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> |
041d3a8c | 59 | #include <linux/migrate.h> |
c0a32fc5 | 60 | #include <linux/page-debug-flags.h> |
8bd75c77 | 61 | #include <linux/sched/rt.h> |
1da177e4 LT |
62 | |
63 | #include <asm/tlbflush.h> | |
ac924c60 | 64 | #include <asm/div64.h> |
1da177e4 LT |
65 | #include "internal.h" |
66 | ||
72812019 LS |
67 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
68 | DEFINE_PER_CPU(int, numa_node); | |
69 | EXPORT_PER_CPU_SYMBOL(numa_node); | |
70 | #endif | |
71 | ||
7aac7898 LS |
72 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
73 | /* | |
74 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. | |
75 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. | |
76 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() | |
77 | * defined in <linux/topology.h>. | |
78 | */ | |
79 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ | |
80 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); | |
81 | #endif | |
82 | ||
1da177e4 | 83 | /* |
13808910 | 84 | * Array of node states. |
1da177e4 | 85 | */ |
13808910 CL |
86 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
87 | [N_POSSIBLE] = NODE_MASK_ALL, | |
88 | [N_ONLINE] = { { [0] = 1UL } }, | |
89 | #ifndef CONFIG_NUMA | |
90 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
91 | #ifdef CONFIG_HIGHMEM | |
92 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
20b2f52b LJ |
93 | #endif |
94 | #ifdef CONFIG_MOVABLE_NODE | |
95 | [N_MEMORY] = { { [0] = 1UL } }, | |
13808910 CL |
96 | #endif |
97 | [N_CPU] = { { [0] = 1UL } }, | |
98 | #endif /* NUMA */ | |
99 | }; | |
100 | EXPORT_SYMBOL(node_states); | |
101 | ||
6c231b7b | 102 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 103 | unsigned long totalreserve_pages __read_mostly; |
ab8fabd4 JW |
104 | /* |
105 | * When calculating the number of globally allowed dirty pages, there | |
106 | * is a certain number of per-zone reserves that should not be | |
107 | * considered dirtyable memory. This is the sum of those reserves | |
108 | * over all existing zones that contribute dirtyable memory. | |
109 | */ | |
110 | unsigned long dirty_balance_reserve __read_mostly; | |
111 | ||
1b76b02f | 112 | int percpu_pagelist_fraction; |
dcce284a | 113 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
1da177e4 | 114 | |
452aa699 RW |
115 | #ifdef CONFIG_PM_SLEEP |
116 | /* | |
117 | * The following functions are used by the suspend/hibernate code to temporarily | |
118 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations | |
119 | * while devices are suspended. To avoid races with the suspend/hibernate code, | |
120 | * they should always be called with pm_mutex held (gfp_allowed_mask also should | |
121 | * only be modified with pm_mutex held, unless the suspend/hibernate code is | |
122 | * guaranteed not to run in parallel with that modification). | |
123 | */ | |
c9e664f1 RW |
124 | |
125 | static gfp_t saved_gfp_mask; | |
126 | ||
127 | void pm_restore_gfp_mask(void) | |
452aa699 RW |
128 | { |
129 | WARN_ON(!mutex_is_locked(&pm_mutex)); | |
c9e664f1 RW |
130 | if (saved_gfp_mask) { |
131 | gfp_allowed_mask = saved_gfp_mask; | |
132 | saved_gfp_mask = 0; | |
133 | } | |
452aa699 RW |
134 | } |
135 | ||
c9e664f1 | 136 | void pm_restrict_gfp_mask(void) |
452aa699 | 137 | { |
452aa699 | 138 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
c9e664f1 RW |
139 | WARN_ON(saved_gfp_mask); |
140 | saved_gfp_mask = gfp_allowed_mask; | |
141 | gfp_allowed_mask &= ~GFP_IOFS; | |
452aa699 | 142 | } |
f90ac398 MG |
143 | |
144 | bool pm_suspended_storage(void) | |
145 | { | |
146 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) | |
147 | return false; | |
148 | return true; | |
149 | } | |
452aa699 RW |
150 | #endif /* CONFIG_PM_SLEEP */ |
151 | ||
d9c23400 MG |
152 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
153 | int pageblock_order __read_mostly; | |
154 | #endif | |
155 | ||
d98c7a09 | 156 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 157 | |
1da177e4 LT |
158 | /* |
159 | * results with 256, 32 in the lowmem_reserve sysctl: | |
160 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
161 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
162 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
163 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
164 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
165 | * |
166 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
167 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 168 | */ |
2f1b6248 | 169 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 170 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 171 | 256, |
4b51d669 | 172 | #endif |
fb0e7942 | 173 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 174 | 256, |
fb0e7942 | 175 | #endif |
e53ef38d | 176 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 177 | 32, |
e53ef38d | 178 | #endif |
2a1e274a | 179 | 32, |
2f1b6248 | 180 | }; |
1da177e4 LT |
181 | |
182 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 183 | |
15ad7cdc | 184 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 185 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 186 | "DMA", |
4b51d669 | 187 | #endif |
fb0e7942 | 188 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 189 | "DMA32", |
fb0e7942 | 190 | #endif |
2f1b6248 | 191 | "Normal", |
e53ef38d | 192 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 193 | "HighMem", |
e53ef38d | 194 | #endif |
2a1e274a | 195 | "Movable", |
2f1b6248 CL |
196 | }; |
197 | ||
1da177e4 LT |
198 | int min_free_kbytes = 1024; |
199 | ||
2c85f51d JB |
200 | static unsigned long __meminitdata nr_kernel_pages; |
201 | static unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 202 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 203 | |
0ee332c1 | 204 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
34b71f1e | 205 | /* Movable memory ranges, will also be used by memblock subsystem. */ |
01a178a9 TC |
206 | struct movablemem_map movablemem_map = { |
207 | .acpi = false, | |
208 | .nr_map = 0, | |
209 | }; | |
34b71f1e | 210 | |
0ee332c1 TH |
211 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; |
212 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
213 | static unsigned long __initdata required_kernelcore; | |
214 | static unsigned long __initdata required_movablecore; | |
215 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
6981ec31 | 216 | static unsigned long __meminitdata zone_movable_limit[MAX_NUMNODES]; |
0ee332c1 TH |
217 | |
218 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
219 | int movable_zone; | |
220 | EXPORT_SYMBOL(movable_zone); | |
221 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
c713216d | 222 | |
418508c1 MS |
223 | #if MAX_NUMNODES > 1 |
224 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
62bc62a8 | 225 | int nr_online_nodes __read_mostly = 1; |
418508c1 | 226 | EXPORT_SYMBOL(nr_node_ids); |
62bc62a8 | 227 | EXPORT_SYMBOL(nr_online_nodes); |
418508c1 MS |
228 | #endif |
229 | ||
9ef9acb0 MG |
230 | int page_group_by_mobility_disabled __read_mostly; |
231 | ||
ee6f509c | 232 | void set_pageblock_migratetype(struct page *page, int migratetype) |
b2a0ac88 | 233 | { |
49255c61 MG |
234 | |
235 | if (unlikely(page_group_by_mobility_disabled)) | |
236 | migratetype = MIGRATE_UNMOVABLE; | |
237 | ||
b2a0ac88 MG |
238 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
239 | PB_migrate, PB_migrate_end); | |
240 | } | |
241 | ||
7f33d49a RW |
242 | bool oom_killer_disabled __read_mostly; |
243 | ||
13e7444b | 244 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 245 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 246 | { |
bdc8cb98 DH |
247 | int ret = 0; |
248 | unsigned seq; | |
249 | unsigned long pfn = page_to_pfn(page); | |
b5e6a5a2 | 250 | unsigned long sp, start_pfn; |
c6a57e19 | 251 | |
bdc8cb98 DH |
252 | do { |
253 | seq = zone_span_seqbegin(zone); | |
b5e6a5a2 CS |
254 | start_pfn = zone->zone_start_pfn; |
255 | sp = zone->spanned_pages; | |
108bcc96 | 256 | if (!zone_spans_pfn(zone, pfn)) |
bdc8cb98 DH |
257 | ret = 1; |
258 | } while (zone_span_seqretry(zone, seq)); | |
259 | ||
b5e6a5a2 CS |
260 | if (ret) |
261 | pr_err("page %lu outside zone [ %lu - %lu ]\n", | |
262 | pfn, start_pfn, start_pfn + sp); | |
263 | ||
bdc8cb98 | 264 | return ret; |
c6a57e19 DH |
265 | } |
266 | ||
267 | static int page_is_consistent(struct zone *zone, struct page *page) | |
268 | { | |
14e07298 | 269 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 270 | return 0; |
1da177e4 | 271 | if (zone != page_zone(page)) |
c6a57e19 DH |
272 | return 0; |
273 | ||
274 | return 1; | |
275 | } | |
276 | /* | |
277 | * Temporary debugging check for pages not lying within a given zone. | |
278 | */ | |
279 | static int bad_range(struct zone *zone, struct page *page) | |
280 | { | |
281 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 282 | return 1; |
c6a57e19 DH |
283 | if (!page_is_consistent(zone, page)) |
284 | return 1; | |
285 | ||
1da177e4 LT |
286 | return 0; |
287 | } | |
13e7444b NP |
288 | #else |
289 | static inline int bad_range(struct zone *zone, struct page *page) | |
290 | { | |
291 | return 0; | |
292 | } | |
293 | #endif | |
294 | ||
224abf92 | 295 | static void bad_page(struct page *page) |
1da177e4 | 296 | { |
d936cf9b HD |
297 | static unsigned long resume; |
298 | static unsigned long nr_shown; | |
299 | static unsigned long nr_unshown; | |
300 | ||
2a7684a2 WF |
301 | /* Don't complain about poisoned pages */ |
302 | if (PageHWPoison(page)) { | |
22b751c3 | 303 | page_mapcount_reset(page); /* remove PageBuddy */ |
2a7684a2 WF |
304 | return; |
305 | } | |
306 | ||
d936cf9b HD |
307 | /* |
308 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
309 | * or allow a steady drip of one report per second. | |
310 | */ | |
311 | if (nr_shown == 60) { | |
312 | if (time_before(jiffies, resume)) { | |
313 | nr_unshown++; | |
314 | goto out; | |
315 | } | |
316 | if (nr_unshown) { | |
1e9e6365 HD |
317 | printk(KERN_ALERT |
318 | "BUG: Bad page state: %lu messages suppressed\n", | |
d936cf9b HD |
319 | nr_unshown); |
320 | nr_unshown = 0; | |
321 | } | |
322 | nr_shown = 0; | |
323 | } | |
324 | if (nr_shown++ == 0) | |
325 | resume = jiffies + 60 * HZ; | |
326 | ||
1e9e6365 | 327 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
3dc14741 | 328 | current->comm, page_to_pfn(page)); |
718a3821 | 329 | dump_page(page); |
3dc14741 | 330 | |
4f31888c | 331 | print_modules(); |
1da177e4 | 332 | dump_stack(); |
d936cf9b | 333 | out: |
8cc3b392 | 334 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
22b751c3 | 335 | page_mapcount_reset(page); /* remove PageBuddy */ |
373d4d09 | 336 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 LT |
337 | } |
338 | ||
1da177e4 LT |
339 | /* |
340 | * Higher-order pages are called "compound pages". They are structured thusly: | |
341 | * | |
342 | * The first PAGE_SIZE page is called the "head page". | |
343 | * | |
344 | * The remaining PAGE_SIZE pages are called "tail pages". | |
345 | * | |
6416b9fa WSH |
346 | * All pages have PG_compound set. All tail pages have their ->first_page |
347 | * pointing at the head page. | |
1da177e4 | 348 | * |
41d78ba5 HD |
349 | * The first tail page's ->lru.next holds the address of the compound page's |
350 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
351 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 352 | */ |
d98c7a09 HD |
353 | |
354 | static void free_compound_page(struct page *page) | |
355 | { | |
d85f3385 | 356 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
357 | } |
358 | ||
01ad1c08 | 359 | void prep_compound_page(struct page *page, unsigned long order) |
18229df5 AW |
360 | { |
361 | int i; | |
362 | int nr_pages = 1 << order; | |
363 | ||
364 | set_compound_page_dtor(page, free_compound_page); | |
365 | set_compound_order(page, order); | |
366 | __SetPageHead(page); | |
367 | for (i = 1; i < nr_pages; i++) { | |
368 | struct page *p = page + i; | |
18229df5 | 369 | __SetPageTail(p); |
58a84aa9 | 370 | set_page_count(p, 0); |
18229df5 AW |
371 | p->first_page = page; |
372 | } | |
373 | } | |
374 | ||
59ff4216 | 375 | /* update __split_huge_page_refcount if you change this function */ |
8cc3b392 | 376 | static int destroy_compound_page(struct page *page, unsigned long order) |
1da177e4 LT |
377 | { |
378 | int i; | |
379 | int nr_pages = 1 << order; | |
8cc3b392 | 380 | int bad = 0; |
1da177e4 | 381 | |
0bb2c763 | 382 | if (unlikely(compound_order(page) != order)) { |
224abf92 | 383 | bad_page(page); |
8cc3b392 HD |
384 | bad++; |
385 | } | |
1da177e4 | 386 | |
6d777953 | 387 | __ClearPageHead(page); |
8cc3b392 | 388 | |
18229df5 AW |
389 | for (i = 1; i < nr_pages; i++) { |
390 | struct page *p = page + i; | |
1da177e4 | 391 | |
e713a21d | 392 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
224abf92 | 393 | bad_page(page); |
8cc3b392 HD |
394 | bad++; |
395 | } | |
d85f3385 | 396 | __ClearPageTail(p); |
1da177e4 | 397 | } |
8cc3b392 HD |
398 | |
399 | return bad; | |
1da177e4 | 400 | } |
1da177e4 | 401 | |
17cf4406 NP |
402 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
403 | { | |
404 | int i; | |
405 | ||
6626c5d5 AM |
406 | /* |
407 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
408 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
409 | */ | |
725d704e | 410 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
411 | for (i = 0; i < (1 << order); i++) |
412 | clear_highpage(page + i); | |
413 | } | |
414 | ||
c0a32fc5 SG |
415 | #ifdef CONFIG_DEBUG_PAGEALLOC |
416 | unsigned int _debug_guardpage_minorder; | |
417 | ||
418 | static int __init debug_guardpage_minorder_setup(char *buf) | |
419 | { | |
420 | unsigned long res; | |
421 | ||
422 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { | |
423 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); | |
424 | return 0; | |
425 | } | |
426 | _debug_guardpage_minorder = res; | |
427 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); | |
428 | return 0; | |
429 | } | |
430 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); | |
431 | ||
432 | static inline void set_page_guard_flag(struct page *page) | |
433 | { | |
434 | __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
435 | } | |
436 | ||
437 | static inline void clear_page_guard_flag(struct page *page) | |
438 | { | |
439 | __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); | |
440 | } | |
441 | #else | |
442 | static inline void set_page_guard_flag(struct page *page) { } | |
443 | static inline void clear_page_guard_flag(struct page *page) { } | |
444 | #endif | |
445 | ||
6aa3001b AM |
446 | static inline void set_page_order(struct page *page, int order) |
447 | { | |
4c21e2f2 | 448 | set_page_private(page, order); |
676165a8 | 449 | __SetPageBuddy(page); |
1da177e4 LT |
450 | } |
451 | ||
452 | static inline void rmv_page_order(struct page *page) | |
453 | { | |
676165a8 | 454 | __ClearPageBuddy(page); |
4c21e2f2 | 455 | set_page_private(page, 0); |
1da177e4 LT |
456 | } |
457 | ||
458 | /* | |
459 | * Locate the struct page for both the matching buddy in our | |
460 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
461 | * | |
462 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
463 | * the following equation: | |
464 | * B2 = B1 ^ (1 << O) | |
465 | * For example, if the starting buddy (buddy2) is #8 its order | |
466 | * 1 buddy is #10: | |
467 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
468 | * | |
469 | * 2) Any buddy B will have an order O+1 parent P which | |
470 | * satisfies the following equation: | |
471 | * P = B & ~(1 << O) | |
472 | * | |
d6e05edc | 473 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 | 474 | */ |
1da177e4 | 475 | static inline unsigned long |
43506fad | 476 | __find_buddy_index(unsigned long page_idx, unsigned int order) |
1da177e4 | 477 | { |
43506fad | 478 | return page_idx ^ (1 << order); |
1da177e4 LT |
479 | } |
480 | ||
481 | /* | |
482 | * This function checks whether a page is free && is the buddy | |
483 | * we can do coalesce a page and its buddy if | |
13e7444b | 484 | * (a) the buddy is not in a hole && |
676165a8 | 485 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
486 | * (c) a page and its buddy have the same order && |
487 | * (d) a page and its buddy are in the same zone. | |
676165a8 | 488 | * |
5f24ce5f AA |
489 | * For recording whether a page is in the buddy system, we set ->_mapcount -2. |
490 | * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock. | |
1da177e4 | 491 | * |
676165a8 | 492 | * For recording page's order, we use page_private(page). |
1da177e4 | 493 | */ |
cb2b95e1 AW |
494 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
495 | int order) | |
1da177e4 | 496 | { |
14e07298 | 497 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 498 | return 0; |
13e7444b | 499 | |
cb2b95e1 AW |
500 | if (page_zone_id(page) != page_zone_id(buddy)) |
501 | return 0; | |
502 | ||
c0a32fc5 SG |
503 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
504 | VM_BUG_ON(page_count(buddy) != 0); | |
505 | return 1; | |
506 | } | |
507 | ||
cb2b95e1 | 508 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
a3af9c38 | 509 | VM_BUG_ON(page_count(buddy) != 0); |
6aa3001b | 510 | return 1; |
676165a8 | 511 | } |
6aa3001b | 512 | return 0; |
1da177e4 LT |
513 | } |
514 | ||
515 | /* | |
516 | * Freeing function for a buddy system allocator. | |
517 | * | |
518 | * The concept of a buddy system is to maintain direct-mapped table | |
519 | * (containing bit values) for memory blocks of various "orders". | |
520 | * The bottom level table contains the map for the smallest allocatable | |
521 | * units of memory (here, pages), and each level above it describes | |
522 | * pairs of units from the levels below, hence, "buddies". | |
523 | * At a high level, all that happens here is marking the table entry | |
524 | * at the bottom level available, and propagating the changes upward | |
525 | * as necessary, plus some accounting needed to play nicely with other | |
526 | * parts of the VM system. | |
527 | * At each level, we keep a list of pages, which are heads of continuous | |
5f24ce5f | 528 | * free pages of length of (1 << order) and marked with _mapcount -2. Page's |
4c21e2f2 | 529 | * order is recorded in page_private(page) field. |
1da177e4 | 530 | * So when we are allocating or freeing one, we can derive the state of the |
5f63b720 MN |
531 | * other. That is, if we allocate a small block, and both were |
532 | * free, the remainder of the region must be split into blocks. | |
1da177e4 | 533 | * If a block is freed, and its buddy is also free, then this |
5f63b720 | 534 | * triggers coalescing into a block of larger size. |
1da177e4 | 535 | * |
6d49e352 | 536 | * -- nyc |
1da177e4 LT |
537 | */ |
538 | ||
48db57f8 | 539 | static inline void __free_one_page(struct page *page, |
ed0ae21d MG |
540 | struct zone *zone, unsigned int order, |
541 | int migratetype) | |
1da177e4 LT |
542 | { |
543 | unsigned long page_idx; | |
6dda9d55 | 544 | unsigned long combined_idx; |
43506fad | 545 | unsigned long uninitialized_var(buddy_idx); |
6dda9d55 | 546 | struct page *buddy; |
1da177e4 | 547 | |
d29bb978 CS |
548 | VM_BUG_ON(!zone_is_initialized(zone)); |
549 | ||
224abf92 | 550 | if (unlikely(PageCompound(page))) |
8cc3b392 HD |
551 | if (unlikely(destroy_compound_page(page, order))) |
552 | return; | |
1da177e4 | 553 | |
ed0ae21d MG |
554 | VM_BUG_ON(migratetype == -1); |
555 | ||
1da177e4 LT |
556 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
557 | ||
f2260e6b | 558 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
725d704e | 559 | VM_BUG_ON(bad_range(zone, page)); |
1da177e4 | 560 | |
1da177e4 | 561 | while (order < MAX_ORDER-1) { |
43506fad KC |
562 | buddy_idx = __find_buddy_index(page_idx, order); |
563 | buddy = page + (buddy_idx - page_idx); | |
cb2b95e1 | 564 | if (!page_is_buddy(page, buddy, order)) |
3c82d0ce | 565 | break; |
c0a32fc5 SG |
566 | /* |
567 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, | |
568 | * merge with it and move up one order. | |
569 | */ | |
570 | if (page_is_guard(buddy)) { | |
571 | clear_page_guard_flag(buddy); | |
572 | set_page_private(page, 0); | |
d1ce749a BZ |
573 | __mod_zone_freepage_state(zone, 1 << order, |
574 | migratetype); | |
c0a32fc5 SG |
575 | } else { |
576 | list_del(&buddy->lru); | |
577 | zone->free_area[order].nr_free--; | |
578 | rmv_page_order(buddy); | |
579 | } | |
43506fad | 580 | combined_idx = buddy_idx & page_idx; |
1da177e4 LT |
581 | page = page + (combined_idx - page_idx); |
582 | page_idx = combined_idx; | |
583 | order++; | |
584 | } | |
585 | set_page_order(page, order); | |
6dda9d55 CZ |
586 | |
587 | /* | |
588 | * If this is not the largest possible page, check if the buddy | |
589 | * of the next-highest order is free. If it is, it's possible | |
590 | * that pages are being freed that will coalesce soon. In case, | |
591 | * that is happening, add the free page to the tail of the list | |
592 | * so it's less likely to be used soon and more likely to be merged | |
593 | * as a higher order page | |
594 | */ | |
b7f50cfa | 595 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
6dda9d55 | 596 | struct page *higher_page, *higher_buddy; |
43506fad KC |
597 | combined_idx = buddy_idx & page_idx; |
598 | higher_page = page + (combined_idx - page_idx); | |
599 | buddy_idx = __find_buddy_index(combined_idx, order + 1); | |
0ba8f2d5 | 600 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
6dda9d55 CZ |
601 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
602 | list_add_tail(&page->lru, | |
603 | &zone->free_area[order].free_list[migratetype]); | |
604 | goto out; | |
605 | } | |
606 | } | |
607 | ||
608 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); | |
609 | out: | |
1da177e4 LT |
610 | zone->free_area[order].nr_free++; |
611 | } | |
612 | ||
224abf92 | 613 | static inline int free_pages_check(struct page *page) |
1da177e4 | 614 | { |
92be2e33 NP |
615 | if (unlikely(page_mapcount(page) | |
616 | (page->mapping != NULL) | | |
a3af9c38 | 617 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
618 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE) | |
619 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 620 | bad_page(page); |
79f4b7bf | 621 | return 1; |
8cc3b392 | 622 | } |
22b751c3 | 623 | page_nid_reset_last(page); |
79f4b7bf HD |
624 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
625 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
626 | return 0; | |
1da177e4 LT |
627 | } |
628 | ||
629 | /* | |
5f8dcc21 | 630 | * Frees a number of pages from the PCP lists |
1da177e4 | 631 | * Assumes all pages on list are in same zone, and of same order. |
207f36ee | 632 | * count is the number of pages to free. |
1da177e4 LT |
633 | * |
634 | * If the zone was previously in an "all pages pinned" state then look to | |
635 | * see if this freeing clears that state. | |
636 | * | |
637 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
638 | * pinned" detection logic. | |
639 | */ | |
5f8dcc21 MG |
640 | static void free_pcppages_bulk(struct zone *zone, int count, |
641 | struct per_cpu_pages *pcp) | |
1da177e4 | 642 | { |
5f8dcc21 | 643 | int migratetype = 0; |
a6f9edd6 | 644 | int batch_free = 0; |
72853e29 | 645 | int to_free = count; |
5f8dcc21 | 646 | |
c54ad30c | 647 | spin_lock(&zone->lock); |
93e4a89a | 648 | zone->all_unreclaimable = 0; |
1da177e4 | 649 | zone->pages_scanned = 0; |
f2260e6b | 650 | |
72853e29 | 651 | while (to_free) { |
48db57f8 | 652 | struct page *page; |
5f8dcc21 MG |
653 | struct list_head *list; |
654 | ||
655 | /* | |
a6f9edd6 MG |
656 | * Remove pages from lists in a round-robin fashion. A |
657 | * batch_free count is maintained that is incremented when an | |
658 | * empty list is encountered. This is so more pages are freed | |
659 | * off fuller lists instead of spinning excessively around empty | |
660 | * lists | |
5f8dcc21 MG |
661 | */ |
662 | do { | |
a6f9edd6 | 663 | batch_free++; |
5f8dcc21 MG |
664 | if (++migratetype == MIGRATE_PCPTYPES) |
665 | migratetype = 0; | |
666 | list = &pcp->lists[migratetype]; | |
667 | } while (list_empty(list)); | |
48db57f8 | 668 | |
1d16871d NK |
669 | /* This is the only non-empty list. Free them all. */ |
670 | if (batch_free == MIGRATE_PCPTYPES) | |
671 | batch_free = to_free; | |
672 | ||
a6f9edd6 | 673 | do { |
770c8aaa BZ |
674 | int mt; /* migratetype of the to-be-freed page */ |
675 | ||
a6f9edd6 MG |
676 | page = list_entry(list->prev, struct page, lru); |
677 | /* must delete as __free_one_page list manipulates */ | |
678 | list_del(&page->lru); | |
b12c4ad1 | 679 | mt = get_freepage_migratetype(page); |
a7016235 | 680 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
770c8aaa BZ |
681 | __free_one_page(page, zone, 0, mt); |
682 | trace_mm_page_pcpu_drain(page, 0, mt); | |
194159fb | 683 | if (likely(!is_migrate_isolate_page(page))) { |
97d0da22 WC |
684 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1); |
685 | if (is_migrate_cma(mt)) | |
686 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1); | |
687 | } | |
72853e29 | 688 | } while (--to_free && --batch_free && !list_empty(list)); |
1da177e4 | 689 | } |
c54ad30c | 690 | spin_unlock(&zone->lock); |
1da177e4 LT |
691 | } |
692 | ||
ed0ae21d MG |
693 | static void free_one_page(struct zone *zone, struct page *page, int order, |
694 | int migratetype) | |
1da177e4 | 695 | { |
006d22d9 | 696 | spin_lock(&zone->lock); |
93e4a89a | 697 | zone->all_unreclaimable = 0; |
006d22d9 | 698 | zone->pages_scanned = 0; |
f2260e6b | 699 | |
ed0ae21d | 700 | __free_one_page(page, zone, order, migratetype); |
194159fb | 701 | if (unlikely(!is_migrate_isolate(migratetype))) |
d1ce749a | 702 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
006d22d9 | 703 | spin_unlock(&zone->lock); |
48db57f8 NP |
704 | } |
705 | ||
ec95f53a | 706 | static bool free_pages_prepare(struct page *page, unsigned int order) |
48db57f8 | 707 | { |
1da177e4 | 708 | int i; |
8cc3b392 | 709 | int bad = 0; |
1da177e4 | 710 | |
b413d48a | 711 | trace_mm_page_free(page, order); |
b1eeab67 VN |
712 | kmemcheck_free_shadow(page, order); |
713 | ||
8dd60a3a AA |
714 | if (PageAnon(page)) |
715 | page->mapping = NULL; | |
716 | for (i = 0; i < (1 << order); i++) | |
717 | bad += free_pages_check(page + i); | |
8cc3b392 | 718 | if (bad) |
ec95f53a | 719 | return false; |
689bcebf | 720 | |
3ac7fe5a | 721 | if (!PageHighMem(page)) { |
9858db50 | 722 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
3ac7fe5a TG |
723 | debug_check_no_obj_freed(page_address(page), |
724 | PAGE_SIZE << order); | |
725 | } | |
dafb1367 | 726 | arch_free_page(page, order); |
48db57f8 | 727 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 728 | |
ec95f53a KM |
729 | return true; |
730 | } | |
731 | ||
732 | static void __free_pages_ok(struct page *page, unsigned int order) | |
733 | { | |
734 | unsigned long flags; | |
95e34412 | 735 | int migratetype; |
ec95f53a KM |
736 | |
737 | if (!free_pages_prepare(page, order)) | |
738 | return; | |
739 | ||
c54ad30c | 740 | local_irq_save(flags); |
f8891e5e | 741 | __count_vm_events(PGFREE, 1 << order); |
95e34412 MK |
742 | migratetype = get_pageblock_migratetype(page); |
743 | set_freepage_migratetype(page, migratetype); | |
744 | free_one_page(page_zone(page), page, order, migratetype); | |
c54ad30c | 745 | local_irq_restore(flags); |
1da177e4 LT |
746 | } |
747 | ||
9feedc9d JL |
748 | /* |
749 | * Read access to zone->managed_pages is safe because it's unsigned long, | |
750 | * but we still need to serialize writers. Currently all callers of | |
751 | * __free_pages_bootmem() except put_page_bootmem() should only be used | |
752 | * at boot time. So for shorter boot time, we shift the burden to | |
753 | * put_page_bootmem() to serialize writers. | |
754 | */ | |
af370fb8 | 755 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 | 756 | { |
c3993076 JW |
757 | unsigned int nr_pages = 1 << order; |
758 | unsigned int loop; | |
a226f6c8 | 759 | |
c3993076 JW |
760 | prefetchw(page); |
761 | for (loop = 0; loop < nr_pages; loop++) { | |
762 | struct page *p = &page[loop]; | |
763 | ||
764 | if (loop + 1 < nr_pages) | |
765 | prefetchw(p + 1); | |
766 | __ClearPageReserved(p); | |
767 | set_page_count(p, 0); | |
a226f6c8 | 768 | } |
c3993076 | 769 | |
9feedc9d | 770 | page_zone(page)->managed_pages += 1 << order; |
c3993076 JW |
771 | set_page_refcounted(page); |
772 | __free_pages(page, order); | |
a226f6c8 DH |
773 | } |
774 | ||
47118af0 MN |
775 | #ifdef CONFIG_CMA |
776 | /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */ | |
777 | void __init init_cma_reserved_pageblock(struct page *page) | |
778 | { | |
779 | unsigned i = pageblock_nr_pages; | |
780 | struct page *p = page; | |
781 | ||
782 | do { | |
783 | __ClearPageReserved(p); | |
784 | set_page_count(p, 0); | |
785 | } while (++p, --i); | |
786 | ||
787 | set_page_refcounted(page); | |
788 | set_pageblock_migratetype(page, MIGRATE_CMA); | |
789 | __free_pages(page, pageblock_order); | |
790 | totalram_pages += pageblock_nr_pages; | |
41a79734 MS |
791 | #ifdef CONFIG_HIGHMEM |
792 | if (PageHighMem(page)) | |
793 | totalhigh_pages += pageblock_nr_pages; | |
794 | #endif | |
47118af0 MN |
795 | } |
796 | #endif | |
1da177e4 LT |
797 | |
798 | /* | |
799 | * The order of subdivision here is critical for the IO subsystem. | |
800 | * Please do not alter this order without good reasons and regression | |
801 | * testing. Specifically, as large blocks of memory are subdivided, | |
802 | * the order in which smaller blocks are delivered depends on the order | |
803 | * they're subdivided in this function. This is the primary factor | |
804 | * influencing the order in which pages are delivered to the IO | |
805 | * subsystem according to empirical testing, and this is also justified | |
806 | * by considering the behavior of a buddy system containing a single | |
807 | * large block of memory acted on by a series of small allocations. | |
808 | * This behavior is a critical factor in sglist merging's success. | |
809 | * | |
6d49e352 | 810 | * -- nyc |
1da177e4 | 811 | */ |
085cc7d5 | 812 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
813 | int low, int high, struct free_area *area, |
814 | int migratetype) | |
1da177e4 LT |
815 | { |
816 | unsigned long size = 1 << high; | |
817 | ||
818 | while (high > low) { | |
819 | area--; | |
820 | high--; | |
821 | size >>= 1; | |
725d704e | 822 | VM_BUG_ON(bad_range(zone, &page[size])); |
c0a32fc5 SG |
823 | |
824 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
825 | if (high < debug_guardpage_minorder()) { | |
826 | /* | |
827 | * Mark as guard pages (or page), that will allow to | |
828 | * merge back to allocator when buddy will be freed. | |
829 | * Corresponding page table entries will not be touched, | |
830 | * pages will stay not present in virtual address space | |
831 | */ | |
832 | INIT_LIST_HEAD(&page[size].lru); | |
833 | set_page_guard_flag(&page[size]); | |
834 | set_page_private(&page[size], high); | |
835 | /* Guard pages are not available for any usage */ | |
d1ce749a BZ |
836 | __mod_zone_freepage_state(zone, -(1 << high), |
837 | migratetype); | |
c0a32fc5 SG |
838 | continue; |
839 | } | |
840 | #endif | |
b2a0ac88 | 841 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
842 | area->nr_free++; |
843 | set_page_order(&page[size], high); | |
844 | } | |
1da177e4 LT |
845 | } |
846 | ||
1da177e4 LT |
847 | /* |
848 | * This page is about to be returned from the page allocator | |
849 | */ | |
2a7684a2 | 850 | static inline int check_new_page(struct page *page) |
1da177e4 | 851 | { |
92be2e33 NP |
852 | if (unlikely(page_mapcount(page) | |
853 | (page->mapping != NULL) | | |
a3af9c38 | 854 | (atomic_read(&page->_count) != 0) | |
f212ad7c DN |
855 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP) | |
856 | (mem_cgroup_bad_page_check(page)))) { | |
224abf92 | 857 | bad_page(page); |
689bcebf | 858 | return 1; |
8cc3b392 | 859 | } |
2a7684a2 WF |
860 | return 0; |
861 | } | |
862 | ||
863 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
864 | { | |
865 | int i; | |
866 | ||
867 | for (i = 0; i < (1 << order); i++) { | |
868 | struct page *p = page + i; | |
869 | if (unlikely(check_new_page(p))) | |
870 | return 1; | |
871 | } | |
689bcebf | 872 | |
4c21e2f2 | 873 | set_page_private(page, 0); |
7835e98b | 874 | set_page_refcounted(page); |
cc102509 NP |
875 | |
876 | arch_alloc_page(page, order); | |
1da177e4 | 877 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
878 | |
879 | if (gfp_flags & __GFP_ZERO) | |
880 | prep_zero_page(page, order, gfp_flags); | |
881 | ||
882 | if (order && (gfp_flags & __GFP_COMP)) | |
883 | prep_compound_page(page, order); | |
884 | ||
689bcebf | 885 | return 0; |
1da177e4 LT |
886 | } |
887 | ||
56fd56b8 MG |
888 | /* |
889 | * Go through the free lists for the given migratetype and remove | |
890 | * the smallest available page from the freelists | |
891 | */ | |
728ec980 MG |
892 | static inline |
893 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
56fd56b8 MG |
894 | int migratetype) |
895 | { | |
896 | unsigned int current_order; | |
897 | struct free_area * area; | |
898 | struct page *page; | |
899 | ||
900 | /* Find a page of the appropriate size in the preferred list */ | |
901 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
902 | area = &(zone->free_area[current_order]); | |
903 | if (list_empty(&area->free_list[migratetype])) | |
904 | continue; | |
905 | ||
906 | page = list_entry(area->free_list[migratetype].next, | |
907 | struct page, lru); | |
908 | list_del(&page->lru); | |
909 | rmv_page_order(page); | |
910 | area->nr_free--; | |
56fd56b8 MG |
911 | expand(zone, page, order, current_order, area, migratetype); |
912 | return page; | |
913 | } | |
914 | ||
915 | return NULL; | |
916 | } | |
917 | ||
918 | ||
b2a0ac88 MG |
919 | /* |
920 | * This array describes the order lists are fallen back to when | |
921 | * the free lists for the desirable migrate type are depleted | |
922 | */ | |
47118af0 MN |
923 | static int fallbacks[MIGRATE_TYPES][4] = { |
924 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
925 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
926 | #ifdef CONFIG_CMA | |
927 | [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
928 | [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */ | |
929 | #else | |
930 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
931 | #endif | |
6d4a4916 | 932 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 933 | #ifdef CONFIG_MEMORY_ISOLATION |
6d4a4916 | 934 | [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */ |
194159fb | 935 | #endif |
b2a0ac88 MG |
936 | }; |
937 | ||
c361be55 MG |
938 | /* |
939 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 940 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
941 | * boundary. If alignment is required, use move_freepages_block() |
942 | */ | |
435b405c | 943 | int move_freepages(struct zone *zone, |
b69a7288 AB |
944 | struct page *start_page, struct page *end_page, |
945 | int migratetype) | |
c361be55 MG |
946 | { |
947 | struct page *page; | |
948 | unsigned long order; | |
d100313f | 949 | int pages_moved = 0; |
c361be55 MG |
950 | |
951 | #ifndef CONFIG_HOLES_IN_ZONE | |
952 | /* | |
953 | * page_zone is not safe to call in this context when | |
954 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
955 | * anyway as we check zone boundaries in move_freepages_block(). | |
956 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 957 | * grouping pages by mobility |
c361be55 MG |
958 | */ |
959 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
960 | #endif | |
961 | ||
962 | for (page = start_page; page <= end_page;) { | |
344c790e AL |
963 | /* Make sure we are not inadvertently changing nodes */ |
964 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
965 | ||
c361be55 MG |
966 | if (!pfn_valid_within(page_to_pfn(page))) { |
967 | page++; | |
968 | continue; | |
969 | } | |
970 | ||
971 | if (!PageBuddy(page)) { | |
972 | page++; | |
973 | continue; | |
974 | } | |
975 | ||
976 | order = page_order(page); | |
84be48d8 KS |
977 | list_move(&page->lru, |
978 | &zone->free_area[order].free_list[migratetype]); | |
95e34412 | 979 | set_freepage_migratetype(page, migratetype); |
c361be55 | 980 | page += 1 << order; |
d100313f | 981 | pages_moved += 1 << order; |
c361be55 MG |
982 | } |
983 | ||
d100313f | 984 | return pages_moved; |
c361be55 MG |
985 | } |
986 | ||
ee6f509c | 987 | int move_freepages_block(struct zone *zone, struct page *page, |
68e3e926 | 988 | int migratetype) |
c361be55 MG |
989 | { |
990 | unsigned long start_pfn, end_pfn; | |
991 | struct page *start_page, *end_page; | |
992 | ||
993 | start_pfn = page_to_pfn(page); | |
d9c23400 | 994 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 995 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
996 | end_page = start_page + pageblock_nr_pages - 1; |
997 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
998 | |
999 | /* Do not cross zone boundaries */ | |
108bcc96 | 1000 | if (!zone_spans_pfn(zone, start_pfn)) |
c361be55 | 1001 | start_page = page; |
108bcc96 | 1002 | if (!zone_spans_pfn(zone, end_pfn)) |
c361be55 MG |
1003 | return 0; |
1004 | ||
1005 | return move_freepages(zone, start_page, end_page, migratetype); | |
1006 | } | |
1007 | ||
2f66a68f MG |
1008 | static void change_pageblock_range(struct page *pageblock_page, |
1009 | int start_order, int migratetype) | |
1010 | { | |
1011 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
1012 | ||
1013 | while (nr_pageblocks--) { | |
1014 | set_pageblock_migratetype(pageblock_page, migratetype); | |
1015 | pageblock_page += pageblock_nr_pages; | |
1016 | } | |
1017 | } | |
1018 | ||
b2a0ac88 | 1019 | /* Remove an element from the buddy allocator from the fallback list */ |
0ac3a409 MG |
1020 | static inline struct page * |
1021 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
b2a0ac88 MG |
1022 | { |
1023 | struct free_area * area; | |
1024 | int current_order; | |
1025 | struct page *page; | |
1026 | int migratetype, i; | |
1027 | ||
1028 | /* Find the largest possible block of pages in the other list */ | |
1029 | for (current_order = MAX_ORDER-1; current_order >= order; | |
1030 | --current_order) { | |
6d4a4916 | 1031 | for (i = 0;; i++) { |
b2a0ac88 MG |
1032 | migratetype = fallbacks[start_migratetype][i]; |
1033 | ||
56fd56b8 MG |
1034 | /* MIGRATE_RESERVE handled later if necessary */ |
1035 | if (migratetype == MIGRATE_RESERVE) | |
6d4a4916 | 1036 | break; |
e010487d | 1037 | |
b2a0ac88 MG |
1038 | area = &(zone->free_area[current_order]); |
1039 | if (list_empty(&area->free_list[migratetype])) | |
1040 | continue; | |
1041 | ||
1042 | page = list_entry(area->free_list[migratetype].next, | |
1043 | struct page, lru); | |
1044 | area->nr_free--; | |
1045 | ||
1046 | /* | |
c361be55 | 1047 | * If breaking a large block of pages, move all free |
46dafbca MG |
1048 | * pages to the preferred allocation list. If falling |
1049 | * back for a reclaimable kernel allocation, be more | |
25985edc | 1050 | * aggressive about taking ownership of free pages |
47118af0 MN |
1051 | * |
1052 | * On the other hand, never change migration | |
1053 | * type of MIGRATE_CMA pageblocks nor move CMA | |
1054 | * pages on different free lists. We don't | |
1055 | * want unmovable pages to be allocated from | |
1056 | * MIGRATE_CMA areas. | |
b2a0ac88 | 1057 | */ |
47118af0 MN |
1058 | if (!is_migrate_cma(migratetype) && |
1059 | (unlikely(current_order >= pageblock_order / 2) || | |
1060 | start_migratetype == MIGRATE_RECLAIMABLE || | |
1061 | page_group_by_mobility_disabled)) { | |
1062 | int pages; | |
46dafbca MG |
1063 | pages = move_freepages_block(zone, page, |
1064 | start_migratetype); | |
1065 | ||
1066 | /* Claim the whole block if over half of it is free */ | |
dd5d241e MG |
1067 | if (pages >= (1 << (pageblock_order-1)) || |
1068 | page_group_by_mobility_disabled) | |
46dafbca MG |
1069 | set_pageblock_migratetype(page, |
1070 | start_migratetype); | |
1071 | ||
b2a0ac88 | 1072 | migratetype = start_migratetype; |
c361be55 | 1073 | } |
b2a0ac88 MG |
1074 | |
1075 | /* Remove the page from the freelists */ | |
1076 | list_del(&page->lru); | |
1077 | rmv_page_order(page); | |
b2a0ac88 | 1078 | |
2f66a68f | 1079 | /* Take ownership for orders >= pageblock_order */ |
47118af0 MN |
1080 | if (current_order >= pageblock_order && |
1081 | !is_migrate_cma(migratetype)) | |
2f66a68f | 1082 | change_pageblock_range(page, current_order, |
b2a0ac88 MG |
1083 | start_migratetype); |
1084 | ||
47118af0 MN |
1085 | expand(zone, page, order, current_order, area, |
1086 | is_migrate_cma(migratetype) | |
1087 | ? migratetype : start_migratetype); | |
e0fff1bd MG |
1088 | |
1089 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
1090 | start_migratetype, migratetype); | |
1091 | ||
b2a0ac88 MG |
1092 | return page; |
1093 | } | |
1094 | } | |
1095 | ||
728ec980 | 1096 | return NULL; |
b2a0ac88 MG |
1097 | } |
1098 | ||
56fd56b8 | 1099 | /* |
1da177e4 LT |
1100 | * Do the hard work of removing an element from the buddy allocator. |
1101 | * Call me with the zone->lock already held. | |
1102 | */ | |
b2a0ac88 MG |
1103 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
1104 | int migratetype) | |
1da177e4 | 1105 | { |
1da177e4 LT |
1106 | struct page *page; |
1107 | ||
728ec980 | 1108 | retry_reserve: |
56fd56b8 | 1109 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 1110 | |
728ec980 | 1111 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
56fd56b8 | 1112 | page = __rmqueue_fallback(zone, order, migratetype); |
b2a0ac88 | 1113 | |
728ec980 MG |
1114 | /* |
1115 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
1116 | * is used because __rmqueue_smallest is an inline function | |
1117 | * and we want just one call site | |
1118 | */ | |
1119 | if (!page) { | |
1120 | migratetype = MIGRATE_RESERVE; | |
1121 | goto retry_reserve; | |
1122 | } | |
1123 | } | |
1124 | ||
0d3d062a | 1125 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
b2a0ac88 | 1126 | return page; |
1da177e4 LT |
1127 | } |
1128 | ||
5f63b720 | 1129 | /* |
1da177e4 LT |
1130 | * Obtain a specified number of elements from the buddy allocator, all under |
1131 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
1132 | * Returns the number of new pages which were placed at *list. | |
1133 | */ | |
5f63b720 | 1134 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
b2a0ac88 | 1135 | unsigned long count, struct list_head *list, |
e084b2d9 | 1136 | int migratetype, int cold) |
1da177e4 | 1137 | { |
47118af0 | 1138 | int mt = migratetype, i; |
5f63b720 | 1139 | |
c54ad30c | 1140 | spin_lock(&zone->lock); |
1da177e4 | 1141 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 1142 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 1143 | if (unlikely(page == NULL)) |
1da177e4 | 1144 | break; |
81eabcbe MG |
1145 | |
1146 | /* | |
1147 | * Split buddy pages returned by expand() are received here | |
1148 | * in physical page order. The page is added to the callers and | |
1149 | * list and the list head then moves forward. From the callers | |
1150 | * perspective, the linked list is ordered by page number in | |
1151 | * some conditions. This is useful for IO devices that can | |
1152 | * merge IO requests if the physical pages are ordered | |
1153 | * properly. | |
1154 | */ | |
e084b2d9 MG |
1155 | if (likely(cold == 0)) |
1156 | list_add(&page->lru, list); | |
1157 | else | |
1158 | list_add_tail(&page->lru, list); | |
47118af0 MN |
1159 | if (IS_ENABLED(CONFIG_CMA)) { |
1160 | mt = get_pageblock_migratetype(page); | |
194159fb | 1161 | if (!is_migrate_cma(mt) && !is_migrate_isolate(mt)) |
47118af0 MN |
1162 | mt = migratetype; |
1163 | } | |
b12c4ad1 | 1164 | set_freepage_migratetype(page, mt); |
81eabcbe | 1165 | list = &page->lru; |
d1ce749a BZ |
1166 | if (is_migrate_cma(mt)) |
1167 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, | |
1168 | -(1 << order)); | |
1da177e4 | 1169 | } |
f2260e6b | 1170 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
c54ad30c | 1171 | spin_unlock(&zone->lock); |
085cc7d5 | 1172 | return i; |
1da177e4 LT |
1173 | } |
1174 | ||
4ae7c039 | 1175 | #ifdef CONFIG_NUMA |
8fce4d8e | 1176 | /* |
4037d452 CL |
1177 | * Called from the vmstat counter updater to drain pagesets of this |
1178 | * currently executing processor on remote nodes after they have | |
1179 | * expired. | |
1180 | * | |
879336c3 CL |
1181 | * Note that this function must be called with the thread pinned to |
1182 | * a single processor. | |
8fce4d8e | 1183 | */ |
4037d452 | 1184 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 1185 | { |
4ae7c039 | 1186 | unsigned long flags; |
4037d452 | 1187 | int to_drain; |
4ae7c039 | 1188 | |
4037d452 CL |
1189 | local_irq_save(flags); |
1190 | if (pcp->count >= pcp->batch) | |
1191 | to_drain = pcp->batch; | |
1192 | else | |
1193 | to_drain = pcp->count; | |
2a13515c KM |
1194 | if (to_drain > 0) { |
1195 | free_pcppages_bulk(zone, to_drain, pcp); | |
1196 | pcp->count -= to_drain; | |
1197 | } | |
4037d452 | 1198 | local_irq_restore(flags); |
4ae7c039 CL |
1199 | } |
1200 | #endif | |
1201 | ||
9f8f2172 CL |
1202 | /* |
1203 | * Drain pages of the indicated processor. | |
1204 | * | |
1205 | * The processor must either be the current processor and the | |
1206 | * thread pinned to the current processor or a processor that | |
1207 | * is not online. | |
1208 | */ | |
1209 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 1210 | { |
c54ad30c | 1211 | unsigned long flags; |
1da177e4 | 1212 | struct zone *zone; |
1da177e4 | 1213 | |
ee99c71c | 1214 | for_each_populated_zone(zone) { |
1da177e4 | 1215 | struct per_cpu_pageset *pset; |
3dfa5721 | 1216 | struct per_cpu_pages *pcp; |
1da177e4 | 1217 | |
99dcc3e5 CL |
1218 | local_irq_save(flags); |
1219 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3dfa5721 CL |
1220 | |
1221 | pcp = &pset->pcp; | |
2ff754fa DR |
1222 | if (pcp->count) { |
1223 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1224 | pcp->count = 0; | |
1225 | } | |
3dfa5721 | 1226 | local_irq_restore(flags); |
1da177e4 LT |
1227 | } |
1228 | } | |
1da177e4 | 1229 | |
9f8f2172 CL |
1230 | /* |
1231 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1232 | */ | |
1233 | void drain_local_pages(void *arg) | |
1234 | { | |
1235 | drain_pages(smp_processor_id()); | |
1236 | } | |
1237 | ||
1238 | /* | |
74046494 GBY |
1239 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
1240 | * | |
1241 | * Note that this code is protected against sending an IPI to an offline | |
1242 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: | |
1243 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but | |
1244 | * nothing keeps CPUs from showing up after we populated the cpumask and | |
1245 | * before the call to on_each_cpu_mask(). | |
9f8f2172 CL |
1246 | */ |
1247 | void drain_all_pages(void) | |
1248 | { | |
74046494 GBY |
1249 | int cpu; |
1250 | struct per_cpu_pageset *pcp; | |
1251 | struct zone *zone; | |
1252 | ||
1253 | /* | |
1254 | * Allocate in the BSS so we wont require allocation in | |
1255 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y | |
1256 | */ | |
1257 | static cpumask_t cpus_with_pcps; | |
1258 | ||
1259 | /* | |
1260 | * We don't care about racing with CPU hotplug event | |
1261 | * as offline notification will cause the notified | |
1262 | * cpu to drain that CPU pcps and on_each_cpu_mask | |
1263 | * disables preemption as part of its processing | |
1264 | */ | |
1265 | for_each_online_cpu(cpu) { | |
1266 | bool has_pcps = false; | |
1267 | for_each_populated_zone(zone) { | |
1268 | pcp = per_cpu_ptr(zone->pageset, cpu); | |
1269 | if (pcp->pcp.count) { | |
1270 | has_pcps = true; | |
1271 | break; | |
1272 | } | |
1273 | } | |
1274 | if (has_pcps) | |
1275 | cpumask_set_cpu(cpu, &cpus_with_pcps); | |
1276 | else | |
1277 | cpumask_clear_cpu(cpu, &cpus_with_pcps); | |
1278 | } | |
1279 | on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1); | |
9f8f2172 CL |
1280 | } |
1281 | ||
296699de | 1282 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
1283 | |
1284 | void mark_free_pages(struct zone *zone) | |
1285 | { | |
f623f0db RW |
1286 | unsigned long pfn, max_zone_pfn; |
1287 | unsigned long flags; | |
b2a0ac88 | 1288 | int order, t; |
1da177e4 LT |
1289 | struct list_head *curr; |
1290 | ||
1291 | if (!zone->spanned_pages) | |
1292 | return; | |
1293 | ||
1294 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db | 1295 | |
108bcc96 | 1296 | max_zone_pfn = zone_end_pfn(zone); |
f623f0db RW |
1297 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
1298 | if (pfn_valid(pfn)) { | |
1299 | struct page *page = pfn_to_page(pfn); | |
1300 | ||
7be98234 RW |
1301 | if (!swsusp_page_is_forbidden(page)) |
1302 | swsusp_unset_page_free(page); | |
f623f0db | 1303 | } |
1da177e4 | 1304 | |
b2a0ac88 MG |
1305 | for_each_migratetype_order(order, t) { |
1306 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 1307 | unsigned long i; |
1da177e4 | 1308 | |
f623f0db RW |
1309 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
1310 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 1311 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 1312 | } |
b2a0ac88 | 1313 | } |
1da177e4 LT |
1314 | spin_unlock_irqrestore(&zone->lock, flags); |
1315 | } | |
e2c55dc8 | 1316 | #endif /* CONFIG_PM */ |
1da177e4 | 1317 | |
1da177e4 LT |
1318 | /* |
1319 | * Free a 0-order page | |
fc91668e | 1320 | * cold == 1 ? free a cold page : free a hot page |
1da177e4 | 1321 | */ |
fc91668e | 1322 | void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
1323 | { |
1324 | struct zone *zone = page_zone(page); | |
1325 | struct per_cpu_pages *pcp; | |
1326 | unsigned long flags; | |
5f8dcc21 | 1327 | int migratetype; |
1da177e4 | 1328 | |
ec95f53a | 1329 | if (!free_pages_prepare(page, 0)) |
689bcebf HD |
1330 | return; |
1331 | ||
5f8dcc21 | 1332 | migratetype = get_pageblock_migratetype(page); |
b12c4ad1 | 1333 | set_freepage_migratetype(page, migratetype); |
1da177e4 | 1334 | local_irq_save(flags); |
f8891e5e | 1335 | __count_vm_event(PGFREE); |
da456f14 | 1336 | |
5f8dcc21 MG |
1337 | /* |
1338 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1339 | * Free ISOLATE pages back to the allocator because they are being | |
1340 | * offlined but treat RESERVE as movable pages so we can get those | |
1341 | * areas back if necessary. Otherwise, we may have to free | |
1342 | * excessively into the page allocator | |
1343 | */ | |
1344 | if (migratetype >= MIGRATE_PCPTYPES) { | |
194159fb | 1345 | if (unlikely(is_migrate_isolate(migratetype))) { |
5f8dcc21 MG |
1346 | free_one_page(zone, page, 0, migratetype); |
1347 | goto out; | |
1348 | } | |
1349 | migratetype = MIGRATE_MOVABLE; | |
1350 | } | |
1351 | ||
99dcc3e5 | 1352 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
3dfa5721 | 1353 | if (cold) |
5f8dcc21 | 1354 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
3dfa5721 | 1355 | else |
5f8dcc21 | 1356 | list_add(&page->lru, &pcp->lists[migratetype]); |
1da177e4 | 1357 | pcp->count++; |
48db57f8 | 1358 | if (pcp->count >= pcp->high) { |
5f8dcc21 | 1359 | free_pcppages_bulk(zone, pcp->batch, pcp); |
48db57f8 NP |
1360 | pcp->count -= pcp->batch; |
1361 | } | |
5f8dcc21 MG |
1362 | |
1363 | out: | |
1da177e4 | 1364 | local_irq_restore(flags); |
1da177e4 LT |
1365 | } |
1366 | ||
cc59850e KK |
1367 | /* |
1368 | * Free a list of 0-order pages | |
1369 | */ | |
1370 | void free_hot_cold_page_list(struct list_head *list, int cold) | |
1371 | { | |
1372 | struct page *page, *next; | |
1373 | ||
1374 | list_for_each_entry_safe(page, next, list, lru) { | |
b413d48a | 1375 | trace_mm_page_free_batched(page, cold); |
cc59850e KK |
1376 | free_hot_cold_page(page, cold); |
1377 | } | |
1378 | } | |
1379 | ||
8dfcc9ba NP |
1380 | /* |
1381 | * split_page takes a non-compound higher-order page, and splits it into | |
1382 | * n (1<<order) sub-pages: page[0..n] | |
1383 | * Each sub-page must be freed individually. | |
1384 | * | |
1385 | * Note: this is probably too low level an operation for use in drivers. | |
1386 | * Please consult with lkml before using this in your driver. | |
1387 | */ | |
1388 | void split_page(struct page *page, unsigned int order) | |
1389 | { | |
1390 | int i; | |
1391 | ||
725d704e NP |
1392 | VM_BUG_ON(PageCompound(page)); |
1393 | VM_BUG_ON(!page_count(page)); | |
b1eeab67 VN |
1394 | |
1395 | #ifdef CONFIG_KMEMCHECK | |
1396 | /* | |
1397 | * Split shadow pages too, because free(page[0]) would | |
1398 | * otherwise free the whole shadow. | |
1399 | */ | |
1400 | if (kmemcheck_page_is_tracked(page)) | |
1401 | split_page(virt_to_page(page[0].shadow), order); | |
1402 | #endif | |
1403 | ||
7835e98b NP |
1404 | for (i = 1; i < (1 << order); i++) |
1405 | set_page_refcounted(page + i); | |
8dfcc9ba | 1406 | } |
8dfcc9ba | 1407 | |
8fb74b9f | 1408 | static int __isolate_free_page(struct page *page, unsigned int order) |
748446bb | 1409 | { |
748446bb MG |
1410 | unsigned long watermark; |
1411 | struct zone *zone; | |
2139cbe6 | 1412 | int mt; |
748446bb MG |
1413 | |
1414 | BUG_ON(!PageBuddy(page)); | |
1415 | ||
1416 | zone = page_zone(page); | |
2e30abd1 | 1417 | mt = get_pageblock_migratetype(page); |
748446bb | 1418 | |
194159fb | 1419 | if (!is_migrate_isolate(mt)) { |
2e30abd1 MS |
1420 | /* Obey watermarks as if the page was being allocated */ |
1421 | watermark = low_wmark_pages(zone) + (1 << order); | |
1422 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1423 | return 0; | |
1424 | ||
8fb74b9f | 1425 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2e30abd1 | 1426 | } |
748446bb MG |
1427 | |
1428 | /* Remove page from free list */ | |
1429 | list_del(&page->lru); | |
1430 | zone->free_area[order].nr_free--; | |
1431 | rmv_page_order(page); | |
2139cbe6 | 1432 | |
8fb74b9f | 1433 | /* Set the pageblock if the isolated page is at least a pageblock */ |
748446bb MG |
1434 | if (order >= pageblock_order - 1) { |
1435 | struct page *endpage = page + (1 << order) - 1; | |
47118af0 MN |
1436 | for (; page < endpage; page += pageblock_nr_pages) { |
1437 | int mt = get_pageblock_migratetype(page); | |
194159fb | 1438 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
47118af0 MN |
1439 | set_pageblock_migratetype(page, |
1440 | MIGRATE_MOVABLE); | |
1441 | } | |
748446bb MG |
1442 | } |
1443 | ||
8fb74b9f | 1444 | return 1UL << order; |
1fb3f8ca MG |
1445 | } |
1446 | ||
1447 | /* | |
1448 | * Similar to split_page except the page is already free. As this is only | |
1449 | * being used for migration, the migratetype of the block also changes. | |
1450 | * As this is called with interrupts disabled, the caller is responsible | |
1451 | * for calling arch_alloc_page() and kernel_map_page() after interrupts | |
1452 | * are enabled. | |
1453 | * | |
1454 | * Note: this is probably too low level an operation for use in drivers. | |
1455 | * Please consult with lkml before using this in your driver. | |
1456 | */ | |
1457 | int split_free_page(struct page *page) | |
1458 | { | |
1459 | unsigned int order; | |
1460 | int nr_pages; | |
1461 | ||
1fb3f8ca MG |
1462 | order = page_order(page); |
1463 | ||
8fb74b9f | 1464 | nr_pages = __isolate_free_page(page, order); |
1fb3f8ca MG |
1465 | if (!nr_pages) |
1466 | return 0; | |
1467 | ||
1468 | /* Split into individual pages */ | |
1469 | set_page_refcounted(page); | |
1470 | split_page(page, order); | |
1471 | return nr_pages; | |
748446bb MG |
1472 | } |
1473 | ||
1da177e4 LT |
1474 | /* |
1475 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1476 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1477 | * or two. | |
1478 | */ | |
0a15c3e9 MG |
1479 | static inline |
1480 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
3dd28266 MG |
1481 | struct zone *zone, int order, gfp_t gfp_flags, |
1482 | int migratetype) | |
1da177e4 LT |
1483 | { |
1484 | unsigned long flags; | |
689bcebf | 1485 | struct page *page; |
1da177e4 LT |
1486 | int cold = !!(gfp_flags & __GFP_COLD); |
1487 | ||
689bcebf | 1488 | again: |
48db57f8 | 1489 | if (likely(order == 0)) { |
1da177e4 | 1490 | struct per_cpu_pages *pcp; |
5f8dcc21 | 1491 | struct list_head *list; |
1da177e4 | 1492 | |
1da177e4 | 1493 | local_irq_save(flags); |
99dcc3e5 CL |
1494 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
1495 | list = &pcp->lists[migratetype]; | |
5f8dcc21 | 1496 | if (list_empty(list)) { |
535131e6 | 1497 | pcp->count += rmqueue_bulk(zone, 0, |
5f8dcc21 | 1498 | pcp->batch, list, |
e084b2d9 | 1499 | migratetype, cold); |
5f8dcc21 | 1500 | if (unlikely(list_empty(list))) |
6fb332fa | 1501 | goto failed; |
535131e6 | 1502 | } |
b92a6edd | 1503 | |
5f8dcc21 MG |
1504 | if (cold) |
1505 | page = list_entry(list->prev, struct page, lru); | |
1506 | else | |
1507 | page = list_entry(list->next, struct page, lru); | |
1508 | ||
b92a6edd MG |
1509 | list_del(&page->lru); |
1510 | pcp->count--; | |
7fb1d9fc | 1511 | } else { |
dab48dab AM |
1512 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
1513 | /* | |
1514 | * __GFP_NOFAIL is not to be used in new code. | |
1515 | * | |
1516 | * All __GFP_NOFAIL callers should be fixed so that they | |
1517 | * properly detect and handle allocation failures. | |
1518 | * | |
1519 | * We most definitely don't want callers attempting to | |
4923abf9 | 1520 | * allocate greater than order-1 page units with |
dab48dab AM |
1521 | * __GFP_NOFAIL. |
1522 | */ | |
4923abf9 | 1523 | WARN_ON_ONCE(order > 1); |
dab48dab | 1524 | } |
1da177e4 | 1525 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1526 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1527 | spin_unlock(&zone->lock); |
1528 | if (!page) | |
1529 | goto failed; | |
d1ce749a BZ |
1530 | __mod_zone_freepage_state(zone, -(1 << order), |
1531 | get_pageblock_migratetype(page)); | |
1da177e4 LT |
1532 | } |
1533 | ||
f8891e5e | 1534 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
78afd561 | 1535 | zone_statistics(preferred_zone, zone, gfp_flags); |
a74609fa | 1536 | local_irq_restore(flags); |
1da177e4 | 1537 | |
725d704e | 1538 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1539 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1540 | goto again; |
1da177e4 | 1541 | return page; |
a74609fa NP |
1542 | |
1543 | failed: | |
1544 | local_irq_restore(flags); | |
a74609fa | 1545 | return NULL; |
1da177e4 LT |
1546 | } |
1547 | ||
933e312e AM |
1548 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1549 | ||
b2588c4b | 1550 | static struct { |
933e312e AM |
1551 | struct fault_attr attr; |
1552 | ||
1553 | u32 ignore_gfp_highmem; | |
1554 | u32 ignore_gfp_wait; | |
54114994 | 1555 | u32 min_order; |
933e312e AM |
1556 | } fail_page_alloc = { |
1557 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1558 | .ignore_gfp_wait = 1, |
1559 | .ignore_gfp_highmem = 1, | |
54114994 | 1560 | .min_order = 1, |
933e312e AM |
1561 | }; |
1562 | ||
1563 | static int __init setup_fail_page_alloc(char *str) | |
1564 | { | |
1565 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1566 | } | |
1567 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1568 | ||
deaf386e | 1569 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1570 | { |
54114994 | 1571 | if (order < fail_page_alloc.min_order) |
deaf386e | 1572 | return false; |
933e312e | 1573 | if (gfp_mask & __GFP_NOFAIL) |
deaf386e | 1574 | return false; |
933e312e | 1575 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
deaf386e | 1576 | return false; |
933e312e | 1577 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
deaf386e | 1578 | return false; |
933e312e AM |
1579 | |
1580 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1581 | } | |
1582 | ||
1583 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1584 | ||
1585 | static int __init fail_page_alloc_debugfs(void) | |
1586 | { | |
f4ae40a6 | 1587 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
933e312e | 1588 | struct dentry *dir; |
933e312e | 1589 | |
dd48c085 AM |
1590 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
1591 | &fail_page_alloc.attr); | |
1592 | if (IS_ERR(dir)) | |
1593 | return PTR_ERR(dir); | |
933e312e | 1594 | |
b2588c4b AM |
1595 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
1596 | &fail_page_alloc.ignore_gfp_wait)) | |
1597 | goto fail; | |
1598 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1599 | &fail_page_alloc.ignore_gfp_highmem)) | |
1600 | goto fail; | |
1601 | if (!debugfs_create_u32("min-order", mode, dir, | |
1602 | &fail_page_alloc.min_order)) | |
1603 | goto fail; | |
1604 | ||
1605 | return 0; | |
1606 | fail: | |
dd48c085 | 1607 | debugfs_remove_recursive(dir); |
933e312e | 1608 | |
b2588c4b | 1609 | return -ENOMEM; |
933e312e AM |
1610 | } |
1611 | ||
1612 | late_initcall(fail_page_alloc_debugfs); | |
1613 | ||
1614 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1615 | ||
1616 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1617 | ||
deaf386e | 1618 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
933e312e | 1619 | { |
deaf386e | 1620 | return false; |
933e312e AM |
1621 | } |
1622 | ||
1623 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1624 | ||
1da177e4 | 1625 | /* |
88f5acf8 | 1626 | * Return true if free pages are above 'mark'. This takes into account the order |
1da177e4 LT |
1627 | * of the allocation. |
1628 | */ | |
88f5acf8 MG |
1629 | static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
1630 | int classzone_idx, int alloc_flags, long free_pages) | |
1da177e4 LT |
1631 | { |
1632 | /* free_pages my go negative - that's OK */ | |
d23ad423 | 1633 | long min = mark; |
2cfed075 | 1634 | long lowmem_reserve = z->lowmem_reserve[classzone_idx]; |
1da177e4 LT |
1635 | int o; |
1636 | ||
df0a6daa | 1637 | free_pages -= (1 << order) - 1; |
7fb1d9fc | 1638 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1639 | min -= min / 2; |
7fb1d9fc | 1640 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 | 1641 | min -= min / 4; |
d95ea5d1 BZ |
1642 | #ifdef CONFIG_CMA |
1643 | /* If allocation can't use CMA areas don't use free CMA pages */ | |
1644 | if (!(alloc_flags & ALLOC_CMA)) | |
1645 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); | |
1646 | #endif | |
2cfed075 | 1647 | if (free_pages <= min + lowmem_reserve) |
88f5acf8 | 1648 | return false; |
1da177e4 LT |
1649 | for (o = 0; o < order; o++) { |
1650 | /* At the next order, this order's pages become unavailable */ | |
1651 | free_pages -= z->free_area[o].nr_free << o; | |
1652 | ||
1653 | /* Require fewer higher order pages to be free */ | |
1654 | min >>= 1; | |
1655 | ||
1656 | if (free_pages <= min) | |
88f5acf8 | 1657 | return false; |
1da177e4 | 1658 | } |
88f5acf8 MG |
1659 | return true; |
1660 | } | |
1661 | ||
1662 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
1663 | int classzone_idx, int alloc_flags) | |
1664 | { | |
1665 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1666 | zone_page_state(z, NR_FREE_PAGES)); | |
1667 | } | |
1668 | ||
1669 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, | |
1670 | int classzone_idx, int alloc_flags) | |
1671 | { | |
1672 | long free_pages = zone_page_state(z, NR_FREE_PAGES); | |
1673 | ||
1674 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) | |
1675 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); | |
1676 | ||
1677 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, | |
1678 | free_pages); | |
1da177e4 LT |
1679 | } |
1680 | ||
9276b1bc PJ |
1681 | #ifdef CONFIG_NUMA |
1682 | /* | |
1683 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1684 | * skip over zones that are not allowed by the cpuset, or that have | |
1685 | * been recently (in last second) found to be nearly full. See further | |
1686 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1687 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1688 | * |
1689 | * If the zonelist cache is present in the passed in zonelist, then | |
1690 | * returns a pointer to the allowed node mask (either the current | |
4b0ef1fe | 1691 | * tasks mems_allowed, or node_states[N_MEMORY].) |
9276b1bc PJ |
1692 | * |
1693 | * If the zonelist cache is not available for this zonelist, does | |
1694 | * nothing and returns NULL. | |
1695 | * | |
1696 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1697 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1698 | * | |
1699 | * We hold off even calling zlc_setup, until after we've checked the | |
1700 | * first zone in the zonelist, on the theory that most allocations will | |
1701 | * be satisfied from that first zone, so best to examine that zone as | |
1702 | * quickly as we can. | |
1703 | */ | |
1704 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1705 | { | |
1706 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1707 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1708 | ||
1709 | zlc = zonelist->zlcache_ptr; | |
1710 | if (!zlc) | |
1711 | return NULL; | |
1712 | ||
f05111f5 | 1713 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
9276b1bc PJ |
1714 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
1715 | zlc->last_full_zap = jiffies; | |
1716 | } | |
1717 | ||
1718 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1719 | &cpuset_current_mems_allowed : | |
4b0ef1fe | 1720 | &node_states[N_MEMORY]; |
9276b1bc PJ |
1721 | return allowednodes; |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1726 | * if it is worth looking at further for free memory: | |
1727 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1728 | * bit set in the zonelist_cache fullzones BITMAP). | |
1729 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1730 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1731 | * Return true (non-zero) if zone is worth looking at further, or | |
1732 | * else return false (zero) if it is not. | |
1733 | * | |
1734 | * This check -ignores- the distinction between various watermarks, | |
1735 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1736 | * found to be full for any variation of these watermarks, it will | |
1737 | * be considered full for up to one second by all requests, unless | |
1738 | * we are so low on memory on all allowed nodes that we are forced | |
1739 | * into the second scan of the zonelist. | |
1740 | * | |
1741 | * In the second scan we ignore this zonelist cache and exactly | |
1742 | * apply the watermarks to all zones, even it is slower to do so. | |
1743 | * We are low on memory in the second scan, and should leave no stone | |
1744 | * unturned looking for a free page. | |
1745 | */ | |
dd1a239f | 1746 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1747 | nodemask_t *allowednodes) |
1748 | { | |
1749 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1750 | int i; /* index of *z in zonelist zones */ | |
1751 | int n; /* node that zone *z is on */ | |
1752 | ||
1753 | zlc = zonelist->zlcache_ptr; | |
1754 | if (!zlc) | |
1755 | return 1; | |
1756 | ||
dd1a239f | 1757 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1758 | n = zlc->z_to_n[i]; |
1759 | ||
1760 | /* This zone is worth trying if it is allowed but not full */ | |
1761 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1762 | } | |
1763 | ||
1764 | /* | |
1765 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1766 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1767 | * from that zone don't waste time re-examining it. | |
1768 | */ | |
dd1a239f | 1769 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1770 | { |
1771 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1772 | int i; /* index of *z in zonelist zones */ | |
1773 | ||
1774 | zlc = zonelist->zlcache_ptr; | |
1775 | if (!zlc) | |
1776 | return; | |
1777 | ||
dd1a239f | 1778 | i = z - zonelist->_zonerefs; |
9276b1bc PJ |
1779 | |
1780 | set_bit(i, zlc->fullzones); | |
1781 | } | |
1782 | ||
76d3fbf8 MG |
1783 | /* |
1784 | * clear all zones full, called after direct reclaim makes progress so that | |
1785 | * a zone that was recently full is not skipped over for up to a second | |
1786 | */ | |
1787 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1788 | { | |
1789 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1790 | ||
1791 | zlc = zonelist->zlcache_ptr; | |
1792 | if (!zlc) | |
1793 | return; | |
1794 | ||
1795 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1796 | } | |
1797 | ||
957f822a DR |
1798 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
1799 | { | |
1800 | return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes); | |
1801 | } | |
1802 | ||
1803 | static void __paginginit init_zone_allows_reclaim(int nid) | |
1804 | { | |
1805 | int i; | |
1806 | ||
1807 | for_each_online_node(i) | |
6b187d02 | 1808 | if (node_distance(nid, i) <= RECLAIM_DISTANCE) |
957f822a | 1809 | node_set(i, NODE_DATA(nid)->reclaim_nodes); |
6b187d02 | 1810 | else |
957f822a | 1811 | zone_reclaim_mode = 1; |
957f822a DR |
1812 | } |
1813 | ||
9276b1bc PJ |
1814 | #else /* CONFIG_NUMA */ |
1815 | ||
1816 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1817 | { | |
1818 | return NULL; | |
1819 | } | |
1820 | ||
dd1a239f | 1821 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
9276b1bc PJ |
1822 | nodemask_t *allowednodes) |
1823 | { | |
1824 | return 1; | |
1825 | } | |
1826 | ||
dd1a239f | 1827 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
9276b1bc PJ |
1828 | { |
1829 | } | |
76d3fbf8 MG |
1830 | |
1831 | static void zlc_clear_zones_full(struct zonelist *zonelist) | |
1832 | { | |
1833 | } | |
957f822a DR |
1834 | |
1835 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) | |
1836 | { | |
1837 | return true; | |
1838 | } | |
1839 | ||
1840 | static inline void init_zone_allows_reclaim(int nid) | |
1841 | { | |
1842 | } | |
9276b1bc PJ |
1843 | #endif /* CONFIG_NUMA */ |
1844 | ||
7fb1d9fc | 1845 | /* |
0798e519 | 1846 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1847 | * a page. |
1848 | */ | |
1849 | static struct page * | |
19770b32 | 1850 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
5117f45d | 1851 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
3dd28266 | 1852 | struct zone *preferred_zone, int migratetype) |
753ee728 | 1853 | { |
dd1a239f | 1854 | struct zoneref *z; |
7fb1d9fc | 1855 | struct page *page = NULL; |
54a6eb5c | 1856 | int classzone_idx; |
5117f45d | 1857 | struct zone *zone; |
9276b1bc PJ |
1858 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1859 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1860 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
54a6eb5c | 1861 | |
19770b32 | 1862 | classzone_idx = zone_idx(preferred_zone); |
9276b1bc | 1863 | zonelist_scan: |
7fb1d9fc | 1864 | /* |
9276b1bc | 1865 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1866 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1867 | */ | |
19770b32 MG |
1868 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
1869 | high_zoneidx, nodemask) { | |
e5adfffc | 1870 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
9276b1bc PJ |
1871 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1872 | continue; | |
7fb1d9fc | 1873 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1874 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
cd38b115 | 1875 | continue; |
a756cf59 JW |
1876 | /* |
1877 | * When allocating a page cache page for writing, we | |
1878 | * want to get it from a zone that is within its dirty | |
1879 | * limit, such that no single zone holds more than its | |
1880 | * proportional share of globally allowed dirty pages. | |
1881 | * The dirty limits take into account the zone's | |
1882 | * lowmem reserves and high watermark so that kswapd | |
1883 | * should be able to balance it without having to | |
1884 | * write pages from its LRU list. | |
1885 | * | |
1886 | * This may look like it could increase pressure on | |
1887 | * lower zones by failing allocations in higher zones | |
1888 | * before they are full. But the pages that do spill | |
1889 | * over are limited as the lower zones are protected | |
1890 | * by this very same mechanism. It should not become | |
1891 | * a practical burden to them. | |
1892 | * | |
1893 | * XXX: For now, allow allocations to potentially | |
1894 | * exceed the per-zone dirty limit in the slowpath | |
1895 | * (ALLOC_WMARK_LOW unset) before going into reclaim, | |
1896 | * which is important when on a NUMA setup the allowed | |
1897 | * zones are together not big enough to reach the | |
1898 | * global limit. The proper fix for these situations | |
1899 | * will require awareness of zones in the | |
1900 | * dirty-throttling and the flusher threads. | |
1901 | */ | |
1902 | if ((alloc_flags & ALLOC_WMARK_LOW) && | |
1903 | (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone)) | |
1904 | goto this_zone_full; | |
7fb1d9fc | 1905 | |
41858966 | 1906 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
7fb1d9fc | 1907 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
3148890b | 1908 | unsigned long mark; |
fa5e084e MG |
1909 | int ret; |
1910 | ||
41858966 | 1911 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
fa5e084e MG |
1912 | if (zone_watermark_ok(zone, order, mark, |
1913 | classzone_idx, alloc_flags)) | |
1914 | goto try_this_zone; | |
1915 | ||
e5adfffc KS |
1916 | if (IS_ENABLED(CONFIG_NUMA) && |
1917 | !did_zlc_setup && nr_online_nodes > 1) { | |
cd38b115 MG |
1918 | /* |
1919 | * we do zlc_setup if there are multiple nodes | |
1920 | * and before considering the first zone allowed | |
1921 | * by the cpuset. | |
1922 | */ | |
1923 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1924 | zlc_active = 1; | |
1925 | did_zlc_setup = 1; | |
1926 | } | |
1927 | ||
957f822a DR |
1928 | if (zone_reclaim_mode == 0 || |
1929 | !zone_allows_reclaim(preferred_zone, zone)) | |
fa5e084e MG |
1930 | goto this_zone_full; |
1931 | ||
cd38b115 MG |
1932 | /* |
1933 | * As we may have just activated ZLC, check if the first | |
1934 | * eligible zone has failed zone_reclaim recently. | |
1935 | */ | |
e5adfffc | 1936 | if (IS_ENABLED(CONFIG_NUMA) && zlc_active && |
cd38b115 MG |
1937 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
1938 | continue; | |
1939 | ||
fa5e084e MG |
1940 | ret = zone_reclaim(zone, gfp_mask, order); |
1941 | switch (ret) { | |
1942 | case ZONE_RECLAIM_NOSCAN: | |
1943 | /* did not scan */ | |
cd38b115 | 1944 | continue; |
fa5e084e MG |
1945 | case ZONE_RECLAIM_FULL: |
1946 | /* scanned but unreclaimable */ | |
cd38b115 | 1947 | continue; |
fa5e084e MG |
1948 | default: |
1949 | /* did we reclaim enough */ | |
1950 | if (!zone_watermark_ok(zone, order, mark, | |
1951 | classzone_idx, alloc_flags)) | |
9276b1bc | 1952 | goto this_zone_full; |
0798e519 | 1953 | } |
7fb1d9fc RS |
1954 | } |
1955 | ||
fa5e084e | 1956 | try_this_zone: |
3dd28266 MG |
1957 | page = buffered_rmqueue(preferred_zone, zone, order, |
1958 | gfp_mask, migratetype); | |
0798e519 | 1959 | if (page) |
7fb1d9fc | 1960 | break; |
9276b1bc | 1961 | this_zone_full: |
e5adfffc | 1962 | if (IS_ENABLED(CONFIG_NUMA)) |
9276b1bc | 1963 | zlc_mark_zone_full(zonelist, z); |
54a6eb5c | 1964 | } |
9276b1bc | 1965 | |
e5adfffc | 1966 | if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) { |
9276b1bc PJ |
1967 | /* Disable zlc cache for second zonelist scan */ |
1968 | zlc_active = 0; | |
1969 | goto zonelist_scan; | |
1970 | } | |
b121186a AS |
1971 | |
1972 | if (page) | |
1973 | /* | |
1974 | * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was | |
1975 | * necessary to allocate the page. The expectation is | |
1976 | * that the caller is taking steps that will free more | |
1977 | * memory. The caller should avoid the page being used | |
1978 | * for !PFMEMALLOC purposes. | |
1979 | */ | |
1980 | page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS); | |
1981 | ||
7fb1d9fc | 1982 | return page; |
753ee728 MH |
1983 | } |
1984 | ||
29423e77 DR |
1985 | /* |
1986 | * Large machines with many possible nodes should not always dump per-node | |
1987 | * meminfo in irq context. | |
1988 | */ | |
1989 | static inline bool should_suppress_show_mem(void) | |
1990 | { | |
1991 | bool ret = false; | |
1992 | ||
1993 | #if NODES_SHIFT > 8 | |
1994 | ret = in_interrupt(); | |
1995 | #endif | |
1996 | return ret; | |
1997 | } | |
1998 | ||
a238ab5b DH |
1999 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
2000 | DEFAULT_RATELIMIT_INTERVAL, | |
2001 | DEFAULT_RATELIMIT_BURST); | |
2002 | ||
2003 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) | |
2004 | { | |
a238ab5b DH |
2005 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
2006 | ||
c0a32fc5 SG |
2007 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
2008 | debug_guardpage_minorder() > 0) | |
a238ab5b DH |
2009 | return; |
2010 | ||
2011 | /* | |
2012 | * This documents exceptions given to allocations in certain | |
2013 | * contexts that are allowed to allocate outside current's set | |
2014 | * of allowed nodes. | |
2015 | */ | |
2016 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2017 | if (test_thread_flag(TIF_MEMDIE) || | |
2018 | (current->flags & (PF_MEMALLOC | PF_EXITING))) | |
2019 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2020 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) | |
2021 | filter &= ~SHOW_MEM_FILTER_NODES; | |
2022 | ||
2023 | if (fmt) { | |
3ee9a4f0 JP |
2024 | struct va_format vaf; |
2025 | va_list args; | |
2026 | ||
a238ab5b | 2027 | va_start(args, fmt); |
3ee9a4f0 JP |
2028 | |
2029 | vaf.fmt = fmt; | |
2030 | vaf.va = &args; | |
2031 | ||
2032 | pr_warn("%pV", &vaf); | |
2033 | ||
a238ab5b DH |
2034 | va_end(args); |
2035 | } | |
2036 | ||
3ee9a4f0 JP |
2037 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
2038 | current->comm, order, gfp_mask); | |
a238ab5b DH |
2039 | |
2040 | dump_stack(); | |
2041 | if (!should_suppress_show_mem()) | |
2042 | show_mem(filter); | |
2043 | } | |
2044 | ||
11e33f6a MG |
2045 | static inline int |
2046 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
f90ac398 | 2047 | unsigned long did_some_progress, |
11e33f6a | 2048 | unsigned long pages_reclaimed) |
1da177e4 | 2049 | { |
11e33f6a MG |
2050 | /* Do not loop if specifically requested */ |
2051 | if (gfp_mask & __GFP_NORETRY) | |
2052 | return 0; | |
1da177e4 | 2053 | |
f90ac398 MG |
2054 | /* Always retry if specifically requested */ |
2055 | if (gfp_mask & __GFP_NOFAIL) | |
2056 | return 1; | |
2057 | ||
2058 | /* | |
2059 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim | |
2060 | * making forward progress without invoking OOM. Suspend also disables | |
2061 | * storage devices so kswapd will not help. Bail if we are suspending. | |
2062 | */ | |
2063 | if (!did_some_progress && pm_suspended_storage()) | |
2064 | return 0; | |
2065 | ||
11e33f6a MG |
2066 | /* |
2067 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
2068 | * means __GFP_NOFAIL, but that may not be true in other | |
2069 | * implementations. | |
2070 | */ | |
2071 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
2072 | return 1; | |
2073 | ||
2074 | /* | |
2075 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
2076 | * specified, then we retry until we no longer reclaim any pages | |
2077 | * (above), or we've reclaimed an order of pages at least as | |
2078 | * large as the allocation's order. In both cases, if the | |
2079 | * allocation still fails, we stop retrying. | |
2080 | */ | |
2081 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
2082 | return 1; | |
cf40bd16 | 2083 | |
11e33f6a MG |
2084 | return 0; |
2085 | } | |
933e312e | 2086 | |
11e33f6a MG |
2087 | static inline struct page * |
2088 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
2089 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2090 | nodemask_t *nodemask, struct zone *preferred_zone, |
2091 | int migratetype) | |
11e33f6a MG |
2092 | { |
2093 | struct page *page; | |
2094 | ||
2095 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
ff321fea | 2096 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
11e33f6a | 2097 | schedule_timeout_uninterruptible(1); |
1da177e4 LT |
2098 | return NULL; |
2099 | } | |
6b1de916 | 2100 | |
11e33f6a MG |
2101 | /* |
2102 | * Go through the zonelist yet one more time, keep very high watermark | |
2103 | * here, this is only to catch a parallel oom killing, we must fail if | |
2104 | * we're still under heavy pressure. | |
2105 | */ | |
2106 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
2107 | order, zonelist, high_zoneidx, | |
5117f45d | 2108 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
3dd28266 | 2109 | preferred_zone, migratetype); |
7fb1d9fc | 2110 | if (page) |
11e33f6a MG |
2111 | goto out; |
2112 | ||
4365a567 KH |
2113 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2114 | /* The OOM killer will not help higher order allocs */ | |
2115 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2116 | goto out; | |
03668b3c DR |
2117 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
2118 | if (high_zoneidx < ZONE_NORMAL) | |
2119 | goto out; | |
4365a567 KH |
2120 | /* |
2121 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
2122 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
2123 | * The caller should handle page allocation failure by itself if | |
2124 | * it specifies __GFP_THISNODE. | |
2125 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
2126 | */ | |
2127 | if (gfp_mask & __GFP_THISNODE) | |
2128 | goto out; | |
2129 | } | |
11e33f6a | 2130 | /* Exhausted what can be done so it's blamo time */ |
08ab9b10 | 2131 | out_of_memory(zonelist, gfp_mask, order, nodemask, false); |
11e33f6a MG |
2132 | |
2133 | out: | |
2134 | clear_zonelist_oom(zonelist, gfp_mask); | |
2135 | return page; | |
2136 | } | |
2137 | ||
56de7263 MG |
2138 | #ifdef CONFIG_COMPACTION |
2139 | /* Try memory compaction for high-order allocations before reclaim */ | |
2140 | static struct page * | |
2141 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2142 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2143 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
66199712 | 2144 | int migratetype, bool sync_migration, |
c67fe375 | 2145 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2146 | unsigned long *did_some_progress) |
56de7263 | 2147 | { |
66199712 | 2148 | if (!order) |
56de7263 MG |
2149 | return NULL; |
2150 | ||
aff62249 | 2151 | if (compaction_deferred(preferred_zone, order)) { |
66199712 MG |
2152 | *deferred_compaction = true; |
2153 | return NULL; | |
2154 | } | |
2155 | ||
c06b1fca | 2156 | current->flags |= PF_MEMALLOC; |
56de7263 | 2157 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
c67fe375 | 2158 | nodemask, sync_migration, |
8fb74b9f | 2159 | contended_compaction); |
c06b1fca | 2160 | current->flags &= ~PF_MEMALLOC; |
56de7263 | 2161 | |
1fb3f8ca | 2162 | if (*did_some_progress != COMPACT_SKIPPED) { |
8fb74b9f MG |
2163 | struct page *page; |
2164 | ||
56de7263 MG |
2165 | /* Page migration frees to the PCP lists but we want merging */ |
2166 | drain_pages(get_cpu()); | |
2167 | put_cpu(); | |
2168 | ||
2169 | page = get_page_from_freelist(gfp_mask, nodemask, | |
2170 | order, zonelist, high_zoneidx, | |
cfd19c5a MG |
2171 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2172 | preferred_zone, migratetype); | |
56de7263 | 2173 | if (page) { |
62997027 | 2174 | preferred_zone->compact_blockskip_flush = false; |
4f92e258 MG |
2175 | preferred_zone->compact_considered = 0; |
2176 | preferred_zone->compact_defer_shift = 0; | |
aff62249 RR |
2177 | if (order >= preferred_zone->compact_order_failed) |
2178 | preferred_zone->compact_order_failed = order + 1; | |
56de7263 MG |
2179 | count_vm_event(COMPACTSUCCESS); |
2180 | return page; | |
2181 | } | |
2182 | ||
2183 | /* | |
2184 | * It's bad if compaction run occurs and fails. | |
2185 | * The most likely reason is that pages exist, | |
2186 | * but not enough to satisfy watermarks. | |
2187 | */ | |
2188 | count_vm_event(COMPACTFAIL); | |
66199712 MG |
2189 | |
2190 | /* | |
2191 | * As async compaction considers a subset of pageblocks, only | |
2192 | * defer if the failure was a sync compaction failure. | |
2193 | */ | |
2194 | if (sync_migration) | |
aff62249 | 2195 | defer_compaction(preferred_zone, order); |
56de7263 MG |
2196 | |
2197 | cond_resched(); | |
2198 | } | |
2199 | ||
2200 | return NULL; | |
2201 | } | |
2202 | #else | |
2203 | static inline struct page * | |
2204 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, | |
2205 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2206 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
66199712 | 2207 | int migratetype, bool sync_migration, |
c67fe375 | 2208 | bool *contended_compaction, bool *deferred_compaction, |
66199712 | 2209 | unsigned long *did_some_progress) |
56de7263 MG |
2210 | { |
2211 | return NULL; | |
2212 | } | |
2213 | #endif /* CONFIG_COMPACTION */ | |
2214 | ||
bba90710 MS |
2215 | /* Perform direct synchronous page reclaim */ |
2216 | static int | |
2217 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist, | |
2218 | nodemask_t *nodemask) | |
11e33f6a | 2219 | { |
11e33f6a | 2220 | struct reclaim_state reclaim_state; |
bba90710 | 2221 | int progress; |
11e33f6a MG |
2222 | |
2223 | cond_resched(); | |
2224 | ||
2225 | /* We now go into synchronous reclaim */ | |
2226 | cpuset_memory_pressure_bump(); | |
c06b1fca | 2227 | current->flags |= PF_MEMALLOC; |
11e33f6a MG |
2228 | lockdep_set_current_reclaim_state(gfp_mask); |
2229 | reclaim_state.reclaimed_slab = 0; | |
c06b1fca | 2230 | current->reclaim_state = &reclaim_state; |
11e33f6a | 2231 | |
bba90710 | 2232 | progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
11e33f6a | 2233 | |
c06b1fca | 2234 | current->reclaim_state = NULL; |
11e33f6a | 2235 | lockdep_clear_current_reclaim_state(); |
c06b1fca | 2236 | current->flags &= ~PF_MEMALLOC; |
11e33f6a MG |
2237 | |
2238 | cond_resched(); | |
2239 | ||
bba90710 MS |
2240 | return progress; |
2241 | } | |
2242 | ||
2243 | /* The really slow allocator path where we enter direct reclaim */ | |
2244 | static inline struct page * | |
2245 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
2246 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
2247 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
2248 | int migratetype, unsigned long *did_some_progress) | |
2249 | { | |
2250 | struct page *page = NULL; | |
2251 | bool drained = false; | |
2252 | ||
2253 | *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist, | |
2254 | nodemask); | |
9ee493ce MG |
2255 | if (unlikely(!(*did_some_progress))) |
2256 | return NULL; | |
11e33f6a | 2257 | |
76d3fbf8 | 2258 | /* After successful reclaim, reconsider all zones for allocation */ |
e5adfffc | 2259 | if (IS_ENABLED(CONFIG_NUMA)) |
76d3fbf8 MG |
2260 | zlc_clear_zones_full(zonelist); |
2261 | ||
9ee493ce MG |
2262 | retry: |
2263 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2264 | zonelist, high_zoneidx, |
cfd19c5a MG |
2265 | alloc_flags & ~ALLOC_NO_WATERMARKS, |
2266 | preferred_zone, migratetype); | |
9ee493ce MG |
2267 | |
2268 | /* | |
2269 | * If an allocation failed after direct reclaim, it could be because | |
2270 | * pages are pinned on the per-cpu lists. Drain them and try again | |
2271 | */ | |
2272 | if (!page && !drained) { | |
2273 | drain_all_pages(); | |
2274 | drained = true; | |
2275 | goto retry; | |
2276 | } | |
2277 | ||
11e33f6a MG |
2278 | return page; |
2279 | } | |
2280 | ||
1da177e4 | 2281 | /* |
11e33f6a MG |
2282 | * This is called in the allocator slow-path if the allocation request is of |
2283 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1da177e4 | 2284 | */ |
11e33f6a MG |
2285 | static inline struct page * |
2286 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
2287 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2288 | nodemask_t *nodemask, struct zone *preferred_zone, |
2289 | int migratetype) | |
11e33f6a MG |
2290 | { |
2291 | struct page *page; | |
2292 | ||
2293 | do { | |
2294 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
5117f45d | 2295 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
3dd28266 | 2296 | preferred_zone, migratetype); |
11e33f6a MG |
2297 | |
2298 | if (!page && gfp_mask & __GFP_NOFAIL) | |
0e093d99 | 2299 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
11e33f6a MG |
2300 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
2301 | ||
2302 | return page; | |
2303 | } | |
2304 | ||
2305 | static inline | |
2306 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
99504748 MG |
2307 | enum zone_type high_zoneidx, |
2308 | enum zone_type classzone_idx) | |
1da177e4 | 2309 | { |
dd1a239f MG |
2310 | struct zoneref *z; |
2311 | struct zone *zone; | |
1da177e4 | 2312 | |
11e33f6a | 2313 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
99504748 | 2314 | wakeup_kswapd(zone, order, classzone_idx); |
11e33f6a | 2315 | } |
cf40bd16 | 2316 | |
341ce06f PZ |
2317 | static inline int |
2318 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
2319 | { | |
341ce06f PZ |
2320 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
2321 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1da177e4 | 2322 | |
a56f57ff | 2323 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
e6223a3b | 2324 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
933e312e | 2325 | |
341ce06f PZ |
2326 | /* |
2327 | * The caller may dip into page reserves a bit more if the caller | |
2328 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
2329 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
2330 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
2331 | */ | |
e6223a3b | 2332 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
1da177e4 | 2333 | |
341ce06f | 2334 | if (!wait) { |
5c3240d9 AA |
2335 | /* |
2336 | * Not worth trying to allocate harder for | |
2337 | * __GFP_NOMEMALLOC even if it can't schedule. | |
2338 | */ | |
2339 | if (!(gfp_mask & __GFP_NOMEMALLOC)) | |
2340 | alloc_flags |= ALLOC_HARDER; | |
523b9458 | 2341 | /* |
341ce06f PZ |
2342 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
2343 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
523b9458 | 2344 | */ |
341ce06f | 2345 | alloc_flags &= ~ALLOC_CPUSET; |
c06b1fca | 2346 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
341ce06f PZ |
2347 | alloc_flags |= ALLOC_HARDER; |
2348 | ||
b37f1dd0 MG |
2349 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
2350 | if (gfp_mask & __GFP_MEMALLOC) | |
2351 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
907aed48 MG |
2352 | else if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
2353 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
2354 | else if (!in_interrupt() && | |
2355 | ((current->flags & PF_MEMALLOC) || | |
2356 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
341ce06f | 2357 | alloc_flags |= ALLOC_NO_WATERMARKS; |
1da177e4 | 2358 | } |
d95ea5d1 BZ |
2359 | #ifdef CONFIG_CMA |
2360 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2361 | alloc_flags |= ALLOC_CMA; | |
2362 | #endif | |
341ce06f PZ |
2363 | return alloc_flags; |
2364 | } | |
2365 | ||
072bb0aa MG |
2366 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
2367 | { | |
b37f1dd0 | 2368 | return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS); |
072bb0aa MG |
2369 | } |
2370 | ||
11e33f6a MG |
2371 | static inline struct page * |
2372 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
2373 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
3dd28266 MG |
2374 | nodemask_t *nodemask, struct zone *preferred_zone, |
2375 | int migratetype) | |
11e33f6a MG |
2376 | { |
2377 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
2378 | struct page *page = NULL; | |
2379 | int alloc_flags; | |
2380 | unsigned long pages_reclaimed = 0; | |
2381 | unsigned long did_some_progress; | |
77f1fe6b | 2382 | bool sync_migration = false; |
66199712 | 2383 | bool deferred_compaction = false; |
c67fe375 | 2384 | bool contended_compaction = false; |
1da177e4 | 2385 | |
72807a74 MG |
2386 | /* |
2387 | * In the slowpath, we sanity check order to avoid ever trying to | |
2388 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
2389 | * be using allocators in order of preference for an area that is | |
2390 | * too large. | |
2391 | */ | |
1fc28b70 MG |
2392 | if (order >= MAX_ORDER) { |
2393 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
72807a74 | 2394 | return NULL; |
1fc28b70 | 2395 | } |
1da177e4 | 2396 | |
952f3b51 CL |
2397 | /* |
2398 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
2399 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
2400 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
2401 | * using a larger set of nodes after it has established that the | |
2402 | * allowed per node queues are empty and that nodes are | |
2403 | * over allocated. | |
2404 | */ | |
e5adfffc KS |
2405 | if (IS_ENABLED(CONFIG_NUMA) && |
2406 | (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
952f3b51 CL |
2407 | goto nopage; |
2408 | ||
cc4a6851 | 2409 | restart: |
caf49191 LT |
2410 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
2411 | wake_all_kswapd(order, zonelist, high_zoneidx, | |
2412 | zone_idx(preferred_zone)); | |
1da177e4 | 2413 | |
9bf2229f | 2414 | /* |
7fb1d9fc RS |
2415 | * OK, we're below the kswapd watermark and have kicked background |
2416 | * reclaim. Now things get more complex, so set up alloc_flags according | |
2417 | * to how we want to proceed. | |
9bf2229f | 2418 | */ |
341ce06f | 2419 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
1da177e4 | 2420 | |
f33261d7 DR |
2421 | /* |
2422 | * Find the true preferred zone if the allocation is unconstrained by | |
2423 | * cpusets. | |
2424 | */ | |
2425 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) | |
2426 | first_zones_zonelist(zonelist, high_zoneidx, NULL, | |
2427 | &preferred_zone); | |
2428 | ||
cfa54a0f | 2429 | rebalance: |
341ce06f | 2430 | /* This is the last chance, in general, before the goto nopage. */ |
19770b32 | 2431 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
341ce06f PZ |
2432 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
2433 | preferred_zone, migratetype); | |
7fb1d9fc RS |
2434 | if (page) |
2435 | goto got_pg; | |
1da177e4 | 2436 | |
11e33f6a | 2437 | /* Allocate without watermarks if the context allows */ |
341ce06f | 2438 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
183f6371 MG |
2439 | /* |
2440 | * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds | |
2441 | * the allocation is high priority and these type of | |
2442 | * allocations are system rather than user orientated | |
2443 | */ | |
2444 | zonelist = node_zonelist(numa_node_id(), gfp_mask); | |
2445 | ||
341ce06f PZ |
2446 | page = __alloc_pages_high_priority(gfp_mask, order, |
2447 | zonelist, high_zoneidx, nodemask, | |
2448 | preferred_zone, migratetype); | |
cfd19c5a | 2449 | if (page) { |
341ce06f | 2450 | goto got_pg; |
cfd19c5a | 2451 | } |
1da177e4 LT |
2452 | } |
2453 | ||
2454 | /* Atomic allocations - we can't balance anything */ | |
2455 | if (!wait) | |
2456 | goto nopage; | |
2457 | ||
341ce06f | 2458 | /* Avoid recursion of direct reclaim */ |
c06b1fca | 2459 | if (current->flags & PF_MEMALLOC) |
341ce06f PZ |
2460 | goto nopage; |
2461 | ||
6583bb64 DR |
2462 | /* Avoid allocations with no watermarks from looping endlessly */ |
2463 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
2464 | goto nopage; | |
2465 | ||
77f1fe6b MG |
2466 | /* |
2467 | * Try direct compaction. The first pass is asynchronous. Subsequent | |
2468 | * attempts after direct reclaim are synchronous | |
2469 | */ | |
56de7263 MG |
2470 | page = __alloc_pages_direct_compact(gfp_mask, order, |
2471 | zonelist, high_zoneidx, | |
2472 | nodemask, | |
2473 | alloc_flags, preferred_zone, | |
66199712 | 2474 | migratetype, sync_migration, |
c67fe375 | 2475 | &contended_compaction, |
66199712 MG |
2476 | &deferred_compaction, |
2477 | &did_some_progress); | |
56de7263 MG |
2478 | if (page) |
2479 | goto got_pg; | |
c6a140bf | 2480 | sync_migration = true; |
56de7263 | 2481 | |
31f8d42d LT |
2482 | /* |
2483 | * If compaction is deferred for high-order allocations, it is because | |
2484 | * sync compaction recently failed. In this is the case and the caller | |
2485 | * requested a movable allocation that does not heavily disrupt the | |
2486 | * system then fail the allocation instead of entering direct reclaim. | |
2487 | */ | |
2488 | if ((deferred_compaction || contended_compaction) && | |
caf49191 | 2489 | (gfp_mask & __GFP_NO_KSWAPD)) |
31f8d42d | 2490 | goto nopage; |
66199712 | 2491 | |
11e33f6a MG |
2492 | /* Try direct reclaim and then allocating */ |
2493 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
2494 | zonelist, high_zoneidx, | |
2495 | nodemask, | |
5117f45d | 2496 | alloc_flags, preferred_zone, |
3dd28266 | 2497 | migratetype, &did_some_progress); |
11e33f6a MG |
2498 | if (page) |
2499 | goto got_pg; | |
1da177e4 | 2500 | |
e33c3b5e | 2501 | /* |
11e33f6a MG |
2502 | * If we failed to make any progress reclaiming, then we are |
2503 | * running out of options and have to consider going OOM | |
e33c3b5e | 2504 | */ |
11e33f6a MG |
2505 | if (!did_some_progress) { |
2506 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
7f33d49a RW |
2507 | if (oom_killer_disabled) |
2508 | goto nopage; | |
29fd66d2 DR |
2509 | /* Coredumps can quickly deplete all memory reserves */ |
2510 | if ((current->flags & PF_DUMPCORE) && | |
2511 | !(gfp_mask & __GFP_NOFAIL)) | |
2512 | goto nopage; | |
11e33f6a MG |
2513 | page = __alloc_pages_may_oom(gfp_mask, order, |
2514 | zonelist, high_zoneidx, | |
3dd28266 MG |
2515 | nodemask, preferred_zone, |
2516 | migratetype); | |
11e33f6a MG |
2517 | if (page) |
2518 | goto got_pg; | |
1da177e4 | 2519 | |
03668b3c DR |
2520 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2521 | /* | |
2522 | * The oom killer is not called for high-order | |
2523 | * allocations that may fail, so if no progress | |
2524 | * is being made, there are no other options and | |
2525 | * retrying is unlikely to help. | |
2526 | */ | |
2527 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
2528 | goto nopage; | |
2529 | /* | |
2530 | * The oom killer is not called for lowmem | |
2531 | * allocations to prevent needlessly killing | |
2532 | * innocent tasks. | |
2533 | */ | |
2534 | if (high_zoneidx < ZONE_NORMAL) | |
2535 | goto nopage; | |
2536 | } | |
e2c55dc8 | 2537 | |
ff0ceb9d DR |
2538 | goto restart; |
2539 | } | |
1da177e4 LT |
2540 | } |
2541 | ||
11e33f6a | 2542 | /* Check if we should retry the allocation */ |
a41f24ea | 2543 | pages_reclaimed += did_some_progress; |
f90ac398 MG |
2544 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
2545 | pages_reclaimed)) { | |
11e33f6a | 2546 | /* Wait for some write requests to complete then retry */ |
0e093d99 | 2547 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
1da177e4 | 2548 | goto rebalance; |
3e7d3449 MG |
2549 | } else { |
2550 | /* | |
2551 | * High-order allocations do not necessarily loop after | |
2552 | * direct reclaim and reclaim/compaction depends on compaction | |
2553 | * being called after reclaim so call directly if necessary | |
2554 | */ | |
2555 | page = __alloc_pages_direct_compact(gfp_mask, order, | |
2556 | zonelist, high_zoneidx, | |
2557 | nodemask, | |
2558 | alloc_flags, preferred_zone, | |
66199712 | 2559 | migratetype, sync_migration, |
c67fe375 | 2560 | &contended_compaction, |
66199712 MG |
2561 | &deferred_compaction, |
2562 | &did_some_progress); | |
3e7d3449 MG |
2563 | if (page) |
2564 | goto got_pg; | |
1da177e4 LT |
2565 | } |
2566 | ||
2567 | nopage: | |
a238ab5b | 2568 | warn_alloc_failed(gfp_mask, order, NULL); |
b1eeab67 | 2569 | return page; |
1da177e4 | 2570 | got_pg: |
b1eeab67 VN |
2571 | if (kmemcheck_enabled) |
2572 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
11e33f6a | 2573 | |
072bb0aa | 2574 | return page; |
1da177e4 | 2575 | } |
11e33f6a MG |
2576 | |
2577 | /* | |
2578 | * This is the 'heart' of the zoned buddy allocator. | |
2579 | */ | |
2580 | struct page * | |
2581 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
2582 | struct zonelist *zonelist, nodemask_t *nodemask) | |
2583 | { | |
2584 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
5117f45d | 2585 | struct zone *preferred_zone; |
cc9a6c87 | 2586 | struct page *page = NULL; |
3dd28266 | 2587 | int migratetype = allocflags_to_migratetype(gfp_mask); |
cc9a6c87 | 2588 | unsigned int cpuset_mems_cookie; |
d95ea5d1 | 2589 | int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET; |
6a1a0d3b | 2590 | struct mem_cgroup *memcg = NULL; |
11e33f6a | 2591 | |
dcce284a BH |
2592 | gfp_mask &= gfp_allowed_mask; |
2593 | ||
11e33f6a MG |
2594 | lockdep_trace_alloc(gfp_mask); |
2595 | ||
2596 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2597 | ||
2598 | if (should_fail_alloc_page(gfp_mask, order)) | |
2599 | return NULL; | |
2600 | ||
2601 | /* | |
2602 | * Check the zones suitable for the gfp_mask contain at least one | |
2603 | * valid zone. It's possible to have an empty zonelist as a result | |
2604 | * of GFP_THISNODE and a memoryless node | |
2605 | */ | |
2606 | if (unlikely(!zonelist->_zonerefs->zone)) | |
2607 | return NULL; | |
2608 | ||
6a1a0d3b GC |
2609 | /* |
2610 | * Will only have any effect when __GFP_KMEMCG is set. This is | |
2611 | * verified in the (always inline) callee | |
2612 | */ | |
2613 | if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order)) | |
2614 | return NULL; | |
2615 | ||
cc9a6c87 MG |
2616 | retry_cpuset: |
2617 | cpuset_mems_cookie = get_mems_allowed(); | |
2618 | ||
5117f45d | 2619 | /* The preferred zone is used for statistics later */ |
f33261d7 DR |
2620 | first_zones_zonelist(zonelist, high_zoneidx, |
2621 | nodemask ? : &cpuset_current_mems_allowed, | |
2622 | &preferred_zone); | |
cc9a6c87 MG |
2623 | if (!preferred_zone) |
2624 | goto out; | |
5117f45d | 2625 | |
d95ea5d1 BZ |
2626 | #ifdef CONFIG_CMA |
2627 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
2628 | alloc_flags |= ALLOC_CMA; | |
2629 | #endif | |
5117f45d | 2630 | /* First allocation attempt */ |
11e33f6a | 2631 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
d95ea5d1 | 2632 | zonelist, high_zoneidx, alloc_flags, |
3dd28266 | 2633 | preferred_zone, migratetype); |
21caf2fc ML |
2634 | if (unlikely(!page)) { |
2635 | /* | |
2636 | * Runtime PM, block IO and its error handling path | |
2637 | * can deadlock because I/O on the device might not | |
2638 | * complete. | |
2639 | */ | |
2640 | gfp_mask = memalloc_noio_flags(gfp_mask); | |
11e33f6a | 2641 | page = __alloc_pages_slowpath(gfp_mask, order, |
5117f45d | 2642 | zonelist, high_zoneidx, nodemask, |
3dd28266 | 2643 | preferred_zone, migratetype); |
21caf2fc | 2644 | } |
11e33f6a | 2645 | |
4b4f278c | 2646 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
cc9a6c87 MG |
2647 | |
2648 | out: | |
2649 | /* | |
2650 | * When updating a task's mems_allowed, it is possible to race with | |
2651 | * parallel threads in such a way that an allocation can fail while | |
2652 | * the mask is being updated. If a page allocation is about to fail, | |
2653 | * check if the cpuset changed during allocation and if so, retry. | |
2654 | */ | |
2655 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) | |
2656 | goto retry_cpuset; | |
2657 | ||
6a1a0d3b GC |
2658 | memcg_kmem_commit_charge(page, memcg, order); |
2659 | ||
11e33f6a | 2660 | return page; |
1da177e4 | 2661 | } |
d239171e | 2662 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
1da177e4 LT |
2663 | |
2664 | /* | |
2665 | * Common helper functions. | |
2666 | */ | |
920c7a5d | 2667 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 2668 | { |
945a1113 AM |
2669 | struct page *page; |
2670 | ||
2671 | /* | |
2672 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
2673 | * a highmem page | |
2674 | */ | |
2675 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
2676 | ||
1da177e4 LT |
2677 | page = alloc_pages(gfp_mask, order); |
2678 | if (!page) | |
2679 | return 0; | |
2680 | return (unsigned long) page_address(page); | |
2681 | } | |
1da177e4 LT |
2682 | EXPORT_SYMBOL(__get_free_pages); |
2683 | ||
920c7a5d | 2684 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 | 2685 | { |
945a1113 | 2686 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
1da177e4 | 2687 | } |
1da177e4 LT |
2688 | EXPORT_SYMBOL(get_zeroed_page); |
2689 | ||
920c7a5d | 2690 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 2691 | { |
b5810039 | 2692 | if (put_page_testzero(page)) { |
1da177e4 | 2693 | if (order == 0) |
fc91668e | 2694 | free_hot_cold_page(page, 0); |
1da177e4 LT |
2695 | else |
2696 | __free_pages_ok(page, order); | |
2697 | } | |
2698 | } | |
2699 | ||
2700 | EXPORT_SYMBOL(__free_pages); | |
2701 | ||
920c7a5d | 2702 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
2703 | { |
2704 | if (addr != 0) { | |
725d704e | 2705 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
2706 | __free_pages(virt_to_page((void *)addr), order); |
2707 | } | |
2708 | } | |
2709 | ||
2710 | EXPORT_SYMBOL(free_pages); | |
2711 | ||
6a1a0d3b GC |
2712 | /* |
2713 | * __free_memcg_kmem_pages and free_memcg_kmem_pages will free | |
2714 | * pages allocated with __GFP_KMEMCG. | |
2715 | * | |
2716 | * Those pages are accounted to a particular memcg, embedded in the | |
2717 | * corresponding page_cgroup. To avoid adding a hit in the allocator to search | |
2718 | * for that information only to find out that it is NULL for users who have no | |
2719 | * interest in that whatsoever, we provide these functions. | |
2720 | * | |
2721 | * The caller knows better which flags it relies on. | |
2722 | */ | |
2723 | void __free_memcg_kmem_pages(struct page *page, unsigned int order) | |
2724 | { | |
2725 | memcg_kmem_uncharge_pages(page, order); | |
2726 | __free_pages(page, order); | |
2727 | } | |
2728 | ||
2729 | void free_memcg_kmem_pages(unsigned long addr, unsigned int order) | |
2730 | { | |
2731 | if (addr != 0) { | |
2732 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
2733 | __free_memcg_kmem_pages(virt_to_page((void *)addr), order); | |
2734 | } | |
2735 | } | |
2736 | ||
ee85c2e1 AK |
2737 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
2738 | { | |
2739 | if (addr) { | |
2740 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2741 | unsigned long used = addr + PAGE_ALIGN(size); | |
2742 | ||
2743 | split_page(virt_to_page((void *)addr), order); | |
2744 | while (used < alloc_end) { | |
2745 | free_page(used); | |
2746 | used += PAGE_SIZE; | |
2747 | } | |
2748 | } | |
2749 | return (void *)addr; | |
2750 | } | |
2751 | ||
2be0ffe2 TT |
2752 | /** |
2753 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2754 | * @size: the number of bytes to allocate | |
2755 | * @gfp_mask: GFP flags for the allocation | |
2756 | * | |
2757 | * This function is similar to alloc_pages(), except that it allocates the | |
2758 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2759 | * allocate memory in power-of-two pages. | |
2760 | * | |
2761 | * This function is also limited by MAX_ORDER. | |
2762 | * | |
2763 | * Memory allocated by this function must be released by free_pages_exact(). | |
2764 | */ | |
2765 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2766 | { | |
2767 | unsigned int order = get_order(size); | |
2768 | unsigned long addr; | |
2769 | ||
2770 | addr = __get_free_pages(gfp_mask, order); | |
ee85c2e1 | 2771 | return make_alloc_exact(addr, order, size); |
2be0ffe2 TT |
2772 | } |
2773 | EXPORT_SYMBOL(alloc_pages_exact); | |
2774 | ||
ee85c2e1 AK |
2775 | /** |
2776 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous | |
2777 | * pages on a node. | |
b5e6ab58 | 2778 | * @nid: the preferred node ID where memory should be allocated |
ee85c2e1 AK |
2779 | * @size: the number of bytes to allocate |
2780 | * @gfp_mask: GFP flags for the allocation | |
2781 | * | |
2782 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling | |
2783 | * back. | |
2784 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, | |
2785 | * but is not exact. | |
2786 | */ | |
2787 | void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) | |
2788 | { | |
2789 | unsigned order = get_order(size); | |
2790 | struct page *p = alloc_pages_node(nid, gfp_mask, order); | |
2791 | if (!p) | |
2792 | return NULL; | |
2793 | return make_alloc_exact((unsigned long)page_address(p), order, size); | |
2794 | } | |
2795 | EXPORT_SYMBOL(alloc_pages_exact_nid); | |
2796 | ||
2be0ffe2 TT |
2797 | /** |
2798 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2799 | * @virt: the value returned by alloc_pages_exact. | |
2800 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2801 | * | |
2802 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2803 | */ | |
2804 | void free_pages_exact(void *virt, size_t size) | |
2805 | { | |
2806 | unsigned long addr = (unsigned long)virt; | |
2807 | unsigned long end = addr + PAGE_ALIGN(size); | |
2808 | ||
2809 | while (addr < end) { | |
2810 | free_page(addr); | |
2811 | addr += PAGE_SIZE; | |
2812 | } | |
2813 | } | |
2814 | EXPORT_SYMBOL(free_pages_exact); | |
2815 | ||
e0fb5815 ZY |
2816 | /** |
2817 | * nr_free_zone_pages - count number of pages beyond high watermark | |
2818 | * @offset: The zone index of the highest zone | |
2819 | * | |
2820 | * nr_free_zone_pages() counts the number of counts pages which are beyond the | |
2821 | * high watermark within all zones at or below a given zone index. For each | |
2822 | * zone, the number of pages is calculated as: | |
2823 | * present_pages - high_pages | |
2824 | */ | |
ebec3862 | 2825 | static unsigned long nr_free_zone_pages(int offset) |
1da177e4 | 2826 | { |
dd1a239f | 2827 | struct zoneref *z; |
54a6eb5c MG |
2828 | struct zone *zone; |
2829 | ||
e310fd43 | 2830 | /* Just pick one node, since fallback list is circular */ |
ebec3862 | 2831 | unsigned long sum = 0; |
1da177e4 | 2832 | |
0e88460d | 2833 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
1da177e4 | 2834 | |
54a6eb5c | 2835 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
b40da049 | 2836 | unsigned long size = zone->managed_pages; |
41858966 | 2837 | unsigned long high = high_wmark_pages(zone); |
e310fd43 MB |
2838 | if (size > high) |
2839 | sum += size - high; | |
1da177e4 LT |
2840 | } |
2841 | ||
2842 | return sum; | |
2843 | } | |
2844 | ||
e0fb5815 ZY |
2845 | /** |
2846 | * nr_free_buffer_pages - count number of pages beyond high watermark | |
2847 | * | |
2848 | * nr_free_buffer_pages() counts the number of pages which are beyond the high | |
2849 | * watermark within ZONE_DMA and ZONE_NORMAL. | |
1da177e4 | 2850 | */ |
ebec3862 | 2851 | unsigned long nr_free_buffer_pages(void) |
1da177e4 | 2852 | { |
af4ca457 | 2853 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 2854 | } |
c2f1a551 | 2855 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 | 2856 | |
e0fb5815 ZY |
2857 | /** |
2858 | * nr_free_pagecache_pages - count number of pages beyond high watermark | |
2859 | * | |
2860 | * nr_free_pagecache_pages() counts the number of pages which are beyond the | |
2861 | * high watermark within all zones. | |
1da177e4 | 2862 | */ |
ebec3862 | 2863 | unsigned long nr_free_pagecache_pages(void) |
1da177e4 | 2864 | { |
2a1e274a | 2865 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 2866 | } |
08e0f6a9 CL |
2867 | |
2868 | static inline void show_node(struct zone *zone) | |
1da177e4 | 2869 | { |
e5adfffc | 2870 | if (IS_ENABLED(CONFIG_NUMA)) |
25ba77c1 | 2871 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 2872 | } |
1da177e4 | 2873 | |
1da177e4 LT |
2874 | void si_meminfo(struct sysinfo *val) |
2875 | { | |
2876 | val->totalram = totalram_pages; | |
2877 | val->sharedram = 0; | |
d23ad423 | 2878 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 2879 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
2880 | val->totalhigh = totalhigh_pages; |
2881 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
2882 | val->mem_unit = PAGE_SIZE; |
2883 | } | |
2884 | ||
2885 | EXPORT_SYMBOL(si_meminfo); | |
2886 | ||
2887 | #ifdef CONFIG_NUMA | |
2888 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2889 | { | |
2890 | pg_data_t *pgdat = NODE_DATA(nid); | |
2891 | ||
2892 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 2893 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 2894 | #ifdef CONFIG_HIGHMEM |
b40da049 | 2895 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages; |
d23ad423 CL |
2896 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
2897 | NR_FREE_PAGES); | |
98d2b0eb CL |
2898 | #else |
2899 | val->totalhigh = 0; | |
2900 | val->freehigh = 0; | |
2901 | #endif | |
1da177e4 LT |
2902 | val->mem_unit = PAGE_SIZE; |
2903 | } | |
2904 | #endif | |
2905 | ||
ddd588b5 | 2906 | /* |
7bf02ea2 DR |
2907 | * Determine whether the node should be displayed or not, depending on whether |
2908 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). | |
ddd588b5 | 2909 | */ |
7bf02ea2 | 2910 | bool skip_free_areas_node(unsigned int flags, int nid) |
ddd588b5 DR |
2911 | { |
2912 | bool ret = false; | |
cc9a6c87 | 2913 | unsigned int cpuset_mems_cookie; |
ddd588b5 DR |
2914 | |
2915 | if (!(flags & SHOW_MEM_FILTER_NODES)) | |
2916 | goto out; | |
2917 | ||
cc9a6c87 MG |
2918 | do { |
2919 | cpuset_mems_cookie = get_mems_allowed(); | |
2920 | ret = !node_isset(nid, cpuset_current_mems_allowed); | |
2921 | } while (!put_mems_allowed(cpuset_mems_cookie)); | |
ddd588b5 DR |
2922 | out: |
2923 | return ret; | |
2924 | } | |
2925 | ||
1da177e4 LT |
2926 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
2927 | ||
377e4f16 RV |
2928 | static void show_migration_types(unsigned char type) |
2929 | { | |
2930 | static const char types[MIGRATE_TYPES] = { | |
2931 | [MIGRATE_UNMOVABLE] = 'U', | |
2932 | [MIGRATE_RECLAIMABLE] = 'E', | |
2933 | [MIGRATE_MOVABLE] = 'M', | |
2934 | [MIGRATE_RESERVE] = 'R', | |
2935 | #ifdef CONFIG_CMA | |
2936 | [MIGRATE_CMA] = 'C', | |
2937 | #endif | |
194159fb | 2938 | #ifdef CONFIG_MEMORY_ISOLATION |
377e4f16 | 2939 | [MIGRATE_ISOLATE] = 'I', |
194159fb | 2940 | #endif |
377e4f16 RV |
2941 | }; |
2942 | char tmp[MIGRATE_TYPES + 1]; | |
2943 | char *p = tmp; | |
2944 | int i; | |
2945 | ||
2946 | for (i = 0; i < MIGRATE_TYPES; i++) { | |
2947 | if (type & (1 << i)) | |
2948 | *p++ = types[i]; | |
2949 | } | |
2950 | ||
2951 | *p = '\0'; | |
2952 | printk("(%s) ", tmp); | |
2953 | } | |
2954 | ||
1da177e4 LT |
2955 | /* |
2956 | * Show free area list (used inside shift_scroll-lock stuff) | |
2957 | * We also calculate the percentage fragmentation. We do this by counting the | |
2958 | * memory on each free list with the exception of the first item on the list. | |
ddd588b5 DR |
2959 | * Suppresses nodes that are not allowed by current's cpuset if |
2960 | * SHOW_MEM_FILTER_NODES is passed. | |
1da177e4 | 2961 | */ |
7bf02ea2 | 2962 | void show_free_areas(unsigned int filter) |
1da177e4 | 2963 | { |
c7241913 | 2964 | int cpu; |
1da177e4 LT |
2965 | struct zone *zone; |
2966 | ||
ee99c71c | 2967 | for_each_populated_zone(zone) { |
7bf02ea2 | 2968 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 2969 | continue; |
c7241913 JS |
2970 | show_node(zone); |
2971 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 2972 | |
6b482c67 | 2973 | for_each_online_cpu(cpu) { |
1da177e4 LT |
2974 | struct per_cpu_pageset *pageset; |
2975 | ||
99dcc3e5 | 2976 | pageset = per_cpu_ptr(zone->pageset, cpu); |
1da177e4 | 2977 | |
3dfa5721 CL |
2978 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
2979 | cpu, pageset->pcp.high, | |
2980 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
2981 | } |
2982 | } | |
2983 | ||
a731286d KM |
2984 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
2985 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
7b854121 | 2986 | " unevictable:%lu" |
b76146ed | 2987 | " dirty:%lu writeback:%lu unstable:%lu\n" |
3701b033 | 2988 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
d1ce749a BZ |
2989 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
2990 | " free_cma:%lu\n", | |
4f98a2fe | 2991 | global_page_state(NR_ACTIVE_ANON), |
4f98a2fe | 2992 | global_page_state(NR_INACTIVE_ANON), |
a731286d KM |
2993 | global_page_state(NR_ISOLATED_ANON), |
2994 | global_page_state(NR_ACTIVE_FILE), | |
4f98a2fe | 2995 | global_page_state(NR_INACTIVE_FILE), |
a731286d | 2996 | global_page_state(NR_ISOLATED_FILE), |
7b854121 | 2997 | global_page_state(NR_UNEVICTABLE), |
b1e7a8fd | 2998 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 2999 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 3000 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 3001 | global_page_state(NR_FREE_PAGES), |
3701b033 KM |
3002 | global_page_state(NR_SLAB_RECLAIMABLE), |
3003 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 3004 | global_page_state(NR_FILE_MAPPED), |
4b02108a | 3005 | global_page_state(NR_SHMEM), |
a25700a5 | 3006 | global_page_state(NR_PAGETABLE), |
d1ce749a BZ |
3007 | global_page_state(NR_BOUNCE), |
3008 | global_page_state(NR_FREE_CMA_PAGES)); | |
1da177e4 | 3009 | |
ee99c71c | 3010 | for_each_populated_zone(zone) { |
1da177e4 LT |
3011 | int i; |
3012 | ||
7bf02ea2 | 3013 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3014 | continue; |
1da177e4 LT |
3015 | show_node(zone); |
3016 | printk("%s" | |
3017 | " free:%lukB" | |
3018 | " min:%lukB" | |
3019 | " low:%lukB" | |
3020 | " high:%lukB" | |
4f98a2fe RR |
3021 | " active_anon:%lukB" |
3022 | " inactive_anon:%lukB" | |
3023 | " active_file:%lukB" | |
3024 | " inactive_file:%lukB" | |
7b854121 | 3025 | " unevictable:%lukB" |
a731286d KM |
3026 | " isolated(anon):%lukB" |
3027 | " isolated(file):%lukB" | |
1da177e4 | 3028 | " present:%lukB" |
9feedc9d | 3029 | " managed:%lukB" |
4a0aa73f KM |
3030 | " mlocked:%lukB" |
3031 | " dirty:%lukB" | |
3032 | " writeback:%lukB" | |
3033 | " mapped:%lukB" | |
4b02108a | 3034 | " shmem:%lukB" |
4a0aa73f KM |
3035 | " slab_reclaimable:%lukB" |
3036 | " slab_unreclaimable:%lukB" | |
c6a7f572 | 3037 | " kernel_stack:%lukB" |
4a0aa73f KM |
3038 | " pagetables:%lukB" |
3039 | " unstable:%lukB" | |
3040 | " bounce:%lukB" | |
d1ce749a | 3041 | " free_cma:%lukB" |
4a0aa73f | 3042 | " writeback_tmp:%lukB" |
1da177e4 LT |
3043 | " pages_scanned:%lu" |
3044 | " all_unreclaimable? %s" | |
3045 | "\n", | |
3046 | zone->name, | |
88f5acf8 | 3047 | K(zone_page_state(zone, NR_FREE_PAGES)), |
41858966 MG |
3048 | K(min_wmark_pages(zone)), |
3049 | K(low_wmark_pages(zone)), | |
3050 | K(high_wmark_pages(zone)), | |
4f98a2fe RR |
3051 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
3052 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
3053 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
3054 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
7b854121 | 3055 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
a731286d KM |
3056 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
3057 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
1da177e4 | 3058 | K(zone->present_pages), |
9feedc9d | 3059 | K(zone->managed_pages), |
4a0aa73f KM |
3060 | K(zone_page_state(zone, NR_MLOCK)), |
3061 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
3062 | K(zone_page_state(zone, NR_WRITEBACK)), | |
3063 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
4b02108a | 3064 | K(zone_page_state(zone, NR_SHMEM)), |
4a0aa73f KM |
3065 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
3066 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
c6a7f572 KM |
3067 | zone_page_state(zone, NR_KERNEL_STACK) * |
3068 | THREAD_SIZE / 1024, | |
4a0aa73f KM |
3069 | K(zone_page_state(zone, NR_PAGETABLE)), |
3070 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
3071 | K(zone_page_state(zone, NR_BOUNCE)), | |
d1ce749a | 3072 | K(zone_page_state(zone, NR_FREE_CMA_PAGES)), |
4a0aa73f | 3073 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
1da177e4 | 3074 | zone->pages_scanned, |
93e4a89a | 3075 | (zone->all_unreclaimable ? "yes" : "no") |
1da177e4 LT |
3076 | ); |
3077 | printk("lowmem_reserve[]:"); | |
3078 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3079 | printk(" %lu", zone->lowmem_reserve[i]); | |
3080 | printk("\n"); | |
3081 | } | |
3082 | ||
ee99c71c | 3083 | for_each_populated_zone(zone) { |
8f9de51a | 3084 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
377e4f16 | 3085 | unsigned char types[MAX_ORDER]; |
1da177e4 | 3086 | |
7bf02ea2 | 3087 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
ddd588b5 | 3088 | continue; |
1da177e4 LT |
3089 | show_node(zone); |
3090 | printk("%s: ", zone->name); | |
1da177e4 LT |
3091 | |
3092 | spin_lock_irqsave(&zone->lock, flags); | |
3093 | for (order = 0; order < MAX_ORDER; order++) { | |
377e4f16 RV |
3094 | struct free_area *area = &zone->free_area[order]; |
3095 | int type; | |
3096 | ||
3097 | nr[order] = area->nr_free; | |
8f9de51a | 3098 | total += nr[order] << order; |
377e4f16 RV |
3099 | |
3100 | types[order] = 0; | |
3101 | for (type = 0; type < MIGRATE_TYPES; type++) { | |
3102 | if (!list_empty(&area->free_list[type])) | |
3103 | types[order] |= 1 << type; | |
3104 | } | |
1da177e4 LT |
3105 | } |
3106 | spin_unlock_irqrestore(&zone->lock, flags); | |
377e4f16 | 3107 | for (order = 0; order < MAX_ORDER; order++) { |
8f9de51a | 3108 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
377e4f16 RV |
3109 | if (nr[order]) |
3110 | show_migration_types(types[order]); | |
3111 | } | |
1da177e4 LT |
3112 | printk("= %lukB\n", K(total)); |
3113 | } | |
3114 | ||
e6f3602d LW |
3115 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
3116 | ||
1da177e4 LT |
3117 | show_swap_cache_info(); |
3118 | } | |
3119 | ||
19770b32 MG |
3120 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
3121 | { | |
3122 | zoneref->zone = zone; | |
3123 | zoneref->zone_idx = zone_idx(zone); | |
3124 | } | |
3125 | ||
1da177e4 LT |
3126 | /* |
3127 | * Builds allocation fallback zone lists. | |
1a93205b CL |
3128 | * |
3129 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 3130 | */ |
f0c0b2b8 KH |
3131 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
3132 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 3133 | { |
1a93205b CL |
3134 | struct zone *zone; |
3135 | ||
98d2b0eb | 3136 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 3137 | zone_type++; |
02a68a5e CL |
3138 | |
3139 | do { | |
2f6726e5 | 3140 | zone_type--; |
070f8032 | 3141 | zone = pgdat->node_zones + zone_type; |
1a93205b | 3142 | if (populated_zone(zone)) { |
dd1a239f MG |
3143 | zoneref_set_zone(zone, |
3144 | &zonelist->_zonerefs[nr_zones++]); | |
070f8032 | 3145 | check_highest_zone(zone_type); |
1da177e4 | 3146 | } |
02a68a5e | 3147 | |
2f6726e5 | 3148 | } while (zone_type); |
070f8032 | 3149 | return nr_zones; |
1da177e4 LT |
3150 | } |
3151 | ||
f0c0b2b8 KH |
3152 | |
3153 | /* | |
3154 | * zonelist_order: | |
3155 | * 0 = automatic detection of better ordering. | |
3156 | * 1 = order by ([node] distance, -zonetype) | |
3157 | * 2 = order by (-zonetype, [node] distance) | |
3158 | * | |
3159 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
3160 | * the same zonelist. So only NUMA can configure this param. | |
3161 | */ | |
3162 | #define ZONELIST_ORDER_DEFAULT 0 | |
3163 | #define ZONELIST_ORDER_NODE 1 | |
3164 | #define ZONELIST_ORDER_ZONE 2 | |
3165 | ||
3166 | /* zonelist order in the kernel. | |
3167 | * set_zonelist_order() will set this to NODE or ZONE. | |
3168 | */ | |
3169 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3170 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
3171 | ||
3172 | ||
1da177e4 | 3173 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
3174 | /* The value user specified ....changed by config */ |
3175 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3176 | /* string for sysctl */ | |
3177 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
3178 | char numa_zonelist_order[16] = "default"; | |
3179 | ||
3180 | /* | |
3181 | * interface for configure zonelist ordering. | |
3182 | * command line option "numa_zonelist_order" | |
3183 | * = "[dD]efault - default, automatic configuration. | |
3184 | * = "[nN]ode - order by node locality, then by zone within node | |
3185 | * = "[zZ]one - order by zone, then by locality within zone | |
3186 | */ | |
3187 | ||
3188 | static int __parse_numa_zonelist_order(char *s) | |
3189 | { | |
3190 | if (*s == 'd' || *s == 'D') { | |
3191 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
3192 | } else if (*s == 'n' || *s == 'N') { | |
3193 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
3194 | } else if (*s == 'z' || *s == 'Z') { | |
3195 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
3196 | } else { | |
3197 | printk(KERN_WARNING | |
3198 | "Ignoring invalid numa_zonelist_order value: " | |
3199 | "%s\n", s); | |
3200 | return -EINVAL; | |
3201 | } | |
3202 | return 0; | |
3203 | } | |
3204 | ||
3205 | static __init int setup_numa_zonelist_order(char *s) | |
3206 | { | |
ecb256f8 VL |
3207 | int ret; |
3208 | ||
3209 | if (!s) | |
3210 | return 0; | |
3211 | ||
3212 | ret = __parse_numa_zonelist_order(s); | |
3213 | if (ret == 0) | |
3214 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); | |
3215 | ||
3216 | return ret; | |
f0c0b2b8 KH |
3217 | } |
3218 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
3219 | ||
3220 | /* | |
3221 | * sysctl handler for numa_zonelist_order | |
3222 | */ | |
3223 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
8d65af78 | 3224 | void __user *buffer, size_t *length, |
f0c0b2b8 KH |
3225 | loff_t *ppos) |
3226 | { | |
3227 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
3228 | int ret; | |
443c6f14 | 3229 | static DEFINE_MUTEX(zl_order_mutex); |
f0c0b2b8 | 3230 | |
443c6f14 | 3231 | mutex_lock(&zl_order_mutex); |
f0c0b2b8 | 3232 | if (write) |
443c6f14 | 3233 | strcpy(saved_string, (char*)table->data); |
8d65af78 | 3234 | ret = proc_dostring(table, write, buffer, length, ppos); |
f0c0b2b8 | 3235 | if (ret) |
443c6f14 | 3236 | goto out; |
f0c0b2b8 KH |
3237 | if (write) { |
3238 | int oldval = user_zonelist_order; | |
3239 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
3240 | /* | |
3241 | * bogus value. restore saved string | |
3242 | */ | |
3243 | strncpy((char*)table->data, saved_string, | |
3244 | NUMA_ZONELIST_ORDER_LEN); | |
3245 | user_zonelist_order = oldval; | |
4eaf3f64 HL |
3246 | } else if (oldval != user_zonelist_order) { |
3247 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 3248 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
3249 | mutex_unlock(&zonelists_mutex); |
3250 | } | |
f0c0b2b8 | 3251 | } |
443c6f14 AK |
3252 | out: |
3253 | mutex_unlock(&zl_order_mutex); | |
3254 | return ret; | |
f0c0b2b8 KH |
3255 | } |
3256 | ||
3257 | ||
62bc62a8 | 3258 | #define MAX_NODE_LOAD (nr_online_nodes) |
f0c0b2b8 KH |
3259 | static int node_load[MAX_NUMNODES]; |
3260 | ||
1da177e4 | 3261 | /** |
4dc3b16b | 3262 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
3263 | * @node: node whose fallback list we're appending |
3264 | * @used_node_mask: nodemask_t of already used nodes | |
3265 | * | |
3266 | * We use a number of factors to determine which is the next node that should | |
3267 | * appear on a given node's fallback list. The node should not have appeared | |
3268 | * already in @node's fallback list, and it should be the next closest node | |
3269 | * according to the distance array (which contains arbitrary distance values | |
3270 | * from each node to each node in the system), and should also prefer nodes | |
3271 | * with no CPUs, since presumably they'll have very little allocation pressure | |
3272 | * on them otherwise. | |
3273 | * It returns -1 if no node is found. | |
3274 | */ | |
f0c0b2b8 | 3275 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 3276 | { |
4cf808eb | 3277 | int n, val; |
1da177e4 | 3278 | int min_val = INT_MAX; |
00ef2d2f | 3279 | int best_node = NUMA_NO_NODE; |
a70f7302 | 3280 | const struct cpumask *tmp = cpumask_of_node(0); |
1da177e4 | 3281 | |
4cf808eb LT |
3282 | /* Use the local node if we haven't already */ |
3283 | if (!node_isset(node, *used_node_mask)) { | |
3284 | node_set(node, *used_node_mask); | |
3285 | return node; | |
3286 | } | |
1da177e4 | 3287 | |
4b0ef1fe | 3288 | for_each_node_state(n, N_MEMORY) { |
1da177e4 LT |
3289 | |
3290 | /* Don't want a node to appear more than once */ | |
3291 | if (node_isset(n, *used_node_mask)) | |
3292 | continue; | |
3293 | ||
1da177e4 LT |
3294 | /* Use the distance array to find the distance */ |
3295 | val = node_distance(node, n); | |
3296 | ||
4cf808eb LT |
3297 | /* Penalize nodes under us ("prefer the next node") */ |
3298 | val += (n < node); | |
3299 | ||
1da177e4 | 3300 | /* Give preference to headless and unused nodes */ |
a70f7302 RR |
3301 | tmp = cpumask_of_node(n); |
3302 | if (!cpumask_empty(tmp)) | |
1da177e4 LT |
3303 | val += PENALTY_FOR_NODE_WITH_CPUS; |
3304 | ||
3305 | /* Slight preference for less loaded node */ | |
3306 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
3307 | val += node_load[n]; | |
3308 | ||
3309 | if (val < min_val) { | |
3310 | min_val = val; | |
3311 | best_node = n; | |
3312 | } | |
3313 | } | |
3314 | ||
3315 | if (best_node >= 0) | |
3316 | node_set(best_node, *used_node_mask); | |
3317 | ||
3318 | return best_node; | |
3319 | } | |
3320 | ||
f0c0b2b8 KH |
3321 | |
3322 | /* | |
3323 | * Build zonelists ordered by node and zones within node. | |
3324 | * This results in maximum locality--normal zone overflows into local | |
3325 | * DMA zone, if any--but risks exhausting DMA zone. | |
3326 | */ | |
3327 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 3328 | { |
f0c0b2b8 | 3329 | int j; |
1da177e4 | 3330 | struct zonelist *zonelist; |
f0c0b2b8 | 3331 | |
54a6eb5c | 3332 | zonelist = &pgdat->node_zonelists[0]; |
dd1a239f | 3333 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
54a6eb5c MG |
3334 | ; |
3335 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3336 | MAX_NR_ZONES - 1); | |
dd1a239f MG |
3337 | zonelist->_zonerefs[j].zone = NULL; |
3338 | zonelist->_zonerefs[j].zone_idx = 0; | |
f0c0b2b8 KH |
3339 | } |
3340 | ||
523b9458 CL |
3341 | /* |
3342 | * Build gfp_thisnode zonelists | |
3343 | */ | |
3344 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
3345 | { | |
523b9458 CL |
3346 | int j; |
3347 | struct zonelist *zonelist; | |
3348 | ||
54a6eb5c MG |
3349 | zonelist = &pgdat->node_zonelists[1]; |
3350 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
dd1a239f MG |
3351 | zonelist->_zonerefs[j].zone = NULL; |
3352 | zonelist->_zonerefs[j].zone_idx = 0; | |
523b9458 CL |
3353 | } |
3354 | ||
f0c0b2b8 KH |
3355 | /* |
3356 | * Build zonelists ordered by zone and nodes within zones. | |
3357 | * This results in conserving DMA zone[s] until all Normal memory is | |
3358 | * exhausted, but results in overflowing to remote node while memory | |
3359 | * may still exist in local DMA zone. | |
3360 | */ | |
3361 | static int node_order[MAX_NUMNODES]; | |
3362 | ||
3363 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
3364 | { | |
f0c0b2b8 KH |
3365 | int pos, j, node; |
3366 | int zone_type; /* needs to be signed */ | |
3367 | struct zone *z; | |
3368 | struct zonelist *zonelist; | |
3369 | ||
54a6eb5c MG |
3370 | zonelist = &pgdat->node_zonelists[0]; |
3371 | pos = 0; | |
3372 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
3373 | for (j = 0; j < nr_nodes; j++) { | |
3374 | node = node_order[j]; | |
3375 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
3376 | if (populated_zone(z)) { | |
dd1a239f MG |
3377 | zoneref_set_zone(z, |
3378 | &zonelist->_zonerefs[pos++]); | |
54a6eb5c | 3379 | check_highest_zone(zone_type); |
f0c0b2b8 KH |
3380 | } |
3381 | } | |
f0c0b2b8 | 3382 | } |
dd1a239f MG |
3383 | zonelist->_zonerefs[pos].zone = NULL; |
3384 | zonelist->_zonerefs[pos].zone_idx = 0; | |
f0c0b2b8 KH |
3385 | } |
3386 | ||
3387 | static int default_zonelist_order(void) | |
3388 | { | |
3389 | int nid, zone_type; | |
3390 | unsigned long low_kmem_size,total_size; | |
3391 | struct zone *z; | |
3392 | int average_size; | |
3393 | /* | |
88393161 | 3394 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
f0c0b2b8 KH |
3395 | * If they are really small and used heavily, the system can fall |
3396 | * into OOM very easily. | |
e325c90f | 3397 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
f0c0b2b8 KH |
3398 | */ |
3399 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
3400 | low_kmem_size = 0; | |
3401 | total_size = 0; | |
3402 | for_each_online_node(nid) { | |
3403 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3404 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3405 | if (populated_zone(z)) { | |
3406 | if (zone_type < ZONE_NORMAL) | |
3407 | low_kmem_size += z->present_pages; | |
3408 | total_size += z->present_pages; | |
e325c90f DR |
3409 | } else if (zone_type == ZONE_NORMAL) { |
3410 | /* | |
3411 | * If any node has only lowmem, then node order | |
3412 | * is preferred to allow kernel allocations | |
3413 | * locally; otherwise, they can easily infringe | |
3414 | * on other nodes when there is an abundance of | |
3415 | * lowmem available to allocate from. | |
3416 | */ | |
3417 | return ZONELIST_ORDER_NODE; | |
f0c0b2b8 KH |
3418 | } |
3419 | } | |
3420 | } | |
3421 | if (!low_kmem_size || /* there are no DMA area. */ | |
3422 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
3423 | return ZONELIST_ORDER_NODE; | |
3424 | /* | |
3425 | * look into each node's config. | |
3426 | * If there is a node whose DMA/DMA32 memory is very big area on | |
3427 | * local memory, NODE_ORDER may be suitable. | |
3428 | */ | |
37b07e41 | 3429 | average_size = total_size / |
4b0ef1fe | 3430 | (nodes_weight(node_states[N_MEMORY]) + 1); |
f0c0b2b8 KH |
3431 | for_each_online_node(nid) { |
3432 | low_kmem_size = 0; | |
3433 | total_size = 0; | |
3434 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
3435 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
3436 | if (populated_zone(z)) { | |
3437 | if (zone_type < ZONE_NORMAL) | |
3438 | low_kmem_size += z->present_pages; | |
3439 | total_size += z->present_pages; | |
3440 | } | |
3441 | } | |
3442 | if (low_kmem_size && | |
3443 | total_size > average_size && /* ignore small node */ | |
3444 | low_kmem_size > total_size * 70/100) | |
3445 | return ZONELIST_ORDER_NODE; | |
3446 | } | |
3447 | return ZONELIST_ORDER_ZONE; | |
3448 | } | |
3449 | ||
3450 | static void set_zonelist_order(void) | |
3451 | { | |
3452 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
3453 | current_zonelist_order = default_zonelist_order(); | |
3454 | else | |
3455 | current_zonelist_order = user_zonelist_order; | |
3456 | } | |
3457 | ||
3458 | static void build_zonelists(pg_data_t *pgdat) | |
3459 | { | |
3460 | int j, node, load; | |
3461 | enum zone_type i; | |
1da177e4 | 3462 | nodemask_t used_mask; |
f0c0b2b8 KH |
3463 | int local_node, prev_node; |
3464 | struct zonelist *zonelist; | |
3465 | int order = current_zonelist_order; | |
1da177e4 LT |
3466 | |
3467 | /* initialize zonelists */ | |
523b9458 | 3468 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 | 3469 | zonelist = pgdat->node_zonelists + i; |
dd1a239f MG |
3470 | zonelist->_zonerefs[0].zone = NULL; |
3471 | zonelist->_zonerefs[0].zone_idx = 0; | |
1da177e4 LT |
3472 | } |
3473 | ||
3474 | /* NUMA-aware ordering of nodes */ | |
3475 | local_node = pgdat->node_id; | |
62bc62a8 | 3476 | load = nr_online_nodes; |
1da177e4 LT |
3477 | prev_node = local_node; |
3478 | nodes_clear(used_mask); | |
f0c0b2b8 | 3479 | |
f0c0b2b8 KH |
3480 | memset(node_order, 0, sizeof(node_order)); |
3481 | j = 0; | |
3482 | ||
1da177e4 LT |
3483 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
3484 | /* | |
3485 | * We don't want to pressure a particular node. | |
3486 | * So adding penalty to the first node in same | |
3487 | * distance group to make it round-robin. | |
3488 | */ | |
957f822a DR |
3489 | if (node_distance(local_node, node) != |
3490 | node_distance(local_node, prev_node)) | |
f0c0b2b8 KH |
3491 | node_load[node] = load; |
3492 | ||
1da177e4 LT |
3493 | prev_node = node; |
3494 | load--; | |
f0c0b2b8 KH |
3495 | if (order == ZONELIST_ORDER_NODE) |
3496 | build_zonelists_in_node_order(pgdat, node); | |
3497 | else | |
3498 | node_order[j++] = node; /* remember order */ | |
3499 | } | |
1da177e4 | 3500 | |
f0c0b2b8 KH |
3501 | if (order == ZONELIST_ORDER_ZONE) { |
3502 | /* calculate node order -- i.e., DMA last! */ | |
3503 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 3504 | } |
523b9458 CL |
3505 | |
3506 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
3507 | } |
3508 | ||
9276b1bc | 3509 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 3510 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3511 | { |
54a6eb5c MG |
3512 | struct zonelist *zonelist; |
3513 | struct zonelist_cache *zlc; | |
dd1a239f | 3514 | struct zoneref *z; |
9276b1bc | 3515 | |
54a6eb5c MG |
3516 | zonelist = &pgdat->node_zonelists[0]; |
3517 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
3518 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
dd1a239f MG |
3519 | for (z = zonelist->_zonerefs; z->zone; z++) |
3520 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
9276b1bc PJ |
3521 | } |
3522 | ||
7aac7898 LS |
3523 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3524 | /* | |
3525 | * Return node id of node used for "local" allocations. | |
3526 | * I.e., first node id of first zone in arg node's generic zonelist. | |
3527 | * Used for initializing percpu 'numa_mem', which is used primarily | |
3528 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. | |
3529 | */ | |
3530 | int local_memory_node(int node) | |
3531 | { | |
3532 | struct zone *zone; | |
3533 | ||
3534 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), | |
3535 | gfp_zone(GFP_KERNEL), | |
3536 | NULL, | |
3537 | &zone); | |
3538 | return zone->node; | |
3539 | } | |
3540 | #endif | |
f0c0b2b8 | 3541 | |
1da177e4 LT |
3542 | #else /* CONFIG_NUMA */ |
3543 | ||
f0c0b2b8 KH |
3544 | static void set_zonelist_order(void) |
3545 | { | |
3546 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
3547 | } | |
3548 | ||
3549 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 3550 | { |
19655d34 | 3551 | int node, local_node; |
54a6eb5c MG |
3552 | enum zone_type j; |
3553 | struct zonelist *zonelist; | |
1da177e4 LT |
3554 | |
3555 | local_node = pgdat->node_id; | |
1da177e4 | 3556 | |
54a6eb5c MG |
3557 | zonelist = &pgdat->node_zonelists[0]; |
3558 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
1da177e4 | 3559 | |
54a6eb5c MG |
3560 | /* |
3561 | * Now we build the zonelist so that it contains the zones | |
3562 | * of all the other nodes. | |
3563 | * We don't want to pressure a particular node, so when | |
3564 | * building the zones for node N, we make sure that the | |
3565 | * zones coming right after the local ones are those from | |
3566 | * node N+1 (modulo N) | |
3567 | */ | |
3568 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
3569 | if (!node_online(node)) | |
3570 | continue; | |
3571 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3572 | MAX_NR_ZONES - 1); | |
1da177e4 | 3573 | } |
54a6eb5c MG |
3574 | for (node = 0; node < local_node; node++) { |
3575 | if (!node_online(node)) | |
3576 | continue; | |
3577 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
3578 | MAX_NR_ZONES - 1); | |
3579 | } | |
3580 | ||
dd1a239f MG |
3581 | zonelist->_zonerefs[j].zone = NULL; |
3582 | zonelist->_zonerefs[j].zone_idx = 0; | |
1da177e4 LT |
3583 | } |
3584 | ||
9276b1bc | 3585 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 3586 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc | 3587 | { |
54a6eb5c | 3588 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
9276b1bc PJ |
3589 | } |
3590 | ||
1da177e4 LT |
3591 | #endif /* CONFIG_NUMA */ |
3592 | ||
99dcc3e5 CL |
3593 | /* |
3594 | * Boot pageset table. One per cpu which is going to be used for all | |
3595 | * zones and all nodes. The parameters will be set in such a way | |
3596 | * that an item put on a list will immediately be handed over to | |
3597 | * the buddy list. This is safe since pageset manipulation is done | |
3598 | * with interrupts disabled. | |
3599 | * | |
3600 | * The boot_pagesets must be kept even after bootup is complete for | |
3601 | * unused processors and/or zones. They do play a role for bootstrapping | |
3602 | * hotplugged processors. | |
3603 | * | |
3604 | * zoneinfo_show() and maybe other functions do | |
3605 | * not check if the processor is online before following the pageset pointer. | |
3606 | * Other parts of the kernel may not check if the zone is available. | |
3607 | */ | |
3608 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
3609 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
1f522509 | 3610 | static void setup_zone_pageset(struct zone *zone); |
99dcc3e5 | 3611 | |
4eaf3f64 HL |
3612 | /* |
3613 | * Global mutex to protect against size modification of zonelists | |
3614 | * as well as to serialize pageset setup for the new populated zone. | |
3615 | */ | |
3616 | DEFINE_MUTEX(zonelists_mutex); | |
3617 | ||
9b1a4d38 | 3618 | /* return values int ....just for stop_machine() */ |
4ed7e022 | 3619 | static int __build_all_zonelists(void *data) |
1da177e4 | 3620 | { |
6811378e | 3621 | int nid; |
99dcc3e5 | 3622 | int cpu; |
9adb62a5 | 3623 | pg_data_t *self = data; |
9276b1bc | 3624 | |
7f9cfb31 BL |
3625 | #ifdef CONFIG_NUMA |
3626 | memset(node_load, 0, sizeof(node_load)); | |
3627 | #endif | |
9adb62a5 JL |
3628 | |
3629 | if (self && !node_online(self->node_id)) { | |
3630 | build_zonelists(self); | |
3631 | build_zonelist_cache(self); | |
3632 | } | |
3633 | ||
9276b1bc | 3634 | for_each_online_node(nid) { |
7ea1530a CL |
3635 | pg_data_t *pgdat = NODE_DATA(nid); |
3636 | ||
3637 | build_zonelists(pgdat); | |
3638 | build_zonelist_cache(pgdat); | |
9276b1bc | 3639 | } |
99dcc3e5 CL |
3640 | |
3641 | /* | |
3642 | * Initialize the boot_pagesets that are going to be used | |
3643 | * for bootstrapping processors. The real pagesets for | |
3644 | * each zone will be allocated later when the per cpu | |
3645 | * allocator is available. | |
3646 | * | |
3647 | * boot_pagesets are used also for bootstrapping offline | |
3648 | * cpus if the system is already booted because the pagesets | |
3649 | * are needed to initialize allocators on a specific cpu too. | |
3650 | * F.e. the percpu allocator needs the page allocator which | |
3651 | * needs the percpu allocator in order to allocate its pagesets | |
3652 | * (a chicken-egg dilemma). | |
3653 | */ | |
7aac7898 | 3654 | for_each_possible_cpu(cpu) { |
99dcc3e5 CL |
3655 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
3656 | ||
7aac7898 LS |
3657 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
3658 | /* | |
3659 | * We now know the "local memory node" for each node-- | |
3660 | * i.e., the node of the first zone in the generic zonelist. | |
3661 | * Set up numa_mem percpu variable for on-line cpus. During | |
3662 | * boot, only the boot cpu should be on-line; we'll init the | |
3663 | * secondary cpus' numa_mem as they come on-line. During | |
3664 | * node/memory hotplug, we'll fixup all on-line cpus. | |
3665 | */ | |
3666 | if (cpu_online(cpu)) | |
3667 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); | |
3668 | #endif | |
3669 | } | |
3670 | ||
6811378e YG |
3671 | return 0; |
3672 | } | |
3673 | ||
4eaf3f64 HL |
3674 | /* |
3675 | * Called with zonelists_mutex held always | |
3676 | * unless system_state == SYSTEM_BOOTING. | |
3677 | */ | |
9adb62a5 | 3678 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
6811378e | 3679 | { |
f0c0b2b8 KH |
3680 | set_zonelist_order(); |
3681 | ||
6811378e | 3682 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 3683 | __build_all_zonelists(NULL); |
68ad8df4 | 3684 | mminit_verify_zonelist(); |
6811378e YG |
3685 | cpuset_init_current_mems_allowed(); |
3686 | } else { | |
183ff22b | 3687 | /* we have to stop all cpus to guarantee there is no user |
6811378e | 3688 | of zonelist */ |
e9959f0f | 3689 | #ifdef CONFIG_MEMORY_HOTPLUG |
9adb62a5 JL |
3690 | if (zone) |
3691 | setup_zone_pageset(zone); | |
e9959f0f | 3692 | #endif |
9adb62a5 | 3693 | stop_machine(__build_all_zonelists, pgdat, NULL); |
6811378e YG |
3694 | /* cpuset refresh routine should be here */ |
3695 | } | |
bd1e22b8 | 3696 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
3697 | /* |
3698 | * Disable grouping by mobility if the number of pages in the | |
3699 | * system is too low to allow the mechanism to work. It would be | |
3700 | * more accurate, but expensive to check per-zone. This check is | |
3701 | * made on memory-hotadd so a system can start with mobility | |
3702 | * disabled and enable it later | |
3703 | */ | |
d9c23400 | 3704 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
3705 | page_group_by_mobility_disabled = 1; |
3706 | else | |
3707 | page_group_by_mobility_disabled = 0; | |
3708 | ||
3709 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
3710 | "Total pages: %ld\n", | |
62bc62a8 | 3711 | nr_online_nodes, |
f0c0b2b8 | 3712 | zonelist_order_name[current_zonelist_order], |
9ef9acb0 | 3713 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
3714 | vm_total_pages); |
3715 | #ifdef CONFIG_NUMA | |
3716 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
3717 | #endif | |
1da177e4 LT |
3718 | } |
3719 | ||
3720 | /* | |
3721 | * Helper functions to size the waitqueue hash table. | |
3722 | * Essentially these want to choose hash table sizes sufficiently | |
3723 | * large so that collisions trying to wait on pages are rare. | |
3724 | * But in fact, the number of active page waitqueues on typical | |
3725 | * systems is ridiculously low, less than 200. So this is even | |
3726 | * conservative, even though it seems large. | |
3727 | * | |
3728 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
3729 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
3730 | */ | |
3731 | #define PAGES_PER_WAITQUEUE 256 | |
3732 | ||
cca448fe | 3733 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 3734 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
3735 | { |
3736 | unsigned long size = 1; | |
3737 | ||
3738 | pages /= PAGES_PER_WAITQUEUE; | |
3739 | ||
3740 | while (size < pages) | |
3741 | size <<= 1; | |
3742 | ||
3743 | /* | |
3744 | * Once we have dozens or even hundreds of threads sleeping | |
3745 | * on IO we've got bigger problems than wait queue collision. | |
3746 | * Limit the size of the wait table to a reasonable size. | |
3747 | */ | |
3748 | size = min(size, 4096UL); | |
3749 | ||
3750 | return max(size, 4UL); | |
3751 | } | |
cca448fe YG |
3752 | #else |
3753 | /* | |
3754 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
3755 | * a suitable size for its wait_table. So we use the maximum size now. | |
3756 | * | |
3757 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
3758 | * | |
3759 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
3760 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
3761 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
3762 | * | |
3763 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
3764 | * or more by the traditional way. (See above). It equals: | |
3765 | * | |
3766 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
3767 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
3768 | * powerpc (64K page size) : = (32G +16M)byte. | |
3769 | */ | |
3770 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
3771 | { | |
3772 | return 4096UL; | |
3773 | } | |
3774 | #endif | |
1da177e4 LT |
3775 | |
3776 | /* | |
3777 | * This is an integer logarithm so that shifts can be used later | |
3778 | * to extract the more random high bits from the multiplicative | |
3779 | * hash function before the remainder is taken. | |
3780 | */ | |
3781 | static inline unsigned long wait_table_bits(unsigned long size) | |
3782 | { | |
3783 | return ffz(~size); | |
3784 | } | |
3785 | ||
3786 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
3787 | ||
6d3163ce AH |
3788 | /* |
3789 | * Check if a pageblock contains reserved pages | |
3790 | */ | |
3791 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) | |
3792 | { | |
3793 | unsigned long pfn; | |
3794 | ||
3795 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
3796 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) | |
3797 | return 1; | |
3798 | } | |
3799 | return 0; | |
3800 | } | |
3801 | ||
56fd56b8 | 3802 | /* |
d9c23400 | 3803 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
41858966 MG |
3804 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
3805 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
56fd56b8 MG |
3806 | * higher will lead to a bigger reserve which will get freed as contiguous |
3807 | * blocks as reclaim kicks in | |
3808 | */ | |
3809 | static void setup_zone_migrate_reserve(struct zone *zone) | |
3810 | { | |
6d3163ce | 3811 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
56fd56b8 | 3812 | struct page *page; |
78986a67 MG |
3813 | unsigned long block_migratetype; |
3814 | int reserve; | |
56fd56b8 | 3815 | |
d0215638 MH |
3816 | /* |
3817 | * Get the start pfn, end pfn and the number of blocks to reserve | |
3818 | * We have to be careful to be aligned to pageblock_nr_pages to | |
3819 | * make sure that we always check pfn_valid for the first page in | |
3820 | * the block. | |
3821 | */ | |
56fd56b8 | 3822 | start_pfn = zone->zone_start_pfn; |
108bcc96 | 3823 | end_pfn = zone_end_pfn(zone); |
d0215638 | 3824 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
41858966 | 3825 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
d9c23400 | 3826 | pageblock_order; |
56fd56b8 | 3827 | |
78986a67 MG |
3828 | /* |
3829 | * Reserve blocks are generally in place to help high-order atomic | |
3830 | * allocations that are short-lived. A min_free_kbytes value that | |
3831 | * would result in more than 2 reserve blocks for atomic allocations | |
3832 | * is assumed to be in place to help anti-fragmentation for the | |
3833 | * future allocation of hugepages at runtime. | |
3834 | */ | |
3835 | reserve = min(2, reserve); | |
3836 | ||
d9c23400 | 3837 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
3838 | if (!pfn_valid(pfn)) |
3839 | continue; | |
3840 | page = pfn_to_page(pfn); | |
3841 | ||
344c790e AL |
3842 | /* Watch out for overlapping nodes */ |
3843 | if (page_to_nid(page) != zone_to_nid(zone)) | |
3844 | continue; | |
3845 | ||
56fd56b8 MG |
3846 | block_migratetype = get_pageblock_migratetype(page); |
3847 | ||
938929f1 MG |
3848 | /* Only test what is necessary when the reserves are not met */ |
3849 | if (reserve > 0) { | |
3850 | /* | |
3851 | * Blocks with reserved pages will never free, skip | |
3852 | * them. | |
3853 | */ | |
3854 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); | |
3855 | if (pageblock_is_reserved(pfn, block_end_pfn)) | |
3856 | continue; | |
56fd56b8 | 3857 | |
938929f1 MG |
3858 | /* If this block is reserved, account for it */ |
3859 | if (block_migratetype == MIGRATE_RESERVE) { | |
3860 | reserve--; | |
3861 | continue; | |
3862 | } | |
3863 | ||
3864 | /* Suitable for reserving if this block is movable */ | |
3865 | if (block_migratetype == MIGRATE_MOVABLE) { | |
3866 | set_pageblock_migratetype(page, | |
3867 | MIGRATE_RESERVE); | |
3868 | move_freepages_block(zone, page, | |
3869 | MIGRATE_RESERVE); | |
3870 | reserve--; | |
3871 | continue; | |
3872 | } | |
56fd56b8 MG |
3873 | } |
3874 | ||
3875 | /* | |
3876 | * If the reserve is met and this is a previous reserved block, | |
3877 | * take it back | |
3878 | */ | |
3879 | if (block_migratetype == MIGRATE_RESERVE) { | |
3880 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
3881 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
3882 | } | |
3883 | } | |
3884 | } | |
ac0e5b7a | 3885 | |
1da177e4 LT |
3886 | /* |
3887 | * Initially all pages are reserved - free ones are freed | |
3888 | * up by free_all_bootmem() once the early boot process is | |
3889 | * done. Non-atomic initialization, single-pass. | |
3890 | */ | |
c09b4240 | 3891 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 3892 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 3893 | { |
1da177e4 | 3894 | struct page *page; |
29751f69 AW |
3895 | unsigned long end_pfn = start_pfn + size; |
3896 | unsigned long pfn; | |
86051ca5 | 3897 | struct zone *z; |
1da177e4 | 3898 | |
22b31eec HD |
3899 | if (highest_memmap_pfn < end_pfn - 1) |
3900 | highest_memmap_pfn = end_pfn - 1; | |
3901 | ||
86051ca5 | 3902 | z = &NODE_DATA(nid)->node_zones[zone]; |
cbe8dd4a | 3903 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
3904 | /* |
3905 | * There can be holes in boot-time mem_map[]s | |
3906 | * handed to this function. They do not | |
3907 | * exist on hotplugged memory. | |
3908 | */ | |
3909 | if (context == MEMMAP_EARLY) { | |
3910 | if (!early_pfn_valid(pfn)) | |
3911 | continue; | |
3912 | if (!early_pfn_in_nid(pfn, nid)) | |
3913 | continue; | |
3914 | } | |
d41dee36 AW |
3915 | page = pfn_to_page(pfn); |
3916 | set_page_links(page, zone, nid, pfn); | |
708614e6 | 3917 | mminit_verify_page_links(page, zone, nid, pfn); |
7835e98b | 3918 | init_page_count(page); |
22b751c3 MG |
3919 | page_mapcount_reset(page); |
3920 | page_nid_reset_last(page); | |
1da177e4 | 3921 | SetPageReserved(page); |
b2a0ac88 MG |
3922 | /* |
3923 | * Mark the block movable so that blocks are reserved for | |
3924 | * movable at startup. This will force kernel allocations | |
3925 | * to reserve their blocks rather than leaking throughout | |
3926 | * the address space during boot when many long-lived | |
56fd56b8 MG |
3927 | * kernel allocations are made. Later some blocks near |
3928 | * the start are marked MIGRATE_RESERVE by | |
3929 | * setup_zone_migrate_reserve() | |
86051ca5 KH |
3930 | * |
3931 | * bitmap is created for zone's valid pfn range. but memmap | |
3932 | * can be created for invalid pages (for alignment) | |
3933 | * check here not to call set_pageblock_migratetype() against | |
3934 | * pfn out of zone. | |
b2a0ac88 | 3935 | */ |
86051ca5 | 3936 | if ((z->zone_start_pfn <= pfn) |
108bcc96 | 3937 | && (pfn < zone_end_pfn(z)) |
86051ca5 | 3938 | && !(pfn & (pageblock_nr_pages - 1))) |
56fd56b8 | 3939 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 3940 | |
1da177e4 LT |
3941 | INIT_LIST_HEAD(&page->lru); |
3942 | #ifdef WANT_PAGE_VIRTUAL | |
3943 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
3944 | if (!is_highmem_idx(zone)) | |
3212c6be | 3945 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 3946 | #endif |
1da177e4 LT |
3947 | } |
3948 | } | |
3949 | ||
1e548deb | 3950 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 3951 | { |
b2a0ac88 MG |
3952 | int order, t; |
3953 | for_each_migratetype_order(order, t) { | |
3954 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
3955 | zone->free_area[order].nr_free = 0; |
3956 | } | |
3957 | } | |
3958 | ||
3959 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
3960 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 3961 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
3962 | #endif |
3963 | ||
4ed7e022 | 3964 | static int __meminit zone_batchsize(struct zone *zone) |
e7c8d5c9 | 3965 | { |
3a6be87f | 3966 | #ifdef CONFIG_MMU |
e7c8d5c9 CL |
3967 | int batch; |
3968 | ||
3969 | /* | |
3970 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 3971 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
3972 | * |
3973 | * OK, so we don't know how big the cache is. So guess. | |
3974 | */ | |
b40da049 | 3975 | batch = zone->managed_pages / 1024; |
ba56e91c SR |
3976 | if (batch * PAGE_SIZE > 512 * 1024) |
3977 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
3978 | batch /= 4; /* We effectively *= 4 below */ |
3979 | if (batch < 1) | |
3980 | batch = 1; | |
3981 | ||
3982 | /* | |
0ceaacc9 NP |
3983 | * Clamp the batch to a 2^n - 1 value. Having a power |
3984 | * of 2 value was found to be more likely to have | |
3985 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 3986 | * |
0ceaacc9 NP |
3987 | * For example if 2 tasks are alternately allocating |
3988 | * batches of pages, one task can end up with a lot | |
3989 | * of pages of one half of the possible page colors | |
3990 | * and the other with pages of the other colors. | |
e7c8d5c9 | 3991 | */ |
9155203a | 3992 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
ba56e91c | 3993 | |
e7c8d5c9 | 3994 | return batch; |
3a6be87f DH |
3995 | |
3996 | #else | |
3997 | /* The deferral and batching of frees should be suppressed under NOMMU | |
3998 | * conditions. | |
3999 | * | |
4000 | * The problem is that NOMMU needs to be able to allocate large chunks | |
4001 | * of contiguous memory as there's no hardware page translation to | |
4002 | * assemble apparent contiguous memory from discontiguous pages. | |
4003 | * | |
4004 | * Queueing large contiguous runs of pages for batching, however, | |
4005 | * causes the pages to actually be freed in smaller chunks. As there | |
4006 | * can be a significant delay between the individual batches being | |
4007 | * recycled, this leads to the once large chunks of space being | |
4008 | * fragmented and becoming unavailable for high-order allocations. | |
4009 | */ | |
4010 | return 0; | |
4011 | #endif | |
e7c8d5c9 CL |
4012 | } |
4013 | ||
b69a7288 | 4014 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2caaad41 CL |
4015 | { |
4016 | struct per_cpu_pages *pcp; | |
5f8dcc21 | 4017 | int migratetype; |
2caaad41 | 4018 | |
1c6fe946 MD |
4019 | memset(p, 0, sizeof(*p)); |
4020 | ||
3dfa5721 | 4021 | pcp = &p->pcp; |
2caaad41 | 4022 | pcp->count = 0; |
2caaad41 CL |
4023 | pcp->high = 6 * batch; |
4024 | pcp->batch = max(1UL, 1 * batch); | |
5f8dcc21 MG |
4025 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
4026 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
2caaad41 CL |
4027 | } |
4028 | ||
8ad4b1fb RS |
4029 | /* |
4030 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
4031 | * to the value high for the pageset p. | |
4032 | */ | |
4033 | ||
4034 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
4035 | unsigned long high) | |
4036 | { | |
4037 | struct per_cpu_pages *pcp; | |
4038 | ||
3dfa5721 | 4039 | pcp = &p->pcp; |
8ad4b1fb RS |
4040 | pcp->high = high; |
4041 | pcp->batch = max(1UL, high/4); | |
4042 | if ((high/4) > (PAGE_SHIFT * 8)) | |
4043 | pcp->batch = PAGE_SHIFT * 8; | |
4044 | } | |
4045 | ||
4ed7e022 | 4046 | static void __meminit setup_zone_pageset(struct zone *zone) |
319774e2 WF |
4047 | { |
4048 | int cpu; | |
4049 | ||
4050 | zone->pageset = alloc_percpu(struct per_cpu_pageset); | |
4051 | ||
4052 | for_each_possible_cpu(cpu) { | |
4053 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
4054 | ||
4055 | setup_pageset(pcp, zone_batchsize(zone)); | |
4056 | ||
4057 | if (percpu_pagelist_fraction) | |
4058 | setup_pagelist_highmark(pcp, | |
b40da049 | 4059 | (zone->managed_pages / |
319774e2 WF |
4060 | percpu_pagelist_fraction)); |
4061 | } | |
4062 | } | |
4063 | ||
2caaad41 | 4064 | /* |
99dcc3e5 CL |
4065 | * Allocate per cpu pagesets and initialize them. |
4066 | * Before this call only boot pagesets were available. | |
e7c8d5c9 | 4067 | */ |
99dcc3e5 | 4068 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 | 4069 | { |
99dcc3e5 | 4070 | struct zone *zone; |
e7c8d5c9 | 4071 | |
319774e2 WF |
4072 | for_each_populated_zone(zone) |
4073 | setup_zone_pageset(zone); | |
e7c8d5c9 CL |
4074 | } |
4075 | ||
577a32f6 | 4076 | static noinline __init_refok |
cca448fe | 4077 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
4078 | { |
4079 | int i; | |
4080 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 4081 | size_t alloc_size; |
ed8ece2e DH |
4082 | |
4083 | /* | |
4084 | * The per-page waitqueue mechanism uses hashed waitqueues | |
4085 | * per zone. | |
4086 | */ | |
02b694de YG |
4087 | zone->wait_table_hash_nr_entries = |
4088 | wait_table_hash_nr_entries(zone_size_pages); | |
4089 | zone->wait_table_bits = | |
4090 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
4091 | alloc_size = zone->wait_table_hash_nr_entries |
4092 | * sizeof(wait_queue_head_t); | |
4093 | ||
cd94b9db | 4094 | if (!slab_is_available()) { |
cca448fe | 4095 | zone->wait_table = (wait_queue_head_t *) |
8f389a99 | 4096 | alloc_bootmem_node_nopanic(pgdat, alloc_size); |
cca448fe YG |
4097 | } else { |
4098 | /* | |
4099 | * This case means that a zone whose size was 0 gets new memory | |
4100 | * via memory hot-add. | |
4101 | * But it may be the case that a new node was hot-added. In | |
4102 | * this case vmalloc() will not be able to use this new node's | |
4103 | * memory - this wait_table must be initialized to use this new | |
4104 | * node itself as well. | |
4105 | * To use this new node's memory, further consideration will be | |
4106 | * necessary. | |
4107 | */ | |
8691f3a7 | 4108 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
4109 | } |
4110 | if (!zone->wait_table) | |
4111 | return -ENOMEM; | |
ed8ece2e | 4112 | |
02b694de | 4113 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 4114 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
4115 | |
4116 | return 0; | |
ed8ece2e DH |
4117 | } |
4118 | ||
c09b4240 | 4119 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e | 4120 | { |
99dcc3e5 CL |
4121 | /* |
4122 | * per cpu subsystem is not up at this point. The following code | |
4123 | * relies on the ability of the linker to provide the | |
4124 | * offset of a (static) per cpu variable into the per cpu area. | |
4125 | */ | |
4126 | zone->pageset = &boot_pageset; | |
ed8ece2e | 4127 | |
f5335c0f | 4128 | if (zone->present_pages) |
99dcc3e5 CL |
4129 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
4130 | zone->name, zone->present_pages, | |
4131 | zone_batchsize(zone)); | |
ed8ece2e DH |
4132 | } |
4133 | ||
4ed7e022 | 4134 | int __meminit init_currently_empty_zone(struct zone *zone, |
718127cc | 4135 | unsigned long zone_start_pfn, |
a2f3aa02 DH |
4136 | unsigned long size, |
4137 | enum memmap_context context) | |
ed8ece2e DH |
4138 | { |
4139 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
4140 | int ret; |
4141 | ret = zone_wait_table_init(zone, size); | |
4142 | if (ret) | |
4143 | return ret; | |
ed8ece2e DH |
4144 | pgdat->nr_zones = zone_idx(zone) + 1; |
4145 | ||
ed8ece2e DH |
4146 | zone->zone_start_pfn = zone_start_pfn; |
4147 | ||
708614e6 MG |
4148 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
4149 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
4150 | pgdat->node_id, | |
4151 | (unsigned long)zone_idx(zone), | |
4152 | zone_start_pfn, (zone_start_pfn + size)); | |
4153 | ||
1e548deb | 4154 | zone_init_free_lists(zone); |
718127cc YG |
4155 | |
4156 | return 0; | |
ed8ece2e DH |
4157 | } |
4158 | ||
0ee332c1 | 4159 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
4160 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
4161 | /* | |
4162 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
4163 | * Architectures may implement their own version but if add_active_range() | |
4164 | * was used and there are no special requirements, this is a convenient | |
4165 | * alternative | |
4166 | */ | |
f2dbcfa7 | 4167 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
c713216d | 4168 | { |
c13291a5 TH |
4169 | unsigned long start_pfn, end_pfn; |
4170 | int i, nid; | |
c713216d | 4171 | |
c13291a5 | 4172 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
c713216d | 4173 | if (start_pfn <= pfn && pfn < end_pfn) |
c13291a5 | 4174 | return nid; |
cc2559bc KH |
4175 | /* This is a memory hole */ |
4176 | return -1; | |
c713216d MG |
4177 | } |
4178 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
4179 | ||
f2dbcfa7 KH |
4180 | int __meminit early_pfn_to_nid(unsigned long pfn) |
4181 | { | |
cc2559bc KH |
4182 | int nid; |
4183 | ||
4184 | nid = __early_pfn_to_nid(pfn); | |
4185 | if (nid >= 0) | |
4186 | return nid; | |
4187 | /* just returns 0 */ | |
4188 | return 0; | |
f2dbcfa7 KH |
4189 | } |
4190 | ||
cc2559bc KH |
4191 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
4192 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
4193 | { | |
4194 | int nid; | |
4195 | ||
4196 | nid = __early_pfn_to_nid(pfn); | |
4197 | if (nid >= 0 && nid != node) | |
4198 | return false; | |
4199 | return true; | |
4200 | } | |
4201 | #endif | |
f2dbcfa7 | 4202 | |
c713216d MG |
4203 | /** |
4204 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
4205 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
4206 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
4207 | * |
4208 | * If an architecture guarantees that all ranges registered with | |
4209 | * add_active_ranges() contain no holes and may be freed, this | |
4210 | * this function may be used instead of calling free_bootmem() manually. | |
4211 | */ | |
c13291a5 | 4212 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
cc289894 | 4213 | { |
c13291a5 TH |
4214 | unsigned long start_pfn, end_pfn; |
4215 | int i, this_nid; | |
edbe7d23 | 4216 | |
c13291a5 TH |
4217 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
4218 | start_pfn = min(start_pfn, max_low_pfn); | |
4219 | end_pfn = min(end_pfn, max_low_pfn); | |
edbe7d23 | 4220 | |
c13291a5 TH |
4221 | if (start_pfn < end_pfn) |
4222 | free_bootmem_node(NODE_DATA(this_nid), | |
4223 | PFN_PHYS(start_pfn), | |
4224 | (end_pfn - start_pfn) << PAGE_SHIFT); | |
edbe7d23 | 4225 | } |
edbe7d23 | 4226 | } |
edbe7d23 | 4227 | |
c713216d MG |
4228 | /** |
4229 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 4230 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
4231 | * |
4232 | * If an architecture guarantees that all ranges registered with | |
4233 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 4234 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
4235 | */ |
4236 | void __init sparse_memory_present_with_active_regions(int nid) | |
4237 | { | |
c13291a5 TH |
4238 | unsigned long start_pfn, end_pfn; |
4239 | int i, this_nid; | |
c713216d | 4240 | |
c13291a5 TH |
4241 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
4242 | memory_present(this_nid, start_pfn, end_pfn); | |
c713216d MG |
4243 | } |
4244 | ||
4245 | /** | |
4246 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
4247 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
4248 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
4249 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
4250 | * |
4251 | * It returns the start and end page frame of a node based on information | |
4252 | * provided by an arch calling add_active_range(). If called for a node | |
4253 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 4254 | * PFNs will be 0. |
c713216d | 4255 | */ |
a3142c8e | 4256 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
4257 | unsigned long *start_pfn, unsigned long *end_pfn) |
4258 | { | |
c13291a5 | 4259 | unsigned long this_start_pfn, this_end_pfn; |
c713216d | 4260 | int i; |
c13291a5 | 4261 | |
c713216d MG |
4262 | *start_pfn = -1UL; |
4263 | *end_pfn = 0; | |
4264 | ||
c13291a5 TH |
4265 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
4266 | *start_pfn = min(*start_pfn, this_start_pfn); | |
4267 | *end_pfn = max(*end_pfn, this_end_pfn); | |
c713216d MG |
4268 | } |
4269 | ||
633c0666 | 4270 | if (*start_pfn == -1UL) |
c713216d | 4271 | *start_pfn = 0; |
c713216d MG |
4272 | } |
4273 | ||
2a1e274a MG |
4274 | /* |
4275 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
4276 | * assumption is made that zones within a node are ordered in monotonic | |
4277 | * increasing memory addresses so that the "highest" populated zone is used | |
4278 | */ | |
b69a7288 | 4279 | static void __init find_usable_zone_for_movable(void) |
2a1e274a MG |
4280 | { |
4281 | int zone_index; | |
4282 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
4283 | if (zone_index == ZONE_MOVABLE) | |
4284 | continue; | |
4285 | ||
4286 | if (arch_zone_highest_possible_pfn[zone_index] > | |
4287 | arch_zone_lowest_possible_pfn[zone_index]) | |
4288 | break; | |
4289 | } | |
4290 | ||
4291 | VM_BUG_ON(zone_index == -1); | |
4292 | movable_zone = zone_index; | |
4293 | } | |
4294 | ||
4295 | /* | |
4296 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
25985edc | 4297 | * because it is sized independent of architecture. Unlike the other zones, |
2a1e274a MG |
4298 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
4299 | * in each node depending on the size of each node and how evenly kernelcore | |
4300 | * is distributed. This helper function adjusts the zone ranges | |
4301 | * provided by the architecture for a given node by using the end of the | |
4302 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
4303 | * zones within a node are in order of monotonic increases memory addresses | |
4304 | */ | |
b69a7288 | 4305 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
2a1e274a MG |
4306 | unsigned long zone_type, |
4307 | unsigned long node_start_pfn, | |
4308 | unsigned long node_end_pfn, | |
4309 | unsigned long *zone_start_pfn, | |
4310 | unsigned long *zone_end_pfn) | |
4311 | { | |
4312 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
4313 | if (zone_movable_pfn[nid]) { | |
4314 | /* Size ZONE_MOVABLE */ | |
4315 | if (zone_type == ZONE_MOVABLE) { | |
4316 | *zone_start_pfn = zone_movable_pfn[nid]; | |
4317 | *zone_end_pfn = min(node_end_pfn, | |
4318 | arch_zone_highest_possible_pfn[movable_zone]); | |
4319 | ||
4320 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
4321 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
4322 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
4323 | *zone_end_pfn = zone_movable_pfn[nid]; | |
4324 | ||
4325 | /* Check if this whole range is within ZONE_MOVABLE */ | |
4326 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
4327 | *zone_start_pfn = *zone_end_pfn; | |
4328 | } | |
4329 | } | |
4330 | ||
c713216d MG |
4331 | /* |
4332 | * Return the number of pages a zone spans in a node, including holes | |
4333 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
4334 | */ | |
6ea6e688 | 4335 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4336 | unsigned long zone_type, |
4337 | unsigned long *ignored) | |
4338 | { | |
4339 | unsigned long node_start_pfn, node_end_pfn; | |
4340 | unsigned long zone_start_pfn, zone_end_pfn; | |
4341 | ||
4342 | /* Get the start and end of the node and zone */ | |
4343 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
4344 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
4345 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
4346 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4347 | node_start_pfn, node_end_pfn, | |
4348 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
4349 | |
4350 | /* Check that this node has pages within the zone's required range */ | |
4351 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
4352 | return 0; | |
4353 | ||
4354 | /* Move the zone boundaries inside the node if necessary */ | |
4355 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
4356 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
4357 | ||
4358 | /* Return the spanned pages */ | |
4359 | return zone_end_pfn - zone_start_pfn; | |
4360 | } | |
4361 | ||
4362 | /* | |
4363 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 4364 | * then all holes in the requested range will be accounted for. |
c713216d | 4365 | */ |
32996250 | 4366 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
4367 | unsigned long range_start_pfn, |
4368 | unsigned long range_end_pfn) | |
4369 | { | |
96e907d1 TH |
4370 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
4371 | unsigned long start_pfn, end_pfn; | |
4372 | int i; | |
c713216d | 4373 | |
96e907d1 TH |
4374 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
4375 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); | |
4376 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); | |
4377 | nr_absent -= end_pfn - start_pfn; | |
c713216d | 4378 | } |
96e907d1 | 4379 | return nr_absent; |
c713216d MG |
4380 | } |
4381 | ||
4382 | /** | |
4383 | * absent_pages_in_range - Return number of page frames in holes within a range | |
4384 | * @start_pfn: The start PFN to start searching for holes | |
4385 | * @end_pfn: The end PFN to stop searching for holes | |
4386 | * | |
88ca3b94 | 4387 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
4388 | */ |
4389 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
4390 | unsigned long end_pfn) | |
4391 | { | |
4392 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
4393 | } | |
4394 | ||
4395 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 4396 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4397 | unsigned long zone_type, |
4398 | unsigned long *ignored) | |
4399 | { | |
96e907d1 TH |
4400 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
4401 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; | |
9c7cd687 MG |
4402 | unsigned long node_start_pfn, node_end_pfn; |
4403 | unsigned long zone_start_pfn, zone_end_pfn; | |
4404 | ||
4405 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
96e907d1 TH |
4406 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
4407 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); | |
9c7cd687 | 4408 | |
2a1e274a MG |
4409 | adjust_zone_range_for_zone_movable(nid, zone_type, |
4410 | node_start_pfn, node_end_pfn, | |
4411 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 4412 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 4413 | } |
0e0b864e | 4414 | |
6981ec31 TC |
4415 | /** |
4416 | * sanitize_zone_movable_limit - Sanitize the zone_movable_limit array. | |
4417 | * | |
4418 | * zone_movable_limit is initialized as 0. This function will try to get | |
4419 | * the first ZONE_MOVABLE pfn of each node from movablemem_map, and | |
4420 | * assigne them to zone_movable_limit. | |
4421 | * zone_movable_limit[nid] == 0 means no limit for the node. | |
4422 | * | |
4423 | * Note: Each range is represented as [start_pfn, end_pfn) | |
4424 | */ | |
4425 | static void __meminit sanitize_zone_movable_limit(void) | |
4426 | { | |
4427 | int map_pos = 0, i, nid; | |
4428 | unsigned long start_pfn, end_pfn; | |
4429 | ||
4430 | if (!movablemem_map.nr_map) | |
4431 | return; | |
4432 | ||
4433 | /* Iterate all ranges from minimum to maximum */ | |
4434 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
4435 | /* | |
4436 | * If we have found lowest pfn of ZONE_MOVABLE of the node | |
4437 | * specified by user, just go on to check next range. | |
4438 | */ | |
4439 | if (zone_movable_limit[nid]) | |
4440 | continue; | |
4441 | ||
4442 | #ifdef CONFIG_ZONE_DMA | |
4443 | /* Skip DMA memory. */ | |
4444 | if (start_pfn < arch_zone_highest_possible_pfn[ZONE_DMA]) | |
4445 | start_pfn = arch_zone_highest_possible_pfn[ZONE_DMA]; | |
4446 | #endif | |
4447 | ||
4448 | #ifdef CONFIG_ZONE_DMA32 | |
4449 | /* Skip DMA32 memory. */ | |
4450 | if (start_pfn < arch_zone_highest_possible_pfn[ZONE_DMA32]) | |
4451 | start_pfn = arch_zone_highest_possible_pfn[ZONE_DMA32]; | |
4452 | #endif | |
4453 | ||
4454 | #ifdef CONFIG_HIGHMEM | |
4455 | /* Skip lowmem if ZONE_MOVABLE is highmem. */ | |
4456 | if (zone_movable_is_highmem() && | |
4457 | start_pfn < arch_zone_lowest_possible_pfn[ZONE_HIGHMEM]) | |
4458 | start_pfn = arch_zone_lowest_possible_pfn[ZONE_HIGHMEM]; | |
4459 | #endif | |
4460 | ||
4461 | if (start_pfn >= end_pfn) | |
4462 | continue; | |
4463 | ||
4464 | while (map_pos < movablemem_map.nr_map) { | |
4465 | if (end_pfn <= movablemem_map.map[map_pos].start_pfn) | |
4466 | break; | |
4467 | ||
4468 | if (start_pfn >= movablemem_map.map[map_pos].end_pfn) { | |
4469 | map_pos++; | |
4470 | continue; | |
4471 | } | |
4472 | ||
4473 | /* | |
4474 | * The start_pfn of ZONE_MOVABLE is either the minimum | |
4475 | * pfn specified by movablemem_map, or 0, which means | |
4476 | * the node has no ZONE_MOVABLE. | |
4477 | */ | |
4478 | zone_movable_limit[nid] = max(start_pfn, | |
4479 | movablemem_map.map[map_pos].start_pfn); | |
4480 | ||
4481 | break; | |
4482 | } | |
4483 | } | |
4484 | } | |
4485 | ||
0ee332c1 | 4486 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6ea6e688 | 4487 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
4488 | unsigned long zone_type, |
4489 | unsigned long *zones_size) | |
4490 | { | |
4491 | return zones_size[zone_type]; | |
4492 | } | |
4493 | ||
6ea6e688 | 4494 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
4495 | unsigned long zone_type, |
4496 | unsigned long *zholes_size) | |
4497 | { | |
4498 | if (!zholes_size) | |
4499 | return 0; | |
4500 | ||
4501 | return zholes_size[zone_type]; | |
4502 | } | |
0ee332c1 | 4503 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4504 | |
a3142c8e | 4505 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
4506 | unsigned long *zones_size, unsigned long *zholes_size) |
4507 | { | |
4508 | unsigned long realtotalpages, totalpages = 0; | |
4509 | enum zone_type i; | |
4510 | ||
4511 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4512 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
4513 | zones_size); | |
4514 | pgdat->node_spanned_pages = totalpages; | |
4515 | ||
4516 | realtotalpages = totalpages; | |
4517 | for (i = 0; i < MAX_NR_ZONES; i++) | |
4518 | realtotalpages -= | |
4519 | zone_absent_pages_in_node(pgdat->node_id, i, | |
4520 | zholes_size); | |
4521 | pgdat->node_present_pages = realtotalpages; | |
4522 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
4523 | realtotalpages); | |
4524 | } | |
4525 | ||
835c134e MG |
4526 | #ifndef CONFIG_SPARSEMEM |
4527 | /* | |
4528 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
4529 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
4530 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
4531 | * round what is now in bits to nearest long in bits, then return it in |
4532 | * bytes. | |
4533 | */ | |
7c45512d | 4534 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
835c134e MG |
4535 | { |
4536 | unsigned long usemapsize; | |
4537 | ||
7c45512d | 4538 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
d9c23400 MG |
4539 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
4540 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
4541 | usemapsize *= NR_PAGEBLOCK_BITS; |
4542 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
4543 | ||
4544 | return usemapsize / 8; | |
4545 | } | |
4546 | ||
4547 | static void __init setup_usemap(struct pglist_data *pgdat, | |
7c45512d LT |
4548 | struct zone *zone, |
4549 | unsigned long zone_start_pfn, | |
4550 | unsigned long zonesize) | |
835c134e | 4551 | { |
7c45512d | 4552 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
835c134e | 4553 | zone->pageblock_flags = NULL; |
58a01a45 | 4554 | if (usemapsize) |
8f389a99 YL |
4555 | zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat, |
4556 | usemapsize); | |
835c134e MG |
4557 | } |
4558 | #else | |
7c45512d LT |
4559 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
4560 | unsigned long zone_start_pfn, unsigned long zonesize) {} | |
835c134e MG |
4561 | #endif /* CONFIG_SPARSEMEM */ |
4562 | ||
d9c23400 | 4563 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c | 4564 | |
d9c23400 | 4565 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
ca57df79 | 4566 | void __init set_pageblock_order(void) |
d9c23400 | 4567 | { |
955c1cd7 AM |
4568 | unsigned int order; |
4569 | ||
d9c23400 MG |
4570 | /* Check that pageblock_nr_pages has not already been setup */ |
4571 | if (pageblock_order) | |
4572 | return; | |
4573 | ||
955c1cd7 AM |
4574 | if (HPAGE_SHIFT > PAGE_SHIFT) |
4575 | order = HUGETLB_PAGE_ORDER; | |
4576 | else | |
4577 | order = MAX_ORDER - 1; | |
4578 | ||
d9c23400 MG |
4579 | /* |
4580 | * Assume the largest contiguous order of interest is a huge page. | |
955c1cd7 AM |
4581 | * This value may be variable depending on boot parameters on IA64 and |
4582 | * powerpc. | |
d9c23400 MG |
4583 | */ |
4584 | pageblock_order = order; | |
4585 | } | |
4586 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4587 | ||
ba72cb8c MG |
4588 | /* |
4589 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
955c1cd7 AM |
4590 | * is unused as pageblock_order is set at compile-time. See |
4591 | * include/linux/pageblock-flags.h for the values of pageblock_order based on | |
4592 | * the kernel config | |
ba72cb8c | 4593 | */ |
ca57df79 | 4594 | void __init set_pageblock_order(void) |
ba72cb8c | 4595 | { |
ba72cb8c | 4596 | } |
d9c23400 MG |
4597 | |
4598 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
4599 | ||
01cefaef JL |
4600 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
4601 | unsigned long present_pages) | |
4602 | { | |
4603 | unsigned long pages = spanned_pages; | |
4604 | ||
4605 | /* | |
4606 | * Provide a more accurate estimation if there are holes within | |
4607 | * the zone and SPARSEMEM is in use. If there are holes within the | |
4608 | * zone, each populated memory region may cost us one or two extra | |
4609 | * memmap pages due to alignment because memmap pages for each | |
4610 | * populated regions may not naturally algined on page boundary. | |
4611 | * So the (present_pages >> 4) heuristic is a tradeoff for that. | |
4612 | */ | |
4613 | if (spanned_pages > present_pages + (present_pages >> 4) && | |
4614 | IS_ENABLED(CONFIG_SPARSEMEM)) | |
4615 | pages = present_pages; | |
4616 | ||
4617 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; | |
4618 | } | |
4619 | ||
1da177e4 LT |
4620 | /* |
4621 | * Set up the zone data structures: | |
4622 | * - mark all pages reserved | |
4623 | * - mark all memory queues empty | |
4624 | * - clear the memory bitmaps | |
6527af5d MK |
4625 | * |
4626 | * NOTE: pgdat should get zeroed by caller. | |
1da177e4 | 4627 | */ |
b5a0e011 | 4628 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
4629 | unsigned long *zones_size, unsigned long *zholes_size) |
4630 | { | |
2f1b6248 | 4631 | enum zone_type j; |
ed8ece2e | 4632 | int nid = pgdat->node_id; |
1da177e4 | 4633 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 4634 | int ret; |
1da177e4 | 4635 | |
208d54e5 | 4636 | pgdat_resize_init(pgdat); |
8177a420 AA |
4637 | #ifdef CONFIG_NUMA_BALANCING |
4638 | spin_lock_init(&pgdat->numabalancing_migrate_lock); | |
4639 | pgdat->numabalancing_migrate_nr_pages = 0; | |
4640 | pgdat->numabalancing_migrate_next_window = jiffies; | |
4641 | #endif | |
1da177e4 | 4642 | init_waitqueue_head(&pgdat->kswapd_wait); |
5515061d | 4643 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
52d4b9ac | 4644 | pgdat_page_cgroup_init(pgdat); |
5f63b720 | 4645 | |
1da177e4 LT |
4646 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4647 | struct zone *zone = pgdat->node_zones + j; | |
9feedc9d | 4648 | unsigned long size, realsize, freesize, memmap_pages; |
1da177e4 | 4649 | |
c713216d | 4650 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
9feedc9d | 4651 | realsize = freesize = size - zone_absent_pages_in_node(nid, j, |
c713216d | 4652 | zholes_size); |
1da177e4 | 4653 | |
0e0b864e | 4654 | /* |
9feedc9d | 4655 | * Adjust freesize so that it accounts for how much memory |
0e0b864e MG |
4656 | * is used by this zone for memmap. This affects the watermark |
4657 | * and per-cpu initialisations | |
4658 | */ | |
01cefaef | 4659 | memmap_pages = calc_memmap_size(size, realsize); |
9feedc9d JL |
4660 | if (freesize >= memmap_pages) { |
4661 | freesize -= memmap_pages; | |
5594c8c8 YL |
4662 | if (memmap_pages) |
4663 | printk(KERN_DEBUG | |
4664 | " %s zone: %lu pages used for memmap\n", | |
4665 | zone_names[j], memmap_pages); | |
0e0b864e MG |
4666 | } else |
4667 | printk(KERN_WARNING | |
9feedc9d JL |
4668 | " %s zone: %lu pages exceeds freesize %lu\n", |
4669 | zone_names[j], memmap_pages, freesize); | |
0e0b864e | 4670 | |
6267276f | 4671 | /* Account for reserved pages */ |
9feedc9d JL |
4672 | if (j == 0 && freesize > dma_reserve) { |
4673 | freesize -= dma_reserve; | |
d903ef9f | 4674 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6267276f | 4675 | zone_names[0], dma_reserve); |
0e0b864e MG |
4676 | } |
4677 | ||
98d2b0eb | 4678 | if (!is_highmem_idx(j)) |
9feedc9d | 4679 | nr_kernel_pages += freesize; |
01cefaef JL |
4680 | /* Charge for highmem memmap if there are enough kernel pages */ |
4681 | else if (nr_kernel_pages > memmap_pages * 2) | |
4682 | nr_kernel_pages -= memmap_pages; | |
9feedc9d | 4683 | nr_all_pages += freesize; |
1da177e4 LT |
4684 | |
4685 | zone->spanned_pages = size; | |
306f2e9e | 4686 | zone->present_pages = realsize; |
9feedc9d JL |
4687 | /* |
4688 | * Set an approximate value for lowmem here, it will be adjusted | |
4689 | * when the bootmem allocator frees pages into the buddy system. | |
4690 | * And all highmem pages will be managed by the buddy system. | |
4691 | */ | |
4692 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; | |
9614634f | 4693 | #ifdef CONFIG_NUMA |
d5f541ed | 4694 | zone->node = nid; |
9feedc9d | 4695 | zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio) |
9614634f | 4696 | / 100; |
9feedc9d | 4697 | zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100; |
9614634f | 4698 | #endif |
1da177e4 LT |
4699 | zone->name = zone_names[j]; |
4700 | spin_lock_init(&zone->lock); | |
4701 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 4702 | zone_seqlock_init(zone); |
1da177e4 | 4703 | zone->zone_pgdat = pgdat; |
1da177e4 | 4704 | |
ed8ece2e | 4705 | zone_pcp_init(zone); |
bea8c150 | 4706 | lruvec_init(&zone->lruvec); |
1da177e4 LT |
4707 | if (!size) |
4708 | continue; | |
4709 | ||
955c1cd7 | 4710 | set_pageblock_order(); |
7c45512d | 4711 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
a2f3aa02 DH |
4712 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
4713 | size, MEMMAP_EARLY); | |
718127cc | 4714 | BUG_ON(ret); |
76cdd58e | 4715 | memmap_init(size, nid, j, zone_start_pfn); |
1da177e4 | 4716 | zone_start_pfn += size; |
1da177e4 LT |
4717 | } |
4718 | } | |
4719 | ||
577a32f6 | 4720 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 4721 | { |
1da177e4 LT |
4722 | /* Skip empty nodes */ |
4723 | if (!pgdat->node_spanned_pages) | |
4724 | return; | |
4725 | ||
d41dee36 | 4726 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
4727 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
4728 | if (!pgdat->node_mem_map) { | |
e984bb43 | 4729 | unsigned long size, start, end; |
d41dee36 AW |
4730 | struct page *map; |
4731 | ||
e984bb43 BP |
4732 | /* |
4733 | * The zone's endpoints aren't required to be MAX_ORDER | |
4734 | * aligned but the node_mem_map endpoints must be in order | |
4735 | * for the buddy allocator to function correctly. | |
4736 | */ | |
4737 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
108bcc96 | 4738 | end = pgdat_end_pfn(pgdat); |
e984bb43 BP |
4739 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
4740 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
4741 | map = alloc_remap(pgdat->node_id, size); |
4742 | if (!map) | |
8f389a99 | 4743 | map = alloc_bootmem_node_nopanic(pgdat, size); |
e984bb43 | 4744 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 4745 | } |
12d810c1 | 4746 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
4747 | /* |
4748 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
4749 | */ | |
c713216d | 4750 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 4751 | mem_map = NODE_DATA(0)->node_mem_map; |
0ee332c1 | 4752 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d | 4753 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
467bc461 | 4754 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
0ee332c1 | 4755 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 4756 | } |
1da177e4 | 4757 | #endif |
d41dee36 | 4758 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
4759 | } |
4760 | ||
9109fb7b JW |
4761 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
4762 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
1da177e4 | 4763 | { |
9109fb7b JW |
4764 | pg_data_t *pgdat = NODE_DATA(nid); |
4765 | ||
88fdf75d | 4766 | /* pg_data_t should be reset to zero when it's allocated */ |
8783b6e2 | 4767 | WARN_ON(pgdat->nr_zones || pgdat->classzone_idx); |
88fdf75d | 4768 | |
1da177e4 LT |
4769 | pgdat->node_id = nid; |
4770 | pgdat->node_start_pfn = node_start_pfn; | |
957f822a | 4771 | init_zone_allows_reclaim(nid); |
c713216d | 4772 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
4773 | |
4774 | alloc_node_mem_map(pgdat); | |
e8c27ac9 YL |
4775 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
4776 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
4777 | nid, (unsigned long)pgdat, | |
4778 | (unsigned long)pgdat->node_mem_map); | |
4779 | #endif | |
1da177e4 LT |
4780 | |
4781 | free_area_init_core(pgdat, zones_size, zholes_size); | |
4782 | } | |
4783 | ||
0ee332c1 | 4784 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
418508c1 MS |
4785 | |
4786 | #if MAX_NUMNODES > 1 | |
4787 | /* | |
4788 | * Figure out the number of possible node ids. | |
4789 | */ | |
4790 | static void __init setup_nr_node_ids(void) | |
4791 | { | |
4792 | unsigned int node; | |
4793 | unsigned int highest = 0; | |
4794 | ||
4795 | for_each_node_mask(node, node_possible_map) | |
4796 | highest = node; | |
4797 | nr_node_ids = highest + 1; | |
4798 | } | |
4799 | #else | |
4800 | static inline void setup_nr_node_ids(void) | |
4801 | { | |
4802 | } | |
4803 | #endif | |
4804 | ||
1e01979c TH |
4805 | /** |
4806 | * node_map_pfn_alignment - determine the maximum internode alignment | |
4807 | * | |
4808 | * This function should be called after node map is populated and sorted. | |
4809 | * It calculates the maximum power of two alignment which can distinguish | |
4810 | * all the nodes. | |
4811 | * | |
4812 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value | |
4813 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the | |
4814 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is | |
4815 | * shifted, 1GiB is enough and this function will indicate so. | |
4816 | * | |
4817 | * This is used to test whether pfn -> nid mapping of the chosen memory | |
4818 | * model has fine enough granularity to avoid incorrect mapping for the | |
4819 | * populated node map. | |
4820 | * | |
4821 | * Returns the determined alignment in pfn's. 0 if there is no alignment | |
4822 | * requirement (single node). | |
4823 | */ | |
4824 | unsigned long __init node_map_pfn_alignment(void) | |
4825 | { | |
4826 | unsigned long accl_mask = 0, last_end = 0; | |
c13291a5 | 4827 | unsigned long start, end, mask; |
1e01979c | 4828 | int last_nid = -1; |
c13291a5 | 4829 | int i, nid; |
1e01979c | 4830 | |
c13291a5 | 4831 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
1e01979c TH |
4832 | if (!start || last_nid < 0 || last_nid == nid) { |
4833 | last_nid = nid; | |
4834 | last_end = end; | |
4835 | continue; | |
4836 | } | |
4837 | ||
4838 | /* | |
4839 | * Start with a mask granular enough to pin-point to the | |
4840 | * start pfn and tick off bits one-by-one until it becomes | |
4841 | * too coarse to separate the current node from the last. | |
4842 | */ | |
4843 | mask = ~((1 << __ffs(start)) - 1); | |
4844 | while (mask && last_end <= (start & (mask << 1))) | |
4845 | mask <<= 1; | |
4846 | ||
4847 | /* accumulate all internode masks */ | |
4848 | accl_mask |= mask; | |
4849 | } | |
4850 | ||
4851 | /* convert mask to number of pages */ | |
4852 | return ~accl_mask + 1; | |
4853 | } | |
4854 | ||
a6af2bc3 | 4855 | /* Find the lowest pfn for a node */ |
b69a7288 | 4856 | static unsigned long __init find_min_pfn_for_node(int nid) |
c713216d | 4857 | { |
a6af2bc3 | 4858 | unsigned long min_pfn = ULONG_MAX; |
c13291a5 TH |
4859 | unsigned long start_pfn; |
4860 | int i; | |
1abbfb41 | 4861 | |
c13291a5 TH |
4862 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
4863 | min_pfn = min(min_pfn, start_pfn); | |
c713216d | 4864 | |
a6af2bc3 MG |
4865 | if (min_pfn == ULONG_MAX) { |
4866 | printk(KERN_WARNING | |
2bc0d261 | 4867 | "Could not find start_pfn for node %d\n", nid); |
a6af2bc3 MG |
4868 | return 0; |
4869 | } | |
4870 | ||
4871 | return min_pfn; | |
c713216d MG |
4872 | } |
4873 | ||
4874 | /** | |
4875 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4876 | * | |
4877 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 4878 | * add_active_range(). |
c713216d MG |
4879 | */ |
4880 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4881 | { | |
4882 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4883 | } | |
4884 | ||
37b07e41 LS |
4885 | /* |
4886 | * early_calculate_totalpages() | |
4887 | * Sum pages in active regions for movable zone. | |
4b0ef1fe | 4888 | * Populate N_MEMORY for calculating usable_nodes. |
37b07e41 | 4889 | */ |
484f51f8 | 4890 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef | 4891 | { |
7e63efef | 4892 | unsigned long totalpages = 0; |
c13291a5 TH |
4893 | unsigned long start_pfn, end_pfn; |
4894 | int i, nid; | |
4895 | ||
4896 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { | |
4897 | unsigned long pages = end_pfn - start_pfn; | |
7e63efef | 4898 | |
37b07e41 LS |
4899 | totalpages += pages; |
4900 | if (pages) | |
4b0ef1fe | 4901 | node_set_state(nid, N_MEMORY); |
37b07e41 LS |
4902 | } |
4903 | return totalpages; | |
7e63efef MG |
4904 | } |
4905 | ||
2a1e274a MG |
4906 | /* |
4907 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4908 | * is spread evenly between nodes as long as the nodes have enough | |
4909 | * memory. When they don't, some nodes will have more kernelcore than | |
4910 | * others | |
4911 | */ | |
b224ef85 | 4912 | static void __init find_zone_movable_pfns_for_nodes(void) |
2a1e274a MG |
4913 | { |
4914 | int i, nid; | |
4915 | unsigned long usable_startpfn; | |
4916 | unsigned long kernelcore_node, kernelcore_remaining; | |
66918dcd | 4917 | /* save the state before borrow the nodemask */ |
4b0ef1fe | 4918 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
37b07e41 | 4919 | unsigned long totalpages = early_calculate_totalpages(); |
4b0ef1fe | 4920 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
2a1e274a | 4921 | |
7e63efef MG |
4922 | /* |
4923 | * If movablecore was specified, calculate what size of | |
4924 | * kernelcore that corresponds so that memory usable for | |
4925 | * any allocation type is evenly spread. If both kernelcore | |
4926 | * and movablecore are specified, then the value of kernelcore | |
4927 | * will be used for required_kernelcore if it's greater than | |
4928 | * what movablecore would have allowed. | |
4929 | */ | |
4930 | if (required_movablecore) { | |
7e63efef MG |
4931 | unsigned long corepages; |
4932 | ||
4933 | /* | |
4934 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4935 | * was requested by the user | |
4936 | */ | |
4937 | required_movablecore = | |
4938 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4939 | corepages = totalpages - required_movablecore; | |
4940 | ||
4941 | required_kernelcore = max(required_kernelcore, corepages); | |
4942 | } | |
4943 | ||
42f47e27 TC |
4944 | /* |
4945 | * If neither kernelcore/movablecore nor movablemem_map is specified, | |
4946 | * there is no ZONE_MOVABLE. But if movablemem_map is specified, the | |
4947 | * start pfn of ZONE_MOVABLE has been stored in zone_movable_limit[]. | |
4948 | */ | |
4949 | if (!required_kernelcore) { | |
4950 | if (movablemem_map.nr_map) | |
4951 | memcpy(zone_movable_pfn, zone_movable_limit, | |
4952 | sizeof(zone_movable_pfn)); | |
66918dcd | 4953 | goto out; |
42f47e27 | 4954 | } |
2a1e274a MG |
4955 | |
4956 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
2a1e274a MG |
4957 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
4958 | ||
4959 | restart: | |
4960 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4961 | kernelcore_node = required_kernelcore / usable_nodes; | |
4b0ef1fe | 4962 | for_each_node_state(nid, N_MEMORY) { |
c13291a5 TH |
4963 | unsigned long start_pfn, end_pfn; |
4964 | ||
2a1e274a MG |
4965 | /* |
4966 | * Recalculate kernelcore_node if the division per node | |
4967 | * now exceeds what is necessary to satisfy the requested | |
4968 | * amount of memory for the kernel | |
4969 | */ | |
4970 | if (required_kernelcore < kernelcore_node) | |
4971 | kernelcore_node = required_kernelcore / usable_nodes; | |
4972 | ||
4973 | /* | |
4974 | * As the map is walked, we track how much memory is usable | |
4975 | * by the kernel using kernelcore_remaining. When it is | |
4976 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4977 | */ | |
4978 | kernelcore_remaining = kernelcore_node; | |
4979 | ||
4980 | /* Go through each range of PFNs within this node */ | |
c13291a5 | 4981 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
2a1e274a MG |
4982 | unsigned long size_pages; |
4983 | ||
42f47e27 TC |
4984 | /* |
4985 | * Find more memory for kernelcore in | |
4986 | * [zone_movable_pfn[nid], zone_movable_limit[nid]). | |
4987 | */ | |
c13291a5 | 4988 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
2a1e274a MG |
4989 | if (start_pfn >= end_pfn) |
4990 | continue; | |
4991 | ||
42f47e27 TC |
4992 | if (zone_movable_limit[nid]) { |
4993 | end_pfn = min(end_pfn, zone_movable_limit[nid]); | |
4994 | /* No range left for kernelcore in this node */ | |
4995 | if (start_pfn >= end_pfn) { | |
4996 | zone_movable_pfn[nid] = | |
4997 | zone_movable_limit[nid]; | |
4998 | break; | |
4999 | } | |
5000 | } | |
5001 | ||
2a1e274a MG |
5002 | /* Account for what is only usable for kernelcore */ |
5003 | if (start_pfn < usable_startpfn) { | |
5004 | unsigned long kernel_pages; | |
5005 | kernel_pages = min(end_pfn, usable_startpfn) | |
5006 | - start_pfn; | |
5007 | ||
5008 | kernelcore_remaining -= min(kernel_pages, | |
5009 | kernelcore_remaining); | |
5010 | required_kernelcore -= min(kernel_pages, | |
5011 | required_kernelcore); | |
5012 | ||
5013 | /* Continue if range is now fully accounted */ | |
5014 | if (end_pfn <= usable_startpfn) { | |
5015 | ||
5016 | /* | |
5017 | * Push zone_movable_pfn to the end so | |
5018 | * that if we have to rebalance | |
5019 | * kernelcore across nodes, we will | |
5020 | * not double account here | |
5021 | */ | |
5022 | zone_movable_pfn[nid] = end_pfn; | |
5023 | continue; | |
5024 | } | |
5025 | start_pfn = usable_startpfn; | |
5026 | } | |
5027 | ||
5028 | /* | |
5029 | * The usable PFN range for ZONE_MOVABLE is from | |
5030 | * start_pfn->end_pfn. Calculate size_pages as the | |
5031 | * number of pages used as kernelcore | |
5032 | */ | |
5033 | size_pages = end_pfn - start_pfn; | |
5034 | if (size_pages > kernelcore_remaining) | |
5035 | size_pages = kernelcore_remaining; | |
5036 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
5037 | ||
5038 | /* | |
5039 | * Some kernelcore has been met, update counts and | |
5040 | * break if the kernelcore for this node has been | |
5041 | * satisified | |
5042 | */ | |
5043 | required_kernelcore -= min(required_kernelcore, | |
5044 | size_pages); | |
5045 | kernelcore_remaining -= size_pages; | |
5046 | if (!kernelcore_remaining) | |
5047 | break; | |
5048 | } | |
5049 | } | |
5050 | ||
5051 | /* | |
5052 | * If there is still required_kernelcore, we do another pass with one | |
5053 | * less node in the count. This will push zone_movable_pfn[nid] further | |
5054 | * along on the nodes that still have memory until kernelcore is | |
5055 | * satisified | |
5056 | */ | |
5057 | usable_nodes--; | |
5058 | if (usable_nodes && required_kernelcore > usable_nodes) | |
5059 | goto restart; | |
5060 | ||
42f47e27 | 5061 | out: |
2a1e274a MG |
5062 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
5063 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
5064 | zone_movable_pfn[nid] = | |
5065 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
66918dcd | 5066 | |
66918dcd | 5067 | /* restore the node_state */ |
4b0ef1fe | 5068 | node_states[N_MEMORY] = saved_node_state; |
2a1e274a MG |
5069 | } |
5070 | ||
4b0ef1fe LJ |
5071 | /* Any regular or high memory on that node ? */ |
5072 | static void check_for_memory(pg_data_t *pgdat, int nid) | |
37b07e41 | 5073 | { |
37b07e41 LS |
5074 | enum zone_type zone_type; |
5075 | ||
4b0ef1fe LJ |
5076 | if (N_MEMORY == N_NORMAL_MEMORY) |
5077 | return; | |
5078 | ||
5079 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { | |
37b07e41 | 5080 | struct zone *zone = &pgdat->node_zones[zone_type]; |
d0048b0e | 5081 | if (zone->present_pages) { |
4b0ef1fe LJ |
5082 | node_set_state(nid, N_HIGH_MEMORY); |
5083 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && | |
5084 | zone_type <= ZONE_NORMAL) | |
5085 | node_set_state(nid, N_NORMAL_MEMORY); | |
d0048b0e BL |
5086 | break; |
5087 | } | |
37b07e41 | 5088 | } |
37b07e41 LS |
5089 | } |
5090 | ||
c713216d MG |
5091 | /** |
5092 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 5093 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
5094 | * |
5095 | * This will call free_area_init_node() for each active node in the system. | |
5096 | * Using the page ranges provided by add_active_range(), the size of each | |
5097 | * zone in each node and their holes is calculated. If the maximum PFN | |
5098 | * between two adjacent zones match, it is assumed that the zone is empty. | |
5099 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
5100 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
5101 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
5102 | * at arch_max_dma_pfn. | |
5103 | */ | |
5104 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
5105 | { | |
c13291a5 TH |
5106 | unsigned long start_pfn, end_pfn; |
5107 | int i, nid; | |
a6af2bc3 | 5108 | |
c713216d MG |
5109 | /* Record where the zone boundaries are */ |
5110 | memset(arch_zone_lowest_possible_pfn, 0, | |
5111 | sizeof(arch_zone_lowest_possible_pfn)); | |
5112 | memset(arch_zone_highest_possible_pfn, 0, | |
5113 | sizeof(arch_zone_highest_possible_pfn)); | |
5114 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
5115 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
5116 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
5117 | if (i == ZONE_MOVABLE) |
5118 | continue; | |
c713216d MG |
5119 | arch_zone_lowest_possible_pfn[i] = |
5120 | arch_zone_highest_possible_pfn[i-1]; | |
5121 | arch_zone_highest_possible_pfn[i] = | |
5122 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
5123 | } | |
2a1e274a MG |
5124 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
5125 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
5126 | ||
5127 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
5128 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
6981ec31 TC |
5129 | find_usable_zone_for_movable(); |
5130 | sanitize_zone_movable_limit(); | |
b224ef85 | 5131 | find_zone_movable_pfns_for_nodes(); |
c713216d | 5132 | |
c713216d | 5133 | /* Print out the zone ranges */ |
a62e2f4f | 5134 | printk("Zone ranges:\n"); |
2a1e274a MG |
5135 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5136 | if (i == ZONE_MOVABLE) | |
5137 | continue; | |
155cbfc8 | 5138 | printk(KERN_CONT " %-8s ", zone_names[i]); |
72f0ba02 DR |
5139 | if (arch_zone_lowest_possible_pfn[i] == |
5140 | arch_zone_highest_possible_pfn[i]) | |
155cbfc8 | 5141 | printk(KERN_CONT "empty\n"); |
72f0ba02 | 5142 | else |
a62e2f4f BH |
5143 | printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n", |
5144 | arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT, | |
5145 | (arch_zone_highest_possible_pfn[i] | |
5146 | << PAGE_SHIFT) - 1); | |
2a1e274a MG |
5147 | } |
5148 | ||
5149 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
a62e2f4f | 5150 | printk("Movable zone start for each node\n"); |
2a1e274a MG |
5151 | for (i = 0; i < MAX_NUMNODES; i++) { |
5152 | if (zone_movable_pfn[i]) | |
a62e2f4f BH |
5153 | printk(" Node %d: %#010lx\n", i, |
5154 | zone_movable_pfn[i] << PAGE_SHIFT); | |
2a1e274a | 5155 | } |
c713216d | 5156 | |
f2d52fe5 | 5157 | /* Print out the early node map */ |
a62e2f4f | 5158 | printk("Early memory node ranges\n"); |
c13291a5 | 5159 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
a62e2f4f BH |
5160 | printk(" node %3d: [mem %#010lx-%#010lx]\n", nid, |
5161 | start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1); | |
c713216d MG |
5162 | |
5163 | /* Initialise every node */ | |
708614e6 | 5164 | mminit_verify_pageflags_layout(); |
8ef82866 | 5165 | setup_nr_node_ids(); |
c713216d MG |
5166 | for_each_online_node(nid) { |
5167 | pg_data_t *pgdat = NODE_DATA(nid); | |
9109fb7b | 5168 | free_area_init_node(nid, NULL, |
c713216d | 5169 | find_min_pfn_for_node(nid), NULL); |
37b07e41 LS |
5170 | |
5171 | /* Any memory on that node */ | |
5172 | if (pgdat->node_present_pages) | |
4b0ef1fe LJ |
5173 | node_set_state(nid, N_MEMORY); |
5174 | check_for_memory(pgdat, nid); | |
c713216d MG |
5175 | } |
5176 | } | |
2a1e274a | 5177 | |
7e63efef | 5178 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
5179 | { |
5180 | unsigned long long coremem; | |
5181 | if (!p) | |
5182 | return -EINVAL; | |
5183 | ||
5184 | coremem = memparse(p, &p); | |
7e63efef | 5185 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 5186 | |
7e63efef | 5187 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
5188 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
5189 | ||
5190 | return 0; | |
5191 | } | |
ed7ed365 | 5192 | |
7e63efef MG |
5193 | /* |
5194 | * kernelcore=size sets the amount of memory for use for allocations that | |
5195 | * cannot be reclaimed or migrated. | |
5196 | */ | |
5197 | static int __init cmdline_parse_kernelcore(char *p) | |
5198 | { | |
5199 | return cmdline_parse_core(p, &required_kernelcore); | |
5200 | } | |
5201 | ||
5202 | /* | |
5203 | * movablecore=size sets the amount of memory for use for allocations that | |
5204 | * can be reclaimed or migrated. | |
5205 | */ | |
5206 | static int __init cmdline_parse_movablecore(char *p) | |
5207 | { | |
5208 | return cmdline_parse_core(p, &required_movablecore); | |
5209 | } | |
5210 | ||
ed7ed365 | 5211 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 5212 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 5213 | |
27168d38 TC |
5214 | /** |
5215 | * movablemem_map_overlap() - Check if a range overlaps movablemem_map.map[]. | |
5216 | * @start_pfn: start pfn of the range to be checked | |
5217 | * @end_pfn: end pfn of the range to be checked (exclusive) | |
5218 | * | |
5219 | * This function checks if a given memory range [start_pfn, end_pfn) overlaps | |
5220 | * the movablemem_map.map[] array. | |
5221 | * | |
5222 | * Return: index of the first overlapped element in movablemem_map.map[] | |
5223 | * or -1 if they don't overlap each other. | |
5224 | */ | |
5225 | int __init movablemem_map_overlap(unsigned long start_pfn, | |
5226 | unsigned long end_pfn) | |
5227 | { | |
5228 | int overlap; | |
5229 | ||
5230 | if (!movablemem_map.nr_map) | |
5231 | return -1; | |
5232 | ||
5233 | for (overlap = 0; overlap < movablemem_map.nr_map; overlap++) | |
5234 | if (start_pfn < movablemem_map.map[overlap].end_pfn) | |
5235 | break; | |
5236 | ||
5237 | if (overlap == movablemem_map.nr_map || | |
5238 | end_pfn <= movablemem_map.map[overlap].start_pfn) | |
5239 | return -1; | |
5240 | ||
5241 | return overlap; | |
5242 | } | |
5243 | ||
34b71f1e TC |
5244 | /** |
5245 | * insert_movablemem_map - Insert a memory range in to movablemem_map.map. | |
5246 | * @start_pfn: start pfn of the range | |
5247 | * @end_pfn: end pfn of the range | |
5248 | * | |
5249 | * This function will also merge the overlapped ranges, and sort the array | |
5250 | * by start_pfn in monotonic increasing order. | |
5251 | */ | |
27168d38 TC |
5252 | void __init insert_movablemem_map(unsigned long start_pfn, |
5253 | unsigned long end_pfn) | |
34b71f1e TC |
5254 | { |
5255 | int pos, overlap; | |
5256 | ||
5257 | /* | |
5258 | * pos will be at the 1st overlapped range, or the position | |
5259 | * where the element should be inserted. | |
5260 | */ | |
5261 | for (pos = 0; pos < movablemem_map.nr_map; pos++) | |
5262 | if (start_pfn <= movablemem_map.map[pos].end_pfn) | |
5263 | break; | |
5264 | ||
5265 | /* If there is no overlapped range, just insert the element. */ | |
5266 | if (pos == movablemem_map.nr_map || | |
5267 | end_pfn < movablemem_map.map[pos].start_pfn) { | |
5268 | /* | |
5269 | * If pos is not the end of array, we need to move all | |
5270 | * the rest elements backward. | |
5271 | */ | |
5272 | if (pos < movablemem_map.nr_map) | |
5273 | memmove(&movablemem_map.map[pos+1], | |
5274 | &movablemem_map.map[pos], | |
5275 | sizeof(struct movablemem_entry) * | |
5276 | (movablemem_map.nr_map - pos)); | |
5277 | movablemem_map.map[pos].start_pfn = start_pfn; | |
5278 | movablemem_map.map[pos].end_pfn = end_pfn; | |
5279 | movablemem_map.nr_map++; | |
5280 | return; | |
5281 | } | |
5282 | ||
5283 | /* overlap will be at the last overlapped range */ | |
5284 | for (overlap = pos + 1; overlap < movablemem_map.nr_map; overlap++) | |
5285 | if (end_pfn < movablemem_map.map[overlap].start_pfn) | |
5286 | break; | |
5287 | ||
5288 | /* | |
5289 | * If there are more ranges overlapped, we need to merge them, | |
5290 | * and move the rest elements forward. | |
5291 | */ | |
5292 | overlap--; | |
5293 | movablemem_map.map[pos].start_pfn = min(start_pfn, | |
5294 | movablemem_map.map[pos].start_pfn); | |
5295 | movablemem_map.map[pos].end_pfn = max(end_pfn, | |
5296 | movablemem_map.map[overlap].end_pfn); | |
5297 | ||
5298 | if (pos != overlap && overlap + 1 != movablemem_map.nr_map) | |
5299 | memmove(&movablemem_map.map[pos+1], | |
5300 | &movablemem_map.map[overlap+1], | |
5301 | sizeof(struct movablemem_entry) * | |
5302 | (movablemem_map.nr_map - overlap - 1)); | |
5303 | ||
5304 | movablemem_map.nr_map -= overlap - pos; | |
5305 | } | |
5306 | ||
5307 | /** | |
5308 | * movablemem_map_add_region - Add a memory range into movablemem_map. | |
5309 | * @start: physical start address of range | |
5310 | * @end: physical end address of range | |
5311 | * | |
5312 | * This function transform the physical address into pfn, and then add the | |
5313 | * range into movablemem_map by calling insert_movablemem_map(). | |
5314 | */ | |
5315 | static void __init movablemem_map_add_region(u64 start, u64 size) | |
5316 | { | |
5317 | unsigned long start_pfn, end_pfn; | |
5318 | ||
5319 | /* In case size == 0 or start + size overflows */ | |
5320 | if (start + size <= start) | |
5321 | return; | |
5322 | ||
5323 | if (movablemem_map.nr_map >= ARRAY_SIZE(movablemem_map.map)) { | |
5324 | pr_err("movablemem_map: too many entries;" | |
5325 | " ignoring [mem %#010llx-%#010llx]\n", | |
5326 | (unsigned long long) start, | |
5327 | (unsigned long long) (start + size - 1)); | |
5328 | return; | |
5329 | } | |
5330 | ||
5331 | start_pfn = PFN_DOWN(start); | |
5332 | end_pfn = PFN_UP(start + size); | |
5333 | insert_movablemem_map(start_pfn, end_pfn); | |
5334 | } | |
5335 | ||
5336 | /* | |
5337 | * cmdline_parse_movablemem_map - Parse boot option movablemem_map. | |
5338 | * @p: The boot option of the following format: | |
5339 | * movablemem_map=nn[KMG]@ss[KMG] | |
5340 | * | |
5341 | * This option sets the memory range [ss, ss+nn) to be used as movable memory. | |
5342 | * | |
5343 | * Return: 0 on success or -EINVAL on failure. | |
5344 | */ | |
5345 | static int __init cmdline_parse_movablemem_map(char *p) | |
5346 | { | |
5347 | char *oldp; | |
5348 | u64 start_at, mem_size; | |
5349 | ||
5350 | if (!p) | |
5351 | goto err; | |
5352 | ||
01a178a9 TC |
5353 | if (!strcmp(p, "acpi")) |
5354 | movablemem_map.acpi = true; | |
5355 | ||
5356 | /* | |
5357 | * If user decide to use info from BIOS, all the other user specified | |
5358 | * ranges will be ingored. | |
5359 | */ | |
5360 | if (movablemem_map.acpi) { | |
5361 | if (movablemem_map.nr_map) { | |
5362 | memset(movablemem_map.map, 0, | |
5363 | sizeof(struct movablemem_entry) | |
5364 | * movablemem_map.nr_map); | |
5365 | movablemem_map.nr_map = 0; | |
5366 | } | |
5367 | return 0; | |
5368 | } | |
5369 | ||
34b71f1e TC |
5370 | oldp = p; |
5371 | mem_size = memparse(p, &p); | |
5372 | if (p == oldp) | |
5373 | goto err; | |
5374 | ||
5375 | if (*p == '@') { | |
5376 | oldp = ++p; | |
5377 | start_at = memparse(p, &p); | |
5378 | if (p == oldp || *p != '\0') | |
5379 | goto err; | |
5380 | ||
5381 | movablemem_map_add_region(start_at, mem_size); | |
5382 | return 0; | |
5383 | } | |
5384 | err: | |
5385 | return -EINVAL; | |
5386 | } | |
5387 | early_param("movablemem_map", cmdline_parse_movablemem_map); | |
5388 | ||
0ee332c1 | 5389 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
c713216d | 5390 | |
0e0b864e | 5391 | /** |
88ca3b94 RD |
5392 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
5393 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
5394 | * |
5395 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
5396 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
5397 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
5398 | * function may optionally be used to account for unfreeable pages in the |
5399 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
5400 | * smaller per-cpu batchsize. | |
0e0b864e MG |
5401 | */ |
5402 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
5403 | { | |
5404 | dma_reserve = new_dma_reserve; | |
5405 | } | |
5406 | ||
1da177e4 LT |
5407 | void __init free_area_init(unsigned long *zones_size) |
5408 | { | |
9109fb7b | 5409 | free_area_init_node(0, zones_size, |
1da177e4 LT |
5410 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
5411 | } | |
1da177e4 | 5412 | |
1da177e4 LT |
5413 | static int page_alloc_cpu_notify(struct notifier_block *self, |
5414 | unsigned long action, void *hcpu) | |
5415 | { | |
5416 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 5417 | |
8bb78442 | 5418 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
f0cb3c76 | 5419 | lru_add_drain_cpu(cpu); |
9f8f2172 CL |
5420 | drain_pages(cpu); |
5421 | ||
5422 | /* | |
5423 | * Spill the event counters of the dead processor | |
5424 | * into the current processors event counters. | |
5425 | * This artificially elevates the count of the current | |
5426 | * processor. | |
5427 | */ | |
f8891e5e | 5428 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
5429 | |
5430 | /* | |
5431 | * Zero the differential counters of the dead processor | |
5432 | * so that the vm statistics are consistent. | |
5433 | * | |
5434 | * This is only okay since the processor is dead and cannot | |
5435 | * race with what we are doing. | |
5436 | */ | |
2244b95a | 5437 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
5438 | } |
5439 | return NOTIFY_OK; | |
5440 | } | |
1da177e4 LT |
5441 | |
5442 | void __init page_alloc_init(void) | |
5443 | { | |
5444 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
5445 | } | |
5446 | ||
cb45b0e9 HA |
5447 | /* |
5448 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
5449 | * or min_free_kbytes changes. | |
5450 | */ | |
5451 | static void calculate_totalreserve_pages(void) | |
5452 | { | |
5453 | struct pglist_data *pgdat; | |
5454 | unsigned long reserve_pages = 0; | |
2f6726e5 | 5455 | enum zone_type i, j; |
cb45b0e9 HA |
5456 | |
5457 | for_each_online_pgdat(pgdat) { | |
5458 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
5459 | struct zone *zone = pgdat->node_zones + i; | |
5460 | unsigned long max = 0; | |
5461 | ||
5462 | /* Find valid and maximum lowmem_reserve in the zone */ | |
5463 | for (j = i; j < MAX_NR_ZONES; j++) { | |
5464 | if (zone->lowmem_reserve[j] > max) | |
5465 | max = zone->lowmem_reserve[j]; | |
5466 | } | |
5467 | ||
41858966 MG |
5468 | /* we treat the high watermark as reserved pages. */ |
5469 | max += high_wmark_pages(zone); | |
cb45b0e9 | 5470 | |
b40da049 JL |
5471 | if (max > zone->managed_pages) |
5472 | max = zone->managed_pages; | |
cb45b0e9 | 5473 | reserve_pages += max; |
ab8fabd4 JW |
5474 | /* |
5475 | * Lowmem reserves are not available to | |
5476 | * GFP_HIGHUSER page cache allocations and | |
5477 | * kswapd tries to balance zones to their high | |
5478 | * watermark. As a result, neither should be | |
5479 | * regarded as dirtyable memory, to prevent a | |
5480 | * situation where reclaim has to clean pages | |
5481 | * in order to balance the zones. | |
5482 | */ | |
5483 | zone->dirty_balance_reserve = max; | |
cb45b0e9 HA |
5484 | } |
5485 | } | |
ab8fabd4 | 5486 | dirty_balance_reserve = reserve_pages; |
cb45b0e9 HA |
5487 | totalreserve_pages = reserve_pages; |
5488 | } | |
5489 | ||
1da177e4 LT |
5490 | /* |
5491 | * setup_per_zone_lowmem_reserve - called whenever | |
5492 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
5493 | * has a correct pages reserved value, so an adequate number of | |
5494 | * pages are left in the zone after a successful __alloc_pages(). | |
5495 | */ | |
5496 | static void setup_per_zone_lowmem_reserve(void) | |
5497 | { | |
5498 | struct pglist_data *pgdat; | |
2f6726e5 | 5499 | enum zone_type j, idx; |
1da177e4 | 5500 | |
ec936fc5 | 5501 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
5502 | for (j = 0; j < MAX_NR_ZONES; j++) { |
5503 | struct zone *zone = pgdat->node_zones + j; | |
b40da049 | 5504 | unsigned long managed_pages = zone->managed_pages; |
1da177e4 LT |
5505 | |
5506 | zone->lowmem_reserve[j] = 0; | |
5507 | ||
2f6726e5 CL |
5508 | idx = j; |
5509 | while (idx) { | |
1da177e4 LT |
5510 | struct zone *lower_zone; |
5511 | ||
2f6726e5 CL |
5512 | idx--; |
5513 | ||
1da177e4 LT |
5514 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
5515 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
5516 | ||
5517 | lower_zone = pgdat->node_zones + idx; | |
b40da049 | 5518 | lower_zone->lowmem_reserve[j] = managed_pages / |
1da177e4 | 5519 | sysctl_lowmem_reserve_ratio[idx]; |
b40da049 | 5520 | managed_pages += lower_zone->managed_pages; |
1da177e4 LT |
5521 | } |
5522 | } | |
5523 | } | |
cb45b0e9 HA |
5524 | |
5525 | /* update totalreserve_pages */ | |
5526 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5527 | } |
5528 | ||
cfd3da1e | 5529 | static void __setup_per_zone_wmarks(void) |
1da177e4 LT |
5530 | { |
5531 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
5532 | unsigned long lowmem_pages = 0; | |
5533 | struct zone *zone; | |
5534 | unsigned long flags; | |
5535 | ||
5536 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
5537 | for_each_zone(zone) { | |
5538 | if (!is_highmem(zone)) | |
b40da049 | 5539 | lowmem_pages += zone->managed_pages; |
1da177e4 LT |
5540 | } |
5541 | ||
5542 | for_each_zone(zone) { | |
ac924c60 AM |
5543 | u64 tmp; |
5544 | ||
1125b4e3 | 5545 | spin_lock_irqsave(&zone->lock, flags); |
b40da049 | 5546 | tmp = (u64)pages_min * zone->managed_pages; |
ac924c60 | 5547 | do_div(tmp, lowmem_pages); |
1da177e4 LT |
5548 | if (is_highmem(zone)) { |
5549 | /* | |
669ed175 NP |
5550 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
5551 | * need highmem pages, so cap pages_min to a small | |
5552 | * value here. | |
5553 | * | |
41858966 | 5554 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
669ed175 NP |
5555 | * deltas controls asynch page reclaim, and so should |
5556 | * not be capped for highmem. | |
1da177e4 | 5557 | */ |
90ae8d67 | 5558 | unsigned long min_pages; |
1da177e4 | 5559 | |
b40da049 | 5560 | min_pages = zone->managed_pages / 1024; |
90ae8d67 | 5561 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
41858966 | 5562 | zone->watermark[WMARK_MIN] = min_pages; |
1da177e4 | 5563 | } else { |
669ed175 NP |
5564 | /* |
5565 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
5566 | * proportionate to the zone's size. |
5567 | */ | |
41858966 | 5568 | zone->watermark[WMARK_MIN] = tmp; |
1da177e4 LT |
5569 | } |
5570 | ||
41858966 MG |
5571 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
5572 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
49f223a9 | 5573 | |
56fd56b8 | 5574 | setup_zone_migrate_reserve(zone); |
1125b4e3 | 5575 | spin_unlock_irqrestore(&zone->lock, flags); |
1da177e4 | 5576 | } |
cb45b0e9 HA |
5577 | |
5578 | /* update totalreserve_pages */ | |
5579 | calculate_totalreserve_pages(); | |
1da177e4 LT |
5580 | } |
5581 | ||
cfd3da1e MG |
5582 | /** |
5583 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
5584 | * or when memory is hot-{added|removed} | |
5585 | * | |
5586 | * Ensures that the watermark[min,low,high] values for each zone are set | |
5587 | * correctly with respect to min_free_kbytes. | |
5588 | */ | |
5589 | void setup_per_zone_wmarks(void) | |
5590 | { | |
5591 | mutex_lock(&zonelists_mutex); | |
5592 | __setup_per_zone_wmarks(); | |
5593 | mutex_unlock(&zonelists_mutex); | |
5594 | } | |
5595 | ||
55a4462a | 5596 | /* |
556adecb RR |
5597 | * The inactive anon list should be small enough that the VM never has to |
5598 | * do too much work, but large enough that each inactive page has a chance | |
5599 | * to be referenced again before it is swapped out. | |
5600 | * | |
5601 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
5602 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
5603 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
5604 | * the anonymous pages are kept on the inactive list. | |
5605 | * | |
5606 | * total target max | |
5607 | * memory ratio inactive anon | |
5608 | * ------------------------------------- | |
5609 | * 10MB 1 5MB | |
5610 | * 100MB 1 50MB | |
5611 | * 1GB 3 250MB | |
5612 | * 10GB 10 0.9GB | |
5613 | * 100GB 31 3GB | |
5614 | * 1TB 101 10GB | |
5615 | * 10TB 320 32GB | |
5616 | */ | |
1b79acc9 | 5617 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
556adecb | 5618 | { |
96cb4df5 | 5619 | unsigned int gb, ratio; |
556adecb | 5620 | |
96cb4df5 | 5621 | /* Zone size in gigabytes */ |
b40da049 | 5622 | gb = zone->managed_pages >> (30 - PAGE_SHIFT); |
96cb4df5 | 5623 | if (gb) |
556adecb | 5624 | ratio = int_sqrt(10 * gb); |
96cb4df5 MK |
5625 | else |
5626 | ratio = 1; | |
556adecb | 5627 | |
96cb4df5 MK |
5628 | zone->inactive_ratio = ratio; |
5629 | } | |
556adecb | 5630 | |
839a4fcc | 5631 | static void __meminit setup_per_zone_inactive_ratio(void) |
96cb4df5 MK |
5632 | { |
5633 | struct zone *zone; | |
5634 | ||
5635 | for_each_zone(zone) | |
5636 | calculate_zone_inactive_ratio(zone); | |
556adecb RR |
5637 | } |
5638 | ||
1da177e4 LT |
5639 | /* |
5640 | * Initialise min_free_kbytes. | |
5641 | * | |
5642 | * For small machines we want it small (128k min). For large machines | |
5643 | * we want it large (64MB max). But it is not linear, because network | |
5644 | * bandwidth does not increase linearly with machine size. We use | |
5645 | * | |
5646 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
5647 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
5648 | * | |
5649 | * which yields | |
5650 | * | |
5651 | * 16MB: 512k | |
5652 | * 32MB: 724k | |
5653 | * 64MB: 1024k | |
5654 | * 128MB: 1448k | |
5655 | * 256MB: 2048k | |
5656 | * 512MB: 2896k | |
5657 | * 1024MB: 4096k | |
5658 | * 2048MB: 5792k | |
5659 | * 4096MB: 8192k | |
5660 | * 8192MB: 11584k | |
5661 | * 16384MB: 16384k | |
5662 | */ | |
1b79acc9 | 5663 | int __meminit init_per_zone_wmark_min(void) |
1da177e4 LT |
5664 | { |
5665 | unsigned long lowmem_kbytes; | |
5666 | ||
5667 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
5668 | ||
5669 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
5670 | if (min_free_kbytes < 128) | |
5671 | min_free_kbytes = 128; | |
5672 | if (min_free_kbytes > 65536) | |
5673 | min_free_kbytes = 65536; | |
bc75d33f | 5674 | setup_per_zone_wmarks(); |
a6cccdc3 | 5675 | refresh_zone_stat_thresholds(); |
1da177e4 | 5676 | setup_per_zone_lowmem_reserve(); |
556adecb | 5677 | setup_per_zone_inactive_ratio(); |
1da177e4 LT |
5678 | return 0; |
5679 | } | |
bc75d33f | 5680 | module_init(init_per_zone_wmark_min) |
1da177e4 LT |
5681 | |
5682 | /* | |
5683 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
5684 | * that we can call two helper functions whenever min_free_kbytes | |
5685 | * changes. | |
5686 | */ | |
5687 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5688 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5689 | { |
8d65af78 | 5690 | proc_dointvec(table, write, buffer, length, ppos); |
3b1d92c5 | 5691 | if (write) |
bc75d33f | 5692 | setup_per_zone_wmarks(); |
1da177e4 LT |
5693 | return 0; |
5694 | } | |
5695 | ||
9614634f CL |
5696 | #ifdef CONFIG_NUMA |
5697 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5698 | void __user *buffer, size_t *length, loff_t *ppos) |
9614634f CL |
5699 | { |
5700 | struct zone *zone; | |
5701 | int rc; | |
5702 | ||
8d65af78 | 5703 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
9614634f CL |
5704 | if (rc) |
5705 | return rc; | |
5706 | ||
5707 | for_each_zone(zone) | |
b40da049 | 5708 | zone->min_unmapped_pages = (zone->managed_pages * |
9614634f CL |
5709 | sysctl_min_unmapped_ratio) / 100; |
5710 | return 0; | |
5711 | } | |
0ff38490 CL |
5712 | |
5713 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5714 | void __user *buffer, size_t *length, loff_t *ppos) |
0ff38490 CL |
5715 | { |
5716 | struct zone *zone; | |
5717 | int rc; | |
5718 | ||
8d65af78 | 5719 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
0ff38490 CL |
5720 | if (rc) |
5721 | return rc; | |
5722 | ||
5723 | for_each_zone(zone) | |
b40da049 | 5724 | zone->min_slab_pages = (zone->managed_pages * |
0ff38490 CL |
5725 | sysctl_min_slab_ratio) / 100; |
5726 | return 0; | |
5727 | } | |
9614634f CL |
5728 | #endif |
5729 | ||
1da177e4 LT |
5730 | /* |
5731 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
5732 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
5733 | * whenever sysctl_lowmem_reserve_ratio changes. | |
5734 | * | |
5735 | * The reserve ratio obviously has absolutely no relation with the | |
41858966 | 5736 | * minimum watermarks. The lowmem reserve ratio can only make sense |
1da177e4 LT |
5737 | * if in function of the boot time zone sizes. |
5738 | */ | |
5739 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5740 | void __user *buffer, size_t *length, loff_t *ppos) |
1da177e4 | 5741 | { |
8d65af78 | 5742 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
1da177e4 LT |
5743 | setup_per_zone_lowmem_reserve(); |
5744 | return 0; | |
5745 | } | |
5746 | ||
8ad4b1fb RS |
5747 | /* |
5748 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
5749 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
5750 | * can have before it gets flushed back to buddy allocator. | |
5751 | */ | |
5752 | ||
5753 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
8d65af78 | 5754 | void __user *buffer, size_t *length, loff_t *ppos) |
8ad4b1fb RS |
5755 | { |
5756 | struct zone *zone; | |
5757 | unsigned int cpu; | |
5758 | int ret; | |
5759 | ||
8d65af78 | 5760 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
93278814 | 5761 | if (!write || (ret < 0)) |
8ad4b1fb | 5762 | return ret; |
364df0eb | 5763 | for_each_populated_zone(zone) { |
99dcc3e5 | 5764 | for_each_possible_cpu(cpu) { |
8ad4b1fb | 5765 | unsigned long high; |
b40da049 | 5766 | high = zone->managed_pages / percpu_pagelist_fraction; |
99dcc3e5 CL |
5767 | setup_pagelist_highmark( |
5768 | per_cpu_ptr(zone->pageset, cpu), high); | |
8ad4b1fb RS |
5769 | } |
5770 | } | |
5771 | return 0; | |
5772 | } | |
5773 | ||
f034b5d4 | 5774 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
5775 | |
5776 | #ifdef CONFIG_NUMA | |
5777 | static int __init set_hashdist(char *str) | |
5778 | { | |
5779 | if (!str) | |
5780 | return 0; | |
5781 | hashdist = simple_strtoul(str, &str, 0); | |
5782 | return 1; | |
5783 | } | |
5784 | __setup("hashdist=", set_hashdist); | |
5785 | #endif | |
5786 | ||
5787 | /* | |
5788 | * allocate a large system hash table from bootmem | |
5789 | * - it is assumed that the hash table must contain an exact power-of-2 | |
5790 | * quantity of entries | |
5791 | * - limit is the number of hash buckets, not the total allocation size | |
5792 | */ | |
5793 | void *__init alloc_large_system_hash(const char *tablename, | |
5794 | unsigned long bucketsize, | |
5795 | unsigned long numentries, | |
5796 | int scale, | |
5797 | int flags, | |
5798 | unsigned int *_hash_shift, | |
5799 | unsigned int *_hash_mask, | |
31fe62b9 TB |
5800 | unsigned long low_limit, |
5801 | unsigned long high_limit) | |
1da177e4 | 5802 | { |
31fe62b9 | 5803 | unsigned long long max = high_limit; |
1da177e4 LT |
5804 | unsigned long log2qty, size; |
5805 | void *table = NULL; | |
5806 | ||
5807 | /* allow the kernel cmdline to have a say */ | |
5808 | if (!numentries) { | |
5809 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 5810 | numentries = nr_kernel_pages; |
1da177e4 LT |
5811 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
5812 | numentries >>= 20 - PAGE_SHIFT; | |
5813 | numentries <<= 20 - PAGE_SHIFT; | |
5814 | ||
5815 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
5816 | if (scale > PAGE_SHIFT) | |
5817 | numentries >>= (scale - PAGE_SHIFT); | |
5818 | else | |
5819 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
5820 | |
5821 | /* Make sure we've got at least a 0-order allocation.. */ | |
2c85f51d JB |
5822 | if (unlikely(flags & HASH_SMALL)) { |
5823 | /* Makes no sense without HASH_EARLY */ | |
5824 | WARN_ON(!(flags & HASH_EARLY)); | |
5825 | if (!(numentries >> *_hash_shift)) { | |
5826 | numentries = 1UL << *_hash_shift; | |
5827 | BUG_ON(!numentries); | |
5828 | } | |
5829 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
9ab37b8f | 5830 | numentries = PAGE_SIZE / bucketsize; |
1da177e4 | 5831 | } |
6e692ed3 | 5832 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
5833 | |
5834 | /* limit allocation size to 1/16 total memory by default */ | |
5835 | if (max == 0) { | |
5836 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
5837 | do_div(max, bucketsize); | |
5838 | } | |
074b8517 | 5839 | max = min(max, 0x80000000ULL); |
1da177e4 | 5840 | |
31fe62b9 TB |
5841 | if (numentries < low_limit) |
5842 | numentries = low_limit; | |
1da177e4 LT |
5843 | if (numentries > max) |
5844 | numentries = max; | |
5845 | ||
f0d1b0b3 | 5846 | log2qty = ilog2(numentries); |
1da177e4 LT |
5847 | |
5848 | do { | |
5849 | size = bucketsize << log2qty; | |
5850 | if (flags & HASH_EARLY) | |
74768ed8 | 5851 | table = alloc_bootmem_nopanic(size); |
1da177e4 LT |
5852 | else if (hashdist) |
5853 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
5854 | else { | |
1037b83b ED |
5855 | /* |
5856 | * If bucketsize is not a power-of-two, we may free | |
a1dd268c MG |
5857 | * some pages at the end of hash table which |
5858 | * alloc_pages_exact() automatically does | |
1037b83b | 5859 | */ |
264ef8a9 | 5860 | if (get_order(size) < MAX_ORDER) { |
a1dd268c | 5861 | table = alloc_pages_exact(size, GFP_ATOMIC); |
264ef8a9 CM |
5862 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
5863 | } | |
1da177e4 LT |
5864 | } |
5865 | } while (!table && size > PAGE_SIZE && --log2qty); | |
5866 | ||
5867 | if (!table) | |
5868 | panic("Failed to allocate %s hash table\n", tablename); | |
5869 | ||
f241e660 | 5870 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
1da177e4 | 5871 | tablename, |
f241e660 | 5872 | (1UL << log2qty), |
f0d1b0b3 | 5873 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
5874 | size); |
5875 | ||
5876 | if (_hash_shift) | |
5877 | *_hash_shift = log2qty; | |
5878 | if (_hash_mask) | |
5879 | *_hash_mask = (1 << log2qty) - 1; | |
5880 | ||
5881 | return table; | |
5882 | } | |
a117e66e | 5883 | |
835c134e MG |
5884 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
5885 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
5886 | unsigned long pfn) | |
5887 | { | |
5888 | #ifdef CONFIG_SPARSEMEM | |
5889 | return __pfn_to_section(pfn)->pageblock_flags; | |
5890 | #else | |
5891 | return zone->pageblock_flags; | |
5892 | #endif /* CONFIG_SPARSEMEM */ | |
5893 | } | |
5894 | ||
5895 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
5896 | { | |
5897 | #ifdef CONFIG_SPARSEMEM | |
5898 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 5899 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e | 5900 | #else |
c060f943 | 5901 | pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages); |
d9c23400 | 5902 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
5903 | #endif /* CONFIG_SPARSEMEM */ |
5904 | } | |
5905 | ||
5906 | /** | |
d9c23400 | 5907 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
5908 | * @page: The page within the block of interest |
5909 | * @start_bitidx: The first bit of interest to retrieve | |
5910 | * @end_bitidx: The last bit of interest | |
5911 | * returns pageblock_bits flags | |
5912 | */ | |
5913 | unsigned long get_pageblock_flags_group(struct page *page, | |
5914 | int start_bitidx, int end_bitidx) | |
5915 | { | |
5916 | struct zone *zone; | |
5917 | unsigned long *bitmap; | |
5918 | unsigned long pfn, bitidx; | |
5919 | unsigned long flags = 0; | |
5920 | unsigned long value = 1; | |
5921 | ||
5922 | zone = page_zone(page); | |
5923 | pfn = page_to_pfn(page); | |
5924 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5925 | bitidx = pfn_to_bitidx(zone, pfn); | |
5926 | ||
5927 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5928 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
5929 | flags |= value; | |
6220ec78 | 5930 | |
835c134e MG |
5931 | return flags; |
5932 | } | |
5933 | ||
5934 | /** | |
d9c23400 | 5935 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
5936 | * @page: The page within the block of interest |
5937 | * @start_bitidx: The first bit of interest | |
5938 | * @end_bitidx: The last bit of interest | |
5939 | * @flags: The flags to set | |
5940 | */ | |
5941 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
5942 | int start_bitidx, int end_bitidx) | |
5943 | { | |
5944 | struct zone *zone; | |
5945 | unsigned long *bitmap; | |
5946 | unsigned long pfn, bitidx; | |
5947 | unsigned long value = 1; | |
5948 | ||
5949 | zone = page_zone(page); | |
5950 | pfn = page_to_pfn(page); | |
5951 | bitmap = get_pageblock_bitmap(zone, pfn); | |
5952 | bitidx = pfn_to_bitidx(zone, pfn); | |
108bcc96 | 5953 | VM_BUG_ON(!zone_spans_pfn(zone, pfn)); |
835c134e MG |
5954 | |
5955 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
5956 | if (flags & value) | |
5957 | __set_bit(bitidx + start_bitidx, bitmap); | |
5958 | else | |
5959 | __clear_bit(bitidx + start_bitidx, bitmap); | |
5960 | } | |
a5d76b54 KH |
5961 | |
5962 | /* | |
80934513 MK |
5963 | * This function checks whether pageblock includes unmovable pages or not. |
5964 | * If @count is not zero, it is okay to include less @count unmovable pages | |
5965 | * | |
5966 | * PageLRU check wihtout isolation or lru_lock could race so that | |
5967 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't | |
5968 | * expect this function should be exact. | |
a5d76b54 | 5969 | */ |
b023f468 WC |
5970 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
5971 | bool skip_hwpoisoned_pages) | |
49ac8255 KH |
5972 | { |
5973 | unsigned long pfn, iter, found; | |
47118af0 MN |
5974 | int mt; |
5975 | ||
49ac8255 KH |
5976 | /* |
5977 | * For avoiding noise data, lru_add_drain_all() should be called | |
80934513 | 5978 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
49ac8255 KH |
5979 | */ |
5980 | if (zone_idx(zone) == ZONE_MOVABLE) | |
80934513 | 5981 | return false; |
47118af0 MN |
5982 | mt = get_pageblock_migratetype(page); |
5983 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) | |
80934513 | 5984 | return false; |
49ac8255 KH |
5985 | |
5986 | pfn = page_to_pfn(page); | |
5987 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { | |
5988 | unsigned long check = pfn + iter; | |
5989 | ||
29723fcc | 5990 | if (!pfn_valid_within(check)) |
49ac8255 | 5991 | continue; |
29723fcc | 5992 | |
49ac8255 | 5993 | page = pfn_to_page(check); |
97d255c8 MK |
5994 | /* |
5995 | * We can't use page_count without pin a page | |
5996 | * because another CPU can free compound page. | |
5997 | * This check already skips compound tails of THP | |
5998 | * because their page->_count is zero at all time. | |
5999 | */ | |
6000 | if (!atomic_read(&page->_count)) { | |
49ac8255 KH |
6001 | if (PageBuddy(page)) |
6002 | iter += (1 << page_order(page)) - 1; | |
6003 | continue; | |
6004 | } | |
97d255c8 | 6005 | |
b023f468 WC |
6006 | /* |
6007 | * The HWPoisoned page may be not in buddy system, and | |
6008 | * page_count() is not 0. | |
6009 | */ | |
6010 | if (skip_hwpoisoned_pages && PageHWPoison(page)) | |
6011 | continue; | |
6012 | ||
49ac8255 KH |
6013 | if (!PageLRU(page)) |
6014 | found++; | |
6015 | /* | |
6016 | * If there are RECLAIMABLE pages, we need to check it. | |
6017 | * But now, memory offline itself doesn't call shrink_slab() | |
6018 | * and it still to be fixed. | |
6019 | */ | |
6020 | /* | |
6021 | * If the page is not RAM, page_count()should be 0. | |
6022 | * we don't need more check. This is an _used_ not-movable page. | |
6023 | * | |
6024 | * The problematic thing here is PG_reserved pages. PG_reserved | |
6025 | * is set to both of a memory hole page and a _used_ kernel | |
6026 | * page at boot. | |
6027 | */ | |
6028 | if (found > count) | |
80934513 | 6029 | return true; |
49ac8255 | 6030 | } |
80934513 | 6031 | return false; |
49ac8255 KH |
6032 | } |
6033 | ||
6034 | bool is_pageblock_removable_nolock(struct page *page) | |
6035 | { | |
656a0706 MH |
6036 | struct zone *zone; |
6037 | unsigned long pfn; | |
687875fb MH |
6038 | |
6039 | /* | |
6040 | * We have to be careful here because we are iterating over memory | |
6041 | * sections which are not zone aware so we might end up outside of | |
6042 | * the zone but still within the section. | |
656a0706 MH |
6043 | * We have to take care about the node as well. If the node is offline |
6044 | * its NODE_DATA will be NULL - see page_zone. | |
687875fb | 6045 | */ |
656a0706 MH |
6046 | if (!node_online(page_to_nid(page))) |
6047 | return false; | |
6048 | ||
6049 | zone = page_zone(page); | |
6050 | pfn = page_to_pfn(page); | |
108bcc96 | 6051 | if (!zone_spans_pfn(zone, pfn)) |
687875fb MH |
6052 | return false; |
6053 | ||
b023f468 | 6054 | return !has_unmovable_pages(zone, page, 0, true); |
a5d76b54 | 6055 | } |
0c0e6195 | 6056 | |
041d3a8c MN |
6057 | #ifdef CONFIG_CMA |
6058 | ||
6059 | static unsigned long pfn_max_align_down(unsigned long pfn) | |
6060 | { | |
6061 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6062 | pageblock_nr_pages) - 1); | |
6063 | } | |
6064 | ||
6065 | static unsigned long pfn_max_align_up(unsigned long pfn) | |
6066 | { | |
6067 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, | |
6068 | pageblock_nr_pages)); | |
6069 | } | |
6070 | ||
041d3a8c | 6071 | /* [start, end) must belong to a single zone. */ |
bb13ffeb MG |
6072 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
6073 | unsigned long start, unsigned long end) | |
041d3a8c MN |
6074 | { |
6075 | /* This function is based on compact_zone() from compaction.c. */ | |
beb51eaa | 6076 | unsigned long nr_reclaimed; |
041d3a8c MN |
6077 | unsigned long pfn = start; |
6078 | unsigned int tries = 0; | |
6079 | int ret = 0; | |
6080 | ||
be49a6e1 | 6081 | migrate_prep(); |
041d3a8c | 6082 | |
bb13ffeb | 6083 | while (pfn < end || !list_empty(&cc->migratepages)) { |
041d3a8c MN |
6084 | if (fatal_signal_pending(current)) { |
6085 | ret = -EINTR; | |
6086 | break; | |
6087 | } | |
6088 | ||
bb13ffeb MG |
6089 | if (list_empty(&cc->migratepages)) { |
6090 | cc->nr_migratepages = 0; | |
6091 | pfn = isolate_migratepages_range(cc->zone, cc, | |
e46a2879 | 6092 | pfn, end, true); |
041d3a8c MN |
6093 | if (!pfn) { |
6094 | ret = -EINTR; | |
6095 | break; | |
6096 | } | |
6097 | tries = 0; | |
6098 | } else if (++tries == 5) { | |
6099 | ret = ret < 0 ? ret : -EBUSY; | |
6100 | break; | |
6101 | } | |
6102 | ||
beb51eaa MK |
6103 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
6104 | &cc->migratepages); | |
6105 | cc->nr_migratepages -= nr_reclaimed; | |
02c6de8d | 6106 | |
9c620e2b HD |
6107 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
6108 | 0, MIGRATE_SYNC, MR_CMA); | |
041d3a8c | 6109 | } |
2a6f5124 SP |
6110 | if (ret < 0) { |
6111 | putback_movable_pages(&cc->migratepages); | |
6112 | return ret; | |
6113 | } | |
6114 | return 0; | |
041d3a8c MN |
6115 | } |
6116 | ||
6117 | /** | |
6118 | * alloc_contig_range() -- tries to allocate given range of pages | |
6119 | * @start: start PFN to allocate | |
6120 | * @end: one-past-the-last PFN to allocate | |
0815f3d8 MN |
6121 | * @migratetype: migratetype of the underlaying pageblocks (either |
6122 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks | |
6123 | * in range must have the same migratetype and it must | |
6124 | * be either of the two. | |
041d3a8c MN |
6125 | * |
6126 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES | |
6127 | * aligned, however it's the caller's responsibility to guarantee that | |
6128 | * we are the only thread that changes migrate type of pageblocks the | |
6129 | * pages fall in. | |
6130 | * | |
6131 | * The PFN range must belong to a single zone. | |
6132 | * | |
6133 | * Returns zero on success or negative error code. On success all | |
6134 | * pages which PFN is in [start, end) are allocated for the caller and | |
6135 | * need to be freed with free_contig_range(). | |
6136 | */ | |
0815f3d8 MN |
6137 | int alloc_contig_range(unsigned long start, unsigned long end, |
6138 | unsigned migratetype) | |
041d3a8c | 6139 | { |
041d3a8c MN |
6140 | unsigned long outer_start, outer_end; |
6141 | int ret = 0, order; | |
6142 | ||
bb13ffeb MG |
6143 | struct compact_control cc = { |
6144 | .nr_migratepages = 0, | |
6145 | .order = -1, | |
6146 | .zone = page_zone(pfn_to_page(start)), | |
6147 | .sync = true, | |
6148 | .ignore_skip_hint = true, | |
6149 | }; | |
6150 | INIT_LIST_HEAD(&cc.migratepages); | |
6151 | ||
041d3a8c MN |
6152 | /* |
6153 | * What we do here is we mark all pageblocks in range as | |
6154 | * MIGRATE_ISOLATE. Because pageblock and max order pages may | |
6155 | * have different sizes, and due to the way page allocator | |
6156 | * work, we align the range to biggest of the two pages so | |
6157 | * that page allocator won't try to merge buddies from | |
6158 | * different pageblocks and change MIGRATE_ISOLATE to some | |
6159 | * other migration type. | |
6160 | * | |
6161 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we | |
6162 | * migrate the pages from an unaligned range (ie. pages that | |
6163 | * we are interested in). This will put all the pages in | |
6164 | * range back to page allocator as MIGRATE_ISOLATE. | |
6165 | * | |
6166 | * When this is done, we take the pages in range from page | |
6167 | * allocator removing them from the buddy system. This way | |
6168 | * page allocator will never consider using them. | |
6169 | * | |
6170 | * This lets us mark the pageblocks back as | |
6171 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the | |
6172 | * aligned range but not in the unaligned, original range are | |
6173 | * put back to page allocator so that buddy can use them. | |
6174 | */ | |
6175 | ||
6176 | ret = start_isolate_page_range(pfn_max_align_down(start), | |
b023f468 WC |
6177 | pfn_max_align_up(end), migratetype, |
6178 | false); | |
041d3a8c | 6179 | if (ret) |
86a595f9 | 6180 | return ret; |
041d3a8c | 6181 | |
bb13ffeb | 6182 | ret = __alloc_contig_migrate_range(&cc, start, end); |
041d3a8c MN |
6183 | if (ret) |
6184 | goto done; | |
6185 | ||
6186 | /* | |
6187 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES | |
6188 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's | |
6189 | * more, all pages in [start, end) are free in page allocator. | |
6190 | * What we are going to do is to allocate all pages from | |
6191 | * [start, end) (that is remove them from page allocator). | |
6192 | * | |
6193 | * The only problem is that pages at the beginning and at the | |
6194 | * end of interesting range may be not aligned with pages that | |
6195 | * page allocator holds, ie. they can be part of higher order | |
6196 | * pages. Because of this, we reserve the bigger range and | |
6197 | * once this is done free the pages we are not interested in. | |
6198 | * | |
6199 | * We don't have to hold zone->lock here because the pages are | |
6200 | * isolated thus they won't get removed from buddy. | |
6201 | */ | |
6202 | ||
6203 | lru_add_drain_all(); | |
6204 | drain_all_pages(); | |
6205 | ||
6206 | order = 0; | |
6207 | outer_start = start; | |
6208 | while (!PageBuddy(pfn_to_page(outer_start))) { | |
6209 | if (++order >= MAX_ORDER) { | |
6210 | ret = -EBUSY; | |
6211 | goto done; | |
6212 | } | |
6213 | outer_start &= ~0UL << order; | |
6214 | } | |
6215 | ||
6216 | /* Make sure the range is really isolated. */ | |
b023f468 | 6217 | if (test_pages_isolated(outer_start, end, false)) { |
041d3a8c MN |
6218 | pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n", |
6219 | outer_start, end); | |
6220 | ret = -EBUSY; | |
6221 | goto done; | |
6222 | } | |
6223 | ||
49f223a9 MS |
6224 | |
6225 | /* Grab isolated pages from freelists. */ | |
bb13ffeb | 6226 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
041d3a8c MN |
6227 | if (!outer_end) { |
6228 | ret = -EBUSY; | |
6229 | goto done; | |
6230 | } | |
6231 | ||
6232 | /* Free head and tail (if any) */ | |
6233 | if (start != outer_start) | |
6234 | free_contig_range(outer_start, start - outer_start); | |
6235 | if (end != outer_end) | |
6236 | free_contig_range(end, outer_end - end); | |
6237 | ||
6238 | done: | |
6239 | undo_isolate_page_range(pfn_max_align_down(start), | |
0815f3d8 | 6240 | pfn_max_align_up(end), migratetype); |
041d3a8c MN |
6241 | return ret; |
6242 | } | |
6243 | ||
6244 | void free_contig_range(unsigned long pfn, unsigned nr_pages) | |
6245 | { | |
bcc2b02f MS |
6246 | unsigned int count = 0; |
6247 | ||
6248 | for (; nr_pages--; pfn++) { | |
6249 | struct page *page = pfn_to_page(pfn); | |
6250 | ||
6251 | count += page_count(page) != 1; | |
6252 | __free_page(page); | |
6253 | } | |
6254 | WARN(count != 0, "%d pages are still in use!\n", count); | |
041d3a8c MN |
6255 | } |
6256 | #endif | |
6257 | ||
4ed7e022 JL |
6258 | #ifdef CONFIG_MEMORY_HOTPLUG |
6259 | static int __meminit __zone_pcp_update(void *data) | |
6260 | { | |
6261 | struct zone *zone = data; | |
6262 | int cpu; | |
6263 | unsigned long batch = zone_batchsize(zone), flags; | |
6264 | ||
6265 | for_each_possible_cpu(cpu) { | |
6266 | struct per_cpu_pageset *pset; | |
6267 | struct per_cpu_pages *pcp; | |
6268 | ||
6269 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6270 | pcp = &pset->pcp; | |
6271 | ||
6272 | local_irq_save(flags); | |
6273 | if (pcp->count > 0) | |
6274 | free_pcppages_bulk(zone, pcp->count, pcp); | |
5a883813 | 6275 | drain_zonestat(zone, pset); |
4ed7e022 JL |
6276 | setup_pageset(pset, batch); |
6277 | local_irq_restore(flags); | |
6278 | } | |
6279 | return 0; | |
6280 | } | |
6281 | ||
6282 | void __meminit zone_pcp_update(struct zone *zone) | |
6283 | { | |
6284 | stop_machine(__zone_pcp_update, zone, NULL); | |
6285 | } | |
6286 | #endif | |
6287 | ||
340175b7 JL |
6288 | void zone_pcp_reset(struct zone *zone) |
6289 | { | |
6290 | unsigned long flags; | |
5a883813 MK |
6291 | int cpu; |
6292 | struct per_cpu_pageset *pset; | |
340175b7 JL |
6293 | |
6294 | /* avoid races with drain_pages() */ | |
6295 | local_irq_save(flags); | |
6296 | if (zone->pageset != &boot_pageset) { | |
5a883813 MK |
6297 | for_each_online_cpu(cpu) { |
6298 | pset = per_cpu_ptr(zone->pageset, cpu); | |
6299 | drain_zonestat(zone, pset); | |
6300 | } | |
340175b7 JL |
6301 | free_percpu(zone->pageset); |
6302 | zone->pageset = &boot_pageset; | |
6303 | } | |
6304 | local_irq_restore(flags); | |
6305 | } | |
6306 | ||
6dcd73d7 | 6307 | #ifdef CONFIG_MEMORY_HOTREMOVE |
0c0e6195 KH |
6308 | /* |
6309 | * All pages in the range must be isolated before calling this. | |
6310 | */ | |
6311 | void | |
6312 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
6313 | { | |
6314 | struct page *page; | |
6315 | struct zone *zone; | |
6316 | int order, i; | |
6317 | unsigned long pfn; | |
6318 | unsigned long flags; | |
6319 | /* find the first valid pfn */ | |
6320 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
6321 | if (pfn_valid(pfn)) | |
6322 | break; | |
6323 | if (pfn == end_pfn) | |
6324 | return; | |
6325 | zone = page_zone(pfn_to_page(pfn)); | |
6326 | spin_lock_irqsave(&zone->lock, flags); | |
6327 | pfn = start_pfn; | |
6328 | while (pfn < end_pfn) { | |
6329 | if (!pfn_valid(pfn)) { | |
6330 | pfn++; | |
6331 | continue; | |
6332 | } | |
6333 | page = pfn_to_page(pfn); | |
b023f468 WC |
6334 | /* |
6335 | * The HWPoisoned page may be not in buddy system, and | |
6336 | * page_count() is not 0. | |
6337 | */ | |
6338 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { | |
6339 | pfn++; | |
6340 | SetPageReserved(page); | |
6341 | continue; | |
6342 | } | |
6343 | ||
0c0e6195 KH |
6344 | BUG_ON(page_count(page)); |
6345 | BUG_ON(!PageBuddy(page)); | |
6346 | order = page_order(page); | |
6347 | #ifdef CONFIG_DEBUG_VM | |
6348 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
6349 | pfn, 1 << order, end_pfn); | |
6350 | #endif | |
6351 | list_del(&page->lru); | |
6352 | rmv_page_order(page); | |
6353 | zone->free_area[order].nr_free--; | |
0c0e6195 KH |
6354 | for (i = 0; i < (1 << order); i++) |
6355 | SetPageReserved((page+i)); | |
6356 | pfn += (1 << order); | |
6357 | } | |
6358 | spin_unlock_irqrestore(&zone->lock, flags); | |
6359 | } | |
6360 | #endif | |
8d22ba1b WF |
6361 | |
6362 | #ifdef CONFIG_MEMORY_FAILURE | |
6363 | bool is_free_buddy_page(struct page *page) | |
6364 | { | |
6365 | struct zone *zone = page_zone(page); | |
6366 | unsigned long pfn = page_to_pfn(page); | |
6367 | unsigned long flags; | |
6368 | int order; | |
6369 | ||
6370 | spin_lock_irqsave(&zone->lock, flags); | |
6371 | for (order = 0; order < MAX_ORDER; order++) { | |
6372 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
6373 | ||
6374 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
6375 | break; | |
6376 | } | |
6377 | spin_unlock_irqrestore(&zone->lock, flags); | |
6378 | ||
6379 | return order < MAX_ORDER; | |
6380 | } | |
6381 | #endif | |
718a3821 | 6382 | |
51300cef | 6383 | static const struct trace_print_flags pageflag_names[] = { |
718a3821 WF |
6384 | {1UL << PG_locked, "locked" }, |
6385 | {1UL << PG_error, "error" }, | |
6386 | {1UL << PG_referenced, "referenced" }, | |
6387 | {1UL << PG_uptodate, "uptodate" }, | |
6388 | {1UL << PG_dirty, "dirty" }, | |
6389 | {1UL << PG_lru, "lru" }, | |
6390 | {1UL << PG_active, "active" }, | |
6391 | {1UL << PG_slab, "slab" }, | |
6392 | {1UL << PG_owner_priv_1, "owner_priv_1" }, | |
6393 | {1UL << PG_arch_1, "arch_1" }, | |
6394 | {1UL << PG_reserved, "reserved" }, | |
6395 | {1UL << PG_private, "private" }, | |
6396 | {1UL << PG_private_2, "private_2" }, | |
6397 | {1UL << PG_writeback, "writeback" }, | |
6398 | #ifdef CONFIG_PAGEFLAGS_EXTENDED | |
6399 | {1UL << PG_head, "head" }, | |
6400 | {1UL << PG_tail, "tail" }, | |
6401 | #else | |
6402 | {1UL << PG_compound, "compound" }, | |
6403 | #endif | |
6404 | {1UL << PG_swapcache, "swapcache" }, | |
6405 | {1UL << PG_mappedtodisk, "mappedtodisk" }, | |
6406 | {1UL << PG_reclaim, "reclaim" }, | |
718a3821 WF |
6407 | {1UL << PG_swapbacked, "swapbacked" }, |
6408 | {1UL << PG_unevictable, "unevictable" }, | |
6409 | #ifdef CONFIG_MMU | |
6410 | {1UL << PG_mlocked, "mlocked" }, | |
6411 | #endif | |
6412 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED | |
6413 | {1UL << PG_uncached, "uncached" }, | |
6414 | #endif | |
6415 | #ifdef CONFIG_MEMORY_FAILURE | |
6416 | {1UL << PG_hwpoison, "hwpoison" }, | |
be9cd873 GS |
6417 | #endif |
6418 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
6419 | {1UL << PG_compound_lock, "compound_lock" }, | |
718a3821 | 6420 | #endif |
718a3821 WF |
6421 | }; |
6422 | ||
6423 | static void dump_page_flags(unsigned long flags) | |
6424 | { | |
6425 | const char *delim = ""; | |
6426 | unsigned long mask; | |
6427 | int i; | |
6428 | ||
51300cef | 6429 | BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS); |
acc50c11 | 6430 | |
718a3821 WF |
6431 | printk(KERN_ALERT "page flags: %#lx(", flags); |
6432 | ||
6433 | /* remove zone id */ | |
6434 | flags &= (1UL << NR_PAGEFLAGS) - 1; | |
6435 | ||
51300cef | 6436 | for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) { |
718a3821 WF |
6437 | |
6438 | mask = pageflag_names[i].mask; | |
6439 | if ((flags & mask) != mask) | |
6440 | continue; | |
6441 | ||
6442 | flags &= ~mask; | |
6443 | printk("%s%s", delim, pageflag_names[i].name); | |
6444 | delim = "|"; | |
6445 | } | |
6446 | ||
6447 | /* check for left over flags */ | |
6448 | if (flags) | |
6449 | printk("%s%#lx", delim, flags); | |
6450 | ||
6451 | printk(")\n"); | |
6452 | } | |
6453 | ||
6454 | void dump_page(struct page *page) | |
6455 | { | |
6456 | printk(KERN_ALERT | |
6457 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", | |
4e9f64c4 | 6458 | page, atomic_read(&page->_count), page_mapcount(page), |
718a3821 WF |
6459 | page->mapping, page->index); |
6460 | dump_page_flags(page->flags); | |
f212ad7c | 6461 | mem_cgroup_print_bad_page(page); |
718a3821 | 6462 | } |