |
6978 | * be either of the two. | |
041d3a8c MN |
6979 | * |
6980 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
6981 | * aligned, however it's the caller's responsibility to guarantee that | |
6982 | * we are the only thread that changes migrate type of pageblocks the | |
6983 | * pages fall in. | |
6984 | * | |
6985 | * The PFN range must belong to a single zone. | |
6986 | * | |
6987 | * Returns zero on success or negative error code. On success all | |
6988 | * pages which PFN is in [start, end) are allocated for the caller and | |
6989 | * need to be freed with free_contig_range(). | |
6990 | */ | |
0815f3d8 MN |
6991 | int alloc_contig_range(unsigned long start, unsigned long end, |
6992 | unsigned migratetype) | |
041d3a8c | 6993 | { |
041d3a8c | 6994 | unsigned long outer_start, outer_end; |
d00181b9 KS |
6995 | unsigned int order; |
6996 | int ret = 0; | |
041d3a8c | 6997 | |
bb13ffeb MG |
6998 | struct compact_control cc = { |
6999 | .nr_migratepages = 0, | |
7000 | .order = -1, | |
7001 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 7002 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
7003 | .ignore_skip_hint = true, |
7004 | }; | |
7005 | INIT_LIST_HEAD(&cc.migratepages); | |
7006 | ||
041d3a8c MN |
7007 | /* |
7008 | * What we do here is we mark all pageblocks in range as | |
7009 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
7010 | * have different sizes, and due to the way page allocator | |
7011 | * work, we align the range to biggest of the two pages so | |
7012 | * that page allocator won't try to merge buddies from | |
7013 | * different pageblocks and change MIGRATE_ISOLATE to some | |
7014 | * other migration type. | |
7015 | * | |
7016 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
7017 | * migrate the pages from an unaligned range (ie. pages that | |
7018 | * we are interested in). This will put all the pages in | |
7019 | * range back to page allocator as MIGRATE_ISOLATE. | |
7020 | * | |
7021 | * When this is done, we take the pages in range from page | |
7022 | * allocator removing them from the buddy system. This way | |
7023 | * page allocator will never consider using them. | |
7024 | * | |
7025 | * This lets us mark the pageblocks back as | |
7026 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
7027 | * aligned range but not in the unaligned, original range are | |
7028 | * put back to page allocator so that buddy can use them. | |
7029 | */ | |
7030 | ||
7031 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
7032 | pfn_max_align_up(end), migratetype, |
7033 | false); | |
041d3a8c | 7034 | if (ret) |
86a595f9 | 7035 | return ret; |
041d3a8c | 7036 | |
8ef5849f JK |
7037 | /* |
7038 | * In case of -EBUSY, we'd like to know which page causes problem. | |
7039 | * So, just fall through. We will check it in test_pages_isolated(). | |
7040 | */ | |
bb13ffeb | 7041 | ret = __alloc_contig_migrate_range(&cc, start, end); |
8ef5849f | 7042 | if (ret && ret != -EBUSY) |
041d3a8c MN |
7043 | goto done; |
7044 | ||
7045 | /* | |
7046 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
7047 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
7048 | * more, all pages in [start, end) are free in page allocator. | |
7049 | * What we are going to do is to allocate all pages from | |
7050 | * [start, end) (that is remove them from page allocator). | |
7051 | * | |
7052 | * The only problem is that pages at the beginning and at the | |
7053 | * end of interesting range may be not aligned with pages that | |
7054 | * page allocator holds, ie. they can be part of higher order | |
7055 | * pages. Because of this, we reserve the bigger range and | |
7056 | * once this is done free the pages we are not interested in. | |
7057 | * | |
7058 | * We don't have to hold zone->lock here because the pages are | |
7059 | * isolated thus they won't get removed from buddy. | |
7060 | */ | |
7061 | ||
7062 | lru_add_drain_all(); | |
510f5507 | 7063 | drain_all_pages(cc.zone); |
041d3a8c MN |
7064 | |
7065 | order = 0; | |
7066 | outer_start = start; | |
7067 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
7068 | if (++order >= MAX_ORDER) { | |
8ef5849f JK |
7069 | outer_start = start; |
7070 | break; | |
041d3a8c MN |
7071 | } |
7072 | outer_start &= ~0UL << order; | |
7073 | } | |
7074 | ||
8ef5849f JK |
7075 | if (outer_start != start) { |
7076 | order = page_order(pfn_to_page(outer_start)); | |
7077 | ||
7078 | /* | |
7079 | * outer_start page could be small order buddy page and | |
7080 | * it doesn't include start page. Adjust outer_start | |
7081 | * in this case to report failed page properly | |
7082 | * on tracepoint in test_pages_isolated() | |
7083 | */ | |
7084 | if (outer_start + (1UL << order) <= start) | |
7085 | outer_start = start; | |
7086 | } | |
7087 | ||
041d3a8c | 7088 | /* Make sure the range is really isolated. */ |
b023f468 | 7089 | if (test_pages_isolated(outer_start, end, false)) { |
dae803e1 MN |
7090 | pr_info("%s: [%lx, %lx) PFNs busy\n", |
7091 | __func__, outer_start, end); | |
041d3a8c MN |
7092 | ret = -EBUSY; |
7093 | goto done; | |
7094 | } | |
7095 | ||
49f223a9 | 7096 | /* Grab isolated pages from freelists. */ |
bb13ffeb | 7097 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
7098 | if (!outer_end) { |
7099 | ret = -EBUSY; | |
7100 | goto done; | |
7101 | } | |
7102 | ||
7103 | /* Free head and tail (if any) */ | |
7104 | if (start != outer_start) | |
7105 | free_contig_range(outer_start, start - outer_start); | |
7106 | if (end != outer_end) | |
7107 | free_contig_range(end, outer_end - end); | |
7108 | ||
7109 | done: | |
7110 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 7111 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
7112 | return ret; |
7113 | } | |
7114 | ||
7115 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
7116 | { | |
bcc2b02f MS |
7117 | unsigned int count = 0; |
7118 | ||
7119 | for (; nr_pages--; pfn++) { | |
7120 | struct page *page = pfn_to_page(pfn); | |
7121 | ||
7122 | count += page_count(page) != 1; | |
7123 | __free_page(page); | |
7124 | } | |
7125 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
7126 | } |
7127 | #endif | |
7128 | ||
4ed7e022 | 7129 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
7130 | /* |
7131 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
7132 | * page high values need to be recalulated. | |
7133 | */ | |
4ed7e022 JL |
7134 | void __meminit zone_pcp_update(struct zone *zone) |
7135 | { | |
0a647f38 | 7136 | unsigned cpu; |
c8e251fa | 7137 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 7138 | for_each_possible_cpu(cpu) |
169f6c19 CS |
7139 | pageset_set_high_and_batch(zone, |
7140 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 7141 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
7142 | } |
7143 | #endif | |
7144 | ||
340175b7 JL |
7145 | void zone_pcp_reset(struct zone *zone) |
7146 | { | |
7147 | unsigned long flags; | |
5a883813 MK |
7148 | int cpu; |
7149 | struct per_cpu_pageset *pset; | |
340175b7 JL |
7150 | |
7151 | /* avoid races with drain_pages() */ | |
7152 | local_irq_save(flags); | |
7153 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
7154 | for_each_online_cpu(cpu) { |
7155 | pset = per_cpu_ptr(zone->pageset, cpu); | |
7156 | drain_zonestat(zone, pset); | |
7157 | } | |
340175b7 JL |
7158 | free_percpu(zone->pageset); |
7159 | zone->pageset = &boot_pageset; | |
7160 | } | |
7161 | local_irq_restore(flags); | |
7162 | } | |
7163 | ||
6dcd73d7 | 7164 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
7165 | /* |
7166 | * All pages in the range must be isolated before calling this. | |
7167 | */ | |
7168 | void | |
7169 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
7170 | { | |
7171 | struct page *page; | |
7172 | struct zone *zone; | |
7aeb09f9 | 7173 | unsigned int order, i; |
0c0e6195 KH |
7174 | unsigned long pfn; |
7175 | unsigned long flags; | |
7176 | /* find the first valid pfn */ | |
7177 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
7178 | if (pfn_valid(pfn)) | |
7179 | break; | |
7180 | if (pfn == end_pfn) | |
7181 | return; | |
7182 | zone = page_zone(pfn_to_page(pfn)); | |
7183 | spin_lock_irqsave(&zone->lock, flags); | |
7184 | pfn = start_pfn; | |
7185 | while (pfn < end_pfn) { | |
7186 | if (!pfn_valid(pfn)) { | |
7187 | pfn++; | |
7188 | continue; | |
7189 | } | |
7190 | page = pfn_to_page(pfn); | |
b023f468 WC |
7191 | /* |
7192 | * The HWPoisoned page may be not in buddy system, and | |
7193 | * page_count() is not 0. | |
7194 | */ | |
7195 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
7196 | pfn++; | |
7197 | SetPageReserved(page); | |
7198 | continue; | |
7199 | } | |
7200 | ||
0c0e6195 KH |
7201 | BUG_ON(page_count(page)); |
7202 | BUG_ON(!PageBuddy(page)); | |
7203 | order = page_order(page); | |
7204 | #ifdef CONFIG_DEBUG_VM | |
7205 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
7206 | pfn, 1 << order, end_pfn); | |
7207 | #endif | |
7208 | list_del(&page->lru); | |
7209 | rmv_page_order(page); | |
7210 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
7211 | for (i = 0; i < (1 << order); i++) |
7212 | SetPageReserved((page+i)); | |
7213 | pfn += (1 << order); | |
7214 | } | |
7215 | spin_unlock_irqrestore(&zone->lock, flags); | |
7216 | } | |
7217 | #endif | |
8d22ba1b | 7218 | |
8d22ba1b WF |
7219 | bool is_free_buddy_page(struct page *page) |
7220 | { | |
7221 | struct zone *zone = page_zone(page); | |
7222 | unsigned long pfn = page_to_pfn(page); | |
7223 | unsigned long flags; | |
7aeb09f9 | 7224 | unsigned int order; |
8d22ba1b WF |
7225 | |
7226 | spin_lock_irqsave(&zone->lock, flags); | |
7227 | for (order = 0; order < MAX_ORDER; order++) { | |
7228 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
7229 | ||
7230 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
7231 | break; | |
7232 | } | |
7233 | spin_unlock_irqrestore(&zone->lock, flags); | |
7234 | ||
7235 | return order < MAX_ORDER; | |
7236 | } |