Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * linux/mm/page_alloc.c
4 : *
5 : * Manages the free list, the system allocates free pages here.
6 : * Note that kmalloc() lives in slab.c
7 : *
8 : * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
9 : * Swap reorganised 29.12.95, Stephen Tweedie
10 : * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
11 : * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
12 : * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
13 : * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
14 : * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
15 : * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
16 : */
17 :
18 : #include <linux/stddef.h>
19 : #include <linux/mm.h>
20 : #include <linux/highmem.h>
21 : #include <linux/swap.h>
22 : #include <linux/interrupt.h>
23 : #include <linux/pagemap.h>
24 : #include <linux/jiffies.h>
25 : #include <linux/memblock.h>
26 : #include <linux/compiler.h>
27 : #include <linux/kernel.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/topology.h>
37 : #include <linux/sysctl.h>
38 : #include <linux/cpu.h>
39 : #include <linux/cpuset.h>
40 : #include <linux/memory_hotplug.h>
41 : #include <linux/nodemask.h>
42 : #include <linux/vmalloc.h>
43 : #include <linux/vmstat.h>
44 : #include <linux/mempolicy.h>
45 : #include <linux/memremap.h>
46 : #include <linux/stop_machine.h>
47 : #include <linux/random.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/debugobjects.h>
54 : #include <linux/kmemleak.h>
55 : #include <linux/compaction.h>
56 : #include <trace/events/kmem.h>
57 : #include <trace/events/oom.h>
58 : #include <linux/prefetch.h>
59 : #include <linux/mm_inline.h>
60 : #include <linux/mmu_notifier.h>
61 : #include <linux/migrate.h>
62 : #include <linux/hugetlb.h>
63 : #include <linux/sched/rt.h>
64 : #include <linux/sched/mm.h>
65 : #include <linux/page_owner.h>
66 : #include <linux/kthread.h>
67 : #include <linux/memcontrol.h>
68 : #include <linux/ftrace.h>
69 : #include <linux/lockdep.h>
70 : #include <linux/nmi.h>
71 : #include <linux/psi.h>
72 : #include <linux/padata.h>
73 : #include <linux/khugepaged.h>
74 : #include <linux/buffer_head.h>
75 :
76 : #include <asm/sections.h>
77 : #include <asm/tlbflush.h>
78 : #include <asm/div64.h>
79 : #include "internal.h"
80 : #include "shuffle.h"
81 : #include "page_reporting.h"
82 :
83 : /* Free Page Internal flags: for internal, non-pcp variants of free_pages(). */
84 : typedef int __bitwise fpi_t;
85 :
86 : /* No special request */
87 : #define FPI_NONE ((__force fpi_t)0)
88 :
89 : /*
90 : * Skip free page reporting notification for the (possibly merged) page.
91 : * This does not hinder free page reporting from grabbing the page,
92 : * reporting it and marking it "reported" - it only skips notifying
93 : * the free page reporting infrastructure about a newly freed page. For
94 : * example, used when temporarily pulling a page from a freelist and
95 : * putting it back unmodified.
96 : */
97 : #define FPI_SKIP_REPORT_NOTIFY ((__force fpi_t)BIT(0))
98 :
99 : /*
100 : * Place the (possibly merged) page to the tail of the freelist. Will ignore
101 : * page shuffling (relevant code - e.g., memory onlining - is expected to
102 : * shuffle the whole zone).
103 : *
104 : * Note: No code should rely on this flag for correctness - it's purely
105 : * to allow for optimizations when handing back either fresh pages
106 : * (memory onlining) or untouched pages (page isolation, free page
107 : * reporting).
108 : */
109 : #define FPI_TO_TAIL ((__force fpi_t)BIT(1))
110 :
111 : /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
112 : static DEFINE_MUTEX(pcp_batch_high_lock);
113 : #define MIN_PERCPU_PAGELIST_FRACTION (8)
114 :
115 : #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
116 : DEFINE_PER_CPU(int, numa_node);
117 : EXPORT_PER_CPU_SYMBOL(numa_node);
118 : #endif
119 :
120 : DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
121 :
122 : #ifdef CONFIG_HAVE_MEMORYLESS_NODES
123 : /*
124 : * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
125 : * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
126 : * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
127 : * defined in <linux/topology.h>.
128 : */
129 : DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
130 : EXPORT_PER_CPU_SYMBOL(_numa_mem_);
131 : #endif
132 :
133 : /* work_structs for global per-cpu drains */
134 : struct pcpu_drain {
135 : struct zone *zone;
136 : struct work_struct work;
137 : };
138 : static DEFINE_MUTEX(pcpu_drain_mutex);
139 : static DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain);
140 :
141 : #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
142 : volatile unsigned long latent_entropy __latent_entropy;
143 : EXPORT_SYMBOL(latent_entropy);
144 : #endif
145 :
146 : /*
147 : * Array of node states.
148 : */
149 : nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
150 : [N_POSSIBLE] = NODE_MASK_ALL,
151 : [N_ONLINE] = { { [0] = 1UL } },
152 : #ifndef CONFIG_NUMA
153 : [N_NORMAL_MEMORY] = { { [0] = 1UL } },
154 : #ifdef CONFIG_HIGHMEM
155 : [N_HIGH_MEMORY] = { { [0] = 1UL } },
156 : #endif
157 : [N_MEMORY] = { { [0] = 1UL } },
158 : [N_CPU] = { { [0] = 1UL } },
159 : #endif /* NUMA */
160 : };
161 : EXPORT_SYMBOL(node_states);
162 :
163 : atomic_long_t _totalram_pages __read_mostly;
164 : EXPORT_SYMBOL(_totalram_pages);
165 : unsigned long totalreserve_pages __read_mostly;
166 : unsigned long totalcma_pages __read_mostly;
167 :
168 : int percpu_pagelist_fraction;
169 : gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
170 : DEFINE_STATIC_KEY_FALSE(init_on_alloc);
171 : EXPORT_SYMBOL(init_on_alloc);
172 :
173 : DEFINE_STATIC_KEY_FALSE(init_on_free);
174 : EXPORT_SYMBOL(init_on_free);
175 :
176 : static bool _init_on_alloc_enabled_early __read_mostly
177 : = IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON);
178 0 : static int __init early_init_on_alloc(char *buf)
179 : {
180 :
181 0 : return kstrtobool(buf, &_init_on_alloc_enabled_early);
182 : }
183 : early_param("init_on_alloc", early_init_on_alloc);
184 :
185 : static bool _init_on_free_enabled_early __read_mostly
186 : = IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON);
187 0 : static int __init early_init_on_free(char *buf)
188 : {
189 0 : return kstrtobool(buf, &_init_on_free_enabled_early);
190 : }
191 : early_param("init_on_free", early_init_on_free);
192 :
193 : /*
194 : * A cached value of the page's pageblock's migratetype, used when the page is
195 : * put on a pcplist. Used to avoid the pageblock migratetype lookup when
196 : * freeing from pcplists in most cases, at the cost of possibly becoming stale.
197 : * Also the migratetype set in the page does not necessarily match the pcplist
198 : * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
199 : * other index - this ensures that it will be put on the correct CMA freelist.
200 : */
201 179298 : static inline int get_pcppage_migratetype(struct page *page)
202 : {
203 179298 : return page->index;
204 : }
205 :
206 216485 : static inline void set_pcppage_migratetype(struct page *page, int migratetype)
207 : {
208 216485 : page->index = migratetype;
209 83160 : }
210 :
211 : #ifdef CONFIG_PM_SLEEP
212 : /*
213 : * The following functions are used by the suspend/hibernate code to temporarily
214 : * change gfp_allowed_mask in order to avoid using I/O during memory allocations
215 : * while devices are suspended. To avoid races with the suspend/hibernate code,
216 : * they should always be called with system_transition_mutex held
217 : * (gfp_allowed_mask also should only be modified with system_transition_mutex
218 : * held, unless the suspend/hibernate code is guaranteed not to run in parallel
219 : * with that modification).
220 : */
221 :
222 : static gfp_t saved_gfp_mask;
223 :
224 : void pm_restore_gfp_mask(void)
225 : {
226 : WARN_ON(!mutex_is_locked(&system_transition_mutex));
227 : if (saved_gfp_mask) {
228 : gfp_allowed_mask = saved_gfp_mask;
229 : saved_gfp_mask = 0;
230 : }
231 : }
232 :
233 : void pm_restrict_gfp_mask(void)
234 : {
235 : WARN_ON(!mutex_is_locked(&system_transition_mutex));
236 : WARN_ON(saved_gfp_mask);
237 : saved_gfp_mask = gfp_allowed_mask;
238 : gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
239 : }
240 :
241 : bool pm_suspended_storage(void)
242 : {
243 : if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
244 : return false;
245 : return true;
246 : }
247 : #endif /* CONFIG_PM_SLEEP */
248 :
249 : #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
250 : unsigned int pageblock_order __read_mostly;
251 : #endif
252 :
253 : static void __free_pages_ok(struct page *page, unsigned int order,
254 : fpi_t fpi_flags);
255 :
256 : /*
257 : * results with 256, 32 in the lowmem_reserve sysctl:
258 : * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
259 : * 1G machine -> (16M dma, 784M normal, 224M high)
260 : * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
261 : * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
262 : * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
263 : *
264 : * TBD: should special case ZONE_DMA32 machines here - in those we normally
265 : * don't need any ZONE_NORMAL reservation
266 : */
267 : int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
268 : #ifdef CONFIG_ZONE_DMA
269 : [ZONE_DMA] = 256,
270 : #endif
271 : #ifdef CONFIG_ZONE_DMA32
272 : [ZONE_DMA32] = 256,
273 : #endif
274 : [ZONE_NORMAL] = 32,
275 : #ifdef CONFIG_HIGHMEM
276 : [ZONE_HIGHMEM] = 0,
277 : #endif
278 : [ZONE_MOVABLE] = 0,
279 : };
280 :
281 : static char * const zone_names[MAX_NR_ZONES] = {
282 : #ifdef CONFIG_ZONE_DMA
283 : "DMA",
284 : #endif
285 : #ifdef CONFIG_ZONE_DMA32
286 : "DMA32",
287 : #endif
288 : "Normal",
289 : #ifdef CONFIG_HIGHMEM
290 : "HighMem",
291 : #endif
292 : "Movable",
293 : #ifdef CONFIG_ZONE_DEVICE
294 : "Device",
295 : #endif
296 : };
297 :
298 : const char * const migratetype_names[MIGRATE_TYPES] = {
299 : "Unmovable",
300 : "Movable",
301 : "Reclaimable",
302 : "HighAtomic",
303 : #ifdef CONFIG_CMA
304 : "CMA",
305 : #endif
306 : #ifdef CONFIG_MEMORY_ISOLATION
307 : "Isolate",
308 : #endif
309 : };
310 :
311 : compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = {
312 : [NULL_COMPOUND_DTOR] = NULL,
313 : [COMPOUND_PAGE_DTOR] = free_compound_page,
314 : #ifdef CONFIG_HUGETLB_PAGE
315 : [HUGETLB_PAGE_DTOR] = free_huge_page,
316 : #endif
317 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
318 : [TRANSHUGE_PAGE_DTOR] = free_transhuge_page,
319 : #endif
320 : };
321 :
322 : int min_free_kbytes = 1024;
323 : int user_min_free_kbytes = -1;
324 : #ifdef CONFIG_DISCONTIGMEM
325 : /*
326 : * DiscontigMem defines memory ranges as separate pg_data_t even if the ranges
327 : * are not on separate NUMA nodes. Functionally this works but with
328 : * watermark_boost_factor, it can reclaim prematurely as the ranges can be
329 : * quite small. By default, do not boost watermarks on discontigmem as in
330 : * many cases very high-order allocations like THP are likely to be
331 : * unsupported and the premature reclaim offsets the advantage of long-term
332 : * fragmentation avoidance.
333 : */
334 : int watermark_boost_factor __read_mostly;
335 : #else
336 : int watermark_boost_factor __read_mostly = 15000;
337 : #endif
338 : int watermark_scale_factor = 10;
339 :
340 : static unsigned long nr_kernel_pages __initdata;
341 : static unsigned long nr_all_pages __initdata;
342 : static unsigned long dma_reserve __initdata;
343 :
344 : static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
345 : static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
346 : static unsigned long required_kernelcore __initdata;
347 : static unsigned long required_kernelcore_percent __initdata;
348 : static unsigned long required_movablecore __initdata;
349 : static unsigned long required_movablecore_percent __initdata;
350 : static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
351 : static bool mirrored_kernelcore __meminitdata;
352 :
353 : /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
354 : int movable_zone;
355 : EXPORT_SYMBOL(movable_zone);
356 :
357 : #if MAX_NUMNODES > 1
358 : unsigned int nr_node_ids __read_mostly = MAX_NUMNODES;
359 : unsigned int nr_online_nodes __read_mostly = 1;
360 : EXPORT_SYMBOL(nr_node_ids);
361 : EXPORT_SYMBOL(nr_online_nodes);
362 : #endif
363 :
364 : int page_group_by_mobility_disabled __read_mostly;
365 :
366 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
367 : /*
368 : * During boot we initialize deferred pages on-demand, as needed, but once
369 : * page_alloc_init_late() has finished, the deferred pages are all initialized,
370 : * and we can permanently disable that path.
371 : */
372 : static DEFINE_STATIC_KEY_TRUE(deferred_pages);
373 :
374 : /*
375 : * Calling kasan_free_pages() only after deferred memory initialization
376 : * has completed. Poisoning pages during deferred memory init will greatly
377 : * lengthen the process and cause problem in large memory systems as the
378 : * deferred pages initialization is done with interrupt disabled.
379 : *
380 : * Assuming that there will be no reference to those newly initialized
381 : * pages before they are ever allocated, this should have no effect on
382 : * KASAN memory tracking as the poison will be properly inserted at page
383 : * allocation time. The only corner case is when pages are allocated by
384 : * on-demand allocation and then freed again before the deferred pages
385 : * initialization is done, but this is not likely to happen.
386 : */
387 : static inline void kasan_free_nondeferred_pages(struct page *page, int order)
388 : {
389 : if (!static_branch_unlikely(&deferred_pages))
390 : kasan_free_pages(page, order);
391 : }
392 :
393 : /* Returns true if the struct page for the pfn is uninitialised */
394 : static inline bool __meminit early_page_uninitialised(unsigned long pfn)
395 : {
396 : int nid = early_pfn_to_nid(pfn);
397 :
398 : if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
399 : return true;
400 :
401 : return false;
402 : }
403 :
404 : /*
405 : * Returns true when the remaining initialisation should be deferred until
406 : * later in the boot cycle when it can be parallelised.
407 : */
408 : static bool __meminit
409 : defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
410 : {
411 : static unsigned long prev_end_pfn, nr_initialised;
412 :
413 : /*
414 : * prev_end_pfn static that contains the end of previous zone
415 : * No need to protect because called very early in boot before smp_init.
416 : */
417 : if (prev_end_pfn != end_pfn) {
418 : prev_end_pfn = end_pfn;
419 : nr_initialised = 0;
420 : }
421 :
422 : /* Always populate low zones for address-constrained allocations */
423 : if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
424 : return false;
425 :
426 : if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX)
427 : return true;
428 : /*
429 : * We start only with one section of pages, more pages are added as
430 : * needed until the rest of deferred pages are initialized.
431 : */
432 : nr_initialised++;
433 : if ((nr_initialised > PAGES_PER_SECTION) &&
434 : (pfn & (PAGES_PER_SECTION - 1)) == 0) {
435 : NODE_DATA(nid)->first_deferred_pfn = pfn;
436 : return true;
437 : }
438 : return false;
439 : }
440 : #else
441 : #define kasan_free_nondeferred_pages(p, o) kasan_free_pages(p, o)
442 :
443 233 : static inline bool early_page_uninitialised(unsigned long pfn)
444 : {
445 233 : return false;
446 : }
447 :
448 : static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
449 : {
450 : return false;
451 : }
452 : #endif
453 :
454 : /* Return a pointer to the bitmap storing bits affecting a block of pages */
455 148615 : static inline unsigned long *get_pageblock_bitmap(struct page *page,
456 : unsigned long pfn)
457 : {
458 : #ifdef CONFIG_SPARSEMEM
459 148615 : return section_to_usemap(__pfn_to_section(pfn));
460 : #else
461 : return page_zone(page)->pageblock_flags;
462 : #endif /* CONFIG_SPARSEMEM */
463 : }
464 :
465 148615 : static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
466 : {
467 : #ifdef CONFIG_SPARSEMEM
468 148615 : pfn &= (PAGES_PER_SECTION-1);
469 : #else
470 : pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
471 : #endif /* CONFIG_SPARSEMEM */
472 148615 : return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
473 : }
474 :
475 : static __always_inline
476 148322 : unsigned long __get_pfnblock_flags_mask(struct page *page,
477 : unsigned long pfn,
478 : unsigned long mask)
479 : {
480 148322 : unsigned long *bitmap;
481 148322 : unsigned long bitidx, word_bitidx;
482 148322 : unsigned long word;
483 :
484 148322 : bitmap = get_pageblock_bitmap(page, pfn);
485 148322 : bitidx = pfn_to_bitidx(page, pfn);
486 148322 : word_bitidx = bitidx / BITS_PER_LONG;
487 148322 : bitidx &= (BITS_PER_LONG-1);
488 :
489 148322 : word = bitmap[word_bitidx];
490 148322 : return (word >> bitidx) & mask;
491 : }
492 :
493 : /**
494 : * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
495 : * @page: The page within the block of interest
496 : * @pfn: The target page frame number
497 : * @mask: mask of bits that the caller is interested in
498 : *
499 : * Return: pageblock_bits flags
500 : */
501 38 : unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
502 : unsigned long mask)
503 : {
504 38 : return __get_pfnblock_flags_mask(page, pfn, mask);
505 : }
506 :
507 148284 : static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
508 : {
509 148284 : return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK);
510 : }
511 :
512 : /**
513 : * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
514 : * @page: The page within the block of interest
515 : * @flags: The flags to set
516 : * @pfn: The target page frame number
517 : * @mask: mask of bits that the caller is interested in
518 : */
519 293 : void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
520 : unsigned long pfn,
521 : unsigned long mask)
522 : {
523 293 : unsigned long *bitmap;
524 293 : unsigned long bitidx, word_bitidx;
525 293 : unsigned long old_word, word;
526 :
527 293 : BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
528 293 : BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits));
529 :
530 293 : bitmap = get_pageblock_bitmap(page, pfn);
531 293 : bitidx = pfn_to_bitidx(page, pfn);
532 293 : word_bitidx = bitidx / BITS_PER_LONG;
533 293 : bitidx &= (BITS_PER_LONG-1);
534 :
535 586 : VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
536 :
537 293 : mask <<= bitidx;
538 293 : flags <<= bitidx;
539 :
540 293 : word = READ_ONCE(bitmap[word_bitidx]);
541 293 : for (;;) {
542 293 : old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
543 293 : if (word == old_word)
544 : break;
545 : word = old_word;
546 : }
547 293 : }
548 :
549 293 : void set_pageblock_migratetype(struct page *page, int migratetype)
550 : {
551 293 : if (unlikely(page_group_by_mobility_disabled &&
552 : migratetype < MIGRATE_PCPTYPES))
553 0 : migratetype = MIGRATE_UNMOVABLE;
554 :
555 293 : set_pfnblock_flags_mask(page, (unsigned long)migratetype,
556 293 : page_to_pfn(page), MIGRATETYPE_MASK);
557 293 : }
558 :
559 : #ifdef CONFIG_DEBUG_VM
560 290654 : static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
561 : {
562 290654 : int ret = 0;
563 290654 : unsigned seq;
564 290654 : unsigned long pfn = page_to_pfn(page);
565 290654 : unsigned long sp, start_pfn;
566 :
567 290654 : do {
568 290654 : seq = zone_span_seqbegin(zone);
569 290654 : start_pfn = zone->zone_start_pfn;
570 290654 : sp = zone->spanned_pages;
571 581316 : if (!zone_spans_pfn(zone, pfn))
572 0 : ret = 1;
573 0 : } while (zone_span_seqretry(zone, seq));
574 :
575 0 : if (ret)
576 0 : pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
577 : pfn, zone_to_nid(zone), zone->name,
578 : start_pfn, start_pfn + sp);
579 :
580 290654 : return ret;
581 : }
582 :
583 290662 : static int page_is_consistent(struct zone *zone, struct page *page)
584 : {
585 290662 : if (!pfn_valid_within(page_to_pfn(page)))
586 : return 0;
587 290662 : if (zone != page_zone(page))
588 0 : return 0;
589 :
590 : return 1;
591 : }
592 : /*
593 : * Temporary debugging check for pages not lying within a given zone.
594 : */
595 290653 : static int __maybe_unused bad_range(struct zone *zone, struct page *page)
596 : {
597 290653 : if (page_outside_zone_boundaries(zone, page))
598 : return 1;
599 290662 : if (!page_is_consistent(zone, page))
600 0 : return 1;
601 :
602 : return 0;
603 : }
604 : #else
605 : static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
606 : {
607 : return 0;
608 : }
609 : #endif
610 :
611 0 : static void bad_page(struct page *page, const char *reason)
612 : {
613 0 : static unsigned long resume;
614 0 : static unsigned long nr_shown;
615 0 : static unsigned long nr_unshown;
616 :
617 : /*
618 : * Allow a burst of 60 reports, then keep quiet for that minute;
619 : * or allow a steady drip of one report per second.
620 : */
621 0 : if (nr_shown == 60) {
622 0 : if (time_before(jiffies, resume)) {
623 0 : nr_unshown++;
624 0 : goto out;
625 : }
626 0 : if (nr_unshown) {
627 0 : pr_alert(
628 : "BUG: Bad page state: %lu messages suppressed\n",
629 : nr_unshown);
630 0 : nr_unshown = 0;
631 : }
632 0 : nr_shown = 0;
633 : }
634 0 : if (nr_shown++ == 0)
635 0 : resume = jiffies + 60 * HZ;
636 :
637 0 : pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
638 : current->comm, page_to_pfn(page));
639 0 : __dump_page(page, reason);
640 0 : dump_page_owner(page);
641 :
642 0 : print_modules();
643 0 : dump_stack();
644 0 : out:
645 : /* Leave bad fields for debug, except PageBuddy could make trouble */
646 0 : page_mapcount_reset(page); /* remove PageBuddy */
647 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
648 0 : }
649 :
650 : /*
651 : * Higher-order pages are called "compound pages". They are structured thusly:
652 : *
653 : * The first PAGE_SIZE page is called the "head page" and have PG_head set.
654 : *
655 : * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
656 : * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
657 : *
658 : * The first tail page's ->compound_dtor holds the offset in array of compound
659 : * page destructors. See compound_page_dtors.
660 : *
661 : * The first tail page's ->compound_order holds the order of allocation.
662 : * This usage means that zero-order pages may not be compound.
663 : */
664 :
665 63 : void free_compound_page(struct page *page)
666 : {
667 63 : mem_cgroup_uncharge(page);
668 63 : __free_pages_ok(page, compound_order(page), FPI_NONE);
669 63 : }
670 :
671 21783 : void prep_compound_page(struct page *page, unsigned int order)
672 : {
673 21783 : int i;
674 21783 : int nr_pages = 1 << order;
675 :
676 21783 : __SetPageHead(page);
677 129557 : for (i = 1; i < nr_pages; i++) {
678 85991 : struct page *p = page + i;
679 85991 : set_page_count(p, 0);
680 85991 : p->mapping = TAIL_MAPPING;
681 85991 : set_compound_head(p, page);
682 : }
683 :
684 21783 : set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
685 21783 : set_compound_order(page, order);
686 21783 : atomic_set(compound_mapcount_ptr(page), -1);
687 21783 : if (hpage_pincount_available(page))
688 10160 : atomic_set(compound_pincount_ptr(page), 0);
689 21783 : }
690 :
691 : #ifdef CONFIG_DEBUG_PAGEALLOC
692 : unsigned int _debug_guardpage_minorder;
693 :
694 : bool _debug_pagealloc_enabled_early __read_mostly
695 : = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
696 : EXPORT_SYMBOL(_debug_pagealloc_enabled_early);
697 : DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
698 : EXPORT_SYMBOL(_debug_pagealloc_enabled);
699 :
700 : DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
701 :
702 : static int __init early_debug_pagealloc(char *buf)
703 : {
704 : return kstrtobool(buf, &_debug_pagealloc_enabled_early);
705 : }
706 : early_param("debug_pagealloc", early_debug_pagealloc);
707 :
708 : static int __init debug_guardpage_minorder_setup(char *buf)
709 : {
710 : unsigned long res;
711 :
712 : if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
713 : pr_err("Bad debug_guardpage_minorder value\n");
714 : return 0;
715 : }
716 : _debug_guardpage_minorder = res;
717 : pr_info("Setting debug_guardpage_minorder to %lu\n", res);
718 : return 0;
719 : }
720 : early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
721 :
722 : static inline bool set_page_guard(struct zone *zone, struct page *page,
723 : unsigned int order, int migratetype)
724 : {
725 : if (!debug_guardpage_enabled())
726 : return false;
727 :
728 : if (order >= debug_guardpage_minorder())
729 : return false;
730 :
731 : __SetPageGuard(page);
732 : INIT_LIST_HEAD(&page->lru);
733 : set_page_private(page, order);
734 : /* Guard pages are not available for any usage */
735 : __mod_zone_freepage_state(zone, -(1 << order), migratetype);
736 :
737 : return true;
738 : }
739 :
740 : static inline void clear_page_guard(struct zone *zone, struct page *page,
741 : unsigned int order, int migratetype)
742 : {
743 : if (!debug_guardpage_enabled())
744 : return;
745 :
746 : __ClearPageGuard(page);
747 :
748 : set_page_private(page, 0);
749 : if (!is_migrate_isolate(migratetype))
750 : __mod_zone_freepage_state(zone, (1 << order), migratetype);
751 : }
752 : #else
753 60148 : static inline bool set_page_guard(struct zone *zone, struct page *page,
754 60148 : unsigned int order, int migratetype) { return false; }
755 : static inline void clear_page_guard(struct zone *zone, struct page *page,
756 : unsigned int order, int migratetype) {}
757 : #endif
758 :
759 : /*
760 : * Enable static keys related to various memory debugging and hardening options.
761 : * Some override others, and depend on early params that are evaluated in the
762 : * order of appearance. So we need to first gather the full picture of what was
763 : * enabled, and then make decisions.
764 : */
765 1 : void init_mem_debugging_and_hardening(void)
766 : {
767 1 : if (_init_on_alloc_enabled_early) {
768 0 : if (page_poisoning_enabled())
769 : pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, "
770 : "will take precedence over init_on_alloc\n");
771 : else
772 0 : static_branch_enable(&init_on_alloc);
773 : }
774 1 : if (_init_on_free_enabled_early) {
775 0 : if (page_poisoning_enabled())
776 : pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, "
777 : "will take precedence over init_on_free\n");
778 : else
779 0 : static_branch_enable(&init_on_free);
780 : }
781 :
782 : #ifdef CONFIG_PAGE_POISONING
783 : /*
784 : * Page poisoning is debug page alloc for some arches. If
785 : * either of those options are enabled, enable poisoning.
786 : */
787 : if (page_poisoning_enabled() ||
788 : (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) &&
789 : debug_pagealloc_enabled()))
790 : static_branch_enable(&_page_poisoning_enabled);
791 : #endif
792 :
793 : #ifdef CONFIG_DEBUG_PAGEALLOC
794 : if (!debug_pagealloc_enabled())
795 : return;
796 :
797 : static_branch_enable(&_debug_pagealloc_enabled);
798 :
799 : if (!debug_guardpage_minorder())
800 : return;
801 :
802 : static_branch_enable(&_debug_guardpage_enabled);
803 : #endif
804 1 : }
805 :
806 97598 : static inline void set_buddy_order(struct page *page, unsigned int order)
807 : {
808 97598 : set_page_private(page, order);
809 97598 : __SetPageBuddy(page);
810 97598 : }
811 :
812 : /*
813 : * This function checks whether a page is free && is the buddy
814 : * we can coalesce a page and its buddy if
815 : * (a) the buddy is not in a hole (check before calling!) &&
816 : * (b) the buddy is in the buddy system &&
817 : * (c) a page and its buddy have the same order &&
818 : * (d) a page and its buddy are in the same zone.
819 : *
820 : * For recording whether a page is in the buddy system, we set PageBuddy.
821 : * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
822 : *
823 : * For recording page's order, we use page_private(page).
824 : */
825 88465 : static inline bool page_is_buddy(struct page *page, struct page *buddy,
826 : unsigned int order)
827 : {
828 88465 : if (!page_is_guard(buddy) && !PageBuddy(buddy))
829 : return false;
830 :
831 22768 : if (buddy_order(buddy) != order)
832 : return false;
833 :
834 : /*
835 : * zone check is done late to avoid uselessly calculating
836 : * zone/node ids for pages that could never merge.
837 : */
838 19774 : if (page_zone_id(page) != page_zone_id(buddy))
839 : return false;
840 :
841 19774 : VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
842 :
843 : return true;
844 : }
845 :
846 : #ifdef CONFIG_COMPACTION
847 37450 : static inline struct capture_control *task_capc(struct zone *zone)
848 : {
849 37450 : struct capture_control *capc = current->capture_control;
850 :
851 37450 : return unlikely(capc) &&
852 0 : !(current->flags & PF_KTHREAD) &&
853 0 : !capc->page &&
854 37450 : capc->cc->zone == zone ? capc : NULL;
855 : }
856 :
857 : static inline bool
858 51272 : compaction_capture(struct capture_control *capc, struct page *page,
859 : int order, int migratetype)
860 : {
861 0 : if (!capc || order != capc->cc->order)
862 : return false;
863 :
864 : /* Do not accidentally pollute CMA or isolated regions*/
865 0 : if (is_migrate_cma(migratetype) ||
866 0 : is_migrate_isolate(migratetype))
867 : return false;
868 :
869 : /*
870 : * Do not let lower order allocations polluate a movable pageblock.
871 : * This might let an unmovable request use a reclaimable pageblock
872 : * and vice-versa but no more than normal fallback logic which can
873 : * have trouble finding a high-order free page.
874 : */
875 0 : if (order < pageblock_order && migratetype == MIGRATE_MOVABLE)
876 : return false;
877 :
878 0 : capc->page = page;
879 0 : return true;
880 : }
881 :
882 : #else
883 : static inline struct capture_control *task_capc(struct zone *zone)
884 : {
885 : return NULL;
886 : }
887 :
888 : static inline bool
889 : compaction_capture(struct capture_control *capc, struct page *page,
890 : int order, int migratetype)
891 : {
892 : return false;
893 : }
894 : #endif /* CONFIG_COMPACTION */
895 :
896 : /* Used for pages not on another list */
897 91611 : static inline void add_to_free_list(struct page *page, struct zone *zone,
898 : unsigned int order, int migratetype)
899 : {
900 91611 : struct free_area *area = &zone->free_area[order];
901 :
902 91611 : list_add(&page->lru, &area->free_list[migratetype]);
903 91611 : area->nr_free++;
904 31463 : }
905 :
906 : /* Used for pages not on another list */
907 5987 : static inline void add_to_free_list_tail(struct page *page, struct zone *zone,
908 : unsigned int order, int migratetype)
909 : {
910 5987 : struct free_area *area = &zone->free_area[order];
911 :
912 5987 : list_add_tail(&page->lru, &area->free_list[migratetype]);
913 5987 : area->nr_free++;
914 5987 : }
915 :
916 : /*
917 : * Used for pages which are on another list. Move the pages to the tail
918 : * of the list - so the moved pages won't immediately be considered for
919 : * allocation again (e.g., optimization for memory onlining).
920 : */
921 38 : static inline void move_to_free_list(struct page *page, struct zone *zone,
922 : unsigned int order, int migratetype)
923 : {
924 38 : struct free_area *area = &zone->free_area[order];
925 :
926 38 : list_move_tail(&page->lru, &area->free_list[migratetype]);
927 : }
928 :
929 97180 : static inline void del_page_from_free_list(struct page *page, struct zone *zone,
930 : unsigned int order)
931 : {
932 : /* clear reported state and update reported page count */
933 97180 : if (page_reported(page))
934 0 : __ClearPageReported(page);
935 :
936 97180 : list_del(&page->lru);
937 97180 : __ClearPageBuddy(page);
938 97180 : set_page_private(page, 0);
939 97180 : zone->free_area[order].nr_free--;
940 97180 : }
941 :
942 : /*
943 : * If this is not the largest possible page, check if the buddy
944 : * of the next-highest order is free. If it is, it's possible
945 : * that pages are being freed that will coalesce soon. In case,
946 : * that is happening, add the free page to the tail of the list
947 : * so it's less likely to be used soon and more likely to be merged
948 : * as a higher order page
949 : */
950 : static inline bool
951 37217 : buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn,
952 : struct page *page, unsigned int order)
953 : {
954 37217 : struct page *higher_page, *higher_buddy;
955 37217 : unsigned long combined_pfn;
956 :
957 37217 : if (order >= MAX_ORDER - 2)
958 : return false;
959 :
960 37193 : if (!pfn_valid_within(buddy_pfn))
961 : return false;
962 :
963 37193 : combined_pfn = buddy_pfn & pfn;
964 37193 : higher_page = page + (combined_pfn - pfn);
965 37193 : buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
966 37193 : higher_buddy = higher_page + (buddy_pfn - combined_pfn);
967 :
968 37193 : return pfn_valid_within(buddy_pfn) &&
969 37193 : page_is_buddy(higher_page, higher_buddy, order + 1);
970 : }
971 :
972 : /*
973 : * Freeing function for a buddy system allocator.
974 : *
975 : * The concept of a buddy system is to maintain direct-mapped table
976 : * (containing bit values) for memory blocks of various "orders".
977 : * The bottom level table contains the map for the smallest allocatable
978 : * units of memory (here, pages), and each level above it describes
979 : * pairs of units from the levels below, hence, "buddies".
980 : * At a high level, all that happens here is marking the table entry
981 : * at the bottom level available, and propagating the changes upward
982 : * as necessary, plus some accounting needed to play nicely with other
983 : * parts of the VM system.
984 : * At each level, we keep a list of pages, which are heads of continuous
985 : * free pages of length of (1 << order) and marked with PageBuddy.
986 : * Page's order is recorded in page_private(page) field.
987 : * So when we are allocating or freeing one, we can derive the state of the
988 : * other. That is, if we allocate a small block, and both were
989 : * free, the remainder of the region must be split into blocks.
990 : * If a block is freed, and its buddy is also free, then this
991 : * triggers coalescing into a block of larger size.
992 : *
993 : * -- nyc
994 : */
995 :
996 37450 : static inline void __free_one_page(struct page *page,
997 : unsigned long pfn,
998 : struct zone *zone, unsigned int order,
999 : int migratetype, fpi_t fpi_flags)
1000 : {
1001 37450 : struct capture_control *capc = task_capc(zone);
1002 37450 : unsigned long buddy_pfn;
1003 37450 : unsigned long combined_pfn;
1004 37450 : unsigned int max_order;
1005 37450 : struct page *buddy;
1006 37450 : bool to_tail;
1007 :
1008 37450 : max_order = min_t(unsigned int, MAX_ORDER - 1, pageblock_order);
1009 :
1010 37450 : VM_BUG_ON(!zone_is_initialized(zone));
1011 37450 : VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
1012 :
1013 37450 : VM_BUG_ON(migratetype == -1);
1014 37450 : if (likely(!is_migrate_isolate(migratetype)))
1015 37450 : __mod_zone_freepage_state(zone, 1 << order, migratetype);
1016 :
1017 37450 : VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
1018 37450 : VM_BUG_ON_PAGE(bad_range(zone, page), page);
1019 :
1020 37450 : continue_merging:
1021 51470 : while (order < max_order) {
1022 51272 : if (compaction_capture(capc, page, order, migratetype)) {
1023 0 : __mod_zone_freepage_state(zone, -(1 << order),
1024 : migratetype);
1025 0 : return;
1026 : }
1027 51272 : buddy_pfn = __find_buddy_pfn(pfn, order);
1028 51272 : buddy = page + (buddy_pfn - pfn);
1029 :
1030 51272 : if (!pfn_valid_within(buddy_pfn))
1031 : goto done_merging;
1032 51272 : if (!page_is_buddy(page, buddy, order))
1033 37252 : goto done_merging;
1034 : /*
1035 : * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
1036 : * merge with it and move up one order.
1037 : */
1038 14020 : if (page_is_guard(buddy))
1039 14020 : clear_page_guard(zone, buddy, order, migratetype);
1040 : else
1041 14020 : del_page_from_free_list(buddy, zone, order);
1042 14020 : combined_pfn = buddy_pfn & pfn;
1043 14020 : page = page + (combined_pfn - pfn);
1044 14020 : pfn = combined_pfn;
1045 14020 : order++;
1046 : }
1047 198 : if (order < MAX_ORDER - 1) {
1048 : /* If we are here, it means order is >= pageblock_order.
1049 : * We want to prevent merge between freepages on isolate
1050 : * pageblock and normal pageblock. Without this, pageblock
1051 : * isolation could cause incorrect freepage or CMA accounting.
1052 : *
1053 : * We don't want to hit this code for the more frequent
1054 : * low-order merging.
1055 : */
1056 0 : if (unlikely(has_isolate_pageblock(zone))) {
1057 : int buddy_mt;
1058 :
1059 : buddy_pfn = __find_buddy_pfn(pfn, order);
1060 : buddy = page + (buddy_pfn - pfn);
1061 : buddy_mt = get_pageblock_migratetype(buddy);
1062 :
1063 : if (migratetype != buddy_mt
1064 : && (is_migrate_isolate(migratetype) ||
1065 : is_migrate_isolate(buddy_mt)))
1066 : goto done_merging;
1067 : }
1068 0 : max_order = order + 1;
1069 0 : goto continue_merging;
1070 : }
1071 :
1072 198 : done_merging:
1073 37450 : set_buddy_order(page, order);
1074 :
1075 37450 : if (fpi_flags & FPI_TO_TAIL)
1076 : to_tail = true;
1077 37217 : else if (is_shuffle_order(order))
1078 : to_tail = shuffle_pick_tail();
1079 : else
1080 37217 : to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order);
1081 :
1082 37217 : if (to_tail)
1083 5987 : add_to_free_list_tail(page, zone, order, migratetype);
1084 : else
1085 31463 : add_to_free_list(page, zone, order, migratetype);
1086 :
1087 : /* Notify page reporting subsystem of freed page */
1088 37450 : if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY))
1089 37450 : page_reporting_notify_free(order);
1090 : }
1091 :
1092 : /*
1093 : * A bad page could be due to a number of fields. Instead of multiple branches,
1094 : * try and check multiple fields with one check. The caller must do a detailed
1095 : * check if necessary.
1096 : */
1097 719859 : static inline bool page_expected_state(struct page *page,
1098 : unsigned long check_flags)
1099 : {
1100 719859 : if (unlikely(atomic_read(&page->_mapcount) != -1))
1101 : return false;
1102 :
1103 719857 : if (unlikely((unsigned long)page->mapping |
1104 : page_ref_count(page) |
1105 : #ifdef CONFIG_MEMCG
1106 : (unsigned long)page_memcg(page) |
1107 : #endif
1108 : (page->flags & check_flags)))
1109 0 : return false;
1110 :
1111 : return true;
1112 : }
1113 :
1114 0 : static const char *page_bad_reason(struct page *page, unsigned long flags)
1115 : {
1116 0 : const char *bad_reason = NULL;
1117 :
1118 0 : if (unlikely(atomic_read(&page->_mapcount) != -1))
1119 0 : bad_reason = "nonzero mapcount";
1120 0 : if (unlikely(page->mapping != NULL))
1121 0 : bad_reason = "non-NULL mapping";
1122 0 : if (unlikely(page_ref_count(page) != 0))
1123 0 : bad_reason = "nonzero _refcount";
1124 0 : if (unlikely(page->flags & flags)) {
1125 0 : if (flags == PAGE_FLAGS_CHECK_AT_PREP)
1126 : bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set";
1127 : else
1128 0 : bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
1129 : }
1130 : #ifdef CONFIG_MEMCG
1131 : if (unlikely(page_memcg(page)))
1132 : bad_reason = "page still charged to cgroup";
1133 : #endif
1134 0 : return bad_reason;
1135 : }
1136 :
1137 0 : static void check_free_page_bad(struct page *page)
1138 : {
1139 0 : bad_page(page,
1140 : page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE));
1141 0 : }
1142 :
1143 428426 : static inline int check_free_page(struct page *page)
1144 : {
1145 428426 : if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
1146 : return 0;
1147 :
1148 : /* Something has gone sideways, find it */
1149 0 : check_free_page_bad(page);
1150 0 : return 1;
1151 : }
1152 :
1153 68885 : static int free_tail_pages_check(struct page *head_page, struct page *page)
1154 : {
1155 68885 : int ret = 1;
1156 :
1157 : /*
1158 : * We rely page->lru.next never has bit 0 set, unless the page
1159 : * is PageTail(). Let's make sure that's true even for poisoned ->lru.
1160 : */
1161 68885 : BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
1162 :
1163 68885 : if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
1164 : ret = 0;
1165 : goto out;
1166 : }
1167 68885 : switch (page - head_page) {
1168 13475 : case 1:
1169 : /* the first tail page: ->mapping may be compound_mapcount() */
1170 13475 : if (unlikely(compound_mapcount(page))) {
1171 0 : bad_page(page, "nonzero compound_mapcount");
1172 0 : goto out;
1173 : }
1174 : break;
1175 : case 2:
1176 : /*
1177 : * the second tail page: ->mapping is
1178 : * deferred_list.next -- ignore value.
1179 : */
1180 : break;
1181 46729 : default:
1182 46729 : if (page->mapping != TAIL_MAPPING) {
1183 0 : bad_page(page, "corrupted mapping in tail page");
1184 0 : goto out;
1185 : }
1186 : break;
1187 : }
1188 68885 : if (unlikely(!PageTail(page))) {
1189 0 : bad_page(page, "PageTail not set");
1190 0 : goto out;
1191 : }
1192 137770 : if (unlikely(compound_head(page) != head_page)) {
1193 0 : bad_page(page, "compound_head not consistent");
1194 0 : goto out;
1195 : }
1196 : ret = 0;
1197 68885 : out:
1198 68885 : page->mapping = NULL;
1199 68885 : clear_compound_head(page);
1200 68885 : return ret;
1201 : }
1202 :
1203 73017 : static void kernel_init_free_pages(struct page *page, int numpages)
1204 : {
1205 73017 : int i;
1206 :
1207 : /* s390's use of memset() could override KASAN redzones. */
1208 73017 : kasan_disable_current();
1209 230719 : for (i = 0; i < numpages; i++) {
1210 84685 : u8 tag = page_kasan_tag(page + i);
1211 84685 : page_kasan_tag_reset(page + i);
1212 84685 : clear_highpage(page + i);
1213 84685 : page_kasan_tag_set(page + i, tag);
1214 : }
1215 73017 : kasan_enable_current();
1216 73017 : }
1217 :
1218 148328 : static __always_inline bool free_pages_prepare(struct page *page,
1219 : unsigned int order, bool check_free)
1220 : {
1221 148328 : int bad = 0;
1222 :
1223 0 : VM_BUG_ON_PAGE(PageTail(page), page);
1224 :
1225 148328 : trace_mm_page_free(page, order);
1226 :
1227 148341 : if (unlikely(PageHWPoison(page)) && !order) {
1228 : /*
1229 : * Do not let hwpoison pages hit pcplists/buddy
1230 : * Untie memcg state and reset page's owner
1231 : */
1232 : if (memcg_kmem_enabled() && PageMemcgKmem(page))
1233 : __memcg_kmem_uncharge_page(page, order);
1234 : reset_page_owner(page, order);
1235 : return false;
1236 : }
1237 :
1238 : /*
1239 : * Check tail pages before head page information is cleared to
1240 : * avoid checking PageCompound for order-0 pages.
1241 : */
1242 14959 : if (unlikely(order)) {
1243 14957 : bool compound = PageCompound(page);
1244 14957 : int i;
1245 :
1246 14957 : VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1247 :
1248 14957 : if (compound)
1249 13475 : ClearPageDoubleMap(page);
1250 295050 : for (i = 1; i < (1 << order); i++) {
1251 280093 : if (compound)
1252 68885 : bad += free_tail_pages_check(page, page + i);
1253 280093 : if (unlikely(check_free_page(page + i))) {
1254 0 : bad++;
1255 0 : continue;
1256 : }
1257 280093 : (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1258 : }
1259 : }
1260 148341 : if (PageMappingFlags(page))
1261 65808 : page->mapping = NULL;
1262 148341 : if (memcg_kmem_enabled() && PageMemcgKmem(page))
1263 : __memcg_kmem_uncharge_page(page, order);
1264 148341 : if (check_free)
1265 148341 : bad += check_free_page(page);
1266 148340 : if (bad)
1267 : return false;
1268 :
1269 148340 : page_cpupid_reset_last(page);
1270 148340 : page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1271 148340 : reset_page_owner(page, order);
1272 :
1273 148340 : if (!PageHighMem(page)) {
1274 148340 : debug_check_no_locks_freed(page_address(page),
1275 : PAGE_SIZE << order);
1276 148343 : debug_check_no_obj_freed(page_address(page),
1277 : PAGE_SIZE << order);
1278 : }
1279 148329 : if (want_init_on_free())
1280 0 : kernel_init_free_pages(page, 1 << order);
1281 :
1282 148329 : kernel_poison_pages(page, 1 << order);
1283 :
1284 : /*
1285 : * With hardware tag-based KASAN, memory tags must be set before the
1286 : * page becomes unavailable via debug_pagealloc or arch_free_page.
1287 : */
1288 148329 : kasan_free_nondeferred_pages(page, order);
1289 :
1290 : /*
1291 : * arch_free_page() can make the page's contents inaccessible. s390
1292 : * does this. So nothing which can access the page's contents should
1293 : * happen after this.
1294 : */
1295 148329 : arch_free_page(page, order);
1296 :
1297 148329 : debug_pagealloc_unmap_pages(page, 1 << order);
1298 :
1299 133370 : return true;
1300 : }
1301 :
1302 : #ifdef CONFIG_DEBUG_VM
1303 : /*
1304 : * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed
1305 : * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when
1306 : * moved from pcp lists to free lists.
1307 : */
1308 133369 : static bool free_pcp_prepare(struct page *page)
1309 : {
1310 133369 : return free_pages_prepare(page, 0, true);
1311 : }
1312 :
1313 22491 : static bool bulkfree_pcp_prepare(struct page *page)
1314 : {
1315 22491 : if (debug_pagealloc_enabled_static())
1316 : return check_free_page(page);
1317 : else
1318 22491 : return false;
1319 : }
1320 : #else
1321 : /*
1322 : * With DEBUG_VM disabled, order-0 pages being freed are checked only when
1323 : * moving from pcp lists to free list in order to reduce overhead. With
1324 : * debug_pagealloc enabled, they are checked also immediately when being freed
1325 : * to the pcp lists.
1326 : */
1327 : static bool free_pcp_prepare(struct page *page)
1328 : {
1329 : if (debug_pagealloc_enabled_static())
1330 : return free_pages_prepare(page, 0, true);
1331 : else
1332 : return free_pages_prepare(page, 0, false);
1333 : }
1334 :
1335 : static bool bulkfree_pcp_prepare(struct page *page)
1336 : {
1337 : return check_free_page(page);
1338 : }
1339 : #endif /* CONFIG_DEBUG_VM */
1340 :
1341 22491 : static inline void prefetch_buddy(struct page *page)
1342 : {
1343 22491 : unsigned long pfn = page_to_pfn(page);
1344 22491 : unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1345 22491 : struct page *buddy = page + (buddy_pfn - pfn);
1346 :
1347 22491 : prefetch(buddy);
1348 22491 : }
1349 :
1350 : /*
1351 : * Frees a number of pages from the PCP lists
1352 : * Assumes all pages on list are in same zone, and of same order.
1353 : * count is the number of pages to free.
1354 : *
1355 : * If the zone was previously in an "all pages pinned" state then look to
1356 : * see if this freeing clears that state.
1357 : *
1358 : * And clear the zone's pages_scanned counter, to hold off the "all pages are
1359 : * pinned" detection logic.
1360 : */
1361 357 : static void free_pcppages_bulk(struct zone *zone, int count,
1362 : struct per_cpu_pages *pcp)
1363 : {
1364 357 : int migratetype = 0;
1365 357 : int batch_free = 0;
1366 357 : int prefetch_nr = READ_ONCE(pcp->batch);
1367 357 : bool isolated_pageblocks;
1368 357 : struct page *page, *tmp;
1369 357 : LIST_HEAD(head);
1370 :
1371 : /*
1372 : * Ensure proper count is passed which otherwise would stuck in the
1373 : * below while (list_empty(list)) loop.
1374 : */
1375 357 : count = min(pcp->count, count);
1376 14201 : while (count) {
1377 17133 : struct list_head *list;
1378 :
1379 : /*
1380 : * Remove pages from lists in a round-robin fashion. A
1381 : * batch_free count is maintained that is incremented when an
1382 : * empty list is encountered. This is so more pages are freed
1383 : * off fuller lists instead of spinning excessively around empty
1384 : * lists
1385 : */
1386 17133 : do {
1387 17133 : batch_free++;
1388 17133 : if (++migratetype == MIGRATE_PCPTYPES)
1389 5672 : migratetype = 0;
1390 17133 : list = &pcp->lists[migratetype];
1391 17133 : } while (list_empty(list));
1392 :
1393 : /* This is the only non-empty list. Free them all. */
1394 13844 : if (batch_free == MIGRATE_PCPTYPES)
1395 114 : batch_free = count;
1396 :
1397 22491 : do {
1398 22491 : page = list_last_entry(list, struct page, lru);
1399 : /* must delete to avoid corrupting pcp list */
1400 22491 : list_del(&page->lru);
1401 22491 : pcp->count--;
1402 :
1403 22491 : if (bulkfree_pcp_prepare(page))
1404 : continue;
1405 :
1406 22491 : list_add_tail(&page->lru, &head);
1407 :
1408 : /*
1409 : * We are going to put the page back to the global
1410 : * pool, prefetch its buddy to speed up later access
1411 : * under zone->lock. It is believed the overhead of
1412 : * an additional test and calculating buddy_pfn here
1413 : * can be offset by reduced memory latency later. To
1414 : * avoid excessive prefetching due to large count, only
1415 : * prefetch buddy for the first pcp->batch nr of pages.
1416 : */
1417 22491 : if (prefetch_nr) {
1418 22491 : prefetch_buddy(page);
1419 22491 : prefetch_nr--;
1420 : }
1421 22491 : } while (--count && --batch_free && !list_empty(list));
1422 : }
1423 :
1424 357 : spin_lock(&zone->lock);
1425 357 : isolated_pageblocks = has_isolate_pageblock(zone);
1426 :
1427 : /*
1428 : * Use safe version since after __free_one_page(),
1429 : * page->lru.next will not point to original list.
1430 : */
1431 22848 : list_for_each_entry_safe(page, tmp, &head, lru) {
1432 22491 : int mt = get_pcppage_migratetype(page);
1433 : /* MIGRATE_ISOLATE page should not go to pcplists */
1434 22491 : VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1435 : /* Pageblock could have been isolated meanwhile */
1436 22491 : if (unlikely(isolated_pageblocks))
1437 : mt = get_pageblock_migratetype(page);
1438 :
1439 22491 : __free_one_page(page, page_to_pfn(page), zone, 0, mt, FPI_NONE);
1440 22491 : trace_mm_page_pcpu_drain(page, 0, mt);
1441 : }
1442 357 : spin_unlock(&zone->lock);
1443 357 : }
1444 :
1445 14959 : static void free_one_page(struct zone *zone,
1446 : struct page *page, unsigned long pfn,
1447 : unsigned int order,
1448 : int migratetype, fpi_t fpi_flags)
1449 : {
1450 14959 : spin_lock(&zone->lock);
1451 14959 : if (unlikely(has_isolate_pageblock(zone) ||
1452 : is_migrate_isolate(migratetype))) {
1453 : migratetype = get_pfnblock_migratetype(page, pfn);
1454 : }
1455 14959 : __free_one_page(page, pfn, zone, order, migratetype, fpi_flags);
1456 14959 : spin_unlock(&zone->lock);
1457 14959 : }
1458 :
1459 262144 : static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1460 : unsigned long zone, int nid)
1461 : {
1462 262144 : mm_zero_struct_page(page);
1463 262144 : set_page_links(page, zone, nid, pfn);
1464 262144 : init_page_count(page);
1465 262144 : page_mapcount_reset(page);
1466 262144 : page_cpupid_reset_last(page);
1467 262144 : page_kasan_tag_reset(page);
1468 :
1469 262144 : INIT_LIST_HEAD(&page->lru);
1470 : #ifdef WANT_PAGE_VIRTUAL
1471 : /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1472 : if (!is_highmem_idx(zone))
1473 : set_page_address(page, __va(pfn << PAGE_SHIFT));
1474 : #endif
1475 262144 : }
1476 :
1477 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1478 : static void __meminit init_reserved_page(unsigned long pfn)
1479 : {
1480 : pg_data_t *pgdat;
1481 : int nid, zid;
1482 :
1483 : if (!early_page_uninitialised(pfn))
1484 : return;
1485 :
1486 : nid = early_pfn_to_nid(pfn);
1487 : pgdat = NODE_DATA(nid);
1488 :
1489 : for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1490 : struct zone *zone = &pgdat->node_zones[zid];
1491 :
1492 : if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1493 : break;
1494 : }
1495 : __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1496 : }
1497 : #else
1498 59461 : static inline void init_reserved_page(unsigned long pfn)
1499 : {
1500 59461 : }
1501 : #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1502 :
1503 : /*
1504 : * Initialised pages do not have PageReserved set. This function is
1505 : * called for each range allocated by the bootmem allocator and
1506 : * marks the pages PageReserved. The remaining valid pages are later
1507 : * sent to the buddy page allocator.
1508 : */
1509 20 : void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1510 : {
1511 20 : unsigned long start_pfn = PFN_DOWN(start);
1512 20 : unsigned long end_pfn = PFN_UP(end);
1513 :
1514 59481 : for (; start_pfn < end_pfn; start_pfn++) {
1515 59461 : if (pfn_valid(start_pfn)) {
1516 59461 : struct page *page = pfn_to_page(start_pfn);
1517 :
1518 59461 : init_reserved_page(start_pfn);
1519 :
1520 : /* Avoid false-positive PageTail() */
1521 59461 : INIT_LIST_HEAD(&page->lru);
1522 :
1523 : /*
1524 : * no need for atomic set_bit because the struct
1525 : * page is not visible yet so nobody should
1526 : * access it yet.
1527 : */
1528 118922 : __SetPageReserved(page);
1529 : }
1530 : }
1531 20 : }
1532 :
1533 14959 : static void __free_pages_ok(struct page *page, unsigned int order,
1534 : fpi_t fpi_flags)
1535 : {
1536 14959 : unsigned long flags;
1537 14959 : int migratetype;
1538 14959 : unsigned long pfn = page_to_pfn(page);
1539 :
1540 14959 : if (!free_pages_prepare(page, order, true))
1541 : return;
1542 :
1543 14959 : migratetype = get_pfnblock_migratetype(page, pfn);
1544 29918 : local_irq_save(flags);
1545 14959 : __count_vm_events(PGFREE, 1 << order);
1546 14959 : free_one_page(page_zone(page), page, pfn, order, migratetype,
1547 : fpi_flags);
1548 14959 : local_irq_restore(flags);
1549 : }
1550 :
1551 233 : void __free_pages_core(struct page *page, unsigned int order)
1552 : {
1553 233 : unsigned int nr_pages = 1 << order;
1554 233 : struct page *p = page;
1555 233 : unsigned int loop;
1556 :
1557 : /*
1558 : * When initializing the memmap, __init_single_page() sets the refcount
1559 : * of all pages to 1 ("allocated"/"not free"). We have to set the
1560 : * refcount of all involved pages to 0.
1561 : */
1562 233 : prefetchw(p);
1563 202925 : for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1564 202459 : prefetchw(p + 1);
1565 202459 : __ClearPageReserved(p);
1566 202459 : set_page_count(p, 0);
1567 : }
1568 233 : __ClearPageReserved(p);
1569 233 : set_page_count(p, 0);
1570 :
1571 233 : atomic_long_add(nr_pages, &page_zone(page)->managed_pages);
1572 :
1573 : /*
1574 : * Bypass PCP and place fresh pages right to the tail, primarily
1575 : * relevant for memory onlining.
1576 : */
1577 233 : __free_pages_ok(page, order, FPI_TO_TAIL);
1578 233 : }
1579 :
1580 : #ifdef CONFIG_NEED_MULTIPLE_NODES
1581 :
1582 : /*
1583 : * During memory init memblocks map pfns to nids. The search is expensive and
1584 : * this caches recent lookups. The implementation of __early_pfn_to_nid
1585 : * treats start/end as pfns.
1586 : */
1587 : struct mminit_pfnnid_cache {
1588 : unsigned long last_start;
1589 : unsigned long last_end;
1590 : int last_nid;
1591 : };
1592 :
1593 : static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1594 :
1595 : /*
1596 : * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
1597 : */
1598 3 : static int __meminit __early_pfn_to_nid(unsigned long pfn,
1599 : struct mminit_pfnnid_cache *state)
1600 : {
1601 3 : unsigned long start_pfn, end_pfn;
1602 3 : int nid;
1603 :
1604 3 : if (state->last_start <= pfn && pfn < state->last_end)
1605 0 : return state->last_nid;
1606 :
1607 3 : nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
1608 3 : if (nid != NUMA_NO_NODE) {
1609 1 : state->last_start = start_pfn;
1610 1 : state->last_end = end_pfn;
1611 1 : state->last_nid = nid;
1612 : }
1613 :
1614 : return nid;
1615 : }
1616 :
1617 3 : int __meminit early_pfn_to_nid(unsigned long pfn)
1618 : {
1619 3 : static DEFINE_SPINLOCK(early_pfn_lock);
1620 3 : int nid;
1621 :
1622 3 : spin_lock(&early_pfn_lock);
1623 3 : nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1624 3 : if (nid < 0)
1625 2 : nid = first_online_node;
1626 3 : spin_unlock(&early_pfn_lock);
1627 :
1628 3 : return nid;
1629 : }
1630 : #endif /* CONFIG_NEED_MULTIPLE_NODES */
1631 :
1632 233 : void __init memblock_free_pages(struct page *page, unsigned long pfn,
1633 : unsigned int order)
1634 : {
1635 233 : if (early_page_uninitialised(pfn))
1636 : return;
1637 233 : __free_pages_core(page, order);
1638 : }
1639 :
1640 : /*
1641 : * Check that the whole (or subset of) a pageblock given by the interval of
1642 : * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1643 : * with the migration of free compaction scanner. The scanners then need to
1644 : * use only pfn_valid_within() check for arches that allow holes within
1645 : * pageblocks.
1646 : *
1647 : * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1648 : *
1649 : * It's possible on some configurations to have a setup like node0 node1 node0
1650 : * i.e. it's possible that all pages within a zones range of pages do not
1651 : * belong to a single zone. We assume that a border between node0 and node1
1652 : * can occur within a single pageblock, but not a node0 node1 node0
1653 : * interleaving within a single pageblock. It is therefore sufficient to check
1654 : * the first and last page of a pageblock and avoid checking each individual
1655 : * page in a pageblock.
1656 : */
1657 256 : struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1658 : unsigned long end_pfn, struct zone *zone)
1659 : {
1660 256 : struct page *start_page;
1661 256 : struct page *end_page;
1662 :
1663 : /* end_pfn is one past the range we are checking */
1664 256 : end_pfn--;
1665 :
1666 256 : if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1667 : return NULL;
1668 :
1669 256 : start_page = pfn_to_online_page(start_pfn);
1670 256 : if (!start_page)
1671 : return NULL;
1672 :
1673 256 : if (page_zone(start_page) != zone)
1674 : return NULL;
1675 :
1676 256 : end_page = pfn_to_page(end_pfn);
1677 :
1678 : /* This gives a shorter code than deriving page_zone(end_page) */
1679 256 : if (page_zone_id(start_page) != page_zone_id(end_page))
1680 0 : return NULL;
1681 :
1682 : return start_page;
1683 : }
1684 :
1685 1 : void set_zone_contiguous(struct zone *zone)
1686 : {
1687 1 : unsigned long block_start_pfn = zone->zone_start_pfn;
1688 1 : unsigned long block_end_pfn;
1689 :
1690 1 : block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1691 257 : for (; block_start_pfn < zone_end_pfn(zone);
1692 256 : block_start_pfn = block_end_pfn,
1693 256 : block_end_pfn += pageblock_nr_pages) {
1694 :
1695 256 : block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1696 :
1697 256 : if (!__pageblock_pfn_to_page(block_start_pfn,
1698 : block_end_pfn, zone))
1699 : return;
1700 256 : cond_resched();
1701 : }
1702 :
1703 : /* We confirm that there is no hole */
1704 1 : zone->contiguous = true;
1705 : }
1706 :
1707 0 : void clear_zone_contiguous(struct zone *zone)
1708 : {
1709 0 : zone->contiguous = false;
1710 0 : }
1711 :
1712 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1713 : static void __init deferred_free_range(unsigned long pfn,
1714 : unsigned long nr_pages)
1715 : {
1716 : struct page *page;
1717 : unsigned long i;
1718 :
1719 : if (!nr_pages)
1720 : return;
1721 :
1722 : page = pfn_to_page(pfn);
1723 :
1724 : /* Free a large naturally-aligned chunk if possible */
1725 : if (nr_pages == pageblock_nr_pages &&
1726 : (pfn & (pageblock_nr_pages - 1)) == 0) {
1727 : set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1728 : __free_pages_core(page, pageblock_order);
1729 : return;
1730 : }
1731 :
1732 : for (i = 0; i < nr_pages; i++, page++, pfn++) {
1733 : if ((pfn & (pageblock_nr_pages - 1)) == 0)
1734 : set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1735 : __free_pages_core(page, 0);
1736 : }
1737 : }
1738 :
1739 : /* Completion tracking for deferred_init_memmap() threads */
1740 : static atomic_t pgdat_init_n_undone __initdata;
1741 : static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1742 :
1743 : static inline void __init pgdat_init_report_one_done(void)
1744 : {
1745 : if (atomic_dec_and_test(&pgdat_init_n_undone))
1746 : complete(&pgdat_init_all_done_comp);
1747 : }
1748 :
1749 : /*
1750 : * Returns true if page needs to be initialized or freed to buddy allocator.
1751 : *
1752 : * First we check if pfn is valid on architectures where it is possible to have
1753 : * holes within pageblock_nr_pages. On systems where it is not possible, this
1754 : * function is optimized out.
1755 : *
1756 : * Then, we check if a current large page is valid by only checking the validity
1757 : * of the head pfn.
1758 : */
1759 : static inline bool __init deferred_pfn_valid(unsigned long pfn)
1760 : {
1761 : if (!pfn_valid_within(pfn))
1762 : return false;
1763 : if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1764 : return false;
1765 : return true;
1766 : }
1767 :
1768 : /*
1769 : * Free pages to buddy allocator. Try to free aligned pages in
1770 : * pageblock_nr_pages sizes.
1771 : */
1772 : static void __init deferred_free_pages(unsigned long pfn,
1773 : unsigned long end_pfn)
1774 : {
1775 : unsigned long nr_pgmask = pageblock_nr_pages - 1;
1776 : unsigned long nr_free = 0;
1777 :
1778 : for (; pfn < end_pfn; pfn++) {
1779 : if (!deferred_pfn_valid(pfn)) {
1780 : deferred_free_range(pfn - nr_free, nr_free);
1781 : nr_free = 0;
1782 : } else if (!(pfn & nr_pgmask)) {
1783 : deferred_free_range(pfn - nr_free, nr_free);
1784 : nr_free = 1;
1785 : } else {
1786 : nr_free++;
1787 : }
1788 : }
1789 : /* Free the last block of pages to allocator */
1790 : deferred_free_range(pfn - nr_free, nr_free);
1791 : }
1792 :
1793 : /*
1794 : * Initialize struct pages. We minimize pfn page lookups and scheduler checks
1795 : * by performing it only once every pageblock_nr_pages.
1796 : * Return number of pages initialized.
1797 : */
1798 : static unsigned long __init deferred_init_pages(struct zone *zone,
1799 : unsigned long pfn,
1800 : unsigned long end_pfn)
1801 : {
1802 : unsigned long nr_pgmask = pageblock_nr_pages - 1;
1803 : int nid = zone_to_nid(zone);
1804 : unsigned long nr_pages = 0;
1805 : int zid = zone_idx(zone);
1806 : struct page *page = NULL;
1807 :
1808 : for (; pfn < end_pfn; pfn++) {
1809 : if (!deferred_pfn_valid(pfn)) {
1810 : page = NULL;
1811 : continue;
1812 : } else if (!page || !(pfn & nr_pgmask)) {
1813 : page = pfn_to_page(pfn);
1814 : } else {
1815 : page++;
1816 : }
1817 : __init_single_page(page, pfn, zid, nid);
1818 : nr_pages++;
1819 : }
1820 : return (nr_pages);
1821 : }
1822 :
1823 : /*
1824 : * This function is meant to pre-load the iterator for the zone init.
1825 : * Specifically it walks through the ranges until we are caught up to the
1826 : * first_init_pfn value and exits there. If we never encounter the value we
1827 : * return false indicating there are no valid ranges left.
1828 : */
1829 : static bool __init
1830 : deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
1831 : unsigned long *spfn, unsigned long *epfn,
1832 : unsigned long first_init_pfn)
1833 : {
1834 : u64 j;
1835 :
1836 : /*
1837 : * Start out by walking through the ranges in this zone that have
1838 : * already been initialized. We don't need to do anything with them
1839 : * so we just need to flush them out of the system.
1840 : */
1841 : for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
1842 : if (*epfn <= first_init_pfn)
1843 : continue;
1844 : if (*spfn < first_init_pfn)
1845 : *spfn = first_init_pfn;
1846 : *i = j;
1847 : return true;
1848 : }
1849 :
1850 : return false;
1851 : }
1852 :
1853 : /*
1854 : * Initialize and free pages. We do it in two loops: first we initialize
1855 : * struct page, then free to buddy allocator, because while we are
1856 : * freeing pages we can access pages that are ahead (computing buddy
1857 : * page in __free_one_page()).
1858 : *
1859 : * In order to try and keep some memory in the cache we have the loop
1860 : * broken along max page order boundaries. This way we will not cause
1861 : * any issues with the buddy page computation.
1862 : */
1863 : static unsigned long __init
1864 : deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
1865 : unsigned long *end_pfn)
1866 : {
1867 : unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
1868 : unsigned long spfn = *start_pfn, epfn = *end_pfn;
1869 : unsigned long nr_pages = 0;
1870 : u64 j = *i;
1871 :
1872 : /* First we loop through and initialize the page values */
1873 : for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
1874 : unsigned long t;
1875 :
1876 : if (mo_pfn <= *start_pfn)
1877 : break;
1878 :
1879 : t = min(mo_pfn, *end_pfn);
1880 : nr_pages += deferred_init_pages(zone, *start_pfn, t);
1881 :
1882 : if (mo_pfn < *end_pfn) {
1883 : *start_pfn = mo_pfn;
1884 : break;
1885 : }
1886 : }
1887 :
1888 : /* Reset values and now loop through freeing pages as needed */
1889 : swap(j, *i);
1890 :
1891 : for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
1892 : unsigned long t;
1893 :
1894 : if (mo_pfn <= spfn)
1895 : break;
1896 :
1897 : t = min(mo_pfn, epfn);
1898 : deferred_free_pages(spfn, t);
1899 :
1900 : if (mo_pfn <= epfn)
1901 : break;
1902 : }
1903 :
1904 : return nr_pages;
1905 : }
1906 :
1907 : static void __init
1908 : deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
1909 : void *arg)
1910 : {
1911 : unsigned long spfn, epfn;
1912 : struct zone *zone = arg;
1913 : u64 i;
1914 :
1915 : deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn);
1916 :
1917 : /*
1918 : * Initialize and free pages in MAX_ORDER sized increments so that we
1919 : * can avoid introducing any issues with the buddy allocator.
1920 : */
1921 : while (spfn < end_pfn) {
1922 : deferred_init_maxorder(&i, zone, &spfn, &epfn);
1923 : cond_resched();
1924 : }
1925 : }
1926 :
1927 : /* An arch may override for more concurrency. */
1928 : __weak int __init
1929 : deferred_page_init_max_threads(const struct cpumask *node_cpumask)
1930 : {
1931 : return 1;
1932 : }
1933 :
1934 : /* Initialise remaining memory on a node */
1935 : static int __init deferred_init_memmap(void *data)
1936 : {
1937 : pg_data_t *pgdat = data;
1938 : const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1939 : unsigned long spfn = 0, epfn = 0;
1940 : unsigned long first_init_pfn, flags;
1941 : unsigned long start = jiffies;
1942 : struct zone *zone;
1943 : int zid, max_threads;
1944 : u64 i;
1945 :
1946 : /* Bind memory initialisation thread to a local node if possible */
1947 : if (!cpumask_empty(cpumask))
1948 : set_cpus_allowed_ptr(current, cpumask);
1949 :
1950 : pgdat_resize_lock(pgdat, &flags);
1951 : first_init_pfn = pgdat->first_deferred_pfn;
1952 : if (first_init_pfn == ULONG_MAX) {
1953 : pgdat_resize_unlock(pgdat, &flags);
1954 : pgdat_init_report_one_done();
1955 : return 0;
1956 : }
1957 :
1958 : /* Sanity check boundaries */
1959 : BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1960 : BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1961 : pgdat->first_deferred_pfn = ULONG_MAX;
1962 :
1963 : /*
1964 : * Once we unlock here, the zone cannot be grown anymore, thus if an
1965 : * interrupt thread must allocate this early in boot, zone must be
1966 : * pre-grown prior to start of deferred page initialization.
1967 : */
1968 : pgdat_resize_unlock(pgdat, &flags);
1969 :
1970 : /* Only the highest zone is deferred so find it */
1971 : for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1972 : zone = pgdat->node_zones + zid;
1973 : if (first_init_pfn < zone_end_pfn(zone))
1974 : break;
1975 : }
1976 :
1977 : /* If the zone is empty somebody else may have cleared out the zone */
1978 : if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
1979 : first_init_pfn))
1980 : goto zone_empty;
1981 :
1982 : max_threads = deferred_page_init_max_threads(cpumask);
1983 :
1984 : while (spfn < epfn) {
1985 : unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION);
1986 : struct padata_mt_job job = {
1987 : .thread_fn = deferred_init_memmap_chunk,
1988 : .fn_arg = zone,
1989 : .start = spfn,
1990 : .size = epfn_align - spfn,
1991 : .align = PAGES_PER_SECTION,
1992 : .min_chunk = PAGES_PER_SECTION,
1993 : .max_threads = max_threads,
1994 : };
1995 :
1996 : padata_do_multithreaded(&job);
1997 : deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
1998 : epfn_align);
1999 : }
2000 : zone_empty:
2001 : /* Sanity check that the next zone really is unpopulated */
2002 : WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
2003 :
2004 : pr_info("node %d deferred pages initialised in %ums\n",
2005 : pgdat->node_id, jiffies_to_msecs(jiffies - start));
2006 :
2007 : pgdat_init_report_one_done();
2008 : return 0;
2009 : }
2010 :
2011 : /*
2012 : * If this zone has deferred pages, try to grow it by initializing enough
2013 : * deferred pages to satisfy the allocation specified by order, rounded up to
2014 : * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments
2015 : * of SECTION_SIZE bytes by initializing struct pages in increments of
2016 : * PAGES_PER_SECTION * sizeof(struct page) bytes.
2017 : *
2018 : * Return true when zone was grown, otherwise return false. We return true even
2019 : * when we grow less than requested, to let the caller decide if there are
2020 : * enough pages to satisfy the allocation.
2021 : *
2022 : * Note: We use noinline because this function is needed only during boot, and
2023 : * it is called from a __ref function _deferred_grow_zone. This way we are
2024 : * making sure that it is not inlined into permanent text section.
2025 : */
2026 : static noinline bool __init
2027 : deferred_grow_zone(struct zone *zone, unsigned int order)
2028 : {
2029 : unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
2030 : pg_data_t *pgdat = zone->zone_pgdat;
2031 : unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
2032 : unsigned long spfn, epfn, flags;
2033 : unsigned long nr_pages = 0;
2034 : u64 i;
2035 :
2036 : /* Only the last zone may have deferred pages */
2037 : if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
2038 : return false;
2039 :
2040 : pgdat_resize_lock(pgdat, &flags);
2041 :
2042 : /*
2043 : * If someone grew this zone while we were waiting for spinlock, return
2044 : * true, as there might be enough pages already.
2045 : */
2046 : if (first_deferred_pfn != pgdat->first_deferred_pfn) {
2047 : pgdat_resize_unlock(pgdat, &flags);
2048 : return true;
2049 : }
2050 :
2051 : /* If the zone is empty somebody else may have cleared out the zone */
2052 : if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2053 : first_deferred_pfn)) {
2054 : pgdat->first_deferred_pfn = ULONG_MAX;
2055 : pgdat_resize_unlock(pgdat, &flags);
2056 : /* Retry only once. */
2057 : return first_deferred_pfn != ULONG_MAX;
2058 : }
2059 :
2060 : /*
2061 : * Initialize and free pages in MAX_ORDER sized increments so
2062 : * that we can avoid introducing any issues with the buddy
2063 : * allocator.
2064 : */
2065 : while (spfn < epfn) {
2066 : /* update our first deferred PFN for this section */
2067 : first_deferred_pfn = spfn;
2068 :
2069 : nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
2070 : touch_nmi_watchdog();
2071 :
2072 : /* We should only stop along section boundaries */
2073 : if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
2074 : continue;
2075 :
2076 : /* If our quota has been met we can stop here */
2077 : if (nr_pages >= nr_pages_needed)
2078 : break;
2079 : }
2080 :
2081 : pgdat->first_deferred_pfn = spfn;
2082 : pgdat_resize_unlock(pgdat, &flags);
2083 :
2084 : return nr_pages > 0;
2085 : }
2086 :
2087 : /*
2088 : * deferred_grow_zone() is __init, but it is called from
2089 : * get_page_from_freelist() during early boot until deferred_pages permanently
2090 : * disables this call. This is why we have refdata wrapper to avoid warning,
2091 : * and to ensure that the function body gets unloaded.
2092 : */
2093 : static bool __ref
2094 : _deferred_grow_zone(struct zone *zone, unsigned int order)
2095 : {
2096 : return deferred_grow_zone(zone, order);
2097 : }
2098 :
2099 : #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
2100 :
2101 1 : void __init page_alloc_init_late(void)
2102 : {
2103 1 : struct zone *zone;
2104 1 : int nid;
2105 :
2106 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
2107 :
2108 : /* There will be num_node_state(N_MEMORY) threads */
2109 : atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
2110 : for_each_node_state(nid, N_MEMORY) {
2111 : kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
2112 : }
2113 :
2114 : /* Block until all are initialised */
2115 : wait_for_completion(&pgdat_init_all_done_comp);
2116 :
2117 : /*
2118 : * The number of managed pages has changed due to the initialisation
2119 : * so the pcpu batch and high limits needs to be updated or the limits
2120 : * will be artificially small.
2121 : */
2122 : for_each_populated_zone(zone)
2123 : zone_pcp_update(zone);
2124 :
2125 : /*
2126 : * We initialized the rest of the deferred pages. Permanently disable
2127 : * on-demand struct page initialization.
2128 : */
2129 : static_branch_disable(&deferred_pages);
2130 :
2131 : /* Reinit limits that are based on free pages after the kernel is up */
2132 : files_maxfiles_init();
2133 : #endif
2134 :
2135 1 : buffer_init();
2136 :
2137 : /* Discard memblock private memory */
2138 1 : memblock_discard();
2139 :
2140 2 : for_each_node_state(nid, N_MEMORY)
2141 1 : shuffle_free_memory(NODE_DATA(nid));
2142 :
2143 4 : for_each_populated_zone(zone)
2144 1 : set_zone_contiguous(zone);
2145 1 : }
2146 :
2147 : #ifdef CONFIG_CMA
2148 : /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
2149 : void __init init_cma_reserved_pageblock(struct page *page)
2150 : {
2151 : unsigned i = pageblock_nr_pages;
2152 : struct page *p = page;
2153 :
2154 : do {
2155 : __ClearPageReserved(p);
2156 : set_page_count(p, 0);
2157 : } while (++p, --i);
2158 :
2159 : set_pageblock_migratetype(page, MIGRATE_CMA);
2160 :
2161 : if (pageblock_order >= MAX_ORDER) {
2162 : i = pageblock_nr_pages;
2163 : p = page;
2164 : do {
2165 : set_page_refcounted(p);
2166 : __free_pages(p, MAX_ORDER - 1);
2167 : p += MAX_ORDER_NR_PAGES;
2168 : } while (i -= MAX_ORDER_NR_PAGES);
2169 : } else {
2170 : set_page_refcounted(page);
2171 : __free_pages(page, pageblock_order);
2172 : }
2173 :
2174 : adjust_managed_page_count(page, pageblock_nr_pages);
2175 : page_zone(page)->cma_pages += pageblock_nr_pages;
2176 : }
2177 : #endif
2178 :
2179 : /*
2180 : * The order of subdivision here is critical for the IO subsystem.
2181 : * Please do not alter this order without good reasons and regression
2182 : * testing. Specifically, as large blocks of memory are subdivided,
2183 : * the order in which smaller blocks are delivered depends on the order
2184 : * they're subdivided in this function. This is the primary factor
2185 : * influencing the order in which pages are delivered to the IO
2186 : * subsystem according to empirical testing, and this is also justified
2187 : * by considering the behavior of a buddy system containing a single
2188 : * large block of memory acted on by a series of small allocations.
2189 : * This behavior is a critical factor in sglist merging's success.
2190 : *
2191 : * -- nyc
2192 : */
2193 83160 : static inline void expand(struct zone *zone, struct page *page,
2194 : int low, int high, int migratetype)
2195 : {
2196 83160 : unsigned long size = 1 << high;
2197 :
2198 143308 : while (high > low) {
2199 60148 : high--;
2200 60148 : size >>= 1;
2201 60148 : VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
2202 :
2203 : /*
2204 : * Mark as guard pages (or page), that will allow to
2205 : * merge back to allocator when buddy will be freed.
2206 : * Corresponding page table entries will not be touched,
2207 : * pages will stay not present in virtual address space
2208 : */
2209 60148 : if (set_page_guard(zone, &page[size], high, migratetype))
2210 : continue;
2211 :
2212 60148 : add_to_free_list(&page[size], zone, high, migratetype);
2213 60148 : set_buddy_order(&page[size], high);
2214 : }
2215 83160 : }
2216 :
2217 0 : static void check_new_page_bad(struct page *page)
2218 : {
2219 0 : if (unlikely(page->flags & __PG_HWPOISON)) {
2220 : /* Don't complain about hwpoisoned pages */
2221 : page_mapcount_reset(page); /* remove PageBuddy */
2222 : return;
2223 : }
2224 :
2225 0 : bad_page(page,
2226 : page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
2227 : }
2228 :
2229 : /*
2230 : * This page is about to be returned from the page allocator
2231 : */
2232 291457 : static inline int check_new_page(struct page *page)
2233 : {
2234 291457 : if (likely(page_expected_state(page,
2235 : PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
2236 : return 0;
2237 :
2238 0 : check_new_page_bad(page);
2239 0 : return 1;
2240 : }
2241 :
2242 : #ifdef CONFIG_DEBUG_VM
2243 : /*
2244 : * With DEBUG_VM enabled, order-0 pages are checked for expected state when
2245 : * being allocated from pcp lists. With debug_pagealloc also enabled, they are
2246 : * also checked when pcp lists are refilled from the free lists.
2247 : */
2248 59738 : static inline bool check_pcp_refill(struct page *page)
2249 : {
2250 59738 : if (debug_pagealloc_enabled_static())
2251 : return check_new_page(page);
2252 : else
2253 59738 : return false;
2254 : }
2255 :
2256 169567 : static inline bool check_new_pcp(struct page *page)
2257 : {
2258 169567 : return check_new_page(page);
2259 : }
2260 : #else
2261 : /*
2262 : * With DEBUG_VM disabled, free order-0 pages are checked for expected state
2263 : * when pcp lists are being refilled from the free lists. With debug_pagealloc
2264 : * enabled, they are also checked when being allocated from the pcp lists.
2265 : */
2266 : static inline bool check_pcp_refill(struct page *page)
2267 : {
2268 : return check_new_page(page);
2269 : }
2270 : static inline bool check_new_pcp(struct page *page)
2271 : {
2272 : if (debug_pagealloc_enabled_static())
2273 : return check_new_page(page);
2274 : else
2275 : return false;
2276 : }
2277 : #endif /* CONFIG_DEBUG_VM */
2278 :
2279 23422 : static bool check_new_pages(struct page *page, unsigned int order)
2280 : {
2281 23422 : int i;
2282 145284 : for (i = 0; i < (1 << order); i++) {
2283 121862 : struct page *p = page + i;
2284 :
2285 121862 : if (unlikely(check_new_page(p)))
2286 : return true;
2287 : }
2288 :
2289 : return false;
2290 : }
2291 :
2292 193027 : inline void post_alloc_hook(struct page *page, unsigned int order,
2293 : gfp_t gfp_flags)
2294 : {
2295 193027 : set_page_private(page, 0);
2296 193027 : set_page_refcounted(page);
2297 :
2298 193031 : arch_alloc_page(page, order);
2299 193031 : debug_pagealloc_map_pages(page, 1 << order);
2300 193031 : kasan_alloc_pages(page, order);
2301 192995 : kernel_unpoison_pages(page, 1 << order);
2302 192995 : set_page_owner(page, order, gfp_flags);
2303 :
2304 192995 : if (!want_init_on_free() && want_init_on_alloc(gfp_flags))
2305 73017 : kernel_init_free_pages(page, 1 << order);
2306 193015 : }
2307 :
2308 193027 : static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
2309 : unsigned int alloc_flags)
2310 : {
2311 193027 : post_alloc_hook(page, order, gfp_flags);
2312 :
2313 193025 : if (order && (gfp_flags & __GFP_COMP))
2314 21783 : prep_compound_page(page, order);
2315 :
2316 : /*
2317 : * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
2318 : * allocate the page. The expectation is that the caller is taking
2319 : * steps that will free more memory. The caller should avoid the page
2320 : * being used for !PFMEMALLOC purposes.
2321 : */
2322 193025 : if (alloc_flags & ALLOC_NO_WATERMARKS)
2323 0 : set_page_pfmemalloc(page);
2324 : else
2325 193025 : clear_page_pfmemalloc(page);
2326 193025 : }
2327 :
2328 : /*
2329 : * Go through the free lists for the given migratetype and remove
2330 : * the smallest available page from the freelists
2331 : */
2332 : static __always_inline
2333 83198 : struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
2334 : int migratetype)
2335 : {
2336 83198 : unsigned int current_order;
2337 83198 : struct free_area *area;
2338 83198 : struct page *page;
2339 :
2340 : /* Find a page of the appropriate size in the preferred list */
2341 143685 : for (current_order = order; current_order < MAX_ORDER; ++current_order) {
2342 143647 : area = &(zone->free_area[current_order]);
2343 143647 : page = get_page_from_free_area(area, migratetype);
2344 83160 : if (!page)
2345 60487 : continue;
2346 83160 : del_page_from_free_list(page, zone, current_order);
2347 83160 : expand(zone, page, order, current_order, migratetype);
2348 83160 : set_pcppage_migratetype(page, migratetype);
2349 : return page;
2350 : }
2351 :
2352 : return NULL;
2353 : }
2354 :
2355 :
2356 : /*
2357 : * This array describes the order lists are fallen back to when
2358 : * the free lists for the desirable migrate type are depleted
2359 : */
2360 : static int fallbacks[MIGRATE_TYPES][3] = {
2361 : [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
2362 : [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
2363 : [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
2364 : #ifdef CONFIG_CMA
2365 : [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
2366 : #endif
2367 : #ifdef CONFIG_MEMORY_ISOLATION
2368 : [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
2369 : #endif
2370 : };
2371 :
2372 : #ifdef CONFIG_CMA
2373 : static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2374 : unsigned int order)
2375 : {
2376 : return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2377 : }
2378 : #else
2379 : static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2380 : unsigned int order) { return NULL; }
2381 : #endif
2382 :
2383 : /*
2384 : * Move the free pages in a range to the freelist tail of the requested type.
2385 : * Note that start_page and end_pages are not aligned on a pageblock
2386 : * boundary. If alignment is required, use move_freepages_block()
2387 : */
2388 0 : static int move_freepages(struct zone *zone,
2389 : struct page *start_page, struct page *end_page,
2390 : int migratetype, int *num_movable)
2391 : {
2392 0 : struct page *page;
2393 0 : unsigned int order;
2394 0 : int pages_moved = 0;
2395 :
2396 0 : for (page = start_page; page <= end_page;) {
2397 0 : if (!pfn_valid_within(page_to_pfn(page))) {
2398 : page++;
2399 : continue;
2400 : }
2401 :
2402 0 : if (!PageBuddy(page)) {
2403 : /*
2404 : * We assume that pages that could be isolated for
2405 : * migration are movable. But we don't actually try
2406 : * isolating, as that would be expensive.
2407 : */
2408 0 : if (num_movable &&
2409 0 : (PageLRU(page) || __PageMovable(page)))
2410 0 : (*num_movable)++;
2411 :
2412 0 : page++;
2413 0 : continue;
2414 : }
2415 :
2416 : /* Make sure we are not inadvertently changing nodes */
2417 0 : VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2418 0 : VM_BUG_ON_PAGE(page_zone(page) != zone, page);
2419 :
2420 0 : order = buddy_order(page);
2421 0 : move_to_free_list(page, zone, order, migratetype);
2422 0 : page += 1 << order;
2423 0 : pages_moved += 1 << order;
2424 : }
2425 :
2426 0 : return pages_moved;
2427 : }
2428 :
2429 0 : int move_freepages_block(struct zone *zone, struct page *page,
2430 : int migratetype, int *num_movable)
2431 : {
2432 0 : unsigned long start_pfn, end_pfn;
2433 0 : struct page *start_page, *end_page;
2434 :
2435 0 : if (num_movable)
2436 0 : *num_movable = 0;
2437 :
2438 0 : start_pfn = page_to_pfn(page);
2439 0 : start_pfn = start_pfn & ~(pageblock_nr_pages-1);
2440 0 : start_page = pfn_to_page(start_pfn);
2441 0 : end_page = start_page + pageblock_nr_pages - 1;
2442 0 : end_pfn = start_pfn + pageblock_nr_pages - 1;
2443 :
2444 : /* Do not cross zone boundaries */
2445 0 : if (!zone_spans_pfn(zone, start_pfn))
2446 0 : start_page = page;
2447 0 : if (!zone_spans_pfn(zone, end_pfn))
2448 : return 0;
2449 :
2450 0 : return move_freepages(zone, start_page, end_page, migratetype,
2451 : num_movable);
2452 : }
2453 :
2454 38 : static void change_pageblock_range(struct page *pageblock_page,
2455 : int start_order, int migratetype)
2456 : {
2457 38 : int nr_pageblocks = 1 << (start_order - pageblock_order);
2458 :
2459 76 : while (nr_pageblocks--) {
2460 38 : set_pageblock_migratetype(pageblock_page, migratetype);
2461 38 : pageblock_page += pageblock_nr_pages;
2462 : }
2463 38 : }
2464 :
2465 : /*
2466 : * When we are falling back to another migratetype during allocation, try to
2467 : * steal extra free pages from the same pageblocks to satisfy further
2468 : * allocations, instead of polluting multiple pageblocks.
2469 : *
2470 : * If we are stealing a relatively large buddy page, it is likely there will
2471 : * be more free pages in the pageblock, so try to steal them all. For
2472 : * reclaimable and unmovable allocations, we steal regardless of page size,
2473 : * as fragmentation caused by those allocations polluting movable pageblocks
2474 : * is worse than movable allocations stealing from unmovable and reclaimable
2475 : * pageblocks.
2476 : */
2477 38 : static bool can_steal_fallback(unsigned int order, int start_mt)
2478 : {
2479 : /*
2480 : * Leaving this order check is intended, although there is
2481 : * relaxed order check in next check. The reason is that
2482 : * we can actually steal whole pageblock if this condition met,
2483 : * but, below check doesn't guarantee it and that is just heuristic
2484 : * so could be changed anytime.
2485 : */
2486 38 : if (order >= pageblock_order)
2487 : return true;
2488 :
2489 0 : if (order >= pageblock_order / 2 ||
2490 0 : start_mt == MIGRATE_RECLAIMABLE ||
2491 0 : start_mt == MIGRATE_UNMOVABLE ||
2492 : page_group_by_mobility_disabled)
2493 0 : return true;
2494 :
2495 : return false;
2496 : }
2497 :
2498 0 : static inline bool boost_watermark(struct zone *zone)
2499 : {
2500 0 : unsigned long max_boost;
2501 :
2502 0 : if (!watermark_boost_factor)
2503 : return false;
2504 : /*
2505 : * Don't bother in zones that are unlikely to produce results.
2506 : * On small machines, including kdump capture kernels running
2507 : * in a small area, boosting the watermark can cause an out of
2508 : * memory situation immediately.
2509 : */
2510 0 : if ((pageblock_nr_pages * 4) > zone_managed_pages(zone))
2511 : return false;
2512 :
2513 0 : max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
2514 : watermark_boost_factor, 10000);
2515 :
2516 : /*
2517 : * high watermark may be uninitialised if fragmentation occurs
2518 : * very early in boot so do not boost. We do not fall
2519 : * through and boost by pageblock_nr_pages as failing
2520 : * allocations that early means that reclaim is not going
2521 : * to help and it may even be impossible to reclaim the
2522 : * boosted watermark resulting in a hang.
2523 : */
2524 0 : if (!max_boost)
2525 : return false;
2526 :
2527 0 : max_boost = max(pageblock_nr_pages, max_boost);
2528 :
2529 0 : zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
2530 : max_boost);
2531 :
2532 0 : return true;
2533 : }
2534 :
2535 : /*
2536 : * This function implements actual steal behaviour. If order is large enough,
2537 : * we can steal whole pageblock. If not, we first move freepages in this
2538 : * pageblock to our migratetype and determine how many already-allocated pages
2539 : * are there in the pageblock with a compatible migratetype. If at least half
2540 : * of pages are free or compatible, we can change migratetype of the pageblock
2541 : * itself, so pages freed in the future will be put on the correct free list.
2542 : */
2543 38 : static void steal_suitable_fallback(struct zone *zone, struct page *page,
2544 : unsigned int alloc_flags, int start_type, bool whole_block)
2545 : {
2546 38 : unsigned int current_order = buddy_order(page);
2547 38 : int free_pages, movable_pages, alike_pages;
2548 38 : int old_block_type;
2549 :
2550 38 : old_block_type = get_pageblock_migratetype(page);
2551 :
2552 : /*
2553 : * This can happen due to races and we want to prevent broken
2554 : * highatomic accounting.
2555 : */
2556 38 : if (is_migrate_highatomic(old_block_type))
2557 0 : goto single_page;
2558 :
2559 : /* Take ownership for orders >= pageblock_order */
2560 38 : if (current_order >= pageblock_order) {
2561 38 : change_pageblock_range(page, current_order, start_type);
2562 38 : goto single_page;
2563 : }
2564 :
2565 : /*
2566 : * Boost watermarks to increase reclaim pressure to reduce the
2567 : * likelihood of future fallbacks. Wake kswapd now as the node
2568 : * may be balanced overall and kswapd will not wake naturally.
2569 : */
2570 0 : if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD))
2571 0 : set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
2572 :
2573 : /* We are not allowed to try stealing from the whole block */
2574 0 : if (!whole_block)
2575 0 : goto single_page;
2576 :
2577 0 : free_pages = move_freepages_block(zone, page, start_type,
2578 : &movable_pages);
2579 : /*
2580 : * Determine how many pages are compatible with our allocation.
2581 : * For movable allocation, it's the number of movable pages which
2582 : * we just obtained. For other types it's a bit more tricky.
2583 : */
2584 0 : if (start_type == MIGRATE_MOVABLE) {
2585 0 : alike_pages = movable_pages;
2586 : } else {
2587 : /*
2588 : * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2589 : * to MOVABLE pageblock, consider all non-movable pages as
2590 : * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2591 : * vice versa, be conservative since we can't distinguish the
2592 : * exact migratetype of non-movable pages.
2593 : */
2594 0 : if (old_block_type == MIGRATE_MOVABLE)
2595 0 : alike_pages = pageblock_nr_pages
2596 0 : - (free_pages + movable_pages);
2597 : else
2598 : alike_pages = 0;
2599 : }
2600 :
2601 : /* moving whole block can fail due to zone boundary conditions */
2602 0 : if (!free_pages)
2603 0 : goto single_page;
2604 :
2605 : /*
2606 : * If a sufficient number of pages in the block are either free or of
2607 : * comparable migratability as our allocation, claim the whole block.
2608 : */
2609 0 : if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2610 : page_group_by_mobility_disabled)
2611 0 : set_pageblock_migratetype(page, start_type);
2612 :
2613 0 : return;
2614 :
2615 38 : single_page:
2616 38 : move_to_free_list(page, zone, current_order, start_type);
2617 : }
2618 :
2619 : /*
2620 : * Check whether there is a suitable fallback freepage with requested order.
2621 : * If only_stealable is true, this function returns fallback_mt only if
2622 : * we can steal other freepages all together. This would help to reduce
2623 : * fragmentation due to mixed migratetype pages in one pageblock.
2624 : */
2625 38 : int find_suitable_fallback(struct free_area *area, unsigned int order,
2626 : int migratetype, bool only_stealable, bool *can_steal)
2627 : {
2628 38 : int i;
2629 38 : int fallback_mt;
2630 :
2631 38 : if (area->nr_free == 0)
2632 : return -1;
2633 :
2634 38 : *can_steal = false;
2635 76 : for (i = 0;; i++) {
2636 76 : fallback_mt = fallbacks[migratetype][i];
2637 76 : if (fallback_mt == MIGRATE_TYPES)
2638 : break;
2639 :
2640 76 : if (free_area_empty(area, fallback_mt))
2641 38 : continue;
2642 :
2643 38 : if (can_steal_fallback(order, migratetype))
2644 38 : *can_steal = true;
2645 :
2646 38 : if (!only_stealable)
2647 38 : return fallback_mt;
2648 :
2649 0 : if (*can_steal)
2650 0 : return fallback_mt;
2651 : }
2652 :
2653 : return -1;
2654 : }
2655 :
2656 : /*
2657 : * Reserve a pageblock for exclusive use of high-order atomic allocations if
2658 : * there are no empty page blocks that contain a page with a suitable order
2659 : */
2660 0 : static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2661 : unsigned int alloc_order)
2662 : {
2663 0 : int mt;
2664 0 : unsigned long max_managed, flags;
2665 :
2666 : /*
2667 : * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2668 : * Check is race-prone but harmless.
2669 : */
2670 0 : max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
2671 0 : if (zone->nr_reserved_highatomic >= max_managed)
2672 : return;
2673 :
2674 0 : spin_lock_irqsave(&zone->lock, flags);
2675 :
2676 : /* Recheck the nr_reserved_highatomic limit under the lock */
2677 0 : if (zone->nr_reserved_highatomic >= max_managed)
2678 0 : goto out_unlock;
2679 :
2680 : /* Yoink! */
2681 0 : mt = get_pageblock_migratetype(page);
2682 0 : if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2683 : && !is_migrate_cma(mt)) {
2684 0 : zone->nr_reserved_highatomic += pageblock_nr_pages;
2685 0 : set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2686 0 : move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2687 : }
2688 :
2689 0 : out_unlock:
2690 0 : spin_unlock_irqrestore(&zone->lock, flags);
2691 : }
2692 :
2693 : /*
2694 : * Used when an allocation is about to fail under memory pressure. This
2695 : * potentially hurts the reliability of high-order allocations when under
2696 : * intense memory pressure but failed atomic allocations should be easier
2697 : * to recover from than an OOM.
2698 : *
2699 : * If @force is true, try to unreserve a pageblock even though highatomic
2700 : * pageblock is exhausted.
2701 : */
2702 0 : static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2703 : bool force)
2704 : {
2705 0 : struct zonelist *zonelist = ac->zonelist;
2706 0 : unsigned long flags;
2707 0 : struct zoneref *z;
2708 0 : struct zone *zone;
2709 0 : struct page *page;
2710 0 : int order;
2711 0 : bool ret;
2712 :
2713 0 : for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx,
2714 : ac->nodemask) {
2715 : /*
2716 : * Preserve at least one pageblock unless memory pressure
2717 : * is really high.
2718 : */
2719 0 : if (!force && zone->nr_reserved_highatomic <=
2720 : pageblock_nr_pages)
2721 0 : continue;
2722 :
2723 0 : spin_lock_irqsave(&zone->lock, flags);
2724 0 : for (order = 0; order < MAX_ORDER; order++) {
2725 0 : struct free_area *area = &(zone->free_area[order]);
2726 :
2727 0 : page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC);
2728 0 : if (!page)
2729 0 : continue;
2730 :
2731 : /*
2732 : * In page freeing path, migratetype change is racy so
2733 : * we can counter several free pages in a pageblock
2734 : * in this loop althoug we changed the pageblock type
2735 : * from highatomic to ac->migratetype. So we should
2736 : * adjust the count once.
2737 : */
2738 0 : if (is_migrate_highatomic_page(page)) {
2739 : /*
2740 : * It should never happen but changes to
2741 : * locking could inadvertently allow a per-cpu
2742 : * drain to add pages to MIGRATE_HIGHATOMIC
2743 : * while unreserving so be safe and watch for
2744 : * underflows.
2745 : */
2746 0 : zone->nr_reserved_highatomic -= min(
2747 : pageblock_nr_pages,
2748 : zone->nr_reserved_highatomic);
2749 : }
2750 :
2751 : /*
2752 : * Convert to ac->migratetype and avoid the normal
2753 : * pageblock stealing heuristics. Minimally, the caller
2754 : * is doing the work and needs the pages. More
2755 : * importantly, if the block was always converted to
2756 : * MIGRATE_UNMOVABLE or another type then the number
2757 : * of pageblocks that cannot be completely freed
2758 : * may increase.
2759 : */
2760 0 : set_pageblock_migratetype(page, ac->migratetype);
2761 0 : ret = move_freepages_block(zone, page, ac->migratetype,
2762 : NULL);
2763 0 : if (ret) {
2764 0 : spin_unlock_irqrestore(&zone->lock, flags);
2765 0 : return ret;
2766 : }
2767 : }
2768 0 : spin_unlock_irqrestore(&zone->lock, flags);
2769 : }
2770 :
2771 : return false;
2772 : }
2773 :
2774 : /*
2775 : * Try finding a free buddy page on the fallback list and put it on the free
2776 : * list of requested migratetype, possibly along with other pages from the same
2777 : * block, depending on fragmentation avoidance heuristics. Returns true if
2778 : * fallback was found so that __rmqueue_smallest() can grab it.
2779 : *
2780 : * The use of signed ints for order and current_order is a deliberate
2781 : * deviation from the rest of this file, to make the for loop
2782 : * condition simpler.
2783 : */
2784 : static __always_inline bool
2785 38 : __rmqueue_fallback(struct zone *zone, int order, int start_migratetype,
2786 : unsigned int alloc_flags)
2787 : {
2788 38 : struct free_area *area;
2789 38 : int current_order;
2790 38 : int min_order = order;
2791 38 : struct page *page;
2792 38 : int fallback_mt;
2793 38 : bool can_steal;
2794 :
2795 : /*
2796 : * Do not steal pages from freelists belonging to other pageblocks
2797 : * i.e. orders < pageblock_order. If there are no local zones free,
2798 : * the zonelists will be reiterated without ALLOC_NOFRAGMENT.
2799 : */
2800 38 : if (alloc_flags & ALLOC_NOFRAGMENT)
2801 0 : min_order = pageblock_order;
2802 :
2803 : /*
2804 : * Find the largest available free page in the other list. This roughly
2805 : * approximates finding the pageblock with the most free pages, which
2806 : * would be too costly to do exactly.
2807 : */
2808 38 : for (current_order = MAX_ORDER - 1; current_order >= min_order;
2809 0 : --current_order) {
2810 38 : area = &(zone->free_area[current_order]);
2811 38 : fallback_mt = find_suitable_fallback(area, current_order,
2812 : start_migratetype, false, &can_steal);
2813 38 : if (fallback_mt == -1)
2814 0 : continue;
2815 :
2816 : /*
2817 : * We cannot steal all free pages from the pageblock and the
2818 : * requested migratetype is movable. In that case it's better to
2819 : * steal and split the smallest available page instead of the
2820 : * largest available page, because even if the next movable
2821 : * allocation falls back into a different pageblock than this
2822 : * one, it won't cause permanent fragmentation.
2823 : */
2824 38 : if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2825 0 : && current_order > order)
2826 0 : goto find_smallest;
2827 :
2828 38 : goto do_steal;
2829 : }
2830 :
2831 : return false;
2832 :
2833 0 : find_smallest:
2834 0 : for (current_order = order; current_order < MAX_ORDER;
2835 0 : current_order++) {
2836 0 : area = &(zone->free_area[current_order]);
2837 0 : fallback_mt = find_suitable_fallback(area, current_order,
2838 : start_migratetype, false, &can_steal);
2839 0 : if (fallback_mt != -1)
2840 : break;
2841 : }
2842 :
2843 : /*
2844 : * This should not happen - we already found a suitable fallback
2845 : * when looking for the largest page.
2846 : */
2847 0 : VM_BUG_ON(current_order == MAX_ORDER);
2848 :
2849 0 : do_steal:
2850 38 : page = get_page_from_free_area(area, fallback_mt);
2851 :
2852 38 : steal_suitable_fallback(zone, page, alloc_flags, start_migratetype,
2853 : can_steal);
2854 :
2855 38 : trace_mm_page_alloc_extfrag(page, order, current_order,
2856 : start_migratetype, fallback_mt);
2857 :
2858 38 : return true;
2859 :
2860 : }
2861 :
2862 : /*
2863 : * Do the hard work of removing an element from the buddy allocator.
2864 : * Call me with the zone->lock already held.
2865 : */
2866 : static __always_inline struct page *
2867 : __rmqueue(struct zone *zone, unsigned int order, int migratetype,
2868 : unsigned int alloc_flags)
2869 : {
2870 : struct page *page;
2871 :
2872 : if (IS_ENABLED(CONFIG_CMA)) {
2873 : /*
2874 : * Balance movable allocations between regular and CMA areas by
2875 : * allocating from CMA when over half of the zone's free memory
2876 : * is in the CMA area.
2877 : */
2878 : if (alloc_flags & ALLOC_CMA &&
2879 : zone_page_state(zone, NR_FREE_CMA_PAGES) >
2880 : zone_page_state(zone, NR_FREE_PAGES) / 2) {
2881 : page = __rmqueue_cma_fallback(zone, order);
2882 : if (page)
2883 : goto out;
2884 : }
2885 : }
2886 59742 : retry:
2887 166396 : page = __rmqueue_smallest(zone, order, migratetype);
2888 83198 : if (unlikely(!page)) {
2889 38 : if (alloc_flags & ALLOC_CMA)
2890 38 : page = __rmqueue_cma_fallback(zone, order);
2891 :
2892 38 : if (!page && __rmqueue_fallback(zone, order, migratetype,
2893 : alloc_flags))
2894 38 : goto retry;
2895 : }
2896 83160 : out:
2897 83160 : if (page)
2898 83160 : trace_mm_page_alloc_zone_locked(page, order, migratetype);
2899 59738 : return page;
2900 : }
2901 :
2902 : /*
2903 : * Obtain a specified number of elements from the buddy allocator, all under
2904 : * a single hold of the lock, for efficiency. Add them to the supplied list.
2905 : * Returns the number of new pages which were placed at *list.
2906 : */
2907 1086 : static int rmqueue_bulk(struct zone *zone, unsigned int order,
2908 : unsigned long count, struct list_head *list,
2909 : int migratetype, unsigned int alloc_flags)
2910 : {
2911 1086 : int i, alloced = 0;
2912 :
2913 1086 : spin_lock(&zone->lock);
2914 61910 : for (i = 0; i < count; ++i) {
2915 59742 : struct page *page = __rmqueue(zone, order, migratetype,
2916 : alloc_flags);
2917 59738 : if (unlikely(page == NULL))
2918 : break;
2919 :
2920 59738 : if (unlikely(check_pcp_refill(page)))
2921 : continue;
2922 :
2923 : /*
2924 : * Split buddy pages returned by expand() are received here in
2925 : * physical page order. The page is added to the tail of
2926 : * caller's list. From the callers perspective, the linked list
2927 : * is ordered by page number under some conditions. This is
2928 : * useful for IO devices that can forward direction from the
2929 : * head, thus also in the physical page order. This is useful
2930 : * for IO devices that can merge IO requests if the physical
2931 : * pages are ordered properly.
2932 : */
2933 59738 : list_add_tail(&page->lru, list);
2934 59738 : alloced++;
2935 59738 : if (is_migrate_cma(get_pcppage_migratetype(page)))
2936 : __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2937 : -(1 << order));
2938 : }
2939 :
2940 : /*
2941 : * i pages were removed from the buddy list even if some leak due
2942 : * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2943 : * on i. Do not confuse with 'alloced' which is the number of
2944 : * pages added to the pcp list.
2945 : */
2946 1086 : __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2947 1086 : spin_unlock(&zone->lock);
2948 1086 : return alloced;
2949 : }
2950 :
2951 : #ifdef CONFIG_NUMA
2952 : /*
2953 : * Called from the vmstat counter updater to drain pagesets of this
2954 : * currently executing processor on remote nodes after they have
2955 : * expired.
2956 : *
2957 : * Note that this function must be called with the thread pinned to
2958 : * a single processor.
2959 : */
2960 0 : void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2961 : {
2962 0 : unsigned long flags;
2963 0 : int to_drain, batch;
2964 :
2965 0 : local_irq_save(flags);
2966 0 : batch = READ_ONCE(pcp->batch);
2967 0 : to_drain = min(pcp->count, batch);
2968 0 : if (to_drain > 0)
2969 0 : free_pcppages_bulk(zone, to_drain, pcp);
2970 0 : local_irq_restore(flags);
2971 0 : }
2972 : #endif
2973 :
2974 : /*
2975 : * Drain pcplists of the indicated processor and zone.
2976 : *
2977 : * The processor must either be the current processor and the
2978 : * thread pinned to the current processor or a processor that
2979 : * is not online.
2980 : */
2981 0 : static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2982 : {
2983 0 : unsigned long flags;
2984 0 : struct per_cpu_pageset *pset;
2985 0 : struct per_cpu_pages *pcp;
2986 :
2987 0 : local_irq_save(flags);
2988 0 : pset = per_cpu_ptr(zone->pageset, cpu);
2989 :
2990 0 : pcp = &pset->pcp;
2991 0 : if (pcp->count)
2992 0 : free_pcppages_bulk(zone, pcp->count, pcp);
2993 0 : local_irq_restore(flags);
2994 0 : }
2995 :
2996 : /*
2997 : * Drain pcplists of all zones on the indicated processor.
2998 : *
2999 : * The processor must either be the current processor and the
3000 : * thread pinned to the current processor or a processor that
3001 : * is not online.
3002 : */
3003 0 : static void drain_pages(unsigned int cpu)
3004 : {
3005 0 : struct zone *zone;
3006 :
3007 0 : for_each_populated_zone(zone) {
3008 0 : drain_pages_zone(cpu, zone);
3009 : }
3010 0 : }
3011 :
3012 : /*
3013 : * Spill all of this CPU's per-cpu pages back into the buddy allocator.
3014 : *
3015 : * The CPU has to be pinned. When zone parameter is non-NULL, spill just
3016 : * the single zone's pages.
3017 : */
3018 0 : void drain_local_pages(struct zone *zone)
3019 : {
3020 0 : int cpu = smp_processor_id();
3021 :
3022 0 : if (zone)
3023 0 : drain_pages_zone(cpu, zone);
3024 : else
3025 0 : drain_pages(cpu);
3026 0 : }
3027 :
3028 0 : static void drain_local_pages_wq(struct work_struct *work)
3029 : {
3030 0 : struct pcpu_drain *drain;
3031 :
3032 0 : drain = container_of(work, struct pcpu_drain, work);
3033 :
3034 : /*
3035 : * drain_all_pages doesn't use proper cpu hotplug protection so
3036 : * we can race with cpu offline when the WQ can move this from
3037 : * a cpu pinned worker to an unbound one. We can operate on a different
3038 : * cpu which is allright but we also have to make sure to not move to
3039 : * a different one.
3040 : */
3041 0 : preempt_disable();
3042 0 : drain_local_pages(drain->zone);
3043 0 : preempt_enable();
3044 0 : }
3045 :
3046 : /*
3047 : * The implementation of drain_all_pages(), exposing an extra parameter to
3048 : * drain on all cpus.
3049 : *
3050 : * drain_all_pages() is optimized to only execute on cpus where pcplists are
3051 : * not empty. The check for non-emptiness can however race with a free to
3052 : * pcplist that has not yet increased the pcp->count from 0 to 1. Callers
3053 : * that need the guarantee that every CPU has drained can disable the
3054 : * optimizing racy check.
3055 : */
3056 0 : static void __drain_all_pages(struct zone *zone, bool force_all_cpus)
3057 : {
3058 0 : int cpu;
3059 :
3060 : /*
3061 : * Allocate in the BSS so we wont require allocation in
3062 : * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
3063 : */
3064 0 : static cpumask_t cpus_with_pcps;
3065 :
3066 : /*
3067 : * Make sure nobody triggers this path before mm_percpu_wq is fully
3068 : * initialized.
3069 : */
3070 0 : if (WARN_ON_ONCE(!mm_percpu_wq))
3071 : return;
3072 :
3073 : /*
3074 : * Do not drain if one is already in progress unless it's specific to
3075 : * a zone. Such callers are primarily CMA and memory hotplug and need
3076 : * the drain to be complete when the call returns.
3077 : */
3078 0 : if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
3079 0 : if (!zone)
3080 : return;
3081 0 : mutex_lock(&pcpu_drain_mutex);
3082 : }
3083 :
3084 : /*
3085 : * We don't care about racing with CPU hotplug event
3086 : * as offline notification will cause the notified
3087 : * cpu to drain that CPU pcps and on_each_cpu_mask
3088 : * disables preemption as part of its processing
3089 : */
3090 0 : for_each_online_cpu(cpu) {
3091 0 : struct per_cpu_pageset *pcp;
3092 0 : struct zone *z;
3093 0 : bool has_pcps = false;
3094 :
3095 0 : if (force_all_cpus) {
3096 : /*
3097 : * The pcp.count check is racy, some callers need a
3098 : * guarantee that no cpu is missed.
3099 : */
3100 : has_pcps = true;
3101 0 : } else if (zone) {
3102 0 : pcp = per_cpu_ptr(zone->pageset, cpu);
3103 0 : if (pcp->pcp.count)
3104 : has_pcps = true;
3105 : } else {
3106 0 : for_each_populated_zone(z) {
3107 0 : pcp = per_cpu_ptr(z->pageset, cpu);
3108 0 : if (pcp->pcp.count) {
3109 : has_pcps = true;
3110 : break;
3111 : }
3112 : }
3113 : }
3114 :
3115 0 : if (has_pcps)
3116 0 : cpumask_set_cpu(cpu, &cpus_with_pcps);
3117 : else
3118 0 : cpumask_clear_cpu(cpu, &cpus_with_pcps);
3119 : }
3120 :
3121 0 : for_each_cpu(cpu, &cpus_with_pcps) {
3122 0 : struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu);
3123 :
3124 0 : drain->zone = zone;
3125 0 : INIT_WORK(&drain->work, drain_local_pages_wq);
3126 0 : queue_work_on(cpu, mm_percpu_wq, &drain->work);
3127 : }
3128 0 : for_each_cpu(cpu, &cpus_with_pcps)
3129 0 : flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work);
3130 :
3131 0 : mutex_unlock(&pcpu_drain_mutex);
3132 : }
3133 :
3134 : /*
3135 : * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
3136 : *
3137 : * When zone parameter is non-NULL, spill just the single zone's pages.
3138 : *
3139 : * Note that this can be extremely slow as the draining happens in a workqueue.
3140 : */
3141 0 : void drain_all_pages(struct zone *zone)
3142 : {
3143 0 : __drain_all_pages(zone, false);
3144 0 : }
3145 :
3146 : #ifdef CONFIG_HIBERNATION
3147 :
3148 : /*
3149 : * Touch the watchdog for every WD_PAGE_COUNT pages.
3150 : */
3151 : #define WD_PAGE_COUNT (128*1024)
3152 :
3153 : void mark_free_pages(struct zone *zone)
3154 : {
3155 : unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
3156 : unsigned long flags;
3157 : unsigned int order, t;
3158 : struct page *page;
3159 :
3160 : if (zone_is_empty(zone))
3161 : return;
3162 :
3163 : spin_lock_irqsave(&zone->lock, flags);
3164 :
3165 : max_zone_pfn = zone_end_pfn(zone);
3166 : for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
3167 : if (pfn_valid(pfn)) {
3168 : page = pfn_to_page(pfn);
3169 :
3170 : if (!--page_count) {
3171 : touch_nmi_watchdog();
3172 : page_count = WD_PAGE_COUNT;
3173 : }
3174 :
3175 : if (page_zone(page) != zone)
3176 : continue;
3177 :
3178 : if (!swsusp_page_is_forbidden(page))
3179 : swsusp_unset_page_free(page);
3180 : }
3181 :
3182 : for_each_migratetype_order(order, t) {
3183 : list_for_each_entry(page,
3184 : &zone->free_area[order].free_list[t], lru) {
3185 : unsigned long i;
3186 :
3187 : pfn = page_to_pfn(page);
3188 : for (i = 0; i < (1UL << order); i++) {
3189 : if (!--page_count) {
3190 : touch_nmi_watchdog();
3191 : page_count = WD_PAGE_COUNT;
3192 : }
3193 : swsusp_set_page_free(pfn_to_page(pfn + i));
3194 : }
3195 : }
3196 : }
3197 : spin_unlock_irqrestore(&zone->lock, flags);
3198 : }
3199 : #endif /* CONFIG_PM */
3200 :
3201 133369 : static bool free_unref_page_prepare(struct page *page, unsigned long pfn)
3202 : {
3203 133369 : int migratetype;
3204 :
3205 133369 : if (!free_pcp_prepare(page))
3206 : return false;
3207 :
3208 133325 : migratetype = get_pfnblock_migratetype(page, pfn);
3209 133325 : set_pcppage_migratetype(page, migratetype);
3210 133325 : return true;
3211 : }
3212 :
3213 133385 : static void free_unref_page_commit(struct page *page, unsigned long pfn)
3214 : {
3215 133385 : struct zone *zone = page_zone(page);
3216 133385 : struct per_cpu_pages *pcp;
3217 133385 : int migratetype;
3218 :
3219 133385 : migratetype = get_pcppage_migratetype(page);
3220 133385 : __count_vm_event(PGFREE);
3221 :
3222 : /*
3223 : * We only track unmovable, reclaimable and movable on pcp lists.
3224 : * Free ISOLATE pages back to the allocator because they are being
3225 : * offlined but treat HIGHATOMIC as movable pages so we can get those
3226 : * areas back if necessary. Otherwise, we may have to free
3227 : * excessively into the page allocator
3228 : */
3229 133385 : if (migratetype >= MIGRATE_PCPTYPES) {
3230 0 : if (unlikely(is_migrate_isolate(migratetype))) {
3231 : free_one_page(zone, page, pfn, 0, migratetype,
3232 : FPI_NONE);
3233 : return;
3234 : }
3235 0 : migratetype = MIGRATE_MOVABLE;
3236 : }
3237 :
3238 133385 : pcp = &this_cpu_ptr(zone->pageset)->pcp;
3239 133388 : list_add(&page->lru, &pcp->lists[migratetype]);
3240 133388 : pcp->count++;
3241 133388 : if (pcp->count >= READ_ONCE(pcp->high))
3242 357 : free_pcppages_bulk(zone, READ_ONCE(pcp->batch), pcp);
3243 : }
3244 :
3245 : /*
3246 : * Free a 0-order page
3247 : */
3248 65686 : void free_unref_page(struct page *page)
3249 : {
3250 65686 : unsigned long flags;
3251 65686 : unsigned long pfn = page_to_pfn(page);
3252 :
3253 65686 : if (!free_unref_page_prepare(page, pfn))
3254 : return;
3255 :
3256 131363 : local_irq_save(flags);
3257 65684 : free_unref_page_commit(page, pfn);
3258 65685 : local_irq_restore(flags);
3259 : }
3260 :
3261 : /*
3262 : * Free a list of 0-order pages
3263 : */
3264 27546 : void free_unref_page_list(struct list_head *list)
3265 : {
3266 27546 : struct page *page, *next;
3267 27546 : unsigned long flags, pfn;
3268 27546 : int batch_count = 0;
3269 :
3270 : /* Prepare pages for freeing */
3271 95205 : list_for_each_entry_safe(page, next, list, lru) {
3272 67659 : pfn = page_to_pfn(page);
3273 67659 : if (!free_unref_page_prepare(page, pfn))
3274 0 : list_del(&page->lru);
3275 67659 : set_page_private(page, pfn);
3276 : }
3277 :
3278 55092 : local_irq_save(flags);
3279 95247 : list_for_each_entry_safe(page, next, list, lru) {
3280 67701 : unsigned long pfn = page_private(page);
3281 :
3282 67701 : set_page_private(page, 0);
3283 67701 : trace_mm_page_free_batched(page);
3284 67702 : free_unref_page_commit(page, pfn);
3285 :
3286 : /*
3287 : * Guard against excessive IRQ disabled times when we get
3288 : * a large list of pages to free.
3289 : */
3290 67701 : if (++batch_count == SWAP_CLUSTER_MAX) {
3291 662 : local_irq_restore(flags);
3292 662 : batch_count = 0;
3293 68363 : local_irq_save(flags);
3294 : }
3295 : }
3296 27546 : local_irq_restore(flags);
3297 27546 : }
3298 :
3299 : /*
3300 : * split_page takes a non-compound higher-order page, and splits it into
3301 : * n (1<<order) sub-pages: page[0..n]
3302 : * Each sub-page must be freed individually.
3303 : *
3304 : * Note: this is probably too low level an operation for use in drivers.
3305 : * Please consult with lkml before using this in your driver.
3306 : */
3307 12 : void split_page(struct page *page, unsigned int order)
3308 : {
3309 12 : int i;
3310 :
3311 24 : VM_BUG_ON_PAGE(PageCompound(page), page);
3312 12 : VM_BUG_ON_PAGE(!page_count(page), page);
3313 :
3314 424 : for (i = 1; i < (1 << order); i++)
3315 412 : set_page_refcounted(page + i);
3316 12 : split_page_owner(page, 1 << order);
3317 12 : split_page_memcg(page, 1 << order);
3318 12 : }
3319 : EXPORT_SYMBOL_GPL(split_page);
3320 :
3321 0 : int __isolate_free_page(struct page *page, unsigned int order)
3322 : {
3323 0 : unsigned long watermark;
3324 0 : struct zone *zone;
3325 0 : int mt;
3326 :
3327 0 : BUG_ON(!PageBuddy(page));
3328 :
3329 0 : zone = page_zone(page);
3330 0 : mt = get_pageblock_migratetype(page);
3331 :
3332 0 : if (!is_migrate_isolate(mt)) {
3333 : /*
3334 : * Obey watermarks as if the page was being allocated. We can
3335 : * emulate a high-order watermark check with a raised order-0
3336 : * watermark, because we already know our high-order page
3337 : * exists.
3338 : */
3339 0 : watermark = zone->_watermark[WMARK_MIN] + (1UL << order);
3340 0 : if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
3341 : return 0;
3342 :
3343 0 : __mod_zone_freepage_state(zone, -(1UL << order), mt);
3344 : }
3345 :
3346 : /* Remove page from free list */
3347 :
3348 0 : del_page_from_free_list(page, zone, order);
3349 :
3350 : /*
3351 : * Set the pageblock if the isolated page is at least half of a
3352 : * pageblock
3353 : */
3354 0 : if (order >= pageblock_order - 1) {
3355 0 : struct page *endpage = page + (1 << order) - 1;
3356 0 : for (; page < endpage; page += pageblock_nr_pages) {
3357 0 : int mt = get_pageblock_migratetype(page);
3358 0 : if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
3359 0 : && !is_migrate_highatomic(mt))
3360 0 : set_pageblock_migratetype(page,
3361 : MIGRATE_MOVABLE);
3362 : }
3363 : }
3364 :
3365 :
3366 0 : return 1UL << order;
3367 : }
3368 :
3369 : /**
3370 : * __putback_isolated_page - Return a now-isolated page back where we got it
3371 : * @page: Page that was isolated
3372 : * @order: Order of the isolated page
3373 : * @mt: The page's pageblock's migratetype
3374 : *
3375 : * This function is meant to return a page pulled from the free lists via
3376 : * __isolate_free_page back to the free lists they were pulled from.
3377 : */
3378 0 : void __putback_isolated_page(struct page *page, unsigned int order, int mt)
3379 : {
3380 0 : struct zone *zone = page_zone(page);
3381 :
3382 : /* zone lock should be held when this function is called */
3383 0 : lockdep_assert_held(&zone->lock);
3384 :
3385 : /* Return isolated page to tail of freelist. */
3386 0 : __free_one_page(page, page_to_pfn(page), zone, order, mt,
3387 : FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL);
3388 0 : }
3389 :
3390 : /*
3391 : * Update NUMA hit/miss statistics
3392 : *
3393 : * Must be called with interrupts disabled.
3394 : */
3395 193003 : static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
3396 : {
3397 : #ifdef CONFIG_NUMA
3398 193003 : enum numa_stat_item local_stat = NUMA_LOCAL;
3399 :
3400 : /* skip numa counters update if numa stats is disabled */
3401 193003 : if (!static_branch_likely(&vm_numa_stat_key))
3402 : return;
3403 :
3404 193004 : if (zone_to_nid(z) != numa_node_id())
3405 0 : local_stat = NUMA_OTHER;
3406 :
3407 193004 : if (zone_to_nid(z) == zone_to_nid(preferred_zone))
3408 193004 : __inc_numa_state(z, NUMA_HIT);
3409 : else {
3410 0 : __inc_numa_state(z, NUMA_MISS);
3411 0 : __inc_numa_state(preferred_zone, NUMA_FOREIGN);
3412 : }
3413 193006 : __inc_numa_state(z, local_stat);
3414 : #endif
3415 : }
3416 :
3417 : /* Remove page from the per-cpu list, caller must protect the list */
3418 169567 : static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
3419 : unsigned int alloc_flags,
3420 : struct per_cpu_pages *pcp,
3421 : struct list_head *list)
3422 : {
3423 169567 : struct page *page;
3424 :
3425 169567 : do {
3426 169567 : if (list_empty(list)) {
3427 2170 : pcp->count += rmqueue_bulk(zone, 0,
3428 1085 : READ_ONCE(pcp->batch), list,
3429 : migratetype, alloc_flags);
3430 1085 : if (unlikely(list_empty(list)))
3431 : return NULL;
3432 : }
3433 :
3434 169567 : page = list_first_entry(list, struct page, lru);
3435 169567 : list_del(&page->lru);
3436 169567 : pcp->count--;
3437 169567 : } while (check_new_pcp(page));
3438 :
3439 : return page;
3440 : }
3441 :
3442 : /* Lock and remove page from the per-cpu list */
3443 169570 : static struct page *rmqueue_pcplist(struct zone *preferred_zone,
3444 : struct zone *zone, gfp_t gfp_flags,
3445 : int migratetype, unsigned int alloc_flags)
3446 : {
3447 169570 : struct per_cpu_pages *pcp;
3448 169570 : struct list_head *list;
3449 169570 : struct page *page;
3450 169570 : unsigned long flags;
3451 :
3452 339146 : local_irq_save(flags);
3453 169566 : pcp = &this_cpu_ptr(zone->pageset)->pcp;
3454 169567 : list = &pcp->lists[migratetype];
3455 169567 : page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list);
3456 169584 : if (page) {
3457 169584 : __count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
3458 169584 : zone_statistics(preferred_zone, zone);
3459 : }
3460 169587 : local_irq_restore(flags);
3461 169577 : return page;
3462 : }
3463 :
3464 : /*
3465 : * Allocate a page from the given zone. Use pcplists for order-0 allocations.
3466 : */
3467 : static inline
3468 192984 : struct page *rmqueue(struct zone *preferred_zone,
3469 : struct zone *zone, unsigned int order,
3470 : gfp_t gfp_flags, unsigned int alloc_flags,
3471 : int migratetype)
3472 : {
3473 192984 : unsigned long flags;
3474 192984 : struct page *page;
3475 :
3476 192984 : if (likely(order == 0)) {
3477 : /*
3478 : * MIGRATE_MOVABLE pcplist could have the pages on CMA area and
3479 : * we need to skip it when CMA area isn't allowed.
3480 : */
3481 169568 : if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA ||
3482 : migratetype != MIGRATE_MOVABLE) {
3483 169568 : page = rmqueue_pcplist(preferred_zone, zone, gfp_flags,
3484 : migratetype, alloc_flags);
3485 169576 : goto out;
3486 : }
3487 : }
3488 :
3489 : /*
3490 : * We most definitely don't want callers attempting to
3491 : * allocate greater than order-1 page units with __GFP_NOFAIL.
3492 : */
3493 46832 : WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3494 23416 : spin_lock_irqsave(&zone->lock, flags);
3495 :
3496 23422 : do {
3497 23422 : page = NULL;
3498 : /*
3499 : * order-0 request can reach here when the pcplist is skipped
3500 : * due to non-CMA allocation context. HIGHATOMIC area is
3501 : * reserved for high-order atomic allocation, so order-0
3502 : * request should skip it.
3503 : */
3504 23422 : if (order > 0 && alloc_flags & ALLOC_HARDER) {
3505 0 : page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3506 0 : if (page)
3507 0 : trace_mm_page_alloc_zone_locked(page, order, migratetype);
3508 : }
3509 0 : if (!page)
3510 23456 : page = __rmqueue(zone, order, migratetype, alloc_flags);
3511 23422 : } while (page && check_new_pages(page, order));
3512 23422 : spin_unlock(&zone->lock);
3513 23422 : if (!page)
3514 0 : goto failed;
3515 23422 : __mod_zone_freepage_state(zone, -(1 << order),
3516 : get_pcppage_migratetype(page));
3517 :
3518 23422 : __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3519 23422 : zone_statistics(preferred_zone, zone);
3520 23422 : local_irq_restore(flags);
3521 :
3522 192998 : out:
3523 : /* Separate test+clear to avoid unnecessary atomics */
3524 192998 : if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
3525 0 : clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
3526 0 : wakeup_kswapd(zone, 0, 0, zone_idx(zone));
3527 : }
3528 :
3529 192989 : VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3530 : return page;
3531 :
3532 0 : failed:
3533 0 : local_irq_restore(flags);
3534 : return NULL;
3535 : }
3536 :
3537 : #ifdef CONFIG_FAIL_PAGE_ALLOC
3538 :
3539 : static struct {
3540 : struct fault_attr attr;
3541 :
3542 : bool ignore_gfp_highmem;
3543 : bool ignore_gfp_reclaim;
3544 : u32 min_order;
3545 : } fail_page_alloc = {
3546 : .attr = FAULT_ATTR_INITIALIZER,
3547 : .ignore_gfp_reclaim = true,
3548 : .ignore_gfp_highmem = true,
3549 : .min_order = 1,
3550 : };
3551 :
3552 : static int __init setup_fail_page_alloc(char *str)
3553 : {
3554 : return setup_fault_attr(&fail_page_alloc.attr, str);
3555 : }
3556 : __setup("fail_page_alloc=", setup_fail_page_alloc);
3557 :
3558 : static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3559 : {
3560 : if (order < fail_page_alloc.min_order)
3561 : return false;
3562 : if (gfp_mask & __GFP_NOFAIL)
3563 : return false;
3564 : if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3565 : return false;
3566 : if (fail_page_alloc.ignore_gfp_reclaim &&
3567 : (gfp_mask & __GFP_DIRECT_RECLAIM))
3568 : return false;
3569 :
3570 : return should_fail(&fail_page_alloc.attr, 1 << order);
3571 : }
3572 :
3573 : #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3574 :
3575 : static int __init fail_page_alloc_debugfs(void)
3576 : {
3577 : umode_t mode = S_IFREG | 0600;
3578 : struct dentry *dir;
3579 :
3580 : dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3581 : &fail_page_alloc.attr);
3582 :
3583 : debugfs_create_bool("ignore-gfp-wait", mode, dir,
3584 : &fail_page_alloc.ignore_gfp_reclaim);
3585 : debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3586 : &fail_page_alloc.ignore_gfp_highmem);
3587 : debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order);
3588 :
3589 : return 0;
3590 : }
3591 :
3592 : late_initcall(fail_page_alloc_debugfs);
3593 :
3594 : #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
3595 :
3596 : #else /* CONFIG_FAIL_PAGE_ALLOC */
3597 :
3598 192976 : static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3599 : {
3600 192976 : return false;
3601 : }
3602 :
3603 : #endif /* CONFIG_FAIL_PAGE_ALLOC */
3604 :
3605 192976 : noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3606 : {
3607 192976 : return __should_fail_alloc_page(gfp_mask, order);
3608 : }
3609 : ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE);
3610 :
3611 192976 : static inline long __zone_watermark_unusable_free(struct zone *z,
3612 : unsigned int order, unsigned int alloc_flags)
3613 : {
3614 192976 : const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3615 192976 : long unusable_free = (1 << order) - 1;
3616 :
3617 : /*
3618 : * If the caller does not have rights to ALLOC_HARDER then subtract
3619 : * the high-atomic reserves. This will over-estimate the size of the
3620 : * atomic reserve but it avoids a search.
3621 : */
3622 192976 : if (likely(!alloc_harder))
3623 192976 : unusable_free += z->nr_reserved_highatomic;
3624 :
3625 : #ifdef CONFIG_CMA
3626 : /* If allocation can't use CMA areas don't use free CMA pages */
3627 : if (!(alloc_flags & ALLOC_CMA))
3628 : unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES);
3629 : #endif
3630 :
3631 192976 : return unusable_free;
3632 : }
3633 :
3634 : /*
3635 : * Return true if free base pages are above 'mark'. For high-order checks it
3636 : * will return true of the order-0 watermark is reached and there is at least
3637 : * one free page of a suitable size. Checking now avoids taking the zone lock
3638 : * to check in the allocation paths if no pages are free.
3639 : */
3640 23417 : bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3641 : int highest_zoneidx, unsigned int alloc_flags,
3642 : long free_pages)
3643 : {
3644 23417 : long min = mark;
3645 23417 : int o;
3646 23417 : const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3647 :
3648 : /* free_pages may go negative - that's OK */
3649 23417 : free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags);
3650 :
3651 23417 : if (alloc_flags & ALLOC_HIGH)
3652 0 : min -= min / 2;
3653 :
3654 23417 : if (unlikely(alloc_harder)) {
3655 : /*
3656 : * OOM victims can try even harder than normal ALLOC_HARDER
3657 : * users on the grounds that it's definitely going to be in
3658 : * the exit path shortly and free memory. Any allocation it
3659 : * makes during the free path will be small and short-lived.
3660 : */
3661 0 : if (alloc_flags & ALLOC_OOM)
3662 0 : min -= min / 2;
3663 : else
3664 0 : min -= min / 4;
3665 : }
3666 :
3667 : /*
3668 : * Check watermarks for an order-0 allocation request. If these
3669 : * are not met, then a high-order request also cannot go ahead
3670 : * even if a suitable page happened to be free.
3671 : */
3672 23417 : if (free_pages <= min + z->lowmem_reserve[highest_zoneidx])
3673 : return false;
3674 :
3675 : /* If this is an order-0 request then the watermark is fine */
3676 23417 : if (!order)
3677 : return true;
3678 :
3679 : /* For a high-order request, check at least one suitable page is free */
3680 24342 : for (o = order; o < MAX_ORDER; o++) {
3681 24342 : struct free_area *area = &z->free_area[o];
3682 24342 : int mt;
3683 :
3684 24342 : if (!area->nr_free)
3685 927 : continue;
3686 :
3687 30466 : for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3688 30466 : if (!free_area_empty(area, mt))
3689 : return true;
3690 : }
3691 :
3692 : #ifdef CONFIG_CMA
3693 : if ((alloc_flags & ALLOC_CMA) &&
3694 : !free_area_empty(area, MIGRATE_CMA)) {
3695 : return true;
3696 : }
3697 : #endif
3698 0 : if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC))
3699 : return true;
3700 : }
3701 : return false;
3702 : }
3703 :
3704 0 : bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3705 : int highest_zoneidx, unsigned int alloc_flags)
3706 : {
3707 0 : return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3708 0 : zone_page_state(z, NR_FREE_PAGES));
3709 : }
3710 :
3711 192968 : static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3712 : unsigned long mark, int highest_zoneidx,
3713 : unsigned int alloc_flags, gfp_t gfp_mask)
3714 : {
3715 192968 : long free_pages;
3716 :
3717 192968 : free_pages = zone_page_state(z, NR_FREE_PAGES);
3718 :
3719 : /*
3720 : * Fast check for order-0 only. If this fails then the reserves
3721 : * need to be calculated.
3722 : */
3723 192975 : if (!order) {
3724 169559 : long fast_free;
3725 :
3726 169559 : fast_free = free_pages;
3727 169559 : fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags);
3728 169559 : if (fast_free > mark + z->lowmem_reserve[highest_zoneidx])
3729 : return true;
3730 : }
3731 :
3732 23410 : if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3733 : free_pages))
3734 : return true;
3735 : /*
3736 : * Ignore watermark boosting for GFP_ATOMIC order-0 allocations
3737 : * when checking the min watermark. The min watermark is the
3738 : * point where boosting is ignored so that kswapd is woken up
3739 : * when below the low watermark.
3740 : */
3741 0 : if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost
3742 : && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) {
3743 0 : mark = z->_watermark[WMARK_MIN];
3744 0 : return __zone_watermark_ok(z, order, mark, highest_zoneidx,
3745 : alloc_flags, free_pages);
3746 : }
3747 :
3748 : return false;
3749 : }
3750 :
3751 2 : bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3752 : unsigned long mark, int highest_zoneidx)
3753 : {
3754 2 : long free_pages = zone_page_state(z, NR_FREE_PAGES);
3755 :
3756 2 : if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3757 0 : free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3758 :
3759 2 : return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0,
3760 : free_pages);
3761 : }
3762 :
3763 : #ifdef CONFIG_NUMA
3764 0 : static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3765 : {
3766 0 : return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3767 : node_reclaim_distance;
3768 : }
3769 : #else /* CONFIG_NUMA */
3770 : static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3771 : {
3772 : return true;
3773 : }
3774 : #endif /* CONFIG_NUMA */
3775 :
3776 : /*
3777 : * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid
3778 : * fragmentation is subtle. If the preferred zone was HIGHMEM then
3779 : * premature use of a lower zone may cause lowmem pressure problems that
3780 : * are worse than fragmentation. If the next zone is ZONE_DMA then it is
3781 : * probably too small. It only makes sense to spread allocations to avoid
3782 : * fragmentation between the Normal and DMA32 zones.
3783 : */
3784 : static inline unsigned int
3785 192987 : alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
3786 : {
3787 192987 : unsigned int alloc_flags;
3788 :
3789 : /*
3790 : * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD
3791 : * to save a branch.
3792 : */
3793 192987 : alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM);
3794 :
3795 : #ifdef CONFIG_ZONE_DMA32
3796 192987 : if (!zone)
3797 : return alloc_flags;
3798 :
3799 192987 : if (zone_idx(zone) != ZONE_NORMAL)
3800 : return alloc_flags;
3801 :
3802 : /*
3803 : * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and
3804 : * the pointer is within zone->zone_pgdat->node_zones[]. Also assume
3805 : * on UMA that if Normal is populated then so is DMA32.
3806 : */
3807 0 : BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1);
3808 0 : if (nr_online_nodes > 1 && !populated_zone(--zone))
3809 : return alloc_flags;
3810 :
3811 0 : alloc_flags |= ALLOC_NOFRAGMENT;
3812 : #endif /* CONFIG_ZONE_DMA32 */
3813 0 : return alloc_flags;
3814 : }
3815 :
3816 192945 : static inline unsigned int current_alloc_flags(gfp_t gfp_mask,
3817 : unsigned int alloc_flags)
3818 : {
3819 : #ifdef CONFIG_CMA
3820 : unsigned int pflags = current->flags;
3821 :
3822 : if (!(pflags & PF_MEMALLOC_NOCMA) &&
3823 : gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3824 : alloc_flags |= ALLOC_CMA;
3825 :
3826 : #endif
3827 192945 : return alloc_flags;
3828 : }
3829 :
3830 : /*
3831 : * get_page_from_freelist goes through the zonelist trying to allocate
3832 : * a page.
3833 : */
3834 : static struct page *
3835 192979 : get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
3836 : const struct alloc_context *ac)
3837 : {
3838 192979 : struct zoneref *z;
3839 192979 : struct zone *zone;
3840 192979 : struct pglist_data *last_pgdat_dirty_limit = NULL;
3841 192979 : bool no_fallback;
3842 :
3843 : retry:
3844 : /*
3845 : * Scan zonelist, looking for a zone with enough free.
3846 : * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
3847 : */
3848 192979 : no_fallback = alloc_flags & ALLOC_NOFRAGMENT;
3849 192979 : z = ac->preferred_zoneref;
3850 192978 : for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx,
3851 : ac->nodemask) {
3852 192978 : struct page *page;
3853 192978 : unsigned long mark;
3854 :
3855 192978 : if (cpusets_enabled() &&
3856 : (alloc_flags & ALLOC_CPUSET) &&
3857 : !__cpuset_zone_allowed(zone, gfp_mask))
3858 : continue;
3859 : /*
3860 : * When allocating a page cache page for writing, we
3861 : * want to get it from a node that is within its dirty
3862 : * limit, such that no single node holds more than its
3863 : * proportional share of globally allowed dirty pages.
3864 : * The dirty limits take into account the node's
3865 : * lowmem reserves and high watermark so that kswapd
3866 : * should be able to balance it without having to
3867 : * write pages from its LRU list.
3868 : *
3869 : * XXX: For now, allow allocations to potentially
3870 : * exceed the per-node dirty limit in the slowpath
3871 : * (spread_dirty_pages unset) before going into reclaim,
3872 : * which is important when on a NUMA setup the allowed
3873 : * nodes are together not big enough to reach the
3874 : * global limit. The proper fix for these situations
3875 : * will require awareness of nodes in the
3876 : * dirty-throttling and the flusher threads.
3877 : */
3878 192978 : if (ac->spread_dirty_pages) {
3879 1583 : if (last_pgdat_dirty_limit == zone->zone_pgdat)
3880 0 : continue;
3881 :
3882 1583 : if (!node_dirty_ok(zone->zone_pgdat)) {
3883 0 : last_pgdat_dirty_limit = zone->zone_pgdat;
3884 0 : continue;
3885 : }
3886 : }
3887 :
3888 192978 : if (no_fallback && nr_online_nodes > 1 &&
3889 0 : zone != ac->preferred_zoneref->zone) {
3890 0 : int local_nid;
3891 :
3892 : /*
3893 : * If moving to a remote node, retry but allow
3894 : * fragmenting fallbacks. Locality is more important
3895 : * than fragmentation avoidance.
3896 : */
3897 0 : local_nid = zone_to_nid(ac->preferred_zoneref->zone);
3898 0 : if (zone_to_nid(zone) != local_nid) {
3899 0 : alloc_flags &= ~ALLOC_NOFRAGMENT;
3900 0 : goto retry;
3901 : }
3902 : }
3903 :
3904 192978 : mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
3905 192971 : if (!zone_watermark_fast(zone, order, mark,
3906 192978 : ac->highest_zoneidx, alloc_flags,
3907 : gfp_mask)) {
3908 0 : int ret;
3909 :
3910 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3911 : /*
3912 : * Watermark failed for this zone, but see if we can
3913 : * grow this zone if it contains deferred pages.
3914 : */
3915 : if (static_branch_unlikely(&deferred_pages)) {
3916 : if (_deferred_grow_zone(zone, order))
3917 : goto try_this_zone;
3918 : }
3919 : #endif
3920 : /* Checked here to keep the fast path fast */
3921 0 : BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3922 0 : if (alloc_flags & ALLOC_NO_WATERMARKS)
3923 0 : goto try_this_zone;
3924 :
3925 0 : if (node_reclaim_mode == 0 ||
3926 0 : !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
3927 0 : continue;
3928 :
3929 0 : ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
3930 0 : switch (ret) {
3931 0 : case NODE_RECLAIM_NOSCAN:
3932 : /* did not scan */
3933 0 : continue;
3934 0 : case NODE_RECLAIM_FULL:
3935 : /* scanned but unreclaimable */
3936 0 : continue;
3937 0 : default:
3938 : /* did we reclaim enough */
3939 0 : if (zone_watermark_ok(zone, order, mark,
3940 0 : ac->highest_zoneidx, alloc_flags))
3941 0 : goto try_this_zone;
3942 :
3943 0 : continue;
3944 : }
3945 : }
3946 :
3947 192971 : try_this_zone:
3948 192971 : page = rmqueue(ac->preferred_zoneref->zone, zone, order,
3949 : gfp_mask, alloc_flags, ac->migratetype);
3950 192983 : if (page) {
3951 192984 : prep_new_page(page, order, gfp_mask, alloc_flags);
3952 :
3953 : /*
3954 : * If this is a high-order atomic allocation then check
3955 : * if the pageblock should be reserved for the future
3956 : */
3957 192980 : if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3958 0 : reserve_highatomic_pageblock(page, zone, order);
3959 :
3960 192980 : return page;
3961 : } else {
3962 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3963 : /* Try again if zone has deferred pages */
3964 : if (static_branch_unlikely(&deferred_pages)) {
3965 : if (_deferred_grow_zone(zone, order))
3966 : goto try_this_zone;
3967 : }
3968 : #endif
3969 0 : }
3970 : }
3971 :
3972 : /*
3973 : * It's possible on a UMA machine to get through all zones that are
3974 : * fragmented. If avoiding fragmentation, reset and try again.
3975 : */
3976 0 : if (no_fallback) {
3977 0 : alloc_flags &= ~ALLOC_NOFRAGMENT;
3978 0 : goto retry;
3979 : }
3980 :
3981 : return NULL;
3982 : }
3983 :
3984 0 : static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
3985 : {
3986 0 : unsigned int filter = SHOW_MEM_FILTER_NODES;
3987 :
3988 : /*
3989 : * This documents exceptions given to allocations in certain
3990 : * contexts that are allowed to allocate outside current's set
3991 : * of allowed nodes.
3992 : */
3993 0 : if (!(gfp_mask & __GFP_NOMEMALLOC))
3994 0 : if (tsk_is_oom_victim(current) ||
3995 0 : (current->flags & (PF_MEMALLOC | PF_EXITING)))
3996 : filter &= ~SHOW_MEM_FILTER_NODES;
3997 0 : if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
3998 0 : filter &= ~SHOW_MEM_FILTER_NODES;
3999 :
4000 0 : show_mem(filter, nodemask);
4001 0 : }
4002 :
4003 0 : void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
4004 : {
4005 0 : struct va_format vaf;
4006 0 : va_list args;
4007 0 : static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1);
4008 :
4009 0 : if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
4010 0 : return;
4011 :
4012 0 : va_start(args, fmt);
4013 0 : vaf.fmt = fmt;
4014 0 : vaf.va = &args;
4015 0 : pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl",
4016 : current->comm, &vaf, gfp_mask, &gfp_mask,
4017 : nodemask_pr_args(nodemask));
4018 0 : va_end(args);
4019 :
4020 0 : cpuset_print_current_mems_allowed();
4021 0 : pr_cont("\n");
4022 0 : dump_stack();
4023 0 : warn_alloc_show_mem(gfp_mask, nodemask);
4024 : }
4025 :
4026 : static inline struct page *
4027 0 : __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
4028 : unsigned int alloc_flags,
4029 : const struct alloc_context *ac)
4030 : {
4031 0 : struct page *page;
4032 :
4033 0 : page = get_page_from_freelist(gfp_mask, order,
4034 0 : alloc_flags|ALLOC_CPUSET, ac);
4035 : /*
4036 : * fallback to ignore cpuset restriction if our nodes
4037 : * are depleted
4038 : */
4039 0 : if (!page)
4040 0 : page = get_page_from_freelist(gfp_mask, order,
4041 : alloc_flags, ac);
4042 :
4043 0 : return page;
4044 : }
4045 :
4046 : static inline struct page *
4047 0 : __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
4048 : const struct alloc_context *ac, unsigned long *did_some_progress)
4049 : {
4050 0 : struct oom_control oc = {
4051 0 : .zonelist = ac->zonelist,
4052 0 : .nodemask = ac->nodemask,
4053 : .memcg = NULL,
4054 : .gfp_mask = gfp_mask,
4055 : .order = order,
4056 : };
4057 0 : struct page *page;
4058 :
4059 0 : *did_some_progress = 0;
4060 :
4061 : /*
4062 : * Acquire the oom lock. If that fails, somebody else is
4063 : * making progress for us.
4064 : */
4065 0 : if (!mutex_trylock(&oom_lock)) {
4066 0 : *did_some_progress = 1;
4067 0 : schedule_timeout_uninterruptible(1);
4068 0 : return NULL;
4069 : }
4070 :
4071 : /*
4072 : * Go through the zonelist yet one more time, keep very high watermark
4073 : * here, this is only to catch a parallel oom killing, we must fail if
4074 : * we're still under heavy pressure. But make sure that this reclaim
4075 : * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
4076 : * allocation which will never fail due to oom_lock already held.
4077 : */
4078 0 : page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
4079 : ~__GFP_DIRECT_RECLAIM, order,
4080 : ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
4081 0 : if (page)
4082 0 : goto out;
4083 :
4084 : /* Coredumps can quickly deplete all memory reserves */
4085 0 : if (current->flags & PF_DUMPCORE)
4086 0 : goto out;
4087 : /* The OOM killer will not help higher order allocs */
4088 0 : if (order > PAGE_ALLOC_COSTLY_ORDER)
4089 0 : goto out;
4090 : /*
4091 : * We have already exhausted all our reclaim opportunities without any
4092 : * success so it is time to admit defeat. We will skip the OOM killer
4093 : * because it is very likely that the caller has a more reasonable
4094 : * fallback than shooting a random task.
4095 : *
4096 : * The OOM killer may not free memory on a specific node.
4097 : */
4098 0 : if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE))
4099 0 : goto out;
4100 : /* The OOM killer does not needlessly kill tasks for lowmem */
4101 0 : if (ac->highest_zoneidx < ZONE_NORMAL)
4102 0 : goto out;
4103 0 : if (pm_suspended_storage())
4104 : goto out;
4105 : /*
4106 : * XXX: GFP_NOFS allocations should rather fail than rely on
4107 : * other request to make a forward progress.
4108 : * We are in an unfortunate situation where out_of_memory cannot
4109 : * do much for this context but let's try it to at least get
4110 : * access to memory reserved if the current task is killed (see
4111 : * out_of_memory). Once filesystems are ready to handle allocation
4112 : * failures more gracefully we should just bail out here.
4113 : */
4114 :
4115 : /* Exhausted what can be done so it's blame time */
4116 0 : if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
4117 0 : *did_some_progress = 1;
4118 :
4119 : /*
4120 : * Help non-failing allocations by giving them access to memory
4121 : * reserves
4122 : */
4123 0 : if (gfp_mask & __GFP_NOFAIL)
4124 0 : page = __alloc_pages_cpuset_fallback(gfp_mask, order,
4125 : ALLOC_NO_WATERMARKS, ac);
4126 : }
4127 0 : out:
4128 0 : mutex_unlock(&oom_lock);
4129 0 : return page;
4130 : }
4131 :
4132 : /*
4133 : * Maximum number of compaction retries wit a progress before OOM
4134 : * killer is consider as the only way to move forward.
4135 : */
4136 : #define MAX_COMPACT_RETRIES 16
4137 :
4138 : #ifdef CONFIG_COMPACTION
4139 : /* Try memory compaction for high-order allocations before reclaim */
4140 : static struct page *
4141 0 : __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4142 : unsigned int alloc_flags, const struct alloc_context *ac,
4143 : enum compact_priority prio, enum compact_result *compact_result)
4144 : {
4145 0 : struct page *page = NULL;
4146 0 : unsigned long pflags;
4147 0 : unsigned int noreclaim_flag;
4148 :
4149 0 : if (!order)
4150 : return NULL;
4151 :
4152 0 : psi_memstall_enter(&pflags);
4153 0 : noreclaim_flag = memalloc_noreclaim_save();
4154 :
4155 0 : *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
4156 : prio, &page);
4157 :
4158 0 : memalloc_noreclaim_restore(noreclaim_flag);
4159 0 : psi_memstall_leave(&pflags);
4160 :
4161 : /*
4162 : * At least in one zone compaction wasn't deferred or skipped, so let's
4163 : * count a compaction stall
4164 : */
4165 0 : count_vm_event(COMPACTSTALL);
4166 :
4167 : /* Prep a captured page if available */
4168 0 : if (page)
4169 0 : prep_new_page(page, order, gfp_mask, alloc_flags);
4170 :
4171 : /* Try get a page from the freelist if available */
4172 0 : if (!page)
4173 0 : page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4174 :
4175 0 : if (page) {
4176 0 : struct zone *zone = page_zone(page);
4177 :
4178 0 : zone->compact_blockskip_flush = false;
4179 0 : compaction_defer_reset(zone, order, true);
4180 0 : count_vm_event(COMPACTSUCCESS);
4181 0 : return page;
4182 : }
4183 :
4184 : /*
4185 : * It's bad if compaction run occurs and fails. The most likely reason
4186 : * is that pages exist, but not enough to satisfy watermarks.
4187 : */
4188 0 : count_vm_event(COMPACTFAIL);
4189 :
4190 0 : cond_resched();
4191 :
4192 0 : return NULL;
4193 : }
4194 :
4195 : static inline bool
4196 0 : should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
4197 : enum compact_result compact_result,
4198 : enum compact_priority *compact_priority,
4199 : int *compaction_retries)
4200 : {
4201 0 : int max_retries = MAX_COMPACT_RETRIES;
4202 0 : int min_priority;
4203 0 : bool ret = false;
4204 0 : int retries = *compaction_retries;
4205 0 : enum compact_priority priority = *compact_priority;
4206 :
4207 0 : if (!order)
4208 : return false;
4209 :
4210 0 : if (compaction_made_progress(compact_result))
4211 0 : (*compaction_retries)++;
4212 :
4213 : /*
4214 : * compaction considers all the zone as desperately out of memory
4215 : * so it doesn't really make much sense to retry except when the
4216 : * failure could be caused by insufficient priority
4217 : */
4218 0 : if (compaction_failed(compact_result))
4219 0 : goto check_priority;
4220 :
4221 : /*
4222 : * compaction was skipped because there are not enough order-0 pages
4223 : * to work with, so we retry only if it looks like reclaim can help.
4224 : */
4225 0 : if (compaction_needs_reclaim(compact_result)) {
4226 0 : ret = compaction_zonelist_suitable(ac, order, alloc_flags);
4227 0 : goto out;
4228 : }
4229 :
4230 : /*
4231 : * make sure the compaction wasn't deferred or didn't bail out early
4232 : * due to locks contention before we declare that we should give up.
4233 : * But the next retry should use a higher priority if allowed, so
4234 : * we don't just keep bailing out endlessly.
4235 : */
4236 0 : if (compaction_withdrawn(compact_result)) {
4237 0 : goto check_priority;
4238 : }
4239 :
4240 : /*
4241 : * !costly requests are much more important than __GFP_RETRY_MAYFAIL
4242 : * costly ones because they are de facto nofail and invoke OOM
4243 : * killer to move on while costly can fail and users are ready
4244 : * to cope with that. 1/4 retries is rather arbitrary but we
4245 : * would need much more detailed feedback from compaction to
4246 : * make a better decision.
4247 : */
4248 0 : if (order > PAGE_ALLOC_COSTLY_ORDER)
4249 0 : max_retries /= 4;
4250 0 : if (*compaction_retries <= max_retries) {
4251 0 : ret = true;
4252 0 : goto out;
4253 : }
4254 :
4255 : /*
4256 : * Make sure there are attempts at the highest priority if we exhausted
4257 : * all retries or failed at the lower priorities.
4258 : */
4259 0 : check_priority:
4260 0 : min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
4261 0 : MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
4262 :
4263 0 : if (*compact_priority > min_priority) {
4264 0 : (*compact_priority)--;
4265 0 : *compaction_retries = 0;
4266 0 : ret = true;
4267 : }
4268 0 : out:
4269 0 : trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
4270 0 : return ret;
4271 : }
4272 : #else
4273 : static inline struct page *
4274 : __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4275 : unsigned int alloc_flags, const struct alloc_context *ac,
4276 : enum compact_priority prio, enum compact_result *compact_result)
4277 : {
4278 : *compact_result = COMPACT_SKIPPED;
4279 : return NULL;
4280 : }
4281 :
4282 : static inline bool
4283 : should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
4284 : enum compact_result compact_result,
4285 : enum compact_priority *compact_priority,
4286 : int *compaction_retries)
4287 : {
4288 : struct zone *zone;
4289 : struct zoneref *z;
4290 :
4291 : if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
4292 : return false;
4293 :
4294 : /*
4295 : * There are setups with compaction disabled which would prefer to loop
4296 : * inside the allocator rather than hit the oom killer prematurely.
4297 : * Let's give them a good hope and keep retrying while the order-0
4298 : * watermarks are OK.
4299 : */
4300 : for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4301 : ac->highest_zoneidx, ac->nodemask) {
4302 : if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
4303 : ac->highest_zoneidx, alloc_flags))
4304 : return true;
4305 : }
4306 : return false;
4307 : }
4308 : #endif /* CONFIG_COMPACTION */
4309 :
4310 : #ifdef CONFIG_LOCKDEP
4311 : static struct lockdep_map __fs_reclaim_map =
4312 : STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
4313 :
4314 3295452 : static bool __need_reclaim(gfp_t gfp_mask)
4315 : {
4316 : /* no reclaim without waiting on it */
4317 3295452 : if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
4318 : return false;
4319 :
4320 : /* this guy won't enter reclaim */
4321 2852458 : if (current->flags & PF_MEMALLOC)
4322 : return false;
4323 :
4324 2852458 : if (gfp_mask & __GFP_NOLOCKDEP)
4325 0 : return false;
4326 :
4327 : return true;
4328 : }
4329 :
4330 1319244 : void __fs_reclaim_acquire(void)
4331 : {
4332 1319244 : lock_map_acquire(&__fs_reclaim_map);
4333 1319445 : }
4334 :
4335 1319507 : void __fs_reclaim_release(void)
4336 : {
4337 0 : lock_map_release(&__fs_reclaim_map);
4338 1319224 : }
4339 :
4340 1649009 : void fs_reclaim_acquire(gfp_t gfp_mask)
4341 : {
4342 1649009 : gfp_mask = current_gfp_context(gfp_mask);
4343 :
4344 1649047 : if (__need_reclaim(gfp_mask)) {
4345 1427604 : if (gfp_mask & __GFP_FS)
4346 1319226 : __fs_reclaim_acquire();
4347 :
4348 : #ifdef CONFIG_MMU_NOTIFIER
4349 : lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
4350 : lock_map_release(&__mmu_notifier_invalidate_range_start_map);
4351 : #endif
4352 :
4353 : }
4354 1649239 : }
4355 : EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
4356 :
4357 1648965 : void fs_reclaim_release(gfp_t gfp_mask)
4358 : {
4359 1648965 : gfp_mask = current_gfp_context(gfp_mask);
4360 :
4361 1649007 : if (__need_reclaim(gfp_mask)) {
4362 1427774 : if (gfp_mask & __GFP_FS)
4363 1319507 : __fs_reclaim_release();
4364 : }
4365 1648724 : }
4366 : EXPORT_SYMBOL_GPL(fs_reclaim_release);
4367 : #endif
4368 :
4369 : /* Perform direct synchronous page reclaim */
4370 : static unsigned long
4371 0 : __perform_reclaim(gfp_t gfp_mask, unsigned int order,
4372 : const struct alloc_context *ac)
4373 : {
4374 0 : unsigned int noreclaim_flag;
4375 0 : unsigned long pflags, progress;
4376 :
4377 0 : cond_resched();
4378 :
4379 : /* We now go into synchronous reclaim */
4380 0 : cpuset_memory_pressure_bump();
4381 0 : psi_memstall_enter(&pflags);
4382 0 : fs_reclaim_acquire(gfp_mask);
4383 0 : noreclaim_flag = memalloc_noreclaim_save();
4384 :
4385 0 : progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
4386 : ac->nodemask);
4387 :
4388 0 : memalloc_noreclaim_restore(noreclaim_flag);
4389 0 : fs_reclaim_release(gfp_mask);
4390 0 : psi_memstall_leave(&pflags);
4391 :
4392 0 : cond_resched();
4393 :
4394 0 : return progress;
4395 : }
4396 :
4397 : /* The really slow allocator path where we enter direct reclaim */
4398 : static inline struct page *
4399 0 : __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
4400 : unsigned int alloc_flags, const struct alloc_context *ac,
4401 : unsigned long *did_some_progress)
4402 : {
4403 0 : struct page *page = NULL;
4404 0 : bool drained = false;
4405 :
4406 0 : *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
4407 0 : if (unlikely(!(*did_some_progress)))
4408 : return NULL;
4409 :
4410 0 : retry:
4411 0 : page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4412 :
4413 : /*
4414 : * If an allocation failed after direct reclaim, it could be because
4415 : * pages are pinned on the per-cpu lists or in high alloc reserves.
4416 : * Shrink them and try again
4417 : */
4418 0 : if (!page && !drained) {
4419 0 : unreserve_highatomic_pageblock(ac, false);
4420 0 : drain_all_pages(NULL);
4421 0 : drained = true;
4422 0 : goto retry;
4423 : }
4424 :
4425 : return page;
4426 : }
4427 :
4428 0 : static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
4429 : const struct alloc_context *ac)
4430 : {
4431 0 : struct zoneref *z;
4432 0 : struct zone *zone;
4433 0 : pg_data_t *last_pgdat = NULL;
4434 0 : enum zone_type highest_zoneidx = ac->highest_zoneidx;
4435 :
4436 0 : for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx,
4437 : ac->nodemask) {
4438 0 : if (last_pgdat != zone->zone_pgdat)
4439 0 : wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx);
4440 0 : last_pgdat = zone->zone_pgdat;
4441 : }
4442 0 : }
4443 :
4444 : static inline unsigned int
4445 0 : gfp_to_alloc_flags(gfp_t gfp_mask)
4446 : {
4447 0 : unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
4448 :
4449 : /*
4450 : * __GFP_HIGH is assumed to be the same as ALLOC_HIGH
4451 : * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD
4452 : * to save two branches.
4453 : */
4454 0 : BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
4455 0 : BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD);
4456 :
4457 : /*
4458 : * The caller may dip into page reserves a bit more if the caller
4459 : * cannot run direct reclaim, or if the caller has realtime scheduling
4460 : * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
4461 : * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
4462 : */
4463 0 : alloc_flags |= (__force int)
4464 : (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM));
4465 :
4466 0 : if (gfp_mask & __GFP_ATOMIC) {
4467 : /*
4468 : * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
4469 : * if it can't schedule.
4470 : */
4471 0 : if (!(gfp_mask & __GFP_NOMEMALLOC))
4472 0 : alloc_flags |= ALLOC_HARDER;
4473 : /*
4474 : * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
4475 : * comment for __cpuset_node_allowed().
4476 : */
4477 0 : alloc_flags &= ~ALLOC_CPUSET;
4478 0 : } else if (unlikely(rt_task(current)) && !in_interrupt())
4479 0 : alloc_flags |= ALLOC_HARDER;
4480 :
4481 0 : alloc_flags = current_alloc_flags(gfp_mask, alloc_flags);
4482 :
4483 0 : return alloc_flags;
4484 : }
4485 :
4486 0 : static bool oom_reserves_allowed(struct task_struct *tsk)
4487 : {
4488 0 : if (!tsk_is_oom_victim(tsk))
4489 : return false;
4490 :
4491 : /*
4492 : * !MMU doesn't have oom reaper so give access to memory reserves
4493 : * only to the thread with TIF_MEMDIE set
4494 : */
4495 : if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
4496 : return false;
4497 :
4498 : return true;
4499 : }
4500 :
4501 : /*
4502 : * Distinguish requests which really need access to full memory
4503 : * reserves from oom victims which can live with a portion of it
4504 : */
4505 0 : static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
4506 : {
4507 0 : if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
4508 : return 0;
4509 0 : if (gfp_mask & __GFP_MEMALLOC)
4510 : return ALLOC_NO_WATERMARKS;
4511 0 : if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
4512 : return ALLOC_NO_WATERMARKS;
4513 0 : if (!in_interrupt()) {
4514 0 : if (current->flags & PF_MEMALLOC)
4515 : return ALLOC_NO_WATERMARKS;
4516 0 : else if (oom_reserves_allowed(current))
4517 0 : return ALLOC_OOM;
4518 : }
4519 :
4520 : return 0;
4521 : }
4522 :
4523 0 : bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
4524 : {
4525 0 : return !!__gfp_pfmemalloc_flags(gfp_mask);
4526 : }
4527 :
4528 : /*
4529 : * Checks whether it makes sense to retry the reclaim to make a forward progress
4530 : * for the given allocation request.
4531 : *
4532 : * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
4533 : * without success, or when we couldn't even meet the watermark if we
4534 : * reclaimed all remaining pages on the LRU lists.
4535 : *
4536 : * Returns true if a retry is viable or false to enter the oom path.
4537 : */
4538 : static inline bool
4539 0 : should_reclaim_retry(gfp_t gfp_mask, unsigned order,
4540 : struct alloc_context *ac, int alloc_flags,
4541 : bool did_some_progress, int *no_progress_loops)
4542 : {
4543 0 : struct zone *zone;
4544 0 : struct zoneref *z;
4545 0 : bool ret = false;
4546 :
4547 : /*
4548 : * Costly allocations might have made a progress but this doesn't mean
4549 : * their order will become available due to high fragmentation so
4550 : * always increment the no progress counter for them
4551 : */
4552 0 : if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
4553 0 : *no_progress_loops = 0;
4554 : else
4555 0 : (*no_progress_loops)++;
4556 :
4557 : /*
4558 : * Make sure we converge to OOM if we cannot make any progress
4559 : * several times in the row.
4560 : */
4561 0 : if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
4562 : /* Before OOM, exhaust highatomic_reserve */
4563 0 : return unreserve_highatomic_pageblock(ac, true);
4564 : }
4565 :
4566 : /*
4567 : * Keep reclaiming pages while there is a chance this will lead
4568 : * somewhere. If none of the target zones can satisfy our allocation
4569 : * request even if all reclaimable pages are considered then we are
4570 : * screwed and have to go OOM.
4571 : */
4572 0 : for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4573 : ac->highest_zoneidx, ac->nodemask) {
4574 0 : unsigned long available;
4575 0 : unsigned long reclaimable;
4576 0 : unsigned long min_wmark = min_wmark_pages(zone);
4577 0 : bool wmark;
4578 :
4579 0 : available = reclaimable = zone_reclaimable_pages(zone);
4580 0 : available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
4581 :
4582 : /*
4583 : * Would the allocation succeed if we reclaimed all
4584 : * reclaimable pages?
4585 : */
4586 0 : wmark = __zone_watermark_ok(zone, order, min_wmark,
4587 0 : ac->highest_zoneidx, alloc_flags, available);
4588 0 : trace_reclaim_retry_zone(z, order, reclaimable,
4589 : available, min_wmark, *no_progress_loops, wmark);
4590 0 : if (wmark) {
4591 : /*
4592 : * If we didn't make any progress and have a lot of
4593 : * dirty + writeback pages then we should wait for
4594 : * an IO to complete to slow down the reclaim and
4595 : * prevent from pre mature OOM
4596 : */
4597 0 : if (!did_some_progress) {
4598 0 : unsigned long write_pending;
4599 :
4600 0 : write_pending = zone_page_state_snapshot(zone,
4601 : NR_ZONE_WRITE_PENDING);
4602 :
4603 0 : if (2 * write_pending > reclaimable) {
4604 0 : congestion_wait(BLK_RW_ASYNC, HZ/10);
4605 0 : return true;
4606 : }
4607 : }
4608 :
4609 0 : ret = true;
4610 0 : goto out;
4611 : }
4612 : }
4613 :
4614 0 : out:
4615 : /*
4616 : * Memory allocation/reclaim might be called from a WQ context and the
4617 : * current implementation of the WQ concurrency control doesn't
4618 : * recognize that a particular WQ is congested if the worker thread is
4619 : * looping without ever sleeping. Therefore we have to do a short sleep
4620 : * here rather than calling cond_resched().
4621 : */
4622 0 : if (current->flags & PF_WQ_WORKER)
4623 0 : schedule_timeout_uninterruptible(1);
4624 : else
4625 0 : cond_resched();
4626 : return ret;
4627 : }
4628 :
4629 : static inline bool
4630 0 : check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4631 : {
4632 : /*
4633 : * It's possible that cpuset's mems_allowed and the nodemask from
4634 : * mempolicy don't intersect. This should be normally dealt with by
4635 : * policy_nodemask(), but it's possible to race with cpuset update in
4636 : * such a way the check therein was true, and then it became false
4637 : * before we got our cpuset_mems_cookie here.
4638 : * This assumes that for all allocations, ac->nodemask can come only
4639 : * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
4640 : * when it does not intersect with the cpuset restrictions) or the
4641 : * caller can deal with a violated nodemask.
4642 : */
4643 0 : if (cpusets_enabled() && ac->nodemask &&
4644 : !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4645 : ac->nodemask = NULL;
4646 : return true;
4647 : }
4648 :
4649 : /*
4650 : * When updating a task's mems_allowed or mempolicy nodemask, it is
4651 : * possible to race with parallel threads in such a way that our
4652 : * allocation can fail while the mask is being updated. If we are about
4653 : * to fail, check if the cpuset changed during allocation and if so,
4654 : * retry.
4655 : */
4656 0 : if (read_mems_allowed_retry(cpuset_mems_cookie))
4657 : return true;
4658 :
4659 0 : return false;
4660 : }
4661 :
4662 : static inline struct page *
4663 0 : __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4664 : struct alloc_context *ac)
4665 : {
4666 0 : bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4667 0 : const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4668 0 : struct page *page = NULL;
4669 0 : unsigned int alloc_flags;
4670 0 : unsigned long did_some_progress;
4671 0 : enum compact_priority compact_priority;
4672 0 : enum compact_result compact_result;
4673 0 : int compaction_retries;
4674 0 : int no_progress_loops;
4675 0 : unsigned int cpuset_mems_cookie;
4676 0 : int reserve_flags;
4677 :
4678 : /*
4679 : * We also sanity check to catch abuse of atomic reserves being used by
4680 : * callers that are not in atomic context.
4681 : */
4682 0 : if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4683 : (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4684 0 : gfp_mask &= ~__GFP_ATOMIC;
4685 :
4686 0 : retry_cpuset:
4687 0 : compaction_retries = 0;
4688 0 : no_progress_loops = 0;
4689 0 : compact_priority = DEF_COMPACT_PRIORITY;
4690 0 : cpuset_mems_cookie = read_mems_allowed_begin();
4691 :
4692 : /*
4693 : * The fast path uses conservative alloc_flags to succeed only until
4694 : * kswapd needs to be woken up, and to avoid the cost of setting up
4695 : * alloc_flags precisely. So we do that now.
4696 : */
4697 0 : alloc_flags = gfp_to_alloc_flags(gfp_mask);
4698 :
4699 : /*
4700 : * We need to recalculate the starting point for the zonelist iterator
4701 : * because we might have used different nodemask in the fast path, or
4702 : * there was a cpuset modification and we are retrying - otherwise we
4703 : * could end up iterating over non-eligible zones endlessly.
4704 : */
4705 0 : ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4706 : ac->highest_zoneidx, ac->nodemask);
4707 0 : if (!ac->preferred_zoneref->zone)
4708 0 : goto nopage;
4709 :
4710 0 : if (alloc_flags & ALLOC_KSWAPD)
4711 0 : wake_all_kswapds(order, gfp_mask, ac);
4712 :
4713 : /*
4714 : * The adjusted alloc_flags might result in immediate success, so try
4715 : * that first
4716 : */
4717 0 : page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4718 0 : if (page)
4719 0 : goto got_pg;
4720 :
4721 : /*
4722 : * For costly allocations, try direct compaction first, as it's likely
4723 : * that we have enough base pages and don't need to reclaim. For non-
4724 : * movable high-order allocations, do that as well, as compaction will
4725 : * try prevent permanent fragmentation by migrating from blocks of the
4726 : * same migratetype.
4727 : * Don't try this for allocations that are allowed to ignore
4728 : * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
4729 : */
4730 0 : if (can_direct_reclaim &&
4731 0 : (costly_order ||
4732 0 : (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4733 0 : && !gfp_pfmemalloc_allowed(gfp_mask)) {
4734 0 : page = __alloc_pages_direct_compact(gfp_mask, order,
4735 : alloc_flags, ac,
4736 : INIT_COMPACT_PRIORITY,
4737 : &compact_result);
4738 0 : if (page)
4739 0 : goto got_pg;
4740 :
4741 : /*
4742 : * Checks for costly allocations with __GFP_NORETRY, which
4743 : * includes some THP page fault allocations
4744 : */
4745 0 : if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4746 : /*
4747 : * If allocating entire pageblock(s) and compaction
4748 : * failed because all zones are below low watermarks
4749 : * or is prohibited because it recently failed at this
4750 : * order, fail immediately unless the allocator has
4751 : * requested compaction and reclaim retry.
4752 : *
4753 : * Reclaim is
4754 : * - potentially very expensive because zones are far
4755 : * below their low watermarks or this is part of very
4756 : * bursty high order allocations,
4757 : * - not guaranteed to help because isolate_freepages()
4758 : * may not iterate over freed pages as part of its
4759 : * linear scan, and
4760 : * - unlikely to make entire pageblocks free on its
4761 : * own.
4762 : */
4763 0 : if (compact_result == COMPACT_SKIPPED ||
4764 : compact_result == COMPACT_DEFERRED)
4765 0 : goto nopage;
4766 :
4767 : /*
4768 : * Looks like reclaim/compaction is worth trying, but
4769 : * sync compaction could be very expensive, so keep
4770 : * using async compaction.
4771 : */
4772 0 : compact_priority = INIT_COMPACT_PRIORITY;
4773 : }
4774 : }
4775 :
4776 0 : retry:
4777 : /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
4778 0 : if (alloc_flags & ALLOC_KSWAPD)
4779 0 : wake_all_kswapds(order, gfp_mask, ac);
4780 :
4781 0 : reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
4782 0 : if (reserve_flags)
4783 0 : alloc_flags = current_alloc_flags(gfp_mask, reserve_flags);
4784 :
4785 : /*
4786 : * Reset the nodemask and zonelist iterators if memory policies can be
4787 : * ignored. These allocations are high priority and system rather than
4788 : * user oriented.
4789 : */
4790 0 : if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
4791 0 : ac->nodemask = NULL;
4792 0 : ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4793 : ac->highest_zoneidx, ac->nodemask);
4794 : }
4795 :
4796 : /* Attempt with potentially adjusted zonelist and alloc_flags */
4797 0 : page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4798 0 : if (page)
4799 0 : goto got_pg;
4800 :
4801 : /* Caller is not willing to reclaim, we can't balance anything */
4802 0 : if (!can_direct_reclaim)
4803 0 : goto nopage;
4804 :
4805 : /* Avoid recursion of direct reclaim */
4806 0 : if (current->flags & PF_MEMALLOC)
4807 0 : goto nopage;
4808 :
4809 : /* Try direct reclaim and then allocating */
4810 0 : page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
4811 : &did_some_progress);
4812 0 : if (page)
4813 0 : goto got_pg;
4814 :
4815 : /* Try direct compaction and then allocating */
4816 0 : page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
4817 : compact_priority, &compact_result);
4818 0 : if (page)
4819 0 : goto got_pg;
4820 :
4821 : /* Do not loop if specifically requested */
4822 0 : if (gfp_mask & __GFP_NORETRY)
4823 0 : goto nopage;
4824 :
4825 : /*
4826 : * Do not retry costly high order allocations unless they are
4827 : * __GFP_RETRY_MAYFAIL
4828 : */
4829 0 : if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
4830 0 : goto nopage;
4831 :
4832 0 : if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
4833 : did_some_progress > 0, &no_progress_loops))
4834 0 : goto retry;
4835 :
4836 : /*
4837 : * It doesn't make any sense to retry for the compaction if the order-0
4838 : * reclaim is not able to make any progress because the current
4839 : * implementation of the compaction depends on the sufficient amount
4840 : * of free memory (see __compaction_suitable)
4841 : */
4842 0 : if (did_some_progress > 0 &&
4843 0 : should_compact_retry(ac, order, alloc_flags,
4844 : compact_result, &compact_priority,
4845 : &compaction_retries))
4846 0 : goto retry;
4847 :
4848 :
4849 : /* Deal with possible cpuset update races before we start OOM killing */
4850 0 : if (check_retry_cpuset(cpuset_mems_cookie, ac))
4851 : goto retry_cpuset;
4852 :
4853 : /* Reclaim has failed us, start killing things */
4854 0 : page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
4855 0 : if (page)
4856 0 : goto got_pg;
4857 :
4858 : /* Avoid allocations with no watermarks from looping endlessly */
4859 0 : if (tsk_is_oom_victim(current) &&
4860 0 : (alloc_flags & ALLOC_OOM ||
4861 0 : (gfp_mask & __GFP_NOMEMALLOC)))
4862 0 : goto nopage;
4863 :
4864 : /* Retry as long as the OOM killer is making progress */
4865 0 : if (did_some_progress) {
4866 0 : no_progress_loops = 0;
4867 0 : goto retry;
4868 : }
4869 :
4870 0 : nopage:
4871 : /* Deal with possible cpuset update races before we fail */
4872 0 : if (check_retry_cpuset(cpuset_mems_cookie, ac))
4873 : goto retry_cpuset;
4874 :
4875 : /*
4876 : * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
4877 : * we always retry
4878 : */
4879 0 : if (gfp_mask & __GFP_NOFAIL) {
4880 : /*
4881 : * All existing users of the __GFP_NOFAIL are blockable, so warn
4882 : * of any new users that actually require GFP_NOWAIT
4883 : */
4884 0 : if (WARN_ON_ONCE(!can_direct_reclaim))
4885 0 : goto fail;
4886 :
4887 : /*
4888 : * PF_MEMALLOC request from this context is rather bizarre
4889 : * because we cannot reclaim anything and only can loop waiting
4890 : * for somebody to do a work for us
4891 : */
4892 0 : WARN_ON_ONCE(current->flags & PF_MEMALLOC);
4893 :
4894 : /*
4895 : * non failing costly orders are a hard requirement which we
4896 : * are not prepared for much so let's warn about these users
4897 : * so that we can identify them and convert them to something
4898 : * else.
4899 : */
4900 0 : WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
4901 :
4902 : /*
4903 : * Help non-failing allocations by giving them access to memory
4904 : * reserves but do not use ALLOC_NO_WATERMARKS because this
4905 : * could deplete whole memory reserves which would just make
4906 : * the situation worse
4907 : */
4908 0 : page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
4909 0 : if (page)
4910 0 : goto got_pg;
4911 :
4912 0 : cond_resched();
4913 0 : goto retry;
4914 : }
4915 0 : fail:
4916 0 : warn_alloc(gfp_mask, ac->nodemask,
4917 : "page allocation failure: order:%u", order);
4918 0 : got_pg:
4919 0 : return page;
4920 : }
4921 :
4922 192956 : static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
4923 : int preferred_nid, nodemask_t *nodemask,
4924 : struct alloc_context *ac, gfp_t *alloc_mask,
4925 : unsigned int *alloc_flags)
4926 : {
4927 192956 : ac->highest_zoneidx = gfp_zone(gfp_mask);
4928 192960 : ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
4929 192960 : ac->nodemask = nodemask;
4930 192960 : ac->migratetype = gfp_migratetype(gfp_mask);
4931 :
4932 192961 : if (cpusets_enabled()) {
4933 : *alloc_mask |= __GFP_HARDWALL;
4934 : /*
4935 : * When we are in the interrupt context, it is irrelevant
4936 : * to the current task context. It means that any node ok.
4937 : */
4938 : if (!in_interrupt() && !ac->nodemask)
4939 : ac->nodemask = &cpuset_current_mems_allowed;
4940 : else
4941 : *alloc_flags |= ALLOC_CPUSET;
4942 : }
4943 :
4944 192961 : fs_reclaim_acquire(gfp_mask);
4945 192940 : fs_reclaim_release(gfp_mask);
4946 :
4947 192948 : might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
4948 :
4949 192960 : if (should_fail_alloc_page(gfp_mask, order))
4950 : return false;
4951 :
4952 192945 : *alloc_flags = current_alloc_flags(gfp_mask, *alloc_flags);
4953 :
4954 : /* Dirty zone balancing only done in the fast path */
4955 192945 : ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
4956 :
4957 : /*
4958 : * The preferred zone is used for statistics but crucially it is
4959 : * also used as the starting point for the zonelist iterator. It
4960 : * may get reset for allocations that ignore memory policies.
4961 : */
4962 192945 : ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4963 : ac->highest_zoneidx, ac->nodemask);
4964 :
4965 192957 : return true;
4966 : }
4967 :
4968 : /*
4969 : * This is the 'heart' of the zoned buddy allocator.
4970 : */
4971 : struct page *
4972 192957 : __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
4973 : nodemask_t *nodemask)
4974 : {
4975 192957 : struct page *page;
4976 192957 : unsigned int alloc_flags = ALLOC_WMARK_LOW;
4977 192957 : gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
4978 192957 : struct alloc_context ac = { };
4979 :
4980 : /*
4981 : * There are several places where we assume that the order value is sane
4982 : * so bail out early if the request is out of bound.
4983 : */
4984 192957 : if (unlikely(order >= MAX_ORDER)) {
4985 0 : WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
4986 : return NULL;
4987 : }
4988 :
4989 192957 : gfp_mask &= gfp_allowed_mask;
4990 192957 : alloc_mask = gfp_mask;
4991 192957 : if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
4992 : return NULL;
4993 :
4994 : /*
4995 : * Forbid the first pass from falling back to types that fragment
4996 : * memory until all local zones are considered.
4997 : */
4998 192958 : alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask);
4999 :
5000 : /* First allocation attempt */
5001 192958 : page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
5002 192956 : if (likely(page))
5003 192956 : goto out;
5004 :
5005 : /*
5006 : * Apply scoped allocation constraints. This is mainly about GFP_NOFS
5007 : * resp. GFP_NOIO which has to be inherited for all allocation requests
5008 : * from a particular context which has been marked by
5009 : * memalloc_no{fs,io}_{save,restore}.
5010 : */
5011 0 : alloc_mask = current_gfp_context(gfp_mask);
5012 0 : ac.spread_dirty_pages = false;
5013 :
5014 : /*
5015 : * Restore the original nodemask if it was potentially replaced with
5016 : * &cpuset_current_mems_allowed to optimize the fast-path attempt.
5017 : */
5018 0 : ac.nodemask = nodemask;
5019 :
5020 0 : page = __alloc_pages_slowpath(alloc_mask, order, &ac);
5021 :
5022 192956 : out:
5023 192956 : if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
5024 : unlikely(__memcg_kmem_charge_page(page, gfp_mask, order) != 0)) {
5025 : __free_pages(page, order);
5026 : page = NULL;
5027 : }
5028 :
5029 192956 : trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
5030 :
5031 192956 : return page;
5032 : }
5033 : EXPORT_SYMBOL(__alloc_pages_nodemask);
5034 :
5035 : /*
5036 : * Common helper functions. Never use with __GFP_HIGHMEM because the returned
5037 : * address cannot represent highmem pages. Use alloc_pages and then kmap if
5038 : * you need to access high mem.
5039 : */
5040 23127 : unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
5041 : {
5042 23127 : struct page *page;
5043 :
5044 23127 : page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
5045 23126 : if (!page)
5046 : return 0;
5047 23126 : return (unsigned long) page_address(page);
5048 : }
5049 : EXPORT_SYMBOL(__get_free_pages);
5050 :
5051 7356 : unsigned long get_zeroed_page(gfp_t gfp_mask)
5052 : {
5053 7356 : return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
5054 : }
5055 : EXPORT_SYMBOL(get_zeroed_page);
5056 :
5057 46009 : static inline void free_the_page(struct page *page, unsigned int order)
5058 : {
5059 46009 : if (order == 0) /* Via pcp? */
5060 31346 : free_unref_page(page);
5061 : else
5062 14663 : __free_pages_ok(page, order, FPI_NONE);
5063 46009 : }
5064 :
5065 : /**
5066 : * __free_pages - Free pages allocated with alloc_pages().
5067 : * @page: The page pointer returned from alloc_pages().
5068 : * @order: The order of the allocation.
5069 : *
5070 : * This function can free multi-page allocations that are not compound
5071 : * pages. It does not check that the @order passed in matches that of
5072 : * the allocation, so it is easy to leak memory. Freeing more memory
5073 : * than was allocated will probably emit a warning.
5074 : *
5075 : * If the last reference to this page is speculative, it will be released
5076 : * by put_page() which only frees the first page of a non-compound
5077 : * allocation. To prevent the remaining pages from being leaked, we free
5078 : * the subsequent pages here. If you want to use the page's reference
5079 : * count to decide when to free the allocation, you should allocate a
5080 : * compound page, and use put_page() instead of __free_pages().
5081 : *
5082 : * Context: May be called in interrupt context or while holding a normal
5083 : * spinlock, but not in NMI context or while holding a raw spinlock.
5084 : */
5085 46009 : void __free_pages(struct page *page, unsigned int order)
5086 : {
5087 46009 : if (put_page_testzero(page))
5088 46009 : free_the_page(page, order);
5089 0 : else if (!PageHead(page))
5090 0 : while (order-- > 0)
5091 0 : free_the_page(page + (1 << order), order);
5092 46009 : }
5093 : EXPORT_SYMBOL(__free_pages);
5094 :
5095 13040 : void free_pages(unsigned long addr, unsigned int order)
5096 : {
5097 13040 : if (addr != 0) {
5098 13040 : VM_BUG_ON(!virt_addr_valid((void *)addr));
5099 13040 : __free_pages(virt_to_page((void *)addr), order);
5100 : }
5101 13040 : }
5102 :
5103 : EXPORT_SYMBOL(free_pages);
5104 :
5105 : /*
5106 : * Page Fragment:
5107 : * An arbitrary-length arbitrary-offset area of memory which resides
5108 : * within a 0 or higher order page. Multiple fragments within that page
5109 : * are individually refcounted, in the page's reference counter.
5110 : *
5111 : * The page_frag functions below provide a simple allocation framework for
5112 : * page fragments. This is used by the network stack and network device
5113 : * drivers to provide a backing region of memory for use as either an
5114 : * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
5115 : */
5116 0 : static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
5117 : gfp_t gfp_mask)
5118 : {
5119 0 : struct page *page = NULL;
5120 0 : gfp_t gfp = gfp_mask;
5121 :
5122 : #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5123 0 : gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
5124 : __GFP_NOMEMALLOC;
5125 0 : page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
5126 0 : PAGE_FRAG_CACHE_MAX_ORDER);
5127 0 : nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
5128 : #endif
5129 0 : if (unlikely(!page))
5130 0 : page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
5131 :
5132 0 : nc->va = page ? page_address(page) : NULL;
5133 :
5134 0 : return page;
5135 : }
5136 :
5137 0 : void __page_frag_cache_drain(struct page *page, unsigned int count)
5138 : {
5139 0 : VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
5140 :
5141 0 : if (page_ref_sub_and_test(page, count))
5142 0 : free_the_page(page, compound_order(page));
5143 0 : }
5144 : EXPORT_SYMBOL(__page_frag_cache_drain);
5145 :
5146 0 : void *page_frag_alloc_align(struct page_frag_cache *nc,
5147 : unsigned int fragsz, gfp_t gfp_mask,
5148 : unsigned int align_mask)
5149 : {
5150 0 : unsigned int size = PAGE_SIZE;
5151 0 : struct page *page;
5152 0 : int offset;
5153 :
5154 0 : if (unlikely(!nc->va)) {
5155 0 : refill:
5156 0 : page = __page_frag_cache_refill(nc, gfp_mask);
5157 0 : if (!page)
5158 : return NULL;
5159 :
5160 : #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5161 : /* if size can vary use size else just use PAGE_SIZE */
5162 0 : size = nc->size;
5163 : #endif
5164 : /* Even if we own the page, we do not use atomic_set().
5165 : * This would break get_page_unless_zero() users.
5166 : */
5167 0 : page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
5168 :
5169 : /* reset page count bias and offset to start of new frag */
5170 0 : nc->pfmemalloc = page_is_pfmemalloc(page);
5171 0 : nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5172 0 : nc->offset = size;
5173 : }
5174 :
5175 0 : offset = nc->offset - fragsz;
5176 0 : if (unlikely(offset < 0)) {
5177 0 : page = virt_to_page(nc->va);
5178 :
5179 0 : if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
5180 0 : goto refill;
5181 :
5182 0 : if (unlikely(nc->pfmemalloc)) {
5183 0 : free_the_page(page, compound_order(page));
5184 0 : goto refill;
5185 : }
5186 :
5187 : #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5188 : /* if size can vary use size else just use PAGE_SIZE */
5189 0 : size = nc->size;
5190 : #endif
5191 : /* OK, page count is 0, we can safely set it */
5192 0 : set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
5193 :
5194 : /* reset page count bias and offset to start of new frag */
5195 0 : nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5196 0 : offset = size - fragsz;
5197 : }
5198 :
5199 0 : nc->pagecnt_bias--;
5200 0 : offset &= align_mask;
5201 0 : nc->offset = offset;
5202 :
5203 0 : return nc->va + offset;
5204 : }
5205 : EXPORT_SYMBOL(page_frag_alloc_align);
5206 :
5207 : /*
5208 : * Frees a page fragment allocated out of either a compound or order 0 page.
5209 : */
5210 0 : void page_frag_free(void *addr)
5211 : {
5212 0 : struct page *page = virt_to_head_page(addr);
5213 :
5214 0 : if (unlikely(put_page_testzero(page)))
5215 0 : free_the_page(page, compound_order(page));
5216 0 : }
5217 : EXPORT_SYMBOL(page_frag_free);
5218 :
5219 12 : static void *make_alloc_exact(unsigned long addr, unsigned int order,
5220 : size_t size)
5221 : {
5222 12 : if (addr) {
5223 12 : unsigned long alloc_end = addr + (PAGE_SIZE << order);
5224 12 : unsigned long used = addr + PAGE_ALIGN(size);
5225 :
5226 12 : split_page(virt_to_page((void *)addr), order);
5227 154 : while (used < alloc_end) {
5228 142 : free_page(used);
5229 142 : used += PAGE_SIZE;
5230 : }
5231 : }
5232 12 : return (void *)addr;
5233 : }
5234 :
5235 : /**
5236 : * alloc_pages_exact - allocate an exact number physically-contiguous pages.
5237 : * @size: the number of bytes to allocate
5238 : * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP
5239 : *
5240 : * This function is similar to alloc_pages(), except that it allocates the
5241 : * minimum number of pages to satisfy the request. alloc_pages() can only
5242 : * allocate memory in power-of-two pages.
5243 : *
5244 : * This function is also limited by MAX_ORDER.
5245 : *
5246 : * Memory allocated by this function must be released by free_pages_exact().
5247 : *
5248 : * Return: pointer to the allocated area or %NULL in case of error.
5249 : */
5250 12 : void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
5251 : {
5252 12 : unsigned int order = get_order(size);
5253 12 : unsigned long addr;
5254 :
5255 12 : if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
5256 0 : gfp_mask &= ~__GFP_COMP;
5257 :
5258 12 : addr = __get_free_pages(gfp_mask, order);
5259 12 : return make_alloc_exact(addr, order, size);
5260 : }
5261 : EXPORT_SYMBOL(alloc_pages_exact);
5262 :
5263 : /**
5264 : * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
5265 : * pages on a node.
5266 : * @nid: the preferred node ID where memory should be allocated
5267 : * @size: the number of bytes to allocate
5268 : * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP
5269 : *
5270 : * Like alloc_pages_exact(), but try to allocate on node nid first before falling
5271 : * back.
5272 : *
5273 : * Return: pointer to the allocated area or %NULL in case of error.
5274 : */
5275 0 : void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
5276 : {
5277 0 : unsigned int order = get_order(size);
5278 0 : struct page *p;
5279 :
5280 0 : if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
5281 0 : gfp_mask &= ~__GFP_COMP;
5282 :
5283 0 : p = alloc_pages_node(nid, gfp_mask, order);
5284 0 : if (!p)
5285 : return NULL;
5286 0 : return make_alloc_exact((unsigned long)page_address(p), order, size);
5287 : }
5288 :
5289 : /**
5290 : * free_pages_exact - release memory allocated via alloc_pages_exact()
5291 : * @virt: the value returned by alloc_pages_exact.
5292 : * @size: size of allocation, same value as passed to alloc_pages_exact().
5293 : *
5294 : * Release the memory allocated by a previous call to alloc_pages_exact.
5295 : */
5296 0 : void free_pages_exact(void *virt, size_t size)
5297 : {
5298 0 : unsigned long addr = (unsigned long)virt;
5299 0 : unsigned long end = addr + PAGE_ALIGN(size);
5300 :
5301 0 : while (addr < end) {
5302 0 : free_page(addr);
5303 0 : addr += PAGE_SIZE;
5304 : }
5305 0 : }
5306 : EXPORT_SYMBOL(free_pages_exact);
5307 :
5308 : /**
5309 : * nr_free_zone_pages - count number of pages beyond high watermark
5310 : * @offset: The zone index of the highest zone
5311 : *
5312 : * nr_free_zone_pages() counts the number of pages which are beyond the
5313 : * high watermark within all zones at or below a given zone index. For each
5314 : * zone, the number of pages is calculated as:
5315 : *
5316 : * nr_free_zone_pages = managed_pages - high_pages
5317 : *
5318 : * Return: number of pages beyond high watermark.
5319 : */
5320 7 : static unsigned long nr_free_zone_pages(int offset)
5321 : {
5322 7 : struct zoneref *z;
5323 7 : struct zone *zone;
5324 :
5325 : /* Just pick one node, since fallback list is circular */
5326 7 : unsigned long sum = 0;
5327 :
5328 7 : struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
5329 :
5330 14 : for_each_zone_zonelist(zone, z, zonelist, offset) {
5331 7 : unsigned long size = zone_managed_pages(zone);
5332 7 : unsigned long high = high_wmark_pages(zone);
5333 7 : if (size > high)
5334 7 : sum += size - high;
5335 : }
5336 :
5337 7 : return sum;
5338 : }
5339 :
5340 : /**
5341 : * nr_free_buffer_pages - count number of pages beyond high watermark
5342 : *
5343 : * nr_free_buffer_pages() counts the number of pages which are beyond the high
5344 : * watermark within ZONE_DMA and ZONE_NORMAL.
5345 : *
5346 : * Return: number of pages beyond high watermark within ZONE_DMA and
5347 : * ZONE_NORMAL.
5348 : */
5349 6 : unsigned long nr_free_buffer_pages(void)
5350 : {
5351 5 : return nr_free_zone_pages(gfp_zone(GFP_USER));
5352 : }
5353 : EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
5354 :
5355 0 : static inline void show_node(struct zone *zone)
5356 : {
5357 0 : if (IS_ENABLED(CONFIG_NUMA))
5358 0 : printk("Node %d ", zone_to_nid(zone));
5359 0 : }
5360 :
5361 9 : long si_mem_available(void)
5362 : {
5363 9 : long available;
5364 9 : unsigned long pagecache;
5365 9 : unsigned long wmark_low = 0;
5366 9 : unsigned long pages[NR_LRU_LISTS];
5367 9 : unsigned long reclaimable;
5368 9 : struct zone *zone;
5369 9 : int lru;
5370 :
5371 54 : for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
5372 45 : pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
5373 :
5374 36 : for_each_zone(zone)
5375 27 : wmark_low += low_wmark_pages(zone);
5376 :
5377 : /*
5378 : * Estimate the amount of memory available for userspace allocations,
5379 : * without causing swapping.
5380 : */
5381 9 : available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
5382 :
5383 : /*
5384 : * Not all the page cache can be freed, otherwise the system will
5385 : * start swapping. Assume at least half of the page cache, or the
5386 : * low watermark worth of cache, needs to stay.
5387 : */
5388 9 : pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
5389 9 : pagecache -= min(pagecache / 2, wmark_low);
5390 9 : available += pagecache;
5391 :
5392 : /*
5393 : * Part of the reclaimable slab and other kernel memory consists of
5394 : * items that are in use, and cannot be freed. Cap this estimate at the
5395 : * low watermark.
5396 : */
5397 9 : reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) +
5398 9 : global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
5399 9 : available += reclaimable - min(reclaimable / 2, wmark_low);
5400 :
5401 9 : if (available < 0)
5402 : available = 0;
5403 9 : return available;
5404 : }
5405 : EXPORT_SYMBOL_GPL(si_mem_available);
5406 :
5407 14 : void si_meminfo(struct sysinfo *val)
5408 : {
5409 14 : val->totalram = totalram_pages();
5410 14 : val->sharedram = global_node_page_state(NR_SHMEM);
5411 14 : val->freeram = global_zone_page_state(NR_FREE_PAGES);
5412 14 : val->bufferram = nr_blockdev_pages();
5413 14 : val->totalhigh = totalhigh_pages();
5414 14 : val->freehigh = nr_free_highpages();
5415 14 : val->mem_unit = PAGE_SIZE;
5416 14 : }
5417 :
5418 : EXPORT_SYMBOL(si_meminfo);
5419 :
5420 : #ifdef CONFIG_NUMA
5421 0 : void si_meminfo_node(struct sysinfo *val, int nid)
5422 : {
5423 0 : int zone_type; /* needs to be signed */
5424 0 : unsigned long managed_pages = 0;
5425 0 : unsigned long managed_highpages = 0;
5426 0 : unsigned long free_highpages = 0;
5427 0 : pg_data_t *pgdat = NODE_DATA(nid);
5428 :
5429 0 : for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
5430 0 : managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]);
5431 0 : val->totalram = managed_pages;
5432 0 : val->sharedram = node_page_state(pgdat, NR_SHMEM);
5433 0 : val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
5434 : #ifdef CONFIG_HIGHMEM
5435 : for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
5436 : struct zone *zone = &pgdat->node_zones[zone_type];
5437 :
5438 : if (is_highmem(zone)) {
5439 : managed_highpages += zone_managed_pages(zone);
5440 : free_highpages += zone_page_state(zone, NR_FREE_PAGES);
5441 : }
5442 : }
5443 : val->totalhigh = managed_highpages;
5444 : val->freehigh = free_highpages;
5445 : #else
5446 0 : val->totalhigh = managed_highpages;
5447 0 : val->freehigh = free_highpages;
5448 : #endif
5449 0 : val->mem_unit = PAGE_SIZE;
5450 0 : }
5451 : #endif
5452 :
5453 : /*
5454 : * Determine whether the node should be displayed or not, depending on whether
5455 : * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
5456 : */
5457 0 : static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
5458 : {
5459 0 : if (!(flags & SHOW_MEM_FILTER_NODES))
5460 : return false;
5461 :
5462 : /*
5463 : * no node mask - aka implicit memory numa policy. Do not bother with
5464 : * the synchronization - read_mems_allowed_begin - because we do not
5465 : * have to be precise here.
5466 : */
5467 0 : if (!nodemask)
5468 0 : nodemask = &cpuset_current_mems_allowed;
5469 :
5470 0 : return !node_isset(nid, *nodemask);
5471 : }
5472 :
5473 : #define K(x) ((x) << (PAGE_SHIFT-10))
5474 :
5475 0 : static void show_migration_types(unsigned char type)
5476 : {
5477 0 : static const char types[MIGRATE_TYPES] = {
5478 : [MIGRATE_UNMOVABLE] = 'U',
5479 : [MIGRATE_MOVABLE] = 'M',
5480 : [MIGRATE_RECLAIMABLE] = 'E',
5481 : [MIGRATE_HIGHATOMIC] = 'H',
5482 : #ifdef CONFIG_CMA
5483 : [MIGRATE_CMA] = 'C',
5484 : #endif
5485 : #ifdef CONFIG_MEMORY_ISOLATION
5486 : [MIGRATE_ISOLATE] = 'I',
5487 : #endif
5488 : };
5489 0 : char tmp[MIGRATE_TYPES + 1];
5490 0 : char *p = tmp;
5491 0 : int i;
5492 :
5493 0 : for (i = 0; i < MIGRATE_TYPES; i++) {
5494 0 : if (type & (1 << i))
5495 0 : *p++ = types[i];
5496 : }
5497 :
5498 0 : *p = '\0';
5499 0 : printk(KERN_CONT "(%s) ", tmp);
5500 0 : }
5501 :
5502 : /*
5503 : * Show free area list (used inside shift_scroll-lock stuff)
5504 : * We also calculate the percentage fragmentation. We do this by counting the
5505 : * memory on each free list with the exception of the first item on the list.
5506 : *
5507 : * Bits in @filter:
5508 : * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
5509 : * cpuset.
5510 : */
5511 0 : void show_free_areas(unsigned int filter, nodemask_t *nodemask)
5512 : {
5513 0 : unsigned long free_pcp = 0;
5514 0 : int cpu;
5515 0 : struct zone *zone;
5516 0 : pg_data_t *pgdat;
5517 :
5518 0 : for_each_populated_zone(zone) {
5519 0 : if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5520 0 : continue;
5521 :
5522 0 : for_each_online_cpu(cpu)
5523 0 : free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5524 : }
5525 :
5526 0 : printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
5527 : " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
5528 : " unevictable:%lu dirty:%lu writeback:%lu\n"
5529 : " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
5530 : " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
5531 : " free:%lu free_pcp:%lu free_cma:%lu\n",
5532 : global_node_page_state(NR_ACTIVE_ANON),
5533 : global_node_page_state(NR_INACTIVE_ANON),
5534 : global_node_page_state(NR_ISOLATED_ANON),
5535 : global_node_page_state(NR_ACTIVE_FILE),
5536 : global_node_page_state(NR_INACTIVE_FILE),
5537 : global_node_page_state(NR_ISOLATED_FILE),
5538 : global_node_page_state(NR_UNEVICTABLE),
5539 : global_node_page_state(NR_FILE_DIRTY),
5540 : global_node_page_state(NR_WRITEBACK),
5541 : global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B),
5542 : global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B),
5543 : global_node_page_state(NR_FILE_MAPPED),
5544 : global_node_page_state(NR_SHMEM),
5545 : global_node_page_state(NR_PAGETABLE),
5546 : global_zone_page_state(NR_BOUNCE),
5547 : global_zone_page_state(NR_FREE_PAGES),
5548 : free_pcp,
5549 : global_zone_page_state(NR_FREE_CMA_PAGES));
5550 :
5551 0 : for_each_online_pgdat(pgdat) {
5552 0 : if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
5553 0 : continue;
5554 :
5555 0 : printk("Node %d"
5556 : " active_anon:%lukB"
5557 : " inactive_anon:%lukB"
5558 : " active_file:%lukB"
5559 : " inactive_file:%lukB"
5560 : " unevictable:%lukB"
5561 : " isolated(anon):%lukB"
5562 : " isolated(file):%lukB"
5563 : " mapped:%lukB"
5564 : " dirty:%lukB"
5565 : " writeback:%lukB"
5566 : " shmem:%lukB"
5567 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5568 : " shmem_thp: %lukB"
5569 : " shmem_pmdmapped: %lukB"
5570 : " anon_thp: %lukB"
5571 : #endif
5572 : " writeback_tmp:%lukB"
5573 : " kernel_stack:%lukB"
5574 : #ifdef CONFIG_SHADOW_CALL_STACK
5575 : " shadow_call_stack:%lukB"
5576 : #endif
5577 : " pagetables:%lukB"
5578 : " all_unreclaimable? %s"
5579 : "\n",
5580 : pgdat->node_id,
5581 0 : K(node_page_state(pgdat, NR_ACTIVE_ANON)),
5582 0 : K(node_page_state(pgdat, NR_INACTIVE_ANON)),
5583 0 : K(node_page_state(pgdat, NR_ACTIVE_FILE)),
5584 0 : K(node_page_state(pgdat, NR_INACTIVE_FILE)),
5585 0 : K(node_page_state(pgdat, NR_UNEVICTABLE)),
5586 0 : K(node_page_state(pgdat, NR_ISOLATED_ANON)),
5587 0 : K(node_page_state(pgdat, NR_ISOLATED_FILE)),
5588 0 : K(node_page_state(pgdat, NR_FILE_MAPPED)),
5589 0 : K(node_page_state(pgdat, NR_FILE_DIRTY)),
5590 0 : K(node_page_state(pgdat, NR_WRITEBACK)),
5591 0 : K(node_page_state(pgdat, NR_SHMEM)),
5592 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5593 0 : K(node_page_state(pgdat, NR_SHMEM_THPS)),
5594 0 : K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
5595 0 : K(node_page_state(pgdat, NR_ANON_THPS)),
5596 : #endif
5597 0 : K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
5598 : node_page_state(pgdat, NR_KERNEL_STACK_KB),
5599 : #ifdef CONFIG_SHADOW_CALL_STACK
5600 : node_page_state(pgdat, NR_KERNEL_SCS_KB),
5601 : #endif
5602 0 : K(node_page_state(pgdat, NR_PAGETABLE)),
5603 0 : pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
5604 : "yes" : "no");
5605 : }
5606 :
5607 0 : for_each_populated_zone(zone) {
5608 0 : int i;
5609 :
5610 0 : if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5611 0 : continue;
5612 :
5613 : free_pcp = 0;
5614 0 : for_each_online_cpu(cpu)
5615 0 : free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5616 :
5617 0 : show_node(zone);
5618 0 : printk(KERN_CONT
5619 : "%s"
5620 : " free:%lukB"
5621 : " min:%lukB"
5622 : " low:%lukB"
5623 : " high:%lukB"
5624 : " reserved_highatomic:%luKB"
5625 : " active_anon:%lukB"
5626 : " inactive_anon:%lukB"
5627 : " active_file:%lukB"
5628 : " inactive_file:%lukB"
5629 : " unevictable:%lukB"
5630 : " writepending:%lukB"
5631 : " present:%lukB"
5632 : " managed:%lukB"
5633 : " mlocked:%lukB"
5634 : " bounce:%lukB"
5635 : " free_pcp:%lukB"
5636 : " local_pcp:%ukB"
5637 : " free_cma:%lukB"
5638 : "\n",
5639 : zone->name,
5640 0 : K(zone_page_state(zone, NR_FREE_PAGES)),
5641 0 : K(min_wmark_pages(zone)),
5642 0 : K(low_wmark_pages(zone)),
5643 0 : K(high_wmark_pages(zone)),
5644 0 : K(zone->nr_reserved_highatomic),
5645 0 : K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
5646 0 : K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
5647 0 : K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
5648 0 : K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
5649 0 : K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
5650 0 : K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
5651 0 : K(zone->present_pages),
5652 0 : K(zone_managed_pages(zone)),
5653 0 : K(zone_page_state(zone, NR_MLOCK)),
5654 0 : K(zone_page_state(zone, NR_BOUNCE)),
5655 : K(free_pcp),
5656 0 : K(this_cpu_read(zone->pageset->pcp.count)),
5657 0 : K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
5658 0 : printk("lowmem_reserve[]:");
5659 0 : for (i = 0; i < MAX_NR_ZONES; i++)
5660 0 : printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
5661 0 : printk(KERN_CONT "\n");
5662 : }
5663 :
5664 0 : for_each_populated_zone(zone) {
5665 0 : unsigned int order;
5666 0 : unsigned long nr[MAX_ORDER], flags, total = 0;
5667 0 : unsigned char types[MAX_ORDER];
5668 :
5669 0 : if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5670 0 : continue;
5671 0 : show_node(zone);
5672 0 : printk(KERN_CONT "%s: ", zone->name);
5673 :
5674 0 : spin_lock_irqsave(&zone->lock, flags);
5675 0 : for (order = 0; order < MAX_ORDER; order++) {
5676 0 : struct free_area *area = &zone->free_area[order];
5677 0 : int type;
5678 :
5679 0 : nr[order] = area->nr_free;
5680 0 : total += nr[order] << order;
5681 :
5682 0 : types[order] = 0;
5683 0 : for (type = 0; type < MIGRATE_TYPES; type++) {
5684 0 : if (!free_area_empty(area, type))
5685 0 : types[order] |= 1 << type;
5686 : }
5687 : }
5688 0 : spin_unlock_irqrestore(&zone->lock, flags);
5689 0 : for (order = 0; order < MAX_ORDER; order++) {
5690 0 : printk(KERN_CONT "%lu*%lukB ",
5691 : nr[order], K(1UL) << order);
5692 0 : if (nr[order])
5693 0 : show_migration_types(types[order]);
5694 : }
5695 0 : printk(KERN_CONT "= %lukB\n", K(total));
5696 : }
5697 :
5698 0 : hugetlb_show_meminfo();
5699 :
5700 0 : printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
5701 :
5702 0 : show_swap_cache_info();
5703 0 : }
5704 :
5705 2 : static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
5706 : {
5707 2 : zoneref->zone = zone;
5708 2 : zoneref->zone_idx = zone_idx(zone);
5709 : }
5710 :
5711 : /*
5712 : * Builds allocation fallback zone lists.
5713 : *
5714 : * Add all populated zones of a node to the zonelist.
5715 : */
5716 2 : static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
5717 : {
5718 2 : struct zone *zone;
5719 2 : enum zone_type zone_type = MAX_NR_ZONES;
5720 2 : int nr_zones = 0;
5721 :
5722 6 : do {
5723 6 : zone_type--;
5724 6 : zone = pgdat->node_zones + zone_type;
5725 6 : if (managed_zone(zone)) {
5726 2 : zoneref_set_zone(zone, &zonerefs[nr_zones++]);
5727 2 : check_highest_zone(zone_type);
5728 : }
5729 6 : } while (zone_type);
5730 :
5731 2 : return nr_zones;
5732 : }
5733 :
5734 : #ifdef CONFIG_NUMA
5735 :
5736 0 : static int __parse_numa_zonelist_order(char *s)
5737 : {
5738 : /*
5739 : * We used to support different zonlists modes but they turned
5740 : * out to be just not useful. Let's keep the warning in place
5741 : * if somebody still use the cmd line parameter so that we do
5742 : * not fail it silently
5743 : */
5744 0 : if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
5745 0 : pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s);
5746 0 : return -EINVAL;
5747 : }
5748 : return 0;
5749 : }
5750 :
5751 : char numa_zonelist_order[] = "Node";
5752 :
5753 : /*
5754 : * sysctl handler for numa_zonelist_order
5755 : */
5756 0 : int numa_zonelist_order_handler(struct ctl_table *table, int write,
5757 : void *buffer, size_t *length, loff_t *ppos)
5758 : {
5759 0 : if (write)
5760 0 : return __parse_numa_zonelist_order(buffer);
5761 0 : return proc_dostring(table, write, buffer, length, ppos);
5762 : }
5763 :
5764 :
5765 : #define MAX_NODE_LOAD (nr_online_nodes)
5766 : static int node_load[MAX_NUMNODES];
5767 :
5768 : /**
5769 : * find_next_best_node - find the next node that should appear in a given node's fallback list
5770 : * @node: node whose fallback list we're appending
5771 : * @used_node_mask: nodemask_t of already used nodes
5772 : *
5773 : * We use a number of factors to determine which is the next node that should
5774 : * appear on a given node's fallback list. The node should not have appeared
5775 : * already in @node's fallback list, and it should be the next closest node
5776 : * according to the distance array (which contains arbitrary distance values
5777 : * from each node to each node in the system), and should also prefer nodes
5778 : * with no CPUs, since presumably they'll have very little allocation pressure
5779 : * on them otherwise.
5780 : *
5781 : * Return: node id of the found node or %NUMA_NO_NODE if no node is found.
5782 : */
5783 2 : static int find_next_best_node(int node, nodemask_t *used_node_mask)
5784 : {
5785 2 : int n, val;
5786 2 : int min_val = INT_MAX;
5787 2 : int best_node = NUMA_NO_NODE;
5788 :
5789 : /* Use the local node if we haven't already */
5790 2 : if (!node_isset(node, *used_node_mask)) {
5791 1 : node_set(node, *used_node_mask);
5792 1 : return node;
5793 : }
5794 :
5795 2 : for_each_node_state(n, N_MEMORY) {
5796 :
5797 : /* Don't want a node to appear more than once */
5798 1 : if (node_isset(n, *used_node_mask))
5799 1 : continue;
5800 :
5801 : /* Use the distance array to find the distance */
5802 0 : val = node_distance(node, n);
5803 :
5804 : /* Penalize nodes under us ("prefer the next node") */
5805 0 : val += (n < node);
5806 :
5807 : /* Give preference to headless and unused nodes */
5808 0 : if (!cpumask_empty(cpumask_of_node(n)))
5809 0 : val += PENALTY_FOR_NODE_WITH_CPUS;
5810 :
5811 : /* Slight preference for less loaded node */
5812 0 : val *= (MAX_NODE_LOAD*MAX_NUMNODES);
5813 0 : val += node_load[n];
5814 :
5815 0 : if (val < min_val) {
5816 0 : min_val = val;
5817 0 : best_node = n;
5818 : }
5819 : }
5820 :
5821 1 : if (best_node >= 0)
5822 0 : node_set(best_node, *used_node_mask);
5823 :
5824 : return best_node;
5825 : }
5826 :
5827 :
5828 : /*
5829 : * Build zonelists ordered by node and zones within node.
5830 : * This results in maximum locality--normal zone overflows into local
5831 : * DMA zone, if any--but risks exhausting DMA zone.
5832 : */
5833 1 : static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
5834 : unsigned nr_nodes)
5835 : {
5836 1 : struct zoneref *zonerefs;
5837 1 : int i;
5838 :
5839 1 : zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5840 :
5841 2 : for (i = 0; i < nr_nodes; i++) {
5842 1 : int nr_zones;
5843 :
5844 1 : pg_data_t *node = NODE_DATA(node_order[i]);
5845 :
5846 1 : nr_zones = build_zonerefs_node(node, zonerefs);
5847 1 : zonerefs += nr_zones;
5848 : }
5849 1 : zonerefs->zone = NULL;
5850 1 : zonerefs->zone_idx = 0;
5851 1 : }
5852 :
5853 : /*
5854 : * Build gfp_thisnode zonelists
5855 : */
5856 1 : static void build_thisnode_zonelists(pg_data_t *pgdat)
5857 : {
5858 1 : struct zoneref *zonerefs;
5859 1 : int nr_zones;
5860 :
5861 1 : zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
5862 1 : nr_zones = build_zonerefs_node(pgdat, zonerefs);
5863 1 : zonerefs += nr_zones;
5864 1 : zonerefs->zone = NULL;
5865 1 : zonerefs->zone_idx = 0;
5866 1 : }
5867 :
5868 : /*
5869 : * Build zonelists ordered by zone and nodes within zones.
5870 : * This results in conserving DMA zone[s] until all Normal memory is
5871 : * exhausted, but results in overflowing to remote node while memory
5872 : * may still exist in local DMA zone.
5873 : */
5874 :
5875 1 : static void build_zonelists(pg_data_t *pgdat)
5876 : {
5877 1 : static int node_order[MAX_NUMNODES];
5878 1 : int node, load, nr_nodes = 0;
5879 1 : nodemask_t used_mask = NODE_MASK_NONE;
5880 1 : int local_node, prev_node;
5881 :
5882 : /* NUMA-aware ordering of nodes */
5883 1 : local_node = pgdat->node_id;
5884 1 : load = nr_online_nodes;
5885 1 : prev_node = local_node;
5886 :
5887 1 : memset(node_order, 0, sizeof(node_order));
5888 2 : while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5889 : /*
5890 : * We don't want to pressure a particular node.
5891 : * So adding penalty to the first node in same
5892 : * distance group to make it round-robin.
5893 : */
5894 2 : if (node_distance(local_node, node) !=
5895 1 : node_distance(local_node, prev_node))
5896 0 : node_load[node] = load;
5897 :
5898 1 : node_order[nr_nodes++] = node;
5899 1 : prev_node = node;
5900 1 : load--;
5901 : }
5902 :
5903 1 : build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5904 1 : build_thisnode_zonelists(pgdat);
5905 1 : }
5906 :
5907 : #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5908 : /*
5909 : * Return node id of node used for "local" allocations.
5910 : * I.e., first node id of first zone in arg node's generic zonelist.
5911 : * Used for initializing percpu 'numa_mem', which is used primarily
5912 : * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
5913 : */
5914 : int local_memory_node(int node)
5915 : {
5916 : struct zoneref *z;
5917 :
5918 : z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
5919 : gfp_zone(GFP_KERNEL),
5920 : NULL);
5921 : return zone_to_nid(z->zone);
5922 : }
5923 : #endif
5924 :
5925 : static void setup_min_unmapped_ratio(void);
5926 : static void setup_min_slab_ratio(void);
5927 : #else /* CONFIG_NUMA */
5928 :
5929 : static void build_zonelists(pg_data_t *pgdat)
5930 : {
5931 : int node, local_node;
5932 : struct zoneref *zonerefs;
5933 : int nr_zones;
5934 :
5935 : local_node = pgdat->node_id;
5936 :
5937 : zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5938 : nr_zones = build_zonerefs_node(pgdat, zonerefs);
5939 : zonerefs += nr_zones;
5940 :
5941 : /*
5942 : * Now we build the zonelist so that it contains the zones
5943 : * of all the other nodes.
5944 : * We don't want to pressure a particular node, so when
5945 : * building the zones for node N, we make sure that the
5946 : * zones coming right after the local ones are those from
5947 : * node N+1 (modulo N)
5948 : */
5949 : for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5950 : if (!node_online(node))
5951 : continue;
5952 : nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5953 : zonerefs += nr_zones;
5954 : }
5955 : for (node = 0; node < local_node; node++) {
5956 : if (!node_online(node))
5957 : continue;
5958 : nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5959 : zonerefs += nr_zones;
5960 : }
5961 :
5962 : zonerefs->zone = NULL;
5963 : zonerefs->zone_idx = 0;
5964 : }
5965 :
5966 : #endif /* CONFIG_NUMA */
5967 :
5968 : /*
5969 : * Boot pageset table. One per cpu which is going to be used for all
5970 : * zones and all nodes. The parameters will be set in such a way
5971 : * that an item put on a list will immediately be handed over to
5972 : * the buddy list. This is safe since pageset manipulation is done
5973 : * with interrupts disabled.
5974 : *
5975 : * The boot_pagesets must be kept even after bootup is complete for
5976 : * unused processors and/or zones. They do play a role for bootstrapping
5977 : * hotplugged processors.
5978 : *
5979 : * zoneinfo_show() and maybe other functions do
5980 : * not check if the processor is online before following the pageset pointer.
5981 : * Other parts of the kernel may not check if the zone is available.
5982 : */
5983 : static void pageset_init(struct per_cpu_pageset *p);
5984 : /* These effectively disable the pcplists in the boot pageset completely */
5985 : #define BOOT_PAGESET_HIGH 0
5986 : #define BOOT_PAGESET_BATCH 1
5987 : static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
5988 : static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
5989 :
5990 1 : static void __build_all_zonelists(void *data)
5991 : {
5992 1 : int nid;
5993 1 : int __maybe_unused cpu;
5994 1 : pg_data_t *self = data;
5995 1 : static DEFINE_SPINLOCK(lock);
5996 :
5997 1 : spin_lock(&lock);
5998 :
5999 : #ifdef CONFIG_NUMA
6000 1 : memset(node_load, 0, sizeof(node_load));
6001 : #endif
6002 :
6003 : /*
6004 : * This node is hotadded and no memory is yet present. So just
6005 : * building zonelists is fine - no need to touch other nodes.
6006 : */
6007 1 : if (self && !node_online(self->node_id)) {
6008 0 : build_zonelists(self);
6009 : } else {
6010 2 : for_each_online_node(nid) {
6011 1 : pg_data_t *pgdat = NODE_DATA(nid);
6012 :
6013 1 : build_zonelists(pgdat);
6014 : }
6015 :
6016 : #ifdef CONFIG_HAVE_MEMORYLESS_NODES
6017 : /*
6018 : * We now know the "local memory node" for each node--
6019 : * i.e., the node of the first zone in the generic zonelist.
6020 : * Set up numa_mem percpu variable for on-line cpus. During
6021 : * boot, only the boot cpu should be on-line; we'll init the
6022 : * secondary cpus' numa_mem as they come on-line. During
6023 : * node/memory hotplug, we'll fixup all on-line cpus.
6024 : */
6025 : for_each_online_cpu(cpu)
6026 : set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
6027 : #endif
6028 : }
6029 :
6030 1 : spin_unlock(&lock);
6031 1 : }
6032 :
6033 : static noinline void __init
6034 1 : build_all_zonelists_init(void)
6035 : {
6036 1 : int cpu;
6037 :
6038 1 : __build_all_zonelists(NULL);
6039 :
6040 : /*
6041 : * Initialize the boot_pagesets that are going to be used
6042 : * for bootstrapping processors. The real pagesets for
6043 : * each zone will be allocated later when the per cpu
6044 : * allocator is available.
6045 : *
6046 : * boot_pagesets are used also for bootstrapping offline
6047 : * cpus if the system is already booted because the pagesets
6048 : * are needed to initialize allocators on a specific cpu too.
6049 : * F.e. the percpu allocator needs the page allocator which
6050 : * needs the percpu allocator in order to allocate its pagesets
6051 : * (a chicken-egg dilemma).
6052 : */
6053 6 : for_each_possible_cpu(cpu)
6054 4 : pageset_init(&per_cpu(boot_pageset, cpu));
6055 :
6056 1 : mminit_verify_zonelist();
6057 1 : cpuset_init_current_mems_allowed();
6058 1 : }
6059 :
6060 : /*
6061 : * unless system_state == SYSTEM_BOOTING.
6062 : *
6063 : * __ref due to call of __init annotated helper build_all_zonelists_init
6064 : * [protected by SYSTEM_BOOTING].
6065 : */
6066 1 : void __ref build_all_zonelists(pg_data_t *pgdat)
6067 : {
6068 1 : unsigned long vm_total_pages;
6069 :
6070 1 : if (system_state == SYSTEM_BOOTING) {
6071 1 : build_all_zonelists_init();
6072 : } else {
6073 0 : __build_all_zonelists(pgdat);
6074 : /* cpuset refresh routine should be here */
6075 : }
6076 : /* Get the number of free pages beyond high watermark in all zones. */
6077 1 : vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
6078 : /*
6079 : * Disable grouping by mobility if the number of pages in the
6080 : * system is too low to allow the mechanism to work. It would be
6081 : * more accurate, but expensive to check per-zone. This check is
6082 : * made on memory-hotadd so a system can start with mobility
6083 : * disabled and enable it later
6084 : */
6085 1 : if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
6086 0 : page_group_by_mobility_disabled = 1;
6087 : else
6088 1 : page_group_by_mobility_disabled = 0;
6089 :
6090 2 : pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n",
6091 : nr_online_nodes,
6092 : page_group_by_mobility_disabled ? "off" : "on",
6093 : vm_total_pages);
6094 : #ifdef CONFIG_NUMA
6095 1 : pr_info("Policy zone: %s\n", zone_names[policy_zone]);
6096 : #endif
6097 1 : }
6098 :
6099 : /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */
6100 : static bool __meminit
6101 262046 : overlap_memmap_init(unsigned long zone, unsigned long *pfn)
6102 : {
6103 262046 : static struct memblock_region *r;
6104 :
6105 262046 : if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
6106 0 : if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
6107 0 : for_each_mem_region(r) {
6108 0 : if (*pfn < memblock_region_memory_end_pfn(r))
6109 : break;
6110 : }
6111 : }
6112 0 : if (*pfn >= memblock_region_memory_base_pfn(r) &&
6113 0 : memblock_is_mirror(r)) {
6114 0 : *pfn = memblock_region_memory_end_pfn(r);
6115 0 : return true;
6116 : }
6117 : }
6118 : return false;
6119 : }
6120 :
6121 : /*
6122 : * Initially all pages are reserved - free ones are freed
6123 : * up by memblock_free_all() once the early boot process is
6124 : * done. Non-atomic initialization, single-pass.
6125 : *
6126 : * All aligned pageblocks are initialized to the specified migratetype
6127 : * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
6128 : * zone stats (e.g., nr_isolate_pageblock) are touched.
6129 : */
6130 2 : void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone,
6131 : unsigned long start_pfn, unsigned long zone_end_pfn,
6132 : enum meminit_context context,
6133 : struct vmem_altmap *altmap, int migratetype)
6134 : {
6135 2 : unsigned long pfn, end_pfn = start_pfn + size;
6136 2 : struct page *page;
6137 :
6138 2 : if (highest_memmap_pfn < end_pfn - 1)
6139 2 : highest_memmap_pfn = end_pfn - 1;
6140 :
6141 : #ifdef CONFIG_ZONE_DEVICE
6142 : /*
6143 : * Honor reservation requested by the driver for this ZONE_DEVICE
6144 : * memory. We limit the total number of pages to initialize to just
6145 : * those that might contain the memory mapping. We will defer the
6146 : * ZONE_DEVICE page initialization until after we have released
6147 : * the hotplug lock.
6148 : */
6149 : if (zone == ZONE_DEVICE) {
6150 : if (!altmap)
6151 : return;
6152 :
6153 : if (start_pfn == altmap->base_pfn)
6154 : start_pfn += altmap->reserve;
6155 : end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6156 : }
6157 : #endif
6158 :
6159 262048 : for (pfn = start_pfn; pfn < end_pfn; ) {
6160 : /*
6161 : * There can be holes in boot-time mem_map[]s handed to this
6162 : * function. They do not exist on hotplugged memory.
6163 : */
6164 262046 : if (context == MEMINIT_EARLY) {
6165 262046 : if (overlap_memmap_init(zone, &pfn))
6166 0 : continue;
6167 262046 : if (defer_init(nid, pfn, zone_end_pfn))
6168 : break;
6169 : }
6170 :
6171 262046 : page = pfn_to_page(pfn);
6172 262046 : __init_single_page(page, pfn, zone, nid);
6173 262046 : if (context == MEMINIT_HOTPLUG)
6174 0 : __SetPageReserved(page);
6175 :
6176 : /*
6177 : * Usually, we want to mark the pageblock MIGRATE_MOVABLE,
6178 : * such that unmovable allocations won't be scattered all
6179 : * over the place during system boot.
6180 : */
6181 262046 : if (IS_ALIGNED(pfn, pageblock_nr_pages)) {
6182 255 : set_pageblock_migratetype(page, migratetype);
6183 255 : cond_resched();
6184 : }
6185 262046 : pfn++;
6186 : }
6187 2 : }
6188 :
6189 : #ifdef CONFIG_ZONE_DEVICE
6190 : void __ref memmap_init_zone_device(struct zone *zone,
6191 : unsigned long start_pfn,
6192 : unsigned long nr_pages,
6193 : struct dev_pagemap *pgmap)
6194 : {
6195 : unsigned long pfn, end_pfn = start_pfn + nr_pages;
6196 : struct pglist_data *pgdat = zone->zone_pgdat;
6197 : struct vmem_altmap *altmap = pgmap_altmap(pgmap);
6198 : unsigned long zone_idx = zone_idx(zone);
6199 : unsigned long start = jiffies;
6200 : int nid = pgdat->node_id;
6201 :
6202 : if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE))
6203 : return;
6204 :
6205 : /*
6206 : * The call to memmap_init_zone should have already taken care
6207 : * of the pages reserved for the memmap, so we can just jump to
6208 : * the end of that region and start processing the device pages.
6209 : */
6210 : if (altmap) {
6211 : start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6212 : nr_pages = end_pfn - start_pfn;
6213 : }
6214 :
6215 : for (pfn = start_pfn; pfn < end_pfn; pfn++) {
6216 : struct page *page = pfn_to_page(pfn);
6217 :
6218 : __init_single_page(page, pfn, zone_idx, nid);
6219 :
6220 : /*
6221 : * Mark page reserved as it will need to wait for onlining
6222 : * phase for it to be fully associated with a zone.
6223 : *
6224 : * We can use the non-atomic __set_bit operation for setting
6225 : * the flag as we are still initializing the pages.
6226 : */
6227 : __SetPageReserved(page);
6228 :
6229 : /*
6230 : * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer
6231 : * and zone_device_data. It is a bug if a ZONE_DEVICE page is
6232 : * ever freed or placed on a driver-private list.
6233 : */
6234 : page->pgmap = pgmap;
6235 : page->zone_device_data = NULL;
6236 :
6237 : /*
6238 : * Mark the block movable so that blocks are reserved for
6239 : * movable at startup. This will force kernel allocations
6240 : * to reserve their blocks rather than leaking throughout
6241 : * the address space during boot when many long-lived
6242 : * kernel allocations are made.
6243 : *
6244 : * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
6245 : * because this is done early in section_activate()
6246 : */
6247 : if (IS_ALIGNED(pfn, pageblock_nr_pages)) {
6248 : set_pageblock_migratetype(page, MIGRATE_MOVABLE);
6249 : cond_resched();
6250 : }
6251 : }
6252 :
6253 : pr_info("%s initialised %lu pages in %ums\n", __func__,
6254 : nr_pages, jiffies_to_msecs(jiffies - start));
6255 : }
6256 :
6257 : #endif
6258 1 : static void __meminit zone_init_free_lists(struct zone *zone)
6259 : {
6260 1 : unsigned int order, t;
6261 56 : for_each_migratetype_order(order, t) {
6262 44 : INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
6263 44 : zone->free_area[order].nr_free = 0;
6264 : }
6265 1 : }
6266 :
6267 : #if !defined(CONFIG_FLAT_NODE_MEM_MAP)
6268 : /*
6269 : * Only struct pages that correspond to ranges defined by memblock.memory
6270 : * are zeroed and initialized by going through __init_single_page() during
6271 : * memmap_init_zone().
6272 : *
6273 : * But, there could be struct pages that correspond to holes in
6274 : * memblock.memory. This can happen because of the following reasons:
6275 : * - physical memory bank size is not necessarily the exact multiple of the
6276 : * arbitrary section size
6277 : * - early reserved memory may not be listed in memblock.memory
6278 : * - memory layouts defined with memmap= kernel parameter may not align
6279 : * nicely with memmap sections
6280 : *
6281 : * Explicitly initialize those struct pages so that:
6282 : * - PG_Reserved is set
6283 : * - zone and node links point to zone and node that span the page if the
6284 : * hole is in the middle of a zone
6285 : * - zone and node links point to adjacent zone/node if the hole falls on
6286 : * the zone boundary; the pages in such holes will be prepended to the
6287 : * zone/node above the hole except for the trailing pages in the last
6288 : * section that will be appended to the zone/node below.
6289 : */
6290 2 : static u64 __meminit init_unavailable_range(unsigned long spfn,
6291 : unsigned long epfn,
6292 : int zone, int node)
6293 : {
6294 2 : unsigned long pfn;
6295 2 : u64 pgcnt = 0;
6296 :
6297 100 : for (pfn = spfn; pfn < epfn; pfn++) {
6298 98 : if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6299 0 : pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6300 : + pageblock_nr_pages - 1;
6301 0 : continue;
6302 : }
6303 98 : __init_single_page(pfn_to_page(pfn), pfn, zone, node);
6304 98 : __SetPageReserved(pfn_to_page(pfn));
6305 98 : pgcnt++;
6306 : }
6307 :
6308 2 : return pgcnt;
6309 : }
6310 : #else
6311 : static inline u64 init_unavailable_range(unsigned long spfn, unsigned long epfn,
6312 : int zone, int node)
6313 : {
6314 : return 0;
6315 : }
6316 : #endif
6317 :
6318 1 : void __meminit __weak memmap_init_zone(struct zone *zone)
6319 : {
6320 1 : unsigned long zone_start_pfn = zone->zone_start_pfn;
6321 1 : unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages;
6322 1 : int i, nid = zone_to_nid(zone), zone_id = zone_idx(zone);
6323 1 : static unsigned long hole_pfn;
6324 1 : unsigned long start_pfn, end_pfn;
6325 1 : u64 pgcnt = 0;
6326 :
6327 3 : for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6328 2 : start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn);
6329 2 : end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn);
6330 :
6331 2 : if (end_pfn > start_pfn)
6332 2 : memmap_init_range(end_pfn - start_pfn, nid,
6333 : zone_id, start_pfn, zone_end_pfn,
6334 : MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
6335 :
6336 2 : if (hole_pfn < start_pfn)
6337 2 : pgcnt += init_unavailable_range(hole_pfn, start_pfn,
6338 : zone_id, nid);
6339 2 : hole_pfn = end_pfn;
6340 : }
6341 :
6342 : #ifdef CONFIG_SPARSEMEM
6343 : /*
6344 : * Initialize the hole in the range [zone_end_pfn, section_end].
6345 : * If zone boundary falls in the middle of a section, this hole
6346 : * will be re-initialized during the call to this function for the
6347 : * higher zone.
6348 : */
6349 1 : end_pfn = round_up(zone_end_pfn, PAGES_PER_SECTION);
6350 1 : if (hole_pfn < end_pfn)
6351 0 : pgcnt += init_unavailable_range(hole_pfn, end_pfn,
6352 : zone_id, nid);
6353 : #endif
6354 :
6355 1 : if (pgcnt)
6356 1 : pr_info(" %s zone: %llu pages in unavailable ranges\n",
6357 : zone->name, pgcnt);
6358 1 : }
6359 :
6360 2 : static int zone_batchsize(struct zone *zone)
6361 : {
6362 : #ifdef CONFIG_MMU
6363 2 : int batch;
6364 :
6365 : /*
6366 : * The per-cpu-pages pools are set to around 1000th of the
6367 : * size of the zone.
6368 : */
6369 2 : batch = zone_managed_pages(zone) / 1024;
6370 : /* But no more than a meg. */
6371 2 : if (batch * PAGE_SIZE > 1024 * 1024)
6372 : batch = (1024 * 1024) / PAGE_SIZE;
6373 2 : batch /= 4; /* We effectively *= 4 below */
6374 2 : if (batch < 1)
6375 0 : batch = 1;
6376 :
6377 : /*
6378 : * Clamp the batch to a 2^n - 1 value. Having a power
6379 : * of 2 value was found to be more likely to have
6380 : * suboptimal cache aliasing properties in some cases.
6381 : *
6382 : * For example if 2 tasks are alternately allocating
6383 : * batches of pages, one task can end up with a lot
6384 : * of pages of one half of the possible page colors
6385 : * and the other with pages of the other colors.
6386 : */
6387 2 : batch = rounddown_pow_of_two(batch + batch/2) - 1;
6388 :
6389 2 : return batch;
6390 :
6391 : #else
6392 : /* The deferral and batching of frees should be suppressed under NOMMU
6393 : * conditions.
6394 : *
6395 : * The problem is that NOMMU needs to be able to allocate large chunks
6396 : * of contiguous memory as there's no hardware page translation to
6397 : * assemble apparent contiguous memory from discontiguous pages.
6398 : *
6399 : * Queueing large contiguous runs of pages for batching, however,
6400 : * causes the pages to actually be freed in smaller chunks. As there
6401 : * can be a significant delay between the individual batches being
6402 : * recycled, this leads to the once large chunks of space being
6403 : * fragmented and becoming unavailable for high-order allocations.
6404 : */
6405 : return 0;
6406 : #endif
6407 : }
6408 :
6409 : /*
6410 : * pcp->high and pcp->batch values are related and generally batch is lower
6411 : * than high. They are also related to pcp->count such that count is lower
6412 : * than high, and as soon as it reaches high, the pcplist is flushed.
6413 : *
6414 : * However, guaranteeing these relations at all times would require e.g. write
6415 : * barriers here but also careful usage of read barriers at the read side, and
6416 : * thus be prone to error and bad for performance. Thus the update only prevents
6417 : * store tearing. Any new users of pcp->batch and pcp->high should ensure they
6418 : * can cope with those fields changing asynchronously, and fully trust only the
6419 : * pcp->count field on the local CPU with interrupts disabled.
6420 : *
6421 : * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
6422 : * outside of boot time (or some other assurance that no concurrent updaters
6423 : * exist).
6424 : */
6425 4 : static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
6426 : unsigned long batch)
6427 : {
6428 4 : WRITE_ONCE(pcp->batch, batch);
6429 4 : WRITE_ONCE(pcp->high, high);
6430 4 : }
6431 :
6432 8 : static void pageset_init(struct per_cpu_pageset *p)
6433 : {
6434 8 : struct per_cpu_pages *pcp;
6435 8 : int migratetype;
6436 :
6437 8 : memset(p, 0, sizeof(*p));
6438 :
6439 8 : pcp = &p->pcp;
6440 32 : for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
6441 24 : INIT_LIST_HEAD(&pcp->lists[migratetype]);
6442 :
6443 : /*
6444 : * Set batch and high values safe for a boot pageset. A true percpu
6445 : * pageset's initialization will update them subsequently. Here we don't
6446 : * need to be as careful as pageset_update() as nobody can access the
6447 : * pageset yet.
6448 : */
6449 8 : pcp->high = BOOT_PAGESET_HIGH;
6450 8 : pcp->batch = BOOT_PAGESET_BATCH;
6451 8 : }
6452 :
6453 1 : static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high,
6454 : unsigned long batch)
6455 : {
6456 1 : struct per_cpu_pageset *p;
6457 1 : int cpu;
6458 :
6459 6 : for_each_possible_cpu(cpu) {
6460 4 : p = per_cpu_ptr(zone->pageset, cpu);
6461 5 : pageset_update(&p->pcp, high, batch);
6462 : }
6463 1 : }
6464 :
6465 : /*
6466 : * Calculate and set new high and batch values for all per-cpu pagesets of a
6467 : * zone, based on the zone's size and the percpu_pagelist_fraction sysctl.
6468 : */
6469 1 : static void zone_set_pageset_high_and_batch(struct zone *zone)
6470 : {
6471 1 : unsigned long new_high, new_batch;
6472 :
6473 1 : if (percpu_pagelist_fraction) {
6474 0 : new_high = zone_managed_pages(zone) / percpu_pagelist_fraction;
6475 0 : new_batch = max(1UL, new_high / 4);
6476 0 : if ((new_high / 4) > (PAGE_SHIFT * 8))
6477 0 : new_batch = PAGE_SHIFT * 8;
6478 : } else {
6479 1 : new_batch = zone_batchsize(zone);
6480 1 : new_high = 6 * new_batch;
6481 1 : new_batch = max(1UL, 1 * new_batch);
6482 : }
6483 :
6484 1 : if (zone->pageset_high == new_high &&
6485 0 : zone->pageset_batch == new_batch)
6486 : return;
6487 :
6488 1 : zone->pageset_high = new_high;
6489 1 : zone->pageset_batch = new_batch;
6490 :
6491 1 : __zone_set_pageset_high_and_batch(zone, new_high, new_batch);
6492 : }
6493 :
6494 1 : void __meminit setup_zone_pageset(struct zone *zone)
6495 : {
6496 1 : struct per_cpu_pageset *p;
6497 1 : int cpu;
6498 :
6499 1 : zone->pageset = alloc_percpu(struct per_cpu_pageset);
6500 5 : for_each_possible_cpu(cpu) {
6501 4 : p = per_cpu_ptr(zone->pageset, cpu);
6502 4 : pageset_init(p);
6503 : }
6504 :
6505 1 : zone_set_pageset_high_and_batch(zone);
6506 1 : }
6507 :
6508 : /*
6509 : * Allocate per cpu pagesets and initialize them.
6510 : * Before this call only boot pagesets were available.
6511 : */
6512 1 : void __init setup_per_cpu_pageset(void)
6513 : {
6514 1 : struct pglist_data *pgdat;
6515 1 : struct zone *zone;
6516 1 : int __maybe_unused cpu;
6517 :
6518 4 : for_each_populated_zone(zone)
6519 1 : setup_zone_pageset(zone);
6520 :
6521 : #ifdef CONFIG_NUMA
6522 : /*
6523 : * Unpopulated zones continue using the boot pagesets.
6524 : * The numa stats for these pagesets need to be reset.
6525 : * Otherwise, they will end up skewing the stats of
6526 : * the nodes these zones are associated with.
6527 : */
6528 5 : for_each_possible_cpu(cpu) {
6529 4 : struct per_cpu_pageset *pcp = &per_cpu(boot_pageset, cpu);
6530 4 : memset(pcp->vm_numa_stat_diff, 0,
6531 : sizeof(pcp->vm_numa_stat_diff));
6532 : }
6533 : #endif
6534 :
6535 2 : for_each_online_pgdat(pgdat)
6536 1 : pgdat->per_cpu_nodestats =
6537 1 : alloc_percpu(struct per_cpu_nodestat);
6538 1 : }
6539 :
6540 3 : static __meminit void zone_pcp_init(struct zone *zone)
6541 : {
6542 : /*
6543 : * per cpu subsystem is not up at this point. The following code
6544 : * relies on the ability of the linker to provide the
6545 : * offset of a (static) per cpu variable into the per cpu area.
6546 : */
6547 3 : zone->pageset = &boot_pageset;
6548 3 : zone->pageset_high = BOOT_PAGESET_HIGH;
6549 3 : zone->pageset_batch = BOOT_PAGESET_BATCH;
6550 :
6551 3 : if (populated_zone(zone))
6552 1 : printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
6553 : zone->name, zone->present_pages,
6554 : zone_batchsize(zone));
6555 3 : }
6556 :
6557 1 : void __meminit init_currently_empty_zone(struct zone *zone,
6558 : unsigned long zone_start_pfn,
6559 : unsigned long size)
6560 : {
6561 1 : struct pglist_data *pgdat = zone->zone_pgdat;
6562 1 : int zone_idx = zone_idx(zone) + 1;
6563 :
6564 1 : if (zone_idx > pgdat->nr_zones)
6565 1 : pgdat->nr_zones = zone_idx;
6566 :
6567 1 : zone->zone_start_pfn = zone_start_pfn;
6568 :
6569 1 : mminit_dprintk(MMINIT_TRACE, "memmap_init",
6570 : "Initialising map node %d zone %lu pfns %lu -> %lu\n",
6571 : pgdat->node_id,
6572 : (unsigned long)zone_idx(zone),
6573 : zone_start_pfn, (zone_start_pfn + size));
6574 :
6575 1 : zone_init_free_lists(zone);
6576 1 : zone->initialized = 1;
6577 1 : }
6578 :
6579 : /**
6580 : * get_pfn_range_for_nid - Return the start and end page frames for a node
6581 : * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
6582 : * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
6583 : * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
6584 : *
6585 : * It returns the start and end page frame of a node based on information
6586 : * provided by memblock_set_node(). If called for a node
6587 : * with no available memory, a warning is printed and the start and end
6588 : * PFNs will be 0.
6589 : */
6590 1 : void __init get_pfn_range_for_nid(unsigned int nid,
6591 : unsigned long *start_pfn, unsigned long *end_pfn)
6592 : {
6593 1 : unsigned long this_start_pfn, this_end_pfn;
6594 1 : int i;
6595 :
6596 1 : *start_pfn = -1UL;
6597 1 : *end_pfn = 0;
6598 :
6599 3 : for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
6600 2 : *start_pfn = min(*start_pfn, this_start_pfn);
6601 2 : *end_pfn = max(*end_pfn, this_end_pfn);
6602 : }
6603 :
6604 1 : if (*start_pfn == -1UL)
6605 0 : *start_pfn = 0;
6606 1 : }
6607 :
6608 : /*
6609 : * This finds a zone that can be used for ZONE_MOVABLE pages. The
6610 : * assumption is made that zones within a node are ordered in monotonic
6611 : * increasing memory addresses so that the "highest" populated zone is used
6612 : */
6613 1 : static void __init find_usable_zone_for_movable(void)
6614 : {
6615 1 : int zone_index;
6616 3 : for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
6617 3 : if (zone_index == ZONE_MOVABLE)
6618 1 : continue;
6619 :
6620 2 : if (arch_zone_highest_possible_pfn[zone_index] >
6621 2 : arch_zone_lowest_possible_pfn[zone_index])
6622 : break;
6623 : }
6624 :
6625 1 : VM_BUG_ON(zone_index == -1);
6626 1 : movable_zone = zone_index;
6627 1 : }
6628 :
6629 : /*
6630 : * The zone ranges provided by the architecture do not include ZONE_MOVABLE
6631 : * because it is sized independent of architecture. Unlike the other zones,
6632 : * the starting point for ZONE_MOVABLE is not fixed. It may be different
6633 : * in each node depending on the size of each node and how evenly kernelcore
6634 : * is distributed. This helper function adjusts the zone ranges
6635 : * provided by the architecture for a given node by using the end of the
6636 : * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
6637 : * zones within a node are in order of monotonic increases memory addresses
6638 : */
6639 6 : static void __init adjust_zone_range_for_zone_movable(int nid,
6640 : unsigned long zone_type,
6641 : unsigned long node_start_pfn,
6642 : unsigned long node_end_pfn,
6643 : unsigned long *zone_start_pfn,
6644 : unsigned long *zone_end_pfn)
6645 : {
6646 : /* Only adjust if ZONE_MOVABLE is on this node */
6647 6 : if (zone_movable_pfn[nid]) {
6648 : /* Size ZONE_MOVABLE */
6649 0 : if (zone_type == ZONE_MOVABLE) {
6650 0 : *zone_start_pfn = zone_movable_pfn[nid];
6651 0 : *zone_end_pfn = min(node_end_pfn,
6652 : arch_zone_highest_possible_pfn[movable_zone]);
6653 :
6654 : /* Adjust for ZONE_MOVABLE starting within this range */
6655 0 : } else if (!mirrored_kernelcore &&
6656 0 : *zone_start_pfn < zone_movable_pfn[nid] &&
6657 0 : *zone_end_pfn > zone_movable_pfn[nid]) {
6658 0 : *zone_end_pfn = zone_movable_pfn[nid];
6659 :
6660 : /* Check if this whole range is within ZONE_MOVABLE */
6661 0 : } else if (*zone_start_pfn >= zone_movable_pfn[nid])
6662 0 : *zone_start_pfn = *zone_end_pfn;
6663 : }
6664 6 : }
6665 :
6666 : /*
6667 : * Return the number of pages a zone spans in a node, including holes
6668 : * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
6669 : */
6670 3 : static unsigned long __init zone_spanned_pages_in_node(int nid,
6671 : unsigned long zone_type,
6672 : unsigned long node_start_pfn,
6673 : unsigned long node_end_pfn,
6674 : unsigned long *zone_start_pfn,
6675 : unsigned long *zone_end_pfn)
6676 : {
6677 3 : unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6678 3 : unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6679 : /* When hotadd a new node from cpu_up(), the node should be empty */
6680 3 : if (!node_start_pfn && !node_end_pfn)
6681 : return 0;
6682 :
6683 : /* Get the start and end of the zone */
6684 3 : *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6685 3 : *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6686 3 : adjust_zone_range_for_zone_movable(nid, zone_type,
6687 : node_start_pfn, node_end_pfn,
6688 : zone_start_pfn, zone_end_pfn);
6689 :
6690 : /* Check that this node has pages within the zone's required range */
6691 3 : if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
6692 : return 0;
6693 :
6694 : /* Move the zone boundaries inside the node if necessary */
6695 2 : *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
6696 2 : *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
6697 :
6698 : /* Return the spanned pages */
6699 2 : return *zone_end_pfn - *zone_start_pfn;
6700 : }
6701 :
6702 : /*
6703 : * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
6704 : * then all holes in the requested range will be accounted for.
6705 : */
6706 5 : unsigned long __init __absent_pages_in_range(int nid,
6707 : unsigned long range_start_pfn,
6708 : unsigned long range_end_pfn)
6709 : {
6710 5 : unsigned long nr_absent = range_end_pfn - range_start_pfn;
6711 5 : unsigned long start_pfn, end_pfn;
6712 5 : int i;
6713 :
6714 15 : for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6715 10 : start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
6716 10 : end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
6717 10 : nr_absent -= end_pfn - start_pfn;
6718 : }
6719 5 : return nr_absent;
6720 : }
6721 :
6722 : /**
6723 : * absent_pages_in_range - Return number of page frames in holes within a range
6724 : * @start_pfn: The start PFN to start searching for holes
6725 : * @end_pfn: The end PFN to stop searching for holes
6726 : *
6727 : * Return: the number of pages frames in memory holes within a range.
6728 : */
6729 1 : unsigned long __init absent_pages_in_range(unsigned long start_pfn,
6730 : unsigned long end_pfn)
6731 : {
6732 1 : return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
6733 : }
6734 :
6735 : /* Return the number of page frames in holes in a zone on a node */
6736 3 : static unsigned long __init zone_absent_pages_in_node(int nid,
6737 : unsigned long zone_type,
6738 : unsigned long node_start_pfn,
6739 : unsigned long node_end_pfn)
6740 : {
6741 3 : unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6742 3 : unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6743 3 : unsigned long zone_start_pfn, zone_end_pfn;
6744 3 : unsigned long nr_absent;
6745 :
6746 : /* When hotadd a new node from cpu_up(), the node should be empty */
6747 3 : if (!node_start_pfn && !node_end_pfn)
6748 : return 0;
6749 :
6750 3 : zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6751 3 : zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6752 :
6753 3 : adjust_zone_range_for_zone_movable(nid, zone_type,
6754 : node_start_pfn, node_end_pfn,
6755 : &zone_start_pfn, &zone_end_pfn);
6756 3 : nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
6757 :
6758 : /*
6759 : * ZONE_MOVABLE handling.
6760 : * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
6761 : * and vice versa.
6762 : */
6763 3 : if (mirrored_kernelcore && zone_movable_pfn[nid]) {
6764 0 : unsigned long start_pfn, end_pfn;
6765 0 : struct memblock_region *r;
6766 :
6767 0 : for_each_mem_region(r) {
6768 0 : start_pfn = clamp(memblock_region_memory_base_pfn(r),
6769 : zone_start_pfn, zone_end_pfn);
6770 0 : end_pfn = clamp(memblock_region_memory_end_pfn(r),
6771 : zone_start_pfn, zone_end_pfn);
6772 :
6773 0 : if (zone_type == ZONE_MOVABLE &&
6774 0 : memblock_is_mirror(r))
6775 0 : nr_absent += end_pfn - start_pfn;
6776 :
6777 0 : if (zone_type == ZONE_NORMAL &&
6778 0 : !memblock_is_mirror(r))
6779 0 : nr_absent += end_pfn - start_pfn;
6780 : }
6781 : }
6782 :
6783 : return nr_absent;
6784 : }
6785 :
6786 1 : static void __init calculate_node_totalpages(struct pglist_data *pgdat,
6787 : unsigned long node_start_pfn,
6788 : unsigned long node_end_pfn)
6789 : {
6790 1 : unsigned long realtotalpages = 0, totalpages = 0;
6791 1 : enum zone_type i;
6792 :
6793 4 : for (i = 0; i < MAX_NR_ZONES; i++) {
6794 3 : struct zone *zone = pgdat->node_zones + i;
6795 3 : unsigned long zone_start_pfn, zone_end_pfn;
6796 3 : unsigned long spanned, absent;
6797 3 : unsigned long size, real_size;
6798 :
6799 3 : spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
6800 : node_start_pfn,
6801 : node_end_pfn,
6802 : &zone_start_pfn,
6803 : &zone_end_pfn);
6804 3 : absent = zone_absent_pages_in_node(pgdat->node_id, i,
6805 : node_start_pfn,
6806 : node_end_pfn);
6807 :
6808 3 : size = spanned;
6809 3 : real_size = size - absent;
6810 :
6811 3 : if (size)
6812 1 : zone->zone_start_pfn = zone_start_pfn;
6813 : else
6814 2 : zone->zone_start_pfn = 0;
6815 3 : zone->spanned_pages = size;
6816 3 : zone->present_pages = real_size;
6817 :
6818 3 : totalpages += size;
6819 3 : realtotalpages += real_size;
6820 : }
6821 :
6822 1 : pgdat->node_spanned_pages = totalpages;
6823 1 : pgdat->node_present_pages = realtotalpages;
6824 1 : printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
6825 : realtotalpages);
6826 1 : }
6827 :
6828 : #ifndef CONFIG_SPARSEMEM
6829 : /*
6830 : * Calculate the size of the zone->blockflags rounded to an unsigned long
6831 : * Start by making sure zonesize is a multiple of pageblock_order by rounding
6832 : * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
6833 : * round what is now in bits to nearest long in bits, then return it in
6834 : * bytes.
6835 : */
6836 : static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
6837 : {
6838 : unsigned long usemapsize;
6839 :
6840 : zonesize += zone_start_pfn & (pageblock_nr_pages-1);
6841 : usemapsize = roundup(zonesize, pageblock_nr_pages);
6842 : usemapsize = usemapsize >> pageblock_order;
6843 : usemapsize *= NR_PAGEBLOCK_BITS;
6844 : usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
6845 :
6846 : return usemapsize / 8;
6847 : }
6848 :
6849 : static void __ref setup_usemap(struct zone *zone)
6850 : {
6851 : unsigned long usemapsize = usemap_size(zone->zone_start_pfn,
6852 : zone->spanned_pages);
6853 : zone->pageblock_flags = NULL;
6854 : if (usemapsize) {
6855 : zone->pageblock_flags =
6856 : memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
6857 : zone_to_nid(zone));
6858 : if (!zone->pageblock_flags)
6859 : panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
6860 : usemapsize, zone->name, zone_to_nid(zone));
6861 : }
6862 : }
6863 : #else
6864 1 : static inline void setup_usemap(struct zone *zone) {}
6865 : #endif /* CONFIG_SPARSEMEM */
6866 :
6867 : #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
6868 :
6869 : /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
6870 : void __init set_pageblock_order(void)
6871 : {
6872 : unsigned int order;
6873 :
6874 : /* Check that pageblock_nr_pages has not already been setup */
6875 : if (pageblock_order)
6876 : return;
6877 :
6878 : if (HPAGE_SHIFT > PAGE_SHIFT)
6879 : order = HUGETLB_PAGE_ORDER;
6880 : else
6881 : order = MAX_ORDER - 1;
6882 :
6883 : /*
6884 : * Assume the largest contiguous order of interest is a huge page.
6885 : * This value may be variable depending on boot parameters on IA64 and
6886 : * powerpc.
6887 : */
6888 : pageblock_order = order;
6889 : }
6890 : #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6891 :
6892 : /*
6893 : * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
6894 : * is unused as pageblock_order is set at compile-time. See
6895 : * include/linux/pageblock-flags.h for the values of pageblock_order based on
6896 : * the kernel config
6897 : */
6898 2 : void __init set_pageblock_order(void)
6899 : {
6900 2 : }
6901 :
6902 : #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6903 :
6904 3 : static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
6905 : unsigned long present_pages)
6906 : {
6907 3 : unsigned long pages = spanned_pages;
6908 :
6909 : /*
6910 : * Provide a more accurate estimation if there are holes within
6911 : * the zone and SPARSEMEM is in use. If there are holes within the
6912 : * zone, each populated memory region may cost us one or two extra
6913 : * memmap pages due to alignment because memmap pages for each
6914 : * populated regions may not be naturally aligned on page boundary.
6915 : * So the (present_pages >> 4) heuristic is a tradeoff for that.
6916 : */
6917 3 : if (spanned_pages > present_pages + (present_pages >> 4) &&
6918 : IS_ENABLED(CONFIG_SPARSEMEM))
6919 0 : pages = present_pages;
6920 :
6921 3 : return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
6922 : }
6923 :
6924 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6925 1 : static void pgdat_init_split_queue(struct pglist_data *pgdat)
6926 : {
6927 1 : struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
6928 :
6929 1 : spin_lock_init(&ds_queue->split_queue_lock);
6930 1 : INIT_LIST_HEAD(&ds_queue->split_queue);
6931 1 : ds_queue->split_queue_len = 0;
6932 1 : }
6933 : #else
6934 : static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
6935 : #endif
6936 :
6937 : #ifdef CONFIG_COMPACTION
6938 1 : static void pgdat_init_kcompactd(struct pglist_data *pgdat)
6939 : {
6940 2 : init_waitqueue_head(&pgdat->kcompactd_wait);
6941 : }
6942 : #else
6943 : static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
6944 : #endif
6945 :
6946 1 : static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
6947 : {
6948 1 : pgdat_resize_init(pgdat);
6949 :
6950 1 : pgdat_init_split_queue(pgdat);
6951 1 : pgdat_init_kcompactd(pgdat);
6952 :
6953 1 : init_waitqueue_head(&pgdat->kswapd_wait);
6954 1 : init_waitqueue_head(&pgdat->pfmemalloc_wait);
6955 :
6956 1 : pgdat_page_ext_init(pgdat);
6957 1 : lruvec_init(&pgdat->__lruvec);
6958 1 : }
6959 :
6960 3 : static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
6961 : unsigned long remaining_pages)
6962 : {
6963 3 : atomic_long_set(&zone->managed_pages, remaining_pages);
6964 3 : zone_set_nid(zone, nid);
6965 3 : zone->name = zone_names[idx];
6966 3 : zone->zone_pgdat = NODE_DATA(nid);
6967 3 : spin_lock_init(&zone->lock);
6968 3 : zone_seqlock_init(zone);
6969 3 : zone_pcp_init(zone);
6970 3 : }
6971 :
6972 : /*
6973 : * Set up the zone data structures
6974 : * - init pgdat internals
6975 : * - init all zones belonging to this node
6976 : *
6977 : * NOTE: this function is only called during memory hotplug
6978 : */
6979 : #ifdef CONFIG_MEMORY_HOTPLUG
6980 : void __ref free_area_init_core_hotplug(int nid)
6981 : {
6982 : enum zone_type z;
6983 : pg_data_t *pgdat = NODE_DATA(nid);
6984 :
6985 : pgdat_init_internals(pgdat);
6986 : for (z = 0; z < MAX_NR_ZONES; z++)
6987 : zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
6988 : }
6989 : #endif
6990 :
6991 : /*
6992 : * Set up the zone data structures:
6993 : * - mark all pages reserved
6994 : * - mark all memory queues empty
6995 : * - clear the memory bitmaps
6996 : *
6997 : * NOTE: pgdat should get zeroed by caller.
6998 : * NOTE: this function is only called during early init.
6999 : */
7000 1 : static void __init free_area_init_core(struct pglist_data *pgdat)
7001 : {
7002 1 : enum zone_type j;
7003 1 : int nid = pgdat->node_id;
7004 :
7005 1 : pgdat_init_internals(pgdat);
7006 1 : pgdat->per_cpu_nodestats = &boot_nodestats;
7007 :
7008 4 : for (j = 0; j < MAX_NR_ZONES; j++) {
7009 3 : struct zone *zone = pgdat->node_zones + j;
7010 3 : unsigned long size, freesize, memmap_pages;
7011 :
7012 3 : size = zone->spanned_pages;
7013 3 : freesize = zone->present_pages;
7014 :
7015 : /*
7016 : * Adjust freesize so that it accounts for how much memory
7017 : * is used by this zone for memmap. This affects the watermark
7018 : * and per-cpu initialisations
7019 : */
7020 3 : memmap_pages = calc_memmap_size(size, freesize);
7021 3 : if (!is_highmem_idx(j)) {
7022 3 : if (freesize >= memmap_pages) {
7023 3 : freesize -= memmap_pages;
7024 3 : if (memmap_pages)
7025 1 : printk(KERN_DEBUG
7026 : " %s zone: %lu pages used for memmap\n",
7027 : zone_names[j], memmap_pages);
7028 : } else
7029 0 : pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
7030 : zone_names[j], memmap_pages, freesize);
7031 : }
7032 :
7033 : /* Account for reserved pages */
7034 3 : if (j == 0 && freesize > dma_reserve) {
7035 1 : freesize -= dma_reserve;
7036 1 : printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
7037 : zone_names[0], dma_reserve);
7038 : }
7039 :
7040 3 : if (!is_highmem_idx(j))
7041 3 : nr_kernel_pages += freesize;
7042 : /* Charge for highmem memmap if there are enough kernel pages */
7043 : else if (nr_kernel_pages > memmap_pages * 2)
7044 : nr_kernel_pages -= memmap_pages;
7045 3 : nr_all_pages += freesize;
7046 :
7047 : /*
7048 : * Set an approximate value for lowmem here, it will be adjusted
7049 : * when the bootmem allocator frees pages into the buddy system.
7050 : * And all highmem pages will be managed by the buddy system.
7051 : */
7052 3 : zone_init_internals(zone, j, nid, freesize);
7053 :
7054 3 : if (!size)
7055 2 : continue;
7056 :
7057 1 : set_pageblock_order();
7058 1 : setup_usemap(zone);
7059 1 : init_currently_empty_zone(zone, zone->zone_start_pfn, size);
7060 1 : memmap_init_zone(zone);
7061 : }
7062 1 : }
7063 :
7064 : #ifdef CONFIG_FLAT_NODE_MEM_MAP
7065 : static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
7066 : {
7067 : unsigned long __maybe_unused start = 0;
7068 : unsigned long __maybe_unused offset = 0;
7069 :
7070 : /* Skip empty nodes */
7071 : if (!pgdat->node_spanned_pages)
7072 : return;
7073 :
7074 : start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
7075 : offset = pgdat->node_start_pfn - start;
7076 : /* ia64 gets its own node_mem_map, before this, without bootmem */
7077 : if (!pgdat->node_mem_map) {
7078 : unsigned long size, end;
7079 : struct page *map;
7080 :
7081 : /*
7082 : * The zone's endpoints aren't required to be MAX_ORDER
7083 : * aligned but the node_mem_map endpoints must be in order
7084 : * for the buddy allocator to function correctly.
7085 : */
7086 : end = pgdat_end_pfn(pgdat);
7087 : end = ALIGN(end, MAX_ORDER_NR_PAGES);
7088 : size = (end - start) * sizeof(struct page);
7089 : map = memblock_alloc_node(size, SMP_CACHE_BYTES,
7090 : pgdat->node_id);
7091 : if (!map)
7092 : panic("Failed to allocate %ld bytes for node %d memory map\n",
7093 : size, pgdat->node_id);
7094 : pgdat->node_mem_map = map + offset;
7095 : }
7096 : pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
7097 : __func__, pgdat->node_id, (unsigned long)pgdat,
7098 : (unsigned long)pgdat->node_mem_map);
7099 : #ifndef CONFIG_NEED_MULTIPLE_NODES
7100 : /*
7101 : * With no DISCONTIG, the global mem_map is just set as node 0's
7102 : */
7103 : if (pgdat == NODE_DATA(0)) {
7104 : mem_map = NODE_DATA(0)->node_mem_map;
7105 : if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
7106 : mem_map -= offset;
7107 : }
7108 : #endif
7109 : }
7110 : #else
7111 : static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
7112 : #endif /* CONFIG_FLAT_NODE_MEM_MAP */
7113 :
7114 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
7115 : static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
7116 : {
7117 : pgdat->first_deferred_pfn = ULONG_MAX;
7118 : }
7119 : #else
7120 1 : static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
7121 : #endif
7122 :
7123 1 : static void __init free_area_init_node(int nid)
7124 : {
7125 1 : pg_data_t *pgdat = NODE_DATA(nid);
7126 1 : unsigned long start_pfn = 0;
7127 1 : unsigned long end_pfn = 0;
7128 :
7129 : /* pg_data_t should be reset to zero when it's allocated */
7130 2 : WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
7131 :
7132 1 : get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
7133 :
7134 1 : pgdat->node_id = nid;
7135 1 : pgdat->node_start_pfn = start_pfn;
7136 1 : pgdat->per_cpu_nodestats = NULL;
7137 :
7138 1 : pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
7139 : (u64)start_pfn << PAGE_SHIFT,
7140 : end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
7141 1 : calculate_node_totalpages(pgdat, start_pfn, end_pfn);
7142 :
7143 1 : alloc_node_mem_map(pgdat);
7144 1 : pgdat_set_deferred_range(pgdat);
7145 :
7146 1 : free_area_init_core(pgdat);
7147 1 : }
7148 :
7149 0 : void __init free_area_init_memoryless_node(int nid)
7150 : {
7151 0 : free_area_init_node(nid);
7152 0 : }
7153 :
7154 : #if MAX_NUMNODES > 1
7155 : /*
7156 : * Figure out the number of possible node ids.
7157 : */
7158 1 : void __init setup_nr_node_ids(void)
7159 : {
7160 1 : unsigned int highest;
7161 :
7162 1 : highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
7163 1 : nr_node_ids = highest + 1;
7164 1 : }
7165 : #endif
7166 :
7167 : /**
7168 : * node_map_pfn_alignment - determine the maximum internode alignment
7169 : *
7170 : * This function should be called after node map is populated and sorted.
7171 : * It calculates the maximum power of two alignment which can distinguish
7172 : * all the nodes.
7173 : *
7174 : * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
7175 : * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
7176 : * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
7177 : * shifted, 1GiB is enough and this function will indicate so.
7178 : *
7179 : * This is used to test whether pfn -> nid mapping of the chosen memory
7180 : * model has fine enough granularity to avoid incorrect mapping for the
7181 : * populated node map.
7182 : *
7183 : * Return: the determined alignment in pfn's. 0 if there is no alignment
7184 : * requirement (single node).
7185 : */
7186 0 : unsigned long __init node_map_pfn_alignment(void)
7187 : {
7188 0 : unsigned long accl_mask = 0, last_end = 0;
7189 0 : unsigned long start, end, mask;
7190 0 : int last_nid = NUMA_NO_NODE;
7191 0 : int i, nid;
7192 :
7193 0 : for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
7194 0 : if (!start || last_nid < 0 || last_nid == nid) {
7195 0 : last_nid = nid;
7196 0 : last_end = end;
7197 0 : continue;
7198 : }
7199 :
7200 : /*
7201 : * Start with a mask granular enough to pin-point to the
7202 : * start pfn and tick off bits one-by-one until it becomes
7203 : * too coarse to separate the current node from the last.
7204 : */
7205 0 : mask = ~((1 << __ffs(start)) - 1);
7206 0 : while (mask && last_end <= (start & (mask << 1)))
7207 : mask <<= 1;
7208 :
7209 : /* accumulate all internode masks */
7210 0 : accl_mask |= mask;
7211 : }
7212 :
7213 : /* convert mask to number of pages */
7214 0 : return ~accl_mask + 1;
7215 : }
7216 :
7217 : /**
7218 : * find_min_pfn_with_active_regions - Find the minimum PFN registered
7219 : *
7220 : * Return: the minimum PFN based on information provided via
7221 : * memblock_set_node().
7222 : */
7223 1 : unsigned long __init find_min_pfn_with_active_regions(void)
7224 : {
7225 1 : return PHYS_PFN(memblock_start_of_DRAM());
7226 : }
7227 :
7228 : /*
7229 : * early_calculate_totalpages()
7230 : * Sum pages in active regions for movable zone.
7231 : * Populate N_MEMORY for calculating usable_nodes.
7232 : */
7233 1 : static unsigned long __init early_calculate_totalpages(void)
7234 : {
7235 1 : unsigned long totalpages = 0;
7236 1 : unsigned long start_pfn, end_pfn;
7237 1 : int i, nid;
7238 :
7239 3 : for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7240 2 : unsigned long pages = end_pfn - start_pfn;
7241 :
7242 2 : totalpages += pages;
7243 2 : if (pages)
7244 4 : node_set_state(nid, N_MEMORY);
7245 : }
7246 1 : return totalpages;
7247 : }
7248 :
7249 : /*
7250 : * Find the PFN the Movable zone begins in each node. Kernel memory
7251 : * is spread evenly between nodes as long as the nodes have enough
7252 : * memory. When they don't, some nodes will have more kernelcore than
7253 : * others
7254 : */
7255 1 : static void __init find_zone_movable_pfns_for_nodes(void)
7256 : {
7257 1 : int i, nid;
7258 1 : unsigned long usable_startpfn;
7259 1 : unsigned long kernelcore_node, kernelcore_remaining;
7260 : /* save the state before borrow the nodemask */
7261 1 : nodemask_t saved_node_state = node_states[N_MEMORY];
7262 1 : unsigned long totalpages = early_calculate_totalpages();
7263 1 : int usable_nodes = nodes_weight(node_states[N_MEMORY]);
7264 1 : struct memblock_region *r;
7265 :
7266 : /* Need to find movable_zone earlier when movable_node is specified. */
7267 1 : find_usable_zone_for_movable();
7268 :
7269 : /*
7270 : * If movable_node is specified, ignore kernelcore and movablecore
7271 : * options.
7272 : */
7273 1 : if (movable_node_is_enabled()) {
7274 : for_each_mem_region(r) {
7275 : if (!memblock_is_hotpluggable(r))
7276 : continue;
7277 :
7278 : nid = memblock_get_region_node(r);
7279 :
7280 : usable_startpfn = PFN_DOWN(r->base);
7281 : zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7282 : min(usable_startpfn, zone_movable_pfn[nid]) :
7283 : usable_startpfn;
7284 : }
7285 :
7286 : goto out2;
7287 : }
7288 :
7289 : /*
7290 : * If kernelcore=mirror is specified, ignore movablecore option
7291 : */
7292 1 : if (mirrored_kernelcore) {
7293 0 : bool mem_below_4gb_not_mirrored = false;
7294 :
7295 0 : for_each_mem_region(r) {
7296 0 : if (memblock_is_mirror(r))
7297 0 : continue;
7298 :
7299 0 : nid = memblock_get_region_node(r);
7300 :
7301 0 : usable_startpfn = memblock_region_memory_base_pfn(r);
7302 :
7303 0 : if (usable_startpfn < 0x100000) {
7304 0 : mem_below_4gb_not_mirrored = true;
7305 0 : continue;
7306 : }
7307 :
7308 0 : zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7309 0 : min(usable_startpfn, zone_movable_pfn[nid]) :
7310 : usable_startpfn;
7311 : }
7312 :
7313 0 : if (mem_below_4gb_not_mirrored)
7314 0 : pr_warn("This configuration results in unmirrored kernel memory.\n");
7315 :
7316 0 : goto out2;
7317 : }
7318 :
7319 : /*
7320 : * If kernelcore=nn% or movablecore=nn% was specified, calculate the
7321 : * amount of necessary memory.
7322 : */
7323 1 : if (required_kernelcore_percent)
7324 0 : required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
7325 : 10000UL;
7326 1 : if (required_movablecore_percent)
7327 0 : required_movablecore = (totalpages * 100 * required_movablecore_percent) /
7328 : 10000UL;
7329 :
7330 : /*
7331 : * If movablecore= was specified, calculate what size of
7332 : * kernelcore that corresponds so that memory usable for
7333 : * any allocation type is evenly spread. If both kernelcore
7334 : * and movablecore are specified, then the value of kernelcore
7335 : * will be used for required_kernelcore if it's greater than
7336 : * what movablecore would have allowed.
7337 : */
7338 1 : if (required_movablecore) {
7339 0 : unsigned long corepages;
7340 :
7341 : /*
7342 : * Round-up so that ZONE_MOVABLE is at least as large as what
7343 : * was requested by the user
7344 : */
7345 0 : required_movablecore =
7346 0 : roundup(required_movablecore, MAX_ORDER_NR_PAGES);
7347 0 : required_movablecore = min(totalpages, required_movablecore);
7348 0 : corepages = totalpages - required_movablecore;
7349 :
7350 0 : required_kernelcore = max(required_kernelcore, corepages);
7351 : }
7352 :
7353 : /*
7354 : * If kernelcore was not specified or kernelcore size is larger
7355 : * than totalpages, there is no ZONE_MOVABLE.
7356 : */
7357 1 : if (!required_kernelcore || required_kernelcore >= totalpages)
7358 1 : goto out;
7359 :
7360 : /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
7361 0 : usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
7362 :
7363 0 : restart:
7364 : /* Spread kernelcore memory as evenly as possible throughout nodes */
7365 0 : kernelcore_node = required_kernelcore / usable_nodes;
7366 0 : for_each_node_state(nid, N_MEMORY) {
7367 0 : unsigned long start_pfn, end_pfn;
7368 :
7369 : /*
7370 : * Recalculate kernelcore_node if the division per node
7371 : * now exceeds what is necessary to satisfy the requested
7372 : * amount of memory for the kernel
7373 : */
7374 0 : if (required_kernelcore < kernelcore_node)
7375 0 : kernelcore_node = required_kernelcore / usable_nodes;
7376 :
7377 : /*
7378 : * As the map is walked, we track how much memory is usable
7379 : * by the kernel using kernelcore_remaining. When it is
7380 : * 0, the rest of the node is usable by ZONE_MOVABLE
7381 : */
7382 0 : kernelcore_remaining = kernelcore_node;
7383 :
7384 : /* Go through each range of PFNs within this node */
7385 0 : for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7386 0 : unsigned long size_pages;
7387 :
7388 0 : start_pfn = max(start_pfn, zone_movable_pfn[nid]);
7389 0 : if (start_pfn >= end_pfn)
7390 0 : continue;
7391 :
7392 : /* Account for what is only usable for kernelcore */
7393 0 : if (start_pfn < usable_startpfn) {
7394 0 : unsigned long kernel_pages;
7395 0 : kernel_pages = min(end_pfn, usable_startpfn)
7396 : - start_pfn;
7397 :
7398 0 : kernelcore_remaining -= min(kernel_pages,
7399 : kernelcore_remaining);
7400 0 : required_kernelcore -= min(kernel_pages,
7401 : required_kernelcore);
7402 :
7403 : /* Continue if range is now fully accounted */
7404 0 : if (end_pfn <= usable_startpfn) {
7405 :
7406 : /*
7407 : * Push zone_movable_pfn to the end so
7408 : * that if we have to rebalance
7409 : * kernelcore across nodes, we will
7410 : * not double account here
7411 : */
7412 0 : zone_movable_pfn[nid] = end_pfn;
7413 0 : continue;
7414 : }
7415 0 : start_pfn = usable_startpfn;
7416 : }
7417 :
7418 : /*
7419 : * The usable PFN range for ZONE_MOVABLE is from
7420 : * start_pfn->end_pfn. Calculate size_pages as the
7421 : * number of pages used as kernelcore
7422 : */
7423 0 : size_pages = end_pfn - start_pfn;
7424 0 : if (size_pages > kernelcore_remaining)
7425 : size_pages = kernelcore_remaining;
7426 0 : zone_movable_pfn[nid] = start_pfn + size_pages;
7427 :
7428 : /*
7429 : * Some kernelcore has been met, update counts and
7430 : * break if the kernelcore for this node has been
7431 : * satisfied
7432 : */
7433 0 : required_kernelcore -= min(required_kernelcore,
7434 : size_pages);
7435 0 : kernelcore_remaining -= size_pages;
7436 0 : if (!kernelcore_remaining)
7437 : break;
7438 : }
7439 : }
7440 :
7441 : /*
7442 : * If there is still required_kernelcore, we do another pass with one
7443 : * less node in the count. This will push zone_movable_pfn[nid] further
7444 : * along on the nodes that still have memory until kernelcore is
7445 : * satisfied
7446 : */
7447 0 : usable_nodes--;
7448 0 : if (usable_nodes && required_kernelcore > usable_nodes)
7449 0 : goto restart;
7450 :
7451 0 : out2:
7452 : /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
7453 0 : for (nid = 0; nid < MAX_NUMNODES; nid++)
7454 0 : zone_movable_pfn[nid] =
7455 0 : roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
7456 :
7457 0 : out:
7458 : /* restore the node_state */
7459 1 : node_states[N_MEMORY] = saved_node_state;
7460 1 : }
7461 :
7462 : /* Any regular or high memory on that node ? */
7463 1 : static void check_for_memory(pg_data_t *pgdat, int nid)
7464 : {
7465 1 : enum zone_type zone_type;
7466 :
7467 1 : for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
7468 1 : struct zone *zone = &pgdat->node_zones[zone_type];
7469 1 : if (populated_zone(zone)) {
7470 1 : if (IS_ENABLED(CONFIG_HIGHMEM))
7471 : node_set_state(nid, N_HIGH_MEMORY);
7472 1 : if (zone_type <= ZONE_NORMAL)
7473 1 : node_set_state(nid, N_NORMAL_MEMORY);
7474 : break;
7475 : }
7476 : }
7477 1 : }
7478 :
7479 : /*
7480 : * Some architecturs, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For
7481 : * such cases we allow max_zone_pfn sorted in the descending order
7482 : */
7483 1 : bool __weak arch_has_descending_max_zone_pfns(void)
7484 : {
7485 1 : return false;
7486 : }
7487 :
7488 : /**
7489 : * free_area_init - Initialise all pg_data_t and zone data
7490 : * @max_zone_pfn: an array of max PFNs for each zone
7491 : *
7492 : * This will call free_area_init_node() for each active node in the system.
7493 : * Using the page ranges provided by memblock_set_node(), the size of each
7494 : * zone in each node and their holes is calculated. If the maximum PFN
7495 : * between two adjacent zones match, it is assumed that the zone is empty.
7496 : * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
7497 : * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
7498 : * starts where the previous one ended. For example, ZONE_DMA32 starts
7499 : * at arch_max_dma_pfn.
7500 : */
7501 1 : void __init free_area_init(unsigned long *max_zone_pfn)
7502 : {
7503 1 : unsigned long start_pfn, end_pfn;
7504 1 : int i, nid, zone;
7505 1 : bool descending;
7506 :
7507 : /* Record where the zone boundaries are */
7508 1 : memset(arch_zone_lowest_possible_pfn, 0,
7509 : sizeof(arch_zone_lowest_possible_pfn));
7510 1 : memset(arch_zone_highest_possible_pfn, 0,
7511 : sizeof(arch_zone_highest_possible_pfn));
7512 :
7513 1 : start_pfn = find_min_pfn_with_active_regions();
7514 1 : descending = arch_has_descending_max_zone_pfns();
7515 :
7516 4 : for (i = 0; i < MAX_NR_ZONES; i++) {
7517 3 : if (descending)
7518 0 : zone = MAX_NR_ZONES - i - 1;
7519 : else
7520 : zone = i;
7521 :
7522 3 : if (zone == ZONE_MOVABLE)
7523 1 : continue;
7524 :
7525 2 : end_pfn = max(max_zone_pfn[zone], start_pfn);
7526 2 : arch_zone_lowest_possible_pfn[zone] = start_pfn;
7527 2 : arch_zone_highest_possible_pfn[zone] = end_pfn;
7528 :
7529 2 : start_pfn = end_pfn;
7530 : }
7531 :
7532 : /* Find the PFNs that ZONE_MOVABLE begins at in each node */
7533 1 : memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
7534 1 : find_zone_movable_pfns_for_nodes();
7535 :
7536 : /* Print out the zone ranges */
7537 1 : pr_info("Zone ranges:\n");
7538 4 : for (i = 0; i < MAX_NR_ZONES; i++) {
7539 3 : if (i == ZONE_MOVABLE)
7540 1 : continue;
7541 2 : pr_info(" %-8s ", zone_names[i]);
7542 2 : if (arch_zone_lowest_possible_pfn[i] ==
7543 2 : arch_zone_highest_possible_pfn[i])
7544 1 : pr_cont("empty\n");
7545 : else
7546 1 : pr_cont("[mem %#018Lx-%#018Lx]\n",
7547 : (u64)arch_zone_lowest_possible_pfn[i]
7548 : << PAGE_SHIFT,
7549 : ((u64)arch_zone_highest_possible_pfn[i]
7550 : << PAGE_SHIFT) - 1);
7551 : }
7552 :
7553 : /* Print out the PFNs ZONE_MOVABLE begins at in each node */
7554 1 : pr_info("Movable zone start for each node\n");
7555 65 : for (i = 0; i < MAX_NUMNODES; i++) {
7556 64 : if (zone_movable_pfn[i])
7557 0 : pr_info(" Node %d: %#018Lx\n", i,
7558 : (u64)zone_movable_pfn[i] << PAGE_SHIFT);
7559 : }
7560 :
7561 : /*
7562 : * Print out the early node map, and initialize the
7563 : * subsection-map relative to active online memory ranges to
7564 : * enable future "sub-section" extensions of the memory map.
7565 : */
7566 1 : pr_info("Early memory node ranges\n");
7567 3 : for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7568 2 : pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
7569 : (u64)start_pfn << PAGE_SHIFT,
7570 : ((u64)end_pfn << PAGE_SHIFT) - 1);
7571 2 : subsection_map_init(start_pfn, end_pfn - start_pfn);
7572 : }
7573 :
7574 : /* Initialise every node */
7575 1 : mminit_verify_pageflags_layout();
7576 1 : setup_nr_node_ids();
7577 2 : for_each_online_node(nid) {
7578 1 : pg_data_t *pgdat = NODE_DATA(nid);
7579 1 : free_area_init_node(nid);
7580 :
7581 : /* Any memory on that node */
7582 1 : if (pgdat->node_present_pages)
7583 1 : node_set_state(nid, N_MEMORY);
7584 1 : check_for_memory(pgdat, nid);
7585 : }
7586 1 : }
7587 :
7588 0 : static int __init cmdline_parse_core(char *p, unsigned long *core,
7589 : unsigned long *percent)
7590 : {
7591 0 : unsigned long long coremem;
7592 0 : char *endptr;
7593 :
7594 0 : if (!p)
7595 : return -EINVAL;
7596 :
7597 : /* Value may be a percentage of total memory, otherwise bytes */
7598 0 : coremem = simple_strtoull(p, &endptr, 0);
7599 0 : if (*endptr == '%') {
7600 : /* Paranoid check for percent values greater than 100 */
7601 0 : WARN_ON(coremem > 100);
7602 :
7603 0 : *percent = coremem;
7604 : } else {
7605 0 : coremem = memparse(p, &p);
7606 : /* Paranoid check that UL is enough for the coremem value */
7607 0 : WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
7608 :
7609 0 : *core = coremem >> PAGE_SHIFT;
7610 0 : *percent = 0UL;
7611 : }
7612 : return 0;
7613 : }
7614 :
7615 : /*
7616 : * kernelcore=size sets the amount of memory for use for allocations that
7617 : * cannot be reclaimed or migrated.
7618 : */
7619 0 : static int __init cmdline_parse_kernelcore(char *p)
7620 : {
7621 : /* parse kernelcore=mirror */
7622 0 : if (parse_option_str(p, "mirror")) {
7623 0 : mirrored_kernelcore = true;
7624 0 : return 0;
7625 : }
7626 :
7627 0 : return cmdline_parse_core(p, &required_kernelcore,
7628 : &required_kernelcore_percent);
7629 : }
7630 :
7631 : /*
7632 : * movablecore=size sets the amount of memory for use for allocations that
7633 : * can be reclaimed or migrated.
7634 : */
7635 0 : static int __init cmdline_parse_movablecore(char *p)
7636 : {
7637 0 : return cmdline_parse_core(p, &required_movablecore,
7638 : &required_movablecore_percent);
7639 : }
7640 :
7641 : early_param("kernelcore", cmdline_parse_kernelcore);
7642 : early_param("movablecore", cmdline_parse_movablecore);
7643 :
7644 964 : void adjust_managed_page_count(struct page *page, long count)
7645 : {
7646 964 : atomic_long_add(count, &page_zone(page)->managed_pages);
7647 964 : totalram_pages_add(count);
7648 : #ifdef CONFIG_HIGHMEM
7649 : if (PageHighMem(page))
7650 : totalhigh_pages_add(count);
7651 : #endif
7652 964 : }
7653 : EXPORT_SYMBOL(adjust_managed_page_count);
7654 :
7655 4 : unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
7656 : {
7657 4 : void *pos;
7658 4 : unsigned long pages = 0;
7659 :
7660 4 : start = (void *)PAGE_ALIGN((unsigned long)start);
7661 4 : end = (void *)((unsigned long)end & PAGE_MASK);
7662 968 : for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
7663 1928 : struct page *page = virt_to_page(pos);
7664 964 : void *direct_map_addr;
7665 :
7666 : /*
7667 : * 'direct_map_addr' might be different from 'pos'
7668 : * because some architectures' virt_to_page()
7669 : * work with aliases. Getting the direct map
7670 : * address ensures that we get a _writeable_
7671 : * alias for the memset().
7672 : */
7673 964 : direct_map_addr = page_address(page);
7674 : /*
7675 : * Perform a kasan-unchecked memset() since this memory
7676 : * has not been initialized.
7677 : */
7678 964 : direct_map_addr = kasan_reset_tag(direct_map_addr);
7679 964 : if ((unsigned int)poison <= 0xFF)
7680 964 : memset(direct_map_addr, poison, PAGE_SIZE);
7681 :
7682 964 : free_reserved_page(page);
7683 : }
7684 :
7685 4 : if (pages && s)
7686 4 : pr_info("Freeing %s memory: %ldK\n",
7687 : s, pages << (PAGE_SHIFT - 10));
7688 :
7689 4 : return pages;
7690 : }
7691 :
7692 1 : void __init mem_init_print_info(const char *str)
7693 : {
7694 1 : unsigned long physpages, codesize, datasize, rosize, bss_size;
7695 1 : unsigned long init_code_size, init_data_size;
7696 :
7697 1 : physpages = get_num_physpages();
7698 1 : codesize = _etext - _stext;
7699 1 : datasize = _edata - _sdata;
7700 1 : rosize = __end_rodata - __start_rodata;
7701 1 : bss_size = __bss_stop - __bss_start;
7702 1 : init_data_size = __init_end - __init_begin;
7703 1 : init_code_size = _einittext - _sinittext;
7704 :
7705 : /*
7706 : * Detect special cases and adjust section sizes accordingly:
7707 : * 1) .init.* may be embedded into .data sections
7708 : * 2) .init.text.* may be out of [__init_begin, __init_end],
7709 : * please refer to arch/tile/kernel/vmlinux.lds.S.
7710 : * 3) .rodata.* may be embedded into .text or .data sections.
7711 : */
7712 : #define adj_init_size(start, end, size, pos, adj) \
7713 : do { \
7714 : if (start <= pos && pos < end && size > adj) \
7715 : size -= adj; \
7716 : } while (0)
7717 :
7718 1 : adj_init_size(__init_begin, __init_end, init_data_size,
7719 : _sinittext, init_code_size);
7720 1 : adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
7721 1 : adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
7722 1 : adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
7723 1 : adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
7724 :
7725 : #undef adj_init_size
7726 :
7727 3 : pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
7728 : #ifdef CONFIG_HIGHMEM
7729 : ", %luK highmem"
7730 : #endif
7731 : "%s%s)\n",
7732 : nr_free_pages() << (PAGE_SHIFT - 10),
7733 : physpages << (PAGE_SHIFT - 10),
7734 : codesize >> 10, datasize >> 10, rosize >> 10,
7735 : (init_data_size + init_code_size) >> 10, bss_size >> 10,
7736 : (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10),
7737 : totalcma_pages << (PAGE_SHIFT - 10),
7738 : #ifdef CONFIG_HIGHMEM
7739 : totalhigh_pages() << (PAGE_SHIFT - 10),
7740 : #endif
7741 : str ? ", " : "", str ? str : "");
7742 1 : }
7743 :
7744 : /**
7745 : * set_dma_reserve - set the specified number of pages reserved in the first zone
7746 : * @new_dma_reserve: The number of pages to mark reserved
7747 : *
7748 : * The per-cpu batchsize and zone watermarks are determined by managed_pages.
7749 : * In the DMA zone, a significant percentage may be consumed by kernel image
7750 : * and other unfreeable allocations which can skew the watermarks badly. This
7751 : * function may optionally be used to account for unfreeable pages in the
7752 : * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
7753 : * smaller per-cpu batchsize.
7754 : */
7755 1 : void __init set_dma_reserve(unsigned long new_dma_reserve)
7756 : {
7757 1 : dma_reserve = new_dma_reserve;
7758 1 : }
7759 :
7760 0 : static int page_alloc_cpu_dead(unsigned int cpu)
7761 : {
7762 :
7763 0 : lru_add_drain_cpu(cpu);
7764 0 : drain_pages(cpu);
7765 :
7766 : /*
7767 : * Spill the event counters of the dead processor
7768 : * into the current processors event counters.
7769 : * This artificially elevates the count of the current
7770 : * processor.
7771 : */
7772 0 : vm_events_fold_cpu(cpu);
7773 :
7774 : /*
7775 : * Zero the differential counters of the dead processor
7776 : * so that the vm statistics are consistent.
7777 : *
7778 : * This is only okay since the processor is dead and cannot
7779 : * race with what we are doing.
7780 : */
7781 0 : cpu_vm_stats_fold(cpu);
7782 0 : return 0;
7783 : }
7784 :
7785 : #ifdef CONFIG_NUMA
7786 : int hashdist = HASHDIST_DEFAULT;
7787 :
7788 0 : static int __init set_hashdist(char *str)
7789 : {
7790 0 : if (!str)
7791 : return 0;
7792 0 : hashdist = simple_strtoul(str, &str, 0);
7793 0 : return 1;
7794 : }
7795 : __setup("hashdist=", set_hashdist);
7796 : #endif
7797 :
7798 1 : void __init page_alloc_init(void)
7799 : {
7800 1 : int ret;
7801 :
7802 : #ifdef CONFIG_NUMA
7803 1 : if (num_node_state(N_MEMORY) == 1)
7804 1 : hashdist = 0;
7805 : #endif
7806 :
7807 1 : ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
7808 : "mm/page_alloc:dead", NULL,
7809 : page_alloc_cpu_dead);
7810 1 : WARN_ON(ret < 0);
7811 1 : }
7812 :
7813 : /*
7814 : * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
7815 : * or min_free_kbytes changes.
7816 : */
7817 3 : static void calculate_totalreserve_pages(void)
7818 : {
7819 3 : struct pglist_data *pgdat;
7820 3 : unsigned long reserve_pages = 0;
7821 3 : enum zone_type i, j;
7822 :
7823 6 : for_each_online_pgdat(pgdat) {
7824 :
7825 3 : pgdat->totalreserve_pages = 0;
7826 :
7827 12 : for (i = 0; i < MAX_NR_ZONES; i++) {
7828 9 : struct zone *zone = pgdat->node_zones + i;
7829 9 : long max = 0;
7830 9 : unsigned long managed_pages = zone_managed_pages(zone);
7831 :
7832 : /* Find valid and maximum lowmem_reserve in the zone */
7833 27 : for (j = i; j < MAX_NR_ZONES; j++) {
7834 18 : if (zone->lowmem_reserve[j] > max)
7835 : max = zone->lowmem_reserve[j];
7836 : }
7837 :
7838 : /* we treat the high watermark as reserved pages. */
7839 9 : max += high_wmark_pages(zone);
7840 :
7841 9 : if (max > managed_pages)
7842 0 : max = managed_pages;
7843 :
7844 9 : pgdat->totalreserve_pages += max;
7845 :
7846 9 : reserve_pages += max;
7847 : }
7848 : }
7849 3 : totalreserve_pages = reserve_pages;
7850 3 : }
7851 :
7852 : /*
7853 : * setup_per_zone_lowmem_reserve - called whenever
7854 : * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
7855 : * has a correct pages reserved value, so an adequate number of
7856 : * pages are left in the zone after a successful __alloc_pages().
7857 : */
7858 1 : static void setup_per_zone_lowmem_reserve(void)
7859 : {
7860 1 : struct pglist_data *pgdat;
7861 1 : enum zone_type i, j;
7862 :
7863 2 : for_each_online_pgdat(pgdat) {
7864 3 : for (i = 0; i < MAX_NR_ZONES - 1; i++) {
7865 2 : struct zone *zone = &pgdat->node_zones[i];
7866 2 : int ratio = sysctl_lowmem_reserve_ratio[i];
7867 4 : bool clear = !ratio || !zone_managed_pages(zone);
7868 2 : unsigned long managed_pages = 0;
7869 :
7870 5 : for (j = i + 1; j < MAX_NR_ZONES; j++) {
7871 3 : if (clear) {
7872 1 : zone->lowmem_reserve[j] = 0;
7873 : } else {
7874 2 : struct zone *upper_zone = &pgdat->node_zones[j];
7875 :
7876 2 : managed_pages += zone_managed_pages(upper_zone);
7877 2 : zone->lowmem_reserve[j] = managed_pages / ratio;
7878 : }
7879 : }
7880 : }
7881 : }
7882 :
7883 : /* update totalreserve_pages */
7884 1 : calculate_totalreserve_pages();
7885 1 : }
7886 :
7887 2 : static void __setup_per_zone_wmarks(void)
7888 : {
7889 2 : unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
7890 2 : unsigned long lowmem_pages = 0;
7891 2 : struct zone *zone;
7892 2 : unsigned long flags;
7893 :
7894 : /* Calculate total number of !ZONE_HIGHMEM pages */
7895 8 : for_each_zone(zone) {
7896 6 : if (!is_highmem(zone))
7897 6 : lowmem_pages += zone_managed_pages(zone);
7898 : }
7899 :
7900 8 : for_each_zone(zone) {
7901 6 : u64 tmp;
7902 :
7903 6 : spin_lock_irqsave(&zone->lock, flags);
7904 6 : tmp = (u64)pages_min * zone_managed_pages(zone);
7905 6 : do_div(tmp, lowmem_pages);
7906 6 : if (is_highmem(zone)) {
7907 : /*
7908 : * __GFP_HIGH and PF_MEMALLOC allocations usually don't
7909 : * need highmem pages, so cap pages_min to a small
7910 : * value here.
7911 : *
7912 : * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
7913 : * deltas control async page reclaim, and so should
7914 : * not be capped for highmem.
7915 : */
7916 : unsigned long min_pages;
7917 :
7918 : min_pages = zone_managed_pages(zone) / 1024;
7919 : min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
7920 : zone->_watermark[WMARK_MIN] = min_pages;
7921 : } else {
7922 : /*
7923 : * If it's a lowmem zone, reserve a number of pages
7924 : * proportionate to the zone's size.
7925 : */
7926 6 : zone->_watermark[WMARK_MIN] = tmp;
7927 : }
7928 :
7929 : /*
7930 : * Set the kswapd watermarks distance according to the
7931 : * scale factor in proportion to available memory, but
7932 : * ensure a minimum size on small systems.
7933 : */
7934 6 : tmp = max_t(u64, tmp >> 2,
7935 : mult_frac(zone_managed_pages(zone),
7936 : watermark_scale_factor, 10000));
7937 :
7938 6 : zone->watermark_boost = 0;
7939 6 : zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
7940 6 : zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
7941 :
7942 6 : spin_unlock_irqrestore(&zone->lock, flags);
7943 : }
7944 :
7945 : /* update totalreserve_pages */
7946 2 : calculate_totalreserve_pages();
7947 2 : }
7948 :
7949 : /**
7950 : * setup_per_zone_wmarks - called when min_free_kbytes changes
7951 : * or when memory is hot-{added|removed}
7952 : *
7953 : * Ensures that the watermark[min,low,high] values for each zone are set
7954 : * correctly with respect to min_free_kbytes.
7955 : */
7956 2 : void setup_per_zone_wmarks(void)
7957 : {
7958 2 : static DEFINE_SPINLOCK(lock);
7959 :
7960 2 : spin_lock(&lock);
7961 2 : __setup_per_zone_wmarks();
7962 2 : spin_unlock(&lock);
7963 2 : }
7964 :
7965 : /*
7966 : * Initialise min_free_kbytes.
7967 : *
7968 : * For small machines we want it small (128k min). For large machines
7969 : * we want it large (256MB max). But it is not linear, because network
7970 : * bandwidth does not increase linearly with machine size. We use
7971 : *
7972 : * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
7973 : * min_free_kbytes = sqrt(lowmem_kbytes * 16)
7974 : *
7975 : * which yields
7976 : *
7977 : * 16MB: 512k
7978 : * 32MB: 724k
7979 : * 64MB: 1024k
7980 : * 128MB: 1448k
7981 : * 256MB: 2048k
7982 : * 512MB: 2896k
7983 : * 1024MB: 4096k
7984 : * 2048MB: 5792k
7985 : * 4096MB: 8192k
7986 : * 8192MB: 11584k
7987 : * 16384MB: 16384k
7988 : */
7989 1 : int __meminit init_per_zone_wmark_min(void)
7990 : {
7991 1 : unsigned long lowmem_kbytes;
7992 1 : int new_min_free_kbytes;
7993 :
7994 1 : lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
7995 1 : new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
7996 :
7997 1 : if (new_min_free_kbytes > user_min_free_kbytes) {
7998 1 : min_free_kbytes = new_min_free_kbytes;
7999 1 : if (min_free_kbytes < 128)
8000 0 : min_free_kbytes = 128;
8001 1 : if (min_free_kbytes > 262144)
8002 0 : min_free_kbytes = 262144;
8003 : } else {
8004 0 : pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
8005 : new_min_free_kbytes, user_min_free_kbytes);
8006 : }
8007 1 : setup_per_zone_wmarks();
8008 1 : refresh_zone_stat_thresholds();
8009 1 : setup_per_zone_lowmem_reserve();
8010 :
8011 : #ifdef CONFIG_NUMA
8012 1 : setup_min_unmapped_ratio();
8013 1 : setup_min_slab_ratio();
8014 : #endif
8015 :
8016 1 : khugepaged_min_free_kbytes_update();
8017 :
8018 1 : return 0;
8019 : }
8020 : postcore_initcall(init_per_zone_wmark_min)
8021 :
8022 : /*
8023 : * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
8024 : * that we can call two helper functions whenever min_free_kbytes
8025 : * changes.
8026 : */
8027 0 : int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
8028 : void *buffer, size_t *length, loff_t *ppos)
8029 : {
8030 0 : int rc;
8031 :
8032 0 : rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8033 0 : if (rc)
8034 : return rc;
8035 :
8036 0 : if (write) {
8037 0 : user_min_free_kbytes = min_free_kbytes;
8038 0 : setup_per_zone_wmarks();
8039 : }
8040 : return 0;
8041 : }
8042 :
8043 0 : int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
8044 : void *buffer, size_t *length, loff_t *ppos)
8045 : {
8046 0 : int rc;
8047 :
8048 0 : rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8049 0 : if (rc)
8050 : return rc;
8051 :
8052 0 : if (write)
8053 0 : setup_per_zone_wmarks();
8054 :
8055 : return 0;
8056 : }
8057 :
8058 : #ifdef CONFIG_NUMA
8059 1 : static void setup_min_unmapped_ratio(void)
8060 : {
8061 1 : pg_data_t *pgdat;
8062 1 : struct zone *zone;
8063 :
8064 2 : for_each_online_pgdat(pgdat)
8065 1 : pgdat->min_unmapped_pages = 0;
8066 :
8067 4 : for_each_zone(zone)
8068 3 : zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) *
8069 3 : sysctl_min_unmapped_ratio) / 100;
8070 1 : }
8071 :
8072 :
8073 0 : int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
8074 : void *buffer, size_t *length, loff_t *ppos)
8075 : {
8076 0 : int rc;
8077 :
8078 0 : rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8079 0 : if (rc)
8080 : return rc;
8081 :
8082 0 : setup_min_unmapped_ratio();
8083 :
8084 0 : return 0;
8085 : }
8086 :
8087 1 : static void setup_min_slab_ratio(void)
8088 : {
8089 1 : pg_data_t *pgdat;
8090 1 : struct zone *zone;
8091 :
8092 2 : for_each_online_pgdat(pgdat)
8093 1 : pgdat->min_slab_pages = 0;
8094 :
8095 4 : for_each_zone(zone)
8096 3 : zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) *
8097 3 : sysctl_min_slab_ratio) / 100;
8098 1 : }
8099 :
8100 0 : int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
8101 : void *buffer, size_t *length, loff_t *ppos)
8102 : {
8103 0 : int rc;
8104 :
8105 0 : rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8106 0 : if (rc)
8107 : return rc;
8108 :
8109 0 : setup_min_slab_ratio();
8110 :
8111 0 : return 0;
8112 : }
8113 : #endif
8114 :
8115 : /*
8116 : * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
8117 : * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
8118 : * whenever sysctl_lowmem_reserve_ratio changes.
8119 : *
8120 : * The reserve ratio obviously has absolutely no relation with the
8121 : * minimum watermarks. The lowmem reserve ratio can only make sense
8122 : * if in function of the boot time zone sizes.
8123 : */
8124 0 : int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
8125 : void *buffer, size_t *length, loff_t *ppos)
8126 : {
8127 0 : int i;
8128 :
8129 0 : proc_dointvec_minmax(table, write, buffer, length, ppos);
8130 :
8131 0 : for (i = 0; i < MAX_NR_ZONES; i++) {
8132 0 : if (sysctl_lowmem_reserve_ratio[i] < 1)
8133 0 : sysctl_lowmem_reserve_ratio[i] = 0;
8134 : }
8135 :
8136 0 : setup_per_zone_lowmem_reserve();
8137 0 : return 0;
8138 : }
8139 :
8140 : /*
8141 : * percpu_pagelist_fraction - changes the pcp->high for each zone on each
8142 : * cpu. It is the fraction of total pages in each zone that a hot per cpu
8143 : * pagelist can have before it gets flushed back to buddy allocator.
8144 : */
8145 0 : int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
8146 : void *buffer, size_t *length, loff_t *ppos)
8147 : {
8148 0 : struct zone *zone;
8149 0 : int old_percpu_pagelist_fraction;
8150 0 : int ret;
8151 :
8152 0 : mutex_lock(&pcp_batch_high_lock);
8153 0 : old_percpu_pagelist_fraction = percpu_pagelist_fraction;
8154 :
8155 0 : ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
8156 0 : if (!write || ret < 0)
8157 0 : goto out;
8158 :
8159 : /* Sanity checking to avoid pcp imbalance */
8160 0 : if (percpu_pagelist_fraction &&
8161 : percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
8162 0 : percpu_pagelist_fraction = old_percpu_pagelist_fraction;
8163 0 : ret = -EINVAL;
8164 0 : goto out;
8165 : }
8166 :
8167 : /* No change? */
8168 0 : if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
8169 0 : goto out;
8170 :
8171 0 : for_each_populated_zone(zone)
8172 0 : zone_set_pageset_high_and_batch(zone);
8173 0 : out:
8174 0 : mutex_unlock(&pcp_batch_high_lock);
8175 0 : return ret;
8176 : }
8177 :
8178 : #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
8179 : /*
8180 : * Returns the number of pages that arch has reserved but
8181 : * is not known to alloc_large_system_hash().
8182 : */
8183 8 : static unsigned long __init arch_reserved_kernel_pages(void)
8184 : {
8185 8 : return 0;
8186 : }
8187 : #endif
8188 :
8189 : /*
8190 : * Adaptive scale is meant to reduce sizes of hash tables on large memory
8191 : * machines. As memory size is increased the scale is also increased but at
8192 : * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory
8193 : * quadruples the scale is increased by one, which means the size of hash table
8194 : * only doubles, instead of quadrupling as well.
8195 : * Because 32-bit systems cannot have large physical memory, where this scaling
8196 : * makes sense, it is disabled on such platforms.
8197 : */
8198 : #if __BITS_PER_LONG > 32
8199 : #define ADAPT_SCALE_BASE (64ul << 30)
8200 : #define ADAPT_SCALE_SHIFT 2
8201 : #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT)
8202 : #endif
8203 :
8204 : /*
8205 : * allocate a large system hash table from bootmem
8206 : * - it is assumed that the hash table must contain an exact power-of-2
8207 : * quantity of entries
8208 : * - limit is the number of hash buckets, not the total allocation size
8209 : */
8210 11 : void *__init alloc_large_system_hash(const char *tablename,
8211 : unsigned long bucketsize,
8212 : unsigned long numentries,
8213 : int scale,
8214 : int flags,
8215 : unsigned int *_hash_shift,
8216 : unsigned int *_hash_mask,
8217 : unsigned long low_limit,
8218 : unsigned long high_limit)
8219 : {
8220 11 : unsigned long long max = high_limit;
8221 11 : unsigned long log2qty, size;
8222 11 : void *table = NULL;
8223 11 : gfp_t gfp_flags;
8224 11 : bool virt;
8225 :
8226 : /* allow the kernel cmdline to have a say */
8227 11 : if (!numentries) {
8228 : /* round applicable memory size up to nearest megabyte */
8229 8 : numentries = nr_kernel_pages;
8230 8 : numentries -= arch_reserved_kernel_pages();
8231 :
8232 : /* It isn't necessary when PAGE_SIZE >= 1MB */
8233 8 : if (PAGE_SHIFT < 20)
8234 8 : numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
8235 :
8236 : #if __BITS_PER_LONG > 32
8237 8 : if (!high_limit) {
8238 : unsigned long adapt;
8239 :
8240 4 : for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
8241 0 : adapt <<= ADAPT_SCALE_SHIFT)
8242 0 : scale++;
8243 : }
8244 : #endif
8245 :
8246 : /* limit to 1 bucket per 2^scale bytes of low memory */
8247 8 : if (scale > PAGE_SHIFT)
8248 8 : numentries >>= (scale - PAGE_SHIFT);
8249 : else
8250 0 : numentries <<= (PAGE_SHIFT - scale);
8251 :
8252 : /* Make sure we've got at least a 0-order allocation.. */
8253 8 : if (unlikely(flags & HASH_SMALL)) {
8254 : /* Makes no sense without HASH_EARLY */
8255 0 : WARN_ON(!(flags & HASH_EARLY));
8256 0 : if (!(numentries >> *_hash_shift)) {
8257 0 : numentries = 1UL << *_hash_shift;
8258 0 : BUG_ON(!numentries);
8259 : }
8260 8 : } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
8261 0 : numentries = PAGE_SIZE / bucketsize;
8262 : }
8263 11 : numentries = roundup_pow_of_two(numentries);
8264 :
8265 : /* limit allocation size to 1/16 total memory by default */
8266 11 : if (max == 0) {
8267 4 : max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
8268 4 : do_div(max, bucketsize);
8269 : }
8270 11 : max = min(max, 0x80000000ULL);
8271 :
8272 11 : if (numentries < low_limit)
8273 : numentries = low_limit;
8274 11 : if (numentries > max)
8275 : numentries = max;
8276 :
8277 11 : log2qty = ilog2(numentries);
8278 :
8279 17 : gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
8280 11 : do {
8281 11 : virt = false;
8282 11 : size = bucketsize << log2qty;
8283 11 : if (flags & HASH_EARLY) {
8284 3 : if (flags & HASH_ZERO)
8285 3 : table = memblock_alloc(size, SMP_CACHE_BYTES);
8286 : else
8287 0 : table = memblock_alloc_raw(size,
8288 : SMP_CACHE_BYTES);
8289 16 : } else if (get_order(size) >= MAX_ORDER || hashdist) {
8290 0 : table = __vmalloc(size, gfp_flags);
8291 0 : virt = true;
8292 : } else {
8293 : /*
8294 : * If bucketsize is not a power-of-two, we may free
8295 : * some pages at the end of hash table which
8296 : * alloc_pages_exact() automatically does
8297 : */
8298 8 : table = alloc_pages_exact(size, gfp_flags);
8299 8 : kmemleak_alloc(table, size, 1, gfp_flags);
8300 : }
8301 11 : } while (!table && size > PAGE_SIZE && --log2qty);
8302 :
8303 11 : if (!table)
8304 0 : panic("Failed to allocate %s hash table\n", tablename);
8305 :
8306 22 : pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
8307 : tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size,
8308 : virt ? "vmalloc" : "linear");
8309 :
8310 11 : if (_hash_shift)
8311 9 : *_hash_shift = log2qty;
8312 11 : if (_hash_mask)
8313 7 : *_hash_mask = (1 << log2qty) - 1;
8314 :
8315 11 : return table;
8316 : }
8317 :
8318 : /*
8319 : * This function checks whether pageblock includes unmovable pages or not.
8320 : *
8321 : * PageLRU check without isolation or lru_lock could race so that
8322 : * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
8323 : * check without lock_page also may miss some movable non-lru pages at
8324 : * race condition. So you can't expect this function should be exact.
8325 : *
8326 : * Returns a page without holding a reference. If the caller wants to
8327 : * dereference that page (e.g., dumping), it has to make sure that it
8328 : * cannot get removed (e.g., via memory unplug) concurrently.
8329 : *
8330 : */
8331 0 : struct page *has_unmovable_pages(struct zone *zone, struct page *page,
8332 : int migratetype, int flags)
8333 : {
8334 0 : unsigned long iter = 0;
8335 0 : unsigned long pfn = page_to_pfn(page);
8336 0 : unsigned long offset = pfn % pageblock_nr_pages;
8337 :
8338 0 : if (is_migrate_cma_page(page)) {
8339 : /*
8340 : * CMA allocations (alloc_contig_range) really need to mark
8341 : * isolate CMA pageblocks even when they are not movable in fact
8342 : * so consider them movable here.
8343 : */
8344 : if (is_migrate_cma(migratetype))
8345 : return NULL;
8346 :
8347 : return page;
8348 : }
8349 :
8350 0 : for (; iter < pageblock_nr_pages - offset; iter++) {
8351 0 : if (!pfn_valid_within(pfn + iter))
8352 : continue;
8353 :
8354 0 : page = pfn_to_page(pfn + iter);
8355 :
8356 : /*
8357 : * Both, bootmem allocations and memory holes are marked
8358 : * PG_reserved and are unmovable. We can even have unmovable
8359 : * allocations inside ZONE_MOVABLE, for example when
8360 : * specifying "movablecore".
8361 : */
8362 0 : if (PageReserved(page))
8363 0 : return page;
8364 :
8365 : /*
8366 : * If the zone is movable and we have ruled out all reserved
8367 : * pages then it should be reasonably safe to assume the rest
8368 : * is movable.
8369 : */
8370 0 : if (zone_idx(zone) == ZONE_MOVABLE)
8371 0 : continue;
8372 :
8373 : /*
8374 : * Hugepages are not in LRU lists, but they're movable.
8375 : * THPs are on the LRU, but need to be counted as #small pages.
8376 : * We need not scan over tail pages because we don't
8377 : * handle each tail page individually in migration.
8378 : */
8379 0 : if (PageHuge(page) || PageTransCompound(page)) {
8380 0 : struct page *head = compound_head(page);
8381 0 : unsigned int skip_pages;
8382 :
8383 0 : if (PageHuge(page)) {
8384 : if (!hugepage_migration_supported(page_hstate(head)))
8385 : return page;
8386 0 : } else if (!PageLRU(head) && !__PageMovable(head)) {
8387 0 : return page;
8388 : }
8389 :
8390 0 : skip_pages = compound_nr(head) - (page - head);
8391 0 : iter += skip_pages - 1;
8392 0 : continue;
8393 : }
8394 :
8395 : /*
8396 : * We can't use page_count without pin a page
8397 : * because another CPU can free compound page.
8398 : * This check already skips compound tails of THP
8399 : * because their page->_refcount is zero at all time.
8400 : */
8401 0 : if (!page_ref_count(page)) {
8402 0 : if (PageBuddy(page))
8403 0 : iter += (1 << buddy_order(page)) - 1;
8404 0 : continue;
8405 : }
8406 :
8407 : /*
8408 : * The HWPoisoned page may be not in buddy system, and
8409 : * page_count() is not 0.
8410 : */
8411 0 : if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
8412 : continue;
8413 :
8414 : /*
8415 : * We treat all PageOffline() pages as movable when offlining
8416 : * to give drivers a chance to decrement their reference count
8417 : * in MEM_GOING_OFFLINE in order to indicate that these pages
8418 : * can be offlined as there are no direct references anymore.
8419 : * For actually unmovable PageOffline() where the driver does
8420 : * not support this, we will fail later when trying to actually
8421 : * move these pages that still have a reference count > 0.
8422 : * (false negatives in this function only)
8423 : */
8424 0 : if ((flags & MEMORY_OFFLINE) && PageOffline(page))
8425 0 : continue;
8426 :
8427 0 : if (__PageMovable(page) || PageLRU(page))
8428 0 : continue;
8429 :
8430 : /*
8431 : * If there are RECLAIMABLE pages, we need to check
8432 : * it. But now, memory offline itself doesn't call
8433 : * shrink_node_slabs() and it still to be fixed.
8434 : */
8435 : return page;
8436 : }
8437 : return NULL;
8438 : }
8439 :
8440 : #ifdef CONFIG_CONTIG_ALLOC
8441 : static unsigned long pfn_max_align_down(unsigned long pfn)
8442 : {
8443 : return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
8444 : pageblock_nr_pages) - 1);
8445 : }
8446 :
8447 : static unsigned long pfn_max_align_up(unsigned long pfn)
8448 : {
8449 : return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
8450 : pageblock_nr_pages));
8451 : }
8452 :
8453 : /* [start, end) must belong to a single zone. */
8454 : static int __alloc_contig_migrate_range(struct compact_control *cc,
8455 : unsigned long start, unsigned long end)
8456 : {
8457 : /* This function is based on compact_zone() from compaction.c. */
8458 : unsigned int nr_reclaimed;
8459 : unsigned long pfn = start;
8460 : unsigned int tries = 0;
8461 : int ret = 0;
8462 : struct migration_target_control mtc = {
8463 : .nid = zone_to_nid(cc->zone),
8464 : .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
8465 : };
8466 :
8467 : migrate_prep();
8468 :
8469 : while (pfn < end || !list_empty(&cc->migratepages)) {
8470 : if (fatal_signal_pending(current)) {
8471 : ret = -EINTR;
8472 : break;
8473 : }
8474 :
8475 : if (list_empty(&cc->migratepages)) {
8476 : cc->nr_migratepages = 0;
8477 : pfn = isolate_migratepages_range(cc, pfn, end);
8478 : if (!pfn) {
8479 : ret = -EINTR;
8480 : break;
8481 : }
8482 : tries = 0;
8483 : } else if (++tries == 5) {
8484 : ret = ret < 0 ? ret : -EBUSY;
8485 : break;
8486 : }
8487 :
8488 : nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
8489 : &cc->migratepages);
8490 : cc->nr_migratepages -= nr_reclaimed;
8491 :
8492 : ret = migrate_pages(&cc->migratepages, alloc_migration_target,
8493 : NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE);
8494 : }
8495 : if (ret < 0) {
8496 : putback_movable_pages(&cc->migratepages);
8497 : return ret;
8498 : }
8499 : return 0;
8500 : }
8501 :
8502 : /**
8503 : * alloc_contig_range() -- tries to allocate given range of pages
8504 : * @start: start PFN to allocate
8505 : * @end: one-past-the-last PFN to allocate
8506 : * @migratetype: migratetype of the underlaying pageblocks (either
8507 : * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
8508 : * in range must have the same migratetype and it must
8509 : * be either of the two.
8510 : * @gfp_mask: GFP mask to use during compaction
8511 : *
8512 : * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
8513 : * aligned. The PFN range must belong to a single zone.
8514 : *
8515 : * The first thing this routine does is attempt to MIGRATE_ISOLATE all
8516 : * pageblocks in the range. Once isolated, the pageblocks should not
8517 : * be modified by others.
8518 : *
8519 : * Return: zero on success or negative error code. On success all
8520 : * pages which PFN is in [start, end) are allocated for the caller and
8521 : * need to be freed with free_contig_range().
8522 : */
8523 : int alloc_contig_range(unsigned long start, unsigned long end,
8524 : unsigned migratetype, gfp_t gfp_mask)
8525 : {
8526 : unsigned long outer_start, outer_end;
8527 : unsigned int order;
8528 : int ret = 0;
8529 :
8530 : struct compact_control cc = {
8531 : .nr_migratepages = 0,
8532 : .order = -1,
8533 : .zone = page_zone(pfn_to_page(start)),
8534 : .mode = MIGRATE_SYNC,
8535 : .ignore_skip_hint = true,
8536 : .no_set_skip_hint = true,
8537 : .gfp_mask = current_gfp_context(gfp_mask),
8538 : .alloc_contig = true,
8539 : };
8540 : INIT_LIST_HEAD(&cc.migratepages);
8541 :
8542 : /*
8543 : * What we do here is we mark all pageblocks in range as
8544 : * MIGRATE_ISOLATE. Because pageblock and max order pages may
8545 : * have different sizes, and due to the way page allocator
8546 : * work, we align the range to biggest of the two pages so
8547 : * that page allocator won't try to merge buddies from
8548 : * different pageblocks and change MIGRATE_ISOLATE to some
8549 : * other migration type.
8550 : *
8551 : * Once the pageblocks are marked as MIGRATE_ISOLATE, we
8552 : * migrate the pages from an unaligned range (ie. pages that
8553 : * we are interested in). This will put all the pages in
8554 : * range back to page allocator as MIGRATE_ISOLATE.
8555 : *
8556 : * When this is done, we take the pages in range from page
8557 : * allocator removing them from the buddy system. This way
8558 : * page allocator will never consider using them.
8559 : *
8560 : * This lets us mark the pageblocks back as
8561 : * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
8562 : * aligned range but not in the unaligned, original range are
8563 : * put back to page allocator so that buddy can use them.
8564 : */
8565 :
8566 : ret = start_isolate_page_range(pfn_max_align_down(start),
8567 : pfn_max_align_up(end), migratetype, 0);
8568 : if (ret)
8569 : return ret;
8570 :
8571 : drain_all_pages(cc.zone);
8572 :
8573 : /*
8574 : * In case of -EBUSY, we'd like to know which page causes problem.
8575 : * So, just fall through. test_pages_isolated() has a tracepoint
8576 : * which will report the busy page.
8577 : *
8578 : * It is possible that busy pages could become available before
8579 : * the call to test_pages_isolated, and the range will actually be
8580 : * allocated. So, if we fall through be sure to clear ret so that
8581 : * -EBUSY is not accidentally used or returned to caller.
8582 : */
8583 : ret = __alloc_contig_migrate_range(&cc, start, end);
8584 : if (ret && ret != -EBUSY)
8585 : goto done;
8586 : ret =0;
8587 :
8588 : /*
8589 : * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
8590 : * aligned blocks that are marked as MIGRATE_ISOLATE. What's
8591 : * more, all pages in [start, end) are free in page allocator.
8592 : * What we are going to do is to allocate all pages from
8593 : * [start, end) (that is remove them from page allocator).
8594 : *
8595 : * The only problem is that pages at the beginning and at the
8596 : * end of interesting range may be not aligned with pages that
8597 : * page allocator holds, ie. they can be part of higher order
8598 : * pages. Because of this, we reserve the bigger range and
8599 : * once this is done free the pages we are not interested in.
8600 : *
8601 : * We don't have to hold zone->lock here because the pages are
8602 : * isolated thus they won't get removed from buddy.
8603 : */
8604 :
8605 : lru_add_drain_all();
8606 :
8607 : order = 0;
8608 : outer_start = start;
8609 : while (!PageBuddy(pfn_to_page(outer_start))) {
8610 : if (++order >= MAX_ORDER) {
8611 : outer_start = start;
8612 : break;
8613 : }
8614 : outer_start &= ~0UL << order;
8615 : }
8616 :
8617 : if (outer_start != start) {
8618 : order = buddy_order(pfn_to_page(outer_start));
8619 :
8620 : /*
8621 : * outer_start page could be small order buddy page and
8622 : * it doesn't include start page. Adjust outer_start
8623 : * in this case to report failed page properly
8624 : * on tracepoint in test_pages_isolated()
8625 : */
8626 : if (outer_start + (1UL << order) <= start)
8627 : outer_start = start;
8628 : }
8629 :
8630 : /* Make sure the range is really isolated. */
8631 : if (test_pages_isolated(outer_start, end, 0)) {
8632 : pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n",
8633 : __func__, outer_start, end);
8634 : ret = -EBUSY;
8635 : goto done;
8636 : }
8637 :
8638 : /* Grab isolated pages from freelists. */
8639 : outer_end = isolate_freepages_range(&cc, outer_start, end);
8640 : if (!outer_end) {
8641 : ret = -EBUSY;
8642 : goto done;
8643 : }
8644 :
8645 : /* Free head and tail (if any) */
8646 : if (start != outer_start)
8647 : free_contig_range(outer_start, start - outer_start);
8648 : if (end != outer_end)
8649 : free_contig_range(end, outer_end - end);
8650 :
8651 : done:
8652 : undo_isolate_page_range(pfn_max_align_down(start),
8653 : pfn_max_align_up(end), migratetype);
8654 : return ret;
8655 : }
8656 : EXPORT_SYMBOL(alloc_contig_range);
8657 :
8658 : static int __alloc_contig_pages(unsigned long start_pfn,
8659 : unsigned long nr_pages, gfp_t gfp_mask)
8660 : {
8661 : unsigned long end_pfn = start_pfn + nr_pages;
8662 :
8663 : return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE,
8664 : gfp_mask);
8665 : }
8666 :
8667 : static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn,
8668 : unsigned long nr_pages)
8669 : {
8670 : unsigned long i, end_pfn = start_pfn + nr_pages;
8671 : struct page *page;
8672 :
8673 : for (i = start_pfn; i < end_pfn; i++) {
8674 : page = pfn_to_online_page(i);
8675 : if (!page)
8676 : return false;
8677 :
8678 : if (page_zone(page) != z)
8679 : return false;
8680 :
8681 : if (PageReserved(page))
8682 : return false;
8683 :
8684 : if (page_count(page) > 0)
8685 : return false;
8686 :
8687 : if (PageHuge(page))
8688 : return false;
8689 : }
8690 : return true;
8691 : }
8692 :
8693 : static bool zone_spans_last_pfn(const struct zone *zone,
8694 : unsigned long start_pfn, unsigned long nr_pages)
8695 : {
8696 : unsigned long last_pfn = start_pfn + nr_pages - 1;
8697 :
8698 : return zone_spans_pfn(zone, last_pfn);
8699 : }
8700 :
8701 : /**
8702 : * alloc_contig_pages() -- tries to find and allocate contiguous range of pages
8703 : * @nr_pages: Number of contiguous pages to allocate
8704 : * @gfp_mask: GFP mask to limit search and used during compaction
8705 : * @nid: Target node
8706 : * @nodemask: Mask for other possible nodes
8707 : *
8708 : * This routine is a wrapper around alloc_contig_range(). It scans over zones
8709 : * on an applicable zonelist to find a contiguous pfn range which can then be
8710 : * tried for allocation with alloc_contig_range(). This routine is intended
8711 : * for allocation requests which can not be fulfilled with the buddy allocator.
8712 : *
8713 : * The allocated memory is always aligned to a page boundary. If nr_pages is a
8714 : * power of two then the alignment is guaranteed to be to the given nr_pages
8715 : * (e.g. 1GB request would be aligned to 1GB).
8716 : *
8717 : * Allocated pages can be freed with free_contig_range() or by manually calling
8718 : * __free_page() on each allocated page.
8719 : *
8720 : * Return: pointer to contiguous pages on success, or NULL if not successful.
8721 : */
8722 : struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
8723 : int nid, nodemask_t *nodemask)
8724 : {
8725 : unsigned long ret, pfn, flags;
8726 : struct zonelist *zonelist;
8727 : struct zone *zone;
8728 : struct zoneref *z;
8729 :
8730 : zonelist = node_zonelist(nid, gfp_mask);
8731 : for_each_zone_zonelist_nodemask(zone, z, zonelist,
8732 : gfp_zone(gfp_mask), nodemask) {
8733 : spin_lock_irqsave(&zone->lock, flags);
8734 :
8735 : pfn = ALIGN(zone->zone_start_pfn, nr_pages);
8736 : while (zone_spans_last_pfn(zone, pfn, nr_pages)) {
8737 : if (pfn_range_valid_contig(zone, pfn, nr_pages)) {
8738 : /*
8739 : * We release the zone lock here because
8740 : * alloc_contig_range() will also lock the zone
8741 : * at some point. If there's an allocation
8742 : * spinning on this lock, it may win the race
8743 : * and cause alloc_contig_range() to fail...
8744 : */
8745 : spin_unlock_irqrestore(&zone->lock, flags);
8746 : ret = __alloc_contig_pages(pfn, nr_pages,
8747 : gfp_mask);
8748 : if (!ret)
8749 : return pfn_to_page(pfn);
8750 : spin_lock_irqsave(&zone->lock, flags);
8751 : }
8752 : pfn += nr_pages;
8753 : }
8754 : spin_unlock_irqrestore(&zone->lock, flags);
8755 : }
8756 : return NULL;
8757 : }
8758 : #endif /* CONFIG_CONTIG_ALLOC */
8759 :
8760 0 : void free_contig_range(unsigned long pfn, unsigned int nr_pages)
8761 : {
8762 0 : unsigned int count = 0;
8763 :
8764 0 : for (; nr_pages--; pfn++) {
8765 0 : struct page *page = pfn_to_page(pfn);
8766 :
8767 0 : count += page_count(page) != 1;
8768 0 : __free_page(page);
8769 : }
8770 0 : WARN(count != 0, "%d pages are still in use!\n", count);
8771 0 : }
8772 : EXPORT_SYMBOL(free_contig_range);
8773 :
8774 : /*
8775 : * The zone indicated has a new number of managed_pages; batch sizes and percpu
8776 : * page high values need to be recalulated.
8777 : */
8778 0 : void __meminit zone_pcp_update(struct zone *zone)
8779 : {
8780 0 : mutex_lock(&pcp_batch_high_lock);
8781 0 : zone_set_pageset_high_and_batch(zone);
8782 0 : mutex_unlock(&pcp_batch_high_lock);
8783 0 : }
8784 :
8785 : /*
8786 : * Effectively disable pcplists for the zone by setting the high limit to 0
8787 : * and draining all cpus. A concurrent page freeing on another CPU that's about
8788 : * to put the page on pcplist will either finish before the drain and the page
8789 : * will be drained, or observe the new high limit and skip the pcplist.
8790 : *
8791 : * Must be paired with a call to zone_pcp_enable().
8792 : */
8793 0 : void zone_pcp_disable(struct zone *zone)
8794 : {
8795 0 : mutex_lock(&pcp_batch_high_lock);
8796 0 : __zone_set_pageset_high_and_batch(zone, 0, 1);
8797 0 : __drain_all_pages(zone, true);
8798 0 : }
8799 :
8800 0 : void zone_pcp_enable(struct zone *zone)
8801 : {
8802 0 : __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch);
8803 0 : mutex_unlock(&pcp_batch_high_lock);
8804 0 : }
8805 :
8806 0 : void zone_pcp_reset(struct zone *zone)
8807 : {
8808 0 : unsigned long flags;
8809 0 : int cpu;
8810 0 : struct per_cpu_pageset *pset;
8811 :
8812 : /* avoid races with drain_pages() */
8813 0 : local_irq_save(flags);
8814 0 : if (zone->pageset != &boot_pageset) {
8815 0 : for_each_online_cpu(cpu) {
8816 0 : pset = per_cpu_ptr(zone->pageset, cpu);
8817 0 : drain_zonestat(zone, pset);
8818 : }
8819 0 : free_percpu(zone->pageset);
8820 0 : zone->pageset = &boot_pageset;
8821 : }
8822 0 : local_irq_restore(flags);
8823 0 : }
8824 :
8825 : #ifdef CONFIG_MEMORY_HOTREMOVE
8826 : /*
8827 : * All pages in the range must be in a single zone, must not contain holes,
8828 : * must span full sections, and must be isolated before calling this function.
8829 : */
8830 : void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
8831 : {
8832 : unsigned long pfn = start_pfn;
8833 : struct page *page;
8834 : struct zone *zone;
8835 : unsigned int order;
8836 : unsigned long flags;
8837 :
8838 : offline_mem_sections(pfn, end_pfn);
8839 : zone = page_zone(pfn_to_page(pfn));
8840 : spin_lock_irqsave(&zone->lock, flags);
8841 : while (pfn < end_pfn) {
8842 : page = pfn_to_page(pfn);
8843 : /*
8844 : * The HWPoisoned page may be not in buddy system, and
8845 : * page_count() is not 0.
8846 : */
8847 : if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
8848 : pfn++;
8849 : continue;
8850 : }
8851 : /*
8852 : * At this point all remaining PageOffline() pages have a
8853 : * reference count of 0 and can simply be skipped.
8854 : */
8855 : if (PageOffline(page)) {
8856 : BUG_ON(page_count(page));
8857 : BUG_ON(PageBuddy(page));
8858 : pfn++;
8859 : continue;
8860 : }
8861 :
8862 : BUG_ON(page_count(page));
8863 : BUG_ON(!PageBuddy(page));
8864 : order = buddy_order(page);
8865 : del_page_from_free_list(page, zone, order);
8866 : pfn += (1 << order);
8867 : }
8868 : spin_unlock_irqrestore(&zone->lock, flags);
8869 : }
8870 : #endif
8871 :
8872 0 : bool is_free_buddy_page(struct page *page)
8873 : {
8874 0 : struct zone *zone = page_zone(page);
8875 0 : unsigned long pfn = page_to_pfn(page);
8876 0 : unsigned long flags;
8877 0 : unsigned int order;
8878 :
8879 0 : spin_lock_irqsave(&zone->lock, flags);
8880 0 : for (order = 0; order < MAX_ORDER; order++) {
8881 0 : struct page *page_head = page - (pfn & ((1 << order) - 1));
8882 :
8883 0 : if (PageBuddy(page_head) && buddy_order(page_head) >= order)
8884 : break;
8885 : }
8886 0 : spin_unlock_irqrestore(&zone->lock, flags);
8887 :
8888 0 : return order < MAX_ORDER;
8889 : }
8890 :
8891 : #ifdef CONFIG_MEMORY_FAILURE
8892 : /*
8893 : * Break down a higher-order page in sub-pages, and keep our target out of
8894 : * buddy allocator.
8895 : */
8896 : static void break_down_buddy_pages(struct zone *zone, struct page *page,
8897 : struct page *target, int low, int high,
8898 : int migratetype)
8899 : {
8900 : unsigned long size = 1 << high;
8901 : struct page *current_buddy, *next_page;
8902 :
8903 : while (high > low) {
8904 : high--;
8905 : size >>= 1;
8906 :
8907 : if (target >= &page[size]) {
8908 : next_page = page + size;
8909 : current_buddy = page;
8910 : } else {
8911 : next_page = page;
8912 : current_buddy = page + size;
8913 : }
8914 :
8915 : if (set_page_guard(zone, current_buddy, high, migratetype))
8916 : continue;
8917 :
8918 : if (current_buddy != target) {
8919 : add_to_free_list(current_buddy, zone, high, migratetype);
8920 : set_buddy_order(current_buddy, high);
8921 : page = next_page;
8922 : }
8923 : }
8924 : }
8925 :
8926 : /*
8927 : * Take a page that will be marked as poisoned off the buddy allocator.
8928 : */
8929 : bool take_page_off_buddy(struct page *page)
8930 : {
8931 : struct zone *zone = page_zone(page);
8932 : unsigned long pfn = page_to_pfn(page);
8933 : unsigned long flags;
8934 : unsigned int order;
8935 : bool ret = false;
8936 :
8937 : spin_lock_irqsave(&zone->lock, flags);
8938 : for (order = 0; order < MAX_ORDER; order++) {
8939 : struct page *page_head = page - (pfn & ((1 << order) - 1));
8940 : int page_order = buddy_order(page_head);
8941 :
8942 : if (PageBuddy(page_head) && page_order >= order) {
8943 : unsigned long pfn_head = page_to_pfn(page_head);
8944 : int migratetype = get_pfnblock_migratetype(page_head,
8945 : pfn_head);
8946 :
8947 : del_page_from_free_list(page_head, zone, page_order);
8948 : break_down_buddy_pages(zone, page_head, page, 0,
8949 : page_order, migratetype);
8950 : ret = true;
8951 : break;
8952 : }
8953 : if (page_count(page_head) > 0)
8954 : break;
8955 : }
8956 : spin_unlock_irqrestore(&zone->lock, flags);
8957 : return ret;
8958 : }
8959 : #endif
|
