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); | |
7cd2b0a3 | 72 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
c8e251fa | 73 | |
72812019 LS |
74 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
75 | DEFINE_PER_CPU(int, numa_node); | |
76 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
77 | #endif | |
78 | ||
7aac7898 LS |
79 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
80 | /* | |
81 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
82 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
83 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
84 | * defined in <linux/topology.h>. | |
85 | */ | |
86 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
87 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
88 | #endif | |
89 | ||
1da177e4 | 90 | /* |
13808910 | 91 | * Array of node states. |
1da177e4 | 92 | */ |
13808910 CL |
93 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
94 | [N_POSSIBLE] = NODE_MASK_ALL, | |
95 | [N_ONLINE] = { { [0] = 1UL } }, | |
96 | #ifndef CONFIG_NUMA | |
97 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
98 | #ifdef CONFIG_HIGHMEM | |
99 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
100 | #endif |
101 | #ifdef CONFIG_MOVABLE_NODE | |
102 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
103 | #endif |
104 | [N_CPU] = { { [0] = 1UL } }, | |
105 | #endif /* NUMA */ | |
106 | }; | |
107 | EXPORT_SYMBOL(node_states); | |
108 | ||
c3d5f5f0 JL |
109 | /* Protect totalram_pages and zone->managed_pages */ |
110 | static DEFINE_SPINLOCK(managed_page_count_lock); | |
111 | ||
6c231b7b | 112 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 113 | unsigned long totalreserve_pages __read_mostly; |
ab8fabd4 JW |
114 | /* |
115 | * When calculating the number of globally allowed dirty pages, there | |
116 | * is a certain number of per-zone reserves that should not be | |
117 | * considered dirtyable memory. This is the sum of those reserves | |
118 | * over all existing zones that contribute dirtyable memory. | |
119 | */ | |
120 | unsigned long dirty_balance_reserve __read_mostly; | |
121 | ||
1b76b02f | 122 | int percpu_pagelist_fraction; |
dcce284a | 123 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 124 | |
452aa699 RW |
125 | #ifdef CONFIG_PM_SLEEP |
126 | /* | |
127 | * The following functions are used by the suspend/hibernate code to temporarily | |
128 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
129 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
130 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
131 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
132 | * guaranteed not to run in parallel with that modification). | |
133 | */ | |
c9e664f1 RW |
134 | |
135 | static gfp_t saved_gfp_mask; | |
136 | ||
137 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
138 | { |
139 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
140 | if (saved_gfp_mask) { |
141 | gfp_allowed_mask = saved_gfp_mask; | |
142 | saved_gfp_mask = 0; | |
143 | } | |
452aa699 RW |
144 | } |
145 | ||
c9e664f1 | 146 | void pm_restrict_gfp_mask(void) |
452aa699 | 147 | { |
452aa699 | 148 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
149 | WARN_ON(saved_gfp_mask); |
150 | saved_gfp_mask = gfp_allowed_mask; | |
151 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 | 152 | } |
f90ac398 MG |
153 | |
154 | bool pm_suspended_storage(void) | |
155 | { | |
156 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) | |
157 | return false; | |
158 | return true; | |
159 | } | |
452aa699 RW |
160 | #endif /* CONFIG_PM_SLEEP */ |
161 | ||
d9c23400 MG |
162 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
163 | int pageblock_order __read_mostly; | |
164 | #endif | |
165 | ||
d98c7a09 | 166 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 167 | |
1da177e4 LT |
168 | /* |
169 | * results with 256, 32 in the lowmem_reserve sysctl: | |
170 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
171 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
172 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
173 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
174 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
175 | * |
176 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
177 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 178 | */ |
2f1b6248 | 179 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 180 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 181 | 256, |
4b51d669 | 182 | #endif |
fb0e7942 | 183 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 184 | 256, |
fb0e7942 | 185 | #endif |
e53ef38d | 186 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 187 | 32, |
e53ef38d | 188 | #endif |
2a1e274a | 189 | 32, |
2f1b6248 | 190 | }; |
1da177e4 LT |
191 | |
192 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 193 | |
15ad7cdc | 194 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 195 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 196 | "DMA", |
4b51d669 | 197 | #endif |
fb0e7942 | 198 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 199 | "DMA32", |
fb0e7942 | 200 | #endif |
2f1b6248 | 201 | "Normal", |
e53ef38d | 202 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 203 | "HighMem", |
e53ef38d | 204 | #endif |
2a1e274a | 205 | "Movable", |
2f1b6248 CL |
206 | }; |
207 | ||
1da177e4 | 208 | int min_free_kbytes = 1024; |
42aa83cb | 209 | int user_min_free_kbytes = -1; |
1da177e4 | 210 | |
2c85f51d JB |
211 | static unsigned long __meminitdata nr_kernel_pages; |
212 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 213 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 214 | |
0ee332c1 TH |
215 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
216 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
217 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
218 | static unsigned long __initdata required_kernelcore; | |
219 | static unsigned long __initdata required_movablecore; | |
220 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
221 | ||
222 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
223 | int movable_zone; | |
224 | EXPORT_SYMBOL(movable_zone); | |
225 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 226 | |
418508c1 MS |
227 | #if MAX_NUMNODES > 1 |
228 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 229 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 230 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 231 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
232 | #endif |
233 | ||
9ef9acb0 MG |
234 | int page_group_by_mobility_disabled __read_mostly; |
235 | ||
ee6f509c | 236 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 237 | { |
5d0f3f72 KM |
238 | if (unlikely(page_group_by_mobility_disabled && |
239 | migratetype < MIGRATE_PCPTYPES)) | |
49255c61 MG |
240 | migratetype = MIGRATE_UNMOVABLE; |
241 | ||
b2a0ac88 MG |
242 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
243 | PB_migrate, PB_migrate_end); | |
244 | } | |
245 | ||
7f33d49a RW |
246 | bool oom_killer_disabled __read_mostly; |
247 | ||
13e7444b | 248 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 249 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 250 | { |
bdc8cb98 DH |
251 | int ret = 0; |
252 | unsigned seq; | |
253 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 254 | unsigned long sp, start_pfn; |
c6a57e19 | 255 | |
bdc8cb98 DH |
256 | do { |
257 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
258 | start_pfn = zone->zone_start_pfn; |
259 | sp = zone->spanned_pages; | |
108bcc96 | 260 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
261 | ret = 1; |
262 | } while (zone_span_seqretry(zone, seq)); | |
263 | ||
b5e6a5a2 | 264 | if (ret) |
613813e8 DH |
265 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
266 | pfn, zone_to_nid(zone), zone->name, | |
267 | start_pfn, start_pfn + sp); | |
b5e6a5a2 | 268 | |
bdc8cb98 | 269 | return ret; |
c6a57e19 DH |
270 | } |
271 | ||
272 | static int page_is_consistent(struct zone *zone, struct page *page) | |
273 | { | |
14e07298 | 274 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 275 | return 0; |
1da177e4 | 276 | if (zone != page_zone(page)) |
c6a57e19 DH |
277 | return 0; |
278 | ||
279 | return 1; | |
280 | } | |
281 | /* | |
282 | * Temporary debugging check for pages not lying within a given zone. | |
283 | */ | |
284 | static int bad_range(struct zone *zone, struct page *page) | |
285 | { | |
286 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 287 | return 1; |
c6a57e19 DH |
288 | if (!page_is_consistent(zone, page)) |
289 | return 1; | |
290 | ||
1da177e4 LT |
291 | return 0; |
292 | } | |
13e7444b NP |
293 | #else |
294 | static inline int bad_range(struct zone *zone, struct page *page) | |
295 | { | |
296 | return 0; | |
297 | } | |
298 | #endif | |
299 | ||
d230dec1 KS |
300 | static void bad_page(struct page *page, const char *reason, |
301 | unsigned long bad_flags) | |
1da177e4 | 302 | { |
d936cf9b HD |
303 | static unsigned long resume; |
304 | static unsigned long nr_shown; | |
305 | static unsigned long nr_unshown; | |
306 | ||
2a7684a2 WF |
307 | /* Don't complain about poisoned pages */ |
308 | if (PageHWPoison(page)) { | |
22b751c3 | 309 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
310 | return; |
311 | } | |
312 | ||
d936cf9b HD |
313 | /* |
314 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
315 | * or allow a steady drip of one report per second. | |
316 | */ | |
317 | if (nr_shown == 60) { | |
318 | if (time_before(jiffies, resume)) { | |
319 | nr_unshown++; | |
320 | goto out; | |
321 | } | |
322 | if (nr_unshown) { | |
1e9e6365 HD |
323 | printk(KERN_ALERT |
324 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
325 | nr_unshown); |
326 | nr_unshown = 0; | |
327 | } | |
328 | nr_shown = 0; | |
329 | } | |
330 | if (nr_shown++ == 0) | |
331 | resume = jiffies + 60 * HZ; | |
332 | ||
1e9e6365 | 333 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 334 | current->comm, page_to_pfn(page)); |
f0b791a3 | 335 | dump_page_badflags(page, reason, bad_flags); |
3dc14741 | 336 | |
4f31888c | 337 | print_modules(); |
1da177e4 | 338 | dump_stack(); |
d936cf9b | 339 | out: |
8cc3b392 | 340 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 341 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 342 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
343 | } |
344 | ||
1da177e4 LT |
345 | /* |
346 | * Higher-order pages are called "compound pages". They are structured thusly: | |
347 | * | |
348 | * The first PAGE_SIZE page is called the "head page". | |
349 | * | |
350 | * The remaining PAGE_SIZE pages are called "tail pages". | |
351 | * | |
6416b9fa WSH |
352 | * All pages have PG_compound set. All tail pages have their ->first_page |
353 | * pointing at the head page. | |
1da177e4 | 354 | * |
41d78ba5 HD |
355 | * The first tail page's ->lru.next holds the address of the compound page's |
356 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
357 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 358 | */ |
d98c7a09 HD |
359 | |
360 | static void free_compound_page(struct page *page) | |
361 | { | |
d85f3385 | 362 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
363 | } |
364 | ||
01ad1c08 | 365 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
366 | { |
367 | int i; | |
368 | int nr_pages = 1 << order; | |
369 | ||
370 | set_compound_page_dtor(page, free_compound_page); | |
371 | set_compound_order(page, order); | |
372 | __SetPageHead(page); | |
373 | for (i = 1; i < nr_pages; i++) { | |
374 | struct page *p = page + i; | |
58a84aa9 | 375 | set_page_count(p, 0); |
18229df5 | 376 | p->first_page = page; |
668f9abb DR |
377 | /* Make sure p->first_page is always valid for PageTail() */ |
378 | smp_wmb(); | |
379 | __SetPageTail(p); | |
18229df5 AW |
380 | } |
381 | } | |
382 | ||
59ff4216 | 383 | /* update __split_huge_page_refcount if you change this function */ |
8cc3b392 | 384 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
385 | { |
386 | int i; | |
387 | int nr_pages = 1 << order; | |
8cc3b392 | 388 | int bad = 0; |
1da177e4 | 389 | |
0bb2c763 | 390 | if (unlikely(compound_order(page) != order)) { |
f0b791a3 | 391 | bad_page(page, "wrong compound order", 0); |
8cc3b392 HD |
392 | bad++; |
393 | } | |
1da177e4 | 394 | |
6d777953 | 395 | __ClearPageHead(page); |
8cc3b392 | 396 | |
18229df5 AW |
397 | for (i = 1; i < nr_pages; i++) { |
398 | struct page *p = page + i; | |
1da177e4 | 399 | |
f0b791a3 DH |
400 | if (unlikely(!PageTail(p))) { |
401 | bad_page(page, "PageTail not set", 0); | |
402 | bad++; | |
403 | } else if (unlikely(p->first_page != page)) { | |
404 | bad_page(page, "first_page not consistent", 0); | |
8cc3b392 HD |
405 | bad++; |
406 | } | |
d85f3385 | 407 | __ClearPageTail(p); |
1da177e4 | 408 | } |
8cc3b392 HD |
409 | |
410 | return bad; | |
1da177e4 | 411 | } |
1da177e4 | 412 | |
7aeb09f9 MG |
413 | static inline void prep_zero_page(struct page *page, unsigned int order, |
414 | gfp_t gfp_flags) | |
17cf4406 NP |
415 | { |
416 | int i; | |
417 | ||
6626c5d5 AM |
418 | /* |
419 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
420 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
421 | */ | |
725d704e | 422 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
423 | for (i = 0; i < (1 << order); i++) |
424 | clear_highpage(page + i); | |
425 | } | |
426 | ||
c0a32fc5 SG |
427 | #ifdef CONFIG_DEBUG_PAGEALLOC |
428 | unsigned int _debug_guardpage_minorder; | |
429 | ||
430 | static int __init debug_guardpage_minorder_setup(char *buf) | |
431 | { | |
432 | unsigned long res; | |
433 | ||
434 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
435 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
436 | return 0; | |
437 | } | |
438 | _debug_guardpage_minorder = res; | |
439 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
440 | return 0; | |
441 | } | |
442 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
443 | ||
444 | static inline void set_page_guard_flag(struct page *page) | |
445 | { | |
446 | __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
447 | } | |
448 | ||
449 | static inline void clear_page_guard_flag(struct page *page) | |
450 | { | |
451 | __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
452 | } | |
453 | #else | |
454 | static inline void set_page_guard_flag(struct page *page) { } | |
455 | static inline void clear_page_guard_flag(struct page *page) { } | |
456 | #endif | |
457 | ||
7aeb09f9 | 458 | static inline void set_page_order(struct page *page, unsigned int order) |
6aa3001b | 459 | { |
4c21e2f2 | 460 | set_page_private(page, order); |
676165a8 | 461 | __SetPageBuddy(page); |
1da177e4 LT |
462 | } |
463 | ||
464 | static inline void rmv_page_order(struct page *page) | |
465 | { | |
676165a8 | 466 | __ClearPageBuddy(page); |
4c21e2f2 | 467 | set_page_private(page, 0); |
1da177e4 LT |
468 | } |
469 | ||
470 | /* | |
471 | * Locate the struct page for both the matching buddy in our | |
472 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
473 | * | |
474 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
475 | * the following equation: | |
476 | * B2 = B1 ^ (1 << O) | |
477 | * For example, if the starting buddy (buddy2) is #8 its order | |
478 | * 1 buddy is #10: | |
479 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
480 | * | |
481 | * 2) Any buddy B will have an order O+1 parent P which | |
482 | * satisfies the following equation: | |
483 | * P = B & ~(1 << O) | |
484 | * | |
d6e05edc | 485 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 | 486 | */ |
1da177e4 | 487 | static inline unsigned long |
43506fad | 488 | __find_buddy_index(unsigned long page_idx, unsigned int order) |
1da177e4 | 489 | { |
43506fad | 490 | return page_idx ^ (1 << order); |
1da177e4 LT |
491 | } |
492 | ||
493 | /* | |
494 | * This function checks whether a page is free && is the buddy | |
495 | * we can do coalesce a page and its buddy if | |
13e7444b | 496 | * (a) the buddy is not in a hole && |
676165a8 | 497 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
498 | * (c) a page and its buddy have the same order && |
499 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 500 | * |
cf6fe945 WSH |
501 | * For recording whether a page is in the buddy system, we set ->_mapcount |
502 | * PAGE_BUDDY_MAPCOUNT_VALUE. | |
503 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is | |
504 | * serialized by zone->lock. | |
1da177e4 | 505 | * |
676165a8 | 506 | * For recording page's order, we use page_private(page). |
1da177e4 | 507 | */ |
cb2b95e1 | 508 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
7aeb09f9 | 509 | unsigned int order) |
1da177e4 | 510 | { |
14e07298 | 511 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 512 | return 0; |
13e7444b | 513 | |
c0a32fc5 | 514 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
309381fe | 515 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
d34c5fa0 MG |
516 | |
517 | if (page_zone_id(page) != page_zone_id(buddy)) | |
518 | return 0; | |
519 | ||
c0a32fc5 SG |
520 | return 1; |
521 | } | |
522 | ||
cb2b95e1 | 523 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
309381fe | 524 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
d34c5fa0 MG |
525 | |
526 | /* | |
527 | * zone check is done late to avoid uselessly | |
528 | * calculating zone/node ids for pages that could | |
529 | * never merge. | |
530 | */ | |
531 | if (page_zone_id(page) != page_zone_id(buddy)) | |
532 | return 0; | |
533 | ||
6aa3001b | 534 | return 1; |
676165a8 | 535 | } |
6aa3001b | 536 | return 0; |
1da177e4 LT |
537 | } |
538 | ||
539 | /* | |
540 | * Freeing function for a buddy system allocator. | |
541 | * | |
542 | * The concept of a buddy system is to maintain direct-mapped table | |
543 | * (containing bit values) for memory blocks of various "orders". | |
544 | * The bottom level table contains the map for the smallest allocatable | |
545 | * units of memory (here, pages), and each level above it describes | |
546 | * pairs of units from the levels below, hence, "buddies". | |
547 | * At a high level, all that happens here is marking the table entry | |
548 | * at the bottom level available, and propagating the changes upward | |
549 | * as necessary, plus some accounting needed to play nicely with other | |
550 | * parts of the VM system. | |
551 | * At each level, we keep a list of pages, which are heads of continuous | |
cf6fe945 WSH |
552 | * free pages of length of (1 << order) and marked with _mapcount |
553 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) | |
554 | * field. | |
1da177e4 | 555 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
556 | * other. That is, if we allocate a small block, and both were |
557 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 558 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 559 | * triggers coalescing into a block of larger size. |
1da177e4 | 560 | * |
6d49e352 | 561 | * -- nyc |
1da177e4 LT |
562 | */ |
563 | ||
48db57f8 | 564 | static inline void __free_one_page(struct page *page, |
dc4b0caf | 565 | unsigned long pfn, |
ed0ae21d MG |
566 | struct zone *zone, unsigned int order, |
567 | int migratetype) | |
1da177e4 LT |
568 | { |
569 | unsigned long page_idx; | |
6dda9d55 | 570 | unsigned long combined_idx; |
43506fad | 571 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 572 | struct page *buddy; |
1da177e4 | 573 | |
d29bb978 CS |
574 | VM_BUG_ON(!zone_is_initialized(zone)); |
575 | ||
224abf92 | 576 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
577 | if (unlikely(destroy_compound_page(page, order))) |
578 | return; | |
1da177e4 | 579 | |
ed0ae21d MG |
580 | VM_BUG_ON(migratetype == -1); |
581 | ||
dc4b0caf | 582 | page_idx = pfn & ((1 << MAX_ORDER) - 1); |
1da177e4 | 583 | |
309381fe SL |
584 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
585 | VM_BUG_ON_PAGE(bad_range(zone, page), page); | |
1da177e4 | 586 | |
1da177e4 | 587 | while (order < MAX_ORDER-1) { |
43506fad KC |
588 | buddy_idx = __find_buddy_index(page_idx, order); |
589 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 590 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 591 | break; |
c0a32fc5 SG |
592 | /* |
593 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
594 | * merge with it and move up one order. | |
595 | */ | |
596 | if (page_is_guard(buddy)) { | |
597 | clear_page_guard_flag(buddy); | |
598 | set_page_private(page, 0); | |
d1ce749a BZ |
599 | __mod_zone_freepage_state(zone, 1 << order, |
600 | migratetype); | |
c0a32fc5 SG |
601 | } else { |
602 | list_del(&buddy->lru); | |
603 | zone->free_area[order].nr_free--; | |
604 | rmv_page_order(buddy); | |
605 | } | |
43506fad | 606 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
607 | page = page + (combined_idx - page_idx); |
608 | page_idx = combined_idx; | |
609 | order++; | |
610 | } | |
611 | set_page_order(page, order); | |
6dda9d55 CZ |
612 | |
613 | /* | |
614 | * If this is not the largest possible page, check if the buddy | |
615 | * of the next-highest order is free. If it is, it's possible | |
616 | * that pages are being freed that will coalesce soon. In case, | |
617 | * that is happening, add the free page to the tail of the list | |
618 | * so it's less likely to be used soon and more likely to be merged | |
619 | * as a higher order page | |
620 | */ | |
b7f50cfa | 621 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 622 | struct page *higher_page, *higher_buddy; |
43506fad KC |
623 | combined_idx = buddy_idx & page_idx; |
624 | higher_page = page + (combined_idx - page_idx); | |
625 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 626 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
627 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
628 | list_add_tail(&page->lru, | |
629 | &zone->free_area[order].free_list[migratetype]); | |
630 | goto out; | |
631 | } | |
632 | } | |
633 | ||
634 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
635 | out: | |
1da177e4 LT |
636 | zone->free_area[order].nr_free++; |
637 | } | |
638 | ||
224abf92 | 639 | static inline int free_pages_check(struct page *page) |
1da177e4 | 640 | { |
d230dec1 | 641 | const char *bad_reason = NULL; |
f0b791a3 DH |
642 | unsigned long bad_flags = 0; |
643 | ||
644 | if (unlikely(page_mapcount(page))) | |
645 | bad_reason = "nonzero mapcount"; | |
646 | if (unlikely(page->mapping != NULL)) | |
647 | bad_reason = "non-NULL mapping"; | |
648 | if (unlikely(atomic_read(&page->_count) != 0)) | |
649 | bad_reason = "nonzero _count"; | |
650 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { | |
651 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; | |
652 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; | |
653 | } | |
654 | if (unlikely(mem_cgroup_bad_page_check(page))) | |
655 | bad_reason = "cgroup check failed"; | |
656 | if (unlikely(bad_reason)) { | |
657 | bad_page(page, bad_reason, bad_flags); | |
79f4b7bf | 658 | return 1; |
8cc3b392 | 659 | } |
90572890 | 660 | page_cpupid_reset_last(page); |
79f4b7bf HD |
661 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
662 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
663 | return 0; | |
1da177e4 LT |
664 | } |
665 | ||
666 | /* | |
5f8dcc21 | 667 | * Frees a number of pages from the PCP lists |
1da177e4 | 668 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 669 | * count is the number of pages to free. |
1da177e4 LT |
670 | * |
671 | * If the zone was previously in an "all pages pinned" state then look to | |
672 | * see if this freeing clears that state. | |
673 | * | |
674 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
675 | * pinned" detection logic. | |
676 | */ | |
5f8dcc21 MG |
677 | static void free_pcppages_bulk(struct zone *zone, int count, |
678 | struct per_cpu_pages *pcp) | |
1da177e4 | 679 | { |
5f8dcc21 | 680 | int migratetype = 0; |
a6f9edd6 | 681 | int batch_free = 0; |
72853e29 | 682 | int to_free = count; |
5f8dcc21 | 683 | |
c54ad30c | 684 | spin_lock(&zone->lock); |
1da177e4 | 685 | zone->pages_scanned = 0; |
f2260e6b | 686 | |
72853e29 | 687 | while (to_free) { |
48db57f8 | 688 | struct page *page; |
5f8dcc21 MG |
689 | struct list_head *list; |
690 | ||
691 | /* | |
a6f9edd6 MG |
692 | * Remove pages from lists in a round-robin fashion. A |
693 | * batch_free count is maintained that is incremented when an | |
694 | * empty list is encountered. This is so more pages are freed | |
695 | * off fuller lists instead of spinning excessively around empty | |
696 | * lists | |
5f8dcc21 MG |
697 | */ |
698 | do { | |
a6f9edd6 | 699 | batch_free++; |
5f8dcc21 MG |
700 | if (++migratetype == MIGRATE_PCPTYPES) |
701 | migratetype = 0; | |
702 | list = &pcp->lists[migratetype]; | |
703 | } while (list_empty(list)); | |
48db57f8 | 704 | |
1d16871d NK |
705 | /* This is the only non-empty list. Free them all. */ |
706 | if (batch_free == MIGRATE_PCPTYPES) | |
707 | batch_free = to_free; | |
708 | ||
a6f9edd6 | 709 | do { |
770c8aaa BZ |
710 | int mt; /* migratetype of the to-be-freed page */ |
711 | ||
a6f9edd6 MG |
712 | page = list_entry(list->prev, struct page, lru); |
713 | /* must delete as __free_one_page list manipulates */ | |
714 | list_del(&page->lru); | |
b12c4ad1 | 715 | mt = get_freepage_migratetype(page); |
a7016235 | 716 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
dc4b0caf | 717 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
770c8aaa | 718 | trace_mm_page_pcpu_drain(page, 0, mt); |
194159fb | 719 | if (likely(!is_migrate_isolate_page(page))) { |
97d0da22 WC |
720 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1); |
721 | if (is_migrate_cma(mt)) | |
722 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1); | |
723 | } | |
72853e29 | 724 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 725 | } |
c54ad30c | 726 | spin_unlock(&zone->lock); |
1da177e4 LT |
727 | } |
728 | ||
dc4b0caf MG |
729 | static void free_one_page(struct zone *zone, |
730 | struct page *page, unsigned long pfn, | |
7aeb09f9 | 731 | unsigned int order, |
ed0ae21d | 732 | int migratetype) |
1da177e4 | 733 | { |
006d22d9 | 734 | spin_lock(&zone->lock); |
006d22d9 | 735 | zone->pages_scanned = 0; |
f2260e6b | 736 | |
dc4b0caf | 737 | __free_one_page(page, pfn, zone, order, migratetype); |
194159fb | 738 | if (unlikely(!is_migrate_isolate(migratetype))) |
d1ce749a | 739 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
006d22d9 | 740 | spin_unlock(&zone->lock); |
48db57f8 NP |
741 | } |
742 | ||
ec95f53a | 743 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 744 | { |
1da177e4 | 745 | int i; |
8cc3b392 | 746 | int bad = 0; |
1da177e4 | 747 | |
b413d48a | 748 | trace_mm_page_free(page, order); |
b1eeab67 VN |
749 | kmemcheck_free_shadow(page, order); |
750 | ||
8dd60a3a AA |
751 | if (PageAnon(page)) |
752 | page->mapping = NULL; | |
753 | for (i = 0; i < (1 << order); i++) | |
754 | bad += free_pages_check(page + i); | |
8cc3b392 | 755 | if (bad) |
ec95f53a | 756 | return false; |
689bcebf | 757 | |
3ac7fe5a | 758 | if (!PageHighMem(page)) { |
b8af2941 PK |
759 | debug_check_no_locks_freed(page_address(page), |
760 | PAGE_SIZE << order); | |
3ac7fe5a TG |
761 | debug_check_no_obj_freed(page_address(page), |
762 | PAGE_SIZE << order); | |
763 | } | |
dafb1367 | 764 | arch_free_page(page, order); |
48db57f8 | 765 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 766 | |
ec95f53a KM |
767 | return true; |
768 | } | |
769 | ||
770 | static void __free_pages_ok(struct page *page, unsigned int order) | |
771 | { | |
772 | unsigned long flags; | |
95e34412 | 773 | int migratetype; |
dc4b0caf | 774 | unsigned long pfn = page_to_pfn(page); |
ec95f53a KM |
775 | |
776 | if (!free_pages_prepare(page, order)) | |
777 | return; | |
778 | ||
cfc47a28 | 779 | migratetype = get_pfnblock_migratetype(page, pfn); |
c54ad30c | 780 | local_irq_save(flags); |
f8891e5e | 781 | __count_vm_events(PGFREE, 1 << order); |
95e34412 | 782 | set_freepage_migratetype(page, migratetype); |
dc4b0caf | 783 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
c54ad30c | 784 | local_irq_restore(flags); |
1da177e4 LT |
785 | } |
786 | ||
170a5a7e | 787 | void __init __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 | 788 | { |
c3993076 | 789 | unsigned int nr_pages = 1 << order; |
e2d0bd2b | 790 | struct page *p = page; |
c3993076 | 791 | unsigned int loop; |
a226f6c8 | 792 | |
e2d0bd2b YL |
793 | prefetchw(p); |
794 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { | |
795 | prefetchw(p + 1); | |
c3993076 JW |
796 | __ClearPageReserved(p); |
797 | set_page_count(p, 0); | |
a226f6c8 | 798 | } |
e2d0bd2b YL |
799 | __ClearPageReserved(p); |
800 | set_page_count(p, 0); | |
c3993076 | 801 | |
e2d0bd2b | 802 | page_zone(page)->managed_pages += nr_pages; |
c3993076 JW |
803 | set_page_refcounted(page); |
804 | __free_pages(page, order); | |
a226f6c8 DH |
805 | } |
806 | ||
47118af0 | 807 | #ifdef CONFIG_CMA |
9cf510a5 | 808 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
47118af0 MN |
809 | void __init init_cma_reserved_pageblock(struct page *page) |
810 | { | |
811 | unsigned i = pageblock_nr_pages; | |
812 | struct page *p = page; | |
813 | ||
814 | do { | |
815 | __ClearPageReserved(p); | |
816 | set_page_count(p, 0); | |
817 | } while (++p, --i); | |
818 | ||
47118af0 | 819 | set_pageblock_migratetype(page, MIGRATE_CMA); |
dc78327c MN |
820 | |
821 | if (pageblock_order >= MAX_ORDER) { | |
822 | i = pageblock_nr_pages; | |
823 | p = page; | |
824 | do { | |
825 | set_page_refcounted(p); | |
826 | __free_pages(p, MAX_ORDER - 1); | |
827 | p += MAX_ORDER_NR_PAGES; | |
828 | } while (i -= MAX_ORDER_NR_PAGES); | |
829 | } else { | |
830 | set_page_refcounted(page); | |
831 | __free_pages(page, pageblock_order); | |
832 | } | |
833 | ||
3dcc0571 | 834 | adjust_managed_page_count(page, pageblock_nr_pages); |
47118af0 MN |
835 | } |
836 | #endif | |
1da177e4 LT |
837 | |
838 | /* | |
839 | * The order of subdivision here is critical for the IO subsystem. | |
840 | * Please do not alter this order without good reasons and regression | |
841 | * testing. Specifically, as large blocks of memory are subdivided, | |
842 | * the order in which smaller blocks are delivered depends on the order | |
843 | * they're subdivided in this function. This is the primary factor | |
844 | * influencing the order in which pages are delivered to the IO | |
845 | * subsystem according to empirical testing, and this is also justified | |
846 | * by considering the behavior of a buddy system containing a single | |
847 | * large block of memory acted on by a series of small allocations. | |
848 | * This behavior is a critical factor in sglist merging's success. | |
849 | * | |
6d49e352 | 850 | * -- nyc |
1da177e4 | 851 | */ |
085cc7d5 | 852 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
853 | int low, int high, struct free_area *area, |
854 | int migratetype) | |
1da177e4 LT |
855 | { |
856 | unsigned long size = 1 << high; | |
857 | ||
858 | while (high > low) { | |
859 | area--; | |
860 | high--; | |
861 | size >>= 1; | |
309381fe | 862 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
c0a32fc5 SG |
863 | |
864 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
865 | if (high < debug_guardpage_minorder()) { | |
866 | /* | |
867 | * Mark as guard pages (or page), that will allow to | |
868 | * merge back to allocator when buddy will be freed. | |
869 | * Corresponding page table entries will not be touched, | |
870 | * pages will stay not present in virtual address space | |
871 | */ | |
872 | INIT_LIST_HEAD(&page[size].lru); | |
873 | set_page_guard_flag(&page[size]); | |
874 | set_page_private(&page[size], high); | |
875 | /* Guard pages are not available for any usage */ | |
d1ce749a BZ |
876 | __mod_zone_freepage_state(zone, -(1 << high), |
877 | migratetype); | |
c0a32fc5 SG |
878 | continue; |
879 | } | |
880 | #endif | |
b2a0ac88 | 881 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
882 | area->nr_free++; |
883 | set_page_order(&page[size], high); | |
884 | } | |
1da177e4 LT |
885 | } |
886 | ||
1da177e4 LT |
887 | /* |
888 | * This page is about to be returned from the page allocator | |
889 | */ | |
2a7684a2 | 890 | static inline int check_new_page(struct page *page) |
1da177e4 | 891 | { |
d230dec1 | 892 | const char *bad_reason = NULL; |
f0b791a3 DH |
893 | unsigned long bad_flags = 0; |
894 | ||
895 | if (unlikely(page_mapcount(page))) | |
896 | bad_reason = "nonzero mapcount"; | |
897 | if (unlikely(page->mapping != NULL)) | |
898 | bad_reason = "non-NULL mapping"; | |
899 | if (unlikely(atomic_read(&page->_count) != 0)) | |
900 | bad_reason = "nonzero _count"; | |
901 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { | |
902 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; | |
903 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; | |
904 | } | |
905 | if (unlikely(mem_cgroup_bad_page_check(page))) | |
906 | bad_reason = "cgroup check failed"; | |
907 | if (unlikely(bad_reason)) { | |
908 | bad_page(page, bad_reason, bad_flags); | |
689bcebf | 909 | return 1; |
8cc3b392 | 910 | } |
2a7684a2 WF |
911 | return 0; |
912 | } | |
913 | ||
7aeb09f9 | 914 | static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags) |
2a7684a2 WF |
915 | { |
916 | int i; | |
917 | ||
918 | for (i = 0; i < (1 << order); i++) { | |
919 | struct page *p = page + i; | |
920 | if (unlikely(check_new_page(p))) | |
921 | return 1; | |
922 | } | |
689bcebf | 923 | |
4c21e2f2 | 924 | set_page_private(page, 0); |
7835e98b | 925 | set_page_refcounted(page); |
cc102509 NP |
926 | |
927 | arch_alloc_page(page, order); | |
1da177e4 | 928 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
929 | |
930 | if (gfp_flags & __GFP_ZERO) | |
931 | prep_zero_page(page, order, gfp_flags); | |
932 | ||
933 | if (order && (gfp_flags & __GFP_COMP)) | |
934 | prep_compound_page(page, order); | |
935 | ||
689bcebf | 936 | return 0; |
1da177e4 LT |
937 | } |
938 | ||
56fd56b8 MG |
939 | /* |
940 | * Go through the free lists for the given migratetype and remove | |
941 | * the smallest available page from the freelists | |
942 | */ | |
728ec980 MG |
943 | static inline |
944 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
945 | int migratetype) |
946 | { | |
947 | unsigned int current_order; | |
b8af2941 | 948 | struct free_area *area; |
56fd56b8 MG |
949 | struct page *page; |
950 | ||
951 | /* Find a page of the appropriate size in the preferred list */ | |
952 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
953 | area = &(zone->free_area[current_order]); | |
954 | if (list_empty(&area->free_list[migratetype])) | |
955 | continue; | |
956 | ||
957 | page = list_entry(area->free_list[migratetype].next, | |
958 | struct page, lru); | |
959 | list_del(&page->lru); | |
960 | rmv_page_order(page); | |
961 | area->nr_free--; | |
56fd56b8 | 962 | expand(zone, page, order, current_order, area, migratetype); |
5bcc9f86 | 963 | set_freepage_migratetype(page, migratetype); |
56fd56b8 MG |
964 | return page; |
965 | } | |
966 | ||
967 | return NULL; | |
968 | } | |
969 | ||
970 | ||
b2a0ac88 MG |
971 | /* |
972 | * This array describes the order lists are fallen back to when | |
973 | * the free lists for the desirable migrate type are depleted | |
974 | */ | |
47118af0 MN |
975 | static int fallbacks[MIGRATE_TYPES][4] = { |
976 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
977 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
978 | #ifdef CONFIG_CMA | |
979 | [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
980 | [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */ | |
981 | #else | |
982 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
983 | #endif | |
6d4a4916 | 984 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 985 | #ifdef CONFIG_MEMORY_ISOLATION |
6d4a4916 | 986 | [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 987 | #endif |
b2a0ac88 MG |
988 | }; |
989 | ||
c361be55 MG |
990 | /* |
991 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 992 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
993 | * boundary. If alignment is required, use move_freepages_block() |
994 | */ | |
435b405c | 995 | int move_freepages(struct zone *zone, |
b69a7288 AB |
996 | struct page *start_page, struct page *end_page, |
997 | int migratetype) | |
c361be55 MG |
998 | { |
999 | struct page *page; | |
1000 | unsigned long order; | |
d100313f | 1001 | int pages_moved = 0; |
c361be55 MG |
1002 | |
1003 | #ifndef CONFIG_HOLES_IN_ZONE | |
1004 | /* | |
1005 | * page_zone is not safe to call in this context when | |
1006 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
1007 | * anyway as we check zone boundaries in move_freepages_block(). | |
1008 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 1009 | * grouping pages by mobility |
c361be55 MG |
1010 | */ |
1011 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
1012 | #endif | |
1013 | ||
1014 | for (page = start_page; page <= end_page;) { | |
344c790e | 1015 | /* Make sure we are not inadvertently changing nodes */ |
309381fe | 1016 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
344c790e | 1017 | |
c361be55 MG |
1018 | if (!pfn_valid_within(page_to_pfn(page))) { |
1019 | page++; | |
1020 | continue; | |
1021 | } | |
1022 | ||
1023 | if (!PageBuddy(page)) { | |
1024 | page++; | |
1025 | continue; | |
1026 | } | |
1027 | ||
1028 | order = page_order(page); | |
84be48d8 KS |
1029 | list_move(&page->lru, |
1030 | &zone->free_area[order].free_list[migratetype]); | |
95e34412 | 1031 | set_freepage_migratetype(page, migratetype); |
c361be55 | 1032 | page += 1 << order; |
d100313f | 1033 | pages_moved += 1 << order; |
c361be55 MG |
1034 | } |
1035 | ||
d100313f | 1036 | return pages_moved; |
c361be55 MG |
1037 | } |
1038 | ||
ee6f509c | 1039 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 1040 | int migratetype) |
c361be55 MG |
1041 | { |
1042 | unsigned long start_pfn, end_pfn; | |
1043 | struct page *start_page, *end_page; | |
1044 | ||
1045 | start_pfn = page_to_pfn(page); | |
d9c23400 | 1046 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 1047 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
1048 | end_page = start_page + pageblock_nr_pages - 1; |
1049 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
1050 | |
1051 | /* Do not cross zone boundaries */ | |
108bcc96 | 1052 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1053 | start_page = page; |
108bcc96 | 1054 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1055 | return 0; |
1056 | ||
1057 | return move_freepages(zone, start_page, end_page, migratetype); | |
1058 | } | |
1059 | ||
2f66a68f MG |
1060 | static void change_pageblock_range(struct page *pageblock_page, |
1061 | int start_order, int migratetype) | |
1062 | { | |
1063 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1064 | ||
1065 | while (nr_pageblocks--) { | |
1066 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1067 | pageblock_page += pageblock_nr_pages; | |
1068 | } | |
1069 | } | |
1070 | ||
fef903ef SB |
1071 | /* |
1072 | * If breaking a large block of pages, move all free pages to the preferred | |
1073 | * allocation list. If falling back for a reclaimable kernel allocation, be | |
1074 | * more aggressive about taking ownership of free pages. | |
1075 | * | |
1076 | * On the other hand, never change migration type of MIGRATE_CMA pageblocks | |
1077 | * nor move CMA pages to different free lists. We don't want unmovable pages | |
1078 | * to be allocated from MIGRATE_CMA areas. | |
1079 | * | |
1080 | * Returns the new migratetype of the pageblock (or the same old migratetype | |
1081 | * if it was unchanged). | |
1082 | */ | |
1083 | static int try_to_steal_freepages(struct zone *zone, struct page *page, | |
1084 | int start_type, int fallback_type) | |
1085 | { | |
1086 | int current_order = page_order(page); | |
1087 | ||
0cbef29a KM |
1088 | /* |
1089 | * When borrowing from MIGRATE_CMA, we need to release the excess | |
5bcc9f86 VB |
1090 | * buddy pages to CMA itself. We also ensure the freepage_migratetype |
1091 | * is set to CMA so it is returned to the correct freelist in case | |
1092 | * the page ends up being not actually allocated from the pcp lists. | |
0cbef29a | 1093 | */ |
fef903ef SB |
1094 | if (is_migrate_cma(fallback_type)) |
1095 | return fallback_type; | |
1096 | ||
1097 | /* Take ownership for orders >= pageblock_order */ | |
1098 | if (current_order >= pageblock_order) { | |
1099 | change_pageblock_range(page, current_order, start_type); | |
1100 | return start_type; | |
1101 | } | |
1102 | ||
1103 | if (current_order >= pageblock_order / 2 || | |
1104 | start_type == MIGRATE_RECLAIMABLE || | |
1105 | page_group_by_mobility_disabled) { | |
1106 | int pages; | |
1107 | ||
1108 | pages = move_freepages_block(zone, page, start_type); | |
1109 | ||
1110 | /* Claim the whole block if over half of it is free */ | |
1111 | if (pages >= (1 << (pageblock_order-1)) || | |
1112 | page_group_by_mobility_disabled) { | |
1113 | ||
1114 | set_pageblock_migratetype(page, start_type); | |
1115 | return start_type; | |
1116 | } | |
1117 | ||
1118 | } | |
1119 | ||
1120 | return fallback_type; | |
1121 | } | |
1122 | ||
b2a0ac88 | 1123 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 | 1124 | static inline struct page * |
7aeb09f9 | 1125 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
b2a0ac88 | 1126 | { |
b8af2941 | 1127 | struct free_area *area; |
7aeb09f9 | 1128 | unsigned int current_order; |
b2a0ac88 | 1129 | struct page *page; |
fef903ef | 1130 | int migratetype, new_type, i; |
b2a0ac88 MG |
1131 | |
1132 | /* Find the largest possible block of pages in the other list */ | |
7aeb09f9 MG |
1133 | for (current_order = MAX_ORDER-1; |
1134 | current_order >= order && current_order <= MAX_ORDER-1; | |
1135 | --current_order) { | |
6d4a4916 | 1136 | for (i = 0;; i++) { |
b2a0ac88 MG |
1137 | migratetype = fallbacks[start_migratetype][i]; |
1138 | ||
56fd56b8 MG |
1139 | /* MIGRATE_RESERVE handled later if necessary */ |
1140 | if (migratetype == MIGRATE_RESERVE) | |
6d4a4916 | 1141 | break; |
e010487d | 1142 | |
b2a0ac88 MG |
1143 | area = &(zone->free_area[current_order]); |
1144 | if (list_empty(&area->free_list[migratetype])) | |
1145 | continue; | |
1146 | ||
1147 | page = list_entry(area->free_list[migratetype].next, | |
1148 | struct page, lru); | |
1149 | area->nr_free--; | |
1150 | ||
fef903ef SB |
1151 | new_type = try_to_steal_freepages(zone, page, |
1152 | start_migratetype, | |
1153 | migratetype); | |
b2a0ac88 MG |
1154 | |
1155 | /* Remove the page from the freelists */ | |
1156 | list_del(&page->lru); | |
1157 | rmv_page_order(page); | |
b2a0ac88 | 1158 | |
47118af0 | 1159 | expand(zone, page, order, current_order, area, |
0cbef29a | 1160 | new_type); |
5bcc9f86 VB |
1161 | /* The freepage_migratetype may differ from pageblock's |
1162 | * migratetype depending on the decisions in | |
1163 | * try_to_steal_freepages. This is OK as long as it does | |
1164 | * not differ for MIGRATE_CMA type. | |
1165 | */ | |
1166 | set_freepage_migratetype(page, new_type); | |
e0fff1bd | 1167 | |
52c8f6a5 KM |
1168 | trace_mm_page_alloc_extfrag(page, order, current_order, |
1169 | start_migratetype, migratetype, new_type); | |
e0fff1bd | 1170 | |
b2a0ac88 MG |
1171 | return page; |
1172 | } | |
1173 | } | |
1174 | ||
728ec980 | 1175 | return NULL; |
b2a0ac88 MG |
1176 | } |
1177 | ||
56fd56b8 | 1178 | /* |
1da177e4 LT |
1179 | * Do the hard work of removing an element from the buddy allocator. |
1180 | * Call me with the zone->lock already held. | |
1181 | */ | |
b2a0ac88 MG |
1182 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1183 | int migratetype) | |
1da177e4 | 1184 | { |
1da177e4 LT |
1185 | struct page *page; |
1186 | ||
728ec980 | 1187 | retry_reserve: |
56fd56b8 | 1188 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1189 | |
728ec980 | 1190 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 1191 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 1192 | |
728ec980 MG |
1193 | /* |
1194 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1195 | * is used because __rmqueue_smallest is an inline function | |
1196 | * and we want just one call site | |
1197 | */ | |
1198 | if (!page) { | |
1199 | migratetype = MIGRATE_RESERVE; | |
1200 | goto retry_reserve; | |
1201 | } | |
1202 | } | |
1203 | ||
0d3d062a | 1204 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1205 | return page; |
1da177e4 LT |
1206 | } |
1207 | ||
5f63b720 | 1208 | /* |
1da177e4 LT |
1209 | * Obtain a specified number of elements from the buddy allocator, all under |
1210 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1211 | * Returns the number of new pages which were placed at *list. | |
1212 | */ | |
5f63b720 | 1213 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1214 | unsigned long count, struct list_head *list, |
b745bc85 | 1215 | int migratetype, bool cold) |
1da177e4 | 1216 | { |
5bcc9f86 | 1217 | int i; |
5f63b720 | 1218 | |
c54ad30c | 1219 | spin_lock(&zone->lock); |
1da177e4 | 1220 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1221 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1222 | if (unlikely(page == NULL)) |
1da177e4 | 1223 | break; |
81eabcbe MG |
1224 | |
1225 | /* | |
1226 | * Split buddy pages returned by expand() are received here | |
1227 | * in physical page order. The page is added to the callers and | |
1228 | * list and the list head then moves forward. From the callers | |
1229 | * perspective, the linked list is ordered by page number in | |
1230 | * some conditions. This is useful for IO devices that can | |
1231 | * merge IO requests if the physical pages are ordered | |
1232 | * properly. | |
1233 | */ | |
b745bc85 | 1234 | if (likely(!cold)) |
e084b2d9 MG |
1235 | list_add(&page->lru, list); |
1236 | else | |
1237 | list_add_tail(&page->lru, list); | |
81eabcbe | 1238 | list = &page->lru; |
5bcc9f86 | 1239 | if (is_migrate_cma(get_freepage_migratetype(page))) |
d1ce749a BZ |
1240 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
1241 | -(1 << order)); | |
1da177e4 | 1242 | } |
f2260e6b | 1243 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1244 | spin_unlock(&zone->lock); |
085cc7d5 | 1245 | return i; |
1da177e4 LT |
1246 | } |
1247 | ||
4ae7c039 | 1248 | #ifdef CONFIG_NUMA |
8fce4d8e | 1249 | /* |
4037d452 CL |
1250 | * Called from the vmstat counter updater to drain pagesets of this |
1251 | * currently executing processor on remote nodes after they have | |
1252 | * expired. | |
1253 | * | |
879336c3 CL |
1254 | * Note that this function must be called with the thread pinned to |
1255 | * a single processor. | |
8fce4d8e | 1256 | */ |
4037d452 | 1257 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1258 | { |
4ae7c039 | 1259 | unsigned long flags; |
7be12fc9 | 1260 | int to_drain, batch; |
4ae7c039 | 1261 | |
4037d452 | 1262 | local_irq_save(flags); |
998d39cb | 1263 | batch = ACCESS_ONCE(pcp->batch); |
7be12fc9 | 1264 | to_drain = min(pcp->count, batch); |
2a13515c KM |
1265 | if (to_drain > 0) { |
1266 | free_pcppages_bulk(zone, to_drain, pcp); | |
1267 | pcp->count -= to_drain; | |
1268 | } | |
4037d452 | 1269 | local_irq_restore(flags); |
4ae7c039 CL |
1270 | } |
1271 | #endif | |
1272 | ||
9f8f2172 CL |
1273 | /* |
1274 | * Drain pages of the indicated processor. | |
1275 | * | |
1276 | * The processor must either be the current processor and the | |
1277 | * thread pinned to the current processor or a processor that | |
1278 | * is not online. | |
1279 | */ | |
1280 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1281 | { |
c54ad30c | 1282 | unsigned long flags; |
1da177e4 | 1283 | struct zone *zone; |
1da177e4 | 1284 | |
ee99c71c | 1285 | for_each_populated_zone(zone) { |
1da177e4 | 1286 | struct per_cpu_pageset *pset; |
3dfa5721 | 1287 | struct per_cpu_pages *pcp; |
1da177e4 | 1288 | |
99dcc3e5 CL |
1289 | local_irq_save(flags); |
1290 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3dfa5721 CL |
1291 | |
1292 | pcp = &pset->pcp; | |
2ff754fa DR |
1293 | if (pcp->count) { |
1294 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1295 | pcp->count = 0; | |
1296 | } | |
3dfa5721 | 1297 | local_irq_restore(flags); |
1da177e4 LT |
1298 | } |
1299 | } | |
1da177e4 | 1300 | |
9f8f2172 CL |
1301 | /* |
1302 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1303 | */ | |
1304 | void drain_local_pages(void *arg) | |
1305 | { | |
1306 | drain_pages(smp_processor_id()); | |
1307 | } | |
1308 | ||
1309 | /* | |
74046494 GBY |
1310 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
1311 | * | |
1312 | * Note that this code is protected against sending an IPI to an offline | |
1313 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
1314 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
1315 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
1316 | * before the call to on_each_cpu_mask(). | |
9f8f2172 CL |
1317 | */ |
1318 | void drain_all_pages(void) | |
1319 | { | |
74046494 GBY |
1320 | int cpu; |
1321 | struct per_cpu_pageset *pcp; | |
1322 | struct zone *zone; | |
1323 | ||
1324 | /* | |
1325 | * Allocate in the BSS so we wont require allocation in | |
1326 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
1327 | */ | |
1328 | static cpumask_t cpus_with_pcps; | |
1329 | ||
1330 | /* | |
1331 | * We don't care about racing with CPU hotplug event | |
1332 | * as offline notification will cause the notified | |
1333 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
1334 | * disables preemption as part of its processing | |
1335 | */ | |
1336 | for_each_online_cpu(cpu) { | |
1337 | bool has_pcps = false; | |
1338 | for_each_populated_zone(zone) { | |
1339 | pcp = per_cpu_ptr(zone->pageset, cpu); | |
1340 | if (pcp->pcp.count) { | |
1341 | has_pcps = true; | |
1342 | break; | |
1343 | } | |
1344 | } | |
1345 | if (has_pcps) | |
1346 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
1347 | else | |
1348 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
1349 | } | |
1350 | on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1); | |
9f8f2172 CL |
1351 | } |
1352 | ||
296699de | 1353 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1354 | |
1355 | void mark_free_pages(struct zone *zone) | |
1356 | { | |
f623f0db RW |
1357 | unsigned long pfn, max_zone_pfn; |
1358 | unsigned long flags; | |
7aeb09f9 | 1359 | unsigned int order, t; |
1da177e4 LT |
1360 | struct list_head *curr; |
1361 | ||
8080fc03 | 1362 | if (zone_is_empty(zone)) |
1da177e4 LT |
1363 | return; |
1364 | ||
1365 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 1366 | |
108bcc96 | 1367 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
1368 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
1369 | if (pfn_valid(pfn)) { | |
1370 | struct page *page = pfn_to_page(pfn); | |
1371 | ||
7be98234 RW |
1372 | if (!swsusp_page_is_forbidden(page)) |
1373 | swsusp_unset_page_free(page); | |
f623f0db | 1374 | } |
1da177e4 | 1375 | |
b2a0ac88 MG |
1376 | for_each_migratetype_order(order, t) { |
1377 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1378 | unsigned long i; |
1da177e4 | 1379 | |
f623f0db RW |
1380 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1381 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1382 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1383 | } |
b2a0ac88 | 1384 | } |
1da177e4 LT |
1385 | spin_unlock_irqrestore(&zone->lock, flags); |
1386 | } | |
e2c55dc8 | 1387 | #endif /* CONFIG_PM */ |
1da177e4 | 1388 | |
1da177e4 LT |
1389 | /* |
1390 | * Free a 0-order page | |
b745bc85 | 1391 | * cold == true ? free a cold page : free a hot page |
1da177e4 | 1392 | */ |
b745bc85 | 1393 | void free_hot_cold_page(struct page *page, bool cold) |
1da177e4 LT |
1394 | { |
1395 | struct zone *zone = page_zone(page); | |
1396 | struct per_cpu_pages *pcp; | |
1397 | unsigned long flags; | |
dc4b0caf | 1398 | unsigned long pfn = page_to_pfn(page); |
5f8dcc21 | 1399 | int migratetype; |
1da177e4 | 1400 | |
ec95f53a | 1401 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1402 | return; |
1403 | ||
dc4b0caf | 1404 | migratetype = get_pfnblock_migratetype(page, pfn); |
b12c4ad1 | 1405 | set_freepage_migratetype(page, migratetype); |
1da177e4 | 1406 | local_irq_save(flags); |
f8891e5e | 1407 | __count_vm_event(PGFREE); |
da456f14 | 1408 | |
5f8dcc21 MG |
1409 | /* |
1410 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1411 | * Free ISOLATE pages back to the allocator because they are being | |
1412 | * offlined but treat RESERVE as movable pages so we can get those | |
1413 | * areas back if necessary. Otherwise, we may have to free | |
1414 | * excessively into the page allocator | |
1415 | */ | |
1416 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 1417 | if (unlikely(is_migrate_isolate(migratetype))) { |
dc4b0caf | 1418 | free_one_page(zone, page, pfn, 0, migratetype); |
5f8dcc21 MG |
1419 | goto out; |
1420 | } | |
1421 | migratetype = MIGRATE_MOVABLE; | |
1422 | } | |
1423 | ||
99dcc3e5 | 1424 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
b745bc85 | 1425 | if (!cold) |
5f8dcc21 | 1426 | list_add(&page->lru, &pcp->lists[migratetype]); |
b745bc85 MG |
1427 | else |
1428 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1da177e4 | 1429 | pcp->count++; |
48db57f8 | 1430 | if (pcp->count >= pcp->high) { |
998d39cb CS |
1431 | unsigned long batch = ACCESS_ONCE(pcp->batch); |
1432 | free_pcppages_bulk(zone, batch, pcp); | |
1433 | pcp->count -= batch; | |
48db57f8 | 1434 | } |
5f8dcc21 MG |
1435 | |
1436 | out: | |
1da177e4 | 1437 | local_irq_restore(flags); |
1da177e4 LT |
1438 | } |
1439 | ||
cc59850e KK |
1440 | /* |
1441 | * Free a list of 0-order pages | |
1442 | */ | |
b745bc85 | 1443 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
cc59850e KK |
1444 | { |
1445 | struct page *page, *next; | |
1446 | ||
1447 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1448 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1449 | free_hot_cold_page(page, cold); |
1450 | } | |
1451 | } | |
1452 | ||
8dfcc9ba NP |
1453 | /* |
1454 | * split_page takes a non-compound higher-order page, and splits it into | |
1455 | * n (1<<order) sub-pages: page[0..n] | |
1456 | * Each sub-page must be freed individually. | |
1457 | * | |
1458 | * Note: this is probably too low level an operation for use in drivers. | |
1459 | * Please consult with lkml before using this in your driver. | |
1460 | */ | |
1461 | void split_page(struct page *page, unsigned int order) | |
1462 | { | |
1463 | int i; | |
1464 | ||
309381fe SL |
1465 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1466 | VM_BUG_ON_PAGE(!page_count(page), page); | |
b1eeab67 VN |
1467 | |
1468 | #ifdef CONFIG_KMEMCHECK | |
1469 | /* | |
1470 | * Split shadow pages too, because free(page[0]) would | |
1471 | * otherwise free the whole shadow. | |
1472 | */ | |
1473 | if (kmemcheck_page_is_tracked(page)) | |
1474 | split_page(virt_to_page(page[0].shadow), order); | |
1475 | #endif | |
1476 | ||
7835e98b NP |
1477 | for (i = 1; i < (1 << order); i++) |
1478 | set_page_refcounted(page + i); | |
8dfcc9ba | 1479 | } |
5853ff23 | 1480 | EXPORT_SYMBOL_GPL(split_page); |
8dfcc9ba | 1481 | |
8fb74b9f | 1482 | static int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 1483 | { |
748446bb MG |
1484 | unsigned long watermark; |
1485 | struct zone *zone; | |
2139cbe6 | 1486 | int mt; |
748446bb MG |
1487 | |
1488 | BUG_ON(!PageBuddy(page)); | |
1489 | ||
1490 | zone = page_zone(page); | |
2e30abd1 | 1491 | mt = get_pageblock_migratetype(page); |
748446bb | 1492 | |
194159fb | 1493 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
1494 | /* Obey watermarks as if the page was being allocated */ |
1495 | watermark = low_wmark_pages(zone) + (1 << order); | |
1496 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1497 | return 0; | |
1498 | ||
8fb74b9f | 1499 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 1500 | } |
748446bb MG |
1501 | |
1502 | /* Remove page from free list */ | |
1503 | list_del(&page->lru); | |
1504 | zone->free_area[order].nr_free--; | |
1505 | rmv_page_order(page); | |
2139cbe6 | 1506 | |
8fb74b9f | 1507 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
1508 | if (order >= pageblock_order - 1) { |
1509 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
1510 | for (; page < endpage; page += pageblock_nr_pages) { |
1511 | int mt = get_pageblock_migratetype(page); | |
194159fb | 1512 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
1513 | set_pageblock_migratetype(page, |
1514 | MIGRATE_MOVABLE); | |
1515 | } | |
748446bb MG |
1516 | } |
1517 | ||
8fb74b9f | 1518 | return 1UL << order; |
1fb3f8ca MG |
1519 | } |
1520 | ||
1521 | /* | |
1522 | * Similar to split_page except the page is already free. As this is only | |
1523 | * being used for migration, the migratetype of the block also changes. | |
1524 | * As this is called with interrupts disabled, the caller is responsible | |
1525 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1526 | * are enabled. | |
1527 | * | |
1528 | * Note: this is probably too low level an operation for use in drivers. | |
1529 | * Please consult with lkml before using this in your driver. | |
1530 | */ | |
1531 | int split_free_page(struct page *page) | |
1532 | { | |
1533 | unsigned int order; | |
1534 | int nr_pages; | |
1535 | ||
1fb3f8ca MG |
1536 | order = page_order(page); |
1537 | ||
8fb74b9f | 1538 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
1539 | if (!nr_pages) |
1540 | return 0; | |
1541 | ||
1542 | /* Split into individual pages */ | |
1543 | set_page_refcounted(page); | |
1544 | split_page(page, order); | |
1545 | return nr_pages; | |
748446bb MG |
1546 | } |
1547 | ||
1da177e4 LT |
1548 | /* |
1549 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1550 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1551 | * or two. | |
1552 | */ | |
0a15c3e9 MG |
1553 | static inline |
1554 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
7aeb09f9 MG |
1555 | struct zone *zone, unsigned int order, |
1556 | gfp_t gfp_flags, int migratetype) | |
1da177e4 LT |
1557 | { |
1558 | unsigned long flags; | |
689bcebf | 1559 | struct page *page; |
b745bc85 | 1560 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
1da177e4 | 1561 | |
689bcebf | 1562 | again: |
48db57f8 | 1563 | if (likely(order == 0)) { |
1da177e4 | 1564 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1565 | struct list_head *list; |
1da177e4 | 1566 | |
1da177e4 | 1567 | local_irq_save(flags); |
99dcc3e5 CL |
1568 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1569 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1570 | if (list_empty(list)) { |
535131e6 | 1571 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1572 | pcp->batch, list, |
e084b2d9 | 1573 | migratetype, cold); |
5f8dcc21 | 1574 | if (unlikely(list_empty(list))) |
6fb332fa | 1575 | goto failed; |
535131e6 | 1576 | } |
b92a6edd | 1577 | |
5f8dcc21 MG |
1578 | if (cold) |
1579 | page = list_entry(list->prev, struct page, lru); | |
1580 | else | |
1581 | page = list_entry(list->next, struct page, lru); | |
1582 | ||
b92a6edd MG |
1583 | list_del(&page->lru); |
1584 | pcp->count--; | |
7fb1d9fc | 1585 | } else { |
dab48dab AM |
1586 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1587 | /* | |
1588 | * __GFP_NOFAIL is not to be used in new code. | |
1589 | * | |
1590 | * All __GFP_NOFAIL callers should be fixed so that they | |
1591 | * properly detect and handle allocation failures. | |
1592 | * | |
1593 | * We most definitely don't want callers attempting to | |
4923abf9 | 1594 | * allocate greater than order-1 page units with |
dab48dab AM |
1595 | * __GFP_NOFAIL. |
1596 | */ | |
4923abf9 | 1597 | WARN_ON_ONCE(order > 1); |
dab48dab | 1598 | } |
1da177e4 | 1599 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1600 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1601 | spin_unlock(&zone->lock); |
1602 | if (!page) | |
1603 | goto failed; | |
d1ce749a | 1604 | __mod_zone_freepage_state(zone, -(1 << order), |
5bcc9f86 | 1605 | get_freepage_migratetype(page)); |
1da177e4 LT |
1606 | } |
1607 | ||
3a025760 | 1608 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order)); |
27329369 | 1609 | |
f8891e5e | 1610 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1611 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1612 | local_irq_restore(flags); |
1da177e4 | 1613 | |
309381fe | 1614 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
17cf4406 | 1615 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1616 | goto again; |
1da177e4 | 1617 | return page; |
a74609fa NP |
1618 | |
1619 | failed: | |
1620 | local_irq_restore(flags); | |
a74609fa | 1621 | return NULL; |
1da177e4 LT |
1622 | } |
1623 | ||
933e312e AM |
1624 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1625 | ||
b2588c4b | 1626 | static struct { |
933e312e AM |
1627 | struct fault_attr attr; |
1628 | ||
1629 | u32 ignore_gfp_highmem; | |
1630 | u32 ignore_gfp_wait; | |
54114994 | 1631 | u32 min_order; |
933e312e AM |
1632 | } fail_page_alloc = { |
1633 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1634 | .ignore_gfp_wait = 1, |
1635 | .ignore_gfp_highmem = 1, | |
54114994 | 1636 | .min_order = 1, |
933e312e AM |
1637 | }; |
1638 | ||
1639 | static int __init setup_fail_page_alloc(char *str) | |
1640 | { | |
1641 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1642 | } | |
1643 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1644 | ||
deaf386e | 1645 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1646 | { |
54114994 | 1647 | if (order < fail_page_alloc.min_order) |
deaf386e | 1648 | return false; |
933e312e | 1649 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 1650 | return false; |
933e312e | 1651 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 1652 | return false; |
933e312e | 1653 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
deaf386e | 1654 | return false; |
933e312e AM |
1655 | |
1656 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1657 | } | |
1658 | ||
1659 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1660 | ||
1661 | static int __init fail_page_alloc_debugfs(void) | |
1662 | { | |
f4ae40a6 | 1663 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1664 | struct dentry *dir; |
933e312e | 1665 | |
dd48c085 AM |
1666 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1667 | &fail_page_alloc.attr); | |
1668 | if (IS_ERR(dir)) | |
1669 | return PTR_ERR(dir); | |
933e312e | 1670 | |
b2588c4b AM |
1671 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1672 | &fail_page_alloc.ignore_gfp_wait)) | |
1673 | goto fail; | |
1674 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1675 | &fail_page_alloc.ignore_gfp_highmem)) | |
1676 | goto fail; | |
1677 | if (!debugfs_create_u32("min-order", mode, dir, | |
1678 | &fail_page_alloc.min_order)) | |
1679 | goto fail; | |
1680 | ||
1681 | return 0; | |
1682 | fail: | |
dd48c085 | 1683 | debugfs_remove_recursive(dir); |
933e312e | 1684 | |
b2588c4b | 1685 | return -ENOMEM; |
933e312e AM |
1686 | } |
1687 | ||
1688 | late_initcall(fail_page_alloc_debugfs); | |
1689 | ||
1690 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1691 | ||
1692 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1693 | ||
deaf386e | 1694 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1695 | { |
deaf386e | 1696 | return false; |
933e312e AM |
1697 | } |
1698 | ||
1699 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1700 | ||
1da177e4 | 1701 | /* |
88f5acf8 | 1702 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1703 | * of the allocation. |
1704 | */ | |
7aeb09f9 MG |
1705 | static bool __zone_watermark_ok(struct zone *z, unsigned int order, |
1706 | unsigned long mark, int classzone_idx, int alloc_flags, | |
1707 | long free_pages) | |
1da177e4 LT |
1708 | { |
1709 | /* free_pages my go negative - that's OK */ | |
d23ad423 | 1710 | long min = mark; |
2cfed075 | 1711 | long lowmem_reserve = z->lowmem_reserve[classzone_idx]; |
1da177e4 | 1712 | int o; |
026b0814 | 1713 | long free_cma = 0; |
1da177e4 | 1714 | |
df0a6daa | 1715 | free_pages -= (1 << order) - 1; |
7fb1d9fc | 1716 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1717 | min -= min / 2; |
7fb1d9fc | 1718 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 | 1719 | min -= min / 4; |
d95ea5d1 BZ |
1720 | #ifdef CONFIG_CMA |
1721 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
1722 | if (!(alloc_flags & ALLOC_CMA)) | |
026b0814 | 1723 | free_cma = zone_page_state(z, NR_FREE_CMA_PAGES); |
d95ea5d1 | 1724 | #endif |
026b0814 TS |
1725 | |
1726 | if (free_pages - free_cma <= min + lowmem_reserve) | |
88f5acf8 | 1727 | return false; |
1da177e4 LT |
1728 | for (o = 0; o < order; o++) { |
1729 | /* At the next order, this order's pages become unavailable */ | |
1730 | free_pages -= z->free_area[o].nr_free << o; | |
1731 | ||
1732 | /* Require fewer higher order pages to be free */ | |
1733 | min >>= 1; | |
1734 | ||
1735 | if (free_pages <= min) | |
88f5acf8 | 1736 | return false; |
1da177e4 | 1737 | } |
88f5acf8 MG |
1738 | return true; |
1739 | } | |
1740 | ||
7aeb09f9 | 1741 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
88f5acf8 MG |
1742 | int classzone_idx, int alloc_flags) |
1743 | { | |
1744 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1745 | zone_page_state(z, NR_FREE_PAGES)); | |
1746 | } | |
1747 | ||
7aeb09f9 MG |
1748 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
1749 | unsigned long mark, int classzone_idx, int alloc_flags) | |
88f5acf8 MG |
1750 | { |
1751 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1752 | ||
1753 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1754 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1755 | ||
1756 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1757 | free_pages); | |
1da177e4 LT |
1758 | } |
1759 | ||
9276b1bc PJ |
1760 | #ifdef CONFIG_NUMA |
1761 | /* | |
1762 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1763 | * skip over zones that are not allowed by the cpuset, or that have | |
1764 | * been recently (in last second) found to be nearly full. See further | |
1765 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1766 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc | 1767 | * |
a1aeb65a | 1768 | * If the zonelist cache is present in the passed zonelist, then |
9276b1bc | 1769 | * returns a pointer to the allowed node mask (either the current |
4b0ef1fe | 1770 | * tasks mems_allowed, or node_states[N_MEMORY].) |
9276b1bc PJ |
1771 | * |
1772 | * If the zonelist cache is not available for this zonelist, does | |
1773 | * nothing and returns NULL. | |
1774 | * | |
1775 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1776 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1777 | * | |
1778 | * We hold off even calling zlc_setup, until after we've checked the | |
1779 | * first zone in the zonelist, on the theory that most allocations will | |
1780 | * be satisfied from that first zone, so best to examine that zone as | |
1781 | * quickly as we can. | |
1782 | */ | |
1783 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1784 | { | |
1785 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1786 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1787 | ||
1788 | zlc = zonelist->zlcache_ptr; | |
1789 | if (!zlc) | |
1790 | return NULL; | |
1791 | ||
f05111f5 | 1792 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1793 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1794 | zlc->last_full_zap = jiffies; | |
1795 | } | |
1796 | ||
1797 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1798 | &cpuset_current_mems_allowed : | |
4b0ef1fe | 1799 | &node_states[N_MEMORY]; |
9276b1bc PJ |
1800 | return allowednodes; |
1801 | } | |
1802 | ||
1803 | /* | |
1804 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1805 | * if it is worth looking at further for free memory: | |
1806 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1807 | * bit set in the zonelist_cache fullzones BITMAP). | |
1808 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1809 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1810 | * Return true (non-zero) if zone is worth looking at further, or | |
1811 | * else return false (zero) if it is not. | |
1812 | * | |
1813 | * This check -ignores- the distinction between various watermarks, | |
1814 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1815 | * found to be full for any variation of these watermarks, it will | |
1816 | * be considered full for up to one second by all requests, unless | |
1817 | * we are so low on memory on all allowed nodes that we are forced | |
1818 | * into the second scan of the zonelist. | |
1819 | * | |
1820 | * In the second scan we ignore this zonelist cache and exactly | |
1821 | * apply the watermarks to all zones, even it is slower to do so. | |
1822 | * We are low on memory in the second scan, and should leave no stone | |
1823 | * unturned looking for a free page. | |
1824 | */ | |
dd1a239f | 1825 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1826 | nodemask_t *allowednodes) |
1827 | { | |
1828 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1829 | int i; /* index of *z in zonelist zones */ | |
1830 | int n; /* node that zone *z is on */ | |
1831 | ||
1832 | zlc = zonelist->zlcache_ptr; | |
1833 | if (!zlc) | |
1834 | return 1; | |
1835 | ||
dd1a239f | 1836 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1837 | n = zlc->z_to_n[i]; |
1838 | ||
1839 | /* This zone is worth trying if it is allowed but not full */ | |
1840 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1841 | } | |
1842 | ||
1843 | /* | |
1844 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1845 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1846 | * from that zone don't waste time re-examining it. | |
1847 | */ | |
dd1a239f | 1848 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1849 | { |
1850 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1851 | int i; /* index of *z in zonelist zones */ | |
1852 | ||
1853 | zlc = zonelist->zlcache_ptr; | |
1854 | if (!zlc) | |
1855 | return; | |
1856 | ||
dd1a239f | 1857 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1858 | |
1859 | set_bit(i, zlc->fullzones); | |
1860 | } | |
1861 | ||
76d3fbf8 MG |
1862 | /* |
1863 | * clear all zones full, called after direct reclaim makes progress so that | |
1864 | * a zone that was recently full is not skipped over for up to a second | |
1865 | */ | |
1866 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1867 | { | |
1868 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1869 | ||
1870 | zlc = zonelist->zlcache_ptr; | |
1871 | if (!zlc) | |
1872 | return; | |
1873 | ||
1874 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1875 | } | |
1876 | ||
81c0a2bb JW |
1877 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
1878 | { | |
fff4068c | 1879 | return local_zone->node == zone->node; |
81c0a2bb JW |
1880 | } |
1881 | ||
957f822a DR |
1882 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1883 | { | |
5f7a75ac MG |
1884 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) < |
1885 | RECLAIM_DISTANCE; | |
957f822a DR |
1886 | } |
1887 | ||
9276b1bc PJ |
1888 | #else /* CONFIG_NUMA */ |
1889 | ||
1890 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1891 | { | |
1892 | return NULL; | |
1893 | } | |
1894 | ||
dd1a239f | 1895 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1896 | nodemask_t *allowednodes) |
1897 | { | |
1898 | return 1; | |
1899 | } | |
1900 | ||
dd1a239f | 1901 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1902 | { |
1903 | } | |
76d3fbf8 MG |
1904 | |
1905 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1906 | { | |
1907 | } | |
957f822a | 1908 | |
81c0a2bb JW |
1909 | static bool zone_local(struct zone *local_zone, struct zone *zone) |
1910 | { | |
1911 | return true; | |
1912 | } | |
1913 | ||
957f822a DR |
1914 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1915 | { | |
1916 | return true; | |
1917 | } | |
1918 | ||
9276b1bc PJ |
1919 | #endif /* CONFIG_NUMA */ |
1920 | ||
7fb1d9fc | 1921 | /* |
0798e519 | 1922 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1923 | * a page. |
1924 | */ | |
1925 | static struct page * | |
19770b32 | 1926 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1927 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
d8846374 | 1928 | struct zone *preferred_zone, int classzone_idx, int migratetype) |
753ee728 | 1929 | { |
dd1a239f | 1930 | struct zoneref *z; |
7fb1d9fc | 1931 | struct page *page = NULL; |
5117f45d | 1932 | struct zone *zone; |
9276b1bc PJ |
1933 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1934 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1935 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
a6e21b14 MG |
1936 | bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) && |
1937 | (gfp_mask & __GFP_WRITE); | |
54a6eb5c | 1938 | |
9276b1bc | 1939 | zonelist_scan: |
7fb1d9fc | 1940 | /* |
9276b1bc | 1941 | * Scan zonelist, looking for a zone with enough free. |
3b11f0aa | 1942 | * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c. |
7fb1d9fc | 1943 | */ |
19770b32 MG |
1944 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1945 | high_zoneidx, nodemask) { | |
e085dbc5 JW |
1946 | unsigned long mark; |
1947 | ||
e5adfffc | 1948 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
9276b1bc PJ |
1949 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1950 | continue; | |
664eedde MG |
1951 | if (cpusets_enabled() && |
1952 | (alloc_flags & ALLOC_CPUSET) && | |
02a0e53d | 1953 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
cd38b115 | 1954 | continue; |
81c0a2bb JW |
1955 | /* |
1956 | * Distribute pages in proportion to the individual | |
1957 | * zone size to ensure fair page aging. The zone a | |
1958 | * page was allocated in should have no effect on the | |
1959 | * time the page has in memory before being reclaimed. | |
81c0a2bb | 1960 | */ |
3a025760 | 1961 | if (alloc_flags & ALLOC_FAIR) { |
fff4068c | 1962 | if (!zone_local(preferred_zone, zone)) |
81c0a2bb | 1963 | continue; |
3a025760 JW |
1964 | if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0) |
1965 | continue; | |
81c0a2bb | 1966 | } |
a756cf59 JW |
1967 | /* |
1968 | * When allocating a page cache page for writing, we | |
1969 | * want to get it from a zone that is within its dirty | |
1970 | * limit, such that no single zone holds more than its | |
1971 | * proportional share of globally allowed dirty pages. | |
1972 | * The dirty limits take into account the zone's | |
1973 | * lowmem reserves and high watermark so that kswapd | |
1974 | * should be able to balance it without having to | |
1975 | * write pages from its LRU list. | |
1976 | * | |
1977 | * This may look like it could increase pressure on | |
1978 | * lower zones by failing allocations in higher zones | |
1979 | * before they are full. But the pages that do spill | |
1980 | * over are limited as the lower zones are protected | |
1981 | * by this very same mechanism. It should not become | |
1982 | * a practical burden to them. | |
1983 | * | |
1984 | * XXX: For now, allow allocations to potentially | |
1985 | * exceed the per-zone dirty limit in the slowpath | |
1986 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
1987 | * which is important when on a NUMA setup the allowed | |
1988 | * zones are together not big enough to reach the | |
1989 | * global limit. The proper fix for these situations | |
1990 | * will require awareness of zones in the | |
1991 | * dirty-throttling and the flusher threads. | |
1992 | */ | |
a6e21b14 | 1993 | if (consider_zone_dirty && !zone_dirty_ok(zone)) |
800a1e75 | 1994 | continue; |
7fb1d9fc | 1995 | |
e085dbc5 JW |
1996 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
1997 | if (!zone_watermark_ok(zone, order, mark, | |
1998 | classzone_idx, alloc_flags)) { | |
fa5e084e MG |
1999 | int ret; |
2000 | ||
5dab2911 MG |
2001 | /* Checked here to keep the fast path fast */ |
2002 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
2003 | if (alloc_flags & ALLOC_NO_WATERMARKS) | |
2004 | goto try_this_zone; | |
2005 | ||
e5adfffc KS |
2006 | if (IS_ENABLED(CONFIG_NUMA) && |
2007 | !did_zlc_setup && nr_online_nodes > 1) { | |
cd38b115 MG |
2008 | /* |
2009 | * we do zlc_setup if there are multiple nodes | |
2010 | * and before considering the first zone allowed | |
2011 | * by the cpuset. | |
2012 | */ | |
2013 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
2014 | zlc_active = 1; | |
2015 | did_zlc_setup = 1; | |
2016 | } | |
2017 | ||
957f822a DR |
2018 | if (zone_reclaim_mode == 0 || |
2019 | !zone_allows_reclaim(preferred_zone, zone)) | |
fa5e084e MG |
2020 | goto this_zone_full; |
2021 | ||
cd38b115 MG |
2022 | /* |
2023 | * As we may have just activated ZLC, check if the first | |
2024 | * eligible zone has failed zone_reclaim recently. | |
2025 | */ | |
e5adfffc | 2026 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
cd38b115 MG |
2027 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
2028 | continue; | |
2029 | ||
fa5e084e MG |
2030 | ret = zone_reclaim(zone, gfp_mask, order); |
2031 | switch (ret) { | |
2032 | case ZONE_RECLAIM_NOSCAN: | |
2033 | /* did not scan */ | |
cd38b115 | 2034 | continue; |
fa5e084e MG |
2035 | case ZONE_RECLAIM_FULL: |
2036 | /* scanned but unreclaimable */ | |
cd38b115 | 2037 | continue; |
fa5e084e MG |
2038 | default: |
2039 | /* did we reclaim enough */ | |
fed2719e | 2040 | if (zone_watermark_ok(zone, order, mark, |
fa5e084e | 2041 | classzone_idx, alloc_flags)) |
fed2719e MG |
2042 | goto try_this_zone; |
2043 | ||
2044 | /* | |
2045 | * Failed to reclaim enough to meet watermark. | |
2046 | * Only mark the zone full if checking the min | |
2047 | * watermark or if we failed to reclaim just | |
2048 | * 1<<order pages or else the page allocator | |
2049 | * fastpath will prematurely mark zones full | |
2050 | * when the watermark is between the low and | |
2051 | * min watermarks. | |
2052 | */ | |
2053 | if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) || | |
2054 | ret == ZONE_RECLAIM_SOME) | |
9276b1bc | 2055 | goto this_zone_full; |
fed2719e MG |
2056 | |
2057 | continue; | |
0798e519 | 2058 | } |
7fb1d9fc RS |
2059 | } |
2060 | ||
fa5e084e | 2061 | try_this_zone: |
3dd28266 MG |
2062 | page = buffered_rmqueue(preferred_zone, zone, order, |
2063 | gfp_mask, migratetype); | |
0798e519 | 2064 | if (page) |
7fb1d9fc | 2065 | break; |
9276b1bc | 2066 | this_zone_full: |
65bb3719 | 2067 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active) |
9276b1bc | 2068 | zlc_mark_zone_full(zonelist, z); |
54a6eb5c | 2069 | } |
9276b1bc | 2070 | |
e5adfffc | 2071 | if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) { |
9276b1bc PJ |
2072 | /* Disable zlc cache for second zonelist scan */ |
2073 | zlc_active = 0; | |
2074 | goto zonelist_scan; | |
2075 | } | |
b121186a AS |
2076 | |
2077 | if (page) | |
2078 | /* | |
2079 | * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was | |
2080 | * necessary to allocate the page. The expectation is | |
2081 | * that the caller is taking steps that will free more | |
2082 | * memory. The caller should avoid the page being used | |
2083 | * for !PFMEMALLOC purposes. | |
2084 | */ | |
2085 | page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS); | |
2086 | ||
7fb1d9fc | 2087 | return page; |
753ee728 MH |
2088 | } |
2089 | ||
29423e77 DR |
2090 | /* |
2091 | * Large machines with many possible nodes should not always dump per-node | |
2092 | * meminfo in irq context. | |
2093 | */ | |
2094 | static inline bool should_suppress_show_mem(void) | |
2095 | { | |
2096 | bool ret = false; | |
2097 | ||
2098 | #if NODES_SHIFT > 8 | |
2099 | ret = in_interrupt(); | |
2100 | #endif | |
2101 | return ret; | |
2102 | } | |
2103 | ||
a238ab5b DH |
2104 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2105 | DEFAULT_RATELIMIT_INTERVAL, | |
2106 | DEFAULT_RATELIMIT_BURST); | |
2107 | ||
2108 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
2109 | { | |
a238ab5b DH |
2110 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2111 | ||
c0a32fc5 SG |
2112 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2113 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2114 | return; |
2115 | ||
2116 | /* | |
2117 | * This documents exceptions given to allocations in certain | |
2118 | * contexts that are allowed to allocate outside current's set | |
2119 | * of allowed nodes. | |
2120 | */ | |
2121 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2122 | if (test_thread_flag(TIF_MEMDIE) || | |
2123 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2124 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2125 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
2126 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2127 | ||
2128 | if (fmt) { | |
3ee9a4f0 JP |
2129 | struct va_format vaf; |
2130 | va_list args; | |
2131 | ||
a238ab5b | 2132 | va_start(args, fmt); |
3ee9a4f0 JP |
2133 | |
2134 | vaf.fmt = fmt; | |
2135 | vaf.va = &args; | |
2136 | ||
2137 | pr_warn("%pV", &vaf); | |
2138 | ||
a238ab5b DH |
2139 | va_end(args); |
2140 | } | |
2141 | ||
3ee9a4f0 JP |
2142 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
2143 | current->comm, order, gfp_mask); | |
a238ab5b DH |
2144 | |
2145 | dump_stack(); | |
2146 | if (!should_suppress_show_mem()) | |
2147 | show_mem(filter); | |
2148 | } | |
2149 | ||
11e33f6a MG |
2150 | static inline int |
2151 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
f90ac398 | 2152 | unsigned long did_some_progress, |
11e33f6a | 2153 | unsigned long pages_reclaimed) |
1da177e4 | 2154 | { |
11e33f6a MG |
2155 | /* Do not loop if specifically requested */ |
2156 | if (gfp_mask & __GFP_NORETRY) | |
2157 | return 0; | |
1da177e4 | 2158 | |
f90ac398 MG |
2159 | /* Always retry if specifically requested */ |
2160 | if (gfp_mask & __GFP_NOFAIL) | |
2161 | return 1; | |
2162 | ||
2163 | /* | |
2164 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim | |
2165 | * making forward progress without invoking OOM. Suspend also disables | |
2166 | * storage devices so kswapd will not help. Bail if we are suspending. | |
2167 | */ | |
2168 | if (!did_some_progress && pm_suspended_storage()) | |
2169 | return 0; | |
2170 | ||
11e33f6a MG |
2171 | /* |
2172 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
2173 | * means __GFP_NOFAIL, but that may not be true in other | |
2174 | * implementations. | |
2175 | */ | |
2176 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
2177 | return 1; | |
2178 | ||
2179 | /* | |
2180 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
2181 | * specified, then we retry until we no longer reclaim any pages | |
2182 | * (above), or we've reclaimed an order of pages at least as | |
2183 | * large as the allocation's order. In both cases, if the | |
2184 | * allocation still fails, we stop retrying. | |
2185 | */ | |
2186 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
2187 | return 1; | |
cf40bd16 | 2188 | |
11e33f6a MG |
2189 | return 0; |
2190 | } | |
933e312e | 2191 | |
11e33f6a MG |
2192 | static inline struct page * |
2193 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
2194 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 | 2195 | nodemask_t *nodemask, struct zone *preferred_zone, |
d8846374 | 2196 | int classzone_idx, int migratetype) |
11e33f6a MG |
2197 | { |
2198 | struct page *page; | |
2199 | ||
2200 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
ff321fea | 2201 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
11e33f6a | 2202 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2203 | return NULL; |
2204 | } | |
6b1de916 | 2205 | |
11e33f6a MG |
2206 | /* |
2207 | * Go through the zonelist yet one more time, keep very high watermark | |
2208 | * here, this is only to catch a parallel oom killing, we must fail if | |
2209 | * we're still under heavy pressure. | |
2210 | */ | |
2211 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
2212 | order, zonelist, high_zoneidx, | |
5117f45d | 2213 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
d8846374 | 2214 | preferred_zone, classzone_idx, migratetype); |
7fb1d9fc | 2215 | if (page) |
11e33f6a MG |
2216 | goto out; |
2217 | ||
4365a567 KH |
2218 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2219 | /* The OOM killer will not help higher order allocs */ | |
2220 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2221 | goto out; | |
03668b3c DR |
2222 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
2223 | if (high_zoneidx < ZONE_NORMAL) | |
2224 | goto out; | |
4365a567 KH |
2225 | /* |
2226 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
2227 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
2228 | * The caller should handle page allocation failure by itself if | |
2229 | * it specifies __GFP_THISNODE. | |
2230 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
2231 | */ | |
2232 | if (gfp_mask & __GFP_THISNODE) | |
2233 | goto out; | |
2234 | } | |
11e33f6a | 2235 | /* Exhausted what can be done so it's blamo time */ |
08ab9b10 | 2236 | out_of_memory(zonelist, gfp_mask, order, nodemask, false); |
11e33f6a MG |
2237 | |
2238 | out: | |
2239 | clear_zonelist_oom(zonelist, gfp_mask); | |
2240 | return page; | |
2241 | } | |
2242 | ||
56de7263 MG |
2243 | #ifdef CONFIG_COMPACTION |
2244 | /* Try memory compaction for high-order allocations before reclaim */ | |
2245 | static struct page * | |
2246 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2247 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2248 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
d8846374 | 2249 | int classzone_idx, int migratetype, enum migrate_mode mode, |
c67fe375 | 2250 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2251 | unsigned long *did_some_progress) |
56de7263 | 2252 | { |
66199712 | 2253 | if (!order) |
56de7263 MG |
2254 | return NULL; |
2255 | ||
aff62249 | 2256 | if (compaction_deferred(preferred_zone, order)) { |
66199712 MG |
2257 | *deferred_compaction = true; |
2258 | return NULL; | |
2259 | } | |
2260 | ||
c06b1fca | 2261 | current->flags |= PF_MEMALLOC; |
56de7263 | 2262 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
e0b9daeb | 2263 | nodemask, mode, |
8fb74b9f | 2264 | contended_compaction); |
c06b1fca | 2265 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2266 | |
1fb3f8ca | 2267 | if (*did_some_progress != COMPACT_SKIPPED) { |
8fb74b9f MG |
2268 | struct page *page; |
2269 | ||
56de7263 MG |
2270 | /* Page migration frees to the PCP lists but we want merging */ |
2271 | drain_pages(get_cpu()); | |
2272 | put_cpu(); | |
2273 | ||
2274 | page = get_page_from_freelist(gfp_mask, nodemask, | |
2275 | order, zonelist, high_zoneidx, | |
cfd19c5a | 2276 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
d8846374 | 2277 | preferred_zone, classzone_idx, migratetype); |
56de7263 | 2278 | if (page) { |
62997027 | 2279 | preferred_zone->compact_blockskip_flush = false; |
de6c60a6 | 2280 | compaction_defer_reset(preferred_zone, order, true); |
56de7263 MG |
2281 | count_vm_event(COMPACTSUCCESS); |
2282 | return page; | |
2283 | } | |
2284 | ||
2285 | /* | |
2286 | * It's bad if compaction run occurs and fails. | |
2287 | * The most likely reason is that pages exist, | |
2288 | * but not enough to satisfy watermarks. | |
2289 | */ | |
2290 | count_vm_event(COMPACTFAIL); | |
66199712 MG |
2291 | |
2292 | /* | |
2293 | * As async compaction considers a subset of pageblocks, only | |
2294 | * defer if the failure was a sync compaction failure. | |
2295 | */ | |
e0b9daeb | 2296 | if (mode != MIGRATE_ASYNC) |
aff62249 | 2297 | defer_compaction(preferred_zone, order); |
56de7263 MG |
2298 | |
2299 | cond_resched(); | |
2300 | } | |
2301 | ||
2302 | return NULL; | |
2303 | } | |
2304 | #else | |
2305 | static inline struct page * | |
2306 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2307 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2308 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
d8846374 MG |
2309 | int classzone_idx, int migratetype, |
2310 | enum migrate_mode mode, bool *contended_compaction, | |
e0b9daeb | 2311 | bool *deferred_compaction, unsigned long *did_some_progress) |
56de7263 MG |
2312 | { |
2313 | return NULL; | |
2314 | } | |
2315 | #endif /* CONFIG_COMPACTION */ | |
2316 | ||
bba90710 MS |
2317 | /* Perform direct synchronous page reclaim */ |
2318 | static int | |
2319 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, | |
2320 | nodemask_t *nodemask) | |
11e33f6a | 2321 | { |
11e33f6a | 2322 | struct reclaim_state reclaim_state; |
bba90710 | 2323 | int progress; |
11e33f6a MG |
2324 | |
2325 | cond_resched(); | |
2326 | ||
2327 | /* We now go into synchronous reclaim */ | |
2328 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2329 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2330 | lockdep_set_current_reclaim_state(gfp_mask); |
2331 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2332 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2333 | |
bba90710 | 2334 | progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
11e33f6a | 2335 | |
c06b1fca | 2336 | current->reclaim_state = NULL; |
11e33f6a | 2337 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2338 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2339 | |
2340 | cond_resched(); | |
2341 | ||
bba90710 MS |
2342 | return progress; |
2343 | } | |
2344 | ||
2345 | /* The really slow allocator path where we enter direct reclaim */ | |
2346 | static inline struct page * | |
2347 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
2348 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2349 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
d8846374 | 2350 | int classzone_idx, int migratetype, unsigned long *did_some_progress) |
bba90710 MS |
2351 | { |
2352 | struct page *page = NULL; | |
2353 | bool drained = false; | |
2354 | ||
2355 | *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist, | |
2356 | nodemask); | |
9ee493ce MG |
2357 | if (unlikely(!(*did_some_progress))) |
2358 | return NULL; | |
11e33f6a | 2359 | |
76d3fbf8 | 2360 | /* After successful reclaim, reconsider all zones for allocation */ |
e5adfffc | 2361 | if (IS_ENABLED(CONFIG_NUMA)) |
76d3fbf8 MG |
2362 | zlc_clear_zones_full(zonelist); |
2363 | ||
9ee493ce MG |
2364 | retry: |
2365 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2366 | zonelist, high_zoneidx, |
cfd19c5a | 2367 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
d8846374 MG |
2368 | preferred_zone, classzone_idx, |
2369 | migratetype); | |
9ee493ce MG |
2370 | |
2371 | /* | |
2372 | * If an allocation failed after direct reclaim, it could be because | |
2373 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2374 | */ | |
2375 | if (!page && !drained) { | |
2376 | drain_all_pages(); | |
2377 | drained = true; | |
2378 | goto retry; | |
2379 | } | |
2380 | ||
11e33f6a MG |
2381 | return page; |
2382 | } | |
2383 | ||
1da177e4 | 2384 | /* |
11e33f6a MG |
2385 | * This is called in the allocator slow-path if the allocation request is of |
2386 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2387 | */ |
11e33f6a MG |
2388 | static inline struct page * |
2389 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
2390 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 | 2391 | nodemask_t *nodemask, struct zone *preferred_zone, |
d8846374 | 2392 | int classzone_idx, int migratetype) |
11e33f6a MG |
2393 | { |
2394 | struct page *page; | |
2395 | ||
2396 | do { | |
2397 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2398 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
d8846374 | 2399 | preferred_zone, classzone_idx, migratetype); |
11e33f6a MG |
2400 | |
2401 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 2402 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
2403 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2404 | ||
2405 | return page; | |
2406 | } | |
2407 | ||
3a025760 JW |
2408 | static void reset_alloc_batches(struct zonelist *zonelist, |
2409 | enum zone_type high_zoneidx, | |
2410 | struct zone *preferred_zone) | |
1da177e4 | 2411 | { |
dd1a239f MG |
2412 | struct zoneref *z; |
2413 | struct zone *zone; | |
1da177e4 | 2414 | |
81c0a2bb | 2415 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81c0a2bb JW |
2416 | /* |
2417 | * Only reset the batches of zones that were actually | |
3a025760 JW |
2418 | * considered in the fairness pass, we don't want to |
2419 | * trash fairness information for zones that are not | |
81c0a2bb JW |
2420 | * actually part of this zonelist's round-robin cycle. |
2421 | */ | |
fff4068c | 2422 | if (!zone_local(preferred_zone, zone)) |
81c0a2bb JW |
2423 | continue; |
2424 | mod_zone_page_state(zone, NR_ALLOC_BATCH, | |
3a025760 JW |
2425 | high_wmark_pages(zone) - low_wmark_pages(zone) - |
2426 | atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH])); | |
81c0a2bb | 2427 | } |
11e33f6a | 2428 | } |
cf40bd16 | 2429 | |
3a025760 JW |
2430 | static void wake_all_kswapds(unsigned int order, |
2431 | struct zonelist *zonelist, | |
2432 | enum zone_type high_zoneidx, | |
2433 | struct zone *preferred_zone) | |
2434 | { | |
2435 | struct zoneref *z; | |
2436 | struct zone *zone; | |
2437 | ||
2438 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) | |
2439 | wakeup_kswapd(zone, order, zone_idx(preferred_zone)); | |
2440 | } | |
2441 | ||
341ce06f PZ |
2442 | static inline int |
2443 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2444 | { | |
341ce06f | 2445 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
b104a35d | 2446 | const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD)); |
1da177e4 | 2447 | |
a56f57ff | 2448 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2449 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2450 | |
341ce06f PZ |
2451 | /* |
2452 | * The caller may dip into page reserves a bit more if the caller | |
2453 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2454 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
b104a35d | 2455 | * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH). |
341ce06f | 2456 | */ |
e6223a3b | 2457 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2458 | |
b104a35d | 2459 | if (atomic) { |
5c3240d9 | 2460 | /* |
b104a35d DR |
2461 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
2462 | * if it can't schedule. | |
5c3240d9 | 2463 | */ |
b104a35d | 2464 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
5c3240d9 | 2465 | alloc_flags |= ALLOC_HARDER; |
523b9458 | 2466 | /* |
b104a35d DR |
2467 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
2468 | * comment for __cpuset_node_allowed_softwall(). | |
523b9458 | 2469 | */ |
341ce06f | 2470 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2471 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2472 | alloc_flags |= ALLOC_HARDER; |
2473 | ||
b37f1dd0 MG |
2474 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
2475 | if (gfp_mask & __GFP_MEMALLOC) | |
2476 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
2477 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
2478 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
2479 | else if (!in_interrupt() && | |
2480 | ((current->flags & PF_MEMALLOC) || | |
2481 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 2482 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 2483 | } |
d95ea5d1 BZ |
2484 | #ifdef CONFIG_CMA |
2485 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2486 | alloc_flags |= ALLOC_CMA; | |
2487 | #endif | |
341ce06f PZ |
2488 | return alloc_flags; |
2489 | } | |
2490 | ||
072bb0aa MG |
2491 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
2492 | { | |
b37f1dd0 | 2493 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
2494 | } |
2495 | ||
11e33f6a MG |
2496 | static inline struct page * |
2497 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2498 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 | 2499 | nodemask_t *nodemask, struct zone *preferred_zone, |
d8846374 | 2500 | int classzone_idx, int migratetype) |
11e33f6a MG |
2501 | { |
2502 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2503 | struct page *page = NULL; | |
2504 | int alloc_flags; | |
2505 | unsigned long pages_reclaimed = 0; | |
2506 | unsigned long did_some_progress; | |
e0b9daeb | 2507 | enum migrate_mode migration_mode = MIGRATE_ASYNC; |
66199712 | 2508 | bool deferred_compaction = false; |
c67fe375 | 2509 | bool contended_compaction = false; |
1da177e4 | 2510 | |
72807a74 MG |
2511 | /* |
2512 | * In the slowpath, we sanity check order to avoid ever trying to | |
2513 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2514 | * be using allocators in order of preference for an area that is | |
2515 | * too large. | |
2516 | */ | |
1fc28b70 MG |
2517 | if (order >= MAX_ORDER) { |
2518 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2519 | return NULL; |
1fc28b70 | 2520 | } |
1da177e4 | 2521 | |
952f3b51 CL |
2522 | /* |
2523 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2524 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2525 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2526 | * using a larger set of nodes after it has established that the | |
2527 | * allowed per node queues are empty and that nodes are | |
2528 | * over allocated. | |
2529 | */ | |
3a025760 JW |
2530 | if (IS_ENABLED(CONFIG_NUMA) && |
2531 | (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
952f3b51 CL |
2532 | goto nopage; |
2533 | ||
cc4a6851 | 2534 | restart: |
3a025760 JW |
2535 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
2536 | wake_all_kswapds(order, zonelist, high_zoneidx, preferred_zone); | |
1da177e4 | 2537 | |
9bf2229f | 2538 | /* |
7fb1d9fc RS |
2539 | * OK, we're below the kswapd watermark and have kicked background |
2540 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2541 | * to how we want to proceed. | |
9bf2229f | 2542 | */ |
341ce06f | 2543 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2544 | |
f33261d7 DR |
2545 | /* |
2546 | * Find the true preferred zone if the allocation is unconstrained by | |
2547 | * cpusets. | |
2548 | */ | |
d8846374 MG |
2549 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) { |
2550 | struct zoneref *preferred_zoneref; | |
2551 | preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx, | |
2552 | NULL, &preferred_zone); | |
2553 | classzone_idx = zonelist_zone_idx(preferred_zoneref); | |
2554 | } | |
f33261d7 | 2555 | |
cfa54a0f | 2556 | rebalance: |
341ce06f | 2557 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2558 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f | 2559 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
d8846374 | 2560 | preferred_zone, classzone_idx, migratetype); |
7fb1d9fc RS |
2561 | if (page) |
2562 | goto got_pg; | |
1da177e4 | 2563 | |
11e33f6a | 2564 | /* Allocate without watermarks if the context allows */ |
341ce06f | 2565 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
2566 | /* |
2567 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
2568 | * the allocation is high priority and these type of | |
2569 | * allocations are system rather than user orientated | |
2570 | */ | |
2571 | zonelist = node_zonelist(numa_node_id(), gfp_mask); | |
2572 | ||
341ce06f PZ |
2573 | page = __alloc_pages_high_priority(gfp_mask, order, |
2574 | zonelist, high_zoneidx, nodemask, | |
d8846374 | 2575 | preferred_zone, classzone_idx, migratetype); |
cfd19c5a | 2576 | if (page) { |
341ce06f | 2577 | goto got_pg; |
cfd19c5a | 2578 | } |
1da177e4 LT |
2579 | } |
2580 | ||
2581 | /* Atomic allocations - we can't balance anything */ | |
aed0a0e3 DR |
2582 | if (!wait) { |
2583 | /* | |
2584 | * All existing users of the deprecated __GFP_NOFAIL are | |
2585 | * blockable, so warn of any new users that actually allow this | |
2586 | * type of allocation to fail. | |
2587 | */ | |
2588 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); | |
1da177e4 | 2589 | goto nopage; |
aed0a0e3 | 2590 | } |
1da177e4 | 2591 | |
341ce06f | 2592 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2593 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2594 | goto nopage; |
2595 | ||
6583bb64 DR |
2596 | /* Avoid allocations with no watermarks from looping endlessly */ |
2597 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2598 | goto nopage; | |
2599 | ||
77f1fe6b MG |
2600 | /* |
2601 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2602 | * attempts after direct reclaim are synchronous | |
2603 | */ | |
e0b9daeb DR |
2604 | page = __alloc_pages_direct_compact(gfp_mask, order, zonelist, |
2605 | high_zoneidx, nodemask, alloc_flags, | |
d8846374 MG |
2606 | preferred_zone, |
2607 | classzone_idx, migratetype, | |
e0b9daeb | 2608 | migration_mode, &contended_compaction, |
66199712 MG |
2609 | &deferred_compaction, |
2610 | &did_some_progress); | |
56de7263 MG |
2611 | if (page) |
2612 | goto got_pg; | |
75f30861 DR |
2613 | |
2614 | /* | |
2615 | * It can become very expensive to allocate transparent hugepages at | |
2616 | * fault, so use asynchronous memory compaction for THP unless it is | |
2617 | * khugepaged trying to collapse. | |
2618 | */ | |
2619 | if (!(gfp_mask & __GFP_NO_KSWAPD) || (current->flags & PF_KTHREAD)) | |
2620 | migration_mode = MIGRATE_SYNC_LIGHT; | |
56de7263 | 2621 | |
31f8d42d LT |
2622 | /* |
2623 | * If compaction is deferred for high-order allocations, it is because | |
2624 | * sync compaction recently failed. In this is the case and the caller | |
2625 | * requested a movable allocation that does not heavily disrupt the | |
2626 | * system then fail the allocation instead of entering direct reclaim. | |
2627 | */ | |
2628 | if ((deferred_compaction || contended_compaction) && | |
caf49191 | 2629 | (gfp_mask & __GFP_NO_KSWAPD)) |
31f8d42d | 2630 | goto nopage; |
66199712 | 2631 | |
11e33f6a MG |
2632 | /* Try direct reclaim and then allocating */ |
2633 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2634 | zonelist, high_zoneidx, | |
2635 | nodemask, | |
5117f45d | 2636 | alloc_flags, preferred_zone, |
d8846374 MG |
2637 | classzone_idx, migratetype, |
2638 | &did_some_progress); | |
11e33f6a MG |
2639 | if (page) |
2640 | goto got_pg; | |
1da177e4 | 2641 | |
e33c3b5e | 2642 | /* |
11e33f6a MG |
2643 | * If we failed to make any progress reclaiming, then we are |
2644 | * running out of options and have to consider going OOM | |
e33c3b5e | 2645 | */ |
11e33f6a | 2646 | if (!did_some_progress) { |
b9921ecd | 2647 | if (oom_gfp_allowed(gfp_mask)) { |
7f33d49a RW |
2648 | if (oom_killer_disabled) |
2649 | goto nopage; | |
29fd66d2 DR |
2650 | /* Coredumps can quickly deplete all memory reserves */ |
2651 | if ((current->flags & PF_DUMPCORE) && | |
2652 | !(gfp_mask & __GFP_NOFAIL)) | |
2653 | goto nopage; | |
11e33f6a MG |
2654 | page = __alloc_pages_may_oom(gfp_mask, order, |
2655 | zonelist, high_zoneidx, | |
3dd28266 | 2656 | nodemask, preferred_zone, |
d8846374 | 2657 | classzone_idx, migratetype); |
11e33f6a MG |
2658 | if (page) |
2659 | goto got_pg; | |
1da177e4 | 2660 | |
03668b3c DR |
2661 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2662 | /* | |
2663 | * The oom killer is not called for high-order | |
2664 | * allocations that may fail, so if no progress | |
2665 | * is being made, there are no other options and | |
2666 | * retrying is unlikely to help. | |
2667 | */ | |
2668 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2669 | goto nopage; | |
2670 | /* | |
2671 | * The oom killer is not called for lowmem | |
2672 | * allocations to prevent needlessly killing | |
2673 | * innocent tasks. | |
2674 | */ | |
2675 | if (high_zoneidx < ZONE_NORMAL) | |
2676 | goto nopage; | |
2677 | } | |
e2c55dc8 | 2678 | |
ff0ceb9d DR |
2679 | goto restart; |
2680 | } | |
1da177e4 LT |
2681 | } |
2682 | ||
11e33f6a | 2683 | /* Check if we should retry the allocation */ |
a41f24ea | 2684 | pages_reclaimed += did_some_progress; |
f90ac398 MG |
2685 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
2686 | pages_reclaimed)) { | |
11e33f6a | 2687 | /* Wait for some write requests to complete then retry */ |
0e093d99 | 2688 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2689 | goto rebalance; |
3e7d3449 MG |
2690 | } else { |
2691 | /* | |
2692 | * High-order allocations do not necessarily loop after | |
2693 | * direct reclaim and reclaim/compaction depends on compaction | |
2694 | * being called after reclaim so call directly if necessary | |
2695 | */ | |
e0b9daeb DR |
2696 | page = __alloc_pages_direct_compact(gfp_mask, order, zonelist, |
2697 | high_zoneidx, nodemask, alloc_flags, | |
d8846374 MG |
2698 | preferred_zone, |
2699 | classzone_idx, migratetype, | |
e0b9daeb | 2700 | migration_mode, &contended_compaction, |
66199712 MG |
2701 | &deferred_compaction, |
2702 | &did_some_progress); | |
3e7d3449 MG |
2703 | if (page) |
2704 | goto got_pg; | |
1da177e4 LT |
2705 | } |
2706 | ||
2707 | nopage: | |
a238ab5b | 2708 | warn_alloc_failed(gfp_mask, order, NULL); |
b1eeab67 | 2709 | return page; |
1da177e4 | 2710 | got_pg: |
b1eeab67 VN |
2711 | if (kmemcheck_enabled) |
2712 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
11e33f6a | 2713 | |
072bb0aa | 2714 | return page; |
1da177e4 | 2715 | } |
11e33f6a MG |
2716 | |
2717 | /* | |
2718 | * This is the 'heart' of the zoned buddy allocator. | |
2719 | */ | |
2720 | struct page * | |
2721 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2722 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2723 | { | |
2724 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2725 | struct zone *preferred_zone; |
d8846374 | 2726 | struct zoneref *preferred_zoneref; |
cc9a6c87 | 2727 | struct page *page = NULL; |
3dd28266 | 2728 | int migratetype = allocflags_to_migratetype(gfp_mask); |
cc9a6c87 | 2729 | unsigned int cpuset_mems_cookie; |
3a025760 | 2730 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR; |
d8846374 | 2731 | int classzone_idx; |
11e33f6a | 2732 | |
dcce284a BH |
2733 | gfp_mask &= gfp_allowed_mask; |
2734 | ||
11e33f6a MG |
2735 | lockdep_trace_alloc(gfp_mask); |
2736 | ||
2737 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2738 | ||
2739 | if (should_fail_alloc_page(gfp_mask, order)) | |
2740 | return NULL; | |
2741 | ||
2742 | /* | |
2743 | * Check the zones suitable for the gfp_mask contain at least one | |
2744 | * valid zone. It's possible to have an empty zonelist as a result | |
2745 | * of GFP_THISNODE and a memoryless node | |
2746 | */ | |
2747 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2748 | return NULL; | |
2749 | ||
cc9a6c87 | 2750 | retry_cpuset: |
d26914d1 | 2751 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 2752 | |
5117f45d | 2753 | /* The preferred zone is used for statistics later */ |
d8846374 | 2754 | preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx, |
f33261d7 DR |
2755 | nodemask ? : &cpuset_current_mems_allowed, |
2756 | &preferred_zone); | |
cc9a6c87 MG |
2757 | if (!preferred_zone) |
2758 | goto out; | |
d8846374 | 2759 | classzone_idx = zonelist_zone_idx(preferred_zoneref); |
5117f45d | 2760 | |
d95ea5d1 BZ |
2761 | #ifdef CONFIG_CMA |
2762 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2763 | alloc_flags |= ALLOC_CMA; | |
2764 | #endif | |
3a025760 | 2765 | retry: |
5117f45d | 2766 | /* First allocation attempt */ |
11e33f6a | 2767 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
d95ea5d1 | 2768 | zonelist, high_zoneidx, alloc_flags, |
d8846374 | 2769 | preferred_zone, classzone_idx, migratetype); |
21caf2fc | 2770 | if (unlikely(!page)) { |
3a025760 JW |
2771 | /* |
2772 | * The first pass makes sure allocations are spread | |
2773 | * fairly within the local node. However, the local | |
2774 | * node might have free pages left after the fairness | |
2775 | * batches are exhausted, and remote zones haven't | |
2776 | * even been considered yet. Try once more without | |
2777 | * fairness, and include remote zones now, before | |
2778 | * entering the slowpath and waking kswapd: prefer | |
2779 | * spilling to a remote zone over swapping locally. | |
2780 | */ | |
2781 | if (alloc_flags & ALLOC_FAIR) { | |
2782 | reset_alloc_batches(zonelist, high_zoneidx, | |
2783 | preferred_zone); | |
2784 | alloc_flags &= ~ALLOC_FAIR; | |
2785 | goto retry; | |
2786 | } | |
21caf2fc ML |
2787 | /* |
2788 | * Runtime PM, block IO and its error handling path | |
2789 | * can deadlock because I/O on the device might not | |
2790 | * complete. | |
2791 | */ | |
2792 | gfp_mask = memalloc_noio_flags(gfp_mask); | |
11e33f6a | 2793 | page = __alloc_pages_slowpath(gfp_mask, order, |
5117f45d | 2794 | zonelist, high_zoneidx, nodemask, |
d8846374 | 2795 | preferred_zone, classzone_idx, migratetype); |
21caf2fc | 2796 | } |
11e33f6a | 2797 | |
4b4f278c | 2798 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
cc9a6c87 MG |
2799 | |
2800 | out: | |
2801 | /* | |
2802 | * When updating a task's mems_allowed, it is possible to race with | |
2803 | * parallel threads in such a way that an allocation can fail while | |
2804 | * the mask is being updated. If a page allocation is about to fail, | |
2805 | * check if the cpuset changed during allocation and if so, retry. | |
2806 | */ | |
d26914d1 | 2807 | if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 MG |
2808 | goto retry_cpuset; |
2809 | ||
11e33f6a | 2810 | return page; |
1da177e4 | 2811 | } |
d239171e | 2812 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2813 | |
2814 | /* | |
2815 | * Common helper functions. | |
2816 | */ | |
920c7a5d | 2817 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2818 | { |
945a1113 AM |
2819 | struct page *page; |
2820 | ||
2821 | /* | |
2822 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2823 | * a highmem page | |
2824 | */ | |
2825 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2826 | ||
1da177e4 LT |
2827 | page = alloc_pages(gfp_mask, order); |
2828 | if (!page) | |
2829 | return 0; | |
2830 | return (unsigned long) page_address(page); | |
2831 | } | |
1da177e4 LT |
2832 | EXPORT_SYMBOL(__get_free_pages); |
2833 | ||
920c7a5d | 2834 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2835 | { |
945a1113 | 2836 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2837 | } |
1da177e4 LT |
2838 | EXPORT_SYMBOL(get_zeroed_page); |
2839 | ||
920c7a5d | 2840 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2841 | { |
b5810039 | 2842 | if (put_page_testzero(page)) { |
1da177e4 | 2843 | if (order == 0) |
b745bc85 | 2844 | free_hot_cold_page(page, false); |
1da177e4 LT |
2845 | else |
2846 | __free_pages_ok(page, order); | |
2847 | } | |
2848 | } | |
2849 | ||
2850 | EXPORT_SYMBOL(__free_pages); | |
2851 | ||
920c7a5d | 2852 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2853 | { |
2854 | if (addr != 0) { | |
725d704e | 2855 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2856 | __free_pages(virt_to_page((void *)addr), order); |
2857 | } | |
2858 | } | |
2859 | ||
2860 | EXPORT_SYMBOL(free_pages); | |
2861 | ||
6a1a0d3b | 2862 | /* |
52383431 VD |
2863 | * alloc_kmem_pages charges newly allocated pages to the kmem resource counter |
2864 | * of the current memory cgroup. | |
6a1a0d3b | 2865 | * |
52383431 VD |
2866 | * It should be used when the caller would like to use kmalloc, but since the |
2867 | * allocation is large, it has to fall back to the page allocator. | |
2868 | */ | |
2869 | struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order) | |
2870 | { | |
2871 | struct page *page; | |
2872 | struct mem_cgroup *memcg = NULL; | |
2873 | ||
2874 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
2875 | return NULL; | |
2876 | page = alloc_pages(gfp_mask, order); | |
2877 | memcg_kmem_commit_charge(page, memcg, order); | |
2878 | return page; | |
2879 | } | |
2880 | ||
2881 | struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
2882 | { | |
2883 | struct page *page; | |
2884 | struct mem_cgroup *memcg = NULL; | |
2885 | ||
2886 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
2887 | return NULL; | |
2888 | page = alloc_pages_node(nid, gfp_mask, order); | |
2889 | memcg_kmem_commit_charge(page, memcg, order); | |
2890 | return page; | |
2891 | } | |
2892 | ||
2893 | /* | |
2894 | * __free_kmem_pages and free_kmem_pages will free pages allocated with | |
2895 | * alloc_kmem_pages. | |
6a1a0d3b | 2896 | */ |
52383431 | 2897 | void __free_kmem_pages(struct page *page, unsigned int order) |
6a1a0d3b GC |
2898 | { |
2899 | memcg_kmem_uncharge_pages(page, order); | |
2900 | __free_pages(page, order); | |
2901 | } | |
2902 | ||
52383431 | 2903 | void free_kmem_pages(unsigned long addr, unsigned int order) |
6a1a0d3b GC |
2904 | { |
2905 | if (addr != 0) { | |
2906 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
52383431 | 2907 | __free_kmem_pages(virt_to_page((void *)addr), order); |
6a1a0d3b GC |
2908 | } |
2909 | } | |
2910 | ||
ee85c2e1 AK |
2911 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
2912 | { | |
2913 | if (addr) { | |
2914 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2915 | unsigned long used = addr + PAGE_ALIGN(size); | |
2916 | ||
2917 | split_page(virt_to_page((void *)addr), order); | |
2918 | while (used < alloc_end) { | |
2919 | free_page(used); | |
2920 | used += PAGE_SIZE; | |
2921 | } | |
2922 | } | |
2923 | return (void *)addr; | |
2924 | } | |
2925 | ||
2be0ffe2 TT |
2926 | /** |
2927 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2928 | * @size: the number of bytes to allocate | |
2929 | * @gfp_mask: GFP flags for the allocation | |
2930 | * | |
2931 | * This function is similar to alloc_pages(), except that it allocates the | |
2932 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2933 | * allocate memory in power-of-two pages. | |
2934 | * | |
2935 | * This function is also limited by MAX_ORDER. | |
2936 | * | |
2937 | * Memory allocated by this function must be released by free_pages_exact(). | |
2938 | */ | |
2939 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2940 | { | |
2941 | unsigned int order = get_order(size); | |
2942 | unsigned long addr; | |
2943 | ||
2944 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 2945 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
2946 | } |
2947 | EXPORT_SYMBOL(alloc_pages_exact); | |
2948 | ||
ee85c2e1 AK |
2949 | /** |
2950 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
2951 | * pages on a node. | |
b5e6ab58 | 2952 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
2953 | * @size: the number of bytes to allocate |
2954 | * @gfp_mask: GFP flags for the allocation | |
2955 | * | |
2956 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
2957 | * back. | |
2958 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
2959 | * but is not exact. | |
2960 | */ | |
e1931811 | 2961 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
ee85c2e1 AK |
2962 | { |
2963 | unsigned order = get_order(size); | |
2964 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
2965 | if (!p) | |
2966 | return NULL; | |
2967 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
2968 | } | |
ee85c2e1 | 2969 | |
2be0ffe2 TT |
2970 | /** |
2971 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2972 | * @virt: the value returned by alloc_pages_exact. | |
2973 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2974 | * | |
2975 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2976 | */ | |
2977 | void free_pages_exact(void *virt, size_t size) | |
2978 | { | |
2979 | unsigned long addr = (unsigned long)virt; | |
2980 | unsigned long end = addr + PAGE_ALIGN(size); | |
2981 | ||
2982 | while (addr < end) { | |
2983 | free_page(addr); | |
2984 | addr += PAGE_SIZE; | |
2985 | } | |
2986 | } | |
2987 | EXPORT_SYMBOL(free_pages_exact); | |
2988 | ||
e0fb5815 ZY |
2989 | /** |
2990 | * nr_free_zone_pages - count number of pages beyond high watermark | |
2991 | * @offset: The zone index of the highest zone | |
2992 | * | |
2993 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
2994 | * high watermark within all zones at or below a given zone index. For each | |
2995 | * zone, the number of pages is calculated as: | |
834405c3 | 2996 | * managed_pages - high_pages |
e0fb5815 | 2997 | */ |
ebec3862 | 2998 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 2999 | { |
dd1a239f | 3000 | struct zoneref *z; |
54a6eb5c MG |
3001 | struct zone *zone; |
3002 | ||
e310fd43 | 3003 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 3004 | unsigned long sum = 0; |
1da177e4 | 3005 | |
0e88460d | 3006 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 3007 | |
54a6eb5c | 3008 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 3009 | unsigned long size = zone->managed_pages; |
41858966 | 3010 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
3011 | if (size > high) |
3012 | sum += size - high; | |
1da177e4 LT |
3013 | } |
3014 | ||
3015 | return sum; | |
3016 | } | |
3017 | ||
e0fb5815 ZY |
3018 | /** |
3019 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
3020 | * | |
3021 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
3022 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 3023 | */ |
ebec3862 | 3024 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 3025 | { |
af4ca457 | 3026 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 3027 | } |
c2f1a551 | 3028 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 3029 | |
e0fb5815 ZY |
3030 | /** |
3031 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
3032 | * | |
3033 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
3034 | * high watermark within all zones. | |
1da177e4 | 3035 | */ |
ebec3862 | 3036 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 3037 | { |
2a1e274a | 3038 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 3039 | } |
08e0f6a9 CL |
3040 | |
3041 | static inline void show_node(struct zone *zone) | |
1da177e4 | 3042 | { |
e5adfffc | 3043 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 3044 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 3045 | } |
1da177e4 | 3046 | |
1da177e4 LT |
3047 | void si_meminfo(struct sysinfo *val) |
3048 | { | |
3049 | val->totalram = totalram_pages; | |
3050 | val->sharedram = 0; | |
d23ad423 | 3051 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 3052 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
3053 | val->totalhigh = totalhigh_pages; |
3054 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
3055 | val->mem_unit = PAGE_SIZE; |
3056 | } | |
3057 | ||
3058 | EXPORT_SYMBOL(si_meminfo); | |
3059 | ||
3060 | #ifdef CONFIG_NUMA | |
3061 | void si_meminfo_node(struct sysinfo *val, int nid) | |
3062 | { | |
cdd91a77 JL |
3063 | int zone_type; /* needs to be signed */ |
3064 | unsigned long managed_pages = 0; | |
1da177e4 LT |
3065 | pg_data_t *pgdat = NODE_DATA(nid); |
3066 | ||
cdd91a77 JL |
3067 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
3068 | managed_pages += pgdat->node_zones[zone_type].managed_pages; | |
3069 | val->totalram = managed_pages; | |
d23ad423 | 3070 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 3071 | #ifdef CONFIG_HIGHMEM |
b40da049 | 3072 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages; |
d23ad423 CL |
3073 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
3074 | NR_FREE_PAGES); | |
98d2b0eb CL |
3075 | #else |
3076 | val->totalhigh = 0; | |
3077 | val->freehigh = 0; | |
3078 | #endif | |
1da177e4 LT |
3079 | val->mem_unit = PAGE_SIZE; |
3080 | } | |
3081 | #endif | |
3082 | ||
ddd588b5 | 3083 | /* |
7bf02ea2 DR |
3084 | * Determine whether the node should be displayed or not, depending on whether |
3085 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 3086 | */ |
7bf02ea2 | 3087 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
3088 | { |
3089 | bool ret = false; | |
cc9a6c87 | 3090 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
3091 | |
3092 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
3093 | goto out; | |
3094 | ||
cc9a6c87 | 3095 | do { |
d26914d1 | 3096 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 3097 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
d26914d1 | 3098 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
ddd588b5 DR |
3099 | out: |
3100 | return ret; | |
3101 | } | |
3102 | ||
1da177e4 LT |
3103 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
3104 | ||
377e4f16 RV |
3105 | static void show_migration_types(unsigned char type) |
3106 | { | |
3107 | static const char types[MIGRATE_TYPES] = { | |
3108 | [MIGRATE_UNMOVABLE] = 'U', | |
3109 | [MIGRATE_RECLAIMABLE] = 'E', | |
3110 | [MIGRATE_MOVABLE] = 'M', | |
3111 | [MIGRATE_RESERVE] = 'R', | |
3112 | #ifdef CONFIG_CMA | |
3113 | [MIGRATE_CMA] = 'C', | |
3114 | #endif | |
194159fb | 3115 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 3116 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 3117 | #endif |
377e4f16 RV |
3118 | }; |
3119 | char tmp[MIGRATE_TYPES + 1]; | |
3120 | char *p = tmp; | |
3121 | int i; | |
3122 | ||
3123 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
3124 | if (type & (1 << i)) | |
3125 | *p++ = types[i]; | |
3126 | } | |
3127 | ||
3128 | *p = '\0'; | |
3129 | printk("(%s) ", tmp); | |
3130 | } | |
3131 | ||
1da177e4 LT |
3132 | /* |
3133 | * Show free area list (used inside shift_scroll-lock stuff) | |
3134 | * We also calculate the percentage fragmentation. We do this by counting the | |
3135 | * memory on each free list with the exception of the first item on the list. | |
ddd588b5 DR |
3136 | * Suppresses nodes that are not allowed by current's cpuset if |
3137 | * SHOW_MEM_FILTER_NODES is passed. | |
1da177e4 | 3138 | */ |
7bf02ea2 | 3139 | void show_free_areas(unsigned int filter) |
1da177e4 | 3140 | { |
c7241913 | 3141 | int cpu; |
1da177e4 LT |
3142 | struct zone *zone; |
3143 | ||
ee99c71c | 3144 | for_each_populated_zone(zone) { |
7bf02ea2 | 3145 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3146 | continue; |
c7241913 JS |
3147 | show_node(zone); |
3148 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 3149 | |
6b482c67 | 3150 | for_each_online_cpu(cpu) { |
1da177e4 LT |
3151 | struct per_cpu_pageset *pageset; |
3152 | ||
99dcc3e5 | 3153 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 3154 | |
3dfa5721 CL |
3155 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
3156 | cpu, pageset->pcp.high, | |
3157 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
3158 | } |
3159 | } | |
3160 | ||
a731286d KM |
3161 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
3162 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 3163 | " unevictable:%lu" |
b76146ed | 3164 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 3165 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a BZ |
3166 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
3167 | " free_cma:%lu\n", | |
4f98a2fe | 3168 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 3169 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
3170 | global_page_state(NR_ISOLATED_ANON), |
3171 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 3172 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 3173 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 3174 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 3175 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 3176 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 3177 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 3178 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
3179 | global_page_state(NR_SLAB_RECLAIMABLE), |
3180 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 3181 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 3182 | global_page_state(NR_SHMEM), |
a25700a5 | 3183 | global_page_state(NR_PAGETABLE), |
d1ce749a BZ |
3184 | global_page_state(NR_BOUNCE), |
3185 | global_page_state(NR_FREE_CMA_PAGES)); | |
1da177e4 | 3186 | |
ee99c71c | 3187 | for_each_populated_zone(zone) { |
1da177e4 LT |
3188 | int i; |
3189 | ||
7bf02ea2 | 3190 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3191 | continue; |
1da177e4 LT |
3192 | show_node(zone); |
3193 | printk("%s" | |
3194 | " free:%lukB" | |
3195 | " min:%lukB" | |
3196 | " low:%lukB" | |
3197 | " high:%lukB" | |
4f98a2fe RR |
3198 | " active_anon:%lukB" |
3199 | " inactive_anon:%lukB" | |
3200 | " active_file:%lukB" | |
3201 | " inactive_file:%lukB" | |
7b854121 | 3202 | " unevictable:%lukB" |
a731286d KM |
3203 | " isolated(anon):%lukB" |
3204 | " isolated(file):%lukB" | |
1da177e4 | 3205 | " present:%lukB" |
9feedc9d | 3206 | " managed:%lukB" |
4a0aa73f KM |
3207 | " mlocked:%lukB" |
3208 | " dirty:%lukB" | |
3209 | " writeback:%lukB" | |
3210 | " mapped:%lukB" | |
4b02108a | 3211 | " shmem:%lukB" |
4a0aa73f KM |
3212 | " slab_reclaimable:%lukB" |
3213 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3214 | " kernel_stack:%lukB" |
4a0aa73f KM |
3215 | " pagetables:%lukB" |
3216 | " unstable:%lukB" | |
3217 | " bounce:%lukB" | |
d1ce749a | 3218 | " free_cma:%lukB" |
4a0aa73f | 3219 | " writeback_tmp:%lukB" |
1da177e4 LT |
3220 | " pages_scanned:%lu" |
3221 | " all_unreclaimable? %s" | |
3222 | "\n", | |
3223 | zone->name, | |
88f5acf8 | 3224 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3225 | K(min_wmark_pages(zone)), |
3226 | K(low_wmark_pages(zone)), | |
3227 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3228 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3229 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3230 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3231 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3232 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3233 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3234 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3235 | K(zone->present_pages), |
9feedc9d | 3236 | K(zone->managed_pages), |
4a0aa73f KM |
3237 | K(zone_page_state(zone, NR_MLOCK)), |
3238 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3239 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3240 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3241 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3242 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3243 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3244 | zone_page_state(zone, NR_KERNEL_STACK) * |
3245 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3246 | K(zone_page_state(zone, NR_PAGETABLE)), |
3247 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3248 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1ce749a | 3249 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3250 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
1da177e4 | 3251 | zone->pages_scanned, |
6e543d57 | 3252 | (!zone_reclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
3253 | ); |
3254 | printk("lowmem_reserve[]:"); | |
3255 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3256 | printk(" %lu", zone->lowmem_reserve[i]); | |
3257 | printk("\n"); | |
3258 | } | |
3259 | ||
ee99c71c | 3260 | for_each_populated_zone(zone) { |
b8af2941 | 3261 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
377e4f16 | 3262 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3263 | |
7bf02ea2 | 3264 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3265 | continue; |
1da177e4 LT |
3266 | show_node(zone); |
3267 | printk("%s: ", zone->name); | |
1da177e4 LT |
3268 | |
3269 | spin_lock_irqsave(&zone->lock, flags); | |
3270 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3271 | struct free_area *area = &zone->free_area[order]; |
3272 | int type; | |
3273 | ||
3274 | nr[order] = area->nr_free; | |
8f9de51a | 3275 | total += nr[order] << order; |
377e4f16 RV |
3276 | |
3277 | types[order] = 0; | |
3278 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3279 | if (!list_empty(&area->free_list[type])) | |
3280 | types[order] |= 1 << type; | |
3281 | } | |
1da177e4 LT |
3282 | } |
3283 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3284 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3285 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3286 | if (nr[order]) |
3287 | show_migration_types(types[order]); | |
3288 | } | |
1da177e4 LT |
3289 | printk("= %lukB\n", K(total)); |
3290 | } | |
3291 | ||
949f7ec5 DR |
3292 | hugetlb_show_meminfo(); |
3293 | ||
e6f3602d LW |
3294 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
3295 | ||
1da177e4 LT |
3296 | show_swap_cache_info(); |
3297 | } | |
3298 | ||
19770b32 MG |
3299 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
3300 | { | |
3301 | zoneref->zone = zone; | |
3302 | zoneref->zone_idx = zone_idx(zone); | |
3303 | } | |
3304 | ||
1da177e4 LT |
3305 | /* |
3306 | * Builds allocation fallback zone lists. | |
1a93205b CL |
3307 | * |
3308 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 3309 | */ |
f0c0b2b8 | 3310 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
bc732f1d | 3311 | int nr_zones) |
1da177e4 | 3312 | { |
1a93205b | 3313 | struct zone *zone; |
bc732f1d | 3314 | enum zone_type zone_type = MAX_NR_ZONES; |
02a68a5e CL |
3315 | |
3316 | do { | |
2f6726e5 | 3317 | zone_type--; |
070f8032 | 3318 | zone = pgdat->node_zones + zone_type; |
1a93205b | 3319 | if (populated_zone(zone)) { |
dd1a239f MG |
3320 | zoneref_set_zone(zone, |
3321 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 3322 | check_highest_zone(zone_type); |
1da177e4 | 3323 | } |
2f6726e5 | 3324 | } while (zone_type); |
bc732f1d | 3325 | |
070f8032 | 3326 | return nr_zones; |
1da177e4 LT |
3327 | } |
3328 | ||
f0c0b2b8 KH |
3329 | |
3330 | /* | |
3331 | * zonelist_order: | |
3332 | * 0 = automatic detection of better ordering. | |
3333 | * 1 = order by ([node] distance, -zonetype) | |
3334 | * 2 = order by (-zonetype, [node] distance) | |
3335 | * | |
3336 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
3337 | * the same zonelist. So only NUMA can configure this param. | |
3338 | */ | |
3339 | #define ZONELIST_ORDER_DEFAULT 0 | |
3340 | #define ZONELIST_ORDER_NODE 1 | |
3341 | #define ZONELIST_ORDER_ZONE 2 | |
3342 | ||
3343 | /* zonelist order in the kernel. | |
3344 | * set_zonelist_order() will set this to NODE or ZONE. | |
3345 | */ | |
3346 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3347 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
3348 | ||
3349 | ||
1da177e4 | 3350 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
3351 | /* The value user specified ....changed by config */ |
3352 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3353 | /* string for sysctl */ | |
3354 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
3355 | char numa_zonelist_order[16] = "default"; | |
3356 | ||
3357 | /* | |
3358 | * interface for configure zonelist ordering. | |
3359 | * command line option "numa_zonelist_order" | |
3360 | * = "[dD]efault - default, automatic configuration. | |
3361 | * = "[nN]ode - order by node locality, then by zone within node | |
3362 | * = "[zZ]one - order by zone, then by locality within zone | |
3363 | */ | |
3364 | ||
3365 | static int __parse_numa_zonelist_order(char *s) | |
3366 | { | |
3367 | if (*s == 'd' || *s == 'D') { | |
3368 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3369 | } else if (*s == 'n' || *s == 'N') { | |
3370 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
3371 | } else if (*s == 'z' || *s == 'Z') { | |
3372 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
3373 | } else { | |
3374 | printk(KERN_WARNING | |
3375 | "Ignoring invalid numa_zonelist_order value: " | |
3376 | "%s\n", s); | |
3377 | return -EINVAL; | |
3378 | } | |
3379 | return 0; | |
3380 | } | |
3381 | ||
3382 | static __init int setup_numa_zonelist_order(char *s) | |
3383 | { | |
ecb256f8 VL |
3384 | int ret; |
3385 | ||
3386 | if (!s) | |
3387 | return 0; | |
3388 | ||
3389 | ret = __parse_numa_zonelist_order(s); | |
3390 | if (ret == 0) | |
3391 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
3392 | ||
3393 | return ret; | |
f0c0b2b8 KH |
3394 | } |
3395 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
3396 | ||
3397 | /* | |
3398 | * sysctl handler for numa_zonelist_order | |
3399 | */ | |
cccad5b9 | 3400 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
8d65af78 | 3401 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
3402 | loff_t *ppos) |
3403 | { | |
3404 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
3405 | int ret; | |
443c6f14 | 3406 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 3407 | |
443c6f14 | 3408 | mutex_lock(&zl_order_mutex); |
dacbde09 CG |
3409 | if (write) { |
3410 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { | |
3411 | ret = -EINVAL; | |
3412 | goto out; | |
3413 | } | |
3414 | strcpy(saved_string, (char *)table->data); | |
3415 | } | |
8d65af78 | 3416 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 3417 | if (ret) |
443c6f14 | 3418 | goto out; |
f0c0b2b8 KH |
3419 | if (write) { |
3420 | int oldval = user_zonelist_order; | |
dacbde09 CG |
3421 | |
3422 | ret = __parse_numa_zonelist_order((char *)table->data); | |
3423 | if (ret) { | |
f0c0b2b8 KH |
3424 | /* |
3425 | * bogus value. restore saved string | |
3426 | */ | |
dacbde09 | 3427 | strncpy((char *)table->data, saved_string, |
f0c0b2b8 KH |
3428 | NUMA_ZONELIST_ORDER_LEN); |
3429 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
3430 | } else if (oldval != user_zonelist_order) { |
3431 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 3432 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
3433 | mutex_unlock(&zonelists_mutex); |
3434 | } | |
f0c0b2b8 | 3435 | } |
443c6f14 AK |
3436 | out: |
3437 | mutex_unlock(&zl_order_mutex); | |
3438 | return ret; | |
f0c0b2b8 KH |
3439 | } |
3440 | ||
3441 | ||
62bc62a8 | 3442 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
3443 | static int node_load[MAX_NUMNODES]; |
3444 | ||
1da177e4 | 3445 | /** |
4dc3b16b | 3446 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
3447 | * @node: node whose fallback list we're appending |
3448 | * @used_node_mask: nodemask_t of already used nodes | |
3449 | * | |
3450 | * We use a number of factors to determine which is the next node that should | |
3451 | * appear on a given node's fallback list. The node should not have appeared | |
3452 | * already in @node's fallback list, and it should be the next closest node | |
3453 | * according to the distance array (which contains arbitrary distance values | |
3454 | * from each node to each node in the system), and should also prefer nodes | |
3455 | * with no CPUs, since presumably they'll have very little allocation pressure | |
3456 | * on them otherwise. | |
3457 | * It returns -1 if no node is found. | |
3458 | */ | |
f0c0b2b8 | 3459 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 3460 | { |
4cf808eb | 3461 | int n, val; |
1da177e4 | 3462 | int min_val = INT_MAX; |
00ef2d2f | 3463 | int best_node = NUMA_NO_NODE; |
a70f7302 | 3464 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 3465 | |
4cf808eb LT |
3466 | /* Use the local node if we haven't already */ |
3467 | if (!node_isset(node, *used_node_mask)) { | |
3468 | node_set(node, *used_node_mask); | |
3469 | return node; | |
3470 | } | |
1da177e4 | 3471 | |
4b0ef1fe | 3472 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
3473 | |
3474 | /* Don't want a node to appear more than once */ | |
3475 | if (node_isset(n, *used_node_mask)) | |
3476 | continue; | |
3477 | ||
1da177e4 LT |
3478 | /* Use the distance array to find the distance */ |
3479 | val = node_distance(node, n); | |
3480 | ||
4cf808eb LT |
3481 | /* Penalize nodes under us ("prefer the next node") */ |
3482 | val += (n < node); | |
3483 | ||
1da177e4 | 3484 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
3485 | tmp = cpumask_of_node(n); |
3486 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
3487 | val += PENALTY_FOR_NODE_WITH_CPUS; |
3488 | ||
3489 | /* Slight preference for less loaded node */ | |
3490 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
3491 | val += node_load[n]; | |
3492 | ||
3493 | if (val < min_val) { | |
3494 | min_val = val; | |
3495 | best_node = n; | |
3496 | } | |
3497 | } | |
3498 | ||
3499 | if (best_node >= 0) | |
3500 | node_set(best_node, *used_node_mask); | |
3501 | ||
3502 | return best_node; | |
3503 | } | |
3504 | ||
f0c0b2b8 KH |
3505 | |
3506 | /* | |
3507 | * Build zonelists ordered by node and zones within node. | |
3508 | * This results in maximum locality--normal zone overflows into local | |
3509 | * DMA zone, if any--but risks exhausting DMA zone. | |
3510 | */ | |
3511 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 3512 | { |
f0c0b2b8 | 3513 | int j; |
1da177e4 | 3514 | struct zonelist *zonelist; |
f0c0b2b8 | 3515 | |
54a6eb5c | 3516 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 3517 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c | 3518 | ; |
bc732f1d | 3519 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
dd1a239f MG |
3520 | zonelist->_zonerefs[j].zone = NULL; |
3521 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
3522 | } |
3523 | ||
523b9458 CL |
3524 | /* |
3525 | * Build gfp_thisnode zonelists | |
3526 | */ | |
3527 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3528 | { | |
523b9458 CL |
3529 | int j; |
3530 | struct zonelist *zonelist; | |
3531 | ||
54a6eb5c | 3532 | zonelist = &pgdat->node_zonelists[1]; |
bc732f1d | 3533 | j = build_zonelists_node(pgdat, zonelist, 0); |
dd1a239f MG |
3534 | zonelist->_zonerefs[j].zone = NULL; |
3535 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3536 | } |
3537 | ||
f0c0b2b8 KH |
3538 | /* |
3539 | * Build zonelists ordered by zone and nodes within zones. | |
3540 | * This results in conserving DMA zone[s] until all Normal memory is | |
3541 | * exhausted, but results in overflowing to remote node while memory | |
3542 | * may still exist in local DMA zone. | |
3543 | */ | |
3544 | static int node_order[MAX_NUMNODES]; | |
3545 | ||
3546 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3547 | { | |
f0c0b2b8 KH |
3548 | int pos, j, node; |
3549 | int zone_type; /* needs to be signed */ | |
3550 | struct zone *z; | |
3551 | struct zonelist *zonelist; | |
3552 | ||
54a6eb5c MG |
3553 | zonelist = &pgdat->node_zonelists[0]; |
3554 | pos = 0; | |
3555 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3556 | for (j = 0; j < nr_nodes; j++) { | |
3557 | node = node_order[j]; | |
3558 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3559 | if (populated_zone(z)) { | |
dd1a239f MG |
3560 | zoneref_set_zone(z, |
3561 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3562 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3563 | } |
3564 | } | |
f0c0b2b8 | 3565 | } |
dd1a239f MG |
3566 | zonelist->_zonerefs[pos].zone = NULL; |
3567 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3568 | } |
3569 | ||
3570 | static int default_zonelist_order(void) | |
3571 | { | |
3572 | int nid, zone_type; | |
b8af2941 | 3573 | unsigned long low_kmem_size, total_size; |
f0c0b2b8 KH |
3574 | struct zone *z; |
3575 | int average_size; | |
3576 | /* | |
b8af2941 | 3577 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
f0c0b2b8 KH |
3578 | * If they are really small and used heavily, the system can fall |
3579 | * into OOM very easily. | |
e325c90f | 3580 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
f0c0b2b8 KH |
3581 | */ |
3582 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
3583 | low_kmem_size = 0; | |
3584 | total_size = 0; | |
3585 | for_each_online_node(nid) { | |
3586 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3587 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3588 | if (populated_zone(z)) { | |
3589 | if (zone_type < ZONE_NORMAL) | |
4f9f4774 JL |
3590 | low_kmem_size += z->managed_pages; |
3591 | total_size += z->managed_pages; | |
e325c90f DR |
3592 | } else if (zone_type == ZONE_NORMAL) { |
3593 | /* | |
3594 | * If any node has only lowmem, then node order | |
3595 | * is preferred to allow kernel allocations | |
3596 | * locally; otherwise, they can easily infringe | |
3597 | * on other nodes when there is an abundance of | |
3598 | * lowmem available to allocate from. | |
3599 | */ | |
3600 | return ZONELIST_ORDER_NODE; | |
f0c0b2b8 KH |
3601 | } |
3602 | } | |
3603 | } | |
3604 | if (!low_kmem_size || /* there are no DMA area. */ | |
3605 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
3606 | return ZONELIST_ORDER_NODE; | |
3607 | /* | |
3608 | * look into each node's config. | |
b8af2941 PK |
3609 | * If there is a node whose DMA/DMA32 memory is very big area on |
3610 | * local memory, NODE_ORDER may be suitable. | |
3611 | */ | |
37b07e41 | 3612 | average_size = total_size / |
4b0ef1fe | 3613 | (nodes_weight(node_states[N_MEMORY]) + 1); |
f0c0b2b8 KH |
3614 | for_each_online_node(nid) { |
3615 | low_kmem_size = 0; | |
3616 | total_size = 0; | |
3617 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3618 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3619 | if (populated_zone(z)) { | |
3620 | if (zone_type < ZONE_NORMAL) | |
3621 | low_kmem_size += z->present_pages; | |
3622 | total_size += z->present_pages; | |
3623 | } | |
3624 | } | |
3625 | if (low_kmem_size && | |
3626 | total_size > average_size && /* ignore small node */ | |
3627 | low_kmem_size > total_size * 70/100) | |
3628 | return ZONELIST_ORDER_NODE; | |
3629 | } | |
3630 | return ZONELIST_ORDER_ZONE; | |
3631 | } | |
3632 | ||
3633 | static void set_zonelist_order(void) | |
3634 | { | |
3635 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3636 | current_zonelist_order = default_zonelist_order(); | |
3637 | else | |
3638 | current_zonelist_order = user_zonelist_order; | |
3639 | } | |
3640 | ||
3641 | static void build_zonelists(pg_data_t *pgdat) | |
3642 | { | |
3643 | int j, node, load; | |
3644 | enum zone_type i; | |
1da177e4 | 3645 | nodemask_t used_mask; |
f0c0b2b8 KH |
3646 | int local_node, prev_node; |
3647 | struct zonelist *zonelist; | |
3648 | int order = current_zonelist_order; | |
1da177e4 LT |
3649 | |
3650 | /* initialize zonelists */ | |
523b9458 | 3651 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3652 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3653 | zonelist->_zonerefs[0].zone = NULL; |
3654 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3655 | } |
3656 | ||
3657 | /* NUMA-aware ordering of nodes */ | |
3658 | local_node = pgdat->node_id; | |
62bc62a8 | 3659 | load = nr_online_nodes; |
1da177e4 LT |
3660 | prev_node = local_node; |
3661 | nodes_clear(used_mask); | |
f0c0b2b8 | 3662 | |
f0c0b2b8 KH |
3663 | memset(node_order, 0, sizeof(node_order)); |
3664 | j = 0; | |
3665 | ||
1da177e4 LT |
3666 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
3667 | /* | |
3668 | * We don't want to pressure a particular node. | |
3669 | * So adding penalty to the first node in same | |
3670 | * distance group to make it round-robin. | |
3671 | */ | |
957f822a DR |
3672 | if (node_distance(local_node, node) != |
3673 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
3674 | node_load[node] = load; |
3675 | ||
1da177e4 LT |
3676 | prev_node = node; |
3677 | load--; | |
f0c0b2b8 KH |
3678 | if (order == ZONELIST_ORDER_NODE) |
3679 | build_zonelists_in_node_order(pgdat, node); | |
3680 | else | |
3681 | node_order[j++] = node; /* remember order */ | |
3682 | } | |
1da177e4 | 3683 | |
f0c0b2b8 KH |
3684 | if (order == ZONELIST_ORDER_ZONE) { |
3685 | /* calculate node order -- i.e., DMA last! */ | |
3686 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3687 | } |
523b9458 CL |
3688 | |
3689 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3690 | } |
3691 | ||
9276b1bc | 3692 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3693 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3694 | { |
54a6eb5c MG |
3695 | struct zonelist *zonelist; |
3696 | struct zonelist_cache *zlc; | |
dd1a239f | 3697 | struct zoneref *z; |
9276b1bc | 3698 | |
54a6eb5c MG |
3699 | zonelist = &pgdat->node_zonelists[0]; |
3700 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3701 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3702 | for (z = zonelist->_zonerefs; z->zone; z++) |
3703 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
3704 | } |
3705 | ||
7aac7898 LS |
3706 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3707 | /* | |
3708 | * Return node id of node used for "local" allocations. | |
3709 | * I.e., first node id of first zone in arg node's generic zonelist. | |
3710 | * Used for initializing percpu 'numa_mem', which is used primarily | |
3711 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
3712 | */ | |
3713 | int local_memory_node(int node) | |
3714 | { | |
3715 | struct zone *zone; | |
3716 | ||
3717 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
3718 | gfp_zone(GFP_KERNEL), | |
3719 | NULL, | |
3720 | &zone); | |
3721 | return zone->node; | |
3722 | } | |
3723 | #endif | |
f0c0b2b8 | 3724 | |
1da177e4 LT |
3725 | #else /* CONFIG_NUMA */ |
3726 | ||
f0c0b2b8 KH |
3727 | static void set_zonelist_order(void) |
3728 | { | |
3729 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
3730 | } | |
3731 | ||
3732 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 3733 | { |
19655d34 | 3734 | int node, local_node; |
54a6eb5c MG |
3735 | enum zone_type j; |
3736 | struct zonelist *zonelist; | |
1da177e4 LT |
3737 | |
3738 | local_node = pgdat->node_id; | |
1da177e4 | 3739 | |
54a6eb5c | 3740 | zonelist = &pgdat->node_zonelists[0]; |
bc732f1d | 3741 | j = build_zonelists_node(pgdat, zonelist, 0); |
1da177e4 | 3742 | |
54a6eb5c MG |
3743 | /* |
3744 | * Now we build the zonelist so that it contains the zones | |
3745 | * of all the other nodes. | |
3746 | * We don't want to pressure a particular node, so when | |
3747 | * building the zones for node N, we make sure that the | |
3748 | * zones coming right after the local ones are those from | |
3749 | * node N+1 (modulo N) | |
3750 | */ | |
3751 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3752 | if (!node_online(node)) | |
3753 | continue; | |
bc732f1d | 3754 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
1da177e4 | 3755 | } |
54a6eb5c MG |
3756 | for (node = 0; node < local_node; node++) { |
3757 | if (!node_online(node)) | |
3758 | continue; | |
bc732f1d | 3759 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
54a6eb5c MG |
3760 | } |
3761 | ||
dd1a239f MG |
3762 | zonelist->_zonerefs[j].zone = NULL; |
3763 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3764 | } |
3765 | ||
9276b1bc | 3766 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3767 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3768 | { |
54a6eb5c | 3769 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3770 | } |
3771 | ||
1da177e4 LT |
3772 | #endif /* CONFIG_NUMA */ |
3773 | ||
99dcc3e5 CL |
3774 | /* |
3775 | * Boot pageset table. One per cpu which is going to be used for all | |
3776 | * zones and all nodes. The parameters will be set in such a way | |
3777 | * that an item put on a list will immediately be handed over to | |
3778 | * the buddy list. This is safe since pageset manipulation is done | |
3779 | * with interrupts disabled. | |
3780 | * | |
3781 | * The boot_pagesets must be kept even after bootup is complete for | |
3782 | * unused processors and/or zones. They do play a role for bootstrapping | |
3783 | * hotplugged processors. | |
3784 | * | |
3785 | * zoneinfo_show() and maybe other functions do | |
3786 | * not check if the processor is online before following the pageset pointer. | |
3787 | * Other parts of the kernel may not check if the zone is available. | |
3788 | */ | |
3789 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3790 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3791 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3792 | |
4eaf3f64 HL |
3793 | /* |
3794 | * Global mutex to protect against size modification of zonelists | |
3795 | * as well as to serialize pageset setup for the new populated zone. | |
3796 | */ | |
3797 | DEFINE_MUTEX(zonelists_mutex); | |
3798 | ||
9b1a4d38 | 3799 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 3800 | static int __build_all_zonelists(void *data) |
1da177e4 | 3801 | { |
6811378e | 3802 | int nid; |
99dcc3e5 | 3803 | int cpu; |
9adb62a5 | 3804 | pg_data_t *self = data; |
9276b1bc | 3805 | |
7f9cfb31 BL |
3806 | #ifdef CONFIG_NUMA |
3807 | memset(node_load, 0, sizeof(node_load)); | |
3808 | #endif | |
9adb62a5 JL |
3809 | |
3810 | if (self && !node_online(self->node_id)) { | |
3811 | build_zonelists(self); | |
3812 | build_zonelist_cache(self); | |
3813 | } | |
3814 | ||
9276b1bc | 3815 | for_each_online_node(nid) { |
7ea1530a CL |
3816 | pg_data_t *pgdat = NODE_DATA(nid); |
3817 | ||
3818 | build_zonelists(pgdat); | |
3819 | build_zonelist_cache(pgdat); | |
9276b1bc | 3820 | } |
99dcc3e5 CL |
3821 | |
3822 | /* | |
3823 | * Initialize the boot_pagesets that are going to be used | |
3824 | * for bootstrapping processors. The real pagesets for | |
3825 | * each zone will be allocated later when the per cpu | |
3826 | * allocator is available. | |
3827 | * | |
3828 | * boot_pagesets are used also for bootstrapping offline | |
3829 | * cpus if the system is already booted because the pagesets | |
3830 | * are needed to initialize allocators on a specific cpu too. | |
3831 | * F.e. the percpu allocator needs the page allocator which | |
3832 | * needs the percpu allocator in order to allocate its pagesets | |
3833 | * (a chicken-egg dilemma). | |
3834 | */ | |
7aac7898 | 3835 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3836 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3837 | ||
7aac7898 LS |
3838 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3839 | /* | |
3840 | * We now know the "local memory node" for each node-- | |
3841 | * i.e., the node of the first zone in the generic zonelist. | |
3842 | * Set up numa_mem percpu variable for on-line cpus. During | |
3843 | * boot, only the boot cpu should be on-line; we'll init the | |
3844 | * secondary cpus' numa_mem as they come on-line. During | |
3845 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3846 | */ | |
3847 | if (cpu_online(cpu)) | |
3848 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3849 | #endif | |
3850 | } | |
3851 | ||
6811378e YG |
3852 | return 0; |
3853 | } | |
3854 | ||
4eaf3f64 HL |
3855 | /* |
3856 | * Called with zonelists_mutex held always | |
3857 | * unless system_state == SYSTEM_BOOTING. | |
3858 | */ | |
9adb62a5 | 3859 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 3860 | { |
f0c0b2b8 KH |
3861 | set_zonelist_order(); |
3862 | ||
6811378e | 3863 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3864 | __build_all_zonelists(NULL); |
68ad8df4 | 3865 | mminit_verify_zonelist(); |
6811378e YG |
3866 | cpuset_init_current_mems_allowed(); |
3867 | } else { | |
e9959f0f | 3868 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
3869 | if (zone) |
3870 | setup_zone_pageset(zone); | |
e9959f0f | 3871 | #endif |
dd1895e2 CS |
3872 | /* we have to stop all cpus to guarantee there is no user |
3873 | of zonelist */ | |
9adb62a5 | 3874 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
3875 | /* cpuset refresh routine should be here */ |
3876 | } | |
bd1e22b8 | 3877 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3878 | /* |
3879 | * Disable grouping by mobility if the number of pages in the | |
3880 | * system is too low to allow the mechanism to work. It would be | |
3881 | * more accurate, but expensive to check per-zone. This check is | |
3882 | * made on memory-hotadd so a system can start with mobility | |
3883 | * disabled and enable it later | |
3884 | */ | |
d9c23400 | 3885 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3886 | page_group_by_mobility_disabled = 1; |
3887 | else | |
3888 | page_group_by_mobility_disabled = 0; | |
3889 | ||
3890 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
3891 | "Total pages: %ld\n", | |
62bc62a8 | 3892 | nr_online_nodes, |
f0c0b2b8 | 3893 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3894 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3895 | vm_total_pages); |
3896 | #ifdef CONFIG_NUMA | |
3897 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
3898 | #endif | |
1da177e4 LT |
3899 | } |
3900 | ||
3901 | /* | |
3902 | * Helper functions to size the waitqueue hash table. | |
3903 | * Essentially these want to choose hash table sizes sufficiently | |
3904 | * large so that collisions trying to wait on pages are rare. | |
3905 | * But in fact, the number of active page waitqueues on typical | |
3906 | * systems is ridiculously low, less than 200. So this is even | |
3907 | * conservative, even though it seems large. | |
3908 | * | |
3909 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
3910 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
3911 | */ | |
3912 | #define PAGES_PER_WAITQUEUE 256 | |
3913 | ||
cca448fe | 3914 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 3915 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
3916 | { |
3917 | unsigned long size = 1; | |
3918 | ||
3919 | pages /= PAGES_PER_WAITQUEUE; | |
3920 | ||
3921 | while (size < pages) | |
3922 | size <<= 1; | |
3923 | ||
3924 | /* | |
3925 | * Once we have dozens or even hundreds of threads sleeping | |
3926 | * on IO we've got bigger problems than wait queue collision. | |
3927 | * Limit the size of the wait table to a reasonable size. | |
3928 | */ | |
3929 | size = min(size, 4096UL); | |
3930 | ||
3931 | return max(size, 4UL); | |
3932 | } | |
cca448fe YG |
3933 | #else |
3934 | /* | |
3935 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
3936 | * a suitable size for its wait_table. So we use the maximum size now. | |
3937 | * | |
3938 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
3939 | * | |
3940 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
3941 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
3942 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
3943 | * | |
3944 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
3945 | * or more by the traditional way. (See above). It equals: | |
3946 | * | |
3947 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
3948 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
3949 | * powerpc (64K page size) : = (32G +16M)byte. | |
3950 | */ | |
3951 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
3952 | { | |
3953 | return 4096UL; | |
3954 | } | |
3955 | #endif | |
1da177e4 LT |
3956 | |
3957 | /* | |
3958 | * This is an integer logarithm so that shifts can be used later | |
3959 | * to extract the more random high bits from the multiplicative | |
3960 | * hash function before the remainder is taken. | |
3961 | */ | |
3962 | static inline unsigned long wait_table_bits(unsigned long size) | |
3963 | { | |
3964 | return ffz(~size); | |
3965 | } | |
3966 | ||
6d3163ce AH |
3967 | /* |
3968 | * Check if a pageblock contains reserved pages | |
3969 | */ | |
3970 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
3971 | { | |
3972 | unsigned long pfn; | |
3973 | ||
3974 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
3975 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
3976 | return 1; | |
3977 | } | |
3978 | return 0; | |
3979 | } | |
3980 | ||
56fd56b8 | 3981 | /* |
d9c23400 | 3982 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
3983 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
3984 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
3985 | * higher will lead to a bigger reserve which will get freed as contiguous |
3986 | * blocks as reclaim kicks in | |
3987 | */ | |
3988 | static void setup_zone_migrate_reserve(struct zone *zone) | |
3989 | { | |
6d3163ce | 3990 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 3991 | struct page *page; |
78986a67 MG |
3992 | unsigned long block_migratetype; |
3993 | int reserve; | |
943dca1a | 3994 | int old_reserve; |
56fd56b8 | 3995 | |
d0215638 MH |
3996 | /* |
3997 | * Get the start pfn, end pfn and the number of blocks to reserve | |
3998 | * We have to be careful to be aligned to pageblock_nr_pages to | |
3999 | * make sure that we always check pfn_valid for the first page in | |
4000 | * the block. | |
4001 | */ | |
56fd56b8 | 4002 | start_pfn = zone->zone_start_pfn; |
108bcc96 | 4003 | end_pfn = zone_end_pfn(zone); |
d0215638 | 4004 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 4005 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 4006 | pageblock_order; |
56fd56b8 | 4007 | |
78986a67 MG |
4008 | /* |
4009 | * Reserve blocks are generally in place to help high-order atomic | |
4010 | * allocations that are short-lived. A min_free_kbytes value that | |
4011 | * would result in more than 2 reserve blocks for atomic allocations | |
4012 | * is assumed to be in place to help anti-fragmentation for the | |
4013 | * future allocation of hugepages at runtime. | |
4014 | */ | |
4015 | reserve = min(2, reserve); | |
943dca1a YI |
4016 | old_reserve = zone->nr_migrate_reserve_block; |
4017 | ||
4018 | /* When memory hot-add, we almost always need to do nothing */ | |
4019 | if (reserve == old_reserve) | |
4020 | return; | |
4021 | zone->nr_migrate_reserve_block = reserve; | |
78986a67 | 4022 | |
d9c23400 | 4023 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
4024 | if (!pfn_valid(pfn)) |
4025 | continue; | |
4026 | page = pfn_to_page(pfn); | |
4027 | ||
344c790e AL |
4028 | /* Watch out for overlapping nodes */ |
4029 | if (page_to_nid(page) != zone_to_nid(zone)) | |
4030 | continue; | |
4031 | ||
56fd56b8 MG |
4032 | block_migratetype = get_pageblock_migratetype(page); |
4033 | ||
938929f1 MG |
4034 | /* Only test what is necessary when the reserves are not met */ |
4035 | if (reserve > 0) { | |
4036 | /* | |
4037 | * Blocks with reserved pages will never free, skip | |
4038 | * them. | |
4039 | */ | |
4040 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
4041 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
4042 | continue; | |
56fd56b8 | 4043 | |
938929f1 MG |
4044 | /* If this block is reserved, account for it */ |
4045 | if (block_migratetype == MIGRATE_RESERVE) { | |
4046 | reserve--; | |
4047 | continue; | |
4048 | } | |
4049 | ||
4050 | /* Suitable for reserving if this block is movable */ | |
4051 | if (block_migratetype == MIGRATE_MOVABLE) { | |
4052 | set_pageblock_migratetype(page, | |
4053 | MIGRATE_RESERVE); | |
4054 | move_freepages_block(zone, page, | |
4055 | MIGRATE_RESERVE); | |
4056 | reserve--; | |
4057 | continue; | |
4058 | } | |
943dca1a YI |
4059 | } else if (!old_reserve) { |
4060 | /* | |
4061 | * At boot time we don't need to scan the whole zone | |
4062 | * for turning off MIGRATE_RESERVE. | |
4063 | */ | |
4064 | break; | |
56fd56b8 MG |
4065 | } |
4066 | ||
4067 | /* | |
4068 | * If the reserve is met and this is a previous reserved block, | |
4069 | * take it back | |
4070 | */ | |
4071 | if (block_migratetype == MIGRATE_RESERVE) { | |
4072 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4073 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
4074 | } | |
4075 | } | |
4076 | } | |
ac0e5b7a | 4077 | |
1da177e4 LT |
4078 | /* |
4079 | * Initially all pages are reserved - free ones are freed | |
4080 | * up by free_all_bootmem() once the early boot process is | |
4081 | * done. Non-atomic initialization, single-pass. | |
4082 | */ | |
c09b4240 | 4083 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 4084 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 4085 | { |
1da177e4 | 4086 | struct page *page; |
29751f69 AW |
4087 | unsigned long end_pfn = start_pfn + size; |
4088 | unsigned long pfn; | |
86051ca5 | 4089 | struct zone *z; |
1da177e4 | 4090 | |
22b31eec HD |
4091 | if (highest_memmap_pfn < end_pfn - 1) |
4092 | highest_memmap_pfn = end_pfn - 1; | |
4093 | ||
86051ca5 | 4094 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 4095 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
4096 | /* |
4097 | * There can be holes in boot-time mem_map[]s | |
4098 | * handed to this function. They do not | |
4099 | * exist on hotplugged memory. | |
4100 | */ | |
4101 | if (context == MEMMAP_EARLY) { | |
4102 | if (!early_pfn_valid(pfn)) | |
4103 | continue; | |
4104 | if (!early_pfn_in_nid(pfn, nid)) | |
4105 | continue; | |
4106 | } | |
d41dee36 AW |
4107 | page = pfn_to_page(pfn); |
4108 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 4109 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 4110 | init_page_count(page); |
22b751c3 | 4111 | page_mapcount_reset(page); |
90572890 | 4112 | page_cpupid_reset_last(page); |
1da177e4 | 4113 | SetPageReserved(page); |
b2a0ac88 MG |
4114 | /* |
4115 | * Mark the block movable so that blocks are reserved for | |
4116 | * movable at startup. This will force kernel allocations | |
4117 | * to reserve their blocks rather than leaking throughout | |
4118 | * the address space during boot when many long-lived | |
56fd56b8 MG |
4119 | * kernel allocations are made. Later some blocks near |
4120 | * the start are marked MIGRATE_RESERVE by | |
4121 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
4122 | * |
4123 | * bitmap is created for zone's valid pfn range. but memmap | |
4124 | * can be created for invalid pages (for alignment) | |
4125 | * check here not to call set_pageblock_migratetype() against | |
4126 | * pfn out of zone. | |
b2a0ac88 | 4127 | */ |
86051ca5 | 4128 | if ((z->zone_start_pfn <= pfn) |
108bcc96 | 4129 | && (pfn < zone_end_pfn(z)) |
86051ca5 | 4130 | && !(pfn & (pageblock_nr_pages - 1))) |
56fd56b8 | 4131 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 4132 | |
1da177e4 LT |
4133 | INIT_LIST_HEAD(&page->lru); |
4134 | #ifdef WANT_PAGE_VIRTUAL | |
4135 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
4136 | if (!is_highmem_idx(zone)) | |
3212c6be | 4137 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 4138 | #endif |
1da177e4 LT |
4139 | } |
4140 | } | |
4141 | ||
1e548deb | 4142 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 4143 | { |
7aeb09f9 | 4144 | unsigned int order, t; |
b2a0ac88 MG |
4145 | for_each_migratetype_order(order, t) { |
4146 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
4147 | zone->free_area[order].nr_free = 0; |
4148 | } | |
4149 | } | |
4150 | ||
4151 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
4152 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 4153 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
4154 | #endif |
4155 | ||
7cd2b0a3 | 4156 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 | 4157 | { |
3a6be87f | 4158 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
4159 | int batch; |
4160 | ||
4161 | /* | |
4162 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 4163 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
4164 | * |
4165 | * OK, so we don't know how big the cache is. So guess. | |
4166 | */ | |
b40da049 | 4167 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
4168 | if (batch * PAGE_SIZE > 512 * 1024) |
4169 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
4170 | batch /= 4; /* We effectively *= 4 below */ |
4171 | if (batch < 1) | |
4172 | batch = 1; | |
4173 | ||
4174 | /* | |
0ceaacc9 NP |
4175 | * Clamp the batch to a 2^n - 1 value. Having a power |
4176 | * of 2 value was found to be more likely to have | |
4177 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 4178 | * |
0ceaacc9 NP |
4179 | * For example if 2 tasks are alternately allocating |
4180 | * batches of pages, one task can end up with a lot | |
4181 | * of pages of one half of the possible page colors | |
4182 | * and the other with pages of the other colors. | |
e7c8d5c9 | 4183 | */ |
9155203a | 4184 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 4185 | |
e7c8d5c9 | 4186 | return batch; |
3a6be87f DH |
4187 | |
4188 | #else | |
4189 | /* The deferral and batching of frees should be suppressed under NOMMU | |
4190 | * conditions. | |
4191 | * | |
4192 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4193 | * of contiguous memory as there's no hardware page translation to | |
4194 | * assemble apparent contiguous memory from discontiguous pages. | |
4195 | * | |
4196 | * Queueing large contiguous runs of pages for batching, however, | |
4197 | * causes the pages to actually be freed in smaller chunks. As there | |
4198 | * can be a significant delay between the individual batches being | |
4199 | * recycled, this leads to the once large chunks of space being | |
4200 | * fragmented and becoming unavailable for high-order allocations. | |
4201 | */ | |
4202 | return 0; | |
4203 | #endif | |
e7c8d5c9 CL |
4204 | } |
4205 | ||
8d7a8fa9 CS |
4206 | /* |
4207 | * pcp->high and pcp->batch values are related and dependent on one another: | |
4208 | * ->batch must never be higher then ->high. | |
4209 | * The following function updates them in a safe manner without read side | |
4210 | * locking. | |
4211 | * | |
4212 | * Any new users of pcp->batch and pcp->high should ensure they can cope with | |
4213 | * those fields changing asynchronously (acording the the above rule). | |
4214 | * | |
4215 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function | |
4216 | * outside of boot time (or some other assurance that no concurrent updaters | |
4217 | * exist). | |
4218 | */ | |
4219 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, | |
4220 | unsigned long batch) | |
4221 | { | |
4222 | /* start with a fail safe value for batch */ | |
4223 | pcp->batch = 1; | |
4224 | smp_wmb(); | |
4225 | ||
4226 | /* Update high, then batch, in order */ | |
4227 | pcp->high = high; | |
4228 | smp_wmb(); | |
4229 | ||
4230 | pcp->batch = batch; | |
4231 | } | |
4232 | ||
3664033c | 4233 | /* a companion to pageset_set_high() */ |
4008bab7 CS |
4234 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
4235 | { | |
8d7a8fa9 | 4236 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
4008bab7 CS |
4237 | } |
4238 | ||
88c90dbc | 4239 | static void pageset_init(struct per_cpu_pageset *p) |
2caaad41 CL |
4240 | { |
4241 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4242 | int migratetype; |
2caaad41 | 4243 | |
1c6fe946 MD |
4244 | memset(p, 0, sizeof(*p)); |
4245 | ||
3dfa5721 | 4246 | pcp = &p->pcp; |
2caaad41 | 4247 | pcp->count = 0; |
5f8dcc21 MG |
4248 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4249 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4250 | } |
4251 | ||
88c90dbc CS |
4252 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
4253 | { | |
4254 | pageset_init(p); | |
4255 | pageset_set_batch(p, batch); | |
4256 | } | |
4257 | ||
8ad4b1fb | 4258 | /* |
3664033c | 4259 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
8ad4b1fb RS |
4260 | * to the value high for the pageset p. |
4261 | */ | |
3664033c | 4262 | static void pageset_set_high(struct per_cpu_pageset *p, |
8ad4b1fb RS |
4263 | unsigned long high) |
4264 | { | |
8d7a8fa9 CS |
4265 | unsigned long batch = max(1UL, high / 4); |
4266 | if ((high / 4) > (PAGE_SHIFT * 8)) | |
4267 | batch = PAGE_SHIFT * 8; | |
8ad4b1fb | 4268 | |
8d7a8fa9 | 4269 | pageset_update(&p->pcp, high, batch); |
8ad4b1fb RS |
4270 | } |
4271 | ||
7cd2b0a3 DR |
4272 | static void pageset_set_high_and_batch(struct zone *zone, |
4273 | struct per_cpu_pageset *pcp) | |
56cef2b8 | 4274 | { |
56cef2b8 | 4275 | if (percpu_pagelist_fraction) |
3664033c | 4276 | pageset_set_high(pcp, |
56cef2b8 CS |
4277 | (zone->managed_pages / |
4278 | percpu_pagelist_fraction)); | |
4279 | else | |
4280 | pageset_set_batch(pcp, zone_batchsize(zone)); | |
4281 | } | |
4282 | ||
169f6c19 CS |
4283 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
4284 | { | |
4285 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4286 | ||
4287 | pageset_init(pcp); | |
4288 | pageset_set_high_and_batch(zone, pcp); | |
4289 | } | |
4290 | ||
4ed7e022 | 4291 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4292 | { |
4293 | int cpu; | |
319774e2 | 4294 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
56cef2b8 CS |
4295 | for_each_possible_cpu(cpu) |
4296 | zone_pageset_init(zone, cpu); | |
319774e2 WF |
4297 | } |
4298 | ||
2caaad41 | 4299 | /* |
99dcc3e5 CL |
4300 | * Allocate per cpu pagesets and initialize them. |
4301 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4302 | */ |
99dcc3e5 | 4303 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4304 | { |
99dcc3e5 | 4305 | struct zone *zone; |
e7c8d5c9 | 4306 | |
319774e2 WF |
4307 | for_each_populated_zone(zone) |
4308 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4309 | } |
4310 | ||
577a32f6 | 4311 | static noinline __init_refok |
cca448fe | 4312 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4313 | { |
4314 | int i; | |
cca448fe | 4315 | size_t alloc_size; |
ed8ece2e DH |
4316 | |
4317 | /* | |
4318 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4319 | * per zone. | |
4320 | */ | |
02b694de YG |
4321 | zone->wait_table_hash_nr_entries = |
4322 | wait_table_hash_nr_entries(zone_size_pages); | |
4323 | zone->wait_table_bits = | |
4324 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4325 | alloc_size = zone->wait_table_hash_nr_entries |
4326 | * sizeof(wait_queue_head_t); | |
4327 | ||
cd94b9db | 4328 | if (!slab_is_available()) { |
cca448fe | 4329 | zone->wait_table = (wait_queue_head_t *) |
6782832e SS |
4330 | memblock_virt_alloc_node_nopanic( |
4331 | alloc_size, zone->zone_pgdat->node_id); | |
cca448fe YG |
4332 | } else { |
4333 | /* | |
4334 | * This case means that a zone whose size was 0 gets new memory | |
4335 | * via memory hot-add. | |
4336 | * But it may be the case that a new node was hot-added. In | |
4337 | * this case vmalloc() will not be able to use this new node's | |
4338 | * memory - this wait_table must be initialized to use this new | |
4339 | * node itself as well. | |
4340 | * To use this new node's memory, further consideration will be | |
4341 | * necessary. | |
4342 | */ | |
8691f3a7 | 4343 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4344 | } |
4345 | if (!zone->wait_table) | |
4346 | return -ENOMEM; | |
ed8ece2e | 4347 | |
b8af2941 | 4348 | for (i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4349 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4350 | |
4351 | return 0; | |
ed8ece2e DH |
4352 | } |
4353 | ||
c09b4240 | 4354 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4355 | { |
99dcc3e5 CL |
4356 | /* |
4357 | * per cpu subsystem is not up at this point. The following code | |
4358 | * relies on the ability of the linker to provide the | |
4359 | * offset of a (static) per cpu variable into the per cpu area. | |
4360 | */ | |
4361 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4362 | |
b38a8725 | 4363 | if (populated_zone(zone)) |
99dcc3e5 CL |
4364 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4365 | zone->name, zone->present_pages, | |
4366 | zone_batchsize(zone)); | |
ed8ece2e DH |
4367 | } |
4368 | ||
4ed7e022 | 4369 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4370 | unsigned long zone_start_pfn, |
a2f3aa02 DH |
4371 | unsigned long size, |
4372 | enum memmap_context context) | |
ed8ece2e DH |
4373 | { |
4374 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4375 | int ret; |
4376 | ret = zone_wait_table_init(zone, size); | |
4377 | if (ret) | |
4378 | return ret; | |
ed8ece2e DH |
4379 | pgdat->nr_zones = zone_idx(zone) + 1; |
4380 | ||
ed8ece2e DH |
4381 | zone->zone_start_pfn = zone_start_pfn; |
4382 | ||
708614e6 MG |
4383 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4384 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4385 | pgdat->node_id, | |
4386 | (unsigned long)zone_idx(zone), | |
4387 | zone_start_pfn, (zone_start_pfn + size)); | |
4388 | ||
1e548deb | 4389 | zone_init_free_lists(zone); |
718127cc YG |
4390 | |
4391 | return 0; | |
ed8ece2e DH |
4392 | } |
4393 | ||
0ee332c1 | 4394 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
4395 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
4396 | /* | |
4397 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
c713216d | 4398 | */ |
f2dbcfa7 | 4399 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d | 4400 | { |
c13291a5 | 4401 | unsigned long start_pfn, end_pfn; |
e76b63f8 | 4402 | int nid; |
7c243c71 RA |
4403 | /* |
4404 | * NOTE: The following SMP-unsafe globals are only used early in boot | |
4405 | * when the kernel is running single-threaded. | |
4406 | */ | |
4407 | static unsigned long __meminitdata last_start_pfn, last_end_pfn; | |
4408 | static int __meminitdata last_nid; | |
4409 | ||
4410 | if (last_start_pfn <= pfn && pfn < last_end_pfn) | |
4411 | return last_nid; | |
c713216d | 4412 | |
e76b63f8 YL |
4413 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
4414 | if (nid != -1) { | |
4415 | last_start_pfn = start_pfn; | |
4416 | last_end_pfn = end_pfn; | |
4417 | last_nid = nid; | |
4418 | } | |
4419 | ||
4420 | return nid; | |
c713216d MG |
4421 | } |
4422 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
4423 | ||
f2dbcfa7 KH |
4424 | int __meminit early_pfn_to_nid(unsigned long pfn) |
4425 | { | |
cc2559bc KH |
4426 | int nid; |
4427 | ||
4428 | nid = __early_pfn_to_nid(pfn); | |
4429 | if (nid >= 0) | |
4430 | return nid; | |
4431 | /* just returns 0 */ | |
4432 | return 0; | |
f2dbcfa7 KH |
4433 | } |
4434 | ||
cc2559bc KH |
4435 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
4436 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
4437 | { | |
4438 | int nid; | |
4439 | ||
4440 | nid = __early_pfn_to_nid(pfn); | |
4441 | if (nid >= 0 && nid != node) | |
4442 | return false; | |
4443 | return true; | |
4444 | } | |
4445 | #endif | |
f2dbcfa7 | 4446 | |
c713216d | 4447 | /** |
6782832e | 4448 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
88ca3b94 | 4449 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
6782832e | 4450 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
c713216d | 4451 | * |
7d018176 ZZ |
4452 | * If an architecture guarantees that all ranges registered contain no holes |
4453 | * and may be freed, this this function may be used instead of calling | |
4454 | * memblock_free_early_nid() manually. | |
c713216d | 4455 | */ |
c13291a5 | 4456 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 4457 | { |
c13291a5 TH |
4458 | unsigned long start_pfn, end_pfn; |
4459 | int i, this_nid; | |
edbe7d23 | 4460 | |
c13291a5 TH |
4461 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
4462 | start_pfn = min(start_pfn, max_low_pfn); | |
4463 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 4464 | |
c13291a5 | 4465 | if (start_pfn < end_pfn) |
6782832e SS |
4466 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
4467 | (end_pfn - start_pfn) << PAGE_SHIFT, | |
4468 | this_nid); | |
edbe7d23 | 4469 | } |
edbe7d23 | 4470 | } |
edbe7d23 | 4471 | |
c713216d MG |
4472 | /** |
4473 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 4474 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d | 4475 | * |
7d018176 ZZ |
4476 | * If an architecture guarantees that all ranges registered contain no holes and may |
4477 | * be freed, this function may be used instead of calling memory_present() manually. | |
c713216d MG |
4478 | */ |
4479 | void __init sparse_memory_present_with_active_regions(int nid) | |
4480 | { | |
c13291a5 TH |
4481 | unsigned long start_pfn, end_pfn; |
4482 | int i, this_nid; | |
c713216d | 4483 | |
c13291a5 TH |
4484 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
4485 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
4486 | } |
4487 | ||
4488 | /** | |
4489 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
4490 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
4491 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
4492 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
4493 | * |
4494 | * It returns the start and end page frame of a node based on information | |
7d018176 | 4495 | * provided by memblock_set_node(). If called for a node |
c713216d | 4496 | * with no available memory, a warning is printed and the start and end |
88ca3b94 | 4497 | * PFNs will be 0. |
c713216d | 4498 | */ |
a3142c8e | 4499 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
4500 | unsigned long *start_pfn, unsigned long *end_pfn) |
4501 | { | |
c13291a5 | 4502 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 4503 | int i; |
c13291a5 | 4504 | |
c713216d MG |
4505 | *start_pfn = -1UL; |
4506 | *end_pfn = 0; | |
4507 | ||
c13291a5 TH |
4508 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
4509 | *start_pfn = min(*start_pfn, this_start_pfn); | |
4510 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
4511 | } |
4512 | ||
633c0666 | 4513 | if (*start_pfn == -1UL) |
c713216d | 4514 | *start_pfn = 0; |
c713216d MG |
4515 | } |
4516 | ||
2a1e274a MG |
4517 | /* |
4518 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
4519 | * assumption is made that zones within a node are ordered in monotonic | |
4520 | * increasing memory addresses so that the "highest" populated zone is used | |
4521 | */ | |
b69a7288 | 4522 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
4523 | { |
4524 | int zone_index; | |
4525 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
4526 | if (zone_index == ZONE_MOVABLE) | |
4527 | continue; | |
4528 | ||
4529 | if (arch_zone_highest_possible_pfn[zone_index] > | |
4530 | arch_zone_lowest_possible_pfn[zone_index]) | |
4531 | break; | |
4532 | } | |
4533 | ||
4534 | VM_BUG_ON(zone_index == -1); | |
4535 | movable_zone = zone_index; | |
4536 | } | |
4537 | ||
4538 | /* | |
4539 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 4540 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
4541 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
4542 | * in each node depending on the size of each node and how evenly kernelcore | |
4543 | * is distributed. This helper function adjusts the zone ranges | |
4544 | * provided by the architecture for a given node by using the end of the | |
4545 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
4546 | * zones within a node are in order of monotonic increases memory addresses | |
4547 | */ | |
b69a7288 | 4548 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4549 | unsigned long zone_type, |
4550 | unsigned long node_start_pfn, | |
4551 | unsigned long node_end_pfn, | |
4552 | unsigned long *zone_start_pfn, | |
4553 | unsigned long *zone_end_pfn) | |
4554 | { | |
4555 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4556 | if (zone_movable_pfn[nid]) { | |
4557 | /* Size ZONE_MOVABLE */ | |
4558 | if (zone_type == ZONE_MOVABLE) { | |
4559 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4560 | *zone_end_pfn = min(node_end_pfn, | |
4561 | arch_zone_highest_possible_pfn[movable_zone]); | |
4562 | ||
4563 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4564 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4565 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4566 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4567 | ||
4568 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4569 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4570 | *zone_start_pfn = *zone_end_pfn; | |
4571 | } | |
4572 | } | |
4573 | ||
c713216d MG |
4574 | /* |
4575 | * Return the number of pages a zone spans in a node, including holes | |
4576 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4577 | */ | |
6ea6e688 | 4578 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4579 | unsigned long zone_type, |
7960aedd ZY |
4580 | unsigned long node_start_pfn, |
4581 | unsigned long node_end_pfn, | |
c713216d MG |
4582 | unsigned long *ignored) |
4583 | { | |
c713216d MG |
4584 | unsigned long zone_start_pfn, zone_end_pfn; |
4585 | ||
7960aedd | 4586 | /* Get the start and end of the zone */ |
c713216d MG |
4587 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
4588 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4589 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4590 | node_start_pfn, node_end_pfn, | |
4591 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4592 | |
4593 | /* Check that this node has pages within the zone's required range */ | |
4594 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4595 | return 0; | |
4596 | ||
4597 | /* Move the zone boundaries inside the node if necessary */ | |
4598 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4599 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4600 | ||
4601 | /* Return the spanned pages */ | |
4602 | return zone_end_pfn - zone_start_pfn; | |
4603 | } | |
4604 | ||
4605 | /* | |
4606 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4607 | * then all holes in the requested range will be accounted for. |
c713216d | 4608 | */ |
32996250 | 4609 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4610 | unsigned long range_start_pfn, |
4611 | unsigned long range_end_pfn) | |
4612 | { | |
96e907d1 TH |
4613 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4614 | unsigned long start_pfn, end_pfn; | |
4615 | int i; | |
c713216d | 4616 | |
96e907d1 TH |
4617 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4618 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4619 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4620 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4621 | } |
96e907d1 | 4622 | return nr_absent; |
c713216d MG |
4623 | } |
4624 | ||
4625 | /** | |
4626 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4627 | * @start_pfn: The start PFN to start searching for holes | |
4628 | * @end_pfn: The end PFN to stop searching for holes | |
4629 | * | |
88ca3b94 | 4630 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4631 | */ |
4632 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4633 | unsigned long end_pfn) | |
4634 | { | |
4635 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4636 | } | |
4637 | ||
4638 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4639 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4640 | unsigned long zone_type, |
7960aedd ZY |
4641 | unsigned long node_start_pfn, |
4642 | unsigned long node_end_pfn, | |
c713216d MG |
4643 | unsigned long *ignored) |
4644 | { | |
96e907d1 TH |
4645 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4646 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4647 | unsigned long zone_start_pfn, zone_end_pfn; |
4648 | ||
96e907d1 TH |
4649 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4650 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4651 | |
2a1e274a MG |
4652 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4653 | node_start_pfn, node_end_pfn, | |
4654 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4655 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4656 | } |
0e0b864e | 4657 | |
0ee332c1 | 4658 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4659 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d | 4660 | unsigned long zone_type, |
7960aedd ZY |
4661 | unsigned long node_start_pfn, |
4662 | unsigned long node_end_pfn, | |
c713216d MG |
4663 | unsigned long *zones_size) |
4664 | { | |
4665 | return zones_size[zone_type]; | |
4666 | } | |
4667 | ||
6ea6e688 | 4668 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d | 4669 | unsigned long zone_type, |
7960aedd ZY |
4670 | unsigned long node_start_pfn, |
4671 | unsigned long node_end_pfn, | |
c713216d MG |
4672 | unsigned long *zholes_size) |
4673 | { | |
4674 | if (!zholes_size) | |
4675 | return 0; | |
4676 | ||
4677 | return zholes_size[zone_type]; | |
4678 | } | |
20e6926d | 4679 | |
0ee332c1 | 4680 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4681 | |
a3142c8e | 4682 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
7960aedd ZY |
4683 | unsigned long node_start_pfn, |
4684 | unsigned long node_end_pfn, | |
4685 | unsigned long *zones_size, | |
4686 | unsigned long *zholes_size) | |
c713216d MG |
4687 | { |
4688 | unsigned long realtotalpages, totalpages = 0; | |
4689 | enum zone_type i; | |
4690 | ||
4691 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4692 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4693 | node_start_pfn, |
4694 | node_end_pfn, | |
4695 | zones_size); | |
c713216d MG |
4696 | pgdat->node_spanned_pages = totalpages; |
4697 | ||
4698 | realtotalpages = totalpages; | |
4699 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4700 | realtotalpages -= | |
4701 | zone_absent_pages_in_node(pgdat->node_id, i, | |
7960aedd ZY |
4702 | node_start_pfn, node_end_pfn, |
4703 | zholes_size); | |
c713216d MG |
4704 | pgdat->node_present_pages = realtotalpages; |
4705 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4706 | realtotalpages); | |
4707 | } | |
4708 | ||
835c134e MG |
4709 | #ifndef CONFIG_SPARSEMEM |
4710 | /* | |
4711 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4712 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4713 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4714 | * round what is now in bits to nearest long in bits, then return it in |
4715 | * bytes. | |
4716 | */ | |
7c45512d | 4717 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
4718 | { |
4719 | unsigned long usemapsize; | |
4720 | ||
7c45512d | 4721 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
4722 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4723 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4724 | usemapsize *= NR_PAGEBLOCK_BITS; |
4725 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4726 | ||
4727 | return usemapsize / 8; | |
4728 | } | |
4729 | ||
4730 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
4731 | struct zone *zone, |
4732 | unsigned long zone_start_pfn, | |
4733 | unsigned long zonesize) | |
835c134e | 4734 | { |
7c45512d | 4735 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 4736 | zone->pageblock_flags = NULL; |
58a01a45 | 4737 | if (usemapsize) |
6782832e SS |
4738 | zone->pageblock_flags = |
4739 | memblock_virt_alloc_node_nopanic(usemapsize, | |
4740 | pgdat->node_id); | |
835c134e MG |
4741 | } |
4742 | #else | |
7c45512d LT |
4743 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
4744 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
4745 | #endif /* CONFIG_SPARSEMEM */ |
4746 | ||
d9c23400 | 4747 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 4748 | |
d9c23400 | 4749 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
15ca220e | 4750 | void __paginginit set_pageblock_order(void) |
d9c23400 | 4751 | { |
955c1cd7 AM |
4752 | unsigned int order; |
4753 | ||
d9c23400 MG |
4754 | /* Check that pageblock_nr_pages has not already been setup */ |
4755 | if (pageblock_order) | |
4756 | return; | |
4757 | ||
955c1cd7 AM |
4758 | if (HPAGE_SHIFT > PAGE_SHIFT) |
4759 | order = HUGETLB_PAGE_ORDER; | |
4760 | else | |
4761 | order = MAX_ORDER - 1; | |
4762 | ||
d9c23400 MG |
4763 | /* |
4764 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
4765 | * This value may be variable depending on boot parameters on IA64 and |
4766 | * powerpc. | |
d9c23400 MG |
4767 | */ |
4768 | pageblock_order = order; | |
4769 | } | |
4770 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4771 | ||
ba72cb8c MG |
4772 | /* |
4773 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
4774 | * is unused as pageblock_order is set at compile-time. See |
4775 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
4776 | * the kernel config | |
ba72cb8c | 4777 | */ |
15ca220e | 4778 | void __paginginit set_pageblock_order(void) |
ba72cb8c | 4779 | { |
ba72cb8c | 4780 | } |
d9c23400 MG |
4781 | |
4782 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4783 | ||
01cefaef JL |
4784 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
4785 | unsigned long present_pages) | |
4786 | { | |
4787 | unsigned long pages = spanned_pages; | |
4788 | ||
4789 | /* | |
4790 | * Provide a more accurate estimation if there are holes within | |
4791 | * the zone and SPARSEMEM is in use. If there are holes within the | |
4792 | * zone, each populated memory region may cost us one or two extra | |
4793 | * memmap pages due to alignment because memmap pages for each | |
4794 | * populated regions may not naturally algined on page boundary. | |
4795 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
4796 | */ | |
4797 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
4798 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
4799 | pages = present_pages; | |
4800 | ||
4801 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
4802 | } | |
4803 | ||
1da177e4 LT |
4804 | /* |
4805 | * Set up the zone data structures: | |
4806 | * - mark all pages reserved | |
4807 | * - mark all memory queues empty | |
4808 | * - clear the memory bitmaps | |
6527af5d MK |
4809 | * |
4810 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 4811 | */ |
b5a0e011 | 4812 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
7960aedd | 4813 | unsigned long node_start_pfn, unsigned long node_end_pfn, |
1da177e4 LT |
4814 | unsigned long *zones_size, unsigned long *zholes_size) |
4815 | { | |
2f1b6248 | 4816 | enum zone_type j; |
ed8ece2e | 4817 | int nid = pgdat->node_id; |
1da177e4 | 4818 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4819 | int ret; |
1da177e4 | 4820 | |
208d54e5 | 4821 | pgdat_resize_init(pgdat); |
8177a420 AA |
4822 | #ifdef CONFIG_NUMA_BALANCING |
4823 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
4824 | pgdat->numabalancing_migrate_nr_pages = 0; | |
4825 | pgdat->numabalancing_migrate_next_window = jiffies; | |
4826 | #endif | |
1da177e4 | 4827 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 4828 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
52d4b9ac | 4829 | pgdat_page_cgroup_init(pgdat); |
5f63b720 | 4830 | |
1da177e4 LT |
4831 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4832 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 4833 | unsigned long size, realsize, freesize, memmap_pages; |
1da177e4 | 4834 | |
7960aedd ZY |
4835 | size = zone_spanned_pages_in_node(nid, j, node_start_pfn, |
4836 | node_end_pfn, zones_size); | |
9feedc9d | 4837 | realsize = freesize = size - zone_absent_pages_in_node(nid, j, |
7960aedd ZY |
4838 | node_start_pfn, |
4839 | node_end_pfn, | |
c713216d | 4840 | zholes_size); |
1da177e4 | 4841 | |
0e0b864e | 4842 | /* |
9feedc9d | 4843 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
4844 | * is used by this zone for memmap. This affects the watermark |
4845 | * and per-cpu initialisations | |
4846 | */ | |
01cefaef | 4847 | memmap_pages = calc_memmap_size(size, realsize); |
9feedc9d JL |
4848 | if (freesize >= memmap_pages) { |
4849 | freesize -= memmap_pages; | |
5594c8c8 YL |
4850 | if (memmap_pages) |
4851 | printk(KERN_DEBUG | |
4852 | " %s zone: %lu pages used for memmap\n", | |
4853 | zone_names[j], memmap_pages); | |
0e0b864e MG |
4854 | } else |
4855 | printk(KERN_WARNING | |
9feedc9d JL |
4856 | " %s zone: %lu pages exceeds freesize %lu\n", |
4857 | zone_names[j], memmap_pages, freesize); | |
0e0b864e | 4858 | |
6267276f | 4859 | /* Account for reserved pages */ |
9feedc9d JL |
4860 | if (j == 0 && freesize > dma_reserve) { |
4861 | freesize -= dma_reserve; | |
d903ef9f | 4862 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4863 | zone_names[0], dma_reserve); |
0e0b864e MG |
4864 | } |
4865 | ||
98d2b0eb | 4866 | if (!is_highmem_idx(j)) |
9feedc9d | 4867 | nr_kernel_pages += freesize; |
01cefaef JL |
4868 | /* Charge for highmem memmap if there are enough kernel pages */ |
4869 | else if (nr_kernel_pages > memmap_pages * 2) | |
4870 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 4871 | nr_all_pages += freesize; |
1da177e4 LT |
4872 | |
4873 | zone->spanned_pages = size; | |
306f2e9e | 4874 | zone->present_pages = realsize; |
9feedc9d JL |
4875 | /* |
4876 | * Set an approximate value for lowmem here, it will be adjusted | |
4877 | * when the bootmem allocator frees pages into the buddy system. | |
4878 | * And all highmem pages will be managed by the buddy system. | |
4879 | */ | |
4880 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 4881 | #ifdef CONFIG_NUMA |
d5f541ed | 4882 | zone->node = nid; |
9feedc9d | 4883 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 4884 | / 100; |
9feedc9d | 4885 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 4886 | #endif |
1da177e4 LT |
4887 | zone->name = zone_names[j]; |
4888 | spin_lock_init(&zone->lock); | |
4889 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4890 | zone_seqlock_init(zone); |
1da177e4 | 4891 | zone->zone_pgdat = pgdat; |
ed8ece2e | 4892 | zone_pcp_init(zone); |
81c0a2bb JW |
4893 | |
4894 | /* For bootup, initialized properly in watermark setup */ | |
4895 | mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages); | |
4896 | ||
bea8c150 | 4897 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
4898 | if (!size) |
4899 | continue; | |
4900 | ||
955c1cd7 | 4901 | set_pageblock_order(); |
7c45512d | 4902 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
a2f3aa02 DH |
4903 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4904 | size, MEMMAP_EARLY); | |
718127cc | 4905 | BUG_ON(ret); |
76cdd58e | 4906 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 4907 | zone_start_pfn += size; |
1da177e4 LT |
4908 | } |
4909 | } | |
4910 | ||
577a32f6 | 4911 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 4912 | { |
1da177e4 LT |
4913 | /* Skip empty nodes */ |
4914 | if (!pgdat->node_spanned_pages) | |
4915 | return; | |
4916 | ||
d41dee36 | 4917 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
4918 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
4919 | if (!pgdat->node_mem_map) { | |
e984bb43 | 4920 | unsigned long size, start, end; |
d41dee36 AW |
4921 | struct page *map; |
4922 | ||
e984bb43 BP |
4923 | /* |
4924 | * The zone's endpoints aren't required to be MAX_ORDER | |
4925 | * aligned but the node_mem_map endpoints must be in order | |
4926 | * for the buddy allocator to function correctly. | |
4927 | */ | |
4928 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
108bcc96 | 4929 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
4930 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
4931 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
4932 | map = alloc_remap(pgdat->node_id, size); |
4933 | if (!map) | |
6782832e SS |
4934 | map = memblock_virt_alloc_node_nopanic(size, |
4935 | pgdat->node_id); | |
e984bb43 | 4936 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 4937 | } |
12d810c1 | 4938 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
4939 | /* |
4940 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
4941 | */ | |
c713216d | 4942 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 4943 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 4944 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4945 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 4946 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 4947 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4948 | } |
1da177e4 | 4949 | #endif |
d41dee36 | 4950 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
4951 | } |
4952 | ||
9109fb7b JW |
4953 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
4954 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 4955 | { |
9109fb7b | 4956 | pg_data_t *pgdat = NODE_DATA(nid); |
7960aedd ZY |
4957 | unsigned long start_pfn = 0; |
4958 | unsigned long end_pfn = 0; | |
9109fb7b | 4959 | |
88fdf75d | 4960 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 4961 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 4962 | |
1da177e4 LT |
4963 | pgdat->node_id = nid; |
4964 | pgdat->node_start_pfn = node_start_pfn; | |
7960aedd ZY |
4965 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
4966 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); | |
4967 | #endif | |
4968 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, | |
4969 | zones_size, zholes_size); | |
1da177e4 LT |
4970 | |
4971 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
4972 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
4973 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
4974 | nid, (unsigned long)pgdat, | |
4975 | (unsigned long)pgdat->node_mem_map); | |
4976 | #endif | |
1da177e4 | 4977 | |
7960aedd ZY |
4978 | free_area_init_core(pgdat, start_pfn, end_pfn, |
4979 | zones_size, zholes_size); | |
1da177e4 LT |
4980 | } |
4981 | ||
0ee332c1 | 4982 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
4983 | |
4984 | #if MAX_NUMNODES > 1 | |
4985 | /* | |
4986 | * Figure out the number of possible node ids. | |
4987 | */ | |
f9872caf | 4988 | void __init setup_nr_node_ids(void) |
418508c1 MS |
4989 | { |
4990 | unsigned int node; | |
4991 | unsigned int highest = 0; | |
4992 | ||
4993 | for_each_node_mask(node, node_possible_map) | |
4994 | highest = node; | |
4995 | nr_node_ids = highest + 1; | |
4996 | } | |
418508c1 MS |
4997 | #endif |
4998 | ||
1e01979c TH |
4999 | /** |
5000 | * node_map_pfn_alignment - determine the maximum internode alignment | |
5001 | * | |
5002 | * This function should be called after node map is populated and sorted. | |
5003 | * It calculates the maximum power of two alignment which can distinguish | |
5004 | * all the nodes. | |
5005 | * | |
5006 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
5007 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
5008 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
5009 | * shifted, 1GiB is enough and this function will indicate so. | |
5010 | * | |
5011 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
5012 | * model has fine enough granularity to avoid incorrect mapping for the | |
5013 | * populated node map. | |
5014 | * | |
5015 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
5016 | * requirement (single node). | |
5017 | */ | |
5018 | unsigned long __init node_map_pfn_alignment(void) | |
5019 | { | |
5020 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 5021 | unsigned long start, end, mask; |
1e01979c | 5022 | int last_nid = -1; |
c13291a5 | 5023 | int i, nid; |
1e01979c | 5024 | |
c13291a5 | 5025 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
5026 | if (!start || last_nid < 0 || last_nid == nid) { |
5027 | last_nid = nid; | |
5028 | last_end = end; | |
5029 | continue; | |
5030 | } | |
5031 | ||
5032 | /* | |
5033 | * Start with a mask granular enough to pin-point to the | |
5034 | * start pfn and tick off bits one-by-one until it becomes | |
5035 | * too coarse to separate the current node from the last. | |
5036 | */ | |
5037 | mask = ~((1 << __ffs(start)) - 1); | |
5038 | while (mask && last_end <= (start & (mask << 1))) | |
5039 | mask <<= 1; | |
5040 | ||
5041 | /* accumulate all internode masks */ | |
5042 | accl_mask |= mask; | |
5043 | } | |
5044 | ||
5045 | /* convert mask to number of pages */ | |
5046 | return ~accl_mask + 1; | |
5047 | } | |
5048 | ||
a6af2bc3 | 5049 | /* Find the lowest pfn for a node */ |
b69a7288 | 5050 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 5051 | { |
a6af2bc3 | 5052 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
5053 | unsigned long start_pfn; |
5054 | int i; | |
1abbfb41 | 5055 | |
c13291a5 TH |
5056 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
5057 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 5058 | |
a6af2bc3 MG |
5059 | if (min_pfn == ULONG_MAX) { |
5060 | printk(KERN_WARNING | |
2bc0d261 | 5061 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
5062 | return 0; |
5063 | } | |
5064 | ||
5065 | return min_pfn; | |
c713216d MG |
5066 | } |
5067 | ||
5068 | /** | |
5069 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
5070 | * | |
5071 | * It returns the minimum PFN based on information provided via | |
7d018176 | 5072 | * memblock_set_node(). |
c713216d MG |
5073 | */ |
5074 | unsigned long __init find_min_pfn_with_active_regions(void) | |
5075 | { | |
5076 | return find_min_pfn_for_node(MAX_NUMNODES); | |
5077 | } | |
5078 | ||
37b07e41 LS |
5079 | /* |
5080 | * early_calculate_totalpages() | |
5081 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 5082 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 5083 | */ |
484f51f8 | 5084 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 5085 | { |
7e63efef | 5086 | unsigned long totalpages = 0; |
c13291a5 TH |
5087 | unsigned long start_pfn, end_pfn; |
5088 | int i, nid; | |
5089 | ||
5090 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
5091 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 5092 | |
37b07e41 LS |
5093 | totalpages += pages; |
5094 | if (pages) | |
4b0ef1fe | 5095 | node_set_state(nid, N_MEMORY); |
37b07e41 | 5096 | } |
b8af2941 | 5097 | return totalpages; |
7e63efef MG |
5098 | } |
5099 | ||
2a1e274a MG |
5100 | /* |
5101 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
5102 | * is spread evenly between nodes as long as the nodes have enough | |
5103 | * memory. When they don't, some nodes will have more kernelcore than | |
5104 | * others | |
5105 | */ | |
b224ef85 | 5106 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
5107 | { |
5108 | int i, nid; | |
5109 | unsigned long usable_startpfn; | |
5110 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 5111 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 5112 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 5113 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 5114 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
136199f0 | 5115 | struct memblock_region *r; |
b2f3eebe TC |
5116 | |
5117 | /* Need to find movable_zone earlier when movable_node is specified. */ | |
5118 | find_usable_zone_for_movable(); | |
5119 | ||
5120 | /* | |
5121 | * If movable_node is specified, ignore kernelcore and movablecore | |
5122 | * options. | |
5123 | */ | |
5124 | if (movable_node_is_enabled()) { | |
136199f0 EM |
5125 | for_each_memblock(memory, r) { |
5126 | if (!memblock_is_hotpluggable(r)) | |
b2f3eebe TC |
5127 | continue; |
5128 | ||
136199f0 | 5129 | nid = r->nid; |
b2f3eebe | 5130 | |
136199f0 | 5131 | usable_startpfn = PFN_DOWN(r->base); |
b2f3eebe TC |
5132 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
5133 | min(usable_startpfn, zone_movable_pfn[nid]) : | |
5134 | usable_startpfn; | |
5135 | } | |
5136 | ||
5137 | goto out2; | |
5138 | } | |
2a1e274a | 5139 | |
7e63efef | 5140 | /* |
b2f3eebe | 5141 | * If movablecore=nn[KMG] was specified, calculate what size of |
7e63efef MG |
5142 | * kernelcore that corresponds so that memory usable for |
5143 | * any allocation type is evenly spread. If both kernelcore | |
5144 | * and movablecore are specified, then the value of kernelcore | |
5145 | * will be used for required_kernelcore if it's greater than | |
5146 | * what movablecore would have allowed. | |
5147 | */ | |
5148 | if (required_movablecore) { | |
7e63efef MG |
5149 | unsigned long corepages; |
5150 | ||
5151 | /* | |
5152 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
5153 | * was requested by the user | |
5154 | */ | |
5155 | required_movablecore = | |
5156 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
5157 | corepages = totalpages - required_movablecore; | |
5158 | ||
5159 | required_kernelcore = max(required_kernelcore, corepages); | |
5160 | } | |
5161 | ||
20e6926d YL |
5162 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
5163 | if (!required_kernelcore) | |
66918dcd | 5164 | goto out; |
2a1e274a MG |
5165 | |
5166 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
5167 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
5168 | ||
5169 | restart: | |
5170 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
5171 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 5172 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
5173 | unsigned long start_pfn, end_pfn; |
5174 | ||
2a1e274a MG |
5175 | /* |
5176 | * Recalculate kernelcore_node if the division per node | |
5177 | * now exceeds what is necessary to satisfy the requested | |
5178 | * amount of memory for the kernel | |
5179 | */ | |
5180 | if (required_kernelcore < kernelcore_node) | |
5181 | kernelcore_node = required_kernelcore / usable_nodes; | |
5182 | ||
5183 | /* | |
5184 | * As the map is walked, we track how much memory is usable | |
5185 | * by the kernel using kernelcore_remaining. When it is | |
5186 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
5187 | */ | |
5188 | kernelcore_remaining = kernelcore_node; | |
5189 | ||
5190 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 5191 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
5192 | unsigned long size_pages; |
5193 | ||
c13291a5 | 5194 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
5195 | if (start_pfn >= end_pfn) |
5196 | continue; | |
5197 | ||
5198 | /* Account for what is only usable for kernelcore */ | |
5199 | if (start_pfn < usable_startpfn) { | |
5200 | unsigned long kernel_pages; | |
5201 | kernel_pages = min(end_pfn, usable_startpfn) | |
5202 | - start_pfn; | |
5203 | ||
5204 | kernelcore_remaining -= min(kernel_pages, | |
5205 | kernelcore_remaining); | |
5206 | required_kernelcore -= min(kernel_pages, | |
5207 | required_kernelcore); | |
5208 | ||
5209 | /* Continue if range is now fully accounted */ | |
5210 | if (end_pfn <= usable_startpfn) { | |
5211 | ||
5212 | /* | |
5213 | * Push zone_movable_pfn to the end so | |
5214 | * that if we have to rebalance | |
5215 | * kernelcore across nodes, we will | |
5216 | * not double account here | |
5217 | */ | |
5218 | zone_movable_pfn[nid] = end_pfn; | |
5219 | continue; | |
5220 | } | |
5221 | start_pfn = usable_startpfn; | |
5222 | } | |
5223 | ||
5224 | /* | |
5225 | * The usable PFN range for ZONE_MOVABLE is from | |
5226 | * start_pfn->end_pfn. Calculate size_pages as the | |
5227 | * number of pages used as kernelcore | |
5228 | */ | |
5229 | size_pages = end_pfn - start_pfn; | |
5230 | if (size_pages > kernelcore_remaining) | |
5231 | size_pages = kernelcore_remaining; | |
5232 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5233 | ||
5234 | /* | |
5235 | * Some kernelcore has been met, update counts and | |
5236 | * break if the kernelcore for this node has been | |
b8af2941 | 5237 | * satisfied |
2a1e274a MG |
5238 | */ |
5239 | required_kernelcore -= min(required_kernelcore, | |
5240 | size_pages); | |
5241 | kernelcore_remaining -= size_pages; | |
5242 | if (!kernelcore_remaining) | |
5243 | break; | |
5244 | } | |
5245 | } | |
5246 | ||
5247 | /* | |
5248 | * If there is still required_kernelcore, we do another pass with one | |
5249 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5250 | * along on the nodes that still have memory until kernelcore is | |
b8af2941 | 5251 | * satisfied |
2a1e274a MG |
5252 | */ |
5253 | usable_nodes--; | |
5254 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5255 | goto restart; | |
5256 | ||
b2f3eebe | 5257 | out2: |
2a1e274a MG |
5258 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
5259 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5260 | zone_movable_pfn[nid] = | |
5261 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5262 | |
20e6926d | 5263 | out: |
66918dcd | 5264 | /* restore the node_state */ |
4b0ef1fe | 5265 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5266 | } |
5267 | ||
4b0ef1fe LJ |
5268 | /* Any regular or high memory on that node ? */ |
5269 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5270 | { |
37b07e41 LS |
5271 | enum zone_type zone_type; |
5272 | ||
4b0ef1fe LJ |
5273 | if (N_MEMORY == N_NORMAL_MEMORY) |
5274 | return; | |
5275 | ||
5276 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5277 | struct zone *zone = &pgdat->node_zones[zone_type]; |
b38a8725 | 5278 | if (populated_zone(zone)) { |
4b0ef1fe LJ |
5279 | node_set_state(nid, N_HIGH_MEMORY); |
5280 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5281 | zone_type <= ZONE_NORMAL) | |
5282 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5283 | break; |
5284 | } | |
37b07e41 | 5285 | } |
37b07e41 LS |
5286 | } |
5287 | ||
c713216d MG |
5288 | /** |
5289 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 5290 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
5291 | * |
5292 | * This will call free_area_init_node() for each active node in the system. | |
7d018176 | 5293 | * Using the page ranges provided by memblock_set_node(), the size of each |
c713216d MG |
5294 | * zone in each node and their holes is calculated. If the maximum PFN |
5295 | * between two adjacent zones match, it is assumed that the zone is empty. | |
5296 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
5297 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
5298 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
5299 | * at arch_max_dma_pfn. | |
5300 | */ | |
5301 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
5302 | { | |
c13291a5 TH |
5303 | unsigned long start_pfn, end_pfn; |
5304 | int i, nid; | |
a6af2bc3 | 5305 | |
c713216d MG |
5306 | /* Record where the zone boundaries are */ |
5307 | memset(arch_zone_lowest_possible_pfn, 0, | |
5308 | sizeof(arch_zone_lowest_possible_pfn)); | |
5309 | memset(arch_zone_highest_possible_pfn, 0, | |
5310 | sizeof(arch_zone_highest_possible_pfn)); | |
5311 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
5312 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
5313 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
5314 | if (i == ZONE_MOVABLE) |
5315 | continue; | |
c713216d MG |
5316 | arch_zone_lowest_possible_pfn[i] = |
5317 | arch_zone_highest_possible_pfn[i-1]; | |
5318 | arch_zone_highest_possible_pfn[i] = | |
5319 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
5320 | } | |
2a1e274a MG |
5321 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
5322 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
5323 | ||
5324 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
5325 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
b224ef85 | 5326 | find_zone_movable_pfns_for_nodes(); |
c713216d | 5327 | |
c713216d | 5328 | /* Print out the zone ranges */ |
a62e2f4f | 5329 | printk("Zone ranges:\n"); |
2a1e274a MG |
5330 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5331 | if (i == ZONE_MOVABLE) | |
5332 | continue; | |
155cbfc8 | 5333 | printk(KERN_CONT " %-8s ", zone_names[i]); |
72f0ba02 DR |
5334 | if (arch_zone_lowest_possible_pfn[i] == |
5335 | arch_zone_highest_possible_pfn[i]) | |
155cbfc8 | 5336 | printk(KERN_CONT "empty\n"); |
72f0ba02 | 5337 | else |
a62e2f4f BH |
5338 | printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n", |
5339 | arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT, | |
5340 | (arch_zone_highest_possible_pfn[i] | |
5341 | << PAGE_SHIFT) - 1); | |
2a1e274a MG |
5342 | } |
5343 | ||
5344 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
a62e2f4f | 5345 | printk("Movable zone start for each node\n"); |
2a1e274a MG |
5346 | for (i = 0; i < MAX_NUMNODES; i++) { |
5347 | if (zone_movable_pfn[i]) | |
a62e2f4f BH |
5348 | printk(" Node %d: %#010lx\n", i, |
5349 | zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 5350 | } |
c713216d | 5351 | |
f2d52fe5 | 5352 | /* Print out the early node map */ |
a62e2f4f | 5353 | printk("Early memory node ranges\n"); |
c13291a5 | 5354 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
a62e2f4f BH |
5355 | printk(" node %3d: [mem %#010lx-%#010lx]\n", nid, |
5356 | start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
5357 | |
5358 | /* Initialise every node */ | |
708614e6 | 5359 | mminit_verify_pageflags_layout(); |
8ef82866 | 5360 | setup_nr_node_ids(); |
c713216d MG |
5361 | for_each_online_node(nid) { |
5362 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 5363 | free_area_init_node(nid, NULL, |
c713216d | 5364 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
5365 | |
5366 | /* Any memory on that node */ | |
5367 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
5368 | node_set_state(nid, N_MEMORY); |
5369 | check_for_memory(pgdat, nid); | |
c713216d MG |
5370 | } |
5371 | } | |
2a1e274a | 5372 | |
7e63efef | 5373 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
5374 | { |
5375 | unsigned long long coremem; | |
5376 | if (!p) | |
5377 | return -EINVAL; | |
5378 | ||
5379 | coremem = memparse(p, &p); | |
7e63efef | 5380 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 5381 | |
7e63efef | 5382 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
5383 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
5384 | ||
5385 | return 0; | |
5386 | } | |
ed7ed365 | 5387 | |
7e63efef MG |
5388 | /* |
5389 | * kernelcore=size sets the amount of memory for use for allocations that | |
5390 | * cannot be reclaimed or migrated. | |
5391 | */ | |
5392 | static int __init cmdline_parse_kernelcore(char *p) | |
5393 | { | |
5394 | return cmdline_parse_core(p, &required_kernelcore); | |
5395 | } | |
5396 | ||
5397 | /* | |
5398 | * movablecore=size sets the amount of memory for use for allocations that | |
5399 | * can be reclaimed or migrated. | |
5400 | */ | |
5401 | static int __init cmdline_parse_movablecore(char *p) | |
5402 | { | |
5403 | return cmdline_parse_core(p, &required_movablecore); | |
5404 | } | |
5405 | ||
ed7ed365 | 5406 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 5407 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 5408 | |
0ee332c1 | 5409 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5410 | |
c3d5f5f0 JL |
5411 | void adjust_managed_page_count(struct page *page, long count) |
5412 | { | |
5413 | spin_lock(&managed_page_count_lock); | |
5414 | page_zone(page)->managed_pages += count; | |
5415 | totalram_pages += count; | |
3dcc0571 JL |
5416 | #ifdef CONFIG_HIGHMEM |
5417 | if (PageHighMem(page)) | |
5418 | totalhigh_pages += count; | |
5419 | #endif | |
c3d5f5f0 JL |
5420 | spin_unlock(&managed_page_count_lock); |
5421 | } | |
3dcc0571 | 5422 | EXPORT_SYMBOL(adjust_managed_page_count); |
c3d5f5f0 | 5423 | |
11199692 | 5424 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
69afade7 | 5425 | { |
11199692 JL |
5426 | void *pos; |
5427 | unsigned long pages = 0; | |
69afade7 | 5428 | |
11199692 JL |
5429 | start = (void *)PAGE_ALIGN((unsigned long)start); |
5430 | end = (void *)((unsigned long)end & PAGE_MASK); | |
5431 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { | |
dbe67df4 | 5432 | if ((unsigned int)poison <= 0xFF) |
11199692 JL |
5433 | memset(pos, poison, PAGE_SIZE); |
5434 | free_reserved_page(virt_to_page(pos)); | |
69afade7 JL |
5435 | } |
5436 | ||
5437 | if (pages && s) | |
11199692 | 5438 | pr_info("Freeing %s memory: %ldK (%p - %p)\n", |
69afade7 JL |
5439 | s, pages << (PAGE_SHIFT - 10), start, end); |
5440 | ||
5441 | return pages; | |
5442 | } | |
11199692 | 5443 | EXPORT_SYMBOL(free_reserved_area); |
69afade7 | 5444 | |
cfa11e08 JL |
5445 | #ifdef CONFIG_HIGHMEM |
5446 | void free_highmem_page(struct page *page) | |
5447 | { | |
5448 | __free_reserved_page(page); | |
5449 | totalram_pages++; | |
7b4b2a0d | 5450 | page_zone(page)->managed_pages++; |
cfa11e08 JL |
5451 | totalhigh_pages++; |
5452 | } | |
5453 | #endif | |
5454 | ||
7ee3d4e8 JL |
5455 | |
5456 | void __init mem_init_print_info(const char *str) | |
5457 | { | |
5458 | unsigned long physpages, codesize, datasize, rosize, bss_size; | |
5459 | unsigned long init_code_size, init_data_size; | |
5460 | ||
5461 | physpages = get_num_physpages(); | |
5462 | codesize = _etext - _stext; | |
5463 | datasize = _edata - _sdata; | |
5464 | rosize = __end_rodata - __start_rodata; | |
5465 | bss_size = __bss_stop - __bss_start; | |
5466 | init_data_size = __init_end - __init_begin; | |
5467 | init_code_size = _einittext - _sinittext; | |
5468 | ||
5469 | /* | |
5470 | * Detect special cases and adjust section sizes accordingly: | |
5471 | * 1) .init.* may be embedded into .data sections | |
5472 | * 2) .init.text.* may be out of [__init_begin, __init_end], | |
5473 | * please refer to arch/tile/kernel/vmlinux.lds.S. | |
5474 | * 3) .rodata.* may be embedded into .text or .data sections. | |
5475 | */ | |
5476 | #define adj_init_size(start, end, size, pos, adj) \ | |
b8af2941 PK |
5477 | do { \ |
5478 | if (start <= pos && pos < end && size > adj) \ | |
5479 | size -= adj; \ | |
5480 | } while (0) | |
7ee3d4e8 JL |
5481 | |
5482 | adj_init_size(__init_begin, __init_end, init_data_size, | |
5483 | _sinittext, init_code_size); | |
5484 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); | |
5485 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); | |
5486 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); | |
5487 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); | |
5488 | ||
5489 | #undef adj_init_size | |
5490 | ||
5491 | printk("Memory: %luK/%luK available " | |
5492 | "(%luK kernel code, %luK rwdata, %luK rodata, " | |
5493 | "%luK init, %luK bss, %luK reserved" | |
5494 | #ifdef CONFIG_HIGHMEM | |
5495 | ", %luK highmem" | |
5496 | #endif | |
5497 | "%s%s)\n", | |
5498 | nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10), | |
5499 | codesize >> 10, datasize >> 10, rosize >> 10, | |
5500 | (init_data_size + init_code_size) >> 10, bss_size >> 10, | |
5501 | (physpages - totalram_pages) << (PAGE_SHIFT-10), | |
5502 | #ifdef CONFIG_HIGHMEM | |
5503 | totalhigh_pages << (PAGE_SHIFT-10), | |
5504 | #endif | |
5505 | str ? ", " : "", str ? str : ""); | |
5506 | } | |
5507 | ||
0e0b864e | 5508 | /** |
88ca3b94 RD |
5509 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
5510 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
5511 | * |
5512 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
5513 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
5514 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
5515 | * function may optionally be used to account for unfreeable pages in the |
5516 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
5517 | * smaller per-cpu batchsize. | |
0e0b864e MG |
5518 | */ |
5519 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
5520 | { | |
5521 | dma_reserve = new_dma_reserve; | |
5522 | } | |
5523 | ||
1da177e4 LT |
5524 | void __init free_area_init(unsigned long *zones_size) |
5525 | { | |
9109fb7b | 5526 | free_area_init_node(0, zones_size, |
1da177e4 LT |
5527 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
5528 | } | |
1da177e4 | 5529 | |
1da177e4 LT |
5530 | static int page_alloc_cpu_notify(struct notifier_block *self, |
5531 | unsigned long action, void *hcpu) | |
5532 | { | |
5533 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 5534 | |
8bb78442 | 5535 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 5536 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
5537 | drain_pages(cpu); |
5538 | ||
5539 | /* | |
5540 | * Spill the event counters of the dead processor | |
5541 | * into the current processors event counters. | |
5542 | * This artificially elevates the count of the current | |
5543 | * processor. | |
5544 | */ | |
f8891e5e | 5545 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
5546 | |
5547 | /* | |
5548 | * Zero the differential counters of the dead processor | |
5549 | * so that the vm statistics are consistent. | |
5550 | * | |
5551 | * This is only okay since the processor is dead and cannot | |
5552 | * race with what we are doing. | |
5553 | */ | |
2bb921e5 | 5554 | cpu_vm_stats_fold(cpu); |
1da177e4 LT |
5555 | } |
5556 | return NOTIFY_OK; | |
5557 | } | |
1da177e4 LT |
5558 | |
5559 | void __init page_alloc_init(void) | |
5560 | { | |
5561 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
5562 | } | |
5563 | ||
cb45b0e9 HA |
5564 | /* |
5565 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
5566 | * or min_free_kbytes changes. | |
5567 | */ | |
5568 | static void calculate_totalreserve_pages(void) | |
5569 | { | |
5570 | struct pglist_data *pgdat; | |
5571 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5572 | enum zone_type i, j; |
cb45b0e9 HA |
5573 | |
5574 | for_each_online_pgdat(pgdat) { | |
5575 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
5576 | struct zone *zone = pgdat->node_zones + i; | |
5577 | unsigned long max = 0; | |
5578 | ||
5579 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5580 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5581 | if (zone->lowmem_reserve[j] > max) | |
5582 | max = zone->lowmem_reserve[j]; | |
5583 | } | |
5584 | ||
41858966 MG |
5585 | /* we treat the high watermark as reserved pages. */ |
5586 | max += high_wmark_pages(zone); | |
cb45b0e9 | 5587 | |
b40da049 JL |
5588 | if (max > zone->managed_pages) |
5589 | max = zone->managed_pages; | |
cb45b0e9 | 5590 | reserve_pages += max; |
ab8fabd4 JW |
5591 | /* |
5592 | * Lowmem reserves are not available to | |
5593 | * GFP_HIGHUSER page cache allocations and | |
5594 | * kswapd tries to balance zones to their high | |
5595 | * watermark. As a result, neither should be | |
5596 | * regarded as dirtyable memory, to prevent a | |
5597 | * situation where reclaim has to clean pages | |
5598 | * in order to balance the zones. | |
5599 | */ | |
5600 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
5601 | } |
5602 | } | |
ab8fabd4 | 5603 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
5604 | totalreserve_pages = reserve_pages; |
5605 | } | |
5606 | ||
1da177e4 LT |
5607 | /* |
5608 | * setup_per_zone_lowmem_reserve - called whenever | |
5609 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
5610 | * has a correct pages reserved value, so an adequate number of | |
5611 | * pages are left in the zone after a successful __alloc_pages(). | |
5612 | */ | |
5613 | static void setup_per_zone_lowmem_reserve(void) | |
5614 | { | |
5615 | struct pglist_data *pgdat; | |
2f6726e5 | 5616 | enum zone_type j, idx; |
1da177e4 | 5617 | |
ec936fc5 | 5618 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
5619 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5620 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 5621 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
5622 | |
5623 | zone->lowmem_reserve[j] = 0; | |
5624 | ||
2f6726e5 CL |
5625 | idx = j; |
5626 | while (idx) { | |
1da177e4 LT |
5627 | struct zone *lower_zone; |
5628 | ||
2f6726e5 CL |
5629 | idx--; |
5630 | ||
1da177e4 LT |
5631 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
5632 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
5633 | ||
5634 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 5635 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 5636 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 5637 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
5638 | } |
5639 | } | |
5640 | } | |
cb45b0e9 HA |
5641 | |
5642 | /* update totalreserve_pages */ | |
5643 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5644 | } |
5645 | ||
cfd3da1e | 5646 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5647 | { |
5648 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5649 | unsigned long lowmem_pages = 0; | |
5650 | struct zone *zone; | |
5651 | unsigned long flags; | |
5652 | ||
5653 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5654 | for_each_zone(zone) { | |
5655 | if (!is_highmem(zone)) | |
b40da049 | 5656 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
5657 | } |
5658 | ||
5659 | for_each_zone(zone) { | |
ac924c60 AM |
5660 | u64 tmp; |
5661 | ||
1125b4e3 | 5662 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 5663 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 5664 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
5665 | if (is_highmem(zone)) { |
5666 | /* | |
669ed175 NP |
5667 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5668 | * need highmem pages, so cap pages_min to a small | |
5669 | * value here. | |
5670 | * | |
41858966 | 5671 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
5672 | * deltas controls asynch page reclaim, and so should |
5673 | * not be capped for highmem. | |
1da177e4 | 5674 | */ |
90ae8d67 | 5675 | unsigned long min_pages; |
1da177e4 | 5676 | |
b40da049 | 5677 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 5678 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 5679 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5680 | } else { |
669ed175 NP |
5681 | /* |
5682 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5683 | * proportionate to the zone's size. |
5684 | */ | |
41858966 | 5685 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5686 | } |
5687 | ||
41858966 MG |
5688 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
5689 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
49f223a9 | 5690 | |
81c0a2bb JW |
5691 | __mod_zone_page_state(zone, NR_ALLOC_BATCH, |
5692 | high_wmark_pages(zone) - | |
5693 | low_wmark_pages(zone) - | |
5694 | zone_page_state(zone, NR_ALLOC_BATCH)); | |
5695 | ||
56fd56b8 | 5696 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 5697 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5698 | } |
cb45b0e9 HA |
5699 | |
5700 | /* update totalreserve_pages */ | |
5701 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5702 | } |
5703 | ||
cfd3da1e MG |
5704 | /** |
5705 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5706 | * or when memory is hot-{added|removed} | |
5707 | * | |
5708 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5709 | * correctly with respect to min_free_kbytes. | |
5710 | */ | |
5711 | void setup_per_zone_wmarks(void) | |
5712 | { | |
5713 | mutex_lock(&zonelists_mutex); | |
5714 | __setup_per_zone_wmarks(); | |
5715 | mutex_unlock(&zonelists_mutex); | |
5716 | } | |
5717 | ||
55a4462a | 5718 | /* |
556adecb RR |
5719 | * The inactive anon list should be small enough that the VM never has to |
5720 | * do too much work, but large enough that each inactive page has a chance | |
5721 | * to be referenced again before it is swapped out. | |
5722 | * | |
5723 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
5724 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
5725 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
5726 | * the anonymous pages are kept on the inactive list. | |
5727 | * | |
5728 | * total target max | |
5729 | * memory ratio inactive anon | |
5730 | * ------------------------------------- | |
5731 | * 10MB 1 5MB | |
5732 | * 100MB 1 50MB | |
5733 | * 1GB 3 250MB | |
5734 | * 10GB 10 0.9GB | |
5735 | * 100GB 31 3GB | |
5736 | * 1TB 101 10GB | |
5737 | * 10TB 320 32GB | |
5738 | */ | |
1b79acc9 | 5739 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5740 | { |
96cb4df5 | 5741 | unsigned int gb, ratio; |
556adecb | 5742 | |
96cb4df5 | 5743 | /* Zone size in gigabytes */ |
b40da049 | 5744 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 5745 | if (gb) |
556adecb | 5746 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5747 | else |
5748 | ratio = 1; | |
556adecb | 5749 | |
96cb4df5 MK |
5750 | zone->inactive_ratio = ratio; |
5751 | } | |
556adecb | 5752 | |
839a4fcc | 5753 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
5754 | { |
5755 | struct zone *zone; | |
5756 | ||
5757 | for_each_zone(zone) | |
5758 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5759 | } |
5760 | ||
1da177e4 LT |
5761 | /* |
5762 | * Initialise min_free_kbytes. | |
5763 | * | |
5764 | * For small machines we want it small (128k min). For large machines | |
5765 | * we want it large (64MB max). But it is not linear, because network | |
5766 | * bandwidth does not increase linearly with machine size. We use | |
5767 | * | |
b8af2941 | 5768 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
1da177e4 LT |
5769 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
5770 | * | |
5771 | * which yields | |
5772 | * | |
5773 | * 16MB: 512k | |
5774 | * 32MB: 724k | |
5775 | * 64MB: 1024k | |
5776 | * 128MB: 1448k | |
5777 | * 256MB: 2048k | |
5778 | * 512MB: 2896k | |
5779 | * 1024MB: 4096k | |
5780 | * 2048MB: 5792k | |
5781 | * 4096MB: 8192k | |
5782 | * 8192MB: 11584k | |
5783 | * 16384MB: 16384k | |
5784 | */ | |
1b79acc9 | 5785 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
5786 | { |
5787 | unsigned long lowmem_kbytes; | |
5f12733e | 5788 | int new_min_free_kbytes; |
1da177e4 LT |
5789 | |
5790 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5f12733e MH |
5791 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
5792 | ||
5793 | if (new_min_free_kbytes > user_min_free_kbytes) { | |
5794 | min_free_kbytes = new_min_free_kbytes; | |
5795 | if (min_free_kbytes < 128) | |
5796 | min_free_kbytes = 128; | |
5797 | if (min_free_kbytes > 65536) | |
5798 | min_free_kbytes = 65536; | |
5799 | } else { | |
5800 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", | |
5801 | new_min_free_kbytes, user_min_free_kbytes); | |
5802 | } | |
bc75d33f | 5803 | setup_per_zone_wmarks(); |
a6cccdc3 | 5804 | refresh_zone_stat_thresholds(); |
1da177e4 | 5805 | setup_per_zone_lowmem_reserve(); |
556adecb | 5806 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5807 | return 0; |
5808 | } | |
bc75d33f | 5809 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5810 | |
5811 | /* | |
b8af2941 | 5812 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
1da177e4 LT |
5813 | * that we can call two helper functions whenever min_free_kbytes |
5814 | * changes. | |
5815 | */ | |
cccad5b9 | 5816 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5817 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5818 | { |
da8c757b HP |
5819 | int rc; |
5820 | ||
5821 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
5822 | if (rc) | |
5823 | return rc; | |
5824 | ||
5f12733e MH |
5825 | if (write) { |
5826 | user_min_free_kbytes = min_free_kbytes; | |
bc75d33f | 5827 | setup_per_zone_wmarks(); |
5f12733e | 5828 | } |
1da177e4 LT |
5829 | return 0; |
5830 | } | |
5831 | ||
9614634f | 5832 | #ifdef CONFIG_NUMA |
cccad5b9 | 5833 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5834 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5835 | { |
5836 | struct zone *zone; | |
5837 | int rc; | |
5838 | ||
8d65af78 | 5839 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5840 | if (rc) |
5841 | return rc; | |
5842 | ||
5843 | for_each_zone(zone) | |
b40da049 | 5844 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
5845 | sysctl_min_unmapped_ratio) / 100; |
5846 | return 0; | |
5847 | } | |
0ff38490 | 5848 | |
cccad5b9 | 5849 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5850 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5851 | { |
5852 | struct zone *zone; | |
5853 | int rc; | |
5854 | ||
8d65af78 | 5855 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5856 | if (rc) |
5857 | return rc; | |
5858 | ||
5859 | for_each_zone(zone) | |
b40da049 | 5860 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
5861 | sysctl_min_slab_ratio) / 100; |
5862 | return 0; | |
5863 | } | |
9614634f CL |
5864 | #endif |
5865 | ||
1da177e4 LT |
5866 | /* |
5867 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5868 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5869 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5870 | * | |
5871 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5872 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5873 | * if in function of the boot time zone sizes. |
5874 | */ | |
cccad5b9 | 5875 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5876 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5877 | { |
8d65af78 | 5878 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5879 | setup_per_zone_lowmem_reserve(); |
5880 | return 0; | |
5881 | } | |
5882 | ||
8ad4b1fb RS |
5883 | /* |
5884 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
b8af2941 PK |
5885 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
5886 | * pagelist can have before it gets flushed back to buddy allocator. | |
8ad4b1fb | 5887 | */ |
cccad5b9 | 5888 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
8d65af78 | 5889 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5890 | { |
5891 | struct zone *zone; | |
7cd2b0a3 | 5892 | int old_percpu_pagelist_fraction; |
8ad4b1fb RS |
5893 | int ret; |
5894 | ||
7cd2b0a3 DR |
5895 | mutex_lock(&pcp_batch_high_lock); |
5896 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; | |
5897 | ||
8d65af78 | 5898 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7cd2b0a3 DR |
5899 | if (!write || ret < 0) |
5900 | goto out; | |
5901 | ||
5902 | /* Sanity checking to avoid pcp imbalance */ | |
5903 | if (percpu_pagelist_fraction && | |
5904 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { | |
5905 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; | |
5906 | ret = -EINVAL; | |
5907 | goto out; | |
5908 | } | |
5909 | ||
5910 | /* No change? */ | |
5911 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) | |
5912 | goto out; | |
c8e251fa | 5913 | |
364df0eb | 5914 | for_each_populated_zone(zone) { |
7cd2b0a3 DR |
5915 | unsigned int cpu; |
5916 | ||
22a7f12b | 5917 | for_each_possible_cpu(cpu) |
7cd2b0a3 DR |
5918 | pageset_set_high_and_batch(zone, |
5919 | per_cpu_ptr(zone->pageset, cpu)); | |
8ad4b1fb | 5920 | } |
7cd2b0a3 | 5921 | out: |
c8e251fa | 5922 | mutex_unlock(&pcp_batch_high_lock); |
7cd2b0a3 | 5923 | return ret; |
8ad4b1fb RS |
5924 | } |
5925 | ||
f034b5d4 | 5926 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
5927 | |
5928 | #ifdef CONFIG_NUMA | |
5929 | static int __init set_hashdist(char *str) | |
5930 | { | |
5931 | if (!str) | |
5932 | return 0; | |
5933 | hashdist = simple_strtoul(str, &str, 0); | |
5934 | return 1; | |
5935 | } | |
5936 | __setup("hashdist=", set_hashdist); | |
5937 | #endif | |
5938 | ||
5939 | /* | |
5940 | * allocate a large system hash table from bootmem | |
5941 | * - it is assumed that the hash table must contain an exact power-of-2 | |
5942 | * quantity of entries | |
5943 | * - limit is the number of hash buckets, not the total allocation size | |
5944 | */ | |
5945 | void *__init alloc_large_system_hash(const char *tablename, | |
5946 | unsigned long bucketsize, | |
5947 | unsigned long numentries, | |
5948 | int scale, | |
5949 | int flags, | |
5950 | unsigned int *_hash_shift, | |
5951 | unsigned int *_hash_mask, | |
31fe62b9 TB |
5952 | unsigned long low_limit, |
5953 | unsigned long high_limit) | |
1da177e4 | 5954 | { |
31fe62b9 | 5955 | unsigned long long max = high_limit; |
1da177e4 LT |
5956 | unsigned long log2qty, size; |
5957 | void *table = NULL; | |
5958 | ||
5959 | /* allow the kernel cmdline to have a say */ | |
5960 | if (!numentries) { | |
5961 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 5962 | numentries = nr_kernel_pages; |
a7e83318 JZ |
5963 | |
5964 | /* It isn't necessary when PAGE_SIZE >= 1MB */ | |
5965 | if (PAGE_SHIFT < 20) | |
5966 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); | |
1da177e4 LT |
5967 | |
5968 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
5969 | if (scale > PAGE_SHIFT) | |
5970 | numentries >>= (scale - PAGE_SHIFT); | |
5971 | else | |
5972 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
5973 | |
5974 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
5975 | if (unlikely(flags & HASH_SMALL)) { |
5976 | /* Makes no sense without HASH_EARLY */ | |
5977 | WARN_ON(!(flags & HASH_EARLY)); | |
5978 | if (!(numentries >> *_hash_shift)) { | |
5979 | numentries = 1UL << *_hash_shift; | |
5980 | BUG_ON(!numentries); | |
5981 | } | |
5982 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 5983 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 5984 | } |
6e692ed3 | 5985 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
5986 | |
5987 | /* limit allocation size to 1/16 total memory by default */ | |
5988 | if (max == 0) { | |
5989 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
5990 | do_div(max, bucketsize); | |
5991 | } | |
074b8517 | 5992 | max = min(max, 0x80000000ULL); |
1da177e4 | 5993 | |
31fe62b9 TB |
5994 | if (numentries < low_limit) |
5995 | numentries = low_limit; | |
1da177e4 LT |
5996 | if (numentries > max) |
5997 | numentries = max; | |
5998 | ||
f0d1b0b3 | 5999 | log2qty = ilog2(numentries); |
1da177e4 LT |
6000 | |
6001 | do { | |
6002 | size = bucketsize << log2qty; | |
6003 | if (flags & HASH_EARLY) | |
6782832e | 6004 | table = memblock_virt_alloc_nopanic(size, 0); |
1da177e4 LT |
6005 | else if (hashdist) |
6006 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
6007 | else { | |
1037b83b ED |
6008 | /* |
6009 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
6010 | * some pages at the end of hash table which |
6011 | * alloc_pages_exact() automatically does | |
1037b83b | 6012 | */ |
264ef8a9 | 6013 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 6014 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
6015 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
6016 | } | |
1da177e4 LT |
6017 | } |
6018 | } while (!table && size > PAGE_SIZE && --log2qty); | |
6019 | ||
6020 | if (!table) | |
6021 | panic("Failed to allocate %s hash table\n", tablename); | |
6022 | ||
f241e660 | 6023 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 6024 | tablename, |
f241e660 | 6025 | (1UL << log2qty), |
f0d1b0b3 | 6026 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
6027 | size); |
6028 | ||
6029 | if (_hash_shift) | |
6030 | *_hash_shift = log2qty; | |
6031 | if (_hash_mask) | |
6032 | *_hash_mask = (1 << log2qty) - 1; | |
6033 | ||
6034 | return table; | |
6035 | } | |
a117e66e | 6036 | |
835c134e MG |
6037 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
6038 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
6039 | unsigned long pfn) | |
6040 | { | |
6041 | #ifdef CONFIG_SPARSEMEM | |
6042 | return __pfn_to_section(pfn)->pageblock_flags; | |
6043 | #else | |
6044 | return zone->pageblock_flags; | |
6045 | #endif /* CONFIG_SPARSEMEM */ | |
6046 | } | |
6047 | ||
6048 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
6049 | { | |
6050 | #ifdef CONFIG_SPARSEMEM | |
6051 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 6052 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 6053 | #else |
c060f943 | 6054 | pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 6055 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
6056 | #endif /* CONFIG_SPARSEMEM */ |
6057 | } | |
6058 | ||
6059 | /** | |
1aab4d77 | 6060 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e | 6061 | * @page: The page within the block of interest |
1aab4d77 RD |
6062 | * @pfn: The target page frame number |
6063 | * @end_bitidx: The last bit of interest to retrieve | |
6064 | * @mask: mask of bits that the caller is interested in | |
6065 | * | |
6066 | * Return: pageblock_bits flags | |
835c134e | 6067 | */ |
dc4b0caf | 6068 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
e58469ba MG |
6069 | unsigned long end_bitidx, |
6070 | unsigned long mask) | |
835c134e MG |
6071 | { |
6072 | struct zone *zone; | |
6073 | unsigned long *bitmap; | |
dc4b0caf | 6074 | unsigned long bitidx, word_bitidx; |
e58469ba | 6075 | unsigned long word; |
835c134e MG |
6076 | |
6077 | zone = page_zone(page); | |
835c134e MG |
6078 | bitmap = get_pageblock_bitmap(zone, pfn); |
6079 | bitidx = pfn_to_bitidx(zone, pfn); | |
e58469ba MG |
6080 | word_bitidx = bitidx / BITS_PER_LONG; |
6081 | bitidx &= (BITS_PER_LONG-1); | |
835c134e | 6082 | |
e58469ba MG |
6083 | word = bitmap[word_bitidx]; |
6084 | bitidx += end_bitidx; | |
6085 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; | |
835c134e MG |
6086 | } |
6087 | ||
6088 | /** | |
dc4b0caf | 6089 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e | 6090 | * @page: The page within the block of interest |
835c134e | 6091 | * @flags: The flags to set |
1aab4d77 RD |
6092 | * @pfn: The target page frame number |
6093 | * @end_bitidx: The last bit of interest | |
6094 | * @mask: mask of bits that the caller is interested in | |
835c134e | 6095 | */ |
dc4b0caf MG |
6096 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
6097 | unsigned long pfn, | |
e58469ba MG |
6098 | unsigned long end_bitidx, |
6099 | unsigned long mask) | |
835c134e MG |
6100 | { |
6101 | struct zone *zone; | |
6102 | unsigned long *bitmap; | |
dc4b0caf | 6103 | unsigned long bitidx, word_bitidx; |
e58469ba MG |
6104 | unsigned long old_word, word; |
6105 | ||
6106 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); | |
835c134e MG |
6107 | |
6108 | zone = page_zone(page); | |
835c134e MG |
6109 | bitmap = get_pageblock_bitmap(zone, pfn); |
6110 | bitidx = pfn_to_bitidx(zone, pfn); | |
e58469ba MG |
6111 | word_bitidx = bitidx / BITS_PER_LONG; |
6112 | bitidx &= (BITS_PER_LONG-1); | |
6113 | ||
309381fe | 6114 | VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page); |
835c134e | 6115 | |
e58469ba MG |
6116 | bitidx += end_bitidx; |
6117 | mask <<= (BITS_PER_LONG - bitidx - 1); | |
6118 | flags <<= (BITS_PER_LONG - bitidx - 1); | |
6119 | ||
6120 | word = ACCESS_ONCE(bitmap[word_bitidx]); | |
6121 | for (;;) { | |
6122 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); | |
6123 | if (word == old_word) | |
6124 | break; | |
6125 | word = old_word; | |
6126 | } | |
835c134e | 6127 | } |
a5d76b54 KH |
6128 | |
6129 | /* | |
80934513 MK |
6130 | * This function checks whether pageblock includes unmovable pages or not. |
6131 | * If @count is not zero, it is okay to include less @count unmovable pages | |
6132 | * | |
b8af2941 | 6133 | * PageLRU check without isolation or lru_lock could race so that |
80934513 MK |
6134 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
6135 | * expect this function should be exact. | |
a5d76b54 | 6136 | */ |
b023f468 WC |
6137 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
6138 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
6139 | { |
6140 | unsigned long pfn, iter, found; | |
47118af0 MN |
6141 | int mt; |
6142 | ||
49ac8255 KH |
6143 | /* |
6144 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 6145 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
6146 | */ |
6147 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 6148 | return false; |
47118af0 MN |
6149 | mt = get_pageblock_migratetype(page); |
6150 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 6151 | return false; |
49ac8255 KH |
6152 | |
6153 | pfn = page_to_pfn(page); | |
6154 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
6155 | unsigned long check = pfn + iter; | |
6156 | ||
29723fcc | 6157 | if (!pfn_valid_within(check)) |
49ac8255 | 6158 | continue; |
29723fcc | 6159 | |
49ac8255 | 6160 | page = pfn_to_page(check); |
c8721bbb NH |
6161 | |
6162 | /* | |
6163 | * Hugepages are not in LRU lists, but they're movable. | |
6164 | * We need not scan over tail pages bacause we don't | |
6165 | * handle each tail page individually in migration. | |
6166 | */ | |
6167 | if (PageHuge(page)) { | |
6168 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; | |
6169 | continue; | |
6170 | } | |
6171 | ||
97d255c8 MK |
6172 | /* |
6173 | * We can't use page_count without pin a page | |
6174 | * because another CPU can free compound page. | |
6175 | * This check already skips compound tails of THP | |
6176 | * because their page->_count is zero at all time. | |
6177 | */ | |
6178 | if (!atomic_read(&page->_count)) { | |
49ac8255 KH |
6179 | if (PageBuddy(page)) |
6180 | iter += (1 << page_order(page)) - 1; | |
6181 | continue; | |
6182 | } | |
97d255c8 | 6183 | |
b023f468 WC |
6184 | /* |
6185 | * The HWPoisoned page may be not in buddy system, and | |
6186 | * page_count() is not 0. | |
6187 | */ | |
6188 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6189 | continue; | |
6190 | ||
49ac8255 KH |
6191 | if (!PageLRU(page)) |
6192 | found++; | |
6193 | /* | |
6194 | * If there are RECLAIMABLE pages, we need to check it. | |
6195 | * But now, memory offline itself doesn't call shrink_slab() | |
6196 | * and it still to be fixed. | |
6197 | */ | |
6198 | /* | |
6199 | * If the page is not RAM, page_count()should be 0. | |
6200 | * we don't need more check. This is an _used_ not-movable page. | |
6201 | * | |
6202 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6203 | * is set to both of a memory hole page and a _used_ kernel | |
6204 | * page at boot. | |
6205 | */ | |
6206 | if (found > count) | |
80934513 | 6207 | return true; |
49ac8255 | 6208 | } |
80934513 | 6209 | return false; |
49ac8255 KH |
6210 | } |
6211 | ||
6212 | bool is_pageblock_removable_nolock(struct page *page) | |
6213 | { | |
656a0706 MH |
6214 | struct zone *zone; |
6215 | unsigned long pfn; | |
687875fb MH |
6216 | |
6217 | /* | |
6218 | * We have to be careful here because we are iterating over memory | |
6219 | * sections which are not zone aware so we might end up outside of | |
6220 | * the zone but still within the section. | |
656a0706 MH |
6221 | * We have to take care about the node as well. If the node is offline |
6222 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 6223 | */ |
656a0706 MH |
6224 | if (!node_online(page_to_nid(page))) |
6225 | return false; | |
6226 | ||
6227 | zone = page_zone(page); | |
6228 | pfn = page_to_pfn(page); | |
108bcc96 | 6229 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
6230 | return false; |
6231 | ||
b023f468 | 6232 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6233 | } |
0c0e6195 | 6234 | |
041d3a8c MN |
6235 | #ifdef CONFIG_CMA |
6236 | ||
6237 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6238 | { | |
6239 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6240 | pageblock_nr_pages) - 1); | |
6241 | } | |
6242 | ||
6243 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6244 | { | |
6245 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6246 | pageblock_nr_pages)); | |
6247 | } | |
6248 | ||
041d3a8c | 6249 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6250 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6251 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6252 | { |
6253 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6254 | unsigned long nr_reclaimed; |
041d3a8c MN |
6255 | unsigned long pfn = start; |
6256 | unsigned int tries = 0; | |
6257 | int ret = 0; | |
6258 | ||
be49a6e1 | 6259 | migrate_prep(); |
041d3a8c | 6260 | |
bb13ffeb | 6261 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6262 | if (fatal_signal_pending(current)) { |
6263 | ret = -EINTR; | |
6264 | break; | |
6265 | } | |
6266 | ||
bb13ffeb MG |
6267 | if (list_empty(&cc->migratepages)) { |
6268 | cc->nr_migratepages = 0; | |
6269 | pfn = isolate_migratepages_range(cc->zone, cc, | |
e46a2879 | 6270 | pfn, end, true); |
041d3a8c MN |
6271 | if (!pfn) { |
6272 | ret = -EINTR; | |
6273 | break; | |
6274 | } | |
6275 | tries = 0; | |
6276 | } else if (++tries == 5) { | |
6277 | ret = ret < 0 ? ret : -EBUSY; | |
6278 | break; | |
6279 | } | |
6280 | ||
beb51eaa MK |
6281 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6282 | &cc->migratepages); | |
6283 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6284 | |
9c620e2b | 6285 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
e0b9daeb | 6286 | NULL, 0, cc->mode, MR_CMA); |
041d3a8c | 6287 | } |
2a6f5124 SP |
6288 | if (ret < 0) { |
6289 | putback_movable_pages(&cc->migratepages); | |
6290 | return ret; | |
6291 | } | |
6292 | return 0; | |
041d3a8c MN |
6293 | } |
6294 | ||
6295 | /** | |
6296 | * alloc_contig_range() -- tries to allocate given range of pages | |
6297 | * @start: start PFN to allocate | |
6298 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
6299 | * @migratetype: migratetype of the underlaying pageblocks (either |
6300 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
6301 | * in range must have the same migratetype and it must | |
6302 | * be either of the two. | |
041d3a8c MN |
6303 | * |
6304 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
6305 | * aligned, however it's the caller's responsibility to guarantee that | |
6306 | * we are the only thread that changes migrate type of pageblocks the | |
6307 | * pages fall in. | |
6308 | * | |
6309 | * The PFN range must belong to a single zone. | |
6310 | * | |
6311 | * Returns zero on success or negative error code. On success all | |
6312 | * pages which PFN is in [start, end) are allocated for the caller and | |
6313 | * need to be freed with free_contig_range(). | |
6314 | */ | |
0815f3d8 MN |
6315 | int alloc_contig_range(unsigned long start, unsigned long end, |
6316 | unsigned migratetype) | |
041d3a8c | 6317 | { |
041d3a8c MN |
6318 | unsigned long outer_start, outer_end; |
6319 | int ret = 0, order; | |
6320 | ||
bb13ffeb MG |
6321 | struct compact_control cc = { |
6322 | .nr_migratepages = 0, | |
6323 | .order = -1, | |
6324 | .zone = page_zone(pfn_to_page(start)), | |
e0b9daeb | 6325 | .mode = MIGRATE_SYNC, |
bb13ffeb MG |
6326 | .ignore_skip_hint = true, |
6327 | }; | |
6328 | INIT_LIST_HEAD(&cc.migratepages); | |
6329 | ||
041d3a8c MN |
6330 | /* |
6331 | * What we do here is we mark all pageblocks in range as | |
6332 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6333 | * have different sizes, and due to the way page allocator | |
6334 | * work, we align the range to biggest of the two pages so | |
6335 | * that page allocator won't try to merge buddies from | |
6336 | * different pageblocks and change MIGRATE_ISOLATE to some | |
6337 | * other migration type. | |
6338 | * | |
6339 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6340 | * migrate the pages from an unaligned range (ie. pages that | |
6341 | * we are interested in). This will put all the pages in | |
6342 | * range back to page allocator as MIGRATE_ISOLATE. | |
6343 | * | |
6344 | * When this is done, we take the pages in range from page | |
6345 | * allocator removing them from the buddy system. This way | |
6346 | * page allocator will never consider using them. | |
6347 | * | |
6348 | * This lets us mark the pageblocks back as | |
6349 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6350 | * aligned range but not in the unaligned, original range are | |
6351 | * put back to page allocator so that buddy can use them. | |
6352 | */ | |
6353 | ||
6354 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
6355 | pfn_max_align_up(end), migratetype, |
6356 | false); | |
041d3a8c | 6357 | if (ret) |
86a595f9 | 6358 | return ret; |
041d3a8c | 6359 | |
bb13ffeb | 6360 | ret = __alloc_contig_migrate_range(&cc, start, end); |
041d3a8c MN |
6361 | if (ret) |
6362 | goto done; | |
6363 | ||
6364 | /* | |
6365 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
6366 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
6367 | * more, all pages in [start, end) are free in page allocator. | |
6368 | * What we are going to do is to allocate all pages from | |
6369 | * [start, end) (that is remove them from page allocator). | |
6370 | * | |
6371 | * The only problem is that pages at the beginning and at the | |
6372 | * end of interesting range may be not aligned with pages that | |
6373 | * page allocator holds, ie. they can be part of higher order | |
6374 | * pages. Because of this, we reserve the bigger range and | |
6375 | * once this is done free the pages we are not interested in. | |
6376 | * | |
6377 | * We don't have to hold zone->lock here because the pages are | |
6378 | * isolated thus they won't get removed from buddy. | |
6379 | */ | |
6380 | ||
6381 | lru_add_drain_all(); | |
6382 | drain_all_pages(); | |
6383 | ||
6384 | order = 0; | |
6385 | outer_start = start; | |
6386 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
6387 | if (++order >= MAX_ORDER) { | |
6388 | ret = -EBUSY; | |
6389 | goto done; | |
6390 | } | |
6391 | outer_start &= ~0UL << order; | |
6392 | } | |
6393 | ||
6394 | /* Make sure the range is really isolated. */ | |
b023f468 | 6395 | if (test_pages_isolated(outer_start, end, false)) { |
041d3a8c MN |
6396 | pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n", |
6397 | outer_start, end); | |
6398 | ret = -EBUSY; | |
6399 | goto done; | |
6400 | } | |
6401 | ||
49f223a9 MS |
6402 | |
6403 | /* Grab isolated pages from freelists. */ | |
bb13ffeb | 6404 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6405 | if (!outer_end) { |
6406 | ret = -EBUSY; | |
6407 | goto done; | |
6408 | } | |
6409 | ||
6410 | /* Free head and tail (if any) */ | |
6411 | if (start != outer_start) | |
6412 | free_contig_range(outer_start, start - outer_start); | |
6413 | if (end != outer_end) | |
6414 | free_contig_range(end, outer_end - end); | |
6415 | ||
6416 | done: | |
6417 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 6418 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
6419 | return ret; |
6420 | } | |
6421 | ||
6422 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
6423 | { | |
bcc2b02f MS |
6424 | unsigned int count = 0; |
6425 | ||
6426 | for (; nr_pages--; pfn++) { | |
6427 | struct page *page = pfn_to_page(pfn); | |
6428 | ||
6429 | count += page_count(page) != 1; | |
6430 | __free_page(page); | |
6431 | } | |
6432 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
6433 | } |
6434 | #endif | |
6435 | ||
4ed7e022 | 6436 | #ifdef CONFIG_MEMORY_HOTPLUG |
0a647f38 CS |
6437 | /* |
6438 | * The zone indicated has a new number of managed_pages; batch sizes and percpu | |
6439 | * page high values need to be recalulated. | |
6440 | */ | |
4ed7e022 JL |
6441 | void __meminit zone_pcp_update(struct zone *zone) |
6442 | { | |
0a647f38 | 6443 | unsigned cpu; |
c8e251fa | 6444 | mutex_lock(&pcp_batch_high_lock); |
0a647f38 | 6445 | for_each_possible_cpu(cpu) |
169f6c19 CS |
6446 | pageset_set_high_and_batch(zone, |
6447 | per_cpu_ptr(zone->pageset, cpu)); | |
c8e251fa | 6448 | mutex_unlock(&pcp_batch_high_lock); |
4ed7e022 JL |
6449 | } |
6450 | #endif | |
6451 | ||
340175b7 JL |
6452 | void zone_pcp_reset(struct zone *zone) |
6453 | { | |
6454 | unsigned long flags; | |
5a883813 MK |
6455 | int cpu; |
6456 | struct per_cpu_pageset *pset; | |
340175b7 JL |
6457 | |
6458 | /* avoid races with drain_pages() */ | |
6459 | local_irq_save(flags); | |
6460 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
6461 | for_each_online_cpu(cpu) { |
6462 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6463 | drain_zonestat(zone, pset); | |
6464 | } | |
340175b7 JL |
6465 | free_percpu(zone->pageset); |
6466 | zone->pageset = &boot_pageset; | |
6467 | } | |
6468 | local_irq_restore(flags); | |
6469 | } | |
6470 | ||
6dcd73d7 | 6471 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
6472 | /* |
6473 | * All pages in the range must be isolated before calling this. | |
6474 | */ | |
6475 | void | |
6476 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
6477 | { | |
6478 | struct page *page; | |
6479 | struct zone *zone; | |
7aeb09f9 | 6480 | unsigned int order, i; |
0c0e6195 KH |
6481 | unsigned long pfn; |
6482 | unsigned long flags; | |
6483 | /* find the first valid pfn */ | |
6484 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
6485 | if (pfn_valid(pfn)) | |
6486 | break; | |
6487 | if (pfn == end_pfn) | |
6488 | return; | |
6489 | zone = page_zone(pfn_to_page(pfn)); | |
6490 | spin_lock_irqsave(&zone->lock, flags); | |
6491 | pfn = start_pfn; | |
6492 | while (pfn < end_pfn) { | |
6493 | if (!pfn_valid(pfn)) { | |
6494 | pfn++; | |
6495 | continue; | |
6496 | } | |
6497 | page = pfn_to_page(pfn); | |
b023f468 WC |
6498 | /* |
6499 | * The HWPoisoned page may be not in buddy system, and | |
6500 | * page_count() is not 0. | |
6501 | */ | |
6502 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6503 | pfn++; | |
6504 | SetPageReserved(page); | |
6505 | continue; | |
6506 | } | |
6507 | ||
0c0e6195 KH |
6508 | BUG_ON(page_count(page)); |
6509 | BUG_ON(!PageBuddy(page)); | |
6510 | order = page_order(page); | |
6511 | #ifdef CONFIG_DEBUG_VM | |
6512 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
6513 | pfn, 1 << order, end_pfn); | |
6514 | #endif | |
6515 | list_del(&page->lru); | |
6516 | rmv_page_order(page); | |
6517 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
6518 | for (i = 0; i < (1 << order); i++) |
6519 | SetPageReserved((page+i)); | |
6520 | pfn += (1 << order); | |
6521 | } | |
6522 | spin_unlock_irqrestore(&zone->lock, flags); | |
6523 | } | |
6524 | #endif | |
8d22ba1b WF |
6525 | |
6526 | #ifdef CONFIG_MEMORY_FAILURE | |
6527 | bool is_free_buddy_page(struct page *page) | |
6528 | { | |
6529 | struct zone *zone = page_zone(page); | |
6530 | unsigned long pfn = page_to_pfn(page); | |
6531 | unsigned long flags; | |
7aeb09f9 | 6532 | unsigned int order; |
8d22ba1b WF |
6533 | |
6534 | spin_lock_irqsave(&zone->lock, flags); | |
6535 | for (order = 0; order < MAX_ORDER; order++) { | |
6536 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
6537 | ||
6538 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
6539 | break; | |
6540 | } | |
6541 | spin_unlock_irqrestore(&zone->lock, flags); | |
6542 | ||
6543 | return order < MAX_ORDER; | |
6544 | } | |
6545 | #endif | |
718a3821 | 6546 | |
51300cef | 6547 | static const struct trace_print_flags pageflag_names[] = { |
718a3821 WF |
6548 | {1UL << PG_locked, "locked" }, |
6549 | {1UL << PG_error, "error" }, | |
6550 | {1UL << PG_referenced, "referenced" }, | |
6551 | {1UL << PG_uptodate, "uptodate" }, | |
6552 | {1UL << PG_dirty, "dirty" }, | |
6553 | {1UL << PG_lru, "lru" }, | |
6554 | {1UL << PG_active, "active" }, | |
6555 | {1UL << PG_slab, "slab" }, | |
6556 | {1UL << PG_owner_priv_1, "owner_priv_1" }, | |
6557 | {1UL << PG_arch_1, "arch_1" }, | |
6558 | {1UL << PG_reserved, "reserved" }, | |
6559 | {1UL << PG_private, "private" }, | |
6560 | {1UL << PG_private_2, "private_2" }, | |
6561 | {1UL << PG_writeback, "writeback" }, | |
6562 | #ifdef CONFIG_PAGEFLAGS_EXTENDED | |
6563 | {1UL << PG_head, "head" }, | |
6564 | {1UL << PG_tail, "tail" }, | |
6565 | #else | |
6566 | {1UL << PG_compound, "compound" }, | |
6567 | #endif | |
6568 | {1UL << PG_swapcache, "swapcache" }, | |
6569 | {1UL << PG_mappedtodisk, "mappedtodisk" }, | |
6570 | {1UL << PG_reclaim, "reclaim" }, | |
718a3821 WF |
6571 | {1UL << PG_swapbacked, "swapbacked" }, |
6572 | {1UL << PG_unevictable, "unevictable" }, | |
6573 | #ifdef CONFIG_MMU | |
6574 | {1UL << PG_mlocked, "mlocked" }, | |
6575 | #endif | |
6576 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED | |
6577 | {1UL << PG_uncached, "uncached" }, | |
6578 | #endif | |
6579 | #ifdef CONFIG_MEMORY_FAILURE | |
6580 | {1UL << PG_hwpoison, "hwpoison" }, | |
be9cd873 GS |
6581 | #endif |
6582 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6583 | {1UL << PG_compound_lock, "compound_lock" }, | |
718a3821 | 6584 | #endif |
718a3821 WF |
6585 | }; |
6586 | ||
6587 | static void dump_page_flags(unsigned long flags) | |
6588 | { | |
6589 | const char *delim = ""; | |
6590 | unsigned long mask; | |
6591 | int i; | |
6592 | ||
51300cef | 6593 | BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS); |
acc50c11 | 6594 | |
718a3821 WF |
6595 | printk(KERN_ALERT "page flags: %#lx(", flags); |
6596 | ||
6597 | /* remove zone id */ | |
6598 | flags &= (1UL << NR_PAGEFLAGS) - 1; | |
6599 | ||
51300cef | 6600 | for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) { |
718a3821 WF |
6601 | |
6602 | mask = pageflag_names[i].mask; | |
6603 | if ((flags & mask) != mask) | |
6604 | continue; | |
6605 | ||
6606 | flags &= ~mask; | |
6607 | printk("%s%s", delim, pageflag_names[i].name); | |
6608 | delim = "|"; | |
6609 | } | |
6610 | ||
6611 | /* check for left over flags */ | |
6612 | if (flags) | |
6613 | printk("%s%#lx", delim, flags); | |
6614 | ||
6615 | printk(")\n"); | |
6616 | } | |
6617 | ||
d230dec1 KS |
6618 | void dump_page_badflags(struct page *page, const char *reason, |
6619 | unsigned long badflags) | |
718a3821 WF |
6620 | { |
6621 | printk(KERN_ALERT | |
6622 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", | |
4e9f64c4 | 6623 | page, atomic_read(&page->_count), page_mapcount(page), |
718a3821 WF |
6624 | page->mapping, page->index); |
6625 | dump_page_flags(page->flags); | |
f0b791a3 DH |
6626 | if (reason) |
6627 | pr_alert("page dumped because: %s\n", reason); | |
6628 | if (page->flags & badflags) { | |
6629 | pr_alert("bad because of flags:\n"); | |
6630 | dump_page_flags(page->flags & badflags); | |
6631 | } | |
f212ad7c | 6632 | mem_cgroup_print_bad_page(page); |
718a3821 | 6633 | } |
f0b791a3 | 6634 | |
d230dec1 | 6635 | void dump_page(struct page *page, const char *reason) |
f0b791a3 DH |
6636 | { |
6637 | dump_page_badflags(page, reason, 0); | |
6638 | } | |
ed12d845 | 6639 | EXPORT_SYMBOL(dump_page); |