]>
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> |
1da177e4 LT |
28 | #include <linux/module.h> |
29 | #include <linux/suspend.h> | |
30 | #include <linux/pagevec.h> | |
31 | #include <linux/blkdev.h> | |
32 | #include <linux/slab.h> | |
a238ab5b | 33 | #include <linux/ratelimit.h> |
5a3135c2 | 34 | #include <linux/oom.h> |
1da177e4 LT |
35 | #include <linux/notifier.h> |
36 | #include <linux/topology.h> | |
37 | #include <linux/sysctl.h> | |
38 | #include <linux/cpu.h> | |
39 | #include <linux/cpuset.h> | |
bdc8cb98 | 40 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
41 | #include <linux/nodemask.h> |
42 | #include <linux/vmalloc.h> | |
a6cccdc3 | 43 | #include <linux/vmstat.h> |
4be38e35 | 44 | #include <linux/mempolicy.h> |
6811378e | 45 | #include <linux/stop_machine.h> |
c713216d MG |
46 | #include <linux/sort.h> |
47 | #include <linux/pfn.h> | |
3fcfab16 | 48 | #include <linux/backing-dev.h> |
933e312e | 49 | #include <linux/fault-inject.h> |
a5d76b54 | 50 | #include <linux/page-isolation.h> |
52d4b9ac | 51 | #include <linux/page_cgroup.h> |
3ac7fe5a | 52 | #include <linux/debugobjects.h> |
dbb1f81c | 53 | #include <linux/kmemleak.h> |
56de7263 | 54 | #include <linux/compaction.h> |
0d3d062a | 55 | #include <trace/events/kmem.h> |
718a3821 | 56 | #include <linux/ftrace_event.h> |
f212ad7c | 57 | #include <linux/memcontrol.h> |
268bb0ce | 58 | #include <linux/prefetch.h> |
6e543d57 | 59 | #include <linux/mm_inline.h> |
041d3a8c | 60 | #include <linux/migrate.h> |
c0a32fc5 | 61 | #include <linux/page-debug-flags.h> |
949f7ec5 | 62 | #include <linux/hugetlb.h> |
8bd75c77 | 63 | #include <linux/sched/rt.h> |
1da177e4 | 64 | |
7ee3d4e8 | 65 | #include <asm/sections.h> |
1da177e4 | 66 | #include <asm/tlbflush.h> |
ac924c60 | 67 | #include <asm/div64.h> |
1da177e4 LT |
68 | #include "internal.h" |
69 | ||
c8e251fa CS |
70 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
71 | static DEFINE_MUTEX(pcp_batch_high_lock); | |
72 | ||
72812019 LS |
73 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
74 | DEFINE_PER_CPU(int, numa_node); | |
75 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
76 | #endif | |
77 | ||
7aac7898 LS |
78 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
79 | /* | |
80 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
81 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
82 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
83 | * defined in <linux/topology.h>. | |
84 | */ | |
85 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
86 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
87 | #endif | |
88 | ||
1da177e4 | 89 | /* |
13808910 | 90 | * Array of node states. |
1da177e4 | 91 | */ |
13808910 CL |
92 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
93 | [N_POSSIBLE] = NODE_MASK_ALL, | |
94 | [N_ONLINE] = { { [0] = 1UL } }, | |
95 | #ifndef CONFIG_NUMA | |
96 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
97 | #ifdef CONFIG_HIGHMEM | |
98 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
99 | #endif |
100 | #ifdef CONFIG_MOVABLE_NODE | |
101 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
102 | #endif |
103 | [N_CPU] = { { [0] = 1UL } }, | |
104 | #endif /* NUMA */ | |
105 | }; | |
106 | EXPORT_SYMBOL(node_states); | |
107 | ||
c3d5f5f0 JL |
108 | /* Protect totalram_pages and zone->managed_pages */ |
109 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
110 | ||
6c231b7b | 111 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 112 | unsigned long totalreserve_pages __read_mostly; |
ab8fabd4 JW |
113 | /* |
114 | * When calculating the number of globally allowed dirty pages, there | |
115 | * is a certain number of per-zone reserves that should not be | |
116 | * considered dirtyable memory. This is the sum of those reserves | |
117 | * over all existing zones that contribute dirtyable memory. | |
118 | */ | |
119 | unsigned long dirty_balance_reserve __read_mostly; | |
120 | ||
1b76b02f | 121 | int percpu_pagelist_fraction; |
dcce284a | 122 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 123 | |
452aa699 RW |
124 | #ifdef CONFIG_PM_SLEEP |
125 | /* | |
126 | * The following functions are used by the suspend/hibernate code to temporarily | |
127 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
128 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
129 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
130 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
131 | * guaranteed not to run in parallel with that modification). | |
132 | */ | |
c9e664f1 RW |
133 | |
134 | static gfp_t saved_gfp_mask; | |
135 | ||
136 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
137 | { |
138 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
139 | if (saved_gfp_mask) { |
140 | gfp_allowed_mask = saved_gfp_mask; | |
141 | saved_gfp_mask = 0; | |
142 | } | |
452aa699 RW |
143 | } |
144 | ||
c9e664f1 | 145 | void pm_restrict_gfp_mask(void) |
452aa699 | 146 | { |
452aa699 | 147 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
148 | WARN_ON(saved_gfp_mask); |
149 | saved_gfp_mask = gfp_allowed_mask; | |
150 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 | 151 | } |
f90ac398 MG |
152 | |
153 | bool pm_suspended_storage(void) | |
154 | { | |
155 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) | |
156 | return false; | |
157 | return true; | |
158 | } | |
452aa699 RW |
159 | #endif /* CONFIG_PM_SLEEP */ |
160 | ||
d9c23400 MG |
161 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
162 | int pageblock_order __read_mostly; | |
163 | #endif | |
164 | ||
d98c7a09 | 165 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 166 | |
1da177e4 LT |
167 | /* |
168 | * results with 256, 32 in the lowmem_reserve sysctl: | |
169 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
170 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
171 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
172 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
173 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
174 | * |
175 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
176 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 177 | */ |
2f1b6248 | 178 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 179 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 180 | 256, |
4b51d669 | 181 | #endif |
fb0e7942 | 182 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 183 | 256, |
fb0e7942 | 184 | #endif |
e53ef38d | 185 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 186 | 32, |
e53ef38d | 187 | #endif |
2a1e274a | 188 | 32, |
2f1b6248 | 189 | }; |
1da177e4 LT |
190 | |
191 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 192 | |
15ad7cdc | 193 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 194 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 195 | "DMA", |
4b51d669 | 196 | #endif |
fb0e7942 | 197 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 198 | "DMA32", |
fb0e7942 | 199 | #endif |
2f1b6248 | 200 | "Normal", |
e53ef38d | 201 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 202 | "HighMem", |
e53ef38d | 203 | #endif |
2a1e274a | 204 | "Movable", |
2f1b6248 CL |
205 | }; |
206 | ||
1da177e4 | 207 | int min_free_kbytes = 1024; |
42aa83cb | 208 | int user_min_free_kbytes = -1; |
1da177e4 | 209 | |
2c85f51d JB |
210 | static unsigned long __meminitdata nr_kernel_pages; |
211 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 212 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 213 | |
0ee332c1 TH |
214 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
215 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
216 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
217 | static unsigned long __initdata required_kernelcore; | |
218 | static unsigned long __initdata required_movablecore; | |
219 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
220 | ||
221 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
222 | int movable_zone; | |
223 | EXPORT_SYMBOL(movable_zone); | |
224 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 225 | |
418508c1 MS |
226 | #if MAX_NUMNODES > 1 |
227 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 228 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 229 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 230 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
231 | #endif |
232 | ||
9ef9acb0 MG |
233 | int page_group_by_mobility_disabled __read_mostly; |
234 | ||
ee6f509c | 235 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 236 | { |
5d0f3f72 KM |
237 | if (unlikely(page_group_by_mobility_disabled && |
238 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
239 | migratetype = MIGRATE_UNMOVABLE; |
240 | ||
b2a0ac88 MG |
241 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
242 | PB_migrate, PB_migrate_end); | |
243 | } | |
244 | ||
7f33d49a RW |
245 | bool oom_killer_disabled __read_mostly; |
246 | ||
13e7444b | 247 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 248 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 249 | { |
bdc8cb98 DH |
250 | int ret = 0; |
251 | unsigned seq; | |
252 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 253 | unsigned long sp, start_pfn; |
c6a57e19 | 254 | |
bdc8cb98 DH |
255 | do { |
256 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
257 | start_pfn = zone->zone_start_pfn; |
258 | sp = zone->spanned_pages; | |
108bcc96 | 259 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
260 | ret = 1; |
261 | } while (zone_span_seqretry(zone, seq)); | |
262 | ||
b5e6a5a2 | 263 | if (ret) |
613813e8 DH |
264 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
265 | pfn, zone_to_nid(zone), zone->name, | |
266 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 267 | |
bdc8cb98 | 268 | return ret; |
c6a57e19 DH |
269 | } |
270 | ||
271 | static int page_is_consistent(struct zone *zone, struct page *page) | |
272 | { | |
14e07298 | 273 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 274 | return 0; |
1da177e4 | 275 | if (zone != page_zone(page)) |
c6a57e19 DH |
276 | return 0; |
277 | ||
278 | return 1; | |
279 | } | |
280 | /* | |
281 | * Temporary debugging check for pages not lying within a given zone. | |
282 | */ | |
283 | static int bad_range(struct zone *zone, struct page *page) | |
284 | { | |
285 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 286 | return 1; |
c6a57e19 DH |
287 | if (!page_is_consistent(zone, page)) |
288 | return 1; | |
289 | ||
1da177e4 LT |
290 | return 0; |
291 | } | |
13e7444b NP |
292 | #else |
293 | static inline int bad_range(struct zone *zone, struct page *page) | |
294 | { | |
295 | return 0; | |
296 | } | |
297 | #endif | |
298 | ||
d230dec1 KS |
299 | static void bad_page(struct page *page, const char *reason, |
300 | unsigned long bad_flags) | |
1da177e4 | 301 | { |
d936cf9b HD |
302 | static unsigned long resume; |
303 | static unsigned long nr_shown; | |
304 | static unsigned long nr_unshown; | |
305 | ||
2a7684a2 WF |
306 | /* Don't complain about poisoned pages */ |
307 | if (PageHWPoison(page)) { | |
22b751c3 | 308 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
309 | return; |
310 | } | |
311 | ||
d936cf9b HD |
312 | /* |
313 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
314 | * or allow a steady drip of one report per second. | |
315 | */ | |
316 | if (nr_shown == 60) { | |
317 | if (time_before(jiffies, resume)) { | |
318 | nr_unshown++; | |
319 | goto out; | |
320 | } | |
321 | if (nr_unshown) { | |
1e9e6365 HD |
322 | printk(KERN_ALERT |
323 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
324 | nr_unshown); |
325 | nr_unshown = 0; | |
326 | } | |
327 | nr_shown = 0; | |
328 | } | |
329 | if (nr_shown++ == 0) | |
330 | resume = jiffies + 60 * HZ; | |
331 | ||
1e9e6365 | 332 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 333 | current->comm, page_to_pfn(page)); |
f0b791a3 | 334 | dump_page_badflags(page, reason, bad_flags); |
3dc14741 | 335 | |
4f31888c | 336 | print_modules(); |
1da177e4 | 337 | dump_stack(); |
d936cf9b | 338 | out: |
8cc3b392 | 339 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 340 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 341 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
342 | } |
343 | ||
1da177e4 LT |
344 | /* |
345 | * Higher-order pages are called "compound pages". They are structured thusly: | |
346 | * | |
347 | * The first PAGE_SIZE page is called the "head page". | |
348 | * | |
349 | * The remaining PAGE_SIZE pages are called "tail pages". | |
350 | * | |
6416b9fa WSH |
351 | * All pages have PG_compound set. All tail pages have their ->first_page |
352 | * pointing at the head page. | |
1da177e4 | 353 | * |
41d78ba5 HD |
354 | * The first tail page's ->lru.next holds the address of the compound page's |
355 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
356 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 357 | */ |
d98c7a09 HD |
358 | |
359 | static void free_compound_page(struct page *page) | |
360 | { | |
d85f3385 | 361 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
362 | } |
363 | ||
01ad1c08 | 364 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
365 | { |
366 | int i; | |
367 | int nr_pages = 1 << order; | |
368 | ||
369 | set_compound_page_dtor(page, free_compound_page); | |
370 | set_compound_order(page, order); | |
371 | __SetPageHead(page); | |
372 | for (i = 1; i < nr_pages; i++) { | |
373 | struct page *p = page + i; | |
58a84aa9 | 374 | set_page_count(p, 0); |
18229df5 | 375 | p->first_page = page; |
668f9abb DR |
376 | /* Make sure p->first_page is always valid for PageTail() */ |
377 | smp_wmb(); | |
378 | __SetPageTail(p); | |
18229df5 AW |
379 | } |
380 | } | |
381 | ||
59ff4216 | 382 | /* update __split_huge_page_refcount if you change this function */ |
8cc3b392 | 383 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
384 | { |
385 | int i; | |
386 | int nr_pages = 1 << order; | |
8cc3b392 | 387 | int bad = 0; |
1da177e4 | 388 | |
0bb2c763 | 389 | if (unlikely(compound_order(page) != order)) { |
f0b791a3 | 390 | bad_page(page, "wrong compound order", 0); |
8cc3b392 HD |
391 | bad++; |
392 | } | |
1da177e4 | 393 | |
6d777953 | 394 | __ClearPageHead(page); |
8cc3b392 | 395 | |
18229df5 AW |
396 | for (i = 1; i < nr_pages; i++) { |
397 | struct page *p = page + i; | |
1da177e4 | 398 | |
f0b791a3 DH |
399 | if (unlikely(!PageTail(p))) { |
400 | bad_page(page, "PageTail not set", 0); | |
401 | bad++; | |
402 | } else if (unlikely(p->first_page != page)) { | |
403 | bad_page(page, "first_page not consistent", 0); | |
8cc3b392 HD |
404 | bad++; |
405 | } | |
d85f3385 | 406 | __ClearPageTail(p); |
1da177e4 | 407 | } |
8cc3b392 HD |
408 | |
409 | return bad; | |
1da177e4 | 410 | } |
1da177e4 | 411 | |
17cf4406 NP |
412 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
413 | { | |
414 | int i; | |
415 | ||
6626c5d5 AM |
416 | /* |
417 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
418 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
419 | */ | |
725d704e | 420 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
421 | for (i = 0; i < (1 << order); i++) |
422 | clear_highpage(page + i); | |
423 | } | |
424 | ||
c0a32fc5 SG |
425 | #ifdef CONFIG_DEBUG_PAGEALLOC |
426 | unsigned int _debug_guardpage_minorder; | |
427 | ||
428 | static int __init debug_guardpage_minorder_setup(char *buf) | |
429 | { | |
430 | unsigned long res; | |
431 | ||
432 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
433 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
434 | return 0; | |
435 | } | |
436 | _debug_guardpage_minorder = res; | |
437 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
438 | return 0; | |
439 | } | |
440 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
441 | ||
442 | static inline void set_page_guard_flag(struct page *page) | |
443 | { | |
444 | __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
445 | } | |
446 | ||
447 | static inline void clear_page_guard_flag(struct page *page) | |
448 | { | |
449 | __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
450 | } | |
451 | #else | |
452 | static inline void set_page_guard_flag(struct page *page) { } | |
453 | static inline void clear_page_guard_flag(struct page *page) { } | |
454 | #endif | |
455 | ||
6aa3001b AM |
456 | static inline void set_page_order(struct page *page, int order) |
457 | { | |
4c21e2f2 | 458 | set_page_private(page, order); |
676165a8 | 459 | __SetPageBuddy(page); |
1da177e4 LT |
460 | } |
461 | ||
462 | static inline void rmv_page_order(struct page *page) | |
463 | { | |
676165a8 | 464 | __ClearPageBuddy(page); |
4c21e2f2 | 465 | set_page_private(page, 0); |
1da177e4 LT |
466 | } |
467 | ||
468 | /* | |
469 | * Locate the struct page for both the matching buddy in our | |
470 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
471 | * | |
472 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
473 | * the following equation: | |
474 | * B2 = B1 ^ (1 << O) | |
475 | * For example, if the starting buddy (buddy2) is #8 its order | |
476 | * 1 buddy is #10: | |
477 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
478 | * | |
479 | * 2) Any buddy B will have an order O+1 parent P which | |
480 | * satisfies the following equation: | |
481 | * P = B & ~(1 << O) | |
482 | * | |
d6e05edc | 483 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 | 484 | */ |
1da177e4 | 485 | static inline unsigned long |
43506fad | 486 | __find_buddy_index(unsigned long page_idx, unsigned int order) |
1da177e4 | 487 | { |
43506fad | 488 | return page_idx ^ (1 << order); |
1da177e4 LT |
489 | } |
490 | ||
491 | /* | |
492 | * This function checks whether a page is free && is the buddy | |
493 | * we can do coalesce a page and its buddy if | |
13e7444b | 494 | * (a) the buddy is not in a hole && |
676165a8 | 495 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
496 | * (c) a page and its buddy have the same order && |
497 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 498 | * |
cf6fe945 WSH |
499 | * For recording whether a page is in the buddy system, we set ->_mapcount |
500 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
501 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
502 | * serialized by zone->lock. | |
1da177e4 | 503 | * |
676165a8 | 504 | * For recording page's order, we use page_private(page). |
1da177e4 | 505 | */ |
cb2b95e1 AW |
506 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
507 | int order) | |
1da177e4 | 508 | { |
14e07298 | 509 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 510 | return 0; |
13e7444b | 511 | |
c0a32fc5 | 512 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
309381fe | 513 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
d34c5fa0 MG |
514 | |
515 | if (page_zone_id(page) != page_zone_id(buddy)) | |
516 | return 0; | |
517 | ||
c0a32fc5 SG |
518 | return 1; |
519 | } | |
520 | ||
cb2b95e1 | 521 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
309381fe | 522 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
d34c5fa0 MG |
523 | |
524 | /* | |
525 | * zone check is done late to avoid uselessly | |
526 | * calculating zone/node ids for pages that could | |
527 | * never merge. | |
528 | */ | |
529 | if (page_zone_id(page) != page_zone_id(buddy)) | |
530 | return 0; | |
531 | ||
6aa3001b | 532 | return 1; |
676165a8 | 533 | } |
6aa3001b | 534 | return 0; |
1da177e4 LT |
535 | } |
536 | ||
537 | /* | |
538 | * Freeing function for a buddy system allocator. | |
539 | * | |
540 | * The concept of a buddy system is to maintain direct-mapped table | |
541 | * (containing bit values) for memory blocks of various "orders". | |
542 | * The bottom level table contains the map for the smallest allocatable | |
543 | * units of memory (here, pages), and each level above it describes | |
544 | * pairs of units from the levels below, hence, "buddies". | |
545 | * At a high level, all that happens here is marking the table entry | |
546 | * at the bottom level available, and propagating the changes upward | |
547 | * as necessary, plus some accounting needed to play nicely with other | |
548 | * parts of the VM system. | |
549 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
550 | * free pages of length of (1 << order) and marked with _mapcount |
551 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
552 | * field. | |
1da177e4 | 553 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
554 | * other. That is, if we allocate a small block, and both were |
555 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 556 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 557 | * triggers coalescing into a block of larger size. |
1da177e4 | 558 | * |
6d49e352 | 559 | * -- nyc |
1da177e4 LT |
560 | */ |
561 | ||
48db57f8 | 562 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
563 | struct zone *zone, unsigned int order, |
564 | int migratetype) | |
1da177e4 LT |
565 | { |
566 | unsigned long page_idx; | |
6dda9d55 | 567 | unsigned long combined_idx; |
43506fad | 568 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 569 | struct page *buddy; |
1da177e4 | 570 | |
d29bb978 CS |
571 | VM_BUG_ON(!zone_is_initialized(zone)); |
572 | ||
224abf92 | 573 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
574 | if (unlikely(destroy_compound_page(page, order))) |
575 | return; | |
1da177e4 | 576 | |
ed0ae21d MG |
577 | VM_BUG_ON(migratetype == -1); |
578 | ||
1da177e4 LT |
579 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
580 | ||
309381fe SL |
581 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
582 | VM_BUG_ON_PAGE(bad_range(zone, page), page); | |
1da177e4 | 583 | |
1da177e4 | 584 | while (order < MAX_ORDER-1) { |
43506fad KC |
585 | buddy_idx = __find_buddy_index(page_idx, order); |
586 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 587 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 588 | break; |
c0a32fc5 SG |
589 | /* |
590 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
591 | * merge with it and move up one order. | |
592 | */ | |
593 | if (page_is_guard(buddy)) { | |
594 | clear_page_guard_flag(buddy); | |
595 | set_page_private(page, 0); | |
d1ce749a BZ |
596 | __mod_zone_freepage_state(zone, 1 << order, |
597 | migratetype); | |
c0a32fc5 SG |
598 | } else { |
599 | list_del(&buddy->lru); | |
600 | zone->free_area[order].nr_free--; | |
601 | rmv_page_order(buddy); | |
602 | } | |
43506fad | 603 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
604 | page = page + (combined_idx - page_idx); |
605 | page_idx = combined_idx; | |
606 | order++; | |
607 | } | |
608 | set_page_order(page, order); | |
6dda9d55 CZ |
609 | |
610 | /* | |
611 | * If this is not the largest possible page, check if the buddy | |
612 | * of the next-highest order is free. If it is, it's possible | |
613 | * that pages are being freed that will coalesce soon. In case, | |
614 | * that is happening, add the free page to the tail of the list | |
615 | * so it's less likely to be used soon and more likely to be merged | |
616 | * as a higher order page | |
617 | */ | |
b7f50cfa | 618 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 619 | struct page *higher_page, *higher_buddy; |
43506fad KC |
620 | combined_idx = buddy_idx & page_idx; |
621 | higher_page = page + (combined_idx - page_idx); | |
622 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 623 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
624 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
625 | list_add_tail(&page->lru, | |
626 | &zone->free_area[order].free_list[migratetype]); | |
627 | goto out; | |
628 | } | |
629 | } | |
630 | ||
631 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
632 | out: | |
1da177e4 LT |
633 | zone->free_area[order].nr_free++; |
634 | } | |
635 | ||
224abf92 | 636 | static inline int free_pages_check(struct page *page) |
1da177e4 | 637 | { |
d230dec1 | 638 | const char *bad_reason = NULL; |
f0b791a3 DH |
639 | unsigned long bad_flags = 0; |
640 | ||
641 | if (unlikely(page_mapcount(page))) | |
642 | bad_reason = "nonzero mapcount"; | |
643 | if (unlikely(page->mapping != NULL)) | |
644 | bad_reason = "non-NULL mapping"; | |
645 | if (unlikely(atomic_read(&page->_count) != 0)) | |
646 | bad_reason = "nonzero _count"; | |
647 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { | |
648 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
649 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
650 | } | |
651 | if (unlikely(mem_cgroup_bad_page_check(page))) | |
652 | bad_reason = "cgroup check failed"; | |
653 | if (unlikely(bad_reason)) { | |
654 | bad_page(page, bad_reason, bad_flags); | |
79f4b7bf | 655 | return 1; |
8cc3b392 | 656 | } |
90572890 | 657 | page_cpupid_reset_last(page); |
79f4b7bf HD |
658 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
659 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
660 | return 0; | |
1da177e4 LT |
661 | } |
662 | ||
663 | /* | |
5f8dcc21 | 664 | * Frees a number of pages from the PCP lists |
1da177e4 | 665 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 666 | * count is the number of pages to free. |
1da177e4 LT |
667 | * |
668 | * If the zone was previously in an "all pages pinned" state then look to | |
669 | * see if this freeing clears that state. | |
670 | * | |
671 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
672 | * pinned" detection logic. | |
673 | */ | |
5f8dcc21 MG |
674 | static void free_pcppages_bulk(struct zone *zone, int count, |
675 | struct per_cpu_pages *pcp) | |
1da177e4 | 676 | { |
5f8dcc21 | 677 | int migratetype = 0; |
a6f9edd6 | 678 | int batch_free = 0; |
72853e29 | 679 | int to_free = count; |
5f8dcc21 | 680 | |
c54ad30c | 681 | spin_lock(&zone->lock); |
1da177e4 | 682 | zone->pages_scanned = 0; |
f2260e6b | 683 | |
72853e29 | 684 | while (to_free) { |
48db57f8 | 685 | struct page *page; |
5f8dcc21 MG |
686 | struct list_head *list; |
687 | ||
688 | /* | |
a6f9edd6 MG |
689 | * Remove pages from lists in a round-robin fashion. A |
690 | * batch_free count is maintained that is incremented when an | |
691 | * empty list is encountered. This is so more pages are freed | |
692 | * off fuller lists instead of spinning excessively around empty | |
693 | * lists | |
5f8dcc21 MG |
694 | */ |
695 | do { | |
a6f9edd6 | 696 | batch_free++; |
5f8dcc21 MG |
697 | if (++migratetype == MIGRATE_PCPTYPES) |
698 | migratetype = 0; | |
699 | list = &pcp->lists[migratetype]; | |
700 | } while (list_empty(list)); | |
48db57f8 | 701 | |
1d16871d NK |
702 | /* This is the only non-empty list. Free them all. */ |
703 | if (batch_free == MIGRATE_PCPTYPES) | |
704 | batch_free = to_free; | |
705 | ||
a6f9edd6 | 706 | do { |
770c8aaa BZ |
707 | int mt; /* migratetype of the to-be-freed page */ |
708 | ||
a6f9edd6 MG |
709 | page = list_entry(list->prev, struct page, lru); |
710 | /* must delete as __free_one_page list manipulates */ | |
711 | list_del(&page->lru); | |
b12c4ad1 | 712 | mt = get_freepage_migratetype(page); |
a7016235 | 713 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
770c8aaa BZ |
714 | __free_one_page(page, zone, 0, mt); |
715 | trace_mm_page_pcpu_drain(page, 0, mt); | |
194159fb | 716 | if (likely(!is_migrate_isolate_page(page))) { |
97d0da22 WC |
717 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1); |
718 | if (is_migrate_cma(mt)) | |
719 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1); | |
720 | } | |
72853e29 | 721 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 722 | } |
c54ad30c | 723 | spin_unlock(&zone->lock); |
1da177e4 LT |
724 | } |
725 | ||
ed0ae21d MG |
726 | static void free_one_page(struct zone *zone, struct page *page, int order, |
727 | int migratetype) | |
1da177e4 | 728 | { |
006d22d9 | 729 | spin_lock(&zone->lock); |
006d22d9 | 730 | zone->pages_scanned = 0; |
f2260e6b | 731 | |
ed0ae21d | 732 | __free_one_page(page, zone, order, migratetype); |
194159fb | 733 | if (unlikely(!is_migrate_isolate(migratetype))) |
d1ce749a | 734 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
006d22d9 | 735 | spin_unlock(&zone->lock); |
48db57f8 NP |
736 | } |
737 | ||
ec95f53a | 738 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 739 | { |
1da177e4 | 740 | int i; |
8cc3b392 | 741 | int bad = 0; |
1da177e4 | 742 | |
b413d48a | 743 | trace_mm_page_free(page, order); |
b1eeab67 VN |
744 | kmemcheck_free_shadow(page, order); |
745 | ||
8dd60a3a AA |
746 | if (PageAnon(page)) |
747 | page->mapping = NULL; | |
748 | for (i = 0; i < (1 << order); i++) | |
749 | bad += free_pages_check(page + i); | |
8cc3b392 | 750 | if (bad) |
ec95f53a | 751 | return false; |
689bcebf | 752 | |
3ac7fe5a | 753 | if (!PageHighMem(page)) { |
b8af2941 PK |
754 | debug_check_no_locks_freed(page_address(page), |
755 | PAGE_SIZE << order); | |
3ac7fe5a TG |
756 | debug_check_no_obj_freed(page_address(page), |
757 | PAGE_SIZE << order); | |
758 | } | |
dafb1367 | 759 | arch_free_page(page, order); |
48db57f8 | 760 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 761 | |
ec95f53a KM |
762 | return true; |
763 | } | |
764 | ||
765 | static void __free_pages_ok(struct page *page, unsigned int order) | |
766 | { | |
767 | unsigned long flags; | |
95e34412 | 768 | int migratetype; |
ec95f53a KM |
769 | |
770 | if (!free_pages_prepare(page, order)) | |
771 | return; | |
772 | ||
c54ad30c | 773 | local_irq_save(flags); |
f8891e5e | 774 | __count_vm_events(PGFREE, 1 << order); |
95e34412 MK |
775 | migratetype = get_pageblock_migratetype(page); |
776 | set_freepage_migratetype(page, migratetype); | |
777 | free_one_page(page_zone(page), page, order, migratetype); | |
c54ad30c | 778 | local_irq_restore(flags); |
1da177e4 LT |
779 | } |
780 | ||
170a5a7e | 781 | void __init __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 | 782 | { |
c3993076 | 783 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 784 | struct page *p = page; |
c3993076 | 785 | unsigned int loop; |
a226f6c8 | 786 | |
e2d0bd2b YL |
787 | prefetchw(p); |
788 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
789 | prefetchw(p + 1); | |
c3993076 JW |
790 | __ClearPageReserved(p); |
791 | set_page_count(p, 0); | |
a226f6c8 | 792 | } |
e2d0bd2b YL |
793 | __ClearPageReserved(p); |
794 | set_page_count(p, 0); | |
c3993076 | 795 | |
e2d0bd2b | 796 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
797 | set_page_refcounted(page); |
798 | __free_pages(page, order); | |
a226f6c8 DH |
799 | } |
800 | ||
47118af0 | 801 | #ifdef CONFIG_CMA |
9cf510a5 | 802 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
803 | void __init init_cma_reserved_pageblock(struct page *page) |
804 | { | |
805 | unsigned i = pageblock_nr_pages; | |
806 | struct page *p = page; | |
807 | ||
808 | do { | |
809 | __ClearPageReserved(p); | |
810 | set_page_count(p, 0); | |
811 | } while (++p, --i); | |
812 | ||
813 | set_page_refcounted(page); | |
814 | set_pageblock_migratetype(page, MIGRATE_CMA); | |
815 | __free_pages(page, pageblock_order); | |
3dcc0571 | 816 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
817 | } |
818 | #endif | |
1da177e4 LT |
819 | |
820 | /* | |
821 | * The order of subdivision here is critical for the IO subsystem. | |
822 | * Please do not alter this order without good reasons and regression | |
823 | * testing. Specifically, as large blocks of memory are subdivided, | |
824 | * the order in which smaller blocks are delivered depends on the order | |
825 | * they're subdivided in this function. This is the primary factor | |
826 | * influencing the order in which pages are delivered to the IO | |
827 | * subsystem according to empirical testing, and this is also justified | |
828 | * by considering the behavior of a buddy system containing a single | |
829 | * large block of memory acted on by a series of small allocations. | |
830 | * This behavior is a critical factor in sglist merging's success. | |
831 | * | |
6d49e352 | 832 | * -- nyc |
1da177e4 | 833 | */ |
085cc7d5 | 834 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
835 | int low, int high, struct free_area *area, |
836 | int migratetype) | |
1da177e4 LT |
837 | { |
838 | unsigned long size = 1 << high; | |
839 | ||
840 | while (high > low) { | |
841 | area--; | |
842 | high--; | |
843 | size >>= 1; | |
309381fe | 844 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 SG |
845 | |
846 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
847 | if (high < debug_guardpage_minorder()) { | |
848 | /* | |
849 | * Mark as guard pages (or page), that will allow to | |
850 | * merge back to allocator when buddy will be freed. | |
851 | * Corresponding page table entries will not be touched, | |
852 | * pages will stay not present in virtual address space | |
853 | */ | |
854 | INIT_LIST_HEAD(&page[size].lru); | |
855 | set_page_guard_flag(&page[size]); | |
856 | set_page_private(&page[size], high); | |
857 | /* Guard pages are not available for any usage */ | |
d1ce749a BZ |
858 | __mod_zone_freepage_state(zone, -(1 << high), |
859 | migratetype); | |
c0a32fc5 SG |
860 | continue; |
861 | } | |
862 | #endif | |
b2a0ac88 | 863 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
864 | area->nr_free++; |
865 | set_page_order(&page[size], high); | |
866 | } | |
1da177e4 LT |
867 | } |
868 | ||
1da177e4 LT |
869 | /* |
870 | * This page is about to be returned from the page allocator | |
871 | */ | |
2a7684a2 | 872 | static inline int check_new_page(struct page *page) |
1da177e4 | 873 | { |
d230dec1 | 874 | const char *bad_reason = NULL; |
f0b791a3 DH |
875 | unsigned long bad_flags = 0; |
876 | ||
877 | if (unlikely(page_mapcount(page))) | |
878 | bad_reason = "nonzero mapcount"; | |
879 | if (unlikely(page->mapping != NULL)) | |
880 | bad_reason = "non-NULL mapping"; | |
881 | if (unlikely(atomic_read(&page->_count) != 0)) | |
882 | bad_reason = "nonzero _count"; | |
883 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { | |
884 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
885 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
886 | } | |
887 | if (unlikely(mem_cgroup_bad_page_check(page))) | |
888 | bad_reason = "cgroup check failed"; | |
889 | if (unlikely(bad_reason)) { | |
890 | bad_page(page, bad_reason, bad_flags); | |
689bcebf | 891 | return 1; |
8cc3b392 | 892 | } |
2a7684a2 WF |
893 | return 0; |
894 | } | |
895 | ||
896 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
897 | { | |
898 | int i; | |
899 | ||
900 | for (i = 0; i < (1 << order); i++) { | |
901 | struct page *p = page + i; | |
902 | if (unlikely(check_new_page(p))) | |
903 | return 1; | |
904 | } | |
689bcebf | 905 | |
4c21e2f2 | 906 | set_page_private(page, 0); |
7835e98b | 907 | set_page_refcounted(page); |
cc102509 NP |
908 | |
909 | arch_alloc_page(page, order); | |
1da177e4 | 910 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
911 | |
912 | if (gfp_flags & __GFP_ZERO) | |
913 | prep_zero_page(page, order, gfp_flags); | |
914 | ||
915 | if (order && (gfp_flags & __GFP_COMP)) | |
916 | prep_compound_page(page, order); | |
917 | ||
689bcebf | 918 | return 0; |
1da177e4 LT |
919 | } |
920 | ||
56fd56b8 MG |
921 | /* |
922 | * Go through the free lists for the given migratetype and remove | |
923 | * the smallest available page from the freelists | |
924 | */ | |
728ec980 MG |
925 | static inline |
926 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
927 | int migratetype) |
928 | { | |
929 | unsigned int current_order; | |
b8af2941 | 930 | struct free_area *area; |
56fd56b8 MG |
931 | struct page *page; |
932 | ||
933 | /* Find a page of the appropriate size in the preferred list */ | |
934 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
935 | area = &(zone->free_area[current_order]); | |
936 | if (list_empty(&area->free_list[migratetype])) | |
937 | continue; | |
938 | ||
939 | page = list_entry(area->free_list[migratetype].next, | |
940 | struct page, lru); | |
941 | list_del(&page->lru); | |
942 | rmv_page_order(page); | |
943 | area->nr_free--; | |
56fd56b8 | 944 | expand(zone, page, order, current_order, area, migratetype); |
5bcc9f86 | 945 | set_freepage_migratetype(page, migratetype); |
56fd56b8 MG |
946 | return page; |
947 | } | |
948 | ||
949 | return NULL; | |
950 | } | |
951 | ||
952 | ||
b2a0ac88 MG |
953 | /* |
954 | * This array describes the order lists are fallen back to when | |
955 | * the free lists for the desirable migrate type are depleted | |
956 | */ | |
47118af0 MN |
957 | static int fallbacks[MIGRATE_TYPES][4] = { |
958 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
959 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
960 | #ifdef CONFIG_CMA | |
961 | [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
962 | [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */ | |
963 | #else | |
964 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
965 | #endif | |
6d4a4916 | 966 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 967 | #ifdef CONFIG_MEMORY_ISOLATION |
6d4a4916 | 968 | [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 969 | #endif |
b2a0ac88 MG |
970 | }; |
971 | ||
c361be55 MG |
972 | /* |
973 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 974 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
975 | * boundary. If alignment is required, use move_freepages_block() |
976 | */ | |
435b405c | 977 | int move_freepages(struct zone *zone, |
b69a7288 AB |
978 | struct page *start_page, struct page *end_page, |
979 | int migratetype) | |
c361be55 MG |
980 | { |
981 | struct page *page; | |
982 | unsigned long order; | |
d100313f | 983 | int pages_moved = 0; |
c361be55 MG |
984 | |
985 | #ifndef CONFIG_HOLES_IN_ZONE | |
986 | /* | |
987 | * page_zone is not safe to call in this context when | |
988 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
989 | * anyway as we check zone boundaries in move_freepages_block(). | |
990 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 991 | * grouping pages by mobility |
c361be55 MG |
992 | */ |
993 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
994 | #endif | |
995 | ||
996 | for (page = start_page; page <= end_page;) { | |
344c790e | 997 | /* Make sure we are not inadvertently changing nodes */ |
309381fe | 998 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
344c790e | 999 | |
c361be55 MG |
1000 | if (!pfn_valid_within(page_to_pfn(page))) { |
1001 | page++; | |
1002 | continue; | |
1003 | } | |
1004 | ||
1005 | if (!PageBuddy(page)) { | |
1006 | page++; | |
1007 | continue; | |
1008 | } | |
1009 | ||
1010 | order = page_order(page); | |
84be48d8 KS |
1011 | list_move(&page->lru, |
1012 | &zone->free_area[order].free_list[migratetype]); | |
95e34412 | 1013 | set_freepage_migratetype(page, migratetype); |
c361be55 | 1014 | page += 1 << order; |
d100313f | 1015 | pages_moved += 1 << order; |
c361be55 MG |
1016 | } |
1017 | ||
d100313f | 1018 | return pages_moved; |
c361be55 MG |
1019 | } |
1020 | ||
ee6f509c | 1021 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1022 | int migratetype) |
c361be55 MG |
1023 | { |
1024 | unsigned long start_pfn, end_pfn; | |
1025 | struct page *start_page, *end_page; | |
1026 | ||
1027 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1028 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1029 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1030 | end_page = start_page + pageblock_nr_pages - 1; |
1031 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1032 | |
1033 | /* Do not cross zone boundaries */ | |
108bcc96 | 1034 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1035 | start_page = page; |
108bcc96 | 1036 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1037 | return 0; |
1038 | ||
1039 | return move_freepages(zone, start_page, end_page, migratetype); | |
1040 | } | |
1041 | ||
2f66a68f MG |
1042 | static void change_pageblock_range(struct page *pageblock_page, |
1043 | int start_order, int migratetype) | |
1044 | { | |
1045 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1046 | ||
1047 | while (nr_pageblocks--) { | |
1048 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1049 | pageblock_page += pageblock_nr_pages; | |
1050 | } | |
1051 | } | |
1052 | ||
fef903ef SB |
1053 | /* |
1054 | * If breaking a large block of pages, move all free pages to the preferred | |
1055 | * allocation list. If falling back for a reclaimable kernel allocation, be | |
1056 | * more aggressive about taking ownership of free pages. | |
1057 | * | |
1058 | * On the other hand, never change migration type of MIGRATE_CMA pageblocks | |
1059 | * nor move CMA pages to different free lists. We don't want unmovable pages | |
1060 | * to be allocated from MIGRATE_CMA areas. | |
1061 | * | |
1062 | * Returns the new migratetype of the pageblock (or the same old migratetype | |
1063 | * if it was unchanged). | |
1064 | */ | |
1065 | static int try_to_steal_freepages(struct zone *zone, struct page *page, | |
1066 | int start_type, int fallback_type) | |
1067 | { | |
1068 | int current_order = page_order(page); | |
1069 | ||
0cbef29a KM |
1070 | /* |
1071 | * When borrowing from MIGRATE_CMA, we need to release the excess | |
5bcc9f86 VB |
1072 | * buddy pages to CMA itself. We also ensure the freepage_migratetype |
1073 | * is set to CMA so it is returned to the correct freelist in case | |
1074 | * the page ends up being not actually allocated from the pcp lists. | |
0cbef29a | 1075 | */ |
fef903ef SB |
1076 | if (is_migrate_cma(fallback_type)) |
1077 | return fallback_type; | |
1078 | ||
1079 | /* Take ownership for orders >= pageblock_order */ | |
1080 | if (current_order >= pageblock_order) { | |
1081 | change_pageblock_range(page, current_order, start_type); | |
1082 | return start_type; | |
1083 | } | |
1084 | ||
1085 | if (current_order >= pageblock_order / 2 || | |
1086 | start_type == MIGRATE_RECLAIMABLE || | |
1087 | page_group_by_mobility_disabled) { | |
1088 | int pages; | |
1089 | ||
1090 | pages = move_freepages_block(zone, page, start_type); | |
1091 | ||
1092 | /* Claim the whole block if over half of it is free */ | |
1093 | if (pages >= (1 << (pageblock_order-1)) || | |
1094 | page_group_by_mobility_disabled) { | |
1095 | ||
1096 | set_pageblock_migratetype(page, start_type); | |
1097 | return start_type; | |
1098 | } | |
1099 | ||
1100 | } | |
1101 | ||
1102 | return fallback_type; | |
1103 | } | |
1104 | ||
b2a0ac88 | 1105 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
1106 | static inline struct page * |
1107 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 | 1108 | { |
b8af2941 | 1109 | struct free_area *area; |
b2a0ac88 MG |
1110 | int current_order; |
1111 | struct page *page; | |
fef903ef | 1112 | int migratetype, new_type, i; |
b2a0ac88 MG |
1113 | |
1114 | /* Find the largest possible block of pages in the other list */ | |
1115 | for (current_order = MAX_ORDER-1; current_order >= order; | |
1116 | --current_order) { | |
6d4a4916 | 1117 | for (i = 0;; i++) { |
b2a0ac88 MG |
1118 | migratetype = fallbacks[start_migratetype][i]; |
1119 | ||
56fd56b8 MG |
1120 | /* MIGRATE_RESERVE handled later if necessary */ |
1121 | if (migratetype == MIGRATE_RESERVE) | |
6d4a4916 | 1122 | break; |
e010487d | 1123 | |
b2a0ac88 MG |
1124 | area = &(zone->free_area[current_order]); |
1125 | if (list_empty(&area->free_list[migratetype])) | |
1126 | continue; | |
1127 | ||
1128 | page = list_entry(area->free_list[migratetype].next, | |
1129 | struct page, lru); | |
1130 | area->nr_free--; | |
1131 | ||
fef903ef SB |
1132 | new_type = try_to_steal_freepages(zone, page, |
1133 | start_migratetype, | |
1134 | migratetype); | |
b2a0ac88 MG |
1135 | |
1136 | /* Remove the page from the freelists */ | |
1137 | list_del(&page->lru); | |
1138 | rmv_page_order(page); | |
b2a0ac88 | 1139 | |
47118af0 | 1140 | expand(zone, page, order, current_order, area, |
0cbef29a | 1141 | new_type); |
5bcc9f86 VB |
1142 | /* The freepage_migratetype may differ from pageblock's |
1143 | * migratetype depending on the decisions in | |
1144 | * try_to_steal_freepages. This is OK as long as it does | |
1145 | * not differ for MIGRATE_CMA type. | |
1146 | */ | |
1147 | set_freepage_migratetype(page, new_type); | |
e0fff1bd | 1148 | |
52c8f6a5 KM |
1149 | trace_mm_page_alloc_extfrag(page, order, current_order, |
1150 | start_migratetype, migratetype, new_type); | |
e0fff1bd | 1151 | |
b2a0ac88 MG |
1152 | return page; |
1153 | } | |
1154 | } | |
1155 | ||
728ec980 | 1156 | return NULL; |
b2a0ac88 MG |
1157 | } |
1158 | ||
56fd56b8 | 1159 | /* |
1da177e4 LT |
1160 | * Do the hard work of removing an element from the buddy allocator. |
1161 | * Call me with the zone->lock already held. | |
1162 | */ | |
b2a0ac88 MG |
1163 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1164 | int migratetype) | |
1da177e4 | 1165 | { |
1da177e4 LT |
1166 | struct page *page; |
1167 | ||
728ec980 | 1168 | retry_reserve: |
56fd56b8 | 1169 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1170 | |
728ec980 | 1171 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 1172 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 1173 | |
728ec980 MG |
1174 | /* |
1175 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1176 | * is used because __rmqueue_smallest is an inline function | |
1177 | * and we want just one call site | |
1178 | */ | |
1179 | if (!page) { | |
1180 | migratetype = MIGRATE_RESERVE; | |
1181 | goto retry_reserve; | |
1182 | } | |
1183 | } | |
1184 | ||
0d3d062a | 1185 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1186 | return page; |
1da177e4 LT |
1187 | } |
1188 | ||
5f63b720 | 1189 | /* |
1da177e4 LT |
1190 | * Obtain a specified number of elements from the buddy allocator, all under |
1191 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1192 | * Returns the number of new pages which were placed at *list. | |
1193 | */ | |
5f63b720 | 1194 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1195 | unsigned long count, struct list_head *list, |
e084b2d9 | 1196 | int migratetype, int cold) |
1da177e4 | 1197 | { |
5bcc9f86 | 1198 | int i; |
5f63b720 | 1199 | |
c54ad30c | 1200 | spin_lock(&zone->lock); |
1da177e4 | 1201 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1202 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1203 | if (unlikely(page == NULL)) |
1da177e4 | 1204 | break; |
81eabcbe MG |
1205 | |
1206 | /* | |
1207 | * Split buddy pages returned by expand() are received here | |
1208 | * in physical page order. The page is added to the callers and | |
1209 | * list and the list head then moves forward. From the callers | |
1210 | * perspective, the linked list is ordered by page number in | |
1211 | * some conditions. This is useful for IO devices that can | |
1212 | * merge IO requests if the physical pages are ordered | |
1213 | * properly. | |
1214 | */ | |
e084b2d9 MG |
1215 | if (likely(cold == 0)) |
1216 | list_add(&page->lru, list); | |
1217 | else | |
1218 | list_add_tail(&page->lru, list); | |
81eabcbe | 1219 | list = &page->lru; |
5bcc9f86 | 1220 | if (is_migrate_cma(get_freepage_migratetype(page))) |
d1ce749a BZ |
1221 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
1222 | -(1 << order)); | |
1da177e4 | 1223 | } |
f2260e6b | 1224 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1225 | spin_unlock(&zone->lock); |
085cc7d5 | 1226 | return i; |
1da177e4 LT |
1227 | } |
1228 | ||
4ae7c039 | 1229 | #ifdef CONFIG_NUMA |
8fce4d8e | 1230 | /* |
4037d452 CL |
1231 | * Called from the vmstat counter updater to drain pagesets of this |
1232 | * currently executing processor on remote nodes after they have | |
1233 | * expired. | |
1234 | * | |
879336c3 CL |
1235 | * Note that this function must be called with the thread pinned to |
1236 | * a single processor. | |
8fce4d8e | 1237 | */ |
4037d452 | 1238 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1239 | { |
4ae7c039 | 1240 | unsigned long flags; |
4037d452 | 1241 | int to_drain; |
998d39cb | 1242 | unsigned long batch; |
4ae7c039 | 1243 | |
4037d452 | 1244 | local_irq_save(flags); |
998d39cb CS |
1245 | batch = ACCESS_ONCE(pcp->batch); |
1246 | if (pcp->count >= batch) | |
1247 | to_drain = batch; | |
4037d452 CL |
1248 | else |
1249 | to_drain = pcp->count; | |
2a13515c KM |
1250 | if (to_drain > 0) { |
1251 | free_pcppages_bulk(zone, to_drain, pcp); | |
1252 | pcp->count -= to_drain; | |
1253 | } | |
4037d452 | 1254 | local_irq_restore(flags); |
4ae7c039 CL |
1255 | } |
1256 | #endif | |
1257 | ||
9f8f2172 CL |
1258 | /* |
1259 | * Drain pages of the indicated processor. | |
1260 | * | |
1261 | * The processor must either be the current processor and the | |
1262 | * thread pinned to the current processor or a processor that | |
1263 | * is not online. | |
1264 | */ | |
1265 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1266 | { |
c54ad30c | 1267 | unsigned long flags; |
1da177e4 | 1268 | struct zone *zone; |
1da177e4 | 1269 | |
ee99c71c | 1270 | for_each_populated_zone(zone) { |
1da177e4 | 1271 | struct per_cpu_pageset *pset; |
3dfa5721 | 1272 | struct per_cpu_pages *pcp; |
1da177e4 | 1273 | |
99dcc3e5 CL |
1274 | local_irq_save(flags); |
1275 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3dfa5721 CL |
1276 | |
1277 | pcp = &pset->pcp; | |
2ff754fa DR |
1278 | if (pcp->count) { |
1279 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1280 | pcp->count = 0; | |
1281 | } | |
3dfa5721 | 1282 | local_irq_restore(flags); |
1da177e4 LT |
1283 | } |
1284 | } | |
1da177e4 | 1285 | |
9f8f2172 CL |
1286 | /* |
1287 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1288 | */ | |
1289 | void drain_local_pages(void *arg) | |
1290 | { | |
1291 | drain_pages(smp_processor_id()); | |
1292 | } | |
1293 | ||
1294 | /* | |
74046494 GBY |
1295 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
1296 | * | |
1297 | * Note that this code is protected against sending an IPI to an offline | |
1298 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
1299 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
1300 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
1301 | * before the call to on_each_cpu_mask(). | |
9f8f2172 CL |
1302 | */ |
1303 | void drain_all_pages(void) | |
1304 | { | |
74046494 GBY |
1305 | int cpu; |
1306 | struct per_cpu_pageset *pcp; | |
1307 | struct zone *zone; | |
1308 | ||
1309 | /* | |
1310 | * Allocate in the BSS so we wont require allocation in | |
1311 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
1312 | */ | |
1313 | static cpumask_t cpus_with_pcps; | |
1314 | ||
1315 | /* | |
1316 | * We don't care about racing with CPU hotplug event | |
1317 | * as offline notification will cause the notified | |
1318 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
1319 | * disables preemption as part of its processing | |
1320 | */ | |
1321 | for_each_online_cpu(cpu) { | |
1322 | bool has_pcps = false; | |
1323 | for_each_populated_zone(zone) { | |
1324 | pcp = per_cpu_ptr(zone->pageset, cpu); | |
1325 | if (pcp->pcp.count) { | |
1326 | has_pcps = true; | |
1327 | break; | |
1328 | } | |
1329 | } | |
1330 | if (has_pcps) | |
1331 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
1332 | else | |
1333 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
1334 | } | |
1335 | on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1); | |
9f8f2172 CL |
1336 | } |
1337 | ||
296699de | 1338 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1339 | |
1340 | void mark_free_pages(struct zone *zone) | |
1341 | { | |
f623f0db RW |
1342 | unsigned long pfn, max_zone_pfn; |
1343 | unsigned long flags; | |
b2a0ac88 | 1344 | int order, t; |
1da177e4 LT |
1345 | struct list_head *curr; |
1346 | ||
8080fc03 | 1347 | if (zone_is_empty(zone)) |
1da177e4 LT |
1348 | return; |
1349 | ||
1350 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 1351 | |
108bcc96 | 1352 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
1353 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
1354 | if (pfn_valid(pfn)) { | |
1355 | struct page *page = pfn_to_page(pfn); | |
1356 | ||
7be98234 RW |
1357 | if (!swsusp_page_is_forbidden(page)) |
1358 | swsusp_unset_page_free(page); | |
f623f0db | 1359 | } |
1da177e4 | 1360 | |
b2a0ac88 MG |
1361 | for_each_migratetype_order(order, t) { |
1362 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1363 | unsigned long i; |
1da177e4 | 1364 | |
f623f0db RW |
1365 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1366 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1367 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1368 | } |
b2a0ac88 | 1369 | } |
1da177e4 LT |
1370 | spin_unlock_irqrestore(&zone->lock, flags); |
1371 | } | |
e2c55dc8 | 1372 | #endif /* CONFIG_PM */ |
1da177e4 | 1373 | |
1da177e4 LT |
1374 | /* |
1375 | * Free a 0-order page | |
fc91668e | 1376 | * cold == 1 ? free a cold page : free a hot page |
1da177e4 | 1377 | */ |
fc91668e | 1378 | void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1379 | { |
1380 | struct zone *zone = page_zone(page); | |
1381 | struct per_cpu_pages *pcp; | |
1382 | unsigned long flags; | |
5f8dcc21 | 1383 | int migratetype; |
1da177e4 | 1384 | |
ec95f53a | 1385 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1386 | return; |
1387 | ||
5f8dcc21 | 1388 | migratetype = get_pageblock_migratetype(page); |
b12c4ad1 | 1389 | set_freepage_migratetype(page, migratetype); |
1da177e4 | 1390 | local_irq_save(flags); |
f8891e5e | 1391 | __count_vm_event(PGFREE); |
da456f14 | 1392 | |
5f8dcc21 MG |
1393 | /* |
1394 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1395 | * Free ISOLATE pages back to the allocator because they are being | |
1396 | * offlined but treat RESERVE as movable pages so we can get those | |
1397 | * areas back if necessary. Otherwise, we may have to free | |
1398 | * excessively into the page allocator | |
1399 | */ | |
1400 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 1401 | if (unlikely(is_migrate_isolate(migratetype))) { |
5f8dcc21 MG |
1402 | free_one_page(zone, page, 0, migratetype); |
1403 | goto out; | |
1404 | } | |
1405 | migratetype = MIGRATE_MOVABLE; | |
1406 | } | |
1407 | ||
99dcc3e5 | 1408 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3dfa5721 | 1409 | if (cold) |
5f8dcc21 | 1410 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1411 | else |
5f8dcc21 | 1412 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1413 | pcp->count++; |
48db57f8 | 1414 | if (pcp->count >= pcp->high) { |
998d39cb CS |
1415 | unsigned long batch = ACCESS_ONCE(pcp->batch); |
1416 | free_pcppages_bulk(zone, batch, pcp); | |
1417 | pcp->count -= batch; | |
48db57f8 | 1418 | } |
5f8dcc21 MG |
1419 | |
1420 | out: | |
1da177e4 | 1421 | local_irq_restore(flags); |
1da177e4 LT |
1422 | } |
1423 | ||
cc59850e KK |
1424 | /* |
1425 | * Free a list of 0-order pages | |
1426 | */ | |
1427 | void free_hot_cold_page_list(struct list_head *list, int cold) | |
1428 | { | |
1429 | struct page *page, *next; | |
1430 | ||
1431 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1432 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1433 | free_hot_cold_page(page, cold); |
1434 | } | |
1435 | } | |
1436 | ||
8dfcc9ba NP |
1437 | /* |
1438 | * split_page takes a non-compound higher-order page, and splits it into | |
1439 | * n (1<<order) sub-pages: page[0..n] | |
1440 | * Each sub-page must be freed individually. | |
1441 | * | |
1442 | * Note: this is probably too low level an operation for use in drivers. | |
1443 | * Please consult with lkml before using this in your driver. | |
1444 | */ | |
1445 | void split_page(struct page *page, unsigned int order) | |
1446 | { | |
1447 | int i; | |
1448 | ||
309381fe SL |
1449 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1450 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
1451 | |
1452 | #ifdef CONFIG_KMEMCHECK | |
1453 | /* | |
1454 | * Split shadow pages too, because free(page[0]) would | |
1455 | * otherwise free the whole shadow. | |
1456 | */ | |
1457 | if (kmemcheck_page_is_tracked(page)) | |
1458 | split_page(virt_to_page(page[0].shadow), order); | |
1459 | #endif | |
1460 | ||
7835e98b NP |
1461 | for (i = 1; i < (1 << order); i++) |
1462 | set_page_refcounted(page + i); | |
8dfcc9ba | 1463 | } |
5853ff23 | 1464 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 1465 | |
8fb74b9f | 1466 | static int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 1467 | { |
748446bb MG |
1468 | unsigned long watermark; |
1469 | struct zone *zone; | |
2139cbe6 | 1470 | int mt; |
748446bb MG |
1471 | |
1472 | BUG_ON(!PageBuddy(page)); | |
1473 | ||
1474 | zone = page_zone(page); | |
2e30abd1 | 1475 | mt = get_pageblock_migratetype(page); |
748446bb | 1476 | |
194159fb | 1477 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
1478 | /* Obey watermarks as if the page was being allocated */ |
1479 | watermark = low_wmark_pages(zone) + (1 << order); | |
1480 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1481 | return 0; | |
1482 | ||
8fb74b9f | 1483 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 1484 | } |
748446bb MG |
1485 | |
1486 | /* Remove page from free list */ | |
1487 | list_del(&page->lru); | |
1488 | zone->free_area[order].nr_free--; | |
1489 | rmv_page_order(page); | |
2139cbe6 | 1490 | |
8fb74b9f | 1491 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
1492 | if (order >= pageblock_order - 1) { |
1493 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
1494 | for (; page < endpage; page += pageblock_nr_pages) { |
1495 | int mt = get_pageblock_migratetype(page); | |
194159fb | 1496 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
1497 | set_pageblock_migratetype(page, |
1498 | MIGRATE_MOVABLE); | |
1499 | } | |
748446bb MG |
1500 | } |
1501 | ||
8fb74b9f | 1502 | return 1UL << order; |
1fb3f8ca MG |
1503 | } |
1504 | ||
1505 | /* | |
1506 | * Similar to split_page except the page is already free. As this is only | |
1507 | * being used for migration, the migratetype of the block also changes. | |
1508 | * As this is called with interrupts disabled, the caller is responsible | |
1509 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1510 | * are enabled. | |
1511 | * | |
1512 | * Note: this is probably too low level an operation for use in drivers. | |
1513 | * Please consult with lkml before using this in your driver. | |
1514 | */ | |
1515 | int split_free_page(struct page *page) | |
1516 | { | |
1517 | unsigned int order; | |
1518 | int nr_pages; | |
1519 | ||
1fb3f8ca MG |
1520 | order = page_order(page); |
1521 | ||
8fb74b9f | 1522 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
1523 | if (!nr_pages) |
1524 | return 0; | |
1525 | ||
1526 | /* Split into individual pages */ | |
1527 | set_page_refcounted(page); | |
1528 | split_page(page, order); | |
1529 | return nr_pages; | |
748446bb MG |
1530 | } |
1531 | ||
1da177e4 LT |
1532 | /* |
1533 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1534 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1535 | * or two. | |
1536 | */ | |
0a15c3e9 MG |
1537 | static inline |
1538 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1539 | struct zone *zone, int order, gfp_t gfp_flags, |
1540 | int migratetype) | |
1da177e4 LT |
1541 | { |
1542 | unsigned long flags; | |
689bcebf | 1543 | struct page *page; |
1da177e4 LT |
1544 | int cold = !!(gfp_flags & __GFP_COLD); |
1545 | ||
689bcebf | 1546 | again: |
48db57f8 | 1547 | if (likely(order == 0)) { |
1da177e4 | 1548 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1549 | struct list_head *list; |
1da177e4 | 1550 | |
1da177e4 | 1551 | local_irq_save(flags); |
99dcc3e5 CL |
1552 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1553 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1554 | if (list_empty(list)) { |
535131e6 | 1555 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1556 | pcp->batch, list, |
e084b2d9 | 1557 | migratetype, cold); |
5f8dcc21 | 1558 | if (unlikely(list_empty(list))) |
6fb332fa | 1559 | goto failed; |
535131e6 | 1560 | } |
b92a6edd | 1561 | |
5f8dcc21 MG |
1562 | if (cold) |
1563 | page = list_entry(list->prev, struct page, lru); | |
1564 | else | |
1565 | page = list_entry(list->next, struct page, lru); | |
1566 | ||
b92a6edd MG |
1567 | list_del(&page->lru); |
1568 | pcp->count--; | |
7fb1d9fc | 1569 | } else { |
dab48dab AM |
1570 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1571 | /* | |
1572 | * __GFP_NOFAIL is not to be used in new code. | |
1573 | * | |
1574 | * All __GFP_NOFAIL callers should be fixed so that they | |
1575 | * properly detect and handle allocation failures. | |
1576 | * | |
1577 | * We most definitely don't want callers attempting to | |
4923abf9 | 1578 | * allocate greater than order-1 page units with |
dab48dab AM |
1579 | * __GFP_NOFAIL. |
1580 | */ | |
4923abf9 | 1581 | WARN_ON_ONCE(order > 1); |
dab48dab | 1582 | } |
1da177e4 | 1583 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1584 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1585 | spin_unlock(&zone->lock); |
1586 | if (!page) | |
1587 | goto failed; | |
d1ce749a | 1588 | __mod_zone_freepage_state(zone, -(1 << order), |
5bcc9f86 | 1589 | get_freepage_migratetype(page)); |
1da177e4 LT |
1590 | } |
1591 | ||
3a025760 | 1592 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
27329369 | 1593 | |
f8891e5e | 1594 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1595 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1596 | local_irq_restore(flags); |
1da177e4 | 1597 | |
309381fe | 1598 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
17cf4406 | 1599 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1600 | goto again; |
1da177e4 | 1601 | return page; |
a74609fa NP |
1602 | |
1603 | failed: | |
1604 | local_irq_restore(flags); | |
a74609fa | 1605 | return NULL; |
1da177e4 LT |
1606 | } |
1607 | ||
933e312e AM |
1608 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1609 | ||
b2588c4b | 1610 | static struct { |
933e312e AM |
1611 | struct fault_attr attr; |
1612 | ||
1613 | u32 ignore_gfp_highmem; | |
1614 | u32 ignore_gfp_wait; | |
54114994 | 1615 | u32 min_order; |
933e312e AM |
1616 | } fail_page_alloc = { |
1617 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1618 | .ignore_gfp_wait = 1, |
1619 | .ignore_gfp_highmem = 1, | |
54114994 | 1620 | .min_order = 1, |
933e312e AM |
1621 | }; |
1622 | ||
1623 | static int __init setup_fail_page_alloc(char *str) | |
1624 | { | |
1625 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1626 | } | |
1627 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1628 | ||
deaf386e | 1629 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1630 | { |
54114994 | 1631 | if (order < fail_page_alloc.min_order) |
deaf386e | 1632 | return false; |
933e312e | 1633 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 1634 | return false; |
933e312e | 1635 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 1636 | return false; |
933e312e | 1637 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
deaf386e | 1638 | return false; |
933e312e AM |
1639 | |
1640 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1641 | } | |
1642 | ||
1643 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1644 | ||
1645 | static int __init fail_page_alloc_debugfs(void) | |
1646 | { | |
f4ae40a6 | 1647 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1648 | struct dentry *dir; |
933e312e | 1649 | |
dd48c085 AM |
1650 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1651 | &fail_page_alloc.attr); | |
1652 | if (IS_ERR(dir)) | |
1653 | return PTR_ERR(dir); | |
933e312e | 1654 | |
b2588c4b AM |
1655 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1656 | &fail_page_alloc.ignore_gfp_wait)) | |
1657 | goto fail; | |
1658 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1659 | &fail_page_alloc.ignore_gfp_highmem)) | |
1660 | goto fail; | |
1661 | if (!debugfs_create_u32("min-order", mode, dir, | |
1662 | &fail_page_alloc.min_order)) | |
1663 | goto fail; | |
1664 | ||
1665 | return 0; | |
1666 | fail: | |
dd48c085 | 1667 | debugfs_remove_recursive(dir); |
933e312e | 1668 | |
b2588c4b | 1669 | return -ENOMEM; |
933e312e AM |
1670 | } |
1671 | ||
1672 | late_initcall(fail_page_alloc_debugfs); | |
1673 | ||
1674 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1675 | ||
1676 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1677 | ||
deaf386e | 1678 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1679 | { |
deaf386e | 1680 | return false; |
933e312e AM |
1681 | } |
1682 | ||
1683 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1684 | ||
1da177e4 | 1685 | /* |
88f5acf8 | 1686 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1687 | * of the allocation. |
1688 | */ | |
88f5acf8 MG |
1689 | static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
1690 | int classzone_idx, int alloc_flags, long free_pages) | |
1da177e4 LT |
1691 | { |
1692 | /* free_pages my go negative - that's OK */ | |
d23ad423 | 1693 | long min = mark; |
2cfed075 | 1694 | long lowmem_reserve = z->lowmem_reserve[classzone_idx]; |
1da177e4 | 1695 | int o; |
026b0814 | 1696 | long free_cma = 0; |
1da177e4 | 1697 | |
df0a6daa | 1698 | free_pages -= (1 << order) - 1; |
7fb1d9fc | 1699 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1700 | min -= min / 2; |
7fb1d9fc | 1701 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 | 1702 | min -= min / 4; |
d95ea5d1 BZ |
1703 | #ifdef CONFIG_CMA |
1704 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
1705 | if (!(alloc_flags & ALLOC_CMA)) | |
026b0814 | 1706 | free_cma = zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 1707 | #endif |
026b0814 TS |
1708 | |
1709 | if (free_pages - free_cma <= min + lowmem_reserve) | |
88f5acf8 | 1710 | return false; |
1da177e4 LT |
1711 | for (o = 0; o < order; o++) { |
1712 | /* At the next order, this order's pages become unavailable */ | |
1713 | free_pages -= z->free_area[o].nr_free << o; | |
1714 | ||
1715 | /* Require fewer higher order pages to be free */ | |
1716 | min >>= 1; | |
1717 | ||
1718 | if (free_pages <= min) | |
88f5acf8 | 1719 | return false; |
1da177e4 | 1720 | } |
88f5acf8 MG |
1721 | return true; |
1722 | } | |
1723 | ||
1724 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
1725 | int classzone_idx, int alloc_flags) | |
1726 | { | |
1727 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1728 | zone_page_state(z, NR_FREE_PAGES)); | |
1729 | } | |
1730 | ||
1731 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, | |
1732 | int classzone_idx, int alloc_flags) | |
1733 | { | |
1734 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1735 | ||
1736 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1737 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1738 | ||
1739 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1740 | free_pages); | |
1da177e4 LT |
1741 | } |
1742 | ||
9276b1bc PJ |
1743 | #ifdef CONFIG_NUMA |
1744 | /* | |
1745 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1746 | * skip over zones that are not allowed by the cpuset, or that have | |
1747 | * been recently (in last second) found to be nearly full. See further | |
1748 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1749 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc | 1750 | * |
a1aeb65a | 1751 | * If the zonelist cache is present in the passed zonelist, then |
9276b1bc | 1752 | * returns a pointer to the allowed node mask (either the current |
4b0ef1fe | 1753 | * tasks mems_allowed, or node_states[N_MEMORY].) |
9276b1bc PJ |
1754 | * |
1755 | * If the zonelist cache is not available for this zonelist, does | |
1756 | * nothing and returns NULL. | |
1757 | * | |
1758 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1759 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1760 | * | |
1761 | * We hold off even calling zlc_setup, until after we've checked the | |
1762 | * first zone in the zonelist, on the theory that most allocations will | |
1763 | * be satisfied from that first zone, so best to examine that zone as | |
1764 | * quickly as we can. | |
1765 | */ | |
1766 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1767 | { | |
1768 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1769 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1770 | ||
1771 | zlc = zonelist->zlcache_ptr; | |
1772 | if (!zlc) | |
1773 | return NULL; | |
1774 | ||
f05111f5 | 1775 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1776 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1777 | zlc->last_full_zap = jiffies; | |
1778 | } | |
1779 | ||
1780 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1781 | &cpuset_current_mems_allowed : | |
4b0ef1fe | 1782 | &node_states[N_MEMORY]; |
9276b1bc PJ |
1783 | return allowednodes; |
1784 | } | |
1785 | ||
1786 | /* | |
1787 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1788 | * if it is worth looking at further for free memory: | |
1789 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1790 | * bit set in the zonelist_cache fullzones BITMAP). | |
1791 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1792 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1793 | * Return true (non-zero) if zone is worth looking at further, or | |
1794 | * else return false (zero) if it is not. | |
1795 | * | |
1796 | * This check -ignores- the distinction between various watermarks, | |
1797 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1798 | * found to be full for any variation of these watermarks, it will | |
1799 | * be considered full for up to one second by all requests, unless | |
1800 | * we are so low on memory on all allowed nodes that we are forced | |
1801 | * into the second scan of the zonelist. | |
1802 | * | |
1803 | * In the second scan we ignore this zonelist cache and exactly | |
1804 | * apply the watermarks to all zones, even it is slower to do so. | |
1805 | * We are low on memory in the second scan, and should leave no stone | |
1806 | * unturned looking for a free page. | |
1807 | */ | |
dd1a239f | 1808 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1809 | nodemask_t *allowednodes) |
1810 | { | |
1811 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1812 | int i; /* index of *z in zonelist zones */ | |
1813 | int n; /* node that zone *z is on */ | |
1814 | ||
1815 | zlc = zonelist->zlcache_ptr; | |
1816 | if (!zlc) | |
1817 | return 1; | |
1818 | ||
dd1a239f | 1819 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1820 | n = zlc->z_to_n[i]; |
1821 | ||
1822 | /* This zone is worth trying if it is allowed but not full */ | |
1823 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1824 | } | |
1825 | ||
1826 | /* | |
1827 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1828 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1829 | * from that zone don't waste time re-examining it. | |
1830 | */ | |
dd1a239f | 1831 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1832 | { |
1833 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1834 | int i; /* index of *z in zonelist zones */ | |
1835 | ||
1836 | zlc = zonelist->zlcache_ptr; | |
1837 | if (!zlc) | |
1838 | return; | |
1839 | ||
dd1a239f | 1840 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1841 | |
1842 | set_bit(i, zlc->fullzones); | |
1843 | } | |
1844 | ||
76d3fbf8 MG |
1845 | /* |
1846 | * clear all zones full, called after direct reclaim makes progress so that | |
1847 | * a zone that was recently full is not skipped over for up to a second | |
1848 | */ | |
1849 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1850 | { | |
1851 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1852 | ||
1853 | zlc = zonelist->zlcache_ptr; | |
1854 | if (!zlc) | |
1855 | return; | |
1856 | ||
1857 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1858 | } | |
1859 | ||
81c0a2bb JW |
1860 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
1861 | { | |
fff4068c | 1862 | return local_zone->node == zone->node; |
81c0a2bb JW |
1863 | } |
1864 | ||
957f822a DR |
1865 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1866 | { | |
5f7a75ac MG |
1867 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
1868 | RECLAIM_DISTANCE; | |
957f822a DR |
1869 | } |
1870 | ||
9276b1bc PJ |
1871 | #else /* CONFIG_NUMA */ |
1872 | ||
1873 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1874 | { | |
1875 | return NULL; | |
1876 | } | |
1877 | ||
dd1a239f | 1878 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1879 | nodemask_t *allowednodes) |
1880 | { | |
1881 | return 1; | |
1882 | } | |
1883 | ||
dd1a239f | 1884 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1885 | { |
1886 | } | |
76d3fbf8 MG |
1887 | |
1888 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1889 | { | |
1890 | } | |
957f822a | 1891 | |
81c0a2bb JW |
1892 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
1893 | { | |
1894 | return true; | |
1895 | } | |
1896 | ||
957f822a DR |
1897 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1898 | { | |
1899 | return true; | |
1900 | } | |
1901 | ||
9276b1bc PJ |
1902 | #endif /* CONFIG_NUMA */ |
1903 | ||
7fb1d9fc | 1904 | /* |
0798e519 | 1905 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1906 | * a page. |
1907 | */ | |
1908 | static struct page * | |
19770b32 | 1909 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1910 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1911 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1912 | { |
dd1a239f | 1913 | struct zoneref *z; |
7fb1d9fc | 1914 | struct page *page = NULL; |
54a6eb5c | 1915 | int classzone_idx; |
5117f45d | 1916 | struct zone *zone; |
9276b1bc PJ |
1917 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1918 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1919 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1920 | |
19770b32 | 1921 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1922 | zonelist_scan: |
7fb1d9fc | 1923 | /* |
9276b1bc | 1924 | * Scan zonelist, looking for a zone with enough free. |
3b11f0aa | 1925 | * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c. |
7fb1d9fc | 1926 | */ |
19770b32 MG |
1927 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1928 | high_zoneidx, nodemask) { | |
e085dbc5 JW |
1929 | unsigned long mark; |
1930 | ||
e5adfffc | 1931 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
9276b1bc PJ |
1932 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1933 | continue; | |
664eedde MG |
1934 | if (cpusets_enabled() && |
1935 | (alloc_flags & ALLOC_CPUSET) && | |
02a0e53d | 1936 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
cd38b115 | 1937 | continue; |
e085dbc5 | 1938 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
e66f0972 | 1939 | if (unlikely(alloc_flags & ALLOC_NO_WATERMARKS)) |
e085dbc5 | 1940 | goto try_this_zone; |
81c0a2bb JW |
1941 | /* |
1942 | * Distribute pages in proportion to the individual | |
1943 | * zone size to ensure fair page aging. The zone a | |
1944 | * page was allocated in should have no effect on the | |
1945 | * time the page has in memory before being reclaimed. | |
81c0a2bb | 1946 | */ |
3a025760 | 1947 | if (alloc_flags & ALLOC_FAIR) { |
fff4068c | 1948 | if (!zone_local(preferred_zone, zone)) |
81c0a2bb | 1949 | continue; |
3a025760 JW |
1950 | if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0) |
1951 | continue; | |
81c0a2bb | 1952 | } |
a756cf59 JW |
1953 | /* |
1954 | * When allocating a page cache page for writing, we | |
1955 | * want to get it from a zone that is within its dirty | |
1956 | * limit, such that no single zone holds more than its | |
1957 | * proportional share of globally allowed dirty pages. | |
1958 | * The dirty limits take into account the zone's | |
1959 | * lowmem reserves and high watermark so that kswapd | |
1960 | * should be able to balance it without having to | |
1961 | * write pages from its LRU list. | |
1962 | * | |
1963 | * This may look like it could increase pressure on | |
1964 | * lower zones by failing allocations in higher zones | |
1965 | * before they are full. But the pages that do spill | |
1966 | * over are limited as the lower zones are protected | |
1967 | * by this very same mechanism. It should not become | |
1968 | * a practical burden to them. | |
1969 | * | |
1970 | * XXX: For now, allow allocations to potentially | |
1971 | * exceed the per-zone dirty limit in the slowpath | |
1972 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
1973 | * which is important when on a NUMA setup the allowed | |
1974 | * zones are together not big enough to reach the | |
1975 | * global limit. The proper fix for these situations | |
1976 | * will require awareness of zones in the | |
1977 | * dirty-throttling and the flusher threads. | |
1978 | */ | |
1979 | if ((alloc_flags & ALLOC_WMARK_LOW) && | |
1980 | (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone)) | |
800a1e75 | 1981 | continue; |
7fb1d9fc | 1982 | |
e085dbc5 JW |
1983 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
1984 | if (!zone_watermark_ok(zone, order, mark, | |
1985 | classzone_idx, alloc_flags)) { | |
fa5e084e MG |
1986 | int ret; |
1987 | ||
e5adfffc KS |
1988 | if (IS_ENABLED(CONFIG_NUMA) && |
1989 | !did_zlc_setup && nr_online_nodes > 1) { | |
cd38b115 MG |
1990 | /* |
1991 | * we do zlc_setup if there are multiple nodes | |
1992 | * and before considering the first zone allowed | |
1993 | * by the cpuset. | |
1994 | */ | |
1995 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1996 | zlc_active = 1; | |
1997 | did_zlc_setup = 1; | |
1998 | } | |
1999 | ||
957f822a DR |
2000 | if (zone_reclaim_mode == 0 || |
2001 | !zone_allows_reclaim(preferred_zone, zone)) | |
fa5e084e MG |
2002 | goto this_zone_full; |
2003 | ||
cd38b115 MG |
2004 | /* |
2005 | * As we may have just activated ZLC, check if the first | |
2006 | * eligible zone has failed zone_reclaim recently. | |
2007 | */ | |
e5adfffc | 2008 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
cd38b115 MG |
2009 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
2010 | continue; | |
2011 | ||
fa5e084e MG |
2012 | ret = zone_reclaim(zone, gfp_mask, order); |
2013 | switch (ret) { | |
2014 | case ZONE_RECLAIM_NOSCAN: | |
2015 | /* did not scan */ | |
cd38b115 | 2016 | continue; |
fa5e084e MG |
2017 | case ZONE_RECLAIM_FULL: |
2018 | /* scanned but unreclaimable */ | |
cd38b115 | 2019 | continue; |
fa5e084e MG |
2020 | default: |
2021 | /* did we reclaim enough */ | |
fed2719e | 2022 | if (zone_watermark_ok(zone, order, mark, |
fa5e084e | 2023 | classzone_idx, alloc_flags)) |
fed2719e MG |
2024 | goto try_this_zone; |
2025 | ||
2026 | /* | |
2027 | * Failed to reclaim enough to meet watermark. | |
2028 | * Only mark the zone full if checking the min | |
2029 | * watermark or if we failed to reclaim just | |
2030 | * 1<<order pages or else the page allocator | |
2031 | * fastpath will prematurely mark zones full | |
2032 | * when the watermark is between the low and | |
2033 | * min watermarks. | |
2034 | */ | |
2035 | if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) || | |
2036 | ret == ZONE_RECLAIM_SOME) | |
9276b1bc | 2037 | goto this_zone_full; |
fed2719e MG |
2038 | |
2039 | continue; | |
0798e519 | 2040 | } |
7fb1d9fc RS |
2041 | } |
2042 | ||
fa5e084e | 2043 | try_this_zone: |
3dd28266 MG |
2044 | page = buffered_rmqueue(preferred_zone, zone, order, |
2045 | gfp_mask, migratetype); | |
0798e519 | 2046 | if (page) |
7fb1d9fc | 2047 | break; |
9276b1bc | 2048 | this_zone_full: |
65bb3719 | 2049 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active) |
9276b1bc | 2050 | zlc_mark_zone_full(zonelist, z); |
54a6eb5c | 2051 | } |
9276b1bc | 2052 | |
e5adfffc | 2053 | if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) { |
9276b1bc PJ |
2054 | /* Disable zlc cache for second zonelist scan */ |
2055 | zlc_active = 0; | |
2056 | goto zonelist_scan; | |
2057 | } | |
b121186a AS |
2058 | |
2059 | if (page) | |
2060 | /* | |
2061 | * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was | |
2062 | * necessary to allocate the page. The expectation is | |
2063 | * that the caller is taking steps that will free more | |
2064 | * memory. The caller should avoid the page being used | |
2065 | * for !PFMEMALLOC purposes. | |
2066 | */ | |
2067 | page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS); | |
2068 | ||
7fb1d9fc | 2069 | return page; |
753ee728 MH |
2070 | } |
2071 | ||
29423e77 DR |
2072 | /* |
2073 | * Large machines with many possible nodes should not always dump per-node | |
2074 | * meminfo in irq context. | |
2075 | */ | |
2076 | static inline bool should_suppress_show_mem(void) | |
2077 | { | |
2078 | bool ret = false; | |
2079 | ||
2080 | #if NODES_SHIFT > 8 | |
2081 | ret = in_interrupt(); | |
2082 | #endif | |
2083 | return ret; | |
2084 | } | |
2085 | ||
a238ab5b DH |
2086 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2087 | DEFAULT_RATELIMIT_INTERVAL, | |
2088 | DEFAULT_RATELIMIT_BURST); | |
2089 | ||
2090 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
2091 | { | |
a238ab5b DH |
2092 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2093 | ||
c0a32fc5 SG |
2094 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2095 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2096 | return; |
2097 | ||
2098 | /* | |
2099 | * This documents exceptions given to allocations in certain | |
2100 | * contexts that are allowed to allocate outside current's set | |
2101 | * of allowed nodes. | |
2102 | */ | |
2103 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2104 | if (test_thread_flag(TIF_MEMDIE) || | |
2105 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2106 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2107 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
2108 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2109 | ||
2110 | if (fmt) { | |
3ee9a4f0 JP |
2111 | struct va_format vaf; |
2112 | va_list args; | |
2113 | ||
a238ab5b | 2114 | va_start(args, fmt); |
3ee9a4f0 JP |
2115 | |
2116 | vaf.fmt = fmt; | |
2117 | vaf.va = &args; | |
2118 | ||
2119 | pr_warn("%pV", &vaf); | |
2120 | ||
a238ab5b DH |
2121 | va_end(args); |
2122 | } | |
2123 | ||
3ee9a4f0 JP |
2124 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
2125 | current->comm, order, gfp_mask); | |
a238ab5b DH |
2126 | |
2127 | dump_stack(); | |
2128 | if (!should_suppress_show_mem()) | |
2129 | show_mem(filter); | |
2130 | } | |
2131 | ||
11e33f6a MG |
2132 | static inline int |
2133 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
f90ac398 | 2134 | unsigned long did_some_progress, |
11e33f6a | 2135 | unsigned long pages_reclaimed) |
1da177e4 | 2136 | { |
11e33f6a MG |
2137 | /* Do not loop if specifically requested */ |
2138 | if (gfp_mask & __GFP_NORETRY) | |
2139 | return 0; | |
1da177e4 | 2140 | |
f90ac398 MG |
2141 | /* Always retry if specifically requested */ |
2142 | if (gfp_mask & __GFP_NOFAIL) | |
2143 | return 1; | |
2144 | ||
2145 | /* | |
2146 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim | |
2147 | * making forward progress without invoking OOM. Suspend also disables | |
2148 | * storage devices so kswapd will not help. Bail if we are suspending. | |
2149 | */ | |
2150 | if (!did_some_progress && pm_suspended_storage()) | |
2151 | return 0; | |
2152 | ||
11e33f6a MG |
2153 | /* |
2154 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
2155 | * means __GFP_NOFAIL, but that may not be true in other | |
2156 | * implementations. | |
2157 | */ | |
2158 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
2159 | return 1; | |
2160 | ||
2161 | /* | |
2162 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
2163 | * specified, then we retry until we no longer reclaim any pages | |
2164 | * (above), or we've reclaimed an order of pages at least as | |
2165 | * large as the allocation's order. In both cases, if the | |
2166 | * allocation still fails, we stop retrying. | |
2167 | */ | |
2168 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
2169 | return 1; | |
cf40bd16 | 2170 | |
11e33f6a MG |
2171 | return 0; |
2172 | } | |
933e312e | 2173 | |
11e33f6a MG |
2174 | static inline struct page * |
2175 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
2176 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2177 | nodemask_t *nodemask, struct zone *preferred_zone, |
2178 | int migratetype) | |
11e33f6a MG |
2179 | { |
2180 | struct page *page; | |
2181 | ||
2182 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
ff321fea | 2183 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
11e33f6a | 2184 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2185 | return NULL; |
2186 | } | |
6b1de916 | 2187 | |
11e33f6a MG |
2188 | /* |
2189 | * Go through the zonelist yet one more time, keep very high watermark | |
2190 | * here, this is only to catch a parallel oom killing, we must fail if | |
2191 | * we're still under heavy pressure. | |
2192 | */ | |
2193 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
2194 | order, zonelist, high_zoneidx, | |
5117f45d | 2195 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 2196 | preferred_zone, migratetype); |
7fb1d9fc | 2197 | if (page) |
11e33f6a MG |
2198 | goto out; |
2199 | ||
4365a567 KH |
2200 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2201 | /* The OOM killer will not help higher order allocs */ | |
2202 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2203 | goto out; | |
03668b3c DR |
2204 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
2205 | if (high_zoneidx < ZONE_NORMAL) | |
2206 | goto out; | |
4365a567 KH |
2207 | /* |
2208 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
2209 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
2210 | * The caller should handle page allocation failure by itself if | |
2211 | * it specifies __GFP_THISNODE. | |
2212 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
2213 | */ | |
2214 | if (gfp_mask & __GFP_THISNODE) | |
2215 | goto out; | |
2216 | } | |
11e33f6a | 2217 | /* Exhausted what can be done so it's blamo time */ |
08ab9b10 | 2218 | out_of_memory(zonelist, gfp_mask, order, nodemask, false); |
11e33f6a MG |
2219 | |
2220 | out: | |
2221 | clear_zonelist_oom(zonelist, gfp_mask); | |
2222 | return page; | |
2223 | } | |
2224 | ||
56de7263 MG |
2225 | #ifdef CONFIG_COMPACTION |
2226 | /* Try memory compaction for high-order allocations before reclaim */ | |
2227 | static struct page * | |
2228 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2229 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2230 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
e0b9daeb | 2231 | int migratetype, enum migrate_mode mode, |
c67fe375 | 2232 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2233 | unsigned long *did_some_progress) |
56de7263 | 2234 | { |
66199712 | 2235 | if (!order) |
56de7263 MG |
2236 | return NULL; |
2237 | ||
aff62249 | 2238 | if (compaction_deferred(preferred_zone, order)) { |
66199712 MG |
2239 | *deferred_compaction = true; |
2240 | return NULL; | |
2241 | } | |
2242 | ||
c06b1fca | 2243 | current->flags |= PF_MEMALLOC; |
56de7263 | 2244 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
e0b9daeb | 2245 | nodemask, mode, |
8fb74b9f | 2246 | contended_compaction); |
c06b1fca | 2247 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2248 | |
1fb3f8ca | 2249 | if (*did_some_progress != COMPACT_SKIPPED) { |
8fb74b9f MG |
2250 | struct page *page; |
2251 | ||
56de7263 MG |
2252 | /* Page migration frees to the PCP lists but we want merging */ |
2253 | drain_pages(get_cpu()); | |
2254 | put_cpu(); | |
2255 | ||
2256 | page = get_page_from_freelist(gfp_mask, nodemask, | |
2257 | order, zonelist, high_zoneidx, | |
cfd19c5a MG |
2258 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2259 | preferred_zone, migratetype); | |
56de7263 | 2260 | if (page) { |
62997027 | 2261 | preferred_zone->compact_blockskip_flush = false; |
de6c60a6 | 2262 | compaction_defer_reset(preferred_zone, order, true); |
56de7263 MG |
2263 | count_vm_event(COMPACTSUCCESS); |
2264 | return page; | |
2265 | } | |
2266 | ||
2267 | /* | |
2268 | * It's bad if compaction run occurs and fails. | |
2269 | * The most likely reason is that pages exist, | |
2270 | * but not enough to satisfy watermarks. | |
2271 | */ | |
2272 | count_vm_event(COMPACTFAIL); | |
66199712 MG |
2273 | |
2274 | /* | |
2275 | * As async compaction considers a subset of pageblocks, only | |
2276 | * defer if the failure was a sync compaction failure. | |
2277 | */ | |
e0b9daeb | 2278 | if (mode != MIGRATE_ASYNC) |
aff62249 | 2279 | defer_compaction(preferred_zone, order); |
56de7263 MG |
2280 | |
2281 | cond_resched(); | |
2282 | } | |
2283 | ||
2284 | return NULL; | |
2285 | } | |
2286 | #else | |
2287 | static inline struct page * | |
2288 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2289 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2290 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
e0b9daeb DR |
2291 | int migratetype, enum migrate_mode mode, bool *contended_compaction, |
2292 | bool *deferred_compaction, unsigned long *did_some_progress) | |
56de7263 MG |
2293 | { |
2294 | return NULL; | |
2295 | } | |
2296 | #endif /* CONFIG_COMPACTION */ | |
2297 | ||
bba90710 MS |
2298 | /* Perform direct synchronous page reclaim */ |
2299 | static int | |
2300 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, | |
2301 | nodemask_t *nodemask) | |
11e33f6a | 2302 | { |
11e33f6a | 2303 | struct reclaim_state reclaim_state; |
bba90710 | 2304 | int progress; |
11e33f6a MG |
2305 | |
2306 | cond_resched(); | |
2307 | ||
2308 | /* We now go into synchronous reclaim */ | |
2309 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2310 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2311 | lockdep_set_current_reclaim_state(gfp_mask); |
2312 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2313 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2314 | |
bba90710 | 2315 | progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
11e33f6a | 2316 | |
c06b1fca | 2317 | current->reclaim_state = NULL; |
11e33f6a | 2318 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2319 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2320 | |
2321 | cond_resched(); | |
2322 | ||
bba90710 MS |
2323 | return progress; |
2324 | } | |
2325 | ||
2326 | /* The really slow allocator path where we enter direct reclaim */ | |
2327 | static inline struct page * | |
2328 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
2329 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2330 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
2331 | int migratetype, unsigned long *did_some_progress) | |
2332 | { | |
2333 | struct page *page = NULL; | |
2334 | bool drained = false; | |
2335 | ||
2336 | *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist, | |
2337 | nodemask); | |
9ee493ce MG |
2338 | if (unlikely(!(*did_some_progress))) |
2339 | return NULL; | |
11e33f6a | 2340 | |
76d3fbf8 | 2341 | /* After successful reclaim, reconsider all zones for allocation */ |
e5adfffc | 2342 | if (IS_ENABLED(CONFIG_NUMA)) |
76d3fbf8 MG |
2343 | zlc_clear_zones_full(zonelist); |
2344 | ||
9ee493ce MG |
2345 | retry: |
2346 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2347 | zonelist, high_zoneidx, |
cfd19c5a MG |
2348 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2349 | preferred_zone, migratetype); | |
9ee493ce MG |
2350 | |
2351 | /* | |
2352 | * If an allocation failed after direct reclaim, it could be because | |
2353 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2354 | */ | |
2355 | if (!page && !drained) { | |
2356 | drain_all_pages(); | |
2357 | drained = true; | |
2358 | goto retry; | |
2359 | } | |
2360 | ||
11e33f6a MG |
2361 | return page; |
2362 | } | |
2363 | ||
1da177e4 | 2364 | /* |
11e33f6a MG |
2365 | * This is called in the allocator slow-path if the allocation request is of |
2366 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2367 | */ |
11e33f6a MG |
2368 | static inline struct page * |
2369 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
2370 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2371 | nodemask_t *nodemask, struct zone *preferred_zone, |
2372 | int migratetype) | |
11e33f6a MG |
2373 | { |
2374 | struct page *page; | |
2375 | ||
2376 | do { | |
2377 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2378 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 2379 | preferred_zone, migratetype); |
11e33f6a MG |
2380 | |
2381 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 2382 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
2383 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2384 | ||
2385 | return page; | |
2386 | } | |
2387 | ||
3a025760 JW |
2388 | static void reset_alloc_batches(struct zonelist *zonelist, |
2389 | enum zone_type high_zoneidx, | |
2390 | struct zone *preferred_zone) | |
1da177e4 | 2391 | { |
dd1a239f MG |
2392 | struct zoneref *z; |
2393 | struct zone *zone; | |
1da177e4 | 2394 | |
81c0a2bb | 2395 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81c0a2bb JW |
2396 | /* |
2397 | * Only reset the batches of zones that were actually | |
3a025760 JW |
2398 | * considered in the fairness pass, we don't want to |
2399 | * trash fairness information for zones that are not | |
81c0a2bb JW |
2400 | * actually part of this zonelist's round-robin cycle. |
2401 | */ | |
fff4068c | 2402 | if (!zone_local(preferred_zone, zone)) |
81c0a2bb JW |
2403 | continue; |
2404 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
3a025760 JW |
2405 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
2406 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
81c0a2bb | 2407 | } |
11e33f6a | 2408 | } |
cf40bd16 | 2409 | |
3a025760 JW |
2410 | static void wake_all_kswapds(unsigned int order, |
2411 | struct zonelist *zonelist, | |
2412 | enum zone_type high_zoneidx, | |
2413 | struct zone *preferred_zone) | |
2414 | { | |
2415 | struct zoneref *z; | |
2416 | struct zone *zone; | |
2417 | ||
2418 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) | |
2419 | wakeup_kswapd(zone, order, zone_idx(preferred_zone)); | |
2420 | } | |
2421 | ||
341ce06f PZ |
2422 | static inline int |
2423 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2424 | { | |
341ce06f PZ |
2425 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
2426 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 2427 | |
a56f57ff | 2428 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2429 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2430 | |
341ce06f PZ |
2431 | /* |
2432 | * The caller may dip into page reserves a bit more if the caller | |
2433 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2434 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
2435 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
2436 | */ | |
e6223a3b | 2437 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2438 | |
341ce06f | 2439 | if (!wait) { |
5c3240d9 AA |
2440 | /* |
2441 | * Not worth trying to allocate harder for | |
2442 | * __GFP_NOMEMALLOC even if it can't schedule. | |
2443 | */ | |
2444 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2445 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 2446 | /* |
341ce06f PZ |
2447 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
2448 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 2449 | */ |
341ce06f | 2450 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2451 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2452 | alloc_flags |= ALLOC_HARDER; |
2453 | ||
b37f1dd0 MG |
2454 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
2455 | if (gfp_mask & __GFP_MEMALLOC) | |
2456 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
2457 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
2458 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
2459 | else if (!in_interrupt() && | |
2460 | ((current->flags & PF_MEMALLOC) || | |
2461 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 2462 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 2463 | } |
d95ea5d1 BZ |
2464 | #ifdef CONFIG_CMA |
2465 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2466 | alloc_flags |= ALLOC_CMA; | |
2467 | #endif | |
341ce06f PZ |
2468 | return alloc_flags; |
2469 | } | |
2470 | ||
072bb0aa MG |
2471 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
2472 | { | |
b37f1dd0 | 2473 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
2474 | } |
2475 | ||
11e33f6a MG |
2476 | static inline struct page * |
2477 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2478 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2479 | nodemask_t *nodemask, struct zone *preferred_zone, |
2480 | int migratetype) | |
11e33f6a MG |
2481 | { |
2482 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2483 | struct page *page = NULL; | |
2484 | int alloc_flags; | |
2485 | unsigned long pages_reclaimed = 0; | |
2486 | unsigned long did_some_progress; | |
e0b9daeb | 2487 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
66199712 | 2488 | bool deferred_compaction = false; |
c67fe375 | 2489 | bool contended_compaction = false; |
1da177e4 | 2490 | |
72807a74 MG |
2491 | /* |
2492 | * In the slowpath, we sanity check order to avoid ever trying to | |
2493 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2494 | * be using allocators in order of preference for an area that is | |
2495 | * too large. | |
2496 | */ | |
1fc28b70 MG |
2497 | if (order >= MAX_ORDER) { |
2498 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2499 | return NULL; |
1fc28b70 | 2500 | } |
1da177e4 | 2501 | |
952f3b51 CL |
2502 | /* |
2503 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2504 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2505 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2506 | * using a larger set of nodes after it has established that the | |
2507 | * allowed per node queues are empty and that nodes are | |
2508 | * over allocated. | |
2509 | */ | |
3a025760 JW |
2510 | if (IS_ENABLED(CONFIG_NUMA) && |
2511 | (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
952f3b51 CL |
2512 | goto nopage; |
2513 | ||
cc4a6851 | 2514 | restart: |
3a025760 JW |
2515 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
2516 | wake_all_kswapds(order, zonelist, high_zoneidx, preferred_zone); | |
1da177e4 | 2517 | |
9bf2229f | 2518 | /* |
7fb1d9fc RS |
2519 | * OK, we're below the kswapd watermark and have kicked background |
2520 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2521 | * to how we want to proceed. | |
9bf2229f | 2522 | */ |
341ce06f | 2523 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2524 | |
f33261d7 DR |
2525 | /* |
2526 | * Find the true preferred zone if the allocation is unconstrained by | |
2527 | * cpusets. | |
2528 | */ | |
2529 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) | |
2530 | first_zones_zonelist(zonelist, high_zoneidx, NULL, | |
2531 | &preferred_zone); | |
2532 | ||
cfa54a0f | 2533 | rebalance: |
341ce06f | 2534 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2535 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
2536 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
2537 | preferred_zone, migratetype); | |
7fb1d9fc RS |
2538 | if (page) |
2539 | goto got_pg; | |
1da177e4 | 2540 | |
11e33f6a | 2541 | /* Allocate without watermarks if the context allows */ |
341ce06f | 2542 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
2543 | /* |
2544 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
2545 | * the allocation is high priority and these type of | |
2546 | * allocations are system rather than user orientated | |
2547 | */ | |
2548 | zonelist = node_zonelist(numa_node_id(), gfp_mask); | |
2549 | ||
341ce06f PZ |
2550 | page = __alloc_pages_high_priority(gfp_mask, order, |
2551 | zonelist, high_zoneidx, nodemask, | |
2552 | preferred_zone, migratetype); | |
cfd19c5a | 2553 | if (page) { |
341ce06f | 2554 | goto got_pg; |
cfd19c5a | 2555 | } |
1da177e4 LT |
2556 | } |
2557 | ||
2558 | /* Atomic allocations - we can't balance anything */ | |
aed0a0e3 DR |
2559 | if (!wait) { |
2560 | /* | |
2561 | * All existing users of the deprecated __GFP_NOFAIL are | |
2562 | * blockable, so warn of any new users that actually allow this | |
2563 | * type of allocation to fail. | |
2564 | */ | |
2565 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 2566 | goto nopage; |
aed0a0e3 | 2567 | } |
1da177e4 | 2568 | |
341ce06f | 2569 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2570 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2571 | goto nopage; |
2572 | ||
6583bb64 DR |
2573 | /* Avoid allocations with no watermarks from looping endlessly */ |
2574 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2575 | goto nopage; | |
2576 | ||
77f1fe6b MG |
2577 | /* |
2578 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2579 | * attempts after direct reclaim are synchronous | |
2580 | */ | |
e0b9daeb DR |
2581 | page = __alloc_pages_direct_compact(gfp_mask, order, zonelist, |
2582 | high_zoneidx, nodemask, alloc_flags, | |
2583 | preferred_zone, migratetype, | |
2584 | migration_mode, &contended_compaction, | |
66199712 MG |
2585 | &deferred_compaction, |
2586 | &did_some_progress); | |
56de7263 MG |
2587 | if (page) |
2588 | goto got_pg; | |
75f30861 DR |
2589 | |
2590 | /* | |
2591 | * It can become very expensive to allocate transparent hugepages at | |
2592 | * fault, so use asynchronous memory compaction for THP unless it is | |
2593 | * khugepaged trying to collapse. | |
2594 | */ | |
2595 | if (!(gfp_mask & __GFP_NO_KSWAPD) || (current->flags & PF_KTHREAD)) | |
2596 | migration_mode = MIGRATE_SYNC_LIGHT; | |
56de7263 | 2597 | |
31f8d42d LT |
2598 | /* |
2599 | * If compaction is deferred for high-order allocations, it is because | |
2600 | * sync compaction recently failed. In this is the case and the caller | |
2601 | * requested a movable allocation that does not heavily disrupt the | |
2602 | * system then fail the allocation instead of entering direct reclaim. | |
2603 | */ | |
2604 | if ((deferred_compaction || contended_compaction) && | |
caf49191 | 2605 | (gfp_mask & __GFP_NO_KSWAPD)) |
31f8d42d | 2606 | goto nopage; |
66199712 | 2607 | |
11e33f6a MG |
2608 | /* Try direct reclaim and then allocating */ |
2609 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2610 | zonelist, high_zoneidx, | |
2611 | nodemask, | |
5117f45d | 2612 | alloc_flags, preferred_zone, |
3dd28266 | 2613 | migratetype, &did_some_progress); |
11e33f6a MG |
2614 | if (page) |
2615 | goto got_pg; | |
1da177e4 | 2616 | |
e33c3b5e | 2617 | /* |
11e33f6a MG |
2618 | * If we failed to make any progress reclaiming, then we are |
2619 | * running out of options and have to consider going OOM | |
e33c3b5e | 2620 | */ |
11e33f6a | 2621 | if (!did_some_progress) { |
b9921ecd | 2622 | if (oom_gfp_allowed(gfp_mask)) { |
7f33d49a RW |
2623 | if (oom_killer_disabled) |
2624 | goto nopage; | |
29fd66d2 DR |
2625 | /* Coredumps can quickly deplete all memory reserves */ |
2626 | if ((current->flags & PF_DUMPCORE) && | |
2627 | !(gfp_mask & __GFP_NOFAIL)) | |
2628 | goto nopage; | |
11e33f6a MG |
2629 | page = __alloc_pages_may_oom(gfp_mask, order, |
2630 | zonelist, high_zoneidx, | |
3dd28266 MG |
2631 | nodemask, preferred_zone, |
2632 | migratetype); | |
11e33f6a MG |
2633 | if (page) |
2634 | goto got_pg; | |
1da177e4 | 2635 | |
03668b3c DR |
2636 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2637 | /* | |
2638 | * The oom killer is not called for high-order | |
2639 | * allocations that may fail, so if no progress | |
2640 | * is being made, there are no other options and | |
2641 | * retrying is unlikely to help. | |
2642 | */ | |
2643 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2644 | goto nopage; | |
2645 | /* | |
2646 | * The oom killer is not called for lowmem | |
2647 | * allocations to prevent needlessly killing | |
2648 | * innocent tasks. | |
2649 | */ | |
2650 | if (high_zoneidx < ZONE_NORMAL) | |
2651 | goto nopage; | |
2652 | } | |
e2c55dc8 | 2653 | |
ff0ceb9d DR |
2654 | goto restart; |
2655 | } | |
1da177e4 LT |
2656 | } |
2657 | ||
11e33f6a | 2658 | /* Check if we should retry the allocation */ |
a41f24ea | 2659 | pages_reclaimed += did_some_progress; |
f90ac398 MG |
2660 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
2661 | pages_reclaimed)) { | |
11e33f6a | 2662 | /* Wait for some write requests to complete then retry */ |
0e093d99 | 2663 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2664 | goto rebalance; |
3e7d3449 MG |
2665 | } else { |
2666 | /* | |
2667 | * High-order allocations do not necessarily loop after | |
2668 | * direct reclaim and reclaim/compaction depends on compaction | |
2669 | * being called after reclaim so call directly if necessary | |
2670 | */ | |
e0b9daeb DR |
2671 | page = __alloc_pages_direct_compact(gfp_mask, order, zonelist, |
2672 | high_zoneidx, nodemask, alloc_flags, | |
2673 | preferred_zone, migratetype, | |
2674 | migration_mode, &contended_compaction, | |
66199712 MG |
2675 | &deferred_compaction, |
2676 | &did_some_progress); | |
3e7d3449 MG |
2677 | if (page) |
2678 | goto got_pg; | |
1da177e4 LT |
2679 | } |
2680 | ||
2681 | nopage: | |
a238ab5b | 2682 | warn_alloc_failed(gfp_mask, order, NULL); |
b1eeab67 | 2683 | return page; |
1da177e4 | 2684 | got_pg: |
b1eeab67 VN |
2685 | if (kmemcheck_enabled) |
2686 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
11e33f6a | 2687 | |
072bb0aa | 2688 | return page; |
1da177e4 | 2689 | } |
11e33f6a MG |
2690 | |
2691 | /* | |
2692 | * This is the 'heart' of the zoned buddy allocator. | |
2693 | */ | |
2694 | struct page * | |
2695 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2696 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2697 | { | |
2698 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2699 | struct zone *preferred_zone; |
cc9a6c87 | 2700 | struct page *page = NULL; |
3dd28266 | 2701 | int migratetype = allocflags_to_migratetype(gfp_mask); |
cc9a6c87 | 2702 | unsigned int cpuset_mems_cookie; |
3a025760 | 2703 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR; |
11e33f6a | 2704 | |
dcce284a BH |
2705 | gfp_mask &= gfp_allowed_mask; |
2706 | ||
11e33f6a MG |
2707 | lockdep_trace_alloc(gfp_mask); |
2708 | ||
2709 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2710 | ||
2711 | if (should_fail_alloc_page(gfp_mask, order)) | |
2712 | return NULL; | |
2713 | ||
2714 | /* | |
2715 | * Check the zones suitable for the gfp_mask contain at least one | |
2716 | * valid zone. It's possible to have an empty zonelist as a result | |
2717 | * of GFP_THISNODE and a memoryless node | |
2718 | */ | |
2719 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2720 | return NULL; | |
2721 | ||
cc9a6c87 | 2722 | retry_cpuset: |
d26914d1 | 2723 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 2724 | |
5117f45d | 2725 | /* The preferred zone is used for statistics later */ |
f33261d7 DR |
2726 | first_zones_zonelist(zonelist, high_zoneidx, |
2727 | nodemask ? : &cpuset_current_mems_allowed, | |
2728 | &preferred_zone); | |
cc9a6c87 MG |
2729 | if (!preferred_zone) |
2730 | goto out; | |
5117f45d | 2731 | |
d95ea5d1 BZ |
2732 | #ifdef CONFIG_CMA |
2733 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2734 | alloc_flags |= ALLOC_CMA; | |
2735 | #endif | |
3a025760 | 2736 | retry: |
5117f45d | 2737 | /* First allocation attempt */ |
11e33f6a | 2738 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
d95ea5d1 | 2739 | zonelist, high_zoneidx, alloc_flags, |
3dd28266 | 2740 | preferred_zone, migratetype); |
21caf2fc | 2741 | if (unlikely(!page)) { |
3a025760 JW |
2742 | /* |
2743 | * The first pass makes sure allocations are spread | |
2744 | * fairly within the local node. However, the local | |
2745 | * node might have free pages left after the fairness | |
2746 | * batches are exhausted, and remote zones haven't | |
2747 | * even been considered yet. Try once more without | |
2748 | * fairness, and include remote zones now, before | |
2749 | * entering the slowpath and waking kswapd: prefer | |
2750 | * spilling to a remote zone over swapping locally. | |
2751 | */ | |
2752 | if (alloc_flags & ALLOC_FAIR) { | |
2753 | reset_alloc_batches(zonelist, high_zoneidx, | |
2754 | preferred_zone); | |
2755 | alloc_flags &= ~ALLOC_FAIR; | |
2756 | goto retry; | |
2757 | } | |
21caf2fc ML |
2758 | /* |
2759 | * Runtime PM, block IO and its error handling path | |
2760 | * can deadlock because I/O on the device might not | |
2761 | * complete. | |
2762 | */ | |
2763 | gfp_mask = memalloc_noio_flags(gfp_mask); | |
11e33f6a | 2764 | page = __alloc_pages_slowpath(gfp_mask, order, |
5117f45d | 2765 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 2766 | preferred_zone, migratetype); |
21caf2fc | 2767 | } |
11e33f6a | 2768 | |
4b4f278c | 2769 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
cc9a6c87 MG |
2770 | |
2771 | out: | |
2772 | /* | |
2773 | * When updating a task's mems_allowed, it is possible to race with | |
2774 | * parallel threads in such a way that an allocation can fail while | |
2775 | * the mask is being updated. If a page allocation is about to fail, | |
2776 | * check if the cpuset changed during allocation and if so, retry. | |
2777 | */ | |
d26914d1 | 2778 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
2779 | goto retry_cpuset; |
2780 | ||
11e33f6a | 2781 | return page; |
1da177e4 | 2782 | } |
d239171e | 2783 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2784 | |
2785 | /* | |
2786 | * Common helper functions. | |
2787 | */ | |
920c7a5d | 2788 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2789 | { |
945a1113 AM |
2790 | struct page *page; |
2791 | ||
2792 | /* | |
2793 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2794 | * a highmem page | |
2795 | */ | |
2796 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2797 | ||
1da177e4 LT |
2798 | page = alloc_pages(gfp_mask, order); |
2799 | if (!page) | |
2800 | return 0; | |
2801 | return (unsigned long) page_address(page); | |
2802 | } | |
1da177e4 LT |
2803 | EXPORT_SYMBOL(__get_free_pages); |
2804 | ||
920c7a5d | 2805 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2806 | { |
945a1113 | 2807 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2808 | } |
1da177e4 LT |
2809 | EXPORT_SYMBOL(get_zeroed_page); |
2810 | ||
920c7a5d | 2811 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2812 | { |
b5810039 | 2813 | if (put_page_testzero(page)) { |
1da177e4 | 2814 | if (order == 0) |
fc91668e | 2815 | free_hot_cold_page(page, 0); |
1da177e4 LT |
2816 | else |
2817 | __free_pages_ok(page, order); | |
2818 | } | |
2819 | } | |
2820 | ||
2821 | EXPORT_SYMBOL(__free_pages); | |
2822 | ||
920c7a5d | 2823 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2824 | { |
2825 | if (addr != 0) { | |
725d704e | 2826 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2827 | __free_pages(virt_to_page((void *)addr), order); |
2828 | } | |
2829 | } | |
2830 | ||
2831 | EXPORT_SYMBOL(free_pages); | |
2832 | ||
6a1a0d3b | 2833 | /* |
52383431 VD |
2834 | * alloc_kmem_pages charges newly allocated pages to the kmem resource counter |
2835 | * of the current memory cgroup. | |
6a1a0d3b | 2836 | * |
52383431 VD |
2837 | * It should be used when the caller would like to use kmalloc, but since the |
2838 | * allocation is large, it has to fall back to the page allocator. | |
2839 | */ | |
2840 | struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order) | |
2841 | { | |
2842 | struct page *page; | |
2843 | struct mem_cgroup *memcg = NULL; | |
2844 | ||
2845 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
2846 | return NULL; | |
2847 | page = alloc_pages(gfp_mask, order); | |
2848 | memcg_kmem_commit_charge(page, memcg, order); | |
2849 | return page; | |
2850 | } | |
2851 | ||
2852 | struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
2853 | { | |
2854 | struct page *page; | |
2855 | struct mem_cgroup *memcg = NULL; | |
2856 | ||
2857 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
2858 | return NULL; | |
2859 | page = alloc_pages_node(nid, gfp_mask, order); | |
2860 | memcg_kmem_commit_charge(page, memcg, order); | |
2861 | return page; | |
2862 | } | |
2863 | ||
2864 | /* | |
2865 | * __free_kmem_pages and free_kmem_pages will free pages allocated with | |
2866 | * alloc_kmem_pages. | |
6a1a0d3b | 2867 | */ |
52383431 | 2868 | void __free_kmem_pages(struct page *page, unsigned int order) |
6a1a0d3b GC |
2869 | { |
2870 | memcg_kmem_uncharge_pages(page, order); | |
2871 | __free_pages(page, order); | |
2872 | } | |
2873 | ||
52383431 | 2874 | void free_kmem_pages(unsigned long addr, unsigned int order) |
6a1a0d3b GC |
2875 | { |
2876 | if (addr != 0) { | |
2877 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
52383431 | 2878 | __free_kmem_pages(virt_to_page((void *)addr), order); |
6a1a0d3b GC |
2879 | } |
2880 | } | |
2881 | ||
ee85c2e1 AK |
2882 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
2883 | { | |
2884 | if (addr) { | |
2885 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2886 | unsigned long used = addr + PAGE_ALIGN(size); | |
2887 | ||
2888 | split_page(virt_to_page((void *)addr), order); | |
2889 | while (used < alloc_end) { | |
2890 | free_page(used); | |
2891 | used += PAGE_SIZE; | |
2892 | } | |
2893 | } | |
2894 | return (void *)addr; | |
2895 | } | |
2896 | ||
2be0ffe2 TT |
2897 | /** |
2898 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2899 | * @size: the number of bytes to allocate | |
2900 | * @gfp_mask: GFP flags for the allocation | |
2901 | * | |
2902 | * This function is similar to alloc_pages(), except that it allocates the | |
2903 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2904 | * allocate memory in power-of-two pages. | |
2905 | * | |
2906 | * This function is also limited by MAX_ORDER. | |
2907 | * | |
2908 | * Memory allocated by this function must be released by free_pages_exact(). | |
2909 | */ | |
2910 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2911 | { | |
2912 | unsigned int order = get_order(size); | |
2913 | unsigned long addr; | |
2914 | ||
2915 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 2916 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
2917 | } |
2918 | EXPORT_SYMBOL(alloc_pages_exact); | |
2919 | ||
ee85c2e1 AK |
2920 | /** |
2921 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
2922 | * pages on a node. | |
b5e6ab58 | 2923 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
2924 | * @size: the number of bytes to allocate |
2925 | * @gfp_mask: GFP flags for the allocation | |
2926 | * | |
2927 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
2928 | * back. | |
2929 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
2930 | * but is not exact. | |
2931 | */ | |
2932 | void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) | |
2933 | { | |
2934 | unsigned order = get_order(size); | |
2935 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
2936 | if (!p) | |
2937 | return NULL; | |
2938 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
2939 | } | |
2940 | EXPORT_SYMBOL(alloc_pages_exact_nid); | |
2941 | ||
2be0ffe2 TT |
2942 | /** |
2943 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2944 | * @virt: the value returned by alloc_pages_exact. | |
2945 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2946 | * | |
2947 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2948 | */ | |
2949 | void free_pages_exact(void *virt, size_t size) | |
2950 | { | |
2951 | unsigned long addr = (unsigned long)virt; | |
2952 | unsigned long end = addr + PAGE_ALIGN(size); | |
2953 | ||
2954 | while (addr < end) { | |
2955 | free_page(addr); | |
2956 | addr += PAGE_SIZE; | |
2957 | } | |
2958 | } | |
2959 | EXPORT_SYMBOL(free_pages_exact); | |
2960 | ||
e0fb5815 ZY |
2961 | /** |
2962 | * nr_free_zone_pages - count number of pages beyond high watermark | |
2963 | * @offset: The zone index of the highest zone | |
2964 | * | |
2965 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
2966 | * high watermark within all zones at or below a given zone index. For each | |
2967 | * zone, the number of pages is calculated as: | |
834405c3 | 2968 | * managed_pages - high_pages |
e0fb5815 | 2969 | */ |
ebec3862 | 2970 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 2971 | { |
dd1a239f | 2972 | struct zoneref *z; |
54a6eb5c MG |
2973 | struct zone *zone; |
2974 | ||
e310fd43 | 2975 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 2976 | unsigned long sum = 0; |
1da177e4 | 2977 | |
0e88460d | 2978 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2979 | |
54a6eb5c | 2980 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 2981 | unsigned long size = zone->managed_pages; |
41858966 | 2982 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2983 | if (size > high) |
2984 | sum += size - high; | |
1da177e4 LT |
2985 | } |
2986 | ||
2987 | return sum; | |
2988 | } | |
2989 | ||
e0fb5815 ZY |
2990 | /** |
2991 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
2992 | * | |
2993 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
2994 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 2995 | */ |
ebec3862 | 2996 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 2997 | { |
af4ca457 | 2998 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2999 | } |
c2f1a551 | 3000 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 3001 | |
e0fb5815 ZY |
3002 | /** |
3003 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
3004 | * | |
3005 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
3006 | * high watermark within all zones. | |
1da177e4 | 3007 | */ |
ebec3862 | 3008 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 3009 | { |
2a1e274a | 3010 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 3011 | } |
08e0f6a9 CL |
3012 | |
3013 | static inline void show_node(struct zone *zone) | |
1da177e4 | 3014 | { |
e5adfffc | 3015 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 3016 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 3017 | } |
1da177e4 | 3018 | |
1da177e4 LT |
3019 | void si_meminfo(struct sysinfo *val) |
3020 | { | |
3021 | val->totalram = totalram_pages; | |
3022 | val->sharedram = 0; | |
d23ad423 | 3023 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 3024 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
3025 | val->totalhigh = totalhigh_pages; |
3026 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
3027 | val->mem_unit = PAGE_SIZE; |
3028 | } | |
3029 | ||
3030 | EXPORT_SYMBOL(si_meminfo); | |
3031 | ||
3032 | #ifdef CONFIG_NUMA | |
3033 | void si_meminfo_node(struct sysinfo *val, int nid) | |
3034 | { | |
cdd91a77 JL |
3035 | int zone_type; /* needs to be signed */ |
3036 | unsigned long managed_pages = 0; | |
1da177e4 LT |
3037 | pg_data_t *pgdat = NODE_DATA(nid); |
3038 | ||
cdd91a77 JL |
3039 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
3040 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
3041 | val->totalram = managed_pages; | |
d23ad423 | 3042 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 3043 | #ifdef CONFIG_HIGHMEM |
b40da049 | 3044 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages; |
d23ad423 CL |
3045 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
3046 | NR_FREE_PAGES); | |
98d2b0eb CL |
3047 | #else |
3048 | val->totalhigh = 0; | |
3049 | val->freehigh = 0; | |
3050 | #endif | |
1da177e4 LT |
3051 | val->mem_unit = PAGE_SIZE; |
3052 | } | |
3053 | #endif | |
3054 | ||
ddd588b5 | 3055 | /* |
7bf02ea2 DR |
3056 | * Determine whether the node should be displayed or not, depending on whether |
3057 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 3058 | */ |
7bf02ea2 | 3059 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
3060 | { |
3061 | bool ret = false; | |
cc9a6c87 | 3062 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
3063 | |
3064 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
3065 | goto out; | |
3066 | ||
cc9a6c87 | 3067 | do { |
d26914d1 | 3068 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3069 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 3070 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
3071 | out: |
3072 | return ret; | |
3073 | } | |
3074 | ||
1da177e4 LT |
3075 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
3076 | ||
377e4f16 RV |
3077 | static void show_migration_types(unsigned char type) |
3078 | { | |
3079 | static const char types[MIGRATE_TYPES] = { | |
3080 | [MIGRATE_UNMOVABLE] = 'U', | |
3081 | [MIGRATE_RECLAIMABLE] = 'E', | |
3082 | [MIGRATE_MOVABLE] = 'M', | |
3083 | [MIGRATE_RESERVE] = 'R', | |
3084 | #ifdef CONFIG_CMA | |
3085 | [MIGRATE_CMA] = 'C', | |
3086 | #endif | |
194159fb | 3087 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 3088 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 3089 | #endif |
377e4f16 RV |
3090 | }; |
3091 | char tmp[MIGRATE_TYPES + 1]; | |
3092 | char *p = tmp; | |
3093 | int i; | |
3094 | ||
3095 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
3096 | if (type & (1 << i)) | |
3097 | *p++ = types[i]; | |
3098 | } | |
3099 | ||
3100 | *p = '\0'; | |
3101 | printk("(%s) ", tmp); | |
3102 | } | |
3103 | ||
1da177e4 LT |
3104 | /* |
3105 | * Show free area list (used inside shift_scroll-lock stuff) | |
3106 | * We also calculate the percentage fragmentation. We do this by counting the | |
3107 | * memory on each free list with the exception of the first item on the list. | |
ddd588b5 DR |
3108 | * Suppresses nodes that are not allowed by current's cpuset if |
3109 | * SHOW_MEM_FILTER_NODES is passed. | |
1da177e4 | 3110 | */ |
7bf02ea2 | 3111 | void show_free_areas(unsigned int filter) |
1da177e4 | 3112 | { |
c7241913 | 3113 | int cpu; |
1da177e4 LT |
3114 | struct zone *zone; |
3115 | ||
ee99c71c | 3116 | for_each_populated_zone(zone) { |
7bf02ea2 | 3117 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3118 | continue; |
c7241913 JS |
3119 | show_node(zone); |
3120 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 3121 | |
6b482c67 | 3122 | for_each_online_cpu(cpu) { |
1da177e4 LT |
3123 | struct per_cpu_pageset *pageset; |
3124 | ||
99dcc3e5 | 3125 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 3126 | |
3dfa5721 CL |
3127 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
3128 | cpu, pageset->pcp.high, | |
3129 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
3130 | } |
3131 | } | |
3132 | ||
a731286d KM |
3133 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
3134 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 3135 | " unevictable:%lu" |
b76146ed | 3136 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 3137 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a BZ |
3138 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
3139 | " free_cma:%lu\n", | |
4f98a2fe | 3140 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 3141 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
3142 | global_page_state(NR_ISOLATED_ANON), |
3143 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 3144 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 3145 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 3146 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 3147 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 3148 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 3149 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 3150 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
3151 | global_page_state(NR_SLAB_RECLAIMABLE), |
3152 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 3153 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 3154 | global_page_state(NR_SHMEM), |
a25700a5 | 3155 | global_page_state(NR_PAGETABLE), |
d1ce749a BZ |
3156 | global_page_state(NR_BOUNCE), |
3157 | global_page_state(NR_FREE_CMA_PAGES)); | |
1da177e4 | 3158 | |
ee99c71c | 3159 | for_each_populated_zone(zone) { |
1da177e4 LT |
3160 | int i; |
3161 | ||
7bf02ea2 | 3162 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3163 | continue; |
1da177e4 LT |
3164 | show_node(zone); |
3165 | printk("%s" | |
3166 | " free:%lukB" | |
3167 | " min:%lukB" | |
3168 | " low:%lukB" | |
3169 | " high:%lukB" | |
4f98a2fe RR |
3170 | " active_anon:%lukB" |
3171 | " inactive_anon:%lukB" | |
3172 | " active_file:%lukB" | |
3173 | " inactive_file:%lukB" | |
7b854121 | 3174 | " unevictable:%lukB" |
a731286d KM |
3175 | " isolated(anon):%lukB" |
3176 | " isolated(file):%lukB" | |
1da177e4 | 3177 | " present:%lukB" |
9feedc9d | 3178 | " managed:%lukB" |
4a0aa73f KM |
3179 | " mlocked:%lukB" |
3180 | " dirty:%lukB" | |
3181 | " writeback:%lukB" | |
3182 | " mapped:%lukB" | |
4b02108a | 3183 | " shmem:%lukB" |
4a0aa73f KM |
3184 | " slab_reclaimable:%lukB" |
3185 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3186 | " kernel_stack:%lukB" |
4a0aa73f KM |
3187 | " pagetables:%lukB" |
3188 | " unstable:%lukB" | |
3189 | " bounce:%lukB" | |
d1ce749a | 3190 | " free_cma:%lukB" |
4a0aa73f | 3191 | " writeback_tmp:%lukB" |
1da177e4 LT |
3192 | " pages_scanned:%lu" |
3193 | " all_unreclaimable? %s" | |
3194 | "\n", | |
3195 | zone->name, | |
88f5acf8 | 3196 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3197 | K(min_wmark_pages(zone)), |
3198 | K(low_wmark_pages(zone)), | |
3199 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3200 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3201 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3202 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3203 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3204 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3205 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3206 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3207 | K(zone->present_pages), |
9feedc9d | 3208 | K(zone->managed_pages), |
4a0aa73f KM |
3209 | K(zone_page_state(zone, NR_MLOCK)), |
3210 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3211 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3212 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3213 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3214 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3215 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3216 | zone_page_state(zone, NR_KERNEL_STACK) * |
3217 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3218 | K(zone_page_state(zone, NR_PAGETABLE)), |
3219 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3220 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1ce749a | 3221 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3222 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
1da177e4 | 3223 | zone->pages_scanned, |
6e543d57 | 3224 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
3225 | ); |
3226 | printk("lowmem_reserve[]:"); | |
3227 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3228 | printk(" %lu", zone->lowmem_reserve[i]); | |
3229 | printk("\n"); | |
3230 | } | |
3231 | ||
ee99c71c | 3232 | for_each_populated_zone(zone) { |
b8af2941 | 3233 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
377e4f16 | 3234 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3235 | |
7bf02ea2 | 3236 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3237 | continue; |
1da177e4 LT |
3238 | show_node(zone); |
3239 | printk("%s: ", zone->name); | |
1da177e4 LT |
3240 | |
3241 | spin_lock_irqsave(&zone->lock, flags); | |
3242 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3243 | struct free_area *area = &zone->free_area[order]; |
3244 | int type; | |
3245 | ||
3246 | nr[order] = area->nr_free; | |
8f9de51a | 3247 | total += nr[order] << order; |
377e4f16 RV |
3248 | |
3249 | types[order] = 0; | |
3250 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3251 | if (!list_empty(&area->free_list[type])) | |
3252 | types[order] |= 1 << type; | |
3253 | } | |
1da177e4 LT |
3254 | } |
3255 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3256 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3257 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3258 | if (nr[order]) |
3259 | show_migration_types(types[order]); | |
3260 | } | |
1da177e4 LT |
3261 | printk("= %lukB\n", K(total)); |
3262 | } | |
3263 | ||
949f7ec5 DR |
3264 | hugetlb_show_meminfo(); |
3265 | ||
e6f3602d LW |
3266 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
3267 | ||
1da177e4 LT |
3268 | show_swap_cache_info(); |
3269 | } | |
3270 | ||
19770b32 MG |
3271 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
3272 | { | |
3273 | zoneref->zone = zone; | |
3274 | zoneref->zone_idx = zone_idx(zone); | |
3275 | } | |
3276 | ||
1da177e4 LT |
3277 | /* |
3278 | * Builds allocation fallback zone lists. | |
1a93205b CL |
3279 | * |
3280 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 3281 | */ |
f0c0b2b8 | 3282 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 3283 | int nr_zones) |
1da177e4 | 3284 | { |
1a93205b | 3285 | struct zone *zone; |
bc732f1d | 3286 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
3287 | |
3288 | do { | |
2f6726e5 | 3289 | zone_type--; |
070f8032 | 3290 | zone = pgdat->node_zones + zone_type; |
1a93205b | 3291 | if (populated_zone(zone)) { |
dd1a239f MG |
3292 | zoneref_set_zone(zone, |
3293 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 3294 | check_highest_zone(zone_type); |
1da177e4 | 3295 | } |
2f6726e5 | 3296 | } while (zone_type); |
bc732f1d | 3297 | |
070f8032 | 3298 | return nr_zones; |
1da177e4 LT |
3299 | } |
3300 | ||
f0c0b2b8 KH |
3301 | |
3302 | /* | |
3303 | * zonelist_order: | |
3304 | * 0 = automatic detection of better ordering. | |
3305 | * 1 = order by ([node] distance, -zonetype) | |
3306 | * 2 = order by (-zonetype, [node] distance) | |
3307 | * | |
3308 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
3309 | * the same zonelist. So only NUMA can configure this param. | |
3310 | */ | |
3311 | #define ZONELIST_ORDER_DEFAULT 0 | |
3312 | #define ZONELIST_ORDER_NODE 1 | |
3313 | #define ZONELIST_ORDER_ZONE 2 | |
3314 | ||
3315 | /* zonelist order in the kernel. | |
3316 | * set_zonelist_order() will set this to NODE or ZONE. | |
3317 | */ | |
3318 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3319 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
3320 | ||
3321 | ||
1da177e4 | 3322 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
3323 | /* The value user specified ....changed by config */ |
3324 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3325 | /* string for sysctl */ | |
3326 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
3327 | char numa_zonelist_order[16] = "default"; | |
3328 | ||
3329 | /* | |
3330 | * interface for configure zonelist ordering. | |
3331 | * command line option "numa_zonelist_order" | |
3332 | * = "[dD]efault - default, automatic configuration. | |
3333 | * = "[nN]ode - order by node locality, then by zone within node | |
3334 | * = "[zZ]one - order by zone, then by locality within zone | |
3335 | */ | |
3336 | ||
3337 | static int __parse_numa_zonelist_order(char *s) | |
3338 | { | |
3339 | if (*s == 'd' || *s == 'D') { | |
3340 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3341 | } else if (*s == 'n' || *s == 'N') { | |
3342 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
3343 | } else if (*s == 'z' || *s == 'Z') { | |
3344 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
3345 | } else { | |
3346 | printk(KERN_WARNING | |
3347 | "Ignoring invalid numa_zonelist_order value: " | |
3348 | "%s\n", s); | |
3349 | return -EINVAL; | |
3350 | } | |
3351 | return 0; | |
3352 | } | |
3353 | ||
3354 | static __init int setup_numa_zonelist_order(char *s) | |
3355 | { | |
ecb256f8 VL |
3356 | int ret; |
3357 | ||
3358 | if (!s) | |
3359 | return 0; | |
3360 | ||
3361 | ret = __parse_numa_zonelist_order(s); | |
3362 | if (ret == 0) | |
3363 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
3364 | ||
3365 | return ret; | |
f0c0b2b8 KH |
3366 | } |
3367 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
3368 | ||
3369 | /* | |
3370 | * sysctl handler for numa_zonelist_order | |
3371 | */ | |
3372 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 3373 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
3374 | loff_t *ppos) |
3375 | { | |
3376 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
3377 | int ret; | |
443c6f14 | 3378 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 3379 | |
443c6f14 | 3380 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
3381 | if (write) { |
3382 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
3383 | ret = -EINVAL; | |
3384 | goto out; | |
3385 | } | |
3386 | strcpy(saved_string, (char *)table->data); | |
3387 | } | |
8d65af78 | 3388 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 3389 | if (ret) |
443c6f14 | 3390 | goto out; |
f0c0b2b8 KH |
3391 | if (write) { |
3392 | int oldval = user_zonelist_order; | |
dacbde09 CG |
3393 | |
3394 | ret = __parse_numa_zonelist_order((char *)table->data); | |
3395 | if (ret) { | |
f0c0b2b8 KH |
3396 | /* |
3397 | * bogus value. restore saved string | |
3398 | */ | |
dacbde09 | 3399 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
3400 | NUMA_ZONELIST_ORDER_LEN); |
3401 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
3402 | } else if (oldval != user_zonelist_order) { |
3403 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 3404 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
3405 | mutex_unlock(&zonelists_mutex); |
3406 | } | |
f0c0b2b8 | 3407 | } |
443c6f14 AK |
3408 | out: |
3409 | mutex_unlock(&zl_order_mutex); | |
3410 | return ret; | |
f0c0b2b8 KH |
3411 | } |
3412 | ||
3413 | ||
62bc62a8 | 3414 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
3415 | static int node_load[MAX_NUMNODES]; |
3416 | ||
1da177e4 | 3417 | /** |
4dc3b16b | 3418 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
3419 | * @node: node whose fallback list we're appending |
3420 | * @used_node_mask: nodemask_t of already used nodes | |
3421 | * | |
3422 | * We use a number of factors to determine which is the next node that should | |
3423 | * appear on a given node's fallback list. The node should not have appeared | |
3424 | * already in @node's fallback list, and it should be the next closest node | |
3425 | * according to the distance array (which contains arbitrary distance values | |
3426 | * from each node to each node in the system), and should also prefer nodes | |
3427 | * with no CPUs, since presumably they'll have very little allocation pressure | |
3428 | * on them otherwise. | |
3429 | * It returns -1 if no node is found. | |
3430 | */ | |
f0c0b2b8 | 3431 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 3432 | { |
4cf808eb | 3433 | int n, val; |
1da177e4 | 3434 | int min_val = INT_MAX; |
00ef2d2f | 3435 | int best_node = NUMA_NO_NODE; |
a70f7302 | 3436 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 3437 | |
4cf808eb LT |
3438 | /* Use the local node if we haven't already */ |
3439 | if (!node_isset(node, *used_node_mask)) { | |
3440 | node_set(node, *used_node_mask); | |
3441 | return node; | |
3442 | } | |
1da177e4 | 3443 | |
4b0ef1fe | 3444 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
3445 | |
3446 | /* Don't want a node to appear more than once */ | |
3447 | if (node_isset(n, *used_node_mask)) | |
3448 | continue; | |
3449 | ||
1da177e4 LT |
3450 | /* Use the distance array to find the distance */ |
3451 | val = node_distance(node, n); | |
3452 | ||
4cf808eb LT |
3453 | /* Penalize nodes under us ("prefer the next node") */ |
3454 | val += (n < node); | |
3455 | ||
1da177e4 | 3456 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
3457 | tmp = cpumask_of_node(n); |
3458 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
3459 | val += PENALTY_FOR_NODE_WITH_CPUS; |
3460 | ||
3461 | /* Slight preference for less loaded node */ | |
3462 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
3463 | val += node_load[n]; | |
3464 | ||
3465 | if (val < min_val) { | |
3466 | min_val = val; | |
3467 | best_node = n; | |
3468 | } | |
3469 | } | |
3470 | ||
3471 | if (best_node >= 0) | |
3472 | node_set(best_node, *used_node_mask); | |
3473 | ||
3474 | return best_node; | |
3475 | } | |
3476 | ||
f0c0b2b8 KH |
3477 | |
3478 | /* | |
3479 | * Build zonelists ordered by node and zones within node. | |
3480 | * This results in maximum locality--normal zone overflows into local | |
3481 | * DMA zone, if any--but risks exhausting DMA zone. | |
3482 | */ | |
3483 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 3484 | { |
f0c0b2b8 | 3485 | int j; |
1da177e4 | 3486 | struct zonelist *zonelist; |
f0c0b2b8 | 3487 | |
54a6eb5c | 3488 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 3489 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 3490 | ; |
bc732f1d | 3491 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
3492 | zonelist->_zonerefs[j].zone = NULL; |
3493 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
3494 | } |
3495 | ||
523b9458 CL |
3496 | /* |
3497 | * Build gfp_thisnode zonelists | |
3498 | */ | |
3499 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3500 | { | |
523b9458 CL |
3501 | int j; |
3502 | struct zonelist *zonelist; | |
3503 | ||
54a6eb5c | 3504 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 3505 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
3506 | zonelist->_zonerefs[j].zone = NULL; |
3507 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3508 | } |
3509 | ||
f0c0b2b8 KH |
3510 | /* |
3511 | * Build zonelists ordered by zone and nodes within zones. | |
3512 | * This results in conserving DMA zone[s] until all Normal memory is | |
3513 | * exhausted, but results in overflowing to remote node while memory | |
3514 | * may still exist in local DMA zone. | |
3515 | */ | |
3516 | static int node_order[MAX_NUMNODES]; | |
3517 | ||
3518 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3519 | { | |
f0c0b2b8 KH |
3520 | int pos, j, node; |
3521 | int zone_type; /* needs to be signed */ | |
3522 | struct zone *z; | |
3523 | struct zonelist *zonelist; | |
3524 | ||
54a6eb5c MG |
3525 | zonelist = &pgdat->node_zonelists[0]; |
3526 | pos = 0; | |
3527 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3528 | for (j = 0; j < nr_nodes; j++) { | |
3529 | node = node_order[j]; | |
3530 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3531 | if (populated_zone(z)) { | |
dd1a239f MG |
3532 | zoneref_set_zone(z, |
3533 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3534 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3535 | } |
3536 | } | |
f0c0b2b8 | 3537 | } |
dd1a239f MG |
3538 | zonelist->_zonerefs[pos].zone = NULL; |
3539 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3540 | } |
3541 | ||
3542 | static int default_zonelist_order(void) | |
3543 | { | |
3544 | int nid, zone_type; | |
b8af2941 | 3545 | unsigned long low_kmem_size, total_size; |
f0c0b2b8 KH |
3546 | struct zone *z; |
3547 | int average_size; | |
3548 | /* | |
b8af2941 | 3549 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
f0c0b2b8 KH |
3550 | * If they are really small and used heavily, the system can fall |
3551 | * into OOM very easily. | |
e325c90f | 3552 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
f0c0b2b8 KH |
3553 | */ |
3554 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
3555 | low_kmem_size = 0; | |
3556 | total_size = 0; | |
3557 | for_each_online_node(nid) { | |
3558 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3559 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3560 | if (populated_zone(z)) { | |
3561 | if (zone_type < ZONE_NORMAL) | |
4f9f4774 JL |
3562 | low_kmem_size += z->managed_pages; |
3563 | total_size += z->managed_pages; | |
e325c90f DR |
3564 | } else if (zone_type == ZONE_NORMAL) { |
3565 | /* | |
3566 | * If any node has only lowmem, then node order | |
3567 | * is preferred to allow kernel allocations | |
3568 | * locally; otherwise, they can easily infringe | |
3569 | * on other nodes when there is an abundance of | |
3570 | * lowmem available to allocate from. | |
3571 | */ | |
3572 | return ZONELIST_ORDER_NODE; | |
f0c0b2b8 KH |
3573 | } |
3574 | } | |
3575 | } | |
3576 | if (!low_kmem_size || /* there are no DMA area. */ | |
3577 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
3578 | return ZONELIST_ORDER_NODE; | |
3579 | /* | |
3580 | * look into each node's config. | |
b8af2941 PK |
3581 | * If there is a node whose DMA/DMA32 memory is very big area on |
3582 | * local memory, NODE_ORDER may be suitable. | |
3583 | */ | |
37b07e41 | 3584 | average_size = total_size / |
4b0ef1fe | 3585 | (nodes_weight(node_states[N_MEMORY]) + 1); |
f0c0b2b8 KH |
3586 | for_each_online_node(nid) { |
3587 | low_kmem_size = 0; | |
3588 | total_size = 0; | |
3589 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3590 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3591 | if (populated_zone(z)) { | |
3592 | if (zone_type < ZONE_NORMAL) | |
3593 | low_kmem_size += z->present_pages; | |
3594 | total_size += z->present_pages; | |
3595 | } | |
3596 | } | |
3597 | if (low_kmem_size && | |
3598 | total_size > average_size && /* ignore small node */ | |
3599 | low_kmem_size > total_size * 70/100) | |
3600 | return ZONELIST_ORDER_NODE; | |
3601 | } | |
3602 | return ZONELIST_ORDER_ZONE; | |
3603 | } | |
3604 | ||
3605 | static void set_zonelist_order(void) | |
3606 | { | |
3607 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3608 | current_zonelist_order = default_zonelist_order(); | |
3609 | else | |
3610 | current_zonelist_order = user_zonelist_order; | |
3611 | } | |
3612 | ||
3613 | static void build_zonelists(pg_data_t *pgdat) | |
3614 | { | |
3615 | int j, node, load; | |
3616 | enum zone_type i; | |
1da177e4 | 3617 | nodemask_t used_mask; |
f0c0b2b8 KH |
3618 | int local_node, prev_node; |
3619 | struct zonelist *zonelist; | |
3620 | int order = current_zonelist_order; | |
1da177e4 LT |
3621 | |
3622 | /* initialize zonelists */ | |
523b9458 | 3623 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3624 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3625 | zonelist->_zonerefs[0].zone = NULL; |
3626 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3627 | } |
3628 | ||
3629 | /* NUMA-aware ordering of nodes */ | |
3630 | local_node = pgdat->node_id; | |
62bc62a8 | 3631 | load = nr_online_nodes; |
1da177e4 LT |
3632 | prev_node = local_node; |
3633 | nodes_clear(used_mask); | |
f0c0b2b8 | 3634 | |
f0c0b2b8 KH |
3635 | memset(node_order, 0, sizeof(node_order)); |
3636 | j = 0; | |
3637 | ||
1da177e4 LT |
3638 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
3639 | /* | |
3640 | * We don't want to pressure a particular node. | |
3641 | * So adding penalty to the first node in same | |
3642 | * distance group to make it round-robin. | |
3643 | */ | |
957f822a DR |
3644 | if (node_distance(local_node, node) != |
3645 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
3646 | node_load[node] = load; |
3647 | ||
1da177e4 LT |
3648 | prev_node = node; |
3649 | load--; | |
f0c0b2b8 KH |
3650 | if (order == ZONELIST_ORDER_NODE) |
3651 | build_zonelists_in_node_order(pgdat, node); | |
3652 | else | |
3653 | node_order[j++] = node; /* remember order */ | |
3654 | } | |
1da177e4 | 3655 | |
f0c0b2b8 KH |
3656 | if (order == ZONELIST_ORDER_ZONE) { |
3657 | /* calculate node order -- i.e., DMA last! */ | |
3658 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3659 | } |
523b9458 CL |
3660 | |
3661 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3662 | } |
3663 | ||
9276b1bc | 3664 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3665 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3666 | { |
54a6eb5c MG |
3667 | struct zonelist *zonelist; |
3668 | struct zonelist_cache *zlc; | |
dd1a239f | 3669 | struct zoneref *z; |
9276b1bc | 3670 | |
54a6eb5c MG |
3671 | zonelist = &pgdat->node_zonelists[0]; |
3672 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3673 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3674 | for (z = zonelist->_zonerefs; z->zone; z++) |
3675 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
3676 | } |
3677 | ||
7aac7898 LS |
3678 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3679 | /* | |
3680 | * Return node id of node used for "local" allocations. | |
3681 | * I.e., first node id of first zone in arg node's generic zonelist. | |
3682 | * Used for initializing percpu 'numa_mem', which is used primarily | |
3683 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
3684 | */ | |
3685 | int local_memory_node(int node) | |
3686 | { | |
3687 | struct zone *zone; | |
3688 | ||
3689 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
3690 | gfp_zone(GFP_KERNEL), | |
3691 | NULL, | |
3692 | &zone); | |
3693 | return zone->node; | |
3694 | } | |
3695 | #endif | |
f0c0b2b8 | 3696 | |
1da177e4 LT |
3697 | #else /* CONFIG_NUMA */ |
3698 | ||
f0c0b2b8 KH |
3699 | static void set_zonelist_order(void) |
3700 | { | |
3701 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
3702 | } | |
3703 | ||
3704 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 3705 | { |
19655d34 | 3706 | int node, local_node; |
54a6eb5c MG |
3707 | enum zone_type j; |
3708 | struct zonelist *zonelist; | |
1da177e4 LT |
3709 | |
3710 | local_node = pgdat->node_id; | |
1da177e4 | 3711 | |
54a6eb5c | 3712 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 3713 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 3714 | |
54a6eb5c MG |
3715 | /* |
3716 | * Now we build the zonelist so that it contains the zones | |
3717 | * of all the other nodes. | |
3718 | * We don't want to pressure a particular node, so when | |
3719 | * building the zones for node N, we make sure that the | |
3720 | * zones coming right after the local ones are those from | |
3721 | * node N+1 (modulo N) | |
3722 | */ | |
3723 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3724 | if (!node_online(node)) | |
3725 | continue; | |
bc732f1d | 3726 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 3727 | } |
54a6eb5c MG |
3728 | for (node = 0; node < local_node; node++) { |
3729 | if (!node_online(node)) | |
3730 | continue; | |
bc732f1d | 3731 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
3732 | } |
3733 | ||
dd1a239f MG |
3734 | zonelist->_zonerefs[j].zone = NULL; |
3735 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3736 | } |
3737 | ||
9276b1bc | 3738 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3739 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3740 | { |
54a6eb5c | 3741 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3742 | } |
3743 | ||
1da177e4 LT |
3744 | #endif /* CONFIG_NUMA */ |
3745 | ||
99dcc3e5 CL |
3746 | /* |
3747 | * Boot pageset table. One per cpu which is going to be used for all | |
3748 | * zones and all nodes. The parameters will be set in such a way | |
3749 | * that an item put on a list will immediately be handed over to | |
3750 | * the buddy list. This is safe since pageset manipulation is done | |
3751 | * with interrupts disabled. | |
3752 | * | |
3753 | * The boot_pagesets must be kept even after bootup is complete for | |
3754 | * unused processors and/or zones. They do play a role for bootstrapping | |
3755 | * hotplugged processors. | |
3756 | * | |
3757 | * zoneinfo_show() and maybe other functions do | |
3758 | * not check if the processor is online before following the pageset pointer. | |
3759 | * Other parts of the kernel may not check if the zone is available. | |
3760 | */ | |
3761 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3762 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3763 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3764 | |
4eaf3f64 HL |
3765 | /* |
3766 | * Global mutex to protect against size modification of zonelists | |
3767 | * as well as to serialize pageset setup for the new populated zone. | |
3768 | */ | |
3769 | DEFINE_MUTEX(zonelists_mutex); | |
3770 | ||
9b1a4d38 | 3771 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 3772 | static int __build_all_zonelists(void *data) |
1da177e4 | 3773 | { |
6811378e | 3774 | int nid; |
99dcc3e5 | 3775 | int cpu; |
9adb62a5 | 3776 | pg_data_t *self = data; |
9276b1bc | 3777 | |
7f9cfb31 BL |
3778 | #ifdef CONFIG_NUMA |
3779 | memset(node_load, 0, sizeof(node_load)); | |
3780 | #endif | |
9adb62a5 JL |
3781 | |
3782 | if (self && !node_online(self->node_id)) { | |
3783 | build_zonelists(self); | |
3784 | build_zonelist_cache(self); | |
3785 | } | |
3786 | ||
9276b1bc | 3787 | for_each_online_node(nid) { |
7ea1530a CL |
3788 | pg_data_t *pgdat = NODE_DATA(nid); |
3789 | ||
3790 | build_zonelists(pgdat); | |
3791 | build_zonelist_cache(pgdat); | |
9276b1bc | 3792 | } |
99dcc3e5 CL |
3793 | |
3794 | /* | |
3795 | * Initialize the boot_pagesets that are going to be used | |
3796 | * for bootstrapping processors. The real pagesets for | |
3797 | * each zone will be allocated later when the per cpu | |
3798 | * allocator is available. | |
3799 | * | |
3800 | * boot_pagesets are used also for bootstrapping offline | |
3801 | * cpus if the system is already booted because the pagesets | |
3802 | * are needed to initialize allocators on a specific cpu too. | |
3803 | * F.e. the percpu allocator needs the page allocator which | |
3804 | * needs the percpu allocator in order to allocate its pagesets | |
3805 | * (a chicken-egg dilemma). | |
3806 | */ | |
7aac7898 | 3807 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3808 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3809 | ||
7aac7898 LS |
3810 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3811 | /* | |
3812 | * We now know the "local memory node" for each node-- | |
3813 | * i.e., the node of the first zone in the generic zonelist. | |
3814 | * Set up numa_mem percpu variable for on-line cpus. During | |
3815 | * boot, only the boot cpu should be on-line; we'll init the | |
3816 | * secondary cpus' numa_mem as they come on-line. During | |
3817 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3818 | */ | |
3819 | if (cpu_online(cpu)) | |
3820 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3821 | #endif | |
3822 | } | |
3823 | ||
6811378e YG |
3824 | return 0; |
3825 | } | |
3826 | ||
4eaf3f64 HL |
3827 | /* |
3828 | * Called with zonelists_mutex held always | |
3829 | * unless system_state == SYSTEM_BOOTING. | |
3830 | */ | |
9adb62a5 | 3831 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 3832 | { |
f0c0b2b8 KH |
3833 | set_zonelist_order(); |
3834 | ||
6811378e | 3835 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3836 | __build_all_zonelists(NULL); |
68ad8df4 | 3837 | mminit_verify_zonelist(); |
6811378e YG |
3838 | cpuset_init_current_mems_allowed(); |
3839 | } else { | |
e9959f0f | 3840 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
3841 | if (zone) |
3842 | setup_zone_pageset(zone); | |
e9959f0f | 3843 | #endif |
dd1895e2 CS |
3844 | /* we have to stop all cpus to guarantee there is no user |
3845 | of zonelist */ | |
9adb62a5 | 3846 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
3847 | /* cpuset refresh routine should be here */ |
3848 | } | |
bd1e22b8 | 3849 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3850 | /* |
3851 | * Disable grouping by mobility if the number of pages in the | |
3852 | * system is too low to allow the mechanism to work. It would be | |
3853 | * more accurate, but expensive to check per-zone. This check is | |
3854 | * made on memory-hotadd so a system can start with mobility | |
3855 | * disabled and enable it later | |
3856 | */ | |
d9c23400 | 3857 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3858 | page_group_by_mobility_disabled = 1; |
3859 | else | |
3860 | page_group_by_mobility_disabled = 0; | |
3861 | ||
3862 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
3863 | "Total pages: %ld\n", | |
62bc62a8 | 3864 | nr_online_nodes, |
f0c0b2b8 | 3865 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3866 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3867 | vm_total_pages); |
3868 | #ifdef CONFIG_NUMA | |
3869 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
3870 | #endif | |
1da177e4 LT |
3871 | } |
3872 | ||
3873 | /* | |
3874 | * Helper functions to size the waitqueue hash table. | |
3875 | * Essentially these want to choose hash table sizes sufficiently | |
3876 | * large so that collisions trying to wait on pages are rare. | |
3877 | * But in fact, the number of active page waitqueues on typical | |
3878 | * systems is ridiculously low, less than 200. So this is even | |
3879 | * conservative, even though it seems large. | |
3880 | * | |
3881 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
3882 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
3883 | */ | |
3884 | #define PAGES_PER_WAITQUEUE 256 | |
3885 | ||
cca448fe | 3886 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 3887 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
3888 | { |
3889 | unsigned long size = 1; | |
3890 | ||
3891 | pages /= PAGES_PER_WAITQUEUE; | |
3892 | ||
3893 | while (size < pages) | |
3894 | size <<= 1; | |
3895 | ||
3896 | /* | |
3897 | * Once we have dozens or even hundreds of threads sleeping | |
3898 | * on IO we've got bigger problems than wait queue collision. | |
3899 | * Limit the size of the wait table to a reasonable size. | |
3900 | */ | |
3901 | size = min(size, 4096UL); | |
3902 | ||
3903 | return max(size, 4UL); | |
3904 | } | |
cca448fe YG |
3905 | #else |
3906 | /* | |
3907 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
3908 | * a suitable size for its wait_table. So we use the maximum size now. | |
3909 | * | |
3910 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
3911 | * | |
3912 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
3913 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
3914 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
3915 | * | |
3916 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
3917 | * or more by the traditional way. (See above). It equals: | |
3918 | * | |
3919 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
3920 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
3921 | * powerpc (64K page size) : = (32G +16M)byte. | |
3922 | */ | |
3923 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
3924 | { | |
3925 | return 4096UL; | |
3926 | } | |
3927 | #endif | |
1da177e4 LT |
3928 | |
3929 | /* | |
3930 | * This is an integer logarithm so that shifts can be used later | |
3931 | * to extract the more random high bits from the multiplicative | |
3932 | * hash function before the remainder is taken. | |
3933 | */ | |
3934 | static inline unsigned long wait_table_bits(unsigned long size) | |
3935 | { | |
3936 | return ffz(~size); | |
3937 | } | |
3938 | ||
6d3163ce AH |
3939 | /* |
3940 | * Check if a pageblock contains reserved pages | |
3941 | */ | |
3942 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
3943 | { | |
3944 | unsigned long pfn; | |
3945 | ||
3946 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
3947 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
3948 | return 1; | |
3949 | } | |
3950 | return 0; | |
3951 | } | |
3952 | ||
56fd56b8 | 3953 | /* |
d9c23400 | 3954 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
3955 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
3956 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
3957 | * higher will lead to a bigger reserve which will get freed as contiguous |
3958 | * blocks as reclaim kicks in | |
3959 | */ | |
3960 | static void setup_zone_migrate_reserve(struct zone *zone) | |
3961 | { | |
6d3163ce | 3962 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 3963 | struct page *page; |
78986a67 MG |
3964 | unsigned long block_migratetype; |
3965 | int reserve; | |
943dca1a | 3966 | int old_reserve; |
56fd56b8 | 3967 | |
d0215638 MH |
3968 | /* |
3969 | * Get the start pfn, end pfn and the number of blocks to reserve | |
3970 | * We have to be careful to be aligned to pageblock_nr_pages to | |
3971 | * make sure that we always check pfn_valid for the first page in | |
3972 | * the block. | |
3973 | */ | |
56fd56b8 | 3974 | start_pfn = zone->zone_start_pfn; |
108bcc96 | 3975 | end_pfn = zone_end_pfn(zone); |
d0215638 | 3976 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 3977 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 3978 | pageblock_order; |
56fd56b8 | 3979 | |
78986a67 MG |
3980 | /* |
3981 | * Reserve blocks are generally in place to help high-order atomic | |
3982 | * allocations that are short-lived. A min_free_kbytes value that | |
3983 | * would result in more than 2 reserve blocks for atomic allocations | |
3984 | * is assumed to be in place to help anti-fragmentation for the | |
3985 | * future allocation of hugepages at runtime. | |
3986 | */ | |
3987 | reserve = min(2, reserve); | |
943dca1a YI |
3988 | old_reserve = zone->nr_migrate_reserve_block; |
3989 | ||
3990 | /* When memory hot-add, we almost always need to do nothing */ | |
3991 | if (reserve == old_reserve) | |
3992 | return; | |
3993 | zone->nr_migrate_reserve_block = reserve; | |
78986a67 | 3994 | |
d9c23400 | 3995 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
3996 | if (!pfn_valid(pfn)) |
3997 | continue; | |
3998 | page = pfn_to_page(pfn); | |
3999 | ||
344c790e AL |
4000 | /* Watch out for overlapping nodes */ |
4001 | if (page_to_nid(page) != zone_to_nid(zone)) | |
4002 | continue; | |
4003 | ||
56fd56b8 MG |
4004 | block_migratetype = get_pageblock_migratetype(page); |
4005 | ||
938929f1 MG |
4006 | /* Only test what is necessary when the reserves are not met */ |
4007 | if (reserve > 0) { | |
4008 | /* | |
4009 | * Blocks with reserved pages will never free, skip | |
4010 | * them. | |
4011 | */ | |
4012 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
4013 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
4014 | continue; | |
56fd56b8 | 4015 | |
938929f1 MG |
4016 | /* If this block is reserved, account for it */ |
4017 | if (block_migratetype == MIGRATE_RESERVE) { | |
4018 | reserve--; | |
4019 | continue; | |
4020 | } | |
4021 | ||
4022 | /* Suitable for reserving if this block is movable */ | |
4023 | if (block_migratetype == MIGRATE_MOVABLE) { | |
4024 | set_pageblock_migratetype(page, | |
4025 | MIGRATE_RESERVE); | |
4026 | move_freepages_block(zone, page, | |
4027 | MIGRATE_RESERVE); | |
4028 | reserve--; | |
4029 | continue; | |
4030 | } | |
943dca1a YI |
4031 | } else if (!old_reserve) { |
4032 | /* | |
4033 | * At boot time we don't need to scan the whole zone | |
4034 | * for turning off MIGRATE_RESERVE. | |
4035 | */ | |
4036 | break; | |
56fd56b8 MG |
4037 | } |
4038 | ||
4039 | /* | |
4040 | * If the reserve is met and this is a previous reserved block, | |
4041 | * take it back | |
4042 | */ | |
4043 | if (block_migratetype == MIGRATE_RESERVE) { | |
4044 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4045 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
4046 | } | |
4047 | } | |
4048 | } | |
ac0e5b7a | 4049 | |
1da177e4 LT |
4050 | /* |
4051 | * Initially all pages are reserved - free ones are freed | |
4052 | * up by free_all_bootmem() once the early boot process is | |
4053 | * done. Non-atomic initialization, single-pass. | |
4054 | */ | |
c09b4240 | 4055 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 4056 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 4057 | { |
1da177e4 | 4058 | struct page *page; |
29751f69 AW |
4059 | unsigned long end_pfn = start_pfn + size; |
4060 | unsigned long pfn; | |
86051ca5 | 4061 | struct zone *z; |
1da177e4 | 4062 | |
22b31eec HD |
4063 | if (highest_memmap_pfn < end_pfn - 1) |
4064 | highest_memmap_pfn = end_pfn - 1; | |
4065 | ||
86051ca5 | 4066 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 4067 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
4068 | /* |
4069 | * There can be holes in boot-time mem_map[]s | |
4070 | * handed to this function. They do not | |
4071 | * exist on hotplugged memory. | |
4072 | */ | |
4073 | if (context == MEMMAP_EARLY) { | |
4074 | if (!early_pfn_valid(pfn)) | |
4075 | continue; | |
4076 | if (!early_pfn_in_nid(pfn, nid)) | |
4077 | continue; | |
4078 | } | |
d41dee36 AW |
4079 | page = pfn_to_page(pfn); |
4080 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 4081 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 4082 | init_page_count(page); |
22b751c3 | 4083 | page_mapcount_reset(page); |
90572890 | 4084 | page_cpupid_reset_last(page); |
1da177e4 | 4085 | SetPageReserved(page); |
b2a0ac88 MG |
4086 | /* |
4087 | * Mark the block movable so that blocks are reserved for | |
4088 | * movable at startup. This will force kernel allocations | |
4089 | * to reserve their blocks rather than leaking throughout | |
4090 | * the address space during boot when many long-lived | |
56fd56b8 MG |
4091 | * kernel allocations are made. Later some blocks near |
4092 | * the start are marked MIGRATE_RESERVE by | |
4093 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
4094 | * |
4095 | * bitmap is created for zone's valid pfn range. but memmap | |
4096 | * can be created for invalid pages (for alignment) | |
4097 | * check here not to call set_pageblock_migratetype() against | |
4098 | * pfn out of zone. | |
b2a0ac88 | 4099 | */ |
86051ca5 | 4100 | if ((z->zone_start_pfn <= pfn) |
108bcc96 | 4101 | && (pfn < zone_end_pfn(z)) |
86051ca5 | 4102 | && !(pfn & (pageblock_nr_pages - 1))) |
56fd56b8 | 4103 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 4104 | |
1da177e4 LT |
4105 | INIT_LIST_HEAD(&page->lru); |
4106 | #ifdef WANT_PAGE_VIRTUAL | |
4107 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
4108 | if (!is_highmem_idx(zone)) | |
3212c6be | 4109 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 4110 | #endif |
1da177e4 LT |
4111 | } |
4112 | } | |
4113 | ||
1e548deb | 4114 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 4115 | { |
b2a0ac88 MG |
4116 | int order, t; |
4117 | for_each_migratetype_order(order, t) { | |
4118 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
4119 | zone->free_area[order].nr_free = 0; |
4120 | } | |
4121 | } | |
4122 | ||
4123 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
4124 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 4125 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
4126 | #endif |
4127 | ||
4ed7e022 | 4128 | static int __meminit zone_batchsize(struct zone *zone) |
e7c8d5c9 | 4129 | { |
3a6be87f | 4130 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
4131 | int batch; |
4132 | ||
4133 | /* | |
4134 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 4135 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
4136 | * |
4137 | * OK, so we don't know how big the cache is. So guess. | |
4138 | */ | |
b40da049 | 4139 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
4140 | if (batch * PAGE_SIZE > 512 * 1024) |
4141 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
4142 | batch /= 4; /* We effectively *= 4 below */ |
4143 | if (batch < 1) | |
4144 | batch = 1; | |
4145 | ||
4146 | /* | |
0ceaacc9 NP |
4147 | * Clamp the batch to a 2^n - 1 value. Having a power |
4148 | * of 2 value was found to be more likely to have | |
4149 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 4150 | * |
0ceaacc9 NP |
4151 | * For example if 2 tasks are alternately allocating |
4152 | * batches of pages, one task can end up with a lot | |
4153 | * of pages of one half of the possible page colors | |
4154 | * and the other with pages of the other colors. | |
e7c8d5c9 | 4155 | */ |
9155203a | 4156 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 4157 | |
e7c8d5c9 | 4158 | return batch; |
3a6be87f DH |
4159 | |
4160 | #else | |
4161 | /* The deferral and batching of frees should be suppressed under NOMMU | |
4162 | * conditions. | |
4163 | * | |
4164 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4165 | * of contiguous memory as there's no hardware page translation to | |
4166 | * assemble apparent contiguous memory from discontiguous pages. | |
4167 | * | |
4168 | * Queueing large contiguous runs of pages for batching, however, | |
4169 | * causes the pages to actually be freed in smaller chunks. As there | |
4170 | * can be a significant delay between the individual batches being | |
4171 | * recycled, this leads to the once large chunks of space being | |
4172 | * fragmented and becoming unavailable for high-order allocations. | |
4173 | */ | |
4174 | return 0; | |
4175 | #endif | |
e7c8d5c9 CL |
4176 | } |
4177 | ||
8d7a8fa9 CS |
4178 | /* |
4179 | * pcp->high and pcp->batch values are related and dependent on one another: | |
4180 | * ->batch must never be higher then ->high. | |
4181 | * The following function updates them in a safe manner without read side | |
4182 | * locking. | |
4183 | * | |
4184 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
4185 | * those fields changing asynchronously (acording the the above rule). | |
4186 | * | |
4187 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
4188 | * outside of boot time (or some other assurance that no concurrent updaters | |
4189 | * exist). | |
4190 | */ | |
4191 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
4192 | unsigned long batch) | |
4193 | { | |
4194 | /* start with a fail safe value for batch */ | |
4195 | pcp->batch = 1; | |
4196 | smp_wmb(); | |
4197 | ||
4198 | /* Update high, then batch, in order */ | |
4199 | pcp->high = high; | |
4200 | smp_wmb(); | |
4201 | ||
4202 | pcp->batch = batch; | |
4203 | } | |
4204 | ||
3664033c | 4205 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
4206 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
4207 | { | |
8d7a8fa9 | 4208 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
4209 | } |
4210 | ||
88c90dbc | 4211 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
4212 | { |
4213 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4214 | int migratetype; |
2caaad41 | 4215 | |
1c6fe946 MD |
4216 | memset(p, 0, sizeof(*p)); |
4217 | ||
3dfa5721 | 4218 | pcp = &p->pcp; |
2caaad41 | 4219 | pcp->count = 0; |
5f8dcc21 MG |
4220 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4221 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4222 | } |
4223 | ||
88c90dbc CS |
4224 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
4225 | { | |
4226 | pageset_init(p); | |
4227 | pageset_set_batch(p, batch); | |
4228 | } | |
4229 | ||
8ad4b1fb | 4230 | /* |
3664033c | 4231 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
4232 | * to the value high for the pageset p. |
4233 | */ | |
3664033c | 4234 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
4235 | unsigned long high) |
4236 | { | |
8d7a8fa9 CS |
4237 | unsigned long batch = max(1UL, high / 4); |
4238 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
4239 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 4240 | |
8d7a8fa9 | 4241 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
4242 | } |
4243 | ||
169f6c19 CS |
4244 | static void __meminit pageset_set_high_and_batch(struct zone *zone, |
4245 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 4246 | { |
56cef2b8 | 4247 | if (percpu_pagelist_fraction) |
3664033c | 4248 | pageset_set_high(pcp, |
56cef2b8 CS |
4249 | (zone->managed_pages / |
4250 | percpu_pagelist_fraction)); | |
4251 | else | |
4252 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
4253 | } | |
4254 | ||
169f6c19 CS |
4255 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
4256 | { | |
4257 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4258 | ||
4259 | pageset_init(pcp); | |
4260 | pageset_set_high_and_batch(zone, pcp); | |
4261 | } | |
4262 | ||
4ed7e022 | 4263 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4264 | { |
4265 | int cpu; | |
319774e2 | 4266 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
4267 | for_each_possible_cpu(cpu) |
4268 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
4269 | } |
4270 | ||
2caaad41 | 4271 | /* |
99dcc3e5 CL |
4272 | * Allocate per cpu pagesets and initialize them. |
4273 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4274 | */ |
99dcc3e5 | 4275 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4276 | { |
99dcc3e5 | 4277 | struct zone *zone; |
e7c8d5c9 | 4278 | |
319774e2 WF |
4279 | for_each_populated_zone(zone) |
4280 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4281 | } |
4282 | ||
577a32f6 | 4283 | static noinline __init_refok |
cca448fe | 4284 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4285 | { |
4286 | int i; | |
cca448fe | 4287 | size_t alloc_size; |
ed8ece2e DH |
4288 | |
4289 | /* | |
4290 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4291 | * per zone. | |
4292 | */ | |
02b694de YG |
4293 | zone->wait_table_hash_nr_entries = |
4294 | wait_table_hash_nr_entries(zone_size_pages); | |
4295 | zone->wait_table_bits = | |
4296 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4297 | alloc_size = zone->wait_table_hash_nr_entries |
4298 | * sizeof(wait_queue_head_t); | |
4299 | ||
cd94b9db | 4300 | if (!slab_is_available()) { |
cca448fe | 4301 | zone->wait_table = (wait_queue_head_t *) |
6782832e SS |
4302 | memblock_virt_alloc_node_nopanic( |
4303 | alloc_size, zone->zone_pgdat->node_id); | |
cca448fe YG |
4304 | } else { |
4305 | /* | |
4306 | * This case means that a zone whose size was 0 gets new memory | |
4307 | * via memory hot-add. | |
4308 | * But it may be the case that a new node was hot-added. In | |
4309 | * this case vmalloc() will not be able to use this new node's | |
4310 | * memory - this wait_table must be initialized to use this new | |
4311 | * node itself as well. | |
4312 | * To use this new node's memory, further consideration will be | |
4313 | * necessary. | |
4314 | */ | |
8691f3a7 | 4315 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4316 | } |
4317 | if (!zone->wait_table) | |
4318 | return -ENOMEM; | |
ed8ece2e | 4319 | |
b8af2941 | 4320 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4321 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4322 | |
4323 | return 0; | |
ed8ece2e DH |
4324 | } |
4325 | ||
c09b4240 | 4326 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4327 | { |
99dcc3e5 CL |
4328 | /* |
4329 | * per cpu subsystem is not up at this point. The following code | |
4330 | * relies on the ability of the linker to provide the | |
4331 | * offset of a (static) per cpu variable into the per cpu area. | |
4332 | */ | |
4333 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4334 | |
b38a8725 | 4335 | if (populated_zone(zone)) |
99dcc3e5 CL |
4336 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4337 | zone->name, zone->present_pages, | |
4338 | zone_batchsize(zone)); | |
ed8ece2e DH |
4339 | } |
4340 | ||
4ed7e022 | 4341 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4342 | unsigned long zone_start_pfn, |
a2f3aa02 DH |
4343 | unsigned long size, |
4344 | enum memmap_context context) | |
ed8ece2e DH |
4345 | { |
4346 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4347 | int ret; |
4348 | ret = zone_wait_table_init(zone, size); | |
4349 | if (ret) | |
4350 | return ret; | |
ed8ece2e DH |
4351 | pgdat->nr_zones = zone_idx(zone) + 1; |
4352 | ||
ed8ece2e DH |
4353 | zone->zone_start_pfn = zone_start_pfn; |
4354 | ||
708614e6 MG |
4355 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4356 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4357 | pgdat->node_id, | |
4358 | (unsigned long)zone_idx(zone), | |
4359 | zone_start_pfn, (zone_start_pfn + size)); | |
4360 | ||
1e548deb | 4361 | zone_init_free_lists(zone); |
718127cc YG |
4362 | |
4363 | return 0; | |
ed8ece2e DH |
4364 | } |
4365 | ||
0ee332c1 | 4366 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
4367 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
4368 | /* | |
4369 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
4370 | * Architectures may implement their own version but if add_active_range() | |
4371 | * was used and there are no special requirements, this is a convenient | |
4372 | * alternative | |
4373 | */ | |
f2dbcfa7 | 4374 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d | 4375 | { |
c13291a5 | 4376 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 4377 | int nid; |
7c243c71 RA |
4378 | /* |
4379 | * NOTE: The following SMP-unsafe globals are only used early in boot | |
4380 | * when the kernel is running single-threaded. | |
4381 | */ | |
4382 | static unsigned long __meminitdata last_start_pfn, last_end_pfn; | |
4383 | static int __meminitdata last_nid; | |
4384 | ||
4385 | if (last_start_pfn <= pfn && pfn < last_end_pfn) | |
4386 | return last_nid; | |
c713216d | 4387 | |
e76b63f8 YL |
4388 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
4389 | if (nid != -1) { | |
4390 | last_start_pfn = start_pfn; | |
4391 | last_end_pfn = end_pfn; | |
4392 | last_nid = nid; | |
4393 | } | |
4394 | ||
4395 | return nid; | |
c713216d MG |
4396 | } |
4397 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
4398 | ||
f2dbcfa7 KH |
4399 | int __meminit early_pfn_to_nid(unsigned long pfn) |
4400 | { | |
cc2559bc KH |
4401 | int nid; |
4402 | ||
4403 | nid = __early_pfn_to_nid(pfn); | |
4404 | if (nid >= 0) | |
4405 | return nid; | |
4406 | /* just returns 0 */ | |
4407 | return 0; | |
f2dbcfa7 KH |
4408 | } |
4409 | ||
cc2559bc KH |
4410 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
4411 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
4412 | { | |
4413 | int nid; | |
4414 | ||
4415 | nid = __early_pfn_to_nid(pfn); | |
4416 | if (nid >= 0 && nid != node) | |
4417 | return false; | |
4418 | return true; | |
4419 | } | |
4420 | #endif | |
f2dbcfa7 | 4421 | |
c713216d | 4422 | /** |
6782832e | 4423 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 4424 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 4425 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d MG |
4426 | * |
4427 | * If an architecture guarantees that all ranges registered with | |
4428 | * add_active_ranges() contain no holes and may be freed, this | |
6782832e SS |
4429 | * this function may be used instead of calling memblock_free_early_nid() |
4430 | * manually. | |
c713216d | 4431 | */ |
c13291a5 | 4432 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 4433 | { |
c13291a5 TH |
4434 | unsigned long start_pfn, end_pfn; |
4435 | int i, this_nid; | |
edbe7d23 | 4436 | |
c13291a5 TH |
4437 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
4438 | start_pfn = min(start_pfn, max_low_pfn); | |
4439 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 4440 | |
c13291a5 | 4441 | if (start_pfn < end_pfn) |
6782832e SS |
4442 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
4443 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
4444 | this_nid); | |
edbe7d23 | 4445 | } |
edbe7d23 | 4446 | } |
edbe7d23 | 4447 | |
c713216d MG |
4448 | /** |
4449 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 4450 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
4451 | * |
4452 | * If an architecture guarantees that all ranges registered with | |
4453 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 4454 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
4455 | */ |
4456 | void __init sparse_memory_present_with_active_regions(int nid) | |
4457 | { | |
c13291a5 TH |
4458 | unsigned long start_pfn, end_pfn; |
4459 | int i, this_nid; | |
c713216d | 4460 | |
c13291a5 TH |
4461 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
4462 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
4463 | } |
4464 | ||
4465 | /** | |
4466 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
4467 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
4468 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
4469 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
4470 | * |
4471 | * It returns the start and end page frame of a node based on information | |
4472 | * provided by an arch calling add_active_range(). If called for a node | |
4473 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 4474 | * PFNs will be 0. |
c713216d | 4475 | */ |
a3142c8e | 4476 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
4477 | unsigned long *start_pfn, unsigned long *end_pfn) |
4478 | { | |
c13291a5 | 4479 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 4480 | int i; |
c13291a5 | 4481 | |
c713216d MG |
4482 | *start_pfn = -1UL; |
4483 | *end_pfn = 0; | |
4484 | ||
c13291a5 TH |
4485 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
4486 | *start_pfn = min(*start_pfn, this_start_pfn); | |
4487 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
4488 | } |
4489 | ||
633c0666 | 4490 | if (*start_pfn == -1UL) |
c713216d | 4491 | *start_pfn = 0; |
c713216d MG |
4492 | } |
4493 | ||
2a1e274a MG |
4494 | /* |
4495 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
4496 | * assumption is made that zones within a node are ordered in monotonic | |
4497 | * increasing memory addresses so that the "highest" populated zone is used | |
4498 | */ | |
b69a7288 | 4499 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
4500 | { |
4501 | int zone_index; | |
4502 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
4503 | if (zone_index == ZONE_MOVABLE) | |
4504 | continue; | |
4505 | ||
4506 | if (arch_zone_highest_possible_pfn[zone_index] > | |
4507 | arch_zone_lowest_possible_pfn[zone_index]) | |
4508 | break; | |
4509 | } | |
4510 | ||
4511 | VM_BUG_ON(zone_index == -1); | |
4512 | movable_zone = zone_index; | |
4513 | } | |
4514 | ||
4515 | /* | |
4516 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 4517 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
4518 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
4519 | * in each node depending on the size of each node and how evenly kernelcore | |
4520 | * is distributed. This helper function adjusts the zone ranges | |
4521 | * provided by the architecture for a given node by using the end of the | |
4522 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
4523 | * zones within a node are in order of monotonic increases memory addresses | |
4524 | */ | |
b69a7288 | 4525 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4526 | unsigned long zone_type, |
4527 | unsigned long node_start_pfn, | |
4528 | unsigned long node_end_pfn, | |
4529 | unsigned long *zone_start_pfn, | |
4530 | unsigned long *zone_end_pfn) | |
4531 | { | |
4532 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4533 | if (zone_movable_pfn[nid]) { | |
4534 | /* Size ZONE_MOVABLE */ | |
4535 | if (zone_type == ZONE_MOVABLE) { | |
4536 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4537 | *zone_end_pfn = min(node_end_pfn, | |
4538 | arch_zone_highest_possible_pfn[movable_zone]); | |
4539 | ||
4540 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4541 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4542 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4543 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4544 | ||
4545 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4546 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4547 | *zone_start_pfn = *zone_end_pfn; | |
4548 | } | |
4549 | } | |
4550 | ||
c713216d MG |
4551 | /* |
4552 | * Return the number of pages a zone spans in a node, including holes | |
4553 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4554 | */ | |
6ea6e688 | 4555 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4556 | unsigned long zone_type, |
7960aedd ZY |
4557 | unsigned long node_start_pfn, |
4558 | unsigned long node_end_pfn, | |
c713216d MG |
4559 | unsigned long *ignored) |
4560 | { | |
c713216d MG |
4561 | unsigned long zone_start_pfn, zone_end_pfn; |
4562 | ||
7960aedd | 4563 | /* Get the start and end of the zone */ |
c713216d MG |
4564 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
4565 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4566 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4567 | node_start_pfn, node_end_pfn, | |
4568 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4569 | |
4570 | /* Check that this node has pages within the zone's required range */ | |
4571 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4572 | return 0; | |
4573 | ||
4574 | /* Move the zone boundaries inside the node if necessary */ | |
4575 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4576 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4577 | ||
4578 | /* Return the spanned pages */ | |
4579 | return zone_end_pfn - zone_start_pfn; | |
4580 | } | |
4581 | ||
4582 | /* | |
4583 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4584 | * then all holes in the requested range will be accounted for. |
c713216d | 4585 | */ |
32996250 | 4586 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4587 | unsigned long range_start_pfn, |
4588 | unsigned long range_end_pfn) | |
4589 | { | |
96e907d1 TH |
4590 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4591 | unsigned long start_pfn, end_pfn; | |
4592 | int i; | |
c713216d | 4593 | |
96e907d1 TH |
4594 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4595 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4596 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4597 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4598 | } |
96e907d1 | 4599 | return nr_absent; |
c713216d MG |
4600 | } |
4601 | ||
4602 | /** | |
4603 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4604 | * @start_pfn: The start PFN to start searching for holes | |
4605 | * @end_pfn: The end PFN to stop searching for holes | |
4606 | * | |
88ca3b94 | 4607 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4608 | */ |
4609 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4610 | unsigned long end_pfn) | |
4611 | { | |
4612 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4613 | } | |
4614 | ||
4615 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4616 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4617 | unsigned long zone_type, |
7960aedd ZY |
4618 | unsigned long node_start_pfn, |
4619 | unsigned long node_end_pfn, | |
c713216d MG |
4620 | unsigned long *ignored) |
4621 | { | |
96e907d1 TH |
4622 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4623 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4624 | unsigned long zone_start_pfn, zone_end_pfn; |
4625 | ||
96e907d1 TH |
4626 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4627 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4628 | |
2a1e274a MG |
4629 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4630 | node_start_pfn, node_end_pfn, | |
4631 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4632 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4633 | } |
0e0b864e | 4634 | |
0ee332c1 | 4635 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4636 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4637 | unsigned long zone_type, |
7960aedd ZY |
4638 | unsigned long node_start_pfn, |
4639 | unsigned long node_end_pfn, | |
c713216d MG |
4640 | unsigned long *zones_size) |
4641 | { | |
4642 | return zones_size[zone_type]; | |
4643 | } | |
4644 | ||
6ea6e688 | 4645 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4646 | unsigned long zone_type, |
7960aedd ZY |
4647 | unsigned long node_start_pfn, |
4648 | unsigned long node_end_pfn, | |
c713216d MG |
4649 | unsigned long *zholes_size) |
4650 | { | |
4651 | if (!zholes_size) | |
4652 | return 0; | |
4653 | ||
4654 | return zholes_size[zone_type]; | |
4655 | } | |
20e6926d | 4656 | |
0ee332c1 | 4657 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4658 | |
a3142c8e | 4659 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
4660 | unsigned long node_start_pfn, |
4661 | unsigned long node_end_pfn, | |
4662 | unsigned long *zones_size, | |
4663 | unsigned long *zholes_size) | |
c713216d MG |
4664 | { |
4665 | unsigned long realtotalpages, totalpages = 0; | |
4666 | enum zone_type i; | |
4667 | ||
4668 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4669 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4670 | node_start_pfn, |
4671 | node_end_pfn, | |
4672 | zones_size); | |
c713216d MG |
4673 | pgdat->node_spanned_pages = totalpages; |
4674 | ||
4675 | realtotalpages = totalpages; | |
4676 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4677 | realtotalpages -= | |
4678 | zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4679 | node_start_pfn, node_end_pfn, |
4680 | zholes_size); | |
c713216d MG |
4681 | pgdat->node_present_pages = realtotalpages; |
4682 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4683 | realtotalpages); | |
4684 | } | |
4685 | ||
835c134e MG |
4686 | #ifndef CONFIG_SPARSEMEM |
4687 | /* | |
4688 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4689 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4690 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4691 | * round what is now in bits to nearest long in bits, then return it in |
4692 | * bytes. | |
4693 | */ | |
7c45512d | 4694 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
4695 | { |
4696 | unsigned long usemapsize; | |
4697 | ||
7c45512d | 4698 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
4699 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4700 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4701 | usemapsize *= NR_PAGEBLOCK_BITS; |
4702 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4703 | ||
4704 | return usemapsize / 8; | |
4705 | } | |
4706 | ||
4707 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
4708 | struct zone *zone, |
4709 | unsigned long zone_start_pfn, | |
4710 | unsigned long zonesize) | |
835c134e | 4711 | { |
7c45512d | 4712 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 4713 | zone->pageblock_flags = NULL; |
58a01a45 | 4714 | if (usemapsize) |
6782832e SS |
4715 | zone->pageblock_flags = |
4716 | memblock_virt_alloc_node_nopanic(usemapsize, | |
4717 | pgdat->node_id); | |
835c134e MG |
4718 | } |
4719 | #else | |
7c45512d LT |
4720 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
4721 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
4722 | #endif /* CONFIG_SPARSEMEM */ |
4723 | ||
d9c23400 | 4724 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 4725 | |
d9c23400 | 4726 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 4727 | void __paginginit set_pageblock_order(void) |
d9c23400 | 4728 | { |
955c1cd7 AM |
4729 | unsigned int order; |
4730 | ||
d9c23400 MG |
4731 | /* Check that pageblock_nr_pages has not already been setup */ |
4732 | if (pageblock_order) | |
4733 | return; | |
4734 | ||
955c1cd7 AM |
4735 | if (HPAGE_SHIFT > PAGE_SHIFT) |
4736 | order = HUGETLB_PAGE_ORDER; | |
4737 | else | |
4738 | order = MAX_ORDER - 1; | |
4739 | ||
d9c23400 MG |
4740 | /* |
4741 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
4742 | * This value may be variable depending on boot parameters on IA64 and |
4743 | * powerpc. | |
d9c23400 MG |
4744 | */ |
4745 | pageblock_order = order; | |
4746 | } | |
4747 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4748 | ||
ba72cb8c MG |
4749 | /* |
4750 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
4751 | * is unused as pageblock_order is set at compile-time. See |
4752 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
4753 | * the kernel config | |
ba72cb8c | 4754 | */ |
15ca220e | 4755 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 4756 | { |
ba72cb8c | 4757 | } |
d9c23400 MG |
4758 | |
4759 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4760 | ||
01cefaef JL |
4761 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
4762 | unsigned long present_pages) | |
4763 | { | |
4764 | unsigned long pages = spanned_pages; | |
4765 | ||
4766 | /* | |
4767 | * Provide a more accurate estimation if there are holes within | |
4768 | * the zone and SPARSEMEM is in use. If there are holes within the | |
4769 | * zone, each populated memory region may cost us one or two extra | |
4770 | * memmap pages due to alignment because memmap pages for each | |
4771 | * populated regions may not naturally algined on page boundary. | |
4772 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
4773 | */ | |
4774 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
4775 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
4776 | pages = present_pages; | |
4777 | ||
4778 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
4779 | } | |
4780 | ||
1da177e4 LT |
4781 | /* |
4782 | * Set up the zone data structures: | |
4783 | * - mark all pages reserved | |
4784 | * - mark all memory queues empty | |
4785 | * - clear the memory bitmaps | |
6527af5d MK |
4786 | * |
4787 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 4788 | */ |
b5a0e011 | 4789 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
7960aedd | 4790 | unsigned long node_start_pfn, unsigned long node_end_pfn, |
1da177e4 LT |
4791 | unsigned long *zones_size, unsigned long *zholes_size) |
4792 | { | |
2f1b6248 | 4793 | enum zone_type j; |
ed8ece2e | 4794 | int nid = pgdat->node_id; |
1da177e4 | 4795 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4796 | int ret; |
1da177e4 | 4797 | |
208d54e5 | 4798 | pgdat_resize_init(pgdat); |
8177a420 AA |
4799 | #ifdef CONFIG_NUMA_BALANCING |
4800 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
4801 | pgdat->numabalancing_migrate_nr_pages = 0; | |
4802 | pgdat->numabalancing_migrate_next_window = jiffies; | |
4803 | #endif | |
1da177e4 | 4804 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 4805 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
52d4b9ac | 4806 | pgdat_page_cgroup_init(pgdat); |
5f63b720 | 4807 | |
1da177e4 LT |
4808 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4809 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 4810 | unsigned long size, realsize, freesize, memmap_pages; |
1da177e4 | 4811 | |
7960aedd ZY |
4812 | size = zone_spanned_pages_in_node(nid, j, node_start_pfn, |
4813 | node_end_pfn, zones_size); | |
9feedc9d | 4814 | realsize = freesize = size - zone_absent_pages_in_node(nid, j, |
7960aedd ZY |
4815 | node_start_pfn, |
4816 | node_end_pfn, | |
c713216d | 4817 | zholes_size); |
1da177e4 | 4818 | |
0e0b864e | 4819 | /* |
9feedc9d | 4820 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
4821 | * is used by this zone for memmap. This affects the watermark |
4822 | * and per-cpu initialisations | |
4823 | */ | |
01cefaef | 4824 | memmap_pages = calc_memmap_size(size, realsize); |
9feedc9d JL |
4825 | if (freesize >= memmap_pages) { |
4826 | freesize -= memmap_pages; | |
5594c8c8 YL |
4827 | if (memmap_pages) |
4828 | printk(KERN_DEBUG | |
4829 | " %s zone: %lu pages used for memmap\n", | |
4830 | zone_names[j], memmap_pages); | |
0e0b864e MG |
4831 | } else |
4832 | printk(KERN_WARNING | |
9feedc9d JL |
4833 | " %s zone: %lu pages exceeds freesize %lu\n", |
4834 | zone_names[j], memmap_pages, freesize); | |
0e0b864e | 4835 | |
6267276f | 4836 | /* Account for reserved pages */ |
9feedc9d JL |
4837 | if (j == 0 && freesize > dma_reserve) { |
4838 | freesize -= dma_reserve; | |
d903ef9f | 4839 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4840 | zone_names[0], dma_reserve); |
0e0b864e MG |
4841 | } |
4842 | ||
98d2b0eb | 4843 | if (!is_highmem_idx(j)) |
9feedc9d | 4844 | nr_kernel_pages += freesize; |
01cefaef JL |
4845 | /* Charge for highmem memmap if there are enough kernel pages */ |
4846 | else if (nr_kernel_pages > memmap_pages * 2) | |
4847 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 4848 | nr_all_pages += freesize; |
1da177e4 LT |
4849 | |
4850 | zone->spanned_pages = size; | |
306f2e9e | 4851 | zone->present_pages = realsize; |
9feedc9d JL |
4852 | /* |
4853 | * Set an approximate value for lowmem here, it will be adjusted | |
4854 | * when the bootmem allocator frees pages into the buddy system. | |
4855 | * And all highmem pages will be managed by the buddy system. | |
4856 | */ | |
4857 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 4858 | #ifdef CONFIG_NUMA |
d5f541ed | 4859 | zone->node = nid; |
9feedc9d | 4860 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 4861 | / 100; |
9feedc9d | 4862 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 4863 | #endif |
1da177e4 LT |
4864 | zone->name = zone_names[j]; |
4865 | spin_lock_init(&zone->lock); | |
4866 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4867 | zone_seqlock_init(zone); |
1da177e4 | 4868 | zone->zone_pgdat = pgdat; |
ed8ece2e | 4869 | zone_pcp_init(zone); |
81c0a2bb JW |
4870 | |
4871 | /* For bootup, initialized properly in watermark setup */ | |
4872 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
4873 | ||
bea8c150 | 4874 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
4875 | if (!size) |
4876 | continue; | |
4877 | ||
955c1cd7 | 4878 | set_pageblock_order(); |
7c45512d | 4879 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
a2f3aa02 DH |
4880 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4881 | size, MEMMAP_EARLY); | |
718127cc | 4882 | BUG_ON(ret); |
76cdd58e | 4883 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 4884 | zone_start_pfn += size; |
1da177e4 LT |
4885 | } |
4886 | } | |
4887 | ||
577a32f6 | 4888 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 4889 | { |
1da177e4 LT |
4890 | /* Skip empty nodes */ |
4891 | if (!pgdat->node_spanned_pages) | |
4892 | return; | |
4893 | ||
d41dee36 | 4894 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
4895 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
4896 | if (!pgdat->node_mem_map) { | |
e984bb43 | 4897 | unsigned long size, start, end; |
d41dee36 AW |
4898 | struct page *map; |
4899 | ||
e984bb43 BP |
4900 | /* |
4901 | * The zone's endpoints aren't required to be MAX_ORDER | |
4902 | * aligned but the node_mem_map endpoints must be in order | |
4903 | * for the buddy allocator to function correctly. | |
4904 | */ | |
4905 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
108bcc96 | 4906 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
4907 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
4908 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
4909 | map = alloc_remap(pgdat->node_id, size); |
4910 | if (!map) | |
6782832e SS |
4911 | map = memblock_virt_alloc_node_nopanic(size, |
4912 | pgdat->node_id); | |
e984bb43 | 4913 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 4914 | } |
12d810c1 | 4915 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
4916 | /* |
4917 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
4918 | */ | |
c713216d | 4919 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 4920 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 4921 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4922 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 4923 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 4924 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4925 | } |
1da177e4 | 4926 | #endif |
d41dee36 | 4927 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
4928 | } |
4929 | ||
9109fb7b JW |
4930 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
4931 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 4932 | { |
9109fb7b | 4933 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
4934 | unsigned long start_pfn = 0; |
4935 | unsigned long end_pfn = 0; | |
9109fb7b | 4936 | |
88fdf75d | 4937 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 4938 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 4939 | |
1da177e4 LT |
4940 | pgdat->node_id = nid; |
4941 | pgdat->node_start_pfn = node_start_pfn; | |
7960aedd ZY |
4942 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
4943 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
4944 | #endif | |
4945 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
4946 | zones_size, zholes_size); | |
1da177e4 LT |
4947 | |
4948 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
4949 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
4950 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
4951 | nid, (unsigned long)pgdat, | |
4952 | (unsigned long)pgdat->node_mem_map); | |
4953 | #endif | |
1da177e4 | 4954 | |
7960aedd ZY |
4955 | free_area_init_core(pgdat, start_pfn, end_pfn, |
4956 | zones_size, zholes_size); | |
1da177e4 LT |
4957 | } |
4958 | ||
0ee332c1 | 4959 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
4960 | |
4961 | #if MAX_NUMNODES > 1 | |
4962 | /* | |
4963 | * Figure out the number of possible node ids. | |
4964 | */ | |
f9872caf | 4965 | void __init setup_nr_node_ids(void) |
418508c1 MS |
4966 | { |
4967 | unsigned int node; | |
4968 | unsigned int highest = 0; | |
4969 | ||
4970 | for_each_node_mask(node, node_possible_map) | |
4971 | highest = node; | |
4972 | nr_node_ids = highest + 1; | |
4973 | } | |
418508c1 MS |
4974 | #endif |
4975 | ||
1e01979c TH |
4976 | /** |
4977 | * node_map_pfn_alignment - determine the maximum internode alignment | |
4978 | * | |
4979 | * This function should be called after node map is populated and sorted. | |
4980 | * It calculates the maximum power of two alignment which can distinguish | |
4981 | * all the nodes. | |
4982 | * | |
4983 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
4984 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
4985 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
4986 | * shifted, 1GiB is enough and this function will indicate so. | |
4987 | * | |
4988 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
4989 | * model has fine enough granularity to avoid incorrect mapping for the | |
4990 | * populated node map. | |
4991 | * | |
4992 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
4993 | * requirement (single node). | |
4994 | */ | |
4995 | unsigned long __init node_map_pfn_alignment(void) | |
4996 | { | |
4997 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 4998 | unsigned long start, end, mask; |
1e01979c | 4999 | int last_nid = -1; |
c13291a5 | 5000 | int i, nid; |
1e01979c | 5001 | |
c13291a5 | 5002 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5003 | if (!start || last_nid < 0 || last_nid == nid) { |
5004 | last_nid = nid; | |
5005 | last_end = end; | |
5006 | continue; | |
5007 | } | |
5008 | ||
5009 | /* | |
5010 | * Start with a mask granular enough to pin-point to the | |
5011 | * start pfn and tick off bits one-by-one until it becomes | |
5012 | * too coarse to separate the current node from the last. | |
5013 | */ | |
5014 | mask = ~((1 << __ffs(start)) - 1); | |
5015 | while (mask && last_end <= (start & (mask << 1))) | |
5016 | mask <<= 1; | |
5017 | ||
5018 | /* accumulate all internode masks */ | |
5019 | accl_mask |= mask; | |
5020 | } | |
5021 | ||
5022 | /* convert mask to number of pages */ | |
5023 | return ~accl_mask + 1; | |
5024 | } | |
5025 | ||
a6af2bc3 | 5026 | /* Find the lowest pfn for a node */ |
b69a7288 | 5027 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 5028 | { |
a6af2bc3 | 5029 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
5030 | unsigned long start_pfn; |
5031 | int i; | |
1abbfb41 | 5032 | |
c13291a5 TH |
5033 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
5034 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 5035 | |
a6af2bc3 MG |
5036 | if (min_pfn == ULONG_MAX) { |
5037 | printk(KERN_WARNING | |
2bc0d261 | 5038 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
5039 | return 0; |
5040 | } | |
5041 | ||
5042 | return min_pfn; | |
c713216d MG |
5043 | } |
5044 | ||
5045 | /** | |
5046 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
5047 | * | |
5048 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 5049 | * add_active_range(). |
c713216d MG |
5050 | */ |
5051 | unsigned long __init find_min_pfn_with_active_regions(void) | |
5052 | { | |
5053 | return find_min_pfn_for_node(MAX_NUMNODES); | |
5054 | } | |
5055 | ||
37b07e41 LS |
5056 | /* |
5057 | * early_calculate_totalpages() | |
5058 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 5059 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 5060 | */ |
484f51f8 | 5061 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 5062 | { |
7e63efef | 5063 | unsigned long totalpages = 0; |
c13291a5 TH |
5064 | unsigned long start_pfn, end_pfn; |
5065 | int i, nid; | |
5066 | ||
5067 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
5068 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 5069 | |
37b07e41 LS |
5070 | totalpages += pages; |
5071 | if (pages) | |
4b0ef1fe | 5072 | node_set_state(nid, N_MEMORY); |
37b07e41 | 5073 | } |
b8af2941 | 5074 | return totalpages; |
7e63efef MG |
5075 | } |
5076 | ||
2a1e274a MG |
5077 | /* |
5078 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
5079 | * is spread evenly between nodes as long as the nodes have enough | |
5080 | * memory. When they don't, some nodes will have more kernelcore than | |
5081 | * others | |
5082 | */ | |
b224ef85 | 5083 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
5084 | { |
5085 | int i, nid; | |
5086 | unsigned long usable_startpfn; | |
5087 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 5088 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 5089 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 5090 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 5091 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 5092 | struct memblock_region *r; |
b2f3eebe TC |
5093 | |
5094 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
5095 | find_usable_zone_for_movable(); | |
5096 | ||
5097 | /* | |
5098 | * If movable_node is specified, ignore kernelcore and movablecore | |
5099 | * options. | |
5100 | */ | |
5101 | if (movable_node_is_enabled()) { | |
136199f0 EM |
5102 | for_each_memblock(memory, r) { |
5103 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
5104 | continue; |
5105 | ||
136199f0 | 5106 | nid = r->nid; |
b2f3eebe | 5107 | |
136199f0 | 5108 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
5109 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
5110 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5111 | usable_startpfn; | |
5112 | } | |
5113 | ||
5114 | goto out2; | |
5115 | } | |
2a1e274a | 5116 | |
7e63efef | 5117 | /* |
b2f3eebe | 5118 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
5119 | * kernelcore that corresponds so that memory usable for |
5120 | * any allocation type is evenly spread. If both kernelcore | |
5121 | * and movablecore are specified, then the value of kernelcore | |
5122 | * will be used for required_kernelcore if it's greater than | |
5123 | * what movablecore would have allowed. | |
5124 | */ | |
5125 | if (required_movablecore) { | |
7e63efef MG |
5126 | unsigned long corepages; |
5127 | ||
5128 | /* | |
5129 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
5130 | * was requested by the user | |
5131 | */ | |
5132 | required_movablecore = | |
5133 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
5134 | corepages = totalpages - required_movablecore; | |
5135 | ||
5136 | required_kernelcore = max(required_kernelcore, corepages); | |
5137 | } | |
5138 | ||
20e6926d YL |
5139 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
5140 | if (!required_kernelcore) | |
66918dcd | 5141 | goto out; |
2a1e274a MG |
5142 | |
5143 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
5144 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
5145 | ||
5146 | restart: | |
5147 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
5148 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 5149 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
5150 | unsigned long start_pfn, end_pfn; |
5151 | ||
2a1e274a MG |
5152 | /* |
5153 | * Recalculate kernelcore_node if the division per node | |
5154 | * now exceeds what is necessary to satisfy the requested | |
5155 | * amount of memory for the kernel | |
5156 | */ | |
5157 | if (required_kernelcore < kernelcore_node) | |
5158 | kernelcore_node = required_kernelcore / usable_nodes; | |
5159 | ||
5160 | /* | |
5161 | * As the map is walked, we track how much memory is usable | |
5162 | * by the kernel using kernelcore_remaining. When it is | |
5163 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
5164 | */ | |
5165 | kernelcore_remaining = kernelcore_node; | |
5166 | ||
5167 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 5168 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
5169 | unsigned long size_pages; |
5170 | ||
c13291a5 | 5171 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
5172 | if (start_pfn >= end_pfn) |
5173 | continue; | |
5174 | ||
5175 | /* Account for what is only usable for kernelcore */ | |
5176 | if (start_pfn < usable_startpfn) { | |
5177 | unsigned long kernel_pages; | |
5178 | kernel_pages = min(end_pfn, usable_startpfn) | |
5179 | - start_pfn; | |
5180 | ||
5181 | kernelcore_remaining -= min(kernel_pages, | |
5182 | kernelcore_remaining); | |
5183 | required_kernelcore -= min(kernel_pages, | |
5184 | required_kernelcore); | |
5185 | ||
5186 | /* Continue if range is now fully accounted */ | |
5187 | if (end_pfn <= usable_startpfn) { | |
5188 | ||
5189 | /* | |
5190 | * Push zone_movable_pfn to the end so | |
5191 | * that if we have to rebalance | |
5192 | * kernelcore across nodes, we will | |
5193 | * not double account here | |
5194 | */ | |
5195 | zone_movable_pfn[nid] = end_pfn; | |
5196 | continue; | |
5197 | } | |
5198 | start_pfn = usable_startpfn; | |
5199 | } | |
5200 | ||
5201 | /* | |
5202 | * The usable PFN range for ZONE_MOVABLE is from | |
5203 | * start_pfn->end_pfn. Calculate size_pages as the | |
5204 | * number of pages used as kernelcore | |
5205 | */ | |
5206 | size_pages = end_pfn - start_pfn; | |
5207 | if (size_pages > kernelcore_remaining) | |
5208 | size_pages = kernelcore_remaining; | |
5209 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5210 | ||
5211 | /* | |
5212 | * Some kernelcore has been met, update counts and | |
5213 | * break if the kernelcore for this node has been | |
b8af2941 | 5214 | * satisfied |
2a1e274a MG |
5215 | */ |
5216 | required_kernelcore -= min(required_kernelcore, | |
5217 | size_pages); | |
5218 | kernelcore_remaining -= size_pages; | |
5219 | if (!kernelcore_remaining) | |
5220 | break; | |
5221 | } | |
5222 | } | |
5223 | ||
5224 | /* | |
5225 | * If there is still required_kernelcore, we do another pass with one | |
5226 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5227 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 5228 | * satisfied |
2a1e274a MG |
5229 | */ |
5230 | usable_nodes--; | |
5231 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5232 | goto restart; | |
5233 | ||
b2f3eebe | 5234 | out2: |
2a1e274a MG |
5235 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
5236 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5237 | zone_movable_pfn[nid] = | |
5238 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5239 | |
20e6926d | 5240 | out: |
66918dcd | 5241 | /* restore the node_state */ |
4b0ef1fe | 5242 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5243 | } |
5244 | ||
4b0ef1fe LJ |
5245 | /* Any regular or high memory on that node ? */ |
5246 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5247 | { |
37b07e41 LS |
5248 | enum zone_type zone_type; |
5249 | ||
4b0ef1fe LJ |
5250 | if (N_MEMORY == N_NORMAL_MEMORY) |
5251 | return; | |
5252 | ||
5253 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5254 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 5255 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
5256 | node_set_state(nid, N_HIGH_MEMORY); |
5257 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5258 | zone_type <= ZONE_NORMAL) | |
5259 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5260 | break; |
5261 | } | |
37b07e41 | 5262 | } |
37b07e41 LS |
5263 | } |
5264 | ||
c713216d MG |
5265 | /** |
5266 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 5267 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
5268 | * |
5269 | * This will call free_area_init_node() for each active node in the system. | |
5270 | * Using the page ranges provided by add_active_range(), the size of each | |
5271 | * zone in each node and their holes is calculated. If the maximum PFN | |
5272 | * between two adjacent zones match, it is assumed that the zone is empty. | |
5273 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
5274 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
5275 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
5276 | * at arch_max_dma_pfn. | |
5277 | */ | |
5278 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
5279 | { | |
c13291a5 TH |
5280 | unsigned long start_pfn, end_pfn; |
5281 | int i, nid; | |
a6af2bc3 | 5282 | |
c713216d MG |
5283 | /* Record where the zone boundaries are */ |
5284 | memset(arch_zone_lowest_possible_pfn, 0, | |
5285 | sizeof(arch_zone_lowest_possible_pfn)); | |
5286 | memset(arch_zone_highest_possible_pfn, 0, | |
5287 | sizeof(arch_zone_highest_possible_pfn)); | |
5288 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
5289 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
5290 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
5291 | if (i == ZONE_MOVABLE) |
5292 | continue; | |
c713216d MG |
5293 | arch_zone_lowest_possible_pfn[i] = |
5294 | arch_zone_highest_possible_pfn[i-1]; | |
5295 | arch_zone_highest_possible_pfn[i] = | |
5296 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
5297 | } | |
2a1e274a MG |
5298 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
5299 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
5300 | ||
5301 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
5302 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 5303 | find_zone_movable_pfns_for_nodes(); |
c713216d | 5304 | |
c713216d | 5305 | /* Print out the zone ranges */ |
a62e2f4f | 5306 | printk("Zone ranges:\n"); |
2a1e274a MG |
5307 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5308 | if (i == ZONE_MOVABLE) | |
5309 | continue; | |
155cbfc8 | 5310 | printk(KERN_CONT " %-8s ", zone_names[i]); |
72f0ba02 DR |
5311 | if (arch_zone_lowest_possible_pfn[i] == |
5312 | arch_zone_highest_possible_pfn[i]) | |
155cbfc8 | 5313 | printk(KERN_CONT "empty\n"); |
72f0ba02 | 5314 | else |
a62e2f4f BH |
5315 | printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n", |
5316 | arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT, | |
5317 | (arch_zone_highest_possible_pfn[i] | |
5318 | << PAGE_SHIFT) - 1); | |
2a1e274a MG |
5319 | } |
5320 | ||
5321 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
a62e2f4f | 5322 | printk("Movable zone start for each node\n"); |
2a1e274a MG |
5323 | for (i = 0; i < MAX_NUMNODES; i++) { |
5324 | if (zone_movable_pfn[i]) | |
a62e2f4f BH |
5325 | printk(" Node %d: %#010lx\n", i, |
5326 | zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 5327 | } |
c713216d | 5328 | |
f2d52fe5 | 5329 | /* Print out the early node map */ |
a62e2f4f | 5330 | printk("Early memory node ranges\n"); |
c13291a5 | 5331 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
a62e2f4f BH |
5332 | printk(" node %3d: [mem %#010lx-%#010lx]\n", nid, |
5333 | start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
5334 | |
5335 | /* Initialise every node */ | |
708614e6 | 5336 | mminit_verify_pageflags_layout(); |
8ef82866 | 5337 | setup_nr_node_ids(); |
c713216d MG |
5338 | for_each_online_node(nid) { |
5339 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 5340 | free_area_init_node(nid, NULL, |
c713216d | 5341 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
5342 | |
5343 | /* Any memory on that node */ | |
5344 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
5345 | node_set_state(nid, N_MEMORY); |
5346 | check_for_memory(pgdat, nid); | |
c713216d MG |
5347 | } |
5348 | } | |
2a1e274a | 5349 | |
7e63efef | 5350 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
5351 | { |
5352 | unsigned long long coremem; | |
5353 | if (!p) | |
5354 | return -EINVAL; | |
5355 | ||
5356 | coremem = memparse(p, &p); | |
7e63efef | 5357 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 5358 | |
7e63efef | 5359 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
5360 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
5361 | ||
5362 | return 0; | |
5363 | } | |
ed7ed365 | 5364 | |
7e63efef MG |
5365 | /* |
5366 | * kernelcore=size sets the amount of memory for use for allocations that | |
5367 | * cannot be reclaimed or migrated. | |
5368 | */ | |
5369 | static int __init cmdline_parse_kernelcore(char *p) | |
5370 | { | |
5371 | return cmdline_parse_core(p, &required_kernelcore); | |
5372 | } | |
5373 | ||
5374 | /* | |
5375 | * movablecore=size sets the amount of memory for use for allocations that | |
5376 | * can be reclaimed or migrated. | |
5377 | */ | |
5378 | static int __init cmdline_parse_movablecore(char *p) | |
5379 | { | |
5380 | return cmdline_parse_core(p, &required_movablecore); | |
5381 | } | |
5382 | ||
ed7ed365 | 5383 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 5384 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 5385 | |
0ee332c1 | 5386 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5387 | |
c3d5f5f0 JL |
5388 | void adjust_managed_page_count(struct page *page, long count) |
5389 | { | |
5390 | spin_lock(&managed_page_count_lock); | |
5391 | page_zone(page)->managed_pages += count; | |
5392 | totalram_pages += count; | |
3dcc0571 JL |
5393 | #ifdef CONFIG_HIGHMEM |
5394 | if (PageHighMem(page)) | |
5395 | totalhigh_pages += count; | |
5396 | #endif | |
c3d5f5f0 JL |
5397 | spin_unlock(&managed_page_count_lock); |
5398 | } | |
3dcc0571 | 5399 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 5400 | |
11199692 | 5401 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 5402 | { |
11199692 JL |
5403 | void *pos; |
5404 | unsigned long pages = 0; | |
69afade7 | 5405 | |
11199692 JL |
5406 | start = (void *)PAGE_ALIGN((unsigned long)start); |
5407 | end = (void *)((unsigned long)end & PAGE_MASK); | |
5408 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 5409 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
5410 | memset(pos, poison, PAGE_SIZE); |
5411 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
5412 | } |
5413 | ||
5414 | if (pages && s) | |
11199692 | 5415 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
5416 | s, pages << (PAGE_SHIFT - 10), start, end); |
5417 | ||
5418 | return pages; | |
5419 | } | |
11199692 | 5420 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 5421 | |
cfa11e08 JL |
5422 | #ifdef CONFIG_HIGHMEM |
5423 | void free_highmem_page(struct page *page) | |
5424 | { | |
5425 | __free_reserved_page(page); | |
5426 | totalram_pages++; | |
7b4b2a0d | 5427 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
5428 | totalhigh_pages++; |
5429 | } | |
5430 | #endif | |
5431 | ||
7ee3d4e8 JL |
5432 | |
5433 | void __init mem_init_print_info(const char *str) | |
5434 | { | |
5435 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
5436 | unsigned long init_code_size, init_data_size; | |
5437 | ||
5438 | physpages = get_num_physpages(); | |
5439 | codesize = _etext - _stext; | |
5440 | datasize = _edata - _sdata; | |
5441 | rosize = __end_rodata - __start_rodata; | |
5442 | bss_size = __bss_stop - __bss_start; | |
5443 | init_data_size = __init_end - __init_begin; | |
5444 | init_code_size = _einittext - _sinittext; | |
5445 | ||
5446 | /* | |
5447 | * Detect special cases and adjust section sizes accordingly: | |
5448 | * 1) .init.* may be embedded into .data sections | |
5449 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
5450 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
5451 | * 3) .rodata.* may be embedded into .text or .data sections. | |
5452 | */ | |
5453 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
5454 | do { \ |
5455 | if (start <= pos && pos < end && size > adj) \ | |
5456 | size -= adj; \ | |
5457 | } while (0) | |
7ee3d4e8 JL |
5458 | |
5459 | adj_init_size(__init_begin, __init_end, init_data_size, | |
5460 | _sinittext, init_code_size); | |
5461 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
5462 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
5463 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
5464 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
5465 | ||
5466 | #undef adj_init_size | |
5467 | ||
5468 | printk("Memory: %luK/%luK available " | |
5469 | "(%luK kernel code, %luK rwdata, %luK rodata, " | |
5470 | "%luK init, %luK bss, %luK reserved" | |
5471 | #ifdef CONFIG_HIGHMEM | |
5472 | ", %luK highmem" | |
5473 | #endif | |
5474 | "%s%s)\n", | |
5475 | nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10), | |
5476 | codesize >> 10, datasize >> 10, rosize >> 10, | |
5477 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
5478 | (physpages - totalram_pages) << (PAGE_SHIFT-10), | |
5479 | #ifdef CONFIG_HIGHMEM | |
5480 | totalhigh_pages << (PAGE_SHIFT-10), | |
5481 | #endif | |
5482 | str ? ", " : "", str ? str : ""); | |
5483 | } | |
5484 | ||
0e0b864e | 5485 | /** |
88ca3b94 RD |
5486 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
5487 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
5488 | * |
5489 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
5490 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
5491 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
5492 | * function may optionally be used to account for unfreeable pages in the |
5493 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
5494 | * smaller per-cpu batchsize. | |
0e0b864e MG |
5495 | */ |
5496 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
5497 | { | |
5498 | dma_reserve = new_dma_reserve; | |
5499 | } | |
5500 | ||
1da177e4 LT |
5501 | void __init free_area_init(unsigned long *zones_size) |
5502 | { | |
9109fb7b | 5503 | free_area_init_node(0, zones_size, |
1da177e4 LT |
5504 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
5505 | } | |
1da177e4 | 5506 | |
1da177e4 LT |
5507 | static int page_alloc_cpu_notify(struct notifier_block *self, |
5508 | unsigned long action, void *hcpu) | |
5509 | { | |
5510 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 5511 | |
8bb78442 | 5512 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 5513 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
5514 | drain_pages(cpu); |
5515 | ||
5516 | /* | |
5517 | * Spill the event counters of the dead processor | |
5518 | * into the current processors event counters. | |
5519 | * This artificially elevates the count of the current | |
5520 | * processor. | |
5521 | */ | |
f8891e5e | 5522 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
5523 | |
5524 | /* | |
5525 | * Zero the differential counters of the dead processor | |
5526 | * so that the vm statistics are consistent. | |
5527 | * | |
5528 | * This is only okay since the processor is dead and cannot | |
5529 | * race with what we are doing. | |
5530 | */ | |
2bb921e5 | 5531 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
5532 | } |
5533 | return NOTIFY_OK; | |
5534 | } | |
1da177e4 LT |
5535 | |
5536 | void __init page_alloc_init(void) | |
5537 | { | |
5538 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
5539 | } | |
5540 | ||
cb45b0e9 HA |
5541 | /* |
5542 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
5543 | * or min_free_kbytes changes. | |
5544 | */ | |
5545 | static void calculate_totalreserve_pages(void) | |
5546 | { | |
5547 | struct pglist_data *pgdat; | |
5548 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5549 | enum zone_type i, j; |
cb45b0e9 HA |
5550 | |
5551 | for_each_online_pgdat(pgdat) { | |
5552 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
5553 | struct zone *zone = pgdat->node_zones + i; | |
5554 | unsigned long max = 0; | |
5555 | ||
5556 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5557 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5558 | if (zone->lowmem_reserve[j] > max) | |
5559 | max = zone->lowmem_reserve[j]; | |
5560 | } | |
5561 | ||
41858966 MG |
5562 | /* we treat the high watermark as reserved pages. */ |
5563 | max += high_wmark_pages(zone); | |
cb45b0e9 | 5564 | |
b40da049 JL |
5565 | if (max > zone->managed_pages) |
5566 | max = zone->managed_pages; | |
cb45b0e9 | 5567 | reserve_pages += max; |
ab8fabd4 JW |
5568 | /* |
5569 | * Lowmem reserves are not available to | |
5570 | * GFP_HIGHUSER page cache allocations and | |
5571 | * kswapd tries to balance zones to their high | |
5572 | * watermark. As a result, neither should be | |
5573 | * regarded as dirtyable memory, to prevent a | |
5574 | * situation where reclaim has to clean pages | |
5575 | * in order to balance the zones. | |
5576 | */ | |
5577 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
5578 | } |
5579 | } | |
ab8fabd4 | 5580 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
5581 | totalreserve_pages = reserve_pages; |
5582 | } | |
5583 | ||
1da177e4 LT |
5584 | /* |
5585 | * setup_per_zone_lowmem_reserve - called whenever | |
5586 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
5587 | * has a correct pages reserved value, so an adequate number of | |
5588 | * pages are left in the zone after a successful __alloc_pages(). | |
5589 | */ | |
5590 | static void setup_per_zone_lowmem_reserve(void) | |
5591 | { | |
5592 | struct pglist_data *pgdat; | |
2f6726e5 | 5593 | enum zone_type j, idx; |
1da177e4 | 5594 | |
ec936fc5 | 5595 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
5596 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5597 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 5598 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
5599 | |
5600 | zone->lowmem_reserve[j] = 0; | |
5601 | ||
2f6726e5 CL |
5602 | idx = j; |
5603 | while (idx) { | |
1da177e4 LT |
5604 | struct zone *lower_zone; |
5605 | ||
2f6726e5 CL |
5606 | idx--; |
5607 | ||
1da177e4 LT |
5608 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
5609 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
5610 | ||
5611 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 5612 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 5613 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 5614 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
5615 | } |
5616 | } | |
5617 | } | |
cb45b0e9 HA |
5618 | |
5619 | /* update totalreserve_pages */ | |
5620 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5621 | } |
5622 | ||
cfd3da1e | 5623 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5624 | { |
5625 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5626 | unsigned long lowmem_pages = 0; | |
5627 | struct zone *zone; | |
5628 | unsigned long flags; | |
5629 | ||
5630 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5631 | for_each_zone(zone) { | |
5632 | if (!is_highmem(zone)) | |
b40da049 | 5633 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
5634 | } |
5635 | ||
5636 | for_each_zone(zone) { | |
ac924c60 AM |
5637 | u64 tmp; |
5638 | ||
1125b4e3 | 5639 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 5640 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 5641 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
5642 | if (is_highmem(zone)) { |
5643 | /* | |
669ed175 NP |
5644 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5645 | * need highmem pages, so cap pages_min to a small | |
5646 | * value here. | |
5647 | * | |
41858966 | 5648 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
5649 | * deltas controls asynch page reclaim, and so should |
5650 | * not be capped for highmem. | |
1da177e4 | 5651 | */ |
90ae8d67 | 5652 | unsigned long min_pages; |
1da177e4 | 5653 | |
b40da049 | 5654 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 5655 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 5656 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5657 | } else { |
669ed175 NP |
5658 | /* |
5659 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5660 | * proportionate to the zone's size. |
5661 | */ | |
41858966 | 5662 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5663 | } |
5664 | ||
41858966 MG |
5665 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
5666 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
49f223a9 | 5667 | |
81c0a2bb JW |
5668 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
5669 | high_wmark_pages(zone) - | |
5670 | low_wmark_pages(zone) - | |
5671 | zone_page_state(zone, NR_ALLOC_BATCH)); | |
5672 | ||
56fd56b8 | 5673 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 5674 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5675 | } |
cb45b0e9 HA |
5676 | |
5677 | /* update totalreserve_pages */ | |
5678 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5679 | } |
5680 | ||
cfd3da1e MG |
5681 | /** |
5682 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5683 | * or when memory is hot-{added|removed} | |
5684 | * | |
5685 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5686 | * correctly with respect to min_free_kbytes. | |
5687 | */ | |
5688 | void setup_per_zone_wmarks(void) | |
5689 | { | |
5690 | mutex_lock(&zonelists_mutex); | |
5691 | __setup_per_zone_wmarks(); | |
5692 | mutex_unlock(&zonelists_mutex); | |
5693 | } | |
5694 | ||
55a4462a | 5695 | /* |
556adecb RR |
5696 | * The inactive anon list should be small enough that the VM never has to |
5697 | * do too much work, but large enough that each inactive page has a chance | |
5698 | * to be referenced again before it is swapped out. | |
5699 | * | |
5700 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
5701 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
5702 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
5703 | * the anonymous pages are kept on the inactive list. | |
5704 | * | |
5705 | * total target max | |
5706 | * memory ratio inactive anon | |
5707 | * ------------------------------------- | |
5708 | * 10MB 1 5MB | |
5709 | * 100MB 1 50MB | |
5710 | * 1GB 3 250MB | |
5711 | * 10GB 10 0.9GB | |
5712 | * 100GB 31 3GB | |
5713 | * 1TB 101 10GB | |
5714 | * 10TB 320 32GB | |
5715 | */ | |
1b79acc9 | 5716 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5717 | { |
96cb4df5 | 5718 | unsigned int gb, ratio; |
556adecb | 5719 | |
96cb4df5 | 5720 | /* Zone size in gigabytes */ |
b40da049 | 5721 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 5722 | if (gb) |
556adecb | 5723 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5724 | else |
5725 | ratio = 1; | |
556adecb | 5726 | |
96cb4df5 MK |
5727 | zone->inactive_ratio = ratio; |
5728 | } | |
556adecb | 5729 | |
839a4fcc | 5730 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
5731 | { |
5732 | struct zone *zone; | |
5733 | ||
5734 | for_each_zone(zone) | |
5735 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5736 | } |
5737 | ||
1da177e4 LT |
5738 | /* |
5739 | * Initialise min_free_kbytes. | |
5740 | * | |
5741 | * For small machines we want it small (128k min). For large machines | |
5742 | * we want it large (64MB max). But it is not linear, because network | |
5743 | * bandwidth does not increase linearly with machine size. We use | |
5744 | * | |
b8af2941 | 5745 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
5746 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
5747 | * | |
5748 | * which yields | |
5749 | * | |
5750 | * 16MB: 512k | |
5751 | * 32MB: 724k | |
5752 | * 64MB: 1024k | |
5753 | * 128MB: 1448k | |
5754 | * 256MB: 2048k | |
5755 | * 512MB: 2896k | |
5756 | * 1024MB: 4096k | |
5757 | * 2048MB: 5792k | |
5758 | * 4096MB: 8192k | |
5759 | * 8192MB: 11584k | |
5760 | * 16384MB: 16384k | |
5761 | */ | |
1b79acc9 | 5762 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
5763 | { |
5764 | unsigned long lowmem_kbytes; | |
5f12733e | 5765 | int new_min_free_kbytes; |
1da177e4 LT |
5766 | |
5767 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
5768 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
5769 | ||
5770 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
5771 | min_free_kbytes = new_min_free_kbytes; | |
5772 | if (min_free_kbytes < 128) | |
5773 | min_free_kbytes = 128; | |
5774 | if (min_free_kbytes > 65536) | |
5775 | min_free_kbytes = 65536; | |
5776 | } else { | |
5777 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
5778 | new_min_free_kbytes, user_min_free_kbytes); | |
5779 | } | |
bc75d33f | 5780 | setup_per_zone_wmarks(); |
a6cccdc3 | 5781 | refresh_zone_stat_thresholds(); |
1da177e4 | 5782 | setup_per_zone_lowmem_reserve(); |
556adecb | 5783 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5784 | return 0; |
5785 | } | |
bc75d33f | 5786 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5787 | |
5788 | /* | |
b8af2941 | 5789 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
5790 | * that we can call two helper functions whenever min_free_kbytes |
5791 | * changes. | |
5792 | */ | |
b8af2941 | 5793 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, |
8d65af78 | 5794 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5795 | { |
da8c757b HP |
5796 | int rc; |
5797 | ||
5798 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
5799 | if (rc) | |
5800 | return rc; | |
5801 | ||
5f12733e MH |
5802 | if (write) { |
5803 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 5804 | setup_per_zone_wmarks(); |
5f12733e | 5805 | } |
1da177e4 LT |
5806 | return 0; |
5807 | } | |
5808 | ||
9614634f CL |
5809 | #ifdef CONFIG_NUMA |
5810 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5811 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5812 | { |
5813 | struct zone *zone; | |
5814 | int rc; | |
5815 | ||
8d65af78 | 5816 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5817 | if (rc) |
5818 | return rc; | |
5819 | ||
5820 | for_each_zone(zone) | |
b40da049 | 5821 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
5822 | sysctl_min_unmapped_ratio) / 100; |
5823 | return 0; | |
5824 | } | |
0ff38490 CL |
5825 | |
5826 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5827 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5828 | { |
5829 | struct zone *zone; | |
5830 | int rc; | |
5831 | ||
8d65af78 | 5832 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5833 | if (rc) |
5834 | return rc; | |
5835 | ||
5836 | for_each_zone(zone) | |
b40da049 | 5837 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
5838 | sysctl_min_slab_ratio) / 100; |
5839 | return 0; | |
5840 | } | |
9614634f CL |
5841 | #endif |
5842 | ||
1da177e4 LT |
5843 | /* |
5844 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5845 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5846 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5847 | * | |
5848 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5849 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5850 | * if in function of the boot time zone sizes. |
5851 | */ | |
5852 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5853 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5854 | { |
8d65af78 | 5855 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5856 | setup_per_zone_lowmem_reserve(); |
5857 | return 0; | |
5858 | } | |
5859 | ||
8ad4b1fb RS |
5860 | /* |
5861 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
5862 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
5863 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 5864 | */ |
8ad4b1fb | 5865 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, |
8d65af78 | 5866 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5867 | { |
5868 | struct zone *zone; | |
5869 | unsigned int cpu; | |
5870 | int ret; | |
5871 | ||
8d65af78 | 5872 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
93278814 | 5873 | if (!write || (ret < 0)) |
8ad4b1fb | 5874 | return ret; |
c8e251fa CS |
5875 | |
5876 | mutex_lock(&pcp_batch_high_lock); | |
364df0eb | 5877 | for_each_populated_zone(zone) { |
22a7f12b CS |
5878 | unsigned long high; |
5879 | high = zone->managed_pages / percpu_pagelist_fraction; | |
5880 | for_each_possible_cpu(cpu) | |
3664033c CS |
5881 | pageset_set_high(per_cpu_ptr(zone->pageset, cpu), |
5882 | high); | |
8ad4b1fb | 5883 | } |
c8e251fa | 5884 | mutex_unlock(&pcp_batch_high_lock); |
8ad4b1fb RS |
5885 | return 0; |
5886 | } | |
5887 | ||
f034b5d4 | 5888 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
5889 | |
5890 | #ifdef CONFIG_NUMA | |
5891 | static int __init set_hashdist(char *str) | |
5892 | { | |
5893 | if (!str) | |
5894 | return 0; | |
5895 | hashdist = simple_strtoul(str, &str, 0); | |
5896 | return 1; | |
5897 | } | |
5898 | __setup("hashdist=", set_hashdist); | |
5899 | #endif | |
5900 | ||
5901 | /* | |
5902 | * allocate a large system hash table from bootmem | |
5903 | * - it is assumed that the hash table must contain an exact power-of-2 | |
5904 | * quantity of entries | |
5905 | * - limit is the number of hash buckets, not the total allocation size | |
5906 | */ | |
5907 | void *__init alloc_large_system_hash(const char *tablename, | |
5908 | unsigned long bucketsize, | |
5909 | unsigned long numentries, | |
5910 | int scale, | |
5911 | int flags, | |
5912 | unsigned int *_hash_shift, | |
5913 | unsigned int *_hash_mask, | |
31fe62b9 TB |
5914 | unsigned long low_limit, |
5915 | unsigned long high_limit) | |
1da177e4 | 5916 | { |
31fe62b9 | 5917 | unsigned long long max = high_limit; |
1da177e4 LT |
5918 | unsigned long log2qty, size; |
5919 | void *table = NULL; | |
5920 | ||
5921 | /* allow the kernel cmdline to have a say */ | |
5922 | if (!numentries) { | |
5923 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 5924 | numentries = nr_kernel_pages; |
a7e83318 JZ |
5925 | |
5926 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
5927 | if (PAGE_SHIFT < 20) | |
5928 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
5929 | |
5930 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
5931 | if (scale > PAGE_SHIFT) | |
5932 | numentries >>= (scale - PAGE_SHIFT); | |
5933 | else | |
5934 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
5935 | |
5936 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
5937 | if (unlikely(flags & HASH_SMALL)) { |
5938 | /* Makes no sense without HASH_EARLY */ | |
5939 | WARN_ON(!(flags & HASH_EARLY)); | |
5940 | if (!(numentries >> *_hash_shift)) { | |
5941 | numentries = 1UL << *_hash_shift; | |
5942 | BUG_ON(!numentries); | |
5943 | } | |
5944 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 5945 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 5946 | } |
6e692ed3 | 5947 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
5948 | |
5949 | /* limit allocation size to 1/16 total memory by default */ | |
5950 | if (max == 0) { | |
5951 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
5952 | do_div(max, bucketsize); | |
5953 | } | |
074b8517 | 5954 | max = min(max, 0x80000000ULL); |
1da177e4 | 5955 | |
31fe62b9 TB |
5956 | if (numentries < low_limit) |
5957 | numentries = low_limit; | |
1da177e4 LT |
5958 | if (numentries > max) |
5959 | numentries = max; | |
5960 | ||
f0d1b0b3 | 5961 | log2qty = ilog2(numentries); |
1da177e4 LT |
5962 | |
5963 | do { | |
5964 | size = bucketsize << log2qty; | |
5965 | if (flags & HASH_EARLY) | |
6782832e | 5966 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
5967 | else if (hashdist) |
5968 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
5969 | else { | |
1037b83b ED |
5970 | /* |
5971 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
5972 | * some pages at the end of hash table which |
5973 | * alloc_pages_exact() automatically does | |
1037b83b | 5974 | */ |
264ef8a9 | 5975 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 5976 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
5977 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
5978 | } | |
1da177e4 LT |
5979 | } |
5980 | } while (!table && size > PAGE_SIZE && --log2qty); | |
5981 | ||
5982 | if (!table) | |
5983 | panic("Failed to allocate %s hash table\n", tablename); | |
5984 | ||
f241e660 | 5985 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 5986 | tablename, |
f241e660 | 5987 | (1UL << log2qty), |
f0d1b0b3 | 5988 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
5989 | size); |
5990 | ||
5991 | if (_hash_shift) | |
5992 | *_hash_shift = log2qty; | |
5993 | if (_hash_mask) | |
5994 | *_hash_mask = (1 << log2qty) - 1; | |
5995 | ||
5996 | return table; | |
5997 | } | |
a117e66e | 5998 | |
835c134e MG |
5999 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
6000 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
6001 | unsigned long pfn) | |
6002 | { | |
6003 | #ifdef CONFIG_SPARSEMEM | |
6004 | return __pfn_to_section(pfn)->pageblock_flags; | |
6005 | #else | |
6006 | return zone->pageblock_flags; | |
6007 | #endif /* CONFIG_SPARSEMEM */ | |
6008 | } | |
6009 | ||
6010 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
6011 | { | |
6012 | #ifdef CONFIG_SPARSEMEM | |
6013 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 6014 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 6015 | #else |
c060f943 | 6016 | pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 6017 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
6018 | #endif /* CONFIG_SPARSEMEM */ |
6019 | } | |
6020 | ||
6021 | /** | |
d9c23400 | 6022 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
6023 | * @page: The page within the block of interest |
6024 | * @start_bitidx: The first bit of interest to retrieve | |
6025 | * @end_bitidx: The last bit of interest | |
6026 | * returns pageblock_bits flags | |
6027 | */ | |
6028 | unsigned long get_pageblock_flags_group(struct page *page, | |
6029 | int start_bitidx, int end_bitidx) | |
6030 | { | |
6031 | struct zone *zone; | |
6032 | unsigned long *bitmap; | |
6033 | unsigned long pfn, bitidx; | |
6034 | unsigned long flags = 0; | |
6035 | unsigned long value = 1; | |
6036 | ||
6037 | zone = page_zone(page); | |
6038 | pfn = page_to_pfn(page); | |
6039 | bitmap = get_pageblock_bitmap(zone, pfn); | |
6040 | bitidx = pfn_to_bitidx(zone, pfn); | |
6041 | ||
6042 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
6043 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
6044 | flags |= value; | |
6220ec78 | 6045 | |
835c134e MG |
6046 | return flags; |
6047 | } | |
6048 | ||
6049 | /** | |
d9c23400 | 6050 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
6051 | * @page: The page within the block of interest |
6052 | * @start_bitidx: The first bit of interest | |
6053 | * @end_bitidx: The last bit of interest | |
6054 | * @flags: The flags to set | |
6055 | */ | |
6056 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
6057 | int start_bitidx, int end_bitidx) | |
6058 | { | |
6059 | struct zone *zone; | |
6060 | unsigned long *bitmap; | |
6061 | unsigned long pfn, bitidx; | |
6062 | unsigned long value = 1; | |
6063 | ||
6064 | zone = page_zone(page); | |
6065 | pfn = page_to_pfn(page); | |
6066 | bitmap = get_pageblock_bitmap(zone, pfn); | |
6067 | bitidx = pfn_to_bitidx(zone, pfn); | |
309381fe | 6068 | VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page); |
835c134e MG |
6069 | |
6070 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
6071 | if (flags & value) | |
6072 | __set_bit(bitidx + start_bitidx, bitmap); | |
6073 | else | |
6074 | __clear_bit(bitidx + start_bitidx, bitmap); | |
6075 | } | |
a5d76b54 KH |
6076 | |
6077 | /* | |
80934513 MK |
6078 | * This function checks whether pageblock includes unmovable pages or not. |
6079 | * If @count is not zero, it is okay to include less @count unmovable pages | |
6080 | * | |
b8af2941 | 6081 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
6082 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
6083 | * expect this function should be exact. | |
a5d76b54 | 6084 | */ |
b023f468 WC |
6085 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
6086 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
6087 | { |
6088 | unsigned long pfn, iter, found; | |
47118af0 MN |
6089 | int mt; |
6090 | ||
49ac8255 KH |
6091 | /* |
6092 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 6093 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
6094 | */ |
6095 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 6096 | return false; |
47118af0 MN |
6097 | mt = get_pageblock_migratetype(page); |
6098 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 6099 | return false; |
49ac8255 KH |
6100 | |
6101 | pfn = page_to_pfn(page); | |
6102 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
6103 | unsigned long check = pfn + iter; | |
6104 | ||
29723fcc | 6105 | if (!pfn_valid_within(check)) |
49ac8255 | 6106 | continue; |
29723fcc | 6107 | |
49ac8255 | 6108 | page = pfn_to_page(check); |
c8721bbb NH |
6109 | |
6110 | /* | |
6111 | * Hugepages are not in LRU lists, but they're movable. | |
6112 | * We need not scan over tail pages bacause we don't | |
6113 | * handle each tail page individually in migration. | |
6114 | */ | |
6115 | if (PageHuge(page)) { | |
6116 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
6117 | continue; | |
6118 | } | |
6119 | ||
97d255c8 MK |
6120 | /* |
6121 | * We can't use page_count without pin a page | |
6122 | * because another CPU can free compound page. | |
6123 | * This check already skips compound tails of THP | |
6124 | * because their page->_count is zero at all time. | |
6125 | */ | |
6126 | if (!atomic_read(&page->_count)) { | |
49ac8255 KH |
6127 | if (PageBuddy(page)) |
6128 | iter += (1 << page_order(page)) - 1; | |
6129 | continue; | |
6130 | } | |
97d255c8 | 6131 | |
b023f468 WC |
6132 | /* |
6133 | * The HWPoisoned page may be not in buddy system, and | |
6134 | * page_count() is not 0. | |
6135 | */ | |
6136 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6137 | continue; | |
6138 | ||
49ac8255 KH |
6139 | if (!PageLRU(page)) |
6140 | found++; | |
6141 | /* | |
6142 | * If there are RECLAIMABLE pages, we need to check it. | |
6143 | * But now, memory offline itself doesn't call shrink_slab() | |
6144 | * and it still to be fixed. | |
6145 | */ | |
6146 | /* | |
6147 | * If the page is not RAM, page_count()should be 0. | |
6148 | * we don't need more check. This is an _used_ not-movable page. | |
6149 | * | |
6150 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6151 | * is set to both of a memory hole page and a _used_ kernel | |
6152 | * page at boot. | |
6153 | */ | |
6154 | if (found > count) | |
80934513 | 6155 | return true; |
49ac8255 | 6156 | } |
80934513 | 6157 | return false; |
49ac8255 KH |
6158 | } |
6159 | ||
6160 | bool is_pageblock_removable_nolock(struct page *page) | |
6161 | { | |
656a0706 MH |
6162 | struct zone *zone; |
6163 | unsigned long pfn; | |
687875fb MH |
6164 | |
6165 | /* | |
6166 | * We have to be careful here because we are iterating over memory | |
6167 | * sections which are not zone aware so we might end up outside of | |
6168 | * the zone but still within the section. | |
656a0706 MH |
6169 | * We have to take care about the node as well. If the node is offline |
6170 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 6171 | */ |
656a0706 MH |
6172 | if (!node_online(page_to_nid(page))) |
6173 | return false; | |
6174 | ||
6175 | zone = page_zone(page); | |
6176 | pfn = page_to_pfn(page); | |
108bcc96 | 6177 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
6178 | return false; |
6179 | ||
b023f468 | 6180 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6181 | } |
0c0e6195 | 6182 | |
041d3a8c MN |
6183 | #ifdef CONFIG_CMA |
6184 | ||
6185 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6186 | { | |
6187 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6188 | pageblock_nr_pages) - 1); | |
6189 | } | |
6190 | ||
6191 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6192 | { | |
6193 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6194 | pageblock_nr_pages)); | |
6195 | } | |
6196 | ||
041d3a8c | 6197 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6198 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6199 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6200 | { |
6201 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6202 | unsigned long nr_reclaimed; |
041d3a8c MN |
6203 | unsigned long pfn = start; |
6204 | unsigned int tries = 0; | |
6205 | int ret = 0; | |
6206 | ||
be49a6e1 | 6207 | migrate_prep(); |
041d3a8c | 6208 | |
bb13ffeb | 6209 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6210 | if (fatal_signal_pending(current)) { |
6211 | ret = -EINTR; | |
6212 | break; | |
6213 | } | |
6214 | ||
bb13ffeb MG |
6215 | if (list_empty(&cc->migratepages)) { |
6216 | cc->nr_migratepages = 0; | |
6217 | pfn = isolate_migratepages_range(cc->zone, cc, | |
e46a2879 | 6218 | pfn, end, true); |
041d3a8c MN |
6219 | if (!pfn) { |
6220 | ret = -EINTR; | |
6221 | break; | |
6222 | } | |
6223 | tries = 0; | |
6224 | } else if (++tries == 5) { | |
6225 | ret = ret < 0 ? ret : -EBUSY; | |
6226 | break; | |
6227 | } | |
6228 | ||
beb51eaa MK |
6229 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6230 | &cc->migratepages); | |
6231 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6232 | |
9c620e2b | 6233 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 6234 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 6235 | } |
2a6f5124 SP |
6236 | if (ret < 0) { |
6237 | putback_movable_pages(&cc->migratepages); | |
6238 | return ret; | |
6239 | } | |
6240 | return 0; | |
041d3a8c MN |
6241 | } |
6242 | ||
6243 | /** | |
6244 | * alloc_contig_range() -- tries to allocate given range of pages | |
6245 | * @start: start PFN to allocate | |
6246 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
6247 | * @migratetype: migratetype of the underlaying pageblocks (either |
6248 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
6249 | * in range must have the same migratetype and it must | |
6250 | * be either of the two. | |
041d3a8c MN |
6251 | * |
6252 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
6253 | * aligned, however it's the caller's responsibility to guarantee that | |
6254 | * we are the only thread that changes migrate type of pageblocks the | |
6255 | * pages fall in. | |
6256 | * | |
6257 | * The PFN range must belong to a single zone. | |
6258 | * | |
6259 | * Returns zero on success or negative error code. On success all | |
6260 | * pages which PFN is in [start, end) are allocated for the caller and | |
6261 | * need to be freed with free_contig_range(). | |
6262 | */ | |
0815f3d8 MN |
6263 | int alloc_contig_range(unsigned long start, unsigned long end, |
6264 | unsigned migratetype) | |
041d3a8c | 6265 | { |
041d3a8c MN |
6266 | unsigned long outer_start, outer_end; |
6267 | int ret = 0, order; | |
6268 | ||
bb13ffeb MG |
6269 | struct compact_control cc = { |
6270 | .nr_migratepages = 0, | |
6271 | .order = -1, | |
6272 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 6273 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
6274 | .ignore_skip_hint = true, |
6275 | }; | |
6276 | INIT_LIST_HEAD(&cc.migratepages); | |
6277 | ||
041d3a8c MN |
6278 | /* |
6279 | * What we do here is we mark all pageblocks in range as | |
6280 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6281 | * have different sizes, and due to the way page allocator | |
6282 | * work, we align the range to biggest of the two pages so | |
6283 | * that page allocator won't try to merge buddies from | |
6284 | * different pageblocks and change MIGRATE_ISOLATE to some | |
6285 | * other migration type. | |
6286 | * | |
6287 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6288 | * migrate the pages from an unaligned range (ie. pages that | |
6289 | * we are interested in). This will put all the pages in | |
6290 | * range back to page allocator as MIGRATE_ISOLATE. | |
6291 | * | |
6292 | * When this is done, we take the pages in range from page | |
6293 | * allocator removing them from the buddy system. This way | |
6294 | * page allocator will never consider using them. | |
6295 | * | |
6296 | * This lets us mark the pageblocks back as | |
6297 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6298 | * aligned range but not in the unaligned, original range are | |
6299 | * put back to page allocator so that buddy can use them. | |
6300 | */ | |
6301 | ||
6302 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
6303 | pfn_max_align_up(end), migratetype, |
6304 | false); | |
041d3a8c | 6305 | if (ret) |
86a595f9 | 6306 | return ret; |
041d3a8c | 6307 | |
bb13ffeb | 6308 | ret = __alloc_contig_migrate_range(&cc, start, end); |
041d3a8c MN |
6309 | if (ret) |
6310 | goto done; | |
6311 | ||
6312 | /* | |
6313 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
6314 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
6315 | * more, all pages in [start, end) are free in page allocator. | |
6316 | * What we are going to do is to allocate all pages from | |
6317 | * [start, end) (that is remove them from page allocator). | |
6318 | * | |
6319 | * The only problem is that pages at the beginning and at the | |
6320 | * end of interesting range may be not aligned with pages that | |
6321 | * page allocator holds, ie. they can be part of higher order | |
6322 | * pages. Because of this, we reserve the bigger range and | |
6323 | * once this is done free the pages we are not interested in. | |
6324 | * | |
6325 | * We don't have to hold zone->lock here because the pages are | |
6326 | * isolated thus they won't get removed from buddy. | |
6327 | */ | |
6328 | ||
6329 | lru_add_drain_all(); | |
6330 | drain_all_pages(); | |
6331 | ||
6332 | order = 0; | |
6333 | outer_start = start; | |
6334 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
6335 | if (++order >= MAX_ORDER) { | |
6336 | ret = -EBUSY; | |
6337 | goto done; | |
6338 | } | |
6339 | outer_start &= ~0UL << order; | |
6340 | } | |
6341 | ||
6342 | /* Make sure the range is really isolated. */ | |
b023f468 | 6343 | if (test_pages_isolated(outer_start, end, false)) { |
041d3a8c MN |
6344 | pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n", |
6345 | outer_start, end); | |
6346 | ret = -EBUSY; | |
6347 | goto done; | |
6348 | } | |
6349 | ||
49f223a9 MS |
6350 | |
6351 | /* Grab isolated pages from freelists. */ | |
bb13ffeb | 6352 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6353 | if (!outer_end) { |
6354 | ret = -EBUSY; | |
6355 | goto done; | |
6356 | } | |
6357 | ||
6358 | /* Free head and tail (if any) */ | |
6359 | if (start != outer_start) | |
6360 | free_contig_range(outer_start, start - outer_start); | |
6361 | if (end != outer_end) | |
6362 | free_contig_range(end, outer_end - end); | |
6363 | ||
6364 | done: | |
6365 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 6366 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
6367 | return ret; |
6368 | } | |
6369 | ||
6370 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
6371 | { | |
bcc2b02f MS |
6372 | unsigned int count = 0; |
6373 | ||
6374 | for (; nr_pages--; pfn++) { | |
6375 | struct page *page = pfn_to_page(pfn); | |
6376 | ||
6377 | count += page_count(page) != 1; | |
6378 | __free_page(page); | |
6379 | } | |
6380 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
6381 | } |
6382 | #endif | |
6383 | ||
4ed7e022 | 6384 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
6385 | /* |
6386 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
6387 | * page high values need to be recalulated. | |
6388 | */ | |
4ed7e022 JL |
6389 | void __meminit zone_pcp_update(struct zone *zone) |
6390 | { | |
0a647f38 | 6391 | unsigned cpu; |
c8e251fa | 6392 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 6393 | for_each_possible_cpu(cpu) |
169f6c19 CS |
6394 | pageset_set_high_and_batch(zone, |
6395 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 6396 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
6397 | } |
6398 | #endif | |
6399 | ||
340175b7 JL |
6400 | void zone_pcp_reset(struct zone *zone) |
6401 | { | |
6402 | unsigned long flags; | |
5a883813 MK |
6403 | int cpu; |
6404 | struct per_cpu_pageset *pset; | |
340175b7 JL |
6405 | |
6406 | /* avoid races with drain_pages() */ | |
6407 | local_irq_save(flags); | |
6408 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
6409 | for_each_online_cpu(cpu) { |
6410 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6411 | drain_zonestat(zone, pset); | |
6412 | } | |
340175b7 JL |
6413 | free_percpu(zone->pageset); |
6414 | zone->pageset = &boot_pageset; | |
6415 | } | |
6416 | local_irq_restore(flags); | |
6417 | } | |
6418 | ||
6dcd73d7 | 6419 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
6420 | /* |
6421 | * All pages in the range must be isolated before calling this. | |
6422 | */ | |
6423 | void | |
6424 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
6425 | { | |
6426 | struct page *page; | |
6427 | struct zone *zone; | |
6428 | int order, i; | |
6429 | unsigned long pfn; | |
6430 | unsigned long flags; | |
6431 | /* find the first valid pfn */ | |
6432 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
6433 | if (pfn_valid(pfn)) | |
6434 | break; | |
6435 | if (pfn == end_pfn) | |
6436 | return; | |
6437 | zone = page_zone(pfn_to_page(pfn)); | |
6438 | spin_lock_irqsave(&zone->lock, flags); | |
6439 | pfn = start_pfn; | |
6440 | while (pfn < end_pfn) { | |
6441 | if (!pfn_valid(pfn)) { | |
6442 | pfn++; | |
6443 | continue; | |
6444 | } | |
6445 | page = pfn_to_page(pfn); | |
b023f468 WC |
6446 | /* |
6447 | * The HWPoisoned page may be not in buddy system, and | |
6448 | * page_count() is not 0. | |
6449 | */ | |
6450 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6451 | pfn++; | |
6452 | SetPageReserved(page); | |
6453 | continue; | |
6454 | } | |
6455 | ||
0c0e6195 KH |
6456 | BUG_ON(page_count(page)); |
6457 | BUG_ON(!PageBuddy(page)); | |
6458 | order = page_order(page); | |
6459 | #ifdef CONFIG_DEBUG_VM | |
6460 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
6461 | pfn, 1 << order, end_pfn); | |
6462 | #endif | |
6463 | list_del(&page->lru); | |
6464 | rmv_page_order(page); | |
6465 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
6466 | for (i = 0; i < (1 << order); i++) |
6467 | SetPageReserved((page+i)); | |
6468 | pfn += (1 << order); | |
6469 | } | |
6470 | spin_unlock_irqrestore(&zone->lock, flags); | |
6471 | } | |
6472 | #endif | |
8d22ba1b WF |
6473 | |
6474 | #ifdef CONFIG_MEMORY_FAILURE | |
6475 | bool is_free_buddy_page(struct page *page) | |
6476 | { | |
6477 | struct zone *zone = page_zone(page); | |
6478 | unsigned long pfn = page_to_pfn(page); | |
6479 | unsigned long flags; | |
6480 | int order; | |
6481 | ||
6482 | spin_lock_irqsave(&zone->lock, flags); | |
6483 | for (order = 0; order < MAX_ORDER; order++) { | |
6484 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
6485 | ||
6486 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
6487 | break; | |
6488 | } | |
6489 | spin_unlock_irqrestore(&zone->lock, flags); | |
6490 | ||
6491 | return order < MAX_ORDER; | |
6492 | } | |
6493 | #endif | |
718a3821 | 6494 | |
51300cef | 6495 | static const struct trace_print_flags pageflag_names[] = { |
718a3821 WF |
6496 | {1UL << PG_locked, "locked" }, |
6497 | {1UL << PG_error, "error" }, | |
6498 | {1UL << PG_referenced, "referenced" }, | |
6499 | {1UL << PG_uptodate, "uptodate" }, | |
6500 | {1UL << PG_dirty, "dirty" }, | |
6501 | {1UL << PG_lru, "lru" }, | |
6502 | {1UL << PG_active, "active" }, | |
6503 | {1UL << PG_slab, "slab" }, | |
6504 | {1UL << PG_owner_priv_1, "owner_priv_1" }, | |
6505 | {1UL << PG_arch_1, "arch_1" }, | |
6506 | {1UL << PG_reserved, "reserved" }, | |
6507 | {1UL << PG_private, "private" }, | |
6508 | {1UL << PG_private_2, "private_2" }, | |
6509 | {1UL << PG_writeback, "writeback" }, | |
6510 | #ifdef CONFIG_PAGEFLAGS_EXTENDED | |
6511 | {1UL << PG_head, "head" }, | |
6512 | {1UL << PG_tail, "tail" }, | |
6513 | #else | |
6514 | {1UL << PG_compound, "compound" }, | |
6515 | #endif | |
6516 | {1UL << PG_swapcache, "swapcache" }, | |
6517 | {1UL << PG_mappedtodisk, "mappedtodisk" }, | |
6518 | {1UL << PG_reclaim, "reclaim" }, | |
718a3821 WF |
6519 | {1UL << PG_swapbacked, "swapbacked" }, |
6520 | {1UL << PG_unevictable, "unevictable" }, | |
6521 | #ifdef CONFIG_MMU | |
6522 | {1UL << PG_mlocked, "mlocked" }, | |
6523 | #endif | |
6524 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED | |
6525 | {1UL << PG_uncached, "uncached" }, | |
6526 | #endif | |
6527 | #ifdef CONFIG_MEMORY_FAILURE | |
6528 | {1UL << PG_hwpoison, "hwpoison" }, | |
be9cd873 GS |
6529 | #endif |
6530 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6531 | {1UL << PG_compound_lock, "compound_lock" }, | |
718a3821 | 6532 | #endif |
718a3821 WF |
6533 | }; |
6534 | ||
6535 | static void dump_page_flags(unsigned long flags) | |
6536 | { | |
6537 | const char *delim = ""; | |
6538 | unsigned long mask; | |
6539 | int i; | |
6540 | ||
51300cef | 6541 | BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS); |
acc50c11 | 6542 | |
718a3821 WF |
6543 | printk(KERN_ALERT "page flags: %#lx(", flags); |
6544 | ||
6545 | /* remove zone id */ | |
6546 | flags &= (1UL << NR_PAGEFLAGS) - 1; | |
6547 | ||
51300cef | 6548 | for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) { |
718a3821 WF |
6549 | |
6550 | mask = pageflag_names[i].mask; | |
6551 | if ((flags & mask) != mask) | |
6552 | continue; | |
6553 | ||
6554 | flags &= ~mask; | |
6555 | printk("%s%s", delim, pageflag_names[i].name); | |
6556 | delim = "|"; | |
6557 | } | |
6558 | ||
6559 | /* check for left over flags */ | |
6560 | if (flags) | |
6561 | printk("%s%#lx", delim, flags); | |
6562 | ||
6563 | printk(")\n"); | |
6564 | } | |
6565 | ||
d230dec1 KS |
6566 | void dump_page_badflags(struct page *page, const char *reason, |
6567 | unsigned long badflags) | |
718a3821 WF |
6568 | { |
6569 | printk(KERN_ALERT | |
6570 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", | |
4e9f64c4 | 6571 | page, atomic_read(&page->_count), page_mapcount(page), |
718a3821 WF |
6572 | page->mapping, page->index); |
6573 | dump_page_flags(page->flags); | |
f0b791a3 DH |
6574 | if (reason) |
6575 | pr_alert("page dumped because: %s\n", reason); | |
6576 | if (page->flags & badflags) { | |
6577 | pr_alert("bad because of flags:\n"); | |
6578 | dump_page_flags(page->flags & badflags); | |
6579 | } | |
f212ad7c | 6580 | mem_cgroup_print_bad_page(page); |
718a3821 | 6581 | } |
f0b791a3 | 6582 | |
d230dec1 | 6583 | void dump_page(struct page *page, const char *reason) |
f0b791a3 DH |
6584 | { |
6585 | dump_page_badflags(page, reason, 0); | |
6586 | } | |
ed12d845 | 6587 | EXPORT_SYMBOL(dump_page); |