blob: 9711cb8ba13cab94f5f31b49b651e1dfcb67a542
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 | |
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> |
22 | #include <linux/jiffies.h> |
23 | #include <linux/bootmem.h> |
24 | #include <linux/memblock.h> |
25 | #include <linux/compiler.h> |
26 | #include <linux/kernel.h> |
27 | #include <linux/kmemcheck.h> |
28 | #include <linux/kasan.h> |
29 | #include <linux/module.h> |
30 | #include <linux/suspend.h> |
31 | #include <linux/pagevec.h> |
32 | #include <linux/blkdev.h> |
33 | #include <linux/slab.h> |
34 | #include <linux/ratelimit.h> |
35 | #include <linux/oom.h> |
36 | #include <linux/notifier.h> |
37 | #include <linux/topology.h> |
38 | #include <linux/sysctl.h> |
39 | #include <linux/cpu.h> |
40 | #include <linux/cpuset.h> |
41 | #include <linux/memory_hotplug.h> |
42 | #include <linux/nodemask.h> |
43 | #include <linux/vmalloc.h> |
44 | #include <linux/vmstat.h> |
45 | #include <linux/mempolicy.h> |
46 | #include <linux/memremap.h> |
47 | #include <linux/stop_machine.h> |
48 | #include <linux/sort.h> |
49 | #include <linux/pfn.h> |
50 | #include <linux/backing-dev.h> |
51 | #include <linux/fault-inject.h> |
52 | #include <linux/page-isolation.h> |
53 | #include <linux/page_ext.h> |
54 | #include <linux/debugobjects.h> |
55 | #include <linux/kmemleak.h> |
56 | #include <linux/compaction.h> |
57 | #include <trace/events/kmem.h> |
58 | #include <linux/prefetch.h> |
59 | #include <linux/mm_inline.h> |
60 | #include <linux/migrate.h> |
61 | #include <linux/page_ext.h> |
62 | #include <linux/hugetlb.h> |
63 | #include <linux/sched/rt.h> |
64 | #include <linux/page_owner.h> |
65 | #include <linux/kthread.h> |
66 | #include <linux/memcontrol.h> |
67 | #include <linux/psi.h> |
68 | #ifdef CONFIG_AMLOGIC_PAGE_TRACE |
69 | #include <linux/amlogic/page_trace.h> |
70 | #endif /* CONFIG_AMLOGIC_PAGE_TRACE */ |
71 | |
72 | #include <asm/sections.h> |
73 | #include <asm/tlbflush.h> |
74 | #include <asm/div64.h> |
75 | #include "internal.h" |
76 | |
77 | /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */ |
78 | static DEFINE_MUTEX(pcp_batch_high_lock); |
79 | #define MIN_PERCPU_PAGELIST_FRACTION (8) |
80 | |
81 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
82 | DEFINE_PER_CPU(int, numa_node); |
83 | EXPORT_PER_CPU_SYMBOL(numa_node); |
84 | #endif |
85 | |
86 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
87 | /* |
88 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. |
89 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. |
90 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() |
91 | * defined in <linux/topology.h>. |
92 | */ |
93 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ |
94 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); |
95 | int _node_numa_mem_[MAX_NUMNODES]; |
96 | #endif |
97 | |
98 | #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY |
99 | volatile unsigned long latent_entropy __latent_entropy; |
100 | EXPORT_SYMBOL(latent_entropy); |
101 | #endif |
102 | |
103 | /* |
104 | * Array of node states. |
105 | */ |
106 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
107 | [N_POSSIBLE] = NODE_MASK_ALL, |
108 | [N_ONLINE] = { { [0] = 1UL } }, |
109 | #ifndef CONFIG_NUMA |
110 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, |
111 | #ifdef CONFIG_HIGHMEM |
112 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, |
113 | #endif |
114 | #ifdef CONFIG_MOVABLE_NODE |
115 | [N_MEMORY] = { { [0] = 1UL } }, |
116 | #endif |
117 | [N_CPU] = { { [0] = 1UL } }, |
118 | #endif /* NUMA */ |
119 | }; |
120 | EXPORT_SYMBOL(node_states); |
121 | |
122 | /* Protect totalram_pages and zone->managed_pages */ |
123 | static DEFINE_SPINLOCK(managed_page_count_lock); |
124 | |
125 | unsigned long totalram_pages __read_mostly; |
126 | unsigned long totalreserve_pages __read_mostly; |
127 | unsigned long totalcma_pages __read_mostly; |
128 | |
129 | int percpu_pagelist_fraction; |
130 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
131 | |
132 | /* |
133 | * A cached value of the page's pageblock's migratetype, used when the page is |
134 | * put on a pcplist. Used to avoid the pageblock migratetype lookup when |
135 | * freeing from pcplists in most cases, at the cost of possibly becoming stale. |
136 | * Also the migratetype set in the page does not necessarily match the pcplist |
137 | * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any |
138 | * other index - this ensures that it will be put on the correct CMA freelist. |
139 | */ |
140 | static inline int get_pcppage_migratetype(struct page *page) |
141 | { |
142 | return page->index; |
143 | } |
144 | |
145 | static inline void set_pcppage_migratetype(struct page *page, int migratetype) |
146 | { |
147 | page->index = migratetype; |
148 | } |
149 | |
150 | #ifdef CONFIG_PM_SLEEP |
151 | /* |
152 | * The following functions are used by the suspend/hibernate code to temporarily |
153 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations |
154 | * while devices are suspended. To avoid races with the suspend/hibernate code, |
155 | * they should always be called with pm_mutex held (gfp_allowed_mask also should |
156 | * only be modified with pm_mutex held, unless the suspend/hibernate code is |
157 | * guaranteed not to run in parallel with that modification). |
158 | */ |
159 | |
160 | static gfp_t saved_gfp_mask; |
161 | |
162 | void pm_restore_gfp_mask(void) |
163 | { |
164 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
165 | if (saved_gfp_mask) { |
166 | gfp_allowed_mask = saved_gfp_mask; |
167 | saved_gfp_mask = 0; |
168 | } |
169 | } |
170 | |
171 | void pm_restrict_gfp_mask(void) |
172 | { |
173 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
174 | WARN_ON(saved_gfp_mask); |
175 | saved_gfp_mask = gfp_allowed_mask; |
176 | gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS); |
177 | } |
178 | |
179 | bool pm_suspended_storage(void) |
180 | { |
181 | if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
182 | return false; |
183 | return true; |
184 | } |
185 | #endif /* CONFIG_PM_SLEEP */ |
186 | |
187 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
188 | unsigned int pageblock_order __read_mostly; |
189 | #endif |
190 | |
191 | static void __free_pages_ok(struct page *page, unsigned int order); |
192 | |
193 | /* |
194 | * results with 256, 32 in the lowmem_reserve sysctl: |
195 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) |
196 | * 1G machine -> (16M dma, 784M normal, 224M high) |
197 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA |
198 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL |
199 | * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA |
200 | * |
201 | * TBD: should special case ZONE_DMA32 machines here - in those we normally |
202 | * don't need any ZONE_NORMAL reservation |
203 | */ |
204 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
205 | #ifdef CONFIG_ZONE_DMA |
206 | 256, |
207 | #endif |
208 | #ifdef CONFIG_ZONE_DMA32 |
209 | 256, |
210 | #endif |
211 | #ifdef CONFIG_HIGHMEM |
212 | 32, |
213 | #endif |
214 | 32, |
215 | }; |
216 | |
217 | EXPORT_SYMBOL(totalram_pages); |
218 | |
219 | static char * const zone_names[MAX_NR_ZONES] = { |
220 | #ifdef CONFIG_ZONE_DMA |
221 | "DMA", |
222 | #endif |
223 | #ifdef CONFIG_ZONE_DMA32 |
224 | "DMA32", |
225 | #endif |
226 | "Normal", |
227 | #ifdef CONFIG_HIGHMEM |
228 | "HighMem", |
229 | #endif |
230 | "Movable", |
231 | #ifdef CONFIG_ZONE_DEVICE |
232 | "Device", |
233 | #endif |
234 | }; |
235 | |
236 | char * const migratetype_names[MIGRATE_TYPES] = { |
237 | "Unmovable", |
238 | "Movable", |
239 | "Reclaimable", |
240 | "HighAtomic", |
241 | #ifdef CONFIG_CMA |
242 | "CMA", |
243 | #endif |
244 | #ifdef CONFIG_MEMORY_ISOLATION |
245 | "Isolate", |
246 | #endif |
247 | }; |
248 | |
249 | compound_page_dtor * const compound_page_dtors[] = { |
250 | NULL, |
251 | free_compound_page, |
252 | #ifdef CONFIG_HUGETLB_PAGE |
253 | free_huge_page, |
254 | #endif |
255 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
256 | free_transhuge_page, |
257 | #endif |
258 | }; |
259 | |
260 | /* |
261 | * Try to keep at least this much lowmem free. Do not allow normal |
262 | * allocations below this point, only high priority ones. Automatically |
263 | * tuned according to the amount of memory in the system. |
264 | */ |
265 | int min_free_kbytes = 1024; |
266 | int user_min_free_kbytes = -1; |
267 | int watermark_scale_factor = 10; |
268 | |
269 | /* |
270 | * Extra memory for the system to try freeing. Used to temporarily |
271 | * free memory, to make space for new workloads. Anyone can allocate |
272 | * down to the min watermarks controlled by min_free_kbytes above. |
273 | */ |
274 | int extra_free_kbytes = 0; |
275 | |
276 | static unsigned long __meminitdata nr_kernel_pages; |
277 | static unsigned long __meminitdata nr_all_pages; |
278 | static unsigned long __meminitdata dma_reserve; |
279 | |
280 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
281 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; |
282 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; |
283 | static unsigned long __initdata required_kernelcore; |
284 | static unsigned long __initdata required_movablecore; |
285 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
286 | static bool mirrored_kernelcore; |
287 | |
288 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ |
289 | int movable_zone; |
290 | EXPORT_SYMBOL(movable_zone); |
291 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
292 | |
293 | #if MAX_NUMNODES > 1 |
294 | int nr_node_ids __read_mostly = MAX_NUMNODES; |
295 | int nr_online_nodes __read_mostly = 1; |
296 | EXPORT_SYMBOL(nr_node_ids); |
297 | EXPORT_SYMBOL(nr_online_nodes); |
298 | #endif |
299 | |
300 | int page_group_by_mobility_disabled __read_mostly; |
301 | |
302 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
303 | |
304 | /* |
305 | * Determine how many pages need to be initialized durig early boot |
306 | * (non-deferred initialization). |
307 | * The value of first_deferred_pfn will be set later, once non-deferred pages |
308 | * are initialized, but for now set it ULONG_MAX. |
309 | */ |
310 | static inline void reset_deferred_meminit(pg_data_t *pgdat) |
311 | { |
312 | phys_addr_t start_addr, end_addr; |
313 | unsigned long max_pgcnt; |
314 | unsigned long reserved; |
315 | |
316 | /* |
317 | * Initialise at least 2G of a node but also take into account that |
318 | * two large system hashes that can take up 1GB for 0.25TB/node. |
319 | */ |
320 | max_pgcnt = max(2UL << (30 - PAGE_SHIFT), |
321 | (pgdat->node_spanned_pages >> 8)); |
322 | |
323 | /* |
324 | * Compensate the all the memblock reservations (e.g. crash kernel) |
325 | * from the initial estimation to make sure we will initialize enough |
326 | * memory to boot. |
327 | */ |
328 | start_addr = PFN_PHYS(pgdat->node_start_pfn); |
329 | end_addr = PFN_PHYS(pgdat->node_start_pfn + max_pgcnt); |
330 | reserved = memblock_reserved_memory_within(start_addr, end_addr); |
331 | max_pgcnt += PHYS_PFN(reserved); |
332 | |
333 | pgdat->static_init_pgcnt = min(max_pgcnt, pgdat->node_spanned_pages); |
334 | pgdat->first_deferred_pfn = ULONG_MAX; |
335 | } |
336 | |
337 | /* Returns true if the struct page for the pfn is uninitialised */ |
338 | static inline bool __meminit early_page_uninitialised(unsigned long pfn) |
339 | { |
340 | int nid = early_pfn_to_nid(pfn); |
341 | |
342 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn) |
343 | return true; |
344 | |
345 | return false; |
346 | } |
347 | |
348 | /* |
349 | * Returns false when the remaining initialisation should be deferred until |
350 | * later in the boot cycle when it can be parallelised. |
351 | */ |
352 | static inline bool update_defer_init(pg_data_t *pgdat, |
353 | unsigned long pfn, unsigned long zone_end, |
354 | unsigned long *nr_initialised) |
355 | { |
356 | /* Always populate low zones for address-contrained allocations */ |
357 | if (zone_end < pgdat_end_pfn(pgdat)) |
358 | return true; |
359 | (*nr_initialised)++; |
360 | if ((*nr_initialised > pgdat->static_init_pgcnt) && |
361 | (pfn & (PAGES_PER_SECTION - 1)) == 0) { |
362 | pgdat->first_deferred_pfn = pfn; |
363 | return false; |
364 | } |
365 | |
366 | return true; |
367 | } |
368 | #else |
369 | static inline void reset_deferred_meminit(pg_data_t *pgdat) |
370 | { |
371 | } |
372 | |
373 | static inline bool early_page_uninitialised(unsigned long pfn) |
374 | { |
375 | return false; |
376 | } |
377 | |
378 | static inline bool update_defer_init(pg_data_t *pgdat, |
379 | unsigned long pfn, unsigned long zone_end, |
380 | unsigned long *nr_initialised) |
381 | { |
382 | return true; |
383 | } |
384 | #endif |
385 | |
386 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
387 | static inline unsigned long *get_pageblock_bitmap(struct page *page, |
388 | unsigned long pfn) |
389 | { |
390 | #ifdef CONFIG_SPARSEMEM |
391 | return __pfn_to_section(pfn)->pageblock_flags; |
392 | #else |
393 | return page_zone(page)->pageblock_flags; |
394 | #endif /* CONFIG_SPARSEMEM */ |
395 | } |
396 | |
397 | static inline int pfn_to_bitidx(struct page *page, unsigned long pfn) |
398 | { |
399 | #ifdef CONFIG_SPARSEMEM |
400 | pfn &= (PAGES_PER_SECTION-1); |
401 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
402 | #else |
403 | pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages); |
404 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
405 | #endif /* CONFIG_SPARSEMEM */ |
406 | } |
407 | |
408 | /** |
409 | * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages |
410 | * @page: The page within the block of interest |
411 | * @pfn: The target page frame number |
412 | * @end_bitidx: The last bit of interest to retrieve |
413 | * @mask: mask of bits that the caller is interested in |
414 | * |
415 | * Return: pageblock_bits flags |
416 | */ |
417 | static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page, |
418 | unsigned long pfn, |
419 | unsigned long end_bitidx, |
420 | unsigned long mask) |
421 | { |
422 | unsigned long *bitmap; |
423 | unsigned long bitidx, word_bitidx; |
424 | unsigned long word; |
425 | |
426 | bitmap = get_pageblock_bitmap(page, pfn); |
427 | bitidx = pfn_to_bitidx(page, pfn); |
428 | word_bitidx = bitidx / BITS_PER_LONG; |
429 | bitidx &= (BITS_PER_LONG-1); |
430 | |
431 | word = bitmap[word_bitidx]; |
432 | bitidx += end_bitidx; |
433 | return (word >> (BITS_PER_LONG - bitidx - 1)) & mask; |
434 | } |
435 | |
436 | unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn, |
437 | unsigned long end_bitidx, |
438 | unsigned long mask) |
439 | { |
440 | return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask); |
441 | } |
442 | |
443 | static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) |
444 | { |
445 | return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK); |
446 | } |
447 | |
448 | /** |
449 | * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages |
450 | * @page: The page within the block of interest |
451 | * @flags: The flags to set |
452 | * @pfn: The target page frame number |
453 | * @end_bitidx: The last bit of interest |
454 | * @mask: mask of bits that the caller is interested in |
455 | */ |
456 | void set_pfnblock_flags_mask(struct page *page, unsigned long flags, |
457 | unsigned long pfn, |
458 | unsigned long end_bitidx, |
459 | unsigned long mask) |
460 | { |
461 | unsigned long *bitmap; |
462 | unsigned long bitidx, word_bitidx; |
463 | unsigned long old_word, word; |
464 | |
465 | BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4); |
466 | |
467 | bitmap = get_pageblock_bitmap(page, pfn); |
468 | bitidx = pfn_to_bitidx(page, pfn); |
469 | word_bitidx = bitidx / BITS_PER_LONG; |
470 | bitidx &= (BITS_PER_LONG-1); |
471 | |
472 | VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page); |
473 | |
474 | bitidx += end_bitidx; |
475 | mask <<= (BITS_PER_LONG - bitidx - 1); |
476 | flags <<= (BITS_PER_LONG - bitidx - 1); |
477 | |
478 | word = READ_ONCE(bitmap[word_bitidx]); |
479 | for (;;) { |
480 | old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags); |
481 | if (word == old_word) |
482 | break; |
483 | word = old_word; |
484 | } |
485 | } |
486 | |
487 | void set_pageblock_migratetype(struct page *page, int migratetype) |
488 | { |
489 | if (unlikely(page_group_by_mobility_disabled && |
490 | migratetype < MIGRATE_PCPTYPES)) |
491 | migratetype = MIGRATE_UNMOVABLE; |
492 | |
493 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
494 | PB_migrate, PB_migrate_end); |
495 | } |
496 | |
497 | #ifdef CONFIG_DEBUG_VM |
498 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
499 | { |
500 | int ret = 0; |
501 | unsigned seq; |
502 | unsigned long pfn = page_to_pfn(page); |
503 | unsigned long sp, start_pfn; |
504 | |
505 | do { |
506 | seq = zone_span_seqbegin(zone); |
507 | start_pfn = zone->zone_start_pfn; |
508 | sp = zone->spanned_pages; |
509 | if (!zone_spans_pfn(zone, pfn)) |
510 | ret = 1; |
511 | } while (zone_span_seqretry(zone, seq)); |
512 | |
513 | if (ret) |
514 | pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", |
515 | pfn, zone_to_nid(zone), zone->name, |
516 | start_pfn, start_pfn + sp); |
517 | |
518 | return ret; |
519 | } |
520 | |
521 | static int page_is_consistent(struct zone *zone, struct page *page) |
522 | { |
523 | if (!pfn_valid_within(page_to_pfn(page))) |
524 | return 0; |
525 | if (zone != page_zone(page)) |
526 | return 0; |
527 | |
528 | return 1; |
529 | } |
530 | /* |
531 | * Temporary debugging check for pages not lying within a given zone. |
532 | */ |
533 | static int bad_range(struct zone *zone, struct page *page) |
534 | { |
535 | if (page_outside_zone_boundaries(zone, page)) |
536 | return 1; |
537 | if (!page_is_consistent(zone, page)) |
538 | return 1; |
539 | |
540 | return 0; |
541 | } |
542 | #else |
543 | static inline int bad_range(struct zone *zone, struct page *page) |
544 | { |
545 | return 0; |
546 | } |
547 | #endif |
548 | |
549 | static void bad_page(struct page *page, const char *reason, |
550 | unsigned long bad_flags) |
551 | { |
552 | static unsigned long resume; |
553 | static unsigned long nr_shown; |
554 | static unsigned long nr_unshown; |
555 | |
556 | /* |
557 | * Allow a burst of 60 reports, then keep quiet for that minute; |
558 | * or allow a steady drip of one report per second. |
559 | */ |
560 | if (nr_shown == 60) { |
561 | if (time_before(jiffies, resume)) { |
562 | nr_unshown++; |
563 | goto out; |
564 | } |
565 | if (nr_unshown) { |
566 | pr_alert( |
567 | "BUG: Bad page state: %lu messages suppressed\n", |
568 | nr_unshown); |
569 | nr_unshown = 0; |
570 | } |
571 | nr_shown = 0; |
572 | } |
573 | if (nr_shown++ == 0) |
574 | resume = jiffies + 60 * HZ; |
575 | |
576 | pr_alert("BUG: Bad page state in process %s pfn:%05lx\n", |
577 | current->comm, page_to_pfn(page)); |
578 | __dump_page(page, reason); |
579 | bad_flags &= page->flags; |
580 | if (bad_flags) |
581 | pr_alert("bad because of flags: %#lx(%pGp)\n", |
582 | bad_flags, &bad_flags); |
583 | dump_page_owner(page); |
584 | |
585 | print_modules(); |
586 | dump_stack(); |
587 | out: |
588 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
589 | page_mapcount_reset(page); /* remove PageBuddy */ |
590 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
591 | } |
592 | |
593 | /* |
594 | * Higher-order pages are called "compound pages". They are structured thusly: |
595 | * |
596 | * The first PAGE_SIZE page is called the "head page" and have PG_head set. |
597 | * |
598 | * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded |
599 | * in bit 0 of page->compound_head. The rest of bits is pointer to head page. |
600 | * |
601 | * The first tail page's ->compound_dtor holds the offset in array of compound |
602 | * page destructors. See compound_page_dtors. |
603 | * |
604 | * The first tail page's ->compound_order holds the order of allocation. |
605 | * This usage means that zero-order pages may not be compound. |
606 | */ |
607 | |
608 | void free_compound_page(struct page *page) |
609 | { |
610 | __free_pages_ok(page, compound_order(page)); |
611 | } |
612 | |
613 | void prep_compound_page(struct page *page, unsigned int order) |
614 | { |
615 | int i; |
616 | int nr_pages = 1 << order; |
617 | |
618 | set_compound_page_dtor(page, COMPOUND_PAGE_DTOR); |
619 | set_compound_order(page, order); |
620 | __SetPageHead(page); |
621 | for (i = 1; i < nr_pages; i++) { |
622 | struct page *p = page + i; |
623 | set_page_count(p, 0); |
624 | p->mapping = TAIL_MAPPING; |
625 | set_compound_head(p, page); |
626 | } |
627 | atomic_set(compound_mapcount_ptr(page), -1); |
628 | } |
629 | |
630 | #ifdef CONFIG_DEBUG_PAGEALLOC |
631 | unsigned int _debug_guardpage_minorder; |
632 | bool _debug_pagealloc_enabled __read_mostly |
633 | = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT); |
634 | EXPORT_SYMBOL(_debug_pagealloc_enabled); |
635 | bool _debug_guardpage_enabled __read_mostly; |
636 | |
637 | static int __init early_debug_pagealloc(char *buf) |
638 | { |
639 | if (!buf) |
640 | return -EINVAL; |
641 | return kstrtobool(buf, &_debug_pagealloc_enabled); |
642 | } |
643 | early_param("debug_pagealloc", early_debug_pagealloc); |
644 | |
645 | static bool need_debug_guardpage(void) |
646 | { |
647 | /* If we don't use debug_pagealloc, we don't need guard page */ |
648 | if (!debug_pagealloc_enabled()) |
649 | return false; |
650 | |
651 | if (!debug_guardpage_minorder()) |
652 | return false; |
653 | |
654 | return true; |
655 | } |
656 | |
657 | static void init_debug_guardpage(void) |
658 | { |
659 | if (!debug_pagealloc_enabled()) |
660 | return; |
661 | |
662 | if (!debug_guardpage_minorder()) |
663 | return; |
664 | |
665 | _debug_guardpage_enabled = true; |
666 | } |
667 | |
668 | struct page_ext_operations debug_guardpage_ops = { |
669 | .need = need_debug_guardpage, |
670 | .init = init_debug_guardpage, |
671 | }; |
672 | |
673 | static int __init debug_guardpage_minorder_setup(char *buf) |
674 | { |
675 | unsigned long res; |
676 | |
677 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { |
678 | pr_err("Bad debug_guardpage_minorder value\n"); |
679 | return 0; |
680 | } |
681 | _debug_guardpage_minorder = res; |
682 | pr_info("Setting debug_guardpage_minorder to %lu\n", res); |
683 | return 0; |
684 | } |
685 | early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup); |
686 | |
687 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
688 | unsigned int order, int migratetype) |
689 | { |
690 | struct page_ext *page_ext; |
691 | |
692 | if (!debug_guardpage_enabled()) |
693 | return false; |
694 | |
695 | if (order >= debug_guardpage_minorder()) |
696 | return false; |
697 | |
698 | page_ext = lookup_page_ext(page); |
699 | if (unlikely(!page_ext)) |
700 | return false; |
701 | |
702 | __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
703 | |
704 | INIT_LIST_HEAD(&page->lru); |
705 | set_page_private(page, order); |
706 | /* Guard pages are not available for any usage */ |
707 | __mod_zone_freepage_state(zone, -(1 << order), migratetype); |
708 | |
709 | return true; |
710 | } |
711 | |
712 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
713 | unsigned int order, int migratetype) |
714 | { |
715 | struct page_ext *page_ext; |
716 | |
717 | if (!debug_guardpage_enabled()) |
718 | return; |
719 | |
720 | page_ext = lookup_page_ext(page); |
721 | if (unlikely(!page_ext)) |
722 | return; |
723 | |
724 | __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags); |
725 | |
726 | set_page_private(page, 0); |
727 | if (!is_migrate_isolate(migratetype)) |
728 | __mod_zone_freepage_state(zone, (1 << order), migratetype); |
729 | } |
730 | #else |
731 | struct page_ext_operations debug_guardpage_ops; |
732 | static inline bool set_page_guard(struct zone *zone, struct page *page, |
733 | unsigned int order, int migratetype) { return false; } |
734 | static inline void clear_page_guard(struct zone *zone, struct page *page, |
735 | unsigned int order, int migratetype) {} |
736 | #endif |
737 | |
738 | static inline void set_page_order(struct page *page, unsigned int order) |
739 | { |
740 | set_page_private(page, order); |
741 | __SetPageBuddy(page); |
742 | } |
743 | |
744 | static inline void rmv_page_order(struct page *page) |
745 | { |
746 | __ClearPageBuddy(page); |
747 | set_page_private(page, 0); |
748 | } |
749 | |
750 | /* |
751 | * This function checks whether a page is free && is the buddy |
752 | * we can do coalesce a page and its buddy if |
753 | * (a) the buddy is not in a hole && |
754 | * (b) the buddy is in the buddy system && |
755 | * (c) a page and its buddy have the same order && |
756 | * (d) a page and its buddy are in the same zone. |
757 | * |
758 | * For recording whether a page is in the buddy system, we set ->_mapcount |
759 | * PAGE_BUDDY_MAPCOUNT_VALUE. |
760 | * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is |
761 | * serialized by zone->lock. |
762 | * |
763 | * For recording page's order, we use page_private(page). |
764 | */ |
765 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
766 | unsigned int order) |
767 | { |
768 | if (!pfn_valid_within(page_to_pfn(buddy))) |
769 | return 0; |
770 | |
771 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
772 | if (page_zone_id(page) != page_zone_id(buddy)) |
773 | return 0; |
774 | |
775 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
776 | |
777 | return 1; |
778 | } |
779 | |
780 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
781 | /* |
782 | * zone check is done late to avoid uselessly |
783 | * calculating zone/node ids for pages that could |
784 | * never merge. |
785 | */ |
786 | if (page_zone_id(page) != page_zone_id(buddy)) |
787 | return 0; |
788 | |
789 | VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy); |
790 | |
791 | return 1; |
792 | } |
793 | return 0; |
794 | } |
795 | |
796 | /* |
797 | * Freeing function for a buddy system allocator. |
798 | * |
799 | * The concept of a buddy system is to maintain direct-mapped table |
800 | * (containing bit values) for memory blocks of various "orders". |
801 | * The bottom level table contains the map for the smallest allocatable |
802 | * units of memory (here, pages), and each level above it describes |
803 | * pairs of units from the levels below, hence, "buddies". |
804 | * At a high level, all that happens here is marking the table entry |
805 | * at the bottom level available, and propagating the changes upward |
806 | * as necessary, plus some accounting needed to play nicely with other |
807 | * parts of the VM system. |
808 | * At each level, we keep a list of pages, which are heads of continuous |
809 | * free pages of length of (1 << order) and marked with _mapcount |
810 | * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page) |
811 | * field. |
812 | * So when we are allocating or freeing one, we can derive the state of the |
813 | * other. That is, if we allocate a small block, and both were |
814 | * free, the remainder of the region must be split into blocks. |
815 | * If a block is freed, and its buddy is also free, then this |
816 | * triggers coalescing into a block of larger size. |
817 | * |
818 | * -- nyc |
819 | */ |
820 | |
821 | static inline void __free_one_page(struct page *page, |
822 | unsigned long pfn, |
823 | struct zone *zone, unsigned int order, |
824 | int migratetype) |
825 | { |
826 | unsigned long page_idx; |
827 | unsigned long combined_idx; |
828 | unsigned long uninitialized_var(buddy_idx); |
829 | struct page *buddy; |
830 | unsigned int max_order; |
831 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
832 | int buddy_mg; |
833 | |
834 | migratetype = get_pageblock_migratetype(page); |
835 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
836 | |
837 | max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1); |
838 | |
839 | VM_BUG_ON(!zone_is_initialized(zone)); |
840 | VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); |
841 | |
842 | VM_BUG_ON(migratetype == -1); |
843 | if (likely(!is_migrate_isolate(migratetype))) |
844 | __mod_zone_freepage_state(zone, 1 << order, migratetype); |
845 | |
846 | page_idx = pfn & ((1 << MAX_ORDER) - 1); |
847 | |
848 | VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page); |
849 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
850 | |
851 | continue_merging: |
852 | while (order < max_order - 1) { |
853 | buddy_idx = __find_buddy_index(page_idx, order); |
854 | buddy = page + (buddy_idx - page_idx); |
855 | if (!page_is_buddy(page, buddy, order)) |
856 | goto done_merging; |
857 | /* |
858 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, |
859 | * merge with it and move up one order. |
860 | */ |
861 | if (page_is_guard(buddy)) { |
862 | clear_page_guard(zone, buddy, order, migratetype); |
863 | } else { |
864 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
865 | /* |
866 | * Kernel have provided some information about it in |
867 | * /proc/pagetypeinfo, /proc/buddyinfo. But both of them |
868 | * do not have a summary of free pages of each migrate |
869 | * type. |
870 | * Update zone free migrate type change according |
871 | * free_area's nr_free, this information is helpful and |
872 | * can be shown in echo m > /proc/sysrq-trigger. |
873 | */ |
874 | buddy_mg = get_pcppage_migratetype(buddy); |
875 | __mod_zone_migrate_state(zone, -(1 << order), buddy_mg); |
876 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
877 | list_del(&buddy->lru); |
878 | zone->free_area[order].nr_free--; |
879 | rmv_page_order(buddy); |
880 | } |
881 | combined_idx = buddy_idx & page_idx; |
882 | page = page + (combined_idx - page_idx); |
883 | page_idx = combined_idx; |
884 | order++; |
885 | } |
886 | if (max_order < MAX_ORDER) { |
887 | /* If we are here, it means order is >= pageblock_order. |
888 | * We want to prevent merge between freepages on isolate |
889 | * pageblock and normal pageblock. Without this, pageblock |
890 | * isolation could cause incorrect freepage or CMA accounting. |
891 | * |
892 | * We don't want to hit this code for the more frequent |
893 | * low-order merging. |
894 | */ |
895 | if (unlikely(has_isolate_pageblock(zone))) { |
896 | int buddy_mt; |
897 | |
898 | buddy_idx = __find_buddy_index(page_idx, order); |
899 | buddy = page + (buddy_idx - page_idx); |
900 | buddy_mt = get_pageblock_migratetype(buddy); |
901 | |
902 | if (migratetype != buddy_mt |
903 | && (is_migrate_isolate(migratetype) || |
904 | is_migrate_isolate(buddy_mt))) |
905 | goto done_merging; |
906 | } |
907 | max_order++; |
908 | goto continue_merging; |
909 | } |
910 | |
911 | done_merging: |
912 | set_page_order(page, order); |
913 | |
914 | /* |
915 | * If this is not the largest possible page, check if the buddy |
916 | * of the next-highest order is free. If it is, it's possible |
917 | * that pages are being freed that will coalesce soon. In case, |
918 | * that is happening, add the free page to the tail of the list |
919 | * so it's less likely to be used soon and more likely to be merged |
920 | * as a higher order page |
921 | */ |
922 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
923 | struct page *higher_page, *higher_buddy; |
924 | combined_idx = buddy_idx & page_idx; |
925 | higher_page = page + (combined_idx - page_idx); |
926 | buddy_idx = __find_buddy_index(combined_idx, order + 1); |
927 | higher_buddy = higher_page + (buddy_idx - combined_idx); |
928 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
929 | list_add_tail(&page->lru, |
930 | &zone->free_area[order].free_list[migratetype]); |
931 | goto out; |
932 | } |
933 | } |
934 | |
935 | #if defined(CONFIG_AMLOGIC_MEMORY_EXTEND) && defined(CONFIG_KASAN) |
936 | /* |
937 | * always put freed page to tail of buddy system, in |
938 | * order to increase probability of use-after-free |
939 | * for KASAN check. |
940 | */ |
941 | list_add_tail(&page->lru, |
942 | &zone->free_area[order].free_list[migratetype]); |
943 | #else |
944 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); |
945 | #endif |
946 | out: |
947 | zone->free_area[order].nr_free++; |
948 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
949 | set_pcppage_migratetype(page, migratetype); |
950 | __mod_zone_migrate_state(zone, (1 << order), migratetype); |
951 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
952 | } |
953 | |
954 | /* |
955 | * A bad page could be due to a number of fields. Instead of multiple branches, |
956 | * try and check multiple fields with one check. The caller must do a detailed |
957 | * check if necessary. |
958 | */ |
959 | static inline bool page_expected_state(struct page *page, |
960 | unsigned long check_flags) |
961 | { |
962 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
963 | return false; |
964 | |
965 | if (unlikely((unsigned long)page->mapping | |
966 | page_ref_count(page) | |
967 | #ifdef CONFIG_MEMCG |
968 | (unsigned long)page->mem_cgroup | |
969 | #endif |
970 | (page->flags & check_flags))) |
971 | return false; |
972 | |
973 | return true; |
974 | } |
975 | |
976 | static void free_pages_check_bad(struct page *page) |
977 | { |
978 | const char *bad_reason; |
979 | unsigned long bad_flags; |
980 | |
981 | bad_reason = NULL; |
982 | bad_flags = 0; |
983 | |
984 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
985 | bad_reason = "nonzero mapcount"; |
986 | if (unlikely(page->mapping != NULL)) |
987 | bad_reason = "non-NULL mapping"; |
988 | if (unlikely(page_ref_count(page) != 0)) |
989 | bad_reason = "nonzero _refcount"; |
990 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) { |
991 | bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set"; |
992 | bad_flags = PAGE_FLAGS_CHECK_AT_FREE; |
993 | } |
994 | #ifdef CONFIG_MEMCG |
995 | if (unlikely(page->mem_cgroup)) |
996 | bad_reason = "page still charged to cgroup"; |
997 | #endif |
998 | bad_page(page, bad_reason, bad_flags); |
999 | } |
1000 | |
1001 | static inline int free_pages_check(struct page *page) |
1002 | { |
1003 | if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE))) |
1004 | return 0; |
1005 | |
1006 | /* Something has gone sideways, find it */ |
1007 | free_pages_check_bad(page); |
1008 | return 1; |
1009 | } |
1010 | |
1011 | static int free_tail_pages_check(struct page *head_page, struct page *page) |
1012 | { |
1013 | int ret = 1; |
1014 | |
1015 | /* |
1016 | * We rely page->lru.next never has bit 0 set, unless the page |
1017 | * is PageTail(). Let's make sure that's true even for poisoned ->lru. |
1018 | */ |
1019 | BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1); |
1020 | |
1021 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) { |
1022 | ret = 0; |
1023 | goto out; |
1024 | } |
1025 | switch (page - head_page) { |
1026 | case 1: |
1027 | /* the first tail page: ->mapping is compound_mapcount() */ |
1028 | if (unlikely(compound_mapcount(page))) { |
1029 | bad_page(page, "nonzero compound_mapcount", 0); |
1030 | goto out; |
1031 | } |
1032 | break; |
1033 | case 2: |
1034 | /* |
1035 | * the second tail page: ->mapping is |
1036 | * page_deferred_list().next -- ignore value. |
1037 | */ |
1038 | break; |
1039 | default: |
1040 | if (page->mapping != TAIL_MAPPING) { |
1041 | bad_page(page, "corrupted mapping in tail page", 0); |
1042 | goto out; |
1043 | } |
1044 | break; |
1045 | } |
1046 | if (unlikely(!PageTail(page))) { |
1047 | bad_page(page, "PageTail not set", 0); |
1048 | goto out; |
1049 | } |
1050 | if (unlikely(compound_head(page) != head_page)) { |
1051 | bad_page(page, "compound_head not consistent", 0); |
1052 | goto out; |
1053 | } |
1054 | ret = 0; |
1055 | out: |
1056 | page->mapping = NULL; |
1057 | clear_compound_head(page); |
1058 | return ret; |
1059 | } |
1060 | |
1061 | static __always_inline bool free_pages_prepare(struct page *page, |
1062 | unsigned int order, bool check_free) |
1063 | { |
1064 | int bad = 0; |
1065 | |
1066 | VM_BUG_ON_PAGE(PageTail(page), page); |
1067 | |
1068 | trace_mm_page_free(page, order); |
1069 | kmemcheck_free_shadow(page, order); |
1070 | |
1071 | /* |
1072 | * Check tail pages before head page information is cleared to |
1073 | * avoid checking PageCompound for order-0 pages. |
1074 | */ |
1075 | if (unlikely(order)) { |
1076 | bool compound = PageCompound(page); |
1077 | int i; |
1078 | |
1079 | VM_BUG_ON_PAGE(compound && compound_order(page) != order, page); |
1080 | |
1081 | if (compound) |
1082 | ClearPageDoubleMap(page); |
1083 | for (i = 1; i < (1 << order); i++) { |
1084 | if (compound) |
1085 | bad += free_tail_pages_check(page, page + i); |
1086 | if (unlikely(free_pages_check(page + i))) { |
1087 | bad++; |
1088 | continue; |
1089 | } |
1090 | (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; |
1091 | } |
1092 | } |
1093 | if (PageMappingFlags(page)) |
1094 | page->mapping = NULL; |
1095 | if (memcg_kmem_enabled() && PageKmemcg(page)) |
1096 | memcg_kmem_uncharge(page, order); |
1097 | if (check_free) |
1098 | bad += free_pages_check(page); |
1099 | if (bad) |
1100 | return false; |
1101 | |
1102 | page_cpupid_reset_last(page); |
1103 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; |
1104 | reset_page_owner(page, order); |
1105 | |
1106 | if (!PageHighMem(page)) { |
1107 | debug_check_no_locks_freed(page_address(page), |
1108 | PAGE_SIZE << order); |
1109 | debug_check_no_obj_freed(page_address(page), |
1110 | PAGE_SIZE << order); |
1111 | } |
1112 | arch_free_page(page, order); |
1113 | kernel_poison_pages(page, 1 << order, 0); |
1114 | kernel_map_pages(page, 1 << order, 0); |
1115 | kasan_free_pages(page, order); |
1116 | #ifdef CONFIG_AMLOGIC_PAGE_TRACE |
1117 | reset_page_trace(page, order); |
1118 | #endif /* CONFIG_AMLOGIC_PAGE_TRACE */ |
1119 | |
1120 | return true; |
1121 | } |
1122 | |
1123 | #ifdef CONFIG_DEBUG_VM |
1124 | static inline bool free_pcp_prepare(struct page *page) |
1125 | { |
1126 | return free_pages_prepare(page, 0, true); |
1127 | } |
1128 | |
1129 | static inline bool bulkfree_pcp_prepare(struct page *page) |
1130 | { |
1131 | return false; |
1132 | } |
1133 | #else |
1134 | static bool free_pcp_prepare(struct page *page) |
1135 | { |
1136 | return free_pages_prepare(page, 0, false); |
1137 | } |
1138 | |
1139 | static bool bulkfree_pcp_prepare(struct page *page) |
1140 | { |
1141 | return free_pages_check(page); |
1142 | } |
1143 | #endif /* CONFIG_DEBUG_VM */ |
1144 | |
1145 | /* |
1146 | * Frees a number of pages from the PCP lists |
1147 | * Assumes all pages on list are in same zone, and of same order. |
1148 | * count is the number of pages to free. |
1149 | * |
1150 | * If the zone was previously in an "all pages pinned" state then look to |
1151 | * see if this freeing clears that state. |
1152 | * |
1153 | * And clear the zone's pages_scanned counter, to hold off the "all pages are |
1154 | * pinned" detection logic. |
1155 | */ |
1156 | static void free_pcppages_bulk(struct zone *zone, int count, |
1157 | struct per_cpu_pages *pcp) |
1158 | { |
1159 | int migratetype = 0; |
1160 | int batch_free = 0; |
1161 | unsigned long nr_scanned; |
1162 | bool isolated_pageblocks; |
1163 | |
1164 | spin_lock(&zone->lock); |
1165 | isolated_pageblocks = has_isolate_pageblock(zone); |
1166 | nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED); |
1167 | if (nr_scanned) |
1168 | __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned); |
1169 | |
1170 | while (count) { |
1171 | struct page *page; |
1172 | struct list_head *list; |
1173 | |
1174 | /* |
1175 | * Remove pages from lists in a round-robin fashion. A |
1176 | * batch_free count is maintained that is incremented when an |
1177 | * empty list is encountered. This is so more pages are freed |
1178 | * off fuller lists instead of spinning excessively around empty |
1179 | * lists |
1180 | */ |
1181 | do { |
1182 | batch_free++; |
1183 | if (++migratetype == MIGRATE_PCPTYPES) |
1184 | migratetype = 0; |
1185 | list = &pcp->lists[migratetype]; |
1186 | } while (list_empty(list)); |
1187 | |
1188 | /* This is the only non-empty list. Free them all. */ |
1189 | if (batch_free == MIGRATE_PCPTYPES) |
1190 | batch_free = count; |
1191 | |
1192 | do { |
1193 | int mt; /* migratetype of the to-be-freed page */ |
1194 | |
1195 | page = list_last_entry(list, struct page, lru); |
1196 | /* must delete as __free_one_page list manipulates */ |
1197 | list_del(&page->lru); |
1198 | |
1199 | mt = get_pcppage_migratetype(page); |
1200 | /* MIGRATE_ISOLATE page should not go to pcplists */ |
1201 | VM_BUG_ON_PAGE(is_migrate_isolate(mt), page); |
1202 | /* Pageblock could have been isolated meanwhile */ |
1203 | if (unlikely(isolated_pageblocks)) |
1204 | mt = get_pageblock_migratetype(page); |
1205 | |
1206 | if (bulkfree_pcp_prepare(page)) |
1207 | continue; |
1208 | |
1209 | __free_one_page(page, page_to_pfn(page), zone, 0, mt); |
1210 | trace_mm_page_pcpu_drain(page, 0, mt); |
1211 | } while (--count && --batch_free && !list_empty(list)); |
1212 | } |
1213 | spin_unlock(&zone->lock); |
1214 | } |
1215 | |
1216 | static void free_one_page(struct zone *zone, |
1217 | struct page *page, unsigned long pfn, |
1218 | unsigned int order, |
1219 | int migratetype) |
1220 | { |
1221 | unsigned long nr_scanned; |
1222 | spin_lock(&zone->lock); |
1223 | nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED); |
1224 | if (nr_scanned) |
1225 | __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned); |
1226 | |
1227 | if (unlikely(has_isolate_pageblock(zone) || |
1228 | is_migrate_isolate(migratetype))) { |
1229 | migratetype = get_pfnblock_migratetype(page, pfn); |
1230 | } |
1231 | __free_one_page(page, pfn, zone, order, migratetype); |
1232 | spin_unlock(&zone->lock); |
1233 | } |
1234 | |
1235 | static void __meminit __init_single_page(struct page *page, unsigned long pfn, |
1236 | unsigned long zone, int nid) |
1237 | { |
1238 | set_page_links(page, zone, nid, pfn); |
1239 | init_page_count(page); |
1240 | page_mapcount_reset(page); |
1241 | page_cpupid_reset_last(page); |
1242 | |
1243 | INIT_LIST_HEAD(&page->lru); |
1244 | #ifdef WANT_PAGE_VIRTUAL |
1245 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ |
1246 | if (!is_highmem_idx(zone)) |
1247 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1248 | #endif |
1249 | } |
1250 | |
1251 | static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone, |
1252 | int nid) |
1253 | { |
1254 | return __init_single_page(pfn_to_page(pfn), pfn, zone, nid); |
1255 | } |
1256 | |
1257 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1258 | static void init_reserved_page(unsigned long pfn) |
1259 | { |
1260 | pg_data_t *pgdat; |
1261 | int nid, zid; |
1262 | |
1263 | if (!early_page_uninitialised(pfn)) |
1264 | return; |
1265 | |
1266 | nid = early_pfn_to_nid(pfn); |
1267 | pgdat = NODE_DATA(nid); |
1268 | |
1269 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
1270 | struct zone *zone = &pgdat->node_zones[zid]; |
1271 | |
1272 | if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone)) |
1273 | break; |
1274 | } |
1275 | __init_single_pfn(pfn, zid, nid); |
1276 | } |
1277 | #else |
1278 | static inline void init_reserved_page(unsigned long pfn) |
1279 | { |
1280 | } |
1281 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
1282 | |
1283 | /* |
1284 | * Initialised pages do not have PageReserved set. This function is |
1285 | * called for each range allocated by the bootmem allocator and |
1286 | * marks the pages PageReserved. The remaining valid pages are later |
1287 | * sent to the buddy page allocator. |
1288 | */ |
1289 | void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end) |
1290 | { |
1291 | unsigned long start_pfn = PFN_DOWN(start); |
1292 | unsigned long end_pfn = PFN_UP(end); |
1293 | |
1294 | for (; start_pfn < end_pfn; start_pfn++) { |
1295 | if (pfn_valid(start_pfn)) { |
1296 | struct page *page = pfn_to_page(start_pfn); |
1297 | |
1298 | init_reserved_page(start_pfn); |
1299 | |
1300 | /* Avoid false-positive PageTail() */ |
1301 | INIT_LIST_HEAD(&page->lru); |
1302 | |
1303 | SetPageReserved(page); |
1304 | } |
1305 | } |
1306 | } |
1307 | |
1308 | static void __free_pages_ok(struct page *page, unsigned int order) |
1309 | { |
1310 | unsigned long flags; |
1311 | int migratetype; |
1312 | unsigned long pfn = page_to_pfn(page); |
1313 | |
1314 | if (!free_pages_prepare(page, order, true)) |
1315 | return; |
1316 | |
1317 | migratetype = get_pfnblock_migratetype(page, pfn); |
1318 | local_irq_save(flags); |
1319 | __count_vm_events(PGFREE, 1 << order); |
1320 | free_one_page(page_zone(page), page, pfn, order, migratetype); |
1321 | local_irq_restore(flags); |
1322 | } |
1323 | |
1324 | static void __init __free_pages_boot_core(struct page *page, unsigned int order) |
1325 | { |
1326 | unsigned int nr_pages = 1 << order; |
1327 | struct page *p = page; |
1328 | unsigned int loop; |
1329 | |
1330 | prefetchw(p); |
1331 | for (loop = 0; loop < (nr_pages - 1); loop++, p++) { |
1332 | prefetchw(p + 1); |
1333 | __ClearPageReserved(p); |
1334 | set_page_count(p, 0); |
1335 | } |
1336 | __ClearPageReserved(p); |
1337 | set_page_count(p, 0); |
1338 | |
1339 | page_zone(page)->managed_pages += nr_pages; |
1340 | set_page_refcounted(page); |
1341 | __free_pages(page, order); |
1342 | } |
1343 | |
1344 | #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \ |
1345 | defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) |
1346 | |
1347 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata; |
1348 | |
1349 | int __meminit early_pfn_to_nid(unsigned long pfn) |
1350 | { |
1351 | static DEFINE_SPINLOCK(early_pfn_lock); |
1352 | int nid; |
1353 | |
1354 | spin_lock(&early_pfn_lock); |
1355 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache); |
1356 | if (nid < 0) |
1357 | nid = first_online_node; |
1358 | spin_unlock(&early_pfn_lock); |
1359 | |
1360 | return nid; |
1361 | } |
1362 | #endif |
1363 | |
1364 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
1365 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, |
1366 | struct mminit_pfnnid_cache *state) |
1367 | { |
1368 | int nid; |
1369 | |
1370 | nid = __early_pfn_to_nid(pfn, state); |
1371 | if (nid >= 0 && nid != node) |
1372 | return false; |
1373 | return true; |
1374 | } |
1375 | |
1376 | /* Only safe to use early in boot when initialisation is single-threaded */ |
1377 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
1378 | { |
1379 | return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache); |
1380 | } |
1381 | |
1382 | #else |
1383 | |
1384 | static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
1385 | { |
1386 | return true; |
1387 | } |
1388 | static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node, |
1389 | struct mminit_pfnnid_cache *state) |
1390 | { |
1391 | return true; |
1392 | } |
1393 | #endif |
1394 | |
1395 | |
1396 | void __init __free_pages_bootmem(struct page *page, unsigned long pfn, |
1397 | unsigned int order) |
1398 | { |
1399 | if (early_page_uninitialised(pfn)) |
1400 | return; |
1401 | return __free_pages_boot_core(page, order); |
1402 | } |
1403 | |
1404 | /* |
1405 | * Check that the whole (or subset of) a pageblock given by the interval of |
1406 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it |
1407 | * with the migration of free compaction scanner. The scanners then need to |
1408 | * use only pfn_valid_within() check for arches that allow holes within |
1409 | * pageblocks. |
1410 | * |
1411 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. |
1412 | * |
1413 | * It's possible on some configurations to have a setup like node0 node1 node0 |
1414 | * i.e. it's possible that all pages within a zones range of pages do not |
1415 | * belong to a single zone. We assume that a border between node0 and node1 |
1416 | * can occur within a single pageblock, but not a node0 node1 node0 |
1417 | * interleaving within a single pageblock. It is therefore sufficient to check |
1418 | * the first and last page of a pageblock and avoid checking each individual |
1419 | * page in a pageblock. |
1420 | */ |
1421 | struct page *__pageblock_pfn_to_page(unsigned long start_pfn, |
1422 | unsigned long end_pfn, struct zone *zone) |
1423 | { |
1424 | struct page *start_page; |
1425 | struct page *end_page; |
1426 | |
1427 | /* end_pfn is one past the range we are checking */ |
1428 | end_pfn--; |
1429 | |
1430 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) |
1431 | return NULL; |
1432 | |
1433 | start_page = pfn_to_page(start_pfn); |
1434 | |
1435 | if (page_zone(start_page) != zone) |
1436 | return NULL; |
1437 | |
1438 | end_page = pfn_to_page(end_pfn); |
1439 | |
1440 | /* This gives a shorter code than deriving page_zone(end_page) */ |
1441 | if (page_zone_id(start_page) != page_zone_id(end_page)) |
1442 | return NULL; |
1443 | |
1444 | return start_page; |
1445 | } |
1446 | |
1447 | void set_zone_contiguous(struct zone *zone) |
1448 | { |
1449 | unsigned long block_start_pfn = zone->zone_start_pfn; |
1450 | unsigned long block_end_pfn; |
1451 | |
1452 | block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages); |
1453 | for (; block_start_pfn < zone_end_pfn(zone); |
1454 | block_start_pfn = block_end_pfn, |
1455 | block_end_pfn += pageblock_nr_pages) { |
1456 | |
1457 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone)); |
1458 | |
1459 | if (!__pageblock_pfn_to_page(block_start_pfn, |
1460 | block_end_pfn, zone)) |
1461 | return; |
1462 | } |
1463 | |
1464 | /* We confirm that there is no hole */ |
1465 | zone->contiguous = true; |
1466 | } |
1467 | |
1468 | void clear_zone_contiguous(struct zone *zone) |
1469 | { |
1470 | zone->contiguous = false; |
1471 | } |
1472 | |
1473 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1474 | static void __init deferred_free_range(struct page *page, |
1475 | unsigned long pfn, int nr_pages) |
1476 | { |
1477 | int i; |
1478 | |
1479 | if (!page) |
1480 | return; |
1481 | |
1482 | /* Free a large naturally-aligned chunk if possible */ |
1483 | if (nr_pages == pageblock_nr_pages && |
1484 | (pfn & (pageblock_nr_pages - 1)) == 0) { |
1485 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
1486 | __free_pages_boot_core(page, pageblock_order); |
1487 | return; |
1488 | } |
1489 | |
1490 | for (i = 0; i < nr_pages; i++, page++, pfn++) { |
1491 | if ((pfn & (pageblock_nr_pages - 1)) == 0) |
1492 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
1493 | __free_pages_boot_core(page, 0); |
1494 | } |
1495 | } |
1496 | |
1497 | /* Completion tracking for deferred_init_memmap() threads */ |
1498 | static atomic_t pgdat_init_n_undone __initdata; |
1499 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp); |
1500 | |
1501 | static inline void __init pgdat_init_report_one_done(void) |
1502 | { |
1503 | if (atomic_dec_and_test(&pgdat_init_n_undone)) |
1504 | complete(&pgdat_init_all_done_comp); |
1505 | } |
1506 | |
1507 | /* Initialise remaining memory on a node */ |
1508 | static int __init deferred_init_memmap(void *data) |
1509 | { |
1510 | pg_data_t *pgdat = data; |
1511 | int nid = pgdat->node_id; |
1512 | struct mminit_pfnnid_cache nid_init_state = { }; |
1513 | unsigned long start = jiffies; |
1514 | unsigned long nr_pages = 0; |
1515 | unsigned long walk_start, walk_end; |
1516 | int i, zid; |
1517 | struct zone *zone; |
1518 | unsigned long first_init_pfn = pgdat->first_deferred_pfn; |
1519 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
1520 | |
1521 | if (first_init_pfn == ULONG_MAX) { |
1522 | pgdat_init_report_one_done(); |
1523 | return 0; |
1524 | } |
1525 | |
1526 | /* Bind memory initialisation thread to a local node if possible */ |
1527 | if (!cpumask_empty(cpumask)) |
1528 | set_cpus_allowed_ptr(current, cpumask); |
1529 | |
1530 | /* Sanity check boundaries */ |
1531 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn); |
1532 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat)); |
1533 | pgdat->first_deferred_pfn = ULONG_MAX; |
1534 | |
1535 | /* Only the highest zone is deferred so find it */ |
1536 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
1537 | zone = pgdat->node_zones + zid; |
1538 | if (first_init_pfn < zone_end_pfn(zone)) |
1539 | break; |
1540 | } |
1541 | |
1542 | for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) { |
1543 | unsigned long pfn, end_pfn; |
1544 | struct page *page = NULL; |
1545 | struct page *free_base_page = NULL; |
1546 | unsigned long free_base_pfn = 0; |
1547 | int nr_to_free = 0; |
1548 | |
1549 | end_pfn = min(walk_end, zone_end_pfn(zone)); |
1550 | pfn = first_init_pfn; |
1551 | if (pfn < walk_start) |
1552 | pfn = walk_start; |
1553 | if (pfn < zone->zone_start_pfn) |
1554 | pfn = zone->zone_start_pfn; |
1555 | |
1556 | for (; pfn < end_pfn; pfn++) { |
1557 | if (!pfn_valid_within(pfn)) |
1558 | goto free_range; |
1559 | |
1560 | /* |
1561 | * Ensure pfn_valid is checked every |
1562 | * pageblock_nr_pages for memory holes |
1563 | */ |
1564 | if ((pfn & (pageblock_nr_pages - 1)) == 0) { |
1565 | if (!pfn_valid(pfn)) { |
1566 | page = NULL; |
1567 | goto free_range; |
1568 | } |
1569 | } |
1570 | |
1571 | if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) { |
1572 | page = NULL; |
1573 | goto free_range; |
1574 | } |
1575 | |
1576 | /* Minimise pfn page lookups and scheduler checks */ |
1577 | if (page && (pfn & (pageblock_nr_pages - 1)) != 0) { |
1578 | page++; |
1579 | } else { |
1580 | nr_pages += nr_to_free; |
1581 | deferred_free_range(free_base_page, |
1582 | free_base_pfn, nr_to_free); |
1583 | free_base_page = NULL; |
1584 | free_base_pfn = nr_to_free = 0; |
1585 | |
1586 | page = pfn_to_page(pfn); |
1587 | cond_resched(); |
1588 | } |
1589 | |
1590 | if (page->flags) { |
1591 | VM_BUG_ON(page_zone(page) != zone); |
1592 | goto free_range; |
1593 | } |
1594 | |
1595 | __init_single_page(page, pfn, zid, nid); |
1596 | if (!free_base_page) { |
1597 | free_base_page = page; |
1598 | free_base_pfn = pfn; |
1599 | nr_to_free = 0; |
1600 | } |
1601 | nr_to_free++; |
1602 | |
1603 | /* Where possible, batch up pages for a single free */ |
1604 | continue; |
1605 | free_range: |
1606 | /* Free the current block of pages to allocator */ |
1607 | nr_pages += nr_to_free; |
1608 | deferred_free_range(free_base_page, free_base_pfn, |
1609 | nr_to_free); |
1610 | free_base_page = NULL; |
1611 | free_base_pfn = nr_to_free = 0; |
1612 | } |
1613 | /* Free the last block of pages to allocator */ |
1614 | nr_pages += nr_to_free; |
1615 | deferred_free_range(free_base_page, free_base_pfn, nr_to_free); |
1616 | |
1617 | first_init_pfn = max(end_pfn, first_init_pfn); |
1618 | } |
1619 | |
1620 | /* Sanity check that the next zone really is unpopulated */ |
1621 | WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone)); |
1622 | |
1623 | pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages, |
1624 | jiffies_to_msecs(jiffies - start)); |
1625 | |
1626 | pgdat_init_report_one_done(); |
1627 | return 0; |
1628 | } |
1629 | #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ |
1630 | |
1631 | void __init page_alloc_init_late(void) |
1632 | { |
1633 | struct zone *zone; |
1634 | |
1635 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT |
1636 | int nid; |
1637 | |
1638 | /* There will be num_node_state(N_MEMORY) threads */ |
1639 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY)); |
1640 | for_each_node_state(nid, N_MEMORY) { |
1641 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid); |
1642 | } |
1643 | |
1644 | /* Block until all are initialised */ |
1645 | wait_for_completion(&pgdat_init_all_done_comp); |
1646 | |
1647 | /* Reinit limits that are based on free pages after the kernel is up */ |
1648 | files_maxfiles_init(); |
1649 | #endif |
1650 | #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK |
1651 | /* Discard memblock private memory */ |
1652 | memblock_discard(); |
1653 | #endif |
1654 | |
1655 | for_each_populated_zone(zone) |
1656 | set_zone_contiguous(zone); |
1657 | } |
1658 | |
1659 | #ifdef CONFIG_CMA |
1660 | /* Free whole pageblock and set its migration type to MIGRATE_CMA. */ |
1661 | void __init init_cma_reserved_pageblock(struct page *page) |
1662 | { |
1663 | unsigned i = pageblock_nr_pages; |
1664 | struct page *p = page; |
1665 | |
1666 | do { |
1667 | __ClearPageReserved(p); |
1668 | set_page_count(p, 0); |
1669 | } while (++p, --i); |
1670 | |
1671 | set_pageblock_migratetype(page, MIGRATE_CMA); |
1672 | |
1673 | if (pageblock_order >= MAX_ORDER) { |
1674 | i = pageblock_nr_pages; |
1675 | p = page; |
1676 | do { |
1677 | set_page_refcounted(p); |
1678 | __free_pages(p, MAX_ORDER - 1); |
1679 | p += MAX_ORDER_NR_PAGES; |
1680 | } while (i -= MAX_ORDER_NR_PAGES); |
1681 | } else { |
1682 | set_page_refcounted(page); |
1683 | __free_pages(page, pageblock_order); |
1684 | } |
1685 | |
1686 | adjust_managed_page_count(page, pageblock_nr_pages); |
1687 | } |
1688 | #endif |
1689 | |
1690 | /* |
1691 | * The order of subdivision here is critical for the IO subsystem. |
1692 | * Please do not alter this order without good reasons and regression |
1693 | * testing. Specifically, as large blocks of memory are subdivided, |
1694 | * the order in which smaller blocks are delivered depends on the order |
1695 | * they're subdivided in this function. This is the primary factor |
1696 | * influencing the order in which pages are delivered to the IO |
1697 | * subsystem according to empirical testing, and this is also justified |
1698 | * by considering the behavior of a buddy system containing a single |
1699 | * large block of memory acted on by a series of small allocations. |
1700 | * This behavior is a critical factor in sglist merging's success. |
1701 | * |
1702 | * -- nyc |
1703 | */ |
1704 | static inline void expand(struct zone *zone, struct page *page, |
1705 | int low, int high, struct free_area *area, |
1706 | int migratetype) |
1707 | { |
1708 | unsigned long size = 1 << high; |
1709 | |
1710 | while (high > low) { |
1711 | area--; |
1712 | high--; |
1713 | size >>= 1; |
1714 | VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]); |
1715 | |
1716 | /* |
1717 | * Mark as guard pages (or page), that will allow to |
1718 | * merge back to allocator when buddy will be freed. |
1719 | * Corresponding page table entries will not be touched, |
1720 | * pages will stay not present in virtual address space |
1721 | */ |
1722 | if (set_page_guard(zone, &page[size], high, migratetype)) |
1723 | continue; |
1724 | |
1725 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1726 | area->nr_free++; |
1727 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
1728 | set_pcppage_migratetype(&page[size], migratetype); |
1729 | __mod_zone_migrate_state(zone, (1 << high), migratetype); |
1730 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
1731 | set_page_order(&page[size], high); |
1732 | } |
1733 | } |
1734 | |
1735 | static void check_new_page_bad(struct page *page) |
1736 | { |
1737 | const char *bad_reason = NULL; |
1738 | unsigned long bad_flags = 0; |
1739 | |
1740 | if (unlikely(atomic_read(&page->_mapcount) != -1)) |
1741 | bad_reason = "nonzero mapcount"; |
1742 | if (unlikely(page->mapping != NULL)) |
1743 | bad_reason = "non-NULL mapping"; |
1744 | if (unlikely(page_ref_count(page) != 0)) |
1745 | bad_reason = "nonzero _count"; |
1746 | if (unlikely(page->flags & __PG_HWPOISON)) { |
1747 | bad_reason = "HWPoisoned (hardware-corrupted)"; |
1748 | bad_flags = __PG_HWPOISON; |
1749 | /* Don't complain about hwpoisoned pages */ |
1750 | page_mapcount_reset(page); /* remove PageBuddy */ |
1751 | return; |
1752 | } |
1753 | if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) { |
1754 | bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set"; |
1755 | bad_flags = PAGE_FLAGS_CHECK_AT_PREP; |
1756 | } |
1757 | #ifdef CONFIG_MEMCG |
1758 | if (unlikely(page->mem_cgroup)) |
1759 | bad_reason = "page still charged to cgroup"; |
1760 | #endif |
1761 | bad_page(page, bad_reason, bad_flags); |
1762 | } |
1763 | |
1764 | /* |
1765 | * This page is about to be returned from the page allocator |
1766 | */ |
1767 | static inline int check_new_page(struct page *page) |
1768 | { |
1769 | if (likely(page_expected_state(page, |
1770 | PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))) |
1771 | return 0; |
1772 | |
1773 | check_new_page_bad(page); |
1774 | return 1; |
1775 | } |
1776 | |
1777 | static inline bool free_pages_prezeroed(bool poisoned) |
1778 | { |
1779 | return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && |
1780 | page_poisoning_enabled() && poisoned; |
1781 | } |
1782 | |
1783 | #ifdef CONFIG_DEBUG_VM |
1784 | static bool check_pcp_refill(struct page *page) |
1785 | { |
1786 | return false; |
1787 | } |
1788 | |
1789 | static bool check_new_pcp(struct page *page) |
1790 | { |
1791 | return check_new_page(page); |
1792 | } |
1793 | #else |
1794 | static bool check_pcp_refill(struct page *page) |
1795 | { |
1796 | return check_new_page(page); |
1797 | } |
1798 | static bool check_new_pcp(struct page *page) |
1799 | { |
1800 | return false; |
1801 | } |
1802 | #endif /* CONFIG_DEBUG_VM */ |
1803 | |
1804 | static bool check_new_pages(struct page *page, unsigned int order) |
1805 | { |
1806 | int i; |
1807 | for (i = 0; i < (1 << order); i++) { |
1808 | struct page *p = page + i; |
1809 | |
1810 | if (unlikely(check_new_page(p))) |
1811 | return true; |
1812 | } |
1813 | |
1814 | return false; |
1815 | } |
1816 | |
1817 | inline void post_alloc_hook(struct page *page, unsigned int order, |
1818 | gfp_t gfp_flags) |
1819 | { |
1820 | set_page_private(page, 0); |
1821 | set_page_refcounted(page); |
1822 | |
1823 | arch_alloc_page(page, order); |
1824 | kernel_map_pages(page, 1 << order, 1); |
1825 | kernel_poison_pages(page, 1 << order, 1); |
1826 | kasan_alloc_pages(page, order); |
1827 | set_page_owner(page, order, gfp_flags); |
1828 | } |
1829 | |
1830 | static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags, |
1831 | unsigned int alloc_flags) |
1832 | { |
1833 | int i; |
1834 | bool poisoned = true; |
1835 | |
1836 | for (i = 0; i < (1 << order); i++) { |
1837 | struct page *p = page + i; |
1838 | if (poisoned) |
1839 | poisoned &= page_is_poisoned(p); |
1840 | } |
1841 | |
1842 | post_alloc_hook(page, order, gfp_flags); |
1843 | |
1844 | if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) |
1845 | for (i = 0; i < (1 << order); i++) |
1846 | clear_highpage(page + i); |
1847 | |
1848 | if (order && (gfp_flags & __GFP_COMP)) |
1849 | prep_compound_page(page, order); |
1850 | |
1851 | /* |
1852 | * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to |
1853 | * allocate the page. The expectation is that the caller is taking |
1854 | * steps that will free more memory. The caller should avoid the page |
1855 | * being used for !PFMEMALLOC purposes. |
1856 | */ |
1857 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
1858 | set_page_pfmemalloc(page); |
1859 | else |
1860 | clear_page_pfmemalloc(page); |
1861 | } |
1862 | |
1863 | /* |
1864 | * Go through the free lists for the given migratetype and remove |
1865 | * the smallest available page from the freelists |
1866 | */ |
1867 | static inline |
1868 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
1869 | int migratetype) |
1870 | { |
1871 | unsigned int current_order; |
1872 | struct free_area *area; |
1873 | struct page *page; |
1874 | |
1875 | /* Find a page of the appropriate size in the preferred list */ |
1876 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { |
1877 | area = &(zone->free_area[current_order]); |
1878 | page = list_first_entry_or_null(&area->free_list[migratetype], |
1879 | struct page, lru); |
1880 | if (!page) |
1881 | continue; |
1882 | list_del(&page->lru); |
1883 | rmv_page_order(page); |
1884 | area->nr_free--; |
1885 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
1886 | __mod_zone_migrate_state(zone, -(1 << current_order), |
1887 | migratetype); |
1888 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
1889 | expand(zone, page, order, current_order, area, migratetype); |
1890 | set_pcppage_migratetype(page, migratetype); |
1891 | return page; |
1892 | } |
1893 | |
1894 | return NULL; |
1895 | } |
1896 | |
1897 | |
1898 | /* |
1899 | * This array describes the order lists are fallen back to when |
1900 | * the free lists for the desirable migrate type are depleted |
1901 | */ |
1902 | static int fallbacks[MIGRATE_TYPES][4] = { |
1903 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1904 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES }, |
1905 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES }, |
1906 | #ifdef CONFIG_CMA |
1907 | [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */ |
1908 | #endif |
1909 | #ifdef CONFIG_MEMORY_ISOLATION |
1910 | [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */ |
1911 | #endif |
1912 | }; |
1913 | |
1914 | #ifdef CONFIG_CMA |
1915 | static struct page *__rmqueue_cma_fallback(struct zone *zone, |
1916 | unsigned int order) |
1917 | { |
1918 | return __rmqueue_smallest(zone, order, MIGRATE_CMA); |
1919 | } |
1920 | #else |
1921 | static inline struct page *__rmqueue_cma_fallback(struct zone *zone, |
1922 | unsigned int order) { return NULL; } |
1923 | #endif |
1924 | |
1925 | /* |
1926 | * Move the free pages in a range to the free lists of the requested type. |
1927 | * Note that start_page and end_pages are not aligned on a pageblock |
1928 | * boundary. If alignment is required, use move_freepages_block() |
1929 | */ |
1930 | int move_freepages(struct zone *zone, |
1931 | struct page *start_page, struct page *end_page, |
1932 | int migratetype) |
1933 | { |
1934 | struct page *page; |
1935 | unsigned int order; |
1936 | int pages_moved = 0; |
1937 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
1938 | int list_type; |
1939 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
1940 | |
1941 | #ifndef CONFIG_HOLES_IN_ZONE |
1942 | /* |
1943 | * page_zone is not safe to call in this context when |
1944 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant |
1945 | * anyway as we check zone boundaries in move_freepages_block(). |
1946 | * Remove at a later date when no bug reports exist related to |
1947 | * grouping pages by mobility |
1948 | */ |
1949 | VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); |
1950 | #endif |
1951 | |
1952 | for (page = start_page; page <= end_page;) { |
1953 | if (!pfn_valid_within(page_to_pfn(page))) { |
1954 | page++; |
1955 | continue; |
1956 | } |
1957 | |
1958 | /* Make sure we are not inadvertently changing nodes */ |
1959 | VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); |
1960 | |
1961 | if (!PageBuddy(page)) { |
1962 | page++; |
1963 | continue; |
1964 | } |
1965 | |
1966 | order = page_order(page); |
1967 | list_move(&page->lru, |
1968 | &zone->free_area[order].free_list[migratetype]); |
1969 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
1970 | list_type = get_pcppage_migratetype(page); |
1971 | __mod_zone_migrate_state(zone, -(1 << order), list_type); |
1972 | __mod_zone_migrate_state(zone, (1 << order), migratetype); |
1973 | set_pcppage_migratetype(page, migratetype); |
1974 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
1975 | page += 1 << order; |
1976 | pages_moved += 1 << order; |
1977 | } |
1978 | |
1979 | return pages_moved; |
1980 | } |
1981 | |
1982 | int move_freepages_block(struct zone *zone, struct page *page, |
1983 | int migratetype) |
1984 | { |
1985 | unsigned long start_pfn, end_pfn; |
1986 | struct page *start_page, *end_page; |
1987 | |
1988 | start_pfn = page_to_pfn(page); |
1989 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
1990 | start_page = pfn_to_page(start_pfn); |
1991 | end_page = start_page + pageblock_nr_pages - 1; |
1992 | end_pfn = start_pfn + pageblock_nr_pages - 1; |
1993 | |
1994 | /* Do not cross zone boundaries */ |
1995 | if (!zone_spans_pfn(zone, start_pfn)) |
1996 | start_page = page; |
1997 | if (!zone_spans_pfn(zone, end_pfn)) |
1998 | return 0; |
1999 | |
2000 | return move_freepages(zone, start_page, end_page, migratetype); |
2001 | } |
2002 | |
2003 | static void change_pageblock_range(struct page *pageblock_page, |
2004 | int start_order, int migratetype) |
2005 | { |
2006 | int nr_pageblocks = 1 << (start_order - pageblock_order); |
2007 | |
2008 | while (nr_pageblocks--) { |
2009 | set_pageblock_migratetype(pageblock_page, migratetype); |
2010 | pageblock_page += pageblock_nr_pages; |
2011 | } |
2012 | } |
2013 | |
2014 | /* |
2015 | * When we are falling back to another migratetype during allocation, try to |
2016 | * steal extra free pages from the same pageblocks to satisfy further |
2017 | * allocations, instead of polluting multiple pageblocks. |
2018 | * |
2019 | * If we are stealing a relatively large buddy page, it is likely there will |
2020 | * be more free pages in the pageblock, so try to steal them all. For |
2021 | * reclaimable and unmovable allocations, we steal regardless of page size, |
2022 | * as fragmentation caused by those allocations polluting movable pageblocks |
2023 | * is worse than movable allocations stealing from unmovable and reclaimable |
2024 | * pageblocks. |
2025 | */ |
2026 | static bool can_steal_fallback(unsigned int order, int start_mt) |
2027 | { |
2028 | /* |
2029 | * Leaving this order check is intended, although there is |
2030 | * relaxed order check in next check. The reason is that |
2031 | * we can actually steal whole pageblock if this condition met, |
2032 | * but, below check doesn't guarantee it and that is just heuristic |
2033 | * so could be changed anytime. |
2034 | */ |
2035 | if (order >= pageblock_order) |
2036 | return true; |
2037 | |
2038 | if (order >= pageblock_order / 2 || |
2039 | start_mt == MIGRATE_RECLAIMABLE || |
2040 | start_mt == MIGRATE_UNMOVABLE || |
2041 | page_group_by_mobility_disabled) |
2042 | return true; |
2043 | |
2044 | return false; |
2045 | } |
2046 | |
2047 | /* |
2048 | * This function implements actual steal behaviour. If order is large enough, |
2049 | * we can steal whole pageblock. If not, we first move freepages in this |
2050 | * pageblock and check whether half of pages are moved or not. If half of |
2051 | * pages are moved, we can change migratetype of pageblock and permanently |
2052 | * use it's pages as requested migratetype in the future. |
2053 | */ |
2054 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
2055 | static void steal_suitable_fallback(struct zone *zone, struct page *page, |
2056 | int start_type, int *list_type) |
2057 | #else |
2058 | static void steal_suitable_fallback(struct zone *zone, struct page *page, |
2059 | int start_type) |
2060 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
2061 | { |
2062 | unsigned int current_order = page_order(page); |
2063 | int pages; |
2064 | |
2065 | /* Take ownership for orders >= pageblock_order */ |
2066 | if (current_order >= pageblock_order) { |
2067 | change_pageblock_range(page, current_order, start_type); |
2068 | return; |
2069 | } |
2070 | |
2071 | pages = move_freepages_block(zone, page, start_type); |
2072 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
2073 | *list_type = start_type; |
2074 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
2075 | |
2076 | /* Claim the whole block if over half of it is free */ |
2077 | if (pages >= (1 << (pageblock_order-1)) || |
2078 | page_group_by_mobility_disabled) |
2079 | set_pageblock_migratetype(page, start_type); |
2080 | } |
2081 | |
2082 | /* |
2083 | * Check whether there is a suitable fallback freepage with requested order. |
2084 | * If only_stealable is true, this function returns fallback_mt only if |
2085 | * we can steal other freepages all together. This would help to reduce |
2086 | * fragmentation due to mixed migratetype pages in one pageblock. |
2087 | */ |
2088 | int find_suitable_fallback(struct free_area *area, unsigned int order, |
2089 | int migratetype, bool only_stealable, bool *can_steal) |
2090 | { |
2091 | int i; |
2092 | int fallback_mt; |
2093 | |
2094 | if (area->nr_free == 0) |
2095 | return -1; |
2096 | |
2097 | *can_steal = false; |
2098 | for (i = 0;; i++) { |
2099 | fallback_mt = fallbacks[migratetype][i]; |
2100 | if (fallback_mt == MIGRATE_TYPES) |
2101 | break; |
2102 | |
2103 | if (list_empty(&area->free_list[fallback_mt])) |
2104 | continue; |
2105 | |
2106 | if (can_steal_fallback(order, migratetype)) |
2107 | *can_steal = true; |
2108 | |
2109 | if (!only_stealable) |
2110 | return fallback_mt; |
2111 | |
2112 | if (*can_steal) |
2113 | return fallback_mt; |
2114 | } |
2115 | |
2116 | return -1; |
2117 | } |
2118 | |
2119 | /* |
2120 | * Reserve a pageblock for exclusive use of high-order atomic allocations if |
2121 | * there are no empty page blocks that contain a page with a suitable order |
2122 | */ |
2123 | static void reserve_highatomic_pageblock(struct page *page, struct zone *zone, |
2124 | unsigned int alloc_order) |
2125 | { |
2126 | int mt; |
2127 | unsigned long max_managed, flags; |
2128 | |
2129 | /* |
2130 | * Limit the number reserved to 1 pageblock or roughly 1% of a zone. |
2131 | * Check is race-prone but harmless. |
2132 | */ |
2133 | max_managed = (zone->managed_pages / 100) + pageblock_nr_pages; |
2134 | if (zone->nr_reserved_highatomic >= max_managed) |
2135 | return; |
2136 | |
2137 | spin_lock_irqsave(&zone->lock, flags); |
2138 | |
2139 | /* Recheck the nr_reserved_highatomic limit under the lock */ |
2140 | if (zone->nr_reserved_highatomic >= max_managed) |
2141 | goto out_unlock; |
2142 | |
2143 | /* Yoink! */ |
2144 | mt = get_pageblock_migratetype(page); |
2145 | if (mt != MIGRATE_HIGHATOMIC && |
2146 | !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { |
2147 | zone->nr_reserved_highatomic += pageblock_nr_pages; |
2148 | set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); |
2149 | move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); |
2150 | } |
2151 | |
2152 | out_unlock: |
2153 | spin_unlock_irqrestore(&zone->lock, flags); |
2154 | } |
2155 | |
2156 | /* |
2157 | * Used when an allocation is about to fail under memory pressure. This |
2158 | * potentially hurts the reliability of high-order allocations when under |
2159 | * intense memory pressure but failed atomic allocations should be easier |
2160 | * to recover from than an OOM. |
2161 | */ |
2162 | static void unreserve_highatomic_pageblock(const struct alloc_context *ac) |
2163 | { |
2164 | struct zonelist *zonelist = ac->zonelist; |
2165 | unsigned long flags; |
2166 | struct zoneref *z; |
2167 | struct zone *zone; |
2168 | struct page *page; |
2169 | int order; |
2170 | |
2171 | for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx, |
2172 | ac->nodemask) { |
2173 | /* Preserve at least one pageblock */ |
2174 | if (zone->nr_reserved_highatomic <= pageblock_nr_pages) |
2175 | continue; |
2176 | |
2177 | spin_lock_irqsave(&zone->lock, flags); |
2178 | for (order = 0; order < MAX_ORDER; order++) { |
2179 | struct free_area *area = &(zone->free_area[order]); |
2180 | |
2181 | page = list_first_entry_or_null( |
2182 | &area->free_list[MIGRATE_HIGHATOMIC], |
2183 | struct page, lru); |
2184 | if (!page) |
2185 | continue; |
2186 | |
2187 | /* |
2188 | * In page freeing path, migratetype change is racy so |
2189 | * we can counter several free pages in a pageblock |
2190 | * in this loop althoug we changed the pageblock type |
2191 | * from highatomic to ac->migratetype. So we should |
2192 | * adjust the count once. |
2193 | */ |
2194 | if (get_pageblock_migratetype(page) == |
2195 | MIGRATE_HIGHATOMIC) { |
2196 | /* |
2197 | * It should never happen but changes to |
2198 | * locking could inadvertently allow a per-cpu |
2199 | * drain to add pages to MIGRATE_HIGHATOMIC |
2200 | * while unreserving so be safe and watch for |
2201 | * underflows. |
2202 | */ |
2203 | zone->nr_reserved_highatomic -= min( |
2204 | pageblock_nr_pages, |
2205 | zone->nr_reserved_highatomic); |
2206 | } |
2207 | |
2208 | /* |
2209 | * Convert to ac->migratetype and avoid the normal |
2210 | * pageblock stealing heuristics. Minimally, the caller |
2211 | * is doing the work and needs the pages. More |
2212 | * importantly, if the block was always converted to |
2213 | * MIGRATE_UNMOVABLE or another type then the number |
2214 | * of pageblocks that cannot be completely freed |
2215 | * may increase. |
2216 | */ |
2217 | set_pageblock_migratetype(page, ac->migratetype); |
2218 | move_freepages_block(zone, page, ac->migratetype); |
2219 | spin_unlock_irqrestore(&zone->lock, flags); |
2220 | return; |
2221 | } |
2222 | spin_unlock_irqrestore(&zone->lock, flags); |
2223 | } |
2224 | } |
2225 | |
2226 | /* Remove an element from the buddy allocator from the fallback list */ |
2227 | static inline struct page * |
2228 | __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) |
2229 | { |
2230 | struct free_area *area; |
2231 | unsigned int current_order; |
2232 | struct page *page; |
2233 | int fallback_mt; |
2234 | bool can_steal; |
2235 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
2236 | int list_type; |
2237 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
2238 | |
2239 | /* Find the largest possible block of pages in the other list */ |
2240 | for (current_order = MAX_ORDER-1; |
2241 | current_order >= order && current_order <= MAX_ORDER-1; |
2242 | --current_order) { |
2243 | area = &(zone->free_area[current_order]); |
2244 | fallback_mt = find_suitable_fallback(area, current_order, |
2245 | start_migratetype, false, &can_steal); |
2246 | if (fallback_mt == -1) |
2247 | continue; |
2248 | |
2249 | page = list_first_entry(&area->free_list[fallback_mt], |
2250 | struct page, lru); |
2251 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
2252 | /* list_type may change after try_to_steal_freepages */ |
2253 | list_type = fallback_mt; |
2254 | if (can_steal) |
2255 | steal_suitable_fallback(zone, page, start_migratetype, |
2256 | &list_type); |
2257 | #else |
2258 | if (can_steal) |
2259 | steal_suitable_fallback(zone, page, start_migratetype); |
2260 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
2261 | |
2262 | /* Remove the page from the freelists */ |
2263 | area->nr_free--; |
2264 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
2265 | __mod_zone_migrate_state(zone, -(1 << current_order), |
2266 | list_type); |
2267 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
2268 | list_del(&page->lru); |
2269 | rmv_page_order(page); |
2270 | |
2271 | expand(zone, page, order, current_order, area, |
2272 | start_migratetype); |
2273 | /* |
2274 | * The pcppage_migratetype may differ from pageblock's |
2275 | * migratetype depending on the decisions in |
2276 | * find_suitable_fallback(). This is OK as long as it does not |
2277 | * differ for MIGRATE_CMA pageblocks. Those can be used as |
2278 | * fallback only via special __rmqueue_cma_fallback() function |
2279 | */ |
2280 | set_pcppage_migratetype(page, start_migratetype); |
2281 | |
2282 | trace_mm_page_alloc_extfrag(page, order, current_order, |
2283 | start_migratetype, fallback_mt); |
2284 | |
2285 | return page; |
2286 | } |
2287 | |
2288 | return NULL; |
2289 | } |
2290 | |
2291 | /* |
2292 | * Do the hard work of removing an element from the buddy allocator. |
2293 | * Call me with the zone->lock already held. |
2294 | */ |
2295 | #ifdef CONFIG_AMLOGIC_CMA |
2296 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
2297 | int migratetype, bool cma) |
2298 | #else |
2299 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
2300 | int migratetype) |
2301 | #endif /* CONFIG_AMLOGIC_CMA */ |
2302 | { |
2303 | struct page *page; |
2304 | |
2305 | #ifdef CONFIG_AMLOGIC_CMA |
2306 | /* use CMA first */ |
2307 | if (migratetype == MIGRATE_MOVABLE && cma) { |
2308 | page = __rmqueue_cma_fallback(zone, order); |
2309 | if (page) { |
2310 | trace_mm_page_alloc_zone_locked(page, order, |
2311 | MIGRATE_CMA); |
2312 | return page; |
2313 | } |
2314 | } |
2315 | #endif /* CONFIG_AMLOGIC_CMA */ |
2316 | |
2317 | page = __rmqueue_smallest(zone, order, migratetype); |
2318 | if (unlikely(!page)) { |
2319 | #ifndef CONFIG_AMLOGIC_CMA /* no need to try again */ |
2320 | if (migratetype == MIGRATE_MOVABLE) |
2321 | page = __rmqueue_cma_fallback(zone, order); |
2322 | #endif /* !CONFIG_AMLOGIC_CMA */ |
2323 | |
2324 | if (!page) |
2325 | page = __rmqueue_fallback(zone, order, migratetype); |
2326 | } |
2327 | |
2328 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
2329 | return page; |
2330 | } |
2331 | |
2332 | #ifdef CONFIG_AMLOGIC_CMA |
2333 | /* |
2334 | * get page but not cma |
2335 | */ |
2336 | static struct page *rmqueue_no_cma(struct zone *zone, unsigned int order, |
2337 | int migratetype) |
2338 | { |
2339 | struct page *page; |
2340 | |
2341 | spin_lock(&zone->lock); |
2342 | page = __rmqueue_smallest(zone, order, migratetype); |
2343 | if (unlikely(!page)) |
2344 | if (!page) |
2345 | page = __rmqueue_fallback(zone, order, migratetype); |
2346 | WARN_ON(page && is_migrate_cma(get_pcppage_migratetype(page))); |
2347 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); |
2348 | spin_unlock(&zone->lock); |
2349 | return page; |
2350 | } |
2351 | #endif /* CONFIG_AMLOGIC_CMA */ |
2352 | |
2353 | /* |
2354 | * Obtain a specified number of elements from the buddy allocator, all under |
2355 | * a single hold of the lock, for efficiency. Add them to the supplied list. |
2356 | * Returns the number of new pages which were placed at *list. |
2357 | */ |
2358 | #ifdef CONFIG_AMLOGIC_CMA |
2359 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
2360 | unsigned long count, struct list_head *list, |
2361 | int migratetype, bool cold, bool cma) |
2362 | #else |
2363 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
2364 | unsigned long count, struct list_head *list, |
2365 | int migratetype, bool cold) |
2366 | #endif /* CONFIG_AMLOGIC_CMA */ |
2367 | { |
2368 | int i, alloced = 0; |
2369 | |
2370 | spin_lock(&zone->lock); |
2371 | for (i = 0; i < count; ++i) { |
2372 | #ifdef CONFIG_AMLOGIC_CMA |
2373 | struct page *page = __rmqueue(zone, order, migratetype, cma); |
2374 | #else |
2375 | struct page *page = __rmqueue(zone, order, migratetype); |
2376 | #endif /* CONFIG_AMLOGIC_CMA */ |
2377 | if (unlikely(page == NULL)) |
2378 | break; |
2379 | |
2380 | if (unlikely(check_pcp_refill(page))) |
2381 | continue; |
2382 | |
2383 | /* |
2384 | * Split buddy pages returned by expand() are received here |
2385 | * in physical page order. The page is added to the callers and |
2386 | * list and the list head then moves forward. From the callers |
2387 | * perspective, the linked list is ordered by page number in |
2388 | * some conditions. This is useful for IO devices that can |
2389 | * merge IO requests if the physical pages are ordered |
2390 | * properly. |
2391 | */ |
2392 | if (likely(!cold)) |
2393 | list_add(&page->lru, list); |
2394 | else |
2395 | list_add_tail(&page->lru, list); |
2396 | list = &page->lru; |
2397 | alloced++; |
2398 | #ifndef CONFIG_AMLOGIC_MEMORY_EXTEND |
2399 | if (is_migrate_cma(get_pcppage_migratetype(page))) |
2400 | __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, |
2401 | -(1 << order)); |
2402 | #endif /* !CONFIG_AMLOGIC_MEMORY_EXTEND */ |
2403 | } |
2404 | |
2405 | /* |
2406 | * i pages were removed from the buddy list even if some leak due |
2407 | * to check_pcp_refill failing so adjust NR_FREE_PAGES based |
2408 | * on i. Do not confuse with 'alloced' which is the number of |
2409 | * pages added to the pcp list. |
2410 | */ |
2411 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
2412 | spin_unlock(&zone->lock); |
2413 | return alloced; |
2414 | } |
2415 | |
2416 | #ifdef CONFIG_NUMA |
2417 | /* |
2418 | * Called from the vmstat counter updater to drain pagesets of this |
2419 | * currently executing processor on remote nodes after they have |
2420 | * expired. |
2421 | * |
2422 | * Note that this function must be called with the thread pinned to |
2423 | * a single processor. |
2424 | */ |
2425 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
2426 | { |
2427 | unsigned long flags; |
2428 | int to_drain, batch; |
2429 | |
2430 | local_irq_save(flags); |
2431 | batch = READ_ONCE(pcp->batch); |
2432 | to_drain = min(pcp->count, batch); |
2433 | if (to_drain > 0) { |
2434 | free_pcppages_bulk(zone, to_drain, pcp); |
2435 | pcp->count -= to_drain; |
2436 | } |
2437 | local_irq_restore(flags); |
2438 | } |
2439 | #endif |
2440 | |
2441 | /* |
2442 | * Drain pcplists of the indicated processor and zone. |
2443 | * |
2444 | * The processor must either be the current processor and the |
2445 | * thread pinned to the current processor or a processor that |
2446 | * is not online. |
2447 | */ |
2448 | static void drain_pages_zone(unsigned int cpu, struct zone *zone) |
2449 | { |
2450 | unsigned long flags; |
2451 | struct per_cpu_pageset *pset; |
2452 | struct per_cpu_pages *pcp; |
2453 | |
2454 | local_irq_save(flags); |
2455 | pset = per_cpu_ptr(zone->pageset, cpu); |
2456 | |
2457 | pcp = &pset->pcp; |
2458 | if (pcp->count) { |
2459 | free_pcppages_bulk(zone, pcp->count, pcp); |
2460 | pcp->count = 0; |
2461 | } |
2462 | local_irq_restore(flags); |
2463 | } |
2464 | |
2465 | /* |
2466 | * Drain pcplists of all zones on the indicated processor. |
2467 | * |
2468 | * The processor must either be the current processor and the |
2469 | * thread pinned to the current processor or a processor that |
2470 | * is not online. |
2471 | */ |
2472 | static void drain_pages(unsigned int cpu) |
2473 | { |
2474 | struct zone *zone; |
2475 | |
2476 | for_each_populated_zone(zone) { |
2477 | drain_pages_zone(cpu, zone); |
2478 | } |
2479 | } |
2480 | |
2481 | /* |
2482 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. |
2483 | * |
2484 | * The CPU has to be pinned. When zone parameter is non-NULL, spill just |
2485 | * the single zone's pages. |
2486 | */ |
2487 | void drain_local_pages(void *z) |
2488 | { |
2489 | struct zone *zone = (struct zone *)z; |
2490 | int cpu = smp_processor_id(); |
2491 | |
2492 | if (zone) |
2493 | drain_pages_zone(cpu, zone); |
2494 | else |
2495 | drain_pages(cpu); |
2496 | } |
2497 | |
2498 | /* |
2499 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator. |
2500 | * |
2501 | * When zone parameter is non-NULL, spill just the single zone's pages. |
2502 | * |
2503 | * Note that this code is protected against sending an IPI to an offline |
2504 | * CPU but does not guarantee sending an IPI to newly hotplugged CPUs: |
2505 | * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but |
2506 | * nothing keeps CPUs from showing up after we populated the cpumask and |
2507 | * before the call to on_each_cpu_mask(). |
2508 | */ |
2509 | void drain_all_pages(struct zone *zone) |
2510 | { |
2511 | int cpu; |
2512 | |
2513 | /* |
2514 | * Allocate in the BSS so we wont require allocation in |
2515 | * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y |
2516 | */ |
2517 | static cpumask_t cpus_with_pcps; |
2518 | |
2519 | /* |
2520 | * We don't care about racing with CPU hotplug event |
2521 | * as offline notification will cause the notified |
2522 | * cpu to drain that CPU pcps and on_each_cpu_mask |
2523 | * disables preemption as part of its processing |
2524 | */ |
2525 | for_each_online_cpu(cpu) { |
2526 | struct per_cpu_pageset *pcp; |
2527 | struct zone *z; |
2528 | bool has_pcps = false; |
2529 | |
2530 | if (zone) { |
2531 | pcp = per_cpu_ptr(zone->pageset, cpu); |
2532 | if (pcp->pcp.count) |
2533 | has_pcps = true; |
2534 | } else { |
2535 | for_each_populated_zone(z) { |
2536 | pcp = per_cpu_ptr(z->pageset, cpu); |
2537 | if (pcp->pcp.count) { |
2538 | has_pcps = true; |
2539 | break; |
2540 | } |
2541 | } |
2542 | } |
2543 | |
2544 | if (has_pcps) |
2545 | cpumask_set_cpu(cpu, &cpus_with_pcps); |
2546 | else |
2547 | cpumask_clear_cpu(cpu, &cpus_with_pcps); |
2548 | } |
2549 | on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, zone, 1); |
2550 | } |
2551 | |
2552 | #ifdef CONFIG_HIBERNATION |
2553 | |
2554 | void mark_free_pages(struct zone *zone) |
2555 | { |
2556 | unsigned long pfn, max_zone_pfn; |
2557 | unsigned long flags; |
2558 | unsigned int order, t; |
2559 | struct page *page; |
2560 | |
2561 | if (zone_is_empty(zone)) |
2562 | return; |
2563 | |
2564 | spin_lock_irqsave(&zone->lock, flags); |
2565 | |
2566 | max_zone_pfn = zone_end_pfn(zone); |
2567 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2568 | if (pfn_valid(pfn)) { |
2569 | page = pfn_to_page(pfn); |
2570 | |
2571 | if (page_zone(page) != zone) |
2572 | continue; |
2573 | |
2574 | if (!swsusp_page_is_forbidden(page)) |
2575 | swsusp_unset_page_free(page); |
2576 | } |
2577 | |
2578 | for_each_migratetype_order(order, t) { |
2579 | list_for_each_entry(page, |
2580 | &zone->free_area[order].free_list[t], lru) { |
2581 | unsigned long i; |
2582 | |
2583 | pfn = page_to_pfn(page); |
2584 | for (i = 0; i < (1UL << order); i++) |
2585 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
2586 | } |
2587 | } |
2588 | spin_unlock_irqrestore(&zone->lock, flags); |
2589 | } |
2590 | #endif /* CONFIG_PM */ |
2591 | |
2592 | /* |
2593 | * Free a 0-order page |
2594 | * cold == true ? free a cold page : free a hot page |
2595 | */ |
2596 | void free_hot_cold_page(struct page *page, bool cold) |
2597 | { |
2598 | struct zone *zone = page_zone(page); |
2599 | struct per_cpu_pages *pcp; |
2600 | unsigned long flags; |
2601 | unsigned long pfn = page_to_pfn(page); |
2602 | int migratetype; |
2603 | |
2604 | if (!free_pcp_prepare(page)) |
2605 | return; |
2606 | |
2607 | migratetype = get_pfnblock_migratetype(page, pfn); |
2608 | set_pcppage_migratetype(page, migratetype); |
2609 | local_irq_save(flags); |
2610 | __count_vm_event(PGFREE); |
2611 | |
2612 | /* |
2613 | * We only track unmovable, reclaimable and movable on pcp lists. |
2614 | * Free ISOLATE pages back to the allocator because they are being |
2615 | * offlined but treat RESERVE as movable pages so we can get those |
2616 | * areas back if necessary. Otherwise, we may have to free |
2617 | * excessively into the page allocator |
2618 | */ |
2619 | if (migratetype >= MIGRATE_PCPTYPES) { |
2620 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
2621 | if (unlikely(is_migrate_isolate(migratetype)) || |
2622 | unlikely(is_migrate_cma(migratetype))) { |
2623 | free_one_page(zone, page, pfn, 0, migratetype); |
2624 | goto out; |
2625 | } |
2626 | #else |
2627 | if (unlikely(is_migrate_isolate(migratetype))) { |
2628 | free_one_page(zone, page, pfn, 0, migratetype); |
2629 | goto out; |
2630 | } |
2631 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
2632 | migratetype = MIGRATE_MOVABLE; |
2633 | } |
2634 | |
2635 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2636 | #if defined(CONFIG_AMLOGIC_MEMORY_EXTEND) && defined(CONFIG_KASAN) |
2637 | /* |
2638 | * always put freed page to tail of buddy system, in |
2639 | * order to increase probability of use-after-free |
2640 | * for KASAN check. |
2641 | */ |
2642 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
2643 | #else |
2644 | if (!cold) |
2645 | list_add(&page->lru, &pcp->lists[migratetype]); |
2646 | else |
2647 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
2648 | #endif |
2649 | pcp->count++; |
2650 | if (pcp->count >= pcp->high) { |
2651 | unsigned long batch = READ_ONCE(pcp->batch); |
2652 | free_pcppages_bulk(zone, batch, pcp); |
2653 | pcp->count -= batch; |
2654 | } |
2655 | |
2656 | out: |
2657 | local_irq_restore(flags); |
2658 | } |
2659 | |
2660 | /* |
2661 | * Free a list of 0-order pages |
2662 | */ |
2663 | void free_hot_cold_page_list(struct list_head *list, bool cold) |
2664 | { |
2665 | struct page *page, *next; |
2666 | |
2667 | list_for_each_entry_safe(page, next, list, lru) { |
2668 | trace_mm_page_free_batched(page, cold); |
2669 | free_hot_cold_page(page, cold); |
2670 | } |
2671 | } |
2672 | |
2673 | /* |
2674 | * split_page takes a non-compound higher-order page, and splits it into |
2675 | * n (1<<order) sub-pages: page[0..n] |
2676 | * Each sub-page must be freed individually. |
2677 | * |
2678 | * Note: this is probably too low level an operation for use in drivers. |
2679 | * Please consult with lkml before using this in your driver. |
2680 | */ |
2681 | void split_page(struct page *page, unsigned int order) |
2682 | { |
2683 | int i; |
2684 | |
2685 | VM_BUG_ON_PAGE(PageCompound(page), page); |
2686 | VM_BUG_ON_PAGE(!page_count(page), page); |
2687 | |
2688 | #ifdef CONFIG_KMEMCHECK |
2689 | /* |
2690 | * Split shadow pages too, because free(page[0]) would |
2691 | * otherwise free the whole shadow. |
2692 | */ |
2693 | if (kmemcheck_page_is_tracked(page)) |
2694 | split_page(virt_to_page(page[0].shadow), order); |
2695 | #endif |
2696 | |
2697 | for (i = 1; i < (1 << order); i++) |
2698 | set_page_refcounted(page + i); |
2699 | split_page_owner(page, order); |
2700 | } |
2701 | EXPORT_SYMBOL_GPL(split_page); |
2702 | |
2703 | int __isolate_free_page(struct page *page, unsigned int order) |
2704 | { |
2705 | unsigned long watermark; |
2706 | struct zone *zone; |
2707 | int mt; |
2708 | |
2709 | BUG_ON(!PageBuddy(page)); |
2710 | |
2711 | zone = page_zone(page); |
2712 | mt = get_pageblock_migratetype(page); |
2713 | |
2714 | if (!is_migrate_isolate(mt)) { |
2715 | /* |
2716 | * Obey watermarks as if the page was being allocated. We can |
2717 | * emulate a high-order watermark check with a raised order-0 |
2718 | * watermark, because we already know our high-order page |
2719 | * exists. |
2720 | */ |
2721 | watermark = min_wmark_pages(zone) + (1UL << order); |
2722 | if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) |
2723 | return 0; |
2724 | |
2725 | __mod_zone_freepage_state(zone, -(1UL << order), mt); |
2726 | } |
2727 | |
2728 | /* Remove page from free list */ |
2729 | list_del(&page->lru); |
2730 | zone->free_area[order].nr_free--; |
2731 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
2732 | __mod_zone_migrate_state(zone, -(1 << order), |
2733 | get_pcppage_migratetype(page)); |
2734 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
2735 | rmv_page_order(page); |
2736 | |
2737 | /* |
2738 | * Set the pageblock if the isolated page is at least half of a |
2739 | * pageblock |
2740 | */ |
2741 | if (order >= pageblock_order - 1) { |
2742 | struct page *endpage = page + (1 << order) - 1; |
2743 | for (; page < endpage; page += pageblock_nr_pages) { |
2744 | int mt = get_pageblock_migratetype(page); |
2745 | if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)) |
2746 | set_pageblock_migratetype(page, |
2747 | MIGRATE_MOVABLE); |
2748 | } |
2749 | } |
2750 | |
2751 | |
2752 | return 1UL << order; |
2753 | } |
2754 | |
2755 | /* |
2756 | * Update NUMA hit/miss statistics |
2757 | * |
2758 | * Must be called with interrupts disabled. |
2759 | */ |
2760 | static inline void zone_statistics(struct zone *preferred_zone, struct zone *z, |
2761 | gfp_t flags) |
2762 | { |
2763 | #ifdef CONFIG_NUMA |
2764 | enum zone_stat_item local_stat = NUMA_LOCAL; |
2765 | |
2766 | if (z->node != numa_node_id()) |
2767 | local_stat = NUMA_OTHER; |
2768 | |
2769 | if (z->node == preferred_zone->node) |
2770 | __inc_zone_state(z, NUMA_HIT); |
2771 | else { |
2772 | __inc_zone_state(z, NUMA_MISS); |
2773 | __inc_zone_state(preferred_zone, NUMA_FOREIGN); |
2774 | } |
2775 | __inc_zone_state(z, local_stat); |
2776 | #endif |
2777 | } |
2778 | |
2779 | /* |
2780 | * Allocate a page from the given zone. Use pcplists for order-0 allocations. |
2781 | */ |
2782 | static inline |
2783 | struct page *buffered_rmqueue(struct zone *preferred_zone, |
2784 | struct zone *zone, unsigned int order, |
2785 | gfp_t gfp_flags, unsigned int alloc_flags, |
2786 | int migratetype) |
2787 | { |
2788 | unsigned long flags; |
2789 | struct page *page; |
2790 | bool cold = ((gfp_flags & __GFP_COLD) != 0); |
2791 | #ifdef CONFIG_AMLOGIC_CMA |
2792 | bool cma = can_use_cma(gfp_flags); |
2793 | #endif |
2794 | |
2795 | if (likely(order == 0)) { |
2796 | struct per_cpu_pages *pcp; |
2797 | struct list_head *list; |
2798 | |
2799 | local_irq_save(flags); |
2800 | do { |
2801 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
2802 | list = &pcp->lists[migratetype]; |
2803 | if (list_empty(list)) { |
2804 | #ifdef CONFIG_AMLOGIC_CMA |
2805 | pcp->count += rmqueue_bulk(zone, 0, |
2806 | pcp->batch, list, |
2807 | migratetype, cold, |
2808 | cma); |
2809 | #else |
2810 | pcp->count += rmqueue_bulk(zone, 0, |
2811 | pcp->batch, list, |
2812 | migratetype, cold); |
2813 | #endif /* CONFIG_AMLOGIC_CMA */ |
2814 | if (unlikely(list_empty(list))) |
2815 | goto failed; |
2816 | } |
2817 | |
2818 | if (cold) |
2819 | page = list_last_entry(list, struct page, lru); |
2820 | else |
2821 | page = list_first_entry(list, struct page, lru); |
2822 | |
2823 | #ifdef CONFIG_AMLOGIC_CMA |
2824 | /* |
2825 | * USING CMA FIRST POLICY situations: |
2826 | * 1. CMA pages may return to pcp and allocated next |
2827 | * but gfp mask is not suitable for CMA; |
2828 | * 2. MOVABLE pages may return to pcp and allocated next |
2829 | * but gfp mask is suitable for CMA |
2830 | * |
2831 | * For 1, we should replace a none-CMA page |
2832 | * For 2, we should replace with a cma page |
2833 | * before page is deleted from PCP list. |
2834 | */ |
2835 | if (!cma && is_migrate_cma_page(page)) { |
2836 | /* case 1 */ |
2837 | page = rmqueue_no_cma(zone, order, migratetype); |
2838 | if (page) |
2839 | break; |
2840 | goto failed; |
2841 | } else if ((migratetype == MIGRATE_MOVABLE) && |
2842 | (get_pcppage_migratetype(page) != MIGRATE_CMA) && |
2843 | cma) { |
2844 | struct page *tmp_page; |
2845 | |
2846 | spin_lock(&zone->lock); |
2847 | tmp_page = __rmqueue_cma_fallback(zone, order); |
2848 | /* can't alloc cma pages or not ready */ |
2849 | if (!tmp_page || check_new_pcp(page)) { |
2850 | spin_unlock(&zone->lock); |
2851 | goto use_pcp; |
2852 | } |
2853 | page = tmp_page; |
2854 | __mod_zone_freepage_state(zone, -(1 << order), |
2855 | get_pcppage_migratetype(page)); |
2856 | spin_unlock(&zone->lock); |
2857 | goto alloc_success; |
2858 | } |
2859 | use_pcp: |
2860 | #endif /* CONFIG_AMLOGIC_CMA */ |
2861 | |
2862 | list_del(&page->lru); |
2863 | pcp->count--; |
2864 | |
2865 | } while (check_new_pcp(page)); |
2866 | } else { |
2867 | /* |
2868 | * We most definitely don't want callers attempting to |
2869 | * allocate greater than order-1 page units with __GFP_NOFAIL. |
2870 | */ |
2871 | WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1)); |
2872 | spin_lock_irqsave(&zone->lock, flags); |
2873 | |
2874 | do { |
2875 | page = NULL; |
2876 | if (alloc_flags & ALLOC_HARDER) { |
2877 | page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC); |
2878 | if (page) |
2879 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
2880 | } |
2881 | if (!page) |
2882 | #ifdef CONFIG_AMLOGIC_CMA |
2883 | page = __rmqueue(zone, order, |
2884 | migratetype, cma); |
2885 | #else |
2886 | page = __rmqueue(zone, order, migratetype); |
2887 | #endif /* CONFIG_AMLOGIC_CMA */ |
2888 | } while (page && check_new_pages(page, order)); |
2889 | spin_unlock(&zone->lock); |
2890 | if (!page) |
2891 | goto failed; |
2892 | __mod_zone_freepage_state(zone, -(1 << order), |
2893 | get_pcppage_migratetype(page)); |
2894 | } |
2895 | |
2896 | #ifdef CONFIG_AMLOGIC_CMA |
2897 | alloc_success: |
2898 | #endif /* CONFIG_AMLOGIC_CMA */ |
2899 | __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); |
2900 | zone_statistics(preferred_zone, zone, gfp_flags); |
2901 | local_irq_restore(flags); |
2902 | |
2903 | VM_BUG_ON_PAGE(bad_range(zone, page), page); |
2904 | return page; |
2905 | |
2906 | failed: |
2907 | local_irq_restore(flags); |
2908 | return NULL; |
2909 | } |
2910 | |
2911 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
2912 | |
2913 | static struct { |
2914 | struct fault_attr attr; |
2915 | |
2916 | bool ignore_gfp_highmem; |
2917 | bool ignore_gfp_reclaim; |
2918 | u32 min_order; |
2919 | } fail_page_alloc = { |
2920 | .attr = FAULT_ATTR_INITIALIZER, |
2921 | .ignore_gfp_reclaim = true, |
2922 | .ignore_gfp_highmem = true, |
2923 | .min_order = 1, |
2924 | }; |
2925 | |
2926 | static int __init setup_fail_page_alloc(char *str) |
2927 | { |
2928 | return setup_fault_attr(&fail_page_alloc.attr, str); |
2929 | } |
2930 | __setup("fail_page_alloc=", setup_fail_page_alloc); |
2931 | |
2932 | static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
2933 | { |
2934 | if (order < fail_page_alloc.min_order) |
2935 | return false; |
2936 | if (gfp_mask & __GFP_NOFAIL) |
2937 | return false; |
2938 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
2939 | return false; |
2940 | if (fail_page_alloc.ignore_gfp_reclaim && |
2941 | (gfp_mask & __GFP_DIRECT_RECLAIM)) |
2942 | return false; |
2943 | |
2944 | return should_fail(&fail_page_alloc.attr, 1 << order); |
2945 | } |
2946 | |
2947 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
2948 | |
2949 | static int __init fail_page_alloc_debugfs(void) |
2950 | { |
2951 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
2952 | struct dentry *dir; |
2953 | |
2954 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
2955 | &fail_page_alloc.attr); |
2956 | if (IS_ERR(dir)) |
2957 | return PTR_ERR(dir); |
2958 | |
2959 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
2960 | &fail_page_alloc.ignore_gfp_reclaim)) |
2961 | goto fail; |
2962 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, |
2963 | &fail_page_alloc.ignore_gfp_highmem)) |
2964 | goto fail; |
2965 | if (!debugfs_create_u32("min-order", mode, dir, |
2966 | &fail_page_alloc.min_order)) |
2967 | goto fail; |
2968 | |
2969 | return 0; |
2970 | fail: |
2971 | debugfs_remove_recursive(dir); |
2972 | |
2973 | return -ENOMEM; |
2974 | } |
2975 | |
2976 | late_initcall(fail_page_alloc_debugfs); |
2977 | |
2978 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ |
2979 | |
2980 | #else /* CONFIG_FAIL_PAGE_ALLOC */ |
2981 | |
2982 | static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
2983 | { |
2984 | return false; |
2985 | } |
2986 | |
2987 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ |
2988 | |
2989 | /* |
2990 | * Return true if free base pages are above 'mark'. For high-order checks it |
2991 | * will return true of the order-0 watermark is reached and there is at least |
2992 | * one free page of a suitable size. Checking now avoids taking the zone lock |
2993 | * to check in the allocation paths if no pages are free. |
2994 | */ |
2995 | bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
2996 | int classzone_idx, unsigned int alloc_flags, |
2997 | long free_pages) |
2998 | { |
2999 | long min = mark; |
3000 | int o; |
3001 | const bool alloc_harder = (alloc_flags & ALLOC_HARDER); |
3002 | |
3003 | /* free_pages may go negative - that's OK */ |
3004 | free_pages -= (1 << order) - 1; |
3005 | |
3006 | if (alloc_flags & ALLOC_HIGH) |
3007 | min -= min / 2; |
3008 | |
3009 | /* |
3010 | * If the caller does not have rights to ALLOC_HARDER then subtract |
3011 | * the high-atomic reserves. This will over-estimate the size of the |
3012 | * atomic reserve but it avoids a search. |
3013 | */ |
3014 | if (likely(!alloc_harder)) |
3015 | free_pages -= z->nr_reserved_highatomic; |
3016 | else |
3017 | min -= min / 4; |
3018 | |
3019 | #ifdef CONFIG_CMA |
3020 | /* If allocation can't use CMA areas don't use free CMA pages */ |
3021 | #ifndef CONFIG_AMLOGIC_CMA /* always sub cma pages to avoid wm all CMA */ |
3022 | if (!(alloc_flags & ALLOC_CMA)) |
3023 | #endif |
3024 | free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES); |
3025 | #endif |
3026 | |
3027 | /* |
3028 | * Check watermarks for an order-0 allocation request. If these |
3029 | * are not met, then a high-order request also cannot go ahead |
3030 | * even if a suitable page happened to be free. |
3031 | */ |
3032 | #ifdef CONFIG_AMLOGIC_CMA |
3033 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) { |
3034 | /* do not using cma until water mark is low */ |
3035 | if (unlikely(!cma_first_wm_low && free_pages > 0)) { |
3036 | cma_first_wm_low = true; |
3037 | pr_info("Now can use cma, free:%ld, wm:%ld\n", |
3038 | free_pages, |
3039 | min + z->lowmem_reserve[classzone_idx]); |
3040 | } |
3041 | return false; |
3042 | } |
3043 | #else |
3044 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) |
3045 | return false; |
3046 | #endif |
3047 | |
3048 | /* If this is an order-0 request then the watermark is fine */ |
3049 | if (!order) |
3050 | return true; |
3051 | |
3052 | /* For a high-order request, check at least one suitable page is free */ |
3053 | for (o = order; o < MAX_ORDER; o++) { |
3054 | struct free_area *area = &z->free_area[o]; |
3055 | int mt; |
3056 | |
3057 | if (!area->nr_free) |
3058 | continue; |
3059 | |
3060 | for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) { |
3061 | if (!list_empty(&area->free_list[mt])) |
3062 | return true; |
3063 | } |
3064 | |
3065 | #ifdef CONFIG_CMA |
3066 | if ((alloc_flags & ALLOC_CMA) && |
3067 | !list_empty(&area->free_list[MIGRATE_CMA])) { |
3068 | return true; |
3069 | } |
3070 | #endif |
3071 | if (alloc_harder && |
3072 | !list_empty(&area->free_list[MIGRATE_HIGHATOMIC])) |
3073 | return true; |
3074 | } |
3075 | return false; |
3076 | } |
3077 | |
3078 | bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, |
3079 | int classzone_idx, unsigned int alloc_flags) |
3080 | { |
3081 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
3082 | zone_page_state(z, NR_FREE_PAGES)); |
3083 | } |
3084 | |
3085 | static inline bool zone_watermark_fast(struct zone *z, unsigned int order, |
3086 | unsigned long mark, int classzone_idx, unsigned int alloc_flags) |
3087 | { |
3088 | long free_pages = zone_page_state(z, NR_FREE_PAGES); |
3089 | long cma_pages = 0; |
3090 | |
3091 | #ifdef CONFIG_CMA |
3092 | /* If allocation can't use CMA areas don't use free CMA pages */ |
3093 | if (!(alloc_flags & ALLOC_CMA)) |
3094 | cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES); |
3095 | #endif |
3096 | |
3097 | /* |
3098 | * Fast check for order-0 only. If this fails then the reserves |
3099 | * need to be calculated. There is a corner case where the check |
3100 | * passes but only the high-order atomic reserve are free. If |
3101 | * the caller is !atomic then it'll uselessly search the free |
3102 | * list. That corner case is then slower but it is harmless. |
3103 | */ |
3104 | if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx]) |
3105 | return true; |
3106 | |
3107 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
3108 | free_pages); |
3109 | } |
3110 | |
3111 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, |
3112 | unsigned long mark, int classzone_idx) |
3113 | { |
3114 | long free_pages = zone_page_state(z, NR_FREE_PAGES); |
3115 | |
3116 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) |
3117 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); |
3118 | |
3119 | return __zone_watermark_ok(z, order, mark, classzone_idx, 0, |
3120 | free_pages); |
3121 | } |
3122 | |
3123 | #ifdef CONFIG_NUMA |
3124 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3125 | { |
3126 | return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <= |
3127 | RECLAIM_DISTANCE; |
3128 | } |
3129 | #else /* CONFIG_NUMA */ |
3130 | static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone) |
3131 | { |
3132 | return true; |
3133 | } |
3134 | #endif /* CONFIG_NUMA */ |
3135 | |
3136 | /* |
3137 | * get_page_from_freelist goes through the zonelist trying to allocate |
3138 | * a page. |
3139 | */ |
3140 | static struct page * |
3141 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags, |
3142 | const struct alloc_context *ac) |
3143 | { |
3144 | struct zoneref *z = ac->preferred_zoneref; |
3145 | struct zone *zone; |
3146 | struct pglist_data *last_pgdat_dirty_limit = NULL; |
3147 | |
3148 | /* |
3149 | * Scan zonelist, looking for a zone with enough free. |
3150 | * See also __cpuset_node_allowed() comment in kernel/cpuset.c. |
3151 | */ |
3152 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
3153 | ac->nodemask) { |
3154 | struct page *page; |
3155 | unsigned long mark; |
3156 | |
3157 | if (cpusets_enabled() && |
3158 | (alloc_flags & ALLOC_CPUSET) && |
3159 | !__cpuset_zone_allowed(zone, gfp_mask)) |
3160 | continue; |
3161 | /* |
3162 | * When allocating a page cache page for writing, we |
3163 | * want to get it from a node that is within its dirty |
3164 | * limit, such that no single node holds more than its |
3165 | * proportional share of globally allowed dirty pages. |
3166 | * The dirty limits take into account the node's |
3167 | * lowmem reserves and high watermark so that kswapd |
3168 | * should be able to balance it without having to |
3169 | * write pages from its LRU list. |
3170 | * |
3171 | * XXX: For now, allow allocations to potentially |
3172 | * exceed the per-node dirty limit in the slowpath |
3173 | * (spread_dirty_pages unset) before going into reclaim, |
3174 | * which is important when on a NUMA setup the allowed |
3175 | * nodes are together not big enough to reach the |
3176 | * global limit. The proper fix for these situations |
3177 | * will require awareness of nodes in the |
3178 | * dirty-throttling and the flusher threads. |
3179 | */ |
3180 | if (ac->spread_dirty_pages) { |
3181 | if (last_pgdat_dirty_limit == zone->zone_pgdat) |
3182 | continue; |
3183 | |
3184 | if (!node_dirty_ok(zone->zone_pgdat)) { |
3185 | last_pgdat_dirty_limit = zone->zone_pgdat; |
3186 | continue; |
3187 | } |
3188 | } |
3189 | |
3190 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
3191 | if (!zone_watermark_fast(zone, order, mark, |
3192 | ac_classzone_idx(ac), alloc_flags)) { |
3193 | int ret; |
3194 | |
3195 | /* Checked here to keep the fast path fast */ |
3196 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
3197 | if (alloc_flags & ALLOC_NO_WATERMARKS) |
3198 | goto try_this_zone; |
3199 | |
3200 | if (node_reclaim_mode == 0 || |
3201 | !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) |
3202 | continue; |
3203 | |
3204 | ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); |
3205 | switch (ret) { |
3206 | case NODE_RECLAIM_NOSCAN: |
3207 | /* did not scan */ |
3208 | continue; |
3209 | case NODE_RECLAIM_FULL: |
3210 | /* scanned but unreclaimable */ |
3211 | continue; |
3212 | default: |
3213 | /* did we reclaim enough */ |
3214 | if (zone_watermark_ok(zone, order, mark, |
3215 | ac_classzone_idx(ac), alloc_flags)) |
3216 | goto try_this_zone; |
3217 | |
3218 | continue; |
3219 | } |
3220 | } |
3221 | |
3222 | try_this_zone: |
3223 | page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order, |
3224 | gfp_mask, alloc_flags, ac->migratetype); |
3225 | if (page) { |
3226 | prep_new_page(page, order, gfp_mask, alloc_flags); |
3227 | |
3228 | /* |
3229 | * If this is a high-order atomic allocation then check |
3230 | * if the pageblock should be reserved for the future |
3231 | */ |
3232 | if (unlikely(order && (alloc_flags & ALLOC_HARDER))) |
3233 | reserve_highatomic_pageblock(page, zone, order); |
3234 | |
3235 | return page; |
3236 | } |
3237 | } |
3238 | |
3239 | return NULL; |
3240 | } |
3241 | |
3242 | /* |
3243 | * Large machines with many possible nodes should not always dump per-node |
3244 | * meminfo in irq context. |
3245 | */ |
3246 | static inline bool should_suppress_show_mem(void) |
3247 | { |
3248 | bool ret = false; |
3249 | |
3250 | #if NODES_SHIFT > 8 |
3251 | ret = in_interrupt(); |
3252 | #endif |
3253 | return ret; |
3254 | } |
3255 | |
3256 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
3257 | DEFAULT_RATELIMIT_INTERVAL, |
3258 | DEFAULT_RATELIMIT_BURST); |
3259 | |
3260 | void warn_alloc(gfp_t gfp_mask, const char *fmt, ...) |
3261 | { |
3262 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
3263 | struct va_format vaf; |
3264 | va_list args; |
3265 | |
3266 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
3267 | debug_guardpage_minorder() > 0) |
3268 | return; |
3269 | |
3270 | /* |
3271 | * This documents exceptions given to allocations in certain |
3272 | * contexts that are allowed to allocate outside current's set |
3273 | * of allowed nodes. |
3274 | */ |
3275 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
3276 | if (test_thread_flag(TIF_MEMDIE) || |
3277 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
3278 | filter &= ~SHOW_MEM_FILTER_NODES; |
3279 | if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
3280 | filter &= ~SHOW_MEM_FILTER_NODES; |
3281 | |
3282 | pr_warn("%s: ", current->comm); |
3283 | |
3284 | va_start(args, fmt); |
3285 | vaf.fmt = fmt; |
3286 | vaf.va = &args; |
3287 | pr_cont("%pV", &vaf); |
3288 | va_end(args); |
3289 | |
3290 | pr_cont(", mode:%#x(%pGg)\n", gfp_mask, &gfp_mask); |
3291 | |
3292 | dump_stack(); |
3293 | if (!should_suppress_show_mem()) |
3294 | show_mem(filter); |
3295 | } |
3296 | |
3297 | static inline struct page * |
3298 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, |
3299 | const struct alloc_context *ac, unsigned long *did_some_progress) |
3300 | { |
3301 | struct oom_control oc = { |
3302 | .zonelist = ac->zonelist, |
3303 | .nodemask = ac->nodemask, |
3304 | .memcg = NULL, |
3305 | .gfp_mask = gfp_mask, |
3306 | .order = order, |
3307 | }; |
3308 | struct page *page; |
3309 | |
3310 | *did_some_progress = 0; |
3311 | |
3312 | /* |
3313 | * Acquire the oom lock. If that fails, somebody else is |
3314 | * making progress for us. |
3315 | */ |
3316 | if (!mutex_trylock(&oom_lock)) { |
3317 | *did_some_progress = 1; |
3318 | schedule_timeout_uninterruptible(1); |
3319 | return NULL; |
3320 | } |
3321 | |
3322 | /* |
3323 | * Go through the zonelist yet one more time, keep very high watermark |
3324 | * here, this is only to catch a parallel oom killing, we must fail if |
3325 | * we're still under heavy pressure. |
3326 | */ |
3327 | page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order, |
3328 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac); |
3329 | if (page) |
3330 | goto out; |
3331 | |
3332 | if (!(gfp_mask & __GFP_NOFAIL)) { |
3333 | /* Coredumps can quickly deplete all memory reserves */ |
3334 | if (current->flags & PF_DUMPCORE) |
3335 | goto out; |
3336 | /* The OOM killer will not help higher order allocs */ |
3337 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
3338 | goto out; |
3339 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
3340 | if (ac->high_zoneidx < ZONE_NORMAL) |
3341 | goto out; |
3342 | if (pm_suspended_storage()) |
3343 | goto out; |
3344 | /* |
3345 | * XXX: GFP_NOFS allocations should rather fail than rely on |
3346 | * other request to make a forward progress. |
3347 | * We are in an unfortunate situation where out_of_memory cannot |
3348 | * do much for this context but let's try it to at least get |
3349 | * access to memory reserved if the current task is killed (see |
3350 | * out_of_memory). Once filesystems are ready to handle allocation |
3351 | * failures more gracefully we should just bail out here. |
3352 | */ |
3353 | |
3354 | /* The OOM killer may not free memory on a specific node */ |
3355 | if (gfp_mask & __GFP_THISNODE) |
3356 | goto out; |
3357 | } |
3358 | /* Exhausted what can be done so it's blamo time */ |
3359 | if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
3360 | *did_some_progress = 1; |
3361 | |
3362 | if (gfp_mask & __GFP_NOFAIL) { |
3363 | page = get_page_from_freelist(gfp_mask, order, |
3364 | ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac); |
3365 | /* |
3366 | * fallback to ignore cpuset restriction if our nodes |
3367 | * are depleted |
3368 | */ |
3369 | if (!page) |
3370 | page = get_page_from_freelist(gfp_mask, order, |
3371 | ALLOC_NO_WATERMARKS, ac); |
3372 | } |
3373 | } |
3374 | out: |
3375 | mutex_unlock(&oom_lock); |
3376 | return page; |
3377 | } |
3378 | |
3379 | /* |
3380 | * Maximum number of compaction retries wit a progress before OOM |
3381 | * killer is consider as the only way to move forward. |
3382 | */ |
3383 | #define MAX_COMPACT_RETRIES 16 |
3384 | |
3385 | #ifdef CONFIG_COMPACTION |
3386 | /* Try memory compaction for high-order allocations before reclaim */ |
3387 | static struct page * |
3388 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
3389 | unsigned int alloc_flags, const struct alloc_context *ac, |
3390 | enum compact_priority prio, enum compact_result *compact_result) |
3391 | { |
3392 | struct page *page; |
3393 | unsigned long pflags; |
3394 | unsigned int noreclaim_flag = current->flags & PF_MEMALLOC; |
3395 | |
3396 | if (!order) |
3397 | return NULL; |
3398 | |
3399 | psi_memstall_enter(&pflags); |
3400 | current->flags |= PF_MEMALLOC; |
3401 | |
3402 | *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, |
3403 | prio); |
3404 | |
3405 | current->flags = (current->flags & ~PF_MEMALLOC) | noreclaim_flag; |
3406 | psi_memstall_leave(&pflags); |
3407 | |
3408 | if (*compact_result <= COMPACT_INACTIVE) |
3409 | return NULL; |
3410 | |
3411 | /* |
3412 | * At least in one zone compaction wasn't deferred or skipped, so let's |
3413 | * count a compaction stall |
3414 | */ |
3415 | count_vm_event(COMPACTSTALL); |
3416 | |
3417 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
3418 | |
3419 | if (page) { |
3420 | struct zone *zone = page_zone(page); |
3421 | |
3422 | zone->compact_blockskip_flush = false; |
3423 | compaction_defer_reset(zone, order, true); |
3424 | count_vm_event(COMPACTSUCCESS); |
3425 | return page; |
3426 | } |
3427 | |
3428 | /* |
3429 | * It's bad if compaction run occurs and fails. The most likely reason |
3430 | * is that pages exist, but not enough to satisfy watermarks. |
3431 | */ |
3432 | count_vm_event(COMPACTFAIL); |
3433 | |
3434 | cond_resched(); |
3435 | |
3436 | return NULL; |
3437 | } |
3438 | |
3439 | static inline bool |
3440 | should_compact_retry(struct alloc_context *ac, int order, int alloc_flags, |
3441 | enum compact_result compact_result, |
3442 | enum compact_priority *compact_priority, |
3443 | int *compaction_retries) |
3444 | { |
3445 | int max_retries = MAX_COMPACT_RETRIES; |
3446 | int min_priority; |
3447 | |
3448 | if (!order) |
3449 | return false; |
3450 | |
3451 | if (compaction_made_progress(compact_result)) |
3452 | (*compaction_retries)++; |
3453 | |
3454 | /* |
3455 | * compaction considers all the zone as desperately out of memory |
3456 | * so it doesn't really make much sense to retry except when the |
3457 | * failure could be caused by insufficient priority |
3458 | */ |
3459 | if (compaction_failed(compact_result)) |
3460 | goto check_priority; |
3461 | |
3462 | /* |
3463 | * make sure the compaction wasn't deferred or didn't bail out early |
3464 | * due to locks contention before we declare that we should give up. |
3465 | * But do not retry if the given zonelist is not suitable for |
3466 | * compaction. |
3467 | */ |
3468 | if (compaction_withdrawn(compact_result)) |
3469 | return compaction_zonelist_suitable(ac, order, alloc_flags); |
3470 | |
3471 | /* |
3472 | * !costly requests are much more important than __GFP_REPEAT |
3473 | * costly ones because they are de facto nofail and invoke OOM |
3474 | * killer to move on while costly can fail and users are ready |
3475 | * to cope with that. 1/4 retries is rather arbitrary but we |
3476 | * would need much more detailed feedback from compaction to |
3477 | * make a better decision. |
3478 | */ |
3479 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
3480 | max_retries /= 4; |
3481 | if (*compaction_retries <= max_retries) |
3482 | return true; |
3483 | |
3484 | /* |
3485 | * Make sure there are attempts at the highest priority if we exhausted |
3486 | * all retries or failed at the lower priorities. |
3487 | */ |
3488 | check_priority: |
3489 | min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
3490 | MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY; |
3491 | if (*compact_priority > min_priority) { |
3492 | (*compact_priority)--; |
3493 | *compaction_retries = 0; |
3494 | return true; |
3495 | } |
3496 | return false; |
3497 | } |
3498 | #else |
3499 | static inline struct page * |
3500 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
3501 | unsigned int alloc_flags, const struct alloc_context *ac, |
3502 | enum compact_priority prio, enum compact_result *compact_result) |
3503 | { |
3504 | *compact_result = COMPACT_SKIPPED; |
3505 | return NULL; |
3506 | } |
3507 | |
3508 | static inline bool |
3509 | should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags, |
3510 | enum compact_result compact_result, |
3511 | enum compact_priority *compact_priority, |
3512 | int *compaction_retries) |
3513 | { |
3514 | struct zone *zone; |
3515 | struct zoneref *z; |
3516 | |
3517 | if (!order || order > PAGE_ALLOC_COSTLY_ORDER) |
3518 | return false; |
3519 | |
3520 | /* |
3521 | * There are setups with compaction disabled which would prefer to loop |
3522 | * inside the allocator rather than hit the oom killer prematurely. |
3523 | * Let's give them a good hope and keep retrying while the order-0 |
3524 | * watermarks are OK. |
3525 | */ |
3526 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
3527 | ac->nodemask) { |
3528 | if (zone_watermark_ok(zone, 0, min_wmark_pages(zone), |
3529 | ac_classzone_idx(ac), alloc_flags)) |
3530 | return true; |
3531 | } |
3532 | return false; |
3533 | } |
3534 | #endif /* CONFIG_COMPACTION */ |
3535 | |
3536 | /* Perform direct synchronous page reclaim */ |
3537 | static int |
3538 | __perform_reclaim(gfp_t gfp_mask, unsigned int order, |
3539 | const struct alloc_context *ac) |
3540 | { |
3541 | struct reclaim_state reclaim_state; |
3542 | int progress; |
3543 | unsigned long pflags; |
3544 | |
3545 | cond_resched(); |
3546 | |
3547 | /* We now go into synchronous reclaim */ |
3548 | cpuset_memory_pressure_bump(); |
3549 | psi_memstall_enter(&pflags); |
3550 | current->flags |= PF_MEMALLOC; |
3551 | lockdep_set_current_reclaim_state(gfp_mask); |
3552 | reclaim_state.reclaimed_slab = 0; |
3553 | current->reclaim_state = &reclaim_state; |
3554 | |
3555 | progress = try_to_free_pages(ac->zonelist, order, gfp_mask, |
3556 | ac->nodemask); |
3557 | |
3558 | current->reclaim_state = NULL; |
3559 | lockdep_clear_current_reclaim_state(); |
3560 | current->flags &= ~PF_MEMALLOC; |
3561 | psi_memstall_leave(&pflags); |
3562 | |
3563 | cond_resched(); |
3564 | |
3565 | return progress; |
3566 | } |
3567 | |
3568 | /* The really slow allocator path where we enter direct reclaim */ |
3569 | static inline struct page * |
3570 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, |
3571 | unsigned int alloc_flags, const struct alloc_context *ac, |
3572 | unsigned long *did_some_progress) |
3573 | { |
3574 | struct page *page = NULL; |
3575 | bool drained = false; |
3576 | |
3577 | *did_some_progress = __perform_reclaim(gfp_mask, order, ac); |
3578 | if (unlikely(!(*did_some_progress))) |
3579 | return NULL; |
3580 | |
3581 | retry: |
3582 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
3583 | |
3584 | /* |
3585 | * If an allocation failed after direct reclaim, it could be because |
3586 | * pages are pinned on the per-cpu lists or in high alloc reserves. |
3587 | * Shrink them them and try again |
3588 | */ |
3589 | if (!page && !drained) { |
3590 | unreserve_highatomic_pageblock(ac); |
3591 | drain_all_pages(NULL); |
3592 | drained = true; |
3593 | goto retry; |
3594 | } |
3595 | |
3596 | return page; |
3597 | } |
3598 | |
3599 | static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac) |
3600 | { |
3601 | struct zoneref *z; |
3602 | struct zone *zone; |
3603 | pg_data_t *last_pgdat = NULL; |
3604 | |
3605 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
3606 | ac->high_zoneidx, ac->nodemask) { |
3607 | if (last_pgdat != zone->zone_pgdat) |
3608 | wakeup_kswapd(zone, order, ac->high_zoneidx); |
3609 | last_pgdat = zone->zone_pgdat; |
3610 | } |
3611 | } |
3612 | |
3613 | static inline unsigned int |
3614 | gfp_to_alloc_flags(gfp_t gfp_mask) |
3615 | { |
3616 | unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
3617 | |
3618 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
3619 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
3620 | |
3621 | /* |
3622 | * The caller may dip into page reserves a bit more if the caller |
3623 | * cannot run direct reclaim, or if the caller has realtime scheduling |
3624 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
3625 | * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH). |
3626 | */ |
3627 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
3628 | |
3629 | if (gfp_mask & __GFP_ATOMIC) { |
3630 | /* |
3631 | * Not worth trying to allocate harder for __GFP_NOMEMALLOC even |
3632 | * if it can't schedule. |
3633 | */ |
3634 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
3635 | alloc_flags |= ALLOC_HARDER; |
3636 | /* |
3637 | * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the |
3638 | * comment for __cpuset_node_allowed(). |
3639 | */ |
3640 | alloc_flags &= ~ALLOC_CPUSET; |
3641 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
3642 | alloc_flags |= ALLOC_HARDER; |
3643 | |
3644 | #ifdef CONFIG_CMA |
3645 | if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) |
3646 | alloc_flags |= ALLOC_CMA; |
3647 | #endif |
3648 | return alloc_flags; |
3649 | } |
3650 | |
3651 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) |
3652 | { |
3653 | if (unlikely(gfp_mask & __GFP_NOMEMALLOC)) |
3654 | return false; |
3655 | |
3656 | if (gfp_mask & __GFP_MEMALLOC) |
3657 | return true; |
3658 | if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) |
3659 | return true; |
3660 | if (!in_interrupt() && |
3661 | ((current->flags & PF_MEMALLOC) || |
3662 | unlikely(test_thread_flag(TIF_MEMDIE)))) |
3663 | return true; |
3664 | |
3665 | return false; |
3666 | } |
3667 | |
3668 | /* |
3669 | * Checks whether it makes sense to retry the reclaim to make a forward progress |
3670 | * for the given allocation request. |
3671 | * The reclaim feedback represented by did_some_progress (any progress during |
3672 | * the last reclaim round) and no_progress_loops (number of reclaim rounds without |
3673 | * any progress in a row) is considered as well as the reclaimable pages on the |
3674 | * applicable zone list (with a backoff mechanism which is a function of |
3675 | * no_progress_loops). |
3676 | * |
3677 | * Returns true if a retry is viable or false to enter the oom path. |
3678 | */ |
3679 | static inline bool |
3680 | should_reclaim_retry(gfp_t gfp_mask, unsigned order, |
3681 | struct alloc_context *ac, int alloc_flags, |
3682 | bool did_some_progress, int *no_progress_loops) |
3683 | { |
3684 | struct zone *zone; |
3685 | struct zoneref *z; |
3686 | |
3687 | /* |
3688 | * Costly allocations might have made a progress but this doesn't mean |
3689 | * their order will become available due to high fragmentation so |
3690 | * always increment the no progress counter for them |
3691 | */ |
3692 | if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) |
3693 | *no_progress_loops = 0; |
3694 | else |
3695 | (*no_progress_loops)++; |
3696 | |
3697 | /* |
3698 | * Make sure we converge to OOM if we cannot make any progress |
3699 | * several times in the row. |
3700 | */ |
3701 | if (*no_progress_loops > MAX_RECLAIM_RETRIES) |
3702 | return false; |
3703 | |
3704 | /* |
3705 | * Keep reclaiming pages while there is a chance this will lead |
3706 | * somewhere. If none of the target zones can satisfy our allocation |
3707 | * request even if all reclaimable pages are considered then we are |
3708 | * screwed and have to go OOM. |
3709 | */ |
3710 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
3711 | ac->nodemask) { |
3712 | unsigned long available; |
3713 | unsigned long reclaimable; |
3714 | |
3715 | available = reclaimable = zone_reclaimable_pages(zone); |
3716 | available -= DIV_ROUND_UP((*no_progress_loops) * available, |
3717 | MAX_RECLAIM_RETRIES); |
3718 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
3719 | |
3720 | /* |
3721 | * Would the allocation succeed if we reclaimed the whole |
3722 | * available? |
3723 | */ |
3724 | if (__zone_watermark_ok(zone, order, min_wmark_pages(zone), |
3725 | ac_classzone_idx(ac), alloc_flags, available)) { |
3726 | /* |
3727 | * If we didn't make any progress and have a lot of |
3728 | * dirty + writeback pages then we should wait for |
3729 | * an IO to complete to slow down the reclaim and |
3730 | * prevent from pre mature OOM |
3731 | */ |
3732 | if (!did_some_progress) { |
3733 | unsigned long write_pending; |
3734 | |
3735 | write_pending = zone_page_state_snapshot(zone, |
3736 | NR_ZONE_WRITE_PENDING); |
3737 | |
3738 | if (2 * write_pending > reclaimable) { |
3739 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
3740 | return true; |
3741 | } |
3742 | } |
3743 | |
3744 | /* |
3745 | * Memory allocation/reclaim might be called from a WQ |
3746 | * context and the current implementation of the WQ |
3747 | * concurrency control doesn't recognize that |
3748 | * a particular WQ is congested if the worker thread is |
3749 | * looping without ever sleeping. Therefore we have to |
3750 | * do a short sleep here rather than calling |
3751 | * cond_resched(). |
3752 | */ |
3753 | if (current->flags & PF_WQ_WORKER) |
3754 | schedule_timeout_uninterruptible(1); |
3755 | else |
3756 | cond_resched(); |
3757 | |
3758 | return true; |
3759 | } |
3760 | } |
3761 | |
3762 | return false; |
3763 | } |
3764 | |
3765 | static inline struct page * |
3766 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, |
3767 | struct alloc_context *ac) |
3768 | { |
3769 | bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; |
3770 | struct page *page = NULL; |
3771 | unsigned int alloc_flags; |
3772 | unsigned long did_some_progress; |
3773 | enum compact_priority compact_priority; |
3774 | enum compact_result compact_result; |
3775 | int compaction_retries; |
3776 | int no_progress_loops; |
3777 | unsigned int cpuset_mems_cookie; |
3778 | |
3779 | /* |
3780 | * In the slowpath, we sanity check order to avoid ever trying to |
3781 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may |
3782 | * be using allocators in order of preference for an area that is |
3783 | * too large. |
3784 | */ |
3785 | if (order >= MAX_ORDER) { |
3786 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); |
3787 | return NULL; |
3788 | } |
3789 | |
3790 | /* |
3791 | * We also sanity check to catch abuse of atomic reserves being used by |
3792 | * callers that are not in atomic context. |
3793 | */ |
3794 | if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) == |
3795 | (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM))) |
3796 | gfp_mask &= ~__GFP_ATOMIC; |
3797 | |
3798 | retry_cpuset: |
3799 | compaction_retries = 0; |
3800 | no_progress_loops = 0; |
3801 | compact_priority = DEF_COMPACT_PRIORITY; |
3802 | cpuset_mems_cookie = read_mems_allowed_begin(); |
3803 | /* |
3804 | * We need to recalculate the starting point for the zonelist iterator |
3805 | * because we might have used different nodemask in the fast path, or |
3806 | * there was a cpuset modification and we are retrying - otherwise we |
3807 | * could end up iterating over non-eligible zones endlessly. |
3808 | */ |
3809 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
3810 | ac->high_zoneidx, ac->nodemask); |
3811 | if (!ac->preferred_zoneref->zone) |
3812 | goto nopage; |
3813 | |
3814 | |
3815 | /* |
3816 | * The fast path uses conservative alloc_flags to succeed only until |
3817 | * kswapd needs to be woken up, and to avoid the cost of setting up |
3818 | * alloc_flags precisely. So we do that now. |
3819 | */ |
3820 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
3821 | |
3822 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3823 | wake_all_kswapds(order, ac); |
3824 | |
3825 | /* |
3826 | * The adjusted alloc_flags might result in immediate success, so try |
3827 | * that first |
3828 | */ |
3829 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
3830 | if (page) |
3831 | goto got_pg; |
3832 | |
3833 | /* |
3834 | * For costly allocations, try direct compaction first, as it's likely |
3835 | * that we have enough base pages and don't need to reclaim. Don't try |
3836 | * that for allocations that are allowed to ignore watermarks, as the |
3837 | * ALLOC_NO_WATERMARKS attempt didn't yet happen. |
3838 | */ |
3839 | if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER && |
3840 | !gfp_pfmemalloc_allowed(gfp_mask)) { |
3841 | page = __alloc_pages_direct_compact(gfp_mask, order, |
3842 | alloc_flags, ac, |
3843 | INIT_COMPACT_PRIORITY, |
3844 | &compact_result); |
3845 | if (page) |
3846 | goto got_pg; |
3847 | |
3848 | /* |
3849 | * Checks for costly allocations with __GFP_NORETRY, which |
3850 | * includes THP page fault allocations |
3851 | */ |
3852 | if (gfp_mask & __GFP_NORETRY) { |
3853 | /* |
3854 | * If compaction is deferred for high-order allocations, |
3855 | * it is because sync compaction recently failed. If |
3856 | * this is the case and the caller requested a THP |
3857 | * allocation, we do not want to heavily disrupt the |
3858 | * system, so we fail the allocation instead of entering |
3859 | * direct reclaim. |
3860 | */ |
3861 | if (compact_result == COMPACT_DEFERRED) |
3862 | goto nopage; |
3863 | |
3864 | /* |
3865 | * Looks like reclaim/compaction is worth trying, but |
3866 | * sync compaction could be very expensive, so keep |
3867 | * using async compaction. |
3868 | */ |
3869 | compact_priority = INIT_COMPACT_PRIORITY; |
3870 | } |
3871 | } |
3872 | |
3873 | retry: |
3874 | /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */ |
3875 | if (gfp_mask & __GFP_KSWAPD_RECLAIM) |
3876 | wake_all_kswapds(order, ac); |
3877 | |
3878 | if (gfp_pfmemalloc_allowed(gfp_mask)) |
3879 | alloc_flags = ALLOC_NO_WATERMARKS; |
3880 | |
3881 | /* |
3882 | * Reset the zonelist iterators if memory policies can be ignored. |
3883 | * These allocations are high priority and system rather than user |
3884 | * orientated. |
3885 | */ |
3886 | if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) { |
3887 | ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, |
3888 | ac->high_zoneidx, ac->nodemask); |
3889 | } |
3890 | |
3891 | /* Attempt with potentially adjusted zonelist and alloc_flags */ |
3892 | page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac); |
3893 | if (page) |
3894 | goto got_pg; |
3895 | |
3896 | /* Caller is not willing to reclaim, we can't balance anything */ |
3897 | if (!can_direct_reclaim) { |
3898 | /* |
3899 | * All existing users of the __GFP_NOFAIL are blockable, so warn |
3900 | * of any new users that actually allow this type of allocation |
3901 | * to fail. |
3902 | */ |
3903 | WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL); |
3904 | goto nopage; |
3905 | } |
3906 | |
3907 | /* Avoid recursion of direct reclaim */ |
3908 | if (current->flags & PF_MEMALLOC) { |
3909 | /* |
3910 | * __GFP_NOFAIL request from this context is rather bizarre |
3911 | * because we cannot reclaim anything and only can loop waiting |
3912 | * for somebody to do a work for us. |
3913 | */ |
3914 | if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) { |
3915 | cond_resched(); |
3916 | goto retry; |
3917 | } |
3918 | goto nopage; |
3919 | } |
3920 | |
3921 | /* Avoid allocations with no watermarks from looping endlessly */ |
3922 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) |
3923 | goto nopage; |
3924 | |
3925 | |
3926 | /* Try direct reclaim and then allocating */ |
3927 | page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, |
3928 | &did_some_progress); |
3929 | if (page) |
3930 | goto got_pg; |
3931 | |
3932 | /* Try direct compaction and then allocating */ |
3933 | page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac, |
3934 | compact_priority, &compact_result); |
3935 | if (page) |
3936 | goto got_pg; |
3937 | |
3938 | /* Do not loop if specifically requested */ |
3939 | if (gfp_mask & __GFP_NORETRY) |
3940 | goto nopage; |
3941 | |
3942 | /* |
3943 | * Do not retry costly high order allocations unless they are |
3944 | * __GFP_REPEAT |
3945 | */ |
3946 | if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT)) |
3947 | goto nopage; |
3948 | |
3949 | if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, |
3950 | did_some_progress > 0, &no_progress_loops)) |
3951 | goto retry; |
3952 | |
3953 | /* |
3954 | * It doesn't make any sense to retry for the compaction if the order-0 |
3955 | * reclaim is not able to make any progress because the current |
3956 | * implementation of the compaction depends on the sufficient amount |
3957 | * of free memory (see __compaction_suitable) |
3958 | */ |
3959 | if (did_some_progress > 0 && |
3960 | should_compact_retry(ac, order, alloc_flags, |
3961 | compact_result, &compact_priority, |
3962 | &compaction_retries)) |
3963 | goto retry; |
3964 | |
3965 | /* |
3966 | * It's possible we raced with cpuset update so the OOM would be |
3967 | * premature (see below the nopage: label for full explanation). |
3968 | */ |
3969 | if (read_mems_allowed_retry(cpuset_mems_cookie)) |
3970 | goto retry_cpuset; |
3971 | |
3972 | /* Reclaim has failed us, start killing things */ |
3973 | page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); |
3974 | if (page) |
3975 | goto got_pg; |
3976 | |
3977 | /* Retry as long as the OOM killer is making progress */ |
3978 | if (did_some_progress) { |
3979 | no_progress_loops = 0; |
3980 | goto retry; |
3981 | } |
3982 | |
3983 | nopage: |
3984 | /* |
3985 | * When updating a task's mems_allowed or mempolicy nodemask, it is |
3986 | * possible to race with parallel threads in such a way that our |
3987 | * allocation can fail while the mask is being updated. If we are about |
3988 | * to fail, check if the cpuset changed during allocation and if so, |
3989 | * retry. |
3990 | */ |
3991 | if (read_mems_allowed_retry(cpuset_mems_cookie)) |
3992 | goto retry_cpuset; |
3993 | |
3994 | warn_alloc(gfp_mask, |
3995 | "page allocation failure: order:%u", order); |
3996 | got_pg: |
3997 | return page; |
3998 | } |
3999 | |
4000 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
4001 | static inline void should_wakeup_kswap(gfp_t gfp_mask, int order, |
4002 | struct alloc_context *ac) |
4003 | { |
4004 | unsigned long free_pages, free_cma = 0; |
4005 | struct zoneref *z = ac->preferred_zoneref; |
4006 | struct zone *zone; |
4007 | |
4008 | if (!(gfp_mask & __GFP_RECLAIM)) /* not allowed */ |
4009 | return; |
4010 | |
4011 | for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
4012 | ac->nodemask) { |
4013 | free_pages = zone_page_state(zone, NR_FREE_PAGES); |
4014 | #ifdef CONFIG_AMLOGIC_CMA |
4015 | if (can_use_cma(gfp_mask)) |
4016 | free_cma = zone_page_state(zone, NR_FREE_CMA_PAGES); |
4017 | #endif /* CONFIG_AMLOGIC_CMA */ |
4018 | free_pages -= free_cma; |
4019 | /* |
4020 | * wake up kswapd before get pages from buddy, this help to |
4021 | * fast reclaim process and can avoid memory become too low |
4022 | * some times |
4023 | */ |
4024 | if (free_pages <= high_wmark_pages(zone)) |
4025 | wakeup_kswapd(zone, order, ac->high_zoneidx); |
4026 | } |
4027 | } |
4028 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
4029 | |
4030 | /* |
4031 | * This is the 'heart' of the zoned buddy allocator. |
4032 | */ |
4033 | struct page * |
4034 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, |
4035 | struct zonelist *zonelist, nodemask_t *nodemask) |
4036 | { |
4037 | struct page *page; |
4038 | unsigned int alloc_flags = ALLOC_WMARK_LOW; |
4039 | gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */ |
4040 | struct alloc_context ac = { |
4041 | .high_zoneidx = gfp_zone(gfp_mask), |
4042 | .zonelist = zonelist, |
4043 | .nodemask = nodemask, |
4044 | .migratetype = gfpflags_to_migratetype(gfp_mask), |
4045 | }; |
4046 | |
4047 | if (cpusets_enabled()) { |
4048 | alloc_mask |= __GFP_HARDWALL; |
4049 | alloc_flags |= ALLOC_CPUSET; |
4050 | if (!ac.nodemask) |
4051 | ac.nodemask = &cpuset_current_mems_allowed; |
4052 | } |
4053 | |
4054 | gfp_mask &= gfp_allowed_mask; |
4055 | |
4056 | lockdep_trace_alloc(gfp_mask); |
4057 | |
4058 | might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM); |
4059 | |
4060 | if (should_fail_alloc_page(gfp_mask, order)) |
4061 | return NULL; |
4062 | |
4063 | /* |
4064 | * Check the zones suitable for the gfp_mask contain at least one |
4065 | * valid zone. It's possible to have an empty zonelist as a result |
4066 | * of __GFP_THISNODE and a memoryless node |
4067 | */ |
4068 | if (unlikely(!zonelist->_zonerefs->zone)) |
4069 | return NULL; |
4070 | |
4071 | if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE) |
4072 | alloc_flags |= ALLOC_CMA; |
4073 | |
4074 | /* Dirty zone balancing only done in the fast path */ |
4075 | ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE); |
4076 | |
4077 | /* |
4078 | * The preferred zone is used for statistics but crucially it is |
4079 | * also used as the starting point for the zonelist iterator. It |
4080 | * may get reset for allocations that ignore memory policies. |
4081 | */ |
4082 | ac.preferred_zoneref = first_zones_zonelist(ac.zonelist, |
4083 | ac.high_zoneidx, ac.nodemask); |
4084 | if (!ac.preferred_zoneref->zone) { |
4085 | page = NULL; |
4086 | /* |
4087 | * This might be due to race with cpuset_current_mems_allowed |
4088 | * update, so make sure we retry with original nodemask in the |
4089 | * slow path. |
4090 | */ |
4091 | goto no_zone; |
4092 | } |
4093 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
4094 | should_wakeup_kswap(gfp_mask, order, &ac); |
4095 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
4096 | |
4097 | /* First allocation attempt */ |
4098 | page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac); |
4099 | if (likely(page)) |
4100 | goto out; |
4101 | |
4102 | no_zone: |
4103 | /* |
4104 | * Runtime PM, block IO and its error handling path can deadlock |
4105 | * because I/O on the device might not complete. |
4106 | */ |
4107 | alloc_mask = memalloc_noio_flags(gfp_mask); |
4108 | ac.spread_dirty_pages = false; |
4109 | |
4110 | /* |
4111 | * Restore the original nodemask if it was potentially replaced with |
4112 | * &cpuset_current_mems_allowed to optimize the fast-path attempt. |
4113 | */ |
4114 | if (unlikely(ac.nodemask != nodemask)) |
4115 | ac.nodemask = nodemask; |
4116 | |
4117 | page = __alloc_pages_slowpath(alloc_mask, order, &ac); |
4118 | |
4119 | out: |
4120 | if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page && |
4121 | unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) { |
4122 | __free_pages(page, order); |
4123 | page = NULL; |
4124 | } |
4125 | |
4126 | if (kmemcheck_enabled && page) |
4127 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); |
4128 | |
4129 | trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype); |
4130 | #ifdef CONFIG_AMLOGIC_PAGE_TRACE |
4131 | set_page_trace(page, order, gfp_mask, NULL); |
4132 | #endif /* CONFIG_AMLOGIC_PAGE_TRACE */ |
4133 | |
4134 | return page; |
4135 | } |
4136 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
4137 | |
4138 | /* |
4139 | * Common helper functions. |
4140 | */ |
4141 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
4142 | { |
4143 | struct page *page; |
4144 | |
4145 | /* |
4146 | * __get_free_pages() returns a 32-bit address, which cannot represent |
4147 | * a highmem page |
4148 | */ |
4149 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
4150 | |
4151 | page = alloc_pages(gfp_mask, order); |
4152 | if (!page) |
4153 | return 0; |
4154 | return (unsigned long) page_address(page); |
4155 | } |
4156 | EXPORT_SYMBOL(__get_free_pages); |
4157 | |
4158 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
4159 | { |
4160 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
4161 | } |
4162 | EXPORT_SYMBOL(get_zeroed_page); |
4163 | |
4164 | void __free_pages(struct page *page, unsigned int order) |
4165 | { |
4166 | if (put_page_testzero(page)) { |
4167 | if (order == 0) |
4168 | free_hot_cold_page(page, false); |
4169 | else |
4170 | __free_pages_ok(page, order); |
4171 | } |
4172 | } |
4173 | |
4174 | EXPORT_SYMBOL(__free_pages); |
4175 | |
4176 | void free_pages(unsigned long addr, unsigned int order) |
4177 | { |
4178 | if (addr != 0) { |
4179 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
4180 | __free_pages(virt_to_page((void *)addr), order); |
4181 | } |
4182 | } |
4183 | |
4184 | EXPORT_SYMBOL(free_pages); |
4185 | |
4186 | /* |
4187 | * Page Fragment: |
4188 | * An arbitrary-length arbitrary-offset area of memory which resides |
4189 | * within a 0 or higher order page. Multiple fragments within that page |
4190 | * are individually refcounted, in the page's reference counter. |
4191 | * |
4192 | * The page_frag functions below provide a simple allocation framework for |
4193 | * page fragments. This is used by the network stack and network device |
4194 | * drivers to provide a backing region of memory for use as either an |
4195 | * sk_buff->head, or to be used in the "frags" portion of skb_shared_info. |
4196 | */ |
4197 | static struct page *__page_frag_refill(struct page_frag_cache *nc, |
4198 | gfp_t gfp_mask) |
4199 | { |
4200 | struct page *page = NULL; |
4201 | gfp_t gfp = gfp_mask; |
4202 | |
4203 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
4204 | gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY | |
4205 | __GFP_NOMEMALLOC; |
4206 | page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, |
4207 | PAGE_FRAG_CACHE_MAX_ORDER); |
4208 | nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; |
4209 | #endif |
4210 | if (unlikely(!page)) |
4211 | page = alloc_pages_node(NUMA_NO_NODE, gfp, 0); |
4212 | |
4213 | nc->va = page ? page_address(page) : NULL; |
4214 | |
4215 | return page; |
4216 | } |
4217 | |
4218 | void *__alloc_page_frag(struct page_frag_cache *nc, |
4219 | unsigned int fragsz, gfp_t gfp_mask) |
4220 | { |
4221 | unsigned int size = PAGE_SIZE; |
4222 | struct page *page; |
4223 | int offset; |
4224 | |
4225 | if (unlikely(!nc->va)) { |
4226 | refill: |
4227 | page = __page_frag_refill(nc, gfp_mask); |
4228 | if (!page) |
4229 | return NULL; |
4230 | |
4231 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
4232 | /* if size can vary use size else just use PAGE_SIZE */ |
4233 | size = nc->size; |
4234 | #endif |
4235 | /* Even if we own the page, we do not use atomic_set(). |
4236 | * This would break get_page_unless_zero() users. |
4237 | */ |
4238 | page_ref_add(page, size); |
4239 | |
4240 | /* reset page count bias and offset to start of new frag */ |
4241 | nc->pfmemalloc = page_is_pfmemalloc(page); |
4242 | nc->pagecnt_bias = size + 1; |
4243 | nc->offset = size; |
4244 | } |
4245 | |
4246 | offset = nc->offset - fragsz; |
4247 | if (unlikely(offset < 0)) { |
4248 | page = virt_to_page(nc->va); |
4249 | |
4250 | if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) |
4251 | goto refill; |
4252 | |
4253 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) |
4254 | /* if size can vary use size else just use PAGE_SIZE */ |
4255 | size = nc->size; |
4256 | #endif |
4257 | /* OK, page count is 0, we can safely set it */ |
4258 | set_page_count(page, size + 1); |
4259 | |
4260 | /* reset page count bias and offset to start of new frag */ |
4261 | nc->pagecnt_bias = size + 1; |
4262 | offset = size - fragsz; |
4263 | } |
4264 | |
4265 | nc->pagecnt_bias--; |
4266 | nc->offset = offset; |
4267 | |
4268 | return nc->va + offset; |
4269 | } |
4270 | EXPORT_SYMBOL(__alloc_page_frag); |
4271 | |
4272 | /* |
4273 | * Frees a page fragment allocated out of either a compound or order 0 page. |
4274 | */ |
4275 | void __free_page_frag(void *addr) |
4276 | { |
4277 | struct page *page = virt_to_head_page(addr); |
4278 | |
4279 | if (unlikely(put_page_testzero(page))) |
4280 | __free_pages_ok(page, compound_order(page)); |
4281 | } |
4282 | EXPORT_SYMBOL(__free_page_frag); |
4283 | |
4284 | static void *make_alloc_exact(unsigned long addr, unsigned int order, |
4285 | size_t size) |
4286 | { |
4287 | if (addr) { |
4288 | unsigned long alloc_end = addr + (PAGE_SIZE << order); |
4289 | unsigned long used = addr + PAGE_ALIGN(size); |
4290 | |
4291 | split_page(virt_to_page((void *)addr), order); |
4292 | while (used < alloc_end) { |
4293 | free_page(used); |
4294 | used += PAGE_SIZE; |
4295 | } |
4296 | } |
4297 | return (void *)addr; |
4298 | } |
4299 | |
4300 | /** |
4301 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. |
4302 | * @size: the number of bytes to allocate |
4303 | * @gfp_mask: GFP flags for the allocation |
4304 | * |
4305 | * This function is similar to alloc_pages(), except that it allocates the |
4306 | * minimum number of pages to satisfy the request. alloc_pages() can only |
4307 | * allocate memory in power-of-two pages. |
4308 | * |
4309 | * This function is also limited by MAX_ORDER. |
4310 | * |
4311 | * Memory allocated by this function must be released by free_pages_exact(). |
4312 | */ |
4313 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) |
4314 | { |
4315 | unsigned int order = get_order(size); |
4316 | unsigned long addr; |
4317 | |
4318 | addr = __get_free_pages(gfp_mask, order); |
4319 | return make_alloc_exact(addr, order, size); |
4320 | } |
4321 | EXPORT_SYMBOL(alloc_pages_exact); |
4322 | |
4323 | /** |
4324 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous |
4325 | * pages on a node. |
4326 | * @nid: the preferred node ID where memory should be allocated |
4327 | * @size: the number of bytes to allocate |
4328 | * @gfp_mask: GFP flags for the allocation |
4329 | * |
4330 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling |
4331 | * back. |
4332 | */ |
4333 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
4334 | { |
4335 | unsigned int order = get_order(size); |
4336 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
4337 | if (!p) |
4338 | return NULL; |
4339 | return make_alloc_exact((unsigned long)page_address(p), order, size); |
4340 | } |
4341 | |
4342 | /** |
4343 | * free_pages_exact - release memory allocated via alloc_pages_exact() |
4344 | * @virt: the value returned by alloc_pages_exact. |
4345 | * @size: size of allocation, same value as passed to alloc_pages_exact(). |
4346 | * |
4347 | * Release the memory allocated by a previous call to alloc_pages_exact. |
4348 | */ |
4349 | void free_pages_exact(void *virt, size_t size) |
4350 | { |
4351 | unsigned long addr = (unsigned long)virt; |
4352 | unsigned long end = addr + PAGE_ALIGN(size); |
4353 | |
4354 | while (addr < end) { |
4355 | free_page(addr); |
4356 | addr += PAGE_SIZE; |
4357 | } |
4358 | } |
4359 | EXPORT_SYMBOL(free_pages_exact); |
4360 | |
4361 | /** |
4362 | * nr_free_zone_pages - count number of pages beyond high watermark |
4363 | * @offset: The zone index of the highest zone |
4364 | * |
4365 | * nr_free_zone_pages() counts the number of counts pages which are beyond the |
4366 | * high watermark within all zones at or below a given zone index. For each |
4367 | * zone, the number of pages is calculated as: |
4368 | * managed_pages - high_pages |
4369 | */ |
4370 | static unsigned long nr_free_zone_pages(int offset) |
4371 | { |
4372 | struct zoneref *z; |
4373 | struct zone *zone; |
4374 | |
4375 | /* Just pick one node, since fallback list is circular */ |
4376 | unsigned long sum = 0; |
4377 | |
4378 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
4379 | |
4380 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
4381 | unsigned long size = zone->managed_pages; |
4382 | unsigned long high = high_wmark_pages(zone); |
4383 | if (size > high) |
4384 | sum += size - high; |
4385 | } |
4386 | |
4387 | return sum; |
4388 | } |
4389 | |
4390 | /** |
4391 | * nr_free_buffer_pages - count number of pages beyond high watermark |
4392 | * |
4393 | * nr_free_buffer_pages() counts the number of pages which are beyond the high |
4394 | * watermark within ZONE_DMA and ZONE_NORMAL. |
4395 | */ |
4396 | unsigned long nr_free_buffer_pages(void) |
4397 | { |
4398 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
4399 | } |
4400 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
4401 | |
4402 | /** |
4403 | * nr_free_pagecache_pages - count number of pages beyond high watermark |
4404 | * |
4405 | * nr_free_pagecache_pages() counts the number of pages which are beyond the |
4406 | * high watermark within all zones. |
4407 | */ |
4408 | unsigned long nr_free_pagecache_pages(void) |
4409 | { |
4410 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
4411 | } |
4412 | |
4413 | static inline void show_node(struct zone *zone) |
4414 | { |
4415 | if (IS_ENABLED(CONFIG_NUMA)) |
4416 | printk("Node %d ", zone_to_nid(zone)); |
4417 | } |
4418 | |
4419 | long si_mem_available(void) |
4420 | { |
4421 | long available; |
4422 | unsigned long pagecache; |
4423 | unsigned long wmark_low = 0; |
4424 | unsigned long pages[NR_LRU_LISTS]; |
4425 | struct zone *zone; |
4426 | int lru; |
4427 | |
4428 | for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++) |
4429 | pages[lru] = global_node_page_state(NR_LRU_BASE + lru); |
4430 | |
4431 | for_each_zone(zone) |
4432 | wmark_low += zone->watermark[WMARK_LOW]; |
4433 | |
4434 | /* |
4435 | * Estimate the amount of memory available for userspace allocations, |
4436 | * without causing swapping. |
4437 | */ |
4438 | available = global_page_state(NR_FREE_PAGES) - totalreserve_pages; |
4439 | |
4440 | /* |
4441 | * Not all the page cache can be freed, otherwise the system will |
4442 | * start swapping. Assume at least half of the page cache, or the |
4443 | * low watermark worth of cache, needs to stay. |
4444 | */ |
4445 | pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE]; |
4446 | pagecache -= min(pagecache / 2, wmark_low); |
4447 | available += pagecache; |
4448 | |
4449 | /* |
4450 | * Part of the reclaimable slab consists of items that are in use, |
4451 | * and cannot be freed. Cap this estimate at the low watermark. |
4452 | */ |
4453 | available += global_page_state(NR_SLAB_RECLAIMABLE) - |
4454 | min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low); |
4455 | |
4456 | if (available < 0) |
4457 | available = 0; |
4458 | return available; |
4459 | } |
4460 | EXPORT_SYMBOL_GPL(si_mem_available); |
4461 | |
4462 | void si_meminfo(struct sysinfo *val) |
4463 | { |
4464 | val->totalram = totalram_pages; |
4465 | val->sharedram = global_node_page_state(NR_SHMEM); |
4466 | val->freeram = global_page_state(NR_FREE_PAGES); |
4467 | val->bufferram = nr_blockdev_pages(); |
4468 | val->totalhigh = totalhigh_pages; |
4469 | val->freehigh = nr_free_highpages(); |
4470 | val->mem_unit = PAGE_SIZE; |
4471 | } |
4472 | |
4473 | EXPORT_SYMBOL(si_meminfo); |
4474 | |
4475 | #ifdef CONFIG_NUMA |
4476 | void si_meminfo_node(struct sysinfo *val, int nid) |
4477 | { |
4478 | int zone_type; /* needs to be signed */ |
4479 | unsigned long managed_pages = 0; |
4480 | unsigned long managed_highpages = 0; |
4481 | unsigned long free_highpages = 0; |
4482 | pg_data_t *pgdat = NODE_DATA(nid); |
4483 | |
4484 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) |
4485 | managed_pages += pgdat->node_zones[zone_type].managed_pages; |
4486 | val->totalram = managed_pages; |
4487 | val->sharedram = node_page_state(pgdat, NR_SHMEM); |
4488 | val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); |
4489 | #ifdef CONFIG_HIGHMEM |
4490 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
4491 | struct zone *zone = &pgdat->node_zones[zone_type]; |
4492 | |
4493 | if (is_highmem(zone)) { |
4494 | managed_highpages += zone->managed_pages; |
4495 | free_highpages += zone_page_state(zone, NR_FREE_PAGES); |
4496 | } |
4497 | } |
4498 | val->totalhigh = managed_highpages; |
4499 | val->freehigh = free_highpages; |
4500 | #else |
4501 | val->totalhigh = managed_highpages; |
4502 | val->freehigh = free_highpages; |
4503 | #endif |
4504 | val->mem_unit = PAGE_SIZE; |
4505 | } |
4506 | #endif |
4507 | |
4508 | /* |
4509 | * Determine whether the node should be displayed or not, depending on whether |
4510 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). |
4511 | */ |
4512 | bool skip_free_areas_node(unsigned int flags, int nid) |
4513 | { |
4514 | bool ret = false; |
4515 | unsigned int cpuset_mems_cookie; |
4516 | |
4517 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
4518 | goto out; |
4519 | |
4520 | do { |
4521 | cpuset_mems_cookie = read_mems_allowed_begin(); |
4522 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
4523 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
4524 | out: |
4525 | return ret; |
4526 | } |
4527 | |
4528 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
4529 | |
4530 | static void show_migration_types(unsigned char type) |
4531 | { |
4532 | static const char types[MIGRATE_TYPES] = { |
4533 | [MIGRATE_UNMOVABLE] = 'U', |
4534 | [MIGRATE_MOVABLE] = 'M', |
4535 | [MIGRATE_RECLAIMABLE] = 'E', |
4536 | [MIGRATE_HIGHATOMIC] = 'H', |
4537 | #ifdef CONFIG_CMA |
4538 | [MIGRATE_CMA] = 'C', |
4539 | #endif |
4540 | #ifdef CONFIG_MEMORY_ISOLATION |
4541 | [MIGRATE_ISOLATE] = 'I', |
4542 | #endif |
4543 | }; |
4544 | char tmp[MIGRATE_TYPES + 1]; |
4545 | char *p = tmp; |
4546 | int i; |
4547 | |
4548 | for (i = 0; i < MIGRATE_TYPES; i++) { |
4549 | if (type & (1 << i)) |
4550 | *p++ = types[i]; |
4551 | } |
4552 | |
4553 | *p = '\0'; |
4554 | printk(KERN_CONT "(%s) ", tmp); |
4555 | } |
4556 | |
4557 | /* |
4558 | * Show free area list (used inside shift_scroll-lock stuff) |
4559 | * We also calculate the percentage fragmentation. We do this by counting the |
4560 | * memory on each free list with the exception of the first item on the list. |
4561 | * |
4562 | * Bits in @filter: |
4563 | * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's |
4564 | * cpuset. |
4565 | */ |
4566 | void show_free_areas(unsigned int filter) |
4567 | { |
4568 | unsigned long free_pcp = 0; |
4569 | int cpu; |
4570 | struct zone *zone; |
4571 | pg_data_t *pgdat; |
4572 | |
4573 | for_each_populated_zone(zone) { |
4574 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
4575 | continue; |
4576 | |
4577 | for_each_online_cpu(cpu) |
4578 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; |
4579 | } |
4580 | |
4581 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
4582 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" |
4583 | " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
4584 | " slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
4585 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n" |
4586 | #ifdef CONFIG_AMLOGIC_CMA |
4587 | " [cma] driver:%lu anon:%lu file:%lu isolate:%lu total:%lu\n" |
4588 | #endif /* CONFIG_AMLOGIC_CMA */ |
4589 | " free:%lu free_pcp:%lu free_cma:%lu\n", |
4590 | global_node_page_state(NR_ACTIVE_ANON), |
4591 | global_node_page_state(NR_INACTIVE_ANON), |
4592 | global_node_page_state(NR_ISOLATED_ANON), |
4593 | global_node_page_state(NR_ACTIVE_FILE), |
4594 | global_node_page_state(NR_INACTIVE_FILE), |
4595 | global_node_page_state(NR_ISOLATED_FILE), |
4596 | global_node_page_state(NR_UNEVICTABLE), |
4597 | global_node_page_state(NR_FILE_DIRTY), |
4598 | global_node_page_state(NR_WRITEBACK), |
4599 | global_node_page_state(NR_UNSTABLE_NFS), |
4600 | global_page_state(NR_SLAB_RECLAIMABLE), |
4601 | global_page_state(NR_SLAB_UNRECLAIMABLE), |
4602 | global_node_page_state(NR_FILE_MAPPED), |
4603 | global_node_page_state(NR_SHMEM), |
4604 | global_page_state(NR_PAGETABLE), |
4605 | global_page_state(NR_BOUNCE), |
4606 | #ifdef CONFIG_AMLOGIC_CMA |
4607 | get_cma_allocated(), |
4608 | global_page_state(NR_INACTIVE_ANON_CMA) + |
4609 | global_page_state(NR_ACTIVE_ANON_CMA), |
4610 | global_page_state(NR_INACTIVE_FILE_CMA) + |
4611 | global_page_state(NR_ACTIVE_FILE_CMA), |
4612 | global_page_state(NR_CMA_ISOLATED), |
4613 | totalcma_pages, |
4614 | #endif /* CONFIG_AMLOGIC_CMA */ |
4615 | global_page_state(NR_FREE_PAGES), |
4616 | free_pcp, |
4617 | global_page_state(NR_FREE_CMA_PAGES)); |
4618 | |
4619 | for_each_online_pgdat(pgdat) { |
4620 | printk("Node %d" |
4621 | " active_anon:%lukB" |
4622 | " inactive_anon:%lukB" |
4623 | " active_file:%lukB" |
4624 | " inactive_file:%lukB" |
4625 | " unevictable:%lukB" |
4626 | " isolated(anon):%lukB" |
4627 | " isolated(file):%lukB" |
4628 | " mapped:%lukB" |
4629 | " dirty:%lukB" |
4630 | " writeback:%lukB" |
4631 | " shmem:%lukB" |
4632 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4633 | " shmem_thp: %lukB" |
4634 | " shmem_pmdmapped: %lukB" |
4635 | " anon_thp: %lukB" |
4636 | #endif |
4637 | " writeback_tmp:%lukB" |
4638 | " unstable:%lukB" |
4639 | " pages_scanned:%lu" |
4640 | " all_unreclaimable? %s" |
4641 | "\n", |
4642 | pgdat->node_id, |
4643 | K(node_page_state(pgdat, NR_ACTIVE_ANON)), |
4644 | K(node_page_state(pgdat, NR_INACTIVE_ANON)), |
4645 | K(node_page_state(pgdat, NR_ACTIVE_FILE)), |
4646 | K(node_page_state(pgdat, NR_INACTIVE_FILE)), |
4647 | K(node_page_state(pgdat, NR_UNEVICTABLE)), |
4648 | K(node_page_state(pgdat, NR_ISOLATED_ANON)), |
4649 | K(node_page_state(pgdat, NR_ISOLATED_FILE)), |
4650 | K(node_page_state(pgdat, NR_FILE_MAPPED)), |
4651 | K(node_page_state(pgdat, NR_FILE_DIRTY)), |
4652 | K(node_page_state(pgdat, NR_WRITEBACK)), |
4653 | K(node_page_state(pgdat, NR_SHMEM)), |
4654 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4655 | K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR), |
4656 | K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) |
4657 | * HPAGE_PMD_NR), |
4658 | K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR), |
4659 | #endif |
4660 | K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), |
4661 | K(node_page_state(pgdat, NR_UNSTABLE_NFS)), |
4662 | node_page_state(pgdat, NR_PAGES_SCANNED), |
4663 | pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ? |
4664 | "yes" : "no"); |
4665 | } |
4666 | |
4667 | for_each_populated_zone(zone) { |
4668 | int i; |
4669 | |
4670 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
4671 | continue; |
4672 | |
4673 | free_pcp = 0; |
4674 | for_each_online_cpu(cpu) |
4675 | free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count; |
4676 | |
4677 | show_node(zone); |
4678 | printk(KERN_CONT |
4679 | "%s" |
4680 | " free:%lukB" |
4681 | " min:%lukB" |
4682 | " low:%lukB" |
4683 | " high:%lukB" |
4684 | " active_anon:%lukB" |
4685 | " inactive_anon:%lukB" |
4686 | " active_file:%lukB" |
4687 | " inactive_file:%lukB" |
4688 | " unevictable:%lukB" |
4689 | " writepending:%lukB" |
4690 | " present:%lukB" |
4691 | " managed:%lukB" |
4692 | " mlocked:%lukB" |
4693 | " slab_reclaimable:%lukB" |
4694 | " slab_unreclaimable:%lukB" |
4695 | " kernel_stack:%lukB" |
4696 | " pagetables:%lukB" |
4697 | " bounce:%lukB" |
4698 | " free_pcp:%lukB" |
4699 | " local_pcp:%ukB" |
4700 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
4701 | " free_unmovable:%lukB" |
4702 | " free_movable:%lukB" |
4703 | " free_reclaimable:%lukB" |
4704 | " free_highatomic:%lukB" |
4705 | #ifdef CONFIG_MEMORY_ISOLATION |
4706 | " free_isolate:%lukB" |
4707 | #endif |
4708 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
4709 | " free_cma:%lukB" |
4710 | "\n", |
4711 | zone->name, |
4712 | K(zone_page_state(zone, NR_FREE_PAGES)), |
4713 | K(min_wmark_pages(zone)), |
4714 | K(low_wmark_pages(zone)), |
4715 | K(high_wmark_pages(zone)), |
4716 | K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)), |
4717 | K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)), |
4718 | K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)), |
4719 | K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)), |
4720 | K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)), |
4721 | K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)), |
4722 | K(zone->present_pages), |
4723 | K(zone->managed_pages), |
4724 | K(zone_page_state(zone, NR_MLOCK)), |
4725 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
4726 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), |
4727 | zone_page_state(zone, NR_KERNEL_STACK_KB), |
4728 | K(zone_page_state(zone, NR_PAGETABLE)), |
4729 | K(zone_page_state(zone, NR_BOUNCE)), |
4730 | K(free_pcp), |
4731 | K(this_cpu_read(zone->pageset->pcp.count)), |
4732 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
4733 | K(zone_page_state(zone, NR_FREE_UNMOVABLE)), |
4734 | K(zone_page_state(zone, NR_FREE_MOVABLE)), |
4735 | K(zone_page_state(zone, NR_FREE_RECLAIMABLE)), |
4736 | K(zone_page_state(zone, NR_FREE_HIGHATOMIC)), |
4737 | #ifdef CONFIG_MEMORY_ISOLATION |
4738 | K(zone_page_state(zone, NR_FREE_ISOLATE)), |
4739 | #endif |
4740 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
4741 | K(zone_page_state(zone, NR_FREE_CMA_PAGES))); |
4742 | printk("lowmem_reserve[]:"); |
4743 | for (i = 0; i < MAX_NR_ZONES; i++) |
4744 | printk(KERN_CONT " %ld", zone->lowmem_reserve[i]); |
4745 | printk(KERN_CONT "\n"); |
4746 | } |
4747 | |
4748 | for_each_populated_zone(zone) { |
4749 | unsigned int order; |
4750 | unsigned long nr[MAX_ORDER], flags, total = 0; |
4751 | unsigned char types[MAX_ORDER]; |
4752 | |
4753 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
4754 | continue; |
4755 | show_node(zone); |
4756 | printk(KERN_CONT "%s: ", zone->name); |
4757 | |
4758 | spin_lock_irqsave(&zone->lock, flags); |
4759 | for (order = 0; order < MAX_ORDER; order++) { |
4760 | struct free_area *area = &zone->free_area[order]; |
4761 | int type; |
4762 | |
4763 | nr[order] = area->nr_free; |
4764 | total += nr[order] << order; |
4765 | |
4766 | types[order] = 0; |
4767 | for (type = 0; type < MIGRATE_TYPES; type++) { |
4768 | if (!list_empty(&area->free_list[type])) |
4769 | types[order] |= 1 << type; |
4770 | } |
4771 | } |
4772 | spin_unlock_irqrestore(&zone->lock, flags); |
4773 | for (order = 0; order < MAX_ORDER; order++) { |
4774 | printk(KERN_CONT "%lu*%lukB ", |
4775 | nr[order], K(1UL) << order); |
4776 | if (nr[order]) |
4777 | show_migration_types(types[order]); |
4778 | } |
4779 | printk(KERN_CONT "= %lukB\n", K(total)); |
4780 | } |
4781 | |
4782 | hugetlb_show_meminfo(); |
4783 | |
4784 | printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES)); |
4785 | |
4786 | show_swap_cache_info(); |
4787 | } |
4788 | |
4789 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
4790 | { |
4791 | zoneref->zone = zone; |
4792 | zoneref->zone_idx = zone_idx(zone); |
4793 | } |
4794 | |
4795 | /* |
4796 | * Builds allocation fallback zone lists. |
4797 | * |
4798 | * Add all populated zones of a node to the zonelist. |
4799 | */ |
4800 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
4801 | int nr_zones) |
4802 | { |
4803 | struct zone *zone; |
4804 | enum zone_type zone_type = MAX_NR_ZONES; |
4805 | |
4806 | do { |
4807 | zone_type--; |
4808 | zone = pgdat->node_zones + zone_type; |
4809 | if (managed_zone(zone)) { |
4810 | zoneref_set_zone(zone, |
4811 | &zonelist->_zonerefs[nr_zones++]); |
4812 | check_highest_zone(zone_type); |
4813 | } |
4814 | } while (zone_type); |
4815 | |
4816 | return nr_zones; |
4817 | } |
4818 | |
4819 | |
4820 | /* |
4821 | * zonelist_order: |
4822 | * 0 = automatic detection of better ordering. |
4823 | * 1 = order by ([node] distance, -zonetype) |
4824 | * 2 = order by (-zonetype, [node] distance) |
4825 | * |
4826 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create |
4827 | * the same zonelist. So only NUMA can configure this param. |
4828 | */ |
4829 | #define ZONELIST_ORDER_DEFAULT 0 |
4830 | #define ZONELIST_ORDER_NODE 1 |
4831 | #define ZONELIST_ORDER_ZONE 2 |
4832 | |
4833 | /* zonelist order in the kernel. |
4834 | * set_zonelist_order() will set this to NODE or ZONE. |
4835 | */ |
4836 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; |
4837 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; |
4838 | |
4839 | |
4840 | #ifdef CONFIG_NUMA |
4841 | /* The value user specified ....changed by config */ |
4842 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
4843 | /* string for sysctl */ |
4844 | #define NUMA_ZONELIST_ORDER_LEN 16 |
4845 | char numa_zonelist_order[16] = "default"; |
4846 | |
4847 | /* |
4848 | * interface for configure zonelist ordering. |
4849 | * command line option "numa_zonelist_order" |
4850 | * = "[dD]efault - default, automatic configuration. |
4851 | * = "[nN]ode - order by node locality, then by zone within node |
4852 | * = "[zZ]one - order by zone, then by locality within zone |
4853 | */ |
4854 | |
4855 | static int __parse_numa_zonelist_order(char *s) |
4856 | { |
4857 | if (*s == 'd' || *s == 'D') { |
4858 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
4859 | } else if (*s == 'n' || *s == 'N') { |
4860 | user_zonelist_order = ZONELIST_ORDER_NODE; |
4861 | } else if (*s == 'z' || *s == 'Z') { |
4862 | user_zonelist_order = ZONELIST_ORDER_ZONE; |
4863 | } else { |
4864 | pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s); |
4865 | return -EINVAL; |
4866 | } |
4867 | return 0; |
4868 | } |
4869 | |
4870 | static __init int setup_numa_zonelist_order(char *s) |
4871 | { |
4872 | int ret; |
4873 | |
4874 | if (!s) |
4875 | return 0; |
4876 | |
4877 | ret = __parse_numa_zonelist_order(s); |
4878 | if (ret == 0) |
4879 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); |
4880 | |
4881 | return ret; |
4882 | } |
4883 | early_param("numa_zonelist_order", setup_numa_zonelist_order); |
4884 | |
4885 | /* |
4886 | * sysctl handler for numa_zonelist_order |
4887 | */ |
4888 | int numa_zonelist_order_handler(struct ctl_table *table, int write, |
4889 | void __user *buffer, size_t *length, |
4890 | loff_t *ppos) |
4891 | { |
4892 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; |
4893 | int ret; |
4894 | static DEFINE_MUTEX(zl_order_mutex); |
4895 | |
4896 | mutex_lock(&zl_order_mutex); |
4897 | if (write) { |
4898 | if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) { |
4899 | ret = -EINVAL; |
4900 | goto out; |
4901 | } |
4902 | strcpy(saved_string, (char *)table->data); |
4903 | } |
4904 | ret = proc_dostring(table, write, buffer, length, ppos); |
4905 | if (ret) |
4906 | goto out; |
4907 | if (write) { |
4908 | int oldval = user_zonelist_order; |
4909 | |
4910 | ret = __parse_numa_zonelist_order((char *)table->data); |
4911 | if (ret) { |
4912 | /* |
4913 | * bogus value. restore saved string |
4914 | */ |
4915 | strncpy((char *)table->data, saved_string, |
4916 | NUMA_ZONELIST_ORDER_LEN); |
4917 | user_zonelist_order = oldval; |
4918 | } else if (oldval != user_zonelist_order) { |
4919 | mutex_lock(&zonelists_mutex); |
4920 | build_all_zonelists(NULL, NULL); |
4921 | mutex_unlock(&zonelists_mutex); |
4922 | } |
4923 | } |
4924 | out: |
4925 | mutex_unlock(&zl_order_mutex); |
4926 | return ret; |
4927 | } |
4928 | |
4929 | |
4930 | #define MAX_NODE_LOAD (nr_online_nodes) |
4931 | static int node_load[MAX_NUMNODES]; |
4932 | |
4933 | /** |
4934 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
4935 | * @node: node whose fallback list we're appending |
4936 | * @used_node_mask: nodemask_t of already used nodes |
4937 | * |
4938 | * We use a number of factors to determine which is the next node that should |
4939 | * appear on a given node's fallback list. The node should not have appeared |
4940 | * already in @node's fallback list, and it should be the next closest node |
4941 | * according to the distance array (which contains arbitrary distance values |
4942 | * from each node to each node in the system), and should also prefer nodes |
4943 | * with no CPUs, since presumably they'll have very little allocation pressure |
4944 | * on them otherwise. |
4945 | * It returns -1 if no node is found. |
4946 | */ |
4947 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
4948 | { |
4949 | int n, val; |
4950 | int min_val = INT_MAX; |
4951 | int best_node = NUMA_NO_NODE; |
4952 | const struct cpumask *tmp = cpumask_of_node(0); |
4953 | |
4954 | /* Use the local node if we haven't already */ |
4955 | if (!node_isset(node, *used_node_mask)) { |
4956 | node_set(node, *used_node_mask); |
4957 | return node; |
4958 | } |
4959 | |
4960 | for_each_node_state(n, N_MEMORY) { |
4961 | |
4962 | /* Don't want a node to appear more than once */ |
4963 | if (node_isset(n, *used_node_mask)) |
4964 | continue; |
4965 | |
4966 | /* Use the distance array to find the distance */ |
4967 | val = node_distance(node, n); |
4968 | |
4969 | /* Penalize nodes under us ("prefer the next node") */ |
4970 | val += (n < node); |
4971 | |
4972 | /* Give preference to headless and unused nodes */ |
4973 | tmp = cpumask_of_node(n); |
4974 | if (!cpumask_empty(tmp)) |
4975 | val += PENALTY_FOR_NODE_WITH_CPUS; |
4976 | |
4977 | /* Slight preference for less loaded node */ |
4978 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); |
4979 | val += node_load[n]; |
4980 | |
4981 | if (val < min_val) { |
4982 | min_val = val; |
4983 | best_node = n; |
4984 | } |
4985 | } |
4986 | |
4987 | if (best_node >= 0) |
4988 | node_set(best_node, *used_node_mask); |
4989 | |
4990 | return best_node; |
4991 | } |
4992 | |
4993 | |
4994 | /* |
4995 | * Build zonelists ordered by node and zones within node. |
4996 | * This results in maximum locality--normal zone overflows into local |
4997 | * DMA zone, if any--but risks exhausting DMA zone. |
4998 | */ |
4999 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) |
5000 | { |
5001 | int j; |
5002 | struct zonelist *zonelist; |
5003 | |
5004 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
5005 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
5006 | ; |
5007 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
5008 | zonelist->_zonerefs[j].zone = NULL; |
5009 | zonelist->_zonerefs[j].zone_idx = 0; |
5010 | } |
5011 | |
5012 | /* |
5013 | * Build gfp_thisnode zonelists |
5014 | */ |
5015 | static void build_thisnode_zonelists(pg_data_t *pgdat) |
5016 | { |
5017 | int j; |
5018 | struct zonelist *zonelist; |
5019 | |
5020 | zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK]; |
5021 | j = build_zonelists_node(pgdat, zonelist, 0); |
5022 | zonelist->_zonerefs[j].zone = NULL; |
5023 | zonelist->_zonerefs[j].zone_idx = 0; |
5024 | } |
5025 | |
5026 | /* |
5027 | * Build zonelists ordered by zone and nodes within zones. |
5028 | * This results in conserving DMA zone[s] until all Normal memory is |
5029 | * exhausted, but results in overflowing to remote node while memory |
5030 | * may still exist in local DMA zone. |
5031 | */ |
5032 | static int node_order[MAX_NUMNODES]; |
5033 | |
5034 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) |
5035 | { |
5036 | int pos, j, node; |
5037 | int zone_type; /* needs to be signed */ |
5038 | struct zone *z; |
5039 | struct zonelist *zonelist; |
5040 | |
5041 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
5042 | pos = 0; |
5043 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { |
5044 | for (j = 0; j < nr_nodes; j++) { |
5045 | node = node_order[j]; |
5046 | z = &NODE_DATA(node)->node_zones[zone_type]; |
5047 | if (managed_zone(z)) { |
5048 | zoneref_set_zone(z, |
5049 | &zonelist->_zonerefs[pos++]); |
5050 | check_highest_zone(zone_type); |
5051 | } |
5052 | } |
5053 | } |
5054 | zonelist->_zonerefs[pos].zone = NULL; |
5055 | zonelist->_zonerefs[pos].zone_idx = 0; |
5056 | } |
5057 | |
5058 | #if defined(CONFIG_64BIT) |
5059 | /* |
5060 | * Devices that require DMA32/DMA are relatively rare and do not justify a |
5061 | * penalty to every machine in case the specialised case applies. Default |
5062 | * to Node-ordering on 64-bit NUMA machines |
5063 | */ |
5064 | static int default_zonelist_order(void) |
5065 | { |
5066 | return ZONELIST_ORDER_NODE; |
5067 | } |
5068 | #else |
5069 | /* |
5070 | * On 32-bit, the Normal zone needs to be preserved for allocations accessible |
5071 | * by the kernel. If processes running on node 0 deplete the low memory zone |
5072 | * then reclaim will occur more frequency increasing stalls and potentially |
5073 | * be easier to OOM if a large percentage of the zone is under writeback or |
5074 | * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set. |
5075 | * Hence, default to zone ordering on 32-bit. |
5076 | */ |
5077 | static int default_zonelist_order(void) |
5078 | { |
5079 | return ZONELIST_ORDER_ZONE; |
5080 | } |
5081 | #endif /* CONFIG_64BIT */ |
5082 | |
5083 | static void set_zonelist_order(void) |
5084 | { |
5085 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) |
5086 | current_zonelist_order = default_zonelist_order(); |
5087 | else |
5088 | current_zonelist_order = user_zonelist_order; |
5089 | } |
5090 | |
5091 | static void build_zonelists(pg_data_t *pgdat) |
5092 | { |
5093 | int i, node, load; |
5094 | nodemask_t used_mask; |
5095 | int local_node, prev_node; |
5096 | struct zonelist *zonelist; |
5097 | unsigned int order = current_zonelist_order; |
5098 | |
5099 | /* initialize zonelists */ |
5100 | for (i = 0; i < MAX_ZONELISTS; i++) { |
5101 | zonelist = pgdat->node_zonelists + i; |
5102 | zonelist->_zonerefs[0].zone = NULL; |
5103 | zonelist->_zonerefs[0].zone_idx = 0; |
5104 | } |
5105 | |
5106 | /* NUMA-aware ordering of nodes */ |
5107 | local_node = pgdat->node_id; |
5108 | load = nr_online_nodes; |
5109 | prev_node = local_node; |
5110 | nodes_clear(used_mask); |
5111 | |
5112 | memset(node_order, 0, sizeof(node_order)); |
5113 | i = 0; |
5114 | |
5115 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
5116 | /* |
5117 | * We don't want to pressure a particular node. |
5118 | * So adding penalty to the first node in same |
5119 | * distance group to make it round-robin. |
5120 | */ |
5121 | if (node_distance(local_node, node) != |
5122 | node_distance(local_node, prev_node)) |
5123 | node_load[node] = load; |
5124 | |
5125 | prev_node = node; |
5126 | load--; |
5127 | if (order == ZONELIST_ORDER_NODE) |
5128 | build_zonelists_in_node_order(pgdat, node); |
5129 | else |
5130 | node_order[i++] = node; /* remember order */ |
5131 | } |
5132 | |
5133 | if (order == ZONELIST_ORDER_ZONE) { |
5134 | /* calculate node order -- i.e., DMA last! */ |
5135 | build_zonelists_in_zone_order(pgdat, i); |
5136 | } |
5137 | |
5138 | build_thisnode_zonelists(pgdat); |
5139 | } |
5140 | |
5141 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5142 | /* |
5143 | * Return node id of node used for "local" allocations. |
5144 | * I.e., first node id of first zone in arg node's generic zonelist. |
5145 | * Used for initializing percpu 'numa_mem', which is used primarily |
5146 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. |
5147 | */ |
5148 | int local_memory_node(int node) |
5149 | { |
5150 | struct zoneref *z; |
5151 | |
5152 | z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
5153 | gfp_zone(GFP_KERNEL), |
5154 | NULL); |
5155 | return z->zone->node; |
5156 | } |
5157 | #endif |
5158 | |
5159 | static void setup_min_unmapped_ratio(void); |
5160 | static void setup_min_slab_ratio(void); |
5161 | #else /* CONFIG_NUMA */ |
5162 | |
5163 | static void set_zonelist_order(void) |
5164 | { |
5165 | current_zonelist_order = ZONELIST_ORDER_ZONE; |
5166 | } |
5167 | |
5168 | static void build_zonelists(pg_data_t *pgdat) |
5169 | { |
5170 | int node, local_node; |
5171 | enum zone_type j; |
5172 | struct zonelist *zonelist; |
5173 | |
5174 | local_node = pgdat->node_id; |
5175 | |
5176 | zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK]; |
5177 | j = build_zonelists_node(pgdat, zonelist, 0); |
5178 | |
5179 | /* |
5180 | * Now we build the zonelist so that it contains the zones |
5181 | * of all the other nodes. |
5182 | * We don't want to pressure a particular node, so when |
5183 | * building the zones for node N, we make sure that the |
5184 | * zones coming right after the local ones are those from |
5185 | * node N+1 (modulo N) |
5186 | */ |
5187 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { |
5188 | if (!node_online(node)) |
5189 | continue; |
5190 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
5191 | } |
5192 | for (node = 0; node < local_node; node++) { |
5193 | if (!node_online(node)) |
5194 | continue; |
5195 | j = build_zonelists_node(NODE_DATA(node), zonelist, j); |
5196 | } |
5197 | |
5198 | zonelist->_zonerefs[j].zone = NULL; |
5199 | zonelist->_zonerefs[j].zone_idx = 0; |
5200 | } |
5201 | |
5202 | #endif /* CONFIG_NUMA */ |
5203 | |
5204 | /* |
5205 | * Boot pageset table. One per cpu which is going to be used for all |
5206 | * zones and all nodes. The parameters will be set in such a way |
5207 | * that an item put on a list will immediately be handed over to |
5208 | * the buddy list. This is safe since pageset manipulation is done |
5209 | * with interrupts disabled. |
5210 | * |
5211 | * The boot_pagesets must be kept even after bootup is complete for |
5212 | * unused processors and/or zones. They do play a role for bootstrapping |
5213 | * hotplugged processors. |
5214 | * |
5215 | * zoneinfo_show() and maybe other functions do |
5216 | * not check if the processor is online before following the pageset pointer. |
5217 | * Other parts of the kernel may not check if the zone is available. |
5218 | */ |
5219 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); |
5220 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); |
5221 | static void setup_zone_pageset(struct zone *zone); |
5222 | |
5223 | /* |
5224 | * Global mutex to protect against size modification of zonelists |
5225 | * as well as to serialize pageset setup for the new populated zone. |
5226 | */ |
5227 | DEFINE_MUTEX(zonelists_mutex); |
5228 | |
5229 | /* return values int ....just for stop_machine() */ |
5230 | static int __build_all_zonelists(void *data) |
5231 | { |
5232 | int nid; |
5233 | int cpu; |
5234 | pg_data_t *self = data; |
5235 | |
5236 | #ifdef CONFIG_NUMA |
5237 | memset(node_load, 0, sizeof(node_load)); |
5238 | #endif |
5239 | |
5240 | if (self && !node_online(self->node_id)) { |
5241 | build_zonelists(self); |
5242 | } |
5243 | |
5244 | for_each_online_node(nid) { |
5245 | pg_data_t *pgdat = NODE_DATA(nid); |
5246 | |
5247 | build_zonelists(pgdat); |
5248 | } |
5249 | |
5250 | /* |
5251 | * Initialize the boot_pagesets that are going to be used |
5252 | * for bootstrapping processors. The real pagesets for |
5253 | * each zone will be allocated later when the per cpu |
5254 | * allocator is available. |
5255 | * |
5256 | * boot_pagesets are used also for bootstrapping offline |
5257 | * cpus if the system is already booted because the pagesets |
5258 | * are needed to initialize allocators on a specific cpu too. |
5259 | * F.e. the percpu allocator needs the page allocator which |
5260 | * needs the percpu allocator in order to allocate its pagesets |
5261 | * (a chicken-egg dilemma). |
5262 | */ |
5263 | for_each_possible_cpu(cpu) { |
5264 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
5265 | |
5266 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
5267 | /* |
5268 | * We now know the "local memory node" for each node-- |
5269 | * i.e., the node of the first zone in the generic zonelist. |
5270 | * Set up numa_mem percpu variable for on-line cpus. During |
5271 | * boot, only the boot cpu should be on-line; we'll init the |
5272 | * secondary cpus' numa_mem as they come on-line. During |
5273 | * node/memory hotplug, we'll fixup all on-line cpus. |
5274 | */ |
5275 | if (cpu_online(cpu)) |
5276 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
5277 | #endif |
5278 | } |
5279 | |
5280 | return 0; |
5281 | } |
5282 | |
5283 | static noinline void __init |
5284 | build_all_zonelists_init(void) |
5285 | { |
5286 | __build_all_zonelists(NULL); |
5287 | mminit_verify_zonelist(); |
5288 | cpuset_init_current_mems_allowed(); |
5289 | } |
5290 | |
5291 | /* |
5292 | * Called with zonelists_mutex held always |
5293 | * unless system_state == SYSTEM_BOOTING. |
5294 | * |
5295 | * __ref due to (1) call of __meminit annotated setup_zone_pageset |
5296 | * [we're only called with non-NULL zone through __meminit paths] and |
5297 | * (2) call of __init annotated helper build_all_zonelists_init |
5298 | * [protected by SYSTEM_BOOTING]. |
5299 | */ |
5300 | void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone) |
5301 | { |
5302 | set_zonelist_order(); |
5303 | |
5304 | if (system_state == SYSTEM_BOOTING) { |
5305 | build_all_zonelists_init(); |
5306 | } else { |
5307 | #ifdef CONFIG_MEMORY_HOTPLUG |
5308 | if (zone) |
5309 | setup_zone_pageset(zone); |
5310 | #endif |
5311 | /* we have to stop all cpus to guarantee there is no user |
5312 | of zonelist */ |
5313 | stop_machine(__build_all_zonelists, pgdat, NULL); |
5314 | /* cpuset refresh routine should be here */ |
5315 | } |
5316 | vm_total_pages = nr_free_pagecache_pages(); |
5317 | /* |
5318 | * Disable grouping by mobility if the number of pages in the |
5319 | * system is too low to allow the mechanism to work. It would be |
5320 | * more accurate, but expensive to check per-zone. This check is |
5321 | * made on memory-hotadd so a system can start with mobility |
5322 | * disabled and enable it later |
5323 | */ |
5324 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
5325 | page_group_by_mobility_disabled = 1; |
5326 | else |
5327 | page_group_by_mobility_disabled = 0; |
5328 | |
5329 | pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n", |
5330 | nr_online_nodes, |
5331 | zonelist_order_name[current_zonelist_order], |
5332 | page_group_by_mobility_disabled ? "off" : "on", |
5333 | vm_total_pages); |
5334 | #ifdef CONFIG_NUMA |
5335 | pr_info("Policy zone: %s\n", zone_names[policy_zone]); |
5336 | #endif |
5337 | } |
5338 | |
5339 | /* |
5340 | * Initially all pages are reserved - free ones are freed |
5341 | * up by free_all_bootmem() once the early boot process is |
5342 | * done. Non-atomic initialization, single-pass. |
5343 | */ |
5344 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
5345 | unsigned long start_pfn, enum memmap_context context) |
5346 | { |
5347 | struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn)); |
5348 | unsigned long end_pfn = start_pfn + size; |
5349 | pg_data_t *pgdat = NODE_DATA(nid); |
5350 | unsigned long pfn; |
5351 | unsigned long nr_initialised = 0; |
5352 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5353 | struct memblock_region *r = NULL, *tmp; |
5354 | #endif |
5355 | |
5356 | if (highest_memmap_pfn < end_pfn - 1) |
5357 | highest_memmap_pfn = end_pfn - 1; |
5358 | |
5359 | /* |
5360 | * Honor reservation requested by the driver for this ZONE_DEVICE |
5361 | * memory |
5362 | */ |
5363 | if (altmap && start_pfn == altmap->base_pfn) |
5364 | start_pfn += altmap->reserve; |
5365 | |
5366 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
5367 | /* |
5368 | * There can be holes in boot-time mem_map[]s handed to this |
5369 | * function. They do not exist on hotplugged memory. |
5370 | */ |
5371 | if (context != MEMMAP_EARLY) |
5372 | goto not_early; |
5373 | |
5374 | if (!early_pfn_valid(pfn)) |
5375 | continue; |
5376 | if (!early_pfn_in_nid(pfn, nid)) |
5377 | continue; |
5378 | if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised)) |
5379 | break; |
5380 | |
5381 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5382 | /* |
5383 | * Check given memblock attribute by firmware which can affect |
5384 | * kernel memory layout. If zone==ZONE_MOVABLE but memory is |
5385 | * mirrored, it's an overlapped memmap init. skip it. |
5386 | */ |
5387 | if (mirrored_kernelcore && zone == ZONE_MOVABLE) { |
5388 | if (!r || pfn >= memblock_region_memory_end_pfn(r)) { |
5389 | for_each_memblock(memory, tmp) |
5390 | if (pfn < memblock_region_memory_end_pfn(tmp)) |
5391 | break; |
5392 | r = tmp; |
5393 | } |
5394 | if (pfn >= memblock_region_memory_base_pfn(r) && |
5395 | memblock_is_mirror(r)) { |
5396 | /* already initialized as NORMAL */ |
5397 | pfn = memblock_region_memory_end_pfn(r); |
5398 | continue; |
5399 | } |
5400 | } |
5401 | #endif |
5402 | |
5403 | not_early: |
5404 | /* |
5405 | * Mark the block movable so that blocks are reserved for |
5406 | * movable at startup. This will force kernel allocations |
5407 | * to reserve their blocks rather than leaking throughout |
5408 | * the address space during boot when many long-lived |
5409 | * kernel allocations are made. |
5410 | * |
5411 | * bitmap is created for zone's valid pfn range. but memmap |
5412 | * can be created for invalid pages (for alignment) |
5413 | * check here not to call set_pageblock_migratetype() against |
5414 | * pfn out of zone. |
5415 | */ |
5416 | if (!(pfn & (pageblock_nr_pages - 1))) { |
5417 | struct page *page = pfn_to_page(pfn); |
5418 | |
5419 | __init_single_page(page, pfn, zone, nid); |
5420 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
5421 | } else { |
5422 | __init_single_pfn(pfn, zone, nid); |
5423 | } |
5424 | } |
5425 | } |
5426 | |
5427 | static void __meminit zone_init_free_lists(struct zone *zone) |
5428 | { |
5429 | unsigned int order, t; |
5430 | for_each_migratetype_order(order, t) { |
5431 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); |
5432 | zone->free_area[order].nr_free = 0; |
5433 | } |
5434 | } |
5435 | |
5436 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
5437 | #define memmap_init(size, nid, zone, start_pfn) \ |
5438 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
5439 | #endif |
5440 | |
5441 | static int zone_batchsize(struct zone *zone) |
5442 | { |
5443 | #ifdef CONFIG_MMU |
5444 | int batch; |
5445 | |
5446 | /* |
5447 | * The per-cpu-pages pools are set to around 1000th of the |
5448 | * size of the zone. But no more than 1/2 of a meg. |
5449 | * |
5450 | * OK, so we don't know how big the cache is. So guess. |
5451 | */ |
5452 | batch = zone->managed_pages / 1024; |
5453 | if (batch * PAGE_SIZE > 512 * 1024) |
5454 | batch = (512 * 1024) / PAGE_SIZE; |
5455 | batch /= 4; /* We effectively *= 4 below */ |
5456 | if (batch < 1) |
5457 | batch = 1; |
5458 | |
5459 | /* |
5460 | * Clamp the batch to a 2^n - 1 value. Having a power |
5461 | * of 2 value was found to be more likely to have |
5462 | * suboptimal cache aliasing properties in some cases. |
5463 | * |
5464 | * For example if 2 tasks are alternately allocating |
5465 | * batches of pages, one task can end up with a lot |
5466 | * of pages of one half of the possible page colors |
5467 | * and the other with pages of the other colors. |
5468 | */ |
5469 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
5470 | |
5471 | return batch; |
5472 | |
5473 | #else |
5474 | /* The deferral and batching of frees should be suppressed under NOMMU |
5475 | * conditions. |
5476 | * |
5477 | * The problem is that NOMMU needs to be able to allocate large chunks |
5478 | * of contiguous memory as there's no hardware page translation to |
5479 | * assemble apparent contiguous memory from discontiguous pages. |
5480 | * |
5481 | * Queueing large contiguous runs of pages for batching, however, |
5482 | * causes the pages to actually be freed in smaller chunks. As there |
5483 | * can be a significant delay between the individual batches being |
5484 | * recycled, this leads to the once large chunks of space being |
5485 | * fragmented and becoming unavailable for high-order allocations. |
5486 | */ |
5487 | return 0; |
5488 | #endif |
5489 | } |
5490 | |
5491 | /* |
5492 | * pcp->high and pcp->batch values are related and dependent on one another: |
5493 | * ->batch must never be higher then ->high. |
5494 | * The following function updates them in a safe manner without read side |
5495 | * locking. |
5496 | * |
5497 | * Any new users of pcp->batch and pcp->high should ensure they can cope with |
5498 | * those fields changing asynchronously (acording the the above rule). |
5499 | * |
5500 | * mutex_is_locked(&pcp_batch_high_lock) required when calling this function |
5501 | * outside of boot time (or some other assurance that no concurrent updaters |
5502 | * exist). |
5503 | */ |
5504 | static void pageset_update(struct per_cpu_pages *pcp, unsigned long high, |
5505 | unsigned long batch) |
5506 | { |
5507 | /* start with a fail safe value for batch */ |
5508 | pcp->batch = 1; |
5509 | smp_wmb(); |
5510 | |
5511 | /* Update high, then batch, in order */ |
5512 | pcp->high = high; |
5513 | smp_wmb(); |
5514 | |
5515 | pcp->batch = batch; |
5516 | } |
5517 | |
5518 | /* a companion to pageset_set_high() */ |
5519 | static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch) |
5520 | { |
5521 | pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch)); |
5522 | } |
5523 | |
5524 | static void pageset_init(struct per_cpu_pageset *p) |
5525 | { |
5526 | struct per_cpu_pages *pcp; |
5527 | int migratetype; |
5528 | |
5529 | memset(p, 0, sizeof(*p)); |
5530 | |
5531 | pcp = &p->pcp; |
5532 | pcp->count = 0; |
5533 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
5534 | INIT_LIST_HEAD(&pcp->lists[migratetype]); |
5535 | } |
5536 | |
5537 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
5538 | { |
5539 | pageset_init(p); |
5540 | pageset_set_batch(p, batch); |
5541 | } |
5542 | |
5543 | /* |
5544 | * pageset_set_high() sets the high water mark for hot per_cpu_pagelist |
5545 | * to the value high for the pageset p. |
5546 | */ |
5547 | static void pageset_set_high(struct per_cpu_pageset *p, |
5548 | unsigned long high) |
5549 | { |
5550 | unsigned long batch = max(1UL, high / 4); |
5551 | if ((high / 4) > (PAGE_SHIFT * 8)) |
5552 | batch = PAGE_SHIFT * 8; |
5553 | |
5554 | pageset_update(&p->pcp, high, batch); |
5555 | } |
5556 | |
5557 | static void pageset_set_high_and_batch(struct zone *zone, |
5558 | struct per_cpu_pageset *pcp) |
5559 | { |
5560 | if (percpu_pagelist_fraction) |
5561 | pageset_set_high(pcp, |
5562 | (zone->managed_pages / |
5563 | percpu_pagelist_fraction)); |
5564 | else |
5565 | pageset_set_batch(pcp, zone_batchsize(zone)); |
5566 | } |
5567 | |
5568 | static void __meminit zone_pageset_init(struct zone *zone, int cpu) |
5569 | { |
5570 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); |
5571 | |
5572 | pageset_init(pcp); |
5573 | pageset_set_high_and_batch(zone, pcp); |
5574 | } |
5575 | |
5576 | static void __meminit setup_zone_pageset(struct zone *zone) |
5577 | { |
5578 | int cpu; |
5579 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
5580 | for_each_possible_cpu(cpu) |
5581 | zone_pageset_init(zone, cpu); |
5582 | } |
5583 | |
5584 | /* |
5585 | * Allocate per cpu pagesets and initialize them. |
5586 | * Before this call only boot pagesets were available. |
5587 | */ |
5588 | void __init setup_per_cpu_pageset(void) |
5589 | { |
5590 | struct pglist_data *pgdat; |
5591 | struct zone *zone; |
5592 | |
5593 | for_each_populated_zone(zone) |
5594 | setup_zone_pageset(zone); |
5595 | |
5596 | for_each_online_pgdat(pgdat) |
5597 | pgdat->per_cpu_nodestats = |
5598 | alloc_percpu(struct per_cpu_nodestat); |
5599 | } |
5600 | |
5601 | static __meminit void zone_pcp_init(struct zone *zone) |
5602 | { |
5603 | /* |
5604 | * per cpu subsystem is not up at this point. The following code |
5605 | * relies on the ability of the linker to provide the |
5606 | * offset of a (static) per cpu variable into the per cpu area. |
5607 | */ |
5608 | zone->pageset = &boot_pageset; |
5609 | |
5610 | if (populated_zone(zone)) |
5611 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
5612 | zone->name, zone->present_pages, |
5613 | zone_batchsize(zone)); |
5614 | } |
5615 | |
5616 | int __meminit init_currently_empty_zone(struct zone *zone, |
5617 | unsigned long zone_start_pfn, |
5618 | unsigned long size) |
5619 | { |
5620 | struct pglist_data *pgdat = zone->zone_pgdat; |
5621 | |
5622 | pgdat->nr_zones = zone_idx(zone) + 1; |
5623 | |
5624 | zone->zone_start_pfn = zone_start_pfn; |
5625 | |
5626 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
5627 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", |
5628 | pgdat->node_id, |
5629 | (unsigned long)zone_idx(zone), |
5630 | zone_start_pfn, (zone_start_pfn + size)); |
5631 | |
5632 | zone_init_free_lists(zone); |
5633 | zone->initialized = 1; |
5634 | |
5635 | return 0; |
5636 | } |
5637 | |
5638 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
5639 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
5640 | |
5641 | /* |
5642 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. |
5643 | */ |
5644 | int __meminit __early_pfn_to_nid(unsigned long pfn, |
5645 | struct mminit_pfnnid_cache *state) |
5646 | { |
5647 | unsigned long start_pfn, end_pfn; |
5648 | int nid; |
5649 | |
5650 | if (state->last_start <= pfn && pfn < state->last_end) |
5651 | return state->last_nid; |
5652 | |
5653 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn); |
5654 | if (nid != -1) { |
5655 | state->last_start = start_pfn; |
5656 | state->last_end = end_pfn; |
5657 | state->last_nid = nid; |
5658 | } |
5659 | |
5660 | return nid; |
5661 | } |
5662 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ |
5663 | |
5664 | /** |
5665 | * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range |
5666 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
5667 | * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid |
5668 | * |
5669 | * If an architecture guarantees that all ranges registered contain no holes |
5670 | * and may be freed, this this function may be used instead of calling |
5671 | * memblock_free_early_nid() manually. |
5672 | */ |
5673 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
5674 | { |
5675 | unsigned long start_pfn, end_pfn; |
5676 | int i, this_nid; |
5677 | |
5678 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
5679 | start_pfn = min(start_pfn, max_low_pfn); |
5680 | end_pfn = min(end_pfn, max_low_pfn); |
5681 | |
5682 | if (start_pfn < end_pfn) |
5683 | memblock_free_early_nid(PFN_PHYS(start_pfn), |
5684 | (end_pfn - start_pfn) << PAGE_SHIFT, |
5685 | this_nid); |
5686 | } |
5687 | } |
5688 | |
5689 | /** |
5690 | * sparse_memory_present_with_active_regions - Call memory_present for each active range |
5691 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
5692 | * |
5693 | * If an architecture guarantees that all ranges registered contain no holes and may |
5694 | * be freed, this function may be used instead of calling memory_present() manually. |
5695 | */ |
5696 | void __init sparse_memory_present_with_active_regions(int nid) |
5697 | { |
5698 | unsigned long start_pfn, end_pfn; |
5699 | int i, this_nid; |
5700 | |
5701 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
5702 | memory_present(this_nid, start_pfn, end_pfn); |
5703 | } |
5704 | |
5705 | /** |
5706 | * get_pfn_range_for_nid - Return the start and end page frames for a node |
5707 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
5708 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. |
5709 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. |
5710 | * |
5711 | * It returns the start and end page frame of a node based on information |
5712 | * provided by memblock_set_node(). If called for a node |
5713 | * with no available memory, a warning is printed and the start and end |
5714 | * PFNs will be 0. |
5715 | */ |
5716 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
5717 | unsigned long *start_pfn, unsigned long *end_pfn) |
5718 | { |
5719 | unsigned long this_start_pfn, this_end_pfn; |
5720 | int i; |
5721 | |
5722 | *start_pfn = -1UL; |
5723 | *end_pfn = 0; |
5724 | |
5725 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
5726 | *start_pfn = min(*start_pfn, this_start_pfn); |
5727 | *end_pfn = max(*end_pfn, this_end_pfn); |
5728 | } |
5729 | |
5730 | if (*start_pfn == -1UL) |
5731 | *start_pfn = 0; |
5732 | } |
5733 | |
5734 | /* |
5735 | * This finds a zone that can be used for ZONE_MOVABLE pages. The |
5736 | * assumption is made that zones within a node are ordered in monotonic |
5737 | * increasing memory addresses so that the "highest" populated zone is used |
5738 | */ |
5739 | static void __init find_usable_zone_for_movable(void) |
5740 | { |
5741 | int zone_index; |
5742 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { |
5743 | if (zone_index == ZONE_MOVABLE) |
5744 | continue; |
5745 | |
5746 | if (arch_zone_highest_possible_pfn[zone_index] > |
5747 | arch_zone_lowest_possible_pfn[zone_index]) |
5748 | break; |
5749 | } |
5750 | |
5751 | VM_BUG_ON(zone_index == -1); |
5752 | movable_zone = zone_index; |
5753 | } |
5754 | |
5755 | /* |
5756 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE |
5757 | * because it is sized independent of architecture. Unlike the other zones, |
5758 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
5759 | * in each node depending on the size of each node and how evenly kernelcore |
5760 | * is distributed. This helper function adjusts the zone ranges |
5761 | * provided by the architecture for a given node by using the end of the |
5762 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that |
5763 | * zones within a node are in order of monotonic increases memory addresses |
5764 | */ |
5765 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
5766 | unsigned long zone_type, |
5767 | unsigned long node_start_pfn, |
5768 | unsigned long node_end_pfn, |
5769 | unsigned long *zone_start_pfn, |
5770 | unsigned long *zone_end_pfn) |
5771 | { |
5772 | /* Only adjust if ZONE_MOVABLE is on this node */ |
5773 | if (zone_movable_pfn[nid]) { |
5774 | /* Size ZONE_MOVABLE */ |
5775 | if (zone_type == ZONE_MOVABLE) { |
5776 | *zone_start_pfn = zone_movable_pfn[nid]; |
5777 | *zone_end_pfn = min(node_end_pfn, |
5778 | arch_zone_highest_possible_pfn[movable_zone]); |
5779 | |
5780 | /* Adjust for ZONE_MOVABLE starting within this range */ |
5781 | } else if (!mirrored_kernelcore && |
5782 | *zone_start_pfn < zone_movable_pfn[nid] && |
5783 | *zone_end_pfn > zone_movable_pfn[nid]) { |
5784 | *zone_end_pfn = zone_movable_pfn[nid]; |
5785 | |
5786 | /* Check if this whole range is within ZONE_MOVABLE */ |
5787 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) |
5788 | *zone_start_pfn = *zone_end_pfn; |
5789 | } |
5790 | } |
5791 | |
5792 | /* |
5793 | * Return the number of pages a zone spans in a node, including holes |
5794 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() |
5795 | */ |
5796 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
5797 | unsigned long zone_type, |
5798 | unsigned long node_start_pfn, |
5799 | unsigned long node_end_pfn, |
5800 | unsigned long *zone_start_pfn, |
5801 | unsigned long *zone_end_pfn, |
5802 | unsigned long *ignored) |
5803 | { |
5804 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
5805 | if (!node_start_pfn && !node_end_pfn) |
5806 | return 0; |
5807 | |
5808 | /* Get the start and end of the zone */ |
5809 | *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
5810 | *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; |
5811 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5812 | node_start_pfn, node_end_pfn, |
5813 | zone_start_pfn, zone_end_pfn); |
5814 | |
5815 | /* Check that this node has pages within the zone's required range */ |
5816 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn) |
5817 | return 0; |
5818 | |
5819 | /* Move the zone boundaries inside the node if necessary */ |
5820 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn); |
5821 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn); |
5822 | |
5823 | /* Return the spanned pages */ |
5824 | return *zone_end_pfn - *zone_start_pfn; |
5825 | } |
5826 | |
5827 | /* |
5828 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, |
5829 | * then all holes in the requested range will be accounted for. |
5830 | */ |
5831 | unsigned long __meminit __absent_pages_in_range(int nid, |
5832 | unsigned long range_start_pfn, |
5833 | unsigned long range_end_pfn) |
5834 | { |
5835 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
5836 | unsigned long start_pfn, end_pfn; |
5837 | int i; |
5838 | |
5839 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
5840 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); |
5841 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); |
5842 | nr_absent -= end_pfn - start_pfn; |
5843 | } |
5844 | return nr_absent; |
5845 | } |
5846 | |
5847 | /** |
5848 | * absent_pages_in_range - Return number of page frames in holes within a range |
5849 | * @start_pfn: The start PFN to start searching for holes |
5850 | * @end_pfn: The end PFN to stop searching for holes |
5851 | * |
5852 | * It returns the number of pages frames in memory holes within a range. |
5853 | */ |
5854 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, |
5855 | unsigned long end_pfn) |
5856 | { |
5857 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); |
5858 | } |
5859 | |
5860 | /* Return the number of page frames in holes in a zone on a node */ |
5861 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
5862 | unsigned long zone_type, |
5863 | unsigned long node_start_pfn, |
5864 | unsigned long node_end_pfn, |
5865 | unsigned long *ignored) |
5866 | { |
5867 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
5868 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; |
5869 | unsigned long zone_start_pfn, zone_end_pfn; |
5870 | unsigned long nr_absent; |
5871 | |
5872 | /* When hotadd a new node from cpu_up(), the node should be empty */ |
5873 | if (!node_start_pfn && !node_end_pfn) |
5874 | return 0; |
5875 | |
5876 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
5877 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); |
5878 | |
5879 | adjust_zone_range_for_zone_movable(nid, zone_type, |
5880 | node_start_pfn, node_end_pfn, |
5881 | &zone_start_pfn, &zone_end_pfn); |
5882 | nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
5883 | |
5884 | /* |
5885 | * ZONE_MOVABLE handling. |
5886 | * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages |
5887 | * and vice versa. |
5888 | */ |
5889 | if (mirrored_kernelcore && zone_movable_pfn[nid]) { |
5890 | unsigned long start_pfn, end_pfn; |
5891 | struct memblock_region *r; |
5892 | |
5893 | for_each_memblock(memory, r) { |
5894 | start_pfn = clamp(memblock_region_memory_base_pfn(r), |
5895 | zone_start_pfn, zone_end_pfn); |
5896 | end_pfn = clamp(memblock_region_memory_end_pfn(r), |
5897 | zone_start_pfn, zone_end_pfn); |
5898 | |
5899 | if (zone_type == ZONE_MOVABLE && |
5900 | memblock_is_mirror(r)) |
5901 | nr_absent += end_pfn - start_pfn; |
5902 | |
5903 | if (zone_type == ZONE_NORMAL && |
5904 | !memblock_is_mirror(r)) |
5905 | nr_absent += end_pfn - start_pfn; |
5906 | } |
5907 | } |
5908 | |
5909 | return nr_absent; |
5910 | } |
5911 | |
5912 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
5913 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
5914 | unsigned long zone_type, |
5915 | unsigned long node_start_pfn, |
5916 | unsigned long node_end_pfn, |
5917 | unsigned long *zone_start_pfn, |
5918 | unsigned long *zone_end_pfn, |
5919 | unsigned long *zones_size) |
5920 | { |
5921 | unsigned int zone; |
5922 | |
5923 | *zone_start_pfn = node_start_pfn; |
5924 | for (zone = 0; zone < zone_type; zone++) |
5925 | *zone_start_pfn += zones_size[zone]; |
5926 | |
5927 | *zone_end_pfn = *zone_start_pfn + zones_size[zone_type]; |
5928 | |
5929 | return zones_size[zone_type]; |
5930 | } |
5931 | |
5932 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
5933 | unsigned long zone_type, |
5934 | unsigned long node_start_pfn, |
5935 | unsigned long node_end_pfn, |
5936 | unsigned long *zholes_size) |
5937 | { |
5938 | if (!zholes_size) |
5939 | return 0; |
5940 | |
5941 | return zholes_size[zone_type]; |
5942 | } |
5943 | |
5944 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
5945 | |
5946 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
5947 | unsigned long node_start_pfn, |
5948 | unsigned long node_end_pfn, |
5949 | unsigned long *zones_size, |
5950 | unsigned long *zholes_size) |
5951 | { |
5952 | unsigned long realtotalpages = 0, totalpages = 0; |
5953 | enum zone_type i; |
5954 | |
5955 | for (i = 0; i < MAX_NR_ZONES; i++) { |
5956 | struct zone *zone = pgdat->node_zones + i; |
5957 | unsigned long zone_start_pfn, zone_end_pfn; |
5958 | unsigned long size, real_size; |
5959 | |
5960 | size = zone_spanned_pages_in_node(pgdat->node_id, i, |
5961 | node_start_pfn, |
5962 | node_end_pfn, |
5963 | &zone_start_pfn, |
5964 | &zone_end_pfn, |
5965 | zones_size); |
5966 | real_size = size - zone_absent_pages_in_node(pgdat->node_id, i, |
5967 | node_start_pfn, node_end_pfn, |
5968 | zholes_size); |
5969 | if (size) |
5970 | zone->zone_start_pfn = zone_start_pfn; |
5971 | else |
5972 | zone->zone_start_pfn = 0; |
5973 | zone->spanned_pages = size; |
5974 | zone->present_pages = real_size; |
5975 | |
5976 | totalpages += size; |
5977 | realtotalpages += real_size; |
5978 | } |
5979 | |
5980 | pgdat->node_spanned_pages = totalpages; |
5981 | pgdat->node_present_pages = realtotalpages; |
5982 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, |
5983 | realtotalpages); |
5984 | } |
5985 | |
5986 | #ifndef CONFIG_SPARSEMEM |
5987 | /* |
5988 | * Calculate the size of the zone->blockflags rounded to an unsigned long |
5989 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
5990 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally |
5991 | * round what is now in bits to nearest long in bits, then return it in |
5992 | * bytes. |
5993 | */ |
5994 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize) |
5995 | { |
5996 | unsigned long usemapsize; |
5997 | |
5998 | zonesize += zone_start_pfn & (pageblock_nr_pages-1); |
5999 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
6000 | usemapsize = usemapsize >> pageblock_order; |
6001 | usemapsize *= NR_PAGEBLOCK_BITS; |
6002 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); |
6003 | |
6004 | return usemapsize / 8; |
6005 | } |
6006 | |
6007 | static void __init setup_usemap(struct pglist_data *pgdat, |
6008 | struct zone *zone, |
6009 | unsigned long zone_start_pfn, |
6010 | unsigned long zonesize) |
6011 | { |
6012 | unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize); |
6013 | zone->pageblock_flags = NULL; |
6014 | if (usemapsize) |
6015 | zone->pageblock_flags = |
6016 | memblock_virt_alloc_node_nopanic(usemapsize, |
6017 | pgdat->node_id); |
6018 | } |
6019 | #else |
6020 | static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone, |
6021 | unsigned long zone_start_pfn, unsigned long zonesize) {} |
6022 | #endif /* CONFIG_SPARSEMEM */ |
6023 | |
6024 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
6025 | |
6026 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
6027 | void __paginginit set_pageblock_order(void) |
6028 | { |
6029 | unsigned int order; |
6030 | |
6031 | /* Check that pageblock_nr_pages has not already been setup */ |
6032 | if (pageblock_order) |
6033 | return; |
6034 | |
6035 | if (HPAGE_SHIFT > PAGE_SHIFT) |
6036 | order = HUGETLB_PAGE_ORDER; |
6037 | else |
6038 | order = MAX_ORDER - 1; |
6039 | |
6040 | /* |
6041 | * Assume the largest contiguous order of interest is a huge page. |
6042 | * This value may be variable depending on boot parameters on IA64 and |
6043 | * powerpc. |
6044 | */ |
6045 | pageblock_order = order; |
6046 | } |
6047 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
6048 | |
6049 | /* |
6050 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() |
6051 | * is unused as pageblock_order is set at compile-time. See |
6052 | * include/linux/pageblock-flags.h for the values of pageblock_order based on |
6053 | * the kernel config |
6054 | */ |
6055 | void __paginginit set_pageblock_order(void) |
6056 | { |
6057 | } |
6058 | |
6059 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
6060 | |
6061 | static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages, |
6062 | unsigned long present_pages) |
6063 | { |
6064 | unsigned long pages = spanned_pages; |
6065 | |
6066 | /* |
6067 | * Provide a more accurate estimation if there are holes within |
6068 | * the zone and SPARSEMEM is in use. If there are holes within the |
6069 | * zone, each populated memory region may cost us one or two extra |
6070 | * memmap pages due to alignment because memmap pages for each |
6071 | * populated regions may not naturally algined on page boundary. |
6072 | * So the (present_pages >> 4) heuristic is a tradeoff for that. |
6073 | */ |
6074 | if (spanned_pages > present_pages + (present_pages >> 4) && |
6075 | IS_ENABLED(CONFIG_SPARSEMEM)) |
6076 | pages = present_pages; |
6077 | |
6078 | return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT; |
6079 | } |
6080 | |
6081 | /* |
6082 | * Set up the zone data structures: |
6083 | * - mark all pages reserved |
6084 | * - mark all memory queues empty |
6085 | * - clear the memory bitmaps |
6086 | * |
6087 | * NOTE: pgdat should get zeroed by caller. |
6088 | */ |
6089 | static void __paginginit free_area_init_core(struct pglist_data *pgdat) |
6090 | { |
6091 | enum zone_type j; |
6092 | int nid = pgdat->node_id; |
6093 | int ret; |
6094 | |
6095 | pgdat_resize_init(pgdat); |
6096 | #ifdef CONFIG_NUMA_BALANCING |
6097 | spin_lock_init(&pgdat->numabalancing_migrate_lock); |
6098 | pgdat->numabalancing_migrate_nr_pages = 0; |
6099 | pgdat->numabalancing_migrate_next_window = jiffies; |
6100 | #endif |
6101 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6102 | spin_lock_init(&pgdat->split_queue_lock); |
6103 | INIT_LIST_HEAD(&pgdat->split_queue); |
6104 | pgdat->split_queue_len = 0; |
6105 | #endif |
6106 | init_waitqueue_head(&pgdat->kswapd_wait); |
6107 | init_waitqueue_head(&pgdat->pfmemalloc_wait); |
6108 | #ifdef CONFIG_COMPACTION |
6109 | init_waitqueue_head(&pgdat->kcompactd_wait); |
6110 | #endif |
6111 | pgdat_page_ext_init(pgdat); |
6112 | spin_lock_init(&pgdat->lru_lock); |
6113 | lruvec_init(node_lruvec(pgdat)); |
6114 | |
6115 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6116 | struct zone *zone = pgdat->node_zones + j; |
6117 | unsigned long size, realsize, freesize, memmap_pages; |
6118 | unsigned long zone_start_pfn = zone->zone_start_pfn; |
6119 | |
6120 | size = zone->spanned_pages; |
6121 | realsize = freesize = zone->present_pages; |
6122 | |
6123 | /* |
6124 | * Adjust freesize so that it accounts for how much memory |
6125 | * is used by this zone for memmap. This affects the watermark |
6126 | * and per-cpu initialisations |
6127 | */ |
6128 | memmap_pages = calc_memmap_size(size, realsize); |
6129 | if (!is_highmem_idx(j)) { |
6130 | if (freesize >= memmap_pages) { |
6131 | freesize -= memmap_pages; |
6132 | if (memmap_pages) |
6133 | printk(KERN_DEBUG |
6134 | " %s zone: %lu pages used for memmap\n", |
6135 | zone_names[j], memmap_pages); |
6136 | } else |
6137 | pr_warn(" %s zone: %lu pages exceeds freesize %lu\n", |
6138 | zone_names[j], memmap_pages, freesize); |
6139 | } |
6140 | |
6141 | /* Account for reserved pages */ |
6142 | if (j == 0 && freesize > dma_reserve) { |
6143 | freesize -= dma_reserve; |
6144 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
6145 | zone_names[0], dma_reserve); |
6146 | } |
6147 | |
6148 | if (!is_highmem_idx(j)) |
6149 | nr_kernel_pages += freesize; |
6150 | /* Charge for highmem memmap if there are enough kernel pages */ |
6151 | else if (nr_kernel_pages > memmap_pages * 2) |
6152 | nr_kernel_pages -= memmap_pages; |
6153 | nr_all_pages += freesize; |
6154 | |
6155 | /* |
6156 | * Set an approximate value for lowmem here, it will be adjusted |
6157 | * when the bootmem allocator frees pages into the buddy system. |
6158 | * And all highmem pages will be managed by the buddy system. |
6159 | */ |
6160 | zone->managed_pages = is_highmem_idx(j) ? realsize : freesize; |
6161 | #ifdef CONFIG_NUMA |
6162 | zone->node = nid; |
6163 | #endif |
6164 | zone->name = zone_names[j]; |
6165 | zone->zone_pgdat = pgdat; |
6166 | spin_lock_init(&zone->lock); |
6167 | zone_seqlock_init(zone); |
6168 | zone_pcp_init(zone); |
6169 | |
6170 | if (!size) |
6171 | continue; |
6172 | |
6173 | set_pageblock_order(); |
6174 | setup_usemap(pgdat, zone, zone_start_pfn, size); |
6175 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); |
6176 | BUG_ON(ret); |
6177 | memmap_init(size, nid, j, zone_start_pfn); |
6178 | } |
6179 | } |
6180 | |
6181 | static void __ref alloc_node_mem_map(struct pglist_data *pgdat) |
6182 | { |
6183 | unsigned long __maybe_unused start = 0; |
6184 | unsigned long __maybe_unused offset = 0; |
6185 | |
6186 | /* Skip empty nodes */ |
6187 | if (!pgdat->node_spanned_pages) |
6188 | return; |
6189 | |
6190 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
6191 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
6192 | offset = pgdat->node_start_pfn - start; |
6193 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
6194 | if (!pgdat->node_mem_map) { |
6195 | unsigned long size, end; |
6196 | struct page *map; |
6197 | |
6198 | /* |
6199 | * The zone's endpoints aren't required to be MAX_ORDER |
6200 | * aligned but the node_mem_map endpoints must be in order |
6201 | * for the buddy allocator to function correctly. |
6202 | */ |
6203 | end = pgdat_end_pfn(pgdat); |
6204 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
6205 | size = (end - start) * sizeof(struct page); |
6206 | map = alloc_remap(pgdat->node_id, size); |
6207 | if (!map) |
6208 | map = memblock_virt_alloc_node_nopanic(size, |
6209 | pgdat->node_id); |
6210 | pgdat->node_mem_map = map + offset; |
6211 | } |
6212 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
6213 | /* |
6214 | * With no DISCONTIG, the global mem_map is just set as node 0's |
6215 | */ |
6216 | if (pgdat == NODE_DATA(0)) { |
6217 | mem_map = NODE_DATA(0)->node_mem_map; |
6218 | #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM) |
6219 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
6220 | mem_map -= offset; |
6221 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6222 | } |
6223 | #endif |
6224 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
6225 | } |
6226 | |
6227 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
6228 | unsigned long node_start_pfn, unsigned long *zholes_size) |
6229 | { |
6230 | pg_data_t *pgdat = NODE_DATA(nid); |
6231 | unsigned long start_pfn = 0; |
6232 | unsigned long end_pfn = 0; |
6233 | |
6234 | /* pg_data_t should be reset to zero when it's allocated */ |
6235 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx); |
6236 | |
6237 | pgdat->node_id = nid; |
6238 | pgdat->node_start_pfn = node_start_pfn; |
6239 | pgdat->per_cpu_nodestats = NULL; |
6240 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6241 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
6242 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid, |
6243 | (u64)start_pfn << PAGE_SHIFT, |
6244 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0); |
6245 | #else |
6246 | start_pfn = node_start_pfn; |
6247 | #endif |
6248 | calculate_node_totalpages(pgdat, start_pfn, end_pfn, |
6249 | zones_size, zholes_size); |
6250 | |
6251 | alloc_node_mem_map(pgdat); |
6252 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
6253 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", |
6254 | nid, (unsigned long)pgdat, |
6255 | (unsigned long)pgdat->node_mem_map); |
6256 | #endif |
6257 | |
6258 | reset_deferred_meminit(pgdat); |
6259 | free_area_init_core(pgdat); |
6260 | } |
6261 | |
6262 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
6263 | |
6264 | #if MAX_NUMNODES > 1 |
6265 | /* |
6266 | * Figure out the number of possible node ids. |
6267 | */ |
6268 | void __init setup_nr_node_ids(void) |
6269 | { |
6270 | unsigned int highest; |
6271 | |
6272 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES); |
6273 | nr_node_ids = highest + 1; |
6274 | } |
6275 | #endif |
6276 | |
6277 | /** |
6278 | * node_map_pfn_alignment - determine the maximum internode alignment |
6279 | * |
6280 | * This function should be called after node map is populated and sorted. |
6281 | * It calculates the maximum power of two alignment which can distinguish |
6282 | * all the nodes. |
6283 | * |
6284 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value |
6285 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the |
6286 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is |
6287 | * shifted, 1GiB is enough and this function will indicate so. |
6288 | * |
6289 | * This is used to test whether pfn -> nid mapping of the chosen memory |
6290 | * model has fine enough granularity to avoid incorrect mapping for the |
6291 | * populated node map. |
6292 | * |
6293 | * Returns the determined alignment in pfn's. 0 if there is no alignment |
6294 | * requirement (single node). |
6295 | */ |
6296 | unsigned long __init node_map_pfn_alignment(void) |
6297 | { |
6298 | unsigned long accl_mask = 0, last_end = 0; |
6299 | unsigned long start, end, mask; |
6300 | int last_nid = -1; |
6301 | int i, nid; |
6302 | |
6303 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
6304 | if (!start || last_nid < 0 || last_nid == nid) { |
6305 | last_nid = nid; |
6306 | last_end = end; |
6307 | continue; |
6308 | } |
6309 | |
6310 | /* |
6311 | * Start with a mask granular enough to pin-point to the |
6312 | * start pfn and tick off bits one-by-one until it becomes |
6313 | * too coarse to separate the current node from the last. |
6314 | */ |
6315 | mask = ~((1 << __ffs(start)) - 1); |
6316 | while (mask && last_end <= (start & (mask << 1))) |
6317 | mask <<= 1; |
6318 | |
6319 | /* accumulate all internode masks */ |
6320 | accl_mask |= mask; |
6321 | } |
6322 | |
6323 | /* convert mask to number of pages */ |
6324 | return ~accl_mask + 1; |
6325 | } |
6326 | |
6327 | /* Find the lowest pfn for a node */ |
6328 | static unsigned long __init find_min_pfn_for_node(int nid) |
6329 | { |
6330 | unsigned long min_pfn = ULONG_MAX; |
6331 | unsigned long start_pfn; |
6332 | int i; |
6333 | |
6334 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
6335 | min_pfn = min(min_pfn, start_pfn); |
6336 | |
6337 | if (min_pfn == ULONG_MAX) { |
6338 | pr_warn("Could not find start_pfn for node %d\n", nid); |
6339 | return 0; |
6340 | } |
6341 | |
6342 | return min_pfn; |
6343 | } |
6344 | |
6345 | /** |
6346 | * find_min_pfn_with_active_regions - Find the minimum PFN registered |
6347 | * |
6348 | * It returns the minimum PFN based on information provided via |
6349 | * memblock_set_node(). |
6350 | */ |
6351 | unsigned long __init find_min_pfn_with_active_regions(void) |
6352 | { |
6353 | return find_min_pfn_for_node(MAX_NUMNODES); |
6354 | } |
6355 | |
6356 | /* |
6357 | * early_calculate_totalpages() |
6358 | * Sum pages in active regions for movable zone. |
6359 | * Populate N_MEMORY for calculating usable_nodes. |
6360 | */ |
6361 | static unsigned long __init early_calculate_totalpages(void) |
6362 | { |
6363 | unsigned long totalpages = 0; |
6364 | unsigned long start_pfn, end_pfn; |
6365 | int i, nid; |
6366 | |
6367 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
6368 | unsigned long pages = end_pfn - start_pfn; |
6369 | |
6370 | totalpages += pages; |
6371 | if (pages) |
6372 | node_set_state(nid, N_MEMORY); |
6373 | } |
6374 | return totalpages; |
6375 | } |
6376 | |
6377 | /* |
6378 | * Find the PFN the Movable zone begins in each node. Kernel memory |
6379 | * is spread evenly between nodes as long as the nodes have enough |
6380 | * memory. When they don't, some nodes will have more kernelcore than |
6381 | * others |
6382 | */ |
6383 | static void __init find_zone_movable_pfns_for_nodes(void) |
6384 | { |
6385 | int i, nid; |
6386 | unsigned long usable_startpfn; |
6387 | unsigned long kernelcore_node, kernelcore_remaining; |
6388 | /* save the state before borrow the nodemask */ |
6389 | nodemask_t saved_node_state = node_states[N_MEMORY]; |
6390 | unsigned long totalpages = early_calculate_totalpages(); |
6391 | int usable_nodes = nodes_weight(node_states[N_MEMORY]); |
6392 | struct memblock_region *r; |
6393 | |
6394 | /* Need to find movable_zone earlier when movable_node is specified. */ |
6395 | find_usable_zone_for_movable(); |
6396 | |
6397 | /* |
6398 | * If movable_node is specified, ignore kernelcore and movablecore |
6399 | * options. |
6400 | */ |
6401 | if (movable_node_is_enabled()) { |
6402 | for_each_memblock(memory, r) { |
6403 | if (!memblock_is_hotpluggable(r)) |
6404 | continue; |
6405 | |
6406 | nid = r->nid; |
6407 | |
6408 | usable_startpfn = PFN_DOWN(r->base); |
6409 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6410 | min(usable_startpfn, zone_movable_pfn[nid]) : |
6411 | usable_startpfn; |
6412 | } |
6413 | |
6414 | goto out2; |
6415 | } |
6416 | |
6417 | /* |
6418 | * If kernelcore=mirror is specified, ignore movablecore option |
6419 | */ |
6420 | if (mirrored_kernelcore) { |
6421 | bool mem_below_4gb_not_mirrored = false; |
6422 | |
6423 | for_each_memblock(memory, r) { |
6424 | if (memblock_is_mirror(r)) |
6425 | continue; |
6426 | |
6427 | nid = r->nid; |
6428 | |
6429 | usable_startpfn = memblock_region_memory_base_pfn(r); |
6430 | |
6431 | if (usable_startpfn < 0x100000) { |
6432 | mem_below_4gb_not_mirrored = true; |
6433 | continue; |
6434 | } |
6435 | |
6436 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ? |
6437 | min(usable_startpfn, zone_movable_pfn[nid]) : |
6438 | usable_startpfn; |
6439 | } |
6440 | |
6441 | if (mem_below_4gb_not_mirrored) |
6442 | pr_warn("This configuration results in unmirrored kernel memory."); |
6443 | |
6444 | goto out2; |
6445 | } |
6446 | |
6447 | /* |
6448 | * If movablecore=nn[KMG] was specified, calculate what size of |
6449 | * kernelcore that corresponds so that memory usable for |
6450 | * any allocation type is evenly spread. If both kernelcore |
6451 | * and movablecore are specified, then the value of kernelcore |
6452 | * will be used for required_kernelcore if it's greater than |
6453 | * what movablecore would have allowed. |
6454 | */ |
6455 | if (required_movablecore) { |
6456 | unsigned long corepages; |
6457 | |
6458 | /* |
6459 | * Round-up so that ZONE_MOVABLE is at least as large as what |
6460 | * was requested by the user |
6461 | */ |
6462 | required_movablecore = |
6463 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); |
6464 | required_movablecore = min(totalpages, required_movablecore); |
6465 | corepages = totalpages - required_movablecore; |
6466 | |
6467 | required_kernelcore = max(required_kernelcore, corepages); |
6468 | } |
6469 | |
6470 | /* |
6471 | * If kernelcore was not specified or kernelcore size is larger |
6472 | * than totalpages, there is no ZONE_MOVABLE. |
6473 | */ |
6474 | if (!required_kernelcore || required_kernelcore >= totalpages) |
6475 | goto out; |
6476 | |
6477 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ |
6478 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
6479 | |
6480 | restart: |
6481 | /* Spread kernelcore memory as evenly as possible throughout nodes */ |
6482 | kernelcore_node = required_kernelcore / usable_nodes; |
6483 | for_each_node_state(nid, N_MEMORY) { |
6484 | unsigned long start_pfn, end_pfn; |
6485 | |
6486 | /* |
6487 | * Recalculate kernelcore_node if the division per node |
6488 | * now exceeds what is necessary to satisfy the requested |
6489 | * amount of memory for the kernel |
6490 | */ |
6491 | if (required_kernelcore < kernelcore_node) |
6492 | kernelcore_node = required_kernelcore / usable_nodes; |
6493 | |
6494 | /* |
6495 | * As the map is walked, we track how much memory is usable |
6496 | * by the kernel using kernelcore_remaining. When it is |
6497 | * 0, the rest of the node is usable by ZONE_MOVABLE |
6498 | */ |
6499 | kernelcore_remaining = kernelcore_node; |
6500 | |
6501 | /* Go through each range of PFNs within this node */ |
6502 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
6503 | unsigned long size_pages; |
6504 | |
6505 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
6506 | if (start_pfn >= end_pfn) |
6507 | continue; |
6508 | |
6509 | /* Account for what is only usable for kernelcore */ |
6510 | if (start_pfn < usable_startpfn) { |
6511 | unsigned long kernel_pages; |
6512 | kernel_pages = min(end_pfn, usable_startpfn) |
6513 | - start_pfn; |
6514 | |
6515 | kernelcore_remaining -= min(kernel_pages, |
6516 | kernelcore_remaining); |
6517 | required_kernelcore -= min(kernel_pages, |
6518 | required_kernelcore); |
6519 | |
6520 | /* Continue if range is now fully accounted */ |
6521 | if (end_pfn <= usable_startpfn) { |
6522 | |
6523 | /* |
6524 | * Push zone_movable_pfn to the end so |
6525 | * that if we have to rebalance |
6526 | * kernelcore across nodes, we will |
6527 | * not double account here |
6528 | */ |
6529 | zone_movable_pfn[nid] = end_pfn; |
6530 | continue; |
6531 | } |
6532 | start_pfn = usable_startpfn; |
6533 | } |
6534 | |
6535 | /* |
6536 | * The usable PFN range for ZONE_MOVABLE is from |
6537 | * start_pfn->end_pfn. Calculate size_pages as the |
6538 | * number of pages used as kernelcore |
6539 | */ |
6540 | size_pages = end_pfn - start_pfn; |
6541 | if (size_pages > kernelcore_remaining) |
6542 | size_pages = kernelcore_remaining; |
6543 | zone_movable_pfn[nid] = start_pfn + size_pages; |
6544 | |
6545 | /* |
6546 | * Some kernelcore has been met, update counts and |
6547 | * break if the kernelcore for this node has been |
6548 | * satisfied |
6549 | */ |
6550 | required_kernelcore -= min(required_kernelcore, |
6551 | size_pages); |
6552 | kernelcore_remaining -= size_pages; |
6553 | if (!kernelcore_remaining) |
6554 | break; |
6555 | } |
6556 | } |
6557 | |
6558 | /* |
6559 | * If there is still required_kernelcore, we do another pass with one |
6560 | * less node in the count. This will push zone_movable_pfn[nid] further |
6561 | * along on the nodes that still have memory until kernelcore is |
6562 | * satisfied |
6563 | */ |
6564 | usable_nodes--; |
6565 | if (usable_nodes && required_kernelcore > usable_nodes) |
6566 | goto restart; |
6567 | |
6568 | out2: |
6569 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
6570 | for (nid = 0; nid < MAX_NUMNODES; nid++) |
6571 | zone_movable_pfn[nid] = |
6572 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); |
6573 | |
6574 | out: |
6575 | /* restore the node_state */ |
6576 | node_states[N_MEMORY] = saved_node_state; |
6577 | } |
6578 | |
6579 | /* Any regular or high memory on that node ? */ |
6580 | static void check_for_memory(pg_data_t *pgdat, int nid) |
6581 | { |
6582 | enum zone_type zone_type; |
6583 | |
6584 | if (N_MEMORY == N_NORMAL_MEMORY) |
6585 | return; |
6586 | |
6587 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) { |
6588 | struct zone *zone = &pgdat->node_zones[zone_type]; |
6589 | if (populated_zone(zone)) { |
6590 | node_set_state(nid, N_HIGH_MEMORY); |
6591 | if (N_NORMAL_MEMORY != N_HIGH_MEMORY && |
6592 | zone_type <= ZONE_NORMAL) |
6593 | node_set_state(nid, N_NORMAL_MEMORY); |
6594 | break; |
6595 | } |
6596 | } |
6597 | } |
6598 | |
6599 | /** |
6600 | * free_area_init_nodes - Initialise all pg_data_t and zone data |
6601 | * @max_zone_pfn: an array of max PFNs for each zone |
6602 | * |
6603 | * This will call free_area_init_node() for each active node in the system. |
6604 | * Using the page ranges provided by memblock_set_node(), the size of each |
6605 | * zone in each node and their holes is calculated. If the maximum PFN |
6606 | * between two adjacent zones match, it is assumed that the zone is empty. |
6607 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed |
6608 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone |
6609 | * starts where the previous one ended. For example, ZONE_DMA32 starts |
6610 | * at arch_max_dma_pfn. |
6611 | */ |
6612 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) |
6613 | { |
6614 | unsigned long start_pfn, end_pfn; |
6615 | int i, nid; |
6616 | |
6617 | /* Record where the zone boundaries are */ |
6618 | memset(arch_zone_lowest_possible_pfn, 0, |
6619 | sizeof(arch_zone_lowest_possible_pfn)); |
6620 | memset(arch_zone_highest_possible_pfn, 0, |
6621 | sizeof(arch_zone_highest_possible_pfn)); |
6622 | |
6623 | start_pfn = find_min_pfn_with_active_regions(); |
6624 | |
6625 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6626 | if (i == ZONE_MOVABLE) |
6627 | continue; |
6628 | |
6629 | end_pfn = max(max_zone_pfn[i], start_pfn); |
6630 | arch_zone_lowest_possible_pfn[i] = start_pfn; |
6631 | arch_zone_highest_possible_pfn[i] = end_pfn; |
6632 | |
6633 | start_pfn = end_pfn; |
6634 | } |
6635 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
6636 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; |
6637 | |
6638 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ |
6639 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); |
6640 | find_zone_movable_pfns_for_nodes(); |
6641 | |
6642 | /* Print out the zone ranges */ |
6643 | pr_info("Zone ranges:\n"); |
6644 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6645 | if (i == ZONE_MOVABLE) |
6646 | continue; |
6647 | pr_info(" %-8s ", zone_names[i]); |
6648 | if (arch_zone_lowest_possible_pfn[i] == |
6649 | arch_zone_highest_possible_pfn[i]) |
6650 | pr_cont("empty\n"); |
6651 | else |
6652 | pr_cont("[mem %#018Lx-%#018Lx]\n", |
6653 | (u64)arch_zone_lowest_possible_pfn[i] |
6654 | << PAGE_SHIFT, |
6655 | ((u64)arch_zone_highest_possible_pfn[i] |
6656 | << PAGE_SHIFT) - 1); |
6657 | } |
6658 | |
6659 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ |
6660 | pr_info("Movable zone start for each node\n"); |
6661 | for (i = 0; i < MAX_NUMNODES; i++) { |
6662 | if (zone_movable_pfn[i]) |
6663 | pr_info(" Node %d: %#018Lx\n", i, |
6664 | (u64)zone_movable_pfn[i] << PAGE_SHIFT); |
6665 | } |
6666 | |
6667 | /* Print out the early node map */ |
6668 | pr_info("Early memory node ranges\n"); |
6669 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
6670 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid, |
6671 | (u64)start_pfn << PAGE_SHIFT, |
6672 | ((u64)end_pfn << PAGE_SHIFT) - 1); |
6673 | |
6674 | /* Initialise every node */ |
6675 | mminit_verify_pageflags_layout(); |
6676 | setup_nr_node_ids(); |
6677 | for_each_online_node(nid) { |
6678 | pg_data_t *pgdat = NODE_DATA(nid); |
6679 | free_area_init_node(nid, NULL, |
6680 | find_min_pfn_for_node(nid), NULL); |
6681 | |
6682 | /* Any memory on that node */ |
6683 | if (pgdat->node_present_pages) |
6684 | node_set_state(nid, N_MEMORY); |
6685 | check_for_memory(pgdat, nid); |
6686 | } |
6687 | } |
6688 | |
6689 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
6690 | { |
6691 | unsigned long long coremem; |
6692 | if (!p) |
6693 | return -EINVAL; |
6694 | |
6695 | coremem = memparse(p, &p); |
6696 | *core = coremem >> PAGE_SHIFT; |
6697 | |
6698 | /* Paranoid check that UL is enough for the coremem value */ |
6699 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
6700 | |
6701 | return 0; |
6702 | } |
6703 | |
6704 | /* |
6705 | * kernelcore=size sets the amount of memory for use for allocations that |
6706 | * cannot be reclaimed or migrated. |
6707 | */ |
6708 | static int __init cmdline_parse_kernelcore(char *p) |
6709 | { |
6710 | /* parse kernelcore=mirror */ |
6711 | if (parse_option_str(p, "mirror")) { |
6712 | mirrored_kernelcore = true; |
6713 | return 0; |
6714 | } |
6715 | |
6716 | return cmdline_parse_core(p, &required_kernelcore); |
6717 | } |
6718 | |
6719 | /* |
6720 | * movablecore=size sets the amount of memory for use for allocations that |
6721 | * can be reclaimed or migrated. |
6722 | */ |
6723 | static int __init cmdline_parse_movablecore(char *p) |
6724 | { |
6725 | return cmdline_parse_core(p, &required_movablecore); |
6726 | } |
6727 | |
6728 | early_param("kernelcore", cmdline_parse_kernelcore); |
6729 | early_param("movablecore", cmdline_parse_movablecore); |
6730 | |
6731 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
6732 | |
6733 | void adjust_managed_page_count(struct page *page, long count) |
6734 | { |
6735 | spin_lock(&managed_page_count_lock); |
6736 | page_zone(page)->managed_pages += count; |
6737 | totalram_pages += count; |
6738 | #ifdef CONFIG_HIGHMEM |
6739 | if (PageHighMem(page)) |
6740 | totalhigh_pages += count; |
6741 | #endif |
6742 | spin_unlock(&managed_page_count_lock); |
6743 | } |
6744 | EXPORT_SYMBOL(adjust_managed_page_count); |
6745 | |
6746 | unsigned long free_reserved_area(void *start, void *end, int poison, char *s) |
6747 | { |
6748 | void *pos; |
6749 | unsigned long pages = 0; |
6750 | |
6751 | start = (void *)PAGE_ALIGN((unsigned long)start); |
6752 | end = (void *)((unsigned long)end & PAGE_MASK); |
6753 | for (pos = start; pos < end; pos += PAGE_SIZE, pages++) { |
6754 | if ((unsigned int)poison <= 0xFF) |
6755 | memset(pos, poison, PAGE_SIZE); |
6756 | free_reserved_page(virt_to_page(pos)); |
6757 | } |
6758 | |
6759 | if (pages && s) |
6760 | pr_info("Freeing %s memory: %ldK\n", |
6761 | s, pages << (PAGE_SHIFT - 10)); |
6762 | |
6763 | return pages; |
6764 | } |
6765 | EXPORT_SYMBOL(free_reserved_area); |
6766 | |
6767 | #ifdef CONFIG_HIGHMEM |
6768 | void free_highmem_page(struct page *page) |
6769 | { |
6770 | __free_reserved_page(page); |
6771 | totalram_pages++; |
6772 | page_zone(page)->managed_pages++; |
6773 | totalhigh_pages++; |
6774 | } |
6775 | #endif |
6776 | |
6777 | |
6778 | void __init mem_init_print_info(const char *str) |
6779 | { |
6780 | unsigned long physpages, codesize, datasize, rosize, bss_size; |
6781 | unsigned long init_code_size, init_data_size; |
6782 | |
6783 | physpages = get_num_physpages(); |
6784 | codesize = _etext - _stext; |
6785 | datasize = _edata - _sdata; |
6786 | rosize = __end_rodata - __start_rodata; |
6787 | bss_size = __bss_stop - __bss_start; |
6788 | init_data_size = __init_end - __init_begin; |
6789 | init_code_size = _einittext - _sinittext; |
6790 | |
6791 | /* |
6792 | * Detect special cases and adjust section sizes accordingly: |
6793 | * 1) .init.* may be embedded into .data sections |
6794 | * 2) .init.text.* may be out of [__init_begin, __init_end], |
6795 | * please refer to arch/tile/kernel/vmlinux.lds.S. |
6796 | * 3) .rodata.* may be embedded into .text or .data sections. |
6797 | */ |
6798 | #define adj_init_size(start, end, size, pos, adj) \ |
6799 | do { \ |
6800 | if (start <= pos && pos < end && size > adj) \ |
6801 | size -= adj; \ |
6802 | } while (0) |
6803 | |
6804 | adj_init_size(__init_begin, __init_end, init_data_size, |
6805 | _sinittext, init_code_size); |
6806 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size); |
6807 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size); |
6808 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize); |
6809 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize); |
6810 | |
6811 | #undef adj_init_size |
6812 | |
6813 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved" |
6814 | #ifdef CONFIG_HIGHMEM |
6815 | ", %luK highmem" |
6816 | #endif |
6817 | "%s%s)\n", |
6818 | nr_free_pages() << (PAGE_SHIFT - 10), |
6819 | physpages << (PAGE_SHIFT - 10), |
6820 | codesize >> 10, datasize >> 10, rosize >> 10, |
6821 | (init_data_size + init_code_size) >> 10, bss_size >> 10, |
6822 | (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), |
6823 | totalcma_pages << (PAGE_SHIFT - 10), |
6824 | #ifdef CONFIG_HIGHMEM |
6825 | totalhigh_pages << (PAGE_SHIFT - 10), |
6826 | #endif |
6827 | str ? ", " : "", str ? str : ""); |
6828 | } |
6829 | |
6830 | /** |
6831 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
6832 | * @new_dma_reserve: The number of pages to mark reserved |
6833 | * |
6834 | * The per-cpu batchsize and zone watermarks are determined by managed_pages. |
6835 | * In the DMA zone, a significant percentage may be consumed by kernel image |
6836 | * and other unfreeable allocations which can skew the watermarks badly. This |
6837 | * function may optionally be used to account for unfreeable pages in the |
6838 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and |
6839 | * smaller per-cpu batchsize. |
6840 | */ |
6841 | void __init set_dma_reserve(unsigned long new_dma_reserve) |
6842 | { |
6843 | dma_reserve = new_dma_reserve; |
6844 | } |
6845 | |
6846 | void __init free_area_init(unsigned long *zones_size) |
6847 | { |
6848 | free_area_init_node(0, zones_size, |
6849 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
6850 | } |
6851 | |
6852 | static int page_alloc_cpu_notify(struct notifier_block *self, |
6853 | unsigned long action, void *hcpu) |
6854 | { |
6855 | int cpu = (unsigned long)hcpu; |
6856 | |
6857 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
6858 | lru_add_drain_cpu(cpu); |
6859 | drain_pages(cpu); |
6860 | |
6861 | /* |
6862 | * Spill the event counters of the dead processor |
6863 | * into the current processors event counters. |
6864 | * This artificially elevates the count of the current |
6865 | * processor. |
6866 | */ |
6867 | vm_events_fold_cpu(cpu); |
6868 | |
6869 | /* |
6870 | * Zero the differential counters of the dead processor |
6871 | * so that the vm statistics are consistent. |
6872 | * |
6873 | * This is only okay since the processor is dead and cannot |
6874 | * race with what we are doing. |
6875 | */ |
6876 | cpu_vm_stats_fold(cpu); |
6877 | } |
6878 | return NOTIFY_OK; |
6879 | } |
6880 | |
6881 | void __init page_alloc_init(void) |
6882 | { |
6883 | hotcpu_notifier(page_alloc_cpu_notify, 0); |
6884 | } |
6885 | |
6886 | /* |
6887 | * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio |
6888 | * or min_free_kbytes changes. |
6889 | */ |
6890 | static void calculate_totalreserve_pages(void) |
6891 | { |
6892 | struct pglist_data *pgdat; |
6893 | unsigned long reserve_pages = 0; |
6894 | enum zone_type i, j; |
6895 | |
6896 | for_each_online_pgdat(pgdat) { |
6897 | |
6898 | pgdat->totalreserve_pages = 0; |
6899 | |
6900 | for (i = 0; i < MAX_NR_ZONES; i++) { |
6901 | struct zone *zone = pgdat->node_zones + i; |
6902 | long max = 0; |
6903 | |
6904 | /* Find valid and maximum lowmem_reserve in the zone */ |
6905 | for (j = i; j < MAX_NR_ZONES; j++) { |
6906 | if (zone->lowmem_reserve[j] > max) |
6907 | max = zone->lowmem_reserve[j]; |
6908 | } |
6909 | |
6910 | /* we treat the high watermark as reserved pages. */ |
6911 | max += high_wmark_pages(zone); |
6912 | |
6913 | if (max > zone->managed_pages) |
6914 | max = zone->managed_pages; |
6915 | |
6916 | pgdat->totalreserve_pages += max; |
6917 | |
6918 | reserve_pages += max; |
6919 | } |
6920 | } |
6921 | totalreserve_pages = reserve_pages; |
6922 | } |
6923 | |
6924 | /* |
6925 | * setup_per_zone_lowmem_reserve - called whenever |
6926 | * sysctl_lowmem_reserve_ratio changes. Ensures that each zone |
6927 | * has a correct pages reserved value, so an adequate number of |
6928 | * pages are left in the zone after a successful __alloc_pages(). |
6929 | */ |
6930 | static void setup_per_zone_lowmem_reserve(void) |
6931 | { |
6932 | struct pglist_data *pgdat; |
6933 | enum zone_type j, idx; |
6934 | |
6935 | for_each_online_pgdat(pgdat) { |
6936 | for (j = 0; j < MAX_NR_ZONES; j++) { |
6937 | struct zone *zone = pgdat->node_zones + j; |
6938 | unsigned long managed_pages = zone->managed_pages; |
6939 | |
6940 | zone->lowmem_reserve[j] = 0; |
6941 | |
6942 | idx = j; |
6943 | while (idx) { |
6944 | struct zone *lower_zone; |
6945 | |
6946 | idx--; |
6947 | |
6948 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
6949 | sysctl_lowmem_reserve_ratio[idx] = 1; |
6950 | |
6951 | lower_zone = pgdat->node_zones + idx; |
6952 | lower_zone->lowmem_reserve[j] = managed_pages / |
6953 | sysctl_lowmem_reserve_ratio[idx]; |
6954 | managed_pages += lower_zone->managed_pages; |
6955 | } |
6956 | } |
6957 | } |
6958 | |
6959 | /* update totalreserve_pages */ |
6960 | calculate_totalreserve_pages(); |
6961 | } |
6962 | |
6963 | static void __setup_per_zone_wmarks(void) |
6964 | { |
6965 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); |
6966 | unsigned long pages_low = extra_free_kbytes >> (PAGE_SHIFT - 10); |
6967 | unsigned long lowmem_pages = 0; |
6968 | struct zone *zone; |
6969 | unsigned long flags; |
6970 | |
6971 | /* Calculate total number of !ZONE_HIGHMEM pages */ |
6972 | for_each_zone(zone) { |
6973 | if (!is_highmem(zone)) |
6974 | lowmem_pages += zone->managed_pages; |
6975 | } |
6976 | |
6977 | for_each_zone(zone) { |
6978 | u64 min, low; |
6979 | |
6980 | spin_lock_irqsave(&zone->lock, flags); |
6981 | min = (u64)pages_min * zone->managed_pages; |
6982 | do_div(min, lowmem_pages); |
6983 | low = (u64)pages_low * zone->managed_pages; |
6984 | do_div(low, vm_total_pages); |
6985 | |
6986 | if (is_highmem(zone)) { |
6987 | /* |
6988 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
6989 | * need highmem pages, so cap pages_min to a small |
6990 | * value here. |
6991 | * |
6992 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
6993 | * deltas control asynch page reclaim, and so should |
6994 | * not be capped for highmem. |
6995 | */ |
6996 | unsigned long min_pages; |
6997 | |
6998 | min_pages = zone->managed_pages / 1024; |
6999 | min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL); |
7000 | zone->watermark[WMARK_MIN] = min_pages; |
7001 | } else { |
7002 | /* |
7003 | * If it's a lowmem zone, reserve a number of pages |
7004 | * proportionate to the zone's size. |
7005 | */ |
7006 | zone->watermark[WMARK_MIN] = min; |
7007 | } |
7008 | |
7009 | /* |
7010 | * Set the kswapd watermarks distance according to the |
7011 | * scale factor in proportion to available memory, but |
7012 | * ensure a minimum size on small systems. |
7013 | */ |
7014 | min = max_t(u64, min >> 2, |
7015 | mult_frac(zone->managed_pages, |
7016 | watermark_scale_factor, 10000)); |
7017 | |
7018 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + |
7019 | low + min; |
7020 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + |
7021 | low + min * 2; |
7022 | |
7023 | spin_unlock_irqrestore(&zone->lock, flags); |
7024 | } |
7025 | |
7026 | /* update totalreserve_pages */ |
7027 | calculate_totalreserve_pages(); |
7028 | } |
7029 | |
7030 | /** |
7031 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
7032 | * or when memory is hot-{added|removed} |
7033 | * |
7034 | * Ensures that the watermark[min,low,high] values for each zone are set |
7035 | * correctly with respect to min_free_kbytes. |
7036 | */ |
7037 | void setup_per_zone_wmarks(void) |
7038 | { |
7039 | mutex_lock(&zonelists_mutex); |
7040 | __setup_per_zone_wmarks(); |
7041 | mutex_unlock(&zonelists_mutex); |
7042 | } |
7043 | |
7044 | /* |
7045 | * Initialise min_free_kbytes. |
7046 | * |
7047 | * For small machines we want it small (128k min). For large machines |
7048 | * we want it large (64MB max). But it is not linear, because network |
7049 | * bandwidth does not increase linearly with machine size. We use |
7050 | * |
7051 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
7052 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
7053 | * |
7054 | * which yields |
7055 | * |
7056 | * 16MB: 512k |
7057 | * 32MB: 724k |
7058 | * 64MB: 1024k |
7059 | * 128MB: 1448k |
7060 | * 256MB: 2048k |
7061 | * 512MB: 2896k |
7062 | * 1024MB: 4096k |
7063 | * 2048MB: 5792k |
7064 | * 4096MB: 8192k |
7065 | * 8192MB: 11584k |
7066 | * 16384MB: 16384k |
7067 | */ |
7068 | int __meminit init_per_zone_wmark_min(void) |
7069 | { |
7070 | unsigned long lowmem_kbytes; |
7071 | int new_min_free_kbytes; |
7072 | |
7073 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); |
7074 | new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
7075 | |
7076 | if (new_min_free_kbytes > user_min_free_kbytes) { |
7077 | min_free_kbytes = new_min_free_kbytes; |
7078 | if (min_free_kbytes < 128) |
7079 | min_free_kbytes = 128; |
7080 | if (min_free_kbytes > 65536) |
7081 | min_free_kbytes = 65536; |
7082 | } else { |
7083 | pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", |
7084 | new_min_free_kbytes, user_min_free_kbytes); |
7085 | } |
7086 | setup_per_zone_wmarks(); |
7087 | refresh_zone_stat_thresholds(); |
7088 | setup_per_zone_lowmem_reserve(); |
7089 | |
7090 | #ifdef CONFIG_NUMA |
7091 | setup_min_unmapped_ratio(); |
7092 | setup_min_slab_ratio(); |
7093 | #endif |
7094 | |
7095 | return 0; |
7096 | } |
7097 | core_initcall(init_per_zone_wmark_min) |
7098 | |
7099 | /* |
7100 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
7101 | * that we can call two helper functions whenever min_free_kbytes |
7102 | * or extra_free_kbytes changes. |
7103 | */ |
7104 | int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write, |
7105 | void __user *buffer, size_t *length, loff_t *ppos) |
7106 | { |
7107 | int rc; |
7108 | |
7109 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7110 | if (rc) |
7111 | return rc; |
7112 | |
7113 | if (write) { |
7114 | user_min_free_kbytes = min_free_kbytes; |
7115 | setup_per_zone_wmarks(); |
7116 | } |
7117 | return 0; |
7118 | } |
7119 | |
7120 | int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write, |
7121 | void __user *buffer, size_t *length, loff_t *ppos) |
7122 | { |
7123 | int rc; |
7124 | |
7125 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7126 | if (rc) |
7127 | return rc; |
7128 | |
7129 | if (write) |
7130 | setup_per_zone_wmarks(); |
7131 | |
7132 | return 0; |
7133 | } |
7134 | |
7135 | #ifdef CONFIG_NUMA |
7136 | static void setup_min_unmapped_ratio(void) |
7137 | { |
7138 | pg_data_t *pgdat; |
7139 | struct zone *zone; |
7140 | |
7141 | for_each_online_pgdat(pgdat) |
7142 | pgdat->min_unmapped_pages = 0; |
7143 | |
7144 | for_each_zone(zone) |
7145 | zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages * |
7146 | sysctl_min_unmapped_ratio) / 100; |
7147 | } |
7148 | |
7149 | |
7150 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write, |
7151 | void __user *buffer, size_t *length, loff_t *ppos) |
7152 | { |
7153 | int rc; |
7154 | |
7155 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7156 | if (rc) |
7157 | return rc; |
7158 | |
7159 | setup_min_unmapped_ratio(); |
7160 | |
7161 | return 0; |
7162 | } |
7163 | |
7164 | static void setup_min_slab_ratio(void) |
7165 | { |
7166 | pg_data_t *pgdat; |
7167 | struct zone *zone; |
7168 | |
7169 | for_each_online_pgdat(pgdat) |
7170 | pgdat->min_slab_pages = 0; |
7171 | |
7172 | for_each_zone(zone) |
7173 | zone->zone_pgdat->min_slab_pages += (zone->managed_pages * |
7174 | sysctl_min_slab_ratio) / 100; |
7175 | } |
7176 | |
7177 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write, |
7178 | void __user *buffer, size_t *length, loff_t *ppos) |
7179 | { |
7180 | int rc; |
7181 | |
7182 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7183 | if (rc) |
7184 | return rc; |
7185 | |
7186 | setup_min_slab_ratio(); |
7187 | |
7188 | return 0; |
7189 | } |
7190 | #endif |
7191 | |
7192 | /* |
7193 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around |
7194 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() |
7195 | * whenever sysctl_lowmem_reserve_ratio changes. |
7196 | * |
7197 | * The reserve ratio obviously has absolutely no relation with the |
7198 | * minimum watermarks. The lowmem reserve ratio can only make sense |
7199 | * if in function of the boot time zone sizes. |
7200 | */ |
7201 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write, |
7202 | void __user *buffer, size_t *length, loff_t *ppos) |
7203 | { |
7204 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
7205 | setup_per_zone_lowmem_reserve(); |
7206 | return 0; |
7207 | } |
7208 | |
7209 | /* |
7210 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each |
7211 | * cpu. It is the fraction of total pages in each zone that a hot per cpu |
7212 | * pagelist can have before it gets flushed back to buddy allocator. |
7213 | */ |
7214 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, |
7215 | void __user *buffer, size_t *length, loff_t *ppos) |
7216 | { |
7217 | struct zone *zone; |
7218 | int old_percpu_pagelist_fraction; |
7219 | int ret; |
7220 | |
7221 | mutex_lock(&pcp_batch_high_lock); |
7222 | old_percpu_pagelist_fraction = percpu_pagelist_fraction; |
7223 | |
7224 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
7225 | if (!write || ret < 0) |
7226 | goto out; |
7227 | |
7228 | /* Sanity checking to avoid pcp imbalance */ |
7229 | if (percpu_pagelist_fraction && |
7230 | percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) { |
7231 | percpu_pagelist_fraction = old_percpu_pagelist_fraction; |
7232 | ret = -EINVAL; |
7233 | goto out; |
7234 | } |
7235 | |
7236 | /* No change? */ |
7237 | if (percpu_pagelist_fraction == old_percpu_pagelist_fraction) |
7238 | goto out; |
7239 | |
7240 | for_each_populated_zone(zone) { |
7241 | unsigned int cpu; |
7242 | |
7243 | for_each_possible_cpu(cpu) |
7244 | pageset_set_high_and_batch(zone, |
7245 | per_cpu_ptr(zone->pageset, cpu)); |
7246 | } |
7247 | out: |
7248 | mutex_unlock(&pcp_batch_high_lock); |
7249 | return ret; |
7250 | } |
7251 | |
7252 | #ifdef CONFIG_NUMA |
7253 | int hashdist = HASHDIST_DEFAULT; |
7254 | |
7255 | static int __init set_hashdist(char *str) |
7256 | { |
7257 | if (!str) |
7258 | return 0; |
7259 | hashdist = simple_strtoul(str, &str, 0); |
7260 | return 1; |
7261 | } |
7262 | __setup("hashdist=", set_hashdist); |
7263 | #endif |
7264 | |
7265 | #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES |
7266 | /* |
7267 | * Returns the number of pages that arch has reserved but |
7268 | * is not known to alloc_large_system_hash(). |
7269 | */ |
7270 | static unsigned long __init arch_reserved_kernel_pages(void) |
7271 | { |
7272 | return 0; |
7273 | } |
7274 | #endif |
7275 | |
7276 | /* |
7277 | * allocate a large system hash table from bootmem |
7278 | * - it is assumed that the hash table must contain an exact power-of-2 |
7279 | * quantity of entries |
7280 | * - limit is the number of hash buckets, not the total allocation size |
7281 | */ |
7282 | void *__init alloc_large_system_hash(const char *tablename, |
7283 | unsigned long bucketsize, |
7284 | unsigned long numentries, |
7285 | int scale, |
7286 | int flags, |
7287 | unsigned int *_hash_shift, |
7288 | unsigned int *_hash_mask, |
7289 | unsigned long low_limit, |
7290 | unsigned long high_limit) |
7291 | { |
7292 | unsigned long long max = high_limit; |
7293 | unsigned long log2qty, size; |
7294 | void *table = NULL; |
7295 | |
7296 | /* allow the kernel cmdline to have a say */ |
7297 | if (!numentries) { |
7298 | /* round applicable memory size up to nearest megabyte */ |
7299 | numentries = nr_kernel_pages; |
7300 | numentries -= arch_reserved_kernel_pages(); |
7301 | |
7302 | /* It isn't necessary when PAGE_SIZE >= 1MB */ |
7303 | if (PAGE_SHIFT < 20) |
7304 | numentries = round_up(numentries, (1<<20)/PAGE_SIZE); |
7305 | |
7306 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
7307 | if (scale > PAGE_SHIFT) |
7308 | numentries >>= (scale - PAGE_SHIFT); |
7309 | else |
7310 | numentries <<= (PAGE_SHIFT - scale); |
7311 | |
7312 | /* Make sure we've got at least a 0-order allocation.. */ |
7313 | if (unlikely(flags & HASH_SMALL)) { |
7314 | /* Makes no sense without HASH_EARLY */ |
7315 | WARN_ON(!(flags & HASH_EARLY)); |
7316 | if (!(numentries >> *_hash_shift)) { |
7317 | numentries = 1UL << *_hash_shift; |
7318 | BUG_ON(!numentries); |
7319 | } |
7320 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) |
7321 | numentries = PAGE_SIZE / bucketsize; |
7322 | } |
7323 | numentries = roundup_pow_of_two(numentries); |
7324 | |
7325 | /* limit allocation size to 1/16 total memory by default */ |
7326 | if (max == 0) { |
7327 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; |
7328 | do_div(max, bucketsize); |
7329 | } |
7330 | max = min(max, 0x80000000ULL); |
7331 | |
7332 | if (numentries < low_limit) |
7333 | numentries = low_limit; |
7334 | if (numentries > max) |
7335 | numentries = max; |
7336 | |
7337 | log2qty = ilog2(numentries); |
7338 | |
7339 | do { |
7340 | size = bucketsize << log2qty; |
7341 | if (flags & HASH_EARLY) |
7342 | table = memblock_virt_alloc_nopanic(size, 0); |
7343 | else if (hashdist) |
7344 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); |
7345 | else { |
7346 | /* |
7347 | * If bucketsize is not a power-of-two, we may free |
7348 | * some pages at the end of hash table which |
7349 | * alloc_pages_exact() automatically does |
7350 | */ |
7351 | if (get_order(size) < MAX_ORDER) { |
7352 | table = alloc_pages_exact(size, GFP_ATOMIC); |
7353 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
7354 | } |
7355 | } |
7356 | } while (!table && size > PAGE_SIZE && --log2qty); |
7357 | |
7358 | if (!table) |
7359 | panic("Failed to allocate %s hash table\n", tablename); |
7360 | |
7361 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n", |
7362 | tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size); |
7363 | |
7364 | if (_hash_shift) |
7365 | *_hash_shift = log2qty; |
7366 | if (_hash_mask) |
7367 | *_hash_mask = (1 << log2qty) - 1; |
7368 | |
7369 | return table; |
7370 | } |
7371 | |
7372 | /* |
7373 | * This function checks whether pageblock includes unmovable pages or not. |
7374 | * If @count is not zero, it is okay to include less @count unmovable pages |
7375 | * |
7376 | * PageLRU check without isolation or lru_lock could race so that |
7377 | * MIGRATE_MOVABLE block might include unmovable pages. It means you can't |
7378 | * expect this function should be exact. |
7379 | */ |
7380 | bool has_unmovable_pages(struct zone *zone, struct page *page, int count, |
7381 | bool skip_hwpoisoned_pages) |
7382 | { |
7383 | unsigned long pfn, iter, found; |
7384 | int mt; |
7385 | |
7386 | /* |
7387 | * For avoiding noise data, lru_add_drain_all() should be called |
7388 | * If ZONE_MOVABLE, the zone never contains unmovable pages |
7389 | */ |
7390 | if (zone_idx(zone) == ZONE_MOVABLE) |
7391 | return false; |
7392 | mt = get_pageblock_migratetype(page); |
7393 | if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt)) |
7394 | return false; |
7395 | |
7396 | pfn = page_to_pfn(page); |
7397 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { |
7398 | unsigned long check = pfn + iter; |
7399 | |
7400 | if (!pfn_valid_within(check)) |
7401 | continue; |
7402 | |
7403 | page = pfn_to_page(check); |
7404 | |
7405 | /* |
7406 | * Hugepages are not in LRU lists, but they're movable. |
7407 | * We need not scan over tail pages bacause we don't |
7408 | * handle each tail page individually in migration. |
7409 | */ |
7410 | if (PageHuge(page)) { |
7411 | iter = round_up(iter + 1, 1<<compound_order(page)) - 1; |
7412 | continue; |
7413 | } |
7414 | |
7415 | /* |
7416 | * We can't use page_count without pin a page |
7417 | * because another CPU can free compound page. |
7418 | * This check already skips compound tails of THP |
7419 | * because their page->_refcount is zero at all time. |
7420 | */ |
7421 | if (!page_ref_count(page)) { |
7422 | if (PageBuddy(page)) |
7423 | iter += (1 << page_order(page)) - 1; |
7424 | continue; |
7425 | } |
7426 | |
7427 | /* |
7428 | * The HWPoisoned page may be not in buddy system, and |
7429 | * page_count() is not 0. |
7430 | */ |
7431 | if (skip_hwpoisoned_pages && PageHWPoison(page)) |
7432 | continue; |
7433 | |
7434 | if (!PageLRU(page)) |
7435 | found++; |
7436 | /* |
7437 | * If there are RECLAIMABLE pages, we need to check |
7438 | * it. But now, memory offline itself doesn't call |
7439 | * shrink_node_slabs() and it still to be fixed. |
7440 | */ |
7441 | /* |
7442 | * If the page is not RAM, page_count()should be 0. |
7443 | * we don't need more check. This is an _used_ not-movable page. |
7444 | * |
7445 | * The problematic thing here is PG_reserved pages. PG_reserved |
7446 | * is set to both of a memory hole page and a _used_ kernel |
7447 | * page at boot. |
7448 | */ |
7449 | if (found > count) |
7450 | return true; |
7451 | } |
7452 | return false; |
7453 | } |
7454 | |
7455 | bool is_pageblock_removable_nolock(struct page *page) |
7456 | { |
7457 | struct zone *zone; |
7458 | unsigned long pfn; |
7459 | |
7460 | /* |
7461 | * We have to be careful here because we are iterating over memory |
7462 | * sections which are not zone aware so we might end up outside of |
7463 | * the zone but still within the section. |
7464 | * We have to take care about the node as well. If the node is offline |
7465 | * its NODE_DATA will be NULL - see page_zone. |
7466 | */ |
7467 | if (!node_online(page_to_nid(page))) |
7468 | return false; |
7469 | |
7470 | zone = page_zone(page); |
7471 | pfn = page_to_pfn(page); |
7472 | if (!zone_spans_pfn(zone, pfn)) |
7473 | return false; |
7474 | |
7475 | return !has_unmovable_pages(zone, page, 0, true); |
7476 | } |
7477 | |
7478 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
7479 | |
7480 | static unsigned long pfn_max_align_down(unsigned long pfn) |
7481 | { |
7482 | return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES, |
7483 | pageblock_nr_pages) - 1); |
7484 | } |
7485 | |
7486 | static unsigned long pfn_max_align_up(unsigned long pfn) |
7487 | { |
7488 | return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES, |
7489 | pageblock_nr_pages)); |
7490 | } |
7491 | |
7492 | /* [start, end) must belong to a single zone. */ |
7493 | static int __alloc_contig_migrate_range(struct compact_control *cc, |
7494 | unsigned long start, unsigned long end) |
7495 | { |
7496 | /* This function is based on compact_zone() from compaction.c. */ |
7497 | unsigned long nr_reclaimed; |
7498 | unsigned long pfn = start; |
7499 | unsigned int tries = 0; |
7500 | int ret = 0; |
7501 | |
7502 | migrate_prep(); |
7503 | |
7504 | while (pfn < end || !list_empty(&cc->migratepages)) { |
7505 | if (fatal_signal_pending(current)) { |
7506 | ret = -EINTR; |
7507 | break; |
7508 | } |
7509 | |
7510 | if (list_empty(&cc->migratepages)) { |
7511 | cc->nr_migratepages = 0; |
7512 | pfn = isolate_migratepages_range(cc, pfn, end); |
7513 | if (!pfn) { |
7514 | ret = -EINTR; |
7515 | break; |
7516 | } |
7517 | tries = 0; |
7518 | } else if (++tries == 5) { |
7519 | ret = ret < 0 ? ret : -EBUSY; |
7520 | break; |
7521 | } |
7522 | |
7523 | nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, |
7524 | &cc->migratepages); |
7525 | cc->nr_migratepages -= nr_reclaimed; |
7526 | |
7527 | ret = migrate_pages(&cc->migratepages, alloc_migrate_target, |
7528 | NULL, 0, cc->mode, MR_CMA); |
7529 | } |
7530 | if (ret < 0) { |
7531 | putback_movable_pages(&cc->migratepages); |
7532 | return ret; |
7533 | } |
7534 | return 0; |
7535 | } |
7536 | |
7537 | /** |
7538 | * alloc_contig_range() -- tries to allocate given range of pages |
7539 | * @start: start PFN to allocate |
7540 | * @end: one-past-the-last PFN to allocate |
7541 | * @migratetype: migratetype of the underlaying pageblocks (either |
7542 | * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks |
7543 | * in range must have the same migratetype and it must |
7544 | * be either of the two. |
7545 | * |
7546 | * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES |
7547 | * aligned, however it's the caller's responsibility to guarantee that |
7548 | * we are the only thread that changes migrate type of pageblocks the |
7549 | * pages fall in. |
7550 | * |
7551 | * The PFN range must belong to a single zone. |
7552 | * |
7553 | * Returns zero on success or negative error code. On success all |
7554 | * pages which PFN is in [start, end) are allocated for the caller and |
7555 | * need to be freed with free_contig_range(). |
7556 | */ |
7557 | int alloc_contig_range(unsigned long start, unsigned long end, |
7558 | unsigned migratetype) |
7559 | { |
7560 | unsigned long outer_start, outer_end; |
7561 | unsigned int order; |
7562 | int ret = 0; |
7563 | |
7564 | struct compact_control cc = { |
7565 | .nr_migratepages = 0, |
7566 | .order = -1, |
7567 | .zone = page_zone(pfn_to_page(start)), |
7568 | .mode = MIGRATE_SYNC, |
7569 | .ignore_skip_hint = true, |
7570 | }; |
7571 | INIT_LIST_HEAD(&cc.migratepages); |
7572 | |
7573 | /* |
7574 | * What we do here is we mark all pageblocks in range as |
7575 | * MIGRATE_ISOLATE. Because pageblock and max order pages may |
7576 | * have different sizes, and due to the way page allocator |
7577 | * work, we align the range to biggest of the two pages so |
7578 | * that page allocator won't try to merge buddies from |
7579 | * different pageblocks and change MIGRATE_ISOLATE to some |
7580 | * other migration type. |
7581 | * |
7582 | * Once the pageblocks are marked as MIGRATE_ISOLATE, we |
7583 | * migrate the pages from an unaligned range (ie. pages that |
7584 | * we are interested in). This will put all the pages in |
7585 | * range back to page allocator as MIGRATE_ISOLATE. |
7586 | * |
7587 | * When this is done, we take the pages in range from page |
7588 | * allocator removing them from the buddy system. This way |
7589 | * page allocator will never consider using them. |
7590 | * |
7591 | * This lets us mark the pageblocks back as |
7592 | * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the |
7593 | * aligned range but not in the unaligned, original range are |
7594 | * put back to page allocator so that buddy can use them. |
7595 | */ |
7596 | |
7597 | ret = start_isolate_page_range(pfn_max_align_down(start), |
7598 | pfn_max_align_up(end), migratetype, |
7599 | false); |
7600 | if (ret) |
7601 | return ret; |
7602 | |
7603 | /* |
7604 | * In case of -EBUSY, we'd like to know which page causes problem. |
7605 | * So, just fall through. test_pages_isolated() has a tracepoint |
7606 | * which will report the busy page. |
7607 | * |
7608 | * It is possible that busy pages could become available before |
7609 | * the call to test_pages_isolated, and the range will actually be |
7610 | * allocated. So, if we fall through be sure to clear ret so that |
7611 | * -EBUSY is not accidentally used or returned to caller. |
7612 | */ |
7613 | ret = __alloc_contig_migrate_range(&cc, start, end); |
7614 | if (ret && ret != -EBUSY) |
7615 | goto done; |
7616 | ret =0; |
7617 | |
7618 | /* |
7619 | * Pages from [start, end) are within a MAX_ORDER_NR_PAGES |
7620 | * aligned blocks that are marked as MIGRATE_ISOLATE. What's |
7621 | * more, all pages in [start, end) are free in page allocator. |
7622 | * What we are going to do is to allocate all pages from |
7623 | * [start, end) (that is remove them from page allocator). |
7624 | * |
7625 | * The only problem is that pages at the beginning and at the |
7626 | * end of interesting range may be not aligned with pages that |
7627 | * page allocator holds, ie. they can be part of higher order |
7628 | * pages. Because of this, we reserve the bigger range and |
7629 | * once this is done free the pages we are not interested in. |
7630 | * |
7631 | * We don't have to hold zone->lock here because the pages are |
7632 | * isolated thus they won't get removed from buddy. |
7633 | */ |
7634 | |
7635 | lru_add_drain_all(); |
7636 | drain_all_pages(cc.zone); |
7637 | |
7638 | order = 0; |
7639 | outer_start = start; |
7640 | while (!PageBuddy(pfn_to_page(outer_start))) { |
7641 | if (++order >= MAX_ORDER) { |
7642 | outer_start = start; |
7643 | break; |
7644 | } |
7645 | outer_start &= ~0UL << order; |
7646 | } |
7647 | |
7648 | if (outer_start != start) { |
7649 | order = page_order(pfn_to_page(outer_start)); |
7650 | |
7651 | /* |
7652 | * outer_start page could be small order buddy page and |
7653 | * it doesn't include start page. Adjust outer_start |
7654 | * in this case to report failed page properly |
7655 | * on tracepoint in test_pages_isolated() |
7656 | */ |
7657 | if (outer_start + (1UL << order) <= start) |
7658 | outer_start = start; |
7659 | } |
7660 | |
7661 | /* Make sure the range is really isolated. */ |
7662 | if (test_pages_isolated(outer_start, end, false)) { |
7663 | pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n", |
7664 | __func__, outer_start, end); |
7665 | ret = -EBUSY; |
7666 | goto done; |
7667 | } |
7668 | |
7669 | /* Grab isolated pages from freelists. */ |
7670 | outer_end = isolate_freepages_range(&cc, outer_start, end); |
7671 | if (!outer_end) { |
7672 | ret = -EBUSY; |
7673 | goto done; |
7674 | } |
7675 | |
7676 | /* Free head and tail (if any) */ |
7677 | if (start != outer_start) |
7678 | free_contig_range(outer_start, start - outer_start); |
7679 | if (end != outer_end) |
7680 | free_contig_range(end, outer_end - end); |
7681 | |
7682 | done: |
7683 | undo_isolate_page_range(pfn_max_align_down(start), |
7684 | pfn_max_align_up(end), migratetype); |
7685 | return ret; |
7686 | } |
7687 | |
7688 | void free_contig_range(unsigned long pfn, unsigned nr_pages) |
7689 | { |
7690 | unsigned int count = 0; |
7691 | |
7692 | for (; nr_pages--; pfn++) { |
7693 | struct page *page = pfn_to_page(pfn); |
7694 | |
7695 | count += page_count(page) != 1; |
7696 | __free_page(page); |
7697 | } |
7698 | WARN(count != 0, "%d pages are still in use!\n", count); |
7699 | } |
7700 | #endif |
7701 | |
7702 | #ifdef CONFIG_MEMORY_HOTPLUG |
7703 | /* |
7704 | * The zone indicated has a new number of managed_pages; batch sizes and percpu |
7705 | * page high values need to be recalulated. |
7706 | */ |
7707 | void __meminit zone_pcp_update(struct zone *zone) |
7708 | { |
7709 | unsigned cpu; |
7710 | mutex_lock(&pcp_batch_high_lock); |
7711 | for_each_possible_cpu(cpu) |
7712 | pageset_set_high_and_batch(zone, |
7713 | per_cpu_ptr(zone->pageset, cpu)); |
7714 | mutex_unlock(&pcp_batch_high_lock); |
7715 | } |
7716 | #endif |
7717 | |
7718 | void zone_pcp_reset(struct zone *zone) |
7719 | { |
7720 | unsigned long flags; |
7721 | int cpu; |
7722 | struct per_cpu_pageset *pset; |
7723 | |
7724 | /* avoid races with drain_pages() */ |
7725 | local_irq_save(flags); |
7726 | if (zone->pageset != &boot_pageset) { |
7727 | for_each_online_cpu(cpu) { |
7728 | pset = per_cpu_ptr(zone->pageset, cpu); |
7729 | drain_zonestat(zone, pset); |
7730 | } |
7731 | free_percpu(zone->pageset); |
7732 | zone->pageset = &boot_pageset; |
7733 | } |
7734 | local_irq_restore(flags); |
7735 | } |
7736 | |
7737 | #ifdef CONFIG_MEMORY_HOTREMOVE |
7738 | /* |
7739 | * All pages in the range must be in a single zone and isolated |
7740 | * before calling this. |
7741 | */ |
7742 | void |
7743 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
7744 | { |
7745 | struct page *page; |
7746 | struct zone *zone; |
7747 | unsigned int order, i; |
7748 | unsigned long pfn; |
7749 | unsigned long flags; |
7750 | /* find the first valid pfn */ |
7751 | for (pfn = start_pfn; pfn < end_pfn; pfn++) |
7752 | if (pfn_valid(pfn)) |
7753 | break; |
7754 | if (pfn == end_pfn) |
7755 | return; |
7756 | zone = page_zone(pfn_to_page(pfn)); |
7757 | spin_lock_irqsave(&zone->lock, flags); |
7758 | pfn = start_pfn; |
7759 | while (pfn < end_pfn) { |
7760 | if (!pfn_valid(pfn)) { |
7761 | pfn++; |
7762 | continue; |
7763 | } |
7764 | page = pfn_to_page(pfn); |
7765 | /* |
7766 | * The HWPoisoned page may be not in buddy system, and |
7767 | * page_count() is not 0. |
7768 | */ |
7769 | if (unlikely(!PageBuddy(page) && PageHWPoison(page))) { |
7770 | pfn++; |
7771 | SetPageReserved(page); |
7772 | continue; |
7773 | } |
7774 | |
7775 | BUG_ON(page_count(page)); |
7776 | BUG_ON(!PageBuddy(page)); |
7777 | order = page_order(page); |
7778 | #ifdef CONFIG_DEBUG_VM |
7779 | pr_info("remove from free list %lx %d %lx\n", |
7780 | pfn, 1 << order, end_pfn); |
7781 | #endif |
7782 | list_del(&page->lru); |
7783 | rmv_page_order(page); |
7784 | zone->free_area[order].nr_free--; |
7785 | #ifdef CONFIG_AMLOGIC_MEMORY_EXTEND |
7786 | __mod_zone_migrate_state(zone, -(1 << order), |
7787 | get_pcppage_migratetype(page)); |
7788 | #endif /* CONFIG_AMLOGIC_MEMORY_EXTEND */ |
7789 | for (i = 0; i < (1 << order); i++) |
7790 | SetPageReserved((page+i)); |
7791 | pfn += (1 << order); |
7792 | } |
7793 | spin_unlock_irqrestore(&zone->lock, flags); |
7794 | } |
7795 | #endif |
7796 | |
7797 | bool is_free_buddy_page(struct page *page) |
7798 | { |
7799 | struct zone *zone = page_zone(page); |
7800 | unsigned long pfn = page_to_pfn(page); |
7801 | unsigned long flags; |
7802 | unsigned int order; |
7803 | |
7804 | spin_lock_irqsave(&zone->lock, flags); |
7805 | for (order = 0; order < MAX_ORDER; order++) { |
7806 | struct page *page_head = page - (pfn & ((1 << order) - 1)); |
7807 | |
7808 | if (PageBuddy(page_head) && page_order(page_head) >= order) |
7809 | break; |
7810 | } |
7811 | spin_unlock_irqrestore(&zone->lock, flags); |
7812 | |
7813 | return order < MAX_ORDER; |
7814 | } |
7815 |